xref: /dflybsd-src/sys/dev/netif/re/if_re.c (revision 14bf4c38b8c66c6bd7b478a1220fa32cd4b83cbd)
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 1.99 2008/10/30 11:27:40 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 
1022 	if (bootverbose) {
1023 		device_printf(dev, "unknown hwrev 0x%08x, macmode 0x%08x\n",
1024 			      hwrev, macmode);
1025 	}
1026 	return ENXIO;
1027 }
1028 
1029 static void
1030 re_dma_map_desc(void *xarg, bus_dma_segment_t *segs, int nsegs,
1031 		bus_size_t mapsize, int error)
1032 {
1033 	struct re_dmaload_arg *arg = xarg;
1034 	int i;
1035 
1036 	if (error)
1037 		return;
1038 
1039 	if (nsegs > arg->re_nsegs) {
1040 		arg->re_nsegs = 0;
1041 		return;
1042 	}
1043 
1044 	arg->re_nsegs = nsegs;
1045 	for (i = 0; i < nsegs; ++i)
1046 		arg->re_segs[i] = segs[i];
1047 }
1048 
1049 /*
1050  * Map a single buffer address.
1051  */
1052 
1053 static void
1054 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1055 {
1056 	uint32_t *addr;
1057 
1058 	if (error)
1059 		return;
1060 
1061 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1062 	addr = arg;
1063 	*addr = segs->ds_addr;
1064 }
1065 
1066 static int
1067 re_allocmem(device_t dev)
1068 {
1069 	struct re_softc *sc = device_get_softc(dev);
1070 	int error, i;
1071 
1072 	/*
1073 	 * Allocate list data
1074 	 */
1075 	sc->re_ldata.re_tx_mbuf =
1076 	kmalloc(sc->re_tx_desc_cnt * sizeof(struct mbuf *),
1077 		M_DEVBUF, M_ZERO | M_WAITOK);
1078 
1079 	sc->re_ldata.re_rx_mbuf =
1080 	kmalloc(sc->re_rx_desc_cnt * sizeof(struct mbuf *),
1081 		M_DEVBUF, M_ZERO | M_WAITOK);
1082 
1083 	sc->re_ldata.re_rx_paddr =
1084 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_addr_t),
1085 		M_DEVBUF, M_ZERO | M_WAITOK);
1086 
1087 	sc->re_ldata.re_tx_dmamap =
1088 	kmalloc(sc->re_tx_desc_cnt * sizeof(bus_dmamap_t),
1089 		M_DEVBUF, M_ZERO | M_WAITOK);
1090 
1091 	sc->re_ldata.re_rx_dmamap =
1092 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_dmamap_t),
1093 		M_DEVBUF, M_ZERO | M_WAITOK);
1094 
1095 	/*
1096 	 * Allocate the parent bus DMA tag appropriate for PCI.
1097 	 */
1098 	error = bus_dma_tag_create(NULL,	/* parent */
1099 			1, 0,			/* alignment, boundary */
1100 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
1101 			BUS_SPACE_MAXADDR,	/* highaddr */
1102 			NULL, NULL,		/* filter, filterarg */
1103 			MAXBSIZE, RE_MAXSEGS,	/* maxsize, nsegments */
1104 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1105 			BUS_DMA_ALLOCNOW,	/* flags */
1106 			&sc->re_parent_tag);
1107 	if (error) {
1108 		device_printf(dev, "could not allocate parent dma tag\n");
1109 		return error;
1110 	}
1111 
1112 	/* Allocate tag for TX descriptor list. */
1113 	error = bus_dma_tag_create(sc->re_parent_tag,
1114 			RE_RING_ALIGN, 0,
1115 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1116 			NULL, NULL,
1117 			RE_TX_LIST_SZ(sc), 1, RE_TX_LIST_SZ(sc),
1118 			BUS_DMA_ALLOCNOW,
1119 			&sc->re_ldata.re_tx_list_tag);
1120 	if (error) {
1121 		device_printf(dev, "could not allocate TX ring dma tag\n");
1122 		return(error);
1123 	}
1124 
1125 	/* Allocate DMA'able memory for the TX ring */
1126         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
1127 			(void **)&sc->re_ldata.re_tx_list,
1128 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1129 			&sc->re_ldata.re_tx_list_map);
1130         if (error) {
1131 		device_printf(dev, "could not allocate TX ring\n");
1132 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1133 		sc->re_ldata.re_tx_list_tag = NULL;
1134                 return(error);
1135 	}
1136 
1137 	/* Load the map for the TX ring. */
1138 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
1139 			sc->re_ldata.re_tx_list_map,
1140 			sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc),
1141 			re_dma_map_addr, &sc->re_ldata.re_tx_list_addr,
1142 			BUS_DMA_NOWAIT);
1143 	if (error) {
1144 		device_printf(dev, "could not get address of TX ring\n");
1145 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1146 				sc->re_ldata.re_tx_list,
1147 				sc->re_ldata.re_tx_list_map);
1148 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1149 		sc->re_ldata.re_tx_list_tag = NULL;
1150 		return(error);
1151 	}
1152 
1153 	/* Allocate tag for RX descriptor list. */
1154 	error = bus_dma_tag_create(sc->re_parent_tag,
1155 			RE_RING_ALIGN, 0,
1156 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1157 			NULL, NULL,
1158 			RE_RX_LIST_SZ(sc), 1, RE_RX_LIST_SZ(sc),
1159 			BUS_DMA_ALLOCNOW,
1160 			&sc->re_ldata.re_rx_list_tag);
1161 	if (error) {
1162 		device_printf(dev, "could not allocate RX ring dma tag\n");
1163 		return(error);
1164 	}
1165 
1166 	/* Allocate DMA'able memory for the RX ring */
1167         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1168 			(void **)&sc->re_ldata.re_rx_list,
1169 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1170 			&sc->re_ldata.re_rx_list_map);
1171         if (error) {
1172 		device_printf(dev, "could not allocate RX ring\n");
1173 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1174 		sc->re_ldata.re_rx_list_tag = NULL;
1175                 return(error);
1176 	}
1177 
1178 	/* Load the map for the RX ring. */
1179 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1180 			sc->re_ldata.re_rx_list_map,
1181 			sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc),
1182 			re_dma_map_addr, &sc->re_ldata.re_rx_list_addr,
1183 			BUS_DMA_NOWAIT);
1184 	if (error) {
1185 		device_printf(dev, "could not get address of RX ring\n");
1186 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1187 				sc->re_ldata.re_rx_list,
1188 				sc->re_ldata.re_rx_list_map);
1189 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1190 		sc->re_ldata.re_rx_list_tag = NULL;
1191 		return(error);
1192 	}
1193 
1194 	/* Allocate map for RX/TX mbufs. */
1195 	error = bus_dma_tag_create(sc->re_parent_tag,
1196 			ETHER_ALIGN, 0,
1197 			BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1198 			NULL, NULL,
1199 			RE_FRAMELEN_MAX, RE_MAXSEGS, MCLBYTES,
1200 			BUS_DMA_ALLOCNOW,
1201 			&sc->re_ldata.re_mtag);
1202 	if (error) {
1203 		device_printf(dev, "could not allocate buf dma tag\n");
1204 		return(error);
1205 	}
1206 
1207 	/* Create spare DMA map for RX */
1208 	error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1209 			&sc->re_ldata.re_rx_spare);
1210 	if (error) {
1211 		device_printf(dev, "can't create spare DMA map for RX\n");
1212 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1213 		sc->re_ldata.re_mtag = NULL;
1214 		return error;
1215 	}
1216 
1217 	/* Create DMA maps for TX buffers */
1218 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
1219 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1220 				&sc->re_ldata.re_tx_dmamap[i]);
1221 		if (error) {
1222 			device_printf(dev, "can't create DMA map for TX buf\n");
1223 			re_freebufmem(sc, i, 0);
1224 			return(error);
1225 		}
1226 	}
1227 
1228 	/* Create DMA maps for RX buffers */
1229 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1230 		error = bus_dmamap_create(sc->re_ldata.re_mtag, 0,
1231 				&sc->re_ldata.re_rx_dmamap[i]);
1232 		if (error) {
1233 			device_printf(dev, "can't create DMA map for RX buf\n");
1234 			re_freebufmem(sc, sc->re_tx_desc_cnt, i);
1235 			return(error);
1236 		}
1237 	}
1238 
1239 	/* Create jumbo buffer pool for RX if required */
1240 	if (sc->re_caps & RE_C_CONTIGRX) {
1241 		error = re_jpool_alloc(sc);
1242 		if (error) {
1243 			re_jpool_free(sc);
1244 			/* Disable jumbo frame support */
1245 			sc->re_maxmtu = ETHERMTU;
1246 		}
1247 	}
1248 	return(0);
1249 }
1250 
1251 static void
1252 re_freebufmem(struct re_softc *sc, int tx_cnt, int rx_cnt)
1253 {
1254 	int i;
1255 
1256 	/* Destroy all the RX and TX buffer maps */
1257 	if (sc->re_ldata.re_mtag) {
1258 		for (i = 0; i < tx_cnt; i++) {
1259 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1260 					   sc->re_ldata.re_tx_dmamap[i]);
1261 		}
1262 		for (i = 0; i < rx_cnt; i++) {
1263 			bus_dmamap_destroy(sc->re_ldata.re_mtag,
1264 					   sc->re_ldata.re_rx_dmamap[i]);
1265 		}
1266 		bus_dmamap_destroy(sc->re_ldata.re_mtag,
1267 				   sc->re_ldata.re_rx_spare);
1268 		bus_dma_tag_destroy(sc->re_ldata.re_mtag);
1269 		sc->re_ldata.re_mtag = NULL;
1270 	}
1271 }
1272 
1273 static void
1274 re_freemem(device_t dev)
1275 {
1276 	struct re_softc *sc = device_get_softc(dev);
1277 
1278 	/* Unload and free the RX DMA ring memory and map */
1279 	if (sc->re_ldata.re_rx_list_tag) {
1280 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1281 				  sc->re_ldata.re_rx_list_map);
1282 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1283 				sc->re_ldata.re_rx_list,
1284 				sc->re_ldata.re_rx_list_map);
1285 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1286 	}
1287 
1288 	/* Unload and free the TX DMA ring memory and map */
1289 	if (sc->re_ldata.re_tx_list_tag) {
1290 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1291 				  sc->re_ldata.re_tx_list_map);
1292 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1293 				sc->re_ldata.re_tx_list,
1294 				sc->re_ldata.re_tx_list_map);
1295 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1296 	}
1297 
1298 	/* Free RX/TX buf DMA stuffs */
1299 	re_freebufmem(sc, sc->re_tx_desc_cnt, sc->re_rx_desc_cnt);
1300 
1301 	/* Unload and free the stats buffer and map */
1302 	if (sc->re_ldata.re_stag) {
1303 		bus_dmamap_unload(sc->re_ldata.re_stag,
1304 				  sc->re_ldata.re_rx_list_map);
1305 		bus_dmamem_free(sc->re_ldata.re_stag,
1306 				sc->re_ldata.re_stats,
1307 				sc->re_ldata.re_smap);
1308 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1309 	}
1310 
1311 	if (sc->re_caps & RE_C_CONTIGRX)
1312 		re_jpool_free(sc);
1313 
1314 	if (sc->re_parent_tag)
1315 		bus_dma_tag_destroy(sc->re_parent_tag);
1316 
1317 	if (sc->re_ldata.re_tx_mbuf != NULL)
1318 		kfree(sc->re_ldata.re_tx_mbuf, M_DEVBUF);
1319 	if (sc->re_ldata.re_rx_mbuf != NULL)
1320 		kfree(sc->re_ldata.re_rx_mbuf, M_DEVBUF);
1321 	if (sc->re_ldata.re_rx_paddr != NULL)
1322 		kfree(sc->re_ldata.re_rx_paddr, M_DEVBUF);
1323 	if (sc->re_ldata.re_tx_dmamap != NULL)
1324 		kfree(sc->re_ldata.re_tx_dmamap, M_DEVBUF);
1325 	if (sc->re_ldata.re_rx_dmamap != NULL)
1326 		kfree(sc->re_ldata.re_rx_dmamap, M_DEVBUF);
1327 }
1328 
1329 /*
1330  * Attach the interface. Allocate softc structures, do ifmedia
1331  * setup and ethernet/BPF attach.
