Home | History | Annotate | Line # | Download | only in pci
if_dge.c revision 1.38.4.5
      1 /*	$NetBSD: if_dge.c,v 1.38.4.5 2017/02/05 13:40:29 skrll Exp $ */
      2 
      3 /*
      4  * Copyright (c) 2004, SUNET, Swedish University Computer Network.
      5  * All rights reserved.
      6  *
      7  * Written by Anders Magnusson for SUNET, Swedish University Computer Network.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *	This product includes software developed for the NetBSD Project by
     20  *	SUNET, Swedish University Computer Network.
     21  * 4. The name of SUNET may not be used to endorse or promote products
     22  *    derived from this software without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY SUNET ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     26  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     27  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     34  * POSSIBILITY OF SUCH DAMAGE.
     35  */
     36 
     37 /*
     38  * Copyright (c) 2001, 2002, 2003 Wasabi Systems, Inc.
     39  * All rights reserved.
     40  *
     41  * Written by Jason R. Thorpe for Wasabi Systems, Inc.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  * 3. All advertising materials mentioning features or use of this software
     52  *    must display the following acknowledgement:
     53  *	This product includes software developed for the NetBSD Project by
     54  *	Wasabi Systems, Inc.
     55  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     56  *    or promote products derived from this software without specific prior
     57  *    written permission.
     58  *
     59  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     61  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     62  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     63  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     64  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     65  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     66  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     67  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     68  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     69  * POSSIBILITY OF SUCH DAMAGE.
     70  */
     71 
     72 /*
     73  * Device driver for the Intel 82597EX Ten Gigabit Ethernet controller.
     74  *
     75  * TODO (in no specific order):
     76  *	HW VLAN support.
     77  *	TSE offloading (needs kernel changes...)
     78  *	RAIDC (receive interrupt delay adaptation)
     79  *	Use memory > 4GB.
     80  */
     81 
     82 #include <sys/cdefs.h>
     83 __KERNEL_RCSID(0, "$NetBSD: if_dge.c,v 1.38.4.5 2017/02/05 13:40:29 skrll Exp $");
     84 
     85 #include <sys/param.h>
     86 #include <sys/systm.h>
     87 #include <sys/callout.h>
     88 #include <sys/mbuf.h>
     89 #include <sys/malloc.h>
     90 #include <sys/kernel.h>
     91 #include <sys/socket.h>
     92 #include <sys/ioctl.h>
     93 #include <sys/errno.h>
     94 #include <sys/device.h>
     95 #include <sys/queue.h>
     96 
     97 #include <sys/rndsource.h>
     98 
     99 #include <net/if.h>
    100 #include <net/if_dl.h>
    101 #include <net/if_media.h>
    102 #include <net/if_ether.h>
    103 
    104 #include <net/bpf.h>
    105 
    106 #include <netinet/in.h>			/* XXX for struct ip */
    107 #include <netinet/in_systm.h>		/* XXX for struct ip */
    108 #include <netinet/ip.h>			/* XXX for struct ip */
    109 #include <netinet/tcp.h>		/* XXX for struct tcphdr */
    110 
    111 #include <sys/bus.h>
    112 #include <sys/intr.h>
    113 #include <machine/endian.h>
    114 
    115 #include <dev/mii/mii.h>
    116 #include <dev/mii/miivar.h>
    117 #include <dev/mii/mii_bitbang.h>
    118 
    119 #include <dev/pci/pcireg.h>
    120 #include <dev/pci/pcivar.h>
    121 #include <dev/pci/pcidevs.h>
    122 
    123 #include <dev/pci/if_dgereg.h>
    124 
    125 /*
    126  * The receive engine may sometimes become off-by-one when writing back
    127  * chained descriptors.	 Avoid this by allocating a large chunk of
    128  * memory and use if instead (to avoid chained descriptors).
    129  * This only happens with chained descriptors under heavy load.
    130  */
    131 #define DGE_OFFBYONE_RXBUG
    132 
    133 #define DGE_EVENT_COUNTERS
    134 #define DGE_DEBUG
    135 
    136 #ifdef DGE_DEBUG
    137 #define DGE_DEBUG_LINK		0x01
    138 #define DGE_DEBUG_TX		0x02
    139 #define DGE_DEBUG_RX		0x04
    140 #define DGE_DEBUG_CKSUM		0x08
    141 int	dge_debug = 0;
    142 
    143 #define DPRINTF(x, y)	if (dge_debug & (x)) printf y
    144 #else
    145 #define DPRINTF(x, y)	/* nothing */
    146 #endif /* DGE_DEBUG */
    147 
    148 /*
    149  * Transmit descriptor list size. We allow up to 100 DMA segments per
    150  * packet (Intel reports of jumbo frame packets with as
    151  * many as 80 DMA segments when using 16k buffers).
    152  */
    153 #define DGE_NTXSEGS		100
    154 #define DGE_IFQUEUELEN		20000
    155 #define DGE_TXQUEUELEN		2048
    156 #define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
    157 #define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
    158 #define DGE_NTXDESC		1024
    159 #define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
    160 #define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
    161 #define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
    162 
    163 /*
    164  * Receive descriptor list size.
    165  * Packet is of size MCLBYTES, and for jumbo packets buffers may
    166  * be chained.	Due to the nature of the card (high-speed), keep this
    167  * ring large. With 2k buffers the ring can store 400 jumbo packets,
    168  * which at full speed will be received in just under 3ms.
    169  */
    170 #define DGE_NRXDESC		2048
    171 #define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
    172 #define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
    173 /*
    174  * # of descriptors between head and written descriptors.
    175  * This is to work-around two erratas.
    176  */
    177 #define DGE_RXSPACE		10
    178 #define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
    179 /*
    180  * Receive descriptor fetch threshholds. These are values recommended
    181  * by Intel, do not touch them unless you know what you are doing.
    182  */
    183 #define RXDCTL_PTHRESH_VAL	128
    184 #define RXDCTL_HTHRESH_VAL	16
    185 #define RXDCTL_WTHRESH_VAL	16
    186 
    187 
    188 /*
    189  * Tweakable parameters; default values.
    190  */
    191 #define FCRTH	0x30000 /* Send XOFF water mark */
    192 #define FCRTL	0x28000 /* Send XON water mark */
    193 #define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
    194 #define TIDV	0x20	/* Interrupt delay after send, .8192us units */
    195 
    196 /*
    197  * Control structures are DMA'd to the i82597 chip.  We allocate them in
    198  * a single clump that maps to a single DMA segment to make serveral things
    199  * easier.
    200  */
    201 struct dge_control_data {
    202 	/*
    203 	 * The transmit descriptors.
    204 	 */
    205 	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
    206 
    207 	/*
    208 	 * The receive descriptors.
    209 	 */
    210 	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
    211 };
    212 
    213 #define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
    214 #define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
    215 #define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
    216 
    217 /*
    218  * The DGE interface have a higher max MTU size than normal jumbo frames.
    219  */
    220 #define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
    221 
    222 /*
    223  * Software state for transmit jobs.
    224  */
    225 struct dge_txsoft {
    226 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
    227 	bus_dmamap_t txs_dmamap;	/* our DMA map */
    228 	int txs_firstdesc;		/* first descriptor in packet */
    229 	int txs_lastdesc;		/* last descriptor in packet */
    230 	int txs_ndesc;			/* # of descriptors used */
    231 };
    232 
    233 /*
    234  * Software state for receive buffers.	Each descriptor gets a
    235  * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
    236  * more than one buffer, we chain them together.
    237  */
    238 struct dge_rxsoft {
    239 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
    240 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
    241 };
    242 
    243 /*
    244  * Software state per device.
    245  */
    246 struct dge_softc {
    247 	device_t sc_dev;		/* generic device information */
    248 	bus_space_tag_t sc_st;		/* bus space tag */
    249 	bus_space_handle_t sc_sh;	/* bus space handle */
    250 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
    251 	struct ethercom sc_ethercom;	/* ethernet common data */
    252 
    253 	int sc_flags;			/* flags; see below */
    254 	int sc_bus_speed;		/* PCI/PCIX bus speed */
    255 	int sc_pcix_offset;		/* PCIX capability register offset */
    256 
    257 	const struct dge_product *sc_dgep; /* Pointer to the dge_product entry */
    258 	pci_chipset_tag_t sc_pc;
    259 	pcitag_t sc_pt;
    260 	int sc_mmrbc;			/* Max PCIX memory read byte count */
    261 
    262 	void *sc_ih;			/* interrupt cookie */
    263 
    264 	struct ifmedia sc_media;
    265 
    266 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
    267 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
    268 
    269 	int		sc_align_tweak;
    270 
    271 	/*
    272 	 * Software state for the transmit and receive descriptors.
    273 	 */
    274 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
    275 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
    276 
    277 	/*
    278 	 * Control data structures.
