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,
74 [board_pch_cnp] = &e1000_pch_cnp_info,
77 struct e1000_reg_info {
82 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
83 /* General Registers */
85 {E1000_STATUS, "STATUS"},
86 {E1000_CTRL_EXT, "CTRL_EXT"},
88 /* Interrupt Registers */
93 {E1000_RDLEN(0), "RDLEN"},
94 {E1000_RDH(0), "RDH"},
95 {E1000_RDT(0), "RDT"},
97 {E1000_RXDCTL(0), "RXDCTL"},
99 {E1000_RDBAL(0), "RDBAL"},
100 {E1000_RDBAH(0), "RDBAH"},
101 {E1000_RDFH, "RDFH"},
102 {E1000_RDFT, "RDFT"},
103 {E1000_RDFHS, "RDFHS"},
104 {E1000_RDFTS, "RDFTS"},
105 {E1000_RDFPC, "RDFPC"},
108 {E1000_TCTL, "TCTL"},
109 {E1000_TDBAL(0), "TDBAL"},
110 {E1000_TDBAH(0), "TDBAH"},
111 {E1000_TDLEN(0), "TDLEN"},
112 {E1000_TDH(0), "TDH"},
113 {E1000_TDT(0), "TDT"},
114 {E1000_TIDV, "TIDV"},
115 {E1000_TXDCTL(0), "TXDCTL"},
116 {E1000_TADV, "TADV"},
117 {E1000_TARC(0), "TARC"},
118 {E1000_TDFH, "TDFH"},
119 {E1000_TDFT, "TDFT"},
120 {E1000_TDFHS, "TDFHS"},
121 {E1000_TDFTS, "TDFTS"},
122 {E1000_TDFPC, "TDFPC"},
124 /* List Terminator */
129 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
130 * @hw: pointer to the HW structure
132 * When updating the MAC CSR registers, the Manageability Engine (ME) could
133 * be accessing the registers at the same time. Normally, this is handled in
134 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
135 * accesses later than it should which could result in the register to have
136 * an incorrect value. Workaround this by checking the FWSM register which
137 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
138 * and try again a number of times.
140 s32 __ew32_prepare(struct e1000_hw *hw)
142 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
144 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
150 void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
152 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
155 writel(val, hw->hw_addr + reg);
159 * e1000_regdump - register printout routine
160 * @hw: pointer to the HW structure
161 * @reginfo: pointer to the register info table
163 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
169 switch (reginfo->ofs) {
170 case E1000_RXDCTL(0):
171 for (n = 0; n < 2; n++)
172 regs[n] = __er32(hw, E1000_RXDCTL(n));
174 case E1000_TXDCTL(0):
175 for (n = 0; n < 2; n++)
176 regs[n] = __er32(hw, E1000_TXDCTL(n));
179 for (n = 0; n < 2; n++)
180 regs[n] = __er32(hw, E1000_TARC(n));
183 pr_info("%-15s %08x\n",
184 reginfo->name, __er32(hw, reginfo->ofs));
188 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
189 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
192 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
193 struct e1000_buffer *bi)
196 struct e1000_ps_page *ps_page;
198 for (i = 0; i < adapter->rx_ps_pages; i++) {
199 ps_page = &bi->ps_pages[i];
202 pr_info("packet dump for ps_page %d:\n", i);
203 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
204 16, 1, page_address(ps_page->page),
211 * e1000e_dump - Print registers, Tx-ring and Rx-ring
212 * @adapter: board private structure
214 static void e1000e_dump(struct e1000_adapter *adapter)
216 struct net_device *netdev = adapter->netdev;
217 struct e1000_hw *hw = &adapter->hw;
218 struct e1000_reg_info *reginfo;
219 struct e1000_ring *tx_ring = adapter->tx_ring;
220 struct e1000_tx_desc *tx_desc;
225 struct e1000_buffer *buffer_info;
226 struct e1000_ring *rx_ring = adapter->rx_ring;
227 union e1000_rx_desc_packet_split *rx_desc_ps;
228 union e1000_rx_desc_extended *rx_desc;
238 if (!netif_msg_hw(adapter))
241 /* Print netdevice Info */
243 dev_info(&adapter->pdev->dev, "Net device Info\n");
244 pr_info("Device Name state trans_start\n");
245 pr_info("%-15s %016lX %016lX\n", netdev->name,
246 netdev->state, dev_trans_start(netdev));
249 /* Print Registers */
250 dev_info(&adapter->pdev->dev, "Register Dump\n");
251 pr_info(" Register Name Value\n");
252 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
253 reginfo->name; reginfo++) {
254 e1000_regdump(hw, reginfo);
257 /* Print Tx Ring Summary */
258 if (!netdev || !netif_running(netdev))
261 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
262 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
263 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
264 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
265 0, tx_ring->next_to_use, tx_ring->next_to_clean,
266 (unsigned long long)buffer_info->dma,
268 buffer_info->next_to_watch,
269 (unsigned long long)buffer_info->time_stamp);
272 if (!netif_msg_tx_done(adapter))
273 goto rx_ring_summary;
275 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
277 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
279 * Legacy Transmit Descriptor
280 * +--------------------------------------------------------------+
281 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
282 * +--------------------------------------------------------------+
283 * 8 | Special | CSS | Status | CMD | CSO | Length |
284 * +--------------------------------------------------------------+
285 * 63 48 47 36 35 32 31 24 23 16 15 0
287 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
288 * 63 48 47 40 39 32 31 16 15 8 7 0
289 * +----------------------------------------------------------------+
290 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
291 * +----------------------------------------------------------------+
292 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
293 * +----------------------------------------------------------------+
294 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
296 * Extended Data Descriptor (DTYP=0x1)
297 * +----------------------------------------------------------------+
298 * 0 | Buffer Address [63:0] |
299 * +----------------------------------------------------------------+
300 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
301 * +----------------------------------------------------------------+
302 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
304 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
305 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
306 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
307 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
308 const char *next_desc;
309 tx_desc = E1000_TX_DESC(*tx_ring, i);
310 buffer_info = &tx_ring->buffer_info[i];
311 u0 = (struct my_u0 *)tx_desc;
312 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
313 next_desc = " NTC/U";
314 else if (i == tx_ring->next_to_use)
316 else if (i == tx_ring->next_to_clean)
320 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
321 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
322 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
324 (unsigned long long)le64_to_cpu(u0->a),
325 (unsigned long long)le64_to_cpu(u0->b),
326 (unsigned long long)buffer_info->dma,
327 buffer_info->length, buffer_info->next_to_watch,
328 (unsigned long long)buffer_info->time_stamp,
329 buffer_info->skb, next_desc);
331 if (netif_msg_pktdata(adapter) && buffer_info->skb)
332 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
333 16, 1, buffer_info->skb->data,
334 buffer_info->skb->len, true);
337 /* Print Rx Ring Summary */
339 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
340 pr_info("Queue [NTU] [NTC]\n");
341 pr_info(" %5d %5X %5X\n",
342 0, rx_ring->next_to_use, rx_ring->next_to_clean);
345 if (!netif_msg_rx_status(adapter))
348 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
349 switch (adapter->rx_ps_pages) {
353 /* [Extended] Packet Split Receive Descriptor Format
355 * +-----------------------------------------------------+
356 * 0 | Buffer Address 0 [63:0] |
357 * +-----------------------------------------------------+
358 * 8 | Buffer Address 1 [63:0] |
359 * +-----------------------------------------------------+
360 * 16 | Buffer Address 2 [63:0] |
361 * +-----------------------------------------------------+
362 * 24 | Buffer Address 3 [63:0] |
363 * +-----------------------------------------------------+
365 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");
366 /* [Extended] Receive Descriptor (Write-Back) Format
368 * 63 48 47 32 31 13 12 8 7 4 3 0
369 * +------------------------------------------------------+
370 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
371 * | Checksum | Ident | | Queue | | Type |
372 * +------------------------------------------------------+
373 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
374 * +------------------------------------------------------+
375 * 63 48 47 32 31 20 19 0
377 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
378 for (i = 0; i < rx_ring->count; i++) {
379 const char *next_desc;
380 buffer_info = &rx_ring->buffer_info[i];
381 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
382 u1 = (struct my_u1 *)rx_desc_ps;
384 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
386 if (i == rx_ring->next_to_use)
388 else if (i == rx_ring->next_to_clean)
393 if (staterr & E1000_RXD_STAT_DD) {
394 /* Descriptor Done */
395 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
397 (unsigned long long)le64_to_cpu(u1->a),
398 (unsigned long long)le64_to_cpu(u1->b),
399 (unsigned long long)le64_to_cpu(u1->c),
400 (unsigned long long)le64_to_cpu(u1->d),
401 buffer_info->skb, next_desc);
403 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
405 (unsigned long long)le64_to_cpu(u1->a),
406 (unsigned long long)le64_to_cpu(u1->b),
407 (unsigned long long)le64_to_cpu(u1->c),
408 (unsigned long long)le64_to_cpu(u1->d),
409 (unsigned long long)buffer_info->dma,
410 buffer_info->skb, next_desc);
412 if (netif_msg_pktdata(adapter))
413 e1000e_dump_ps_pages(adapter,
420 /* Extended Receive Descriptor (Read) Format
422 * +-----------------------------------------------------+
423 * 0 | Buffer Address [63:0] |
424 * +-----------------------------------------------------+
426 * +-----------------------------------------------------+
428 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
429 /* Extended Receive Descriptor (Write-Back) Format
431 * 63 48 47 32 31 24 23 4 3 0
432 * +------------------------------------------------------+
434 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
435 * | Packet | IP | | | Type |
436 * | Checksum | Ident | | | |
437 * +------------------------------------------------------+
438 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
439 * +------------------------------------------------------+
440 * 63 48 47 32 31 20 19 0
442 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
444 for (i = 0; i < rx_ring->count; i++) {
445 const char *next_desc;
447 buffer_info = &rx_ring->buffer_info[i];
448 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
449 u1 = (struct my_u1 *)rx_desc;
450 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
452 if (i == rx_ring->next_to_use)
454 else if (i == rx_ring->next_to_clean)
459 if (staterr & E1000_RXD_STAT_DD) {
460 /* Descriptor Done */
461 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
463 (unsigned long long)le64_to_cpu(u1->a),
464 (unsigned long long)le64_to_cpu(u1->b),
465 buffer_info->skb, next_desc);
467 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
469 (unsigned long long)le64_to_cpu(u1->a),
470 (unsigned long long)le64_to_cpu(u1->b),
471 (unsigned long long)buffer_info->dma,
472 buffer_info->skb, next_desc);
474 if (netif_msg_pktdata(adapter) &&
476 print_hex_dump(KERN_INFO, "",
477 DUMP_PREFIX_ADDRESS, 16,
479 buffer_info->skb->data,
480 adapter->rx_buffer_len,
488 * e1000_desc_unused - calculate if we have unused descriptors
490 static int e1000_desc_unused(struct e1000_ring *ring)
492 if (ring->next_to_clean > ring->next_to_use)
493 return ring->next_to_clean - ring->next_to_use - 1;
495 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
499 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
500 * @adapter: board private structure
501 * @hwtstamps: time stamp structure to update
502 * @systim: unsigned 64bit system time value.
504 * Convert the system time value stored in the RX/TXSTMP registers into a
505 * hwtstamp which can be used by the upper level time stamping functions.
507 * The 'systim_lock' spinlock is used to protect the consistency of the
508 * system time value. This is needed because reading the 64 bit time
509 * value involves reading two 32 bit registers. The first read latches the
512 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
513 struct skb_shared_hwtstamps *hwtstamps,
519 spin_lock_irqsave(&adapter->systim_lock, flags);
520 ns = timecounter_cyc2time(&adapter->tc, systim);
521 spin_unlock_irqrestore(&adapter->systim_lock, flags);
523 memset(hwtstamps, 0, sizeof(*hwtstamps));
524 hwtstamps->hwtstamp = ns_to_ktime(ns);
528 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
529 * @adapter: board private structure
530 * @status: descriptor extended error and status field
531 * @skb: particular skb to include time stamp
533 * If the time stamp is valid, convert it into the timecounter ns value
534 * and store that result into the shhwtstamps structure which is passed
535 * up the network stack.
537 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
540 struct e1000_hw *hw = &adapter->hw;
543 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
544 !(status & E1000_RXDEXT_STATERR_TST) ||
545 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
548 /* The Rx time stamp registers contain the time stamp. No other
549 * received packet will be time stamped until the Rx time stamp
550 * registers are read. Because only one packet can be time stamped
551 * at a time, the register values must belong to this packet and
552 * therefore none of the other additional attributes need to be
555 rxstmp = (u64)er32(RXSTMPL);
556 rxstmp |= (u64)er32(RXSTMPH) << 32;
557 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
559 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
563 * e1000_receive_skb - helper function to handle Rx indications
564 * @adapter: board private structure
565 * @staterr: descriptor extended error and status field as written by hardware
566 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
567 * @skb: pointer to sk_buff to be indicated to stack
569 static void e1000_receive_skb(struct e1000_adapter *adapter,
570 struct net_device *netdev, struct sk_buff *skb,
571 u32 staterr, __le16 vlan)
573 u16 tag = le16_to_cpu(vlan);
575 e1000e_rx_hwtstamp(adapter, staterr, skb);
577 skb->protocol = eth_type_trans(skb, netdev);
579 if (staterr & E1000_RXD_STAT_VP)
580 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
582 napi_gro_receive(&adapter->napi, skb);
586 * e1000_rx_checksum - Receive Checksum Offload
587 * @adapter: board private structure
588 * @status_err: receive descriptor status and error fields
589 * @csum: receive descriptor csum field
590 * @sk_buff: socket buffer with received data
592 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
595 u16 status = (u16)status_err;
596 u8 errors = (u8)(status_err >> 24);
598 skb_checksum_none_assert(skb);
600 /* Rx checksum disabled */
601 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
604 /* Ignore Checksum bit is set */
605 if (status & E1000_RXD_STAT_IXSM)
608 /* TCP/UDP checksum error bit or IP checksum error bit is set */
609 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
610 /* let the stack verify checksum errors */
611 adapter->hw_csum_err++;
615 /* TCP/UDP Checksum has not been calculated */
616 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
619 /* It must be a TCP or UDP packet with a valid checksum */
620 skb->ip_summed = CHECKSUM_UNNECESSARY;
621 adapter->hw_csum_good++;
624 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
626 struct e1000_adapter *adapter = rx_ring->adapter;
627 struct e1000_hw *hw = &adapter->hw;
628 s32 ret_val = __ew32_prepare(hw);
630 writel(i, rx_ring->tail);
632 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
633 u32 rctl = er32(RCTL);
635 ew32(RCTL, rctl & ~E1000_RCTL_EN);
636 e_err("ME firmware caused invalid RDT - resetting\n");
637 schedule_work(&adapter->reset_task);
641 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
643 struct e1000_adapter *adapter = tx_ring->adapter;
644 struct e1000_hw *hw = &adapter->hw;
645 s32 ret_val = __ew32_prepare(hw);
647 writel(i, tx_ring->tail);
649 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
650 u32 tctl = er32(TCTL);
652 ew32(TCTL, tctl & ~E1000_TCTL_EN);
653 e_err("ME firmware caused invalid TDT - resetting\n");
654 schedule_work(&adapter->reset_task);
659 * e1000_alloc_rx_buffers - Replace used receive buffers
660 * @rx_ring: Rx descriptor ring
662 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
663 int cleaned_count, gfp_t gfp)
665 struct e1000_adapter *adapter = rx_ring->adapter;
666 struct net_device *netdev = adapter->netdev;
667 struct pci_dev *pdev = adapter->pdev;
668 union e1000_rx_desc_extended *rx_desc;
669 struct e1000_buffer *buffer_info;
672 unsigned int bufsz = adapter->rx_buffer_len;
674 i = rx_ring->next_to_use;
675 buffer_info = &rx_ring->buffer_info[i];
677 while (cleaned_count--) {
678 skb = buffer_info->skb;
684 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
686 /* Better luck next round */
687 adapter->alloc_rx_buff_failed++;
691 buffer_info->skb = skb;
693 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
694 adapter->rx_buffer_len,
696 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
697 dev_err(&pdev->dev, "Rx DMA map failed\n");
698 adapter->rx_dma_failed++;
702 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
703 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
705 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
706 /* Force memory writes to complete before letting h/w
707 * know there are new descriptors to fetch. (Only
708 * applicable for weak-ordered memory model archs,
712 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
713 e1000e_update_rdt_wa(rx_ring, i);
715 writel(i, rx_ring->tail);
718 if (i == rx_ring->count)
720 buffer_info = &rx_ring->buffer_info[i];
723 rx_ring->next_to_use = i;
727 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
728 * @rx_ring: Rx descriptor ring
730 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
731 int cleaned_count, gfp_t gfp)
733 struct e1000_adapter *adapter = rx_ring->adapter;
734 struct net_device *netdev = adapter->netdev;
735 struct pci_dev *pdev = adapter->pdev;
736 union e1000_rx_desc_packet_split *rx_desc;
737 struct e1000_buffer *buffer_info;
738 struct e1000_ps_page *ps_page;
742 i = rx_ring->next_to_use;
743 buffer_info = &rx_ring->buffer_info[i];
745 while (cleaned_count--) {
746 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
748 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
749 ps_page = &buffer_info->ps_pages[j];
750 if (j >= adapter->rx_ps_pages) {
751 /* all unused desc entries get hw null ptr */
752 rx_desc->read.buffer_addr[j + 1] =
756 if (!ps_page->page) {
757 ps_page->page = alloc_page(gfp);
758 if (!ps_page->page) {
759 adapter->alloc_rx_buff_failed++;
762 ps_page->dma = dma_map_page(&pdev->dev,
766 if (dma_mapping_error(&pdev->dev,
768 dev_err(&adapter->pdev->dev,
769 "Rx DMA page map failed\n");
770 adapter->rx_dma_failed++;
774 /* Refresh the desc even if buffer_addrs
775 * didn't change because each write-back
778 rx_desc->read.buffer_addr[j + 1] =
779 cpu_to_le64(ps_page->dma);
782 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
786 adapter->alloc_rx_buff_failed++;
790 buffer_info->skb = skb;
791 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
792 adapter->rx_ps_bsize0,
794 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
795 dev_err(&pdev->dev, "Rx DMA map failed\n");
796 adapter->rx_dma_failed++;
798 dev_kfree_skb_any(skb);
799 buffer_info->skb = NULL;
803 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
805 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
806 /* Force memory writes to complete before letting h/w
807 * know there are new descriptors to fetch. (Only
808 * applicable for weak-ordered memory model archs,
812 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
813 e1000e_update_rdt_wa(rx_ring, i << 1);
815 writel(i << 1, rx_ring->tail);
819 if (i == rx_ring->count)
821 buffer_info = &rx_ring->buffer_info[i];
825 rx_ring->next_to_use = i;
829 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
830 * @rx_ring: Rx descriptor ring
831 * @cleaned_count: number of buffers to allocate this pass
834 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
835 int cleaned_count, gfp_t gfp)
837 struct e1000_adapter *adapter = rx_ring->adapter;
838 struct net_device *netdev = adapter->netdev;
839 struct pci_dev *pdev = adapter->pdev;
840 union e1000_rx_desc_extended *rx_desc;
841 struct e1000_buffer *buffer_info;
844 unsigned int bufsz = 256 - 16; /* for skb_reserve */
846 i = rx_ring->next_to_use;
847 buffer_info = &rx_ring->buffer_info[i];
849 while (cleaned_count--) {
850 skb = buffer_info->skb;
856 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
857 if (unlikely(!skb)) {
858 /* Better luck next round */
859 adapter->alloc_rx_buff_failed++;
863 buffer_info->skb = skb;
865 /* allocate a new page if necessary */
866 if (!buffer_info->page) {
867 buffer_info->page = alloc_page(gfp);
868 if (unlikely(!buffer_info->page)) {
869 adapter->alloc_rx_buff_failed++;
874 if (!buffer_info->dma) {
875 buffer_info->dma = dma_map_page(&pdev->dev,
876 buffer_info->page, 0,
879 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
880 adapter->alloc_rx_buff_failed++;
885 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
886 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
888 if (unlikely(++i == rx_ring->count))
890 buffer_info = &rx_ring->buffer_info[i];
893 if (likely(rx_ring->next_to_use != i)) {
894 rx_ring->next_to_use = i;
895 if (unlikely(i-- == 0))
896 i = (rx_ring->count - 1);
898 /* Force memory writes to complete before letting h/w
899 * know there are new descriptors to fetch. (Only
900 * applicable for weak-ordered memory model archs,
904 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
905 e1000e_update_rdt_wa(rx_ring, i);
907 writel(i, rx_ring->tail);
911 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
914 if (netdev->features & NETIF_F_RXHASH)
915 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
919 * e1000_clean_rx_irq - Send received data up the network stack
920 * @rx_ring: Rx descriptor ring
922 * the return value indicates whether actual cleaning was done, there
923 * is no guarantee that everything was cleaned
925 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
928 struct e1000_adapter *adapter = rx_ring->adapter;
929 struct net_device *netdev = adapter->netdev;
930 struct pci_dev *pdev = adapter->pdev;
931 struct e1000_hw *hw = &adapter->hw;
932 union e1000_rx_desc_extended *rx_desc, *next_rxd;
933 struct e1000_buffer *buffer_info, *next_buffer;
936 int cleaned_count = 0;
937 bool cleaned = false;
938 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
940 i = rx_ring->next_to_clean;
941 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
942 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
943 buffer_info = &rx_ring->buffer_info[i];
945 while (staterr & E1000_RXD_STAT_DD) {
948 if (*work_done >= work_to_do)
951 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
953 skb = buffer_info->skb;
954 buffer_info->skb = NULL;
956 prefetch(skb->data - NET_IP_ALIGN);
959 if (i == rx_ring->count)
961 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
964 next_buffer = &rx_ring->buffer_info[i];
968 dma_unmap_single(&pdev->dev, buffer_info->dma,
969 adapter->rx_buffer_len, DMA_FROM_DEVICE);
970 buffer_info->dma = 0;
972 length = le16_to_cpu(rx_desc->wb.upper.length);
974 /* !EOP means multiple descriptors were used to store a single
975 * packet, if that's the case we need to toss it. In fact, we
976 * need to toss every packet with the EOP bit clear and the
977 * next frame that _does_ have the EOP bit set, as it is by
978 * definition only a frame fragment
980 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
981 adapter->flags2 |= FLAG2_IS_DISCARDING;
983 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
984 /* All receives must fit into a single buffer */
985 e_dbg("Receive packet consumed multiple buffers\n");
987 buffer_info->skb = skb;
988 if (staterr & E1000_RXD_STAT_EOP)
989 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
993 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
994 !(netdev->features & NETIF_F_RXALL))) {
996 buffer_info->skb = skb;
1000 /* adjust length to remove Ethernet CRC */
1001 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1002 /* If configured to store CRC, don't subtract FCS,
1003 * but keep the FCS bytes out of the total_rx_bytes
1006 if (netdev->features & NETIF_F_RXFCS)
1007 total_rx_bytes -= 4;
1012 total_rx_bytes += length;
1015 /* code added for copybreak, this should improve
1016 * performance for small packets with large amounts
1017 * of reassembly being done in the stack
1019 if (length < copybreak) {
1020 struct sk_buff *new_skb =
1021 napi_alloc_skb(&adapter->napi, length);
1023 skb_copy_to_linear_data_offset(new_skb,
1029 /* save the skb in buffer_info as good */
1030 buffer_info->skb = skb;
1033 /* else just continue with the old one */
1035 /* end copybreak code */
1036 skb_put(skb, length);
1038 /* Receive Checksum Offload */
1039 e1000_rx_checksum(adapter, staterr, skb);
1041 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1043 e1000_receive_skb(adapter, netdev, skb, staterr,
1044 rx_desc->wb.upper.vlan);
1047 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1049 /* return some buffers to hardware, one at a time is too slow */
1050 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1051 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1056 /* use prefetched values */
1058 buffer_info = next_buffer;
1060 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1062 rx_ring->next_to_clean = i;
1064 cleaned_count = e1000_desc_unused(rx_ring);
1066 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1068 adapter->total_rx_bytes += total_rx_bytes;
1069 adapter->total_rx_packets += total_rx_packets;
1073 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1074 struct e1000_buffer *buffer_info,
1077 struct e1000_adapter *adapter = tx_ring->adapter;
1079 if (buffer_info->dma) {
1080 if (buffer_info->mapped_as_page)
1081 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1082 buffer_info->length, DMA_TO_DEVICE);
1084 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1085 buffer_info->length, DMA_TO_DEVICE);
1086 buffer_info->dma = 0;
1088 if (buffer_info->skb) {
1090 dev_kfree_skb_any(buffer_info->skb);
1092 dev_consume_skb_any(buffer_info->skb);
1093 buffer_info->skb = NULL;
1095 buffer_info->time_stamp = 0;
1098 static void e1000_print_hw_hang(struct work_struct *work)
1100 struct e1000_adapter *adapter = container_of(work,
1101 struct e1000_adapter,
1103 struct net_device *netdev = adapter->netdev;
1104 struct e1000_ring *tx_ring = adapter->tx_ring;
1105 unsigned int i = tx_ring->next_to_clean;
1106 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1107 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1108 struct e1000_hw *hw = &adapter->hw;
1109 u16 phy_status, phy_1000t_status, phy_ext_status;
1112 if (test_bit(__E1000_DOWN, &adapter->state))
1115 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1116 /* May be block on write-back, flush and detect again
1117 * flush pending descriptor writebacks to memory
1119 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1120 /* execute the writes immediately */
1122 /* Due to rare timing issues, write to TIDV again to ensure
1123 * the write is successful
1125 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1126 /* execute the writes immediately */
1128 adapter->tx_hang_recheck = true;
1131 adapter->tx_hang_recheck = false;
1133 if (er32(TDH(0)) == er32(TDT(0))) {
1134 e_dbg("false hang detected, ignoring\n");
1138 /* Real hang detected */
1139 netif_stop_queue(netdev);
1141 e1e_rphy(hw, MII_BMSR, &phy_status);
1142 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1143 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1145 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1147 /* detected Hardware unit hang */
1148 e_err("Detected Hardware Unit Hang:\n"
1151 " next_to_use <%x>\n"
1152 " next_to_clean <%x>\n"
1153 "buffer_info[next_to_clean]:\n"
1154 " time_stamp <%lx>\n"
1155 " next_to_watch <%x>\n"
1157 " next_to_watch.status <%x>\n"
1160 "PHY 1000BASE-T Status <%x>\n"
1161 "PHY Extended Status <%x>\n"
1162 "PCI Status <%x>\n",
1163 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1164 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1165 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1166 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1168 e1000e_dump(adapter);
1170 /* Suggest workaround for known h/w issue */
1171 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1172 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1176 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1177 * @work: pointer to work struct
1179 * This work function polls the TSYNCTXCTL valid bit to determine when a
1180 * timestamp has been taken for the current stored skb. The timestamp must
1181 * be for this skb because only one such packet is allowed in the queue.
