1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/module.h>
7 #include <linux/types.h>
8 #include <linux/init.h>
10 #include <linux/vmalloc.h>
11 #include <linux/pagemap.h>
12 #include <linux/delay.h>
13 #include <linux/netdevice.h>
14 #include <linux/interrupt.h>
15 #include <linux/tcp.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/ethtool.h>
21 #include <linux/if_vlan.h>
22 #include <linux/cpu.h>
23 #include <linux/smp.h>
24 #include <linux/pm_qos.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/aer.h>
27 #include <linux/prefetch.h>
31 #define DRV_EXTRAVERSION "-k"
33 #define DRV_VERSION "3.2.6" DRV_EXTRAVERSION
34 char e1000e_driver_name[] = "e1000e";
35 const char e1000e_driver_version[] = DRV_VERSION;
37 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
38 static int debug = -1;
39 module_param(debug, int, 0);
40 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
42 static const struct e1000_info *e1000_info_tbl[] = {
43 [board_82571] = &e1000_82571_info,
44 [board_82572] = &e1000_82572_info,
45 [board_82573] = &e1000_82573_info,
46 [board_82574] = &e1000_82574_info,
47 [board_82583] = &e1000_82583_info,
48 [board_80003es2lan] = &e1000_es2_info,
49 [board_ich8lan] = &e1000_ich8_info,
50 [board_ich9lan] = &e1000_ich9_info,
51 [board_ich10lan] = &e1000_ich10_info,
52 [board_pchlan] = &e1000_pch_info,
53 [board_pch2lan] = &e1000_pch2_info,
54 [board_pch_lpt] = &e1000_pch_lpt_info,
55 [board_pch_spt] = &e1000_pch_spt_info,
56 [board_pch_cnp] = &e1000_pch_cnp_info,
59 struct e1000_reg_info {
64 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
65 /* General Registers */
67 {E1000_STATUS, "STATUS"},
68 {E1000_CTRL_EXT, "CTRL_EXT"},
70 /* Interrupt Registers */
75 {E1000_RDLEN(0), "RDLEN"},
76 {E1000_RDH(0), "RDH"},
77 {E1000_RDT(0), "RDT"},
79 {E1000_RXDCTL(0), "RXDCTL"},
81 {E1000_RDBAL(0), "RDBAL"},
82 {E1000_RDBAH(0), "RDBAH"},
85 {E1000_RDFHS, "RDFHS"},
86 {E1000_RDFTS, "RDFTS"},
87 {E1000_RDFPC, "RDFPC"},
91 {E1000_TDBAL(0), "TDBAL"},
92 {E1000_TDBAH(0), "TDBAH"},
93 {E1000_TDLEN(0), "TDLEN"},
94 {E1000_TDH(0), "TDH"},
95 {E1000_TDT(0), "TDT"},
97 {E1000_TXDCTL(0), "TXDCTL"},
99 {E1000_TARC(0), "TARC"},
100 {E1000_TDFH, "TDFH"},
101 {E1000_TDFT, "TDFT"},
102 {E1000_TDFHS, "TDFHS"},
103 {E1000_TDFTS, "TDFTS"},
104 {E1000_TDFPC, "TDFPC"},
106 /* List Terminator */
111 * __ew32_prepare - prepare to write to MAC CSR register on certain parts
112 * @hw: pointer to the HW structure
114 * When updating the MAC CSR registers, the Manageability Engine (ME) could
115 * be accessing the registers at the same time. Normally, this is handled in
116 * h/w by an arbiter but on some parts there is a bug that acknowledges Host
117 * accesses later than it should which could result in the register to have
118 * an incorrect value. Workaround this by checking the FWSM register which
119 * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
120 * and try again a number of times.
122 s32 __ew32_prepare(struct e1000_hw *hw)
124 s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
126 while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
132 void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
134 if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
137 writel(val, hw->hw_addr + reg);
141 * e1000_regdump - register printout routine
142 * @hw: pointer to the HW structure
143 * @reginfo: pointer to the register info table
145 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
151 switch (reginfo->ofs) {
152 case E1000_RXDCTL(0):
153 for (n = 0; n < 2; n++)
154 regs[n] = __er32(hw, E1000_RXDCTL(n));
156 case E1000_TXDCTL(0):
157 for (n = 0; n < 2; n++)
158 regs[n] = __er32(hw, E1000_TXDCTL(n));
161 for (n = 0; n < 2; n++)
162 regs[n] = __er32(hw, E1000_TARC(n));
165 pr_info("%-15s %08x\n",
166 reginfo->name, __er32(hw, reginfo->ofs));
170 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
171 pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
174 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
175 struct e1000_buffer *bi)
178 struct e1000_ps_page *ps_page;
180 for (i = 0; i < adapter->rx_ps_pages; i++) {
181 ps_page = &bi->ps_pages[i];
184 pr_info("packet dump for ps_page %d:\n", i);
185 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
186 16, 1, page_address(ps_page->page),
193 * e1000e_dump - Print registers, Tx-ring and Rx-ring
194 * @adapter: board private structure
196 static void e1000e_dump(struct e1000_adapter *adapter)
198 struct net_device *netdev = adapter->netdev;
199 struct e1000_hw *hw = &adapter->hw;
200 struct e1000_reg_info *reginfo;
201 struct e1000_ring *tx_ring = adapter->tx_ring;
202 struct e1000_tx_desc *tx_desc;
207 struct e1000_buffer *buffer_info;
208 struct e1000_ring *rx_ring = adapter->rx_ring;
209 union e1000_rx_desc_packet_split *rx_desc_ps;
210 union e1000_rx_desc_extended *rx_desc;
220 if (!netif_msg_hw(adapter))
223 /* Print netdevice Info */
225 dev_info(&adapter->pdev->dev, "Net device Info\n");
226 pr_info("Device Name state trans_start\n");
227 pr_info("%-15s %016lX %016lX\n", netdev->name,
228 netdev->state, dev_trans_start(netdev));
231 /* Print Registers */
232 dev_info(&adapter->pdev->dev, "Register Dump\n");
233 pr_info(" Register Name Value\n");
234 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
235 reginfo->name; reginfo++) {
236 e1000_regdump(hw, reginfo);
239 /* Print Tx Ring Summary */
240 if (!netdev || !netif_running(netdev))
243 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
244 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
245 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
246 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
247 0, tx_ring->next_to_use, tx_ring->next_to_clean,
248 (unsigned long long)buffer_info->dma,
250 buffer_info->next_to_watch,
251 (unsigned long long)buffer_info->time_stamp);
254 if (!netif_msg_tx_done(adapter))
255 goto rx_ring_summary;
257 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
259 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
261 * Legacy Transmit Descriptor
262 * +--------------------------------------------------------------+
263 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
264 * +--------------------------------------------------------------+
265 * 8 | Special | CSS | Status | CMD | CSO | Length |
266 * +--------------------------------------------------------------+
267 * 63 48 47 36 35 32 31 24 23 16 15 0
269 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
270 * 63 48 47 40 39 32 31 16 15 8 7 0
271 * +----------------------------------------------------------------+
272 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
273 * +----------------------------------------------------------------+
274 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
275 * +----------------------------------------------------------------+
276 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
278 * Extended Data Descriptor (DTYP=0x1)
279 * +----------------------------------------------------------------+
280 * 0 | Buffer Address [63:0] |
281 * +----------------------------------------------------------------+
282 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
283 * +----------------------------------------------------------------+
284 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
286 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
287 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
288 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
289 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
290 const char *next_desc;
291 tx_desc = E1000_TX_DESC(*tx_ring, i);
292 buffer_info = &tx_ring->buffer_info[i];
293 u0 = (struct my_u0 *)tx_desc;
294 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
295 next_desc = " NTC/U";
296 else if (i == tx_ring->next_to_use)
298 else if (i == tx_ring->next_to_clean)
302 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
303 (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
304 ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
306 (unsigned long long)le64_to_cpu(u0->a),
307 (unsigned long long)le64_to_cpu(u0->b),
308 (unsigned long long)buffer_info->dma,
309 buffer_info->length, buffer_info->next_to_watch,
310 (unsigned long long)buffer_info->time_stamp,
311 buffer_info->skb, next_desc);
313 if (netif_msg_pktdata(adapter) && buffer_info->skb)
314 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
315 16, 1, buffer_info->skb->data,
316 buffer_info->skb->len, true);
319 /* Print Rx Ring Summary */
321 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
322 pr_info("Queue [NTU] [NTC]\n");
323 pr_info(" %5d %5X %5X\n",
324 0, rx_ring->next_to_use, rx_ring->next_to_clean);
327 if (!netif_msg_rx_status(adapter))
330 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
331 switch (adapter->rx_ps_pages) {
335 /* [Extended] Packet Split Receive Descriptor Format
337 * +-----------------------------------------------------+
338 * 0 | Buffer Address 0 [63:0] |
339 * +-----------------------------------------------------+
340 * 8 | Buffer Address 1 [63:0] |
341 * +-----------------------------------------------------+
342 * 16 | Buffer Address 2 [63:0] |
343 * +-----------------------------------------------------+
344 * 24 | Buffer Address 3 [63:0] |
345 * +-----------------------------------------------------+
347 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");
348 /* [Extended] Receive Descriptor (Write-Back) Format
350 * 63 48 47 32 31 13 12 8 7 4 3 0
351 * +------------------------------------------------------+
352 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
353 * | Checksum | Ident | | Queue | | Type |
354 * +------------------------------------------------------+
355 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
356 * +------------------------------------------------------+
357 * 63 48 47 32 31 20 19 0
359 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
360 for (i = 0; i < rx_ring->count; i++) {
361 const char *next_desc;
362 buffer_info = &rx_ring->buffer_info[i];
363 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
364 u1 = (struct my_u1 *)rx_desc_ps;
366 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
368 if (i == rx_ring->next_to_use)
370 else if (i == rx_ring->next_to_clean)
375 if (staterr & E1000_RXD_STAT_DD) {
376 /* Descriptor Done */
377 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
379 (unsigned long long)le64_to_cpu(u1->a),
380 (unsigned long long)le64_to_cpu(u1->b),
381 (unsigned long long)le64_to_cpu(u1->c),
382 (unsigned long long)le64_to_cpu(u1->d),
383 buffer_info->skb, next_desc);
385 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
387 (unsigned long long)le64_to_cpu(u1->a),
388 (unsigned long long)le64_to_cpu(u1->b),
389 (unsigned long long)le64_to_cpu(u1->c),
390 (unsigned long long)le64_to_cpu(u1->d),
391 (unsigned long long)buffer_info->dma,
392 buffer_info->skb, next_desc);
394 if (netif_msg_pktdata(adapter))
395 e1000e_dump_ps_pages(adapter,
402 /* Extended Receive Descriptor (Read) Format
404 * +-----------------------------------------------------+
405 * 0 | Buffer Address [63:0] |
406 * +-----------------------------------------------------+
408 * +-----------------------------------------------------+
410 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
411 /* Extended Receive Descriptor (Write-Back) Format
413 * 63 48 47 32 31 24 23 4 3 0
414 * +------------------------------------------------------+
416 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
417 * | Packet | IP | | | Type |
418 * | Checksum | Ident | | | |
419 * +------------------------------------------------------+
420 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
421 * +------------------------------------------------------+
422 * 63 48 47 32 31 20 19 0
424 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
426 for (i = 0; i < rx_ring->count; i++) {
427 const char *next_desc;
429 buffer_info = &rx_ring->buffer_info[i];
430 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
431 u1 = (struct my_u1 *)rx_desc;
432 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
434 if (i == rx_ring->next_to_use)
436 else if (i == rx_ring->next_to_clean)
441 if (staterr & E1000_RXD_STAT_DD) {
442 /* Descriptor Done */
443 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
445 (unsigned long long)le64_to_cpu(u1->a),
446 (unsigned long long)le64_to_cpu(u1->b),
447 buffer_info->skb, next_desc);
449 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
451 (unsigned long long)le64_to_cpu(u1->a),
452 (unsigned long long)le64_to_cpu(u1->b),
453 (unsigned long long)buffer_info->dma,
454 buffer_info->skb, next_desc);
456 if (netif_msg_pktdata(adapter) &&
458 print_hex_dump(KERN_INFO, "",
459 DUMP_PREFIX_ADDRESS, 16,
461 buffer_info->skb->data,
462 adapter->rx_buffer_len,
470 * e1000_desc_unused - calculate if we have unused descriptors
472 static int e1000_desc_unused(struct e1000_ring *ring)
474 if (ring->next_to_clean > ring->next_to_use)
475 return ring->next_to_clean - ring->next_to_use - 1;
477 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
481 * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
482 * @adapter: board private structure
483 * @hwtstamps: time stamp structure to update
484 * @systim: unsigned 64bit system time value.
486 * Convert the system time value stored in the RX/TXSTMP registers into a
487 * hwtstamp which can be used by the upper level time stamping functions.
489 * The 'systim_lock' spinlock is used to protect the consistency of the
490 * system time value. This is needed because reading the 64 bit time
491 * value involves reading two 32 bit registers. The first read latches the
494 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
495 struct skb_shared_hwtstamps *hwtstamps,
501 spin_lock_irqsave(&adapter->systim_lock, flags);
502 ns = timecounter_cyc2time(&adapter->tc, systim);
503 spin_unlock_irqrestore(&adapter->systim_lock, flags);
505 memset(hwtstamps, 0, sizeof(*hwtstamps));
506 hwtstamps->hwtstamp = ns_to_ktime(ns);
510 * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
511 * @adapter: board private structure
512 * @status: descriptor extended error and status field
513 * @skb: particular skb to include time stamp
515 * If the time stamp is valid, convert it into the timecounter ns value
516 * and store that result into the shhwtstamps structure which is passed
517 * up the network stack.
519 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
522 struct e1000_hw *hw = &adapter->hw;
525 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
526 !(status & E1000_RXDEXT_STATERR_TST) ||
527 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
530 /* The Rx time stamp registers contain the time stamp. No other
531 * received packet will be time stamped until the Rx time stamp
532 * registers are read. Because only one packet can be time stamped
533 * at a time, the register values must belong to this packet and
534 * therefore none of the other additional attributes need to be
537 rxstmp = (u64)er32(RXSTMPL);
538 rxstmp |= (u64)er32(RXSTMPH) << 32;
539 e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
541 adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
545 * e1000_receive_skb - helper function to handle Rx indications
546 * @adapter: board private structure
547 * @staterr: descriptor extended error and status field as written by hardware
548 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
549 * @skb: pointer to sk_buff to be indicated to stack
551 static void e1000_receive_skb(struct e1000_adapter *adapter,
552 struct net_device *netdev, struct sk_buff *skb,
553 u32 staterr, __le16 vlan)
555 u16 tag = le16_to_cpu(vlan);
557 e1000e_rx_hwtstamp(adapter, staterr, skb);
559 skb->protocol = eth_type_trans(skb, netdev);
561 if (staterr & E1000_RXD_STAT_VP)
562 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
564 napi_gro_receive(&adapter->napi, skb);
568 * e1000_rx_checksum - Receive Checksum Offload
569 * @adapter: board private structure
570 * @status_err: receive descriptor status and error fields
571 * @csum: receive descriptor csum field
572 * @sk_buff: socket buffer with received data
574 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
577 u16 status = (u16)status_err;
578 u8 errors = (u8)(status_err >> 24);
580 skb_checksum_none_assert(skb);
582 /* Rx checksum disabled */
583 if (!(adapter->netdev->features & NETIF_F_RXCSUM))
586 /* Ignore Checksum bit is set */
587 if (status & E1000_RXD_STAT_IXSM)
590 /* TCP/UDP checksum error bit or IP checksum error bit is set */
591 if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
592 /* let the stack verify checksum errors */
593 adapter->hw_csum_err++;
597 /* TCP/UDP Checksum has not been calculated */
598 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
601 /* It must be a TCP or UDP packet with a valid checksum */
602 skb->ip_summed = CHECKSUM_UNNECESSARY;
603 adapter->hw_csum_good++;
606 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
608 struct e1000_adapter *adapter = rx_ring->adapter;
609 struct e1000_hw *hw = &adapter->hw;
610 s32 ret_val = __ew32_prepare(hw);
612 writel(i, rx_ring->tail);
614 if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
615 u32 rctl = er32(RCTL);
617 ew32(RCTL, rctl & ~E1000_RCTL_EN);
618 e_err("ME firmware caused invalid RDT - resetting\n");
619 schedule_work(&adapter->reset_task);
623 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
625 struct e1000_adapter *adapter = tx_ring->adapter;
626 struct e1000_hw *hw = &adapter->hw;
627 s32 ret_val = __ew32_prepare(hw);
629 writel(i, tx_ring->tail);
631 if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
632 u32 tctl = er32(TCTL);
634 ew32(TCTL, tctl & ~E1000_TCTL_EN);
635 e_err("ME firmware caused invalid TDT - resetting\n");
636 schedule_work(&adapter->reset_task);
641 * e1000_alloc_rx_buffers - Replace used receive buffers
642 * @rx_ring: Rx descriptor ring
644 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
645 int cleaned_count, gfp_t gfp)
647 struct e1000_adapter *adapter = rx_ring->adapter;
648 struct net_device *netdev = adapter->netdev;
649 struct pci_dev *pdev = adapter->pdev;
650 union e1000_rx_desc_extended *rx_desc;
651 struct e1000_buffer *buffer_info;
654 unsigned int bufsz = adapter->rx_buffer_len;
656 i = rx_ring->next_to_use;
657 buffer_info = &rx_ring->buffer_info[i];
659 while (cleaned_count--) {
660 skb = buffer_info->skb;
666 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
668 /* Better luck next round */
669 adapter->alloc_rx_buff_failed++;
673 buffer_info->skb = skb;
675 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
676 adapter->rx_buffer_len,
678 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
679 dev_err(&pdev->dev, "Rx DMA map failed\n");
680 adapter->rx_dma_failed++;
684 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
685 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
687 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
688 /* Force memory writes to complete before letting h/w
689 * know there are new descriptors to fetch. (Only
690 * applicable for weak-ordered memory model archs,
694 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
695 e1000e_update_rdt_wa(rx_ring, i);
697 writel(i, rx_ring->tail);
700 if (i == rx_ring->count)
702 buffer_info = &rx_ring->buffer_info[i];
705 rx_ring->next_to_use = i;
709 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
710 * @rx_ring: Rx descriptor ring
712 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
713 int cleaned_count, gfp_t gfp)
715 struct e1000_adapter *adapter = rx_ring->adapter;
716 struct net_device *netdev = adapter->netdev;
717 struct pci_dev *pdev = adapter->pdev;
718 union e1000_rx_desc_packet_split *rx_desc;
719 struct e1000_buffer *buffer_info;
720 struct e1000_ps_page *ps_page;
724 i = rx_ring->next_to_use;
725 buffer_info = &rx_ring->buffer_info[i];
727 while (cleaned_count--) {
728 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
730 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
731 ps_page = &buffer_info->ps_pages[j];
732 if (j >= adapter->rx_ps_pages) {
733 /* all unused desc entries get hw null ptr */
734 rx_desc->read.buffer_addr[j + 1] =
738 if (!ps_page->page) {
739 ps_page->page = alloc_page(gfp);
740 if (!ps_page->page) {
741 adapter->alloc_rx_buff_failed++;
744 ps_page->dma = dma_map_page(&pdev->dev,
748 if (dma_mapping_error(&pdev->dev,
750 dev_err(&adapter->pdev->dev,
751 "Rx DMA page map failed\n");
752 adapter->rx_dma_failed++;
756 /* Refresh the desc even if buffer_addrs
757 * didn't change because each write-back
760 rx_desc->read.buffer_addr[j + 1] =
761 cpu_to_le64(ps_page->dma);
764 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
768 adapter->alloc_rx_buff_failed++;
772 buffer_info->skb = skb;
773 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
774 adapter->rx_ps_bsize0,
776 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
777 dev_err(&pdev->dev, "Rx DMA map failed\n");
778 adapter->rx_dma_failed++;
780 dev_kfree_skb_any(skb);
781 buffer_info->skb = NULL;
785 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
787 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
788 /* Force memory writes to complete before letting h/w
789 * know there are new descriptors to fetch. (Only
790 * applicable for weak-ordered memory model archs,
794 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
795 e1000e_update_rdt_wa(rx_ring, i << 1);
797 writel(i << 1, rx_ring->tail);
801 if (i == rx_ring->count)
803 buffer_info = &rx_ring->buffer_info[i];
807 rx_ring->next_to_use = i;
811 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
812 * @rx_ring: Rx descriptor ring
813 * @cleaned_count: number of buffers to allocate this pass
816 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
817 int cleaned_count, gfp_t gfp)
819 struct e1000_adapter *adapter = rx_ring->adapter;
820 struct net_device *netdev = adapter->netdev;
821 struct pci_dev *pdev = adapter->pdev;
822 union e1000_rx_desc_extended *rx_desc;
823 struct e1000_buffer *buffer_info;
826 unsigned int bufsz = 256 - 16; /* for skb_reserve */
828 i = rx_ring->next_to_use;
829 buffer_info = &rx_ring->buffer_info[i];
831 while (cleaned_count--) {
832 skb = buffer_info->skb;
838 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
839 if (unlikely(!skb)) {
840 /* Better luck next round */
841 adapter->alloc_rx_buff_failed++;
845 buffer_info->skb = skb;
847 /* allocate a new page if necessary */
848 if (!buffer_info->page) {
849 buffer_info->page = alloc_page(gfp);
850 if (unlikely(!buffer_info->page)) {
851 adapter->alloc_rx_buff_failed++;
856 if (!buffer_info->dma) {
857 buffer_info->dma = dma_map_page(&pdev->dev,
858 buffer_info->page, 0,
861 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
862 adapter->alloc_rx_buff_failed++;
867 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
868 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
870 if (unlikely(++i == rx_ring->count))
872 buffer_info = &rx_ring->buffer_info[i];
875 if (likely(rx_ring->next_to_use != i)) {
876 rx_ring->next_to_use = i;
877 if (unlikely(i-- == 0))
878 i = (rx_ring->count - 1);
880 /* Force memory writes to complete before letting h/w
881 * know there are new descriptors to fetch. (Only
882 * applicable for weak-ordered memory model archs,
886 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
887 e1000e_update_rdt_wa(rx_ring, i);
889 writel(i, rx_ring->tail);
893 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
896 if (netdev->features & NETIF_F_RXHASH)
897 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
901 * e1000_clean_rx_irq - Send received data up the network stack
902 * @rx_ring: Rx descriptor ring
904 * the return value indicates whether actual cleaning was done, there
905 * is no guarantee that everything was cleaned
907 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
910 struct e1000_adapter *adapter = rx_ring->adapter;
911 struct net_device *netdev = adapter->netdev;
912 struct pci_dev *pdev = adapter->pdev;
913 struct e1000_hw *hw = &adapter->hw;
914 union e1000_rx_desc_extended *rx_desc, *next_rxd;
915 struct e1000_buffer *buffer_info, *next_buffer;
918 int cleaned_count = 0;
919 bool cleaned = false;
920 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
922 i = rx_ring->next_to_clean;
923 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
924 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
925 buffer_info = &rx_ring->buffer_info[i];
927 while (staterr & E1000_RXD_STAT_DD) {
930 if (*work_done >= work_to_do)
933 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
935 skb = buffer_info->skb;
936 buffer_info->skb = NULL;
938 prefetch(skb->data - NET_IP_ALIGN);
941 if (i == rx_ring->count)
943 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
946 next_buffer = &rx_ring->buffer_info[i];
950 dma_unmap_single(&pdev->dev, buffer_info->dma,
951 adapter->rx_buffer_len, DMA_FROM_DEVICE);
952 buffer_info->dma = 0;
954 length = le16_to_cpu(rx_desc->wb.upper.length);
956 /* !EOP means multiple descriptors were used to store a single
957 * packet, if that's the case we need to toss it. In fact, we
958 * need to toss every packet with the EOP bit clear and the
959 * next frame that _does_ have the EOP bit set, as it is by
960 * definition only a frame fragment
962 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
963 adapter->flags2 |= FLAG2_IS_DISCARDING;
965 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
966 /* All receives must fit into a single buffer */
967 e_dbg("Receive packet consumed multiple buffers\n");
969 buffer_info->skb = skb;
970 if (staterr & E1000_RXD_STAT_EOP)
971 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
975 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
976 !(netdev->features & NETIF_F_RXALL))) {
978 buffer_info->skb = skb;
982 /* adjust length to remove Ethernet CRC */
983 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
984 /* If configured to store CRC, don't subtract FCS,
985 * but keep the FCS bytes out of the total_rx_bytes
988 if (netdev->features & NETIF_F_RXFCS)
994 total_rx_bytes += length;
997 /* code added for copybreak, this should improve
998 * performance for small packets with large amounts
999 * of reassembly being done in the stack
1001 if (length < copybreak) {
1002 struct sk_buff *new_skb =
1003 napi_alloc_skb(&adapter->napi, length);
1005 skb_copy_to_linear_data_offset(new_skb,
1011 /* save the skb in buffer_info as good */
1012 buffer_info->skb = skb;
1015 /* else just continue with the old one */
1017 /* end copybreak code */
1018 skb_put(skb, length);
1020 /* Receive Checksum Offload */
1021 e1000_rx_checksum(adapter, staterr, skb);
1023 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1025 e1000_receive_skb(adapter, netdev, skb, staterr,
1026 rx_desc->wb.upper.vlan);
1029 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1031 /* return some buffers to hardware, one at a time is too slow */
1032 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1033 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1038 /* use prefetched values */
1040 buffer_info = next_buffer;
1042 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1044 rx_ring->next_to_clean = i;
1046 cleaned_count = e1000_desc_unused(rx_ring);
1048 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1050 adapter->total_rx_bytes += total_rx_bytes;
1051 adapter->total_rx_packets += total_rx_packets;
1055 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1056 struct e1000_buffer *buffer_info,
1059 struct e1000_adapter *adapter = tx_ring->adapter;
1061 if (buffer_info->dma) {
1062 if (buffer_info->mapped_as_page)
1063 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1064 buffer_info->length, DMA_TO_DEVICE);
1066 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1067 buffer_info->length, DMA_TO_DEVICE);
1068 buffer_info->dma = 0;
1070 if (buffer_info->skb) {
1072 dev_kfree_skb_any(buffer_info->skb);
1074 dev_consume_skb_any(buffer_info->skb);
1075 buffer_info->skb = NULL;
1077 buffer_info->time_stamp = 0;
1080 static void e1000_print_hw_hang(struct work_struct *work)
1082 struct e1000_adapter *adapter = container_of(work,
1083 struct e1000_adapter,
1085 struct net_device *netdev = adapter->netdev;
1086 struct e1000_ring *tx_ring = adapter->tx_ring;
1087 unsigned int i = tx_ring->next_to_clean;
1088 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1089 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1090 struct e1000_hw *hw = &adapter->hw;
1091 u16 phy_status, phy_1000t_status, phy_ext_status;
1094 if (test_bit(__E1000_DOWN, &adapter->state))
1097 if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1098 /* May be block on write-back, flush and detect again
1099 * flush pending descriptor writebacks to memory
1101 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1102 /* execute the writes immediately */
1104 /* Due to rare timing issues, write to TIDV again to ensure
1105 * the write is successful
1107 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1108 /* execute the writes immediately */
1110 adapter->tx_hang_recheck = true;
1113 adapter->tx_hang_recheck = false;
1115 if (er32(TDH(0)) == er32(TDT(0))) {
1116 e_dbg("false hang detected, ignoring\n");
1120 /* Real hang detected */
1121 netif_stop_queue(netdev);
1123 e1e_rphy(hw, MII_BMSR, &phy_status);
1124 e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1125 e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1127 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1129 /* detected Hardware unit hang */
1130 e_err("Detected Hardware Unit Hang:\n"
1133 " next_to_use <%x>\n"
1134 " next_to_clean <%x>\n"
1135 "buffer_info[next_to_clean]:\n"
1136 " time_stamp <%lx>\n"
1137 " next_to_watch <%x>\n"
1139 " next_to_watch.status <%x>\n"
1142 "PHY 1000BASE-T Status <%x>\n"
1143 "PHY Extended Status <%x>\n"
1144 "PCI Status <%x>\n",
1145 readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1146 tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1147 eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1148 phy_status, phy_1000t_status, phy_ext_status, pci_status);
1150 e1000e_dump(adapter);
1152 /* Suggest workaround for known h/w issue */
1153 if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1154 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1158 * e1000e_tx_hwtstamp_work - check for Tx time stamp
1159 * @work: pointer to work struct
1161 * This work function polls the TSYNCTXCTL valid bit to determine when a
1162 * timestamp has been taken for the current stored skb. The timestamp must
1163 * be for this skb because only one such packet is allowed in the queue.
