xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision cbc6ab280002ab85b14d16c2e8e3df38130702d5)
1 /*
2  * Copyright (c) 2004
3  *	Joerg Sonnenberger <joerg@bec.de>.  All rights reserved.
4  *
5  * Copyright (c) 1997, 1998-2003
6  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
7  *
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.
22  *
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.
34  *
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 1.98 2008/10/28 07:23:28 sephe Exp $
37  */
38 
39 /*
40  * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
41  *
42  * Written by Bill Paul <wpaul@windriver.com>
43  * Senior Networking Software Engineer
44  * Wind River Systems
45  */
46 
47 /*
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.
52  *
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:
57  *
58  *	o Descriptor based DMA mechanism. Each descriptor represents
59  *	  a single packet fragment. Data buffers may be aligned on
60  *	  any byte boundary.
61  *
62  *	o 64-bit DMA
63  *
64  *	o TCP/IP checksum offload for both RX and TX
65  *
66  *	o High and normal priority transmit DMA rings
67  *
68  *	o VLAN tag insertion and extraction
69  *
70  *	o TCP large send (segmentation offload)
71  *
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.
76  *
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:
80  *
81  *	o 1000Mbps mode
82  *
83  *	o Jumbo frames
84  *
85  * 	o GMII and TBI ports/registers for interfacing with copper
86  *	  or fiber PHYs
87  *
88  *      o RX and TX DMA rings can have up to 1024 descriptors
89  *        (the 8139C+ allows a maximum of 64)
90  *
91  *	o Slight differences in register layout from the 8139C+
92  *
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.
98  *
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.
104  *
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.
112  */
113 
114 #define _IP_VHL
115 
116 #include "opt_polling.h"
117 
118 #include <sys/param.h>
119 #include <sys/bus.h>
120 #include <sys/endian.h>
121 #include <sys/kernel.h>
122 #include <sys/in_cksum.h>
123 #include <sys/interrupt.h>
124 #include <sys/malloc.h>
125 #include <sys/mbuf.h>
126 #include <sys/rman.h>
127 #include <sys/serialize.h>
128 #include <sys/socket.h>
129 #include <sys/sockio.h>
130 #include <sys/sysctl.h>
131 
132 #include <net/bpf.h>
133 #include <net/ethernet.h>
134 #include <net/if.h>
135 #include <net/ifq_var.h>
136 #include <net/if_arp.h>
137 #include <net/if_dl.h>
138 #include <net/if_media.h>
139 #include <net/if_types.h>
140 #include <net/vlan/if_vlan_var.h>
141 #include <net/vlan/if_vlan_ether.h>
142 
143 #include <netinet/ip.h>
144 
145 #include <dev/netif/mii_layer/mii.h>
146 #include <dev/netif/mii_layer/miivar.h>
147 
148 #include <bus/pci/pcidevs.h>
149 #include <bus/pci/pcireg.h>
150 #include <bus/pci/pcivar.h>
151 
152 /* "device miibus" required.  See GENERIC if you get errors here. */
153 #include "miibus_if.h"
154 
155 #include <dev/netif/re/if_rereg.h>
156 #include <dev/netif/re/if_revar.h>
157 
158 #define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
159 
160 /*
161  * Various supported device vendors/types and their names.
162  */
163 static const struct re_type {
164 	uint16_t	re_vid;
165 	uint16_t	re_did;
166 	const char	*re_name;
167 } re_devs[] = {
168 	{ PCI_VENDOR_DLINK, PCI_PRODUCT_DLINK_DGE528T,
169 	  "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
170 
171 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8139,
172 	  "RealTek 8139C+ 10/100BaseTX" },
173 
174 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8101E,
175 	  "RealTek 810x PCIe 10/100baseTX" },
176 
177 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8168,
178 	  "RealTek 8111/8168 PCIe Gigabit Ethernet" },
179 
180 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169,
181 	  "RealTek 8110/8169 Gigabit Ethernet" },
182 
183 	{ PCI_VENDOR_REALTEK, PCI_PRODUCT_REALTEK_RT8169SC,
184 	  "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
185 
186 	{ PCI_VENDOR_COREGA, PCI_PRODUCT_COREGA_CG_LAPCIGT,
187 	  "Corega CG-LAPCIGT Gigabit Ethernet" },
188 
189 	{ PCI_VENDOR_LINKSYS, PCI_PRODUCT_LINKSYS_EG1032,
190 	  "Linksys EG1032 Gigabit Ethernet" },
191 
192 	{ PCI_VENDOR_USR2, PCI_PRODUCT_USR2_997902,
193 	  "US Robotics 997902 Gigabit Ethernet" },
194 
195 	{ 0, 0, NULL }
196 };
197 
198 static const struct re_hwrev re_hwrevs[] = {
199 	{ RE_HWREV_8139CPLUS,	RE_MACVER_UNKN,		ETHERMTU,
200 	  RE_C_HWCSUM | RE_C_8139CP },
201 
202 	{ RE_HWREV_8169,	RE_MACVER_UNKN,		RE_MTU_6K,
203 	  RE_C_HWCSUM | RE_C_8169 },
204 
205 	{ RE_HWREV_8110S,	RE_MACVER_03,		RE_MTU_6K,
206 	  RE_C_HWCSUM | RE_C_8169 },
207 
208 	{ RE_HWREV_8169S,	RE_MACVER_03,		RE_MTU_6K,
209 	  RE_C_HWCSUM | RE_C_8169 },
210 
211 	{ RE_HWREV_8169SB,	RE_MACVER_04,		RE_MTU_6K,
212 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
213 
214 	{ RE_HWREV_8169SC1,	RE_MACVER_05,		RE_MTU_6K,
215 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
216 
217 	{ RE_HWREV_8169SC2,	RE_MACVER_06,		RE_MTU_6K,
218 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
219 
220 	{ RE_HWREV_8168B1,	RE_MACVER_21,		RE_MTU_6K,
221 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT },
222 
223 	{ RE_HWREV_8168B2,	RE_MACVER_23,		RE_MTU_6K,
224 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
225 
226 	{ RE_HWREV_8168B3,	RE_MACVER_23,		RE_MTU_6K,
227 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
228 
229 	{ RE_HWREV_8168C,	RE_MACVER_29,		RE_MTU_6K,
230 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
231 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
232 
233 	{ RE_HWREV_8168CP,	RE_MACVER_2B,		RE_MTU_6K,
234 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
235 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
236 
237 	{ RE_HWREV_8168D,	RE_MACVER_2A,		RE_MTU_9K,
238 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
239 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
240 
241 	{ RE_HWREV_8100E,	RE_MACVER_UNKN,		ETHERMTU,
242 	  RE_C_HWCSUM },
243 
244 	{ RE_HWREV_8101E1,	RE_MACVER_16,		ETHERMTU,
245 	  RE_C_HWCSUM },
246 
247 	{ RE_HWREV_8101E2,	RE_MACVER_16,		ETHERMTU,
248 	  RE_C_HWCSUM },
249 
250 	{ RE_HWREV_8102E,	RE_MACVER_15,		ETHERMTU,
251 	  RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX },
252 
253 	{ RE_HWREV_8102EL,	RE_MACVER_15,		ETHERMTU,
254 	  RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX },
255 
256 	{ RE_HWREV_NULL, 0, 0, 0 }
257 };
258 
259 static int	re_probe(device_t);
260 static int	re_attach(device_t);
261 static int	re_detach(device_t);
262 static int	re_suspend(device_t);
263 static int	re_resume(device_t);
264 static void	re_shutdown(device_t);
265 
266 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
267 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
268 				bus_size_t, int);
269 static int	re_allocmem(device_t);
270 static void	re_freemem(device_t);
271 static void	re_freebufmem(struct re_softc *, int, int);
272 static int	re_encap(struct re_softc *, struct mbuf **, int *);
273 static int	re_newbuf_std(struct re_softc *, int, int);
274 static int	re_newbuf_jumbo(struct re_softc *, int, int);
275 static void	re_setup_rxdesc(struct re_softc *, int);
276 static int	re_rx_list_init(struct re_softc *);
277 static int	re_tx_list_init(struct re_softc *);
278 static int	re_rxeof(struct re_softc *);
279 static int	re_txeof(struct re_softc *);
280 static int	re_tx_collect(struct re_softc *);
281 static void	re_intr(void *);
282 static void	re_tick(void *);
283 static void	re_tick_serialized(void *);
284 
285 static void	re_start(struct ifnet *);
286 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
287 static void	re_init(void *);
288 static void	re_stop(struct re_softc *);
289 static void	re_watchdog(struct ifnet *);
290 static int	re_ifmedia_upd(struct ifnet *);
291 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
292 
293 static void	re_eeprom_putbyte(struct re_softc *, int);
294 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
295 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
296 static void	re_get_eewidth(struct re_softc *);
297 
298 static int	re_gmii_readreg(device_t, int, int);
299 static int	re_gmii_writereg(device_t, int, int, int);
300 
301 static int	re_miibus_readreg(device_t, int, int);
302 static int	re_miibus_writereg(device_t, int, int, int);
303 static void	re_miibus_statchg(device_t);
304 
305 static void	re_setmulti(struct re_softc *);
306 static void	re_reset(struct re_softc *, int);
307 static void	re_get_eaddr(struct re_softc *, uint8_t *);
308 
309 static void	re_setup_hw_im(struct re_softc *);
310 static void	re_setup_sim_im(struct re_softc *);
311 static void	re_disable_hw_im(struct re_softc *);
312 static void	re_disable_sim_im(struct re_softc *);
313 static void	re_config_imtype(struct re_softc *, int);
314 static void	re_setup_intr(struct re_softc *, int, int);
315 
316 static int	re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *);
317 static int	re_sysctl_rxtime(SYSCTL_HANDLER_ARGS);
318 static int	re_sysctl_txtime(SYSCTL_HANDLER_ARGS);
319 static int	re_sysctl_simtime(SYSCTL_HANDLER_ARGS);
320 static int	re_sysctl_imtype(SYSCTL_HANDLER_ARGS);
321 
322 static int	re_jpool_alloc(struct re_softc *);
323 static void	re_jpool_free(struct re_softc *);
324 static struct re_jbuf *re_jbuf_alloc(struct re_softc *);
325 static void	re_jbuf_free(void *);
326 static void	re_jbuf_ref(void *);
327 
328 #ifdef RE_DIAG
329 static int	re_diag(struct re_softc *);
330 #endif
331 
332 #ifdef DEVICE_POLLING
333 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
334 #endif
335 
336 static device_method_t re_methods[] = {
337 	/* Device interface */
338 	DEVMETHOD(device_probe,		re_probe),
339 	DEVMETHOD(device_attach,	re_attach),
340 	DEVMETHOD(device_detach,	re_detach),
341 	DEVMETHOD(device_suspend,	re_suspend),
342 	DEVMETHOD(device_resume,	re_resume),
343 	DEVMETHOD(device_shutdown,	re_shutdown),
344 
345 	/* bus interface */
346 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
347 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
348 
349 	/* MII interface */
350 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
351 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
352 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
353 
354 	{ 0, 0 }
355 };
356 
357 static driver_t re_driver = {
358 	"re",
359 	re_methods,
360 	sizeof(struct re_softc)
361 };
362 
363 static devclass_t re_devclass;
364 
365 DECLARE_DUMMY_MODULE(if_re);
366 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
367 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
368 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
369 
370 static int	re_rx_desc_count = RE_RX_DESC_CNT_DEF;
371 static int	re_tx_desc_count = RE_TX_DESC_CNT_DEF;
372 
373 TUNABLE_INT("hw.re.rx_desc_count", &re_rx_desc_count);
374 TUNABLE_INT("hw.re.tx_desc_count", &re_tx_desc_count);
375 
376 #define EE_SET(x)	\
377 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
378 
379 #define EE_CLR(x)	\
380 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
381 
382 static __inline void
383 re_free_rxchain(struct re_softc *sc)
384 {
385 	if (sc->re_head != NULL) {
386 		m_freem(sc->re_head);
387 		sc->re_head = sc->re_tail = NULL;
388 	}
389 }
390 
391 /*
392  * Send a read command and address to the EEPROM, check for ACK.
393  */
394 static void
395 re_eeprom_putbyte(struct re_softc *sc, int addr)
396 {
397 	int d, i;
398 
399 	d = addr | (RE_9346_READ << sc->re_eewidth);
400 
401 	/*
402 	 * Feed in each bit and strobe the clock.
403 	 */
404 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
405 		if (d & i)
406 			EE_SET(RE_EE_DATAIN);
407 		else
408 			EE_CLR(RE_EE_DATAIN);
409 		DELAY(100);
410 		EE_SET(RE_EE_CLK);
411 		DELAY(150);
412 		EE_CLR(RE_EE_CLK);
413 		DELAY(100);
414 	}
415 }
416 
417 /*
418  * Read a word of data stored in the EEPROM at address 'addr.'
419  */
420 static void
421 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
422 {
423 	int i;
424 	uint16_t word = 0;
425 
426 	/*
427 	 * Send address of word we want to read.
428 	 */
429 	re_eeprom_putbyte(sc, addr);
430 
431 	/*
432 	 * Start reading bits from EEPROM.
433 	 */
434 	for (i = 0x8000; i != 0; i >>= 1) {
435 		EE_SET(RE_EE_CLK);
436 		DELAY(100);
437 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
438 			word |= i;
439 		EE_CLR(RE_EE_CLK);
440 		DELAY(100);
441 	}
442 
443 	*dest = word;
444 }
445 
446 /*
447  * Read a sequence of words from the EEPROM.
