xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision b5523eac)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Sepherosa Ziehau <sepherosa@gmail.com> and
8  * Matthew Dillon <dillon@apollo.backplane.com>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in
18  *    the documentation and/or other materials provided with the
19  *    distribution.
20  * 3. Neither the name of The DragonFly Project nor the names of its
21  *    contributors may be used to endorse or promote products derived
22  *    from this software without specific, prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
27  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
28  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
29  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
30  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
31  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
32  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
33  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
34  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
40  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
41  *
42  * Permission to use, copy, modify, and distribute this software for any
43  * purpose with or without fee is hereby granted, provided that the above
44  * copyright notice and this permission notice appear in all copies.
45  *
46  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
47  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
48  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
49  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
50  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
51  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
52  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
53  */
54 
55 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
56 
57 #include "opt_ifpoll.h"
58 
59 #include <sys/param.h>
60 #include <sys/endian.h>
61 #include <sys/kernel.h>
62 #include <sys/bus.h>
63 #include <sys/interrupt.h>
64 #include <sys/proc.h>
65 #include <sys/rman.h>
66 #include <sys/serialize.h>
67 #include <sys/socket.h>
68 #include <sys/sockio.h>
69 #include <sys/sysctl.h>
70 
71 #include <net/ethernet.h>
72 #include <net/if.h>
73 #include <net/bpf.h>
74 #include <net/if_arp.h>
75 #include <net/if_dl.h>
76 #include <net/if_media.h>
77 #include <net/if_poll.h>
78 #include <net/ifq_var.h>
79 #include <net/if_types.h>
80 #include <net/if_var.h>
81 #include <net/vlan/if_vlan_var.h>
82 #include <net/vlan/if_vlan_ether.h>
83 
84 #include <bus/pci/pcireg.h>
85 #include <bus/pci/pcivar.h>
86 #include "pcidevs.h"
87 
88 #include <dev/netif/mii_layer/mii.h>
89 #include <dev/netif/mii_layer/miivar.h>
90 
91 #include "miibus_if.h"
92 
93 #include <dev/netif/nfe/if_nfereg.h>
94 #include <dev/netif/nfe/if_nfevar.h>
95 
96 #define NFE_CSUM
97 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
98 
99 static int	nfe_probe(device_t);
100 static int	nfe_attach(device_t);
101 static int	nfe_detach(device_t);
102 static void	nfe_shutdown(device_t);
103 static int	nfe_resume(device_t);
104 static int	nfe_suspend(device_t);
105 
106 static int	nfe_miibus_readreg(device_t, int, int);
107 static void	nfe_miibus_writereg(device_t, int, int, int);
108 static void	nfe_miibus_statchg(device_t);
109 
110 #ifdef IFPOLL_ENABLE
111 static void	nfe_npoll(struct ifnet *, struct ifpoll_info *);
112 static void	nfe_npoll_compat(struct ifnet *, void *, int);
113 static void	nfe_disable_intrs(struct nfe_softc *);
114 #endif
115 static void	nfe_intr(void *);
116 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
117 static int	nfe_rxeof(struct nfe_softc *);
118 static int	nfe_txeof(struct nfe_softc *, int);
119 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
120 			  struct mbuf *);
121 static void	nfe_start(struct ifnet *, struct ifaltq_subque *);
122 static void	nfe_watchdog(struct ifnet *);
123 static void	nfe_init(void *);
124 static void	nfe_stop(struct nfe_softc *);
125 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
126 static void	nfe_jfree(void *);
127 static void	nfe_jref(void *);
128 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
129 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
130 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
131 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
132 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
133 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
134 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
135 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
136 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
137 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
138 static int	nfe_ifmedia_upd(struct ifnet *);
139 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
140 static void	nfe_setmulti(struct nfe_softc *);
141 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
142 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
143 static void	nfe_powerup(device_t);
144 static void	nfe_mac_reset(struct nfe_softc *);
145 static void	nfe_tick(void *);
146 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
147 				     int, bus_addr_t);
148 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
149 				     int);
150 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
151 			       int);
152 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
153 				 int);
154 static void	nfe_enable_intrs(struct nfe_softc *);
155 
156 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
157 
158 #define NFE_DEBUG
159 #ifdef NFE_DEBUG
160 
161 static int	nfe_debug = 0;
162 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
163 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
164 /*
165  * hw timer simulated interrupt moderation @4000Hz.  Negative values
166  * disable the timer when the discrete interrupt rate falls below
167  * the moderation rate.
168  *
169  * XXX 8000Hz might be better but if the interrupt is shared it can
170  *     blow out the cpu.
171  */
172 static int	nfe_imtime = -250;	/* uS */
173 
174 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
175 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
176 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
177 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
178 
179 #define DPRINTF(sc, fmt, ...) do {		\
180 	if ((sc)->sc_debug) {			\
181 		if_printf(&(sc)->arpcom.ac_if,	\
182 			  fmt, __VA_ARGS__);	\
183 	}					\
184 } while (0)
185 
186 #define DPRINTFN(sc, lv, fmt, ...) do {		\
187 	if ((sc)->sc_debug >= (lv)) {		\
188 		if_printf(&(sc)->arpcom.ac_if,	\
189 			  fmt, __VA_ARGS__);	\
190 	}					\
191 } while (0)
192 
193 #else	/* !NFE_DEBUG */
194 
195 #define DPRINTF(sc, fmt, ...)
196 #define DPRINTFN(sc, lv, fmt, ...)
197 
198 #endif	/* NFE_DEBUG */
199 
200 static const struct nfe_dev {
201 	uint16_t	vid;
202 	uint16_t	did;
203 	const char	*desc;
204 } nfe_devices[] = {
205 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
206 	  "NVIDIA nForce Fast Ethernet" },
207 
208 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
209 	  "NVIDIA nForce2 Fast Ethernet" },
210 
211 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
212 	  "NVIDIA nForce3 Gigabit Ethernet" },
213 
214 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
215 	   chipset, and possibly also the 400R; it might be both nForce2- and
216 	   nForce3-based boards can use the same MCPs (= southbridges) */
217 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
218 	  "NVIDIA nForce3 Gigabit Ethernet" },
219 
220 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
221 	  "NVIDIA nForce3 Gigabit Ethernet" },
222 
223 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
224 	  "NVIDIA nForce3 Gigabit Ethernet" },
225 
226 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
227 	  "NVIDIA nForce3 Gigabit Ethernet" },
228 
229 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
230 	  "NVIDIA CK804 Gigabit Ethernet" },
231 
232 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
233 	  "NVIDIA CK804 Gigabit Ethernet" },
234 
235 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
236 	  "NVIDIA MCP04 Gigabit Ethernet" },
237 
238 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
239 	  "NVIDIA MCP04 Gigabit Ethernet" },
240 
241 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
242 	  "NVIDIA MCP51 Gigabit Ethernet" },
243 
244 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
245 	  "NVIDIA MCP51 Gigabit Ethernet" },
246 
247 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
248 	  "NVIDIA MCP55 Gigabit Ethernet" },
249 
250 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
251 	  "NVIDIA MCP55 Gigabit Ethernet" },
252 
253 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
254 	  "NVIDIA MCP61 Gigabit Ethernet" },
255 
256 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
257 	  "NVIDIA MCP61 Gigabit Ethernet" },
258 
259 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
260 	  "NVIDIA MCP61 Gigabit Ethernet" },
261 
262 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
263 	  "NVIDIA MCP61 Gigabit Ethernet" },
264 
265 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
266 	  "NVIDIA MCP65 Gigabit Ethernet" },
267 
268 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
269 	  "NVIDIA MCP65 Gigabit Ethernet" },
270 
271 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
272 	  "NVIDIA MCP65 Gigabit Ethernet" },
273 
274 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
275 	  "NVIDIA MCP65 Gigabit Ethernet" },
276 
277 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
278 	  "NVIDIA MCP67 Gigabit Ethernet" },
279 
280 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
281 	  "NVIDIA MCP67 Gigabit Ethernet" },
282 
283 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
284 	  "NVIDIA MCP67 Gigabit Ethernet" },
285 
286 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
287 	  "NVIDIA MCP67 Gigabit Ethernet" },
288 
289 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
290 	  "NVIDIA MCP73 Gigabit Ethernet" },
291 
292 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
293 	  "NVIDIA MCP73 Gigabit Ethernet" },
294 
295 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
296 	  "NVIDIA MCP73 Gigabit Ethernet" },
297 
298 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
299 	  "NVIDIA MCP73 Gigabit Ethernet" },
300 
301 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
302 	  "NVIDIA MCP77 Gigabit Ethernet" },
303 
304 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
305 	  "NVIDIA MCP77 Gigabit Ethernet" },
306 
307 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
308 	  "NVIDIA MCP77 Gigabit Ethernet" },
309 
310 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
311 	  "NVIDIA MCP77 Gigabit Ethernet" },
312 
313 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
314 	  "NVIDIA MCP79 Gigabit Ethernet" },
315 
316 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
317 	  "NVIDIA MCP79 Gigabit Ethernet" },
318 
319 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
320 	  "NVIDIA MCP79 Gigabit Ethernet" },
321 
322 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
323 	  "NVIDIA MCP79 Gigabit Ethernet" },
324 
325 	{ 0, 0, NULL }
326 };
327 
328 static device_method_t nfe_methods[] = {
329 	/* Device interface */
330 	DEVMETHOD(device_probe,		nfe_probe),
331 	DEVMETHOD(device_attach,	nfe_attach),
332 	DEVMETHOD(device_detach,	nfe_detach),
333 	DEVMETHOD(device_suspend,	nfe_suspend),
334 	DEVMETHOD(device_resume,	nfe_resume),
335 	DEVMETHOD(device_shutdown,	nfe_shutdown),
336 
337 	/* Bus interface */
338 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
339 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
340 
341 	/* MII interface */
342 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
343 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
344 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
345 
346 	DEVMETHOD_END
347 };
348 
349 static driver_t nfe_driver = {
350 	"nfe",
351 	nfe_methods,
352 	sizeof(struct nfe_softc)
353 };
354 
355 static devclass_t	nfe_devclass;
356 
357 DECLARE_DUMMY_MODULE(if_nfe);
358 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
359 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, NULL, NULL);
360 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, NULL, NULL);
361 
362 /*
363  * NOTE: NFE_WORDALIGN support is guesswork right now.