1183 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1185 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1187 struct e1000_hw *hw = &adapter->hw;
1189 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1190 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1191 struct skb_shared_hwtstamps shhwtstamps;
1194 txstmp = er32(TXSTMPL);
1195 txstmp |= (u64)er32(TXSTMPH) << 32;
1197 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1199 /* Clear the global tx_hwtstamp_skb pointer and force writes
1200 * prior to notifying the stack of a Tx timestamp.
1202 adapter->tx_hwtstamp_skb = NULL;
1203 wmb(); /* force write prior to skb_tstamp_tx */
1205 skb_tstamp_tx(skb, &shhwtstamps);
1206 dev_consume_skb_any(skb);
1207 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1208 + adapter->tx_timeout_factor * HZ)) {
1209 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1210 adapter->tx_hwtstamp_skb = NULL;
1211 adapter->tx_hwtstamp_timeouts++;
1212 e_warn("clearing Tx timestamp hang\n");
1214 /* reschedule to check later */
1215 schedule_work(&adapter->tx_hwtstamp_work);
1220 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1221 * @tx_ring: Tx descriptor ring
1223 * the return value indicates whether actual cleaning was done, there
1224 * is no guarantee that everything was cleaned
1226 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1228 struct e1000_adapter *adapter = tx_ring->adapter;
1229 struct net_device *netdev = adapter->netdev;
1230 struct e1000_hw *hw = &adapter->hw;
1231 struct e1000_tx_desc *tx_desc, *eop_desc;
1232 struct e1000_buffer *buffer_info;
1233 unsigned int i, eop;
1234 unsigned int count = 0;
1235 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1236 unsigned int bytes_compl = 0, pkts_compl = 0;
1238 i = tx_ring->next_to_clean;
1239 eop = tx_ring->buffer_info[i].next_to_watch;
1240 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1242 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1243 (count < tx_ring->count)) {
1244 bool cleaned = false;
1246 dma_rmb(); /* read buffer_info after eop_desc */
1247 for (; !cleaned; count++) {
1248 tx_desc = E1000_TX_DESC(*tx_ring, i);
1249 buffer_info = &tx_ring->buffer_info[i];
1250 cleaned = (i == eop);
1253 total_tx_packets += buffer_info->segs;
1254 total_tx_bytes += buffer_info->bytecount;
1255 if (buffer_info->skb) {
1256 bytes_compl += buffer_info->skb->len;
1261 e1000_put_txbuf(tx_ring, buffer_info, false);
1262 tx_desc->upper.data = 0;
1265 if (i == tx_ring->count)
1269 if (i == tx_ring->next_to_use)
1271 eop = tx_ring->buffer_info[i].next_to_watch;
1272 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1275 tx_ring->next_to_clean = i;
1277 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1279 #define TX_WAKE_THRESHOLD 32
1280 if (count && netif_carrier_ok(netdev) &&
1281 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1282 /* Make sure that anybody stopping the queue after this
1283 * sees the new next_to_clean.
1287 if (netif_queue_stopped(netdev) &&
1288 !(test_bit(__E1000_DOWN, &adapter->state))) {
1289 netif_wake_queue(netdev);
1290 ++adapter->restart_queue;
1294 if (adapter->detect_tx_hung) {
1295 /* Detect a transmit hang in hardware, this serializes the
1296 * check with the clearing of time_stamp and movement of i
1298 adapter->detect_tx_hung = false;
1299 if (tx_ring->buffer_info[i].time_stamp &&
1300 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1301 + (adapter->tx_timeout_factor * HZ)) &&
1302 !(er32(STATUS) & E1000_STATUS_TXOFF))
1303 schedule_work(&adapter->print_hang_task);
1305 adapter->tx_hang_recheck = false;
1307 adapter->total_tx_bytes += total_tx_bytes;
1308 adapter->total_tx_packets += total_tx_packets;
1309 return count < tx_ring->count;
1313 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1314 * @rx_ring: Rx descriptor ring
1316 * the return value indicates whether actual cleaning was done, there
1317 * is no guarantee that everything was cleaned
1319 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1322 struct e1000_adapter *adapter = rx_ring->adapter;
1323 struct e1000_hw *hw = &adapter->hw;
1324 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1325 struct net_device *netdev = adapter->netdev;
1326 struct pci_dev *pdev = adapter->pdev;
1327 struct e1000_buffer *buffer_info, *next_buffer;
1328 struct e1000_ps_page *ps_page;
1329 struct sk_buff *skb;
1331 u32 length, staterr;
1332 int cleaned_count = 0;
1333 bool cleaned = false;
1334 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1336 i = rx_ring->next_to_clean;
1337 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1338 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1339 buffer_info = &rx_ring->buffer_info[i];
1341 while (staterr & E1000_RXD_STAT_DD) {
1342 if (*work_done >= work_to_do)
1345 skb = buffer_info->skb;
1346 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1348 /* in the packet split case this is header only */
1349 prefetch(skb->data - NET_IP_ALIGN);
1352 if (i == rx_ring->count)
1354 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1357 next_buffer = &rx_ring->buffer_info[i];
1361 dma_unmap_single(&pdev->dev, buffer_info->dma,
1362 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1363 buffer_info->dma = 0;
1365 /* see !EOP comment in other Rx routine */
1366 if (!(staterr & E1000_RXD_STAT_EOP))
1367 adapter->flags2 |= FLAG2_IS_DISCARDING;
1369 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1370 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1371 dev_kfree_skb_irq(skb);
1372 if (staterr & E1000_RXD_STAT_EOP)
1373 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1377 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1378 !(netdev->features & NETIF_F_RXALL))) {
1379 dev_kfree_skb_irq(skb);
1383 length = le16_to_cpu(rx_desc->wb.middle.length0);
1386 e_dbg("Last part of the packet spanning multiple descriptors\n");
1387 dev_kfree_skb_irq(skb);
1392 skb_put(skb, length);
1395 /* this looks ugly, but it seems compiler issues make
1396 * it more efficient than reusing j
1398 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1400 /* page alloc/put takes too long and effects small
1401 * packet throughput, so unsplit small packets and
1402 * save the alloc/put only valid in softirq (napi)
1403 * context to call kmap_*
1405 if (l1 && (l1 <= copybreak) &&
1406 ((length + l1) <= adapter->rx_ps_bsize0)) {
1409 ps_page = &buffer_info->ps_pages[0];
1411 /* there is no documentation about how to call
1412 * kmap_atomic, so we can't hold the mapping
1415 dma_sync_single_for_cpu(&pdev->dev,
1419 vaddr = kmap_atomic(ps_page->page);
1420 memcpy(skb_tail_pointer(skb), vaddr, l1);
1421 kunmap_atomic(vaddr);
1422 dma_sync_single_for_device(&pdev->dev,
1427 /* remove the CRC */
1428 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1429 if (!(netdev->features & NETIF_F_RXFCS))
1438 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1439 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1443 ps_page = &buffer_info->ps_pages[j];
1444 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1447 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1448 ps_page->page = NULL;
1450 skb->data_len += length;
1451 skb->truesize += PAGE_SIZE;
1454 /* strip the ethernet crc, problem is we're using pages now so
1455 * this whole operation can get a little cpu intensive
1457 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1458 if (!(netdev->features & NETIF_F_RXFCS))
1459 pskb_trim(skb, skb->len - 4);
1463 total_rx_bytes += skb->len;
1466 e1000_rx_checksum(adapter, staterr, skb);
1468 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1470 if (rx_desc->wb.upper.header_status &
1471 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1472 adapter->rx_hdr_split++;
1474 e1000_receive_skb(adapter, netdev, skb, staterr,
1475 rx_desc->wb.middle.vlan);
1478 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1479 buffer_info->skb = NULL;
1481 /* return some buffers to hardware, one at a time is too slow */
1482 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1483 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1488 /* use prefetched values */
1490 buffer_info = next_buffer;
1492 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1494 rx_ring->next_to_clean = i;
1496 cleaned_count = e1000_desc_unused(rx_ring);
1498 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1500 adapter->total_rx_bytes += total_rx_bytes;
1501 adapter->total_rx_packets += total_rx_packets;
1506 * e1000_consume_page - helper function
1508 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1513 skb->data_len += length;
1514 skb->truesize += PAGE_SIZE;
1518 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1519 * @adapter: board private structure
1521 * the return value indicates whether actual cleaning was done, there
1522 * is no guarantee that everything was cleaned
1524 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1527 struct e1000_adapter *adapter = rx_ring->adapter;
1528 struct net_device *netdev = adapter->netdev;
1529 struct pci_dev *pdev = adapter->pdev;
1530 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1531 struct e1000_buffer *buffer_info, *next_buffer;
1532 u32 length, staterr;
1534 int cleaned_count = 0;
1535 bool cleaned = false;
1536 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1537 struct skb_shared_info *shinfo;
1539 i = rx_ring->next_to_clean;
1540 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1541 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1542 buffer_info = &rx_ring->buffer_info[i];
1544 while (staterr & E1000_RXD_STAT_DD) {
1545 struct sk_buff *skb;
1547 if (*work_done >= work_to_do)
1550 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1552 skb = buffer_info->skb;
1553 buffer_info->skb = NULL;
1556 if (i == rx_ring->count)
1558 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1561 next_buffer = &rx_ring->buffer_info[i];
1565 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1567 buffer_info->dma = 0;
1569 length = le16_to_cpu(rx_desc->wb.upper.length);
1571 /* errors is only valid for DD + EOP descriptors */
1572 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1573 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1574 !(netdev->features & NETIF_F_RXALL)))) {
1575 /* recycle both page and skb */
1576 buffer_info->skb = skb;
1577 /* an error means any chain goes out the window too */
1578 if (rx_ring->rx_skb_top)
1579 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1580 rx_ring->rx_skb_top = NULL;
1583 #define rxtop (rx_ring->rx_skb_top)
1584 if (!(staterr & E1000_RXD_STAT_EOP)) {
1585 /* this descriptor is only the beginning (or middle) */
1587 /* this is the beginning of a chain */
1589 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1592 /* this is the middle of a chain */
1593 shinfo = skb_shinfo(rxtop);
1594 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1595 buffer_info->page, 0,
1597 /* re-use the skb, only consumed the page */
1598 buffer_info->skb = skb;
1600 e1000_consume_page(buffer_info, rxtop, length);
1604 /* end of the chain */
1605 shinfo = skb_shinfo(rxtop);
1606 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1607 buffer_info->page, 0,
1609 /* re-use the current skb, we only consumed the
1612 buffer_info->skb = skb;
1615 e1000_consume_page(buffer_info, skb, length);
1617 /* no chain, got EOP, this buf is the packet
1618 * copybreak to save the put_page/alloc_page
1620 if (length <= copybreak &&
1621 skb_tailroom(skb) >= length) {
1623 vaddr = kmap_atomic(buffer_info->page);
1624 memcpy(skb_tail_pointer(skb), vaddr,
1626 kunmap_atomic(vaddr);
1627 /* re-use the page, so don't erase
1630 skb_put(skb, length);
1632 skb_fill_page_desc(skb, 0,
1633 buffer_info->page, 0,
1635 e1000_consume_page(buffer_info, skb,
1641 /* Receive Checksum Offload */
1642 e1000_rx_checksum(adapter, staterr, skb);
1644 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1646 /* probably a little skewed due to removing CRC */
1647 total_rx_bytes += skb->len;
1650 /* eth type trans needs skb->data to point to something */
1651 if (!pskb_may_pull(skb, ETH_HLEN)) {
1652 e_err("pskb_may_pull failed.\n");
1653 dev_kfree_skb_irq(skb);
1657 e1000_receive_skb(adapter, netdev, skb, staterr,
1658 rx_desc->wb.upper.vlan);
1661 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1663 /* return some buffers to hardware, one at a time is too slow */
1664 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1665 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1670 /* use prefetched values */
1672 buffer_info = next_buffer;
1674 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1676 rx_ring->next_to_clean = i;
1678 cleaned_count = e1000_desc_unused(rx_ring);
1680 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1682 adapter->total_rx_bytes += total_rx_bytes;
1683 adapter->total_rx_packets += total_rx_packets;
1688 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1689 * @rx_ring: Rx descriptor ring
1691 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1693 struct e1000_adapter *adapter = rx_ring->adapter;
1694 struct e1000_buffer *buffer_info;
1695 struct e1000_ps_page *ps_page;
1696 struct pci_dev *pdev = adapter->pdev;
1699 /* Free all the Rx ring sk_buffs */
1700 for (i = 0; i < rx_ring->count; i++) {
1701 buffer_info = &rx_ring->buffer_info[i];
1702 if (buffer_info->dma) {
1703 if (adapter->clean_rx == e1000_clean_rx_irq)
1704 dma_unmap_single(&pdev->dev, buffer_info->dma,
1705 adapter->rx_buffer_len,
1707 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1708 dma_unmap_page(&pdev->dev, buffer_info->dma,
1709 PAGE_SIZE, DMA_FROM_DEVICE);
1710 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1711 dma_unmap_single(&pdev->dev, buffer_info->dma,
1712 adapter->rx_ps_bsize0,
1714 buffer_info->dma = 0;
1717 if (buffer_info->page) {
1718 put_page(buffer_info->page);
1719 buffer_info->page = NULL;
1722 if (buffer_info->skb) {
1723 dev_kfree_skb(buffer_info->skb);
1724 buffer_info->skb = NULL;
1727 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1728 ps_page = &buffer_info->ps_pages[j];
1731 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1734 put_page(ps_page->page);
1735 ps_page->page = NULL;
1739 /* there also may be some cached data from a chained receive */
1740 if (rx_ring->rx_skb_top) {
1741 dev_kfree_skb(rx_ring->rx_skb_top);
1742 rx_ring->rx_skb_top = NULL;
1745 /* Zero out the descriptor ring */
1746 memset(rx_ring->desc, 0, rx_ring->size);
1748 rx_ring->next_to_clean = 0;
1749 rx_ring->next_to_use = 0;
1750 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1753 static void e1000e_downshift_workaround(struct work_struct *work)
1755 struct e1000_adapter *adapter = container_of(work,
1756 struct e1000_adapter,
1759 if (test_bit(__E1000_DOWN, &adapter->state))
1762 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1766 * e1000_intr_msi - Interrupt Handler
1767 * @irq: interrupt number
1768 * @data: pointer to a network interface device structure
1770 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1772 struct net_device *netdev = data;
1773 struct e1000_adapter *adapter = netdev_priv(netdev);
1774 struct e1000_hw *hw = &adapter->hw;
1775 u32 icr = er32(ICR);
1777 /* read ICR disables interrupts using IAM */
1778 if (icr & E1000_ICR_LSC) {
1779 hw->mac.get_link_status = true;
1780 /* ICH8 workaround-- Call gig speed drop workaround on cable
1781 * disconnect (LSC) before accessing any PHY registers
1783 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1784 (!(er32(STATUS) & E1000_STATUS_LU)))
1785 schedule_work(&adapter->downshift_task);
1787 /* 80003ES2LAN workaround-- For packet buffer work-around on
1788 * link down event; disable receives here in the ISR and reset
1789 * adapter in watchdog
1791 if (netif_carrier_ok(netdev) &&
1792 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1793 /* disable receives */
1794 u32 rctl = er32(RCTL);
1796 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1797 adapter->flags |= FLAG_RESTART_NOW;
1799 /* guard against interrupt when we're going down */
1800 if (!test_bit(__E1000_DOWN, &adapter->state))
1801 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1804 /* Reset on uncorrectable ECC error */
1805 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1806 u32 pbeccsts = er32(PBECCSTS);
1808 adapter->corr_errors +=
1809 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1810 adapter->uncorr_errors +=
1811 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1812 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1814 /* Do the reset outside of interrupt context */
1815 schedule_work(&adapter->reset_task);
1817 /* return immediately since reset is imminent */
1821 if (napi_schedule_prep(&adapter->napi)) {
1822 adapter->total_tx_bytes = 0;
1823 adapter->total_tx_packets = 0;
1824 adapter->total_rx_bytes = 0;
1825 adapter->total_rx_packets = 0;
1826 __napi_schedule(&adapter->napi);
1833 * e1000_intr - Interrupt Handler
1834 * @irq: interrupt number
1835 * @data: pointer to a network interface device structure
1837 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1839 struct net_device *netdev = data;
1840 struct e1000_adapter *adapter = netdev_priv(netdev);
1841 struct e1000_hw *hw = &adapter->hw;
1842 u32 rctl, icr = er32(ICR);
1844 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1845 return IRQ_NONE; /* Not our interrupt */
1847 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1848 * not set, then the adapter didn't send an interrupt
1850 if (!(icr & E1000_ICR_INT_ASSERTED))
1853 /* Interrupt Auto-Mask...upon reading ICR,
1854 * interrupts are masked. No need for the
1858 if (icr & E1000_ICR_LSC) {
1859 hw->mac.get_link_status = true;
1860 /* ICH8 workaround-- Call gig speed drop workaround on cable
1861 * disconnect (LSC) before accessing any PHY registers
1863 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1864 (!(er32(STATUS) & E1000_STATUS_LU)))
1865 schedule_work(&adapter->downshift_task);
1867 /* 80003ES2LAN workaround--
1868 * For packet buffer work-around on link down event;
1869 * disable receives here in the ISR and
1870 * reset adapter in watchdog
1872 if (netif_carrier_ok(netdev) &&
1873 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1874 /* disable receives */
1876 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1877 adapter->flags |= FLAG_RESTART_NOW;
1879 /* guard against interrupt when we're going down */
1880 if (!test_bit(__E1000_DOWN, &adapter->state))
1881 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1884 /* Reset on uncorrectable ECC error */
1885 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1886 u32 pbeccsts = er32(PBECCSTS);
1888 adapter->corr_errors +=
1889 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1890 adapter->uncorr_errors +=
1891 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1892 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1894 /* Do the reset outside of interrupt context */
1895 schedule_work(&adapter->reset_task);
1897 /* return immediately since reset is imminent */
1901 if (napi_schedule_prep(&adapter->napi)) {
1902 adapter->total_tx_bytes = 0;
1903 adapter->total_tx_packets = 0;
1904 adapter->total_rx_bytes = 0;
1905 adapter->total_rx_packets = 0;
1906 __napi_schedule(&adapter->napi);
1912 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1914 struct net_device *netdev = data;
1915 struct e1000_adapter *adapter = netdev_priv(netdev);
1916 struct e1000_hw *hw = &adapter->hw;
1917 u32 icr = er32(ICR);
1919 if (icr & adapter->eiac_mask)
1920 ew32(ICS, (icr & adapter->eiac_mask));
1922 if (icr & E1000_ICR_LSC) {
1923 hw->mac.get_link_status = true;
1924 /* guard against interrupt when we're going down */
1925 if (!test_bit(__E1000_DOWN, &adapter->state))
1926 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1929 if (!test_bit(__E1000_DOWN, &adapter->state))
1930 ew32(IMS, E1000_IMS_OTHER | IMS_OTHER_MASK);
1935 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1937 struct net_device *netdev = data;
1938 struct e1000_adapter *adapter = netdev_priv(netdev);
1939 struct e1000_hw *hw = &adapter->hw;
1940 struct e1000_ring *tx_ring = adapter->tx_ring;
1942 adapter->total_tx_bytes = 0;
1943 adapter->total_tx_packets = 0;
1945 if (!e1000_clean_tx_irq(tx_ring))
1946 /* Ring was not completely cleaned, so fire another interrupt */
1947 ew32(ICS, tx_ring->ims_val);
1949 if (!test_bit(__E1000_DOWN, &adapter->state))
1950 ew32(IMS, adapter->tx_ring->ims_val);
1955 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1957 struct net_device *netdev = data;
1958 struct e1000_adapter *adapter = netdev_priv(netdev);
1959 struct e1000_ring *rx_ring = adapter->rx_ring;
1961 /* Write the ITR value calculated at the end of the
1962 * previous interrupt.
