Add a new csum flag to tell IP defragmenter that csum_data does _not_
[dragonfly/port-amd64.git] / sys / dev / netif / re / if_re.c
blob6bdb91d6763aa945d53ec25ddcb84bc417c31094
1 /*
2 * Copyright (c) 2004
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
10 * are met:
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 1.36 2007/08/14 13:30:35 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
44 * Wind River Systems
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
60 * any byte boundary.
62 * o 64-bit DMA
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
75 * chips.
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:
81 * o 1000Mbps mode
83 * o Jumbo frames
85 * o GMII and TBI ports/registers for interfacing with copper
86 * or fiber PHYs
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'
97 * copper gigE PHY.
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>
117 #include <sys/bus.h>
118 #include <sys/endian.h>
119 #include <sys/kernel.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 /* #include <sys/module.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>
129 #include <net/bpf.h>
130 #include <net/ethernet.h>
131 #include <net/if.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>
139 #include <dev/netif/mii_layer/mii.h>
140 #include <dev/netif/mii_layer/miivar.h>
142 #include <bus/pci/pcidevs.h>
143 #include <bus/pci/pcireg.h>
144 #include <bus/pci/pcivar.h>
146 /* "device miibus" required. See GENERIC if you get errors here. */
147 #include "miibus_if.h"
149 #include <dev/netif/re/if_rereg.h>
150 #include <dev/netif/re/if_revar.h>
152 #define RE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
153 #if 0
154 #define RE_DISABLE_HWCSUM
155 #endif
158 * Various supported device vendors/types and their names.
160 static const struct re_type re_devs[] = {
161 { PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T, RE_HWREV_8169S,
162 "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
163 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139, RE_HWREV_8139CPLUS,
164 "RealTek 8139C+ 10/100BaseTX" },
165 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E, RE_HWREV_8101E,
166 "RealTek 8101E PCIe 10/100baseTX" },
167 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN1,
168 "RealTek 8168/8111B PCIe Gigabit Ethernet" },
169 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN2,
170 "RealTek 8168/8111B PCIe Gigabit Ethernet" },
171 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168, RE_HWREV_8168_SPIN3,
172 "RealTek 8168B/8111B PCIe Gigabit Ethernet" },
173 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169,
174 "RealTek 8169 Gigabit Ethernet" },
175 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169S,
176 "RealTek 8169S Single-chip Gigabit Ethernet" },
177 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SB,
178 "RealTek 8169SB/8110SB Single-chip Gigabit Ethernet" },
179 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8169_8110SC,
180 "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
181 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC, RE_HWREV_8169_8110SC,
182 "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
183 { PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169, RE_HWREV_8110S,
184 "RealTek 8110S Single-chip Gigabit Ethernet" },
185 { PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT, RE_HWREV_8169S,
186 "Corega CG-LAPCIGT Gigabit Ethernet" },
187 { PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032, RE_HWREV_8169S,
188 "Linksys EG1032 Gigabit Ethernet" },
189 { PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902, RE_HWREV_8169S,
190 "US Robotics 997902 Gigabit Ethernet" },
191 { 0, 0, 0, NULL }
194 static const struct re_hwrev re_hwrevs[] = {
195 { RE_HWREV_8139CPLUS, RE_8139CPLUS, RE_F_HASMPC, "C+" },
196 { RE_HWREV_8168_SPIN1, RE_8169, RE_F_PCIE, "8168" },
197 { RE_HWREV_8168_SPIN2, RE_8169, RE_F_PCIE, "8168" },
198 { RE_HWREV_8168_SPIN3, RE_8169, RE_F_PCIE, "8168" },
199 { RE_HWREV_8169, RE_8169, RE_F_HASMPC, "8169" },
200 { RE_HWREV_8169S, RE_8169, RE_F_HASMPC, "8169S" },
201 { RE_HWREV_8110S, RE_8169, RE_F_HASMPC, "8110S" },
202 { RE_HWREV_8169_8110SB, RE_8169, RE_F_HASMPC, "8169SB" },
203 { RE_HWREV_8169_8110SC, RE_8169, 0, "8169SC" },
204 { RE_HWREV_8100E, RE_8169, RE_F_HASMPC, "8100E" },
205 { RE_HWREV_8101E, RE_8169, RE_F_PCIE, "8101E" },
206 { 0, 0, 0, NULL }
209 static int re_probe(device_t);
210 static int re_attach(device_t);
211 static int re_detach(device_t);
213 static int re_encap(struct re_softc *, struct mbuf **, int *, int *);
215 static void re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
216 static void re_dma_map_desc(void *, bus_dma_segment_t *, int,
217 bus_size_t, int);
218 static int re_allocmem(device_t, struct re_softc *);
219 static int re_newbuf(struct re_softc *, int, struct mbuf *);
220 static int re_rx_list_init(struct re_softc *);
221 static int re_tx_list_init(struct re_softc *);
222 static void re_rxeof(struct re_softc *);
223 static void re_txeof(struct re_softc *);
224 static void re_intr(void *);
225 static void re_tick(void *);
226 static void re_tick_serialized(void *);
227 static void re_start(struct ifnet *);
228 static int re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
229 static void re_init(void *);
230 static void re_stop(struct re_softc *);
231 static void re_watchdog(struct ifnet *);
232 static int re_suspend(device_t);
233 static int re_resume(device_t);
234 static void re_shutdown(device_t);
235 static int re_ifmedia_upd(struct ifnet *);
236 static void re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
238 static void re_eeprom_putbyte(struct re_softc *, int);
239 static void re_eeprom_getword(struct re_softc *, int, u_int16_t *);
240 static void re_read_eeprom(struct re_softc *, caddr_t, int, int);
241 static int re_gmii_readreg(device_t, int, int);
242 static int re_gmii_writereg(device_t, int, int, int);
244 static int re_miibus_readreg(device_t, int, int);
245 static int re_miibus_writereg(device_t, int, int, int);
246 static void re_miibus_statchg(device_t);
248 static void re_setmulti(struct re_softc *);
249 static void re_reset(struct re_softc *);
251 #ifdef RE_DIAG
252 static int re_diag(struct re_softc *);
253 #endif
255 #ifdef DEVICE_POLLING
256 static void re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
257 #endif
259 static int re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS);
261 static device_method_t re_methods[] = {
262 /* Device interface */
263 DEVMETHOD(device_probe, re_probe),
264 DEVMETHOD(device_attach, re_attach),
265 DEVMETHOD(device_detach, re_detach),
266 DEVMETHOD(device_suspend, re_suspend),
267 DEVMETHOD(device_resume, re_resume),
268 DEVMETHOD(device_shutdown, re_shutdown),
270 /* bus interface */
271 DEVMETHOD(bus_print_child, bus_generic_print_child),
272 DEVMETHOD(bus_driver_added, bus_generic_driver_added),
274 /* MII interface */
275 DEVMETHOD(miibus_readreg, re_miibus_readreg),
276 DEVMETHOD(miibus_writereg, re_miibus_writereg),
277 DEVMETHOD(miibus_statchg, re_miibus_statchg),
279 { 0, 0 }
282 static driver_t re_driver = {
283 "re",
284 re_methods,
285 sizeof(struct re_softc)
288 static devclass_t re_devclass;
290 DECLARE_DUMMY_MODULE(if_re);
291 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
292 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
293 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
295 #define EE_SET(x) \
296 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
298 #define EE_CLR(x) \
299 CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
302 * Send a read command and address to the EEPROM, check for ACK.
304 static void
305 re_eeprom_putbyte(struct re_softc *sc, int addr)
307 int d, i;
309 d = addr | (RE_9346_READ << sc->re_eewidth);
312 * Feed in each bit and strobe the clock.
314 for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
315 if (d & i)
316 EE_SET(RE_EE_DATAIN);
317 else
318 EE_CLR(RE_EE_DATAIN);
319 DELAY(100);
320 EE_SET(RE_EE_CLK);
321 DELAY(150);
322 EE_CLR(RE_EE_CLK);
323 DELAY(100);
328 * Read a word of data stored in the EEPROM at address 'addr.'
330 static void
331 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
333 int i;
334 uint16_t word = 0;
337 * Send address of word we want to read.
339 re_eeprom_putbyte(sc, addr);
342 * Start reading bits from EEPROM.
344 for (i = 0x8000; i != 0; i >>= 1) {
345 EE_SET(RE_EE_CLK);
346 DELAY(100);
347 if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
348 word |= i;
349 EE_CLR(RE_EE_CLK);
350 DELAY(100);
353 *dest = word;
357 * Read a sequence of words from the EEPROM.