364  */
365 static int
366 nfe_probe(device_t dev)
367 {
368 	const struct nfe_dev *n;
369 	uint16_t vid, did;
370 
371 	vid = pci_get_vendor(dev);
372 	did = pci_get_device(dev);
373 	for (n = nfe_devices; n->desc != NULL; ++n) {
374 		if (vid == n->vid && did == n->did) {
375 			struct nfe_softc *sc = device_get_softc(dev);
376 
377 			switch (did) {
378 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
379 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
380 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
381 				sc->sc_caps = NFE_NO_PWRCTL |
382 					      NFE_FIX_EADDR;
383 				break;
384 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
385 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
386 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
387 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
388 				sc->sc_caps = NFE_JUMBO_SUP |
389 					      NFE_HW_CSUM |
390 					      NFE_NO_PWRCTL |
391 					      NFE_FIX_EADDR;
392 				break;
393 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
394 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
395 				sc->sc_caps = NFE_FIX_EADDR;
396 				/* FALL THROUGH */
397 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
398 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
399 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
400 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
401 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
402 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
403 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
404 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
405 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
406 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
407 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
408 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
409 				sc->sc_caps |= NFE_40BIT_ADDR;
410 				break;
411 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
412 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
413 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
414 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
415 				sc->sc_caps = NFE_JUMBO_SUP |
416 					      NFE_40BIT_ADDR |
417 					      NFE_HW_CSUM |
418 					      NFE_NO_PWRCTL |
419 					      NFE_FIX_EADDR;
420 				break;
421 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
422 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
423 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
424 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
425 				sc->sc_caps = NFE_JUMBO_SUP |
426 					      NFE_40BIT_ADDR;
427 				break;
428 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
429 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
430 				sc->sc_caps = NFE_JUMBO_SUP |
431 					      NFE_40BIT_ADDR |
432 					      NFE_HW_CSUM |
433 					      NFE_HW_VLAN |
434 					      NFE_FIX_EADDR;
435 				break;
436 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
437 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
438 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
439 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
440 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
441 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
442 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
443 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
444 				sc->sc_caps = NFE_40BIT_ADDR |
445 					      NFE_HW_CSUM |
446 					      NFE_WORDALIGN;
447 				break;
448 			}
449 
450 			device_set_desc(dev, n->desc);
451 			device_set_async_attach(dev, TRUE);
452 			return 0;
453 		}
454 	}
455 	return ENXIO;
456 }
457 
458 static int
459 nfe_attach(device_t dev)
460 {
461 	struct nfe_softc *sc = device_get_softc(dev);
462 	struct ifnet *ifp = &sc->arpcom.ac_if;
463 	struct sysctl_ctx_list *ctx;
464 	struct sysctl_oid *tree;
465 	uint8_t eaddr[ETHER_ADDR_LEN];
466 	bus_addr_t lowaddr;
467 	int error;
468 
469 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
470 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
471 
472 	/*
473 	 * Initialize sysctl variables
474 	 */
475 	sc->sc_rx_ring_count = nfe_rx_ring_count;
476 	sc->sc_tx_ring_count = nfe_tx_ring_count;
477 	sc->sc_debug = nfe_debug;
478 	if (nfe_imtime < 0) {
479 		sc->sc_flags |= NFE_F_DYN_IM;
480 		sc->sc_imtime = -nfe_imtime;
481 	} else {
482 		sc->sc_imtime = nfe_imtime;
483 	}
484 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
485 
486 	sc->sc_mem_rid = PCIR_BAR(0);
487 
488 	if (sc->sc_caps & NFE_40BIT_ADDR)
489 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
490 	else if (sc->sc_caps & NFE_JUMBO_SUP)
491 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
492 
493 #ifndef BURN_BRIDGES
494 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
495 		uint32_t mem, irq;
496 
497 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
498 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
499 
500 		device_printf(dev, "chip is in D%d power mode "
501 		    "-- setting to D0\n", pci_get_powerstate(dev));
502 
503 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
504 
505 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
506 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
507 	}
508 #endif	/* !BURN_BRIDGE */
509 
510 	/* Enable bus mastering */
511 	pci_enable_busmaster(dev);
512 
513 	/* Allocate IO memory */
514 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
515 						&sc->sc_mem_rid, RF_ACTIVE);
516 	if (sc->sc_mem_res == NULL) {
517 		device_printf(dev, "could not allocate io memory\n");
518 		return ENXIO;
519 	}
520 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
521 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
522 
523 	/* Allocate IRQ */
524 	sc->sc_irq_rid = 0;
525 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
526 						&sc->sc_irq_rid,
527 						RF_SHAREABLE | RF_ACTIVE);
528 	if (sc->sc_irq_res == NULL) {
529 		device_printf(dev, "could not allocate irq\n");
530 		error = ENXIO;
531 		goto fail;
532 	}
533 
534 	/* Disable WOL */
535 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
536 
537 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
538 		nfe_powerup(dev);
539 
540 	nfe_get_macaddr(sc, eaddr);
541 
542 	/*
543 	 * Allocate top level DMA tag
544 	 */
545 	if (sc->sc_caps & NFE_40BIT_ADDR)
546 		lowaddr = NFE_BUS_SPACE_MAXADDR;
547 	else
548 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
549 	error = bus_dma_tag_create(NULL,	/* parent */
550 			1, 0,			/* alignment, boundary */
551 			lowaddr,		/* lowaddr */
552 			BUS_SPACE_MAXADDR,	/* highaddr */
553 			NULL, NULL,		/* filter, filterarg */
554 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
555 			0,			/* nsegments */
556 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
557 			0,			/* flags */
558 			&sc->sc_dtag);
559 	if (error) {
560 		device_printf(dev, "could not allocate parent dma tag\n");
561 		goto fail;
562 	}
563 
564 	/*
565 	 * Allocate Tx and Rx rings.
