Home | History | Annotate | Line # | Download | only in pci
if_dge.c revision 1.58
      1 /*	$NetBSD: if_dge.c,v 1.58 2020/03/01 15:11:31 thorpej 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 1.58 2020/03/01 15:11:31 thorpej Exp $");
     84 
     85 
     86 
     87 #include <sys/param.h>
     88 #include <sys/systm.h>
     89 #include <sys/callout.h>
     90 #include <sys/mbuf.h>
     91 #include <sys/malloc.h>
     92 #include <sys/kernel.h>
     93 #include <sys/socket.h>
     94 #include <sys/ioctl.h>
     95 #include <sys/errno.h>
     96 #include <sys/device.h>
     97 #include <sys/queue.h>
     98 #include <sys/rndsource.h>
     99 
    100 #include <net/if.h>
    101 #include <net/if_dl.h>
    102 #include <net/if_media.h>
    103 #include <net/if_ether.h>
    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 	const bus_addr_t __rxaddr = sc->sc_bugmap->dm_segs[0].ds_addr +	\
    407 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf);		\
    408 									\
    409 	__rxd->dr_baddrl = htole32(__rxaddr);				\
    410 	__rxd->dr_baddrh = htole32(((uint64_t)__rxaddr) >> 32);		\
    411 	__rxd->dr_len = 0;						\
    412 	__rxd->dr_cksum = 0;						\
    413 	__rxd->dr_status = 0;						\
    414 	__rxd->dr_errors = 0;						\
    415 	__rxd->dr_special = 0;						\
    416 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); \
    417 									\
    418 	CSR_WRITE((sc), DGE_RDT, (x));					\
    419 } while (/*CONSTCOND*/0)
    420 #else
    421 #define DGE_INIT_RXDESC(sc, x)						\
    422 do {									\
    423 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
    424 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
    425 	struct mbuf *__m = __rxs->rxs_mbuf;				\
    426 									\
    427 	/*								\
    428 	 * Note: We scoot the packet forward 2 bytes in the buffer	\
    429 	 * so that the payload after the Ethernet header is aligned	\
    430 	 * to a 4-byte boundary.					\
    431 	 *								\
    432 	 * XXX BRAINDAMAGE ALERT!					\
    433 	 * The stupid chip uses the same size for every buffer, which	\
    434 	 * is set in the Receive Control register.  We are using the 2K \
    435 	 * size option, but what we REALLY want is (2K - 2)!  For this	\
    436 	 * reason, we can't "scoot" packets longer than the standard	\
    437 	 * Ethernet MTU.  On strict-alignment platforms, if the total	\
    438 	 * size exceeds (2K - 2) we set align_tweak to 0 and let	\
    439 	 * the upper layer copy the headers.				\
    440 	 */								\
    441 	__m->m_data = __m->m_ext.ext_buf + (sc)->sc_align_tweak;	\
    442 									\
    443 	const bus_addr_t __rxaddr =					\
    444 	    __rxs->rxs_dmamap->dm_segs[0].ds_addr +			\
    445 	    (sc)->sc_align_tweak;					\
    446 									\
    447 	__rxd->dr_baddrl = htole32(__rxaddr);				\
    448 	__rxd->dr_baddrh = htole32(((uint64_t)__rxaddr) >> 32);		\
    449 	__rxd->dr_len = 0;						\
    450 	__rxd->dr_cksum = 0;						\
    451 	__rxd->dr_status = 0;						\
    452 	__rxd->dr_errors = 0;						\
    453 	__rxd->dr_special = 0;						\
    454 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); \
    455 									\
    456 	CSR_WRITE((sc), DGE_RDT, (x));					\
    457 } while (/*CONSTCOND*/0)
    458 #endif
    459 
    460 #ifdef DGE_OFFBYONE_RXBUG
    461 /*
    462  * Allocation constants.  Much memory may be used for this.
    463  */
    464 #ifndef DGE_BUFFER_SIZE
    465 #define DGE_BUFFER_SIZE DGE_MAX_MTU
    466 #endif
    467 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
    468 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
    469 
    470 struct rxbugentry {
    471 	SLIST_ENTRY(rxbugentry) rb_entry;
    472 	int rb_slot;
    473 };
    474 
    475 static int
    476 dge_alloc_rcvmem(struct dge_softc *sc)
    477 {
    478 	char *kva;
    479 	bus_dma_segment_t seg;
    480 	int i, rseg, state, error;
    481 	struct rxbugentry *entry;
    482 
    483 	state = error = 0;
    484 
    485 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
    486 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
    487 		aprint_error_dev(sc->sc_dev, "can't alloc rx buffers\n");
    488 		return ENOBUFS;
    489 	}
    490 
    491 	state = 1;
    492 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
    493 	    BUS_DMA_NOWAIT)) {
    494 		aprint_error_dev(sc->sc_dev,
    495 		    "can't map DMA buffers (%d bytes)\n", (int)DGE_RXMEM);
    496 		error = ENOBUFS;
    497 		goto out;
    498 	}
    499 
    500 	state = 2;
    501 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
    502 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
    503 		aprint_error_dev(sc->sc_dev, "can't create DMA map\n");
    504 		error = ENOBUFS;
    505 		goto out;
    506 	}
    507 
    508 	state = 3;
    509 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
    510 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
    511 		aprint_error_dev(sc->sc_dev, "can't load DMA map\n");
    512 		error = ENOBUFS;
    513 		goto out;
    514 	}
    515 
    516 	state = 4;
    517 	sc->sc_bugbuf = (void *)kva;
    518 	SLIST_INIT(&sc->sc_buglist);
    519 
    520 	/*
    521 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
    522 	 * in an array.
    523 	 */
    524 	entry = malloc(sizeof(*entry) * DGE_NBUFFERS, M_DEVBUF, M_WAITOK);
    525 	sc->sc_entry = entry;
    526 	for (i = 0; i < DGE_NBUFFERS; i++) {
    527 		entry[i].rb_slot = i;
    528 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
    529 	}
    530 out:
    531 	if (error != 0) {
    532 		switch (state) {
    533 		case 4:
    534 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
    535 			/* FALLTHROUGH */
    536 		case 3:
    537 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
    538 			/* FALLTHROUGH */
    539 		case 2:
    540 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
    541 			/* FALLTHROUGH */
    542 		case 1:
    543 			bus_dmamem_free(sc->sc_dmat, &seg, rseg);
    544 			break;
    545 		default:
    546 			break;
    547 		}
    548 	}
    549 
    550 	return error;
    551 }
    552 
    553 /*
    554  * Allocate a jumbo buffer.
    555  */
    556 static void *
    557 dge_getbuf(struct dge_softc *sc)
    558 {
    559 	struct rxbugentry *entry;
    560 
    561 	entry = SLIST_FIRST(&sc->sc_buglist);
    562 
    563 	if (entry == NULL) {
    564 		printf("%s: no free RX buffers\n", device_xname(sc->sc_dev));
    565 		return NULL;
    566 	}
    567 
    568 	SLIST_REMOVE_HEAD(&sc->sc_buglist, rb_entry);
    569 	return (char *)sc->sc_bugbuf + entry->rb_slot * DGE_BUFFER_SIZE;
    570 }
    571 
    572 /*
    573  * Release a jumbo buffer.
    574  */
    575 static void
    576 dge_freebuf(struct mbuf *m, void *buf, size_t size, void *arg)
    577 {
    578 	struct rxbugentry *entry;
    579 	struct dge_softc *sc;
    580 	int i, s;
    581 
    582 	/* Extract the softc struct pointer. */
    583 	sc = (struct dge_softc *)arg;
    584 
    585 	if (sc == NULL)
    586 		panic("dge_freebuf: can't find softc pointer!");
    587 
    588 	/* calculate the slot this buffer belongs to */
    589 
    590 	i = ((char *)buf - (char *)sc->sc_bugbuf) / DGE_BUFFER_SIZE;
    591 
    592 	if ((i < 0) || (i >= DGE_NBUFFERS))
    593 		panic("dge_freebuf: asked to free buffer %d!", i);
    594 
    595 	s = splvm();
    596 	entry = sc->sc_entry + i;
    597 	SLIST_INSERT_HEAD(&sc->sc_buglist, entry, rb_entry);
    598 
    599 	if (__predict_true(m != NULL))
    600 		pool_cache_put(mb_cache, m);
    601 	splx(s);
    602 }
    603 #endif
    604 
    605 static void	dge_start(struct ifnet *);
    606 static void	dge_watchdog(struct ifnet *);
    607 static int	dge_ioctl(struct ifnet *, u_long, void *);
    608 static int	dge_init(struct ifnet *);
    609 static void	dge_stop(struct ifnet *, int);
    610 
    611 static bool	dge_shutdown(device_t, int);
    612 
    613 static void	dge_reset(struct dge_softc *);
    614 static void	dge_rxdrain(struct dge_softc *);
    615 static int	dge_add_rxbuf(struct dge_softc *, int);
    616 
    617 static void	dge_set_filter(struct dge_softc *);
    618 
    619 static int	dge_intr(void *);
    620 static void	dge_txintr(struct dge_softc *);
    621 static void	dge_rxintr(struct dge_softc *);
    622 static void	dge_linkintr(struct dge_softc *, uint32_t);
    623 
    624 static int	dge_match(device_t, cfdata_t, void *);
    625 static void	dge_attach(device_t, device_t, void *);
    626 
    627 static int	dge_read_eeprom(struct dge_softc *sc);
    628 static int	dge_eeprom_clockin(struct dge_softc *sc);
    629 static void	dge_eeprom_clockout(struct dge_softc *sc, int bit);
    630 static uint16_t	dge_eeprom_word(struct dge_softc *sc, int addr);
    631 static int	dge_xgmii_mediachange(struct ifnet *);
    632 static void	dge_xgmii_mediastatus(struct ifnet *, struct ifmediareq *);
    633 static void	dge_xgmii_reset(struct dge_softc *);
    634 static void	dge_xgmii_writereg(struct dge_softc *, int, int, int);
    635 
    636 
    637 CFATTACH_DECL_NEW(dge, sizeof(struct dge_softc),
    638     dge_match, dge_attach, NULL, NULL);
    639 
    640 #ifdef DGE_EVENT_COUNTERS
    641 #if DGE_NTXSEGS > 100
    642 #error Update dge_txseg_evcnt_names
    643 #endif
    644 static char (*dge_txseg_evcnt_names)[DGE_NTXSEGS][8 /* "txseg00" + \0 */];
    645 #endif /* DGE_EVENT_COUNTERS */
    646 
    647 /*
    648  * Devices supported by this driver.
