1/*-
2 * Copyright (c) 1997, 1998-2003
3 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by Bill Paul.
16 * 4. Neither the name of the author nor the names of any co-contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33#include <sys/cdefs.h>
34__FBSDID("$FreeBSD$");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38 *
39 * Written by Bill Paul <wpaul@windriver.com>
40 * Senior Networking Software Engineer
41 * Wind River Systems
42 */
43
44/*
45 * This driver is designed to support RealTek's next generation of
46 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49 *
50 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51 * with the older 8139 family, however it also supports a special
52 * C+ mode of operation that provides several new performance enhancing
53 * features. These include:
54 *
55 *	o Descriptor based DMA mechanism. Each descriptor represents
56 *	  a single packet fragment. Data buffers may be aligned on
57 *	  any byte boundary.
58 *
59 *	o 64-bit DMA
60 *
61 *	o TCP/IP checksum offload for both RX and TX
62 *
63 *	o High and normal priority transmit DMA rings
64 *
65 *	o VLAN tag insertion and extraction
66 *
67 *	o TCP large send (segmentation offload)
68 *
69 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70 * programming API is fairly straightforward. The RX filtering, EEPROM
71 * access and PHY access is the same as it is on the older 8139 series
72 * chips.
73 *
74 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75 * same programming API and feature set as the 8139C+ with the following
76 * differences and additions:
77 *
78 *	o 1000Mbps mode
79 *
80 *	o Jumbo frames
81 *
82 *	o GMII and TBI ports/registers for interfacing with copper
83 *	  or fiber PHYs
84 *
85 *	o RX and TX DMA rings can have up to 1024 descriptors
86 *	  (the 8139C+ allows a maximum of 64)
87 *
88 *	o Slight differences in register layout from the 8139C+
89 *
90 * The TX start and timer interrupt registers are at different locations
91 * on the 8169 than they are on the 8139C+. Also, the status word in the
92 * RX descriptor has a slightly different bit layout. The 8169 does not
93 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94 * copper gigE PHY.
95 *
96 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97 * (the 'S' stands for 'single-chip'). These devices have the same
98 * programming API as the older 8169, but also have some vendor-specific
99 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101 *
102 * This driver takes advantage of the RX and TX checksum offload and
103 * VLAN tag insertion/extraction features. It also implements TX
104 * interrupt moderation using the timer interrupt registers, which
105 * significantly reduces TX interrupt load. There is also support
106 * for jumbo frames, however the 8169/8169S/8110S can not transmit
107 * jumbo frames larger than 7440, so the max MTU possible with this
108 * driver is 7422 bytes.
109 */
110
111#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/sysctl.h>
127#include <sys/taskqueue.h>
128
129#include <net/if.h>
130#include <net/if_arp.h>
131#include <net/ethernet.h>
132#include <net/if_dl.h>
133#include <net/if_media.h>
134#include <net/if_types.h>
135#include <net/if_vlan_var.h>
136
137#include <net/bpf.h>
138
139#include <machine/bus.h>
140#include <machine/resource.h>
141#include <sys/bus.h>
142#include <sys/rman.h>
143
144#include <dev/mii/mii.h>
145#include <dev/mii/miivar.h>
146
147#include <dev/pci/pcireg.h>
148#include <dev/pci/pcivar.h>
149
150#include <pci/if_rlreg.h>
151
152MODULE_DEPEND(re, pci, 1, 1, 1);
153MODULE_DEPEND(re, ether, 1, 1, 1);
154MODULE_DEPEND(re, miibus, 1, 1, 1);
155
156/* "device miibus" required.  See GENERIC if you get errors here. */
157#include "miibus_if.h"
158
159/* Tunables. */
160static int intr_filter = 0;
161TUNABLE_INT("hw.re.intr_filter", &intr_filter);
162static int msi_disable = 0;
163TUNABLE_INT("hw.re.msi_disable", &msi_disable);
164static int msix_disable = 0;
165TUNABLE_INT("hw.re.msix_disable", &msix_disable);
166static int prefer_iomap = 0;
167TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
168
169#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
170
171/*
172 * Various supported device vendors/types and their names.
173 */
174static const struct rl_type const re_devs[] = {
175	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
178	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
179	{ RT_VENDORID, RT_DEVICEID_8139, 0,
180	    "RealTek 8139C+ 10/100BaseTX" },
181	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
182	    "RealTek 810xE PCIe 10/100baseTX" },
183	{ RT_VENDORID, RT_DEVICEID_8168, 0,
184	    "RealTek 8168/8111 B/C/CP/D/DP/E/F PCIe Gigabit Ethernet" },
185	{ RT_VENDORID, RT_DEVICEID_8169, 0,
186	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
187	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
188	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
189	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
190	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
192	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193	{ USR_VENDORID, USR_DEVICEID_997902, 0,
194	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
195};
196
197static const struct rl_hwrev const re_hwrevs[] = {
198	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
199	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
200	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
201	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
202	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
203	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
204	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
205	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
206	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
207	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
208	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
209	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
210	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
211	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
213	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
214	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
215	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
216	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
217	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
218	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
219	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
220	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
221	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
222	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
223	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
224	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
225	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
226	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
227	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
228	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
229	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
230	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
231	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
232	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
233	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
234	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
235	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
236	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
237	{ 0, 0, NULL, 0 }
238};
239
240static int re_probe		(device_t);
241static int re_attach		(device_t);
242static int re_detach		(device_t);
243
244static int re_encap		(struct rl_softc *, struct mbuf **);
245
246static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
247static int re_allocmem		(device_t, struct rl_softc *);
248static __inline void re_discard_rxbuf
249				(struct rl_softc *, int);
250static int re_newbuf		(struct rl_softc *, int);
251static int re_jumbo_newbuf	(struct rl_softc *, int);
252static int re_rx_list_init	(struct rl_softc *);
253static int re_jrx_list_init	(struct rl_softc *);
254static int re_tx_list_init	(struct rl_softc *);
255#ifdef RE_FIXUP_RX
256static __inline void re_fixup_rx
257				(struct mbuf *);
258#endif
259static int re_rxeof		(struct rl_softc *, int *);
260static void re_txeof		(struct rl_softc *);
261#ifdef DEVICE_POLLING
262static int re_poll		(struct ifnet *, enum poll_cmd, int);
263static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
264#endif
265static int re_intr		(void *);
266static void re_intr_msi		(void *);
267static void re_tick		(void *);
268static void re_int_task		(void *, int);
269static void re_start		(struct ifnet *);
270static void re_start_locked	(struct ifnet *);
271static int re_ioctl		(struct ifnet *, u_long, caddr_t);
272static void re_init		(void *);
273static void re_init_locked	(struct rl_softc *);
274static void re_stop		(struct rl_softc *);
275static void re_watchdog		(struct rl_softc *);
276static int re_suspend		(device_t);
277static int re_resume		(device_t);
278static int re_shutdown		(device_t);
279static int re_ifmedia_upd	(struct ifnet *);
280static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
281
282static void re_eeprom_putbyte	(struct rl_softc *, int);
283static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
284static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
285static int re_gmii_readreg	(device_t, int, int);
286static int re_gmii_writereg	(device_t, int, int, int);
287
288static int re_miibus_readreg	(device_t, int, int);
289static int re_miibus_writereg	(device_t, int, int, int);
290static void re_miibus_statchg	(device_t);
291
292static void re_set_jumbo	(struct rl_softc *, int);
293static void re_set_rxmode		(struct rl_softc *);
294static void re_reset		(struct rl_softc *);
295static void re_setwol		(struct rl_softc *);
296static void re_clrwol		(struct rl_softc *);
297static void re_set_linkspeed	(struct rl_softc *);
298
299#ifdef DEV_NETMAP	/* see ixgbe.c for details */
300#include <dev/netmap/if_re_netmap.h>
301#endif /* !DEV_NETMAP */
302
303#ifdef RE_DIAG
304static int re_diag		(struct rl_softc *);
305#endif
306
307static void re_add_sysctls	(struct rl_softc *);
308static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
309static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
310static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
311
312static device_method_t re_methods[] = {
313	/* Device interface */
314	DEVMETHOD(device_probe,		re_probe),
315	DEVMETHOD(device_attach,	re_attach),
316	DEVMETHOD(device_detach,	re_detach),
317	DEVMETHOD(device_suspend,	re_suspend),
318	DEVMETHOD(device_resume,	re_resume),
319	DEVMETHOD(device_shutdown,	re_shutdown),
320
321	/* MII interface */
322	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
323	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
324	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
325
326	DEVMETHOD_END
327};
328
329static driver_t re_driver = {
330	"re",
331	re_methods,
332	sizeof(struct rl_softc)
333};
334
335static devclass_t re_devclass;
336
337DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
338DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
339
340#define EE_SET(x)					\
341	CSR_WRITE_1(sc, RL_EECMD,			\
342		CSR_READ_1(sc, RL_EECMD) | x)
343
344#define EE_CLR(x)					\
345	CSR_WRITE_1(sc, RL_EECMD,			\
346		CSR_READ_1(sc, RL_EECMD) & ~x)
347
348/*
349 * Send a read command and address to the EEPROM, check for ACK.
