xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision 3f625015)
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 1.9 2007/05/01 23:48:03 dillon 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 
60 #include <sys/param.h>
61 #include <sys/endian.h>
62 #include <sys/kernel.h>
63 #include <sys/bus.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/ifq_var.h>
78 #include <net/if_types.h>
79 #include <net/if_var.h>
80 #include <net/vlan/if_vlan_var.h>
81 
82 #include <bus/pci/pcireg.h>
83 #include <bus/pci/pcivar.h>
84 #include <bus/pci/pcidevs.h>
85 
86 #include <dev/netif/mii_layer/mii.h>
87 #include <dev/netif/mii_layer/miivar.h>
88 
89 #include "miibus_if.h"
90 
91 #include "if_nfereg.h"
92 #include "if_nfevar.h"
93 
94 static int	nfe_probe(device_t);
95 static int	nfe_attach(device_t);
96 static int	nfe_detach(device_t);
97 static void	nfe_shutdown(device_t);
98 static int	nfe_resume(device_t);
99 static int	nfe_suspend(device_t);
100 
101 static int	nfe_miibus_readreg(device_t, int, int);
102 static void	nfe_miibus_writereg(device_t, int, int, int);
103 static void	nfe_miibus_statchg(device_t);
104 
105 #ifdef DEVICE_POLLING
106 static void	nfe_poll(struct ifnet *, enum poll_cmd, int);
107 #endif
108 static void	nfe_intr(void *);
109 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
110 static void	nfe_rxeof(struct nfe_softc *);
111 static void	nfe_txeof(struct nfe_softc *);
112 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
113 			  struct mbuf *);
114 static void	nfe_start(struct ifnet *);
115 static void	nfe_watchdog(struct ifnet *);
116 static void	nfe_init(void *);
117 static void	nfe_stop(struct nfe_softc *);
118 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
119 static void	nfe_jfree(void *);
120 static void	nfe_jref(void *);
121 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
122 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
123 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
124 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
125 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
126 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
127 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
128 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
129 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
130 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
131 static int	nfe_ifmedia_upd(struct ifnet *);
132 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
133 static void	nfe_setmulti(struct nfe_softc *);
134 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
135 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
136 static void	nfe_tick(void *);
137 static void	nfe_ring_dma_addr(void *, bus_dma_segment_t *, int, int);
138 static void	nfe_buf_dma_addr(void *, bus_dma_segment_t *, int, bus_size_t,
139 				 int);
140 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
141 				     int, bus_addr_t);
142 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
143 				     int);
144 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
145 			       int);
146 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
147 				 int);
148 
149 #define NFE_DEBUG
150 #ifdef NFE_DEBUG
151 
152 static int	nfe_debug = 0;
153 
154 SYSCTL_NODE(_hw, OID_AUTO, nfe, CTLFLAG_RD, 0, "nVidia GigE parameters");
155 SYSCTL_INT(_hw_nfe, OID_AUTO, debug, CTLFLAG_RW, &nfe_debug, 0,
156 	   "control debugging printfs");
157 
158 #define DPRINTF(sc, fmt, ...) do {		\
159 	if (nfe_debug) {			\
160 		if_printf(&(sc)->arpcom.ac_if,	\
161 			  fmt, __VA_ARGS__);	\
162 	}					\
163 } while (0)
164 
165 #define DPRINTFN(sc, lv, fmt, ...) do {		\
166 	if (nfe_debug >= (lv)) {		\
167 		if_printf(&(sc)->arpcom.ac_if,	\
168 			  fmt, __VA_ARGS__);	\
169 	}					\
170 } while (0)
171 
172 #else	/* !NFE_DEBUG */
173 
174 #define DPRINTF(sc, fmt, ...)
175 #define DPRINTFN(sc, lv, fmt, ...)
176 
177 #endif	/* NFE_DEBUG */
178 
179 struct nfe_dma_ctx {
180 	int			nsegs;
181 	bus_dma_segment_t	*segs;
182 };
183 
184 static const struct nfe_dev {
185 	uint16_t	vid;
186 	uint16_t	did;
187 	const char	*desc;
188 } nfe_devices[] = {
189 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
190 	  "NVIDIA nForce Fast Ethernet" },
191 
192 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
193 	  "NVIDIA nForce2 Fast Ethernet" },
194 
195 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
196 	  "NVIDIA nForce3 Gigabit Ethernet" },
197 
198 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
199 	   chipset, and possibly also the 400R; it might be both nForce2- and
200 	   nForce3-based boards can use the same MCPs (= southbridges) */
201 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
202 	  "NVIDIA nForce3 Gigabit Ethernet" },
203 
204 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
205 	  "NVIDIA nForce3 Gigabit Ethernet" },
206 
207 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
208 	  "NVIDIA nForce3 Gigabit Ethernet" },
209 
210 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
211 	  "NVIDIA nForce3 Gigabit Ethernet" },
212 
213 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
214 	  "NVIDIA CK804 Gigabit Ethernet" },
215 
216 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
217 	  "NVIDIA CK804 Gigabit Ethernet" },
218 
219 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
220 	  "NVIDIA MCP04 Gigabit Ethernet" },
221 
222 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
223 	  "NVIDIA MCP04 Gigabit Ethernet" },
224 
225 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
226 	  "NVIDIA MCP51 Gigabit Ethernet" },
227 
228 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
229 	  "NVIDIA MCP51 Gigabit Ethernet" },
230 
231 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
232 	  "NVIDIA MCP55 Gigabit Ethernet" },
233 
234 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
235 	  "NVIDIA MCP55 Gigabit Ethernet" },
236 
237 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
238 	  "NVIDIA MCP61 Gigabit Ethernet" },
239 
240 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
241 	  "NVIDIA MCP61 Gigabit Ethernet" },
242 
243 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
244 	  "NVIDIA MCP61 Gigabit Ethernet" },
245 
246 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
247 	  "NVIDIA MCP61 Gigabit Ethernet" },
248 
249 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
250 	  "NVIDIA MCP65 Gigabit Ethernet" },
251 
252 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
253 	  "NVIDIA MCP65 Gigabit Ethernet" },
254 
255 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
256 	  "NVIDIA MCP65 Gigabit Ethernet" },
257 
258 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
259 	  "NVIDIA MCP65 Gigabit Ethernet" },
260 
261 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
262 	  "NVIDIA MCP67 Gigabit Ethernet" },
263 
264 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
265 	  "NVIDIA MCP67 Gigabit Ethernet" },
266 
267 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
268 	  "NVIDIA MCP67 Gigabit Ethernet" },
269 
270 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
271 	  "NVIDIA MCP67 Gigabit Ethernet" }
272 };
273 
274 static device_method_t nfe_methods[] = {
275 	/* Device interface */
276 	DEVMETHOD(device_probe,		nfe_probe),
277 	DEVMETHOD(device_attach,	nfe_attach),
278 	DEVMETHOD(device_detach,	nfe_detach),
279 	DEVMETHOD(device_suspend,	nfe_suspend),
280 	DEVMETHOD(device_resume,	nfe_resume),
281 	DEVMETHOD(device_shutdown,	nfe_shutdown),
282 
283 	/* Bus interface */
284 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
285 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
286 
287 	/* MII interface */
288 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