    649  */
    650 static const struct dge_product {
    651 	pci_vendor_id_t dgep_vendor;
    652 	pci_product_id_t dgep_product;
    653 	const char *dgep_name;
    654 	int dgep_flags;
    655 #define DGEP_F_10G_LR	  0x01
    656 #define DGEP_F_10G_SR	  0x02
    657 } dge_products[] = {
    658 	{ PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX,
    659 	  "Intel i82597EX 10GbE-LR Ethernet",
    660 	  DGEP_F_10G_LR },
    661 
    662 	{ PCI_VENDOR_INTEL,  PCI_PRODUCT_INTEL_82597EX_SR,
    663 	  "Intel i82597EX 10GbE-SR Ethernet",
    664 	  DGEP_F_10G_SR },
    665 
    666 	{ 0,	    0,
    667 	  NULL,
    668 	  0 },
    669 };
    670 
    671 static const struct dge_product *
    672 dge_lookup(const struct pci_attach_args *pa)
    673 {
    674 	const struct dge_product *dgep;
    675 
    676 	for (dgep = dge_products; dgep->dgep_name != NULL; dgep++) {
    677 		if (PCI_VENDOR(pa->pa_id) == dgep->dgep_vendor &&
    678 		    PCI_PRODUCT(pa->pa_id) == dgep->dgep_product)
    679 			return dgep;
    680 		}
    681 	return NULL;
    682 }
    683 
    684 static int
    685 dge_match(device_t parent, cfdata_t cf, void *aux)
    686 {
    687 	struct pci_attach_args *pa = aux;
    688 
    689 	if (dge_lookup(pa) != NULL)
    690 		return 1;
    691 
    692 	return 0;
    693 }
    694 
    695 static void
    696 dge_attach(device_t parent, device_t self, void *aux)
    697 {
    698 	struct dge_softc *sc = device_private(self);
    699 	struct pci_attach_args *pa = aux;
    700 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
    701 	pci_chipset_tag_t pc = pa->pa_pc;
    702 	pci_intr_handle_t ih;
    703 	const char *intrstr = NULL;
    704 	bus_dma_segment_t seg;
    705 	int i, rseg, error;
    706 	uint8_t enaddr[ETHER_ADDR_LEN];
    707 	pcireg_t preg, memtype;
    708 	uint32_t reg;
    709 	char intrbuf[PCI_INTRSTR_LEN];
    710 	const struct dge_product *dgep;
    711 
    712 	sc->sc_dgep = dgep = dge_lookup(pa);
    713 	if (dgep == NULL) {
    714 		printf("\n");
    715 		panic("dge_attach: impossible");
    716 	}
    717 
    718 	sc->sc_dev = self;
    719 	sc->sc_pc = pa->pa_pc;
    720 	sc->sc_pt = pa->pa_tag;
    721 
    722 	if (pci_dma64_available(pa))
    723 		sc->sc_dmat = pa->pa_dmat64;
    724 	else
    725 		sc->sc_dmat = pa->pa_dmat;
    726 
    727 	pci_aprint_devinfo_fancy(pa, "Ethernet controller",
    728 		dgep->dgep_name, 1);
    729 
    730 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
    731 	if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
    732 	    &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
    733 		aprint_error_dev(sc->sc_dev,
    734 		    "unable to map device registers\n");
    735 		return;
    736 	}
    737 
    738 	/* Enable bus mastering */
    739 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
    740 	preg |= PCI_COMMAND_MASTER_ENABLE;
    741 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
    742 
    743 	/*
    744 	 * Map and establish our interrupt.
    745 	 */
    746 	if (pci_intr_map(pa, &ih)) {
    747 		aprint_error_dev(sc->sc_dev, "unable to map interrupt\n");
    748 		return;
    749 	}
    750 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
    751 	sc->sc_ih = pci_intr_establish_xname(pc, ih, IPL_NET, dge_intr, sc,
    752 	    device_xname(self));
    753 	if (sc->sc_ih == NULL) {
    754 		aprint_error_dev(sc->sc_dev, "unable to establish interrupt");
    755 		if (intrstr != NULL)
    756 			aprint_error(" at %s", intrstr);
    757 		aprint_error("\n");
    758 		return;
    759 	}
    760 	aprint_normal_dev(sc->sc_dev, "interrupting at %s\n", intrstr);
    761 
    762 	/*
    763 	 * Determine a few things about the bus we're connected to.
    764 	 */
    765 	reg = CSR_READ(sc, DGE_STATUS);
    766 	if (reg & STATUS_BUS64)
    767 		sc->sc_flags |= DGE_F_BUS64;
    768 
    769 	sc->sc_flags |= DGE_F_PCIX;
    770 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
    771 			       PCI_CAP_PCIX,
    772 			       &sc->sc_pcix_offset, NULL) == 0)
    773 		aprint_error_dev(sc->sc_dev, "unable to find PCIX "
    774 		    "capability\n");
    775 
    776 	if (sc->sc_flags & DGE_F_PCIX) {
    777 		switch (reg & STATUS_PCIX_MSK) {
    778 		case STATUS_PCIX_66:
    779 			sc->sc_bus_speed = 66;
    780 			break;
    781 		case STATUS_PCIX_100:
    782 			sc->sc_bus_speed = 100;
    783 			break;
    784 		case STATUS_PCIX_133:
    785 			sc->sc_bus_speed = 133;
    786 			break;
    787 		default:
    788 			aprint_error_dev(sc->sc_dev,
    789 			    "unknown PCIXSPD %d; assuming 66MHz\n",
    790 			    reg & STATUS_PCIX_MSK);
    791 			sc->sc_bus_speed = 66;
    792 		}
    793 	} else
    794 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
    795 	aprint_verbose_dev(sc->sc_dev, "%d-bit %dMHz %s bus\n",
    796 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
    797 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
    798 
    799 	/*
    800 	 * Allocate the control data structures, and create and load the
    801 	 * DMA map for it.
    802 	 */
    803 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    804 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
    805 	    0)) != 0) {
    806 		aprint_error_dev(sc->sc_dev,
    807 		    "unable to allocate control data, error = %d\n",
    808 		    error);
    809 		goto fail_0;
    810 	}
    811 
    812 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
    813 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
    814 	    0)) != 0) {
    815 		aprint_error_dev(sc->sc_dev,
    816 		    "unable to map control data, error = %d\n", error);
    817 		goto fail_1;
    818 	}
    819 
    820 	if ((error = bus_dmamap_create(sc->sc_dmat,
    821 	    sizeof(struct dge_control_data), 1,
    822 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
    823 		aprint_error_dev(sc->sc_dev, "unable to create control data "
    824 		    "DMA map, error = %d\n", error);
    825 		goto fail_2;
    826 	}
    827 
    828 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
    829 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
    830 	    0)) != 0) {
    831 		aprint_error_dev(sc->sc_dev,
    832 		    "unable to load control data DMA map, error = %d\n",
    833 		    error);
    834 		goto fail_3;
    835 	}
    836 
    837 #ifdef DGE_OFFBYONE_RXBUG
    838 	if (dge_alloc_rcvmem(sc) != 0)
    839 		return; /* Already complained */
    840 #endif
    841 	/*
    842 	 * Create the transmit buffer DMA maps.
    843 	 */
    844 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
    845 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
    846 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
    847 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
    848 			aprint_error_dev(sc->sc_dev, "unable to create Tx DMA map %d, "
    849 			    "error = %d\n", i, error);
    850 			goto fail_4;
    851 		}
    852 	}
    853 
    854 	/*
    855 	 * Create the receive buffer DMA maps.
    856 	 */
    857 	for (i = 0; i < DGE_NRXDESC; i++) {
    858 #ifdef DGE_OFFBYONE_RXBUG
    859 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
    860 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    861 #else
    862 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
    863 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    864 #endif
    865 			aprint_error_dev(sc->sc_dev, "unable to create Rx DMA "
    866 			    "map %d, error = %d\n", i, error);
    867 			goto fail_5;
    868 		}
    869 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
    870 	}
    871 
    872 	/*
    873 	 * Set bits in ctrl0 register.
    874 	 * Should get the software defined pins out of EEPROM?
