xref: /dflybsd-src/sys/dev/netif/nfe/if_nfe.c (revision d54592ee9e96c920b951af2e00cd72c0081ccae3)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 /*	$DragonFly: src/sys/dev/netif/nfe/if_nfe.c,v 1.43 2008/07/27 10:06:56 sephe Exp $	*/
3 
4 /*
5  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
6  *
7  * This code is derived from software contributed to The DragonFly Project
8  * by Sepherosa Ziehau <sepherosa@gmail.com> and
9  * Matthew Dillon <dillon@apollo.backplane.com>
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in
19  *    the documentation and/or other materials provided with the
20  *    distribution.
21  * 3. Neither the name of The DragonFly Project nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific, prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
28  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
29  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
30  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
31  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
32  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
33  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
34  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
35  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
41  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
42  *
43  * Permission to use, copy, modify, and distribute this software for any
44  * purpose with or without fee is hereby granted, provided that the above
45  * copyright notice and this permission notice appear in all copies.
46  *
47  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
48  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
49  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
50  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
51  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
52  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
53  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
54  */
55 
56 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
57 
58 #include "opt_polling.h"
59 #include "opt_ethernet.h"
60 
61 #include <sys/param.h>
62 #include <sys/endian.h>
63 #include <sys/kernel.h>
64 #include <sys/bus.h>
65 #include <sys/interrupt.h>
66 #include <sys/proc.h>
67 #include <sys/rman.h>
68 #include <sys/serialize.h>
69 #include <sys/socket.h>
70 #include <sys/sockio.h>
71 #include <sys/sysctl.h>
72 
73 #include <net/ethernet.h>
74 #include <net/if.h>
75 #include <net/bpf.h>
76 #include <net/if_arp.h>
77 #include <net/if_dl.h>
78 #include <net/if_media.h>
79 #include <net/ifq_var.h>
80 #include <net/if_types.h>
81 #include <net/if_var.h>
82 #include <net/vlan/if_vlan_var.h>
83 #include <net/vlan/if_vlan_ether.h>
84 
85 #include <bus/pci/pcireg.h>
86 #include <bus/pci/pcivar.h>
87 #include <bus/pci/pcidevs.h>
88 
89 #include <dev/netif/mii_layer/mii.h>
90 #include <dev/netif/mii_layer/miivar.h>
91 
92 #include "miibus_if.h"
93 
94 #include <dev/netif/nfe/if_nfereg.h>
95 #include <dev/netif/nfe/if_nfevar.h>
96 
97 #define NFE_CSUM
98 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
99 
100 static int	nfe_probe(device_t);
101 static int	nfe_attach(device_t);
102 static int	nfe_detach(device_t);
103 static void	nfe_shutdown(device_t);
104 static int	nfe_resume(device_t);
105 static int	nfe_suspend(device_t);
106 
107 static int	nfe_miibus_readreg(device_t, int, int);
108 static void	nfe_miibus_writereg(device_t, int, int, int);
109 static void	nfe_miibus_statchg(device_t);
110 
111 #ifdef DEVICE_POLLING
112 static void	nfe_poll(struct ifnet *, enum poll_cmd, int);
113 #endif
114 static void	nfe_intr(void *);
115 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
116 static int	nfe_rxeof(struct nfe_softc *);
117 static int	nfe_txeof(struct nfe_softc *, int);
118 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
119 			  struct mbuf *);
120 static void	nfe_start(struct ifnet *);
121 static void	nfe_watchdog(struct ifnet *);
122 static void	nfe_init(void *);
123 static void	nfe_stop(struct nfe_softc *);
124 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
125 static void	nfe_jfree(void *);
126 static void	nfe_jref(void *);
127 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
128 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
129 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
130 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
131 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
132 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
133 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
134 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
135 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
136 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
137 static int	nfe_ifmedia_upd(struct ifnet *);
138 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
139 static void	nfe_setmulti(struct nfe_softc *);
140 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
141 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
142 static void	nfe_powerup(device_t);
143 static void	nfe_mac_reset(struct nfe_softc *);
144 static void	nfe_tick(void *);
145 static void	nfe_ring_dma_addr(void *, bus_dma_segment_t *, int, int);
146 static void	nfe_buf_dma_addr(void *, bus_dma_segment_t *, int, bus_size_t,
147 				 int);
148 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
149 				     int, bus_addr_t);
150 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
151 				     int);
152 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
153 			       int);
154 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
155 				 int);
156 static void	nfe_enable_intrs(struct nfe_softc *);
157 static void	nfe_disable_intrs(struct nfe_softc *);
158 
159 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
160 
161 #define NFE_DEBUG
162 #ifdef NFE_DEBUG
163 
164 static int	nfe_debug = 0;
165 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
166 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
167 /* hw timer simulated interrupt moderation @8000Hz */
168 static int	nfe_imtime = -125;
169 
170 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
171 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
172 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
173 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
174 
175 #define DPRINTF(sc, fmt, ...) do {		\
176 	if ((sc)->sc_debug) {			\
177 		if_printf(&(sc)->arpcom.ac_if,	\
178 			  fmt, __VA_ARGS__);	\
179 	}					\
180 } while (0)
181 
182 #define DPRINTFN(sc, lv, fmt, ...) do {		\
183 	if ((sc)->sc_debug >= (lv)) {		\
184 		if_printf(&(sc)->arpcom.ac_if,	\
185 			  fmt, __VA_ARGS__);	\
186 	}					\
187 } while (0)
188 
189 #else	/* !NFE_DEBUG */
190 
191 #define DPRINTF(sc, fmt, ...)
192 #define DPRINTFN(sc, lv, fmt, ...)
193 
194 #endif	/* NFE_DEBUG */
195 
196 struct nfe_dma_ctx {
197 	int			nsegs;
198 	bus_dma_segment_t	*segs;
199 };
200 
201 static const struct nfe_dev {
202 	uint16_t	vid;
203 	uint16_t	did;
204 	const char	*desc;
205 } nfe_devices[] = {
206 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
207 	  "NVIDIA nForce Fast Ethernet" },
208 
209 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
210 	  "NVIDIA nForce2 Fast Ethernet" },
211 
212 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
213 	  "NVIDIA nForce3 Gigabit Ethernet" },
214 
215 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
216 	   chipset, and possibly also the 400R; it might be both nForce2- and
217 	   nForce3-based boards can use the same MCPs (= southbridges) */
218 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
219 	  "NVIDIA nForce3 Gigabit Ethernet" },
220 
221 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
222 	  "NVIDIA nForce3 Gigabit Ethernet" },
223 
224 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
225 	  "NVIDIA nForce3 Gigabit Ethernet" },
226 
227 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
228 	  "NVIDIA nForce3 Gigabit Ethernet" },
229 
230 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
231 	  "NVIDIA CK804 Gigabit Ethernet" },
232 
233 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
234 	  "NVIDIA CK804 Gigabit Ethernet" },
235 
236 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
237 	  "NVIDIA MCP04 Gigabit Ethernet" },
238 
239 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
240 	  "NVIDIA MCP04 Gigabit Ethernet" },
241 
242 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
243 	  "NVIDIA MCP51 Gigabit Ethernet" },
244 
245 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
246 	  "NVIDIA MCP51 Gigabit Ethernet" },
247 
248 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
249 	  "NVIDIA MCP55 Gigabit Ethernet" },
250 
251 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
252 	  "NVIDIA MCP55 Gigabit Ethernet" },
253 
254 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
255 	  "NVIDIA MCP61 Gigabit Ethernet" },
256 
257 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
258 	  "NVIDIA MCP61 Gigabit Ethernet" },
259 
260 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
261 	  "NVIDIA MCP61 Gigabit Ethernet" },
262 
263 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
264 	  "NVIDIA MCP61 Gigabit Ethernet" },
265 
266 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
267 	  "NVIDIA MCP65 Gigabit Ethernet" },
268 
269 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
270 	  "NVIDIA MCP65 Gigabit Ethernet" },
271 
272 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
273 	  "NVIDIA MCP65 Gigabit Ethernet" },
274 
275 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
276 	  "NVIDIA MCP65 Gigabit Ethernet" },
277 
278 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
279 	  "NVIDIA MCP67 Gigabit Ethernet" },
280 
281 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
282 	  "NVIDIA MCP67 Gigabit Ethernet" },
283 
284 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