289 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
290 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
291 
292 	{ 0, 0 }
293 };
294 
295 static driver_t nfe_driver = {
296 	"nfe",
297 	nfe_methods,
298 	sizeof(struct nfe_softc)
299 };
300 
301 static devclass_t	nfe_devclass;
302 
303 DECLARE_DUMMY_MODULE(if_nfe);
304 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
305 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, 0, 0);
306 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, 0, 0);
307 
308 static int
309 nfe_probe(device_t dev)
310 {
311 	const struct nfe_dev *n;
312 	uint16_t vid, did;
313 
314 	vid = pci_get_vendor(dev);
315 	did = pci_get_device(dev);
316 	for (n = nfe_devices; n->desc != NULL; ++n) {
317 		if (vid == n->vid && did == n->did) {
318 			struct nfe_softc *sc = device_get_softc(dev);
319 
320 			switch (did) {
321 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
322 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
323 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
324 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
325 				sc->sc_flags = NFE_JUMBO_SUP |
326 					       NFE_HW_CSUM;
327 				break;
328 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
329 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
330 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
331 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
332 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
333 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
334 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
335 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
336 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
337 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
338 				sc->sc_flags = NFE_40BIT_ADDR;
339 				break;
340 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
341 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
342 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
343 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
344 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
345 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
346 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
347 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
348 				sc->sc_flags = NFE_JUMBO_SUP |
349 					       NFE_40BIT_ADDR |
350 					       NFE_HW_CSUM;
351 				break;
352 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
353 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
354 				sc->sc_flags = NFE_JUMBO_SUP |
355 					       NFE_40BIT_ADDR |
356 					       NFE_HW_CSUM |
357 					       NFE_HW_VLAN;
358 				break;
359 			}
360 
361 			/* Enable jumbo frames for adapters that support it */
362 			if (sc->sc_flags & NFE_JUMBO_SUP)
363 				sc->sc_flags |= NFE_USE_JUMBO;
364 
365 			device_set_desc(dev, n->desc);
366 			device_set_async_attach(dev, TRUE);
367 			return 0;
368 		}
369 	}
370 	return ENXIO;
371 }
372 
373 static int
374 nfe_attach(device_t dev)
375 {
376 	struct nfe_softc *sc = device_get_softc(dev);
377 	struct ifnet *ifp = &sc->arpcom.ac_if;
378 	uint8_t eaddr[ETHER_ADDR_LEN];
379 	int error;
380 
381 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
382 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
383 
384 	sc->sc_mem_rid = PCIR_BAR(0);
385 
386 #ifndef BURN_BRIDGES
387 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
388 		uint32_t mem, irq;
389 
390 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
391 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
392 
393 		device_printf(dev, "chip is in D%d power mode "
394 		    "-- setting to D0\n", pci_get_powerstate(dev));
395 
396 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
397 
398 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
399 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
400 	}
401 #endif	/* !BURN_BRIDGE */
402 
403 	/* Enable bus mastering */
404 	pci_enable_busmaster(dev);
405 
406 	/* Allocate IO memory */
407 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
408 						&sc->sc_mem_rid, RF_ACTIVE);
409 	if (sc->sc_mem_res == NULL) {
410 		device_printf(dev, "cound not allocate io memory\n");
411 		return ENXIO;
412 	}
413 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
414 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
415 
416 	/* Allocate IRQ */
417 	sc->sc_irq_rid = 0;
418 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
419 						&sc->sc_irq_rid,
420 						RF_SHAREABLE | RF_ACTIVE);
421 	if (sc->sc_irq_res == NULL) {
422 		device_printf(dev, "could not allocate irq\n");
423 		error = ENXIO;
424 		goto fail;
425 	}
426 
427 	nfe_get_macaddr(sc, eaddr);
428 
429 	/*
430 	 * Allocate Tx and Rx rings.
431 	 */
432 	error = nfe_alloc_tx_ring(sc, &sc->txq);
433 	if (error) {
434 		device_printf(dev, "could not allocate Tx ring\n");
435 		goto fail;
436 	}
437 
438 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
439 	if (error) {
440 		device_printf(dev, "could not allocate Rx ring\n");
441 		goto fail;
442 	}
443 
444 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
445 			      nfe_ifmedia_sts);
446 	if (error) {
447 		device_printf(dev, "MII without any phy\n");
448 		goto fail;
449 	}
450 
451 	ifp->if_softc = sc;
452 	ifp->if_mtu = ETHERMTU;
453 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
454 	ifp->if_ioctl = nfe_ioctl;
455 	ifp->if_start = nfe_start;
456 #ifdef DEVICE_POLLING
457 	ifp->if_poll = nfe_poll;
458 #endif
459 	ifp->if_watchdog = nfe_watchdog;
460 	ifp->if_init = nfe_init;
461 	ifq_set_maxlen(&ifp->if_snd, NFE_IFQ_MAXLEN);
462 	ifq_set_ready(&ifp->if_snd);
463 
464 	ifp->if_capabilities = IFCAP_VLAN_MTU;
465 
466 #if 0
467 	if (sc->sc_flags & NFE_USE_JUMBO)
468 		ifp->if_hardmtu = NFE_JUMBO_MTU;
469 #endif
470 
471 	if (sc->sc_flags & NFE_HW_VLAN)
472 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
473 
474 #ifdef NFE_CSUM
475 	if (sc->sc_flags & NFE_HW_CSUM) {
476 #if 0
477 		ifp->if_capabilities |= IFCAP_CSUM_IPv4 | IFCAP_CSUM_TCPv4 |
478 		    IFCAP_CSUM_UDPv4;
479 #else
480 		ifp->if_capabilities = IFCAP_HWCSUM;
481 		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
482 #endif
483 	}
484 #endif
485 	ifp->if_capenable = ifp->if_capabilities;
486 
487 	callout_init(&sc->sc_tick_ch);
488 
489 	ether_ifattach(ifp, eaddr, NULL);
490 
491 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
492 			       &sc->sc_ih, ifp->if_serializer);
493 	if (error) {
494 		device_printf(dev, "could not setup intr\n");
495 		ether_ifdetach(ifp);
496 		goto fail;
497 	}
498 
499 	return 0;
500 fail:
501 	nfe_detach(dev);
502 	return error;
503 }
504 
505 static int
506 nfe_detach(device_t dev)
507 {
508 	struct nfe_softc *sc = device_get_softc(dev);
509 
510 	if (device_is_attached(dev)) {
511 		struct ifnet *ifp = &sc->arpcom.ac_if;
512 
513 		lwkt_serialize_enter(ifp->if_serializer);
514 		nfe_stop(sc);
515 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
516 		lwkt_serialize_exit(ifp->if_serializer);
517 
518 		ether_ifdetach(ifp);
519 	}
520 
521 	if (sc->sc_miibus != NULL)
522 		device_delete_child(dev, sc->sc_miibus);
523 	bus_generic_detach(dev);
524 
525 	if (sc->sc_irq_res != NULL) {
526 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
527 				     sc->sc_irq_res);
528 	}
529 
530 	if (sc->sc_mem_res != NULL) {
531 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
532 				     sc->sc_mem_res);
533 	}
534 
535 	nfe_free_tx_ring(sc, &sc->txq);
536 	nfe_free_rx_ring(sc, &sc->rxq);
537 
538 	return 0;
539 }
540 
541 static void
542 nfe_shutdown(device_t dev)
543 {
544 	struct nfe_softc *sc = device_get_softc(dev);
545 	struct ifnet *ifp = &sc->arpcom.ac_if;
546 
547 	lwkt_serialize_enter(ifp->if_serializer);
