3 * Joerg Sonnenberger <joerg@bec.de>. All rights reserved.
5 * Copyright (c) 1997, 1998-2003
6 * Bill Paul <wpaul@windriver.com>. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by Bill Paul.
19 * 4. Neither the name of the author nor the names of any co-contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
27 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33 * THE POSSIBILITY OF SUCH DAMAGE.
35 * $FreeBSD: src/sys/dev/re/if_re.c,v 1.25 2004/06/09 14:34:01 naddy Exp $
36 * $DragonFly: src/sys/dev/netif/re/if_re.c,v 1.43 2008/05/16 13:19:12 sephe Exp $
40 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
42 * Written by Bill Paul <wpaul@windriver.com>
43 * Senior Networking Software Engineer
48 * This driver is designed to support RealTek's next generation of
49 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
50 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
51 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
53 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
54 * with the older 8139 family, however it also supports a special
55 * C+ mode of operation that provides several new performance enhancing
56 * features. These include:
58 * o Descriptor based DMA mechanism. Each descriptor represents
59 * a single packet fragment. Data buffers may be aligned on
64 * o TCP/IP checksum offload for both RX and TX
66 * o High and normal priority transmit DMA rings
68 * o VLAN tag insertion and extraction
70 * o TCP large send (segmentation offload)
72 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
73 * programming API is fairly straightforward. The RX filtering, EEPROM
74 * access and PHY access is the same as it is on the older 8139 series
77 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
78 * same programming API and feature set as the 8139C+ with the following
79 * differences and additions:
85 * o GMII and TBI ports/registers for interfacing with copper
88 * o RX and TX DMA rings can have up to 1024 descriptors
89 * (the 8139C+ allows a maximum of 64)
91 * o Slight differences in register layout from the 8139C+
93 * The TX start and timer interrupt registers are at different locations
94 * on the 8169 than they are on the 8139C+. Also, the status word in the
95 * RX descriptor has a slightly different bit layout. The 8169 does not
96 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
99 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
100 * (the 'S' stands for 'single-chip'). These devices have the same
101 * programming API as the older 8169, but also have some vendor-specific
102 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
103 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
105 * This driver takes advantage of the RX and TX checksum offload and
106 * VLAN tag insertion/extraction features. It also implements TX
107 * interrupt moderation using the timer interrupt registers, which
108 * significantly reduces TX interrupt load. There is also support
109 * for jumbo frames, however the 8169/8169S/8110S can not transmit
110 * jumbo frames larger than 7440, so the max MTU possible with this
111 * driver is 7422 bytes.
114 #include "opt_polling.h"
116 #include <sys/param.h>
118 #include <sys/endian.h>
119 #include <sys/kernel.h>
120 #include <sys/interrupt.h>
121 #include <sys/malloc.h>
122 #include <sys/mbuf.h>
123 #include <sys/rman.h>
124 #include <sys/serialize.h>
125 #include <sys/socket.h>
126 #include <sys/sockio.h>
127 #include <sys/sysctl.h>
130 #include <net/ethernet.h>
132 #include <net/ifq_var.h>
133 #include <net/if_arp.h>
134 #include <net/if_dl.h>
135 #include <net/if_media.h>
136 #include <net/if_types.h>
137 #include <net/vlan/if_vlan_var.h>
138 #include <net/vlan/if_vlan_ether.h>
140 #include <dev/netif/mii_layer/mii.h>
141 #include <dev/netif/mii_layer/miivar.h>
143 #include <bus/pci/pcidevs.h>
144 #include <bus/pci/pcireg.h>
145 #include <bus/pci/pcivar.h>
147 /* "device miibus" required. See GENERIC if you get errors here. */
148 #include "miibus_if.h"
150 #include <dev/netif/re/if_rereg.h>
151 #include <dev/netif/re/if_revar.h>
153 #define RE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
155 #define RE_DISABLE_HWCSUM
159 * Various supported device vendors/types and their names.
161 static const struct re_type re_devs
[] = {
162 { PCI_VENDOR_DLINK
, PCI_PRODUCT_DLINK_DGE528T
, RE_HWREV_8169S
,
163 "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
164 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8139
, RE_HWREV_8139CPLUS
,
165 "RealTek 8139C+ 10/100BaseTX" },
166 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8101E
, RE_HWREV_8101E
,
167 "RealTek 8101E PCIe 10/100baseTX" },
168 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8168
, RE_HWREV_8168_SPIN1
,
169 "RealTek 8168/8111B PCIe Gigabit Ethernet" },
170 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8168
, RE_HWREV_8168_SPIN2
,
171 "RealTek 8168/8111B PCIe Gigabit Ethernet" },
172 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8168
, RE_HWREV_8168_SPIN3
,
173 "RealTek 8168B/8111B PCIe Gigabit Ethernet" },
174 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8168
, RE_HWREV_8168C
,
175 "RealTek 8168C/8111C PCIe Gigabit Ethernet" },
176 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169
, RE_HWREV_8169
,
177 "RealTek 8169 Gigabit Ethernet" },
178 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169
, RE_HWREV_8169S
,
179 "RealTek 8169S Single-chip Gigabit Ethernet" },
180 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169
, RE_HWREV_8169_8110SB
,
181 "RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
182 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169
, RE_HWREV_8169_8110SC
,
183 "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
184 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169SC
, RE_HWREV_8169_8110SC
,
185 "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
186 { PCI_VENDOR_REALTEK
, PCI_PRODUCT_REALTEK_RT8169
, RE_HWREV_8110S
,
187 "RealTek 8110S Single-chip Gigabit Ethernet" },
188 { PCI_VENDOR_COREGA
, PCI_PRODUCT_COREGA_CG_LAPCIGT
, RE_HWREV_8169S
,
189 "Corega CG-LAPCIGT Gigabit Ethernet" },
190 { PCI_VENDOR_LINKSYS
, PCI_PRODUCT_LINKSYS_EG1032
, RE_HWREV_8169S
,
191 "Linksys EG1032 Gigabit Ethernet" },
192 { PCI_VENDOR_USR2
, PCI_PRODUCT_USR2_997902
, RE_HWREV_8169S
,
193 "US Robotics 997902 Gigabit Ethernet" },
197 static const struct re_hwrev re_hwrevs
[] = {
198 { RE_HWREV_8139CPLUS
, RE_8139CPLUS
, RE_F_HASMPC
, "C+" },
199 { RE_HWREV_8168_SPIN1
, RE_8169
, RE_F_PCIE
, "8168" },
200 { RE_HWREV_8168_SPIN2
, RE_8169
, RE_F_PCIE
, "8168" },
201 { RE_HWREV_8168_SPIN3
, RE_8169
, RE_F_PCIE
, "8168" },
202 { RE_HWREV_8168C
, RE_8169
, RE_F_PCIE
, "8168C" },
203 { RE_HWREV_8169
, RE_8169
, RE_F_HASMPC
, "8169" },
204 { RE_HWREV_8169S
, RE_8169
, RE_F_HASMPC
, "8169S" },
205 { RE_HWREV_8110S
, RE_8169
, RE_F_HASMPC
, "8110S" },
206 { RE_HWREV_8169_8110SB
, RE_8169
, RE_F_HASMPC
, "8169SB" },
207 { RE_HWREV_8169_8110SC
, RE_8169
, 0, "8169SC" },
208 { RE_HWREV_8100E
, RE_8169
, RE_F_HASMPC
, "8100E" },
209 { RE_HWREV_8101E
, RE_8169
, RE_F_PCIE
, "8101E" },
213 static int re_probe(device_t
);
214 static int re_attach(device_t
);
215 static int re_detach(device_t
);
217 static int re_encap(struct re_softc
*, struct mbuf
**, int *, int *);
219 static void re_dma_map_addr(void *, bus_dma_segment_t
*, int, int);
220 static void re_dma_map_desc(void *, bus_dma_segment_t
*, int,
222 static int re_allocmem(device_t
, struct re_softc
*);
223 static int re_newbuf(struct re_softc
*, int, struct mbuf
*);
224 static int re_rx_list_init(struct re_softc
*);
225 static int re_tx_list_init(struct re_softc
*);
226 static void re_rxeof(struct re_softc
*);
227 static void re_txeof(struct re_softc
*);
228 static void re_intr(void *);
229 static void re_tick(void *);
230 static void re_tick_serialized(void *);
231 static void re_start(struct ifnet
*);
232 static int re_ioctl(struct ifnet
*, u_long
, caddr_t
, struct ucred
*);
233 static void re_init(void *);
234 static void re_stop(struct re_softc
*);
235 static void re_watchdog(struct ifnet
*);
236 static int re_suspend(device_t
);
237 static int re_resume(device_t
);
238 static void re_shutdown(device_t
);
239 static int re_ifmedia_upd(struct ifnet
*);
240 static void re_ifmedia_sts(struct ifnet
*, struct ifmediareq
*);
242 static void re_eeprom_putbyte(struct re_softc
*, int);
243 static void re_eeprom_getword(struct re_softc
*, int, u_int16_t
*);
244 static void re_read_eeprom(struct re_softc
*, caddr_t
, int, int);
245 static int re_gmii_readreg(device_t
, int, int);
246 static int re_gmii_writereg(device_t
, int, int, int);
248 static int re_miibus_readreg(device_t
, int, int);
249 static int re_miibus_writereg(device_t
, int, int, int);
250 static void re_miibus_statchg(device_t
);
252 static void re_setmulti(struct re_softc
*);
253 static void re_reset(struct re_softc
*);
256 static int re_diag(struct re_softc
*);
259 #ifdef DEVICE_POLLING
260 static void re_poll(struct ifnet
*ifp
, enum poll_cmd cmd
, int count
);
263 static int re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS
);
265 static device_method_t re_methods
[] = {
266 /* Device interface */
267 DEVMETHOD(device_probe
, re_probe
),
268 DEVMETHOD(device_attach
, re_attach
),
269 DEVMETHOD(device_detach
, re_detach
),
270 DEVMETHOD(device_suspend
, re_suspend
),
271 DEVMETHOD(device_resume
, re_resume
),
272 DEVMETHOD(device_shutdown
, re_shutdown
),
275 DEVMETHOD(bus_print_child
, bus_generic_print_child
),
276 DEVMETHOD(bus_driver_added
, bus_generic_driver_added
),
279 DEVMETHOD(miibus_readreg
, re_miibus_readreg
),
280 DEVMETHOD(miibus_writereg
, re_miibus_writereg
),
281 DEVMETHOD(miibus_statchg
, re_miibus_statchg
),
286 static driver_t re_driver
= {
289 sizeof(struct re_softc
)
292 static devclass_t re_devclass
;
294 DECLARE_DUMMY_MODULE(if_re
);
295 DRIVER_MODULE(if_re
, pci
, re_driver
, re_devclass
, 0, 0);
296 DRIVER_MODULE(if_re
, cardbus
, re_driver
, re_devclass
, 0, 0);
297 DRIVER_MODULE(miibus
, re
, miibus_driver
, miibus_devclass
, 0, 0);
300 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
303 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
306 * Send a read command and address to the EEPROM, check for ACK.
