Home | History | Annotate | Line # | Download | only in pci
if_dge.c revision 1.34
      1 /*	$NetBSD: if_dge.c,v 1.34 2012/02/02 19:43:05 tls 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 1.34 2012/02/02 19:43:05 tls 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 
     99 #include <sys/rnd.h>
    100 
    101 #include <net/if.h>
    102 #include <net/if_dl.h>
    103 #include <net/if_media.h>
    104 #include <net/if_ether.h>
    105 
    106 #include <net/bpf.h>
    107 
    108 #include <netinet/in.h>			/* XXX for struct ip */
    109 #include <netinet/in_systm.h>		/* XXX for struct ip */
    110 #include <netinet/ip.h>			/* XXX for struct ip */
    111 #include <netinet/tcp.h>		/* XXX for struct tcphdr */
    112 
    113 #include <sys/bus.h>
    114 #include <sys/intr.h>
    115 #include <machine/endian.h>
    116 
    117 #include <dev/mii/mii.h>
    118 #include <dev/mii/miivar.h>
    119 #include <dev/mii/mii_bitbang.h>
    120 
    121 #include <dev/pci/pcireg.h>
    122 #include <dev/pci/pcivar.h>
    123 #include <dev/pci/pcidevs.h>
    124 
    125 #include <dev/pci/if_dgereg.h>
    126 
    127 /*
    128  * The receive engine may sometimes become off-by-one when writing back
    129  * chained descriptors.	 Avoid this by allocating a large chunk of
    130  * memory and use if instead (to avoid chained descriptors).
    131  * This only happens with chained descriptors under heavy load.
    132  */
    133 #define DGE_OFFBYONE_RXBUG
    134 
    135 #define DGE_EVENT_COUNTERS
    136 #define DGE_DEBUG
    137 
    138 #ifdef DGE_DEBUG
    139 #define DGE_DEBUG_LINK		0x01
    140 #define DGE_DEBUG_TX		0x02
    141 #define DGE_DEBUG_RX		0x04
    142 #define DGE_DEBUG_CKSUM		0x08
    143 int	dge_debug = 0;
    144 
    145 #define DPRINTF(x, y)	if (dge_debug & (x)) printf y
    146 #else
    147 #define DPRINTF(x, y)	/* nothing */
    148 #endif /* DGE_DEBUG */
    149 
    150 /*
    151  * Transmit descriptor list size. We allow up to 100 DMA segments per
    152  * packet (Intel reports of jumbo frame packets with as
    153  * many as 80 DMA segments when using 16k buffers).
    154  */
    155 #define DGE_NTXSEGS		100
    156 #define DGE_IFQUEUELEN		20000
    157 #define DGE_TXQUEUELEN		2048
    158 #define DGE_TXQUEUELEN_MASK	(DGE_TXQUEUELEN - 1)
    159 #define DGE_TXQUEUE_GC		(DGE_TXQUEUELEN / 8)
    160 #define DGE_NTXDESC		1024
    161 #define DGE_NTXDESC_MASK		(DGE_NTXDESC - 1)
    162 #define DGE_NEXTTX(x)		(((x) + 1) & DGE_NTXDESC_MASK)
    163 #define DGE_NEXTTXS(x)		(((x) + 1) & DGE_TXQUEUELEN_MASK)
    164 
    165 /*
    166  * Receive descriptor list size.
    167  * Packet is of size MCLBYTES, and for jumbo packets buffers may
    168  * be chained.	Due to the nature of the card (high-speed), keep this
    169  * ring large. With 2k buffers the ring can store 400 jumbo packets,
    170  * which at full speed will be received in just under 3ms.
    171  */
    172 #define DGE_NRXDESC		2048
    173 #define DGE_NRXDESC_MASK	(DGE_NRXDESC - 1)
    174 #define DGE_NEXTRX(x)		(((x) + 1) & DGE_NRXDESC_MASK)
    175 /*
    176  * # of descriptors between head and written descriptors.
    177  * This is to work-around two erratas.
    178  */
    179 #define DGE_RXSPACE		10
    180 #define DGE_PREVRX(x)		(((x) - DGE_RXSPACE) & DGE_NRXDESC_MASK)
    181 /*
    182  * Receive descriptor fetch threshholds. These are values recommended
    183  * by Intel, do not touch them unless you know what you are doing.
    184  */
    185 #define RXDCTL_PTHRESH_VAL	128
    186 #define RXDCTL_HTHRESH_VAL	16
    187 #define RXDCTL_WTHRESH_VAL	16
    188 
    189 
    190 /*
    191  * Tweakable parameters; default values.
    192  */
    193 #define FCRTH	0x30000 /* Send XOFF water mark */
    194 #define FCRTL	0x28000 /* Send XON water mark */
    195 #define RDTR	0x20	/* Interrupt delay after receive, .8192us units */
    196 #define TIDV	0x20	/* Interrupt delay after send, .8192us units */
    197 
    198 /*
    199  * Control structures are DMA'd to the i82597 chip.  We allocate them in
    200  * a single clump that maps to a single DMA segment to make serveral things
    201  * easier.
    202  */
    203 struct dge_control_data {
    204 	/*
    205 	 * The transmit descriptors.
    206 	 */
    207 	struct dge_tdes wcd_txdescs[DGE_NTXDESC];
    208 
    209 	/*
    210 	 * The receive descriptors.
    211 	 */
    212 	struct dge_rdes wcd_rxdescs[DGE_NRXDESC];
    213 };
    214 
    215 #define DGE_CDOFF(x)	offsetof(struct dge_control_data, x)
    216 #define DGE_CDTXOFF(x)	DGE_CDOFF(wcd_txdescs[(x)])
    217 #define DGE_CDRXOFF(x)	DGE_CDOFF(wcd_rxdescs[(x)])
    218 
    219 /*
    220  * The DGE interface have a higher max MTU size than normal jumbo frames.
    221  */
    222 #define DGE_MAX_MTU	16288	/* Max MTU size for this interface */
    223 
    224 /*
    225  * Software state for transmit jobs.
    226  */
    227 struct dge_txsoft {
    228 	struct mbuf *txs_mbuf;		/* head of our mbuf chain */
    229 	bus_dmamap_t txs_dmamap;	/* our DMA map */
    230 	int txs_firstdesc;		/* first descriptor in packet */
    231 	int txs_lastdesc;		/* last descriptor in packet */
    232 	int txs_ndesc;			/* # of descriptors used */
    233 };
    234 
    235 /*
    236  * Software state for receive buffers.	Each descriptor gets a
    237  * 2k (MCLBYTES) buffer and a DMA map.	For packets which fill
    238  * more than one buffer, we chain them together.
    239  */
    240 struct dge_rxsoft {
    241 	struct mbuf *rxs_mbuf;		/* head of our mbuf chain */
    242 	bus_dmamap_t rxs_dmamap;	/* our DMA map */
    243 };
    244 
    245 /*
    246  * Software state per device.
    247  */
    248 struct dge_softc {
    249 	struct device sc_dev;		/* generic device information */
    250 	bus_space_tag_t sc_st;		/* bus space tag */
    251 	bus_space_handle_t sc_sh;	/* bus space handle */
    252 	bus_dma_tag_t sc_dmat;		/* bus DMA tag */
    253 	struct ethercom sc_ethercom;	/* ethernet common data */
    254 
    255 	int sc_flags;			/* flags; see below */
    256 	int sc_bus_speed;		/* PCI/PCIX bus speed */
    257 	int sc_pcix_offset;		/* PCIX capability register offset */
    258 
    259 	pci_chipset_tag_t sc_pc;
    260 	pcitag_t sc_pt;
    261 	int sc_mmrbc;			/* Max PCIX memory read byte count */
    262 
    263 	void *sc_ih;			/* interrupt cookie */
    264 
    265 	struct ifmedia sc_media;
    266 
    267 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
    268 #define sc_cddma	sc_cddmamap->dm_segs[0].ds_addr
    269 
    270 	int		sc_align_tweak;
    271 
    272 	/*
    273 	 * Software state for the transmit and receive descriptors.
    274 	 */
    275 	struct dge_txsoft sc_txsoft[DGE_TXQUEUELEN];
    276 	struct dge_rxsoft sc_rxsoft[DGE_NRXDESC];
    277 
    278 	/*
    279 	 * Control data structures.
    280 	 */
    281 	struct dge_control_data *sc_control_data;
    282 #define sc_txdescs	sc_control_data->wcd_txdescs
    283 #define sc_rxdescs	sc_control_data->wcd_rxdescs
    284 
    285 #ifdef DGE_EVENT_COUNTERS
    286 	/* Event counters. */
    287 	struct evcnt sc_ev_txsstall;	/* Tx stalled due to no txs */
    288 	struct evcnt sc_ev_txdstall;	/* Tx stalled due to no txd */
    289 	struct evcnt sc_ev_txforceintr; /* Tx interrupts forced */
    290 	struct evcnt sc_ev_txdw;	/* Tx descriptor interrupts */
    291 	struct evcnt sc_ev_txqe;	/* Tx queue empty interrupts */
    292 	struct evcnt sc_ev_rxintr;	/* Rx interrupts */
    293 	struct evcnt sc_ev_linkintr;	/* Link interrupts */
    294 
    295 	struct evcnt sc_ev_rxipsum;	/* IP checksums checked in-bound */
    296 	struct evcnt sc_ev_rxtusum;	/* TCP/UDP cksums checked in-bound */
    297 	struct evcnt sc_ev_txipsum;	/* IP checksums comp. out-bound */
    298 	struct evcnt sc_ev_txtusum;	/* TCP/UDP cksums comp. out-bound */
    299 
    300 	struct evcnt sc_ev_txctx_init;	/* Tx cksum context cache initialized */
    301 	struct evcnt sc_ev_txctx_hit;	/* Tx cksum context cache hit */
    302 	struct evcnt sc_ev_txctx_miss;	/* Tx cksum context cache miss */
    303 
    304 	struct evcnt sc_ev_txseg[DGE_NTXSEGS]; /* Tx packets w/ N segments */
    305 	struct evcnt sc_ev_txdrop;	/* Tx packets dropped (too many segs) */
    306 #endif /* DGE_EVENT_COUNTERS */
    307 
    308 	int	sc_txfree;		/* number of free Tx descriptors */
    309 	int	sc_txnext;		/* next ready Tx descriptor */
    310 
    311 	int	sc_txsfree;		/* number of free Tx jobs */
    312 	int	sc_txsnext;		/* next free Tx job */
    313 	int	sc_txsdirty;		/* dirty Tx jobs */
    314 
    315 	uint32_t sc_txctx_ipcs;		/* cached Tx IP cksum ctx */
    316 	uint32_t sc_txctx_tucs;		/* cached Tx TCP/UDP cksum ctx */
    317 
    318 	int	sc_rxptr;		/* next ready Rx descriptor/queue ent */
    319 	int	sc_rxdiscard;
    320 	int	sc_rxlen;
    321 	struct mbuf *sc_rxhead;
    322 	struct mbuf *sc_rxtail;
    323 	struct mbuf **sc_rxtailp;
    324 
    325 	uint32_t sc_ctrl0;		/* prototype CTRL0 register */
    326 	uint32_t sc_icr;		/* prototype interrupt bits */
    327 	uint32_t sc_tctl;		/* prototype TCTL register */
    328 	uint32_t sc_rctl;		/* prototype RCTL register */
    329 
    330 	int sc_mchash_type;		/* multicast filter offset */
    331 
    332 	uint16_t sc_eeprom[EEPROM_SIZE];
    333 
    334 	krndsource_t rnd_source; /* random source */
    335 #ifdef DGE_OFFBYONE_RXBUG
    336 	void *sc_bugbuf;
    337 	SLIST_HEAD(, rxbugentry) sc_buglist;
    338 	bus_dmamap_t sc_bugmap;
    339 	struct rxbugentry *sc_entry;
    340 #endif
    341 };
    342 
    343 #define DGE_RXCHAIN_RESET(sc)						\
    344 do {									\
    345 	(sc)->sc_rxtailp = &(sc)->sc_rxhead;				\
    346 	*(sc)->sc_rxtailp = NULL;					\
    347 	(sc)->sc_rxlen = 0;						\