1165 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1167 struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1169 struct e1000_hw *hw = &adapter->hw;
1171 if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1172 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1173 struct skb_shared_hwtstamps shhwtstamps;
1176 txstmp = er32(TXSTMPL);
1177 txstmp |= (u64)er32(TXSTMPH) << 32;
1179 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1181 /* Clear the global tx_hwtstamp_skb pointer and force writes
1182 * prior to notifying the stack of a Tx timestamp.
1184 adapter->tx_hwtstamp_skb = NULL;
1185 wmb(); /* force write prior to skb_tstamp_tx */
1187 skb_tstamp_tx(skb, &shhwtstamps);
1188 dev_consume_skb_any(skb);
1189 } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1190 + adapter->tx_timeout_factor * HZ)) {
1191 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1192 adapter->tx_hwtstamp_skb = NULL;
1193 adapter->tx_hwtstamp_timeouts++;
1194 e_warn("clearing Tx timestamp hang\n");
1196 /* reschedule to check later */
1197 schedule_work(&adapter->tx_hwtstamp_work);
1202 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1203 * @tx_ring: Tx descriptor ring
1205 * the return value indicates whether actual cleaning was done, there
1206 * is no guarantee that everything was cleaned
1208 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1210 struct e1000_adapter *adapter = tx_ring->adapter;
1211 struct net_device *netdev = adapter->netdev;
1212 struct e1000_hw *hw = &adapter->hw;
1213 struct e1000_tx_desc *tx_desc, *eop_desc;
1214 struct e1000_buffer *buffer_info;
1215 unsigned int i, eop;
1216 unsigned int count = 0;
1217 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1218 unsigned int bytes_compl = 0, pkts_compl = 0;
1220 i = tx_ring->next_to_clean;
1221 eop = tx_ring->buffer_info[i].next_to_watch;
1222 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1224 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1225 (count < tx_ring->count)) {
1226 bool cleaned = false;
1228 dma_rmb(); /* read buffer_info after eop_desc */
1229 for (; !cleaned; count++) {
1230 tx_desc = E1000_TX_DESC(*tx_ring, i);
1231 buffer_info = &tx_ring->buffer_info[i];
1232 cleaned = (i == eop);
1235 total_tx_packets += buffer_info->segs;
1236 total_tx_bytes += buffer_info->bytecount;
1237 if (buffer_info->skb) {
1238 bytes_compl += buffer_info->skb->len;
1243 e1000_put_txbuf(tx_ring, buffer_info, false);
1244 tx_desc->upper.data = 0;
1247 if (i == tx_ring->count)
1251 if (i == tx_ring->next_to_use)
1253 eop = tx_ring->buffer_info[i].next_to_watch;
1254 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1257 tx_ring->next_to_clean = i;
1259 netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1261 #define TX_WAKE_THRESHOLD 32
1262 if (count && netif_carrier_ok(netdev) &&
1263 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1264 /* Make sure that anybody stopping the queue after this
1265 * sees the new next_to_clean.
1269 if (netif_queue_stopped(netdev) &&
1270 !(test_bit(__E1000_DOWN, &adapter->state))) {
1271 netif_wake_queue(netdev);
1272 ++adapter->restart_queue;
1276 if (adapter->detect_tx_hung) {
1277 /* Detect a transmit hang in hardware, this serializes the
1278 * check with the clearing of time_stamp and movement of i
1280 adapter->detect_tx_hung = false;
1281 if (tx_ring->buffer_info[i].time_stamp &&
1282 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1283 + (adapter->tx_timeout_factor * HZ)) &&
1284 !(er32(STATUS) & E1000_STATUS_TXOFF))
1285 schedule_work(&adapter->print_hang_task);
1287 adapter->tx_hang_recheck = false;
1289 adapter->total_tx_bytes += total_tx_bytes;
1290 adapter->total_tx_packets += total_tx_packets;
1291 return count < tx_ring->count;
1295 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1296 * @rx_ring: Rx descriptor ring
1298 * the return value indicates whether actual cleaning was done, there
1299 * is no guarantee that everything was cleaned
1301 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1304 struct e1000_adapter *adapter = rx_ring->adapter;
1305 struct e1000_hw *hw = &adapter->hw;
1306 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1307 struct net_device *netdev = adapter->netdev;
1308 struct pci_dev *pdev = adapter->pdev;
1309 struct e1000_buffer *buffer_info, *next_buffer;
1310 struct e1000_ps_page *ps_page;
1311 struct sk_buff *skb;
1313 u32 length, staterr;
1314 int cleaned_count = 0;
1315 bool cleaned = false;
1316 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1318 i = rx_ring->next_to_clean;
1319 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1320 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1321 buffer_info = &rx_ring->buffer_info[i];
1323 while (staterr & E1000_RXD_STAT_DD) {
1324 if (*work_done >= work_to_do)
1327 skb = buffer_info->skb;
1328 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1330 /* in the packet split case this is header only */
1331 prefetch(skb->data - NET_IP_ALIGN);
1334 if (i == rx_ring->count)
1336 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1339 next_buffer = &rx_ring->buffer_info[i];
1343 dma_unmap_single(&pdev->dev, buffer_info->dma,
1344 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1345 buffer_info->dma = 0;
1347 /* see !EOP comment in other Rx routine */
1348 if (!(staterr & E1000_RXD_STAT_EOP))
1349 adapter->flags2 |= FLAG2_IS_DISCARDING;
1351 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1352 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1353 dev_kfree_skb_irq(skb);
1354 if (staterr & E1000_RXD_STAT_EOP)
1355 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1359 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1360 !(netdev->features & NETIF_F_RXALL))) {
1361 dev_kfree_skb_irq(skb);
1365 length = le16_to_cpu(rx_desc->wb.middle.length0);
1368 e_dbg("Last part of the packet spanning multiple descriptors\n");
1369 dev_kfree_skb_irq(skb);
1374 skb_put(skb, length);
1377 /* this looks ugly, but it seems compiler issues make
1378 * it more efficient than reusing j
1380 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1382 /* page alloc/put takes too long and effects small
1383 * packet throughput, so unsplit small packets and
1384 * save the alloc/put only valid in softirq (napi)
1385 * context to call kmap_*
1387 if (l1 && (l1 <= copybreak) &&
1388 ((length + l1) <= adapter->rx_ps_bsize0)) {
1391 ps_page = &buffer_info->ps_pages[0];
1393 /* there is no documentation about how to call
1394 * kmap_atomic, so we can't hold the mapping
1397 dma_sync_single_for_cpu(&pdev->dev,
1401 vaddr = kmap_atomic(ps_page->page);
1402 memcpy(skb_tail_pointer(skb), vaddr, l1);
1403 kunmap_atomic(vaddr);
1404 dma_sync_single_for_device(&pdev->dev,
1409 /* remove the CRC */
1410 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1411 if (!(netdev->features & NETIF_F_RXFCS))
1420 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1421 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1425 ps_page = &buffer_info->ps_pages[j];
1426 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1429 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1430 ps_page->page = NULL;
1432 skb->data_len += length;
1433 skb->truesize += PAGE_SIZE;
1436 /* strip the ethernet crc, problem is we're using pages now so
1437 * this whole operation can get a little cpu intensive
1439 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1440 if (!(netdev->features & NETIF_F_RXFCS))
1441 pskb_trim(skb, skb->len - 4);
1445 total_rx_bytes += skb->len;
1448 e1000_rx_checksum(adapter, staterr, skb);
1450 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1452 if (rx_desc->wb.upper.header_status &
1453 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1454 adapter->rx_hdr_split++;
1456 e1000_receive_skb(adapter, netdev, skb, staterr,
1457 rx_desc->wb.middle.vlan);
1460 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1461 buffer_info->skb = NULL;
1463 /* return some buffers to hardware, one at a time is too slow */
1464 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1465 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1470 /* use prefetched values */
1472 buffer_info = next_buffer;
1474 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1476 rx_ring->next_to_clean = i;
1478 cleaned_count = e1000_desc_unused(rx_ring);
1480 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1482 adapter->total_rx_bytes += total_rx_bytes;
1483 adapter->total_rx_packets += total_rx_packets;
1488 * e1000_consume_page - helper function
1490 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1495 skb->data_len += length;
1496 skb->truesize += PAGE_SIZE;
1500 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1501 * @adapter: board private structure
1503 * the return value indicates whether actual cleaning was done, there
1504 * is no guarantee that everything was cleaned
1506 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1509 struct e1000_adapter *adapter = rx_ring->adapter;
1510 struct net_device *netdev = adapter->netdev;
1511 struct pci_dev *pdev = adapter->pdev;
1512 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1513 struct e1000_buffer *buffer_info, *next_buffer;
1514 u32 length, staterr;
1516 int cleaned_count = 0;
1517 bool cleaned = false;
1518 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1519 struct skb_shared_info *shinfo;
1521 i = rx_ring->next_to_clean;
1522 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1523 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1524 buffer_info = &rx_ring->buffer_info[i];
1526 while (staterr & E1000_RXD_STAT_DD) {
1527 struct sk_buff *skb;
1529 if (*work_done >= work_to_do)
1532 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
1534 skb = buffer_info->skb;
1535 buffer_info->skb = NULL;
1538 if (i == rx_ring->count)
1540 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1543 next_buffer = &rx_ring->buffer_info[i];
1547 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1549 buffer_info->dma = 0;
1551 length = le16_to_cpu(rx_desc->wb.upper.length);
1553 /* errors is only valid for DD + EOP descriptors */
1554 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1555 ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1556 !(netdev->features & NETIF_F_RXALL)))) {
1557 /* recycle both page and skb */
1558 buffer_info->skb = skb;
1559 /* an error means any chain goes out the window too */
1560 if (rx_ring->rx_skb_top)
1561 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1562 rx_ring->rx_skb_top = NULL;
1565 #define rxtop (rx_ring->rx_skb_top)
1566 if (!(staterr & E1000_RXD_STAT_EOP)) {
1567 /* this descriptor is only the beginning (or middle) */
1569 /* this is the beginning of a chain */
1571 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1574 /* this is the middle of a chain */
1575 shinfo = skb_shinfo(rxtop);
1576 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1577 buffer_info->page, 0,
1579 /* re-use the skb, only consumed the page */
1580 buffer_info->skb = skb;
1582 e1000_consume_page(buffer_info, rxtop, length);
1586 /* end of the chain */
1587 shinfo = skb_shinfo(rxtop);
1588 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1589 buffer_info->page, 0,
1591 /* re-use the current skb, we only consumed the
1594 buffer_info->skb = skb;
1597 e1000_consume_page(buffer_info, skb, length);
1599 /* no chain, got EOP, this buf is the packet
1600 * copybreak to save the put_page/alloc_page
1602 if (length <= copybreak &&
1603 skb_tailroom(skb) >= length) {
1605 vaddr = kmap_atomic(buffer_info->page);
1606 memcpy(skb_tail_pointer(skb), vaddr,
1608 kunmap_atomic(vaddr);
1609 /* re-use the page, so don't erase
1612 skb_put(skb, length);
1614 skb_fill_page_desc(skb, 0,
1615 buffer_info->page, 0,
1617 e1000_consume_page(buffer_info, skb,
1623 /* Receive Checksum Offload */
1624 e1000_rx_checksum(adapter, staterr, skb);
1626 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1628 /* probably a little skewed due to removing CRC */
1629 total_rx_bytes += skb->len;
1632 /* eth type trans needs skb->data to point to something */
1633 if (!pskb_may_pull(skb, ETH_HLEN)) {
1634 e_err("pskb_may_pull failed.\n");
1635 dev_kfree_skb_irq(skb);
1639 e1000_receive_skb(adapter, netdev, skb, staterr,
1640 rx_desc->wb.upper.vlan);
1643 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1645 /* return some buffers to hardware, one at a time is too slow */
1646 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1647 adapter->alloc_rx_buf(rx_ring, cleaned_count,
1652 /* use prefetched values */
1654 buffer_info = next_buffer;
1656 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1658 rx_ring->next_to_clean = i;
1660 cleaned_count = e1000_desc_unused(rx_ring);
1662 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1664 adapter->total_rx_bytes += total_rx_bytes;
1665 adapter->total_rx_packets += total_rx_packets;
1670 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1671 * @rx_ring: Rx descriptor ring
1673 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1675 struct e1000_adapter *adapter = rx_ring->adapter;
1676 struct e1000_buffer *buffer_info;
1677 struct e1000_ps_page *ps_page;
1678 struct pci_dev *pdev = adapter->pdev;
1681 /* Free all the Rx ring sk_buffs */
1682 for (i = 0; i < rx_ring->count; i++) {
1683 buffer_info = &rx_ring->buffer_info[i];
1684 if (buffer_info->dma) {
1685 if (adapter->clean_rx == e1000_clean_rx_irq)
1686 dma_unmap_single(&pdev->dev, buffer_info->dma,
1687 adapter->rx_buffer_len,
1689 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1690 dma_unmap_page(&pdev->dev, buffer_info->dma,
1691 PAGE_SIZE, DMA_FROM_DEVICE);
1692 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1693 dma_unmap_single(&pdev->dev, buffer_info->dma,
1694 adapter->rx_ps_bsize0,
1696 buffer_info->dma = 0;
1699 if (buffer_info->page) {
1700 put_page(buffer_info->page);
1701 buffer_info->page = NULL;
1704 if (buffer_info->skb) {
1705 dev_kfree_skb(buffer_info->skb);
1706 buffer_info->skb = NULL;
1709 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1710 ps_page = &buffer_info->ps_pages[j];
1713 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1716 put_page(ps_page->page);
1717 ps_page->page = NULL;
1721 /* there also may be some cached data from a chained receive */
1722 if (rx_ring->rx_skb_top) {
1723 dev_kfree_skb(rx_ring->rx_skb_top);
1724 rx_ring->rx_skb_top = NULL;
1727 /* Zero out the descriptor ring */
1728 memset(rx_ring->desc, 0, rx_ring->size);
1730 rx_ring->next_to_clean = 0;
1731 rx_ring->next_to_use = 0;
1732 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1735 static void e1000e_downshift_workaround(struct work_struct *work)
1737 struct e1000_adapter *adapter = container_of(work,
1738 struct e1000_adapter,
1741 if (test_bit(__E1000_DOWN, &adapter->state))
1744 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1748 * e1000_intr_msi - Interrupt Handler
1749 * @irq: interrupt number
1750 * @data: pointer to a network interface device structure
1752 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1754 struct net_device *netdev = data;
1755 struct e1000_adapter *adapter = netdev_priv(netdev);
1756 struct e1000_hw *hw = &adapter->hw;
1757 u32 icr = er32(ICR);
1759 /* read ICR disables interrupts using IAM */
1760 if (icr & E1000_ICR_LSC) {
1761 hw->mac.get_link_status = true;
1762 /* ICH8 workaround-- Call gig speed drop workaround on cable
1763 * disconnect (LSC) before accessing any PHY registers
1765 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1766 (!(er32(STATUS) & E1000_STATUS_LU)))
1767 schedule_work(&adapter->downshift_task);
1769 /* 80003ES2LAN workaround-- For packet buffer work-around on
1770 * link down event; disable receives here in the ISR and reset
1771 * adapter in watchdog
1773 if (netif_carrier_ok(netdev) &&
1774 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1775 /* disable receives */
1776 u32 rctl = er32(RCTL);
1778 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1779 adapter->flags |= FLAG_RESTART_NOW;
1781 /* guard against interrupt when we're going down */
1782 if (!test_bit(__E1000_DOWN, &adapter->state))
1783 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1786 /* Reset on uncorrectable ECC error */
1787 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1788 u32 pbeccsts = er32(PBECCSTS);
1790 adapter->corr_errors +=
1791 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1792 adapter->uncorr_errors +=
1793 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1794 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1796 /* Do the reset outside of interrupt context */
1797 schedule_work(&adapter->reset_task);
1799 /* return immediately since reset is imminent */
1803 if (napi_schedule_prep(&adapter->napi)) {
1804 adapter->total_tx_bytes = 0;
1805 adapter->total_tx_packets = 0;
1806 adapter->total_rx_bytes = 0;
1807 adapter->total_rx_packets = 0;
1808 __napi_schedule(&adapter->napi);
1815 * e1000_intr - Interrupt Handler
1816 * @irq: interrupt number
1817 * @data: pointer to a network interface device structure
1819 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1821 struct net_device *netdev = data;
1822 struct e1000_adapter *adapter = netdev_priv(netdev);
1823 struct e1000_hw *hw = &adapter->hw;
1824 u32 rctl, icr = er32(ICR);
1826 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1827 return IRQ_NONE; /* Not our interrupt */
1829 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1830 * not set, then the adapter didn't send an interrupt
1832 if (!(icr & E1000_ICR_INT_ASSERTED))
1835 /* Interrupt Auto-Mask...upon reading ICR,
1836 * interrupts are masked. No need for the
1840 if (icr & E1000_ICR_LSC) {
1841 hw->mac.get_link_status = true;
1842 /* ICH8 workaround-- Call gig speed drop workaround on cable
1843 * disconnect (LSC) before accessing any PHY registers
1845 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1846 (!(er32(STATUS) & E1000_STATUS_LU)))
1847 schedule_work(&adapter->downshift_task);
1849 /* 80003ES2LAN workaround--
1850 * For packet buffer work-around on link down event;
1851 * disable receives here in the ISR and
1852 * reset adapter in watchdog
1854 if (netif_carrier_ok(netdev) &&
1855 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1856 /* disable receives */
1858 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1859 adapter->flags |= FLAG_RESTART_NOW;
1861 /* guard against interrupt when we're going down */
1862 if (!test_bit(__E1000_DOWN, &adapter->state))
1863 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1866 /* Reset on uncorrectable ECC error */
1867 if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1868 u32 pbeccsts = er32(PBECCSTS);
1870 adapter->corr_errors +=
1871 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1872 adapter->uncorr_errors +=
1873 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1874 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1876 /* Do the reset outside of interrupt context */
1877 schedule_work(&adapter->reset_task);
1879 /* return immediately since reset is imminent */
1883 if (napi_schedule_prep(&adapter->napi)) {
1884 adapter->total_tx_bytes = 0;
1885 adapter->total_tx_packets = 0;
1886 adapter->total_rx_bytes = 0;
1887 adapter->total_rx_packets = 0;
1888 __napi_schedule(&adapter->napi);
1894 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1896 struct net_device *netdev = data;
1897 struct e1000_adapter *adapter = netdev_priv(netdev);
1898 struct e1000_hw *hw = &adapter->hw;
1899 u32 icr = er32(ICR);
1901 if (icr & adapter->eiac_mask)
1902 ew32(ICS, (icr & adapter->eiac_mask));
1904 if (icr & E1000_ICR_LSC) {
1905 hw->mac.get_link_status = true;
1906 /* guard against interrupt when we're going down */
1907 if (!test_bit(__E1000_DOWN, &adapter->state))
1908 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1911 if (!test_bit(__E1000_DOWN, &adapter->state))
1912 ew32(IMS, E1000_IMS_OTHER | IMS_OTHER_MASK);
1917 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1919 struct net_device *netdev = data;
1920 struct e1000_adapter *adapter = netdev_priv(netdev);
1921 struct e1000_hw *hw = &adapter->hw;
1922 struct e1000_ring *tx_ring = adapter->tx_ring;
1924 adapter->total_tx_bytes = 0;
1925 adapter->total_tx_packets = 0;
1927 if (!e1000_clean_tx_irq(tx_ring))
1928 /* Ring was not completely cleaned, so fire another interrupt */
1929 ew32(ICS, tx_ring->ims_val);
1931 if (!test_bit(__E1000_DOWN, &adapter->state))
1932 ew32(IMS, adapter->tx_ring->ims_val);
1937 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1939 struct net_device *netdev = data;
1940 struct e1000_adapter *adapter = netdev_priv(netdev);
1941 struct e1000_ring *rx_ring = adapter->rx_ring;
1943 /* Write the ITR value calculated at the end of the
1944 * previous interrupt.