285 	  "NVIDIA MCP67 Gigabit Ethernet" },
286 
287 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
288 	  "NVIDIA MCP67 Gigabit Ethernet" },
289 
290 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
291 	  "NVIDIA MCP73 Gigabit Ethernet" },
292 
293 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
294 	  "NVIDIA MCP73 Gigabit Ethernet" },
295 
296 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
297 	  "NVIDIA MCP73 Gigabit Ethernet" },
298 
299 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
300 	  "NVIDIA MCP73 Gigabit Ethernet" },
301 
302 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
303 	  "NVIDIA MCP77 Gigabit Ethernet" },
304 
305 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
306 	  "NVIDIA MCP77 Gigabit Ethernet" },
307 
308 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
309 	  "NVIDIA MCP77 Gigabit Ethernet" },
310 
311 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
312 	  "NVIDIA MCP77 Gigabit Ethernet" },
313 
314 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
315 	  "NVIDIA MCP79 Gigabit Ethernet" },
316 
317 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
318 	  "NVIDIA MCP79 Gigabit Ethernet" },
319 
320 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
321 	  "NVIDIA MCP79 Gigabit Ethernet" },
322 
323 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
324 	  "NVIDIA MCP79 Gigabit Ethernet" },
325 
326 	{ 0, 0, NULL }
327 };
328 
329 static device_method_t nfe_methods[] = {
330 	/* Device interface */
331 	DEVMETHOD(device_probe,		nfe_probe),
332 	DEVMETHOD(device_attach,	nfe_attach),
333 	DEVMETHOD(device_detach,	nfe_detach),
334 	DEVMETHOD(device_suspend,	nfe_suspend),
335 	DEVMETHOD(device_resume,	nfe_resume),
336 	DEVMETHOD(device_shutdown,	nfe_shutdown),
337 
338 	/* Bus interface */
339 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
340 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
341 
342 	/* MII interface */
343 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
344 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
345 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
346 
347 	{ 0, 0 }
348 };
349 
350 static driver_t nfe_driver = {
351 	"nfe",
352 	nfe_methods,
353 	sizeof(struct nfe_softc)
354 };
355 
356 static devclass_t	nfe_devclass;
357 
358 DECLARE_DUMMY_MODULE(if_nfe);
359 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
360 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, 0, 0);
361 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, 0, 0);
362 
363 static int
364 nfe_probe(device_t dev)
365 {
366 	const struct nfe_dev *n;
367 	uint16_t vid, did;
368 
369 	vid = pci_get_vendor(dev);
370 	did = pci_get_device(dev);
371 	for (n = nfe_devices; n->desc != NULL; ++n) {
372 		if (vid == n->vid && did == n->did) {
373 			struct nfe_softc *sc = device_get_softc(dev);
374 
375 			switch (did) {
376 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
377 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
378 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
379 				sc->sc_caps = NFE_NO_PWRCTL |
380 					      NFE_FIX_EADDR;
381 				break;
382 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
383 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
384 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
385 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
386 				sc->sc_caps = NFE_JUMBO_SUP |
387 					      NFE_HW_CSUM |
388 					      NFE_NO_PWRCTL |
389 					      NFE_FIX_EADDR;
390 				break;
391 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
392 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
393 				sc->sc_caps = NFE_FIX_EADDR;
394 				/* FALL THROUGH */
395 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
396 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
397 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
398 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
399 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
400 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
401 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
402 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
403 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
404 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
405 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
406 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
407 				sc->sc_caps |= NFE_40BIT_ADDR;
408 				break;
409 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
410 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
411 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
412 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
413 				sc->sc_caps = NFE_JUMBO_SUP |
414 					      NFE_40BIT_ADDR |
415 					      NFE_HW_CSUM |
416 					      NFE_NO_PWRCTL |
417 					      NFE_FIX_EADDR;
418 				break;
419 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
420 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
421 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
422 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
423 				sc->sc_caps = NFE_JUMBO_SUP |
424 					      NFE_40BIT_ADDR;
425 				break;
426 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
427 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
428 				sc->sc_caps = NFE_JUMBO_SUP |
429 					      NFE_40BIT_ADDR |
430 					      NFE_HW_CSUM |
431 					      NFE_HW_VLAN |
432 					      NFE_FIX_EADDR;
433 				break;
434 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
435 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
436 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
437 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
438 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
439 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
440 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
441 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
442 				sc->sc_caps = NFE_40BIT_ADDR |
443 					      NFE_HW_CSUM;
444 				break;
445 			}
446 
447 			device_set_desc(dev, n->desc);
448 			device_set_async_attach(dev, TRUE);
449 			return 0;
450 		}
451 	}
452 	return ENXIO;
453 }
454 
455 static int
456 nfe_attach(device_t dev)
457 {
458 	struct nfe_softc *sc = device_get_softc(dev);
459 	struct ifnet *ifp = &sc->arpcom.ac_if;
460 	uint8_t eaddr[ETHER_ADDR_LEN];
461 	int error;
462 
463 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
464 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
465 
466 	/*
467 	 * Initialize sysctl variables
468 	 */
469 	sc->sc_rx_ring_count = nfe_rx_ring_count;
470 	sc->sc_tx_ring_count = nfe_tx_ring_count;
471 	sc->sc_debug = nfe_debug;
472 	if (nfe_imtime < 0) {
473 		sc->sc_flags |= NFE_F_DYN_IM;
474 		sc->sc_imtime = -nfe_imtime;
475 	} else {
476 		sc->sc_imtime = nfe_imtime;
477 	}
478 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
479 
480 	sc->sc_mem_rid = PCIR_BAR(0);
481 
482 	if (sc->sc_caps & NFE_40BIT_ADDR)
483 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
484 	else if (sc->sc_caps & NFE_JUMBO_SUP)
485 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
486 
487 #ifndef BURN_BRIDGES
488 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
489 		uint32_t mem, irq;
490 
491 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
492 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
493 
494 		device_printf(dev, "chip is in D%d power mode "
495 		    "-- setting to D0\n", pci_get_powerstate(dev));
496 
497 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
498 
499 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
500 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
501 	}
502 #endif	/* !BURN_BRIDGE */
503 
504 	/* Enable bus mastering */
505 	pci_enable_busmaster(dev);
506 
507 	/* Allocate IO memory */
508 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
509 						&sc->sc_mem_rid, RF_ACTIVE);
510 	if (sc->sc_mem_res == NULL) {
511 		device_printf(dev, "cound not allocate io memory\n");
512 		return ENXIO;
513 	}
514 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
515 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
516 
517 	/* Allocate IRQ */
518 	sc->sc_irq_rid = 0;
519 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
520 						&sc->sc_irq_rid,
521 						RF_SHAREABLE | RF_ACTIVE);
522 	if (sc->sc_irq_res == NULL) {
523 		device_printf(dev, "could not allocate irq\n");
524 		error = ENXIO;
525 		goto fail;
526 	}
527 
528 	/* Disable WOL */
529 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
530 
531 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
532 		nfe_powerup(dev);
533 
534 	nfe_get_macaddr(sc, eaddr);
535 
536 	/*
537 	 * Allocate Tx and Rx rings.
538 	 */
539 	error = nfe_alloc_tx_ring(sc, &sc->txq);
540 	if (error) {
541 		device_printf(dev, "could not allocate Tx ring\n");
542 		goto fail;
543 	}
544 
545 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
546 	if (error) {
547 		device_printf(dev, "could not allocate Rx ring\n");
548 		goto fail;
549 	}
550 
551 	/*
552 	 * Create sysctl tree
553 	 */
554 	sysctl_ctx_init(&sc->sc_sysctl_ctx);
555 	sc->sc_sysctl_tree = SYSCTL_ADD_NODE(&sc->sc_sysctl_ctx,
556 					     SYSCTL_STATIC_CHILDREN(_hw),
557 					     OID_AUTO,
558 					     device_get_nameunit(dev),
559 					     CTLFLAG_RD, 0, "");
560 	if (sc->sc_sysctl_tree == NULL) {
561 		device_printf(dev, "can't add sysctl node\n");
562 		error = ENXIO;
563 		goto fail;
564 	}
565 	SYSCTL_ADD_PROC(&sc->sc_sysctl_ctx,
566 			SYSCTL_CHILDREN(sc->sc_sysctl_tree),
567 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
568 			sc, 0, nfe_sysctl_imtime, "I",
569 			"Interrupt moderation time (usec).  "
570 			"0 to disable interrupt moderation.");
571 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
572 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
573 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