1332  */
1333 static int
1334 re_attach(device_t dev)
1335 {
1336 	struct re_softc	*sc = device_get_softc(dev);
1337 	struct ifnet *ifp;
1338 	uint8_t eaddr[ETHER_ADDR_LEN];
1339 	int error = 0, rid, qlen;
1340 
1341 	callout_init(&sc->re_timer);
1342 	sc->re_dev = dev;
1343 
1344 	if (RE_IS_8139CP(sc)) {
1345 		sc->re_rx_desc_cnt = RE_RX_DESC_CNT_8139CP;
1346 		sc->re_tx_desc_cnt = RE_TX_DESC_CNT_8139CP;
1347 	} else {
1348 		sc->re_rx_desc_cnt = re_rx_desc_count;
1349 		if (sc->re_rx_desc_cnt > RE_RX_DESC_CNT_MAX)
1350 			sc->re_rx_desc_cnt = RE_RX_DESC_CNT_MAX;
1351 
1352 		sc->re_tx_desc_cnt = re_tx_desc_count;
1353 		if (sc->re_tx_desc_cnt > RE_TX_DESC_CNT_MAX)
1354 			sc->re_tx_desc_cnt = RE_TX_DESC_CNT_MAX;
1355 	}
1356 
1357 	qlen = RE_IFQ_MAXLEN;
1358 	if (sc->re_tx_desc_cnt > qlen)
1359 		qlen = sc->re_tx_desc_cnt;
1360 
1361 	sc->re_rxbuf_size = MCLBYTES;
1362 	sc->re_newbuf = re_newbuf_std;
1363 
1364 	sc->re_tx_time = 5;		/* 125us */
1365 	sc->re_rx_time = 2;		/* 50us */
1366 	if (sc->re_caps & RE_C_PCIE)
1367 		sc->re_sim_time = 75;	/* 75us */
1368 	else
1369 		sc->re_sim_time = 125;	/* 125us */
1370 	sc->re_imtype = RE_IMTYPE_SIM;	/* simulated interrupt moderation */
1371 	re_config_imtype(sc, sc->re_imtype);
1372 
1373 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1374 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1375 					     SYSCTL_STATIC_CHILDREN(_hw),
1376 					     OID_AUTO,
1377 					     device_get_nameunit(dev),
1378 					     CTLFLAG_RD, 0, "");
1379 	if (sc->re_sysctl_tree == NULL) {
1380 		device_printf(dev, "can't add sysctl node\n");
1381 		error = ENXIO;
1382 		goto fail;
1383 	}
1384 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1385 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1386 		       "rx_desc_count", CTLFLAG_RD, &sc->re_rx_desc_cnt,
1387 		       0, "RX desc count");
1388 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1389 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1390 		       "tx_desc_count", CTLFLAG_RD, &sc->re_tx_desc_cnt,
1391 		       0, "TX desc count");
1392 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1393 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1394 			OID_AUTO, "sim_time",
1395 			CTLTYPE_INT | CTLFLAG_RW,
1396 			sc, 0, re_sysctl_simtime, "I",
1397 			"Simulated interrupt moderation time (usec).");
1398 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1399 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1400 			OID_AUTO, "imtype",
1401 			CTLTYPE_INT | CTLFLAG_RW,
1402 			sc, 0, re_sysctl_imtype, "I",
1403 			"Interrupt moderation type -- "
1404 			"0:disable, 1:simulated, "
1405 			"2:hardware(if supported)");
1406 	if (sc->re_caps & RE_C_HWIM) {
1407 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1408 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1409 				OID_AUTO, "hw_rxtime",
1410 				CTLTYPE_INT | CTLFLAG_RW,
1411 				sc, 0, re_sysctl_rxtime, "I",
1412 				"Hardware interrupt moderation time "
1413 				"(unit: 25usec).");
1414 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1415 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1416 				OID_AUTO, "hw_txtime",
1417 				CTLTYPE_INT | CTLFLAG_RW,
1418 				sc, 0, re_sysctl_txtime, "I",
1419 				"Hardware interrupt moderation time "
1420 				"(unit: 25usec).");
1421 	}
1422 
1423 #ifndef BURN_BRIDGES
1424 	/*
1425 	 * Handle power management nonsense.
1426 	 */
1427 
1428 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1429 		uint32_t membase, irq;
1430 
1431 		/* Save important PCI config data. */
1432 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1433 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1434 
1435 		/* Reset the power state. */
1436 		device_printf(dev, "chip is in D%d power mode "
1437 		    "-- setting to D0\n", pci_get_powerstate(dev));
1438 
1439 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1440 
1441 		/* Restore PCI config data. */
1442 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1443 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1444 	}
1445 #endif
1446 	/*
1447 	 * Map control/status registers.
1448 	 */
1449 	pci_enable_busmaster(dev);
1450 
1451 	rid = RE_PCI_LOIO;
1452 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1453 					    RF_ACTIVE);
1454 
1455 	if (sc->re_res == NULL) {
1456 		device_printf(dev, "couldn't map ports\n");
1457 		error = ENXIO;
1458 		goto fail;
1459 	}
1460 
1461 	sc->re_btag = rman_get_bustag(sc->re_res);
1462 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1463 
1464 	/* Allocate interrupt */
1465 	rid = 0;
1466 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1467 					    RF_SHAREABLE | RF_ACTIVE);
1468 
1469 	if (sc->re_irq == NULL) {
1470 		device_printf(dev, "couldn't map interrupt\n");
1471 		error = ENXIO;
1472 		goto fail;
1473 	}
1474 
1475 	/* Reset the adapter. */
1476 	re_reset(sc, 0);
1477 
1478 	if (RE_IS_8139CP(sc)) {
1479 		sc->re_bus_speed = 33; /* XXX */
1480 	} else if (sc->re_caps & RE_C_PCIE) {
1481 		sc->re_bus_speed = 125;
1482 	} else {
1483 		uint8_t cfg2;
1484 
1485 		cfg2 = CSR_READ_1(sc, RE_CFG2);
1486 		switch (cfg2 & RE_CFG2_PCICLK_MASK) {
1487 		case RE_CFG2_PCICLK_33MHZ:
1488 			sc->re_bus_speed = 33;
1489 			break;
1490 		case RE_CFG2_PCICLK_66MHZ:
1491 			sc->re_bus_speed = 66;
1492 			break;
1493 		default:
1494 			device_printf(dev, "unknown bus speed, assume 33MHz\n");
1495 			sc->re_bus_speed = 33;
1496 			break;
1497 		}
1498 		if (cfg2 & RE_CFG2_PCI64)
1499 			sc->re_caps |= RE_C_PCI64;
1500 	}
1501 	device_printf(dev, "Hardware rev. 0x%08x; MAC ver. 0x%02x; "
1502 		      "PCI%s %dMHz\n",
1503 		      sc->re_hwrev, sc->re_macver,
1504 		      (sc->re_caps & RE_C_PCIE) ?
1505 		      "-E" : ((sc->re_caps & RE_C_PCI64) ? "64" : "32"),
1506 		      sc->re_bus_speed);
1507 
1508 	/*
1509 	 * NOTE:
1510 	 * DO NOT try to adjust config1 and config5 which was spotted in
1511 	 * Realtek's Linux drivers.  It will _permanently_ damage certain
1512 	 * cards EEPROM, e.g. one of my 8168B (0x38000000) card ...
1513 	 */
1514 
1515 	re_get_eaddr(sc, eaddr);
1516 
1517 	if (!RE_IS_8139CP(sc)) {
1518 		/* Set RX length mask */
1519 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1520 		sc->re_txstart = RE_GTXSTART;
1521 	} else {
1522 		/* Set RX length mask */
1523 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1524 		sc->re_txstart = RE_TXSTART;
1525 	}
1526 
1527 	/* Allocate DMA stuffs */
1528 	error = re_allocmem(dev);
1529 	if (error)
1530 		goto fail;
1531 
1532 	/*
1533 	 * Apply some magic PCI settings from Realtek ...
1534 	 */
1535 	if (RE_IS_8169(sc)) {
1536 		CSR_WRITE_1(sc, 0x82, 1);
1537 		pci_write_config(dev, PCIR_CACHELNSZ, 0x8, 1);
1538 	}
1539 	pci_write_config(dev, PCIR_LATTIMER, 0x40, 1);
1540 
1541 	if (sc->re_caps & RE_C_MAC2) {
1542 		/*
1543 		 * Following part is extracted from Realtek BSD driver v176.
1544 		 * However, this does _not_ make much/any sense:
1545 		 * 8168C's PCI Express device control is located at 0x78,
1546 		 * so the reading from 0x79 (higher part of 0x78) and setting
1547 		 * the 4~6bits intend to enlarge the "max read request size"
1548 		 * (we will do it).  The content of the rest part of this
1549 		 * register is not meaningful to other PCI registers, so
1550 		 * writing the value to 0x54 could be completely wrong.
1551 		 * 0x80 is the lower part of PCI Express device status, non-
1552 		 * reserved bits are RW1C, writing 0 to them will not have
1553 		 * any effect at all.
1554 		 */
1555 #ifdef foo
1556 		uint8_t val;
1557 
1558 		val = pci_read_config(dev, 0x79, 1);
1559 		val = (val & ~0x70) | 0x50;
1560 		pci_write_config(dev, 0x54, val, 1);
1561 		pci_write_config(dev, 0x80, 0, 1);
1562 #endif
1563 	}
1564 
1565 	/*
1566 	 * Apply some PHY fixup from Realtek ...