548 	nfe_stop(sc);
549 	lwkt_serialize_exit(ifp->if_serializer);
550 }
551 
552 static int
553 nfe_suspend(device_t dev)
554 {
555 	struct nfe_softc *sc = device_get_softc(dev);
556 	struct ifnet *ifp = &sc->arpcom.ac_if;
557 
558 	lwkt_serialize_enter(ifp->if_serializer);
559 	nfe_stop(sc);
560 	lwkt_serialize_exit(ifp->if_serializer);
561 
562 	return 0;
563 }
564 
565 static int
566 nfe_resume(device_t dev)
567 {
568 	struct nfe_softc *sc = device_get_softc(dev);
569 	struct ifnet *ifp = &sc->arpcom.ac_if;
570 
571 	lwkt_serialize_enter(ifp->if_serializer);
572 	if (ifp->if_flags & IFF_UP) {
573 		nfe_init(sc);
574 		if (ifp->if_flags & IFF_RUNNING)
575 			ifp->if_start(ifp);
576 	}
577 	lwkt_serialize_exit(ifp->if_serializer);
578 
579 	return 0;
580 }
581 
582 static void
583 nfe_miibus_statchg(device_t dev)
584 {
585 	struct nfe_softc *sc = device_get_softc(dev);
586 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
587 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
588 
589 	phy = NFE_READ(sc, NFE_PHY_IFACE);
590 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
591 
592 	seed = NFE_READ(sc, NFE_RNDSEED);
593 	seed &= ~NFE_SEED_MASK;
594 
595 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
596 		phy  |= NFE_PHY_HDX;	/* half-duplex */
597 		misc |= NFE_MISC1_HDX;
598 	}
599 
600 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
601 	case IFM_1000_T:	/* full-duplex only */
602 		link |= NFE_MEDIA_1000T;
603 		seed |= NFE_SEED_1000T;
604 		phy  |= NFE_PHY_1000T;
605 		break;
606 	case IFM_100_TX:
607 		link |= NFE_MEDIA_100TX;
608 		seed |= NFE_SEED_100TX;
609 		phy  |= NFE_PHY_100TX;
610 		break;
611 	case IFM_10_T:
612 		link |= NFE_MEDIA_10T;
613 		seed |= NFE_SEED_10T;
614 		break;
615 	}
616 
617 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
618 
619 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
620 	NFE_WRITE(sc, NFE_MISC1, misc);
621 	NFE_WRITE(sc, NFE_LINKSPEED, link);
622 }
623 
624 static int
625 nfe_miibus_readreg(device_t dev, int phy, int reg)
626 {
627 	struct nfe_softc *sc = device_get_softc(dev);
628 	uint32_t val;
629 	int ntries;
630 
631 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
632 
633 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
634 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
635 		DELAY(100);
636 	}
637 
638 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
639 
640 	for (ntries = 0; ntries < 1000; ntries++) {
641 		DELAY(100);
642 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
643 			break;
644 	}
645 	if (ntries == 1000) {
646 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
647 		return 0;
648 	}
649 
650 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
651 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
652 		return 0;
653 	}
654 
655 	val = NFE_READ(sc, NFE_PHY_DATA);
656 	if (val != 0xffffffff && val != 0)
657 		sc->mii_phyaddr = phy;
658 
659 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
660 
661 	return val;
662 }
663 
664 static void
665 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
666 {
667 	struct nfe_softc *sc = device_get_softc(dev);
668 	uint32_t ctl;
669 	int ntries;
670 
671 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
672 
673 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
674 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
675 		DELAY(100);
676 	}
677 
678 	NFE_WRITE(sc, NFE_PHY_DATA, val);
679 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
680 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
681 
682 	for (ntries = 0; ntries < 1000; ntries++) {
683 		DELAY(100);
684 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
685 			break;
686 	}
687 
688 #ifdef NFE_DEBUG
689 	if (ntries == 1000)
690 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
691 #endif
692 }
693 
694 #ifdef DEVICE_POLLING
695 
696 static void
697 nfe_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
698 {
699 	struct nfe_softc *sc = ifp->if_softc;
700 
701 	switch(cmd) {
702 	case POLL_REGISTER:
703 		/* Disable interrupts */
704 		NFE_WRITE(sc, NFE_IRQ_MASK, 0);
705 		break;
706 	case POLL_DEREGISTER:
707 		/* enable interrupts */
708 		NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
709 		break;
710 	case POLL_AND_CHECK_STATUS:
711 		/* fall through */
712 	case POLL_ONLY:
713 		if (ifp->if_flags & IFF_RUNNING) {
714 			nfe_rxeof(sc);
715 			nfe_txeof(sc);
716 		}
717 		break;
718 	}
719 }
720 
721 #endif
722 
723 static void
724 nfe_intr(void *arg)
725 {
726 	struct nfe_softc *sc = arg;
727 	struct ifnet *ifp = &sc->arpcom.ac_if;
728 	uint32_t r;
729 
730 	r = NFE_READ(sc, NFE_IRQ_STATUS);
731 	if (r == 0)
732 		return;	/* not for us */
733 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
734 
735 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
736 
737 	if (r & NFE_IRQ_LINK) {
738 		NFE_READ(sc, NFE_PHY_STATUS);
739 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
740 		DPRINTF(sc, "link state changed %s\n", "");
741 	}
742 
743 	if (ifp->if_flags & IFF_RUNNING) {
744 		/* check Rx ring */
745 		nfe_rxeof(sc);
746 
747 		/* check Tx ring */
748 		nfe_txeof(sc);
749 	}
750 }
751 
752 static int
753 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
754 {
755 	struct nfe_softc *sc = ifp->if_softc;
756 	struct ifreq *ifr = (struct ifreq *)data;
757 	struct mii_data *mii;
758 	int error = 0, mask;
759 
760 	switch (cmd) {
761 	case SIOCSIFMTU:
762 		/* XXX NFE_USE_JUMBO should be set here */
763 		break;
764 	case SIOCSIFFLAGS:
765 		if (ifp->if_flags & IFF_UP) {
766 			/*
767 			 * If only the PROMISC or ALLMULTI flag changes, then
768 			 * don't do a full re-init of the chip, just update
769 			 * the Rx filter.
770 			 */
771 			if ((ifp->if_flags & IFF_RUNNING) &&
772 			    ((ifp->if_flags ^ sc->sc_if_flags) &
773 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
774 				nfe_setmulti(sc);
775 			} else {
776 				if (!(ifp->if_flags & IFF_RUNNING))
777 					nfe_init(sc);
778 			}
779 		} else {
780 			if (ifp->if_flags & IFF_RUNNING)
781 				nfe_stop(sc);
782 		}
783 		sc->sc_if_flags = ifp->if_flags;
784 		break;
785 	case SIOCADDMULTI:
786 	case SIOCDELMULTI:
787 		if (ifp->if_flags & IFF_RUNNING)
788 			nfe_setmulti(sc);
789 		break;
790 	case SIOCSIFMEDIA:
791 	case SIOCGIFMEDIA:
792 		mii = device_get_softc(sc->sc_miibus);
793 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
794 		break;
795         case SIOCSIFCAP:
796 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
797 		if (mask & IFCAP_HWCSUM) {
798 			if (IFCAP_HWCSUM & ifp->if_capenable)
799 				ifp->if_capenable &= ~IFCAP_HWCSUM;
800 			else
801 				ifp->if_capenable |= IFCAP_HWCSUM;
802 		}
803 		break;
804 	default:
805 		error = ether_ioctl(ifp, cmd, data);
806 		break;
807 	}
808 	return error;
809 }
810 
811 static void
812 nfe_rxeof(struct nfe_softc *sc)
813 {
814 	struct ifnet *ifp = &sc->arpcom.ac_if;
815 	struct nfe_rx_ring *ring = &sc->rxq;
816 	int reap;
817 
818 	reap = 0;
819 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
820 
821 	for (;;) {
822 		struct nfe_rx_data *data = &ring->data[ring->cur];
823 		struct mbuf *m;
824 		uint16_t flags;
825 		int len, error;
826 
827 		if (sc->sc_flags & NFE_40BIT_ADDR) {
828 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
829 
830 			flags = le16toh(desc64->flags);
831 			len = le16toh(desc64->length) & 0x3fff;
832 		} else {
833 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
834 
835 			flags = le16toh(desc32->flags);
836 			len = le16toh(desc32->length) & 0x3fff;
837 		}
838 
839 		if (flags & NFE_RX_READY)
840 			break;
841 
842 		reap = 1;
843 
844 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
845 			if (!(flags & NFE_RX_VALID_V1))