1946 if (rx_ring->set_itr) {
1947 u32 itr = rx_ring->itr_val ?
1948 1000000000 / (rx_ring->itr_val * 256) : 0;
1950 writel(itr, rx_ring->itr_register);
1951 rx_ring->set_itr = 0;
1954 if (napi_schedule_prep(&adapter->napi)) {
1955 adapter->total_rx_bytes = 0;
1956 adapter->total_rx_packets = 0;
1957 __napi_schedule(&adapter->napi);
1963 * e1000_configure_msix - Configure MSI-X hardware
1965 * e1000_configure_msix sets up the hardware to properly
1966 * generate MSI-X interrupts.
1968 static void e1000_configure_msix(struct e1000_adapter *adapter)
1970 struct e1000_hw *hw = &adapter->hw;
1971 struct e1000_ring *rx_ring = adapter->rx_ring;
1972 struct e1000_ring *tx_ring = adapter->tx_ring;
1974 u32 ctrl_ext, ivar = 0;
1976 adapter->eiac_mask = 0;
1978 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1979 if (hw->mac.type == e1000_82574) {
1980 u32 rfctl = er32(RFCTL);
1982 rfctl |= E1000_RFCTL_ACK_DIS;
1986 /* Configure Rx vector */
1987 rx_ring->ims_val = E1000_IMS_RXQ0;
1988 adapter->eiac_mask |= rx_ring->ims_val;
1989 if (rx_ring->itr_val)
1990 writel(1000000000 / (rx_ring->itr_val * 256),
1991 rx_ring->itr_register);
1993 writel(1, rx_ring->itr_register);
1994 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1996 /* Configure Tx vector */
1997 tx_ring->ims_val = E1000_IMS_TXQ0;
1999 if (tx_ring->itr_val)
2000 writel(1000000000 / (tx_ring->itr_val * 256),
2001 tx_ring->itr_register);
2003 writel(1, tx_ring->itr_register);
2004 adapter->eiac_mask |= tx_ring->ims_val;
2005 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2007 /* set vector for Other Causes, e.g. link changes */
2009 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2010 if (rx_ring->itr_val)
2011 writel(1000000000 / (rx_ring->itr_val * 256),
2012 hw->hw_addr + E1000_EITR_82574(vector));
2014 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2016 /* Cause Tx interrupts on every write back */
2021 /* enable MSI-X PBA support */
2022 ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2023 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2024 ew32(CTRL_EXT, ctrl_ext);
2028 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2030 if (adapter->msix_entries) {
2031 pci_disable_msix(adapter->pdev);
2032 kfree(adapter->msix_entries);
2033 adapter->msix_entries = NULL;
2034 } else if (adapter->flags & FLAG_MSI_ENABLED) {
2035 pci_disable_msi(adapter->pdev);
2036 adapter->flags &= ~FLAG_MSI_ENABLED;
2041 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2043 * Attempt to configure interrupts using the best available
2044 * capabilities of the hardware and kernel.
2046 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2051 switch (adapter->int_mode) {
2052 case E1000E_INT_MODE_MSIX:
2053 if (adapter->flags & FLAG_HAS_MSIX) {
2054 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2055 adapter->msix_entries = kcalloc(adapter->num_vectors,
2059 if (adapter->msix_entries) {
2060 struct e1000_adapter *a = adapter;
2062 for (i = 0; i < adapter->num_vectors; i++)
2063 adapter->msix_entries[i].entry = i;
2065 err = pci_enable_msix_range(a->pdev,
2072 /* MSI-X failed, so fall through and try MSI */
2073 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
2074 e1000e_reset_interrupt_capability(adapter);
2076 adapter->int_mode = E1000E_INT_MODE_MSI;
2078 case E1000E_INT_MODE_MSI:
2079 if (!pci_enable_msi(adapter->pdev)) {
2080 adapter->flags |= FLAG_MSI_ENABLED;
2082 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2083 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
2086 case E1000E_INT_MODE_LEGACY:
2087 /* Don't do anything; this is the system default */
2091 /* store the number of vectors being used */
2092 adapter->num_vectors = 1;
2096 * e1000_request_msix - Initialize MSI-X interrupts
2098 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2101 static int e1000_request_msix(struct e1000_adapter *adapter)
2103 struct net_device *netdev = adapter->netdev;
2104 int err = 0, vector = 0;
2106 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2107 snprintf(adapter->rx_ring->name,
2108 sizeof(adapter->rx_ring->name) - 1,
2109 "%s-rx-0", netdev->name);
2111 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2112 err = request_irq(adapter->msix_entries[vector].vector,
2113 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2117 adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2118 E1000_EITR_82574(vector);
2119 adapter->rx_ring->itr_val = adapter->itr;
2122 if (strlen(netdev->name) < (IFNAMSIZ - 5))
2123 snprintf(adapter->tx_ring->name,
2124 sizeof(adapter->tx_ring->name) - 1,
2125 "%s-tx-0", netdev->name);
2127 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2128 err = request_irq(adapter->msix_entries[vector].vector,
2129 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2133 adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2134 E1000_EITR_82574(vector);
2135 adapter->tx_ring->itr_val = adapter->itr;
2138 err = request_irq(adapter->msix_entries[vector].vector,
2139 e1000_msix_other, 0, netdev->name, netdev);
2143 e1000_configure_msix(adapter);
2149 * e1000_request_irq - initialize interrupts
2151 * Attempts to configure interrupts using the best available
2152 * capabilities of the hardware and kernel.
2154 static int e1000_request_irq(struct e1000_adapter *adapter)
2156 struct net_device *netdev = adapter->netdev;
2159 if (adapter->msix_entries) {
2160 err = e1000_request_msix(adapter);
2163 /* fall back to MSI */
2164 e1000e_reset_interrupt_capability(adapter);
2165 adapter->int_mode = E1000E_INT_MODE_MSI;
2166 e1000e_set_interrupt_capability(adapter);
2168 if (adapter->flags & FLAG_MSI_ENABLED) {
2169 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2170 netdev->name, netdev);
2174 /* fall back to legacy interrupt */
2175 e1000e_reset_interrupt_capability(adapter);
2176 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2179 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2180 netdev->name, netdev);
2182 e_err("Unable to allocate interrupt, Error: %d\n", err);
2187 static void e1000_free_irq(struct e1000_adapter *adapter)
2189 struct net_device *netdev = adapter->netdev;
2191 if (adapter->msix_entries) {
2194 free_irq(adapter->msix_entries[vector].vector, netdev);
2197 free_irq(adapter->msix_entries[vector].vector, netdev);
2200 /* Other Causes interrupt vector */
2201 free_irq(adapter->msix_entries[vector].vector, netdev);
2205 free_irq(adapter->pdev->irq, netdev);
2209 * e1000_irq_disable - Mask off interrupt generation on the NIC
2211 static void e1000_irq_disable(struct e1000_adapter *adapter)
2213 struct e1000_hw *hw = &adapter->hw;
2216 if (adapter->msix_entries)
2217 ew32(EIAC_82574, 0);
2220 if (adapter->msix_entries) {
2223 for (i = 0; i < adapter->num_vectors; i++)
2224 synchronize_irq(adapter->msix_entries[i].vector);
2226 synchronize_irq(adapter->pdev->irq);
2231 * e1000_irq_enable - Enable default interrupt generation settings
2233 static void e1000_irq_enable(struct e1000_adapter *adapter)
2235 struct e1000_hw *hw = &adapter->hw;
2237 if (adapter->msix_entries) {
2238 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2239 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER |
2241 } else if (hw->mac.type >= e1000_pch_lpt) {
2242 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2244 ew32(IMS, IMS_ENABLE_MASK);
2250 * e1000e_get_hw_control - get control of the h/w from f/w
2251 * @adapter: address of board private structure
2253 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2254 * For ASF and Pass Through versions of f/w this means that
2255 * the driver is loaded. For AMT version (only with 82573)
2256 * of the f/w this means that the network i/f is open.
2258 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2260 struct e1000_hw *hw = &adapter->hw;
2264 /* Let firmware know the driver has taken over */
2265 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2267 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2268 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2269 ctrl_ext = er32(CTRL_EXT);
2270 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2275 * e1000e_release_hw_control - release control of the h/w to f/w
2276 * @adapter: address of board private structure
2278 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2279 * For ASF and Pass Through versions of f/w this means that the
2280 * driver is no longer loaded. For AMT version (only with 82573) i
2281 * of the f/w this means that the network i/f is closed.
2284 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2286 struct e1000_hw *hw = &adapter->hw;
2290 /* Let firmware taken over control of h/w */
2291 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2293 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2294 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2295 ctrl_ext = er32(CTRL_EXT);
2296 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2301 * e1000_alloc_ring_dma - allocate memory for a ring structure
2303 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2304 struct e1000_ring *ring)
2306 struct pci_dev *pdev = adapter->pdev;
2308 ring->desc = dma_zalloc_coherent(&pdev->dev, ring->size, &ring->dma,
2317 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2318 * @tx_ring: Tx descriptor ring
2320 * Return 0 on success, negative on failure
2322 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2324 struct e1000_adapter *adapter = tx_ring->adapter;
2325 int err = -ENOMEM, size;
2327 size = sizeof(struct e1000_buffer) * tx_ring->count;
2328 tx_ring->buffer_info = vzalloc(size);
2329 if (!tx_ring->buffer_info)
2332 /* round up to nearest 4K */
2333 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2334 tx_ring->size = ALIGN(tx_ring->size, 4096);
2336 err = e1000_alloc_ring_dma(adapter, tx_ring);
2340 tx_ring->next_to_use = 0;
2341 tx_ring->next_to_clean = 0;
2345 vfree(tx_ring->buffer_info);
2346 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2351 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2352 * @rx_ring: Rx descriptor ring
2354 * Returns 0 on success, negative on failure
2356 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2358 struct e1000_adapter *adapter = rx_ring->adapter;
2359 struct e1000_buffer *buffer_info;
2360 int i, size, desc_len, err = -ENOMEM;
2362 size = sizeof(struct e1000_buffer) * rx_ring->count;
2363 rx_ring->buffer_info = vzalloc(size);
2364 if (!rx_ring->buffer_info)
2367 for (i = 0; i < rx_ring->count; i++) {
2368 buffer_info = &rx_ring->buffer_info[i];
2369 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2370 sizeof(struct e1000_ps_page),
2372 if (!buffer_info->ps_pages)
2376 desc_len = sizeof(union e1000_rx_desc_packet_split);
2378 /* Round up to nearest 4K */
2379 rx_ring->size = rx_ring->count * desc_len;
2380 rx_ring->size = ALIGN(rx_ring->size, 4096);
2382 err = e1000_alloc_ring_dma(adapter, rx_ring);
2386 rx_ring->next_to_clean = 0;
2387 rx_ring->next_to_use = 0;
2388 rx_ring->rx_skb_top = NULL;
2393 for (i = 0; i < rx_ring->count; i++) {
2394 buffer_info = &rx_ring->buffer_info[i];
2395 kfree(buffer_info->ps_pages);
2398 vfree(rx_ring->buffer_info);
2399 e_err("Unable to allocate memory for the receive descriptor ring\n");
2404 * e1000_clean_tx_ring - Free Tx Buffers
2405 * @tx_ring: Tx descriptor ring
2407 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2409 struct e1000_adapter *adapter = tx_ring->adapter;
2410 struct e1000_buffer *buffer_info;
2414 for (i = 0; i < tx_ring->count; i++) {
2415 buffer_info = &tx_ring->buffer_info[i];
2416 e1000_put_txbuf(tx_ring, buffer_info, false);
2419 netdev_reset_queue(adapter->netdev);
2420 size = sizeof(struct e1000_buffer) * tx_ring->count;
2421 memset(tx_ring->buffer_info, 0, size);
2423 memset(tx_ring->desc, 0, tx_ring->size);
2425 tx_ring->next_to_use = 0;
2426 tx_ring->next_to_clean = 0;
2430 * e1000e_free_tx_resources - Free Tx Resources per Queue
2431 * @tx_ring: Tx descriptor ring
2433 * Free all transmit software resources
2435 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2437 struct e1000_adapter *adapter = tx_ring->adapter;
2438 struct pci_dev *pdev = adapter->pdev;
2440 e1000_clean_tx_ring(tx_ring);
2442 vfree(tx_ring->buffer_info);
2443 tx_ring->buffer_info = NULL;
2445 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2447 tx_ring->desc = NULL;
2451 * e1000e_free_rx_resources - Free Rx Resources
2452 * @rx_ring: Rx descriptor ring
2454 * Free all receive software resources
2456 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2458 struct e1000_adapter *adapter = rx_ring->adapter;
2459 struct pci_dev *pdev = adapter->pdev;
2462 e1000_clean_rx_ring(rx_ring);
2464 for (i = 0; i < rx_ring->count; i++)
2465 kfree(rx_ring->buffer_info[i].ps_pages);
2467 vfree(rx_ring->buffer_info);
2468 rx_ring->buffer_info = NULL;
2470 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2472 rx_ring->desc = NULL;
2476 * e1000_update_itr - update the dynamic ITR value based on statistics
2477 * @adapter: pointer to adapter
2478 * @itr_setting: current adapter->itr
2479 * @packets: the number of packets during this measurement interval
2480 * @bytes: the number of bytes during this measurement interval
2482 * Stores a new ITR value based on packets and byte
2483 * counts during the last interrupt. The advantage of per interrupt
2484 * computation is faster updates and more accurate ITR for the current
2485 * traffic pattern. Constants in this function were computed
2486 * based on theoretical maximum wire speed and thresholds were set based
2487 * on testing data as well as attempting to minimize response time
2488 * while increasing bulk throughput. This functionality is controlled
2489 * by the InterruptThrottleRate module parameter.
2491 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2493 unsigned int retval = itr_setting;
2498 switch (itr_setting) {
2499 case lowest_latency:
2500 /* handle TSO and jumbo frames */
2501 if (bytes / packets > 8000)
2502 retval = bulk_latency;
2503 else if ((packets < 5) && (bytes > 512))
2504 retval = low_latency;
2506 case low_latency: /* 50 usec aka 20000 ints/s */
2507 if (bytes > 10000) {
2508 /* this if handles the TSO accounting */
2509 if (bytes / packets > 8000)
2510 retval = bulk_latency;
2511 else if ((packets < 10) || ((bytes / packets) > 1200))
2512 retval = bulk_latency;
2513 else if ((packets > 35))
2514 retval = lowest_latency;
2515 } else if (bytes / packets > 2000) {
2516 retval = bulk_latency;
2517 } else if (packets <= 2 && bytes < 512) {
2518 retval = lowest_latency;
2521 case bulk_latency: /* 250 usec aka 4000 ints/s */
2522 if (bytes > 25000) {
2524 retval = low_latency;
2525 } else if (bytes < 6000) {
2526 retval = low_latency;
2534 static void e1000_set_itr(struct e1000_adapter *adapter)
2537 u32 new_itr = adapter->itr;
2539 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2540 if (adapter->link_speed != SPEED_1000) {
2546 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2551 adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2552 adapter->total_tx_packets,
2553 adapter->total_tx_bytes);
2554 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2555 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2556 adapter->tx_itr = low_latency;
2558 adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2559 adapter->total_rx_packets,
2560 adapter->total_rx_bytes);
2561 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2562 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2563 adapter->rx_itr = low_latency;
2565 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2567 /* counts and packets in update_itr are dependent on these numbers */
2568 switch (current_itr) {
2569 case lowest_latency:
2573 new_itr = 20000; /* aka hwitr = ~200 */
2583 if (new_itr != adapter->itr) {
2584 /* this attempts to bias the interrupt rate towards Bulk
2585 * by adding intermediate steps when interrupt rate is
2588 new_itr = new_itr > adapter->itr ?
2589 min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2590 adapter->itr = new_itr;
2591 adapter->rx_ring->itr_val = new_itr;
2592 if (adapter->msix_entries)
2593 adapter->rx_ring->set_itr = 1;
2595 e1000e_write_itr(adapter, new_itr);
2600 * e1000e_write_itr - write the ITR value to the appropriate registers
2601 * @adapter: address of board private structure
2602 * @itr: new ITR value to program
2604 * e1000e_write_itr determines if the adapter is in MSI-X mode
2605 * and, if so, writes the EITR registers with the ITR value.
2606 * Otherwise, it writes the ITR value into the ITR register.
2608 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2610 struct e1000_hw *hw = &adapter->hw;
2611 u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2613 if (adapter->msix_entries) {
2616 for (vector = 0; vector < adapter->num_vectors; vector++)
2617 writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2624 * e1000_alloc_queues - Allocate memory for all rings
2625 * @adapter: board private structure to initialize
2627 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2629 int size = sizeof(struct e1000_ring);
2631 adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2632 if (!adapter->tx_ring)
2634 adapter->tx_ring->count = adapter->tx_ring_count;
2635 adapter->tx_ring->adapter = adapter;
2637 adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2638 if (!adapter->rx_ring)
2640 adapter->rx_ring->count = adapter->rx_ring_count;
2641 adapter->rx_ring->adapter = adapter;
2645 e_err("Unable to allocate memory for queues\n");
2646 kfree(adapter->rx_ring);
2647 kfree(adapter->tx_ring);
2652 * e1000e_poll - NAPI Rx polling callback
2653 * @napi: struct associated with this polling callback
2654 * @weight: number of packets driver is allowed to process this poll
2656 static int e1000e_poll(struct napi_struct *napi, int weight)
2658 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2660 struct e1000_hw *hw = &adapter->hw;
2661 struct net_device *poll_dev = adapter->netdev;
2662 int tx_cleaned = 1, work_done = 0;
2664 adapter = netdev_priv(poll_dev);
2666 if (!adapter->msix_entries ||
2667 (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2668 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2670 adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2675 /* If weight not fully consumed, exit the polling mode */
2676 if (work_done < weight) {
2677 if (adapter->itr_setting & 3)
2678 e1000_set_itr(adapter);
2679 napi_complete_done(napi, work_done);
2680 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2681 if (adapter->msix_entries)
2682 ew32(IMS, adapter->rx_ring->ims_val);
2684 e1000_irq_enable(adapter);
2691 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2692 __always_unused __be16 proto, u16 vid)
2694 struct e1000_adapter *adapter = netdev_priv(netdev);
2695 struct e1000_hw *hw = &adapter->hw;
2698 /* don't update vlan cookie if already programmed */
2699 if ((adapter->hw.mng_cookie.status &
2700 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2701 (vid == adapter->mng_vlan_id))
2704 /* add VID to filter table */
2705 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2706 index = (vid >> 5) & 0x7F;
2707 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2708 vfta |= BIT((vid & 0x1F));
2709 hw->mac.ops.write_vfta(hw, index, vfta);
2712 set_bit(vid, adapter->active_vlans);
2717 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2718 __always_unused __be16 proto, u16 vid)
2720 struct e1000_adapter *adapter = netdev_priv(netdev);
2721 struct e1000_hw *hw = &adapter->hw;
2724 if ((adapter->hw.mng_cookie.status &
2725 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2726 (vid == adapter->mng_vlan_id)) {
2727 /* release control to f/w */
2728 e1000e_release_hw_control(adapter);
2732 /* remove VID from filter table */
2733 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2734 index = (vid >> 5) & 0x7F;
2735 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2736 vfta &= ~BIT((vid & 0x1F));
2737 hw->mac.ops.write_vfta(hw, index, vfta);
2740 clear_bit(vid, adapter->active_vlans);
2746 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2747 * @adapter: board private structure to initialize
2749 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2751 struct net_device *netdev = adapter->netdev;
2752 struct e1000_hw *hw = &adapter->hw;
2755 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2756 /* disable VLAN receive filtering */
2758 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2761 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2762 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2763 adapter->mng_vlan_id);
2764 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2770 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2771 * @adapter: board private structure to initialize
2773 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2775 struct e1000_hw *hw = &adapter->hw;
2778 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2779 /* enable VLAN receive filtering */
2781 rctl |= E1000_RCTL_VFE;
2782 rctl &= ~E1000_RCTL_CFIEN;
2788 * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2789 * @adapter: board private structure to initialize
2791 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2793 struct e1000_hw *hw = &adapter->hw;
2796 /* disable VLAN tag insert/strip */
2798 ctrl &= ~E1000_CTRL_VME;
2803 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2804 * @adapter: board private structure to initialize
2806 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2808 struct e1000_hw *hw = &adapter->hw;
2811 /* enable VLAN tag insert/strip */
2813 ctrl |= E1000_CTRL_VME;
2817 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2819 struct net_device *netdev = adapter->netdev;
2820 u16 vid = adapter->hw.mng_cookie.vlan_id;
2821 u16 old_vid = adapter->mng_vlan_id;
2823 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2824 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2825 adapter->mng_vlan_id = vid;
2828 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2829 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2832 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2836 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2838 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2839 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2842 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2844 struct e1000_hw *hw = &adapter->hw;
2845 u32 manc, manc2h, mdef, i, j;
2847 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2852 /* enable receiving management packets to the host. this will probably
2853 * generate destination unreachable messages from the host OS, but
2854 * the packets will be handled on SMBUS
2856 manc |= E1000_MANC_EN_MNG2HOST;
2857 manc2h = er32(MANC2H);
2859 switch (hw->mac.type) {
2861 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2865 /* Check if IPMI pass-through decision filter already exists;
2868 for (i = 0, j = 0; i < 8; i++) {
2869 mdef = er32(MDEF(i));
2871 /* Ignore filters with anything other than IPMI ports */
2872 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2875 /* Enable this decision filter in MANC2H */
2882 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2885 /* Create new decision filter in an empty filter */
2886 for (i = 0, j = 0; i < 8; i++)
2887 if (er32(MDEF(i)) == 0) {
2888 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2889 E1000_MDEF_PORT_664));
2896 e_warn("Unable to create IPMI pass-through filter\n");
2900 ew32(MANC2H, manc2h);
2905 * e1000_configure_tx - Configure Transmit Unit after Reset
2906 * @adapter: board private structure
2908 * Configure the Tx unit of the MAC after a reset.
