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)
1076 struct e1000_adapter *adapter = tx_ring->adapter;
1078 if (buffer_info->dma) {
1079 if (buffer_info->mapped_as_page)
1080 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1081 buffer_info->length, DMA_TO_DEVICE);
1083 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1084 buffer_info->length, DMA_TO_DEVICE);
1085 buffer_info->dma = 0;
1087 if (buffer_info->skb) {
1088 dev_kfree_skb_any(buffer_info->skb);
1089 buffer_info->skb = NULL;
1091 buffer_info->time_stamp = 0;
1094 static void e1000_print_hw_hang(struct work_struct *work)
1096 struct e1000_adapter *adapter = container_of(work,
1097 struct e1000_adapter,
1099 struct net_device *netdev = adapter->netdev;
1100 struct e1000_ring *tx_ring = adapter->tx_ring;
1101 unsigned int i = tx_ring->next_to_clean;
1102 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1103 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1104 struct e1000_hw *hw = &adapter->hw;
1105 u16 phy_status, phy_1000t_status, phy_ext_status;
1108 if (test_bit(__E1000_DOWN, &adapter->state))
1111 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1112 /* May be block on write-back, flush and detect again
1113 * flush pending descriptor writebacks to memory
1115 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1116 /* execute the writes immediately */
1118 /* Due to rare timing issues, write to TIDV again to ensure
1119 * the write is successful
1121 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1122 /* execute the writes immediately */
1124 adapter->tx_hang_recheck = true;
1127 adapter->tx_hang_recheck = false;
1129 if (er32(TDH(0)) == er32(TDT(0))) {
1130 e_dbg("false hang detected, ignoring\n");
1134 /* Real hang detected */
1135 netif_stop_queue(netdev);
1137 e1e_rphy(hw, MII_BMSR, &phy_status);
1138 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1139 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1141 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1143 /* detected Hardware unit hang */
1144 e_err("Detected Hardware Unit Hang:\n"
1147 " next_to_use <%x>\n"
1148 " next_to_clean <%x>\n"
1149 "buffer_info[next_to_clean]:\n"
1150 " time_stamp <%lx>\n"
1151 " next_to_watch <%x>\n"
1153 " next_to_watch.status <%x>\n"
1156 "PHY 1000BASE-T Status <%x>\n"
1157 "PHY Extended Status <%x>\n"
1158 "PCI Status <%x>\n",
1159 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1160 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1161 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1162 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1164 e1000e_dump(adapter);
1166 /* Suggest workaround for known h/w issue */
1167 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1168 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1172 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1173 * @work: pointer to work struct
1175 * This work function polls the TSYNCTXCTL valid bit to determine when a
1176 * timestamp has been taken for the current stored skb. The timestamp must
1177 * be for this skb because only one such packet is allowed in the queue.
1179 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1181 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1183 struct e1000_hw *hw = &adapter->hw;
1185 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1186 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1187 struct skb_shared_hwtstamps shhwtstamps;
1190 txstmp = er32(TXSTMPL);
1191 txstmp |= (u64)er32(TXSTMPH) << 32;
1193 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1195 /* Clear the global tx_hwtstamp_skb pointer and force writes
1196 * prior to notifying the stack of a Tx timestamp.
1198 adapter->tx_hwtstamp_skb = NULL;
1199 wmb(); /* force write prior to skb_tstamp_tx */
1201 skb_tstamp_tx(skb, &shhwtstamps);
1202 dev_kfree_skb_any(skb);
1203 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1204 + adapter->tx_timeout_factor * HZ)) {
1205 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1206 adapter->tx_hwtstamp_skb = NULL;
1207 adapter->tx_hwtstamp_timeouts++;
1208 e_warn("clearing Tx timestamp hang\n");
1210 /* reschedule to check later */
1211 schedule_work(&adapter->tx_hwtstamp_work);
1216 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1217 * @tx_ring: Tx descriptor ring
1219 * the return value indicates whether actual cleaning was done, there
1220 * is no guarantee that everything was cleaned
1222 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1224 struct e1000_adapter *adapter = tx_ring->adapter;
1225 struct net_device *netdev = adapter->netdev;
1226 struct e1000_hw *hw = &adapter->hw;
1227 struct e1000_tx_desc *tx_desc, *eop_desc;
1228 struct e1000_buffer *buffer_info;
1229 unsigned int i, eop;
1230 unsigned int count = 0;
1231 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1232 unsigned int bytes_compl = 0, pkts_compl = 0;
1234 i = tx_ring->next_to_clean;
1235 eop = tx_ring->buffer_info[i].next_to_watch;
1236 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1238 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1239 (count < tx_ring->count)) {
1240 bool cleaned = false;
1242 dma_rmb(); /* read buffer_info after eop_desc */
1243 for (; !cleaned; count++) {
1244 tx_desc = E1000_TX_DESC(*tx_ring, i);
1245 buffer_info = &tx_ring->buffer_info[i];
1246 cleaned = (i == eop);
1249 total_tx_packets += buffer_info->segs;
1250 total_tx_bytes += buffer_info->bytecount;
1251 if (buffer_info->skb) {
1252 bytes_compl += buffer_info->skb->len;
1257 e1000_put_txbuf(tx_ring, buffer_info);
1258 tx_desc->upper.data = 0;
1261 if (i == tx_ring->count)
1265 if (i == tx_ring->next_to_use)
1267 eop = tx_ring->buffer_info[i].next_to_watch;
1268 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1271 tx_ring->next_to_clean = i;
1273 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1275 #define TX_WAKE_THRESHOLD 32
1276 if (count && netif_carrier_ok(netdev) &&
1277 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1278 /* Make sure that anybody stopping the queue after this
1279 * sees the new next_to_clean.
1283 if (netif_queue_stopped(netdev) &&
1284 !(test_bit(__E1000_DOWN, &adapter->state))) {
1285 netif_wake_queue(netdev);
1286 ++adapter->restart_queue;
1290 if (adapter->detect_tx_hung) {
1291 /* Detect a transmit hang in hardware, this serializes the
1292 * check with the clearing of time_stamp and movement of i
1294 adapter->detect_tx_hung = false;
1295 if (tx_ring->buffer_info[i].time_stamp &&
1296 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1297 + (adapter->tx_timeout_factor * HZ)) &&
1298 !(er32(STATUS) & E1000_STATUS_TXOFF))
1299 schedule_work(&adapter->print_hang_task);
1301 adapter->tx_hang_recheck = false;
1303 adapter->total_tx_bytes += total_tx_bytes;
1304 adapter->total_tx_packets += total_tx_packets;
1305 return count < tx_ring->count;
1309 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1310 * @rx_ring: Rx descriptor ring
1312 * the return value indicates whether actual cleaning was done, there
1313 * is no guarantee that everything was cleaned
1315 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1318 struct e1000_adapter *adapter = rx_ring->adapter;
1319 struct e1000_hw *hw = &adapter->hw;
1320 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1321 struct net_device *netdev = adapter->netdev;
1322 struct pci_dev *pdev = adapter->pdev;
1323 struct e1000_buffer *buffer_info, *next_buffer;
1324 struct e1000_ps_page *ps_page;
1325 struct sk_buff *skb;
1327 u32 length, staterr;
1328 int cleaned_count = 0;
1329 bool cleaned = false;
1330 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1332 i = rx_ring->next_to_clean;
1333 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1334 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1335 buffer_info = &rx_ring->buffer_info[i];
1337 while (staterr & E1000_RXD_STAT_DD) {
1338 if (*work_done >= work_to_do)
1341 skb = buffer_info->skb;
1342 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1344 /* in the packet split case this is header only */
1345 prefetch(skb->data - NET_IP_ALIGN);
1348 if (i == rx_ring->count)
1350 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1353 next_buffer = &rx_ring->buffer_info[i];
1357 dma_unmap_single(&pdev->dev, buffer_info->dma,
1358 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1359 buffer_info->dma = 0;
1361 /* see !EOP comment in other Rx routine */
1362 if (!(staterr & E1000_RXD_STAT_EOP))
1363 adapter->flags2 |= FLAG2_IS_DISCARDING;
1365 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1366 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1367 dev_kfree_skb_irq(skb);
1368 if (staterr & E1000_RXD_STAT_EOP)
1369 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1373 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1374 !(netdev->features & NETIF_F_RXALL))) {
1375 dev_kfree_skb_irq(skb);
1379 length = le16_to_cpu(rx_desc->wb.middle.length0);
1382 e_dbg("Last part of the packet spanning multiple descriptors\n");
1383 dev_kfree_skb_irq(skb);
1388 skb_put(skb, length);
1391 /* this looks ugly, but it seems compiler issues make
1392 * it more efficient than reusing j
1394 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1396 /* page alloc/put takes too long and effects small
1397 * packet throughput, so unsplit small packets and
1398 * save the alloc/put only valid in softirq (napi)
1399 * context to call kmap_*
1401 if (l1 && (l1 <= copybreak) &&
1402 ((length + l1) <= adapter->rx_ps_bsize0)) {
1405 ps_page = &buffer_info->ps_pages[0];
1407 /* there is no documentation about how to call
1408 * kmap_atomic, so we can't hold the mapping
1411 dma_sync_single_for_cpu(&pdev->dev,
1415 vaddr = kmap_atomic(ps_page->page);
1416 memcpy(skb_tail_pointer(skb), vaddr, l1);
1417 kunmap_atomic(vaddr);
1418 dma_sync_single_for_device(&pdev->dev,
1423 /* remove the CRC */
1424 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1425 if (!(netdev->features & NETIF_F_RXFCS))
1434 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1435 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1439 ps_page = &buffer_info->ps_pages[j];
1440 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1443 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1444 ps_page->page = NULL;
1446 skb->data_len += length;
1447 skb->truesize += PAGE_SIZE;
1450 /* strip the ethernet crc, problem is we're using pages now so
1451 * this whole operation can get a little cpu intensive
1453 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1454 if (!(netdev->features & NETIF_F_RXFCS))
1455 pskb_trim(skb, skb->len - 4);
1459 total_rx_bytes += skb->len;
1462 e1000_rx_checksum(adapter, staterr, skb);
1464 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1466 if (rx_desc->wb.upper.header_status &
1467 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1468 adapter->rx_hdr_split++;
1470 e1000_receive_skb(adapter, netdev, skb, staterr,
1471 rx_desc->wb.middle.vlan);
1474 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1475 buffer_info->skb = NULL;
1477 /* return some buffers to hardware, one at a time is too slow */
1478 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1479 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1484 /* use prefetched values */
1486 buffer_info = next_buffer;
1488 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1490 rx_ring->next_to_clean = i;
1492 cleaned_count = e1000_desc_unused(rx_ring);
1494 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1496 adapter->total_rx_bytes += total_rx_bytes;
1497 adapter->total_rx_packets += total_rx_packets;
1502 * e1000_consume_page - helper function
1504 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1509 skb->data_len += length;
1510 skb->truesize += PAGE_SIZE;
1514 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1515 * @adapter: board private structure
1517 * the return value indicates whether actual cleaning was done, there
1518 * is no guarantee that everything was cleaned
1520 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1523 struct e1000_adapter *adapter = rx_ring->adapter;
1524 struct net_device *netdev = adapter->netdev;
1525 struct pci_dev *pdev = adapter->pdev;
1526 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1527 struct e1000_buffer *buffer_info, *next_buffer;
1528 u32 length, staterr;
1530 int cleaned_count = 0;
1531 bool cleaned = false;
1532 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1533 struct skb_shared_info *shinfo;
1535 i = rx_ring->next_to_clean;
1536 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1537 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1538 buffer_info = &rx_ring->buffer_info[i];
1540 while (staterr & E1000_RXD_STAT_DD) {
1541 struct sk_buff *skb;
1543 if (*work_done >= work_to_do)
1546 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1548 skb = buffer_info->skb;
1549 buffer_info->skb = NULL;
1552 if (i == rx_ring->count)
1554 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1557 next_buffer = &rx_ring->buffer_info[i];
1561 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1563 buffer_info->dma = 0;
1565 length = le16_to_cpu(rx_desc->wb.upper.length);
1567 /* errors is only valid for DD + EOP descriptors */
1568 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1569 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1570 !(netdev->features & NETIF_F_RXALL)))) {
1571 /* recycle both page and skb */
1572 buffer_info->skb = skb;
1573 /* an error means any chain goes out the window too */
1574 if (rx_ring->rx_skb_top)
1575 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1576 rx_ring->rx_skb_top = NULL;
1579 #define rxtop (rx_ring->rx_skb_top)
1580 if (!(staterr & E1000_RXD_STAT_EOP)) {
1581 /* this descriptor is only the beginning (or middle) */
1583 /* this is the beginning of a chain */
1585 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1588 /* this is the middle of a chain */
1589 shinfo = skb_shinfo(rxtop);
1590 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1591 buffer_info->page, 0,
1593 /* re-use the skb, only consumed the page */
1594 buffer_info->skb = skb;
1596 e1000_consume_page(buffer_info, rxtop, length);
1600 /* end of the chain */
1601 shinfo = skb_shinfo(rxtop);
1602 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1603 buffer_info->page, 0,
1605 /* re-use the current skb, we only consumed the
1608 buffer_info->skb = skb;
1611 e1000_consume_page(buffer_info, skb, length);
1613 /* no chain, got EOP, this buf is the packet
1614 * copybreak to save the put_page/alloc_page
1616 if (length <= copybreak &&
1617 skb_tailroom(skb) >= length) {
1619 vaddr = kmap_atomic(buffer_info->page);
1620 memcpy(skb_tail_pointer(skb), vaddr,
1622 kunmap_atomic(vaddr);
1623 /* re-use the page, so don't erase
1626 skb_put(skb, length);
1628 skb_fill_page_desc(skb, 0,
1629 buffer_info->page, 0,
1631 e1000_consume_page(buffer_info, skb,
1637 /* Receive Checksum Offload */
1638 e1000_rx_checksum(adapter, staterr, skb);
1640 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1642 /* probably a little skewed due to removing CRC */
1643 total_rx_bytes += skb->len;
1646 /* eth type trans needs skb->data to point to something */
1647 if (!pskb_may_pull(skb, ETH_HLEN)) {
1648 e_err("pskb_may_pull failed.\n");
1649 dev_kfree_skb_irq(skb);
1653 e1000_receive_skb(adapter, netdev, skb, staterr,
1654 rx_desc->wb.upper.vlan);
1657 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1659 /* return some buffers to hardware, one at a time is too slow */
1660 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1661 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1666 /* use prefetched values */
1668 buffer_info = next_buffer;
1670 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1672 rx_ring->next_to_clean = i;
1674 cleaned_count = e1000_desc_unused(rx_ring);
1676 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1678 adapter->total_rx_bytes += total_rx_bytes;
1679 adapter->total_rx_packets += total_rx_packets;
1684 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1685 * @rx_ring: Rx descriptor ring
1687 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1689 struct e1000_adapter *adapter = rx_ring->adapter;
1690 struct e1000_buffer *buffer_info;
1691 struct e1000_ps_page *ps_page;
1692 struct pci_dev *pdev = adapter->pdev;
1695 /* Free all the Rx ring sk_buffs */
1696 for (i = 0; i < rx_ring->count; i++) {
1697 buffer_info = &rx_ring->buffer_info[i];
1698 if (buffer_info->dma) {
1699 if (adapter->clean_rx == e1000_clean_rx_irq)
1700 dma_unmap_single(&pdev->dev, buffer_info->dma,
1701 adapter->rx_buffer_len,
1703 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1704 dma_unmap_page(&pdev->dev, buffer_info->dma,
1705 PAGE_SIZE, DMA_FROM_DEVICE);
1706 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1707 dma_unmap_single(&pdev->dev, buffer_info->dma,
1708 adapter->rx_ps_bsize0,
1710 buffer_info->dma = 0;
1713 if (buffer_info->page) {
1714 put_page(buffer_info->page);
1715 buffer_info->page = NULL;
1718 if (buffer_info->skb) {
1719 dev_kfree_skb(buffer_info->skb);
1720 buffer_info->skb = NULL;
1723 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1724 ps_page = &buffer_info->ps_pages[j];
1727 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1730 put_page(ps_page->page);
1731 ps_page->page = NULL;
1735 /* there also may be some cached data from a chained receive */
1736 if (rx_ring->rx_skb_top) {
1737 dev_kfree_skb(rx_ring->rx_skb_top);
1738 rx_ring->rx_skb_top = NULL;
1741 /* Zero out the descriptor ring */
1742 memset(rx_ring->desc, 0, rx_ring->size);
1744 rx_ring->next_to_clean = 0;
1745 rx_ring->next_to_use = 0;
1746 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1749 static void e1000e_downshift_workaround(struct work_struct *work)
1751 struct e1000_adapter *adapter = container_of(work,
1752 struct e1000_adapter,
1755 if (test_bit(__E1000_DOWN, &adapter->state))
1758 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1762 * e1000_intr_msi - Interrupt Handler
1763 * @irq: interrupt number
1764 * @data: pointer to a network interface device structure
1766 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1768 struct net_device *netdev = data;
1769 struct e1000_adapter *adapter = netdev_priv(netdev);
1770 struct e1000_hw *hw = &adapter->hw;
1771 u32 icr = er32(ICR);
1773 /* read ICR disables interrupts using IAM */
1774 if (icr & E1000_ICR_LSC) {
1775 hw->mac.get_link_status = true;
1776 /* ICH8 workaround-- Call gig speed drop workaround on cable
1777 * disconnect (LSC) before accessing any PHY registers
1779 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1780 (!(er32(STATUS) & E1000_STATUS_LU)))
1781 schedule_work(&adapter->downshift_task);
1783 /* 80003ES2LAN workaround-- For packet buffer work-around on
1784 * link down event; disable receives here in the ISR and reset
1785 * adapter in watchdog
1787 if (netif_carrier_ok(netdev) &&
1788 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1789 /* disable receives */
1790 u32 rctl = er32(RCTL);
1792 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1793 adapter->flags |= FLAG_RESTART_NOW;
1795 /* guard against interrupt when we're going down */
1796 if (!test_bit(__E1000_DOWN, &adapter->state))
1797 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1800 /* Reset on uncorrectable ECC error */
1801 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1802 u32 pbeccsts = er32(PBECCSTS);
1804 adapter->corr_errors +=
1805 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1806 adapter->uncorr_errors +=
1807 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1808 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1810 /* Do the reset outside of interrupt context */
1811 schedule_work(&adapter->reset_task);
1813 /* return immediately since reset is imminent */
1817 if (napi_schedule_prep(&adapter->napi)) {
1818 adapter->total_tx_bytes = 0;
1819 adapter->total_tx_packets = 0;
1820 adapter->total_rx_bytes = 0;
1821 adapter->total_rx_packets = 0;
1822 __napi_schedule(&adapter->napi);
1829 * e1000_intr - Interrupt Handler
1830 * @irq: interrupt number
1831 * @data: pointer to a network interface device structure
1833 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1835 struct net_device *netdev = data;
1836 struct e1000_adapter *adapter = netdev_priv(netdev);
1837 struct e1000_hw *hw = &adapter->hw;
1838 u32 rctl, icr = er32(ICR);
1840 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1841 return IRQ_NONE; /* Not our interrupt */
1843 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1844 * not set, then the adapter didn't send an interrupt
1846 if (!(icr & E1000_ICR_INT_ASSERTED))
1849 /* Interrupt Auto-Mask...upon reading ICR,
1850 * interrupts are masked. No need for the
1854 if (icr & E1000_ICR_LSC) {
1855 hw->mac.get_link_status = true;
1856 /* ICH8 workaround-- Call gig speed drop workaround on cable
1857 * disconnect (LSC) before accessing any PHY registers
1859 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1860 (!(er32(STATUS) & E1000_STATUS_LU)))
1861 schedule_work(&adapter->downshift_task);
1863 /* 80003ES2LAN workaround--
1864 * For packet buffer work-around on link down event;
1865 * disable receives here in the ISR and
1866 * reset adapter in watchdog
1868 if (netif_carrier_ok(netdev) &&
1869 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1870 /* disable receives */
1872 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1873 adapter->flags |= FLAG_RESTART_NOW;
1875 /* guard against interrupt when we're going down */
1876 if (!test_bit(__E1000_DOWN, &adapter->state))
1877 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1880 /* Reset on uncorrectable ECC error */
1881 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1882 u32 pbeccsts = er32(PBECCSTS);
1884 adapter->corr_errors +=
1885 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1886 adapter->uncorr_errors +=
1887 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1888 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1890 /* Do the reset outside of interrupt context */
1891 schedule_work(&adapter->reset_task);
1893 /* return immediately since reset is imminent */
1897 if (napi_schedule_prep(&adapter->napi)) {
1898 adapter->total_tx_bytes = 0;
1899 adapter->total_tx_packets = 0;
1900 adapter->total_rx_bytes = 0;
1901 adapter->total_rx_packets = 0;
1902 __napi_schedule(&adapter->napi);
1908 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1910 struct net_device *netdev = data;
1911 struct e1000_adapter *adapter = netdev_priv(netdev);
1912 struct e1000_hw *hw = &adapter->hw;
1914 hw->mac.get_link_status = true;
1916 /* guard against interrupt when we're going down */
1917 if (!test_bit(__E1000_DOWN, &adapter->state)) {
1918 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1919 ew32(IMS, E1000_IMS_OTHER);
1925 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1927 struct net_device *netdev = data;
1928 struct e1000_adapter *adapter = netdev_priv(netdev);
1929 struct e1000_hw *hw = &adapter->hw;
1930 struct e1000_ring *tx_ring = adapter->tx_ring;
1932 adapter->total_tx_bytes = 0;
1933 adapter->total_tx_packets = 0;
1935 if (!e1000_clean_tx_irq(tx_ring))
1936 /* Ring was not completely cleaned, so fire another interrupt */
1937 ew32(ICS, tx_ring->ims_val);
1939 if (!test_bit(__E1000_DOWN, &adapter->state))
1940 ew32(IMS, adapter->tx_ring->ims_val);
1945 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1947 struct net_device *netdev = data;
1948 struct e1000_adapter *adapter = netdev_priv(netdev);
1949 struct e1000_ring *rx_ring = adapter->rx_ring;
1951 /* Write the ITR value calculated at the end of the
1952 * previous interrupt.