566 	 */
567 	error = nfe_alloc_tx_ring(sc, &sc->txq);
568 	if (error) {
569 		device_printf(dev, "could not allocate Tx ring\n");
570 		goto fail;
571 	}
572 
573 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
574 	if (error) {
575 		device_printf(dev, "could not allocate Rx ring\n");
576 		goto fail;
577 	}
578 
579 	/*
580 	 * Create sysctl tree
581 	 */
582 	ctx = device_get_sysctl_ctx(dev);
583 	tree = device_get_sysctl_tree(dev);
584 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
585 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
586 			sc, 0, nfe_sysctl_imtime, "I",
587 			"Interrupt moderation time (usec).  "
588 			"0 to disable interrupt moderation.");
589 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
590 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
591 		       0, "RX ring count");
592 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
593 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
594 		       0, "TX ring count");
595 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
596 		       "debug", CTLFLAG_RW, &sc->sc_debug,
597 		       0, "control debugging printfs");
598 
599 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
600 			      nfe_ifmedia_sts);
601 	if (error) {
602 		device_printf(dev, "MII without any phy\n");
603 		goto fail;
604 	}
605 
606 	ifp->if_softc = sc;
607 	ifp->if_mtu = ETHERMTU;
608 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
609 	ifp->if_ioctl = nfe_ioctl;
610 	ifp->if_start = nfe_start;
611 #ifdef IFPOLL_ENABLE
612 	ifp->if_npoll = nfe_npoll;
613 #endif
614 	ifp->if_watchdog = nfe_watchdog;
615 	ifp->if_init = nfe_init;
616 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
617 	ifq_set_ready(&ifp->if_snd);
618 
619 	ifp->if_capabilities = IFCAP_VLAN_MTU;
620 
621 	if (sc->sc_caps & NFE_HW_VLAN)
622 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
623 
624 #ifdef NFE_CSUM
625 	if (sc->sc_caps & NFE_HW_CSUM) {
626 		ifp->if_capabilities |= IFCAP_HWCSUM;
627 		ifp->if_hwassist = NFE_CSUM_FEATURES;
628 	}
629 #else
630 	sc->sc_caps &= ~NFE_HW_CSUM;
631 #endif
632 	ifp->if_capenable = ifp->if_capabilities;
633 
634 	callout_init(&sc->sc_tick_ch);
635 
636 	ether_ifattach(ifp, eaddr, NULL);
637 
638 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
639 
640 #ifdef IFPOLL_ENABLE
641 	ifpoll_compat_setup(&sc->sc_npoll, ctx, (struct sysctl_oid *)tree,
642 	    device_get_unit(dev), ifp->if_serializer);
643 #endif
644 
645 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
646 			       &sc->sc_ih, ifp->if_serializer);
647 	if (error) {
648 		device_printf(dev, "could not setup intr\n");
649 		ether_ifdetach(ifp);
650 		goto fail;
651 	}
652 
653 	return 0;
654 fail:
655 	nfe_detach(dev);
656 	return error;
657 }
658 
659 static int
660 nfe_detach(device_t dev)
661 {
662 	struct nfe_softc *sc = device_get_softc(dev);
663 
664 	if (device_is_attached(dev)) {
665 		struct ifnet *ifp = &sc->arpcom.ac_if;
666 
667 		lwkt_serialize_enter(ifp->if_serializer);
668 		nfe_stop(sc);
669 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
670 		lwkt_serialize_exit(ifp->if_serializer);
671 
672 		ether_ifdetach(ifp);
673 	}
674 
675 	if (sc->sc_miibus != NULL)
676 		device_delete_child(dev, sc->sc_miibus);
677 	bus_generic_detach(dev);
678 
679 	if (sc->sc_irq_res != NULL) {
680 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
681 				     sc->sc_irq_res);
682 	}
683 
684 	if (sc->sc_mem_res != NULL) {
685 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
686 				     sc->sc_mem_res);
687 	}
688 
689 	nfe_free_tx_ring(sc, &sc->txq);
690 	nfe_free_rx_ring(sc, &sc->rxq);
691 	if (sc->sc_dtag != NULL)
692 		bus_dma_tag_destroy(sc->sc_dtag);
693 
694 	return 0;
695 }
696 
697 static void
698 nfe_shutdown(device_t dev)
699 {
700 	struct nfe_softc *sc = device_get_softc(dev);
701 	struct ifnet *ifp = &sc->arpcom.ac_if;
702 
703 	lwkt_serialize_enter(ifp->if_serializer);
704 	nfe_stop(sc);
705 	lwkt_serialize_exit(ifp->if_serializer);
706 }
707 
708 static int
709 nfe_suspend(device_t dev)
710 {
711 	struct nfe_softc *sc = device_get_softc(dev);
712 	struct ifnet *ifp = &sc->arpcom.ac_if;
713 
714 	lwkt_serialize_enter(ifp->if_serializer);
715 	nfe_stop(sc);
716 	lwkt_serialize_exit(ifp->if_serializer);
717 
718 	return 0;
719 }
720 
721 static int
722 nfe_resume(device_t dev)
723 {
724 	struct nfe_softc *sc = device_get_softc(dev);
725 	struct ifnet *ifp = &sc->arpcom.ac_if;
726 
727 	lwkt_serialize_enter(ifp->if_serializer);
728 	if (ifp->if_flags & IFF_UP)
729 		nfe_init(sc);
730 	lwkt_serialize_exit(ifp->if_serializer);
731 
732 	return 0;
733 }
734 
735 static void
736 nfe_miibus_statchg(device_t dev)
737 {
738 	struct nfe_softc *sc = device_get_softc(dev);
739 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
740 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
741 
742 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
743 
744 	phy = NFE_READ(sc, NFE_PHY_IFACE);
745 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
746 
747 	seed = NFE_READ(sc, NFE_RNDSEED);
748 	seed &= ~NFE_SEED_MASK;
749 
750 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
751 		phy  |= NFE_PHY_HDX;	/* half-duplex */
752 		misc |= NFE_MISC1_HDX;
753 	}
754 
755 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
756 	case IFM_1000_T:	/* full-duplex only */
757 		link |= NFE_MEDIA_1000T;
758 		seed |= NFE_SEED_1000T;
759 		phy  |= NFE_PHY_1000T;
760 		break;
761 	case IFM_100_TX:
762 		link |= NFE_MEDIA_100TX;
763 		seed |= NFE_SEED_100TX;
764 		phy  |= NFE_PHY_100TX;
765 		break;
766 	case IFM_10_T:
767 		link |= NFE_MEDIA_10T;
768 		seed |= NFE_SEED_10T;
769 		break;
770 	}
771 
772 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
773 
774 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
775 	NFE_WRITE(sc, NFE_MISC1, misc);
776 	NFE_WRITE(sc, NFE_LINKSPEED, link);
777 }
778 
779 static int
780 nfe_miibus_readreg(device_t dev, int phy, int reg)
781 {
782 	struct nfe_softc *sc = device_get_softc(dev);
783 	uint32_t val;
784 	int ntries;
785 
786 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
787 
788 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
789 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
790 		DELAY(100);
791 	}
792 
793 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
794 
795 	for (ntries = 0; ntries < 1000; ntries++) {
796 		DELAY(100);
797 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
798 			break;
799 	}
800 	if (ntries == 1000) {
801 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
802 		return 0;
803 	}
804 
805 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
806 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
807 		return 0;
808 	}
809 
810 	val = NFE_READ(sc, NFE_PHY_DATA);
811 	if (val != 0xffffffff && val != 0)
812 		sc->mii_phyaddr = phy;
813 
814 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
815 
816 	return val;
817 }
818 
819 static void
820 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
821 {
822 	struct nfe_softc *sc = device_get_softc(dev);
823 	uint32_t ctl;
824 	int ntries;
825 
826 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
827 
828 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
829 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
830 		DELAY(100);
831 	}
832 
833 	NFE_WRITE(sc, NFE_PHY_DATA, val);
834 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
835 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
836 
837 	for (ntries = 0; ntries < 1000; ntries++) {
838 		DELAY(100);
839 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
840 			break;
841 	}
842 
843 #ifdef NFE_DEBUG
844 	if (ntries == 1000)
845 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
846 #endif
847 }
848 
849 #ifdef IFPOLL_ENABLE
850 
851 static void
852 nfe_npoll_compat(struct ifnet *ifp, void *arg __unused, int count __unused)
853 {
854 	struct nfe_softc *sc = ifp->if_softc;
855 
856 	ASSERT_SERIALIZED(ifp->if_serializer);
857 
858 	nfe_rxeof(sc);
859 	nfe_txeof(sc, 1);
860 }
861 
862 static void
863 nfe_disable_intrs(struct nfe_softc *sc)
864 {
865 	/* Disable interrupts */
866 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
867 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
868 	sc->sc_npoll.ifpc_stcount = 0;
869 }
870 
871 static void
872 nfe_npoll(struct ifnet *ifp, struct ifpoll_info *info)
873 {
874 	struct nfe_softc *sc = ifp->if_softc;
875 
876 	ASSERT_SERIALIZED(ifp->if_serializer);
877 
878 	if (info != NULL) {
879 		int cpuid = sc->sc_npoll.ifpc_cpuid;
880 
881 		info->ifpi_rx[cpuid].poll_func = nfe_npoll_compat;
882 		info->ifpi_rx[cpuid].arg = NULL;
883 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
884 
885 		if (ifp->if_flags & IFF_RUNNING)
886 			nfe_disable_intrs(sc);
887 		ifq_set_cpuid(&ifp->if_snd, cpuid);
888 	} else {
889 		if (ifp->if_flags & IFF_RUNNING)
890 			nfe_enable_intrs(sc);
891 		ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
892 	}
893 }
894 
895 #endif	/* IFPOLL_ENABLE */
896 
897 static void
898 nfe_intr(void *arg)
899 {
900 	struct nfe_softc *sc = arg;
901 	struct ifnet *ifp = &sc->arpcom.ac_if;
902 	uint32_t r;
903 
904 	r = NFE_READ(sc, NFE_IRQ_STATUS);
905 	if (r == 0)
906 		return;	/* not for us */
907 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
908 
909 	if (sc->sc_rate_second != time_uptime) {
910 		/*
911 		 * Calculate sc_rate_avg - interrupts per second.
912 		 */
913 		sc->sc_rate_second = time_uptime;
914 		if (sc->sc_rate_avg < sc->sc_rate_acc)
915 			sc->sc_rate_avg = sc->sc_rate_acc;
916 		else
917 			sc->sc_rate_avg = (sc->sc_rate_avg * 3 +
918 					   sc->sc_rate_acc) / 4;
919 		sc->sc_rate_acc = 0;
920 	} else if (sc->sc_rate_avg < sc->sc_rate_acc) {
921 		/*
922 		 * Don't wait for a tick to roll over if we are taking
923 		 * a lot of interrupts.
