xref: /dragonfly/sys/dev/netif/re/if_re.c (revision 8aaa1e58)
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 	{ PCI_VENDOR_TTTECH, PCI_PRODUCT_TTTECH_MC322,
196 	  "TTTech MC322 Gigabit Ethernet" },
197 
198 	{ 0, 0, NULL }
199 };
200 
201 static const struct re_hwrev re_hwrevs[] = {
202 	{ RE_HWREV_8139CPLUS,	RE_MACVER_UNKN,		ETHERMTU,
203 	  RE_C_HWCSUM | RE_C_8139CP | RE_C_FASTE },
204 
205 	{ RE_HWREV_8169,	RE_MACVER_UNKN,		ETHERMTU,
206 	  RE_C_HWCSUM | RE_C_8169 },
207 
208 	{ RE_HWREV_8110S,	RE_MACVER_03,		RE_MTU_6K,
209 	  RE_C_HWCSUM | RE_C_8169 },
210 
211 	{ RE_HWREV_8169S,	RE_MACVER_03,		RE_MTU_6K,
212 	  RE_C_HWCSUM | RE_C_8169 },
213 
214 	{ RE_HWREV_8169SB,	RE_MACVER_04,		RE_MTU_6K,
215 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
216 
217 	{ RE_HWREV_8169SC1,	RE_MACVER_05,		RE_MTU_6K,
218 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
219 
220 	{ RE_HWREV_8169SC2,	RE_MACVER_06,		RE_MTU_6K,
221 	  RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_8169 },
222 
223 	{ RE_HWREV_8168B1,	RE_MACVER_21,		RE_MTU_6K,
224 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT },
225 
226 	{ RE_HWREV_8168B2,	RE_MACVER_23,		RE_MTU_6K,
227 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
228 
229 	{ RE_HWREV_8168B3,	RE_MACVER_23,		RE_MTU_6K,
230 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_PHYPMGT | RE_C_AUTOPAD },
231 
232 	{ RE_HWREV_8168C,	RE_MACVER_29,		RE_MTU_6K,
233 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
234 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
235 
236 	{ RE_HWREV_8168CP,	RE_MACVER_2B,		RE_MTU_6K,
237 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
238 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
239 
240 	{ RE_HWREV_8168D,	RE_MACVER_2A,		RE_MTU_9K,
241 	  RE_C_HWIM | RE_C_HWCSUM | RE_C_MAC2 | RE_C_PHYPMGT |
242 	  RE_C_AUTOPAD | RE_C_CONTIGRX | RE_C_STOP_RXTX },
243 
244 	{ RE_HWREV_8100E,	RE_MACVER_UNKN,		ETHERMTU,
245 	  RE_C_HWCSUM | RE_C_FASTE },
246 
247 	{ RE_HWREV_8101E1,	RE_MACVER_16,		ETHERMTU,
248 	  RE_C_HWCSUM | RE_C_FASTE },
249 
250 	{ RE_HWREV_8101E2,	RE_MACVER_16,		ETHERMTU,
251 	  RE_C_HWCSUM | RE_C_FASTE },
252 
253 	{ RE_HWREV_8102E,	RE_MACVER_15,		ETHERMTU,
254 	  RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX |
255 	  RE_C_FASTE },
256 
257 	{ RE_HWREV_8102EL,	RE_MACVER_15,		ETHERMTU,
258 	  RE_C_HWCSUM | RE_C_MAC2 | RE_C_AUTOPAD | RE_C_STOP_RXTX |
259 	  RE_C_FASTE },
260 
261 	{ RE_HWREV_NULL, 0, 0, 0 }
262 };
263 
264 static int	re_probe(device_t);
265 static int	re_attach(device_t);
266 static int	re_detach(device_t);
267 static int	re_suspend(device_t);
268 static int	re_resume(device_t);
269 static void	re_shutdown(device_t);
270 
271 static void	re_dma_map_addr(void *, bus_dma_segment_t *, int, int);
272 static void	re_dma_map_desc(void *, bus_dma_segment_t *, int,
273 				bus_size_t, int);
274 static int	re_allocmem(device_t);
275 static void	re_freemem(device_t);
276 static void	re_freebufmem(struct re_softc *, int, int);
277 static int	re_encap(struct re_softc *, struct mbuf **, int *);
278 static int	re_newbuf_std(struct re_softc *, int, int);
279 static int	re_newbuf_jumbo(struct re_softc *, int, int);
280 static void	re_setup_rxdesc(struct re_softc *, int);
281 static int	re_rx_list_init(struct re_softc *);
282 static int	re_tx_list_init(struct re_softc *);
283 static int	re_rxeof(struct re_softc *);
284 static int	re_txeof(struct re_softc *);
285 static int	re_tx_collect(struct re_softc *);
286 static void	re_intr(void *);
287 static void	re_tick(void *);
288 static void	re_tick_serialized(void *);
289 
290 static void	re_start(struct ifnet *);
291 static int	re_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
292 static void	re_init(void *);
293 static void	re_stop(struct re_softc *);
294 static void	re_watchdog(struct ifnet *);
295 static int	re_ifmedia_upd(struct ifnet *);
296 static void	re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
297 
298 static void	re_eeprom_putbyte(struct re_softc *, int);
299 static void	re_eeprom_getword(struct re_softc *, int, u_int16_t *);
300 static void	re_read_eeprom(struct re_softc *, caddr_t, int, int);
301 static void	re_get_eewidth(struct re_softc *);
302 
303 static int	re_gmii_readreg(device_t, int, int);
304 static int	re_gmii_writereg(device_t, int, int, int);
305 
306 static int	re_miibus_readreg(device_t, int, int);
307 static int	re_miibus_writereg(device_t, int, int, int);
308 static void	re_miibus_statchg(device_t);
309 
310 static void	re_setmulti(struct re_softc *);
311 static void	re_reset(struct re_softc *, int);
312 static void	re_get_eaddr(struct re_softc *, uint8_t *);
313 
314 static void	re_setup_hw_im(struct re_softc *);
315 static void	re_setup_sim_im(struct re_softc *);
316 static void	re_disable_hw_im(struct re_softc *);
317 static void	re_disable_sim_im(struct re_softc *);
318 static void	re_config_imtype(struct re_softc *, int);
319 static void	re_setup_intr(struct re_softc *, int, int);
320 
321 static int	re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *);
322 static int	re_sysctl_rxtime(SYSCTL_HANDLER_ARGS);
323 static int	re_sysctl_txtime(SYSCTL_HANDLER_ARGS);
324 static int	re_sysctl_simtime(SYSCTL_HANDLER_ARGS);
325 static int	re_sysctl_imtype(SYSCTL_HANDLER_ARGS);
326 
327 static int	re_jpool_alloc(struct re_softc *);
328 static void	re_jpool_free(struct re_softc *);
329 static struct re_jbuf *re_jbuf_alloc(struct re_softc *);
330 static void	re_jbuf_free(void *);
331 static void	re_jbuf_ref(void *);
332 
333 #ifdef RE_DIAG
334 static int	re_diag(struct re_softc *);
335 #endif
336 
337 #ifdef DEVICE_POLLING
338 static void	re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
339 #endif
340 
341 static device_method_t re_methods[] = {
342 	/* Device interface */
343 	DEVMETHOD(device_probe,		re_probe),
344 	DEVMETHOD(device_attach,	re_attach),
345 	DEVMETHOD(device_detach,	re_detach),
346 	DEVMETHOD(device_suspend,	re_suspend),
347 	DEVMETHOD(device_resume,	re_resume),
348 	DEVMETHOD(device_shutdown,	re_shutdown),
349 
350 	/* bus interface */
351 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
352 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
353 
354 	/* MII interface */
355 	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
356 	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
357 	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
358 
359 	{ 0, 0 }
360 };
361 
362 static driver_t re_driver = {
363 	"re",
364 	re_methods,
365 	sizeof(struct re_softc)
366 };
367 
368 static devclass_t re_devclass;
369 
370 DECLARE_DUMMY_MODULE(if_re);
371 MODULE_DEPEND(if_re, miibus, 1, 1, 1);
372 DRIVER_MODULE(if_re, pci, re_driver, re_devclass, 0, 0);
373 DRIVER_MODULE(if_re, cardbus, re_driver, re_devclass, 0, 0);
374 DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
375 
376 static int	re_rx_desc_count = RE_RX_DESC_CNT_DEF;
377 static int	re_tx_desc_count = RE_TX_DESC_CNT_DEF;
378 
379 TUNABLE_INT("hw.re.rx_desc_count", &re_rx_desc_count);
380 TUNABLE_INT("hw.re.tx_desc_count", &re_tx_desc_count);
381 
382 #define EE_SET(x)	\
383 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) | (x))
384 
385 #define EE_CLR(x)	\
386 	CSR_WRITE_1(sc, RE_EECMD, CSR_READ_1(sc, RE_EECMD) & ~(x))
387 
388 static __inline void
389 re_free_rxchain(struct re_softc *sc)
390 {
391 	if (sc->re_head != NULL) {
392 		m_freem(sc->re_head);
393 		sc->re_head = sc->re_tail = NULL;
394 	}
395 }
396 
397 /*
398  * Send a read command and address to the EEPROM, check for ACK.
399  */
400 static void
401 re_eeprom_putbyte(struct re_softc *sc, int addr)
402 {
403 	int d, i;
404 
405 	d = addr | (RE_9346_READ << sc->re_eewidth);
406 
407 	/*
408 	 * Feed in each bit and strobe the clock.
409 	 */
410 	for (i = 1 << (sc->re_eewidth + 3); i; i >>= 1) {
411 		if (d & i)
412 			EE_SET(RE_EE_DATAIN);
413 		else
414 			EE_CLR(RE_EE_DATAIN);
415 		DELAY(100);
416 		EE_SET(RE_EE_CLK);
417 		DELAY(150);
418 		EE_CLR(RE_EE_CLK);
419 		DELAY(100);
420 	}
421 }
422 
423 /*
424  * Read a word of data stored in the EEPROM at address 'addr.'
425  */
426 static void
427 re_eeprom_getword(struct re_softc *sc, int addr, uint16_t *dest)
428 {
429 	int i;
430 	uint16_t word = 0;
431 
432 	/*
433 	 * Send address of word we want to read.
434 	 */
435 	re_eeprom_putbyte(sc, addr);
436 
437 	/*
438 	 * Start reading bits from EEPROM.
439 	 */
440 	for (i = 0x8000; i != 0; i >>= 1) {
441 		EE_SET(RE_EE_CLK);
442 		DELAY(100);
443 		if (CSR_READ_1(sc, RE_EECMD) & RE_EE_DATAOUT)
444 			word |= i;
445 		EE_CLR(RE_EE_CLK);
446 		DELAY(100);
447 	}
448 
449 	*dest = word;
450 }
451 
452 /*
453  * Read a sequence of words from the EEPROM.
454  */
455 static void
456 re_read_eeprom(struct re_softc *sc, caddr_t dest, int off, int cnt)
457 {
458 	int i;
459 	uint16_t word = 0, *ptr;
460 
461 	CSR_SETBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
462 	DELAY(100);
463 
464 	for (i = 0; i < cnt; i++) {
465 		CSR_SETBIT_1(sc, RE_EECMD, RE_EE_SEL);
466 		re_eeprom_getword(sc, off + i, &word);
467 		CSR_CLRBIT_1(sc, RE_EECMD, RE_EE_SEL);
468 		ptr = (uint16_t *)(dest + (i * 2));
469 		*ptr = word;
470 	}
471 
472 	CSR_CLRBIT_1(sc, RE_EECMD, RE_EEMODE_PROGRAM);
473 }
474 
475 static void
476 re_get_eewidth(struct re_softc *sc)
477 {
478 	uint16_t re_did = 0;
479 
480 	sc->re_eewidth = 6;
481 	re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
482 	if (re_did != 0x8129)
483 		sc->re_eewidth = 8;
484 }
485 
486 static int
487 re_gmii_readreg(device_t dev, int phy, int reg)
488 {
489 	struct re_softc *sc = device_get_softc(dev);
490 	u_int32_t rval;
491 	int i;
492 
493 	if (phy != 1)
494 		return(0);
495 
496 	/* Let the rgephy driver read the GMEDIASTAT register */
497 
498 	if (reg == RE_GMEDIASTAT)
499 		return(CSR_READ_1(sc, RE_GMEDIASTAT));
500 
501 	CSR_WRITE_4(sc, RE_PHYAR, reg << 16);
502 	DELAY(1000);
503 
504 	for (i = 0; i < RE_TIMEOUT; i++) {
505 		rval = CSR_READ_4(sc, RE_PHYAR);
506 		if (rval & RE_PHYAR_BUSY)
507 			break;
508 		DELAY(100);
509 	}
510 
511 	if (i == RE_TIMEOUT) {
512 		device_printf(dev, "PHY read failed\n");
513 		return(0);
514 	}
515 
516 	return(rval & RE_PHYAR_PHYDATA);
517 }
518 
519 static int
520 re_gmii_writereg(device_t dev, int phy, int reg, int data)
521 {
522 	struct re_softc *sc = device_get_softc(dev);
523 	uint32_t rval;
524 	int i;
525 
526 	CSR_WRITE_4(sc, RE_PHYAR,
527 		    (reg << 16) | (data & RE_PHYAR_PHYDATA) | RE_PHYAR_BUSY);
528 	DELAY(1000);
529 
530 	for (i = 0; i < RE_TIMEOUT; i++) {
531 		rval = CSR_READ_4(sc, RE_PHYAR);
532 		if ((rval & RE_PHYAR_BUSY) == 0)
533 			break;
534 		DELAY(100);
535 	}
536 
537 	if (i == RE_TIMEOUT)
538 		device_printf(dev, "PHY write failed\n");
539 
540 	return(0);
541 }
542 
543 static int
544 re_miibus_readreg(device_t dev, int phy, int reg)
545 {
546 	struct re_softc	*sc = device_get_softc(dev);
547 	uint16_t rval = 0;
548 	uint16_t re8139_reg = 0;
549 
550 	if (!RE_IS_8139CP(sc)) {
551 		rval = re_gmii_readreg(dev, phy, reg);
552 		return(rval);
553 	}
554 
555 	/* Pretend the internal PHY is only at address 0 */
556 	if (phy)
557 		return(0);
558 
559 	switch(reg) {
560 	case MII_BMCR:
561 		re8139_reg = RE_BMCR;
562 		break;
563 	case MII_BMSR:
564 		re8139_reg = RE_BMSR;
565 		break;
566 	case MII_ANAR:
567 		re8139_reg = RE_ANAR;
568 		break;
569 	case MII_ANER:
570 		re8139_reg = RE_ANER;
571 		break;
572 	case MII_ANLPAR:
573 		re8139_reg = RE_LPAR;
574 		break;
575 	case MII_PHYIDR1:
576 	case MII_PHYIDR2:
577 		return(0);
578 	/*
579 	 * Allow the rlphy driver to read the media status
580 	 * register. If we have a link partner which does not
581 	 * support NWAY, this is the register which will tell
582 	 * us the results of parallel detection.
583 	 */
584 	case RE_MEDIASTAT:
585 		return(CSR_READ_1(sc, RE_MEDIASTAT));
586 	default:
587 		device_printf(dev, "bad phy register\n");
588 		return(0);
589 	}
590 	rval = CSR_READ_2(sc, re8139_reg);
591 	if (re8139_reg == RE_BMCR) {
592 		/* 8139C+ has different bit layout. */
593 		rval &= ~(BMCR_LOOP | BMCR_ISO);
594 	}
595 	return(rval);
596 }
597 
598 static int
599 re_miibus_writereg(device_t dev, int phy, int reg, int data)
600 {
601 	struct re_softc *sc= device_get_softc(dev);
602 	u_int16_t re8139_reg = 0;
603 
604 	if (!RE_IS_8139CP(sc))
605 		return(re_gmii_writereg(dev, phy, reg, data));
606 
607 	/* Pretend the internal PHY is only at address 0 */
608 	if (phy)
609 		return(0);
610 
611 	switch(reg) {
612 	case MII_BMCR:
613 		re8139_reg = RE_BMCR;
614 		/* 8139C+ has different bit layout. */
615 		data &= ~(BMCR_LOOP | BMCR_ISO);
616 		break;
617 	case MII_BMSR:
618 		re8139_reg = RE_BMSR;
619 		break;
620 	case MII_ANAR:
621 		re8139_reg = RE_ANAR;
622 		break;
623 	case MII_ANER:
624 		re8139_reg = RE_ANER;
625 		break;
626 	case MII_ANLPAR:
627 		re8139_reg = RE_LPAR;
628 		break;
629 	case MII_PHYIDR1:
630 	case MII_PHYIDR2:
631 		return(0);
632 	default:
633 		device_printf(dev, "bad phy register\n");
634 		return(0);
635 	}
636 	CSR_WRITE_2(sc, re8139_reg, data);
637 	return(0);
638 }
639 
640 static void
641 re_miibus_statchg(device_t dev)
642 {
643 }
644 
645 /*
646  * Program the 64-bit multicast hash filter.
