1 /* Intel PRO/1000 Linux driver
2 * Copyright(c) 1999 - 2015 Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * The full GNU General Public License is included in this distribution in
14 * the file called "COPYING".
16 * Contact Information:
17 * Linux NICS <linux.nics@intel.com>
18 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
19 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 #include <linux/module.h>
25 #include <linux/types.h>
26 #include <linux/init.h>
27 #include <linux/pci.h>
28 #include <linux/vmalloc.h>
29 #include <linux/pagemap.h>
30 #include <linux/delay.h>
31 #include <linux/netdevice.h>
32 #include <linux/interrupt.h>
33 #include <linux/tcp.h>
34 #include <linux/ipv6.h>
35 #include <linux/slab.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/cpu.h>
41 #include <linux/smp.h>
42 #include <linux/pm_qos.h>
43 #include <linux/pm_runtime.h>
44 #include <linux/aer.h>
45 #include <linux/prefetch.h>
49 #define DRV_EXTRAVERSION "-k"
51 #define DRV_VERSION "3.2.6" DRV_EXTRAVERSION
52 char e1000e_driver_name[] = "e1000e";
53 const char e1000e_driver_version[] = DRV_VERSION;
55 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
56 static int debug = -1;
57 module_param(debug, int, 0);
58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
60 static const struct e1000_info *e1000_info_tbl[] = {
61 [board_82571] = &e1000_82571_info,
62 [board_82572] = &e1000_82572_info,
63 [board_82573] = &e1000_82573_info,
64 [board_82574] = &e1000_82574_info,
65 [board_82583] = &e1000_82583_info,
66 [board_80003es2lan] = &e1000_es2_info,
67 [board_ich8lan] = &e1000_ich8_info,
68 [board_ich9lan] = &e1000_ich9_info,
69 [board_ich10lan] = &e1000_ich10_info,
70 [board_pchlan] = &e1000_pch_info,
71 [board_pch2lan] = &e1000_pch2_info,
72 [board_pch_lpt] = &e1000_pch_lpt_info,
73 [board_pch_spt] = &e1000_pch_spt_info,
76 struct e1000_reg_info {
81 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
82 /* General Registers */
84 {E1000_STATUS, "STATUS"},
85 {E1000_CTRL_EXT, "CTRL_EXT"},
87 /* Interrupt Registers */
92 {E1000_RDLEN(0), "RDLEN"},
93 {E1000_RDH(0), "RDH"},
94 {E1000_RDT(0), "RDT"},
96 {E1000_RXDCTL(0), "RXDCTL"},
98 {E1000_RDBAL(0), "RDBAL"},
99 {E1000_RDBAH(0), "RDBAH"},
100 {E1000_RDFH, "RDFH"},
101 {E1000_RDFT, "RDFT"},
102 {E1000_RDFHS, "RDFHS"},
103 {E1000_RDFTS, "RDFTS"},
104 {E1000_RDFPC, "RDFPC"},
107 {E1000_TCTL, "TCTL"},
108 {E1000_TDBAL(0), "TDBAL"},
109 {E1000_TDBAH(0), "TDBAH"},
110 {E1000_TDLEN(0), "TDLEN"},
111 {E1000_TDH(0), "TDH"},
112 {E1000_TDT(0), "TDT"},
113 {E1000_TIDV, "TIDV"},
114 {E1000_TXDCTL(0), "TXDCTL"},
115 {E1000_TADV, "TADV"},
116 {E1000_TARC(0), "TARC"},
117 {E1000_TDFH, "TDFH"},
118 {E1000_TDFT, "TDFT"},
119 {E1000_TDFHS, "TDFHS"},
120 {E1000_TDFTS, "TDFTS"},
121 {E1000_TDFPC, "TDFPC"},
123 /* List Terminator */
128 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
129 * @hw: pointer to the HW structure
131 * When updating the MAC CSR registers, the Manageability Engine (ME) could
132 * be accessing the registers at the same time. Normally, this is handled in
133 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
134 * accesses later than it should which could result in the register to have
135 * an incorrect value. Workaround this by checking the FWSM register which
136 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
137 * and try again a number of times.
139 s32 __ew32_prepare(struct e1000_hw *hw)
141 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
143 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
149 void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
151 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
154 writel(val, hw->hw_addr + reg);
158 * e1000_regdump - register printout routine
159 * @hw: pointer to the HW structure
160 * @reginfo: pointer to the register info table
162 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
168 switch (reginfo->ofs) {
169 case E1000_RXDCTL(0):
170 for (n = 0; n < 2; n++)
171 regs[n] = __er32(hw, E1000_RXDCTL(n));
173 case E1000_TXDCTL(0):
174 for (n = 0; n < 2; n++)
175 regs[n] = __er32(hw, E1000_TXDCTL(n));
178 for (n = 0; n < 2; n++)
179 regs[n] = __er32(hw, E1000_TARC(n));
182 pr_info("%-15s %08x\n",
183 reginfo->name, __er32(hw, reginfo->ofs));
187 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
188 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
191 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
192 struct e1000_buffer *bi)
195 struct e1000_ps_page *ps_page;
197 for (i = 0; i < adapter->rx_ps_pages; i++) {
198 ps_page = &bi->ps_pages[i];
201 pr_info("packet dump for ps_page %d:\n", i);
202 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
203 16, 1, page_address(ps_page->page),
210 * e1000e_dump - Print registers, Tx-ring and Rx-ring
211 * @adapter: board private structure
213 static void e1000e_dump(struct e1000_adapter *adapter)
215 struct net_device *netdev = adapter->netdev;
216 struct e1000_hw *hw = &adapter->hw;
217 struct e1000_reg_info *reginfo;
218 struct e1000_ring *tx_ring = adapter->tx_ring;
219 struct e1000_tx_desc *tx_desc;
224 struct e1000_buffer *buffer_info;
225 struct e1000_ring *rx_ring = adapter->rx_ring;
226 union e1000_rx_desc_packet_split *rx_desc_ps;
227 union e1000_rx_desc_extended *rx_desc;
237 if (!netif_msg_hw(adapter))
240 /* Print netdevice Info */
242 dev_info(&adapter->pdev->dev, "Net device Info\n");
243 pr_info("Device Name state trans_start last_rx\n");
244 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
245 netdev->state, dev_trans_start(netdev), netdev->last_rx);
248 /* Print Registers */
249 dev_info(&adapter->pdev->dev, "Register Dump\n");
250 pr_info(" Register Name Value\n");
251 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
252 reginfo->name; reginfo++) {
253 e1000_regdump(hw, reginfo);
256 /* Print Tx Ring Summary */
257 if (!netdev || !netif_running(netdev))
260 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
261 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
262 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
263 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
264 0, tx_ring->next_to_use, tx_ring->next_to_clean,
265 (unsigned long long)buffer_info->dma,
267 buffer_info->next_to_watch,
268 (unsigned long long)buffer_info->time_stamp);
271 if (!netif_msg_tx_done(adapter))
272 goto rx_ring_summary;
274 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
276 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
278 * Legacy Transmit Descriptor
279 * +--------------------------------------------------------------+
280 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
281 * +--------------------------------------------------------------+
282 * 8 | Special | CSS | Status | CMD | CSO | Length |
283 * +--------------------------------------------------------------+
284 * 63 48 47 36 35 32 31 24 23 16 15 0
286 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
287 * 63 48 47 40 39 32 31 16 15 8 7 0
288 * +----------------------------------------------------------------+
289 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
290 * +----------------------------------------------------------------+
291 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
292 * +----------------------------------------------------------------+
293 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
295 * Extended Data Descriptor (DTYP=0x1)
296 * +----------------------------------------------------------------+
297 * 0 | Buffer Address [63:0] |
298 * +----------------------------------------------------------------+
299 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
300 * +----------------------------------------------------------------+
301 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
303 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
304 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
305 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
306 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
307 const char *next_desc;
308 tx_desc = E1000_TX_DESC(*tx_ring, i);
309 buffer_info = &tx_ring->buffer_info[i];
310 u0 = (struct my_u0 *)tx_desc;
311 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
312 next_desc = " NTC/U";
313 else if (i == tx_ring->next_to_use)
315 else if (i == tx_ring->next_to_clean)
319 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
320 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
321 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
323 (unsigned long long)le64_to_cpu(u0->a),
324 (unsigned long long)le64_to_cpu(u0->b),
325 (unsigned long long)buffer_info->dma,
326 buffer_info->length, buffer_info->next_to_watch,
327 (unsigned long long)buffer_info->time_stamp,
328 buffer_info->skb, next_desc);
330 if (netif_msg_pktdata(adapter) && buffer_info->skb)
331 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
332 16, 1, buffer_info->skb->data,
333 buffer_info->skb->len, true);
336 /* Print Rx Ring Summary */
338 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
339 pr_info("Queue [NTU] [NTC]\n");
340 pr_info(" %5d %5X %5X\n",
341 0, rx_ring->next_to_use, rx_ring->next_to_clean);
344 if (!netif_msg_rx_status(adapter))
347 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
348 switch (adapter->rx_ps_pages) {
352 /* [Extended] Packet Split Receive Descriptor Format
354 * +-----------------------------------------------------+
355 * 0 | Buffer Address 0 [63:0] |
356 * +-----------------------------------------------------+
357 * 8 | Buffer Address 1 [63:0] |
358 * +-----------------------------------------------------+
359 * 16 | Buffer Address 2 [63:0] |
360 * +-----------------------------------------------------+
361 * 24 | Buffer Address 3 [63:0] |
362 * +-----------------------------------------------------+
364 pr_info("R [desc] [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma ] [bi->skb] <-- Ext Pkt Split format\n");
365 /* [Extended] Receive Descriptor (Write-Back) Format
367 * 63 48 47 32 31 13 12 8 7 4 3 0
368 * +------------------------------------------------------+
369 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
370 * | Checksum | Ident | | Queue | | Type |
371 * +------------------------------------------------------+
372 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
373 * +------------------------------------------------------+
374 * 63 48 47 32 31 20 19 0
376 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
377 for (i = 0; i < rx_ring->count; i++) {
378 const char *next_desc;
379 buffer_info = &rx_ring->buffer_info[i];
380 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
381 u1 = (struct my_u1 *)rx_desc_ps;
383 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
385 if (i == rx_ring->next_to_use)
387 else if (i == rx_ring->next_to_clean)
392 if (staterr & E1000_RXD_STAT_DD) {
393 /* Descriptor Done */
394 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
396 (unsigned long long)le64_to_cpu(u1->a),
397 (unsigned long long)le64_to_cpu(u1->b),
398 (unsigned long long)le64_to_cpu(u1->c),
399 (unsigned long long)le64_to_cpu(u1->d),
400 buffer_info->skb, next_desc);
402 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
404 (unsigned long long)le64_to_cpu(u1->a),
405 (unsigned long long)le64_to_cpu(u1->b),
406 (unsigned long long)le64_to_cpu(u1->c),
407 (unsigned long long)le64_to_cpu(u1->d),
408 (unsigned long long)buffer_info->dma,
409 buffer_info->skb, next_desc);
411 if (netif_msg_pktdata(adapter))
412 e1000e_dump_ps_pages(adapter,
419 /* Extended Receive Descriptor (Read) Format
421 * +-----------------------------------------------------+
422 * 0 | Buffer Address [63:0] |
423 * +-----------------------------------------------------+
425 * +-----------------------------------------------------+
427 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
428 /* Extended Receive Descriptor (Write-Back) Format
430 * 63 48 47 32 31 24 23 4 3 0
431 * +------------------------------------------------------+
433 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
434 * | Packet | IP | | | Type |
435 * | Checksum | Ident | | | |
436 * +------------------------------------------------------+
437 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
438 * +------------------------------------------------------+
439 * 63 48 47 32 31 20 19 0
441 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
443 for (i = 0; i < rx_ring->count; i++) {
444 const char *next_desc;
446 buffer_info = &rx_ring->buffer_info[i];
447 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
448 u1 = (struct my_u1 *)rx_desc;
449 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
451 if (i == rx_ring->next_to_use)
453 else if (i == rx_ring->next_to_clean)
458 if (staterr & E1000_RXD_STAT_DD) {
459 /* Descriptor Done */
460 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
462 (unsigned long long)le64_to_cpu(u1->a),
463 (unsigned long long)le64_to_cpu(u1->b),
464 buffer_info->skb, next_desc);
466 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
468 (unsigned long long)le64_to_cpu(u1->a),
469 (unsigned long long)le64_to_cpu(u1->b),
470 (unsigned long long)buffer_info->dma,
471 buffer_info->skb, next_desc);
473 if (netif_msg_pktdata(adapter) &&
475 print_hex_dump(KERN_INFO, "",
476 DUMP_PREFIX_ADDRESS, 16,
478 buffer_info->skb->data,
479 adapter->rx_buffer_len,
487 * e1000_desc_unused - calculate if we have unused descriptors
489 static int e1000_desc_unused(struct e1000_ring *ring)
491 if (ring->next_to_clean > ring->next_to_use)
492 return ring->next_to_clean - ring->next_to_use - 1;
494 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
498 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
499 * @adapter: board private structure
500 * @hwtstamps: time stamp structure to update
501 * @systim: unsigned 64bit system time value.
503 * Convert the system time value stored in the RX/TXSTMP registers into a
504 * hwtstamp which can be used by the upper level time stamping functions.
506 * The 'systim_lock' spinlock is used to protect the consistency of the
507 * system time value. This is needed because reading the 64 bit time
508 * value involves reading two 32 bit registers. The first read latches the
511 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
512 struct skb_shared_hwtstamps *hwtstamps,
518 spin_lock_irqsave(&adapter->systim_lock, flags);
519 ns = timecounter_cyc2time(&adapter->tc, systim);
520 spin_unlock_irqrestore(&adapter->systim_lock, flags);
522 memset(hwtstamps, 0, sizeof(*hwtstamps));
523 hwtstamps->hwtstamp = ns_to_ktime(ns);
527 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
528 * @adapter: board private structure
529 * @status: descriptor extended error and status field
530 * @skb: particular skb to include time stamp
532 * If the time stamp is valid, convert it into the timecounter ns value
533 * and store that result into the shhwtstamps structure which is passed
534 * up the network stack.
536 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
539 struct e1000_hw *hw = &adapter->hw;
542 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
543 !(status & E1000_RXDEXT_STATERR_TST) ||
544 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
547 /* The Rx time stamp registers contain the time stamp. No other
548 * received packet will be time stamped until the Rx time stamp
549 * registers are read. Because only one packet can be time stamped
550 * at a time, the register values must belong to this packet and
551 * therefore none of the other additional attributes need to be
554 rxstmp = (u64)er32(RXSTMPL);
555 rxstmp |= (u64)er32(RXSTMPH) << 32;
556 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
558 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
562 * e1000_receive_skb - helper function to handle Rx indications
563 * @adapter: board private structure
564 * @staterr: descriptor extended error and status field as written by hardware
565 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
566 * @skb: pointer to sk_buff to be indicated to stack
568 static void e1000_receive_skb(struct e1000_adapter *adapter,
569 struct net_device *netdev, struct sk_buff *skb,
570 u32 staterr, __le16 vlan)
572 u16 tag = le16_to_cpu(vlan);
574 e1000e_rx_hwtstamp(adapter, staterr, skb);
576 skb->protocol = eth_type_trans(skb, netdev);
578 if (staterr & E1000_RXD_STAT_VP)
579 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
581 napi_gro_receive(&adapter->napi, skb);
585 * e1000_rx_checksum - Receive Checksum Offload
586 * @adapter: board private structure
587 * @status_err: receive descriptor status and error fields
588 * @csum: receive descriptor csum field
589 * @sk_buff: socket buffer with received data
591 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
594 u16 status = (u16)status_err;
595 u8 errors = (u8)(status_err >> 24);
597 skb_checksum_none_assert(skb);
599 /* Rx checksum disabled */
600 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
603 /* Ignore Checksum bit is set */
604 if (status & E1000_RXD_STAT_IXSM)
607 /* TCP/UDP checksum error bit or IP checksum error bit is set */
608 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
609 /* let the stack verify checksum errors */
610 adapter->hw_csum_err++;
614 /* TCP/UDP Checksum has not been calculated */
615 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
618 /* It must be a TCP or UDP packet with a valid checksum */
619 skb->ip_summed = CHECKSUM_UNNECESSARY;
620 adapter->hw_csum_good++;
623 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
625 struct e1000_adapter *adapter = rx_ring->adapter;
626 struct e1000_hw *hw = &adapter->hw;
627 s32 ret_val = __ew32_prepare(hw);
629 writel(i, rx_ring->tail);
631 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
632 u32 rctl = er32(RCTL);
634 ew32(RCTL, rctl & ~E1000_RCTL_EN);
635 e_err("ME firmware caused invalid RDT - resetting\n");
636 schedule_work(&adapter->reset_task);
640 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
642 struct e1000_adapter *adapter = tx_ring->adapter;
643 struct e1000_hw *hw = &adapter->hw;
644 s32 ret_val = __ew32_prepare(hw);
646 writel(i, tx_ring->tail);
648 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
649 u32 tctl = er32(TCTL);
651 ew32(TCTL, tctl & ~E1000_TCTL_EN);
652 e_err("ME firmware caused invalid TDT - resetting\n");
653 schedule_work(&adapter->reset_task);
658 * e1000_alloc_rx_buffers - Replace used receive buffers
659 * @rx_ring: Rx descriptor ring
661 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
662 int cleaned_count, gfp_t gfp)
664 struct e1000_adapter *adapter = rx_ring->adapter;
665 struct net_device *netdev = adapter->netdev;
666 struct pci_dev *pdev = adapter->pdev;
667 union e1000_rx_desc_extended *rx_desc;
668 struct e1000_buffer *buffer_info;
671 unsigned int bufsz = adapter->rx_buffer_len;
673 i = rx_ring->next_to_use;
674 buffer_info = &rx_ring->buffer_info[i];
676 while (cleaned_count--) {
677 skb = buffer_info->skb;
683 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
685 /* Better luck next round */
686 adapter->alloc_rx_buff_failed++;
690 buffer_info->skb = skb;
692 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
693 adapter->rx_buffer_len,
695 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
696 dev_err(&pdev->dev, "Rx DMA map failed\n");
697 adapter->rx_dma_failed++;
701 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
702 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
704 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
705 /* Force memory writes to complete before letting h/w
706 * know there are new descriptors to fetch. (Only
707 * applicable for weak-ordered memory model archs,
711 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
712 e1000e_update_rdt_wa(rx_ring, i);
714 writel(i, rx_ring->tail);
717 if (i == rx_ring->count)
719 buffer_info = &rx_ring->buffer_info[i];
722 rx_ring->next_to_use = i;
726 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
727 * @rx_ring: Rx descriptor ring
729 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
730 int cleaned_count, gfp_t gfp)
732 struct e1000_adapter *adapter = rx_ring->adapter;
733 struct net_device *netdev = adapter->netdev;
734 struct pci_dev *pdev = adapter->pdev;
735 union e1000_rx_desc_packet_split *rx_desc;
736 struct e1000_buffer *buffer_info;
737 struct e1000_ps_page *ps_page;
741 i = rx_ring->next_to_use;
742 buffer_info = &rx_ring->buffer_info[i];
744 while (cleaned_count--) {
745 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
747 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
748 ps_page = &buffer_info->ps_pages[j];
749 if (j >= adapter->rx_ps_pages) {
750 /* all unused desc entries get hw null ptr */
751 rx_desc->read.buffer_addr[j + 1] =
755 if (!ps_page->page) {
756 ps_page->page = alloc_page(gfp);
757 if (!ps_page->page) {
758 adapter->alloc_rx_buff_failed++;
761 ps_page->dma = dma_map_page(&pdev->dev,
765 if (dma_mapping_error(&pdev->dev,
767 dev_err(&adapter->pdev->dev,
768 "Rx DMA page map failed\n");
769 adapter->rx_dma_failed++;
773 /* Refresh the desc even if buffer_addrs
774 * didn't change because each write-back
777 rx_desc->read.buffer_addr[j + 1] =
778 cpu_to_le64(ps_page->dma);
781 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
785 adapter->alloc_rx_buff_failed++;
789 buffer_info->skb = skb;
790 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
791 adapter->rx_ps_bsize0,
793 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
794 dev_err(&pdev->dev, "Rx DMA map failed\n");
795 adapter->rx_dma_failed++;
797 dev_kfree_skb_any(skb);
798 buffer_info->skb = NULL;
802 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
804 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
805 /* Force memory writes to complete before letting h/w
806 * know there are new descriptors to fetch. (Only
807 * applicable for weak-ordered memory model archs,
811 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
812 e1000e_update_rdt_wa(rx_ring, i << 1);
814 writel(i << 1, rx_ring->tail);
818 if (i == rx_ring->count)
820 buffer_info = &rx_ring->buffer_info[i];
824 rx_ring->next_to_use = i;
828 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
829 * @rx_ring: Rx descriptor ring
830 * @cleaned_count: number of buffers to allocate this pass
833 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
834 int cleaned_count, gfp_t gfp)
836 struct e1000_adapter *adapter = rx_ring->adapter;
837 struct net_device *netdev = adapter->netdev;
838 struct pci_dev *pdev = adapter->pdev;
839 union e1000_rx_desc_extended *rx_desc;
840 struct e1000_buffer *buffer_info;
843 unsigned int bufsz = 256 - 16; /* for skb_reserve */
845 i = rx_ring->next_to_use;
846 buffer_info = &rx_ring->buffer_info[i];
848 while (cleaned_count--) {
849 skb = buffer_info->skb;
855 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
856 if (unlikely(!skb)) {
857 /* Better luck next round */
858 adapter->alloc_rx_buff_failed++;
862 buffer_info->skb = skb;
864 /* allocate a new page if necessary */
865 if (!buffer_info->page) {
866 buffer_info->page = alloc_page(gfp);
867 if (unlikely(!buffer_info->page)) {
868 adapter->alloc_rx_buff_failed++;
873 if (!buffer_info->dma) {
874 buffer_info->dma = dma_map_page(&pdev->dev,
875 buffer_info->page, 0,
878 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
879 adapter->alloc_rx_buff_failed++;
884 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
885 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
887 if (unlikely(++i == rx_ring->count))
889 buffer_info = &rx_ring->buffer_info[i];
892 if (likely(rx_ring->next_to_use != i)) {
893 rx_ring->next_to_use = i;
894 if (unlikely(i-- == 0))
895 i = (rx_ring->count - 1);
897 /* Force memory writes to complete before letting h/w
898 * know there are new descriptors to fetch. (Only
899 * applicable for weak-ordered memory model archs,
903 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
904 e1000e_update_rdt_wa(rx_ring, i);
906 writel(i, rx_ring->tail);
910 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
913 if (netdev->features & NETIF_F_RXHASH)
914 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
918 * e1000_clean_rx_irq - Send received data up the network stack
919 * @rx_ring: Rx descriptor ring
921 * the return value indicates whether actual cleaning was done, there
922 * is no guarantee that everything was cleaned
924 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
927 struct e1000_adapter *adapter = rx_ring->adapter;
928 struct net_device *netdev = adapter->netdev;
929 struct pci_dev *pdev = adapter->pdev;
930 struct e1000_hw *hw = &adapter->hw;
931 union e1000_rx_desc_extended *rx_desc, *next_rxd;
932 struct e1000_buffer *buffer_info, *next_buffer;
935 int cleaned_count = 0;
936 bool cleaned = false;
937 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
939 i = rx_ring->next_to_clean;
940 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
941 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
942 buffer_info = &rx_ring->buffer_info[i];
944 while (staterr & E1000_RXD_STAT_DD) {
947 if (*work_done >= work_to_do)
950 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
952 skb = buffer_info->skb;
953 buffer_info->skb = NULL;
955 prefetch(skb->data - NET_IP_ALIGN);
958 if (i == rx_ring->count)
960 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
963 next_buffer = &rx_ring->buffer_info[i];
967 dma_unmap_single(&pdev->dev, buffer_info->dma,
968 adapter->rx_buffer_len, DMA_FROM_DEVICE);
969 buffer_info->dma = 0;
971 length = le16_to_cpu(rx_desc->wb.upper.length);
973 /* !EOP means multiple descriptors were used to store a single
974 * packet, if that's the case we need to toss it. In fact, we
975 * need to toss every packet with the EOP bit clear and the
976 * next frame that _does_ have the EOP bit set, as it is by
977 * definition only a frame fragment
979 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
980 adapter->flags2 |= FLAG2_IS_DISCARDING;
982 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
983 /* All receives must fit into a single buffer */
984 e_dbg("Receive packet consumed multiple buffers\n");
986 buffer_info->skb = skb;
987 if (staterr & E1000_RXD_STAT_EOP)
988 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
992 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
993 !(netdev->features & NETIF_F_RXALL))) {
995 buffer_info->skb = skb;
999 /* adjust length to remove Ethernet CRC */
1000 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1001 /* If configured to store CRC, don't subtract FCS,
1002 * but keep the FCS bytes out of the total_rx_bytes
1005 if (netdev->features & NETIF_F_RXFCS)
1006 total_rx_bytes -= 4;
1011 total_rx_bytes += length;
1014 /* code added for copybreak, this should improve
1015 * performance for small packets with large amounts
1016 * of reassembly being done in the stack
1018 if (length < copybreak) {
1019 struct sk_buff *new_skb =
1020 napi_alloc_skb(&adapter->napi, length);
1022 skb_copy_to_linear_data_offset(new_skb,
1028 /* save the skb in buffer_info as good */
1029 buffer_info->skb = skb;
1032 /* else just continue with the old one */
1034 /* end copybreak code */
1035 skb_put(skb, length);
1037 /* Receive Checksum Offload */
1038 e1000_rx_checksum(adapter, staterr, skb);
1040 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1042 e1000_receive_skb(adapter, netdev, skb, staterr,
1043 rx_desc->wb.upper.vlan);
1046 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1048 /* return some buffers to hardware, one at a time is too slow */
1049 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1050 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1055 /* use prefetched values */
1057 buffer_info = next_buffer;
1059 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1061 rx_ring->next_to_clean = i;
1063 cleaned_count = e1000_desc_unused(rx_ring);
1065 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1067 adapter->total_rx_bytes += total_rx_bytes;
1068 adapter->total_rx_packets += total_rx_packets;
1072 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1073 struct e1000_buffer *buffer_info)
1075 struct e1000_adapter *adapter = tx_ring->adapter;
1077 if (buffer_info->dma) {
1078 if (buffer_info->mapped_as_page)
1079 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1080 buffer_info->length, DMA_TO_DEVICE);
1082 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1083 buffer_info->length, DMA_TO_DEVICE);
1084 buffer_info->dma = 0;
1086 if (buffer_info->skb) {
1087 dev_kfree_skb_any(buffer_info->skb);
1088 buffer_info->skb = NULL;
1090 buffer_info->time_stamp = 0;
1093 static void e1000_print_hw_hang(struct work_struct *work)
1095 struct e1000_adapter *adapter = container_of(work,
1096 struct e1000_adapter,
1098 struct net_device *netdev = adapter->netdev;
1099 struct e1000_ring *tx_ring = adapter->tx_ring;
1100 unsigned int i = tx_ring->next_to_clean;
1101 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1102 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103 struct e1000_hw *hw = &adapter->hw;
1104 u16 phy_status, phy_1000t_status, phy_ext_status;
1107 if (test_bit(__E1000_DOWN, &adapter->state))
1110 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1111 /* May be block on write-back, flush and detect again
1112 * flush pending descriptor writebacks to memory
1114 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1115 /* execute the writes immediately */
1117 /* Due to rare timing issues, write to TIDV again to ensure
1118 * the write is successful
1120 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1121 /* execute the writes immediately */
1123 adapter->tx_hang_recheck = true;
1126 adapter->tx_hang_recheck = false;
1128 if (er32(TDH(0)) == er32(TDT(0))) {
1129 e_dbg("false hang detected, ignoring\n");
1133 /* Real hang detected */
1134 netif_stop_queue(netdev);
1136 e1e_rphy(hw, MII_BMSR, &phy_status);
1137 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1138 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1140 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1142 /* detected Hardware unit hang */
1143 e_err("Detected Hardware Unit Hang:\n"
1146 " next_to_use <%x>\n"
1147 " next_to_clean <%x>\n"
1148 "buffer_info[next_to_clean]:\n"
1149 " time_stamp <%lx>\n"
1150 " next_to_watch <%x>\n"
1152 " next_to_watch.status <%x>\n"
1155 "PHY 1000BASE-T Status <%x>\n"
1156 "PHY Extended Status <%x>\n"
1157 "PCI Status <%x>\n",
1158 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1159 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1160 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1161 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1163 e1000e_dump(adapter);
1165 /* Suggest workaround for known h/w issue */
1166 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1167 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1171 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1172 * @work: pointer to work struct
1174 * This work function polls the TSYNCTXCTL valid bit to determine when a
1175 * timestamp has been taken for the current stored skb. The timestamp must
1176 * be for this skb because only one such packet is allowed in the queue.