846 				goto skip;
847 
848 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
849 				flags &= ~NFE_RX_ERROR;
850 				len--;	/* fix buffer length */
851 			}
852 		} else {
853 			if (!(flags & NFE_RX_VALID_V2))
854 				goto skip;
855 
856 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
857 				flags &= ~NFE_RX_ERROR;
858 				len--;	/* fix buffer length */
859 			}
860 		}
861 
862 		if (flags & NFE_RX_ERROR) {
863 			ifp->if_ierrors++;
864 			goto skip;
865 		}
866 
867 		m = data->m;
868 
869 		if (sc->sc_flags & NFE_USE_JUMBO)
870 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
871 		else
872 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
873 		if (error) {
874 			ifp->if_ierrors++;
875 			goto skip;
876 		}
877 
878 		/* finalize mbuf */
879 		m->m_pkthdr.len = m->m_len = len;
880 		m->m_pkthdr.rcvif = ifp;
881 
882 #ifdef notyet
883 		if (sc->sc_flags & NFE_HW_CSUM) {
884 			if (flags & NFE_RX_IP_CSUMOK)
885 				m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
886 			if (flags & NFE_RX_UDP_CSUMOK)
887 				m->m_pkthdr.csum_flags |= M_UDP_CSUM_IN_OK;
888 			if (flags & NFE_RX_TCP_CSUMOK)
889 				m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK;
890 		}
891 #elif defined(NFE_CSUM)
892 		if ((sc->sc_flags & NFE_HW_CSUM) && (flags & NFE_RX_CSUMOK))
893 			m->m_pkthdr.csum_flags = M_IPV4_CSUM_IN_OK;
894 #endif
895 
896 		ifp->if_ipackets++;
897 		ifp->if_input(ifp, m);
898 skip:
899 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
900 		sc->rxq.cur = (sc->rxq.cur + 1) % NFE_RX_RING_COUNT;
901 	}
902 
903 	if (reap)
904 		bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
905 }
906 
907 static void
908 nfe_txeof(struct nfe_softc *sc)
909 {
910 	struct ifnet *ifp = &sc->arpcom.ac_if;
911 	struct nfe_tx_ring *ring = &sc->txq;
912 	struct nfe_tx_data *data = NULL;
913 
914 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
915 	while (ring->next != ring->cur) {
916 		uint16_t flags;
917 
918 		if (sc->sc_flags & NFE_40BIT_ADDR)
919 			flags = le16toh(ring->desc64[ring->next].flags);
920 		else
921 			flags = le16toh(ring->desc32[ring->next].flags);
922 
923 		if (flags & NFE_TX_VALID)
924 			break;
925 
926 		data = &ring->data[ring->next];
927 
928 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
929 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
930 				goto skip;
931 
932 			if ((flags & NFE_TX_ERROR_V1) != 0) {
933 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
934 					  NFE_V1_TXERR);
935 				ifp->if_oerrors++;
936 			} else {
937 				ifp->if_opackets++;
938 			}
939 		} else {
940 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
941 				goto skip;
942 
943 			if ((flags & NFE_TX_ERROR_V2) != 0) {
944 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
945 					  NFE_V2_TXERR);
946 				ifp->if_oerrors++;
947 			} else {
948 				ifp->if_opackets++;
949 			}
950 		}
951 
952 		if (data->m == NULL) {	/* should not get there */
953 			if_printf(ifp,
954 				  "last fragment bit w/o associated mbuf!\n");
955 			goto skip;
956 		}
957 
958 		/* last fragment of the mbuf chain transmitted */
959 		bus_dmamap_sync(ring->data_tag, data->map,
960 				BUS_DMASYNC_POSTWRITE);
961 		bus_dmamap_unload(ring->data_tag, data->map);
962 		m_freem(data->m);
963 		data->m = NULL;
964 
965 		ifp->if_timer = 0;
966 skip:
967 		ring->queued--;
968 		KKASSERT(ring->queued >= 0);
969 		ring->next = (ring->next + 1) % NFE_TX_RING_COUNT;
970 	}
971 
972 	if (data != NULL) {	/* at least one slot freed */
973 		ifp->if_flags &= ~IFF_OACTIVE;
974 		ifp->if_start(ifp);
975 	}
976 }
977 
978 static int
979 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
980 {
981 	struct nfe_dma_ctx ctx;
982 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
983 	struct nfe_tx_data *data, *data_map;
984 	bus_dmamap_t map;
985 	struct nfe_desc64 *desc64 = NULL;
986 	struct nfe_desc32 *desc32 = NULL;
987 	uint16_t flags = 0;
988 	uint32_t vtag = 0;
989 	int error, i, j;
990 
991 	data = &ring->data[ring->cur];
992 	map = data->map;
993 	data_map = data;	/* Remember who owns the DMA map */
994 
995 	ctx.nsegs = NFE_MAX_SCATTER;
996 	ctx.segs = segs;
997 	error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
998 				     nfe_buf_dma_addr, &ctx, BUS_DMA_NOWAIT);
999 	if (error && error != EFBIG) {
1000 		if_printf(&sc->arpcom.ac_if, "could not map TX mbuf\n");
1001 		goto back;
1002 	}
1003 
1004 	if (error) {	/* error == EFBIG */
1005 		struct mbuf *m_new;
1006 
1007 		m_new = m_defrag(m0, MB_DONTWAIT);
1008 		if (m_new == NULL) {
1009 			if_printf(&sc->arpcom.ac_if,
1010 				  "could not defrag TX mbuf\n");
1011 			error = ENOBUFS;
1012 			goto back;
1013 		} else {
1014 			m0 = m_new;
1015 		}
1016 
1017 		ctx.nsegs = NFE_MAX_SCATTER;
1018 		ctx.segs = segs;
1019 		error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
1020 					     nfe_buf_dma_addr, &ctx,
1021 					     BUS_DMA_NOWAIT);
1022 		if (error) {
1023 			if_printf(&sc->arpcom.ac_if,
1024 				  "could not map defraged TX mbuf\n");
1025 			goto back;
1026 		}
1027 	}
1028 
1029 	error = 0;
1030 
1031 	if (ring->queued + ctx.nsegs >= NFE_TX_RING_COUNT - 1) {
1032 		bus_dmamap_unload(ring->data_tag, map);
1033 		error = ENOBUFS;
1034 		goto back;
1035 	}
1036 
1037 	/* setup h/w VLAN tagging */
1038 	if ((m0->m_flags & (M_PROTO1 | M_PKTHDR)) == (M_PROTO1 | M_PKTHDR) &&
1039 	    m0->m_pkthdr.rcvif != NULL &&
1040 	    m0->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1041 		struct ifvlan *ifv = m0->m_pkthdr.rcvif->if_softc;
1042 
1043 		if (ifv != NULL)
1044 			vtag = NFE_TX_VTAG | htons(ifv->ifv_tag);
1045 	}
1046 
1047 #ifdef NFE_CSUM
1048 	if (m0->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
1049 		flags |= NFE_TX_IP_CSUM;
1050 	if (m0->m_pkthdr.csum_flags & (M_TCPV4_CSUM_OUT | M_UDPV4_CSUM_OUT))
1051 		flags |= NFE_TX_TCP_CSUM;
1052 #endif
1053 
1054 	/*
1055 	 * XXX urm. somebody is unaware of how hardware works.  You
1056 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1057 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1058 	 * the hardware may encounter a partially initialized chain that
1059 	 * is marked as being ready to go when it in fact is not ready to
1060 	 * go.
1061 	 */
1062 
1063 	for (i = 0; i < ctx.nsegs; i++) {
1064 		j = (ring->cur + i) % NFE_TX_RING_COUNT;
1065 		data = &ring->data[j];
1066 
1067 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1068 			desc64 = &ring->desc64[j];
1069 #if defined(__LP64__)
1070 			desc64->physaddr[0] =
1071 			    htole32(segs[i].ds_addr >> 32);
1072 #endif
1073 			desc64->physaddr[1] =
1074 			    htole32(segs[i].ds_addr & 0xffffffff);
1075 			desc64->length = htole16(segs[i].ds_len - 1);
1076 			desc64->vtag = htole32(vtag);
1077 			desc64->flags = htole16(flags);
1078 		} else {
1079 			desc32 = &ring->desc32[j];
1080 			desc32->physaddr = htole32(segs[i].ds_addr);
1081 			desc32->length = htole16(segs[i].ds_len - 1);
1082 			desc32->flags = htole16(flags);
1083 		}
1084 
1085 		/* csum flags and vtag belong to the first fragment only */
1086 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1087 		vtag = 0;
1088 
1089 		ring->queued++;
1090 		KKASSERT(ring->queued <= NFE_TX_RING_COUNT);
1091 	}
1092 
1093 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1094 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1095 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1096 	} else {
1097 		if (sc->sc_flags & NFE_JUMBO_SUP)
1098 			flags = NFE_TX_LASTFRAG_V2;
1099 		else
1100 			flags = NFE_TX_LASTFRAG_V1;
1101 		desc32->flags |= htole16(flags);
1102 	}
1103 
1104 	/*
1105 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1106 	 * whole mess until the first descriptor in the map is flagged.