1567 	 */
1568 	if (sc->re_hwrev == RE_HWREV_8110S) {
1569 		CSR_WRITE_1(sc, 0x82, 1);
1570 		re_miibus_writereg(dev, 1, 0xb, 0);
1571 	}
1572 	if (sc->re_caps & RE_C_PHYPMGT) {
1573 		/* Power up PHY */
1574 		re_miibus_writereg(dev, 1, 0x1f, 0);
1575 		re_miibus_writereg(dev, 1, 0xe, 0);
1576 	}
1577 
1578 	/* Do MII setup */
1579 	if (mii_phy_probe(dev, &sc->re_miibus,
1580 	    re_ifmedia_upd, re_ifmedia_sts)) {
1581 		device_printf(dev, "MII without any phy!\n");
1582 		error = ENXIO;
1583 		goto fail;
1584 	}
1585 
1586 	ifp = &sc->arpcom.ac_if;
1587 	ifp->if_softc = sc;
1588 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1589 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1590 	ifp->if_ioctl = re_ioctl;
1591 	ifp->if_start = re_start;
1592 #ifdef DEVICE_POLLING
1593 	ifp->if_poll = re_poll;
1594 #endif
1595 	ifp->if_watchdog = re_watchdog;
1596 	ifp->if_init = re_init;
1597 	if (!RE_IS_8139CP(sc)) /* XXX */
1598 		ifp->if_baudrate = 1000000000;
1599 	else
1600 		ifp->if_baudrate = 100000000;
1601 	ifq_set_maxlen(&ifp->if_snd, qlen);
1602 	ifq_set_ready(&ifp->if_snd);
1603 
1604 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1605 	if (sc->re_caps & RE_C_HWCSUM)
1606 		ifp->if_capabilities |= IFCAP_HWCSUM;
1607 
1608 	ifp->if_capenable = ifp->if_capabilities;
1609 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1610 		ifp->if_hwassist = RE_CSUM_FEATURES;
1611 	else
1612 		ifp->if_hwassist = 0;
1613 
1614 	/*
1615 	 * Call MI attach routine.
1616 	 */
1617 	ether_ifattach(ifp, eaddr, NULL);
1618 
1619 #ifdef RE_DIAG
1620 	/*
1621 	 * Perform hardware diagnostic on the original RTL8169.
1622 	 * Some 32-bit cards were incorrectly wired and would
1623 	 * malfunction if plugged into a 64-bit slot.
1624 	 */
1625 	if (sc->re_hwrev == RE_HWREV_8169) {
1626 		lwkt_serialize_enter(ifp->if_serializer);
1627 		error = re_diag(sc);
1628 		lwkt_serialize_exit(ifp->if_serializer);
1629 
1630 		if (error) {
1631 			device_printf(dev, "hardware diagnostic failure\n");
1632 			ether_ifdetach(ifp);
1633 			goto fail;
1634 		}
1635 	}
1636 #endif	/* RE_DIAG */
1637 
1638 	/* Hook interrupt last to avoid having to lock softc */
1639 	error = bus_setup_intr(dev, sc->re_irq, INTR_MPSAFE, re_intr, sc,
1640 			       &sc->re_intrhand, ifp->if_serializer);
1641 
1642 	if (error) {
1643 		device_printf(dev, "couldn't set up irq\n");
1644 		ether_ifdetach(ifp);
1645 		goto fail;
1646 	}
1647 
1648 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1649 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1650 
1651 fail:
1652 	if (error)
1653 		re_detach(dev);
1654 
1655 	return (error);
1656 }
1657 
1658 /*
1659  * Shutdown hardware and free up resources. This can be called any
1660  * time after the mutex has been initialized. It is called in both
1661  * the error case in attach and the normal detach case so it needs
1662  * to be careful about only freeing resources that have actually been
1663  * allocated.
1664  */
1665 static int
1666 re_detach(device_t dev)
1667 {
1668 	struct re_softc *sc = device_get_softc(dev);
1669 	struct ifnet *ifp = &sc->arpcom.ac_if;
1670 
1671 	/* These should only be active if attach succeeded */
1672 	if (device_is_attached(dev)) {
1673 		lwkt_serialize_enter(ifp->if_serializer);
1674 		re_stop(sc);
1675 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1676 		lwkt_serialize_exit(ifp->if_serializer);
1677 
1678 		ether_ifdetach(ifp);
1679 	}
1680 	if (sc->re_miibus)
1681 		device_delete_child(dev, sc->re_miibus);
1682 	bus_generic_detach(dev);
1683 
1684 	if (sc->re_sysctl_tree != NULL)
1685 		sysctl_ctx_free(&sc->re_sysctl_ctx);
1686 
1687 	if (sc->re_irq)
1688 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1689 	if (sc->re_res) {
1690 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1691 				     sc->re_res);
1692 	}
1693 
1694 	/* Free DMA stuffs */
1695 	re_freemem(dev);
1696 
1697 	return(0);
1698 }
1699 
1700 static void
1701 re_setup_rxdesc(struct re_softc *sc, int idx)
1702 {
1703 	bus_addr_t paddr;
1704 	uint32_t cmdstat;
1705 	struct re_desc *d;
1706 
1707 	paddr = sc->re_ldata.re_rx_paddr[idx];
1708 	d = &sc->re_ldata.re_rx_list[idx];
1709 
1710 	d->re_bufaddr_lo = htole32(RE_ADDR_LO(paddr));
1711 	d->re_bufaddr_hi = htole32(RE_ADDR_HI(paddr));
1712 
1713 	cmdstat = sc->re_rxbuf_size | RE_RDESC_CMD_OWN;
1714 	if (idx == (sc->re_rx_desc_cnt - 1))
1715 		cmdstat |= RE_RDESC_CMD_EOR;
1716 	d->re_cmdstat = htole32(cmdstat);
1717 }
1718 
1719 static int
1720 re_newbuf_std(struct re_softc *sc, int idx, int init)
1721 {
1722 	struct re_dmaload_arg arg;
1723 	bus_dma_segment_t seg;
1724 	bus_dmamap_t map;
1725 	struct mbuf *m;
1726 	int error;
1727 
1728 	m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
1729 	if (m == NULL) {
1730 		error = ENOBUFS;
1731 
1732 		if (init) {
1733 			if_printf(&sc->arpcom.ac_if, "m_getcl failed\n");
1734 			return error;
1735 		} else {
1736 			goto back;
1737 		}
1738 	}
1739 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1740 
1741 	/*
1742 	 * NOTE:
1743 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1744 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1745 	 */
1746 
1747 	arg.re_nsegs = 1;
1748 	arg.re_segs = &seg;
1749         error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag,
1750 				     sc->re_ldata.re_rx_spare, m,
1751 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1752 	if (error || arg.re_nsegs == 0) {
1753 		if (!error) {
1754 			if_printf(&sc->arpcom.ac_if, "too many segments?!\n");
1755 			bus_dmamap_unload(sc->re_ldata.re_mtag,
1756 					  sc->re_ldata.re_rx_spare);
1757 			error = EFBIG;
1758 		}
1759 		m_freem(m);
1760 
1761 		if (init) {
1762 			if_printf(&sc->arpcom.ac_if, "can't load RX mbuf\n");
1763 			return error;
1764 		} else {
1765 			goto back;
1766 		}
1767 	}
1768 
1769 	if (!init) {
1770 		bus_dmamap_sync(sc->re_ldata.re_mtag,
1771 				sc->re_ldata.re_rx_dmamap[idx],
1772 				BUS_DMASYNC_POSTREAD);
1773 		bus_dmamap_unload(sc->re_ldata.re_mtag,
1774 				  sc->re_ldata.re_rx_dmamap[idx]);
1775 	}
1776 	sc->re_ldata.re_rx_mbuf[idx] = m;
1777 	sc->re_ldata.re_rx_paddr[idx] = seg.ds_addr;
1778 
1779 	map = sc->re_ldata.re_rx_dmamap[idx];
1780 	sc->re_ldata.re_rx_dmamap[idx] = sc->re_ldata.re_rx_spare;
1781 	sc->re_ldata.re_rx_spare = map;
1782 back:
1783 	re_setup_rxdesc(sc, idx);
1784 	return error;
1785 }
1786 
1787 static int
1788 re_newbuf_jumbo(struct re_softc *sc, int idx, int init)
1789 {
1790 	struct mbuf *m;
1791 	struct re_jbuf *jbuf;
1792 	int error = 0;
1793 
1794 	MGETHDR(m, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1795 	if (m == NULL) {
1796 		error = ENOBUFS;
1797 		if (init) {
1798 			if_printf(&sc->arpcom.ac_if, "MGETHDR failed\n");
1799 			return error;
1800 		} else {
1801 			goto back;
1802 		}
1803 	}
1804 
1805 	jbuf = re_jbuf_alloc(sc);
1806 	if (jbuf == NULL) {
1807 		m_freem(m);
1808 
1809 		error = ENOBUFS;
1810 		if (init) {
1811 			if_printf(&sc->arpcom.ac_if, "jpool is empty\n");
1812 			return error;
1813 		} else {
1814 			goto back;
1815 		}
1816 	}
1817 
1818 	m->m_ext.ext_arg = jbuf;
1819 	m->m_ext.ext_buf = jbuf->re_buf;
1820 	m->m_ext.ext_free = re_jbuf_free;
1821 	m->m_ext.ext_ref = re_jbuf_ref;
1822 	m->m_ext.ext_size = sc->re_rxbuf_size;
1823 
1824 	m->m_data = m->m_ext.ext_buf;
1825 	m->m_flags |= M_EXT;
1826 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
1827 
1828 	/*
1829 	 * NOTE:
1830 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1831 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1832 	 */
1833 
1834 	sc->re_ldata.re_rx_mbuf[idx] = m;
1835 	sc->re_ldata.re_rx_paddr[idx] = jbuf->re_paddr;
1836 back:
1837 	re_setup_rxdesc(sc, idx);
1838 	return error;
1839 }
1840 
1841 static int
1842 re_tx_list_init(struct re_softc *sc)
1843 {
1844 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
1845 
1846 	/* Flush the TX descriptors */
1847 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1848 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1849 
1850 	sc->re_ldata.re_tx_prodidx = 0;
1851 	sc->re_ldata.re_tx_considx = 0;
1852 	sc->re_ldata.re_tx_free = sc->re_tx_desc_cnt;
1853 
1854 	return(0);
1855 }
1856 
1857 static int
1858 re_rx_list_init(struct re_softc *sc)
1859 {
1860 	int i, error;
1861 
1862 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc));
1863 
1864 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1865 		error = sc->re_newbuf(sc, i, 1);
1866 		if (error)
1867 			return(error);
1868 	}
1869 
1870 	/* Flush the RX descriptors */
1871 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1872 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1873 
1874 	sc->re_ldata.re_rx_prodidx = 0;
1875 	sc->re_head = sc->re_tail = NULL;
1876 
1877 	return(0);
1878 }
1879 
1880 #define RE_IP4_PACKET	0x1
1881 #define RE_TCP_PACKET	0x2
1882 #define RE_UDP_PACKET	0x4
1883 
1884 static __inline uint8_t
1885 re_packet_type(struct re_softc *sc, uint32_t rxstat, uint32_t rxctrl)
1886 {
1887 	uint8_t packet_type = 0;
1888 
1889 	if (sc->re_caps & RE_C_MAC2) {
1890 		if (rxctrl & RE_RDESC_CTL_PROTOIP4)
1891 			packet_type |= RE_IP4_PACKET;
1892 	} else {
1893 		if (rxstat & RE_RDESC_STAT_PROTOID)
1894 			packet_type |= RE_IP4_PACKET;
1895 	}
1896 	if (RE_TCPPKT(rxstat))
1897 		packet_type |= RE_TCP_PACKET;
1898 	else if (RE_UDPPKT(rxstat))
1899 		packet_type |= RE_UDP_PACKET;
1900 	return packet_type;
1901 }
1902 
1903 /*
1904  * RX handler for C+ and 8169. For the gigE chips, we support
1905  * the reception of jumbo frames that have been fragmented
1906  * across multiple 2K mbuf cluster buffers.