1178 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1180 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1182 struct e1000_hw *hw = &adapter->hw;
1184 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1185 struct skb_shared_hwtstamps shhwtstamps;
1188 txstmp = er32(TXSTMPL);
1189 txstmp |= (u64)er32(TXSTMPH) << 32;
1191 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1193 skb_tstamp_tx(adapter->tx_hwtstamp_skb, &shhwtstamps);
1194 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1195 adapter->tx_hwtstamp_skb = NULL;
1196 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1197 + adapter->tx_timeout_factor * HZ)) {
1198 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1199 adapter->tx_hwtstamp_skb = NULL;
1200 adapter->tx_hwtstamp_timeouts++;
1201 e_warn("clearing Tx timestamp hang\n");
1203 /* reschedule to check later */
1204 schedule_work(&adapter->tx_hwtstamp_work);
1209 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1210 * @tx_ring: Tx descriptor ring
1212 * the return value indicates whether actual cleaning was done, there
1213 * is no guarantee that everything was cleaned
1215 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1217 struct e1000_adapter *adapter = tx_ring->adapter;
1218 struct net_device *netdev = adapter->netdev;
1219 struct e1000_hw *hw = &adapter->hw;
1220 struct e1000_tx_desc *tx_desc, *eop_desc;
1221 struct e1000_buffer *buffer_info;
1222 unsigned int i, eop;
1223 unsigned int count = 0;
1224 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1225 unsigned int bytes_compl = 0, pkts_compl = 0;
1227 i = tx_ring->next_to_clean;
1228 eop = tx_ring->buffer_info[i].next_to_watch;
1229 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1231 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1232 (count < tx_ring->count)) {
1233 bool cleaned = false;
1235 dma_rmb(); /* read buffer_info after eop_desc */
1236 for (; !cleaned; count++) {
1237 tx_desc = E1000_TX_DESC(*tx_ring, i);
1238 buffer_info = &tx_ring->buffer_info[i];
1239 cleaned = (i == eop);
1242 total_tx_packets += buffer_info->segs;
1243 total_tx_bytes += buffer_info->bytecount;
1244 if (buffer_info->skb) {
1245 bytes_compl += buffer_info->skb->len;
1250 e1000_put_txbuf(tx_ring, buffer_info);
1251 tx_desc->upper.data = 0;
1254 if (i == tx_ring->count)
1258 if (i == tx_ring->next_to_use)
1260 eop = tx_ring->buffer_info[i].next_to_watch;
1261 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1264 tx_ring->next_to_clean = i;
1266 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1268 #define TX_WAKE_THRESHOLD 32
1269 if (count && netif_carrier_ok(netdev) &&
1270 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1271 /* Make sure that anybody stopping the queue after this
1272 * sees the new next_to_clean.
1276 if (netif_queue_stopped(netdev) &&
1277 !(test_bit(__E1000_DOWN, &adapter->state))) {
1278 netif_wake_queue(netdev);
1279 ++adapter->restart_queue;
1283 if (adapter->detect_tx_hung) {
1284 /* Detect a transmit hang in hardware, this serializes the
1285 * check with the clearing of time_stamp and movement of i
1287 adapter->detect_tx_hung = false;
1288 if (tx_ring->buffer_info[i].time_stamp &&
1289 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1290 + (adapter->tx_timeout_factor * HZ)) &&
1291 !(er32(STATUS) & E1000_STATUS_TXOFF))
1292 schedule_work(&adapter->print_hang_task);
1294 adapter->tx_hang_recheck = false;
1296 adapter->total_tx_bytes += total_tx_bytes;
1297 adapter->total_tx_packets += total_tx_packets;
1298 return count < tx_ring->count;
1302 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1303 * @rx_ring: Rx descriptor ring
1305 * the return value indicates whether actual cleaning was done, there
1306 * is no guarantee that everything was cleaned
1308 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1311 struct e1000_adapter *adapter = rx_ring->adapter;
1312 struct e1000_hw *hw = &adapter->hw;
1313 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1314 struct net_device *netdev = adapter->netdev;
1315 struct pci_dev *pdev = adapter->pdev;
1316 struct e1000_buffer *buffer_info, *next_buffer;
1317 struct e1000_ps_page *ps_page;
1318 struct sk_buff *skb;
1320 u32 length, staterr;
1321 int cleaned_count = 0;
1322 bool cleaned = false;
1323 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1325 i = rx_ring->next_to_clean;
1326 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1327 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1328 buffer_info = &rx_ring->buffer_info[i];
1330 while (staterr & E1000_RXD_STAT_DD) {
1331 if (*work_done >= work_to_do)
1334 skb = buffer_info->skb;
1335 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1337 /* in the packet split case this is header only */
1338 prefetch(skb->data - NET_IP_ALIGN);
1341 if (i == rx_ring->count)
1343 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1346 next_buffer = &rx_ring->buffer_info[i];
1350 dma_unmap_single(&pdev->dev, buffer_info->dma,
1351 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1352 buffer_info->dma = 0;
1354 /* see !EOP comment in other Rx routine */
1355 if (!(staterr & E1000_RXD_STAT_EOP))
1356 adapter->flags2 |= FLAG2_IS_DISCARDING;
1358 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1359 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1360 dev_kfree_skb_irq(skb);
1361 if (staterr & E1000_RXD_STAT_EOP)
1362 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1366 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1367 !(netdev->features & NETIF_F_RXALL))) {
1368 dev_kfree_skb_irq(skb);
1372 length = le16_to_cpu(rx_desc->wb.middle.length0);
1375 e_dbg("Last part of the packet spanning multiple descriptors\n");
1376 dev_kfree_skb_irq(skb);
1381 skb_put(skb, length);
1384 /* this looks ugly, but it seems compiler issues make
1385 * it more efficient than reusing j
1387 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1389 /* page alloc/put takes too long and effects small
1390 * packet throughput, so unsplit small packets and
1391 * save the alloc/put only valid in softirq (napi)
1392 * context to call kmap_*
1394 if (l1 && (l1 <= copybreak) &&
1395 ((length + l1) <= adapter->rx_ps_bsize0)) {
1398 ps_page = &buffer_info->ps_pages[0];
1400 /* there is no documentation about how to call
1401 * kmap_atomic, so we can't hold the mapping
1404 dma_sync_single_for_cpu(&pdev->dev,
1408 vaddr = kmap_atomic(ps_page->page);
1409 memcpy(skb_tail_pointer(skb), vaddr, l1);
1410 kunmap_atomic(vaddr);
1411 dma_sync_single_for_device(&pdev->dev,
1416 /* remove the CRC */
1417 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1418 if (!(netdev->features & NETIF_F_RXFCS))
1427 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1428 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1432 ps_page = &buffer_info->ps_pages[j];
1433 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1436 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1437 ps_page->page = NULL;
1439 skb->data_len += length;
1440 skb->truesize += PAGE_SIZE;
1443 /* strip the ethernet crc, problem is we're using pages now so
1444 * this whole operation can get a little cpu intensive
1446 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1447 if (!(netdev->features & NETIF_F_RXFCS))
1448 pskb_trim(skb, skb->len - 4);
1452 total_rx_bytes += skb->len;
1455 e1000_rx_checksum(adapter, staterr, skb);
1457 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1459 if (rx_desc->wb.upper.header_status &
1460 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1461 adapter->rx_hdr_split++;
1463 e1000_receive_skb(adapter, netdev, skb, staterr,
1464 rx_desc->wb.middle.vlan);
1467 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1468 buffer_info->skb = NULL;
1470 /* return some buffers to hardware, one at a time is too slow */
1471 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1472 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1477 /* use prefetched values */
1479 buffer_info = next_buffer;
1481 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1483 rx_ring->next_to_clean = i;
1485 cleaned_count = e1000_desc_unused(rx_ring);
1487 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1489 adapter->total_rx_bytes += total_rx_bytes;
1490 adapter->total_rx_packets += total_rx_packets;
1495 * e1000_consume_page - helper function
1497 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1502 skb->data_len += length;
1503 skb->truesize += PAGE_SIZE;
1507 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1508 * @adapter: board private structure
1510 * the return value indicates whether actual cleaning was done, there
1511 * is no guarantee that everything was cleaned
1513 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1516 struct e1000_adapter *adapter = rx_ring->adapter;
1517 struct net_device *netdev = adapter->netdev;
1518 struct pci_dev *pdev = adapter->pdev;
1519 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1520 struct e1000_buffer *buffer_info, *next_buffer;
1521 u32 length, staterr;
1523 int cleaned_count = 0;
1524 bool cleaned = false;
1525 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1526 struct skb_shared_info *shinfo;
1528 i = rx_ring->next_to_clean;
1529 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1530 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1531 buffer_info = &rx_ring->buffer_info[i];
1533 while (staterr & E1000_RXD_STAT_DD) {
1534 struct sk_buff *skb;
1536 if (*work_done >= work_to_do)
1539 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1541 skb = buffer_info->skb;
1542 buffer_info->skb = NULL;
1545 if (i == rx_ring->count)
1547 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1550 next_buffer = &rx_ring->buffer_info[i];
1554 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1556 buffer_info->dma = 0;
1558 length = le16_to_cpu(rx_desc->wb.upper.length);
1560 /* errors is only valid for DD + EOP descriptors */
1561 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1562 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1563 !(netdev->features & NETIF_F_RXALL)))) {
1564 /* recycle both page and skb */
1565 buffer_info->skb = skb;
1566 /* an error means any chain goes out the window too */
1567 if (rx_ring->rx_skb_top)
1568 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1569 rx_ring->rx_skb_top = NULL;
1572 #define rxtop (rx_ring->rx_skb_top)
1573 if (!(staterr & E1000_RXD_STAT_EOP)) {
1574 /* this descriptor is only the beginning (or middle) */
1576 /* this is the beginning of a chain */
1578 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1581 /* this is the middle of a chain */
1582 shinfo = skb_shinfo(rxtop);
1583 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1584 buffer_info->page, 0,
1586 /* re-use the skb, only consumed the page */
1587 buffer_info->skb = skb;
1589 e1000_consume_page(buffer_info, rxtop, length);
1593 /* end of the chain */
1594 shinfo = skb_shinfo(rxtop);
1595 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1596 buffer_info->page, 0,
1598 /* re-use the current skb, we only consumed the
1601 buffer_info->skb = skb;
1604 e1000_consume_page(buffer_info, skb, length);
1606 /* no chain, got EOP, this buf is the packet
1607 * copybreak to save the put_page/alloc_page
1609 if (length <= copybreak &&
1610 skb_tailroom(skb) >= length) {
1612 vaddr = kmap_atomic(buffer_info->page);
1613 memcpy(skb_tail_pointer(skb), vaddr,
1615 kunmap_atomic(vaddr);
1616 /* re-use the page, so don't erase
1619 skb_put(skb, length);
1621 skb_fill_page_desc(skb, 0,
1622 buffer_info->page, 0,
1624 e1000_consume_page(buffer_info, skb,
1630 /* Receive Checksum Offload */
1631 e1000_rx_checksum(adapter, staterr, skb);
1633 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1635 /* probably a little skewed due to removing CRC */
1636 total_rx_bytes += skb->len;
1639 /* eth type trans needs skb->data to point to something */
1640 if (!pskb_may_pull(skb, ETH_HLEN)) {
1641 e_err("pskb_may_pull failed.\n");
1642 dev_kfree_skb_irq(skb);
1646 e1000_receive_skb(adapter, netdev, skb, staterr,
1647 rx_desc->wb.upper.vlan);
1650 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1652 /* return some buffers to hardware, one at a time is too slow */
1653 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1654 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1659 /* use prefetched values */
1661 buffer_info = next_buffer;
1663 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1665 rx_ring->next_to_clean = i;
1667 cleaned_count = e1000_desc_unused(rx_ring);
1669 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1671 adapter->total_rx_bytes += total_rx_bytes;
1672 adapter->total_rx_packets += total_rx_packets;
1677 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1678 * @rx_ring: Rx descriptor ring
1680 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1682 struct e1000_adapter *adapter = rx_ring->adapter;
1683 struct e1000_buffer *buffer_info;
1684 struct e1000_ps_page *ps_page;
1685 struct pci_dev *pdev = adapter->pdev;
1688 /* Free all the Rx ring sk_buffs */
1689 for (i = 0; i < rx_ring->count; i++) {
1690 buffer_info = &rx_ring->buffer_info[i];
1691 if (buffer_info->dma) {
1692 if (adapter->clean_rx == e1000_clean_rx_irq)
1693 dma_unmap_single(&pdev->dev, buffer_info->dma,
1694 adapter->rx_buffer_len,
1696 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1697 dma_unmap_page(&pdev->dev, buffer_info->dma,
1698 PAGE_SIZE, DMA_FROM_DEVICE);
1699 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1700 dma_unmap_single(&pdev->dev, buffer_info->dma,
1701 adapter->rx_ps_bsize0,
1703 buffer_info->dma = 0;
1706 if (buffer_info->page) {
1707 put_page(buffer_info->page);
1708 buffer_info->page = NULL;
1711 if (buffer_info->skb) {
1712 dev_kfree_skb(buffer_info->skb);
1713 buffer_info->skb = NULL;
1716 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1717 ps_page = &buffer_info->ps_pages[j];
1720 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1723 put_page(ps_page->page);
1724 ps_page->page = NULL;
1728 /* there also may be some cached data from a chained receive */
1729 if (rx_ring->rx_skb_top) {
1730 dev_kfree_skb(rx_ring->rx_skb_top);
1731 rx_ring->rx_skb_top = NULL;
1734 /* Zero out the descriptor ring */
1735 memset(rx_ring->desc, 0, rx_ring->size);
1737 rx_ring->next_to_clean = 0;
1738 rx_ring->next_to_use = 0;
1739 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1742 static void e1000e_downshift_workaround(struct work_struct *work)
1744 struct e1000_adapter *adapter = container_of(work,
1745 struct e1000_adapter,
1748 if (test_bit(__E1000_DOWN, &adapter->state))
1751 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1755 * e1000_intr_msi - Interrupt Handler
1756 * @irq: interrupt number
1757 * @data: pointer to a network interface device structure
1759 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1761 struct net_device *netdev = data;
1762 struct e1000_adapter *adapter = netdev_priv(netdev);
1763 struct e1000_hw *hw = &adapter->hw;
1764 u32 icr = er32(ICR);
1766 /* read ICR disables interrupts using IAM */
1767 if (icr & E1000_ICR_LSC) {
1768 hw->mac.get_link_status = true;
1769 /* ICH8 workaround-- Call gig speed drop workaround on cable
1770 * disconnect (LSC) before accessing any PHY registers
1772 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1773 (!(er32(STATUS) & E1000_STATUS_LU)))
1774 schedule_work(&adapter->downshift_task);
1776 /* 80003ES2LAN workaround-- For packet buffer work-around on
1777 * link down event; disable receives here in the ISR and reset
1778 * adapter in watchdog
1780 if (netif_carrier_ok(netdev) &&
1781 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1782 /* disable receives */
1783 u32 rctl = er32(RCTL);
1785 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1786 adapter->flags |= FLAG_RESTART_NOW;
1788 /* guard against interrupt when we're going down */
1789 if (!test_bit(__E1000_DOWN, &adapter->state))
1790 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1793 /* Reset on uncorrectable ECC error */
1794 if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1795 (hw->mac.type == e1000_pch_spt))) {
1796 u32 pbeccsts = er32(PBECCSTS);
1798 adapter->corr_errors +=
1799 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1800 adapter->uncorr_errors +=
1801 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1802 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1804 /* Do the reset outside of interrupt context */
1805 schedule_work(&adapter->reset_task);
1807 /* return immediately since reset is imminent */
1811 if (napi_schedule_prep(&adapter->napi)) {
1812 adapter->total_tx_bytes = 0;
1813 adapter->total_tx_packets = 0;
1814 adapter->total_rx_bytes = 0;
1815 adapter->total_rx_packets = 0;
1816 __napi_schedule(&adapter->napi);
1823 * e1000_intr - Interrupt Handler
1824 * @irq: interrupt number
1825 * @data: pointer to a network interface device structure
1827 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1829 struct net_device *netdev = data;
1830 struct e1000_adapter *adapter = netdev_priv(netdev);
1831 struct e1000_hw *hw = &adapter->hw;
1832 u32 rctl, icr = er32(ICR);
1834 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1835 return IRQ_NONE; /* Not our interrupt */
1837 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1838 * not set, then the adapter didn't send an interrupt
1840 if (!(icr & E1000_ICR_INT_ASSERTED))
1843 /* Interrupt Auto-Mask...upon reading ICR,
1844 * interrupts are masked. No need for the
1848 if (icr & E1000_ICR_LSC) {
1849 hw->mac.get_link_status = true;
1850 /* ICH8 workaround-- Call gig speed drop workaround on cable
1851 * disconnect (LSC) before accessing any PHY registers
1853 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1854 (!(er32(STATUS) & E1000_STATUS_LU)))
1855 schedule_work(&adapter->downshift_task);
1857 /* 80003ES2LAN workaround--
1858 * For packet buffer work-around on link down event;
1859 * disable receives here in the ISR and
1860 * reset adapter in watchdog
1862 if (netif_carrier_ok(netdev) &&
1863 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1864 /* disable receives */
1866 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1867 adapter->flags |= FLAG_RESTART_NOW;
1869 /* guard against interrupt when we're going down */
1870 if (!test_bit(__E1000_DOWN, &adapter->state))
1871 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1874 /* Reset on uncorrectable ECC error */
1875 if ((icr & E1000_ICR_ECCER) && ((hw->mac.type == e1000_pch_lpt) ||
1876 (hw->mac.type == e1000_pch_spt))) {
1877 u32 pbeccsts = er32(PBECCSTS);
1879 adapter->corr_errors +=
1880 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1881 adapter->uncorr_errors +=
1882 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1883 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1885 /* Do the reset outside of interrupt context */
1886 schedule_work(&adapter->reset_task);
1888 /* return immediately since reset is imminent */
1892 if (napi_schedule_prep(&adapter->napi)) {
1893 adapter->total_tx_bytes = 0;
1894 adapter->total_tx_packets = 0;
1895 adapter->total_rx_bytes = 0;
1896 adapter->total_rx_packets = 0;
1897 __napi_schedule(&adapter->napi);
1903 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1905 struct net_device *netdev = data;
1906 struct e1000_adapter *adapter = netdev_priv(netdev);
1907 struct e1000_hw *hw = &adapter->hw;
1909 hw->mac.get_link_status = true;
1911 /* guard against interrupt when we're going down */
1912 if (!test_bit(__E1000_DOWN, &adapter->state)) {
1913 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1914 ew32(IMS, E1000_IMS_OTHER);
1920 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1922 struct net_device *netdev = data;
1923 struct e1000_adapter *adapter = netdev_priv(netdev);
1924 struct e1000_hw *hw = &adapter->hw;
1925 struct e1000_ring *tx_ring = adapter->tx_ring;
1927 adapter->total_tx_bytes = 0;
1928 adapter->total_tx_packets = 0;
1930 if (!e1000_clean_tx_irq(tx_ring))
1931 /* Ring was not completely cleaned, so fire another interrupt */
1932 ew32(ICS, tx_ring->ims_val);
1934 if (!test_bit(__E1000_DOWN, &adapter->state))
1935 ew32(IMS, adapter->tx_ring->ims_val);
1940 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1942 struct net_device *netdev = data;
1943 struct e1000_adapter *adapter = netdev_priv(netdev);
1944 struct e1000_ring *rx_ring = adapter->rx_ring;
1946 /* Write the ITR value calculated at the end of the
1947 * previous interrupt.