    279 	 */
    280 	struct dge_control_data *sc_control_data;
    281 #define sc_txdescs	sc_control_data->wcd_txdescs
    282 #define sc_rxdescs	sc_control_data->wcd_rxdescs
    283 
    284 #ifdef DGE_EVENT_COUNTERS
    285 	/* Event counters. */
    286 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
    287 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
    288 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
    289 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
    290 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
    291 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
    292 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
    293 
    294 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
    295 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
    296 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
    297 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
    298 
    299 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
    300 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
    301 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
    302 
    303 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
    304 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
    305 #endif /* DGE_EVENT_COUNTERS */
    306 
    307 	int	sc_txfree;		/* number of free Tx descriptors */
    308 	int	sc_txnext;		/* next ready Tx descriptor */
    309 
    310 	int	sc_txsfree;		/* number of free Tx jobs */
    311 	int	sc_txsnext;		/* next free Tx job */
    312 	int	sc_txsdirty;		/* dirty Tx jobs */
    313 
    314 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
    315 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
    316 
    317 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
    318 	int	sc_rxdiscard;
    319 	int	sc_rxlen;
    320 	struct mbuf *sc_rxhead;
    321 	struct mbuf *sc_rxtail;
    322 	struct mbuf **sc_rxtailp;
    323 
    324 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
    325 	uint32_t sc_icr;		/* prototype interrupt bits */
    326 	uint32_t sc_tctl;		/* prototype TCTL register */
    327 	uint32_t sc_rctl;		/* prototype RCTL register */
    328 
    329 	int sc_mchash_type;		/* multicast filter offset */
    330 
    331 	uint16_t sc_eeprom[EEPROM_SIZE];
    332 
    333 	krndsource_t rnd_source; /* random source */
    334 #ifdef DGE_OFFBYONE_RXBUG
    335 	void *sc_bugbuf;
    336 	SLIST_HEAD(, rxbugentry) sc_buglist;
    337 	bus_dmamap_t sc_bugmap;
    338 	struct rxbugentry *sc_entry;
    339 #endif
    340 };
    341 
    342 #define DGE_RXCHAIN_RESET(sc)						\
    343 do {									\
    344 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
    345 	*(sc)->sc_rxtailp = NULL;					\
    346 	(sc)->sc_rxlen = 0;						\
    347 } while (/*CONSTCOND*/0)
    348 
    349 #define DGE_RXCHAIN_LINK(sc, m)						\
    350 do {									\
    351 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
    352 	(sc)->sc_rxtailp = &(m)->m_next;				\
    353 } while (/*CONSTCOND*/0)
    354 
    355 /* sc_flags */
    356 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
    357 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
    358 
    359 #ifdef DGE_EVENT_COUNTERS
    360 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
    361 #else
    362 #define DGE_EVCNT_INCR(ev)	/* nothing */
    363 #endif
    364 
    365 #define CSR_READ(sc, reg)						\
    366 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
    367 #define CSR_WRITE(sc, reg, val)						\
    368 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
    369 
    370 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
    371 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
    372 
    373 #define DGE_CDTXSYNC(sc, x, n, ops)					\
    374 do {									\
    375 	int __x, __n;							\
    376 									\
    377 	__x = (x);							\
    378 	__n = (n);							\
    379 									\
    380 	/* If it will wrap around, sync to the end of the ring. */	\
    381 	if ((__x + __n) > DGE_NTXDESC) {				\
    382 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
    383 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
    384 		    (DGE_NTXDESC - __x), (ops));			\
    385 		__n -= (DGE_NTXDESC - __x);				\
    386 		__x = 0;						\
    387 	}								\
    388 									\
    389 	/* Now sync whatever is left. */				\
    390 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
    391 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
    392 } while (/*CONSTCOND*/0)
    393 
    394 #define DGE_CDRXSYNC(sc, x, ops)						\
    395 do {									\
    396 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
    397 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
    398 } while (/*CONSTCOND*/0)
    399 
    400 #ifdef DGE_OFFBYONE_RXBUG
    401 #define DGE_INIT_RXDESC(sc, x)						\
    402 do {									\
    403 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
    404 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
    405 	struct mbuf *__m = __rxs->rxs_mbuf;				\
    406 									\
    407 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
    408 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
    409 	__rxd->dr_baddrh = 0;						\
    410 	__rxd->dr_len = 0;						\
    411 	__rxd->dr_cksum = 0;						\
    412 	__rxd->dr_status = 0;						\
    413 	__rxd->dr_errors = 0;						\
    414 	__rxd->dr_special = 0;						\
    415 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
    416 									\
    417 	CSR_WRITE((sc), DGE_RDT, (x));					\
    418 } while (/*CONSTCOND*/0)
    419 #else
    420 #define DGE_INIT_RXDESC(sc, x)						\
    421 do {									\
    422 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
    423 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
    424 	struct mbuf *__m = __rxs->rxs_mbuf;				\
    425 									\
    426 	/*								\
    427 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
    428 	 * so that the payload after the Ethernet header is aligned	\
    429 	 * to a 4-byte boundary.					\
    430 	 *								\
    431 	 * XXX BRAINDAMAGE ALERT!					\
    432 	 * The stupid chip uses the same size for every buffer, which	\
    433 	 * is set in the Receive Control register.  We are using the 2K \
    434 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
    435 	 * reason, we can't "scoot" packets longer than the standard	\
    436 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
    437 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
    438 	 * the upper layer copy the headers.				\
    439 	 */								\
    440 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
    441 									\
    442 	__rxd->dr_baddrl =						\
    443 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
    444 		(sc)->sc_align_tweak);					\
    445 	__rxd->dr_baddrh = 0;						\
    446 	__rxd->dr_len = 0;						\
    447 	__rxd->dr_cksum = 0;						\
    448 	__rxd->dr_status = 0;						\
    449 	__rxd->dr_errors = 0;						\
    450 	__rxd->dr_special = 0;						\
    451 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
    452 									\
    453 	CSR_WRITE((sc), DGE_RDT, (x));					\
    454 } while (/*CONSTCOND*/0)
    455 #endif
    456 
    457 #ifdef DGE_OFFBYONE_RXBUG
    458 /*
    459  * Allocation constants.  Much memory may be used for this.
    460  */
    461 #ifndef DGE_BUFFER_SIZE
    462 #define DGE_BUFFER_SIZE DGE_MAX_MTU
    463 #endif
    464 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
    465 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
    466 
    467 struct rxbugentry {
    468 	SLIST_ENTRY(rxbugentry) rb_entry;
    469 	int rb_slot;
    470 };
    471 
    472 static int
    473 dge_alloc_rcvmem(struct dge_softc *sc)
    474 {
    475 	char *kva;
    476 	bus_dma_segment_t seg;
    477 	int i, rseg, state, error;
    478 	struct rxbugentry *entry;
    479 
    480 	state = error = 0;
    481 
    482 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
    483 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
    484 		aprint_error_dev(sc->sc_dev, "can't alloc rx buffers\n");
    485 		return ENOBUFS;
    486 	}
    487 
    488 	state = 1;
    489 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
    490 	    BUS_DMA_NOWAIT)) {
    491 		aprint_error_dev(sc->sc_dev, "can't map DMA buffers (%d bytes)\n",
    492 		    (int)DGE_RXMEM);
    493 		error = ENOBUFS;
    494 		goto out;
    495 	}
    496 
    497 	state = 2;
    498 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
    499 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
    500 		aprint_error_dev(sc->sc_dev, "can't create DMA map\n");
    501 		error = ENOBUFS;
    502 		goto out;
    503 	}
    504 
    505 	state = 3;
    506 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
    507 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
    508 		aprint_error_dev(sc->sc_dev, "can't load DMA map\n");
    509 		error = ENOBUFS;
    510 		goto out;
    511 	}
    512 
    513 	state = 4;
    514 	sc->sc_bugbuf = (void *)kva;
    515 	SLIST_INIT(&sc->sc_buglist);
    516 
    517 	/*
    518 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
    519 	 * in an array.
    520 	 */
    521 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
    522 	    M_DEVBUF, M_NOWAIT)) == NULL) {
    523 		error = ENOBUFS;
    524 		goto out;
    525 	}
    526 	sc->sc_entry = entry;
    527 	for (i = 0; i < DGE_NBUFFERS; i++) {
    528 		entry[i].rb_slot = i;
    529 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
    530 	}
    531 out:
    532 	if (error != 0) {
    533 		switch (state) {
    534 		case 4:
    535 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
    536 		case 3:
    537 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
    538 		case 2:
    539 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
    540 		case 1:
    541 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
    542 			break;
    543 		default:
    544 			break;
    545 		}
    546 	}
    547 
    548 	return error;
    549 }
    550 
    551 /*
    552  * Allocate a jumbo buffer.
    553  */
    554 static void *
    555 dge_getbuf(struct dge_softc *sc)
    556 {
    557 	struct rxbugentry *entry;
    558 
    559 	entry = SLIST_FIRST(&sc->sc_buglist);
    560 
    561 	if (entry == NULL) {
    562 		printf("%s: no free RX buffers\n", device_xname(sc->sc_dev));
    563 		return(NULL);
    564 	}
    565 
    566 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
    567 	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
    568 }
    569 
    570 /*
    571  * Release a jumbo buffer.
    572  */
    573 static void
    574 dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
    575 {
    576 	struct rxbugentry *entry;
    577 	struct dge_softc *sc;
    578 	int i, s;
    579 
    580 	/* Extract the softc struct pointer. */
    581 	sc = (struct dge_softc *)arg;
    582 
    583 	if (sc == NULL)
    584 		panic("dge_freebuf: can't find softc pointer!");
    585 
    586 	/* calculate the slot this buffer belongs to */
    587 
    588 	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
    589 
    590 	if ((i < 0) || (i >= DGE_NBUFFERS))
    591 		panic("dge_freebuf: asked to free buffer %d!", i);
    592 
    593 	s = splvm();
    594 	entry = sc->sc_entry + i;
    595 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
    596 
    597 	if (__predict_true(m != NULL))
    598 		pool_cache_put(mb_cache, m);
    599 	splx(s);
    600 }
    601 #endif
    602 
    603 static void	dge_start(struct ifnet *);
    604 static void	dge_watchdog(struct ifnet *);
    605 static int	dge_ioctl(struct ifnet *, u_long, void *);
    606 static int	dge_init(struct ifnet *);
    607 static void	dge_stop(struct ifnet *, int);
    608 
    609 static bool	dge_shutdown(device_t, int);
    610 
    611 static void	dge_reset(struct dge_softc *);
    612 static void	dge_rxdrain(struct dge_softc *);
    613 static int	dge_add_rxbuf(struct dge_softc *, int);
    614 
    615 static void	dge_set_filter(struct dge_softc *);
    616 
    617 static int	dge_intr(void *);
    618 static void	dge_txintr(struct dge_softc *);
    619 static void	dge_rxintr(struct dge_softc *);
    620 static void	dge_linkintr(struct dge_softc *, uint32_t);
    621 
    622 static int	dge_match(device_t, cfdata_t, void *);
    623 static void	dge_attach(device_t, device_t, void *);
    624 
    625 static int	dge_read_eeprom(struct dge_softc *sc);
    626 static int	dge_eeprom_clockin(struct dge_softc *sc);
    627 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
    628 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
    629 static int	dge_xgmii_mediachange(struct ifnet *);
    630 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
    631 static void	dge_xgmii_reset(struct dge_softc *);
    632 static void	dge_xgmii_writereg(struct dge_softc *, int, int, int);
    633 
    634 
    635 CFATTACH_DECL_NEW(dge, sizeof(struct dge_softc),
    636     dge_match, dge_attach, NULL, NULL);
    637 
    638 #ifdef DGE_EVENT_COUNTERS
    639 #if DGE_NTXSEGS > 100
    640 #error Update dge_txseg_evcnt_names
    641 #endif
    642 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
    643 #endif /* DGE_EVENT_COUNTERS */
    644 
    645 /*
    646  * Devices supported by this driver.