    348 } while (/*CONSTCOND*/0)
    349 
    350 #define DGE_RXCHAIN_LINK(sc, m)						\
    351 do {									\
    352 	*(sc)->sc_rxtailp = (sc)->sc_rxtail = (m);			\
    353 	(sc)->sc_rxtailp = &(m)->m_next;				\
    354 } while (/*CONSTCOND*/0)
    355 
    356 /* sc_flags */
    357 #define DGE_F_BUS64		0x20	/* bus is 64-bit */
    358 #define DGE_F_PCIX		0x40	/* bus is PCI-X */
    359 
    360 #ifdef DGE_EVENT_COUNTERS
    361 #define DGE_EVCNT_INCR(ev)	(ev)->ev_count++
    362 #else
    363 #define DGE_EVCNT_INCR(ev)	/* nothing */
    364 #endif
    365 
    366 #define CSR_READ(sc, reg)						\
    367 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (reg))
    368 #define CSR_WRITE(sc, reg, val)						\
    369 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, (reg), (val))
    370 
    371 #define DGE_CDTXADDR(sc, x)	((sc)->sc_cddma + DGE_CDTXOFF((x)))
    372 #define DGE_CDRXADDR(sc, x)	((sc)->sc_cddma + DGE_CDRXOFF((x)))
    373 
    374 #define DGE_CDTXSYNC(sc, x, n, ops)					\
    375 do {									\
    376 	int __x, __n;							\
    377 									\
    378 	__x = (x);							\
    379 	__n = (n);							\
    380 									\
    381 	/* If it will wrap around, sync to the end of the ring. */	\
    382 	if ((__x + __n) > DGE_NTXDESC) {				\
    383 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
    384 		    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) *		\
    385 		    (DGE_NTXDESC - __x), (ops));			\
    386 		__n -= (DGE_NTXDESC - __x);				\
    387 		__x = 0;						\
    388 	}								\
    389 									\
    390 	/* Now sync whatever is left. */				\
    391 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
    392 	    DGE_CDTXOFF(__x), sizeof(struct dge_tdes) * __n, (ops));	\
    393 } while (/*CONSTCOND*/0)
    394 
    395 #define DGE_CDRXSYNC(sc, x, ops)						\
    396 do {									\
    397 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
    398 	   DGE_CDRXOFF((x)), sizeof(struct dge_rdes), (ops));		\
    399 } while (/*CONSTCOND*/0)
    400 
    401 #ifdef DGE_OFFBYONE_RXBUG
    402 #define DGE_INIT_RXDESC(sc, x)						\
    403 do {									\
    404 	struct dge_rxsoft *__rxs = &(sc)->sc_rxsoft[(x)];		\
    405 	struct dge_rdes *__rxd = &(sc)->sc_rxdescs[(x)];		\
    406 	struct mbuf *__m = __rxs->rxs_mbuf;				\
    407 									\
    408 	__rxd->dr_baddrl = htole32(sc->sc_bugmap->dm_segs[0].ds_addr +	\
    409 	    (mtod((__m), char *) - (char *)sc->sc_bugbuf));		\
    410 	__rxd->dr_baddrh = 0;						\
    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 	__rxd->dr_baddrl =						\
    444 	    htole32(__rxs->rxs_dmamap->dm_segs[0].ds_addr +		\
    445 		(sc)->sc_align_tweak);					\
    446 	__rxd->dr_baddrh = 0;						\
    447 	__rxd->dr_len = 0;						\
    448 	__rxd->dr_cksum = 0;						\
    449 	__rxd->dr_status = 0;						\
    450 	__rxd->dr_errors = 0;						\
    451 	__rxd->dr_special = 0;						\
    452 	DGE_CDRXSYNC((sc), (x), BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE); \
    453 									\
    454 	CSR_WRITE((sc), DGE_RDT, (x));					\
    455 } while (/*CONSTCOND*/0)
    456 #endif
    457 
    458 #ifdef DGE_OFFBYONE_RXBUG
    459 /*
    460  * Allocation constants.  Much memory may be used for this.
    461  */
    462 #ifndef DGE_BUFFER_SIZE
    463 #define DGE_BUFFER_SIZE DGE_MAX_MTU
    464 #endif
    465 #define DGE_NBUFFERS	(4*DGE_NRXDESC)
    466 #define DGE_RXMEM	(DGE_NBUFFERS*DGE_BUFFER_SIZE)
    467 
    468 struct rxbugentry {
    469 	SLIST_ENTRY(rxbugentry) rb_entry;
    470 	int rb_slot;
    471 };
    472 
    473 static int
    474 dge_alloc_rcvmem(struct dge_softc *sc)
    475 {
    476 	char *ptr, *kva;
    477 	bus_dma_segment_t seg;
    478 	int i, rseg, state, error;
    479 	struct rxbugentry *entry;
    480 
    481 	state = error = 0;
    482 
    483 	if (bus_dmamem_alloc(sc->sc_dmat, DGE_RXMEM, PAGE_SIZE, 0,
    484 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
    485 		aprint_error_dev(&sc->sc_dev, "can't alloc rx buffers\n");
    486 		return ENOBUFS;
    487 	}
    488 
    489 	state = 1;
    490 	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, DGE_RXMEM, (void **)&kva,
    491 	    BUS_DMA_NOWAIT)) {
    492 		aprint_error_dev(&sc->sc_dev, "can't map DMA buffers (%d bytes)\n",
    493 		    (int)DGE_RXMEM);
    494 		error = ENOBUFS;
    495 		goto out;
    496 	}
    497 
    498 	state = 2;
    499 	if (bus_dmamap_create(sc->sc_dmat, DGE_RXMEM, 1, DGE_RXMEM, 0,
    500 	    BUS_DMA_NOWAIT, &sc->sc_bugmap)) {
    501 		aprint_error_dev(&sc->sc_dev, "can't create DMA map\n");
    502 		error = ENOBUFS;
    503 		goto out;
    504 	}
    505 
    506 	state = 3;
    507 	if (bus_dmamap_load(sc->sc_dmat, sc->sc_bugmap,
    508 	    kva, DGE_RXMEM, NULL, BUS_DMA_NOWAIT)) {
    509 		aprint_error_dev(&sc->sc_dev, "can't load DMA map\n");
    510 		error = ENOBUFS;
    511 		goto out;
    512 	}
    513 
    514 	state = 4;
    515 	sc->sc_bugbuf = (void *)kva;
    516 	SLIST_INIT(&sc->sc_buglist);
    517 
    518 	/*
    519 	 * Now divide it up into DGE_BUFFER_SIZE pieces and save the addresses
    520 	 * in an array.
    521 	 */
    522 	ptr = sc->sc_bugbuf;
    523 	if ((entry = malloc(sizeof(*entry) * DGE_NBUFFERS,
    524 	    M_DEVBUF, M_NOWAIT)) == NULL) {
    525 		error = ENOBUFS;
    526 		goto out;
    527 	}
    528 	sc->sc_entry = entry;
    529 	for (i = 0; i < DGE_NBUFFERS; i++) {
    530 		entry[i].rb_slot = i;
    531 		SLIST_INSERT_HEAD(&sc->sc_buglist, &entry[i], rb_entry);
    532 	}
    533 out:
    534 	if (error != 0) {
    535 		switch (state) {
    536 		case 4:
    537 			bus_dmamap_unload(sc->sc_dmat, sc->sc_bugmap);
    538 		case 3:
    539 			bus_dmamap_destroy(sc->sc_dmat, sc->sc_bugmap);
    540 		case 2:
    541 			bus_dmamem_unmap(sc->sc_dmat, kva, DGE_RXMEM);
    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(device_t, int, int, int);
    635 
    636 
    637 CFATTACH_DECL(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 static int
    648 dge_match(device_t parent, cfdata_t cf, void *aux)
    649 {
    650 	struct pci_attach_args *pa = aux;
    651 
    652 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
    653 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82597EX)
    654 		return (1);
    655 
    656 	return (0);
    657 }
    658 
    659 static void
    660 dge_attach(device_t parent, device_t self, void *aux)
    661 {
    662 	struct dge_softc *sc = device_private(self);
    663 	struct pci_attach_args *pa = aux;
    664 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
    665 	pci_chipset_tag_t pc = pa->pa_pc;
    666 	pci_intr_handle_t ih;
    667 	const char *intrstr = NULL;
    668 	bus_dma_segment_t seg;
    669 	int i, rseg, error;
    670 	uint8_t enaddr[ETHER_ADDR_LEN];
    671 	pcireg_t preg, memtype;
    672 	uint32_t reg;
    673 
    674 	sc->sc_dmat = pa->pa_dmat;
    675 	sc->sc_pc = pa->pa_pc;
    676 	sc->sc_pt = pa->pa_tag;
    677 
    678 	pci_aprint_devinfo_fancy(pa, "Ethernet controller",
    679 		"Intel i82597EX 10GbE-LR Ethernet", 1);
    680 
    681 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, DGE_PCI_BAR);
    682         if (pci_mapreg_map(pa, DGE_PCI_BAR, memtype, 0,
    683             &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
    684                 aprint_error_dev(&sc->sc_dev, "unable to map device registers\n");
    685                 return;
    686         }
    687 
    688 	/* Enable bus mastering */
    689 	preg = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
    690 	preg |= PCI_COMMAND_MASTER_ENABLE;
    691 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, preg);
    692 
    693 	/*
    694 	 * Map and establish our interrupt.
    695 	 */
    696 	if (pci_intr_map(pa, &ih)) {
    697 		aprint_error_dev(&sc->sc_dev, "unable to map interrupt\n");
    698 		return;
    699 	}
    700 	intrstr = pci_intr_string(pc, ih);
    701 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, dge_intr, sc);
    702 	if (sc->sc_ih == NULL) {
    703 		aprint_error_dev(&sc->sc_dev, "unable to establish interrupt");
    704 		if (intrstr != NULL)
    705 			aprint_error(" at %s", intrstr);
    706 		aprint_error("\n");
    707 		return;
    708 	}
    709 	aprint_normal_dev(&sc->sc_dev, "interrupting at %s\n", intrstr);
    710 
    711 	/*
    712 	 * Determine a few things about the bus we're connected to.
    713 	 */
    714 	reg = CSR_READ(sc, DGE_STATUS);
    715 	if (reg & STATUS_BUS64)
    716 		sc->sc_flags |= DGE_F_BUS64;
    717 
    718 	sc->sc_flags |= DGE_F_PCIX;
    719 	if (pci_get_capability(pa->pa_pc, pa->pa_tag,
    720 			       PCI_CAP_PCIX,
    721 			       &sc->sc_pcix_offset, NULL) == 0)
    722 		aprint_error_dev(&sc->sc_dev, "unable to find PCIX "
    723 		    "capability\n");
    724 
    725 	if (sc->sc_flags & DGE_F_PCIX) {
    726 		switch (reg & STATUS_PCIX_MSK) {
    727 		case STATUS_PCIX_66:
    728 			sc->sc_bus_speed = 66;
    729 			break;
    730 		case STATUS_PCIX_100:
    731 			sc->sc_bus_speed = 100;
    732 			break;
    733 		case STATUS_PCIX_133:
    734 			sc->sc_bus_speed = 133;
    735 			break;
    736 		default:
    737 			aprint_error_dev(&sc->sc_dev,
    738 			    "unknown PCIXSPD %d; assuming 66MHz\n",
    739 			    reg & STATUS_PCIX_MSK);
    740 			sc->sc_bus_speed = 66;
    741 		}
    742 	} else
    743 		sc->sc_bus_speed = (reg & STATUS_BUS64) ? 66 : 33;
    744 	aprint_verbose_dev(&sc->sc_dev, "%d-bit %dMHz %s bus\n",
    745 	    (sc->sc_flags & DGE_F_BUS64) ? 64 : 32, sc->sc_bus_speed,
    746 	    (sc->sc_flags & DGE_F_PCIX) ? "PCIX" : "PCI");
    747 
    748 	/*
    749 	 * Allocate the control data structures, and create and load the
    750 	 * DMA map for it.