350 */
351static void
352re_eeprom_putbyte(struct rl_softc *sc, int addr)
353{
354	int			d, i;
355
356	d = addr | (RL_9346_READ << sc->rl_eewidth);
357
358	/*
359	 * Feed in each bit and strobe the clock.
360	 */
361
362	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
363		if (d & i) {
364			EE_SET(RL_EE_DATAIN);
365		} else {
366			EE_CLR(RL_EE_DATAIN);
367		}
368		DELAY(100);
369		EE_SET(RL_EE_CLK);
370		DELAY(150);
371		EE_CLR(RL_EE_CLK);
372		DELAY(100);
373	}
374}
375
376/*
377 * Read a word of data stored in the EEPROM at address 'addr.'
378 */
379static void
380re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
381{
382	int			i;
383	u_int16_t		word = 0;
384
385	/*
386	 * Send address of word we want to read.
387	 */
388	re_eeprom_putbyte(sc, addr);
389
390	/*
391	 * Start reading bits from EEPROM.
392	 */
393	for (i = 0x8000; i; i >>= 1) {
394		EE_SET(RL_EE_CLK);
395		DELAY(100);
396		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
397			word |= i;
398		EE_CLR(RL_EE_CLK);
399		DELAY(100);
400	}
401
402	*dest = word;
403}
404
405/*
406 * Read a sequence of words from the EEPROM.
407 */
408static void
409re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
410{
411	int			i;
412	u_int16_t		word = 0, *ptr;
413
414	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
415
416        DELAY(100);
417
418	for (i = 0; i < cnt; i++) {
419		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
420		re_eeprom_getword(sc, off + i, &word);
421		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
422		ptr = (u_int16_t *)(dest + (i * 2));
423                *ptr = word;
424	}
425
426	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
427}
428
429static int
430re_gmii_readreg(device_t dev, int phy, int reg)
431{
432	struct rl_softc		*sc;
433	u_int32_t		rval;
434	int			i;
435
436	sc = device_get_softc(dev);
437
438	/* Let the rgephy driver read the GMEDIASTAT register */
439
440	if (reg == RL_GMEDIASTAT) {
441		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
442		return (rval);
443	}
444
445	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
446
447	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
448		rval = CSR_READ_4(sc, RL_PHYAR);
449		if (rval & RL_PHYAR_BUSY)
450			break;
451		DELAY(25);
452	}
453
454	if (i == RL_PHY_TIMEOUT) {
455		device_printf(sc->rl_dev, "PHY read failed\n");
456		return (0);
457	}
458
459	/*
460	 * Controller requires a 20us delay to process next MDIO request.
461	 */
462	DELAY(20);
463
464	return (rval & RL_PHYAR_PHYDATA);
465}
466
467static int
468re_gmii_writereg(device_t dev, int phy, int reg, int data)
469{
470	struct rl_softc		*sc;
471	u_int32_t		rval;
472	int			i;
473
474	sc = device_get_softc(dev);
475
476	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
477	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
478
479	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
480		rval = CSR_READ_4(sc, RL_PHYAR);
481		if (!(rval & RL_PHYAR_BUSY))
482			break;
483		DELAY(25);
484	}
485
486	if (i == RL_PHY_TIMEOUT) {
487		device_printf(sc->rl_dev, "PHY write failed\n");
488		return (0);
489	}
490
491	/*
492	 * Controller requires a 20us delay to process next MDIO request.
493	 */
494	DELAY(20);
495
496	return (0);
497}
498
499static int
500re_miibus_readreg(device_t dev, int phy, int reg)
501{
502	struct rl_softc		*sc;
503	u_int16_t		rval = 0;
504	u_int16_t		re8139_reg = 0;
505
506	sc = device_get_softc(dev);
507
508	if (sc->rl_type == RL_8169) {
509		rval = re_gmii_readreg(dev, phy, reg);
510		return (rval);
511	}
512
513	switch (reg) {
514	case MII_BMCR:
515		re8139_reg = RL_BMCR;
516		break;
517	case MII_BMSR:
518		re8139_reg = RL_BMSR;
519		break;
520	case MII_ANAR:
521		re8139_reg = RL_ANAR;
522		break;
523	case MII_ANER:
524		re8139_reg = RL_ANER;
525		break;
526	case MII_ANLPAR:
527		re8139_reg = RL_LPAR;
528		break;
529	case MII_PHYIDR1:
530	case MII_PHYIDR2:
531		return (0);
532	/*
533	 * Allow the rlphy driver to read the media status
534	 * register. If we have a link partner which does not
535	 * support NWAY, this is the register which will tell
536	 * us the results of parallel detection.
537	 */
538	case RL_MEDIASTAT:
539		rval = CSR_READ_1(sc, RL_MEDIASTAT);
540		return (rval);
541	default:
542		device_printf(sc->rl_dev, "bad phy register\n");
543		return (0);
544	}
545	rval = CSR_READ_2(sc, re8139_reg);
546	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
547		/* 8139C+ has different bit layout. */
548		rval &= ~(BMCR_LOOP | BMCR_ISO);
549	}
550	return (rval);
551}
552
553static int
554re_miibus_writereg(device_t dev, int phy, int reg, int data)
555{
556	struct rl_softc		*sc;
557	u_int16_t		re8139_reg = 0;
558	int			rval = 0;
559
560	sc = device_get_softc(dev);
561
562	if (sc->rl_type == RL_8169) {
563		rval = re_gmii_writereg(dev, phy, reg, data);
564		return (rval);
565	}
566
567	switch (reg) {
568	case MII_BMCR:
569		re8139_reg = RL_BMCR;
570		if (sc->rl_type == RL_8139CPLUS) {
571			/* 8139C+ has different bit layout. */
572			data &= ~(BMCR_LOOP | BMCR_ISO);
573		}
574		break;
575	case MII_BMSR:
576		re8139_reg = RL_BMSR;
577		break;
578	case MII_ANAR:
579		re8139_reg = RL_ANAR;
580		break;
581	case MII_ANER:
582		re8139_reg = RL_ANER;
583		break;
584	case MII_ANLPAR:
585		re8139_reg = RL_LPAR;
586		break;
587	case MII_PHYIDR1:
588	case MII_PHYIDR2:
589		return (0);
590		break;
591	default:
592		device_printf(sc->rl_dev, "bad phy register\n");
593		return (0);
594	}
595	CSR_WRITE_2(sc, re8139_reg, data);
596	return (0);
597}
598
599static void
600re_miibus_statchg(device_t dev)
601{
602	struct rl_softc		*sc;
603	struct ifnet		*ifp;
604	struct mii_data		*mii;
605
606	sc = device_get_softc(dev);
607	mii = device_get_softc(sc->rl_miibus);
608	ifp = sc->rl_ifp;
609	if (mii == NULL || ifp == NULL ||
610	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
611		return;
612
613	sc->rl_flags &= ~RL_FLAG_LINK;
614	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
615	    (IFM_ACTIVE | IFM_AVALID)) {
616		switch (IFM_SUBTYPE(mii->mii_media_active)) {
617		case IFM_10_T:
618		case IFM_100_TX:
619			sc->rl_flags |= RL_FLAG_LINK;
620			break;
621		case IFM_1000_T:
622			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
623				break;
624			sc->rl_flags |= RL_FLAG_LINK;
625			break;
626		default:
627			break;
628		}
629	}
630	/*
631	 * RealTek controllers does not provide any interface to
632	 * Tx/Rx MACs for resolved speed, duplex and flow-control
633	 * parameters.
634	 */
635}
636
637/*
638 * Set the RX configuration and 64-bit multicast hash filter.
639 */
640static void
641re_set_rxmode(struct rl_softc *sc)
642{
643	struct ifnet		*ifp;
644	struct ifmultiaddr	*ifma;
645	uint32_t		hashes[2] = { 0, 0 };
646	uint32_t		h, rxfilt;
647
648	RL_LOCK_ASSERT(sc);
649
650	ifp = sc->rl_ifp;
651
652	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
653
654	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
655		if (ifp->if_flags & IFF_PROMISC)
656			rxfilt |= RL_RXCFG_RX_ALLPHYS;
657		/*
658		 * Unlike other hardwares, we have to explicitly set
659		 * RL_RXCFG_RX_MULTI to receive multicast frames in
660		 * promiscuous mode.