    875 	 */
    876 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
    877 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
    878 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
    879 
    880 	/*
    881 	 * Reset the chip to a known state.
    882 	 */
    883 	dge_reset(sc);
    884 
    885 	/*
    886 	 * Reset the PHY.
    887 	 */
    888 	dge_xgmii_reset(sc);
    889 
    890 	/*
    891 	 * Read in EEPROM data.
    892 	 */
    893 	if (dge_read_eeprom(sc)) {
    894 		aprint_error_dev(sc->sc_dev, "couldn't read EEPROM\n");
    895 		return;
    896 	}
    897 
    898 	/*
    899 	 * Get the ethernet address.
    900 	 */
    901 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
    902 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
    903 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
    904 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
    905 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
    906 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
    907 
    908 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    909 	    ether_sprintf(enaddr));
    910 
    911 	/*
    912 	 * Setup media stuff.
    913 	 */
    914 	sc->sc_ethercom.ec_ifmedia = &sc->sc_media;
    915 	ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
    916 	    dge_xgmii_mediastatus);
    917 	if (dgep->dgep_flags & DGEP_F_10G_SR) {
    918 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_10G_SR, 0, NULL);
    919 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_10G_SR);
    920 	} else { /* XXX default is LR */
    921 		ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_10G_LR, 0, NULL);
    922 		ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_10G_LR);
    923 	}
    924 
    925 	ifp = &sc->sc_ethercom.ec_if;
    926 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    927 	ifp->if_softc = sc;
    928 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    929 	ifp->if_ioctl = dge_ioctl;
    930 	ifp->if_start = dge_start;
    931 	ifp->if_watchdog = dge_watchdog;
    932 	ifp->if_init = dge_init;
    933 	ifp->if_stop = dge_stop;
    934 	IFQ_SET_MAXLEN(&ifp->if_snd, uimax(DGE_IFQUEUELEN, IFQ_MAXLEN));
    935 	IFQ_SET_READY(&ifp->if_snd);
    936 
    937 	sc->sc_ethercom.ec_capabilities |=
    938 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
    939 
    940 	/*
    941 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
    942 	 */
    943 	ifp->if_capabilities |=
    944 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    945 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    946 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
    947 
    948 	/*
    949 	 * Attach the interface.
    950 	 */
    951 	if_attach(ifp);
    952 	if_deferred_start_init(ifp, NULL);
    953 	ether_ifattach(ifp, enaddr);
    954 	rnd_attach_source(&sc->rnd_source, device_xname(sc->sc_dev),
    955 	    RND_TYPE_NET, RND_FLAG_DEFAULT);
    956 
    957 #ifdef DGE_EVENT_COUNTERS
    958 	/* Fix segment event naming */
    959 	if (dge_txseg_evcnt_names == NULL) {
    960 		dge_txseg_evcnt_names =
    961 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
    962 		for (i = 0; i < DGE_NTXSEGS; i++)
    963 			snprintf((*dge_txseg_evcnt_names)[i],
    964 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
    965 	}
    966 
    967 	/* Attach event counters. */
    968 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
    969 	    NULL, device_xname(sc->sc_dev), "txsstall");
    970 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
    971 	    NULL, device_xname(sc->sc_dev), "txdstall");
    972 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
    973 	    NULL, device_xname(sc->sc_dev), "txforceintr");
    974 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
    975 	    NULL, device_xname(sc->sc_dev), "txdw");
    976 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
    977 	    NULL, device_xname(sc->sc_dev), "txqe");
    978 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
    979 	    NULL, device_xname(sc->sc_dev), "rxintr");
    980 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
    981 	    NULL, device_xname(sc->sc_dev), "linkintr");
    982 
    983 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
    984 	    NULL, device_xname(sc->sc_dev), "rxipsum");
    985 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
    986 	    NULL, device_xname(sc->sc_dev), "rxtusum");
    987 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
    988 	    NULL, device_xname(sc->sc_dev), "txipsum");
    989 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
    990 	    NULL, device_xname(sc->sc_dev), "txtusum");
    991 
    992 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
    993 	    NULL, device_xname(sc->sc_dev), "txctx init");
    994 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
    995 	    NULL, device_xname(sc->sc_dev), "txctx hit");
    996 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
    997 	    NULL, device_xname(sc->sc_dev), "txctx miss");
    998 
    999 	for (i = 0; i < DGE_NTXSEGS; i++)
   1000 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
   1001 		    NULL, device_xname(sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
   1002 
   1003 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
   1004 	    NULL, device_xname(sc->sc_dev), "txdrop");
   1005 
   1006 #endif /* DGE_EVENT_COUNTERS */
   1007 
   1008 	/*
   1009 	 * Make sure the interface is shutdown during reboot.
   1010 	 */
   1011 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
   1012 		pmf_class_network_register(self, ifp);
   1013 	else
   1014 		aprint_error_dev(self, "couldn't establish power handler\n");
   1015 
   1016 	return;
   1017 
   1018 	/*
   1019 	 * Free any resources we've allocated during the failed attach
   1020 	 * attempt.  Do this in reverse order and fall through.
   1021 	 */
   1022  fail_5:
   1023 	for (i = 0; i < DGE_NRXDESC; i++) {
   1024 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
   1025 			bus_dmamap_destroy(sc->sc_dmat,
   1026 			    sc->sc_rxsoft[i].rxs_dmamap);
   1027 	}
   1028  fail_4:
   1029 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
   1030 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
   1031 			bus_dmamap_destroy(sc->sc_dmat,
   1032 			    sc->sc_txsoft[i].txs_dmamap);
   1033 	}
   1034 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
   1035  fail_3:
   1036 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
   1037  fail_2:
   1038 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
   1039 	    sizeof(struct dge_control_data));
   1040  fail_1:
   1041 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
   1042  fail_0:
   1043 	return;
   1044 }
   1045 
   1046 /*
   1047  * dge_shutdown:
   1048  *
   1049  *	Make sure the interface is stopped at reboot time.
   1050  */
   1051 static bool
   1052 dge_shutdown(device_t self, int howto)
   1053 {
   1054 	struct dge_softc *sc;
   1055 
   1056 	sc = device_private(self);
   1057 	dge_stop(&sc->sc_ethercom.ec_if, 1);
   1058 
   1059 	return true;
   1060 }
   1061 
   1062 /*
   1063  * dge_tx_cksum:
   1064  *
   1065  *	Set up TCP/IP checksumming parameters for the
   1066  *	specified packet.
   1067  */
   1068 static int
   1069 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
   1070 {
   1071 	struct mbuf *m0 = txs->txs_mbuf;
   1072 	struct dge_ctdes *t;
   1073 	uint32_t ipcs, tucs;
   1074 	struct ether_header *eh;
   1075 	int offset, iphl;
   1076 	uint8_t fields = 0;
   1077 
   1078 	/*
   1079 	 * XXX It would be nice if the mbuf pkthdr had offset
   1080 	 * fields for the protocol headers.
   1081 	 */
   1082 
   1083 	eh = mtod(m0, struct ether_header *);
   1084 	switch (htons(eh->ether_type)) {
   1085 	case ETHERTYPE_IP:
   1086 		offset = ETHER_HDR_LEN;
   1087 		break;
   1088 
   1089 	case ETHERTYPE_VLAN:
   1090 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
   1091 		break;
   1092 
   1093 	default:
   1094 		/*
   1095 		 * Don't support this protocol or encapsulation.
   1096 		 */
   1097 		*fieldsp = 0;
   1098 		return 0;
   1099 	}
   1100 
   1101 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
   1102 
   1103 	/*
   1104 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
   1105 	 * offload feature, if we load the context descriptor, we
   1106 	 * MUST provide valid values for IPCSS and TUCSS fields.
   1107 	 */
   1108 
   1109 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
   1110 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
   1111 		fields |= TDESC_POPTS_IXSM;
   1112 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1113 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1114 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1115 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
   1116 		/* Use the cached value. */
   1117 		ipcs = sc->sc_txctx_ipcs;
   1118 	} else {
   1119 		/* Just initialize it to the likely value anyway. */
   1120 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1121 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1122 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1123 	}
   1124 	DPRINTF(DGE_DEBUG_CKSUM,
   1125 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
   1126 	    device_xname(sc->sc_dev), offset, ipcs));
   1127 
   1128 	offset += iphl;
   1129 
   1130 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4 | M_CSUM_UDPv4)) {
   1131 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
   1132 		fields |= TDESC_POPTS_TXSM;
   1133 		tucs = DGE_TCPIP_TUCSS(offset) |
   1134 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
   1135 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1136 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
   1137 		/* Use the cached value. */
   1138 		tucs = sc->sc_txctx_tucs;
   1139 	} else {
   1140 		/* Just initialize it to a valid TCP context. */
   1141 		tucs = DGE_TCPIP_TUCSS(offset) |
   1142 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
   1143 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1144 	}
   1145 
   1146 	DPRINTF(DGE_DEBUG_CKSUM,
   1147 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
   1148 	    device_xname(sc->sc_dev), offset, tucs));
   1149 
   1150 	if (sc->sc_txctx_ipcs == ipcs &&
   1151 	    sc->sc_txctx_tucs == tucs) {
   1152 		/* Cached context is fine. */
   1153 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
   1154 	} else {
   1155 		/* Fill in the context descriptor. */
   1156 #ifdef DGE_EVENT_COUNTERS
   1157 		if (sc->sc_txctx_ipcs == 0xffffffff &&
   1158 		    sc->sc_txctx_tucs == 0xffffffff)
   1159 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
   1160 		else
   1161 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
   1162 #endif
   1163 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
   1164 		t->dc_tcpip_ipcs = htole32(ipcs);
   1165 		t->dc_tcpip_tucs = htole32(tucs);
   1166 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
   1167 		t->dc_tcpip_seg = 0;
   1168 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
   1169 
   1170 		sc->sc_txctx_ipcs = ipcs;
   1171 		sc->sc_txctx_tucs = tucs;
   1172 
   1173 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
   1174 		txs->txs_ndesc++;
   1175 	}
   1176 
   1177 	*fieldsp = fields;
   1178 
   1179 	return 0;
   1180 }
   1181 
   1182 /*
   1183  * dge_start:		[ifnet interface function]
   1184  *
   1185  *	Start packet transmission on the interface.