1954 if (rx_ring->set_itr) {
1955 u32 itr = rx_ring->itr_val ?
1956 1000000000 / (rx_ring->itr_val * 256) : 0;
1958 writel(itr, rx_ring->itr_register);
1959 rx_ring->set_itr = 0;
1962 if (napi_schedule_prep(&adapter->napi)) {
1963 adapter->total_rx_bytes = 0;
1964 adapter->total_rx_packets = 0;
1965 __napi_schedule(&adapter->napi);
1971 * e1000_configure_msix - Configure MSI-X hardware
1973 * e1000_configure_msix sets up the hardware to properly
1974 * generate MSI-X interrupts.
1976 static void e1000_configure_msix(struct e1000_adapter *adapter)
1978 struct e1000_hw *hw = &adapter->hw;
1979 struct e1000_ring *rx_ring = adapter->rx_ring;
1980 struct e1000_ring *tx_ring = adapter->tx_ring;
1982 u32 ctrl_ext, ivar = 0;
1984 adapter->eiac_mask = 0;
1986 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1987 if (hw->mac.type == e1000_82574) {
1988 u32 rfctl = er32(RFCTL);
1990 rfctl |= E1000_RFCTL_ACK_DIS;
1994 /* Configure Rx vector */
1995 rx_ring->ims_val = E1000_IMS_RXQ0;
1996 adapter->eiac_mask |= rx_ring->ims_val;
1997 if (rx_ring->itr_val)
1998 writel(1000000000 / (rx_ring->itr_val * 256),
1999 rx_ring->itr_register);
2001 writel(1, rx_ring->itr_register);
2002 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
2004 /* Configure Tx vector */
2005 tx_ring->ims_val = E1000_IMS_TXQ0;
2007 if (tx_ring->itr_val)
2008 writel(1000000000 / (tx_ring->itr_val * 256),
2009 tx_ring->itr_register);
2011 writel(1, tx_ring->itr_register);
2012 adapter->eiac_mask |= tx_ring->ims_val;
2013 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2015 /* set vector for Other Causes, e.g. link changes */
2017 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2018 if (rx_ring->itr_val)
2019 writel(1000000000 / (rx_ring->itr_val * 256),
2020 hw->hw_addr + E1000_EITR_82574(vector));
2022 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2023 adapter->eiac_mask |= E1000_IMS_OTHER;
2025 /* Cause Tx interrupts on every write back */
2030 /* enable MSI-X PBA support */
2031 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2032 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2033 ew32(CTRL_EXT, ctrl_ext);
2037 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2039 if (adapter->msix_entries) {
2040 pci_disable_msix(adapter->pdev);
2041 kfree(adapter->msix_entries);
2042 adapter->msix_entries = NULL;
2043 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2044 pci_disable_msi(adapter->pdev);
2045 adapter->flags &= ~FLAG_MSI_ENABLED;
2050 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2052 * Attempt to configure interrupts using the best available
2053 * capabilities of the hardware and kernel.
2055 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2060 switch (adapter->int_mode) {
2061 case E1000E_INT_MODE_MSIX:
2062 if (adapter->flags & FLAG_HAS_MSIX) {
2063 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2064 adapter->msix_entries = kcalloc(adapter->num_vectors,
2068 if (adapter->msix_entries) {
2069 struct e1000_adapter *a = adapter;
2071 for (i = 0; i < adapter->num_vectors; i++)
2072 adapter->msix_entries[i].entry = i;
2074 err = pci_enable_msix_range(a->pdev,
2081 /* MSI-X failed, so fall through and try MSI */
2082 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2083 e1000e_reset_interrupt_capability(adapter);
2085 adapter->int_mode = E1000E_INT_MODE_MSI;
2087 case E1000E_INT_MODE_MSI:
2088 if (!pci_enable_msi(adapter->pdev)) {
2089 adapter->flags |= FLAG_MSI_ENABLED;
2091 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2092 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2095 case E1000E_INT_MODE_LEGACY:
2096 /* Don't do anything; this is the system default */
2100 /* store the number of vectors being used */
2101 adapter->num_vectors = 1;
2105 * e1000_request_msix - Initialize MSI-X interrupts
2107 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2110 static int e1000_request_msix(struct e1000_adapter *adapter)
2112 struct net_device *netdev = adapter->netdev;
2113 int err = 0, vector = 0;
2115 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2116 snprintf(adapter->rx_ring->name,
2117 sizeof(adapter->rx_ring->name) - 1,
2118 "%s-rx-0", netdev->name);
2120 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2121 err = request_irq(adapter->msix_entries[vector].vector,
2122 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2126 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2127 E1000_EITR_82574(vector);
2128 adapter->rx_ring->itr_val = adapter->itr;
2131 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2132 snprintf(adapter->tx_ring->name,
2133 sizeof(adapter->tx_ring->name) - 1,
2134 "%s-tx-0", netdev->name);
2136 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2137 err = request_irq(adapter->msix_entries[vector].vector,
2138 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2142 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2143 E1000_EITR_82574(vector);
2144 adapter->tx_ring->itr_val = adapter->itr;
2147 err = request_irq(adapter->msix_entries[vector].vector,
2148 e1000_msix_other, 0, netdev->name, netdev);
2152 e1000_configure_msix(adapter);
2158 * e1000_request_irq - initialize interrupts
2160 * Attempts to configure interrupts using the best available
2161 * capabilities of the hardware and kernel.
2163 static int e1000_request_irq(struct e1000_adapter *adapter)
2165 struct net_device *netdev = adapter->netdev;
2168 if (adapter->msix_entries) {
2169 err = e1000_request_msix(adapter);
2172 /* fall back to MSI */
2173 e1000e_reset_interrupt_capability(adapter);
2174 adapter->int_mode = E1000E_INT_MODE_MSI;
2175 e1000e_set_interrupt_capability(adapter);
2177 if (adapter->flags & FLAG_MSI_ENABLED) {
2178 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2179 netdev->name, netdev);
2183 /* fall back to legacy interrupt */
2184 e1000e_reset_interrupt_capability(adapter);
2185 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2188 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2189 netdev->name, netdev);
2191 e_err("Unable to allocate interrupt, Error: %d\n", err);
2196 static void e1000_free_irq(struct e1000_adapter *adapter)
2198 struct net_device *netdev = adapter->netdev;
2200 if (adapter->msix_entries) {
2203 free_irq(adapter->msix_entries[vector].vector, netdev);
2206 free_irq(adapter->msix_entries[vector].vector, netdev);
2209 /* Other Causes interrupt vector */
2210 free_irq(adapter->msix_entries[vector].vector, netdev);
2214 free_irq(adapter->pdev->irq, netdev);
2218 * e1000_irq_disable - Mask off interrupt generation on the NIC
2220 static void e1000_irq_disable(struct e1000_adapter *adapter)
2222 struct e1000_hw *hw = &adapter->hw;
2225 if (adapter->msix_entries)
2226 ew32(EIAC_82574, 0);
2229 if (adapter->msix_entries) {
2232 for (i = 0; i < adapter->num_vectors; i++)
2233 synchronize_irq(adapter->msix_entries[i].vector);
2235 synchronize_irq(adapter->pdev->irq);
2240 * e1000_irq_enable - Enable default interrupt generation settings
2242 static void e1000_irq_enable(struct e1000_adapter *adapter)
2244 struct e1000_hw *hw = &adapter->hw;
2246 if (adapter->msix_entries) {
2247 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2248 ew32(IMS, adapter->eiac_mask | E1000_IMS_LSC);
2249 } else if (hw->mac.type >= e1000_pch_lpt) {
2250 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2252 ew32(IMS, IMS_ENABLE_MASK);
2258 * e1000e_get_hw_control - get control of the h/w from f/w
2259 * @adapter: address of board private structure
2261 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2262 * For ASF and Pass Through versions of f/w this means that
2263 * the driver is loaded. For AMT version (only with 82573)
2264 * of the f/w this means that the network i/f is open.
2266 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2268 struct e1000_hw *hw = &adapter->hw;
2272 /* Let firmware know the driver has taken over */
2273 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2275 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2276 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2277 ctrl_ext = er32(CTRL_EXT);
2278 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2283 * e1000e_release_hw_control - release control of the h/w to f/w
2284 * @adapter: address of board private structure
2286 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2287 * For ASF and Pass Through versions of f/w this means that the
2288 * driver is no longer loaded. For AMT version (only with 82573) i
2289 * of the f/w this means that the network i/f is closed.
2292 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2294 struct e1000_hw *hw = &adapter->hw;
2298 /* Let firmware taken over control of h/w */
2299 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2301 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2302 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2303 ctrl_ext = er32(CTRL_EXT);
2304 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2309 * e1000_alloc_ring_dma - allocate memory for a ring structure
2311 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2312 struct e1000_ring *ring)
2314 struct pci_dev *pdev = adapter->pdev;
2316 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2325 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2326 * @tx_ring: Tx descriptor ring
2328 * Return 0 on success, negative on failure
2330 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2332 struct e1000_adapter *adapter = tx_ring->adapter;
2333 int err = -ENOMEM, size;
2335 size = sizeof(struct e1000_buffer) * tx_ring->count;
2336 tx_ring->buffer_info = vzalloc(size);
2337 if (!tx_ring->buffer_info)
2340 /* round up to nearest 4K */
2341 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2342 tx_ring->size = ALIGN(tx_ring->size, 4096);
2344 err = e1000_alloc_ring_dma(adapter, tx_ring);
2348 tx_ring->next_to_use = 0;
2349 tx_ring->next_to_clean = 0;
2353 vfree(tx_ring->buffer_info);
2354 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2359 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2360 * @rx_ring: Rx descriptor ring
2362 * Returns 0 on success, negative on failure
2364 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2366 struct e1000_adapter *adapter = rx_ring->adapter;
2367 struct e1000_buffer *buffer_info;
2368 int i, size, desc_len, err = -ENOMEM;
2370 size = sizeof(struct e1000_buffer) * rx_ring->count;
2371 rx_ring->buffer_info = vzalloc(size);
2372 if (!rx_ring->buffer_info)
2375 for (i = 0; i < rx_ring->count; i++) {
2376 buffer_info = &rx_ring->buffer_info[i];
2377 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2378 sizeof(struct e1000_ps_page),
2380 if (!buffer_info->ps_pages)
2384 desc_len = sizeof(union e1000_rx_desc_packet_split);
2386 /* Round up to nearest 4K */
2387 rx_ring->size = rx_ring->count * desc_len;
2388 rx_ring->size = ALIGN(rx_ring->size, 4096);
2390 err = e1000_alloc_ring_dma(adapter, rx_ring);
2394 rx_ring->next_to_clean = 0;
2395 rx_ring->next_to_use = 0;
2396 rx_ring->rx_skb_top = NULL;
2401 for (i = 0; i < rx_ring->count; i++) {
2402 buffer_info = &rx_ring->buffer_info[i];
2403 kfree(buffer_info->ps_pages);
2406 vfree(rx_ring->buffer_info);
2407 e_err("Unable to allocate memory for the receive descriptor ring\n");
2412 * e1000_clean_tx_ring - Free Tx Buffers
2413 * @tx_ring: Tx descriptor ring
2415 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2417 struct e1000_adapter *adapter = tx_ring->adapter;
2418 struct e1000_buffer *buffer_info;
2422 for (i = 0; i < tx_ring->count; i++) {
2423 buffer_info = &tx_ring->buffer_info[i];
2424 e1000_put_txbuf(tx_ring, buffer_info);
2427 netdev_reset_queue(adapter->netdev);
2428 size = sizeof(struct e1000_buffer) * tx_ring->count;
2429 memset(tx_ring->buffer_info, 0, size);
2431 memset(tx_ring->desc, 0, tx_ring->size);
2433 tx_ring->next_to_use = 0;
2434 tx_ring->next_to_clean = 0;
2438 * e1000e_free_tx_resources - Free Tx Resources per Queue
2439 * @tx_ring: Tx descriptor ring
2441 * Free all transmit software resources
2443 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2445 struct e1000_adapter *adapter = tx_ring->adapter;
2446 struct pci_dev *pdev = adapter->pdev;
2448 e1000_clean_tx_ring(tx_ring);
2450 vfree(tx_ring->buffer_info);
2451 tx_ring->buffer_info = NULL;
2453 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2455 tx_ring->desc = NULL;
2459 * e1000e_free_rx_resources - Free Rx Resources
2460 * @rx_ring: Rx descriptor ring
2462 * Free all receive software resources
2464 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2466 struct e1000_adapter *adapter = rx_ring->adapter;
2467 struct pci_dev *pdev = adapter->pdev;
2470 e1000_clean_rx_ring(rx_ring);
2472 for (i = 0; i < rx_ring->count; i++)
2473 kfree(rx_ring->buffer_info[i].ps_pages);
2475 vfree(rx_ring->buffer_info);
2476 rx_ring->buffer_info = NULL;
2478 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2480 rx_ring->desc = NULL;
2484 * e1000_update_itr - update the dynamic ITR value based on statistics
2485 * @adapter: pointer to adapter
2486 * @itr_setting: current adapter->itr
2487 * @packets: the number of packets during this measurement interval
2488 * @bytes: the number of bytes during this measurement interval
2490 * Stores a new ITR value based on packets and byte
2491 * counts during the last interrupt. The advantage of per interrupt
2492 * computation is faster updates and more accurate ITR for the current
2493 * traffic pattern. Constants in this function were computed
2494 * based on theoretical maximum wire speed and thresholds were set based
2495 * on testing data as well as attempting to minimize response time
2496 * while increasing bulk throughput. This functionality is controlled
2497 * by the InterruptThrottleRate module parameter.
2499 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2501 unsigned int retval = itr_setting;
2506 switch (itr_setting) {
2507 case lowest_latency:
2508 /* handle TSO and jumbo frames */
2509 if (bytes / packets > 8000)
2510 retval = bulk_latency;
2511 else if ((packets < 5) && (bytes > 512))
2512 retval = low_latency;
2514 case low_latency: /* 50 usec aka 20000 ints/s */
2515 if (bytes > 10000) {
2516 /* this if handles the TSO accounting */
2517 if (bytes / packets > 8000)
2518 retval = bulk_latency;
2519 else if ((packets < 10) || ((bytes / packets) > 1200))
2520 retval = bulk_latency;
2521 else if ((packets > 35))
2522 retval = lowest_latency;
2523 } else if (bytes / packets > 2000) {
2524 retval = bulk_latency;
2525 } else if (packets <= 2 && bytes < 512) {
2526 retval = lowest_latency;
2529 case bulk_latency: /* 250 usec aka 4000 ints/s */
2530 if (bytes > 25000) {
2532 retval = low_latency;
2533 } else if (bytes < 6000) {
2534 retval = low_latency;
2542 static void e1000_set_itr(struct e1000_adapter *adapter)
2545 u32 new_itr = adapter->itr;
2547 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2548 if (adapter->link_speed != SPEED_1000) {
2554 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2559 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2560 adapter->total_tx_packets,
2561 adapter->total_tx_bytes);
2562 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2563 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2564 adapter->tx_itr = low_latency;
2566 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2567 adapter->total_rx_packets,
2568 adapter->total_rx_bytes);
2569 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2570 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2571 adapter->rx_itr = low_latency;
2573 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2575 /* counts and packets in update_itr are dependent on these numbers */
2576 switch (current_itr) {
2577 case lowest_latency:
2581 new_itr = 20000; /* aka hwitr = ~200 */
2591 if (new_itr != adapter->itr) {
2592 /* this attempts to bias the interrupt rate towards Bulk
2593 * by adding intermediate steps when interrupt rate is
2596 new_itr = new_itr > adapter->itr ?
2597 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2598 adapter->itr = new_itr;
2599 adapter->rx_ring->itr_val = new_itr;
2600 if (adapter->msix_entries)
2601 adapter->rx_ring->set_itr = 1;
2603 e1000e_write_itr(adapter, new_itr);
2608 * e1000e_write_itr - write the ITR value to the appropriate registers
2609 * @adapter: address of board private structure
2610 * @itr: new ITR value to program
2612 * e1000e_write_itr determines if the adapter is in MSI-X mode
2613 * and, if so, writes the EITR registers with the ITR value.
2614 * Otherwise, it writes the ITR value into the ITR register.