309 re_eeprom_putbyte(struct re_softc
*sc
, int addr
)
313 d
= addr
| (RE_9346_READ
<< sc
->re_eewidth
);
316 * Feed in each bit and strobe the clock.
318 for (i
= 1 << (sc
->re_eewidth
+ 3); i
; i
>>= 1) {
320 EE_SET(RE_EE_DATAIN
);
322 EE_CLR(RE_EE_DATAIN
);
332 * Read a word of data stored in the EEPROM at address 'addr.'
335 re_eeprom_getword(struct re_softc
*sc
, int addr
, uint16_t *dest
)
341 * Send address of word we want to read.
343 re_eeprom_putbyte(sc
, addr
);
346 * Start reading bits from EEPROM.
348 for (i
= 0x8000; i
!= 0; i
>>= 1) {
351 if (CSR_READ_1(sc
, RE_EECMD
) & RE_EE_DATAOUT
)
361 * Read a sequence of words from the EEPROM.
364 re_read_eeprom(struct re_softc
*sc
, caddr_t dest
, int off
, int cnt
)
367 uint16_t word
= 0, *ptr
;
369 CSR_SETBIT_1(sc
, RE_EECMD
, RE_EEMODE_PROGRAM
);
372 for (i
= 0; i
< cnt
; i
++) {
373 CSR_SETBIT_1(sc
, RE_EECMD
, RE_EE_SEL
);
374 re_eeprom_getword(sc
, off
+ i
, &word
);
375 CSR_CLRBIT_1(sc
, RE_EECMD
, RE_EE_SEL
);
376 ptr
= (uint16_t *)(dest
+ (i
* 2));
380 CSR_CLRBIT_1(sc
, RE_EECMD
, RE_EEMODE_PROGRAM
);
384 re_gmii_readreg(device_t dev
, int phy
, int reg
)
386 struct re_softc
*sc
= device_get_softc(dev
);
393 /* Let the rgephy driver read the GMEDIASTAT register */
395 if (reg
== RE_GMEDIASTAT
)
396 return(CSR_READ_1(sc
, RE_GMEDIASTAT
));
398 CSR_WRITE_4(sc
, RE_PHYAR
, reg
<< 16);
401 for (i
= 0; i
< RE_TIMEOUT
; i
++) {
402 rval
= CSR_READ_4(sc
, RE_PHYAR
);
403 if (rval
& RE_PHYAR_BUSY
)
408 if (i
== RE_TIMEOUT
) {
409 device_printf(dev
, "PHY read failed\n");
413 return(rval
& RE_PHYAR_PHYDATA
);
417 re_gmii_writereg(device_t dev
, int phy
, int reg
, int data
)
419 struct re_softc
*sc
= device_get_softc(dev
);
423 CSR_WRITE_4(sc
, RE_PHYAR
,
424 (reg
<< 16) | (data
& RE_PHYAR_PHYDATA
) | RE_PHYAR_BUSY
);
427 for (i
= 0; i
< RE_TIMEOUT
; i
++) {
428 rval
= CSR_READ_4(sc
, RE_PHYAR
);
429 if ((rval
& RE_PHYAR_BUSY
) == 0)
435 device_printf(dev
, "PHY write failed\n");
441 re_miibus_readreg(device_t dev
, int phy
, int reg
)
443 struct re_softc
*sc
= device_get_softc(dev
);
445 uint16_t re8139_reg
= 0;
447 if (sc
->re_type
== RE_8169
) {
448 rval
= re_gmii_readreg(dev
, phy
, reg
);
452 /* Pretend the internal PHY is only at address 0 */
458 re8139_reg
= RE_BMCR
;
461 re8139_reg
= RE_BMSR
;
464 re8139_reg
= RE_ANAR
;
467 re8139_reg
= RE_ANER
;
470 re8139_reg
= RE_LPAR
;
476 * Allow the rlphy driver to read the media status
477 * register. If we have a link partner which does not
478 * support NWAY, this is the register which will tell
479 * us the results of parallel detection.
482 return(CSR_READ_1(sc
, RE_MEDIASTAT
));
484 device_printf(dev
, "bad phy register\n");
487 rval
= CSR_READ_2(sc
, re8139_reg
);
488 if (sc
->re_type
== RE_8139CPLUS
&& re8139_reg
== RE_BMCR
) {
489 /* 8139C+ has different bit layout. */
490 rval
&= ~(BMCR_LOOP
| BMCR_ISO
);
496 re_miibus_writereg(device_t dev
, int phy
, int reg
, int data
)
498 struct re_softc
*sc
= device_get_softc(dev
);
499 u_int16_t re8139_reg
= 0;
501 if (sc
->re_type
== RE_8169
)
502 return(re_gmii_writereg(dev
, phy
, reg
, data
));
504 /* Pretend the internal PHY is only at address 0 */
510 re8139_reg
= RE_BMCR
;
511 if (sc
->re_type
== RE_8139CPLUS
) {
512 /* 8139C+ has different bit layout. */
513 data
&= ~(BMCR_LOOP
| BMCR_ISO
);
517 re8139_reg
= RE_BMSR
;
520 re8139_reg
= RE_ANAR
;
523 re8139_reg
= RE_ANER
;
526 re8139_reg
= RE_LPAR
;
532 device_printf(dev
, "bad phy register\n");
535 CSR_WRITE_2(sc
, re8139_reg
, data
);
540 re_miibus_statchg(device_t dev
)
545 * Program the 64-bit multicast hash filter.
548 re_setmulti(struct re_softc
*sc
)
550 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
552 uint32_t hashes
[2] = { 0, 0 };
553 struct ifmultiaddr
*ifma
;
557 rxfilt
= CSR_READ_4(sc
, RE_RXCFG
);
559 if (ifp
->if_flags
& IFF_ALLMULTI
|| ifp
->if_flags
& IFF_PROMISC
) {
560 rxfilt
|= RE_RXCFG_RX_MULTI
;
561 CSR_WRITE_4(sc
, RE_RXCFG
, rxfilt
);
562 CSR_WRITE_4(sc
, RE_MAR0
, 0xFFFFFFFF);
563 CSR_WRITE_4(sc
, RE_MAR4
, 0xFFFFFFFF);
567 /* first, zot all the existing hash bits */
568 CSR_WRITE_4(sc
, RE_MAR0
, 0);
569 CSR_WRITE_4(sc
, RE_MAR4
, 0);
571 /* now program new ones */
572 LIST_FOREACH(ifma
, &ifp
->if_multiaddrs
, ifma_link
) {
573 if (ifma
->ifma_addr
->sa_family
!= AF_LINK
)
575 h
= ether_crc32_be(LLADDR((struct sockaddr_dl
*)
576 ifma
->ifma_addr
), ETHER_ADDR_LEN
) >> 26;
578 hashes
[0] |= (1 << h
);
580 hashes
[1] |= (1 << (h
- 32));
585 rxfilt
|= RE_RXCFG_RX_MULTI
;
587 rxfilt
&= ~RE_RXCFG_RX_MULTI
;
589 CSR_WRITE_4(sc
, RE_RXCFG
, rxfilt
);
592 * For some unfathomable reason, RealTek decided to reverse
593 * the order of the multicast hash registers in the PCI Express
594 * parts. This means we have to write the hash pattern in reverse
595 * order for those devices.
597 if (sc
->re_flags
& RE_F_PCIE
) {
598 CSR_WRITE_4(sc
, RE_MAR0
, bswap32(hashes
[0]));
599 CSR_WRITE_4(sc
, RE_MAR4
, bswap32(hashes
[1]));
601 CSR_WRITE_4(sc
, RE_MAR0
, hashes
[0]);
602 CSR_WRITE_4(sc
, RE_MAR4
, hashes
[1]);
607 re_reset(struct re_softc
*sc
)
611 CSR_WRITE_1(sc
, RE_COMMAND
, RE_CMD_RESET
);
613 for (i
= 0; i
< RE_TIMEOUT
; i
++) {
615 if ((CSR_READ_1(sc
, RE_COMMAND
) & RE_CMD_RESET
) == 0)
619 if_printf(&sc
->arpcom
.ac_if
, "reset never completed!\n");
621 CSR_WRITE_1(sc
, 0x82, 1);
626 * The following routine is designed to test for a defect on some
627 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
628 * lines connected to the bus, however for a 32-bit only card, they
629 * should be pulled high. The result of this defect is that the
630 * NIC will not work right if you plug it into a 64-bit slot: DMA
631 * operations will be done with 64-bit transfers, which will fail
632 * because the 64-bit data lines aren't connected.
634 * There's no way to work around this (short of talking a soldering
635 * iron to the board), however we can detect it. The method we use
636 * here is to put the NIC into digital loopback mode, set the receiver
637 * to promiscuous mode, and then try to send a frame. We then compare
638 * the frame data we sent to what was received. If the data matches,
639 * then the NIC is working correctly, otherwise we know the user has
640 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
641 * slot. In the latter case, there's no way the NIC can work correctly,
642 * so we print out a message on the console and abort the device attach.
646 re_diag(struct re_softc
*sc
)
648 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
650 struct ether_header
*eh
;
651 struct re_desc
*cur_rx
;
654 int total_len
, i
, error
= 0, phyaddr
;
655 uint8_t dst
[ETHER_ADDR_LEN
] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
656 uint8_t src
[ETHER_ADDR_LEN
] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
658 /* Allocate a single mbuf */
660 MGETHDR(m0
, MB_DONTWAIT
, MT_DATA
);
665 * Initialize the NIC in test mode. This sets the chip up
666 * so that it can send and receive frames, but performs the
667 * following special functions:
668 * - Puts receiver in promiscuous mode
669 * - Enables digital loopback mode
670 * - Leaves interrupts turned off
673 ifp
->if_flags
|= IFF_PROMISC
;
678 if (sc
->re_type
== RE_8169
)
683 re_miibus_writereg(sc
->re_dev
, phyaddr
, MII_BMCR
, BMCR_RESET
);
684 for (i
= 0; i
< RE_TIMEOUT
; i
++) {
685 status
= re_miibus_readreg(sc
->re_dev
, phyaddr
, MII_BMCR
);
686 if (!(status
& BMCR_RESET
))
690 re_miibus_writereg(sc
->re_dev
, phyaddr
, MII_BMCR
, BMCR_LOOP
);
691 CSR_WRITE_2(sc
, RE_ISR
, RE_INTRS_DIAG
);
695 /* Put some data in the mbuf */
697 eh
= mtod(m0
, struct ether_header
*);
698 bcopy (dst
, eh
->ether_dhost
, ETHER_ADDR_LEN
);
699 bcopy (src
, eh
->ether_shost
, ETHER_ADDR_LEN
);
700 eh
->ether_type
= htons(ETHERTYPE_IP
);
701 m0
->m_pkthdr
.len
= m0
->m_len
= ETHER_MIN_LEN
- ETHER_CRC_LEN
;
704 * Queue the packet, start transmission.