   1186  */
   1187 static void
   1188 dge_start(struct ifnet *ifp)
   1189 {
   1190 	struct dge_softc *sc = ifp->if_softc;
   1191 	struct mbuf *m0;
   1192 	struct dge_txsoft *txs;
   1193 	bus_dmamap_t dmamap;
   1194 	int error, nexttx, lasttx = -1, ofree, seg;
   1195 	uint32_t cksumcmd;
   1196 	uint8_t cksumfields;
   1197 
   1198 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
   1199 		return;
   1200 
   1201 	/*
   1202 	 * Remember the previous number of free descriptors.
   1203 	 */
   1204 	ofree = sc->sc_txfree;
   1205 
   1206 	/*
   1207 	 * Loop through the send queue, setting up transmit descriptors
   1208 	 * until we drain the queue, or use up all available transmit
   1209 	 * descriptors.
   1210 	 */
   1211 	for (;;) {
   1212 		/* Grab a packet off the queue. */
   1213 		IFQ_POLL(&ifp->if_snd, m0);
   1214 		if (m0 == NULL)
   1215 			break;
   1216 
   1217 		DPRINTF(DGE_DEBUG_TX,
   1218 		    ("%s: TX: have packet to transmit: %p\n",
   1219 		    device_xname(sc->sc_dev), m0));
   1220 
   1221 		/* Get a work queue entry. */
   1222 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
   1223 			dge_txintr(sc);
   1224 			if (sc->sc_txsfree == 0) {
   1225 				DPRINTF(DGE_DEBUG_TX,
   1226 				    ("%s: TX: no free job descriptors\n",
   1227 					device_xname(sc->sc_dev)));
   1228 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
   1229 				break;
   1230 			}
   1231 		}
   1232 
   1233 		txs = &sc->sc_txsoft[sc->sc_txsnext];
   1234 		dmamap = txs->txs_dmamap;
   1235 
   1236 		/*
   1237 		 * Load the DMA map.  If this fails, the packet either
   1238 		 * didn't fit in the allotted number of segments, or we
   1239 		 * were short on resources.  For the too-many-segments
   1240 		 * case, we simply report an error and drop the packet,
   1241 		 * since we can't sanely copy a jumbo packet to a single
   1242 		 * buffer.
   1243 		 */
   1244 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
   1245 		    BUS_DMA_WRITE | BUS_DMA_NOWAIT);
   1246 		if (error) {
   1247 			if (error == EFBIG) {
   1248 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
   1249 				printf("%s: Tx packet consumes too many "
   1250 				    "DMA segments, dropping...\n",
   1251 				    device_xname(sc->sc_dev));
   1252 				IFQ_DEQUEUE(&ifp->if_snd, m0);
   1253 				m_freem(m0);
   1254 				continue;
   1255 			}
   1256 			/*
   1257 			 * Short on resources, just stop for now.
   1258 			 */
   1259 			DPRINTF(DGE_DEBUG_TX,
   1260 			    ("%s: TX: dmamap load failed: %d\n",
   1261 			    device_xname(sc->sc_dev), error));
   1262 			break;
   1263 		}
   1264 
   1265 		/*
   1266 		 * Ensure we have enough descriptors free to describe
   1267 		 * the packet.  Note, we always reserve one descriptor
   1268 		 * at the end of the ring due to the semantics of the
   1269 		 * TDT register, plus one more in the event we need
   1270 		 * to re-load checksum offload context.
   1271 		 */
   1272 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
   1273 			/*
   1274 			 * Not enough free descriptors to transmit this
   1275 			 * packet.  We haven't committed anything yet,
   1276 			 * so just unload the DMA map, put the packet
   1277 			 * pack on the queue, and punt.  Notify the upper
   1278 			 * layer that there are no more slots left.
   1279 			 */
   1280 			DPRINTF(DGE_DEBUG_TX,
   1281 			    ("%s: TX: need %d descriptors, have %d\n",
   1282 			    device_xname(sc->sc_dev), dmamap->dm_nsegs,
   1283 			    sc->sc_txfree - 1));
   1284 			ifp->if_flags |= IFF_OACTIVE;
   1285 			bus_dmamap_unload(sc->sc_dmat, dmamap);
   1286 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
   1287 			break;
   1288 		}
   1289 
   1290 		IFQ_DEQUEUE(&ifp->if_snd, m0);
   1291 
   1292 		/*
   1293 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
   1294 		 */
   1295 
   1296 		/* Sync the DMA map. */
   1297 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
   1298 		    BUS_DMASYNC_PREWRITE);
   1299 
   1300 		DPRINTF(DGE_DEBUG_TX,
   1301 		    ("%s: TX: packet has %d DMA segments\n",
   1302 		    device_xname(sc->sc_dev), dmamap->dm_nsegs));
   1303 
   1304 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
   1305 
   1306 		/*
   1307 		 * Store a pointer to the packet so that we can free it
   1308 		 * later.
   1309 		 *
   1310 		 * Initially, we consider the number of descriptors the
   1311 		 * packet uses the number of DMA segments.  This may be
   1312 		 * incremented by 1 if we do checksum offload (a descriptor
   1313 		 * is used to set the checksum context).
   1314 		 */
   1315 		txs->txs_mbuf = m0;
   1316 		txs->txs_firstdesc = sc->sc_txnext;
   1317 		txs->txs_ndesc = dmamap->dm_nsegs;
   1318 
   1319 		/*
   1320 		 * Set up checksum offload parameters for
   1321 		 * this packet.
   1322 		 */
   1323 		if (m0->m_pkthdr.csum_flags &
   1324 		    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4)) {
   1325 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
   1326 				/* Error message already displayed. */
   1327 				bus_dmamap_unload(sc->sc_dmat, dmamap);
   1328 				continue;
   1329 			}
   1330 		} else {
   1331 			cksumfields = 0;
   1332 		}
   1333 
   1334 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
   1335 
   1336 		/*
   1337 		 * Initialize the transmit descriptor.
   1338 		 */
   1339 		for (nexttx = sc->sc_txnext, seg = 0;
   1340 		     seg < dmamap->dm_nsegs;
   1341 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
   1342 			sc->sc_txdescs[nexttx].dt_baddrh =
   1343 			    htole32(((uint64_t)dmamap->dm_segs[seg].ds_addr) >> 32);
   1344 			sc->sc_txdescs[nexttx].dt_baddrl =
   1345 			    htole32(dmamap->dm_segs[seg].ds_addr);
   1346 			sc->sc_txdescs[nexttx].dt_ctl =
   1347 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
   1348 			sc->sc_txdescs[nexttx].dt_status = 0;
   1349 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
   1350 			sc->sc_txdescs[nexttx].dt_vlan = 0;
   1351 			lasttx = nexttx;
   1352 
   1353 			DPRINTF(DGE_DEBUG_TX,
   1354 			    ("%s: TX: desc %d: high 0x%08lx, low 0x%08lx, len 0x%04lx\n",
   1355 			    device_xname(sc->sc_dev), nexttx,
   1356 			    (unsigned long)(((uint64_t)dmamap->dm_segs[seg].ds_addr) >> 32),
   1357 			    (unsigned long)((uint32_t)dmamap->dm_segs[seg].ds_addr),
   1358 			    (unsigned long)dmamap->dm_segs[seg].ds_len));
   1359 		}
   1360 
   1361 		KASSERT(lasttx != -1);
   1362 
   1363 		/*
   1364 		 * Set up the command byte on the last descriptor of
   1365 		 * the packet.  If we're in the interrupt delay window,
   1366 		 * delay the interrupt.
   1367 		 */
   1368 		sc->sc_txdescs[lasttx].dt_ctl |=
   1369 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
   1370 
   1371 		txs->txs_lastdesc = lasttx;
   1372 
   1373 		DPRINTF(DGE_DEBUG_TX,
   1374 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(sc->sc_dev),
   1375 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
   1376 
   1377 		/* Sync the descriptors we're using. */
   1378 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
   1379 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
   1380 
   1381 		/* Give the packet to the chip. */
   1382 		CSR_WRITE(sc, DGE_TDT, nexttx);
   1383 
   1384 		DPRINTF(DGE_DEBUG_TX,
   1385 		    ("%s: TX: TDT -> %d\n", device_xname(sc->sc_dev), nexttx));
   1386 
   1387 		DPRINTF(DGE_DEBUG_TX,
   1388 		    ("%s: TX: finished transmitting packet, job %d\n",
   1389 		    device_xname(sc->sc_dev), sc->sc_txsnext));
   1390 
   1391 		/* Advance the tx pointer. */
   1392 		sc->sc_txfree -= txs->txs_ndesc;
   1393 		sc->sc_txnext = nexttx;
   1394 
   1395 		sc->sc_txsfree--;
   1396 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
   1397 
   1398 		/* Pass the packet to any BPF listeners. */
   1399 		bpf_mtap(ifp, m0, BPF_D_OUT);
   1400 	}
   1401 
   1402 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
   1403 		/* No more slots; notify upper layer. */
   1404 		ifp->if_flags |= IFF_OACTIVE;
   1405 	}
   1406 
   1407 	if (sc->sc_txfree != ofree) {
   1408 		/* Set a watchdog timer in case the chip flakes out. */
   1409 		ifp->if_timer = 5;
   1410 	}
   1411 }
   1412 
   1413 /*
   1414  * dge_watchdog:		[ifnet interface function]
   1415  *
   1416  *	Watchdog timer handler.
