xref: /netbsd-src/sys/dev/pci/if_nfe.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: if_nfe.c,v 1.55 2012/01/30 19:41:20 drochner Exp $	*/
2 /*	$OpenBSD: if_nfe.c,v 1.77 2008/02/05 16:52:50 brad Exp $	*/
3 
4 /*-
5  * Copyright (c) 2006, 2007 Damien Bergamini <damien.bergamini@free.fr>
6  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
22 
23 #include <sys/cdefs.h>
24 __KERNEL_RCSID(0, "$NetBSD: if_nfe.c,v 1.55 2012/01/30 19:41:20 drochner Exp $");
25 
26 #include "opt_inet.h"
27 #include "vlan.h"
28 
29 #include <sys/param.h>
30 #include <sys/endian.h>
31 #include <sys/systm.h>
32 #include <sys/types.h>
33 #include <sys/sockio.h>
34 #include <sys/mbuf.h>
35 #include <sys/mutex.h>
36 #include <sys/queue.h>
37 #include <sys/kernel.h>
38 #include <sys/device.h>
39 #include <sys/callout.h>
40 #include <sys/socket.h>
41 
42 #include <sys/bus.h>
43 
44 #include <net/if.h>
45 #include <net/if_dl.h>
46 #include <net/if_media.h>
47 #include <net/if_ether.h>
48 #include <net/if_arp.h>
49 
50 #ifdef INET
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/in_var.h>
54 #include <netinet/ip.h>
55 #include <netinet/if_inarp.h>
56 #endif
57 
58 #if NVLAN > 0
59 #include <net/if_types.h>
60 #endif
61 
62 #include <net/bpf.h>
63 
64 #include <dev/mii/mii.h>
65 #include <dev/mii/miivar.h>
66 
67 #include <dev/pci/pcireg.h>
68 #include <dev/pci/pcivar.h>
69 #include <dev/pci/pcidevs.h>
70 
71 #include <dev/pci/if_nfereg.h>
72 #include <dev/pci/if_nfevar.h>
73 
74 static int nfe_ifflags_cb(struct ethercom *);
75 
76 int	nfe_match(device_t, cfdata_t, void *);
77 void	nfe_attach(device_t, device_t, void *);
78 int	nfe_detach(device_t, int);
79 void	nfe_power(int, void *);
80 void	nfe_miibus_statchg(device_t);
81 int	nfe_miibus_readreg(device_t, int, int);
82 void	nfe_miibus_writereg(device_t, int, int, int);
83 int	nfe_intr(void *);
84 int	nfe_ioctl(struct ifnet *, u_long, void *);
85 void	nfe_txdesc32_sync(struct nfe_softc *, struct nfe_desc32 *, int);
86 void	nfe_txdesc64_sync(struct nfe_softc *, struct nfe_desc64 *, int);
87 void	nfe_txdesc32_rsync(struct nfe_softc *, int, int, int);
88 void	nfe_txdesc64_rsync(struct nfe_softc *, int, int, int);
89 void	nfe_rxdesc32_sync(struct nfe_softc *, struct nfe_desc32 *, int);
90 void	nfe_rxdesc64_sync(struct nfe_softc *, struct nfe_desc64 *, int);
91 void	nfe_rxeof(struct nfe_softc *);
92 void	nfe_txeof(struct nfe_softc *);
93 int	nfe_encap(struct nfe_softc *, struct mbuf *);
94 void	nfe_start(struct ifnet *);
95 void	nfe_watchdog(struct ifnet *);
96 int	nfe_init(struct ifnet *);
97 void	nfe_stop(struct ifnet *, int);
98 struct	nfe_jbuf *nfe_jalloc(struct nfe_softc *, int);
99 void	nfe_jfree(struct mbuf *, void *, size_t, void *);
100 int	nfe_jpool_alloc(struct nfe_softc *);
101 void	nfe_jpool_free(struct nfe_softc *);
102 int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
103 void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
104 void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
105 int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
106 void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
107 void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
108 void	nfe_setmulti(struct nfe_softc *);
109 void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
110 void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
111 void	nfe_tick(void *);
112 void	nfe_poweron(device_t);
113 bool	nfe_resume(device_t, const pmf_qual_t *);
114 
115 CFATTACH_DECL_NEW(nfe, sizeof(struct nfe_softc),
116     nfe_match, nfe_attach, nfe_detach, NULL);
117 
118 /* #define NFE_NO_JUMBO */
119 
120 #ifdef NFE_DEBUG
121 int nfedebug = 0;
122 #define DPRINTF(x)	do { if (nfedebug) printf x; } while (0)
123 #define DPRINTFN(n,x)	do { if (nfedebug >= (n)) printf x; } while (0)
124 #else
125 #define DPRINTF(x)
126 #define DPRINTFN(n,x)
127 #endif
128 
129 /* deal with naming differences */
130 
131 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN2 \
132 	PCI_PRODUCT_NVIDIA_NFORCE2_400_LAN1
133 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN3 \
134 	PCI_PRODUCT_NVIDIA_NFORCE2_400_LAN2
135 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN5 \
136 	PCI_PRODUCT_NVIDIA_NFORCE3_250_LAN
137 
138 #define	PCI_PRODUCT_NVIDIA_CK804_LAN1 \
139 	PCI_PRODUCT_NVIDIA_NFORCE4_LAN1
140 #define	PCI_PRODUCT_NVIDIA_CK804_LAN2 \
141 	PCI_PRODUCT_NVIDIA_NFORCE4_LAN2
142 
143 #define	PCI_PRODUCT_NVIDIA_MCP51_LAN1 \
144 	PCI_PRODUCT_NVIDIA_NFORCE430_LAN1
145 #define	PCI_PRODUCT_NVIDIA_MCP51_LAN2 \
146 	PCI_PRODUCT_NVIDIA_NFORCE430_LAN2
147 
148 #ifdef	_LP64
149 #define	__LP64__ 1
150 #endif
151 
152 const struct nfe_product {
153 	pci_vendor_id_t		vendor;
154 	pci_product_id_t	product;
155 } nfe_devices[] = {
156 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN },
157 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN },
158 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1 },
159 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2 },
160 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3 },
161 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4 },
162 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5 },
163 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1 },
164 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2 },
165 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1 },
166 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2 },
167 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1 },
168 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2 },
169 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1 },
170 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2 },
171 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1 },
172 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2 },
173 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3 },
174 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4 },
175 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1 },
176 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2 },
177 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3 },
178 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4 },
179 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1 },
180 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2 },
181 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3 },
182 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4 },
183 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1 },
184 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2 },
185 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3 },
186 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4 },
187 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1 },
188 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2 },
189 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3 },
190 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4 },
191 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1 },
192 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2 },
193 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3 },
194 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4 }
195 };
196 
197 int
198 nfe_match(device_t dev, cfdata_t match, void *aux)
199 {
200 	struct pci_attach_args *pa = aux;
201 	const struct nfe_product *np;
202 	int i;
203 
204 	for (i = 0; i < __arraycount(nfe_devices); i++) {
205 		np = &nfe_devices[i];
206 		if (PCI_VENDOR(pa->pa_id) == np->vendor &&
207 		    PCI_PRODUCT(pa->pa_id) == np->product)
208 			return 1;
209 	}
210 	return 0;
211 }
212 
213 void
214 nfe_attach(device_t parent, device_t self, void *aux)
215 {
216 	struct nfe_softc *sc = device_private(self);
217 	struct pci_attach_args *pa = aux;
218 	pci_chipset_tag_t pc = pa->pa_pc;
219 	pci_intr_handle_t ih;
220 	const char *intrstr;
221 	struct ifnet *ifp;
222 	pcireg_t memtype, csr;
223 	int mii_flags = 0;
224 
225 	sc->sc_dev = self;
226 	sc->sc_pc = pa->pa_pc;
227 	pci_aprint_devinfo(pa, NULL);
228 
229 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, NFE_PCI_BA);
230 	switch (memtype) {
231 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
232 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
233 		if (pci_mapreg_map(pa, NFE_PCI_BA, memtype, 0, &sc->sc_memt,
234 		    &sc->sc_memh, NULL, &sc->sc_mems) == 0)
235 			break;
236 		/* FALLTHROUGH */
237 	default:
238 		aprint_error_dev(self, "could not map mem space\n");
239 		return;
240 	}
241 
242 	if (pci_intr_map(pa, &ih) != 0) {
243 		aprint_error_dev(self, "could not map interrupt\n");
244 		goto fail;
245 	}
246 
247 	intrstr = pci_intr_string(pc, ih);
248 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, nfe_intr, sc);
249 	if (sc->sc_ih == NULL) {
250 		aprint_error_dev(self, "could not establish interrupt");
251 		if (intrstr != NULL)
252 			aprint_error(" at %s", intrstr);
253 		aprint_error("\n");
254 		goto fail;
255 	}
256 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
257 
258 	sc->sc_dmat = pa->pa_dmat;
259 
260 	csr = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
261 	csr |= PCI_COMMAND_MASTER_ENABLE;
262 	pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, csr);
263 
264 	sc->sc_flags = 0;
265 
266 	switch (PCI_PRODUCT(pa->pa_id)) {
267 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
268 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
269 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
270 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
271 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_HW_CSUM;
272 		break;
273 	case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
274 	case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
275 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_PWR_MGMT;
276 		break;
277 	case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
278 	case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
279 	case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
280 	case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
281 	case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
282 	case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
283 	case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
284 	case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
285 	case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
286 	case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
287 	case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
288 	case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
289 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_CORRECT_MACADDR |
290 		    NFE_PWR_MGMT;
291 		break;
292 	case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
293 	case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
294 	case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
295 	case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
296 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_HW_CSUM |
297 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
298 		break;
299 	case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
300 	case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
301 	case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
302 	case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
303 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM |
304 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
305 		break;
306 	case PCI_PRODUCT_NVIDIA_CK804_LAN1:
307 	case PCI_PRODUCT_NVIDIA_CK804_LAN2:
308 	case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
309 	case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
310 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM;
311 		break;
312 	case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
313 	case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
314 	case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
315 	case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
316 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR |
317 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
318 		mii_flags = MIIF_DOPAUSE;
319 		break;
320 	case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
321 	case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
322 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM |
323 		    NFE_HW_VLAN | NFE_PWR_MGMT;
324 		break;
325 	}
326 
327 	nfe_poweron(self);
328 
329 #ifndef NFE_NO_JUMBO
330 	/* enable jumbo frames for adapters that support it */
331 	if (sc->sc_flags & NFE_JUMBO_SUP)
332 		sc->sc_flags |= NFE_USE_JUMBO;
333 #endif
334 
335 	/* Check for reversed ethernet address */
336 	if ((NFE_READ(sc, NFE_TX_UNK) & NFE_MAC_ADDR_INORDER) != 0)
337 		sc->sc_flags |= NFE_CORRECT_MACADDR;
338 
339 	nfe_get_macaddr(sc, sc->sc_enaddr);
340 	aprint_normal_dev(self, "Ethernet address %s\n",
341 	    ether_sprintf(sc->sc_enaddr));
342 
343 	/*
344 	 * Allocate Tx and Rx rings.