359 static void
360 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
362 int i;
363 uint16_t word = 0, *ptr;
365 CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
366 DELAY(100);
368 for (i = 0; i < cnt; i++) {
369 CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
370 re_eeprom_getword(sc, off + i, &word);
371 CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
372 ptr = (uint16_t *)(dest + (i * 2));
373 *ptr = word;
376 CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
379 static int
380 re_gmii_readreg(device_t dev, int phy, int reg)
382 struct re_softc *sc = device_get_softc(dev);
383 u_int32_t rval;
384 int i;
386 if (phy != 1)
387 return(0);
389 /* Let the rgephy driver read the GMEDIASTAT register */
391 if (reg == RE_GMEDIASTAT)
392 return(CSR_READ_1(sc, RE_GMEDIASTAT));
394 CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
395 DELAY(1000);
397 for (i = 0; i < RE_TIMEOUT; i++) {
398 rval = CSR_READ_4(sc, RE_PHYAR);
399 if (rval & RE_PHYAR_BUSY)
400 break;
401 DELAY(100);
404 if (i == RE_TIMEOUT) {
405 device_printf(dev, "PHY read failed\n");
406 return(0);
409 return(rval & RE_PHYAR_PHYDATA);
412 static int
413 re_gmii_writereg(device_t dev, int phy, int reg, int data)
415 struct re_softc *sc = device_get_softc(dev);
416 uint32_t rval;
417 int i;
419 CSR_WRITE_4(sc, RE_PHYAR,
420 (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
421 DELAY(1000);
423 for (i = 0; i < RE_TIMEOUT; i++) {
424 rval = CSR_READ_4(sc, RE_PHYAR);
425 if ((rval & RE_PHYAR_BUSY) == 0)
426 break;
427 DELAY(100);
430 if (i == RE_TIMEOUT)
431 device_printf(dev, "PHY write failed\n");
433 return(0);
436 static int
437 re_miibus_readreg(device_t dev, int phy, int reg)
439 struct re_softc *sc = device_get_softc(dev);
440 uint16_t rval = 0;
441 uint16_t re8139_reg = 0;
443 if (sc->re_type == RE_8169) {
444 rval = re_gmii_readreg(dev, phy, reg);
445 return(rval);
448 /* Pretend the internal PHY is only at address 0 */
449 if (phy)
450 return(0);
452 switch(reg) {
453 case MII_BMCR:
454 re8139_reg = RE_BMCR;
455 break;
456 case MII_BMSR:
457 re8139_reg = RE_BMSR;
458 break;
459 case MII_ANAR:
460 re8139_reg = RE_ANAR;
461 break;
462 case MII_ANER:
463 re8139_reg = RE_ANER;
464 break;
465 case MII_ANLPAR:
466 re8139_reg = RE_LPAR;
467 break;
468 case MII_PHYIDR1:
469 case MII_PHYIDR2:
470 return(0);
472 * Allow the rlphy driver to read the media status
473 * register. If we have a link partner which does not
474 * support NWAY, this is the register which will tell
475 * us the results of parallel detection.
477 case RE_MEDIASTAT:
478 return(CSR_READ_1(sc, RE_MEDIASTAT));
479 default:
480 device_printf(dev, "bad phy register\n");
481 return(0);
483 rval = CSR_READ_2(sc, re8139_reg);
484 if (sc->re_type == RE_8139CPLUS && re8139_reg == RE_BMCR) {
485 /* 8139C+ has different bit layout. */
486 rval &= ~(BMCR_LOOP | BMCR_ISO);
488 return(rval);
491 static int
492 re_miibus_writereg(device_t dev, int phy, int reg, int data)
494 struct re_softc *sc= device_get_softc(dev);
495 u_int16_t re8139_reg = 0;
497 if (sc->re_type == RE_8169)
498 return(re_gmii_writereg(dev, phy, reg, data));
500 /* Pretend the internal PHY is only at address 0 */
501 if (phy)
502 return(0);
504 switch(reg) {
505 case MII_BMCR:
506 re8139_reg = RE_BMCR;
507 if (sc->re_type == RE_8139CPLUS) {
508 /* 8139C+ has different bit layout. */
509 data &= ~(BMCR_LOOP | BMCR_ISO);
511 break;
512 case MII_BMSR:
513 re8139_reg = RE_BMSR;
514 break;
515 case MII_ANAR:
516 re8139_reg = RE_ANAR;
517 break;
518 case MII_ANER:
519 re8139_reg = RE_ANER;
520 break;
521 case MII_ANLPAR:
522 re8139_reg = RE_LPAR;
523 break;
524 case MII_PHYIDR1:
525 case MII_PHYIDR2:
526 return(0);
527 default:
528 device_printf(dev, "bad phy register\n");
529 return(0);
531 CSR_WRITE_2(sc, re8139_reg, data);
532 return(0);
535 static void
536 re_miibus_statchg(device_t dev)
541 * Program the 64-bit multicast hash filter.
543 static void
544 re_setmulti(struct re_softc *sc)
546 struct ifnet *ifp = &sc->arpcom.ac_if;
547 int h = 0;
548 uint32_t hashes[2] = { 0, 0 };
549 struct ifmultiaddr *ifma;
550 uint32_t rxfilt;
551 int mcnt = 0;
553 rxfilt = CSR_READ_4(sc, RE_RXCFG);
555 if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
556 rxfilt |= RE_RXCFG_RX_MULTI;
557 CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
558 CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
559 CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
560 return;
563 /* first, zot all the existing hash bits */
564 CSR_WRITE_4(sc, RE_MAR0, 0);
565 CSR_WRITE_4(sc, RE_MAR4, 0);
567 /* now program new ones */
568 LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
569 if (ifma->ifma_addr->sa_family != AF_LINK)
570 continue;
571 h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
572 ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
573 if (h < 32)
574 hashes[0] |= (1 << h);
575 else
576 hashes[1] |= (1 << (h - 32));
577 mcnt++;
580 if (mcnt)
581 rxfilt |= RE_RXCFG_RX_MULTI;
582 else
583 rxfilt &= ~RE_RXCFG_RX_MULTI;
585 CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
588 * For some unfathomable reason, RealTek decided to reverse
589 * the order of the multicast hash registers in the PCI Express
590 * parts. This means we have to write the hash pattern in reverse
591 * order for those devices.
593 if (sc->re_flags & RE_F_PCIE) {
594 CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
595 CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
596 } else {
597 CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
598 CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
602 static void
603 re_reset(struct re_softc *sc)
605 int i;
607 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
609 for (i = 0; i < RE_TIMEOUT; i++) {
610 DELAY(10);
611 if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
612 break;
614 if (i == RE_TIMEOUT)
615 if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
617 CSR_WRITE_1(sc, 0x82, 1);
620 #ifdef RE_DIAG
622 * The following routine is designed to test for a defect on some
623 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
624 * lines connected to the bus, however for a 32-bit only card, they
625 * should be pulled high. The result of this defect is that the
626 * NIC will not work right if you plug it into a 64-bit slot: DMA
627 * operations will be done with 64-bit transfers, which will fail
628 * because the 64-bit data lines aren't connected.
630 * There's no way to work around this (short of talking a soldering
631 * iron to the board), however we can detect it. The method we use
632 * here is to put the NIC into digital loopback mode, set the receiver
633 * to promiscuous mode, and then try to send a frame. We then compare
634 * the frame data we sent to what was received. If the data matches,
635 * then the NIC is working correctly, otherwise we know the user has
636 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
637 * slot. In the latter case, there's no way the NIC can work correctly,
638 * so we print out a message on the console and abort the device attach.
641 static int
642 re_diag(struct re_softc *sc)
644 struct ifnet *ifp = &sc->arpcom.ac_if;
645 struct mbuf *m0;
646 struct ether_header *eh;
647 struct re_desc *cur_rx;
648 uint16_t status;
649 uint32_t rxstat;
650 int total_len, i, error = 0, phyaddr;
651 uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
652 uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
654 /* Allocate a single mbuf */
656 MGETHDR(m0, MB_DONTWAIT, MT_DATA);
657 if (m0 == NULL)
658 return(ENOBUFS);
661 * Initialize the NIC in test mode. This sets the chip up
662 * so that it can send and receive frames, but performs the
663 * following special functions:
664 * - Puts receiver in promiscuous mode
665 * - Enables digital loopback mode
666 * - Leaves interrupts turned off
669 ifp->if_flags |= IFF_PROMISC;
670 sc->re_testmode = 1;
671 re_reset(sc);
672 re_init(sc);
673 sc->re_link = 1;
674 if (sc->re_type == RE_8169)
675 phyaddr = 1;
676 else
677 phyaddr = 0;
679 re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
680 for (i = 0; i < RE_TIMEOUT; i++) {
681 status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
682 if (!(status & BMCR_RESET))
683 break;
686 re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
687 CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
689 DELAY(100000);
691 /* Put some data in the mbuf */
693 eh = mtod(m0, struct ether_header *);
694 bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
695 bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
696 eh->ether_type = htons(ETHERTYPE_IP);
697 m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
700 * Queue the packet, start transmission.
701 * Note: ifq_handoff() ultimately calls re_start() for us.