448  */
449 static void
450 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
451 {
452 	int i;
453 	uint16_t word = 0, *ptr;
454 
455 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
456 	DELAY(100);
457 
458 	for (i = 0; i < cnt; i++) {
459 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
460 		re_eeprom_getword(sc, off + i, &word);
461 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
462 		ptr = (uint16_t *)(dest + (i * 2));
463 		*ptr = word;
464 	}
465 
466 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
467 }
468 
469 static void
470 re_get_eewidth(struct re_softc *sc)
471 {
472 	uint16_t re_did = 0;
473 
474 	sc->re_eewidth = 6;
475 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
476 	if (re_did != 0x8129)
477 		sc->re_eewidth = 8;
478 }
479 
480 static int
481 re_gmii_readreg(device_t dev, int phy, int reg)
482 {
483 	struct re_softc *sc = device_get_softc(dev);
484 	u_int32_t rval;
485 	int i;
486 
487 	if (phy != 1)
488 		return(0);
489 
490 	/* Let the rgephy driver read the GMEDIASTAT register */
491 
492 	if (reg == RE_GMEDIASTAT)
493 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
494 
495 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
496 	DELAY(1000);
497 
498 	for (i = 0; i < RE_TIMEOUT; i++) {
499 		rval = CSR_READ_4(sc, RE_PHYAR);
500 		if (rval & RE_PHYAR_BUSY)
501 			break;
502 		DELAY(100);
503 	}
504 
505 	if (i == RE_TIMEOUT) {
506 		device_printf(dev, "PHY read failed\n");
507 		return(0);
508 	}
509 
510 	return(rval & RE_PHYAR_PHYDATA);
511 }
512 
513 static int
514 re_gmii_writereg(device_t dev, int phy, int reg, int data)
515 {
516 	struct re_softc *sc = device_get_softc(dev);
517 	uint32_t rval;
518 	int i;
519 
520 	CSR_WRITE_4(sc, RE_PHYAR,
521 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
522 	DELAY(1000);
523 
524 	for (i = 0; i < RE_TIMEOUT; i++) {
525 		rval = CSR_READ_4(sc, RE_PHYAR);
526 		if ((rval & RE_PHYAR_BUSY) == 0)
527 			break;
528 		DELAY(100);
529 	}
530 
531 	if (i == RE_TIMEOUT)
532 		device_printf(dev, "PHY write failed\n");
533 
534 	return(0);
535 }
536 
537 static int
538 re_miibus_readreg(device_t dev, int phy, int reg)
539 {
540 	struct re_softc	*sc = device_get_softc(dev);
541 	uint16_t rval = 0;
542 	uint16_t re8139_reg = 0;
543 
544 	if (!RE_IS_8139CP(sc)) {
545 		rval = re_gmii_readreg(dev, phy, reg);
546 		return(rval);
547 	}
548 
549 	/* Pretend the internal PHY is only at address 0 */
550 	if (phy)
551 		return(0);
552 
553 	switch(reg) {
554 	case MII_BMCR:
555 		re8139_reg = RE_BMCR;
556 		break;
557 	case MII_BMSR:
558 		re8139_reg = RE_BMSR;
559 		break;
560 	case MII_ANAR:
561 		re8139_reg = RE_ANAR;
562 		break;
563 	case MII_ANER:
564 		re8139_reg = RE_ANER;
565 		break;
566 	case MII_ANLPAR:
567 		re8139_reg = RE_LPAR;
568 		break;
569 	case MII_PHYIDR1:
570 	case MII_PHYIDR2:
571 		return(0);
572 	/*
573 	 * Allow the rlphy driver to read the media status
574 	 * register. If we have a link partner which does not
575 	 * support NWAY, this is the register which will tell
576 	 * us the results of parallel detection.
577 	 */
578 	case RE_MEDIASTAT:
579 		return(CSR_READ_1(sc, RE_MEDIASTAT));
580 	default:
581 		device_printf(dev, "bad phy register\n");
582 		return(0);
583 	}
584 	rval = CSR_READ_2(sc, re8139_reg);
585 	if (re8139_reg == RE_BMCR) {
586 		/* 8139C+ has different bit layout. */
587 		rval &= ~(BMCR_LOOP | BMCR_ISO);
588 	}
589 	return(rval);
590 }
591 
592 static int
593 re_miibus_writereg(device_t dev, int phy, int reg, int data)
594 {
595 	struct re_softc *sc= device_get_softc(dev);
596 	u_int16_t re8139_reg = 0;
597 
598 	if (!RE_IS_8139CP(sc))
599 		return(re_gmii_writereg(dev, phy, reg, data));
600 
601 	/* Pretend the internal PHY is only at address 0 */
602 	if (phy)
603 		return(0);
604 
605 	switch(reg) {
606 	case MII_BMCR:
607 		re8139_reg = RE_BMCR;
608 		/* 8139C+ has different bit layout. */
609 		data &= ~(BMCR_LOOP | BMCR_ISO);
610 		break;
611 	case MII_BMSR:
612 		re8139_reg = RE_BMSR;
613 		break;
614 	case MII_ANAR:
615 		re8139_reg = RE_ANAR;
616 		break;
617 	case MII_ANER:
618 		re8139_reg = RE_ANER;
619 		break;
620 	case MII_ANLPAR:
621 		re8139_reg = RE_LPAR;
622 		break;
623 	case MII_PHYIDR1:
624 	case MII_PHYIDR2:
625 		return(0);
626 	default:
627 		device_printf(dev, "bad phy register\n");
628 		return(0);
629 	}
630 	CSR_WRITE_2(sc, re8139_reg, data);
631 	return(0);
632 }
633 
634 static void
635 re_miibus_statchg(device_t dev)
636 {
637 }
638 
639 /*
640  * Program the 64-bit multicast hash filter.
641  */
642 static void
643 re_setmulti(struct re_softc *sc)
644 {
645 	struct ifnet *ifp = &sc->arpcom.ac_if;
646 	int h = 0;
647 	uint32_t hashes[2] = { 0, 0 };
648 	struct ifmultiaddr *ifma;
649 	uint32_t rxfilt;
650 	int mcnt = 0;
651 
652 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
653 
654 	/* Set the individual bit to receive frames for this host only. */
655 	rxfilt |= RE_RXCFG_RX_INDIV;
656 	/* Set capture broadcast bit to capture broadcast frames. */
657 	rxfilt |= RE_RXCFG_RX_BROAD;
658 
659 	rxfilt &= ~(RE_RXCFG_RX_ALLPHYS | RE_RXCFG_RX_MULTI);
660 	if ((ifp->if_flags & IFF_ALLMULTI) || (ifp->if_flags & IFF_PROMISC)) {
661 		rxfilt |= RE_RXCFG_RX_MULTI;
662 
663 		/* If we want promiscuous mode, set the allframes bit. */
664 		if (ifp->if_flags & IFF_PROMISC)
665 			rxfilt |= RE_RXCFG_RX_ALLPHYS;
666 
667 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
668 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
669 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
670 		return;
671 	}
672 
673 	/* first, zot all the existing hash bits */
674 	CSR_WRITE_4(sc, RE_MAR0, 0);
675 	CSR_WRITE_4(sc, RE_MAR4, 0);
676 
677 	/* now program new ones */
678 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
679 		if (ifma->ifma_addr->sa_family != AF_LINK)
680 			continue;
681 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
682 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
683 		if (h < 32)
684 			hashes[0] |= (1 << h);
685 		else
686 			hashes[1] |= (1 << (h - 32));
687 		mcnt++;
688 	}
689 
690 	if (mcnt)
691 		rxfilt |= RE_RXCFG_RX_MULTI;
692 	else
693 		rxfilt &= ~RE_RXCFG_RX_MULTI;
694 
695 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
696 
697 	/*
698 	 * For some unfathomable reason, RealTek decided to reverse
699 	 * the order of the multicast hash registers in the PCI Express
700 	 * parts. This means we have to write the hash pattern in reverse
701 	 * order for those devices.
702 	 */
703 	if (sc->re_caps & RE_C_PCIE) {
704 		CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
705 		CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
706 	} else {
707 		CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
708 		CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
709 	}
710 }
711 
712 static void
713 re_reset(struct re_softc *sc, int running)
714 {
715 	int i;
716 
717 	if ((sc->re_caps & RE_C_STOP_RXTX) && running) {
718 		CSR_WRITE_1(sc, RE_COMMAND,
719 			    RE_CMD_STOPREQ | RE_CMD_TX_ENB | RE_CMD_RX_ENB);
720 		DELAY(100);
721 	}
722 
723 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
724 
725 	for (i = 0; i < RE_TIMEOUT; i++) {
726 		DELAY(10);
727 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
728 			break;
729 	}
730 	if (i == RE_TIMEOUT)
731 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
732 }
733 
734 #ifdef RE_DIAG
735 /*
736  * The following routine is designed to test for a defect on some
737  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
738  * lines connected to the bus, however for a 32-bit only card, they
739  * should be pulled high. The result of this defect is that the
740  * NIC will not work right if you plug it into a 64-bit slot: DMA
741  * operations will be done with 64-bit transfers, which will fail
742  * because the 64-bit data lines aren't connected.
743  *
744  * There's no way to work around this (short of talking a soldering
745  * iron to the board), however we can detect it. The method we use
746  * here is to put the NIC into digital loopback mode, set the receiver
747  * to promiscuous mode, and then try to send a frame. We then compare
748  * the frame data we sent to what was received. If the data matches,
749  * then the NIC is working correctly, otherwise we know the user has
750  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
751  * slot. In the latter case, there's no way the NIC can work correctly,
752  * so we print out a message on the console and abort the device attach.
753  */
754 
755 static int
756 re_diag(struct re_softc *sc)
757 {
758 	struct ifnet *ifp = &sc->arpcom.ac_if;
759 	struct mbuf *m0;
760 	struct ether_header *eh;
761 	struct re_desc *cur_rx;
762 	uint16_t status;
763 	uint32_t rxstat;
764 	int total_len, i, error = 0, phyaddr;
765 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
766 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
767 
768 	/* Allocate a single mbuf */
769 
770 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
771 	if (m0 == NULL)
772 		return(ENOBUFS);
773 
774 	/*
775 	 * Initialize the NIC in test mode. This sets the chip up
776 	 * so that it can send and receive frames, but performs the
777 	 * following special functions:
778 	 * - Puts receiver in promiscuous mode
779 	 * - Enables digital loopback mode
780 	 * - Leaves interrupts turned off
781 	 */
782 
783 	ifp->if_flags |= IFF_PROMISC;
784 	sc->re_flags |= RE_F_TESTMODE;
785 	re_init(sc);
786 	sc->re_flags |= RE_F_LINKED;
787 	if (!RE_IS_8139CP(sc))
788 		phyaddr = 1;
789 	else
790 		phyaddr = 0;
791 
792 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
793 	for (i = 0; i < RE_TIMEOUT; i++) {
794 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
795 		if (!(status & BMCR_RESET))
796 			break;
797 	}
798 
799 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
800 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
801 
802 	DELAY(100000);
803 
804 	/* Put some data in the mbuf */
805 
806 	eh = mtod(m0, struct ether_header *);
807 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
808 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
809 	eh->ether_type = htons(ETHERTYPE_IP);
810 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
811 
812 	/*
813 	 * Queue the packet, start transmission.
814 	 * Note: ifq_handoff() ultimately calls re_start() for us.
815 	 */
816 
817 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
818 	error = ifq_handoff(ifp, m0, NULL);
819 	if (error) {
820 		m0 = NULL;
821 		goto done;
822 	}
823 	m0 = NULL;
824 
825 	/* Wait for it to propagate through the chip */
826 
827 	DELAY(100000);
828 	for (i = 0; i < RE_TIMEOUT; i++) {
829 		status = CSR_READ_2(sc, RE_ISR);
830 		CSR_WRITE_2(sc, RE_ISR, status);
831 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
832 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
833 			break;
834 		DELAY(10);
835 	}
836 
837 	if (i == RE_TIMEOUT) {
838 		if_printf(ifp, "diagnostic failed to receive packet "
839 			  "in loopback mode\n");
840 		error = EIO;
841 		goto done;
842 	}
843 
844 	/*
845 	 * The packet should have been dumped into the first
846 	 * entry in the RX DMA ring. Grab it from there.
847 	 */
848 
849 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
850 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
851 	bus_dmamap_sync(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0],
852 			BUS_DMASYNC_POSTWRITE);
853 	bus_dmamap_unload(sc->re_ldata.re_mtag, sc->re_ldata.re_rx_dmamap[0]);
854 
855 	m0 = sc->re_ldata.re_rx_mbuf[0];
856 	sc->re_ldata.re_rx_mbuf[0] = NULL;
857 	eh = mtod(m0, struct ether_header *);
858 
859 	cur_rx = &sc->re_ldata.re_rx_list[0];
860 	total_len = RE_RXBYTES(cur_rx);
861 	rxstat = le32toh(cur_rx->re_cmdstat);
862 
863 	if (total_len != ETHER_MIN_LEN) {
864 		if_printf(ifp, "diagnostic failed, received short packet\n");
865 		error = EIO;
866 		goto done;
867 	}
868 
869 	/* Test that the received packet data matches what we sent. */
870 
871 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
872 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
873 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
874 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
875 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
876 		    dst, ":", src, ":", ETHERTYPE_IP);
877 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
878 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
879 		    ntohs(eh->ether_type));
880 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
881 		    "into a 64-bit PCI slot.\n");
882 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
883 		    "for proper operation.\n");
884 		if_printf(ifp, "Read the re(4) man page for more details.\n");
885 		error = EIO;
886 	}
887 
888 done:
889 	/* Turn interface off, release resources */
890 
891 	sc->re_flags &= ~(RE_F_LINKED | RE_F_TESTMODE);
892 	ifp->if_flags &= ~IFF_PROMISC;
893 	re_stop(sc);
894 	if (m0 != NULL)
895 		m_freem(m0);
896 
897 	return (error);
898 }
899 #endif	/* RE_DIAG */
900 
901 /*
902  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
903  * IDs against our list and return a device name if we find a match.
904  */
905 static int
906 re_probe(device_t dev)
907 {
908 	const struct re_type *t;
909 	const struct re_hwrev *hw_rev;
910 	struct re_softc *sc;
911 	int rid;
912 	uint32_t hwrev, macmode, txcfg;
913 	uint16_t vendor, product;
914 
915 	vendor = pci_get_vendor(dev);
916 	product = pci_get_device(dev);
917 
918 	/*
919 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
920 	 * Rev.2 is supported by sk(4).
921 	 */
922 	if (vendor == PCI_VENDOR_LINKSYS &&
923 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
924 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
925 		return ENXIO;
926 
927 	for (t = re_devs; t->re_name != NULL; t++) {
928 		if (product == t->re_did && vendor == t->re_vid)
929 			break;
930 	}
931 
932 	/*
933 	 * Check if we found a RealTek device.
934 	 */
935 	if (t->re_name == NULL)
936 		return ENXIO;
937 
938 	/*
939 	 * Temporarily map the I/O space so we can read the chip ID register.
940 	 */
941 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
942 	rid = RE_PCI_LOIO;
943 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
944 					    RF_ACTIVE);
945 	if (sc->re_res == NULL) {
946 		device_printf(dev, "couldn't map ports/memory\n");
947 		kfree(sc, M_TEMP);
948 		return ENXIO;
949 	}
950 
951 	sc->re_btag = rman_get_bustag(sc->re_res);
952 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
953 
954 	txcfg = CSR_READ_4(sc, RE_TXCFG);
955 	hwrev = txcfg & RE_TXCFG_HWREV;
956 	macmode = txcfg & RE_TXCFG_MACMODE;
957 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
958 	kfree(sc, M_TEMP);
959 
960 	/*
961 	 * and continue matching for the specific chip...