   1417  */
   1418 static void
   1419 dge_watchdog(struct ifnet *ifp)
   1420 {
   1421 	struct dge_softc *sc = ifp->if_softc;
   1422 
   1423 	/*
   1424 	 * Since we're using delayed interrupts, sweep up
   1425 	 * before we report an error.
   1426 	 */
   1427 	dge_txintr(sc);
   1428 
   1429 	if (sc->sc_txfree != DGE_NTXDESC) {
   1430 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
   1431 		    device_xname(sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
   1432 		    sc->sc_txnext);
   1433 		if_statinc(ifp, if_oerrors);
   1434 
   1435 		/* Reset the interface. */
   1436 		(void) dge_init(ifp);
   1437 	}
   1438 
   1439 	/* Try to get more packets going. */
   1440 	dge_start(ifp);
   1441 }
   1442 
   1443 /*
   1444  * dge_ioctl:		[ifnet interface function]
   1445  *
   1446  *	Handle control requests from the operator.
   1447  */
   1448 static int
   1449 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   1450 {
   1451 	struct dge_softc *sc = ifp->if_softc;
   1452 	struct ifreq *ifr = (struct ifreq *) data;
   1453 	pcireg_t preg;
   1454 	int s, error, mmrbc;
   1455 
   1456 	s = splnet();
   1457 
   1458 	switch (cmd) {
   1459 	case SIOCSIFMTU:
   1460 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
   1461 			error = EINVAL;
   1462 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
   1463 			break;
   1464 		else if (ifp->if_flags & IFF_UP)
   1465 			error = (*ifp->if_init)(ifp);
   1466 		else
   1467 			error = 0;
   1468 		break;
   1469 
   1470 	case SIOCSIFFLAGS:
   1471 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   1472 			break;
   1473 		/* extract link flags */
   1474 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1475 		    (ifp->if_flags & IFF_LINK1) == 0)
   1476 			mmrbc = PCIX_MMRBC_512;
   1477 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1478 		    (ifp->if_flags & IFF_LINK1) != 0)
   1479 			mmrbc = PCIX_MMRBC_1024;
   1480 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
   1481 		    (ifp->if_flags & IFF_LINK1) == 0)
   1482 			mmrbc = PCIX_MMRBC_2048;
   1483 		else
   1484 			mmrbc = PCIX_MMRBC_4096;
   1485 		if (mmrbc != sc->sc_mmrbc) {
   1486 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
   1487 			preg &= ~PCIX_MMRBC_MSK;
   1488 			preg |= mmrbc;
   1489 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
   1490 			sc->sc_mmrbc = mmrbc;
   1491 		}
   1492 		/* FALLTHROUGH */
   1493 	default:
   1494 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
   1495 			break;
   1496 
   1497 		error = 0;
   1498 
   1499 		if (cmd == SIOCSIFCAP)
   1500 			error = (*ifp->if_init)(ifp);
   1501 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
   1502 			;
   1503 		else if (ifp->if_flags & IFF_RUNNING) {
   1504 			/*
   1505 			 * Multicast list has changed; set the hardware filter
   1506 			 * accordingly.
   1507 			 */
   1508 			dge_set_filter(sc);
   1509 		}
   1510 		break;
   1511 	}
   1512 
   1513 	/* Try to get more packets going. */
   1514 	dge_start(ifp);
   1515 
   1516 	splx(s);
   1517 	return error;
   1518 }
   1519 
   1520 /*
   1521  * dge_intr:
   1522  *
   1523  *	Interrupt service routine.
   1524  */
   1525 static int
   1526 dge_intr(void *arg)
   1527 {
   1528 	struct dge_softc *sc = arg;
   1529 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1530 	uint32_t icr;
   1531 	int wantinit, handled = 0;
   1532 
   1533 	for (wantinit = 0; wantinit == 0;) {
   1534 		icr = CSR_READ(sc, DGE_ICR);
   1535 		if ((icr & sc->sc_icr) == 0)
   1536 			break;
   1537 
   1538 		rnd_add_uint32(&sc->rnd_source, icr);
   1539 
   1540 		handled = 1;
   1541 
   1542 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1543 		if (icr & (ICR_RXDMT0 | ICR_RXT0)) {
   1544 			DPRINTF(DGE_DEBUG_RX,
   1545 			    ("%s: RX: got Rx intr 0x%08x\n",
   1546 			    device_xname(sc->sc_dev),
   1547 			    icr & (ICR_RXDMT0 | ICR_RXT0)));
   1548 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
   1549 		}
   1550 #endif
   1551 		dge_rxintr(sc);
   1552 
   1553 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1554 		if (icr & ICR_TXDW) {
   1555 			DPRINTF(DGE_DEBUG_TX,
   1556 			    ("%s: TX: got TXDW interrupt\n",
   1557 			    device_xname(sc->sc_dev)));
   1558 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
   1559 		}
   1560 		if (icr & ICR_TXQE)
   1561 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
   1562 #endif
   1563 		dge_txintr(sc);
   1564 
   1565 		if (icr & (ICR_LSC | ICR_RXSEQ)) {
   1566 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
   1567 			dge_linkintr(sc, icr);
   1568 		}
   1569 
   1570 		if (icr & ICR_RXO) {
   1571 			printf("%s: Receive overrun\n",
   1572 			    device_xname(sc->sc_dev));
   1573 			wantinit = 1;
   1574 		}
   1575 	}
   1576 
   1577 	if (handled) {
   1578 		if (wantinit)
   1579 			dge_init(ifp);
   1580 
   1581 		/* Try to get more packets going. */
   1582 		if_schedule_deferred_start(ifp);
   1583 	}
   1584 
   1585 	return handled;
   1586 }
   1587 
   1588 /*
   1589  * dge_txintr:
   1590  *
   1591  *	Helper; handle transmit interrupts.
   1592  */
   1593 static void
   1594 dge_txintr(struct dge_softc *sc)
   1595 {
   1596 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1597 	struct dge_txsoft *txs;
   1598 	uint8_t status;
   1599 	int i;
   1600 
   1601 	ifp->if_flags &= ~IFF_OACTIVE;
   1602 
   1603 	/*
   1604 	 * Go through the Tx list and free mbufs for those
   1605 	 * frames which have been transmitted.
   1606 	 */
   1607 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
   1608 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
   1609 		txs = &sc->sc_txsoft[i];
   1610 
   1611 		DPRINTF(DGE_DEBUG_TX,
   1612 		    ("%s: TX: checking job %d\n", device_xname(sc->sc_dev), i));
   1613 
   1614 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
   1615 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
   1616 
   1617 		status =
   1618 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
   1619 		if ((status & TDESC_STA_DD) == 0) {
   1620 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
   1621 			    BUS_DMASYNC_PREREAD);
   1622 			break;
   1623 		}
   1624 
   1625 		DPRINTF(DGE_DEBUG_TX,
   1626 		    ("%s: TX: job %d done: descs %d..%d\n",
   1627 		    device_xname(sc->sc_dev), i, txs->txs_firstdesc,
   1628 		    txs->txs_lastdesc));
   1629 
   1630 		if_statinc(ifp, if_opackets);
   1631 		sc->sc_txfree += txs->txs_ndesc;
   1632 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
   1633 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1634 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   1635 		m_freem(txs->txs_mbuf);
   1636 		txs->txs_mbuf = NULL;
   1637 	}
   1638 
   1639 	/* Update the dirty transmit buffer pointer. */
   1640 	sc->sc_txsdirty = i;
   1641 	DPRINTF(DGE_DEBUG_TX,
   1642 	    ("%s: TX: txsdirty -> %d\n", device_xname(sc->sc_dev), i));
   1643 
   1644 	/*
   1645 	 * If there are no more pending transmissions, cancel the watchdog
   1646 	 * timer.
   1647 	 */
   1648 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
   1649 		ifp->if_timer = 0;
   1650 }
   1651 
   1652 /*
   1653  * dge_rxintr:
   1654  *
   1655  *	Helper; handle receive interrupts.