647  */
648 static void
649 re_setmulti(struct re_softc *sc)
650 {
651 	struct ifnet *ifp = &sc->arpcom.ac_if;
652 	int h = 0;
653 	uint32_t hashes[2] = { 0, 0 };
654 	struct ifmultiaddr *ifma;
655 	uint32_t rxfilt;
656 	int mcnt = 0;
657 
658 	rxfilt = CSR_READ_4(sc, RE_RXCFG);
659 
660 	/* Set the individual bit to receive frames for this host only. */
661 	rxfilt |= RE_RXCFG_RX_INDIV;
662 	/* Set capture broadcast bit to capture broadcast frames. */
663 	rxfilt |= RE_RXCFG_RX_BROAD;
664 
665 	rxfilt &= ~(RE_RXCFG_RX_ALLPHYS | RE_RXCFG_RX_MULTI);
666 	if ((ifp->if_flags & IFF_ALLMULTI) || (ifp->if_flags & IFF_PROMISC)) {
667 		rxfilt |= RE_RXCFG_RX_MULTI;
668 
669 		/* If we want promiscuous mode, set the allframes bit. */
670 		if (ifp->if_flags & IFF_PROMISC)
671 			rxfilt |= RE_RXCFG_RX_ALLPHYS;
672 
673 		CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
674 		CSR_WRITE_4(sc, RE_MAR0, 0xFFFFFFFF);
675 		CSR_WRITE_4(sc, RE_MAR4, 0xFFFFFFFF);
676 		return;
677 	}
678 
679 	/* first, zot all the existing hash bits */
680 	CSR_WRITE_4(sc, RE_MAR0, 0);
681 	CSR_WRITE_4(sc, RE_MAR4, 0);
682 
683 	/* now program new ones */
684 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
685 		if (ifma->ifma_addr->sa_family != AF_LINK)
686 			continue;
687 		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
688 		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
689 		if (h < 32)
690 			hashes[0] |= (1 << h);
691 		else
692 			hashes[1] |= (1 << (h - 32));
693 		mcnt++;
694 	}
695 
696 	if (mcnt)
697 		rxfilt |= RE_RXCFG_RX_MULTI;
698 	else
699 		rxfilt &= ~RE_RXCFG_RX_MULTI;
700 
701 	CSR_WRITE_4(sc, RE_RXCFG, rxfilt);
702 
703 	/*
704 	 * For some unfathomable reason, RealTek decided to reverse
705 	 * the order of the multicast hash registers in the PCI Express
706 	 * parts. This means we have to write the hash pattern in reverse
707 	 * order for those devices.
708 	 */
709 	if (sc->re_caps & RE_C_PCIE) {
710 		CSR_WRITE_4(sc, RE_MAR0, bswap32(hashes[0]));
711 		CSR_WRITE_4(sc, RE_MAR4, bswap32(hashes[1]));
712 	} else {
713 		CSR_WRITE_4(sc, RE_MAR0, hashes[0]);
714 		CSR_WRITE_4(sc, RE_MAR4, hashes[1]);
715 	}
716 }
717 
718 static void
719 re_reset(struct re_softc *sc, int running)
720 {
721 	int i;
722 
723 	if ((sc->re_caps & RE_C_STOP_RXTX) && running) {
724 		CSR_WRITE_1(sc, RE_COMMAND,
725 			    RE_CMD_STOPREQ | RE_CMD_TX_ENB | RE_CMD_RX_ENB);
726 		DELAY(100);
727 	}
728 
729 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_RESET);
730 
731 	for (i = 0; i < RE_TIMEOUT; i++) {
732 		DELAY(10);
733 		if ((CSR_READ_1(sc, RE_COMMAND) & RE_CMD_RESET) == 0)
734 			break;
735 	}
736 	if (i == RE_TIMEOUT)
737 		if_printf(&sc->arpcom.ac_if, "reset never completed!\n");
738 }
739 
740 #ifdef RE_DIAG
741 /*
742  * The following routine is designed to test for a defect on some
743  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
744  * lines connected to the bus, however for a 32-bit only card, they
745  * should be pulled high. The result of this defect is that the
746  * NIC will not work right if you plug it into a 64-bit slot: DMA
747  * operations will be done with 64-bit transfers, which will fail
748  * because the 64-bit data lines aren't connected.
749  *
750  * There's no way to work around this (short of talking a soldering
751  * iron to the board), however we can detect it. The method we use
752  * here is to put the NIC into digital loopback mode, set the receiver
753  * to promiscuous mode, and then try to send a frame. We then compare
754  * the frame data we sent to what was received. If the data matches,
755  * then the NIC is working correctly, otherwise we know the user has
756  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
757  * slot. In the latter case, there's no way the NIC can work correctly,
758  * so we print out a message on the console and abort the device attach.
759  */
760 
761 static int
762 re_diag(struct re_softc *sc)
763 {
764 	struct ifnet *ifp = &sc->arpcom.ac_if;
765 	struct mbuf *m0;
766 	struct ether_header *eh;
767 	struct re_desc *cur_rx;
768 	uint16_t status;
769 	uint32_t rxstat;
770 	int total_len, i, error = 0, phyaddr;
771 	uint8_t dst[ETHER_ADDR_LEN] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
772 	uint8_t src[ETHER_ADDR_LEN] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
773 
774 	/* Allocate a single mbuf */
775 
776 	MGETHDR(m0, MB_DONTWAIT, MT_DATA);
777 	if (m0 == NULL)
778 		return(ENOBUFS);
779 
780 	/*
781 	 * Initialize the NIC in test mode. This sets the chip up
782 	 * so that it can send and receive frames, but performs the
783 	 * following special functions:
784 	 * - Puts receiver in promiscuous mode
785 	 * - Enables digital loopback mode
786 	 * - Leaves interrupts turned off
787 	 */
788 
789 	ifp->if_flags |= IFF_PROMISC;
790 	sc->re_flags |= RE_F_TESTMODE;
791 	re_init(sc);
792 	sc->re_flags |= RE_F_LINKED;
793 	if (!RE_IS_8139CP(sc))
794 		phyaddr = 1;
795 	else
796 		phyaddr = 0;
797 
798 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_RESET);
799 	for (i = 0; i < RE_TIMEOUT; i++) {
800 		status = re_miibus_readreg(sc->re_dev, phyaddr, MII_BMCR);
801 		if (!(status & BMCR_RESET))
802 			break;
803 	}
804 
805 	re_miibus_writereg(sc->re_dev, phyaddr, MII_BMCR, BMCR_LOOP);
806 	CSR_WRITE_2(sc, RE_ISR, RE_INTRS_DIAG);
807 
808 	DELAY(100000);
809 
810 	/* Put some data in the mbuf */
811 
812 	eh = mtod(m0, struct ether_header *);
813 	bcopy (dst, eh->ether_dhost, ETHER_ADDR_LEN);
814 	bcopy (src, eh->ether_shost, ETHER_ADDR_LEN);
815 	eh->ether_type = htons(ETHERTYPE_IP);
816 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
817 
818 	/*
819 	 * Queue the packet, start transmission.
820 	 * Note: ifq_handoff() ultimately calls re_start() for us.
821 	 */
822 
823 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
824 	error = ifq_handoff(ifp, m0, NULL);
825 	if (error) {
826 		m0 = NULL;
827 		goto done;
828 	}
829 	m0 = NULL;
830 
831 	/* Wait for it to propagate through the chip */
832 
833 	DELAY(100000);
834 	for (i = 0; i < RE_TIMEOUT; i++) {
835 		status = CSR_READ_2(sc, RE_ISR);
836 		CSR_WRITE_2(sc, RE_ISR, status);
837 		if ((status & (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK)) ==
838 		    (RE_ISR_TIMEOUT_EXPIRED|RE_ISR_RX_OK))
839 			break;
840 		DELAY(10);
841 	}
842 
843 	if (i == RE_TIMEOUT) {
844 		if_printf(ifp, "diagnostic failed to receive packet "
845 			  "in loopback mode\n");
846 		error = EIO;
847 		goto done;
848 	}
849 
850 	/*
851 	 * The packet should have been dumped into the first
852 	 * entry in the RX DMA ring. Grab it from there.
853 	 */
854 
855 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
856 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
857 	bus_dmamap_sync(sc->re_ldata.re_rx_mtag, sc->re_ldata.re_rx_dmamap[0],
858 			BUS_DMASYNC_POSTREAD);
859 	bus_dmamap_unload(sc->re_ldata.re_rx_mtag,
860 			  sc->re_ldata.re_rx_dmamap[0]);
861 
862 	m0 = sc->re_ldata.re_rx_mbuf[0];
863 	sc->re_ldata.re_rx_mbuf[0] = NULL;
864 	eh = mtod(m0, struct ether_header *);
865 
866 	cur_rx = &sc->re_ldata.re_rx_list[0];
867 	total_len = RE_RXBYTES(cur_rx);
868 	rxstat = le32toh(cur_rx->re_cmdstat);
869 
870 	if (total_len != ETHER_MIN_LEN) {
871 		if_printf(ifp, "diagnostic failed, received short packet\n");
872 		error = EIO;
873 		goto done;
874 	}
875 
876 	/* Test that the received packet data matches what we sent. */
877 
878 	if (bcmp(eh->ether_dhost, dst, ETHER_ADDR_LEN) ||
879 	    bcmp(eh->ether_shost, &src, ETHER_ADDR_LEN) ||
880 	    be16toh(eh->ether_type) != ETHERTYPE_IP) {
881 		if_printf(ifp, "WARNING, DMA FAILURE!\n");
882 		if_printf(ifp, "expected TX data: %6D/%6D/0x%x\n",
883 		    dst, ":", src, ":", ETHERTYPE_IP);
884 		if_printf(ifp, "received RX data: %6D/%6D/0x%x\n",
885 		    eh->ether_dhost, ":",  eh->ether_shost, ":",
886 		    ntohs(eh->ether_type));
887 		if_printf(ifp, "You may have a defective 32-bit NIC plugged "
888 		    "into a 64-bit PCI slot.\n");
889 		if_printf(ifp, "Please re-install the NIC in a 32-bit slot "
890 		    "for proper operation.\n");
891 		if_printf(ifp, "Read the re(4) man page for more details.\n");
892 		error = EIO;
893 	}
894 
895 done:
896 	/* Turn interface off, release resources */
897 
898 	sc->re_flags &= ~(RE_F_LINKED | RE_F_TESTMODE);
899 	ifp->if_flags &= ~IFF_PROMISC;
900 	re_stop(sc);
901 	if (m0 != NULL)
902 		m_freem(m0);
903 
904 	return (error);
905 }
906 #endif	/* RE_DIAG */
907 
908 /*
909  * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
910  * IDs against our list and return a device name if we find a match.
911  */
912 static int
913 re_probe(device_t dev)
914 {
915 	const struct re_type *t;
916 	const struct re_hwrev *hw_rev;
917 	struct re_softc *sc;
918 	int rid;
919 	uint32_t hwrev, macmode, txcfg;
920 	uint16_t vendor, product;
921 
922 	vendor = pci_get_vendor(dev);
923 	product = pci_get_device(dev);
924 
925 	/*
926 	 * Only attach to rev.3 of the Linksys EG1032 adapter.
927 	 * Rev.2 is supported by sk(4).
928 	 */
929 	if (vendor == PCI_VENDOR_LINKSYS &&
930 	    product == PCI_PRODUCT_LINKSYS_EG1032 &&
931 	    pci_get_subdevice(dev) != PCI_SUBDEVICE_LINKSYS_EG1032_REV3)
932 		return ENXIO;
933 
934 	if (vendor == PCI_VENDOR_REALTEK &&
935 	    product == PCI_PRODUCT_REALTEK_RT8139 &&
936 	    pci_get_revid(dev) != PCI_REVID_REALTEK_RT8139CP) {
937 		/* Poor 8139 */
938 		return ENXIO;
939 	}
940 
941 	for (t = re_devs; t->re_name != NULL; t++) {
942 		if (product == t->re_did && vendor == t->re_vid)
943 			break;
944 	}
945 
946 	/*
947 	 * Check if we found a RealTek device.
948 	 */
949 	if (t->re_name == NULL)
950 		return ENXIO;
951 
952 	/*
953 	 * Temporarily map the I/O space so we can read the chip ID register.
954 	 */
955 	sc = kmalloc(sizeof(*sc), M_TEMP, M_WAITOK | M_ZERO);
956 	rid = RE_PCI_LOIO;
957 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
958 					    RF_ACTIVE);
959 	if (sc->re_res == NULL) {
960 		device_printf(dev, "couldn't map ports/memory\n");
961 		kfree(sc, M_TEMP);
962 		return ENXIO;
963 	}
964 
965 	sc->re_btag = rman_get_bustag(sc->re_res);
966 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
967 
968 	txcfg = CSR_READ_4(sc, RE_TXCFG);
969 	hwrev = txcfg & RE_TXCFG_HWREV;
970 	macmode = txcfg & RE_TXCFG_MACMODE;
971 	bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO, sc->re_res);
972 	kfree(sc, M_TEMP);
973 
974 	/*
975 	 * and continue matching for the specific chip...
976 	 */
977 	for (hw_rev = re_hwrevs; hw_rev->re_hwrev != RE_HWREV_NULL; hw_rev++) {
978 		if (hw_rev->re_hwrev == hwrev) {
979 			sc = device_get_softc(dev);
980 
981 			sc->re_hwrev = hw_rev->re_hwrev;
982 			sc->re_macver = hw_rev->re_macver;
983 			sc->re_caps = hw_rev->re_caps;
984 			sc->re_maxmtu = hw_rev->re_maxmtu;
985 
986 			/*
987 			 * Apply chip property fixup
988 			 */
989 			switch (sc->re_hwrev) {
990 			case RE_HWREV_8101E1:
991 			case RE_HWREV_8101E2:
992 				if (macmode == 0)
993 					sc->re_macver = RE_MACVER_11;
994 				else if (macmode == 0x200000)
995 					sc->re_macver = RE_MACVER_12;
996 				break;
997 			case RE_HWREV_8102E:
998 			case RE_HWREV_8102EL:
999 				if (macmode == 0)
1000 					sc->re_macver = RE_MACVER_13;
1001 				else if (macmode == 0x100000)
1002 					sc->re_macver = RE_MACVER_14;
1003 				break;
1004 			case RE_HWREV_8168B2:
1005 			case RE_HWREV_8168B3:
1006 				if (macmode == 0)
1007 					sc->re_macver = RE_MACVER_22;
1008 				break;
1009 			case RE_HWREV_8168C:
1010 				if (macmode == 0)
1011 					sc->re_macver = RE_MACVER_24;
1012 				else if (macmode == 0x200000)
1013 					sc->re_macver = RE_MACVER_25;
1014 				else if (macmode == 0x300000)
1015 					sc->re_macver = RE_MACVER_27;
1016 				break;
1017 			case RE_HWREV_8168CP:
1018 				if (macmode == 0)
1019 					sc->re_macver = RE_MACVER_26;
1020 				else if (macmode == 0x100000)
1021 					sc->re_macver = RE_MACVER_28;
1022 				break;
1023 			}
1024 			if (pci_is_pcie(dev))
1025 				sc->re_caps |= RE_C_PCIE;
1026 
1027 			device_set_desc(dev, t->re_name);
1028 			return 0;
1029 		}
1030 	}
1031 
1032 	if (bootverbose) {
1033 		device_printf(dev, "unknown hwrev 0x%08x, macmode 0x%08x\n",
1034 			      hwrev, macmode);
1035 	}
1036 	return ENXIO;
1037 }
1038 
1039 static void
1040 re_dma_map_desc(void *xarg, bus_dma_segment_t *segs, int nsegs,
1041 		bus_size_t mapsize, int error)
1042 {
1043 	struct re_dmaload_arg *arg = xarg;
1044 	int i;
1045 
1046 	if (error)
1047 		return;
1048 
1049 	if (nsegs > arg->re_nsegs) {
1050 		arg->re_nsegs = 0;
1051 		return;
1052 	}
1053 
1054 	arg->re_nsegs = nsegs;
1055 	for (i = 0; i < nsegs; ++i)
1056 		arg->re_segs[i] = segs[i];
1057 }
1058 
1059 /*
1060  * Map a single buffer address.
