1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/interrupt.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/mii.h>
46 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/cpu.h>
49 #include <linux/smp.h>
50 #include <linux/pm_qos.h>
51 #include <linux/pm_runtime.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
57 #define DRV_EXTRAVERSION "-k"
59 #define DRV_VERSION "2.1.4" DRV_EXTRAVERSION
60 char e1000e_driver_name
[] = "e1000e";
61 const char e1000e_driver_version
[] = DRV_VERSION
;
63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
64 static int debug
= -1;
65 module_param(debug
, int, 0);
66 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
68 static void e1000e_disable_aspm(struct pci_dev
*pdev
, u16 state
);
70 static const struct e1000_info
*e1000_info_tbl
[] = {
71 [board_82571
] = &e1000_82571_info
,
72 [board_82572
] = &e1000_82572_info
,
73 [board_82573
] = &e1000_82573_info
,
74 [board_82574
] = &e1000_82574_info
,
75 [board_82583
] = &e1000_82583_info
,
76 [board_80003es2lan
] = &e1000_es2_info
,
77 [board_ich8lan
] = &e1000_ich8_info
,
78 [board_ich9lan
] = &e1000_ich9_info
,
79 [board_ich10lan
] = &e1000_ich10_info
,
80 [board_pchlan
] = &e1000_pch_info
,
81 [board_pch2lan
] = &e1000_pch2_info
,
82 [board_pch_lpt
] = &e1000_pch_lpt_info
,
85 struct e1000_reg_info
{
90 #define E1000_RDFH 0x02410 /* Rx Data FIFO Head - RW */
91 #define E1000_RDFT 0x02418 /* Rx Data FIFO Tail - RW */
92 #define E1000_RDFHS 0x02420 /* Rx Data FIFO Head Saved - RW */
93 #define E1000_RDFTS 0x02428 /* Rx Data FIFO Tail Saved - RW */
94 #define E1000_RDFPC 0x02430 /* Rx Data FIFO Packet Count - RW */
96 #define E1000_TDFH 0x03410 /* Tx Data FIFO Head - RW */
97 #define E1000_TDFT 0x03418 /* Tx Data FIFO Tail - RW */
98 #define E1000_TDFHS 0x03420 /* Tx Data FIFO Head Saved - RW */
99 #define E1000_TDFTS 0x03428 /* Tx Data FIFO Tail Saved - RW */
100 #define E1000_TDFPC 0x03430 /* Tx Data FIFO Packet Count - RW */
102 static const struct e1000_reg_info e1000_reg_info_tbl
[] = {
104 /* General Registers */
105 {E1000_CTRL
, "CTRL"},
106 {E1000_STATUS
, "STATUS"},
107 {E1000_CTRL_EXT
, "CTRL_EXT"},
109 /* Interrupt Registers */
113 {E1000_RCTL
, "RCTL"},
114 {E1000_RDLEN(0), "RDLEN"},
115 {E1000_RDH(0), "RDH"},
116 {E1000_RDT(0), "RDT"},
117 {E1000_RDTR
, "RDTR"},
118 {E1000_RXDCTL(0), "RXDCTL"},
120 {E1000_RDBAL(0), "RDBAL"},
121 {E1000_RDBAH(0), "RDBAH"},
122 {E1000_RDFH
, "RDFH"},
123 {E1000_RDFT
, "RDFT"},
124 {E1000_RDFHS
, "RDFHS"},
125 {E1000_RDFTS
, "RDFTS"},
126 {E1000_RDFPC
, "RDFPC"},
129 {E1000_TCTL
, "TCTL"},
130 {E1000_TDBAL(0), "TDBAL"},
131 {E1000_TDBAH(0), "TDBAH"},
132 {E1000_TDLEN(0), "TDLEN"},
133 {E1000_TDH(0), "TDH"},
134 {E1000_TDT(0), "TDT"},
135 {E1000_TIDV
, "TIDV"},
136 {E1000_TXDCTL(0), "TXDCTL"},
137 {E1000_TADV
, "TADV"},
138 {E1000_TARC(0), "TARC"},
139 {E1000_TDFH
, "TDFH"},
140 {E1000_TDFT
, "TDFT"},
141 {E1000_TDFHS
, "TDFHS"},
142 {E1000_TDFTS
, "TDFTS"},
143 {E1000_TDFPC
, "TDFPC"},
145 /* List Terminator */
150 * e1000_regdump - register printout routine
152 static void e1000_regdump(struct e1000_hw
*hw
, struct e1000_reg_info
*reginfo
)
158 switch (reginfo
->ofs
) {
159 case E1000_RXDCTL(0):
160 for (n
= 0; n
< 2; n
++)
161 regs
[n
] = __er32(hw
, E1000_RXDCTL(n
));
163 case E1000_TXDCTL(0):
164 for (n
= 0; n
< 2; n
++)
165 regs
[n
] = __er32(hw
, E1000_TXDCTL(n
));
168 for (n
= 0; n
< 2; n
++)
169 regs
[n
] = __er32(hw
, E1000_TARC(n
));
172 pr_info("%-15s %08x\n",
173 reginfo
->name
, __er32(hw
, reginfo
->ofs
));
177 snprintf(rname
, 16, "%s%s", reginfo
->name
, "[0-1]");
178 pr_info("%-15s %08x %08x\n", rname
, regs
[0], regs
[1]);
181 static void e1000e_dump_ps_pages(struct e1000_adapter
*adapter
,
182 struct e1000_buffer
*bi
)
185 struct e1000_ps_page
*ps_page
;
187 for (i
= 0; i
< adapter
->rx_ps_pages
; i
++) {
188 ps_page
= &bi
->ps_pages
[i
];
191 pr_info("packet dump for ps_page %d:\n", i
);
192 print_hex_dump(KERN_INFO
, "", DUMP_PREFIX_ADDRESS
,
193 16, 1, page_address(ps_page
->page
),
200 * e1000e_dump - Print registers, Tx-ring and Rx-ring
202 static void e1000e_dump(struct e1000_adapter
*adapter
)
204 struct net_device
*netdev
= adapter
->netdev
;
205 struct e1000_hw
*hw
= &adapter
->hw
;
206 struct e1000_reg_info
*reginfo
;
207 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
208 struct e1000_tx_desc
*tx_desc
;
213 struct e1000_buffer
*buffer_info
;
214 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
215 union e1000_rx_desc_packet_split
*rx_desc_ps
;
216 union e1000_rx_desc_extended
*rx_desc
;
226 if (!netif_msg_hw(adapter
))
229 /* Print netdevice Info */
231 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
232 pr_info("Device Name state trans_start last_rx\n");
233 pr_info("%-15s %016lX %016lX %016lX\n",
234 netdev
->name
, netdev
->state
, netdev
->trans_start
,
238 /* Print Registers */
239 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
240 pr_info(" Register Name Value\n");
241 for (reginfo
= (struct e1000_reg_info
*)e1000_reg_info_tbl
;
242 reginfo
->name
; reginfo
++) {
243 e1000_regdump(hw
, reginfo
);
246 /* Print Tx Ring Summary */
247 if (!netdev
|| !netif_running(netdev
))
250 dev_info(&adapter
->pdev
->dev
, "Tx Ring Summary\n");
251 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
252 buffer_info
= &tx_ring
->buffer_info
[tx_ring
->next_to_clean
];
253 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
254 0, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
255 (unsigned long long)buffer_info
->dma
,
257 buffer_info
->next_to_watch
,
258 (unsigned long long)buffer_info
->time_stamp
);
261 if (!netif_msg_tx_done(adapter
))
262 goto rx_ring_summary
;
264 dev_info(&adapter
->pdev
->dev
, "Tx Ring Dump\n");
266 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
268 * Legacy Transmit Descriptor
269 * +--------------------------------------------------------------+
270 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
271 * +--------------------------------------------------------------+
272 * 8 | Special | CSS | Status | CMD | CSO | Length |
273 * +--------------------------------------------------------------+
274 * 63 48 47 36 35 32 31 24 23 16 15 0
276 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
277 * 63 48 47 40 39 32 31 16 15 8 7 0
278 * +----------------------------------------------------------------+
279 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
280 * +----------------------------------------------------------------+
281 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
282 * +----------------------------------------------------------------+
283 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
285 * Extended Data Descriptor (DTYP=0x1)
286 * +----------------------------------------------------------------+
287 * 0 | Buffer Address [63:0] |
288 * +----------------------------------------------------------------+
289 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
290 * +----------------------------------------------------------------+
291 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
293 pr_info("Tl[desc] [address 63:0 ] [SpeCssSCmCsLen] [bi->dma ] leng ntw timestamp bi->skb <-- Legacy format\n");
294 pr_info("Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Context format\n");
295 pr_info("Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen] [bi->dma ] leng ntw timestamp bi->skb <-- Ext Data format\n");
296 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
297 const char *next_desc
;
298 tx_desc
= E1000_TX_DESC(*tx_ring
, i
);
299 buffer_info
= &tx_ring
->buffer_info
[i
];
300 u0
= (struct my_u0
*)tx_desc
;
301 if (i
== tx_ring
->next_to_use
&& i
== tx_ring
->next_to_clean
)
302 next_desc
= " NTC/U";
303 else if (i
== tx_ring
->next_to_use
)
305 else if (i
== tx_ring
->next_to_clean
)
309 pr_info("T%c[0x%03X] %016llX %016llX %016llX %04X %3X %016llX %p%s\n",
310 (!(le64_to_cpu(u0
->b
) & (1 << 29)) ? 'l' :
311 ((le64_to_cpu(u0
->b
) & (1 << 20)) ? 'd' : 'c')),
313 (unsigned long long)le64_to_cpu(u0
->a
),
314 (unsigned long long)le64_to_cpu(u0
->b
),
315 (unsigned long long)buffer_info
->dma
,
316 buffer_info
->length
, buffer_info
->next_to_watch
,
317 (unsigned long long)buffer_info
->time_stamp
,
318 buffer_info
->skb
, next_desc
);
320 if (netif_msg_pktdata(adapter
) && buffer_info
->skb
)
321 print_hex_dump(KERN_INFO
, "", DUMP_PREFIX_ADDRESS
,
322 16, 1, buffer_info
->skb
->data
,
323 buffer_info
->skb
->len
, true);
326 /* Print Rx Ring Summary */
328 dev_info(&adapter
->pdev
->dev
, "Rx Ring Summary\n");
329 pr_info("Queue [NTU] [NTC]\n");
330 pr_info(" %5d %5X %5X\n",
331 0, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
334 if (!netif_msg_rx_status(adapter
))
337 dev_info(&adapter
->pdev
->dev
, "Rx Ring Dump\n");
338 switch (adapter
->rx_ps_pages
) {
342 /* [Extended] Packet Split Receive Descriptor Format
344 * +-----------------------------------------------------+
345 * 0 | Buffer Address 0 [63:0] |
346 * +-----------------------------------------------------+
347 * 8 | Buffer Address 1 [63:0] |
348 * +-----------------------------------------------------+
349 * 16 | Buffer Address 2 [63:0] |
350 * +-----------------------------------------------------+
351 * 24 | Buffer Address 3 [63:0] |
352 * +-----------------------------------------------------+
354 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");
355 /* [Extended] Receive Descriptor (Write-Back) Format
357 * 63 48 47 32 31 13 12 8 7 4 3 0
358 * +------------------------------------------------------+
359 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
360 * | Checksum | Ident | | Queue | | Type |
361 * +------------------------------------------------------+
362 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
363 * +------------------------------------------------------+
364 * 63 48 47 32 31 20 19 0
366 pr_info("RWB[desc] [ck ipid mrqhsh] [vl l0 ee es] [ l3 l2 l1 hs] [reserved ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
367 for (i
= 0; i
< rx_ring
->count
; i
++) {
368 const char *next_desc
;
369 buffer_info
= &rx_ring
->buffer_info
[i
];
370 rx_desc_ps
= E1000_RX_DESC_PS(*rx_ring
, i
);
371 u1
= (struct my_u1
*)rx_desc_ps
;
373 le32_to_cpu(rx_desc_ps
->wb
.middle
.status_error
);
375 if (i
== rx_ring
->next_to_use
)
377 else if (i
== rx_ring
->next_to_clean
)
382 if (staterr
& E1000_RXD_STAT_DD
) {
383 /* Descriptor Done */
384 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX ---------------- %p%s\n",
386 (unsigned long long)le64_to_cpu(u1
->a
),
387 (unsigned long long)le64_to_cpu(u1
->b
),
388 (unsigned long long)le64_to_cpu(u1
->c
),
389 (unsigned long long)le64_to_cpu(u1
->d
),
390 buffer_info
->skb
, next_desc
);
392 pr_info("%s[0x%03X] %016llX %016llX %016llX %016llX %016llX %p%s\n",
394 (unsigned long long)le64_to_cpu(u1
->a
),
395 (unsigned long long)le64_to_cpu(u1
->b
),
396 (unsigned long long)le64_to_cpu(u1
->c
),
397 (unsigned long long)le64_to_cpu(u1
->d
),
398 (unsigned long long)buffer_info
->dma
,
399 buffer_info
->skb
, next_desc
);
401 if (netif_msg_pktdata(adapter
))
402 e1000e_dump_ps_pages(adapter
,
409 /* Extended Receive Descriptor (Read) Format
411 * +-----------------------------------------------------+
412 * 0 | Buffer Address [63:0] |
413 * +-----------------------------------------------------+
415 * +-----------------------------------------------------+
417 pr_info("R [desc] [buf addr 63:0 ] [reserved 63:0 ] [bi->dma ] [bi->skb] <-- Ext (Read) format\n");
418 /* Extended Receive Descriptor (Write-Back) Format
420 * 63 48 47 32 31 24 23 4 3 0
421 * +------------------------------------------------------+
423 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
424 * | Packet | IP | | | Type |
425 * | Checksum | Ident | | | |
426 * +------------------------------------------------------+
427 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
428 * +------------------------------------------------------+
429 * 63 48 47 32 31 20 19 0
431 pr_info("RWB[desc] [cs ipid mrq] [vt ln xe xs] [bi->skb] <-- Ext (Write-Back) format\n");
433 for (i
= 0; i
< rx_ring
->count
; i
++) {
434 const char *next_desc
;
436 buffer_info
= &rx_ring
->buffer_info
[i
];
437 rx_desc
= E1000_RX_DESC_EXT(*rx_ring
, i
);
438 u1
= (struct my_u1
*)rx_desc
;
439 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
441 if (i
== rx_ring
->next_to_use
)
443 else if (i
== rx_ring
->next_to_clean
)
448 if (staterr
& E1000_RXD_STAT_DD
) {
449 /* Descriptor Done */
450 pr_info("%s[0x%03X] %016llX %016llX ---------------- %p%s\n",
452 (unsigned long long)le64_to_cpu(u1
->a
),
453 (unsigned long long)le64_to_cpu(u1
->b
),
454 buffer_info
->skb
, next_desc
);
456 pr_info("%s[0x%03X] %016llX %016llX %016llX %p%s\n",
458 (unsigned long long)le64_to_cpu(u1
->a
),
459 (unsigned long long)le64_to_cpu(u1
->b
),
460 (unsigned long long)buffer_info
->dma
,
461 buffer_info
->skb
, next_desc
);
463 if (netif_msg_pktdata(adapter
) &&
465 print_hex_dump(KERN_INFO
, "",
466 DUMP_PREFIX_ADDRESS
, 16,
468 buffer_info
->skb
->data
,
469 adapter
->rx_buffer_len
,
477 * e1000_desc_unused - calculate if we have unused descriptors
479 static int e1000_desc_unused(struct e1000_ring
*ring
)
481 if (ring
->next_to_clean
> ring
->next_to_use
)
482 return ring
->next_to_clean
- ring
->next_to_use
- 1;
484 return ring
->count
+ ring
->next_to_clean
- ring
->next_to_use
- 1;
488 * e1000_receive_skb - helper function to handle Rx indications
489 * @adapter: board private structure
490 * @status: descriptor status field as written by hardware
491 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
492 * @skb: pointer to sk_buff to be indicated to stack
494 static void e1000_receive_skb(struct e1000_adapter
*adapter
,
495 struct net_device
*netdev
, struct sk_buff
*skb
,
496 u8 status
, __le16 vlan
)
498 u16 tag
= le16_to_cpu(vlan
);
499 skb
->protocol
= eth_type_trans(skb
, netdev
);
501 if (status
& E1000_RXD_STAT_VP
)
502 __vlan_hwaccel_put_tag(skb
, tag
);
504 napi_gro_receive(&adapter
->napi
, skb
);
508 * e1000_rx_checksum - Receive Checksum Offload
509 * @adapter: board private structure
510 * @status_err: receive descriptor status and error fields
511 * @csum: receive descriptor csum field
512 * @sk_buff: socket buffer with received data
514 static void e1000_rx_checksum(struct e1000_adapter
*adapter
, u32 status_err
,
517 u16 status
= (u16
)status_err
;
518 u8 errors
= (u8
)(status_err
>> 24);
520 skb_checksum_none_assert(skb
);
522 /* Rx checksum disabled */
523 if (!(adapter
->netdev
->features
& NETIF_F_RXCSUM
))
526 /* Ignore Checksum bit is set */
527 if (status
& E1000_RXD_STAT_IXSM
)
530 /* TCP/UDP checksum error bit or IP checksum error bit is set */
531 if (errors
& (E1000_RXD_ERR_TCPE
| E1000_RXD_ERR_IPE
)) {
532 /* let the stack verify checksum errors */
533 adapter
->hw_csum_err
++;
537 /* TCP/UDP Checksum has not been calculated */
538 if (!(status
& (E1000_RXD_STAT_TCPCS
| E1000_RXD_STAT_UDPCS
)))
541 /* It must be a TCP or UDP packet with a valid checksum */
542 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
543 adapter
->hw_csum_good
++;
546 static void e1000e_update_rdt_wa(struct e1000_ring
*rx_ring
, unsigned int i
)
548 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
549 struct e1000_hw
*hw
= &adapter
->hw
;
550 s32 ret_val
= __ew32_prepare(hw
);
552 writel(i
, rx_ring
->tail
);
554 if (unlikely(!ret_val
&& (i
!= readl(rx_ring
->tail
)))) {
555 u32 rctl
= er32(RCTL
);
556 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
557 e_err("ME firmware caused invalid RDT - resetting\n");
558 schedule_work(&adapter
->reset_task
);
562 static void e1000e_update_tdt_wa(struct e1000_ring
*tx_ring
, unsigned int i
)
564 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
565 struct e1000_hw
*hw
= &adapter
->hw
;
566 s32 ret_val
= __ew32_prepare(hw
);
568 writel(i
, tx_ring
->tail
);
570 if (unlikely(!ret_val
&& (i
!= readl(tx_ring
->tail
)))) {
571 u32 tctl
= er32(TCTL
);
572 ew32(TCTL
, tctl
& ~E1000_TCTL_EN
);
573 e_err("ME firmware caused invalid TDT - resetting\n");
574 schedule_work(&adapter
->reset_task
);
579 * e1000_alloc_rx_buffers - Replace used receive buffers
580 * @rx_ring: Rx descriptor ring
582 static void e1000_alloc_rx_buffers(struct e1000_ring
*rx_ring
,
583 int cleaned_count
, gfp_t gfp
)
585 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
586 struct net_device
*netdev
= adapter
->netdev
;
587 struct pci_dev
*pdev
= adapter
->pdev
;
588 union e1000_rx_desc_extended
*rx_desc
;
589 struct e1000_buffer
*buffer_info
;
592 unsigned int bufsz
= adapter
->rx_buffer_len
;
594 i
= rx_ring
->next_to_use
;
595 buffer_info
= &rx_ring
->buffer_info
[i
];
597 while (cleaned_count
--) {
598 skb
= buffer_info
->skb
;
604 skb
= __netdev_alloc_skb_ip_align(netdev
, bufsz
, gfp
);
606 /* Better luck next round */
607 adapter
->alloc_rx_buff_failed
++;
611 buffer_info
->skb
= skb
;
613 buffer_info
->dma
= dma_map_single(&pdev
->dev
, skb
->data
,
614 adapter
->rx_buffer_len
,
616 if (dma_mapping_error(&pdev
->dev
, buffer_info
->dma
)) {
617 dev_err(&pdev
->dev
, "Rx DMA map failed\n");
618 adapter
->rx_dma_failed
++;
622 rx_desc
= E1000_RX_DESC_EXT(*rx_ring
, i
);
623 rx_desc
->read
.buffer_addr
= cpu_to_le64(buffer_info
->dma
);
625 if (unlikely(!(i
& (E1000_RX_BUFFER_WRITE
- 1)))) {
627 * Force memory writes to complete before letting h/w
628 * know there are new descriptors to fetch. (Only
629 * applicable for weak-ordered memory model archs,
633 if (adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
634 e1000e_update_rdt_wa(rx_ring
, i
);
636 writel(i
, rx_ring
->tail
);
639 if (i
== rx_ring
->count
)
641 buffer_info
= &rx_ring
->buffer_info
[i
];
644 rx_ring
->next_to_use
= i
;
648 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
649 * @rx_ring: Rx descriptor ring
651 static void e1000_alloc_rx_buffers_ps(struct e1000_ring
*rx_ring
,
652 int cleaned_count
, gfp_t gfp
)
654 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
655 struct net_device
*netdev
= adapter
->netdev
;
656 struct pci_dev
*pdev
= adapter
->pdev
;
657 union e1000_rx_desc_packet_split
*rx_desc
;
658 struct e1000_buffer
*buffer_info
;
659 struct e1000_ps_page
*ps_page
;
663 i
= rx_ring
->next_to_use
;
664 buffer_info
= &rx_ring
->buffer_info
[i
];
666 while (cleaned_count
--) {
667 rx_desc
= E1000_RX_DESC_PS(*rx_ring
, i
);
669 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
670 ps_page
= &buffer_info
->ps_pages
[j
];
671 if (j
>= adapter
->rx_ps_pages
) {
672 /* all unused desc entries get hw null ptr */
673 rx_desc
->read
.buffer_addr
[j
+ 1] =
677 if (!ps_page
->page
) {
678 ps_page
->page
= alloc_page(gfp
);
679 if (!ps_page
->page
) {
680 adapter
->alloc_rx_buff_failed
++;
683 ps_page
->dma
= dma_map_page(&pdev
->dev
,
687 if (dma_mapping_error(&pdev
->dev
,
689 dev_err(&adapter
->pdev
->dev
,
690 "Rx DMA page map failed\n");
691 adapter
->rx_dma_failed
++;
696 * Refresh the desc even if buffer_addrs
697 * didn't change because each write-back
700 rx_desc
->read
.buffer_addr
[j
+ 1] =
701 cpu_to_le64(ps_page
->dma
);
704 skb
= __netdev_alloc_skb_ip_align(netdev
,
705 adapter
->rx_ps_bsize0
,
709 adapter
->alloc_rx_buff_failed
++;
713 buffer_info
->skb
= skb
;
714 buffer_info
->dma
= dma_map_single(&pdev
->dev
, skb
->data
,
715 adapter
->rx_ps_bsize0
,
717 if (dma_mapping_error(&pdev
->dev
, buffer_info
->dma
)) {
718 dev_err(&pdev
->dev
, "Rx DMA map failed\n");
719 adapter
->rx_dma_failed
++;
721 dev_kfree_skb_any(skb
);
722 buffer_info
->skb
= NULL
;
726 rx_desc
->read
.buffer_addr
[0] = cpu_to_le64(buffer_info
->dma
);
728 if (unlikely(!(i
& (E1000_RX_BUFFER_WRITE
- 1)))) {
730 * Force memory writes to complete before letting h/w
731 * know there are new descriptors to fetch. (Only
732 * applicable for weak-ordered memory model archs,
736 if (adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
737 e1000e_update_rdt_wa(rx_ring
, i
<< 1);
739 writel(i
<< 1, rx_ring
->tail
);
743 if (i
== rx_ring
->count
)
745 buffer_info
= &rx_ring
->buffer_info
[i
];
749 rx_ring
->next_to_use
= i
;
753 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
754 * @rx_ring: Rx descriptor ring
755 * @cleaned_count: number of buffers to allocate this pass
758 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring
*rx_ring
,
759 int cleaned_count
, gfp_t gfp
)
761 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
762 struct net_device
*netdev
= adapter
->netdev
;
763 struct pci_dev
*pdev
= adapter
->pdev
;
764 union e1000_rx_desc_extended
*rx_desc
;
765 struct e1000_buffer
*buffer_info
;
768 unsigned int bufsz
= 256 - 16 /* for skb_reserve */;
770 i
= rx_ring
->next_to_use
;
771 buffer_info
= &rx_ring
->buffer_info
[i
];
773 while (cleaned_count
--) {
774 skb
= buffer_info
->skb
;
780 skb
= __netdev_alloc_skb_ip_align(netdev
, bufsz
, gfp
);
781 if (unlikely(!skb
)) {
782 /* Better luck next round */
783 adapter
->alloc_rx_buff_failed
++;
787 buffer_info
->skb
= skb
;
789 /* allocate a new page if necessary */
790 if (!buffer_info
->page
) {
791 buffer_info
->page
= alloc_page(gfp
);
792 if (unlikely(!buffer_info
->page
)) {
793 adapter
->alloc_rx_buff_failed
++;
798 if (!buffer_info
->dma
)
799 buffer_info
->dma
= dma_map_page(&pdev
->dev
,
800 buffer_info
->page
, 0,
804 rx_desc
= E1000_RX_DESC_EXT(*rx_ring
, i
);
805 rx_desc
->read
.buffer_addr
= cpu_to_le64(buffer_info
->dma
);
807 if (unlikely(++i
== rx_ring
->count
))
809 buffer_info
= &rx_ring
->buffer_info
[i
];
812 if (likely(rx_ring
->next_to_use
!= i
)) {
813 rx_ring
->next_to_use
= i
;
814 if (unlikely(i
-- == 0))
815 i
= (rx_ring
->count
- 1);
817 /* Force memory writes to complete before letting h/w
818 * know there are new descriptors to fetch. (Only
819 * applicable for weak-ordered memory model archs,
822 if (adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
823 e1000e_update_rdt_wa(rx_ring
, i
);
825 writel(i
, rx_ring
->tail
);
829 static inline void e1000_rx_hash(struct net_device
*netdev
, __le32 rss
,
832 if (netdev
->features
& NETIF_F_RXHASH
)
833 skb
->rxhash
= le32_to_cpu(rss
);
837 * e1000_clean_rx_irq - Send received data up the network stack
838 * @rx_ring: Rx descriptor ring
840 * the return value indicates whether actual cleaning was done, there
841 * is no guarantee that everything was cleaned
843 static bool e1000_clean_rx_irq(struct e1000_ring
*rx_ring
, int *work_done
,
846 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
847 struct net_device
*netdev
= adapter
->netdev
;
848 struct pci_dev
*pdev
= adapter
->pdev
;
849 struct e1000_hw
*hw
= &adapter
->hw
;
850 union e1000_rx_desc_extended
*rx_desc
, *next_rxd
;
851 struct e1000_buffer
*buffer_info
, *next_buffer
;
854 int cleaned_count
= 0;
855 bool cleaned
= false;
856 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
858 i
= rx_ring
->next_to_clean
;
859 rx_desc
= E1000_RX_DESC_EXT(*rx_ring
, i
);
860 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
861 buffer_info
= &rx_ring
->buffer_info
[i
];
863 while (staterr
& E1000_RXD_STAT_DD
) {
866 if (*work_done
>= work_to_do
)
869 rmb(); /* read descriptor and rx_buffer_info after status DD */
871 skb
= buffer_info
->skb
;
872 buffer_info
->skb
= NULL
;
874 prefetch(skb
->data
- NET_IP_ALIGN
);
877 if (i
== rx_ring
->count
)
879 next_rxd
= E1000_RX_DESC_EXT(*rx_ring
, i
);
882 next_buffer
= &rx_ring
->buffer_info
[i
];
886 dma_unmap_single(&pdev
->dev
,
888 adapter
->rx_buffer_len
,
890 buffer_info
->dma
= 0;
892 length
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
895 * !EOP means multiple descriptors were used to store a single
896 * packet, if that's the case we need to toss it. In fact, we
897 * need to toss every packet with the EOP bit clear and the
898 * next frame that _does_ have the EOP bit set, as it is by
899 * definition only a frame fragment
901 if (unlikely(!(staterr
& E1000_RXD_STAT_EOP
)))
902 adapter
->flags2
|= FLAG2_IS_DISCARDING
;
904 if (adapter
->flags2
& FLAG2_IS_DISCARDING
) {
905 /* All receives must fit into a single buffer */
906 e_dbg("Receive packet consumed multiple buffers\n");
908 buffer_info
->skb
= skb
;
909 if (staterr
& E1000_RXD_STAT_EOP
)
910 adapter
->flags2