   1656  */
   1657 static void
   1658 dge_rxintr(struct dge_softc *sc)
   1659 {
   1660 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1661 	struct dge_rxsoft *rxs;
   1662 	struct mbuf *m;
   1663 	int i, len;
   1664 	uint8_t status, errors;
   1665 
   1666 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
   1667 		rxs = &sc->sc_rxsoft[i];
   1668 
   1669 		DPRINTF(DGE_DEBUG_RX,
   1670 		    ("%s: RX: checking descriptor %d\n",
   1671 		    device_xname(sc->sc_dev), i));
   1672 
   1673 		DGE_CDRXSYNC(sc, i,
   1674 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
   1675 
   1676 		status = sc->sc_rxdescs[i].dr_status;
   1677 		errors = sc->sc_rxdescs[i].dr_errors;
   1678 		len = le16toh(sc->sc_rxdescs[i].dr_len);
   1679 
   1680 		if ((status & RDESC_STS_DD) == 0) {
   1681 			/* We have processed all of the receive descriptors. */
   1682 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1683 			break;
   1684 		}
   1685 
   1686 		if (__predict_false(sc->sc_rxdiscard)) {
   1687 			DPRINTF(DGE_DEBUG_RX,
   1688 			    ("%s: RX: discarding contents of descriptor %d\n",
   1689 			    device_xname(sc->sc_dev), i));
   1690 			DGE_INIT_RXDESC(sc, i);
   1691 			if (status & RDESC_STS_EOP) {
   1692 				/* Reset our state. */
   1693 				DPRINTF(DGE_DEBUG_RX,
   1694 				    ("%s: RX: resetting rxdiscard -> 0\n",
   1695 				    device_xname(sc->sc_dev)));
   1696 				sc->sc_rxdiscard = 0;
   1697 			}
   1698 			continue;
   1699 		}
   1700 
   1701 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1702 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
   1703 
   1704 		m = rxs->rxs_mbuf;
   1705 
   1706 		/*
   1707 		 * Add a new receive buffer to the ring.
   1708 		 */
   1709 		if (dge_add_rxbuf(sc, i) != 0) {
   1710 			/*
   1711 			 * Failed, throw away what we've done so
   1712 			 * far, and discard the rest of the packet.
   1713 			 */
   1714 			if_statinc(ifp, if_ierrors);
   1715 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1716 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   1717 			DGE_INIT_RXDESC(sc, i);
   1718 			if ((status & RDESC_STS_EOP) == 0)
   1719 				sc->sc_rxdiscard = 1;
   1720 			if (sc->sc_rxhead != NULL)
   1721 				m_freem(sc->sc_rxhead);
   1722 			DGE_RXCHAIN_RESET(sc);
   1723 			DPRINTF(DGE_DEBUG_RX,
   1724 			    ("%s: RX: Rx buffer allocation failed, "
   1725 			    "dropping packet%s\n", device_xname(sc->sc_dev),
   1726 			    sc->sc_rxdiscard ? " (discard)" : ""));
   1727 			continue;
   1728 		}
   1729 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
   1730 
   1731 		DGE_RXCHAIN_LINK(sc, m);
   1732 
   1733 		m->m_len = len;
   1734 
   1735 		DPRINTF(DGE_DEBUG_RX,
   1736 		    ("%s: RX: buffer at %p len %d\n",
   1737 		    device_xname(sc->sc_dev), m->m_data, len));
   1738 
   1739 		/*
   1740 		 * If this is not the end of the packet, keep
   1741 		 * looking.
   1742 		 */
   1743 		if ((status & RDESC_STS_EOP) == 0) {
   1744 			sc->sc_rxlen += len;
   1745 			DPRINTF(DGE_DEBUG_RX,
   1746 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
   1747 			    device_xname(sc->sc_dev), sc->sc_rxlen));
   1748 			continue;
   1749 		}
   1750 
   1751 		/*
   1752 		 * Okay, we have the entire packet now...
   1753 		 */
   1754 		*sc->sc_rxtailp = NULL;
   1755 		m = sc->sc_rxhead;
   1756 		len += sc->sc_rxlen;
   1757 
   1758 		DGE_RXCHAIN_RESET(sc);
   1759 
   1760 		DPRINTF(DGE_DEBUG_RX,
   1761 		    ("%s: RX: have entire packet, len -> %d\n",
   1762 		    device_xname(sc->sc_dev), len));
   1763 
   1764 		/*
   1765 		 * If an error occurred, update stats and drop the packet.
   1766 		 */
   1767 		if (errors & (RDESC_ERR_CE | RDESC_ERR_SE | RDESC_ERR_P |
   1768 		    RDESC_ERR_RXE)) {
   1769 			if_statinc(ifp, if_ierrors);
   1770 			if (errors & RDESC_ERR_SE)
   1771 				printf("%s: symbol error\n",
   1772 				    device_xname(sc->sc_dev));
   1773 			else if (errors & RDESC_ERR_P)
   1774 				printf("%s: parity error\n",
   1775 				    device_xname(sc->sc_dev));
   1776 			else if (errors & RDESC_ERR_CE)
   1777 				printf("%s: CRC error\n",
   1778 				    device_xname(sc->sc_dev));
   1779 			m_freem(m);
   1780 			continue;
   1781 		}
   1782 
   1783 		/*
   1784 		 * No errors.  Receive the packet.
   1785 		 */
   1786 		m_set_rcvif(m, ifp);
   1787 		m->m_pkthdr.len = len;
   1788 
   1789 		/*
   1790 		 * Set up checksum info for this packet.
   1791 		 */
   1792 		if (status & RDESC_STS_IPCS) {
   1793 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
   1794 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1795 			if (errors & RDESC_ERR_IPE)
   1796 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1797 		}
   1798 		if (status & RDESC_STS_TCPCS) {
   1799 			/*
   1800 			 * Note: we don't know if this was TCP or UDP,
   1801 			 * so we just set both bits, and expect the
   1802 			 * upper layers to deal.
   1803 			 */
   1804 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
   1805 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4 | M_CSUM_UDPv4;
   1806 			if (errors & RDESC_ERR_TCPE)
   1807 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1808 		}
   1809 
   1810 		/* Pass it on. */
   1811 		if_percpuq_enqueue(ifp->if_percpuq, m);
   1812 	}
   1813 
   1814 	/* Update the receive pointer. */
   1815 	sc->sc_rxptr = i;
   1816 
   1817 	DPRINTF(DGE_DEBUG_RX,
   1818 	    ("%s: RX: rxptr -> %d\n", device_xname(sc->sc_dev), i));
   1819 }
   1820 
   1821 /*
   1822  * dge_linkintr:
   1823  *
   1824  *	Helper; handle link interrupts.
   1825  */
   1826 static void
   1827 dge_linkintr(struct dge_softc *sc, uint32_t icr)
   1828 {
   1829 	uint32_t status;
   1830 
   1831 	if (icr & ICR_LSC) {
   1832 		status = CSR_READ(sc, DGE_STATUS);
   1833 		if (status & STATUS_LINKUP) {
   1834 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
   1835 			    device_xname(sc->sc_dev)));
   1836 		} else {
   1837 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
   1838 			    device_xname(sc->sc_dev)));
   1839 		}
   1840 	} else if (icr & ICR_RXSEQ) {
   1841 		DPRINTF(DGE_DEBUG_LINK,
   1842 		    ("%s: LINK: Receive sequence error\n",
   1843 		    device_xname(sc->sc_dev)));
   1844 	}
   1845 	/* XXX - fix errata */
   1846 }
   1847 
   1848 /*
   1849  * dge_reset:
   1850  *
   1851  *	Reset the i82597 chip.
   1852  */
   1853 static void
   1854 dge_reset(struct dge_softc *sc)
   1855 {
   1856 	int i;
   1857 
   1858 	/*
   1859 	 * Do a chip reset.
   1860 	 */
   1861 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
   1862 
   1863 	delay(10000);
   1864 
   1865 	for (i = 0; i < 1000; i++) {
   1866 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
   1867 			break;
   1868 		delay(20);
   1869 	}
   1870 
   1871 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
   1872 		printf("%s: WARNING: reset failed to complete\n",
   1873 		    device_xname(sc->sc_dev));
   1874 	/*
   1875 	 * Reset the EEPROM logic.
   1876 	 * This will cause the chip to reread its default values,
   1877 	 * which doesn't happen otherwise (errata).
   1878 	 */
   1879 	CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
   1880 	delay(10000);
   1881 }
   1882 
   1883 /*
   1884  * dge_init:		[ifnet interface function]
   1885  *
   1886  *	Initialize the interface.  Must be called at splnet().
   1887  */
   1888 static int
   1889 dge_init(struct ifnet *ifp)
   1890 {
   1891 	struct dge_softc *sc = ifp->if_softc;
   1892 	struct dge_rxsoft *rxs;
   1893 	int i, error = 0;
   1894 	uint32_t reg;
   1895 
   1896 	/*
   1897 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
   1898 	 * There is a small but measurable benefit to avoiding the adjusment
   1899 	 * of the descriptor so that the headers are aligned, for normal mtu,
   1900 	 * on such platforms.  One possibility is that the DMA itself is
   1901 	 * slightly more efficient if the front of the entire packet (instead
   1902 	 * of the front of the headers) is aligned.
   1903 	 *
   1904 	 * Note we must always set align_tweak to 0 if we are using
   1905 	 * jumbo frames.