704 CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
705 error = ifq_handoff(ifp, m0, NULL);
706 if (error) {
707 m0 = NULL;
708 goto done;
710 m0 = NULL;
712 /* Wait for it to propagate through the chip */
714 DELAY(100000);
715 for (i = 0; i < RE_TIMEOUT; i++) {
716 status = CSR_READ_2(sc, RE_ISR);
717 CSR_WRITE_2(sc, RE_ISR, status);
718 if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
719 (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
720 break;
721 DELAY(10);
724 if (i == RE_TIMEOUT) {
725 if_printf(ifp, "diagnostic failed to receive packet "
726 "in loopback mode\n");
727 error = EIO;
728 goto done;
732 * The packet should have been dumped into the first
733 * entry in the RX DMA ring. Grab it from there.
736 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
737 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
738 bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
739 BUS_DMASYNC_POSTWRITE);
740 bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
742 m0 = sc->re_ldata.re_rx_mbuf[0];
743 sc->re_ldata.re_rx_mbuf[0] = NULL;
744 eh = mtod(m0, struct ether_header *);
746 cur_rx = &sc->re_ldata.re_rx_list[0];
747 total_len = RE_RXBYTES(cur_rx);
748 rxstat = le32toh(cur_rx->re_cmdstat);
750 if (total_len != ETHER_MIN_LEN) {
751 if_printf(ifp, "diagnostic failed, received short packet\n");
752 error = EIO;
753 goto done;
756 /* Test that the received packet data matches what we sent. */
758 if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
759 bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
760 be16toh(eh->ether_type) != ETHERTYPE_IP) {
761 if_printf(ifp, "WARNING, DMA FAILURE!\n");
762 if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
763 dst, ":", src, ":", ETHERTYPE_IP);
764 if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
765 eh->ether_dhost, ":", eh->ether_shost, ":",
766 ntohs(eh->ether_type));
767 if_printf(ifp, "You may have a defective 32-bit NIC plugged "
768 "into a 64-bit PCI slot.\n");
769 if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
770 "for proper operation.\n");
771 if_printf(ifp, "Read the re(4) man page for more details.\n");
772 error = EIO;
775 done:
776 /* Turn interface off, release resources */
778 sc->re_testmode = 0;
779 sc->re_link = 0;
780 ifp->if_flags &= ~IFF_PROMISC;
781 re_stop(sc);
782 if (m0 != NULL)
783 m_freem(m0);
785 return (error);
787 #endif /* RE_DIAG */
790 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
791 * IDs against our list and return a device name if we find a match.
793 static int
794 re_probe(device_t dev)
796 const struct re_type *t;
797 struct re_softc *sc;
798 int rid;
799 uint32_t hwrev;
800 uint16_t vendor, product;
802 t = re_devs;
804 vendor = pci_get_vendor(dev);
805 product = pci_get_device(dev);
808 * Only attach to rev.3 of the Linksys EG1032 adapter.
809 * Rev.2 is supported by sk(4).
811 if (vendor == PCI_VENDOR_LINKSYS &&
812 product == PCI_PRODUCT_LINKSYS_EG1032 &&
813 pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
814 return ENXIO;
816 for (t = re_devs; t->re_name != NULL; t++) {
817 if (product == t->re_did && vendor == t->re_vid)
818 break;
822 * Check if we found a RealTek device.
824 if (t->re_name == NULL)
825 return(ENXIO);
828 * Temporarily map the I/O space so we can read the chip ID register.
830 sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
831 rid = RE_PCI_LOIO;
832 sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
833 RF_ACTIVE);
834 if (sc->re_res == NULL) {
835 device_printf(dev, "couldn't map ports/memory\n");
836 kfree(sc, M_TEMP);
837 return(ENXIO);
840 sc->re_btag = rman_get_bustag(sc->re_res);
841 sc->re_bhandle = rman_get_bushandle(sc->re_res);
843 hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
844 bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
845 kfree(sc, M_TEMP);
848 * and continue matching for the specific chip...
850 for (; t->re_name != NULL; t++) {
851 if (product == t->re_did && vendor == t->re_vid &&
852 t->re_basetype == hwrev) {
853 device_set_desc(dev, t->re_name);
854 return(0);
857 return(ENXIO);
861 * This routine takes the segment list provided as the result of
862 * a bus_dma_map_load() operation and assigns the addresses/lengths
863 * to RealTek DMA descriptors. This can be called either by the RX
864 * code or the TX code. In the RX case, we'll probably wind up mapping
865 * at most one segment. For the TX case, there could be any number of
866 * segments since TX packets may span multiple mbufs. In either case,
867 * if the number of segments is larger than the re_maxsegs limit
868 * specified by the caller, we abort the mapping operation. Sadly,
869 * whoever designed the buffer mapping API did not provide a way to
870 * return an error from here, so we have to fake it a bit.
873 static void
874 re_dma_map_desc(void *arg, bus_dma_segment_t *segs, int nseg,
875 bus_size_t mapsize, int error)
877 struct re_dmaload_arg *ctx;
878 struct re_desc *d = NULL;
879 int i = 0, idx;
880 uint32_t cmdstat;
882 if (error)
883 return;
885 ctx = arg;
887 /* Signal error to caller if there's too many segments */
888 if (nseg > ctx->re_maxsegs) {
889 ctx->re_maxsegs = 0;
890 return;
894 * Map the segment array into descriptors. Note that we set the
895 * start-of-frame and end-of-frame markers for either TX or RX, but
896 * they really only have meaning in the TX case. (In the RX case,
897 * it's the chip that tells us where packets begin and end.)
898 * We also keep track of the end of the ring and set the
899 * end-of-ring bits as needed, and we set the ownership bits
900 * in all except the very first descriptor. (The caller will
901 * set this descriptor later when it start transmission or
902 * reception.)
904 idx = ctx->re_idx;
905 for (;;) {
906 d = &ctx->re_ring[idx];
907 if (le32toh(d->re_cmdstat) & RE_RDESC_STAT_OWN) {
908 ctx->re_maxsegs = 0;
909 return;
911 cmdstat = segs[i].ds_len;
912 d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
913 d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
914 if (i == 0)
915 cmdstat |= RE_TDESC_CMD_SOF;
916 else
917 cmdstat |= RE_TDESC_CMD_OWN;
918 if (idx == (RE_RX_DESC_CNT - 1))
919 cmdstat |= RE_TDESC_CMD_EOR;
920 d->re_cmdstat = htole32(cmdstat | ctx->re_flags);
921 i++;
922 if (i == nseg)
923 break;
924 RE_DESC_INC(idx);
927 d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
928 ctx->re_maxsegs = nseg;
929 ctx->re_idx = idx;
933 * Map a single buffer address.
936 static void
937 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
939 uint32_t *addr;
941 if (error)
942 return;
944 KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
945 addr = arg;
946 *addr = segs->ds_addr;
949 static int
950 re_allocmem(device_t dev, struct re_softc *sc)
952 int error, i, nseg;
955 * Allocate map for RX mbufs.
957 nseg = 32;
958 error = bus_dma_tag_create(sc->re_parent_tag, ETHER_ALIGN, 0,
959 BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
960 NULL, MCLBYTES * nseg, nseg, MCLBYTES, BUS_DMA_ALLOCNOW,
961 &sc->re_ldata.re_mtag);
962 if (error) {
963 device_printf(dev, "could not allocate dma tag\n");
964 return(error);
968 * Allocate map for TX descriptor list.
970 error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
971 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
972 NULL, RE_TX_LIST_SZ, 1, RE_TX_LIST_SZ, BUS_DMA_ALLOCNOW,
973 &sc->re_ldata.re_tx_list_tag);
974 if (error) {
975 device_printf(dev, "could not allocate dma tag\n");
976 return(error);
979 /* Allocate DMA'able memory for the TX ring */
981 error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
982 (void **)&sc->re_ldata.re_tx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
983 &sc->re_ldata.re_tx_list_map);
984 if (error) {
985 device_printf(dev, "could not allocate TX ring\n");
986 return(error);
989 /* Load the map for the TX ring. */
991 error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
992 sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
993 RE_TX_LIST_SZ, re_dma_map_addr,
994 &sc->re_ldata.re_tx_list_addr, BUS_DMA_NOWAIT);
995 if (error) {
996 device_printf(dev, "could not get addres of TX ring\n");
997 return(error);
1000 /* Create DMA maps for TX buffers */
1002 for (i = 0; i < RE_TX_DESC_CNT; i++) {
1003 error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1004 &sc->re_ldata.re_tx_dmamap[i]);
1005 if (error) {
1006 device_printf(dev, "can't create DMA map for TX\n");
1007 return(error);
1012 * Allocate map for RX descriptor list.