345 	 */
346 	if (nfe_alloc_tx_ring(sc, &sc->txq) != 0) {
347 		aprint_error_dev(self, "could not allocate Tx ring\n");
348 		goto fail;
349 	}
350 
351 	mutex_init(&sc->rxq.mtx, MUTEX_DEFAULT, IPL_NET);
352 
353 	if (nfe_alloc_rx_ring(sc, &sc->rxq) != 0) {
354 		aprint_error_dev(self, "could not allocate Rx ring\n");
355 		nfe_free_tx_ring(sc, &sc->txq);
356 		goto fail;
357 	}
358 
359 	ifp = &sc->sc_ethercom.ec_if;
360 	ifp->if_softc = sc;
361 	ifp->if_mtu = ETHERMTU;
362 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
363 	ifp->if_ioctl = nfe_ioctl;
364 	ifp->if_start = nfe_start;
365 	ifp->if_stop = nfe_stop;
366 	ifp->if_watchdog = nfe_watchdog;
367 	ifp->if_init = nfe_init;
368 	ifp->if_baudrate = IF_Gbps(1);
369 	IFQ_SET_MAXLEN(&ifp->if_snd, NFE_IFQ_MAXLEN);
370 	IFQ_SET_READY(&ifp->if_snd);
371 	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
372 
373 	if (sc->sc_flags & NFE_USE_JUMBO)
374 		sc->sc_ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
375 
376 #if NVLAN > 0
377 	if (sc->sc_flags & NFE_HW_VLAN)
378 		sc->sc_ethercom.ec_capabilities |=
379 			ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
380 #endif
381 	if (sc->sc_flags & NFE_HW_CSUM) {
382 		ifp->if_capabilities |=
383 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
384 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
385 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
386 	}
387 
388 	sc->sc_mii.mii_ifp = ifp;
389 	sc->sc_mii.mii_readreg = nfe_miibus_readreg;
390 	sc->sc_mii.mii_writereg = nfe_miibus_writereg;
391 	sc->sc_mii.mii_statchg = nfe_miibus_statchg;
392 
393 	sc->sc_ethercom.ec_mii = &sc->sc_mii;
394 	ifmedia_init(&sc->sc_mii.mii_media, 0, ether_mediachange,
395 	    ether_mediastatus);
396 
397 	mii_attach(self, &sc->sc_mii, 0xffffffff, MII_PHY_ANY, 0, mii_flags);
398 
399 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
400 		aprint_error_dev(self, "no PHY found!\n");
401 		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER | IFM_MANUAL,
402 		    0, NULL);
403 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER | IFM_MANUAL);
404 	} else
405 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER | IFM_AUTO);
406 
407 	if_attach(ifp);
408 	ether_ifattach(ifp, sc->sc_enaddr);
409 	ether_set_ifflags_cb(&sc->sc_ethercom, nfe_ifflags_cb);
410 
411 	callout_init(&sc->sc_tick_ch, 0);
412 	callout_setfunc(&sc->sc_tick_ch, nfe_tick, sc);
413 
414 	if (pmf_device_register(self, NULL, nfe_resume))
415 		pmf_class_network_register(self, ifp);
416 	else
417 		aprint_error_dev(self, "couldn't establish power handler\n");
418 
419 	return;
420 
421 fail:
422 	if (sc->sc_ih != NULL) {
423 		pci_intr_disestablish(pc, sc->sc_ih);
424 		sc->sc_ih = NULL;
425 	}
426 	if (sc->sc_mems != 0) {
427 		bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
428 		sc->sc_mems = 0;
429 	}
430 }
431 
432 int
433 nfe_detach(device_t self, int flags)
434 {
435 	struct nfe_softc *sc = device_private(self);
436 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
437 	int s;
438 
439 	s = splnet();
440 
441 	nfe_stop(ifp, 1);
442 
443 	pmf_device_deregister(self);
444 	callout_destroy(&sc->sc_tick_ch);
445 	ether_ifdetach(ifp);
446 	if_detach(ifp);
447 	mii_detach(&sc->sc_mii, MII_PHY_ANY, MII_OFFSET_ANY);
448 
449 	nfe_free_rx_ring(sc, &sc->rxq);
450 	mutex_destroy(&sc->rxq.mtx);
451 	nfe_free_tx_ring(sc, &sc->txq);
452 
453 	if (sc->sc_ih != NULL) {
454 		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
455 		sc->sc_ih = NULL;
456 	}
457 
458 	if ((sc->sc_flags & NFE_CORRECT_MACADDR) != 0) {
459 		nfe_set_macaddr(sc, sc->sc_enaddr);
460 	} else {
461 		NFE_WRITE(sc, NFE_MACADDR_LO,
462 		    sc->sc_enaddr[0] <<  8 | sc->sc_enaddr[1]);
463 		NFE_WRITE(sc, NFE_MACADDR_HI,
464 		    sc->sc_enaddr[2] << 24 | sc->sc_enaddr[3] << 16 |
465 		    sc->sc_enaddr[4] <<  8 | sc->sc_enaddr[5]);
466 	}
467 
468 	if (sc->sc_mems != 0) {
469 		bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
470 		sc->sc_mems = 0;
471 	}
472 
473 	splx(s);
474 
475 	return 0;
476 }
477 
478 void
479 nfe_miibus_statchg(device_t dev)
480 {
481 	struct nfe_softc *sc = device_private(dev);
482 	struct mii_data *mii = &sc->sc_mii;
483 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
484 
485 	phy = NFE_READ(sc, NFE_PHY_IFACE);
486 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
487 
488 	seed = NFE_READ(sc, NFE_RNDSEED);
489 	seed &= ~NFE_SEED_MASK;
490 
491 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
492 		phy  |= NFE_PHY_HDX;	/* half-duplex */
493 		misc |= NFE_MISC1_HDX;
494 	}
495 
496 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
497 	case IFM_1000_T:	/* full-duplex only */
498 		link |= NFE_MEDIA_1000T;
499 		seed |= NFE_SEED_1000T;
500 		phy  |= NFE_PHY_1000T;
501 		break;
502 	case IFM_100_TX:
503 		link |= NFE_MEDIA_100TX;
504 		seed |= NFE_SEED_100TX;
505 		phy  |= NFE_PHY_100TX;
506 		break;
507 	case IFM_10_T:
508 		link |= NFE_MEDIA_10T;
509 		seed |= NFE_SEED_10T;
510 		break;
511 	}
512 
513 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
514 
515 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
516 	NFE_WRITE(sc, NFE_MISC1, misc);
517 	NFE_WRITE(sc, NFE_LINKSPEED, link);
518 }
519 
520 int
521 nfe_miibus_readreg(device_t dev, int phy, int reg)
522 {
523 	struct nfe_softc *sc = device_private(dev);
524 	uint32_t val;
525 	int ntries;
526 
527 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
528 
529 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
530 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
531 		DELAY(100);
532 	}
533 
534 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
535 
536 	for (ntries = 0; ntries < 1000; ntries++) {
537 		DELAY(100);
538 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
539 			break;
540 	}
541 	if (ntries == 1000) {
542 		DPRINTFN(2, ("%s: timeout waiting for PHY\n",
543 		    device_xname(sc->sc_dev)));
544 		return 0;
545 	}
546 
547 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
548 		DPRINTFN(2, ("%s: could not read PHY\n",
549 		    device_xname(sc->sc_dev)));
550 		return 0;
551 	}
552 
553 	val = NFE_READ(sc, NFE_PHY_DATA);
554 	if (val != 0xffffffff && val != 0)
555 		sc->mii_phyaddr = phy;
556 
557 	DPRINTFN(2, ("%s: mii read phy %d reg 0x%x ret 0x%x\n",
558 	    device_xname(sc->sc_dev), phy, reg, val));
559 
560 	return val;
561 }
562 
563 void
564 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
565 {
566 	struct nfe_softc *sc = device_private(dev);
567 	uint32_t ctl;
568 	int ntries;
569 
570 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
571 
572 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
573 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
574 		DELAY(100);
575 	}
576 
577 	NFE_WRITE(sc, NFE_PHY_DATA, val);
578 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
579 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
580 
581 	for (ntries = 0; ntries < 1000; ntries++) {
582 		DELAY(100);
583 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
584 			break;
585 	}
586 #ifdef NFE_DEBUG
587 	if (nfedebug >= 2 && ntries == 1000)