574 		       0, "RX ring count");
575 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
576 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
577 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
578 		       0, "TX ring count");
579 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
580 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
581 		       "debug", CTLFLAG_RW, &sc->sc_debug,
582 		       0, "control debugging printfs");
583 
584 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
585 			      nfe_ifmedia_sts);
586 	if (error) {
587 		device_printf(dev, "MII without any phy\n");
588 		goto fail;
589 	}
590 
591 	ifp->if_softc = sc;
592 	ifp->if_mtu = ETHERMTU;
593 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
594 	ifp->if_ioctl = nfe_ioctl;
595 	ifp->if_start = nfe_start;
596 #ifdef DEVICE_POLLING
597 	ifp->if_poll = nfe_poll;
598 #endif
599 	ifp->if_watchdog = nfe_watchdog;
600 	ifp->if_init = nfe_init;
601 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
602 	ifq_set_ready(&ifp->if_snd);
603 
604 	ifp->if_capabilities = IFCAP_VLAN_MTU;
605 
606 	if (sc->sc_caps & NFE_HW_VLAN)
607 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
608 
609 #ifdef NFE_CSUM
610 	if (sc->sc_caps & NFE_HW_CSUM) {
611 		ifp->if_capabilities |= IFCAP_HWCSUM;
612 		ifp->if_hwassist = NFE_CSUM_FEATURES;
613 	}
614 #else
615 	sc->sc_caps &= ~NFE_HW_CSUM;
616 #endif
617 	ifp->if_capenable = ifp->if_capabilities;
618 
619 	callout_init(&sc->sc_tick_ch);
620 
621 	ether_ifattach(ifp, eaddr, NULL);
622 
623 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
624 			       &sc->sc_ih, ifp->if_serializer);
625 	if (error) {
626 		device_printf(dev, "could not setup intr\n");
627 		ether_ifdetach(ifp);
628 		goto fail;
629 	}
630 
631 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->sc_irq_res));
632 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
633 
634 	return 0;
635 fail:
636 	nfe_detach(dev);
637 	return error;
638 }
639 
640 static int
641 nfe_detach(device_t dev)
642 {
643 	struct nfe_softc *sc = device_get_softc(dev);
644 
645 	if (device_is_attached(dev)) {
646 		struct ifnet *ifp = &sc->arpcom.ac_if;
647 
648 		lwkt_serialize_enter(ifp->if_serializer);
649 		nfe_stop(sc);
650 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
651 		lwkt_serialize_exit(ifp->if_serializer);
652 
653 		ether_ifdetach(ifp);
654 	}
655 
656 	if (sc->sc_miibus != NULL)
657 		device_delete_child(dev, sc->sc_miibus);
658 	bus_generic_detach(dev);
659 
660 	if (sc->sc_sysctl_tree != NULL)
661 		sysctl_ctx_free(&sc->sc_sysctl_ctx);
662 
663 	if (sc->sc_irq_res != NULL) {
664 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
665 				     sc->sc_irq_res);
666 	}
667 
668 	if (sc->sc_mem_res != NULL) {
669 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
670 				     sc->sc_mem_res);
671 	}
672 
673 	nfe_free_tx_ring(sc, &sc->txq);
674 	nfe_free_rx_ring(sc, &sc->rxq);
675 
676 	return 0;
677 }
678 
679 static void
680 nfe_shutdown(device_t dev)
681 {
682 	struct nfe_softc *sc = device_get_softc(dev);
683 	struct ifnet *ifp = &sc->arpcom.ac_if;
684 
685 	lwkt_serialize_enter(ifp->if_serializer);
686 	nfe_stop(sc);
687 	lwkt_serialize_exit(ifp->if_serializer);
688 }
689 
690 static int
691 nfe_suspend(device_t dev)
692 {
693 	struct nfe_softc *sc = device_get_softc(dev);
694 	struct ifnet *ifp = &sc->arpcom.ac_if;
695 
696 	lwkt_serialize_enter(ifp->if_serializer);
697 	nfe_stop(sc);
698 	lwkt_serialize_exit(ifp->if_serializer);
699 
700 	return 0;
701 }
702 
703 static int
704 nfe_resume(device_t dev)
705 {
706 	struct nfe_softc *sc = device_get_softc(dev);
707 	struct ifnet *ifp = &sc->arpcom.ac_if;
708 
709 	lwkt_serialize_enter(ifp->if_serializer);
710 	if (ifp->if_flags & IFF_UP)
711 		nfe_init(sc);
712 	lwkt_serialize_exit(ifp->if_serializer);
713 
714 	return 0;
715 }
716 
717 static void
718 nfe_miibus_statchg(device_t dev)
719 {
720 	struct nfe_softc *sc = device_get_softc(dev);
721 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
722 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
723 
724 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
725 
726 	phy = NFE_READ(sc, NFE_PHY_IFACE);
727 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
728 
729 	seed = NFE_READ(sc, NFE_RNDSEED);
730 	seed &= ~NFE_SEED_MASK;
731 
732 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
733 		phy  |= NFE_PHY_HDX;	/* half-duplex */
734 		misc |= NFE_MISC1_HDX;
735 	}
736 
737 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
738 	case IFM_1000_T:	/* full-duplex only */
739 		link |= NFE_MEDIA_1000T;
740 		seed |= NFE_SEED_1000T;
741 		phy  |= NFE_PHY_1000T;
742 		break;
743 	case IFM_100_TX:
744 		link |= NFE_MEDIA_100TX;
745 		seed |= NFE_SEED_100TX;
746 		phy  |= NFE_PHY_100TX;
747 		break;
748 	case IFM_10_T:
749 		link |= NFE_MEDIA_10T;
750 		seed |= NFE_SEED_10T;
751 		break;
752 	}
753 
754 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
755 
756 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
757 	NFE_WRITE(sc, NFE_MISC1, misc);
758 	NFE_WRITE(sc, NFE_LINKSPEED, link);
759 }
760 
761 static int
762 nfe_miibus_readreg(device_t dev, int phy, int reg)
763 {
764 	struct nfe_softc *sc = device_get_softc(dev);
765 	uint32_t val;
766 	int ntries;
767 
768 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
769 
770 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
771 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
772 		DELAY(100);
773 	}
774 
775 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
776 
777 	for (ntries = 0; ntries < 1000; ntries++) {
778 		DELAY(100);
779 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
780 			break;
781 	}
782 	if (ntries == 1000) {
783 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
784 		return 0;
785 	}
786 
787 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
788 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
789 		return 0;
790 	}
791 
792 	val = NFE_READ(sc, NFE_PHY_DATA);
793 	if (val != 0xffffffff && val != 0)
794 		sc->mii_phyaddr = phy;
795 
796 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
797 
798 	return val;
799 }
800 
801 static void
802 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
803 {
804 	struct nfe_softc *sc = device_get_softc(dev);
805 	uint32_t ctl;
806 	int ntries;
807 
808 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
809 
810 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
811 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
812 		DELAY(100);
813 	}
814 
815 	NFE_WRITE(sc, NFE_PHY_DATA, val);
816 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
817 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
818 
819 	for (ntries = 0; ntries < 1000; ntries++) {
820 		DELAY(100);
821 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
822 			break;
823 	}
824 
825 #ifdef NFE_DEBUG
826 	if (ntries == 1000)
827 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
828 #endif
829 }
830 
831 #ifdef DEVICE_POLLING
832 
833 static void
834 nfe_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
835 {
836 	struct nfe_softc *sc = ifp->if_softc;
837 
838 	ASSERT_SERIALIZED(ifp->if_serializer);
839 
840 	switch(cmd) {
841 	case POLL_REGISTER:
842 		nfe_disable_intrs(sc);
843 		break;
844 
845 	case POLL_DEREGISTER:
846 		nfe_enable_intrs(sc);
847 		break;
848 
849 	case POLL_AND_CHECK_STATUS:
850 		/* fall through */
851 	case POLL_ONLY:
852 		if (ifp->if_flags & IFF_RUNNING) {
853 			nfe_rxeof(sc);
854 			nfe_txeof(sc, 1);
855 		}
856 		break;
857 	}
858 }
859 
860 #endif
861 
862 static void
863 nfe_intr(void *arg)
864 {
865 	struct nfe_softc *sc = arg;
866 	struct ifnet *ifp = &sc->arpcom.ac_if;
867 	uint32_t r;
868 
869 	r = NFE_READ(sc, NFE_IRQ_STATUS);
870 	if (r == 0)
871 		return;	/* not for us */
872 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
873 
874 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
875 
876 	if (r & NFE_IRQ_LINK) {
877 		NFE_READ(sc, NFE_PHY_STATUS);
878 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
879 		DPRINTF(sc, "link state changed %s\n", "");
880 	}
881 
882 	if (ifp->if_flags & IFF_RUNNING) {
883 		int ret;
884 
885 		/* check Rx ring */
886 		ret = nfe_rxeof(sc);
887 
888 		/* check Tx ring */
889 		ret |= nfe_txeof(sc, 1);
890 
891 		if (sc->sc_flags & NFE_F_DYN_IM) {
892 			if (ret && (sc->sc_flags & NFE_F_IRQ_TIMER) == 0) {
893 				/*
894 				 * Assume that using hardware timer could reduce
895 				 * the interrupt rate.
896 				 */
897 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
898 				sc->sc_flags |= NFE_F_IRQ_TIMER;
899 			} else if (!ret && (sc->sc_flags & NFE_F_IRQ_TIMER)) {
900 				/*
901 				 * Nothing needs to be processed, fall back to
902 				 * use TX/RX interrupts.
903 				 */
904 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
905 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
906 			}
907 		}
908 	}
909 }
910 
911 static int
912 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
913 {
914 	struct nfe_softc *sc = ifp->if_softc;
915 	struct ifreq *ifr = (struct ifreq *)data;
916 	struct mii_data *mii;
917 	int error = 0, mask, jumbo_cap;
918 
919 	ASSERT_SERIALIZED(ifp->if_serializer);
920 
921 	switch (cmd) {
922 	case SIOCSIFMTU:
923 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
924 			jumbo_cap = 1;
925 		else
926 			jumbo_cap = 0;
927 
928 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
929 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
930 			return EINVAL;
931 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
932 			ifp->if_mtu = ifr->ifr_mtu;
933 			if (ifp->if_flags & IFF_RUNNING)
934 				nfe_init(sc);
935 		}
936 		break;
937 	case SIOCSIFFLAGS:
938 		if (ifp->if_flags & IFF_UP) {
939 			/*
940 			 * If only the PROMISC or ALLMULTI flag changes, then
941 			 * don't do a full re-init of the chip, just update
942 			 * the Rx filter.