705 * Note: ifq_handoff() ultimately calls re_start() for us.
708 CSR_WRITE_2(sc
, RE_ISR
, 0xFFFF);
709 error
= ifq_handoff(ifp
, m0
, NULL
);
716 /* Wait for it to propagate through the chip */
719 for (i
= 0; i
< RE_TIMEOUT
; i
++) {
720 status
= CSR_READ_2(sc
, RE_ISR
);
721 CSR_WRITE_2(sc
, RE_ISR
, status
);
722 if ((status
& (RE_ISR_TIMEOUT_EXPIRED
|RE_ISR_RX_OK
)) ==
723 (RE_ISR_TIMEOUT_EXPIRED
|RE_ISR_RX_OK
))
728 if (i
== RE_TIMEOUT
) {
729 if_printf(ifp
, "diagnostic failed to receive packet "
730 "in loopback mode\n");
736 * The packet should have been dumped into the first
737 * entry in the RX DMA ring. Grab it from there.
740 bus_dmamap_sync(sc
->re_ldata
.re_rx_list_tag
,
741 sc
->re_ldata
.re_rx_list_map
, BUS_DMASYNC_POSTREAD
);
742 bus_dmamap_sync(sc
->re_ldata
.re_mtag
, sc
->re_ldata
.re_rx_dmamap
[0],
743 BUS_DMASYNC_POSTWRITE
);
744 bus_dmamap_unload(sc
->re_ldata
.re_mtag
, sc
->re_ldata
.re_rx_dmamap
[0]);
746 m0
= sc
->re_ldata
.re_rx_mbuf
[0];
747 sc
->re_ldata
.re_rx_mbuf
[0] = NULL
;
748 eh
= mtod(m0
, struct ether_header
*);
750 cur_rx
= &sc
->re_ldata
.re_rx_list
[0];
751 total_len
= RE_RXBYTES(cur_rx
);
752 rxstat
= le32toh(cur_rx
->re_cmdstat
);
754 if (total_len
!= ETHER_MIN_LEN
) {
755 if_printf(ifp
, "diagnostic failed, received short packet\n");
760 /* Test that the received packet data matches what we sent. */
762 if (bcmp(eh
->ether_dhost
, dst
, ETHER_ADDR_LEN
) ||
763 bcmp(eh
->ether_shost
, &src
, ETHER_ADDR_LEN
) ||
764 be16toh(eh
->ether_type
) != ETHERTYPE_IP
) {
765 if_printf(ifp
, "WARNING, DMA FAILURE!\n");
766 if_printf(ifp
, "expected TX data: %6D/%6D/0x%x\n",
767 dst
, ":", src
, ":", ETHERTYPE_IP
);
768 if_printf(ifp
, "received RX data: %6D/%6D/0x%x\n",
769 eh
->ether_dhost
, ":", eh
->ether_shost
, ":",
770 ntohs(eh
->ether_type
));
771 if_printf(ifp
, "You may have a defective 32-bit NIC plugged "
772 "into a 64-bit PCI slot.\n");
773 if_printf(ifp
, "Please re-install the NIC in a 32-bit slot "
774 "for proper operation.\n");
775 if_printf(ifp
, "Read the re(4) man page for more details.\n");
780 /* Turn interface off, release resources */
784 ifp
->if_flags
&= ~IFF_PROMISC
;
794 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
795 * IDs against our list and return a device name if we find a match.
798 re_probe(device_t dev
)
800 const struct re_type
*t
;
804 uint16_t vendor
, product
;
808 vendor
= pci_get_vendor(dev
);
809 product
= pci_get_device(dev
);
812 * Only attach to rev.3 of the Linksys EG1032 adapter.
813 * Rev.2 is supported by sk(4).
815 if (vendor
== PCI_VENDOR_LINKSYS
&&
816 product
== PCI_PRODUCT_LINKSYS_EG1032
&&
817 pci_get_subdevice(dev
) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3
)
820 for (t
= re_devs
; t
->re_name
!= NULL
; t
++) {
821 if (product
== t
->re_did
&& vendor
== t
->re_vid
)
826 * Check if we found a RealTek device.
828 if (t
->re_name
== NULL
)
832 * Temporarily map the I/O space so we can read the chip ID register.
834 sc
= kmalloc(sizeof(*sc
), M_TEMP
, M_WAITOK
| M_ZERO
);
836 sc
->re_res
= bus_alloc_resource_any(dev
, SYS_RES_IOPORT
, &rid
,
838 if (sc
->re_res
== NULL
) {
839 device_printf(dev
, "couldn't map ports/memory\n");
844 sc
->re_btag
= rman_get_bustag(sc
->re_res
);
845 sc
->re_bhandle
= rman_get_bushandle(sc
->re_res
);
847 hwrev
= CSR_READ_4(sc
, RE_TXCFG
) & RE_TXCFG_HWREV
;
848 bus_release_resource(dev
, SYS_RES_IOPORT
, RE_PCI_LOIO
, sc
->re_res
);
852 * and continue matching for the specific chip...
854 for (; t
->re_name
!= NULL
; t
++) {
855 if (product
== t
->re_did
&& vendor
== t
->re_vid
&&
856 t
->re_basetype
== hwrev
) {
857 device_set_desc(dev
, t
->re_name
);
863 kprintf("re: unknown hwrev %#x\n", hwrev
);
868 * This routine takes the segment list provided as the result of
869 * a bus_dma_map_load() operation and assigns the addresses/lengths
870 * to RealTek DMA descriptors. This can be called either by the RX
871 * code or the TX code. In the RX case, we'll probably wind up mapping
872 * at most one segment. For the TX case, there could be any number of
873 * segments since TX packets may span multiple mbufs. In either case,
874 * if the number of segments is larger than the re_maxsegs limit
875 * specified by the caller, we abort the mapping operation. Sadly,
876 * whoever designed the buffer mapping API did not provide a way to
877 * return an error from here, so we have to fake it a bit.
881 re_dma_map_desc(void *arg
, bus_dma_segment_t
*segs
, int nseg
,
882 bus_size_t mapsize
, int error
)
884 struct re_dmaload_arg
*ctx
;
885 struct re_desc
*d
= NULL
;
894 /* Signal error to caller if there's too many segments */
895 if (nseg
> ctx
->re_maxsegs
) {
901 * Map the segment array into descriptors. Note that we set the
902 * start-of-frame and end-of-frame markers for either TX or RX, but
903 * they really only have meaning in the TX case. (In the RX case,
904 * it's the chip that tells us where packets begin and end.)
905 * We also keep track of the end of the ring and set the
906 * end-of-ring bits as needed, and we set the ownership bits
907 * in all except the very first descriptor. (The caller will
908 * set this descriptor later when it start transmission or
913 d
= &ctx
->re_ring
[idx
];
914 if (le32toh(d
->re_cmdstat
) & RE_RDESC_STAT_OWN
) {
918 cmdstat
= segs
[i
].ds_len
;
919 d
->re_bufaddr_lo
= htole32(RE_ADDR_LO(segs
[i
].ds_addr
));
920 d
->re_bufaddr_hi
= htole32(RE_ADDR_HI(segs
[i
].ds_addr
));
922 cmdstat
|= RE_TDESC_CMD_SOF
;
924 cmdstat
|= RE_TDESC_CMD_OWN
;
925 if (idx
== (RE_RX_DESC_CNT
- 1))
926 cmdstat
|= RE_TDESC_CMD_EOR
;
927 d
->re_cmdstat
= htole32(cmdstat
| ctx
->re_flags
);
934 d
->re_cmdstat
|= htole32(RE_TDESC_CMD_EOF
);
935 ctx
->re_maxsegs
= nseg
;
940 * Map a single buffer address.
944 re_dma_map_addr(void *arg
, bus_dma_segment_t
*segs
, int nseg
, int error
)
951 KASSERT(nseg
== 1, ("too many DMA segments, %d should be 1", nseg
));
953 *addr
= segs
->ds_addr
;
957 re_allocmem(device_t dev
, struct re_softc
*sc
)
962 * Allocate map for RX mbufs.
965 error
= bus_dma_tag_create(sc
->re_parent_tag
, ETHER_ALIGN
, 0,
966 BUS_SPACE_MAXADDR_32BIT
, BUS_SPACE_MAXADDR
, NULL
,
967 NULL
, MCLBYTES
* nseg
, nseg
, MCLBYTES
, BUS_DMA_ALLOCNOW
,
968 &sc
->re_ldata
.re_mtag
);
970 device_printf(dev
, "could not allocate dma tag\n");
975 * Allocate map for TX descriptor list.
977 error
= bus_dma_tag_create(sc
->re_parent_tag
, RE_RING_ALIGN
,
978 0, BUS_SPACE_MAXADDR_32BIT
, BUS_SPACE_MAXADDR
, NULL
,
979 NULL
, RE_TX_LIST_SZ
, 1, RE_TX_LIST_SZ
, BUS_DMA_ALLOCNOW
,
980 &sc
->re_ldata
.re_tx_list_tag
);
982 device_printf(dev
, "could not allocate dma tag\n");
986 /* Allocate DMA'able memory for the TX ring */
988 error
= bus_dmamem_alloc(sc
->re_ldata
.re_tx_list_tag
,
989 (void **)&sc
->re_ldata
.re_tx_list
, BUS_DMA_WAITOK
| BUS_DMA_ZERO
,
990 &sc
->re_ldata
.re_tx_list_map
);
992 device_printf(dev
, "could not allocate TX ring\n");
996 /* Load the map for the TX ring. */
998 error
= bus_dmamap_load(sc
->re_ldata
.re_tx_list_tag
,
999 sc
->re_ldata
.re_tx_list_map
, sc
->re_ldata
.re_tx_list
,
1000 RE_TX_LIST_SZ
, re_dma_map_addr
,
1001 &sc
->re_ldata
.re_tx_list_addr
, BUS_DMA_NOWAIT
);
1003 device_printf(dev
, "could not get address of TX ring\n");
1007 /* Create DMA maps for TX buffers */
1009 for (i
= 0; i
< RE_TX_DESC_CNT
; i
++) {
1010 error
= bus_dmamap_create(sc
->re_ldata
.re_mtag
, 0,
1011 &sc
->re_ldata
.re_tx_dmamap
[i
]);
1013 device_printf(dev
, "can't create DMA map for TX\n");
1019 * Allocate map for RX descriptor list.