2910 static void e1000_configure_tx(struct e1000_adapter *adapter)
2912 struct e1000_hw *hw = &adapter->hw;
2913 struct e1000_ring *tx_ring = adapter->tx_ring;
2915 u32 tdlen, tctl, tarc;
2917 /* Setup the HW Tx Head and Tail descriptor pointers */
2918 tdba = tx_ring->dma;
2919 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2920 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2921 ew32(TDBAH(0), (tdba >> 32));
2922 ew32(TDLEN(0), tdlen);
2925 tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2926 tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2928 writel(0, tx_ring->head);
2929 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2930 e1000e_update_tdt_wa(tx_ring, 0);
2932 writel(0, tx_ring->tail);
2934 /* Set the Tx Interrupt Delay register */
2935 ew32(TIDV, adapter->tx_int_delay);
2936 /* Tx irq moderation */
2937 ew32(TADV, adapter->tx_abs_int_delay);
2939 if (adapter->flags2 & FLAG2_DMA_BURST) {
2940 u32 txdctl = er32(TXDCTL(0));
2942 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2943 E1000_TXDCTL_WTHRESH);
2944 /* set up some performance related parameters to encourage the
2945 * hardware to use the bus more efficiently in bursts, depends
2946 * on the tx_int_delay to be enabled,
2947 * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2948 * hthresh = 1 ==> prefetch when one or more available
2949 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2950 * BEWARE: this seems to work but should be considered first if
2951 * there are Tx hangs or other Tx related bugs
2953 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2954 ew32(TXDCTL(0), txdctl);
2956 /* erratum work around: set txdctl the same for both queues */
2957 ew32(TXDCTL(1), er32(TXDCTL(0)));
2959 /* Program the Transmit Control Register */
2961 tctl &= ~E1000_TCTL_CT;
2962 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2963 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2965 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2966 tarc = er32(TARC(0));
2967 /* set the speed mode bit, we'll clear it if we're not at
2968 * gigabit link later
2970 #define SPEED_MODE_BIT BIT(21)
2971 tarc |= SPEED_MODE_BIT;
2972 ew32(TARC(0), tarc);
2975 /* errata: program both queues to unweighted RR */
2976 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2977 tarc = er32(TARC(0));
2979 ew32(TARC(0), tarc);
2980 tarc = er32(TARC(1));
2982 ew32(TARC(1), tarc);
2985 /* Setup Transmit Descriptor Settings for eop descriptor */
2986 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2988 /* only set IDE if we are delaying interrupts using the timers */
2989 if (adapter->tx_int_delay)
2990 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2992 /* enable Report Status bit */
2993 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2997 hw->mac.ops.config_collision_dist(hw);
2999 /* SPT and KBL Si errata workaround to avoid data corruption */
3000 if (hw->mac.type == e1000_pch_spt) {
3003 reg_val = er32(IOSFPC);
3004 reg_val |= E1000_RCTL_RDMTS_HEX;
3005 ew32(IOSFPC, reg_val);
3007 reg_val = er32(TARC(0));
3008 /* SPT and KBL Si errata workaround to avoid Tx hang.
3009 * Dropping the number of outstanding requests from
3010 * 3 to 2 in order to avoid a buffer overrun.
3012 reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
3013 reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
3014 ew32(TARC(0), reg_val);
3019 * e1000_setup_rctl - configure the receive control registers
3020 * @adapter: Board private structure
3022 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3023 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3024 static void e1000_setup_rctl(struct e1000_adapter *adapter)
3026 struct e1000_hw *hw = &adapter->hw;
3030 /* Workaround Si errata on PCHx - configure jumbo frame flow.
3031 * If jumbo frames not set, program related MAC/PHY registers
3034 if (hw->mac.type >= e1000_pch2lan) {
3037 if (adapter->netdev->mtu > ETH_DATA_LEN)
3038 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3040 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3043 e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3046 /* Program MC offset vector base */
3048 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3049 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3050 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3051 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3053 /* Do not Store bad packets */
3054 rctl &= ~E1000_RCTL_SBP;
3056 /* Enable Long Packet receive */
3057 if (adapter->netdev->mtu <= ETH_DATA_LEN)
3058 rctl &= ~E1000_RCTL_LPE;
3060 rctl |= E1000_RCTL_LPE;
3062 /* Some systems expect that the CRC is included in SMBUS traffic. The
3063 * hardware strips the CRC before sending to both SMBUS (BMC) and to
3064 * host memory when this is enabled
3066 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3067 rctl |= E1000_RCTL_SECRC;
3069 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3070 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3073 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3076 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3078 e1e_rphy(hw, 22, &phy_data);
3080 phy_data |= BIT(14);
3081 e1e_wphy(hw, 0x10, 0x2823);
3082 e1e_wphy(hw, 0x11, 0x0003);
3083 e1e_wphy(hw, 22, phy_data);
3086 /* Setup buffer sizes */
3087 rctl &= ~E1000_RCTL_SZ_4096;
3088 rctl |= E1000_RCTL_BSEX;
3089 switch (adapter->rx_buffer_len) {
3092 rctl |= E1000_RCTL_SZ_2048;
3093 rctl &= ~E1000_RCTL_BSEX;
3096 rctl |= E1000_RCTL_SZ_4096;
3099 rctl |= E1000_RCTL_SZ_8192;
3102 rctl |= E1000_RCTL_SZ_16384;
3106 /* Enable Extended Status in all Receive Descriptors */
3107 rfctl = er32(RFCTL);
3108 rfctl |= E1000_RFCTL_EXTEN;
3111 /* 82571 and greater support packet-split where the protocol
3112 * header is placed in skb->data and the packet data is
3113 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3114 * In the case of a non-split, skb->data is linearly filled,
3115 * followed by the page buffers. Therefore, skb->data is
3116 * sized to hold the largest protocol header.
3118 * allocations using alloc_page take too long for regular MTU
3119 * so only enable packet split for jumbo frames
3121 * Using pages when the page size is greater than 16k wastes
3122 * a lot of memory, since we allocate 3 pages at all times
3125 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3126 if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3127 adapter->rx_ps_pages = pages;
3129 adapter->rx_ps_pages = 0;
3131 if (adapter->rx_ps_pages) {
3134 /* Enable Packet split descriptors */
3135 rctl |= E1000_RCTL_DTYP_PS;
3137 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3139 switch (adapter->rx_ps_pages) {
3141 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3144 psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3147 psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3151 ew32(PSRCTL, psrctl);
3154 /* This is useful for sniffing bad packets. */
3155 if (adapter->netdev->features & NETIF_F_RXALL) {
3156 /* UPE and MPE will be handled by normal PROMISC logic
3157 * in e1000e_set_rx_mode
3159 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3160 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3161 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3163 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3164 E1000_RCTL_DPF | /* Allow filtered pause */
3165 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3166 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3167 * and that breaks VLANs.
3172 /* just started the receive unit, no need to restart */
3173 adapter->flags &= ~FLAG_RESTART_NOW;
3177 * e1000_configure_rx - Configure Receive Unit after Reset
3178 * @adapter: board private structure
3180 * Configure the Rx unit of the MAC after a reset.
3182 static void e1000_configure_rx(struct e1000_adapter *adapter)
3184 struct e1000_hw *hw = &adapter->hw;
3185 struct e1000_ring *rx_ring = adapter->rx_ring;
3187 u32 rdlen, rctl, rxcsum, ctrl_ext;
3189 if (adapter->rx_ps_pages) {
3190 /* this is a 32 byte descriptor */
3191 rdlen = rx_ring->count *
3192 sizeof(union e1000_rx_desc_packet_split);
3193 adapter->clean_rx = e1000_clean_rx_irq_ps;
3194 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3195 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3196 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3197 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3198 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3200 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3201 adapter->clean_rx = e1000_clean_rx_irq;
3202 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3205 /* disable receives while setting up the descriptors */
3207 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3208 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3210 usleep_range(10000, 20000);
3212 if (adapter->flags2 & FLAG2_DMA_BURST) {
3213 /* set the writeback threshold (only takes effect if the RDTR
3214 * is set). set GRAN=1 and write back up to 0x4 worth, and
3215 * enable prefetching of 0x20 Rx descriptors
3221 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3222 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3225 /* set the Receive Delay Timer Register */
3226 ew32(RDTR, adapter->rx_int_delay);
3228 /* irq moderation */
3229 ew32(RADV, adapter->rx_abs_int_delay);
3230 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3231 e1000e_write_itr(adapter, adapter->itr);
3233 ctrl_ext = er32(CTRL_EXT);
3234 /* Auto-Mask interrupts upon ICR access */
3235 ctrl_ext |= E1000_CTRL_EXT_IAME;
3236 ew32(IAM, 0xffffffff);
3237 ew32(CTRL_EXT, ctrl_ext);
3240 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3241 * the Base and Length of the Rx Descriptor Ring
3243 rdba = rx_ring->dma;
3244 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3245 ew32(RDBAH(0), (rdba >> 32));
3246 ew32(RDLEN(0), rdlen);
3249 rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3250 rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3252 writel(0, rx_ring->head);
3253 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3254 e1000e_update_rdt_wa(rx_ring, 0);
3256 writel(0, rx_ring->tail);
3258 /* Enable Receive Checksum Offload for TCP and UDP */
3259 rxcsum = er32(RXCSUM);
3260 if (adapter->netdev->features & NETIF_F_RXCSUM)
3261 rxcsum |= E1000_RXCSUM_TUOFL;
3263 rxcsum &= ~E1000_RXCSUM_TUOFL;
3264 ew32(RXCSUM, rxcsum);
3266 /* With jumbo frames, excessive C-state transition latencies result
3267 * in dropped transactions.
3269 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3271 ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3272 adapter->max_frame_size) * 8 / 1000;
3274 if (adapter->flags & FLAG_IS_ICH) {
3275 u32 rxdctl = er32(RXDCTL(0));
3277 ew32(RXDCTL(0), rxdctl | 0x3 | BIT(8));
3280 dev_info(&adapter->pdev->dev,
3281 "Some CPU C-states have been disabled in order to enable jumbo frames\n");
3282 pm_qos_update_request(&adapter->pm_qos_req, lat);
3284 pm_qos_update_request(&adapter->pm_qos_req,
3285 PM_QOS_DEFAULT_VALUE);
3288 /* Enable Receives */
3293 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3294 * @netdev: network interface device structure
3296 * Writes multicast address list to the MTA hash table.
3297 * Returns: -ENOMEM on failure
3298 * 0 on no addresses written
3299 * X on writing X addresses to MTA
3301 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3303 struct e1000_adapter *adapter = netdev_priv(netdev);
3304 struct e1000_hw *hw = &adapter->hw;
3305 struct netdev_hw_addr *ha;
3309 if (netdev_mc_empty(netdev)) {
3310 /* nothing to program, so clear mc list */
3311 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3315 mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3319 /* update_mc_addr_list expects a packed array of only addresses. */
3321 netdev_for_each_mc_addr(ha, netdev)
3322 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3324 hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3327 return netdev_mc_count(netdev);
3331 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3332 * @netdev: network interface device structure
3334 * Writes unicast address list to the RAR table.
3335 * Returns: -ENOMEM on failure/insufficient address space
3336 * 0 on no addresses written
3337 * X on writing X addresses to the RAR table
3339 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3341 struct e1000_adapter *adapter = netdev_priv(netdev);
3342 struct e1000_hw *hw = &adapter->hw;
3343 unsigned int rar_entries;
3346 rar_entries = hw->mac.ops.rar_get_count(hw);
3348 /* save a rar entry for our hardware address */
3351 /* save a rar entry for the LAA workaround */
3352 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3355 /* return ENOMEM indicating insufficient memory for addresses */
3356 if (netdev_uc_count(netdev) > rar_entries)
3359 if (!netdev_uc_empty(netdev) && rar_entries) {
3360 struct netdev_hw_addr *ha;
3362 /* write the addresses in reverse order to avoid write
3365 netdev_for_each_uc_addr(ha, netdev) {
3370 ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3377 /* zero out the remaining RAR entries not used above */
3378 for (; rar_entries > 0; rar_entries--) {
3379 ew32(RAH(rar_entries), 0);
3380 ew32(RAL(rar_entries), 0);
3388 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3389 * @netdev: network interface device structure
3391 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3392 * address list or the network interface flags are updated. This routine is
3393 * responsible for configuring the hardware for proper unicast, multicast,
3394 * promiscuous mode, and all-multi behavior.
3396 static void e1000e_set_rx_mode(struct net_device *netdev)
3398 struct e1000_adapter *adapter = netdev_priv(netdev);
3399 struct e1000_hw *hw = &adapter->hw;
3402 if (pm_runtime_suspended(netdev->dev.parent))
3405 /* Check for Promiscuous and All Multicast modes */
3408 /* clear the affected bits */
3409 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3411 if (netdev->flags & IFF_PROMISC) {
3412 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3413 /* Do not hardware filter VLANs in promisc mode */
3414 e1000e_vlan_filter_disable(adapter);
3418 if (netdev->flags & IFF_ALLMULTI) {
3419 rctl |= E1000_RCTL_MPE;
3421 /* Write addresses to the MTA, if the attempt fails
3422 * then we should just turn on promiscuous mode so
3423 * that we can at least receive multicast traffic
3425 count = e1000e_write_mc_addr_list(netdev);
3427 rctl |= E1000_RCTL_MPE;
3429 e1000e_vlan_filter_enable(adapter);
3430 /* Write addresses to available RAR registers, if there is not
3431 * sufficient space to store all the addresses then enable
3432 * unicast promiscuous mode
3434 count = e1000e_write_uc_addr_list(netdev);
3436 rctl |= E1000_RCTL_UPE;
3441 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3442 e1000e_vlan_strip_enable(adapter);
3444 e1000e_vlan_strip_disable(adapter);
3447 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3449 struct e1000_hw *hw = &adapter->hw;
3454 netdev_rss_key_fill(rss_key, sizeof(rss_key));
3455 for (i = 0; i < 10; i++)
3456 ew32(RSSRK(i), rss_key[i]);
3458 /* Direct all traffic to queue 0 */
3459 for (i = 0; i < 32; i++)
3462 /* Disable raw packet checksumming so that RSS hash is placed in
3463 * descriptor on writeback.
3465 rxcsum = er32(RXCSUM);
3466 rxcsum |= E1000_RXCSUM_PCSD;
3468 ew32(RXCSUM, rxcsum);
3470 mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3471 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3472 E1000_MRQC_RSS_FIELD_IPV6 |
3473 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3474 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3480 * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3481 * @adapter: board private structure
3482 * @timinca: pointer to returned time increment attributes
3484 * Get attributes for incrementing the System Time Register SYSTIML/H at
3485 * the default base frequency, and set the cyclecounter shift value.
3487 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3489 struct e1000_hw *hw = &adapter->hw;
3490 u32 incvalue, incperiod, shift;
3492 /* Make sure clock is enabled on I217/I218/I219 before checking
3495 if ((hw->mac.type >= e1000_pch_lpt) &&
3496 !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3497 !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3498 u32 fextnvm7 = er32(FEXTNVM7);
3500 if (!(fextnvm7 & BIT(0))) {
3501 ew32(FEXTNVM7, fextnvm7 | BIT(0));
3506 switch (hw->mac.type) {
3508 /* Stable 96MHz frequency */
3509 incperiod = INCPERIOD_96MHZ;
3510 incvalue = INCVALUE_96MHZ;
3511 shift = INCVALUE_SHIFT_96MHZ;
3512 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3515 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3516 /* Stable 96MHz frequency */
3517 incperiod = INCPERIOD_96MHZ;
3518 incvalue = INCVALUE_96MHZ;
3519 shift = INCVALUE_SHIFT_96MHZ;
3520 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3522 /* Stable 25MHz frequency */
3523 incperiod = INCPERIOD_25MHZ;
3524 incvalue = INCVALUE_25MHZ;
3525 shift = INCVALUE_SHIFT_25MHZ;
3526 adapter->cc.shift = shift;
3530 /* Stable 24MHz frequency */
3531 incperiod = INCPERIOD_24MHZ;
3532 incvalue = INCVALUE_24MHZ;
3533 shift = INCVALUE_SHIFT_24MHZ;
3534 adapter->cc.shift = shift;
3537 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3538 /* Stable 24MHz frequency */
3539 incperiod = INCPERIOD_24MHZ;
3540 incvalue = INCVALUE_24MHZ;
3541 shift = INCVALUE_SHIFT_24MHZ;
3542 adapter->cc.shift = shift;
3544 /* Stable 38400KHz frequency */
3545 incperiod = INCPERIOD_38400KHZ;
3546 incvalue = INCVALUE_38400KHZ;
3547 shift = INCVALUE_SHIFT_38400KHZ;
3548 adapter->cc.shift = shift;
3553 /* Stable 25MHz frequency */
3554 incperiod = INCPERIOD_25MHZ;
3555 incvalue = INCVALUE_25MHZ;
3556 shift = INCVALUE_SHIFT_25MHZ;
3557 adapter->cc.shift = shift;
3563 *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3564 ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3570 * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3571 * @adapter: board private structure
3573 * Outgoing time stamping can be enabled and disabled. Play nice and
3574 * disable it when requested, although it shouldn't cause any overhead
3575 * when no packet needs it. At most one packet in the queue may be
3576 * marked for time stamping, otherwise it would be impossible to tell
3577 * for sure to which packet the hardware time stamp belongs.
3579 * Incoming time stamping has to be configured via the hardware filters.
3580 * Not all combinations are supported, in particular event type has to be
3581 * specified. Matching the kind of event packet is not supported, with the
3582 * exception of "all V2 events regardless of level 2 or 4".
3584 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3585 struct hwtstamp_config *config)
3587 struct e1000_hw *hw = &adapter->hw;
3588 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3589 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3596 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3599 /* flags reserved for future extensions - must be zero */
3603 switch (config->tx_type) {
3604 case HWTSTAMP_TX_OFF:
3607 case HWTSTAMP_TX_ON:
3613 switch (config->rx_filter) {
3614 case HWTSTAMP_FILTER_NONE:
3617 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3618 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3619 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3622 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3623 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3624 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3627 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3628 /* Also time stamps V2 L2 Path Delay Request/Response */
3629 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3630 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3633 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3634 /* Also time stamps V2 L2 Path Delay Request/Response. */
3635 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3636 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3639 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3640 /* Hardware cannot filter just V2 L4 Sync messages;
3641 * fall-through to V2 (both L2 and L4) Sync.
3643 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3644 /* Also time stamps V2 Path Delay Request/Response. */
3645 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3646 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3650 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3651 /* Hardware cannot filter just V2 L4 Delay Request messages;
3652 * fall-through to V2 (both L2 and L4) Delay Request.
3654 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3655 /* Also time stamps V2 Path Delay Request/Response. */
3656 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3657 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3661 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3662 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3663 /* Hardware cannot filter just V2 L4 or L2 Event messages;
3664 * fall-through to all V2 (both L2 and L4) Events.
3666 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3667 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3668 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3672 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3673 /* For V1, the hardware can only filter Sync messages or
3674 * Delay Request messages but not both so fall-through to
3675 * time stamp all packets.
3677 case HWTSTAMP_FILTER_NTP_ALL:
3678 case HWTSTAMP_FILTER_ALL:
3681 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3682 config->rx_filter = HWTSTAMP_FILTER_ALL;
3688 adapter->hwtstamp_config = *config;
3690 /* enable/disable Tx h/w time stamping */
3691 regval = er32(TSYNCTXCTL);
3692 regval &= ~E1000_TSYNCTXCTL_ENABLED;
3693 regval |= tsync_tx_ctl;
3694 ew32(TSYNCTXCTL, regval);
3695 if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3696 (regval & E1000_TSYNCTXCTL_ENABLED)) {
3697 e_err("Timesync Tx Control register not set as expected\n");
3701 /* enable/disable Rx h/w time stamping */
3702 regval = er32(TSYNCRXCTL);
3703 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3704 regval |= tsync_rx_ctl;
3705 ew32(TSYNCRXCTL, regval);
3706 if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3707 E1000_TSYNCRXCTL_TYPE_MASK)) !=
3708 (regval & (E1000_TSYNCRXCTL_ENABLED |
3709 E1000_TSYNCRXCTL_TYPE_MASK))) {
3710 e_err("Timesync Rx Control register not set as expected\n");
3714 /* L2: define ethertype filter for time stamped packets */
3716 rxmtrl |= ETH_P_1588;
3718 /* define which PTP packets get time stamped */
3719 ew32(RXMTRL, rxmtrl);
3721 /* Filter by destination port */
3723 rxudp = PTP_EV_PORT;
3724 cpu_to_be16s(&rxudp);
3730 /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3738 * e1000_configure - configure the hardware for Rx and Tx
3739 * @adapter: private board structure
3741 static void e1000_configure(struct e1000_adapter *adapter)
3743 struct e1000_ring *rx_ring = adapter->rx_ring;
3745 e1000e_set_rx_mode(adapter->netdev);
3747 e1000_restore_vlan(adapter);
3748 e1000_init_manageability_pt(adapter);
3750 e1000_configure_tx(adapter);
3752 if (adapter->netdev->features & NETIF_F_RXHASH)
3753 e1000e_setup_rss_hash(adapter);
3754 e1000_setup_rctl(adapter);
3755 e1000_configure_rx(adapter);
3756 adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3760 * e1000e_power_up_phy - restore link in case the phy was powered down
3761 * @adapter: address of board private structure
3763 * The phy may be powered down to save power and turn off link when the
3764 * driver is unloaded and wake on lan is not enabled (among others)
3765 * *** this routine MUST be followed by a call to e1000e_reset ***
3767 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3769 if (adapter->hw.phy.ops.power_up)
3770 adapter->hw.phy.ops.power_up(&adapter->hw);
3772 adapter->hw.mac.ops.setup_link(&adapter->hw);
3776 * e1000_power_down_phy - Power down the PHY
3778 * Power down the PHY so no link is implied when interface is down.
3779 * The PHY cannot be powered down if management or WoL is active.
3781 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3783 if (adapter->hw.phy.ops.power_down)
3784 adapter->hw.phy.ops.power_down(&adapter->hw);
3788 * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3790 * We want to clear all pending descriptors from the TX ring.
3791 * zeroing happens when the HW reads the regs. We assign the ring itself as
3792 * the data of the next descriptor. We don't care about the data we are about
3795 static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3797 struct e1000_hw *hw = &adapter->hw;
3798 struct e1000_ring *tx_ring = adapter->tx_ring;
3799 struct e1000_tx_desc *tx_desc = NULL;
3800 u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3804 ew32(TCTL, tctl | E1000_TCTL_EN);
3806 BUG_ON(tdt != tx_ring->next_to_use);
3807 tx_desc = E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3808 tx_desc->buffer_addr = tx_ring->dma;
3810 tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3811 tx_desc->upper.data = 0;
3812 /* flush descriptors to memory before notifying the HW */
3814 tx_ring->next_to_use++;
3815 if (tx_ring->next_to_use == tx_ring->count)
3816 tx_ring->next_to_use = 0;
3817 ew32(TDT(0), tx_ring->next_to_use);
3819 usleep_range(200, 250);
3823 * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3825 * Mark all descriptors in the RX ring as consumed and disable the rx ring
3827 static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3830 struct e1000_hw *hw = &adapter->hw;
3833 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3835 usleep_range(100, 150);
3837 rxdctl = er32(RXDCTL(0));
3838 /* zero the lower 14 bits (prefetch and host thresholds) */
3839 rxdctl &= 0xffffc000;
3841 /* update thresholds: prefetch threshold to 31, host threshold to 1
3842 * and make sure the granularity is "descriptors" and not "cache lines"
3844 rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3846 ew32(RXDCTL(0), rxdctl);
3847 /* momentarily enable the RX ring for the changes to take effect */
3848 ew32(RCTL, rctl | E1000_RCTL_EN);
3850 usleep_range(100, 150);
3851 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3855 * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3857 * In i219, the descriptor rings must be emptied before resetting the HW
3858 * or before changing the device state to D3 during runtime (runtime PM).
3860 * Failure to do this will cause the HW to enter a unit hang state which can
3861 * only be released by PCI reset on the device
3865 static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3868 u32 fext_nvm11, tdlen;
3869 struct e1000_hw *hw = &adapter->hw;
3871 /* First, disable MULR fix in FEXTNVM11 */
3872 fext_nvm11 = er32(FEXTNVM11);
3873 fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3874 ew32(FEXTNVM11, fext_nvm11);
3875 /* do nothing if we're not in faulty state, or if the queue is empty */
3876 tdlen = er32(TDLEN(0));
3877 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3879 if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3881 e1000_flush_tx_ring(adapter);
3882 /* recheck, maybe the fault is caused by the rx ring */
3883 pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3885 if (hang_state & FLUSH_DESC_REQUIRED)
3886 e1000_flush_rx_ring(adapter);
3890 * e1000e_systim_reset - reset the timesync registers after a hardware reset
3891 * @adapter: board private structure
3893 * When the MAC is reset, all hardware bits for timesync will be reset to the
3894 * default values. This function will restore the settings last in place.