924 		 */
925 		sc->sc_rate_avg = sc->sc_rate_acc;
926 	}
927 
928 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
929 
930 	if (r & NFE_IRQ_LINK) {
931 		NFE_READ(sc, NFE_PHY_STATUS);
932 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
933 		DPRINTF(sc, "link state changed %s\n", "");
934 	}
935 
936 	if (ifp->if_flags & IFF_RUNNING) {
937 		int ret;
938 		int rate;
939 
940 		/* check Rx ring */
941 		ret = nfe_rxeof(sc);
942 
943 		/* check Tx ring */
944 		ret |= nfe_txeof(sc, 1);
945 
946 		/* update the rate accumulator */
947 		if (ret)
948 			++sc->sc_rate_acc;
949 
950 		if (sc->sc_flags & NFE_F_DYN_IM) {
951 			rate = 1000000 / sc->sc_imtime;
952 			if ((sc->sc_flags & NFE_F_IRQ_TIMER) == 0 &&
953 			    sc->sc_rate_avg > rate) {
954 				/*
955 				 * Use the hardware timer to reduce the
956 				 * interrupt rate if the discrete interrupt
957 				 * rate has exceeded our threshold.
958 				 */
959 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
960 				sc->sc_flags |= NFE_F_IRQ_TIMER;
961 			} else if ((sc->sc_flags & NFE_F_IRQ_TIMER) &&
962 				   sc->sc_rate_avg <= rate) {
963 				/*
964 				 * Use discrete TX/RX interrupts if the rate
965 				 * has fallen below our threshold.
966 				 */
967 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
968 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
969 
970 				/*
971 				 * Recollect, mainly to avoid the possible race
972 				 * introduced by changing interrupt masks.
973 				 */
974 				nfe_rxeof(sc);
975 				nfe_txeof(sc, 1);
976 			}
977 		}
978 	}
979 }
980 
981 static int
982 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
983 {
984 	struct nfe_softc *sc = ifp->if_softc;
985 	struct ifreq *ifr = (struct ifreq *)data;
986 	struct mii_data *mii;
987 	int error = 0, mask, jumbo_cap;
988 
989 	ASSERT_SERIALIZED(ifp->if_serializer);
990 
991 	switch (cmd) {
992 	case SIOCSIFMTU:
993 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
994 			jumbo_cap = 1;
995 		else
996 			jumbo_cap = 0;
997 
998 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
999 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
1000 			return EINVAL;
1001 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
1002 			ifp->if_mtu = ifr->ifr_mtu;
1003 			if (ifp->if_flags & IFF_RUNNING)
1004 				nfe_init(sc);
1005 		}
1006 		break;
1007 	case SIOCSIFFLAGS:
1008 		if (ifp->if_flags & IFF_UP) {
1009 			/*
1010 			 * If only the PROMISC or ALLMULTI flag changes, then
1011 			 * don't do a full re-init of the chip, just update
1012 			 * the Rx filter.
1013 			 */
1014 			if ((ifp->if_flags & IFF_RUNNING) &&
1015 			    ((ifp->if_flags ^ sc->sc_if_flags) &
1016 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1017 				nfe_setmulti(sc);
1018 			} else {
1019 				if (!(ifp->if_flags & IFF_RUNNING))
1020 					nfe_init(sc);
1021 			}
1022 		} else {
1023 			if (ifp->if_flags & IFF_RUNNING)
1024 				nfe_stop(sc);
1025 		}
1026 		sc->sc_if_flags = ifp->if_flags;
1027 		break;
1028 	case SIOCADDMULTI:
1029 	case SIOCDELMULTI:
1030 		if (ifp->if_flags & IFF_RUNNING)
1031 			nfe_setmulti(sc);
1032 		break;
1033 	case SIOCSIFMEDIA:
1034 	case SIOCGIFMEDIA:
1035 		mii = device_get_softc(sc->sc_miibus);
1036 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1037 		break;
1038         case SIOCSIFCAP:
1039 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
1040 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
1041 			ifp->if_capenable ^= mask;
1042 			if (IFCAP_TXCSUM & ifp->if_capenable)
1043 				ifp->if_hwassist = NFE_CSUM_FEATURES;
1044 			else
1045 				ifp->if_hwassist = 0;
1046 
1047 			if (ifp->if_flags & IFF_RUNNING)
1048 				nfe_init(sc);
1049 		}
1050 		break;
1051 	default:
1052 		error = ether_ioctl(ifp, cmd, data);
1053 		break;
1054 	}
1055 	return error;
1056 }
1057 
1058 static int
1059 nfe_rxeof(struct nfe_softc *sc)
1060 {
1061 	struct ifnet *ifp = &sc->arpcom.ac_if;
1062 	struct nfe_rx_ring *ring = &sc->rxq;
1063 	int reap;
1064 
1065 	reap = 0;
1066 	for (;;) {
1067 		struct nfe_rx_data *data = &ring->data[ring->cur];
1068 		struct mbuf *m;
1069 		uint16_t flags;
1070 		int len, error;
1071 
1072 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1073 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1074 
1075 			flags = le16toh(desc64->flags);
1076 			len = le16toh(desc64->length) & 0x3fff;
1077 		} else {
1078 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1079 
1080 			flags = le16toh(desc32->flags);
1081 			len = le16toh(desc32->length) & 0x3fff;
1082 		}
1083 
1084 		if (flags & NFE_RX_READY)
1085 			break;
1086 
1087 		reap = 1;
1088 
1089 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1090 			if (!(flags & NFE_RX_VALID_V1))
1091 				goto skip;
1092 
1093 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1094 				flags &= ~NFE_RX_ERROR;
1095 				len--;	/* fix buffer length */
1096 			}
1097 		} else {
1098 			if (!(flags & NFE_RX_VALID_V2))
1099 				goto skip;
1100 
1101 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1102 				flags &= ~NFE_RX_ERROR;
1103 				len--;	/* fix buffer length */
1104 			}
1105 		}
1106 
1107 		if (flags & NFE_RX_ERROR) {
1108 			IFNET_STAT_INC(ifp, ierrors, 1);
1109 			goto skip;
1110 		}
1111 
1112 		m = data->m;
1113 
1114 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1115 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1116 		else
1117 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1118 		if (error) {
1119 			IFNET_STAT_INC(ifp, ierrors, 1);
1120 			goto skip;
1121 		}
1122 
1123 		/* finalize mbuf */
1124 		m->m_pkthdr.len = m->m_len = len;
1125 		m->m_pkthdr.rcvif = ifp;
1126 
1127 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1128 		    (flags & NFE_RX_CSUMOK)) {
1129 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1130 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1131 							  CSUM_IP_VALID;
1132 			}
1133 
1134 			if (flags &
1135 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1136 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1137 							  CSUM_PSEUDO_HDR |
1138 							  CSUM_FRAG_NOT_CHECKED;
1139 				m->m_pkthdr.csum_data = 0xffff;
1140 			}
1141 		}
1142 
1143 		IFNET_STAT_INC(ifp, ipackets, 1);
1144 		ifp->if_input(ifp, m, NULL, -1);
1145 skip:
1146 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1147 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1148 	}
1149 	return reap;
1150 }
1151 
1152 static int
1153 nfe_txeof(struct nfe_softc *sc, int start)
1154 {
1155 	struct ifnet *ifp = &sc->arpcom.ac_if;
1156 	struct nfe_tx_ring *ring = &sc->txq;
1157 	struct nfe_tx_data *data = NULL;
1158 
1159 	while (ring->next != ring->cur) {
1160 		uint16_t flags;
1161 
1162 		if (sc->sc_caps & NFE_40BIT_ADDR)
1163 			flags = le16toh(ring->desc64[ring->next].flags);
1164 		else
1165 			flags = le16toh(ring->desc32[ring->next].flags);
1166 
1167 		if (flags & NFE_TX_VALID)
1168 			break;
1169 
1170 		data = &ring->data[ring->next];
1171 
1172 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1173 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1174 				goto skip;
1175 
1176 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1177 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
1178 					  NFE_V1_TXERR);
1179 				IFNET_STAT_INC(ifp, oerrors, 1);
1180 			} else {
1181 				IFNET_STAT_INC(ifp, opackets, 1);
1182 			}
1183 		} else {
1184 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1185 				goto skip;
1186 
1187 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1188 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
1189 					  NFE_V2_TXERR);
1190 				IFNET_STAT_INC(ifp, oerrors, 1);
1191 			} else {
1192 				IFNET_STAT_INC(ifp, opackets, 1);
1193 			}
1194 		}
1195 
1196 		if (data->m == NULL) {	/* should not get there */
1197 			if_printf(ifp,
1198 				  "last fragment bit w/o associated mbuf!\n");
1199 			goto skip;
1200 		}
1201 
1202 		/* last fragment of the mbuf chain transmitted */
1203 		bus_dmamap_unload(ring->data_tag, data->map);
1204 		m_freem(data->m);
1205 		data->m = NULL;
1206 skip:
1207 		ring->queued--;
1208 		KKASSERT(ring->queued >= 0);
1209 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1210 	}
1211 
1212 	if (sc->sc_tx_ring_count - ring->queued >=
1213 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1214 		ifq_clr_oactive(&ifp->if_snd);
1215 
1216 	if (ring->queued == 0)
1217 		ifp->if_timer = 0;
1218 
1219 	if (start && !ifq_is_empty(&ifp->if_snd))
1220 		if_devstart(ifp);
1221 
1222 	if (data != NULL)
1223 		return 1;
1224 	else
1225 		return 0;
1226 }
1227 
1228 static int
1229 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1230 {
1231 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1232 	struct nfe_tx_data *data, *data_map;
1233 	bus_dmamap_t map;
1234 	struct nfe_desc64 *desc64 = NULL;
1235 	struct nfe_desc32 *desc32 = NULL;
1236 	uint16_t flags = 0;
1237 	uint32_t vtag = 0;
1238 	int error, i, j, maxsegs, nsegs;
1239 
1240 	data = &ring->data[ring->cur];
1241 	map = data->map;
1242 	data_map = data;	/* Remember who owns the DMA map */
1243 
1244 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1245 	if (maxsegs > NFE_MAX_SCATTER)
1246 		maxsegs = NFE_MAX_SCATTER;
1247 	KASSERT(maxsegs >= sc->sc_tx_spare,
1248 		("not enough segments %d,%d", maxsegs, sc->sc_tx_spare));
1249 
1250 	error = bus_dmamap_load_mbuf_defrag(ring->data_tag, map, &m0,
1251 			segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1252 	if (error)
1253 		goto back;
1254 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1255 
1256 	error = 0;
1257 
1258 	/* setup h/w VLAN tagging */
1259 	if (m0->m_flags & M_VLANTAG)
1260 		vtag = m0->m_pkthdr.ether_vlantag;
1261 
1262 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1263 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1264 			flags |= NFE_TX_IP_CSUM;
1265 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1266 			flags |= NFE_TX_TCP_CSUM;
1267 	}
1268 
1269 	/*
1270 	 * XXX urm. somebody is unaware of how hardware works.  You
1271 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1272 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1273 	 * the hardware may encounter a partially initialized chain that
1274 	 * is marked as being ready to go when it in fact is not ready to
1275 	 * go.