    647  */
    648 static const struct dge_product {
    649   pci_vendor_id_t      dgep_vendor;
    650   pci_product_id_t  dgep_product;
    651   const char     *dgep_name;
    652   int         dgep_flags;
    653 #define DGEP_F_10G_LR     0x01
    654 #define DGEP_F_10G_SR     0x02
    655 } dge_products[] = {
    656   { PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX,
    657     "Intel i82597EX 10GbE-LR Ethernet",
    658     DGEP_F_10G_LR },
    659 
    660   { PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX_SR,
    661     "Intel i82597EX 10GbE-SR Ethernet",
    662     DGEP_F_10G_SR },
    663 
    664   { 0,        0,
    665     NULL,
    666     0 },
    667 };
    668 
    669 static const struct dge_product *
    670 dge_lookup(const struct pci_attach_args *pa)
    671 {
    672 	const struct dge_product *dgep;
    673 
    674 	for (dgep = dge_products; dgep->dgep_name != NULL; dgep++) {
    675 		if (PCI_VENDOR(pa->pa_id) == dgep->dgep_vendor &&
    676 		    PCI_PRODUCT(pa->pa_id) == dgep->dgep_product)
    677 			return dgep;
    678 		}
    679 	return NULL;
    680 }
    681 
    682 static int
    683 dge_match(device_t parent, cfdata_t cf, void *aux)
    684 {
    685 	struct pci_attach_args *pa = aux;
    686 
    687 	if (dge_lookup(pa) != NULL)
    688 		return (1);
    689 
    690 	return (0);
    691 }
    692 
    693 static void
    694 dge_attach(device_t parent, device_t self, void *aux)
    695 {
    696 	struct dge_softc *sc = device_private(self);
    697 	struct pci_attach_args *pa = aux;
    698 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
    699 	pci_chipset_tag_t pc = pa->pa_pc;
    700 	pci_intr_handle_t ih;
    701 	const char *intrstr = NULL;
    702 	bus_dma_segment_t seg;
    703 	int i, rseg, error;
    704 	uint8_t enaddr[ETHER_ADDR_LEN];
    705 	pcireg_t preg, memtype;
    706 	uint32_t reg;
    707 	char intrbuf[PCI_INTRSTR_LEN];
    708 	const struct dge_product *dgep;
    709 
    710 	sc->sc_dgep = dgep = dge_lookup(pa);
    711 	if (dgep == NULL) {
    712 		printf("\n");
    713 		panic("dge_attach: impossible");
    714 	}
    715 
    716 	sc->sc_dev = self;
    717 	sc->sc_dmat = pa->pa_dmat;
    718 	sc->sc_pc = pa->pa_pc;
    719 	sc->sc_pt = pa->pa_tag;
    720 
    721 	pci_aprint_devinfo_fancy(pa, "Ethernet controller",
    722 		dgep->dgep_name, 1);
    723 
    724 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
    725         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
    726             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
    727                 aprint_error_dev(sc->sc_dev,
    728 		    "unable to map device registers\n");
    729                 return;
    730         }
    731 
    732 	/* Enable bus mastering */
    733 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
    734 	preg |= PCI_COMMAND_MASTER_ENABLE;
    735 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
    736 
    737 	/*
    738 	 * Map and establish our interrupt.
    739 	 */
    740 	if (pci_intr_map(pa, &ih)) {
    741 		aprint_error_dev(sc->sc_dev, "unable to map interrupt\n");
    742 		return;
    743 	}
    744 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
    745 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
    746 	if (sc->sc_ih == NULL) {
    747 		aprint_error_dev(sc->sc_dev, "unable to establish interrupt");
    748 		if (intrstr != NULL)
    749 			aprint_error(" at %s", intrstr);
    750 		aprint_error("\n");
    751 		return;
    752 	}
    753 	aprint_normal_dev(sc->sc_dev, "interrupting at %s\n", intrstr);
    754 
    755 	/*
    756 	 * Determine a few things about the bus we're connected to.
    757 	 */
    758 	reg = CSR_READ(sc, DGE_STATUS);
    759 	if (reg & STATUS_BUS64)
    760 		sc->sc_flags |= DGE_F_BUS64;
    761 
    762 	sc->sc_flags |= DGE_F_PCIX;
    763 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
    764 			       PCI_CAP_PCIX,
    765 			       &sc->sc_pcix_offset, NULL) == 0)
    766 		aprint_error_dev(sc->sc_dev, "unable to find PCIX "
    767 		    "capability\n");
    768 
    769 	if (sc->sc_flags & DGE_F_PCIX) {
    770 		switch (reg & STATUS_PCIX_MSK) {
    771 		case STATUS_PCIX_66:
    772 			sc->sc_bus_speed = 66;
    773 			break;
    774 		case STATUS_PCIX_100:
    775 			sc->sc_bus_speed = 100;
    776 			break;
    777 		case STATUS_PCIX_133:
    778 			sc->sc_bus_speed = 133;
    779 			break;
    780 		default:
    781 			aprint_error_dev(sc->sc_dev,
    782 			    "unknown PCIXSPD %d; assuming 66MHz\n",
    783 			    reg & STATUS_PCIX_MSK);
    784 			sc->sc_bus_speed = 66;
    785 		}
    786 	} else
    787 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
    788 	aprint_verbose_dev(sc->sc_dev, "%d-bit %dMHz %s bus\n",
    789 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
    790 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
    791 
    792 	/*
    793 	 * Allocate the control data structures, and create and load the
    794 	 * DMA map for it.
    795 	 */
    796 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    797 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
    798 	    0)) != 0) {
    799 		aprint_error_dev(sc->sc_dev,
    800 		    "unable to allocate control data, error = %d\n",
    801 		    error);
    802 		goto fail_0;
    803 	}
    804 
    805 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
    806 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
    807 	    0)) != 0) {
    808 		aprint_error_dev(sc->sc_dev, "unable to map control data, error = %d\n",
    809 		    error);
    810 		goto fail_1;
    811 	}
    812 
    813 	if ((error = bus_dmamap_create(sc->sc_dmat,
    814 	    sizeof(struct dge_control_data), 1,
    815 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
    816 		aprint_error_dev(sc->sc_dev, "unable to create control data DMA map, "
    817 		    "error = %d\n", error);
    818 		goto fail_2;
    819 	}
    820 
    821 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
    822 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
    823 	    0)) != 0) {
    824 		aprint_error_dev(sc->sc_dev,
    825 		    "unable to load control data DMA map, error = %d\n",
    826 		    error);
    827 		goto fail_3;
    828 	}
    829 
    830 #ifdef DGE_OFFBYONE_RXBUG
    831 	if (dge_alloc_rcvmem(sc) != 0)
    832 		return; /* Already complained */
    833 #endif
    834 	/*
    835 	 * Create the transmit buffer DMA maps.
    836 	 */
    837 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
    838 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
    839 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
    840 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
    841 			aprint_error_dev(sc->sc_dev, "unable to create Tx DMA map %d, "
    842 			    "error = %d\n", i, error);
    843 			goto fail_4;
    844 		}
    845 	}
    846 
    847 	/*
    848 	 * Create the receive buffer DMA maps.
    849 	 */
    850 	for (i = 0; i < DGE_NRXDESC; i++) {
    851 #ifdef DGE_OFFBYONE_RXBUG
    852 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
    853 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    854 #else
    855 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
    856 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    857 #endif
    858 			aprint_error_dev(sc->sc_dev, "unable to create Rx DMA map %d, "
    859 			    "error = %d\n", i, error);
    860 			goto fail_5;
    861 		}
    862 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
    863 	}
    864 
    865 	/*
    866 	 * Set bits in ctrl0 register.
    867 	 * Should get the software defined pins out of EEPROM?
    868 	 */
    869 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
    870 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
    871 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
    872 
    873 	/*
    874 	 * Reset the chip to a known state.
    875 	 */
    876 	dge_reset(sc);
    877 
    878 	/*
    879 	 * Reset the PHY.
    880 	 */
    881 	dge_xgmii_reset(sc);
    882 
    883 	/*
    884 	 * Read in EEPROM data.
    885 	 */
    886 	if (dge_read_eeprom(sc)) {
    887 		aprint_error_dev(sc->sc_dev, "couldn't read EEPROM\n");
    888 		return;
    889 	}
    890 
    891 	/*
    892 	 * Get the ethernet address.
    893 	 */
    894 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
    895 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
    896 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
    897 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
    898 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
    899 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
    900 
    901 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    902 	    ether_sprintf(enaddr));
    903 
    904 	/*
    905 	 * Setup media stuff.
    906 	 */
    907         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
    908             dge_xgmii_mediastatus);
    909 	if (dgep->dgep_flags & DGEP_F_10G_SR) {
    910 		ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_SR, 0, NULL);
    911 		ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_SR);
    912 	} else { /* XXX default is LR */
    913 		ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
    914 		ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
    915 	}
    916 
    917 	ifp = &sc->sc_ethercom.ec_if;
    918 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    919 	ifp->if_softc = sc;
    920 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    921 	ifp->if_ioctl = dge_ioctl;
    922 	ifp->if_start = dge_start;
    923 	ifp->if_watchdog = dge_watchdog;
    924 	ifp->if_init = dge_init;
    925 	ifp->if_stop = dge_stop;
    926 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
    927 	IFQ_SET_READY(&ifp->if_snd);
    928 
    929 	sc->sc_ethercom.ec_capabilities |=
    930 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
    931 
    932 	/*
    933 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
    934 	 */
    935 	ifp->if_capabilities |=
    936 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    937 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    938 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
    939 
    940 	/*
    941 	 * Attach the interface.