3895 * Since the clock SYSTIME registers are reset, we will simply restore the
3896 * cyclecounter to the kernel real clock time.
3898 static void e1000e_systim_reset(struct e1000_adapter *adapter)
3900 struct ptp_clock_info *info = &adapter->ptp_clock_info;
3901 struct e1000_hw *hw = &adapter->hw;
3902 unsigned long flags;
3906 if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3909 if (info->adjfreq) {
3910 /* restore the previous ptp frequency delta */
3911 ret_val = info->adjfreq(info, adapter->ptp_delta);
3913 /* set the default base frequency if no adjustment possible */
3914 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3916 ew32(TIMINCA, timinca);
3920 dev_warn(&adapter->pdev->dev,
3921 "Failed to restore TIMINCA clock rate delta: %d\n",
3926 /* reset the systim ns time counter */
3927 spin_lock_irqsave(&adapter->systim_lock, flags);
3928 timecounter_init(&adapter->tc, &adapter->cc,
3929 ktime_to_ns(ktime_get_real()));
3930 spin_unlock_irqrestore(&adapter->systim_lock, flags);
3932 /* restore the previous hwtstamp configuration settings */
3933 e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3937 * e1000e_reset - bring the hardware into a known good state
3939 * This function boots the hardware and enables some settings that
3940 * require a configuration cycle of the hardware - those cannot be
3941 * set/changed during runtime. After reset the device needs to be
3942 * properly configured for Rx, Tx etc.
3944 void e1000e_reset(struct e1000_adapter *adapter)
3946 struct e1000_mac_info *mac = &adapter->hw.mac;
3947 struct e1000_fc_info *fc = &adapter->hw.fc;
3948 struct e1000_hw *hw = &adapter->hw;
3949 u32 tx_space, min_tx_space, min_rx_space;
3950 u32 pba = adapter->pba;
3953 /* reset Packet Buffer Allocation to default */
3956 if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3957 /* To maintain wire speed transmits, the Tx FIFO should be
3958 * large enough to accommodate two full transmit packets,
3959 * rounded up to the next 1KB and expressed in KB. Likewise,
3960 * the Rx FIFO should be large enough to accommodate at least
3961 * one full receive packet and is similarly rounded up and
3965 /* upper 16 bits has Tx packet buffer allocation size in KB */
3966 tx_space = pba >> 16;
3967 /* lower 16 bits has Rx packet buffer allocation size in KB */
3969 /* the Tx fifo also stores 16 bytes of information about the Tx
3970 * but don't include ethernet FCS because hardware appends it
3972 min_tx_space = (adapter->max_frame_size +
3973 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3974 min_tx_space = ALIGN(min_tx_space, 1024);
3975 min_tx_space >>= 10;
3976 /* software strips receive CRC, so leave room for it */
3977 min_rx_space = adapter->max_frame_size;
3978 min_rx_space = ALIGN(min_rx_space, 1024);
3979 min_rx_space >>= 10;
3981 /* If current Tx allocation is less than the min Tx FIFO size,
3982 * and the min Tx FIFO size is less than the current Rx FIFO
3983 * allocation, take space away from current Rx allocation
3985 if ((tx_space < min_tx_space) &&
3986 ((min_tx_space - tx_space) < pba)) {
3987 pba -= min_tx_space - tx_space;
3989 /* if short on Rx space, Rx wins and must trump Tx
3992 if (pba < min_rx_space)
3999 /* flow control settings
4001 * The high water mark must be low enough to fit one full frame
4002 * (or the size used for early receive) above it in the Rx FIFO.
4003 * Set it to the lower of:
4004 * - 90% of the Rx FIFO size, and
4005 * - the full Rx FIFO size minus one full frame
4007 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4008 fc->pause_time = 0xFFFF;
4010 fc->pause_time = E1000_FC_PAUSE_TIME;
4011 fc->send_xon = true;
4012 fc->current_mode = fc->requested_mode;
4014 switch (hw->mac.type) {
4016 case e1000_ich10lan:
4017 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4020 fc->high_water = 0x2800;
4021 fc->low_water = fc->high_water - 8;
4026 hwm = min(((pba << 10) * 9 / 10),
4027 ((pba << 10) - adapter->max_frame_size));
4029 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4030 fc->low_water = fc->high_water - 8;
4033 /* Workaround PCH LOM adapter hangs with certain network
4034 * loads. If hangs persist, try disabling Tx flow control.
4036 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4037 fc->high_water = 0x3500;
4038 fc->low_water = 0x1500;
4040 fc->high_water = 0x5000;
4041 fc->low_water = 0x3000;
4043 fc->refresh_time = 0x1000;
4049 fc->refresh_time = 0x0400;
4051 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4052 fc->high_water = 0x05C20;
4053 fc->low_water = 0x05048;
4054 fc->pause_time = 0x0650;
4060 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4061 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4065 /* Alignment of Tx data is on an arbitrary byte boundary with the
4066 * maximum size per Tx descriptor limited only to the transmit
4067 * allocation of the packet buffer minus 96 bytes with an upper
4068 * limit of 24KB due to receive synchronization limitations.
4070 adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4073 /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4074 * fit in receive buffer.
4076 if (adapter->itr_setting & 0x3) {
4077 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4078 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4079 dev_info(&adapter->pdev->dev,
4080 "Interrupt Throttle Rate off\n");
4081 adapter->flags2 |= FLAG2_DISABLE_AIM;
4082 e1000e_write_itr(adapter, 0);
4084 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4085 dev_info(&adapter->pdev->dev,
4086 "Interrupt Throttle Rate on\n");
4087 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4088 adapter->itr = 20000;
4089 e1000e_write_itr(adapter, adapter->itr);
4093 if (hw->mac.type >= e1000_pch_spt)
4094 e1000_flush_desc_rings(adapter);
4095 /* Allow time for pending master requests to run */
4096 mac->ops.reset_hw(hw);
4098 /* For parts with AMT enabled, let the firmware know
4099 * that the network interface is in control
4101 if (adapter->flags & FLAG_HAS_AMT)
4102 e1000e_get_hw_control(adapter);
4106 if (mac->ops.init_hw(hw))
4107 e_err("Hardware Error\n");
4109 e1000_update_mng_vlan(adapter);
4111 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4112 ew32(VET, ETH_P_8021Q);
4114 e1000e_reset_adaptive(hw);
4116 /* restore systim and hwtstamp settings */
4117 e1000e_systim_reset(adapter);
4119 /* Set EEE advertisement as appropriate */
4120 if (adapter->flags2 & FLAG2_HAS_EEE) {
4124 switch (hw->phy.type) {
4125 case e1000_phy_82579:
4126 adv_addr = I82579_EEE_ADVERTISEMENT;
4128 case e1000_phy_i217:
4129 adv_addr = I217_EEE_ADVERTISEMENT;
4132 dev_err(&adapter->pdev->dev,
4133 "Invalid PHY type setting EEE advertisement\n");
4137 ret_val = hw->phy.ops.acquire(hw);
4139 dev_err(&adapter->pdev->dev,
4140 "EEE advertisement - unable to acquire PHY\n");
4144 e1000_write_emi_reg_locked(hw, adv_addr,
4145 hw->dev_spec.ich8lan.eee_disable ?
4146 0 : adapter->eee_advert);
4148 hw->phy.ops.release(hw);
4151 if (!netif_running(adapter->netdev) &&
4152 !test_bit(__E1000_TESTING, &adapter->state))
4153 e1000_power_down_phy(adapter);
4155 e1000_get_phy_info(hw);
4157 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4158 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4160 /* speed up time to link by disabling smart power down, ignore
4161 * the return value of this function because there is nothing
4162 * different we would do if it failed
4164 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4165 phy_data &= ~IGP02E1000_PM_SPD;
4166 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4168 if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4171 /* Fextnvm7 @ 0xe4[2] = 1 */
4172 reg = er32(FEXTNVM7);
4173 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4174 ew32(FEXTNVM7, reg);
4175 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4176 reg = er32(FEXTNVM9);
4177 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4178 E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4179 ew32(FEXTNVM9, reg);
4185 * e1000e_trigger_lsc - trigger an LSC interrupt
4188 * Fire a link status change interrupt to start the watchdog.
4190 static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4192 struct e1000_hw *hw = &adapter->hw;
4194 if (adapter->msix_entries)
4195 ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
4197 ew32(ICS, E1000_ICS_LSC);
4200 void e1000e_up(struct e1000_adapter *adapter)
4202 /* hardware has been reset, we need to reload some things */
4203 e1000_configure(adapter);
4205 clear_bit(__E1000_DOWN, &adapter->state);
4207 if (adapter->msix_entries)
4208 e1000_configure_msix(adapter);
4209 e1000_irq_enable(adapter);
4211 netif_start_queue(adapter->netdev);
4213 e1000e_trigger_lsc(adapter);
4216 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4218 struct e1000_hw *hw = &adapter->hw;
4220 if (!(adapter->flags2 & FLAG2_DMA_BURST))
4223 /* flush pending descriptor writebacks to memory */
4224 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4225 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4227 /* execute the writes immediately */
4230 /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4231 * write is successful
4233 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4234 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4236 /* execute the writes immediately */
4240 static void e1000e_update_stats(struct e1000_adapter *adapter);
4243 * e1000e_down - quiesce the device and optionally reset the hardware
4244 * @adapter: board private structure
4245 * @reset: boolean flag to reset the hardware or not
4247 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4249 struct net_device *netdev = adapter->netdev;
4250 struct e1000_hw *hw = &adapter->hw;
4253 /* signal that we're down so the interrupt handler does not
4254 * reschedule our watchdog timer
4256 set_bit(__E1000_DOWN, &adapter->state);
4258 netif_carrier_off(netdev);
4260 /* disable receives in the hardware */
4262 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4263 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4264 /* flush and sleep below */
4266 netif_stop_queue(netdev);
4268 /* disable transmits in the hardware */
4270 tctl &= ~E1000_TCTL_EN;
4273 /* flush both disables and wait for them to finish */
4275 usleep_range(10000, 20000);
4277 e1000_irq_disable(adapter);
4279 napi_synchronize(&adapter->napi);
4281 del_timer_sync(&adapter->watchdog_timer);
4282 del_timer_sync(&adapter->phy_info_timer);
4284 spin_lock(&adapter->stats64_lock);
4285 e1000e_update_stats(adapter);
4286 spin_unlock(&adapter->stats64_lock);
4288 e1000e_flush_descriptors(adapter);
4290 adapter->link_speed = 0;
4291 adapter->link_duplex = 0;
4293 /* Disable Si errata workaround on PCHx for jumbo frame flow */
4294 if ((hw->mac.type >= e1000_pch2lan) &&
4295 (adapter->netdev->mtu > ETH_DATA_LEN) &&
4296 e1000_lv_jumbo_workaround_ich8lan(hw, false))
4297 e_dbg("failed to disable jumbo frame workaround mode\n");
4299 if (!pci_channel_offline(adapter->pdev)) {
4301 e1000e_reset(adapter);
4302 else if (hw->mac.type >= e1000_pch_spt)
4303 e1000_flush_desc_rings(adapter);
4305 e1000_clean_tx_ring(adapter->tx_ring);
4306 e1000_clean_rx_ring(adapter->rx_ring);
4309 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4312 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4313 usleep_range(1000, 2000);
4314 e1000e_down(adapter, true);
4316 clear_bit(__E1000_RESETTING, &adapter->state);
4320 * e1000e_sanitize_systim - sanitize raw cycle counter reads
4321 * @hw: pointer to the HW structure
4322 * @systim: time value read, sanitized and returned
4324 * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4325 * check to see that the time is incrementing at a reasonable
4326 * rate and is a multiple of incvalue.
4328 static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim)
4330 u64 time_delta, rem, temp;
4335 incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4336 for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4337 /* latch SYSTIMH on read of SYSTIML */
4338 systim_next = (u64)er32(SYSTIML);
4339 systim_next |= (u64)er32(SYSTIMH) << 32;
4341 time_delta = systim_next - systim;
4343 /* VMWare users have seen incvalue of zero, don't div / 0 */
4344 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4346 systim = systim_next;
4348 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4356 * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4357 * @cc: cyclecounter structure
4359 static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4361 struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4363 struct e1000_hw *hw = &adapter->hw;
4364 u32 systimel, systimeh;
4366 /* SYSTIMH latching upon SYSTIML read does not work well.
4367 * This means that if SYSTIML overflows after we read it but before
4368 * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4369 * will experience a huge non linear increment in the systime value
4370 * to fix that we test for overflow and if true, we re-read systime.
4372 systimel = er32(SYSTIML);
4373 systimeh = er32(SYSTIMH);
4374 /* Is systimel is so large that overflow is possible? */
4375 if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4376 u32 systimel_2 = er32(SYSTIML);
4377 if (systimel > systimel_2) {
4378 /* There was an overflow, read again SYSTIMH, and use
4381 systimeh = er32(SYSTIMH);
4382 systimel = systimel_2;
4385 systim = (u64)systimel;
4386 systim |= (u64)systimeh << 32;
4388 if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4389 systim = e1000e_sanitize_systim(hw, systim);
4395 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4396 * @adapter: board private structure to initialize
4398 * e1000_sw_init initializes the Adapter private data structure.
4399 * Fields are initialized based on PCI device information and
4400 * OS network device settings (MTU size).
4402 static int e1000_sw_init(struct e1000_adapter *adapter)
4404 struct net_device *netdev = adapter->netdev;
4406 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4407 adapter->rx_ps_bsize0 = 128;
4408 adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4409 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4410 adapter->tx_ring_count = E1000_DEFAULT_TXD;
4411 adapter->rx_ring_count = E1000_DEFAULT_RXD;
4413 spin_lock_init(&adapter->stats64_lock);
4415 e1000e_set_interrupt_capability(adapter);
4417 if (e1000_alloc_queues(adapter))
4420 /* Setup hardware time stamping cyclecounter */
4421 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4422 adapter->cc.read = e1000e_cyclecounter_read;
4423 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4424 adapter->cc.mult = 1;
4425 /* cc.shift set in e1000e_get_base_tininca() */
4427 spin_lock_init(&adapter->systim_lock);
4428 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4431 /* Explicitly disable IRQ since the NIC can be in any state. */
4432 e1000_irq_disable(adapter);
4434 set_bit(__E1000_DOWN, &adapter->state);
4439 * e1000_intr_msi_test - Interrupt Handler
4440 * @irq: interrupt number
4441 * @data: pointer to a network interface device structure
4443 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4445 struct net_device *netdev = data;
4446 struct e1000_adapter *adapter = netdev_priv(netdev);
4447 struct e1000_hw *hw = &adapter->hw;
4448 u32 icr = er32(ICR);
4450 e_dbg("icr is %08X\n", icr);
4451 if (icr & E1000_ICR_RXSEQ) {
4452 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4453 /* Force memory writes to complete before acknowledging the
4454 * interrupt is handled.
4463 * e1000_test_msi_interrupt - Returns 0 for successful test
4464 * @adapter: board private struct
4466 * code flow taken from tg3.c
4468 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4470 struct net_device *netdev = adapter->netdev;
4471 struct e1000_hw *hw = &adapter->hw;
4474 /* poll_enable hasn't been called yet, so don't need disable */
4475 /* clear any pending events */
4478 /* free the real vector and request a test handler */
4479 e1000_free_irq(adapter);
4480 e1000e_reset_interrupt_capability(adapter);
4482 /* Assume that the test fails, if it succeeds then the test
4483 * MSI irq handler will unset this flag
4485 adapter->flags |= FLAG_MSI_TEST_FAILED;
4487 err = pci_enable_msi(adapter->pdev);
4489 goto msi_test_failed;
4491 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4492 netdev->name, netdev);
4494 pci_disable_msi(adapter->pdev);
4495 goto msi_test_failed;
4498 /* Force memory writes to complete before enabling and firing an
4503 e1000_irq_enable(adapter);
4505 /* fire an unusual interrupt on the test handler */
4506 ew32(ICS, E1000_ICS_RXSEQ);
4510 e1000_irq_disable(adapter);
4512 rmb(); /* read flags after interrupt has been fired */
4514 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4515 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4516 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4518 e_dbg("MSI interrupt test succeeded!\n");
4521 free_irq(adapter->pdev->irq, netdev);
4522 pci_disable_msi(adapter->pdev);
4525 e1000e_set_interrupt_capability(adapter);
4526 return e1000_request_irq(adapter);
4530 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4531 * @adapter: board private struct
4533 * code flow taken from tg3.c, called with e1000 interrupts disabled.
4535 static int e1000_test_msi(struct e1000_adapter *adapter)
4540 if (!(adapter->flags & FLAG_MSI_ENABLED))
4543 /* disable SERR in case the MSI write causes a master abort */
4544 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4545 if (pci_cmd & PCI_COMMAND_SERR)
4546 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4547 pci_cmd & ~PCI_COMMAND_SERR);
4549 err = e1000_test_msi_interrupt(adapter);
4551 /* re-enable SERR */
4552 if (pci_cmd & PCI_COMMAND_SERR) {
4553 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4554 pci_cmd |= PCI_COMMAND_SERR;
4555 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4562 * e1000e_open - Called when a network interface is made active
4563 * @netdev: network interface device structure
4565 * Returns 0 on success, negative value on failure
4567 * The open entry point is called when a network interface is made
4568 * active by the system (IFF_UP). At this point all resources needed
4569 * for transmit and receive operations are allocated, the interrupt
4570 * handler is registered with the OS, the watchdog timer is started,
4571 * and the stack is notified that the interface is ready.
4573 int e1000e_open(struct net_device *netdev)
4575 struct e1000_adapter *adapter = netdev_priv(netdev);
4576 struct e1000_hw *hw = &adapter->hw;
4577 struct pci_dev *pdev = adapter->pdev;
4580 /* disallow open during test */
4581 if (test_bit(__E1000_TESTING, &adapter->state))
4584 pm_runtime_get_sync(&pdev->dev);
4586 netif_carrier_off(netdev);
4588 /* allocate transmit descriptors */
4589 err = e1000e_setup_tx_resources(adapter->tx_ring);
4593 /* allocate receive descriptors */
4594 err = e1000e_setup_rx_resources(adapter->rx_ring);
4598 /* If AMT is enabled, let the firmware know that the network
4599 * interface is now open and reset the part to a known state.
4601 if (adapter->flags & FLAG_HAS_AMT) {
4602 e1000e_get_hw_control(adapter);
4603 e1000e_reset(adapter);
4606 e1000e_power_up_phy(adapter);
4608 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4609 if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4610 e1000_update_mng_vlan(adapter);
4612 /* DMA latency requirement to workaround jumbo issue */
4613 pm_qos_add_request(&adapter->pm_qos_req, PM_QOS_CPU_DMA_LATENCY,
4614 PM_QOS_DEFAULT_VALUE);
4616 /* before we allocate an interrupt, we must be ready to handle it.
4617 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4618 * as soon as we call pci_request_irq, so we have to setup our
4619 * clean_rx handler before we do so.
4621 e1000_configure(adapter);
4623 err = e1000_request_irq(adapter);
4627 /* Work around PCIe errata with MSI interrupts causing some chipsets to
4628 * ignore e1000e MSI messages, which means we need to test our MSI
4631 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4632 err = e1000_test_msi(adapter);
4634 e_err("Interrupt allocation failed\n");
4639 /* From here on the code is the same as e1000e_up() */
4640 clear_bit(__E1000_DOWN, &adapter->state);
4642 napi_enable(&adapter->napi);
4644 e1000_irq_enable(adapter);
4646 adapter->tx_hang_recheck = false;
4647 netif_start_queue(netdev);
4649 hw->mac.get_link_status = true;
4650 pm_runtime_put(&pdev->dev);
4652 e1000e_trigger_lsc(adapter);
4657 pm_qos_remove_request(&adapter->pm_qos_req);
4658 e1000e_release_hw_control(adapter);
4659 e1000_power_down_phy(adapter);
4660 e1000e_free_rx_resources(adapter->rx_ring);
4662 e1000e_free_tx_resources(adapter->tx_ring);
4664 e1000e_reset(adapter);
4665 pm_runtime_put_sync(&pdev->dev);
4671 * e1000e_close - Disables a network interface
4672 * @netdev: network interface device structure
4674 * Returns 0, this is not allowed to fail
4676 * The close entry point is called when an interface is de-activated
4677 * by the OS. The hardware is still under the drivers control, but
4678 * needs to be disabled. A global MAC reset is issued to stop the
4679 * hardware, and all transmit and receive resources are freed.
4681 int e1000e_close(struct net_device *netdev)
4683 struct e1000_adapter *adapter = netdev_priv(netdev);
4684 struct pci_dev *pdev = adapter->pdev;
4685 int count = E1000_CHECK_RESET_COUNT;
4687 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4688 usleep_range(10000, 20000);
4690 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4692 pm_runtime_get_sync(&pdev->dev);
4694 if (!test_bit(__E1000_DOWN, &adapter->state)) {
4695 e1000e_down(adapter, true);
4696 e1000_free_irq(adapter);
4698 /* Link status message must follow this format */
4699 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
4702 napi_disable(&adapter->napi);
4704 e1000e_free_tx_resources(adapter->tx_ring);
4705 e1000e_free_rx_resources(adapter->rx_ring);
4707 /* kill manageability vlan ID if supported, but not if a vlan with
4708 * the same ID is registered on the host OS (let 8021q kill it)
4710 if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4711 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4712 adapter->mng_vlan_id);
4714 /* If AMT is enabled, let the firmware know that the network
4715 * interface is now closed
4717 if ((adapter->flags & FLAG_HAS_AMT) &&
4718 !test_bit(__E1000_TESTING, &adapter->state))
4719 e1000e_release_hw_control(adapter);
4721 pm_qos_remove_request(&adapter->pm_qos_req);
4723 pm_runtime_put_sync(&pdev->dev);
4729 * e1000_set_mac - Change the Ethernet Address of the NIC
4730 * @netdev: network interface device structure
4731 * @p: pointer to an address structure
4733 * Returns 0 on success, negative on failure
4735 static int e1000_set_mac(struct net_device *netdev, void *p)
4737 struct e1000_adapter *adapter = netdev_priv(netdev);
4738 struct e1000_hw *hw = &adapter->hw;
4739 struct sockaddr *addr = p;
4741 if (!is_valid_ether_addr(addr->sa_data))
4742 return -EADDRNOTAVAIL;
4744 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4745 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4747 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4749 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4750 /* activate the work around */
4751 e1000e_set_laa_state_82571(&adapter->hw, 1);
4753 /* Hold a copy of the LAA in RAR[14] This is done so that
4754 * between the time RAR[0] gets clobbered and the time it
4755 * gets fixed (in e1000_watchdog), the actual LAA is in one
4756 * of the RARs and no incoming packets directed to this port
4757 * are dropped. Eventually the LAA will be in RAR[0] and
4760 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4761 adapter->hw.mac.rar_entry_count - 1);
4768 * e1000e_update_phy_task - work thread to update phy
4769 * @work: pointer to our work struct
4771 * this worker thread exists because we must acquire a
4772 * semaphore to read the phy, which we could msleep while
4773 * waiting for it, and we can't msleep in a timer.
4775 static void e1000e_update_phy_task(struct work_struct *work)
4777 struct e1000_adapter *adapter = container_of(work,
4778 struct e1000_adapter,
4780 struct e1000_hw *hw = &adapter->hw;
4782 if (test_bit(__E1000_DOWN, &adapter->state))
4785 e1000_get_phy_info(hw);
4787 /* Enable EEE on 82579 after link up */
4788 if (hw->phy.type >= e1000_phy_82579)
4789 e1000_set_eee_pchlan(hw);
4793 * e1000_update_phy_info - timre call-back to update PHY info
4794 * @data: pointer to adapter cast into an unsigned long
4796 * Need to wait a few seconds after link up to get diagnostic information from
4799 static void e1000_update_phy_info(struct timer_list *t)
4801 struct e1000_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4803 if (test_bit(__E1000_DOWN, &adapter->state))
4806 schedule_work(&adapter->update_phy_task);
4810 * e1000e_update_phy_stats - Update the PHY statistics counters
4811 * @adapter: board private structure
4813 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4815 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4817 struct e1000_hw *hw = &adapter->hw;
4821 ret_val = hw->phy.ops.acquire(hw);
4825 /* A page set is expensive so check if already on desired page.