1061  */
1062 
1063 static void
1064 re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1065 {
1066 	uint32_t *addr;
1067 
1068 	if (error)
1069 		return;
1070 
1071 	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
1072 	addr = arg;
1073 	*addr = segs->ds_addr;
1074 }
1075 
1076 static int
1077 re_allocmem(device_t dev)
1078 {
1079 	struct re_softc *sc = device_get_softc(dev);
1080 	int error, i;
1081 
1082 	/*
1083 	 * Allocate list data
1084 	 */
1085 	sc->re_ldata.re_tx_mbuf =
1086 	kmalloc(sc->re_tx_desc_cnt * sizeof(struct mbuf *),
1087 		M_DEVBUF, M_ZERO | M_WAITOK);
1088 
1089 	sc->re_ldata.re_rx_mbuf =
1090 	kmalloc(sc->re_rx_desc_cnt * sizeof(struct mbuf *),
1091 		M_DEVBUF, M_ZERO | M_WAITOK);
1092 
1093 	sc->re_ldata.re_rx_paddr =
1094 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_addr_t),
1095 		M_DEVBUF, M_ZERO | M_WAITOK);
1096 
1097 	sc->re_ldata.re_tx_dmamap =
1098 	kmalloc(sc->re_tx_desc_cnt * sizeof(bus_dmamap_t),
1099 		M_DEVBUF, M_ZERO | M_WAITOK);
1100 
1101 	sc->re_ldata.re_rx_dmamap =
1102 	kmalloc(sc->re_rx_desc_cnt * sizeof(bus_dmamap_t),
1103 		M_DEVBUF, M_ZERO | M_WAITOK);
1104 
1105 	/*
1106 	 * Allocate the parent bus DMA tag appropriate for PCI.
1107 	 */
1108 	error = bus_dma_tag_create(NULL,	/* parent */
1109 			1, 0,			/* alignment, boundary */
1110 			BUS_SPACE_MAXADDR,	/* lowaddr */
1111 			BUS_SPACE_MAXADDR,	/* highaddr */
1112 			NULL, NULL,		/* filter, filterarg */
1113 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
1114 			0,			/* nsegments */
1115 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
1116 			0,			/* flags */
1117 			&sc->re_parent_tag);
1118 	if (error) {
1119 		device_printf(dev, "could not allocate parent dma tag\n");
1120 		return error;
1121 	}
1122 
1123 	/* Allocate tag for TX descriptor list. */
1124 	error = bus_dma_tag_create(sc->re_parent_tag,
1125 			RE_RING_ALIGN, 0,
1126 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1127 			NULL, NULL,
1128 			RE_TX_LIST_SZ(sc), 1, RE_TX_LIST_SZ(sc),
1129 			0, &sc->re_ldata.re_tx_list_tag);
1130 	if (error) {
1131 		device_printf(dev, "could not allocate TX ring dma tag\n");
1132 		return(error);
1133 	}
1134 
1135 	/* Allocate DMA'able memory for the TX ring */
1136         error = bus_dmamem_alloc(sc->re_ldata.re_tx_list_tag,
1137 			(void **)&sc->re_ldata.re_tx_list,
1138 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1139 			&sc->re_ldata.re_tx_list_map);
1140         if (error) {
1141 		device_printf(dev, "could not allocate TX ring\n");
1142 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1143 		sc->re_ldata.re_tx_list_tag = NULL;
1144                 return(error);
1145 	}
1146 
1147 	/* Load the map for the TX ring. */
1148 	error = bus_dmamap_load(sc->re_ldata.re_tx_list_tag,
1149 			sc->re_ldata.re_tx_list_map,
1150 			sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc),
1151 			re_dma_map_addr, &sc->re_ldata.re_tx_list_addr,
1152 			BUS_DMA_NOWAIT);
1153 	if (error) {
1154 		device_printf(dev, "could not get address of TX ring\n");
1155 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1156 				sc->re_ldata.re_tx_list,
1157 				sc->re_ldata.re_tx_list_map);
1158 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1159 		sc->re_ldata.re_tx_list_tag = NULL;
1160 		return(error);
1161 	}
1162 
1163 	/* Allocate tag for RX descriptor list. */
1164 	error = bus_dma_tag_create(sc->re_parent_tag,
1165 			RE_RING_ALIGN, 0,
1166 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1167 			NULL, NULL,
1168 			RE_RX_LIST_SZ(sc), 1, RE_RX_LIST_SZ(sc),
1169 			0, &sc->re_ldata.re_rx_list_tag);
1170 	if (error) {
1171 		device_printf(dev, "could not allocate RX ring dma tag\n");
1172 		return(error);
1173 	}
1174 
1175 	/* Allocate DMA'able memory for the RX ring */
1176         error = bus_dmamem_alloc(sc->re_ldata.re_rx_list_tag,
1177 			(void **)&sc->re_ldata.re_rx_list,
1178 			BUS_DMA_WAITOK | BUS_DMA_ZERO,
1179 			&sc->re_ldata.re_rx_list_map);
1180         if (error) {
1181 		device_printf(dev, "could not allocate RX ring\n");
1182 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1183 		sc->re_ldata.re_rx_list_tag = NULL;
1184                 return(error);
1185 	}
1186 
1187 	/* Load the map for the RX ring. */
1188 	error = bus_dmamap_load(sc->re_ldata.re_rx_list_tag,
1189 			sc->re_ldata.re_rx_list_map,
1190 			sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc),
1191 			re_dma_map_addr, &sc->re_ldata.re_rx_list_addr,
1192 			BUS_DMA_NOWAIT);
1193 	if (error) {
1194 		device_printf(dev, "could not get address of RX ring\n");
1195 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1196 				sc->re_ldata.re_rx_list,
1197 				sc->re_ldata.re_rx_list_map);
1198 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1199 		sc->re_ldata.re_rx_list_tag = NULL;
1200 		return(error);
1201 	}
1202 
1203 	/* Allocate maps for TX mbufs. */
1204 	error = bus_dma_tag_create(sc->re_parent_tag,
1205 			1, 0,
1206 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1207 			NULL, NULL,
1208 			RE_FRAMELEN_MAX, RE_MAXSEGS, MCLBYTES,
1209 			BUS_DMA_ALLOCNOW,
1210 			&sc->re_ldata.re_tx_mtag);
1211 	if (error) {
1212 		device_printf(dev, "could not allocate TX buf dma tag\n");
1213 		return(error);
1214 	}
1215 
1216 	/* Create DMA maps for TX buffers */
1217 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
1218 		error = bus_dmamap_create(sc->re_ldata.re_tx_mtag, 0,
1219 				&sc->re_ldata.re_tx_dmamap[i]);
1220 		if (error) {
1221 			device_printf(dev, "can't create DMA map for TX buf\n");
1222 			re_freebufmem(sc, i, 0);
1223 			return(error);
1224 		}
1225 	}
1226 
1227 	/* Allocate maps for RX mbufs. */
1228 	error = bus_dma_tag_create(sc->re_parent_tag,
1229 			RE_RXBUF_ALIGN, 0,
1230 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1231 			NULL, NULL,
1232 			MCLBYTES, 1, MCLBYTES,
1233 			BUS_DMA_ALLOCNOW,
1234 			&sc->re_ldata.re_rx_mtag);
1235 	if (error) {
1236 		device_printf(dev, "could not allocate RX buf dma tag\n");
1237 		return(error);
1238 	}
1239 
1240 	/* Create spare DMA map for RX */
1241 	error = bus_dmamap_create(sc->re_ldata.re_rx_mtag, 0,
1242 			&sc->re_ldata.re_rx_spare);
1243 	if (error) {
1244 		device_printf(dev, "can't create spare DMA map for RX\n");
1245 		bus_dma_tag_destroy(sc->re_ldata.re_rx_mtag);
1246 		sc->re_ldata.re_rx_mtag = NULL;
1247 		return error;
1248 	}
1249 
1250 	/* Create DMA maps for RX buffers */
1251 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1252 		error = bus_dmamap_create(sc->re_ldata.re_rx_mtag, 0,
1253 				&sc->re_ldata.re_rx_dmamap[i]);
1254 		if (error) {
1255 			device_printf(dev, "can't create DMA map for RX buf\n");
1256 			re_freebufmem(sc, sc->re_tx_desc_cnt, i);
1257 			return(error);
1258 		}
1259 	}
1260 
1261 	/* Create jumbo buffer pool for RX if required */
1262 	if (sc->re_caps & RE_C_CONTIGRX) {
1263 		error = re_jpool_alloc(sc);
1264 		if (error) {
1265 			re_jpool_free(sc);
1266 			/* Disable jumbo frame support */
1267 			sc->re_maxmtu = ETHERMTU;
1268 		}
1269 	}
1270 	return(0);
1271 }
1272 
1273 static void
1274 re_freebufmem(struct re_softc *sc, int tx_cnt, int rx_cnt)
1275 {
1276 	int i;
1277 
1278 	/* Destroy all the RX and TX buffer maps */
1279 	if (sc->re_ldata.re_tx_mtag) {
1280 		for (i = 0; i < tx_cnt; i++) {
1281 			bus_dmamap_destroy(sc->re_ldata.re_tx_mtag,
1282 					   sc->re_ldata.re_tx_dmamap[i]);
1283 		}
1284 		bus_dma_tag_destroy(sc->re_ldata.re_tx_mtag);
1285 		sc->re_ldata.re_tx_mtag = NULL;
1286 	}
1287 
1288 	if (sc->re_ldata.re_rx_mtag) {
1289 		for (i = 0; i < rx_cnt; i++) {
1290 			bus_dmamap_destroy(sc->re_ldata.re_rx_mtag,
1291 					   sc->re_ldata.re_rx_dmamap[i]);
1292 		}
1293 		bus_dmamap_destroy(sc->re_ldata.re_rx_mtag,
1294 				   sc->re_ldata.re_rx_spare);
1295 		bus_dma_tag_destroy(sc->re_ldata.re_rx_mtag);
1296 		sc->re_ldata.re_rx_mtag = NULL;
1297 	}
1298 }
1299 
1300 static void
1301 re_freemem(device_t dev)
1302 {
1303 	struct re_softc *sc = device_get_softc(dev);
1304 
1305 	/* Unload and free the RX DMA ring memory and map */
1306 	if (sc->re_ldata.re_rx_list_tag) {
1307 		bus_dmamap_unload(sc->re_ldata.re_rx_list_tag,
1308 				  sc->re_ldata.re_rx_list_map);
1309 		bus_dmamem_free(sc->re_ldata.re_rx_list_tag,
1310 				sc->re_ldata.re_rx_list,
1311 				sc->re_ldata.re_rx_list_map);
1312 		bus_dma_tag_destroy(sc->re_ldata.re_rx_list_tag);
1313 	}
1314 
1315 	/* Unload and free the TX DMA ring memory and map */
1316 	if (sc->re_ldata.re_tx_list_tag) {
1317 		bus_dmamap_unload(sc->re_ldata.re_tx_list_tag,
1318 				  sc->re_ldata.re_tx_list_map);
1319 		bus_dmamem_free(sc->re_ldata.re_tx_list_tag,
1320 				sc->re_ldata.re_tx_list,
1321 				sc->re_ldata.re_tx_list_map);
1322 		bus_dma_tag_destroy(sc->re_ldata.re_tx_list_tag);
1323 	}
1324 
1325 	/* Free RX/TX buf DMA stuffs */
1326 	re_freebufmem(sc, sc->re_tx_desc_cnt, sc->re_rx_desc_cnt);
1327 
1328 	/* Unload and free the stats buffer and map */
1329 	if (sc->re_ldata.re_stag) {
1330 		bus_dmamap_unload(sc->re_ldata.re_stag,
1331 				  sc->re_ldata.re_rx_list_map);
1332 		bus_dmamem_free(sc->re_ldata.re_stag,
1333 				sc->re_ldata.re_stats,
1334 				sc->re_ldata.re_smap);
1335 		bus_dma_tag_destroy(sc->re_ldata.re_stag);
1336 	}
1337 
1338 	if (sc->re_caps & RE_C_CONTIGRX)
1339 		re_jpool_free(sc);
1340 
1341 	if (sc->re_parent_tag)
1342 		bus_dma_tag_destroy(sc->re_parent_tag);
1343 
1344 	if (sc->re_ldata.re_tx_mbuf != NULL)
1345 		kfree(sc->re_ldata.re_tx_mbuf, M_DEVBUF);
1346 	if (sc->re_ldata.re_rx_mbuf != NULL)
1347 		kfree(sc->re_ldata.re_rx_mbuf, M_DEVBUF);
1348 	if (sc->re_ldata.re_rx_paddr != NULL)
1349 		kfree(sc->re_ldata.re_rx_paddr, M_DEVBUF);
1350 	if (sc->re_ldata.re_tx_dmamap != NULL)
1351 		kfree(sc->re_ldata.re_tx_dmamap, M_DEVBUF);
1352 	if (sc->re_ldata.re_rx_dmamap != NULL)
1353 		kfree(sc->re_ldata.re_rx_dmamap, M_DEVBUF);
1354 }
1355 
1356 /*
1357  * Attach the interface. Allocate softc structures, do ifmedia
1358  * setup and ethernet/BPF attach.
1359  */
1360 static int
1361 re_attach(device_t dev)
1362 {
1363 	struct re_softc	*sc = device_get_softc(dev);
1364 	struct ifnet *ifp;
1365 	uint8_t eaddr[ETHER_ADDR_LEN];
1366 	int error = 0, rid, qlen;
1367 
1368 	callout_init(&sc->re_timer);
1369 	sc->re_dev = dev;
1370 
1371 	if (RE_IS_8139CP(sc)) {
1372 		sc->re_rx_desc_cnt = RE_RX_DESC_CNT_8139CP;
1373 		sc->re_tx_desc_cnt = RE_TX_DESC_CNT_8139CP;
1374 	} else {
1375 		sc->re_rx_desc_cnt = re_rx_desc_count;
1376 		if (sc->re_rx_desc_cnt > RE_RX_DESC_CNT_MAX)
1377 			sc->re_rx_desc_cnt = RE_RX_DESC_CNT_MAX;
1378 
1379 		sc->re_tx_desc_cnt = re_tx_desc_count;
1380 		if (sc->re_tx_desc_cnt > RE_TX_DESC_CNT_MAX)
1381 			sc->re_tx_desc_cnt = RE_TX_DESC_CNT_MAX;
1382 	}
1383 
1384 	qlen = RE_IFQ_MAXLEN;
1385 	if (sc->re_tx_desc_cnt > qlen)
1386 		qlen = sc->re_tx_desc_cnt;
1387 
1388 	sc->re_rxbuf_size = MCLBYTES;
1389 	sc->re_newbuf = re_newbuf_std;
1390 
1391 	sc->re_tx_time = 5;		/* 125us */
1392 	sc->re_rx_time = 2;		/* 50us */
1393 	if (sc->re_caps & RE_C_PCIE)
1394 		sc->re_sim_time = 75;	/* 75us */
1395 	else
1396 		sc->re_sim_time = 125;	/* 125us */
1397 	if (!RE_IS_8139CP(sc)) {
1398 		/* simulated interrupt moderation */
1399 		sc->re_imtype = RE_IMTYPE_SIM;
1400 	} else {
1401 		sc->re_imtype = RE_IMTYPE_NONE;
1402 	}
1403 	re_config_imtype(sc, sc->re_imtype);
1404 
1405 	sysctl_ctx_init(&sc->re_sysctl_ctx);
1406 	sc->re_sysctl_tree = SYSCTL_ADD_NODE(&sc->re_sysctl_ctx,
1407 					     SYSCTL_STATIC_CHILDREN(_hw),
1408 					     OID_AUTO,
1409 					     device_get_nameunit(dev),
1410 					     CTLFLAG_RD, 0, "");
1411 	if (sc->re_sysctl_tree == NULL) {
1412 		device_printf(dev, "can't add sysctl node\n");
1413 		error = ENXIO;
1414 		goto fail;
1415 	}
1416 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1417 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1418 		       "rx_desc_count", CTLFLAG_RD, &sc->re_rx_desc_cnt,
1419 		       0, "RX desc count");
1420 	SYSCTL_ADD_INT(&sc->re_sysctl_ctx,
1421 		       SYSCTL_CHILDREN(sc->re_sysctl_tree), OID_AUTO,
1422 		       "tx_desc_count", CTLFLAG_RD, &sc->re_tx_desc_cnt,
1423 		       0, "TX desc count");
1424 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1425 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1426 			OID_AUTO, "sim_time",
1427 			CTLTYPE_INT | CTLFLAG_RW,
1428 			sc, 0, re_sysctl_simtime, "I",
1429 			"Simulated interrupt moderation time (usec).");
1430 	SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1431 			SYSCTL_CHILDREN(sc->re_sysctl_tree),
1432 			OID_AUTO, "imtype",
1433 			CTLTYPE_INT | CTLFLAG_RW,
1434 			sc, 0, re_sysctl_imtype, "I",
1435 			"Interrupt moderation type -- "
1436 			"0:disable, 1:simulated, "
1437 			"2:hardware(if supported)");
1438 	if (sc->re_caps & RE_C_HWIM) {
1439 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1440 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1441 				OID_AUTO, "hw_rxtime",
1442 				CTLTYPE_INT | CTLFLAG_RW,
1443 				sc, 0, re_sysctl_rxtime, "I",
1444 				"Hardware interrupt moderation time "
1445 				"(unit: 25usec).");
1446 		SYSCTL_ADD_PROC(&sc->re_sysctl_ctx,
1447 				SYSCTL_CHILDREN(sc->re_sysctl_tree),
1448 				OID_AUTO, "hw_txtime",
1449 				CTLTYPE_INT | CTLFLAG_RW,
1450 				sc, 0, re_sysctl_txtime, "I",
1451 				"Hardware interrupt moderation time "
1452 				"(unit: 25usec).");
1453 	}
1454 
1455 #ifndef BURN_BRIDGES
1456 	/*
1457 	 * Handle power management nonsense.