1014 error = bus_dma_tag_create(sc->re_parent_tag, RE_RING_ALIGN,
1015 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1016 NULL, RE_RX_LIST_SZ, 1, RE_RX_LIST_SZ, BUS_DMA_ALLOCNOW,
1017 &sc->re_ldata.re_rx_list_tag);
1018 if (error) {
1019 device_printf(dev, "could not allocate dma tag\n");
1020 return(error);
1023 /* Allocate DMA'able memory for the RX ring */
1025 error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1026 (void **)&sc->re_ldata.re_rx_list, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1027 &sc->re_ldata.re_rx_list_map);
1028 if (error) {
1029 device_printf(dev, "could not allocate RX ring\n");
1030 return(error);
1033 /* Load the map for the RX ring. */
1035 error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1036 sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
1037 RE_RX_LIST_SZ, re_dma_map_addr,
1038 &sc->re_ldata.re_rx_list_addr, BUS_DMA_NOWAIT);
1039 if (error) {
1040 device_printf(dev, "could not get address of RX ring\n");
1041 return(error);
1044 /* Create DMA maps for RX buffers */
1046 for (i = 0; i < RE_RX_DESC_CNT; i++) {
1047 error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1048 &sc->re_ldata.re_rx_dmamap[i]);
1049 if (error) {
1050 device_printf(dev, "can't create DMA map for RX\n");
1051 return(ENOMEM);
1055 return(0);
1059 * Attach the interface. Allocate softc structures, do ifmedia
1060 * setup and ethernet/BPF attach.
1062 static int
1063 re_attach(device_t dev)
1065 struct re_softc *sc = device_get_softc(dev);
1066 struct ifnet *ifp;
1067 const struct re_hwrev *hw_rev;
1068 uint8_t eaddr[ETHER_ADDR_LEN];
1069 uint16_t as[ETHER_ADDR_LEN / 2];
1070 uint16_t re_did = 0;
1071 uint32_t hwrev;
1072 int error = 0, rid, i;
1074 callout_init(&sc->re_timer);
1075 #ifdef RE_DIAG
1076 sc->re_dev = dev;
1077 #endif
1079 RE_ENABLE_TX_MODERATION(sc);
1081 sysctl_ctx_init(&sc->re_sysctl_ctx);
1082 sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1083 SYSCTL_STATIC_CHILDREN(_hw),
1084 OID_AUTO,
1085 device_get_nameunit(dev),
1086 CTLFLAG_RD, 0, "");
1087 if (sc->re_sysctl_tree == NULL) {
1088 device_printf(dev, "can't add sysctl node\n");
1089 error = ENXIO;
1090 goto fail;
1092 SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1093 SYSCTL_CHILDREN(sc->re_sysctl_tree),
1094 OID_AUTO, "tx_moderation",
1095 CTLTYPE_INT | CTLFLAG_RW,
1096 sc, 0, re_sysctl_tx_moderation, "I",
1097 "Enable/Disable TX moderation");
1099 #ifndef BURN_BRIDGES
1101 * Handle power management nonsense.
1104 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1105 uint32_t membase, irq;
1107 /* Save important PCI config data. */
1108 membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1109 irq = pci_read_config(dev, PCIR_INTLINE, 4);
1111 /* Reset the power state. */
1112 device_printf(dev, "chip is in D%d power mode "
1113 "-- setting to D0\n", pci_get_powerstate(dev));
1115 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1117 /* Restore PCI config data. */
1118 pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1119 pci_write_config(dev, PCIR_INTLINE, irq, 4);
1121 #endif
1123 * Map control/status registers.
1125 pci_enable_busmaster(dev);
1127 rid = RE_PCI_LOIO;
1128 sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1129 RF_ACTIVE);
1131 if (sc->re_res == NULL) {
1132 device_printf(dev, "couldn't map ports\n");
1133 error = ENXIO;
1134 goto fail;
1137 sc->re_btag = rman_get_bustag(sc->re_res);
1138 sc->re_bhandle = rman_get_bushandle(sc->re_res);
1140 /* Allocate interrupt */
1141 rid = 0;
1142 sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1143 RF_SHAREABLE | RF_ACTIVE);
1145 if (sc->re_irq == NULL) {
1146 device_printf(dev, "couldn't map interrupt\n");
1147 error = ENXIO;
1148 goto fail;
1151 /* Reset the adapter. */
1152 re_reset(sc);
1154 hwrev = CSR_READ_4(sc, RE_TXCFG) & RE_TXCFG_HWREV;
1155 for (hw_rev = re_hwrevs; hw_rev->re_desc != NULL; hw_rev++) {
1156 if (hw_rev->re_rev == hwrev) {
1157 sc->re_type = hw_rev->re_type;
1158 sc->re_flags = hw_rev->re_flags;
1159 break;
1163 sc->re_eewidth = 6;
1164 re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1165 if (re_did != 0x8129)
1166 sc->re_eewidth = 8;
1169 * Get station address from the EEPROM.
1171 re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
1172 for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1173 as[i] = le16toh(as[i]);
1174 bcopy(as, eaddr, sizeof(eaddr));
1176 if (sc->re_type == RE_8169) {
1177 /* Set RX length mask */
1178 sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1179 sc->re_txstart = RE_GTXSTART;
1180 } else {
1181 /* Set RX length mask */
1182 sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1183 sc->re_txstart = RE_TXSTART;
1187 * Allocate the parent bus DMA tag appropriate for PCI.
1189 #define RE_NSEG_NEW 32
1190 error = bus_dma_tag_create(NULL, /* parent */
1191 1, 0, /* alignment, boundary */
1192 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1193 BUS_SPACE_MAXADDR, /* highaddr */
1194 NULL, NULL, /* filter, filterarg */
1195 MAXBSIZE, RE_NSEG_NEW, /* maxsize, nsegments */
1196 BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1197 BUS_DMA_ALLOCNOW, /* flags */
1198 &sc->re_parent_tag);
1199 if (error)
1200 goto fail;
1202 error = re_allocmem(dev, sc);
1204 if (error)
1205 goto fail;
1207 /* Do MII setup */
1208 if (mii_phy_probe(dev, &sc->re_miibus,
1209 re_ifmedia_upd, re_ifmedia_sts)) {
1210 device_printf(dev, "MII without any phy!\n");
1211 error = ENXIO;
1212 goto fail;
1215 ifp = &sc->arpcom.ac_if;
1216 ifp->if_softc = sc;
1217 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1218 ifp->if_mtu = ETHERMTU;
1219 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1220 ifp->if_ioctl = re_ioctl;
1221 ifp->if_capabilities = IFCAP_VLAN_MTU;
1222 ifp->if_start = re_start;
1223 ifp->if_capabilities |= IFCAP_HWCSUM|IFCAP_VLAN_HWTAGGING;
1224 #ifdef DEVICE_POLLING
1225 ifp->if_poll = re_poll;
1226 #endif
1227 ifp->if_watchdog = re_watchdog;
1228 ifp->if_init = re_init;
1229 if (sc->re_type == RE_8169)
1230 ifp->if_baudrate = 1000000000;
1231 else
1232 ifp->if_baudrate = 100000000;
1233 ifq_set_maxlen(&ifp->if_snd, RE_IFQ_MAXLEN);
1234 ifq_set_ready(&ifp->if_snd);
1236 #ifdef RE_DISABLE_HWCSUM
1237 ifp->if_capenable = ifp->if_capabilities & ~IFCAP_HWCSUM;
1238 ifp->if_hwassist = 0;
1239 #else
1240 ifp->if_capenable = ifp->if_capabilities;
1241 ifp->if_hwassist = RE_CSUM_FEATURES;
1242 #endif /* RE_DISABLE_HWCSUM */
1245 * Call MI attach routine.
1247 ether_ifattach(ifp, eaddr, NULL);
1249 #ifdef RE_DIAG
1251 * Perform hardware diagnostic on the original RTL8169.
1252 * Some 32-bit cards were incorrectly wired and would
1253 * malfunction if plugged into a 64-bit slot.
1255 if (hwrev == RE_HWREV_8169) {
1256 lwkt_serialize_enter(ifp->if_serializer);
1257 error = re_diag(sc);
1258 lwkt_serialize_exit(ifp->if_serializer);
1260 if (error) {
1261 device_printf(dev, "hardware diagnostic failure\n");
1262 ether_ifdetach(ifp);
1263 goto fail;
1266 #endif /* RE_DIAG */
1268 /* Hook interrupt last to avoid having to lock softc */
1269 error = bus_setup_intr(dev, sc->re_irq, INTR_NETSAFE, re_intr, sc,
1270 &sc->re_intrhand, ifp->if_serializer);
1272 if (error) {
1273 device_printf(dev, "couldn't set up irq\n");
1274 ether_ifdetach(ifp);
1275 goto fail;
1278 fail:
1279 if (error)
1280 re_detach(dev);
1282 return (error);
1286 * Shutdown hardware and free up resources. This can be called any
1287 * time after the mutex has been initialized. It is called in both
1288 * the error case in attach and the normal detach case so it needs
1289 * to be careful about only freeing resources that have actually been
1290 * allocated.