&= ~FLAG2_IS_DISCARDING
;
914 if (unlikely((staterr
& E1000_RXDEXT_ERR_FRAME_ERR_MASK
) &&
915 !(netdev
->features
& NETIF_F_RXALL
))) {
917 buffer_info
->skb
= skb
;
921 /* adjust length to remove Ethernet CRC */
922 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
)) {
923 /* If configured to store CRC, don't subtract FCS,
924 * but keep the FCS bytes out of the total_rx_bytes
927 if (netdev
->features
& NETIF_F_RXFCS
)
933 total_rx_bytes
+= length
;
937 * code added for copybreak, this should improve
938 * performance for small packets with large amounts
939 * of reassembly being done in the stack
941 if (length
< copybreak
) {
942 struct sk_buff
*new_skb
=
943 netdev_alloc_skb_ip_align(netdev
, length
);
945 skb_copy_to_linear_data_offset(new_skb
,
951 /* save the skb in buffer_info as good */
952 buffer_info
->skb
= skb
;
955 /* else just continue with the old one */
957 /* end copybreak code */
958 skb_put(skb
, length
);
960 /* Receive Checksum Offload */
961 e1000_rx_checksum(adapter
, staterr
, skb
);
963 e1000_rx_hash(netdev
, rx_desc
->wb
.lower
.hi_dword
.rss
, skb
);
965 e1000_receive_skb(adapter
, netdev
, skb
, staterr
,
966 rx_desc
->wb
.upper
.vlan
);
969 rx_desc
->wb
.upper
.status_error
&= cpu_to_le32(~0xFF);
971 /* return some buffers to hardware, one at a time is too slow */
972 if (cleaned_count
>= E1000_RX_BUFFER_WRITE
) {
973 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
,
978 /* use prefetched values */
980 buffer_info
= next_buffer
;
982 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
984 rx_ring
->next_to_clean
= i
;
986 cleaned_count
= e1000_desc_unused(rx_ring
);
988 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
, GFP_ATOMIC
);
990 adapter
->total_rx_bytes
+= total_rx_bytes
;
991 adapter
->total_rx_packets
+= total_rx_packets
;
995 static void e1000_put_txbuf(struct e1000_ring
*tx_ring
,
996 struct e1000_buffer
*buffer_info
)
998 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
1000 if (buffer_info
->dma
) {
1001 if (buffer_info
->mapped_as_page
)
1002 dma_unmap_page(&adapter
->pdev
->dev
, buffer_info
->dma
,
1003 buffer_info
->length
, DMA_TO_DEVICE
);
1005 dma_unmap_single(&adapter
->pdev
->dev
, buffer_info
->dma
,
1006 buffer_info
->length
, DMA_TO_DEVICE
);
1007 buffer_info
->dma
= 0;
1009 if (buffer_info
->skb
) {
1010 dev_kfree_skb_any(buffer_info
->skb
);
1011 buffer_info
->skb
= NULL
;
1013 buffer_info
->time_stamp
= 0;
1016 static void e1000_print_hw_hang(struct work_struct
*work
)
1018 struct e1000_adapter
*adapter
= container_of(work
,
1019 struct e1000_adapter
,
1021 struct net_device
*netdev
= adapter
->netdev
;
1022 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1023 unsigned int i
= tx_ring
->next_to_clean
;
1024 unsigned int eop
= tx_ring
->buffer_info
[i
].next_to_watch
;
1025 struct e1000_tx_desc
*eop_desc
= E1000_TX_DESC(*tx_ring
, eop
);
1026 struct e1000_hw
*hw
= &adapter
->hw
;
1027 u16 phy_status
, phy_1000t_status
, phy_ext_status
;
1030 if (test_bit(__E1000_DOWN
, &adapter
->state
))
1033 if (!adapter
->tx_hang_recheck
&&
1034 (adapter
->flags2
& FLAG2_DMA_BURST
)) {
1036 * May be block on write-back, flush and detect again
1037 * flush pending descriptor writebacks to memory
1039 ew32(TIDV
, adapter
->tx_int_delay
| E1000_TIDV_FPD
);
1040 /* execute the writes immediately */
1043 * Due to rare timing issues, write to TIDV again to ensure
1044 * the write is successful
1046 ew32(TIDV
, adapter
->tx_int_delay
| E1000_TIDV_FPD
);
1047 /* execute the writes immediately */
1049 adapter
->tx_hang_recheck
= true;
1052 /* Real hang detected */
1053 adapter
->tx_hang_recheck
= false;
1054 netif_stop_queue(netdev
);
1056 e1e_rphy(hw
, PHY_STATUS
, &phy_status
);
1057 e1e_rphy(hw
, PHY_1000T_STATUS
, &phy_1000t_status
);
1058 e1e_rphy(hw
, PHY_EXT_STATUS
, &phy_ext_status
);
1060 pci_read_config_word(adapter
->pdev
, PCI_STATUS
, &pci_status
);
1062 /* detected Hardware unit hang */
1063 e_err("Detected Hardware Unit Hang:\n"
1066 " next_to_use <%x>\n"
1067 " next_to_clean <%x>\n"
1068 "buffer_info[next_to_clean]:\n"
1069 " time_stamp <%lx>\n"
1070 " next_to_watch <%x>\n"
1072 " next_to_watch.status <%x>\n"
1075 "PHY 1000BASE-T Status <%x>\n"
1076 "PHY Extended Status <%x>\n"
1077 "PCI Status <%x>\n",
1078 readl(tx_ring
->head
),
1079 readl(tx_ring
->tail
),
1080 tx_ring
->next_to_use
,
1081 tx_ring
->next_to_clean
,
1082 tx_ring
->buffer_info
[eop
].time_stamp
,
1085 eop_desc
->upper
.fields
.status
,
1092 /* Suggest workaround for known h/w issue */
1093 if ((hw
->mac
.type
== e1000_pchlan
) && (er32(CTRL
) & E1000_CTRL_TFCE
))
1094 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1098 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1099 * @tx_ring: Tx descriptor ring
1101 * the return value indicates whether actual cleaning was done, there
1102 * is no guarantee that everything was cleaned
1104 static bool e1000_clean_tx_irq(struct e1000_ring
*tx_ring
)
1106 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
1107 struct net_device
*netdev
= adapter
->netdev
;
1108 struct e1000_hw
*hw
= &adapter
->hw
;
1109 struct e1000_tx_desc
*tx_desc
, *eop_desc
;
1110 struct e1000_buffer
*buffer_info
;
1111 unsigned int i
, eop
;
1112 unsigned int count
= 0;
1113 unsigned int total_tx_bytes
= 0, total_tx_packets
= 0;
1114 unsigned int bytes_compl
= 0, pkts_compl
= 0;
1116 i
= tx_ring
->next_to_clean
;
1117 eop
= tx_ring
->buffer_info
[i
].next_to_watch
;
1118 eop_desc
= E1000_TX_DESC(*tx_ring
, eop
);
1120 while ((eop_desc
->upper
.data
& cpu_to_le32(E1000_TXD_STAT_DD
)) &&
1121 (count
< tx_ring
->count
)) {
1122 bool cleaned
= false;
1123 rmb(); /* read buffer_info after eop_desc */
1124 for (; !cleaned
; count
++) {
1125 tx_desc
= E1000_TX_DESC(*tx_ring
, i
);
1126 buffer_info
= &tx_ring
->buffer_info
[i
];
1127 cleaned
= (i
== eop
);
1130 total_tx_packets
+= buffer_info
->segs
;
1131 total_tx_bytes
+= buffer_info
->bytecount
;
1132 if (buffer_info
->skb
) {
1133 bytes_compl
+= buffer_info
->skb
->len
;
1138 e1000_put_txbuf(tx_ring
, buffer_info
);
1139 tx_desc
->upper
.data
= 0;
1142 if (i
== tx_ring
->count
)
1146 if (i
== tx_ring
->next_to_use
)
1148 eop
= tx_ring
->buffer_info
[i
].next_to_watch
;
1149 eop_desc
= E1000_TX_DESC(*tx_ring
, eop
);
1152 tx_ring
->next_to_clean
= i
;
1154 netdev_completed_queue(netdev
, pkts_compl
, bytes_compl
);
1156 #define TX_WAKE_THRESHOLD 32
1157 if (count
&& netif_carrier_ok(netdev
) &&
1158 e1000_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
) {
1159 /* Make sure that anybody stopping the queue after this
1160 * sees the new next_to_clean.
1164 if (netif_queue_stopped(netdev
) &&
1165 !(test_bit(__E1000_DOWN
, &adapter
->state
))) {
1166 netif_wake_queue(netdev
);
1167 ++adapter
->restart_queue
;
1171 if (adapter
->detect_tx_hung
) {
1173 * Detect a transmit hang in hardware, this serializes the
1174 * check with the clearing of time_stamp and movement of i
1176 adapter
->detect_tx_hung
= false;
1177 if (tx_ring
->buffer_info
[i
].time_stamp
&&
1178 time_after(jiffies
, tx_ring
->buffer_info
[i
].time_stamp
1179 + (adapter
->tx_timeout_factor
* HZ
)) &&
1180 !(er32(STATUS
) & E1000_STATUS_TXOFF
))
1181 schedule_work(&adapter
->print_hang_task
);
1183 adapter
->tx_hang_recheck
= false;
1185 adapter
->total_tx_bytes
+= total_tx_bytes
;
1186 adapter
->total_tx_packets
+= total_tx_packets
;
1187 return count
< tx_ring
->count
;
1191 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1192 * @rx_ring: Rx descriptor ring
1194 * the return value indicates whether actual cleaning was done, there
1195 * is no guarantee that everything was cleaned
1197 static bool e1000_clean_rx_irq_ps(struct e1000_ring
*rx_ring
, int *work_done
,
1200 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
1201 struct e1000_hw
*hw
= &adapter
->hw
;
1202 union e1000_rx_desc_packet_split
*rx_desc
, *next_rxd
;
1203 struct net_device
*netdev
= adapter
->netdev
;
1204 struct pci_dev
*pdev
= adapter
->pdev
;
1205 struct e1000_buffer
*buffer_info
, *next_buffer
;
1206 struct e1000_ps_page
*ps_page
;
1207 struct sk_buff
*skb
;
1209 u32 length
, staterr
;
1210 int cleaned_count
= 0;
1211 bool cleaned
= false;
1212 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
1214 i
= rx_ring
->next_to_clean
;
1215 rx_desc
= E1000_RX_DESC_PS(*rx_ring
, i
);
1216 staterr
= le32_to_cpu(rx_desc
->wb
.middle
.status_error
);
1217 buffer_info
= &rx_ring
->buffer_info
[i
];
1219 while (staterr
& E1000_RXD_STAT_DD
) {
1220 if (*work_done
>= work_to_do
)
1223 skb
= buffer_info
->skb
;
1224 rmb(); /* read descriptor and rx_buffer_info after status DD */
1226 /* in the packet split case this is header only */
1227 prefetch(skb
->data
- NET_IP_ALIGN
);
1230 if (i
== rx_ring
->count
)
1232 next_rxd
= E1000_RX_DESC_PS(*rx_ring
, i
);
1235 next_buffer
= &rx_ring
->buffer_info
[i
];
1239 dma_unmap_single(&pdev
->dev
, buffer_info
->dma
,
1240 adapter
->rx_ps_bsize0
, DMA_FROM_DEVICE
);
1241 buffer_info
->dma
= 0;
1243 /* see !EOP comment in other Rx routine */
1244 if (!(staterr
& E1000_RXD_STAT_EOP
))
1245 adapter
->flags2
|= FLAG2_IS_DISCARDING
;
1247 if (adapter
->flags2
& FLAG2_IS_DISCARDING
) {
1248 e_dbg("Packet Split buffers didn't pick up the full packet\n");
1249 dev_kfree_skb_irq(skb
);
1250 if (staterr
& E1000_RXD_STAT_EOP
)
1251 adapter
->flags2
&= ~FLAG2_IS_DISCARDING
;
1255 if (unlikely((staterr
& E1000_RXDEXT_ERR_FRAME_ERR_MASK
) &&
1256 !(netdev
->features
& NETIF_F_RXALL
))) {
1257 dev_kfree_skb_irq(skb
);
1261 length
= le16_to_cpu(rx_desc
->wb
.middle
.length0
);
1264 e_dbg("Last part of the packet spanning multiple descriptors\n");
1265 dev_kfree_skb_irq(skb
);
1270 skb_put(skb
, length
);
1274 * this looks ugly, but it seems compiler issues make
1275 * it more efficient than reusing j
1277 int l1
= le16_to_cpu(rx_desc
->wb
.upper
.length
[0]);
1280 * page alloc/put takes too long and effects small
1281 * packet throughput, so unsplit small packets and
1282 * save the alloc/put only valid in softirq (napi)
1283 * context to call kmap_*
1285 if (l1
&& (l1
<= copybreak
) &&
1286 ((length
+ l1
) <= adapter
->rx_ps_bsize0
)) {
1289 ps_page
= &buffer_info
->ps_pages
[0];
1292 * there is no documentation about how to call
1293 * kmap_atomic, so we can't hold the mapping
1296 dma_sync_single_for_cpu(&pdev
->dev
,
1300 vaddr
= kmap_atomic(ps_page
->page
);
1301 memcpy(skb_tail_pointer(skb
), vaddr
, l1
);
1302 kunmap_atomic(vaddr
);
1303 dma_sync_single_for_device(&pdev
->dev
,
1308 /* remove the CRC */
1309 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
)) {
1310 if (!(netdev
->features
& NETIF_F_RXFCS
))
1319 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
1320 length
= le16_to_cpu(rx_desc
->wb
.upper
.length
[j
]);
1324 ps_page
= &buffer_info
->ps_pages
[j
];
1325 dma_unmap_page(&pdev
->dev
, ps_page
->dma
, PAGE_SIZE
,
1328 skb_fill_page_desc(skb
, j
, ps_page
->page
, 0, length
);
1329 ps_page
->page
= NULL
;
1331 skb
->data_len
+= length
;
1332 skb
->truesize
+= PAGE_SIZE
;
1335 /* strip the ethernet crc, problem is we're using pages now so
1336 * this whole operation can get a little cpu intensive
1338 if (!(adapter
->flags2
& FLAG2_CRC_STRIPPING
)) {
1339 if (!(netdev
->features
& NETIF_F_RXFCS
))
1340 pskb_trim(skb
, skb
->len
- 4);
1344 total_rx_bytes
+= skb
->len
;
1347 e1000_rx_checksum(adapter
, staterr
, skb
);
1349 e1000_rx_hash(netdev
, rx_desc
->wb
.lower
.hi_dword
.rss
, skb
);
1351 if (rx_desc
->wb
.upper
.header_status
&
1352 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP
))
1353 adapter
->rx_hdr_split
++;
1355 e1000_receive_skb(adapter
, netdev
, skb
,
1356 staterr
, rx_desc
->wb
.middle
.vlan
);
1359 rx_desc
->wb
.middle
.status_error
&= cpu_to_le32(~0xFF);
1360 buffer_info
->skb
= NULL
;
1362 /* return some buffers to hardware, one at a time is too slow */
1363 if (cleaned_count
>= E1000_RX_BUFFER_WRITE
) {
1364 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
,
1369 /* use prefetched values */
1371 buffer_info
= next_buffer
;
1373 staterr
= le32_to_cpu(rx_desc
->wb
.middle
.status_error
);
1375 rx_ring
->next_to_clean
= i
;
1377 cleaned_count
= e1000_desc_unused(rx_ring
);
1379 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
, GFP_ATOMIC
);
1381 adapter
->total_rx_bytes
+= total_rx_bytes
;
1382 adapter
->total_rx_packets
+= total_rx_packets
;
1387 * e1000_consume_page - helper function
1389 static void e1000_consume_page(struct e1000_buffer
*bi
, struct sk_buff
*skb
,
1394 skb
->data_len
+= length
;
1395 skb
->truesize
+= PAGE_SIZE
;
1399 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1400 * @adapter: board private structure
1402 * the return value indicates whether actual cleaning was done, there
1403 * is no guarantee that everything was cleaned
1405 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring
*rx_ring
, int *work_done
,
1408 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
1409 struct net_device
*netdev
= adapter
->netdev
;
1410 struct pci_dev
*pdev
= adapter
->pdev
;
1411 union e1000_rx_desc_extended
*rx_desc
, *next_rxd
;
1412 struct e1000_buffer
*buffer_info
, *next_buffer
;
1413 u32 length
, staterr
;
1415 int cleaned_count
= 0;
1416 bool cleaned
= false;
1417 unsigned int total_rx_bytes
=0, total_rx_packets
=0;
1419 i
= rx_ring
->next_to_clean
;
1420 rx_desc
= E1000_RX_DESC_EXT(*rx_ring
, i
);
1421 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
1422 buffer_info
= &rx_ring
->buffer_info
[i
];
1424 while (staterr
& E1000_RXD_STAT_DD
) {
1425 struct sk_buff
*skb
;
1427 if (*work_done
>= work_to_do
)
1430 rmb(); /* read descriptor and rx_buffer_info after status DD */
1432 skb
= buffer_info
->skb
;
1433 buffer_info
->skb
= NULL
;
1436 if (i
== rx_ring
->count
)
1438 next_rxd
= E1000_RX_DESC_EXT(*rx_ring
, i
);
1441 next_buffer
= &rx_ring
->buffer_info
[i
];
1445 dma_unmap_page(&pdev
->dev
, buffer_info
->dma
, PAGE_SIZE
,
1447 buffer_info
->dma
= 0;
1449 length
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
1451 /* errors is only valid for DD + EOP descriptors */
1452 if (unlikely((staterr
& E1000_RXD_STAT_EOP
) &&
1453 ((staterr
& E1000_RXDEXT_ERR_FRAME_ERR_MASK
) &&
1454 !(netdev
->features
& NETIF_F_RXALL
)))) {
1455 /* recycle both page and skb */
1456 buffer_info
->skb
= skb
;
1457 /* an error means any chain goes out the window too */
1458 if (rx_ring
->rx_skb_top
)
1459 dev_kfree_skb_irq(rx_ring
->rx_skb_top
);
1460 rx_ring
->rx_skb_top
= NULL
;
1464 #define rxtop (rx_ring->rx_skb_top)
1465 if (!(staterr
& E1000_RXD_STAT_EOP
)) {
1466 /* this descriptor is only the beginning (or middle) */
1468 /* this is the beginning of a chain */
1470 skb_fill_page_desc(rxtop
, 0, buffer_info
->page
,
1473 /* this is the middle of a chain */
1474 skb_fill_page_desc(rxtop
,
1475 skb_shinfo(rxtop
)->nr_frags
,
1476 buffer_info
->page
, 0, length
);
1477 /* re-use the skb, only consumed the page */
1478 buffer_info
->skb
= skb
;
1480 e1000_consume_page(buffer_info
, rxtop
, length
);
1484 /* end of the chain */
1485 skb_fill_page_desc(rxtop
,
1486 skb_shinfo(rxtop
)->nr_frags
,
1487 buffer_info
->page
, 0, length
);
1488 /* re-use the current skb, we only consumed the
1490 buffer_info
->skb
= skb
;
1493 e1000_consume_page(buffer_info
, skb
, length
);
1495 /* no chain, got EOP, this buf is the packet
1496 * copybreak to save the put_page/alloc_page */
1497 if (length
<= copybreak
&&
1498 skb_tailroom(skb
) >= length
) {
1500 vaddr
= kmap_atomic(buffer_info
->page
);
1501 memcpy(skb_tail_pointer(skb
), vaddr
,
1503 kunmap_atomic(vaddr
);
1504 /* re-use the page, so don't erase
1505 * buffer_info->page */
1506 skb_put(skb
, length
);
1508 skb_fill_page_desc(skb
, 0,
1509 buffer_info
->page
, 0,
1511 e1000_consume_page(buffer_info
, skb
,
1517 /* Receive Checksum Offload */
1518 e1000_rx_checksum(adapter
, staterr
, skb
);
1520 e1000_rx_hash(netdev
, rx_desc
->wb
.lower
.hi_dword
.rss
, skb
);
1522 /* probably a little skewed due to removing CRC */
1523 total_rx_bytes
+= skb
->len
;
1526 /* eth type trans needs skb->data to point to something */
1527 if (!pskb_may_pull(skb
, ETH_HLEN
)) {
1528 e_err("pskb_may_pull failed.\n");
1529 dev_kfree_skb_irq(skb
);
1533 e1000_receive_skb(adapter
, netdev
, skb
, staterr
,
1534 rx_desc
->wb
.upper
.vlan
);
1537 rx_desc
->wb
.upper
.status_error
&= cpu_to_le32(~0xFF);
1539 /* return some buffers to hardware, one at a time is too slow */
1540 if (unlikely(cleaned_count
>= E1000_RX_BUFFER_WRITE
)) {
1541 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
,
1546 /* use prefetched values */
1548 buffer_info
= next_buffer
;
1550 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
1552 rx_ring
->next_to_clean
= i
;
1554 cleaned_count
= e1000_desc_unused(rx_ring
);
1556 adapter
->alloc_rx_buf(rx_ring
, cleaned_count
, GFP_ATOMIC
);
1558 adapter
->total_rx_bytes
+= total_rx_bytes
;
1559 adapter
->total_rx_packets
+= total_rx_packets
;
1564 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1565 * @rx_ring: Rx descriptor ring
1567 static void e1000_clean_rx_ring(struct e1000_ring
*rx_ring
)
1569 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
1570 struct e1000_buffer
*buffer_info
;
1571 struct e1000_ps_page
*ps_page
;
1572 struct pci_dev
*pdev
= adapter
->pdev
;
1575 /* Free all the Rx ring sk_buffs */
1576 for (i
= 0; i
< rx_ring
->count
; i
++) {
1577 buffer_info
= &rx_ring
->buffer_info
[i
];
1578 if (buffer_info
->dma
) {
1579 if (adapter
->clean_rx
== e1000_clean_rx_irq
)
1580 dma_unmap_single(&pdev
->dev
, buffer_info
->dma
,
1581 adapter
->rx_buffer_len
,
1583 else if (adapter
->clean_rx
== e1000_clean_jumbo_rx_irq
)
1584 dma_unmap_page(&pdev
->dev
, buffer_info
->dma
,
1587 else if (adapter
->clean_rx
== e1000_clean_rx_irq_ps
)
1588 dma_unmap_single(&pdev
->dev
, buffer_info
->dma
,
1589 adapter
->rx_ps_bsize0
,
1591 buffer_info
->dma
= 0;
1594 if (buffer_info
->page
) {
1595 put_page(buffer_info
->page
);
1596 buffer_info
->page
= NULL
;
1599 if (buffer_info
->skb
) {
1600 dev_kfree_skb(buffer_info
->skb
);
1601 buffer_info
->skb
= NULL
;
1604 for (j
= 0; j
< PS_PAGE_BUFFERS
; j
++) {
1605 ps_page
= &buffer_info
->ps_pages
[j
];
1608 dma_unmap_page(&pdev
->dev
, ps_page
->dma
, PAGE_SIZE
,
1611 put_page(ps_page
->page
);
1612 ps_page
->page
= NULL
;
1616 /* there also may be some cached data from a chained receive */
1617 if (rx_ring
->rx_skb_top
) {
1618 dev_kfree_skb(rx_ring
->rx_skb_top
);
1619 rx_ring
->rx_skb_top
= NULL
;
1622 /* Zero out the descriptor ring */
1623 memset(rx_ring
->desc
, 0, rx_ring
->size
);
1625 rx_ring
->next_to_clean
= 0;
1626 rx_ring
->next_to_use
= 0;
1627 adapter
->flags2
&= ~FLAG2_IS_DISCARDING
;
1629 writel(0, rx_ring
->head
);
1630 if (rx_ring
->adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
1631 e1000e_update_rdt_wa(rx_ring
, 0);
1633 writel(0, rx_ring
->tail
);
1636 static void e1000e_downshift_workaround(struct work_struct
*work
)
1638 struct e1000_adapter
*adapter
= container_of(work
,
1639 struct e1000_adapter
, downshift_task
);
1641 if (test_bit(__E1000_DOWN
, &adapter
->state
))
1644 e1000e_gig_downshift_workaround_ich8lan(&adapter
->hw
);
1648 * e1000_intr_msi - Interrupt Handler
1649 * @irq: interrupt number
1650 * @data: pointer to a network interface device structure
1652 static irqreturn_t
e1000_intr_msi(int irq
, void *data
)
1654 struct net_device
*netdev
= data
;
1655 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1656 struct e1000_hw
*hw
= &adapter
->hw
;
1657 u32 icr
= er32(ICR
);
1660 * read ICR disables interrupts using IAM
1663 if (icr
& E1000_ICR_LSC
) {
1664 hw
->mac
.get_link_status
= true;
1666 * ICH8 workaround-- Call gig speed drop workaround on cable
1667 * disconnect (LSC) before accessing any PHY registers
1669 if ((adapter
->flags
& FLAG_LSC_GIG_SPEED_DROP
) &&
1670 (!(er32(STATUS
) & E1000_STATUS_LU
)))
1671 schedule_work(&adapter
->downshift_task
);
1674 * 80003ES2LAN workaround-- For packet buffer work-around on
1675 * link down event; disable receives here in the ISR and reset
1676 * adapter in watchdog
1678 if (netif_carrier_ok(netdev
) &&
1679 adapter
->flags
& FLAG_RX_NEEDS_RESTART
) {
1680 /* disable receives */
1681 u32 rctl
= er32(RCTL
);
1682 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
1683 adapter
->flags
|= FLAG_RX_RESTART_NOW
;
1685 /* guard against interrupt when we're going down */
1686 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1687 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1690 if (napi_schedule_prep(&adapter
->napi
)) {
1691 adapter
->total_tx_bytes
= 0;
1692 adapter
->total_tx_packets
= 0;
1693 adapter
->total_rx_bytes
= 0;
1694 adapter
->total_rx_packets
= 0;
1695 __napi_schedule(&adapter
->napi
);
1702 * e1000_intr - Interrupt Handler
1703 * @irq: interrupt number
1704 * @data: pointer to a network interface device structure
1706 static irqreturn_t
e1000_intr(int irq
, void *data
)
1708 struct net_device
*netdev
= data
;
1709 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1710 struct e1000_hw
*hw
= &adapter
->hw
;
1711 u32 rctl
, icr
= er32(ICR
);
1713 if (!icr
|| test_bit(__E1000_DOWN
, &adapter
->state
))
1714 return IRQ_NONE
; /* Not our interrupt */
1717 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1718 * not set, then the adapter didn't send an interrupt
1720 if (!(icr
& E1000_ICR_INT_ASSERTED
))
1724 * Interrupt Auto-Mask...upon reading ICR,
1725 * interrupts are masked. No need for the
1729 if (icr
& E1000_ICR_LSC
) {
1730 hw
->mac
.get_link_status
= true;
1732 * ICH8 workaround-- Call gig speed drop workaround on cable
1733 * disconnect (LSC) before accessing any PHY registers
1735 if ((adapter
->flags
& FLAG_LSC_GIG_SPEED_DROP
) &&
1736 (!(er32(STATUS
) & E1000_STATUS_LU
)))
1737 schedule_work(&adapter
->downshift_task
);
1740 * 80003ES2LAN workaround--
1741 * For packet buffer work-around on link down event;
1742 * disable receives here in the ISR and
1743 * reset adapter in watchdog
1745 if (netif_carrier_ok(netdev
) &&
1746 (adapter
->flags
& FLAG_RX_NEEDS_RESTART
)) {
1747 /* disable receives */
1749 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
1750 adapter
->flags
|= FLAG_RX_RESTART_NOW
;
1752 /* guard against interrupt when we're going down */
1753 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1754 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1757 if (napi_schedule_prep(&adapter
->napi
)) {
1758 adapter
->total_tx_bytes
= 0;
1759 adapter
->total_tx_packets
= 0;
1760 adapter
->total_rx_bytes
= 0;
1761 adapter
->total_rx_packets
= 0;
1762 __napi_schedule(&adapter
->napi
);
1768 static irqreturn_t
e1000_msix_other(int irq
, void *data
)
1770 struct net_device
*netdev
= data
;
1771 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1772 struct e1000_hw
*hw
= &adapter
->hw
;
1773 u32 icr
= er32(ICR
);
1775 if (!(icr
& E1000_ICR_INT_ASSERTED
)) {
1776 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1777 ew32(IMS
, E1000_IMS_OTHER
);
1781 if (icr
& adapter
->eiac_mask
)
1782 ew32(ICS
, (icr
& adapter
->eiac_mask
));
1784 if (icr
& E1000_ICR_OTHER
) {
1785 if (!(icr
& E1000_ICR_LSC
))
1786 goto no_link_interrupt
;
1787 hw
->mac
.get_link_status
= true;
1788 /* guard against interrupt when we're going down */
1789 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1790 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
1794 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
1795 ew32(IMS
, E1000_IMS_LSC
| E1000_IMS_OTHER
);
1801 static irqreturn_t
e1000_intr_msix_tx(int irq
, void *data
)
1803 struct net_device
*netdev
= data
;
1804 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1805 struct e1000_hw
*hw
= &adapter
->hw
;
1806 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1809 adapter
->total_tx_bytes
= 0;
1810 adapter
->total_tx_packets
= 0;
1812 if (!e1000_clean_tx_irq(tx_ring
))
1813 /* Ring was not completely cleaned, so fire another interrupt */
1814 ew32(ICS
, tx_ring
->ims_val
);
1819 static irqreturn_t
e1000_intr_msix_rx(int irq
, void *data
)
1821 struct net_device
*netdev
= data
;
1822 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
1823 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1825 /* Write the ITR value calculated at the end of the
1826 * previous interrupt.