943 			 */
944 			if ((ifp->if_flags & IFF_RUNNING) &&
945 			    ((ifp->if_flags ^ sc->sc_if_flags) &
946 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
947 				nfe_setmulti(sc);
948 			} else {
949 				if (!(ifp->if_flags & IFF_RUNNING))
950 					nfe_init(sc);
951 			}
952 		} else {
953 			if (ifp->if_flags & IFF_RUNNING)
954 				nfe_stop(sc);
955 		}
956 		sc->sc_if_flags = ifp->if_flags;
957 		break;
958 	case SIOCADDMULTI:
959 	case SIOCDELMULTI:
960 		if (ifp->if_flags & IFF_RUNNING)
961 			nfe_setmulti(sc);
962 		break;
963 	case SIOCSIFMEDIA:
964 	case SIOCGIFMEDIA:
965 		mii = device_get_softc(sc->sc_miibus);
966 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
967 		break;
968         case SIOCSIFCAP:
969 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
970 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
971 			ifp->if_capenable ^= mask;
972 			if (IFCAP_TXCSUM & ifp->if_capenable)
973 				ifp->if_hwassist = NFE_CSUM_FEATURES;
974 			else
975 				ifp->if_hwassist = 0;
976 
977 			if (ifp->if_flags & IFF_RUNNING)
978 				nfe_init(sc);
979 		}
980 		break;
981 	default:
982 		error = ether_ioctl(ifp, cmd, data);
983 		break;
984 	}
985 	return error;
986 }
987 
988 static int
989 nfe_rxeof(struct nfe_softc *sc)
990 {
991 	struct ifnet *ifp = &sc->arpcom.ac_if;
992 	struct nfe_rx_ring *ring = &sc->rxq;
993 	int reap;
994 #ifdef ETHER_INPUT_CHAIN
995 	struct mbuf_chain chain[MAXCPU];
996 #endif
997 
998 	reap = 0;
999 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
1000 
1001 #ifdef ETHER_INPUT_CHAIN
1002 	ether_input_chain_init(chain);
1003 #endif
1004 
1005 	for (;;) {
1006 		struct nfe_rx_data *data = &ring->data[ring->cur];
1007 		struct mbuf *m;
1008 		uint16_t flags;
1009 		int len, error;
1010 
1011 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1012 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1013 
1014 			flags = le16toh(desc64->flags);
1015 			len = le16toh(desc64->length) & 0x3fff;
1016 		} else {
1017 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1018 
1019 			flags = le16toh(desc32->flags);
1020 			len = le16toh(desc32->length) & 0x3fff;
1021 		}
1022 
1023 		if (flags & NFE_RX_READY)
1024 			break;
1025 
1026 		reap = 1;
1027 
1028 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1029 			if (!(flags & NFE_RX_VALID_V1))
1030 				goto skip;
1031 
1032 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1033 				flags &= ~NFE_RX_ERROR;
1034 				len--;	/* fix buffer length */
1035 			}
1036 		} else {
1037 			if (!(flags & NFE_RX_VALID_V2))
1038 				goto skip;
1039 
1040 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1041 				flags &= ~NFE_RX_ERROR;
1042 				len--;	/* fix buffer length */
1043 			}
1044 		}
1045 
1046 		if (flags & NFE_RX_ERROR) {
1047 			ifp->if_ierrors++;
1048 			goto skip;
1049 		}
1050 
1051 		m = data->m;
1052 
1053 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1054 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1055 		else
1056 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1057 		if (error) {
1058 			ifp->if_ierrors++;
1059 			goto skip;
1060 		}
1061 
1062 		/* finalize mbuf */
1063 		m->m_pkthdr.len = m->m_len = len;
1064 		m->m_pkthdr.rcvif = ifp;
1065 
1066 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1067 		    (flags & NFE_RX_CSUMOK)) {
1068 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1069 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1070 							  CSUM_IP_VALID;
1071 			}
1072 
1073 			if (flags &
1074 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1075 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1076 							  CSUM_PSEUDO_HDR |
1077 							  CSUM_FRAG_NOT_CHECKED;
1078 				m->m_pkthdr.csum_data = 0xffff;
1079 			}
1080 		}
1081 
1082 		ifp->if_ipackets++;
1083 #ifdef ETHER_INPUT_CHAIN
1084 		ether_input_chain2(ifp, m, chain);
1085 #else
1086 		ifp->if_input(ifp, m);
1087 #endif
1088 skip:
1089 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1090 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1091 	}
1092 
1093 	if (reap) {
1094 		bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1095 #ifdef ETHER_INPUT_CHAIN
1096 		ether_input_dispatch(chain);
1097 #endif
1098 	}
1099 	return reap;
1100 }
1101 
1102 static int
1103 nfe_txeof(struct nfe_softc *sc, int start)
1104 {
1105 	struct ifnet *ifp = &sc->arpcom.ac_if;
1106 	struct nfe_tx_ring *ring = &sc->txq;
1107 	struct nfe_tx_data *data = NULL;
1108 
1109 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
1110 	while (ring->next != ring->cur) {
1111 		uint16_t flags;
1112 
1113 		if (sc->sc_caps & NFE_40BIT_ADDR)
1114 			flags = le16toh(ring->desc64[ring->next].flags);
1115 		else
1116 			flags = le16toh(ring->desc32[ring->next].flags);
1117 
1118 		if (flags & NFE_TX_VALID)
1119 			break;
1120 
1121 		data = &ring->data[ring->next];
1122 
1123 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1124 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1125 				goto skip;
1126 
1127 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1128 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
1129 					  NFE_V1_TXERR);
1130 				ifp->if_oerrors++;
1131 			} else {
1132 				ifp->if_opackets++;
1133 			}
1134 		} else {
1135 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1136 				goto skip;
1137 
1138 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1139 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
1140 					  NFE_V2_TXERR);
1141 				ifp->if_oerrors++;
1142 			} else {
1143 				ifp->if_opackets++;
1144 			}
1145 		}
1146 
1147 		if (data->m == NULL) {	/* should not get there */
1148 			if_printf(ifp,
1149 				  "last fragment bit w/o associated mbuf!\n");
1150 			goto skip;
1151 		}
1152 
1153 		/* last fragment of the mbuf chain transmitted */
1154 		bus_dmamap_sync(ring->data_tag, data->map,
1155 				BUS_DMASYNC_POSTWRITE);
1156 		bus_dmamap_unload(ring->data_tag, data->map);
1157 		m_freem(data->m);
1158 		data->m = NULL;
1159 skip:
1160 		ring->queued--;
1161 		KKASSERT(ring->queued >= 0);
1162 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1163 	}
1164 
1165 	if (sc->sc_tx_ring_count - ring->queued >=
1166 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1167 		ifp->if_flags &= ~IFF_OACTIVE;
1168 
1169 	if (ring->queued == 0)
1170 		ifp->if_timer = 0;
1171 
1172 	if (start && !ifq_is_empty(&ifp->if_snd))
1173 		if_devstart(ifp);
1174 
1175 	if (data != NULL)
1176 		return 1;
1177 	else
1178 		return 0;
1179 }
1180 
1181 static int
1182 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1183 {
1184 	struct nfe_dma_ctx ctx;
1185 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1186 	struct nfe_tx_data *data, *data_map;
1187 	bus_dmamap_t map;
1188 	struct nfe_desc64 *desc64 = NULL;
1189 	struct nfe_desc32 *desc32 = NULL;
1190 	uint16_t flags = 0;
1191 	uint32_t vtag = 0;
1192 	int error, i, j, maxsegs;
1193 
1194 	data = &ring->data[ring->cur];
1195 	map = data->map;
1196 	data_map = data;	/* Remember who owns the DMA map */
1197 
1198 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1199 	if (maxsegs > NFE_MAX_SCATTER)
1200 		maxsegs = NFE_MAX_SCATTER;
1201 	KASSERT(maxsegs >= sc->sc_tx_spare,
1202 		("no enough segments %d,%d\n", maxsegs, sc->sc_tx_spare));
1203 
1204 	ctx.nsegs = maxsegs;
1205 	ctx.segs = segs;
1206 	error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
1207 				     nfe_buf_dma_addr, &ctx, BUS_DMA_NOWAIT);
1208 	if (!error && ctx.nsegs == 0) {
1209 		bus_dmamap_unload(ring->data_tag, map);
1210 		error = EFBIG;
1211 	}
1212 	if (error && error != EFBIG) {
1213 		if_printf(&sc->arpcom.ac_if, "could not map TX mbuf\n");
1214 		goto back;
1215 	}
1216 	if (error) {	/* error == EFBIG */
1217 		struct mbuf *m_new;
1218 
1219 		m_new = m_defrag(m0, MB_DONTWAIT);
1220 		if (m_new == NULL) {
1221 			if_printf(&sc->arpcom.ac_if,
1222 				  "could not defrag TX mbuf\n");
1223 			error = ENOBUFS;
1224 			goto back;
1225 		} else {
1226 			m0 = m_new;
1227 		}
1228 
1229 		ctx.nsegs = maxsegs;
1230 		ctx.segs = segs;
1231 		error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
1232 					     nfe_buf_dma_addr, &ctx,
1233 					     BUS_DMA_NOWAIT);
1234 		if (error || ctx.nsegs == 0) {
1235 			if (!error) {
1236 				bus_dmamap_unload(ring->data_tag, map);
1237 				error = EFBIG;
1238 			}
1239 			if_printf(&sc->arpcom.ac_if,
1240 				  "could not map defraged TX mbuf\n");
1241 			goto back;
1242 		}
1243 	}
1244 
1245 	error = 0;
1246 
1247 	/* setup h/w VLAN tagging */
1248 	if (m0->m_flags & M_VLANTAG)
1249 		vtag = m0->m_pkthdr.ether_vlantag;
1250 
1251 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1252 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1253 			flags |= NFE_TX_IP_CSUM;
1254 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1255 			flags |= NFE_TX_TCP_CSUM;
1256 	}
1257 
1258 	/*
1259 	 * XXX urm. somebody is unaware of how hardware works.  You
1260 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1261 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1262 	 * the hardware may encounter a partially initialized chain that
1263 	 * is marked as being ready to go when it in fact is not ready to
1264 	 * go.
