xref: /netbsd-src/sys/dev/pci/if_kse.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: if_kse.c,v 1.11 2007/11/16 10:31:57 nisimura Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 Tohru Nishimura
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Tohru Nishimura.
17  * 4. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: if_kse.c,v 1.11 2007/11/16 10:31:57 nisimura Exp $");
34 
35 #include "bpfilter.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/callout.h>
40 #include <sys/mbuf.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
43 #include <sys/ioctl.h>
44 #include <sys/errno.h>
45 #include <sys/device.h>
46 #include <sys/queue.h>
47 
48 #include <machine/endian.h>
49 #include <sys/bus.h>
50 #include <sys/intr.h>
51 
52 #include <net/if.h>
53 #include <net/if_media.h>
54 #include <net/if_dl.h>
55 #include <net/if_ether.h>
56 
57 #if NBPFILTER > 0
58 #include <net/bpf.h>
59 #endif
60 
61 #include <dev/pci/pcivar.h>
62 #include <dev/pci/pcireg.h>
63 #include <dev/pci/pcidevs.h>
64 
65 #define CSR_READ_4(sc, off) \
66 	    bus_space_read_4(sc->sc_st, sc->sc_sh, off)
67 #define CSR_WRITE_4(sc, off, val) \
68 	    bus_space_write_4(sc->sc_st, sc->sc_sh, off, val)
69 #define CSR_READ_2(sc, off) \
70 	    bus_space_read_2(sc->sc_st, sc->sc_sh, off)
71 #define CSR_WRITE_2(sc, off, val) \
72 	    bus_space_write_2(sc->sc_st, sc->sc_sh, off, val)
73 
74 #define MDTXC	0x000	/* DMA transmit control */
75 #define MDRXC	0x004	/* DMA receive control */
76 #define MDTSC	0x008	/* DMA transmit start */
77 #define MDRSC	0x00c	/* DMA receive start */
78 #define TDLB	0x010	/* transmit descriptor list base */
79 #define RDLB	0x014	/* receive descriptor list base */
80 #define MTR0	0x020	/* multicast table 31:0 */
81 #define MTR1	0x024	/* multicast table 63:32 */
82 #define INTEN	0x028	/* interrupt enable */
83 #define INTST	0x02c	/* interrupt status */
84 #define MARL	0x200	/* MAC address low */
85 #define MARM	0x202	/* MAC address middle */
86 #define MARH	0x204	/* MAC address high */
87 #define GRR	0x216	/* global reset */
88 #define CIDR	0x400	/* chip ID and enable */
89 #define CGCR	0x40a	/* chip global control */
90 #define IACR	0x4a0	/* indirect access control */
91 #define IADR1	0x4a2	/* indirect access data 66:63 */
92 #define IADR2	0x4a4	/* indirect access data 47:32 */
93 #define IADR3	0x4a6	/* indirect access data 63:48 */
94 #define IADR4	0x4a8	/* indirect access data 15:0 */
95 #define IADR5	0x4aa	/* indirect access data 31:16 */
96 #define P1CR4	0x512	/* port 1 control 4 */
97 #define P1SR	0x514	/* port 1 status */
98 #define P2CR4	0x532	/* port 2 control 4 */
99 #define P2SR	0x534	/* port 2 status */
100 
101 #define TXC_BS_MSK	0x3f000000	/* burst size */
102 #define TXC_BS_SFT	(24)		/* 1,2,4,8,16,32 or 0 for unlimited */
103 #define TXC_UCG		(1U<<18)	/* generate UDP checksum */
104 #define TXC_TCG		(1U<<17)	/* generate TCP checksum */
105 #define TXC_ICG		(1U<<16)	/* generate IP checksum */
106 #define TXC_FCE		(1U<<9)		/* enable flowcontrol */
107 #define TXC_EP		(1U<<2)		/* enable automatic padding */
108 #define TXC_AC		(1U<<1)		/* add CRC to frame */
109 #define TXC_TEN		(1)		/* enable DMA to run */
110 
111 #define RXC_BS_MSK	0x3f000000	/* burst size */
112 #define RXC_BS_SFT	(24)		/* 1,2,4,8,16,32 or 0 for unlimited */
113 #define RXC_IHAE	(1U<<19)	/* IP header alignment enable */
114 #define RXC_UCC		(1U<<18)	/* run UDP checksum */
115 #define RXC_TCC		(1U<<17)	/* run TDP checksum */
116 #define RXC_ICC		(1U<<16)	/* run IP checksum */
117 #define RXC_FCE		(1U<<9)		/* enable flowcontrol */
118 #define RXC_RB		(1U<<6)		/* receive broadcast frame */
119 #define RXC_RM		(1U<<5)		/* receive multicast frame */
120 #define RXC_RU		(1U<<4)		/* receive unicast frame */
121 #define RXC_RE		(1U<<3)		/* accept error frame */
122 #define RXC_RA		(1U<<2)		/* receive all frame */
123 #define RXC_MHTE	(1U<<1)		/* use multicast hash table */
124 #define RXC_REN		(1)		/* enable DMA to run */
125 
126 #define INT_DMLCS	(1U<<31)	/* link status change */
127 #define INT_DMTS	(1U<<30)	/* sending desc. has posted Tx done */
128 #define INT_DMRS	(1U<<29)	/* frame was received */
129 #define INT_DMRBUS	(1U<<27)	/* Rx descriptor pool is full */
130 
131 #define T0_OWN		(1U<<31)	/* desc is ready to Tx */
132 
133 #define R0_OWN		(1U<<31)	/* desc is empty */
134 #define R0_FS		(1U<<30)	/* first segment of frame */
135 #define R0_LS		(1U<<29)	/* last segment of frame */
136 #define R0_IPE		(1U<<28)	/* IP checksum error */
137 #define R0_TCPE		(1U<<27)	/* TCP checksum error */
138 #define R0_UDPE		(1U<<26)	/* UDP checksum error */
139 #define R0_ES		(1U<<25)	/* error summary */
140 #define R0_MF		(1U<<24)	/* multicast frame */
141 #define R0_SPN		0x00300000	/* 21:20 switch port 1/2 */
142 #define R0_ALIGN	0x00300000	/* 21:20 (KSZ8692P) Rx align amount */
143 #define R0_RE		(1U<<19)	/* MII reported error */
144 #define R0_TL		(1U<<18)	/* frame too long, beyond 1518 */
145 #define R0_RF		(1U<<17)	/* damaged runt frame */
146 #define R0_CE		(1U<<16)	/* CRC error */
147 #define R0_FT		(1U<<15)	/* frame type */
148 #define R0_FL_MASK	0x7ff		/* frame length 10:0 */
149 
150 #define T1_IC		(1U<<31)	/* post interrupt on complete */
151 #define T1_FS		(1U<<30)	/* first segment of frame */
152 #define T1_LS		(1U<<29)	/* last segment of frame */
153 #define T1_IPCKG	(1U<<28)	/* generate IP checksum */
154 #define T1_TCPCKG	(1U<<27)	/* generate TCP checksum */
155 #define T1_UDPCKG	(1U<<26)	/* generate UDP checksum */
156 #define T1_TER		(1U<<25)	/* end of ring */
157 #define T1_SPN		0x00300000	/* 21:20 switch port 1/2 */
158 #define T1_TBS_MASK	0x7ff		/* segment size 10:0 */
159 
160 #define R1_RER		(1U<<25)	/* end of ring */
161 #define R1_RBS_MASK	0x7fc		/* segment size 10:0 */
162 
163 #define KSE_NTXSEGS		16
164 #define KSE_TXQUEUELEN		64
165 #define KSE_TXQUEUELEN_MASK	(KSE_TXQUEUELEN - 1)
166 #define KSE_TXQUEUE_GC		(KSE_TXQUEUELEN / 4)
167 #define KSE_NTXDESC		256
168 #define KSE_NTXDESC_MASK	(KSE_NTXDESC - 1)
169 #define KSE_NEXTTX(x)		(((x) + 1) & KSE_NTXDESC_MASK)
170 #define KSE_NEXTTXS(x)		(((x) + 1) & KSE_TXQUEUELEN_MASK)
171 
172 #define KSE_NRXDESC		64
173 #define KSE_NRXDESC_MASK	(KSE_NRXDESC - 1)
174 #define KSE_NEXTRX(x)		(((x) + 1) & KSE_NRXDESC_MASK)
175 
176 struct tdes {
177 	uint32_t t0, t1, t2, t3;
178 };
179 
180 struct rdes {
181 	uint32_t r0, r1, r2, r3;
182 };
183 
184 struct kse_control_data {
185 	struct tdes kcd_txdescs[KSE_NTXDESC];
186 	struct rdes kcd_rxdescs[KSE_NRXDESC];
187 };
188 #define KSE_CDOFF(x)		offsetof(struct kse_control_data, x)
189 #define KSE_CDTXOFF(x)		KSE_CDOFF(kcd_txdescs[(x)])
190 #define KSE_CDRXOFF(x)		KSE_CDOFF(kcd_rxdescs[(x)])
191 
192 struct kse_txsoft {
193 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
194 	bus_dmamap_t txs_dmamap;	/* our DMA map */