1107 	 */
1108 	for (i = ctx.nsegs - 1; i >= 0; --i) {
1109 		j = (ring->cur + i) % NFE_TX_RING_COUNT;
1110 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1111 			desc64 = &ring->desc64[j];
1112 			desc64->flags |= htole16(NFE_TX_VALID);
1113 		} else {
1114 			desc32 = &ring->desc32[j];
1115 			desc32->flags |= htole16(NFE_TX_VALID);
1116 		}
1117 	}
1118 	ring->cur = (ring->cur + ctx.nsegs) % NFE_TX_RING_COUNT;
1119 
1120 	/* Exchange DMA map */
1121 	data_map->map = data->map;
1122 	data->map = map;
1123 	data->m = m0;
1124 
1125 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1126 back:
1127 	if (error)
1128 		m_freem(m0);
1129 	return error;
1130 }
1131 
1132 static void
1133 nfe_start(struct ifnet *ifp)
1134 {
1135 	struct nfe_softc *sc = ifp->if_softc;
1136 	struct nfe_tx_ring *ring = &sc->txq;
1137 	int count = 0;
1138 	struct mbuf *m0;
1139 
1140 	if (ifp->if_flags & IFF_OACTIVE)
1141 		return;
1142 
1143 	if (ifq_is_empty(&ifp->if_snd))
1144 		return;
1145 
1146 	for (;;) {
1147 		m0 = ifq_dequeue(&ifp->if_snd, NULL);
1148 		if (m0 == NULL)
1149 			break;
1150 
1151 		BPF_MTAP(ifp, m0);
1152 
1153 		if (nfe_encap(sc, ring, m0) != 0) {
1154 			ifp->if_flags |= IFF_OACTIVE;
1155 			break;
1156 		}
1157 		++count;
1158 
1159 		/*
1160 		 * NOTE:
1161 		 * `m0' may be freed in nfe_encap(), so
1162 		 * it should not be touched any more.
1163 		 */
1164 	}
1165 	if (count == 0)	/* nothing sent */
1166 		return;
1167 
1168 	/* Sync TX descriptor ring */
1169 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1170 
1171 	/* Kick Tx */
1172 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1173 
1174 	/*
1175 	 * Set a timeout in case the chip goes out to lunch.
1176 	 */
1177 	ifp->if_timer = 5;
1178 }
1179 
1180 static void
1181 nfe_watchdog(struct ifnet *ifp)
1182 {
1183 	struct nfe_softc *sc = ifp->if_softc;
1184 
1185 	if (ifp->if_flags & IFF_RUNNING) {
1186 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1187 		nfe_txeof(sc);
1188 		return;
1189 	}
1190 
1191 	if_printf(ifp, "watchdog timeout\n");
1192 
1193 	nfe_init(ifp->if_softc);
1194 
1195 	ifp->if_oerrors++;
1196 
1197 	if (!ifq_is_empty(&ifp->if_snd))
1198 		ifp->if_start(ifp);
1199 }
1200 
1201 static void
1202 nfe_init(void *xsc)
1203 {
1204 	struct nfe_softc *sc = xsc;
1205 	struct ifnet *ifp = &sc->arpcom.ac_if;
1206 	uint32_t tmp;
1207 	int error;
1208 
1209 	nfe_stop(sc);
1210 
1211 	error = nfe_init_tx_ring(sc, &sc->txq);
1212 	if (error) {
1213 		nfe_stop(sc);
1214 		return;
1215 	}
1216 
1217 	error = nfe_init_rx_ring(sc, &sc->rxq);
1218 	if (error) {
1219 		nfe_stop(sc);
1220 		return;
1221 	}
1222 
1223 	NFE_WRITE(sc, NFE_TX_UNK, 0);
1224 	NFE_WRITE(sc, NFE_STATUS, 0);
1225 
1226 	sc->rxtxctl = NFE_RXTX_BIT2;
1227 	if (sc->sc_flags & NFE_40BIT_ADDR)
1228 		sc->rxtxctl |= NFE_RXTX_V3MAGIC;
1229 	else if (sc->sc_flags & NFE_JUMBO_SUP)
1230 		sc->rxtxctl |= NFE_RXTX_V2MAGIC;
1231 #ifdef NFE_CSUM
1232 	if (sc->sc_flags & NFE_HW_CSUM)
1233 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1234 #endif
1235 
1236 	/*
1237 	 * Although the adapter is capable of stripping VLAN tags from received
1238 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1239 	 * purpose.  This will be done in software by our network stack.
1240 	 */
1241 	if (sc->sc_flags & NFE_HW_VLAN)
1242 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1243 
1244 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1245 	DELAY(10);
1246 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1247 
1248 	if (sc->sc_flags & NFE_HW_VLAN)
1249 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1250 
1251 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1252 
1253 	/* set MAC address */
1254 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1255 
1256 	/* tell MAC where rings are in memory */
1257 #ifdef __LP64__
1258 	NFE_WRITE(sc, NFE_RX_RING_ADDR_HI, sc->rxq.physaddr >> 32);
1259 #endif
1260 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, sc->rxq.physaddr & 0xffffffff);
1261 #ifdef __LP64__
1262 	NFE_WRITE(sc, NFE_TX_RING_ADDR_HI, sc->txq.physaddr >> 32);
1263 #endif
1264 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, sc->txq.physaddr & 0xffffffff);
1265 
1266 	NFE_WRITE(sc, NFE_RING_SIZE,
1267 	    (NFE_RX_RING_COUNT - 1) << 16 |
1268 	    (NFE_TX_RING_COUNT - 1));
1269 
1270 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1271 
1272 	/* force MAC to wakeup */
1273 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1274 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1275 	DELAY(10);
1276 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1277 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1278 
1279 	/*
1280 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
1281 	 * It is unclear how wide the timer is.  Base programming does
1282 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
1283 	 * we don't get any interrupt moderation.  TX moderation is
1284 	 * possible by using the timer interrupt instead of TX_DONE.
1285 	 *
1286 	 * It is unclear whether there are other bits that can be
1287 	 * set to make the NFE device actually do interrupt moderation
1288 	 * on the RX side.
1289 	 *
1290 	 * For now set a 128uS interval as a placemark, but don't use
1291 	 * the timer.