1949 if (rx_ring->set_itr) {
1950 u32 itr = rx_ring->itr_val ?
1951 1000000000 / (rx_ring->itr_val * 256) : 0;
1953 writel(itr, rx_ring->itr_register);
1954 rx_ring->set_itr = 0;
1957 if (napi_schedule_prep(&adapter->napi)) {
1958 adapter->total_rx_bytes = 0;
1959 adapter->total_rx_packets = 0;
1960 __napi_schedule(&adapter->napi);
1966 * e1000_configure_msix - Configure MSI-X hardware
1968 * e1000_configure_msix sets up the hardware to properly
1969 * generate MSI-X interrupts.
1971 static void e1000_configure_msix(struct e1000_adapter *adapter)
1973 struct e1000_hw *hw = &adapter->hw;
1974 struct e1000_ring *rx_ring = adapter->rx_ring;
1975 struct e1000_ring *tx_ring = adapter->tx_ring;
1977 u32 ctrl_ext, ivar = 0;
1979 adapter->eiac_mask = 0;
1981 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1982 if (hw->mac.type == e1000_82574) {
1983 u32 rfctl = er32(RFCTL);
1985 rfctl |= E1000_RFCTL_ACK_DIS;
1989 /* Configure Rx vector */
1990 rx_ring->ims_val = E1000_IMS_RXQ0;
1991 adapter->eiac_mask |= rx_ring->ims_val;
1992 if (rx_ring->itr_val)
1993 writel(1000000000 / (rx_ring->itr_val * 256),
1994 rx_ring->itr_register);
1996 writel(1, rx_ring->itr_register);
1997 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1999 /* Configure Tx vector */
2000 tx_ring->ims_val = E1000_IMS_TXQ0;
2002 if (tx_ring->itr_val)
2003 writel(1000000000 / (tx_ring->itr_val * 256),
2004 tx_ring->itr_register);
2006 writel(1, tx_ring->itr_register);
2007 adapter->eiac_mask |= tx_ring->ims_val;
2008 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2010 /* set vector for Other Causes, e.g. link changes */
2012 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2013 if (rx_ring->itr_val)
2014 writel(1000000000 / (rx_ring->itr_val * 256),
2015 hw->hw_addr + E1000_EITR_82574(vector));
2017 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2018 adapter->eiac_mask |= E1000_IMS_OTHER;
2020 /* Cause Tx interrupts on every write back */
2025 /* enable MSI-X PBA support */
2026 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2027 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2028 ew32(CTRL_EXT, ctrl_ext);
2032 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2034 if (adapter->msix_entries) {
2035 pci_disable_msix(adapter->pdev);
2036 kfree(adapter->msix_entries);
2037 adapter->msix_entries = NULL;
2038 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2039 pci_disable_msi(adapter->pdev);
2040 adapter->flags &= ~FLAG_MSI_ENABLED;
2045 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2047 * Attempt to configure interrupts using the best available
2048 * capabilities of the hardware and kernel.
2050 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2055 switch (adapter->int_mode) {
2056 case E1000E_INT_MODE_MSIX:
2057 if (adapter->flags & FLAG_HAS_MSIX) {
2058 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2059 adapter->msix_entries = kcalloc(adapter->num_vectors,
2063 if (adapter->msix_entries) {
2064 struct e1000_adapter *a = adapter;
2066 for (i = 0; i < adapter->num_vectors; i++)
2067 adapter->msix_entries[i].entry = i;
2069 err = pci_enable_msix_range(a->pdev,
2076 /* MSI-X failed, so fall through and try MSI */
2077 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2078 e1000e_reset_interrupt_capability(adapter);
2080 adapter->int_mode = E1000E_INT_MODE_MSI;
2082 case E1000E_INT_MODE_MSI:
2083 if (!pci_enable_msi(adapter->pdev)) {
2084 adapter->flags |= FLAG_MSI_ENABLED;
2086 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2087 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2090 case E1000E_INT_MODE_LEGACY:
2091 /* Don't do anything; this is the system default */
2095 /* store the number of vectors being used */
2096 adapter->num_vectors = 1;
2100 * e1000_request_msix - Initialize MSI-X interrupts
2102 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2105 static int e1000_request_msix(struct e1000_adapter *adapter)
2107 struct net_device *netdev = adapter->netdev;
2108 int err = 0, vector = 0;
2110 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2111 snprintf(adapter->rx_ring->name,
2112 sizeof(adapter->rx_ring->name) - 1,
2113 "%s-rx-0", netdev->name);
2115 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2116 err = request_irq(adapter->msix_entries[vector].vector,
2117 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2121 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2122 E1000_EITR_82574(vector);
2123 adapter->rx_ring->itr_val = adapter->itr;
2126 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2127 snprintf(adapter->tx_ring->name,
2128 sizeof(adapter->tx_ring->name) - 1,
2129 "%s-tx-0", netdev->name);
2131 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2132 err = request_irq(adapter->msix_entries[vector].vector,
2133 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2137 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2138 E1000_EITR_82574(vector);
2139 adapter->tx_ring->itr_val = adapter->itr;
2142 err = request_irq(adapter->msix_entries[vector].vector,
2143 e1000_msix_other, 0, netdev->name, netdev);
2147 e1000_configure_msix(adapter);
2153 * e1000_request_irq - initialize interrupts
2155 * Attempts to configure interrupts using the best available
2156 * capabilities of the hardware and kernel.
2158 static int e1000_request_irq(struct e1000_adapter *adapter)
2160 struct net_device *netdev = adapter->netdev;
2163 if (adapter->msix_entries) {
2164 err = e1000_request_msix(adapter);
2167 /* fall back to MSI */
2168 e1000e_reset_interrupt_capability(adapter);
2169 adapter->int_mode = E1000E_INT_MODE_MSI;
2170 e1000e_set_interrupt_capability(adapter);
2172 if (adapter->flags & FLAG_MSI_ENABLED) {
2173 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2174 netdev->name, netdev);
2178 /* fall back to legacy interrupt */
2179 e1000e_reset_interrupt_capability(adapter);
2180 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2183 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2184 netdev->name, netdev);
2186 e_err("Unable to allocate interrupt, Error: %d\n", err);
2191 static void e1000_free_irq(struct e1000_adapter *adapter)
2193 struct net_device *netdev = adapter->netdev;
2195 if (adapter->msix_entries) {
2198 free_irq(adapter->msix_entries[vector].vector, netdev);
2201 free_irq(adapter->msix_entries[vector].vector, netdev);
2204 /* Other Causes interrupt vector */
2205 free_irq(adapter->msix_entries[vector].vector, netdev);
2209 free_irq(adapter->pdev->irq, netdev);
2213 * e1000_irq_disable - Mask off interrupt generation on the NIC
2215 static void e1000_irq_disable(struct e1000_adapter *adapter)
2217 struct e1000_hw *hw = &adapter->hw;
2220 if (adapter->msix_entries)
2221 ew32(EIAC_82574, 0);
2224 if (adapter->msix_entries) {
2227 for (i = 0; i < adapter->num_vectors; i++)
2228 synchronize_irq(adapter->msix_entries[i].vector);
2230 synchronize_irq(adapter->pdev->irq);
2235 * e1000_irq_enable - Enable default interrupt generation settings
2237 static void e1000_irq_enable(struct e1000_adapter *adapter)
2239 struct e1000_hw *hw = &adapter->hw;
2241 if (adapter->msix_entries) {
2242 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2243 ew32(IMS, adapter->eiac_mask | E1000_IMS_LSC);
2244 } else if ((hw->mac.type == e1000_pch_lpt) ||
2245 (hw->mac.type == e1000_pch_spt)) {
2246 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2248 ew32(IMS, IMS_ENABLE_MASK);
2254 * e1000e_get_hw_control - get control of the h/w from f/w
2255 * @adapter: address of board private structure
2257 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2258 * For ASF and Pass Through versions of f/w this means that
2259 * the driver is loaded. For AMT version (only with 82573)
2260 * of the f/w this means that the network i/f is open.
2262 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2264 struct e1000_hw *hw = &adapter->hw;
2268 /* Let firmware know the driver has taken over */
2269 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2271 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2272 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2273 ctrl_ext = er32(CTRL_EXT);
2274 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2279 * e1000e_release_hw_control - release control of the h/w to f/w
2280 * @adapter: address of board private structure
2282 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2283 * For ASF and Pass Through versions of f/w this means that the
2284 * driver is no longer loaded. For AMT version (only with 82573) i
2285 * of the f/w this means that the network i/f is closed.
2288 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2290 struct e1000_hw *hw = &adapter->hw;
2294 /* Let firmware taken over control of h/w */
2295 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2297 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2298 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2299 ctrl_ext = er32(CTRL_EXT);
2300 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2305 * e1000_alloc_ring_dma - allocate memory for a ring structure
2307 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2308 struct e1000_ring *ring)
2310 struct pci_dev *pdev = adapter->pdev;
2312 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2321 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2322 * @tx_ring: Tx descriptor ring
2324 * Return 0 on success, negative on failure
2326 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2328 struct e1000_adapter *adapter = tx_ring->adapter;
2329 int err = -ENOMEM, size;
2331 size = sizeof(struct e1000_buffer) * tx_ring->count;
2332 tx_ring->buffer_info = vzalloc(size);
2333 if (!tx_ring->buffer_info)
2336 /* round up to nearest 4K */
2337 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2338 tx_ring->size = ALIGN(tx_ring->size, 4096);
2340 err = e1000_alloc_ring_dma(adapter, tx_ring);
2344 tx_ring->next_to_use = 0;
2345 tx_ring->next_to_clean = 0;
2349 vfree(tx_ring->buffer_info);
2350 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2355 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2356 * @rx_ring: Rx descriptor ring
2358 * Returns 0 on success, negative on failure
2360 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2362 struct e1000_adapter *adapter = rx_ring->adapter;
2363 struct e1000_buffer *buffer_info;
2364 int i, size, desc_len, err = -ENOMEM;
2366 size = sizeof(struct e1000_buffer) * rx_ring->count;
2367 rx_ring->buffer_info = vzalloc(size);
2368 if (!rx_ring->buffer_info)
2371 for (i = 0; i < rx_ring->count; i++) {
2372 buffer_info = &rx_ring->buffer_info[i];
2373 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2374 sizeof(struct e1000_ps_page),
2376 if (!buffer_info->ps_pages)
2380 desc_len = sizeof(union e1000_rx_desc_packet_split);
2382 /* Round up to nearest 4K */
2383 rx_ring->size = rx_ring->count * desc_len;
2384 rx_ring->size = ALIGN(rx_ring->size, 4096);
2386 err = e1000_alloc_ring_dma(adapter, rx_ring);
2390 rx_ring->next_to_clean = 0;
2391 rx_ring->next_to_use = 0;
2392 rx_ring->rx_skb_top = NULL;
2397 for (i = 0; i < rx_ring->count; i++) {
2398 buffer_info = &rx_ring->buffer_info[i];
2399 kfree(buffer_info->ps_pages);
2402 vfree(rx_ring->buffer_info);
2403 e_err("Unable to allocate memory for the receive descriptor ring\n");
2408 * e1000_clean_tx_ring - Free Tx Buffers
2409 * @tx_ring: Tx descriptor ring
2411 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2413 struct e1000_adapter *adapter = tx_ring->adapter;
2414 struct e1000_buffer *buffer_info;
2418 for (i = 0; i < tx_ring->count; i++) {
2419 buffer_info = &tx_ring->buffer_info[i];
2420 e1000_put_txbuf(tx_ring, buffer_info);
2423 netdev_reset_queue(adapter->netdev);
2424 size = sizeof(struct e1000_buffer) * tx_ring->count;
2425 memset(tx_ring->buffer_info, 0, size);
2427 memset(tx_ring->desc, 0, tx_ring->size);
2429 tx_ring->next_to_use = 0;
2430 tx_ring->next_to_clean = 0;
2434 * e1000e_free_tx_resources - Free Tx Resources per Queue
2435 * @tx_ring: Tx descriptor ring
2437 * Free all transmit software resources
2439 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2441 struct e1000_adapter *adapter = tx_ring->adapter;
2442 struct pci_dev *pdev = adapter->pdev;
2444 e1000_clean_tx_ring(tx_ring);
2446 vfree(tx_ring->buffer_info);
2447 tx_ring->buffer_info = NULL;
2449 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2451 tx_ring->desc = NULL;
2455 * e1000e_free_rx_resources - Free Rx Resources
2456 * @rx_ring: Rx descriptor ring
2458 * Free all receive software resources
2460 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2462 struct e1000_adapter *adapter = rx_ring->adapter;
2463 struct pci_dev *pdev = adapter->pdev;
2466 e1000_clean_rx_ring(rx_ring);
2468 for (i = 0; i < rx_ring->count; i++)
2469 kfree(rx_ring->buffer_info[i].ps_pages);
2471 vfree(rx_ring->buffer_info);
2472 rx_ring->buffer_info = NULL;
2474 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2476 rx_ring->desc = NULL;
2480 * e1000_update_itr - update the dynamic ITR value based on statistics
2481 * @adapter: pointer to adapter
2482 * @itr_setting: current adapter->itr
2483 * @packets: the number of packets during this measurement interval
2484 * @bytes: the number of bytes during this measurement interval
2486 * Stores a new ITR value based on packets and byte
2487 * counts during the last interrupt. The advantage of per interrupt
2488 * computation is faster updates and more accurate ITR for the current
2489 * traffic pattern. Constants in this function were computed
2490 * based on theoretical maximum wire speed and thresholds were set based
2491 * on testing data as well as attempting to minimize response time
2492 * while increasing bulk throughput. This functionality is controlled
2493 * by the InterruptThrottleRate module parameter.
2495 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2497 unsigned int retval = itr_setting;
2502 switch (itr_setting) {
2503 case lowest_latency:
2504 /* handle TSO and jumbo frames */
2505 if (bytes / packets > 8000)
2506 retval = bulk_latency;
2507 else if ((packets < 5) && (bytes > 512))
2508 retval = low_latency;
2510 case low_latency: /* 50 usec aka 20000 ints/s */
2511 if (bytes > 10000) {
2512 /* this if handles the TSO accounting */
2513 if (bytes / packets > 8000)
2514 retval = bulk_latency;
2515 else if ((packets < 10) || ((bytes / packets) > 1200))
2516 retval = bulk_latency;
2517 else if ((packets > 35))
2518 retval = lowest_latency;
2519 } else if (bytes / packets > 2000) {
2520 retval = bulk_latency;
2521 } else if (packets <= 2 && bytes < 512) {
2522 retval = lowest_latency;
2525 case bulk_latency: /* 250 usec aka 4000 ints/s */
2526 if (bytes > 25000) {
2528 retval = low_latency;
2529 } else if (bytes < 6000) {
2530 retval = low_latency;
2538 static void e1000_set_itr(struct e1000_adapter *adapter)
2541 u32 new_itr = adapter->itr;
2543 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2544 if (adapter->link_speed != SPEED_1000) {
2550 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2555 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2556 adapter->total_tx_packets,
2557 adapter->total_tx_bytes);
2558 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2559 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2560 adapter->tx_itr = low_latency;
2562 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2563 adapter->total_rx_packets,
2564 adapter->total_rx_bytes);
2565 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2566 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2567 adapter->rx_itr = low_latency;
2569 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2571 /* counts and packets in update_itr are dependent on these numbers */
2572 switch (current_itr) {
2573 case lowest_latency:
2577 new_itr = 20000; /* aka hwitr = ~200 */
2587 if (new_itr != adapter->itr) {
2588 /* this attempts to bias the interrupt rate towards Bulk
2589 * by adding intermediate steps when interrupt rate is
2592 new_itr = new_itr > adapter->itr ?
2593 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2594 adapter->itr = new_itr;
2595 adapter->rx_ring->itr_val = new_itr;
2596 if (adapter->msix_entries)
2597 adapter->rx_ring->set_itr = 1;
2599 e1000e_write_itr(adapter, new_itr);
2604 * e1000e_write_itr - write the ITR value to the appropriate registers
2605 * @adapter: address of board private structure
2606 * @itr: new ITR value to program
2608 * e1000e_write_itr determines if the adapter is in MSI-X mode
2609 * and, if so, writes the EITR registers with the ITR value.
2610 * Otherwise, it writes the ITR value into the ITR register.
2612 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2614 struct e1000_hw *hw = &adapter->hw;
2615 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2617 if (adapter->msix_entries) {
2620 for (vector = 0; vector < adapter->num_vectors; vector++)
2621 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2628 * e1000_alloc_queues - Allocate memory for all rings
2629 * @adapter: board private structure to initialize
2631 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2633 int size = sizeof(struct e1000_ring);
2635 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2636 if (!adapter->tx_ring)
2638 adapter->tx_ring->count = adapter->tx_ring_count;
2639 adapter->tx_ring->adapter = adapter;
2641 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2642 if (!adapter->rx_ring)
2644 adapter->rx_ring->count = adapter->rx_ring_count;
2645 adapter->rx_ring->adapter = adapter;
2649 e_err("Unable to allocate memory for queues\n");
2650 kfree(adapter->rx_ring);
2651 kfree(adapter->tx_ring);
2656 * e1000e_poll - NAPI Rx polling callback
2657 * @napi: struct associated with this polling callback
2658 * @weight: number of packets driver is allowed to process this poll
2660 static int e1000e_poll(struct napi_struct *napi, int weight)
2662 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2664 struct e1000_hw *hw = &adapter->hw;
2665 struct net_device *poll_dev = adapter->netdev;
2666 int tx_cleaned = 1, work_done = 0;
2668 adapter = netdev_priv(poll_dev);
2670 if (!adapter->msix_entries ||
2671 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2672 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2674 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2679 /* If weight not fully consumed, exit the polling mode */
2680 if (work_done < weight) {
2681 if (adapter->itr_setting & 3)
2682 e1000_set_itr(adapter);
2683 napi_complete_done(napi, work_done);
2684 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2685 if (adapter->msix_entries)
2686 ew32(IMS, adapter->rx_ring->ims_val);
2688 e1000_irq_enable(adapter);
2695 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2696 __always_unused __be16 proto, u16 vid)
2698 struct e1000_adapter *adapter = netdev_priv(netdev);
2699 struct e1000_hw *hw = &adapter->hw;
2702 /* don't update vlan cookie if already programmed */
2703 if ((adapter->hw.mng_cookie.status &
2704 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2705 (vid == adapter->mng_vlan_id))
2708 /* add VID to filter table */
2709 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2710 index = (vid >> 5) & 0x7F;
2711 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2712 vfta |= BIT((vid & 0x1F));
2713 hw->mac.ops.write_vfta(hw, index, vfta);
2716 set_bit(vid, adapter->active_vlans);
2721 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2722 __always_unused __be16 proto, u16 vid)
2724 struct e1000_adapter *adapter = netdev_priv(netdev);
2725 struct e1000_hw *hw = &adapter->hw;
2728 if ((adapter->hw.mng_cookie.status &
2729 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2730 (vid == adapter->mng_vlan_id)) {
2731 /* release control to f/w */
2732 e1000e_release_hw_control(adapter);
2736 /* remove VID from filter table */
2737 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2738 index = (vid >> 5) & 0x7F;
2739 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2740 vfta &= ~BIT((vid & 0x1F));
2741 hw->mac.ops.write_vfta(hw, index, vfta);
2744 clear_bit(vid, adapter->active_vlans);
2750 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2751 * @adapter: board private structure to initialize
2753 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2755 struct net_device *netdev = adapter->netdev;
2756 struct e1000_hw *hw = &adapter->hw;
2759 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2760 /* disable VLAN receive filtering */
2762 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2765 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2766 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2767 adapter->mng_vlan_id);
2768 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2774 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2775 * @adapter: board private structure to initialize
2777 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2779 struct e1000_hw *hw = &adapter->hw;
2782 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2783 /* enable VLAN receive filtering */
2785 rctl |= E1000_RCTL_VFE;
2786 rctl &= ~E1000_RCTL_CFIEN;
2792 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2793 * @adapter: board private structure to initialize
2795 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2797 struct e1000_hw *hw = &adapter->hw;
2800 /* disable VLAN tag insert/strip */
2802 ctrl &= ~E1000_CTRL_VME;
2807 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2808 * @adapter: board private structure to initialize
2810 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2812 struct e1000_hw *hw = &adapter->hw;
2815 /* enable VLAN tag insert/strip */
2817 ctrl |= E1000_CTRL_VME;
2821 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2823 struct net_device *netdev = adapter->netdev;
2824 u16 vid = adapter->hw.mng_cookie.vlan_id;
2825 u16 old_vid = adapter->mng_vlan_id;
2827 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2828 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2829 adapter->mng_vlan_id = vid;
2832 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2833 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2836 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2840 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2842 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2843 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2846 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2848 struct e1000_hw *hw = &adapter->hw;
2849 u32 manc, manc2h, mdef, i, j;
2851 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2856 /* enable receiving management packets to the host. this will probably
2857 * generate destination unreachable messages from the host OS, but
2858 * the packets will be handled on SMBUS
2860 manc |= E1000_MANC_EN_MNG2HOST;
2861 manc2h = er32(MANC2H);
2863 switch (hw->mac.type) {
2865 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2869 /* Check if IPMI pass-through decision filter already exists;
2872 for (i = 0, j = 0; i < 8; i++) {
2873 mdef = er32(MDEF(i));
2875 /* Ignore filters with anything other than IPMI ports */
2876 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2879 /* Enable this decision filter in MANC2H */
2886 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2889 /* Create new decision filter in an empty filter */
2890 for (i = 0, j = 0; i < 8; i++)
2891 if (er32(MDEF(i)) == 0) {
2892 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2893 E1000_MDEF_PORT_664));
2900 e_warn("Unable to create IPMI pass-through filter\n");
2904 ew32(MANC2H, manc2h);
2909 * e1000_configure_tx - Configure Transmit Unit after Reset
2910 * @adapter: board private structure
2912 * Configure the Tx unit of the MAC after a reset.