1458 	 */
1459 
1460 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1461 		uint32_t membase, irq;
1462 
1463 		/* Save important PCI config data. */
1464 		membase = pci_read_config(dev, RE_PCI_LOMEM, 4);
1465 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1466 
1467 		/* Reset the power state. */
1468 		device_printf(dev, "chip is in D%d power mode "
1469 		    "-- setting to D0\n", pci_get_powerstate(dev));
1470 
1471 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1472 
1473 		/* Restore PCI config data. */
1474 		pci_write_config(dev, RE_PCI_LOMEM, membase, 4);
1475 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1476 	}
1477 #endif
1478 	/*
1479 	 * Map control/status registers.
1480 	 */
1481 	pci_enable_busmaster(dev);
1482 
1483 	rid = RE_PCI_LOIO;
1484 	sc->re_res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid,
1485 					    RF_ACTIVE);
1486 
1487 	if (sc->re_res == NULL) {
1488 		device_printf(dev, "couldn't map ports\n");
1489 		error = ENXIO;
1490 		goto fail;
1491 	}
1492 
1493 	sc->re_btag = rman_get_bustag(sc->re_res);
1494 	sc->re_bhandle = rman_get_bushandle(sc->re_res);
1495 
1496 	/* Allocate interrupt */
1497 	rid = 0;
1498 	sc->re_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1499 					    RF_SHAREABLE | RF_ACTIVE);
1500 
1501 	if (sc->re_irq == NULL) {
1502 		device_printf(dev, "couldn't map interrupt\n");
1503 		error = ENXIO;
1504 		goto fail;
1505 	}
1506 
1507 	/* Reset the adapter. */
1508 	re_reset(sc, 0);
1509 
1510 	if (RE_IS_8139CP(sc)) {
1511 		sc->re_bus_speed = 33; /* XXX */
1512 	} else if (sc->re_caps & RE_C_PCIE) {
1513 		sc->re_bus_speed = 125;
1514 	} else {
1515 		uint8_t cfg2;
1516 
1517 		cfg2 = CSR_READ_1(sc, RE_CFG2);
1518 		switch (cfg2 & RE_CFG2_PCICLK_MASK) {
1519 		case RE_CFG2_PCICLK_33MHZ:
1520 			sc->re_bus_speed = 33;
1521 			break;
1522 		case RE_CFG2_PCICLK_66MHZ:
1523 			sc->re_bus_speed = 66;
1524 			break;
1525 		default:
1526 			device_printf(dev, "unknown bus speed, assume 33MHz\n");
1527 			sc->re_bus_speed = 33;
1528 			break;
1529 		}
1530 		if (cfg2 & RE_CFG2_PCI64)
1531 			sc->re_caps |= RE_C_PCI64;
1532 	}
1533 	device_printf(dev, "Hardware rev. 0x%08x; MAC ver. 0x%02x; "
1534 		      "PCI%s %dMHz\n",
1535 		      sc->re_hwrev, sc->re_macver,
1536 		      (sc->re_caps & RE_C_PCIE) ?
1537 		      "-E" : ((sc->re_caps & RE_C_PCI64) ? "64" : "32"),
1538 		      sc->re_bus_speed);
1539 
1540 	/*
1541 	 * NOTE:
1542 	 * DO NOT try to adjust config1 and config5 which was spotted in
1543 	 * Realtek's Linux drivers.  It will _permanently_ damage certain
1544 	 * cards EEPROM, e.g. one of my 8168B (0x38000000) card ...
1545 	 */
1546 
1547 	re_get_eaddr(sc, eaddr);
1548 
1549 	if (!RE_IS_8139CP(sc)) {
1550 		/* Set RX length mask */
1551 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
1552 		sc->re_txstart = RE_GTXSTART;
1553 	} else {
1554 		/* Set RX length mask */
1555 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
1556 		sc->re_txstart = RE_TXSTART;
1557 	}
1558 
1559 	/* Allocate DMA stuffs */
1560 	error = re_allocmem(dev);
1561 	if (error)
1562 		goto fail;
1563 
1564 	/*
1565 	 * Apply some magic PCI settings from Realtek ...
1566 	 */
1567 	if (RE_IS_8169(sc)) {
1568 		CSR_WRITE_1(sc, 0x82, 1);
1569 		pci_write_config(dev, PCIR_CACHELNSZ, 0x8, 1);
1570 	}
1571 	pci_write_config(dev, PCIR_LATTIMER, 0x40, 1);
1572 
1573 	if (sc->re_caps & RE_C_MAC2) {
1574 		/*
1575 		 * Following part is extracted from Realtek BSD driver v176.
1576 		 * However, this does _not_ make much/any sense:
1577 		 * 8168C's PCI Express device control is located at 0x78,
1578 		 * so the reading from 0x79 (higher part of 0x78) and setting
1579 		 * the 4~6bits intend to enlarge the "max read request size"
1580 		 * (we will do it).  The content of the rest part of this
1581 		 * register is not meaningful to other PCI registers, so
1582 		 * writing the value to 0x54 could be completely wrong.
1583 		 * 0x80 is the lower part of PCI Express device status, non-
1584 		 * reserved bits are RW1C, writing 0 to them will not have
1585 		 * any effect at all.
1586 		 */
1587 #ifdef foo
1588 		uint8_t val;
1589 
1590 		val = pci_read_config(dev, 0x79, 1);
1591 		val = (val & ~0x70) | 0x50;
1592 		pci_write_config(dev, 0x54, val, 1);
1593 		pci_write_config(dev, 0x80, 0, 1);
1594 #endif
1595 	}
1596 
1597 	/*
1598 	 * Apply some PHY fixup from Realtek ...
1599 	 */
1600 	if (sc->re_hwrev == RE_HWREV_8110S) {
1601 		CSR_WRITE_1(sc, 0x82, 1);
1602 		re_miibus_writereg(dev, 1, 0xb, 0);
1603 	}
1604 	if (sc->re_caps & RE_C_PHYPMGT) {
1605 		/* Power up PHY */
1606 		re_miibus_writereg(dev, 1, 0x1f, 0);
1607 		re_miibus_writereg(dev, 1, 0xe, 0);
1608 	}
1609 
1610 	/* Do MII setup */
1611 	if (mii_phy_probe(dev, &sc->re_miibus,
1612 	    re_ifmedia_upd, re_ifmedia_sts)) {
1613 		device_printf(dev, "MII without any phy!\n");
1614 		error = ENXIO;
1615 		goto fail;
1616 	}
1617 
1618 	ifp = &sc->arpcom.ac_if;
1619 	ifp->if_softc = sc;
1620 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1621 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1622 	ifp->if_ioctl = re_ioctl;
1623 	ifp->if_start = re_start;
1624 #ifdef DEVICE_POLLING
1625 	ifp->if_poll = re_poll;
1626 #endif
1627 	ifp->if_watchdog = re_watchdog;
1628 	ifp->if_init = re_init;
1629 	if (!RE_IS_8139CP(sc)) /* XXX */
1630 		ifp->if_baudrate = 1000000000;
1631 	else
1632 		ifp->if_baudrate = 100000000;
1633 	ifq_set_maxlen(&ifp->if_snd, qlen);
1634 	ifq_set_ready(&ifp->if_snd);
1635 
1636 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1637 	if (sc->re_caps & RE_C_HWCSUM)
1638 		ifp->if_capabilities |= IFCAP_HWCSUM;
1639 
1640 	ifp->if_capenable = ifp->if_capabilities;
1641 	if (ifp->if_capabilities & IFCAP_HWCSUM)
1642 		ifp->if_hwassist = RE_CSUM_FEATURES;
1643 	else
1644 		ifp->if_hwassist = 0;
1645 
1646 	/*
1647 	 * Call MI attach routine.
1648 	 */
1649 	ether_ifattach(ifp, eaddr, NULL);
1650 
1651 #ifdef RE_DIAG
1652 	/*
1653 	 * Perform hardware diagnostic on the original RTL8169.
1654 	 * Some 32-bit cards were incorrectly wired and would
1655 	 * malfunction if plugged into a 64-bit slot.
1656 	 */
1657 	if (sc->re_hwrev == RE_HWREV_8169) {
1658 		lwkt_serialize_enter(ifp->if_serializer);
1659 		error = re_diag(sc);
1660 		lwkt_serialize_exit(ifp->if_serializer);
1661 
1662 		if (error) {
1663 			device_printf(dev, "hardware diagnostic failure\n");
1664 			ether_ifdetach(ifp);
1665 			goto fail;
1666 		}
1667 	}
1668 #endif	/* RE_DIAG */
1669 
1670 	/* Hook interrupt last to avoid having to lock softc */
1671 	error = bus_setup_intr(dev, sc->re_irq, INTR_MPSAFE, re_intr, sc,
1672 			       &sc->re_intrhand, ifp->if_serializer);
1673 
1674 	if (error) {
1675 		device_printf(dev, "couldn't set up irq\n");
1676 		ether_ifdetach(ifp);
1677 		goto fail;
1678 	}
1679 
1680 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->re_irq));
1681 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
1682 
1683 fail:
1684 	if (error)
1685 		re_detach(dev);
1686 
1687 	return (error);
1688 }
1689 
1690 /*
1691  * Shutdown hardware and free up resources. This can be called any
1692  * time after the mutex has been initialized. It is called in both
1693  * the error case in attach and the normal detach case so it needs
1694  * to be careful about only freeing resources that have actually been
1695  * allocated.
1696  */
1697 static int
1698 re_detach(device_t dev)
1699 {
1700 	struct re_softc *sc = device_get_softc(dev);
1701 	struct ifnet *ifp = &sc->arpcom.ac_if;
1702 
1703 	/* These should only be active if attach succeeded */
1704 	if (device_is_attached(dev)) {
1705 		lwkt_serialize_enter(ifp->if_serializer);
1706 		re_stop(sc);
1707 		bus_teardown_intr(dev, sc->re_irq, sc->re_intrhand);
1708 		lwkt_serialize_exit(ifp->if_serializer);
1709 
1710 		ether_ifdetach(ifp);
1711 	}
1712 	if (sc->re_miibus)
1713 		device_delete_child(dev, sc->re_miibus);
1714 	bus_generic_detach(dev);
1715 
1716 	if (sc->re_sysctl_tree != NULL)
1717 		sysctl_ctx_free(&sc->re_sysctl_ctx);
1718 
1719 	if (sc->re_irq)
1720 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->re_irq);
1721 	if (sc->re_res) {
1722 		bus_release_resource(dev, SYS_RES_IOPORT, RE_PCI_LOIO,
1723 				     sc->re_res);
1724 	}
1725 
1726 	/* Free DMA stuffs */
1727 	re_freemem(dev);
1728 
1729 	return(0);
1730 }
1731 
1732 static void
1733 re_setup_rxdesc(struct re_softc *sc, int idx)
1734 {
1735 	bus_addr_t paddr;
1736 	uint32_t cmdstat;
1737 	struct re_desc *d;
1738 
1739 	paddr = sc->re_ldata.re_rx_paddr[idx];
1740 	d = &sc->re_ldata.re_rx_list[idx];
1741 
1742 	d->re_bufaddr_lo = htole32(RE_ADDR_LO(paddr));
1743 	d->re_bufaddr_hi = htole32(RE_ADDR_HI(paddr));
1744 
1745 	cmdstat = sc->re_rxbuf_size | RE_RDESC_CMD_OWN;
1746 	if (idx == (sc->re_rx_desc_cnt - 1))
1747 		cmdstat |= RE_RDESC_CMD_EOR;
1748 	d->re_cmdstat = htole32(cmdstat);
1749 }
1750 
1751 static int
1752 re_newbuf_std(struct re_softc *sc, int idx, int init)
1753 {
1754 	struct re_dmaload_arg arg;
1755 	bus_dma_segment_t seg;
1756 	bus_dmamap_t map;
1757 	struct mbuf *m;
1758 	int error;
1759 
1760 	m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
1761 	if (m == NULL) {
1762 		error = ENOBUFS;
1763 
1764 		if (init) {
1765 			if_printf(&sc->arpcom.ac_if, "m_getcl failed\n");
1766 			return error;
1767 		} else {
1768 			goto back;
1769 		}
1770 	}
1771 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1772 
1773 	/*
1774 	 * NOTE:
1775 	 * re(4) chips need address of the receive buffer to be 8-byte
1776 	 * aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1777 	 */
1778 
1779 	arg.re_nsegs = 1;
1780 	arg.re_segs = &seg;
1781 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_rx_mtag,
1782 				     sc->re_ldata.re_rx_spare, m,
1783 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
1784 	if (error || arg.re_nsegs == 0) {
1785 		if (!error) {
1786 			if_printf(&sc->arpcom.ac_if, "too many segments?!\n");
1787 			bus_dmamap_unload(sc->re_ldata.re_rx_mtag,
1788 					  sc->re_ldata.re_rx_spare);
1789 			error = EFBIG;
1790 		}
1791 		m_freem(m);
1792 
1793 		if (init) {
1794 			if_printf(&sc->arpcom.ac_if, "can't load RX mbuf\n");
1795 			return error;
1796 		} else {
1797 			goto back;
1798 		}
1799 	}
1800 
1801 	if (!init) {
1802 		bus_dmamap_sync(sc->re_ldata.re_rx_mtag,
1803 				sc->re_ldata.re_rx_dmamap[idx],
1804 				BUS_DMASYNC_POSTREAD);
1805 		bus_dmamap_unload(sc->re_ldata.re_rx_mtag,
1806 				  sc->re_ldata.re_rx_dmamap[idx]);
1807 	}
1808 	sc->re_ldata.re_rx_mbuf[idx] = m;
1809 	sc->re_ldata.re_rx_paddr[idx] = seg.ds_addr;
1810 
1811 	map = sc->re_ldata.re_rx_dmamap[idx];
1812 	sc->re_ldata.re_rx_dmamap[idx] = sc->re_ldata.re_rx_spare;
1813 	sc->re_ldata.re_rx_spare = map;
1814 back:
1815 	re_setup_rxdesc(sc, idx);
1816 	return error;
1817 }
1818 
1819 static int
1820 re_newbuf_jumbo(struct re_softc *sc, int idx, int init)
1821 {
1822 	struct mbuf *m;
1823 	struct re_jbuf *jbuf;
1824 	int error = 0;
1825 
1826 	MGETHDR(m, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1827 	if (m == NULL) {
1828 		error = ENOBUFS;
1829 		if (init) {
1830 			if_printf(&sc->arpcom.ac_if, "MGETHDR failed\n");
1831 			return error;
1832 		} else {
1833 			goto back;
1834 		}
1835 	}
1836 
1837 	jbuf = re_jbuf_alloc(sc);
1838 	if (jbuf == NULL) {
1839 		m_freem(m);
1840 
1841 		error = ENOBUFS;
1842 		if (init) {
1843 			if_printf(&sc->arpcom.ac_if, "jpool is empty\n");
1844 			return error;
1845 		} else {
1846 			goto back;
1847 		}
1848 	}
1849 
1850 	m->m_ext.ext_arg = jbuf;
1851 	m->m_ext.ext_buf = jbuf->re_buf;
1852 	m->m_ext.ext_free = re_jbuf_free;
1853 	m->m_ext.ext_ref = re_jbuf_ref;
1854 	m->m_ext.ext_size = sc->re_rxbuf_size;
1855 
1856 	m->m_data = m->m_ext.ext_buf;
1857 	m->m_flags |= M_EXT;
1858 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
1859 
1860 	/*
1861 	 * NOTE:
1862 	 * Some re(4) chips(e.g. RTL8101E) need address of the receive buffer
1863 	 * to be 8-byte aligned, so don't call m_adj(m, ETHER_ALIGN) here.