1828 if (rx_ring
->set_itr
) {
1829 writel(1000000000 / (rx_ring
->itr_val
* 256),
1830 rx_ring
->itr_register
);
1831 rx_ring
->set_itr
= 0;
1834 if (napi_schedule_prep(&adapter
->napi
)) {
1835 adapter
->total_rx_bytes
= 0;
1836 adapter
->total_rx_packets
= 0;
1837 __napi_schedule(&adapter
->napi
);
1843 * e1000_configure_msix - Configure MSI-X hardware
1845 * e1000_configure_msix sets up the hardware to properly
1846 * generate MSI-X interrupts.
1848 static void e1000_configure_msix(struct e1000_adapter
*adapter
)
1850 struct e1000_hw
*hw
= &adapter
->hw
;
1851 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
1852 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
1854 u32 ctrl_ext
, ivar
= 0;
1856 adapter
->eiac_mask
= 0;
1858 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1859 if (hw
->mac
.type
== e1000_82574
) {
1860 u32 rfctl
= er32(RFCTL
);
1861 rfctl
|= E1000_RFCTL_ACK_DIS
;
1865 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1866 /* Configure Rx vector */
1867 rx_ring
->ims_val
= E1000_IMS_RXQ0
;
1868 adapter
->eiac_mask
|= rx_ring
->ims_val
;
1869 if (rx_ring
->itr_val
)
1870 writel(1000000000 / (rx_ring
->itr_val
* 256),
1871 rx_ring
->itr_register
);
1873 writel(1, rx_ring
->itr_register
);
1874 ivar
= E1000_IVAR_INT_ALLOC_VALID
| vector
;
1876 /* Configure Tx vector */
1877 tx_ring
->ims_val
= E1000_IMS_TXQ0
;
1879 if (tx_ring
->itr_val
)
1880 writel(1000000000 / (tx_ring
->itr_val
* 256),
1881 tx_ring
->itr_register
);
1883 writel(1, tx_ring
->itr_register
);
1884 adapter
->eiac_mask
|= tx_ring
->ims_val
;
1885 ivar
|= ((E1000_IVAR_INT_ALLOC_VALID
| vector
) << 8);
1887 /* set vector for Other Causes, e.g. link changes */
1889 ivar
|= ((E1000_IVAR_INT_ALLOC_VALID
| vector
) << 16);
1890 if (rx_ring
->itr_val
)
1891 writel(1000000000 / (rx_ring
->itr_val
* 256),
1892 hw
->hw_addr
+ E1000_EITR_82574(vector
));
1894 writel(1, hw
->hw_addr
+ E1000_EITR_82574(vector
));
1896 /* Cause Tx interrupts on every write back */
1901 /* enable MSI-X PBA support */
1902 ctrl_ext
= er32(CTRL_EXT
);
1903 ctrl_ext
|= E1000_CTRL_EXT_PBA_CLR
;
1905 /* Auto-Mask Other interrupts upon ICR read */
1906 #define E1000_EIAC_MASK_82574 0x01F00000
1907 ew32(IAM
, ~E1000_EIAC_MASK_82574
| E1000_IMS_OTHER
);
1908 ctrl_ext
|= E1000_CTRL_EXT_EIAME
;
1909 ew32(CTRL_EXT
, ctrl_ext
);
1913 void e1000e_reset_interrupt_capability(struct e1000_adapter
*adapter
)
1915 if (adapter
->msix_entries
) {
1916 pci_disable_msix(adapter
->pdev
);
1917 kfree(adapter
->msix_entries
);
1918 adapter
->msix_entries
= NULL
;
1919 } else if (adapter
->flags
& FLAG_MSI_ENABLED
) {
1920 pci_disable_msi(adapter
->pdev
);
1921 adapter
->flags
&= ~FLAG_MSI_ENABLED
;
1926 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1928 * Attempt to configure interrupts using the best available
1929 * capabilities of the hardware and kernel.
1931 void e1000e_set_interrupt_capability(struct e1000_adapter
*adapter
)
1936 switch (adapter
->int_mode
) {
1937 case E1000E_INT_MODE_MSIX
:
1938 if (adapter
->flags
& FLAG_HAS_MSIX
) {
1939 adapter
->num_vectors
= 3; /* RxQ0, TxQ0 and other */
1940 adapter
->msix_entries
= kcalloc(adapter
->num_vectors
,
1941 sizeof(struct msix_entry
),
1943 if (adapter
->msix_entries
) {
1944 for (i
= 0; i
< adapter
->num_vectors
; i
++)
1945 adapter
->msix_entries
[i
].entry
= i
;
1947 err
= pci_enable_msix(adapter
->pdev
,
1948 adapter
->msix_entries
,
1949 adapter
->num_vectors
);
1953 /* MSI-X failed, so fall through and try MSI */
1954 e_err("Failed to initialize MSI-X interrupts. Falling back to MSI interrupts.\n");
1955 e1000e_reset_interrupt_capability(adapter
);
1957 adapter
->int_mode
= E1000E_INT_MODE_MSI
;
1959 case E1000E_INT_MODE_MSI
:
1960 if (!pci_enable_msi(adapter
->pdev
)) {
1961 adapter
->flags
|= FLAG_MSI_ENABLED
;
1963 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
1964 e_err("Failed to initialize MSI interrupts. Falling back to legacy interrupts.\n");
1967 case E1000E_INT_MODE_LEGACY
:
1968 /* Don't do anything; this is the system default */
1972 /* store the number of vectors being used */
1973 adapter
->num_vectors
= 1;
1977 * e1000_request_msix - Initialize MSI-X interrupts
1979 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1982 static int e1000_request_msix(struct e1000_adapter
*adapter
)
1984 struct net_device
*netdev
= adapter
->netdev
;
1985 int err
= 0, vector
= 0;
1987 if (strlen(netdev
->name
) < (IFNAMSIZ
- 5))
1988 snprintf(adapter
->rx_ring
->name
,
1989 sizeof(adapter
->rx_ring
->name
) - 1,
1990 "%s-rx-0", netdev
->name
);
1992 memcpy(adapter
->rx_ring
->name
, netdev
->name
, IFNAMSIZ
);
1993 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1994 e1000_intr_msix_rx
, 0, adapter
->rx_ring
->name
,
1998 adapter
->rx_ring
->itr_register
= adapter
->hw
.hw_addr
+
1999 E1000_EITR_82574(vector
);
2000 adapter
->rx_ring
->itr_val
= adapter
->itr
;
2003 if (strlen(netdev
->name
) < (IFNAMSIZ
- 5))
2004 snprintf(adapter
->tx_ring
->name
,
2005 sizeof(adapter
->tx_ring
->name
) - 1,
2006 "%s-tx-0", netdev
->name
);
2008 memcpy(adapter
->tx_ring
->name
, netdev
->name
, IFNAMSIZ
);
2009 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
2010 e1000_intr_msix_tx
, 0, adapter
->tx_ring
->name
,
2014 adapter
->tx_ring
->itr_register
= adapter
->hw
.hw_addr
+
2015 E1000_EITR_82574(vector
);
2016 adapter
->tx_ring
->itr_val
= adapter
->itr
;
2019 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
2020 e1000_msix_other
, 0, netdev
->name
, netdev
);
2024 e1000_configure_msix(adapter
);
2030 * e1000_request_irq - initialize interrupts
2032 * Attempts to configure interrupts using the best available
2033 * capabilities of the hardware and kernel.
2035 static int e1000_request_irq(struct e1000_adapter
*adapter
)
2037 struct net_device
*netdev
= adapter
->netdev
;
2040 if (adapter
->msix_entries
) {
2041 err
= e1000_request_msix(adapter
);
2044 /* fall back to MSI */
2045 e1000e_reset_interrupt_capability(adapter
);
2046 adapter
->int_mode
= E1000E_INT_MODE_MSI
;
2047 e1000e_set_interrupt_capability(adapter
);
2049 if (adapter
->flags
& FLAG_MSI_ENABLED
) {
2050 err
= request_irq(adapter
->pdev
->irq
, e1000_intr_msi
, 0,
2051 netdev
->name
, netdev
);
2055 /* fall back to legacy interrupt */
2056 e1000e_reset_interrupt_capability(adapter
);
2057 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
2060 err
= request_irq(adapter
->pdev
->irq
, e1000_intr
, IRQF_SHARED
,
2061 netdev
->name
, netdev
);
2063 e_err("Unable to allocate interrupt, Error: %d\n", err
);
2068 static void e1000_free_irq(struct e1000_adapter
*adapter
)
2070 struct net_device
*netdev
= adapter
->netdev
;
2072 if (adapter
->msix_entries
) {
2075 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
2078 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
2081 /* Other Causes interrupt vector */
2082 free_irq(adapter
->msix_entries
[vector
].vector
, netdev
);
2086 free_irq(adapter
->pdev
->irq
, netdev
);
2090 * e1000_irq_disable - Mask off interrupt generation on the NIC
2092 static void e1000_irq_disable(struct e1000_adapter
*adapter
)
2094 struct e1000_hw
*hw
= &adapter
->hw
;
2097 if (adapter
->msix_entries
)
2098 ew32(EIAC_82574
, 0);
2101 if (adapter
->msix_entries
) {
2103 for (i
= 0; i
< adapter
->num_vectors
; i
++)
2104 synchronize_irq(adapter
->msix_entries
[i
].vector
);
2106 synchronize_irq(adapter
->pdev
->irq
);
2111 * e1000_irq_enable - Enable default interrupt generation settings
2113 static void e1000_irq_enable(struct e1000_adapter
*adapter
)
2115 struct e1000_hw
*hw
= &adapter
->hw
;
2117 if (adapter
->msix_entries
) {
2118 ew32(EIAC_82574
, adapter
->eiac_mask
& E1000_EIAC_MASK_82574
);
2119 ew32(IMS
, adapter
->eiac_mask
| E1000_IMS_OTHER
| E1000_IMS_LSC
);
2121 ew32(IMS
, IMS_ENABLE_MASK
);
2127 * e1000e_get_hw_control - get control of the h/w from f/w
2128 * @adapter: address of board private structure
2130 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2131 * For ASF and Pass Through versions of f/w this means that
2132 * the driver is loaded. For AMT version (only with 82573)
2133 * of the f/w this means that the network i/f is open.
2135 void e1000e_get_hw_control(struct e1000_adapter
*adapter
)
2137 struct e1000_hw
*hw
= &adapter
->hw
;
2141 /* Let firmware know the driver has taken over */
2142 if (adapter
->flags
& FLAG_HAS_SWSM_ON_LOAD
) {
2144 ew32(SWSM
, swsm
| E1000_SWSM_DRV_LOAD
);
2145 } else if (adapter
->flags
& FLAG_HAS_CTRLEXT_ON_LOAD
) {
2146 ctrl_ext
= er32(CTRL_EXT
);
2147 ew32(CTRL_EXT
, ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
2152 * e1000e_release_hw_control - release control of the h/w to f/w
2153 * @adapter: address of board private structure
2155 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2156 * For ASF and Pass Through versions of f/w this means that the
2157 * driver is no longer loaded. For AMT version (only with 82573) i
2158 * of the f/w this means that the network i/f is closed.
2161 void e1000e_release_hw_control(struct e1000_adapter
*adapter
)
2163 struct e1000_hw
*hw
= &adapter
->hw
;
2167 /* Let firmware taken over control of h/w */
2168 if (adapter
->flags
& FLAG_HAS_SWSM_ON_LOAD
) {
2170 ew32(SWSM
, swsm
& ~E1000_SWSM_DRV_LOAD
);
2171 } else if (adapter
->flags
& FLAG_HAS_CTRLEXT_ON_LOAD
) {
2172 ctrl_ext
= er32(CTRL_EXT
);
2173 ew32(CTRL_EXT
, ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
2178 * e1000_alloc_ring_dma - allocate memory for a ring structure
2180 static int e1000_alloc_ring_dma(struct e1000_adapter
*adapter
,
2181 struct e1000_ring
*ring
)
2183 struct pci_dev
*pdev
= adapter
->pdev
;
2185 ring
->desc
= dma_alloc_coherent(&pdev
->dev
, ring
->size
, &ring
->dma
,
2194 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2195 * @tx_ring: Tx descriptor ring
2197 * Return 0 on success, negative on failure
2199 int e1000e_setup_tx_resources(struct e1000_ring
*tx_ring
)
2201 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
2202 int err
= -ENOMEM
, size
;
2204 size
= sizeof(struct e1000_buffer
) * tx_ring
->count
;
2205 tx_ring
->buffer_info
= vzalloc(size
);
2206 if (!tx_ring
->buffer_info
)
2209 /* round up to nearest 4K */
2210 tx_ring
->size
= tx_ring
->count
* sizeof(struct e1000_tx_desc
);
2211 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
2213 err
= e1000_alloc_ring_dma(adapter
, tx_ring
);
2217 tx_ring
->next_to_use
= 0;
2218 tx_ring
->next_to_clean
= 0;
2222 vfree(tx_ring
->buffer_info
);
2223 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2228 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2229 * @rx_ring: Rx descriptor ring
2231 * Returns 0 on success, negative on failure
2233 int e1000e_setup_rx_resources(struct e1000_ring
*rx_ring
)
2235 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
2236 struct e1000_buffer
*buffer_info
;
2237 int i
, size
, desc_len
, err
= -ENOMEM
;
2239 size
= sizeof(struct e1000_buffer
) * rx_ring
->count
;
2240 rx_ring
->buffer_info
= vzalloc(size
);
2241 if (!rx_ring
->buffer_info
)
2244 for (i
= 0; i
< rx_ring
->count
; i
++) {
2245 buffer_info
= &rx_ring
->buffer_info
[i
];
2246 buffer_info
->ps_pages
= kcalloc(PS_PAGE_BUFFERS
,
2247 sizeof(struct e1000_ps_page
),
2249 if (!buffer_info
->ps_pages
)
2253 desc_len
= sizeof(union e1000_rx_desc_packet_split
);
2255 /* Round up to nearest 4K */
2256 rx_ring
->size
= rx_ring
->count
* desc_len
;
2257 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
2259 err
= e1000_alloc_ring_dma(adapter
, rx_ring
);
2263 rx_ring
->next_to_clean
= 0;
2264 rx_ring
->next_to_use
= 0;
2265 rx_ring
->rx_skb_top
= NULL
;
2270 for (i
= 0; i
< rx_ring
->count
; i
++) {
2271 buffer_info
= &rx_ring
->buffer_info
[i
];
2272 kfree(buffer_info
->ps_pages
);
2275 vfree(rx_ring
->buffer_info
);
2276 e_err("Unable to allocate memory for the receive descriptor ring\n");
2281 * e1000_clean_tx_ring - Free Tx Buffers
2282 * @tx_ring: Tx descriptor ring
2284 static void e1000_clean_tx_ring(struct e1000_ring
*tx_ring
)
2286 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
2287 struct e1000_buffer
*buffer_info
;
2291 for (i
= 0; i
< tx_ring
->count
; i
++) {
2292 buffer_info
= &tx_ring
->buffer_info
[i
];
2293 e1000_put_txbuf(tx_ring
, buffer_info
);
2296 netdev_reset_queue(adapter
->netdev
);
2297 size
= sizeof(struct e1000_buffer
) * tx_ring
->count
;
2298 memset(tx_ring
->buffer_info
, 0, size
);
2300 memset(tx_ring
->desc
, 0, tx_ring
->size
);
2302 tx_ring
->next_to_use
= 0;
2303 tx_ring
->next_to_clean
= 0;
2305 writel(0, tx_ring
->head
);
2306 if (tx_ring
->adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
2307 e1000e_update_tdt_wa(tx_ring
, 0);
2309 writel(0, tx_ring
->tail
);
2313 * e1000e_free_tx_resources - Free Tx Resources per Queue
2314 * @tx_ring: Tx descriptor ring
2316 * Free all transmit software resources
2318 void e1000e_free_tx_resources(struct e1000_ring
*tx_ring
)
2320 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
2321 struct pci_dev
*pdev
= adapter
->pdev
;
2323 e1000_clean_tx_ring(tx_ring
);
2325 vfree(tx_ring
->buffer_info
);
2326 tx_ring
->buffer_info
= NULL
;
2328 dma_free_coherent(&pdev
->dev
, tx_ring
->size
, tx_ring
->desc
,
2330 tx_ring
->desc
= NULL
;
2334 * e1000e_free_rx_resources - Free Rx Resources
2335 * @rx_ring: Rx descriptor ring
2337 * Free all receive software resources
2339 void e1000e_free_rx_resources(struct e1000_ring
*rx_ring
)
2341 struct e1000_adapter
*adapter
= rx_ring
->adapter
;
2342 struct pci_dev
*pdev
= adapter
->pdev
;
2345 e1000_clean_rx_ring(rx_ring
);
2347 for (i
= 0; i
< rx_ring
->count
; i
++)
2348 kfree(rx_ring
->buffer_info
[i
].ps_pages
);
2350 vfree(rx_ring
->buffer_info
);
2351 rx_ring
->buffer_info
= NULL
;
2353 dma_free_coherent(&pdev
->dev
, rx_ring
->size
, rx_ring
->desc
,
2355 rx_ring
->desc
= NULL
;
2359 * e1000_update_itr - update the dynamic ITR value based on statistics
2360 * @adapter: pointer to adapter
2361 * @itr_setting: current adapter->itr
2362 * @packets: the number of packets during this measurement interval
2363 * @bytes: the number of bytes during this measurement interval
2365 * Stores a new ITR value based on packets and byte
2366 * counts during the last interrupt. The advantage of per interrupt
2367 * computation is faster updates and more accurate ITR for the current
2368 * traffic pattern. Constants in this function were computed
2369 * based on theoretical maximum wire speed and thresholds were set based
2370 * on testing data as well as attempting to minimize response time
2371 * while increasing bulk throughput. This functionality is controlled
2372 * by the InterruptThrottleRate module parameter.
2374 static unsigned int e1000_update_itr(struct e1000_adapter
*adapter
,
2375 u16 itr_setting
, int packets
,
2378 unsigned int retval
= itr_setting
;
2383 switch (itr_setting
) {
2384 case lowest_latency
:
2385 /* handle TSO and jumbo frames */
2386 if (bytes
/packets
> 8000)
2387 retval
= bulk_latency
;
2388 else if ((packets
< 5) && (bytes
> 512))
2389 retval
= low_latency
;
2391 case low_latency
: /* 50 usec aka 20000 ints/s */
2392 if (bytes
> 10000) {
2393 /* this if handles the TSO accounting */
2394 if (bytes
/packets
> 8000)
2395 retval
= bulk_latency
;
2396 else if ((packets
< 10) || ((bytes
/packets
) > 1200))
2397 retval
= bulk_latency
;
2398 else if ((packets
> 35))
2399 retval
= lowest_latency
;
2400 } else if (bytes
/packets
> 2000) {
2401 retval
= bulk_latency
;
2402 } else if (packets
<= 2 && bytes
< 512) {
2403 retval
= lowest_latency
;
2406 case bulk_latency
: /* 250 usec aka 4000 ints/s */
2407 if (bytes
> 25000) {
2409 retval
= low_latency
;
2410 } else if (bytes
< 6000) {
2411 retval
= low_latency
;
2419 static void e1000_set_itr(struct e1000_adapter
*adapter
)
2421 struct e1000_hw
*hw
= &adapter
->hw
;
2423 u32 new_itr
= adapter
->itr
;
2425 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2426 if (adapter
->link_speed
!= SPEED_1000
) {
2432 if (adapter
->flags2
& FLAG2_DISABLE_AIM
) {
2437 adapter
->tx_itr
= e1000_update_itr(adapter
,
2439 adapter
->total_tx_packets
,
2440 adapter
->total_tx_bytes
);
2441 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2442 if (adapter
->itr_setting
== 3 && adapter
->tx_itr
== lowest_latency
)
2443 adapter
->tx_itr
= low_latency
;
2445 adapter
->rx_itr
= e1000_update_itr(adapter
,
2447 adapter
->total_rx_packets
,
2448 adapter
->total_rx_bytes
);
2449 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2450 if (adapter
->itr_setting
== 3 && adapter
->rx_itr
== lowest_latency
)
2451 adapter
->rx_itr
= low_latency
;
2453 current_itr
= max(adapter
->rx_itr
, adapter
->tx_itr
);
2455 switch (current_itr
) {
2456 /* counts and packets in update_itr are dependent on these numbers */
2457 case lowest_latency
:
2461 new_itr
= 20000; /* aka hwitr = ~200 */
2471 if (new_itr
!= adapter
->itr
) {
2473 * this attempts to bias the interrupt rate towards Bulk
2474 * by adding intermediate steps when interrupt rate is
2477 new_itr
= new_itr
> adapter
->itr
?
2478 min(adapter
->itr
+ (new_itr
>> 2), new_itr
) :
2480 adapter
->itr
= new_itr
;
2481 adapter
->rx_ring
->itr_val
= new_itr
;
2482 if (adapter
->msix_entries
)
2483 adapter
->rx_ring
->set_itr
= 1;
2486 ew32(ITR
, 1000000000 / (new_itr
* 256));
2493 * e1000e_write_itr - write the ITR value to the appropriate registers
2494 * @adapter: address of board private structure
2495 * @itr: new ITR value to program
2497 * e1000e_write_itr determines if the adapter is in MSI-X mode
2498 * and, if so, writes the EITR registers with the ITR value.
2499 * Otherwise, it writes the ITR value into the ITR register.