2914 static void e1000_configure_tx(struct e1000_adapter *adapter)
2916 struct e1000_hw *hw = &adapter->hw;
2917 struct e1000_ring *tx_ring = adapter->tx_ring;
2919 u32 tdlen, tctl, tarc;
2921 /* Setup the HW Tx Head and Tail descriptor pointers */
2922 tdba = tx_ring->dma;
2923 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2924 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2925 ew32(TDBAH(0), (tdba >> 32));
2926 ew32(TDLEN(0), tdlen);
2929 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2930 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2932 writel(0, tx_ring->head);
2933 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2934 e1000e_update_tdt_wa(tx_ring, 0);
2936 writel(0, tx_ring->tail);
2938 /* Set the Tx Interrupt Delay register */
2939 ew32(TIDV, adapter->tx_int_delay);
2940 /* Tx irq moderation */
2941 ew32(TADV, adapter->tx_abs_int_delay);
2943 if (adapter->flags2 & FLAG2_DMA_BURST) {
2944 u32 txdctl = er32(TXDCTL(0));
2946 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2947 E1000_TXDCTL_WTHRESH);
2948 /* set up some performance related parameters to encourage the
2949 * hardware to use the bus more efficiently in bursts, depends
2950 * on the tx_int_delay to be enabled,
2951 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2952 * hthresh = 1 ==> prefetch when one or more available
2953 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2954 * BEWARE: this seems to work but should be considered first if
2955 * there are Tx hangs or other Tx related bugs
2957 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2958 ew32(TXDCTL(0), txdctl);
2960 /* erratum work around: set txdctl the same for both queues */
2961 ew32(TXDCTL(1), er32(TXDCTL(0)));
2963 /* Program the Transmit Control Register */
2965 tctl &= ~E1000_TCTL_CT;
2966 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2967 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2969 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2970 tarc = er32(TARC(0));
2971 /* set the speed mode bit, we'll clear it if we're not at
2972 * gigabit link later
2974 #define SPEED_MODE_BIT BIT(21)
2975 tarc |= SPEED_MODE_BIT;
2976 ew32(TARC(0), tarc);
2979 /* errata: program both queues to unweighted RR */
2980 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2981 tarc = er32(TARC(0));
2983 ew32(TARC(0), tarc);
2984 tarc = er32(TARC(1));
2986 ew32(TARC(1), tarc);
2989 /* Setup Transmit Descriptor Settings for eop descriptor */
2990 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2992 /* only set IDE if we are delaying interrupts using the timers */
2993 if (adapter->tx_int_delay)
2994 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2996 /* enable Report Status bit */
2997 adapter->txd_cmd |= E1000_TXD_CMD_RS;
3001 hw->mac.ops.config_collision_dist(hw);
3003 /* SPT Si errata workaround to avoid data corruption */
3004 if (hw->mac.type == e1000_pch_spt) {
3007 reg_val = er32(IOSFPC);
3008 reg_val |= E1000_RCTL_RDMTS_HEX;
3009 ew32(IOSFPC, reg_val);
3011 reg_val = er32(TARC(0));
3012 reg_val |= E1000_TARC0_CB_MULTIQ_3_REQ;
3013 ew32(TARC(0), reg_val);
3018 * e1000_setup_rctl - configure the receive control registers
3019 * @adapter: Board private structure
3021 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3022 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3023 static void e1000_setup_rctl(struct e1000_adapter *adapter)
3025 struct e1000_hw *hw = &adapter->hw;
3029 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3030 * If jumbo frames not set, program related MAC/PHY registers
3033 if (hw->mac.type >= e1000_pch2lan) {
3036 if (adapter->netdev->mtu > ETH_DATA_LEN)
3037 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3039 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3042 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3045 /* Program MC offset vector base */
3047 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3048 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3049 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3050 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3052 /* Do not Store bad packets */
3053 rctl &= ~E1000_RCTL_SBP;
3055 /* Enable Long Packet receive */
3056 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3057 rctl &= ~E1000_RCTL_LPE;
3059 rctl |= E1000_RCTL_LPE;
3061 /* Some systems expect that the CRC is included in SMBUS traffic. The
3062 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3063 * host memory when this is enabled
3065 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3066 rctl |= E1000_RCTL_SECRC;
3068 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3069 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3072 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3075 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3077 e1e_rphy(hw, 22, &phy_data);
3079 phy_data |= BIT(14);
3080 e1e_wphy(hw, 0x10, 0x2823);
3081 e1e_wphy(hw, 0x11, 0x0003);
3082 e1e_wphy(hw, 22, phy_data);
3085 /* Setup buffer sizes */
3086 rctl &= ~E1000_RCTL_SZ_4096;
3087 rctl |= E1000_RCTL_BSEX;
3088 switch (adapter->rx_buffer_len) {
3091 rctl |= E1000_RCTL_SZ_2048;
3092 rctl &= ~E1000_RCTL_BSEX;
3095 rctl |= E1000_RCTL_SZ_4096;
3098 rctl |= E1000_RCTL_SZ_8192;
3101 rctl |= E1000_RCTL_SZ_16384;
3105 /* Enable Extended Status in all Receive Descriptors */
3106 rfctl = er32(RFCTL);
3107 rfctl |= E1000_RFCTL_EXTEN;
3110 /* 82571 and greater support packet-split where the protocol
3111 * header is placed in skb->data and the packet data is
3112 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3113 * In the case of a non-split, skb->data is linearly filled,
3114 * followed by the page buffers. Therefore, skb->data is
3115 * sized to hold the largest protocol header.
3117 * allocations using alloc_page take too long for regular MTU
3118 * so only enable packet split for jumbo frames
3120 * Using pages when the page size is greater than 16k wastes
3121 * a lot of memory, since we allocate 3 pages at all times
3124 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3125 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3126 adapter->rx_ps_pages = pages;
3128 adapter->rx_ps_pages = 0;
3130 if (adapter->rx_ps_pages) {
3133 /* Enable Packet split descriptors */
3134 rctl |= E1000_RCTL_DTYP_PS;
3136 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3138 switch (adapter->rx_ps_pages) {
3140 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3143 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3146 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3150 ew32(PSRCTL, psrctl);
3153 /* This is useful for sniffing bad packets. */
3154 if (adapter->netdev->features & NETIF_F_RXALL) {
3155 /* UPE and MPE will be handled by normal PROMISC logic
3156 * in e1000e_set_rx_mode
3158 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3159 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3160 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3162 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3163 E1000_RCTL_DPF | /* Allow filtered pause */
3164 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3165 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3166 * and that breaks VLANs.
3171 /* just started the receive unit, no need to restart */
3172 adapter->flags &= ~FLAG_RESTART_NOW;
3176 * e1000_configure_rx - Configure Receive Unit after Reset
3177 * @adapter: board private structure
3179 * Configure the Rx unit of the MAC after a reset.
3181 static void e1000_configure_rx(struct e1000_adapter *adapter)
3183 struct e1000_hw *hw = &adapter->hw;
3184 struct e1000_ring *rx_ring = adapter->rx_ring;
3186 u32 rdlen, rctl, rxcsum, ctrl_ext;
3188 if (adapter->rx_ps_pages) {
3189 /* this is a 32 byte descriptor */
3190 rdlen = rx_ring->count *
3191 sizeof(union e1000_rx_desc_packet_split);
3192 adapter->clean_rx = e1000_clean_rx_irq_ps;
3193 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3194 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3195 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3196 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3197 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3199 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3200 adapter->clean_rx = e1000_clean_rx_irq;
3201 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3204 /* disable receives while setting up the descriptors */
3206 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3207 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3209 usleep_range(10000, 20000);
3211 if (adapter->flags2 & FLAG2_DMA_BURST) {
3212 /* set the writeback threshold (only takes effect if the RDTR
3213 * is set). set GRAN=1 and write back up to 0x4 worth, and
3214 * enable prefetching of 0x20 Rx descriptors
3220 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3221 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3223 /* override the delay timers for enabling bursting, only if
3224 * the value was not set by the user via module options
3226 if (adapter->rx_int_delay == DEFAULT_RDTR)
3227 adapter->rx_int_delay = BURST_RDTR;
3228 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3229 adapter->rx_abs_int_delay = BURST_RADV;
3232 /* set the Receive Delay Timer Register */
3233 ew32(RDTR, adapter->rx_int_delay);
3235 /* irq moderation */
3236 ew32(RADV, adapter->rx_abs_int_delay);
3237 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3238 e1000e_write_itr(adapter, adapter->itr);
3240 ctrl_ext = er32(CTRL_EXT);
3241 /* Auto-Mask interrupts upon ICR access */
3242 ctrl_ext |= E1000_CTRL_EXT_IAME;
3243 ew32(IAM, 0xffffffff);
3244 ew32(CTRL_EXT, ctrl_ext);
3247 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3248 * the Base and Length of the Rx Descriptor Ring
3250 rdba = rx_ring->dma;
3251 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3252 ew32(RDBAH(0), (rdba >> 32));
3253 ew32(RDLEN(0), rdlen);
3256 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3257 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3259 writel(0, rx_ring->head);
3260 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3261 e1000e_update_rdt_wa(rx_ring, 0);
3263 writel(0, rx_ring->tail);
3265 /* Enable Receive Checksum Offload for TCP and UDP */
3266 rxcsum = er32(RXCSUM);
3267 if (adapter->netdev->features & NETIF_F_RXCSUM)
3268 rxcsum |= E1000_RXCSUM_TUOFL;
3270 rxcsum &= ~E1000_RXCSUM_TUOFL;
3271 ew32(RXCSUM, rxcsum);
3273 /* With jumbo frames, excessive C-state transition latencies result
3274 * in dropped transactions.
3276 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3278 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3279 adapter->max_frame_size) * 8 / 1000;
3281 if (adapter->flags & FLAG_IS_ICH) {
3282 u32 rxdctl = er32(RXDCTL(0));
3284 ew32(RXDCTL(0), rxdctl | 0x3);
3287 pm_qos_update_request(&adapter->pm_qos_req, lat);
3289 pm_qos_update_request(&adapter->pm_qos_req,
3290 PM_QOS_DEFAULT_VALUE);
3293 /* Enable Receives */
3298 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3299 * @netdev: network interface device structure
3301 * Writes multicast address list to the MTA hash table.
3302 * Returns: -ENOMEM on failure
3303 * 0 on no addresses written
3304 * X on writing X addresses to MTA
3306 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3308 struct e1000_adapter *adapter = netdev_priv(netdev);
3309 struct e1000_hw *hw = &adapter->hw;
3310 struct netdev_hw_addr *ha;
3314 if (netdev_mc_empty(netdev)) {
3315 /* nothing to program, so clear mc list */
3316 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3320 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3324 /* update_mc_addr_list expects a packed array of only addresses. */
3326 netdev_for_each_mc_addr(ha, netdev)
3327 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3329 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3332 return netdev_mc_count(netdev);
3336 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3337 * @netdev: network interface device structure
3339 * Writes unicast address list to the RAR table.
3340 * Returns: -ENOMEM on failure/insufficient address space
3341 * 0 on no addresses written
3342 * X on writing X addresses to the RAR table
3344 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3346 struct e1000_adapter *adapter = netdev_priv(netdev);
3347 struct e1000_hw *hw = &adapter->hw;
3348 unsigned int rar_entries;
3351 rar_entries = hw->mac.ops.rar_get_count(hw);
3353 /* save a rar entry for our hardware address */
3356 /* save a rar entry for the LAA workaround */
3357 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3360 /* return ENOMEM indicating insufficient memory for addresses */
3361 if (netdev_uc_count(netdev) > rar_entries)
3364 if (!netdev_uc_empty(netdev) && rar_entries) {
3365 struct netdev_hw_addr *ha;
3367 /* write the addresses in reverse order to avoid write
3370 netdev_for_each_uc_addr(ha, netdev) {
3375 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3382 /* zero out the remaining RAR entries not used above */
3383 for (; rar_entries > 0; rar_entries--) {
3384 ew32(RAH(rar_entries), 0);
3385 ew32(RAL(rar_entries), 0);
3393 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3394 * @netdev: network interface device structure
3396 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3397 * address list or the network interface flags are updated. This routine is
3398 * responsible for configuring the hardware for proper unicast, multicast,
3399 * promiscuous mode, and all-multi behavior.
3401 static void e1000e_set_rx_mode(struct net_device *netdev)
3403 struct e1000_adapter *adapter = netdev_priv(netdev);
3404 struct e1000_hw *hw = &adapter->hw;
3407 if (pm_runtime_suspended(netdev->dev.parent))
3410 /* Check for Promiscuous and All Multicast modes */
3413 /* clear the affected bits */
3414 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3416 if (netdev->flags & IFF_PROMISC) {
3417 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3418 /* Do not hardware filter VLANs in promisc mode */
3419 e1000e_vlan_filter_disable(adapter);
3423 if (netdev->flags & IFF_ALLMULTI) {
3424 rctl |= E1000_RCTL_MPE;
3426 /* Write addresses to the MTA, if the attempt fails
3427 * then we should just turn on promiscuous mode so
3428 * that we can at least receive multicast traffic
3430 count = e1000e_write_mc_addr_list(netdev);
3432 rctl |= E1000_RCTL_MPE;
3434 e1000e_vlan_filter_enable(adapter);
3435 /* Write addresses to available RAR registers, if there is not
3436 * sufficient space to store all the addresses then enable
3437 * unicast promiscuous mode
3439 count = e1000e_write_uc_addr_list(netdev);
3441 rctl |= E1000_RCTL_UPE;
3446 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3447 e1000e_vlan_strip_enable(adapter);
3449 e1000e_vlan_strip_disable(adapter);
3452 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3454 struct e1000_hw *hw = &adapter->hw;
3459 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3460 for (i = 0; i < 10; i++)
3461 ew32(RSSRK(i), rss_key[i]);
3463 /* Direct all traffic to queue 0 */
3464 for (i = 0; i < 32; i++)
3467 /* Disable raw packet checksumming so that RSS hash is placed in
3468 * descriptor on writeback.
3470 rxcsum = er32(RXCSUM);
3471 rxcsum |= E1000_RXCSUM_PCSD;
3473 ew32(RXCSUM, rxcsum);
3475 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3476 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3477 E1000_MRQC_RSS_FIELD_IPV6 |
3478 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3479 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3485 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3486 * @adapter: board private structure
3487 * @timinca: pointer to returned time increment attributes
3489 * Get attributes for incrementing the System Time Register SYSTIML/H at
3490 * the default base frequency, and set the cyclecounter shift value.
3492 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3494 struct e1000_hw *hw = &adapter->hw;
3495 u32 incvalue, incperiod, shift;
3497 /* Make sure clock is enabled on I217/I218/I219 before checking
3500 if (((hw->mac.type == e1000_pch_lpt) ||
3501 (hw->mac.type == e1000_pch_spt)) &&
3502 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3503 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3504 u32 fextnvm7 = er32(FEXTNVM7);
3506 if (!(fextnvm7 & BIT(0))) {
3507 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3512 switch (hw->mac.type) {
3515 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3516 /* Stable 96MHz frequency */
3517 incperiod = INCPERIOD_96MHz;
3518 incvalue = INCVALUE_96MHz;
3519 shift = INCVALUE_SHIFT_96MHz;
3520 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHz;
3522 /* Stable 25MHz frequency */
3523 incperiod = INCPERIOD_25MHz;
3524 incvalue = INCVALUE_25MHz;
3525 shift = INCVALUE_SHIFT_25MHz;
3526 adapter->cc.shift = shift;
3530 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3531 /* Stable 24MHz frequency */
3532 incperiod = INCPERIOD_24MHz;
3533 incvalue = INCVALUE_24MHz;
3534 shift = INCVALUE_SHIFT_24MHz;
3535 adapter->cc.shift = shift;
3541 /* Stable 25MHz frequency */
3542 incperiod = INCPERIOD_25MHz;
3543 incvalue = INCVALUE_25MHz;
3544 shift = INCVALUE_SHIFT_25MHz;
3545 adapter->cc.shift = shift;
3551 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3552 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3558 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3559 * @adapter: board private structure
3561 * Outgoing time stamping can be enabled and disabled. Play nice and
3562 * disable it when requested, although it shouldn't cause any overhead
3563 * when no packet needs it. At most one packet in the queue may be
3564 * marked for time stamping, otherwise it would be impossible to tell
3565 * for sure to which packet the hardware time stamp belongs.
3567 * Incoming time stamping has to be configured via the hardware filters.
3568 * Not all combinations are supported, in particular event type has to be
3569 * specified. Matching the kind of event packet is not supported, with the
3570 * exception of "all V2 events regardless of level 2 or 4".
3572 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3573 struct hwtstamp_config *config)
3575 struct e1000_hw *hw = &adapter->hw;
3576 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3577 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3584 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3587 /* flags reserved for future extensions - must be zero */
3591 switch (config->tx_type) {
3592 case HWTSTAMP_TX_OFF:
3595 case HWTSTAMP_TX_ON:
3601 switch (config->rx_filter) {
3602 case HWTSTAMP_FILTER_NONE:
3605 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3606 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3607 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3610 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3611 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3612 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3615 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3616 /* Also time stamps V2 L2 Path Delay Request/Response */
3617 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3618 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3621 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3622 /* Also time stamps V2 L2 Path Delay Request/Response. */
3623 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3624 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3627 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3628 /* Hardware cannot filter just V2 L4 Sync messages;
3629 * fall-through to V2 (both L2 and L4) Sync.
3631 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3632 /* Also time stamps V2 Path Delay Request/Response. */
3633 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3634 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3638 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3639 /* Hardware cannot filter just V2 L4 Delay Request messages;
3640 * fall-through to V2 (both L2 and L4) Delay Request.
3642 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3643 /* Also time stamps V2 Path Delay Request/Response. */
3644 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3645 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3649 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3650 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3651 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3652 * fall-through to all V2 (both L2 and L4) Events.
3654 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3655 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3656 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3660 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3661 /* For V1, the hardware can only filter Sync messages or
3662 * Delay Request messages but not both so fall-through to
3663 * time stamp all packets.
3665 case HWTSTAMP_FILTER_ALL:
3668 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3669 config->rx_filter = HWTSTAMP_FILTER_ALL;
3675 adapter->hwtstamp_config = *config;
3677 /* enable/disable Tx h/w time stamping */
3678 regval = er32(TSYNCTXCTL);
3679 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3680 regval |= tsync_tx_ctl;
3681 ew32(TSYNCTXCTL, regval);
3682 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3683 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3684 e_err("Timesync Tx Control register not set as expected\n");
3688 /* enable/disable Rx h/w time stamping */
3689 regval = er32(TSYNCRXCTL);
3690 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3691 regval |= tsync_rx_ctl;
3692 ew32(TSYNCRXCTL, regval);
3693 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3694 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3695 (regval & (E1000_TSYNCRXCTL_ENABLED |
3696 E1000_TSYNCRXCTL_TYPE_MASK))) {
3697 e_err("Timesync Rx Control register not set as expected\n");
3701 /* L2: define ethertype filter for time stamped packets */
3703 rxmtrl |= ETH_P_1588;
3705 /* define which PTP packets get time stamped */
3706 ew32(RXMTRL, rxmtrl);
3708 /* Filter by destination port */
3710 rxudp = PTP_EV_PORT;
3711 cpu_to_be16s(&rxudp);
3717 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3725 * e1000_configure - configure the hardware for Rx and Tx
3726 * @adapter: private board structure
3728 static void e1000_configure(struct e1000_adapter *adapter)
3730 struct e1000_ring *rx_ring = adapter->rx_ring;
3732 e1000e_set_rx_mode(adapter->netdev);
3734 e1000_restore_vlan(adapter);
3735 e1000_init_manageability_pt(adapter);
3737 e1000_configure_tx(adapter);
3739 if (adapter->netdev->features & NETIF_F_RXHASH)
3740 e1000e_setup_rss_hash(adapter);
3741 e1000_setup_rctl(adapter);
3742 e1000_configure_rx(adapter);
3743 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3747 * e1000e_power_up_phy - restore link in case the phy was powered down
3748 * @adapter: address of board private structure
3750 * The phy may be powered down to save power and turn off link when the
3751 * driver is unloaded and wake on lan is not enabled (among others)
3752 * *** this routine MUST be followed by a call to e1000e_reset ***
3754 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3756 if (adapter->hw.phy.ops.power_up)
3757 adapter->hw.phy.ops.power_up(&adapter->hw);
3759 adapter->hw.mac.ops.setup_link(&adapter->hw);
3763 * e1000_power_down_phy - Power down the PHY
3765 * Power down the PHY so no link is implied when interface is down.
3766 * The PHY cannot be powered down if management or WoL is active.
3768 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3770 if (adapter->hw.phy.ops.power_down)
3771 adapter->hw.phy.ops.power_down(&adapter->hw);
3775 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3777 * We want to clear all pending descriptors from the TX ring.
3778 * zeroing happens when the HW reads the regs. We assign the ring itself as
3779 * the data of the next descriptor. We don't care about the data we are about
3782 static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3784 struct e1000_hw *hw = &adapter->hw;
3785 struct e1000_ring *tx_ring = adapter->tx_ring;
3786 struct e1000_tx_desc *tx_desc = NULL;
3787 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3791 ew32(TCTL, tctl | E1000_TCTL_EN);
3793 BUG_ON(tdt != tx_ring->next_to_use);
3794 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3795 tx_desc->buffer_addr = tx_ring->dma;
3797 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3798 tx_desc->upper.data = 0;
3799 /* flush descriptors to memory before notifying the HW */
3801 tx_ring->next_to_use++;
3802 if (tx_ring->next_to_use == tx_ring->count)
3803 tx_ring->next_to_use = 0;
3804 ew32(TDT(0), tx_ring->next_to_use);
3806 usleep_range(200, 250);
3810 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3812 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3814 static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3817 struct e1000_hw *hw = &adapter->hw;
3820 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3822 usleep_range(100, 150);
3824 rxdctl = er32(RXDCTL(0));
3825 /* zero the lower 14 bits (prefetch and host thresholds) */
3826 rxdctl &= 0xffffc000;
3828 /* update thresholds: prefetch threshold to 31, host threshold to 1
3829 * and make sure the granularity is "descriptors" and not "cache lines"
3831 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3833 ew32(RXDCTL(0), rxdctl);
3834 /* momentarily enable the RX ring for the changes to take effect */
3835 ew32(RCTL, rctl | E1000_RCTL_EN);
3837 usleep_range(100, 150);
3838 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3842 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3844 * In i219, the descriptor rings must be emptied before resetting the HW
3845 * or before changing the device state to D3 during runtime (runtime PM).