195 	int txs_firstdesc;		/* first descriptor in packet */
196 	int txs_lastdesc;		/* last descriptor in packet */
197 	int txs_ndesc;			/* # of descriptors used */
198 };
199 
200 struct kse_rxsoft {
201 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
202 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
203 };
204 
205 struct kse_softc {
206 	struct device sc_dev;		/* generic device information */
207 	bus_space_tag_t sc_st;		/* bus space tag */
208 	bus_space_handle_t sc_sh;	/* bus space handle */
209 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
210 	struct ethercom sc_ethercom;	/* Ethernet common data */
211 	void *sc_ih;			/* interrupt cookie */
212 
213 	struct ifmedia sc_media;	/* ifmedia information */
214 	int sc_media_status;		/* PHY */
215 	int sc_media_active;		/* PHY */
216 	callout_t  sc_callout;		/* MII tick callout */
217 	callout_t  sc_stat_ch;		/* statistics counter callout */
218 
219 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
220 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
221 
222 	struct kse_control_data *sc_control_data;
223 #define sc_txdescs	sc_control_data->kcd_txdescs
224 #define sc_rxdescs	sc_control_data->kcd_rxdescs
225 
226 	struct kse_txsoft sc_txsoft[KSE_TXQUEUELEN];
227 	struct kse_rxsoft sc_rxsoft[KSE_NRXDESC];
228 	int sc_txfree;			/* number of free Tx descriptors */
229 	int sc_txnext;			/* next ready Tx descriptor */
230 	int sc_txsfree;			/* number of free Tx jobs */
231 	int sc_txsnext;			/* next ready Tx job */
232 	int sc_txsdirty;		/* dirty Tx jobs */
233 	int sc_rxptr;			/* next ready Rx descriptor/descsoft */
234 
235 	uint32_t sc_txc, sc_rxc;
236 	uint32_t sc_t1csum;
237 	int sc_mcsum;
238 	uint32_t sc_inten;
239 
240 	uint32_t sc_chip;
241 	uint8_t sc_altmac[16][ETHER_ADDR_LEN];
242 	uint16_t sc_vlan[16];
243 
244 #ifdef KSE_EVENT_COUNTERS
245 	struct ksext {
246 		char evcntname[3][8];
247 		struct evcnt pev[3][34];
248 	} sc_ext;			/* switch statistics */
249 #endif
250 };
251 
252 #define KSE_CDTXADDR(sc, x)	((sc)->sc_cddma + KSE_CDTXOFF((x)))
253 #define KSE_CDRXADDR(sc, x)	((sc)->sc_cddma + KSE_CDRXOFF((x)))
254 
255 #define KSE_CDTXSYNC(sc, x, n, ops)					\
256 do {									\
257 	int __x, __n;							\
258 									\
259 	__x = (x);							\
260 	__n = (n);							\
261 									\
262 	/* If it will wrap around, sync to the end of the ring. */	\
263 	if ((__x + __n) > KSE_NTXDESC) {				\
264 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
265 		    KSE_CDTXOFF(__x), sizeof(struct tdes) *		\
266 		    (KSE_NTXDESC - __x), (ops));			\
267 		__n -= (KSE_NTXDESC - __x);				\
268 		__x = 0;						\
269 	}								\
270 									\
271 	/* Now sync whatever is left. */				\
272 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
273 	    KSE_CDTXOFF(__x), sizeof(struct tdes) * __n, (ops));	\
274 } while (/*CONSTCOND*/0)
275 
276 #define KSE_CDRXSYNC(sc, x, ops)					\
277 do {									\
278 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
279 	    KSE_CDRXOFF((x)), sizeof(struct rdes), (ops));		\
280 } while (/*CONSTCOND*/0)
281 
282 #define KSE_INIT_RXDESC(sc, x)						\
283 do {									\
284 	struct kse_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
285 	struct rdes *__rxd = &(sc)->sc_rxdescs[(x)];			\
286 	struct mbuf *__m = __rxs->rxs_mbuf;				\
287 									\
288 	__m->m_data = __m->m_ext.ext_buf;				\
289 	__rxd->r2 = __rxs->rxs_dmamap->dm_segs[0].ds_addr;		\
290 	__rxd->r1 = R1_RBS_MASK /* __m->m_ext.ext_size */;		\
291 	__rxd->r0 = R0_OWN;						\
292 	KSE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
293 } while (/*CONSTCOND*/0)
294 
295 u_int kse_burstsize = 8;	/* DMA burst length tuning knob */
296 
297 #ifdef KSEDIAGNOSTIC
298 u_int kse_monitor_rxintr;	/* fragmented UDP csum HW bug hook */
299 #endif
300 
301 static int kse_match(struct device *, struct cfdata *, void *);
302 static void kse_attach(struct device *, struct device *, void *);
303 
304 CFATTACH_DECL(kse, sizeof(struct kse_softc),
305     kse_match, kse_attach, NULL, NULL);
306 
307 static int kse_ioctl(struct ifnet *, u_long, void *);
308 static void kse_start(struct ifnet *);
309 static void kse_watchdog(struct ifnet *);
310 static int kse_init(struct ifnet *);
311 static void kse_stop(struct ifnet *, int);
312 static void kse_reset(struct kse_softc *);
313 static void kse_set_filter(struct kse_softc *);
314 static int add_rxbuf(struct kse_softc *, int);
315 static void rxdrain(struct kse_softc *);
316 static int kse_intr(void *);
317 static void rxintr(struct kse_softc *);
318 static void txreap(struct kse_softc *);
319 static void lnkchg(struct kse_softc *);
320 static int ifmedia_upd(struct ifnet *);
321 static void ifmedia_sts(struct ifnet *, struct ifmediareq *);
322 static void phy_tick(void *);
323 static int ifmedia2_upd(struct ifnet *);
324 static void ifmedia2_sts(struct ifnet *, struct ifmediareq *);
325 #ifdef KSE_EVENT_COUNTERS
326 static void stat_tick(void *);
327 static void zerostats(struct kse_softc *);
328 #endif
329 
330 static int
331 kse_match(struct device *parent, struct cfdata *match, void *aux)
332 {
333 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
334 
335 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_MICREL &&
336 	     (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_MICREL_KSZ8842 ||
337 	      PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_MICREL_KSZ8841) &&
338 	    PCI_CLASS(pa->pa_class) == PCI_CLASS_NETWORK)
339 		return 1;
340 
341 	return 0;
342 }
343 
344 static void
345 kse_attach(struct device *parent, struct device *self, void *aux)
346 {
347 	struct kse_softc *sc = (struct kse_softc *)self;
348 	struct pci_attach_args *pa = aux;
349 	pci_chipset_tag_t pc = pa->pa_pc;
350 	pci_intr_handle_t ih;
351 	const char *intrstr;
352 	struct ifnet *ifp;
353 	struct ifmedia *ifm;
354 	uint8_t enaddr[ETHER_ADDR_LEN];
355 	bus_dma_segment_t seg;
356 	int i, p, error, nseg;
357 	pcireg_t pmode;
358 	int pmreg;
359 
360 	if (pci_mapreg_map(pa, 0x10,
361 	    PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT,
362 	    0, &sc->sc_st, &sc->sc_sh, NULL, NULL) != 0) {
363 		printf(": unable to map device registers\n");
364 		return;
365 	}
366 
367 	sc->sc_dmat = pa->pa_dmat;
368 
369 	/* Make sure bus mastering is enabled. */
370 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
371 	    pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG) |
372 	    PCI_COMMAND_MASTER_ENABLE);
373 
374 	/* Get it out of power save mode, if needed. */
375 	if (pci_get_capability(pc, pa->pa_tag, PCI_CAP_PWRMGMT, &pmreg, 0)) {
376 		pmode = pci_conf_read(pc, pa->pa_tag, pmreg + PCI_PMCSR) &
377 		    PCI_PMCSR_STATE_MASK;
378 		if (pmode == PCI_PMCSR_STATE_D3) {
379 			/*
380 			 * The card has lost all configuration data in
381 			 * this state, so punt.