588 		printf("could not write to PHY\n");
589 #endif
590 }
591 
592 int
593 nfe_intr(void *arg)
594 {
595 	struct nfe_softc *sc = arg;
596 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
597 	uint32_t r;
598 	int handled;
599 
600 	if ((ifp->if_flags & IFF_UP) == 0)
601 		return 0;
602 
603 	handled = 0;
604 
605 	for (;;) {
606 		r = NFE_READ(sc, NFE_IRQ_STATUS);
607 		if ((r & NFE_IRQ_WANTED) == 0)
608 			break;
609 
610 		NFE_WRITE(sc, NFE_IRQ_STATUS, r);
611 		handled = 1;
612 		DPRINTFN(5, ("nfe_intr: interrupt register %x\n", r));
613 
614 		if ((r & (NFE_IRQ_RXERR|NFE_IRQ_RX_NOBUF|NFE_IRQ_RX)) != 0) {
615 			/* check Rx ring */
616 			nfe_rxeof(sc);
617 		}
618 		if ((r & (NFE_IRQ_TXERR|NFE_IRQ_TXERR2|NFE_IRQ_TX_DONE)) != 0) {
619 			/* check Tx ring */
620 			nfe_txeof(sc);
621 		}
622 		if ((r & NFE_IRQ_LINK) != 0) {
623 			NFE_READ(sc, NFE_PHY_STATUS);
624 			NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
625 			DPRINTF(("%s: link state changed\n",
626 			    device_xname(sc->sc_dev)));
627 		}
628 	}
629 
630 	if (handled && !IF_IS_EMPTY(&ifp->if_snd))
631 		nfe_start(ifp);
632 
633 	return handled;
634 }
635 
636 static int
637 nfe_ifflags_cb(struct ethercom *ec)
638 {
639 	struct ifnet *ifp = &ec->ec_if;
640 	struct nfe_softc *sc = ifp->if_softc;
641 	int change = ifp->if_flags ^ sc->sc_if_flags;
642 
643 	/*
644 	 * If only the PROMISC flag changes, then
645 	 * don't do a full re-init of the chip, just update
646 	 * the Rx filter.
647 	 */
648 	if ((change & ~(IFF_CANTCHANGE|IFF_DEBUG)) != 0)
649 		return ENETRESET;
650 	else if ((change & IFF_PROMISC) != 0)
651 		nfe_setmulti(sc);
652 
653 	return 0;
654 }
655 
656 int
657 nfe_ioctl(struct ifnet *ifp, u_long cmd, void *data)
658 {
659 	struct nfe_softc *sc = ifp->if_softc;
660 	struct ifaddr *ifa = (struct ifaddr *)data;
661 	int s, error = 0;
662 
663 	s = splnet();
664 
665 	switch (cmd) {
666 	case SIOCINITIFADDR:
667 		ifp->if_flags |= IFF_UP;
668 		nfe_init(ifp);
669 		switch (ifa->ifa_addr->sa_family) {
670 #ifdef INET
671 		case AF_INET:
672 			arp_ifinit(ifp, ifa);
673 			break;
674 #endif
675 		default:
676 			break;
677 		}
678 		break;
679 	default:
680 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
681 			break;
682 
683 		error = 0;
684 
685 		if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
686 			;
687 		else if (ifp->if_flags & IFF_RUNNING)
688 			nfe_setmulti(sc);
689 		break;
690 	}
691 	sc->sc_if_flags = ifp->if_flags;
692 
693 	splx(s);
694 
695 	return error;
696 }
697 
698 void
699 nfe_txdesc32_sync(struct nfe_softc *sc, struct nfe_desc32 *desc32, int ops)
700 {
701 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
702 	    (char *)desc32 - (char *)sc->txq.desc32,
703 	    sizeof (struct nfe_desc32), ops);
704 }
705 
706 void
707 nfe_txdesc64_sync(struct nfe_softc *sc, struct nfe_desc64 *desc64, int ops)
708 {
709 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
710 	    (char *)desc64 - (char *)sc->txq.desc64,
711 	    sizeof (struct nfe_desc64), ops);
712 }
713 
714 void
715 nfe_txdesc32_rsync(struct nfe_softc *sc, int start, int end, int ops)
716 {
717 	if (end > start) {
718 		bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
719 		    (char *)&sc->txq.desc32[start] - (char *)sc->txq.desc32,
720 		    (char *)&sc->txq.desc32[end] -
721 		    (char *)&sc->txq.desc32[start], ops);
722 		return;
723 	}
724 	/* sync from 'start' to end of ring */
725 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
726 	    (char *)&sc->txq.desc32[start] - (char *)sc->txq.desc32,
727 	    (char *)&sc->txq.desc32[NFE_TX_RING_COUNT] -
728 	    (char *)&sc->txq.desc32[start], ops);
729 
730 	/* sync from start of ring to 'end' */
731 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map, 0,
732 	    (char *)&sc->txq.desc32[end] - (char *)sc->txq.desc32, ops);
733 }
734 
735 void
736 nfe_txdesc64_rsync(struct nfe_softc *sc, int start, int end, int ops)
737 {
738 	if (end > start) {
739 		bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
740 		    (char *)&sc->txq.desc64[start] - (char *)sc->txq.desc64,
741 		    (char *)&sc->txq.desc64[end] -
742 		    (char *)&sc->txq.desc64[start], ops);
743 		return;
744 	}
745 	/* sync from 'start' to end of ring */
746 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
747 	    (char *)&sc->txq.desc64[start] - (char *)sc->txq.desc64,
748 	    (char *)&sc->txq.desc64[NFE_TX_RING_COUNT] -
749 	    (char *)&sc->txq.desc64[start], ops);
750 
751 	/* sync from start of ring to 'end' */
752 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map, 0,
753 	    (char *)&sc->txq.desc64[end] - (char *)sc->txq.desc64, ops);
754 }
755 
756 void
757 nfe_rxdesc32_sync(struct nfe_softc *sc, struct nfe_desc32 *desc32, int ops)
758 {
759 	bus_dmamap_sync(sc->sc_dmat, sc->rxq.map,
760 	    (char *)desc32 - (char *)sc->rxq.desc32,
761 	    sizeof (struct nfe_desc32), ops);
762 }
763 
764 void
765 nfe_rxdesc64_sync(struct nfe_softc *sc, struct nfe_desc64 *desc64, int ops)
766 {
767 	bus_dmamap_sync(sc->sc_dmat, sc->rxq.map,
768 	    (char *)desc64 - (char *)sc->rxq.desc64,
769 	    sizeof (struct nfe_desc64), ops);
770 }
771 
772 void
773 nfe_rxeof(struct nfe_softc *sc)
774 {
775 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
776 	struct nfe_desc32 *desc32;
777 	struct nfe_desc64 *desc64;
778 	struct nfe_rx_data *data;
779 	struct nfe_jbuf *jbuf;
780 	struct mbuf *m, *mnew;
781 	bus_addr_t physaddr;
782 	uint16_t flags;
783 	int error, len, i;
784 
785 	desc32 = NULL;
786 	desc64 = NULL;
787 	for (i = sc->rxq.cur;; i = NFE_RX_NEXTDESC(i)) {
788 		data = &sc->rxq.data[i];
789 
790 		if (sc->sc_flags & NFE_40BIT_ADDR) {
791 			desc64 = &sc->rxq.desc64[i];
792 			nfe_rxdesc64_sync(sc, desc64,
793 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
794 
795 			flags = le16toh(desc64->flags);
796 			len = le16toh(desc64->length) & 0x3fff;
797 		} else {
798 			desc32 = &sc->rxq.desc32[i];
799 			nfe_rxdesc32_sync(sc, desc32,
800 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
801 
802 			flags = le16toh(desc32->flags);
803 			len = le16toh(desc32->length) & 0x3fff;
804 		}
805 
806 		if ((flags & NFE_RX_READY) != 0)
807 			break;
808 
809 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
810 			if ((flags & NFE_RX_VALID_V1) == 0)
811 				goto skip;
812 
813 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
814 				flags &= ~NFE_RX_ERROR;
815 				len--;	/* fix buffer length */
816 			}
817 		} else {
818 			if ((flags & NFE_RX_VALID_V2) == 0)
819 				goto skip;
820 
821 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
822 				flags &= ~NFE_RX_ERROR;
823 				len--;	/* fix buffer length */
824 			}
825 		}
826 
827 		if (flags & NFE_RX_ERROR) {
828 			ifp->if_ierrors++;
829 			goto skip;
830 		}
831 
832 		/*
833 		 * Try to allocate a new mbuf for this ring element and load
834 		 * it before processing the current mbuf. If the ring element
835 		 * cannot be loaded, drop the received packet and reuse the
836 		 * old mbuf. In the unlikely case that the old mbuf can't be
837 		 * reloaded either, explicitly panic.