1964 if (rx_ring->set_itr) {
1965 u32 itr = rx_ring->itr_val ?
1966 1000000000 / (rx_ring->itr_val * 256) : 0;
1968 writel(itr, rx_ring->itr_register);
1969 rx_ring->set_itr = 0;
1972 if (napi_schedule_prep(&adapter->napi)) {
1973 adapter->total_rx_bytes = 0;
1974 adapter->total_rx_packets = 0;
1975 __napi_schedule(&adapter->napi);
1981 * e1000_configure_msix - Configure MSI-X hardware
1983 * e1000_configure_msix sets up the hardware to properly
1984 * generate MSI-X interrupts.
1986 static void e1000_configure_msix(struct e1000_adapter *adapter)
1988 struct e1000_hw *hw = &adapter->hw;
1989 struct e1000_ring *rx_ring = adapter->rx_ring;
1990 struct e1000_ring *tx_ring = adapter->tx_ring;
1992 u32 ctrl_ext, ivar = 0;
1994 adapter->eiac_mask = 0;
1996 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1997 if (hw->mac.type == e1000_82574) {
1998 u32 rfctl = er32(RFCTL);
2000 rfctl |= E1000_RFCTL_ACK_DIS;
2004 /* Configure Rx vector */
2005 rx_ring->ims_val = E1000_IMS_RXQ0;
2006 adapter->eiac_mask |= rx_ring->ims_val;
2007 if (rx_ring->itr_val)
2008 writel(1000000000 / (rx_ring->itr_val * 256),
2009 rx_ring->itr_register);
2011 writel(1, rx_ring->itr_register);
2012 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
2014 /* Configure Tx vector */
2015 tx_ring->ims_val = E1000_IMS_TXQ0;
2017 if (tx_ring->itr_val)
2018 writel(1000000000 / (tx_ring->itr_val * 256),
2019 tx_ring->itr_register);
2021 writel(1, tx_ring->itr_register);
2022 adapter->eiac_mask |= tx_ring->ims_val;
2023 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2025 /* set vector for Other Causes, e.g. link changes */
2027 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2028 if (rx_ring->itr_val)
2029 writel(1000000000 / (rx_ring->itr_val * 256),
2030 hw->hw_addr + E1000_EITR_82574(vector));
2032 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2034 /* Cause Tx interrupts on every write back */
2039 /* enable MSI-X PBA support */
2040 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2041 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2042 ew32(CTRL_EXT, ctrl_ext);
2046 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2048 if (adapter->msix_entries) {
2049 pci_disable_msix(adapter->pdev);
2050 kfree(adapter->msix_entries);
2051 adapter->msix_entries = NULL;
2052 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2053 pci_disable_msi(adapter->pdev);
2054 adapter->flags &= ~FLAG_MSI_ENABLED;
2059 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2061 * Attempt to configure interrupts using the best available
2062 * capabilities of the hardware and kernel.
2064 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2069 switch (adapter->int_mode) {
2070 case E1000E_INT_MODE_MSIX:
2071 if (adapter->flags & FLAG_HAS_MSIX) {
2072 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2073 adapter->msix_entries = kcalloc(adapter->num_vectors,
2077 if (adapter->msix_entries) {
2078 struct e1000_adapter *a = adapter;
2080 for (i = 0; i < adapter->num_vectors; i++)
2081 adapter->msix_entries[i].entry = i;
2083 err = pci_enable_msix_range(a->pdev,
2090 /* MSI-X failed, so fall through and try MSI */
2091 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2092 e1000e_reset_interrupt_capability(adapter);
2094 adapter->int_mode = E1000E_INT_MODE_MSI;
2096 case E1000E_INT_MODE_MSI:
2097 if (!pci_enable_msi(adapter->pdev)) {
2098 adapter->flags |= FLAG_MSI_ENABLED;
2100 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2101 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2104 case E1000E_INT_MODE_LEGACY:
2105 /* Don't do anything; this is the system default */
2109 /* store the number of vectors being used */
2110 adapter->num_vectors = 1;
2114 * e1000_request_msix - Initialize MSI-X interrupts
2116 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2119 static int e1000_request_msix(struct e1000_adapter *adapter)
2121 struct net_device *netdev = adapter->netdev;
2122 int err = 0, vector = 0;
2124 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2125 snprintf(adapter->rx_ring->name,
2126 sizeof(adapter->rx_ring->name) - 1,
2127 "%s-rx-0", netdev->name);
2129 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2130 err = request_irq(adapter->msix_entries[vector].vector,
2131 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2135 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2136 E1000_EITR_82574(vector);
2137 adapter->rx_ring->itr_val = adapter->itr;
2140 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2141 snprintf(adapter->tx_ring->name,
2142 sizeof(adapter->tx_ring->name) - 1,
2143 "%s-tx-0", netdev->name);
2145 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2146 err = request_irq(adapter->msix_entries[vector].vector,
2147 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2151 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2152 E1000_EITR_82574(vector);
2153 adapter->tx_ring->itr_val = adapter->itr;
2156 err = request_irq(adapter->msix_entries[vector].vector,
2157 e1000_msix_other, 0, netdev->name, netdev);
2161 e1000_configure_msix(adapter);
2167 * e1000_request_irq - initialize interrupts
2169 * Attempts to configure interrupts using the best available
2170 * capabilities of the hardware and kernel.
2172 static int e1000_request_irq(struct e1000_adapter *adapter)
2174 struct net_device *netdev = adapter->netdev;
2177 if (adapter->msix_entries) {
2178 err = e1000_request_msix(adapter);
2181 /* fall back to MSI */
2182 e1000e_reset_interrupt_capability(adapter);
2183 adapter->int_mode = E1000E_INT_MODE_MSI;
2184 e1000e_set_interrupt_capability(adapter);
2186 if (adapter->flags & FLAG_MSI_ENABLED) {
2187 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2188 netdev->name, netdev);
2192 /* fall back to legacy interrupt */
2193 e1000e_reset_interrupt_capability(adapter);
2194 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2197 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2198 netdev->name, netdev);
2200 e_err("Unable to allocate interrupt, Error: %d\n", err);
2205 static void e1000_free_irq(struct e1000_adapter *adapter)
2207 struct net_device *netdev = adapter->netdev;
2209 if (adapter->msix_entries) {
2212 free_irq(adapter->msix_entries[vector].vector, netdev);
2215 free_irq(adapter->msix_entries[vector].vector, netdev);
2218 /* Other Causes interrupt vector */
2219 free_irq(adapter->msix_entries[vector].vector, netdev);
2223 free_irq(adapter->pdev->irq, netdev);
2227 * e1000_irq_disable - Mask off interrupt generation on the NIC
2229 static void e1000_irq_disable(struct e1000_adapter *adapter)
2231 struct e1000_hw *hw = &adapter->hw;
2234 if (adapter->msix_entries)
2235 ew32(EIAC_82574, 0);
2238 if (adapter->msix_entries) {
2241 for (i = 0; i < adapter->num_vectors; i++)
2242 synchronize_irq(adapter->msix_entries[i].vector);
2244 synchronize_irq(adapter->pdev->irq);
2249 * e1000_irq_enable - Enable default interrupt generation settings
2251 static void e1000_irq_enable(struct e1000_adapter *adapter)
2253 struct e1000_hw *hw = &adapter->hw;
2255 if (adapter->msix_entries) {
2256 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2257 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER |
2259 } else if (hw->mac.type >= e1000_pch_lpt) {
2260 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2262 ew32(IMS, IMS_ENABLE_MASK);
2268 * e1000e_get_hw_control - get control of the h/w from f/w
2269 * @adapter: address of board private structure
2271 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2272 * For ASF and Pass Through versions of f/w this means that
2273 * the driver is loaded. For AMT version (only with 82573)
2274 * of the f/w this means that the network i/f is open.
2276 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2278 struct e1000_hw *hw = &adapter->hw;
2282 /* Let firmware know the driver has taken over */
2283 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2285 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2286 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2287 ctrl_ext = er32(CTRL_EXT);
2288 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2293 * e1000e_release_hw_control - release control of the h/w to f/w
2294 * @adapter: address of board private structure
2296 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2297 * For ASF and Pass Through versions of f/w this means that the
2298 * driver is no longer loaded. For AMT version (only with 82573) i
2299 * of the f/w this means that the network i/f is closed.
2302 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2304 struct e1000_hw *hw = &adapter->hw;
2308 /* Let firmware taken over control of h/w */
2309 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2311 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2312 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2313 ctrl_ext = er32(CTRL_EXT);
2314 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2319 * e1000_alloc_ring_dma - allocate memory for a ring structure
2321 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2322 struct e1000_ring *ring)
2324 struct pci_dev *pdev = adapter->pdev;
2326 ring->desc = dma_zalloc_coherent(&pdev->dev, ring->size, &ring->dma,
2335 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2336 * @tx_ring: Tx descriptor ring
2338 * Return 0 on success, negative on failure
2340 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2342 struct e1000_adapter *adapter = tx_ring->adapter;
2343 int err = -ENOMEM, size;
2345 size = sizeof(struct e1000_buffer) * tx_ring->count;
2346 tx_ring->buffer_info = vzalloc(size);
2347 if (!tx_ring->buffer_info)
2350 /* round up to nearest 4K */
2351 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2352 tx_ring->size = ALIGN(tx_ring->size, 4096);
2354 err = e1000_alloc_ring_dma(adapter, tx_ring);
2358 tx_ring->next_to_use = 0;
2359 tx_ring->next_to_clean = 0;
2363 vfree(tx_ring->buffer_info);
2364 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2369 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2370 * @rx_ring: Rx descriptor ring
2372 * Returns 0 on success, negative on failure
2374 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2376 struct e1000_adapter *adapter = rx_ring->adapter;
2377 struct e1000_buffer *buffer_info;
2378 int i, size, desc_len, err = -ENOMEM;
2380 size = sizeof(struct e1000_buffer) * rx_ring->count;
2381 rx_ring->buffer_info = vzalloc(size);
2382 if (!rx_ring->buffer_info)
2385 for (i = 0; i < rx_ring->count; i++) {
2386 buffer_info = &rx_ring->buffer_info[i];
2387 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2388 sizeof(struct e1000_ps_page),
2390 if (!buffer_info->ps_pages)
2394 desc_len = sizeof(union e1000_rx_desc_packet_split);
2396 /* Round up to nearest 4K */
2397 rx_ring->size = rx_ring->count * desc_len;
2398 rx_ring->size = ALIGN(rx_ring->size, 4096);
2400 err = e1000_alloc_ring_dma(adapter, rx_ring);
2404 rx_ring->next_to_clean = 0;
2405 rx_ring->next_to_use = 0;
2406 rx_ring->rx_skb_top = NULL;
2411 for (i = 0; i < rx_ring->count; i++) {
2412 buffer_info = &rx_ring->buffer_info[i];
2413 kfree(buffer_info->ps_pages);
2416 vfree(rx_ring->buffer_info);
2417 e_err("Unable to allocate memory for the receive descriptor ring\n");
2422 * e1000_clean_tx_ring - Free Tx Buffers
2423 * @tx_ring: Tx descriptor ring
2425 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2427 struct e1000_adapter *adapter = tx_ring->adapter;
2428 struct e1000_buffer *buffer_info;
2432 for (i = 0; i < tx_ring->count; i++) {
2433 buffer_info = &tx_ring->buffer_info[i];
2434 e1000_put_txbuf(tx_ring, buffer_info, false);
2437 netdev_reset_queue(adapter->netdev);
2438 size = sizeof(struct e1000_buffer) * tx_ring->count;
2439 memset(tx_ring->buffer_info, 0, size);
2441 memset(tx_ring->desc, 0, tx_ring->size);
2443 tx_ring->next_to_use = 0;
2444 tx_ring->next_to_clean = 0;
2448 * e1000e_free_tx_resources - Free Tx Resources per Queue
2449 * @tx_ring: Tx descriptor ring
2451 * Free all transmit software resources
2453 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2455 struct e1000_adapter *adapter = tx_ring->adapter;
2456 struct pci_dev *pdev = adapter->pdev;
2458 e1000_clean_tx_ring(tx_ring);
2460 vfree(tx_ring->buffer_info);
2461 tx_ring->buffer_info = NULL;
2463 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2465 tx_ring->desc = NULL;
2469 * e1000e_free_rx_resources - Free Rx Resources
2470 * @rx_ring: Rx descriptor ring
2472 * Free all receive software resources
2474 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2476 struct e1000_adapter *adapter = rx_ring->adapter;
2477 struct pci_dev *pdev = adapter->pdev;
2480 e1000_clean_rx_ring(rx_ring);
2482 for (i = 0; i < rx_ring->count; i++)
2483 kfree(rx_ring->buffer_info[i].ps_pages);
2485 vfree(rx_ring->buffer_info);
2486 rx_ring->buffer_info = NULL;
2488 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2490 rx_ring->desc = NULL;
2494 * e1000_update_itr - update the dynamic ITR value based on statistics
2495 * @adapter: pointer to adapter
2496 * @itr_setting: current adapter->itr
2497 * @packets: the number of packets during this measurement interval
2498 * @bytes: the number of bytes during this measurement interval
2500 * Stores a new ITR value based on packets and byte
2501 * counts during the last interrupt. The advantage of per interrupt
2502 * computation is faster updates and more accurate ITR for the current
2503 * traffic pattern. Constants in this function were computed
2504 * based on theoretical maximum wire speed and thresholds were set based
2505 * on testing data as well as attempting to minimize response time
2506 * while increasing bulk throughput. This functionality is controlled
2507 * by the InterruptThrottleRate module parameter.
2509 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2511 unsigned int retval = itr_setting;
2516 switch (itr_setting) {
2517 case lowest_latency:
2518 /* handle TSO and jumbo frames */
2519 if (bytes / packets > 8000)
2520 retval = bulk_latency;
2521 else if ((packets < 5) && (bytes > 512))
2522 retval = low_latency;
2524 case low_latency: /* 50 usec aka 20000 ints/s */
2525 if (bytes > 10000) {
2526 /* this if handles the TSO accounting */
2527 if (bytes / packets > 8000)
2528 retval = bulk_latency;
2529 else if ((packets < 10) || ((bytes / packets) > 1200))
2530 retval = bulk_latency;
2531 else if ((packets > 35))
2532 retval = lowest_latency;
2533 } else if (bytes / packets > 2000) {
2534 retval = bulk_latency;
2535 } else if (packets <= 2 && bytes < 512) {
2536 retval = lowest_latency;
2539 case bulk_latency: /* 250 usec aka 4000 ints/s */
2540 if (bytes > 25000) {
2542 retval = low_latency;
2543 } else if (bytes < 6000) {
2544 retval = low_latency;
2552 static void e1000_set_itr(struct e1000_adapter *adapter)
2555 u32 new_itr = adapter->itr;
2557 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2558 if (adapter->link_speed != SPEED_1000) {
2564 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2569 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2570 adapter->total_tx_packets,
2571 adapter->total_tx_bytes);
2572 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2573 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2574 adapter->tx_itr = low_latency;
2576 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2577 adapter->total_rx_packets,
2578 adapter->total_rx_bytes);
2579 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2580 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2581 adapter->rx_itr = low_latency;
2583 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2585 /* counts and packets in update_itr are dependent on these numbers */
2586 switch (current_itr) {
2587 case lowest_latency:
2591 new_itr = 20000; /* aka hwitr = ~200 */
2601 if (new_itr != adapter->itr) {
2602 /* this attempts to bias the interrupt rate towards Bulk
2603 * by adding intermediate steps when interrupt rate is
2606 new_itr = new_itr > adapter->itr ?
2607 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2608 adapter->itr = new_itr;
2609 adapter->rx_ring->itr_val = new_itr;
2610 if (adapter->msix_entries)
2611 adapter->rx_ring->set_itr = 1;
2613 e1000e_write_itr(adapter, new_itr);
2618 * e1000e_write_itr - write the ITR value to the appropriate registers
2619 * @adapter: address of board private structure
2620 * @itr: new ITR value to program
2622 * e1000e_write_itr determines if the adapter is in MSI-X mode
2623 * and, if so, writes the EITR registers with the ITR value.
2624 * Otherwise, it writes the ITR value into the ITR register.
2626 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2628 struct e1000_hw *hw = &adapter->hw;
2629 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2631 if (adapter->msix_entries) {
2634 for (vector = 0; vector < adapter->num_vectors; vector++)
2635 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2642 * e1000_alloc_queues - Allocate memory for all rings
2643 * @adapter: board private structure to initialize
2645 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2647 int size = sizeof(struct e1000_ring);
2649 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2650 if (!adapter->tx_ring)
2652 adapter->tx_ring->count = adapter->tx_ring_count;
2653 adapter->tx_ring->adapter = adapter;
2655 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2656 if (!adapter->rx_ring)
2658 adapter->rx_ring->count = adapter->rx_ring_count;
2659 adapter->rx_ring->adapter = adapter;
2663 e_err("Unable to allocate memory for queues\n");
2664 kfree(adapter->rx_ring);
2665 kfree(adapter->tx_ring);
2670 * e1000e_poll - NAPI Rx polling callback
2671 * @napi: struct associated with this polling callback
2672 * @weight: number of packets driver is allowed to process this poll
2674 static int e1000e_poll(struct napi_struct *napi, int weight)
2676 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2678 struct e1000_hw *hw = &adapter->hw;
2679 struct net_device *poll_dev = adapter->netdev;
2680 int tx_cleaned = 1, work_done = 0;
2682 adapter = netdev_priv(poll_dev);
2684 if (!adapter->msix_entries ||
2685 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2686 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2688 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2693 /* If weight not fully consumed, exit the polling mode */
2694 if (work_done < weight) {
2695 if (adapter->itr_setting & 3)
2696 e1000_set_itr(adapter);
2697 napi_complete_done(napi, work_done);
2698 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2699 if (adapter->msix_entries)
2700 ew32(IMS, adapter->rx_ring->ims_val);
2702 e1000_irq_enable(adapter);
2709 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2710 __always_unused __be16 proto, u16 vid)
2712 struct e1000_adapter *adapter = netdev_priv(netdev);
2713 struct e1000_hw *hw = &adapter->hw;
2716 /* don't update vlan cookie if already programmed */
2717 if ((adapter->hw.mng_cookie.status &
2718 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2719 (vid == adapter->mng_vlan_id))
2722 /* add VID to filter table */
2723 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2724 index = (vid >> 5) & 0x7F;
2725 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2726 vfta |= BIT((vid & 0x1F));
2727 hw->mac.ops.write_vfta(hw, index, vfta);
2730 set_bit(vid, adapter->active_vlans);
2735 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2736 __always_unused __be16 proto, u16 vid)
2738 struct e1000_adapter *adapter = netdev_priv(netdev);
2739 struct e1000_hw *hw = &adapter->hw;
2742 if ((adapter->hw.mng_cookie.status &
2743 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2744 (vid == adapter->mng_vlan_id)) {
2745 /* release control to f/w */
2746 e1000e_release_hw_control(adapter);
2750 /* remove VID from filter table */
2751 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2752 index = (vid >> 5) & 0x7F;
2753 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2754 vfta &= ~BIT((vid & 0x1F));
2755 hw->mac.ops.write_vfta(hw, index, vfta);
2758 clear_bit(vid, adapter->active_vlans);
2764 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2765 * @adapter: board private structure to initialize
2767 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2769 struct net_device *netdev = adapter->netdev;
2770 struct e1000_hw *hw = &adapter->hw;
2773 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2774 /* disable VLAN receive filtering */
2776 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2779 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2780 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2781 adapter->mng_vlan_id);
2782 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2788 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2789 * @adapter: board private structure to initialize
2791 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2793 struct e1000_hw *hw = &adapter->hw;
2796 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2797 /* enable VLAN receive filtering */
2799 rctl |= E1000_RCTL_VFE;
2800 rctl &= ~E1000_RCTL_CFIEN;
2806 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2807 * @adapter: board private structure to initialize
2809 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2811 struct e1000_hw *hw = &adapter->hw;
2814 /* disable VLAN tag insert/strip */
2816 ctrl &= ~E1000_CTRL_VME;
2821 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2822 * @adapter: board private structure to initialize
2824 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2826 struct e1000_hw *hw = &adapter->hw;
2829 /* enable VLAN tag insert/strip */
2831 ctrl |= E1000_CTRL_VME;
2835 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2837 struct net_device *netdev = adapter->netdev;
2838 u16 vid = adapter->hw.mng_cookie.vlan_id;
2839 u16 old_vid = adapter->mng_vlan_id;
2841 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2842 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2843 adapter->mng_vlan_id = vid;
2846 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2847 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2850 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2854 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2856 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2857 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2860 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2862 struct e1000_hw *hw = &adapter->hw;
2863 u32 manc, manc2h, mdef, i, j;
2865 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2870 /* enable receiving management packets to the host. this will probably
2871 * generate destination unreachable messages from the host OS, but
2872 * the packets will be handled on SMBUS
2874 manc |= E1000_MANC_EN_MNG2HOST;
2875 manc2h = er32(MANC2H);
2877 switch (hw->mac.type) {
2879 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2883 /* Check if IPMI pass-through decision filter already exists;
2886 for (i = 0, j = 0; i < 8; i++) {
2887 mdef = er32(MDEF(i));
2889 /* Ignore filters with anything other than IPMI ports */
2890 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2893 /* Enable this decision filter in MANC2H */
2900 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2903 /* Create new decision filter in an empty filter */
2904 for (i = 0, j = 0; i < 8; i++)
2905 if (er32(MDEF(i)) == 0) {
2906 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2907 E1000_MDEF_PORT_664));
2914 e_warn("Unable to create IPMI pass-through filter\n");
2918 ew32(MANC2H, manc2h);
2923 * e1000_configure_tx - Configure Transmit Unit after Reset
2924 * @adapter: board private structure
2926 * Configure the Tx unit of the MAC after a reset.