382 			 */
383 			printf("%s: unable to wake from power state D3\n",
384 			    sc->sc_dev.dv_xname);
385 			return;
386 		}
387 		if (pmode != PCI_PMCSR_STATE_D0) {
388 			printf("%s: waking up from power date D%d\n",
389 			    sc->sc_dev.dv_xname, pmode);
390 			pci_conf_write(pc, pa->pa_tag, pmreg + PCI_PMCSR,
391 			    PCI_PMCSR_STATE_D0);
392 		}
393 	}
394 
395 	sc->sc_chip = PCI_PRODUCT(pa->pa_id);
396 	printf(": Micrel KSZ%04x Ethernet (rev. 0x%02x)\n",
397 	    sc->sc_chip, PCI_REVISION(pa->pa_class));
398 
399 	/*
400 	 * Read the Ethernet address from the EEPROM.
401 	 */
402 	i = CSR_READ_2(sc, MARL);
403 	enaddr[5] = i; enaddr[4] = i >> 8;
404 	i = CSR_READ_2(sc, MARM);
405 	enaddr[3] = i; enaddr[2] = i >> 8;
406 	i = CSR_READ_2(sc, MARH);
407 	enaddr[1] = i; enaddr[0] = i >> 8;
408 	printf("%s: Ethernet address: %s\n",
409 		sc->sc_dev.dv_xname, ether_sprintf(enaddr));
410 
411 	/*
412 	 * Enable chip function.
413 	 */
414 	CSR_WRITE_2(sc, CIDR, 1);
415 
416 	/*
417 	 * Map and establish our interrupt.
418 	 */
419 	if (pci_intr_map(pa, &ih)) {
420 		printf("%s: unable to map interrupt\n", sc->sc_dev.dv_xname);
421 		return;
422 	}
423 	intrstr = pci_intr_string(pc, ih);
424 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, kse_intr, sc);
425 	if (sc->sc_ih == NULL) {
426 		printf("%s: unable to establish interrupt",
427 		    sc->sc_dev.dv_xname);
428 		if (intrstr != NULL)
429 			printf(" at %s", intrstr);
430 		printf("\n");
431 		return;
432 	}
433 	printf("%s: interrupting at %s\n", sc->sc_dev.dv_xname, intrstr);
434 
435 	/*
436 	 * Allocate the control data structures, and create and load the
437 	 * DMA map for it.
438 	 */
439 	error = bus_dmamem_alloc(sc->sc_dmat,
440 	    sizeof(struct kse_control_data), PAGE_SIZE, 0, &seg, 1, &nseg, 0);
441 	if (error != 0) {
442 		printf("%s: unable to allocate control data, error = %d\n",
443 		    sc->sc_dev.dv_xname, error);
444 		goto fail_0;
445 	}
446 	error = bus_dmamem_map(sc->sc_dmat, &seg, nseg,
447 	    sizeof(struct kse_control_data), (void **)&sc->sc_control_data,
448 	    BUS_DMA_COHERENT);
449 	if (error != 0) {
450 		printf("%s: unable to map control data, error = %d\n",
451 		    sc->sc_dev.dv_xname, error);
452 		goto fail_1;
453 	}
454 	error = bus_dmamap_create(sc->sc_dmat,
455 	    sizeof(struct kse_control_data), 1,
456 	    sizeof(struct kse_control_data), 0, 0, &sc->sc_cddmamap);
457 	if (error != 0) {
458 		printf("%s: unable to create control data DMA map, "
459 		    "error = %d\n", sc->sc_dev.dv_xname, error);
460 		goto fail_2;
461 	}
462 	error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
463 	    sc->sc_control_data, sizeof(struct kse_control_data), NULL, 0);
464 	if (error != 0) {
465 		printf("%s: unable to load control data DMA map, error = %d\n",
466 		    sc->sc_dev.dv_xname, error);
467 		goto fail_3;
468 	}
469 	for (i = 0; i < KSE_TXQUEUELEN; i++) {
470 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES,
471 		    KSE_NTXSEGS, MCLBYTES, 0, 0,
472 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
473 			printf("%s: unable to create tx DMA map %d, "
474 			    "error = %d\n", sc->sc_dev.dv_xname, i, error);
475 			goto fail_4;
476 		}
477 	}
478 	for (i = 0; i < KSE_NRXDESC; i++) {
479 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES,
480 		    1, MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
481 			printf("%s: unable to create rx DMA map %d, "
482 			    "error = %d\n", sc->sc_dev.dv_xname, i, error);
483 			goto fail_5;
484 		}
485 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
486 	}
487 
488 	callout_init(&sc->sc_callout, 0);
489 	callout_init(&sc->sc_stat_ch, 0);
490 
491 	ifm = &sc->sc_media;
492 	if (sc->sc_chip == 0x8841) {
493 		ifmedia_init(ifm, 0, ifmedia_upd, ifmedia_sts);
494 		ifmedia_add(ifm, IFM_ETHER|IFM_10_T, 0, NULL);
495 		ifmedia_add(ifm, IFM_ETHER|IFM_10_T|IFM_FDX, 0, NULL);
496 		ifmedia_add(ifm, IFM_ETHER|IFM_100_TX, 0, NULL);
497 		ifmedia_add(ifm, IFM_ETHER|IFM_100_TX|IFM_FDX, 0, NULL);
498 		ifmedia_add(ifm, IFM_ETHER|IFM_AUTO, 0, NULL);
499 		ifmedia_set(ifm, IFM_ETHER|IFM_AUTO);
500 	}
501 	else {
502 		ifmedia_init(ifm, 0, ifmedia2_upd, ifmedia2_sts);
503 		ifmedia_add(ifm, IFM_ETHER|IFM_AUTO, 0, NULL);
504 		ifmedia_set(ifm, IFM_ETHER|IFM_AUTO);
505 	}
506 
507 	printf("%s: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto\n",
508 	    sc->sc_dev.dv_xname);
509 
510 	ifp = &sc->sc_ethercom.ec_if;
511 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
512 	ifp->if_softc = sc;
513 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
514 	ifp->if_ioctl = kse_ioctl;
515 	ifp->if_start = kse_start;
516 	ifp->if_watchdog = kse_watchdog;
517 	ifp->if_init = kse_init;
518 	ifp->if_stop = kse_stop;
519 	IFQ_SET_READY(&ifp->if_snd);
520 
521 	/*
522 	 * KSZ8842 can handle 802.1Q VLAN-sized frames,
523 	 * can do IPv4, TCPv4, and UDPv4 checksums in hardware.