2501 void e1000e_write_itr(struct e1000_adapter
*adapter
, u32 itr
)
2503 struct e1000_hw
*hw
= &adapter
->hw
;
2504 u32 new_itr
= itr
? 1000000000 / (itr
* 256) : 0;
2506 if (adapter
->msix_entries
) {
2509 for (vector
= 0; vector
< adapter
->num_vectors
; vector
++)
2510 writel(new_itr
, hw
->hw_addr
+ E1000_EITR_82574(vector
));
2517 * e1000_alloc_queues - Allocate memory for all rings
2518 * @adapter: board private structure to initialize
2520 static int __devinit
e1000_alloc_queues(struct e1000_adapter
*adapter
)
2522 int size
= sizeof(struct e1000_ring
);
2524 adapter
->tx_ring
= kzalloc(size
, GFP_KERNEL
);
2525 if (!adapter
->tx_ring
)
2527 adapter
->tx_ring
->count
= adapter
->tx_ring_count
;
2528 adapter
->tx_ring
->adapter
= adapter
;
2530 adapter
->rx_ring
= kzalloc(size
, GFP_KERNEL
);
2531 if (!adapter
->rx_ring
)
2533 adapter
->rx_ring
->count
= adapter
->rx_ring_count
;
2534 adapter
->rx_ring
->adapter
= adapter
;
2538 e_err("Unable to allocate memory for queues\n");
2539 kfree(adapter
->rx_ring
);
2540 kfree(adapter
->tx_ring
);
2545 * e1000e_poll - NAPI Rx polling callback
2546 * @napi: struct associated with this polling callback
2547 * @weight: number of packets driver is allowed to process this poll
2549 static int e1000e_poll(struct napi_struct
*napi
, int weight
)
2551 struct e1000_adapter
*adapter
= container_of(napi
, struct e1000_adapter
,
2553 struct e1000_hw
*hw
= &adapter
->hw
;
2554 struct net_device
*poll_dev
= adapter
->netdev
;
2555 int tx_cleaned
= 1, work_done
= 0;
2557 adapter
= netdev_priv(poll_dev
);
2559 if (!adapter
->msix_entries
||
2560 (adapter
->rx_ring
->ims_val
& adapter
->tx_ring
->ims_val
))
2561 tx_cleaned
= e1000_clean_tx_irq(adapter
->tx_ring
);
2563 adapter
->clean_rx(adapter
->rx_ring
, &work_done
, weight
);
2568 /* If weight not fully consumed, exit the polling mode */
2569 if (work_done
< weight
) {
2570 if (adapter
->itr_setting
& 3)
2571 e1000_set_itr(adapter
);
2572 napi_complete(napi
);
2573 if (!test_bit(__E1000_DOWN
, &adapter
->state
)) {
2574 if (adapter
->msix_entries
)
2575 ew32(IMS
, adapter
->rx_ring
->ims_val
);
2577 e1000_irq_enable(adapter
);
2584 static int e1000_vlan_rx_add_vid(struct net_device
*netdev
, u16 vid
)
2586 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2587 struct e1000_hw
*hw
= &adapter
->hw
;
2590 /* don't update vlan cookie if already programmed */
2591 if ((adapter
->hw
.mng_cookie
.status
&
2592 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) &&
2593 (vid
== adapter
->mng_vlan_id
))
2596 /* add VID to filter table */
2597 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2598 index
= (vid
>> 5) & 0x7F;
2599 vfta
= E1000_READ_REG_ARRAY(hw
, E1000_VFTA
, index
);
2600 vfta
|= (1 << (vid
& 0x1F));
2601 hw
->mac
.ops
.write_vfta(hw
, index
, vfta
);
2604 set_bit(vid
, adapter
->active_vlans
);
2609 static int e1000_vlan_rx_kill_vid(struct net_device
*netdev
, u16 vid
)
2611 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
2612 struct e1000_hw
*hw
= &adapter
->hw
;
2615 if ((adapter
->hw
.mng_cookie
.status
&
2616 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) &&
2617 (vid
== adapter
->mng_vlan_id
)) {
2618 /* release control to f/w */
2619 e1000e_release_hw_control(adapter
);
2623 /* remove VID from filter table */
2624 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2625 index
= (vid
>> 5) & 0x7F;
2626 vfta
= E1000_READ_REG_ARRAY(hw
, E1000_VFTA
, index
);
2627 vfta
&= ~(1 << (vid
& 0x1F));
2628 hw
->mac
.ops
.write_vfta(hw
, index
, vfta
);
2631 clear_bit(vid
, adapter
->active_vlans
);
2637 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2638 * @adapter: board private structure to initialize
2640 static void e1000e_vlan_filter_disable(struct e1000_adapter
*adapter
)
2642 struct net_device
*netdev
= adapter
->netdev
;
2643 struct e1000_hw
*hw
= &adapter
->hw
;
2646 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2647 /* disable VLAN receive filtering */
2649 rctl
&= ~(E1000_RCTL_VFE
| E1000_RCTL_CFIEN
);
2652 if (adapter
->mng_vlan_id
!= (u16
)E1000_MNG_VLAN_NONE
) {
2653 e1000_vlan_rx_kill_vid(netdev
, adapter
->mng_vlan_id
);
2654 adapter
->mng_vlan_id
= E1000_MNG_VLAN_NONE
;
2660 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2661 * @adapter: board private structure to initialize
2663 static void e1000e_vlan_filter_enable(struct e1000_adapter
*adapter
)
2665 struct e1000_hw
*hw
= &adapter
->hw
;
2668 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
) {
2669 /* enable VLAN receive filtering */
2671 rctl
|= E1000_RCTL_VFE
;
2672 rctl
&= ~E1000_RCTL_CFIEN
;
2678 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2679 * @adapter: board private structure to initialize
2681 static void e1000e_vlan_strip_disable(struct e1000_adapter
*adapter
)
2683 struct e1000_hw
*hw
= &adapter
->hw
;
2686 /* disable VLAN tag insert/strip */
2688 ctrl
&= ~E1000_CTRL_VME
;
2693 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2694 * @adapter: board private structure to initialize
2696 static void e1000e_vlan_strip_enable(struct e1000_adapter
*adapter
)
2698 struct e1000_hw
*hw
= &adapter
->hw
;
2701 /* enable VLAN tag insert/strip */
2703 ctrl
|= E1000_CTRL_VME
;
2707 static void e1000_update_mng_vlan(struct e1000_adapter
*adapter
)
2709 struct net_device
*netdev
= adapter
->netdev
;
2710 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
2711 u16 old_vid
= adapter
->mng_vlan_id
;
2713 if (adapter
->hw
.mng_cookie
.status
&
2714 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
2715 e1000_vlan_rx_add_vid(netdev
, vid
);
2716 adapter
->mng_vlan_id
= vid
;
2719 if ((old_vid
!= (u16
)E1000_MNG_VLAN_NONE
) && (vid
!= old_vid
))
2720 e1000_vlan_rx_kill_vid(netdev
, old_vid
);
2723 static void e1000_restore_vlan(struct e1000_adapter
*adapter
)
2727 e1000_vlan_rx_add_vid(adapter
->netdev
, 0);
2729 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
2730 e1000_vlan_rx_add_vid(adapter
->netdev
, vid
);
2733 static void e1000_init_manageability_pt(struct e1000_adapter
*adapter
)
2735 struct e1000_hw
*hw
= &adapter
->hw
;
2736 u32 manc
, manc2h
, mdef
, i
, j
;
2738 if (!(adapter
->flags
& FLAG_MNG_PT_ENABLED
))
2744 * enable receiving management packets to the host. this will probably
2745 * generate destination unreachable messages from the host OS, but
2746 * the packets will be handled on SMBUS
2748 manc
|= E1000_MANC_EN_MNG2HOST
;
2749 manc2h
= er32(MANC2H
);
2751 switch (hw
->mac
.type
) {
2753 manc2h
|= (E1000_MANC2H_PORT_623
| E1000_MANC2H_PORT_664
);
2758 * Check if IPMI pass-through decision filter already exists;
2761 for (i
= 0, j
= 0; i
< 8; i
++) {
2762 mdef
= er32(MDEF(i
));
2764 /* Ignore filters with anything other than IPMI ports */
2765 if (mdef
& ~(E1000_MDEF_PORT_623
| E1000_MDEF_PORT_664
))
2768 /* Enable this decision filter in MANC2H */
2775 if (j
== (E1000_MDEF_PORT_623
| E1000_MDEF_PORT_664
))
2778 /* Create new decision filter in an empty filter */
2779 for (i
= 0, j
= 0; i
< 8; i
++)
2780 if (er32(MDEF(i
)) == 0) {
2781 ew32(MDEF(i
), (E1000_MDEF_PORT_623
|
2782 E1000_MDEF_PORT_664
));
2789 e_warn("Unable to create IPMI pass-through filter\n");
2793 ew32(MANC2H
, manc2h
);
2798 * e1000_configure_tx - Configure Transmit Unit after Reset
2799 * @adapter: board private structure
2801 * Configure the Tx unit of the MAC after a reset.
2803 static void e1000_configure_tx(struct e1000_adapter
*adapter
)
2805 struct e1000_hw
*hw
= &adapter
->hw
;
2806 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
2810 /* Setup the HW Tx Head and Tail descriptor pointers */
2811 tdba
= tx_ring
->dma
;
2812 tdlen
= tx_ring
->count
* sizeof(struct e1000_tx_desc
);
2813 ew32(TDBAL(0), (tdba
& DMA_BIT_MASK(32)));
2814 ew32(TDBAH(0), (tdba
>> 32));
2815 ew32(TDLEN(0), tdlen
);
2818 tx_ring
->head
= adapter
->hw
.hw_addr
+ E1000_TDH(0);
2819 tx_ring
->tail
= adapter
->hw
.hw_addr
+ E1000_TDT(0);
2821 /* Set the Tx Interrupt Delay register */
2822 ew32(TIDV
, adapter
->tx_int_delay
);
2823 /* Tx irq moderation */
2824 ew32(TADV
, adapter
->tx_abs_int_delay
);
2826 if (adapter
->flags2
& FLAG2_DMA_BURST
) {
2827 u32 txdctl
= er32(TXDCTL(0));
2828 txdctl
&= ~(E1000_TXDCTL_PTHRESH
| E1000_TXDCTL_HTHRESH
|
2829 E1000_TXDCTL_WTHRESH
);
2831 * set up some performance related parameters to encourage the
2832 * hardware to use the bus more efficiently in bursts, depends
2833 * on the tx_int_delay to be enabled,
2834 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2835 * hthresh = 1 ==> prefetch when one or more available
2836 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2837 * BEWARE: this seems to work but should be considered first if
2838 * there are Tx hangs or other Tx related bugs
2840 txdctl
|= E1000_TXDCTL_DMA_BURST_ENABLE
;
2841 ew32(TXDCTL(0), txdctl
);
2843 /* erratum work around: set txdctl the same for both queues */
2844 ew32(TXDCTL(1), er32(TXDCTL(0)));
2846 if (adapter
->flags
& FLAG_TARC_SPEED_MODE_BIT
) {
2847 tarc
= er32(TARC(0));
2849 * set the speed mode bit, we'll clear it if we're not at
2850 * gigabit link later
2852 #define SPEED_MODE_BIT (1 << 21)
2853 tarc
|= SPEED_MODE_BIT
;
2854 ew32(TARC(0), tarc
);
2857 /* errata: program both queues to unweighted RR */
2858 if (adapter
->flags
& FLAG_TARC_SET_BIT_ZERO
) {
2859 tarc
= er32(TARC(0));
2861 ew32(TARC(0), tarc
);
2862 tarc
= er32(TARC(1));
2864 ew32(TARC(1), tarc
);
2867 /* Setup Transmit Descriptor Settings for eop descriptor */
2868 adapter
->txd_cmd
= E1000_TXD_CMD_EOP
| E1000_TXD_CMD_IFCS
;
2870 /* only set IDE if we are delaying interrupts using the timers */
2871 if (adapter
->tx_int_delay
)
2872 adapter
->txd_cmd
|= E1000_TXD_CMD_IDE
;
2874 /* enable Report Status bit */
2875 adapter
->txd_cmd
|= E1000_TXD_CMD_RS
;
2877 hw
->mac
.ops
.config_collision_dist(hw
);
2881 * e1000_setup_rctl - configure the receive control registers
2882 * @adapter: Board private structure
2884 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2885 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2886 static void e1000_setup_rctl(struct e1000_adapter
*adapter
)
2888 struct e1000_hw
*hw
= &adapter
->hw
;
2892 /* Workaround Si errata on PCHx - configure jumbo frame flow */
2893 if (hw
->mac
.type
>= e1000_pch2lan
) {
2896 if (adapter
->netdev
->mtu
> ETH_DATA_LEN
)
2897 ret_val
= e1000_lv_jumbo_workaround_ich8lan(hw
, true);
2899 ret_val
= e1000_lv_jumbo_workaround_ich8lan(hw
, false);
2902 e_dbg("failed to enable jumbo frame workaround mode\n");
2905 /* Program MC offset vector base */
2907 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
2908 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
|
2909 E1000_RCTL_LBM_NO
| E1000_RCTL_RDMTS_HALF
|
2910 (adapter
->hw
.mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
2912 /* Do not Store bad packets */
2913 rctl
&= ~E1000_RCTL_SBP
;
2915 /* Enable Long Packet receive */
2916 if (adapter
->netdev
->mtu
<= ETH_DATA_LEN
)
2917 rctl
&= ~E1000_RCTL_LPE
;
2919 rctl
|= E1000_RCTL_LPE
;
2921 /* Some systems expect that the CRC is included in SMBUS traffic. The
2922 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2923 * host memory when this is enabled
2925 if (adapter
->flags2
& FLAG2_CRC_STRIPPING
)
2926 rctl
|= E1000_RCTL_SECRC
;
2928 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2929 if ((hw
->phy
.type
== e1000_phy_82577
) && (rctl
& E1000_RCTL_LPE
)) {
2932 e1e_rphy(hw
, PHY_REG(770, 26), &phy_data
);
2934 phy_data
|= (1 << 2);
2935 e1e_wphy(hw
, PHY_REG(770, 26), phy_data
);
2937 e1e_rphy(hw
, 22, &phy_data
);
2939 phy_data
|= (1 << 14);
2940 e1e_wphy(hw
, 0x10, 0x2823);
2941 e1e_wphy(hw
, 0x11, 0x0003);
2942 e1e_wphy(hw
, 22, phy_data
);
2945 /* Setup buffer sizes */
2946 rctl
&= ~E1000_RCTL_SZ_4096
;
2947 rctl
|= E1000_RCTL_BSEX
;
2948 switch (adapter
->rx_buffer_len
) {
2951 rctl
|= E1000_RCTL_SZ_2048
;
2952 rctl
&= ~E1000_RCTL_BSEX
;
2955 rctl
|= E1000_RCTL_SZ_4096
;
2958 rctl
|= E1000_RCTL_SZ_8192
;
2961 rctl
|= E1000_RCTL_SZ_16384
;
2965 /* Enable Extended Status in all Receive Descriptors */
2966 rfctl
= er32(RFCTL
);
2967 rfctl
|= E1000_RFCTL_EXTEN
;
2971 * 82571 and greater support packet-split where the protocol
2972 * header is placed in skb->data and the packet data is
2973 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2974 * In the case of a non-split, skb->data is linearly filled,
2975 * followed by the page buffers. Therefore, skb->data is
2976 * sized to hold the largest protocol header.
2978 * allocations using alloc_page take too long for regular MTU
2979 * so only enable packet split for jumbo frames
2981 * Using pages when the page size is greater than 16k wastes
2982 * a lot of memory, since we allocate 3 pages at all times
2985 pages
= PAGE_USE_COUNT(adapter
->netdev
->mtu
);
2986 if ((pages
<= 3) && (PAGE_SIZE
<= 16384) && (rctl
& E1000_RCTL_LPE
))
2987 adapter
->rx_ps_pages
= pages
;
2989 adapter
->rx_ps_pages
= 0;
2991 if (adapter
->rx_ps_pages
) {
2994 /* Enable Packet split descriptors */
2995 rctl
|= E1000_RCTL_DTYP_PS
;
2997 psrctl
|= adapter
->rx_ps_bsize0
>>
2998 E1000_PSRCTL_BSIZE0_SHIFT
;
3000 switch (adapter
->rx_ps_pages
) {
3002 psrctl
|= PAGE_SIZE
<<
3003 E1000_PSRCTL_BSIZE3_SHIFT
;
3005 psrctl
|= PAGE_SIZE
<<
3006 E1000_PSRCTL_BSIZE2_SHIFT
;
3008 psrctl
|= PAGE_SIZE
>>
3009 E1000_PSRCTL_BSIZE1_SHIFT
;
3013 ew32(PSRCTL
, psrctl
);
3016 /* This is useful for sniffing bad packets. */
3017 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3018 /* UPE and MPE will be handled by normal PROMISC logic
3019 * in e1000e_set_rx_mode */
3020 rctl
|= (E1000_RCTL_SBP
| /* Receive bad packets */
3021 E1000_RCTL_BAM
| /* RX All Bcast Pkts */
3022 E1000_RCTL_PMCF
); /* RX All MAC Ctrl Pkts */
3024 rctl
&= ~(E1000_RCTL_VFE
| /* Disable VLAN filter */
3025 E1000_RCTL_DPF
| /* Allow filtered pause */
3026 E1000_RCTL_CFIEN
); /* Dis VLAN CFIEN Filter */
3027 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3028 * and that breaks VLANs.
3033 /* just started the receive unit, no need to restart */
3034 adapter
->flags
&= ~FLAG_RX_RESTART_NOW
;
3038 * e1000_configure_rx - Configure Receive Unit after Reset
3039 * @adapter: board private structure
3041 * Configure the Rx unit of the MAC after a reset.
3043 static void e1000_configure_rx(struct e1000_adapter
*adapter
)
3045 struct e1000_hw
*hw
= &adapter
->hw
;
3046 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
3048 u32 rdlen
, rctl
, rxcsum
, ctrl_ext
;
3050 if (adapter
->rx_ps_pages
) {
3051 /* this is a 32 byte descriptor */
3052 rdlen
= rx_ring
->count
*
3053 sizeof(union e1000_rx_desc_packet_split
);
3054 adapter
->clean_rx
= e1000_clean_rx_irq_ps
;
3055 adapter
->alloc_rx_buf
= e1000_alloc_rx_buffers_ps
;
3056 } else if (adapter
->netdev
->mtu
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) {
3057 rdlen
= rx_ring
->count
* sizeof(union e1000_rx_desc_extended
);
3058 adapter
->clean_rx
= e1000_clean_jumbo_rx_irq
;
3059 adapter
->alloc_rx_buf
= e1000_alloc_jumbo_rx_buffers
;
3061 rdlen
= rx_ring
->count
* sizeof(union e1000_rx_desc_extended
);
3062 adapter
->clean_rx
= e1000_clean_rx_irq
;
3063 adapter
->alloc_rx_buf
= e1000_alloc_rx_buffers
;
3066 /* disable receives while setting up the descriptors */
3068 if (!(adapter
->flags2
& FLAG2_NO_DISABLE_RX
))
3069 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
3071 usleep_range(10000, 20000);
3073 if (adapter
->flags2
& FLAG2_DMA_BURST
) {
3075 * set the writeback threshold (only takes effect if the RDTR
3076 * is set). set GRAN=1 and write back up to 0x4 worth, and
3077 * enable prefetching of 0x20 Rx descriptors
3083 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE
);
3084 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE
);
3087 * override the delay timers for enabling bursting, only if
3088 * the value was not set by the user via module options
3090 if (adapter
->rx_int_delay
== DEFAULT_RDTR
)
3091 adapter
->rx_int_delay
= BURST_RDTR
;
3092 if (adapter
->rx_abs_int_delay
== DEFAULT_RADV
)
3093 adapter
->rx_abs_int_delay
= BURST_RADV
;
3096 /* set the Receive Delay Timer Register */
3097 ew32(RDTR
, adapter
->rx_int_delay
);
3099 /* irq moderation */
3100 ew32(RADV
, adapter
->rx_abs_int_delay
);
3101 if ((adapter
->itr_setting
!= 0) && (adapter
->itr
!= 0))
3102 e1000e_write_itr(adapter
, adapter
->itr
);
3104 ctrl_ext
= er32(CTRL_EXT
);
3105 /* Auto-Mask interrupts upon ICR access */
3106 ctrl_ext
|= E1000_CTRL_EXT_IAME
;
3107 ew32(IAM
, 0xffffffff);
3108 ew32(CTRL_EXT
, ctrl_ext
);
3112 * Setup the HW Rx Head and Tail Descriptor Pointers and
3113 * the Base and Length of the Rx Descriptor Ring
3115 rdba
= rx_ring
->dma
;
3116 ew32(RDBAL(0), (rdba
& DMA_BIT_MASK(32)));
3117 ew32(RDBAH(0), (rdba
>> 32));
3118 ew32(RDLEN(0), rdlen
);
3121 rx_ring
->head
= adapter
->hw
.hw_addr
+ E1000_RDH(0);
3122 rx_ring
->tail
= adapter
->hw
.hw_addr
+ E1000_RDT(0);
3124 /* Enable Receive Checksum Offload for TCP and UDP */
3125 rxcsum
= er32(RXCSUM
);
3126 if (adapter
->netdev
->features
& NETIF_F_RXCSUM
)
3127 rxcsum
|= E1000_RXCSUM_TUOFL
;
3129 rxcsum
&= ~E1000_RXCSUM_TUOFL
;
3130 ew32(RXCSUM
, rxcsum
);
3132 if (adapter
->hw
.mac
.type
== e1000_pch2lan
) {
3134 * With jumbo frames, excessive C-state transition
3135 * latencies result in dropped transactions.
3137 if (adapter
->netdev
->mtu
> ETH_DATA_LEN
) {
3138 u32 rxdctl
= er32(RXDCTL(0));
3139 ew32(RXDCTL(0), rxdctl
| 0x3);
3140 pm_qos_update_request(&adapter
->netdev
->pm_qos_req
, 55);
3142 pm_qos_update_request(&adapter
->netdev
->pm_qos_req
,
3143 PM_QOS_DEFAULT_VALUE
);
3147 /* Enable Receives */
3152 * e1000e_write_mc_addr_list - write multicast addresses to MTA
3153 * @netdev: network interface device structure
3155 * Writes multicast address list to the MTA hash table.
3156 * Returns: -ENOMEM on failure
3157 * 0 on no addresses written
3158 * X on writing X addresses to MTA
3160 static int e1000e_write_mc_addr_list(struct net_device
*netdev
)
3162 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3163 struct e1000_hw
*hw
= &adapter
->hw
;
3164 struct netdev_hw_addr
*ha
;
3168 if (netdev_mc_empty(netdev
)) {
3169 /* nothing to program, so clear mc list */
3170 hw
->mac
.ops
.update_mc_addr_list(hw
, NULL
, 0);
3174 mta_list
= kzalloc(netdev_mc_count(netdev
) * ETH_ALEN
, GFP_ATOMIC
);
3178 /* update_mc_addr_list expects a packed array of only addresses. */
3180 netdev_for_each_mc_addr(ha
, netdev
)
3181 memcpy(mta_list
+ (i
++ * ETH_ALEN
), ha
->addr
, ETH_ALEN
);
3183 hw
->mac
.ops
.update_mc_addr_list(hw
, mta_list
, i
);
3186 return netdev_mc_count(netdev
);
3190 * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3191 * @netdev: network interface device structure
3193 * Writes unicast address list to the RAR table.
3194 * Returns: -ENOMEM on failure/insufficient address space
3195 * 0 on no addresses written
3196 * X on writing X addresses to the RAR table
3198 static int e1000e_write_uc_addr_list(struct net_device
*netdev
)
3200 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3201 struct e1000_hw
*hw
= &adapter
->hw
;
3202 unsigned int rar_entries
= hw
->mac
.rar_entry_count
;
3205 /* save a rar entry for our hardware address */
3208 /* save a rar entry for the LAA workaround */
3209 if (adapter
->flags
& FLAG_RESET_OVERWRITES_LAA
)
3212 /* return ENOMEM indicating insufficient memory for addresses */
3213 if (netdev_uc_count(netdev
) > rar_entries
)
3216 if (!netdev_uc_empty(netdev
) && rar_entries
) {
3217 struct netdev_hw_addr
*ha
;
3220 * write the addresses in reverse order to avoid write
3223 netdev_for_each_uc_addr(ha
, netdev
) {
3226 hw
->mac
.ops
.rar_set(hw
, ha
->addr
, rar_entries
--);
3231 /* zero out the remaining RAR entries not used above */
3232 for (; rar_entries
> 0; rar_entries
--) {
3233 ew32(RAH(rar_entries
), 0);
3234 ew32(RAL(rar_entries
), 0);
3242 * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3243 * @netdev: network interface device structure
3245 * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3246 * address list or the network interface flags are updated. This routine is
3247 * responsible for configuring the hardware for proper unicast, multicast,
3248 * promiscuous mode, and all-multi behavior.
3250 static void e1000e_set_rx_mode(struct net_device
*netdev
)
3252 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3253 struct e1000_hw
*hw
= &adapter
->hw
;
3256 /* Check for Promiscuous and All Multicast modes */
3259 /* clear the affected bits */
3260 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
);
3262 if (netdev
->flags
& IFF_PROMISC
) {
3263 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
3264 /* Do not hardware filter VLANs in promisc mode */
3265 e1000e_vlan_filter_disable(adapter
);
3269 if (netdev
->flags
& IFF_ALLMULTI
) {
3270 rctl
|= E1000_RCTL_MPE
;
3273 * Write addresses to the MTA, if the attempt fails
3274 * then we should just turn on promiscuous mode so
3275 * that we can at least receive multicast traffic
3277 count
= e1000e_write_mc_addr_list(netdev
);
3279 rctl
|= E1000_RCTL_MPE
;
3281 e1000e_vlan_filter_enable(adapter
);
3283 * Write addresses to available RAR registers, if there is not
3284 * sufficient space to store all the addresses then enable
3285 * unicast promiscuous mode
3287 count
= e1000e_write_uc_addr_list(netdev
);
3289 rctl
|= E1000_RCTL_UPE
;
3294 if (netdev
->features
& NETIF_F_HW_VLAN_RX
)
3295 e1000e_vlan_strip_enable(adapter
);
3297 e1000e_vlan_strip_disable(adapter
);
3300 static void e1000e_setup_rss_hash(struct e1000_adapter
*adapter
)
3302 struct e1000_hw
*hw
= &adapter
->hw
;
3305 static const u32 rsskey
[10] = {
3306 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3307 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3310 /* Fill out hash function seed */
3311 for (i
= 0; i
< 10; i
++)
3312 ew32(RSSRK(i
), rsskey
[i
]);
3314 /* Direct all traffic to queue 0 */
3315 for (i
= 0; i
< 32; i
++)
3319 * Disable raw packet checksumming so that RSS hash is placed in
3320 * descriptor on writeback.
3322 rxcsum
= er32(RXCSUM
);
3323 rxcsum
|= E1000_RXCSUM_PCSD
;
3325 ew32(RXCSUM
, rxcsum
);
3327 mrqc
= (E1000_MRQC_RSS_FIELD_IPV4
|
3328 E1000_MRQC_RSS_FIELD_IPV4_TCP
|
3329 E1000_MRQC_RSS_FIELD_IPV6
|
3330 E1000_MRQC_RSS_FIELD_IPV6_TCP
|
3331 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX
);
3337 * e1000_configure - configure the hardware for Rx and Tx
3338 * @adapter: private board structure
3340 static void e1000_configure(struct e1000_adapter
*adapter
)
3342 struct e1000_ring
*rx_ring
= adapter
->rx_ring
;
3344 e1000e_set_rx_mode(adapter
->netdev
);
3346 e1000_restore_vlan(adapter
);
3347 e1000_init_manageability_pt(adapter
);
3349 e1000_configure_tx(adapter
);
3351 if (adapter
->netdev
->features
& NETIF_F_RXHASH
)
3352 e1000e_setup_rss_hash(adapter
);
3353 e1000_setup_rctl(adapter
);
3354 e1000_configure_rx(adapter
);
3355 adapter
->alloc_rx_buf(rx_ring
, e1000_desc_unused(rx_ring
), GFP_KERNEL
);
3359 * e1000e_power_up_phy - restore link in case the phy was powered down
3360 * @adapter: address of board private structure
3362 * The phy may be powered down to save power and turn off link when the
3363 * driver is unloaded and wake on lan is not enabled (among others)
3364 * *** this routine MUST be followed by a call to e1000e_reset ***
3366 void e1000e_power_up_phy(struct e1000_adapter
*adapter
)
3368 if (adapter
->hw
.phy
.ops
.power_up
)
3369 adapter
->hw
.phy
.ops
.power_up(&adapter
->hw
);
3371 adapter
->hw
.mac
.ops
.setup_link(&adapter
->hw
);
3375 * e1000_power_down_phy - Power down the PHY
3377 * Power down the PHY so no link is implied when interface is down.
3378 * The PHY cannot be powered down if management or WoL is active.
3380 static void e1000_power_down_phy(struct e1000_adapter
*adapter
)
3382 /* WoL is enabled */
3386 if (adapter
->hw
.phy
.ops
.power_down
)
3387 adapter
->hw
.phy
.ops
.power_down(&adapter
->hw
);
3391 * e1000e_reset - bring the hardware into a known good state
3393 * This function boots the hardware and enables some settings that
3394 * require a configuration cycle of the hardware - those cannot be
3395 * set/changed during runtime. After reset the device needs to be
3396 * properly configured for Rx, Tx etc.
3398 void e1000e_reset(struct e1000_adapter
*adapter
)
3400 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
3401 struct e1000_fc_info
*fc
= &adapter
->hw
.fc
;
3402 struct e1000_hw
*hw
= &adapter
->hw
;
3403 u32 tx_space
, min_tx_space
, min_rx_space
;
3404 u32 pba
= adapter
->pba
;
3407 /* reset Packet Buffer Allocation to default */
3410 if (adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) {
3412 * To maintain wire speed transmits, the Tx FIFO should be
3413 * large enough to accommodate two full transmit packets,
3414 * rounded up to the next 1KB and expressed in KB. Likewise,
3415 * the Rx FIFO should be large enough to accommodate at least
3416 * one full receive packet and is similarly rounded up and
3420 /* upper 16 bits has Tx packet buffer allocation size in KB */
3421 tx_space
= pba
>> 16;
3422 /* lower 16 bits has Rx packet buffer allocation size in KB */
3425 * the Tx fifo also stores 16 bytes of information about the Tx
3426 * but don't include ethernet FCS because hardware appends it
3428 min_tx_space
= (adapter
->max_frame_size
+
3429 sizeof(struct e1000_tx_desc
) -
3431 min_tx_space
= ALIGN(min_tx_space
, 1024);
3432 min_tx_space
>>= 10;
3433 /* software strips receive CRC, so leave room for it */
3434 min_rx_space
= adapter
->max_frame_size
;
3435 min_rx_space
= ALIGN(min_rx_space
, 1024);
3436 min_rx_space
>>= 10;
3439 * If current Tx allocation is less than the min Tx FIFO size,
3440 * and the min Tx FIFO size is less than the current Rx FIFO
3441 * allocation, take space away from current Rx allocation
3443 if ((tx_space
< min_tx_space
) &&
3444 ((min_tx_space
- tx_space
) < pba
)) {
3445 pba
-= min_tx_space
- tx_space
;
3448 * if short on Rx space, Rx wins and must trump Tx
3451 if (pba
< min_rx_space
)
3459 * flow control settings
3461 * The high water mark must be low enough to fit one full frame
3462 * (or the size used for early receive) above it in the Rx FIFO.