4826 * If not, set to the page with the PHY status registers.
4829 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4833 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4834 ret_val = hw->phy.ops.set_page(hw,
4835 HV_STATS_PAGE << IGP_PAGE_SHIFT);
4840 /* Single Collision Count */
4841 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4842 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4844 adapter->stats.scc += phy_data;
4846 /* Excessive Collision Count */
4847 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4848 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4850 adapter->stats.ecol += phy_data;
4852 /* Multiple Collision Count */
4853 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4854 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4856 adapter->stats.mcc += phy_data;
4858 /* Late Collision Count */
4859 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4860 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4862 adapter->stats.latecol += phy_data;
4864 /* Collision Count - also used for adaptive IFS */
4865 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4866 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4868 hw->mac.collision_delta = phy_data;
4871 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4872 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4874 adapter->stats.dc += phy_data;
4876 /* Transmit with no CRS */
4877 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4878 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4880 adapter->stats.tncrs += phy_data;
4883 hw->phy.ops.release(hw);
4887 * e1000e_update_stats - Update the board statistics counters
4888 * @adapter: board private structure
4890 static void e1000e_update_stats(struct e1000_adapter *adapter)
4892 struct net_device *netdev = adapter->netdev;
4893 struct e1000_hw *hw = &adapter->hw;
4894 struct pci_dev *pdev = adapter->pdev;
4896 /* Prevent stats update while adapter is being reset, or if the pci
4897 * connection is down.
4899 if (adapter->link_speed == 0)
4901 if (pci_channel_offline(pdev))
4904 adapter->stats.crcerrs += er32(CRCERRS);
4905 adapter->stats.gprc += er32(GPRC);
4906 adapter->stats.gorc += er32(GORCL);
4907 er32(GORCH); /* Clear gorc */
4908 adapter->stats.bprc += er32(BPRC);
4909 adapter->stats.mprc += er32(MPRC);
4910 adapter->stats.roc += er32(ROC);
4912 adapter->stats.mpc += er32(MPC);
4914 /* Half-duplex statistics */
4915 if (adapter->link_duplex == HALF_DUPLEX) {
4916 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4917 e1000e_update_phy_stats(adapter);
4919 adapter->stats.scc += er32(SCC);
4920 adapter->stats.ecol += er32(ECOL);
4921 adapter->stats.mcc += er32(MCC);
4922 adapter->stats.latecol += er32(LATECOL);
4923 adapter->stats.dc += er32(DC);
4925 hw->mac.collision_delta = er32(COLC);
4927 if ((hw->mac.type != e1000_82574) &&
4928 (hw->mac.type != e1000_82583))
4929 adapter->stats.tncrs += er32(TNCRS);
4931 adapter->stats.colc += hw->mac.collision_delta;
4934 adapter->stats.xonrxc += er32(XONRXC);
4935 adapter->stats.xontxc += er32(XONTXC);
4936 adapter->stats.xoffrxc += er32(XOFFRXC);
4937 adapter->stats.xofftxc += er32(XOFFTXC);
4938 adapter->stats.gptc += er32(GPTC);
4939 adapter->stats.gotc += er32(GOTCL);
4940 er32(GOTCH); /* Clear gotc */
4941 adapter->stats.rnbc += er32(RNBC);
4942 adapter->stats.ruc += er32(RUC);
4944 adapter->stats.mptc += er32(MPTC);
4945 adapter->stats.bptc += er32(BPTC);
4947 /* used for adaptive IFS */
4949 hw->mac.tx_packet_delta = er32(TPT);
4950 adapter->stats.tpt += hw->mac.tx_packet_delta;
4952 adapter->stats.algnerrc += er32(ALGNERRC);
4953 adapter->stats.rxerrc += er32(RXERRC);
4954 adapter->stats.cexterr += er32(CEXTERR);
4955 adapter->stats.tsctc += er32(TSCTC);
4956 adapter->stats.tsctfc += er32(TSCTFC);
4958 /* Fill out the OS statistics structure */
4959 netdev->stats.multicast = adapter->stats.mprc;
4960 netdev->stats.collisions = adapter->stats.colc;
4964 /* RLEC on some newer hardware can be incorrect so build
4965 * our own version based on RUC and ROC
4967 netdev->stats.rx_errors = adapter->stats.rxerrc +
4968 adapter->stats.crcerrs + adapter->stats.algnerrc +
4969 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
4970 netdev->stats.rx_length_errors = adapter->stats.ruc +
4972 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4973 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4974 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4977 netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
4978 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4979 netdev->stats.tx_window_errors = adapter->stats.latecol;
4980 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4982 /* Tx Dropped needs to be maintained elsewhere */
4984 /* Management Stats */
4985 adapter->stats.mgptc += er32(MGTPTC);
4986 adapter->stats.mgprc += er32(MGTPRC);
4987 adapter->stats.mgpdc += er32(MGTPDC);
4989 /* Correctable ECC Errors */
4990 if (hw->mac.type >= e1000_pch_lpt) {
4991 u32 pbeccsts = er32(PBECCSTS);
4993 adapter->corr_errors +=
4994 pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
4995 adapter->uncorr_errors +=
4996 (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
4997 E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
5002 * e1000_phy_read_status - Update the PHY register status snapshot
5003 * @adapter: board private structure
5005 static void e1000_phy_read_status(struct e1000_adapter *adapter)
5007 struct e1000_hw *hw = &adapter->hw;
5008 struct e1000_phy_regs *phy = &adapter->phy_regs;
5010 if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5011 (er32(STATUS) & E1000_STATUS_LU) &&
5012 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5015 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5016 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5017 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5018 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5019 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5020 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5021 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5022 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5024 e_warn("Error reading PHY register\n");
5026 /* Do not read PHY registers if link is not up
5027 * Set values to typical power-on defaults
5029 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5030 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5031 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5033 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5034 ADVERTISE_ALL | ADVERTISE_CSMA);
5036 phy->expansion = EXPANSION_ENABLENPAGE;
5037 phy->ctrl1000 = ADVERTISE_1000FULL;
5039 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5043 static void e1000_print_link_info(struct e1000_adapter *adapter)
5045 struct e1000_hw *hw = &adapter->hw;
5046 u32 ctrl = er32(CTRL);
5048 /* Link status message must follow this format for user tools */
5049 pr_info("%s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5050 adapter->netdev->name, adapter->link_speed,
5051 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5052 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5053 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5054 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5057 static bool e1000e_has_link(struct e1000_adapter *adapter)
5059 struct e1000_hw *hw = &adapter->hw;
5060 bool link_active = false;
5063 /* get_link_status is set on LSC (link status) interrupt or
5064 * Rx sequence error interrupt. get_link_status will stay
5065 * true until the check_for_link establishes link
5066 * for copper adapters ONLY
5068 switch (hw->phy.media_type) {
5069 case e1000_media_type_copper:
5070 if (hw->mac.get_link_status) {
5071 ret_val = hw->mac.ops.check_for_link(hw);
5072 link_active = !hw->mac.get_link_status;
5077 case e1000_media_type_fiber:
5078 ret_val = hw->mac.ops.check_for_link(hw);
5079 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5081 case e1000_media_type_internal_serdes:
5082 ret_val = hw->mac.ops.check_for_link(hw);
5083 link_active = hw->mac.serdes_has_link;
5086 case e1000_media_type_unknown:
5090 if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5091 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5092 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5093 e_info("Gigabit has been disabled, downgrading speed\n");
5099 static void e1000e_enable_receives(struct e1000_adapter *adapter)
5101 /* make sure the receive unit is started */
5102 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5103 (adapter->flags & FLAG_RESTART_NOW)) {
5104 struct e1000_hw *hw = &adapter->hw;
5105 u32 rctl = er32(RCTL);
5107 ew32(RCTL, rctl | E1000_RCTL_EN);
5108 adapter->flags &= ~FLAG_RESTART_NOW;
5112 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5114 struct e1000_hw *hw = &adapter->hw;
5116 /* With 82574 controllers, PHY needs to be checked periodically
5117 * for hung state and reset, if two calls return true
5119 if (e1000_check_phy_82574(hw))
5120 adapter->phy_hang_count++;
5122 adapter->phy_hang_count = 0;
5124 if (adapter->phy_hang_count > 1) {
5125 adapter->phy_hang_count = 0;
5126 e_dbg("PHY appears hung - resetting\n");
5127 schedule_work(&adapter->reset_task);
5132 * e1000_watchdog - Timer Call-back
5133 * @data: pointer to adapter cast into an unsigned long
5135 static void e1000_watchdog(struct timer_list *t)
5137 struct e1000_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5139 /* Do the rest outside of interrupt context */
5140 schedule_work(&adapter->watchdog_task);
5142 /* TODO: make this use queue_delayed_work() */
5145 static void e1000_watchdog_task(struct work_struct *work)
5147 struct e1000_adapter *adapter = container_of(work,
5148 struct e1000_adapter,
5150 struct net_device *netdev = adapter->netdev;
5151 struct e1000_mac_info *mac = &adapter->hw.mac;
5152 struct e1000_phy_info *phy = &adapter->hw.phy;
5153 struct e1000_ring *tx_ring = adapter->tx_ring;
5154 struct e1000_hw *hw = &adapter->hw;
5157 if (test_bit(__E1000_DOWN, &adapter->state))
5160 link = e1000e_has_link(adapter);
5161 if ((netif_carrier_ok(netdev)) && link) {
5162 /* Cancel scheduled suspend requests. */
5163 pm_runtime_resume(netdev->dev.parent);
5165 e1000e_enable_receives(adapter);
5169 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5170 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5171 e1000_update_mng_vlan(adapter);
5174 if (!netif_carrier_ok(netdev)) {
5177 /* Cancel scheduled suspend requests. */
5178 pm_runtime_resume(netdev->dev.parent);
5180 /* update snapshot of PHY registers on LSC */
5181 e1000_phy_read_status(adapter);
5182 mac->ops.get_link_up_info(&adapter->hw,
5183 &adapter->link_speed,
5184 &adapter->link_duplex);
5185 e1000_print_link_info(adapter);
5187 /* check if SmartSpeed worked */
5188 e1000e_check_downshift(hw);
5189 if (phy->speed_downgraded)
5191 "Link Speed was downgraded by SmartSpeed\n");
5193 /* On supported PHYs, check for duplex mismatch only
5194 * if link has autonegotiated at 10/100 half
5196 if ((hw->phy.type == e1000_phy_igp_3 ||
5197 hw->phy.type == e1000_phy_bm) &&
5199 (adapter->link_speed == SPEED_10 ||
5200 adapter->link_speed == SPEED_100) &&
5201 (adapter->link_duplex == HALF_DUPLEX)) {
5204 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5206 if (!(autoneg_exp & EXPANSION_NWAY))
5207 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
5210 /* adjust timeout factor according to speed/duplex */
5211 adapter->tx_timeout_factor = 1;
5212 switch (adapter->link_speed) {
5215 adapter->tx_timeout_factor = 16;
5219 adapter->tx_timeout_factor = 10;
5223 /* workaround: re-program speed mode bit after
5226 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5230 tarc0 = er32(TARC(0));
5231 tarc0 &= ~SPEED_MODE_BIT;
5232 ew32(TARC(0), tarc0);
5235 /* disable TSO for pcie and 10/100 speeds, to avoid
5236 * some hardware issues
5238 if (!(adapter->flags & FLAG_TSO_FORCE)) {
5239 switch (adapter->link_speed) {
5242 e_info("10/100 speed: disabling TSO\n");
5243 netdev->features &= ~NETIF_F_TSO;
5244 netdev->features &= ~NETIF_F_TSO6;
5247 netdev->features |= NETIF_F_TSO;
5248 netdev->features |= NETIF_F_TSO6;
5256 /* enable transmits in the hardware, need to do this
5257 * after setting TARC(0)
5260 tctl |= E1000_TCTL_EN;
5263 /* Perform any post-link-up configuration before
5264 * reporting link up.
5266 if (phy->ops.cfg_on_link_up)
5267 phy->ops.cfg_on_link_up(hw);
5269 netif_carrier_on(netdev);
5271 if (!test_bit(__E1000_DOWN, &adapter->state))
5272 mod_timer(&adapter->phy_info_timer,
5273 round_jiffies(jiffies + 2 * HZ));
5276 if (netif_carrier_ok(netdev)) {
5277 adapter->link_speed = 0;
5278 adapter->link_duplex = 0;
5279 /* Link status message must follow this format */
5280 pr_info("%s NIC Link is Down\n", adapter->netdev->name);
5281 netif_carrier_off(netdev);
5282 if (!test_bit(__E1000_DOWN, &adapter->state))
5283 mod_timer(&adapter->phy_info_timer,
5284 round_jiffies(jiffies + 2 * HZ));
5286 /* 8000ES2LAN requires a Rx packet buffer work-around
5287 * on link down event; reset the controller to flush
5288 * the Rx packet buffer.
5290 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5291 adapter->flags |= FLAG_RESTART_NOW;
5293 pm_schedule_suspend(netdev->dev.parent,
5299 spin_lock(&adapter->stats64_lock);
5300 e1000e_update_stats(adapter);
5302 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5303 adapter->tpt_old = adapter->stats.tpt;
5304 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5305 adapter->colc_old = adapter->stats.colc;
5307 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5308 adapter->gorc_old = adapter->stats.gorc;
5309 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5310 adapter->gotc_old = adapter->stats.gotc;
5311 spin_unlock(&adapter->stats64_lock);
5313 /* If the link is lost the controller stops DMA, but
5314 * if there is queued Tx work it cannot be done. So
5315 * reset the controller to flush the Tx packet buffers.
5317 if (!netif_carrier_ok(netdev) &&
5318 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5319 adapter->flags |= FLAG_RESTART_NOW;
5321 /* If reset is necessary, do it outside of interrupt context. */
5322 if (adapter->flags & FLAG_RESTART_NOW) {
5323 schedule_work(&adapter->reset_task);
5324 /* return immediately since reset is imminent */
5328 e1000e_update_adaptive(&adapter->hw);
5330 /* Simple mode for Interrupt Throttle Rate (ITR) */
5331 if (adapter->itr_setting == 4) {
5332 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5333 * Total asymmetrical Tx or Rx gets ITR=8000;
5334 * everyone else is between 2000-8000.
5336 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5337 u32 dif = (adapter->gotc > adapter->gorc ?
5338 adapter->gotc - adapter->gorc :
5339 adapter->gorc - adapter->gotc) / 10000;
5340 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5342 e1000e_write_itr(adapter, itr);
5345 /* Cause software interrupt to ensure Rx ring is cleaned */
5346 if (adapter->msix_entries)
5347 ew32(ICS, adapter->rx_ring->ims_val);
5349 ew32(ICS, E1000_ICS_RXDMT0);
5351 /* flush pending descriptors to memory before detecting Tx hang */
5352 e1000e_flush_descriptors(adapter);
5354 /* Force detection of hung controller every watchdog period */
5355 adapter->detect_tx_hung = true;
5357 /* With 82571 controllers, LAA may be overwritten due to controller
5358 * reset from the other port. Set the appropriate LAA in RAR[0]
5360 if (e1000e_get_laa_state_82571(hw))
5361 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5363 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5364 e1000e_check_82574_phy_workaround(adapter);
5366 /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5367 if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5368 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5369 (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5371 adapter->rx_hwtstamp_cleared++;
5373 adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5377 /* Reset the timer */
5378 if (!test_bit(__E1000_DOWN, &adapter->state))
5379 mod_timer(&adapter->watchdog_timer,
5380 round_jiffies(jiffies + 2 * HZ));
5383 #define E1000_TX_FLAGS_CSUM 0x00000001
5384 #define E1000_TX_FLAGS_VLAN 0x00000002
5385 #define E1000_TX_FLAGS_TSO 0x00000004
5386 #define E1000_TX_FLAGS_IPV4 0x00000008
5387 #define E1000_TX_FLAGS_NO_FCS 0x00000010
5388 #define E1000_TX_FLAGS_HWTSTAMP 0x00000020
5389 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
5390 #define E1000_TX_FLAGS_VLAN_SHIFT 16
5392 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5395 struct e1000_context_desc *context_desc;
5396 struct e1000_buffer *buffer_info;
5400 u8 ipcss, ipcso, tucss, tucso, hdr_len;
5403 if (!skb_is_gso(skb))
5406 err = skb_cow_head(skb, 0);
5410 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5411 mss = skb_shinfo(skb)->gso_size;
5412 if (protocol == htons(ETH_P_IP)) {
5413 struct iphdr *iph = ip_hdr(skb);
5416 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5418 cmd_length = E1000_TXD_CMD_IP;
5419 ipcse = skb_transport_offset(skb) - 1;
5420 } else if (skb_is_gso_v6(skb)) {
5421 ipv6_hdr(skb)->payload_len = 0;
5422 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5423 &ipv6_hdr(skb)->daddr,
5427 ipcss = skb_network_offset(skb);
5428 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5429 tucss = skb_transport_offset(skb);
5430 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5432 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5433 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5435 i = tx_ring->next_to_use;
5436 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5437 buffer_info = &tx_ring->buffer_info[i];
5439 context_desc->lower_setup.ip_fields.ipcss = ipcss;
5440 context_desc->lower_setup.ip_fields.ipcso = ipcso;
5441 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5442 context_desc->upper_setup.tcp_fields.tucss = tucss;
5443 context_desc->upper_setup.tcp_fields.tucso = tucso;
5444 context_desc->upper_setup.tcp_fields.tucse = 0;
5445 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5446 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5447 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5449 buffer_info->time_stamp = jiffies;
5450 buffer_info->next_to_watch = i;
5453 if (i == tx_ring->count)
5455 tx_ring->next_to_use = i;
5460 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5463 struct e1000_adapter *adapter = tx_ring->adapter;
5464 struct e1000_context_desc *context_desc;
5465 struct e1000_buffer *buffer_info;
5468 u32 cmd_len = E1000_TXD_CMD_DEXT;
5470 if (skb->ip_summed != CHECKSUM_PARTIAL)
5474 case cpu_to_be16(ETH_P_IP):
5475 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5476 cmd_len |= E1000_TXD_CMD_TCP;
5478 case cpu_to_be16(ETH_P_IPV6):
5479 /* XXX not handling all IPV6 headers */
5480 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5481 cmd_len |= E1000_TXD_CMD_TCP;
5484 if (unlikely(net_ratelimit()))
5485 e_warn("checksum_partial proto=%x!\n",
5486 be16_to_cpu(protocol));
5490 css = skb_checksum_start_offset(skb);
5492 i = tx_ring->next_to_use;
5493 buffer_info = &tx_ring->buffer_info[i];
5494 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5496 context_desc->lower_setup.ip_config = 0;
5497 context_desc->upper_setup.tcp_fields.tucss = css;
5498 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5499 context_desc->upper_setup.tcp_fields.tucse = 0;
5500 context_desc->tcp_seg_setup.data = 0;
5501 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5503 buffer_info->time_stamp = jiffies;
5504 buffer_info->next_to_watch = i;
5507 if (i == tx_ring->count)
5509 tx_ring->next_to_use = i;
5514 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5515 unsigned int first, unsigned int max_per_txd,
5516 unsigned int nr_frags)
5518 struct e1000_adapter *adapter = tx_ring->adapter;
5519 struct pci_dev *pdev = adapter->pdev;
5520 struct e1000_buffer *buffer_info;
5521 unsigned int len = skb_headlen(skb);
5522 unsigned int offset = 0, size, count = 0, i;
5523 unsigned int f, bytecount, segs;
5525 i = tx_ring->next_to_use;
5528 buffer_info = &tx_ring->buffer_info[i];
5529 size = min(len, max_per_txd);
5531 buffer_info->length = size;
5532 buffer_info->time_stamp = jiffies;
5533 buffer_info->next_to_watch = i;
5534 buffer_info->dma = dma_map_single(&pdev->dev,
5536 size, DMA_TO_DEVICE);
5537 buffer_info->mapped_as_page = false;
5538 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5547 if (i == tx_ring->count)
5552 for (f = 0; f < nr_frags; f++) {
5553 const struct skb_frag_struct *frag;
5555 frag = &skb_shinfo(skb)->frags[f];
5556 len = skb_frag_size(frag);
5561 if (i == tx_ring->count)
5564 buffer_info = &tx_ring->buffer_info[i];
5565 size = min(len, max_per_txd);
5567 buffer_info->length = size;
5568 buffer_info->time_stamp = jiffies;
5569 buffer_info->next_to_watch = i;
5570 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5573 buffer_info->mapped_as_page = true;
5574 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5583 segs = skb_shinfo(skb)->gso_segs ? : 1;
5584 /* multiply data chunks by size of headers */
5585 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5587 tx_ring->buffer_info[i].skb = skb;
5588 tx_ring->buffer_info[i].segs = segs;
5589 tx_ring->buffer_info[i].bytecount = bytecount;
5590 tx_ring->buffer_info[first].next_to_watch = i;
5595 dev_err(&pdev->dev, "Tx DMA map failed\n");
5596 buffer_info->dma = 0;
5602 i += tx_ring->count;
5604 buffer_info = &tx_ring->buffer_info[i];
5605 e1000_put_txbuf(tx_ring, buffer_info, true);
5611 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5613 struct e1000_adapter *adapter = tx_ring->adapter;
5614 struct e1000_tx_desc *tx_desc = NULL;
5615 struct e1000_buffer *buffer_info;
5616 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5619 if (tx_flags & E1000_TX_FLAGS_TSO) {
5620 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5622 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5624 if (tx_flags & E1000_TX_FLAGS_IPV4)
5625 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5628 if (tx_flags & E1000_TX_FLAGS_CSUM) {
5629 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5630 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5633 if (tx_flags & E1000_TX_FLAGS_VLAN) {
5634 txd_lower |= E1000_TXD_CMD_VLE;
5635 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5638 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5639 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5641 if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5642 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5643 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5646 i = tx_ring->next_to_use;
5649 buffer_info = &tx_ring->buffer_info[i];
5650 tx_desc = E1000_TX_DESC(*tx_ring, i);
5651 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5652 tx_desc->lower.data = cpu_to_le32(txd_lower |
5653 buffer_info->length);
5654 tx_desc->upper.data = cpu_to_le32(txd_upper);
5657 if (i == tx_ring->count)
5659 } while (--count > 0);
5661 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5663 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5664 if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5665 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5667 /* Force memory writes to complete before letting h/w
5668 * know there are new descriptors to fetch. (Only
5669 * applicable for weak-ordered memory model archs,
5674 tx_ring->next_to_use = i;
5677 #define MINIMUM_DHCP_PACKET_SIZE 282
5678 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5679 struct sk_buff *skb)
5681 struct e1000_hw *hw = &adapter->hw;
5684 if (skb_vlan_tag_present(skb) &&
5685 !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5686 (adapter->hw.mng_cookie.status &
5687 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5690 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5693 if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5697 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5700 if (ip->protocol != IPPROTO_UDP)
5703 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5704 if (ntohs(udp->dest) != 67)
5707 offset = (u8 *)udp + 8 - skb->data;
5708 length = skb->len - offset;
5709 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5715 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5717 struct e1000_adapter *adapter = tx_ring->adapter;
5719 netif_stop_queue(adapter->netdev);
5720 /* Herbert's original patch had:
5721 * smp_mb__after_netif_stop_queue();
5722 * but since that doesn't exist yet, just open code it.
5726 /* We need to check again in a case another CPU has just
5727 * made room available.