838 		 */
839 		MGETHDR(mnew, M_DONTWAIT, MT_DATA);
840 		if (mnew == NULL) {
841 			ifp->if_ierrors++;
842 			goto skip;
843 		}
844 
845 		if (sc->sc_flags & NFE_USE_JUMBO) {
846 			physaddr =
847 			    sc->rxq.jbuf[sc->rxq.jbufmap[i]].physaddr;
848 			if ((jbuf = nfe_jalloc(sc, i)) == NULL) {
849 				if (len > MCLBYTES) {
850 					m_freem(mnew);
851 					ifp->if_ierrors++;
852 					goto skip1;
853 				}
854 				MCLGET(mnew, M_DONTWAIT);
855 				if ((mnew->m_flags & M_EXT) == 0) {
856 					m_freem(mnew);
857 					ifp->if_ierrors++;
858 					goto skip1;
859 				}
860 
861 				(void)memcpy(mtod(mnew, void *),
862 				    mtod(data->m, const void *), len);
863 				m = mnew;
864 				goto mbufcopied;
865 			} else {
866 				MEXTADD(mnew, jbuf->buf, NFE_JBYTES, 0, nfe_jfree, sc);
867 				bus_dmamap_sync(sc->sc_dmat, sc->rxq.jmap,
868 				    mtod(data->m, char *) - (char *)sc->rxq.jpool,
869 				    NFE_JBYTES, BUS_DMASYNC_POSTREAD);
870 
871 				physaddr = jbuf->physaddr;
872 			}
873 		} else {
874 			MCLGET(mnew, M_DONTWAIT);
875 			if ((mnew->m_flags & M_EXT) == 0) {
876 				m_freem(mnew);
877 				ifp->if_ierrors++;
878 				goto skip;
879 			}
880 
881 			bus_dmamap_sync(sc->sc_dmat, data->map, 0,
882 			    data->map->dm_mapsize, BUS_DMASYNC_POSTREAD);
883 			bus_dmamap_unload(sc->sc_dmat, data->map);
884 
885 			error = bus_dmamap_load(sc->sc_dmat, data->map,
886 			    mtod(mnew, void *), MCLBYTES, NULL,
887 			    BUS_DMA_READ | BUS_DMA_NOWAIT);
888 			if (error != 0) {
889 				m_freem(mnew);
890 
891 				/* try to reload the old mbuf */
892 				error = bus_dmamap_load(sc->sc_dmat, data->map,
893 				    mtod(data->m, void *), MCLBYTES, NULL,
894 				    BUS_DMA_READ | BUS_DMA_NOWAIT);
895 				if (error != 0) {
896 					/* very unlikely that it will fail.. */
897 					panic("%s: could not load old rx mbuf",
898 					    device_xname(sc->sc_dev));
899 				}
900 				ifp->if_ierrors++;
901 				goto skip;
902 			}
903 			physaddr = data->map->dm_segs[0].ds_addr;
904 		}
905 
906 		/*
907 		 * New mbuf successfully loaded, update Rx ring and continue
908 		 * processing.
909 		 */
910 		m = data->m;
911 		data->m = mnew;
912 
913 mbufcopied:
914 		/* finalize mbuf */
915 		m->m_pkthdr.len = m->m_len = len;
916 		m->m_pkthdr.rcvif = ifp;
917 
918 		if ((sc->sc_flags & NFE_HW_CSUM) != 0) {
919 			/*
920 			 * XXX
921 			 * no way to check M_CSUM_IPv4_BAD or non-IPv4 packets?
922 			 */
923 			if (flags & NFE_RX_IP_CSUMOK) {
924 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
925 				DPRINTFN(3, ("%s: ip4csum-rx ok\n",
926 				    device_xname(sc->sc_dev)));
927 			}
928 			/*
929 			 * XXX
930 			 * no way to check M_CSUM_TCP_UDP_BAD or
931 			 * other protocols?
932 			 */
933 			if (flags & NFE_RX_UDP_CSUMOK) {
934 				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
935 				DPRINTFN(3, ("%s: udp4csum-rx ok\n",
936 				    device_xname(sc->sc_dev)));
937 			} else if (flags & NFE_RX_TCP_CSUMOK) {
938 				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
939 				DPRINTFN(3, ("%s: tcp4csum-rx ok\n",
940 				    device_xname(sc->sc_dev)));
941 			}
942 		}
943 		bpf_mtap(ifp, m);
944 		ifp->if_ipackets++;
945 		(*ifp->if_input)(ifp, m);
946 
947 skip1:
948 		/* update mapping address in h/w descriptor */
949 		if (sc->sc_flags & NFE_40BIT_ADDR) {
950 #if defined(__LP64__)
951 			desc64->physaddr[0] = htole32(physaddr >> 32);
952 #endif
953 			desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
954 		} else {
955 			desc32->physaddr = htole32(physaddr);
956 		}
957 
958 skip:
959 		if (sc->sc_flags & NFE_40BIT_ADDR) {
960 			desc64->length = htole16(sc->rxq.bufsz);
961 			desc64->flags = htole16(NFE_RX_READY);
962 
963 			nfe_rxdesc64_sync(sc, desc64,
964 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
965 		} else {
966 			desc32->length = htole16(sc->rxq.bufsz);
967 			desc32->flags = htole16(NFE_RX_READY);
968 
969 			nfe_rxdesc32_sync(sc, desc32,
970 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
971 		}
972 	}
973 	/* update current RX pointer */
974 	sc->rxq.cur = i;
975 }
976 
977 void
978 nfe_txeof(struct nfe_softc *sc)
979 {
980 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
981 	struct nfe_desc32 *desc32;
982 	struct nfe_desc64 *desc64;
983 	struct nfe_tx_data *data = NULL;
984 	int i;
985 	uint16_t flags;
986 	char buf[128];
987 
988 	for (i = sc->txq.next;
989 	    sc->txq.queued > 0;
990 	    i = NFE_TX_NEXTDESC(i), sc->txq.queued--) {
991 		if (sc->sc_flags & NFE_40BIT_ADDR) {
992 			desc64 = &sc->txq.desc64[i];
993 			nfe_txdesc64_sync(sc, desc64,
994 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
995 
996 			flags = le16toh(desc64->flags);
997 		} else {
998 			desc32 = &sc->txq.desc32[i];
999 			nfe_txdesc32_sync(sc, desc32,
1000 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1001 
1002 			flags = le16toh(desc32->flags);
1003 		}
1004 
1005 		if ((flags & NFE_TX_VALID) != 0)
1006 			break;
1007 
1008 		data = &sc->txq.data[i];
1009 
1010 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1011 			if ((flags & NFE_TX_LASTFRAG_V1) == 0 &&
1012 			    data->m == NULL)
1013 				continue;
1014 
1015 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1016 				snprintb(buf, sizeof(buf), NFE_V1_TXERR, flags);
1017 				aprint_error_dev(sc->sc_dev, "tx v1 error %s\n",
1018 				    buf);
1019 				ifp->if_oerrors++;
1020 			} else
1021 				ifp->if_opackets++;
1022 		} else {
1023 			if ((flags & NFE_TX_LASTFRAG_V2) == 0 &&
1024 			    data->m == NULL)
1025 				continue;
1026 
1027 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1028 				snprintb(buf, sizeof(buf), NFE_V2_TXERR, flags);
1029 				aprint_error_dev(sc->sc_dev, "tx v2 error %s\n",
1030 				    buf);
1031 				ifp->if_oerrors++;
1032 			} else
1033 				ifp->if_opackets++;
1034 		}
1035 
1036 		if (data->m == NULL) {	/* should not get there */
1037 			aprint_error_dev(sc->sc_dev,
1038 			    "last fragment bit w/o associated mbuf!\n");
1039 			continue;
1040 		}
1041 
1042 		/* last fragment of the mbuf chain transmitted */
1043 		bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1044 		    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1045 		bus_dmamap_unload(sc->sc_dmat, data->active);