1907  */
1908 static int
1909 re_rxeof(struct re_softc *sc)
1910 {
1911 	struct ifnet *ifp = &sc->arpcom.ac_if;
1912 	struct mbuf *m;
1913 	struct re_desc 	*cur_rx;
1914 	uint32_t rxstat, rxctrl;
1915 	int i, total_len, rx = 0;
1916 	struct mbuf_chain chain[MAXCPU];
1917 
1918 	/* Invalidate the descriptor memory */
1919 
1920 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1921 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1922 
1923 	ether_input_chain_init(chain);
1924 
1925 	for (i = sc->re_ldata.re_rx_prodidx;
1926 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0; RE_RXDESC_INC(sc, i)) {
1927 		cur_rx = &sc->re_ldata.re_rx_list[i];
1928 		m = sc->re_ldata.re_rx_mbuf[i];
1929 		total_len = RE_RXBYTES(cur_rx);
1930 		rxstat = le32toh(cur_rx->re_cmdstat);
1931 		rxctrl = le32toh(cur_rx->re_control);
1932 
1933 		rx = 1;
1934 
1935 #ifdef INVARIANTS
1936 		if (sc->re_flags & RE_F_USE_JPOOL)
1937 			KKASSERT(rxstat & RE_RDESC_STAT_EOF);
1938 #endif
1939 
1940 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1941 			if (sc->re_flags & RE_F_DROP_RXFRAG) {
1942 				re_setup_rxdesc(sc, i);
1943 				continue;
1944 			}
1945 
1946 			if (sc->re_newbuf(sc, i, 0)) {
1947 				/* Drop upcoming fragments */
1948 				sc->re_flags |= RE_F_DROP_RXFRAG;
1949 				continue;
1950 			}
1951 
1952 			m->m_len = MCLBYTES;
1953 			if (sc->re_head == NULL) {
1954 				sc->re_head = sc->re_tail = m;
1955 			} else {
1956 				sc->re_tail->m_next = m;
1957 				sc->re_tail = m;
1958 			}
1959 			continue;
1960 		} else if (sc->re_flags & RE_F_DROP_RXFRAG) {
1961 			/*
1962 			 * Last fragment of a multi-fragment packet.
1963 			 *
1964 			 * Since error already happened, this fragment
1965 			 * must be dropped as well as the fragment chain.
1966 			 */
1967 			re_setup_rxdesc(sc, i);
1968 			re_free_rxchain(sc);
1969 			sc->re_flags &= ~RE_F_DROP_RXFRAG;
1970 			continue;
1971 		}
1972 
1973 		/*
1974 		 * NOTE: for the 8139C+, the frame length field
1975 		 * is always 12 bits in size, but for the gigE chips,
1976 		 * it is 13 bits (since the max RX frame length is 16K).
1977 		 * Unfortunately, all 32 bits in the status word
1978 		 * were already used, so to make room for the extra
1979 		 * length bit, RealTek took out the 'frame alignment
1980 		 * error' bit and shifted the other status bits
1981 		 * over one slot. The OWN, EOR, FS and LS bits are
1982 		 * still in the same places. We have already extracted
1983 		 * the frame length and checked the OWN bit, so rather
1984 		 * than using an alternate bit mapping, we shift the
1985 		 * status bits one space to the right so we can evaluate
1986 		 * them using the 8169 status as though it was in the
1987 		 * same format as that of the 8139C+.
1988 		 */
1989 		if (!RE_IS_8139CP(sc))
1990 			rxstat >>= 1;
1991 
1992 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
1993 			ifp->if_ierrors++;
1994 			/*
1995 			 * If this is part of a multi-fragment packet,
1996 			 * discard all the pieces.
1997 			 */
1998 			re_free_rxchain(sc);
1999 			re_setup_rxdesc(sc, i);
2000 			continue;
2001 		}
2002 
2003 		/*
2004 		 * If allocating a replacement mbuf fails,
2005 		 * reload the current one.
2006 		 */
2007 
2008 		if (sc->re_newbuf(sc, i, 0)) {
2009 			ifp->if_ierrors++;
2010 			continue;
2011 		}
2012 
2013 		if (sc->re_head != NULL) {
2014 			m->m_len = total_len % MCLBYTES;
2015 			/*
2016 			 * Special case: if there's 4 bytes or less
2017 			 * in this buffer, the mbuf can be discarded:
2018 			 * the last 4 bytes is the CRC, which we don't
2019 			 * care about anyway.
2020 			 */
2021 			if (m->m_len <= ETHER_CRC_LEN) {
2022 				sc->re_tail->m_len -=
2023 				    (ETHER_CRC_LEN - m->m_len);
2024 				m_freem(m);
2025 			} else {
2026 				m->m_len -= ETHER_CRC_LEN;
2027 				sc->re_tail->m_next = m;
2028 			}
2029 			m = sc->re_head;
2030 			sc->re_head = sc->re_tail = NULL;
2031 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2032 		} else {
2033 			m->m_pkthdr.len = m->m_len =
2034 			    (total_len - ETHER_CRC_LEN);
2035 		}
2036 
2037 		ifp->if_ipackets++;
2038 		m->m_pkthdr.rcvif = ifp;
2039 
2040 		/* Do RX checksumming if enabled */
2041 
2042 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2043 			uint8_t packet_type;
2044 
2045 			packet_type = re_packet_type(sc, rxstat, rxctrl);
2046 
2047 			/* Check IP header checksum */
2048 			if (packet_type & RE_IP4_PACKET) {
2049 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2050 				if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
2051 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2052 			}
2053 
2054 			/* Check TCP/UDP checksum */
2055 			if (((packet_type & RE_TCP_PACKET) &&
2056 			     (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
2057 			    ((packet_type & RE_UDP_PACKET) &&
2058 			     (rxstat & RE_RDESC_STAT_UDPSUMBAD) == 0)) {
2059 				m->m_pkthdr.csum_flags |=
2060 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
2061 				    CSUM_FRAG_NOT_CHECKED;
2062 				m->m_pkthdr.csum_data = 0xffff;
2063 			}
2064 		}
2065 
2066 		if (rxctrl & RE_RDESC_CTL_HASTAG) {
2067 			m->m_flags |= M_VLANTAG;
2068 			m->m_pkthdr.ether_vlantag =
2069 				be16toh((rxctrl & RE_RDESC_CTL_TAGDATA));
2070 		}
2071 		ether_input_chain(ifp, m, chain);
2072 	}
2073 
2074 	ether_input_dispatch(chain);
2075 
2076 	/* Flush the RX DMA ring */
2077 
2078 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
2079 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
2080 
2081 	sc->re_ldata.re_rx_prodidx = i;
2082 
2083 	return rx;
2084 }
2085 
2086 #undef RE_IP4_PACKET
2087 #undef RE_TCP_PACKET
2088 #undef RE_UDP_PACKET
2089 
2090 static int
2091 re_tx_collect(struct re_softc *sc)
2092 {
2093 	struct ifnet *ifp = &sc->arpcom.ac_if;
2094 	uint32_t txstat;
2095 	int idx, tx = 0;
2096 
2097 	/* Invalidate the TX descriptor list */
2098 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2099 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
2100 
2101 	for (idx = sc->re_ldata.re_tx_considx;
2102 	     sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt;
2103 	     RE_TXDESC_INC(sc, idx)) {
2104 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
2105 		if (txstat & RE_TDESC_CMD_OWN)
2106 			break;
2107 
2108 		tx = 1;
2109 
2110 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
2111 
2112 		/*
2113 		 * We only stash mbufs in the last descriptor
2114 		 * in a fragment chain, which also happens to
2115 		 * be the only place where the TX status bits
2116 		 * are valid.
2117 		 */
2118 		if (txstat & RE_TDESC_CMD_EOF) {
2119 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
2120 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
2121 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2122 			    sc->re_ldata.re_tx_dmamap[idx]);
2123 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
2124 			    RE_TDESC_STAT_COLCNT))
2125 				ifp->if_collisions++;
2126 			if (txstat & RE_TDESC_STAT_TXERRSUM)
2127 				ifp->if_oerrors++;
2128 			else
2129 				ifp->if_opackets++;
2130 		}
2131 		sc->re_ldata.re_tx_free++;
2132 	}
2133 	sc->re_ldata.re_tx_considx = idx;
2134 
2135 	return tx;
2136 }
2137 
2138 static int
2139 re_txeof(struct re_softc *sc)
2140 {
2141 	struct ifnet *ifp = &sc->arpcom.ac_if;
2142 	int tx;
2143 
2144 	tx = re_tx_collect(sc);
2145 
2146 	/* There is enough free TX descs */
2147 	if (sc->re_ldata.re_tx_free > RE_TXDESC_SPARE)
2148 		ifp->if_flags &= ~IFF_OACTIVE;
2149 
2150 	/*
2151 	 * Some chips will ignore a second TX request issued while an
2152 	 * existing transmission is in progress. If the transmitter goes
2153 	 * idle but there are still packets waiting to be sent, we need
2154 	 * to restart the channel here to flush them out. This only seems
2155 	 * to be required with the PCIe devices.
2156 	 */
2157 	if (sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt)
2158 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2159 	else
2160 		ifp->if_timer = 0;
2161 
2162 	return tx;
2163 }
2164 
2165 static void
2166 re_tick(void *xsc)
2167 {
2168 	struct re_softc *sc = xsc;
2169 
2170 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
2171 	re_tick_serialized(xsc);
2172 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
2173 }
2174 
2175 static void
2176 re_tick_serialized(void *xsc)
2177 {
2178 	struct re_softc *sc = xsc;
2179 	struct ifnet *ifp = &sc->arpcom.ac_if;
2180 	struct mii_data *mii;
2181 
2182 	ASSERT_SERIALIZED(ifp->if_serializer);
2183 
2184 	mii = device_get_softc(sc->re_miibus);
2185 	mii_tick(mii);
2186 	if (sc->re_flags & RE_F_LINKED) {
2187 		if (!(mii->mii_media_status & IFM_ACTIVE))
2188 			sc->re_flags &= ~RE_F_LINKED;
2189 	} else {
2190 		if (mii->mii_media_status & IFM_ACTIVE &&
2191 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2192 			sc->re_flags |= RE_F_LINKED;
2193 			if (!ifq_is_empty(&ifp->if_snd))
2194 				if_devstart(ifp);
2195 		}
2196 	}
2197 
2198 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2199 }
2200 
2201 #ifdef DEVICE_POLLING
2202 
2203 static void
2204 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2205 {
2206 	struct re_softc *sc = ifp->if_softc;
2207 
2208 	ASSERT_SERIALIZED(ifp->if_serializer);
2209 
2210 	switch(cmd) {
2211 	case POLL_REGISTER:
2212 		/* disable interrupts */
2213 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2214 		break;
2215 
2216 	case POLL_DEREGISTER:
2217 		/* enable interrupts */
2218 		re_setup_intr(sc, 1, sc->re_imtype);
2219 		break;
2220 
2221 	default:
2222 		sc->rxcycles = count;
2223 		re_rxeof(sc);
2224 		re_txeof(sc);
2225 
2226 		if (!ifq_is_empty(&ifp->if_snd))
2227 			if_devstart(ifp);
2228 
2229 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2230 			uint16_t       status;
2231 
2232 			status = CSR_READ_2(sc, RE_ISR);
2233 			if (status == 0xffff)
2234 				return;
2235 			if (status)
2236 				CSR_WRITE_2(sc, RE_ISR, status);
2237 
2238 			/*
2239 			 * XXX check behaviour on receiver stalls.