962 	 */
963 	for (hw_rev = re_hwrevs; hw_rev->re_hwrev != RE_HWREV_NULL; hw_rev++) {
964 		if (hw_rev->re_hwrev == hwrev) {
965 			sc = device_get_softc(dev);
966 
967 			sc->re_hwrev = hw_rev->re_hwrev;
968 			sc->re_macver = hw_rev->re_macver;
969 			sc->re_caps = hw_rev->re_caps;
970 			sc->re_maxmtu = hw_rev->re_maxmtu;
971 			sc->re_swcsum_lim = RE_SWCSUM_UNLIMITED;
972 
973 			/*
974 			 * Apply chip property fixup
975 			 */
976 			switch (sc->re_hwrev) {
977 			case RE_HWREV_8169:
978 				sc->re_swcsum_lim = RE_SWCSUM_LIM_8169;
979 				break;
980 			case RE_HWREV_8101E1:
981 			case RE_HWREV_8101E2:
982 				if (macmode == 0)
983 					sc->re_macver = RE_MACVER_11;
984 				else if (macmode == 0x200000)
985 					sc->re_macver = RE_MACVER_12;
986 				break;
987 			case RE_HWREV_8102E:
988 			case RE_HWREV_8102EL:
989 				if (macmode == 0)
990 					sc->re_macver = RE_MACVER_13;
991 				else if (macmode == 0x100000)
992 					sc->re_macver = RE_MACVER_14;
993 				break;
994 			case RE_HWREV_8168B2:
995 			case RE_HWREV_8168B3:
996 				if (macmode == 0)
997 					sc->re_macver = RE_MACVER_22;
998 				break;
999 			case RE_HWREV_8168C:
1000 				if (macmode == 0)
1001 					sc->re_macver = RE_MACVER_24;
1002 				else if (macmode == 0x200000)
1003 					sc->re_macver = RE_MACVER_25;
1004 				else if (macmode == 0x300000)
1005 					sc->re_macver = RE_MACVER_27;
1006 				break;
1007 			case RE_HWREV_8168CP:
1008 				if (macmode == 0)
1009 					sc->re_macver = RE_MACVER_26;
1010 				else if (macmode == 0x100000)
1011 					sc->re_macver = RE_MACVER_28;
1012 				break;
1013 			}
1014 			if (pci_get_pciecap_ptr(dev) != 0)
1015 				sc->re_caps |= RE_C_PCIE;
1016 
1017 			device_set_desc(dev, t->re_name);
1018 			return 0;
1019 		}
1020 	}
1021 	device_printf(dev, "unknown hwrev 0x%08x, macmode 0x%08x\n",
1022 		      hwrev, macmode);
1023 	return ENXIO;
1024 }
1025 
1026 static void
1027 re_dma_map_desc(void *xarg, bus_dma_segment_t *segs, int nsegs,
1028 		bus_size_t mapsize, int error)
1029 {
1030 	struct re_dmaload_arg *arg = xarg;
1031 	int i;
1032 
1033 	if (error)
1034 		return;
1035 
1036 	if (nsegs > arg->re_nsegs) {
1037 		arg->re_nsegs = 0;
1038 		return;
1039 	}
1040 
1041 	arg->re_nsegs = nsegs;
1042 	for (i = 0; i < nsegs; ++i)
1043 		arg->re_segs[i] = segs[i];
1044 }
1045 
1046 /*
1047  * Map a single buffer address.
1048  */
1049 
1050 static void
1051 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1052 {
1053 	uint32_t *addr;
1054 
1055 	if (error)
1056 		return;
1057 
1058 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1059 	addr = arg;
1060 	*addr = segs->ds_addr;
1061 }
1062 
1063 static int
1064 re_allocmem(device_t dev)
1065 {
1066 	struct re_softc *sc = device_get_softc(dev);
1067 	int error, i;
1068 
1069 	/*
1070 	 * Allocate list data
1071 	 */
1072 	sc->re_ldata.re_tx_mbuf =
1073 	kmalloc(sc->re_tx_desc_cnt * sizeof(struct mbuf *),
1074 		M_DEVBUF, M_ZERO | M_WAITOK);
1075 
1076 	sc->re_ldata.re_rx_mbuf =
1077 	kmalloc(sc->re_rx_desc_cnt * sizeof(struct mbuf *),
1078 		M_DEVBUF, M_ZERO | M_WAITOK);
1079 
1080 	sc->re_ldata.re_rx_paddr =
1081 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_addr_t),
1082 		M_DEVBUF, M_ZERO | M_WAITOK);
1083 
1084 	sc->re_ldata.re_tx_dmamap =
1085 	kmalloc(sc->re_tx_desc_cnt * sizeof(bus_dmamap_t),
1086 		M_DEVBUF, M_ZERO | M_WAITOK);
1087 
1088 	sc->re_ldata.re_rx_dmamap =
1089 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_dmamap_t),
1090 		M_DEVBUF, M_ZERO | M_WAITOK);
1091 
1092 	/*
1093 	 * Allocate the parent bus DMA tag appropriate for PCI.
1094 	 */
1095 	error = bus_dma_tag_create(NULL,	/* parent */
1096 			1, 0,			/* alignment, boundary */
1097 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1098 			BUS_SPACE_MAXADDR,	/* highaddr */
1099 			NULL, NULL,		/* filter, filterarg */
1100 			MAXBSIZE, RE_MAXSEGS,	/* maxsize, nsegments */
1101 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1102 			BUS_DMA_ALLOCNOW,	/* flags */
1103 			&sc->re_parent_tag);
1104 	if (error) {
1105 		device_printf(dev, "could not allocate parent dma tag\n");
1106 		return error;
1107 	}
1108 
1109 	/* Allocate tag for TX descriptor list. */
1110 	error = bus_dma_tag_create(sc->re_parent_tag,
1111 			RE_RING_ALIGN, 0,
1112 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1113 			NULL, NULL,
1114 			RE_TX_LIST_SZ(sc), 1, RE_TX_LIST_SZ(sc),
1115 			BUS_DMA_ALLOCNOW,
1116 			&sc->re_ldata.re_tx_list_tag);
1117 	if (error) {
1118 		device_printf(dev, "could not allocate TX ring dma tag\n");
1119 		return(error);
1120 	}
1121 
1122 	/* Allocate DMA'able memory for the TX ring */
1123         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
1124 			(void **)&sc->re_ldata.re_tx_list,
1125 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1126 			&sc->re_ldata.re_tx_list_map);
1127         if (error) {
1128 		device_printf(dev, "could not allocate TX ring\n");
1129 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1130 		sc->re_ldata.re_tx_list_tag = NULL;
1131                 return(error);
1132 	}
1133 
1134 	/* Load the map for the TX ring. */
1135 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
1136 			sc->re_ldata.re_tx_list_map,
1137 			sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc),
1138 			re_dma_map_addr, &sc->re_ldata.re_tx_list_addr,
1139 			BUS_DMA_NOWAIT);
1140 	if (error) {
1141 		device_printf(dev, "could not get address of TX ring\n");
1142 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1143 				sc->re_ldata.re_tx_list,
1144 				sc->re_ldata.re_tx_list_map);
1145 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1146 		sc->re_ldata.re_tx_list_tag = NULL;
1147 		return(error);
1148 	}
1149 
1150 	/* Allocate tag for RX descriptor list. */
1151 	error = bus_dma_tag_create(sc->re_parent_tag,
1152 			RE_RING_ALIGN, 0,
1153 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1154 			NULL, NULL,
1155 			RE_RX_LIST_SZ(sc), 1, RE_RX_LIST_SZ(sc),
1156 			BUS_DMA_ALLOCNOW,
1157 			&sc->re_ldata.re_rx_list_tag);
1158 	if (error) {
1159 		device_printf(dev, "could not allocate RX ring dma tag\n");
1160 		return(error);
1161 	}
1162 
1163 	/* Allocate DMA'able memory for the RX ring */
1164         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1165 			(void **)&sc->re_ldata.re_rx_list,
1166 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1167 			&sc->re_ldata.re_rx_list_map);
1168         if (error) {
1169 		device_printf(dev, "could not allocate RX ring\n");
1170 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1171 		sc->re_ldata.re_rx_list_tag = NULL;
1172                 return(error);
1173 	}
1174 
1175 	/* Load the map for the RX ring. */
1176 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1177 			sc->re_ldata.re_rx_list_map,
1178 			sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc),
1179 			re_dma_map_addr, &sc->re_ldata.re_rx_list_addr,
1180 			BUS_DMA_NOWAIT);
1181 	if (error) {
1182 		device_printf(dev, "could not get address of RX ring\n");
1183 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1184 				sc->re_ldata.re_rx_list,
1185 				sc->re_ldata.re_rx_list_map);
1186 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1187 		sc->re_ldata.re_rx_list_tag = NULL;
1188 		return(error);
1189 	}
1190 
1191 	/* Allocate map for RX/TX mbufs. */
1192 	error = bus_dma_tag_create(sc->re_parent_tag,
1193 			ETHER_ALIGN, 0,
1194 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1195 			NULL, NULL,
1196 			RE_FRAMELEN_MAX, RE_MAXSEGS, MCLBYTES,
1197 			BUS_DMA_ALLOCNOW,
1198 			&sc->re_ldata.re_mtag);
1199 	if (error) {
1200 		device_printf(dev, "could not allocate buf dma tag\n");
1201 		return(error);
1202 	}
1203 
1204 	/* Create spare DMA map for RX */
1205 	error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1206 			&sc->re_ldata.re_rx_spare);
1207 	if (error) {
1208 		device_printf(dev, "can't create spare DMA map for RX\n");
1209 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1210 		sc->re_ldata.re_mtag = NULL;
1211 		return error;
1212 	}
1213 
1214 	/* Create DMA maps for TX buffers */
1215 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
1216 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1217 				&sc->re_ldata.re_tx_dmamap[i]);
1218 		if (error) {
1219 			device_printf(dev, "can't create DMA map for TX buf\n");
1220 			re_freebufmem(sc, i, 0);
1221 			return(error);
1222 		}
1223 	}
1224 
1225 	/* Create DMA maps for RX buffers */
1226 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1227 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1228 				&sc->re_ldata.re_rx_dmamap[i]);
1229 		if (error) {
1230 			device_printf(dev, "can't create DMA map for RX buf\n");
1231 			re_freebufmem(sc, sc->re_tx_desc_cnt, i);
1232 			return(error);
1233 		}
1234 	}
1235 
1236 	/* Create jumbo buffer pool for RX if required */
1237 	if (sc->re_caps & RE_C_CONTIGRX) {
1238 		error = re_jpool_alloc(sc);
1239 		if (error) {
1240 			re_jpool_free(sc);
1241 			/* Disable jumbo frame support */
1242 			sc->re_maxmtu = ETHERMTU;
1243 		}
1244 	}
1245 	return(0);
1246 }
1247 
1248 static void
1249 re_freebufmem(struct re_softc *sc, int tx_cnt, int rx_cnt)
1250 {
1251 	int i;
1252 
1253 	/* Destroy all the RX and TX buffer maps */
1254 	if (sc->re_ldata.re_mtag) {
1255 		for (i = 0; i < tx_cnt; i++) {
1256 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1257 					   sc->re_ldata.re_tx_dmamap[i]);
1258 		}
1259 		for (i = 0; i < rx_cnt; i++) {
1260 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1261 					   sc->re_ldata.re_rx_dmamap[i]);
1262 		}
1263 		bus_dmamap_destroy(sc->re_ldata.re_mtag,
1264 				   sc->re_ldata.re_rx_spare);
1265 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1266 		sc->re_ldata.re_mtag = NULL;
1267 	}
1268 }
1269 
1270 static void
1271 re_freemem(device_t dev)
1272 {
1273 	struct re_softc *sc = device_get_softc(dev);
1274 
1275 	/* Unload and free the RX DMA ring memory and map */
1276 	if (sc->re_ldata.re_rx_list_tag) {
1277 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1278 				  sc->re_ldata.re_rx_list_map);
1279 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1280 				sc->re_ldata.re_rx_list,
1281 				sc->re_ldata.re_rx_list_map);
1282 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1283 	}
1284 
1285 	/* Unload and free the TX DMA ring memory and map */
1286 	if (sc->re_ldata.re_tx_list_tag) {
1287 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1288 				  sc->re_ldata.re_tx_list_map);
1289 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1290 				sc->re_ldata.re_tx_list,
1291 				sc->re_ldata.re_tx_list_map);
1292 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1293 	}
1294 
1295 	/* Free RX/TX buf DMA stuffs */
1296 	re_freebufmem(sc, sc->re_tx_desc_cnt, sc->re_rx_desc_cnt);
1297 
1298 	/* Unload and free the stats buffer and map */
1299 	if (sc->re_ldata.re_stag) {
1300 		bus_dmamap_unload(sc->re_ldata.re_stag,
1301 				  sc->re_ldata.re_rx_list_map);
1302 		bus_dmamem_free(sc->re_ldata.re_stag,
1303 				sc->re_ldata.re_stats,
1304 				sc->re_ldata.re_smap);
1305 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1306 	}
1307 
1308 	if (sc->re_caps & RE_C_CONTIGRX)
1309 		re_jpool_free(sc);
1310 
1311 	if (sc->re_parent_tag)
1312 		bus_dma_tag_destroy(sc->re_parent_tag);
1313 
1314 	if (sc->re_ldata.re_tx_mbuf != NULL)
1315 		kfree(sc->re_ldata.re_tx_mbuf, M_DEVBUF);
1316 	if (sc->re_ldata.re_rx_mbuf != NULL)
1317 		kfree(sc->re_ldata.re_rx_mbuf, M_DEVBUF);
1318 	if (sc->re_ldata.re_rx_paddr != NULL)
1319 		kfree(sc->re_ldata.re_rx_paddr, M_DEVBUF);
1320 	if (sc->re_ldata.re_tx_dmamap != NULL)
1321 		kfree(sc->re_ldata.re_tx_dmamap, M_DEVBUF);
1322 	if (sc->re_ldata.re_rx_dmamap != NULL)
1323 		kfree(sc->re_ldata.re_rx_dmamap, M_DEVBUF);
1324 }
1325 
1326 /*
1327  * Attach the interface. Allocate softc structures, do ifmedia
1328  * setup and ethernet/BPF attach.