1276 	 */
1277 
1278 	for (i = 0; i < nsegs; i++) {
1279 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1280 		data = &ring->data[j];
1281 
1282 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1283 			desc64 = &ring->desc64[j];
1284 			desc64->physaddr[0] =
1285 			    htole32(NFE_ADDR_HI(segs[i].ds_addr));
1286 			desc64->physaddr[1] =
1287 			    htole32(NFE_ADDR_LO(segs[i].ds_addr));
1288 			desc64->length = htole16(segs[i].ds_len - 1);
1289 			desc64->vtag = htole32(vtag);
1290 			desc64->flags = htole16(flags);
1291 		} else {
1292 			desc32 = &ring->desc32[j];
1293 			desc32->physaddr = htole32(segs[i].ds_addr);
1294 			desc32->length = htole16(segs[i].ds_len - 1);
1295 			desc32->flags = htole16(flags);
1296 		}
1297 
1298 		/* csum flags and vtag belong to the first fragment only */
1299 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1300 		vtag = 0;
1301 
1302 		ring->queued++;
1303 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1304 	}
1305 
1306 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1307 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1308 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1309 	} else {
1310 		if (sc->sc_caps & NFE_JUMBO_SUP)
1311 			flags = NFE_TX_LASTFRAG_V2;
1312 		else
1313 			flags = NFE_TX_LASTFRAG_V1;
1314 		desc32->flags |= htole16(flags);
1315 	}
1316 
1317 	/*
1318 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1319 	 * whole mess until the first descriptor in the map is flagged.
1320 	 */
1321 	for (i = nsegs - 1; i >= 0; --i) {
1322 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1323 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1324 			desc64 = &ring->desc64[j];
1325 			desc64->flags |= htole16(NFE_TX_VALID);
1326 		} else {
1327 			desc32 = &ring->desc32[j];
1328 			desc32->flags |= htole16(NFE_TX_VALID);
1329 		}
1330 	}
1331 	ring->cur = (ring->cur + nsegs) % sc->sc_tx_ring_count;
1332 
1333 	/* Exchange DMA map */
1334 	data_map->map = data->map;
1335 	data->map = map;
1336 	data->m = m0;
1337 back:
1338 	if (error)
1339 		m_freem(m0);
1340 	return error;
1341 }
1342 
1343 static void
1344 nfe_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1345 {
1346 	struct nfe_softc *sc = ifp->if_softc;
1347 	struct nfe_tx_ring *ring = &sc->txq;
1348 	int count = 0, oactive = 0;
1349 	struct mbuf *m0;
1350 
1351 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1352 	ASSERT_SERIALIZED(ifp->if_serializer);
1353 
1354 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1355 		return;
1356 
1357 	for (;;) {
1358 		int error;
1359 
1360 		if (sc->sc_tx_ring_count - ring->queued <
1361 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1362 			if (oactive) {
1363 				ifq_set_oactive(&ifp->if_snd);
1364 				break;
1365 			}
1366 
1367 			nfe_txeof(sc, 0);
1368 			oactive = 1;
1369 			continue;
1370 		}
1371 
1372 		m0 = ifq_dequeue(&ifp->if_snd);
1373 		if (m0 == NULL)
1374 			break;
1375 
1376 		ETHER_BPF_MTAP(ifp, m0);
1377 
1378 		error = nfe_encap(sc, ring, m0);
1379 		if (error) {
1380 			IFNET_STAT_INC(ifp, oerrors, 1);
1381 			if (error == EFBIG) {
1382 				if (oactive) {
1383 					ifq_set_oactive(&ifp->if_snd);
1384 					break;
1385 				}
1386 				nfe_txeof(sc, 0);
1387 				oactive = 1;
1388 			}
1389 			continue;
1390 		} else {
1391 			oactive = 0;
1392 		}
1393 		++count;
1394 
1395 		/*
1396 		 * NOTE:
1397 		 * `m0' may be freed in nfe_encap(), so
1398 		 * it should not be touched any more.
1399 		 */
1400 	}
1401 
1402 	if (count == 0)	/* nothing sent */
1403 		return;
1404 
1405 	/* Kick Tx */
1406 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1407 
1408 	/*
1409 	 * Set a timeout in case the chip goes out to lunch.
1410 	 */
1411 	ifp->if_timer = 5;
1412 }
1413 
1414 static void
1415 nfe_watchdog(struct ifnet *ifp)
1416 {
1417 	struct nfe_softc *sc = ifp->if_softc;
1418 
1419 	ASSERT_SERIALIZED(ifp->if_serializer);
1420 
1421 	if (ifp->if_flags & IFF_RUNNING) {
1422 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1423 		nfe_txeof(sc, 1);
1424 		return;
1425 	}
1426 
1427 	if_printf(ifp, "watchdog timeout\n");
1428 
1429 	nfe_init(ifp->if_softc);
1430 
1431 	IFNET_STAT_INC(ifp, oerrors, 1);
1432 }
1433 
1434 static void
1435 nfe_init(void *xsc)
1436 {
1437 	struct nfe_softc *sc = xsc;
1438 	struct ifnet *ifp = &sc->arpcom.ac_if;
1439 	uint32_t tmp;
1440 	int error;
1441 
1442 	ASSERT_SERIALIZED(ifp->if_serializer);
1443 
1444 	nfe_stop(sc);
1445 
1446 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1447 		nfe_mac_reset(sc);
1448 
1449 	/*
1450 	 * NOTE:
1451 	 * Switching between jumbo frames and normal frames should
1452 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1453 	 */
1454 	if (ifp->if_mtu > ETHERMTU) {
1455 		sc->sc_flags |= NFE_F_USE_JUMBO;
1456 		sc->rxq.bufsz = NFE_JBYTES;
1457 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1458 		if (bootverbose)
1459 			if_printf(ifp, "use jumbo frames\n");
1460 	} else {
1461 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1462 		sc->rxq.bufsz = MCLBYTES;
1463 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1464 		if (bootverbose)
1465 			if_printf(ifp, "use non-jumbo frames\n");
1466 	}
1467 
1468 	error = nfe_init_tx_ring(sc, &sc->txq);
1469 	if (error) {
1470 		nfe_stop(sc);
1471 		return;
1472 	}
1473 
1474 	error = nfe_init_rx_ring(sc, &sc->rxq);
1475 	if (error) {
1476 		nfe_stop(sc);
1477 		return;
1478 	}
1479 
1480 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1481 	NFE_WRITE(sc, NFE_STATUS, 0);
1482 
1483 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1484 
1485 	if (ifp->if_capenable & IFCAP_RXCSUM)
1486 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1487 
1488 	/*
1489 	 * Although the adapter is capable of stripping VLAN tags from received
1490 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1491 	 * purpose.  This will be done in software by our network stack.