2240 			 */
2241 
2242 			if (status & RE_ISR_SYSTEM_ERR)
2243 				re_init(sc);
2244 		}
2245 		break;
2246 	}
2247 }
2248 #endif /* DEVICE_POLLING */
2249 
2250 static void
2251 re_intr(void *arg)
2252 {
2253 	struct re_softc	*sc = arg;
2254 	struct ifnet *ifp = &sc->arpcom.ac_if;
2255 	uint16_t status;
2256 	int rx, tx;
2257 
2258 	ASSERT_SERIALIZED(ifp->if_serializer);
2259 
2260 	if ((sc->re_flags & RE_F_SUSPENDED) ||
2261 	    (ifp->if_flags & IFF_RUNNING) == 0)
2262 		return;
2263 
2264 	rx = tx = 0;
2265 	for (;;) {
2266 		status = CSR_READ_2(sc, RE_ISR);
2267 		/* If the card has gone away the read returns 0xffff. */
2268 		if (status == 0xffff)
2269 			break;
2270 		if (status)
2271 			CSR_WRITE_2(sc, RE_ISR, status);
2272 
2273 		if ((status & sc->re_intrs) == 0)
2274 			break;
2275 
2276 		if (status & (sc->re_rx_ack | RE_ISR_RX_ERR))
2277 			rx |= re_rxeof(sc);
2278 
2279 		if (status & (sc->re_tx_ack | RE_ISR_TX_ERR))
2280 			tx |= re_txeof(sc);
2281 
2282 		if (status & RE_ISR_SYSTEM_ERR)
2283 			re_init(sc);
2284 
2285 		if (status & RE_ISR_LINKCHG) {
2286 			callout_stop(&sc->re_timer);
2287 			re_tick_serialized(sc);
2288 		}
2289 	}
2290 
2291 	if (sc->re_imtype == RE_IMTYPE_SIM) {
2292 		if ((sc->re_flags & RE_F_TIMER_INTR)) {
2293 			if ((tx | rx) == 0) {
2294 				/*
2295 				 * Nothing needs to be processed, fallback
2296 				 * to use TX/RX interrupts.
2297 				 */
2298 				re_setup_intr(sc, 1, RE_IMTYPE_NONE);
2299 
2300 				/*
2301 				 * Recollect, mainly to avoid the possible
2302 				 * race introduced by changing interrupt
2303 				 * masks.
2304 				 */
2305 				re_rxeof(sc);
2306 				tx = re_txeof(sc);
2307 			} else {
2308 				CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
2309 			}
2310 		} else if (tx | rx) {
2311 			/*
2312 			 * Assume that using simulated interrupt moderation
2313 			 * (hardware timer based) could reduce the interript
2314 			 * rate.
2315 			 */
2316 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
2317 		}
2318 	}
2319 
2320 	if (tx && !ifq_is_empty(&ifp->if_snd))
2321 		if_devstart(ifp);
2322 }
2323 
2324 static int
2325 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx0)
2326 {
2327 	struct ifnet *ifp = &sc->arpcom.ac_if;
2328 	struct mbuf *m;
2329 	struct re_dmaload_arg arg;
2330 	bus_dma_segment_t segs[RE_MAXSEGS];
2331 	bus_dmamap_t map;
2332 	int error, maxsegs, idx, i;
2333 	struct re_desc *d, *tx_ring;
2334 	uint32_t cmd_csum, ctl_csum, vlantag;
2335 
2336 	KASSERT(sc->re_ldata.re_tx_free > RE_TXDESC_SPARE,
2337 		("not enough free TX desc\n"));
2338 
2339 	m = *m_head;
2340 	map = sc->re_ldata.re_tx_dmamap[*idx0];
2341 
2342 	/*
2343 	 * Set up checksum offload. Note: checksum offload bits must
2344 	 * appear in all descriptors of a multi-descriptor transmit
2345 	 * attempt. (This is according to testing done with an 8169
2346 	 * chip. I'm not sure if this is a requirement or a bug.)
2347 	 */
2348 	cmd_csum = ctl_csum = 0;
2349 	if (m->m_pkthdr.csum_flags & CSUM_IP) {
2350 		cmd_csum |= RE_TDESC_CMD_IPCSUM;
2351 		ctl_csum |= RE_TDESC_CTL_IPCSUM;
2352 	}
2353 	if (m->m_pkthdr.csum_flags & CSUM_TCP) {
2354 		cmd_csum |= RE_TDESC_CMD_TCPCSUM;
2355 		ctl_csum |= RE_TDESC_CTL_TCPCSUM;
2356 	}
2357 	if (m->m_pkthdr.csum_flags & CSUM_UDP) {
2358 		cmd_csum |= RE_TDESC_CMD_UDPCSUM;
2359 		ctl_csum |= RE_TDESC_CTL_UDPCSUM;
2360 	}
2361 
2362 	/* For MAC2 chips, csum flags are set on re_control */
2363 	if (sc->re_caps & RE_C_MAC2)
2364 		cmd_csum = 0;
2365 	else
2366 		ctl_csum = 0;
2367 
2368 	if (m->m_pkthdr.len > sc->re_swcsum_lim &&
2369 	    (m->m_pkthdr.csum_flags & (CSUM_DELAY_IP | CSUM_DELAY_DATA))) {
2370 		struct ether_header *eh;
2371 		struct ip *ip;
2372 		u_short offset;
2373 
2374 		m = m_pullup(m, sizeof(struct ether_header *));
2375 		if (m == NULL) {
2376 			*m_head = NULL;
2377 			return ENOBUFS;
2378 		}
2379 		eh = mtod(m, struct ether_header *);
2380 
2381 		/* XXX */
2382 		if (eh->ether_type == ETHERTYPE_VLAN)
2383 			offset = sizeof(struct ether_vlan_header);
2384 		else
2385 			offset = sizeof(struct ether_header);
2386 
2387 		m = m_pullup(m, offset + sizeof(struct ip *));
2388 		if (m == NULL) {
2389 			*m_head = NULL;
2390 			return ENOBUFS;
2391 		}
2392 		ip = (struct ip *)(mtod(m, uint8_t *) + offset);
2393 
2394 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2395 			u_short csum;
2396 
2397 			offset += IP_VHL_HL(ip->ip_vhl) << 2;
2398 			csum = in_cksum_skip(m, ntohs(ip->ip_len), offset);
2399 			if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
2400 				csum = 0xffff;
2401 			offset += m->m_pkthdr.csum_data;        /* checksum offset */
2402 			*(u_short *)(m->m_data + offset) = csum;
2403 
2404 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2405 		}
2406 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_IP) {
2407 			ip->ip_sum = 0;
2408 			if (ip->ip_vhl == IP_VHL_BORING) {
2409 				ip->ip_sum = in_cksum_hdr(ip);
2410 			} else {
2411 				ip->ip_sum =
2412 				in_cksum(m, IP_VHL_HL(ip->ip_vhl) << 2);
2413 			}
2414 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
2415 		}
2416 		*m_head = m; /* 'm' may be changed by above two m_pullup() */
2417 
2418 		/* Clear hardware CSUM flags */
2419 		cmd_csum = ctl_csum = 0;
2420 	}
2421 
2422 	if ((sc->re_caps & RE_C_AUTOPAD) == 0) {
2423 		/*
2424 		 * With some of the RealTek chips, using the checksum offload
2425 		 * support in conjunction with the autopadding feature results
2426 		 * in the transmission of corrupt frames. For example, if we
2427 		 * need to send a really small IP fragment that's less than 60
2428 		 * bytes in size, and IP header checksumming is enabled, the
2429 		 * resulting ethernet frame that appears on the wire will
2430 		 * have garbled payload. To work around this, if TX checksum
2431 		 * offload is enabled, we always manually pad short frames out
2432 		 * to the minimum ethernet frame size.
2433 		 *
2434 		 * Note: this appears unnecessary for TCP, and doing it for TCP
2435 		 * with PCIe adapters seems to result in bad checksums.
2436 		 */
2437 		if ((m->m_pkthdr.csum_flags &
2438 		     (CSUM_DELAY_IP | CSUM_DELAY_DATA)) &&
2439 		    (m->m_pkthdr.csum_flags & CSUM_TCP) == 0 &&
2440 		    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
2441 			error = m_devpad(m, RE_MIN_FRAMELEN);
2442 			if (error)
2443 				goto back;
2444 		}
2445 	}
2446 
2447 	vlantag = 0;
2448 	if (m->m_flags & M_VLANTAG) {
2449 		vlantag = htobe16(m->m_pkthdr.ether_vlantag) |
2450 			  RE_TDESC_CTL_INSTAG;
2451 	}
2452 
2453 	maxsegs = sc->re_ldata.re_tx_free;
2454 	if (maxsegs > RE_MAXSEGS)
2455 		maxsegs = RE_MAXSEGS;
2456 
2457 	arg.re_nsegs = maxsegs;
2458 	arg.re_segs = segs;
2459 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2460 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2461 	if (error && error != EFBIG) {
2462 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
2463 		goto back;
2464 	}
2465 
2466 	/*
2467 	 * Too many segments to map, coalesce into a single mbuf
2468 	 */
2469 	if (!error && arg.re_nsegs == 0) {
2470 		bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2471 		error = EFBIG;
2472 	}
2473 	if (error) {
2474 		struct mbuf *m_new;
2475 
2476 		m_new = m_defrag(m, MB_DONTWAIT);
2477 		if (m_new == NULL) {
2478 			if_printf(ifp, "can't defrag TX mbuf\n");
2479 			error = ENOBUFS;
2480 			goto back;
2481 		} else {
2482 			*m_head = m = m_new;
2483 		}
2484 
2485 		arg.re_nsegs = maxsegs;
2486 		arg.re_segs = segs;
2487 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_mtag, map, m,
2488 					     re_dma_map_desc, &arg,
2489 					     BUS_DMA_NOWAIT);
2490 		if (error || arg.re_nsegs == 0) {
2491 			if (!error) {
2492 				bus_dmamap_unload(sc->re_ldata.re_mtag, map);
2493 				error = EFBIG;
2494 			}
2495 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
2496 			goto back;
2497 		}
2498 	}
2499 	bus_dmamap_sync(sc->re_ldata.re_mtag, map, BUS_DMASYNC_PREWRITE);
2500 
2501 	/*
2502 	 * Map the segment array into descriptors.  We also keep track
2503 	 * of the end of the ring and set the end-of-ring bits as needed,
2504 	 * and we set the ownership bits in all except the very first
2505 	 * descriptor, whose ownership bits will be turned on later.