1329  */
1330 static int
1331 re_attach(device_t dev)
1332 {
1333 	struct re_softc	*sc = device_get_softc(dev);
1334 	struct ifnet *ifp;
1335 	uint8_t eaddr[ETHER_ADDR_LEN];
1336 	int error = 0, rid, qlen;
1337 
1338 	callout_init(&sc->re_timer);
1339 	sc->re_dev = dev;
1340 
1341 	if (RE_IS_8139CP(sc)) {
1342 		sc->re_rx_desc_cnt = RE_RX_DESC_CNT_8139CP;
1343 		sc->re_tx_desc_cnt = RE_TX_DESC_CNT_8139CP;
1344 	} else {
1345 		sc->re_rx_desc_cnt = re_rx_desc_count;
1346 		if (sc->re_rx_desc_cnt > RE_RX_DESC_CNT_MAX)
1347 			sc->re_rx_desc_cnt = RE_RX_DESC_CNT_MAX;
1348 
1349 		sc->re_tx_desc_cnt = re_tx_desc_count;
1350 		if (sc->re_tx_desc_cnt > RE_TX_DESC_CNT_MAX)
1351 			sc->re_tx_desc_cnt = RE_TX_DESC_CNT_MAX;
1352 	}
1353 
1354 	qlen = RE_IFQ_MAXLEN;
1355 	if (sc->re_tx_desc_cnt > qlen)
1356 		qlen = sc->re_tx_desc_cnt;
1357 
1358 	sc->re_rxbuf_size = MCLBYTES;
1359 	sc->re_newbuf = re_newbuf_std;
1360 
1361 	sc->re_tx_time = 5;		/* 125us */
1362 	sc->re_rx_time = 2;		/* 50us */
1363 	if (sc->re_caps & RE_C_PCIE)
1364 		sc->re_sim_time = 75;	/* 75us */
1365 	else
1366 		sc->re_sim_time = 125;	/* 125us */
1367 	sc->re_imtype = RE_IMTYPE_SIM;	/* simulated interrupt moderation */
1368 	re_config_imtype(sc, sc->re_imtype);
1369 
1370 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1371 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1372 					     SYSCTL_STATIC_CHILDREN(_hw),
1373 					     OID_AUTO,
1374 					     device_get_nameunit(dev),
1375 					     CTLFLAG_RD, 0, "");
1376 	if (sc->re_sysctl_tree == NULL) {
1377 		device_printf(dev, "can't add sysctl node\n");
1378 		error = ENXIO;
1379 		goto fail;
1380 	}
1381 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1382 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1383 		       "rx_desc_count", CTLFLAG_RD, &sc->re_rx_desc_cnt,
1384 		       0, "RX desc count");
1385 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1386 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1387 		       "tx_desc_count", CTLFLAG_RD, &sc->re_tx_desc_cnt,
1388 		       0, "TX desc count");
1389 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1390 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1391 			OID_AUTO, "sim_time",
1392 			CTLTYPE_INT | CTLFLAG_RW,
1393 			sc, 0, re_sysctl_simtime, "I",
1394 			"Simulated interrupt moderation time (usec).");
1395 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1396 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1397 			OID_AUTO, "imtype",
1398 			CTLTYPE_INT | CTLFLAG_RW,
1399 			sc, 0, re_sysctl_imtype, "I",
1400 			"Interrupt moderation type -- "
1401 			"0:disable, 1:simulated, "
1402 			"2:hardware(if supported)");
1403 	if (sc->re_caps & RE_C_HWIM) {
1404 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1405 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1406 				OID_AUTO, "hw_rxtime",
1407 				CTLTYPE_INT | CTLFLAG_RW,
1408 				sc, 0, re_sysctl_rxtime, "I",
1409 				"Hardware interrupt moderation time "
1410 				"(unit: 25usec).");
1411 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1412 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1413 				OID_AUTO, "hw_txtime",
1414 				CTLTYPE_INT | CTLFLAG_RW,
1415 				sc, 0, re_sysctl_txtime, "I",
1416 				"Hardware interrupt moderation time "
1417 				"(unit: 25usec).");
1418 	}
1419 
1420 #ifndef BURN_BRIDGES
1421 	/*
1422 	 * Handle power management nonsense.
1423 	 */
1424 
1425 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1426 		uint32_t membase, irq;
1427 
1428 		/* Save important PCI config data. */
1429 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1430 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1431 
1432 		/* Reset the power state. */
1433 		device_printf(dev, "chip is in D%d power mode "
1434 		    "-- setting to D0\n", pci_get_powerstate(dev));
1435 
1436 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1437 
1438 		/* Restore PCI config data. */
1439 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1440 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1441 	}
1442 #endif
1443 	/*
1444 	 * Map control/status registers.
1445 	 */
1446 	pci_enable_busmaster(dev);
1447 
1448 	rid = RE_PCI_LOIO;
1449 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1450 					    RF_ACTIVE);
1451 
1452 	if (sc->re_res == NULL) {
1453 		device_printf(dev, "couldn't map ports\n");
1454 		error = ENXIO;
1455 		goto fail;
1456 	}
1457 
1458 	sc->re_btag = rman_get_bustag(sc->re_res);
1459 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1460 
1461 	/* Allocate interrupt */
1462 	rid = 0;
1463 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1464 					    RF_SHAREABLE | RF_ACTIVE);
1465 
1466 	if (sc->re_irq == NULL) {
1467 		device_printf(dev, "couldn't map interrupt\n");
1468 		error = ENXIO;
1469 		goto fail;
1470 	}
1471 
1472 	/* Reset the adapter. */
1473 	re_reset(sc, 0);
1474 
1475 	if (RE_IS_8139CP(sc)) {
1476 		sc->re_bus_speed = 33; /* XXX */
1477 	} else if (sc->re_caps & RE_C_PCIE) {
1478 		sc->re_bus_speed = 125;
1479 	} else {
1480 		uint8_t cfg2;
1481 
1482 		cfg2 = CSR_READ_1(sc, RE_CFG2);
1483 		switch (cfg2 & RE_CFG2_PCICLK_MASK) {
1484 		case RE_CFG2_PCICLK_33MHZ:
1485 			sc->re_bus_speed = 33;
1486 			break;
1487 		case RE_CFG2_PCICLK_66MHZ:
1488 			sc->re_bus_speed = 66;
1489 			break;
1490 		default:
1491 			device_printf(dev, "unknown bus speed, assume 33MHz\n");
1492 			sc->re_bus_speed = 33;
1493 			break;
1494 		}
1495 		if (cfg2 & RE_CFG2_PCI64)
1496 			sc->re_caps |= RE_C_PCI64;
1497 	}
1498 	device_printf(dev, "Hardware rev. 0x%08x; MAC ver. 0x%02x; "
1499 		      "PCI%s %dMHz\n",
1500 		      sc->re_hwrev, sc->re_macver,
1501 		      (sc->re_caps & RE_C_PCIE) ?
1502 		      "-E" : ((sc->re_caps & RE_C_PCI64) ? "64" : "32"),
1503 		      sc->re_bus_speed);
1504 
1505 	/*
1506 	 * NOTE:
1507 	 * DO NOT try to adjust config1 and config5 which was spotted in
1508 	 * Realtek's Linux drivers.  It will _permanently_ damage certain
1509 	 * cards EEPROM, e.g. one of my 8168B (0x38000000) card ...
1510 	 */
1511 
1512 	re_get_eaddr(sc, eaddr);
1513 
1514 	if (!RE_IS_8139CP(sc)) {
1515 		/* Set RX length mask */
1516 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1517 		sc->re_txstart = RE_GTXSTART;
1518 	} else {
1519 		/* Set RX length mask */
1520 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1521 		sc->re_txstart = RE_TXSTART;
1522 	}
1523 
1524 	/* Allocate DMA stuffs */
1525 	error = re_allocmem(dev);
1526 	if (error)
1527 		goto fail;
1528 
1529 	/*
1530 	 * Apply some magic PCI settings from Realtek ...
1531 	 */
1532 	if (RE_IS_8169(sc)) {
1533 		CSR_WRITE_1(sc, 0x82, 1);
1534 		pci_write_config(dev, PCIR_CACHELNSZ, 0x8, 1);
1535 	}
1536 	pci_write_config(dev, PCIR_LATTIMER, 0x40, 1);
1537 
1538 	if (sc->re_caps & RE_C_MAC2) {
1539 		/*
1540 		 * Following part is extracted from Realtek BSD driver v176.
1541 		 * However, this does _not_ make much/any sense:
1542 		 * 8168C's PCI Express device control is located at 0x78,
1543 		 * so the reading from 0x79 (higher part of 0x78) and setting
1544 		 * the 4~6bits intend to enlarge the "max read request size"
1545 		 * (we will do it).  The content of the rest part of this
1546 		 * register is not meaningful to other PCI registers, so
1547 		 * writing the value to 0x54 could be completely wrong.
1548 		 * 0x80 is the lower part of PCI Express device status, non-
1549 		 * reserved bits are RW1C, writing 0 to them will not have
1550 		 * any effect at all.
1551 		 */
1552 #ifdef foo
1553 		uint8_t val;
1554 
1555 		val = pci_read_config(dev, 0x79, 1);
1556 		val = (val & ~0x70) | 0x50;
1557 		pci_write_config(dev, 0x54, val, 1);
1558 		pci_write_config(dev, 0x80, 0, 1);
1559 #endif
1560 	}
1561 
1562 	/*
1563 	 * Apply some PHY fixup from Realtek ...
1564 	 */
1565 	if (sc->re_hwrev == RE_HWREV_8110S) {
1566 		CSR_WRITE_1(sc, 0x82, 1);
1567 		re_miibus_writereg(dev, 1, 0xb, 0);
1568 	}
1569 	if (sc->re_caps & RE_C_PHYPMGT) {
1570 		/* Power up PHY */
1571 		re_miibus_writereg(dev, 1, 0x1f, 0);
1572 		re_miibus_writereg(dev, 1, 0xe, 0);
1573 	}
1574 
1575 	/* Do MII setup */
1576 	if (mii_phy_probe(dev, &sc->re_miibus,
1577 	    re_ifmedia_upd, re_ifmedia_sts)) {
1578 		device_printf(dev, "MII without any phy!\n");
1579 		error = ENXIO;
1580 		goto fail;
1581 	}
1582 
1583 	ifp = &sc->arpcom.ac_if;
1584 	ifp->if_softc = sc;
1585 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1586 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1587 	ifp->if_ioctl = re_ioctl;
1588 	ifp->if_start = re_start;
1589 #ifdef DEVICE_POLLING
1590 	ifp->if_poll = re_poll;
1591 #endif
1592 	ifp->if_watchdog = re_watchdog;
1593 	ifp->if_init = re_init;
1594 	if (!RE_IS_8139CP(sc)) /* XXX */
1595 		ifp->if_baudrate = 1000000000;
1596 	else
1597 		ifp->if_baudrate = 100000000;
1598 	ifq_set_maxlen(&ifp->if_snd, qlen);
1599 	ifq_set_ready(&ifp->if_snd);
1600 
1601 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1602 	if (sc->re_caps & RE_C_HWCSUM)
1603 		ifp->if_capabilities |= IFCAP_HWCSUM;
1604 
1605 	ifp->if_capenable = ifp->if_capabilities;
1606 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1607 		ifp->if_hwassist = RE_CSUM_FEATURES;
1608 	else
1609 		ifp->if_hwassist = 0;
1610 
1611 	/*
1612 	 * Call MI attach routine.
1613 	 */
1614 	ether_ifattach(ifp, eaddr, NULL);
1615 
1616 #ifdef RE_DIAG
1617 	/*
1618 	 * Perform hardware diagnostic on the original RTL8169.
1619 	 * Some 32-bit cards were incorrectly wired and would
1620 	 * malfunction if plugged into a 64-bit slot.
1621 	 */
1622 	if (sc->re_hwrev == RE_HWREV_8169) {
1623 		lwkt_serialize_enter(ifp->if_serializer);
1624 		error = re_diag(sc);
1625 		lwkt_serialize_exit(ifp->if_serializer);
1626 
1627 		if (error) {
1628 			device_printf(dev, "hardware diagnostic failure\n");
1629 			ether_ifdetach(ifp);
1630 			goto fail;
1631 		}
1632 	}
1633 #endif	/* RE_DIAG */
1634 
1635 	/* Hook interrupt last to avoid having to lock softc */
1636 	error = bus_setup_intr(dev, sc->re_irq, INTR_MPSAFE, re_intr, sc,
1637 			       &sc->re_intrhand, ifp->if_serializer);
1638 
1639 	if (error) {
1640 		device_printf(dev, "couldn't set up irq\n");
1641 		ether_ifdetach(ifp);
1642 		goto fail;
1643 	}
1644 
1645 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1646 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1647 
1648 fail:
1649 	if (error)
1650 		re_detach(dev);
1651 
1652 	return (error);
1653 }
1654 
1655 /*
1656  * Shutdown hardware and free up resources. This can be called any
1657  * time after the mutex has been initialized. It is called in both
1658  * the error case in attach and the normal detach case so it needs
1659  * to be careful about only freeing resources that have actually been
1660  * allocated.