2616 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2618 struct e1000_hw *hw = &adapter->hw;
2619 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2621 if (adapter->msix_entries) {
2624 for (vector = 0; vector < adapter->num_vectors; vector++)
2625 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2632 * e1000_alloc_queues - Allocate memory for all rings
2633 * @adapter: board private structure to initialize
2635 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2637 int size = sizeof(struct e1000_ring);
2639 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2640 if (!adapter->tx_ring)
2642 adapter->tx_ring->count = adapter->tx_ring_count;
2643 adapter->tx_ring->adapter = adapter;
2645 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2646 if (!adapter->rx_ring)
2648 adapter->rx_ring->count = adapter->rx_ring_count;
2649 adapter->rx_ring->adapter = adapter;
2653 e_err("Unable to allocate memory for queues\n");
2654 kfree(adapter->rx_ring);
2655 kfree(adapter->tx_ring);
2660 * e1000e_poll - NAPI Rx polling callback
2661 * @napi: struct associated with this polling callback
2662 * @weight: number of packets driver is allowed to process this poll
2664 static int e1000e_poll(struct napi_struct *napi, int weight)
2666 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2668 struct e1000_hw *hw = &adapter->hw;
2669 struct net_device *poll_dev = adapter->netdev;
2670 int tx_cleaned = 1, work_done = 0;
2672 adapter = netdev_priv(poll_dev);
2674 if (!adapter->msix_entries ||
2675 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2676 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2678 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2683 /* If weight not fully consumed, exit the polling mode */
2684 if (work_done < weight) {
2685 if (adapter->itr_setting & 3)
2686 e1000_set_itr(adapter);
2687 napi_complete_done(napi, work_done);
2688 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2689 if (adapter->msix_entries)
2690 ew32(IMS, adapter->rx_ring->ims_val);
2692 e1000_irq_enable(adapter);
2699 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2700 __always_unused __be16 proto, u16 vid)
2702 struct e1000_adapter *adapter = netdev_priv(netdev);
2703 struct e1000_hw *hw = &adapter->hw;
2706 /* don't update vlan cookie if already programmed */
2707 if ((adapter->hw.mng_cookie.status &
2708 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2709 (vid == adapter->mng_vlan_id))
2712 /* add VID to filter table */
2713 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2714 index = (vid >> 5) & 0x7F;
2715 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2716 vfta |= BIT((vid & 0x1F));
2717 hw->mac.ops.write_vfta(hw, index, vfta);
2720 set_bit(vid, adapter->active_vlans);
2725 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2726 __always_unused __be16 proto, u16 vid)
2728 struct e1000_adapter *adapter = netdev_priv(netdev);
2729 struct e1000_hw *hw = &adapter->hw;
2732 if ((adapter->hw.mng_cookie.status &
2733 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2734 (vid == adapter->mng_vlan_id)) {
2735 /* release control to f/w */
2736 e1000e_release_hw_control(adapter);
2740 /* remove VID from filter table */
2741 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2742 index = (vid >> 5) & 0x7F;
2743 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2744 vfta &= ~BIT((vid & 0x1F));
2745 hw->mac.ops.write_vfta(hw, index, vfta);
2748 clear_bit(vid, adapter->active_vlans);
2754 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2755 * @adapter: board private structure to initialize
2757 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2759 struct net_device *netdev = adapter->netdev;
2760 struct e1000_hw *hw = &adapter->hw;
2763 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2764 /* disable VLAN receive filtering */
2766 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2769 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2770 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2771 adapter->mng_vlan_id);
2772 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2778 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2779 * @adapter: board private structure to initialize
2781 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2783 struct e1000_hw *hw = &adapter->hw;
2786 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2787 /* enable VLAN receive filtering */
2789 rctl |= E1000_RCTL_VFE;
2790 rctl &= ~E1000_RCTL_CFIEN;
2796 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2797 * @adapter: board private structure to initialize
2799 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2801 struct e1000_hw *hw = &adapter->hw;
2804 /* disable VLAN tag insert/strip */
2806 ctrl &= ~E1000_CTRL_VME;
2811 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2812 * @adapter: board private structure to initialize
2814 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2816 struct e1000_hw *hw = &adapter->hw;
2819 /* enable VLAN tag insert/strip */
2821 ctrl |= E1000_CTRL_VME;
2825 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2827 struct net_device *netdev = adapter->netdev;
2828 u16 vid = adapter->hw.mng_cookie.vlan_id;
2829 u16 old_vid = adapter->mng_vlan_id;
2831 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2832 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2833 adapter->mng_vlan_id = vid;
2836 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2837 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2840 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2844 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2846 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2847 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2850 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2852 struct e1000_hw *hw = &adapter->hw;
2853 u32 manc, manc2h, mdef, i, j;
2855 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2860 /* enable receiving management packets to the host. this will probably
2861 * generate destination unreachable messages from the host OS, but
2862 * the packets will be handled on SMBUS
2864 manc |= E1000_MANC_EN_MNG2HOST;
2865 manc2h = er32(MANC2H);
2867 switch (hw->mac.type) {
2869 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2873 /* Check if IPMI pass-through decision filter already exists;
2876 for (i = 0, j = 0; i < 8; i++) {
2877 mdef = er32(MDEF(i));
2879 /* Ignore filters with anything other than IPMI ports */
2880 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2883 /* Enable this decision filter in MANC2H */
2890 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2893 /* Create new decision filter in an empty filter */
2894 for (i = 0, j = 0; i < 8; i++)
2895 if (er32(MDEF(i)) == 0) {
2896 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2897 E1000_MDEF_PORT_664));
2904 e_warn("Unable to create IPMI pass-through filter\n");
2908 ew32(MANC2H, manc2h);
2913 * e1000_configure_tx - Configure Transmit Unit after Reset
2914 * @adapter: board private structure
2916 * Configure the Tx unit of the MAC after a reset.
2918 static void e1000_configure_tx(struct e1000_adapter *adapter)
2920 struct e1000_hw *hw = &adapter->hw;
2921 struct e1000_ring *tx_ring = adapter->tx_ring;
2923 u32 tdlen, tctl, tarc;
2925 /* Setup the HW Tx Head and Tail descriptor pointers */
2926 tdba = tx_ring->dma;
2927 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2928 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2929 ew32(TDBAH(0), (tdba >> 32));
2930 ew32(TDLEN(0), tdlen);
2933 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2934 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2936 writel(0, tx_ring->head);
2937 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2938 e1000e_update_tdt_wa(tx_ring, 0);
2940 writel(0, tx_ring->tail);
2942 /* Set the Tx Interrupt Delay register */
2943 ew32(TIDV, adapter->tx_int_delay);
2944 /* Tx irq moderation */
2945 ew32(TADV, adapter->tx_abs_int_delay);
2947 if (adapter->flags2 & FLAG2_DMA_BURST) {
2948 u32 txdctl = er32(TXDCTL(0));
2950 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2951 E1000_TXDCTL_WTHRESH);
2952 /* set up some performance related parameters to encourage the
2953 * hardware to use the bus more efficiently in bursts, depends
2954 * on the tx_int_delay to be enabled,
2955 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2956 * hthresh = 1 ==> prefetch when one or more available
2957 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2958 * BEWARE: this seems to work but should be considered first if
2959 * there are Tx hangs or other Tx related bugs
2961 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2962 ew32(TXDCTL(0), txdctl);
2964 /* erratum work around: set txdctl the same for both queues */
2965 ew32(TXDCTL(1), er32(TXDCTL(0)));
2967 /* Program the Transmit Control Register */
2969 tctl &= ~E1000_TCTL_CT;
2970 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2971 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2973 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2974 tarc = er32(TARC(0));
2975 /* set the speed mode bit, we'll clear it if we're not at
2976 * gigabit link later
2978 #define SPEED_MODE_BIT BIT(21)
2979 tarc |= SPEED_MODE_BIT;
2980 ew32(TARC(0), tarc);
2983 /* errata: program both queues to unweighted RR */
2984 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2985 tarc = er32(TARC(0));
2987 ew32(TARC(0), tarc);
2988 tarc = er32(TARC(1));
2990 ew32(TARC(1), tarc);
2993 /* Setup Transmit Descriptor Settings for eop descriptor */
2994 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2996 /* only set IDE if we are delaying interrupts using the timers */
2997 if (adapter->tx_int_delay)
2998 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3000 /* enable Report Status bit */
3001 adapter->txd_cmd |= E1000_TXD_CMD_RS;
3005 hw->mac.ops.config_collision_dist(hw);
3007 /* SPT and CNP Si errata workaround to avoid data corruption */
3008 if (hw->mac.type >= e1000_pch_spt) {
3011 reg_val = er32(IOSFPC);
3012 reg_val |= E1000_RCTL_RDMTS_HEX;
3013 ew32(IOSFPC, reg_val);
3015 reg_val = er32(TARC(0));
3016 reg_val |= E1000_TARC0_CB_MULTIQ_3_REQ;
3017 ew32(TARC(0), reg_val);
3022 * e1000_setup_rctl - configure the receive control registers
3023 * @adapter: Board private structure
3025 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3026 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3027 static void e1000_setup_rctl(struct e1000_adapter *adapter)
3029 struct e1000_hw *hw = &adapter->hw;
3033 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3034 * If jumbo frames not set, program related MAC/PHY registers
3037 if (hw->mac.type >= e1000_pch2lan) {
3040 if (adapter->netdev->mtu > ETH_DATA_LEN)
3041 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3043 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3046 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3049 /* Program MC offset vector base */
3051 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3052 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3053 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3054 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3056 /* Do not Store bad packets */
3057 rctl &= ~E1000_RCTL_SBP;
3059 /* Enable Long Packet receive */
3060 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3061 rctl &= ~E1000_RCTL_LPE;
3063 rctl |= E1000_RCTL_LPE;
3065 /* Some systems expect that the CRC is included in SMBUS traffic. The
3066 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3067 * host memory when this is enabled
3069 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3070 rctl |= E1000_RCTL_SECRC;
3072 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3073 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3076 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3079 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3081 e1e_rphy(hw, 22, &phy_data);
3083 phy_data |= BIT(14);
3084 e1e_wphy(hw, 0x10, 0x2823);
3085 e1e_wphy(hw, 0x11, 0x0003);
3086 e1e_wphy(hw, 22, phy_data);
3089 /* Setup buffer sizes */
3090 rctl &= ~E1000_RCTL_SZ_4096;
3091 rctl |= E1000_RCTL_BSEX;
3092 switch (adapter->rx_buffer_len) {
3095 rctl |= E1000_RCTL_SZ_2048;
3096 rctl &= ~E1000_RCTL_BSEX;
3099 rctl |= E1000_RCTL_SZ_4096;
3102 rctl |= E1000_RCTL_SZ_8192;
3105 rctl |= E1000_RCTL_SZ_16384;
3109 /* Enable Extended Status in all Receive Descriptors */
3110 rfctl = er32(RFCTL);
3111 rfctl |= E1000_RFCTL_EXTEN;
3114 /* 82571 and greater support packet-split where the protocol
3115 * header is placed in skb->data and the packet data is
3116 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3117 * In the case of a non-split, skb->data is linearly filled,
3118 * followed by the page buffers. Therefore, skb->data is
3119 * sized to hold the largest protocol header.
3121 * allocations using alloc_page take too long for regular MTU
3122 * so only enable packet split for jumbo frames
3124 * Using pages when the page size is greater than 16k wastes
3125 * a lot of memory, since we allocate 3 pages at all times
3128 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3129 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3130 adapter->rx_ps_pages = pages;
3132 adapter->rx_ps_pages = 0;
3134 if (adapter->rx_ps_pages) {
3137 /* Enable Packet split descriptors */
3138 rctl |= E1000_RCTL_DTYP_PS;
3140 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3142 switch (adapter->rx_ps_pages) {
3144 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3147 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3150 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3154 ew32(PSRCTL, psrctl);
3157 /* This is useful for sniffing bad packets. */
3158 if (adapter->netdev->features & NETIF_F_RXALL) {
3159 /* UPE and MPE will be handled by normal PROMISC logic
3160 * in e1000e_set_rx_mode
3162 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3163 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3164 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3166 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3167 E1000_RCTL_DPF | /* Allow filtered pause */
3168 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3169 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3170 * and that breaks VLANs.
3175 /* just started the receive unit, no need to restart */
3176 adapter->flags &= ~FLAG_RESTART_NOW;
3180 * e1000_configure_rx - Configure Receive Unit after Reset
3181 * @adapter: board private structure
3183 * Configure the Rx unit of the MAC after a reset.
3185 static void e1000_configure_rx(struct e1000_adapter *adapter)
3187 struct e1000_hw *hw = &adapter->hw;
3188 struct e1000_ring *rx_ring = adapter->rx_ring;
3190 u32 rdlen, rctl, rxcsum, ctrl_ext;
3192 if (adapter->rx_ps_pages) {
3193 /* this is a 32 byte descriptor */
3194 rdlen = rx_ring->count *
3195 sizeof(union e1000_rx_desc_packet_split);
3196 adapter->clean_rx = e1000_clean_rx_irq_ps;
3197 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3198 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3199 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3200 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3201 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3203 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3204 adapter->clean_rx = e1000_clean_rx_irq;
3205 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3208 /* disable receives while setting up the descriptors */
3210 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3211 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3213 usleep_range(10000, 20000);
3215 if (adapter->flags2 & FLAG2_DMA_BURST) {
3216 /* set the writeback threshold (only takes effect if the RDTR
3217 * is set). set GRAN=1 and write back up to 0x4 worth, and
3218 * enable prefetching of 0x20 Rx descriptors
3224 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3225 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3227 /* override the delay timers for enabling bursting, only if
3228 * the value was not set by the user via module options
3230 if (adapter->rx_int_delay == DEFAULT_RDTR)
3231 adapter->rx_int_delay = BURST_RDTR;
3232 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3233 adapter->rx_abs_int_delay = BURST_RADV;
3236 /* set the Receive Delay Timer Register */
3237 ew32(RDTR, adapter->rx_int_delay);
3239 /* irq moderation */
3240 ew32(RADV, adapter->rx_abs_int_delay);
3241 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3242 e1000e_write_itr(adapter, adapter->itr);
3244 ctrl_ext = er32(CTRL_EXT);
3245 /* Auto-Mask interrupts upon ICR access */
3246 ctrl_ext |= E1000_CTRL_EXT_IAME;
3247 ew32(IAM, 0xffffffff);
3248 ew32(CTRL_EXT, ctrl_ext);
3251 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3252 * the Base and Length of the Rx Descriptor Ring
3254 rdba = rx_ring->dma;
3255 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3256 ew32(RDBAH(0), (rdba >> 32));
3257 ew32(RDLEN(0), rdlen);
3260 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3261 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3263 writel(0, rx_ring->head);
3264 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3265 e1000e_update_rdt_wa(rx_ring, 0);
3267 writel(0, rx_ring->tail);
3269 /* Enable Receive Checksum Offload for TCP and UDP */
3270 rxcsum = er32(RXCSUM);
3271 if (adapter->netdev->features & NETIF_F_RXCSUM)
3272 rxcsum |= E1000_RXCSUM_TUOFL;
3274 rxcsum &= ~E1000_RXCSUM_TUOFL;
3275 ew32(RXCSUM, rxcsum);
3277 /* With jumbo frames, excessive C-state transition latencies result
3278 * in dropped transactions.
3280 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3282 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3283 adapter->max_frame_size) * 8 / 1000;
3285 if (adapter->flags & FLAG_IS_ICH) {
3286 u32 rxdctl = er32(RXDCTL(0));
3288 ew32(RXDCTL(0), rxdctl | 0x3);
3291 pm_qos_update_request(&adapter->pm_qos_req, lat);
3293 pm_qos_update_request(&adapter->pm_qos_req,
3294 PM_QOS_DEFAULT_VALUE);
3297 /* Enable Receives */
3302 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3303 * @netdev: network interface device structure
3305 * Writes multicast address list to the MTA hash table.
3306 * Returns: -ENOMEM on failure
3307 * 0 on no addresses written
3308 * X on writing X addresses to MTA
3310 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3312 struct e1000_adapter *adapter = netdev_priv(netdev);
3313 struct e1000_hw *hw = &adapter->hw;
3314 struct netdev_hw_addr *ha;
3318 if (netdev_mc_empty(netdev)) {
3319 /* nothing to program, so clear mc list */
3320 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3324 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3328 /* update_mc_addr_list expects a packed array of only addresses. */
3330 netdev_for_each_mc_addr(ha, netdev)
3331 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3333 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3336 return netdev_mc_count(netdev);
3340 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3341 * @netdev: network interface device structure
3343 * Writes unicast address list to the RAR table.
3344 * Returns: -ENOMEM on failure/insufficient address space
3345 * 0 on no addresses written
3346 * X on writing X addresses to the RAR table
3348 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3350 struct e1000_adapter *adapter = netdev_priv(netdev);
3351 struct e1000_hw *hw = &adapter->hw;
3352 unsigned int rar_entries;
3355 rar_entries = hw->mac.ops.rar_get_count(hw);
3357 /* save a rar entry for our hardware address */
3360 /* save a rar entry for the LAA workaround */
3361 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3364 /* return ENOMEM indicating insufficient memory for addresses */
3365 if (netdev_uc_count(netdev) > rar_entries)
3368 if (!netdev_uc_empty(netdev) && rar_entries) {
3369 struct netdev_hw_addr *ha;
3371 /* write the addresses in reverse order to avoid write
3374 netdev_for_each_uc_addr(ha, netdev) {
3379 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3386 /* zero out the remaining RAR entries not used above */
3387 for (; rar_entries > 0; rar_entries--) {
3388 ew32(RAH(rar_entries), 0);
3389 ew32(RAL(rar_entries), 0);
3397 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3398 * @netdev: network interface device structure
3400 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3401 * address list or the network interface flags are updated. This routine is
3402 * responsible for configuring the hardware for proper unicast, multicast,
3403 * promiscuous mode, and all-multi behavior.
3405 static void e1000e_set_rx_mode(struct net_device *netdev)
3407 struct e1000_adapter *adapter = netdev_priv(netdev);
3408 struct e1000_hw *hw = &adapter->hw;
3411 if (pm_runtime_suspended(netdev->dev.parent))
3414 /* Check for Promiscuous and All Multicast modes */
3417 /* clear the affected bits */
3418 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3420 if (netdev->flags & IFF_PROMISC) {
3421 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3422 /* Do not hardware filter VLANs in promisc mode */
3423 e1000e_vlan_filter_disable(adapter);
3427 if (netdev->flags & IFF_ALLMULTI) {
3428 rctl |= E1000_RCTL_MPE;
3430 /* Write addresses to the MTA, if the attempt fails
3431 * then we should just turn on promiscuous mode so
3432 * that we can at least receive multicast traffic
3434 count = e1000e_write_mc_addr_list(netdev);
3436 rctl |= E1000_RCTL_MPE;
3438 e1000e_vlan_filter_enable(adapter);
3439 /* Write addresses to available RAR registers, if there is not
3440 * sufficient space to store all the addresses then enable
3441 * unicast promiscuous mode
3443 count = e1000e_write_uc_addr_list(netdev);
3445 rctl |= E1000_RCTL_UPE;
3450 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3451 e1000e_vlan_strip_enable(adapter);
3453 e1000e_vlan_strip_disable(adapter);
3456 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3458 struct e1000_hw *hw = &adapter->hw;
3463 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3464 for (i = 0; i < 10; i++)
3465 ew32(RSSRK(i), rss_key[i]);
3467 /* Direct all traffic to queue 0 */
3468 for (i = 0; i < 32; i++)
3471 /* Disable raw packet checksumming so that RSS hash is placed in
3472 * descriptor on writeback.
3474 rxcsum = er32(RXCSUM);
3475 rxcsum |= E1000_RXCSUM_PCSD;
3477 ew32(RXCSUM, rxcsum);
3479 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3480 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3481 E1000_MRQC_RSS_FIELD_IPV6 |
3482 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3483 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3489 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3490 * @adapter: board private structure
3491 * @timinca: pointer to returned time increment attributes
3493 * Get attributes for incrementing the System Time Register SYSTIML/H at
3494 * the default base frequency, and set the cyclecounter shift value.
3496 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3498 struct e1000_hw *hw = &adapter->hw;
3499 u32 incvalue, incperiod, shift;
3501 /* Make sure clock is enabled on I217/I218/I219 before checking
3504 if ((hw->mac.type >= e1000_pch_lpt) &&
3505 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3506 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3507 u32 fextnvm7 = er32(FEXTNVM7);
3509 if (!(fextnvm7 & BIT(0))) {
3510 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3515 switch (hw->mac.type) {
3517 /* Stable 96MHz frequency */
3518 incperiod = INCPERIOD_96MHZ;
3519 incvalue = INCVALUE_96MHZ;
3520 shift = INCVALUE_SHIFT_96MHZ;
3521 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3524 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3525 /* Stable 96MHz frequency */
3526 incperiod = INCPERIOD_96MHZ;
3527 incvalue = INCVALUE_96MHZ;
3528 shift = INCVALUE_SHIFT_96MHZ;
3529 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3531 /* Stable 25MHz frequency */
3532 incperiod = INCPERIOD_25MHZ;
3533 incvalue = INCVALUE_25MHZ;
3534 shift = INCVALUE_SHIFT_25MHZ;
3535 adapter->cc.shift = shift;
3539 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3540 /* Stable 24MHz frequency */
3541 incperiod = INCPERIOD_24MHZ;
3542 incvalue = INCVALUE_24MHZ;
3543 shift = INCVALUE_SHIFT_24MHZ;
3544 adapter->cc.shift = shift;
3549 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3550 /* Stable 24MHz frequency */
3551 incperiod = INCPERIOD_24MHZ;
3552 incvalue = INCVALUE_24MHZ;
3553 shift = INCVALUE_SHIFT_24MHZ;
3554 adapter->cc.shift = shift;
3556 /* Stable 38400KHz frequency */
3557 incperiod = INCPERIOD_38400KHZ;
3558 incvalue = INCVALUE_38400KHZ;
3559 shift = INCVALUE_SHIFT_38400KHZ;
3560 adapter->cc.shift = shift;
3565 /* Stable 25MHz frequency */
3566 incperiod = INCPERIOD_25MHZ;
3567 incvalue = INCVALUE_25MHZ;
3568 shift = INCVALUE_SHIFT_25MHZ;
3569 adapter->cc.shift = shift;
3575 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3576 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3582 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3583 * @adapter: board private structure
3585 * Outgoing time stamping can be enabled and disabled. Play nice and
3586 * disable it when requested, although it shouldn't cause any overhead
3587 * when no packet needs it. At most one packet in the queue may be
3588 * marked for time stamping, otherwise it would be impossible to tell
3589 * for sure to which packet the hardware time stamp belongs.