5729 if (e1000_desc_unused(tx_ring) < size)
5733 netif_start_queue(adapter->netdev);
5734 ++adapter->restart_queue;
5738 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5740 BUG_ON(size > tx_ring->count);
5742 if (e1000_desc_unused(tx_ring) >= size)
5744 return __e1000_maybe_stop_tx(tx_ring, size);
5747 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5748 struct net_device *netdev)
5750 struct e1000_adapter *adapter = netdev_priv(netdev);
5751 struct e1000_ring *tx_ring = adapter->tx_ring;
5753 unsigned int tx_flags = 0;
5754 unsigned int len = skb_headlen(skb);
5755 unsigned int nr_frags;
5760 __be16 protocol = vlan_get_protocol(skb);
5762 if (test_bit(__E1000_DOWN, &adapter->state)) {
5763 dev_kfree_skb_any(skb);
5764 return NETDEV_TX_OK;
5767 if (skb->len <= 0) {
5768 dev_kfree_skb_any(skb);
5769 return NETDEV_TX_OK;
5772 /* The minimum packet size with TCTL.PSP set is 17 bytes so
5773 * pad skb in order to meet this minimum size requirement
5775 if (skb_put_padto(skb, 17))
5776 return NETDEV_TX_OK;
5778 mss = skb_shinfo(skb)->gso_size;
5782 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5783 * points to just header, pull a few bytes of payload from
5784 * frags into skb->data
5786 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5787 /* we do this workaround for ES2LAN, but it is un-necessary,
5788 * avoiding it could save a lot of cycles
5790 if (skb->data_len && (hdr_len == len)) {
5791 unsigned int pull_size;
5793 pull_size = min_t(unsigned int, 4, skb->data_len);
5794 if (!__pskb_pull_tail(skb, pull_size)) {
5795 e_err("__pskb_pull_tail failed.\n");
5796 dev_kfree_skb_any(skb);
5797 return NETDEV_TX_OK;
5799 len = skb_headlen(skb);
5803 /* reserve a descriptor for the offload context */
5804 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5808 count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5810 nr_frags = skb_shinfo(skb)->nr_frags;
5811 for (f = 0; f < nr_frags; f++)
5812 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5813 adapter->tx_fifo_limit);
5815 if (adapter->hw.mac.tx_pkt_filtering)
5816 e1000_transfer_dhcp_info(adapter, skb);
5818 /* need: count + 2 desc gap to keep tail from touching
5819 * head, otherwise try next time
5821 if (e1000_maybe_stop_tx(tx_ring, count + 2))
5822 return NETDEV_TX_BUSY;
5824 if (skb_vlan_tag_present(skb)) {
5825 tx_flags |= E1000_TX_FLAGS_VLAN;
5826 tx_flags |= (skb_vlan_tag_get(skb) <<
5827 E1000_TX_FLAGS_VLAN_SHIFT);
5830 first = tx_ring->next_to_use;
5832 tso = e1000_tso(tx_ring, skb, protocol);
5834 dev_kfree_skb_any(skb);
5835 return NETDEV_TX_OK;
5839 tx_flags |= E1000_TX_FLAGS_TSO;
5840 else if (e1000_tx_csum(tx_ring, skb, protocol))
5841 tx_flags |= E1000_TX_FLAGS_CSUM;
5843 /* Old method was to assume IPv4 packet by default if TSO was enabled.
5844 * 82571 hardware supports TSO capabilities for IPv6 as well...
5845 * no longer assume, we must.
5847 if (protocol == htons(ETH_P_IP))
5848 tx_flags |= E1000_TX_FLAGS_IPV4;
5850 if (unlikely(skb->no_fcs))
5851 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5853 /* if count is 0 then mapping error has occurred */
5854 count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5857 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5858 (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5859 if (!adapter->tx_hwtstamp_skb) {
5860 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5861 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5862 adapter->tx_hwtstamp_skb = skb_get(skb);
5863 adapter->tx_hwtstamp_start = jiffies;
5864 schedule_work(&adapter->tx_hwtstamp_work);
5866 adapter->tx_hwtstamp_skipped++;
5870 skb_tx_timestamp(skb);
5872 netdev_sent_queue(netdev, skb->len);
5873 e1000_tx_queue(tx_ring, tx_flags, count);
5874 /* Make sure there is space in the ring for the next send. */
5875 e1000_maybe_stop_tx(tx_ring,
5877 DIV_ROUND_UP(PAGE_SIZE,
5878 adapter->tx_fifo_limit) + 2));
5880 if (!skb->xmit_more ||
5881 netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5882 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5883 e1000e_update_tdt_wa(tx_ring,
5884 tx_ring->next_to_use);
5886 writel(tx_ring->next_to_use, tx_ring->tail);
5888 /* we need this if more than one processor can write
5889 * to our tail at a time, it synchronizes IO on
5895 dev_kfree_skb_any(skb);
5896 tx_ring->buffer_info[first].time_stamp = 0;
5897 tx_ring->next_to_use = first;
5900 return NETDEV_TX_OK;
5904 * e1000_tx_timeout - Respond to a Tx Hang
5905 * @netdev: network interface device structure
5907 static void e1000_tx_timeout(struct net_device *netdev)
5909 struct e1000_adapter *adapter = netdev_priv(netdev);
5911 /* Do the reset outside of interrupt context */
5912 adapter->tx_timeout_count++;
5913 schedule_work(&adapter->reset_task);
5916 static void e1000_reset_task(struct work_struct *work)
5918 struct e1000_adapter *adapter;
5919 adapter = container_of(work, struct e1000_adapter, reset_task);
5921 /* don't run the task if already down */
5922 if (test_bit(__E1000_DOWN, &adapter->state))
5925 if (!(adapter->flags & FLAG_RESTART_NOW)) {
5926 e1000e_dump(adapter);
5927 e_err("Reset adapter unexpectedly\n");
5929 e1000e_reinit_locked(adapter);
5933 * e1000_get_stats64 - Get System Network Statistics
5934 * @netdev: network interface device structure
5935 * @stats: rtnl_link_stats64 pointer
5937 * Returns the address of the device statistics structure.
5939 void e1000e_get_stats64(struct net_device *netdev,
5940 struct rtnl_link_stats64 *stats)
5942 struct e1000_adapter *adapter = netdev_priv(netdev);
5944 spin_lock(&adapter->stats64_lock);
5945 e1000e_update_stats(adapter);
5946 /* Fill out the OS statistics structure */
5947 stats->rx_bytes = adapter->stats.gorc;
5948 stats->rx_packets = adapter->stats.gprc;
5949 stats->tx_bytes = adapter->stats.gotc;
5950 stats->tx_packets = adapter->stats.gptc;
5951 stats->multicast = adapter->stats.mprc;
5952 stats->collisions = adapter->stats.colc;
5956 /* RLEC on some newer hardware can be incorrect so build
5957 * our own version based on RUC and ROC
5959 stats->rx_errors = adapter->stats.rxerrc +
5960 adapter->stats.crcerrs + adapter->stats.algnerrc +
5961 adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5962 stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
5963 stats->rx_crc_errors = adapter->stats.crcerrs;
5964 stats->rx_frame_errors = adapter->stats.algnerrc;
5965 stats->rx_missed_errors = adapter->stats.mpc;
5968 stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5969 stats->tx_aborted_errors = adapter->stats.ecol;
5970 stats->tx_window_errors = adapter->stats.latecol;
5971 stats->tx_carrier_errors = adapter->stats.tncrs;
5973 /* Tx Dropped needs to be maintained elsewhere */
5975 spin_unlock(&adapter->stats64_lock);
5979 * e1000_change_mtu - Change the Maximum Transfer Unit
5980 * @netdev: network interface device structure
5981 * @new_mtu: new value for maximum frame size
5983 * Returns 0 on success, negative on failure
5985 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5987 struct e1000_adapter *adapter = netdev_priv(netdev);
5988 int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
5990 /* Jumbo frame support */
5991 if ((new_mtu > ETH_DATA_LEN) &&
5992 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5993 e_err("Jumbo Frames not supported.\n");
5997 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5998 if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5999 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6000 (new_mtu > ETH_DATA_LEN)) {
6001 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6005 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6006 usleep_range(1000, 2000);
6007 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6008 adapter->max_frame_size = max_frame;
6009 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6010 netdev->mtu = new_mtu;
6012 pm_runtime_get_sync(netdev->dev.parent);
6014 if (netif_running(netdev))
6015 e1000e_down(adapter, true);
6017 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6018 * means we reserve 2 more, this pushes us to allocate from the next
6020 * i.e. RXBUFFER_2048 --> size-4096 slab
6021 * However with the new *_jumbo_rx* routines, jumbo receives will use
6025 if (max_frame <= 2048)
6026 adapter->rx_buffer_len = 2048;
6028 adapter->rx_buffer_len = 4096;
6030 /* adjust allocation if LPE protects us, and we aren't using SBP */
6031 if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6032 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6034 if (netif_running(netdev))
6037 e1000e_reset(adapter);
6039 pm_runtime_put_sync(netdev->dev.parent);
6041 clear_bit(__E1000_RESETTING, &adapter->state);
6046 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6049 struct e1000_adapter *adapter = netdev_priv(netdev);
6050 struct mii_ioctl_data *data = if_mii(ifr);
6052 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6057 data->phy_id = adapter->hw.phy.addr;
6060 e1000_phy_read_status(adapter);
6062 switch (data->reg_num & 0x1F) {
6064 data->val_out = adapter->phy_regs.bmcr;
6067 data->val_out = adapter->phy_regs.bmsr;
6070 data->val_out = (adapter->hw.phy.id >> 16);
6073 data->val_out = (adapter->hw.phy.id & 0xFFFF);
6076 data->val_out = adapter->phy_regs.advertise;
6079 data->val_out = adapter->phy_regs.lpa;
6082 data->val_out = adapter->phy_regs.expansion;
6085 data->val_out = adapter->phy_regs.ctrl1000;
6088 data->val_out = adapter->phy_regs.stat1000;
6091 data->val_out = adapter->phy_regs.estatus;
6105 * e1000e_hwtstamp_ioctl - control hardware time stamping
6106 * @netdev: network interface device structure
6107 * @ifreq: interface request
6109 * Outgoing time stamping can be enabled and disabled. Play nice and
6110 * disable it when requested, although it shouldn't cause any overhead
6111 * when no packet needs it. At most one packet in the queue may be
6112 * marked for time stamping, otherwise it would be impossible to tell
6113 * for sure to which packet the hardware time stamp belongs.
6115 * Incoming time stamping has to be configured via the hardware filters.
6116 * Not all combinations are supported, in particular event type has to be
6117 * specified. Matching the kind of event packet is not supported, with the
6118 * exception of "all V2 events regardless of level 2 or 4".
6120 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6122 struct e1000_adapter *adapter = netdev_priv(netdev);
6123 struct hwtstamp_config config;
6126 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6129 ret_val = e1000e_config_hwtstamp(adapter, &config);
6133 switch (config.rx_filter) {
6134 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6135 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6136 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6137 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6138 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6139 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6140 /* With V2 type filters which specify a Sync or Delay Request,
6141 * Path Delay Request/Response messages are also time stamped
6142 * by hardware so notify the caller the requested packets plus
6143 * some others are time stamped.
6145 config.rx_filter = HWTSTAMP_FILTER_SOME;
6151 return copy_to_user(ifr->ifr_data, &config,
6152 sizeof(config)) ? -EFAULT : 0;
6155 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6157 struct e1000_adapter *adapter = netdev_priv(netdev);
6159 return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6160 sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6163 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6169 return e1000_mii_ioctl(netdev, ifr, cmd);
6171 return e1000e_hwtstamp_set(netdev, ifr);
6173 return e1000e_hwtstamp_get(netdev, ifr);
6179 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6181 struct e1000_hw *hw = &adapter->hw;
6182 u32 i, mac_reg, wuc;
6183 u16 phy_reg, wuc_enable;
6186 /* copy MAC RARs to PHY RARs */
6187 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6189 retval = hw->phy.ops.acquire(hw);
6191 e_err("Could not acquire PHY\n");
6195 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6196 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6200 /* copy MAC MTA to PHY MTA - only needed for pchlan */
6201 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6202 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6203 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6204 (u16)(mac_reg & 0xFFFF));
6205 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6206 (u16)((mac_reg >> 16) & 0xFFFF));
6209 /* configure PHY Rx Control register */
6210 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6211 mac_reg = er32(RCTL);
6212 if (mac_reg & E1000_RCTL_UPE)
6213 phy_reg |= BM_RCTL_UPE;
6214 if (mac_reg & E1000_RCTL_MPE)
6215 phy_reg |= BM_RCTL_MPE;
6216 phy_reg &= ~(BM_RCTL_MO_MASK);
6217 if (mac_reg & E1000_RCTL_MO_3)
6218 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6219 << BM_RCTL_MO_SHIFT);
6220 if (mac_reg & E1000_RCTL_BAM)
6221 phy_reg |= BM_RCTL_BAM;
6222 if (mac_reg & E1000_RCTL_PMCF)
6223 phy_reg |= BM_RCTL_PMCF;
6224 mac_reg = er32(CTRL);
6225 if (mac_reg & E1000_CTRL_RFCE)
6226 phy_reg |= BM_RCTL_RFCE;
6227 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6229 wuc = E1000_WUC_PME_EN;
6230 if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6231 wuc |= E1000_WUC_APME;
6233 /* enable PHY wakeup in MAC register */
6235 ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6236 E1000_WUC_PME_STATUS | wuc));
6238 /* configure and enable PHY wakeup in PHY registers */
6239 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6240 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6242 /* activate PHY wakeup */
6243 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6244 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6246 e_err("Could not set PHY Host Wakeup bit\n");
6248 hw->phy.ops.release(hw);
6253 static void e1000e_flush_lpic(struct pci_dev *pdev)
6255 struct net_device *netdev = pci_get_drvdata(pdev);
6256 struct e1000_adapter *adapter = netdev_priv(netdev);
6257 struct e1000_hw *hw = &adapter->hw;
6260 pm_runtime_get_sync(netdev->dev.parent);
6262 ret_val = hw->phy.ops.acquire(hw);
6266 pr_info("EEE TX LPI TIMER: %08X\n",
6267 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6269 hw->phy.ops.release(hw);
6272 pm_runtime_put_sync(netdev->dev.parent);
6275 static int e1000e_pm_freeze(struct device *dev)
6277 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6278 struct e1000_adapter *adapter = netdev_priv(netdev);
6280 netif_device_detach(netdev);
6282 if (netif_running(netdev)) {
6283 int count = E1000_CHECK_RESET_COUNT;
6285 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6286 usleep_range(10000, 20000);
6288 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6290 /* Quiesce the device without resetting the hardware */
6291 e1000e_down(adapter, false);
6292 e1000_free_irq(adapter);
6294 e1000e_reset_interrupt_capability(adapter);
6296 /* Allow time for pending master requests to run */
6297 e1000e_disable_pcie_master(&adapter->hw);
6302 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6304 struct net_device *netdev = pci_get_drvdata(pdev);
6305 struct e1000_adapter *adapter = netdev_priv(netdev);
6306 struct e1000_hw *hw = &adapter->hw;
6307 u32 ctrl, ctrl_ext, rctl, status;
6308 /* Runtime suspend should only enable wakeup for link changes */
6309 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6312 status = er32(STATUS);
6313 if (status & E1000_STATUS_LU)
6314 wufc &= ~E1000_WUFC_LNKC;
6317 e1000_setup_rctl(adapter);
6318 e1000e_set_rx_mode(netdev);
6320 /* turn on all-multi mode if wake on multicast is enabled */
6321 if (wufc & E1000_WUFC_MC) {
6323 rctl |= E1000_RCTL_MPE;
6328 ctrl |= E1000_CTRL_ADVD3WUC;
6329 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6330 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6333 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6334 adapter->hw.phy.media_type ==
6335 e1000_media_type_internal_serdes) {
6336 /* keep the laser running in D3 */
6337 ctrl_ext = er32(CTRL_EXT);
6338 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6339 ew32(CTRL_EXT, ctrl_ext);
6343 e1000e_power_up_phy(adapter);
6345 if (adapter->flags & FLAG_IS_ICH)
6346 e1000_suspend_workarounds_ich8lan(&adapter->hw);
6348 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6349 /* enable wakeup by the PHY */
6350 retval = e1000_init_phy_wakeup(adapter, wufc);
6354 /* enable wakeup by the MAC */
6356 ew32(WUC, E1000_WUC_PME_EN);
6362 e1000_power_down_phy(adapter);
6365 if (adapter->hw.phy.type == e1000_phy_igp_3) {
6366 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6367 } else if (hw->mac.type >= e1000_pch_lpt) {
6368 if (!(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6369 /* ULP does not support wake from unicast, multicast
6372 retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6378 /* Ensure that the appropriate bits are set in LPI_CTRL
6381 if ((hw->phy.type >= e1000_phy_i217) &&
6382 adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6385 retval = hw->phy.ops.acquire(hw);
6387 retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6390 if (adapter->eee_advert &
6391 hw->dev_spec.ich8lan.eee_lp_ability &
6392 I82579_EEE_100_SUPPORTED)
6393 lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6394 if (adapter->eee_advert &
6395 hw->dev_spec.ich8lan.eee_lp_ability &
6396 I82579_EEE_1000_SUPPORTED)
6397 lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6399 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6403 hw->phy.ops.release(hw);
6406 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6407 * would have already happened in close and is redundant.
6409 e1000e_release_hw_control(adapter);
6411 pci_clear_master(pdev);
6413 /* The pci-e switch on some quad port adapters will report a
6414 * correctable error when the MAC transitions from D0 to D3. To
6415 * prevent this we need to mask off the correctable errors on the
6416 * downstream port of the pci-e switch.
6418 * We don't have the associated upstream bridge while assigning
6419 * the PCI device into guest. For example, the KVM on power is
6422 if (adapter->flags & FLAG_IS_QUAD_PORT) {
6423 struct pci_dev *us_dev = pdev->bus->self;
6429 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6430 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6431 (devctl & ~PCI_EXP_DEVCTL_CERE));
6433 pci_save_state(pdev);
6434 pci_prepare_to_sleep(pdev);
6436 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6443 * __e1000e_disable_aspm - Disable ASPM states
6444 * @pdev: pointer to PCI device struct
6445 * @state: bit-mask of ASPM states to disable
6446 * @locked: indication if this context holds pci_bus_sem locked.
6448 * Some devices *must* have certain ASPM states disabled per hardware errata.
6450 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6452 struct pci_dev *parent = pdev->bus->self;
6453 u16 aspm_dis_mask = 0;
6454 u16 pdev_aspmc, parent_aspmc;
6457 case PCIE_LINK_STATE_L0S:
6458 case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6459 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6460 /* fall-through - can't have L1 without L0s */
6461 case PCIE_LINK_STATE_L1:
6462 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6468 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6469 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6472 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6474 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6477 /* Nothing to do if the ASPM states to be disabled already are */
6478 if (!(pdev_aspmc & aspm_dis_mask) &&
6479 (!parent || !(parent_aspmc & aspm_dis_mask)))
6482 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6483 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6485 (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6488 #ifdef CONFIG_PCIEASPM
6490 pci_disable_link_state_locked(pdev, state);
6492 pci_disable_link_state(pdev, state);
6494 /* Double-check ASPM control. If not disabled by the above, the
6495 * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6496 * not enabled); override by writing PCI config space directly.
6498 pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6499 pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6501 if (!(aspm_dis_mask & pdev_aspmc))
6505 /* Both device and parent should have the same ASPM setting.
6506 * Disable ASPM in downstream component first and then upstream.
6508 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6511 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6516 * e1000e_disable_aspm - Disable ASPM states.
6517 * @pdev: pointer to PCI device struct
6518 * @state: bit-mask of ASPM states to disable
6520 * This function acquires the pci_bus_sem!
6521 * Some devices *must* have certain ASPM states disabled per hardware errata.
6523 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6525 __e1000e_disable_aspm(pdev, state, 0);
6529 * e1000e_disable_aspm_locked Disable ASPM states.
6530 * @pdev: pointer to PCI device struct
6531 * @state: bit-mask of ASPM states to disable
6533 * This function must be called with pci_bus_sem acquired!
6534 * Some devices *must* have certain ASPM states disabled per hardware errata.
6536 static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6538 __e1000e_disable_aspm(pdev, state, 1);
6542 static int __e1000_resume(struct pci_dev *pdev)
6544 struct net_device *netdev = pci_get_drvdata(pdev);
6545 struct e1000_adapter *adapter = netdev_priv(netdev);
6546 struct e1000_hw *hw = &adapter->hw;
6547 u16 aspm_disable_flag = 0;
6549 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6550 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6551 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6552 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6553 if (aspm_disable_flag)
6554 e1000e_disable_aspm(pdev, aspm_disable_flag);
6556 pci_set_master(pdev);
6558 if (hw->mac.type >= e1000_pch2lan)
6559 e1000_resume_workarounds_pchlan(&adapter->hw);
6561 e1000e_power_up_phy(adapter);
6563 /* report the system wakeup cause from S3/S4 */
6564 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6567 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6569 e_info("PHY Wakeup cause - %s\n",
6570 phy_data & E1000_WUS_EX ? "Unicast Packet" :
6571 phy_data & E1000_WUS_MC ? "Multicast Packet" :
6572 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6573 phy_data & E1000_WUS_MAG ? "Magic Packet" :
6574 phy_data & E1000_WUS_LNKC ?
6575 "Link Status Change" : "other");
6577 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6579 u32 wus = er32(WUS);
6582 e_info("MAC Wakeup cause - %s\n",
6583 wus & E1000_WUS_EX ? "Unicast Packet" :
6584 wus & E1000_WUS_MC ? "Multicast Packet" :
6585 wus & E1000_WUS_BC ? "Broadcast Packet" :
6586 wus & E1000_WUS_MAG ? "Magic Packet" :
6587 wus & E1000_WUS_LNKC ? "Link Status Change" :
6593 e1000e_reset(adapter);
6595 e1000_init_manageability_pt(adapter);
6597 /* If the controller has AMT, do not set DRV_LOAD until the interface
6598 * is up. For all other cases, let the f/w know that the h/w is now
6599 * under the control of the driver.