1292 	 */
1293 	NFE_WRITE(sc, NFE_IMTIMER, NFE_IM_DEFAULT);
1294 
1295 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1296 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1297 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1298 
1299 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1300 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1301 
1302 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1303 	NFE_WRITE(sc, NFE_WOL_CTL, NFE_WOL_MAGIC);
1304 
1305 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1306 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1307 	DELAY(10);
1308 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1309 
1310 	/* set Rx filter */
1311 	nfe_setmulti(sc);
1312 
1313 	nfe_ifmedia_upd(ifp);
1314 
1315 	/* enable Rx */
1316 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1317 
1318 	/* enable Tx */
1319 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1320 
1321 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1322 
1323 #ifdef DEVICE_POLLING
1324 	if ((ifp->if_flags & IFF_POLLING) == 0)
1325 #endif
1326 	/* enable interrupts */
1327 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1328 
1329 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1330 
1331 	ifp->if_flags |= IFF_RUNNING;
1332 	ifp->if_flags &= ~IFF_OACTIVE;
1333 }
1334 
1335 static void
1336 nfe_stop(struct nfe_softc *sc)
1337 {
1338 	struct ifnet *ifp = &sc->arpcom.ac_if;
1339 
1340 	callout_stop(&sc->sc_tick_ch);
1341 
1342 	ifp->if_timer = 0;
1343 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1344 
1345 	/* Abort Tx */
1346 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1347 
1348 	/* Disable Rx */
1349 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1350 
1351 	/* Disable interrupts */
1352 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1353 
1354 	/* Reset Tx and Rx rings */
1355 	nfe_reset_tx_ring(sc, &sc->txq);
1356 	nfe_reset_rx_ring(sc, &sc->rxq);
1357 }
1358 
1359 static int
1360 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1361 {
1362 	int i, j, error, descsize;
1363 	void **desc;
1364 
1365 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1366 		desc = (void **)&ring->desc64;
1367 		descsize = sizeof(struct nfe_desc64);
1368 	} else {
1369 		desc = (void **)&ring->desc32;
1370 		descsize = sizeof(struct nfe_desc32);
1371 	}
1372 
1373 	ring->bufsz = MCLBYTES;
1374 	ring->cur = ring->next = 0;
1375 
1376 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1377 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1378 				   NULL, NULL,
1379 				   NFE_RX_RING_COUNT * descsize, 1,
1380 				   NFE_RX_RING_COUNT * descsize,
1381 				   0, &ring->tag);
1382 	if (error) {
1383 		if_printf(&sc->arpcom.ac_if,
1384 			  "could not create desc RX DMA tag\n");
1385 		return error;
1386 	}
1387 
1388 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1389 				 &ring->map);
1390 	if (error) {
1391 		if_printf(&sc->arpcom.ac_if,
1392 			  "could not allocate RX desc DMA memory\n");
1393 		bus_dma_tag_destroy(ring->tag);
1394 		ring->tag = NULL;
1395 		return error;
1396 	}
1397 
1398 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1399 				NFE_RX_RING_COUNT * descsize,
1400 				nfe_ring_dma_addr, &ring->physaddr,
1401 				BUS_DMA_WAITOK);
1402 	if (error) {
1403 		if_printf(&sc->arpcom.ac_if,
1404 			  "could not load RX desc DMA map\n");
1405 		bus_dmamem_free(ring->tag, *desc, ring->map);
1406 		bus_dma_tag_destroy(ring->tag);
1407 		ring->tag = NULL;
1408 		return error;
1409 	}
1410 
1411 	if (sc->sc_flags & NFE_USE_JUMBO) {
1412 		ring->bufsz = NFE_JBYTES;
1413 
1414 		error = nfe_jpool_alloc(sc, ring);
1415 		if (error) {
1416 			if_printf(&sc->arpcom.ac_if,
1417 				  "could not allocate jumbo frames\n");
1418 			return error;
1419 		}
1420 	}
1421 
1422 	error = bus_dma_tag_create(NULL, 1, 0,
1423 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1424 				   NULL, NULL,
1425 				   MCLBYTES, 1, MCLBYTES,
1426 				   0, &ring->data_tag);
1427 	if (error) {
1428 		if_printf(&sc->arpcom.ac_if,
1429 			  "could not create RX mbuf DMA tag\n");
1430 		return error;
1431 	}
1432 
1433 	/* Create a spare RX mbuf DMA map */
1434 	error = bus_dmamap_create(ring->data_tag, 0, &ring->data_tmpmap);
1435 	if (error) {
1436 		if_printf(&sc->arpcom.ac_if,
1437 			  "could not create spare RX mbuf DMA map\n");
1438 		bus_dma_tag_destroy(ring->data_tag);
1439 		ring->data_tag = NULL;
1440 		return error;
1441 	}
1442 
1443 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1444 		error = bus_dmamap_create(ring->data_tag, 0,
1445 					  &ring->data[i].map);
1446 		if (error) {
1447 			if_printf(&sc->arpcom.ac_if,
1448 				  "could not create %dth RX mbuf DMA mapn", i);
1449 			goto fail;
1450 		}
1451 	}
1452 	return 0;
1453 fail:
1454 	for (j = 0; j < i; ++j)
1455 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1456 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1457 	bus_dma_tag_destroy(ring->data_tag);
1458 	ring->data_tag = NULL;
1459 	return error;
1460 }
1461 
1462 static void
1463 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1464 {
1465 	int i;
1466 
1467 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1468 		struct nfe_rx_data *data = &ring->data[i];
1469 
1470 		if (data->m != NULL) {
1471 			bus_dmamap_unload(ring->data_tag, data->map);
1472 			m_freem(data->m);
1473 			data->m = NULL;
1474 		}
1475 	}
1476 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1477 
1478 	ring->cur = ring->next = 0;
1479 }
1480 
1481 static int
1482 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1483 {
1484 	int i;
1485 
1486 	for (i = 0; i < NFE_RX_RING_COUNT; ++i) {
1487 		int error;
1488 
1489 		/* XXX should use a function pointer */
1490 		if (sc->sc_flags & NFE_USE_JUMBO)
1491 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1492 		else
1493 			error = nfe_newbuf_std(sc, ring, i, 1);
1494 		if (error) {
1495 			if_printf(&sc->arpcom.ac_if,
1496 				  "could not allocate RX buffer\n");
1497 			return error;
1498 		}
1499 
1500 		nfe_set_ready_rxdesc(sc, ring, i);
1501 	}
1502 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1503 
1504 	return 0;
1505 }
1506 
1507 static void
1508 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1509 {
1510 	if (ring->data_tag != NULL) {
1511 		struct nfe_rx_data *data;
1512 		int i;
1513 
1514 		for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1515 			data = &ring->data[i];
1516 
1517 			if (data->m != NULL) {
1518 				bus_dmamap_unload(ring->data_tag, data->map);
1519 				m_freem(data->m);
1520 			}
1521 			bus_dmamap_destroy(ring->data_tag, data->map);
1522 		}
1523 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1524 		bus_dma_tag_destroy(ring->data_tag);
1525 	}
1526 
1527 	nfe_jpool_free(sc, ring);
1528 
1529 	if (ring->tag != NULL) {
1530 		void *desc;
1531 
1532 		if (sc->sc_flags & NFE_40BIT_ADDR)
1533 			desc = ring->desc64;
1534 		else
1535 			desc = ring->desc32;
1536 
1537 		bus_dmamap_unload(ring->tag, ring->map);
1538 		bus_dmamem_free(ring->tag, desc, ring->map);
1539 		bus_dma_tag_destroy(ring->tag);
1540 	}
1541 }
1542 
1543 static struct nfe_jbuf *
1544 nfe_jalloc(struct nfe_softc *sc)
1545 {
1546 	struct ifnet *ifp = &sc->arpcom.ac_if;
1547 	struct nfe_jbuf *jbuf;
1548 
1549 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1550 
1551 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1552 	if (jbuf != NULL) {
1553 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1554 		jbuf->inuse = 1;
1555 	} else {
1556 		if_printf(ifp, "no free jumbo buffer\n");
1557 	}
1558 
1559 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1560 
1561 	return jbuf;
1562 }
1563 
1564 static void
1565 nfe_jfree(void *arg)
1566 {
1567 	struct nfe_jbuf *jbuf = arg;
1568 	struct nfe_softc *sc = jbuf->sc;
1569 	struct nfe_rx_ring *ring = jbuf->ring;
1570 
1571 	if (&ring->jbuf[jbuf->slot] != jbuf)
1572 		panic("%s: free wrong jumbo buffer\n", __func__);
1573 	else if (jbuf->inuse == 0)
1574 		panic("%s: jumbo buffer already freed\n", __func__);
1575 
1576 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1577 	atomic_subtract_int(&jbuf->inuse, 1);
1578 	if (jbuf->inuse == 0)
1579 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1580 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1581 }
1582 
1583 static void
1584 nfe_jref(void *arg)
1585 {
1586 	struct nfe_jbuf *jbuf = arg;
1587 	struct nfe_rx_ring *ring = jbuf->ring;
1588 
1589 	if (&ring->jbuf[jbuf->slot] != jbuf)
1590 		panic("%s: ref wrong jumbo buffer\n", __func__);
1591 	else if (jbuf->inuse == 0)
1592 		panic("%s: jumbo buffer already freed\n", __func__);
1593 
1594 	atomic_add_int(&jbuf->inuse, 1);
1595 }
1596 
1597 static int
1598 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1599 {
1600 	struct nfe_jbuf *jbuf;
1601 	bus_addr_t physaddr;
1602 	caddr_t buf;
1603 	int i, error;
1604 
1605 	/*
1606 	 * Allocate a big chunk of DMA'able memory.