1292 static int
1293 re_detach(device_t dev)
1295 struct re_softc *sc = device_get_softc(dev);
1296 struct ifnet *ifp = &sc->arpcom.ac_if;
1297 int i;
1299 /* These should only be active if attach succeeded */
1300 if (device_is_attached(dev)) {
1301 lwkt_serialize_enter(ifp->if_serializer);
1302 re_stop(sc);
1303 bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1304 lwkt_serialize_exit(ifp->if_serializer);
1306 ether_ifdetach(ifp);
1308 if (sc->re_miibus)
1309 device_delete_child(dev, sc->re_miibus);
1310 bus_generic_detach(dev);
1312 if (sc->re_irq)
1313 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1314 if (sc->re_res) {
1315 bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1316 sc->re_res);
1319 /* Unload and free the RX DMA ring memory and map */
1321 if (sc->re_ldata.re_rx_list_tag) {
1322 bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1323 sc->re_ldata.re_rx_list_map);
1324 bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1325 sc->re_ldata.re_rx_list,
1326 sc->re_ldata.re_rx_list_map);
1327 bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1330 /* Unload and free the TX DMA ring memory and map */
1332 if (sc->re_ldata.re_tx_list_tag) {
1333 bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1334 sc->re_ldata.re_tx_list_map);
1335 bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1336 sc->re_ldata.re_tx_list,
1337 sc->re_ldata.re_tx_list_map);
1338 bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1341 /* Destroy all the RX and TX buffer maps */
1343 if (sc->re_ldata.re_mtag) {
1344 for (i = 0; i < RE_TX_DESC_CNT; i++)
1345 bus_dmamap_destroy(sc->re_ldata.re_mtag,
1346 sc->re_ldata.re_tx_dmamap[i]);
1347 for (i = 0; i < RE_RX_DESC_CNT; i++)
1348 bus_dmamap_destroy(sc->re_ldata.re_mtag,
1349 sc->re_ldata.re_rx_dmamap[i]);
1350 bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1353 /* Unload and free the stats buffer and map */
1355 if (sc->re_ldata.re_stag) {
1356 bus_dmamap_unload(sc->re_ldata.re_stag,
1357 sc->re_ldata.re_rx_list_map);
1358 bus_dmamem_free(sc->re_ldata.re_stag,
1359 sc->re_ldata.re_stats,
1360 sc->re_ldata.re_smap);
1361 bus_dma_tag_destroy(sc->re_ldata.re_stag);
1364 if (sc->re_parent_tag)
1365 bus_dma_tag_destroy(sc->re_parent_tag);
1367 return(0);
1370 static int
1371 re_newbuf(struct re_softc *sc, int idx, struct mbuf *m)
1373 struct re_dmaload_arg arg;
1374 struct mbuf *n = NULL;
1375 int error;
1377 if (m == NULL) {
1378 n = m_getcl(MB_DONTWAIT, MT_DATA, M_PKTHDR);
1379 if (n == NULL)
1380 return(ENOBUFS);
1381 m = n;
1382 } else
1383 m->m_data = m->m_ext.ext_buf;
1385 m->m_len = m->m_pkthdr.len = MCLBYTES;
1388 * NOTE:
1389 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1390 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1393 arg.sc = sc;
1394 arg.re_idx = idx;
1395 arg.re_maxsegs = 1;
1396 arg.re_flags = 0;
1397 arg.re_ring = sc->re_ldata.re_rx_list;
1399 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1400 sc->re_ldata.re_rx_dmamap[idx], m, re_dma_map_desc,
1401 &arg, BUS_DMA_NOWAIT);
1402 if (error || arg.re_maxsegs != 1) {
1403 if (n != NULL)
1404 m_freem(n);
1405 return (ENOMEM);
1408 sc->re_ldata.re_rx_list[idx].re_cmdstat |= htole32(RE_RDESC_CMD_OWN);
1409 sc->re_ldata.re_rx_mbuf[idx] = m;
1411 bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[idx],
1412 BUS_DMASYNC_PREREAD);
1414 return(0);
1417 static int
1418 re_tx_list_init(struct re_softc *sc)
1420 bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ);
1421 bzero(&sc->re_ldata.re_tx_mbuf, RE_TX_DESC_CNT * sizeof(struct mbuf *));
1423 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1424 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1425 sc->re_ldata.re_tx_prodidx = 0;
1426 sc->re_ldata.re_tx_considx = 0;
1427 sc->re_ldata.re_tx_free = RE_TX_DESC_CNT;
1429 return(0);
1432 static int
1433 re_rx_list_init(struct re_softc *sc)
1435 int i, error;
1437 bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ);
1438 bzero(&sc->re_ldata.re_rx_mbuf, RE_RX_DESC_CNT * sizeof(struct mbuf *));
1440 for (i = 0; i < RE_RX_DESC_CNT; i++) {
1441 error = re_newbuf(sc, i, NULL);
1442 if (error)
1443 return(error);
1446 /* Flush the RX descriptors */
1448 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1449 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1451 sc->re_ldata.re_rx_prodidx = 0;
1452 sc->re_head = sc->re_tail = NULL;
1454 return(0);
1458 * RX handler for C+ and 8169. For the gigE chips, we support
1459 * the reception of jumbo frames that have been fragmented
1460 * across multiple 2K mbuf cluster buffers.
1462 static void
1463 re_rxeof(struct re_softc *sc)
1465 struct ifnet *ifp = &sc->arpcom.ac_if;
1466 struct mbuf *m;
1467 struct re_desc *cur_rx;
1468 uint32_t rxstat, rxvlan;
1469 int i, total_len;
1471 /* Invalidate the descriptor memory */
1473 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1474 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1476 for (i = sc->re_ldata.re_rx_prodidx;
1477 RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0 ; RE_DESC_INC(i)) {
1478 cur_rx = &sc->re_ldata.re_rx_list[i];
1479 m = sc->re_ldata.re_rx_mbuf[i];
1480 total_len = RE_RXBYTES(cur_rx);
1481 rxstat = le32toh(cur_rx->re_cmdstat);
1482 rxvlan = le32toh(cur_rx->re_vlanctl);
1484 /* Invalidate the RX mbuf and unload its map */
1486 bus_dmamap_sync(sc->re_ldata.re_mtag,
1487 sc->re_ldata.re_rx_dmamap[i],
1488 BUS_DMASYNC_POSTWRITE);
1489 bus_dmamap_unload(sc->re_ldata.re_mtag,
1490 sc->re_ldata.re_rx_dmamap[i]);
1492 if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1493 m->m_len = MCLBYTES - ETHER_ALIGN;
1494 if (sc->re_head == NULL) {
1495 sc->re_head = sc->re_tail = m;
1496 } else {
1497 sc->re_tail->m_next = m;
1498 sc->re_tail = m;
1500 re_newbuf(sc, i, NULL);
1501 continue;
1505 * NOTE: for the 8139C+, the frame length field
1506 * is always 12 bits in size, but for the gigE chips,
1507 * it is 13 bits (since the max RX frame length is 16K).
1508 * Unfortunately, all 32 bits in the status word
1509 * were already used, so to make room for the extra
1510 * length bit, RealTek took out the 'frame alignment
1511 * error' bit and shifted the other status bits
1512 * over one slot. The OWN, EOR, FS and LS bits are
1513 * still in the same places. We have already extracted
1514 * the frame length and checked the OWN bit, so rather
1515 * than using an alternate bit mapping, we shift the
1516 * status bits one space to the right so we can evaluate
1517 * them using the 8169 status as though it was in the
1518 * same format as that of the 8139C+.
1520 if (sc->re_type == RE_8169)
1521 rxstat >>= 1;
1523 if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1524 ifp->if_ierrors++;
1526 * If this is part of a multi-fragment packet,
1527 * discard all the pieces.
1529 if (sc->re_head != NULL) {
1530 m_freem(sc->re_head);
1531 sc->re_head = sc->re_tail = NULL;
1533 re_newbuf(sc, i, m);
1534 continue;
1538 * If allocating a replacement mbuf fails,
1539 * reload the current one.
1542 if (re_newbuf(sc, i, NULL)) {
1543 ifp->if_ierrors++;
1544 if (sc->re_head != NULL) {
1545 m_freem(sc->re_head);
1546 sc->re_head = sc->re_tail = NULL;
1548 re_newbuf(sc, i, m);
1549 continue;
1552 if (sc->re_head != NULL) {
1553 m->m_len = total_len % (MCLBYTES - ETHER_ALIGN);
1555 * Special case: if there's 4 bytes or less
1556 * in this buffer, the mbuf can be discarded:
1557 * the last 4 bytes is the CRC, which we don't
1558 * care about anyway.