1661  */
1662 static int
1663 re_detach(device_t dev)
1664 {
1665 	struct re_softc *sc = device_get_softc(dev);
1666 	struct ifnet *ifp = &sc->arpcom.ac_if;
1667 
1668 	/* These should only be active if attach succeeded */
1669 	if (device_is_attached(dev)) {
1670 		lwkt_serialize_enter(ifp->if_serializer);
1671 		re_stop(sc);
1672 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1673 		lwkt_serialize_exit(ifp->if_serializer);
1674 
1675 		ether_ifdetach(ifp);
1676 	}
1677 	if (sc->re_miibus)
1678 		device_delete_child(dev, sc->re_miibus);
1679 	bus_generic_detach(dev);
1680 
1681 	if (sc->re_sysctl_tree != NULL)
1682 		sysctl_ctx_free(&sc->re_sysctl_ctx);
1683 
1684 	if (sc->re_irq)
1685 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1686 	if (sc->re_res) {
1687 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1688 				     sc->re_res);
1689 	}
1690 
1691 	/* Free DMA stuffs */
1692 	re_freemem(dev);
1693 
1694 	return(0);
1695 }
1696 
1697 static void
1698 re_setup_rxdesc(struct re_softc *sc, int idx)
1699 {
1700 	bus_addr_t paddr;
1701 	uint32_t cmdstat;
1702 	struct re_desc *d;
1703 
1704 	paddr = sc->re_ldata.re_rx_paddr[idx];
1705 	d = &sc->re_ldata.re_rx_list[idx];
1706 
1707 	d->re_bufaddr_lo = htole32(RE_ADDR_LO(paddr));
1708 	d->re_bufaddr_hi = htole32(RE_ADDR_HI(paddr));
1709 
1710 	cmdstat = sc->re_rxbuf_size | RE_RDESC_CMD_OWN;
1711 	if (idx == (sc->re_rx_desc_cnt - 1))
1712 		cmdstat |= RE_RDESC_CMD_EOR;
1713 	d->re_cmdstat = htole32(cmdstat);
1714 }
1715 
1716 static int
1717 re_newbuf_std(struct re_softc *sc, int idx, int init)
1718 {
1719 	struct re_dmaload_arg arg;
1720 	bus_dma_segment_t seg;
1721 	bus_dmamap_t map;
1722 	struct mbuf *m;
1723 	int error;
1724 
1725 	m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
1726 	if (m == NULL) {
1727 		error = ENOBUFS;
1728 
1729 		if (init) {
1730 			if_printf(&sc->arpcom.ac_if, "m_getcl failed\n");
1731 			return error;
1732 		} else {
1733 			goto back;
1734 		}
1735 	}
1736 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1737 
1738 	/*
1739 	 * NOTE:
1740 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1741 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1742 	 */
1743 
1744 	arg.re_nsegs = 1;
1745 	arg.re_segs = &seg;
1746         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1747 				     sc->re_ldata.re_rx_spare, m,
1748 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1749 	if (error || arg.re_nsegs == 0) {
1750 		if (!error) {
1751 			if_printf(&sc->arpcom.ac_if, "too many segments?!\n");
1752 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1753 					  sc->re_ldata.re_rx_spare);
1754 			error = EFBIG;
1755 		}
1756 		m_freem(m);
1757 
1758 		if (init) {
1759 			if_printf(&sc->arpcom.ac_if, "can't load RX mbuf\n");
1760 			return error;
1761 		} else {
1762 			goto back;
1763 		}
1764 	}
1765 
1766 	if (!init) {
1767 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1768 				sc->re_ldata.re_rx_dmamap[idx],
1769 				BUS_DMASYNC_POSTREAD);
1770 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1771 				  sc->re_ldata.re_rx_dmamap[idx]);
1772 	}
1773 	sc->re_ldata.re_rx_mbuf[idx] = m;
1774 	sc->re_ldata.re_rx_paddr[idx] = seg.ds_addr;
1775 
1776 	map = sc->re_ldata.re_rx_dmamap[idx];
1777 	sc->re_ldata.re_rx_dmamap[idx] = sc->re_ldata.re_rx_spare;
1778 	sc->re_ldata.re_rx_spare = map;
1779 back:
1780 	re_setup_rxdesc(sc, idx);
1781 	return error;
1782 }
1783 
1784 static int
1785 re_newbuf_jumbo(struct re_softc *sc, int idx, int init)
1786 {
1787 	struct mbuf *m;
1788 	struct re_jbuf *jbuf;
1789 	int error = 0;
1790 
1791 	MGETHDR(m, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1792 	if (m == NULL) {
1793 		error = ENOBUFS;
1794 		if (init) {
1795 			if_printf(&sc->arpcom.ac_if, "MGETHDR failed\n");
1796 			return error;
1797 		} else {
1798 			goto back;
1799 		}
1800 	}
1801 
1802 	jbuf = re_jbuf_alloc(sc);
1803 	if (jbuf == NULL) {
1804 		m_freem(m);
1805 
1806 		error = ENOBUFS;
1807 		if (init) {
1808 			if_printf(&sc->arpcom.ac_if, "jpool is empty\n");
1809 			return error;
1810 		} else {
1811 			goto back;
1812 		}
1813 	}
1814 
1815 	m->m_ext.ext_arg = jbuf;
1816 	m->m_ext.ext_buf = jbuf->re_buf;
1817 	m->m_ext.ext_free = re_jbuf_free;
1818 	m->m_ext.ext_ref = re_jbuf_ref;
1819 	m->m_ext.ext_size = sc->re_rxbuf_size;
1820 
1821 	m->m_data = m->m_ext.ext_buf;
1822 	m->m_flags |= M_EXT;
1823 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
1824 
1825 	/*
1826 	 * NOTE:
1827 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1828 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1829 	 */
1830 
1831 	sc->re_ldata.re_rx_mbuf[idx] = m;
1832 	sc->re_ldata.re_rx_paddr[idx] = jbuf->re_paddr;
1833 back:
1834 	re_setup_rxdesc(sc, idx);
1835 	return error;
1836 }
1837 
1838 static int
1839 re_tx_list_init(struct re_softc *sc)
1840 {
1841 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
1842 
1843 	/* Flush the TX descriptors */
1844 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1845 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1846 
1847 	sc->re_ldata.re_tx_prodidx = 0;
1848 	sc->re_ldata.re_tx_considx = 0;
1849 	sc->re_ldata.re_tx_free = sc->re_tx_desc_cnt;
1850 
1851 	return(0);
1852 }
1853 
1854 static int
1855 re_rx_list_init(struct re_softc *sc)
1856 {
1857 	int i, error;
1858 
1859 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc));
1860 
1861 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1862 		error = sc->re_newbuf(sc, i, 1);
1863 		if (error)
1864 			return(error);
1865 	}
1866 
1867 	/* Flush the RX descriptors */
1868 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1869 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1870 
1871 	sc->re_ldata.re_rx_prodidx = 0;
1872 	sc->re_head = sc->re_tail = NULL;
1873 
1874 	return(0);
1875 }
1876 
1877 #define RE_IP4_PACKET	0x1
1878 #define RE_TCP_PACKET	0x2
1879 #define RE_UDP_PACKET	0x4
1880 
1881 static __inline uint8_t
1882 re_packet_type(struct re_softc *sc, uint32_t rxstat, uint32_t rxctrl)
1883 {
1884 	uint8_t packet_type = 0;
1885 
1886 	if (sc->re_caps & RE_C_MAC2) {
1887 		if (rxctrl & RE_RDESC_CTL_PROTOIP4)
1888 			packet_type |= RE_IP4_PACKET;
1889 	} else {
1890 		if (rxstat & RE_RDESC_STAT_PROTOID)
1891 			packet_type |= RE_IP4_PACKET;
1892 	}
1893 	if (RE_TCPPKT(rxstat))
1894 		packet_type |= RE_TCP_PACKET;
1895 	else if (RE_UDPPKT(rxstat))
1896 		packet_type |= RE_UDP_PACKET;
1897 	return packet_type;
1898 }
1899 
1900 /*
1901  * RX handler for C+ and 8169. For the gigE chips, we support
1902  * the reception of jumbo frames that have been fragmented
1903  * across multiple 2K mbuf cluster buffers.
1904  */
1905 static int
1906 re_rxeof(struct re_softc *sc)
1907 {
1908 	struct ifnet *ifp = &sc->arpcom.ac_if;
1909 	struct mbuf *m;
1910 	struct re_desc 	*cur_rx;
1911 	uint32_t rxstat, rxctrl;
1912 	int i, total_len, rx = 0;
1913 	struct mbuf_chain chain[MAXCPU];
1914 
1915 	/* Invalidate the descriptor memory */
1916 
1917 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1918 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1919 
1920 	ether_input_chain_init(chain);
1921 
1922 	for (i = sc->re_ldata.re_rx_prodidx;
1923 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0; RE_RXDESC_INC(sc, i)) {
1924 		cur_rx = &sc->re_ldata.re_rx_list[i];
1925 		m = sc->re_ldata.re_rx_mbuf[i];
1926 		total_len = RE_RXBYTES(cur_rx);
1927 		rxstat = le32toh(cur_rx->re_cmdstat);
1928 		rxctrl = le32toh(cur_rx->re_control);
1929 
1930 		rx = 1;
1931 
1932 #ifdef INVARIANTS
1933 		if (sc->re_flags & RE_F_USE_JPOOL)
1934 			KKASSERT(rxstat & RE_RDESC_STAT_EOF);
1935 #endif
1936 
1937 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1938 			if (sc->re_flags & RE_F_DROP_RXFRAG) {
1939 				re_setup_rxdesc(sc, i);
1940 				continue;
1941 			}
1942 
1943 			if (sc->re_newbuf(sc, i, 0)) {
1944 				/* Drop upcoming fragments */
1945 				sc->re_flags |= RE_F_DROP_RXFRAG;
1946 				continue;
1947 			}
1948 
1949 			m->m_len = MCLBYTES;
1950 			if (sc->re_head == NULL) {
1951 				sc->re_head = sc->re_tail = m;
1952 			} else {
1953 				sc->re_tail->m_next = m;
1954 				sc->re_tail = m;
1955 			}
1956 			continue;
1957 		} else if (sc->re_flags & RE_F_DROP_RXFRAG) {
1958 			/*
1959 			 * Last fragment of a multi-fragment packet.
1960 			 *
1961 			 * Since error already happened, this fragment
1962 			 * must be dropped as well as the fragment chain.
1963 			 */
1964 			re_setup_rxdesc(sc, i);
1965 			re_free_rxchain(sc);
1966 			sc->re_flags &= ~RE_F_DROP_RXFRAG;
1967 			continue;
1968 		}
1969 
1970 		/*
1971 		 * NOTE: for the 8139C+, the frame length field
1972 		 * is always 12 bits in size, but for the gigE chips,
1973 		 * it is 13 bits (since the max RX frame length is 16K).
1974 		 * Unfortunately, all 32 bits in the status word
1975 		 * were already used, so to make room for the extra
1976 		 * length bit, RealTek took out the 'frame alignment
1977 		 * error' bit and shifted the other status bits
1978 		 * over one slot. The OWN, EOR, FS and LS bits are
1979 		 * still in the same places. We have already extracted
1980 		 * the frame length and checked the OWN bit, so rather
1981 		 * than using an alternate bit mapping, we shift the
1982 		 * status bits one space to the right so we can evaluate
1983 		 * them using the 8169 status as though it was in the
1984 		 * same format as that of the 8139C+.
1985 		 */
1986 		if (!RE_IS_8139CP(sc))
1987 			rxstat >>= 1;
1988 
1989 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1990 			ifp->if_ierrors++;
1991 			/*
1992 			 * If this is part of a multi-fragment packet,
1993 			 * discard all the pieces.
1994 			 */
1995 			re_free_rxchain(sc);
1996 			re_setup_rxdesc(sc, i);
1997 			continue;
1998 		}
1999 
2000 		/*
2001 		 * If allocating a replacement mbuf fails,
2002 		 * reload the current one.
2003 		 */
2004 
2005 		if (sc->re_newbuf(sc, i, 0)) {
2006 			ifp->if_ierrors++;
2007 			continue;
2008 		}
2009 
2010 		if (sc->re_head != NULL) {
2011 			m->m_len = total_len % MCLBYTES;
2012 			/*
2013 			 * Special case: if there's 4 bytes or less
2014 			 * in this buffer, the mbuf can be discarded:
2015 			 * the last 4 bytes is the CRC, which we don't
2016 			 * care about anyway.
2017 			 */
2018 			if (m->m_len <= ETHER_CRC_LEN) {
2019 				sc->re_tail->m_len -=
2020 				    (ETHER_CRC_LEN - m->m_len);
2021 				m_freem(m);
2022 			} else {
2023 				m->m_len -= ETHER_CRC_LEN;
2024 				sc->re_tail->m_next = m;
2025 			}
2026 			m = sc->re_head;
2027 			sc->re_head = sc->re_tail = NULL;
2028 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2029 		} else {
2030 			m->m_pkthdr.len = m->m_len =
2031 			    (total_len - ETHER_CRC_LEN);
2032 		}
2033 
2034 		ifp->if_ipackets++;
2035 		m->m_pkthdr.rcvif = ifp;
2036 
2037 		/* Do RX checksumming if enabled */
2038 
2039 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2040 			uint8_t packet_type;
2041 
2042 			packet_type = re_packet_type(sc, rxstat, rxctrl);
2043 
2044 			/* Check IP header checksum */
2045 			if (packet_type & RE_IP4_PACKET) {
2046 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2047 				if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
2048 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2049 			}
2050 
2051 			/* Check TCP/UDP checksum */
2052 			if (((packet_type & RE_TCP_PACKET) &&
2053 			     (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
2054 			    ((packet_type & RE_UDP_PACKET) &&
2055 			     (rxstat & RE_RDESC_STAT_UDPSUMBAD) == 0)) {
2056 				m->m_pkthdr.csum_flags |=
2057 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
2058 				    CSUM_FRAG_NOT_CHECKED;
2059 				m->m_pkthdr.csum_data = 0xffff;
2060 			}
2061 		}
2062 
2063 		if (rxctrl & RE_RDESC_CTL_HASTAG) {
2064 			m->m_flags |= M_VLANTAG;
2065 			m->m_pkthdr.ether_vlantag =
2066 				be16toh((rxctrl & RE_RDESC_CTL_TAGDATA));
2067 		}
2068 		ether_input_chain(ifp, m, chain);
2069 	}
2070 
2071 	ether_input_dispatch(chain);
2072 
2073 	/* Flush the RX DMA ring */
2074 
2075 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
2076 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
2077 
2078 	sc->re_ldata.re_rx_prodidx = i;
2079 
2080 	return rx;
2081 }
2082 
2083 #undef RE_IP4_PACKET
2084 #undef RE_TCP_PACKET
2085 #undef RE_UDP_PACKET
2086 
2087 static int
2088 re_tx_collect(struct re_softc *sc)
2089 {
2090 	struct ifnet *ifp = &sc->arpcom.ac_if;
2091 	uint32_t txstat;
2092 	int idx, tx = 0;
2093 
2094 	/* Invalidate the TX descriptor list */
2095 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2096 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
2097 
2098 	for (idx = sc->re_ldata.re_tx_considx;
2099 	     sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt;
2100 	     RE_TXDESC_INC(sc, idx)) {
2101 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
2102 		if (txstat & RE_TDESC_CMD_OWN)
2103 			break;
2104 
2105 		tx = 1;
2106 
2107 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
2108 
2109 		/*
2110 		 * We only stash mbufs in the last descriptor
2111 		 * in a fragment chain, which also happens to
2112 		 * be the only place where the TX status bits
2113 		 * are valid.
2114 		 */
2115 		if (txstat & RE_TDESC_CMD_EOF) {
2116 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
2117 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
2118 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2119 			    sc->re_ldata.re_tx_dmamap[idx]);
2120 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
2121 			    RE_TDESC_STAT_COLCNT))
2122 				ifp->if_collisions++;
2123 			if (txstat & RE_TDESC_STAT_TXERRSUM)
2124 				ifp->if_oerrors++;
2125 			else
2126 				ifp->if_opackets++;
2127 		}
2128 		sc->re_ldata.re_tx_free++;
2129 	}
2130 	sc->re_ldata.re_tx_considx = idx;
2131 
2132 	return tx;
2133 }
2134 
2135 static int
2136 re_txeof(struct re_softc *sc)
2137 {
2138 	struct ifnet *ifp = &sc->arpcom.ac_if;
2139 	int tx;
2140 
2141 	tx = re_tx_collect(sc);
2142 
2143 	/* There is enough free TX descs */
2144 	if (sc->re_ldata.re_tx_free > RE_TXDESC_SPARE)
2145 		ifp->if_flags &= ~IFF_OACTIVE;
2146 
2147 	/*
2148 	 * Some chips will ignore a second TX request issued while an
2149 	 * existing transmission is in progress. If the transmitter goes
2150 	 * idle but there are still packets waiting to be sent, we need
2151 	 * to restart the channel here to flush them out. This only seems
2152 	 * to be required with the PCIe devices.
2153 	 */
2154 	if (sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt)
2155 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2156 	else
2157 		ifp->if_timer = 0;
2158 
2159 	return tx;
2160 }
2161 
2162 static void
2163 re_tick(void *xsc)
2164 {
2165 	struct re_softc *sc = xsc;
2166 
2167 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
2168 	re_tick_serialized(xsc);
2169 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
2170 }
2171 
2172 static void
2173 re_tick_serialized(void *xsc)
2174 {
2175 	struct re_softc *sc = xsc;
2176 	struct ifnet *ifp = &sc->arpcom.ac_if;
2177 	struct mii_data *mii;
2178 
2179 	ASSERT_SERIALIZED(ifp->if_serializer);
2180 
2181 	mii = device_get_softc(sc->re_miibus);
2182 	mii_tick(mii);
2183 	if (sc->re_flags & RE_F_LINKED) {
2184 		if (!(mii->mii_media_status & IFM_ACTIVE))
2185 			sc->re_flags &= ~RE_F_LINKED;
2186 	} else {
2187 		if (mii->mii_media_status & IFM_ACTIVE &&
2188 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2189 			sc->re_flags |= RE_F_LINKED;
2190 			if (!ifq_is_empty(&ifp->if_snd))
2191 				if_devstart(ifp);
2192 		}
2193 	}
2194 
2195 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2196 }
2197 
2198 #ifdef DEVICE_POLLING
2199 
2200 static void
2201 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2202 {
2203 	struct re_softc *sc = ifp->if_softc;
2204 
2205 	ASSERT_SERIALIZED(ifp->if_serializer);
2206 
2207 	switch(cmd) {
2208 	case POLL_REGISTER:
2209 		/* disable interrupts */
2210 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2211 		break;
2212 
2213 	case POLL_DEREGISTER:
2214 		/* enable interrupts */
2215 		re_setup_intr(sc, 1, sc->re_imtype);
2216 		break;
2217 
2218 	default:
2219 		sc->rxcycles = count;
2220 		re_rxeof(sc);
2221 		re_txeof(sc);
2222 
2223 		if (!ifq_is_empty(&ifp->if_snd))
2224 			if_devstart(ifp);
2225 
2226 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2227 			uint16_t       status;
2228 
2229 			status = CSR_READ_2(sc, RE_ISR);
2230 			if (status == 0xffff)
2231 				return;
2232 			if (status)
2233 				CSR_WRITE_2(sc, RE_ISR, status);
2234 
2235 			/*
2236 			 * XXX check behaviour on receiver stalls.