524 	 */
525 	sc->sc_ethercom.ec_capabilities |= ETHERCAP_VLAN_MTU;
526 	ifp->if_capabilities |=
527 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
528 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
529 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
530 
531 	if_attach(ifp);
532 	ether_ifattach(ifp, enaddr);
533 
534 	p = (sc->sc_chip == 0x8842) ? 3 : 1;
535 #ifdef KSE_EVENT_COUNTERS
536 	for (i = 0; i < p; i++) {
537 		struct ksext *ee = &sc->sc_ext;
538 		sprintf(ee->evcntname[i], "%s.%d", sc->sc_dev.dv_xname, i+1);
539 		evcnt_attach_dynamic(&ee->pev[i][0], EVCNT_TYPE_MISC,
540 		    NULL, ee->evcntname[i], "RxLoPriotyByte");
541 		evcnt_attach_dynamic(&ee->pev[i][1], EVCNT_TYPE_MISC,
542 		    NULL, ee->evcntname[i], "RxHiPriotyByte");
543 		evcnt_attach_dynamic(&ee->pev[i][2], EVCNT_TYPE_MISC,
544 		    NULL, ee->evcntname[i], "RxUndersizePkt");
545 		evcnt_attach_dynamic(&ee->pev[i][3], EVCNT_TYPE_MISC,
546 		    NULL, ee->evcntname[i], "RxFragments");
547 		evcnt_attach_dynamic(&ee->pev[i][4], EVCNT_TYPE_MISC,
548 		    NULL, ee->evcntname[i], "RxOversize");
549 		evcnt_attach_dynamic(&ee->pev[i][5], EVCNT_TYPE_MISC,
550 		    NULL, ee->evcntname[i], "RxJabbers");
551 		evcnt_attach_dynamic(&ee->pev[i][6], EVCNT_TYPE_MISC,
552 		    NULL, ee->evcntname[i], "RxSymbolError");
553 		evcnt_attach_dynamic(&ee->pev[i][7], EVCNT_TYPE_MISC,
554 		    NULL, ee->evcntname[i], "RxCRCError");
555 		evcnt_attach_dynamic(&ee->pev[i][8], EVCNT_TYPE_MISC,
556 		    NULL, ee->evcntname[i], "RxAlignmentError");
557 		evcnt_attach_dynamic(&ee->pev[i][9], EVCNT_TYPE_MISC,
558 		    NULL, ee->evcntname[i], "RxControl8808Pkts");
559 		evcnt_attach_dynamic(&ee->pev[i][10], EVCNT_TYPE_MISC,
560 		    NULL, ee->evcntname[i], "RxPausePkts");
561 		evcnt_attach_dynamic(&ee->pev[i][11], EVCNT_TYPE_MISC,
562 		    NULL, ee->evcntname[i], "RxBroadcast");
563 		evcnt_attach_dynamic(&ee->pev[i][12], EVCNT_TYPE_MISC,
564 		    NULL, ee->evcntname[i], "RxMulticast");
565 		evcnt_attach_dynamic(&ee->pev[i][13], EVCNT_TYPE_MISC,
566 		    NULL, ee->evcntname[i], "RxUnicast");
567 		evcnt_attach_dynamic(&ee->pev[i][14], EVCNT_TYPE_MISC,
568 		    NULL, ee->evcntname[i], "Rx64Octets");
569 		evcnt_attach_dynamic(&ee->pev[i][15], EVCNT_TYPE_MISC,
570 		    NULL, ee->evcntname[i], "Rx65To127Octets");
571 		evcnt_attach_dynamic(&ee->pev[i][16], EVCNT_TYPE_MISC,
572 		    NULL, ee->evcntname[i], "Rx128To255Octets");
573 		evcnt_attach_dynamic(&ee->pev[i][17], EVCNT_TYPE_MISC,
574 		    NULL, ee->evcntname[i], "Rx255To511Octets");
575 		evcnt_attach_dynamic(&ee->pev[i][18], EVCNT_TYPE_MISC,
576 		    NULL, ee->evcntname[i], "Rx512To1023Octets");
577 		evcnt_attach_dynamic(&ee->pev[i][19], EVCNT_TYPE_MISC,
578 		    NULL, ee->evcntname[i], "Rx1024To1522Octets");
579 		evcnt_attach_dynamic(&ee->pev[i][20], EVCNT_TYPE_MISC,
580 		    NULL, ee->evcntname[i], "TxLoPriotyByte");
581 		evcnt_attach_dynamic(&ee->pev[i][21], EVCNT_TYPE_MISC,
582 		    NULL, ee->evcntname[i], "TxHiPriotyByte");
583 		evcnt_attach_dynamic(&ee->pev[i][22], EVCNT_TYPE_MISC,
584 		    NULL, ee->evcntname[i], "TxLateCollision");
585 		evcnt_attach_dynamic(&ee->pev[i][23], EVCNT_TYPE_MISC,
586 		    NULL, ee->evcntname[i], "TxPausePkts");
587 		evcnt_attach_dynamic(&ee->pev[i][24], EVCNT_TYPE_MISC,
588 		    NULL, ee->evcntname[i], "TxBroadcastPkts");
589 		evcnt_attach_dynamic(&ee->pev[i][25], EVCNT_TYPE_MISC,
590 		    NULL, ee->evcntname[i], "TxMulticastPkts");
591 		evcnt_attach_dynamic(&ee->pev[i][26], EVCNT_TYPE_MISC,
592 		    NULL, ee->evcntname[i], "TxUnicastPkts");
593 		evcnt_attach_dynamic(&ee->pev[i][27], EVCNT_TYPE_MISC,
594 		    NULL, ee->evcntname[i], "TxDeferred");
595 		evcnt_attach_dynamic(&ee->pev[i][28], EVCNT_TYPE_MISC,
596 		    NULL, ee->evcntname[i], "TxTotalCollision");
597 		evcnt_attach_dynamic(&ee->pev[i][29], EVCNT_TYPE_MISC,
598 		    NULL, ee->evcntname[i], "TxExcessiveCollision");
599 		evcnt_attach_dynamic(&ee->pev[i][30], EVCNT_TYPE_MISC,
600 		    NULL, ee->evcntname[i], "TxSingleCollision");
601 		evcnt_attach_dynamic(&ee->pev[i][31], EVCNT_TYPE_MISC,
602 		    NULL, ee->evcntname[i], "TxMultipleCollision");
603 		evcnt_attach_dynamic(&ee->pev[i][32], EVCNT_TYPE_MISC,
604 		    NULL, ee->evcntname[i], "TxDropPkts");
605 		evcnt_attach_dynamic(&ee->pev[i][33], EVCNT_TYPE_MISC,
606 		    NULL, ee->evcntname[i], "RxDropPkts");
607 	}
608 #endif
609 	return;
610 
611  fail_5:
612 	for (i = 0; i < KSE_NRXDESC; i++) {
613 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
614 			bus_dmamap_destroy(sc->sc_dmat,
615 			    sc->sc_rxsoft[i].rxs_dmamap);
616 	}
617  fail_4:
618 	for (i = 0; i < KSE_TXQUEUELEN; i++) {
619 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
620 			bus_dmamap_destroy(sc->sc_dmat,
621 			    sc->sc_txsoft[i].txs_dmamap);
622 	}
623 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
624  fail_3:
625 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
626  fail_2:
627 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
628 	    sizeof(struct kse_control_data));
629  fail_1:
630 	bus_dmamem_free(sc->sc_dmat, &seg, nseg);
631  fail_0:
632 	return;
633 }
634 
635 static int
636 kse_ioctl(struct ifnet *ifp, u_long cmd, void *data)
637 {
638 	struct kse_softc *sc = ifp->if_softc;
639 	struct ifreq *ifr = (struct ifreq *)data;
640 	int s, error;
641 
642 	s = splnet();
643 
644 	switch (cmd) {
645 	case SIOCSIFMEDIA:
646 	case SIOCGIFMEDIA:
647 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
648 		break;
649 
650 	default:
651 		error = ether_ioctl(ifp, cmd, data);
652 		if (cmd == ENETRESET) {
653 			/*
654 			 * Multicast list has changed; set the hardware filter
655 			 * accordingly.