1492 	 */
1493 	if (sc->sc_caps & NFE_HW_VLAN)
1494 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1495 
1496 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1497 	DELAY(10);
1498 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1499 
1500 	if (sc->sc_caps & NFE_HW_VLAN)
1501 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1502 
1503 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1504 
1505 	/* set MAC address */
1506 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1507 
1508 	/* tell MAC where rings are in memory */
1509 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1510 		NFE_WRITE(sc, NFE_RX_RING_ADDR_HI,
1511 			  NFE_ADDR_HI(sc->rxq.physaddr));
1512 	}
1513 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, NFE_ADDR_LO(sc->rxq.physaddr));
1514 
1515 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1516 		NFE_WRITE(sc, NFE_TX_RING_ADDR_HI,
1517 			  NFE_ADDR_HI(sc->txq.physaddr));
1518 	}
1519 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, NFE_ADDR_LO(sc->txq.physaddr));
1520 
1521 	NFE_WRITE(sc, NFE_RING_SIZE,
1522 	    (sc->sc_rx_ring_count - 1) << 16 |
1523 	    (sc->sc_tx_ring_count - 1));
1524 
1525 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1526 
1527 	/* force MAC to wakeup */
1528 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1529 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1530 	DELAY(10);
1531 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1532 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1533 
1534 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1535 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1536 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1537 
1538 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1539 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1540 
1541 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1542 
1543 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1544 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1545 	DELAY(10);
1546 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1547 
1548 	/* set Rx filter */
1549 	nfe_setmulti(sc);
1550 
1551 	nfe_ifmedia_upd(ifp);
1552 
1553 	/* enable Rx */
1554 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1555 
1556 	/* enable Tx */
1557 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1558 
1559 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1560 
1561 #ifdef IFPOLL_ENABLE
1562 	if (ifp->if_flags & IFF_NPOLLING)
1563 		nfe_disable_intrs(sc);
1564 	else
1565 #endif
1566 	nfe_enable_intrs(sc);
1567 
1568 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1569 
1570 	ifp->if_flags |= IFF_RUNNING;
1571 	ifq_clr_oactive(&ifp->if_snd);
1572 
1573 	/*
1574 	 * If we had stuff in the tx ring before its all cleaned out now
1575 	 * so we are not going to get an interrupt, jump-start any pending
1576 	 * output.
1577 	 */
1578 	if (!ifq_is_empty(&ifp->if_snd))
1579 		if_devstart(ifp);
1580 }
1581 
1582 static void
1583 nfe_stop(struct nfe_softc *sc)
1584 {
1585 	struct ifnet *ifp = &sc->arpcom.ac_if;
1586 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1587 	int i;
1588 
1589 	ASSERT_SERIALIZED(ifp->if_serializer);
1590 
1591 	callout_stop(&sc->sc_tick_ch);
1592 
1593 	ifp->if_timer = 0;
1594 	ifp->if_flags &= ~IFF_RUNNING;
1595 	ifq_clr_oactive(&ifp->if_snd);
1596 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1597 
1598 #define WAITMAX	50000
1599 
1600 	/*
1601 	 * Abort Tx
1602 	 */
1603 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1604 	for (i = 0; i < WAITMAX; ++i) {
1605 		DELAY(100);
1606 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1607 			break;
1608 	}
1609 	if (i == WAITMAX)
1610 		if_printf(ifp, "can't stop TX\n");
1611 	DELAY(100);
1612 
1613 	/*
1614 	 * Disable Rx
1615 	 */
1616 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1617 	for (i = 0; i < WAITMAX; ++i) {
1618 		DELAY(100);
1619 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1620 			break;
1621 	}
1622 	if (i == WAITMAX)
1623 		if_printf(ifp, "can't stop RX\n");
1624 	DELAY(100);
1625 
1626 #undef WAITMAX
1627 
1628 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1629 	DELAY(10);
1630 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1631 
1632 	/* Disable interrupts */
1633 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1634 
1635 	/* Reset Tx and Rx rings */
1636 	nfe_reset_tx_ring(sc, &sc->txq);
1637 	nfe_reset_rx_ring(sc, &sc->rxq);
1638 }
1639 
1640 static int
1641 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1642 {
1643 	int i, j, error, descsize;
1644 	bus_dmamem_t dmem;
1645 	void **desc;
1646 
1647 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1648 		desc = (void *)&ring->desc64;
1649 		descsize = sizeof(struct nfe_desc64);
1650 	} else {
1651 		desc = (void *)&ring->desc32;
1652 		descsize = sizeof(struct nfe_desc32);
1653 	}
1654 
1655 	ring->bufsz = MCLBYTES;
1656 	ring->cur = ring->next = 0;
1657 
1658 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1659 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1660 				    sc->sc_rx_ring_count * descsize,
1661 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1662 	if (error) {
1663 		if_printf(&sc->arpcom.ac_if,
1664 			  "could not create RX desc ring\n");
1665 		return error;
1666 	}
1667 	ring->tag = dmem.dmem_tag;
1668 	ring->map = dmem.dmem_map;
1669 	*desc = dmem.dmem_addr;
1670 	ring->physaddr = dmem.dmem_busaddr;
1671 
1672 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1673 		ring->jbuf =
1674 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1675 			M_DEVBUF, M_WAITOK | M_ZERO);
1676 
1677 		error = nfe_jpool_alloc(sc, ring);
1678 		if (error) {
1679 			if_printf(&sc->arpcom.ac_if,
1680 				  "could not allocate jumbo frames\n");
1681 			kfree(ring->jbuf, M_DEVBUF);
1682 			ring->jbuf = NULL;
1683 			/* Allow jumbo frame allocation to fail */
1684 		}
1685 	}
1686 
1687 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1688 			     M_DEVBUF, M_WAITOK | M_ZERO);
1689 
1690 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1691 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1692 				   NULL, NULL,
1693 				   MCLBYTES, 1, MCLBYTES,
1694 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
1695 				   &ring->data_tag);
1696 	if (error) {
1697 		if_printf(&sc->arpcom.ac_if,
1698 			  "could not create RX mbuf DMA tag\n");
1699 		return error;
1700 	}
1701 
1702 	/* Create a spare RX mbuf DMA map */
1703 	error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1704 				  &ring->data_tmpmap);
1705 	if (error) {
1706 		if_printf(&sc->arpcom.ac_if,
1707 			  "could not create spare RX mbuf DMA map\n");
1708 		bus_dma_tag_destroy(ring->data_tag);
1709 		ring->data_tag = NULL;
1710 		return error;
1711 	}
1712 
1713 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1714 		error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1715 					  &ring->data[i].map);
1716 		if (error) {
1717 			if_printf(&sc->arpcom.ac_if,
1718 				  "could not create %dth RX mbuf DMA mapn", i);
1719 			goto fail;
1720 		}
1721 	}
1722 	return 0;
1723 fail:
1724 	for (j = 0; j < i; ++j)
1725 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1726 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1727 	bus_dma_tag_destroy(ring->data_tag);
1728 	ring->data_tag = NULL;
1729 	return error;
1730 }
1731 
1732 static void
1733 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1734 {
1735 	int i;
1736 
1737 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1738 		struct nfe_rx_data *data = &ring->data[i];
1739 
1740 		if (data->m != NULL) {
1741 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1742 				bus_dmamap_unload(ring->data_tag, data->map);
1743 			m_freem(data->m);
1744 			data->m = NULL;
1745 		}
1746 	}
1747 
1748 	ring->cur = ring->next = 0;
1749 }
1750 
1751 static int
1752 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1753 {
1754 	int i;
1755 
1756 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1757 		int error;
1758 
1759 		/* XXX should use a function pointer */
1760 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1761 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1762 		else
1763 			error = nfe_newbuf_std(sc, ring, i, 1);
1764 		if (error) {
1765 			if_printf(&sc->arpcom.ac_if,
1766 				  "could not allocate RX buffer\n");
1767 			return error;
1768 		}
1769 		nfe_set_ready_rxdesc(sc, ring, i);
1770 	}
1771 	return 0;
1772 }
1773 
1774 static void
1775 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1776 {
1777 	if (ring->data_tag != NULL) {
1778 		struct nfe_rx_data *data;
1779 		int i;
1780 
1781 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1782 			data = &ring->data[i];
1783 
1784 			if (data->m != NULL) {
1785 				bus_dmamap_unload(ring->data_tag, data->map);
1786 				m_freem(data->m);
1787 			}
1788 			bus_dmamap_destroy(ring->data_tag, data->map);
1789 		}
1790 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1791 		bus_dma_tag_destroy(ring->data_tag);
1792 	}
1793 
1794 	nfe_jpool_free(sc, ring);
1795 
1796 	if (ring->jbuf != NULL)
1797 		kfree(ring->jbuf, M_DEVBUF);
1798 	if (ring->data != NULL)
1799 		kfree(ring->data, M_DEVBUF);
1800 
1801 	if (ring->tag != NULL) {
1802 		void *desc;
1803 
1804 		if (sc->sc_caps & NFE_40BIT_ADDR)
1805 			desc = ring->desc64;
1806 		else
1807 			desc = ring->desc32;
1808 
1809 		bus_dmamap_unload(ring->tag, ring->map);
1810 		bus_dmamem_free(ring->tag, desc, ring->map);
1811 		bus_dma_tag_destroy(ring->tag);
1812 	}
1813 }
1814 
1815 static struct nfe_jbuf *
1816 nfe_jalloc(struct nfe_softc *sc)
1817 {
1818 	struct ifnet *ifp = &sc->arpcom.ac_if;
1819 	struct nfe_jbuf *jbuf;
1820 
1821 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1822 
1823 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1824 	if (jbuf != NULL) {
1825 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1826 		jbuf->inuse = 1;
1827 	} else {
1828 		if_printf(ifp, "no free jumbo buffer\n");
1829 	}
1830 
1831 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1832 
1833 	return jbuf;
1834 }
1835 
1836 static void
1837 nfe_jfree(void *arg)
1838 {
1839 	struct nfe_jbuf *jbuf = arg;
1840 	struct nfe_softc *sc = jbuf->sc;
1841 	struct nfe_rx_ring *ring = jbuf->ring;
1842 
1843 	if (&ring->jbuf[jbuf->slot] != jbuf)
1844 		panic("%s: free wrong jumbo buffer", __func__);
1845 	else if (jbuf->inuse == 0)
1846 		panic("%s: jumbo buffer already freed", __func__);
1847 
1848 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1849 	atomic_subtract_int(&jbuf->inuse, 1);
1850 	if (jbuf->inuse == 0)
1851 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1852 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1853 }
1854 
1855 static void
1856 nfe_jref(void *arg)
1857 {
1858 	struct nfe_jbuf *jbuf = arg;
1859 	struct nfe_rx_ring *ring = jbuf->ring;
1860 
1861 	if (&ring->jbuf[jbuf->slot] != jbuf)
1862 		panic("%s: ref wrong jumbo buffer", __func__);
1863 	else if (jbuf->inuse == 0)
1864 		panic("%s: jumbo buffer already freed", __func__);
1865 
1866 	atomic_add_int(&jbuf->inuse, 1);
1867 }
1868 
1869 static int
1870 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1871 {
1872 	struct nfe_jbuf *jbuf;
1873 	bus_dmamem_t dmem;
1874 	bus_addr_t physaddr;
1875 	caddr_t buf;
1876 	int i, error;
1877 
1878 	/*
1879 	 * Allocate a big chunk of DMA'able memory.