1021 error
= bus_dma_tag_create(sc
->re_parent_tag
, RE_RING_ALIGN
,
1022 0, BUS_SPACE_MAXADDR_32BIT
, BUS_SPACE_MAXADDR
, NULL
,
1023 NULL
, RE_RX_LIST_SZ
, 1, RE_RX_LIST_SZ
, BUS_DMA_ALLOCNOW
,
1024 &sc
->re_ldata
.re_rx_list_tag
);
1026 device_printf(dev
, "could not allocate dma tag\n");
1030 /* Allocate DMA'able memory for the RX ring */
1032 error
= bus_dmamem_alloc(sc
->re_ldata
.re_rx_list_tag
,
1033 (void **)&sc
->re_ldata
.re_rx_list
, BUS_DMA_WAITOK
| BUS_DMA_ZERO
,
1034 &sc
->re_ldata
.re_rx_list_map
);
1036 device_printf(dev
, "could not allocate RX ring\n");
1040 /* Load the map for the RX ring. */
1042 error
= bus_dmamap_load(sc
->re_ldata
.re_rx_list_tag
,
1043 sc
->re_ldata
.re_rx_list_map
, sc
->re_ldata
.re_rx_list
,
1044 RE_RX_LIST_SZ
, re_dma_map_addr
,
1045 &sc
->re_ldata
.re_rx_list_addr
, BUS_DMA_NOWAIT
);
1047 device_printf(dev
, "could not get address of RX ring\n");
1051 /* Create DMA maps for RX buffers */
1053 for (i
= 0; i
< RE_RX_DESC_CNT
; i
++) {
1054 error
= bus_dmamap_create(sc
->re_ldata
.re_mtag
, 0,
1055 &sc
->re_ldata
.re_rx_dmamap
[i
]);
1057 device_printf(dev
, "can't create DMA map for RX\n");
1066 * Attach the interface. Allocate softc structures, do ifmedia
1067 * setup and ethernet/BPF attach.
1070 re_attach(device_t dev
)
1072 struct re_softc
*sc
= device_get_softc(dev
);
1074 const struct re_hwrev
*hw_rev
;
1075 uint8_t eaddr
[ETHER_ADDR_LEN
];
1076 uint16_t as
[ETHER_ADDR_LEN
/ 2];
1077 uint16_t re_did
= 0;
1079 int error
= 0, rid
, i
;
1081 callout_init(&sc
->re_timer
);
1086 RE_ENABLE_TX_MODERATION(sc
);
1088 sysctl_ctx_init(&sc
->re_sysctl_ctx
);
1089 sc
->re_sysctl_tree
= SYSCTL_ADD_NODE(&sc
->re_sysctl_ctx
,
1090 SYSCTL_STATIC_CHILDREN(_hw
),
1092 device_get_nameunit(dev
),
1094 if (sc
->re_sysctl_tree
== NULL
) {
1095 device_printf(dev
, "can't add sysctl node\n");
1099 SYSCTL_ADD_PROC(&sc
->re_sysctl_ctx
,
1100 SYSCTL_CHILDREN(sc
->re_sysctl_tree
),
1101 OID_AUTO
, "tx_moderation",
1102 CTLTYPE_INT
| CTLFLAG_RW
,
1103 sc
, 0, re_sysctl_tx_moderation
, "I",
1104 "Enable/Disable TX moderation");
1106 #ifndef BURN_BRIDGES
1108 * Handle power management nonsense.
1111 if (pci_get_powerstate(dev
) != PCI_POWERSTATE_D0
) {
1112 uint32_t membase
, irq
;
1114 /* Save important PCI config data. */
1115 membase
= pci_read_config(dev
, RE_PCI_LOMEM
, 4);
1116 irq
= pci_read_config(dev
, PCIR_INTLINE
, 4);
1118 /* Reset the power state. */
1119 device_printf(dev
, "chip is in D%d power mode "
1120 "-- setting to D0\n", pci_get_powerstate(dev
));
1122 pci_set_powerstate(dev
, PCI_POWERSTATE_D0
);
1124 /* Restore PCI config data. */
1125 pci_write_config(dev
, RE_PCI_LOMEM
, membase
, 4);
1126 pci_write_config(dev
, PCIR_INTLINE
, irq
, 4);
1130 * Map control/status registers.
1132 pci_enable_busmaster(dev
);
1135 sc
->re_res
= bus_alloc_resource_any(dev
, SYS_RES_IOPORT
, &rid
,
1138 if (sc
->re_res
== NULL
) {
1139 device_printf(dev
, "couldn't map ports\n");
1144 sc
->re_btag
= rman_get_bustag(sc
->re_res
);
1145 sc
->re_bhandle
= rman_get_bushandle(sc
->re_res
);
1147 /* Allocate interrupt */
1149 sc
->re_irq
= bus_alloc_resource_any(dev
, SYS_RES_IRQ
, &rid
,
1150 RF_SHAREABLE
| RF_ACTIVE
);
1152 if (sc
->re_irq
== NULL
) {
1153 device_printf(dev
, "couldn't map interrupt\n");
1158 /* Reset the adapter. */
1161 hwrev
= CSR_READ_4(sc
, RE_TXCFG
) & RE_TXCFG_HWREV
;
1162 for (hw_rev
= re_hwrevs
; hw_rev
->re_desc
!= NULL
; hw_rev
++) {
1163 if (hw_rev
->re_rev
== hwrev
) {
1164 sc
->re_type
= hw_rev
->re_type
;
1165 sc
->re_flags
= hw_rev
->re_flags
;
1171 re_read_eeprom(sc
, (caddr_t
)&re_did
, 0, 1);
1172 if (re_did
!= 0x8129)
1176 * Get station address from the EEPROM.
1178 re_read_eeprom(sc
, (caddr_t
)as
, RE_EE_EADDR
, 3);
1179 for (i
= 0; i
< ETHER_ADDR_LEN
/ 2; i
++)
1180 as
[i
] = le16toh(as
[i
]);
1181 bcopy(as
, eaddr
, sizeof(eaddr
));
1183 if (sc
->re_type
== RE_8169
) {
1184 /* Set RX length mask */
1185 sc
->re_rxlenmask
= RE_RDESC_STAT_GFRAGLEN
;
1186 sc
->re_txstart
= RE_GTXSTART
;
1188 /* Set RX length mask */
1189 sc
->re_rxlenmask
= RE_RDESC_STAT_FRAGLEN
;
1190 sc
->re_txstart
= RE_TXSTART
;
1194 * Allocate the parent bus DMA tag appropriate for PCI.
1196 #define RE_NSEG_NEW 32
1197 error
= bus_dma_tag_create(NULL
, /* parent */
1198 1, 0, /* alignment, boundary */
1199 BUS_SPACE_MAXADDR_32BIT
,/* lowaddr */
1200 BUS_SPACE_MAXADDR
, /* highaddr */
1201 NULL
, NULL
, /* filter, filterarg */
1202 MAXBSIZE
, RE_NSEG_NEW
, /* maxsize, nsegments */
1203 BUS_SPACE_MAXSIZE_32BIT
,/* maxsegsize */
1204 BUS_DMA_ALLOCNOW
, /* flags */
1205 &sc
->re_parent_tag
);
1209 error
= re_allocmem(dev
, sc
);
1215 if (mii_phy_probe(dev
, &sc
->re_miibus
,
1216 re_ifmedia_upd
, re_ifmedia_sts
)) {
1217 device_printf(dev
, "MII without any phy!\n");
1222 ifp
= &sc
->arpcom
.ac_if
;
1224 if_initname(ifp
, device_get_name(dev
), device_get_unit(dev
));
1225 ifp
->if_mtu
= ETHERMTU
;
1226 ifp
->if_flags
= IFF_BROADCAST
| IFF_SIMPLEX
| IFF_MULTICAST
;
1227 ifp
->if_ioctl
= re_ioctl
;
1228 ifp
->if_start
= re_start
;
1229 ifp
->if_capabilities
= IFCAP_VLAN_MTU
| IFCAP_VLAN_HWTAGGING
;
1230 if (hwrev
!= RE_HWREV_8168C
) /* XXX does not work yet */
1231 ifp
->if_capabilities
|= IFCAP_HWCSUM
;
1232 #ifdef DEVICE_POLLING
1233 ifp
->if_poll
= re_poll
;
1235 ifp
->if_watchdog
= re_watchdog
;
1236 ifp
->if_init
= re_init
;
1237 if (sc
->re_type
== RE_8169
)
1238 ifp
->if_baudrate
= 1000000000;
1240 ifp
->if_baudrate
= 100000000;
1241 ifq_set_maxlen(&ifp
->if_snd
, RE_IFQ_MAXLEN
);
1242 ifq_set_ready(&ifp
->if_snd
);
1244 #ifdef RE_DISABLE_HWCSUM
1245 ifp
->if_capenable
= ifp
->if_capabilities
& ~IFCAP_HWCSUM
;
1246 ifp
->if_hwassist
= 0;
1248 ifp
->if_capenable
= ifp
->if_capabilities
;
1249 if (ifp
->if_capabilities
& IFCAP_HWCSUM
)
1250 ifp
->if_hwassist
= RE_CSUM_FEATURES
;
1252 ifp
->if_hwassist
= 0;
1253 #endif /* RE_DISABLE_HWCSUM */
1256 * Call MI attach routine.
1258 ether_ifattach(ifp
, eaddr
, NULL
);
1262 * Perform hardware diagnostic on the original RTL8169.
1263 * Some 32-bit cards were incorrectly wired and would
1264 * malfunction if plugged into a 64-bit slot.