1046 		m_freem(data->m);
1047 		data->m = NULL;
1048 	}
1049 
1050 	sc->txq.next = i;
1051 
1052 	if (sc->txq.queued < NFE_TX_RING_COUNT) {
1053 		/* at least one slot freed */
1054 		ifp->if_flags &= ~IFF_OACTIVE;
1055 	}
1056 
1057 	if (sc->txq.queued == 0) {
1058 		/* all queued packets are sent */
1059 		ifp->if_timer = 0;
1060 	}
1061 }
1062 
1063 int
1064 nfe_encap(struct nfe_softc *sc, struct mbuf *m0)
1065 {
1066 	struct nfe_desc32 *desc32;
1067 	struct nfe_desc64 *desc64;
1068 	struct nfe_tx_data *data;
1069 	bus_dmamap_t map;
1070 	uint16_t flags, csumflags;
1071 #if NVLAN > 0
1072 	struct m_tag *mtag;
1073 	uint32_t vtag = 0;
1074 #endif
1075 	int error, i, first;
1076 
1077 	desc32 = NULL;
1078 	desc64 = NULL;
1079 	data = NULL;
1080 
1081 	flags = 0;
1082 	csumflags = 0;
1083 	first = sc->txq.cur;
1084 
1085 	map = sc->txq.data[first].map;
1086 
1087 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m0, BUS_DMA_NOWAIT);
1088 	if (error != 0) {
1089 		aprint_error_dev(sc->sc_dev, "could not map mbuf (error %d)\n",
1090 		    error);
1091 		return error;
1092 	}
1093 
1094 	if (sc->txq.queued + map->dm_nsegs >= NFE_TX_RING_COUNT - 1) {
1095 		bus_dmamap_unload(sc->sc_dmat, map);
1096 		return ENOBUFS;
1097 	}
1098 
1099 #if NVLAN > 0
1100 	/* setup h/w VLAN tagging */
1101 	if ((mtag = VLAN_OUTPUT_TAG(&sc->sc_ethercom, m0)) != NULL)
1102 		vtag = NFE_TX_VTAG | VLAN_TAG_VALUE(mtag);
1103 #endif
1104 	if ((sc->sc_flags & NFE_HW_CSUM) != 0) {
1105 		if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4)
1106 			csumflags |= NFE_TX_IP_CSUM;
1107 		if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4 | M_CSUM_UDPv4))
1108 			csumflags |= NFE_TX_TCP_UDP_CSUM;
1109 	}
1110 
1111 	for (i = 0; i < map->dm_nsegs; i++) {
1112 		data = &sc->txq.data[sc->txq.cur];
1113 
1114 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1115 			desc64 = &sc->txq.desc64[sc->txq.cur];
1116 #if defined(__LP64__)
1117 			desc64->physaddr[0] =
1118 			    htole32(map->dm_segs[i].ds_addr >> 32);
1119 #endif
1120 			desc64->physaddr[1] =
1121 			    htole32(map->dm_segs[i].ds_addr & 0xffffffff);
1122 			desc64->length = htole16(map->dm_segs[i].ds_len - 1);
1123 			desc64->flags = htole16(flags);
1124 			desc64->vtag = 0;
1125 		} else {
1126 			desc32 = &sc->txq.desc32[sc->txq.cur];
1127 
1128 			desc32->physaddr = htole32(map->dm_segs[i].ds_addr);
1129 			desc32->length = htole16(map->dm_segs[i].ds_len - 1);
1130 			desc32->flags = htole16(flags);
1131 		}
1132 
1133 		/*
1134 		 * Setting of the valid bit in the first descriptor is
1135 		 * deferred until the whole chain is fully setup.
1136 		 */
1137 		flags |= NFE_TX_VALID;
1138 
1139 		sc->txq.queued++;
1140 		sc->txq.cur = NFE_TX_NEXTDESC(sc->txq.cur);
1141 	}
1142 
1143 	/* the whole mbuf chain has been setup */
1144 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1145 		/* fix last descriptor */
1146 		flags |= NFE_TX_LASTFRAG_V2;
1147 		desc64->flags = htole16(flags);
1148 
1149 		/* Checksum flags and vtag belong to the first fragment only. */
1150 #if NVLAN > 0
1151 		sc->txq.desc64[first].vtag = htole32(vtag);
1152 #endif
1153 		sc->txq.desc64[first].flags |= htole16(csumflags);
1154 
1155 		/* finally, set the valid bit in the first descriptor */
1156 		sc->txq.desc64[first].flags |= htole16(NFE_TX_VALID);
1157 	} else {
1158 		/* fix last descriptor */
1159 		if (sc->sc_flags & NFE_JUMBO_SUP)
1160 			flags |= NFE_TX_LASTFRAG_V2;
1161 		else
1162 			flags |= NFE_TX_LASTFRAG_V1;
1163 		desc32->flags = htole16(flags);
1164 
1165 		/* Checksum flags belong to the first fragment only. */
1166 		sc->txq.desc32[first].flags |= htole16(csumflags);
1167 
1168 		/* finally, set the valid bit in the first descriptor */
1169 		sc->txq.desc32[first].flags |= htole16(NFE_TX_VALID);
1170 	}
1171 
1172 	data->m = m0;
1173 	data->active = map;
1174 
1175 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1176 	    BUS_DMASYNC_PREWRITE);
1177 
1178 	return 0;
1179 }
1180 
1181 void
1182 nfe_start(struct ifnet *ifp)
1183 {
1184 	struct nfe_softc *sc = ifp->if_softc;
1185 	int old = sc->txq.queued;
1186 	struct mbuf *m0;
1187 
1188 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
1189 		return;
1190 
1191 	for (;;) {
1192 		IFQ_POLL(&ifp->if_snd, m0);
1193 		if (m0 == NULL)
1194 			break;
1195 
1196 		if (nfe_encap(sc, m0) != 0) {
1197 			ifp->if_flags |= IFF_OACTIVE;
1198 			break;
1199 		}
1200 
1201 		/* packet put in h/w queue, remove from s/w queue */
1202 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1203 
1204 		bpf_mtap(ifp, m0);
1205 	}
1206 
1207 	if (sc->txq.queued != old) {
1208 		/* packets are queued */
1209 		if (sc->sc_flags & NFE_40BIT_ADDR)
1210 			nfe_txdesc64_rsync(sc, old, sc->txq.cur,
1211 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1212 		else
1213 			nfe_txdesc32_rsync(sc, old, sc->txq.cur,
1214 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1215 		/* kick Tx */
1216 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1217 
1218 		/*
1219 		 * Set a timeout in case the chip goes out to lunch.
1220 		 */
1221 		ifp->if_timer = 5;
1222 	}
1223 }
1224 
1225 void
1226 nfe_watchdog(struct ifnet *ifp)
1227 {
1228 	struct nfe_softc *sc = ifp->if_softc;
1229 
1230 	aprint_error_dev(sc->sc_dev, "watchdog timeout\n");
1231 
1232 	ifp->if_flags &= ~IFF_RUNNING;
1233 	nfe_init(ifp);
1234 
1235 	ifp->if_oerrors++;
1236 }
1237 
1238 int
1239 nfe_init(struct ifnet *ifp)
1240 {
1241 	struct nfe_softc *sc = ifp->if_softc;
1242 	uint32_t tmp;
1243 	int rc = 0, s;
1244 
1245 	if (ifp->if_flags & IFF_RUNNING)
1246 		return 0;
1247 
1248 	nfe_stop(ifp, 0);
1249 
1250 	NFE_WRITE(sc, NFE_TX_UNK, 0);
1251 	NFE_WRITE(sc, NFE_STATUS, 0);
1252 
1253 	sc->rxtxctl = NFE_RXTX_BIT2;
1254 	if (sc->sc_flags & NFE_40BIT_ADDR)
1255 		sc->rxtxctl |= NFE_RXTX_V3MAGIC;
1256 	else if (sc->sc_flags & NFE_JUMBO_SUP)
1257 		sc->rxtxctl |= NFE_RXTX_V2MAGIC;
1258 	if (sc->sc_flags & NFE_HW_CSUM)
1259 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1260 #if NVLAN > 0
1261 	/*
1262 	 * Although the adapter is capable of stripping VLAN tags from received
1263 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1264 	 * purpose.  This will be done in software by our network stack.