1880 	 */
1881 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1882 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1883 				    NFE_JPOOL_SIZE(sc),
1884 				    BUS_DMA_WAITOK, &dmem);
1885 	if (error) {
1886 		if_printf(&sc->arpcom.ac_if,
1887 			  "could not create jumbo buffer\n");
1888 		return error;
1889 	}
1890 	ring->jtag = dmem.dmem_tag;
1891 	ring->jmap = dmem.dmem_map;
1892 	ring->jpool = dmem.dmem_addr;
1893 	physaddr = dmem.dmem_busaddr;
1894 
1895 	/* ..and split it into 9KB chunks */
1896 	SLIST_INIT(&ring->jfreelist);
1897 
1898 	buf = ring->jpool;
1899 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1900 		jbuf = &ring->jbuf[i];
1901 
1902 		jbuf->sc = sc;
1903 		jbuf->ring = ring;
1904 		jbuf->inuse = 0;
1905 		jbuf->slot = i;
1906 		jbuf->buf = buf;
1907 		jbuf->physaddr = physaddr;
1908 
1909 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1910 
1911 		buf += NFE_JBYTES;
1912 		physaddr += NFE_JBYTES;
1913 	}
1914 
1915 	return 0;
1916 }
1917 
1918 static void
1919 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1920 {
1921 	if (ring->jtag != NULL) {
1922 		bus_dmamap_unload(ring->jtag, ring->jmap);
1923 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1924 		bus_dma_tag_destroy(ring->jtag);
1925 	}
1926 }
1927 
1928 static int
1929 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1930 {
1931 	int i, j, error, descsize;
1932 	bus_dmamem_t dmem;
1933 	void **desc;
1934 
1935 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1936 		desc = (void *)&ring->desc64;
1937 		descsize = sizeof(struct nfe_desc64);
1938 	} else {
1939 		desc = (void *)&ring->desc32;
1940 		descsize = sizeof(struct nfe_desc32);
1941 	}
1942 
1943 	ring->queued = 0;
1944 	ring->cur = ring->next = 0;
1945 
1946 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1947 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1948 				    sc->sc_tx_ring_count * descsize,
1949 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1950 	if (error) {
1951 		if_printf(&sc->arpcom.ac_if,
1952 			  "could not create TX desc ring\n");
1953 		return error;
1954 	}
1955 	ring->tag = dmem.dmem_tag;
1956 	ring->map = dmem.dmem_map;
1957 	*desc = dmem.dmem_addr;
1958 	ring->physaddr = dmem.dmem_busaddr;
1959 
1960 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
1961 			     M_DEVBUF, M_WAITOK | M_ZERO);
1962 
1963 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1964 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1965 			NULL, NULL,
1966 			NFE_JBYTES, NFE_MAX_SCATTER, MCLBYTES,
1967 			BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1968 			&ring->data_tag);
1969 	if (error) {
1970 		if_printf(&sc->arpcom.ac_if,
1971 			  "could not create TX buf DMA tag\n");
1972 		return error;
1973 	}
1974 
1975 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1976 		error = bus_dmamap_create(ring->data_tag,
1977 				BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1978 				&ring->data[i].map);
1979 		if (error) {
1980 			if_printf(&sc->arpcom.ac_if,
1981 				  "could not create %dth TX buf DMA map\n", i);
1982 			goto fail;
1983 		}
1984 	}
1985 
1986 	return 0;
1987 fail:
1988 	for (j = 0; j < i; ++j)
1989 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1990 	bus_dma_tag_destroy(ring->data_tag);
1991 	ring->data_tag = NULL;
1992 	return error;
1993 }
1994 
1995 static void
1996 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1997 {
1998 	int i;
1999 
2000 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
2001 		struct nfe_tx_data *data = &ring->data[i];
2002 
2003 		if (sc->sc_caps & NFE_40BIT_ADDR)
2004 			ring->desc64[i].flags = 0;
2005 		else
2006 			ring->desc32[i].flags = 0;
2007 
2008 		if (data->m != NULL) {
2009 			bus_dmamap_unload(ring->data_tag, data->map);
2010 			m_freem(data->m);
2011 			data->m = NULL;
2012 		}
2013 	}
2014 
2015 	ring->queued = 0;
2016 	ring->cur = ring->next = 0;
2017 }
2018 
2019 static int
2020 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2021 		 struct nfe_tx_ring *ring __unused)
2022 {
2023 	return 0;
2024 }
2025 
2026 static void
2027 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2028 {
2029 	if (ring->data_tag != NULL) {
2030 		struct nfe_tx_data *data;
2031 		int i;
2032 
2033 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2034 			data = &ring->data[i];
2035 
2036 			if (data->m != NULL) {
2037 				bus_dmamap_unload(ring->data_tag, data->map);
2038 				m_freem(data->m);
2039 			}
2040 			bus_dmamap_destroy(ring->data_tag, data->map);
2041 		}
2042 
2043 		bus_dma_tag_destroy(ring->data_tag);
2044 	}
2045 
2046 	if (ring->data != NULL)
2047 		kfree(ring->data, M_DEVBUF);
2048 
2049 	if (ring->tag != NULL) {
2050 		void *desc;
2051 
2052 		if (sc->sc_caps & NFE_40BIT_ADDR)
2053 			desc = ring->desc64;
2054 		else
2055 			desc = ring->desc32;
2056 
2057 		bus_dmamap_unload(ring->tag, ring->map);
2058 		bus_dmamem_free(ring->tag, desc, ring->map);
2059 		bus_dma_tag_destroy(ring->tag);
2060 	}
2061 }
2062 
2063 static int
2064 nfe_ifmedia_upd(struct ifnet *ifp)
2065 {
2066 	struct nfe_softc *sc = ifp->if_softc;
2067 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2068 
2069 	ASSERT_SERIALIZED(ifp->if_serializer);
2070 
2071 	if (mii->mii_instance != 0) {
2072 		struct mii_softc *miisc;
2073 
2074 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2075 			mii_phy_reset(miisc);
2076 	}
2077 	mii_mediachg(mii);
2078 
2079 	return 0;
2080 }
2081 
2082 static void
2083 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2084 {
2085 	struct nfe_softc *sc = ifp->if_softc;
2086 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2087 
2088 	ASSERT_SERIALIZED(ifp->if_serializer);
2089 
2090 	mii_pollstat(mii);
2091 	ifmr->ifm_status = mii->mii_media_status;
2092 	ifmr->ifm_active = mii->mii_media_active;
2093 }
2094 
2095 static void
2096 nfe_setmulti(struct nfe_softc *sc)
2097 {
2098 	struct ifnet *ifp = &sc->arpcom.ac_if;
2099 	struct ifmultiaddr *ifma;
2100 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2101 	uint32_t filter = NFE_RXFILTER_MAGIC;
2102 	int i;
2103 
2104 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2105 		bzero(addr, ETHER_ADDR_LEN);
2106 		bzero(mask, ETHER_ADDR_LEN);
2107 		goto done;
2108 	}
2109 
2110 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2111 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2112 
2113 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2114 		caddr_t maddr;
2115 
2116 		if (ifma->ifma_addr->sa_family != AF_LINK)
2117 			continue;
2118 
2119 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2120 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2121 			addr[i] &= maddr[i];
2122 			mask[i] &= ~maddr[i];
2123 		}
2124 	}
2125 
2126 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2127 		mask[i] |= addr[i];
2128 
2129 done:
2130 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2131 
2132 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2133 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2134 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2135 	    addr[5] <<  8 | addr[4]);
2136 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2137 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2138 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2139 	    mask[5] <<  8 | mask[4]);
2140 
2141 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2142 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2143 }
2144 
2145 static void
2146 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2147 {
2148 	uint32_t lo, hi;
2149 
2150 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2151 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2152 	if (sc->sc_caps & NFE_FIX_EADDR) {
2153 		addr[0] = (lo >> 8) & 0xff;
2154 		addr[1] = (lo & 0xff);
2155 
2156 		addr[2] = (hi >> 24) & 0xff;
2157 		addr[3] = (hi >> 16) & 0xff;
2158 		addr[4] = (hi >>  8) & 0xff;
2159 		addr[5] = (hi & 0xff);
2160 	} else {
2161 		addr[0] = (hi & 0xff);
2162 		addr[1] = (hi >>  8) & 0xff;
2163 		addr[2] = (hi >> 16) & 0xff;
2164 		addr[3] = (hi >> 24) & 0xff;
2165 
2166 		addr[4] = (lo & 0xff);
2167 		addr[5] = (lo >>  8) & 0xff;
2168 	}
2169 }
2170 
2171 static void
2172 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2173 {
2174 	NFE_WRITE(sc, NFE_MACADDR_LO,
2175 	    addr[5] <<  8 | addr[4]);
2176 	NFE_WRITE(sc, NFE_MACADDR_HI,
2177 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2178 }
2179 
2180 static void
2181 nfe_tick(void *arg)
2182 {
2183 	struct nfe_softc *sc = arg;
2184 	struct ifnet *ifp = &sc->arpcom.ac_if;
2185 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2186 
2187 	lwkt_serialize_enter(ifp->if_serializer);
2188 
2189 	mii_tick(mii);
2190 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2191 
2192 	lwkt_serialize_exit(ifp->if_serializer);
2193 }
2194 
2195 static int
2196 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2197 	       int wait)
2198 {
2199 	struct nfe_rx_data *data = &ring->data[idx];
2200 	bus_dma_segment_t seg;
2201 	bus_dmamap_t map;
2202 	struct mbuf *m;
2203 	int nsegs, error;
2204 
2205 	m = m_getcl(wait ? M_WAITOK : M_NOWAIT, MT_DATA, M_PKTHDR);
2206 	if (m == NULL)
2207 		return ENOBUFS;
2208 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2209 
2210 	/*
2211 	 * Aligning the payload improves access times.