1266 if (hwrev
== RE_HWREV_8169
) {
1267 lwkt_serialize_enter(ifp
->if_serializer
);
1268 error
= re_diag(sc
);
1269 lwkt_serialize_exit(ifp
->if_serializer
);
1272 device_printf(dev
, "hardware diagnostic failure\n");
1273 ether_ifdetach(ifp
);
1277 #endif /* RE_DIAG */
1279 /* Hook interrupt last to avoid having to lock softc */
1280 error
= bus_setup_intr(dev
, sc
->re_irq
, INTR_NETSAFE
, re_intr
, sc
,
1281 &sc
->re_intrhand
, ifp
->if_serializer
);
1284 device_printf(dev
, "couldn't set up irq\n");
1285 ether_ifdetach(ifp
);
1289 ifp
->if_cpuid
= ithread_cpuid(rman_get_start(sc
->re_irq
));
1290 KKASSERT(ifp
->if_cpuid
>= 0 && ifp
->if_cpuid
< ncpus
);
1300 * Shutdown hardware and free up resources. This can be called any
1301 * time after the mutex has been initialized. It is called in both
1302 * the error case in attach and the normal detach case so it needs
1303 * to be careful about only freeing resources that have actually been
1307 re_detach(device_t dev
)
1309 struct re_softc
*sc
= device_get_softc(dev
);
1310 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1313 /* These should only be active if attach succeeded */
1314 if (device_is_attached(dev
)) {
1315 lwkt_serialize_enter(ifp
->if_serializer
);
1317 bus_teardown_intr(dev
, sc
->re_irq
, sc
->re_intrhand
);
1318 lwkt_serialize_exit(ifp
->if_serializer
);
1320 ether_ifdetach(ifp
);
1323 device_delete_child(dev
, sc
->re_miibus
);
1324 bus_generic_detach(dev
);
1327 bus_release_resource(dev
, SYS_RES_IRQ
, 0, sc
->re_irq
);
1329 bus_release_resource(dev
, SYS_RES_IOPORT
, RE_PCI_LOIO
,
1333 /* Unload and free the RX DMA ring memory and map */
1335 if (sc
->re_ldata
.re_rx_list_tag
) {
1336 bus_dmamap_unload(sc
->re_ldata
.re_rx_list_tag
,
1337 sc
->re_ldata
.re_rx_list_map
);
1338 bus_dmamem_free(sc
->re_ldata
.re_rx_list_tag
,
1339 sc
->re_ldata
.re_rx_list
,
1340 sc
->re_ldata
.re_rx_list_map
);
1341 bus_dma_tag_destroy(sc
->re_ldata
.re_rx_list_tag
);
1344 /* Unload and free the TX DMA ring memory and map */
1346 if (sc
->re_ldata
.re_tx_list_tag
) {
1347 bus_dmamap_unload(sc
->re_ldata
.re_tx_list_tag
,
1348 sc
->re_ldata
.re_tx_list_map
);
1349 bus_dmamem_free(sc
->re_ldata
.re_tx_list_tag
,
1350 sc
->re_ldata
.re_tx_list
,
1351 sc
->re_ldata
.re_tx_list_map
);
1352 bus_dma_tag_destroy(sc
->re_ldata
.re_tx_list_tag
);
1355 /* Destroy all the RX and TX buffer maps */
1357 if (sc
->re_ldata
.re_mtag
) {
1358 for (i
= 0; i
< RE_TX_DESC_CNT
; i
++)
1359 bus_dmamap_destroy(sc
->re_ldata
.re_mtag
,
1360 sc
->re_ldata
.re_tx_dmamap
[i
]);
1361 for (i
= 0; i
< RE_RX_DESC_CNT
; i
++)
1362 bus_dmamap_destroy(sc
->re_ldata
.re_mtag
,
1363 sc
->re_ldata
.re_rx_dmamap
[i
]);
1364 bus_dma_tag_destroy(sc
->re_ldata
.re_mtag
);
1367 /* Unload and free the stats buffer and map */
1369 if (sc
->re_ldata
.re_stag
) {
1370 bus_dmamap_unload(sc
->re_ldata
.re_stag
,
1371 sc
->re_ldata
.re_rx_list_map
);
1372 bus_dmamem_free(sc
->re_ldata
.re_stag
,
1373 sc
->re_ldata
.re_stats
,
1374 sc
->re_ldata
.re_smap
);
1375 bus_dma_tag_destroy(sc
->re_ldata
.re_stag
);
1378 if (sc
->re_parent_tag
)
1379 bus_dma_tag_destroy(sc
->re_parent_tag
);
1385 re_newbuf(struct re_softc
*sc
, int idx
, struct mbuf
*m
)
1387 struct re_dmaload_arg arg
;
1388 struct mbuf
*n
= NULL
;
1392 n
= m_getcl(MB_DONTWAIT
, MT_DATA
, M_PKTHDR
);
1397 m
->m_data
= m
->m_ext
.ext_buf
;
1399 m
->m_len
= m
->m_pkthdr
.len
= MCLBYTES
;
1403 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1404 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1411 arg
.re_ring
= sc
->re_ldata
.re_rx_list
;
1413 error
= bus_dmamap_load_mbuf(sc
->re_ldata
.re_mtag
,
1414 sc
->re_ldata
.re_rx_dmamap
[idx
], m
, re_dma_map_desc
,
1415 &arg
, BUS_DMA_NOWAIT
);
1416 if (error
|| arg
.re_maxsegs
!= 1) {
1422 sc
->re_ldata
.re_rx_list
[idx
].re_cmdstat
|= htole32(RE_RDESC_CMD_OWN
);
1423 sc
->re_ldata
.re_rx_mbuf
[idx
] = m
;
1425 bus_dmamap_sync(sc
->re_ldata
.re_mtag
, sc
->re_ldata
.re_rx_dmamap
[idx
],
1426 BUS_DMASYNC_PREREAD
);
1432 re_tx_list_init(struct re_softc
*sc
)
1434 bzero(sc
->re_ldata
.re_tx_list
, RE_TX_LIST_SZ
);
1435 bzero(&sc
->re_ldata
.re_tx_mbuf
, RE_TX_DESC_CNT
* sizeof(struct mbuf
*));
1437 bus_dmamap_sync(sc
->re_ldata
.re_tx_list_tag
,
1438 sc
->re_ldata
.re_tx_list_map
, BUS_DMASYNC_PREWRITE
);
1439 sc
->re_ldata
.re_tx_prodidx
= 0;
1440 sc
->re_ldata
.re_tx_considx
= 0;
1441 sc
->re_ldata
.re_tx_free
= RE_TX_DESC_CNT
;
1447 re_rx_list_init(struct re_softc
*sc
)
1451 bzero(sc
->re_ldata
.re_rx_list
, RE_RX_LIST_SZ
);
1452 bzero(&sc
->re_ldata
.re_rx_mbuf
, RE_RX_DESC_CNT
* sizeof(struct mbuf
*));
1454 for (i
= 0; i
< RE_RX_DESC_CNT
; i
++) {
1455 error
= re_newbuf(sc
, i
, NULL
);
1460 /* Flush the RX descriptors */
1462 bus_dmamap_sync(sc
->re_ldata
.re_rx_list_tag
,
1463 sc
->re_ldata
.re_rx_list_map
, BUS_DMASYNC_PREWRITE
);
1465 sc
->re_ldata
.re_rx_prodidx
= 0;
1466 sc
->re_head
= sc
->re_tail
= NULL
;
1472 * RX handler for C+ and 8169. For the gigE chips, we support
1473 * the reception of jumbo frames that have been fragmented
1474 * across multiple 2K mbuf cluster buffers.
1477 re_rxeof(struct re_softc
*sc
)
1479 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1481 struct re_desc
*cur_rx
;
1482 uint32_t rxstat
, rxvlan
;
1485 /* Invalidate the descriptor memory */
1487 bus_dmamap_sync(sc
->re_ldata
.re_rx_list_tag
,
1488 sc
->re_ldata
.re_rx_list_map
, BUS_DMASYNC_POSTREAD
);
1490 for (i
= sc
->re_ldata
.re_rx_prodidx
;
1491 RE_OWN(&sc
->re_ldata
.re_rx_list
[i
]) == 0 ; RE_DESC_INC(i
)) {
1492 cur_rx
= &sc
->re_ldata
.re_rx_list
[i
];
1493 m
= sc
->re_ldata
.re_rx_mbuf
[i
];
1494 total_len
= RE_RXBYTES(cur_rx
);
1495 rxstat
= le32toh(cur_rx
->re_cmdstat
);
1496 rxvlan
= le32toh(cur_rx
->re_vlanctl
);
1498 /* Invalidate the RX mbuf and unload its map */
1500 bus_dmamap_sync(sc
->re_ldata
.re_mtag
,
1501 sc
->re_ldata
.re_rx_dmamap
[i
],
1502 BUS_DMASYNC_POSTWRITE
);
1503 bus_dmamap_unload(sc
->re_ldata
.re_mtag
,
1504 sc
->re_ldata
.re_rx_dmamap
[i
]);
1506 if ((rxstat
& RE_RDESC_STAT_EOF
) == 0) {
1507 m
->m_len
= MCLBYTES
- ETHER_ALIGN
;
1508 if (sc
->re_head
== NULL
) {
1509 sc
->re_head
= sc
->re_tail
= m
;
1511 sc
->re_tail
->m_next
= m
;
1514 re_newbuf(sc
, i
, NULL
);
1519 * NOTE: for the 8139C+, the frame length field
1520 * is always 12 bits in size, but for the gigE chips,
1521 * it is 13 bits (since the max RX frame length is 16K).
1522 * Unfortunately, all 32 bits in the status word
1523 * were already used, so to make room for the extra
1524 * length bit, RealTek took out the 'frame alignment
1525 * error' bit and shifted the other status bits
1526 * over one slot. The OWN, EOR, FS and LS bits are
1527 * still in the same places. We have already extracted
1528 * the frame length and checked the OWN bit, so rather
1529 * than using an alternate bit mapping, we shift the
1530 * status bits one space to the right so we can evaluate
1531 * them using the 8169 status as though it was in the
1532 * same format as that of the 8139C+.
1534 if (sc
->re_type
== RE_8169
)
1537 if (rxstat
& RE_RDESC_STAT_RXERRSUM
) {
1540 * If this is part of a multi-fragment packet,
1541 * discard all the pieces.
1543 if (sc
->re_head
!= NULL
) {
1544 m_freem(sc
->re_head
);
1545 sc
->re_head
= sc
->re_tail
= NULL
;
1547 re_newbuf(sc
, i
, m
);
1552 * If allocating a replacement mbuf fails,
1553 * reload the current one.
1556 if (re_newbuf(sc
, i
, NULL
)) {
1558 if (sc
->re_head
!= NULL
) {
1559 m_freem(sc
->re_head
);
1560 sc
->re_head
= sc
->re_tail
= NULL
;
1562 re_newbuf(sc
, i
, m
);
1566 if (sc
->re_head
!= NULL
) {
1567 m
->m_len
= total_len
% (MCLBYTES
- ETHER_ALIGN
);
1569 * Special case: if there's 4 bytes or less
1570 * in this buffer, the mbuf can be discarded:
1571 * the last 4 bytes is the CRC, which we don't
1572 * care about anyway.