1265 	 */
1266 
1267 	for (i = 0; i < ctx.nsegs; i++) {
1268 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1269 		data = &ring->data[j];
1270 
1271 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1272 			desc64 = &ring->desc64[j];
1273 #if defined(__LP64__)
1274 			desc64->physaddr[0] =
1275 			    htole32(segs[i].ds_addr >> 32);
1276 #endif
1277 			desc64->physaddr[1] =
1278 			    htole32(segs[i].ds_addr & 0xffffffff);
1279 			desc64->length = htole16(segs[i].ds_len - 1);
1280 			desc64->vtag = htole32(vtag);
1281 			desc64->flags = htole16(flags);
1282 		} else {
1283 			desc32 = &ring->desc32[j];
1284 			desc32->physaddr = htole32(segs[i].ds_addr);
1285 			desc32->length = htole16(segs[i].ds_len - 1);
1286 			desc32->flags = htole16(flags);
1287 		}
1288 
1289 		/* csum flags and vtag belong to the first fragment only */
1290 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1291 		vtag = 0;
1292 
1293 		ring->queued++;
1294 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1295 	}
1296 
1297 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1298 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1299 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1300 	} else {
1301 		if (sc->sc_caps & NFE_JUMBO_SUP)
1302 			flags = NFE_TX_LASTFRAG_V2;
1303 		else
1304 			flags = NFE_TX_LASTFRAG_V1;
1305 		desc32->flags |= htole16(flags);
1306 	}
1307 
1308 	/*
1309 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1310 	 * whole mess until the first descriptor in the map is flagged.
1311 	 */
1312 	for (i = ctx.nsegs - 1; i >= 0; --i) {
1313 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1314 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1315 			desc64 = &ring->desc64[j];
1316 			desc64->flags |= htole16(NFE_TX_VALID);
1317 		} else {
1318 			desc32 = &ring->desc32[j];
1319 			desc32->flags |= htole16(NFE_TX_VALID);
1320 		}
1321 	}
1322 	ring->cur = (ring->cur + ctx.nsegs) % sc->sc_tx_ring_count;
1323 
1324 	/* Exchange DMA map */
1325 	data_map->map = data->map;
1326 	data->map = map;
1327 	data->m = m0;
1328 
1329 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1330 back:
1331 	if (error)
1332 		m_freem(m0);
1333 	return error;
1334 }
1335 
1336 static void
1337 nfe_start(struct ifnet *ifp)
1338 {
1339 	struct nfe_softc *sc = ifp->if_softc;
1340 	struct nfe_tx_ring *ring = &sc->txq;
1341 	int count = 0, oactive = 0;
1342 	struct mbuf *m0;
1343 
1344 	ASSERT_SERIALIZED(ifp->if_serializer);
1345 
1346 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
1347 		return;
1348 
1349 	for (;;) {
1350 		int error;
1351 
1352 		if (sc->sc_tx_ring_count - ring->queued <
1353 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1354 			if (oactive) {
1355 				ifp->if_flags |= IFF_OACTIVE;
1356 				break;
1357 			}
1358 
1359 			nfe_txeof(sc, 0);
1360 			oactive = 1;
1361 			continue;
1362 		}
1363 
1364 		m0 = ifq_dequeue(&ifp->if_snd, NULL);
1365 		if (m0 == NULL)
1366 			break;
1367 
1368 		ETHER_BPF_MTAP(ifp, m0);
1369 
1370 		error = nfe_encap(sc, ring, m0);
1371 		if (error) {
1372 			ifp->if_oerrors++;
1373 			if (error == EFBIG) {
1374 				if (oactive) {
1375 					ifp->if_flags |= IFF_OACTIVE;
1376 					break;
1377 				}
1378 				nfe_txeof(sc, 0);
1379 				oactive = 1;
1380 			}
1381 			continue;
1382 		} else {
1383 			oactive = 0;
1384 		}
1385 		++count;
1386 
1387 		/*
1388 		 * NOTE:
1389 		 * `m0' may be freed in nfe_encap(), so
1390 		 * it should not be touched any more.
1391 		 */
1392 	}
1393 	if (count == 0)	/* nothing sent */
1394 		return;
1395 
1396 	/* Sync TX descriptor ring */
1397 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1398 
1399 	/* Kick Tx */
1400 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1401 
1402 	/*
1403 	 * Set a timeout in case the chip goes out to lunch.
1404 	 */
1405 	ifp->if_timer = 5;
1406 }
1407 
1408 static void
1409 nfe_watchdog(struct ifnet *ifp)
1410 {
1411 	struct nfe_softc *sc = ifp->if_softc;
1412 
1413 	ASSERT_SERIALIZED(ifp->if_serializer);
1414 
1415 	if (ifp->if_flags & IFF_RUNNING) {
1416 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1417 		nfe_txeof(sc, 1);
1418 		return;
1419 	}
1420 
1421 	if_printf(ifp, "watchdog timeout\n");
1422 
1423 	nfe_init(ifp->if_softc);
1424 
1425 	ifp->if_oerrors++;
1426 }
1427 
1428 static void
1429 nfe_init(void *xsc)
1430 {
1431 	struct nfe_softc *sc = xsc;
1432 	struct ifnet *ifp = &sc->arpcom.ac_if;
1433 	uint32_t tmp;
1434 	int error;
1435 
1436 	ASSERT_SERIALIZED(ifp->if_serializer);
1437 
1438 	nfe_stop(sc);
1439 
1440 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1441 		nfe_mac_reset(sc);
1442 
1443 	/*
1444 	 * NOTE:
1445 	 * Switching between jumbo frames and normal frames should
1446 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1447 	 */
1448 	if (ifp->if_mtu > ETHERMTU) {
1449 		sc->sc_flags |= NFE_F_USE_JUMBO;
1450 		sc->rxq.bufsz = NFE_JBYTES;
1451 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1452 		if (bootverbose)
1453 			if_printf(ifp, "use jumbo frames\n");
1454 	} else {
1455 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1456 		sc->rxq.bufsz = MCLBYTES;
1457 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1458 		if (bootverbose)
1459 			if_printf(ifp, "use non-jumbo frames\n");
1460 	}
1461 
1462 	error = nfe_init_tx_ring(sc, &sc->txq);
1463 	if (error) {
1464 		nfe_stop(sc);
1465 		return;
1466 	}
1467 
1468 	error = nfe_init_rx_ring(sc, &sc->rxq);
1469 	if (error) {
1470 		nfe_stop(sc);
1471 		return;
1472 	}
1473 
1474 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1475 	NFE_WRITE(sc, NFE_STATUS, 0);
1476 
1477 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1478 
1479 	if (ifp->if_capenable & IFCAP_RXCSUM)
1480 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1481 
1482 	/*
1483 	 * Although the adapter is capable of stripping VLAN tags from received
1484 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1485 	 * purpose.  This will be done in software by our network stack.
1486 	 */
1487 	if (sc->sc_caps & NFE_HW_VLAN)
1488 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1489 
1490 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1491 	DELAY(10);
1492 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1493 
1494 	if (sc->sc_caps & NFE_HW_VLAN)
1495 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1496 
1497 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1498 
1499 	/* set MAC address */
1500 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1501 
1502 	/* tell MAC where rings are in memory */
1503 #ifdef __LP64__
1504 	NFE_WRITE(sc, NFE_RX_RING_ADDR_HI, sc->rxq.physaddr >> 32);
1505 #endif
1506 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, sc->rxq.physaddr & 0xffffffff);
1507 #ifdef __LP64__
1508 	NFE_WRITE(sc, NFE_TX_RING_ADDR_HI, sc->txq.physaddr >> 32);
1509 #endif
1510 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, sc->txq.physaddr & 0xffffffff);
1511 
1512 	NFE_WRITE(sc, NFE_RING_SIZE,
1513 	    (sc->sc_rx_ring_count - 1) << 16 |
1514 	    (sc->sc_tx_ring_count - 1));
1515 
1516 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1517 
1518 	/* force MAC to wakeup */
1519 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1520 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1521 	DELAY(10);
1522 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1523 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1524 
1525 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1526 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1527 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1528 
1529 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1530 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1531 
1532 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1533 
1534 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1535 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1536 	DELAY(10);
1537 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1538 
1539 	/* set Rx filter */
1540 	nfe_setmulti(sc);
1541 
1542 	nfe_ifmedia_upd(ifp);
1543 
1544 	/* enable Rx */
1545 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1546 
1547 	/* enable Tx */
1548 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1549 
1550 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1551 
1552 #ifdef DEVICE_POLLING
1553 	if ((ifp->if_flags & IFF_POLLING))
1554 		nfe_disable_intrs(sc);
1555 	else
1556 #endif
1557 	nfe_enable_intrs(sc);
1558 
1559 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1560 
1561 	ifp->if_flags |= IFF_RUNNING;
1562 	ifp->if_flags &= ~IFF_OACTIVE;
1563 
1564 	/*
1565 	 * If we had stuff in the tx ring before its all cleaned out now
1566 	 * so we are not going to get an interrupt, jump-start any pending
1567 	 * output.