2928 static void e1000_configure_tx(struct e1000_adapter *adapter)
2930 struct e1000_hw *hw = &adapter->hw;
2931 struct e1000_ring *tx_ring = adapter->tx_ring;
2933 u32 tdlen, tctl, tarc;
2935 /* Setup the HW Tx Head and Tail descriptor pointers */
2936 tdba = tx_ring->dma;
2937 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2938 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2939 ew32(TDBAH(0), (tdba >> 32));
2940 ew32(TDLEN(0), tdlen);
2943 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2944 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2946 writel(0, tx_ring->head);
2947 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2948 e1000e_update_tdt_wa(tx_ring, 0);
2950 writel(0, tx_ring->tail);
2952 /* Set the Tx Interrupt Delay register */
2953 ew32(TIDV, adapter->tx_int_delay);
2954 /* Tx irq moderation */
2955 ew32(TADV, adapter->tx_abs_int_delay);
2957 if (adapter->flags2 & FLAG2_DMA_BURST) {
2958 u32 txdctl = er32(TXDCTL(0));
2960 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2961 E1000_TXDCTL_WTHRESH);
2962 /* set up some performance related parameters to encourage the
2963 * hardware to use the bus more efficiently in bursts, depends
2964 * on the tx_int_delay to be enabled,
2965 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2966 * hthresh = 1 ==> prefetch when one or more available
2967 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2968 * BEWARE: this seems to work but should be considered first if
2969 * there are Tx hangs or other Tx related bugs
2971 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2972 ew32(TXDCTL(0), txdctl);
2974 /* erratum work around: set txdctl the same for both queues */
2975 ew32(TXDCTL(1), er32(TXDCTL(0)));
2977 /* Program the Transmit Control Register */
2979 tctl &= ~E1000_TCTL_CT;
2980 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2981 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2983 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2984 tarc = er32(TARC(0));
2985 /* set the speed mode bit, we'll clear it if we're not at
2986 * gigabit link later
2988 #define SPEED_MODE_BIT BIT(21)
2989 tarc |= SPEED_MODE_BIT;
2990 ew32(TARC(0), tarc);
2993 /* errata: program both queues to unweighted RR */
2994 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2995 tarc = er32(TARC(0));
2997 ew32(TARC(0), tarc);
2998 tarc = er32(TARC(1));
3000 ew32(TARC(1), tarc);
3003 /* Setup Transmit Descriptor Settings for eop descriptor */
3004 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
3006 /* only set IDE if we are delaying interrupts using the timers */
3007 if (adapter->tx_int_delay)
3008 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3010 /* enable Report Status bit */
3011 adapter->txd_cmd |= E1000_TXD_CMD_RS;
3015 hw->mac.ops.config_collision_dist(hw);
3017 /* SPT and KBL Si errata workaround to avoid data corruption */
3018 if (hw->mac.type == e1000_pch_spt) {
3021 reg_val = er32(IOSFPC);
3022 reg_val |= E1000_RCTL_RDMTS_HEX;
3023 ew32(IOSFPC, reg_val);
3025 reg_val = er32(TARC(0));
3026 /* SPT and KBL Si errata workaround to avoid Tx hang.
3027 * Dropping the number of outstanding requests from
3028 * 3 to 2 in order to avoid a buffer overrun.
3030 reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
3031 reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
3032 ew32(TARC(0), reg_val);
3037 * e1000_setup_rctl - configure the receive control registers
3038 * @adapter: Board private structure
3040 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3041 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3042 static void e1000_setup_rctl(struct e1000_adapter *adapter)
3044 struct e1000_hw *hw = &adapter->hw;
3048 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3049 * If jumbo frames not set, program related MAC/PHY registers
3052 if (hw->mac.type >= e1000_pch2lan) {
3055 if (adapter->netdev->mtu > ETH_DATA_LEN)
3056 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3058 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3061 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3064 /* Program MC offset vector base */
3066 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3067 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3068 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3069 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3071 /* Do not Store bad packets */
3072 rctl &= ~E1000_RCTL_SBP;
3074 /* Enable Long Packet receive */
3075 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3076 rctl &= ~E1000_RCTL_LPE;
3078 rctl |= E1000_RCTL_LPE;
3080 /* Some systems expect that the CRC is included in SMBUS traffic. The
3081 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3082 * host memory when this is enabled
3084 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3085 rctl |= E1000_RCTL_SECRC;
3087 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3088 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3091 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3094 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3096 e1e_rphy(hw, 22, &phy_data);
3098 phy_data |= BIT(14);
3099 e1e_wphy(hw, 0x10, 0x2823);
3100 e1e_wphy(hw, 0x11, 0x0003);
3101 e1e_wphy(hw, 22, phy_data);
3104 /* Setup buffer sizes */
3105 rctl &= ~E1000_RCTL_SZ_4096;
3106 rctl |= E1000_RCTL_BSEX;
3107 switch (adapter->rx_buffer_len) {
3110 rctl |= E1000_RCTL_SZ_2048;
3111 rctl &= ~E1000_RCTL_BSEX;
3114 rctl |= E1000_RCTL_SZ_4096;
3117 rctl |= E1000_RCTL_SZ_8192;
3120 rctl |= E1000_RCTL_SZ_16384;
3124 /* Enable Extended Status in all Receive Descriptors */
3125 rfctl = er32(RFCTL);
3126 rfctl |= E1000_RFCTL_EXTEN;
3129 /* 82571 and greater support packet-split where the protocol
3130 * header is placed in skb->data and the packet data is
3131 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3132 * In the case of a non-split, skb->data is linearly filled,
3133 * followed by the page buffers. Therefore, skb->data is
3134 * sized to hold the largest protocol header.
3136 * allocations using alloc_page take too long for regular MTU
3137 * so only enable packet split for jumbo frames
3139 * Using pages when the page size is greater than 16k wastes
3140 * a lot of memory, since we allocate 3 pages at all times
3143 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3144 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3145 adapter->rx_ps_pages = pages;
3147 adapter->rx_ps_pages = 0;
3149 if (adapter->rx_ps_pages) {
3152 /* Enable Packet split descriptors */
3153 rctl |= E1000_RCTL_DTYP_PS;
3155 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3157 switch (adapter->rx_ps_pages) {
3159 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3162 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3165 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3169 ew32(PSRCTL, psrctl);
3172 /* This is useful for sniffing bad packets. */
3173 if (adapter->netdev->features & NETIF_F_RXALL) {
3174 /* UPE and MPE will be handled by normal PROMISC logic
3175 * in e1000e_set_rx_mode
3177 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3178 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3179 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3181 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3182 E1000_RCTL_DPF | /* Allow filtered pause */
3183 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3184 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3185 * and that breaks VLANs.
3190 /* just started the receive unit, no need to restart */
3191 adapter->flags &= ~FLAG_RESTART_NOW;
3195 * e1000_configure_rx - Configure Receive Unit after Reset
3196 * @adapter: board private structure
3198 * Configure the Rx unit of the MAC after a reset.
3200 static void e1000_configure_rx(struct e1000_adapter *adapter)
3202 struct e1000_hw *hw = &adapter->hw;
3203 struct e1000_ring *rx_ring = adapter->rx_ring;
3205 u32 rdlen, rctl, rxcsum, ctrl_ext;
3207 if (adapter->rx_ps_pages) {
3208 /* this is a 32 byte descriptor */
3209 rdlen = rx_ring->count *
3210 sizeof(union e1000_rx_desc_packet_split);
3211 adapter->clean_rx = e1000_clean_rx_irq_ps;
3212 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3213 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3214 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3215 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3216 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3218 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3219 adapter->clean_rx = e1000_clean_rx_irq;
3220 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3223 /* disable receives while setting up the descriptors */
3225 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3226 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3228 usleep_range(10000, 20000);
3230 if (adapter->flags2 & FLAG2_DMA_BURST) {
3231 /* set the writeback threshold (only takes effect if the RDTR
3232 * is set). set GRAN=1 and write back up to 0x4 worth, and
3233 * enable prefetching of 0x20 Rx descriptors
3239 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3240 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3243 /* set the Receive Delay Timer Register */
3244 ew32(RDTR, adapter->rx_int_delay);
3246 /* irq moderation */
3247 ew32(RADV, adapter->rx_abs_int_delay);
3248 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3249 e1000e_write_itr(adapter, adapter->itr);
3251 ctrl_ext = er32(CTRL_EXT);
3252 /* Auto-Mask interrupts upon ICR access */
3253 ctrl_ext |= E1000_CTRL_EXT_IAME;
3254 ew32(IAM, 0xffffffff);
3255 ew32(CTRL_EXT, ctrl_ext);
3258 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3259 * the Base and Length of the Rx Descriptor Ring
3261 rdba = rx_ring->dma;
3262 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3263 ew32(RDBAH(0), (rdba >> 32));
3264 ew32(RDLEN(0), rdlen);
3267 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3268 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3270 writel(0, rx_ring->head);
3271 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3272 e1000e_update_rdt_wa(rx_ring, 0);
3274 writel(0, rx_ring->tail);
3276 /* Enable Receive Checksum Offload for TCP and UDP */
3277 rxcsum = er32(RXCSUM);
3278 if (adapter->netdev->features & NETIF_F_RXCSUM)
3279 rxcsum |= E1000_RXCSUM_TUOFL;
3281 rxcsum &= ~E1000_RXCSUM_TUOFL;
3282 ew32(RXCSUM, rxcsum);
3284 /* With jumbo frames, excessive C-state transition latencies result
3285 * in dropped transactions.
3287 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3289 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3290 adapter->max_frame_size) * 8 / 1000;
3292 if (adapter->flags & FLAG_IS_ICH) {
3293 u32 rxdctl = er32(RXDCTL(0));
3295 ew32(RXDCTL(0), rxdctl | 0x3 | BIT(8));
3298 dev_info(&adapter->pdev->dev,
3299 "Some CPU C-states have been disabled in order to enable jumbo frames\n");
3300 pm_qos_update_request(&adapter->pm_qos_req, lat);
3302 pm_qos_update_request(&adapter->pm_qos_req,
3303 PM_QOS_DEFAULT_VALUE);
3306 /* Enable Receives */
3311 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3312 * @netdev: network interface device structure
3314 * Writes multicast address list to the MTA hash table.
3315 * Returns: -ENOMEM on failure
3316 * 0 on no addresses written
3317 * X on writing X addresses to MTA
3319 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3321 struct e1000_adapter *adapter = netdev_priv(netdev);
3322 struct e1000_hw *hw = &adapter->hw;
3323 struct netdev_hw_addr *ha;
3327 if (netdev_mc_empty(netdev)) {
3328 /* nothing to program, so clear mc list */
3329 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3333 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3337 /* update_mc_addr_list expects a packed array of only addresses. */
3339 netdev_for_each_mc_addr(ha, netdev)
3340 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3342 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3345 return netdev_mc_count(netdev);
3349 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3350 * @netdev: network interface device structure
3352 * Writes unicast address list to the RAR table.
3353 * Returns: -ENOMEM on failure/insufficient address space
3354 * 0 on no addresses written
3355 * X on writing X addresses to the RAR table
3357 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3359 struct e1000_adapter *adapter = netdev_priv(netdev);
3360 struct e1000_hw *hw = &adapter->hw;
3361 unsigned int rar_entries;
3364 rar_entries = hw->mac.ops.rar_get_count(hw);
3366 /* save a rar entry for our hardware address */
3369 /* save a rar entry for the LAA workaround */
3370 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3373 /* return ENOMEM indicating insufficient memory for addresses */
3374 if (netdev_uc_count(netdev) > rar_entries)
3377 if (!netdev_uc_empty(netdev) && rar_entries) {
3378 struct netdev_hw_addr *ha;
3380 /* write the addresses in reverse order to avoid write
3383 netdev_for_each_uc_addr(ha, netdev) {
3388 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3395 /* zero out the remaining RAR entries not used above */
3396 for (; rar_entries > 0; rar_entries--) {
3397 ew32(RAH(rar_entries), 0);
3398 ew32(RAL(rar_entries), 0);
3406 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3407 * @netdev: network interface device structure
3409 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3410 * address list or the network interface flags are updated. This routine is
3411 * responsible for configuring the hardware for proper unicast, multicast,
3412 * promiscuous mode, and all-multi behavior.
3414 static void e1000e_set_rx_mode(struct net_device *netdev)
3416 struct e1000_adapter *adapter = netdev_priv(netdev);
3417 struct e1000_hw *hw = &adapter->hw;
3420 if (pm_runtime_suspended(netdev->dev.parent))
3423 /* Check for Promiscuous and All Multicast modes */
3426 /* clear the affected bits */
3427 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3429 if (netdev->flags & IFF_PROMISC) {
3430 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3431 /* Do not hardware filter VLANs in promisc mode */
3432 e1000e_vlan_filter_disable(adapter);
3436 if (netdev->flags & IFF_ALLMULTI) {
3437 rctl |= E1000_RCTL_MPE;
3439 /* Write addresses to the MTA, if the attempt fails
3440 * then we should just turn on promiscuous mode so
3441 * that we can at least receive multicast traffic
3443 count = e1000e_write_mc_addr_list(netdev);
3445 rctl |= E1000_RCTL_MPE;
3447 e1000e_vlan_filter_enable(adapter);
3448 /* Write addresses to available RAR registers, if there is not
3449 * sufficient space to store all the addresses then enable
3450 * unicast promiscuous mode
3452 count = e1000e_write_uc_addr_list(netdev);
3454 rctl |= E1000_RCTL_UPE;
3459 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3460 e1000e_vlan_strip_enable(adapter);
3462 e1000e_vlan_strip_disable(adapter);
3465 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3467 struct e1000_hw *hw = &adapter->hw;
3472 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3473 for (i = 0; i < 10; i++)
3474 ew32(RSSRK(i), rss_key[i]);
3476 /* Direct all traffic to queue 0 */
3477 for (i = 0; i < 32; i++)
3480 /* Disable raw packet checksumming so that RSS hash is placed in
3481 * descriptor on writeback.
3483 rxcsum = er32(RXCSUM);
3484 rxcsum |= E1000_RXCSUM_PCSD;
3486 ew32(RXCSUM, rxcsum);
3488 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3489 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3490 E1000_MRQC_RSS_FIELD_IPV6 |
3491 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3492 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3498 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3499 * @adapter: board private structure
3500 * @timinca: pointer to returned time increment attributes
3502 * Get attributes for incrementing the System Time Register SYSTIML/H at
3503 * the default base frequency, and set the cyclecounter shift value.
3505 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3507 struct e1000_hw *hw = &adapter->hw;
3508 u32 incvalue, incperiod, shift;
3510 /* Make sure clock is enabled on I217/I218/I219 before checking
3513 if ((hw->mac.type >= e1000_pch_lpt) &&
3514 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3515 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3516 u32 fextnvm7 = er32(FEXTNVM7);
3518 if (!(fextnvm7 & BIT(0))) {
3519 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3524 switch (hw->mac.type) {
3526 /* Stable 96MHz frequency */
3527 incperiod = INCPERIOD_96MHZ;
3528 incvalue = INCVALUE_96MHZ;
3529 shift = INCVALUE_SHIFT_96MHZ;
3530 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3533 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3534 /* Stable 96MHz frequency */
3535 incperiod = INCPERIOD_96MHZ;
3536 incvalue = INCVALUE_96MHZ;
3537 shift = INCVALUE_SHIFT_96MHZ;
3538 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3540 /* Stable 25MHz frequency */
3541 incperiod = INCPERIOD_25MHZ;
3542 incvalue = INCVALUE_25MHZ;
3543 shift = INCVALUE_SHIFT_25MHZ;
3544 adapter->cc.shift = shift;
3548 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3549 /* Stable 24MHz frequency */
3550 incperiod = INCPERIOD_24MHZ;
3551 incvalue = INCVALUE_24MHZ;
3552 shift = INCVALUE_SHIFT_24MHZ;
3553 adapter->cc.shift = shift;
3558 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3559 /* Stable 24MHz frequency */
3560 incperiod = INCPERIOD_24MHZ;
3561 incvalue = INCVALUE_24MHZ;
3562 shift = INCVALUE_SHIFT_24MHZ;
3563 adapter->cc.shift = shift;
3565 /* Stable 38400KHz frequency */
3566 incperiod = INCPERIOD_38400KHZ;
3567 incvalue = INCVALUE_38400KHZ;
3568 shift = INCVALUE_SHIFT_38400KHZ;
3569 adapter->cc.shift = shift;
3574 /* Stable 25MHz frequency */
3575 incperiod = INCPERIOD_25MHZ;
3576 incvalue = INCVALUE_25MHZ;
3577 shift = INCVALUE_SHIFT_25MHZ;
3578 adapter->cc.shift = shift;
3584 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3585 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3591 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3592 * @adapter: board private structure
3594 * Outgoing time stamping can be enabled and disabled. Play nice and
3595 * disable it when requested, although it shouldn't cause any overhead
3596 * when no packet needs it. At most one packet in the queue may be
3597 * marked for time stamping, otherwise it would be impossible to tell
3598 * for sure to which packet the hardware time stamp belongs.
3600 * Incoming time stamping has to be configured via the hardware filters.
3601 * Not all combinations are supported, in particular event type has to be
3602 * specified. Matching the kind of event packet is not supported, with the
3603 * exception of "all V2 events regardless of level 2 or 4".
3605 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3606 struct hwtstamp_config *config)
3608 struct e1000_hw *hw = &adapter->hw;
3609 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3610 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3617 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3620 /* flags reserved for future extensions - must be zero */
3624 switch (config->tx_type) {
3625 case HWTSTAMP_TX_OFF:
3628 case HWTSTAMP_TX_ON:
3634 switch (config->rx_filter) {
3635 case HWTSTAMP_FILTER_NONE:
3638 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3639 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3640 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3643 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3644 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3645 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3648 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3649 /* Also time stamps V2 L2 Path Delay Request/Response */
3650 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3651 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3654 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3655 /* Also time stamps V2 L2 Path Delay Request/Response. */
3656 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3657 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3660 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3661 /* Hardware cannot filter just V2 L4 Sync messages;
3662 * fall-through to V2 (both L2 and L4) Sync.
3664 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3665 /* Also time stamps V2 Path Delay Request/Response. */
3666 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3667 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3671 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3672 /* Hardware cannot filter just V2 L4 Delay Request messages;
3673 * fall-through to V2 (both L2 and L4) Delay Request.
3675 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3676 /* Also time stamps V2 Path Delay Request/Response. */
3677 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3678 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3682 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3683 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3684 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3685 * fall-through to all V2 (both L2 and L4) Events.
3687 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3688 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3689 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3693 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3694 /* For V1, the hardware can only filter Sync messages or
3695 * Delay Request messages but not both so fall-through to
3696 * time stamp all packets.
3698 case HWTSTAMP_FILTER_NTP_ALL:
3699 case HWTSTAMP_FILTER_ALL:
3702 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3703 config->rx_filter = HWTSTAMP_FILTER_ALL;
3709 adapter->hwtstamp_config = *config;
3711 /* enable/disable Tx h/w time stamping */
3712 regval = er32(TSYNCTXCTL);
3713 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3714 regval |= tsync_tx_ctl;
3715 ew32(TSYNCTXCTL, regval);
3716 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3717 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3718 e_err("Timesync Tx Control register not set as expected\n");
3722 /* enable/disable Rx h/w time stamping */
3723 regval = er32(TSYNCRXCTL);
3724 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3725 regval |= tsync_rx_ctl;
3726 ew32(TSYNCRXCTL, regval);
3727 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3728 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3729 (regval & (E1000_TSYNCRXCTL_ENABLED |
3730 E1000_TSYNCRXCTL_TYPE_MASK))) {
3731 e_err("Timesync Rx Control register not set as expected\n");
3735 /* L2: define ethertype filter for time stamped packets */
3737 rxmtrl |= ETH_P_1588;
3739 /* define which PTP packets get time stamped */
3740 ew32(RXMTRL, rxmtrl);
3742 /* Filter by destination port */
3744 rxudp = PTP_EV_PORT;
3745 cpu_to_be16s(&rxudp);
3751 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3759 * e1000_configure - configure the hardware for Rx and Tx
3760 * @adapter: private board structure
3762 static void e1000_configure(struct e1000_adapter *adapter)
3764 struct e1000_ring *rx_ring = adapter->rx_ring;
3766 e1000e_set_rx_mode(adapter->netdev);
3768 e1000_restore_vlan(adapter);
3769 e1000_init_manageability_pt(adapter);
3771 e1000_configure_tx(adapter);
3773 if (adapter->netdev->features & NETIF_F_RXHASH)
3774 e1000e_setup_rss_hash(adapter);
3775 e1000_setup_rctl(adapter);
3776 e1000_configure_rx(adapter);
3777 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3781 * e1000e_power_up_phy - restore link in case the phy was powered down
3782 * @adapter: address of board private structure
3784 * The phy may be powered down to save power and turn off link when the
3785 * driver is unloaded and wake on lan is not enabled (among others)
3786 * *** this routine MUST be followed by a call to e1000e_reset ***
3788 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3790 if (adapter->hw.phy.ops.power_up)
3791 adapter->hw.phy.ops.power_up(&adapter->hw);
3793 adapter->hw.mac.ops.setup_link(&adapter->hw);
3797 * e1000_power_down_phy - Power down the PHY
3799 * Power down the PHY so no link is implied when interface is down.
3800 * The PHY cannot be powered down if management or WoL is active.
3802 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3804 if (adapter->hw.phy.ops.power_down)
3805 adapter->hw.phy.ops.power_down(&adapter->hw);
3809 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3811 * We want to clear all pending descriptors from the TX ring.
3812 * zeroing happens when the HW reads the regs. We assign the ring itself as
3813 * the data of the next descriptor. We don't care about the data we are about
3816 static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3818 struct e1000_hw *hw = &adapter->hw;
3819 struct e1000_ring *tx_ring = adapter->tx_ring;
3820 struct e1000_tx_desc *tx_desc = NULL;
3821 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3825 ew32(TCTL, tctl | E1000_TCTL_EN);
3827 BUG_ON(tdt != tx_ring->next_to_use);
3828 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3829 tx_desc->buffer_addr = tx_ring->dma;
3831 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3832 tx_desc->upper.data = 0;
3833 /* flush descriptors to memory before notifying the HW */
3835 tx_ring->next_to_use++;
3836 if (tx_ring->next_to_use == tx_ring->count)
3837 tx_ring->next_to_use = 0;
3838 ew32(TDT(0), tx_ring->next_to_use);
3840 usleep_range(200, 250);
3844 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3846 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3848 static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3851 struct e1000_hw *hw = &adapter->hw;
3854 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3856 usleep_range(100, 150);
3858 rxdctl = er32(RXDCTL(0));
3859 /* zero the lower 14 bits (prefetch and host thresholds) */
3860 rxdctl &= 0xffffc000;
3862 /* update thresholds: prefetch threshold to 31, host threshold to 1
3863 * and make sure the granularity is "descriptors" and not "cache lines"
3865 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3867 ew32(RXDCTL(0), rxdctl);
3868 /* momentarily enable the RX ring for the changes to take effect */
3869 ew32(RCTL, rctl | E1000_RCTL_EN);
3871 usleep_range(100, 150);
3872 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3876 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3878 * In i219, the descriptor rings must be emptied before resetting the HW
3879 * or before changing the device state to D3 during runtime (runtime PM).
3881 * Failure to do this will cause the HW to enter a unit hang state which can
3882 * only be released by PCI reset on the device
3886 static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3889 u32 fext_nvm11, tdlen;
3890 struct e1000_hw *hw = &adapter->hw;
3892 /* First, disable MULR fix in FEXTNVM11 */
3893 fext_nvm11 = er32(FEXTNVM11);
3894 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3895 ew32(FEXTNVM11, fext_nvm11);
3896 /* do nothing if we're not in faulty state, or if the queue is empty */
3897 tdlen = er32(TDLEN(0));
3898 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3900 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3902 e1000_flush_tx_ring(adapter);
3903 /* recheck, maybe the fault is caused by the rx ring */
3904 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3906 if (hang_state & FLUSH_DESC_REQUIRED)
3907 e1000_flush_rx_ring(adapter);
3911 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3912 * @adapter: board private structure
3914 * When the MAC is reset, all hardware bits for timesync will be reset to the
3915 * default values. This function will restore the settings last in place.