    751 	 */
    752 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    753 	    sizeof(struct dge_control_data), PAGE_SIZE, 0, &seg, 1, &rseg,
    754 	    0)) != 0) {
    755 		aprint_error_dev(&sc->sc_dev,
    756 		    "unable to allocate control data, error = %d\n",
    757 		    error);
    758 		goto fail_0;
    759 	}
    760 
    761 	if ((error = bus_dmamem_map(sc->sc_dmat, &seg, rseg,
    762 	    sizeof(struct dge_control_data), (void **)&sc->sc_control_data,
    763 	    0)) != 0) {
    764 		aprint_error_dev(&sc->sc_dev, "unable to map control data, error = %d\n",
    765 		    error);
    766 		goto fail_1;
    767 	}
    768 
    769 	if ((error = bus_dmamap_create(sc->sc_dmat,
    770 	    sizeof(struct dge_control_data), 1,
    771 	    sizeof(struct dge_control_data), 0, 0, &sc->sc_cddmamap)) != 0) {
    772 		aprint_error_dev(&sc->sc_dev, "unable to create control data DMA map, "
    773 		    "error = %d\n", error);
    774 		goto fail_2;
    775 	}
    776 
    777 	if ((error = bus_dmamap_load(sc->sc_dmat, sc->sc_cddmamap,
    778 	    sc->sc_control_data, sizeof(struct dge_control_data), NULL,
    779 	    0)) != 0) {
    780 		aprint_error_dev(&sc->sc_dev,
    781 		    "unable to load control data DMA map, error = %d\n",
    782 		    error);
    783 		goto fail_3;
    784 	}
    785 
    786 #ifdef DGE_OFFBYONE_RXBUG
    787 	if (dge_alloc_rcvmem(sc) != 0)
    788 		return; /* Already complained */
    789 #endif
    790 	/*
    791 	 * Create the transmit buffer DMA maps.
    792 	 */
    793 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
    794 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_MAX_MTU,
    795 		    DGE_NTXSEGS, MCLBYTES, 0, 0,
    796 		    &sc->sc_txsoft[i].txs_dmamap)) != 0) {
    797 			aprint_error_dev(&sc->sc_dev, "unable to create Tx DMA map %d, "
    798 			    "error = %d\n", i, error);
    799 			goto fail_4;
    800 		}
    801 	}
    802 
    803 	/*
    804 	 * Create the receive buffer DMA maps.
    805 	 */
    806 	for (i = 0; i < DGE_NRXDESC; i++) {
    807 #ifdef DGE_OFFBYONE_RXBUG
    808 		if ((error = bus_dmamap_create(sc->sc_dmat, DGE_BUFFER_SIZE, 1,
    809 		    DGE_BUFFER_SIZE, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    810 #else
    811 		if ((error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
    812 		    MCLBYTES, 0, 0, &sc->sc_rxsoft[i].rxs_dmamap)) != 0) {
    813 #endif
    814 			aprint_error_dev(&sc->sc_dev, "unable to create Rx DMA map %d, "
    815 			    "error = %d\n", i, error);
    816 			goto fail_5;
    817 		}
    818 		sc->sc_rxsoft[i].rxs_mbuf = NULL;
    819 	}
    820 
    821 	/*
    822 	 * Set bits in ctrl0 register.
    823 	 * Should get the software defined pins out of EEPROM?
    824 	 */
    825 	sc->sc_ctrl0 |= CTRL0_RPE | CTRL0_TPE; /* XON/XOFF */
    826 	sc->sc_ctrl0 |= CTRL0_SDP3_DIR | CTRL0_SDP2_DIR | CTRL0_SDP1_DIR |
    827 	    CTRL0_SDP0_DIR | CTRL0_SDP3 | CTRL0_SDP2 | CTRL0_SDP0;
    828 
    829 	/*
    830 	 * Reset the chip to a known state.
    831 	 */
    832 	dge_reset(sc);
    833 
    834 	/*
    835 	 * Reset the PHY.
    836 	 */
    837 	dge_xgmii_reset(sc);
    838 
    839 	/*
    840 	 * Read in EEPROM data.
    841 	 */
    842 	if (dge_read_eeprom(sc)) {
    843 		aprint_error_dev(&sc->sc_dev, "couldn't read EEPROM\n");
    844 		return;
    845 	}
    846 
    847 	/*
    848 	 * Get the ethernet address.
    849 	 */
    850 	enaddr[0] = sc->sc_eeprom[EE_ADDR01] & 0377;
    851 	enaddr[1] = sc->sc_eeprom[EE_ADDR01] >> 8;
    852 	enaddr[2] = sc->sc_eeprom[EE_ADDR23] & 0377;
    853 	enaddr[3] = sc->sc_eeprom[EE_ADDR23] >> 8;
    854 	enaddr[4] = sc->sc_eeprom[EE_ADDR45] & 0377;
    855 	enaddr[5] = sc->sc_eeprom[EE_ADDR45] >> 8;
    856 
    857 	aprint_normal_dev(&sc->sc_dev, "Ethernet address %s\n",
    858 	    ether_sprintf(enaddr));
    859 
    860 	/*
    861 	 * Setup media stuff.
    862 	 */
    863         ifmedia_init(&sc->sc_media, IFM_IMASK, dge_xgmii_mediachange,
    864             dge_xgmii_mediastatus);
    865         ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_10G_LR, 0, NULL);
    866         ifmedia_set(&sc->sc_media, IFM_ETHER|IFM_10G_LR);
    867 
    868 	ifp = &sc->sc_ethercom.ec_if;
    869 	strlcpy(ifp->if_xname, device_xname(&sc->sc_dev), IFNAMSIZ);
    870 	ifp->if_softc = sc;
    871 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    872 	ifp->if_ioctl = dge_ioctl;
    873 	ifp->if_start = dge_start;
    874 	ifp->if_watchdog = dge_watchdog;
    875 	ifp->if_init = dge_init;
    876 	ifp->if_stop = dge_stop;
    877 	IFQ_SET_MAXLEN(&ifp->if_snd, max(DGE_IFQUEUELEN, IFQ_MAXLEN));
    878 	IFQ_SET_READY(&ifp->if_snd);
    879 
    880 	sc->sc_ethercom.ec_capabilities |=
    881 	    ETHERCAP_JUMBO_MTU | ETHERCAP_VLAN_MTU;
    882 
    883 	/*
    884 	 * We can perform TCPv4 and UDPv4 checkums in-bound.
    885 	 */
    886 	ifp->if_capabilities |=
    887 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    888 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    889 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
    890 
    891 	/*
    892 	 * Attach the interface.
    893 	 */
    894 	if_attach(ifp);
    895 	ether_ifattach(ifp, enaddr);
    896 	rnd_attach_source(&sc->rnd_source, device_xname(&sc->sc_dev),
    897 	    RND_TYPE_NET, 0);
    898 
    899 #ifdef DGE_EVENT_COUNTERS
    900 	/* Fix segment event naming */
    901 	if (dge_txseg_evcnt_names == NULL) {
    902 		dge_txseg_evcnt_names =
    903 		    malloc(sizeof(*dge_txseg_evcnt_names), M_DEVBUF, M_WAITOK);
    904 		for (i = 0; i < DGE_NTXSEGS; i++)
    905 			snprintf((*dge_txseg_evcnt_names)[i],
    906 			    sizeof((*dge_txseg_evcnt_names)[i]), "txseg%d", i);
    907 	}
    908 
    909 	/* Attach event counters. */
    910 	evcnt_attach_dynamic(&sc->sc_ev_txsstall, EVCNT_TYPE_MISC,
    911 	    NULL, device_xname(&sc->sc_dev), "txsstall");
    912 	evcnt_attach_dynamic(&sc->sc_ev_txdstall, EVCNT_TYPE_MISC,
    913 	    NULL, device_xname(&sc->sc_dev), "txdstall");
    914 	evcnt_attach_dynamic(&sc->sc_ev_txforceintr, EVCNT_TYPE_MISC,
    915 	    NULL, device_xname(&sc->sc_dev), "txforceintr");
    916 	evcnt_attach_dynamic(&sc->sc_ev_txdw, EVCNT_TYPE_INTR,
    917 	    NULL, device_xname(&sc->sc_dev), "txdw");
    918 	evcnt_attach_dynamic(&sc->sc_ev_txqe, EVCNT_TYPE_INTR,
    919 	    NULL, device_xname(&sc->sc_dev), "txqe");
    920 	evcnt_attach_dynamic(&sc->sc_ev_rxintr, EVCNT_TYPE_INTR,
    921 	    NULL, device_xname(&sc->sc_dev), "rxintr");
    922 	evcnt_attach_dynamic(&sc->sc_ev_linkintr, EVCNT_TYPE_INTR,
    923 	    NULL, device_xname(&sc->sc_dev), "linkintr");
    924 
    925 	evcnt_attach_dynamic(&sc->sc_ev_rxipsum, EVCNT_TYPE_MISC,
    926 	    NULL, device_xname(&sc->sc_dev), "rxipsum");
    927 	evcnt_attach_dynamic(&sc->sc_ev_rxtusum, EVCNT_TYPE_MISC,
    928 	    NULL, device_xname(&sc->sc_dev), "rxtusum");
    929 	evcnt_attach_dynamic(&sc->sc_ev_txipsum, EVCNT_TYPE_MISC,
    930 	    NULL, device_xname(&sc->sc_dev), "txipsum");
    931 	evcnt_attach_dynamic(&sc->sc_ev_txtusum, EVCNT_TYPE_MISC,
    932 	    NULL, device_xname(&sc->sc_dev), "txtusum");
    933 
    934 	evcnt_attach_dynamic(&sc->sc_ev_txctx_init, EVCNT_TYPE_MISC,
    935 	    NULL, device_xname(&sc->sc_dev), "txctx init");
    936 	evcnt_attach_dynamic(&sc->sc_ev_txctx_hit, EVCNT_TYPE_MISC,
    937 	    NULL, device_xname(&sc->sc_dev), "txctx hit");
    938 	evcnt_attach_dynamic(&sc->sc_ev_txctx_miss, EVCNT_TYPE_MISC,
    939 	    NULL, device_xname(&sc->sc_dev), "txctx miss");
    940 
    941 	for (i = 0; i < DGE_NTXSEGS; i++)
    942 		evcnt_attach_dynamic(&sc->sc_ev_txseg[i], EVCNT_TYPE_MISC,
    943 		    NULL, device_xname(&sc->sc_dev), (*dge_txseg_evcnt_names)[i]);
    944 
    945 	evcnt_attach_dynamic(&sc->sc_ev_txdrop, EVCNT_TYPE_MISC,
    946 	    NULL, device_xname(&sc->sc_dev), "txdrop");
    947 
    948 #endif /* DGE_EVENT_COUNTERS */
    949 
    950 	/*
    951 	 * Make sure the interface is shutdown during reboot.
    952 	 */
    953 	if (pmf_device_register1(self, NULL, NULL, dge_shutdown))
    954 		pmf_class_network_register(self, ifp);
    955 	else
    956 		aprint_error_dev(self, "couldn't establish power handler\n");
    957 
    958 	return;
    959 
    960 	/*
    961 	 * Free any resources we've allocated during the failed attach
    962 	 * attempt.  Do this in reverse order and fall through.
    963 	 */
    964  fail_5:
    965 	for (i = 0; i < DGE_NRXDESC; i++) {
    966 		if (sc->sc_rxsoft[i].rxs_dmamap != NULL)
    967 			bus_dmamap_destroy(sc->sc_dmat,
    968 			    sc->sc_rxsoft[i].rxs_dmamap);
    969 	}
    970  fail_4:
    971 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
    972 		if (sc->sc_txsoft[i].txs_dmamap != NULL)
    973 			bus_dmamap_destroy(sc->sc_dmat,
    974 			    sc->sc_txsoft[i].txs_dmamap);
    975 	}
    976 	bus_dmamap_unload(sc->sc_dmat, sc->sc_cddmamap);
    977  fail_3:
    978 	bus_dmamap_destroy(sc->sc_dmat, sc->sc_cddmamap);
    979  fail_2:
    980 	bus_dmamem_unmap(sc->sc_dmat, (void *)sc->sc_control_data,
    981 	    sizeof(struct dge_control_data));
    982  fail_1:
    983 	bus_dmamem_free(sc->sc_dmat, &seg, rseg);
    984  fail_0:
    985 	return;
    986 }
    987 
    988 /*
    989  * dge_shutdown:
    990  *
    991  *	Make sure the interface is stopped at reboot time.