    942 	 */
    943 	if_attach(ifp);
    944 	if_deferred_start_init(ifp, NULL);
    945 	ether_ifattach(ifp, enaddr);
    946 	rnd_attach_source(&sc->rnd_source, device_xname(sc->sc_dev),
    947 	    RND_TYPE_NET, RND_FLAG_DEFAULT);
    948 
    949 #ifdef DGE_EVENT_COUNTERS
    950 	/* Fix segment event naming */
    951 	if (dge_txseg_evcnt_names == NULL) {
    952 		dge_txseg_evcnt_names =
    953 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
    954 		for (i = 0; i < DGE_NTXSEGS; i++)
    955 			snprintf((*dge_txseg_evcnt_names)[i],
    956 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
    957 	}
    958 
    959 	/* Attach event counters. */
    960 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
    961 	    NULL, device_xname(sc->sc_dev), "txsstall");
    962 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
    963 	    NULL, device_xname(sc->sc_dev), "txdstall");
    964 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
    965 	    NULL, device_xname(sc->sc_dev), "txforceintr");
    966 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
    967 	    NULL, device_xname(sc->sc_dev), "txdw");
    968 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
    969 	    NULL, device_xname(sc->sc_dev), "txqe");
    970 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
    971 	    NULL, device_xname(sc->sc_dev), "rxintr");
    972 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
    973 	    NULL, device_xname(sc->sc_dev), "linkintr");
    974 
    975 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
    976 	    NULL, device_xname(sc->sc_dev), "rxipsum");
    977 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
    978 	    NULL, device_xname(sc->sc_dev), "rxtusum");
    979 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
    980 	    NULL, device_xname(sc->sc_dev), "txipsum");
    981 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
    982 	    NULL, device_xname(sc->sc_dev), "txtusum");
    983 
    984 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
    985 	    NULL, device_xname(sc->sc_dev), "txctx init");
    986 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
    987 	    NULL, device_xname(sc->sc_dev), "txctx hit");
    988 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
    989 	    NULL, device_xname(sc->sc_dev), "txctx miss");
    990 
    991 	for (i = 0; i < DGE_NTXSEGS; i++)
    992 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
    993 		    NULL, device_xname(sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
    994 
    995 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
    996 	    NULL, device_xname(sc->sc_dev), "txdrop");
    997 
    998 #endif /* DGE_EVENT_COUNTERS */
    999 
   1000 	/*
   1001 	 * Make sure the interface is shutdown during reboot.
   1002 	 */
   1003 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
   1004 		pmf_class_network_register(self, ifp);
   1005 	else
   1006 		aprint_error_dev(self, "couldn't establish power handler\n");
   1007 
   1008 	return;
   1009 
   1010 	/*
   1011 	 * Free any resources we've allocated during the failed attach
   1012 	 * attempt.  Do this in reverse order and fall through.
   1013 	 */
   1014  fail_5:
   1015 	for (i = 0; i < DGE_NRXDESC; i++) {
   1016 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
   1017 			bus_dmamap_destroy(sc->sc_dmat,
   1018 			    sc->sc_rxsoft[i].rxs_dmamap);
   1019 	}
   1020  fail_4:
   1021 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
   1022 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
   1023 			bus_dmamap_destroy(sc->sc_dmat,
   1024 			    sc->sc_txsoft[i].txs_dmamap);
   1025 	}
   1026 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
   1027  fail_3:
   1028 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
   1029  fail_2:
   1030 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
   1031 	    sizeof(struct dge_control_data));
   1032  fail_1:
   1033 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
   1034  fail_0:
   1035 	return;
   1036 }
   1037 
   1038 /*
   1039  * dge_shutdown:
   1040  *
   1041  *	Make sure the interface is stopped at reboot time.
   1042  */
   1043 static bool
   1044 dge_shutdown(device_t self, int howto)
   1045 {
   1046 	struct dge_softc *sc;
   1047 
   1048 	sc = device_private(self);
   1049 	dge_stop(&sc->sc_ethercom.ec_if, 1);
   1050 
   1051 	return true;
   1052 }
   1053 
   1054 /*
   1055  * dge_tx_cksum:
   1056  *
   1057  *	Set up TCP/IP checksumming parameters for the
   1058  *	specified packet.
   1059  */
   1060 static int
   1061 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
   1062 {
   1063 	struct mbuf *m0 = txs->txs_mbuf;
   1064 	struct dge_ctdes *t;
   1065 	uint32_t ipcs, tucs;
   1066 	struct ether_header *eh;
   1067 	int offset, iphl;
   1068 	uint8_t fields = 0;
   1069 
   1070 	/*
   1071 	 * XXX It would be nice if the mbuf pkthdr had offset
   1072 	 * fields for the protocol headers.
   1073 	 */
   1074 
   1075 	eh = mtod(m0, struct ether_header *);
   1076 	switch (htons(eh->ether_type)) {
   1077 	case ETHERTYPE_IP:
   1078 		offset = ETHER_HDR_LEN;
   1079 		break;
   1080 
   1081 	case ETHERTYPE_VLAN:
   1082 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
   1083 		break;
   1084 
   1085 	default:
   1086 		/*
   1087 		 * Don't support this protocol or encapsulation.
   1088 		 */
   1089 		*fieldsp = 0;
   1090 		return (0);
   1091 	}
   1092 
   1093 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
   1094 
   1095 	/*
   1096 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
   1097 	 * offload feature, if we load the context descriptor, we
   1098 	 * MUST provide valid values for IPCSS and TUCSS fields.
   1099 	 */
   1100 
   1101 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
   1102 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
   1103 		fields |= TDESC_POPTS_IXSM;
   1104 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1105 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1106 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1107 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
   1108 		/* Use the cached value. */
   1109 		ipcs = sc->sc_txctx_ipcs;
   1110 	} else {
   1111 		/* Just initialize it to the likely value anyway. */
   1112 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1113 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1114 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1115 	}
   1116 	DPRINTF(DGE_DEBUG_CKSUM,
   1117 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
   1118 	    device_xname(sc->sc_dev), offset, ipcs));
   1119 
   1120 	offset += iphl;
   1121 
   1122 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
   1123 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
   1124 		fields |= TDESC_POPTS_TXSM;
   1125 		tucs = DGE_TCPIP_TUCSS(offset) |
   1126 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
   1127 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1128 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
   1129 		/* Use the cached value. */
   1130 		tucs = sc->sc_txctx_tucs;
   1131 	} else {
   1132 		/* Just initialize it to a valid TCP context. */
   1133 		tucs = DGE_TCPIP_TUCSS(offset) |
   1134 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
   1135 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1136 	}
   1137 
   1138 	DPRINTF(DGE_DEBUG_CKSUM,
   1139 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
   1140 	    device_xname(sc->sc_dev), offset, tucs));
   1141 
   1142 	if (sc->sc_txctx_ipcs == ipcs &&
   1143 	    sc->sc_txctx_tucs == tucs) {
   1144 		/* Cached context is fine. */
   1145 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
   1146 	} else {
   1147 		/* Fill in the context descriptor. */
   1148 #ifdef DGE_EVENT_COUNTERS
   1149 		if (sc->sc_txctx_ipcs == 0xffffffff &&
   1150 		    sc->sc_txctx_tucs == 0xffffffff)
   1151 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
   1152 		else
   1153 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
   1154 #endif
   1155 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
   1156 		t->dc_tcpip_ipcs = htole32(ipcs);
   1157 		t->dc_tcpip_tucs = htole32(tucs);
   1158 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
   1159 		t->dc_tcpip_seg = 0;
   1160 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
   1161 
   1162 		sc->sc_txctx_ipcs = ipcs;
   1163 		sc->sc_txctx_tucs = tucs;
   1164 
   1165 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
   1166 		txs->txs_ndesc++;
   1167 	}
   1168 
   1169 	*fieldsp = fields;
   1170 
   1171 	return (0);
   1172 }
   1173 
   1174 /*
   1175  * dge_start:		[ifnet interface function]
   1176  *
   1177  *	Start packet transmission on the interface.
   1178  */
   1179 static void
   1180 dge_start(struct ifnet *ifp)
   1181 {
   1182 	struct dge_softc *sc = ifp->if_softc;
   1183 	struct mbuf *m0;
   1184 	struct dge_txsoft *txs;
   1185 	bus_dmamap_t dmamap;
   1186 	int error, nexttx, lasttx = -1, ofree, seg;
   1187 	uint32_t cksumcmd;
   1188 	uint8_t cksumfields;
   1189 
   1190 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
   1191 		return;
   1192 
   1193 	/*
   1194 	 * Remember the previous number of free descriptors.
   1195 	 */
   1196 	ofree = sc->sc_txfree;
   1197 
   1198 	/*
   1199 	 * Loop through the send queue, setting up transmit descriptors
   1200 	 * until we drain the queue, or use up all available transmit
   1201 	 * descriptors.
   1202 	 */
   1203 	for (;;) {
   1204 		/* Grab a packet off the queue. */
   1205 		IFQ_POLL(&ifp->if_snd, m0);
   1206 		if (m0 == NULL)
   1207 			break;
   1208 
   1209 		DPRINTF(DGE_DEBUG_TX,
   1210 		    ("%s: TX: have packet to transmit: %p\n",
   1211 		    device_xname(sc->sc_dev), m0));
   1212 
   1213 		/* Get a work queue entry. */
   1214 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
   1215 			dge_txintr(sc);
   1216 			if (sc->sc_txsfree == 0) {
   1217 				DPRINTF(DGE_DEBUG_TX,
   1218 				    ("%s: TX: no free job descriptors\n",
   1219 					device_xname(sc->sc_dev)));
   1220 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
   1221 				break;
   1222 			}
   1223 		}
   1224 
   1225 		txs = &sc->sc_txsoft[sc->sc_txsnext];
   1226 		dmamap = txs->txs_dmamap;
   1227 
   1228 		/*
   1229 		 * Load the DMA map.  If this fails, the packet either
   1230 		 * didn't fit in the allotted number of segments, or we
   1231 		 * were short on resources.  For the too-many-segments
   1232 		 * case, we simply report an error and drop the packet,
   1233 		 * since we can't sanely copy a jumbo packet to a single
   1234 		 * buffer.
   1235 		 */
   1236 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
   1237 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1238 		if (error) {
   1239 			if (error == EFBIG) {
   1240 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
   1241 				printf("%s: Tx packet consumes too many "
   1242 				    "DMA segments, dropping...\n",
   1243 				    device_xname(sc->sc_dev));
   1244 				IFQ_DEQUEUE(&ifp->if_snd, m0);
   1245 				m_freem(m0);
   1246 				continue;
   1247 			}
   1248 			/*
   1249 			 * Short on resources, just stop for now.