1560 if (m->m_len <= ETHER_CRC_LEN) {
1561 sc->re_tail->m_len -=
1562 (ETHER_CRC_LEN - m->m_len);
1563 m_freem(m);
1564 } else {
1565 m->m_len -= ETHER_CRC_LEN;
1566 sc->re_tail->m_next = m;
1568 m = sc->re_head;
1569 sc->re_head = sc->re_tail = NULL;
1570 m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
1571 } else
1572 m->m_pkthdr.len = m->m_len =
1573 (total_len - ETHER_CRC_LEN);
1575 ifp->if_ipackets++;
1576 m->m_pkthdr.rcvif = ifp;
1578 /* Do RX checksumming if enabled */
1580 if (ifp->if_capenable & IFCAP_RXCSUM) {
1582 /* Check IP header checksum */
1583 if (rxstat & RE_RDESC_STAT_PROTOID)
1584 m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
1585 if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
1586 m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1588 /* Check TCP/UDP checksum */
1589 if ((RE_TCPPKT(rxstat) &&
1590 (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
1591 (RE_UDPPKT(rxstat) &&
1592 (rxstat & RE_RDESC_STAT_UDPSUMBAD)) == 0) {
1593 m->m_pkthdr.csum_flags |=
1594 CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
1595 CSUM_FRAG_NOT_CHECKED;
1596 m->m_pkthdr.csum_data = 0xffff;
1600 if (rxvlan & RE_RDESC_VLANCTL_TAG) {
1601 VLAN_INPUT_TAG(m,
1602 be16toh((rxvlan & RE_RDESC_VLANCTL_DATA)));
1603 } else {
1604 ifp->if_input(ifp, m);
1608 /* Flush the RX DMA ring */
1610 bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1611 sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1613 sc->re_ldata.re_rx_prodidx = i;
1616 static void
1617 re_txeof(struct re_softc *sc)
1619 struct ifnet *ifp = &sc->arpcom.ac_if;
1620 uint32_t txstat;
1621 int idx;
1623 /* Invalidate the TX descriptor list */
1625 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1626 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
1628 for (idx = sc->re_ldata.re_tx_considx;
1629 sc->re_ldata.re_tx_free < RE_TX_DESC_CNT; RE_DESC_INC(idx)) {
1630 txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
1631 if (txstat & RE_TDESC_CMD_OWN)
1632 break;
1634 sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
1637 * We only stash mbufs in the last descriptor
1638 * in a fragment chain, which also happens to
1639 * be the only place where the TX status bits
1640 * are valid.
1642 if (txstat & RE_TDESC_CMD_EOF) {
1643 m_freem(sc->re_ldata.re_tx_mbuf[idx]);
1644 sc->re_ldata.re_tx_mbuf[idx] = NULL;
1645 bus_dmamap_unload(sc->re_ldata.re_mtag,
1646 sc->re_ldata.re_tx_dmamap[idx]);
1647 if (txstat & (RE_TDESC_STAT_EXCESSCOL|
1648 RE_TDESC_STAT_COLCNT))
1649 ifp->if_collisions++;
1650 if (txstat & RE_TDESC_STAT_TXERRSUM)
1651 ifp->if_oerrors++;
1652 else
1653 ifp->if_opackets++;
1655 sc->re_ldata.re_tx_free++;
1658 /* No changes made to the TX ring, so no flush needed */
1659 if (sc->re_ldata.re_tx_free) {
1660 sc->re_ldata.re_tx_considx = idx;
1661 ifp->if_flags &= ~IFF_OACTIVE;
1662 ifp->if_timer = 0;
1666 * Some chips will ignore a second TX request issued while an
1667 * existing transmission is in progress. If the transmitter goes
1668 * idle but there are still packets waiting to be sent, we need
1669 * to restart the channel here to flush them out. This only seems
1670 * to be required with the PCIe devices.
1672 if (sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1673 CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
1676 * If not all descriptors have been released reaped yet,
1677 * reload the timer so that we will eventually get another
1678 * interrupt that will cause us to re-enter this routine.
1679 * This is done in case the transmitter has gone idle.
1681 if (RE_TX_MODERATION_IS_ENABLED(sc) &&
1682 sc->re_ldata.re_tx_free < RE_TX_DESC_CNT)
1683 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
1686 static void
1687 re_tick(void *xsc)
1689 struct re_softc *sc = xsc;
1691 lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
1692 re_tick_serialized(xsc);
1693 lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
1696 static void
1697 re_tick_serialized(void *xsc)
1699 struct re_softc *sc = xsc;
1700 struct ifnet *ifp = &sc->arpcom.ac_if;
1701 struct mii_data *mii;
1703 mii = device_get_softc(sc->re_miibus);
1704 mii_tick(mii);
1705 if (sc->re_link) {
1706 if (!(mii->mii_media_status & IFM_ACTIVE))
1707 sc->re_link = 0;
1708 } else {
1709 if (mii->mii_media_status & IFM_ACTIVE &&
1710 IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
1711 sc->re_link = 1;
1712 if (!ifq_is_empty(&ifp->if_snd))
1713 ifp->if_start(ifp);
1717 callout_reset(&sc->re_timer, hz, re_tick, sc);
1720 #ifdef DEVICE_POLLING
1722 static void
1723 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1725 struct re_softc *sc = ifp->if_softc;
1727 switch(cmd) {
1728 case POLL_REGISTER:
1729 /* disable interrupts */
1730 CSR_WRITE_2(sc, RE_IMR, 0x0000);
1731 break;
1732 case POLL_DEREGISTER:
1733 /* enable interrupts */
1734 CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
1735 break;
1736 default:
1737 sc->rxcycles = count;
1738 re_rxeof(sc);
1739 re_txeof(sc);
1741 if (!ifq_is_empty(&ifp->if_snd))
1742 (*ifp->if_start)(ifp);
1744 if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
1745 uint16_t status;
1747 status = CSR_READ_2(sc, RE_ISR);
1748 if (status == 0xffff)
1749 return;
1750 if (status)
1751 CSR_WRITE_2(sc, RE_ISR, status);
1754 * XXX check behaviour on receiver stalls.
1757 if (status & RE_ISR_SYSTEM_ERR) {
1758 re_reset(sc);
1759 re_init(sc);
1762 break;
1765 #endif /* DEVICE_POLLING */
1767 static void
1768 re_intr(void *arg)
1770 struct re_softc *sc = arg;
1771 struct ifnet *ifp = &sc->arpcom.ac_if;
1772 uint16_t status;
1774 if (sc->suspended || (ifp->if_flags & IFF_UP) == 0)
1775 return;
1777 for (;;) {
1778 status = CSR_READ_2(sc, RE_ISR);
1779 /* If the card has gone away the read returns 0xffff. */
1780 if (status == 0xffff)
1781 break;
1782 if (status)
1783 CSR_WRITE_2(sc, RE_ISR, status);
1785 if ((status & sc->re_intrs) == 0)
1786 break;
1788 if (status & (RE_ISR_RX_OK | RE_ISR_RX_ERR | RE_ISR_FIFO_OFLOW))
1789 re_rxeof(sc);
1791 if ((status & sc->re_tx_ack) ||
1792 (status & RE_ISR_TX_ERR) ||
1793 (status & RE_ISR_TX_DESC_UNAVAIL))
1794 re_txeof(sc);
1796 if (status & RE_ISR_SYSTEM_ERR) {
1797 re_reset(sc);
1798 re_init(sc);
1801 if (status & RE_ISR_LINKCHG) {
1802 callout_stop(&sc->re_timer);
1803 re_tick_serialized(sc);
1807 if (!ifq_is_empty(&ifp->if_snd))
1808 (*ifp->if_start)(ifp);
1811 static int
1812 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx, int *called_defrag)
1814 struct ifnet *ifp = &sc->arpcom.ac_if;
1815 struct mbuf *m, *m_new = NULL;
1816 struct re_dmaload_arg arg;
1817 bus_dmamap_t map;
1818 int error;
1820 *called_defrag = 0;
1821 if (sc->re_ldata.re_tx_free <= 4)
1822 return(EFBIG);
1824 m = *m_head;
1827 * Set up checksum offload. Note: checksum offload bits must
1828 * appear in all descriptors of a multi-descriptor transmit
1829 * attempt. (This is according to testing done with an 8169
1830 * chip. I'm not sure if this is a requirement or a bug.)
1833 arg.re_flags = 0;
1835 if (m->m_pkthdr.csum_flags & CSUM_IP)
1836 arg.re_flags |= RE_TDESC_CMD_IPCSUM;
1837 if (m->m_pkthdr.csum_flags & CSUM_TCP)
1838 arg.re_flags |= RE_TDESC_CMD_TCPCSUM;
1839 if (m->m_pkthdr.csum_flags & CSUM_UDP)
1840 arg.re_flags |= RE_TDESC_CMD_UDPCSUM;
1842 arg.sc = sc;
1843 arg.re_idx = *idx;
1844 arg.re_maxsegs = sc->re_ldata.re_tx_free;
1845 if (arg.re_maxsegs > 4)
1846 arg.re_maxsegs -= 4;
1847 arg.re_ring = sc->re_ldata.re_tx_list;
1849 map = sc->re_ldata.re_tx_dmamap[*idx];
1852 * With some of the RealTek chips, using the checksum offload
1853 * support in conjunction with the autopadding feature results
1854 * in the transmission of corrupt frames. For example, if we
1855 * need to send a really small IP fragment that's less than 60
1856 * bytes in size, and IP header checksumming is enabled, the
1857 * resulting ethernet frame that appears on the wire will
1858 * have garbled payload. To work around this, if TX checksum
1859 * offload is enabled, we always manually pad short frames out
1860 * to the minimum ethernet frame size. We do this by pretending
1861 * the mbuf chain has too many fragments so the coalescing code
1862 * below can assemble the packet into a single buffer that's
1863 * padded out to the mininum frame size.
1865 * Note: this appears unnecessary for TCP, and doing it for TCP
1866 * with PCIe adapters seems to result in bad checksums.