    992  */
    993 static bool
    994 dge_shutdown(device_t self, int howto)
    995 {
    996 	struct dge_softc *sc;
    997 
    998 	sc = device_private(self);
    999 	dge_stop(&sc->sc_ethercom.ec_if, 1);
   1000 
   1001 	return true;
   1002 }
   1003 
   1004 /*
   1005  * dge_tx_cksum:
   1006  *
   1007  *	Set up TCP/IP checksumming parameters for the
   1008  *	specified packet.
   1009  */
   1010 static int
   1011 dge_tx_cksum(struct dge_softc *sc, struct dge_txsoft *txs, uint8_t *fieldsp)
   1012 {
   1013 	struct mbuf *m0 = txs->txs_mbuf;
   1014 	struct dge_ctdes *t;
   1015 	uint32_t ipcs, tucs;
   1016 	struct ether_header *eh;
   1017 	int offset, iphl;
   1018 	uint8_t fields = 0;
   1019 
   1020 	/*
   1021 	 * XXX It would be nice if the mbuf pkthdr had offset
   1022 	 * fields for the protocol headers.
   1023 	 */
   1024 
   1025 	eh = mtod(m0, struct ether_header *);
   1026 	switch (htons(eh->ether_type)) {
   1027 	case ETHERTYPE_IP:
   1028 		offset = ETHER_HDR_LEN;
   1029 		break;
   1030 
   1031 	case ETHERTYPE_VLAN:
   1032 		offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
   1033 		break;
   1034 
   1035 	default:
   1036 		/*
   1037 		 * Don't support this protocol or encapsulation.
   1038 		 */
   1039 		*fieldsp = 0;
   1040 		return (0);
   1041 	}
   1042 
   1043 	iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
   1044 
   1045 	/*
   1046 	 * NOTE: Even if we're not using the IP or TCP/UDP checksum
   1047 	 * offload feature, if we load the context descriptor, we
   1048 	 * MUST provide valid values for IPCSS and TUCSS fields.
   1049 	 */
   1050 
   1051 	if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4) {
   1052 		DGE_EVCNT_INCR(&sc->sc_ev_txipsum);
   1053 		fields |= TDESC_POPTS_IXSM;
   1054 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1055 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1056 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1057 	} else if (__predict_true(sc->sc_txctx_ipcs != 0xffffffff)) {
   1058 		/* Use the cached value. */
   1059 		ipcs = sc->sc_txctx_ipcs;
   1060 	} else {
   1061 		/* Just initialize it to the likely value anyway. */
   1062 		ipcs = DGE_TCPIP_IPCSS(offset) |
   1063 		    DGE_TCPIP_IPCSO(offset + offsetof(struct ip, ip_sum)) |
   1064 		    DGE_TCPIP_IPCSE(offset + iphl - 1);
   1065 	}
   1066 	DPRINTF(DGE_DEBUG_CKSUM,
   1067 	    ("%s: CKSUM: offset %d ipcs 0x%x\n",
   1068 	    device_xname(&sc->sc_dev), offset, ipcs));
   1069 
   1070 	offset += iphl;
   1071 
   1072 	if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) {
   1073 		DGE_EVCNT_INCR(&sc->sc_ev_txtusum);
   1074 		fields |= TDESC_POPTS_TXSM;
   1075 		tucs = DGE_TCPIP_TUCSS(offset) |
   1076 		   DGE_TCPIP_TUCSO(offset + M_CSUM_DATA_IPv4_OFFSET(m0->m_pkthdr.csum_data)) |
   1077 		   DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1078 	} else if (__predict_true(sc->sc_txctx_tucs != 0xffffffff)) {
   1079 		/* Use the cached value. */
   1080 		tucs = sc->sc_txctx_tucs;
   1081 	} else {
   1082 		/* Just initialize it to a valid TCP context. */
   1083 		tucs = DGE_TCPIP_TUCSS(offset) |
   1084 		    DGE_TCPIP_TUCSO(offset + offsetof(struct tcphdr, th_sum)) |
   1085 		    DGE_TCPIP_TUCSE(0) /* rest of packet */;
   1086 	}
   1087 
   1088 	DPRINTF(DGE_DEBUG_CKSUM,
   1089 	    ("%s: CKSUM: offset %d tucs 0x%x\n",
   1090 	    device_xname(&sc->sc_dev), offset, tucs));
   1091 
   1092 	if (sc->sc_txctx_ipcs == ipcs &&
   1093 	    sc->sc_txctx_tucs == tucs) {
   1094 		/* Cached context is fine. */
   1095 		DGE_EVCNT_INCR(&sc->sc_ev_txctx_hit);
   1096 	} else {
   1097 		/* Fill in the context descriptor. */
   1098 #ifdef DGE_EVENT_COUNTERS
   1099 		if (sc->sc_txctx_ipcs == 0xffffffff &&
   1100 		    sc->sc_txctx_tucs == 0xffffffff)
   1101 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_init);
   1102 		else
   1103 			DGE_EVCNT_INCR(&sc->sc_ev_txctx_miss);
   1104 #endif
   1105 		t = (struct dge_ctdes *)&sc->sc_txdescs[sc->sc_txnext];
   1106 		t->dc_tcpip_ipcs = htole32(ipcs);
   1107 		t->dc_tcpip_tucs = htole32(tucs);
   1108 		t->dc_tcpip_cmdlen = htole32(TDESC_DTYP_CTD);
   1109 		t->dc_tcpip_seg = 0;
   1110 		DGE_CDTXSYNC(sc, sc->sc_txnext, 1, BUS_DMASYNC_PREWRITE);
   1111 
   1112 		sc->sc_txctx_ipcs = ipcs;
   1113 		sc->sc_txctx_tucs = tucs;
   1114 
   1115 		sc->sc_txnext = DGE_NEXTTX(sc->sc_txnext);
   1116 		txs->txs_ndesc++;
   1117 	}
   1118 
   1119 	*fieldsp = fields;
   1120 
   1121 	return (0);
   1122 }
   1123 
   1124 /*
   1125  * dge_start:		[ifnet interface function]
   1126  *
   1127  *	Start packet transmission on the interface.
   1128  */
   1129 static void
   1130 dge_start(struct ifnet *ifp)
   1131 {
   1132 	struct dge_softc *sc = ifp->if_softc;
   1133 	struct mbuf *m0;
   1134 	struct dge_txsoft *txs;
   1135 	bus_dmamap_t dmamap;
   1136 	int error, nexttx, lasttx = -1, ofree, seg;
   1137 	uint32_t cksumcmd;
   1138 	uint8_t cksumfields;
   1139 
   1140 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
   1141 		return;
   1142 
   1143 	/*
   1144 	 * Remember the previous number of free descriptors.
   1145 	 */
   1146 	ofree = sc->sc_txfree;
   1147 
   1148 	/*
   1149 	 * Loop through the send queue, setting up transmit descriptors
   1150 	 * until we drain the queue, or use up all available transmit
   1151 	 * descriptors.
   1152 	 */
   1153 	for (;;) {
   1154 		/* Grab a packet off the queue. */
   1155 		IFQ_POLL(&ifp->if_snd, m0);
   1156 		if (m0 == NULL)
   1157 			break;
   1158 
   1159 		DPRINTF(DGE_DEBUG_TX,
   1160 		    ("%s: TX: have packet to transmit: %p\n",
   1161 		    device_xname(&sc->sc_dev), m0));
   1162 
   1163 		/* Get a work queue entry. */
   1164 		if (sc->sc_txsfree < DGE_TXQUEUE_GC) {
   1165 			dge_txintr(sc);
   1166 			if (sc->sc_txsfree == 0) {
   1167 				DPRINTF(DGE_DEBUG_TX,
   1168 				    ("%s: TX: no free job descriptors\n",
   1169 					device_xname(&sc->sc_dev)));
   1170 				DGE_EVCNT_INCR(&sc->sc_ev_txsstall);
   1171 				break;
   1172 			}
   1173 		}
   1174 
   1175 		txs = &sc->sc_txsoft[sc->sc_txsnext];
   1176 		dmamap = txs->txs_dmamap;
   1177 
   1178 		/*
   1179 		 * Load the DMA map.  If this fails, the packet either
   1180 		 * didn't fit in the allotted number of segments, or we
   1181 		 * were short on resources.  For the too-many-segments
   1182 		 * case, we simply report an error and drop the packet,
   1183 		 * since we can't sanely copy a jumbo packet to a single
   1184 		 * buffer.
   1185 		 */
   1186 		error = bus_dmamap_load_mbuf(sc->sc_dmat, dmamap, m0,
   1187 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1188 		if (error) {
   1189 			if (error == EFBIG) {
   1190 				DGE_EVCNT_INCR(&sc->sc_ev_txdrop);
   1191 				printf("%s: Tx packet consumes too many "
   1192 				    "DMA segments, dropping...\n",
   1193 				    device_xname(&sc->sc_dev));
   1194 				IFQ_DEQUEUE(&ifp->if_snd, m0);
   1195 				m_freem(m0);
   1196 				continue;
   1197 			}
   1198 			/*
   1199 			 * Short on resources, just stop for now.
   1200 			 */
   1201 			DPRINTF(DGE_DEBUG_TX,
   1202 			    ("%s: TX: dmamap load failed: %d\n",
   1203 			    device_xname(&sc->sc_dev), error));
   1204 			break;
   1205 		}
   1206 
   1207 		/*
   1208 		 * Ensure we have enough descriptors free to describe
   1209 		 * the packet.  Note, we always reserve one descriptor
   1210 		 * at the end of the ring due to the semantics of the
   1211 		 * TDT register, plus one more in the event we need
   1212 		 * to re-load checksum offload context.
   1213 		 */
   1214 		if (dmamap->dm_nsegs > (sc->sc_txfree - 2)) {
   1215 			/*
   1216 			 * Not enough free descriptors to transmit this
   1217 			 * packet.  We haven't committed anything yet,
   1218 			 * so just unload the DMA map, put the packet
   1219 			 * pack on the queue, and punt.  Notify the upper
   1220 			 * layer that there are no more slots left.
   1221 			 */
   1222 			DPRINTF(DGE_DEBUG_TX,
   1223 			    ("%s: TX: need %d descriptors, have %d\n",
   1224 			    device_xname(&sc->sc_dev), dmamap->dm_nsegs,
   1225 			    sc->sc_txfree - 1));
   1226 			ifp->if_flags |= IFF_OACTIVE;
   1227 			bus_dmamap_unload(sc->sc_dmat, dmamap);
   1228 			DGE_EVCNT_INCR(&sc->sc_ev_txdstall);
   1229 			break;
   1230 		}
   1231 
   1232 		IFQ_DEQUEUE(&ifp->if_snd, m0);
   1233 
   1234 		/*
   1235 		 * WE ARE NOW COMMITTED TO TRANSMITTING THE PACKET.
   1236 		 */
   1237 
   1238 		/* Sync the DMA map. */
   1239 		bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
   1240 		    BUS_DMASYNC_PREWRITE);
   1241 
   1242 		DPRINTF(DGE_DEBUG_TX,
   1243 		    ("%s: TX: packet has %d DMA segments\n",
   1244 		    device_xname(&sc->sc_dev), dmamap->dm_nsegs));
   1245 
   1246 		DGE_EVCNT_INCR(&sc->sc_ev_txseg[dmamap->dm_nsegs - 1]);
   1247 
   1248 		/*
   1249 		 * Store a pointer to the packet so that we can free it
   1250 		 * later.
   1251 		 *
   1252 		 * Initially, we consider the number of descriptors the
   1253 		 * packet uses the number of DMA segments.  This may be
   1254 		 * incremented by 1 if we do checksum offload (a descriptor
   1255 		 * is used to set the checksum context).
   1256 		 */
   1257 		txs->txs_mbuf = m0;
   1258 		txs->txs_firstdesc = sc->sc_txnext;
   1259 		txs->txs_ndesc = dmamap->dm_nsegs;
   1260 
   1261 		/*
   1262 		 * Set up checksum offload parameters for
   1263 		 * this packet.