1568 	 */
1569 	if (!ifq_is_empty(&ifp->if_snd))
1570 		if_devstart(ifp);
1571 }
1572 
1573 static void
1574 nfe_stop(struct nfe_softc *sc)
1575 {
1576 	struct ifnet *ifp = &sc->arpcom.ac_if;
1577 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1578 	int i;
1579 
1580 	ASSERT_SERIALIZED(ifp->if_serializer);
1581 
1582 	callout_stop(&sc->sc_tick_ch);
1583 
1584 	ifp->if_timer = 0;
1585 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1586 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1587 
1588 #define WAITMAX	50000
1589 
1590 	/*
1591 	 * Abort Tx
1592 	 */
1593 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1594 	for (i = 0; i < WAITMAX; ++i) {
1595 		DELAY(100);
1596 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1597 			break;
1598 	}
1599 	if (i == WAITMAX)
1600 		if_printf(ifp, "can't stop TX\n");
1601 	DELAY(100);
1602 
1603 	/*
1604 	 * Disable Rx
1605 	 */
1606 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1607 	for (i = 0; i < WAITMAX; ++i) {
1608 		DELAY(100);
1609 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1610 			break;
1611 	}
1612 	if (i == WAITMAX)
1613 		if_printf(ifp, "can't stop RX\n");
1614 	DELAY(100);
1615 
1616 #undef WAITMAX
1617 
1618 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1619 	DELAY(10);
1620 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1621 
1622 	/* Disable interrupts */
1623 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1624 
1625 	/* Reset Tx and Rx rings */
1626 	nfe_reset_tx_ring(sc, &sc->txq);
1627 	nfe_reset_rx_ring(sc, &sc->rxq);
1628 }
1629 
1630 static int
1631 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1632 {
1633 	int i, j, error, descsize;
1634 	void **desc;
1635 
1636 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1637 		desc = (void **)&ring->desc64;
1638 		descsize = sizeof(struct nfe_desc64);
1639 	} else {
1640 		desc = (void **)&ring->desc32;
1641 		descsize = sizeof(struct nfe_desc32);
1642 	}
1643 
1644 	ring->bufsz = MCLBYTES;
1645 	ring->cur = ring->next = 0;
1646 
1647 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1648 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1649 				   NULL, NULL,
1650 				   sc->sc_rx_ring_count * descsize, 1,
1651 				   BUS_SPACE_MAXSIZE_32BIT,
1652 				   0, &ring->tag);
1653 	if (error) {
1654 		if_printf(&sc->arpcom.ac_if,
1655 			  "could not create desc RX DMA tag\n");
1656 		return error;
1657 	}
1658 
1659 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1660 				 &ring->map);
1661 	if (error) {
1662 		if_printf(&sc->arpcom.ac_if,
1663 			  "could not allocate RX desc DMA memory\n");
1664 		bus_dma_tag_destroy(ring->tag);
1665 		ring->tag = NULL;
1666 		return error;
1667 	}
1668 
1669 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1670 				sc->sc_rx_ring_count * descsize,
1671 				nfe_ring_dma_addr, &ring->physaddr,
1672 				BUS_DMA_WAITOK);
1673 	if (error) {
1674 		if_printf(&sc->arpcom.ac_if,
1675 			  "could not load RX desc DMA map\n");
1676 		bus_dmamem_free(ring->tag, *desc, ring->map);
1677 		bus_dma_tag_destroy(ring->tag);
1678 		ring->tag = NULL;
1679 		return error;
1680 	}
1681 
1682 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1683 		ring->jbuf =
1684 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1685 			M_DEVBUF, M_WAITOK | M_ZERO);
1686 
1687 		error = nfe_jpool_alloc(sc, ring);
1688 		if (error) {
1689 			if_printf(&sc->arpcom.ac_if,
1690 				  "could not allocate jumbo frames\n");
1691 			kfree(ring->jbuf, M_DEVBUF);
1692 			ring->jbuf = NULL;
1693 			/* Allow jumbo frame allocation to fail */
1694 		}
1695 	}
1696 
1697 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1698 			     M_DEVBUF, M_WAITOK | M_ZERO);
1699 
1700 	error = bus_dma_tag_create(NULL, 1, 0,
1701 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1702 				   NULL, NULL,
1703 				   MCLBYTES, 1, BUS_SPACE_MAXSIZE_32BIT,
1704 				   BUS_DMA_ALLOCNOW, &ring->data_tag);
1705 	if (error) {
1706 		if_printf(&sc->arpcom.ac_if,
1707 			  "could not create RX mbuf DMA tag\n");
1708 		return error;
1709 	}
1710 
1711 	/* Create a spare RX mbuf DMA map */
1712 	error = bus_dmamap_create(ring->data_tag, 0, &ring->data_tmpmap);
1713 	if (error) {
1714 		if_printf(&sc->arpcom.ac_if,
1715 			  "could not create spare RX mbuf DMA map\n");
1716 		bus_dma_tag_destroy(ring->data_tag);
1717 		ring->data_tag = NULL;
1718 		return error;
1719 	}
1720 
1721 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1722 		error = bus_dmamap_create(ring->data_tag, 0,
1723 					  &ring->data[i].map);
1724 		if (error) {
1725 			if_printf(&sc->arpcom.ac_if,
1726 				  "could not create %dth RX mbuf DMA mapn", i);
1727 			goto fail;
1728 		}
1729 	}
1730 	return 0;
1731 fail:
1732 	for (j = 0; j < i; ++j)
1733 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1734 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1735 	bus_dma_tag_destroy(ring->data_tag);
1736 	ring->data_tag = NULL;
1737 	return error;
1738 }
1739 
1740 static void
1741 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1742 {
1743 	int i;
1744 
1745 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1746 		struct nfe_rx_data *data = &ring->data[i];
1747 
1748 		if (data->m != NULL) {
1749 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1750 				bus_dmamap_unload(ring->data_tag, data->map);
1751 			m_freem(data->m);
1752 			data->m = NULL;
1753 		}
1754 	}
1755 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1756 
1757 	ring->cur = ring->next = 0;
1758 }
1759 
1760 static int
1761 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1762 {
1763 	int i;
1764 
1765 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1766 		int error;
1767 
1768 		/* XXX should use a function pointer */
1769 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1770 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1771 		else
1772 			error = nfe_newbuf_std(sc, ring, i, 1);
1773 		if (error) {
1774 			if_printf(&sc->arpcom.ac_if,
1775 				  "could not allocate RX buffer\n");
1776 			return error;
1777 		}
1778 
1779 		nfe_set_ready_rxdesc(sc, ring, i);
1780 	}
1781 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1782 
1783 	return 0;
1784 }
1785 
1786 static void
1787 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1788 {
1789 	if (ring->data_tag != NULL) {
1790 		struct nfe_rx_data *data;
1791 		int i;
1792 
1793 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1794 			data = &ring->data[i];
1795 
1796 			if (data->m != NULL) {
1797 				bus_dmamap_unload(ring->data_tag, data->map);
1798 				m_freem(data->m);
1799 			}
1800 			bus_dmamap_destroy(ring->data_tag, data->map);
1801 		}
1802 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1803 		bus_dma_tag_destroy(ring->data_tag);
1804 	}
1805 
1806 	nfe_jpool_free(sc, ring);
1807 
1808 	if (ring->jbuf != NULL)
1809 		kfree(ring->jbuf, M_DEVBUF);
1810 	if (ring->data != NULL)
1811 		kfree(ring->data, M_DEVBUF);
1812 
1813 	if (ring->tag != NULL) {
1814 		void *desc;
1815 
1816 		if (sc->sc_caps & NFE_40BIT_ADDR)
1817 			desc = ring->desc64;
1818 		else
1819 			desc = ring->desc32;
1820 
1821 		bus_dmamap_unload(ring->tag, ring->map);
1822 		bus_dmamem_free(ring->tag, desc, ring->map);
1823 		bus_dma_tag_destroy(ring->tag);
1824 	}
1825 }
1826 
1827 static struct nfe_jbuf *
1828 nfe_jalloc(struct nfe_softc *sc)
1829 {
1830 	struct ifnet *ifp = &sc->arpcom.ac_if;
1831 	struct nfe_jbuf *jbuf;
1832 
1833 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1834 
1835 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1836 	if (jbuf != NULL) {
1837 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1838 		jbuf->inuse = 1;
1839 	} else {
1840 		if_printf(ifp, "no free jumbo buffer\n");
1841 	}
1842 
1843 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1844 
1845 	return jbuf;
1846 }
1847 
1848 static void
1849 nfe_jfree(void *arg)
1850 {
1851 	struct nfe_jbuf *jbuf = arg;
1852 	struct nfe_softc *sc = jbuf->sc;
1853 	struct nfe_rx_ring *ring = jbuf->ring;
1854 
1855 	if (&ring->jbuf[jbuf->slot] != jbuf)
1856 		panic("%s: free wrong jumbo buffer\n", __func__);
1857 	else if (jbuf->inuse == 0)
1858 		panic("%s: jumbo buffer already freed\n", __func__);
1859 
1860 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1861 	atomic_subtract_int(&jbuf->inuse, 1);
1862 	if (jbuf->inuse == 0)
1863 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1864 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1865 }
1866 
1867 static void
1868 nfe_jref(void *arg)
1869 {
1870 	struct nfe_jbuf *jbuf = arg;
1871 	struct nfe_rx_ring *ring = jbuf->ring;
1872 
1873 	if (&ring->jbuf[jbuf->slot] != jbuf)
1874 		panic("%s: ref wrong jumbo buffer\n", __func__);
1875 	else if (jbuf->inuse == 0)
1876 		panic("%s: jumbo buffer already freed\n", __func__);
1877 
1878 	atomic_add_int(&jbuf->inuse, 1);
1879 }
1880 
1881 static int
1882 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1883 {
1884 	struct nfe_jbuf *jbuf;
1885 	bus_addr_t physaddr;
1886 	caddr_t buf;
1887 	int i, error;
1888 
1889 	/*
1890 	 * Allocate a big chunk of DMA'able memory.