3916 * Since the clock SYSTIME registers are reset, we will simply restore the
3917 * cyclecounter to the kernel real clock time.
3919 static void e1000e_systim_reset(struct e1000_adapter *adapter)
3921 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3922 struct e1000_hw *hw = &adapter->hw;
3923 unsigned long flags;
3927 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3930 if (info->adjfreq) {
3931 /* restore the previous ptp frequency delta */
3932 ret_val = info->adjfreq(info, adapter->ptp_delta);
3934 /* set the default base frequency if no adjustment possible */
3935 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3937 ew32(TIMINCA, timinca);
3941 dev_warn(&adapter->pdev->dev,
3942 "Failed to restore TIMINCA clock rate delta: %d\n",
3947 /* reset the systim ns time counter */
3948 spin_lock_irqsave(&adapter->systim_lock, flags);
3949 timecounter_init(&adapter->tc, &adapter->cc,
3950 ktime_to_ns(ktime_get_real()));
3951 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3953 /* restore the previous hwtstamp configuration settings */
3954 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3958 * e1000e_reset - bring the hardware into a known good state
3960 * This function boots the hardware and enables some settings that
3961 * require a configuration cycle of the hardware - those cannot be
3962 * set/changed during runtime. After reset the device needs to be
3963 * properly configured for Rx, Tx etc.
3965 void e1000e_reset(struct e1000_adapter *adapter)
3967 struct e1000_mac_info *mac = &adapter->hw.mac;
3968 struct e1000_fc_info *fc = &adapter->hw.fc;
3969 struct e1000_hw *hw = &adapter->hw;
3970 u32 tx_space, min_tx_space, min_rx_space;
3971 u32 pba = adapter->pba;
3974 /* reset Packet Buffer Allocation to default */
3977 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3978 /* To maintain wire speed transmits, the Tx FIFO should be
3979 * large enough to accommodate two full transmit packets,
3980 * rounded up to the next 1KB and expressed in KB. Likewise,
3981 * the Rx FIFO should be large enough to accommodate at least
3982 * one full receive packet and is similarly rounded up and
3986 /* upper 16 bits has Tx packet buffer allocation size in KB */
3987 tx_space = pba >> 16;
3988 /* lower 16 bits has Rx packet buffer allocation size in KB */
3990 /* the Tx fifo also stores 16 bytes of information about the Tx
3991 * but don't include ethernet FCS because hardware appends it
3993 min_tx_space = (adapter->max_frame_size +
3994 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3995 min_tx_space = ALIGN(min_tx_space, 1024);
3996 min_tx_space >>= 10;
3997 /* software strips receive CRC, so leave room for it */
3998 min_rx_space = adapter->max_frame_size;
3999 min_rx_space = ALIGN(min_rx_space, 1024);
4000 min_rx_space >>= 10;
4002 /* If current Tx allocation is less than the min Tx FIFO size,
4003 * and the min Tx FIFO size is less than the current Rx FIFO
4004 * allocation, take space away from current Rx allocation
4006 if ((tx_space < min_tx_space) &&
4007 ((min_tx_space - tx_space) < pba)) {
4008 pba -= min_tx_space - tx_space;
4010 /* if short on Rx space, Rx wins and must trump Tx
4013 if (pba < min_rx_space)
4020 /* flow control settings
4022 * The high water mark must be low enough to fit one full frame
4023 * (or the size used for early receive) above it in the Rx FIFO.
4024 * Set it to the lower of:
4025 * - 90% of the Rx FIFO size, and
4026 * - the full Rx FIFO size minus one full frame
4028 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4029 fc->pause_time = 0xFFFF;
4031 fc->pause_time = E1000_FC_PAUSE_TIME;
4032 fc->send_xon = true;
4033 fc->current_mode = fc->requested_mode;
4035 switch (hw->mac.type) {
4037 case e1000_ich10lan:
4038 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4041 fc->high_water = 0x2800;
4042 fc->low_water = fc->high_water - 8;
4047 hwm = min(((pba << 10) * 9 / 10),
4048 ((pba << 10) - adapter->max_frame_size));
4050 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4051 fc->low_water = fc->high_water - 8;
4054 /* Workaround PCH LOM adapter hangs with certain network
4055 * loads. If hangs persist, try disabling Tx flow control.
4057 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4058 fc->high_water = 0x3500;
4059 fc->low_water = 0x1500;
4061 fc->high_water = 0x5000;
4062 fc->low_water = 0x3000;
4064 fc->refresh_time = 0x1000;
4070 fc->refresh_time = 0x0400;
4072 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4073 fc->high_water = 0x05C20;
4074 fc->low_water = 0x05048;
4075 fc->pause_time = 0x0650;
4081 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4082 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4086 /* Alignment of Tx data is on an arbitrary byte boundary with the
4087 * maximum size per Tx descriptor limited only to the transmit
4088 * allocation of the packet buffer minus 96 bytes with an upper
4089 * limit of 24KB due to receive synchronization limitations.
4091 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4094 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4095 * fit in receive buffer.
4097 if (adapter->itr_setting & 0x3) {
4098 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4099 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4100 dev_info(&adapter->pdev->dev,
4101 "Interrupt Throttle Rate off\n");
4102 adapter->flags2 |= FLAG2_DISABLE_AIM;
4103 e1000e_write_itr(adapter, 0);
4105 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4106 dev_info(&adapter->pdev->dev,
4107 "Interrupt Throttle Rate on\n");
4108 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4109 adapter->itr = 20000;
4110 e1000e_write_itr(adapter, adapter->itr);
4114 if (hw->mac.type >= e1000_pch_spt)
4115 e1000_flush_desc_rings(adapter);
4116 /* Allow time for pending master requests to run */
4117 mac->ops.reset_hw(hw);
4119 /* For parts with AMT enabled, let the firmware know
4120 * that the network interface is in control
4122 if (adapter->flags & FLAG_HAS_AMT)
4123 e1000e_get_hw_control(adapter);
4127 if (mac->ops.init_hw(hw))
4128 e_err("Hardware Error\n");
4130 e1000_update_mng_vlan(adapter);
4132 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4133 ew32(VET, ETH_P_8021Q);
4135 e1000e_reset_adaptive(hw);
4137 /* restore systim and hwtstamp settings */
4138 e1000e_systim_reset(adapter);
4140 /* Set EEE advertisement as appropriate */
4141 if (adapter->flags2 & FLAG2_HAS_EEE) {
4145 switch (hw->phy.type) {
4146 case e1000_phy_82579:
4147 adv_addr = I82579_EEE_ADVERTISEMENT;
4149 case e1000_phy_i217:
4150 adv_addr = I217_EEE_ADVERTISEMENT;
4153 dev_err(&adapter->pdev->dev,
4154 "Invalid PHY type setting EEE advertisement\n");
4158 ret_val = hw->phy.ops.acquire(hw);
4160 dev_err(&adapter->pdev->dev,
4161 "EEE advertisement - unable to acquire PHY\n");
4165 e1000_write_emi_reg_locked(hw, adv_addr,
4166 hw->dev_spec.ich8lan.eee_disable ?
4167 0 : adapter->eee_advert);
4169 hw->phy.ops.release(hw);
4172 if (!netif_running(adapter->netdev) &&
4173 !test_bit(__E1000_TESTING, &adapter->state))
4174 e1000_power_down_phy(adapter);
4176 e1000_get_phy_info(hw);
4178 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4179 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4181 /* speed up time to link by disabling smart power down, ignore
4182 * the return value of this function because there is nothing
4183 * different we would do if it failed
4185 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4186 phy_data &= ~IGP02E1000_PM_SPD;
4187 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4189 if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4192 /* Fextnvm7 @ 0xe4[2] = 1 */
4193 reg = er32(FEXTNVM7);
4194 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4195 ew32(FEXTNVM7, reg);
4196 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4197 reg = er32(FEXTNVM9);
4198 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4199 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4200 ew32(FEXTNVM9, reg);
4206 * e1000e_trigger_lsc - trigger an LSC interrupt
4209 * Fire a link status change interrupt to start the watchdog.
4211 static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4213 struct e1000_hw *hw = &adapter->hw;
4215 if (adapter->msix_entries)
4216 ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
4218 ew32(ICS, E1000_ICS_LSC);
4221 void e1000e_up(struct e1000_adapter *adapter)
4223 /* hardware has been reset, we need to reload some things */
4224 e1000_configure(adapter);
4226 clear_bit(__E1000_DOWN, &adapter->state);
4228 if (adapter->msix_entries)
4229 e1000_configure_msix(adapter);
4230 e1000_irq_enable(adapter);
4232 netif_start_queue(adapter->netdev);
4234 e1000e_trigger_lsc(adapter);
4237 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4239 struct e1000_hw *hw = &adapter->hw;
4241 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4244 /* flush pending descriptor writebacks to memory */
4245 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4246 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4248 /* execute the writes immediately */
4251 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4252 * write is successful
4254 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4255 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4257 /* execute the writes immediately */
4261 static void e1000e_update_stats(struct e1000_adapter *adapter);
4264 * e1000e_down - quiesce the device and optionally reset the hardware
4265 * @adapter: board private structure
4266 * @reset: boolean flag to reset the hardware or not
4268 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4270 struct net_device *netdev = adapter->netdev;
4271 struct e1000_hw *hw = &adapter->hw;
4274 /* signal that we're down so the interrupt handler does not
4275 * reschedule our watchdog timer
4277 set_bit(__E1000_DOWN, &adapter->state);
4279 netif_carrier_off(netdev);
4281 /* disable receives in the hardware */
4283 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4284 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4285 /* flush and sleep below */
4287 netif_stop_queue(netdev);
4289 /* disable transmits in the hardware */
4291 tctl &= ~E1000_TCTL_EN;
4294 /* flush both disables and wait for them to finish */
4296 usleep_range(10000, 20000);
4298 e1000_irq_disable(adapter);
4300 napi_synchronize(&adapter->napi);
4302 del_timer_sync(&adapter->watchdog_timer);
4303 del_timer_sync(&adapter->phy_info_timer);
4305 spin_lock(&adapter->stats64_lock);
4306 e1000e_update_stats(adapter);
4307 spin_unlock(&adapter->stats64_lock);
4309 e1000e_flush_descriptors(adapter);
4311 adapter->link_speed = 0;
4312 adapter->link_duplex = 0;
4314 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4315 if ((hw->mac.type >= e1000_pch2lan) &&
4316 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4317 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4318 e_dbg("failed to disable jumbo frame workaround mode\n");
4320 if (!pci_channel_offline(adapter->pdev)) {
4322 e1000e_reset(adapter);
4323 else if (hw->mac.type >= e1000_pch_spt)
4324 e1000_flush_desc_rings(adapter);
4326 e1000_clean_tx_ring(adapter->tx_ring);
4327 e1000_clean_rx_ring(adapter->rx_ring);
4330 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4333 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4334 usleep_range(1000, 2000);
4335 e1000e_down(adapter, true);
4337 clear_bit(__E1000_RESETTING, &adapter->state);
4341 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4342 * @hw: pointer to the HW structure
4343 * @systim: time value read, sanitized and returned
4345 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4346 * check to see that the time is incrementing at a reasonable
4347 * rate and is a multiple of incvalue.
4349 static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim)
4351 u64 time_delta, rem, temp;
4356 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4357 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4358 /* latch SYSTIMH on read of SYSTIML */
4359 systim_next = (u64)er32(SYSTIML);
4360 systim_next |= (u64)er32(SYSTIMH) << 32;
4362 time_delta = systim_next - systim;
4364 /* VMWare users have seen incvalue of zero, don't div / 0 */
4365 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4367 systim = systim_next;
4369 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4377 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4378 * @cc: cyclecounter structure
4380 static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4382 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4384 struct e1000_hw *hw = &adapter->hw;
4385 u32 systimel, systimeh;
4387 /* SYSTIMH latching upon SYSTIML read does not work well.
4388 * This means that if SYSTIML overflows after we read it but before
4389 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4390 * will experience a huge non linear increment in the systime value
4391 * to fix that we test for overflow and if true, we re-read systime.
4393 systimel = er32(SYSTIML);
4394 systimeh = er32(SYSTIMH);
4395 /* Is systimel is so large that overflow is possible? */
4396 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4397 u32 systimel_2 = er32(SYSTIML);
4398 if (systimel > systimel_2) {
4399 /* There was an overflow, read again SYSTIMH, and use
4402 systimeh = er32(SYSTIMH);
4403 systimel = systimel_2;
4406 systim = (u64)systimel;
4407 systim |= (u64)systimeh << 32;
4409 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4410 systim = e1000e_sanitize_systim(hw, systim);
4416 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4417 * @adapter: board private structure to initialize
4419 * e1000_sw_init initializes the Adapter private data structure.
4420 * Fields are initialized based on PCI device information and
4421 * OS network device settings (MTU size).
4423 static int e1000_sw_init(struct e1000_adapter *adapter)
4425 struct net_device *netdev = adapter->netdev;
4427 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4428 adapter->rx_ps_bsize0 = 128;
4429 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4430 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4431 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4432 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4434 spin_lock_init(&adapter->stats64_lock);
4436 e1000e_set_interrupt_capability(adapter);
4438 if (e1000_alloc_queues(adapter))
4441 /* Setup hardware time stamping cyclecounter */
4442 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4443 adapter->cc.read = e1000e_cyclecounter_read;
4444 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4445 adapter->cc.mult = 1;
4446 /* cc.shift set in e1000e_get_base_tininca() */
4448 spin_lock_init(&adapter->systim_lock);
4449 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4452 /* Explicitly disable IRQ since the NIC can be in any state. */
4453 e1000_irq_disable(adapter);
4455 set_bit(__E1000_DOWN, &adapter->state);
4460 * e1000_intr_msi_test - Interrupt Handler
4461 * @irq: interrupt number
4462 * @data: pointer to a network interface device structure
4464 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4466 struct net_device *netdev = data;
4467 struct e1000_adapter *adapter = netdev_priv(netdev);
4468 struct e1000_hw *hw = &adapter->hw;
4469 u32 icr = er32(ICR);
4471 e_dbg("icr is %08X\n", icr);
4472 if (icr & E1000_ICR_RXSEQ) {
4473 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4474 /* Force memory writes to complete before acknowledging the
4475 * interrupt is handled.
4484 * e1000_test_msi_interrupt - Returns 0 for successful test
4485 * @adapter: board private struct
4487 * code flow taken from tg3.c
4489 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4491 struct net_device *netdev = adapter->netdev;
4492 struct e1000_hw *hw = &adapter->hw;
4495 /* poll_enable hasn't been called yet, so don't need disable */
4496 /* clear any pending events */
4499 /* free the real vector and request a test handler */
4500 e1000_free_irq(adapter);
4501 e1000e_reset_interrupt_capability(adapter);
4503 /* Assume that the test fails, if it succeeds then the test
4504 * MSI irq handler will unset this flag
4506 adapter->flags |= FLAG_MSI_TEST_FAILED;
4508 err = pci_enable_msi(adapter->pdev);
4510 goto msi_test_failed;
4512 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4513 netdev->name, netdev);
4515 pci_disable_msi(adapter->pdev);
4516 goto msi_test_failed;
4519 /* Force memory writes to complete before enabling and firing an
4524 e1000_irq_enable(adapter);
4526 /* fire an unusual interrupt on the test handler */
4527 ew32(ICS, E1000_ICS_RXSEQ);
4531 e1000_irq_disable(adapter);
4533 rmb(); /* read flags after interrupt has been fired */
4535 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4536 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4537 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4539 e_dbg("MSI interrupt test succeeded!\n");
4542 free_irq(adapter->pdev->irq, netdev);
4543 pci_disable_msi(adapter->pdev);
4546 e1000e_set_interrupt_capability(adapter);
4547 return e1000_request_irq(adapter);
4551 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4552 * @adapter: board private struct
4554 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4556 static int e1000_test_msi(struct e1000_adapter *adapter)
4561 if (!(adapter->flags & FLAG_MSI_ENABLED))
4564 /* disable SERR in case the MSI write causes a master abort */
4565 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4566 if (pci_cmd & PCI_COMMAND_SERR)
4567 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4568 pci_cmd & ~PCI_COMMAND_SERR);
4570 err = e1000_test_msi_interrupt(adapter);
4572 /* re-enable SERR */
4573 if (pci_cmd & PCI_COMMAND_SERR) {
4574 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4575 pci_cmd |= PCI_COMMAND_SERR;
4576 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4583 * e1000e_open - Called when a network interface is made active
4584 * @netdev: network interface device structure
4586 * Returns 0 on success, negative value on failure
4588 * The open entry point is called when a network interface is made
4589 * active by the system (IFF_UP). At this point all resources needed
4590 * for transmit and receive operations are allocated, the interrupt
4591 * handler is registered with the OS, the watchdog timer is started,
4592 * and the stack is notified that the interface is ready.
4594 int e1000e_open(struct net_device *netdev)
4596 struct e1000_adapter *adapter = netdev_priv(netdev);
4597 struct e1000_hw *hw = &adapter->hw;
4598 struct pci_dev *pdev = adapter->pdev;
4601 /* disallow open during test */
4602 if (test_bit(__E1000_TESTING, &adapter->state))
4605 pm_runtime_get_sync(&pdev->dev);
4607 netif_carrier_off(netdev);
4609 /* allocate transmit descriptors */
4610 err = e1000e_setup_tx_resources(adapter->tx_ring);
4614 /* allocate receive descriptors */
4615 err = e1000e_setup_rx_resources(adapter->rx_ring);
4619 /* If AMT is enabled, let the firmware know that the network
4620 * interface is now open and reset the part to a known state.
4622 if (adapter->flags & FLAG_HAS_AMT) {
4623 e1000e_get_hw_control(adapter);
4624 e1000e_reset(adapter);
4627 e1000e_power_up_phy(adapter);
4629 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4630 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4631 e1000_update_mng_vlan(adapter);
4633 /* DMA latency requirement to workaround jumbo issue */
4634 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4635 PM_QOS_DEFAULT_VALUE);
4637 /* before we allocate an interrupt, we must be ready to handle it.
4638 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4639 * as soon as we call pci_request_irq, so we have to setup our
4640 * clean_rx handler before we do so.
4642 e1000_configure(adapter);
4644 err = e1000_request_irq(adapter);
4648 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4649 * ignore e1000e MSI messages, which means we need to test our MSI
4652 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4653 err = e1000_test_msi(adapter);
4655 e_err("Interrupt allocation failed\n");
4660 /* From here on the code is the same as e1000e_up() */
4661 clear_bit(__E1000_DOWN, &adapter->state);
4663 napi_enable(&adapter->napi);
4665 e1000_irq_enable(adapter);
4667 adapter->tx_hang_recheck = false;
4668 netif_start_queue(netdev);
4670 hw->mac.get_link_status = true;
4671 pm_runtime_put(&pdev->dev);
4673 e1000e_trigger_lsc(adapter);
4678 pm_qos_remove_request(&adapter->pm_qos_req);
4679 e1000e_release_hw_control(adapter);
4680 e1000_power_down_phy(adapter);
4681 e1000e_free_rx_resources(adapter->rx_ring);
4683 e1000e_free_tx_resources(adapter->tx_ring);
4685 e1000e_reset(adapter);
4686 pm_runtime_put_sync(&pdev->dev);
4692 * e1000e_close - Disables a network interface
4693 * @netdev: network interface device structure
4695 * Returns 0, this is not allowed to fail
4697 * The close entry point is called when an interface is de-activated
4698 * by the OS. The hardware is still under the drivers control, but
4699 * needs to be disabled. A global MAC reset is issued to stop the
4700 * hardware, and all transmit and receive resources are freed.
4702 int e1000e_close(struct net_device *netdev)
4704 struct e1000_adapter *adapter = netdev_priv(netdev);
4705 struct pci_dev *pdev = adapter->pdev;
4706 int count = E1000_CHECK_RESET_COUNT;
4708 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4709 usleep_range(10000, 20000);
4711 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4713 pm_runtime_get_sync(&pdev->dev);
4715 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4716 e1000e_down(adapter, true);
4717 e1000_free_irq(adapter);
4719 /* Link status message must follow this format */
4720 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4723 napi_disable(&adapter->napi);
4725 e1000e_free_tx_resources(adapter->tx_ring);
4726 e1000e_free_rx_resources(adapter->rx_ring);
4728 /* kill manageability vlan ID if supported, but not if a vlan with
4729 * the same ID is registered on the host OS (let 8021q kill it)
4731 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4732 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4733 adapter->mng_vlan_id);
4735 /* If AMT is enabled, let the firmware know that the network
4736 * interface is now closed
4738 if ((adapter->flags & FLAG_HAS_AMT) &&
4739 !test_bit(__E1000_TESTING, &adapter->state))
4740 e1000e_release_hw_control(adapter);
4742 pm_qos_remove_request(&adapter->pm_qos_req);
4744 pm_runtime_put_sync(&pdev->dev);
4750 * e1000_set_mac - Change the Ethernet Address of the NIC
4751 * @netdev: network interface device structure
4752 * @p: pointer to an address structure
4754 * Returns 0 on success, negative on failure
4756 static int e1000_set_mac(struct net_device *netdev, void *p)
4758 struct e1000_adapter *adapter = netdev_priv(netdev);
4759 struct e1000_hw *hw = &adapter->hw;
4760 struct sockaddr *addr = p;
4762 if (!is_valid_ether_addr(addr->sa_data))
4763 return -EADDRNOTAVAIL;
4765 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4766 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4768 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4770 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4771 /* activate the work around */
4772 e1000e_set_laa_state_82571(&adapter->hw, 1);
4774 /* Hold a copy of the LAA in RAR[14] This is done so that
4775 * between the time RAR[0] gets clobbered and the time it
4776 * gets fixed (in e1000_watchdog), the actual LAA is in one
4777 * of the RARs and no incoming packets directed to this port
4778 * are dropped. Eventually the LAA will be in RAR[0] and
4781 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4782 adapter->hw.mac.rar_entry_count - 1);
4789 * e1000e_update_phy_task - work thread to update phy
4790 * @work: pointer to our work struct
4792 * this worker thread exists because we must acquire a
4793 * semaphore to read the phy, which we could msleep while
4794 * waiting for it, and we can't msleep in a timer.