   1264 		 */
   1265 		if (m0->m_pkthdr.csum_flags &
   1266 		    (M_CSUM_IPv4|M_CSUM_TCPv4|M_CSUM_UDPv4)) {
   1267 			if (dge_tx_cksum(sc, txs, &cksumfields) != 0) {
   1268 				/* Error message already displayed. */
   1269 				bus_dmamap_unload(sc->sc_dmat, dmamap);
   1270 				continue;
   1271 			}
   1272 		} else {
   1273 			cksumfields = 0;
   1274 		}
   1275 
   1276 		cksumcmd = TDESC_DCMD_IDE | TDESC_DTYP_DATA;
   1277 
   1278 		/*
   1279 		 * Initialize the transmit descriptor.
   1280 		 */
   1281 		for (nexttx = sc->sc_txnext, seg = 0;
   1282 		     seg < dmamap->dm_nsegs;
   1283 		     seg++, nexttx = DGE_NEXTTX(nexttx)) {
   1284 			/*
   1285 			 * Note: we currently only use 32-bit DMA
   1286 			 * addresses.
   1287 			 */
   1288 			sc->sc_txdescs[nexttx].dt_baddrh = 0;
   1289 			sc->sc_txdescs[nexttx].dt_baddrl =
   1290 			    htole32(dmamap->dm_segs[seg].ds_addr);
   1291 			sc->sc_txdescs[nexttx].dt_ctl =
   1292 			    htole32(cksumcmd | dmamap->dm_segs[seg].ds_len);
   1293 			sc->sc_txdescs[nexttx].dt_status = 0;
   1294 			sc->sc_txdescs[nexttx].dt_popts = cksumfields;
   1295 			sc->sc_txdescs[nexttx].dt_vlan = 0;
   1296 			lasttx = nexttx;
   1297 
   1298 			DPRINTF(DGE_DEBUG_TX,
   1299 			    ("%s: TX: desc %d: low 0x%08lx, len 0x%04lx\n",
   1300 			    device_xname(&sc->sc_dev), nexttx,
   1301 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_addr),
   1302 			    (unsigned long)le32toh(dmamap->dm_segs[seg].ds_len)));
   1303 		}
   1304 
   1305 		KASSERT(lasttx != -1);
   1306 
   1307 		/*
   1308 		 * Set up the command byte on the last descriptor of
   1309 		 * the packet.  If we're in the interrupt delay window,
   1310 		 * delay the interrupt.
   1311 		 */
   1312 		sc->sc_txdescs[lasttx].dt_ctl |=
   1313 		    htole32(TDESC_DCMD_EOP | TDESC_DCMD_RS);
   1314 
   1315 		txs->txs_lastdesc = lasttx;
   1316 
   1317 		DPRINTF(DGE_DEBUG_TX,
   1318 		    ("%s: TX: desc %d: cmdlen 0x%08x\n", device_xname(&sc->sc_dev),
   1319 		    lasttx, le32toh(sc->sc_txdescs[lasttx].dt_ctl)));
   1320 
   1321 		/* Sync the descriptors we're using. */
   1322 		DGE_CDTXSYNC(sc, sc->sc_txnext, dmamap->dm_nsegs,
   1323 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1324 
   1325 		/* Give the packet to the chip. */
   1326 		CSR_WRITE(sc, DGE_TDT, nexttx);
   1327 
   1328 		DPRINTF(DGE_DEBUG_TX,
   1329 		    ("%s: TX: TDT -> %d\n", device_xname(&sc->sc_dev), nexttx));
   1330 
   1331 		DPRINTF(DGE_DEBUG_TX,
   1332 		    ("%s: TX: finished transmitting packet, job %d\n",
   1333 		    device_xname(&sc->sc_dev), sc->sc_txsnext));
   1334 
   1335 		/* Advance the tx pointer. */
   1336 		sc->sc_txfree -= txs->txs_ndesc;
   1337 		sc->sc_txnext = nexttx;
   1338 
   1339 		sc->sc_txsfree--;
   1340 		sc->sc_txsnext = DGE_NEXTTXS(sc->sc_txsnext);
   1341 
   1342 		/* Pass the packet to any BPF listeners. */
   1343 		bpf_mtap(ifp, m0);
   1344 	}
   1345 
   1346 	if (sc->sc_txsfree == 0 || sc->sc_txfree <= 2) {
   1347 		/* No more slots; notify upper layer. */
   1348 		ifp->if_flags |= IFF_OACTIVE;
   1349 	}
   1350 
   1351 	if (sc->sc_txfree != ofree) {
   1352 		/* Set a watchdog timer in case the chip flakes out. */
   1353 		ifp->if_timer = 5;
   1354 	}
   1355 }
   1356 
   1357 /*
   1358  * dge_watchdog:		[ifnet interface function]
   1359  *
   1360  *	Watchdog timer handler.
   1361  */
   1362 static void
   1363 dge_watchdog(struct ifnet *ifp)
   1364 {
   1365 	struct dge_softc *sc = ifp->if_softc;
   1366 
   1367 	/*
   1368 	 * Since we're using delayed interrupts, sweep up
   1369 	 * before we report an error.
   1370 	 */
   1371 	dge_txintr(sc);
   1372 
   1373 	if (sc->sc_txfree != DGE_NTXDESC) {
   1374 		printf("%s: device timeout (txfree %d txsfree %d txnext %d)\n",
   1375 		    device_xname(&sc->sc_dev), sc->sc_txfree, sc->sc_txsfree,
   1376 		    sc->sc_txnext);
   1377 		ifp->if_oerrors++;
   1378 
   1379 		/* Reset the interface. */
   1380 		(void) dge_init(ifp);
   1381 	}
   1382 
   1383 	/* Try to get more packets going. */
   1384 	dge_start(ifp);
   1385 }
   1386 
   1387 /*
   1388  * dge_ioctl:		[ifnet interface function]
   1389  *
   1390  *	Handle control requests from the operator.
   1391  */
   1392 static int
   1393 dge_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   1394 {
   1395 	struct dge_softc *sc = ifp->if_softc;
   1396 	struct ifreq *ifr = (struct ifreq *) data;
   1397 	pcireg_t preg;
   1398 	int s, error, mmrbc;
   1399 
   1400 	s = splnet();
   1401 
   1402 	switch (cmd) {
   1403 	case SIOCSIFMEDIA:
   1404 	case SIOCGIFMEDIA:
   1405 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
   1406 		break;
   1407 
   1408 	case SIOCSIFMTU:
   1409 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > DGE_MAX_MTU)
   1410 			error = EINVAL;
   1411 		else if ((error = ifioctl_common(ifp, cmd, data)) != ENETRESET)
   1412 			break;
   1413 		else if (ifp->if_flags & IFF_UP)
   1414 			error = (*ifp->if_init)(ifp);
   1415 		else
   1416 			error = 0;
   1417 		break;
   1418 
   1419         case SIOCSIFFLAGS:
   1420 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   1421 			break;
   1422 		/* extract link flags */
   1423 		if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1424 		    (ifp->if_flags & IFF_LINK1) == 0)
   1425 			mmrbc = PCIX_MMRBC_512;
   1426 		else if ((ifp->if_flags & IFF_LINK0) == 0 &&
   1427 		    (ifp->if_flags & IFF_LINK1) != 0)
   1428 			mmrbc = PCIX_MMRBC_1024;
   1429 		else if ((ifp->if_flags & IFF_LINK0) != 0 &&
   1430 		    (ifp->if_flags & IFF_LINK1) == 0)
   1431 			mmrbc = PCIX_MMRBC_2048;
   1432 		else
   1433 			mmrbc = PCIX_MMRBC_4096;
   1434 		if (mmrbc != sc->sc_mmrbc) {
   1435 			preg = pci_conf_read(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD);
   1436 			preg &= ~PCIX_MMRBC_MSK;
   1437 			preg |= mmrbc;
   1438 			pci_conf_write(sc->sc_pc, sc->sc_pt,DGE_PCIX_CMD, preg);
   1439 			sc->sc_mmrbc = mmrbc;
   1440 		}
   1441                 /* FALLTHROUGH */
   1442 	default:
   1443 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
   1444 			break;
   1445 
   1446 		error = 0;
   1447 
   1448 		if (cmd == SIOCSIFCAP)
   1449 			error = (*ifp->if_init)(ifp);
   1450 		else if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
   1451 			;
   1452 		else if (ifp->if_flags & IFF_RUNNING) {
   1453 			/*
   1454 			 * Multicast list has changed; set the hardware filter
   1455 			 * accordingly.
   1456 			 */
   1457 			dge_set_filter(sc);
   1458 		}
   1459 		break;
   1460 	}
   1461 
   1462 	/* Try to get more packets going. */
   1463 	dge_start(ifp);
   1464 
   1465 	splx(s);
   1466 	return (error);
   1467 }
   1468 
   1469 /*
   1470  * dge_intr:
   1471  *
   1472  *	Interrupt service routine.
   1473  */
   1474 static int
   1475 dge_intr(void *arg)
   1476 {
   1477 	struct dge_softc *sc = arg;
   1478 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1479 	uint32_t icr;
   1480 	int wantinit, handled = 0;
   1481 
   1482 	for (wantinit = 0; wantinit == 0;) {
   1483 		icr = CSR_READ(sc, DGE_ICR);
   1484 		if ((icr & sc->sc_icr) == 0)
   1485 			break;
   1486 
   1487 		rnd_add_uint32(&sc->rnd_source, icr);
   1488 
   1489 		handled = 1;
   1490 
   1491 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1492 		if (icr & (ICR_RXDMT0|ICR_RXT0)) {
   1493 			DPRINTF(DGE_DEBUG_RX,
   1494 			    ("%s: RX: got Rx intr 0x%08x\n",
   1495 			    device_xname(&sc->sc_dev),
   1496 			    icr & (ICR_RXDMT0|ICR_RXT0)));
   1497 			DGE_EVCNT_INCR(&sc->sc_ev_rxintr);
   1498 		}
   1499 #endif
   1500 		dge_rxintr(sc);
   1501 
   1502 #if defined(DGE_DEBUG) || defined(DGE_EVENT_COUNTERS)
   1503 		if (icr & ICR_TXDW) {
   1504 			DPRINTF(DGE_DEBUG_TX,
   1505 			    ("%s: TX: got TXDW interrupt\n",
   1506 			    device_xname(&sc->sc_dev)));
   1507 			DGE_EVCNT_INCR(&sc->sc_ev_txdw);
   1508 		}
   1509 		if (icr & ICR_TXQE)
   1510 			DGE_EVCNT_INCR(&sc->sc_ev_txqe);
   1511 #endif
   1512 		dge_txintr(sc);
   1513 
   1514 		if (icr & (ICR_LSC|ICR_RXSEQ)) {
   1515 			DGE_EVCNT_INCR(&sc->sc_ev_linkintr);
   1516 			dge_linkintr(sc, icr);
   1517 		}
   1518 
   1519 		if (icr & ICR_RXO) {
   1520 			printf("%s: Receive overrun\n", device_xname(&sc->sc_dev));
   1521 			wantinit = 1;
   1522 		}
   1523 	}
   1524 
   1525 	if (handled) {
   1526 		if (wantinit)
   1527 			dge_init(ifp);
   1528 
   1529 		/* Try to get more packets going. */
   1530 		dge_start(ifp);
   1531 	}
   1532 
   1533 	return (handled);
   1534 }
   1535 
   1536 /*
   1537  * dge_txintr:
   1538  *
   1539  *	Helper; handle transmit interrupts.
   1540  */
   1541 static void
   1542 dge_txintr(struct dge_softc *sc)
   1543 {
   1544 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1545 	struct dge_txsoft *txs;
   1546 	uint8_t status;
   1547 	int i;
   1548 
   1549 	ifp->if_flags &= ~IFF_OACTIVE;
   1550 
   1551 	/*
   1552 	 * Go through the Tx list and free mbufs for those
   1553 	 * frames which have been transmitted.