   1906 	 */
   1907 #ifdef __NO_STRICT_ALIGNMENT
   1908 	sc->sc_align_tweak = 0;
   1909 #else
   1910 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
   1911 		sc->sc_align_tweak = 0;
   1912 	else
   1913 		sc->sc_align_tweak = 2;
   1914 #endif /* __NO_STRICT_ALIGNMENT */
   1915 
   1916 	/* Cancel any pending I/O. */
   1917 	dge_stop(ifp, 0);
   1918 
   1919 	/* Reset the chip to a known state. */
   1920 	dge_reset(sc);
   1921 
   1922 	/* Initialize the transmit descriptor ring. */
   1923 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
   1924 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
   1925 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
   1926 	sc->sc_txfree = DGE_NTXDESC;
   1927 	sc->sc_txnext = 0;
   1928 
   1929 	sc->sc_txctx_ipcs = 0xffffffff;
   1930 	sc->sc_txctx_tucs = 0xffffffff;
   1931 
   1932 	CSR_WRITE(sc, DGE_TDBAH, ((uint64_t)DGE_CDTXADDR(sc, 0)) >> 32);
   1933 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
   1934 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
   1935 	CSR_WRITE(sc, DGE_TDH, 0);
   1936 	CSR_WRITE(sc, DGE_TDT, 0);
   1937 	CSR_WRITE(sc, DGE_TIDV, TIDV);
   1938 
   1939 #if 0
   1940 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
   1941 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
   1942 #endif
   1943 	CSR_WRITE(sc, DGE_RXDCTL,
   1944 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
   1945 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
   1946 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
   1947 
   1948 	/* Initialize the transmit job descriptors. */
   1949 	for (i = 0; i < DGE_TXQUEUELEN; i++)
   1950 		sc->sc_txsoft[i].txs_mbuf = NULL;
   1951 	sc->sc_txsfree = DGE_TXQUEUELEN;
   1952 	sc->sc_txsnext = 0;
   1953 	sc->sc_txsdirty = 0;
   1954 
   1955 	/*
   1956 	 * Initialize the receive descriptor and receive job
   1957 	 * descriptor rings.
   1958 	 */
   1959 	CSR_WRITE(sc, DGE_RDBAH, ((uint64_t)DGE_CDRXADDR(sc, 0)) >> 32);
   1960 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
   1961 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
   1962 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
   1963 	CSR_WRITE(sc, DGE_RDT, 0);
   1964 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
   1965 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
   1966 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
   1967 
   1968 	for (i = 0; i < DGE_NRXDESC; i++) {
   1969 		rxs = &sc->sc_rxsoft[i];
   1970 		if (rxs->rxs_mbuf == NULL) {
   1971 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
   1972 				printf("%s: unable to allocate or map rx "
   1973 				    "buffer %d, error = %d\n",
   1974 				    device_xname(sc->sc_dev), i, error);
   1975 				/*
   1976 				 * XXX Should attempt to run with fewer receive
   1977 				 * XXX buffers instead of just failing.
   1978 				 */
   1979 				dge_rxdrain(sc);
   1980 				goto out;
   1981 			}
   1982 		}
   1983 		DGE_INIT_RXDESC(sc, i);
   1984 	}
   1985 	sc->sc_rxptr = DGE_RXSPACE;
   1986 	sc->sc_rxdiscard = 0;
   1987 	DGE_RXCHAIN_RESET(sc);
   1988 
   1989 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
   1990 		sc->sc_ctrl0 |= CTRL0_JFE;
   1991 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
   1992 	}
   1993 
   1994 	/* Write the control registers. */
   1995 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
   1996 
   1997 	/*
   1998 	 * Set up checksum offload parameters.
   1999 	 */
   2000 	reg = CSR_READ(sc, DGE_RXCSUM);
   2001 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
   2002 		reg |= RXCSUM_IPOFL;
   2003 	else
   2004 		reg &= ~RXCSUM_IPOFL;
   2005 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
   2006 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
   2007 	else {
   2008 		reg &= ~RXCSUM_TUOFL;
   2009 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
   2010 			reg &= ~RXCSUM_IPOFL;
   2011 	}
   2012 	CSR_WRITE(sc, DGE_RXCSUM, reg);
   2013 
   2014 	/*
   2015 	 * Set up the interrupt registers.
   2016 	 */
   2017 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
   2018 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
   2019 	    ICR_RXO | ICR_RXT0;
   2020 
   2021 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
   2022 
   2023 	/*
   2024 	 * Set up the transmit control register.
   2025 	 */
   2026 	sc->sc_tctl = TCTL_TCE | TCTL_TPDE | TCTL_TXEN;
   2027 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
   2028 
   2029 	/*
   2030 	 * Set up the receive control register; we actually program
   2031 	 * the register when we set the receive filter.  Use multicast
   2032 	 * address offset type 0.
   2033 	 */
   2034 	sc->sc_mchash_type = 0;
   2035 
   2036 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
   2037 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
   2038 
   2039 #ifdef DGE_OFFBYONE_RXBUG
   2040 	sc->sc_rctl |= RCTL_BSIZE_16k;
   2041 #else
   2042 	switch (MCLBYTES) {
   2043 	case 2048:
   2044 		sc->sc_rctl |= RCTL_BSIZE_2k;
   2045 		break;
   2046 	case 4096:
   2047 		sc->sc_rctl |= RCTL_BSIZE_4k;
   2048 		break;
   2049 	case 8192:
   2050 		sc->sc_rctl |= RCTL_BSIZE_8k;
   2051 		break;
   2052 	case 16384:
   2053 		sc->sc_rctl |= RCTL_BSIZE_16k;
   2054 		break;
   2055 	default:
   2056 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
   2057 	}
   2058 #endif
   2059 
   2060 	/* Set the receive filter. */
   2061 	/* Also sets RCTL */
   2062 	dge_set_filter(sc);
   2063 
   2064 	/* ...all done! */
   2065 	ifp->if_flags |= IFF_RUNNING;
   2066 	ifp->if_flags &= ~IFF_OACTIVE;
   2067 
   2068  out:
   2069 	if (error)
   2070 		printf("%s: interface not running\n", device_xname(sc->sc_dev));
   2071 	return error;
   2072 }
   2073 
   2074 /*
   2075  * dge_rxdrain:
   2076  *
   2077  *	Drain the receive queue.
   2078  */
   2079 static void
   2080 dge_rxdrain(struct dge_softc *sc)
   2081 {
   2082 	struct dge_rxsoft *rxs;
   2083 	int i;
   2084 
   2085 	for (i = 0; i < DGE_NRXDESC; i++) {
   2086 		rxs = &sc->sc_rxsoft[i];
   2087 		if (rxs->rxs_mbuf != NULL) {
   2088 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2089 			m_freem(rxs->rxs_mbuf);
   2090 			rxs->rxs_mbuf = NULL;
   2091 		}
   2092 	}
   2093 }
   2094 
   2095 /*
   2096  * dge_stop:		[ifnet interface function]
   2097  *
   2098  *	Stop transmission on the interface.
   2099  */
   2100 static void
   2101 dge_stop(struct ifnet *ifp, int disable)
   2102 {
   2103 	struct dge_softc *sc = ifp->if_softc;
   2104 	struct dge_txsoft *txs;
   2105 	int i;
   2106 
   2107 	/* Stop the transmit and receive processes. */
   2108 	CSR_WRITE(sc, DGE_TCTL, 0);
   2109 	CSR_WRITE(sc, DGE_RCTL, 0);
   2110 
   2111 	/* Release any queued transmit buffers. */
   2112 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
   2113 		txs = &sc->sc_txsoft[i];
   2114 		if (txs->txs_mbuf != NULL) {
   2115 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   2116 			m_freem(txs->txs_mbuf);
   2117 			txs->txs_mbuf = NULL;
   2118 		}
   2119 	}
   2120 
   2121 	/* Mark the interface as down and cancel the watchdog timer. */
   2122 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2123 	ifp->if_timer = 0;
   2124 
   2125 	if (disable)
   2126 		dge_rxdrain(sc);
   2127 }
   2128 
   2129 /*
   2130  * dge_add_rxbuf:
   2131  *
   2132  *	Add a receive buffer to the indiciated descriptor.
   2133  */
   2134 static int
   2135 dge_add_rxbuf(struct dge_softc *sc, int idx)
   2136 {
   2137 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
   2138 	struct mbuf *m;
   2139 	int error;
   2140 #ifdef DGE_OFFBYONE_RXBUG
   2141 	void *buf;
   2142 #endif
   2143 
   2144 	MGETHDR(m, M_DONTWAIT, MT_DATA);
   2145 	if (m == NULL)
   2146 		return ENOBUFS;
   2147 
   2148 #ifdef DGE_OFFBYONE_RXBUG
   2149 	if ((buf = dge_getbuf(sc)) == NULL)
   2150 		return ENOBUFS;
   2151 
   2152 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
   2153 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
   2154 	m->m_flags |= M_EXT_RW;
   2155 
   2156 	if (rxs->rxs_mbuf != NULL)
   2157 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2158 	rxs->rxs_mbuf = m;
   2159 
   2160 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
   2161 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ | BUS_DMA_NOWAIT);
   2162 #else
   2163 	MCLGET(m, M_DONTWAIT);
   2164 	if ((m->m_flags & M_EXT) == 0) {
   2165 		m_freem(m);
   2166 		return ENOBUFS;
   2167 	}
   2168 
   2169 	if (rxs->rxs_mbuf != NULL)
   2170 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2171 
   2172 	rxs->rxs_mbuf = m;
   2173 
   2174 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
   2175 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
   2176 	    BUS_DMA_READ | BUS_DMA_NOWAIT);
   2177 #endif
   2178 	if (error) {
   2179 		printf("%s: unable to load rx DMA map %d, error = %d\n",
   2180 		    device_xname(sc->sc_dev), idx, error);
   2181 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
   2182 	}
   2183 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   2184 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   2185 
   2186 	return 0;
   2187 }
   2188 
   2189 /*
   2190  * dge_set_ral:
   2191  *
   2192  *	Set an entry in the receive address list.