6601 if (!(adapter->flags & FLAG_HAS_AMT))
6602 e1000e_get_hw_control(adapter);
6607 #ifdef CONFIG_PM_SLEEP
6608 static int e1000e_pm_thaw(struct device *dev)
6610 struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6611 struct e1000_adapter *adapter = netdev_priv(netdev);
6613 e1000e_set_interrupt_capability(adapter);
6614 if (netif_running(netdev)) {
6615 u32 err = e1000_request_irq(adapter);
6623 netif_device_attach(netdev);
6628 static int e1000e_pm_suspend(struct device *dev)
6630 struct pci_dev *pdev = to_pci_dev(dev);
6633 e1000e_flush_lpic(pdev);
6635 e1000e_pm_freeze(dev);
6637 rc = __e1000_shutdown(pdev, false);
6639 e1000e_pm_thaw(dev);
6644 static int e1000e_pm_resume(struct device *dev)
6646 struct pci_dev *pdev = to_pci_dev(dev);
6649 rc = __e1000_resume(pdev);
6653 return e1000e_pm_thaw(dev);
6655 #endif /* CONFIG_PM_SLEEP */
6657 static int e1000e_pm_runtime_idle(struct device *dev)
6659 struct pci_dev *pdev = to_pci_dev(dev);
6660 struct net_device *netdev = pci_get_drvdata(pdev);
6661 struct e1000_adapter *adapter = netdev_priv(netdev);
6664 eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6666 if (!e1000e_has_link(adapter)) {
6667 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6668 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6674 static int e1000e_pm_runtime_resume(struct device *dev)
6676 struct pci_dev *pdev = to_pci_dev(dev);
6677 struct net_device *netdev = pci_get_drvdata(pdev);
6678 struct e1000_adapter *adapter = netdev_priv(netdev);
6681 rc = __e1000_resume(pdev);
6685 if (netdev->flags & IFF_UP)
6691 static int e1000e_pm_runtime_suspend(struct device *dev)
6693 struct pci_dev *pdev = to_pci_dev(dev);
6694 struct net_device *netdev = pci_get_drvdata(pdev);
6695 struct e1000_adapter *adapter = netdev_priv(netdev);
6697 if (netdev->flags & IFF_UP) {
6698 int count = E1000_CHECK_RESET_COUNT;
6700 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6701 usleep_range(10000, 20000);
6703 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6705 /* Down the device without resetting the hardware */
6706 e1000e_down(adapter, false);
6709 if (__e1000_shutdown(pdev, true)) {
6710 e1000e_pm_runtime_resume(dev);
6716 #endif /* CONFIG_PM */
6718 static void e1000_shutdown(struct pci_dev *pdev)
6720 e1000e_flush_lpic(pdev);
6722 e1000e_pm_freeze(&pdev->dev);
6724 __e1000_shutdown(pdev, false);
6727 #ifdef CONFIG_NET_POLL_CONTROLLER
6729 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
6731 struct net_device *netdev = data;
6732 struct e1000_adapter *adapter = netdev_priv(netdev);
6734 if (adapter->msix_entries) {
6735 int vector, msix_irq;
6738 msix_irq = adapter->msix_entries[vector].vector;
6739 if (disable_hardirq(msix_irq))
6740 e1000_intr_msix_rx(msix_irq, netdev);
6741 enable_irq(msix_irq);
6744 msix_irq = adapter->msix_entries[vector].vector;
6745 if (disable_hardirq(msix_irq))
6746 e1000_intr_msix_tx(msix_irq, netdev);
6747 enable_irq(msix_irq);
6750 msix_irq = adapter->msix_entries[vector].vector;
6751 if (disable_hardirq(msix_irq))
6752 e1000_msix_other(msix_irq, netdev);
6753 enable_irq(msix_irq);
6761 * @netdev: network interface device structure
6763 * Polling 'interrupt' - used by things like netconsole to send skbs
6764 * without having to re-enable interrupts. It's not called while
6765 * the interrupt routine is executing.
6767 static void e1000_netpoll(struct net_device *netdev)
6769 struct e1000_adapter *adapter = netdev_priv(netdev);
6771 switch (adapter->int_mode) {
6772 case E1000E_INT_MODE_MSIX:
6773 e1000_intr_msix(adapter->pdev->irq, netdev);
6775 case E1000E_INT_MODE_MSI:
6776 if (disable_hardirq(adapter->pdev->irq))
6777 e1000_intr_msi(adapter->pdev->irq, netdev);
6778 enable_irq(adapter->pdev->irq);
6780 default: /* E1000E_INT_MODE_LEGACY */
6781 if (disable_hardirq(adapter->pdev->irq))
6782 e1000_intr(adapter->pdev->irq, netdev);
6783 enable_irq(adapter->pdev->irq);
6790 * e1000_io_error_detected - called when PCI error is detected
6791 * @pdev: Pointer to PCI device
6792 * @state: The current pci connection state
6794 * This function is called after a PCI bus error affecting
6795 * this device has been detected.
6797 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
6798 pci_channel_state_t state)
6800 struct net_device *netdev = pci_get_drvdata(pdev);
6801 struct e1000_adapter *adapter = netdev_priv(netdev);
6803 netif_device_detach(netdev);
6805 if (state == pci_channel_io_perm_failure)
6806 return PCI_ERS_RESULT_DISCONNECT;
6808 if (netif_running(netdev))
6809 e1000e_down(adapter, true);
6810 pci_disable_device(pdev);
6812 /* Request a slot slot reset. */
6813 return PCI_ERS_RESULT_NEED_RESET;
6817 * e1000_io_slot_reset - called after the pci bus has been reset.
6818 * @pdev: Pointer to PCI device
6820 * Restart the card from scratch, as if from a cold-boot. Implementation
6821 * resembles the first-half of the e1000e_pm_resume routine.
6823 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
6825 struct net_device *netdev = pci_get_drvdata(pdev);
6826 struct e1000_adapter *adapter = netdev_priv(netdev);
6827 struct e1000_hw *hw = &adapter->hw;
6828 u16 aspm_disable_flag = 0;
6830 pci_ers_result_t result;
6832 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6833 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6834 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6835 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6836 if (aspm_disable_flag)
6837 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
6839 err = pci_enable_device_mem(pdev);
6842 "Cannot re-enable PCI device after reset.\n");
6843 result = PCI_ERS_RESULT_DISCONNECT;
6845 pdev->state_saved = true;
6846 pci_restore_state(pdev);
6847 pci_set_master(pdev);
6849 pci_enable_wake(pdev, PCI_D3hot, 0);
6850 pci_enable_wake(pdev, PCI_D3cold, 0);
6852 e1000e_reset(adapter);
6854 result = PCI_ERS_RESULT_RECOVERED;
6857 pci_cleanup_aer_uncorrect_error_status(pdev);
6863 * e1000_io_resume - called when traffic can start flowing again.
6864 * @pdev: Pointer to PCI device
6866 * This callback is called when the error recovery driver tells us that
6867 * its OK to resume normal operation. Implementation resembles the
6868 * second-half of the e1000e_pm_resume routine.
6870 static void e1000_io_resume(struct pci_dev *pdev)
6872 struct net_device *netdev = pci_get_drvdata(pdev);
6873 struct e1000_adapter *adapter = netdev_priv(netdev);
6875 e1000_init_manageability_pt(adapter);
6877 if (netif_running(netdev))
6880 netif_device_attach(netdev);
6882 /* If the controller has AMT, do not set DRV_LOAD until the interface
6883 * is up. For all other cases, let the f/w know that the h/w is now
6884 * under the control of the driver.
6886 if (!(adapter->flags & FLAG_HAS_AMT))
6887 e1000e_get_hw_control(adapter);
6890 static void e1000_print_device_info(struct e1000_adapter *adapter)
6892 struct e1000_hw *hw = &adapter->hw;
6893 struct net_device *netdev = adapter->netdev;
6895 u8 pba_str[E1000_PBANUM_LENGTH];
6897 /* print bus type/speed/width info */
6898 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
6900 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
6904 e_info("Intel(R) PRO/%s Network Connection\n",
6905 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
6906 ret_val = e1000_read_pba_string_generic(hw, pba_str,
6907 E1000_PBANUM_LENGTH);
6909 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
6910 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
6911 hw->mac.type, hw->phy.type, pba_str);
6914 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
6916 struct e1000_hw *hw = &adapter->hw;
6920 if (hw->mac.type != e1000_82573)
6923 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6925 if (!ret_val && (!(buf & BIT(0)))) {
6926 /* Deep Smart Power Down (DSPD) */
6927 dev_warn(&adapter->pdev->dev,
6928 "Warning: detected DSPD enabled in EEPROM\n");
6932 static netdev_features_t e1000_fix_features(struct net_device *netdev,
6933 netdev_features_t features)
6935 struct e1000_adapter *adapter = netdev_priv(netdev);
6936 struct e1000_hw *hw = &adapter->hw;
6938 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6939 if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
6940 features &= ~NETIF_F_RXFCS;
6942 /* Since there is no support for separate Rx/Tx vlan accel
6943 * enable/disable make sure Tx flag is always in same state as Rx.
6945 if (features & NETIF_F_HW_VLAN_CTAG_RX)
6946 features |= NETIF_F_HW_VLAN_CTAG_TX;
6948 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
6953 static int e1000_set_features(struct net_device *netdev,
6954 netdev_features_t features)
6956 struct e1000_adapter *adapter = netdev_priv(netdev);
6957 netdev_features_t changed = features ^ netdev->features;
6959 if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6960 adapter->flags |= FLAG_TSO_FORCE;
6962 if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
6963 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6967 if (changed & NETIF_F_RXFCS) {
6968 if (features & NETIF_F_RXFCS) {
6969 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6971 /* We need to take it back to defaults, which might mean
6972 * stripping is still disabled at the adapter level.
6974 if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6975 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6977 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6981 netdev->features = features;
6983 if (netif_running(netdev))
6984 e1000e_reinit_locked(adapter);
6986 e1000e_reset(adapter);
6991 static const struct net_device_ops e1000e_netdev_ops = {
6992 .ndo_open = e1000e_open,
6993 .ndo_stop = e1000e_close,
6994 .ndo_start_xmit = e1000_xmit_frame,
6995 .ndo_get_stats64 = e1000e_get_stats64,
6996 .ndo_set_rx_mode = e1000e_set_rx_mode,
6997 .ndo_set_mac_address = e1000_set_mac,
6998 .ndo_change_mtu = e1000_change_mtu,
6999 .ndo_do_ioctl = e1000_ioctl,
7000 .ndo_tx_timeout = e1000_tx_timeout,
7001 .ndo_validate_addr = eth_validate_addr,
7003 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
7004 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
7005 #ifdef CONFIG_NET_POLL_CONTROLLER
7006 .ndo_poll_controller = e1000_netpoll,
7008 .ndo_set_features = e1000_set_features,
7009 .ndo_fix_features = e1000_fix_features,
7010 .ndo_features_check = passthru_features_check,
7014 * e1000_probe - Device Initialization Routine
7015 * @pdev: PCI device information struct
7016 * @ent: entry in e1000_pci_tbl
7018 * Returns 0 on success, negative on failure
7020 * e1000_probe initializes an adapter identified by a pci_dev structure.
7021 * The OS initialization, configuring of the adapter private structure,
7022 * and a hardware reset occur.
7024 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7026 struct net_device *netdev;
7027 struct e1000_adapter *adapter;
7028 struct e1000_hw *hw;
7029 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7030 resource_size_t mmio_start, mmio_len;
7031 resource_size_t flash_start, flash_len;
7032 static int cards_found;
7033 u16 aspm_disable_flag = 0;
7034 int bars, i, err, pci_using_dac;
7035 u16 eeprom_data = 0;
7036 u16 eeprom_apme_mask = E1000_EEPROM_APME;
7039 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7040 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7041 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7042 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7043 if (aspm_disable_flag)
7044 e1000e_disable_aspm(pdev, aspm_disable_flag);
7046 err = pci_enable_device_mem(pdev);
7051 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7055 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7058 "No usable DMA configuration, aborting\n");
7063 bars = pci_select_bars(pdev, IORESOURCE_MEM);
7064 err = pci_request_selected_regions_exclusive(pdev, bars,
7065 e1000e_driver_name);
7069 /* AER (Advanced Error Reporting) hooks */
7070 pci_enable_pcie_error_reporting(pdev);
7072 pci_set_master(pdev);
7073 /* PCI config space info */
7074 err = pci_save_state(pdev);
7076 goto err_alloc_etherdev;
7079 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7081 goto err_alloc_etherdev;
7083 SET_NETDEV_DEV(netdev, &pdev->dev);
7085 netdev->irq = pdev->irq;
7087 pci_set_drvdata(pdev, netdev);
7088 adapter = netdev_priv(netdev);
7090 adapter->netdev = netdev;
7091 adapter->pdev = pdev;
7093 adapter->pba = ei->pba;
7094 adapter->flags = ei->flags;
7095 adapter->flags2 = ei->flags2;
7096 adapter->hw.adapter = adapter;
7097 adapter->hw.mac.type = ei->mac;
7098 adapter->max_hw_frame_size = ei->max_hw_frame_size;
7099 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7101 mmio_start = pci_resource_start(pdev, 0);
7102 mmio_len = pci_resource_len(pdev, 0);
7105 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7106 if (!adapter->hw.hw_addr)
7109 if ((adapter->flags & FLAG_HAS_FLASH) &&
7110 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7111 (hw->mac.type < e1000_pch_spt)) {
7112 flash_start = pci_resource_start(pdev, 1);
7113 flash_len = pci_resource_len(pdev, 1);
7114 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7115 if (!adapter->hw.flash_address)
7119 /* Set default EEE advertisement */
7120 if (adapter->flags2 & FLAG2_HAS_EEE)
7121 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7123 /* construct the net_device struct */
7124 netdev->netdev_ops = &e1000e_netdev_ops;
7125 e1000e_set_ethtool_ops(netdev);
7126 netdev->watchdog_timeo = 5 * HZ;
7127 netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7128 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7130 netdev->mem_start = mmio_start;
7131 netdev->mem_end = mmio_start + mmio_len;
7133 adapter->bd_number = cards_found++;
7135 e1000e_check_options(adapter);
7137 /* setup adapter struct */
7138 err = e1000_sw_init(adapter);
7142 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7143 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7144 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7146 err = ei->get_variants(adapter);
7150 if ((adapter->flags & FLAG_IS_ICH) &&
7151 (adapter->flags & FLAG_READ_ONLY_NVM) &&
7152 (hw->mac.type < e1000_pch_spt))
7153 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7155 hw->mac.ops.get_bus_info(&adapter->hw);
7157 adapter->hw.phy.autoneg_wait_to_complete = 0;
7159 /* Copper options */
7160 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7161 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7162 adapter->hw.phy.disable_polarity_correction = 0;
7163 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7166 if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7167 dev_info(&pdev->dev,
7168 "PHY reset is blocked due to SOL/IDER session.\n");
7170 /* Set initial default active device features */
7171 netdev->features = (NETIF_F_SG |
7172 NETIF_F_HW_VLAN_CTAG_RX |
7173 NETIF_F_HW_VLAN_CTAG_TX |
7180 /* Set user-changeable features (subset of all device features) */
7181 netdev->hw_features = netdev->features;
7182 netdev->hw_features |= NETIF_F_RXFCS;
7183 netdev->priv_flags |= IFF_SUPP_NOFCS;
7184 netdev->hw_features |= NETIF_F_RXALL;
7186 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7187 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7189 netdev->vlan_features |= (NETIF_F_SG |
7194 netdev->priv_flags |= IFF_UNICAST_FLT;
7196 if (pci_using_dac) {
7197 netdev->features |= NETIF_F_HIGHDMA;
7198 netdev->vlan_features |= NETIF_F_HIGHDMA;
7201 /* MTU range: 68 - max_hw_frame_size */
7202 netdev->min_mtu = ETH_MIN_MTU;
7203 netdev->max_mtu = adapter->max_hw_frame_size -
7204 (VLAN_ETH_HLEN + ETH_FCS_LEN);
7206 if (e1000e_enable_mng_pass_thru(&adapter->hw))
7207 adapter->flags |= FLAG_MNG_PT_ENABLED;
7209 /* before reading the NVM, reset the controller to
7210 * put the device in a known good starting state
7212 adapter->hw.mac.ops.reset_hw(&adapter->hw);
7214 /* systems with ASPM and others may see the checksum fail on the first
7215 * attempt. Let's give it a few tries
7218 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7221 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7227 e1000_eeprom_checks(adapter);
7229 /* copy the MAC address */
7230 if (e1000e_read_mac_addr(&adapter->hw))
7232 "NVM Read Error while reading MAC address\n");
7234 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7236 if (!is_valid_ether_addr(netdev->dev_addr)) {
7237 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7243 timer_setup(&adapter->watchdog_timer, e1000_watchdog, 0);
7244 timer_setup(&adapter->phy_info_timer, e1000_update_phy_info, 0);
7246 INIT_WORK(&adapter->reset_task, e1000_reset_task);
7247 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7248 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7249 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7250 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7252 /* Initialize link parameters. User can change them with ethtool */
7253 adapter->hw.mac.autoneg = 1;
7254 adapter->fc_autoneg = true;
7255 adapter->hw.fc.requested_mode = e1000_fc_default;
7256 adapter->hw.fc.current_mode = e1000_fc_default;
7257 adapter->hw.phy.autoneg_advertised = 0x2f;
7259 /* Initial Wake on LAN setting - If APM wake is enabled in
7260 * the EEPROM, enable the ACPI Magic Packet filter
7262 if (adapter->flags & FLAG_APME_IN_WUC) {
7263 /* APME bit in EEPROM is mapped to WUC.APME */
7264 eeprom_data = er32(WUC);
7265 eeprom_apme_mask = E1000_WUC_APME;
7266 if ((hw->mac.type > e1000_ich10lan) &&
7267 (eeprom_data & E1000_WUC_PHY_WAKE))
7268 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7269 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7270 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7271 (adapter->hw.bus.func == 1))
7272 ret_val = e1000_read_nvm(&adapter->hw,
7273 NVM_INIT_CONTROL3_PORT_B,
7276 ret_val = e1000_read_nvm(&adapter->hw,
7277 NVM_INIT_CONTROL3_PORT_A,
7281 /* fetch WoL from EEPROM */
7283 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7284 else if (eeprom_data & eeprom_apme_mask)
7285 adapter->eeprom_wol |= E1000_WUFC_MAG;
7287 /* now that we have the eeprom settings, apply the special cases
7288 * where the eeprom may be wrong or the board simply won't support
7289 * wake on lan on a particular port
7291 if (!(adapter->flags & FLAG_HAS_WOL))
7292 adapter->eeprom_wol = 0;
7294 /* initialize the wol settings based on the eeprom settings */
7295 adapter->wol = adapter->eeprom_wol;
7297 /* make sure adapter isn't asleep if manageability is enabled */
7298 if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7299 (hw->mac.ops.check_mng_mode(hw)))
7300 device_wakeup_enable(&pdev->dev);
7302 /* save off EEPROM version number */
7303 ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7306 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7307 adapter->eeprom_vers = 0;
7310 /* init PTP hardware clock */
7311 e1000e_ptp_init(adapter);
7313 /* reset the hardware with the new settings */
7314 e1000e_reset(adapter);
7316 /* If the controller has AMT, do not set DRV_LOAD until the interface
7317 * is up. For all other cases, let the f/w know that the h/w is now
7318 * under the control of the driver.
7320 if (!(adapter->flags & FLAG_HAS_AMT))
7321 e1000e_get_hw_control(adapter);
7323 strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7324 err = register_netdev(netdev);
7328 /* carrier off reporting is important to ethtool even BEFORE open */
7329 netif_carrier_off(netdev);
7331 e1000_print_device_info(adapter);
7333 if (pci_dev_run_wake(pdev))
7334 pm_runtime_put_noidle(&pdev->dev);
7339 if (!(adapter->flags & FLAG_HAS_AMT))
7340 e1000e_release_hw_control(adapter);
7342 if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7343 e1000_phy_hw_reset(&adapter->hw);
7345 kfree(adapter->tx_ring);
7346 kfree(adapter->rx_ring);
7348 if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7349 iounmap(adapter->hw.flash_address);
7350 e1000e_reset_interrupt_capability(adapter);
7352 iounmap(adapter->hw.hw_addr);
7354 free_netdev(netdev);
7356 pci_release_mem_regions(pdev);
7359 pci_disable_device(pdev);
7364 * e1000_remove - Device Removal Routine
7365 * @pdev: PCI device information struct
7367 * e1000_remove is called by the PCI subsystem to alert the driver
7368 * that it should release a PCI device. The could be caused by a
7369 * Hot-Plug event, or because the driver is going to be removed from
7372 static void e1000_remove(struct pci_dev *pdev)
7374 struct net_device *netdev = pci_get_drvdata(pdev);
7375 struct e1000_adapter *adapter = netdev_priv(netdev);
7376 bool down = test_bit(__E1000_DOWN, &adapter->state);
7378 e1000e_ptp_remove(adapter);
7380 /* The timers may be rescheduled, so explicitly disable them
7381 * from being rescheduled.
7384 set_bit(__E1000_DOWN, &adapter->state);
7385 del_timer_sync(&adapter->watchdog_timer);
7386 del_timer_sync(&adapter->phy_info_timer);
7388 cancel_work_sync(&adapter->reset_task);
7389 cancel_work_sync(&adapter->watchdog_task);
7390 cancel_work_sync(&adapter->downshift_task);
7391 cancel_work_sync(&adapter->update_phy_task);
7392 cancel_work_sync(&adapter->print_hang_task);
7394 if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7395 cancel_work_sync(&adapter->tx_hwtstamp_work);
7396 if (adapter->tx_hwtstamp_skb) {
7397 dev_consume_skb_any(adapter->tx_hwtstamp_skb);
7398 adapter->tx_hwtstamp_skb = NULL;
7402 /* Don't lie to e1000_close() down the road. */
7404 clear_bit(__E1000_DOWN, &adapter->state);
7405 unregister_netdev(netdev);
7407 if (pci_dev_run_wake(pdev))
7408 pm_runtime_get_noresume(&pdev->dev);
7410 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7411 * would have already happened in close and is redundant.
7413 e1000e_release_hw_control(adapter);
7415 e1000e_reset_interrupt_capability(adapter);
7416 kfree(adapter->tx_ring);
7417 kfree(adapter->rx_ring);
7419 iounmap(adapter->hw.hw_addr);
7420 if ((adapter->hw.flash_address) &&
7421 (adapter->hw.mac.type < e1000_pch_spt))
7422 iounmap(adapter->hw.flash_address);
7423 pci_release_mem_regions(pdev);
7425 free_netdev(netdev);
7428 pci_disable_pcie_error_reporting(pdev);
7430 pci_disable_device(pdev);
7433 /* PCI Error Recovery (ERS) */
7434 static const struct pci_error_handlers e1000_err_handler = {
7435 .error_detected = e1000_io_error_detected,
7436 .slot_reset = e1000_io_slot_reset,
7437 .resume = e1000_io_resume,
7440 static const struct pci_device_id e1000_pci_tbl[] = {
7441 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7442 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7443 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7444 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7446 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7447 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7448 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7449 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7450 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7452 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7453 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7454 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7455 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7457 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7458 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7459 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7461 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7462 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7463 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7465 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7466 board_80003es2lan },
7467 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7468 board_80003es2lan },
7469 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7470 board_80003es2lan },
7471 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7472 board_80003es2lan },
7474 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7475 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7476 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7477 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7478 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7479 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7480 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7481 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7483 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7484 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7485 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7486 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7487 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7488 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7489 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7490 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7491 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7493 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7494 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7495 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7497 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7498 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7499 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7501 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7502 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7503 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7504 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7506 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7507 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7509 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7510 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7511 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7512 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7513 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7514 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7515 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7516 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7517 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7518 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7519 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7520 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7521 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7522 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7523 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7524 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7525 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7526 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7527 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7528 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7529 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7530 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM8), board_pch_cnp },
7531 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V8), board_pch_cnp },
7532 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM9), board_pch_cnp },
7533 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V9), board_pch_cnp },
7535 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7537 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7539 static const struct dev_pm_ops e1000_pm_ops = {
7540 #ifdef CONFIG_PM_SLEEP
7541 .suspend = e1000e_pm_suspend,
7542 .resume = e1000e_pm_resume,
7543 .freeze = e1000e_pm_freeze,
7544 .thaw = e1000e_pm_thaw,
7545 .poweroff = e1000e_pm_suspend,
7546 .restore = e1000e_pm_resume,
7548 SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7549 e1000e_pm_runtime_idle)
7552 /* PCI Device API Driver */
7553 static struct pci_driver e1000_driver = {
7554 .name = e1000e_driver_name,
7555 .id_table = e1000_pci_tbl,
7556 .probe = e1000_probe,
7557 .remove = e1000_remove,
7559 .pm = &e1000_pm_ops,
7561 .shutdown = e1000_shutdown,
7562 .err_handler = &e1000_err_handler
7566 * e1000_init_module - Driver Registration Routine
7568 * e1000_init_module is the first routine called when the driver is
7569 * loaded. All it does is register with the PCI subsystem.
7571 static int __init e1000_init_module(void)
7573 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
7574 e1000e_driver_version);
7575 pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7577 return pci_register_driver(&e1000_driver);
7579 module_init(e1000_init_module);
7582 * e1000_exit_module - Driver Exit Cleanup Routine
7584 * e1000_exit_module is called just before the driver is removed
7587 static void __exit e1000_exit_module(void)
7589 pci_unregister_driver(&e1000_driver);
7591 module_exit(e1000_exit_module);
7593 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7594 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7595 MODULE_LICENSE("GPL");
7596 MODULE_VERSION(DRV_VERSION);