2212 	 */
2213 	if (sc->sc_caps & NFE_WORDALIGN)
2214 		m_adj(m, ETHER_ALIGN);
2215 
2216 	error = bus_dmamap_load_mbuf_segment(ring->data_tag, ring->data_tmpmap,
2217 			m, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
2218 	if (error) {
2219 		m_freem(m);
2220 		if (wait) {
2221 			if_printf(&sc->arpcom.ac_if,
2222 				  "could map RX mbuf %d\n", error);
2223 		}
2224 		return error;
2225 	}
2226 
2227 	if (data->m != NULL) {
2228 		/* Sync and unload originally mapped mbuf */
2229 		bus_dmamap_sync(ring->data_tag, data->map,
2230 				BUS_DMASYNC_POSTREAD);
2231 		bus_dmamap_unload(ring->data_tag, data->map);
2232 	}
2233 
2234 	/* Swap this DMA map with tmp DMA map */
2235 	map = data->map;
2236 	data->map = ring->data_tmpmap;
2237 	ring->data_tmpmap = map;
2238 
2239 	/* Caller is assumed to have collected the old mbuf */
2240 	data->m = m;
2241 
2242 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2243 	return 0;
2244 }
2245 
2246 static int
2247 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2248 		 int wait)
2249 {
2250 	struct nfe_rx_data *data = &ring->data[idx];
2251 	struct nfe_jbuf *jbuf;
2252 	struct mbuf *m;
2253 
2254 	MGETHDR(m, wait ? M_WAITOK : M_NOWAIT, MT_DATA);
2255 	if (m == NULL)
2256 		return ENOBUFS;
2257 
2258 	jbuf = nfe_jalloc(sc);
2259 	if (jbuf == NULL) {
2260 		m_freem(m);
2261 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2262 		    "-- packet dropped!\n");
2263 		return ENOBUFS;
2264 	}
2265 
2266 	m->m_ext.ext_arg = jbuf;
2267 	m->m_ext.ext_buf = jbuf->buf;
2268 	m->m_ext.ext_free = nfe_jfree;
2269 	m->m_ext.ext_ref = nfe_jref;
2270 	m->m_ext.ext_size = NFE_JBYTES;
2271 
2272 	m->m_data = m->m_ext.ext_buf;
2273 	m->m_flags |= M_EXT;
2274 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2275 
2276 	/*
2277 	 * Aligning the payload improves access times.
2278 	 */
2279 	if (sc->sc_caps & NFE_WORDALIGN)
2280 		m_adj(m, ETHER_ALIGN);
2281 
2282 	/* Caller is assumed to have collected the old mbuf */
2283 	data->m = m;
2284 
2285 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2286 	return 0;
2287 }
2288 
2289 static void
2290 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2291 		     bus_addr_t physaddr)
2292 {
2293 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2294 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2295 
2296 		desc64->physaddr[0] = htole32(NFE_ADDR_HI(physaddr));
2297 		desc64->physaddr[1] = htole32(NFE_ADDR_LO(physaddr));
2298 	} else {
2299 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2300 
2301 		desc32->physaddr = htole32(physaddr);
2302 	}
2303 }
2304 
2305 static void
2306 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2307 {
2308 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2309 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2310 
2311 		desc64->length = htole16(ring->bufsz);
2312 		desc64->flags = htole16(NFE_RX_READY);
2313 	} else {
2314 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2315 
2316 		desc32->length = htole16(ring->bufsz);
2317 		desc32->flags = htole16(NFE_RX_READY);
2318 	}
2319 }
2320 
2321 static int
2322 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2323 {
2324 	struct nfe_softc *sc = arg1;
2325 	struct ifnet *ifp = &sc->arpcom.ac_if;
2326 	uint32_t flags;
2327 	int error, v;
2328 
2329 	lwkt_serialize_enter(ifp->if_serializer);
2330 
2331 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2332 	v = sc->sc_imtime;
2333 	if (sc->sc_flags & NFE_F_DYN_IM)
2334 		v = -v;
2335 
2336 	error = sysctl_handle_int(oidp, &v, 0, req);
2337 	if (error || req->newptr == NULL)
2338 		goto back;
2339 
2340 	if (v < 0) {
2341 		flags |= NFE_F_DYN_IM;
2342 		v = -v;
2343 	}
2344 
2345 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2346 		if (NFE_IMTIME(v) == 0)
2347 			v = 0;
2348 		sc->sc_imtime = v;
2349 		sc->sc_flags = flags;
2350 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2351 
2352 		if ((ifp->if_flags & (IFF_NPOLLING | IFF_RUNNING))
2353 		    == IFF_RUNNING) {
2354 			nfe_enable_intrs(sc);
2355 		}
2356 	}
2357 back:
2358 	lwkt_serialize_exit(ifp->if_serializer);
2359 	return error;
2360 }
2361 
2362 static void
2363 nfe_powerup(device_t dev)
2364 {
2365 	struct nfe_softc *sc = device_get_softc(dev);
2366 	uint32_t pwr_state;
2367 	uint16_t did;
2368 
2369 	/*
2370 	 * Bring MAC and PHY out of low power state
2371 	 */
2372 
2373 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2374 
2375 	did = pci_get_device(dev);
2376 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2377 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2378 	    pci_get_revid(dev) >= 0xa3)
2379 		pwr_state |= NFE_PWRUP_REV_A3;
2380 
2381 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2382 }
2383 
2384 static void
2385 nfe_mac_reset(struct nfe_softc *sc)
2386 {
2387 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2388 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2389 
2390 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2391 
2392 	/* Save several registers for later restoration */
2393 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2394 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2395 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2396 
2397 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2398 	DELAY(100);
2399 
2400 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2401 	DELAY(100);
2402 
2403 	/* Restore saved registers */
2404 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2405 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2406 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2407 
2408 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2409 }
2410 
2411 static void
2412 nfe_enable_intrs(struct nfe_softc *sc)
2413 {
2414 	/*
2415 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2416 	 * It is unclear how wide the timer is.  Base programming does
2417 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2418 	 * we don't get any interrupt moderation.  TX moderation is
2419 	 * possible by using the timer interrupt instead of TX_DONE.
2420 	 *
2421 	 * It is unclear whether there are other bits that can be
2422 	 * set to make the NFE device actually do interrupt moderation
2423 	 * on the RX side.
2424 	 *
2425 	 * For now set a 128uS interval as a placemark, but don't use
2426 	 * the timer.
2427 	 */
2428 	if (sc->sc_imtime == 0)
2429 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2430 	else
2431 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2432 
2433 	/* Enable interrupts */
2434 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2435 
2436 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2437 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2438 	else
2439 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2440 }
2441