2506 	 */
2507 	tx_ring = sc->re_ldata.re_tx_list;
2508 	idx = *idx0;
2509 	i = 0;
2510 	for (;;) {
2511 		uint32_t cmdstat;
2512 
2513 		d = &tx_ring[idx];
2514 
2515 		cmdstat = segs[i].ds_len;
2516 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
2517 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
2518 		if (i == 0)
2519 			cmdstat |= RE_TDESC_CMD_SOF;
2520 		else
2521 			cmdstat |= RE_TDESC_CMD_OWN;
2522 		if (idx == (sc->re_tx_desc_cnt - 1))
2523 			cmdstat |= RE_TDESC_CMD_EOR;
2524 		d->re_cmdstat = htole32(cmdstat | cmd_csum);
2525 		d->re_control = htole32(ctl_csum | vlantag);
2526 
2527 		i++;
2528 		if (i == arg.re_nsegs)
2529 			break;
2530 		RE_TXDESC_INC(sc, idx);
2531 	}
2532 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
2533 
2534 	/* Transfer ownership of packet to the chip. */
2535 	d->re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2536 	if (*idx0 != idx)
2537 		tx_ring[*idx0].re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2538 
2539 	/*
2540 	 * Insure that the map for this transmission
2541 	 * is placed at the array index of the last descriptor
2542 	 * in this chain.
2543 	 */
2544 	sc->re_ldata.re_tx_dmamap[*idx0] = sc->re_ldata.re_tx_dmamap[idx];
2545 	sc->re_ldata.re_tx_dmamap[idx] = map;
2546 
2547 	sc->re_ldata.re_tx_mbuf[idx] = m;
2548 	sc->re_ldata.re_tx_free -= arg.re_nsegs;
2549 
2550 	RE_TXDESC_INC(sc, idx);
2551 	*idx0 = idx;
2552 back:
2553 	if (error) {
2554 		m_freem(m);
2555 		*m_head = NULL;
2556 	}
2557 	return error;
2558 }
2559 
2560 /*
2561  * Main transmit routine for C+ and gigE NICs.
2562  */
2563 
2564 static void
2565 re_start(struct ifnet *ifp)
2566 {
2567 	struct re_softc	*sc = ifp->if_softc;
2568 	struct mbuf *m_head;
2569 	int idx, need_trans, oactive, error;
2570 
2571 	ASSERT_SERIALIZED(ifp->if_serializer);
2572 
2573 	if ((sc->re_flags & RE_F_LINKED) == 0) {
2574 		ifq_purge(&ifp->if_snd);
2575 		return;
2576 	}
2577 
2578 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2579 		return;
2580 
2581 	idx = sc->re_ldata.re_tx_prodidx;
2582 
2583 	need_trans = 0;
2584 	oactive = 0;
2585 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2586 		if (sc->re_ldata.re_tx_free <= RE_TXDESC_SPARE) {
2587 			if (!oactive) {
2588 				if (re_tx_collect(sc)) {
2589 					oactive = 1;
2590 					continue;
2591 				}
2592 			}
2593 			ifp->if_flags |= IFF_OACTIVE;
2594 			break;
2595 		}
2596 
2597 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2598 		if (m_head == NULL)
2599 			break;
2600 
2601 		error = re_encap(sc, &m_head, &idx);
2602 		if (error) {
2603 			/* m_head is freed by re_encap(), if we reach here */
2604 			ifp->if_oerrors++;
2605 
2606 			if (error == EFBIG && !oactive) {
2607 				if (re_tx_collect(sc)) {
2608 					oactive = 1;
2609 					continue;
2610 				}
2611 			}
2612 			ifp->if_flags |= IFF_OACTIVE;
2613 			break;
2614 		}
2615 
2616 		oactive = 0;
2617 		need_trans = 1;
2618 
2619 		/*
2620 		 * If there's a BPF listener, bounce a copy of this frame
2621 		 * to him.
2622 		 */
2623 		ETHER_BPF_MTAP(ifp, m_head);
2624 	}
2625 
2626 	if (!need_trans)
2627 		return;
2628 
2629 	/* Flush the TX descriptors */
2630 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2631 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2632 
2633 	sc->re_ldata.re_tx_prodidx = idx;
2634 
2635 	/*
2636 	 * RealTek put the TX poll request register in a different
2637 	 * location on the 8169 gigE chip. I don't know why.
2638 	 */
2639 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2640 
2641 	/*
2642 	 * Set a timeout in case the chip goes out to lunch.
2643 	 */
2644 	ifp->if_timer = 5;
2645 }
2646 
2647 static void
2648 re_init(void *xsc)
2649 {
2650 	struct re_softc *sc = xsc;
2651 	struct ifnet *ifp = &sc->arpcom.ac_if;
2652 	struct mii_data *mii;
2653 	int error, framelen;
2654 
2655 	ASSERT_SERIALIZED(ifp->if_serializer);
2656 
2657 	mii = device_get_softc(sc->re_miibus);
2658 
2659 	/*
2660 	 * Cancel pending I/O and free all RX/TX buffers.
2661 	 */
2662 	re_stop(sc);
2663 
2664 	if (sc->re_caps & RE_C_CONTIGRX) {
2665 		if (ifp->if_mtu > ETHERMTU) {
2666 			KKASSERT(sc->re_ldata.re_jbuf != NULL);
2667 			sc->re_flags |= RE_F_USE_JPOOL;
2668 			sc->re_rxbuf_size = RE_FRAMELEN_MAX;
2669 			sc->re_newbuf = re_newbuf_jumbo;
2670 		} else {
2671 			sc->re_flags &= ~RE_F_USE_JPOOL;
2672 			sc->re_rxbuf_size = MCLBYTES;
2673 			sc->re_newbuf = re_newbuf_std;
2674 		}
2675 	}
2676 
2677 	/*
2678 	 * Adjust max read request size according to MTU.
2679 	 * Mainly to improve TX performance for common case (ETHERMTU).
2680 	 */
2681 	if (sc->re_caps & RE_C_PCIE) {
2682 		if (ifp->if_mtu > ETHERMTU) {
2683 			/*
2684 			 * 512 seems to be the only value that works
2685 			 * reliably with jumbo frame
2686 			 */
2687 			pcie_set_max_readrq(sc->re_dev,
2688 				PCIEM_DEVCTL_MAX_READRQ_512);
2689 		} else {
2690 			pcie_set_max_readrq(sc->re_dev,
2691 				PCIEM_DEVCTL_MAX_READRQ_4096);
2692 		}
2693 	}
2694 
2695 	/*
2696 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2697 	 * RX checksum offload. We must configure the C+ register
2698 	 * before all others.
2699 	 */
2700 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2701 		    RE_CPLUSCMD_PCI_MRW |
2702 		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING ?
2703 		     RE_CPLUSCMD_VLANSTRIP : 0) |
2704 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2705 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2706 
2707 	/*
2708 	 * Init our MAC address.  Even though the chipset
2709 	 * documentation doesn't mention it, we need to enter "Config
2710 	 * register write enable" mode to modify the ID registers.
2711 	 */
2712 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2713 	CSR_WRITE_4(sc, RE_IDR0,
2714 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2715 	CSR_WRITE_2(sc, RE_IDR4,
2716 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2717 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2718 
2719 	/*
2720 	 * For C+ mode, initialize the RX descriptors and mbufs.
2721 	 */
2722 	error = re_rx_list_init(sc);
2723 	if (error) {
2724 		re_stop(sc);
2725 		return;
2726 	}
2727 	error = re_tx_list_init(sc);
2728 	if (error) {
2729 		re_stop(sc);
2730 		return;
2731 	}
2732 
2733 	/*
2734 	 * Load the addresses of the RX and TX lists into the chip.
2735 	 */
2736 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2737 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2738 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2739 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2740 
2741 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2742 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2743 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2744 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2745 
2746 	/*
2747 	 * Enable transmit and receive.
2748 	 */
2749 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2750 
2751 	/*
2752 	 * Set the initial TX and RX configuration.
2753 	 */
2754 	if (sc->re_flags & RE_F_TESTMODE) {
2755 		if (!RE_IS_8139CP(sc))
2756 			CSR_WRITE_4(sc, RE_TXCFG,
2757 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2758 		else
2759 			CSR_WRITE_4(sc, RE_TXCFG,
2760 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2761 	} else
2762 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2763 
2764 	framelen = RE_FRAMELEN(ifp->if_mtu);
2765 	if (framelen < MCLBYTES)
2766 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(MCLBYTES, 128));
2767 	else
2768 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(framelen, 128));
2769 
2770 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2771 
2772 	/*
2773 	 * Program the multicast filter, if necessary.
2774 	 */
2775 	re_setmulti(sc);
2776 
2777 #ifdef DEVICE_POLLING
2778 	/*
2779 	 * Disable interrupts if we are polling.
2780 	 */
2781 	if (ifp->if_flags & IFF_POLLING)
2782 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2783 	else	/* otherwise ... */
2784 #endif /* DEVICE_POLLING */
2785 	/*
2786 	 * Enable interrupts.
2787 	 */
2788 	if (sc->re_flags & RE_F_TESTMODE)
2789 		CSR_WRITE_2(sc, RE_IMR, 0);
2790 	else
2791 		re_setup_intr(sc, 1, sc->re_imtype);
2792 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2793 
2794 	/* Start RX/TX process. */
2795 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2796 
2797 #ifdef notdef
2798 	/* Enable receiver and transmitter. */
2799 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2800 #endif
2801 
2802 	/*
2803 	 * For 8169 gigE NICs, set the max allowed RX packet
2804 	 * size so we can receive jumbo frames.
2805 	 */
2806 	if (!RE_IS_8139CP(sc)) {
2807 		if (sc->re_caps & RE_C_CONTIGRX)
2808 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, sc->re_rxbuf_size);
2809 		else
2810 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2811 	}
2812 
2813 	if (sc->re_flags & RE_F_TESTMODE)
2814 		return;
2815 
2816 	mii_mediachg(mii);
2817 
2818 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2819 
2820 	ifp->if_flags |= IFF_RUNNING;
2821 	ifp->if_flags &= ~IFF_OACTIVE;
2822 
2823 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2824 }
2825 
2826 /*
2827  * Set media options.
2828  */
2829 static int
2830 re_ifmedia_upd(struct ifnet *ifp)
2831 {
2832 	struct re_softc *sc = ifp->if_softc;
2833 	struct mii_data *mii;
2834 
2835 	ASSERT_SERIALIZED(ifp->if_serializer);
2836 
2837 	mii = device_get_softc(sc->re_miibus);
2838 	mii_mediachg(mii);
2839 
2840 	return(0);
2841 }
2842 
2843 /*
2844  * Report current media status.