661		 */
662		rxfilt |= RL_RXCFG_RX_MULTI;
663		hashes[0] = hashes[1] = 0xffffffff;
664		goto done;
665	}
666
667	if_maddr_rlock(ifp);
668	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
669		if (ifma->ifma_addr->sa_family != AF_LINK)
670			continue;
671		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
672		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
673		if (h < 32)
674			hashes[0] |= (1 << h);
675		else
676			hashes[1] |= (1 << (h - 32));
677	}
678	if_maddr_runlock(ifp);
679
680	if (hashes[0] != 0 || hashes[1] != 0) {
681		/*
682		 * For some unfathomable reason, RealTek decided to
683		 * reverse the order of the multicast hash registers
684		 * in the PCI Express parts.  This means we have to
685		 * write the hash pattern in reverse order for those
686		 * devices.
687		 */
688		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
689			h = bswap32(hashes[0]);
690			hashes[0] = bswap32(hashes[1]);
691			hashes[1] = h;
692		}
693		rxfilt |= RL_RXCFG_RX_MULTI;
694	}
695
696done:
697	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
698	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
699	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
700}
701
702static void
703re_reset(struct rl_softc *sc)
704{
705	int			i;
706
707	RL_LOCK_ASSERT(sc);
708
709	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
710
711	for (i = 0; i < RL_TIMEOUT; i++) {
712		DELAY(10);
713		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
714			break;
715	}
716	if (i == RL_TIMEOUT)
717		device_printf(sc->rl_dev, "reset never completed!\n");
718
719	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
720		CSR_WRITE_1(sc, 0x82, 1);
721	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
722		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
723}
724
725#ifdef RE_DIAG
726
727/*
728 * The following routine is designed to test for a defect on some
729 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
730 * lines connected to the bus, however for a 32-bit only card, they
731 * should be pulled high. The result of this defect is that the
732 * NIC will not work right if you plug it into a 64-bit slot: DMA
733 * operations will be done with 64-bit transfers, which will fail
734 * because the 64-bit data lines aren't connected.
735 *
736 * There's no way to work around this (short of talking a soldering
737 * iron to the board), however we can detect it. The method we use
738 * here is to put the NIC into digital loopback mode, set the receiver
739 * to promiscuous mode, and then try to send a frame. We then compare
740 * the frame data we sent to what was received. If the data matches,
741 * then the NIC is working correctly, otherwise we know the user has
742 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
743 * slot. In the latter case, there's no way the NIC can work correctly,
744 * so we print out a message on the console and abort the device attach.
745 */
746
747static int
748re_diag(struct rl_softc *sc)
749{
750	struct ifnet		*ifp = sc->rl_ifp;
751	struct mbuf		*m0;
752	struct ether_header	*eh;
753	struct rl_desc		*cur_rx;
754	u_int16_t		status;
755	u_int32_t		rxstat;
756	int			total_len, i, error = 0, phyaddr;
757	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
758	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
759
760	/* Allocate a single mbuf */
761	MGETHDR(m0, M_DONTWAIT, MT_DATA);
762	if (m0 == NULL)
763		return (ENOBUFS);
764
765	RL_LOCK(sc);
766
767	/*
768	 * Initialize the NIC in test mode. This sets the chip up
769	 * so that it can send and receive frames, but performs the
770	 * following special functions:
771	 * - Puts receiver in promiscuous mode
772	 * - Enables digital loopback mode
773	 * - Leaves interrupts turned off
774	 */
775
776	ifp->if_flags |= IFF_PROMISC;
777	sc->rl_testmode = 1;
778	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
779	re_init_locked(sc);
780	sc->rl_flags |= RL_FLAG_LINK;
781	if (sc->rl_type == RL_8169)
782		phyaddr = 1;
783	else
784		phyaddr = 0;
785
786	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
787	for (i = 0; i < RL_TIMEOUT; i++) {
788		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
789		if (!(status & BMCR_RESET))
790			break;
791	}
792
793	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
794	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
795
796	DELAY(100000);
797
798	/* Put some data in the mbuf */
799
800	eh = mtod(m0, struct ether_header *);
801	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
802	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
803	eh->ether_type = htons(ETHERTYPE_IP);
804	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
805
806	/*
807	 * Queue the packet, start transmission.
808	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
809	 */
810
811	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
812	RL_UNLOCK(sc);
813	/* XXX: re_diag must not be called when in ALTQ mode */
814	IF_HANDOFF(&ifp->if_snd, m0, ifp);
815	RL_LOCK(sc);
816	m0 = NULL;
817
818	/* Wait for it to propagate through the chip */
819
820	DELAY(100000);
821	for (i = 0; i < RL_TIMEOUT; i++) {
822		status = CSR_READ_2(sc, RL_ISR);
823		CSR_WRITE_2(sc, RL_ISR, status);
824		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
825		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
826			break;
827		DELAY(10);
828	}
829
830	if (i == RL_TIMEOUT) {
831		device_printf(sc->rl_dev,
832		    "diagnostic failed, failed to receive packet in"
833		    " loopback mode\n");
834		error = EIO;
835		goto done;
836	}
837
838	/*
839	 * The packet should have been dumped into the first
840	 * entry in the RX DMA ring. Grab it from there.
841	 */
842
843	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
844	    sc->rl_ldata.rl_rx_list_map,
845	    BUS_DMASYNC_POSTREAD);
846	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
847	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
848	    BUS_DMASYNC_POSTREAD);
849	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
850	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
851
852	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
853	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
854	eh = mtod(m0, struct ether_header *);
855
856	cur_rx = &sc->rl_ldata.rl_rx_list[0];
857	total_len = RL_RXBYTES(cur_rx);
858	rxstat = le32toh(cur_rx->rl_cmdstat);
859
860	if (total_len != ETHER_MIN_LEN) {
861		device_printf(sc->rl_dev,
862		    "diagnostic failed, received short packet\n");
863		error = EIO;
864		goto done;
865	}
866
867	/* Test that the received packet data matches what we sent. */
868
869	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
870	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
871	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
872		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
873		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
874		    dst, ":", src, ":", ETHERTYPE_IP);
875		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
876		    eh->ether_dhost, ":", eh->ether_shost, ":",
877		    ntohs(eh->ether_type));
878		device_printf(sc->rl_dev, "You may have a defective 32-bit "
879		    "NIC plugged into a 64-bit PCI slot.\n");
880		device_printf(sc->rl_dev, "Please re-install the NIC in a "
881		    "32-bit slot for proper operation.\n");
882		device_printf(sc->rl_dev, "Read the re(4) man page for more "
883		    "details.\n");
884		error = EIO;
885	}
886
887done:
888	/* Turn interface off, release resources */
889
890	sc->rl_testmode = 0;
891	sc->rl_flags &= ~RL_FLAG_LINK;
892	ifp->if_flags &= ~IFF_PROMISC;
893	re_stop(sc);
894	if (m0 != NULL)
895		m_freem(m0);
896
897	RL_UNLOCK(sc);
898
899	return (error);
900}
901
902#endif
903
904/*
905 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
906 * IDs against our list and return a device name if we find a match.
907 */
908static int
909re_probe(device_t dev)
910{
911	const struct rl_type	*t;
912	uint16_t		devid, vendor;
913	uint16_t		revid, sdevid;
914	int			i;
915
916	vendor = pci_get_vendor(dev);
917	devid = pci_get_device(dev);
918	revid = pci_get_revid(dev);
919	sdevid = pci_get_subdevice(dev);
920
921	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
922		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
923			/*
924			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
925			 * Rev. 2 is supported by sk(4).
926			 */
927			return (ENXIO);
928		}
929	}
930
931	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
932		if (revid != 0x20) {
933			/* 8139, let rl(4) take care of this device. */
934			return (ENXIO);
935		}
936	}
937
938	t = re_devs;
939	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
940		if (vendor == t->rl_vid && devid == t->rl_did) {
941			device_set_desc(dev, t->rl_name);
942			return (BUS_PROBE_DEFAULT);
943		}
944	}
945
946	return (ENXIO);
947}
948
949/*
950 * Map a single buffer address.