4796 static void e1000e_update_phy_task(struct work_struct *work)
4798 struct e1000_adapter *adapter = container_of(work,
4799 struct e1000_adapter,
4801 struct e1000_hw *hw = &adapter->hw;
4803 if (test_bit(__E1000_DOWN, &adapter->state))
4806 e1000_get_phy_info(hw);
4808 /* Enable EEE on 82579 after link up */
4809 if (hw->phy.type >= e1000_phy_82579)
4810 e1000_set_eee_pchlan(hw);
4814 * e1000_update_phy_info - timre call-back to update PHY info
4815 * @data: pointer to adapter cast into an unsigned long
4817 * Need to wait a few seconds after link up to get diagnostic information from
4820 static void e1000_update_phy_info(struct timer_list *t)
4822 struct e1000_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4824 if (test_bit(__E1000_DOWN, &adapter->state))
4827 schedule_work(&adapter->update_phy_task);
4831 * e1000e_update_phy_stats - Update the PHY statistics counters
4832 * @adapter: board private structure
4834 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4836 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4838 struct e1000_hw *hw = &adapter->hw;
4842 ret_val = hw->phy.ops.acquire(hw);
4846 /* A page set is expensive so check if already on desired page.
4847 * If not, set to the page with the PHY status registers.
4850 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4854 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4855 ret_val = hw->phy.ops.set_page(hw,
4856 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4861 /* Single Collision Count */
4862 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4863 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4865 adapter->stats.scc += phy_data;
4867 /* Excessive Collision Count */
4868 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4869 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4871 adapter->stats.ecol += phy_data;
4873 /* Multiple Collision Count */
4874 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4875 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4877 adapter->stats.mcc += phy_data;
4879 /* Late Collision Count */
4880 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4881 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4883 adapter->stats.latecol += phy_data;
4885 /* Collision Count - also used for adaptive IFS */
4886 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4887 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4889 hw->mac.collision_delta = phy_data;
4892 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4893 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4895 adapter->stats.dc += phy_data;
4897 /* Transmit with no CRS */
4898 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4899 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4901 adapter->stats.tncrs += phy_data;
4904 hw->phy.ops.release(hw);
4908 * e1000e_update_stats - Update the board statistics counters
4909 * @adapter: board private structure
4911 static void e1000e_update_stats(struct e1000_adapter *adapter)
4913 struct net_device *netdev = adapter->netdev;
4914 struct e1000_hw *hw = &adapter->hw;
4915 struct pci_dev *pdev = adapter->pdev;
4917 /* Prevent stats update while adapter is being reset, or if the pci
4918 * connection is down.
4920 if (adapter->link_speed == 0)
4922 if (pci_channel_offline(pdev))
4925 adapter->stats.crcerrs += er32(CRCERRS);
4926 adapter->stats.gprc += er32(GPRC);
4927 adapter->stats.gorc += er32(GORCL);
4928 er32(GORCH); /* Clear gorc */
4929 adapter->stats.bprc += er32(BPRC);
4930 adapter->stats.mprc += er32(MPRC);
4931 adapter->stats.roc += er32(ROC);
4933 adapter->stats.mpc += er32(MPC);
4935 /* Half-duplex statistics */
4936 if (adapter->link_duplex == HALF_DUPLEX) {
4937 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4938 e1000e_update_phy_stats(adapter);
4940 adapter->stats.scc += er32(SCC);
4941 adapter->stats.ecol += er32(ECOL);
4942 adapter->stats.mcc += er32(MCC);
4943 adapter->stats.latecol += er32(LATECOL);
4944 adapter->stats.dc += er32(DC);
4946 hw->mac.collision_delta = er32(COLC);
4948 if ((hw->mac.type != e1000_82574) &&
4949 (hw->mac.type != e1000_82583))
4950 adapter->stats.tncrs += er32(TNCRS);
4952 adapter->stats.colc += hw->mac.collision_delta;
4955 adapter->stats.xonrxc += er32(XONRXC);
4956 adapter->stats.xontxc += er32(XONTXC);
4957 adapter->stats.xoffrxc += er32(XOFFRXC);
4958 adapter->stats.xofftxc += er32(XOFFTXC);
4959 adapter->stats.gptc += er32(GPTC);
4960 adapter->stats.gotc += er32(GOTCL);
4961 er32(GOTCH); /* Clear gotc */
4962 adapter->stats.rnbc += er32(RNBC);
4963 adapter->stats.ruc += er32(RUC);
4965 adapter->stats.mptc += er32(MPTC);
4966 adapter->stats.bptc += er32(BPTC);
4968 /* used for adaptive IFS */
4970 hw->mac.tx_packet_delta = er32(TPT);
4971 adapter->stats.tpt += hw->mac.tx_packet_delta;
4973 adapter->stats.algnerrc += er32(ALGNERRC);
4974 adapter->stats.rxerrc += er32(RXERRC);
4975 adapter->stats.cexterr += er32(CEXTERR);
4976 adapter->stats.tsctc += er32(TSCTC);
4977 adapter->stats.tsctfc += er32(TSCTFC);
4979 /* Fill out the OS statistics structure */
4980 netdev->stats.multicast = adapter->stats.mprc;
4981 netdev->stats.collisions = adapter->stats.colc;
4985 /* RLEC on some newer hardware can be incorrect so build
4986 * our own version based on RUC and ROC
4988 netdev->stats.rx_errors = adapter->stats.rxerrc +
4989 adapter->stats.crcerrs + adapter->stats.algnerrc +
4990 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4991 netdev->stats.rx_length_errors = adapter->stats.ruc +
4993 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4994 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4995 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4998 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4999 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
5000 netdev->stats.tx_window_errors = adapter->stats.latecol;
5001 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
5003 /* Tx Dropped needs to be maintained elsewhere */
5005 /* Management Stats */
5006 adapter->stats.mgptc += er32(MGTPTC);
5007 adapter->stats.mgprc += er32(MGTPRC);
5008 adapter->stats.mgpdc += er32(MGTPDC);
5010 /* Correctable ECC Errors */
5011 if (hw->mac.type >= e1000_pch_lpt) {
5012 u32 pbeccsts = er32(PBECCSTS);
5014 adapter->corr_errors +=
5015 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
5016 adapter->uncorr_errors +=
5017 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
5018 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
5023 * e1000_phy_read_status - Update the PHY register status snapshot
5024 * @adapter: board private structure
5026 static void e1000_phy_read_status(struct e1000_adapter *adapter)
5028 struct e1000_hw *hw = &adapter->hw;
5029 struct e1000_phy_regs *phy = &adapter->phy_regs;
5031 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5032 (er32(STATUS) & E1000_STATUS_LU) &&
5033 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5036 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5037 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5038 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5039 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5040 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5041 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5042 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5043 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5045 e_warn("Error reading PHY register\n");
5047 /* Do not read PHY registers if link is not up
5048 * Set values to typical power-on defaults
5050 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5051 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5052 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5054 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5055 ADVERTISE_ALL | ADVERTISE_CSMA);
5057 phy->expansion = EXPANSION_ENABLENPAGE;
5058 phy->ctrl1000 = ADVERTISE_1000FULL;
5060 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5064 static void e1000_print_link_info(struct e1000_adapter *adapter)
5066 struct e1000_hw *hw = &adapter->hw;
5067 u32 ctrl = er32(CTRL);
5069 /* Link status message must follow this format for user tools */
5070 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5071 adapter->netdev->name, adapter->link_speed,
5072 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5073 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5074 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5075 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5078 static bool e1000e_has_link(struct e1000_adapter *adapter)
5080 struct e1000_hw *hw = &adapter->hw;
5081 bool link_active = false;
5084 /* get_link_status is set on LSC (link status) interrupt or
5085 * Rx sequence error interrupt. get_link_status will stay
5086 * true until the check_for_link establishes link
5087 * for copper adapters ONLY
5089 switch (hw->phy.media_type) {
5090 case e1000_media_type_copper:
5091 if (hw->mac.get_link_status) {
5092 ret_val = hw->mac.ops.check_for_link(hw);
5093 link_active = !hw->mac.get_link_status;
5098 case e1000_media_type_fiber:
5099 ret_val = hw->mac.ops.check_for_link(hw);
5100 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5102 case e1000_media_type_internal_serdes:
5103 ret_val = hw->mac.ops.check_for_link(hw);
5104 link_active = hw->mac.serdes_has_link;
5107 case e1000_media_type_unknown:
5111 if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5112 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5113 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5114 e_info("Gigabit has been disabled, downgrading speed\n");
5120 static void e1000e_enable_receives(struct e1000_adapter *adapter)
5122 /* make sure the receive unit is started */
5123 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5124 (adapter->flags & FLAG_RESTART_NOW)) {
5125 struct e1000_hw *hw = &adapter->hw;
5126 u32 rctl = er32(RCTL);
5128 ew32(RCTL, rctl | E1000_RCTL_EN);
5129 adapter->flags &= ~FLAG_RESTART_NOW;
5133 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5135 struct e1000_hw *hw = &adapter->hw;
5137 /* With 82574 controllers, PHY needs to be checked periodically
5138 * for hung state and reset, if two calls return true
5140 if (e1000_check_phy_82574(hw))
5141 adapter->phy_hang_count++;
5143 adapter->phy_hang_count = 0;
5145 if (adapter->phy_hang_count > 1) {
5146 adapter->phy_hang_count = 0;
5147 e_dbg("PHY appears hung - resetting\n");
5148 schedule_work(&adapter->reset_task);
5153 * e1000_watchdog - Timer Call-back
5154 * @data: pointer to adapter cast into an unsigned long
5156 static void e1000_watchdog(struct timer_list *t)
5158 struct e1000_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5160 /* Do the rest outside of interrupt context */
5161 schedule_work(&adapter->watchdog_task);
5163 /* TODO: make this use queue_delayed_work() */
5166 static void e1000_watchdog_task(struct work_struct *work)
5168 struct e1000_adapter *adapter = container_of(work,
5169 struct e1000_adapter,
5171 struct net_device *netdev = adapter->netdev;
5172 struct e1000_mac_info *mac = &adapter->hw.mac;
5173 struct e1000_phy_info *phy = &adapter->hw.phy;
5174 struct e1000_ring *tx_ring = adapter->tx_ring;
5175 struct e1000_hw *hw = &adapter->hw;
5178 if (test_bit(__E1000_DOWN, &adapter->state))
5181 link = e1000e_has_link(adapter);
5182 if ((netif_carrier_ok(netdev)) && link) {
5183 /* Cancel scheduled suspend requests. */
5184 pm_runtime_resume(netdev->dev.parent);
5186 e1000e_enable_receives(adapter);
5190 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5191 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5192 e1000_update_mng_vlan(adapter);
5195 if (!netif_carrier_ok(netdev)) {
5198 /* Cancel scheduled suspend requests. */
5199 pm_runtime_resume(netdev->dev.parent);
5201 /* update snapshot of PHY registers on LSC */
5202 e1000_phy_read_status(adapter);
5203 mac->ops.get_link_up_info(&adapter->hw,
5204 &adapter->link_speed,
5205 &adapter->link_duplex);
5206 e1000_print_link_info(adapter);
5208 /* check if SmartSpeed worked */
5209 e1000e_check_downshift(hw);
5210 if (phy->speed_downgraded)
5212 "Link Speed was downgraded by SmartSpeed\n");
5214 /* On supported PHYs, check for duplex mismatch only
5215 * if link has autonegotiated at 10/100 half
5217 if ((hw->phy.type == e1000_phy_igp_3 ||
5218 hw->phy.type == e1000_phy_bm) &&
5220 (adapter->link_speed == SPEED_10 ||
5221 adapter->link_speed == SPEED_100) &&
5222 (adapter->link_duplex == HALF_DUPLEX)) {
5225 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5227 if (!(autoneg_exp & EXPANSION_NWAY))
5228 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5231 /* adjust timeout factor according to speed/duplex */
5232 adapter->tx_timeout_factor = 1;
5233 switch (adapter->link_speed) {
5236 adapter->tx_timeout_factor = 16;
5240 adapter->tx_timeout_factor = 10;
5244 /* workaround: re-program speed mode bit after
5247 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5251 tarc0 = er32(TARC(0));
5252 tarc0 &= ~SPEED_MODE_BIT;
5253 ew32(TARC(0), tarc0);
5256 /* disable TSO for pcie and 10/100 speeds, to avoid
5257 * some hardware issues
5259 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5260 switch (adapter->link_speed) {
5263 e_info("10/100 speed: disabling TSO\n");
5264 netdev->features &= ~NETIF_F_TSO;
5265 netdev->features &= ~NETIF_F_TSO6;
5268 netdev->features |= NETIF_F_TSO;
5269 netdev->features |= NETIF_F_TSO6;
5277 /* enable transmits in the hardware, need to do this
5278 * after setting TARC(0)
5281 tctl |= E1000_TCTL_EN;
5284 /* Perform any post-link-up configuration before
5285 * reporting link up.
5287 if (phy->ops.cfg_on_link_up)
5288 phy->ops.cfg_on_link_up(hw);
5290 netif_carrier_on(netdev);
5292 if (!test_bit(__E1000_DOWN, &adapter->state))
5293 mod_timer(&adapter->phy_info_timer,
5294 round_jiffies(jiffies + 2 * HZ));
5297 if (netif_carrier_ok(netdev)) {
5298 adapter->link_speed = 0;
5299 adapter->link_duplex = 0;
5300 /* Link status message must follow this format */
5301 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5302 netif_carrier_off(netdev);
5303 if (!test_bit(__E1000_DOWN, &adapter->state))
5304 mod_timer(&adapter->phy_info_timer,
5305 round_jiffies(jiffies + 2 * HZ));
5307 /* 8000ES2LAN requires a Rx packet buffer work-around
5308 * on link down event; reset the controller to flush
5309 * the Rx packet buffer.
5311 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5312 adapter->flags |= FLAG_RESTART_NOW;
5314 pm_schedule_suspend(netdev->dev.parent,
5320 spin_lock(&adapter->stats64_lock);
5321 e1000e_update_stats(adapter);
5323 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5324 adapter->tpt_old = adapter->stats.tpt;
5325 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5326 adapter->colc_old = adapter->stats.colc;
5328 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5329 adapter->gorc_old = adapter->stats.gorc;
5330 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5331 adapter->gotc_old = adapter->stats.gotc;
5332 spin_unlock(&adapter->stats64_lock);
5334 /* If the link is lost the controller stops DMA, but
5335 * if there is queued Tx work it cannot be done. So
5336 * reset the controller to flush the Tx packet buffers.
5338 if (!netif_carrier_ok(netdev) &&
5339 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5340 adapter->flags |= FLAG_RESTART_NOW;
5342 /* If reset is necessary, do it outside of interrupt context. */
5343 if (adapter->flags & FLAG_RESTART_NOW) {
5344 schedule_work(&adapter->reset_task);
5345 /* return immediately since reset is imminent */
5349 e1000e_update_adaptive(&adapter->hw);
5351 /* Simple mode for Interrupt Throttle Rate (ITR) */
5352 if (adapter->itr_setting == 4) {
5353 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5354 * Total asymmetrical Tx or Rx gets ITR=8000;
5355 * everyone else is between 2000-8000.
5357 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5358 u32 dif = (adapter->gotc > adapter->gorc ?
5359 adapter->gotc - adapter->gorc :
5360 adapter->gorc - adapter->gotc) / 10000;
5361 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5363 e1000e_write_itr(adapter, itr);
5366 /* Cause software interrupt to ensure Rx ring is cleaned */
5367 if (adapter->msix_entries)
5368 ew32(ICS, adapter->rx_ring->ims_val);
5370 ew32(ICS, E1000_ICS_RXDMT0);
5372 /* flush pending descriptors to memory before detecting Tx hang */
5373 e1000e_flush_descriptors(adapter);
5375 /* Force detection of hung controller every watchdog period */
5376 adapter->detect_tx_hung = true;
5378 /* With 82571 controllers, LAA may be overwritten due to controller
5379 * reset from the other port. Set the appropriate LAA in RAR[0]
5381 if (e1000e_get_laa_state_82571(hw))
5382 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5384 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5385 e1000e_check_82574_phy_workaround(adapter);
5387 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5388 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5389 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5390 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5392 adapter->rx_hwtstamp_cleared++;
5394 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5398 /* Reset the timer */
5399 if (!test_bit(__E1000_DOWN, &adapter->state))
5400 mod_timer(&adapter->watchdog_timer,
5401 round_jiffies(jiffies + 2 * HZ));
5404 #define E1000_TX_FLAGS_CSUM 0x00000001
5405 #define E1000_TX_FLAGS_VLAN 0x00000002
5406 #define E1000_TX_FLAGS_TSO 0x00000004
5407 #define E1000_TX_FLAGS_IPV4 0x00000008
5408 #define E1000_TX_FLAGS_NO_FCS 0x00000010
5409 #define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5410 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5411 #define E1000_TX_FLAGS_VLAN_SHIFT 16
5413 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5416 struct e1000_context_desc *context_desc;
5417 struct e1000_buffer *buffer_info;
5421 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5424 if (!skb_is_gso(skb))
5427 err = skb_cow_head(skb, 0);
5431 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5432 mss = skb_shinfo(skb)->gso_size;
5433 if (protocol == htons(ETH_P_IP)) {
5434 struct iphdr *iph = ip_hdr(skb);
5437 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5439 cmd_length = E1000_TXD_CMD_IP;
5440 ipcse = skb_transport_offset(skb) - 1;
5441 } else if (skb_is_gso_v6(skb)) {
5442 ipv6_hdr(skb)->payload_len = 0;
5443 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5444 &ipv6_hdr(skb)->daddr,
5448 ipcss = skb_network_offset(skb);
5449 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5450 tucss = skb_transport_offset(skb);
5451 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5453 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5454 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5456 i = tx_ring->next_to_use;
5457 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5458 buffer_info = &tx_ring->buffer_info[i];
5460 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5461 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5462 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5463 context_desc->upper_setup.tcp_fields.tucss = tucss;
5464 context_desc->upper_setup.tcp_fields.tucso = tucso;
5465 context_desc->upper_setup.tcp_fields.tucse = 0;
5466 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5467 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5468 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5470 buffer_info->time_stamp = jiffies;
5471 buffer_info->next_to_watch = i;
5474 if (i == tx_ring->count)
5476 tx_ring->next_to_use = i;
5481 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5484 struct e1000_adapter *adapter = tx_ring->adapter;
5485 struct e1000_context_desc *context_desc;
5486 struct e1000_buffer *buffer_info;
5489 u32 cmd_len = E1000_TXD_CMD_DEXT;
5491 if (skb->ip_summed != CHECKSUM_PARTIAL)
5495 case cpu_to_be16(ETH_P_IP):
5496 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5497 cmd_len |= E1000_TXD_CMD_TCP;
5499 case cpu_to_be16(ETH_P_IPV6):
5500 /* XXX not handling all IPV6 headers */
5501 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5502 cmd_len |= E1000_TXD_CMD_TCP;
5505 if (unlikely(net_ratelimit()))
5506 e_warn("checksum_partial proto=%x!\n",
5507 be16_to_cpu(protocol));
5511 css = skb_checksum_start_offset(skb);
5513 i = tx_ring->next_to_use;
5514 buffer_info = &tx_ring->buffer_info[i];
5515 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5517 context_desc->lower_setup.ip_config = 0;
5518 context_desc->upper_setup.tcp_fields.tucss = css;
5519 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5520 context_desc->upper_setup.tcp_fields.tucse = 0;
5521 context_desc->tcp_seg_setup.data = 0;
5522 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5524 buffer_info->time_stamp = jiffies;
5525 buffer_info->next_to_watch = i;
5528 if (i == tx_ring->count)
5530 tx_ring->next_to_use = i;
5535 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5536 unsigned int first, unsigned int max_per_txd,
5537 unsigned int nr_frags)
5539 struct e1000_adapter *adapter = tx_ring->adapter;
5540 struct pci_dev *pdev = adapter->pdev;
5541 struct e1000_buffer *buffer_info;
5542 unsigned int len = skb_headlen(skb);
5543 unsigned int offset = 0, size, count = 0, i;
5544 unsigned int f, bytecount, segs;
5546 i = tx_ring->next_to_use;
5549 buffer_info = &tx_ring->buffer_info[i];
5550 size = min(len, max_per_txd);
5552 buffer_info->length = size;
5553 buffer_info->time_stamp = jiffies;
5554 buffer_info->next_to_watch = i;
5555 buffer_info->dma = dma_map_single(&pdev->dev,
5557 size, DMA_TO_DEVICE);
5558 buffer_info->mapped_as_page = false;
5559 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5568 if (i == tx_ring->count)
5573 for (f = 0; f < nr_frags; f++) {
5574 const struct skb_frag_struct *frag;
5576 frag = &skb_shinfo(skb)->frags[f];
5577 len = skb_frag_size(frag);
5582 if (i == tx_ring->count)
5585 buffer_info = &tx_ring->buffer_info[i];
5586 size = min(len, max_per_txd);
5588 buffer_info->length = size;
5589 buffer_info->time_stamp = jiffies;
5590 buffer_info->next_to_watch = i;
5591 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5594 buffer_info->mapped_as_page = true;
5595 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5604 segs = skb_shinfo(skb)->gso_segs ? : 1;
5605 /* multiply data chunks by size of headers */
5606 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5608 tx_ring->buffer_info[i].skb = skb;
5609 tx_ring->buffer_info[i].segs = segs;
5610 tx_ring->buffer_info[i].bytecount = bytecount;
5611 tx_ring->buffer_info[first].next_to_watch = i;
5616 dev_err(&pdev->dev, "Tx DMA map failed\n");
5617 buffer_info->dma = 0;
5623 i += tx_ring->count;
5625 buffer_info = &tx_ring->buffer_info[i];
5626 e1000_put_txbuf(tx_ring, buffer_info, true);
5632 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5634 struct e1000_adapter *adapter = tx_ring->adapter;
5635 struct e1000_tx_desc *tx_desc = NULL;
5636 struct e1000_buffer *buffer_info;
5637 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5640 if (tx_flags & E1000_TX_FLAGS_TSO) {
5641 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5643 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5645 if (tx_flags & E1000_TX_FLAGS_IPV4)
5646 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5649 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5650 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5651 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5654 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5655 txd_lower |= E1000_TXD_CMD_VLE;
5656 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5659 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5660 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5662 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5663 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5664 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5667 i = tx_ring->next_to_use;
5670 buffer_info = &tx_ring->buffer_info[i];
5671 tx_desc = E1000_TX_DESC(*tx_ring, i);
5672 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5673 tx_desc->lower.data = cpu_to_le32(txd_lower |
5674 buffer_info->length);
5675 tx_desc->upper.data = cpu_to_le32(txd_upper);
5678 if (i == tx_ring->count)
5680 } while (--count > 0);
5682 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5684 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5685 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5686 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5688 /* Force memory writes to complete before letting h/w
5689 * know there are new descriptors to fetch. (Only
5690 * applicable for weak-ordered memory model archs,
5695 tx_ring->next_to_use = i;
5698 #define MINIMUM_DHCP_PACKET_SIZE 282
5699 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5700 struct sk_buff *skb)
5702 struct e1000_hw *hw = &adapter->hw;
5705 if (skb_vlan_tag_present(skb) &&
5706 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5707 (adapter->hw.mng_cookie.status &
5708 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5711 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5714 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5718 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5721 if (ip->protocol != IPPROTO_UDP)
5724 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5725 if (ntohs(udp->dest) != 67)
5728 offset = (u8 *)udp + 8 - skb->data;
5729 length = skb->len - offset;
5730 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5736 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5738 struct e1000_adapter *adapter = tx_ring->adapter;
5740 netif_stop_queue(adapter->netdev);
5741 /* Herbert's original patch had:
5742 * smp_mb__after_netif_stop_queue();
5743 * but since that doesn't exist yet, just open code it.