1868 if (arg.re_flags && !(arg.re_flags & RE_TDESC_CMD_TCPCSUM) &&
1869 m->m_pkthdr.len < RE_MIN_FRAMELEN) {
1870 error = EFBIG;
1871 } else {
1872 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1873 m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1876 if (error && error != EFBIG) {
1877 if_printf(ifp, "can't map mbuf (error %d)\n", error);
1878 return(ENOBUFS);
1881 /* Too many segments to map, coalesce into a single mbuf */
1883 if (error || arg.re_maxsegs == 0) {
1884 m_new = m_defrag_nofree(m, MB_DONTWAIT);
1885 if (m_new == NULL) {
1886 return(1);
1887 } else {
1888 m = m_new;
1889 *m_head = m;
1893 * Manually pad short frames, and zero the pad space
1894 * to avoid leaking data.
1896 if (m_new->m_pkthdr.len < RE_MIN_FRAMELEN) {
1897 bzero(mtod(m_new, char *) + m_new->m_pkthdr.len,
1898 RE_MIN_FRAMELEN - m_new->m_pkthdr.len);
1899 m_new->m_pkthdr.len += RE_MIN_FRAMELEN -
1900 m_new->m_pkthdr.len;
1901 m_new->m_len = m_new->m_pkthdr.len;
1904 *called_defrag = 1;
1905 arg.sc = sc;
1906 arg.re_idx = *idx;
1907 arg.re_maxsegs = sc->re_ldata.re_tx_free;
1908 arg.re_ring = sc->re_ldata.re_tx_list;
1910 error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map,
1911 m, re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1912 if (error) {
1913 m_freem(m);
1914 if_printf(ifp, "can't map mbuf (error %d)\n", error);
1915 return(EFBIG);
1920 * Insure that the map for this transmission
1921 * is placed at the array index of the last descriptor
1922 * in this chain.
1924 sc->re_ldata.re_tx_dmamap[*idx] =
1925 sc->re_ldata.re_tx_dmamap[arg.re_idx];
1926 sc->re_ldata.re_tx_dmamap[arg.re_idx] = map;
1928 sc->re_ldata.re_tx_mbuf[arg.re_idx] = m;
1929 sc->re_ldata.re_tx_free -= arg.re_maxsegs;
1932 * Set up hardware VLAN tagging. Note: vlan tag info must
1933 * appear in the first descriptor of a multi-descriptor
1934 * transmission attempt.
1937 if ((m->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1938 m->m_pkthdr.rcvif != NULL &&
1939 m->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1940 struct ifvlan *ifv;
1941 ifv = m->m_pkthdr.rcvif->if_softc;
1942 if (ifv != NULL)
1943 sc->re_ldata.re_tx_list[*idx].re_vlanctl =
1944 htole32(htobe16(ifv->ifv_tag) | RE_TDESC_VLANCTL_TAG);
1947 /* Transfer ownership of packet to the chip. */
1949 sc->re_ldata.re_tx_list[arg.re_idx].re_cmdstat |=
1950 htole32(RE_TDESC_CMD_OWN);
1951 if (*idx != arg.re_idx)
1952 sc->re_ldata.re_tx_list[*idx].re_cmdstat |=
1953 htole32(RE_TDESC_CMD_OWN);
1955 RE_DESC_INC(arg.re_idx);
1956 *idx = arg.re_idx;
1958 return(0);
1962 * Main transmit routine for C+ and gigE NICs.
1965 static void
1966 re_start(struct ifnet *ifp)
1968 struct re_softc *sc = ifp->if_softc;
1969 struct mbuf *m_head;
1970 struct mbuf *m_head2;
1971 int called_defrag, idx, need_trans;
1973 if (!sc->re_link || (ifp->if_flags & IFF_OACTIVE))
1974 return;
1976 idx = sc->re_ldata.re_tx_prodidx;
1978 need_trans = 0;
1979 while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
1980 m_head = ifq_poll(&ifp->if_snd);
1981 if (m_head == NULL)
1982 break;
1983 m_head2 = m_head;
1984 if (re_encap(sc, &m_head2, &idx, &called_defrag)) {
1986 * If we could not encapsulate the defragged packet,
1987 * the returned m_head2 is garbage and we must dequeue
1988 * and throw away the original packet.
1990 if (called_defrag) {
1991 ifq_dequeue(&ifp->if_snd, m_head);
1992 m_freem(m_head);
1994 ifp->if_flags |= IFF_OACTIVE;
1995 break;
1999 * Clean out the packet we encapsulated. If we defragged
2000 * the packet the m_head2 is the one that got encapsulated
2001 * and the original must be thrown away. Otherwise m_head2
2002 * *IS* the original.
2004 ifq_dequeue(&ifp->if_snd, m_head);
2005 if (called_defrag)
2006 m_freem(m_head);
2007 need_trans = 1;
2010 * If there's a BPF listener, bounce a copy of this frame
2011 * to him.
2013 BPF_MTAP(ifp, m_head2);
2016 if (!need_trans) {
2017 if (RE_TX_MODERATION_IS_ENABLED(sc) &&
2018 sc->re_ldata.re_tx_free != RE_TX_DESC_CNT)
2019 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2020 return;
2023 /* Flush the TX descriptors */
2024 bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2025 sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2027 sc->re_ldata.re_tx_prodidx = idx;
2030 * RealTek put the TX poll request register in a different
2031 * location on the 8169 gigE chip. I don't know why.
2033 CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2035 if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2037 * Use the countdown timer for interrupt moderation.
2038 * 'TX done' interrupts are disabled. Instead, we reset the
2039 * countdown timer, which will begin counting until it hits
2040 * the value in the TIMERINT register, and then trigger an
2041 * interrupt. Each time we write to the TIMERCNT register,
2042 * the timer count is reset to 0.
2044 CSR_WRITE_4(sc, RE_TIMERCNT, 1);
2048 * Set a timeout in case the chip goes out to lunch.
2050 ifp->if_timer = 5;
2053 static void
2054 re_init(void *xsc)
2056 struct re_softc *sc = xsc;
2057 struct ifnet *ifp = &sc->arpcom.ac_if;
2058 struct mii_data *mii;
2059 uint32_t rxcfg = 0;
2061 mii = device_get_softc(sc->re_miibus);
2064 * Cancel pending I/O and free all RX/TX buffers.
2066 re_stop(sc);
2069 * Enable C+ RX and TX mode, as well as VLAN stripping and
2070 * RX checksum offload. We must configure the C+ register
2071 * before all others.
2073 CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2074 RE_CPLUSCMD_PCI_MRW | RE_CPLUSCMD_VLANSTRIP |
2075 (ifp->if_capenable & IFCAP_RXCSUM ?
2076 RE_CPLUSCMD_RXCSUM_ENB : 0));
2079 * Init our MAC address. Even though the chipset
2080 * documentation doesn't mention it, we need to enter "Config
2081 * register write enable" mode to modify the ID registers.
2083 CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2084 CSR_WRITE_4(sc, RE_IDR0,
2085 htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2086 CSR_WRITE_2(sc, RE_IDR4,
2087 htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2088 CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2091 * For C+ mode, initialize the RX descriptors and mbufs.
2093 re_rx_list_init(sc);
2094 re_tx_list_init(sc);
2097 * Load the addresses of the RX and TX lists into the chip.
2099 CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2100 RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2101 CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2102 RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2104 CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2105 RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2106 CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2107 RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2110 * Enable transmit and receive.
2112 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2115 * Set the initial TX and RX configuration.
2117 if (sc->re_testmode) {
2118 if (sc->re_type == RE_8169)
2119 CSR_WRITE_4(sc, RE_TXCFG,
2120 RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2121 else
2122 CSR_WRITE_4(sc, RE_TXCFG,
2123 RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2124 } else
2125 CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2127 CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, 16);
2129 CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2131 /* Set the individual bit to receive frames for this host only. */
2132 rxcfg = CSR_READ_4(sc, RE_RXCFG);
2133 rxcfg |= RE_RXCFG_RX_INDIV;
2135 /* If we want promiscuous mode, set the allframes bit. */
2136 if (ifp->if_flags & IFF_PROMISC) {
2137 rxcfg |= RE_RXCFG_RX_ALLPHYS;
2138 CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2139 } else {
2140 rxcfg &= ~RE_RXCFG_RX_ALLPHYS;
2141 CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2145 * Set capture broadcast bit to capture broadcast frames.
2147 if (ifp->if_flags & IFF_BROADCAST) {
2148 rxcfg |= RE_RXCFG_RX_BROAD;
2149 CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2150 } else {
2151 rxcfg &= ~RE_RXCFG_RX_BROAD;
2152 CSR_WRITE_4(sc, RE_RXCFG, rxcfg);
2156 * Program the multicast filter, if necessary.
2158 re_setmulti(sc);
2160 #ifdef DEVICE_POLLING
2162 * Disable interrupts if we are polling.