   2193  */
   2194 static void
   2195 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
   2196 {
   2197 	uint32_t ral_lo, ral_hi;
   2198 
   2199 	if (enaddr != NULL) {
   2200 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
   2201 		    (enaddr[3] << 24);
   2202 		ral_hi = enaddr[4] | (enaddr[5] << 8);
   2203 		ral_hi |= RAH_AV;
   2204 	} else {
   2205 		ral_lo = 0;
   2206 		ral_hi = 0;
   2207 	}
   2208 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
   2209 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
   2210 }
   2211 
   2212 /*
   2213  * dge_mchash:
   2214  *
   2215  *	Compute the hash of the multicast address for the 4096-bit
   2216  *	multicast filter.
   2217  */
   2218 static uint32_t
   2219 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
   2220 {
   2221 	static const int lo_shift[4] = { 4, 3, 2, 0 };
   2222 	static const int hi_shift[4] = { 4, 5, 6, 8 };
   2223 	uint32_t hash;
   2224 
   2225 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
   2226 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
   2227 
   2228 	return (hash & 0xfff);
   2229 }
   2230 
   2231 /*
   2232  * dge_set_filter:
   2233  *
   2234  *	Set up the receive filter.
   2235  */
   2236 static void
   2237 dge_set_filter(struct dge_softc *sc)
   2238 {
   2239 	struct ethercom *ec = &sc->sc_ethercom;
   2240 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   2241 	struct ether_multi *enm;
   2242 	struct ether_multistep step;
   2243 	uint32_t hash, reg, bit;
   2244 	int i;
   2245 
   2246 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
   2247 
   2248 	if (ifp->if_flags & IFF_BROADCAST)
   2249 		sc->sc_rctl |= RCTL_BAM;
   2250 	if (ifp->if_flags & IFF_PROMISC) {
   2251 		sc->sc_rctl |= RCTL_UPE;
   2252 		goto allmulti;
   2253 	}
   2254 
   2255 	/*
   2256 	 * Set the station address in the first RAL slot, and
   2257 	 * clear the remaining slots.
   2258 	 */
   2259 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
   2260 	for (i = 1; i < RA_TABSIZE; i++)
   2261 		dge_set_ral(sc, NULL, i);
   2262 
   2263 	/* Clear out the multicast table. */
   2264 	for (i = 0; i < MC_TABSIZE; i++)
   2265 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
   2266 
   2267 	ETHER_LOCK(ec);
   2268 	ETHER_FIRST_MULTI(step, ec, enm);
   2269 	while (enm != NULL) {
   2270 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
   2271 			/*
   2272 			 * We must listen to a range of multicast addresses.
   2273 			 * For now, just accept all multicasts, rather than
   2274 			 * trying to set only those filter bits needed to match
   2275 			 * the range.  (At this time, the only use of address
   2276 			 * ranges is for IP multicast routing, for which the
   2277 			 * range is big enough to require all bits set.)
   2278 			 */
   2279 			ETHER_UNLOCK(ec);
   2280 			goto allmulti;
   2281 		}
   2282 
   2283 		hash = dge_mchash(sc, enm->enm_addrlo);
   2284 
   2285 		reg = (hash >> 5) & 0x7f;
   2286 		bit = hash & 0x1f;
   2287 
   2288 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
   2289 		hash |= 1U << bit;
   2290 
   2291 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
   2292 
   2293 		ETHER_NEXT_MULTI(step, enm);
   2294 	}
   2295 	ETHER_UNLOCK(ec);
   2296 
   2297 	ifp->if_flags &= ~IFF_ALLMULTI;
   2298 	goto setit;
   2299 
   2300  allmulti:
   2301 	ifp->if_flags |= IFF_ALLMULTI;
   2302 	sc->sc_rctl |= RCTL_MPE;
   2303 
   2304  setit:
   2305 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
   2306 }
   2307 
   2308 /*
   2309  * Read in the EEPROM info and verify checksum.
   2310  */
   2311 int
   2312 dge_read_eeprom(struct dge_softc *sc)
   2313 {
   2314 	uint16_t cksum;
   2315 	int i;
   2316 
   2317 	cksum = 0;
   2318 	for (i = 0; i < EEPROM_SIZE; i++) {
   2319 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
   2320 		cksum += sc->sc_eeprom[i];
   2321 	}
   2322 	return cksum != EEPROM_CKSUM;
   2323 }
   2324 
   2325 
   2326 /*
   2327  * Read a 16-bit word from address addr in the serial EEPROM.
   2328  */
   2329 uint16_t
   2330 dge_eeprom_word(struct dge_softc *sc, int addr)
   2331 {
   2332 	uint32_t reg;
   2333 	uint16_t rval = 0;
   2334 	int i;
   2335 
   2336 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK | EECD_DI | EECD_CS);
   2337 
   2338 	/* Lower clock pulse (and data in to chip) */
   2339 	CSR_WRITE(sc, DGE_EECD, reg);
   2340 	/* Select chip */
   2341 	CSR_WRITE(sc, DGE_EECD, reg | EECD_CS);
   2342 
   2343 	/* Send read command */
   2344 	dge_eeprom_clockout(sc, 1);
   2345 	dge_eeprom_clockout(sc, 1);
   2346 	dge_eeprom_clockout(sc, 0);
   2347 
   2348 	/* Send address */
   2349 	for (i = 5; i >= 0; i--)
   2350 		dge_eeprom_clockout(sc, (addr >> i) & 1);
   2351 
   2352 	/* Read data */
   2353 	for (i = 0; i < 16; i++) {
   2354 		rval <<= 1;
   2355 		rval |= dge_eeprom_clockin(sc);
   2356 	}
   2357 
   2358 	/* Deselect chip */
   2359 	CSR_WRITE(sc, DGE_EECD, reg);
   2360 
   2361 	return rval;
   2362 }
   2363 
   2364 /*
   2365  * Clock out a single bit to the EEPROM.
   2366  */
   2367 void
   2368 dge_eeprom_clockout(struct dge_softc *sc, int bit)
   2369 {
   2370 	int reg;
   2371 
   2372 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI | EECD_SK);
   2373 	if (bit)
   2374 		reg |= EECD_DI;
   2375 
   2376 	CSR_WRITE(sc, DGE_EECD, reg);
   2377 	delay(2);
   2378 	CSR_WRITE(sc, DGE_EECD, reg | EECD_SK);
   2379 	delay(2);
   2380 	CSR_WRITE(sc, DGE_EECD, reg);
   2381 	delay(2);
   2382 }
   2383 
   2384 /*
   2385  * Clock in a single bit from EEPROM.
   2386  */
   2387 int
   2388 dge_eeprom_clockin(struct dge_softc *sc)
   2389 {
   2390 	int reg, rv;
   2391 
   2392 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI | EECD_DO | EECD_SK);
   2393 
   2394 	CSR_WRITE(sc, DGE_EECD, reg | EECD_SK); /* Raise clock */
   2395 	delay(2);
   2396 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
   2397 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
   2398 	delay(2);
   2399 
   2400 	return rv;
   2401 }
   2402 
   2403 static void
   2404 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
   2405 {
   2406 	struct dge_softc *sc = ifp->if_softc;
   2407 
   2408 	ifmr->ifm_status = IFM_AVALID;
   2409 	if (sc->sc_dgep->dgep_flags & DGEP_F_10G_SR ) {
   2410 		ifmr->ifm_active = IFM_ETHER | IFM_10G_SR;
   2411 	} else {
   2412 		ifmr->ifm_active = IFM_ETHER | IFM_10G_LR;
   2413 	}
   2414 
   2415 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
   2416 		ifmr->ifm_status |= IFM_ACTIVE;
   2417 }
   2418 
   2419 static inline int
   2420 phwait(struct dge_softc *sc, int p, int r, int d, int type)
   2421 {
   2422 	int i, mdic;
   2423 
   2424 	CSR_WRITE(sc, DGE_MDIO,
   2425 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
   2426 	for (i = 0; i < 10; i++) {
   2427 		delay(10);
   2428 		if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
   2429 			break;
   2430 	}
   2431 	return mdic;
   2432 }
   2433 
   2434 static void
   2435 dge_xgmii_writereg(struct dge_softc *sc, int phy, int reg, int val)
   2436 {
   2437 	int mdic;
   2438 
   2439 	CSR_WRITE(sc, DGE_MDIRW, val);
   2440 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
   2441 		printf("%s: address cycle timeout; phy %d reg %d\n",
   2442 		    device_xname(sc->sc_dev), phy, reg);
   2443 		return;
   2444 	}
   2445 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
   2446 		printf("%s: write cycle timeout; phy %d reg %d\n",
   2447 		    device_xname(sc->sc_dev), phy, reg);
   2448 		return;
   2449 	}
   2450 }
   2451 
   2452 static void
   2453 dge_xgmii_reset(struct dge_softc *sc)
   2454 {
   2455 	dge_xgmii_writereg(sc, 0, 0, BMCR_RESET);
   2456 }
   2457 
   2458 static int
   2459 dge_xgmii_mediachange(struct ifnet *ifp)
   2460 {
   2461 	return 0;
   2462 }
   2463