1265 	 */
1266 	if (sc->sc_flags & NFE_HW_VLAN)
1267 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1268 #endif
1269 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1270 	DELAY(10);
1271 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1272 
1273 #if NVLAN
1274 	if (sc->sc_flags & NFE_HW_VLAN)
1275 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1276 #endif
1277 
1278 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1279 
1280 	/* set MAC address */
1281 	nfe_set_macaddr(sc, sc->sc_enaddr);
1282 
1283 	/* tell MAC where rings are in memory */
1284 #ifdef __LP64__
1285 	NFE_WRITE(sc, NFE_RX_RING_ADDR_HI, sc->rxq.physaddr >> 32);
1286 #endif
1287 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, sc->rxq.physaddr & 0xffffffff);
1288 #ifdef __LP64__
1289 	NFE_WRITE(sc, NFE_TX_RING_ADDR_HI, sc->txq.physaddr >> 32);
1290 #endif
1291 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, sc->txq.physaddr & 0xffffffff);
1292 
1293 	NFE_WRITE(sc, NFE_RING_SIZE,
1294 	    (NFE_RX_RING_COUNT - 1) << 16 |
1295 	    (NFE_TX_RING_COUNT - 1));
1296 
1297 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1298 
1299 	/* force MAC to wakeup */
1300 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1301 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1302 	DELAY(10);
1303 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1304 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1305 
1306 	s = splnet();
1307 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1308 	nfe_intr(sc); /* XXX clear IRQ status registers */
1309 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1310 	splx(s);
1311 
1312 #if 1
1313 	/* configure interrupts coalescing/mitigation */
1314 	NFE_WRITE(sc, NFE_IMTIMER, NFE_IM_DEFAULT);
1315 #else
1316 	/* no interrupt mitigation: one interrupt per packet */
1317 	NFE_WRITE(sc, NFE_IMTIMER, 970);
1318 #endif
1319 
1320 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1321 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1322 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1323 
1324 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1325 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1326 
1327 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1328 	NFE_WRITE(sc, NFE_WOL_CTL, NFE_WOL_ENABLE);
1329 
1330 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1331 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1332 	DELAY(10);
1333 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1334 
1335 	/* set Rx filter */
1336 	nfe_setmulti(sc);
1337 
1338 	if ((rc = ether_mediachange(ifp)) != 0)
1339 		goto out;
1340 
1341 	nfe_tick(sc);
1342 
1343 	/* enable Rx */
1344 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1345 
1346 	/* enable Tx */
1347 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1348 
1349 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1350 
1351 	/* enable interrupts */
1352 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1353 
1354 	callout_schedule(&sc->sc_tick_ch, hz);
1355 
1356 	ifp->if_flags |= IFF_RUNNING;
1357 	ifp->if_flags &= ~IFF_OACTIVE;
1358 
1359 out:
1360 	return rc;
1361 }
1362 
1363 void
1364 nfe_stop(struct ifnet *ifp, int disable)
1365 {
1366 	struct nfe_softc *sc = ifp->if_softc;
1367 
1368 	callout_stop(&sc->sc_tick_ch);
1369 
1370 	ifp->if_timer = 0;
1371 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1372 
1373 	mii_down(&sc->sc_mii);
1374 
1375 	/* abort Tx */
1376 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1377 
1378 	/* disable Rx */
1379 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1380 
1381 	/* disable interrupts */
1382 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1383 
1384 	/* reset Tx and Rx rings */
1385 	nfe_reset_tx_ring(sc, &sc->txq);
1386 	nfe_reset_rx_ring(sc, &sc->rxq);
1387 }
1388 
1389 int
1390 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1391 {
1392 	struct nfe_desc32 *desc32;
1393 	struct nfe_desc64 *desc64;
1394 	struct nfe_rx_data *data;
1395 	struct nfe_jbuf *jbuf;
1396 	void **desc;
1397 	bus_addr_t physaddr;
1398 	int i, nsegs, error, descsize;
1399 
1400 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1401 		desc = (void **)&ring->desc64;
1402 		descsize = sizeof (struct nfe_desc64);
1403 	} else {
1404 		desc = (void **)&ring->desc32;
1405 		descsize = sizeof (struct nfe_desc32);
1406 	}
1407 
1408 	ring->cur = ring->next = 0;
1409 	ring->bufsz = MCLBYTES;
1410 
1411 	error = bus_dmamap_create(sc->sc_dmat, NFE_RX_RING_COUNT * descsize, 1,
1412 	    NFE_RX_RING_COUNT * descsize, 0, BUS_DMA_NOWAIT, &ring->map);
1413 	if (error != 0) {
1414 		aprint_error_dev(sc->sc_dev,
1415 		    "could not create desc DMA map\n");
1416 		ring->map = NULL;
1417 		goto fail;
1418 	}
1419 
1420 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_RX_RING_COUNT * descsize,
1421 	    PAGE_SIZE, 0, &ring->seg, 1, &nsegs, BUS_DMA_NOWAIT);
1422 	if (error != 0) {
1423 		aprint_error_dev(sc->sc_dev,
1424 		    "could not allocate DMA memory\n");
1425 		goto fail;
1426 	}
1427 
1428 	error = bus_dmamem_map(sc->sc_dmat, &ring->seg, nsegs,
1429 	    NFE_RX_RING_COUNT * descsize, (void **)desc, BUS_DMA_NOWAIT);
1430 	if (error != 0) {
1431 		aprint_error_dev(sc->sc_dev,
1432 		    "could not map desc DMA memory\n");
1433 		goto fail;
1434 	}
1435 
1436 	error = bus_dmamap_load(sc->sc_dmat, ring->map, *desc,
1437 	    NFE_RX_RING_COUNT * descsize, NULL, BUS_DMA_NOWAIT);
1438 	if (error != 0) {
1439 		aprint_error_dev(sc->sc_dev, "could not load desc DMA map\n");
1440 		goto fail;
1441 	}
1442 
1443 	memset(*desc, 0, NFE_RX_RING_COUNT * descsize);
1444 	ring->physaddr = ring->map->dm_segs[0].ds_addr;
1445 
1446 	if (sc->sc_flags & NFE_USE_JUMBO) {
1447 		ring->bufsz = NFE_JBYTES;
1448 		if ((error = nfe_jpool_alloc(sc)) != 0) {
1449 			aprint_error_dev(sc->sc_dev,
1450 			    "could not allocate jumbo frames\n");
1451 			goto fail;
1452 		}
1453 	}
1454 
1455 	/*
1456 	 * Pre-allocate Rx buffers and populate Rx ring.