1891 	 */
1892 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1893 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1894 				   NULL, NULL,
1895 				   NFE_JPOOL_SIZE(sc), 1,
1896 				   BUS_SPACE_MAXSIZE_32BIT,
1897 				   0, &ring->jtag);
1898 	if (error) {
1899 		if_printf(&sc->arpcom.ac_if,
1900 			  "could not create jumbo DMA tag\n");
1901 		return error;
1902 	}
1903 
1904 	error = bus_dmamem_alloc(ring->jtag, (void **)&ring->jpool,
1905 				 BUS_DMA_WAITOK, &ring->jmap);
1906 	if (error) {
1907 		if_printf(&sc->arpcom.ac_if,
1908 			  "could not allocate jumbo DMA memory\n");
1909 		bus_dma_tag_destroy(ring->jtag);
1910 		ring->jtag = NULL;
1911 		return error;
1912 	}
1913 
1914 	error = bus_dmamap_load(ring->jtag, ring->jmap, ring->jpool,
1915 				NFE_JPOOL_SIZE(sc),
1916 				nfe_ring_dma_addr, &physaddr, BUS_DMA_WAITOK);
1917 	if (error) {
1918 		if_printf(&sc->arpcom.ac_if,
1919 			  "could not load jumbo DMA map\n");
1920 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1921 		bus_dma_tag_destroy(ring->jtag);
1922 		ring->jtag = NULL;
1923 		return error;
1924 	}
1925 
1926 	/* ..and split it into 9KB chunks */
1927 	SLIST_INIT(&ring->jfreelist);
1928 
1929 	buf = ring->jpool;
1930 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1931 		jbuf = &ring->jbuf[i];
1932 
1933 		jbuf->sc = sc;
1934 		jbuf->ring = ring;
1935 		jbuf->inuse = 0;
1936 		jbuf->slot = i;
1937 		jbuf->buf = buf;
1938 		jbuf->physaddr = physaddr;
1939 
1940 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1941 
1942 		buf += NFE_JBYTES;
1943 		physaddr += NFE_JBYTES;
1944 	}
1945 
1946 	return 0;
1947 }
1948 
1949 static void
1950 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1951 {
1952 	if (ring->jtag != NULL) {
1953 		bus_dmamap_unload(ring->jtag, ring->jmap);
1954 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1955 		bus_dma_tag_destroy(ring->jtag);
1956 	}
1957 }
1958 
1959 static int
1960 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1961 {
1962 	int i, j, error, descsize;
1963 	void **desc;
1964 
1965 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1966 		desc = (void **)&ring->desc64;
1967 		descsize = sizeof(struct nfe_desc64);
1968 	} else {
1969 		desc = (void **)&ring->desc32;
1970 		descsize = sizeof(struct nfe_desc32);
1971 	}
1972 
1973 	ring->queued = 0;
1974 	ring->cur = ring->next = 0;
1975 
1976 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1977 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1978 				   NULL, NULL,
1979 				   sc->sc_tx_ring_count * descsize, 1,
1980 				   BUS_SPACE_MAXSIZE_32BIT,
1981 				   0, &ring->tag);
1982 	if (error) {
1983 		if_printf(&sc->arpcom.ac_if,
1984 			  "could not create TX desc DMA map\n");
1985 		return error;
1986 	}
1987 
1988 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1989 				 &ring->map);
1990 	if (error) {
1991 		if_printf(&sc->arpcom.ac_if,
1992 			  "could not allocate TX desc DMA memory\n");
1993 		bus_dma_tag_destroy(ring->tag);
1994 		ring->tag = NULL;
1995 		return error;
1996 	}
1997 
1998 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1999 				sc->sc_tx_ring_count * descsize,
2000 				nfe_ring_dma_addr, &ring->physaddr,
2001 				BUS_DMA_WAITOK);
2002 	if (error) {
2003 		if_printf(&sc->arpcom.ac_if,
2004 			  "could not load TX desc DMA map\n");
2005 		bus_dmamem_free(ring->tag, *desc, ring->map);
2006 		bus_dma_tag_destroy(ring->tag);
2007 		ring->tag = NULL;
2008 		return error;
2009 	}
2010 
2011 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
2012 			     M_DEVBUF, M_WAITOK | M_ZERO);
2013 
2014 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
2015 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
2016 				   NULL, NULL,
2017 				   NFE_JBYTES, NFE_MAX_SCATTER,
2018 				   BUS_SPACE_MAXSIZE_32BIT,
2019 				   BUS_DMA_ALLOCNOW, &ring->data_tag);
2020 	if (error) {
2021 		if_printf(&sc->arpcom.ac_if,
2022 			  "could not create TX buf DMA tag\n");
2023 		return error;
2024 	}
2025 
2026 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
2027 		error = bus_dmamap_create(ring->data_tag, 0,
2028 					  &ring->data[i].map);
2029 		if (error) {
2030 			if_printf(&sc->arpcom.ac_if,
2031 				  "could not create %dth TX buf DMA map\n", i);
2032 			goto fail;
2033 		}
2034 	}
2035 
2036 	return 0;
2037 fail:
2038 	for (j = 0; j < i; ++j)
2039 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
2040 	bus_dma_tag_destroy(ring->data_tag);
2041 	ring->data_tag = NULL;
2042 	return error;
2043 }
2044 
2045 static void
2046 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2047 {
2048 	int i;
2049 
2050 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
2051 		struct nfe_tx_data *data = &ring->data[i];
2052 
2053 		if (sc->sc_caps & NFE_40BIT_ADDR)
2054 			ring->desc64[i].flags = 0;
2055 		else
2056 			ring->desc32[i].flags = 0;
2057 
2058 		if (data->m != NULL) {
2059 			bus_dmamap_sync(ring->data_tag, data->map,
2060 					BUS_DMASYNC_POSTWRITE);
2061 			bus_dmamap_unload(ring->data_tag, data->map);
2062 			m_freem(data->m);
2063 			data->m = NULL;
2064 		}
2065 	}
2066 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
2067 
2068 	ring->queued = 0;
2069 	ring->cur = ring->next = 0;
2070 }
2071 
2072 static int
2073 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2074 		 struct nfe_tx_ring *ring __unused)
2075 {
2076 	return 0;
2077 }
2078 
2079 static void
2080 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2081 {
2082 	if (ring->data_tag != NULL) {
2083 		struct nfe_tx_data *data;
2084 		int i;
2085 
2086 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2087 			data = &ring->data[i];
2088 
2089 			if (data->m != NULL) {
2090 				bus_dmamap_unload(ring->data_tag, data->map);
2091 				m_freem(data->m);
2092 			}
2093 			bus_dmamap_destroy(ring->data_tag, data->map);
2094 		}
2095 
2096 		bus_dma_tag_destroy(ring->data_tag);
2097 	}
2098 
2099 	if (ring->data != NULL)
2100 		kfree(ring->data, M_DEVBUF);
2101 
2102 	if (ring->tag != NULL) {
2103 		void *desc;
2104 
2105 		if (sc->sc_caps & NFE_40BIT_ADDR)
2106 			desc = ring->desc64;
2107 		else
2108 			desc = ring->desc32;
2109 
2110 		bus_dmamap_unload(ring->tag, ring->map);
2111 		bus_dmamem_free(ring->tag, desc, ring->map);
2112 		bus_dma_tag_destroy(ring->tag);
2113 	}
2114 }
2115 
2116 static int
2117 nfe_ifmedia_upd(struct ifnet *ifp)
2118 {
2119 	struct nfe_softc *sc = ifp->if_softc;
2120 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2121 
2122 	ASSERT_SERIALIZED(ifp->if_serializer);
2123 
2124 	if (mii->mii_instance != 0) {
2125 		struct mii_softc *miisc;
2126 
2127 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2128 			mii_phy_reset(miisc);
2129 	}
2130 	mii_mediachg(mii);
2131 
2132 	return 0;
2133 }
2134 
2135 static void
2136 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2137 {
2138 	struct nfe_softc *sc = ifp->if_softc;
2139 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2140 
2141 	ASSERT_SERIALIZED(ifp->if_serializer);
2142 
2143 	mii_pollstat(mii);
2144 	ifmr->ifm_status = mii->mii_media_status;
2145 	ifmr->ifm_active = mii->mii_media_active;
2146 }
2147 
2148 static void
2149 nfe_setmulti(struct nfe_softc *sc)
2150 {
2151 	struct ifnet *ifp = &sc->arpcom.ac_if;
2152 	struct ifmultiaddr *ifma;
2153 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2154 	uint32_t filter = NFE_RXFILTER_MAGIC;
2155 	int i;
2156 
2157 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2158 		bzero(addr, ETHER_ADDR_LEN);
2159 		bzero(mask, ETHER_ADDR_LEN);
2160 		goto done;
2161 	}
2162 
2163 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2164 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2165 
2166 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2167 		caddr_t maddr;
2168 
2169 		if (ifma->ifma_addr->sa_family != AF_LINK)
2170 			continue;
2171 
2172 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2173 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2174 			addr[i] &= maddr[i];
2175 			mask[i] &= ~maddr[i];
2176 		}
2177 	}
2178 
2179 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2180 		mask[i] |= addr[i];
2181 
2182 done:
2183 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2184 
2185 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2186 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2187 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2188 	    addr[5] <<  8 | addr[4]);
2189 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2190 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2191 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2192 	    mask[5] <<  8 | mask[4]);
2193 
2194 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2195 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2196 }
2197 
2198 static void
2199 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2200 {
2201 	uint32_t lo, hi;
2202 
2203 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2204 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2205 	if (sc->sc_caps & NFE_FIX_EADDR) {
2206 		addr[0] = (lo >> 8) & 0xff;
2207 		addr[1] = (lo & 0xff);
2208 
2209 		addr[2] = (hi >> 24) & 0xff;
2210 		addr[3] = (hi >> 16) & 0xff;
2211 		addr[4] = (hi >>  8) & 0xff;
2212 		addr[5] = (hi & 0xff);
2213 	} else {
2214 		addr[0] = (hi & 0xff);
2215 		addr[1] = (hi >>  8) & 0xff;