1864 	 */
1865 
1866 	sc->re_ldata.re_rx_mbuf[idx] = m;
1867 	sc->re_ldata.re_rx_paddr[idx] = jbuf->re_paddr;
1868 back:
1869 	re_setup_rxdesc(sc, idx);
1870 	return error;
1871 }
1872 
1873 static int
1874 re_tx_list_init(struct re_softc *sc)
1875 {
1876 	bzero(sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
1877 
1878 	/* Flush the TX descriptors */
1879 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
1880 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
1881 
1882 	sc->re_ldata.re_tx_prodidx = 0;
1883 	sc->re_ldata.re_tx_considx = 0;
1884 	sc->re_ldata.re_tx_free = sc->re_tx_desc_cnt;
1885 
1886 	return(0);
1887 }
1888 
1889 static int
1890 re_rx_list_init(struct re_softc *sc)
1891 {
1892 	int i, error;
1893 
1894 	bzero(sc->re_ldata.re_rx_list, RE_RX_LIST_SZ(sc));
1895 
1896 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
1897 		error = sc->re_newbuf(sc, i, 1);
1898 		if (error)
1899 			return(error);
1900 	}
1901 
1902 	/* Flush the RX descriptors */
1903 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1904 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
1905 
1906 	sc->re_ldata.re_rx_prodidx = 0;
1907 	sc->re_head = sc->re_tail = NULL;
1908 
1909 	return(0);
1910 }
1911 
1912 #define RE_IP4_PACKET	0x1
1913 #define RE_TCP_PACKET	0x2
1914 #define RE_UDP_PACKET	0x4
1915 
1916 static __inline uint8_t
1917 re_packet_type(struct re_softc *sc, uint32_t rxstat, uint32_t rxctrl)
1918 {
1919 	uint8_t packet_type = 0;
1920 
1921 	if (sc->re_caps & RE_C_MAC2) {
1922 		if (rxctrl & RE_RDESC_CTL_PROTOIP4)
1923 			packet_type |= RE_IP4_PACKET;
1924 	} else {
1925 		if (rxstat & RE_RDESC_STAT_PROTOID)
1926 			packet_type |= RE_IP4_PACKET;
1927 	}
1928 	if (RE_TCPPKT(rxstat))
1929 		packet_type |= RE_TCP_PACKET;
1930 	else if (RE_UDPPKT(rxstat))
1931 		packet_type |= RE_UDP_PACKET;
1932 	return packet_type;
1933 }
1934 
1935 /*
1936  * RX handler for C+ and 8169. For the gigE chips, we support
1937  * the reception of jumbo frames that have been fragmented
1938  * across multiple 2K mbuf cluster buffers.
1939  */
1940 static int
1941 re_rxeof(struct re_softc *sc)
1942 {
1943 	struct ifnet *ifp = &sc->arpcom.ac_if;
1944 	struct mbuf *m;
1945 	struct re_desc 	*cur_rx;
1946 	uint32_t rxstat, rxctrl;
1947 	int i, total_len, rx = 0;
1948 	struct mbuf_chain chain[MAXCPU];
1949 
1950 	/* Invalidate the descriptor memory */
1951 
1952 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
1953 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_POSTREAD);
1954 
1955 	ether_input_chain_init(chain);
1956 
1957 	for (i = sc->re_ldata.re_rx_prodidx;
1958 	     RE_OWN(&sc->re_ldata.re_rx_list[i]) == 0; RE_RXDESC_INC(sc, i)) {
1959 		cur_rx = &sc->re_ldata.re_rx_list[i];
1960 		m = sc->re_ldata.re_rx_mbuf[i];
1961 		total_len = RE_RXBYTES(cur_rx);
1962 		rxstat = le32toh(cur_rx->re_cmdstat);
1963 		rxctrl = le32toh(cur_rx->re_control);
1964 
1965 		rx = 1;
1966 
1967 #ifdef INVARIANTS
1968 		if (sc->re_flags & RE_F_USE_JPOOL)
1969 			KKASSERT(rxstat & RE_RDESC_STAT_EOF);
1970 #endif
1971 
1972 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
1973 			if (sc->re_flags & RE_F_DROP_RXFRAG) {
1974 				re_setup_rxdesc(sc, i);
1975 				continue;
1976 			}
1977 
1978 			if (sc->re_newbuf(sc, i, 0)) {
1979 				/* Drop upcoming fragments */
1980 				sc->re_flags |= RE_F_DROP_RXFRAG;
1981 				continue;
1982 			}
1983 
1984 			m->m_len = MCLBYTES;
1985 			if (sc->re_head == NULL) {
1986 				sc->re_head = sc->re_tail = m;
1987 			} else {
1988 				sc->re_tail->m_next = m;
1989 				sc->re_tail = m;
1990 			}
1991 			continue;
1992 		} else if (sc->re_flags & RE_F_DROP_RXFRAG) {
1993 			/*
1994 			 * Last fragment of a multi-fragment packet.
1995 			 *
1996 			 * Since error already happened, this fragment
1997 			 * must be dropped as well as the fragment chain.
1998 			 */
1999 			re_setup_rxdesc(sc, i);
2000 			re_free_rxchain(sc);
2001 			sc->re_flags &= ~RE_F_DROP_RXFRAG;
2002 			continue;
2003 		}
2004 
2005 		/*
2006 		 * NOTE: for the 8139C+, the frame length field
2007 		 * is always 12 bits in size, but for the gigE chips,
2008 		 * it is 13 bits (since the max RX frame length is 16K).
2009 		 * Unfortunately, all 32 bits in the status word
2010 		 * were already used, so to make room for the extra
2011 		 * length bit, RealTek took out the 'frame alignment
2012 		 * error' bit and shifted the other status bits
2013 		 * over one slot. The OWN, EOR, FS and LS bits are
2014 		 * still in the same places. We have already extracted
2015 		 * the frame length and checked the OWN bit, so rather
2016 		 * than using an alternate bit mapping, we shift the
2017 		 * status bits one space to the right so we can evaluate
2018 		 * them using the 8169 status as though it was in the
2019 		 * same format as that of the 8139C+.
2020 		 */
2021 		if (!RE_IS_8139CP(sc))
2022 			rxstat >>= 1;
2023 
2024 		if (rxstat & RE_RDESC_STAT_RXERRSUM) {
2025 			ifp->if_ierrors++;
2026 			/*
2027 			 * If this is part of a multi-fragment packet,
2028 			 * discard all the pieces.
2029 			 */
2030 			re_free_rxchain(sc);
2031 			re_setup_rxdesc(sc, i);
2032 			continue;
2033 		}
2034 
2035 		/*
2036 		 * If allocating a replacement mbuf fails,
2037 		 * reload the current one.
2038 		 */
2039 
2040 		if (sc->re_newbuf(sc, i, 0)) {
2041 			ifp->if_ierrors++;
2042 			continue;
2043 		}
2044 
2045 		if (sc->re_head != NULL) {
2046 			m->m_len = total_len % MCLBYTES;
2047 			/*
2048 			 * Special case: if there's 4 bytes or less
2049 			 * in this buffer, the mbuf can be discarded:
2050 			 * the last 4 bytes is the CRC, which we don't
2051 			 * care about anyway.
2052 			 */
2053 			if (m->m_len <= ETHER_CRC_LEN) {
2054 				sc->re_tail->m_len -=
2055 				    (ETHER_CRC_LEN - m->m_len);
2056 				m_freem(m);
2057 			} else {
2058 				m->m_len -= ETHER_CRC_LEN;
2059 				sc->re_tail->m_next = m;
2060 			}
2061 			m = sc->re_head;
2062 			sc->re_head = sc->re_tail = NULL;
2063 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2064 		} else {
2065 			m->m_pkthdr.len = m->m_len =
2066 			    (total_len - ETHER_CRC_LEN);
2067 		}
2068 
2069 		ifp->if_ipackets++;
2070 		m->m_pkthdr.rcvif = ifp;
2071 
2072 		/* Do RX checksumming if enabled */
2073 
2074 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2075 			uint8_t packet_type;
2076 
2077 			packet_type = re_packet_type(sc, rxstat, rxctrl);
2078 
2079 			/* Check IP header checksum */
2080 			if (packet_type & RE_IP4_PACKET) {
2081 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2082 				if ((rxstat & RE_RDESC_STAT_IPSUMBAD) == 0)
2083 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2084 			}
2085 
2086 			/* Check TCP/UDP checksum */
2087 			if (((packet_type & RE_TCP_PACKET) &&
2088 			     (rxstat & RE_RDESC_STAT_TCPSUMBAD) == 0) ||
2089 			    ((packet_type & RE_UDP_PACKET) &&
2090 			     (rxstat & RE_RDESC_STAT_UDPSUMBAD) == 0)) {
2091 				m->m_pkthdr.csum_flags |=
2092 				    CSUM_DATA_VALID|CSUM_PSEUDO_HDR|
2093 				    CSUM_FRAG_NOT_CHECKED;
2094 				m->m_pkthdr.csum_data = 0xffff;
2095 			}
2096 		}
2097 
2098 		if (rxctrl & RE_RDESC_CTL_HASTAG) {
2099 			m->m_flags |= M_VLANTAG;
2100 			m->m_pkthdr.ether_vlantag =
2101 				be16toh((rxctrl & RE_RDESC_CTL_TAGDATA));
2102 		}
2103 		ether_input_chain(ifp, m, chain);
2104 	}
2105 
2106 	ether_input_dispatch(chain);
2107 
2108 	/* Flush the RX DMA ring */
2109 
2110 	bus_dmamap_sync(sc->re_ldata.re_rx_list_tag,
2111 			sc->re_ldata.re_rx_list_map, BUS_DMASYNC_PREWRITE);
2112 
2113 	sc->re_ldata.re_rx_prodidx = i;
2114 
2115 	return rx;
2116 }
2117 
2118 #undef RE_IP4_PACKET
2119 #undef RE_TCP_PACKET
2120 #undef RE_UDP_PACKET
2121 
2122 static int
2123 re_tx_collect(struct re_softc *sc)
2124 {
2125 	struct ifnet *ifp = &sc->arpcom.ac_if;
2126 	uint32_t txstat;
2127 	int idx, tx = 0;
2128 
2129 	/* Invalidate the TX descriptor list */
2130 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2131 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_POSTREAD);
2132 
2133 	for (idx = sc->re_ldata.re_tx_considx;
2134 	     sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt;
2135 	     RE_TXDESC_INC(sc, idx)) {
2136 		txstat = le32toh(sc->re_ldata.re_tx_list[idx].re_cmdstat);
2137 		if (txstat & RE_TDESC_CMD_OWN)
2138 			break;
2139 
2140 		tx = 1;
2141 
2142 		sc->re_ldata.re_tx_list[idx].re_bufaddr_lo = 0;
2143 
2144 		/*
2145 		 * We only stash mbufs in the last descriptor
2146 		 * in a fragment chain, which also happens to
2147 		 * be the only place where the TX status bits
2148 		 * are valid.
2149 		 */
2150 		if (txstat & RE_TDESC_CMD_EOF) {
2151 			bus_dmamap_unload(sc->re_ldata.re_tx_mtag,
2152 			    sc->re_ldata.re_tx_dmamap[idx]);
2153 			m_freem(sc->re_ldata.re_tx_mbuf[idx]);
2154 			sc->re_ldata.re_tx_mbuf[idx] = NULL;
2155 			if (txstat & (RE_TDESC_STAT_EXCESSCOL|
2156 			    RE_TDESC_STAT_COLCNT))
2157 				ifp->if_collisions++;
2158 			if (txstat & RE_TDESC_STAT_TXERRSUM)
2159 				ifp->if_oerrors++;
2160 			else
2161 				ifp->if_opackets++;
2162 		}
2163 		sc->re_ldata.re_tx_free++;
2164 	}
2165 	sc->re_ldata.re_tx_considx = idx;
2166 
2167 	return tx;
2168 }
2169 
2170 static int
2171 re_txeof(struct re_softc *sc)
2172 {
2173 	struct ifnet *ifp = &sc->arpcom.ac_if;
2174 	int tx;
2175 
2176 	tx = re_tx_collect(sc);
2177 
2178 	/* There is enough free TX descs */
2179 	if (sc->re_ldata.re_tx_free > RE_TXDESC_SPARE)
2180 		ifp->if_flags &= ~IFF_OACTIVE;
2181 
2182 	/*
2183 	 * Some chips will ignore a second TX request issued while an
2184 	 * existing transmission is in progress. If the transmitter goes
2185 	 * idle but there are still packets waiting to be sent, we need
2186 	 * to restart the channel here to flush them out. This only seems
2187 	 * to be required with the PCIe devices.
2188 	 */
2189 	if (sc->re_ldata.re_tx_free < sc->re_tx_desc_cnt)
2190 		CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2191 	else
2192 		ifp->if_timer = 0;
2193 
2194 	return tx;
2195 }
2196 
2197 static void
2198 re_tick(void *xsc)
2199 {
2200 	struct re_softc *sc = xsc;
2201 
2202 	lwkt_serialize_enter(sc->arpcom.ac_if.if_serializer);
2203 	re_tick_serialized(xsc);
2204 	lwkt_serialize_exit(sc->arpcom.ac_if.if_serializer);
2205 }
2206 
2207 static void
2208 re_tick_serialized(void *xsc)
2209 {
2210 	struct re_softc *sc = xsc;
2211 	struct ifnet *ifp = &sc->arpcom.ac_if;
2212 	struct mii_data *mii;
2213 
2214 	ASSERT_SERIALIZED(ifp->if_serializer);
2215 
2216 	mii = device_get_softc(sc->re_miibus);
2217 	mii_tick(mii);
2218 	if (sc->re_flags & RE_F_LINKED) {
2219 		if (!(mii->mii_media_status & IFM_ACTIVE))
2220 			sc->re_flags &= ~RE_F_LINKED;
2221 	} else {
2222 		if (mii->mii_media_status & IFM_ACTIVE &&
2223 		    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
2224 			sc->re_flags |= RE_F_LINKED;
2225 			if (!ifq_is_empty(&ifp->if_snd))
2226 				if_devstart(ifp);
2227 		}
2228 	}
2229 
2230 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2231 }
2232 
2233 #ifdef DEVICE_POLLING
2234 
2235 static void
2236 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2237 {
2238 	struct re_softc *sc = ifp->if_softc;
2239 
2240 	ASSERT_SERIALIZED(ifp->if_serializer);
2241 
2242 	switch(cmd) {
2243 	case POLL_REGISTER:
2244 		/* disable interrupts */
2245 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2246 		break;
2247 
2248 	case POLL_DEREGISTER:
2249 		/* enable interrupts */
2250 		re_setup_intr(sc, 1, sc->re_imtype);
2251 		break;
2252 
2253 	default:
2254 		sc->rxcycles = count;
2255 		re_rxeof(sc);
2256 		re_txeof(sc);
2257 
2258 		if (!ifq_is_empty(&ifp->if_snd))
2259 			if_devstart(ifp);
2260 
2261 		if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2262 			uint16_t       status;
2263 
2264 			status = CSR_READ_2(sc, RE_ISR);
2265 			if (status == 0xffff)
2266 				return;
2267 			if (status)
2268 				CSR_WRITE_2(sc, RE_ISR, status);
2269 
2270 			/*
2271 			 * XXX check behaviour on receiver stalls.