1457 	 */
1458 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1459 		data = &sc->rxq.data[i];
1460 
1461 		MGETHDR(data->m, M_DONTWAIT, MT_DATA);
1462 		if (data->m == NULL) {
1463 			aprint_error_dev(sc->sc_dev,
1464 			    "could not allocate rx mbuf\n");
1465 			error = ENOMEM;
1466 			goto fail;
1467 		}
1468 
1469 		if (sc->sc_flags & NFE_USE_JUMBO) {
1470 			if ((jbuf = nfe_jalloc(sc, i)) == NULL) {
1471 				aprint_error_dev(sc->sc_dev,
1472 				    "could not allocate jumbo buffer\n");
1473 				goto fail;
1474 			}
1475 			MEXTADD(data->m, jbuf->buf, NFE_JBYTES, 0, nfe_jfree,
1476 			    sc);
1477 
1478 			physaddr = jbuf->physaddr;
1479 		} else {
1480 			error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
1481 			    MCLBYTES, 0, BUS_DMA_NOWAIT, &data->map);
1482 			if (error != 0) {
1483 				aprint_error_dev(sc->sc_dev,
1484 				    "could not create DMA map\n");
1485 				data->map = NULL;
1486 				goto fail;
1487 			}
1488 			MCLGET(data->m, M_DONTWAIT);
1489 			if (!(data->m->m_flags & M_EXT)) {
1490 				aprint_error_dev(sc->sc_dev,
1491 				    "could not allocate mbuf cluster\n");
1492 				error = ENOMEM;
1493 				goto fail;
1494 			}
1495 
1496 			error = bus_dmamap_load(sc->sc_dmat, data->map,
1497 			    mtod(data->m, void *), MCLBYTES, NULL,
1498 			    BUS_DMA_READ | BUS_DMA_NOWAIT);
1499 			if (error != 0) {
1500 				aprint_error_dev(sc->sc_dev,
1501 				    "could not load rx buf DMA map");
1502 				goto fail;
1503 			}
1504 			physaddr = data->map->dm_segs[0].ds_addr;
1505 		}
1506 
1507 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1508 			desc64 = &sc->rxq.desc64[i];
1509 #if defined(__LP64__)
1510 			desc64->physaddr[0] = htole32(physaddr >> 32);
1511 #endif
1512 			desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
1513 			desc64->length = htole16(sc->rxq.bufsz);
1514 			desc64->flags = htole16(NFE_RX_READY);
1515 		} else {
1516 			desc32 = &sc->rxq.desc32[i];
1517 			desc32->physaddr = htole32(physaddr);
1518 			desc32->length = htole16(sc->rxq.bufsz);
1519 			desc32->flags = htole16(NFE_RX_READY);
1520 		}
1521 	}
1522 
1523 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1524 	    BUS_DMASYNC_PREWRITE);
1525 
1526 	return 0;
1527 
1528 fail:	nfe_free_rx_ring(sc, ring);
1529 	return error;
1530 }
1531 
1532 void
1533 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1534 {
1535 	int i;
1536 
1537 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1538 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1539 			ring->desc64[i].length = htole16(ring->bufsz);
1540 			ring->desc64[i].flags = htole16(NFE_RX_READY);
1541 		} else {
1542 			ring->desc32[i].length = htole16(ring->bufsz);
1543 			ring->desc32[i].flags = htole16(NFE_RX_READY);
1544 		}
1545 	}
1546 
1547 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1548 	    BUS_DMASYNC_PREWRITE);
1549 
1550 	ring->cur = ring->next = 0;
1551 }
1552 
1553 void
1554 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1555 {
1556 	struct nfe_rx_data *data;
1557 	void *desc;
1558 	int i, descsize;
1559 
1560 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1561 		desc = ring->desc64;
1562 		descsize = sizeof (struct nfe_desc64);
1563 	} else {
1564 		desc = ring->desc32;
1565 		descsize = sizeof (struct nfe_desc32);
1566 	}
1567 
1568 	if (desc != NULL) {
1569 		bus_dmamap_sync(sc->sc_dmat, ring->map, 0,
1570 		    ring->map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1571 		bus_dmamap_unload(sc->sc_dmat, ring->map);
1572 		bus_dmamem_unmap(sc->sc_dmat, (void *)desc,
1573 		    NFE_RX_RING_COUNT * descsize);
1574 		bus_dmamem_free(sc->sc_dmat, &ring->seg, 1);
1575 	}
1576 
1577 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1578 		data = &ring->data[i];
1579 
1580 		if (data->map != NULL) {
1581 			bus_dmamap_sync(sc->sc_dmat, data->map, 0,
1582 			    data->map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1583 			bus_dmamap_unload(sc->sc_dmat, data->map);
1584 			bus_dmamap_destroy(sc->sc_dmat, data->map);
1585 		}
1586 		if (data->m != NULL)
1587 			m_freem(data->m);
1588 	}
1589 
1590 	nfe_jpool_free(sc);
1591 }
1592 
1593 struct nfe_jbuf *
1594 nfe_jalloc(struct nfe_softc *sc, int i)
1595 {
1596 	struct nfe_jbuf *jbuf;
1597 
1598 	mutex_enter(&sc->rxq.mtx);
1599 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1600 	if (jbuf != NULL)
1601 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1602 	mutex_exit(&sc->rxq.mtx);
1603 	if (jbuf == NULL)
1604 		return NULL;
1605 	sc->rxq.jbufmap[i] =
1606 	    ((char *)jbuf->buf - (char *)sc->rxq.jpool) / NFE_JBYTES;
1607 	return jbuf;
1608 }
1609 
1610 /*
1611  * This is called automatically by the network stack when the mbuf is freed.
1612  * Caution must be taken that the NIC might be reset by the time the mbuf is
1613  * freed.
1614  */
1615 void
1616 nfe_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1617 {
1618 	struct nfe_softc *sc = arg;
1619 	struct nfe_jbuf *jbuf;
1620 	int i;
1621 
1622 	/* find the jbuf from the base pointer */
1623 	i = ((char *)buf - (char *)sc->rxq.jpool) / NFE_JBYTES;
1624 	if (i < 0 || i >= NFE_JPOOL_COUNT) {
1625 		aprint_error_dev(sc->sc_dev,
1626 		    "request to free a buffer (%p) not managed by us\n", buf);
1627 		return;
1628 	}
1629 	jbuf = &sc->rxq.jbuf[i];
1630 
1631 	/* ..and put it back in the free list */
1632 	mutex_enter(&sc->rxq.mtx);
1633 	SLIST_INSERT_HEAD(&sc->rxq.jfreelist, jbuf, jnext);
1634 	mutex_exit(&sc->rxq.mtx);
1635 
1636 	if (m != NULL)
1637 		pool_cache_put(mb_cache, m);
1638 }
1639 
1640 int
1641 nfe_jpool_alloc(struct nfe_softc *sc)
1642 {
1643 	struct nfe_rx_ring *ring = &sc->rxq;
1644 	struct nfe_jbuf *jbuf;
1645 	bus_addr_t physaddr;
1646 	char *buf;
1647 	int i, nsegs, error;
1648 
1649 	/*
1650 	 * Allocate a big chunk of DMA'able memory.
1651 	 */
1652 	error = bus_dmamap_create(sc->sc_dmat, NFE_JPOOL_SIZE, 1,
1653 	    NFE_JPOOL_SIZE, 0, BUS_DMA_NOWAIT, &ring->jmap);
1654 	if (error != 0) {
1655 		aprint_error_dev(sc->sc_dev,
1656 		    "could not create jumbo DMA map\n");
1657 		ring->jmap = NULL;
1658 		goto fail;
1659 	}
1660 
1661 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_JPOOL_SIZE, PAGE_SIZE, 0,
1662 	    &ring->jseg, 1, &nsegs, BUS_DMA_NOWAIT);
1663 	if (error != 0) {
1664 		aprint_error_dev(sc->sc_dev,
1665 		    "could not allocate jumbo DMA memory\n");
1666 		goto fail;
1667 	}
1668 
1669 	error = bus_dmamem_map(sc->sc_dmat, &ring->jseg, nsegs, NFE_JPOOL_SIZE,
1670 	    &ring->jpool, BUS_DMA_NOWAIT);
1671 	if (error != 0) {
1672 		aprint_error_dev(sc->sc_dev,
1673 		    "could not map jumbo DMA memory\n");
1674 		goto fail;
1675 	}
1676 
1677 	error = bus_dmamap_load(sc->sc_dmat, ring->jmap, ring->jpool,
1678 	    NFE_JPOOL_SIZE, NULL, BUS_DMA_READ | BUS_DMA_NOWAIT);
1679 	if (error != 0) {
1680 		aprint_error_dev(sc->sc_dev,
1681 		    "could not load jumbo DMA map\n");
1682 		goto fail;
1683 	}
1684 
1685 	/* ..and split it into 9KB chunks */
1686 	SLIST_INIT(&ring->jfreelist);
1687 
1688 	buf = ring->jpool;
1689 	physaddr = ring->jmap->dm_segs[0].ds_addr;
1690 	for (i = 0; i < NFE_JPOOL_COUNT; i++) {
1691 		jbuf = &ring->jbuf[i];
1692 
1693 		jbuf->buf = buf;
1694 		jbuf->physaddr = physaddr;
1695 
1696 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1697 
1698 		buf += NFE_JBYTES;
1699 		physaddr += NFE_JBYTES;
1700 	}
1701 
1702 	return 0;
1703 
1704 fail:	nfe_jpool_free(sc);
1705 	return error;
1706 }
1707 
1708 void
1709 nfe_jpool_free(struct nfe_softc *sc)
1710 {
1711 	struct nfe_rx_ring *ring = &sc->rxq;
1712 
1713 	if (ring->jmap != NULL) {
1714 		bus_dmamap_sync(sc->sc_dmat, ring->jmap, 0,
1715 		    ring->jmap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1716 		bus_dmamap_unload(sc->sc_dmat, ring->jmap);
1717 		bus_dmamap_destroy(sc->sc_dmat, ring->jmap);
1718 		ring->jmap = NULL;
1719 	}
1720 	if (ring->jpool != NULL) {
1721 		bus_dmamem_unmap(sc->sc_dmat, ring->jpool, NFE_JPOOL_SIZE);
1722 		bus_dmamem_free(sc->sc_dmat, &ring->jseg, 1);
1723 		ring->jpool = NULL;
1724 	}
1725 }
1726 
1727 int