1574 if (m
->m_len
<= ETHER_CRC_LEN
) {
1575 sc
->re_tail
->m_len
-=
1576 (ETHER_CRC_LEN
- m
->m_len
);
1579 m
->m_len
-= ETHER_CRC_LEN
;
1580 sc
->re_tail
->m_next
= m
;
1583 sc
->re_head
= sc
->re_tail
= NULL
;
1584 m
->m_pkthdr
.len
= total_len
- ETHER_CRC_LEN
;
1586 m
->m_pkthdr
.len
= m
->m_len
=
1587 (total_len
- ETHER_CRC_LEN
);
1590 m
->m_pkthdr
.rcvif
= ifp
;
1592 /* Do RX checksumming if enabled */
1594 if (ifp
->if_capenable
& IFCAP_RXCSUM
) {
1596 /* Check IP header checksum */
1597 if (rxstat
& RE_RDESC_STAT_PROTOID
)
1598 m
->m_pkthdr
.csum_flags
|= CSUM_IP_CHECKED
;
1599 if ((rxstat
& RE_RDESC_STAT_IPSUMBAD
) == 0)
1600 m
->m_pkthdr
.csum_flags
|= CSUM_IP_VALID
;
1602 /* Check TCP/UDP checksum */
1603 if ((RE_TCPPKT(rxstat
) &&
1604 (rxstat
& RE_RDESC_STAT_TCPSUMBAD
) == 0) ||
1605 (RE_UDPPKT(rxstat
) &&
1606 (rxstat
& RE_RDESC_STAT_UDPSUMBAD
)) == 0) {
1607 m
->m_pkthdr
.csum_flags
|=
1608 CSUM_DATA_VALID
|CSUM_PSEUDO_HDR
|
1609 CSUM_FRAG_NOT_CHECKED
;
1610 m
->m_pkthdr
.csum_data
= 0xffff;
1614 if (rxvlan
& RE_RDESC_VLANCTL_TAG
) {
1615 m
->m_flags
|= M_VLANTAG
;
1616 m
->m_pkthdr
.ether_vlantag
=
1617 be16toh((rxvlan
& RE_RDESC_VLANCTL_DATA
));
1619 ifp
->if_input(ifp
, m
);
1622 /* Flush the RX DMA ring */
1624 bus_dmamap_sync(sc
->re_ldata
.re_rx_list_tag
,
1625 sc
->re_ldata
.re_rx_list_map
, BUS_DMASYNC_PREWRITE
);
1627 sc
->re_ldata
.re_rx_prodidx
= i
;
1631 re_txeof(struct re_softc
*sc
)
1633 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1637 /* Invalidate the TX descriptor list */
1639 bus_dmamap_sync(sc
->re_ldata
.re_tx_list_tag
,
1640 sc
->re_ldata
.re_tx_list_map
, BUS_DMASYNC_POSTREAD
);
1642 for (idx
= sc
->re_ldata
.re_tx_considx
;
1643 sc
->re_ldata
.re_tx_free
< RE_TX_DESC_CNT
; RE_DESC_INC(idx
)) {
1644 txstat
= le32toh(sc
->re_ldata
.re_tx_list
[idx
].re_cmdstat
);
1645 if (txstat
& RE_TDESC_CMD_OWN
)
1648 sc
->re_ldata
.re_tx_list
[idx
].re_bufaddr_lo
= 0;
1651 * We only stash mbufs in the last descriptor
1652 * in a fragment chain, which also happens to
1653 * be the only place where the TX status bits
1656 if (txstat
& RE_TDESC_CMD_EOF
) {
1657 m_freem(sc
->re_ldata
.re_tx_mbuf
[idx
]);
1658 sc
->re_ldata
.re_tx_mbuf
[idx
] = NULL
;
1659 bus_dmamap_unload(sc
->re_ldata
.re_mtag
,
1660 sc
->re_ldata
.re_tx_dmamap
[idx
]);
1661 if (txstat
& (RE_TDESC_STAT_EXCESSCOL
|
1662 RE_TDESC_STAT_COLCNT
))
1663 ifp
->if_collisions
++;
1664 if (txstat
& RE_TDESC_STAT_TXERRSUM
)
1669 sc
->re_ldata
.re_tx_free
++;
1672 /* No changes made to the TX ring, so no flush needed */
1673 if (sc
->re_ldata
.re_tx_free
) {
1674 sc
->re_ldata
.re_tx_considx
= idx
;
1675 ifp
->if_flags
&= ~IFF_OACTIVE
;
1680 * Some chips will ignore a second TX request issued while an
1681 * existing transmission is in progress. If the transmitter goes
1682 * idle but there are still packets waiting to be sent, we need
1683 * to restart the channel here to flush them out. This only seems
1684 * to be required with the PCIe devices.
1686 if (sc
->re_ldata
.re_tx_free
< RE_TX_DESC_CNT
)
1687 CSR_WRITE_1(sc
, sc
->re_txstart
, RE_TXSTART_START
);
1690 * If not all descriptors have been released reaped yet,
1691 * reload the timer so that we will eventually get another
1692 * interrupt that will cause us to re-enter this routine.
1693 * This is done in case the transmitter has gone idle.
1695 if (RE_TX_MODERATION_IS_ENABLED(sc
) &&
1696 sc
->re_ldata
.re_tx_free
< RE_TX_DESC_CNT
)
1697 CSR_WRITE_4(sc
, RE_TIMERCNT
, 1);
1703 struct re_softc
*sc
= xsc
;
1705 lwkt_serialize_enter(sc
->arpcom
.ac_if
.if_serializer
);
1706 re_tick_serialized(xsc
);
1707 lwkt_serialize_exit(sc
->arpcom
.ac_if
.if_serializer
);
1711 re_tick_serialized(void *xsc
)
1713 struct re_softc
*sc
= xsc
;
1714 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1715 struct mii_data
*mii
;
1717 mii
= device_get_softc(sc
->re_miibus
);
1720 if (!(mii
->mii_media_status
& IFM_ACTIVE
))
1723 if (mii
->mii_media_status
& IFM_ACTIVE
&&
1724 IFM_SUBTYPE(mii
->mii_media_active
) != IFM_NONE
) {
1726 if (!ifq_is_empty(&ifp
->if_snd
))
1731 callout_reset(&sc
->re_timer
, hz
, re_tick
, sc
);
1734 #ifdef DEVICE_POLLING
1737 re_poll(struct ifnet
*ifp
, enum poll_cmd cmd
, int count
)
1739 struct re_softc
*sc
= ifp
->if_softc
;
1743 /* disable interrupts */
1744 CSR_WRITE_2(sc
, RE_IMR
, 0x0000);
1746 case POLL_DEREGISTER
:
1747 /* enable interrupts */
1748 CSR_WRITE_2(sc
, RE_IMR
, sc
->re_intrs
);
1751 sc
->rxcycles
= count
;
1755 if (!ifq_is_empty(&ifp
->if_snd
))
1758 if (cmd
== POLL_AND_CHECK_STATUS
) { /* also check status register */
1761 status
= CSR_READ_2(sc
, RE_ISR
);
1762 if (status
== 0xffff)
1765 CSR_WRITE_2(sc
, RE_ISR
, status
);
1768 * XXX check behaviour on receiver stalls.
1771 if (status
& RE_ISR_SYSTEM_ERR
) {
1779 #endif /* DEVICE_POLLING */
1784 struct re_softc
*sc
= arg
;
1785 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1788 if (sc
->suspended
|| (ifp
->if_flags
& IFF_UP
) == 0)
1792 status
= CSR_READ_2(sc
, RE_ISR
);
1793 /* If the card has gone away the read returns 0xffff. */
1794 if (status
== 0xffff)
1797 CSR_WRITE_2(sc
, RE_ISR
, status
);
1799 if ((status
& sc
->re_intrs
) == 0)
1802 if (status
& (RE_ISR_RX_OK
| RE_ISR_RX_ERR
| RE_ISR_FIFO_OFLOW
))
1805 if ((status
& sc
->re_tx_ack
) ||
1806 (status
& RE_ISR_TX_ERR
) ||
1807 (status
& RE_ISR_TX_DESC_UNAVAIL
))
1810 if (status
& RE_ISR_SYSTEM_ERR
) {
1815 if (status
& RE_ISR_LINKCHG
) {
1816 callout_stop(&sc
->re_timer
);
1817 re_tick_serialized(sc
);
1821 if (!ifq_is_empty(&ifp
->if_snd
))
1826 re_encap(struct re_softc
*sc
, struct mbuf
**m_head
, int *idx
, int *called_defrag
)
1828 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
1829 struct mbuf
*m
, *m_new
= NULL
;
1830 struct re_dmaload_arg arg
;
1834 KASSERT(sc
->re_ldata
.re_tx_free
> 4, ("not enough free TX desc\n"));
1840 * Set up checksum offload. Note: checksum offload bits must
1841 * appear in all descriptors of a multi-descriptor transmit
1842 * attempt. (This is according to testing done with an 8169
1843 * chip. I'm not sure if this is a requirement or a bug.)
1848 if (m
->m_pkthdr
.csum_flags
& CSUM_IP
)
1849 arg
.re_flags
|= RE_TDESC_CMD_IPCSUM
;
1850 if (m
->m_pkthdr
.csum_flags
& CSUM_TCP
)
1851 arg
.re_flags
|= RE_TDESC_CMD_TCPCSUM
;
1852 if (m
->m_pkthdr
.csum_flags
& CSUM_UDP
)
1853 arg
.re_flags
|= RE_TDESC_CMD_UDPCSUM
;
1857 arg
.re_maxsegs
= sc
->re_ldata
.re_tx_free
;
1858 if (arg
.re_maxsegs
> 4)
1859 arg
.re_maxsegs
-= 4;
1860 arg
.re_ring
= sc
->re_ldata
.re_tx_list
;
1862 map
= sc
->re_ldata
.re_tx_dmamap
[*idx
];
1865 * With some of the RealTek chips, using the checksum offload
1866 * support in conjunction with the autopadding feature results
1867 * in the transmission of corrupt frames. For example, if we
1868 * need to send a really small IP fragment that's less than 60
1869 * bytes in size, and IP header checksumming is enabled, the
1870 * resulting ethernet frame that appears on the wire will
1871 * have garbled payload. To work around this, if TX checksum
1872 * offload is enabled, we always manually pad short frames out
1873 * to the minimum ethernet frame size. We do this by pretending
1874 * the mbuf chain has too many fragments so the coalescing code
1875 * below can assemble the packet into a single buffer that's
1876 * padded out to the mininum frame size.