2237 			 */
2238 
2239 			if (status & RE_ISR_SYSTEM_ERR)
2240 				re_init(sc);
2241 		}
2242 		break;
2243 	}
2244 }
2245 #endif /* DEVICE_POLLING */
2246 
2247 static void
2248 re_intr(void *arg)
2249 {
2250 	struct re_softc	*sc = arg;
2251 	struct ifnet *ifp = &sc->arpcom.ac_if;
2252 	uint16_t status;
2253 	int rx, tx;
2254 
2255 	ASSERT_SERIALIZED(ifp->if_serializer);
2256 
2257 	if ((sc->re_flags & RE_F_SUSPENDED) ||
2258 	    (ifp->if_flags & IFF_RUNNING) == 0)
2259 		return;
2260 
2261 	rx = tx = 0;
2262 	for (;;) {
2263 		status = CSR_READ_2(sc, RE_ISR);
2264 		/* If the card has gone away the read returns 0xffff. */
2265 		if (status == 0xffff)
2266 			break;
2267 		if (status)
2268 			CSR_WRITE_2(sc, RE_ISR, status);
2269 
2270 		if ((status & sc->re_intrs) == 0)
2271 			break;
2272 
2273 		if (status & (sc->re_rx_ack | RE_ISR_RX_ERR))
2274 			rx |= re_rxeof(sc);
2275 
2276 		if (status & (sc->re_tx_ack | RE_ISR_TX_ERR))
2277 			tx |= re_txeof(sc);
2278 
2279 		if (status & RE_ISR_SYSTEM_ERR)
2280 			re_init(sc);
2281 
2282 		if (status & RE_ISR_LINKCHG) {
2283 			callout_stop(&sc->re_timer);
2284 			re_tick_serialized(sc);
2285 		}
2286 	}
2287 
2288 	if (sc->re_imtype == RE_IMTYPE_SIM) {
2289 		if ((sc->re_flags & RE_F_TIMER_INTR)) {
2290 			if ((tx | rx) == 0) {
2291 				/*
2292 				 * Nothing needs to be processed, fallback
2293 				 * to use TX/RX interrupts.
2294 				 */
2295 				re_setup_intr(sc, 1, RE_IMTYPE_NONE);
2296 
2297 				/*
2298 				 * Recollect, mainly to avoid the possible
2299 				 * race introduced by changing interrupt
2300 				 * masks.
2301 				 */
2302 				re_rxeof(sc);
2303 				tx = re_txeof(sc);
2304 			} else {
2305 				CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
2306 			}
2307 		} else if (tx | rx) {
2308 			/*
2309 			 * Assume that using simulated interrupt moderation
2310 			 * (hardware timer based) could reduce the interript
2311 			 * rate.
2312 			 */
2313 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
2314 		}
2315 	}
2316 
2317 	if (tx && !ifq_is_empty(&ifp->if_snd))
2318 		if_devstart(ifp);
2319 }
2320 
2321 static int
2322 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx0)
2323 {
2324 	struct ifnet *ifp = &sc->arpcom.ac_if;
2325 	struct mbuf *m;
2326 	struct re_dmaload_arg arg;
2327 	bus_dma_segment_t segs[RE_MAXSEGS];
2328 	bus_dmamap_t map;
2329 	int error, maxsegs, idx, i;
2330 	struct re_desc *d, *tx_ring;
2331 	uint32_t cmd_csum, ctl_csum, vlantag;
2332 
2333 	KASSERT(sc->re_ldata.re_tx_free > RE_TXDESC_SPARE,
2334 		("not enough free TX desc\n"));
2335 
2336 	m = *m_head;
2337 	map = sc->re_ldata.re_tx_dmamap[*idx0];
2338 
2339 	/*
2340 	 * Set up checksum offload. Note: checksum offload bits must
2341 	 * appear in all descriptors of a multi-descriptor transmit
2342 	 * attempt. (This is according to testing done with an 8169
2343 	 * chip. I'm not sure if this is a requirement or a bug.)
2344 	 */
2345 	cmd_csum = ctl_csum = 0;
2346 	if (m->m_pkthdr.csum_flags & CSUM_IP) {
2347 		cmd_csum |= RE_TDESC_CMD_IPCSUM;
2348 		ctl_csum |= RE_TDESC_CTL_IPCSUM;
2349 	}
2350 	if (m->m_pkthdr.csum_flags & CSUM_TCP) {
2351 		cmd_csum |= RE_TDESC_CMD_TCPCSUM;
2352 		ctl_csum |= RE_TDESC_CTL_TCPCSUM;
2353 	}
2354 	if (m->m_pkthdr.csum_flags & CSUM_UDP) {
2355 		cmd_csum |= RE_TDESC_CMD_UDPCSUM;
2356 		ctl_csum |= RE_TDESC_CTL_UDPCSUM;
2357 	}
2358 
2359 	/* For MAC2 chips, csum flags are set on re_control */
2360 	if (sc->re_caps & RE_C_MAC2)
2361 		cmd_csum = 0;
2362 	else
2363 		ctl_csum = 0;
2364 
2365 	if (m->m_pkthdr.len > sc->re_swcsum_lim &&
2366 	    (m->m_pkthdr.csum_flags & (CSUM_DELAY_IP | CSUM_DELAY_DATA))) {
2367 		struct ether_header *eh;
2368 		struct ip *ip;
2369 		u_short offset;
2370 
2371 		m = m_pullup(m, sizeof(struct ether_header *));
2372 		if (m == NULL) {
2373 			*m_head = NULL;
2374 			return ENOBUFS;
2375 		}
2376 		eh = mtod(m, struct ether_header *);
2377 
2378 		/* XXX */
2379 		if (eh->ether_type == ETHERTYPE_VLAN)
2380 			offset = sizeof(struct ether_vlan_header);
2381 		else
2382 			offset = sizeof(struct ether_header);
2383 
2384 		m = m_pullup(m, offset + sizeof(struct ip *));
2385 		if (m == NULL) {
2386 			*m_head = NULL;
2387 			return ENOBUFS;
2388 		}
2389 		ip = (struct ip *)(mtod(m, uint8_t *) + offset);
2390 
2391 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2392 			u_short csum;
2393 
2394 			offset += IP_VHL_HL(ip->ip_vhl) << 2;
2395 			csum = in_cksum_skip(m, ntohs(ip->ip_len), offset);
2396 			if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
2397 				csum = 0xffff;
2398 			offset += m->m_pkthdr.csum_data;        /* checksum offset */
2399 			*(u_short *)(m->m_data + offset) = csum;
2400 
2401 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2402 		}
2403 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_IP) {
2404 			ip->ip_sum = 0;
2405 			if (ip->ip_vhl == IP_VHL_BORING) {
2406 				ip->ip_sum = in_cksum_hdr(ip);
2407 			} else {
2408 				ip->ip_sum =
2409 				in_cksum(m, IP_VHL_HL(ip->ip_vhl) << 2);
2410 			}
2411 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
2412 		}
2413 		*m_head = m; /* 'm' may be changed by above two m_pullup() */
2414 
2415 		/* Clear hardware CSUM flags */
2416 		cmd_csum = ctl_csum = 0;
2417 	}
2418 
2419 	if ((sc->re_caps & RE_C_AUTOPAD) == 0) {
2420 		/*
2421 		 * With some of the RealTek chips, using the checksum offload
2422 		 * support in conjunction with the autopadding feature results
2423 		 * in the transmission of corrupt frames. For example, if we
2424 		 * need to send a really small IP fragment that's less than 60
2425 		 * bytes in size, and IP header checksumming is enabled, the
2426 		 * resulting ethernet frame that appears on the wire will
2427 		 * have garbled payload. To work around this, if TX checksum
2428 		 * offload is enabled, we always manually pad short frames out
2429 		 * to the minimum ethernet frame size.
2430 		 *
2431 		 * Note: this appears unnecessary for TCP, and doing it for TCP
2432 		 * with PCIe adapters seems to result in bad checksums.
2433 		 */
2434 		if ((m->m_pkthdr.csum_flags &
2435 		     (CSUM_DELAY_IP | CSUM_DELAY_DATA)) &&
2436 		    (m->m_pkthdr.csum_flags & CSUM_TCP) == 0 &&
2437 		    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
2438 			error = m_devpad(m, RE_MIN_FRAMELEN);
2439 			if (error)
2440 				goto back;
2441 		}
2442 	}
2443 
2444 	vlantag = 0;
2445 	if (m->m_flags & M_VLANTAG) {
2446 		vlantag = htobe16(m->m_pkthdr.ether_vlantag) |
2447 			  RE_TDESC_CTL_INSTAG;
2448 	}
2449 
2450 	maxsegs = sc->re_ldata.re_tx_free;
2451 	if (maxsegs > RE_MAXSEGS)
2452 		maxsegs = RE_MAXSEGS;
2453 
2454 	arg.re_nsegs = maxsegs;
2455 	arg.re_segs = segs;
2456 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2457 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2458 	if (error && error != EFBIG) {
2459 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
2460 		goto back;
2461 	}
2462 
2463 	/*
2464 	 * Too many segments to map, coalesce into a single mbuf
2465 	 */
2466 	if (!error && arg.re_nsegs == 0) {
2467 		bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2468 		error = EFBIG;
2469 	}
2470 	if (error) {
2471 		struct mbuf *m_new;
2472 
2473 		m_new = m_defrag(m, MB_DONTWAIT);
2474 		if (m_new == NULL) {
2475 			if_printf(ifp, "can't defrag TX mbuf\n");
2476 			error = ENOBUFS;
2477 			goto back;
2478 		} else {
2479 			*m_head = m = m_new;
2480 		}
2481 
2482 		arg.re_nsegs = maxsegs;
2483 		arg.re_segs = segs;
2484 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2485 					     re_dma_map_desc, &arg,
2486 					     BUS_DMA_NOWAIT);
2487 		if (error || arg.re_nsegs == 0) {
2488 			if (!error) {
2489 				bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2490 				error = EFBIG;
2491 			}
2492 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
2493 			goto back;
2494 		}
2495 	}
2496 	bus_dmamap_sync(sc->re_ldata.re_mtag, map, BUS_DMASYNC_PREWRITE);
2497 
2498 	/*
2499 	 * Map the segment array into descriptors.  We also keep track
2500 	 * of the end of the ring and set the end-of-ring bits as needed,
2501 	 * and we set the ownership bits in all except the very first
2502 	 * descriptor, whose ownership bits will be turned on later.
2503 	 */
2504 	tx_ring = sc->re_ldata.re_tx_list;
2505 	idx = *idx0;
2506 	i = 0;
2507 	for (;;) {
2508 		uint32_t cmdstat;
2509 
2510 		d = &tx_ring[idx];
2511 
2512 		cmdstat = segs[i].ds_len;
2513 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
2514 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
2515 		if (i == 0)
2516 			cmdstat |= RE_TDESC_CMD_SOF;
2517 		else
2518 			cmdstat |= RE_TDESC_CMD_OWN;
2519 		if (idx == (sc->re_tx_desc_cnt - 1))
2520 			cmdstat |= RE_TDESC_CMD_EOR;
2521 		d->re_cmdstat = htole32(cmdstat | cmd_csum);
2522 		d->re_control = htole32(ctl_csum | vlantag);
2523 
2524 		i++;
2525 		if (i == arg.re_nsegs)
2526 			break;
2527 		RE_TXDESC_INC(sc, idx);
2528 	}
2529 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
2530 
2531 	/* Transfer ownership of packet to the chip. */
2532 	d->re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2533 	if (*idx0 != idx)
2534 		tx_ring[*idx0].re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2535 
2536 	/*
2537 	 * Insure that the map for this transmission
2538 	 * is placed at the array index of the last descriptor
2539 	 * in this chain.
2540 	 */
2541 	sc->re_ldata.re_tx_dmamap[*idx0] = sc->re_ldata.re_tx_dmamap[idx];
2542 	sc->re_ldata.re_tx_dmamap[idx] = map;
2543 
2544 	sc->re_ldata.re_tx_mbuf[idx] = m;
2545 	sc->re_ldata.re_tx_free -= arg.re_nsegs;
2546 
2547 	RE_TXDESC_INC(sc, idx);
2548 	*idx0 = idx;
2549 back:
2550 	if (error) {
2551 		m_freem(m);
2552 		*m_head = NULL;
2553 	}
2554 	return error;
2555 }
2556 
2557 /*
2558  * Main transmit routine for C+ and gigE NICs.
2559  */
2560 
2561 static void
2562 re_start(struct ifnet *ifp)
2563 {
2564 	struct re_softc	*sc = ifp->if_softc;
2565 	struct mbuf *m_head;
2566 	int idx, need_trans, oactive, error;
2567 
2568 	ASSERT_SERIALIZED(ifp->if_serializer);
2569 
2570 	if ((sc->re_flags & RE_F_LINKED) == 0) {
2571 		ifq_purge(&ifp->if_snd);
2572 		return;
2573 	}
2574 
2575 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2576 		return;
2577 
2578 	idx = sc->re_ldata.re_tx_prodidx;
2579 
2580 	need_trans = 0;
2581 	oactive = 0;
2582 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2583 		if (sc->re_ldata.re_tx_free <= RE_TXDESC_SPARE) {
2584 			if (!oactive) {
2585 				if (re_tx_collect(sc)) {
2586 					oactive = 1;
2587 					continue;
2588 				}
2589 			}
2590 			ifp->if_flags |= IFF_OACTIVE;
2591 			break;
2592 		}
2593 
2594 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2595 		if (m_head == NULL)
2596 			break;
2597 
2598 		error = re_encap(sc, &m_head, &idx);
2599 		if (error) {
2600 			/* m_head is freed by re_encap(), if we reach here */
2601 			ifp->if_oerrors++;
2602 
2603 			if (error == EFBIG && !oactive) {
2604 				if (re_tx_collect(sc)) {
2605 					oactive = 1;
2606 					continue;
2607 				}
2608 			}
2609 			ifp->if_flags |= IFF_OACTIVE;
2610 			break;
2611 		}
2612 
2613 		oactive = 0;
2614 		need_trans = 1;
2615 
2616 		/*
2617 		 * If there's a BPF listener, bounce a copy of this frame
2618 		 * to him.
2619 		 */
2620 		ETHER_BPF_MTAP(ifp, m_head);
2621 	}
2622 
2623 	if (!need_trans)
2624 		return;
2625 
2626 	/* Flush the TX descriptors */
2627 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2628 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2629 
2630 	sc->re_ldata.re_tx_prodidx = idx;
2631 
2632 	/*
2633 	 * RealTek put the TX poll request register in a different
2634 	 * location on the 8169 gigE chip. I don't know why.