3591 * Incoming time stamping has to be configured via the hardware filters.
3592 * Not all combinations are supported, in particular event type has to be
3593 * specified. Matching the kind of event packet is not supported, with the
3594 * exception of "all V2 events regardless of level 2 or 4".
3596 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3597 struct hwtstamp_config *config)
3599 struct e1000_hw *hw = &adapter->hw;
3600 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3601 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3608 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3611 /* flags reserved for future extensions - must be zero */
3615 switch (config->tx_type) {
3616 case HWTSTAMP_TX_OFF:
3619 case HWTSTAMP_TX_ON:
3625 switch (config->rx_filter) {
3626 case HWTSTAMP_FILTER_NONE:
3629 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3630 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3631 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3634 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3635 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3636 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3639 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3640 /* Also time stamps V2 L2 Path Delay Request/Response */
3641 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3642 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3645 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3646 /* Also time stamps V2 L2 Path Delay Request/Response. */
3647 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3648 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3651 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3652 /* Hardware cannot filter just V2 L4 Sync messages;
3653 * fall-through to V2 (both L2 and L4) Sync.
3655 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3656 /* Also time stamps V2 Path Delay Request/Response. */
3657 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3658 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3662 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3663 /* Hardware cannot filter just V2 L4 Delay Request messages;
3664 * fall-through to V2 (both L2 and L4) Delay Request.
3666 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3667 /* Also time stamps V2 Path Delay Request/Response. */
3668 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3669 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3673 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3674 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3675 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3676 * fall-through to all V2 (both L2 and L4) Events.
3678 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3679 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3680 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3684 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3685 /* For V1, the hardware can only filter Sync messages or
3686 * Delay Request messages but not both so fall-through to
3687 * time stamp all packets.
3689 case HWTSTAMP_FILTER_NTP_ALL:
3690 case HWTSTAMP_FILTER_ALL:
3693 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3694 config->rx_filter = HWTSTAMP_FILTER_ALL;
3700 adapter->hwtstamp_config = *config;
3702 /* enable/disable Tx h/w time stamping */
3703 regval = er32(TSYNCTXCTL);
3704 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3705 regval |= tsync_tx_ctl;
3706 ew32(TSYNCTXCTL, regval);
3707 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3708 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3709 e_err("Timesync Tx Control register not set as expected\n");
3713 /* enable/disable Rx h/w time stamping */
3714 regval = er32(TSYNCRXCTL);
3715 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3716 regval |= tsync_rx_ctl;
3717 ew32(TSYNCRXCTL, regval);
3718 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3719 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3720 (regval & (E1000_TSYNCRXCTL_ENABLED |
3721 E1000_TSYNCRXCTL_TYPE_MASK))) {
3722 e_err("Timesync Rx Control register not set as expected\n");
3726 /* L2: define ethertype filter for time stamped packets */
3728 rxmtrl |= ETH_P_1588;
3730 /* define which PTP packets get time stamped */
3731 ew32(RXMTRL, rxmtrl);
3733 /* Filter by destination port */
3735 rxudp = PTP_EV_PORT;
3736 cpu_to_be16s(&rxudp);
3742 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3750 * e1000_configure - configure the hardware for Rx and Tx
3751 * @adapter: private board structure
3753 static void e1000_configure(struct e1000_adapter *adapter)
3755 struct e1000_ring *rx_ring = adapter->rx_ring;
3757 e1000e_set_rx_mode(adapter->netdev);
3759 e1000_restore_vlan(adapter);
3760 e1000_init_manageability_pt(adapter);
3762 e1000_configure_tx(adapter);
3764 if (adapter->netdev->features & NETIF_F_RXHASH)
3765 e1000e_setup_rss_hash(adapter);
3766 e1000_setup_rctl(adapter);
3767 e1000_configure_rx(adapter);
3768 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3772 * e1000e_power_up_phy - restore link in case the phy was powered down
3773 * @adapter: address of board private structure
3775 * The phy may be powered down to save power and turn off link when the
3776 * driver is unloaded and wake on lan is not enabled (among others)
3777 * *** this routine MUST be followed by a call to e1000e_reset ***
3779 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3781 if (adapter->hw.phy.ops.power_up)
3782 adapter->hw.phy.ops.power_up(&adapter->hw);
3784 adapter->hw.mac.ops.setup_link(&adapter->hw);
3788 * e1000_power_down_phy - Power down the PHY
3790 * Power down the PHY so no link is implied when interface is down.
3791 * The PHY cannot be powered down if management or WoL is active.
3793 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3795 if (adapter->hw.phy.ops.power_down)
3796 adapter->hw.phy.ops.power_down(&adapter->hw);
3800 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3802 * We want to clear all pending descriptors from the TX ring.
3803 * zeroing happens when the HW reads the regs. We assign the ring itself as
3804 * the data of the next descriptor. We don't care about the data we are about
3807 static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3809 struct e1000_hw *hw = &adapter->hw;
3810 struct e1000_ring *tx_ring = adapter->tx_ring;
3811 struct e1000_tx_desc *tx_desc = NULL;
3812 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3816 ew32(TCTL, tctl | E1000_TCTL_EN);
3818 BUG_ON(tdt != tx_ring->next_to_use);
3819 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3820 tx_desc->buffer_addr = tx_ring->dma;
3822 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3823 tx_desc->upper.data = 0;
3824 /* flush descriptors to memory before notifying the HW */
3826 tx_ring->next_to_use++;
3827 if (tx_ring->next_to_use == tx_ring->count)
3828 tx_ring->next_to_use = 0;
3829 ew32(TDT(0), tx_ring->next_to_use);
3831 usleep_range(200, 250);
3835 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3837 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3839 static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3842 struct e1000_hw *hw = &adapter->hw;
3845 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3847 usleep_range(100, 150);
3849 rxdctl = er32(RXDCTL(0));
3850 /* zero the lower 14 bits (prefetch and host thresholds) */
3851 rxdctl &= 0xffffc000;
3853 /* update thresholds: prefetch threshold to 31, host threshold to 1
3854 * and make sure the granularity is "descriptors" and not "cache lines"
3856 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3858 ew32(RXDCTL(0), rxdctl);
3859 /* momentarily enable the RX ring for the changes to take effect */
3860 ew32(RCTL, rctl | E1000_RCTL_EN);
3862 usleep_range(100, 150);
3863 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3867 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3869 * In i219, the descriptor rings must be emptied before resetting the HW
3870 * or before changing the device state to D3 during runtime (runtime PM).
3872 * Failure to do this will cause the HW to enter a unit hang state which can
3873 * only be released by PCI reset on the device
3877 static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3880 u32 fext_nvm11, tdlen;
3881 struct e1000_hw *hw = &adapter->hw;
3883 /* First, disable MULR fix in FEXTNVM11 */
3884 fext_nvm11 = er32(FEXTNVM11);
3885 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3886 ew32(FEXTNVM11, fext_nvm11);
3887 /* do nothing if we're not in faulty state, or if the queue is empty */
3888 tdlen = er32(TDLEN(0));
3889 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3891 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3893 e1000_flush_tx_ring(adapter);
3894 /* recheck, maybe the fault is caused by the rx ring */
3895 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3897 if (hang_state & FLUSH_DESC_REQUIRED)
3898 e1000_flush_rx_ring(adapter);
3902 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3903 * @adapter: board private structure
3905 * When the MAC is reset, all hardware bits for timesync will be reset to the
3906 * default values. This function will restore the settings last in place.
3907 * Since the clock SYSTIME registers are reset, we will simply restore the
3908 * cyclecounter to the kernel real clock time.
3910 static void e1000e_systim_reset(struct e1000_adapter *adapter)
3912 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3913 struct e1000_hw *hw = &adapter->hw;
3914 unsigned long flags;
3918 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3921 if (info->adjfreq) {
3922 /* restore the previous ptp frequency delta */
3923 ret_val = info->adjfreq(info, adapter->ptp_delta);
3925 /* set the default base frequency if no adjustment possible */
3926 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3928 ew32(TIMINCA, timinca);
3932 dev_warn(&adapter->pdev->dev,
3933 "Failed to restore TIMINCA clock rate delta: %d\n",
3938 /* reset the systim ns time counter */
3939 spin_lock_irqsave(&adapter->systim_lock, flags);
3940 timecounter_init(&adapter->tc, &adapter->cc,
3941 ktime_to_ns(ktime_get_real()));
3942 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3944 /* restore the previous hwtstamp configuration settings */
3945 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3949 * e1000e_reset - bring the hardware into a known good state
3951 * This function boots the hardware and enables some settings that
3952 * require a configuration cycle of the hardware - those cannot be
3953 * set/changed during runtime. After reset the device needs to be
3954 * properly configured for Rx, Tx etc.
3956 void e1000e_reset(struct e1000_adapter *adapter)
3958 struct e1000_mac_info *mac = &adapter->hw.mac;
3959 struct e1000_fc_info *fc = &adapter->hw.fc;
3960 struct e1000_hw *hw = &adapter->hw;
3961 u32 tx_space, min_tx_space, min_rx_space;
3962 u32 pba = adapter->pba;
3965 /* reset Packet Buffer Allocation to default */
3968 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3969 /* To maintain wire speed transmits, the Tx FIFO should be
3970 * large enough to accommodate two full transmit packets,
3971 * rounded up to the next 1KB and expressed in KB. Likewise,
3972 * the Rx FIFO should be large enough to accommodate at least
3973 * one full receive packet and is similarly rounded up and
3977 /* upper 16 bits has Tx packet buffer allocation size in KB */
3978 tx_space = pba >> 16;
3979 /* lower 16 bits has Rx packet buffer allocation size in KB */
3981 /* the Tx fifo also stores 16 bytes of information about the Tx
3982 * but don't include ethernet FCS because hardware appends it
3984 min_tx_space = (adapter->max_frame_size +
3985 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3986 min_tx_space = ALIGN(min_tx_space, 1024);
3987 min_tx_space >>= 10;
3988 /* software strips receive CRC, so leave room for it */
3989 min_rx_space = adapter->max_frame_size;
3990 min_rx_space = ALIGN(min_rx_space, 1024);
3991 min_rx_space >>= 10;
3993 /* If current Tx allocation is less than the min Tx FIFO size,
3994 * and the min Tx FIFO size is less than the current Rx FIFO
3995 * allocation, take space away from current Rx allocation
3997 if ((tx_space < min_tx_space) &&
3998 ((min_tx_space - tx_space) < pba)) {
3999 pba -= min_tx_space - tx_space;
4001 /* if short on Rx space, Rx wins and must trump Tx
4004 if (pba < min_rx_space)
4011 /* flow control settings
4013 * The high water mark must be low enough to fit one full frame
4014 * (or the size used for early receive) above it in the Rx FIFO.
4015 * Set it to the lower of:
4016 * - 90% of the Rx FIFO size, and
4017 * - the full Rx FIFO size minus one full frame
4019 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4020 fc->pause_time = 0xFFFF;
4022 fc->pause_time = E1000_FC_PAUSE_TIME;
4023 fc->send_xon = true;
4024 fc->current_mode = fc->requested_mode;
4026 switch (hw->mac.type) {
4028 case e1000_ich10lan:
4029 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4032 fc->high_water = 0x2800;
4033 fc->low_water = fc->high_water - 8;
4038 hwm = min(((pba << 10) * 9 / 10),
4039 ((pba << 10) - adapter->max_frame_size));
4041 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4042 fc->low_water = fc->high_water - 8;
4045 /* Workaround PCH LOM adapter hangs with certain network
4046 * loads. If hangs persist, try disabling Tx flow control.
4048 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4049 fc->high_water = 0x3500;
4050 fc->low_water = 0x1500;
4052 fc->high_water = 0x5000;
4053 fc->low_water = 0x3000;
4055 fc->refresh_time = 0x1000;
4061 fc->refresh_time = 0x0400;
4063 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4064 fc->high_water = 0x05C20;
4065 fc->low_water = 0x05048;
4066 fc->pause_time = 0x0650;
4072 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4073 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4077 /* Alignment of Tx data is on an arbitrary byte boundary with the
4078 * maximum size per Tx descriptor limited only to the transmit
4079 * allocation of the packet buffer minus 96 bytes with an upper
4080 * limit of 24KB due to receive synchronization limitations.
4082 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4085 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4086 * fit in receive buffer.
4088 if (adapter->itr_setting & 0x3) {
4089 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4090 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4091 dev_info(&adapter->pdev->dev,
4092 "Interrupt Throttle Rate off\n");
4093 adapter->flags2 |= FLAG2_DISABLE_AIM;
4094 e1000e_write_itr(adapter, 0);
4096 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4097 dev_info(&adapter->pdev->dev,
4098 "Interrupt Throttle Rate on\n");
4099 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4100 adapter->itr = 20000;
4101 e1000e_write_itr(adapter, adapter->itr);
4105 if (hw->mac.type >= e1000_pch_spt)
4106 e1000_flush_desc_rings(adapter);
4107 /* Allow time for pending master requests to run */
4108 mac->ops.reset_hw(hw);
4110 /* For parts with AMT enabled, let the firmware know
4111 * that the network interface is in control
4113 if (adapter->flags & FLAG_HAS_AMT)
4114 e1000e_get_hw_control(adapter);
4118 if (mac->ops.init_hw(hw))
4119 e_err("Hardware Error\n");
4121 e1000_update_mng_vlan(adapter);
4123 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4124 ew32(VET, ETH_P_8021Q);
4126 e1000e_reset_adaptive(hw);
4128 /* restore systim and hwtstamp settings */
4129 e1000e_systim_reset(adapter);
4131 /* Set EEE advertisement as appropriate */
4132 if (adapter->flags2 & FLAG2_HAS_EEE) {
4136 switch (hw->phy.type) {
4137 case e1000_phy_82579:
4138 adv_addr = I82579_EEE_ADVERTISEMENT;
4140 case e1000_phy_i217:
4141 adv_addr = I217_EEE_ADVERTISEMENT;
4144 dev_err(&adapter->pdev->dev,
4145 "Invalid PHY type setting EEE advertisement\n");
4149 ret_val = hw->phy.ops.acquire(hw);
4151 dev_err(&adapter->pdev->dev,
4152 "EEE advertisement - unable to acquire PHY\n");
4156 e1000_write_emi_reg_locked(hw, adv_addr,
4157 hw->dev_spec.ich8lan.eee_disable ?
4158 0 : adapter->eee_advert);
4160 hw->phy.ops.release(hw);
4163 if (!netif_running(adapter->netdev) &&
4164 !test_bit(__E1000_TESTING, &adapter->state))
4165 e1000_power_down_phy(adapter);
4167 e1000_get_phy_info(hw);
4169 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4170 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4172 /* speed up time to link by disabling smart power down, ignore
4173 * the return value of this function because there is nothing
4174 * different we would do if it failed
4176 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4177 phy_data &= ~IGP02E1000_PM_SPD;
4178 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4180 if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4183 /* Fextnvm7 @ 0xe4[2] = 1 */
4184 reg = er32(FEXTNVM7);
4185 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4186 ew32(FEXTNVM7, reg);
4187 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4188 reg = er32(FEXTNVM9);
4189 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4190 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4191 ew32(FEXTNVM9, reg);
4197 * e1000e_trigger_lsc - trigger an LSC interrupt
4200 * Fire a link status change interrupt to start the watchdog.
4202 static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4204 struct e1000_hw *hw = &adapter->hw;
4206 if (adapter->msix_entries)
4207 ew32(ICS, E1000_ICS_OTHER);
4209 ew32(ICS, E1000_ICS_LSC);
4212 void e1000e_up(struct e1000_adapter *adapter)
4214 /* hardware has been reset, we need to reload some things */
4215 e1000_configure(adapter);
4217 clear_bit(__E1000_DOWN, &adapter->state);
4219 if (adapter->msix_entries)
4220 e1000_configure_msix(adapter);
4221 e1000_irq_enable(adapter);
4223 netif_start_queue(adapter->netdev);
4225 e1000e_trigger_lsc(adapter);
4228 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4230 struct e1000_hw *hw = &adapter->hw;
4232 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4235 /* flush pending descriptor writebacks to memory */
4236 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4237 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4239 /* execute the writes immediately */
4242 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4243 * write is successful
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 */
4252 static void e1000e_update_stats(struct e1000_adapter *adapter);
4255 * e1000e_down - quiesce the device and optionally reset the hardware
4256 * @adapter: board private structure
4257 * @reset: boolean flag to reset the hardware or not
4259 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4261 struct net_device *netdev = adapter->netdev;
4262 struct e1000_hw *hw = &adapter->hw;
4265 /* signal that we're down so the interrupt handler does not
4266 * reschedule our watchdog timer
4268 set_bit(__E1000_DOWN, &adapter->state);
4270 netif_carrier_off(netdev);
4272 /* disable receives in the hardware */
4274 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4275 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4276 /* flush and sleep below */
4278 netif_stop_queue(netdev);
4280 /* disable transmits in the hardware */
4282 tctl &= ~E1000_TCTL_EN;
4285 /* flush both disables and wait for them to finish */
4287 usleep_range(10000, 20000);
4289 e1000_irq_disable(adapter);
4291 napi_synchronize(&adapter->napi);
4293 del_timer_sync(&adapter->watchdog_timer);
4294 del_timer_sync(&adapter->phy_info_timer);
4296 spin_lock(&adapter->stats64_lock);
4297 e1000e_update_stats(adapter);
4298 spin_unlock(&adapter->stats64_lock);
4300 e1000e_flush_descriptors(adapter);
4302 adapter->link_speed = 0;
4303 adapter->link_duplex = 0;
4305 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4306 if ((hw->mac.type >= e1000_pch2lan) &&
4307 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4308 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4309 e_dbg("failed to disable jumbo frame workaround mode\n");
4311 if (!pci_channel_offline(adapter->pdev)) {
4313 e1000e_reset(adapter);
4314 else if (hw->mac.type >= e1000_pch_spt)
4315 e1000_flush_desc_rings(adapter);
4317 e1000_clean_tx_ring(adapter->tx_ring);
4318 e1000_clean_rx_ring(adapter->rx_ring);
4321 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4324 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4325 usleep_range(1000, 2000);
4326 e1000e_down(adapter, true);
4328 clear_bit(__E1000_RESETTING, &adapter->state);
4332 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4333 * @hw: pointer to the HW structure
4334 * @systim: time value read, sanitized and returned
4336 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4337 * check to see that the time is incrementing at a reasonable
4338 * rate and is a multiple of incvalue.
4340 static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim)
4342 u64 time_delta, rem, temp;
4347 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4348 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4349 /* latch SYSTIMH on read of SYSTIML */
4350 systim_next = (u64)er32(SYSTIML);
4351 systim_next |= (u64)er32(SYSTIMH) << 32;
4353 time_delta = systim_next - systim;
4355 /* VMWare users have seen incvalue of zero, don't div / 0 */
4356 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4358 systim = systim_next;
4360 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4368 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4369 * @cc: cyclecounter structure
4371 static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4373 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4375 struct e1000_hw *hw = &adapter->hw;
4376 u32 systimel, systimeh;
4378 /* SYSTIMH latching upon SYSTIML read does not work well.
4379 * This means that if SYSTIML overflows after we read it but before
4380 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4381 * will experience a huge non linear increment in the systime value
4382 * to fix that we test for overflow and if true, we re-read systime.