656 			 */
657 			if (ifp->if_flags & IFF_RUNNING)
658 				kse_set_filter(sc);
659 			error = 0;
660 		}
661 		break;
662 	}
663 
664 	kse_start(ifp);
665 
666 	splx(s);
667 	return error;
668 }
669 
670 static int
671 kse_init(struct ifnet *ifp)
672 {
673 	struct kse_softc *sc = ifp->if_softc;
674 	uint32_t paddr;
675 	int i, error = 0;
676 
677 	/* cancel pending I/O */
678 	kse_stop(ifp, 0);
679 
680 	/* reset all registers but PCI configuration */
681 	kse_reset(sc);
682 
683 	/* craft Tx descriptor ring */
684 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
685 	for (i = 0, paddr = KSE_CDTXADDR(sc, 1); i < KSE_NTXDESC - 1; i++) {
686 		sc->sc_txdescs[i].t3 = paddr;
687 		paddr += sizeof(struct tdes);
688 	}
689 	sc->sc_txdescs[KSE_NTXDESC - 1].t3 = KSE_CDTXADDR(sc, 0);
690 	KSE_CDTXSYNC(sc, 0, KSE_NTXDESC,
691 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
692 	sc->sc_txfree = KSE_NTXDESC;
693 	sc->sc_txnext = 0;
694 
695 	for (i = 0; i < KSE_TXQUEUELEN; i++)
696 		sc->sc_txsoft[i].txs_mbuf = NULL;
697 	sc->sc_txsfree = KSE_TXQUEUELEN;
698 	sc->sc_txsnext = 0;
699 	sc->sc_txsdirty = 0;
700 
701 	/* craft Rx descriptor ring */
702 	memset(sc->sc_rxdescs, 0, sizeof(sc->sc_rxdescs));
703 	for (i = 0, paddr = KSE_CDRXADDR(sc, 1); i < KSE_NRXDESC - 1; i++) {
704 		sc->sc_rxdescs[i].r3 = paddr;
705 		paddr += sizeof(struct rdes);
706 	}
707 	sc->sc_rxdescs[KSE_NRXDESC - 1].r3 = KSE_CDRXADDR(sc, 0);
708 	for (i = 0; i < KSE_NRXDESC; i++) {
709 		if (sc->sc_rxsoft[i].rxs_mbuf == NULL) {
710 			if ((error = add_rxbuf(sc, i)) != 0) {
711 				printf("%s: unable to allocate or map rx "
712 				    "buffer %d, error = %d\n",
713 				     sc->sc_dev.dv_xname, i, error);
714 				rxdrain(sc);
715 				goto out;
716 			}
717 		}
718 		else
719 			KSE_INIT_RXDESC(sc, i);
720 	}
721 	sc->sc_rxptr = 0;
722 
723 	/* hand Tx/Rx rings to HW */
724 	CSR_WRITE_4(sc, TDLB, KSE_CDTXADDR(sc, 0));
725 	CSR_WRITE_4(sc, RDLB, KSE_CDRXADDR(sc, 0));
726 
727 	sc->sc_txc = TXC_TEN | TXC_EP | TXC_AC | TXC_FCE;
728 	sc->sc_rxc = RXC_REN | RXC_RU | RXC_FCE;
729 	if (ifp->if_flags & IFF_PROMISC)
730 		sc->sc_rxc |= RXC_RA;
731 	if (ifp->if_flags & IFF_BROADCAST)
732 		sc->sc_rxc |= RXC_RB;
733 	sc->sc_t1csum = sc->sc_mcsum = 0;
734 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
735 		sc->sc_rxc |= RXC_ICC;
736 		sc->sc_mcsum |= M_CSUM_IPv4;
737 	}
738 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
739 		sc->sc_txc |= TXC_ICG;
740 		sc->sc_t1csum |= T1_IPCKG;
741 	}
742 	if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
743 		sc->sc_rxc |= RXC_TCC;
744 		sc->sc_mcsum |= M_CSUM_TCPv4;
745 	}
746 	if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
747 		sc->sc_txc |= TXC_TCG;
748 		sc->sc_t1csum |= T1_TCPCKG;
749 	}
750 	if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
751 		sc->sc_rxc |= RXC_UCC;
752 		sc->sc_mcsum |= M_CSUM_UDPv4;
753 	}
754 	if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
755 		sc->sc_txc |= TXC_UCG;
756 		sc->sc_t1csum |= T1_UDPCKG;
757 	}
758 	sc->sc_txc |= (kse_burstsize << TXC_BS_SFT);
759 	sc->sc_rxc |= (kse_burstsize << RXC_BS_SFT);
760 
761 	/* build multicast hash filter if necessary */
762 	kse_set_filter(sc);
763 
764 	/* set current media */
765 	(void)ifmedia_upd(ifp);
766 
767 	/* enable transmitter and receiver */
768 	CSR_WRITE_4(sc, MDTXC, sc->sc_txc);
769 	CSR_WRITE_4(sc, MDRXC, sc->sc_rxc);
770 	CSR_WRITE_4(sc, MDRSC, 1);
771 
772 	/* enable interrupts */
773 	sc->sc_inten = INT_DMTS|INT_DMRS|INT_DMRBUS;
774 	if (sc->sc_chip == 0x8841)
775 		sc->sc_inten |= INT_DMLCS;
776 	CSR_WRITE_4(sc, INTST, ~0);
777 	CSR_WRITE_4(sc, INTEN, sc->sc_inten);
778 
779 	ifp->if_flags |= IFF_RUNNING;
780 	ifp->if_flags &= ~IFF_OACTIVE;
781 
782 	if (sc->sc_chip == 0x8841) {
783 		/* start one second timer */
784 		callout_reset(&sc->sc_callout, hz, phy_tick, sc);
785 	}
786 #ifdef KSE_EVENT_COUNTERS
787 	/* start statistics gather 1 minute timer */
788 	zerostats(sc);
789 	callout_reset(&sc->sc_stat_ch, hz * 60, stat_tick, sc);
790 #endif
791 
792  out:
793 	if (error) {
794 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
795 		ifp->if_timer = 0;
796 		printf("%s: interface not running\n", sc->sc_dev.dv_xname);
797 	}
798 	return error;
799 }
800 
801 static void
802 kse_stop(struct ifnet *ifp, int disable)
803 {
804 	struct kse_softc *sc = ifp->if_softc;
805 	struct kse_txsoft *txs;
806 	int i;
807 
808 	if (sc->sc_chip == 0x8841)
809 		callout_stop(&sc->sc_callout);
810 	callout_stop(&sc->sc_stat_ch);
811 
812 	sc->sc_txc &= ~TXC_TEN;
813 	sc->sc_rxc &= ~RXC_REN;
814 	CSR_WRITE_4(sc, MDTXC, sc->sc_txc);
815 	CSR_WRITE_4(sc, MDRXC, sc->sc_rxc);
816 
817 	for (i = 0; i < KSE_TXQUEUELEN; i++) {
818 		txs = &sc->sc_txsoft[i];
819 		if (txs->txs_mbuf != NULL) {
820 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
821 			m_freem(txs->txs_mbuf);
822 			txs->txs_mbuf = NULL;
823 		}
824 	}
825 
826 	if (disable)
827 		rxdrain(sc);
828 
829 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
830 	ifp->if_timer = 0;
831 }
832 
833 static void
834 kse_reset(struct kse_softc *sc)
835 {
836 
837 	CSR_WRITE_2(sc, GRR, 1);
838 	delay(1000); /* PDF does not mention the delay amount */
839 	CSR_WRITE_2(sc, GRR, 0);
840 
841 	CSR_WRITE_2(sc, CIDR, 1);
842 }
843 
844 static void
845 kse_watchdog(struct ifnet *ifp)
846 {
847 	struct kse_softc *sc = ifp->if_softc;
848 
849 	/*
850 	 * Since we're not interrupting every packet, sweep
851 	 * up before we report an error.