3847 * Failure to do this will cause the HW to enter a unit hang state which can
3848 * only be released by PCI reset on the device
3852 static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3855 u32 fext_nvm11, tdlen;
3856 struct e1000_hw *hw = &adapter->hw;
3858 /* First, disable MULR fix in FEXTNVM11 */
3859 fext_nvm11 = er32(FEXTNVM11);
3860 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3861 ew32(FEXTNVM11, fext_nvm11);
3862 /* do nothing if we're not in faulty state, or if the queue is empty */
3863 tdlen = er32(TDLEN(0));
3864 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3866 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3868 e1000_flush_tx_ring(adapter);
3869 /* recheck, maybe the fault is caused by the rx ring */
3870 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3872 if (hang_state & FLUSH_DESC_REQUIRED)
3873 e1000_flush_rx_ring(adapter);
3877 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3878 * @adapter: board private structure
3880 * When the MAC is reset, all hardware bits for timesync will be reset to the
3881 * default values. This function will restore the settings last in place.
3882 * Since the clock SYSTIME registers are reset, we will simply restore the
3883 * cyclecounter to the kernel real clock time.
3885 static void e1000e_systim_reset(struct e1000_adapter *adapter)
3887 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3888 struct e1000_hw *hw = &adapter->hw;
3889 unsigned long flags;
3893 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3896 if (info->adjfreq) {
3897 /* restore the previous ptp frequency delta */
3898 ret_val = info->adjfreq(info, adapter->ptp_delta);
3900 /* set the default base frequency if no adjustment possible */
3901 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3903 ew32(TIMINCA, timinca);
3907 dev_warn(&adapter->pdev->dev,
3908 "Failed to restore TIMINCA clock rate delta: %d\n",
3913 /* reset the systim ns time counter */
3914 spin_lock_irqsave(&adapter->systim_lock, flags);
3915 timecounter_init(&adapter->tc, &adapter->cc,
3916 ktime_to_ns(ktime_get_real()));
3917 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3919 /* restore the previous hwtstamp configuration settings */
3920 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3924 * e1000e_reset - bring the hardware into a known good state
3926 * This function boots the hardware and enables some settings that
3927 * require a configuration cycle of the hardware - those cannot be
3928 * set/changed during runtime. After reset the device needs to be
3929 * properly configured for Rx, Tx etc.
3931 void e1000e_reset(struct e1000_adapter *adapter)
3933 struct e1000_mac_info *mac = &adapter->hw.mac;
3934 struct e1000_fc_info *fc = &adapter->hw.fc;
3935 struct e1000_hw *hw = &adapter->hw;
3936 u32 tx_space, min_tx_space, min_rx_space;
3937 u32 pba = adapter->pba;
3940 /* reset Packet Buffer Allocation to default */
3943 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3944 /* To maintain wire speed transmits, the Tx FIFO should be
3945 * large enough to accommodate two full transmit packets,
3946 * rounded up to the next 1KB and expressed in KB. Likewise,
3947 * the Rx FIFO should be large enough to accommodate at least
3948 * one full receive packet and is similarly rounded up and
3952 /* upper 16 bits has Tx packet buffer allocation size in KB */
3953 tx_space = pba >> 16;
3954 /* lower 16 bits has Rx packet buffer allocation size in KB */
3956 /* the Tx fifo also stores 16 bytes of information about the Tx
3957 * but don't include ethernet FCS because hardware appends it
3959 min_tx_space = (adapter->max_frame_size +
3960 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3961 min_tx_space = ALIGN(min_tx_space, 1024);
3962 min_tx_space >>= 10;
3963 /* software strips receive CRC, so leave room for it */
3964 min_rx_space = adapter->max_frame_size;
3965 min_rx_space = ALIGN(min_rx_space, 1024);
3966 min_rx_space >>= 10;
3968 /* If current Tx allocation is less than the min Tx FIFO size,
3969 * and the min Tx FIFO size is less than the current Rx FIFO
3970 * allocation, take space away from current Rx allocation
3972 if ((tx_space < min_tx_space) &&
3973 ((min_tx_space - tx_space) < pba)) {
3974 pba -= min_tx_space - tx_space;
3976 /* if short on Rx space, Rx wins and must trump Tx
3979 if (pba < min_rx_space)
3986 /* flow control settings
3988 * The high water mark must be low enough to fit one full frame
3989 * (or the size used for early receive) above it in the Rx FIFO.
3990 * Set it to the lower of:
3991 * - 90% of the Rx FIFO size, and
3992 * - the full Rx FIFO size minus one full frame
3994 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3995 fc->pause_time = 0xFFFF;
3997 fc->pause_time = E1000_FC_PAUSE_TIME;
3998 fc->send_xon = true;
3999 fc->current_mode = fc->requested_mode;
4001 switch (hw->mac.type) {
4003 case e1000_ich10lan:
4004 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4007 fc->high_water = 0x2800;
4008 fc->low_water = fc->high_water - 8;
4013 hwm = min(((pba << 10) * 9 / 10),
4014 ((pba << 10) - adapter->max_frame_size));
4016 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4017 fc->low_water = fc->high_water - 8;
4020 /* Workaround PCH LOM adapter hangs with certain network
4021 * loads. If hangs persist, try disabling Tx flow control.
4023 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4024 fc->high_water = 0x3500;
4025 fc->low_water = 0x1500;
4027 fc->high_water = 0x5000;
4028 fc->low_water = 0x3000;
4030 fc->refresh_time = 0x1000;
4035 fc->refresh_time = 0x0400;
4037 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4038 fc->high_water = 0x05C20;
4039 fc->low_water = 0x05048;
4040 fc->pause_time = 0x0650;
4046 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4047 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4051 /* Alignment of Tx data is on an arbitrary byte boundary with the
4052 * maximum size per Tx descriptor limited only to the transmit
4053 * allocation of the packet buffer minus 96 bytes with an upper
4054 * limit of 24KB due to receive synchronization limitations.
4056 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4059 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4060 * fit in receive buffer.
4062 if (adapter->itr_setting & 0x3) {
4063 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4064 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4065 dev_info(&adapter->pdev->dev,
4066 "Interrupt Throttle Rate off\n");
4067 adapter->flags2 |= FLAG2_DISABLE_AIM;
4068 e1000e_write_itr(adapter, 0);
4070 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4071 dev_info(&adapter->pdev->dev,
4072 "Interrupt Throttle Rate on\n");
4073 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4074 adapter->itr = 20000;
4075 e1000e_write_itr(adapter, adapter->itr);
4079 if (hw->mac.type == e1000_pch_spt)
4080 e1000_flush_desc_rings(adapter);
4081 /* Allow time for pending master requests to run */
4082 mac->ops.reset_hw(hw);
4084 /* For parts with AMT enabled, let the firmware know
4085 * that the network interface is in control
4087 if (adapter->flags & FLAG_HAS_AMT)
4088 e1000e_get_hw_control(adapter);
4092 if (mac->ops.init_hw(hw))
4093 e_err("Hardware Error\n");
4095 e1000_update_mng_vlan(adapter);
4097 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4098 ew32(VET, ETH_P_8021Q);
4100 e1000e_reset_adaptive(hw);
4102 /* restore systim and hwtstamp settings */
4103 e1000e_systim_reset(adapter);
4105 /* Set EEE advertisement as appropriate */
4106 if (adapter->flags2 & FLAG2_HAS_EEE) {
4110 switch (hw->phy.type) {
4111 case e1000_phy_82579:
4112 adv_addr = I82579_EEE_ADVERTISEMENT;
4114 case e1000_phy_i217:
4115 adv_addr = I217_EEE_ADVERTISEMENT;
4118 dev_err(&adapter->pdev->dev,
4119 "Invalid PHY type setting EEE advertisement\n");
4123 ret_val = hw->phy.ops.acquire(hw);
4125 dev_err(&adapter->pdev->dev,
4126 "EEE advertisement - unable to acquire PHY\n");
4130 e1000_write_emi_reg_locked(hw, adv_addr,
4131 hw->dev_spec.ich8lan.eee_disable ?
4132 0 : adapter->eee_advert);
4134 hw->phy.ops.release(hw);
4137 if (!netif_running(adapter->netdev) &&
4138 !test_bit(__E1000_TESTING, &adapter->state))
4139 e1000_power_down_phy(adapter);
4141 e1000_get_phy_info(hw);
4143 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4144 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4146 /* speed up time to link by disabling smart power down, ignore
4147 * the return value of this function because there is nothing
4148 * different we would do if it failed
4150 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4151 phy_data &= ~IGP02E1000_PM_SPD;
4152 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4154 if (hw->mac.type == e1000_pch_spt && adapter->int_mode == 0) {
4157 /* Fextnvm7 @ 0xe4[2] = 1 */
4158 reg = er32(FEXTNVM7);
4159 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4160 ew32(FEXTNVM7, reg);
4161 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4162 reg = er32(FEXTNVM9);
4163 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4164 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4165 ew32(FEXTNVM9, reg);
4171 * e1000e_trigger_lsc - trigger an LSC interrupt
4174 * Fire a link status change interrupt to start the watchdog.
4176 static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4178 struct e1000_hw *hw = &adapter->hw;
4180 if (adapter->msix_entries)
4181 ew32(ICS, E1000_ICS_OTHER);
4183 ew32(ICS, E1000_ICS_LSC);
4186 void e1000e_up(struct e1000_adapter *adapter)
4188 /* hardware has been reset, we need to reload some things */
4189 e1000_configure(adapter);
4191 clear_bit(__E1000_DOWN, &adapter->state);
4193 if (adapter->msix_entries)
4194 e1000_configure_msix(adapter);
4195 e1000_irq_enable(adapter);
4197 netif_start_queue(adapter->netdev);
4199 e1000e_trigger_lsc(adapter);
4202 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4204 struct e1000_hw *hw = &adapter->hw;
4206 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4209 /* flush pending descriptor writebacks to memory */
4210 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4211 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4213 /* execute the writes immediately */
4216 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4217 * write is successful
4219 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4220 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4222 /* execute the writes immediately */
4226 static void e1000e_update_stats(struct e1000_adapter *adapter);
4229 * e1000e_down - quiesce the device and optionally reset the hardware
4230 * @adapter: board private structure
4231 * @reset: boolean flag to reset the hardware or not
4233 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4235 struct net_device *netdev = adapter->netdev;
4236 struct e1000_hw *hw = &adapter->hw;
4239 /* signal that we're down so the interrupt handler does not
4240 * reschedule our watchdog timer
4242 set_bit(__E1000_DOWN, &adapter->state);
4244 netif_carrier_off(netdev);
4246 /* disable receives in the hardware */
4248 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4249 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4250 /* flush and sleep below */
4252 netif_stop_queue(netdev);
4254 /* disable transmits in the hardware */
4256 tctl &= ~E1000_TCTL_EN;
4259 /* flush both disables and wait for them to finish */
4261 usleep_range(10000, 20000);
4263 e1000_irq_disable(adapter);
4265 napi_synchronize(&adapter->napi);
4267 del_timer_sync(&adapter->watchdog_timer);
4268 del_timer_sync(&adapter->phy_info_timer);
4270 spin_lock(&adapter->stats64_lock);
4271 e1000e_update_stats(adapter);
4272 spin_unlock(&adapter->stats64_lock);
4274 e1000e_flush_descriptors(adapter);
4276 adapter->link_speed = 0;
4277 adapter->link_duplex = 0;
4279 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4280 if ((hw->mac.type >= e1000_pch2lan) &&
4281 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4282 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4283 e_dbg("failed to disable jumbo frame workaround mode\n");
4285 if (!pci_channel_offline(adapter->pdev)) {
4287 e1000e_reset(adapter);
4288 else if (hw->mac.type == e1000_pch_spt)
4289 e1000_flush_desc_rings(adapter);
4291 e1000_clean_tx_ring(adapter->tx_ring);
4292 e1000_clean_rx_ring(adapter->rx_ring);
4295 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4298 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4299 usleep_range(1000, 2000);
4300 e1000e_down(adapter, true);
4302 clear_bit(__E1000_RESETTING, &adapter->state);
4306 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4307 * @cc: cyclecounter structure
4309 static cycle_t e1000e_cyclecounter_read(const struct cyclecounter *cc)
4311 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4313 struct e1000_hw *hw = &adapter->hw;
4314 u32 systimel, systimeh;
4315 cycle_t systim, systim_next;
4316 /* SYSTIMH latching upon SYSTIML read does not work well.
4317 * This means that if SYSTIML overflows after we read it but before
4318 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4319 * will experience a huge non linear increment in the systime value
4320 * to fix that we test for overflow and if true, we re-read systime.
4322 systimel = er32(SYSTIML);
4323 systimeh = er32(SYSTIMH);
4324 /* Is systimel is so large that overflow is possible? */
4325 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4326 u32 systimel_2 = er32(SYSTIML);
4327 if (systimel > systimel_2) {
4328 /* There was an overflow, read again SYSTIMH, and use
4331 systimeh = er32(SYSTIMH);
4332 systimel = systimel_2;
4335 systim = (cycle_t)systimel;
4336 systim |= (cycle_t)systimeh << 32;
4338 if ((hw->mac.type == e1000_82574) || (hw->mac.type == e1000_82583)) {
4339 u64 time_delta, rem, temp;
4343 /* errata for 82574/82583 possible bad bits read from SYSTIMH/L
4344 * check to see that the time is incrementing at a reasonable
4345 * rate and is a multiple of incvalue
4347 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4348 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4349 /* latch SYSTIMH on read of SYSTIML */
4350 systim_next = (cycle_t)er32(SYSTIML);
4351 systim_next |= (cycle_t)er32(SYSTIMH) << 32;
4353 time_delta = systim_next - systim;
4355 /* VMWare users have seen incvalue of zero, don't div / 0 */
4356 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4358 systim = systim_next;
4360 if ((time_delta < E1000_82574_SYSTIM_EPSILON) &&
4369 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4370 * @adapter: board private structure to initialize
4372 * e1000_sw_init initializes the Adapter private data structure.
4373 * Fields are initialized based on PCI device information and
4374 * OS network device settings (MTU size).
4376 static int e1000_sw_init(struct e1000_adapter *adapter)
4378 struct net_device *netdev = adapter->netdev;
4380 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4381 adapter->rx_ps_bsize0 = 128;
4382 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4383 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4384 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4385 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4387 spin_lock_init(&adapter->stats64_lock);
4389 e1000e_set_interrupt_capability(adapter);
4391 if (e1000_alloc_queues(adapter))
4394 /* Setup hardware time stamping cyclecounter */
4395 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4396 adapter->cc.read = e1000e_cyclecounter_read;
4397 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4398 adapter->cc.mult = 1;
4399 /* cc.shift set in e1000e_get_base_tininca() */
4401 spin_lock_init(&adapter->systim_lock);
4402 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4405 /* Explicitly disable IRQ since the NIC can be in any state. */
4406 e1000_irq_disable(adapter);
4408 set_bit(__E1000_DOWN, &adapter->state);
4413 * e1000_intr_msi_test - Interrupt Handler
4414 * @irq: interrupt number
4415 * @data: pointer to a network interface device structure
4417 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4419 struct net_device *netdev = data;
4420 struct e1000_adapter *adapter = netdev_priv(netdev);
4421 struct e1000_hw *hw = &adapter->hw;
4422 u32 icr = er32(ICR);
4424 e_dbg("icr is %08X\n", icr);
4425 if (icr & E1000_ICR_RXSEQ) {
4426 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4427 /* Force memory writes to complete before acknowledging the
4428 * interrupt is handled.
4437 * e1000_test_msi_interrupt - Returns 0 for successful test
4438 * @adapter: board private struct
4440 * code flow taken from tg3.c
4442 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4444 struct net_device *netdev = adapter->netdev;
4445 struct e1000_hw *hw = &adapter->hw;
4448 /* poll_enable hasn't been called yet, so don't need disable */
4449 /* clear any pending events */
4452 /* free the real vector and request a test handler */
4453 e1000_free_irq(adapter);
4454 e1000e_reset_interrupt_capability(adapter);
4456 /* Assume that the test fails, if it succeeds then the test
4457 * MSI irq handler will unset this flag
4459 adapter->flags |= FLAG_MSI_TEST_FAILED;
4461 err = pci_enable_msi(adapter->pdev);
4463 goto msi_test_failed;
4465 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4466 netdev->name, netdev);
4468 pci_disable_msi(adapter->pdev);
4469 goto msi_test_failed;
4472 /* Force memory writes to complete before enabling and firing an
4477 e1000_irq_enable(adapter);
4479 /* fire an unusual interrupt on the test handler */
4480 ew32(ICS, E1000_ICS_RXSEQ);
4484 e1000_irq_disable(adapter);
4486 rmb(); /* read flags after interrupt has been fired */
4488 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4489 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4490 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4492 e_dbg("MSI interrupt test succeeded!\n");
4495 free_irq(adapter->pdev->irq, netdev);
4496 pci_disable_msi(adapter->pdev);
4499 e1000e_set_interrupt_capability(adapter);
4500 return e1000_request_irq(adapter);
4504 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4505 * @adapter: board private struct
4507 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4509 static int e1000_test_msi(struct e1000_adapter *adapter)
4514 if (!(adapter->flags & FLAG_MSI_ENABLED))
4517 /* disable SERR in case the MSI write causes a master abort */
4518 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4519 if (pci_cmd & PCI_COMMAND_SERR)
4520 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4521 pci_cmd & ~PCI_COMMAND_SERR);
4523 err = e1000_test_msi_interrupt(adapter);
4525 /* re-enable SERR */
4526 if (pci_cmd & PCI_COMMAND_SERR) {
4527 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4528 pci_cmd |= PCI_COMMAND_SERR;
4529 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4536 * e1000e_open - Called when a network interface is made active
4537 * @netdev: network interface device structure
4539 * Returns 0 on success, negative value on failure
4541 * The open entry point is called when a network interface is made
4542 * active by the system (IFF_UP). At this point all resources needed
4543 * for transmit and receive operations are allocated, the interrupt
4544 * handler is registered with the OS, the watchdog timer is started,
4545 * and the stack is notified that the interface is ready.
4547 int e1000e_open(struct net_device *netdev)
4549 struct e1000_adapter *adapter = netdev_priv(netdev);
4550 struct e1000_hw *hw = &adapter->hw;
4551 struct pci_dev *pdev = adapter->pdev;
4554 /* disallow open during test */
4555 if (test_bit(__E1000_TESTING, &adapter->state))
4558 pm_runtime_get_sync(&pdev->dev);
4560 netif_carrier_off(netdev);
4562 /* allocate transmit descriptors */
4563 err = e1000e_setup_tx_resources(adapter->tx_ring);
4567 /* allocate receive descriptors */
4568 err = e1000e_setup_rx_resources(adapter->rx_ring);
4572 /* If AMT is enabled, let the firmware know that the network
4573 * interface is now open and reset the part to a known state.
4575 if (adapter->flags & FLAG_HAS_AMT) {
4576 e1000e_get_hw_control(adapter);
4577 e1000e_reset(adapter);
4580 e1000e_power_up_phy(adapter);
4582 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4583 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4584 e1000_update_mng_vlan(adapter);
4586 /* DMA latency requirement to workaround jumbo issue */
4587 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4588 PM_QOS_DEFAULT_VALUE);
4590 /* before we allocate an interrupt, we must be ready to handle it.
4591 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4592 * as soon as we call pci_request_irq, so we have to setup our
4593 * clean_rx handler before we do so.
4595 e1000_configure(adapter);
4597 err = e1000_request_irq(adapter);
4601 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4602 * ignore e1000e MSI messages, which means we need to test our MSI
4605 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4606 err = e1000_test_msi(adapter);
4608 e_err("Interrupt allocation failed\n");
4613 /* From here on the code is the same as e1000e_up() */
4614 clear_bit(__E1000_DOWN, &adapter->state);
4616 napi_enable(&adapter->napi);
4618 e1000_irq_enable(adapter);
4620 adapter->tx_hang_recheck = false;
4621 netif_start_queue(netdev);
4623 hw->mac.get_link_status = true;
4624 pm_runtime_put(&pdev->dev);
4626 e1000e_trigger_lsc(adapter);
4631 pm_qos_remove_request(&adapter->pm_qos_req);
4632 e1000e_release_hw_control(adapter);
4633 e1000_power_down_phy(adapter);
4634 e1000e_free_rx_resources(adapter->rx_ring);
4636 e1000e_free_tx_resources(adapter->tx_ring);
4638 e1000e_reset(adapter);
4639 pm_runtime_put_sync(&pdev->dev);
4645 * e1000e_close - Disables a network interface
4646 * @netdev: network interface device structure
4648 * Returns 0, this is not allowed to fail
4650 * The close entry point is called when an interface is de-activated
4651 * by the OS. The hardware is still under the drivers control, but
4652 * needs to be disabled. A global MAC reset is issued to stop the
4653 * hardware, and all transmit and receive resources are freed.
4655 int e1000e_close(struct net_device *netdev)
4657 struct e1000_adapter *adapter = netdev_priv(netdev);
4658 struct pci_dev *pdev = adapter->pdev;
4659 int count = E1000_CHECK_RESET_COUNT;
4661 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4662 usleep_range(10000, 20000);
4664 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4666 pm_runtime_get_sync(&pdev->dev);
4668 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4669 e1000e_down(adapter, true);
4670 e1000_free_irq(adapter);
4672 /* Link status message must follow this format */
4673 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4676 napi_disable(&adapter->napi);
4678 e1000e_free_tx_resources(adapter->tx_ring);
4679 e1000e_free_rx_resources(adapter->rx_ring);
4681 /* kill manageability vlan ID if supported, but not if a vlan with
4682 * the same ID is registered on the host OS (let 8021q kill it)
4684 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4685 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4686 adapter->mng_vlan_id);
4688 /* If AMT is enabled, let the firmware know that the network
4689 * interface is now closed
4691 if ((adapter->flags & FLAG_HAS_AMT) &&
4692 !test_bit(__E1000_TESTING, &adapter->state))
4693 e1000e_release_hw_control(adapter);
4695 pm_qos_remove_request(&adapter->pm_qos_req);
4697 pm_runtime_put_sync(&pdev->dev);
4703 * e1000_set_mac - Change the Ethernet Address of the NIC
4704 * @netdev: network interface device structure
4705 * @p: pointer to an address structure
4707 * Returns 0 on success, negative on failure
4709 static int e1000_set_mac(struct net_device *netdev, void *p)
4711 struct e1000_adapter *adapter = netdev_priv(netdev);
4712 struct e1000_hw *hw = &adapter->hw;
4713 struct sockaddr *addr = p;
4715 if (!is_valid_ether_addr(addr->sa_data))
4716 return -EADDRNOTAVAIL;
4718 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4719 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4721 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4723 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4724 /* activate the work around */
4725 e1000e_set_laa_state_82571(&adapter->hw, 1);
4727 /* Hold a copy of the LAA in RAR[14] This is done so that
4728 * between the time RAR[0] gets clobbered and the time it
4729 * gets fixed (in e1000_watchdog), the actual LAA is in one
4730 * of the RARs and no incoming packets directed to this port
4731 * are dropped. Eventually the LAA will be in RAR[0] and
4734 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4735 adapter->hw.mac.rar_entry_count - 1);
4742 * e1000e_update_phy_task - work thread to update phy
4743 * @work: pointer to our work struct
4745 * this worker thread exists because we must acquire a
4746 * semaphore to read the phy, which we could msleep while
4747 * waiting for it, and we can't msleep in a timer.