   1250 			 */
   1251 			DPRINTF(DGE_DEBUG_TX,
   1252 			    ("%s: TX: dmamap load failed: %d\n",
   1253 			    device_xname(sc->sc_dev), error));
   1254 			break;
   1255 		}
   1256 
   1257 		/*
   1258 		 * Ensure we have enough descriptors free to describe
   1259 		 * the packet.  Note, we always reserve one descriptor
   1260 		 * at the end of the ring due to the semantics of the
   1261 		 * TDT register, plus one more in the event we need
   1262 		 * to re-load checksum offload context.
   1263 		 */
   1264 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
   1265 			/*
   1266 			 * Not enough free descriptors to transmit this
   1267 			 * packet.  We haven't committed anything yet,
   1268 			 * so just unload the DMA map, put the packet
   1269 			 * pack on the queue, and punt.  Notify the upper
   1270 			 * layer that there are no more slots left.
   1271 			 */
   1272 			DPRINTF(DGE_DEBUG_TX,
   1273 			    ("%s: TX: need %d descriptors, have %d\n",
   1274 			    device_xname(sc->sc_dev), dmamap->dm_nsegs,
   1275 			    sc->sc_txfree - 1));
   1276 			ifp->if_flags |= IFF_OACTIVE;
   1277 			bus_dmamap_unload(sc->sc_dmat, dmamap);
   1278 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
   1279 			break;
   1280 		}
   1281 
   1282 		IFQ_DEQUEUE(&ifp->if_snd, m0);
   1283 
   1284 		/*
   1285 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
   1286 		 */
   1287 
   1288 		/* Sync the DMA map. */
   1289 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
   1290 		    BUS_DMASYNC_PREWRITE);
   1291 
   1292 		DPRINTF(DGE_DEBUG_TX,
   1293 		    ("%s: TX: packet has %d DMA segments\n",
   1294 		    device_xname(sc->sc_dev), dmamap->dm_nsegs));
   1295 
   1296 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
   1297 
   1298 		/*
   1299 		 * Store a pointer to the packet so that we can free it
   1300 		 * later.
   1301 		 *
   1302 		 * Initially, we consider the number of descriptors the
   1303 		 * packet uses the number of DMA segments.  This may be
   1304 		 * incremented by 1 if we do checksum offload (a descriptor
   1305 		 * is used to set the checksum context).
   1306 		 */
   1307 		txs->txs_mbuf = m0;
   1308 		txs->txs_firstdesc = sc->sc_txnext;
   1309 		txs->txs_ndesc = dmamap->dm_nsegs;
   1310 
   1311 		/*
   1312 		 * Set up checksum offload parameters for
   1313 		 * this packet.
   1314 		 */
   1315 		if (m0->m_pkthdr.csum_flags &
   1316 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
   1317 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
   1318 				/* Error message already displayed. */
   1319 				bus_dmamap_unload(sc->sc_dmat, dmamap);
   1320 				continue;
   1321 			}
   1322 		} else {
   1323 			cksumfields = 0;
   1324 		}
   1325 
   1326 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
   1327 
   1328 		/*
   1329 		 * Initialize the transmit descriptor.
   1330 		 */
   1331 		for (nexttx = sc->sc_txnext, seg = 0;
   1332 		     seg < dmamap->dm_nsegs;
   1333 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
   1334 			/*
   1335 			 * Note: we currently only use 32-bit DMA
   1336 			 * addresses.
   1337 			 */
   1338 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
   1339 			sc->sc_txdescs[nexttx].dt_baddrl =
   1340 			    htole32(dmamap->dm_segs[seg].ds_addr);
   1341 			sc->sc_txdescs[nexttx].dt_ctl =
   1342 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
   1343 			sc->sc_txdescs[nexttx].dt_status = 0;
   1344 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
   1345 			sc->sc_txdescs[nexttx].dt_vlan = 0;
   1346 			lasttx = nexttx;
   1347 
   1348 			DPRINTF(DGE_DEBUG_TX,
   1349 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
   1350 			    device_xname(sc->sc_dev), nexttx,
   1351 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_addr),
   1352 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_len)));
   1353 		}
   1354 
   1355 		KASSERT(lasttx != -1);
   1356 
   1357 		/*
   1358 		 * Set up the command byte on the last descriptor of
   1359 		 * the packet.  If we're in the interrupt delay window,
   1360 		 * delay the interrupt.
   1361 		 */
   1362 		sc->sc_txdescs[lasttx].dt_ctl |=
   1363 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
   1364 
   1365 		txs->txs_lastdesc = lasttx;
   1366 
   1367 		DPRINTF(DGE_DEBUG_TX,
   1368 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(sc->sc_dev),
   1369 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
   1370 
   1371 		/* Sync the descriptors we're using. */
   1372 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
   1373 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1374 
   1375 		/* Give the packet to the chip. */
   1376 		CSR_WRITE(sc, DGE_TDT, nexttx);
   1377 
   1378 		DPRINTF(DGE_DEBUG_TX,
   1379 		    ("%s: TX: TDT -> %d\n", device_xname(sc->sc_dev), nexttx));
   1380 
   1381 		DPRINTF(DGE_DEBUG_TX,
   1382 		    ("%s: TX: finished transmitting packet, job %d\n",
   1383 		    device_xname(sc->sc_dev), sc->sc_txsnext));
   1384 
   1385 		/* Advance the tx pointer. */
   1386 		sc->sc_txfree -= txs->txs_ndesc;
   1387 		sc->sc_txnext = nexttx;
   1388 
   1389 		sc->sc_txsfree--;
   1390 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
   1391 
   1392 		/* Pass the packet to any BPF listeners. */
   1393 		bpf_mtap(ifp, m0);
   1394 	}
   1395 
   1396 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
   1397 		/* No more slots; notify upper layer. */
   1398 		ifp->if_flags |= IFF_OACTIVE;
   1399 	}
   1400 
   1401 	if (sc->sc_txfree != ofree) {
   1402 		/* Set a watchdog timer in case the chip flakes out. */
   1403 		ifp->if_timer = 5;
   1404 	}
   1405 }
   1406 
   1407 /*
   1408  * dge_watchdog:		[ifnet interface function]
   1409  *
   1410  *	Watchdog timer handler.
   1411  */
   1412 static void
   1413 dge_watchdog(struct ifnet *ifp)
   1414 {
   1415 	struct dge_softc *sc = ifp->if_softc;
   1416 
   1417 	/*
   1418 	 * Since we're using delayed interrupts, sweep up
   1419 	 * before we report an error.
   1420 	 */
   1421 	dge_txintr(sc);
   1422 
   1423 	if (sc->sc_txfree != DGE_NTXDESC) {
   1424 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
   1425 		    device_xname(sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
   1426 		    sc->sc_txnext);
   1427 		ifp->if_oerrors++;
   1428 
   1429 		/* Reset the interface. */
   1430 		(void) dge_init(ifp);
   1431 	}
   1432 
   1433 	/* Try to get more packets going. */
   1434 	dge_start(ifp);
   1435 }
   1436 
   1437 /*
   1438  * dge_ioctl:		[ifnet interface function]
   1439  *
   1440  *	Handle control requests from the operator.
   1441  */
   1442 static int
   1443 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   1444 {
   1445 	struct dge_softc *sc = ifp->if_softc;
   1446 	struct ifreq *ifr = (struct ifreq *) data;
   1447 	pcireg_t preg;
   1448 	int s, error, mmrbc;
   1449 
   1450 	s = splnet();
   1451 
   1452 	switch (cmd) {
   1453 	case SIOCSIFMEDIA:
   1454 	case SIOCGIFMEDIA:
   1455 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
   1456 		break;
   1457 
   1458 	case SIOCSIFMTU:
   1459 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
   1460 			error = EINVAL;
   1461 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
   1462 			break;
   1463 		else if (ifp->if_flags & IFF_UP)
   1464 			error = (*ifp->if_init)(ifp);
   1465 		else
   1466 			error = 0;
   1467 		break;
   1468 
   1469         case SIOCSIFFLAGS:
   1470 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   1471 			break;
   1472 		/* extract link flags */
   1473 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1474 		    (ifp->if_flags & IFF_LINK1) == 0)
   1475 			mmrbc = PCIX_MMRBC_512;
   1476 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1477 		    (ifp->if_flags & IFF_LINK1) != 0)
   1478 			mmrbc = PCIX_MMRBC_1024;
   1479 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
   1480 		    (ifp->if_flags & IFF_LINK1) == 0)
   1481 			mmrbc = PCIX_MMRBC_2048;
   1482 		else
   1483 			mmrbc = PCIX_MMRBC_4096;
   1484 		if (mmrbc != sc->sc_mmrbc) {
   1485 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
   1486 			preg &= ~PCIX_MMRBC_MSK;
   1487 			preg |= mmrbc;
   1488 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
   1489 			sc->sc_mmrbc = mmrbc;
   1490 		}
   1491                 /* FALLTHROUGH */
   1492 	default:
   1493 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
   1494 			break;
   1495 
   1496 		error = 0;
   1497 
   1498 		if (cmd == SIOCSIFCAP)
   1499 			error = (*ifp->if_init)(ifp);
   1500 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
   1501 			;
   1502 		else if (ifp->if_flags & IFF_RUNNING) {
   1503 			/*
   1504 			 * Multicast list has changed; set the hardware filter
   1505 			 * accordingly.
   1506 			 */
   1507 			dge_set_filter(sc);
   1508 		}
   1509 		break;
   1510 	}
   1511 
   1512 	/* Try to get more packets going. */
   1513 	dge_start(ifp);
   1514 
   1515 	splx(s);
   1516 	return (error);
   1517 }
   1518 
   1519 /*
   1520  * dge_intr:
   1521  *
   1522  *	Interrupt service routine.