1878 * Note: this appears unnecessary for TCP, and doing it for TCP
1879 * with PCIe adapters seems to result in bad checksums.
1881 if (arg
.re_flags
&& !(arg
.re_flags
& RE_TDESC_CMD_TCPCSUM
) &&
1882 m
->m_pkthdr
.len
< RE_MIN_FRAMELEN
) {
1885 error
= bus_dmamap_load_mbuf(sc
->re_ldata
.re_mtag
, map
,
1886 m
, re_dma_map_desc
, &arg
, BUS_DMA_NOWAIT
);
1889 if (error
&& error
!= EFBIG
) {
1890 if_printf(ifp
, "can't map mbuf (error %d)\n", error
);
1894 /* Too many segments to map, coalesce into a single mbuf */
1896 if (error
|| arg
.re_maxsegs
== 0) {
1897 m_new
= m_defrag_nofree(m
, MB_DONTWAIT
);
1898 if (m_new
== NULL
) {
1906 * Manually pad short frames, and zero the pad space
1907 * to avoid leaking data.
1909 if (m_new
->m_pkthdr
.len
< RE_MIN_FRAMELEN
) {
1910 bzero(mtod(m_new
, char *) + m_new
->m_pkthdr
.len
,
1911 RE_MIN_FRAMELEN
- m_new
->m_pkthdr
.len
);
1912 m_new
->m_pkthdr
.len
+= RE_MIN_FRAMELEN
-
1913 m_new
->m_pkthdr
.len
;
1914 m_new
->m_len
= m_new
->m_pkthdr
.len
;
1920 arg
.re_maxsegs
= sc
->re_ldata
.re_tx_free
;
1921 arg
.re_ring
= sc
->re_ldata
.re_tx_list
;
1923 error
= bus_dmamap_load_mbuf(sc
->re_ldata
.re_mtag
, map
,
1924 m
, re_dma_map_desc
, &arg
, BUS_DMA_NOWAIT
);
1927 if_printf(ifp
, "can't map mbuf (error %d)\n", error
);
1933 * Insure that the map for this transmission
1934 * is placed at the array index of the last descriptor
1937 sc
->re_ldata
.re_tx_dmamap
[*idx
] =
1938 sc
->re_ldata
.re_tx_dmamap
[arg
.re_idx
];
1939 sc
->re_ldata
.re_tx_dmamap
[arg
.re_idx
] = map
;
1941 sc
->re_ldata
.re_tx_mbuf
[arg
.re_idx
] = m
;
1942 sc
->re_ldata
.re_tx_free
-= arg
.re_maxsegs
;
1945 * Set up hardware VLAN tagging. Note: vlan tag info must
1946 * appear in the first descriptor of a multi-descriptor
1947 * transmission attempt.
1950 if (m
->m_flags
& M_VLANTAG
) {
1951 sc
->re_ldata
.re_tx_list
[*idx
].re_vlanctl
=
1952 htole32(htobe16(m
->m_pkthdr
.ether_vlantag
) |
1953 RE_TDESC_VLANCTL_TAG
);
1956 /* Transfer ownership of packet to the chip. */
1958 sc
->re_ldata
.re_tx_list
[arg
.re_idx
].re_cmdstat
|=
1959 htole32(RE_TDESC_CMD_OWN
);
1960 if (*idx
!= arg
.re_idx
)
1961 sc
->re_ldata
.re_tx_list
[*idx
].re_cmdstat
|=
1962 htole32(RE_TDESC_CMD_OWN
);
1964 RE_DESC_INC(arg
.re_idx
);
1971 * Main transmit routine for C+ and gigE NICs.
1975 re_start(struct ifnet
*ifp
)
1977 struct re_softc
*sc
= ifp
->if_softc
;
1978 struct mbuf
*m_head
;
1979 struct mbuf
*m_head2
;
1980 int called_defrag
, idx
, need_trans
;
1983 ifq_purge(&ifp
->if_snd
);
1987 if ((ifp
->if_flags
& (IFF_OACTIVE
| IFF_RUNNING
)) != IFF_RUNNING
)
1990 idx
= sc
->re_ldata
.re_tx_prodidx
;
1993 while (sc
->re_ldata
.re_tx_mbuf
[idx
] == NULL
) {
1994 if (sc
->re_ldata
.re_tx_free
<= 4) {
1995 ifp
->if_flags
|= IFF_OACTIVE
;
1999 m_head
= ifq_dequeue(&ifp
->if_snd
, NULL
);
2004 if (re_encap(sc
, &m_head2
, &idx
, &called_defrag
)) {
2006 * If we could not encapsulate the defragged packet,
2007 * the returned m_head2 is garbage and we must dequeue
2008 * and throw away the original packet.
2012 ifp
->if_flags
|= IFF_OACTIVE
;
2017 * Clean out the packet we encapsulated. If we defragged
2018 * the packet the m_head2 is the one that got encapsulated
2019 * and the original must be thrown away. Otherwise m_head2
2020 * *IS* the original.
2027 * If there's a BPF listener, bounce a copy of this frame
2030 ETHER_BPF_MTAP(ifp
, m_head2
);
2034 if (RE_TX_MODERATION_IS_ENABLED(sc
) &&
2035 sc
->re_ldata
.re_tx_free
!= RE_TX_DESC_CNT
)
2036 CSR_WRITE_4(sc
, RE_TIMERCNT
, 1);
2040 /* Flush the TX descriptors */
2041 bus_dmamap_sync(sc
->re_ldata
.re_tx_list_tag
,
2042 sc
->re_ldata
.re_tx_list_map
, BUS_DMASYNC_PREWRITE
);
2044 sc
->re_ldata
.re_tx_prodidx
= idx
;
2047 * RealTek put the TX poll request register in a different
2048 * location on the 8169 gigE chip. I don't know why.
2050 CSR_WRITE_1(sc
, sc
->re_txstart
, RE_TXSTART_START
);
2052 if (RE_TX_MODERATION_IS_ENABLED(sc
)) {
2054 * Use the countdown timer for interrupt moderation.
2055 * 'TX done' interrupts are disabled. Instead, we reset the
2056 * countdown timer, which will begin counting until it hits
2057 * the value in the TIMERINT register, and then trigger an
2058 * interrupt. Each time we write to the TIMERCNT register,
2059 * the timer count is reset to 0.
2061 CSR_WRITE_4(sc
, RE_TIMERCNT
, 1);
2065 * Set a timeout in case the chip goes out to lunch.
2073 struct re_softc
*sc
= xsc
;
2074 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
2075 struct mii_data
*mii
;
2078 mii
= device_get_softc(sc
->re_miibus
);
2081 * Cancel pending I/O and free all RX/TX buffers.
2086 * Enable C+ RX and TX mode, as well as VLAN stripping and
2087 * RX checksum offload. We must configure the C+ register
2088 * before all others.
2090 CSR_WRITE_2(sc
, RE_CPLUS_CMD
, RE_CPLUSCMD_RXENB
| RE_CPLUSCMD_TXENB
|
2091 RE_CPLUSCMD_PCI_MRW
| RE_CPLUSCMD_VLANSTRIP
|
2092 (ifp
->if_capenable
& IFCAP_RXCSUM
?
2093 RE_CPLUSCMD_RXCSUM_ENB
: 0));
2096 * Init our MAC address. Even though the chipset
2097 * documentation doesn't mention it, we need to enter "Config
2098 * register write enable" mode to modify the ID registers.
2100 CSR_WRITE_1(sc
, RE_EECMD
, RE_EEMODE_WRITECFG
);
2101 CSR_WRITE_4(sc
, RE_IDR0
,
2102 htole32(*(uint32_t *)(&sc
->arpcom
.ac_enaddr
[0])));
2103 CSR_WRITE_2(sc
, RE_IDR4
,
2104 htole16(*(uint16_t *)(&sc
->arpcom
.ac_enaddr
[4])));
2105 CSR_WRITE_1(sc
, RE_EECMD
, RE_EEMODE_OFF
);
2108 * For C+ mode, initialize the RX descriptors and mbufs.
2110 re_rx_list_init(sc
);
2111 re_tx_list_init(sc
);
2114 * Load the addresses of the RX and TX lists into the chip.
2116 CSR_WRITE_4(sc
, RE_RXLIST_ADDR_HI
,
2117 RE_ADDR_HI(sc
->re_ldata
.re_rx_list_addr
));
2118 CSR_WRITE_4(sc
, RE_RXLIST_ADDR_LO
,
2119 RE_ADDR_LO(sc
->re_ldata
.re_rx_list_addr
));
2121 CSR_WRITE_4(sc
, RE_TXLIST_ADDR_HI
,
2122 RE_ADDR_HI(sc
->re_ldata
.re_tx_list_addr
));
2123 CSR_WRITE_4(sc
, RE_TXLIST_ADDR_LO
,
2124 RE_ADDR_LO(sc
->re_ldata
.re_tx_list_addr
));
2127 * Enable transmit and receive.
2129 CSR_WRITE_1(sc
, RE_COMMAND
, RE_CMD_TX_ENB
|RE_CMD_RX_ENB
);
2132 * Set the initial TX and RX configuration.
2134 if (sc
->re_testmode
) {
2135 if (sc
->re_type
== RE_8169
)
2136 CSR_WRITE_4(sc
, RE_TXCFG
,
2137 RE_TXCFG_CONFIG
| RE_LOOPTEST_ON
);
2139 CSR_WRITE_4(sc
, RE_TXCFG
,
2140 RE_TXCFG_CONFIG
| RE_LOOPTEST_ON_CPLUS
);
2142 CSR_WRITE_4(sc
, RE_TXCFG
, RE_TXCFG_CONFIG
);
2144 CSR_WRITE_1(sc
, RE_EARLY_TX_THRESH
, 16);
2146 CSR_WRITE_4(sc
, RE_RXCFG
, RE_RXCFG_CONFIG
);
2148 /* Set the individual bit to receive frames for this host only. */
2149 rxcfg
= CSR_READ_4(sc
, RE_RXCFG
);
2150 rxcfg
|= RE_RXCFG_RX_INDIV
;
2152 /* If we want promiscuous mode, set the allframes bit. */
2153 if (ifp
->if_flags
& IFF_PROMISC
) {
2154 rxcfg
|= RE_RXCFG_RX_ALLPHYS
;
2155 CSR_WRITE_4(sc
, RE_RXCFG
, rxcfg
);
2157 rxcfg
&= ~RE_RXCFG_RX_ALLPHYS
;
2158 CSR_WRITE_4(sc
, RE_RXCFG
, rxcfg
);
2162 * Set capture broadcast bit to capture broadcast frames.