3463 * Set it to the lower of:
3464 * - 90% of the Rx FIFO size, and
3465 * - the full Rx FIFO size minus one full frame
3467 if (adapter
->flags
& FLAG_DISABLE_FC_PAUSE_TIME
)
3468 fc
->pause_time
= 0xFFFF;
3470 fc
->pause_time
= E1000_FC_PAUSE_TIME
;
3471 fc
->send_xon
= true;
3472 fc
->current_mode
= fc
->requested_mode
;
3474 switch (hw
->mac
.type
) {
3476 case e1000_ich10lan
:
3477 if (adapter
->netdev
->mtu
> ETH_DATA_LEN
) {
3480 fc
->high_water
= 0x2800;
3481 fc
->low_water
= fc
->high_water
- 8;
3486 hwm
= min(((pba
<< 10) * 9 / 10),
3487 ((pba
<< 10) - adapter
->max_frame_size
));
3489 fc
->high_water
= hwm
& E1000_FCRTH_RTH
; /* 8-byte granularity */
3490 fc
->low_water
= fc
->high_water
- 8;
3494 * Workaround PCH LOM adapter hangs with certain network
3495 * loads. If hangs persist, try disabling Tx flow control.
3497 if (adapter
->netdev
->mtu
> ETH_DATA_LEN
) {
3498 fc
->high_water
= 0x3500;
3499 fc
->low_water
= 0x1500;
3501 fc
->high_water
= 0x5000;
3502 fc
->low_water
= 0x3000;
3504 fc
->refresh_time
= 0x1000;
3508 fc
->high_water
= 0x05C20;
3509 fc
->low_water
= 0x05048;
3510 fc
->pause_time
= 0x0650;
3511 fc
->refresh_time
= 0x0400;
3512 if (adapter
->netdev
->mtu
> ETH_DATA_LEN
) {
3520 * Alignment of Tx data is on an arbitrary byte boundary with the
3521 * maximum size per Tx descriptor limited only to the transmit
3522 * allocation of the packet buffer minus 96 bytes with an upper
3523 * limit of 24KB due to receive synchronization limitations.
3525 adapter
->tx_fifo_limit
= min_t(u32
, ((er32(PBA
) >> 16) << 10) - 96,
3529 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3530 * fit in receive buffer.
3532 if (adapter
->itr_setting
& 0x3) {
3533 if ((adapter
->max_frame_size
* 2) > (pba
<< 10)) {
3534 if (!(adapter
->flags2
& FLAG2_DISABLE_AIM
)) {
3535 dev_info(&adapter
->pdev
->dev
,
3536 "Interrupt Throttle Rate turned off\n");
3537 adapter
->flags2
|= FLAG2_DISABLE_AIM
;
3538 e1000e_write_itr(adapter
, 0);
3540 } else if (adapter
->flags2
& FLAG2_DISABLE_AIM
) {
3541 dev_info(&adapter
->pdev
->dev
,
3542 "Interrupt Throttle Rate turned on\n");
3543 adapter
->flags2
&= ~FLAG2_DISABLE_AIM
;
3544 adapter
->itr
= 20000;
3545 e1000e_write_itr(adapter
, adapter
->itr
);
3549 /* Allow time for pending master requests to run */
3550 mac
->ops
.reset_hw(hw
);
3553 * For parts with AMT enabled, let the firmware know
3554 * that the network interface is in control
3556 if (adapter
->flags
& FLAG_HAS_AMT
)
3557 e1000e_get_hw_control(adapter
);
3561 if (mac
->ops
.init_hw(hw
))
3562 e_err("Hardware Error\n");
3564 e1000_update_mng_vlan(adapter
);
3566 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3567 ew32(VET
, ETH_P_8021Q
);
3569 e1000e_reset_adaptive(hw
);
3571 if (!netif_running(adapter
->netdev
) &&
3572 !test_bit(__E1000_TESTING
, &adapter
->state
)) {
3573 e1000_power_down_phy(adapter
);
3577 e1000_get_phy_info(hw
);
3579 if ((adapter
->flags
& FLAG_HAS_SMART_POWER_DOWN
) &&
3580 !(adapter
->flags
& FLAG_SMART_POWER_DOWN
)) {
3583 * speed up time to link by disabling smart power down, ignore
3584 * the return value of this function because there is nothing
3585 * different we would do if it failed
3587 e1e_rphy(hw
, IGP02E1000_PHY_POWER_MGMT
, &phy_data
);
3588 phy_data
&= ~IGP02E1000_PM_SPD
;
3589 e1e_wphy(hw
, IGP02E1000_PHY_POWER_MGMT
, phy_data
);
3593 int e1000e_up(struct e1000_adapter
*adapter
)
3595 struct e1000_hw
*hw
= &adapter
->hw
;
3597 /* hardware has been reset, we need to reload some things */
3598 e1000_configure(adapter
);
3600 clear_bit(__E1000_DOWN
, &adapter
->state
);
3602 if (adapter
->msix_entries
)
3603 e1000_configure_msix(adapter
);
3604 e1000_irq_enable(adapter
);
3606 netif_start_queue(adapter
->netdev
);
3608 /* fire a link change interrupt to start the watchdog */
3609 if (adapter
->msix_entries
)
3610 ew32(ICS
, E1000_ICS_LSC
| E1000_ICR_OTHER
);
3612 ew32(ICS
, E1000_ICS_LSC
);
3617 static void e1000e_flush_descriptors(struct e1000_adapter
*adapter
)
3619 struct e1000_hw
*hw
= &adapter
->hw
;
3621 if (!(adapter
->flags2
& FLAG2_DMA_BURST
))
3624 /* flush pending descriptor writebacks to memory */
3625 ew32(TIDV
, adapter
->tx_int_delay
| E1000_TIDV_FPD
);
3626 ew32(RDTR
, adapter
->rx_int_delay
| E1000_RDTR_FPD
);
3628 /* execute the writes immediately */
3632 * due to rare timing issues, write to TIDV/RDTR again to ensure the
3633 * write is successful
3635 ew32(TIDV
, adapter
->tx_int_delay
| E1000_TIDV_FPD
);
3636 ew32(RDTR
, adapter
->rx_int_delay
| E1000_RDTR_FPD
);
3638 /* execute the writes immediately */
3642 static void e1000e_update_stats(struct e1000_adapter
*adapter
);
3644 void e1000e_down(struct e1000_adapter
*adapter
)
3646 struct net_device
*netdev
= adapter
->netdev
;
3647 struct e1000_hw
*hw
= &adapter
->hw
;
3651 * signal that we're down so the interrupt handler does not
3652 * reschedule our watchdog timer
3654 set_bit(__E1000_DOWN
, &adapter
->state
);
3656 /* disable receives in the hardware */
3658 if (!(adapter
->flags2
& FLAG2_NO_DISABLE_RX
))
3659 ew32(RCTL
, rctl
& ~E1000_RCTL_EN
);
3660 /* flush and sleep below */
3662 netif_stop_queue(netdev
);
3664 /* disable transmits in the hardware */
3666 tctl
&= ~E1000_TCTL_EN
;
3669 /* flush both disables and wait for them to finish */
3671 usleep_range(10000, 20000);
3673 e1000_irq_disable(adapter
);
3675 del_timer_sync(&adapter
->watchdog_timer
);
3676 del_timer_sync(&adapter
->phy_info_timer
);
3678 netif_carrier_off(netdev
);
3680 spin_lock(&adapter
->stats64_lock
);
3681 e1000e_update_stats(adapter
);
3682 spin_unlock(&adapter
->stats64_lock
);
3684 e1000e_flush_descriptors(adapter
);
3685 e1000_clean_tx_ring(adapter
->tx_ring
);
3686 e1000_clean_rx_ring(adapter
->rx_ring
);
3688 adapter
->link_speed
= 0;
3689 adapter
->link_duplex
= 0;
3691 if (!pci_channel_offline(adapter
->pdev
))
3692 e1000e_reset(adapter
);
3695 * TODO: for power management, we could drop the link and
3696 * pci_disable_device here.
3700 void e1000e_reinit_locked(struct e1000_adapter
*adapter
)
3703 while (test_and_set_bit(__E1000_RESETTING
, &adapter
->state
))
3704 usleep_range(1000, 2000);
3705 e1000e_down(adapter
);
3707 clear_bit(__E1000_RESETTING
, &adapter
->state
);
3711 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3712 * @adapter: board private structure to initialize
3714 * e1000_sw_init initializes the Adapter private data structure.
3715 * Fields are initialized based on PCI device information and
3716 * OS network device settings (MTU size).
3718 static int __devinit
e1000_sw_init(struct e1000_adapter
*adapter
)
3720 struct net_device
*netdev
= adapter
->netdev
;
3722 adapter
->rx_buffer_len
= ETH_FRAME_LEN
+ VLAN_HLEN
+ ETH_FCS_LEN
;
3723 adapter
->rx_ps_bsize0
= 128;
3724 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3725 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
3726 adapter
->tx_ring_count
= E1000_DEFAULT_TXD
;
3727 adapter
->rx_ring_count
= E1000_DEFAULT_RXD
;
3729 spin_lock_init(&adapter
->stats64_lock
);
3731 e1000e_set_interrupt_capability(adapter
);
3733 if (e1000_alloc_queues(adapter
))
3736 /* Explicitly disable IRQ since the NIC can be in any state. */
3737 e1000_irq_disable(adapter
);
3739 set_bit(__E1000_DOWN
, &adapter
->state
);
3744 * e1000_intr_msi_test - Interrupt Handler
3745 * @irq: interrupt number
3746 * @data: pointer to a network interface device structure
3748 static irqreturn_t
e1000_intr_msi_test(int irq
, void *data
)
3750 struct net_device
*netdev
= data
;
3751 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3752 struct e1000_hw
*hw
= &adapter
->hw
;
3753 u32 icr
= er32(ICR
);
3755 e_dbg("icr is %08X\n", icr
);
3756 if (icr
& E1000_ICR_RXSEQ
) {
3757 adapter
->flags
&= ~FLAG_MSI_TEST_FAILED
;
3759 * Force memory writes to complete before acknowledging the
3760 * interrupt is handled.
3769 * e1000_test_msi_interrupt - Returns 0 for successful test
3770 * @adapter: board private struct
3772 * code flow taken from tg3.c
3774 static int e1000_test_msi_interrupt(struct e1000_adapter
*adapter
)
3776 struct net_device
*netdev
= adapter
->netdev
;
3777 struct e1000_hw
*hw
= &adapter
->hw
;
3780 /* poll_enable hasn't been called yet, so don't need disable */
3781 /* clear any pending events */
3784 /* free the real vector and request a test handler */
3785 e1000_free_irq(adapter
);
3786 e1000e_reset_interrupt_capability(adapter
);
3788 /* Assume that the test fails, if it succeeds then the test
3789 * MSI irq handler will unset this flag */
3790 adapter
->flags
|= FLAG_MSI_TEST_FAILED
;
3792 err
= pci_enable_msi(adapter
->pdev
);
3794 goto msi_test_failed
;
3796 err
= request_irq(adapter
->pdev
->irq
, e1000_intr_msi_test
, 0,
3797 netdev
->name
, netdev
);
3799 pci_disable_msi(adapter
->pdev
);
3800 goto msi_test_failed
;
3804 * Force memory writes to complete before enabling and firing an
3809 e1000_irq_enable(adapter
);
3811 /* fire an unusual interrupt on the test handler */
3812 ew32(ICS
, E1000_ICS_RXSEQ
);
3816 e1000_irq_disable(adapter
);
3818 rmb(); /* read flags after interrupt has been fired */
3820 if (adapter
->flags
& FLAG_MSI_TEST_FAILED
) {
3821 adapter
->int_mode
= E1000E_INT_MODE_LEGACY
;
3822 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3824 e_dbg("MSI interrupt test succeeded!\n");
3827 free_irq(adapter
->pdev
->irq
, netdev
);
3828 pci_disable_msi(adapter
->pdev
);
3831 e1000e_set_interrupt_capability(adapter
);
3832 return e1000_request_irq(adapter
);
3836 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3837 * @adapter: board private struct
3839 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3841 static int e1000_test_msi(struct e1000_adapter
*adapter
)
3846 if (!(adapter
->flags
& FLAG_MSI_ENABLED
))
3849 /* disable SERR in case the MSI write causes a master abort */
3850 pci_read_config_word(adapter
->pdev
, PCI_COMMAND
, &pci_cmd
);
3851 if (pci_cmd
& PCI_COMMAND_SERR
)
3852 pci_write_config_word(adapter
->pdev
, PCI_COMMAND
,
3853 pci_cmd
& ~PCI_COMMAND_SERR
);
3855 err
= e1000_test_msi_interrupt(adapter
);
3857 /* re-enable SERR */
3858 if (pci_cmd
& PCI_COMMAND_SERR
) {
3859 pci_read_config_word(adapter
->pdev
, PCI_COMMAND
, &pci_cmd
);
3860 pci_cmd
|= PCI_COMMAND_SERR
;
3861 pci_write_config_word(adapter
->pdev
, PCI_COMMAND
, pci_cmd
);
3868 * e1000_open - Called when a network interface is made active
3869 * @netdev: network interface device structure
3871 * Returns 0 on success, negative value on failure
3873 * The open entry point is called when a network interface is made
3874 * active by the system (IFF_UP). At this point all resources needed
3875 * for transmit and receive operations are allocated, the interrupt
3876 * handler is registered with the OS, the watchdog timer is started,
3877 * and the stack is notified that the interface is ready.
3879 static int e1000_open(struct net_device
*netdev
)
3881 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3882 struct e1000_hw
*hw
= &adapter
->hw
;
3883 struct pci_dev
*pdev
= adapter
->pdev
;
3886 /* disallow open during test */
3887 if (test_bit(__E1000_TESTING
, &adapter
->state
))
3890 pm_runtime_get_sync(&pdev
->dev
);
3892 netif_carrier_off(netdev
);
3894 /* allocate transmit descriptors */
3895 err
= e1000e_setup_tx_resources(adapter
->tx_ring
);
3899 /* allocate receive descriptors */
3900 err
= e1000e_setup_rx_resources(adapter
->rx_ring
);
3905 * If AMT is enabled, let the firmware know that the network
3906 * interface is now open and reset the part to a known state.
3908 if (adapter
->flags
& FLAG_HAS_AMT
) {
3909 e1000e_get_hw_control(adapter
);
3910 e1000e_reset(adapter
);
3913 e1000e_power_up_phy(adapter
);
3915 adapter
->mng_vlan_id
= E1000_MNG_VLAN_NONE
;
3916 if ((adapter
->hw
.mng_cookie
.status
&
3917 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
))
3918 e1000_update_mng_vlan(adapter
);
3920 /* DMA latency requirement to workaround jumbo issue */
3921 if (adapter
->hw
.mac
.type
== e1000_pch2lan
)
3922 pm_qos_add_request(&adapter
->netdev
->pm_qos_req
,
3923 PM_QOS_CPU_DMA_LATENCY
,
3924 PM_QOS_DEFAULT_VALUE
);
3927 * before we allocate an interrupt, we must be ready to handle it.
3928 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3929 * as soon as we call pci_request_irq, so we have to setup our
3930 * clean_rx handler before we do so.
3932 e1000_configure(adapter
);
3934 err
= e1000_request_irq(adapter
);
3939 * Work around PCIe errata with MSI interrupts causing some chipsets to
3940 * ignore e1000e MSI messages, which means we need to test our MSI
3943 if (adapter
->int_mode
!= E1000E_INT_MODE_LEGACY
) {
3944 err
= e1000_test_msi(adapter
);
3946 e_err("Interrupt allocation failed\n");
3951 /* From here on the code is the same as e1000e_up() */
3952 clear_bit(__E1000_DOWN
, &adapter
->state
);
3954 napi_enable(&adapter
->napi
);
3956 e1000_irq_enable(adapter
);
3958 adapter
->tx_hang_recheck
= false;
3959 netif_start_queue(netdev
);
3961 adapter
->idle_check
= true;
3962 pm_runtime_put(&pdev
->dev
);
3964 /* fire a link status change interrupt to start the watchdog */
3965 if (adapter
->msix_entries
)
3966 ew32(ICS
, E1000_ICS_LSC
| E1000_ICR_OTHER
);
3968 ew32(ICS
, E1000_ICS_LSC
);
3973 e1000e_release_hw_control(adapter
);
3974 e1000_power_down_phy(adapter
);
3975 e1000e_free_rx_resources(adapter
->rx_ring
);
3977 e1000e_free_tx_resources(adapter
->tx_ring
);
3979 e1000e_reset(adapter
);
3980 pm_runtime_put_sync(&pdev
->dev
);
3986 * e1000_close - Disables a network interface
3987 * @netdev: network interface device structure
3989 * Returns 0, this is not allowed to fail
3991 * The close entry point is called when an interface is de-activated
3992 * by the OS. The hardware is still under the drivers control, but
3993 * needs to be disabled. A global MAC reset is issued to stop the
3994 * hardware, and all transmit and receive resources are freed.
3996 static int e1000_close(struct net_device
*netdev
)
3998 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
3999 struct pci_dev
*pdev
= adapter
->pdev
;
4000 int count
= E1000_CHECK_RESET_COUNT
;
4002 while (test_bit(__E1000_RESETTING
, &adapter
->state
) && count
--)
4003 usleep_range(10000, 20000);
4005 WARN_ON(test_bit(__E1000_RESETTING
, &adapter
->state
));
4007 pm_runtime_get_sync(&pdev
->dev
);
4009 napi_disable(&adapter
->napi
);
4011 if (!test_bit(__E1000_DOWN
, &adapter
->state
)) {
4012 e1000e_down(adapter
);
4013 e1000_free_irq(adapter
);
4015 e1000_power_down_phy(adapter
);
4017 e1000e_free_tx_resources(adapter
->tx_ring
);
4018 e1000e_free_rx_resources(adapter
->rx_ring
);
4021 * kill manageability vlan ID if supported, but not if a vlan with
4022 * the same ID is registered on the host OS (let 8021q kill it)
4024 if (adapter
->hw
.mng_cookie
.status
&
4025 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
)
4026 e1000_vlan_rx_kill_vid(netdev
, adapter
->mng_vlan_id
);
4029 * If AMT is enabled, let the firmware know that the network
4030 * interface is now closed
4032 if ((adapter
->flags
& FLAG_HAS_AMT
) &&
4033 !test_bit(__E1000_TESTING
, &adapter
->state
))
4034 e1000e_release_hw_control(adapter
);
4036 if (adapter
->hw
.mac
.type
== e1000_pch2lan
)
4037 pm_qos_remove_request(&adapter
->netdev
->pm_qos_req
);
4039 pm_runtime_put_sync(&pdev
->dev
);
4044 * e1000_set_mac - Change the Ethernet Address of the NIC
4045 * @netdev: network interface device structure
4046 * @p: pointer to an address structure
4048 * Returns 0 on success, negative on failure
4050 static int e1000_set_mac(struct net_device
*netdev
, void *p
)
4052 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
4053 struct e1000_hw
*hw
= &adapter
->hw
;
4054 struct sockaddr
*addr
= p
;
4056 if (!is_valid_ether_addr(addr
->sa_data
))
4057 return -EADDRNOTAVAIL
;
4059 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
4060 memcpy(adapter
->hw
.mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
4062 hw
->mac
.ops
.rar_set(&adapter
->hw
, adapter
->hw
.mac
.addr
, 0);
4064 if (adapter
->flags
& FLAG_RESET_OVERWRITES_LAA
) {
4065 /* activate the work around */
4066 e1000e_set_laa_state_82571(&adapter
->hw
, 1);
4069 * Hold a copy of the LAA in RAR[14] This is done so that
4070 * between the time RAR[0] gets clobbered and the time it
4071 * gets fixed (in e1000_watchdog), the actual LAA is in one
4072 * of the RARs and no incoming packets directed to this port
4073 * are dropped. Eventually the LAA will be in RAR[0] and
4076 hw
->mac
.ops
.rar_set(&adapter
->hw
, adapter
->hw
.mac
.addr
,
4077 adapter
->hw
.mac
.rar_entry_count
- 1);
4084 * e1000e_update_phy_task - work thread to update phy
4085 * @work: pointer to our work struct
4087 * this worker thread exists because we must acquire a
4088 * semaphore to read the phy, which we could msleep while
4089 * waiting for it, and we can't msleep in a timer.
4091 static void e1000e_update_phy_task(struct work_struct
*work
)
4093 struct e1000_adapter
*adapter
= container_of(work
,
4094 struct e1000_adapter
, update_phy_task
);
4096 if (test_bit(__E1000_DOWN
, &adapter
->state
))
4099 e1000_get_phy_info(&adapter
->hw
);
4103 * Need to wait a few seconds after link up to get diagnostic information from
4106 static void e1000_update_phy_info(unsigned long data
)
4108 struct e1000_adapter
*adapter
= (struct e1000_adapter
*) data
;
4110 if (test_bit(__E1000_DOWN
, &adapter
->state
))
4113 schedule_work(&adapter
->update_phy_task
);
4117 * e1000e_update_phy_stats - Update the PHY statistics counters
4118 * @adapter: board private structure
4120 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4122 static void e1000e_update_phy_stats(struct e1000_adapter
*adapter
)
4124 struct e1000_hw
*hw
= &adapter
->hw
;
4128 ret_val
= hw
->phy
.ops
.acquire(hw
);
4133 * A page set is expensive so check if already on desired page.
4134 * If not, set to the page with the PHY status registers.
4137 ret_val
= e1000e_read_phy_reg_mdic(hw
, IGP01E1000_PHY_PAGE_SELECT
,
4141 if (phy_data
!= (HV_STATS_PAGE
<< IGP_PAGE_SHIFT
)) {
4142 ret_val
= hw
->phy
.ops
.set_page(hw
,
4143 HV_STATS_PAGE
<< IGP_PAGE_SHIFT
);
4148 /* Single Collision Count */
4149 hw
->phy
.ops
.read_reg_page(hw
, HV_SCC_UPPER
, &phy_data
);
4150 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_SCC_LOWER
, &phy_data
);
4152 adapter
->stats
.scc
+= phy_data
;
4154 /* Excessive Collision Count */
4155 hw
->phy
.ops
.read_reg_page(hw
, HV_ECOL_UPPER
, &phy_data
);
4156 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_ECOL_LOWER
, &phy_data
);
4158 adapter
->stats
.ecol
+= phy_data
;
4160 /* Multiple Collision Count */
4161 hw
->phy
.ops
.read_reg_page(hw
, HV_MCC_UPPER
, &phy_data
);
4162 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_MCC_LOWER
, &phy_data
);
4164 adapter
->stats
.mcc
+= phy_data
;
4166 /* Late Collision Count */
4167 hw
->phy
.ops
.read_reg_page(hw
, HV_LATECOL_UPPER
, &phy_data
);
4168 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_LATECOL_LOWER
, &phy_data
);
4170 adapter
->stats
.latecol
+= phy_data
;
4172 /* Collision Count - also used for adaptive IFS */
4173 hw
->phy
.ops
.read_reg_page(hw
, HV_COLC_UPPER
, &phy_data
);
4174 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_COLC_LOWER
, &phy_data
);
4176 hw
->mac
.collision_delta
= phy_data
;
4179 hw
->phy
.ops
.read_reg_page(hw
, HV_DC_UPPER
, &phy_data
);
4180 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_DC_LOWER
, &phy_data
);
4182 adapter
->stats
.dc
+= phy_data
;
4184 /* Transmit with no CRS */
4185 hw
->phy
.ops
.read_reg_page(hw
, HV_TNCRS_UPPER
, &phy_data
);
4186 ret_val
= hw
->phy
.ops
.read_reg_page(hw
, HV_TNCRS_LOWER
, &phy_data
);
4188 adapter
->stats
.tncrs
+= phy_data
;
4191 hw
->phy
.ops
.release(hw
);
4195 * e1000e_update_stats - Update the board statistics counters
4196 * @adapter: board private structure
4198 static void e1000e_update_stats(struct e1000_adapter
*adapter
)
4200 struct net_device
*netdev
= adapter
->netdev
;
4201 struct e1000_hw
*hw
= &adapter
->hw
;
4202 struct pci_dev
*pdev
= adapter
->pdev
;
4205 * Prevent stats update while adapter is being reset, or if the pci
4206 * connection is down.