2272 			 */
2273 
2274 			if (status & RE_ISR_SYSTEM_ERR)
2275 				re_init(sc);
2276 		}
2277 		break;
2278 	}
2279 }
2280 #endif /* DEVICE_POLLING */
2281 
2282 static void
2283 re_intr(void *arg)
2284 {
2285 	struct re_softc	*sc = arg;
2286 	struct ifnet *ifp = &sc->arpcom.ac_if;
2287 	uint16_t status;
2288 	int rx, tx;
2289 
2290 	ASSERT_SERIALIZED(ifp->if_serializer);
2291 
2292 	if ((sc->re_flags & RE_F_SUSPENDED) ||
2293 	    (ifp->if_flags & IFF_RUNNING) == 0)
2294 		return;
2295 
2296 	rx = tx = 0;
2297 	for (;;) {
2298 		status = CSR_READ_2(sc, RE_ISR);
2299 		/* If the card has gone away the read returns 0xffff. */
2300 		if (status == 0xffff)
2301 			break;
2302 		if (status)
2303 			CSR_WRITE_2(sc, RE_ISR, status);
2304 
2305 		if ((status & sc->re_intrs) == 0)
2306 			break;
2307 
2308 		if (status & (sc->re_rx_ack | RE_ISR_RX_ERR))
2309 			rx |= re_rxeof(sc);
2310 
2311 		if (status & (sc->re_tx_ack | RE_ISR_TX_ERR))
2312 			tx |= re_txeof(sc);
2313 
2314 		if (status & RE_ISR_SYSTEM_ERR)
2315 			re_init(sc);
2316 
2317 		if (status & RE_ISR_LINKCHG) {
2318 			callout_stop(&sc->re_timer);
2319 			re_tick_serialized(sc);
2320 		}
2321 	}
2322 
2323 	if (sc->re_imtype == RE_IMTYPE_SIM) {
2324 		if ((sc->re_flags & RE_F_TIMER_INTR)) {
2325 			if ((tx | rx) == 0) {
2326 				/*
2327 				 * Nothing needs to be processed, fallback
2328 				 * to use TX/RX interrupts.
2329 				 */
2330 				re_setup_intr(sc, 1, RE_IMTYPE_NONE);
2331 
2332 				/*
2333 				 * Recollect, mainly to avoid the possible
2334 				 * race introduced by changing interrupt
2335 				 * masks.
2336 				 */
2337 				re_rxeof(sc);
2338 				tx = re_txeof(sc);
2339 			} else {
2340 				CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
2341 			}
2342 		} else if (tx | rx) {
2343 			/*
2344 			 * Assume that using simulated interrupt moderation
2345 			 * (hardware timer based) could reduce the interript
2346 			 * rate.
2347 			 */
2348 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
2349 		}
2350 	}
2351 
2352 	if (tx && !ifq_is_empty(&ifp->if_snd))
2353 		if_devstart(ifp);
2354 }
2355 
2356 static int
2357 re_encap(struct re_softc *sc, struct mbuf **m_head, int *idx0)
2358 {
2359 	struct ifnet *ifp = &sc->arpcom.ac_if;
2360 	struct mbuf *m;
2361 	struct re_dmaload_arg arg;
2362 	bus_dma_segment_t segs[RE_MAXSEGS];
2363 	bus_dmamap_t map;
2364 	int error, maxsegs, idx, i;
2365 	struct re_desc *d, *tx_ring;
2366 	uint32_t cmd_csum, ctl_csum, vlantag;
2367 
2368 	KASSERT(sc->re_ldata.re_tx_free > RE_TXDESC_SPARE,
2369 		("not enough free TX desc\n"));
2370 
2371 	m = *m_head;
2372 	map = sc->re_ldata.re_tx_dmamap[*idx0];
2373 
2374 	/*
2375 	 * Set up checksum offload. Note: checksum offload bits must
2376 	 * appear in all descriptors of a multi-descriptor transmit
2377 	 * attempt. (This is according to testing done with an 8169
2378 	 * chip. I'm not sure if this is a requirement or a bug.)
2379 	 */
2380 	cmd_csum = ctl_csum = 0;
2381 	if (m->m_pkthdr.csum_flags & CSUM_IP) {
2382 		cmd_csum |= RE_TDESC_CMD_IPCSUM;
2383 		ctl_csum |= RE_TDESC_CTL_IPCSUM;
2384 	}
2385 	if (m->m_pkthdr.csum_flags & CSUM_TCP) {
2386 		cmd_csum |= RE_TDESC_CMD_TCPCSUM;
2387 		ctl_csum |= RE_TDESC_CTL_TCPCSUM;
2388 	}
2389 	if (m->m_pkthdr.csum_flags & CSUM_UDP) {
2390 		cmd_csum |= RE_TDESC_CMD_UDPCSUM;
2391 		ctl_csum |= RE_TDESC_CTL_UDPCSUM;
2392 	}
2393 
2394 	/* For MAC2 chips, csum flags are set on re_control */
2395 	if (sc->re_caps & RE_C_MAC2)
2396 		cmd_csum = 0;
2397 	else
2398 		ctl_csum = 0;
2399 
2400 	if ((sc->re_caps & RE_C_AUTOPAD) == 0) {
2401 		/*
2402 		 * With some of the RealTek chips, using the checksum offload
2403 		 * support in conjunction with the autopadding feature results
2404 		 * in the transmission of corrupt frames. For example, if we
2405 		 * need to send a really small IP fragment that's less than 60
2406 		 * bytes in size, and IP header checksumming is enabled, the
2407 		 * resulting ethernet frame that appears on the wire will
2408 		 * have garbled payload. To work around this, if TX checksum
2409 		 * offload is enabled, we always manually pad short frames out
2410 		 * to the minimum ethernet frame size.
2411 		 *
2412 		 * Note: this appears unnecessary for TCP, and doing it for TCP
2413 		 * with PCIe adapters seems to result in bad checksums.
2414 		 */
2415 		if ((m->m_pkthdr.csum_flags &
2416 		     (CSUM_DELAY_IP | CSUM_DELAY_DATA)) &&
2417 		    (m->m_pkthdr.csum_flags & CSUM_TCP) == 0 &&
2418 		    m->m_pkthdr.len < RE_MIN_FRAMELEN) {
2419 			error = m_devpad(m, RE_MIN_FRAMELEN);
2420 			if (error)
2421 				goto back;
2422 		}
2423 	}
2424 
2425 	vlantag = 0;
2426 	if (m->m_flags & M_VLANTAG) {
2427 		vlantag = htobe16(m->m_pkthdr.ether_vlantag) |
2428 			  RE_TDESC_CTL_INSTAG;
2429 	}
2430 
2431 	maxsegs = sc->re_ldata.re_tx_free;
2432 	if (maxsegs > RE_MAXSEGS)
2433 		maxsegs = RE_MAXSEGS;
2434 
2435 	arg.re_nsegs = maxsegs;
2436 	arg.re_segs = segs;
2437 	error = bus_dmamap_load_mbuf(sc->re_ldata.re_tx_mtag, map, m,
2438 				     re_dma_map_desc, &arg, BUS_DMA_NOWAIT);
2439 	if (error && error != EFBIG) {
2440 		if_printf(ifp, "can't map mbuf (error %d)\n", error);
2441 		goto back;
2442 	}
2443 
2444 	/*
2445 	 * Too many segments to map, coalesce into a single mbuf
2446 	 */
2447 	if (!error && arg.re_nsegs == 0) {
2448 		bus_dmamap_unload(sc->re_ldata.re_tx_mtag, map);
2449 		error = EFBIG;
2450 	}
2451 	if (error) {
2452 		struct mbuf *m_new;
2453 
2454 		m_new = m_defrag(m, MB_DONTWAIT);
2455 		if (m_new == NULL) {
2456 			if_printf(ifp, "can't defrag TX mbuf\n");
2457 			error = ENOBUFS;
2458 			goto back;
2459 		} else {
2460 			*m_head = m = m_new;
2461 		}
2462 
2463 		arg.re_nsegs = maxsegs;
2464 		arg.re_segs = segs;
2465 		error = bus_dmamap_load_mbuf(sc->re_ldata.re_tx_mtag, map, m,
2466 					     re_dma_map_desc, &arg,
2467 					     BUS_DMA_NOWAIT);
2468 		if (error || arg.re_nsegs == 0) {
2469 			if (!error) {
2470 				bus_dmamap_unload(sc->re_ldata.re_tx_mtag, map);
2471 				error = EFBIG;
2472 			}
2473 			if_printf(ifp, "can't map mbuf (error %d)\n", error);
2474 			goto back;
2475 		}
2476 	}
2477 	bus_dmamap_sync(sc->re_ldata.re_tx_mtag, map, BUS_DMASYNC_PREWRITE);
2478 
2479 	/*
2480 	 * Map the segment array into descriptors.  We also keep track
2481 	 * of the end of the ring and set the end-of-ring bits as needed,
2482 	 * and we set the ownership bits in all except the very first
2483 	 * descriptor, whose ownership bits will be turned on later.
2484 	 */
2485 	tx_ring = sc->re_ldata.re_tx_list;
2486 	idx = *idx0;
2487 	i = 0;
2488 	for (;;) {
2489 		uint32_t cmdstat;
2490 
2491 		d = &tx_ring[idx];
2492 
2493 		cmdstat = segs[i].ds_len;
2494 		d->re_bufaddr_lo = htole32(RE_ADDR_LO(segs[i].ds_addr));
2495 		d->re_bufaddr_hi = htole32(RE_ADDR_HI(segs[i].ds_addr));
2496 		if (i == 0)
2497 			cmdstat |= RE_TDESC_CMD_SOF;
2498 		else
2499 			cmdstat |= RE_TDESC_CMD_OWN;
2500 		if (idx == (sc->re_tx_desc_cnt - 1))
2501 			cmdstat |= RE_TDESC_CMD_EOR;
2502 		d->re_cmdstat = htole32(cmdstat | cmd_csum);
2503 		d->re_control = htole32(ctl_csum | vlantag);
2504 
2505 		i++;
2506 		if (i == arg.re_nsegs)
2507 			break;
2508 		RE_TXDESC_INC(sc, idx);
2509 	}
2510 	d->re_cmdstat |= htole32(RE_TDESC_CMD_EOF);
2511 
2512 	/* Transfer ownership of packet to the chip. */
2513 	d->re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2514 	if (*idx0 != idx)
2515 		tx_ring[*idx0].re_cmdstat |= htole32(RE_TDESC_CMD_OWN);
2516 
2517 	/*
2518 	 * Insure that the map for this transmission
2519 	 * is placed at the array index of the last descriptor
2520 	 * in this chain.
2521 	 */
2522 	sc->re_ldata.re_tx_dmamap[*idx0] = sc->re_ldata.re_tx_dmamap[idx];
2523 	sc->re_ldata.re_tx_dmamap[idx] = map;
2524 
2525 	sc->re_ldata.re_tx_mbuf[idx] = m;
2526 	sc->re_ldata.re_tx_free -= arg.re_nsegs;
2527 
2528 	RE_TXDESC_INC(sc, idx);
2529 	*idx0 = idx;
2530 back:
2531 	if (error) {
2532 		m_freem(m);
2533 		*m_head = NULL;
2534 	}
2535 	return error;
2536 }
2537 
2538 /*
2539  * Main transmit routine for C+ and gigE NICs.
2540  */
2541 
2542 static void
2543 re_start(struct ifnet *ifp)
2544 {
2545 	struct re_softc	*sc = ifp->if_softc;
2546 	struct mbuf *m_head;
2547 	int idx, need_trans, oactive, error;
2548 
2549 	ASSERT_SERIALIZED(ifp->if_serializer);
2550 
2551 	if ((sc->re_flags & RE_F_LINKED) == 0) {
2552 		ifq_purge(&ifp->if_snd);
2553 		return;
2554 	}
2555 
2556 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
2557 		return;
2558 
2559 	idx = sc->re_ldata.re_tx_prodidx;
2560 
2561 	need_trans = 0;
2562 	oactive = 0;
2563 	while (sc->re_ldata.re_tx_mbuf[idx] == NULL) {
2564 		if (sc->re_ldata.re_tx_free <= RE_TXDESC_SPARE) {
2565 			if (!oactive) {
2566 				if (re_tx_collect(sc)) {
2567 					oactive = 1;
2568 					continue;
2569 				}
2570 			}
2571 			ifp->if_flags |= IFF_OACTIVE;
2572 			break;
2573 		}
2574 
2575 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2576 		if (m_head == NULL)
2577 			break;
2578 
2579 		error = re_encap(sc, &m_head, &idx);
2580 		if (error) {
2581 			/* m_head is freed by re_encap(), if we reach here */
2582 			ifp->if_oerrors++;
2583 
2584 			if (error == EFBIG && !oactive) {
2585 				if (re_tx_collect(sc)) {
2586 					oactive = 1;
2587 					continue;
2588 				}
2589 			}
2590 			ifp->if_flags |= IFF_OACTIVE;
2591 			break;
2592 		}
2593 
2594 		oactive = 0;
2595 		need_trans = 1;
2596 
2597 		/*
2598 		 * If there's a BPF listener, bounce a copy of this frame
2599 		 * to him.
2600 		 */
2601 		ETHER_BPF_MTAP(ifp, m_head);
2602 	}
2603 
2604 	if (!need_trans)
2605 		return;
2606 
2607 	/* Flush the TX descriptors */
2608 	bus_dmamap_sync(sc->re_ldata.re_tx_list_tag,
2609 			sc->re_ldata.re_tx_list_map, BUS_DMASYNC_PREWRITE);
2610 
2611 	sc->re_ldata.re_tx_prodidx = idx;
2612 
2613 	/*
2614 	 * RealTek put the TX poll request register in a different
2615 	 * location on the 8169 gigE chip. I don't know why.
2616 	 */
2617 	CSR_WRITE_1(sc, sc->re_txstart, RE_TXSTART_START);
2618 
2619 	/*
2620 	 * Set a timeout in case the chip goes out to lunch.
2621 	 */
2622 	ifp->if_timer = 5;
2623 }
2624 
2625 static void
2626 re_init(void *xsc)
2627 {
2628 	struct re_softc *sc = xsc;
2629 	struct ifnet *ifp = &sc->arpcom.ac_if;
2630 	struct mii_data *mii;
2631 	int error, framelen;
2632 
2633 	ASSERT_SERIALIZED(ifp->if_serializer);
2634 
2635 	mii = device_get_softc(sc->re_miibus);
2636 
2637 	/*
2638 	 * Cancel pending I/O and free all RX/TX buffers.
2639 	 */
2640 	re_stop(sc);
2641 
2642 	if (sc->re_caps & RE_C_CONTIGRX) {
2643 		if (ifp->if_mtu > ETHERMTU) {
2644 			KKASSERT(sc->re_ldata.re_jbuf != NULL);
2645 			sc->re_flags |= RE_F_USE_JPOOL;
2646 			sc->re_rxbuf_size = RE_FRAMELEN_MAX;
2647 			sc->re_newbuf = re_newbuf_jumbo;
2648 		} else {
2649 			sc->re_flags &= ~RE_F_USE_JPOOL;
2650 			sc->re_rxbuf_size = MCLBYTES;
2651 			sc->re_newbuf = re_newbuf_std;
2652 		}
2653 	}
2654 
2655 	/*
2656 	 * Adjust max read request size according to MTU; mainly to
2657 	 * improve TX performance for common case (ETHERMTU) on GigE
2658 	 * NICs.  However, this could _not_ be done on 10/100 only
2659 	 * NICs; their DMA engines will malfunction using non-default
2660 	 * max read request size.
2661 	 */
2662 	if ((sc->re_caps & (RE_C_PCIE | RE_C_FASTE)) == RE_C_PCIE) {
2663 		if (ifp->if_mtu > ETHERMTU) {
2664 			/*
2665 			 * 512 seems to be the only value that works
2666 			 * reliably with jumbo frame
2667 			 */
2668 			pcie_set_max_readrq(sc->re_dev,
2669 				PCIEM_DEVCTL_MAX_READRQ_512);
2670 		} else {
2671 			pcie_set_max_readrq(sc->re_dev,
2672 				PCIEM_DEVCTL_MAX_READRQ_4096);
2673 		}
2674 	}
2675 
2676 	/*
2677 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
2678 	 * RX checksum offload. We must configure the C+ register
2679 	 * before all others.