951 */
952
953static void
954re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
955{
956	bus_addr_t		*addr;
957
958	if (error)
959		return;
960
961	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
962	addr = arg;
963	*addr = segs->ds_addr;
964}
965
966static int
967re_allocmem(device_t dev, struct rl_softc *sc)
968{
969	bus_addr_t		lowaddr;
970	bus_size_t		rx_list_size, tx_list_size;
971	int			error;
972	int			i;
973
974	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
975	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
976
977	/*
978	 * Allocate the parent bus DMA tag appropriate for PCI.
979	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
980	 * register should be set. However some RealTek chips are known
981	 * to be buggy on DAC handling, therefore disable DAC by limiting
982	 * DMA address space to 32bit. PCIe variants of RealTek chips
983	 * may not have the limitation.
984	 */
985	lowaddr = BUS_SPACE_MAXADDR;
986	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
987		lowaddr = BUS_SPACE_MAXADDR_32BIT;
988	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
989	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
990	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
991	    NULL, NULL, &sc->rl_parent_tag);
992	if (error) {
993		device_printf(dev, "could not allocate parent DMA tag\n");
994		return (error);
995	}
996
997	/*
998	 * Allocate map for TX mbufs.
999	 */
1000	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1001	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1002	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1003	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1004	if (error) {
1005		device_printf(dev, "could not allocate TX DMA tag\n");
1006		return (error);
1007	}
1008
1009	/*
1010	 * Allocate map for RX mbufs.
1011	 */
1012
1013	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1014		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1015		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1016		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1017		    &sc->rl_ldata.rl_jrx_mtag);
1018		if (error) {
1019			device_printf(dev,
1020			    "could not allocate jumbo RX DMA tag\n");
1021			return (error);
1022		}
1023	}
1024	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1025	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1026	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1027	if (error) {
1028		device_printf(dev, "could not allocate RX DMA tag\n");
1029		return (error);
1030	}
1031
1032	/*
1033	 * Allocate map for TX descriptor list.
1034	 */
1035	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1036	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1037	    NULL, tx_list_size, 1, tx_list_size, 0,
1038	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1039	if (error) {
1040		device_printf(dev, "could not allocate TX DMA ring tag\n");
1041		return (error);
1042	}
1043
1044	/* Allocate DMA'able memory for the TX ring */
1045
1046	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1047	    (void **)&sc->rl_ldata.rl_tx_list,
1048	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1049	    &sc->rl_ldata.rl_tx_list_map);
1050	if (error) {
1051		device_printf(dev, "could not allocate TX DMA ring\n");
1052		return (error);
1053	}
1054
1055	/* Load the map for the TX ring. */
1056
1057	sc->rl_ldata.rl_tx_list_addr = 0;
1058	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1059	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1060	     tx_list_size, re_dma_map_addr,
1061	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1062	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1063		device_printf(dev, "could not load TX DMA ring\n");
1064		return (ENOMEM);
1065	}
1066
1067	/* Create DMA maps for TX buffers */
1068
1069	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1070		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1071		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1072		if (error) {
1073			device_printf(dev, "could not create DMA map for TX\n");
1074			return (error);
1075		}
1076	}
1077
1078	/*
1079	 * Allocate map for RX descriptor list.
1080	 */
1081	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1082	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1083	    NULL, rx_list_size, 1, rx_list_size, 0,
1084	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1085	if (error) {
1086		device_printf(dev, "could not create RX DMA ring tag\n");
1087		return (error);
1088	}
1089
1090	/* Allocate DMA'able memory for the RX ring */
1091
1092	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1093	    (void **)&sc->rl_ldata.rl_rx_list,
1094	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1095	    &sc->rl_ldata.rl_rx_list_map);
1096	if (error) {
1097		device_printf(dev, "could not allocate RX DMA ring\n");
1098		return (error);
1099	}
1100
1101	/* Load the map for the RX ring. */
1102
1103	sc->rl_ldata.rl_rx_list_addr = 0;
1104	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1105	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1106	     rx_list_size, re_dma_map_addr,
1107	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1108	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1109		device_printf(dev, "could not load RX DMA ring\n");
1110		return (ENOMEM);
1111	}
1112
1113	/* Create DMA maps for RX buffers */
1114
1115	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1116		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1117		    &sc->rl_ldata.rl_jrx_sparemap);
1118		if (error) {
1119			device_printf(dev,
1120			    "could not create spare DMA map for jumbo RX\n");
1121			return (error);
1122		}
1123		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1124			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1125			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1126			if (error) {
1127				device_printf(dev,
1128				    "could not create DMA map for jumbo RX\n");
1129				return (error);
1130			}
1131		}
1132	}
1133	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1134	    &sc->rl_ldata.rl_rx_sparemap);
1135	if (error) {
1136		device_printf(dev, "could not create spare DMA map for RX\n");
1137		return (error);
1138	}
1139	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1140		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1141		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1142		if (error) {
1143			device_printf(dev, "could not create DMA map for RX\n");
1144			return (error);
1145		}
1146	}
1147
1148	/* Create DMA map for statistics. */
1149	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1150	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1151	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1152	    &sc->rl_ldata.rl_stag);
1153	if (error) {
1154		device_printf(dev, "could not create statistics DMA tag\n");
1155		return (error);
1156	}
1157	/* Allocate DMA'able memory for statistics. */
1158	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1159	    (void **)&sc->rl_ldata.rl_stats,
1160	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1161	    &sc->rl_ldata.rl_smap);
1162	if (error) {
1163		device_printf(dev,
1164		    "could not allocate statistics DMA memory\n");
1165		return (error);
1166	}
1167	/* Load the map for statistics. */
1168	sc->rl_ldata.rl_stats_addr = 0;
1169	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1170	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1171	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1172	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1173		device_printf(dev, "could not load statistics DMA memory\n");
1174		return (ENOMEM);
1175	}
1176
1177	return (0);
1178}
1179
1180/*
1181 * Attach the interface. Allocate softc structures, do ifmedia
1182 * setup and ethernet/BPF attach.
1183 */
1184static int
1185re_attach(device_t dev)
1186{
1187	u_char			eaddr[ETHER_ADDR_LEN];
1188	u_int16_t		as[ETHER_ADDR_LEN / 2];
1189	struct rl_softc		*sc;
1190	struct ifnet		*ifp;
1191	const struct rl_hwrev	*hw_rev;
1192	u_int32_t		cap, ctl;
1193	int			hwrev;
1194	u_int16_t		devid, re_did = 0;
1195	int			error = 0, i, phy, rid;
1196	int			msic, msixc, reg;
1197	uint8_t			cfg;
1198
1199	sc = device_get_softc(dev);
1200	sc->rl_dev = dev;
1201
1202	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1203	    MTX_DEF);
1204	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1205
1206	/*
1207	 * Map control/status registers.
1208	 */
1209	pci_enable_busmaster(dev);
1210
1211	devid = pci_get_device(dev);
1212	/*
1213	 * Prefer memory space register mapping over IO space.
1214	 * Because RTL8169SC does not seem to work when memory mapping
1215	 * is used always activate io mapping.