4749 static void e1000e_update_phy_task(struct work_struct *work)
4751 struct e1000_adapter *adapter = container_of(work,
4752 struct e1000_adapter,
4754 struct e1000_hw *hw = &adapter->hw;
4756 if (test_bit(__E1000_DOWN, &adapter->state))
4759 e1000_get_phy_info(hw);
4761 /* Enable EEE on 82579 after link up */
4762 if (hw->phy.type >= e1000_phy_82579)
4763 e1000_set_eee_pchlan(hw);
4767 * e1000_update_phy_info - timre call-back to update PHY info
4768 * @data: pointer to adapter cast into an unsigned long
4770 * Need to wait a few seconds after link up to get diagnostic information from
4773 static void e1000_update_phy_info(unsigned long data)
4775 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4777 if (test_bit(__E1000_DOWN, &adapter->state))
4780 schedule_work(&adapter->update_phy_task);
4784 * e1000e_update_phy_stats - Update the PHY statistics counters
4785 * @adapter: board private structure
4787 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4789 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4791 struct e1000_hw *hw = &adapter->hw;
4795 ret_val = hw->phy.ops.acquire(hw);
4799 /* A page set is expensive so check if already on desired page.
4800 * If not, set to the page with the PHY status registers.
4803 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4807 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4808 ret_val = hw->phy.ops.set_page(hw,
4809 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4814 /* Single Collision Count */
4815 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4816 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4818 adapter->stats.scc += phy_data;
4820 /* Excessive Collision Count */
4821 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4822 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4824 adapter->stats.ecol += phy_data;
4826 /* Multiple Collision Count */
4827 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4828 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4830 adapter->stats.mcc += phy_data;
4832 /* Late Collision Count */
4833 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4834 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4836 adapter->stats.latecol += phy_data;
4838 /* Collision Count - also used for adaptive IFS */
4839 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4840 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4842 hw->mac.collision_delta = phy_data;
4845 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4846 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4848 adapter->stats.dc += phy_data;
4850 /* Transmit with no CRS */
4851 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4852 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4854 adapter->stats.tncrs += phy_data;
4857 hw->phy.ops.release(hw);
4861 * e1000e_update_stats - Update the board statistics counters
4862 * @adapter: board private structure
4864 static void e1000e_update_stats(struct e1000_adapter *adapter)
4866 struct net_device *netdev = adapter->netdev;
4867 struct e1000_hw *hw = &adapter->hw;
4868 struct pci_dev *pdev = adapter->pdev;
4870 /* Prevent stats update while adapter is being reset, or if the pci
4871 * connection is down.
4873 if (adapter->link_speed == 0)
4875 if (pci_channel_offline(pdev))
4878 adapter->stats.crcerrs += er32(CRCERRS);
4879 adapter->stats.gprc += er32(GPRC);
4880 adapter->stats.gorc += er32(GORCL);
4881 er32(GORCH); /* Clear gorc */
4882 adapter->stats.bprc += er32(BPRC);
4883 adapter->stats.mprc += er32(MPRC);
4884 adapter->stats.roc += er32(ROC);
4886 adapter->stats.mpc += er32(MPC);
4888 /* Half-duplex statistics */
4889 if (adapter->link_duplex == HALF_DUPLEX) {
4890 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4891 e1000e_update_phy_stats(adapter);
4893 adapter->stats.scc += er32(SCC);
4894 adapter->stats.ecol += er32(ECOL);
4895 adapter->stats.mcc += er32(MCC);
4896 adapter->stats.latecol += er32(LATECOL);
4897 adapter->stats.dc += er32(DC);
4899 hw->mac.collision_delta = er32(COLC);
4901 if ((hw->mac.type != e1000_82574) &&
4902 (hw->mac.type != e1000_82583))
4903 adapter->stats.tncrs += er32(TNCRS);
4905 adapter->stats.colc += hw->mac.collision_delta;
4908 adapter->stats.xonrxc += er32(XONRXC);
4909 adapter->stats.xontxc += er32(XONTXC);
4910 adapter->stats.xoffrxc += er32(XOFFRXC);
4911 adapter->stats.xofftxc += er32(XOFFTXC);
4912 adapter->stats.gptc += er32(GPTC);
4913 adapter->stats.gotc += er32(GOTCL);
4914 er32(GOTCH); /* Clear gotc */
4915 adapter->stats.rnbc += er32(RNBC);
4916 adapter->stats.ruc += er32(RUC);
4918 adapter->stats.mptc += er32(MPTC);
4919 adapter->stats.bptc += er32(BPTC);
4921 /* used for adaptive IFS */
4923 hw->mac.tx_packet_delta = er32(TPT);
4924 adapter->stats.tpt += hw->mac.tx_packet_delta;
4926 adapter->stats.algnerrc += er32(ALGNERRC);
4927 adapter->stats.rxerrc += er32(RXERRC);
4928 adapter->stats.cexterr += er32(CEXTERR);
4929 adapter->stats.tsctc += er32(TSCTC);
4930 adapter->stats.tsctfc += er32(TSCTFC);
4932 /* Fill out the OS statistics structure */
4933 netdev->stats.multicast = adapter->stats.mprc;
4934 netdev->stats.collisions = adapter->stats.colc;
4938 /* RLEC on some newer hardware can be incorrect so build
4939 * our own version based on RUC and ROC
4941 netdev->stats.rx_errors = adapter->stats.rxerrc +
4942 adapter->stats.crcerrs + adapter->stats.algnerrc +
4943 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4944 netdev->stats.rx_length_errors = adapter->stats.ruc +
4946 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4947 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4948 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4951 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4952 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4953 netdev->stats.tx_window_errors = adapter->stats.latecol;
4954 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4956 /* Tx Dropped needs to be maintained elsewhere */
4958 /* Management Stats */
4959 adapter->stats.mgptc += er32(MGTPTC);
4960 adapter->stats.mgprc += er32(MGTPRC);
4961 adapter->stats.mgpdc += er32(MGTPDC);
4963 /* Correctable ECC Errors */
4964 if ((hw->mac.type == e1000_pch_lpt) ||
4965 (hw->mac.type == e1000_pch_spt)) {
4966 u32 pbeccsts = er32(PBECCSTS);
4968 adapter->corr_errors +=
4969 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4970 adapter->uncorr_errors +=
4971 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4972 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
4977 * e1000_phy_read_status - Update the PHY register status snapshot
4978 * @adapter: board private structure
4980 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4982 struct e1000_hw *hw = &adapter->hw;
4983 struct e1000_phy_regs *phy = &adapter->phy_regs;
4985 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
4986 (er32(STATUS) & E1000_STATUS_LU) &&
4987 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4990 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
4991 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
4992 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
4993 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
4994 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
4995 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
4996 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
4997 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
4999 e_warn("Error reading PHY register\n");
5001 /* Do not read PHY registers if link is not up
5002 * Set values to typical power-on defaults
5004 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5005 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5006 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5008 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5009 ADVERTISE_ALL | ADVERTISE_CSMA);
5011 phy->expansion = EXPANSION_ENABLENPAGE;
5012 phy->ctrl1000 = ADVERTISE_1000FULL;
5014 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5018 static void e1000_print_link_info(struct e1000_adapter *adapter)
5020 struct e1000_hw *hw = &adapter->hw;
5021 u32 ctrl = er32(CTRL);
5023 /* Link status message must follow this format for user tools */
5024 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5025 adapter->netdev->name, adapter->link_speed,
5026 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5027 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5028 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5029 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5032 static bool e1000e_has_link(struct e1000_adapter *adapter)
5034 struct e1000_hw *hw = &adapter->hw;
5035 bool link_active = false;
5038 /* get_link_status is set on LSC (link status) interrupt or
5039 * Rx sequence error interrupt. get_link_status will stay
5040 * false until the check_for_link establishes link
5041 * for copper adapters ONLY
5043 switch (hw->phy.media_type) {
5044 case e1000_media_type_copper:
5045 if (hw->mac.get_link_status) {
5046 ret_val = hw->mac.ops.check_for_link(hw);
5047 link_active = !hw->mac.get_link_status;
5052 case e1000_media_type_fiber:
5053 ret_val = hw->mac.ops.check_for_link(hw);
5054 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5056 case e1000_media_type_internal_serdes:
5057 ret_val = hw->mac.ops.check_for_link(hw);
5058 link_active = adapter->hw.mac.serdes_has_link;
5061 case e1000_media_type_unknown:
5065 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5066 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5067 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5068 e_info("Gigabit has been disabled, downgrading speed\n");
5074 static void e1000e_enable_receives(struct e1000_adapter *adapter)
5076 /* make sure the receive unit is started */
5077 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5078 (adapter->flags & FLAG_RESTART_NOW)) {
5079 struct e1000_hw *hw = &adapter->hw;
5080 u32 rctl = er32(RCTL);
5082 ew32(RCTL, rctl | E1000_RCTL_EN);
5083 adapter->flags &= ~FLAG_RESTART_NOW;
5087 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5089 struct e1000_hw *hw = &adapter->hw;
5091 /* With 82574 controllers, PHY needs to be checked periodically
5092 * for hung state and reset, if two calls return true
5094 if (e1000_check_phy_82574(hw))
5095 adapter->phy_hang_count++;
5097 adapter->phy_hang_count = 0;
5099 if (adapter->phy_hang_count > 1) {
5100 adapter->phy_hang_count = 0;
5101 e_dbg("PHY appears hung - resetting\n");
5102 schedule_work(&adapter->reset_task);
5107 * e1000_watchdog - Timer Call-back
5108 * @data: pointer to adapter cast into an unsigned long
5110 static void e1000_watchdog(unsigned long data)
5112 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
5114 /* Do the rest outside of interrupt context */
5115 schedule_work(&adapter->watchdog_task);
5117 /* TODO: make this use queue_delayed_work() */
5120 static void e1000_watchdog_task(struct work_struct *work)
5122 struct e1000_adapter *adapter = container_of(work,
5123 struct e1000_adapter,
5125 struct net_device *netdev = adapter->netdev;
5126 struct e1000_mac_info *mac = &adapter->hw.mac;
5127 struct e1000_phy_info *phy = &adapter->hw.phy;
5128 struct e1000_ring *tx_ring = adapter->tx_ring;
5129 struct e1000_hw *hw = &adapter->hw;
5132 if (test_bit(__E1000_DOWN, &adapter->state))
5135 link = e1000e_has_link(adapter);
5136 if ((netif_carrier_ok(netdev)) && link) {
5137 /* Cancel scheduled suspend requests. */
5138 pm_runtime_resume(netdev->dev.parent);
5140 e1000e_enable_receives(adapter);
5144 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5145 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5146 e1000_update_mng_vlan(adapter);
5149 if (!netif_carrier_ok(netdev)) {
5152 /* Cancel scheduled suspend requests. */
5153 pm_runtime_resume(netdev->dev.parent);
5155 /* update snapshot of PHY registers on LSC */
5156 e1000_phy_read_status(adapter);
5157 mac->ops.get_link_up_info(&adapter->hw,
5158 &adapter->link_speed,
5159 &adapter->link_duplex);
5160 e1000_print_link_info(adapter);
5162 /* check if SmartSpeed worked */
5163 e1000e_check_downshift(hw);
5164 if (phy->speed_downgraded)
5166 "Link Speed was downgraded by SmartSpeed\n");
5168 /* On supported PHYs, check for duplex mismatch only
5169 * if link has autonegotiated at 10/100 half
5171 if ((hw->phy.type == e1000_phy_igp_3 ||
5172 hw->phy.type == e1000_phy_bm) &&
5174 (adapter->link_speed == SPEED_10 ||
5175 adapter->link_speed == SPEED_100) &&
5176 (adapter->link_duplex == HALF_DUPLEX)) {
5179 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5181 if (!(autoneg_exp & EXPANSION_NWAY))
5182 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5185 /* adjust timeout factor according to speed/duplex */
5186 adapter->tx_timeout_factor = 1;
5187 switch (adapter->link_speed) {
5190 adapter->tx_timeout_factor = 16;
5194 adapter->tx_timeout_factor = 10;
5198 /* workaround: re-program speed mode bit after
5201 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5205 tarc0 = er32(TARC(0));
5206 tarc0 &= ~SPEED_MODE_BIT;
5207 ew32(TARC(0), tarc0);
5210 /* disable TSO for pcie and 10/100 speeds, to avoid
5211 * some hardware issues
5213 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5214 switch (adapter->link_speed) {
5217 e_info("10/100 speed: disabling TSO\n");
5218 netdev->features &= ~NETIF_F_TSO;
5219 netdev->features &= ~NETIF_F_TSO6;
5222 netdev->features |= NETIF_F_TSO;
5223 netdev->features |= NETIF_F_TSO6;
5231 /* enable transmits in the hardware, need to do this
5232 * after setting TARC(0)
5235 tctl |= E1000_TCTL_EN;
5238 /* Perform any post-link-up configuration before
5239 * reporting link up.
5241 if (phy->ops.cfg_on_link_up)
5242 phy->ops.cfg_on_link_up(hw);
5244 netif_carrier_on(netdev);
5246 if (!test_bit(__E1000_DOWN, &adapter->state))
5247 mod_timer(&adapter->phy_info_timer,
5248 round_jiffies(jiffies + 2 * HZ));
5251 if (netif_carrier_ok(netdev)) {
5252 adapter->link_speed = 0;
5253 adapter->link_duplex = 0;
5254 /* Link status message must follow this format */
5255 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5256 netif_carrier_off(netdev);
5257 if (!test_bit(__E1000_DOWN, &adapter->state))
5258 mod_timer(&adapter->phy_info_timer,
5259 round_jiffies(jiffies + 2 * HZ));
5261 /* 8000ES2LAN requires a Rx packet buffer work-around
5262 * on link down event; reset the controller to flush
5263 * the Rx packet buffer.
5265 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5266 adapter->flags |= FLAG_RESTART_NOW;
5268 pm_schedule_suspend(netdev->dev.parent,
5274 spin_lock(&adapter->stats64_lock);
5275 e1000e_update_stats(adapter);
5277 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5278 adapter->tpt_old = adapter->stats.tpt;
5279 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5280 adapter->colc_old = adapter->stats.colc;
5282 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5283 adapter->gorc_old = adapter->stats.gorc;
5284 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5285 adapter->gotc_old = adapter->stats.gotc;
5286 spin_unlock(&adapter->stats64_lock);
5288 /* If the link is lost the controller stops DMA, but
5289 * if there is queued Tx work it cannot be done. So
5290 * reset the controller to flush the Tx packet buffers.
5292 if (!netif_carrier_ok(netdev) &&
5293 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5294 adapter->flags |= FLAG_RESTART_NOW;
5296 /* If reset is necessary, do it outside of interrupt context. */
5297 if (adapter->flags & FLAG_RESTART_NOW) {
5298 schedule_work(&adapter->reset_task);
5299 /* return immediately since reset is imminent */
5303 e1000e_update_adaptive(&adapter->hw);
5305 /* Simple mode for Interrupt Throttle Rate (ITR) */
5306 if (adapter->itr_setting == 4) {
5307 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5308 * Total asymmetrical Tx or Rx gets ITR=8000;
5309 * everyone else is between 2000-8000.
5311 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5312 u32 dif = (adapter->gotc > adapter->gorc ?
5313 adapter->gotc - adapter->gorc :
5314 adapter->gorc - adapter->gotc) / 10000;
5315 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5317 e1000e_write_itr(adapter, itr);
5320 /* Cause software interrupt to ensure Rx ring is cleaned */
5321 if (adapter->msix_entries)
5322 ew32(ICS, adapter->rx_ring->ims_val);
5324 ew32(ICS, E1000_ICS_RXDMT0);
5326 /* flush pending descriptors to memory before detecting Tx hang */
5327 e1000e_flush_descriptors(adapter);
5329 /* Force detection of hung controller every watchdog period */
5330 adapter->detect_tx_hung = true;
5332 /* With 82571 controllers, LAA may be overwritten due to controller
5333 * reset from the other port. Set the appropriate LAA in RAR[0]
5335 if (e1000e_get_laa_state_82571(hw))
5336 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5338 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5339 e1000e_check_82574_phy_workaround(adapter);
5341 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5342 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5343 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5344 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5346 adapter->rx_hwtstamp_cleared++;
5348 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5352 /* Reset the timer */
5353 if (!test_bit(__E1000_DOWN, &adapter->state))
5354 mod_timer(&adapter->watchdog_timer,
5355 round_jiffies(jiffies + 2 * HZ));
5358 #define E1000_TX_FLAGS_CSUM 0x00000001
5359 #define E1000_TX_FLAGS_VLAN 0x00000002
5360 #define E1000_TX_FLAGS_TSO 0x00000004
5361 #define E1000_TX_FLAGS_IPV4 0x00000008
5362 #define E1000_TX_FLAGS_NO_FCS 0x00000010
5363 #define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5364 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5365 #define E1000_TX_FLAGS_VLAN_SHIFT 16
5367 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5370 struct e1000_context_desc *context_desc;
5371 struct e1000_buffer *buffer_info;
5375 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5378 if (!skb_is_gso(skb))
5381 err = skb_cow_head(skb, 0);
5385 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5386 mss = skb_shinfo(skb)->gso_size;
5387 if (protocol == htons(ETH_P_IP)) {
5388 struct iphdr *iph = ip_hdr(skb);
5391 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5393 cmd_length = E1000_TXD_CMD_IP;
5394 ipcse = skb_transport_offset(skb) - 1;
5395 } else if (skb_is_gso_v6(skb)) {
5396 ipv6_hdr(skb)->payload_len = 0;
5397 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5398 &ipv6_hdr(skb)->daddr,
5402 ipcss = skb_network_offset(skb);
5403 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5404 tucss = skb_transport_offset(skb);
5405 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5407 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5408 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5410 i = tx_ring->next_to_use;
5411 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5412 buffer_info = &tx_ring->buffer_info[i];
5414 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5415 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5416 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5417 context_desc->upper_setup.tcp_fields.tucss = tucss;
5418 context_desc->upper_setup.tcp_fields.tucso = tucso;
5419 context_desc->upper_setup.tcp_fields.tucse = 0;
5420 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5421 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5422 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5424 buffer_info->time_stamp = jiffies;
5425 buffer_info->next_to_watch = i;
5428 if (i == tx_ring->count)
5430 tx_ring->next_to_use = i;
5435 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5438 struct e1000_adapter *adapter = tx_ring->adapter;
5439 struct e1000_context_desc *context_desc;
5440 struct e1000_buffer *buffer_info;
5443 u32 cmd_len = E1000_TXD_CMD_DEXT;
5445 if (skb->ip_summed != CHECKSUM_PARTIAL)
5449 case cpu_to_be16(ETH_P_IP):
5450 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5451 cmd_len |= E1000_TXD_CMD_TCP;
5453 case cpu_to_be16(ETH_P_IPV6):
5454 /* XXX not handling all IPV6 headers */
5455 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5456 cmd_len |= E1000_TXD_CMD_TCP;
5459 if (unlikely(net_ratelimit()))
5460 e_warn("checksum_partial proto=%x!\n",
5461 be16_to_cpu(protocol));
5465 css = skb_checksum_start_offset(skb);
5467 i = tx_ring->next_to_use;
5468 buffer_info = &tx_ring->buffer_info[i];
5469 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5471 context_desc->lower_setup.ip_config = 0;
5472 context_desc->upper_setup.tcp_fields.tucss = css;
5473 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5474 context_desc->upper_setup.tcp_fields.tucse = 0;
5475 context_desc->tcp_seg_setup.data = 0;
5476 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5478 buffer_info->time_stamp = jiffies;
5479 buffer_info->next_to_watch = i;
5482 if (i == tx_ring->count)
5484 tx_ring->next_to_use = i;
5489 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5490 unsigned int first, unsigned int max_per_txd,
5491 unsigned int nr_frags)
5493 struct e1000_adapter *adapter = tx_ring->adapter;
5494 struct pci_dev *pdev = adapter->pdev;
5495 struct e1000_buffer *buffer_info;
5496 unsigned int len = skb_headlen(skb);
5497 unsigned int offset = 0, size, count = 0, i;
5498 unsigned int f, bytecount, segs;
5500 i = tx_ring->next_to_use;
5503 buffer_info = &tx_ring->buffer_info[i];
5504 size = min(len, max_per_txd);
5506 buffer_info->length = size;
5507 buffer_info->time_stamp = jiffies;
5508 buffer_info->next_to_watch = i;
5509 buffer_info->dma = dma_map_single(&pdev->dev,
5511 size, DMA_TO_DEVICE);
5512 buffer_info->mapped_as_page = false;
5513 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5522 if (i == tx_ring->count)
5527 for (f = 0; f < nr_frags; f++) {
5528 const struct skb_frag_struct *frag;
5530 frag = &skb_shinfo(skb)->frags[f];
5531 len = skb_frag_size(frag);
5536 if (i == tx_ring->count)
5539 buffer_info = &tx_ring->buffer_info[i];
5540 size = min(len, max_per_txd);
5542 buffer_info->length = size;
5543 buffer_info->time_stamp = jiffies;
5544 buffer_info->next_to_watch = i;
5545 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5548 buffer_info->mapped_as_page = true;
5549 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5558 segs = skb_shinfo(skb)->gso_segs ? : 1;
5559 /* multiply data chunks by size of headers */
5560 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5562 tx_ring->buffer_info[i].skb = skb;
5563 tx_ring->buffer_info[i].segs = segs;
5564 tx_ring->buffer_info[i].bytecount = bytecount;
5565 tx_ring->buffer_info[first].next_to_watch = i;
5570 dev_err(&pdev->dev, "Tx DMA map failed\n");
5571 buffer_info->dma = 0;
5577 i += tx_ring->count;
5579 buffer_info = &tx_ring->buffer_info[i];
5580 e1000_put_txbuf(tx_ring, buffer_info);
5586 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5588 struct e1000_adapter *adapter = tx_ring->adapter;
5589 struct e1000_tx_desc *tx_desc = NULL;
5590 struct e1000_buffer *buffer_info;
5591 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5594 if (tx_flags & E1000_TX_FLAGS_TSO) {
5595 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5597 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5599 if (tx_flags & E1000_TX_FLAGS_IPV4)
5600 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5603 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5604 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5605 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5608 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5609 txd_lower |= E1000_TXD_CMD_VLE;
5610 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5613 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5614 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5616 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5617 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5618 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5621 i = tx_ring->next_to_use;
5624 buffer_info = &tx_ring->buffer_info[i];
5625 tx_desc = E1000_TX_DESC(*tx_ring, i);
5626 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5627 tx_desc->lower.data = cpu_to_le32(txd_lower |
5628 buffer_info->length);
5629 tx_desc->upper.data = cpu_to_le32(txd_upper);
5632 if (i == tx_ring->count)
5634 } while (--count > 0);
5636 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5638 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5639 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5640 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5642 /* Force memory writes to complete before letting h/w
5643 * know there are new descriptors to fetch. (Only
5644 * applicable for weak-ordered memory model archs,
5649 tx_ring->next_to_use = i;
5652 #define MINIMUM_DHCP_PACKET_SIZE 282
5653 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5654 struct sk_buff *skb)
5656 struct e1000_hw *hw = &adapter->hw;
5659 if (skb_vlan_tag_present(skb) &&
5660 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5661 (adapter->hw.mng_cookie.status &
5662 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5665 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5668 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5672 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5675 if (ip->protocol != IPPROTO_UDP)
5678 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5679 if (ntohs(udp->dest) != 67)
5682 offset = (u8 *)udp + 8 - skb->data;
5683 length = skb->len - offset;
5684 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5690 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5692 struct e1000_adapter *adapter = tx_ring->adapter;
5694 netif_stop_queue(adapter->netdev);
5695 /* Herbert's original patch had:
5696 * smp_mb__after_netif_stop_queue();
5697 * but since that doesn't exist yet, just open code it.