2680 	 */
2681 	CSR_WRITE_2(sc, RE_CPLUS_CMD, RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB |
2682 		    RE_CPLUSCMD_PCI_MRW |
2683 		    (ifp->if_capenable & IFCAP_VLAN_HWTAGGING ?
2684 		     RE_CPLUSCMD_VLANSTRIP : 0) |
2685 		    (ifp->if_capenable & IFCAP_RXCSUM ?
2686 		     RE_CPLUSCMD_RXCSUM_ENB : 0));
2687 
2688 	/*
2689 	 * Init our MAC address.  Even though the chipset
2690 	 * documentation doesn't mention it, we need to enter "Config
2691 	 * register write enable" mode to modify the ID registers.
2692 	 */
2693 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_WRITECFG);
2694 	CSR_WRITE_4(sc, RE_IDR0,
2695 	    htole32(*(uint32_t *)(&sc->arpcom.ac_enaddr[0])));
2696 	CSR_WRITE_2(sc, RE_IDR4,
2697 	    htole16(*(uint16_t *)(&sc->arpcom.ac_enaddr[4])));
2698 	CSR_WRITE_1(sc, RE_EECMD, RE_EEMODE_OFF);
2699 
2700 	/*
2701 	 * For C+ mode, initialize the RX descriptors and mbufs.
2702 	 */
2703 	error = re_rx_list_init(sc);
2704 	if (error) {
2705 		re_stop(sc);
2706 		return;
2707 	}
2708 	error = re_tx_list_init(sc);
2709 	if (error) {
2710 		re_stop(sc);
2711 		return;
2712 	}
2713 
2714 	/*
2715 	 * Load the addresses of the RX and TX lists into the chip.
2716 	 */
2717 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_HI,
2718 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_addr));
2719 	CSR_WRITE_4(sc, RE_RXLIST_ADDR_LO,
2720 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_addr));
2721 
2722 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_HI,
2723 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_addr));
2724 	CSR_WRITE_4(sc, RE_TXLIST_ADDR_LO,
2725 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_addr));
2726 
2727 	/*
2728 	 * Enable transmit and receive.
2729 	 */
2730 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2731 
2732 	/*
2733 	 * Set the initial TX and RX configuration.
2734 	 */
2735 	if (sc->re_flags & RE_F_TESTMODE) {
2736 		if (!RE_IS_8139CP(sc))
2737 			CSR_WRITE_4(sc, RE_TXCFG,
2738 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON);
2739 		else
2740 			CSR_WRITE_4(sc, RE_TXCFG,
2741 				    RE_TXCFG_CONFIG | RE_LOOPTEST_ON_CPLUS);
2742 	} else
2743 		CSR_WRITE_4(sc, RE_TXCFG, RE_TXCFG_CONFIG);
2744 
2745 	framelen = RE_FRAMELEN(ifp->if_mtu);
2746 	if (framelen < MCLBYTES)
2747 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(MCLBYTES, 128));
2748 	else
2749 		CSR_WRITE_1(sc, RE_EARLY_TX_THRESH, howmany(framelen, 128));
2750 
2751 	CSR_WRITE_4(sc, RE_RXCFG, RE_RXCFG_CONFIG);
2752 
2753 	/*
2754 	 * Program the multicast filter, if necessary.
2755 	 */
2756 	re_setmulti(sc);
2757 
2758 #ifdef DEVICE_POLLING
2759 	/*
2760 	 * Disable interrupts if we are polling.
2761 	 */
2762 	if (ifp->if_flags & IFF_POLLING)
2763 		re_setup_intr(sc, 0, RE_IMTYPE_NONE);
2764 	else	/* otherwise ... */
2765 #endif /* DEVICE_POLLING */
2766 	/*
2767 	 * Enable interrupts.
2768 	 */
2769 	if (sc->re_flags & RE_F_TESTMODE)
2770 		CSR_WRITE_2(sc, RE_IMR, 0);
2771 	else
2772 		re_setup_intr(sc, 1, sc->re_imtype);
2773 	CSR_WRITE_2(sc, RE_ISR, sc->re_intrs);
2774 
2775 	/* Start RX/TX process. */
2776 	CSR_WRITE_4(sc, RE_MISSEDPKT, 0);
2777 
2778 #ifdef notdef
2779 	/* Enable receiver and transmitter. */
2780 	CSR_WRITE_1(sc, RE_COMMAND, RE_CMD_TX_ENB|RE_CMD_RX_ENB);
2781 #endif
2782 
2783 	/*
2784 	 * For 8169 gigE NICs, set the max allowed RX packet
2785 	 * size so we can receive jumbo frames.
2786 	 */
2787 	if (!RE_IS_8139CP(sc)) {
2788 		if (sc->re_caps & RE_C_CONTIGRX)
2789 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, sc->re_rxbuf_size);
2790 		else
2791 			CSR_WRITE_2(sc, RE_MAXRXPKTLEN, 16383);
2792 	}
2793 
2794 	if (sc->re_flags & RE_F_TESTMODE)
2795 		return;
2796 
2797 	mii_mediachg(mii);
2798 
2799 	CSR_WRITE_1(sc, RE_CFG1, RE_CFG1_DRVLOAD|RE_CFG1_FULLDUPLEX);
2800 
2801 	ifp->if_flags |= IFF_RUNNING;
2802 	ifp->if_flags &= ~IFF_OACTIVE;
2803 
2804 	callout_reset(&sc->re_timer, hz, re_tick, sc);
2805 }
2806 
2807 /*
2808  * Set media options.
2809  */
2810 static int
2811 re_ifmedia_upd(struct ifnet *ifp)
2812 {
2813 	struct re_softc *sc = ifp->if_softc;
2814 	struct mii_data *mii;
2815 
2816 	ASSERT_SERIALIZED(ifp->if_serializer);
2817 
2818 	mii = device_get_softc(sc->re_miibus);
2819 	mii_mediachg(mii);
2820 
2821 	return(0);
2822 }
2823 
2824 /*
2825  * Report current media status.
2826  */
2827 static void
2828 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2829 {
2830 	struct re_softc *sc = ifp->if_softc;
2831 	struct mii_data *mii;
2832 
2833 	ASSERT_SERIALIZED(ifp->if_serializer);
2834 
2835 	mii = device_get_softc(sc->re_miibus);
2836 
2837 	mii_pollstat(mii);
2838 	ifmr->ifm_active = mii->mii_media_active;
2839 	ifmr->ifm_status = mii->mii_media_status;
2840 }
2841 
2842 static int
2843 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
2844 {
2845 	struct re_softc *sc = ifp->if_softc;
2846 	struct ifreq *ifr = (struct ifreq *) data;
2847 	struct mii_data *mii;
2848 	int error = 0, mask;
2849 
2850 	ASSERT_SERIALIZED(ifp->if_serializer);
2851 
2852 	switch(command) {
2853 	case SIOCSIFMTU:
2854 		if (ifr->ifr_mtu > sc->re_maxmtu) {
2855 			error = EINVAL;
2856 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
2857 			ifp->if_mtu = ifr->ifr_mtu;
2858 			if (ifp->if_flags & IFF_RUNNING)
2859 				ifp->if_init(sc);
2860 		}
2861 		break;
2862 
2863 	case SIOCSIFFLAGS:
2864 		if (ifp->if_flags & IFF_UP) {
2865 			if (ifp->if_flags & IFF_RUNNING) {
2866 				if ((ifp->if_flags ^ sc->re_if_flags) &
2867 				    (IFF_PROMISC | IFF_ALLMULTI))
2868 					re_setmulti(sc);
2869 			} else {
2870 				re_init(sc);
2871 			}
2872 		} else if (ifp->if_flags & IFF_RUNNING) {
2873 			re_stop(sc);
2874 		}
2875 		sc->re_if_flags = ifp->if_flags;
2876 		break;
2877 
2878 	case SIOCADDMULTI:
2879 	case SIOCDELMULTI:
2880 		re_setmulti(sc);
2881 		break;
2882 
2883 	case SIOCGIFMEDIA:
2884 	case SIOCSIFMEDIA:
2885 		mii = device_get_softc(sc->re_miibus);
2886 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
2887 		break;
2888 
2889 	case SIOCSIFCAP:
2890 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) &
2891 		       ifp->if_capabilities;
2892 		ifp->if_capenable ^= mask;
2893 
2894 		if (mask & IFCAP_HWCSUM) {
2895 			if (ifp->if_capenable & IFCAP_TXCSUM)
2896 				ifp->if_hwassist = RE_CSUM_FEATURES;
2897 			else
2898 				ifp->if_hwassist = 0;
2899 		}
2900 		if (mask && (ifp->if_flags & IFF_RUNNING))
2901 			re_init(sc);
2902 		break;
2903 
2904 	default:
2905 		error = ether_ioctl(ifp, command, data);
2906 		break;
2907 	}
2908 	return(error);
2909 }
2910 
2911 static void
2912 re_watchdog(struct ifnet *ifp)
2913 {
2914 	struct re_softc *sc = ifp->if_softc;
2915 
2916 	ASSERT_SERIALIZED(ifp->if_serializer);
2917 
2918 	if_printf(ifp, "watchdog timeout\n");
2919 
2920 	ifp->if_oerrors++;
2921 
2922 	re_txeof(sc);
2923 	re_rxeof(sc);
2924 
2925 	re_init(sc);
2926 
2927 	if (!ifq_is_empty(&ifp->if_snd))
2928 		if_devstart(ifp);
2929 }
2930 
2931 /*
2932  * Stop the adapter and free any mbufs allocated to the
2933  * RX and TX lists.
2934  */
2935 static void
2936 re_stop(struct re_softc *sc)
2937 {
2938 	struct ifnet *ifp = &sc->arpcom.ac_if;
2939 	int i;
2940 
2941 	ASSERT_SERIALIZED(ifp->if_serializer);
2942 
2943 	/* Reset the adapter. */
2944 	re_reset(sc, ifp->if_flags & IFF_RUNNING);
2945 
2946 	ifp->if_timer = 0;
2947 	callout_stop(&sc->re_timer);
2948 
2949 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2950 	sc->re_flags &= ~(RE_F_TIMER_INTR | RE_F_DROP_RXFRAG | RE_F_LINKED);
2951 
2952 	CSR_WRITE_1(sc, RE_COMMAND, 0x00);
2953 	CSR_WRITE_2(sc, RE_IMR, 0x0000);
2954 	CSR_WRITE_2(sc, RE_ISR, 0xFFFF);
2955 
2956 	re_free_rxchain(sc);
2957 
2958 	/* Free the TX list buffers. */
2959 	for (i = 0; i < sc->re_tx_desc_cnt; i++) {
2960 		if (sc->re_ldata.re_tx_mbuf[i] != NULL) {
2961 			bus_dmamap_unload(sc->re_ldata.re_tx_mtag,
2962 					  sc->re_ldata.re_tx_dmamap[i]);
2963 			m_freem(sc->re_ldata.re_tx_mbuf[i]);
2964 			sc->re_ldata.re_tx_mbuf[i] = NULL;
2965 		}
2966 	}
2967 
2968 	/* Free the RX list buffers. */
2969 	for (i = 0; i < sc->re_rx_desc_cnt; i++) {
2970 		if (sc->re_ldata.re_rx_mbuf[i] != NULL) {
2971 			if ((sc->re_flags & RE_F_USE_JPOOL) == 0) {
2972 				bus_dmamap_unload(sc->re_ldata.re_rx_mtag,
2973 						  sc->re_ldata.re_rx_dmamap[i]);
2974 			}
2975 			m_freem(sc->re_ldata.re_rx_mbuf[i]);
2976 			sc->re_ldata.re_rx_mbuf[i] = NULL;
2977 		}
2978 	}
2979 }
2980 
2981 /*
2982  * Device suspend routine.  Stop the interface and save some PCI
2983  * settings in case the BIOS doesn't restore them properly on
2984  * resume.
2985  */
2986 static int
2987 re_suspend(device_t dev)
2988 {
2989 #ifndef BURN_BRIDGES
2990 	int i;
2991 #endif
2992 	struct re_softc *sc = device_get_softc(dev);
2993 	struct ifnet *ifp = &sc->arpcom.ac_if;
2994 
2995 	lwkt_serialize_enter(ifp->if_serializer);
2996 
2997 	re_stop(sc);
2998 
2999 #ifndef BURN_BRIDGES
3000 	for (i = 0; i < 5; i++)
3001 		sc->saved_maps[i] = pci_read_config(dev, PCIR_MAPS + i * 4, 4);
3002 	sc->saved_biosaddr = pci_read_config(dev, PCIR_BIOS, 4);
3003 	sc->saved_intline = pci_read_config(dev, PCIR_INTLINE, 1);
3004 	sc->saved_cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3005 	sc->saved_lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3006 #endif
3007 
3008 	sc->re_flags |= RE_F_SUSPENDED;
3009 
3010 	lwkt_serialize_exit(ifp->if_serializer);
3011 
3012 	return (0);
3013 }
3014 
3015 /*
3016  * Device resume routine.  Restore some PCI settings in case the BIOS
3017  * doesn't, re-enable busmastering, and restart the interface if
3018  * appropriate.
3019  */
3020 static int
3021 re_resume(device_t dev)
3022 {
3023 	struct re_softc *sc = device_get_softc(dev);
3024 	struct ifnet *ifp = &sc->arpcom.ac_if;
3025 #ifndef BURN_BRIDGES
3026 	int i;
3027 #endif
3028 
3029 	lwkt_serialize_enter(ifp->if_serializer);
3030 
3031 #ifndef BURN_BRIDGES
3032 	/* better way to do this? */
3033 	for (i = 0; i < 5; i++)
3034 		pci_write_config(dev, PCIR_MAPS + i * 4, sc->saved_maps[i], 4);
3035 	pci_write_config(dev, PCIR_BIOS, sc->saved_biosaddr, 4);
3036 	pci_write_config(dev, PCIR_INTLINE, sc->saved_intline, 1);
3037 	pci_write_config(dev, PCIR_CACHELNSZ, sc->saved_cachelnsz, 1);
3038 	pci_write_config(dev, PCIR_LATTIMER, sc->saved_lattimer, 1);
3039 
3040 	/* reenable busmastering */
3041 	pci_enable_busmaster(dev);
3042 	pci_enable_io(dev, SYS_RES_IOPORT);
3043 #endif
3044 
3045 	/* reinitialize interface if necessary */
3046 	if (ifp->if_flags & IFF_UP)
3047 		re_init(sc);
3048 
3049 	sc->re_flags &= ~RE_F_SUSPENDED;
3050 
3051 	lwkt_serialize_exit(ifp->if_serializer);
3052 
3053 	return (0);
3054 }
3055 
3056 /*
3057  * Stop all chip I/O so that the kernel's probe routines don't
3058  * get confused by errant DMAs when rebooting.