1216	 */
1217	if (devid == RT_DEVICEID_8169SC)
1218		prefer_iomap = 1;
1219	if (prefer_iomap == 0) {
1220		sc->rl_res_id = PCIR_BAR(1);
1221		sc->rl_res_type = SYS_RES_MEMORY;
1222		/* RTL8168/8101E seems to use different BARs. */
1223		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1224			sc->rl_res_id = PCIR_BAR(2);
1225	} else {
1226		sc->rl_res_id = PCIR_BAR(0);
1227		sc->rl_res_type = SYS_RES_IOPORT;
1228	}
1229	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1230	    &sc->rl_res_id, RF_ACTIVE);
1231	if (sc->rl_res == NULL && prefer_iomap == 0) {
1232		sc->rl_res_id = PCIR_BAR(0);
1233		sc->rl_res_type = SYS_RES_IOPORT;
1234		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1235		    &sc->rl_res_id, RF_ACTIVE);
1236	}
1237	if (sc->rl_res == NULL) {
1238		device_printf(dev, "couldn't map ports/memory\n");
1239		error = ENXIO;
1240		goto fail;
1241	}
1242
1243	sc->rl_btag = rman_get_bustag(sc->rl_res);
1244	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1245
1246	msic = pci_msi_count(dev);
1247	msixc = pci_msix_count(dev);
1248	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1249		sc->rl_flags |= RL_FLAG_PCIE;
1250		sc->rl_expcap = reg;
1251	}
1252	if (bootverbose) {
1253		device_printf(dev, "MSI count : %d\n", msic);
1254		device_printf(dev, "MSI-X count : %d\n", msixc);
1255	}
1256	if (msix_disable > 0)
1257		msixc = 0;
1258	if (msi_disable > 0)
1259		msic = 0;
1260	/* Prefer MSI-X to MSI. */
1261	if (msixc > 0) {
1262		msixc = 1;
1263		rid = PCIR_BAR(4);
1264		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1265		    &rid, RF_ACTIVE);
1266		if (sc->rl_res_pba == NULL) {
1267			device_printf(sc->rl_dev,
1268			    "could not allocate MSI-X PBA resource\n");
1269		}
1270		if (sc->rl_res_pba != NULL &&
1271		    pci_alloc_msix(dev, &msixc) == 0) {
1272			if (msixc == 1) {
1273				device_printf(dev, "Using %d MSI-X message\n",
1274				    msixc);
1275				sc->rl_flags |= RL_FLAG_MSIX;
1276			} else
1277				pci_release_msi(dev);
1278		}
1279		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1280			if (sc->rl_res_pba != NULL)
1281				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1282				    sc->rl_res_pba);
1283			sc->rl_res_pba = NULL;
1284			msixc = 0;
1285		}
1286	}
1287	/* Prefer MSI to INTx. */
1288	if (msixc == 0 && msic > 0) {
1289		msic = 1;
1290		if (pci_alloc_msi(dev, &msic) == 0) {
1291			if (msic == RL_MSI_MESSAGES) {
1292				device_printf(dev, "Using %d MSI message\n",
1293				    msic);
1294				sc->rl_flags |= RL_FLAG_MSI;
1295				/* Explicitly set MSI enable bit. */
1296				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1297				cfg = CSR_READ_1(sc, RL_CFG2);
1298				cfg |= RL_CFG2_MSI;
1299				CSR_WRITE_1(sc, RL_CFG2, cfg);
1300				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1301			} else
1302				pci_release_msi(dev);
1303		}
1304		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1305			msic = 0;
1306	}
1307
1308	/* Allocate interrupt */
1309	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1310		rid = 0;
1311		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1312		    RF_SHAREABLE | RF_ACTIVE);
1313		if (sc->rl_irq[0] == NULL) {
1314			device_printf(dev, "couldn't allocate IRQ resources\n");
1315			error = ENXIO;
1316			goto fail;
1317		}
1318	} else {
1319		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1320			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1321			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1322			if (sc->rl_irq[i] == NULL) {
1323				device_printf(dev,
1324				    "couldn't llocate IRQ resources for "
1325				    "message %d\n", rid);
1326				error = ENXIO;
1327				goto fail;
1328			}
1329		}
1330	}
1331
1332	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1333		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1334		cfg = CSR_READ_1(sc, RL_CFG2);
1335		if ((cfg & RL_CFG2_MSI) != 0) {
1336			device_printf(dev, "turning off MSI enable bit.\n");
1337			cfg &= ~RL_CFG2_MSI;
1338			CSR_WRITE_1(sc, RL_CFG2, cfg);
1339		}
1340		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1341	}
1342
1343	/* Disable ASPM L0S/L1. */
1344	if (sc->rl_expcap != 0) {
1345		cap = pci_read_config(dev, sc->rl_expcap +
1346		    PCIR_EXPRESS_LINK_CAP, 2);
1347		if ((cap & PCIM_LINK_CAP_ASPM) != 0) {
1348			ctl = pci_read_config(dev, sc->rl_expcap +
1349			    PCIR_EXPRESS_LINK_CTL, 2);
1350			if ((ctl & 0x0003) != 0) {
1351				ctl &= ~0x0003;
1352				pci_write_config(dev, sc->rl_expcap +
1353				    PCIR_EXPRESS_LINK_CTL, ctl, 2);
1354				device_printf(dev, "ASPM disabled\n");
1355			}
1356		} else
1357			device_printf(dev, "no ASPM capability\n");
1358	}
1359
1360	hw_rev = re_hwrevs;
1361	hwrev = CSR_READ_4(sc, RL_TXCFG);
1362	switch (hwrev & 0x70000000) {
1363	case 0x00000000:
1364	case 0x10000000:
1365		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1366		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1367		break;
1368	default:
1369		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1370		hwrev &= RL_TXCFG_HWREV;
1371		break;
1372	}
1373	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1374	while (hw_rev->rl_desc != NULL) {
1375		if (hw_rev->rl_rev == hwrev) {
1376			sc->rl_type = hw_rev->rl_type;
1377			sc->rl_hwrev = hw_rev;
1378			break;
1379		}
1380		hw_rev++;
1381	}
1382	if (hw_rev->rl_desc == NULL) {
1383		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1384		error = ENXIO;
1385		goto fail;
1386	}
1387
1388	switch (hw_rev->rl_rev) {
1389	case RL_HWREV_8139CPLUS:
1390		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1391		break;
1392	case RL_HWREV_8100E:
1393	case RL_HWREV_8101E:
1394		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1395		break;
1396	case RL_HWREV_8102E:
1397	case RL_HWREV_8102EL:
1398	case RL_HWREV_8102EL_SPIN1:
1399		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1400		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1401		    RL_FLAG_AUTOPAD;
1402		break;
1403	case RL_HWREV_8103E:
1404		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1405		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1406		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1407		break;
1408	case RL_HWREV_8401E:
1409	case RL_HWREV_8105E:
1410	case RL_HWREV_8105E_SPIN1:
1411		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1412		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1413		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1414		break;
1415	case RL_HWREV_8402:
1416		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1417		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1418		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1419		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1420		break;
1421	case RL_HWREV_8168B_SPIN1:
1422	case RL_HWREV_8168B_SPIN2:
1423		sc->rl_flags |= RL_FLAG_WOLRXENB;
1424		/* FALLTHROUGH */
1425	case RL_HWREV_8168B_SPIN3:
1426		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1427		break;
1428	case RL_HWREV_8168C_SPIN2:
1429		sc->rl_flags |= RL_FLAG_MACSLEEP;
1430		/* FALLTHROUGH */
1431	case RL_HWREV_8168C:
1432		if ((hwrev & 0x00700000) == 0x00200000)
1433			sc->rl_flags |= RL_FLAG_MACSLEEP;
1434		/* FALLTHROUGH */
1435	case RL_HWREV_8168CP:
1436		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1437		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1438		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1439		break;
1440	case RL_HWREV_8168D:
1441		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1442		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1443		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1444		    RL_FLAG_WOL_MANLINK;
1445		break;
1446	case RL_HWREV_8168DP:
1447		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1448		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1449		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1450		break;
1451	case RL_HWREV_8168E:
1452		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1453		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1454		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1455		    RL_FLAG_WOL_MANLINK;
1456		break;
1457	case RL_HWREV_8168E_VL:
1458	case RL_HWREV_8168F:
1459	case RL_HWREV_8411:
1460		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1461		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1462		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1463		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1464		break;
1465	case RL_HWREV_8169_8110SB:
1466	case RL_HWREV_8169_8110SBL:
1467	case RL_HWREV_8169_8110SC:
1468	case RL_HWREV_8169_8110SCE:
1469		sc->rl_flags |= RL_FLAG_PHYWAKE;
1470		/* FALLTHROUGH */
1471	case RL_HWREV_8169:
1472	case RL_HWREV_8169S:
1473	case RL_HWREV_8110S:
1474		sc->rl_flags |= RL_FLAG_MACRESET;
1475		break;
1476	default:
1477		break;
1478	}
1479
1480	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1481		sc->rl_cfg0 = RL_8139_CFG0;
1482		sc->rl_cfg1 = RL_8139_CFG1;
1483		sc->rl_cfg2 = 0;
1484		sc->rl_cfg3 = RL_8139_CFG3;
1485		sc->rl_cfg4 = RL_8139_CFG4;
1486		sc->rl_cfg5 = RL_8139_CFG5;
1487	} else {
1488		sc->rl_cfg0 = RL_CFG0;
1489		sc->rl_cfg1 = RL_CFG1;
1490		sc->rl_cfg2 = RL_CFG2;
1491		sc->rl_cfg3 = RL_CFG3;
1492		sc->rl_cfg4 = RL_CFG4;
1493		sc->rl_cfg5 = RL_CFG5;
1494	}
1495
1496	/* Reset the adapter. */
1497	RL_LOCK(sc);
1498	re_reset(sc);
1499	RL_UNLOCK(sc);
1500
1501	/* Enable PME. */
1502	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1503	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1504	cfg |= RL_CFG1_PME;
1505	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1506	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1507	cfg &= RL_CFG5_PME_STS;
1508	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1509	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1510
1511	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1512		/*
1513		 * XXX Should have a better way to extract station
1514		 * address from EEPROM.