2845  */
2846 static void
2847 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2848 {
2849 	struct re_softc *sc = ifp->if_softc;
2850 	struct mii_data *mii;
2851 
2852 	ASSERT_SERIALIZED(ifp->if_serializer);
2853 
2854 	mii = device_get_softc(sc->re_miibus);
2855 
2856 	mii_pollstat(mii);
2857 	ifmr->ifm_active = mii->mii_media_active;
2858 	ifmr->ifm_status = mii->mii_media_status;
2859 }
2860 
2861 static int
2862 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2863 {
2864 	struct re_softc *sc = ifp->if_softc;
2865 	struct ifreq *ifr = (struct ifreq *) data;
2866 	struct mii_data *mii;
2867 	int error = 0, mask;
2868 
2869 	ASSERT_SERIALIZED(ifp->if_serializer);
2870 
2871 	switch(command) {
2872 	case SIOCSIFMTU:
2873 		if (ifr->ifr_mtu > sc->re_maxmtu) {
2874 			error = EINVAL;
2875 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
2876 			ifp->if_mtu = ifr->ifr_mtu;
2877 			if (ifp->if_flags & IFF_RUNNING)
2878 				ifp->if_init(sc);
2879 		}
2880 		break;
2881 
2882 	case SIOCSIFFLAGS:
2883 		if (ifp->if_flags & IFF_UP) {
2884 			if (ifp->if_flags & IFF_RUNNING) {
2885 				if ((ifp->if_flags ^ sc->re_if_flags) &
2886 				    (IFF_PROMISC | IFF_ALLMULTI))
2887 					re_setmulti(sc);
2888 			} else {
2889 				re_init(sc);
2890 			}
2891 		} else if (ifp->if_flags & IFF_RUNNING) {
2892 			re_stop(sc);
2893 		}
2894 		sc->re_if_flags = ifp->if_flags;
2895 		break;
2896 
2897 	case SIOCADDMULTI:
2898 	case SIOCDELMULTI:
2899 		re_setmulti(sc);
2900 		break;
2901 
2902 	case SIOCGIFMEDIA:
2903 	case SIOCSIFMEDIA:
2904 		mii = device_get_softc(sc->re_miibus);
2905 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2906 		break;
2907 
2908 	case SIOCSIFCAP:
2909 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) &
2910 		       ifp->if_capabilities;
2911 		ifp->if_capenable ^= mask;
2912 
2913 		if (mask & IFCAP_HWCSUM) {
2914 			if (ifp->if_capenable & IFCAP_TXCSUM)
2915 				ifp->if_hwassist = RE_CSUM_FEATURES;
2916 			else
2917 				ifp->if_hwassist = 0;
2918 		}
2919 		if (mask && (ifp->if_flags & IFF_RUNNING))
2920 			re_init(sc);
2921 		break;
2922 
2923 	default:
2924 		error = ether_ioctl(ifp, command, data);
2925 		break;
2926 	}
2927 	return(error);
2928 }
2929 
2930 static void
2931 re_watchdog(struct ifnet *ifp)
2932 {
2933 	struct re_softc *sc = ifp->if_softc;
2934 
2935 	ASSERT_SERIALIZED(ifp->if_serializer);
2936 
2937 	if_printf(ifp, "watchdog timeout\n");
2938 
2939 	ifp->if_oerrors++;
2940 
2941 	re_txeof(sc);
2942 	re_rxeof(sc);
2943 
2944 	re_init(sc);
2945 
2946 	if (!ifq_is_empty(&ifp->if_snd))
2947 		if_devstart(ifp);
2948 }
2949 
2950 /*
2951  * Stop the adapter and free any mbufs allocated to the
2952  * RX and TX lists.
2953  */
2954 static void
2955 re_stop(struct re_softc *sc)
2956 {
2957 	struct ifnet *ifp = &sc->arpcom.ac_if;
2958 	int i;
2959 
2960 	ASSERT_SERIALIZED(ifp->if_serializer);
2961 
2962 	/* Reset the adapter. */
2963 	re_reset(sc, ifp->if_flags & IFF_RUNNING);
2964 
2965 	ifp->if_timer = 0;
2966 	callout_stop(&sc->re_timer);
2967 
2968 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2969 	sc->re_flags &= ~(RE_F_TIMER_INTR | RE_F_DROP_RXFRAG | RE_F_LINKED);
2970 
2971 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2972 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2973 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2974 
2975 	re_free_rxchain(sc);
2976 
2977 	/* Free the TX list buffers. */
2978 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
2979 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2980 			bus_dmamap_unload(sc->re_ldata.re_mtag,
2981 					  sc->re_ldata.re_tx_dmamap[i]);
2982 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2983 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2984 		}
2985 	}
2986 
2987 	/* Free the RX list buffers. */
2988 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
2989 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2990 			if ((sc->re_flags & RE_F_USE_JPOOL) == 0) {
2991 				bus_dmamap_unload(sc->re_ldata.re_mtag,
2992 						  sc->re_ldata.re_rx_dmamap[i]);
2993 			}
2994 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2995 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2996 		}
2997 	}
2998 }
2999 
3000 /*
3001  * Device suspend routine.  Stop the interface and save some PCI
3002  * settings in case the BIOS doesn't restore them properly on
3003  * resume.
3004  */
3005 static int
3006 re_suspend(device_t dev)
3007 {
3008 #ifndef BURN_BRIDGES
3009 	int i;
3010 #endif
3011 	struct re_softc *sc = device_get_softc(dev);
3012 	struct ifnet *ifp = &sc->arpcom.ac_if;
3013 
3014 	lwkt_serialize_enter(ifp->if_serializer);
3015 
3016 	re_stop(sc);
3017 
3018 #ifndef BURN_BRIDGES
3019 	for (i = 0; i < 5; i++)
3020 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
3021 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
3022 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
3023 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3024 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3025 #endif
3026 
3027 	sc->re_flags |= RE_F_SUSPENDED;
3028 
3029 	lwkt_serialize_exit(ifp->if_serializer);
3030 
3031 	return (0);
3032 }
3033 
3034 /*
3035  * Device resume routine.  Restore some PCI settings in case the BIOS
3036  * doesn't, re-enable busmastering, and restart the interface if
3037  * appropriate.
3038  */
3039 static int
3040 re_resume(device_t dev)
3041 {
3042 	struct re_softc *sc = device_get_softc(dev);
3043 	struct ifnet *ifp = &sc->arpcom.ac_if;
3044 #ifndef BURN_BRIDGES
3045 	int i;
3046 #endif
3047 
3048 	lwkt_serialize_enter(ifp->if_serializer);
3049 
3050 #ifndef BURN_BRIDGES
3051 	/* better way to do this? */
3052 	for (i = 0; i < 5; i++)
3053 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
3054 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
3055 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
3056 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
3057 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
3058 
3059 	/* reenable busmastering */
3060 	pci_enable_busmaster(dev);
3061 	pci_enable_io(dev, SYS_RES_IOPORT);
3062 #endif
3063 
3064 	/* reinitialize interface if necessary */
3065 	if (ifp->if_flags & IFF_UP)
3066 		re_init(sc);
3067 
3068 	sc->re_flags &= ~RE_F_SUSPENDED;
3069 
3070 	lwkt_serialize_exit(ifp->if_serializer);
3071 
3072 	return (0);
3073 }
3074 
3075 /*
3076  * Stop all chip I/O so that the kernel's probe routines don't
3077  * get confused by errant DMAs when rebooting.
3078  */
3079 static void
3080 re_shutdown(device_t dev)
3081 {
3082 	struct re_softc *sc = device_get_softc(dev);
3083 	struct ifnet *ifp = &sc->arpcom.ac_if;
3084 
3085 	lwkt_serialize_enter(ifp->if_serializer);
3086 	re_stop(sc);
3087 	lwkt_serialize_exit(ifp->if_serializer);
3088 }
3089 
3090 static int
3091 re_sysctl_rxtime(SYSCTL_HANDLER_ARGS)
3092 {
3093 	struct re_softc *sc = arg1;
3094 
3095 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_rx_time);
3096 }
3097 
3098 static int
3099 re_sysctl_txtime(SYSCTL_HANDLER_ARGS)
3100 {
3101 	struct re_softc *sc = arg1;
3102 
3103 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_tx_time);
3104 }
3105 
3106 static int
3107 re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *hwtime)
3108 {
3109 	struct re_softc *sc = arg1;
3110 	struct ifnet *ifp = &sc->arpcom.ac_if;
3111 	int error, v;
3112 
3113 	lwkt_serialize_enter(ifp->if_serializer);
3114 
3115 	v = *hwtime;
3116 	error = sysctl_handle_int(oidp, &v, 0, req);
3117 	if (error || req->newptr == NULL)
3118 		goto back;
3119 
3120 	if (v <= 0) {
3121 		error = EINVAL;
3122 		goto back;
3123 	}
3124 
3125 	if (v != *hwtime) {
3126 		*hwtime = v;
3127 
3128 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3129 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_HW)
3130 			re_setup_hw_im(sc);
3131 	}
3132 back:
3133 	lwkt_serialize_exit(ifp->if_serializer);
3134 	return error;
3135 }
3136 
3137 static int
3138 re_sysctl_simtime(SYSCTL_HANDLER_ARGS)
3139 {
3140 	struct re_softc *sc = arg1;
3141 	struct ifnet *ifp = &sc->arpcom.ac_if;
3142 	int error, v;
3143 
3144 	lwkt_serialize_enter(ifp->if_serializer);
3145 
3146 	v = sc->re_sim_time;
3147 	error = sysctl_handle_int(oidp, &v, 0, req);
3148 	if (error || req->newptr == NULL)
3149 		goto back;
3150 
3151 	if (v <= 0) {
3152 		error = EINVAL;
3153 		goto back;
3154 	}
3155 
3156 	if (v != sc->re_sim_time) {
3157 		sc->re_sim_time = v;
3158 
3159 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3160 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_SIM) {
3161 #ifdef foo
3162 			int reg;
3163 
3164 			/*
3165 			 * Following code causes various strange
3166 			 * performance problems.  Hmm ...
3167 			 */
3168 			CSR_WRITE_2(sc, RE_IMR, 0);
3169 			if (!RE_IS_8139CP(sc))
3170 				reg = RE_TIMERINT_8169;
3171 			else
3172 				reg = RE_TIMERINT;
3173 			CSR_WRITE_4(sc, reg, 0);
3174 			CSR_READ_4(sc, reg); /* flush */
3175 
3176 			CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3177 			re_setup_sim_im(sc);
3178 #else
3179 			re_setup_intr(sc, 0, RE_IMTYPE_NONE);
3180 			DELAY(10);
3181 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
3182 #endif
3183 		}
3184 	}
3185 back:
3186 	lwkt_serialize_exit(ifp->if_serializer);
3187 	return error;
3188 }
3189 
3190 static int
3191 re_sysctl_imtype(SYSCTL_HANDLER_ARGS)
3192 {
3193 	struct re_softc *sc = arg1;
3194 	struct ifnet *ifp = &sc->arpcom.ac_if;
3195 	int error, v;
3196 
3197 	lwkt_serialize_enter(ifp->if_serializer);
3198 
3199 	v = sc->re_imtype;
3200 	error = sysctl_handle_int(oidp, &v, 0, req);
3201 	if (error || req->newptr == NULL)
3202 		goto back;
3203 
3204 	if (v != RE_IMTYPE_HW && v != RE_IMTYPE_SIM && v != RE_IMTYPE_NONE) {
3205 		error = EINVAL;
3206 		goto back;
3207 	}
3208 	if (v == RE_IMTYPE_HW && (sc->re_caps & RE_C_HWIM) == 0) {
3209 		/* Can't do hardware interrupt moderation */
3210 		error = EOPNOTSUPP;
3211 		goto back;
3212 	}
3213 
3214 	if (v != sc->re_imtype) {
3215 		sc->re_imtype = v;
3216 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3217 		    IFF_RUNNING)
3218 			re_setup_intr(sc, 1, sc->re_imtype);
3219 	}
3220 back:
3221 	lwkt_serialize_exit(ifp->if_serializer);
3222 	return error;
3223 }
3224 
3225 static void
3226 re_setup_hw_im(struct re_softc *sc)
3227 {
3228 	KKASSERT(sc->re_caps & RE_C_HWIM);
3229 
3230 	/*
3231 	 * Interrupt moderation
3232 	 *
3233 	 * 0xABCD
3234 	 * A - unknown (maybe TX related)
3235 	 * B - TX timer (unit: 25us)
3236 	 * C - unknown (maybe RX related)
3237 	 * D - RX timer (unit: 25us)
3238 	 *
3239 	 *
3240 	 * re(4)'s interrupt moderation is actually controlled by
3241 	 * two variables, like most other NICs (bge, bce etc.)