3059  */
3060 static void
3061 re_shutdown(device_t dev)
3062 {
3063 	struct re_softc *sc = device_get_softc(dev);
3064 	struct ifnet *ifp = &sc->arpcom.ac_if;
3065 
3066 	lwkt_serialize_enter(ifp->if_serializer);
3067 	re_stop(sc);
3068 	lwkt_serialize_exit(ifp->if_serializer);
3069 }
3070 
3071 static int
3072 re_sysctl_rxtime(SYSCTL_HANDLER_ARGS)
3073 {
3074 	struct re_softc *sc = arg1;
3075 
3076 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_rx_time);
3077 }
3078 
3079 static int
3080 re_sysctl_txtime(SYSCTL_HANDLER_ARGS)
3081 {
3082 	struct re_softc *sc = arg1;
3083 
3084 	return re_sysctl_hwtime(oidp, arg1, arg2, req, &sc->re_tx_time);
3085 }
3086 
3087 static int
3088 re_sysctl_hwtime(SYSCTL_HANDLER_ARGS, int *hwtime)
3089 {
3090 	struct re_softc *sc = arg1;
3091 	struct ifnet *ifp = &sc->arpcom.ac_if;
3092 	int error, v;
3093 
3094 	lwkt_serialize_enter(ifp->if_serializer);
3095 
3096 	v = *hwtime;
3097 	error = sysctl_handle_int(oidp, &v, 0, req);
3098 	if (error || req->newptr == NULL)
3099 		goto back;
3100 
3101 	if (v <= 0) {
3102 		error = EINVAL;
3103 		goto back;
3104 	}
3105 
3106 	if (v != *hwtime) {
3107 		*hwtime = v;
3108 
3109 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3110 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_HW)
3111 			re_setup_hw_im(sc);
3112 	}
3113 back:
3114 	lwkt_serialize_exit(ifp->if_serializer);
3115 	return error;
3116 }
3117 
3118 static int
3119 re_sysctl_simtime(SYSCTL_HANDLER_ARGS)
3120 {
3121 	struct re_softc *sc = arg1;
3122 	struct ifnet *ifp = &sc->arpcom.ac_if;
3123 	int error, v;
3124 
3125 	lwkt_serialize_enter(ifp->if_serializer);
3126 
3127 	v = sc->re_sim_time;
3128 	error = sysctl_handle_int(oidp, &v, 0, req);
3129 	if (error || req->newptr == NULL)
3130 		goto back;
3131 
3132 	if (v <= 0) {
3133 		error = EINVAL;
3134 		goto back;
3135 	}
3136 
3137 	if (v != sc->re_sim_time) {
3138 		sc->re_sim_time = v;
3139 
3140 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3141 		    IFF_RUNNING && sc->re_imtype == RE_IMTYPE_SIM) {
3142 #ifdef foo
3143 			int reg;
3144 
3145 			/*
3146 			 * Following code causes various strange
3147 			 * performance problems.  Hmm ...
3148 			 */
3149 			CSR_WRITE_2(sc, RE_IMR, 0);
3150 			if (!RE_IS_8139CP(sc))
3151 				reg = RE_TIMERINT_8169;
3152 			else
3153 				reg = RE_TIMERINT;
3154 			CSR_WRITE_4(sc, reg, 0);
3155 			CSR_READ_4(sc, reg); /* flush */
3156 
3157 			CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3158 			re_setup_sim_im(sc);
3159 #else
3160 			re_setup_intr(sc, 0, RE_IMTYPE_NONE);
3161 			DELAY(10);
3162 			re_setup_intr(sc, 1, RE_IMTYPE_SIM);
3163 #endif
3164 		}
3165 	}
3166 back:
3167 	lwkt_serialize_exit(ifp->if_serializer);
3168 	return error;
3169 }
3170 
3171 static int
3172 re_sysctl_imtype(SYSCTL_HANDLER_ARGS)
3173 {
3174 	struct re_softc *sc = arg1;
3175 	struct ifnet *ifp = &sc->arpcom.ac_if;
3176 	int error, v;
3177 
3178 	lwkt_serialize_enter(ifp->if_serializer);
3179 
3180 	v = sc->re_imtype;
3181 	error = sysctl_handle_int(oidp, &v, 0, req);
3182 	if (error || req->newptr == NULL)
3183 		goto back;
3184 
3185 	if (v != RE_IMTYPE_HW && v != RE_IMTYPE_SIM && v != RE_IMTYPE_NONE) {
3186 		error = EINVAL;
3187 		goto back;
3188 	}
3189 	if (v == RE_IMTYPE_HW && (sc->re_caps & RE_C_HWIM) == 0) {
3190 		/* Can't do hardware interrupt moderation */
3191 		error = EOPNOTSUPP;
3192 		goto back;
3193 	}
3194 
3195 	if (v != sc->re_imtype) {
3196 		sc->re_imtype = v;
3197 		if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) ==
3198 		    IFF_RUNNING)
3199 			re_setup_intr(sc, 1, sc->re_imtype);
3200 	}
3201 back:
3202 	lwkt_serialize_exit(ifp->if_serializer);
3203 	return error;
3204 }
3205 
3206 static void
3207 re_setup_hw_im(struct re_softc *sc)
3208 {
3209 	KKASSERT(sc->re_caps & RE_C_HWIM);
3210 
3211 	/*
3212 	 * Interrupt moderation
3213 	 *
3214 	 * 0xABCD
3215 	 * A - unknown (maybe TX related)
3216 	 * B - TX timer (unit: 25us)
3217 	 * C - unknown (maybe RX related)
3218 	 * D - RX timer (unit: 25us)
3219 	 *
3220 	 *
3221 	 * re(4)'s interrupt moderation is actually controlled by
3222 	 * two variables, like most other NICs (bge, bce etc.)
3223 	 * o  timer
3224 	 * o  number of packets [P]
3225 	 *
3226 	 * The logic relationship between these two variables is
3227 	 * similar to other NICs too:
3228 	 * if (timer expire || packets > [P])
3229 	 *     Interrupt is delivered
3230 	 *
3231 	 * Currently we only know how to set 'timer', but not
3232 	 * 'number of packets', which should be ~30, as far as I
3233 	 * tested (sink ~900Kpps, interrupt rate is 30KHz)
3234 	 */
3235 	CSR_WRITE_2(sc, RE_IM,
3236 		    RE_IM_RXTIME(sc->re_rx_time) |
3237 		    RE_IM_TXTIME(sc->re_tx_time) |
3238 		    RE_IM_MAGIC);
3239 }
3240 
3241 static void
3242 re_disable_hw_im(struct re_softc *sc)
3243 {
3244 	if (sc->re_caps & RE_C_HWIM)
3245 		CSR_WRITE_2(sc, RE_IM, 0);
3246 }
3247 
3248 static void
3249 re_setup_sim_im(struct re_softc *sc)
3250 {
3251 	if (!RE_IS_8139CP(sc)) {
3252 		uint32_t ticks;
3253 
3254 		/*
3255 		 * Datasheet says tick decreases at bus speed,
3256 		 * but it seems the clock runs a little bit
3257 		 * faster, so we do some compensation here.
3258 		 */
3259 		ticks = (sc->re_sim_time * sc->re_bus_speed * 8) / 5;
3260 		CSR_WRITE_4(sc, RE_TIMERINT_8169, ticks);
3261 	} else {
3262 		CSR_WRITE_4(sc, RE_TIMERINT, 0x400); /* XXX */
3263 	}
3264 	CSR_WRITE_4(sc, RE_TIMERCNT, 1); /* reload */
3265 	sc->re_flags |= RE_F_TIMER_INTR;
3266 }
3267 
3268 static void
3269 re_disable_sim_im(struct re_softc *sc)
3270 {
3271 	if (!RE_IS_8139CP(sc))
3272 		CSR_WRITE_4(sc, RE_TIMERINT_8169, 0);
3273 	else
3274 		CSR_WRITE_4(sc, RE_TIMERINT, 0);
3275 	sc->re_flags &= ~RE_F_TIMER_INTR;
3276 }
3277 
3278 static void
3279 re_config_imtype(struct re_softc *sc, int imtype)
3280 {
3281 	switch (imtype) {
3282 	case RE_IMTYPE_HW:
3283 		KKASSERT(sc->re_caps & RE_C_HWIM);
3284 		/* FALL THROUGH */
3285 	case RE_IMTYPE_NONE:
3286 		sc->re_intrs = RE_INTRS;
3287 		sc->re_rx_ack = RE_ISR_RX_OK | RE_ISR_FIFO_OFLOW |
3288 				RE_ISR_RX_OVERRUN;
3289 		sc->re_tx_ack = RE_ISR_TX_OK;
3290 		break;
3291 
3292 	case RE_IMTYPE_SIM:
3293 		sc->re_intrs = RE_INTRS_TIMER;
3294 		sc->re_rx_ack = RE_ISR_TIMEOUT_EXPIRED;
3295 		sc->re_tx_ack = RE_ISR_TIMEOUT_EXPIRED;
3296 		break;
3297 
3298 	default:
3299 		panic("%s: unknown imtype %d\n",
3300 		      sc->arpcom.ac_if.if_xname, imtype);
3301 	}
3302 }
3303 
3304 static void
3305 re_setup_intr(struct re_softc *sc, int enable_intrs, int imtype)
3306 {
3307 	re_config_imtype(sc, imtype);
3308 
3309 	if (enable_intrs)
3310 		CSR_WRITE_2(sc, RE_IMR, sc->re_intrs);
3311 	else
3312 		CSR_WRITE_2(sc, RE_IMR, 0);
3313 
3314 	switch (imtype) {
3315 	case RE_IMTYPE_NONE:
3316 		re_disable_sim_im(sc);
3317 		re_disable_hw_im(sc);
3318 		break;
3319 
3320 	case RE_IMTYPE_HW:
3321 		KKASSERT(sc->re_caps & RE_C_HWIM);
3322 		re_disable_sim_im(sc);
3323 		re_setup_hw_im(sc);
3324 		break;
3325 
3326 	case RE_IMTYPE_SIM:
3327 		re_disable_hw_im(sc);
3328 		re_setup_sim_im(sc);
3329 		break;
3330 
3331 	default:
3332 		panic("%s: unknown imtype %d\n",
3333 		      sc->arpcom.ac_if.if_xname, imtype);
3334 	}
3335 }
3336 
3337 static void
3338 re_get_eaddr(struct re_softc *sc, uint8_t *eaddr)
3339 {
3340 	int i;
3341 
3342 	if (sc->re_macver == RE_MACVER_11 || sc->re_macver == RE_MACVER_12) {
3343 		uint16_t re_did;
3344 
3345 		re_get_eewidth(sc);
3346 		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
3347 		if (re_did == 0x8128) {
3348 			uint16_t as[ETHER_ADDR_LEN / 2];
3349 
3350 			/*
3351 			 * Get station address from the EEPROM.
3352 			 */
3353 			re_read_eeprom(sc, (caddr_t)as, RE_EE_EADDR, 3);
3354 			for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
3355 				as[i] = le16toh(as[i]);
3356 			bcopy(as, eaddr, sizeof(eaddr));
3357 			return;
3358 		}
3359 	}
3360 
3361 	/*
3362 	 * Get station address from IDRx.
3363 	 */
3364 	for (i = 0; i < ETHER_ADDR_LEN; ++i)
3365 		eaddr[i] = CSR_READ_1(sc, RE_IDR0 + i);
3366 }
3367 
3368 static int
3369 re_jpool_alloc(struct re_softc *sc)
3370 {
3371 	struct re_list_data *ldata = &sc->re_ldata;
3372 	struct re_jbuf *jbuf;
3373 	bus_addr_t paddr;
3374 	bus_size_t jpool_size;
3375 	caddr_t buf;
3376 	int i, error;
3377 
3378 	lwkt_serialize_init(&ldata->re_jbuf_serializer);
3379 
3380 	ldata->re_jbuf = kmalloc(sizeof(struct re_jbuf) * RE_JBUF_COUNT(sc),
3381 				 M_DEVBUF, M_WAITOK | M_ZERO);
3382 
3383 	jpool_size = RE_JBUF_COUNT(sc) * RE_JBUF_SIZE;
3384 
3385 	error = bus_dma_tag_create(sc->re_parent_tag,
3386 			RE_RXBUF_ALIGN, 0,	/* alignment, boundary */
3387 			BUS_SPACE_MAXADDR,	/* lowaddr */
3388 			BUS_SPACE_MAXADDR,	/* highaddr */
3389 			NULL, NULL,		/* filter, filterarg */
3390 			jpool_size, 1,		/* maxsize, nsegments */
3391 			jpool_size,		/* maxsegsize */
3392 			0,			/* flags */
3393 			&ldata->re_jpool_tag);
3394 	if (error) {
3395 		device_printf(sc->re_dev, "could not allocate jumbo dma tag\n");
3396 		return error;
3397 	}
3398 
3399 	error = bus_dmamem_alloc(ldata->re_jpool_tag, (void **)&ldata->re_jpool,
3400 				 BUS_DMA_WAITOK, &ldata->re_jpool_map);
3401 	if (error) {
3402 		device_printf(sc->re_dev,
3403 			      "could not allocate jumbo dma memory\n");
3404 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3405 		ldata->re_jpool_tag = NULL;
3406 		return error;
3407 	}
3408 
3409 	error = bus_dmamap_load(ldata->re_jpool_tag, ldata->re_jpool_map,
3410 				ldata->re_jpool, jpool_size,
3411 				re_dma_map_addr, &paddr, BUS_DMA_WAITOK);
3412 	if (error) {
3413 		device_printf(sc->re_dev, "could not load jumbo dma map\n");
3414 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3415 				ldata->re_jpool_map);
3416 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3417 		ldata->re_jpool_tag = NULL;
3418 		return error;
3419 	}
3420 
3421 	/* ..and split it into 9KB chunks */
3422 	SLIST_INIT(&ldata->re_jbuf_free);
3423 
3424 	buf = ldata->re_jpool;
3425 	for (i = 0; i < RE_JBUF_COUNT(sc); i++) {
3426 		jbuf = &ldata->re_jbuf[i];
3427 
3428 		jbuf->re_sc = sc;
3429 		jbuf->re_inuse = 0;
3430 		jbuf->re_slot = i;
3431 		jbuf->re_buf = buf;
3432 		jbuf->re_paddr = paddr;
3433 
3434 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3435 
3436 		buf += RE_JBUF_SIZE;
3437 		paddr += RE_JBUF_SIZE;
3438 	}
3439 	return 0;
3440 }
3441 
3442 static void
3443 re_jpool_free(struct re_softc *sc)
3444 {
3445 	struct re_list_data *ldata = &sc->re_ldata;
3446 
3447 	if (ldata->re_jpool_tag != NULL) {
3448 		bus_dmamap_unload(ldata->re_jpool_tag, ldata->re_jpool_map);
3449 		bus_dmamem_free(ldata->re_jpool_tag, ldata->re_jpool,
3450 				ldata->re_jpool_map);
3451 		bus_dma_tag_destroy(ldata->re_jpool_tag);
3452 		ldata->re_jpool_tag = NULL;
3453 	}
3454 
3455 	if (ldata->re_jbuf != NULL) {
3456 		kfree(ldata->re_jbuf, M_DEVBUF);
3457 		ldata->re_jbuf = NULL;
3458 	}
3459 }
3460 
3461 static struct re_jbuf *
3462 re_jbuf_alloc(struct re_softc *sc)
3463 {
3464 	struct re_list_data *ldata = &sc->re_ldata;
3465 	struct re_jbuf *jbuf;
3466 
3467 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3468 
3469 	jbuf = SLIST_FIRST(&ldata->re_jbuf_free);
3470 	if (jbuf != NULL) {
3471 		SLIST_REMOVE_HEAD(&ldata->re_jbuf_free, re_link);
3472 		jbuf->re_inuse = 1;
3473 	}
3474 
3475 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3476 
3477 	return jbuf;
3478 }
3479 
3480 static void
3481 re_jbuf_free(void *arg)
3482 {
3483 	struct re_jbuf *jbuf = arg;
3484 	struct re_softc *sc = jbuf->re_sc;
3485 	struct re_list_data *ldata = &sc->re_ldata;
3486 
3487 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3488 		panic("%s: free wrong jumbo buffer\n",
3489 		      sc->arpcom.ac_if.if_xname);
3490 	} else if (jbuf->re_inuse == 0) {
3491 		panic("%s: jumbo buffer already freed\n",
3492 		      sc->arpcom.ac_if.if_xname);
3493 	}
3494 
3495 	lwkt_serialize_enter(&ldata->re_jbuf_serializer);
3496 	atomic_subtract_int(&jbuf->re_inuse, 1);
3497 	if (jbuf->re_inuse == 0)
3498 		SLIST_INSERT_HEAD(&ldata->re_jbuf_free, jbuf, re_link);
3499 	lwkt_serialize_exit(&ldata->re_jbuf_serializer);
3500 }
3501 
3502 static void
3503 re_jbuf_ref(void *arg)
3504 {
3505 	struct re_jbuf *jbuf = arg;
3506 	struct re_softc *sc = jbuf->re_sc;
3507 	struct re_list_data *ldata = &sc->re_ldata;
3508 
3509 	if (&ldata->re_jbuf[jbuf->re_slot] != jbuf) {
3510 		panic("%s: ref wrong jumbo buffer\n",
3511 		      sc->arpcom.ac_if.if_xname);
3512 	} else if (jbuf->re_inuse == 0) {
3513 		panic("%s: jumbo buffer already freed\n",
3514 		      sc->arpcom.ac_if.if_xname);
3515 	}
3516 	atomic_add_int(&jbuf->re_inuse, 1);
3517 }
3518