4384 systimel = er32(SYSTIML);
4385 systimeh = er32(SYSTIMH);
4386 /* Is systimel is so large that overflow is possible? */
4387 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4388 u32 systimel_2 = er32(SYSTIML);
4389 if (systimel > systimel_2) {
4390 /* There was an overflow, read again SYSTIMH, and use
4393 systimeh = er32(SYSTIMH);
4394 systimel = systimel_2;
4397 systim = (u64)systimel;
4398 systim |= (u64)systimeh << 32;
4400 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4401 systim = e1000e_sanitize_systim(hw, systim);
4407 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4408 * @adapter: board private structure to initialize
4410 * e1000_sw_init initializes the Adapter private data structure.
4411 * Fields are initialized based on PCI device information and
4412 * OS network device settings (MTU size).
4414 static int e1000_sw_init(struct e1000_adapter *adapter)
4416 struct net_device *netdev = adapter->netdev;
4418 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4419 adapter->rx_ps_bsize0 = 128;
4420 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4421 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4422 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4423 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4425 spin_lock_init(&adapter->stats64_lock);
4427 e1000e_set_interrupt_capability(adapter);
4429 if (e1000_alloc_queues(adapter))
4432 /* Setup hardware time stamping cyclecounter */
4433 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4434 adapter->cc.read = e1000e_cyclecounter_read;
4435 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4436 adapter->cc.mult = 1;
4437 /* cc.shift set in e1000e_get_base_tininca() */
4439 spin_lock_init(&adapter->systim_lock);
4440 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4443 /* Explicitly disable IRQ since the NIC can be in any state. */
4444 e1000_irq_disable(adapter);
4446 set_bit(__E1000_DOWN, &adapter->state);
4451 * e1000_intr_msi_test - Interrupt Handler
4452 * @irq: interrupt number
4453 * @data: pointer to a network interface device structure
4455 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4457 struct net_device *netdev = data;
4458 struct e1000_adapter *adapter = netdev_priv(netdev);
4459 struct e1000_hw *hw = &adapter->hw;
4460 u32 icr = er32(ICR);
4462 e_dbg("icr is %08X\n", icr);
4463 if (icr & E1000_ICR_RXSEQ) {
4464 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4465 /* Force memory writes to complete before acknowledging the
4466 * interrupt is handled.
4475 * e1000_test_msi_interrupt - Returns 0 for successful test
4476 * @adapter: board private struct
4478 * code flow taken from tg3.c
4480 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4482 struct net_device *netdev = adapter->netdev;
4483 struct e1000_hw *hw = &adapter->hw;
4486 /* poll_enable hasn't been called yet, so don't need disable */
4487 /* clear any pending events */
4490 /* free the real vector and request a test handler */
4491 e1000_free_irq(adapter);
4492 e1000e_reset_interrupt_capability(adapter);
4494 /* Assume that the test fails, if it succeeds then the test
4495 * MSI irq handler will unset this flag
4497 adapter->flags |= FLAG_MSI_TEST_FAILED;
4499 err = pci_enable_msi(adapter->pdev);
4501 goto msi_test_failed;
4503 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4504 netdev->name, netdev);
4506 pci_disable_msi(adapter->pdev);
4507 goto msi_test_failed;
4510 /* Force memory writes to complete before enabling and firing an
4515 e1000_irq_enable(adapter);
4517 /* fire an unusual interrupt on the test handler */
4518 ew32(ICS, E1000_ICS_RXSEQ);
4522 e1000_irq_disable(adapter);
4524 rmb(); /* read flags after interrupt has been fired */
4526 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4527 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4528 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4530 e_dbg("MSI interrupt test succeeded!\n");
4533 free_irq(adapter->pdev->irq, netdev);
4534 pci_disable_msi(adapter->pdev);
4537 e1000e_set_interrupt_capability(adapter);
4538 return e1000_request_irq(adapter);
4542 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4543 * @adapter: board private struct
4545 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4547 static int e1000_test_msi(struct e1000_adapter *adapter)
4552 if (!(adapter->flags & FLAG_MSI_ENABLED))
4555 /* disable SERR in case the MSI write causes a master abort */
4556 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4557 if (pci_cmd & PCI_COMMAND_SERR)
4558 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4559 pci_cmd & ~PCI_COMMAND_SERR);
4561 err = e1000_test_msi_interrupt(adapter);
4563 /* re-enable SERR */
4564 if (pci_cmd & PCI_COMMAND_SERR) {
4565 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4566 pci_cmd |= PCI_COMMAND_SERR;
4567 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4574 * e1000e_open - Called when a network interface is made active
4575 * @netdev: network interface device structure
4577 * Returns 0 on success, negative value on failure
4579 * The open entry point is called when a network interface is made
4580 * active by the system (IFF_UP). At this point all resources needed
4581 * for transmit and receive operations are allocated, the interrupt
4582 * handler is registered with the OS, the watchdog timer is started,
4583 * and the stack is notified that the interface is ready.
4585 int e1000e_open(struct net_device *netdev)
4587 struct e1000_adapter *adapter = netdev_priv(netdev);
4588 struct e1000_hw *hw = &adapter->hw;
4589 struct pci_dev *pdev = adapter->pdev;
4592 /* disallow open during test */
4593 if (test_bit(__E1000_TESTING, &adapter->state))
4596 pm_runtime_get_sync(&pdev->dev);
4598 netif_carrier_off(netdev);
4600 /* allocate transmit descriptors */
4601 err = e1000e_setup_tx_resources(adapter->tx_ring);
4605 /* allocate receive descriptors */
4606 err = e1000e_setup_rx_resources(adapter->rx_ring);
4610 /* If AMT is enabled, let the firmware know that the network
4611 * interface is now open and reset the part to a known state.
4613 if (adapter->flags & FLAG_HAS_AMT) {
4614 e1000e_get_hw_control(adapter);
4615 e1000e_reset(adapter);
4618 e1000e_power_up_phy(adapter);
4620 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4621 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4622 e1000_update_mng_vlan(adapter);
4624 /* DMA latency requirement to workaround jumbo issue */
4625 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4626 PM_QOS_DEFAULT_VALUE);
4628 /* before we allocate an interrupt, we must be ready to handle it.
4629 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4630 * as soon as we call pci_request_irq, so we have to setup our
4631 * clean_rx handler before we do so.
4633 e1000_configure(adapter);
4635 err = e1000_request_irq(adapter);
4639 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4640 * ignore e1000e MSI messages, which means we need to test our MSI
4643 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4644 err = e1000_test_msi(adapter);
4646 e_err("Interrupt allocation failed\n");
4651 /* From here on the code is the same as e1000e_up() */
4652 clear_bit(__E1000_DOWN, &adapter->state);
4654 napi_enable(&adapter->napi);
4656 e1000_irq_enable(adapter);
4658 adapter->tx_hang_recheck = false;
4659 netif_start_queue(netdev);
4661 hw->mac.get_link_status = true;
4662 pm_runtime_put(&pdev->dev);
4664 e1000e_trigger_lsc(adapter);
4669 pm_qos_remove_request(&adapter->pm_qos_req);
4670 e1000e_release_hw_control(adapter);
4671 e1000_power_down_phy(adapter);
4672 e1000e_free_rx_resources(adapter->rx_ring);
4674 e1000e_free_tx_resources(adapter->tx_ring);
4676 e1000e_reset(adapter);
4677 pm_runtime_put_sync(&pdev->dev);
4683 * e1000e_close - Disables a network interface
4684 * @netdev: network interface device structure
4686 * Returns 0, this is not allowed to fail
4688 * The close entry point is called when an interface is de-activated
4689 * by the OS. The hardware is still under the drivers control, but
4690 * needs to be disabled. A global MAC reset is issued to stop the
4691 * hardware, and all transmit and receive resources are freed.
4693 int e1000e_close(struct net_device *netdev)
4695 struct e1000_adapter *adapter = netdev_priv(netdev);
4696 struct pci_dev *pdev = adapter->pdev;
4697 int count = E1000_CHECK_RESET_COUNT;
4699 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4700 usleep_range(10000, 20000);
4702 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4704 pm_runtime_get_sync(&pdev->dev);
4706 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4707 e1000e_down(adapter, true);
4708 e1000_free_irq(adapter);
4710 /* Link status message must follow this format */
4711 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4714 napi_disable(&adapter->napi);
4716 e1000e_free_tx_resources(adapter->tx_ring);
4717 e1000e_free_rx_resources(adapter->rx_ring);
4719 /* kill manageability vlan ID if supported, but not if a vlan with
4720 * the same ID is registered on the host OS (let 8021q kill it)
4722 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4723 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4724 adapter->mng_vlan_id);
4726 /* If AMT is enabled, let the firmware know that the network
4727 * interface is now closed
4729 if ((adapter->flags & FLAG_HAS_AMT) &&
4730 !test_bit(__E1000_TESTING, &adapter->state))
4731 e1000e_release_hw_control(adapter);
4733 pm_qos_remove_request(&adapter->pm_qos_req);
4735 pm_runtime_put_sync(&pdev->dev);
4741 * e1000_set_mac - Change the Ethernet Address of the NIC
4742 * @netdev: network interface device structure
4743 * @p: pointer to an address structure
4745 * Returns 0 on success, negative on failure
4747 static int e1000_set_mac(struct net_device *netdev, void *p)
4749 struct e1000_adapter *adapter = netdev_priv(netdev);
4750 struct e1000_hw *hw = &adapter->hw;
4751 struct sockaddr *addr = p;
4753 if (!is_valid_ether_addr(addr->sa_data))
4754 return -EADDRNOTAVAIL;
4756 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4757 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4759 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4761 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4762 /* activate the work around */
4763 e1000e_set_laa_state_82571(&adapter->hw, 1);
4765 /* Hold a copy of the LAA in RAR[14] This is done so that
4766 * between the time RAR[0] gets clobbered and the time it
4767 * gets fixed (in e1000_watchdog), the actual LAA is in one
4768 * of the RARs and no incoming packets directed to this port
4769 * are dropped. Eventually the LAA will be in RAR[0] and
4772 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4773 adapter->hw.mac.rar_entry_count - 1);
4780 * e1000e_update_phy_task - work thread to update phy
4781 * @work: pointer to our work struct
4783 * this worker thread exists because we must acquire a
4784 * semaphore to read the phy, which we could msleep while
4785 * waiting for it, and we can't msleep in a timer.
4787 static void e1000e_update_phy_task(struct work_struct *work)
4789 struct e1000_adapter *adapter = container_of(work,
4790 struct e1000_adapter,
4792 struct e1000_hw *hw = &adapter->hw;
4794 if (test_bit(__E1000_DOWN, &adapter->state))
4797 e1000_get_phy_info(hw);
4799 /* Enable EEE on 82579 after link up */
4800 if (hw->phy.type >= e1000_phy_82579)
4801 e1000_set_eee_pchlan(hw);
4805 * e1000_update_phy_info - timre call-back to update PHY info
4806 * @data: pointer to adapter cast into an unsigned long
4808 * Need to wait a few seconds after link up to get diagnostic information from
4811 static void e1000_update_phy_info(unsigned long data)
4813 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
4815 if (test_bit(__E1000_DOWN, &adapter->state))
4818 schedule_work(&adapter->update_phy_task);
4822 * e1000e_update_phy_stats - Update the PHY statistics counters
4823 * @adapter: board private structure
4825 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4827 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4829 struct e1000_hw *hw = &adapter->hw;
4833 ret_val = hw->phy.ops.acquire(hw);
4837 /* A page set is expensive so check if already on desired page.
4838 * If not, set to the page with the PHY status registers.
4841 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4845 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4846 ret_val = hw->phy.ops.set_page(hw,
4847 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4852 /* Single Collision Count */
4853 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4854 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4856 adapter->stats.scc += phy_data;
4858 /* Excessive Collision Count */
4859 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4860 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4862 adapter->stats.ecol += phy_data;
4864 /* Multiple Collision Count */
4865 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4866 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4868 adapter->stats.mcc += phy_data;
4870 /* Late Collision Count */
4871 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4872 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4874 adapter->stats.latecol += phy_data;
4876 /* Collision Count - also used for adaptive IFS */
4877 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4878 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4880 hw->mac.collision_delta = phy_data;
4883 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4884 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4886 adapter->stats.dc += phy_data;
4888 /* Transmit with no CRS */
4889 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4890 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4892 adapter->stats.tncrs += phy_data;
4895 hw->phy.ops.release(hw);
4899 * e1000e_update_stats - Update the board statistics counters
4900 * @adapter: board private structure
4902 static void e1000e_update_stats(struct e1000_adapter *adapter)
4904 struct net_device *netdev = adapter->netdev;
4905 struct e1000_hw *hw = &adapter->hw;
4906 struct pci_dev *pdev = adapter->pdev;
4908 /* Prevent stats update while adapter is being reset, or if the pci
4909 * connection is down.
4911 if (adapter->link_speed == 0)
4913 if (pci_channel_offline(pdev))
4916 adapter->stats.crcerrs += er32(CRCERRS);
4917 adapter->stats.gprc += er32(GPRC);
4918 adapter->stats.gorc += er32(GORCL);
4919 er32(GORCH); /* Clear gorc */
4920 adapter->stats.bprc += er32(BPRC);
4921 adapter->stats.mprc += er32(MPRC);
4922 adapter->stats.roc += er32(ROC);
4924 adapter->stats.mpc += er32(MPC);
4926 /* Half-duplex statistics */
4927 if (adapter->link_duplex == HALF_DUPLEX) {
4928 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4929 e1000e_update_phy_stats(adapter);
4931 adapter->stats.scc += er32(SCC);
4932 adapter->stats.ecol += er32(ECOL);
4933 adapter->stats.mcc += er32(MCC);
4934 adapter->stats.latecol += er32(LATECOL);
4935 adapter->stats.dc += er32(DC);
4937 hw->mac.collision_delta = er32(COLC);
4939 if ((hw->mac.type != e1000_82574) &&
4940 (hw->mac.type != e1000_82583))
4941 adapter->stats.tncrs += er32(TNCRS);
4943 adapter->stats.colc += hw->mac.collision_delta;
4946 adapter->stats.xonrxc += er32(XONRXC);
4947 adapter->stats.xontxc += er32(XONTXC);
4948 adapter->stats.xoffrxc += er32(XOFFRXC);
4949 adapter->stats.xofftxc += er32(XOFFTXC);
4950 adapter->stats.gptc += er32(GPTC);
4951 adapter->stats.gotc += er32(GOTCL);
4952 er32(GOTCH); /* Clear gotc */
4953 adapter->stats.rnbc += er32(RNBC);
4954 adapter->stats.ruc += er32(RUC);
4956 adapter->stats.mptc += er32(MPTC);
4957 adapter->stats.bptc += er32(BPTC);
4959 /* used for adaptive IFS */
4961 hw->mac.tx_packet_delta = er32(TPT);
4962 adapter->stats.tpt += hw->mac.tx_packet_delta;
4964 adapter->stats.algnerrc += er32(ALGNERRC);
4965 adapter->stats.rxerrc += er32(RXERRC);
4966 adapter->stats.cexterr += er32(CEXTERR);
4967 adapter->stats.tsctc += er32(TSCTC);
4968 adapter->stats.tsctfc += er32(TSCTFC);
4970 /* Fill out the OS statistics structure */
4971 netdev->stats.multicast = adapter->stats.mprc;
4972 netdev->stats.collisions = adapter->stats.colc;
4976 /* RLEC on some newer hardware can be incorrect so build
4977 * our own version based on RUC and ROC
4979 netdev->stats.rx_errors = adapter->stats.rxerrc +
4980 adapter->stats.crcerrs + adapter->stats.algnerrc +
4981 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4982 netdev->stats.rx_length_errors = adapter->stats.ruc +
4984 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4985 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4986 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4989 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4990 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4991 netdev->stats.tx_window_errors = adapter->stats.latecol;
4992 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4994 /* Tx Dropped needs to be maintained elsewhere */
4996 /* Management Stats */
4997 adapter->stats.mgptc += er32(MGTPTC);
4998 adapter->stats.mgprc += er32(MGTPRC);
4999 adapter->stats.mgpdc += er32(MGTPDC);
5001 /* Correctable ECC Errors */
5002 if (hw->mac.type >= e1000_pch_lpt) {
5003 u32 pbeccsts = er32(PBECCSTS);
5005 adapter->corr_errors +=
5006 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
5007 adapter->uncorr_errors +=
5008 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
5009 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
5014 * e1000_phy_read_status - Update the PHY register status snapshot
5015 * @adapter: board private structure
5017 static void e1000_phy_read_status(struct e1000_adapter *adapter)
5019 struct e1000_hw *hw = &adapter->hw;
5020 struct e1000_phy_regs *phy = &adapter->phy_regs;
5022 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5023 (er32(STATUS) & E1000_STATUS_LU) &&
5024 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5027 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5028 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5029 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5030 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5031 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5032 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5033 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5034 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5036 e_warn("Error reading PHY register\n");
5038 /* Do not read PHY registers if link is not up
5039 * Set values to typical power-on defaults
5041 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5042 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5043 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5045 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5046 ADVERTISE_ALL | ADVERTISE_CSMA);
5048 phy->expansion = EXPANSION_ENABLENPAGE;
5049 phy->ctrl1000 = ADVERTISE_1000FULL;
5051 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5055 static void e1000_print_link_info(struct e1000_adapter *adapter)
5057 struct e1000_hw *hw = &adapter->hw;
5058 u32 ctrl = er32(CTRL);
5060 /* Link status message must follow this format for user tools */
5061 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5062 adapter->netdev->name, adapter->link_speed,
5063 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5064 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5065 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5066 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5069 static bool e1000e_has_link(struct e1000_adapter *adapter)
5071 struct e1000_hw *hw = &adapter->hw;
5072 bool link_active = false;
5075 /* get_link_status is set on LSC (link status) interrupt or
5076 * Rx sequence error interrupt. get_link_status will stay
5077 * false until the check_for_link establishes link
5078 * for copper adapters ONLY
5080 switch (hw->phy.media_type) {
5081 case e1000_media_type_copper:
5082 if (hw->mac.get_link_status) {
5083 ret_val = hw->mac.ops.check_for_link(hw);
5084 link_active = !hw->mac.get_link_status;
5089 case e1000_media_type_fiber:
5090 ret_val = hw->mac.ops.check_for_link(hw);
5091 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5093 case e1000_media_type_internal_serdes:
5094 ret_val = hw->mac.ops.check_for_link(hw);
5095 link_active = adapter->hw.mac.serdes_has_link;
5098 case e1000_media_type_unknown:
5102 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5103 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5104 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5105 e_info("Gigabit has been disabled, downgrading speed\n");
5111 static void e1000e_enable_receives(struct e1000_adapter *adapter)
5113 /* make sure the receive unit is started */
5114 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5115 (adapter->flags & FLAG_RESTART_NOW)) {
5116 struct e1000_hw *hw = &adapter->hw;
5117 u32 rctl = er32(RCTL);
5119 ew32(RCTL, rctl | E1000_RCTL_EN);
5120 adapter->flags &= ~FLAG_RESTART_NOW;
5124 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5126 struct e1000_hw *hw = &adapter->hw;
5128 /* With 82574 controllers, PHY needs to be checked periodically
5129 * for hung state and reset, if two calls return true
5131 if (e1000_check_phy_82574(hw))
5132 adapter->phy_hang_count++;
5134 adapter->phy_hang_count = 0;
5136 if (adapter->phy_hang_count > 1) {
5137 adapter->phy_hang_count = 0;
5138 e_dbg("PHY appears hung - resetting\n");
5139 schedule_work(&adapter->reset_task);
5144 * e1000_watchdog - Timer Call-back
5145 * @data: pointer to adapter cast into an unsigned long
5147 static void e1000_watchdog(unsigned long data)
5149 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
5151 /* Do the rest outside of interrupt context */
5152 schedule_work(&adapter->watchdog_task);
5154 /* TODO: make this use queue_delayed_work() */
5157 static void e1000_watchdog_task(struct work_struct *work)
5159 struct e1000_adapter *adapter = container_of(work,
5160 struct e1000_adapter,
5162 struct net_device *netdev = adapter->netdev;
5163 struct e1000_mac_info *mac = &adapter->hw.mac;
5164 struct e1000_phy_info *phy = &adapter->hw.phy;
5165 struct e1000_ring *tx_ring = adapter->tx_ring;
5166 struct e1000_hw *hw = &adapter->hw;
5169 if (test_bit(__E1000_DOWN, &adapter->state))
5172 link = e1000e_has_link(adapter);
5173 if ((netif_carrier_ok(netdev)) && link) {
5174 /* Cancel scheduled suspend requests. */
5175 pm_runtime_resume(netdev->dev.parent);
5177 e1000e_enable_receives(adapter);
5181 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5182 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5183 e1000_update_mng_vlan(adapter);
5186 if (!netif_carrier_ok(netdev)) {
5189 /* Cancel scheduled suspend requests. */
5190 pm_runtime_resume(netdev->dev.parent);
5192 /* update snapshot of PHY registers on LSC */
5193 e1000_phy_read_status(adapter);
5194 mac->ops.get_link_up_info(&adapter->hw,
5195 &adapter->link_speed,
5196 &adapter->link_duplex);
5197 e1000_print_link_info(adapter);
5199 /* check if SmartSpeed worked */
5200 e1000e_check_downshift(hw);
5201 if (phy->speed_downgraded)
5203 "Link Speed was downgraded by SmartSpeed\n");
5205 /* On supported PHYs, check for duplex mismatch only
5206 * if link has autonegotiated at 10/100 half
5208 if ((hw->phy.type == e1000_phy_igp_3 ||
5209 hw->phy.type == e1000_phy_bm) &&
5211 (adapter->link_speed == SPEED_10 ||
5212 adapter->link_speed == SPEED_100) &&
5213 (adapter->link_duplex == HALF_DUPLEX)) {
5216 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5218 if (!(autoneg_exp & EXPANSION_NWAY))
5219 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5222 /* adjust timeout factor according to speed/duplex */
5223 adapter->tx_timeout_factor = 1;
5224 switch (adapter->link_speed) {
5227 adapter->tx_timeout_factor = 16;
5231 adapter->tx_timeout_factor = 10;
5235 /* workaround: re-program speed mode bit after
5238 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5242 tarc0 = er32(TARC(0));
5243 tarc0 &= ~SPEED_MODE_BIT;
5244 ew32(TARC(0), tarc0);
5247 /* disable TSO for pcie and 10/100 speeds, to avoid
5248 * some hardware issues
5250 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5251 switch (adapter->link_speed) {
5254 e_info("10/100 speed: disabling TSO\n");
5255 netdev->features &= ~NETIF_F_TSO;
5256 netdev->features &= ~NETIF_F_TSO6;
5259 netdev->features |= NETIF_F_TSO;
5260 netdev->features |= NETIF_F_TSO6;
5268 /* enable transmits in the hardware, need to do this
5269 * after setting TARC(0)
5272 tctl |= E1000_TCTL_EN;
5275 /* Perform any post-link-up configuration before
5276 * reporting link up.