2216 		addr[2] = (hi >> 16) & 0xff;
2217 		addr[3] = (hi >> 24) & 0xff;
2218 
2219 		addr[4] = (lo & 0xff);
2220 		addr[5] = (lo >>  8) & 0xff;
2221 	}
2222 }
2223 
2224 static void
2225 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2226 {
2227 	NFE_WRITE(sc, NFE_MACADDR_LO,
2228 	    addr[5] <<  8 | addr[4]);
2229 	NFE_WRITE(sc, NFE_MACADDR_HI,
2230 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2231 }
2232 
2233 static void
2234 nfe_tick(void *arg)
2235 {
2236 	struct nfe_softc *sc = arg;
2237 	struct ifnet *ifp = &sc->arpcom.ac_if;
2238 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2239 
2240 	lwkt_serialize_enter(ifp->if_serializer);
2241 
2242 	mii_tick(mii);
2243 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2244 
2245 	lwkt_serialize_exit(ifp->if_serializer);
2246 }
2247 
2248 static void
2249 nfe_ring_dma_addr(void *arg, bus_dma_segment_t *seg, int nseg, int error)
2250 {
2251 	if (error)
2252 		return;
2253 
2254 	KASSERT(nseg == 1, ("too many segments, should be 1\n"));
2255 
2256 	*((uint32_t *)arg) = seg->ds_addr;
2257 }
2258 
2259 static void
2260 nfe_buf_dma_addr(void *arg, bus_dma_segment_t *segs, int nsegs,
2261 		 bus_size_t mapsz __unused, int error)
2262 {
2263 	struct nfe_dma_ctx *ctx = arg;
2264 	int i;
2265 
2266 	if (error)
2267 		return;
2268 
2269 	if (nsegs > ctx->nsegs) {
2270 		ctx->nsegs = 0;
2271 		return;
2272 	}
2273 
2274 	ctx->nsegs = nsegs;
2275 	for (i = 0; i < nsegs; ++i)
2276 		ctx->segs[i] = segs[i];
2277 }
2278 
2279 static int
2280 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2281 	       int wait)
2282 {
2283 	struct nfe_rx_data *data = &ring->data[idx];
2284 	struct nfe_dma_ctx ctx;
2285 	bus_dma_segment_t seg;
2286 	bus_dmamap_t map;
2287 	struct mbuf *m;
2288 	int error;
2289 
2290 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2291 	if (m == NULL)
2292 		return ENOBUFS;
2293 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2294 
2295 	ctx.nsegs = 1;
2296 	ctx.segs = &seg;
2297 	error = bus_dmamap_load_mbuf(ring->data_tag, ring->data_tmpmap,
2298 				     m, nfe_buf_dma_addr, &ctx,
2299 				     wait ? BUS_DMA_WAITOK : BUS_DMA_NOWAIT);
2300 	if (error || ctx.nsegs == 0) {
2301 		if (!error) {
2302 			bus_dmamap_unload(ring->data_tag, ring->data_tmpmap);
2303 			error = EFBIG;
2304 			if_printf(&sc->arpcom.ac_if, "too many segments?!\n");
2305 		}
2306 		m_freem(m);
2307 
2308 		if (wait) {
2309 			if_printf(&sc->arpcom.ac_if,
2310 				  "could map RX mbuf %d\n", error);
2311 		}
2312 		return error;
2313 	}
2314 
2315 	/* Unload originally mapped mbuf */
2316 	bus_dmamap_unload(ring->data_tag, data->map);
2317 
2318 	/* Swap this DMA map with tmp DMA map */
2319 	map = data->map;
2320 	data->map = ring->data_tmpmap;
2321 	ring->data_tmpmap = map;
2322 
2323 	/* Caller is assumed to have collected the old mbuf */
2324 	data->m = m;
2325 
2326 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2327 
2328 	bus_dmamap_sync(ring->data_tag, data->map, BUS_DMASYNC_PREREAD);
2329 	return 0;
2330 }
2331 
2332 static int
2333 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2334 		 int wait)
2335 {
2336 	struct nfe_rx_data *data = &ring->data[idx];
2337 	struct nfe_jbuf *jbuf;
2338 	struct mbuf *m;
2339 
2340 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2341 	if (m == NULL)
2342 		return ENOBUFS;
2343 
2344 	jbuf = nfe_jalloc(sc);
2345 	if (jbuf == NULL) {
2346 		m_freem(m);
2347 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2348 		    "-- packet dropped!\n");
2349 		return ENOBUFS;
2350 	}
2351 
2352 	m->m_ext.ext_arg = jbuf;
2353 	m->m_ext.ext_buf = jbuf->buf;
2354 	m->m_ext.ext_free = nfe_jfree;
2355 	m->m_ext.ext_ref = nfe_jref;
2356 	m->m_ext.ext_size = NFE_JBYTES;
2357 
2358 	m->m_data = m->m_ext.ext_buf;
2359 	m->m_flags |= M_EXT;
2360 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2361 
2362 	/* Caller is assumed to have collected the old mbuf */
2363 	data->m = m;
2364 
2365 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2366 
2367 	bus_dmamap_sync(ring->jtag, ring->jmap, BUS_DMASYNC_PREREAD);
2368 	return 0;
2369 }
2370 
2371 static void
2372 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2373 		     bus_addr_t physaddr)
2374 {
2375 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2376 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2377 
2378 #if defined(__LP64__)
2379 		desc64->physaddr[0] = htole32(physaddr >> 32);
2380 #endif
2381 		desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
2382 	} else {
2383 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2384 
2385 		desc32->physaddr = htole32(physaddr);
2386 	}
2387 }
2388 
2389 static void
2390 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2391 {
2392 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2393 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2394 
2395 		desc64->length = htole16(ring->bufsz);
2396 		desc64->flags = htole16(NFE_RX_READY);
2397 	} else {
2398 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2399 
2400 		desc32->length = htole16(ring->bufsz);
2401 		desc32->flags = htole16(NFE_RX_READY);
2402 	}
2403 }
2404 
2405 static int
2406 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2407 {
2408 	struct nfe_softc *sc = arg1;
2409 	struct ifnet *ifp = &sc->arpcom.ac_if;
2410 	uint32_t flags;
2411 	int error, v;
2412 
2413 	lwkt_serialize_enter(ifp->if_serializer);
2414 
2415 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2416 	v = sc->sc_imtime;
2417 	if (sc->sc_flags & NFE_F_DYN_IM)
2418 		v = -v;
2419 
2420 	error = sysctl_handle_int(oidp, &v, 0, req);
2421 	if (error || req->newptr == NULL)
2422 		goto back;
2423 
2424 	if (v < 0) {
2425 		flags |= NFE_F_DYN_IM;
2426 		v = -v;
2427 	}
2428 
2429 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2430 		int old_imtime = sc->sc_imtime;
2431 		uint32_t old_flags = sc->sc_flags;
2432 
2433 		sc->sc_imtime = v;
2434 		sc->sc_flags = flags;
2435 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2436 
2437 		if ((ifp->if_flags & (IFF_POLLING | IFF_RUNNING))
2438 		    == IFF_RUNNING) {
2439 			if (old_imtime * sc->sc_imtime == 0 ||
2440 			    (old_flags ^ sc->sc_flags)) {
2441 				ifp->if_init(sc);
2442 			} else {
2443 				NFE_WRITE(sc, NFE_IMTIMER,
2444 					  NFE_IMTIME(sc->sc_imtime));
2445 			}
2446 		}
2447 	}
2448 back:
2449 	lwkt_serialize_exit(ifp->if_serializer);
2450 	return error;
2451 }
2452 
2453 static void
2454 nfe_powerup(device_t dev)
2455 {
2456 	struct nfe_softc *sc = device_get_softc(dev);
2457 	uint32_t pwr_state;
2458 	uint16_t did;
2459 
2460 	/*
2461 	 * Bring MAC and PHY out of low power state
2462 	 */
2463 
2464 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2465 
2466 	did = pci_get_device(dev);
2467 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2468 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2469 	    pci_get_revid(dev) >= 0xa3)
2470 		pwr_state |= NFE_PWRUP_REV_A3;
2471 
2472 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2473 }
2474 
2475 static void
2476 nfe_mac_reset(struct nfe_softc *sc)
2477 {
2478 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2479 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2480 
2481 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2482 
2483 	/* Save several registers for later restoration */
2484 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2485 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2486 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2487 
2488 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2489 	DELAY(100);
2490 
2491 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2492 	DELAY(100);
2493 
2494 	/* Restore saved registers */
2495 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2496 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2497 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2498 
2499 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2500 }
2501 
2502 static void
2503 nfe_enable_intrs(struct nfe_softc *sc)
2504 {
2505 	/*
2506 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2507 	 * It is unclear how wide the timer is.  Base programming does
2508 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2509 	 * we don't get any interrupt moderation.  TX moderation is
2510 	 * possible by using the timer interrupt instead of TX_DONE.
2511 	 *
2512 	 * It is unclear whether there are other bits that can be
2513 	 * set to make the NFE device actually do interrupt moderation
2514 	 * on the RX side.
2515 	 *
2516 	 * For now set a 128uS interval as a placemark, but don't use
2517 	 * the timer.
2518 	 */
2519 	if (sc->sc_imtime == 0)
2520 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2521 	else
2522 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2523 
2524 	/* Enable interrupts */
2525 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2526 
2527 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2528 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2529 	else
2530 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2531 }
2532 
2533 static void
2534 nfe_disable_intrs(struct nfe_softc *sc)
2535 {
2536 	/* Disable interrupts */
2537 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
2538 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2539 }
2540