5701 /* We need to check again in a case another CPU has just
5702 * made room available.
5704 if (e1000_desc_unused(tx_ring) < size)
5708 netif_start_queue(adapter->netdev);
5709 ++adapter->restart_queue;
5713 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5715 BUG_ON(size > tx_ring->count);
5717 if (e1000_desc_unused(tx_ring) >= size)
5719 return __e1000_maybe_stop_tx(tx_ring, size);
5722 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5723 struct net_device *netdev)
5725 struct e1000_adapter *adapter = netdev_priv(netdev);
5726 struct e1000_ring *tx_ring = adapter->tx_ring;
5728 unsigned int tx_flags = 0;
5729 unsigned int len = skb_headlen(skb);
5730 unsigned int nr_frags;
5735 __be16 protocol = vlan_get_protocol(skb);
5737 if (test_bit(__E1000_DOWN, &adapter->state)) {
5738 dev_kfree_skb_any(skb);
5739 return NETDEV_TX_OK;
5742 if (skb->len <= 0) {
5743 dev_kfree_skb_any(skb);
5744 return NETDEV_TX_OK;
5747 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5748 * pad skb in order to meet this minimum size requirement
5750 if (skb_put_padto(skb, 17))
5751 return NETDEV_TX_OK;
5753 mss = skb_shinfo(skb)->gso_size;
5757 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5758 * points to just header, pull a few bytes of payload from
5759 * frags into skb->data
5761 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5762 /* we do this workaround for ES2LAN, but it is un-necessary,
5763 * avoiding it could save a lot of cycles
5765 if (skb->data_len && (hdr_len == len)) {
5766 unsigned int pull_size;
5768 pull_size = min_t(unsigned int, 4, skb->data_len);
5769 if (!__pskb_pull_tail(skb, pull_size)) {
5770 e_err("__pskb_pull_tail failed.\n");
5771 dev_kfree_skb_any(skb);
5772 return NETDEV_TX_OK;
5774 len = skb_headlen(skb);
5778 /* reserve a descriptor for the offload context */
5779 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5783 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5785 nr_frags = skb_shinfo(skb)->nr_frags;
5786 for (f = 0; f < nr_frags; f++)
5787 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5788 adapter->tx_fifo_limit);
5790 if (adapter->hw.mac.tx_pkt_filtering)
5791 e1000_transfer_dhcp_info(adapter, skb);
5793 /* need: count + 2 desc gap to keep tail from touching
5794 * head, otherwise try next time
5796 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5797 return NETDEV_TX_BUSY;
5799 if (skb_vlan_tag_present(skb)) {
5800 tx_flags |= E1000_TX_FLAGS_VLAN;
5801 tx_flags |= (skb_vlan_tag_get(skb) <<
5802 E1000_TX_FLAGS_VLAN_SHIFT);
5805 first = tx_ring->next_to_use;
5807 tso = e1000_tso(tx_ring, skb, protocol);
5809 dev_kfree_skb_any(skb);
5810 return NETDEV_TX_OK;
5814 tx_flags |= E1000_TX_FLAGS_TSO;
5815 else if (e1000_tx_csum(tx_ring, skb, protocol))
5816 tx_flags |= E1000_TX_FLAGS_CSUM;
5818 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5819 * 82571 hardware supports TSO capabilities for IPv6 as well...
5820 * no longer assume, we must.
5822 if (protocol == htons(ETH_P_IP))
5823 tx_flags |= E1000_TX_FLAGS_IPV4;
5825 if (unlikely(skb->no_fcs))
5826 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5828 /* if count is 0 then mapping error has occurred */
5829 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5832 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5833 (adapter->flags & FLAG_HAS_HW_TIMESTAMP) &&
5834 !adapter->tx_hwtstamp_skb) {
5835 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5836 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5837 adapter->tx_hwtstamp_skb = skb_get(skb);
5838 adapter->tx_hwtstamp_start = jiffies;
5839 schedule_work(&adapter->tx_hwtstamp_work);
5841 skb_tx_timestamp(skb);
5844 netdev_sent_queue(netdev, skb->len);
5845 e1000_tx_queue(tx_ring, tx_flags, count);
5846 /* Make sure there is space in the ring for the next send. */
5847 e1000_maybe_stop_tx(tx_ring,
5849 DIV_ROUND_UP(PAGE_SIZE,
5850 adapter->tx_fifo_limit) + 2));
5852 if (!skb->xmit_more ||
5853 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5854 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5855 e1000e_update_tdt_wa(tx_ring,
5856 tx_ring->next_to_use);
5858 writel(tx_ring->next_to_use, tx_ring->tail);
5860 /* we need this if more than one processor can write
5861 * to our tail at a time, it synchronizes IO on
5867 dev_kfree_skb_any(skb);
5868 tx_ring->buffer_info[first].time_stamp = 0;
5869 tx_ring->next_to_use = first;
5872 return NETDEV_TX_OK;
5876 * e1000_tx_timeout - Respond to a Tx Hang
5877 * @netdev: network interface device structure
5879 static void e1000_tx_timeout(struct net_device *netdev)
5881 struct e1000_adapter *adapter = netdev_priv(netdev);
5883 /* Do the reset outside of interrupt context */
5884 adapter->tx_timeout_count++;
5885 schedule_work(&adapter->reset_task);
5888 static void e1000_reset_task(struct work_struct *work)
5890 struct e1000_adapter *adapter;
5891 adapter = container_of(work, struct e1000_adapter, reset_task);
5893 /* don't run the task if already down */
5894 if (test_bit(__E1000_DOWN, &adapter->state))
5897 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5898 e1000e_dump(adapter);
5899 e_err("Reset adapter unexpectedly\n");
5901 e1000e_reinit_locked(adapter);
5905 * e1000_get_stats64 - Get System Network Statistics
5906 * @netdev: network interface device structure
5907 * @stats: rtnl_link_stats64 pointer
5909 * Returns the address of the device statistics structure.
5911 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5912 struct rtnl_link_stats64 *stats)
5914 struct e1000_adapter *adapter = netdev_priv(netdev);
5916 memset(stats, 0, sizeof(struct rtnl_link_stats64));
5917 spin_lock(&adapter->stats64_lock);
5918 e1000e_update_stats(adapter);
5919 /* Fill out the OS statistics structure */
5920 stats->rx_bytes = adapter->stats.gorc;
5921 stats->rx_packets = adapter->stats.gprc;
5922 stats->tx_bytes = adapter->stats.gotc;
5923 stats->tx_packets = adapter->stats.gptc;
5924 stats->multicast = adapter->stats.mprc;
5925 stats->collisions = adapter->stats.colc;
5929 /* RLEC on some newer hardware can be incorrect so build
5930 * our own version based on RUC and ROC
5932 stats->rx_errors = adapter->stats.rxerrc +
5933 adapter->stats.crcerrs + adapter->stats.algnerrc +
5934 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5935 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5936 stats->rx_crc_errors = adapter->stats.crcerrs;
5937 stats->rx_frame_errors = adapter->stats.algnerrc;
5938 stats->rx_missed_errors = adapter->stats.mpc;
5941 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5942 stats->tx_aborted_errors = adapter->stats.ecol;
5943 stats->tx_window_errors = adapter->stats.latecol;
5944 stats->tx_carrier_errors = adapter->stats.tncrs;
5946 /* Tx Dropped needs to be maintained elsewhere */
5948 spin_unlock(&adapter->stats64_lock);
5953 * e1000_change_mtu - Change the Maximum Transfer Unit
5954 * @netdev: network interface device structure
5955 * @new_mtu: new value for maximum frame size
5957 * Returns 0 on success, negative on failure
5959 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5961 struct e1000_adapter *adapter = netdev_priv(netdev);
5962 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
5964 /* Jumbo frame support */
5965 if ((max_frame > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) &&
5966 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5967 e_err("Jumbo Frames not supported.\n");
5971 /* Supported frame sizes */
5972 if ((new_mtu < (VLAN_ETH_ZLEN + ETH_FCS_LEN)) ||
5973 (max_frame > adapter->max_hw_frame_size)) {
5974 e_err("Unsupported MTU setting\n");
5978 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5979 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5980 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5981 (new_mtu > ETH_DATA_LEN)) {
5982 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5986 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5987 usleep_range(1000, 2000);
5988 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5989 adapter->max_frame_size = max_frame;
5990 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5991 netdev->mtu = new_mtu;
5993 pm_runtime_get_sync(netdev->dev.parent);
5995 if (netif_running(netdev))
5996 e1000e_down(adapter, true);
5998 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5999 * means we reserve 2 more, this pushes us to allocate from the next
6001 * i.e. RXBUFFER_2048 --> size-4096 slab
6002 * However with the new *_jumbo_rx* routines, jumbo receives will use
6006 if (max_frame <= 2048)
6007 adapter->rx_buffer_len = 2048;
6009 adapter->rx_buffer_len = 4096;
6011 /* adjust allocation if LPE protects us, and we aren't using SBP */
6012 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6013 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6015 if (netif_running(netdev))
6018 e1000e_reset(adapter);
6020 pm_runtime_put_sync(netdev->dev.parent);
6022 clear_bit(__E1000_RESETTING, &adapter->state);
6027 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6030 struct e1000_adapter *adapter = netdev_priv(netdev);
6031 struct mii_ioctl_data *data = if_mii(ifr);
6033 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6038 data->phy_id = adapter->hw.phy.addr;
6041 e1000_phy_read_status(adapter);
6043 switch (data->reg_num & 0x1F) {
6045 data->val_out = adapter->phy_regs.bmcr;
6048 data->val_out = adapter->phy_regs.bmsr;
6051 data->val_out = (adapter->hw.phy.id >> 16);
6054 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6057 data->val_out = adapter->phy_regs.advertise;
6060 data->val_out = adapter->phy_regs.lpa;
6063 data->val_out = adapter->phy_regs.expansion;
6066 data->val_out = adapter->phy_regs.ctrl1000;
6069 data->val_out = adapter->phy_regs.stat1000;
6072 data->val_out = adapter->phy_regs.estatus;
6086 * e1000e_hwtstamp_ioctl - control hardware time stamping
6087 * @netdev: network interface device structure
6088 * @ifreq: interface request
6090 * Outgoing time stamping can be enabled and disabled. Play nice and
6091 * disable it when requested, although it shouldn't cause any overhead
6092 * when no packet needs it. At most one packet in the queue may be
6093 * marked for time stamping, otherwise it would be impossible to tell
6094 * for sure to which packet the hardware time stamp belongs.
6096 * Incoming time stamping has to be configured via the hardware filters.
6097 * Not all combinations are supported, in particular event type has to be
6098 * specified. Matching the kind of event packet is not supported, with the
6099 * exception of "all V2 events regardless of level 2 or 4".
6101 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6103 struct e1000_adapter *adapter = netdev_priv(netdev);
6104 struct hwtstamp_config config;
6107 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6110 ret_val = e1000e_config_hwtstamp(adapter, &config);
6114 switch (config.rx_filter) {
6115 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6116 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6117 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6118 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6119 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6120 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6121 /* With V2 type filters which specify a Sync or Delay Request,
6122 * Path Delay Request/Response messages are also time stamped
6123 * by hardware so notify the caller the requested packets plus
6124 * some others are time stamped.
6126 config.rx_filter = HWTSTAMP_FILTER_SOME;
6132 return copy_to_user(ifr->ifr_data, &config,
6133 sizeof(config)) ? -EFAULT : 0;
6136 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6138 struct e1000_adapter *adapter = netdev_priv(netdev);
6140 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6141 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6144 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6150 return e1000_mii_ioctl(netdev, ifr, cmd);
6152 return e1000e_hwtstamp_set(netdev, ifr);
6154 return e1000e_hwtstamp_get(netdev, ifr);
6160 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6162 struct e1000_hw *hw = &adapter->hw;
6163 u32 i, mac_reg, wuc;
6164 u16 phy_reg, wuc_enable;
6167 /* copy MAC RARs to PHY RARs */
6168 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6170 retval = hw->phy.ops.acquire(hw);
6172 e_err("Could not acquire PHY\n");
6176 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6177 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6181 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6182 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6183 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6184 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6185 (u16)(mac_reg & 0xFFFF));
6186 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6187 (u16)((mac_reg >> 16) & 0xFFFF));
6190 /* configure PHY Rx Control register */
6191 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6192 mac_reg = er32(RCTL);
6193 if (mac_reg & E1000_RCTL_UPE)
6194 phy_reg |= BM_RCTL_UPE;
6195 if (mac_reg & E1000_RCTL_MPE)
6196 phy_reg |= BM_RCTL_MPE;
6197 phy_reg &= ~(BM_RCTL_MO_MASK);
6198 if (mac_reg & E1000_RCTL_MO_3)
6199 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6200 << BM_RCTL_MO_SHIFT);
6201 if (mac_reg & E1000_RCTL_BAM)
6202 phy_reg |= BM_RCTL_BAM;
6203 if (mac_reg & E1000_RCTL_PMCF)
6204 phy_reg |= BM_RCTL_PMCF;
6205 mac_reg = er32(CTRL);
6206 if (mac_reg & E1000_CTRL_RFCE)
6207 phy_reg |= BM_RCTL_RFCE;
6208 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6210 wuc = E1000_WUC_PME_EN;
6211 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6212 wuc |= E1000_WUC_APME;
6214 /* enable PHY wakeup in MAC register */
6216 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6217 E1000_WUC_PME_STATUS | wuc));
6219 /* configure and enable PHY wakeup in PHY registers */
6220 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6221 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6223 /* activate PHY wakeup */
6224 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6225 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6227 e_err("Could not set PHY Host Wakeup bit\n");
6229 hw->phy.ops.release(hw);
6234 static void e1000e_flush_lpic(struct pci_dev *pdev)
6236 struct net_device *netdev = pci_get_drvdata(pdev);
6237 struct e1000_adapter *adapter = netdev_priv(netdev);
6238 struct e1000_hw *hw = &adapter->hw;
6241 pm_runtime_get_sync(netdev->dev.parent);
6243 ret_val = hw->phy.ops.acquire(hw);
6247 pr_info("EEE TX LPI TIMER: %08X\n",
6248 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6250 hw->phy.ops.release(hw);
6253 pm_runtime_put_sync(netdev->dev.parent);
6256 static int e1000e_pm_freeze(struct device *dev)
6258 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6259 struct e1000_adapter *adapter = netdev_priv(netdev);
6261 netif_device_detach(netdev);
6263 if (netif_running(netdev)) {
6264 int count = E1000_CHECK_RESET_COUNT;
6266 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6267 usleep_range(10000, 20000);
6269 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6271 /* Quiesce the device without resetting the hardware */
6272 e1000e_down(adapter, false);
6273 e1000_free_irq(adapter);
6275 e1000e_reset_interrupt_capability(adapter);
6277 /* Allow time for pending master requests to run */
6278 e1000e_disable_pcie_master(&adapter->hw);
6283 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6285 struct net_device *netdev = pci_get_drvdata(pdev);
6286 struct e1000_adapter *adapter = netdev_priv(netdev);
6287 struct e1000_hw *hw = &adapter->hw;
6288 u32 ctrl, ctrl_ext, rctl, status;
6289 /* Runtime suspend should only enable wakeup for link changes */
6290 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6293 status = er32(STATUS);
6294 if (status & E1000_STATUS_LU)
6295 wufc &= ~E1000_WUFC_LNKC;
6298 e1000_setup_rctl(adapter);
6299 e1000e_set_rx_mode(netdev);
6301 /* turn on all-multi mode if wake on multicast is enabled */
6302 if (wufc & E1000_WUFC_MC) {
6304 rctl |= E1000_RCTL_MPE;
6309 ctrl |= E1000_CTRL_ADVD3WUC;
6310 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6311 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6314 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6315 adapter->hw.phy.media_type ==
6316 e1000_media_type_internal_serdes) {
6317 /* keep the laser running in D3 */
6318 ctrl_ext = er32(CTRL_EXT);
6319 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6320 ew32(CTRL_EXT, ctrl_ext);
6324 e1000e_power_up_phy(adapter);
6326 if (adapter->flags & FLAG_IS_ICH)
6327 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6329 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6330 /* enable wakeup by the PHY */
6331 retval = e1000_init_phy_wakeup(adapter, wufc);
6335 /* enable wakeup by the MAC */
6337 ew32(WUC, E1000_WUC_PME_EN);
6343 e1000_power_down_phy(adapter);
6346 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6347 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6348 } else if ((hw->mac.type == e1000_pch_lpt) ||
6349 (hw->mac.type == e1000_pch_spt)) {
6350 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6351 /* ULP does not support wake from unicast, multicast
6354 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6360 /* Ensure that the appropriate bits are set in LPI_CTRL
6363 if ((hw->phy.type >= e1000_phy_i217) &&
6364 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6367 retval = hw->phy.ops.acquire(hw);
6369 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6372 if (adapter->eee_advert &
6373 hw->dev_spec.ich8lan.eee_lp_ability &
6374 I82579_EEE_100_SUPPORTED)
6375 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6376 if (adapter->eee_advert &
6377 hw->dev_spec.ich8lan.eee_lp_ability &
6378 I82579_EEE_1000_SUPPORTED)
6379 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6381 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6385 hw->phy.ops.release(hw);
6388 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6389 * would have already happened in close and is redundant.
6391 e1000e_release_hw_control(adapter);
6393 pci_clear_master(pdev);
6395 /* The pci-e switch on some quad port adapters will report a
6396 * correctable error when the MAC transitions from D0 to D3. To
6397 * prevent this we need to mask off the correctable errors on the
6398 * downstream port of the pci-e switch.
6400 * We don't have the associated upstream bridge while assigning
6401 * the PCI device into guest. For example, the KVM on power is
6404 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6405 struct pci_dev *us_dev = pdev->bus->self;
6411 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6412 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6413 (devctl & ~PCI_EXP_DEVCTL_CERE));
6415 pci_save_state(pdev);
6416 pci_prepare_to_sleep(pdev);
6418 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6425 * __e1000e_disable_aspm - Disable ASPM states
6426 * @pdev: pointer to PCI device struct
6427 * @state: bit-mask of ASPM states to disable
6428 * @locked: indication if this context holds pci_bus_sem locked.
6430 * Some devices *must* have certain ASPM states disabled per hardware errata.
6432 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6434 struct pci_dev *parent = pdev->bus->self;
6435 u16 aspm_dis_mask = 0;
6436 u16 pdev_aspmc, parent_aspmc;
6439 case PCIE_LINK_STATE_L0S:
6440 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6441 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6442 /* fall-through - can't have L1 without L0s */
6443 case PCIE_LINK_STATE_L1:
6444 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6450 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6451 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6454 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6456 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6459 /* Nothing to do if the ASPM states to be disabled already are */
6460 if (!(pdev_aspmc & aspm_dis_mask) &&
6461 (!parent || !(parent_aspmc & aspm_dis_mask)))
6464 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6465 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6467 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6470 #ifdef CONFIG_PCIEASPM
6472 pci_disable_link_state_locked(pdev, state);
6474 pci_disable_link_state(pdev, state);
6476 /* Double-check ASPM control. If not disabled by the above, the
6477 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6478 * not enabled); override by writing PCI config space directly.
6480 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6481 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6483 if (!(aspm_dis_mask & pdev_aspmc))
6487 /* Both device and parent should have the same ASPM setting.
6488 * Disable ASPM in downstream component first and then upstream.
6490 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6493 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6498 * e1000e_disable_aspm - Disable ASPM states.
6499 * @pdev: pointer to PCI device struct
6500 * @state: bit-mask of ASPM states to disable
6502 * This function acquires the pci_bus_sem!
6503 * Some devices *must* have certain ASPM states disabled per hardware errata.
6505 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6507 __e1000e_disable_aspm(pdev, state, 0);
6511 * e1000e_disable_aspm_locked Disable ASPM states.
6512 * @pdev: pointer to PCI device struct
6513 * @state: bit-mask of ASPM states to disable
6515 * This function must be called with pci_bus_sem acquired!
6516 * Some devices *must* have certain ASPM states disabled per hardware errata.
6518 static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6520 __e1000e_disable_aspm(pdev, state, 1);
6524 static int __e1000_resume(struct pci_dev *pdev)
6526 struct net_device *netdev = pci_get_drvdata(pdev);
6527 struct e1000_adapter *adapter = netdev_priv(netdev);
6528 struct e1000_hw *hw = &adapter->hw;
6529 u16 aspm_disable_flag = 0;
6531 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6532 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6533 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6534 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6535 if (aspm_disable_flag)
6536 e1000e_disable_aspm(pdev, aspm_disable_flag);
6538 pci_set_master(pdev);
6540 if (hw->mac.type >= e1000_pch2lan)
6541 e1000_resume_workarounds_pchlan(&adapter->hw);
6543 e1000e_power_up_phy(adapter);
6545 /* report the system wakeup cause from S3/S4 */
6546 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6549 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6551 e_info("PHY Wakeup cause - %s\n",
6552 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6553 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6554 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6555 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6556 phy_data & E1000_WUS_LNKC ?
6557 "Link Status Change" : "other");
6559 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6561 u32 wus = er32(WUS);
6564 e_info("MAC Wakeup cause - %s\n",
6565 wus & E1000_WUS_EX ? "Unicast Packet" :
6566 wus & E1000_WUS_MC ? "Multicast Packet" :
6567 wus & E1000_WUS_BC ? "Broadcast Packet" :
6568 wus & E1000_WUS_MAG ? "Magic Packet" :
6569 wus & E1000_WUS_LNKC ? "Link Status Change" :
6575 e1000e_reset(adapter);
6577 e1000_init_manageability_pt(adapter);
6579 /* If the controller has AMT, do not set DRV_LOAD until the interface
6580 * is up. For all other cases, let the f/w know that the h/w is now
6581 * under the control of the driver.