5278 if (phy->ops.cfg_on_link_up)
5279 phy->ops.cfg_on_link_up(hw);
5281 netif_carrier_on(netdev);
5283 if (!test_bit(__E1000_DOWN, &adapter->state))
5284 mod_timer(&adapter->phy_info_timer,
5285 round_jiffies(jiffies + 2 * HZ));
5288 if (netif_carrier_ok(netdev)) {
5289 adapter->link_speed = 0;
5290 adapter->link_duplex = 0;
5291 /* Link status message must follow this format */
5292 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5293 netif_carrier_off(netdev);
5294 if (!test_bit(__E1000_DOWN, &adapter->state))
5295 mod_timer(&adapter->phy_info_timer,
5296 round_jiffies(jiffies + 2 * HZ));
5298 /* 8000ES2LAN requires a Rx packet buffer work-around
5299 * on link down event; reset the controller to flush
5300 * the Rx packet buffer.
5302 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5303 adapter->flags |= FLAG_RESTART_NOW;
5305 pm_schedule_suspend(netdev->dev.parent,
5311 spin_lock(&adapter->stats64_lock);
5312 e1000e_update_stats(adapter);
5314 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5315 adapter->tpt_old = adapter->stats.tpt;
5316 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5317 adapter->colc_old = adapter->stats.colc;
5319 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5320 adapter->gorc_old = adapter->stats.gorc;
5321 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5322 adapter->gotc_old = adapter->stats.gotc;
5323 spin_unlock(&adapter->stats64_lock);
5325 /* If the link is lost the controller stops DMA, but
5326 * if there is queued Tx work it cannot be done. So
5327 * reset the controller to flush the Tx packet buffers.
5329 if (!netif_carrier_ok(netdev) &&
5330 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5331 adapter->flags |= FLAG_RESTART_NOW;
5333 /* If reset is necessary, do it outside of interrupt context. */
5334 if (adapter->flags & FLAG_RESTART_NOW) {
5335 schedule_work(&adapter->reset_task);
5336 /* return immediately since reset is imminent */
5340 e1000e_update_adaptive(&adapter->hw);
5342 /* Simple mode for Interrupt Throttle Rate (ITR) */
5343 if (adapter->itr_setting == 4) {
5344 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5345 * Total asymmetrical Tx or Rx gets ITR=8000;
5346 * everyone else is between 2000-8000.
5348 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5349 u32 dif = (adapter->gotc > adapter->gorc ?
5350 adapter->gotc - adapter->gorc :
5351 adapter->gorc - adapter->gotc) / 10000;
5352 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5354 e1000e_write_itr(adapter, itr);
5357 /* Cause software interrupt to ensure Rx ring is cleaned */
5358 if (adapter->msix_entries)
5359 ew32(ICS, adapter->rx_ring->ims_val);
5361 ew32(ICS, E1000_ICS_RXDMT0);
5363 /* flush pending descriptors to memory before detecting Tx hang */
5364 e1000e_flush_descriptors(adapter);
5366 /* Force detection of hung controller every watchdog period */
5367 adapter->detect_tx_hung = true;
5369 /* With 82571 controllers, LAA may be overwritten due to controller
5370 * reset from the other port. Set the appropriate LAA in RAR[0]
5372 if (e1000e_get_laa_state_82571(hw))
5373 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5375 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5376 e1000e_check_82574_phy_workaround(adapter);
5378 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5379 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5380 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5381 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5383 adapter->rx_hwtstamp_cleared++;
5385 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5389 /* Reset the timer */
5390 if (!test_bit(__E1000_DOWN, &adapter->state))
5391 mod_timer(&adapter->watchdog_timer,
5392 round_jiffies(jiffies + 2 * HZ));
5395 #define E1000_TX_FLAGS_CSUM 0x00000001
5396 #define E1000_TX_FLAGS_VLAN 0x00000002
5397 #define E1000_TX_FLAGS_TSO 0x00000004
5398 #define E1000_TX_FLAGS_IPV4 0x00000008
5399 #define E1000_TX_FLAGS_NO_FCS 0x00000010
5400 #define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5401 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5402 #define E1000_TX_FLAGS_VLAN_SHIFT 16
5404 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5407 struct e1000_context_desc *context_desc;
5408 struct e1000_buffer *buffer_info;
5412 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5415 if (!skb_is_gso(skb))
5418 err = skb_cow_head(skb, 0);
5422 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5423 mss = skb_shinfo(skb)->gso_size;
5424 if (protocol == htons(ETH_P_IP)) {
5425 struct iphdr *iph = ip_hdr(skb);
5428 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5430 cmd_length = E1000_TXD_CMD_IP;
5431 ipcse = skb_transport_offset(skb) - 1;
5432 } else if (skb_is_gso_v6(skb)) {
5433 ipv6_hdr(skb)->payload_len = 0;
5434 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5435 &ipv6_hdr(skb)->daddr,
5439 ipcss = skb_network_offset(skb);
5440 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5441 tucss = skb_transport_offset(skb);
5442 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5444 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5445 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5447 i = tx_ring->next_to_use;
5448 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5449 buffer_info = &tx_ring->buffer_info[i];
5451 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5452 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5453 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5454 context_desc->upper_setup.tcp_fields.tucss = tucss;
5455 context_desc->upper_setup.tcp_fields.tucso = tucso;
5456 context_desc->upper_setup.tcp_fields.tucse = 0;
5457 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5458 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5459 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5461 buffer_info->time_stamp = jiffies;
5462 buffer_info->next_to_watch = i;
5465 if (i == tx_ring->count)
5467 tx_ring->next_to_use = i;
5472 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5475 struct e1000_adapter *adapter = tx_ring->adapter;
5476 struct e1000_context_desc *context_desc;
5477 struct e1000_buffer *buffer_info;
5480 u32 cmd_len = E1000_TXD_CMD_DEXT;
5482 if (skb->ip_summed != CHECKSUM_PARTIAL)
5486 case cpu_to_be16(ETH_P_IP):
5487 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5488 cmd_len |= E1000_TXD_CMD_TCP;
5490 case cpu_to_be16(ETH_P_IPV6):
5491 /* XXX not handling all IPV6 headers */
5492 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5493 cmd_len |= E1000_TXD_CMD_TCP;
5496 if (unlikely(net_ratelimit()))
5497 e_warn("checksum_partial proto=%x!\n",
5498 be16_to_cpu(protocol));
5502 css = skb_checksum_start_offset(skb);
5504 i = tx_ring->next_to_use;
5505 buffer_info = &tx_ring->buffer_info[i];
5506 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5508 context_desc->lower_setup.ip_config = 0;
5509 context_desc->upper_setup.tcp_fields.tucss = css;
5510 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5511 context_desc->upper_setup.tcp_fields.tucse = 0;
5512 context_desc->tcp_seg_setup.data = 0;
5513 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5515 buffer_info->time_stamp = jiffies;
5516 buffer_info->next_to_watch = i;
5519 if (i == tx_ring->count)
5521 tx_ring->next_to_use = i;
5526 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5527 unsigned int first, unsigned int max_per_txd,
5528 unsigned int nr_frags)
5530 struct e1000_adapter *adapter = tx_ring->adapter;
5531 struct pci_dev *pdev = adapter->pdev;
5532 struct e1000_buffer *buffer_info;
5533 unsigned int len = skb_headlen(skb);
5534 unsigned int offset = 0, size, count = 0, i;
5535 unsigned int f, bytecount, segs;
5537 i = tx_ring->next_to_use;
5540 buffer_info = &tx_ring->buffer_info[i];
5541 size = min(len, max_per_txd);
5543 buffer_info->length = size;
5544 buffer_info->time_stamp = jiffies;
5545 buffer_info->next_to_watch = i;
5546 buffer_info->dma = dma_map_single(&pdev->dev,
5548 size, DMA_TO_DEVICE);
5549 buffer_info->mapped_as_page = false;
5550 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5559 if (i == tx_ring->count)
5564 for (f = 0; f < nr_frags; f++) {
5565 const struct skb_frag_struct *frag;
5567 frag = &skb_shinfo(skb)->frags[f];
5568 len = skb_frag_size(frag);
5573 if (i == tx_ring->count)
5576 buffer_info = &tx_ring->buffer_info[i];
5577 size = min(len, max_per_txd);
5579 buffer_info->length = size;
5580 buffer_info->time_stamp = jiffies;
5581 buffer_info->next_to_watch = i;
5582 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5585 buffer_info->mapped_as_page = true;
5586 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5595 segs = skb_shinfo(skb)->gso_segs ? : 1;
5596 /* multiply data chunks by size of headers */
5597 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5599 tx_ring->buffer_info[i].skb = skb;
5600 tx_ring->buffer_info[i].segs = segs;
5601 tx_ring->buffer_info[i].bytecount = bytecount;
5602 tx_ring->buffer_info[first].next_to_watch = i;
5607 dev_err(&pdev->dev, "Tx DMA map failed\n");
5608 buffer_info->dma = 0;
5614 i += tx_ring->count;
5616 buffer_info = &tx_ring->buffer_info[i];
5617 e1000_put_txbuf(tx_ring, buffer_info);
5623 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5625 struct e1000_adapter *adapter = tx_ring->adapter;
5626 struct e1000_tx_desc *tx_desc = NULL;
5627 struct e1000_buffer *buffer_info;
5628 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5631 if (tx_flags & E1000_TX_FLAGS_TSO) {
5632 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5634 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5636 if (tx_flags & E1000_TX_FLAGS_IPV4)
5637 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5640 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5641 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5642 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5645 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5646 txd_lower |= E1000_TXD_CMD_VLE;
5647 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5650 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5651 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5653 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5654 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5655 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5658 i = tx_ring->next_to_use;
5661 buffer_info = &tx_ring->buffer_info[i];
5662 tx_desc = E1000_TX_DESC(*tx_ring, i);
5663 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5664 tx_desc->lower.data = cpu_to_le32(txd_lower |
5665 buffer_info->length);
5666 tx_desc->upper.data = cpu_to_le32(txd_upper);
5669 if (i == tx_ring->count)
5671 } while (--count > 0);
5673 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5675 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5676 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5677 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5679 /* Force memory writes to complete before letting h/w
5680 * know there are new descriptors to fetch. (Only
5681 * applicable for weak-ordered memory model archs,
5686 tx_ring->next_to_use = i;
5689 #define MINIMUM_DHCP_PACKET_SIZE 282
5690 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5691 struct sk_buff *skb)
5693 struct e1000_hw *hw = &adapter->hw;
5696 if (skb_vlan_tag_present(skb) &&
5697 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5698 (adapter->hw.mng_cookie.status &
5699 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5702 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5705 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5709 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5712 if (ip->protocol != IPPROTO_UDP)
5715 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5716 if (ntohs(udp->dest) != 67)
5719 offset = (u8 *)udp + 8 - skb->data;
5720 length = skb->len - offset;
5721 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5727 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5729 struct e1000_adapter *adapter = tx_ring->adapter;
5731 netif_stop_queue(adapter->netdev);
5732 /* Herbert's original patch had:
5733 * smp_mb__after_netif_stop_queue();
5734 * but since that doesn't exist yet, just open code it.
5738 /* We need to check again in a case another CPU has just
5739 * made room available.
5741 if (e1000_desc_unused(tx_ring) < size)
5745 netif_start_queue(adapter->netdev);
5746 ++adapter->restart_queue;
5750 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5752 BUG_ON(size > tx_ring->count);
5754 if (e1000_desc_unused(tx_ring) >= size)
5756 return __e1000_maybe_stop_tx(tx_ring, size);
5759 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5760 struct net_device *netdev)
5762 struct e1000_adapter *adapter = netdev_priv(netdev);
5763 struct e1000_ring *tx_ring = adapter->tx_ring;
5765 unsigned int tx_flags = 0;
5766 unsigned int len = skb_headlen(skb);
5767 unsigned int nr_frags;
5772 __be16 protocol = vlan_get_protocol(skb);
5774 if (test_bit(__E1000_DOWN, &adapter->state)) {
5775 dev_kfree_skb_any(skb);
5776 return NETDEV_TX_OK;
5779 if (skb->len <= 0) {
5780 dev_kfree_skb_any(skb);
5781 return NETDEV_TX_OK;
5784 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5785 * pad skb in order to meet this minimum size requirement
5787 if (skb_put_padto(skb, 17))
5788 return NETDEV_TX_OK;
5790 mss = skb_shinfo(skb)->gso_size;
5794 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5795 * points to just header, pull a few bytes of payload from
5796 * frags into skb->data
5798 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5799 /* we do this workaround for ES2LAN, but it is un-necessary,
5800 * avoiding it could save a lot of cycles
5802 if (skb->data_len && (hdr_len == len)) {
5803 unsigned int pull_size;
5805 pull_size = min_t(unsigned int, 4, skb->data_len);
5806 if (!__pskb_pull_tail(skb, pull_size)) {
5807 e_err("__pskb_pull_tail failed.\n");
5808 dev_kfree_skb_any(skb);
5809 return NETDEV_TX_OK;
5811 len = skb_headlen(skb);
5815 /* reserve a descriptor for the offload context */
5816 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5820 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5822 nr_frags = skb_shinfo(skb)->nr_frags;
5823 for (f = 0; f < nr_frags; f++)
5824 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5825 adapter->tx_fifo_limit);
5827 if (adapter->hw.mac.tx_pkt_filtering)
5828 e1000_transfer_dhcp_info(adapter, skb);
5830 /* need: count + 2 desc gap to keep tail from touching
5831 * head, otherwise try next time
5833 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5834 return NETDEV_TX_BUSY;
5836 if (skb_vlan_tag_present(skb)) {
5837 tx_flags |= E1000_TX_FLAGS_VLAN;
5838 tx_flags |= (skb_vlan_tag_get(skb) <<
5839 E1000_TX_FLAGS_VLAN_SHIFT);
5842 first = tx_ring->next_to_use;
5844 tso = e1000_tso(tx_ring, skb, protocol);
5846 dev_kfree_skb_any(skb);
5847 return NETDEV_TX_OK;
5851 tx_flags |= E1000_TX_FLAGS_TSO;
5852 else if (e1000_tx_csum(tx_ring, skb, protocol))
5853 tx_flags |= E1000_TX_FLAGS_CSUM;
5855 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5856 * 82571 hardware supports TSO capabilities for IPv6 as well...
5857 * no longer assume, we must.
5859 if (protocol == htons(ETH_P_IP))
5860 tx_flags |= E1000_TX_FLAGS_IPV4;
5862 if (unlikely(skb->no_fcs))
5863 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5865 /* if count is 0 then mapping error has occurred */
5866 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5869 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5870 (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5871 if (!adapter->tx_hwtstamp_skb) {
5872 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5873 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5874 adapter->tx_hwtstamp_skb = skb_get(skb);
5875 adapter->tx_hwtstamp_start = jiffies;
5876 schedule_work(&adapter->tx_hwtstamp_work);
5878 adapter->tx_hwtstamp_skipped++;
5882 skb_tx_timestamp(skb);
5884 netdev_sent_queue(netdev, skb->len);
5885 e1000_tx_queue(tx_ring, tx_flags, count);
5886 /* Make sure there is space in the ring for the next send. */
5887 e1000_maybe_stop_tx(tx_ring,
5889 DIV_ROUND_UP(PAGE_SIZE,
5890 adapter->tx_fifo_limit) + 2));
5892 if (!skb->xmit_more ||
5893 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5894 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5895 e1000e_update_tdt_wa(tx_ring,
5896 tx_ring->next_to_use);
5898 writel(tx_ring->next_to_use, tx_ring->tail);
5900 /* we need this if more than one processor can write
5901 * to our tail at a time, it synchronizes IO on
5907 dev_kfree_skb_any(skb);
5908 tx_ring->buffer_info[first].time_stamp = 0;
5909 tx_ring->next_to_use = first;
5912 return NETDEV_TX_OK;
5916 * e1000_tx_timeout - Respond to a Tx Hang
5917 * @netdev: network interface device structure
5919 static void e1000_tx_timeout(struct net_device *netdev)
5921 struct e1000_adapter *adapter = netdev_priv(netdev);
5923 /* Do the reset outside of interrupt context */
5924 adapter->tx_timeout_count++;
5925 schedule_work(&adapter->reset_task);
5928 static void e1000_reset_task(struct work_struct *work)
5930 struct e1000_adapter *adapter;
5931 adapter = container_of(work, struct e1000_adapter, reset_task);
5933 /* don't run the task if already down */
5934 if (test_bit(__E1000_DOWN, &adapter->state))
5937 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5938 e1000e_dump(adapter);
5939 e_err("Reset adapter unexpectedly\n");
5941 e1000e_reinit_locked(adapter);
5945 * e1000_get_stats64 - Get System Network Statistics
5946 * @netdev: network interface device structure
5947 * @stats: rtnl_link_stats64 pointer
5949 * Returns the address of the device statistics structure.