   1554 	 */
   1555 	for (i = sc->sc_txsdirty; sc->sc_txsfree != DGE_TXQUEUELEN;
   1556 	     i = DGE_NEXTTXS(i), sc->sc_txsfree++) {
   1557 		txs = &sc->sc_txsoft[i];
   1558 
   1559 		DPRINTF(DGE_DEBUG_TX,
   1560 		    ("%s: TX: checking job %d\n", device_xname(&sc->sc_dev), i));
   1561 
   1562 		DGE_CDTXSYNC(sc, txs->txs_firstdesc, txs->txs_dmamap->dm_nsegs,
   1563 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1564 
   1565 		status =
   1566 		    sc->sc_txdescs[txs->txs_lastdesc].dt_status;
   1567 		if ((status & TDESC_STA_DD) == 0) {
   1568 			DGE_CDTXSYNC(sc, txs->txs_lastdesc, 1,
   1569 			    BUS_DMASYNC_PREREAD);
   1570 			break;
   1571 		}
   1572 
   1573 		DPRINTF(DGE_DEBUG_TX,
   1574 		    ("%s: TX: job %d done: descs %d..%d\n",
   1575 		    device_xname(&sc->sc_dev), i, txs->txs_firstdesc,
   1576 		    txs->txs_lastdesc));
   1577 
   1578 		ifp->if_opackets++;
   1579 		sc->sc_txfree += txs->txs_ndesc;
   1580 		bus_dmamap_sync(sc->sc_dmat, txs->txs_dmamap,
   1581 		    0, txs->txs_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1582 		bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   1583 		m_freem(txs->txs_mbuf);
   1584 		txs->txs_mbuf = NULL;
   1585 	}
   1586 
   1587 	/* Update the dirty transmit buffer pointer. */
   1588 	sc->sc_txsdirty = i;
   1589 	DPRINTF(DGE_DEBUG_TX,
   1590 	    ("%s: TX: txsdirty -> %d\n", device_xname(&sc->sc_dev), i));
   1591 
   1592 	/*
   1593 	 * If there are no more pending transmissions, cancel the watchdog
   1594 	 * timer.
   1595 	 */
   1596 	if (sc->sc_txsfree == DGE_TXQUEUELEN)
   1597 		ifp->if_timer = 0;
   1598 }
   1599 
   1600 /*
   1601  * dge_rxintr:
   1602  *
   1603  *	Helper; handle receive interrupts.
   1604  */
   1605 static void
   1606 dge_rxintr(struct dge_softc *sc)
   1607 {
   1608 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   1609 	struct dge_rxsoft *rxs;
   1610 	struct mbuf *m;
   1611 	int i, len;
   1612 	uint8_t status, errors;
   1613 
   1614 	for (i = sc->sc_rxptr;; i = DGE_NEXTRX(i)) {
   1615 		rxs = &sc->sc_rxsoft[i];
   1616 
   1617 		DPRINTF(DGE_DEBUG_RX,
   1618 		    ("%s: RX: checking descriptor %d\n",
   1619 		    device_xname(&sc->sc_dev), i));
   1620 
   1621 		DGE_CDRXSYNC(sc, i, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1622 
   1623 		status = sc->sc_rxdescs[i].dr_status;
   1624 		errors = sc->sc_rxdescs[i].dr_errors;
   1625 		len = le16toh(sc->sc_rxdescs[i].dr_len);
   1626 
   1627 		if ((status & RDESC_STS_DD) == 0) {
   1628 			/*
   1629 			 * We have processed all of the receive descriptors.
   1630 			 */
   1631 			DGE_CDRXSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1632 			break;
   1633 		}
   1634 
   1635 		if (__predict_false(sc->sc_rxdiscard)) {
   1636 			DPRINTF(DGE_DEBUG_RX,
   1637 			    ("%s: RX: discarding contents of descriptor %d\n",
   1638 			    device_xname(&sc->sc_dev), i));
   1639 			DGE_INIT_RXDESC(sc, i);
   1640 			if (status & RDESC_STS_EOP) {
   1641 				/* Reset our state. */
   1642 				DPRINTF(DGE_DEBUG_RX,
   1643 				    ("%s: RX: resetting rxdiscard -> 0\n",
   1644 				    device_xname(&sc->sc_dev)));
   1645 				sc->sc_rxdiscard = 0;
   1646 			}
   1647 			continue;
   1648 		}
   1649 
   1650 		bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1651 		    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
   1652 
   1653 		m = rxs->rxs_mbuf;
   1654 
   1655 		/*
   1656 		 * Add a new receive buffer to the ring.
   1657 		 */
   1658 		if (dge_add_rxbuf(sc, i) != 0) {
   1659 			/*
   1660 			 * Failed, throw away what we've done so
   1661 			 * far, and discard the rest of the packet.
   1662 			 */
   1663 			ifp->if_ierrors++;
   1664 			bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   1665 			    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   1666 			DGE_INIT_RXDESC(sc, i);
   1667 			if ((status & RDESC_STS_EOP) == 0)
   1668 				sc->sc_rxdiscard = 1;
   1669 			if (sc->sc_rxhead != NULL)
   1670 				m_freem(sc->sc_rxhead);
   1671 			DGE_RXCHAIN_RESET(sc);
   1672 			DPRINTF(DGE_DEBUG_RX,
   1673 			    ("%s: RX: Rx buffer allocation failed, "
   1674 			    "dropping packet%s\n", device_xname(&sc->sc_dev),
   1675 			    sc->sc_rxdiscard ? " (discard)" : ""));
   1676 			continue;
   1677 		}
   1678 		DGE_INIT_RXDESC(sc, DGE_PREVRX(i)); /* Write the descriptor */
   1679 
   1680 		DGE_RXCHAIN_LINK(sc, m);
   1681 
   1682 		m->m_len = len;
   1683 
   1684 		DPRINTF(DGE_DEBUG_RX,
   1685 		    ("%s: RX: buffer at %p len %d\n",
   1686 		    device_xname(&sc->sc_dev), m->m_data, len));
   1687 
   1688 		/*
   1689 		 * If this is not the end of the packet, keep
   1690 		 * looking.
   1691 		 */
   1692 		if ((status & RDESC_STS_EOP) == 0) {
   1693 			sc->sc_rxlen += len;
   1694 			DPRINTF(DGE_DEBUG_RX,
   1695 			    ("%s: RX: not yet EOP, rxlen -> %d\n",
   1696 			    device_xname(&sc->sc_dev), sc->sc_rxlen));
   1697 			continue;
   1698 		}
   1699 
   1700 		/*
   1701 		 * Okay, we have the entire packet now...
   1702 		 */
   1703 		*sc->sc_rxtailp = NULL;
   1704 		m = sc->sc_rxhead;
   1705 		len += sc->sc_rxlen;
   1706 
   1707 		DGE_RXCHAIN_RESET(sc);
   1708 
   1709 		DPRINTF(DGE_DEBUG_RX,
   1710 		    ("%s: RX: have entire packet, len -> %d\n",
   1711 		    device_xname(&sc->sc_dev), len));
   1712 
   1713 		/*
   1714 		 * If an error occurred, update stats and drop the packet.
   1715 		 */
   1716 		if (errors &
   1717 		     (RDESC_ERR_CE|RDESC_ERR_SE|RDESC_ERR_P|RDESC_ERR_RXE)) {
   1718 			ifp->if_ierrors++;
   1719 			if (errors & RDESC_ERR_SE)
   1720 				printf("%s: symbol error\n",
   1721 				    device_xname(&sc->sc_dev));
   1722 			else if (errors & RDESC_ERR_P)
   1723 				printf("%s: parity error\n",
   1724 				    device_xname(&sc->sc_dev));
   1725 			else if (errors & RDESC_ERR_CE)
   1726 				printf("%s: CRC error\n",
   1727 				    device_xname(&sc->sc_dev));
   1728 			m_freem(m);
   1729 			continue;
   1730 		}
   1731 
   1732 		/*
   1733 		 * No errors.  Receive the packet.
   1734 		 */
   1735 		m->m_pkthdr.rcvif = ifp;
   1736 		m->m_pkthdr.len = len;
   1737 
   1738 		/*
   1739 		 * Set up checksum info for this packet.
   1740 		 */
   1741 		if (status & RDESC_STS_IPCS) {
   1742 			DGE_EVCNT_INCR(&sc->sc_ev_rxipsum);
   1743 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1744 			if (errors & RDESC_ERR_IPE)
   1745 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1746 		}
   1747 		if (status & RDESC_STS_TCPCS) {
   1748 			/*
   1749 			 * Note: we don't know if this was TCP or UDP,
   1750 			 * so we just set both bits, and expect the
   1751 			 * upper layers to deal.
   1752 			 */
   1753 			DGE_EVCNT_INCR(&sc->sc_ev_rxtusum);
   1754 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_UDPv4;
   1755 			if (errors & RDESC_ERR_TCPE)
   1756 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1757 		}
   1758 
   1759 		ifp->if_ipackets++;
   1760 
   1761 		/* Pass this up to any BPF listeners. */
   1762 		bpf_mtap(ifp, m);
   1763 
   1764 		/* Pass it on. */
   1765 		(*ifp->if_input)(ifp, m);
   1766 	}
   1767 
   1768 	/* Update the receive pointer. */
   1769 	sc->sc_rxptr = i;
   1770 
   1771 	DPRINTF(DGE_DEBUG_RX,
   1772 	    ("%s: RX: rxptr -> %d\n", device_xname(&sc->sc_dev), i));
   1773 }
   1774 
   1775 /*
   1776  * dge_linkintr:
   1777  *
   1778  *	Helper; handle link interrupts.
   1779  */
   1780 static void
   1781 dge_linkintr(struct dge_softc *sc, uint32_t icr)
   1782 {
   1783 	uint32_t status;
   1784 
   1785 	if (icr & ICR_LSC) {
   1786 		status = CSR_READ(sc, DGE_STATUS);
   1787 		if (status & STATUS_LINKUP) {
   1788 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> up\n",
   1789 			    device_xname(&sc->sc_dev)));
   1790 		} else {
   1791 			DPRINTF(DGE_DEBUG_LINK, ("%s: LINK: LSC -> down\n",
   1792 			    device_xname(&sc->sc_dev)));
   1793 		}
   1794 	} else if (icr & ICR_RXSEQ) {
   1795 		DPRINTF(DGE_DEBUG_LINK,
   1796 		    ("%s: LINK: Receive sequence error\n",
   1797 		    device_xname(&sc->sc_dev)));
   1798 	}
   1799 	/* XXX - fix errata */
   1800 }
   1801 
   1802 /*
   1803  * dge_reset:
   1804  *
   1805  *	Reset the i82597 chip.
   1806  */
   1807 static void
   1808 dge_reset(struct dge_softc *sc)
   1809 {
   1810 	int i;
   1811 
   1812 	/*
   1813 	 * Do a chip reset.
   1814 	 */
   1815 	CSR_WRITE(sc, DGE_CTRL0, CTRL0_RST | sc->sc_ctrl0);
   1816 
   1817 	delay(10000);
   1818 
   1819 	for (i = 0; i < 1000; i++) {
   1820 		if ((CSR_READ(sc, DGE_CTRL0) & CTRL0_RST) == 0)
   1821 			break;
   1822 		delay(20);
   1823 	}
   1824 
   1825 	if (CSR_READ(sc, DGE_CTRL0) & CTRL0_RST)
   1826 		printf("%s: WARNING: reset failed to complete\n",
   1827 		    device_xname(&sc->sc_dev));
   1828         /*
   1829          * Reset the EEPROM logic.
   1830          * This will cause the chip to reread its default values,
   1831 	 * which doesn't happen otherwise (errata).
   1832          */
   1833         CSR_WRITE(sc, DGE_CTRL1, CTRL1_EE_RST);
   1834         delay(10000);
   1835 }
   1836 
   1837 /*
   1838  * dge_init:		[ifnet interface function]
   1839  *
   1840  *	Initialize the interface.  Must be called at splnet().
   1841  */
   1842 static int
   1843 dge_init(struct ifnet *ifp)
   1844 {
   1845 	struct dge_softc *sc = ifp->if_softc;
   1846 	struct dge_rxsoft *rxs;
   1847 	int i, error = 0;
   1848 	uint32_t reg;
   1849 
   1850 	/*
   1851 	 * *_HDR_ALIGNED_P is constant 1 if __NO_STRICT_ALIGMENT is set.