   1523  */
   1524 static int
   1525 dge_intr(void *arg)
   1526 {
   1527 	struct dge_softc *sc = arg;
   1528 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1529 	uint32_t icr;
   1530 	int wantinit, handled = 0;
   1531 
   1532 	for (wantinit = 0; wantinit == 0;) {
   1533 		icr = CSR_READ(sc, DGE_ICR);
   1534 		if ((icr & sc->sc_icr) == 0)
   1535 			break;
   1536 
   1537 		rnd_add_uint32(&sc->rnd_source, icr);
   1538 
   1539 		handled = 1;
   1540 
   1541 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1542 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
   1543 			DPRINTF(DGE_DEBUG_RX,
   1544 			    ("%s: RX: got Rx intr 0x%08x\n",
   1545 			    device_xname(sc->sc_dev),
   1546 			    icr & (ICR_RXDMT0|ICR_RXT0)));
   1547 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
   1548 		}
   1549 #endif
   1550 		dge_rxintr(sc);
   1551 
   1552 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1553 		if (icr & ICR_TXDW) {
   1554 			DPRINTF(DGE_DEBUG_TX,
   1555 			    ("%s: TX: got TXDW interrupt\n",
   1556 			    device_xname(sc->sc_dev)));
   1557 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
   1558 		}
   1559 		if (icr & ICR_TXQE)
   1560 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
   1561 #endif
   1562 		dge_txintr(sc);
   1563 
   1564 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
   1565 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
   1566 			dge_linkintr(sc, icr);
   1567 		}
   1568 
   1569 		if (icr & ICR_RXO) {
   1570 			printf("%s: Receive overrun\n", device_xname(sc->sc_dev));
   1571 			wantinit = 1;
   1572 		}
   1573 	}
   1574 
   1575 	if (handled) {
   1576 		if (wantinit)
   1577 			dge_init(ifp);
   1578 
   1579 		/* Try to get more packets going. */
   1580 		if_schedule_deferred_start(ifp);
   1581 	}
   1582 
   1583 	return (handled);
   1584 }
   1585 
   1586 /*
   1587  * dge_txintr:
   1588  *
   1589  *	Helper; handle transmit interrupts.
   1590  */
   1591 static void
   1592 dge_txintr(struct dge_softc *sc)
   1593 {
   1594 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1595 	struct dge_txsoft *txs;
   1596 	uint8_t status;
   1597 	int i;
   1598 
   1599 	ifp->if_flags &= ~IFF_OACTIVE;
   1600 
   1601 	/*
   1602 	 * Go through the Tx list and free mbufs for those
   1603 	 * frames which have been transmitted.
   1604 	 */
   1605 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
   1606 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
   1607 		txs = &sc->sc_txsoft[i];
   1608 
   1609 		DPRINTF(DGE_DEBUG_TX,
   1610 		    ("%s: TX: checking job %d\n", device_xname(sc->sc_dev), i));
   1611 
   1612 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
   1613 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1614 
   1615 		status =
   1616 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
   1617 		if ((status & TDESC_STA_DD) == 0) {
   1618 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
   1619 			    BUS_DMASYNC_PREREAD);
   1620 			break;
   1621 		}
   1622 
   1623 		DPRINTF(DGE_DEBUG_TX,
   1624 		    ("%s: TX: job %d done: descs %d..%d\n",
   1625 		    device_xname(sc->sc_dev), i, txs->txs_firstdesc,
   1626 		    txs->txs_lastdesc));
   1627 
   1628 		ifp->if_opackets++;
   1629 		sc->sc_txfree += txs->txs_ndesc;
   1630 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
   1631 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1632 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   1633 		m_freem(txs->txs_mbuf);
   1634 		txs->txs_mbuf = NULL;
   1635 	}
   1636 
   1637 	/* Update the dirty transmit buffer pointer. */
   1638 	sc->sc_txsdirty = i;
   1639 	DPRINTF(DGE_DEBUG_TX,
   1640 	    ("%s: TX: txsdirty -> %d\n", device_xname(sc->sc_dev), i));
   1641 
   1642 	/*
   1643 	 * If there are no more pending transmissions, cancel the watchdog
   1644 	 * timer.
   1645 	 */
   1646 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
   1647 		ifp->if_timer = 0;
   1648 }
   1649 
   1650 /*
   1651  * dge_rxintr:
   1652  *
   1653  *	Helper; handle receive interrupts.
   1654  */
   1655 static void
   1656 dge_rxintr(struct dge_softc *sc)
   1657 {
   1658 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1659 	struct dge_rxsoft *rxs;
   1660 	struct mbuf *m;
   1661 	int i, len;
   1662 	uint8_t status, errors;
   1663 
   1664 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
   1665 		rxs = &sc->sc_rxsoft[i];
   1666 
   1667 		DPRINTF(DGE_DEBUG_RX,
   1668 		    ("%s: RX: checking descriptor %d\n",
   1669 		    device_xname(sc->sc_dev), i));
   1670 
   1671 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1672 
   1673 		status = sc->sc_rxdescs[i].dr_status;
   1674 		errors = sc->sc_rxdescs[i].dr_errors;
   1675 		len = le16toh(sc->sc_rxdescs[i].dr_len);
   1676 
   1677 		if ((status & RDESC_STS_DD) == 0) {
   1678 			/*
   1679 			 * We have processed all of the receive descriptors.
   1680 			 */
   1681 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1682 			break;
   1683 		}
   1684 
   1685 		if (__predict_false(sc->sc_rxdiscard)) {
   1686 			DPRINTF(DGE_DEBUG_RX,
   1687 			    ("%s: RX: discarding contents of descriptor %d\n",
   1688 			    device_xname(sc->sc_dev), i));
   1689 			DGE_INIT_RXDESC(sc, i);
   1690 			if (status & RDESC_STS_EOP) {
   1691 				/* Reset our state. */
   1692 				DPRINTF(DGE_DEBUG_RX,
   1693 				    ("%s: RX: resetting rxdiscard -> 0\n",
   1694 				    device_xname(sc->sc_dev)));
   1695 				sc->sc_rxdiscard = 0;
   1696 			}
   1697 			continue;
   1698 		}
   1699 
   1700 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1701 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
   1702 
   1703 		m = rxs->rxs_mbuf;
   1704 
   1705 		/*
   1706 		 * Add a new receive buffer to the ring.
   1707 		 */
   1708 		if (dge_add_rxbuf(sc, i) != 0) {
   1709 			/*
   1710 			 * Failed, throw away what we've done so
   1711 			 * far, and discard the rest of the packet.
   1712 			 */
   1713 			ifp->if_ierrors++;
   1714 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1715 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   1716 			DGE_INIT_RXDESC(sc, i);
   1717 			if ((status & RDESC_STS_EOP) == 0)
   1718 				sc->sc_rxdiscard = 1;
   1719 			if (sc->sc_rxhead != NULL)
   1720 				m_freem(sc->sc_rxhead);
   1721 			DGE_RXCHAIN_RESET(sc);
   1722 			DPRINTF(DGE_DEBUG_RX,
   1723 			    ("%s: RX: Rx buffer allocation failed, "
   1724 			    "dropping packet%s\n", device_xname(sc->sc_dev),
   1725 			    sc->sc_rxdiscard ? " (discard)" : ""));
   1726 			continue;
   1727 		}
   1728 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
   1729 
   1730 		DGE_RXCHAIN_LINK(sc, m);
   1731 
   1732 		m->m_len = len;
   1733 
   1734 		DPRINTF(DGE_DEBUG_RX,
   1735 		    ("%s: RX: buffer at %p len %d\n",
   1736 		    device_xname(sc->sc_dev), m->m_data, len));
   1737 
   1738 		/*
   1739 		 * If this is not the end of the packet, keep
   1740 		 * looking.
   1741 		 */
   1742 		if ((status & RDESC_STS_EOP) == 0) {
   1743 			sc->sc_rxlen += len;
   1744 			DPRINTF(DGE_DEBUG_RX,
   1745 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
   1746 			    device_xname(sc->sc_dev), sc->sc_rxlen));
   1747 			continue;
   1748 		}
   1749 
   1750 		/*
   1751 		 * Okay, we have the entire packet now...
   1752 		 */
   1753 		*sc->sc_rxtailp = NULL;
   1754 		m = sc->sc_rxhead;
   1755 		len += sc->sc_rxlen;
   1756 
   1757 		DGE_RXCHAIN_RESET(sc);
   1758 
   1759 		DPRINTF(DGE_DEBUG_RX,
   1760 		    ("%s: RX: have entire packet, len -> %d\n",
   1761 		    device_xname(sc->sc_dev), len));
   1762 
   1763 		/*
   1764 		 * If an error occurred, update stats and drop the packet.
   1765 		 */
   1766 		if (errors &
   1767 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
   1768 			ifp->if_ierrors++;
   1769 			if (errors & RDESC_ERR_SE)
   1770 				printf("%s: symbol error\n",
   1771 				    device_xname(sc->sc_dev));
   1772 			else if (errors & RDESC_ERR_P)
   1773 				printf("%s: parity error\n",
   1774 				    device_xname(sc->sc_dev));
   1775 			else if (errors & RDESC_ERR_CE)
   1776 				printf("%s: CRC error\n",
   1777 				    device_xname(sc->sc_dev));
   1778 			m_freem(m);
   1779 			continue;
   1780 		}
   1781 
   1782 		/*
   1783 		 * No errors.  Receive the packet.
   1784 		 */
   1785 		m_set_rcvif(m, ifp);
   1786 		m->m_pkthdr.len = len;
   1787 
   1788 		/*
   1789 		 * Set up checksum info for this packet.
   1790 		 */
   1791 		if (status & RDESC_STS_IPCS) {
   1792 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
   1793 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1794 			if (errors & RDESC_ERR_IPE)
   1795 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1796 		}
   1797 		if (status & RDESC_STS_TCPCS) {
   1798 			/*
   1799 			 * Note: we don't know if this was TCP or UDP,
   1800 			 * so we just set both bits, and expect the
   1801 			 * upper layers to deal.
   1802 			 */
   1803 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
   1804 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
   1805 			if (errors & RDESC_ERR_TCPE)
   1806 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1807 		}
   1808 
   1809 		/* Pass it on. */
   1810 		if_percpuq_enqueue(ifp->if_percpuq, m);
   1811 	}
   1812 
   1813 	/* Update the receive pointer. */
   1814 	sc->sc_rxptr = i;
   1815 
   1816 	DPRINTF(DGE_DEBUG_RX,
   1817 	    ("%s: RX: rxptr -> %d\n", device_xname(sc->sc_dev), i));
   1818 }
   1819 
   1820 /*
   1821  * dge_linkintr:
   1822  *
   1823  *	Helper; handle link interrupts.