5951 void e1000e_get_stats64(struct net_device *netdev,
5952 struct rtnl_link_stats64 *stats)
5954 struct e1000_adapter *adapter = netdev_priv(netdev);
5956 spin_lock(&adapter->stats64_lock);
5957 e1000e_update_stats(adapter);
5958 /* Fill out the OS statistics structure */
5959 stats->rx_bytes = adapter->stats.gorc;
5960 stats->rx_packets = adapter->stats.gprc;
5961 stats->tx_bytes = adapter->stats.gotc;
5962 stats->tx_packets = adapter->stats.gptc;
5963 stats->multicast = adapter->stats.mprc;
5964 stats->collisions = adapter->stats.colc;
5968 /* RLEC on some newer hardware can be incorrect so build
5969 * our own version based on RUC and ROC
5971 stats->rx_errors = adapter->stats.rxerrc +
5972 adapter->stats.crcerrs + adapter->stats.algnerrc +
5973 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5974 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5975 stats->rx_crc_errors = adapter->stats.crcerrs;
5976 stats->rx_frame_errors = adapter->stats.algnerrc;
5977 stats->rx_missed_errors = adapter->stats.mpc;
5980 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5981 stats->tx_aborted_errors = adapter->stats.ecol;
5982 stats->tx_window_errors = adapter->stats.latecol;
5983 stats->tx_carrier_errors = adapter->stats.tncrs;
5985 /* Tx Dropped needs to be maintained elsewhere */
5987 spin_unlock(&adapter->stats64_lock);
5991 * e1000_change_mtu - Change the Maximum Transfer Unit
5992 * @netdev: network interface device structure
5993 * @new_mtu: new value for maximum frame size
5995 * Returns 0 on success, negative on failure
5997 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5999 struct e1000_adapter *adapter = netdev_priv(netdev);
6000 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
6002 /* Jumbo frame support */
6003 if ((new_mtu > ETH_DATA_LEN) &&
6004 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
6005 e_err("Jumbo Frames not supported.\n");
6009 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6010 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
6011 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6012 (new_mtu > ETH_DATA_LEN)) {
6013 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6017 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6018 usleep_range(1000, 2000);
6019 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6020 adapter->max_frame_size = max_frame;
6021 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6022 netdev->mtu = new_mtu;
6024 pm_runtime_get_sync(netdev->dev.parent);
6026 if (netif_running(netdev))
6027 e1000e_down(adapter, true);
6029 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6030 * means we reserve 2 more, this pushes us to allocate from the next
6032 * i.e. RXBUFFER_2048 --> size-4096 slab
6033 * However with the new *_jumbo_rx* routines, jumbo receives will use
6037 if (max_frame <= 2048)
6038 adapter->rx_buffer_len = 2048;
6040 adapter->rx_buffer_len = 4096;
6042 /* adjust allocation if LPE protects us, and we aren't using SBP */
6043 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6044 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6046 if (netif_running(netdev))
6049 e1000e_reset(adapter);
6051 pm_runtime_put_sync(netdev->dev.parent);
6053 clear_bit(__E1000_RESETTING, &adapter->state);
6058 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6061 struct e1000_adapter *adapter = netdev_priv(netdev);
6062 struct mii_ioctl_data *data = if_mii(ifr);
6064 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6069 data->phy_id = adapter->hw.phy.addr;
6072 e1000_phy_read_status(adapter);
6074 switch (data->reg_num & 0x1F) {
6076 data->val_out = adapter->phy_regs.bmcr;
6079 data->val_out = adapter->phy_regs.bmsr;
6082 data->val_out = (adapter->hw.phy.id >> 16);
6085 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6088 data->val_out = adapter->phy_regs.advertise;
6091 data->val_out = adapter->phy_regs.lpa;
6094 data->val_out = adapter->phy_regs.expansion;
6097 data->val_out = adapter->phy_regs.ctrl1000;
6100 data->val_out = adapter->phy_regs.stat1000;
6103 data->val_out = adapter->phy_regs.estatus;
6117 * e1000e_hwtstamp_ioctl - control hardware time stamping
6118 * @netdev: network interface device structure
6119 * @ifreq: interface request
6121 * Outgoing time stamping can be enabled and disabled. Play nice and
6122 * disable it when requested, although it shouldn't cause any overhead
6123 * when no packet needs it. At most one packet in the queue may be
6124 * marked for time stamping, otherwise it would be impossible to tell
6125 * for sure to which packet the hardware time stamp belongs.
6127 * Incoming time stamping has to be configured via the hardware filters.
6128 * Not all combinations are supported, in particular event type has to be
6129 * specified. Matching the kind of event packet is not supported, with the
6130 * exception of "all V2 events regardless of level 2 or 4".
6132 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6134 struct e1000_adapter *adapter = netdev_priv(netdev);
6135 struct hwtstamp_config config;
6138 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6141 ret_val = e1000e_config_hwtstamp(adapter, &config);
6145 switch (config.rx_filter) {
6146 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6147 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6148 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6149 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6150 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6151 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6152 /* With V2 type filters which specify a Sync or Delay Request,
6153 * Path Delay Request/Response messages are also time stamped
6154 * by hardware so notify the caller the requested packets plus
6155 * some others are time stamped.
6157 config.rx_filter = HWTSTAMP_FILTER_SOME;
6163 return copy_to_user(ifr->ifr_data, &config,
6164 sizeof(config)) ? -EFAULT : 0;
6167 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6169 struct e1000_adapter *adapter = netdev_priv(netdev);
6171 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6172 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6175 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6181 return e1000_mii_ioctl(netdev, ifr, cmd);
6183 return e1000e_hwtstamp_set(netdev, ifr);
6185 return e1000e_hwtstamp_get(netdev, ifr);
6191 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6193 struct e1000_hw *hw = &adapter->hw;
6194 u32 i, mac_reg, wuc;
6195 u16 phy_reg, wuc_enable;
6198 /* copy MAC RARs to PHY RARs */
6199 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6201 retval = hw->phy.ops.acquire(hw);
6203 e_err("Could not acquire PHY\n");
6207 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6208 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6212 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6213 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6214 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6215 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6216 (u16)(mac_reg & 0xFFFF));
6217 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6218 (u16)((mac_reg >> 16) & 0xFFFF));
6221 /* configure PHY Rx Control register */
6222 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6223 mac_reg = er32(RCTL);
6224 if (mac_reg & E1000_RCTL_UPE)
6225 phy_reg |= BM_RCTL_UPE;
6226 if (mac_reg & E1000_RCTL_MPE)
6227 phy_reg |= BM_RCTL_MPE;
6228 phy_reg &= ~(BM_RCTL_MO_MASK);
6229 if (mac_reg & E1000_RCTL_MO_3)
6230 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6231 << BM_RCTL_MO_SHIFT);
6232 if (mac_reg & E1000_RCTL_BAM)
6233 phy_reg |= BM_RCTL_BAM;
6234 if (mac_reg & E1000_RCTL_PMCF)
6235 phy_reg |= BM_RCTL_PMCF;
6236 mac_reg = er32(CTRL);
6237 if (mac_reg & E1000_CTRL_RFCE)
6238 phy_reg |= BM_RCTL_RFCE;
6239 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6241 wuc = E1000_WUC_PME_EN;
6242 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6243 wuc |= E1000_WUC_APME;
6245 /* enable PHY wakeup in MAC register */
6247 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6248 E1000_WUC_PME_STATUS | wuc));
6250 /* configure and enable PHY wakeup in PHY registers */
6251 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6252 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6254 /* activate PHY wakeup */
6255 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6256 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6258 e_err("Could not set PHY Host Wakeup bit\n");
6260 hw->phy.ops.release(hw);
6265 static void e1000e_flush_lpic(struct pci_dev *pdev)
6267 struct net_device *netdev = pci_get_drvdata(pdev);
6268 struct e1000_adapter *adapter = netdev_priv(netdev);
6269 struct e1000_hw *hw = &adapter->hw;
6272 pm_runtime_get_sync(netdev->dev.parent);
6274 ret_val = hw->phy.ops.acquire(hw);
6278 pr_info("EEE TX LPI TIMER: %08X\n",
6279 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6281 hw->phy.ops.release(hw);
6284 pm_runtime_put_sync(netdev->dev.parent);
6287 static int e1000e_pm_freeze(struct device *dev)
6289 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6290 struct e1000_adapter *adapter = netdev_priv(netdev);
6292 netif_device_detach(netdev);
6294 if (netif_running(netdev)) {
6295 int count = E1000_CHECK_RESET_COUNT;
6297 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6298 usleep_range(10000, 20000);
6300 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6302 /* Quiesce the device without resetting the hardware */
6303 e1000e_down(adapter, false);
6304 e1000_free_irq(adapter);
6306 e1000e_reset_interrupt_capability(adapter);
6308 /* Allow time for pending master requests to run */
6309 e1000e_disable_pcie_master(&adapter->hw);
6314 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6316 struct net_device *netdev = pci_get_drvdata(pdev);
6317 struct e1000_adapter *adapter = netdev_priv(netdev);
6318 struct e1000_hw *hw = &adapter->hw;
6319 u32 ctrl, ctrl_ext, rctl, status;
6320 /* Runtime suspend should only enable wakeup for link changes */
6321 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6324 status = er32(STATUS);
6325 if (status & E1000_STATUS_LU)
6326 wufc &= ~E1000_WUFC_LNKC;
6329 e1000_setup_rctl(adapter);
6330 e1000e_set_rx_mode(netdev);
6332 /* turn on all-multi mode if wake on multicast is enabled */
6333 if (wufc & E1000_WUFC_MC) {
6335 rctl |= E1000_RCTL_MPE;
6340 ctrl |= E1000_CTRL_ADVD3WUC;
6341 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6342 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6345 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6346 adapter->hw.phy.media_type ==
6347 e1000_media_type_internal_serdes) {
6348 /* keep the laser running in D3 */
6349 ctrl_ext = er32(CTRL_EXT);
6350 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6351 ew32(CTRL_EXT, ctrl_ext);
6355 e1000e_power_up_phy(adapter);
6357 if (adapter->flags & FLAG_IS_ICH)
6358 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6360 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6361 /* enable wakeup by the PHY */
6362 retval = e1000_init_phy_wakeup(adapter, wufc);
6366 /* enable wakeup by the MAC */
6368 ew32(WUC, E1000_WUC_PME_EN);
6374 e1000_power_down_phy(adapter);
6377 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6378 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6379 } else if (hw->mac.type >= e1000_pch_lpt) {
6380 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6381 /* ULP does not support wake from unicast, multicast
6384 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6390 /* Ensure that the appropriate bits are set in LPI_CTRL
6393 if ((hw->phy.type >= e1000_phy_i217) &&
6394 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6397 retval = hw->phy.ops.acquire(hw);
6399 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6402 if (adapter->eee_advert &
6403 hw->dev_spec.ich8lan.eee_lp_ability &
6404 I82579_EEE_100_SUPPORTED)
6405 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6406 if (adapter->eee_advert &
6407 hw->dev_spec.ich8lan.eee_lp_ability &
6408 I82579_EEE_1000_SUPPORTED)
6409 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6411 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6415 hw->phy.ops.release(hw);
6418 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6419 * would have already happened in close and is redundant.
6421 e1000e_release_hw_control(adapter);
6423 pci_clear_master(pdev);
6425 /* The pci-e switch on some quad port adapters will report a
6426 * correctable error when the MAC transitions from D0 to D3. To
6427 * prevent this we need to mask off the correctable errors on the
6428 * downstream port of the pci-e switch.
6430 * We don't have the associated upstream bridge while assigning
6431 * the PCI device into guest. For example, the KVM on power is
6434 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6435 struct pci_dev *us_dev = pdev->bus->self;
6441 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6442 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6443 (devctl & ~PCI_EXP_DEVCTL_CERE));
6445 pci_save_state(pdev);
6446 pci_prepare_to_sleep(pdev);
6448 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6455 * __e1000e_disable_aspm - Disable ASPM states
6456 * @pdev: pointer to PCI device struct
6457 * @state: bit-mask of ASPM states to disable
6458 * @locked: indication if this context holds pci_bus_sem locked.
6460 * Some devices *must* have certain ASPM states disabled per hardware errata.
6462 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6464 struct pci_dev *parent = pdev->bus->self;
6465 u16 aspm_dis_mask = 0;
6466 u16 pdev_aspmc, parent_aspmc;
6469 case PCIE_LINK_STATE_L0S:
6470 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6471 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6472 /* fall-through - can't have L1 without L0s */
6473 case PCIE_LINK_STATE_L1:
6474 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6480 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6481 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6484 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6486 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6489 /* Nothing to do if the ASPM states to be disabled already are */
6490 if (!(pdev_aspmc & aspm_dis_mask) &&
6491 (!parent || !(parent_aspmc & aspm_dis_mask)))
6494 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6495 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6497 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6500 #ifdef CONFIG_PCIEASPM
6502 pci_disable_link_state_locked(pdev, state);
6504 pci_disable_link_state(pdev, state);
6506 /* Double-check ASPM control. If not disabled by the above, the
6507 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6508 * not enabled); override by writing PCI config space directly.
6510 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6511 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6513 if (!(aspm_dis_mask & pdev_aspmc))
6517 /* Both device and parent should have the same ASPM setting.
6518 * Disable ASPM in downstream component first and then upstream.
6520 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6523 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6528 * e1000e_disable_aspm - Disable ASPM states.
6529 * @pdev: pointer to PCI device struct
6530 * @state: bit-mask of ASPM states to disable
6532 * This function acquires the pci_bus_sem!
6533 * Some devices *must* have certain ASPM states disabled per hardware errata.
6535 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6537 __e1000e_disable_aspm(pdev, state, 0);
6541 * e1000e_disable_aspm_locked Disable ASPM states.
6542 * @pdev: pointer to PCI device struct
6543 * @state: bit-mask of ASPM states to disable
6545 * This function must be called with pci_bus_sem acquired!
6546 * Some devices *must* have certain ASPM states disabled per hardware errata.
6548 static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6550 __e1000e_disable_aspm(pdev, state, 1);
6554 static int __e1000_resume(struct pci_dev *pdev)
6556 struct net_device *netdev = pci_get_drvdata(pdev);
6557 struct e1000_adapter *adapter = netdev_priv(netdev);
6558 struct e1000_hw *hw = &adapter->hw;
6559 u16 aspm_disable_flag = 0;
6561 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6562 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6563 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6564 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6565 if (aspm_disable_flag)
6566 e1000e_disable_aspm(pdev, aspm_disable_flag);
6568 pci_set_master(pdev);
6570 if (hw->mac.type >= e1000_pch2lan)
6571 e1000_resume_workarounds_pchlan(&adapter->hw);
6573 e1000e_power_up_phy(adapter);
6575 /* report the system wakeup cause from S3/S4 */
6576 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6579 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6581 e_info("PHY Wakeup cause - %s\n",
6582 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6583 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6584 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6585 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6586 phy_data & E1000_WUS_LNKC ?
6587 "Link Status Change" : "other");
6589 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6591 u32 wus = er32(WUS);
6594 e_info("MAC Wakeup cause - %s\n",
6595 wus & E1000_WUS_EX ? "Unicast Packet" :
6596 wus & E1000_WUS_MC ? "Multicast Packet" :
6597 wus & E1000_WUS_BC ? "Broadcast Packet" :
6598 wus & E1000_WUS_MAG ? "Magic Packet" :
6599 wus & E1000_WUS_LNKC ? "Link Status Change" :
6605 e1000e_reset(adapter);
6607 e1000_init_manageability_pt(adapter);
6609 /* If the controller has AMT, do not set DRV_LOAD until the interface
6610 * is up. For all other cases, let the f/w know that the h/w is now
6611 * under the control of the driver.
6613 if (!(adapter->flags & FLAG_HAS_AMT))
6614 e1000e_get_hw_control(adapter);
6619 #ifdef CONFIG_PM_SLEEP
6620 static int e1000e_pm_thaw(struct device *dev)
6622 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6623 struct e1000_adapter *adapter = netdev_priv(netdev);
6625 e1000e_set_interrupt_capability(adapter);
6626 if (netif_running(netdev)) {
6627 u32 err = e1000_request_irq(adapter);
6635 netif_device_attach(netdev);
6640 static int e1000e_pm_suspend(struct device *dev)
6642 struct pci_dev *pdev = to_pci_dev(dev);
6645 e1000e_flush_lpic(pdev);
6647 e1000e_pm_freeze(dev);
6649 rc = __e1000_shutdown(pdev, false);
6651 e1000e_pm_thaw(dev);
6656 static int e1000e_pm_resume(struct device *dev)
6658 struct pci_dev *pdev = to_pci_dev(dev);
6661 rc = __e1000_resume(pdev);
6665 return e1000e_pm_thaw(dev);
6667 #endif /* CONFIG_PM_SLEEP */
6669 static int e1000e_pm_runtime_idle(struct device *dev)
6671 struct pci_dev *pdev = to_pci_dev(dev);
6672 struct net_device *netdev = pci_get_drvdata(pdev);
6673 struct e1000_adapter *adapter = netdev_priv(netdev);
6676 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6678 if (!e1000e_has_link(adapter)) {
6679 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6680 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6686 static int e1000e_pm_runtime_resume(struct device *dev)
6688 struct pci_dev *pdev = to_pci_dev(dev);
6689 struct net_device *netdev = pci_get_drvdata(pdev);
6690 struct e1000_adapter *adapter = netdev_priv(netdev);
6693 rc = __e1000_resume(pdev);
6697 if (netdev->flags & IFF_UP)
6703 static int e1000e_pm_runtime_suspend(struct device *dev)
6705 struct pci_dev *pdev = to_pci_dev(dev);
6706 struct net_device *netdev = pci_get_drvdata(pdev);
6707 struct e1000_adapter *adapter = netdev_priv(netdev);
6709 if (netdev->flags & IFF_UP) {
6710 int count = E1000_CHECK_RESET_COUNT;
6712 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6713 usleep_range(10000, 20000);
6715 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6717 /* Down the device without resetting the hardware */
6718 e1000e_down(adapter, false);
6721 if (__e1000_shutdown(pdev, true)) {
6722 e1000e_pm_runtime_resume(dev);
6728 #endif /* CONFIG_PM */
6730 static void e1000_shutdown(struct pci_dev *pdev)
6732 e1000e_flush_lpic(pdev);
6734 e1000e_pm_freeze(&pdev->dev);
6736 __e1000_shutdown(pdev, false);
6739 #ifdef CONFIG_NET_POLL_CONTROLLER
6741 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6743 struct net_device *netdev = data;
6744 struct e1000_adapter *adapter = netdev_priv(netdev);
6746 if (adapter->msix_entries) {
6747 int vector, msix_irq;
6750 msix_irq = adapter->msix_entries[vector].vector;
6751 if (disable_hardirq(msix_irq))
6752 e1000_intr_msix_rx(msix_irq, netdev);
6753 enable_irq(msix_irq);
6756 msix_irq = adapter->msix_entries[vector].vector;
6757 if (disable_hardirq(msix_irq))
6758 e1000_intr_msix_tx(msix_irq, netdev);
6759 enable_irq(msix_irq);
6762 msix_irq = adapter->msix_entries[vector].vector;
6763 if (disable_hardirq(msix_irq))
6764 e1000_msix_other(msix_irq, netdev);
6765 enable_irq(msix_irq);
6773 * @netdev: network interface device structure
6775 * Polling 'interrupt' - used by things like netconsole to send skbs
6776 * without having to re-enable interrupts. It's not called while
6777 * the interrupt routine is executing.
6779 static void e1000_netpoll(struct net_device *netdev)
6781 struct e1000_adapter *adapter = netdev_priv(netdev);
6783 switch (adapter->int_mode) {
6784 case E1000E_INT_MODE_MSIX:
6785 e1000_intr_msix(adapter->pdev->irq, netdev);
6787 case E1000E_INT_MODE_MSI:
6788 if (disable_hardirq(adapter->pdev->irq))
6789 e1000_intr_msi(adapter->pdev->irq, netdev);
6790 enable_irq(adapter->pdev->irq);
6792 default: /* E1000E_INT_MODE_LEGACY */
6793 if (disable_hardirq(adapter->pdev->irq))
6794 e1000_intr(adapter->pdev->irq, netdev);
6795 enable_irq(adapter->pdev->irq);
6802 * e1000_io_error_detected - called when PCI error is detected
6803 * @pdev: Pointer to PCI device
6804 * @state: The current pci connection state
6806 * This function is called after a PCI bus error affecting
6807 * this device has been detected.