1515		 */
1516		for (i = 0; i < ETHER_ADDR_LEN; i++)
1517			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1518	} else {
1519		sc->rl_eewidth = RL_9356_ADDR_LEN;
1520		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1521		if (re_did != 0x8129)
1522			sc->rl_eewidth = RL_9346_ADDR_LEN;
1523
1524		/*
1525		 * Get station address from the EEPROM.
1526		 */
1527		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1528		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1529			as[i] = le16toh(as[i]);
1530		bcopy(as, eaddr, ETHER_ADDR_LEN);
1531	}
1532
1533	if (sc->rl_type == RL_8169) {
1534		/* Set RX length mask and number of descriptors. */
1535		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1536		sc->rl_txstart = RL_GTXSTART;
1537		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1538		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1539	} else {
1540		/* Set RX length mask and number of descriptors. */
1541		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1542		sc->rl_txstart = RL_TXSTART;
1543		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1544		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1545	}
1546
1547	error = re_allocmem(dev, sc);
1548	if (error)
1549		goto fail;
1550	re_add_sysctls(sc);
1551
1552	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1553	if (ifp == NULL) {
1554		device_printf(dev, "can not if_alloc()\n");
1555		error = ENOSPC;
1556		goto fail;
1557	}
1558
1559	/* Take controller out of deep sleep mode. */
1560	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1561		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1562			CSR_WRITE_1(sc, RL_GPIO,
1563			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1564		else
1565			CSR_WRITE_1(sc, RL_GPIO,
1566			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1567	}
1568
1569	/* Take PHY out of power down mode. */
1570	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1571		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1572		if (hw_rev->rl_rev == RL_HWREV_8401E)
1573			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1574	}
1575	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1576		re_gmii_writereg(dev, 1, 0x1f, 0);
1577		re_gmii_writereg(dev, 1, 0x0e, 0);
1578	}
1579
1580	ifp->if_softc = sc;
1581	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1582	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1583	ifp->if_ioctl = re_ioctl;
1584	ifp->if_start = re_start;
1585	/*
1586	 * RTL8168/8111C generates wrong IP checksummed frame if the
1587	 * packet has IP options so disable TX IP checksum offloading.
1588	 */
1589	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1590	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
1591		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1592	else
1593		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1594	ifp->if_hwassist |= CSUM_TSO;
1595	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1596	ifp->if_capenable = ifp->if_capabilities;
1597	ifp->if_init = re_init;
1598	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1599	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1600	IFQ_SET_READY(&ifp->if_snd);
1601
1602	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1603
1604#define	RE_PHYAD_INTERNAL	 0
1605
1606	/* Do MII setup. */
1607	phy = RE_PHYAD_INTERNAL;
1608	if (sc->rl_type == RL_8169)
1609		phy = 1;
1610	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1611	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1612	if (error != 0) {
1613		device_printf(dev, "attaching PHYs failed\n");
1614		goto fail;
1615	}
1616
1617	/*
1618	 * Call MI attach routine.
1619	 */
1620	ether_ifattach(ifp, eaddr);
1621
1622	/* VLAN capability setup */
1623	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1624	if (ifp->if_capabilities & IFCAP_HWCSUM)
1625		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1626	/* Enable WOL if PM is supported. */
1627	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1628		ifp->if_capabilities |= IFCAP_WOL;
1629	ifp->if_capenable = ifp->if_capabilities;
1630	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1631	/*
1632	 * Don't enable TSO by default.  It is known to generate
1633	 * corrupted TCP segments(bad TCP options) under certain
1634	 * circumtances.
1635	 */
1636	ifp->if_hwassist &= ~CSUM_TSO;
1637	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1638#ifdef DEVICE_POLLING
1639	ifp->if_capabilities |= IFCAP_POLLING;
1640#endif
1641	/*
1642	 * Tell the upper layer(s) we support long frames.
1643	 * Must appear after the call to ether_ifattach() because
1644	 * ether_ifattach() sets ifi_hdrlen to the default value.
1645	 */
1646	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1647
1648#ifdef DEV_NETMAP
1649	re_netmap_attach(sc);
1650#endif /* DEV_NETMAP */
1651#ifdef RE_DIAG
1652	/*
1653	 * Perform hardware diagnostic on the original RTL8169.
1654	 * Some 32-bit cards were incorrectly wired and would
1655	 * malfunction if plugged into a 64-bit slot.
1656	 */
1657
1658	if (hwrev == RL_HWREV_8169) {
1659		error = re_diag(sc);
1660		if (error) {
1661			device_printf(dev,
1662		    	"attach aborted due to hardware diag failure\n");
1663			ether_ifdetach(ifp);
1664			goto fail;
1665		}
1666	}
1667#endif
1668
1669#ifdef RE_TX_MODERATION
1670	intr_filter = 1;
1671#endif
1672	/* Hook interrupt last to avoid having to lock softc */
1673	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1674	    intr_filter == 0) {
1675		error = bus_setup_intr(dev, sc->rl_irq[0],
1676		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1677		    &sc->rl_intrhand[0]);
1678	} else {
1679		error = bus_setup_intr(dev, sc->rl_irq[0],
1680		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1681		    &sc->rl_intrhand[0]);
1682	}
1683	if (error) {
1684		device_printf(dev, "couldn't set up irq\n");
1685		ether_ifdetach(ifp);
1686	}
1687
1688fail:
1689
1690	if (error)
1691		re_detach(dev);
1692
1693	return (error);
1694}
1695
1696/*
1697 * Shutdown hardware and free up resources. This can be called any
1698 * time after the mutex has been initialized. It is called in both
1699 * the error case in attach and the normal detach case so it needs
1700 * to be careful about only freeing resources that have actually been
1701 * allocated.
1702 */
1703static int
1704re_detach(device_t dev)
1705{
1706	struct rl_softc		*sc;
1707	struct ifnet		*ifp;
1708	int			i, rid;
1709
1710	sc = device_get_softc(dev);
1711	ifp = sc->rl_ifp;
1712	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1713
1714	/* These should only be active if attach succeeded */
1715	if (device_is_attached(dev)) {
1716#ifdef DEVICE_POLLING
1717		if (ifp->if_capenable & IFCAP_POLLING)
1718			ether_poll_deregister(ifp);
1719#endif
1720		RL_LOCK(sc);
1721#if 0
1722		sc->suspended = 1;
1723#endif
1724		re_stop(sc);
1725		RL_UNLOCK(sc);
1726		callout_drain(&sc->rl_stat_callout);
1727		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1728		/*
1729		 * Force off the IFF_UP flag here, in case someone
1730		 * still had a BPF descriptor attached to this
1731		 * interface. If they do, ether_ifdetach() will cause
1732		 * the BPF code to try and clear the promisc mode
1733		 * flag, which will bubble down to re_ioctl(),
1734		 * which will try to call re_init() again. This will
1735		 * turn the NIC back on and restart the MII ticker,
1736		 * which will panic the system when the kernel tries
1737		 * to invoke the re_tick() function that isn't there
1738		 * anymore.
1739		 */
1740		ifp->if_flags &= ~IFF_UP;
1741		ether_ifdetach(ifp);
1742	}
1743	if (sc->rl_miibus)
1744		device_delete_child(dev, sc->rl_miibus);
1745	bus_generic_detach(dev);
1746
1747	/*
1748	 * The rest is resource deallocation, so we should already be
1749	 * stopped here.