2164 if (ifp
->if_flags
& IFF_BROADCAST
) {
2165 rxcfg
|= RE_RXCFG_RX_BROAD
;
2166 CSR_WRITE_4(sc
, RE_RXCFG
, rxcfg
);
2168 rxcfg
&= ~RE_RXCFG_RX_BROAD
;
2169 CSR_WRITE_4(sc
, RE_RXCFG
, rxcfg
);
2173 * Program the multicast filter, if necessary.
2177 #ifdef DEVICE_POLLING
2179 * Disable interrupts if we are polling.
2181 if (ifp
->if_flags
& IFF_POLLING
)
2182 CSR_WRITE_2(sc
, RE_IMR
, 0);
2183 else /* otherwise ... */
2184 #endif /* DEVICE_POLLING */
2186 * Enable interrupts.
2188 if (sc
->re_testmode
)
2189 CSR_WRITE_2(sc
, RE_IMR
, 0);
2191 CSR_WRITE_2(sc
, RE_IMR
, sc
->re_intrs
);
2192 CSR_WRITE_2(sc
, RE_ISR
, sc
->re_intrs
);
2194 /* Set initial TX threshold */
2195 sc
->re_txthresh
= RE_TX_THRESH_INIT
;
2197 /* Start RX/TX process. */
2198 if (sc
->re_flags
& RE_F_HASMPC
)
2199 CSR_WRITE_4(sc
, RE_MISSEDPKT
, 0);
2201 /* Enable receiver and transmitter. */
2202 CSR_WRITE_1(sc
, RE_COMMAND
, RE_CMD_TX_ENB
|RE_CMD_RX_ENB
);
2205 if (RE_TX_MODERATION_IS_ENABLED(sc
)) {
2207 * Initialize the timer interrupt register so that
2208 * a timer interrupt will be generated once the timer
2209 * reaches a certain number of ticks. The timer is
2210 * reloaded on each transmit. This gives us TX interrupt
2211 * moderation, which dramatically improves TX frame rate.
2213 if (sc
->re_type
== RE_8169
)
2214 CSR_WRITE_4(sc
, RE_TIMERINT_8169
, 0x800);
2216 CSR_WRITE_4(sc
, RE_TIMERINT
, 0x400);
2220 * For 8169 gigE NICs, set the max allowed RX packet
2221 * size so we can receive jumbo frames.
2223 if (sc
->re_type
== RE_8169
)
2224 CSR_WRITE_2(sc
, RE_MAXRXPKTLEN
, 16383);
2226 if (sc
->re_testmode
) {
2232 CSR_WRITE_1(sc
, RE_CFG1
, RE_CFG1_DRVLOAD
|RE_CFG1_FULLDUPLEX
);
2234 ifp
->if_flags
|= IFF_RUNNING
;
2235 ifp
->if_flags
&= ~IFF_OACTIVE
;
2238 callout_reset(&sc
->re_timer
, hz
, re_tick
, sc
);
2242 * Set media options.
2245 re_ifmedia_upd(struct ifnet
*ifp
)
2247 struct re_softc
*sc
= ifp
->if_softc
;
2248 struct mii_data
*mii
;
2250 mii
= device_get_softc(sc
->re_miibus
);
2257 * Report current media status.
2260 re_ifmedia_sts(struct ifnet
*ifp
, struct ifmediareq
*ifmr
)
2262 struct re_softc
*sc
= ifp
->if_softc
;
2263 struct mii_data
*mii
;
2265 mii
= device_get_softc(sc
->re_miibus
);
2268 ifmr
->ifm_active
= mii
->mii_media_active
;
2269 ifmr
->ifm_status
= mii
->mii_media_status
;
2273 re_ioctl(struct ifnet
*ifp
, u_long command
, caddr_t data
, struct ucred
*cr
)
2275 struct re_softc
*sc
= ifp
->if_softc
;
2276 struct ifreq
*ifr
= (struct ifreq
*) data
;
2277 struct mii_data
*mii
;
2282 if (ifr
->ifr_mtu
> RE_JUMBO_MTU
)
2284 ifp
->if_mtu
= ifr
->ifr_mtu
;
2287 if (ifp
->if_flags
& IFF_UP
)
2289 else if (ifp
->if_flags
& IFF_RUNNING
)
2299 mii
= device_get_softc(sc
->re_miibus
);
2300 error
= ifmedia_ioctl(ifp
, ifr
, &mii
->mii_media
, command
);
2303 ifp
->if_capenable
&= ~(IFCAP_HWCSUM
);
2304 ifp
->if_capenable
|=
2305 ifr
->ifr_reqcap
& (IFCAP_HWCSUM
);
2306 if (ifp
->if_capenable
& IFCAP_TXCSUM
)
2307 ifp
->if_hwassist
= RE_CSUM_FEATURES
;
2309 ifp
->if_hwassist
= 0;
2310 if (ifp
->if_flags
& IFF_RUNNING
)
2314 error
= ether_ioctl(ifp
, command
, data
);
2321 re_watchdog(struct ifnet
*ifp
)
2323 struct re_softc
*sc
= ifp
->if_softc
;
2325 if_printf(ifp
, "watchdog timeout\n");
2334 if (!ifq_is_empty(&ifp
->if_snd
))
2339 * Stop the adapter and free any mbufs allocated to the
2343 re_stop(struct re_softc
*sc
)
2345 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
2349 callout_stop(&sc
->re_timer
);
2351 ifp
->if_flags
&= ~(IFF_RUNNING
| IFF_OACTIVE
);
2353 CSR_WRITE_1(sc
, RE_COMMAND
, 0x00);
2354 CSR_WRITE_2(sc
, RE_IMR
, 0x0000);
2355 CSR_WRITE_2(sc
, RE_ISR
, 0xFFFF);
2357 if (sc
->re_head
!= NULL
) {
2358 m_freem(sc
->re_head
);
2359 sc
->re_head
= sc
->re_tail
= NULL
;
2362 /* Free the TX list buffers. */
2363 for (i
= 0; i
< RE_TX_DESC_CNT
; i
++) {
2364 if (sc
->re_ldata
.re_tx_mbuf
[i
] != NULL
) {
2365 bus_dmamap_unload(sc
->re_ldata
.re_mtag
,
2366 sc
->re_ldata
.re_tx_dmamap
[i
]);
2367 m_freem(sc
->re_ldata
.re_tx_mbuf
[i
]);
2368 sc
->re_ldata
.re_tx_mbuf
[i
] = NULL
;
2372 /* Free the RX list buffers. */
2373 for (i
= 0; i
< RE_RX_DESC_CNT
; i
++) {
2374 if (sc
->re_ldata
.re_rx_mbuf
[i
] != NULL
) {
2375 bus_dmamap_unload(sc
->re_ldata
.re_mtag
,
2376 sc
->re_ldata
.re_rx_dmamap
[i
]);
2377 m_freem(sc
->re_ldata
.re_rx_mbuf
[i
]);
2378 sc
->re_ldata
.re_rx_mbuf
[i
] = NULL
;
2384 * Device suspend routine. Stop the interface and save some PCI
2385 * settings in case the BIOS doesn't restore them properly on
2389 re_suspend(device_t dev
)
2391 #ifndef BURN_BRIDGES
2394 struct re_softc
*sc
= device_get_softc(dev
);
2398 #ifndef BURN_BRIDGES
2399 for (i
= 0; i
< 5; i
++)
2400 sc
->saved_maps
[i
] = pci_read_config(dev
, PCIR_MAPS
+ i
* 4, 4);
2401 sc
->saved_biosaddr
= pci_read_config(dev
, PCIR_BIOS
, 4);
2402 sc
->saved_intline
= pci_read_config(dev
, PCIR_INTLINE
, 1);
2403 sc
->saved_cachelnsz
= pci_read_config(dev
, PCIR_CACHELNSZ
, 1);
2404 sc
->saved_lattimer
= pci_read_config(dev
, PCIR_LATTIMER
, 1);
2413 * Device resume routine. Restore some PCI settings in case the BIOS
2414 * doesn't, re-enable busmastering, and restart the interface if
2418 re_resume(device_t dev
)
2420 struct re_softc
*sc
= device_get_softc(dev
);
2421 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
2422 #ifndef BURN_BRIDGES
2426 #ifndef BURN_BRIDGES
2427 /* better way to do this? */
2428 for (i
= 0; i
< 5; i
++)
2429 pci_write_config(dev
, PCIR_MAPS
+ i
* 4, sc
->saved_maps
[i
], 4);
2430 pci_write_config(dev
, PCIR_BIOS
, sc
->saved_biosaddr
, 4);
2431 pci_write_config(dev
, PCIR_INTLINE
, sc
->saved_intline
, 1);
2432 pci_write_config(dev
, PCIR_CACHELNSZ
, sc
->saved_cachelnsz
, 1);
2433 pci_write_config(dev
, PCIR_LATTIMER
, sc
->saved_lattimer
, 1);
2435 /* reenable busmastering */
2436 pci_enable_busmaster(dev
);
2437 pci_enable_io(dev
, SYS_RES_IOPORT
);
2440 /* reinitialize interface if necessary */
2441 if (ifp
->if_flags
& IFF_UP
)
2450 * Stop all chip I/O so that the kernel's probe routines don't
2451 * get confused by errant DMAs when rebooting.
2454 re_shutdown(device_t dev
)
2456 struct re_softc
*sc
= device_get_softc(dev
);
2457 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
2459 lwkt_serialize_enter(ifp
->if_serializer
);
2461 lwkt_serialize_exit(ifp
->if_serializer
);
2465 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS
)
2467 struct re_softc
*sc
= arg1
;
2468 struct ifnet
*ifp
= &sc
->arpcom
.ac_if
;
2469 int error
= 0, mod
, mod_old
;
2471 lwkt_serialize_enter(ifp
->if_serializer
);
2473 mod_old
= mod
= RE_TX_MODERATION_IS_ENABLED(sc
);
2475 error
= sysctl_handle_int(oidp
, &mod
, 0, req
);
2476 if (error
|| req
->newptr
== NULL
|| mod
== mod_old
)
2478 if (mod
!= 0 && mod
!= 1) {
2484 RE_ENABLE_TX_MODERATION(sc
);
2486 RE_DISABLE_TX_MODERATION(sc
);
2488 if ((ifp
->if_flags
& (IFF_RUNNING
| IFF_UP
)) == (IFF_RUNNING
| IFF_UP
))
2491 lwkt_serialize_exit(ifp
->if_serializer
);