1607 	 */
1608 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1609 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1610 				   NULL, NULL,
1611 				   NFE_JPOOL_SIZE, 1, NFE_JPOOL_SIZE,
1612 				   0, &ring->jtag);
1613 	if (error) {
1614 		if_printf(&sc->arpcom.ac_if,
1615 			  "could not create jumbo DMA tag\n");
1616 		return error;
1617 	}
1618 
1619 	error = bus_dmamem_alloc(ring->jtag, (void **)&ring->jpool,
1620 				 BUS_DMA_WAITOK, &ring->jmap);
1621 	if (error) {
1622 		if_printf(&sc->arpcom.ac_if,
1623 			  "could not allocate jumbo DMA memory\n");
1624 		bus_dma_tag_destroy(ring->jtag);
1625 		ring->jtag = NULL;
1626 		return error;
1627 	}
1628 
1629 	error = bus_dmamap_load(ring->jtag, ring->jmap, ring->jpool,
1630 				NFE_JPOOL_SIZE, nfe_ring_dma_addr, &physaddr,
1631 				BUS_DMA_WAITOK);
1632 	if (error) {
1633 		if_printf(&sc->arpcom.ac_if,
1634 			  "could not load jumbo DMA map\n");
1635 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1636 		bus_dma_tag_destroy(ring->jtag);
1637 		ring->jtag = NULL;
1638 		return error;
1639 	}
1640 
1641 	/* ..and split it into 9KB chunks */
1642 	SLIST_INIT(&ring->jfreelist);
1643 
1644 	buf = ring->jpool;
1645 	for (i = 0; i < NFE_JPOOL_COUNT; i++) {
1646 		jbuf = &ring->jbuf[i];
1647 
1648 		jbuf->sc = sc;
1649 		jbuf->ring = ring;
1650 		jbuf->inuse = 0;
1651 		jbuf->slot = i;
1652 		jbuf->buf = buf;
1653 		jbuf->physaddr = physaddr;
1654 
1655 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1656 
1657 		buf += NFE_JBYTES;
1658 		physaddr += NFE_JBYTES;
1659 	}
1660 
1661 	return 0;
1662 }
1663 
1664 static void
1665 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1666 {
1667 	if (ring->jtag != NULL) {
1668 		bus_dmamap_unload(ring->jtag, ring->jmap);
1669 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1670 		bus_dma_tag_destroy(ring->jtag);
1671 	}
1672 }
1673 
1674 static int
1675 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1676 {
1677 	int i, j, error, descsize;
1678 	void **desc;
1679 
1680 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1681 		desc = (void **)&ring->desc64;
1682 		descsize = sizeof(struct nfe_desc64);
1683 	} else {
1684 		desc = (void **)&ring->desc32;
1685 		descsize = sizeof(struct nfe_desc32);
1686 	}
1687 
1688 	ring->queued = 0;
1689 	ring->cur = ring->next = 0;
1690 
1691 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1692 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1693 				   NULL, NULL,
1694 				   NFE_TX_RING_COUNT * descsize, 1,
1695 				   NFE_TX_RING_COUNT * descsize,
1696 				   0, &ring->tag);
1697 	if (error) {
1698 		if_printf(&sc->arpcom.ac_if,
1699 			  "could not create TX desc DMA map\n");
1700 		return error;
1701 	}
1702 
1703 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1704 				 &ring->map);
1705 	if (error) {
1706 		if_printf(&sc->arpcom.ac_if,
1707 			  "could not allocate TX desc DMA memory\n");
1708 		bus_dma_tag_destroy(ring->tag);
1709 		ring->tag = NULL;
1710 		return error;
1711 	}
1712 
1713 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1714 				NFE_TX_RING_COUNT * descsize,
1715 				nfe_ring_dma_addr, &ring->physaddr,
1716 				BUS_DMA_WAITOK);
1717 	if (error) {
1718 		if_printf(&sc->arpcom.ac_if,
1719 			  "could not load TX desc DMA map\n");
1720 		bus_dmamem_free(ring->tag, *desc, ring->map);
1721 		bus_dma_tag_destroy(ring->tag);
1722 		ring->tag = NULL;
1723 		return error;
1724 	}
1725 
1726 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1727 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1728 				   NULL, NULL,
1729 				   NFE_JBYTES * NFE_MAX_SCATTER,
1730 				   NFE_MAX_SCATTER, NFE_JBYTES,
1731 				   0, &ring->data_tag);
1732 	if (error) {
1733 		if_printf(&sc->arpcom.ac_if,
1734 			  "could not create TX buf DMA tag\n");
1735 		return error;
1736 	}
1737 
1738 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1739 		error = bus_dmamap_create(ring->data_tag, 0,
1740 					  &ring->data[i].map);
1741 		if (error) {
1742 			if_printf(&sc->arpcom.ac_if,
1743 				  "could not create %dth TX buf DMA map\n", i);
1744 			goto fail;
1745 		}
1746 	}
1747 
1748 	return 0;
1749 fail:
1750 	for (j = 0; j < i; ++j)
1751 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1752 	bus_dma_tag_destroy(ring->data_tag);
1753 	ring->data_tag = NULL;
1754 	return error;
1755 }
1756 
1757 static void
1758 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1759 {
1760 	int i;
1761 
1762 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1763 		struct nfe_tx_data *data = &ring->data[i];
1764 
1765 		if (sc->sc_flags & NFE_40BIT_ADDR)
1766 			ring->desc64[i].flags = 0;
1767 		else
1768 			ring->desc32[i].flags = 0;
1769 
1770 		if (data->m != NULL) {
1771 			bus_dmamap_sync(ring->data_tag, data->map,
1772 					BUS_DMASYNC_POSTWRITE);
1773 			bus_dmamap_unload(ring->data_tag, data->map);
1774 			m_freem(data->m);
1775 			data->m = NULL;
1776 		}
1777 	}
1778 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1779 
1780 	ring->queued = 0;
1781 	ring->cur = ring->next = 0;
1782 }
1783 
1784 static int
1785 nfe_init_tx_ring(struct nfe_softc *sc __unused,
1786 		 struct nfe_tx_ring *ring __unused)
1787 {
1788 	return 0;
1789 }
1790 
1791 static void
1792 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1793 {
1794 	if (ring->data_tag != NULL) {
1795 		struct nfe_tx_data *data;
1796 		int i;
1797 
1798 		for (i = 0; i < NFE_TX_RING_COUNT; ++i) {
1799 			data = &ring->data[i];
1800 
1801 			if (data->m != NULL) {
1802 				bus_dmamap_unload(ring->data_tag, data->map);
1803 				m_freem(data->m);
1804 			}
1805 			bus_dmamap_destroy(ring->data_tag, data->map);
1806 		}
1807 
1808 		bus_dma_tag_destroy(ring->data_tag);
1809 	}
1810 
1811 	if (ring->tag != NULL) {
1812 		void *desc;
1813 
1814 		if (sc->sc_flags & NFE_40BIT_ADDR)
1815 			desc = ring->desc64;
1816 		else
1817 			desc = ring->desc32;
1818 
1819 		bus_dmamap_unload(ring->tag, ring->map);
1820 		bus_dmamem_free(ring->tag, desc, ring->map);
1821 		bus_dma_tag_destroy(ring->tag);
1822 	}
1823 }
1824 
1825 static int
1826 nfe_ifmedia_upd(struct ifnet *ifp)
1827 {
1828 	struct nfe_softc *sc = ifp->if_softc;
1829 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1830 
1831 	if (mii->mii_instance != 0) {
1832 		struct mii_softc *miisc;
1833 
1834 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
1835 			mii_phy_reset(miisc);
1836 	}
1837 	mii_mediachg(mii);
1838 
1839 	return 0;
1840 }
1841 
1842 static void
1843 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1844 {
1845 	struct nfe_softc *sc = ifp->if_softc;
1846 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1847 
1848 	mii_pollstat(mii);
1849 	ifmr->ifm_status = mii->mii_media_status;
1850 	ifmr->ifm_active = mii->mii_media_active;
1851 }