4208 if (adapter
->link_speed
== 0)
4210 if (pci_channel_offline(pdev
))
4213 adapter
->stats
.crcerrs
+= er32(CRCERRS
);
4214 adapter
->stats
.gprc
+= er32(GPRC
);
4215 adapter
->stats
.gorc
+= er32(GORCL
);
4216 er32(GORCH
); /* Clear gorc */
4217 adapter
->stats
.bprc
+= er32(BPRC
);
4218 adapter
->stats
.mprc
+= er32(MPRC
);
4219 adapter
->stats
.roc
+= er32(ROC
);
4221 adapter
->stats
.mpc
+= er32(MPC
);
4223 /* Half-duplex statistics */
4224 if (adapter
->link_duplex
== HALF_DUPLEX
) {
4225 if (adapter
->flags2
& FLAG2_HAS_PHY_STATS
) {
4226 e1000e_update_phy_stats(adapter
);
4228 adapter
->stats
.scc
+= er32(SCC
);
4229 adapter
->stats
.ecol
+= er32(ECOL
);
4230 adapter
->stats
.mcc
+= er32(MCC
);
4231 adapter
->stats
.latecol
+= er32(LATECOL
);
4232 adapter
->stats
.dc
+= er32(DC
);
4234 hw
->mac
.collision_delta
= er32(COLC
);
4236 if ((hw
->mac
.type
!= e1000_82574
) &&
4237 (hw
->mac
.type
!= e1000_82583
))
4238 adapter
->stats
.tncrs
+= er32(TNCRS
);
4240 adapter
->stats
.colc
+= hw
->mac
.collision_delta
;
4243 adapter
->stats
.xonrxc
+= er32(XONRXC
);
4244 adapter
->stats
.xontxc
+= er32(XONTXC
);
4245 adapter
->stats
.xoffrxc
+= er32(XOFFRXC
);
4246 adapter
->stats
.xofftxc
+= er32(XOFFTXC
);
4247 adapter
->stats
.gptc
+= er32(GPTC
);
4248 adapter
->stats
.gotc
+= er32(GOTCL
);
4249 er32(GOTCH
); /* Clear gotc */
4250 adapter
->stats
.rnbc
+= er32(RNBC
);
4251 adapter
->stats
.ruc
+= er32(RUC
);
4253 adapter
->stats
.mptc
+= er32(MPTC
);
4254 adapter
->stats
.bptc
+= er32(BPTC
);
4256 /* used for adaptive IFS */
4258 hw
->mac
.tx_packet_delta
= er32(TPT
);
4259 adapter
->stats
.tpt
+= hw
->mac
.tx_packet_delta
;
4261 adapter
->stats
.algnerrc
+= er32(ALGNERRC
);
4262 adapter
->stats
.rxerrc
+= er32(RXERRC
);
4263 adapter
->stats
.cexterr
+= er32(CEXTERR
);
4264 adapter
->stats
.tsctc
+= er32(TSCTC
);
4265 adapter
->stats
.tsctfc
+= er32(TSCTFC
);
4267 /* Fill out the OS statistics structure */
4268 netdev
->stats
.multicast
= adapter
->stats
.mprc
;
4269 netdev
->stats
.collisions
= adapter
->stats
.colc
;
4274 * RLEC on some newer hardware can be incorrect so build
4275 * our own version based on RUC and ROC
4277 netdev
->stats
.rx_errors
= adapter
->stats
.rxerrc
+
4278 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
4279 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
4280 adapter
->stats
.cexterr
;
4281 netdev
->stats
.rx_length_errors
= adapter
->stats
.ruc
+
4283 netdev
->stats
.rx_crc_errors
= adapter
->stats
.crcerrs
;
4284 netdev
->stats
.rx_frame_errors
= adapter
->stats
.algnerrc
;
4285 netdev
->stats
.rx_missed_errors
= adapter
->stats
.mpc
;
4288 netdev
->stats
.tx_errors
= adapter
->stats
.ecol
+
4289 adapter
->stats
.latecol
;
4290 netdev
->stats
.tx_aborted_errors
= adapter
->stats
.ecol
;
4291 netdev
->stats
.tx_window_errors
= adapter
->stats
.latecol
;
4292 netdev
->stats
.tx_carrier_errors
= adapter
->stats
.tncrs
;
4294 /* Tx Dropped needs to be maintained elsewhere */
4296 /* Management Stats */
4297 adapter
->stats
.mgptc
+= er32(MGTPTC
);
4298 adapter
->stats
.mgprc
+= er32(MGTPRC
);
4299 adapter
->stats
.mgpdc
+= er32(MGTPDC
);
4303 * e1000_phy_read_status - Update the PHY register status snapshot
4304 * @adapter: board private structure
4306 static void e1000_phy_read_status(struct e1000_adapter
*adapter
)
4308 struct e1000_hw
*hw
= &adapter
->hw
;
4309 struct e1000_phy_regs
*phy
= &adapter
->phy_regs
;
4311 if ((er32(STATUS
) & E1000_STATUS_LU
) &&
4312 (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)) {
4315 ret_val
= e1e_rphy(hw
, PHY_CONTROL
, &phy
->bmcr
);
4316 ret_val
|= e1e_rphy(hw
, PHY_STATUS
, &phy
->bmsr
);
4317 ret_val
|= e1e_rphy(hw
, PHY_AUTONEG_ADV
, &phy
->advertise
);
4318 ret_val
|= e1e_rphy(hw
, PHY_LP_ABILITY
, &phy
->lpa
);
4319 ret_val
|= e1e_rphy(hw
, PHY_AUTONEG_EXP
, &phy
->expansion
);
4320 ret_val
|= e1e_rphy(hw
, PHY_1000T_CTRL
, &phy
->ctrl1000
);
4321 ret_val
|= e1e_rphy(hw
, PHY_1000T_STATUS
, &phy
->stat1000
);
4322 ret_val
|= e1e_rphy(hw
, PHY_EXT_STATUS
, &phy
->estatus
);
4324 e_warn("Error reading PHY register\n");
4327 * Do not read PHY registers if link is not up
4328 * Set values to typical power-on defaults
4330 phy
->bmcr
= (BMCR_SPEED1000
| BMCR_ANENABLE
| BMCR_FULLDPLX
);
4331 phy
->bmsr
= (BMSR_100FULL
| BMSR_100HALF
| BMSR_10FULL
|
4332 BMSR_10HALF
| BMSR_ESTATEN
| BMSR_ANEGCAPABLE
|
4334 phy
->advertise
= (ADVERTISE_PAUSE_ASYM
| ADVERTISE_PAUSE_CAP
|
4335 ADVERTISE_ALL
| ADVERTISE_CSMA
);
4337 phy
->expansion
= EXPANSION_ENABLENPAGE
;
4338 phy
->ctrl1000
= ADVERTISE_1000FULL
;
4340 phy
->estatus
= (ESTATUS_1000_TFULL
| ESTATUS_1000_THALF
);
4344 static void e1000_print_link_info(struct e1000_adapter
*adapter
)
4346 struct e1000_hw
*hw
= &adapter
->hw
;
4347 u32 ctrl
= er32(CTRL
);
4349 /* Link status message must follow this format for user tools */
4350 printk(KERN_INFO
"e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4351 adapter
->netdev
->name
,
4352 adapter
->link_speed
,
4353 adapter
->link_duplex
== FULL_DUPLEX
? "Full" : "Half",
4354 (ctrl
& E1000_CTRL_TFCE
) && (ctrl
& E1000_CTRL_RFCE
) ? "Rx/Tx" :
4355 (ctrl
& E1000_CTRL_RFCE
) ? "Rx" :
4356 (ctrl
& E1000_CTRL_TFCE
) ? "Tx" : "None");
4359 static bool e1000e_has_link(struct e1000_adapter
*adapter
)
4361 struct e1000_hw
*hw
= &adapter
->hw
;
4362 bool link_active
= false;
4366 * get_link_status is set on LSC (link status) interrupt or
4367 * Rx sequence error interrupt. get_link_status will stay
4368 * false until the check_for_link establishes link
4369 * for copper adapters ONLY
4371 switch (hw
->phy
.media_type
) {
4372 case e1000_media_type_copper
:
4373 if (hw
->mac
.get_link_status
) {
4374 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
4375 link_active
= !hw
->mac
.get_link_status
;
4380 case e1000_media_type_fiber
:
4381 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
4382 link_active
= !!(er32(STATUS
) & E1000_STATUS_LU
);
4384 case e1000_media_type_internal_serdes
:
4385 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
4386 link_active
= adapter
->hw
.mac
.serdes_has_link
;
4389 case e1000_media_type_unknown
:
4393 if ((ret_val
== E1000_ERR_PHY
) && (hw
->phy
.type
== e1000_phy_igp_3
) &&
4394 (er32(CTRL
) & E1000_PHY_CTRL_GBE_DISABLE
)) {
4395 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4396 e_info("Gigabit has been disabled, downgrading speed\n");
4402 static void e1000e_enable_receives(struct e1000_adapter
*adapter
)
4404 /* make sure the receive unit is started */
4405 if ((adapter
->flags
& FLAG_RX_NEEDS_RESTART
) &&
4406 (adapter
->flags
& FLAG_RX_RESTART_NOW
)) {
4407 struct e1000_hw
*hw
= &adapter
->hw
;
4408 u32 rctl
= er32(RCTL
);
4409 ew32(RCTL
, rctl
| E1000_RCTL_EN
);
4410 adapter
->flags
&= ~FLAG_RX_RESTART_NOW
;
4414 static void e1000e_check_82574_phy_workaround(struct e1000_adapter
*adapter
)
4416 struct e1000_hw
*hw
= &adapter
->hw
;
4419 * With 82574 controllers, PHY needs to be checked periodically
4420 * for hung state and reset, if two calls return true
4422 if (e1000_check_phy_82574(hw
))
4423 adapter
->phy_hang_count
++;
4425 adapter
->phy_hang_count
= 0;
4427 if (adapter
->phy_hang_count
> 1) {
4428 adapter
->phy_hang_count
= 0;
4429 schedule_work(&adapter
->reset_task
);
4434 * e1000_watchdog - Timer Call-back
4435 * @data: pointer to adapter cast into an unsigned long
4437 static void e1000_watchdog(unsigned long data
)
4439 struct e1000_adapter
*adapter
= (struct e1000_adapter
*) data
;
4441 /* Do the rest outside of interrupt context */
4442 schedule_work(&adapter
->watchdog_task
);
4444 /* TODO: make this use queue_delayed_work() */
4447 static void e1000_watchdog_task(struct work_struct
*work
)
4449 struct e1000_adapter
*adapter
= container_of(work
,
4450 struct e1000_adapter
, watchdog_task
);
4451 struct net_device
*netdev
= adapter
->netdev
;
4452 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
4453 struct e1000_phy_info
*phy
= &adapter
->hw
.phy
;
4454 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
4455 struct e1000_hw
*hw
= &adapter
->hw
;
4458 if (test_bit(__E1000_DOWN
, &adapter
->state
))
4461 link
= e1000e_has_link(adapter
);
4462 if ((netif_carrier_ok(netdev
)) && link
) {
4463 /* Cancel scheduled suspend requests. */
4464 pm_runtime_resume(netdev
->dev
.parent
);
4466 e1000e_enable_receives(adapter
);
4470 if ((e1000e_enable_tx_pkt_filtering(hw
)) &&
4471 (adapter
->mng_vlan_id
!= adapter
->hw
.mng_cookie
.vlan_id
))
4472 e1000_update_mng_vlan(adapter
);
4475 if (!netif_carrier_ok(netdev
)) {
4478 /* Cancel scheduled suspend requests. */
4479 pm_runtime_resume(netdev
->dev
.parent
);
4481 /* update snapshot of PHY registers on LSC */
4482 e1000_phy_read_status(adapter
);
4483 mac
->ops
.get_link_up_info(&adapter
->hw
,
4484 &adapter
->link_speed
,
4485 &adapter
->link_duplex
);
4486 e1000_print_link_info(adapter
);
4488 * On supported PHYs, check for duplex mismatch only
4489 * if link has autonegotiated at 10/100 half
4491 if ((hw
->phy
.type
== e1000_phy_igp_3
||
4492 hw
->phy
.type
== e1000_phy_bm
) &&
4493 (hw
->mac
.autoneg
== true) &&
4494 (adapter
->link_speed
== SPEED_10
||
4495 adapter
->link_speed
== SPEED_100
) &&
4496 (adapter
->link_duplex
== HALF_DUPLEX
)) {
4499 e1e_rphy(hw
, PHY_AUTONEG_EXP
, &autoneg_exp
);
4501 if (!(autoneg_exp
& NWAY_ER_LP_NWAY_CAPS
))
4502 e_info("Autonegotiated half duplex but link partner cannot autoneg. Try forcing full duplex if link gets many collisions.\n");
4505 /* adjust timeout factor according to speed/duplex */
4506 adapter
->tx_timeout_factor
= 1;
4507 switch (adapter
->link_speed
) {
4510 adapter
->tx_timeout_factor
= 16;
4514 adapter
->tx_timeout_factor
= 10;
4519 * workaround: re-program speed mode bit after
4522 if ((adapter
->flags
& FLAG_TARC_SPEED_MODE_BIT
) &&
4525 tarc0
= er32(TARC(0));
4526 tarc0
&= ~SPEED_MODE_BIT
;
4527 ew32(TARC(0), tarc0
);
4531 * disable TSO for pcie and 10/100 speeds, to avoid
4532 * some hardware issues
4534 if (!(adapter
->flags
& FLAG_TSO_FORCE
)) {
4535 switch (adapter
->link_speed
) {
4538 e_info("10/100 speed: disabling TSO\n");
4539 netdev
->features
&= ~NETIF_F_TSO
;
4540 netdev
->features
&= ~NETIF_F_TSO6
;
4543 netdev
->features
|= NETIF_F_TSO
;
4544 netdev
->features
|= NETIF_F_TSO6
;
4553 * enable transmits in the hardware, need to do this
4554 * after setting TARC(0)
4557 tctl
|= E1000_TCTL_EN
;
4561 * Perform any post-link-up configuration before
4562 * reporting link up.
4564 if (phy
->ops
.cfg_on_link_up
)
4565 phy
->ops
.cfg_on_link_up(hw
);
4567 netif_carrier_on(netdev
);
4569 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
4570 mod_timer(&adapter
->phy_info_timer
,
4571 round_jiffies(jiffies
+ 2 * HZ
));
4574 if (netif_carrier_ok(netdev
)) {
4575 adapter
->link_speed
= 0;
4576 adapter
->link_duplex
= 0;
4577 /* Link status message must follow this format */
4578 printk(KERN_INFO
"e1000e: %s NIC Link is Down\n",
4579 adapter
->netdev
->name
);
4580 netif_carrier_off(netdev
);
4581 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
4582 mod_timer(&adapter
->phy_info_timer
,
4583 round_jiffies(jiffies
+ 2 * HZ
));
4585 if (adapter
->flags
& FLAG_RX_NEEDS_RESTART
)
4586 schedule_work(&adapter
->reset_task
);
4588 pm_schedule_suspend(netdev
->dev
.parent
,
4594 spin_lock(&adapter
->stats64_lock
);
4595 e1000e_update_stats(adapter
);
4597 mac
->tx_packet_delta
= adapter
->stats
.tpt
- adapter
->tpt_old
;
4598 adapter
->tpt_old
= adapter
->stats
.tpt
;
4599 mac
->collision_delta
= adapter
->stats
.colc
- adapter
->colc_old
;
4600 adapter
->colc_old
= adapter
->stats
.colc
;
4602 adapter
->gorc
= adapter
->stats
.gorc
- adapter
->gorc_old
;
4603 adapter
->gorc_old
= adapter
->stats
.gorc
;
4604 adapter
->gotc
= adapter
->stats
.gotc
- adapter
->gotc_old
;
4605 adapter
->gotc_old
= adapter
->stats
.gotc
;
4606 spin_unlock(&adapter
->stats64_lock
);
4608 e1000e_update_adaptive(&adapter
->hw
);
4610 if (!netif_carrier_ok(netdev
) &&
4611 (e1000_desc_unused(tx_ring
) + 1 < tx_ring
->count
)) {
4613 * We've lost link, so the controller stops DMA,
4614 * but we've got queued Tx work that's never going
4615 * to get done, so reset controller to flush Tx.
4616 * (Do the reset outside of interrupt context).
4618 schedule_work(&adapter
->reset_task
);
4619 /* return immediately since reset is imminent */
4623 /* Simple mode for Interrupt Throttle Rate (ITR) */
4624 if (adapter
->itr_setting
== 4) {
4626 * Symmetric Tx/Rx gets a reduced ITR=2000;
4627 * Total asymmetrical Tx or Rx gets ITR=8000;
4628 * everyone else is between 2000-8000.
4630 u32 goc
= (adapter
->gotc
+ adapter
->gorc
) / 10000;
4631 u32 dif
= (adapter
->gotc
> adapter
->gorc
?
4632 adapter
->gotc
- adapter
->gorc
:
4633 adapter
->gorc
- adapter
->gotc
) / 10000;
4634 u32 itr
= goc
> 0 ? (dif
* 6000 / goc
+ 2000) : 8000;
4636 e1000e_write_itr(adapter
, itr
);
4639 /* Cause software interrupt to ensure Rx ring is cleaned */
4640 if (adapter
->msix_entries
)
4641 ew32(ICS
, adapter
->rx_ring
->ims_val
);
4643 ew32(ICS
, E1000_ICS_RXDMT0
);
4645 /* flush pending descriptors to memory before detecting Tx hang */
4646 e1000e_flush_descriptors(adapter
);
4648 /* Force detection of hung controller every watchdog period */
4649 adapter
->detect_tx_hung
= true;
4652 * With 82571 controllers, LAA may be overwritten due to controller
4653 * reset from the other port. Set the appropriate LAA in RAR[0]
4655 if (e1000e_get_laa_state_82571(hw
))
4656 hw
->mac
.ops
.rar_set(hw
, adapter
->hw
.mac
.addr
, 0);
4658 if (adapter
->flags2
& FLAG2_CHECK_PHY_HANG
)
4659 e1000e_check_82574_phy_workaround(adapter
);
4661 /* Reset the timer */
4662 if (!test_bit(__E1000_DOWN
, &adapter
->state
))
4663 mod_timer(&adapter
->watchdog_timer
,
4664 round_jiffies(jiffies
+ 2 * HZ
));
4667 #define E1000_TX_FLAGS_CSUM 0x00000001
4668 #define E1000_TX_FLAGS_VLAN 0x00000002
4669 #define E1000_TX_FLAGS_TSO 0x00000004
4670 #define E1000_TX_FLAGS_IPV4 0x00000008
4671 #define E1000_TX_FLAGS_NO_FCS 0x00000010
4672 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4673 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4675 static int e1000_tso(struct e1000_ring
*tx_ring
, struct sk_buff
*skb
)
4677 struct e1000_context_desc
*context_desc
;
4678 struct e1000_buffer
*buffer_info
;
4682 u8 ipcss
, ipcso
, tucss
, tucso
, hdr_len
;
4684 if (!skb_is_gso(skb
))
4687 if (skb_header_cloned(skb
)) {
4688 int err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4694 hdr_len
= skb_transport_offset(skb
) + tcp_hdrlen(skb
);
4695 mss
= skb_shinfo(skb
)->gso_size
;
4696 if (skb
->protocol
== htons(ETH_P_IP
)) {
4697 struct iphdr
*iph
= ip_hdr(skb
);
4700 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
4702 cmd_length
= E1000_TXD_CMD_IP
;
4703 ipcse
= skb_transport_offset(skb
) - 1;
4704 } else if (skb_is_gso_v6(skb
)) {
4705 ipv6_hdr(skb
)->payload_len
= 0;
4706 tcp_hdr(skb
)->check
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4707 &ipv6_hdr(skb
)->daddr
,
4711 ipcss
= skb_network_offset(skb
);
4712 ipcso
= (void *)&(ip_hdr(skb
)->check
) - (void *)skb
->data
;
4713 tucss
= skb_transport_offset(skb
);
4714 tucso
= (void *)&(tcp_hdr(skb
)->check
) - (void *)skb
->data
;
4716 cmd_length
|= (E1000_TXD_CMD_DEXT
| E1000_TXD_CMD_TSE
|
4717 E1000_TXD_CMD_TCP
| (skb
->len
- (hdr_len
)));
4719 i
= tx_ring
->next_to_use
;
4720 context_desc
= E1000_CONTEXT_DESC(*tx_ring
, i
);
4721 buffer_info
= &tx_ring
->buffer_info
[i
];
4723 context_desc
->lower_setup
.ip_fields
.ipcss
= ipcss
;
4724 context_desc
->lower_setup
.ip_fields
.ipcso
= ipcso
;
4725 context_desc
->lower_setup
.ip_fields
.ipcse
= cpu_to_le16(ipcse
);
4726 context_desc
->upper_setup
.tcp_fields
.tucss
= tucss
;
4727 context_desc
->upper_setup
.tcp_fields
.tucso
= tucso
;
4728 context_desc
->upper_setup
.tcp_fields
.tucse
= 0;
4729 context_desc
->tcp_seg_setup
.fields
.mss
= cpu_to_le16(mss
);
4730 context_desc
->tcp_seg_setup
.fields
.hdr_len
= hdr_len
;
4731 context_desc
->cmd_and_length
= cpu_to_le32(cmd_length
);
4733 buffer_info
->time_stamp
= jiffies
;
4734 buffer_info
->next_to_watch
= i
;
4737 if (i
== tx_ring
->count
)
4739 tx_ring
->next_to_use
= i
;
4744 static bool e1000_tx_csum(struct e1000_ring
*tx_ring
, struct sk_buff
*skb
)
4746 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
4747 struct e1000_context_desc
*context_desc
;
4748 struct e1000_buffer
*buffer_info
;
4751 u32 cmd_len
= E1000_TXD_CMD_DEXT
;
4754 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
4757 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
))
4758 protocol
= vlan_eth_hdr(skb
)->h_vlan_encapsulated_proto
;
4760 protocol
= skb
->protocol
;
4763 case cpu_to_be16(ETH_P_IP
):
4764 if (ip_hdr(skb
)->protocol
== IPPROTO_TCP
)
4765 cmd_len
|= E1000_TXD_CMD_TCP
;
4767 case cpu_to_be16(ETH_P_IPV6
):
4768 /* XXX not handling all IPV6 headers */
4769 if (ipv6_hdr(skb
)->nexthdr
== IPPROTO_TCP
)
4770 cmd_len
|= E1000_TXD_CMD_TCP
;
4773 if (unlikely(net_ratelimit()))
4774 e_warn("checksum_partial proto=%x!\n",
4775 be16_to_cpu(protocol
));
4779 css
= skb_checksum_start_offset(skb
);
4781 i
= tx_ring
->next_to_use
;
4782 buffer_info
= &tx_ring
->buffer_info
[i
];
4783 context_desc
= E1000_CONTEXT_DESC(*tx_ring
, i
);
4785 context_desc
->lower_setup
.ip_config
= 0;
4786 context_desc
->upper_setup
.tcp_fields
.tucss
= css
;
4787 context_desc
->upper_setup
.tcp_fields
.tucso
=
4788 css
+ skb
->csum_offset
;
4789 context_desc
->upper_setup
.tcp_fields
.tucse
= 0;
4790 context_desc
->tcp_seg_setup
.data
= 0;
4791 context_desc
->cmd_and_length
= cpu_to_le32(cmd_len
);
4793 buffer_info
->time_stamp
= jiffies
;
4794 buffer_info
->next_to_watch
= i
;
4797 if (i
== tx_ring
->count
)
4799 tx_ring
->next_to_use
= i
;
4804 static int e1000_tx_map(struct e1000_ring
*tx_ring
, struct sk_buff
*skb
,
4805 unsigned int first
, unsigned int max_per_txd
,
4806 unsigned int nr_frags
)
4808 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
4809 struct pci_dev
*pdev
= adapter
->pdev
;
4810 struct e1000_buffer
*buffer_info
;
4811 unsigned int len
= skb_headlen(skb
);
4812 unsigned int offset
= 0, size
, count
= 0, i
;
4813 unsigned int f
, bytecount
, segs
;
4815 i
= tx_ring
->next_to_use
;
4818 buffer_info
= &tx_ring
->buffer_info
[i
];
4819 size
= min(len
, max_per_txd
);
4821 buffer_info
->length
= size
;
4822 buffer_info
->time_stamp
= jiffies
;
4823 buffer_info
->next_to_watch
= i
;
4824 buffer_info
->dma
= dma_map_single(&pdev
->dev
,
4826 size
, DMA_TO_DEVICE
);
4827 buffer_info
->mapped_as_page
= false;
4828 if (dma_mapping_error(&pdev
->dev
, buffer_info
->dma
))
4837 if (i
== tx_ring
->count
)
4842 for (f
= 0; f
< nr_frags
; f
++) {
4843 const struct skb_frag_struct
*frag
;
4845 frag
= &skb_shinfo(skb
)->frags
[f
];
4846 len
= skb_frag_size(frag
);
4851 if (i
== tx_ring
->count
)
4854 buffer_info
= &tx_ring
->buffer_info
[i
];
4855 size
= min(len
, max_per_txd
);
4857 buffer_info
->length
= size
;
4858 buffer_info
->time_stamp
= jiffies
;
4859 buffer_info
->next_to_watch
= i
;
4860 buffer_info
->dma
= skb_frag_dma_map(&pdev
->dev
, frag
,
4861 offset
, size
, DMA_TO_DEVICE
);
4862 buffer_info
->mapped_as_page
= true;
4863 if (dma_mapping_error(&pdev
->dev
, buffer_info
->dma
))
4872 segs
= skb_shinfo(skb
)->gso_segs
? : 1;
4873 /* multiply data chunks by size of headers */
4874 bytecount
= ((segs
- 1) * skb_headlen(skb
)) + skb
->len
;
4876 tx_ring
->buffer_info
[i
].skb
= skb
;
4877 tx_ring
->buffer_info
[i
].segs
= segs
;
4878 tx_ring
->buffer_info
[i
].bytecount
= bytecount
;
4879 tx_ring
->buffer_info
[first
].next_to_watch
= i
;
4884 dev_err(&pdev
->dev
, "Tx DMA map failed\n");
4885 buffer_info
->dma
= 0;
4891 i
+= tx_ring
->count
;
4893 buffer_info
= &tx_ring
->buffer_info
[i
];
4894 e1000_put_txbuf(tx_ring
, buffer_info
);
4900 static void e1000_tx_queue(struct e1000_ring
*tx_ring
, int tx_flags
, int count
)
4902 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
4903 struct e1000_tx_desc
*tx_desc
= NULL
;
4904 struct e1000_buffer
*buffer_info
;
4905 u32 txd_upper
= 0, txd_lower
= E1000_TXD_CMD_IFCS
;
4908 if (tx_flags
& E1000_TX_FLAGS_TSO
) {
4909 txd_lower
|= E1000_TXD_CMD_DEXT
| E1000_TXD_DTYP_D
|
4911 txd_upper
|= E1000_TXD_POPTS_TXSM
<< 8;
4913 if (tx_flags
& E1000_TX_FLAGS_IPV4
)
4914 txd_upper
|= E1000_TXD_POPTS_IXSM
<< 8;
4917 if (tx_flags
& E1000_TX_FLAGS_CSUM
) {
4918 txd_lower
|= E1000_TXD_CMD_DEXT
| E1000_TXD_DTYP_D
;
4919 txd_upper
|= E1000_TXD_POPTS_TXSM
<< 8;
4922 if (tx_flags
& E1000_TX_FLAGS_VLAN
) {
4923 txd_lower
|= E1000_TXD_CMD_VLE
;
4924 txd_upper
|= (tx_flags
& E1000_TX_FLAGS_VLAN_MASK
);
4927 if (unlikely(tx_flags
& E1000_TX_FLAGS_NO_FCS
))
4928 txd_lower
&= ~(E1000_TXD_CMD_IFCS
);
4930 i
= tx_ring
->next_to_use
;
4933 buffer_info
= &tx_ring
->buffer_info
[i
];
4934 tx_desc
= E1000_TX_DESC(*tx_ring
, i
);
4935 tx_desc
->buffer_addr
= cpu_to_le64(buffer_info
->dma
);
4936 tx_desc
->lower
.data
=
4937 cpu_to_le32(txd_lower
| buffer_info
->length
);
4938 tx_desc
->upper
.data
= cpu_to_le32(txd_upper
);
4941 if (i
== tx_ring
->count
)
4943 } while (--count
> 0);
4945 tx_desc
->lower
.data
|= cpu_to_le32(adapter
->txd_cmd
);
4947 /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4948 if (unlikely(tx_flags
& E1000_TX_FLAGS_NO_FCS
))
4949 tx_desc
->lower
.data
&= ~(cpu_to_le32(E1000_TXD_CMD_IFCS
));
4952 * Force memory writes to complete before letting h/w
4953 * know there are new descriptors to fetch. (Only
4954 * applicable for weak-ordered memory model archs,
4959 tx_ring
->next_to_use
= i
;
4961 if (adapter
->flags2
& FLAG2_PCIM2PCI_ARBITER_WA
)
4962 e1000e_update_tdt_wa(tx_ring
, i
);
4964 writel(i
, tx_ring
->tail
);
4967 * we need this if more than one processor can write to our tail
4968 * at a time, it synchronizes IO on IA64/Altix systems
4973 #define MINIMUM_DHCP_PACKET_SIZE 282
4974 static int e1000_transfer_dhcp_info(struct e1000_adapter
*adapter
,
4975 struct sk_buff
*skb
)
4977 struct e1000_hw
*hw
= &adapter
->hw
;
4980 if (vlan_tx_tag_present(skb
)) {
4981 if (!((vlan_tx_tag_get(skb
) == adapter
->hw
.mng_cookie
.vlan_id
) &&
4982 (adapter
->hw
.mng_cookie
.status
&
4983 E1000_MNG_DHCP_COOKIE_STATUS_VLAN
)))
4987 if (skb
->len
<= MINIMUM_DHCP_PACKET_SIZE
)
4990 if (((struct ethhdr
*) skb
->data
)->h_proto
!= htons(ETH_P_IP
))
4994 const struct iphdr
*ip
= (struct iphdr
*)((u8
*)skb
->data
+14);
4997 if (ip
->protocol
!= IPPROTO_UDP
)
5000 udp
= (struct udphdr
*)((u8
*)ip
+ (ip
->ihl
<< 2));
5001 if (ntohs(udp
->dest
) != 67)
5004 offset
= (u8
*)udp
+ 8 - skb
->data
;
5005 length
= skb
->len
- offset
;
5006 return e1000e_mng_write_dhcp_info(hw
, (u8
*)udp
+ 8, length
);
5012 static int __e1000_maybe_stop_tx(struct e1000_ring
*tx_ring
, int size
)
5014 struct e1000_adapter
*adapter
= tx_ring
->adapter
;
5016 netif_stop_queue(adapter
->netdev
);
5018 * Herbert's original patch had:
5019 * smp_mb__after_netif_stop_queue();
5020 * but since that doesn't exist yet, just open code it.