   1824  */
   1825 static void
   1826 dge_linkintr(struct dge_softc *sc, uint32_t icr)
   1827 {
   1828 	uint32_t status;
   1829 
   1830 	if (icr & ICR_LSC) {
   1831 		status = CSR_READ(sc, DGE_STATUS);
   1832 		if (status & STATUS_LINKUP) {
   1833 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
   1834 			    device_xname(sc->sc_dev)));
   1835 		} else {
   1836 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
   1837 			    device_xname(sc->sc_dev)));
   1838 		}
   1839 	} else if (icr & ICR_RXSEQ) {
   1840 		DPRINTF(DGE_DEBUG_LINK,
   1841 		    ("%s: LINK: Receive sequence error\n",
   1842 		    device_xname(sc->sc_dev)));
   1843 	}
   1844 	/* XXX - fix errata */
   1845 }
   1846 
   1847 /*
   1848  * dge_reset:
   1849  *
   1850  *	Reset the i82597 chip.
   1851  */
   1852 static void
   1853 dge_reset(struct dge_softc *sc)
   1854 {
   1855 	int i;
   1856 
   1857 	/*
   1858 	 * Do a chip reset.
   1859 	 */
   1860 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
   1861 
   1862 	delay(10000);
   1863 
   1864 	for (i = 0; i < 1000; i++) {
   1865 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
   1866 			break;
   1867 		delay(20);
   1868 	}
   1869 
   1870 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
   1871 		printf("%s: WARNING: reset failed to complete\n",
   1872 		    device_xname(sc->sc_dev));
   1873         /*
   1874          * Reset the EEPROM logic.
   1875          * This will cause the chip to reread its default values,
   1876 	 * which doesn't happen otherwise (errata).
   1877          */
   1878         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
   1879         delay(10000);
   1880 }
   1881 
   1882 /*
   1883  * dge_init:		[ifnet interface function]
   1884  *
   1885  *	Initialize the interface.  Must be called at splnet().
   1886  */
   1887 static int
   1888 dge_init(struct ifnet *ifp)
   1889 {
   1890 	struct dge_softc *sc = ifp->if_softc;
   1891 	struct dge_rxsoft *rxs;
   1892 	int i, error = 0;
   1893 	uint32_t reg;
   1894 
   1895 	/*
   1896 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
   1897 	 * There is a small but measurable benefit to avoiding the adjusment
   1898 	 * of the descriptor so that the headers are aligned, for normal mtu,
   1899 	 * on such platforms.  One possibility is that the DMA itself is
   1900 	 * slightly more efficient if the front of the entire packet (instead
   1901 	 * of the front of the headers) is aligned.
   1902 	 *
   1903 	 * Note we must always set align_tweak to 0 if we are using
   1904 	 * jumbo frames.
   1905 	 */
   1906 #ifdef __NO_STRICT_ALIGNMENT
   1907 	sc->sc_align_tweak = 0;
   1908 #else
   1909 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
   1910 		sc->sc_align_tweak = 0;
   1911 	else
   1912 		sc->sc_align_tweak = 2;
   1913 #endif /* __NO_STRICT_ALIGNMENT */
   1914 
   1915 	/* Cancel any pending I/O. */
   1916 	dge_stop(ifp, 0);
   1917 
   1918 	/* Reset the chip to a known state. */
   1919 	dge_reset(sc);
   1920 
   1921 	/* Initialize the transmit descriptor ring. */
   1922 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
   1923 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
   1924 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1925 	sc->sc_txfree = DGE_NTXDESC;
   1926 	sc->sc_txnext = 0;
   1927 
   1928 	sc->sc_txctx_ipcs = 0xffffffff;
   1929 	sc->sc_txctx_tucs = 0xffffffff;
   1930 
   1931 	CSR_WRITE(sc, DGE_TDBAH, 0);
   1932 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
   1933 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
   1934 	CSR_WRITE(sc, DGE_TDH, 0);
   1935 	CSR_WRITE(sc, DGE_TDT, 0);
   1936 	CSR_WRITE(sc, DGE_TIDV, TIDV);
   1937 
   1938 #if 0
   1939 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
   1940 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
   1941 #endif
   1942 	CSR_WRITE(sc, DGE_RXDCTL,
   1943 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
   1944 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
   1945 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
   1946 
   1947 	/* Initialize the transmit job descriptors. */
   1948 	for (i = 0; i < DGE_TXQUEUELEN; i++)
   1949 		sc->sc_txsoft[i].txs_mbuf = NULL;
   1950 	sc->sc_txsfree = DGE_TXQUEUELEN;
   1951 	sc->sc_txsnext = 0;
   1952 	sc->sc_txsdirty = 0;
   1953 
   1954 	/*
   1955 	 * Initialize the receive descriptor and receive job
   1956 	 * descriptor rings.
   1957 	 */
   1958 	CSR_WRITE(sc, DGE_RDBAH, 0);
   1959 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
   1960 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
   1961 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
   1962 	CSR_WRITE(sc, DGE_RDT, 0);
   1963 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
   1964 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
   1965 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
   1966 
   1967 	for (i = 0; i < DGE_NRXDESC; i++) {
   1968 		rxs = &sc->sc_rxsoft[i];
   1969 		if (rxs->rxs_mbuf == NULL) {
   1970 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
   1971 				printf("%s: unable to allocate or map rx "
   1972 				    "buffer %d, error = %d\n",
   1973 				    device_xname(sc->sc_dev), i, error);
   1974 				/*
   1975 				 * XXX Should attempt to run with fewer receive
   1976 				 * XXX buffers instead of just failing.
   1977 				 */
   1978 				dge_rxdrain(sc);
   1979 				goto out;
   1980 			}
   1981 		}
   1982 		DGE_INIT_RXDESC(sc, i);
   1983 	}
   1984 	sc->sc_rxptr = DGE_RXSPACE;
   1985 	sc->sc_rxdiscard = 0;
   1986 	DGE_RXCHAIN_RESET(sc);
   1987 
   1988 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
   1989 		sc->sc_ctrl0 |= CTRL0_JFE;
   1990 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
   1991 	}
   1992 
   1993 	/* Write the control registers. */
   1994 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
   1995 
   1996 	/*
   1997 	 * Set up checksum offload parameters.
   1998 	 */
   1999 	reg = CSR_READ(sc, DGE_RXCSUM);
   2000 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
   2001 		reg |= RXCSUM_IPOFL;
   2002 	else
   2003 		reg &= ~RXCSUM_IPOFL;
   2004 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
   2005 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
   2006 	else {
   2007 		reg &= ~RXCSUM_TUOFL;
   2008 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
   2009 			reg &= ~RXCSUM_IPOFL;
   2010 	}
   2011 	CSR_WRITE(sc, DGE_RXCSUM, reg);
   2012 
   2013 	/*
   2014 	 * Set up the interrupt registers.
   2015 	 */
   2016 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
   2017 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
   2018 	    ICR_RXO | ICR_RXT0;
   2019 
   2020 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
   2021 
   2022 	/*
   2023 	 * Set up the transmit control register.
   2024 	 */
   2025 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
   2026 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
   2027 
   2028 	/*
   2029 	 * Set up the receive control register; we actually program
   2030 	 * the register when we set the receive filter.  Use multicast
   2031 	 * address offset type 0.
   2032 	 */
   2033 	sc->sc_mchash_type = 0;
   2034 
   2035 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
   2036 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
   2037 
   2038 #ifdef DGE_OFFBYONE_RXBUG
   2039 	sc->sc_rctl |= RCTL_BSIZE_16k;
   2040 #else
   2041 	switch(MCLBYTES) {
   2042 	case 2048:
   2043 		sc->sc_rctl |= RCTL_BSIZE_2k;
   2044 		break;
   2045 	case 4096:
   2046 		sc->sc_rctl |= RCTL_BSIZE_4k;
   2047 		break;
   2048 	case 8192:
   2049 		sc->sc_rctl |= RCTL_BSIZE_8k;
   2050 		break;
   2051 	case 16384:
   2052 		sc->sc_rctl |= RCTL_BSIZE_16k;
   2053 		break;
   2054 	default:
   2055 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
   2056 	}
   2057 #endif
   2058 
   2059 	/* Set the receive filter. */
   2060 	/* Also sets RCTL */
   2061 	dge_set_filter(sc);
   2062 
   2063 	/* ...all done! */
   2064 	ifp->if_flags |= IFF_RUNNING;
   2065 	ifp->if_flags &= ~IFF_OACTIVE;
   2066 
   2067  out:
   2068 	if (error)
   2069 		printf("%s: interface not running\n", device_xname(sc->sc_dev));
   2070 	return (error);
   2071 }
   2072 
   2073 /*
   2074  * dge_rxdrain:
   2075  *
   2076  *	Drain the receive queue.
   2077  */
   2078 static void
   2079 dge_rxdrain(struct dge_softc *sc)
   2080 {
   2081 	struct dge_rxsoft *rxs;
   2082 	int i;
   2083 
   2084 	for (i = 0; i < DGE_NRXDESC; i++) {
   2085 		rxs = &sc->sc_rxsoft[i];
   2086 		if (rxs->rxs_mbuf != NULL) {
   2087 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2088 			m_freem(rxs->rxs_mbuf);
   2089 			rxs->rxs_mbuf = NULL;
   2090 		}
   2091 	}
   2092 }
   2093 
   2094 /*
   2095  * dge_stop:		[ifnet interface function]
   2096  *
   2097  *	Stop transmission on the interface.
   2098  */
   2099 static void
   2100 dge_stop(struct ifnet *ifp, int disable)
   2101 {
   2102 	struct dge_softc *sc = ifp->if_softc;
   2103 	struct dge_txsoft *txs;
   2104 	int i;
   2105 
   2106 	/* Stop the transmit and receive processes. */
   2107 	CSR_WRITE(sc, DGE_TCTL, 0);
   2108 	CSR_WRITE(sc, DGE_RCTL, 0);
   2109 
   2110 	/* Release any queued transmit buffers. */
   2111 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
   2112 		txs = &sc->sc_txsoft[i];
   2113 		if (txs->txs_mbuf != NULL) {
   2114 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   2115 			m_freem(txs->txs_mbuf);
   2116 			txs->txs_mbuf = NULL;
   2117 		}
   2118 	}
   2119 
   2120 	/* Mark the interface as down and cancel the watchdog timer. */
   2121 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2122 	ifp->if_timer = 0;
   2123 
   2124 	if (disable)
   2125 		dge_rxdrain(sc);
   2126 }
   2127 
   2128 /*
   2129  * dge_add_rxbuf:
   2130  *
   2131  *	Add a receive buffer to the indiciated descriptor.