1852 
1853 static void
1854 nfe_setmulti(struct nfe_softc *sc)
1855 {
1856 	struct ifnet *ifp = &sc->arpcom.ac_if;
1857 	struct ifmultiaddr *ifma;
1858 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
1859 	uint32_t filter = NFE_RXFILTER_MAGIC;
1860 	int i;
1861 
1862 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1863 		bzero(addr, ETHER_ADDR_LEN);
1864 		bzero(mask, ETHER_ADDR_LEN);
1865 		goto done;
1866 	}
1867 
1868 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
1869 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
1870 
1871 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1872 		caddr_t maddr;
1873 
1874 		if (ifma->ifma_addr->sa_family != AF_LINK)
1875 			continue;
1876 
1877 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
1878 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
1879 			addr[i] &= maddr[i];
1880 			mask[i] &= ~maddr[i];
1881 		}
1882 	}
1883 
1884 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1885 		mask[i] |= addr[i];
1886 
1887 done:
1888 	addr[0] |= 0x01;	/* make sure multicast bit is set */
1889 
1890 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
1891 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1892 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
1893 	    addr[5] <<  8 | addr[4]);
1894 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
1895 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
1896 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
1897 	    mask[5] <<  8 | mask[4]);
1898 
1899 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
1900 	NFE_WRITE(sc, NFE_RXFILTER, filter);
1901 }
1902 
1903 static void
1904 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
1905 {
1906 	uint32_t tmp;
1907 
1908 	tmp = NFE_READ(sc, NFE_MACADDR_LO);
1909 	addr[0] = (tmp >> 8) & 0xff;
1910 	addr[1] = (tmp & 0xff);
1911 
1912 	tmp = NFE_READ(sc, NFE_MACADDR_HI);
1913 	addr[2] = (tmp >> 24) & 0xff;
1914 	addr[3] = (tmp >> 16) & 0xff;
1915 	addr[4] = (tmp >>  8) & 0xff;
1916 	addr[5] = (tmp & 0xff);
1917 }
1918 
1919 static void
1920 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
1921 {
1922 	NFE_WRITE(sc, NFE_MACADDR_LO,
1923 	    addr[5] <<  8 | addr[4]);
1924 	NFE_WRITE(sc, NFE_MACADDR_HI,
1925 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1926 }
1927 
1928 static void
1929 nfe_tick(void *arg)
1930 {
1931 	struct nfe_softc *sc = arg;
1932 	struct ifnet *ifp = &sc->arpcom.ac_if;
1933 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1934 
1935 	lwkt_serialize_enter(ifp->if_serializer);
1936 
1937 	mii_tick(mii);
1938 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1939 
1940 	lwkt_serialize_exit(ifp->if_serializer);
1941 }
1942 
1943 static void
1944 nfe_ring_dma_addr(void *arg, bus_dma_segment_t *seg, int nseg, int error)
1945 {
1946 	if (error)
1947 		return;
1948 
1949 	KASSERT(nseg == 1, ("too many segments, should be 1\n"));
1950 
1951 	*((uint32_t *)arg) = seg->ds_addr;
1952 }
1953 
1954 static void
1955 nfe_buf_dma_addr(void *arg, bus_dma_segment_t *segs, int nsegs,
1956 		 bus_size_t mapsz __unused, int error)
1957 {
1958 	struct nfe_dma_ctx *ctx = arg;
1959 	int i;
1960 
1961 	if (error)
1962 		return;
1963 
1964 	KASSERT(nsegs <= ctx->nsegs,
1965 		("too many segments(%d), should be <= %d\n",
1966 		 nsegs, ctx->nsegs));
1967 
1968 	ctx->nsegs = nsegs;
1969 	for (i = 0; i < nsegs; ++i)
1970 		ctx->segs[i] = segs[i];
1971 }
1972 
1973 static int
1974 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
1975 	       int wait)
1976 {
1977 	struct nfe_rx_data *data = &ring->data[idx];
1978 	struct nfe_dma_ctx ctx;
1979 	bus_dma_segment_t seg;
1980 	bus_dmamap_t map;
1981 	struct mbuf *m;
1982 	int error;
1983 
1984 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
1985 	if (m == NULL)
1986 		return ENOBUFS;
1987 	m->m_len = m->m_pkthdr.len = MCLBYTES;
1988 
1989 	ctx.nsegs = 1;
1990 	ctx.segs = &seg;
1991 	error = bus_dmamap_load_mbuf(ring->data_tag, ring->data_tmpmap,
1992 				     m, nfe_buf_dma_addr, &ctx,
1993 				     wait ? BUS_DMA_WAITOK : BUS_DMA_NOWAIT);
1994 	if (error) {
1995 		m_freem(m);
1996 		if_printf(&sc->arpcom.ac_if, "could map RX mbuf %d\n", error);
1997 		return error;
1998 	}
1999 
2000 	/* Unload originally mapped mbuf */
2001 	bus_dmamap_unload(ring->data_tag, data->map);
2002 
2003 	/* Swap this DMA map with tmp DMA map */
2004 	map = data->map;
2005 	data->map = ring->data_tmpmap;
2006 	ring->data_tmpmap = map;
2007 
2008 	/* Caller is assumed to have collected the old mbuf */
2009 	data->m = m;
2010 
2011 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2012 
2013 	bus_dmamap_sync(ring->data_tag, data->map, BUS_DMASYNC_PREREAD);
2014 	return 0;
2015 }
2016 
2017 static int
2018 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2019 		 int wait)
2020 {
2021 	struct nfe_rx_data *data = &ring->data[idx];
2022 	struct nfe_jbuf *jbuf;
2023 	struct mbuf *m;
2024 
2025 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2026 	if (m == NULL)
2027 		return ENOBUFS;
2028 
2029 	jbuf = nfe_jalloc(sc);
2030 	if (jbuf == NULL) {
2031 		m_freem(m);
2032 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2033 		    "-- packet dropped!\n");
2034 		return ENOBUFS;
2035 	}
2036 
2037 	m->m_ext.ext_arg = jbuf;
2038 	m->m_ext.ext_buf = jbuf->buf;
2039 	m->m_ext.ext_free = nfe_jfree;
2040 	m->m_ext.ext_ref = nfe_jref;
2041 	m->m_ext.ext_size = NFE_JBYTES;
2042 
2043 	m->m_data = m->m_ext.ext_buf;
2044 	m->m_flags |= M_EXT;
2045 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2046 
2047 	/* Caller is assumed to have collected the old mbuf */
2048 	data->m = m;
2049 
2050 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2051 
2052 	bus_dmamap_sync(ring->jtag, ring->jmap, BUS_DMASYNC_PREREAD);
2053 	return 0;
2054 }
2055 
2056 static void
2057 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2058 		     bus_addr_t physaddr)
2059 {
2060 	if (sc->sc_flags & NFE_40BIT_ADDR) {
2061 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2062 
2063 #if defined(__LP64__)
2064 		desc64->physaddr[0] = htole32(physaddr >> 32);
2065 #endif
2066 		desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
2067 	} else {
2068 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2069 
2070 		desc32->physaddr = htole32(physaddr);
2071 	}
2072 }
2073 
2074 static void
2075 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2076 {
2077 	if (sc->sc_flags & NFE_40BIT_ADDR) {
2078 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2079 
2080 		desc64->length = htole16(ring->bufsz);
2081 		desc64->flags = htole16(NFE_RX_READY);
2082 	} else {
2083 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2084 
2085 		desc32->length = htole16(ring->bufsz);
2086 		desc32->flags = htole16(NFE_RX_READY);
2087 	}
2088 }
2089