   1852 	 * There is a small but measurable benefit to avoiding the adjusment
   1853 	 * of the descriptor so that the headers are aligned, for normal mtu,
   1854 	 * on such platforms.  One possibility is that the DMA itself is
   1855 	 * slightly more efficient if the front of the entire packet (instead
   1856 	 * of the front of the headers) is aligned.
   1857 	 *
   1858 	 * Note we must always set align_tweak to 0 if we are using
   1859 	 * jumbo frames.
   1860 	 */
   1861 #ifdef __NO_STRICT_ALIGNMENT
   1862 	sc->sc_align_tweak = 0;
   1863 #else
   1864 	if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN) > (MCLBYTES - 2))
   1865 		sc->sc_align_tweak = 0;
   1866 	else
   1867 		sc->sc_align_tweak = 2;
   1868 #endif /* __NO_STRICT_ALIGNMENT */
   1869 
   1870 	/* Cancel any pending I/O. */
   1871 	dge_stop(ifp, 0);
   1872 
   1873 	/* Reset the chip to a known state. */
   1874 	dge_reset(sc);
   1875 
   1876 	/* Initialize the transmit descriptor ring. */
   1877 	memset(sc->sc_txdescs, 0, sizeof(sc->sc_txdescs));
   1878 	DGE_CDTXSYNC(sc, 0, DGE_NTXDESC,
   1879 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1880 	sc->sc_txfree = DGE_NTXDESC;
   1881 	sc->sc_txnext = 0;
   1882 
   1883 	sc->sc_txctx_ipcs = 0xffffffff;
   1884 	sc->sc_txctx_tucs = 0xffffffff;
   1885 
   1886 	CSR_WRITE(sc, DGE_TDBAH, 0);
   1887 	CSR_WRITE(sc, DGE_TDBAL, DGE_CDTXADDR(sc, 0));
   1888 	CSR_WRITE(sc, DGE_TDLEN, sizeof(sc->sc_txdescs));
   1889 	CSR_WRITE(sc, DGE_TDH, 0);
   1890 	CSR_WRITE(sc, DGE_TDT, 0);
   1891 	CSR_WRITE(sc, DGE_TIDV, TIDV);
   1892 
   1893 #if 0
   1894 	CSR_WRITE(sc, DGE_TXDCTL, TXDCTL_PTHRESH(0) |
   1895 	    TXDCTL_HTHRESH(0) | TXDCTL_WTHRESH(0));
   1896 #endif
   1897 	CSR_WRITE(sc, DGE_RXDCTL,
   1898 	    RXDCTL_PTHRESH(RXDCTL_PTHRESH_VAL) |
   1899 	    RXDCTL_HTHRESH(RXDCTL_HTHRESH_VAL) |
   1900 	    RXDCTL_WTHRESH(RXDCTL_WTHRESH_VAL));
   1901 
   1902 	/* Initialize the transmit job descriptors. */
   1903 	for (i = 0; i < DGE_TXQUEUELEN; i++)
   1904 		sc->sc_txsoft[i].txs_mbuf = NULL;
   1905 	sc->sc_txsfree = DGE_TXQUEUELEN;
   1906 	sc->sc_txsnext = 0;
   1907 	sc->sc_txsdirty = 0;
   1908 
   1909 	/*
   1910 	 * Initialize the receive descriptor and receive job
   1911 	 * descriptor rings.
   1912 	 */
   1913 	CSR_WRITE(sc, DGE_RDBAH, 0);
   1914 	CSR_WRITE(sc, DGE_RDBAL, DGE_CDRXADDR(sc, 0));
   1915 	CSR_WRITE(sc, DGE_RDLEN, sizeof(sc->sc_rxdescs));
   1916 	CSR_WRITE(sc, DGE_RDH, DGE_RXSPACE);
   1917 	CSR_WRITE(sc, DGE_RDT, 0);
   1918 	CSR_WRITE(sc, DGE_RDTR, RDTR | 0x80000000);
   1919 	CSR_WRITE(sc, DGE_FCRTL, FCRTL | FCRTL_XONE);
   1920 	CSR_WRITE(sc, DGE_FCRTH, FCRTH);
   1921 
   1922 	for (i = 0; i < DGE_NRXDESC; i++) {
   1923 		rxs = &sc->sc_rxsoft[i];
   1924 		if (rxs->rxs_mbuf == NULL) {
   1925 			if ((error = dge_add_rxbuf(sc, i)) != 0) {
   1926 				printf("%s: unable to allocate or map rx "
   1927 				    "buffer %d, error = %d\n",
   1928 				    device_xname(&sc->sc_dev), i, error);
   1929 				/*
   1930 				 * XXX Should attempt to run with fewer receive
   1931 				 * XXX buffers instead of just failing.
   1932 				 */
   1933 				dge_rxdrain(sc);
   1934 				goto out;
   1935 			}
   1936 		}
   1937 		DGE_INIT_RXDESC(sc, i);
   1938 	}
   1939 	sc->sc_rxptr = DGE_RXSPACE;
   1940 	sc->sc_rxdiscard = 0;
   1941 	DGE_RXCHAIN_RESET(sc);
   1942 
   1943 	if (sc->sc_ethercom.ec_capabilities & ETHERCAP_JUMBO_MTU) {
   1944 		sc->sc_ctrl0 |= CTRL0_JFE;
   1945 		CSR_WRITE(sc, DGE_MFS, ETHER_MAX_LEN_JUMBO << 16);
   1946 	}
   1947 
   1948 	/* Write the control registers. */
   1949 	CSR_WRITE(sc, DGE_CTRL0, sc->sc_ctrl0);
   1950 
   1951 	/*
   1952 	 * Set up checksum offload parameters.
   1953 	 */
   1954 	reg = CSR_READ(sc, DGE_RXCSUM);
   1955 	if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
   1956 		reg |= RXCSUM_IPOFL;
   1957 	else
   1958 		reg &= ~RXCSUM_IPOFL;
   1959 	if (ifp->if_capenable & (IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx))
   1960 		reg |= RXCSUM_IPOFL | RXCSUM_TUOFL;
   1961 	else {
   1962 		reg &= ~RXCSUM_TUOFL;
   1963 		if ((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) == 0)
   1964 			reg &= ~RXCSUM_IPOFL;
   1965 	}
   1966 	CSR_WRITE(sc, DGE_RXCSUM, reg);
   1967 
   1968 	/*
   1969 	 * Set up the interrupt registers.
   1970 	 */
   1971 	CSR_WRITE(sc, DGE_IMC, 0xffffffffU);
   1972 	sc->sc_icr = ICR_TXDW | ICR_LSC | ICR_RXSEQ | ICR_RXDMT0 |
   1973 	    ICR_RXO | ICR_RXT0;
   1974 
   1975 	CSR_WRITE(sc, DGE_IMS, sc->sc_icr);
   1976 
   1977 	/*
   1978 	 * Set up the transmit control register.
   1979 	 */
   1980 	sc->sc_tctl = TCTL_TCE|TCTL_TPDE|TCTL_TXEN;
   1981 	CSR_WRITE(sc, DGE_TCTL, sc->sc_tctl);
   1982 
   1983 	/*
   1984 	 * Set up the receive control register; we actually program
   1985 	 * the register when we set the receive filter.  Use multicast
   1986 	 * address offset type 0.
   1987 	 */
   1988 	sc->sc_mchash_type = 0;
   1989 
   1990 	sc->sc_rctl = RCTL_RXEN | RCTL_RDMTS_12 | RCTL_RPDA_MC |
   1991 	    RCTL_CFF | RCTL_SECRC | RCTL_MO(sc->sc_mchash_type);
   1992 
   1993 #ifdef DGE_OFFBYONE_RXBUG
   1994 	sc->sc_rctl |= RCTL_BSIZE_16k;
   1995 #else
   1996 	switch(MCLBYTES) {
   1997 	case 2048:
   1998 		sc->sc_rctl |= RCTL_BSIZE_2k;
   1999 		break;
   2000 	case 4096:
   2001 		sc->sc_rctl |= RCTL_BSIZE_4k;
   2002 		break;
   2003 	case 8192:
   2004 		sc->sc_rctl |= RCTL_BSIZE_8k;
   2005 		break;
   2006 	case 16384:
   2007 		sc->sc_rctl |= RCTL_BSIZE_16k;
   2008 		break;
   2009 	default:
   2010 		panic("dge_init: MCLBYTES %d unsupported", MCLBYTES);
   2011 	}
   2012 #endif
   2013 
   2014 	/* Set the receive filter. */
   2015 	/* Also sets RCTL */
   2016 	dge_set_filter(sc);
   2017 
   2018 	/* ...all done! */
   2019 	ifp->if_flags |= IFF_RUNNING;
   2020 	ifp->if_flags &= ~IFF_OACTIVE;
   2021 
   2022  out:
   2023 	if (error)
   2024 		printf("%s: interface not running\n", device_xname(&sc->sc_dev));
   2025 	return (error);
   2026 }
   2027 
   2028 /*
   2029  * dge_rxdrain:
   2030  *
   2031  *	Drain the receive queue.
   2032  */
   2033 static void
   2034 dge_rxdrain(struct dge_softc *sc)
   2035 {
   2036 	struct dge_rxsoft *rxs;
   2037 	int i;
   2038 
   2039 	for (i = 0; i < DGE_NRXDESC; i++) {
   2040 		rxs = &sc->sc_rxsoft[i];
   2041 		if (rxs->rxs_mbuf != NULL) {
   2042 			bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2043 			m_freem(rxs->rxs_mbuf);
   2044 			rxs->rxs_mbuf = NULL;
   2045 		}
   2046 	}
   2047 }
   2048 
   2049 /*
   2050  * dge_stop:		[ifnet interface function]
   2051  *
   2052  *	Stop transmission on the interface.
   2053  */
   2054 static void
   2055 dge_stop(struct ifnet *ifp, int disable)
   2056 {
   2057 	struct dge_softc *sc = ifp->if_softc;
   2058 	struct dge_txsoft *txs;
   2059 	int i;
   2060 
   2061 	/* Stop the transmit and receive processes. */
   2062 	CSR_WRITE(sc, DGE_TCTL, 0);
   2063 	CSR_WRITE(sc, DGE_RCTL, 0);
   2064 
   2065 	/* Release any queued transmit buffers. */
   2066 	for (i = 0; i < DGE_TXQUEUELEN; i++) {
   2067 		txs = &sc->sc_txsoft[i];
   2068 		if (txs->txs_mbuf != NULL) {
   2069 			bus_dmamap_unload(sc->sc_dmat, txs->txs_dmamap);
   2070 			m_freem(txs->txs_mbuf);
   2071 			txs->txs_mbuf = NULL;
   2072 		}
   2073 	}
   2074 
   2075 	/* Mark the interface as down and cancel the watchdog timer. */
   2076 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2077 	ifp->if_timer = 0;
   2078 
   2079 	if (disable)
   2080 		dge_rxdrain(sc);
   2081 }
   2082 
   2083 /*
   2084  * dge_add_rxbuf:
   2085  *
   2086  *	Add a receive buffer to the indiciated descriptor.