6583 if (!(adapter->flags & FLAG_HAS_AMT))
6584 e1000e_get_hw_control(adapter);
6589 #ifdef CONFIG_PM_SLEEP
6590 static int e1000e_pm_thaw(struct device *dev)
6592 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6593 struct e1000_adapter *adapter = netdev_priv(netdev);
6595 e1000e_set_interrupt_capability(adapter);
6596 if (netif_running(netdev)) {
6597 u32 err = e1000_request_irq(adapter);
6605 netif_device_attach(netdev);
6610 static int e1000e_pm_suspend(struct device *dev)
6612 struct pci_dev *pdev = to_pci_dev(dev);
6614 e1000e_flush_lpic(pdev);
6616 e1000e_pm_freeze(dev);
6618 return __e1000_shutdown(pdev, false);
6621 static int e1000e_pm_resume(struct device *dev)
6623 struct pci_dev *pdev = to_pci_dev(dev);
6626 rc = __e1000_resume(pdev);
6630 return e1000e_pm_thaw(dev);
6632 #endif /* CONFIG_PM_SLEEP */
6634 static int e1000e_pm_runtime_idle(struct device *dev)
6636 struct pci_dev *pdev = to_pci_dev(dev);
6637 struct net_device *netdev = pci_get_drvdata(pdev);
6638 struct e1000_adapter *adapter = netdev_priv(netdev);
6641 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6643 if (!e1000e_has_link(adapter)) {
6644 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6645 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6651 static int e1000e_pm_runtime_resume(struct device *dev)
6653 struct pci_dev *pdev = to_pci_dev(dev);
6654 struct net_device *netdev = pci_get_drvdata(pdev);
6655 struct e1000_adapter *adapter = netdev_priv(netdev);
6658 rc = __e1000_resume(pdev);
6662 if (netdev->flags & IFF_UP)
6668 static int e1000e_pm_runtime_suspend(struct device *dev)
6670 struct pci_dev *pdev = to_pci_dev(dev);
6671 struct net_device *netdev = pci_get_drvdata(pdev);
6672 struct e1000_adapter *adapter = netdev_priv(netdev);
6674 if (netdev->flags & IFF_UP) {
6675 int count = E1000_CHECK_RESET_COUNT;
6677 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6678 usleep_range(10000, 20000);
6680 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6682 /* Down the device without resetting the hardware */
6683 e1000e_down(adapter, false);
6686 if (__e1000_shutdown(pdev, true)) {
6687 e1000e_pm_runtime_resume(dev);
6693 #endif /* CONFIG_PM */
6695 static void e1000_shutdown(struct pci_dev *pdev)
6697 e1000e_flush_lpic(pdev);
6699 e1000e_pm_freeze(&pdev->dev);
6701 __e1000_shutdown(pdev, false);
6704 #ifdef CONFIG_NET_POLL_CONTROLLER
6706 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6708 struct net_device *netdev = data;
6709 struct e1000_adapter *adapter = netdev_priv(netdev);
6711 if (adapter->msix_entries) {
6712 int vector, msix_irq;
6715 msix_irq = adapter->msix_entries[vector].vector;
6716 disable_irq(msix_irq);
6717 e1000_intr_msix_rx(msix_irq, netdev);
6718 enable_irq(msix_irq);
6721 msix_irq = adapter->msix_entries[vector].vector;
6722 disable_irq(msix_irq);
6723 e1000_intr_msix_tx(msix_irq, netdev);
6724 enable_irq(msix_irq);
6727 msix_irq = adapter->msix_entries[vector].vector;
6728 disable_irq(msix_irq);
6729 e1000_msix_other(msix_irq, netdev);
6730 enable_irq(msix_irq);
6738 * @netdev: network interface device structure
6740 * Polling 'interrupt' - used by things like netconsole to send skbs
6741 * without having to re-enable interrupts. It's not called while
6742 * the interrupt routine is executing.
6744 static void e1000_netpoll(struct net_device *netdev)
6746 struct e1000_adapter *adapter = netdev_priv(netdev);
6748 switch (adapter->int_mode) {
6749 case E1000E_INT_MODE_MSIX:
6750 e1000_intr_msix(adapter->pdev->irq, netdev);
6752 case E1000E_INT_MODE_MSI:
6753 disable_irq(adapter->pdev->irq);
6754 e1000_intr_msi(adapter->pdev->irq, netdev);
6755 enable_irq(adapter->pdev->irq);
6757 default: /* E1000E_INT_MODE_LEGACY */
6758 disable_irq(adapter->pdev->irq);
6759 e1000_intr(adapter->pdev->irq, netdev);
6760 enable_irq(adapter->pdev->irq);
6767 * e1000_io_error_detected - called when PCI error is detected
6768 * @pdev: Pointer to PCI device
6769 * @state: The current pci connection state
6771 * This function is called after a PCI bus error affecting
6772 * this device has been detected.
6774 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6775 pci_channel_state_t state)
6777 struct net_device *netdev = pci_get_drvdata(pdev);
6778 struct e1000_adapter *adapter = netdev_priv(netdev);
6780 netif_device_detach(netdev);
6782 if (state == pci_channel_io_perm_failure)
6783 return PCI_ERS_RESULT_DISCONNECT;
6785 if (netif_running(netdev))
6786 e1000e_down(adapter, true);
6787 pci_disable_device(pdev);
6789 /* Request a slot slot reset. */
6790 return PCI_ERS_RESULT_NEED_RESET;
6794 * e1000_io_slot_reset - called after the pci bus has been reset.
6795 * @pdev: Pointer to PCI device
6797 * Restart the card from scratch, as if from a cold-boot. Implementation
6798 * resembles the first-half of the e1000e_pm_resume routine.
6800 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6802 struct net_device *netdev = pci_get_drvdata(pdev);
6803 struct e1000_adapter *adapter = netdev_priv(netdev);
6804 struct e1000_hw *hw = &adapter->hw;
6805 u16 aspm_disable_flag = 0;
6807 pci_ers_result_t result;
6809 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6810 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6811 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6812 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6813 if (aspm_disable_flag)
6814 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6816 err = pci_enable_device_mem(pdev);
6819 "Cannot re-enable PCI device after reset.\n");
6820 result = PCI_ERS_RESULT_DISCONNECT;
6822 pdev->state_saved = true;
6823 pci_restore_state(pdev);
6824 pci_set_master(pdev);
6826 pci_enable_wake(pdev, PCI_D3hot, 0);
6827 pci_enable_wake(pdev, PCI_D3cold, 0);
6829 e1000e_reset(adapter);
6831 result = PCI_ERS_RESULT_RECOVERED;
6834 pci_cleanup_aer_uncorrect_error_status(pdev);
6840 * e1000_io_resume - called when traffic can start flowing again.
6841 * @pdev: Pointer to PCI device
6843 * This callback is called when the error recovery driver tells us that
6844 * its OK to resume normal operation. Implementation resembles the
6845 * second-half of the e1000e_pm_resume routine.
6847 static void e1000_io_resume(struct pci_dev *pdev)
6849 struct net_device *netdev = pci_get_drvdata(pdev);
6850 struct e1000_adapter *adapter = netdev_priv(netdev);
6852 e1000_init_manageability_pt(adapter);
6854 if (netif_running(netdev))
6857 netif_device_attach(netdev);
6859 /* If the controller has AMT, do not set DRV_LOAD until the interface
6860 * is up. For all other cases, let the f/w know that the h/w is now
6861 * under the control of the driver.
6863 if (!(adapter->flags & FLAG_HAS_AMT))
6864 e1000e_get_hw_control(adapter);
6867 static void e1000_print_device_info(struct e1000_adapter *adapter)
6869 struct e1000_hw *hw = &adapter->hw;
6870 struct net_device *netdev = adapter->netdev;
6872 u8 pba_str[E1000_PBANUM_LENGTH];
6874 /* print bus type/speed/width info */
6875 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6877 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6881 e_info("Intel(R) PRO/%s Network Connection\n",
6882 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6883 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6884 E1000_PBANUM_LENGTH);
6886 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6887 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6888 hw->mac.type, hw->phy.type, pba_str);
6891 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6893 struct e1000_hw *hw = &adapter->hw;
6897 if (hw->mac.type != e1000_82573)
6900 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6902 if (!ret_val && (!(buf & BIT(0)))) {
6903 /* Deep Smart Power Down (DSPD) */
6904 dev_warn(&adapter->pdev->dev,
6905 "Warning: detected DSPD enabled in EEPROM\n");
6909 static netdev_features_t e1000_fix_features(struct net_device *netdev,
6910 netdev_features_t features)
6912 struct e1000_adapter *adapter = netdev_priv(netdev);
6913 struct e1000_hw *hw = &adapter->hw;
6915 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6916 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6917 features &= ~NETIF_F_RXFCS;
6919 /* Since there is no support for separate Rx/Tx vlan accel
6920 * enable/disable make sure Tx flag is always in same state as Rx.
6922 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6923 features |= NETIF_F_HW_VLAN_CTAG_TX;
6925 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6930 static int e1000_set_features(struct net_device *netdev,
6931 netdev_features_t features)
6933 struct e1000_adapter *adapter = netdev_priv(netdev);
6934 netdev_features_t changed = features ^ netdev->features;
6936 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6937 adapter->flags |= FLAG_TSO_FORCE;
6939 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6940 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6944 if (changed & NETIF_F_RXFCS) {
6945 if (features & NETIF_F_RXFCS) {
6946 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6948 /* We need to take it back to defaults, which might mean
6949 * stripping is still disabled at the adapter level.
6951 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6952 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6954 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6958 netdev->features = features;
6960 if (netif_running(netdev))
6961 e1000e_reinit_locked(adapter);
6963 e1000e_reset(adapter);
6968 static const struct net_device_ops e1000e_netdev_ops = {
6969 .ndo_open = e1000e_open,
6970 .ndo_stop = e1000e_close,
6971 .ndo_start_xmit = e1000_xmit_frame,
6972 .ndo_get_stats64 = e1000e_get_stats64,
6973 .ndo_set_rx_mode = e1000e_set_rx_mode,
6974 .ndo_set_mac_address = e1000_set_mac,
6975 .ndo_change_mtu = e1000_change_mtu,
6976 .ndo_do_ioctl = e1000_ioctl,
6977 .ndo_tx_timeout = e1000_tx_timeout,
6978 .ndo_validate_addr = eth_validate_addr,
6980 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
6981 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
6982 #ifdef CONFIG_NET_POLL_CONTROLLER
6983 .ndo_poll_controller = e1000_netpoll,
6985 .ndo_set_features = e1000_set_features,
6986 .ndo_fix_features = e1000_fix_features,
6987 .ndo_features_check = passthru_features_check,
6991 * e1000_probe - Device Initialization Routine
6992 * @pdev: PCI device information struct
6993 * @ent: entry in e1000_pci_tbl
6995 * Returns 0 on success, negative on failure
6997 * e1000_probe initializes an adapter identified by a pci_dev structure.
6998 * The OS initialization, configuring of the adapter private structure,
6999 * and a hardware reset occur.
7001 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7003 struct net_device *netdev;
7004 struct e1000_adapter *adapter;
7005 struct e1000_hw *hw;
7006 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7007 resource_size_t mmio_start, mmio_len;
7008 resource_size_t flash_start, flash_len;
7009 static int cards_found;
7010 u16 aspm_disable_flag = 0;
7011 int bars, i, err, pci_using_dac;
7012 u16 eeprom_data = 0;
7013 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7016 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7017 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7018 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7019 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7020 if (aspm_disable_flag)
7021 e1000e_disable_aspm(pdev, aspm_disable_flag);
7023 err = pci_enable_device_mem(pdev);
7028 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7032 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7035 "No usable DMA configuration, aborting\n");
7040 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7041 err = pci_request_selected_regions_exclusive(pdev, bars,
7042 e1000e_driver_name);
7046 /* AER (Advanced Error Reporting) hooks */
7047 pci_enable_pcie_error_reporting(pdev);
7049 pci_set_master(pdev);
7050 /* PCI config space info */
7051 err = pci_save_state(pdev);
7053 goto err_alloc_etherdev;
7056 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7058 goto err_alloc_etherdev;
7060 SET_NETDEV_DEV(netdev, &pdev->dev);
7062 netdev->irq = pdev->irq;
7064 pci_set_drvdata(pdev, netdev);
7065 adapter = netdev_priv(netdev);
7067 adapter->netdev = netdev;
7068 adapter->pdev = pdev;
7070 adapter->pba = ei->pba;
7071 adapter->flags = ei->flags;
7072 adapter->flags2 = ei->flags2;
7073 adapter->hw.adapter = adapter;
7074 adapter->hw.mac.type = ei->mac;
7075 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7076 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7078 mmio_start = pci_resource_start(pdev, 0);
7079 mmio_len = pci_resource_len(pdev, 0);
7082 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7083 if (!adapter->hw.hw_addr)
7086 if ((adapter->flags & FLAG_HAS_FLASH) &&
7087 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7088 (hw->mac.type < e1000_pch_spt)) {
7089 flash_start = pci_resource_start(pdev, 1);
7090 flash_len = pci_resource_len(pdev, 1);
7091 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7092 if (!adapter->hw.flash_address)
7096 /* Set default EEE advertisement */
7097 if (adapter->flags2 & FLAG2_HAS_EEE)
7098 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7100 /* construct the net_device struct */
7101 netdev->netdev_ops = &e1000e_netdev_ops;
7102 e1000e_set_ethtool_ops(netdev);
7103 netdev->watchdog_timeo = 5 * HZ;
7104 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7105 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7107 netdev->mem_start = mmio_start;
7108 netdev->mem_end = mmio_start + mmio_len;
7110 adapter->bd_number = cards_found++;
7112 e1000e_check_options(adapter);
7114 /* setup adapter struct */
7115 err = e1000_sw_init(adapter);
7119 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7120 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7121 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7123 err = ei->get_variants(adapter);
7127 if ((adapter->flags & FLAG_IS_ICH) &&
7128 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7129 (hw->mac.type < e1000_pch_spt))
7130 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7132 hw->mac.ops.get_bus_info(&adapter->hw);
7134 adapter->hw.phy.autoneg_wait_to_complete = 0;
7136 /* Copper options */
7137 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7138 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7139 adapter->hw.phy.disable_polarity_correction = 0;
7140 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7143 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7144 dev_info(&pdev->dev,
7145 "PHY reset is blocked due to SOL/IDER session.\n");
7147 /* Set initial default active device features */
7148 netdev->features = (NETIF_F_SG |
7149 NETIF_F_HW_VLAN_CTAG_RX |
7150 NETIF_F_HW_VLAN_CTAG_TX |
7157 /* Set user-changeable features (subset of all device features) */
7158 netdev->hw_features = netdev->features;
7159 netdev->hw_features |= NETIF_F_RXFCS;
7160 netdev->priv_flags |= IFF_SUPP_NOFCS;
7161 netdev->hw_features |= NETIF_F_RXALL;
7163 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7164 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7166 netdev->vlan_features |= (NETIF_F_SG |
7171 netdev->priv_flags |= IFF_UNICAST_FLT;
7173 if (pci_using_dac) {
7174 netdev->features |= NETIF_F_HIGHDMA;
7175 netdev->vlan_features |= NETIF_F_HIGHDMA;
7178 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7179 adapter->flags |= FLAG_MNG_PT_ENABLED;
7181 /* before reading the NVM, reset the controller to
7182 * put the device in a known good starting state
7184 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7186 /* systems with ASPM and others may see the checksum fail on the first
7187 * attempt. Let's give it a few tries
7190 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7193 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7199 e1000_eeprom_checks(adapter);
7201 /* copy the MAC address */
7202 if (e1000e_read_mac_addr(&adapter->hw))
7204 "NVM Read Error while reading MAC address\n");
7206 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7208 if (!is_valid_ether_addr(netdev->dev_addr)) {
7209 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7215 init_timer(&adapter->watchdog_timer);
7216 adapter->watchdog_timer.function = e1000_watchdog;
7217 adapter->watchdog_timer.data = (unsigned long)adapter;
7219 init_timer(&adapter->phy_info_timer);
7220 adapter->phy_info_timer.function = e1000_update_phy_info;
7221 adapter->phy_info_timer.data = (unsigned long)adapter;
7223 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7224 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7225 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7226 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7227 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7229 /* Initialize link parameters. User can change them with ethtool */
7230 adapter->hw.mac.autoneg = 1;
7231 adapter->fc_autoneg = true;
7232 adapter->hw.fc.requested_mode = e1000_fc_default;
7233 adapter->hw.fc.current_mode = e1000_fc_default;
7234 adapter->hw.phy.autoneg_advertised = 0x2f;
7236 /* Initial Wake on LAN setting - If APM wake is enabled in
7237 * the EEPROM, enable the ACPI Magic Packet filter
7239 if (adapter->flags & FLAG_APME_IN_WUC) {
7240 /* APME bit in EEPROM is mapped to WUC.APME */
7241 eeprom_data = er32(WUC);
7242 eeprom_apme_mask = E1000_WUC_APME;
7243 if ((hw->mac.type > e1000_ich10lan) &&
7244 (eeprom_data & E1000_WUC_PHY_WAKE))
7245 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7246 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7247 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7248 (adapter->hw.bus.func == 1))
7249 ret_val = e1000_read_nvm(&adapter->hw,
7250 NVM_INIT_CONTROL3_PORT_B,
7253 ret_val = e1000_read_nvm(&adapter->hw,
7254 NVM_INIT_CONTROL3_PORT_A,
7258 /* fetch WoL from EEPROM */
7260 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7261 else if (eeprom_data & eeprom_apme_mask)
7262 adapter->eeprom_wol |= E1000_WUFC_MAG;
7264 /* now that we have the eeprom settings, apply the special cases
7265 * where the eeprom may be wrong or the board simply won't support
7266 * wake on lan on a particular port
7268 if (!(adapter->flags & FLAG_HAS_WOL))
7269 adapter->eeprom_wol = 0;
7271 /* initialize the wol settings based on the eeprom settings */
7272 adapter->wol = adapter->eeprom_wol;
7274 /* make sure adapter isn't asleep if manageability is enabled */
7275 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7276 (hw->mac.ops.check_mng_mode(hw)))
7277 device_wakeup_enable(&pdev->dev);
7279 /* save off EEPROM version number */
7280 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7283 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7284 adapter->eeprom_vers = 0;
7287 /* init PTP hardware clock */
7288 e1000e_ptp_init(adapter);
7290 /* reset the hardware with the new settings */
7291 e1000e_reset(adapter);
7293 /* If the controller has AMT, do not set DRV_LOAD until the interface
7294 * is up. For all other cases, let the f/w know that the h/w is now
7295 * under the control of the driver.
7297 if (!(adapter->flags & FLAG_HAS_AMT))
7298 e1000e_get_hw_control(adapter);
7300 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7301 err = register_netdev(netdev);
7305 /* carrier off reporting is important to ethtool even BEFORE open */
7306 netif_carrier_off(netdev);
7308 e1000_print_device_info(adapter);
7310 if (pci_dev_run_wake(pdev))
7311 pm_runtime_put_noidle(&pdev->dev);
7316 if (!(adapter->flags & FLAG_HAS_AMT))
7317 e1000e_release_hw_control(adapter);
7319 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7320 e1000_phy_hw_reset(&adapter->hw);
7322 kfree(adapter->tx_ring);
7323 kfree(adapter->rx_ring);
7325 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7326 iounmap(adapter->hw.flash_address);
7327 e1000e_reset_interrupt_capability(adapter);
7329 iounmap(adapter->hw.hw_addr);
7331 free_netdev(netdev);
7333 pci_release_mem_regions(pdev);
7336 pci_disable_device(pdev);
7341 * e1000_remove - Device Removal Routine
7342 * @pdev: PCI device information struct
7344 * e1000_remove is called by the PCI subsystem to alert the driver
7345 * that it should release a PCI device. The could be caused by a
7346 * Hot-Plug event, or because the driver is going to be removed from
7349 static void e1000_remove(struct pci_dev *pdev)
7351 struct net_device *netdev = pci_get_drvdata(pdev);
7352 struct e1000_adapter *adapter = netdev_priv(netdev);
7353 bool down = test_bit(__E1000_DOWN, &adapter->state);
7355 e1000e_ptp_remove(adapter);
7357 /* The timers may be rescheduled, so explicitly disable them
7358 * from being rescheduled.
7361 set_bit(__E1000_DOWN, &adapter->state);
7362 del_timer_sync(&adapter->watchdog_timer);
7363 del_timer_sync(&adapter->phy_info_timer);
7365 cancel_work_sync(&adapter->reset_task);
7366 cancel_work_sync(&adapter->watchdog_task);
7367 cancel_work_sync(&adapter->downshift_task);
7368 cancel_work_sync(&adapter->update_phy_task);
7369 cancel_work_sync(&adapter->print_hang_task);
7371 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7372 cancel_work_sync(&adapter->tx_hwtstamp_work);
7373 if (adapter->tx_hwtstamp_skb) {
7374 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
7375 adapter->tx_hwtstamp_skb = NULL;
7379 /* Don't lie to e1000_close() down the road. */
7381 clear_bit(__E1000_DOWN, &adapter->state);
7382 unregister_netdev(netdev);
7384 if (pci_dev_run_wake(pdev))
7385 pm_runtime_get_noresume(&pdev->dev);
7387 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7388 * would have already happened in close and is redundant.
7390 e1000e_release_hw_control(adapter);
7392 e1000e_reset_interrupt_capability(adapter);
7393 kfree(adapter->tx_ring);
7394 kfree(adapter->rx_ring);
7396 iounmap(adapter->hw.hw_addr);
7397 if ((adapter->hw.flash_address) &&
7398 (adapter->hw.mac.type < e1000_pch_spt))
7399 iounmap(adapter->hw.flash_address);
7400 pci_release_mem_regions(pdev);
7402 free_netdev(netdev);
7405 pci_disable_pcie_error_reporting(pdev);
7407 pci_disable_device(pdev);
7410 /* PCI Error Recovery (ERS) */
7411 static const struct pci_error_handlers e1000_err_handler = {
7412 .error_detected = e1000_io_error_detected,
7413 .slot_reset = e1000_io_slot_reset,
7414 .resume = e1000_io_resume,
7417 static const struct pci_device_id e1000_pci_tbl[] = {
7418 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7419 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7420 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7421 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7423 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7424 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7425 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7426 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7427 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7429 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7430 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7431 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7432 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7434 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7435 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7436 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7438 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7439 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7440 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7442 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7443 board_80003es2lan },
7444 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7445 board_80003es2lan },
7446 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7447 board_80003es2lan },
7448 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7449 board_80003es2lan },
7451 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7452 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7453 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7454 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7455 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7456 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7457 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7460 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7462 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7467 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7479 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7481 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7483 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7487 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7489 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7490 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7492 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7493 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7494 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7495 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7504 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7506 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7508 static const struct dev_pm_ops e1000_pm_ops = {
7509 #ifdef CONFIG_PM_SLEEP
7510 .suspend = e1000e_pm_suspend,
7511 .resume = e1000e_pm_resume,
7512 .freeze = e1000e_pm_freeze,
7513 .thaw = e1000e_pm_thaw,
7514 .poweroff = e1000e_pm_suspend,
7515 .restore = e1000e_pm_resume,
7517 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7518 e1000e_pm_runtime_idle)
7521 /* PCI Device API Driver */
7522 static struct pci_driver e1000_driver = {
7523 .name = e1000e_driver_name,
7524 .id_table = e1000_pci_tbl,
7525 .probe = e1000_probe,
7526 .remove = e1000_remove,
7528 .pm = &e1000_pm_ops,
7530 .shutdown = e1000_shutdown,
7531 .err_handler = &e1000_err_handler
7535 * e1000_init_module - Driver Registration Routine
7537 * e1000_init_module is the first routine called when the driver is
7538 * loaded. All it does is register with the PCI subsystem.
7540 static int __init e1000_init_module(void)
7542 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7543 e1000e_driver_version);
7544 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7546 return pci_register_driver(&e1000_driver);
7548 module_init(e1000_init_module);
7551 * e1000_exit_module - Driver Exit Cleanup Routine
7553 * e1000_exit_module is called just before the driver is removed
7556 static void __exit e1000_exit_module(void)
7558 pci_unregister_driver(&e1000_driver);
7560 module_exit(e1000_exit_module);
7562 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7563 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7564 MODULE_LICENSE("GPL");
7565 MODULE_VERSION(DRV_VERSION);