   2132  */
   2133 static int
   2134 dge_add_rxbuf(struct dge_softc *sc, int idx)
   2135 {
   2136 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
   2137 	struct mbuf *m;
   2138 	int error;
   2139 #ifdef DGE_OFFBYONE_RXBUG
   2140 	void *buf;
   2141 #endif
   2142 
   2143 	MGETHDR(m, M_DONTWAIT, MT_DATA);
   2144 	if (m == NULL)
   2145 		return (ENOBUFS);
   2146 
   2147 #ifdef DGE_OFFBYONE_RXBUG
   2148 	if ((buf = dge_getbuf(sc)) == NULL)
   2149 		return ENOBUFS;
   2150 
   2151 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
   2152 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
   2153 	m->m_flags |= M_EXT_RW;
   2154 
   2155 	if (rxs->rxs_mbuf != NULL)
   2156 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2157 	rxs->rxs_mbuf = m;
   2158 
   2159 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
   2160 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
   2161 #else
   2162 	MCLGET(m, M_DONTWAIT);
   2163 	if ((m->m_flags & M_EXT) == 0) {
   2164 		m_freem(m);
   2165 		return (ENOBUFS);
   2166 	}
   2167 
   2168 	if (rxs->rxs_mbuf != NULL)
   2169 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2170 
   2171 	rxs->rxs_mbuf = m;
   2172 
   2173 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
   2174 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
   2175 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   2176 #endif
   2177 	if (error) {
   2178 		printf("%s: unable to load rx DMA map %d, error = %d\n",
   2179 		    device_xname(sc->sc_dev), idx, error);
   2180 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
   2181 	}
   2182 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   2183 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   2184 
   2185 	return (0);
   2186 }
   2187 
   2188 /*
   2189  * dge_set_ral:
   2190  *
   2191  *	Set an entry in the receive address list.
   2192  */
   2193 static void
   2194 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
   2195 {
   2196 	uint32_t ral_lo, ral_hi;
   2197 
   2198 	if (enaddr != NULL) {
   2199 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
   2200 		    (enaddr[3] << 24);
   2201 		ral_hi = enaddr[4] | (enaddr[5] << 8);
   2202 		ral_hi |= RAH_AV;
   2203 	} else {
   2204 		ral_lo = 0;
   2205 		ral_hi = 0;
   2206 	}
   2207 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
   2208 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
   2209 }
   2210 
   2211 /*
   2212  * dge_mchash:
   2213  *
   2214  *	Compute the hash of the multicast address for the 4096-bit
   2215  *	multicast filter.
   2216  */
   2217 static uint32_t
   2218 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
   2219 {
   2220 	static const int lo_shift[4] = { 4, 3, 2, 0 };
   2221 	static const int hi_shift[4] = { 4, 5, 6, 8 };
   2222 	uint32_t hash;
   2223 
   2224 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
   2225 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
   2226 
   2227 	return (hash & 0xfff);
   2228 }
   2229 
   2230 /*
   2231  * dge_set_filter:
   2232  *
   2233  *	Set up the receive filter.
   2234  */
   2235 static void
   2236 dge_set_filter(struct dge_softc *sc)
   2237 {
   2238 	struct ethercom *ec = &sc->sc_ethercom;
   2239 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   2240 	struct ether_multi *enm;
   2241 	struct ether_multistep step;
   2242 	uint32_t hash, reg, bit;
   2243 	int i;
   2244 
   2245 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
   2246 
   2247 	if (ifp->if_flags & IFF_BROADCAST)
   2248 		sc->sc_rctl |= RCTL_BAM;
   2249 	if (ifp->if_flags & IFF_PROMISC) {
   2250 		sc->sc_rctl |= RCTL_UPE;
   2251 		goto allmulti;
   2252 	}
   2253 
   2254 	/*
   2255 	 * Set the station address in the first RAL slot, and
   2256 	 * clear the remaining slots.
   2257 	 */
   2258 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
   2259 	for (i = 1; i < RA_TABSIZE; i++)
   2260 		dge_set_ral(sc, NULL, i);
   2261 
   2262 	/* Clear out the multicast table. */
   2263 	for (i = 0; i < MC_TABSIZE; i++)
   2264 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
   2265 
   2266 	ETHER_FIRST_MULTI(step, ec, enm);
   2267 	while (enm != NULL) {
   2268 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
   2269 			/*
   2270 			 * We must listen to a range of multicast addresses.
   2271 			 * For now, just accept all multicasts, rather than
   2272 			 * trying to set only those filter bits needed to match
   2273 			 * the range.  (At this time, the only use of address
   2274 			 * ranges is for IP multicast routing, for which the
   2275 			 * range is big enough to require all bits set.)
   2276 			 */
   2277 			goto allmulti;
   2278 		}
   2279 
   2280 		hash = dge_mchash(sc, enm->enm_addrlo);
   2281 
   2282 		reg = (hash >> 5) & 0x7f;
   2283 		bit = hash & 0x1f;
   2284 
   2285 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
   2286 		hash |= 1U << bit;
   2287 
   2288 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
   2289 
   2290 		ETHER_NEXT_MULTI(step, enm);
   2291 	}
   2292 
   2293 	ifp->if_flags &= ~IFF_ALLMULTI;
   2294 	goto setit;
   2295 
   2296  allmulti:
   2297 	ifp->if_flags |= IFF_ALLMULTI;
   2298 	sc->sc_rctl |= RCTL_MPE;
   2299 
   2300  setit:
   2301 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
   2302 }
   2303 
   2304 /*
   2305  * Read in the EEPROM info and verify checksum.
   2306  */
   2307 int
   2308 dge_read_eeprom(struct dge_softc *sc)
   2309 {
   2310 	uint16_t cksum;
   2311 	int i;
   2312 
   2313 	cksum = 0;
   2314 	for (i = 0; i < EEPROM_SIZE; i++) {
   2315 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
   2316 		cksum += sc->sc_eeprom[i];
   2317 	}
   2318 	return cksum != EEPROM_CKSUM;
   2319 }
   2320 
   2321 
   2322 /*
   2323  * Read a 16-bit word from address addr in the serial EEPROM.
   2324  */
   2325 uint16_t
   2326 dge_eeprom_word(struct dge_softc *sc, int addr)
   2327 {
   2328 	uint32_t reg;
   2329 	uint16_t rval = 0;
   2330 	int i;
   2331 
   2332 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
   2333 
   2334 	/* Lower clock pulse (and data in to chip) */
   2335 	CSR_WRITE(sc, DGE_EECD, reg);
   2336 	/* Select chip */
   2337 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
   2338 
   2339 	/* Send read command */
   2340 	dge_eeprom_clockout(sc, 1);
   2341 	dge_eeprom_clockout(sc, 1);
   2342 	dge_eeprom_clockout(sc, 0);
   2343 
   2344 	/* Send address */
   2345 	for (i = 5; i >= 0; i--)
   2346 		dge_eeprom_clockout(sc, (addr >> i) & 1);
   2347 
   2348 	/* Read data */
   2349 	for (i = 0; i < 16; i++) {
   2350 		rval <<= 1;
   2351 		rval |= dge_eeprom_clockin(sc);
   2352 	}
   2353 
   2354 	/* Deselect chip */
   2355 	CSR_WRITE(sc, DGE_EECD, reg);
   2356 
   2357 	return rval;
   2358 }
   2359 
   2360 /*
   2361  * Clock out a single bit to the EEPROM.
   2362  */
   2363 void
   2364 dge_eeprom_clockout(struct dge_softc *sc, int bit)
   2365 {
   2366 	int reg;
   2367 
   2368 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
   2369 	if (bit)
   2370 		reg |= EECD_DI;
   2371 
   2372 	CSR_WRITE(sc, DGE_EECD, reg);
   2373 	delay(2);
   2374 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
   2375 	delay(2);
   2376 	CSR_WRITE(sc, DGE_EECD, reg);
   2377 	delay(2);
   2378 }
   2379 
   2380 /*
   2381  * Clock in a single bit from EEPROM.
   2382  */
   2383 int
   2384 dge_eeprom_clockin(struct dge_softc *sc)
   2385 {
   2386 	int reg, rv;
   2387 
   2388 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
   2389 
   2390 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
   2391 	delay(2);
   2392 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
   2393 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
   2394 	delay(2);
   2395 
   2396 	return rv;
   2397 }
   2398 
   2399 static void
   2400 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
   2401 {
   2402 	struct dge_softc *sc = ifp->if_softc;
   2403 
   2404 	ifmr->ifm_status = IFM_AVALID;
   2405 	if (sc->sc_dgep->dgep_flags & DGEP_F_10G_SR ) {
   2406 		ifmr->ifm_active = IFM_ETHER|IFM_10G_SR;
   2407 	} else {
   2408 		ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
   2409 	}
   2410 
   2411 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
   2412 		ifmr->ifm_status |= IFM_ACTIVE;
   2413 }
   2414 
   2415 static inline int
   2416 phwait(struct dge_softc *sc, int p, int r, int d, int type)
   2417 {
   2418         int i, mdic;
   2419 
   2420         CSR_WRITE(sc, DGE_MDIO,
   2421 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
   2422         for (i = 0; i < 10; i++) {
   2423                 delay(10);
   2424                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
   2425                         break;
   2426         }
   2427         return mdic;
   2428 }
   2429 
   2430 static void
   2431 dge_xgmii_writereg(struct dge_softc *sc, int phy, int reg, int val)
   2432 {
   2433 	int mdic;
   2434 
   2435 	CSR_WRITE(sc, DGE_MDIRW, val);
   2436 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
   2437 		printf("%s: address cycle timeout; phy %d reg %d\n",
   2438 		    device_xname(sc->sc_dev), phy, reg);
   2439 		return;
   2440 	}
   2441 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
   2442 		printf("%s: write cycle timeout; phy %d reg %d\n",
   2443 		    device_xname(sc->sc_dev), phy, reg);
   2444 		return;
   2445 	}
   2446 }
   2447 
   2448 static void
   2449 dge_xgmii_reset(struct dge_softc *sc)
   2450 {
   2451 	dge_xgmii_writereg(sc, 0, 0, BMCR_RESET);
   2452 }
   2453 
   2454 static int
   2455 dge_xgmii_mediachange(struct ifnet *ifp)
   2456 {
   2457 	return 0;
   2458 }
   2459