5747 /* We need to check again in a case another CPU has just
5748 * made room available.
5750 if (e1000_desc_unused(tx_ring) < size)
5754 netif_start_queue(adapter->netdev);
5755 ++adapter->restart_queue;
5759 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5761 BUG_ON(size > tx_ring->count);
5763 if (e1000_desc_unused(tx_ring) >= size)
5765 return __e1000_maybe_stop_tx(tx_ring, size);
5768 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5769 struct net_device *netdev)
5771 struct e1000_adapter *adapter = netdev_priv(netdev);
5772 struct e1000_ring *tx_ring = adapter->tx_ring;
5774 unsigned int tx_flags = 0;
5775 unsigned int len = skb_headlen(skb);
5776 unsigned int nr_frags;
5781 __be16 protocol = vlan_get_protocol(skb);
5783 if (test_bit(__E1000_DOWN, &adapter->state)) {
5784 dev_kfree_skb_any(skb);
5785 return NETDEV_TX_OK;
5788 if (skb->len <= 0) {
5789 dev_kfree_skb_any(skb);
5790 return NETDEV_TX_OK;
5793 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5794 * pad skb in order to meet this minimum size requirement
5796 if (skb_put_padto(skb, 17))
5797 return NETDEV_TX_OK;
5799 mss = skb_shinfo(skb)->gso_size;
5803 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5804 * points to just header, pull a few bytes of payload from
5805 * frags into skb->data
5807 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5808 /* we do this workaround for ES2LAN, but it is un-necessary,
5809 * avoiding it could save a lot of cycles
5811 if (skb->data_len && (hdr_len == len)) {
5812 unsigned int pull_size;
5814 pull_size = min_t(unsigned int, 4, skb->data_len);
5815 if (!__pskb_pull_tail(skb, pull_size)) {
5816 e_err("__pskb_pull_tail failed.\n");
5817 dev_kfree_skb_any(skb);
5818 return NETDEV_TX_OK;
5820 len = skb_headlen(skb);
5824 /* reserve a descriptor for the offload context */
5825 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5829 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5831 nr_frags = skb_shinfo(skb)->nr_frags;
5832 for (f = 0; f < nr_frags; f++)
5833 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5834 adapter->tx_fifo_limit);
5836 if (adapter->hw.mac.tx_pkt_filtering)
5837 e1000_transfer_dhcp_info(adapter, skb);
5839 /* need: count + 2 desc gap to keep tail from touching
5840 * head, otherwise try next time
5842 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5843 return NETDEV_TX_BUSY;
5845 if (skb_vlan_tag_present(skb)) {
5846 tx_flags |= E1000_TX_FLAGS_VLAN;
5847 tx_flags |= (skb_vlan_tag_get(skb) <<
5848 E1000_TX_FLAGS_VLAN_SHIFT);
5851 first = tx_ring->next_to_use;
5853 tso = e1000_tso(tx_ring, skb, protocol);
5855 dev_kfree_skb_any(skb);
5856 return NETDEV_TX_OK;
5860 tx_flags |= E1000_TX_FLAGS_TSO;
5861 else if (e1000_tx_csum(tx_ring, skb, protocol))
5862 tx_flags |= E1000_TX_FLAGS_CSUM;
5864 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5865 * 82571 hardware supports TSO capabilities for IPv6 as well...
5866 * no longer assume, we must.
5868 if (protocol == htons(ETH_P_IP))
5869 tx_flags |= E1000_TX_FLAGS_IPV4;
5871 if (unlikely(skb->no_fcs))
5872 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5874 /* if count is 0 then mapping error has occurred */
5875 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5878 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5879 (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5880 if (!adapter->tx_hwtstamp_skb) {
5881 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5882 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5883 adapter->tx_hwtstamp_skb = skb_get(skb);
5884 adapter->tx_hwtstamp_start = jiffies;
5885 schedule_work(&adapter->tx_hwtstamp_work);
5887 adapter->tx_hwtstamp_skipped++;
5891 skb_tx_timestamp(skb);
5893 netdev_sent_queue(netdev, skb->len);
5894 e1000_tx_queue(tx_ring, tx_flags, count);
5895 /* Make sure there is space in the ring for the next send. */
5896 e1000_maybe_stop_tx(tx_ring,
5898 DIV_ROUND_UP(PAGE_SIZE,
5899 adapter->tx_fifo_limit) + 2));
5901 if (!skb->xmit_more ||
5902 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5903 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5904 e1000e_update_tdt_wa(tx_ring,
5905 tx_ring->next_to_use);
5907 writel(tx_ring->next_to_use, tx_ring->tail);
5909 /* we need this if more than one processor can write
5910 * to our tail at a time, it synchronizes IO on
5916 dev_kfree_skb_any(skb);
5917 tx_ring->buffer_info[first].time_stamp = 0;
5918 tx_ring->next_to_use = first;
5921 return NETDEV_TX_OK;
5925 * e1000_tx_timeout - Respond to a Tx Hang
5926 * @netdev: network interface device structure
5928 static void e1000_tx_timeout(struct net_device *netdev)
5930 struct e1000_adapter *adapter = netdev_priv(netdev);
5932 /* Do the reset outside of interrupt context */
5933 adapter->tx_timeout_count++;
5934 schedule_work(&adapter->reset_task);
5937 static void e1000_reset_task(struct work_struct *work)
5939 struct e1000_adapter *adapter;
5940 adapter = container_of(work, struct e1000_adapter, reset_task);
5942 /* don't run the task if already down */
5943 if (test_bit(__E1000_DOWN, &adapter->state))
5946 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5947 e1000e_dump(adapter);
5948 e_err("Reset adapter unexpectedly\n");
5950 e1000e_reinit_locked(adapter);
5954 * e1000_get_stats64 - Get System Network Statistics
5955 * @netdev: network interface device structure
5956 * @stats: rtnl_link_stats64 pointer
5958 * Returns the address of the device statistics structure.
5960 void e1000e_get_stats64(struct net_device *netdev,
5961 struct rtnl_link_stats64 *stats)
5963 struct e1000_adapter *adapter = netdev_priv(netdev);
5965 spin_lock(&adapter->stats64_lock);
5966 e1000e_update_stats(adapter);
5967 /* Fill out the OS statistics structure */
5968 stats->rx_bytes = adapter->stats.gorc;
5969 stats->rx_packets = adapter->stats.gprc;
5970 stats->tx_bytes = adapter->stats.gotc;
5971 stats->tx_packets = adapter->stats.gptc;
5972 stats->multicast = adapter->stats.mprc;
5973 stats->collisions = adapter->stats.colc;
5977 /* RLEC on some newer hardware can be incorrect so build
5978 * our own version based on RUC and ROC
5980 stats->rx_errors = adapter->stats.rxerrc +
5981 adapter->stats.crcerrs + adapter->stats.algnerrc +
5982 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5983 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5984 stats->rx_crc_errors = adapter->stats.crcerrs;
5985 stats->rx_frame_errors = adapter->stats.algnerrc;
5986 stats->rx_missed_errors = adapter->stats.mpc;
5989 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5990 stats->tx_aborted_errors = adapter->stats.ecol;
5991 stats->tx_window_errors = adapter->stats.latecol;
5992 stats->tx_carrier_errors = adapter->stats.tncrs;
5994 /* Tx Dropped needs to be maintained elsewhere */
5996 spin_unlock(&adapter->stats64_lock);
6000 * e1000_change_mtu - Change the Maximum Transfer Unit
6001 * @netdev: network interface device structure
6002 * @new_mtu: new value for maximum frame size
6004 * Returns 0 on success, negative on failure
6006 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
6008 struct e1000_adapter *adapter = netdev_priv(netdev);
6009 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
6011 /* Jumbo frame support */
6012 if ((new_mtu > ETH_DATA_LEN) &&
6013 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
6014 e_err("Jumbo Frames not supported.\n");
6018 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6019 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
6020 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6021 (new_mtu > ETH_DATA_LEN)) {
6022 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6026 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6027 usleep_range(1000, 2000);
6028 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6029 adapter->max_frame_size = max_frame;
6030 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6031 netdev->mtu = new_mtu;
6033 pm_runtime_get_sync(netdev->dev.parent);
6035 if (netif_running(netdev))
6036 e1000e_down(adapter, true);
6038 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6039 * means we reserve 2 more, this pushes us to allocate from the next
6041 * i.e. RXBUFFER_2048 --> size-4096 slab
6042 * However with the new *_jumbo_rx* routines, jumbo receives will use
6046 if (max_frame <= 2048)
6047 adapter->rx_buffer_len = 2048;
6049 adapter->rx_buffer_len = 4096;
6051 /* adjust allocation if LPE protects us, and we aren't using SBP */
6052 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6053 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6055 if (netif_running(netdev))
6058 e1000e_reset(adapter);
6060 pm_runtime_put_sync(netdev->dev.parent);
6062 clear_bit(__E1000_RESETTING, &adapter->state);
6067 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6070 struct e1000_adapter *adapter = netdev_priv(netdev);
6071 struct mii_ioctl_data *data = if_mii(ifr);
6073 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6078 data->phy_id = adapter->hw.phy.addr;
6081 e1000_phy_read_status(adapter);
6083 switch (data->reg_num & 0x1F) {
6085 data->val_out = adapter->phy_regs.bmcr;
6088 data->val_out = adapter->phy_regs.bmsr;
6091 data->val_out = (adapter->hw.phy.id >> 16);
6094 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6097 data->val_out = adapter->phy_regs.advertise;
6100 data->val_out = adapter->phy_regs.lpa;
6103 data->val_out = adapter->phy_regs.expansion;
6106 data->val_out = adapter->phy_regs.ctrl1000;
6109 data->val_out = adapter->phy_regs.stat1000;
6112 data->val_out = adapter->phy_regs.estatus;
6126 * e1000e_hwtstamp_ioctl - control hardware time stamping
6127 * @netdev: network interface device structure
6128 * @ifreq: interface request
6130 * Outgoing time stamping can be enabled and disabled. Play nice and
6131 * disable it when requested, although it shouldn't cause any overhead
6132 * when no packet needs it. At most one packet in the queue may be
6133 * marked for time stamping, otherwise it would be impossible to tell
6134 * for sure to which packet the hardware time stamp belongs.
6136 * Incoming time stamping has to be configured via the hardware filters.
6137 * Not all combinations are supported, in particular event type has to be
6138 * specified. Matching the kind of event packet is not supported, with the
6139 * exception of "all V2 events regardless of level 2 or 4".
6141 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6143 struct e1000_adapter *adapter = netdev_priv(netdev);
6144 struct hwtstamp_config config;
6147 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6150 ret_val = e1000e_config_hwtstamp(adapter, &config);
6154 switch (config.rx_filter) {
6155 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6156 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6157 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6158 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6159 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6160 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6161 /* With V2 type filters which specify a Sync or Delay Request,
6162 * Path Delay Request/Response messages are also time stamped
6163 * by hardware so notify the caller the requested packets plus
6164 * some others are time stamped.
6166 config.rx_filter = HWTSTAMP_FILTER_SOME;
6172 return copy_to_user(ifr->ifr_data, &config,
6173 sizeof(config)) ? -EFAULT : 0;
6176 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6178 struct e1000_adapter *adapter = netdev_priv(netdev);
6180 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6181 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6184 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6190 return e1000_mii_ioctl(netdev, ifr, cmd);
6192 return e1000e_hwtstamp_set(netdev, ifr);
6194 return e1000e_hwtstamp_get(netdev, ifr);
6200 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6202 struct e1000_hw *hw = &adapter->hw;
6203 u32 i, mac_reg, wuc;
6204 u16 phy_reg, wuc_enable;
6207 /* copy MAC RARs to PHY RARs */
6208 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6210 retval = hw->phy.ops.acquire(hw);
6212 e_err("Could not acquire PHY\n");
6216 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6217 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6221 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6222 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6223 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6224 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6225 (u16)(mac_reg & 0xFFFF));
6226 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6227 (u16)((mac_reg >> 16) & 0xFFFF));
6230 /* configure PHY Rx Control register */
6231 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6232 mac_reg = er32(RCTL);
6233 if (mac_reg & E1000_RCTL_UPE)
6234 phy_reg |= BM_RCTL_UPE;
6235 if (mac_reg & E1000_RCTL_MPE)
6236 phy_reg |= BM_RCTL_MPE;
6237 phy_reg &= ~(BM_RCTL_MO_MASK);
6238 if (mac_reg & E1000_RCTL_MO_3)
6239 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6240 << BM_RCTL_MO_SHIFT);
6241 if (mac_reg & E1000_RCTL_BAM)
6242 phy_reg |= BM_RCTL_BAM;
6243 if (mac_reg & E1000_RCTL_PMCF)
6244 phy_reg |= BM_RCTL_PMCF;
6245 mac_reg = er32(CTRL);
6246 if (mac_reg & E1000_CTRL_RFCE)
6247 phy_reg |= BM_RCTL_RFCE;
6248 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6250 wuc = E1000_WUC_PME_EN;
6251 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6252 wuc |= E1000_WUC_APME;
6254 /* enable PHY wakeup in MAC register */
6256 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6257 E1000_WUC_PME_STATUS | wuc));
6259 /* configure and enable PHY wakeup in PHY registers */
6260 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6261 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6263 /* activate PHY wakeup */
6264 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6265 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6267 e_err("Could not set PHY Host Wakeup bit\n");
6269 hw->phy.ops.release(hw);
6274 static void e1000e_flush_lpic(struct pci_dev *pdev)
6276 struct net_device *netdev = pci_get_drvdata(pdev);
6277 struct e1000_adapter *adapter = netdev_priv(netdev);
6278 struct e1000_hw *hw = &adapter->hw;
6281 pm_runtime_get_sync(netdev->dev.parent);
6283 ret_val = hw->phy.ops.acquire(hw);
6287 pr_info("EEE TX LPI TIMER: %08X\n",
6288 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6290 hw->phy.ops.release(hw);
6293 pm_runtime_put_sync(netdev->dev.parent);
6296 static int e1000e_pm_freeze(struct device *dev)
6298 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6299 struct e1000_adapter *adapter = netdev_priv(netdev);
6301 netif_device_detach(netdev);
6303 if (netif_running(netdev)) {
6304 int count = E1000_CHECK_RESET_COUNT;
6306 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6307 usleep_range(10000, 20000);
6309 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6311 /* Quiesce the device without resetting the hardware */
6312 e1000e_down(adapter, false);
6313 e1000_free_irq(adapter);
6315 e1000e_reset_interrupt_capability(adapter);
6317 /* Allow time for pending master requests to run */
6318 e1000e_disable_pcie_master(&adapter->hw);
6323 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6325 struct net_device *netdev = pci_get_drvdata(pdev);
6326 struct e1000_adapter *adapter = netdev_priv(netdev);
6327 struct e1000_hw *hw = &adapter->hw;
6328 u32 ctrl, ctrl_ext, rctl, status;
6329 /* Runtime suspend should only enable wakeup for link changes */
6330 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6333 status = er32(STATUS);
6334 if (status & E1000_STATUS_LU)
6335 wufc &= ~E1000_WUFC_LNKC;
6338 e1000_setup_rctl(adapter);
6339 e1000e_set_rx_mode(netdev);
6341 /* turn on all-multi mode if wake on multicast is enabled */
6342 if (wufc & E1000_WUFC_MC) {
6344 rctl |= E1000_RCTL_MPE;
6349 ctrl |= E1000_CTRL_ADVD3WUC;
6350 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6351 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6354 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6355 adapter->hw.phy.media_type ==
6356 e1000_media_type_internal_serdes) {
6357 /* keep the laser running in D3 */
6358 ctrl_ext = er32(CTRL_EXT);
6359 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6360 ew32(CTRL_EXT, ctrl_ext);
6364 e1000e_power_up_phy(adapter);
6366 if (adapter->flags & FLAG_IS_ICH)
6367 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6369 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6370 /* enable wakeup by the PHY */
6371 retval = e1000_init_phy_wakeup(adapter, wufc);
6375 /* enable wakeup by the MAC */
6377 ew32(WUC, E1000_WUC_PME_EN);
6383 e1000_power_down_phy(adapter);
6386 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6387 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6388 } else if (hw->mac.type >= e1000_pch_lpt) {
6389 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6390 /* ULP does not support wake from unicast, multicast
6393 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6399 /* Ensure that the appropriate bits are set in LPI_CTRL
6402 if ((hw->phy.type >= e1000_phy_i217) &&
6403 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6406 retval = hw->phy.ops.acquire(hw);
6408 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6411 if (adapter->eee_advert &
6412 hw->dev_spec.ich8lan.eee_lp_ability &
6413 I82579_EEE_100_SUPPORTED)
6414 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6415 if (adapter->eee_advert &
6416 hw->dev_spec.ich8lan.eee_lp_ability &
6417 I82579_EEE_1000_SUPPORTED)
6418 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6420 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6424 hw->phy.ops.release(hw);
6427 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6428 * would have already happened in close and is redundant.
6430 e1000e_release_hw_control(adapter);
6432 pci_clear_master(pdev);
6434 /* The pci-e switch on some quad port adapters will report a
6435 * correctable error when the MAC transitions from D0 to D3. To
6436 * prevent this we need to mask off the correctable errors on the
6437 * downstream port of the pci-e switch.
6439 * We don't have the associated upstream bridge while assigning
6440 * the PCI device into guest. For example, the KVM on power is
6443 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6444 struct pci_dev *us_dev = pdev->bus->self;
6450 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6451 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6452 (devctl & ~PCI_EXP_DEVCTL_CERE));
6454 pci_save_state(pdev);
6455 pci_prepare_to_sleep(pdev);
6457 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6464 * __e1000e_disable_aspm - Disable ASPM states
6465 * @pdev: pointer to PCI device struct
6466 * @state: bit-mask of ASPM states to disable
6467 * @locked: indication if this context holds pci_bus_sem locked.
6469 * Some devices *must* have certain ASPM states disabled per hardware errata.
6471 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6473 struct pci_dev *parent = pdev->bus->self;
6474 u16 aspm_dis_mask = 0;
6475 u16 pdev_aspmc, parent_aspmc;
6478 case PCIE_LINK_STATE_L0S:
6479 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6480 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6481 /* fall-through - can't have L1 without L0s */
6482 case PCIE_LINK_STATE_L1:
6483 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6489 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6490 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6493 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6495 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6498 /* Nothing to do if the ASPM states to be disabled already are */
6499 if (!(pdev_aspmc & aspm_dis_mask) &&
6500 (!parent || !(parent_aspmc & aspm_dis_mask)))
6503 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6504 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6506 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6509 #ifdef CONFIG_PCIEASPM
6511 pci_disable_link_state_locked(pdev, state);
6513 pci_disable_link_state(pdev, state);
6515 /* Double-check ASPM control. If not disabled by the above, the
6516 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6517 * not enabled); override by writing PCI config space directly.
6519 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6520 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6522 if (!(aspm_dis_mask & pdev_aspmc))
6526 /* Both device and parent should have the same ASPM setting.
6527 * Disable ASPM in downstream component first and then upstream.
6529 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6532 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6537 * e1000e_disable_aspm - Disable ASPM states.
6538 * @pdev: pointer to PCI device struct
6539 * @state: bit-mask of ASPM states to disable
6541 * This function acquires the pci_bus_sem!
6542 * Some devices *must* have certain ASPM states disabled per hardware errata.
6544 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6546 __e1000e_disable_aspm(pdev, state, 0);
6550 * e1000e_disable_aspm_locked Disable ASPM states.
6551 * @pdev: pointer to PCI device struct
6552 * @state: bit-mask of ASPM states to disable
6554 * This function must be called with pci_bus_sem acquired!
6555 * Some devices *must* have certain ASPM states disabled per hardware errata.
6557 static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6559 __e1000e_disable_aspm(pdev, state, 1);
6563 static int __e1000_resume(struct pci_dev *pdev)
6565 struct net_device *netdev = pci_get_drvdata(pdev);
6566 struct e1000_adapter *adapter = netdev_priv(netdev);
6567 struct e1000_hw *hw = &adapter->hw;
6568 u16 aspm_disable_flag = 0;
6570 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6571 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6572 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6573 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6574 if (aspm_disable_flag)
6575 e1000e_disable_aspm(pdev, aspm_disable_flag);
6577 pci_set_master(pdev);
6579 if (hw->mac.type >= e1000_pch2lan)
6580 e1000_resume_workarounds_pchlan(&adapter->hw);
6582 e1000e_power_up_phy(adapter);
6584 /* report the system wakeup cause from S3/S4 */
6585 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6588 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6590 e_info("PHY Wakeup cause - %s\n",
6591 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6592 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6593 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6594 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6595 phy_data & E1000_WUS_LNKC ?
6596 "Link Status Change" : "other");
6598 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6600 u32 wus = er32(WUS);
6603 e_info("MAC Wakeup cause - %s\n",
6604 wus & E1000_WUS_EX ? "Unicast Packet" :
6605 wus & E1000_WUS_MC ? "Multicast Packet" :
6606 wus & E1000_WUS_BC ? "Broadcast Packet" :
6607 wus & E1000_WUS_MAG ? "Magic Packet" :
6608 wus & E1000_WUS_LNKC ? "Link Status Change" :
6614 e1000e_reset(adapter);
6616 e1000_init_manageability_pt(adapter);
6618 /* If the controller has AMT, do not set DRV_LOAD until the interface
6619 * is up. For all other cases, let the f/w know that the h/w is now
6620 * under the control of the driver.