3242 	 * o  timer
3243 	 * o  number of packets [P]
3244 	 *
3245 	 * The logic relationship between these two variables is
3246 	 * similar to other NICs too:
3247 	 * if (timer expire || packets > [P])
3248 	 *     Interrupt is delivered
3249 	 *
3250 	 * Currently we only know how to set 'timer', but not
3251 	 * 'number of packets', which should be ~30, as far as I
3252 	 * tested (sink ~900Kpps, interrupt rate is 30KHz)
3253 	 */
3254 	CSR_WRITE_2(sc, RE_IM,
3255 		    RE_IM_RXTIME(sc->re_rx_time) |
3256 		    RE_IM_TXTIME(sc->re_tx_time) |
3257 		    RE_IM_MAGIC);
3258 }
3259 
3260 static void
3261 re_disable_hw_im(struct re_softc *sc)
3262 {
3263 	if (sc->re_caps & RE_C_HWIM)
3264 		CSR_WRITE_2(sc, RE_IM, 0);
3265 }
3266 
3267 static void
3268 re_setup_sim_im(struct re_softc *sc)
3269 {
3270 	if (!RE_IS_8139CP(sc)) {
3271 		uint32_t ticks;
3272 
3273 		/*
3274 		 * Datasheet says tick decreases at bus speed,
3275 		 * but it seems the clock runs a little bit
3276 		 * faster, so we do some compensation here.
3277 		 */
3278 		ticks = (sc->re_sim_time * sc->re_bus_speed * 8) / 5;
3279 		CSR_WRITE_4(sc, RE_TIMERINT_8169, ticks);
3280 	} else {
3281 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400); /* XXX */
3282 	}
3283 	CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
3284 	sc->re_flags |= RE_F_TIMER_INTR;
3285 }
3286 
3287 static void
3288 re_disable_sim_im(struct re_softc *sc)
3289 {
3290 	if (!RE_IS_8139CP(sc))
3291 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0);
3292 	else
3293 		CSR_WRITE_4(sc, RE_TIMERINT, 0);
3294 	sc->re_flags &= ~RE_F_TIMER_INTR;
3295 }
3296 
3297 static void
3298 re_config_imtype(struct re_softc *sc, int imtype)
3299 {
3300 	switch (imtype) {
3301 	case RE_IMTYPE_HW:
3302 		KKASSERT(sc->re_caps & RE_C_HWIM);
3303 		/* FALL THROUGH */
3304 	case RE_IMTYPE_NONE:
3305 		sc->re_intrs = RE_INTRS;
3306 		sc->re_rx_ack = RE_ISR_RX_OK | RE_ISR_FIFO_OFLOW |
3307 				RE_ISR_RX_OVERRUN;
3308 		sc->re_tx_ack = RE_ISR_TX_OK;
3309 		break;
3310 
3311 	case RE_IMTYPE_SIM:
3312 		sc->re_intrs = RE_INTRS_TIMER;
3313 		sc->re_rx_ack = RE_ISR_TIMEOUT_EXPIRED;
3314 		sc->re_tx_ack = RE_ISR_TIMEOUT_EXPIRED;
3315 		break;
3316 
3317 	default:
3318 		panic("%s: unknown imtype %d\n",
3319 		      sc->arpcom.ac_if.if_xname, imtype);
3320 	}
3321 }
3322 
3323 static void
3324 re_setup_intr(struct re_softc *sc, int enable_intrs, int imtype)
3325 {
3326 	re_config_imtype(sc, imtype);
3327 
3328 	if (enable_intrs)
3329 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3330 	else
3331 		CSR_WRITE_2(sc, RE_IMR, 0);
3332 
3333 	switch (imtype) {
3334 	case RE_IMTYPE_NONE:
3335 		re_disable_sim_im(sc);
3336 		re_disable_hw_im(sc);
3337 		break;
3338 
3339 	case RE_IMTYPE_HW:
3340 		KKASSERT(sc->re_caps & RE_C_HWIM);
3341 		re_disable_sim_im(sc);
3342 		re_setup_hw_im(sc);
3343 		break;
3344 
3345 	case RE_IMTYPE_SIM:
3346 		re_disable_hw_im(sc);
3347 		re_setup_sim_im(sc);
3348 		break;
3349 
3350 	default:
3351 		panic("%s: unknown imtype %d\n",
3352 		      sc->arpcom.ac_if.if_xname, imtype);
3353 	}
3354 }
3355 
3356 static void
3357 re_get_eaddr(struct re_softc *sc, uint8_t *eaddr)
3358 {
3359 	int i;
3360 
3361 	if (sc->re_macver == RE_MACVER_11 || sc->re_macver == RE_MACVER_12) {
3362 		uint16_t re_did;
3363 
3364 		re_get_eewidth(sc);
3365 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
3366 		if (re_did == 0x8128) {
3367 			uint16_t as[ETHER_ADDR_LEN / 2];
3368 
3369 			/*
3370 			 * Get station address from the EEPROM.
3371 			 */
3372 			re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
3373 			for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
3374 				as[i] = le16toh(as[i]);
3375 			bcopy(as, eaddr, sizeof(eaddr));
3376 			return;
3377 		}
3378 	}
3379 
3380 	/*
3381 	 * Get station address from IDRx.
3382 	 */
3383 	for (i = 0; i < ETHER_ADDR_LEN; ++i)
3384 		eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
3385 }
3386 
3387 static int
3388 re_jpool_alloc(struct re_softc *sc)
3389 {
3390 	struct re_list_data *ldata = &sc->re_ldata;
3391 	struct re_jbuf *jbuf;
3392 	bus_addr_t paddr;
3393 	bus_size_t jpool_size;
3394 	caddr_t buf;
3395 	int i, error;
3396 
3397 	lwkt_serialize_init(&ldata->re_jbuf_serializer);
3398 
3399 	ldata->re_jbuf = kmalloc(sizeof(struct re_jbuf) * RE_JBUF_COUNT(sc),
3400 				 M_DEVBUF, M_WAITOK | M_ZERO);
3401 
3402 	jpool_size = RE_JBUF_COUNT(sc) * RE_JBUF_SIZE;
3403 
3404 	error = bus_dma_tag_create(sc->re_parent_tag,
3405 			RE_BUF_ALIGN, 0,	/* alignment, boundary */
3406 			BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
3407 			BUS_SPACE_MAXADDR,	/* highaddr */
3408 			NULL, NULL,		/* filter, filterarg */
3409 			jpool_size, 1,		/* nsegments, maxsize */
3410 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
3411 			BUS_DMA_ALLOCNOW,	/* flags */
3412 			&ldata->re_jpool_tag);
3413 	if (error) {
3414 		device_printf(sc->re_dev, "could not allocate jumbo dma tag\n");
3415 		return error;
3416 	}
3417 
3418 	error = bus_dmamem_alloc(ldata->re_jpool_tag, (void **)&ldata->re_jpool,
3419 				 BUS_DMA_WAITOK, &ldata->re_jpool_map);
3420 	if (error) {
3421 		device_printf(sc->re_dev,
3422 			      "could not allocate jumbo dma memory\n");
3423 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3424 		ldata->re_jpool_tag = NULL;
3425 		return error;
3426 	}
3427 
3428 	error = bus_dmamap_load(ldata->re_jpool_tag, ldata->re_jpool_map,
3429 				ldata->re_jpool, jpool_size,
3430 				re_dma_map_addr, &paddr, BUS_DMA_WAITOK);
3431 	if (error) {
3432 		device_printf(sc->re_dev, "could not load jumbo dma map\n");
3433 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3434 				ldata->re_jpool_map);
3435 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3436 		ldata->re_jpool_tag = NULL;
3437 		return error;
3438 	}
3439 
3440 	/* ..and split it into 9KB chunks */
3441 	SLIST_INIT(&ldata->re_jbuf_free);
3442 
3443 	buf = ldata->re_jpool;
3444 	for (i = 0; i < RE_JBUF_COUNT(sc); i++) {
3445 		jbuf = &ldata->re_jbuf[i];
3446 
3447 		jbuf->re_sc = sc;
3448 		jbuf->re_inuse = 0;
3449 		jbuf->re_slot = i;
3450 		jbuf->re_buf = buf;
3451 		jbuf->re_paddr = paddr;
3452 
3453 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3454 
3455 		buf += RE_JBUF_SIZE;
3456 		paddr += RE_JBUF_SIZE;
3457 	}
3458 	return 0;
3459 }
3460 
3461 static void
3462 re_jpool_free(struct re_softc *sc)
3463 {
3464 	struct re_list_data *ldata = &sc->re_ldata;
3465 
3466 	if (ldata->re_jpool_tag != NULL) {
3467 		bus_dmamap_unload(ldata->re_jpool_tag, ldata->re_jpool_map);
3468 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3469 				ldata->re_jpool_map);
3470 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3471 		ldata->re_jpool_tag = NULL;
3472 	}
3473 
3474 	if (ldata->re_jbuf != NULL) {
3475 		kfree(ldata->re_jbuf, M_DEVBUF);
3476 		ldata->re_jbuf = NULL;
3477 	}
3478 }
3479 
3480 static struct re_jbuf *
3481 re_jbuf_alloc(struct re_softc *sc)
3482 {
3483 	struct re_list_data *ldata = &sc->re_ldata;
3484 	struct re_jbuf *jbuf;
3485 
3486 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3487 
3488 	jbuf = SLIST_FIRST(&ldata->re_jbuf_free);
3489 	if (jbuf != NULL) {
3490 		SLIST_REMOVE_HEAD(&ldata->re_jbuf_free, re_link);
3491 		jbuf->re_inuse = 1;
3492 	}
3493 
3494 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3495 
3496 	return jbuf;
3497 }
3498 
3499 static void
3500 re_jbuf_free(void *arg)
3501 {
3502 	struct re_jbuf *jbuf = arg;
3503 	struct re_softc *sc = jbuf->re_sc;
3504 	struct re_list_data *ldata = &sc->re_ldata;
3505 
3506 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3507 		panic("%s: free wrong jumbo buffer\n",
3508 		      sc->arpcom.ac_if.if_xname);
3509 	} else if (jbuf->re_inuse == 0) {
3510 		panic("%s: jumbo buffer already freed\n",
3511 		      sc->arpcom.ac_if.if_xname);
3512 	}
3513 
3514 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3515 	atomic_subtract_int(&jbuf->re_inuse, 1);
3516 	if (jbuf->re_inuse == 0)
3517 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3518 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3519 }
3520 
3521 static void
3522 re_jbuf_ref(void *arg)
3523 {
3524 	struct re_jbuf *jbuf = arg;
3525 	struct re_softc *sc = jbuf->re_sc;
3526 	struct re_list_data *ldata = &sc->re_ldata;
3527 
3528 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3529 		panic("%s: ref wrong jumbo buffer\n",
3530 		      sc->arpcom.ac_if.if_xname);
3531 	} else if (jbuf->re_inuse == 0) {
3532 		panic("%s: jumbo buffer already freed\n",
3533 		      sc->arpcom.ac_if.if_xname);
3534 	}
3535 	atomic_add_int(&jbuf->re_inuse, 1);
3536 }
3537