852 	 */
853 	txreap(sc);
854 
855 	if (sc->sc_txfree != KSE_NTXDESC) {
856 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
857 		    sc->sc_dev.dv_xname, sc->sc_txfree, sc->sc_txsfree,
858 		    sc->sc_txnext);
859 		ifp->if_oerrors++;
860 
861 		/* Reset the interface. */
862 		kse_init(ifp);
863 	}
864 	else if (ifp->if_flags & IFF_DEBUG)
865 		printf("%s: recovered from device timeout\n",
866 		    sc->sc_dev.dv_xname);
867 
868 	/* Try to get more packets going. */
869 	kse_start(ifp);
870 }
871 
872 static void
873 kse_start(struct ifnet *ifp)
874 {
875 	struct kse_softc *sc = ifp->if_softc;
876 	struct mbuf *m0, *m;
877 	struct kse_txsoft *txs;
878 	bus_dmamap_t dmamap;
879 	int error, nexttx, lasttx, ofree, seg;
880 	uint32_t tdes0;
881 
882 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
883 		return;
884 
885 	/*
886 	 * Remember the previous number of free descriptors.
887 	 */
888 	ofree = sc->sc_txfree;
889 
890 	/*
891 	 * Loop through the send queue, setting up transmit descriptors
892 	 * until we drain the queue, or use up all available transmit
893 	 * descriptors.
894 	 */
895 	for (;;) {
896 		IFQ_POLL(&ifp->if_snd, m0);
897 		if (m0 == NULL)
898 			break;
899 
900 		if (sc->sc_txsfree < KSE_TXQUEUE_GC) {
901 			txreap(sc);
902 			if (sc->sc_txsfree == 0)
903 				break;
904 		}
905 		txs = &sc->sc_txsoft[sc->sc_txsnext];
906 		dmamap = txs->txs_dmamap;
907 
908 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
909 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
910 		if (error) {
911 			if (error == EFBIG) {
912 				printf("%s: Tx packet consumes too many "
913 				    "DMA segments, dropping...\n",
914 				    sc->sc_dev.dv_xname);
915 				    IFQ_DEQUEUE(&ifp->if_snd, m0);
916 				    m_freem(m0);
917 				    continue;
918 			}
919 			/* Short on resources, just stop for now. */
920 			break;
921 		}
922 
923 		if (dmamap->dm_nsegs > sc->sc_txfree) {
924 			/*
925 			 * Not enough free descriptors to transmit this
926 			 * packet.  We haven't committed anything yet,
927 			 * so just unload the DMA map, put the packet
928 			 * back on the queue, and punt.	 Notify the upper
929 			 * layer that there are not more slots left.
930 			 */
931 			ifp->if_flags |= IFF_OACTIVE;
932 			bus_dmamap_unload(sc->sc_dmat, dmamap);
933 			break;
934 		}
935 
936 		IFQ_DEQUEUE(&ifp->if_snd, m0);
937 
938 		/*
939 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
940 		 */
941 
942 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
943 		    BUS_DMASYNC_PREWRITE);
944 
945 		lasttx = -1; tdes0 = 0;
946 		for (nexttx = sc->sc_txnext, seg = 0;
947 		     seg < dmamap->dm_nsegs;
948 		     seg++, nexttx = KSE_NEXTTX(nexttx)) {
949 			struct tdes *tdes = &sc->sc_txdescs[nexttx];
950 			/*
951 			 * If this is the first descriptor we're
952 			 * enqueueing, don't set the OWN bit just
953 			 * yet.	 That could cause a race condition.
954 			 * We'll do it below.
955 			 */
956 			tdes->t2 = dmamap->dm_segs[seg].ds_addr;
957 			tdes->t1 = sc->sc_t1csum
958 			     | (dmamap->dm_segs[seg].ds_len & T1_TBS_MASK);
959 			tdes->t0 = tdes0;
960 			tdes0 |= T0_OWN;
961 			lasttx = nexttx;
962 		}
963 
964 		/*
965 		 * Outgoing NFS mbuf must be unloaded when Tx completed.
966 		 * Without T1_IC NFS mbuf is left unack'ed for excessive
967 		 * time and NFS stops to proceed until kse_watchdog()
968 		 * calls txreap() to reclaim the unack'ed mbuf.
969 		 * It's painful to traverse every mbuf chain to determine
970 		 * whether someone is waiting for Tx completion.
971 		 */
972 		m = m0;
973 		do {
974 			if ((m->m_flags & M_EXT) && m->m_ext.ext_free) {
975 				sc->sc_txdescs[lasttx].t1 |= T1_IC;
976 				break;
977 			}
978 		} while ((m = m->m_next) != NULL);
979 
980 		/* write last T0_OWN bit of the 1st segment */
981 		sc->sc_txdescs[lasttx].t1 |= T1_LS;
982 		sc->sc_txdescs[sc->sc_txnext].t1 |= T1_FS;
983 		sc->sc_txdescs[sc->sc_txnext].t0 = T0_OWN;
984 		KSE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
985 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
986 
987 		/* tell DMA start transmit */
988 		CSR_WRITE_4(sc, MDTSC, 1);
989 
990 		txs->txs_mbuf = m0;
991 		txs->txs_firstdesc = sc->sc_txnext;
992 		txs->txs_lastdesc = lasttx;
993 		txs->txs_ndesc = dmamap->dm_nsegs;
994 
995 		sc->sc_txfree -= txs->txs_ndesc;
996 		sc->sc_txnext = nexttx;
997 		sc->sc_txsfree--;
998 		sc->sc_txsnext = KSE_NEXTTXS(sc->sc_txsnext);
999 #if NBPFILTER > 0
1000 		/*
1001 		 * Pass the packet to any BPF listeners.
1002 		 */
1003 		if (ifp->if_bpf)
1004 			bpf_mtap(ifp->if_bpf, m0);
1005 #endif /* NBPFILTER > 0 */
1006 	}
1007 
1008 	if (sc->sc_txsfree == 0 || sc->sc_txfree == 0) {
1009 		/* No more slots left; notify upper layer. */
1010 		ifp->if_flags |= IFF_OACTIVE;
1011 	}
1012 	if (sc->sc_txfree != ofree) {
1013 		/* Set a watchdog timer in case the chip flakes out. */
1014 		ifp->if_timer = 5;
1015 	}
1016 }
1017 
1018 static void
1019 kse_set_filter(struct kse_softc *sc)
1020 {
1021 	struct ether_multistep step;
1022 	struct ether_multi *enm;
1023 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1024 	uint32_t h, hashes[2];
1025 
1026 	sc->sc_rxc &= ~(RXC_MHTE | RXC_RM);
1027 	ifp->if_flags &= ~IFF_ALLMULTI;
1028 	if (ifp->if_flags & IFF_PROMISC)
1029 		return;
1030 
1031 	ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
1032 	if (enm == NULL)
1033 		return;
1034 	hashes[0] = hashes[1] = 0;
1035 	do {
1036 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1037 			/*
1038 			 * We must listen to a range of multicast addresses.
1039 			 * For now, just accept all multicasts, rather than
1040 			 * trying to set only those filter bits needed to match
1041 			 * the range.  (At this time, the only use of address
1042 			 * ranges is for IP multicast routing, for which the
1043 			 * range is big enough to require all bits set.)