6809 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6810 pci_channel_state_t state)
6812 struct net_device *netdev = pci_get_drvdata(pdev);
6813 struct e1000_adapter *adapter = netdev_priv(netdev);
6815 netif_device_detach(netdev);
6817 if (state == pci_channel_io_perm_failure)
6818 return PCI_ERS_RESULT_DISCONNECT;
6820 if (netif_running(netdev))
6821 e1000e_down(adapter, true);
6822 pci_disable_device(pdev);
6824 /* Request a slot slot reset. */
6825 return PCI_ERS_RESULT_NEED_RESET;
6829 * e1000_io_slot_reset - called after the pci bus has been reset.
6830 * @pdev: Pointer to PCI device
6832 * Restart the card from scratch, as if from a cold-boot. Implementation
6833 * resembles the first-half of the e1000e_pm_resume routine.
6835 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6837 struct net_device *netdev = pci_get_drvdata(pdev);
6838 struct e1000_adapter *adapter = netdev_priv(netdev);
6839 struct e1000_hw *hw = &adapter->hw;
6840 u16 aspm_disable_flag = 0;
6842 pci_ers_result_t result;
6844 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6845 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6846 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6847 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6848 if (aspm_disable_flag)
6849 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6851 err = pci_enable_device_mem(pdev);
6854 "Cannot re-enable PCI device after reset.\n");
6855 result = PCI_ERS_RESULT_DISCONNECT;
6857 pdev->state_saved = true;
6858 pci_restore_state(pdev);
6859 pci_set_master(pdev);
6861 pci_enable_wake(pdev, PCI_D3hot, 0);
6862 pci_enable_wake(pdev, PCI_D3cold, 0);
6864 e1000e_reset(adapter);
6866 result = PCI_ERS_RESULT_RECOVERED;
6869 pci_cleanup_aer_uncorrect_error_status(pdev);
6875 * e1000_io_resume - called when traffic can start flowing again.
6876 * @pdev: Pointer to PCI device
6878 * This callback is called when the error recovery driver tells us that
6879 * its OK to resume normal operation. Implementation resembles the
6880 * second-half of the e1000e_pm_resume routine.
6882 static void e1000_io_resume(struct pci_dev *pdev)
6884 struct net_device *netdev = pci_get_drvdata(pdev);
6885 struct e1000_adapter *adapter = netdev_priv(netdev);
6887 e1000_init_manageability_pt(adapter);
6889 if (netif_running(netdev))
6892 netif_device_attach(netdev);
6894 /* If the controller has AMT, do not set DRV_LOAD until the interface
6895 * is up. For all other cases, let the f/w know that the h/w is now
6896 * under the control of the driver.
6898 if (!(adapter->flags & FLAG_HAS_AMT))
6899 e1000e_get_hw_control(adapter);
6902 static void e1000_print_device_info(struct e1000_adapter *adapter)
6904 struct e1000_hw *hw = &adapter->hw;
6905 struct net_device *netdev = adapter->netdev;
6907 u8 pba_str[E1000_PBANUM_LENGTH];
6909 /* print bus type/speed/width info */
6910 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6912 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6916 e_info("Intel(R) PRO/%s Network Connection\n",
6917 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6918 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6919 E1000_PBANUM_LENGTH);
6921 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6922 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6923 hw->mac.type, hw->phy.type, pba_str);
6926 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6928 struct e1000_hw *hw = &adapter->hw;
6932 if (hw->mac.type != e1000_82573)
6935 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6937 if (!ret_val && (!(buf & BIT(0)))) {
6938 /* Deep Smart Power Down (DSPD) */
6939 dev_warn(&adapter->pdev->dev,
6940 "Warning: detected DSPD enabled in EEPROM\n");
6944 static netdev_features_t e1000_fix_features(struct net_device *netdev,
6945 netdev_features_t features)
6947 struct e1000_adapter *adapter = netdev_priv(netdev);
6948 struct e1000_hw *hw = &adapter->hw;
6950 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6951 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6952 features &= ~NETIF_F_RXFCS;
6954 /* Since there is no support for separate Rx/Tx vlan accel
6955 * enable/disable make sure Tx flag is always in same state as Rx.
6957 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6958 features |= NETIF_F_HW_VLAN_CTAG_TX;
6960 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6965 static int e1000_set_features(struct net_device *netdev,
6966 netdev_features_t features)
6968 struct e1000_adapter *adapter = netdev_priv(netdev);
6969 netdev_features_t changed = features ^ netdev->features;
6971 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6972 adapter->flags |= FLAG_TSO_FORCE;
6974 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6975 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6979 if (changed & NETIF_F_RXFCS) {
6980 if (features & NETIF_F_RXFCS) {
6981 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6983 /* We need to take it back to defaults, which might mean
6984 * stripping is still disabled at the adapter level.
6986 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6987 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6989 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6993 netdev->features = features;
6995 if (netif_running(netdev))
6996 e1000e_reinit_locked(adapter);
6998 e1000e_reset(adapter);
7003 static const struct net_device_ops e1000e_netdev_ops = {
7004 .ndo_open = e1000e_open,
7005 .ndo_stop = e1000e_close,
7006 .ndo_start_xmit = e1000_xmit_frame,
7007 .ndo_get_stats64 = e1000e_get_stats64,
7008 .ndo_set_rx_mode = e1000e_set_rx_mode,
7009 .ndo_set_mac_address = e1000_set_mac,
7010 .ndo_change_mtu = e1000_change_mtu,
7011 .ndo_do_ioctl = e1000_ioctl,
7012 .ndo_tx_timeout = e1000_tx_timeout,
7013 .ndo_validate_addr = eth_validate_addr,
7015 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
7016 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
7017 #ifdef CONFIG_NET_POLL_CONTROLLER
7018 .ndo_poll_controller = e1000_netpoll,
7020 .ndo_set_features = e1000_set_features,
7021 .ndo_fix_features = e1000_fix_features,
7022 .ndo_features_check = passthru_features_check,
7026 * e1000_probe - Device Initialization Routine
7027 * @pdev: PCI device information struct
7028 * @ent: entry in e1000_pci_tbl
7030 * Returns 0 on success, negative on failure
7032 * e1000_probe initializes an adapter identified by a pci_dev structure.
7033 * The OS initialization, configuring of the adapter private structure,
7034 * and a hardware reset occur.
7036 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7038 struct net_device *netdev;
7039 struct e1000_adapter *adapter;
7040 struct e1000_hw *hw;
7041 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7042 resource_size_t mmio_start, mmio_len;
7043 resource_size_t flash_start, flash_len;
7044 static int cards_found;
7045 u16 aspm_disable_flag = 0;
7046 int bars, i, err, pci_using_dac;
7047 u16 eeprom_data = 0;
7048 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7051 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7052 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7053 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7054 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7055 if (aspm_disable_flag)
7056 e1000e_disable_aspm(pdev, aspm_disable_flag);
7058 err = pci_enable_device_mem(pdev);
7063 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7067 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7070 "No usable DMA configuration, aborting\n");
7075 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7076 err = pci_request_selected_regions_exclusive(pdev, bars,
7077 e1000e_driver_name);
7081 /* AER (Advanced Error Reporting) hooks */
7082 pci_enable_pcie_error_reporting(pdev);
7084 pci_set_master(pdev);
7085 /* PCI config space info */
7086 err = pci_save_state(pdev);
7088 goto err_alloc_etherdev;
7091 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7093 goto err_alloc_etherdev;
7095 SET_NETDEV_DEV(netdev, &pdev->dev);
7097 netdev->irq = pdev->irq;
7099 pci_set_drvdata(pdev, netdev);
7100 adapter = netdev_priv(netdev);
7102 adapter->netdev = netdev;
7103 adapter->pdev = pdev;
7105 adapter->pba = ei->pba;
7106 adapter->flags = ei->flags;
7107 adapter->flags2 = ei->flags2;
7108 adapter->hw.adapter = adapter;
7109 adapter->hw.mac.type = ei->mac;
7110 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7111 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7113 mmio_start = pci_resource_start(pdev, 0);
7114 mmio_len = pci_resource_len(pdev, 0);
7117 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7118 if (!adapter->hw.hw_addr)
7121 if ((adapter->flags & FLAG_HAS_FLASH) &&
7122 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7123 (hw->mac.type < e1000_pch_spt)) {
7124 flash_start = pci_resource_start(pdev, 1);
7125 flash_len = pci_resource_len(pdev, 1);
7126 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7127 if (!adapter->hw.flash_address)
7131 /* Set default EEE advertisement */
7132 if (adapter->flags2 & FLAG2_HAS_EEE)
7133 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7135 /* construct the net_device struct */
7136 netdev->netdev_ops = &e1000e_netdev_ops;
7137 e1000e_set_ethtool_ops(netdev);
7138 netdev->watchdog_timeo = 5 * HZ;
7139 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7140 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7142 netdev->mem_start = mmio_start;
7143 netdev->mem_end = mmio_start + mmio_len;
7145 adapter->bd_number = cards_found++;
7147 e1000e_check_options(adapter);
7149 /* setup adapter struct */
7150 err = e1000_sw_init(adapter);
7154 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7155 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7156 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7158 err = ei->get_variants(adapter);
7162 if ((adapter->flags & FLAG_IS_ICH) &&
7163 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7164 (hw->mac.type < e1000_pch_spt))
7165 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7167 hw->mac.ops.get_bus_info(&adapter->hw);
7169 adapter->hw.phy.autoneg_wait_to_complete = 0;
7171 /* Copper options */
7172 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7173 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7174 adapter->hw.phy.disable_polarity_correction = 0;
7175 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7178 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7179 dev_info(&pdev->dev,
7180 "PHY reset is blocked due to SOL/IDER session.\n");
7182 /* Set initial default active device features */
7183 netdev->features = (NETIF_F_SG |
7184 NETIF_F_HW_VLAN_CTAG_RX |
7185 NETIF_F_HW_VLAN_CTAG_TX |
7192 /* Set user-changeable features (subset of all device features) */
7193 netdev->hw_features = netdev->features;
7194 netdev->hw_features |= NETIF_F_RXFCS;
7195 netdev->priv_flags |= IFF_SUPP_NOFCS;
7196 netdev->hw_features |= NETIF_F_RXALL;
7198 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7199 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7201 netdev->vlan_features |= (NETIF_F_SG |
7206 netdev->priv_flags |= IFF_UNICAST_FLT;
7208 if (pci_using_dac) {
7209 netdev->features |= NETIF_F_HIGHDMA;
7210 netdev->vlan_features |= NETIF_F_HIGHDMA;
7213 /* MTU range: 68 - max_hw_frame_size */
7214 netdev->min_mtu = ETH_MIN_MTU;
7215 netdev->max_mtu = adapter->max_hw_frame_size -
7216 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7218 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7219 adapter->flags |= FLAG_MNG_PT_ENABLED;
7221 /* before reading the NVM, reset the controller to
7222 * put the device in a known good starting state
7224 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7226 /* systems with ASPM and others may see the checksum fail on the first
7227 * attempt. Let's give it a few tries
7230 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7233 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7239 e1000_eeprom_checks(adapter);
7241 /* copy the MAC address */
7242 if (e1000e_read_mac_addr(&adapter->hw))
7244 "NVM Read Error while reading MAC address\n");
7246 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7248 if (!is_valid_ether_addr(netdev->dev_addr)) {
7249 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7255 init_timer(&adapter->watchdog_timer);
7256 adapter->watchdog_timer.function = e1000_watchdog;
7257 adapter->watchdog_timer.data = (unsigned long)adapter;
7259 init_timer(&adapter->phy_info_timer);
7260 adapter->phy_info_timer.function = e1000_update_phy_info;
7261 adapter->phy_info_timer.data = (unsigned long)adapter;
7263 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7264 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7265 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7266 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7267 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7269 /* Initialize link parameters. User can change them with ethtool */
7270 adapter->hw.mac.autoneg = 1;
7271 adapter->fc_autoneg = true;
7272 adapter->hw.fc.requested_mode = e1000_fc_default;
7273 adapter->hw.fc.current_mode = e1000_fc_default;
7274 adapter->hw.phy.autoneg_advertised = 0x2f;
7276 /* Initial Wake on LAN setting - If APM wake is enabled in
7277 * the EEPROM, enable the ACPI Magic Packet filter
7279 if (adapter->flags & FLAG_APME_IN_WUC) {
7280 /* APME bit in EEPROM is mapped to WUC.APME */
7281 eeprom_data = er32(WUC);
7282 eeprom_apme_mask = E1000_WUC_APME;
7283 if ((hw->mac.type > e1000_ich10lan) &&
7284 (eeprom_data & E1000_WUC_PHY_WAKE))
7285 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7286 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7287 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7288 (adapter->hw.bus.func == 1))
7289 ret_val = e1000_read_nvm(&adapter->hw,
7290 NVM_INIT_CONTROL3_PORT_B,
7293 ret_val = e1000_read_nvm(&adapter->hw,
7294 NVM_INIT_CONTROL3_PORT_A,
7298 /* fetch WoL from EEPROM */
7300 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7301 else if (eeprom_data & eeprom_apme_mask)
7302 adapter->eeprom_wol |= E1000_WUFC_MAG;
7304 /* now that we have the eeprom settings, apply the special cases
7305 * where the eeprom may be wrong or the board simply won't support
7306 * wake on lan on a particular port
7308 if (!(adapter->flags & FLAG_HAS_WOL))
7309 adapter->eeprom_wol = 0;
7311 /* initialize the wol settings based on the eeprom settings */
7312 adapter->wol = adapter->eeprom_wol;
7314 /* make sure adapter isn't asleep if manageability is enabled */
7315 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7316 (hw->mac.ops.check_mng_mode(hw)))
7317 device_wakeup_enable(&pdev->dev);
7319 /* save off EEPROM version number */
7320 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7323 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7324 adapter->eeprom_vers = 0;
7327 /* init PTP hardware clock */
7328 e1000e_ptp_init(adapter);
7330 /* reset the hardware with the new settings */
7331 e1000e_reset(adapter);
7333 /* If the controller has AMT, do not set DRV_LOAD until the interface
7334 * is up. For all other cases, let the f/w know that the h/w is now
7335 * under the control of the driver.
7337 if (!(adapter->flags & FLAG_HAS_AMT))
7338 e1000e_get_hw_control(adapter);
7340 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7341 err = register_netdev(netdev);
7345 /* carrier off reporting is important to ethtool even BEFORE open */
7346 netif_carrier_off(netdev);
7348 e1000_print_device_info(adapter);
7350 if (pci_dev_run_wake(pdev))
7351 pm_runtime_put_noidle(&pdev->dev);
7356 if (!(adapter->flags & FLAG_HAS_AMT))
7357 e1000e_release_hw_control(adapter);
7359 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7360 e1000_phy_hw_reset(&adapter->hw);
7362 kfree(adapter->tx_ring);
7363 kfree(adapter->rx_ring);
7365 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7366 iounmap(adapter->hw.flash_address);
7367 e1000e_reset_interrupt_capability(adapter);
7369 iounmap(adapter->hw.hw_addr);
7371 free_netdev(netdev);
7373 pci_release_mem_regions(pdev);
7376 pci_disable_device(pdev);
7381 * e1000_remove - Device Removal Routine
7382 * @pdev: PCI device information struct
7384 * e1000_remove is called by the PCI subsystem to alert the driver
7385 * that it should release a PCI device. The could be caused by a
7386 * Hot-Plug event, or because the driver is going to be removed from
7389 static void e1000_remove(struct pci_dev *pdev)
7391 struct net_device *netdev = pci_get_drvdata(pdev);
7392 struct e1000_adapter *adapter = netdev_priv(netdev);
7393 bool down = test_bit(__E1000_DOWN, &adapter->state);
7395 e1000e_ptp_remove(adapter);
7397 /* The timers may be rescheduled, so explicitly disable them
7398 * from being rescheduled.
7401 set_bit(__E1000_DOWN, &adapter->state);
7402 del_timer_sync(&adapter->watchdog_timer);
7403 del_timer_sync(&adapter->phy_info_timer);
7405 cancel_work_sync(&adapter->reset_task);
7406 cancel_work_sync(&adapter->watchdog_task);
7407 cancel_work_sync(&adapter->downshift_task);
7408 cancel_work_sync(&adapter->update_phy_task);
7409 cancel_work_sync(&adapter->print_hang_task);
7411 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7412 cancel_work_sync(&adapter->tx_hwtstamp_work);
7413 if (adapter->tx_hwtstamp_skb) {
7414 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
7415 adapter->tx_hwtstamp_skb = NULL;
7419 /* Don't lie to e1000_close() down the road. */
7421 clear_bit(__E1000_DOWN, &adapter->state);
7422 unregister_netdev(netdev);
7424 if (pci_dev_run_wake(pdev))
7425 pm_runtime_get_noresume(&pdev->dev);
7427 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7428 * would have already happened in close and is redundant.
7430 e1000e_release_hw_control(adapter);
7432 e1000e_reset_interrupt_capability(adapter);
7433 kfree(adapter->tx_ring);
7434 kfree(adapter->rx_ring);
7436 iounmap(adapter->hw.hw_addr);
7437 if ((adapter->hw.flash_address) &&
7438 (adapter->hw.mac.type < e1000_pch_spt))
7439 iounmap(adapter->hw.flash_address);
7440 pci_release_mem_regions(pdev);
7442 free_netdev(netdev);
7445 pci_disable_pcie_error_reporting(pdev);
7447 pci_disable_device(pdev);
7450 /* PCI Error Recovery (ERS) */
7451 static const struct pci_error_handlers e1000_err_handler = {
7452 .error_detected = e1000_io_error_detected,
7453 .slot_reset = e1000_io_slot_reset,
7454 .resume = e1000_io_resume,
7457 static const struct pci_device_id e1000_pci_tbl[] = {
7458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7459 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7460 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7464 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7466 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7467 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7470 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7472 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7479 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7482 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7483 board_80003es2lan },
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7485 board_80003es2lan },
7486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7487 board_80003es2lan },
7488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7489 board_80003es2lan },
7491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7492 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7493 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7494 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7495 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7496 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7500 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7503 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7504 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7505 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7507 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7508 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7511 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7514 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7515 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7516 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7518 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7519 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7520 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7521 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7523 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7524 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7526 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7527 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7528 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7529 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7530 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7531 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7532 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7533 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7534 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7535 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7536 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7537 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7538 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7539 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7540 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7541 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7542 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7543 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7544 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7545 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7546 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7548 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7550 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7552 static const struct dev_pm_ops e1000_pm_ops = {
7553 #ifdef CONFIG_PM_SLEEP
7554 .suspend = e1000e_pm_suspend,
7555 .resume = e1000e_pm_resume,
7556 .freeze = e1000e_pm_freeze,
7557 .thaw = e1000e_pm_thaw,
7558 .poweroff = e1000e_pm_suspend,
7559 .restore = e1000e_pm_resume,
7561 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7562 e1000e_pm_runtime_idle)
7565 /* PCI Device API Driver */
7566 static struct pci_driver e1000_driver = {
7567 .name = e1000e_driver_name,
7568 .id_table = e1000_pci_tbl,
7569 .probe = e1000_probe,
7570 .remove = e1000_remove,
7572 .pm = &e1000_pm_ops,
7574 .shutdown = e1000_shutdown,
7575 .err_handler = &e1000_err_handler
7579 * e1000_init_module - Driver Registration Routine
7581 * e1000_init_module is the first routine called when the driver is
7582 * loaded. All it does is register with the PCI subsystem.
7584 static int __init e1000_init_module(void)
7586 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7587 e1000e_driver_version);
7588 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7590 return pci_register_driver(&e1000_driver);
7592 module_init(e1000_init_module);
7595 * e1000_exit_module - Driver Exit Cleanup Routine
7597 * e1000_exit_module is called just before the driver is removed
7600 static void __exit e1000_exit_module(void)
7602 pci_unregister_driver(&e1000_driver);
7604 module_exit(e1000_exit_module);
7606 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7607 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7608 MODULE_LICENSE("GPL");
7609 MODULE_VERSION(DRV_VERSION);