1750	 */
1751
1752	if (sc->rl_intrhand[0] != NULL) {
1753		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1754		sc->rl_intrhand[0] = NULL;
1755	}
1756	if (ifp != NULL)
1757		if_free(ifp);
1758	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1759		rid = 0;
1760	else
1761		rid = 1;
1762	if (sc->rl_irq[0] != NULL) {
1763		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1764		sc->rl_irq[0] = NULL;
1765	}
1766	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1767		pci_release_msi(dev);
1768	if (sc->rl_res_pba) {
1769		rid = PCIR_BAR(4);
1770		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1771	}
1772	if (sc->rl_res)
1773		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1774		    sc->rl_res);
1775
1776	/* Unload and free the RX DMA ring memory and map */
1777
1778	if (sc->rl_ldata.rl_rx_list_tag) {
1779		if (sc->rl_ldata.rl_rx_list_map)
1780			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1781			    sc->rl_ldata.rl_rx_list_map);
1782		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1783			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1784			    sc->rl_ldata.rl_rx_list,
1785			    sc->rl_ldata.rl_rx_list_map);
1786		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1787	}
1788
1789	/* Unload and free the TX DMA ring memory and map */
1790
1791	if (sc->rl_ldata.rl_tx_list_tag) {
1792		if (sc->rl_ldata.rl_tx_list_map)
1793			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1794			    sc->rl_ldata.rl_tx_list_map);
1795		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1796			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1797			    sc->rl_ldata.rl_tx_list,
1798			    sc->rl_ldata.rl_tx_list_map);
1799		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1800	}
1801
1802	/* Destroy all the RX and TX buffer maps */
1803
1804	if (sc->rl_ldata.rl_tx_mtag) {
1805		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1806			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1807				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1808				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1809		}
1810		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1811	}
1812	if (sc->rl_ldata.rl_rx_mtag) {
1813		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1814			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1815				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1816				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1817		}
1818		if (sc->rl_ldata.rl_rx_sparemap)
1819			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1820			    sc->rl_ldata.rl_rx_sparemap);
1821		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1822	}
1823	if (sc->rl_ldata.rl_jrx_mtag) {
1824		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1825			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1826				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1827				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1828		}
1829		if (sc->rl_ldata.rl_jrx_sparemap)
1830			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1831			    sc->rl_ldata.rl_jrx_sparemap);
1832		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1833	}
1834	/* Unload and free the stats buffer and map */
1835
1836	if (sc->rl_ldata.rl_stag) {
1837		if (sc->rl_ldata.rl_smap)
1838			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1839			    sc->rl_ldata.rl_smap);
1840		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1841			bus_dmamem_free(sc->rl_ldata.rl_stag,
1842			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1843		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1844	}
1845
1846#ifdef DEV_NETMAP
1847	netmap_detach(ifp);
1848#endif /* DEV_NETMAP */
1849	if (sc->rl_parent_tag)
1850		bus_dma_tag_destroy(sc->rl_parent_tag);
1851
1852	mtx_destroy(&sc->rl_mtx);
1853
1854	return (0);
1855}
1856
1857static __inline void
1858re_discard_rxbuf(struct rl_softc *sc, int idx)
1859{
1860	struct rl_desc		*desc;
1861	struct rl_rxdesc	*rxd;
1862	uint32_t		cmdstat;
1863
1864	if (sc->rl_ifp->if_mtu > RL_MTU &&
1865	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1866		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1867	else
1868		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1869	desc = &sc->rl_ldata.rl_rx_list[idx];
1870	desc->rl_vlanctl = 0;
1871	cmdstat = rxd->rx_size;
1872	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1873		cmdstat |= RL_RDESC_CMD_EOR;
1874	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1875}
1876
1877static int
1878re_newbuf(struct rl_softc *sc, int idx)
1879{
1880	struct mbuf		*m;
1881	struct rl_rxdesc	*rxd;
1882	bus_dma_segment_t	segs[1];
1883	bus_dmamap_t		map;
1884	struct rl_desc		*desc;
1885	uint32_t		cmdstat;
1886	int			error, nsegs;
1887
1888	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1889	if (m == NULL)
1890		return (ENOBUFS);
1891
1892	m->m_len = m->m_pkthdr.len = MCLBYTES;
1893#ifdef RE_FIXUP_RX
1894	/*
1895	 * This is part of an evil trick to deal with non-x86 platforms.
1896	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1897	 * boundaries, but that will hose non-x86 machines. To get around
1898	 * this, we leave some empty space at the start of each buffer
1899	 * and for non-x86 hosts, we copy the buffer back six bytes
1900	 * to achieve word alignment. This is slightly more efficient
1901	 * than allocating a new buffer, copying the contents, and
1902	 * discarding the old buffer.
1903	 */
1904	m_adj(m, RE_ETHER_ALIGN);
1905#endif
1906	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1907	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1908	if (error != 0) {
1909		m_freem(m);
1910		return (ENOBUFS);
1911	}
1912	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1913
1914	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1915	if (rxd->rx_m != NULL) {
1916		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1917		    BUS_DMASYNC_POSTREAD);
1918		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1919	}
1920
1921	rxd->rx_m = m;
1922	map = rxd->rx_dmamap;
1923	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1924	rxd->rx_size = segs[0].ds_len;
1925	sc->rl_ldata.rl_rx_sparemap = map;
1926	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1927	    BUS_DMASYNC_PREREAD);
1928
1929	desc = &sc->rl_ldata.rl_rx_list[idx];
1930	desc->rl_vlanctl = 0;
1931	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1932	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1933	cmdstat = segs[0].ds_len;
1934	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1935		cmdstat |= RL_RDESC_CMD_EOR;
1936	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1937
1938	return (0);
1939}
1940
1941static int
1942re_jumbo_newbuf(struct rl_softc *sc, int idx)
1943{
1944	struct mbuf		*m;
1945	struct rl_rxdesc	*rxd;
1946	bus_dma_segment_t	segs[1];
1947	bus_dmamap_t		map;
1948	struct rl_desc		*desc;
1949	uint32_t		cmdstat;
1950	int			error, nsegs;
1951
1952	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1953	if (m == NULL)
1954		return (ENOBUFS);
1955	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1956#ifdef RE_FIXUP_RX
1957	m_adj(m, RE_ETHER_ALIGN);
1958#endif
1959	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1960	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1961	if (error != 0) {
1962		m_freem(m);
1963		return (ENOBUFS);
1964	}
1965	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1966
1967	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1968	if (rxd->rx_m != NULL) {
1969		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1970		    BUS_DMASYNC_POSTREAD);
1971		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1972	}
1973
1974	rxd->rx_m = m;
1975	map = rxd->rx_dmamap;
1976	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1977	rxd->rx_size = segs[0].ds_len;
1978	sc->rl_ldata.rl_jrx_sparemap = map;
1979	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1980	    BUS_DMASYNC_PREREAD);
1981
1982	desc = &sc->rl_ldata.rl_rx_list[idx];
1983	desc->rl_vlanctl = 0;
1984	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1985	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1986	cmdstat = segs[0].ds_len;
1987	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1988		cmdstat |= RL_RDESC_CMD_EOR;
1989	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1990
1991	return (0);
1992}
1993
1994#ifdef RE_FIXUP_RX
1995static __inline void
1996re_fixup_rx(struct mbuf *m)
1997{
1998	int                     i;
1999	uint16_t                *src, *dst;
2000
2001	src = mtod(m, uint16_t *);
2002	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2003
2004	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2005		*dst++ = *src++;
2006
2007	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2008}
2009#endif
2010
2011static int
2012re_tx_list_init(struct rl_softc *sc)
2013{
2014	struct rl_desc		*desc;
2015	int			i;
2016
2017	RL_LOCK_ASSERT(sc);
2018
2019	bzero(sc->rl_ldata.rl_tx_list,
2020	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2021	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2022		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2023#ifdef DEV_NETMAP
2024	re_netmap_tx_init(sc);
2025#endif /* DEV_NETMAP */
2026	/* Set EOR. */
2027	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2028	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2029
2030	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2031	    sc->rl_ldata.rl_tx_list_map,
2032	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2033
2034	sc->rl_ldata.rl_tx_prodidx = 0;
2035	sc->rl_ldata.rl_tx_considx = 0;
2036	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2037
2038	return (0);
2039}
2040
2041static int
2042re_rx_list_init(struct rl_softc *sc)
2043{
2044	int			error, i;
2045
2046	bzero(sc->rl_ldata.rl_rx_list,
2047	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2048	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2049		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2050		if ((error = re_newbuf(sc, i)) != 0)
2051			return (error);
2052	}
2053#ifdef DEV_NETMAP
2054	re_netmap_rx_init(sc);
2055#endif /* DEV_NETMAP */
2056
2057	/* Flush the RX descriptors */
2058
2059	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2060	    sc->rl_ldata.rl_rx_list_map,
2061	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2062
2063	sc->rl_ldata.rl_rx_prodidx = 0;
2064	sc->rl_head = sc->rl_tail = NULL;
2065	sc->rl_int_rx_act = 0;
2066
2067	return (0);
2068}
2069
2070static int
2071re_jrx_list_init(struct rl_softc *sc)
2072{
2073	int			error, i;
2074
2075	bzero(sc->rl_ldata.rl_rx_list,
2076	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2077	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2078		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2079		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2080			return (error);
2081	}
2082
2083	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2084	    sc->rl_ldata.rl_rx_list_map,
2085	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2086
2087	sc->rl_ldata.rl_rx_prodidx = 0;
2088	sc->rl_head = sc->rl_tail = NULL;
2089	sc->rl_int_rx_act = 0;
2090
2091	return (0);
2092}
2093
2094/*
2095 * RX handler for C+ and 8169. For the gigE chips, we support
2096 * the reception of jumbo frames that have been fragmented
2097 * across multiple 2K mbuf cluster buffers.