2164 if (ifp->if_flags & IFF_POLLING)
2165 CSR_WRITE_2(sc, RE_IMR, 0);
2166 else /* otherwise ... */
2167 #endif /* DEVICE_POLLING */
2169 * Enable interrupts.
2171 if (sc->re_testmode)
2172 CSR_WRITE_2(sc, RE_IMR, 0);
2173 else
2174 CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
2175 CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2177 /* Set initial TX threshold */
2178 sc->re_txthresh = RE_TX_THRESH_INIT;
2180 /* Start RX/TX process. */
2181 if (sc->re_flags & RE_F_HASMPC)
2182 CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2183 #ifdef notdef
2184 /* Enable receiver and transmitter. */
2185 CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2186 #endif
2188 if (RE_TX_MODERATION_IS_ENABLED(sc)) {
2190 * Initialize the timer interrupt register so that
2191 * a timer interrupt will be generated once the timer
2192 * reaches a certain number of ticks. The timer is
2193 * reloaded on each transmit. This gives us TX interrupt
2194 * moderation, which dramatically improves TX frame rate.
2196 if (sc->re_type == RE_8169)
2197 CSR_WRITE_4(sc, RE_TIMERINT_8169, 0x800);
2198 else
2199 CSR_WRITE_4(sc, RE_TIMERINT, 0x400);
2203 * For 8169 gigE NICs, set the max allowed RX packet
2204 * size so we can receive jumbo frames.
2206 if (sc->re_type == RE_8169)
2207 CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2209 if (sc->re_testmode) {
2210 return;
2213 mii_mediachg(mii);
2215 CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2217 ifp->if_flags |= IFF_RUNNING;
2218 ifp->if_flags &= ~IFF_OACTIVE;
2220 sc->re_link = 0;
2221 callout_reset(&sc->re_timer, hz, re_tick, sc);
2225 * Set media options.
2227 static int
2228 re_ifmedia_upd(struct ifnet *ifp)
2230 struct re_softc *sc = ifp->if_softc;
2231 struct mii_data *mii;
2233 mii = device_get_softc(sc->re_miibus);
2234 mii_mediachg(mii);
2236 return(0);
2240 * Report current media status.
2242 static void
2243 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2245 struct re_softc *sc = ifp->if_softc;
2246 struct mii_data *mii;
2248 mii = device_get_softc(sc->re_miibus);
2250 mii_pollstat(mii);
2251 ifmr->ifm_active = mii->mii_media_active;
2252 ifmr->ifm_status = mii->mii_media_status;
2255 static int
2256 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2258 struct re_softc *sc = ifp->if_softc;
2259 struct ifreq *ifr = (struct ifreq *) data;
2260 struct mii_data *mii;
2261 int error = 0;
2263 switch(command) {
2264 case SIOCSIFMTU:
2265 if (ifr->ifr_mtu > RE_JUMBO_MTU)
2266 error = EINVAL;
2267 ifp->if_mtu = ifr->ifr_mtu;
2268 break;
2269 case SIOCSIFFLAGS:
2270 if (ifp->if_flags & IFF_UP)
2271 re_init(sc);
2272 else if (ifp->if_flags & IFF_RUNNING)
2273 re_stop(sc);
2274 break;
2275 case SIOCADDMULTI:
2276 case SIOCDELMULTI:
2277 re_setmulti(sc);
2278 error = 0;
2279 break;
2280 case SIOCGIFMEDIA:
2281 case SIOCSIFMEDIA:
2282 mii = device_get_softc(sc->re_miibus);
2283 error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2284 break;
2285 case SIOCSIFCAP:
2286 ifp->if_capenable &= ~(IFCAP_HWCSUM);
2287 ifp->if_capenable |=
2288 ifr->ifr_reqcap & (IFCAP_HWCSUM);
2289 if (ifp->if_capenable & IFCAP_TXCSUM)
2290 ifp->if_hwassist = RE_CSUM_FEATURES;
2291 else
2292 ifp->if_hwassist = 0;
2293 if (ifp->if_flags & IFF_RUNNING)
2294 re_init(sc);
2295 break;
2296 default:
2297 error = ether_ioctl(ifp, command, data);
2298 break;
2300 return(error);
2303 static void
2304 re_watchdog(struct ifnet *ifp)
2306 struct re_softc *sc = ifp->if_softc;
2308 if_printf(ifp, "watchdog timeout\n");
2310 ifp->if_oerrors++;
2312 re_txeof(sc);
2313 re_rxeof(sc);
2315 re_init(sc);
2317 if (!ifq_is_empty(&ifp->if_snd))
2318 ifp->if_start(ifp);
2322 * Stop the adapter and free any mbufs allocated to the
2323 * RX and TX lists.
2325 static void
2326 re_stop(struct re_softc *sc)
2328 struct ifnet *ifp = &sc->arpcom.ac_if;
2329 int i;
2331 ifp->if_timer = 0;
2332 callout_stop(&sc->re_timer);
2334 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2336 CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2337 CSR_WRITE_2(sc, RE_IMR, 0x0000);
2338 CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2340 if (sc->re_head != NULL) {
2341 m_freem(sc->re_head);
2342 sc->re_head = sc->re_tail = NULL;
2345 /* Free the TX list buffers. */
2346 for (i = 0; i < RE_TX_DESC_CNT; i++) {
2347 if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2348 bus_dmamap_unload(sc->re_ldata.re_mtag,
2349 sc->re_ldata.re_tx_dmamap[i]);
2350 m_freem(sc->re_ldata.re_tx_mbuf[i]);
2351 sc->re_ldata.re_tx_mbuf[i] = NULL;
2355 /* Free the RX list buffers. */
2356 for (i = 0; i < RE_RX_DESC_CNT; i++) {
2357 if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2358 bus_dmamap_unload(sc->re_ldata.re_mtag,
2359 sc->re_ldata.re_rx_dmamap[i]);
2360 m_freem(sc->re_ldata.re_rx_mbuf[i]);
2361 sc->re_ldata.re_rx_mbuf[i] = NULL;
2367 * Device suspend routine. Stop the interface and save some PCI
2368 * settings in case the BIOS doesn't restore them properly on
2369 * resume.
2371 static int
2372 re_suspend(device_t dev)
2374 #ifndef BURN_BRIDGES
2375 int i;
2376 #endif
2377 struct re_softc *sc = device_get_softc(dev);
2379 re_stop(sc);
2381 #ifndef BURN_BRIDGES
2382 for (i = 0; i < 5; i++)
2383 sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
2384 sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
2385 sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
2386 sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
2387 sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
2388 #endif
2390 sc->suspended = 1;
2392 return (0);
2396 * Device resume routine. Restore some PCI settings in case the BIOS
2397 * doesn't, re-enable busmastering, and restart the interface if
2398 * appropriate.
2400 static int
2401 re_resume(device_t dev)
2403 struct re_softc *sc = device_get_softc(dev);
2404 struct ifnet *ifp = &sc->arpcom.ac_if;
2405 #ifndef BURN_BRIDGES
2406 int i;
2407 #endif
2409 #ifndef BURN_BRIDGES
2410 /* better way to do this? */
2411 for (i = 0; i < 5; i++)
2412 pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
2413 pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
2414 pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
2415 pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
2416 pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
2418 /* reenable busmastering */
2419 pci_enable_busmaster(dev);
2420 pci_enable_io(dev, SYS_RES_IOPORT);
2421 #endif
2423 /* reinitialize interface if necessary */
2424 if (ifp->if_flags & IFF_UP)
2425 re_init(sc);
2427 sc->suspended = 0;
2429 return (0);
2433 * Stop all chip I/O so that the kernel's probe routines don't
2434 * get confused by errant DMAs when rebooting.
2436 static void
2437 re_shutdown(device_t dev)
2439 struct re_softc *sc = device_get_softc(dev);
2440 struct ifnet *ifp = &sc->arpcom.ac_if;
2442 lwkt_serialize_enter(ifp->if_serializer);
2443 re_stop(sc);
2444 lwkt_serialize_exit(ifp->if_serializer);
2447 static int
2448 re_sysctl_tx_moderation(SYSCTL_HANDLER_ARGS)
2450 struct re_softc *sc = arg1;
2451 struct ifnet *ifp = &sc->arpcom.ac_if;
2452 int error = 0, mod, mod_old;
2454 lwkt_serialize_enter(ifp->if_serializer);
2456 mod_old = mod = RE_TX_MODERATION_IS_ENABLED(sc);
2458 error = sysctl_handle_int(oidp, &mod, 0, req);
2459 if (error || req->newptr == NULL || mod == mod_old)
2460 goto back;
2461 if (mod != 0 && mod != 1) {
2462 error = EINVAL;
2463 goto back;
2466 if (mod)
2467 RE_ENABLE_TX_MODERATION(sc);
2468 else
2469 RE_DISABLE_TX_MODERATION(sc);
2471 if ((ifp->if_flags & (IFF_RUNNING | IFF_UP)) == (IFF_RUNNING | IFF_UP))
2472 re_init(sc);
2473 back:
2474 lwkt_serialize_exit(ifp->if_serializer);
2475 return error;