5025 * We need to check again in a case another CPU has just
5026 * made room available.
5028 if (e1000_desc_unused(tx_ring
) < size
)
5032 netif_start_queue(adapter
->netdev
);
5033 ++adapter
->restart_queue
;
5037 static int e1000_maybe_stop_tx(struct e1000_ring
*tx_ring
, int size
)
5039 BUG_ON(size
> tx_ring
->count
);
5041 if (e1000_desc_unused(tx_ring
) >= size
)
5043 return __e1000_maybe_stop_tx(tx_ring
, size
);
5046 static netdev_tx_t
e1000_xmit_frame(struct sk_buff
*skb
,
5047 struct net_device
*netdev
)
5049 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5050 struct e1000_ring
*tx_ring
= adapter
->tx_ring
;
5052 unsigned int tx_flags
= 0;
5053 unsigned int len
= skb_headlen(skb
);
5054 unsigned int nr_frags
;
5060 if (test_bit(__E1000_DOWN
, &adapter
->state
)) {
5061 dev_kfree_skb_any(skb
);
5062 return NETDEV_TX_OK
;
5065 if (skb
->len
<= 0) {
5066 dev_kfree_skb_any(skb
);
5067 return NETDEV_TX_OK
;
5070 mss
= skb_shinfo(skb
)->gso_size
;
5075 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5076 * points to just header, pull a few bytes of payload from
5077 * frags into skb->data
5079 hdr_len
= skb_transport_offset(skb
) + tcp_hdrlen(skb
);
5081 * we do this workaround for ES2LAN, but it is un-necessary,
5082 * avoiding it could save a lot of cycles
5084 if (skb
->data_len
&& (hdr_len
== len
)) {
5085 unsigned int pull_size
;
5087 pull_size
= min_t(unsigned int, 4, skb
->data_len
);
5088 if (!__pskb_pull_tail(skb
, pull_size
)) {
5089 e_err("__pskb_pull_tail failed.\n");
5090 dev_kfree_skb_any(skb
);
5091 return NETDEV_TX_OK
;
5093 len
= skb_headlen(skb
);
5097 /* reserve a descriptor for the offload context */
5098 if ((mss
) || (skb
->ip_summed
== CHECKSUM_PARTIAL
))
5102 count
+= DIV_ROUND_UP(len
, adapter
->tx_fifo_limit
);
5104 nr_frags
= skb_shinfo(skb
)->nr_frags
;
5105 for (f
= 0; f
< nr_frags
; f
++)
5106 count
+= DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb
)->frags
[f
]),
5107 adapter
->tx_fifo_limit
);
5109 if (adapter
->hw
.mac
.tx_pkt_filtering
)
5110 e1000_transfer_dhcp_info(adapter
, skb
);
5113 * need: count + 2 desc gap to keep tail from touching
5114 * head, otherwise try next time
5116 if (e1000_maybe_stop_tx(tx_ring
, count
+ 2))
5117 return NETDEV_TX_BUSY
;
5119 if (vlan_tx_tag_present(skb
)) {
5120 tx_flags
|= E1000_TX_FLAGS_VLAN
;
5121 tx_flags
|= (vlan_tx_tag_get(skb
) << E1000_TX_FLAGS_VLAN_SHIFT
);
5124 first
= tx_ring
->next_to_use
;
5126 tso
= e1000_tso(tx_ring
, skb
);
5128 dev_kfree_skb_any(skb
);
5129 return NETDEV_TX_OK
;
5133 tx_flags
|= E1000_TX_FLAGS_TSO
;
5134 else if (e1000_tx_csum(tx_ring
, skb
))
5135 tx_flags
|= E1000_TX_FLAGS_CSUM
;
5138 * Old method was to assume IPv4 packet by default if TSO was enabled.
5139 * 82571 hardware supports TSO capabilities for IPv6 as well...
5140 * no longer assume, we must.
5142 if (skb
->protocol
== htons(ETH_P_IP
))
5143 tx_flags
|= E1000_TX_FLAGS_IPV4
;
5145 if (unlikely(skb
->no_fcs
))
5146 tx_flags
|= E1000_TX_FLAGS_NO_FCS
;
5148 /* if count is 0 then mapping error has occurred */
5149 count
= e1000_tx_map(tx_ring
, skb
, first
, adapter
->tx_fifo_limit
,
5152 skb_tx_timestamp(skb
);
5154 netdev_sent_queue(netdev
, skb
->len
);
5155 e1000_tx_queue(tx_ring
, tx_flags
, count
);
5156 /* Make sure there is space in the ring for the next send. */
5157 e1000_maybe_stop_tx(tx_ring
,
5159 DIV_ROUND_UP(PAGE_SIZE
,
5160 adapter
->tx_fifo_limit
) + 2));
5162 dev_kfree_skb_any(skb
);
5163 tx_ring
->buffer_info
[first
].time_stamp
= 0;
5164 tx_ring
->next_to_use
= first
;
5167 return NETDEV_TX_OK
;
5171 * e1000_tx_timeout - Respond to a Tx Hang
5172 * @netdev: network interface device structure
5174 static void e1000_tx_timeout(struct net_device
*netdev
)
5176 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5178 /* Do the reset outside of interrupt context */
5179 adapter
->tx_timeout_count
++;
5180 schedule_work(&adapter
->reset_task
);
5183 static void e1000_reset_task(struct work_struct
*work
)
5185 struct e1000_adapter
*adapter
;
5186 adapter
= container_of(work
, struct e1000_adapter
, reset_task
);
5188 /* don't run the task if already down */
5189 if (test_bit(__E1000_DOWN
, &adapter
->state
))
5192 if (!((adapter
->flags
& FLAG_RX_NEEDS_RESTART
) &&
5193 (adapter
->flags
& FLAG_RX_RESTART_NOW
))) {
5194 e1000e_dump(adapter
);
5195 e_err("Reset adapter\n");
5197 e1000e_reinit_locked(adapter
);
5201 * e1000_get_stats64 - Get System Network Statistics
5202 * @netdev: network interface device structure
5203 * @stats: rtnl_link_stats64 pointer
5205 * Returns the address of the device statistics structure.
5207 struct rtnl_link_stats64
*e1000e_get_stats64(struct net_device
*netdev
,
5208 struct rtnl_link_stats64
*stats
)
5210 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5212 memset(stats
, 0, sizeof(struct rtnl_link_stats64
));
5213 spin_lock(&adapter
->stats64_lock
);
5214 e1000e_update_stats(adapter
);
5215 /* Fill out the OS statistics structure */
5216 stats
->rx_bytes
= adapter
->stats
.gorc
;
5217 stats
->rx_packets
= adapter
->stats
.gprc
;
5218 stats
->tx_bytes
= adapter
->stats
.gotc
;
5219 stats
->tx_packets
= adapter
->stats
.gptc
;
5220 stats
->multicast
= adapter
->stats
.mprc
;
5221 stats
->collisions
= adapter
->stats
.colc
;
5226 * RLEC on some newer hardware can be incorrect so build
5227 * our own version based on RUC and ROC
5229 stats
->rx_errors
= adapter
->stats
.rxerrc
+
5230 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
5231 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
5232 adapter
->stats
.cexterr
;
5233 stats
->rx_length_errors
= adapter
->stats
.ruc
+
5235 stats
->rx_crc_errors
= adapter
->stats
.crcerrs
;
5236 stats
->rx_frame_errors
= adapter
->stats
.algnerrc
;
5237 stats
->rx_missed_errors
= adapter
->stats
.mpc
;
5240 stats
->tx_errors
= adapter
->stats
.ecol
+
5241 adapter
->stats
.latecol
;
5242 stats
->tx_aborted_errors
= adapter
->stats
.ecol
;
5243 stats
->tx_window_errors
= adapter
->stats
.latecol
;
5244 stats
->tx_carrier_errors
= adapter
->stats
.tncrs
;
5246 /* Tx Dropped needs to be maintained elsewhere */
5248 spin_unlock(&adapter
->stats64_lock
);
5253 * e1000_change_mtu - Change the Maximum Transfer Unit
5254 * @netdev: network interface device structure
5255 * @new_mtu: new value for maximum frame size
5257 * Returns 0 on success, negative on failure
5259 static int e1000_change_mtu(struct net_device
*netdev
, int new_mtu
)
5261 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5262 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
5264 /* Jumbo frame support */
5265 if ((max_frame
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) &&
5266 !(adapter
->flags
& FLAG_HAS_JUMBO_FRAMES
)) {
5267 e_err("Jumbo Frames not supported.\n");
5271 /* Supported frame sizes */
5272 if ((new_mtu
< ETH_ZLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
) ||
5273 (max_frame
> adapter
->max_hw_frame_size
)) {
5274 e_err("Unsupported MTU setting\n");
5278 /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5279 if ((adapter
->hw
.mac
.type
>= e1000_pch2lan
) &&
5280 !(adapter
->flags2
& FLAG2_CRC_STRIPPING
) &&
5281 (new_mtu
> ETH_DATA_LEN
)) {
5282 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5286 while (test_and_set_bit(__E1000_RESETTING
, &adapter
->state
))
5287 usleep_range(1000, 2000);
5288 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5289 adapter
->max_frame_size
= max_frame
;
5290 e_info("changing MTU from %d to %d\n", netdev
->mtu
, new_mtu
);
5291 netdev
->mtu
= new_mtu
;
5292 if (netif_running(netdev
))
5293 e1000e_down(adapter
);
5296 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5297 * means we reserve 2 more, this pushes us to allocate from the next
5299 * i.e. RXBUFFER_2048 --> size-4096 slab
5300 * However with the new *_jumbo_rx* routines, jumbo receives will use
5304 if (max_frame
<= 2048)
5305 adapter
->rx_buffer_len
= 2048;
5307 adapter
->rx_buffer_len
= 4096;
5309 /* adjust allocation if LPE protects us, and we aren't using SBP */
5310 if ((max_frame
== ETH_FRAME_LEN
+ ETH_FCS_LEN
) ||
5311 (max_frame
== ETH_FRAME_LEN
+ VLAN_HLEN
+ ETH_FCS_LEN
))
5312 adapter
->rx_buffer_len
= ETH_FRAME_LEN
+ VLAN_HLEN
5315 if (netif_running(netdev
))
5318 e1000e_reset(adapter
);
5320 clear_bit(__E1000_RESETTING
, &adapter
->state
);
5325 static int e1000_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
,
5328 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5329 struct mii_ioctl_data
*data
= if_mii(ifr
);
5331 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
5336 data
->phy_id
= adapter
->hw
.phy
.addr
;
5339 e1000_phy_read_status(adapter
);
5341 switch (data
->reg_num
& 0x1F) {
5343 data
->val_out
= adapter
->phy_regs
.bmcr
;
5346 data
->val_out
= adapter
->phy_regs
.bmsr
;
5349 data
->val_out
= (adapter
->hw
.phy
.id
>> 16);
5352 data
->val_out
= (adapter
->hw
.phy
.id
& 0xFFFF);
5355 data
->val_out
= adapter
->phy_regs
.advertise
;
5358 data
->val_out
= adapter
->phy_regs
.lpa
;
5361 data
->val_out
= adapter
->phy_regs
.expansion
;
5364 data
->val_out
= adapter
->phy_regs
.ctrl1000
;
5367 data
->val_out
= adapter
->phy_regs
.stat1000
;
5370 data
->val_out
= adapter
->phy_regs
.estatus
;
5383 static int e1000_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
5389 return e1000_mii_ioctl(netdev
, ifr
, cmd
);
5395 static int e1000_init_phy_wakeup(struct e1000_adapter
*adapter
, u32 wufc
)
5397 struct e1000_hw
*hw
= &adapter
->hw
;
5399 u16 phy_reg
, wuc_enable
;
5402 /* copy MAC RARs to PHY RARs */
5403 e1000_copy_rx_addrs_to_phy_ich8lan(hw
);
5405 retval
= hw
->phy
.ops
.acquire(hw
);
5407 e_err("Could not acquire PHY\n");
5411 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5412 retval
= e1000_enable_phy_wakeup_reg_access_bm(hw
, &wuc_enable
);
5416 /* copy MAC MTA to PHY MTA - only needed for pchlan */
5417 for (i
= 0; i
< adapter
->hw
.mac
.mta_reg_count
; i
++) {
5418 mac_reg
= E1000_READ_REG_ARRAY(hw
, E1000_MTA
, i
);
5419 hw
->phy
.ops
.write_reg_page(hw
, BM_MTA(i
),
5420 (u16
)(mac_reg
& 0xFFFF));
5421 hw
->phy
.ops
.write_reg_page(hw
, BM_MTA(i
) + 1,
5422 (u16
)((mac_reg
>> 16) & 0xFFFF));
5425 /* configure PHY Rx Control register */
5426 hw
->phy
.ops
.read_reg_page(&adapter
->hw
, BM_RCTL
, &phy_reg
);
5427 mac_reg
= er32(RCTL
);
5428 if (mac_reg
& E1000_RCTL_UPE
)
5429 phy_reg
|= BM_RCTL_UPE
;
5430 if (mac_reg
& E1000_RCTL_MPE
)
5431 phy_reg
|= BM_RCTL_MPE
;
5432 phy_reg
&= ~(BM_RCTL_MO_MASK
);
5433 if (mac_reg
& E1000_RCTL_MO_3
)
5434 phy_reg
|= (((mac_reg
& E1000_RCTL_MO_3
) >> E1000_RCTL_MO_SHIFT
)
5435 << BM_RCTL_MO_SHIFT
);
5436 if (mac_reg
& E1000_RCTL_BAM
)
5437 phy_reg
|= BM_RCTL_BAM
;
5438 if (mac_reg
& E1000_RCTL_PMCF
)
5439 phy_reg
|= BM_RCTL_PMCF
;
5440 mac_reg
= er32(CTRL
);
5441 if (mac_reg
& E1000_CTRL_RFCE
)
5442 phy_reg
|= BM_RCTL_RFCE
;
5443 hw
->phy
.ops
.write_reg_page(&adapter
->hw
, BM_RCTL
, phy_reg
);
5445 /* enable PHY wakeup in MAC register */
5447 ew32(WUC
, E1000_WUC_PHY_WAKE
| E1000_WUC_PME_EN
);
5449 /* configure and enable PHY wakeup in PHY registers */
5450 hw
->phy
.ops
.write_reg_page(&adapter
->hw
, BM_WUFC
, wufc
);
5451 hw
->phy
.ops
.write_reg_page(&adapter
->hw
, BM_WUC
, E1000_WUC_PME_EN
);
5453 /* activate PHY wakeup */
5454 wuc_enable
|= BM_WUC_ENABLE_BIT
| BM_WUC_HOST_WU_BIT
;
5455 retval
= e1000_disable_phy_wakeup_reg_access_bm(hw
, &wuc_enable
);
5457 e_err("Could not set PHY Host Wakeup bit\n");
5459 hw
->phy
.ops
.release(hw
);
5464 static int __e1000_shutdown(struct pci_dev
*pdev
, bool *enable_wake
,
5467 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5468 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5469 struct e1000_hw
*hw
= &adapter
->hw
;
5470 u32 ctrl
, ctrl_ext
, rctl
, status
;
5471 /* Runtime suspend should only enable wakeup for link changes */
5472 u32 wufc
= runtime
? E1000_WUFC_LNKC
: adapter
->wol
;
5475 netif_device_detach(netdev
);
5477 if (netif_running(netdev
)) {
5478 int count
= E1000_CHECK_RESET_COUNT
;
5480 while (test_bit(__E1000_RESETTING
, &adapter
->state
) && count
--)
5481 usleep_range(10000, 20000);
5483 WARN_ON(test_bit(__E1000_RESETTING
, &adapter
->state
));
5484 e1000e_down(adapter
);
5485 e1000_free_irq(adapter
);
5487 e1000e_reset_interrupt_capability(adapter
);
5489 retval
= pci_save_state(pdev
);
5493 status
= er32(STATUS
);
5494 if (status
& E1000_STATUS_LU
)
5495 wufc
&= ~E1000_WUFC_LNKC
;
5498 e1000_setup_rctl(adapter
);
5499 e1000e_set_rx_mode(netdev
);
5501 /* turn on all-multi mode if wake on multicast is enabled */
5502 if (wufc
& E1000_WUFC_MC
) {
5504 rctl
|= E1000_RCTL_MPE
;
5509 /* advertise wake from D3Cold */
5510 #define E1000_CTRL_ADVD3WUC 0x00100000
5511 /* phy power management enable */
5512 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5513 ctrl
|= E1000_CTRL_ADVD3WUC
;
5514 if (!(adapter
->flags2
& FLAG2_HAS_PHY_WAKEUP
))
5515 ctrl
|= E1000_CTRL_EN_PHY_PWR_MGMT
;
5518 if (adapter
->hw
.phy
.media_type
== e1000_media_type_fiber
||
5519 adapter
->hw
.phy
.media_type
==
5520 e1000_media_type_internal_serdes
) {
5521 /* keep the laser running in D3 */
5522 ctrl_ext
= er32(CTRL_EXT
);
5523 ctrl_ext
|= E1000_CTRL_EXT_SDP3_DATA
;
5524 ew32(CTRL_EXT
, ctrl_ext
);
5527 if (adapter
->flags
& FLAG_IS_ICH
)
5528 e1000_suspend_workarounds_ich8lan(&adapter
->hw
);
5530 /* Allow time for pending master requests to run */
5531 e1000e_disable_pcie_master(&adapter
->hw
);
5533 if (adapter
->flags2
& FLAG2_HAS_PHY_WAKEUP
) {
5534 /* enable wakeup by the PHY */
5535 retval
= e1000_init_phy_wakeup(adapter
, wufc
);
5539 /* enable wakeup by the MAC */
5541 ew32(WUC
, E1000_WUC_PME_EN
);
5548 *enable_wake
= !!wufc
;
5550 /* make sure adapter isn't asleep if manageability is enabled */
5551 if ((adapter
->flags
& FLAG_MNG_PT_ENABLED
) ||
5552 (hw
->mac
.ops
.check_mng_mode(hw
)))
5553 *enable_wake
= true;
5555 if (adapter
->hw
.phy
.type
== e1000_phy_igp_3
)
5556 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter
->hw
);
5559 * Release control of h/w to f/w. If f/w is AMT enabled, this
5560 * would have already happened in close and is redundant.
5562 e1000e_release_hw_control(adapter
);
5564 pci_disable_device(pdev
);
5569 static void e1000_power_off(struct pci_dev
*pdev
, bool sleep
, bool wake
)
5571 if (sleep
&& wake
) {
5572 pci_prepare_to_sleep(pdev
);
5576 pci_wake_from_d3(pdev
, wake
);
5577 pci_set_power_state(pdev
, PCI_D3hot
);
5580 static void e1000_complete_shutdown(struct pci_dev
*pdev
, bool sleep
,
5583 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5584 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5587 * The pci-e switch on some quad port adapters will report a
5588 * correctable error when the MAC transitions from D0 to D3. To
5589 * prevent this we need to mask off the correctable errors on the
5590 * downstream port of the pci-e switch.
5592 if (adapter
->flags
& FLAG_IS_QUAD_PORT
) {
5593 struct pci_dev
*us_dev
= pdev
->bus
->self
;
5596 pcie_capability_read_word(us_dev
, PCI_EXP_DEVCTL
, &devctl
);
5597 pcie_capability_write_word(us_dev
, PCI_EXP_DEVCTL
,
5598 (devctl
& ~PCI_EXP_DEVCTL_CERE
));
5600 e1000_power_off(pdev
, sleep
, wake
);
5602 pcie_capability_write_word(us_dev
, PCI_EXP_DEVCTL
, devctl
);
5604 e1000_power_off(pdev
, sleep
, wake
);
5608 #ifdef CONFIG_PCIEASPM
5609 static void __e1000e_disable_aspm(struct pci_dev
*pdev
, u16 state
)
5611 pci_disable_link_state_locked(pdev
, state
);
5614 static void __e1000e_disable_aspm(struct pci_dev
*pdev
, u16 state
)
5617 * Both device and parent should have the same ASPM setting.
5618 * Disable ASPM in downstream component first and then upstream.
5620 pcie_capability_clear_word(pdev
, PCI_EXP_LNKCTL
, state
);
5622 if (pdev
->bus
->self
)
5623 pcie_capability_clear_word(pdev
->bus
->self
, PCI_EXP_LNKCTL
,
5627 static void e1000e_disable_aspm(struct pci_dev
*pdev
, u16 state
)
5629 dev_info(&pdev
->dev
, "Disabling ASPM %s %s\n",
5630 (state
& PCIE_LINK_STATE_L0S
) ? "L0s" : "",
5631 (state
& PCIE_LINK_STATE_L1
) ? "L1" : "");
5633 __e1000e_disable_aspm(pdev
, state
);
5637 static bool e1000e_pm_ready(struct e1000_adapter
*adapter
)
5639 return !!adapter
->tx_ring
->buffer_info
;
5642 static int __e1000_resume(struct pci_dev
*pdev
)
5644 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5645 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5646 struct e1000_hw
*hw
= &adapter
->hw
;
5647 u16 aspm_disable_flag
= 0;
5650 if (adapter
->flags2
& FLAG2_DISABLE_ASPM_L0S
)
5651 aspm_disable_flag
= PCIE_LINK_STATE_L0S
;
5652 if (adapter
->flags2
& FLAG2_DISABLE_ASPM_L1
)
5653 aspm_disable_flag
|= PCIE_LINK_STATE_L1
;
5654 if (aspm_disable_flag
)
5655 e1000e_disable_aspm(pdev
, aspm_disable_flag
);
5657 pci_set_power_state(pdev
, PCI_D0
);
5658 pci_restore_state(pdev
);
5659 pci_save_state(pdev
);
5661 e1000e_set_interrupt_capability(adapter
);
5662 if (netif_running(netdev
)) {
5663 err
= e1000_request_irq(adapter
);
5668 if (hw
->mac
.type
>= e1000_pch2lan
)
5669 e1000_resume_workarounds_pchlan(&adapter
->hw
);
5671 e1000e_power_up_phy(adapter
);
5673 /* report the system wakeup cause from S3/S4 */
5674 if (adapter
->flags2
& FLAG2_HAS_PHY_WAKEUP
) {
5677 e1e_rphy(&adapter
->hw
, BM_WUS
, &phy_data
);
5679 e_info("PHY Wakeup cause - %s\n",
5680 phy_data
& E1000_WUS_EX
? "Unicast Packet" :
5681 phy_data
& E1000_WUS_MC
? "Multicast Packet" :
5682 phy_data
& E1000_WUS_BC
? "Broadcast Packet" :
5683 phy_data
& E1000_WUS_MAG
? "Magic Packet" :
5684 phy_data
& E1000_WUS_LNKC
?
5685 "Link Status Change" : "other");
5687 e1e_wphy(&adapter
->hw
, BM_WUS
, ~0);
5689 u32 wus
= er32(WUS
);
5691 e_info("MAC Wakeup cause - %s\n",
5692 wus
& E1000_WUS_EX
? "Unicast Packet" :
5693 wus
& E1000_WUS_MC
? "Multicast Packet" :
5694 wus
& E1000_WUS_BC
? "Broadcast Packet" :
5695 wus
& E1000_WUS_MAG
? "Magic Packet" :
5696 wus
& E1000_WUS_LNKC
? "Link Status Change" :
5702 e1000e_reset(adapter
);
5704 e1000_init_manageability_pt(adapter
);
5706 if (netif_running(netdev
))
5709 netif_device_attach(netdev
);
5712 * If the controller has AMT, do not set DRV_LOAD until the interface
5713 * is up. For all other cases, let the f/w know that the h/w is now
5714 * under the control of the driver.
5716 if (!(adapter
->flags
& FLAG_HAS_AMT
))
5717 e1000e_get_hw_control(adapter
);
5722 #ifdef CONFIG_PM_SLEEP
5723 static int e1000_suspend(struct device
*dev
)
5725 struct pci_dev
*pdev
= to_pci_dev(dev
);
5729 retval
= __e1000_shutdown(pdev
, &wake
, false);
5731 e1000_complete_shutdown(pdev
, true, wake
);
5736 static int e1000_resume(struct device
*dev
)
5738 struct pci_dev
*pdev
= to_pci_dev(dev
);
5739 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5740 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5742 if (e1000e_pm_ready(adapter
))
5743 adapter
->idle_check
= true;
5745 return __e1000_resume(pdev
);
5747 #endif /* CONFIG_PM_SLEEP */
5749 #ifdef CONFIG_PM_RUNTIME
5750 static int e1000_runtime_suspend(struct device
*dev
)
5752 struct pci_dev
*pdev
= to_pci_dev(dev
);
5753 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5754 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5756 if (e1000e_pm_ready(adapter
)) {
5759 __e1000_shutdown(pdev
, &wake
, true);
5765 static int e1000_idle(struct device
*dev
)
5767 struct pci_dev
*pdev
= to_pci_dev(dev
);
5768 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5769 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5771 if (!e1000e_pm_ready(adapter
))
5774 if (adapter
->idle_check
) {
5775 adapter
->idle_check
= false;
5776 if (!e1000e_has_link(adapter
))
5777 pm_schedule_suspend(dev
, MSEC_PER_SEC
);
5783 static int e1000_runtime_resume(struct device
*dev
)
5785 struct pci_dev
*pdev
= to_pci_dev(dev
);
5786 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5787 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5789 if (!e1000e_pm_ready(adapter
))
5792 adapter
->idle_check
= !dev
->power
.runtime_auto
;
5793 return __e1000_resume(pdev
);
5795 #endif /* CONFIG_PM_RUNTIME */
5796 #endif /* CONFIG_PM */
5798 static void e1000_shutdown(struct pci_dev
*pdev
)
5802 __e1000_shutdown(pdev
, &wake
, false);
5804 if (system_state
== SYSTEM_POWER_OFF
)
5805 e1000_complete_shutdown(pdev
, false, wake
);
5808 #ifdef CONFIG_NET_POLL_CONTROLLER
5810 static irqreturn_t
e1000_intr_msix(int irq
, void *data
)
5812 struct net_device
*netdev
= data
;
5813 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5815 if (adapter
->msix_entries
) {
5816 int vector
, msix_irq
;
5819 msix_irq
= adapter
->msix_entries
[vector
].vector
;
5820 disable_irq(msix_irq
);
5821 e1000_intr_msix_rx(msix_irq
, netdev
);
5822 enable_irq(msix_irq
);
5825 msix_irq
= adapter
->msix_entries
[vector
].vector
;
5826 disable_irq(msix_irq
);
5827 e1000_intr_msix_tx(msix_irq
, netdev
);
5828 enable_irq(msix_irq
);
5831 msix_irq
= adapter
->msix_entries
[vector
].vector
;
5832 disable_irq(msix_irq
);
5833 e1000_msix_other(msix_irq
, netdev
);
5834 enable_irq(msix_irq
);
5841 * Polling 'interrupt' - used by things like netconsole to send skbs
5842 * without having to re-enable interrupts. It's not called while
5843 * the interrupt routine is executing.