1728 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1729 {
1730 	int i, nsegs, error;
1731 	void **desc;
1732 	int descsize;
1733 
1734 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1735 		desc = (void **)&ring->desc64;
1736 		descsize = sizeof (struct nfe_desc64);
1737 	} else {
1738 		desc = (void **)&ring->desc32;
1739 		descsize = sizeof (struct nfe_desc32);
1740 	}
1741 
1742 	ring->queued = 0;
1743 	ring->cur = ring->next = 0;
1744 
1745 	error = bus_dmamap_create(sc->sc_dmat, NFE_TX_RING_COUNT * descsize, 1,
1746 	    NFE_TX_RING_COUNT * descsize, 0, BUS_DMA_NOWAIT, &ring->map);
1747 
1748 	if (error != 0) {
1749 		aprint_error_dev(sc->sc_dev,
1750 		    "could not create desc DMA map\n");
1751 		ring->map = NULL;
1752 		goto fail;
1753 	}
1754 
1755 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_TX_RING_COUNT * descsize,
1756 	    PAGE_SIZE, 0, &ring->seg, 1, &nsegs, BUS_DMA_NOWAIT);
1757 	if (error != 0) {
1758 		aprint_error_dev(sc->sc_dev,
1759 		    "could not allocate DMA memory\n");
1760 		goto fail;
1761 	}
1762 
1763 	error = bus_dmamem_map(sc->sc_dmat, &ring->seg, nsegs,
1764 	    NFE_TX_RING_COUNT * descsize, (void **)desc, BUS_DMA_NOWAIT);
1765 	if (error != 0) {
1766 		aprint_error_dev(sc->sc_dev,
1767 		    "could not map desc DMA memory\n");
1768 		goto fail;
1769 	}
1770 
1771 	error = bus_dmamap_load(sc->sc_dmat, ring->map, *desc,
1772 	    NFE_TX_RING_COUNT * descsize, NULL, BUS_DMA_NOWAIT);
1773 	if (error != 0) {
1774 		aprint_error_dev(sc->sc_dev, "could not load desc DMA map\n");
1775 		goto fail;
1776 	}
1777 
1778 	memset(*desc, 0, NFE_TX_RING_COUNT * descsize);
1779 	ring->physaddr = ring->map->dm_segs[0].ds_addr;
1780 
1781 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1782 		error = bus_dmamap_create(sc->sc_dmat, NFE_JBYTES,
1783 		    NFE_MAX_SCATTER, NFE_JBYTES, 0, BUS_DMA_NOWAIT,
1784 		    &ring->data[i].map);
1785 		if (error != 0) {
1786 			aprint_error_dev(sc->sc_dev,
1787 			    "could not create DMA map\n");
1788 			ring->data[i].map = NULL;
1789 			goto fail;
1790 		}
1791 	}
1792 
1793 	return 0;
1794 
1795 fail:	nfe_free_tx_ring(sc, ring);
1796 	return error;
1797 }
1798 
1799 void
1800 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1801 {
1802 	struct nfe_tx_data *data;
1803 	int i;
1804 
1805 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1806 		if (sc->sc_flags & NFE_40BIT_ADDR)
1807 			ring->desc64[i].flags = 0;
1808 		else
1809 			ring->desc32[i].flags = 0;
1810 
1811 		data = &ring->data[i];
1812 
1813 		if (data->m != NULL) {
1814 			bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1815 			    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1816 			bus_dmamap_unload(sc->sc_dmat, data->active);
1817 			m_freem(data->m);
1818 			data->m = NULL;
1819 		}
1820 	}
1821 
1822 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1823 	    BUS_DMASYNC_PREWRITE);
1824 
1825 	ring->queued = 0;
1826 	ring->cur = ring->next = 0;
1827 }
1828 
1829 void
1830 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1831 {
1832 	struct nfe_tx_data *data;
1833 	void *desc;
1834 	int i, descsize;
1835 
1836 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1837 		desc = ring->desc64;
1838 		descsize = sizeof (struct nfe_desc64);
1839 	} else {
1840 		desc = ring->desc32;
1841 		descsize = sizeof (struct nfe_desc32);
1842 	}
1843 
1844 	if (desc != NULL) {
1845 		bus_dmamap_sync(sc->sc_dmat, ring->map, 0,
1846 		    ring->map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1847 		bus_dmamap_unload(sc->sc_dmat, ring->map);
1848 		bus_dmamem_unmap(sc->sc_dmat, (void *)desc,
1849 		    NFE_TX_RING_COUNT * descsize);
1850 		bus_dmamem_free(sc->sc_dmat, &ring->seg, 1);
1851 	}
1852 
1853 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1854 		data = &ring->data[i];
1855 
1856 		if (data->m != NULL) {
1857 			bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1858 			    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1859 			bus_dmamap_unload(sc->sc_dmat, data->active);
1860 			m_freem(data->m);
1861 		}
1862 	}
1863 
1864 	/* ..and now actually destroy the DMA mappings */
1865 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1866 		data = &ring->data[i];
1867 		if (data->map == NULL)
1868 			continue;
1869 		bus_dmamap_destroy(sc->sc_dmat, data->map);
1870 	}
1871 }
1872 
1873 void
1874 nfe_setmulti(struct nfe_softc *sc)
1875 {
1876 	struct ethercom *ec = &sc->sc_ethercom;
1877 	struct ifnet *ifp = &ec->ec_if;
1878 	struct ether_multi *enm;
1879 	struct ether_multistep step;
1880 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
1881 	uint32_t filter = NFE_RXFILTER_MAGIC;
1882 	int i;
1883 
1884 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1885 		memset(addr, 0, ETHER_ADDR_LEN);
1886 		memset(mask, 0, ETHER_ADDR_LEN);
1887 		goto done;
1888 	}
1889 
1890 	memcpy(addr, etherbroadcastaddr, ETHER_ADDR_LEN);
1891 	memcpy(mask, etherbroadcastaddr, ETHER_ADDR_LEN);
1892 
1893 	ETHER_FIRST_MULTI(step, ec, enm);
1894 	while (enm != NULL) {
1895 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1896 			ifp->if_flags |= IFF_ALLMULTI;
1897 			memset(addr, 0, ETHER_ADDR_LEN);
1898 			memset(mask, 0, ETHER_ADDR_LEN);
1899 			goto done;
1900 		}
1901 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
1902 			addr[i] &=  enm->enm_addrlo[i];
1903 			mask[i] &= ~enm->enm_addrlo[i];
1904 		}
1905 		ETHER_NEXT_MULTI(step, enm);
1906 	}
1907 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1908 		mask[i] |= addr[i];
1909 
1910 done:
1911 	addr[0] |= 0x01;	/* make sure multicast bit is set */
1912 
1913 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
1914 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1915 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
1916 	    addr[5] <<  8 | addr[4]);
1917 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
1918 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
1919 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
1920 	    mask[5] <<  8 | mask[4]);
1921 
1922 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
1923 	NFE_WRITE(sc, NFE_RXFILTER, filter);
1924 }
1925 
1926 void
1927 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
1928 {
1929 	uint32_t tmp;
1930 
1931 	if ((sc->sc_flags & NFE_CORRECT_MACADDR) != 0) {
1932 		tmp = NFE_READ(sc, NFE_MACADDR_HI);
1933 		addr[0] = (tmp & 0xff);
1934 		addr[1] = (tmp >>  8) & 0xff;
1935 		addr[2] = (tmp >> 16) & 0xff;
1936 		addr[3] = (tmp >> 24) & 0xff;
1937 
1938 		tmp = NFE_READ(sc, NFE_MACADDR_LO);
1939 		addr[4] = (tmp & 0xff);
1940 		addr[5] = (tmp >> 8) & 0xff;
1941 
1942 	} else {
1943 		tmp = NFE_READ(sc, NFE_MACADDR_LO);
1944 		addr[0] = (tmp >> 8) & 0xff;
1945 		addr[1] = (tmp & 0xff);
1946 
1947 		tmp = NFE_READ(sc, NFE_MACADDR_HI);
1948 		addr[2] = (tmp >> 24) & 0xff;
1949 		addr[3] = (tmp >> 16) & 0xff;
1950 		addr[4] = (tmp >>  8) & 0xff;
1951 		addr[5] = (tmp & 0xff);
1952 	}
1953 }
1954 
1955 void
1956 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
1957 {
1958 	NFE_WRITE(sc, NFE_MACADDR_LO,
1959 	    addr[5] <<  8 | addr[4]);
1960 	NFE_WRITE(sc, NFE_MACADDR_HI,
1961 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1962 }
1963 
1964 void
1965 nfe_tick(void *arg)
1966 {
1967 	struct nfe_softc *sc = arg;
1968 	int s;
1969 
1970 	s = splnet();
1971 	mii_tick(&sc->sc_mii);
1972 	splx(s);
1973 
1974 	callout_schedule(&sc->sc_tick_ch, hz);
1975 }
1976 
1977 void
1978 nfe_poweron(device_t self)
1979 {
1980 	struct nfe_softc *sc = device_private(self);
1981 
1982 	if ((sc->sc_flags & NFE_PWR_MGMT) != 0) {
1983 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | NFE_RXTX_BIT2);
1984 		NFE_WRITE(sc, NFE_MAC_RESET, NFE_MAC_RESET_MAGIC);
1985 		DELAY(100);
1986 		NFE_WRITE(sc, NFE_MAC_RESET, 0);
1987 		DELAY(100);
1988 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT2);
1989 		NFE_WRITE(sc, NFE_PWR2_CTL,
1990 		    NFE_READ(sc, NFE_PWR2_CTL) & ~NFE_PWR2_WAKEUP_MASK);
1991 	}
1992 }
1993 
1994 bool
1995 nfe_resume(device_t dv, const pmf_qual_t *qual)
1996 {
1997 	nfe_poweron(dv);
1998 
1999 	return true;
2000 }
2001