6622 if (!(adapter->flags & FLAG_HAS_AMT))
6623 e1000e_get_hw_control(adapter);
6628 #ifdef CONFIG_PM_SLEEP
6629 static int e1000e_pm_thaw(struct device *dev)
6631 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6632 struct e1000_adapter *adapter = netdev_priv(netdev);
6634 e1000e_set_interrupt_capability(adapter);
6635 if (netif_running(netdev)) {
6636 u32 err = e1000_request_irq(adapter);
6644 netif_device_attach(netdev);
6649 static int e1000e_pm_suspend(struct device *dev)
6651 struct pci_dev *pdev = to_pci_dev(dev);
6654 e1000e_flush_lpic(pdev);
6656 e1000e_pm_freeze(dev);
6658 rc = __e1000_shutdown(pdev, false);
6660 e1000e_pm_thaw(dev);
6665 static int e1000e_pm_resume(struct device *dev)
6667 struct pci_dev *pdev = to_pci_dev(dev);
6670 rc = __e1000_resume(pdev);
6674 return e1000e_pm_thaw(dev);
6676 #endif /* CONFIG_PM_SLEEP */
6678 static int e1000e_pm_runtime_idle(struct device *dev)
6680 struct pci_dev *pdev = to_pci_dev(dev);
6681 struct net_device *netdev = pci_get_drvdata(pdev);
6682 struct e1000_adapter *adapter = netdev_priv(netdev);
6685 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6687 if (!e1000e_has_link(adapter)) {
6688 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6689 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6695 static int e1000e_pm_runtime_resume(struct device *dev)
6697 struct pci_dev *pdev = to_pci_dev(dev);
6698 struct net_device *netdev = pci_get_drvdata(pdev);
6699 struct e1000_adapter *adapter = netdev_priv(netdev);
6702 rc = __e1000_resume(pdev);
6706 if (netdev->flags & IFF_UP)
6712 static int e1000e_pm_runtime_suspend(struct device *dev)
6714 struct pci_dev *pdev = to_pci_dev(dev);
6715 struct net_device *netdev = pci_get_drvdata(pdev);
6716 struct e1000_adapter *adapter = netdev_priv(netdev);
6718 if (netdev->flags & IFF_UP) {
6719 int count = E1000_CHECK_RESET_COUNT;
6721 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6722 usleep_range(10000, 20000);
6724 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6726 /* Down the device without resetting the hardware */
6727 e1000e_down(adapter, false);
6730 if (__e1000_shutdown(pdev, true)) {
6731 e1000e_pm_runtime_resume(dev);
6737 #endif /* CONFIG_PM */
6739 static void e1000_shutdown(struct pci_dev *pdev)
6741 e1000e_flush_lpic(pdev);
6743 e1000e_pm_freeze(&pdev->dev);
6745 __e1000_shutdown(pdev, false);
6748 #ifdef CONFIG_NET_POLL_CONTROLLER
6750 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6752 struct net_device *netdev = data;
6753 struct e1000_adapter *adapter = netdev_priv(netdev);
6755 if (adapter->msix_entries) {
6756 int vector, msix_irq;
6759 msix_irq = adapter->msix_entries[vector].vector;
6760 if (disable_hardirq(msix_irq))
6761 e1000_intr_msix_rx(msix_irq, netdev);
6762 enable_irq(msix_irq);
6765 msix_irq = adapter->msix_entries[vector].vector;
6766 if (disable_hardirq(msix_irq))
6767 e1000_intr_msix_tx(msix_irq, netdev);
6768 enable_irq(msix_irq);
6771 msix_irq = adapter->msix_entries[vector].vector;
6772 if (disable_hardirq(msix_irq))
6773 e1000_msix_other(msix_irq, netdev);
6774 enable_irq(msix_irq);
6782 * @netdev: network interface device structure
6784 * Polling 'interrupt' - used by things like netconsole to send skbs
6785 * without having to re-enable interrupts. It's not called while
6786 * the interrupt routine is executing.
6788 static void e1000_netpoll(struct net_device *netdev)
6790 struct e1000_adapter *adapter = netdev_priv(netdev);
6792 switch (adapter->int_mode) {
6793 case E1000E_INT_MODE_MSIX:
6794 e1000_intr_msix(adapter->pdev->irq, netdev);
6796 case E1000E_INT_MODE_MSI:
6797 if (disable_hardirq(adapter->pdev->irq))
6798 e1000_intr_msi(adapter->pdev->irq, netdev);
6799 enable_irq(adapter->pdev->irq);
6801 default: /* E1000E_INT_MODE_LEGACY */
6802 if (disable_hardirq(adapter->pdev->irq))
6803 e1000_intr(adapter->pdev->irq, netdev);
6804 enable_irq(adapter->pdev->irq);
6811 * e1000_io_error_detected - called when PCI error is detected
6812 * @pdev: Pointer to PCI device
6813 * @state: The current pci connection state
6815 * This function is called after a PCI bus error affecting
6816 * this device has been detected.
6818 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6819 pci_channel_state_t state)
6821 struct net_device *netdev = pci_get_drvdata(pdev);
6822 struct e1000_adapter *adapter = netdev_priv(netdev);
6824 netif_device_detach(netdev);
6826 if (state == pci_channel_io_perm_failure)
6827 return PCI_ERS_RESULT_DISCONNECT;
6829 if (netif_running(netdev))
6830 e1000e_down(adapter, true);
6831 pci_disable_device(pdev);
6833 /* Request a slot slot reset. */
6834 return PCI_ERS_RESULT_NEED_RESET;
6838 * e1000_io_slot_reset - called after the pci bus has been reset.
6839 * @pdev: Pointer to PCI device
6841 * Restart the card from scratch, as if from a cold-boot. Implementation
6842 * resembles the first-half of the e1000e_pm_resume routine.
6844 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6846 struct net_device *netdev = pci_get_drvdata(pdev);
6847 struct e1000_adapter *adapter = netdev_priv(netdev);
6848 struct e1000_hw *hw = &adapter->hw;
6849 u16 aspm_disable_flag = 0;
6851 pci_ers_result_t result;
6853 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6854 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6855 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6856 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6857 if (aspm_disable_flag)
6858 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6860 err = pci_enable_device_mem(pdev);
6863 "Cannot re-enable PCI device after reset.\n");
6864 result = PCI_ERS_RESULT_DISCONNECT;
6866 pdev->state_saved = true;
6867 pci_restore_state(pdev);
6868 pci_set_master(pdev);
6870 pci_enable_wake(pdev, PCI_D3hot, 0);
6871 pci_enable_wake(pdev, PCI_D3cold, 0);
6873 e1000e_reset(adapter);
6875 result = PCI_ERS_RESULT_RECOVERED;
6878 pci_cleanup_aer_uncorrect_error_status(pdev);
6884 * e1000_io_resume - called when traffic can start flowing again.
6885 * @pdev: Pointer to PCI device
6887 * This callback is called when the error recovery driver tells us that
6888 * its OK to resume normal operation. Implementation resembles the
6889 * second-half of the e1000e_pm_resume routine.
6891 static void e1000_io_resume(struct pci_dev *pdev)
6893 struct net_device *netdev = pci_get_drvdata(pdev);
6894 struct e1000_adapter *adapter = netdev_priv(netdev);
6896 e1000_init_manageability_pt(adapter);
6898 if (netif_running(netdev))
6901 netif_device_attach(netdev);
6903 /* If the controller has AMT, do not set DRV_LOAD until the interface
6904 * is up. For all other cases, let the f/w know that the h/w is now
6905 * under the control of the driver.
6907 if (!(adapter->flags & FLAG_HAS_AMT))
6908 e1000e_get_hw_control(adapter);
6911 static void e1000_print_device_info(struct e1000_adapter *adapter)
6913 struct e1000_hw *hw = &adapter->hw;
6914 struct net_device *netdev = adapter->netdev;
6916 u8 pba_str[E1000_PBANUM_LENGTH];
6918 /* print bus type/speed/width info */
6919 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6921 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6925 e_info("Intel(R) PRO/%s Network Connection\n",
6926 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6927 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6928 E1000_PBANUM_LENGTH);
6930 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6931 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6932 hw->mac.type, hw->phy.type, pba_str);
6935 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6937 struct e1000_hw *hw = &adapter->hw;
6941 if (hw->mac.type != e1000_82573)
6944 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6946 if (!ret_val && (!(buf & BIT(0)))) {
6947 /* Deep Smart Power Down (DSPD) */
6948 dev_warn(&adapter->pdev->dev,
6949 "Warning: detected DSPD enabled in EEPROM\n");
6953 static netdev_features_t e1000_fix_features(struct net_device *netdev,
6954 netdev_features_t features)
6956 struct e1000_adapter *adapter = netdev_priv(netdev);
6957 struct e1000_hw *hw = &adapter->hw;
6959 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6960 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6961 features &= ~NETIF_F_RXFCS;
6963 /* Since there is no support for separate Rx/Tx vlan accel
6964 * enable/disable make sure Tx flag is always in same state as Rx.
6966 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6967 features |= NETIF_F_HW_VLAN_CTAG_TX;
6969 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6974 static int e1000_set_features(struct net_device *netdev,
6975 netdev_features_t features)
6977 struct e1000_adapter *adapter = netdev_priv(netdev);
6978 netdev_features_t changed = features ^ netdev->features;
6980 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6981 adapter->flags |= FLAG_TSO_FORCE;
6983 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6984 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6988 if (changed & NETIF_F_RXFCS) {
6989 if (features & NETIF_F_RXFCS) {
6990 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6992 /* We need to take it back to defaults, which might mean
6993 * stripping is still disabled at the adapter level.
6995 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6996 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6998 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7002 netdev->features = features;
7004 if (netif_running(netdev))
7005 e1000e_reinit_locked(adapter);
7007 e1000e_reset(adapter);
7012 static const struct net_device_ops e1000e_netdev_ops = {
7013 .ndo_open = e1000e_open,
7014 .ndo_stop = e1000e_close,
7015 .ndo_start_xmit = e1000_xmit_frame,
7016 .ndo_get_stats64 = e1000e_get_stats64,
7017 .ndo_set_rx_mode = e1000e_set_rx_mode,
7018 .ndo_set_mac_address = e1000_set_mac,
7019 .ndo_change_mtu = e1000_change_mtu,
7020 .ndo_do_ioctl = e1000_ioctl,
7021 .ndo_tx_timeout = e1000_tx_timeout,
7022 .ndo_validate_addr = eth_validate_addr,
7024 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
7025 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
7026 #ifdef CONFIG_NET_POLL_CONTROLLER
7027 .ndo_poll_controller = e1000_netpoll,
7029 .ndo_set_features = e1000_set_features,
7030 .ndo_fix_features = e1000_fix_features,
7031 .ndo_features_check = passthru_features_check,
7035 * e1000_probe - Device Initialization Routine
7036 * @pdev: PCI device information struct
7037 * @ent: entry in e1000_pci_tbl
7039 * Returns 0 on success, negative on failure
7041 * e1000_probe initializes an adapter identified by a pci_dev structure.
7042 * The OS initialization, configuring of the adapter private structure,
7043 * and a hardware reset occur.
7045 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7047 struct net_device *netdev;
7048 struct e1000_adapter *adapter;
7049 struct e1000_hw *hw;
7050 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7051 resource_size_t mmio_start, mmio_len;
7052 resource_size_t flash_start, flash_len;
7053 static int cards_found;
7054 u16 aspm_disable_flag = 0;
7055 int bars, i, err, pci_using_dac;
7056 u16 eeprom_data = 0;
7057 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7060 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7061 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7062 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7063 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7064 if (aspm_disable_flag)
7065 e1000e_disable_aspm(pdev, aspm_disable_flag);
7067 err = pci_enable_device_mem(pdev);
7072 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7076 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7079 "No usable DMA configuration, aborting\n");
7084 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7085 err = pci_request_selected_regions_exclusive(pdev, bars,
7086 e1000e_driver_name);
7090 /* AER (Advanced Error Reporting) hooks */
7091 pci_enable_pcie_error_reporting(pdev);
7093 pci_set_master(pdev);
7094 /* PCI config space info */
7095 err = pci_save_state(pdev);
7097 goto err_alloc_etherdev;
7100 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7102 goto err_alloc_etherdev;
7104 SET_NETDEV_DEV(netdev, &pdev->dev);
7106 netdev->irq = pdev->irq;
7108 pci_set_drvdata(pdev, netdev);
7109 adapter = netdev_priv(netdev);
7111 adapter->netdev = netdev;
7112 adapter->pdev = pdev;
7114 adapter->pba = ei->pba;
7115 adapter->flags = ei->flags;
7116 adapter->flags2 = ei->flags2;
7117 adapter->hw.adapter = adapter;
7118 adapter->hw.mac.type = ei->mac;
7119 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7120 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7122 mmio_start = pci_resource_start(pdev, 0);
7123 mmio_len = pci_resource_len(pdev, 0);
7126 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7127 if (!adapter->hw.hw_addr)
7130 if ((adapter->flags & FLAG_HAS_FLASH) &&
7131 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7132 (hw->mac.type < e1000_pch_spt)) {
7133 flash_start = pci_resource_start(pdev, 1);
7134 flash_len = pci_resource_len(pdev, 1);
7135 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7136 if (!adapter->hw.flash_address)
7140 /* Set default EEE advertisement */
7141 if (adapter->flags2 & FLAG2_HAS_EEE)
7142 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7144 /* construct the net_device struct */
7145 netdev->netdev_ops = &e1000e_netdev_ops;
7146 e1000e_set_ethtool_ops(netdev);
7147 netdev->watchdog_timeo = 5 * HZ;
7148 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7149 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7151 netdev->mem_start = mmio_start;
7152 netdev->mem_end = mmio_start + mmio_len;
7154 adapter->bd_number = cards_found++;
7156 e1000e_check_options(adapter);
7158 /* setup adapter struct */
7159 err = e1000_sw_init(adapter);
7163 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7164 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7165 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7167 err = ei->get_variants(adapter);
7171 if ((adapter->flags & FLAG_IS_ICH) &&
7172 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7173 (hw->mac.type < e1000_pch_spt))
7174 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7176 hw->mac.ops.get_bus_info(&adapter->hw);
7178 adapter->hw.phy.autoneg_wait_to_complete = 0;
7180 /* Copper options */
7181 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7182 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7183 adapter->hw.phy.disable_polarity_correction = 0;
7184 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7187 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7188 dev_info(&pdev->dev,
7189 "PHY reset is blocked due to SOL/IDER session.\n");
7191 /* Set initial default active device features */
7192 netdev->features = (NETIF_F_SG |
7193 NETIF_F_HW_VLAN_CTAG_RX |
7194 NETIF_F_HW_VLAN_CTAG_TX |
7201 /* Set user-changeable features (subset of all device features) */
7202 netdev->hw_features = netdev->features;
7203 netdev->hw_features |= NETIF_F_RXFCS;
7204 netdev->priv_flags |= IFF_SUPP_NOFCS;
7205 netdev->hw_features |= NETIF_F_RXALL;
7207 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7208 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7210 netdev->vlan_features |= (NETIF_F_SG |
7215 netdev->priv_flags |= IFF_UNICAST_FLT;
7217 if (pci_using_dac) {
7218 netdev->features |= NETIF_F_HIGHDMA;
7219 netdev->vlan_features |= NETIF_F_HIGHDMA;
7222 /* MTU range: 68 - max_hw_frame_size */
7223 netdev->min_mtu = ETH_MIN_MTU;
7224 netdev->max_mtu = adapter->max_hw_frame_size -
7225 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7227 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7228 adapter->flags |= FLAG_MNG_PT_ENABLED;
7230 /* before reading the NVM, reset the controller to
7231 * put the device in a known good starting state
7233 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7235 /* systems with ASPM and others may see the checksum fail on the first
7236 * attempt. Let's give it a few tries
7239 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7242 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7248 e1000_eeprom_checks(adapter);
7250 /* copy the MAC address */
7251 if (e1000e_read_mac_addr(&adapter->hw))
7253 "NVM Read Error while reading MAC address\n");
7255 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7257 if (!is_valid_ether_addr(netdev->dev_addr)) {
7258 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7264 timer_setup(&adapter->watchdog_timer, e1000_watchdog, 0);
7265 timer_setup(&adapter->phy_info_timer, e1000_update_phy_info, 0);
7267 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7268 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7269 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7270 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7271 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7273 /* Initialize link parameters. User can change them with ethtool */
7274 adapter->hw.mac.autoneg = 1;
7275 adapter->fc_autoneg = true;
7276 adapter->hw.fc.requested_mode = e1000_fc_default;
7277 adapter->hw.fc.current_mode = e1000_fc_default;
7278 adapter->hw.phy.autoneg_advertised = 0x2f;
7280 /* Initial Wake on LAN setting - If APM wake is enabled in
7281 * the EEPROM, enable the ACPI Magic Packet filter
7283 if (adapter->flags & FLAG_APME_IN_WUC) {
7284 /* APME bit in EEPROM is mapped to WUC.APME */
7285 eeprom_data = er32(WUC);
7286 eeprom_apme_mask = E1000_WUC_APME;
7287 if ((hw->mac.type > e1000_ich10lan) &&
7288 (eeprom_data & E1000_WUC_PHY_WAKE))
7289 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7290 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7291 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7292 (adapter->hw.bus.func == 1))
7293 ret_val = e1000_read_nvm(&adapter->hw,
7294 NVM_INIT_CONTROL3_PORT_B,
7297 ret_val = e1000_read_nvm(&adapter->hw,
7298 NVM_INIT_CONTROL3_PORT_A,
7302 /* fetch WoL from EEPROM */
7304 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7305 else if (eeprom_data & eeprom_apme_mask)
7306 adapter->eeprom_wol |= E1000_WUFC_MAG;
7308 /* now that we have the eeprom settings, apply the special cases
7309 * where the eeprom may be wrong or the board simply won't support
7310 * wake on lan on a particular port
7312 if (!(adapter->flags & FLAG_HAS_WOL))
7313 adapter->eeprom_wol = 0;
7315 /* initialize the wol settings based on the eeprom settings */
7316 adapter->wol = adapter->eeprom_wol;
7318 /* make sure adapter isn't asleep if manageability is enabled */
7319 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7320 (hw->mac.ops.check_mng_mode(hw)))
7321 device_wakeup_enable(&pdev->dev);
7323 /* save off EEPROM version number */
7324 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7327 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7328 adapter->eeprom_vers = 0;
7331 /* init PTP hardware clock */
7332 e1000e_ptp_init(adapter);
7334 /* reset the hardware with the new settings */
7335 e1000e_reset(adapter);
7337 /* If the controller has AMT, do not set DRV_LOAD until the interface
7338 * is up. For all other cases, let the f/w know that the h/w is now
7339 * under the control of the driver.
7341 if (!(adapter->flags & FLAG_HAS_AMT))
7342 e1000e_get_hw_control(adapter);
7344 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7345 err = register_netdev(netdev);
7349 /* carrier off reporting is important to ethtool even BEFORE open */
7350 netif_carrier_off(netdev);
7352 e1000_print_device_info(adapter);
7354 if (pci_dev_run_wake(pdev))
7355 pm_runtime_put_noidle(&pdev->dev);
7360 if (!(adapter->flags & FLAG_HAS_AMT))
7361 e1000e_release_hw_control(adapter);
7363 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7364 e1000_phy_hw_reset(&adapter->hw);
7366 kfree(adapter->tx_ring);
7367 kfree(adapter->rx_ring);
7369 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7370 iounmap(adapter->hw.flash_address);
7371 e1000e_reset_interrupt_capability(adapter);
7373 iounmap(adapter->hw.hw_addr);
7375 free_netdev(netdev);
7377 pci_release_mem_regions(pdev);
7380 pci_disable_device(pdev);
7385 * e1000_remove - Device Removal Routine
7386 * @pdev: PCI device information struct
7388 * e1000_remove is called by the PCI subsystem to alert the driver
7389 * that it should release a PCI device. The could be caused by a
7390 * Hot-Plug event, or because the driver is going to be removed from
7393 static void e1000_remove(struct pci_dev *pdev)
7395 struct net_device *netdev = pci_get_drvdata(pdev);
7396 struct e1000_adapter *adapter = netdev_priv(netdev);
7397 bool down = test_bit(__E1000_DOWN, &adapter->state);
7399 e1000e_ptp_remove(adapter);
7401 /* The timers may be rescheduled, so explicitly disable them
7402 * from being rescheduled.
7405 set_bit(__E1000_DOWN, &adapter->state);
7406 del_timer_sync(&adapter->watchdog_timer);
7407 del_timer_sync(&adapter->phy_info_timer);
7409 cancel_work_sync(&adapter->reset_task);
7410 cancel_work_sync(&adapter->watchdog_task);
7411 cancel_work_sync(&adapter->downshift_task);
7412 cancel_work_sync(&adapter->update_phy_task);
7413 cancel_work_sync(&adapter->print_hang_task);
7415 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7416 cancel_work_sync(&adapter->tx_hwtstamp_work);
7417 if (adapter->tx_hwtstamp_skb) {
7418 dev_consume_skb_any(adapter->tx_hwtstamp_skb);
7419 adapter->tx_hwtstamp_skb = NULL;
7423 /* Don't lie to e1000_close() down the road. */
7425 clear_bit(__E1000_DOWN, &adapter->state);
7426 unregister_netdev(netdev);
7428 if (pci_dev_run_wake(pdev))
7429 pm_runtime_get_noresume(&pdev->dev);
7431 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7432 * would have already happened in close and is redundant.
7434 e1000e_release_hw_control(adapter);
7436 e1000e_reset_interrupt_capability(adapter);
7437 kfree(adapter->tx_ring);
7438 kfree(adapter->rx_ring);
7440 iounmap(adapter->hw.hw_addr);
7441 if ((adapter->hw.flash_address) &&
7442 (adapter->hw.mac.type < e1000_pch_spt))
7443 iounmap(adapter->hw.flash_address);
7444 pci_release_mem_regions(pdev);
7446 free_netdev(netdev);
7449 pci_disable_pcie_error_reporting(pdev);
7451 pci_disable_device(pdev);
7454 /* PCI Error Recovery (ERS) */
7455 static const struct pci_error_handlers e1000_err_handler = {
7456 .error_detected = e1000_io_error_detected,
7457 .slot_reset = e1000_io_slot_reset,
7458 .resume = e1000_io_resume,
7461 static const struct pci_device_id e1000_pci_tbl[] = {
7462 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7467 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7468 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7473 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7479 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7482 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7483 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7487 board_80003es2lan },
7488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7489 board_80003es2lan },
7490 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7491 board_80003es2lan },
7492 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7493 board_80003es2lan },
7495 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7504 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7505 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7507 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7508 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7509 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7511 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7514 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7515 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7516 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7518 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7519 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7520 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7522 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7523 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7524 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7525 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7527 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7528 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7530 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7531 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7532 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7533 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7534 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7535 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7536 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7537 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7538 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7539 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7540 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7541 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7542 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7543 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7544 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7545 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7546 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7547 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7548 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7549 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7550 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7551 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM8), board_pch_cnp },
7552 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V8), board_pch_cnp },
7553 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM9), board_pch_cnp },
7554 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V9), board_pch_cnp },
7556 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7558 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7560 static const struct dev_pm_ops e1000_pm_ops = {
7561 #ifdef CONFIG_PM_SLEEP
7562 .suspend = e1000e_pm_suspend,
7563 .resume = e1000e_pm_resume,
7564 .freeze = e1000e_pm_freeze,
7565 .thaw = e1000e_pm_thaw,
7566 .poweroff = e1000e_pm_suspend,
7567 .restore = e1000e_pm_resume,
7569 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7570 e1000e_pm_runtime_idle)
7573 /* PCI Device API Driver */
7574 static struct pci_driver e1000_driver = {
7575 .name = e1000e_driver_name,
7576 .id_table = e1000_pci_tbl,
7577 .probe = e1000_probe,
7578 .remove = e1000_remove,
7580 .pm = &e1000_pm_ops,
7582 .shutdown = e1000_shutdown,
7583 .err_handler = &e1000_err_handler
7587 * e1000_init_module - Driver Registration Routine
7589 * e1000_init_module is the first routine called when the driver is
7590 * loaded. All it does is register with the PCI subsystem.
7592 static int __init e1000_init_module(void)
7594 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7595 e1000e_driver_version);
7596 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7598 return pci_register_driver(&e1000_driver);
7600 module_init(e1000_init_module);
7603 * e1000_exit_module - Driver Exit Cleanup Routine
7605 * e1000_exit_module is called just before the driver is removed
7608 static void __exit e1000_exit_module(void)
7610 pci_unregister_driver(&e1000_driver);
7612 module_exit(e1000_exit_module);
7614 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7615 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7616 MODULE_LICENSE("GPL");
7617 MODULE_VERSION(DRV_VERSION);