5845 static void e1000_netpoll(struct net_device
*netdev
)
5847 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5849 switch (adapter
->int_mode
) {
5850 case E1000E_INT_MODE_MSIX
:
5851 e1000_intr_msix(adapter
->pdev
->irq
, netdev
);
5853 case E1000E_INT_MODE_MSI
:
5854 disable_irq(adapter
->pdev
->irq
);
5855 e1000_intr_msi(adapter
->pdev
->irq
, netdev
);
5856 enable_irq(adapter
->pdev
->irq
);
5858 default: /* E1000E_INT_MODE_LEGACY */
5859 disable_irq(adapter
->pdev
->irq
);
5860 e1000_intr(adapter
->pdev
->irq
, netdev
);
5861 enable_irq(adapter
->pdev
->irq
);
5868 * e1000_io_error_detected - called when PCI error is detected
5869 * @pdev: Pointer to PCI device
5870 * @state: The current pci connection state
5872 * This function is called after a PCI bus error affecting
5873 * this device has been detected.
5875 static pci_ers_result_t
e1000_io_error_detected(struct pci_dev
*pdev
,
5876 pci_channel_state_t state
)
5878 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5879 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5881 netif_device_detach(netdev
);
5883 if (state
== pci_channel_io_perm_failure
)
5884 return PCI_ERS_RESULT_DISCONNECT
;
5886 if (netif_running(netdev
))
5887 e1000e_down(adapter
);
5888 pci_disable_device(pdev
);
5890 /* Request a slot slot reset. */
5891 return PCI_ERS_RESULT_NEED_RESET
;
5895 * e1000_io_slot_reset - called after the pci bus has been reset.
5896 * @pdev: Pointer to PCI device
5898 * Restart the card from scratch, as if from a cold-boot. Implementation
5899 * resembles the first-half of the e1000_resume routine.
5901 static pci_ers_result_t
e1000_io_slot_reset(struct pci_dev
*pdev
)
5903 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5904 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5905 struct e1000_hw
*hw
= &adapter
->hw
;
5906 u16 aspm_disable_flag
= 0;
5908 pci_ers_result_t result
;
5910 if (adapter
->flags2
& FLAG2_DISABLE_ASPM_L0S
)
5911 aspm_disable_flag
= PCIE_LINK_STATE_L0S
;
5912 if (adapter
->flags2
& FLAG2_DISABLE_ASPM_L1
)
5913 aspm_disable_flag
|= PCIE_LINK_STATE_L1
;
5914 if (aspm_disable_flag
)
5915 e1000e_disable_aspm(pdev
, aspm_disable_flag
);
5917 err
= pci_enable_device_mem(pdev
);
5920 "Cannot re-enable PCI device after reset.\n");
5921 result
= PCI_ERS_RESULT_DISCONNECT
;
5923 pci_set_master(pdev
);
5924 pdev
->state_saved
= true;
5925 pci_restore_state(pdev
);
5927 pci_enable_wake(pdev
, PCI_D3hot
, 0);
5928 pci_enable_wake(pdev
, PCI_D3cold
, 0);
5930 e1000e_reset(adapter
);
5932 result
= PCI_ERS_RESULT_RECOVERED
;
5935 pci_cleanup_aer_uncorrect_error_status(pdev
);
5941 * e1000_io_resume - called when traffic can start flowing again.
5942 * @pdev: Pointer to PCI device
5944 * This callback is called when the error recovery driver tells us that
5945 * its OK to resume normal operation. Implementation resembles the
5946 * second-half of the e1000_resume routine.
5948 static void e1000_io_resume(struct pci_dev
*pdev
)
5950 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5951 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
5953 e1000_init_manageability_pt(adapter
);
5955 if (netif_running(netdev
)) {
5956 if (e1000e_up(adapter
)) {
5958 "can't bring device back up after reset\n");
5963 netif_device_attach(netdev
);
5966 * If the controller has AMT, do not set DRV_LOAD until the interface
5967 * is up. For all other cases, let the f/w know that the h/w is now
5968 * under the control of the driver.
5970 if (!(adapter
->flags
& FLAG_HAS_AMT
))
5971 e1000e_get_hw_control(adapter
);
5975 static void e1000_print_device_info(struct e1000_adapter
*adapter
)
5977 struct e1000_hw
*hw
= &adapter
->hw
;
5978 struct net_device
*netdev
= adapter
->netdev
;
5980 u8 pba_str
[E1000_PBANUM_LENGTH
];
5982 /* print bus type/speed/width info */
5983 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5985 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ? "Width x4" :
5989 e_info("Intel(R) PRO/%s Network Connection\n",
5990 (hw
->phy
.type
== e1000_phy_ife
) ? "10/100" : "1000");
5991 ret_val
= e1000_read_pba_string_generic(hw
, pba_str
,
5992 E1000_PBANUM_LENGTH
);
5994 strlcpy((char *)pba_str
, "Unknown", sizeof(pba_str
));
5995 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5996 hw
->mac
.type
, hw
->phy
.type
, pba_str
);
5999 static void e1000_eeprom_checks(struct e1000_adapter
*adapter
)
6001 struct e1000_hw
*hw
= &adapter
->hw
;
6005 if (hw
->mac
.type
!= e1000_82573
)
6008 ret_val
= e1000_read_nvm(hw
, NVM_INIT_CONTROL2_REG
, 1, &buf
);
6010 if (!ret_val
&& (!(buf
& (1 << 0)))) {
6011 /* Deep Smart Power Down (DSPD) */
6012 dev_warn(&adapter
->pdev
->dev
,
6013 "Warning: detected DSPD enabled in EEPROM\n");
6017 static int e1000_set_features(struct net_device
*netdev
,
6018 netdev_features_t features
)
6020 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
6021 netdev_features_t changed
= features
^ netdev
->features
;
6023 if (changed
& (NETIF_F_TSO
| NETIF_F_TSO6
))
6024 adapter
->flags
|= FLAG_TSO_FORCE
;
6026 if (!(changed
& (NETIF_F_HW_VLAN_RX
| NETIF_F_HW_VLAN_TX
|
6027 NETIF_F_RXCSUM
| NETIF_F_RXHASH
| NETIF_F_RXFCS
|
6031 if (changed
& NETIF_F_RXFCS
) {
6032 if (features
& NETIF_F_RXFCS
) {
6033 adapter
->flags2
&= ~FLAG2_CRC_STRIPPING
;
6035 /* We need to take it back to defaults, which might mean
6036 * stripping is still disabled at the adapter level.
6038 if (adapter
->flags2
& FLAG2_DFLT_CRC_STRIPPING
)
6039 adapter
->flags2
|= FLAG2_CRC_STRIPPING
;
6041 adapter
->flags2
&= ~FLAG2_CRC_STRIPPING
;
6045 netdev
->features
= features
;
6047 if (netif_running(netdev
))
6048 e1000e_reinit_locked(adapter
);
6050 e1000e_reset(adapter
);
6055 static const struct net_device_ops e1000e_netdev_ops
= {
6056 .ndo_open
= e1000_open
,
6057 .ndo_stop
= e1000_close
,
6058 .ndo_start_xmit
= e1000_xmit_frame
,
6059 .ndo_get_stats64
= e1000e_get_stats64
,
6060 .ndo_set_rx_mode
= e1000e_set_rx_mode
,
6061 .ndo_set_mac_address
= e1000_set_mac
,
6062 .ndo_change_mtu
= e1000_change_mtu
,
6063 .ndo_do_ioctl
= e1000_ioctl
,
6064 .ndo_tx_timeout
= e1000_tx_timeout
,
6065 .ndo_validate_addr
= eth_validate_addr
,
6067 .ndo_vlan_rx_add_vid
= e1000_vlan_rx_add_vid
,
6068 .ndo_vlan_rx_kill_vid
= e1000_vlan_rx_kill_vid
,
6069 #ifdef CONFIG_NET_POLL_CONTROLLER
6070 .ndo_poll_controller
= e1000_netpoll
,
6072 .ndo_set_features
= e1000_set_features
,
6076 * e1000_probe - Device Initialization Routine
6077 * @pdev: PCI device information struct
6078 * @ent: entry in e1000_pci_tbl
6080 * Returns 0 on success, negative on failure
6082 * e1000_probe initializes an adapter identified by a pci_dev structure.
6083 * The OS initialization, configuring of the adapter private structure,
6084 * and a hardware reset occur.
6086 static int __devinit
e1000_probe(struct pci_dev
*pdev
,
6087 const struct pci_device_id
*ent
)
6089 struct net_device
*netdev
;
6090 struct e1000_adapter
*adapter
;
6091 struct e1000_hw
*hw
;
6092 const struct e1000_info
*ei
= e1000_info_tbl
[ent
->driver_data
];
6093 resource_size_t mmio_start
, mmio_len
;
6094 resource_size_t flash_start
, flash_len
;
6095 static int cards_found
;
6096 u16 aspm_disable_flag
= 0;
6097 int i
, err
, pci_using_dac
;
6098 u16 eeprom_data
= 0;
6099 u16 eeprom_apme_mask
= E1000_EEPROM_APME
;
6101 if (ei
->flags2
& FLAG2_DISABLE_ASPM_L0S
)
6102 aspm_disable_flag
= PCIE_LINK_STATE_L0S
;
6103 if (ei
->flags2
& FLAG2_DISABLE_ASPM_L1
)
6104 aspm_disable_flag
|= PCIE_LINK_STATE_L1
;
6105 if (aspm_disable_flag
)
6106 e1000e_disable_aspm(pdev
, aspm_disable_flag
);
6108 err
= pci_enable_device_mem(pdev
);
6113 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(64));
6115 err
= dma_set_coherent_mask(&pdev
->dev
, DMA_BIT_MASK(64));
6119 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(32));
6121 err
= dma_set_coherent_mask(&pdev
->dev
,
6124 dev_err(&pdev
->dev
, "No usable DMA configuration, aborting\n");
6130 err
= pci_request_selected_regions_exclusive(pdev
,
6131 pci_select_bars(pdev
, IORESOURCE_MEM
),
6132 e1000e_driver_name
);
6136 /* AER (Advanced Error Reporting) hooks */
6137 pci_enable_pcie_error_reporting(pdev
);
6139 pci_set_master(pdev
);
6140 /* PCI config space info */
6141 err
= pci_save_state(pdev
);
6143 goto err_alloc_etherdev
;
6146 netdev
= alloc_etherdev(sizeof(struct e1000_adapter
));
6148 goto err_alloc_etherdev
;
6150 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
6152 netdev
->irq
= pdev
->irq
;
6154 pci_set_drvdata(pdev
, netdev
);
6155 adapter
= netdev_priv(netdev
);
6157 adapter
->netdev
= netdev
;
6158 adapter
->pdev
= pdev
;
6160 adapter
->pba
= ei
->pba
;
6161 adapter
->flags
= ei
->flags
;
6162 adapter
->flags2
= ei
->flags2
;
6163 adapter
->hw
.adapter
= adapter
;
6164 adapter
->hw
.mac
.type
= ei
->mac
;
6165 adapter
->max_hw_frame_size
= ei
->max_hw_frame_size
;
6166 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
6168 mmio_start
= pci_resource_start(pdev
, 0);
6169 mmio_len
= pci_resource_len(pdev
, 0);
6172 adapter
->hw
.hw_addr
= ioremap(mmio_start
, mmio_len
);
6173 if (!adapter
->hw
.hw_addr
)
6176 if ((adapter
->flags
& FLAG_HAS_FLASH
) &&
6177 (pci_resource_flags(pdev
, 1) & IORESOURCE_MEM
)) {
6178 flash_start
= pci_resource_start(pdev
, 1);
6179 flash_len
= pci_resource_len(pdev
, 1);
6180 adapter
->hw
.flash_address
= ioremap(flash_start
, flash_len
);
6181 if (!adapter
->hw
.flash_address
)
6185 /* construct the net_device struct */
6186 netdev
->netdev_ops
= &e1000e_netdev_ops
;
6187 e1000e_set_ethtool_ops(netdev
);
6188 netdev
->watchdog_timeo
= 5 * HZ
;
6189 netif_napi_add(netdev
, &adapter
->napi
, e1000e_poll
, 64);
6190 strlcpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
));
6192 netdev
->mem_start
= mmio_start
;
6193 netdev
->mem_end
= mmio_start
+ mmio_len
;
6195 adapter
->bd_number
= cards_found
++;
6197 e1000e_check_options(adapter
);
6199 /* setup adapter struct */
6200 err
= e1000_sw_init(adapter
);
6204 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
6205 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
6206 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
6208 err
= ei
->get_variants(adapter
);
6212 if ((adapter
->flags
& FLAG_IS_ICH
) &&
6213 (adapter
->flags
& FLAG_READ_ONLY_NVM
))
6214 e1000e_write_protect_nvm_ich8lan(&adapter
->hw
);
6216 hw
->mac
.ops
.get_bus_info(&adapter
->hw
);
6218 adapter
->hw
.phy
.autoneg_wait_to_complete
= 0;
6220 /* Copper options */
6221 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
) {
6222 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
6223 adapter
->hw
.phy
.disable_polarity_correction
= 0;
6224 adapter
->hw
.phy
.ms_type
= e1000_ms_hw_default
;
6227 if (hw
->phy
.ops
.check_reset_block
&& hw
->phy
.ops
.check_reset_block(hw
))
6228 dev_info(&pdev
->dev
,
6229 "PHY reset is blocked due to SOL/IDER session.\n");
6231 /* Set initial default active device features */
6232 netdev
->features
= (NETIF_F_SG
|
6233 NETIF_F_HW_VLAN_RX
|
6234 NETIF_F_HW_VLAN_TX
|
6241 /* Set user-changeable features (subset of all device features) */
6242 netdev
->hw_features
= netdev
->features
;
6243 netdev
->hw_features
|= NETIF_F_RXFCS
;
6244 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
6245 netdev
->hw_features
|= NETIF_F_RXALL
;
6247 if (adapter
->flags
& FLAG_HAS_HW_VLAN_FILTER
)
6248 netdev
->features
|= NETIF_F_HW_VLAN_FILTER
;
6250 netdev
->vlan_features
|= (NETIF_F_SG
|
6255 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
6257 if (pci_using_dac
) {
6258 netdev
->features
|= NETIF_F_HIGHDMA
;
6259 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
6262 if (e1000e_enable_mng_pass_thru(&adapter
->hw
))
6263 adapter
->flags
|= FLAG_MNG_PT_ENABLED
;
6266 * before reading the NVM, reset the controller to
6267 * put the device in a known good starting state
6269 adapter
->hw
.mac
.ops
.reset_hw(&adapter
->hw
);
6272 * systems with ASPM and others may see the checksum fail on the first
6273 * attempt. Let's give it a few tries
6276 if (e1000_validate_nvm_checksum(&adapter
->hw
) >= 0)
6279 dev_err(&pdev
->dev
, "The NVM Checksum Is Not Valid\n");
6285 e1000_eeprom_checks(adapter
);
6287 /* copy the MAC address */
6288 if (e1000e_read_mac_addr(&adapter
->hw
))
6290 "NVM Read Error while reading MAC address\n");
6292 memcpy(netdev
->dev_addr
, adapter
->hw
.mac
.addr
, netdev
->addr_len
);
6293 memcpy(netdev
->perm_addr
, adapter
->hw
.mac
.addr
, netdev
->addr_len
);
6295 if (!is_valid_ether_addr(netdev
->perm_addr
)) {
6296 dev_err(&pdev
->dev
, "Invalid MAC Address: %pM\n",
6302 init_timer(&adapter
->watchdog_timer
);
6303 adapter
->watchdog_timer
.function
= e1000_watchdog
;
6304 adapter
->watchdog_timer
.data
= (unsigned long) adapter
;
6306 init_timer(&adapter
->phy_info_timer
);
6307 adapter
->phy_info_timer
.function
= e1000_update_phy_info
;
6308 adapter
->phy_info_timer
.data
= (unsigned long) adapter
;
6310 INIT_WORK(&adapter
->reset_task
, e1000_reset_task
);
6311 INIT_WORK(&adapter
->watchdog_task
, e1000_watchdog_task
);
6312 INIT_WORK(&adapter
->downshift_task
, e1000e_downshift_workaround
);
6313 INIT_WORK(&adapter
->update_phy_task
, e1000e_update_phy_task
);
6314 INIT_WORK(&adapter
->print_hang_task
, e1000_print_hw_hang
);
6316 /* Initialize link parameters. User can change them with ethtool */
6317 adapter
->hw
.mac
.autoneg
= 1;
6318 adapter
->fc_autoneg
= true;
6319 adapter
->hw
.fc
.requested_mode
= e1000_fc_default
;
6320 adapter
->hw
.fc
.current_mode
= e1000_fc_default
;
6321 adapter
->hw
.phy
.autoneg_advertised
= 0x2f;
6323 /* ring size defaults */
6324 adapter
->rx_ring
->count
= E1000_DEFAULT_RXD
;
6325 adapter
->tx_ring
->count
= E1000_DEFAULT_TXD
;
6328 * Initial Wake on LAN setting - If APM wake is enabled in
6329 * the EEPROM, enable the ACPI Magic Packet filter
6331 if (adapter
->flags
& FLAG_APME_IN_WUC
) {
6332 /* APME bit in EEPROM is mapped to WUC.APME */
6333 eeprom_data
= er32(WUC
);
6334 eeprom_apme_mask
= E1000_WUC_APME
;
6335 if ((hw
->mac
.type
> e1000_ich10lan
) &&
6336 (eeprom_data
& E1000_WUC_PHY_WAKE
))
6337 adapter
->flags2
|= FLAG2_HAS_PHY_WAKEUP
;
6338 } else if (adapter
->flags
& FLAG_APME_IN_CTRL3
) {
6339 if (adapter
->flags
& FLAG_APME_CHECK_PORT_B
&&
6340 (adapter
->hw
.bus
.func
== 1))
6341 e1000_read_nvm(&adapter
->hw
, NVM_INIT_CONTROL3_PORT_B
,
6344 e1000_read_nvm(&adapter
->hw
, NVM_INIT_CONTROL3_PORT_A
,
6348 /* fetch WoL from EEPROM */
6349 if (eeprom_data
& eeprom_apme_mask
)
6350 adapter
->eeprom_wol
|= E1000_WUFC_MAG
;
6353 * now that we have the eeprom settings, apply the special cases
6354 * where the eeprom may be wrong or the board simply won't support
6355 * wake on lan on a particular port
6357 if (!(adapter
->flags
& FLAG_HAS_WOL
))
6358 adapter
->eeprom_wol
= 0;
6360 /* initialize the wol settings based on the eeprom settings */
6361 adapter
->wol
= adapter
->eeprom_wol
;
6362 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
6364 /* save off EEPROM version number */
6365 e1000_read_nvm(&adapter
->hw
, 5, 1, &adapter
->eeprom_vers
);
6367 /* reset the hardware with the new settings */
6368 e1000e_reset(adapter
);
6371 * If the controller has AMT, do not set DRV_LOAD until the interface
6372 * is up. For all other cases, let the f/w know that the h/w is now
6373 * under the control of the driver.
6375 if (!(adapter
->flags
& FLAG_HAS_AMT
))
6376 e1000e_get_hw_control(adapter
);
6378 strlcpy(netdev
->name
, "eth%d", sizeof(netdev
->name
));
6379 err
= register_netdev(netdev
);
6383 /* carrier off reporting is important to ethtool even BEFORE open */
6384 netif_carrier_off(netdev
);
6386 e1000_print_device_info(adapter
);
6388 if (pci_dev_run_wake(pdev
))
6389 pm_runtime_put_noidle(&pdev
->dev
);
6394 if (!(adapter
->flags
& FLAG_HAS_AMT
))
6395 e1000e_release_hw_control(adapter
);
6397 if (hw
->phy
.ops
.check_reset_block
&& !hw
->phy
.ops
.check_reset_block(hw
))
6398 e1000_phy_hw_reset(&adapter
->hw
);
6400 kfree(adapter
->tx_ring
);
6401 kfree(adapter
->rx_ring
);
6403 if (adapter
->hw
.flash_address
)
6404 iounmap(adapter
->hw
.flash_address
);
6405 e1000e_reset_interrupt_capability(adapter
);
6407 iounmap(adapter
->hw
.hw_addr
);
6409 free_netdev(netdev
);
6411 pci_release_selected_regions(pdev
,
6412 pci_select_bars(pdev
, IORESOURCE_MEM
));
6415 pci_disable_device(pdev
);
6420 * e1000_remove - Device Removal Routine
6421 * @pdev: PCI device information struct
6423 * e1000_remove is called by the PCI subsystem to alert the driver
6424 * that it should release a PCI device. The could be caused by a
6425 * Hot-Plug event, or because the driver is going to be removed from
6428 static void __devexit
e1000_remove(struct pci_dev
*pdev
)
6430 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6431 struct e1000_adapter
*adapter
= netdev_priv(netdev
);
6432 bool down
= test_bit(__E1000_DOWN
, &adapter
->state
);
6435 * The timers may be rescheduled, so explicitly disable them
6436 * from being rescheduled.
6439 set_bit(__E1000_DOWN
, &adapter
->state
);
6440 del_timer_sync(&adapter
->watchdog_timer
);
6441 del_timer_sync(&adapter
->phy_info_timer
);
6443 cancel_work_sync(&adapter
->reset_task
);
6444 cancel_work_sync(&adapter
->watchdog_task
);
6445 cancel_work_sync(&adapter
->downshift_task
);
6446 cancel_work_sync(&adapter
->update_phy_task
);
6447 cancel_work_sync(&adapter
->print_hang_task
);
6449 if (!(netdev
->flags
& IFF_UP
))
6450 e1000_power_down_phy(adapter
);
6452 /* Don't lie to e1000_close() down the road. */
6454 clear_bit(__E1000_DOWN
, &adapter
->state
);
6455 unregister_netdev(netdev
);
6457 if (pci_dev_run_wake(pdev
))
6458 pm_runtime_get_noresume(&pdev
->dev
);
6461 * Release control of h/w to f/w. If f/w is AMT enabled, this
6462 * would have already happened in close and is redundant.
6464 e1000e_release_hw_control(adapter
);
6466 e1000e_reset_interrupt_capability(adapter
);
6467 kfree(adapter
->tx_ring
);
6468 kfree(adapter
->rx_ring
);
6470 iounmap(adapter
->hw
.hw_addr
);
6471 if (adapter
->hw
.flash_address
)
6472 iounmap(adapter
->hw
.flash_address
);
6473 pci_release_selected_regions(pdev
,
6474 pci_select_bars(pdev
, IORESOURCE_MEM
));
6476 free_netdev(netdev
);
6479 pci_disable_pcie_error_reporting(pdev
);
6481 pci_disable_device(pdev
);
6484 /* PCI Error Recovery (ERS) */
6485 static const struct pci_error_handlers e1000_err_handler
= {
6486 .error_detected
= e1000_io_error_detected
,
6487 .slot_reset
= e1000_io_slot_reset
,
6488 .resume
= e1000_io_resume
,
6491 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl
) = {
6492 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_COPPER
), board_82571
},
6493 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_FIBER
), board_82571
},
6494 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_COPPER
), board_82571
},
6495 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_COPPER_LP
), board_82571
},
6496 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_QUAD_FIBER
), board_82571
},
6497 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES
), board_82571
},
6498 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES_DUAL
), board_82571
},
6499 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571EB_SERDES_QUAD
), board_82571
},
6500 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82571PT_QUAD_COPPER
), board_82571
},
6502 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI
), board_82572
},
6503 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_COPPER
), board_82572
},
6504 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_FIBER
), board_82572
},
6505 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82572EI_SERDES
), board_82572
},
6507 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573E
), board_82573
},
6508 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573E_IAMT
), board_82573
},
6509 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82573L
), board_82573
},
6511 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82574L
), board_82574
},
6512 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82574LA
), board_82574
},
6513 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82583V
), board_82583
},
6515 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_COPPER_DPT
),
6516 board_80003es2lan
},
6517 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_COPPER_SPT
),
6518 board_80003es2lan
},
6519 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_SERDES_DPT
),
6520 board_80003es2lan
},
6521 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_80003ES2LAN_SERDES_SPT
),
6522 board_80003es2lan
},
6524 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE
), board_ich8lan
},
6525 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE_G
), board_ich8lan
},
6526 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IFE_GT
), board_ich8lan
},
6527 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_AMT
), board_ich8lan
},
6528 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_C
), board_ich8lan
},
6529 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_M
), board_ich8lan
},
6530 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_IGP_M_AMT
), board_ich8lan
},
6531 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH8_82567V_3
), board_ich8lan
},
6533 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE
), board_ich9lan
},
6534 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE_G
), board_ich9lan
},
6535 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IFE_GT
), board_ich9lan
},
6536 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_AMT
), board_ich9lan
},
6537 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_C
), board_ich9lan
},
6538 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_BM
), board_ich9lan
},
6539 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M
), board_ich9lan
},
6540 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M_AMT
), board_ich9lan
},
6541 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH9_IGP_M_V
), board_ich9lan
},
6543 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_LM
), board_ich9lan
},
6544 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_LF
), board_ich9lan
},
6545 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_R_BM_V
), board_ich9lan
},
6547 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_D_BM_LM
), board_ich10lan
},
6548 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_D_BM_LF
), board_ich10lan
},
6549 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_ICH10_D_BM_V
), board_ich10lan
},
6551 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_M_HV_LM
), board_pchlan
},
6552 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_M_HV_LC
), board_pchlan
},
6553 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_D_HV_DM
), board_pchlan
},
6554 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_D_HV_DC
), board_pchlan
},
6556 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH2_LV_LM
), board_pch2lan
},
6557 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH2_LV_V
), board_pch2lan
},
6559 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_LPT_I217_LM
), board_pch_lpt
},
6560 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_PCH_LPT_I217_V
), board_pch_lpt
},
6562 { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6564 MODULE_DEVICE_TABLE(pci
, e1000_pci_tbl
);
6567 static const struct dev_pm_ops e1000_pm_ops
= {
6568 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend
, e1000_resume
)
6569 SET_RUNTIME_PM_OPS(e1000_runtime_suspend
,
6570 e1000_runtime_resume
, e1000_idle
)
6574 /* PCI Device API Driver */
6575 static struct pci_driver e1000_driver
= {
6576 .name
= e1000e_driver_name
,
6577 .id_table
= e1000_pci_tbl
,
6578 .probe
= e1000_probe
,
6579 .remove
= __devexit_p(e1000_remove
),
6582 .pm
= &e1000_pm_ops
,
6585 .shutdown
= e1000_shutdown
,
6586 .err_handler
= &e1000_err_handler
6590 * e1000_init_module - Driver Registration Routine
6592 * e1000_init_module is the first routine called when the driver is
6593 * loaded. All it does is register with the PCI subsystem.
6595 static int __init
e1000_init_module(void)
6598 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6599 e1000e_driver_version
);
6600 pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6601 ret
= pci_register_driver(&e1000_driver
);
6605 module_init(e1000_init_module
);
6608 * e1000_exit_module - Driver Exit Cleanup Routine
6610 * e1000_exit_module is called just before the driver is removed
6613 static void __exit
e1000_exit_module(void)
6615 pci_unregister_driver(&e1000_driver
);
6617 module_exit(e1000_exit_module
);
6620 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6621 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6622 MODULE_LICENSE("GPL");
6623 MODULE_VERSION(DRV_VERSION
);