1044 			 */
1045 			goto allmulti;
1046 		}
1047 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN) >> 26;
1048 		hashes[h >> 5] |= 1 << (h & 0x1f);
1049 		ETHER_NEXT_MULTI(step, enm);
1050 	} while (enm != NULL);
1051 	sc->sc_rxc |= RXC_MHTE;
1052 	CSR_WRITE_4(sc, MTR0, hashes[0]);
1053 	CSR_WRITE_4(sc, MTR1, hashes[1]);
1054 	return;
1055  allmulti:
1056 	sc->sc_rxc |= RXC_RM;
1057 	ifp->if_flags |= IFF_ALLMULTI;
1058 }
1059 
1060 static int
1061 add_rxbuf(struct kse_softc *sc, int idx)
1062 {
1063 	struct kse_rxsoft *rxs = &sc->sc_rxsoft[idx];
1064 	struct mbuf *m;
1065 	int error;
1066 
1067 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1068 	if (m == NULL)
1069 		return ENOBUFS;
1070 
1071 	MCLGET(m, M_DONTWAIT);
1072 	if ((m->m_flags & M_EXT) == 0) {
1073 		m_freem(m);
1074 		return ENOBUFS;
1075 	}
1076 
1077 	if (rxs->rxs_mbuf != NULL)
1078 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1079 
1080 	rxs->rxs_mbuf = m;
1081 
1082 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap,
1083 	    m->m_ext.ext_buf, m->m_ext.ext_size, NULL, BUS_DMA_NOWAIT);
1084 	if (error) {
1085 		printf("%s: can't load rx DMA map %d, error = %d\n",
1086 		    sc->sc_dev.dv_xname, idx, error);
1087 		panic("kse_add_rxbuf");
1088 	}
1089 
1090 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1091 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
1092 
1093 	KSE_INIT_RXDESC(sc, idx);
1094 
1095 	return 0;
1096 }
1097 
1098 static void
1099 rxdrain(struct kse_softc *sc)
1100 {
1101 	struct kse_rxsoft *rxs;
1102 	int i;
1103 
1104 	for (i = 0; i < KSE_NRXDESC; i++) {
1105 		rxs = &sc->sc_rxsoft[i];
1106 		if (rxs->rxs_mbuf != NULL) {
1107 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
1108 			m_freem(rxs->rxs_mbuf);
1109 			rxs->rxs_mbuf = NULL;
1110 		}
1111 	}
1112 }
1113 
1114 static int
1115 kse_intr(void *arg)
1116 {
1117 	struct kse_softc *sc = arg;
1118 	uint32_t isr;
1119 
1120 	if ((isr = CSR_READ_4(sc, INTST)) == 0)
1121 		return 0;
1122 
1123 	if (isr & INT_DMRS)
1124 		rxintr(sc);
1125 	if (isr & INT_DMTS)
1126 		txreap(sc);
1127 	if (isr & INT_DMLCS)
1128 		lnkchg(sc);
1129 	if (isr & INT_DMRBUS)
1130 		printf("%s: Rx descriptor full\n", sc->sc_dev.dv_xname);
1131 
1132 	CSR_WRITE_4(sc, INTST, isr);
1133 	return 1;
1134 }
1135 
1136 static void
1137 rxintr(struct kse_softc *sc)
1138 {
1139 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1140 	struct kse_rxsoft *rxs;
1141 	struct mbuf *m;
1142 	uint32_t rxstat;
1143 	int i, len;
1144 
1145 	for (i = sc->sc_rxptr; /*CONSTCOND*/ 1; i = KSE_NEXTRX(i)) {
1146 		rxs = &sc->sc_rxsoft[i];
1147 
1148 		KSE_CDRXSYNC(sc, i,
1149 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1150 
1151 		rxstat = sc->sc_rxdescs[i].r0;
1152 
1153 		if (rxstat & R0_OWN) /* desc is left empty */
1154 			break;
1155 
1156 		/* R0_FS|R0_LS must have been marked for this desc */
1157 
1158 		if (rxstat & R0_ES) {
1159 			ifp->if_ierrors++;
1160 #define PRINTERR(bit, str)						\
1161 			if (rxstat & (bit))				\
1162 				printf("%s: receive error: %s\n",	\
1163 				    sc->sc_dev.dv_xname, str)
1164 			PRINTERR(R0_TL, "frame too long");
1165 			PRINTERR(R0_RF, "runt frame");
1166 			PRINTERR(R0_CE, "bad FCS");
1167 #undef PRINTERR
1168 			KSE_INIT_RXDESC(sc, i);
1169 			continue;
1170 		}
1171 
1172 		/* HW errata; frame might be too small or too large */
1173 
1174 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
1175 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
1176 
1177 		len = rxstat & R0_FL_MASK;
1178 		len -= ETHER_CRC_LEN;	/* trim CRC off */
1179 		m = rxs->rxs_mbuf;
1180 
1181 		if (add_rxbuf(sc, i) != 0) {
1182 			ifp->if_ierrors++;
1183 			KSE_INIT_RXDESC(sc, i);
1184 			bus_dmamap_sync(sc->sc_dmat,
1185 			    rxs->rxs_dmamap, 0,
1186 			    rxs->rxs_dmamap->dm_mapsize,
1187 			    BUS_DMASYNC_PREREAD);
1188 			continue;
1189 		}
1190 
1191 		ifp->if_ipackets++;
1192 		m->m_pkthdr.rcvif = ifp;
1193 		m->m_pkthdr.len = m->m_len = len;
1194 
1195 		if (sc->sc_mcsum) {
1196 			m->m_pkthdr.csum_flags |= sc->sc_mcsum;
1197 			if (rxstat & R0_IPE)
1198 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1199 			if (rxstat & (R0_TCPE | R0_UDPE))
1200 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1201 		}
1202 #if NBPFILTER > 0
1203 		if (ifp->if_bpf)
1204 			bpf_mtap(ifp->if_bpf, m);
1205 #endif /* NBPFILTER > 0 */
1206 		(*ifp->if_input)(ifp, m);
1207 #ifdef KSEDIAGNOSTIC
1208 		if (kse_monitor_rxintr > 0) {
1209 			printf("m stat %x data %p len %d\n",
1210 			    rxstat, m->m_data, m->m_len);
1211 		}
1212 #endif
1213 	}
1214 	sc->sc_rxptr = i;
1215 }
1216 
1217 static void
1218 txreap(struct kse_softc *sc)
1219 {
1220 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
1221 	struct kse_txsoft *txs;
1222 	uint32_t txstat;
1223 	int i;
1224 
1225 	ifp->if_flags &= ~IFF_OACTIVE;
1226 
1227 	for (i = sc->sc_txsdirty; sc->sc_txsfree != KSE_TXQUEUELEN;
1228 	     i = KSE_NEXTTXS(i), sc->sc_txsfree++) {
1229 		txs = &sc->sc_txsoft[i];
1230 
1231 		KSE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_ndesc,
1232 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1233 
1234 		txstat = sc->sc_txdescs[txs->txs_lastdesc].t0;
1235 
1236 		if (txstat & T0_OWN) /* desc is still in use */
1237 			break;
1238 
1239 		/* there is no way to tell transmission status per frame */
1240 
1241 		ifp->if_opackets++;
1242 
1243 		sc->sc_txfree += txs->txs_ndesc;
1244 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
1245 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1246 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
1247 		m_freem(txs->txs_mbuf);
1248 		txs->txs_mbuf = NULL;
1249 	}
1250 	sc->sc_txsdirty = i;
1251 	if (sc->sc_txsfree == KSE_TXQUEUELEN)