2635 	 */
2636 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2637 
2638 	/*
2639 	 * Set a timeout in case the chip goes out to lunch.
2640 	 */
2641 	ifp->if_timer = 5;
2642 }
2643 
2644 static void
2645 re_init(void *xsc)
2646 {
2647 	struct re_softc *sc = xsc;
2648 	struct ifnet *ifp = &sc->arpcom.ac_if;
2649 	struct mii_data *mii;
2650 	int error, framelen;
2651 
2652 	ASSERT_SERIALIZED(ifp->if_serializer);
2653 
2654 	mii = device_get_softc(sc->re_miibus);
2655 
2656 	/*
2657 	 * Cancel pending I/O and free all RX/TX buffers.
2658 	 */
2659 	re_stop(sc);
2660 
2661 	if (sc->re_caps & RE_C_CONTIGRX) {
2662 		if (ifp->if_mtu > ETHERMTU) {
2663 			KKASSERT(sc->re_ldata.re_jbuf != NULL);
2664 			sc->re_flags |= RE_F_USE_JPOOL;
2665 			sc->re_rxbuf_size = RE_FRAMELEN_MAX;
2666 			sc->re_newbuf = re_newbuf_jumbo;
2667 		} else {
2668 			sc->re_flags &= ~RE_F_USE_JPOOL;
2669 			sc->re_rxbuf_size = MCLBYTES;
2670 			sc->re_newbuf = re_newbuf_std;
2671 		}
2672 	}
2673 
2674 	/*
2675 	 * Adjust max read request size according to MTU.
2676 	 * Mainly to improve TX performance for common case (ETHERMTU).
2677 	 */
2678 	if (sc->re_caps & RE_C_PCIE) {
2679 		if (ifp->if_mtu > ETHERMTU) {
2680 			/*
2681 			 * 512 seems to be the only value that works
2682 			 * reliably with jumbo frame
2683 			 */
2684 			pcie_set_max_readrq(sc->re_dev,
2685 				PCIEM_DEVCTL_MAX_READRQ_512);
2686 		} else {
2687 			pcie_set_max_readrq(sc->re_dev,
2688 				PCIEM_DEVCTL_MAX_READRQ_4096);
2689 		}
2690 	}
2691 
2692 	/*
2693 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2694 	 * RX checksum offload. We must configure the C+ register
2695 	 * before all others.
2696 	 */
2697 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2698 		    RE_CPLUSCMD_PCI_MRW |
2699 		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING ?
2700 		     RE_CPLUSCMD_VLANSTRIP : 0) |
2701 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2702 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2703 
2704 	/*
2705 	 * Init our MAC address.  Even though the chipset
2706 	 * documentation doesn't mention it, we need to enter "Config
2707 	 * register write enable" mode to modify the ID registers.
2708 	 */
2709 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2710 	CSR_WRITE_4(sc, RE_IDR0,
2711 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2712 	CSR_WRITE_2(sc, RE_IDR4,
2713 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2714 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2715 
2716 	/*
2717 	 * For C+ mode, initialize the RX descriptors and mbufs.
2718 	 */
2719 	error = re_rx_list_init(sc);
2720 	if (error) {
2721 		re_stop(sc);
2722 		return;
2723 	}
2724 	error = re_tx_list_init(sc);
2725 	if (error) {
2726 		re_stop(sc);
2727 		return;
2728 	}
2729 
2730 	/*
2731 	 * Load the addresses of the RX and TX lists into the chip.
2732 	 */
2733 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2734 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2735 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2736 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2737 
2738 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2739 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2740 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2741 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2742 
2743 	/*
2744 	 * Enable transmit and receive.
2745 	 */
2746 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2747 
2748 	/*
2749 	 * Set the initial TX and RX configuration.
2750 	 */
2751 	if (sc->re_flags & RE_F_TESTMODE) {
2752 		if (!RE_IS_8139CP(sc))
2753 			CSR_WRITE_4(sc, RE_TXCFG,
2754 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2755 		else
2756 			CSR_WRITE_4(sc, RE_TXCFG,
2757 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2758 	} else
2759 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2760 
2761 	framelen = RE_FRAMELEN(ifp->if_mtu);
2762 	if (framelen < MCLBYTES)
2763 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(MCLBYTES, 128));
2764 	else
2765 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(framelen, 128));
2766 
2767 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2768 
2769 	/*
2770 	 * Program the multicast filter, if necessary.
2771 	 */
2772 	re_setmulti(sc);
2773 
2774 #ifdef DEVICE_POLLING
2775 	/*
2776 	 * Disable interrupts if we are polling.
2777 	 */
2778 	if (ifp->if_flags & IFF_POLLING)
2779 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2780 	else	/* otherwise ... */
2781 #endif /* DEVICE_POLLING */
2782 	/*
2783 	 * Enable interrupts.
2784 	 */
2785 	if (sc->re_flags & RE_F_TESTMODE)
2786 		CSR_WRITE_2(sc, RE_IMR, 0);
2787 	else
2788 		re_setup_intr(sc, 1, sc->re_imtype);
2789 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2790 
2791 	/* Start RX/TX process. */
2792 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2793 
2794 #ifdef notdef
2795 	/* Enable receiver and transmitter. */
2796 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2797 #endif
2798 
2799 	/*
2800 	 * For 8169 gigE NICs, set the max allowed RX packet
2801 	 * size so we can receive jumbo frames.
2802 	 */
2803 	if (!RE_IS_8139CP(sc)) {
2804 		if (sc->re_caps & RE_C_CONTIGRX)
2805 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, sc->re_rxbuf_size);
2806 		else
2807 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2808 	}
2809 
2810 	if (sc->re_flags & RE_F_TESTMODE)
2811 		return;
2812 
2813 	mii_mediachg(mii);
2814 
2815 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2816 
2817 	ifp->if_flags |= IFF_RUNNING;
2818 	ifp->if_flags &= ~IFF_OACTIVE;
2819 
2820 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2821 }
2822 
2823 /*
2824  * Set media options.
2825  */
2826 static int
2827 re_ifmedia_upd(struct ifnet *ifp)
2828 {
2829 	struct re_softc *sc = ifp->if_softc;
2830 	struct mii_data *mii;
2831 
2832 	ASSERT_SERIALIZED(ifp->if_serializer);
2833 
2834 	mii = device_get_softc(sc->re_miibus);
2835 	mii_mediachg(mii);
2836 
2837 	return(0);
2838 }
2839 
2840 /*
2841  * Report current media status.
2842  */
2843 static void
2844 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2845 {
2846 	struct re_softc *sc = ifp->if_softc;
2847 	struct mii_data *mii;
2848 
2849 	ASSERT_SERIALIZED(ifp->if_serializer);
2850 
2851 	mii = device_get_softc(sc->re_miibus);
2852 
2853 	mii_pollstat(mii);
2854 	ifmr->ifm_active = mii->mii_media_active;
2855 	ifmr->ifm_status = mii->mii_media_status;
2856 }
2857 
2858 static int
2859 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2860 {
2861 	struct re_softc *sc = ifp->if_softc;
2862 	struct ifreq *ifr = (struct ifreq *) data;
2863 	struct mii_data *mii;
2864 	int error = 0, mask;
2865 
2866 	ASSERT_SERIALIZED(ifp->if_serializer);
2867 
2868 	switch(command) {
2869 	case SIOCSIFMTU:
2870 		if (ifr->ifr_mtu > sc->re_maxmtu) {
2871 			error = EINVAL;
2872 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
2873 			ifp->if_mtu = ifr->ifr_mtu;
2874 			if (ifp->if_flags & IFF_RUNNING)
2875 				ifp->if_init(sc);
2876 		}
2877 		break;
2878 
2879 	case SIOCSIFFLAGS:
2880 		if (ifp->if_flags & IFF_UP) {
2881 			if (ifp->if_flags & IFF_RUNNING) {
2882 				if ((ifp->if_flags ^ sc->re_if_flags) &
2883 				    (IFF_PROMISC | IFF_ALLMULTI))
2884 					re_setmulti(sc);
2885 			} else {
2886 				re_init(sc);
2887 			}
2888 		} else if (ifp->if_flags & IFF_RUNNING) {
2889 			re_stop(sc);
2890 		}
2891 		sc->re_if_flags = ifp->if_flags;
2892 		break;
2893 
2894 	case SIOCADDMULTI:
2895 	case SIOCDELMULTI:
2896 		re_setmulti(sc);
2897 		break;
2898 
2899 	case SIOCGIFMEDIA:
2900 	case SIOCSIFMEDIA:
2901 		mii = device_get_softc(sc->re_miibus);
2902 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2903 		break;
2904 
2905 	case SIOCSIFCAP:
2906 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) &
2907 		       ifp->if_capabilities;
2908 		ifp->if_capenable ^= mask;
2909 
2910 		if (mask & IFCAP_HWCSUM) {
2911 			if (ifp->if_capenable & IFCAP_TXCSUM)
2912 				ifp->if_hwassist = RE_CSUM_FEATURES;
2913 			else
2914 				ifp->if_hwassist = 0;
2915 		}
2916 		if (mask && (ifp->if_flags & IFF_RUNNING))
2917 			re_init(sc);
2918 		break;
2919 
2920 	default:
2921 		error = ether_ioctl(ifp, command, data);
2922 		break;
2923 	}
2924 	return(error);
2925 }
2926 
2927 static void
2928 re_watchdog(struct ifnet *ifp)
2929 {
2930 	struct re_softc *sc = ifp->if_softc;
2931 
2932 	ASSERT_SERIALIZED(ifp->if_serializer);
2933 
2934 	if_printf(ifp, "watchdog timeout\n");
2935 
2936 	ifp->if_oerrors++;
2937 
2938 	re_txeof(sc);
2939 	re_rxeof(sc);
2940 
2941 	re_init(sc);
2942 
2943 	if (!ifq_is_empty(&ifp->if_snd))
2944 		if_devstart(ifp);
2945 }
2946 
2947 /*
2948  * Stop the adapter and free any mbufs allocated to the
2949  * RX and TX lists.
2950  */
2951 static void
2952 re_stop(struct re_softc *sc)
2953 {
2954 	struct ifnet *ifp = &sc->arpcom.ac_if;
2955 	int i;
2956 
2957 	ASSERT_SERIALIZED(ifp->if_serializer);
2958 
2959 	/* Reset the adapter. */
2960 	re_reset(sc, ifp->if_flags & IFF_RUNNING);
2961 
2962 	ifp->if_timer = 0;
2963 	callout_stop(&sc->re_timer);
2964 
2965 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2966 	sc->re_flags &= ~(RE_F_TIMER_INTR | RE_F_DROP_RXFRAG | RE_F_LINKED);
2967 
2968 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2969 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2970 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2971 
2972 	re_free_rxchain(sc);
2973 
2974 	/* Free the TX list buffers. */
2975 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
2976 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2977 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2978 					  sc->re_ldata.re_tx_dmamap[i]);
2979 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2980 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2981 		}
2982 	}
2983 
2984 	/* Free the RX list buffers. */
2985 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
2986 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2987 			if ((sc->re_flags & RE_F_USE_JPOOL) == 0) {
2988 				bus_dmamap_unload(sc->re_ldata.re_mtag,
2989 						  sc->re_ldata.re_rx_dmamap[i]);
2990 			}
2991 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2992 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2993 		}
2994 	}
2995 }
2996 
2997 /*
2998  * Device suspend routine.  Stop the interface and save some PCI
2999  * settings in case the BIOS doesn't restore them properly on
3000  * resume.
3001  */
3002 static int
3003 re_suspend(device_t dev)
3004 {
3005 #ifndef BURN_BRIDGES
3006 	int i;
3007 #endif
3008 	struct re_softc *sc = device_get_softc(dev);
3009 	struct ifnet *ifp = &sc->arpcom.ac_if;
3010 
3011 	lwkt_serialize_enter(ifp->if_serializer);
3012 
3013 	re_stop(sc);
3014 
3015 #ifndef BURN_BRIDGES
3016 	for (i = 0; i < 5; i++)
3017 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
3018 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
3019 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
3020 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3021 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3022 #endif
3023 
3024 	sc->re_flags |= RE_F_SUSPENDED;
3025 
3026 	lwkt_serialize_exit(ifp->if_serializer);
3027 
3028 	return (0);
3029 }
3030 
3031 /*
3032  * Device resume routine.  Restore some PCI settings in case the BIOS
3033  * doesn't, re-enable busmastering, and restart the interface if
3034  * appropriate.
3035  */
3036 static int
3037 re_resume(device_t dev)
3038 {
3039 	struct re_softc *sc = device_get_softc(dev);
3040 	struct ifnet *ifp = &sc->arpcom.ac_if;
3041 #ifndef BURN_BRIDGES
3042 	int i;
3043 #endif
3044 
3045 	lwkt_serialize_enter(ifp->if_serializer);
3046 
3047 #ifndef BURN_BRIDGES
3048 	/* better way to do this? */
3049 	for (i = 0; i < 5; i++)
3050 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
3051 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
3052 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
3053 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
3054 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
3055 
3056 	/* reenable busmastering */
3057 	pci_enable_busmaster(dev);
3058 	pci_enable_io(dev, SYS_RES_IOPORT);
3059 #endif
3060 
3061 	/* reinitialize interface if necessary */
3062 	if (ifp->if_flags & IFF_UP)
3063 		re_init(sc);
3064 
3065 	sc->re_flags &= ~RE_F_SUSPENDED;
3066 
3067 	lwkt_serialize_exit(ifp->if_serializer);
3068 
3069 	return (0);
3070 }
3071 
3072 /*
3073  * Stop all chip I/O so that the kernel's probe routines don't
3074  * get confused by errant DMAs when rebooting.