   2087  */
   2088 static int
   2089 dge_add_rxbuf(struct dge_softc *sc, int idx)
   2090 {
   2091 	struct dge_rxsoft *rxs = &sc->sc_rxsoft[idx];
   2092 	struct mbuf *m;
   2093 	int error;
   2094 #ifdef DGE_OFFBYONE_RXBUG
   2095 	void *buf;
   2096 #endif
   2097 
   2098 	MGETHDR(m, M_DONTWAIT, MT_DATA);
   2099 	if (m == NULL)
   2100 		return (ENOBUFS);
   2101 
   2102 #ifdef DGE_OFFBYONE_RXBUG
   2103 	if ((buf = dge_getbuf(sc)) == NULL)
   2104 		return ENOBUFS;
   2105 
   2106 	m->m_len = m->m_pkthdr.len = DGE_BUFFER_SIZE;
   2107 	MEXTADD(m, buf, DGE_BUFFER_SIZE, M_DEVBUF, dge_freebuf, sc);
   2108 	m->m_flags |= M_EXT_RW;
   2109 
   2110 	if (rxs->rxs_mbuf != NULL)
   2111 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2112 	rxs->rxs_mbuf = m;
   2113 
   2114 	error = bus_dmamap_load(sc->sc_dmat, rxs->rxs_dmamap, buf,
   2115 	    DGE_BUFFER_SIZE, NULL, BUS_DMA_READ|BUS_DMA_NOWAIT);
   2116 #else
   2117 	MCLGET(m, M_DONTWAIT);
   2118 	if ((m->m_flags & M_EXT) == 0) {
   2119 		m_freem(m);
   2120 		return (ENOBUFS);
   2121 	}
   2122 
   2123 	if (rxs->rxs_mbuf != NULL)
   2124 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   2125 
   2126 	rxs->rxs_mbuf = m;
   2127 
   2128 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
   2129 	error = bus_dmamap_load_mbuf(sc->sc_dmat, rxs->rxs_dmamap, m,
   2130 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   2131 #endif
   2132 	if (error) {
   2133 		printf("%s: unable to load rx DMA map %d, error = %d\n",
   2134 		    device_xname(&sc->sc_dev), idx, error);
   2135 		panic("dge_add_rxbuf");	/* XXX XXX XXX */
   2136 	}
   2137 	bus_dmamap_sync(sc->sc_dmat, rxs->rxs_dmamap, 0,
   2138 	    rxs->rxs_dmamap->dm_mapsize, BUS_DMASYNC_PREREAD);
   2139 
   2140 	return (0);
   2141 }
   2142 
   2143 /*
   2144  * dge_set_ral:
   2145  *
   2146  *	Set an entry in the receive address list.
   2147  */
   2148 static void
   2149 dge_set_ral(struct dge_softc *sc, const uint8_t *enaddr, int idx)
   2150 {
   2151 	uint32_t ral_lo, ral_hi;
   2152 
   2153 	if (enaddr != NULL) {
   2154 		ral_lo = enaddr[0] | (enaddr[1] << 8) | (enaddr[2] << 16) |
   2155 		    (enaddr[3] << 24);
   2156 		ral_hi = enaddr[4] | (enaddr[5] << 8);
   2157 		ral_hi |= RAH_AV;
   2158 	} else {
   2159 		ral_lo = 0;
   2160 		ral_hi = 0;
   2161 	}
   2162 	CSR_WRITE(sc, RA_ADDR(DGE_RAL, idx), ral_lo);
   2163 	CSR_WRITE(sc, RA_ADDR(DGE_RAH, idx), ral_hi);
   2164 }
   2165 
   2166 /*
   2167  * dge_mchash:
   2168  *
   2169  *	Compute the hash of the multicast address for the 4096-bit
   2170  *	multicast filter.
   2171  */
   2172 static uint32_t
   2173 dge_mchash(struct dge_softc *sc, const uint8_t *enaddr)
   2174 {
   2175 	static const int lo_shift[4] = { 4, 3, 2, 0 };
   2176 	static const int hi_shift[4] = { 4, 5, 6, 8 };
   2177 	uint32_t hash;
   2178 
   2179 	hash = (enaddr[4] >> lo_shift[sc->sc_mchash_type]) |
   2180 	    (((uint16_t) enaddr[5]) << hi_shift[sc->sc_mchash_type]);
   2181 
   2182 	return (hash & 0xfff);
   2183 }
   2184 
   2185 /*
   2186  * dge_set_filter:
   2187  *
   2188  *	Set up the receive filter.
   2189  */
   2190 static void
   2191 dge_set_filter(struct dge_softc *sc)
   2192 {
   2193 	struct ethercom *ec = &sc->sc_ethercom;
   2194 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
   2195 	struct ether_multi *enm;
   2196 	struct ether_multistep step;
   2197 	uint32_t hash, reg, bit;
   2198 	int i;
   2199 
   2200 	sc->sc_rctl &= ~(RCTL_BAM | RCTL_UPE | RCTL_MPE);
   2201 
   2202 	if (ifp->if_flags & IFF_BROADCAST)
   2203 		sc->sc_rctl |= RCTL_BAM;
   2204 	if (ifp->if_flags & IFF_PROMISC) {
   2205 		sc->sc_rctl |= RCTL_UPE;
   2206 		goto allmulti;
   2207 	}
   2208 
   2209 	/*
   2210 	 * Set the station address in the first RAL slot, and
   2211 	 * clear the remaining slots.
   2212 	 */
   2213 	dge_set_ral(sc, CLLADDR(ifp->if_sadl), 0);
   2214 	for (i = 1; i < RA_TABSIZE; i++)
   2215 		dge_set_ral(sc, NULL, i);
   2216 
   2217 	/* Clear out the multicast table. */
   2218 	for (i = 0; i < MC_TABSIZE; i++)
   2219 		CSR_WRITE(sc, DGE_MTA + (i << 2), 0);
   2220 
   2221 	ETHER_FIRST_MULTI(step, ec, enm);
   2222 	while (enm != NULL) {
   2223 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
   2224 			/*
   2225 			 * We must listen to a range of multicast addresses.
   2226 			 * For now, just accept all multicasts, rather than
   2227 			 * trying to set only those filter bits needed to match
   2228 			 * the range.  (At this time, the only use of address
   2229 			 * ranges is for IP multicast routing, for which the
   2230 			 * range is big enough to require all bits set.)
   2231 			 */
   2232 			goto allmulti;
   2233 		}
   2234 
   2235 		hash = dge_mchash(sc, enm->enm_addrlo);
   2236 
   2237 		reg = (hash >> 5) & 0x7f;
   2238 		bit = hash & 0x1f;
   2239 
   2240 		hash = CSR_READ(sc, DGE_MTA + (reg << 2));
   2241 		hash |= 1U << bit;
   2242 
   2243 		CSR_WRITE(sc, DGE_MTA + (reg << 2), hash);
   2244 
   2245 		ETHER_NEXT_MULTI(step, enm);
   2246 	}
   2247 
   2248 	ifp->if_flags &= ~IFF_ALLMULTI;
   2249 	goto setit;
   2250 
   2251  allmulti:
   2252 	ifp->if_flags |= IFF_ALLMULTI;
   2253 	sc->sc_rctl |= RCTL_MPE;
   2254 
   2255  setit:
   2256 	CSR_WRITE(sc, DGE_RCTL, sc->sc_rctl);
   2257 }
   2258 
   2259 /*
   2260  * Read in the EEPROM info and verify checksum.
   2261  */
   2262 int
   2263 dge_read_eeprom(struct dge_softc *sc)
   2264 {
   2265 	uint16_t cksum;
   2266 	int i;
   2267 
   2268 	cksum = 0;
   2269 	for (i = 0; i < EEPROM_SIZE; i++) {
   2270 		sc->sc_eeprom[i] = dge_eeprom_word(sc, i);
   2271 		cksum += sc->sc_eeprom[i];
   2272 	}
   2273 	return cksum != EEPROM_CKSUM;
   2274 }
   2275 
   2276 
   2277 /*
   2278  * Read a 16-bit word from address addr in the serial EEPROM.
   2279  */
   2280 uint16_t
   2281 dge_eeprom_word(struct dge_softc *sc, int addr)
   2282 {
   2283 	uint32_t reg;
   2284 	uint16_t rval = 0;
   2285 	int i;
   2286 
   2287 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_SK|EECD_DI|EECD_CS);
   2288 
   2289 	/* Lower clock pulse (and data in to chip) */
   2290 	CSR_WRITE(sc, DGE_EECD, reg);
   2291 	/* Select chip */
   2292 	CSR_WRITE(sc, DGE_EECD, reg|EECD_CS);
   2293 
   2294 	/* Send read command */
   2295 	dge_eeprom_clockout(sc, 1);
   2296 	dge_eeprom_clockout(sc, 1);
   2297 	dge_eeprom_clockout(sc, 0);
   2298 
   2299 	/* Send address */
   2300 	for (i = 5; i >= 0; i--)
   2301 		dge_eeprom_clockout(sc, (addr >> i) & 1);
   2302 
   2303 	/* Read data */
   2304 	for (i = 0; i < 16; i++) {
   2305 		rval <<= 1;
   2306 		rval |= dge_eeprom_clockin(sc);
   2307 	}
   2308 
   2309 	/* Deselect chip */
   2310 	CSR_WRITE(sc, DGE_EECD, reg);
   2311 
   2312 	return rval;
   2313 }
   2314 
   2315 /*
   2316  * Clock out a single bit to the EEPROM.
   2317  */
   2318 void
   2319 dge_eeprom_clockout(struct dge_softc *sc, int bit)
   2320 {
   2321 	int reg;
   2322 
   2323 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_SK);
   2324 	if (bit)
   2325 		reg |= EECD_DI;
   2326 
   2327 	CSR_WRITE(sc, DGE_EECD, reg);
   2328 	delay(2);
   2329 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK);
   2330 	delay(2);
   2331 	CSR_WRITE(sc, DGE_EECD, reg);
   2332 	delay(2);
   2333 }
   2334 
   2335 /*
   2336  * Clock in a single bit from EEPROM.
   2337  */
   2338 int
   2339 dge_eeprom_clockin(struct dge_softc *sc)
   2340 {
   2341 	int reg, rv;
   2342 
   2343 	reg = CSR_READ(sc, DGE_EECD) & ~(EECD_DI|EECD_DO|EECD_SK);
   2344 
   2345 	CSR_WRITE(sc, DGE_EECD, reg|EECD_SK); /* Raise clock */
   2346 	delay(2);
   2347 	rv = (CSR_READ(sc, DGE_EECD) & EECD_DO) != 0; /* Get bit */
   2348 	CSR_WRITE(sc, DGE_EECD, reg); /* Lower clock */
   2349 	delay(2);
   2350 
   2351 	return rv;
   2352 }
   2353 
   2354 static void
   2355 dge_xgmii_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
   2356 {
   2357 	struct dge_softc *sc = ifp->if_softc;
   2358 
   2359 	ifmr->ifm_status = IFM_AVALID;
   2360 	ifmr->ifm_active = IFM_ETHER|IFM_10G_LR;
   2361 
   2362 	if (CSR_READ(sc, DGE_STATUS) & STATUS_LINKUP)
   2363 		ifmr->ifm_status |= IFM_ACTIVE;
   2364 }
   2365 
   2366 static inline int
   2367 phwait(struct dge_softc *sc, int p, int r, int d, int type)
   2368 {
   2369         int i, mdic;
   2370 
   2371         CSR_WRITE(sc, DGE_MDIO,
   2372 	    MDIO_PHY(p) | MDIO_REG(r) | MDIO_DEV(d) | type | MDIO_CMD);
   2373         for (i = 0; i < 10; i++) {
   2374                 delay(10);
   2375                 if (((mdic = CSR_READ(sc, DGE_MDIO)) & MDIO_CMD) == 0)
   2376                         break;
   2377         }
   2378         return mdic;
   2379 }
   2380 
   2381 
   2382 static void
   2383 dge_xgmii_writereg(device_t self, int phy, int reg, int val)
   2384 {
   2385 	struct dge_softc *sc = device_private(self);
   2386 	int mdic;
   2387 
   2388 	CSR_WRITE(sc, DGE_MDIRW, val);
   2389 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_ADDR)) & MDIO_CMD)) {
   2390 		printf("%s: address cycle timeout; phy %d reg %d\n",
   2391 		    device_xname(&sc->sc_dev), phy, reg);
   2392 		return;
   2393 	}
   2394 	if (((mdic = phwait(sc, phy, reg, 1, MDIO_WRITE)) & MDIO_CMD)) {
   2395 		printf("%s: read cycle timeout; phy %d reg %d\n",
   2396 		    device_xname(&sc->sc_dev), phy, reg);
   2397 		return;
   2398 	}
   2399 }
   2400 
   2401 static void
   2402 dge_xgmii_reset(struct dge_softc *sc)
   2403 {
   2404 	dge_xgmii_writereg((void *)sc, 0, 0, BMCR_RESET);
   2405 }
   2406 
   2407 static int
   2408 dge_xgmii_mediachange(struct ifnet *ifp)
   2409 {
   2410 	return 0;
   2411 }
   2412