1252 		ifp->if_timer = 0;
1253 }
1254 
1255 static void
1256 lnkchg(struct kse_softc *sc)
1257 {
1258 	struct ifmediareq ifmr;
1259 
1260 #if 0 /* rambling link status */
1261 	printf("%s: link %s\n", sc->sc_dev.dv_xname,
1262 	    (CSR_READ_2(sc, P1SR) & (1U << 5)) ? "up" : "down");
1263 #endif
1264 	ifmedia_sts(&sc->sc_ethercom.ec_if, &ifmr);
1265 }
1266 
1267 static int
1268 ifmedia_upd(struct ifnet *ifp)
1269 {
1270 	struct kse_softc *sc = ifp->if_softc;
1271 	struct ifmedia *ifm = &sc->sc_media;
1272 	uint16_t ctl;
1273 
1274 	ctl = 0;
1275 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
1276 		ctl |= (1U << 13); /* restart AN */
1277 		ctl |= (1U << 7);  /* enable AN */
1278 		ctl |= (1U << 4);  /* advertise flow control pause */
1279 		ctl |= (1U << 3) | (1U << 2) | (1U << 1) | (1U << 0);
1280 	}
1281 	else {
1282 		if (IFM_SUBTYPE(ifm->ifm_media) == IFM_100_TX)
1283 			ctl |= (1U << 6);
1284 		if (ifm->ifm_media & IFM_FDX)
1285 			ctl |= (1U << 5);
1286 	}
1287 	CSR_WRITE_2(sc, P1CR4, ctl);
1288 
1289 	sc->sc_media_active = IFM_NONE;
1290 	sc->sc_media_status = IFM_AVALID;
1291 
1292 	return 0;
1293 }
1294 
1295 static void
1296 ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1297 {
1298 	struct kse_softc *sc = ifp->if_softc;
1299 	struct ifmedia *ifm = &sc->sc_media;
1300 	uint16_t ctl, sts, result;
1301 
1302 	ifmr->ifm_status = IFM_AVALID;
1303 	ifmr->ifm_active = IFM_ETHER;
1304 
1305 	ctl = CSR_READ_2(sc, P1CR4);
1306 	sts = CSR_READ_2(sc, P1SR);
1307 	if ((sts & (1U << 5)) == 0) {
1308 		ifmr->ifm_active |= IFM_NONE;
1309 		goto out; /* link is down */
1310 	}
1311 	ifmr->ifm_status |= IFM_ACTIVE;
1312 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
1313 		if ((sts & (1U << 6)) == 0) {
1314 			ifmr->ifm_active |= IFM_NONE;
1315 			goto out; /* negotiation in progress */
1316 		}
1317 		result = ctl & sts & 017;
1318 		if (result & (1U << 3))
1319 			ifmr->ifm_active |= IFM_100_TX|IFM_FDX;
1320 		else if (result & (1U << 2))
1321 			ifmr->ifm_active |= IFM_100_TX;
1322 		else if (result & (1U << 1))
1323 			ifmr->ifm_active |= IFM_10_T|IFM_FDX;
1324 		else if (result & (1U << 0))
1325 			ifmr->ifm_active |= IFM_10_T;
1326 		else
1327 			ifmr->ifm_active |= IFM_NONE;
1328 		if (ctl & (1U << 4))
1329 			ifmr->ifm_active |= IFM_FLOW | IFM_ETH_RXPAUSE;
1330 		if (sts & (1U << 4))
1331 			ifmr->ifm_active |= IFM_FLOW | IFM_ETH_TXPAUSE;
1332 	}
1333 	else {
1334 		ifmr->ifm_active |= (sts & (1U << 10)) ? IFM_100_TX : IFM_10_T;
1335 		if (sts & (1U << 9))
1336 			ifmr->ifm_active |= IFM_FDX;
1337 		if (sts & (1U << 12))
1338 			ifmr->ifm_active |= IFM_FLOW | IFM_ETH_RXPAUSE;
1339 		if (sts & (1U << 11))
1340 			ifmr->ifm_active |= IFM_FLOW | IFM_ETH_TXPAUSE;
1341 	}
1342 
1343   out:
1344 	sc->sc_media_status = ifmr->ifm_status;
1345 	sc->sc_media_active = ifmr->ifm_active;
1346 }
1347 
1348 static void
1349 phy_tick(void *arg)
1350 {
1351 	struct kse_softc *sc = arg;
1352 	struct ifmediareq ifmr;
1353 	int s;
1354 
1355 	s = splnet();
1356 	ifmedia_sts(&sc->sc_ethercom.ec_if, &ifmr);
1357 	splx(s);
1358 
1359 	callout_reset(&sc->sc_callout, hz, phy_tick, sc);
1360 }
1361 
1362 static int
1363 ifmedia2_upd(struct ifnet *ifp)
1364 {
1365 	struct kse_softc *sc = ifp->if_softc;
1366 
1367 	sc->sc_media_status = IFM_AVALID;
1368 	sc->sc_media_active = IFM_NONE;
1369 	return 0;
1370 }
1371 
1372 static void
1373 ifmedia2_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1374 {
1375 	struct kse_softc *sc = ifp->if_softc;
1376 	int p1sts, p2sts;
1377 
1378 	ifmr->ifm_status = IFM_AVALID;
1379 	ifmr->ifm_active = IFM_ETHER;
1380 	p1sts = CSR_READ_2(sc, P1SR);
1381 	p2sts = CSR_READ_2(sc, P2SR);
1382 	if (((p1sts | p2sts) & (1U << 5)) == 0)
1383 		ifmr->ifm_active |= IFM_NONE;
1384 	else {
1385 		ifmr->ifm_status |= IFM_ACTIVE;
1386 		ifmr->ifm_active |= IFM_100_TX|IFM_FDX;
1387 		ifmr->ifm_active |= IFM_FLOW|IFM_ETH_RXPAUSE|IFM_ETH_TXPAUSE;
1388 	}
1389 	sc->sc_media_status = ifmr->ifm_status;
1390 	sc->sc_media_active = ifmr->ifm_active;
1391 }
1392 
1393 #ifdef KSE_EVENT_COUNTERS
1394 static void
1395 stat_tick(arg)
1396 	void *arg;
1397 {
1398 	struct kse_softc *sc = arg;
1399 	struct ksext *ee = &sc->sc_ext;
1400 	int nport, p, i, val;
1401 
1402 	nport = (sc->sc_chip == 0x8842) ? 3 : 1;
1403 	for (p = 0; p < nport; p++) {
1404 		for (i = 0; i < 32; i++) {
1405 			val = 0x1c00 | (p * 0x20 + i);
1406 			CSR_WRITE_2(sc, IACR, val);
1407 			do {
1408 				val = CSR_READ_2(sc, IADR5) << 16;
1409 			} while ((val & (1U << 30)) == 0);
1410 			if (val & (1U << 31)) {
1411 				(void)CSR_READ_2(sc, IADR4);
1412 				val = 0x3fffffff; /* has made overflow */
1413 			}
1414 			else {
1415 				val &= 0x3fff0000;		/* 29:16 */
1416 				val |= CSR_READ_2(sc, IADR4);	/* 15:0 */
1417 			}
1418 			ee->pev[p][i].ev_count += val; /* i (0-31) */
1419 		}
1420 		CSR_WRITE_2(sc, IACR, 0x1c00 + 0x100 + p);
1421 		ee->pev[p][32].ev_count = CSR_READ_2(sc, IADR4); /* 32 */
1422 		CSR_WRITE_2(sc, IACR, 0x1c00 + 0x100 + p * 3 + 1);
1423 		ee->pev[p][33].ev_count = CSR_READ_2(sc, IADR4); /* 33 */
1424 	}
1425 	callout_reset(&sc->sc_stat_ch, hz * 60, stat_tick, arg);
1426 }
1427 
1428 static void
1429 zerostats(struct kse_softc *sc)
1430 {
1431 	struct ksext *ee = &sc->sc_ext;
1432 	int nport, p, i, val;
1433 
1434 	/* make sure all the HW counters get zero */
1435 	nport = (sc->sc_chip == 0x8842) ? 3 : 1;
1436 	for (p = 0; p < nport; p++) {
1437 		for (i = 0; i < 31; i++) {
1438 			val = 0x1c00 | (p * 0x20 + i);
1439 			CSR_WRITE_2(sc, IACR, val);
1440 			do {
1441 				val = CSR_READ_2(sc, IADR5) << 16;
1442 			} while ((val & (1U << 30)) == 0);
1443 			(void)CSR_READ_2(sc, IADR4);
1444 			ee->pev[p][i].ev_count = 0;
1445 		}
1446 	}
1447 }
1448 #endif
1449