3075  */
3076 static void
3077 re_shutdown(device_t dev)
3078 {
3079 	struct re_softc *sc = device_get_softc(dev);
3080 	struct ifnet *ifp = &sc->arpcom.ac_if;
3081 
3082 	lwkt_serialize_enter(ifp->if_serializer);
3083 	re_stop(sc);
3084 	lwkt_serialize_exit(ifp->if_serializer);
3085 }
3086 
3087 static int
3088 re_sysctl_rxtime(SYSCTL_HANDLER_ARGS)
3089 {
3090 	struct re_softc *sc = arg1;
3091 
3092 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_rx_time);
3093 }
3094 
3095 static int
3096 re_sysctl_txtime(SYSCTL_HANDLER_ARGS)
3097 {
3098 	struct re_softc *sc = arg1;
3099 
3100 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_tx_time);
3101 }
3102 
3103 static int
3104 re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *hwtime)
3105 {
3106 	struct re_softc *sc = arg1;
3107 	struct ifnet *ifp = &sc->arpcom.ac_if;
3108 	int error, v;
3109 
3110 	lwkt_serialize_enter(ifp->if_serializer);
3111 
3112 	v = *hwtime;
3113 	error = sysctl_handle_int(oidp, &v, 0, req);
3114 	if (error || req->newptr == NULL)
3115 		goto back;
3116 
3117 	if (v <= 0) {
3118 		error = EINVAL;
3119 		goto back;
3120 	}
3121 
3122 	if (v != *hwtime) {
3123 		*hwtime = v;
3124 
3125 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3126 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_HW)
3127 			re_setup_hw_im(sc);
3128 	}
3129 back:
3130 	lwkt_serialize_exit(ifp->if_serializer);
3131 	return error;
3132 }
3133 
3134 static int
3135 re_sysctl_simtime(SYSCTL_HANDLER_ARGS)
3136 {
3137 	struct re_softc *sc = arg1;
3138 	struct ifnet *ifp = &sc->arpcom.ac_if;
3139 	int error, v;
3140 
3141 	lwkt_serialize_enter(ifp->if_serializer);
3142 
3143 	v = sc->re_sim_time;
3144 	error = sysctl_handle_int(oidp, &v, 0, req);
3145 	if (error || req->newptr == NULL)
3146 		goto back;
3147 
3148 	if (v <= 0) {
3149 		error = EINVAL;
3150 		goto back;
3151 	}
3152 
3153 	if (v != sc->re_sim_time) {
3154 		sc->re_sim_time = v;
3155 
3156 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3157 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_SIM) {
3158 #ifdef foo
3159 			int reg;
3160 
3161 			/*
3162 			 * Following code causes various strange
3163 			 * performance problems.  Hmm ...
3164 			 */
3165 			CSR_WRITE_2(sc, RE_IMR, 0);
3166 			if (!RE_IS_8139CP(sc))
3167 				reg = RE_TIMERINT_8169;
3168 			else
3169 				reg = RE_TIMERINT;
3170 			CSR_WRITE_4(sc, reg, 0);
3171 			CSR_READ_4(sc, reg); /* flush */
3172 
3173 			CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3174 			re_setup_sim_im(sc);
3175 #else
3176 			re_setup_intr(sc, 0, RE_IMTYPE_NONE);
3177 			DELAY(10);
3178 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
3179 #endif
3180 		}
3181 	}
3182 back:
3183 	lwkt_serialize_exit(ifp->if_serializer);
3184 	return error;
3185 }
3186 
3187 static int
3188 re_sysctl_imtype(SYSCTL_HANDLER_ARGS)
3189 {
3190 	struct re_softc *sc = arg1;
3191 	struct ifnet *ifp = &sc->arpcom.ac_if;
3192 	int error, v;
3193 
3194 	lwkt_serialize_enter(ifp->if_serializer);
3195 
3196 	v = sc->re_imtype;
3197 	error = sysctl_handle_int(oidp, &v, 0, req);
3198 	if (error || req->newptr == NULL)
3199 		goto back;
3200 
3201 	if (v != RE_IMTYPE_HW && v != RE_IMTYPE_SIM && v != RE_IMTYPE_NONE) {
3202 		error = EINVAL;
3203 		goto back;
3204 	}
3205 	if (v == RE_IMTYPE_HW && (sc->re_caps & RE_C_HWIM) == 0) {
3206 		/* Can't do hardware interrupt moderation */
3207 		error = EOPNOTSUPP;
3208 		goto back;
3209 	}
3210 
3211 	if (v != sc->re_imtype) {
3212 		sc->re_imtype = v;
3213 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3214 		    IFF_RUNNING)
3215 			re_setup_intr(sc, 1, sc->re_imtype);
3216 	}
3217 back:
3218 	lwkt_serialize_exit(ifp->if_serializer);
3219 	return error;
3220 }
3221 
3222 static void
3223 re_setup_hw_im(struct re_softc *sc)
3224 {
3225 	KKASSERT(sc->re_caps & RE_C_HWIM);
3226 
3227 	/*
3228 	 * Interrupt moderation
3229 	 *
3230 	 * 0xABCD
3231 	 * A - unknown (maybe TX related)
3232 	 * B - TX timer (unit: 25us)
3233 	 * C - unknown (maybe RX related)
3234 	 * D - RX timer (unit: 25us)
3235 	 *
3236 	 *
3237 	 * re(4)'s interrupt moderation is actually controlled by
3238 	 * two variables, like most other NICs (bge, bce etc.)
3239 	 * o  timer
3240 	 * o  number of packets [P]
3241 	 *
3242 	 * The logic relationship between these two variables is
3243 	 * similar to other NICs too:
3244 	 * if (timer expire || packets > [P])
3245 	 *     Interrupt is delivered
3246 	 *
3247 	 * Currently we only know how to set 'timer', but not
3248 	 * 'number of packets', which should be ~30, as far as I
3249 	 * tested (sink ~900Kpps, interrupt rate is 30KHz)
3250 	 */
3251 	CSR_WRITE_2(sc, RE_IM,
3252 		    RE_IM_RXTIME(sc->re_rx_time) |
3253 		    RE_IM_TXTIME(sc->re_tx_time) |
3254 		    RE_IM_MAGIC);
3255 }
3256 
3257 static void
3258 re_disable_hw_im(struct re_softc *sc)
3259 {
3260 	if (sc->re_caps & RE_C_HWIM)
3261 		CSR_WRITE_2(sc, RE_IM, 0);
3262 }
3263 
3264 static void
3265 re_setup_sim_im(struct re_softc *sc)
3266 {
3267 	if (!RE_IS_8139CP(sc)) {
3268 		uint32_t ticks;
3269 
3270 		/*
3271 		 * Datasheet says tick decreases at bus speed,
3272 		 * but it seems the clock runs a little bit
3273 		 * faster, so we do some compensation here.
3274 		 */
3275 		ticks = (sc->re_sim_time * sc->re_bus_speed * 8) / 5;
3276 		CSR_WRITE_4(sc, RE_TIMERINT_8169, ticks);
3277 	} else {
3278 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400); /* XXX */
3279 	}
3280 	CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
3281 	sc->re_flags |= RE_F_TIMER_INTR;
3282 }
3283 
3284 static void
3285 re_disable_sim_im(struct re_softc *sc)
3286 {
3287 	if (!RE_IS_8139CP(sc))
3288 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0);
3289 	else
3290 		CSR_WRITE_4(sc, RE_TIMERINT, 0);
3291 	sc->re_flags &= ~RE_F_TIMER_INTR;
3292 }
3293 
3294 static void
3295 re_config_imtype(struct re_softc *sc, int imtype)
3296 {
3297 	switch (imtype) {
3298 	case RE_IMTYPE_HW:
3299 		KKASSERT(sc->re_caps & RE_C_HWIM);
3300 		/* FALL THROUGH */
3301 	case RE_IMTYPE_NONE:
3302 		sc->re_intrs = RE_INTRS;
3303 		sc->re_rx_ack = RE_ISR_RX_OK | RE_ISR_FIFO_OFLOW |
3304 				RE_ISR_RX_OVERRUN;
3305 		sc->re_tx_ack = RE_ISR_TX_OK;
3306 		break;
3307 
3308 	case RE_IMTYPE_SIM:
3309 		sc->re_intrs = RE_INTRS_TIMER;
3310 		sc->re_rx_ack = RE_ISR_TIMEOUT_EXPIRED;
3311 		sc->re_tx_ack = RE_ISR_TIMEOUT_EXPIRED;
3312 		break;
3313 
3314 	default:
3315 		panic("%s: unknown imtype %d\n",
3316 		      sc->arpcom.ac_if.if_xname, imtype);
3317 	}
3318 }
3319 
3320 static void
3321 re_setup_intr(struct re_softc *sc, int enable_intrs, int imtype)
3322 {
3323 	re_config_imtype(sc, imtype);
3324 
3325 	if (enable_intrs)
3326 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3327 	else
3328 		CSR_WRITE_2(sc, RE_IMR, 0);
3329 
3330 	switch (imtype) {
3331 	case RE_IMTYPE_NONE:
3332 		re_disable_sim_im(sc);
3333 		re_disable_hw_im(sc);
3334 		break;
3335 
3336 	case RE_IMTYPE_HW:
3337 		KKASSERT(sc->re_caps & RE_C_HWIM);
3338 		re_disable_sim_im(sc);
3339 		re_setup_hw_im(sc);
3340 		break;
3341 
3342 	case RE_IMTYPE_SIM:
3343 		re_disable_hw_im(sc);
3344 		re_setup_sim_im(sc);
3345 		break;
3346 
3347 	default:
3348 		panic("%s: unknown imtype %d\n",
3349 		      sc->arpcom.ac_if.if_xname, imtype);
3350 	}
3351 }
3352 
3353 static void
3354 re_get_eaddr(struct re_softc *sc, uint8_t *eaddr)
3355 {
3356 	int i;
3357 
3358 	if (sc->re_macver == RE_MACVER_11 || sc->re_macver == RE_MACVER_12) {
3359 		uint16_t re_did;
3360 
3361 		re_get_eewidth(sc);
3362 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
3363 		if (re_did == 0x8128) {
3364 			uint16_t as[ETHER_ADDR_LEN / 2];
3365 
3366 			/*
3367 			 * Get station address from the EEPROM.
3368 			 */
3369 			re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
3370 			for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
3371 				as[i] = le16toh(as[i]);
3372 			bcopy(as, eaddr, sizeof(eaddr));
3373 			return;
3374 		}
3375 	}
3376 
3377 	/*
3378 	 * Get station address from IDRx.
3379 	 */
3380 	for (i = 0; i < ETHER_ADDR_LEN; ++i)
3381 		eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
3382 }
3383 
3384 static int
3385 re_jpool_alloc(struct re_softc *sc)
3386 {
3387 	struct re_list_data *ldata = &sc->re_ldata;
3388 	struct re_jbuf *jbuf;
3389 	bus_addr_t paddr;
3390 	bus_size_t jpool_size;
3391 	caddr_t buf;
3392 	int i, error;
3393 
3394 	lwkt_serialize_init(&ldata->re_jbuf_serializer);
3395 
3396 	ldata->re_jbuf = kmalloc(sizeof(struct re_jbuf) * RE_JBUF_COUNT(sc),
3397 				 M_DEVBUF, M_WAITOK | M_ZERO);
3398 
3399 	jpool_size = RE_JBUF_COUNT(sc) * RE_JBUF_SIZE;
3400 
3401 	error = bus_dma_tag_create(sc->re_parent_tag,
3402 			RE_BUF_ALIGN, 0,	/* alignment, boundary */
3403 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
3404 			BUS_SPACE_MAXADDR,	/* highaddr */
3405 			NULL, NULL,		/* filter, filterarg */
3406 			jpool_size, 1,		/* nsegments, maxsize */
3407 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
3408 			BUS_DMA_ALLOCNOW,	/* flags */
3409 			&ldata->re_jpool_tag);
3410 	if (error) {
3411 		device_printf(sc->re_dev, "could not allocate jumbo dma tag\n");
3412 		return error;
3413 	}
3414 
3415 	error = bus_dmamem_alloc(ldata->re_jpool_tag, (void **)&ldata->re_jpool,
3416 				 BUS_DMA_WAITOK, &ldata->re_jpool_map);
3417 	if (error) {
3418 		device_printf(sc->re_dev,
3419 			      "could not allocate jumbo dma memory\n");
3420 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3421 		ldata->re_jpool_tag = NULL;
3422 		return error;
3423 	}
3424 
3425 	error = bus_dmamap_load(ldata->re_jpool_tag, ldata->re_jpool_map,
3426 				ldata->re_jpool, jpool_size,
3427 				re_dma_map_addr, &paddr, BUS_DMA_WAITOK);
3428 	if (error) {
3429 		device_printf(sc->re_dev, "could not load jumbo dma map\n");
3430 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3431 				ldata->re_jpool_map);
3432 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3433 		ldata->re_jpool_tag = NULL;
3434 		return error;
3435 	}
3436 
3437 	/* ..and split it into 9KB chunks */
3438 	SLIST_INIT(&ldata->re_jbuf_free);
3439 
3440 	buf = ldata->re_jpool;
3441 	for (i = 0; i < RE_JBUF_COUNT(sc); i++) {
3442 		jbuf = &ldata->re_jbuf[i];
3443 
3444 		jbuf->re_sc = sc;
3445 		jbuf->re_inuse = 0;
3446 		jbuf->re_slot = i;
3447 		jbuf->re_buf = buf;
3448 		jbuf->re_paddr = paddr;
3449 
3450 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3451 
3452 		buf += RE_JBUF_SIZE;
3453 		paddr += RE_JBUF_SIZE;
3454 	}
3455 	return 0;
3456 }
3457 
3458 static void
3459 re_jpool_free(struct re_softc *sc)
3460 {
3461 	struct re_list_data *ldata = &sc->re_ldata;
3462 
3463 	if (ldata->re_jpool_tag != NULL) {
3464 		bus_dmamap_unload(ldata->re_jpool_tag, ldata->re_jpool_map);
3465 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3466 				ldata->re_jpool_map);
3467 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3468 		ldata->re_jpool_tag = NULL;
3469 	}
3470 
3471 	if (ldata->re_jbuf != NULL) {
3472 		kfree(ldata->re_jbuf, M_DEVBUF);
3473 		ldata->re_jbuf = NULL;
3474 	}
3475 }
3476 
3477 static struct re_jbuf *
3478 re_jbuf_alloc(struct re_softc *sc)
3479 {
3480 	struct re_list_data *ldata = &sc->re_ldata;
3481 	struct re_jbuf *jbuf;
3482 
3483 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3484 
3485 	jbuf = SLIST_FIRST(&ldata->re_jbuf_free);
3486 	if (jbuf != NULL) {
3487 		SLIST_REMOVE_HEAD(&ldata->re_jbuf_free, re_link);
3488 		jbuf->re_inuse = 1;
3489 	}
3490 
3491 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3492 
3493 	return jbuf;
3494 }
3495 
3496 static void
3497 re_jbuf_free(void *arg)
3498 {
3499 	struct re_jbuf *jbuf = arg;
3500 	struct re_softc *sc = jbuf->re_sc;
3501 	struct re_list_data *ldata = &sc->re_ldata;
3502 
3503 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3504 		panic("%s: free wrong jumbo buffer\n",
3505 		      sc->arpcom.ac_if.if_xname);
3506 	} else if (jbuf->re_inuse == 0) {
3507 		panic("%s: jumbo buffer already freed\n",
3508 		      sc->arpcom.ac_if.if_xname);
3509 	}
3510 
3511 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3512 	atomic_subtract_int(&jbuf->re_inuse, 1);
3513 	if (jbuf->re_inuse == 0)
3514 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3515 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3516 }
3517 
3518 static void
3519 re_jbuf_ref(void *arg)
3520 {
3521 	struct re_jbuf *jbuf = arg;
3522 	struct re_softc *sc = jbuf->re_sc;
3523 	struct re_list_data *ldata = &sc->re_ldata;
3524 
3525 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3526 		panic("%s: ref wrong jumbo buffer\n",
3527 		      sc->arpcom.ac_if.if_xname);
3528 	} else if (jbuf->re_inuse == 0) {
3529 		panic("%s: jumbo buffer already freed\n",
3530 		      sc->arpcom.ac_if.if_xname);
3531 	}
3532 	atomic_add_int(&jbuf->re_inuse, 1);
3533 }
3534