2098 */
2099static int
2100re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2101{
2102	struct mbuf		*m;
2103	struct ifnet		*ifp;
2104	int			i, rxerr, total_len;
2105	struct rl_desc		*cur_rx;
2106	u_int32_t		rxstat, rxvlan;
2107	int			jumbo, maxpkt = 16, rx_npkts = 0;
2108
2109	RL_LOCK_ASSERT(sc);
2110
2111	ifp = sc->rl_ifp;
2112#ifdef DEV_NETMAP
2113	if (ifp->if_capenable & IFCAP_NETMAP) {
2114		NA(ifp)->rx_rings->nr_kflags |= NKR_PENDINTR;
2115		selwakeuppri(&NA(ifp)->rx_rings->si, PI_NET);
2116		return 0;
2117	}
2118#endif /* DEV_NETMAP */
2119	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2120		jumbo = 1;
2121	else
2122		jumbo = 0;
2123
2124	/* Invalidate the descriptor memory */
2125
2126	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2127	    sc->rl_ldata.rl_rx_list_map,
2128	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2129
2130	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2131	    i = RL_RX_DESC_NXT(sc, i)) {
2132		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2133			break;
2134		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2135		rxstat = le32toh(cur_rx->rl_cmdstat);
2136		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2137			break;
2138		total_len = rxstat & sc->rl_rxlenmask;
2139		rxvlan = le32toh(cur_rx->rl_vlanctl);
2140		if (jumbo != 0)
2141			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2142		else
2143			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2144
2145		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2146		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2147		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2148			/*
2149			 * RTL8168C or later controllers do not
2150			 * support multi-fragment packet.
2151			 */
2152			re_discard_rxbuf(sc, i);
2153			continue;
2154		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2155			if (re_newbuf(sc, i) != 0) {
2156				/*
2157				 * If this is part of a multi-fragment packet,
2158				 * discard all the pieces.
2159				 */
2160				if (sc->rl_head != NULL) {
2161					m_freem(sc->rl_head);
2162					sc->rl_head = sc->rl_tail = NULL;
2163				}
2164				re_discard_rxbuf(sc, i);
2165				continue;
2166			}
2167			m->m_len = RE_RX_DESC_BUFLEN;
2168			if (sc->rl_head == NULL)
2169				sc->rl_head = sc->rl_tail = m;
2170			else {
2171				m->m_flags &= ~M_PKTHDR;
2172				sc->rl_tail->m_next = m;
2173				sc->rl_tail = m;
2174			}
2175			continue;
2176		}
2177
2178		/*
2179		 * NOTE: for the 8139C+, the frame length field
2180		 * is always 12 bits in size, but for the gigE chips,
2181		 * it is 13 bits (since the max RX frame length is 16K).
2182		 * Unfortunately, all 32 bits in the status word
2183		 * were already used, so to make room for the extra
2184		 * length bit, RealTek took out the 'frame alignment
2185		 * error' bit and shifted the other status bits
2186		 * over one slot. The OWN, EOR, FS and LS bits are
2187		 * still in the same places. We have already extracted
2188		 * the frame length and checked the OWN bit, so rather
2189		 * than using an alternate bit mapping, we shift the
2190		 * status bits one space to the right so we can evaluate
2191		 * them using the 8169 status as though it was in the
2192		 * same format as that of the 8139C+.
2193		 */
2194		if (sc->rl_type == RL_8169)
2195			rxstat >>= 1;
2196
2197		/*
2198		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2199		 * set, but if CRC is clear, it will still be a valid frame.
2200		 */
2201		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2202			rxerr = 1;
2203			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2204			    total_len > 8191 &&
2205			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2206				rxerr = 0;
2207			if (rxerr != 0) {
2208				ifp->if_ierrors++;
2209				/*
2210				 * If this is part of a multi-fragment packet,
2211				 * discard all the pieces.
2212				 */
2213				if (sc->rl_head != NULL) {
2214					m_freem(sc->rl_head);
2215					sc->rl_head = sc->rl_tail = NULL;
2216				}
2217				re_discard_rxbuf(sc, i);
2218				continue;
2219			}
2220		}
2221
2222		/*
2223		 * If allocating a replacement mbuf fails,
2224		 * reload the current one.
2225		 */
2226		if (jumbo != 0)
2227			rxerr = re_jumbo_newbuf(sc, i);
2228		else
2229			rxerr = re_newbuf(sc, i);
2230		if (rxerr != 0) {
2231			ifp->if_iqdrops++;
2232			if (sc->rl_head != NULL) {
2233				m_freem(sc->rl_head);
2234				sc->rl_head = sc->rl_tail = NULL;
2235			}
2236			re_discard_rxbuf(sc, i);
2237			continue;
2238		}
2239
2240		if (sc->rl_head != NULL) {
2241			if (jumbo != 0)
2242				m->m_len = total_len;
2243			else {
2244				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2245				if (m->m_len == 0)
2246					m->m_len = RE_RX_DESC_BUFLEN;
2247			}
2248			/*
2249			 * Special case: if there's 4 bytes or less
2250			 * in this buffer, the mbuf can be discarded:
2251			 * the last 4 bytes is the CRC, which we don't
2252			 * care about anyway.
2253			 */
2254			if (m->m_len <= ETHER_CRC_LEN) {
2255				sc->rl_tail->m_len -=
2256				    (ETHER_CRC_LEN - m->m_len);
2257				m_freem(m);
2258			} else {
2259				m->m_len -= ETHER_CRC_LEN;
2260				m->m_flags &= ~M_PKTHDR;
2261				sc->rl_tail->m_next = m;
2262			}
2263			m = sc->rl_head;
2264			sc->rl_head = sc->rl_tail = NULL;
2265			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2266		} else
2267			m->m_pkthdr.len = m->m_len =
2268			    (total_len - ETHER_CRC_LEN);
2269
2270#ifdef RE_FIXUP_RX
2271		re_fixup_rx(m);
2272#endif
2273		ifp->if_ipackets++;
2274		m->m_pkthdr.rcvif = ifp;
2275
2276		/* Do RX checksumming if enabled */
2277
2278		if (ifp->if_capenable & IFCAP_RXCSUM) {
2279			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2280				/* Check IP header checksum */
2281				if (rxstat & RL_RDESC_STAT_PROTOID)
2282					m->m_pkthdr.csum_flags |=
2283					    CSUM_IP_CHECKED;
2284				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2285					m->m_pkthdr.csum_flags |=
2286					    CSUM_IP_VALID;
2287
2288				/* Check TCP/UDP checksum */
2289				if ((RL_TCPPKT(rxstat) &&
2290				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2291				    (RL_UDPPKT(rxstat) &&
2292				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2293					m->m_pkthdr.csum_flags |=
2294						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2295					m->m_pkthdr.csum_data = 0xffff;
2296				}
2297			} else {
2298				/*
2299				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2300				 */
2301				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2302				    (rxvlan & RL_RDESC_IPV4))
2303					m->m_pkthdr.csum_flags |=
2304					    CSUM_IP_CHECKED;
2305				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2306				    (rxvlan & RL_RDESC_IPV4))
2307					m->m_pkthdr.csum_flags |=
2308					    CSUM_IP_VALID;
2309				if (((rxstat & RL_RDESC_STAT_TCP) &&
2310				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2311				    ((rxstat & RL_RDESC_STAT_UDP) &&
2312				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2313					m->m_pkthdr.csum_flags |=
2314						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2315					m->m_pkthdr.csum_data = 0xffff;
2316				}
2317			}
2318		}
2319		maxpkt--;
2320		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2321			m->m_pkthdr.ether_vtag =
2322			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2323			m->m_flags |= M_VLANTAG;
2324		}
2325		RL_UNLOCK(sc);
2326		(*ifp->if_input)(ifp, m);
2327		RL_LOCK(sc);
2328		rx_npkts++;
2329	}
2330
2331	/* Flush the RX DMA ring */
2332
2333	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2334	    sc->rl_ldata.rl_rx_list_map,
2335	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2336
2337	sc->rl_ldata.rl_rx_prodidx = i;
2338
2339	if (rx_npktsp != NULL)
2340		*rx_npktsp = rx_npkts;
2341	if (maxpkt)
2342		return (EAGAIN);
2343
2344	return (0);
2345}
2346
2347static void
2348re_txeof(struct rl_softc *sc)
2349{
2350	struct ifnet		*ifp;
2351	struct rl_txdesc	*txd;
2352	u_int32_t		txstat;
2353	int			cons;
2354
2355	cons = sc->rl_ldata.rl_tx_considx;
2356	if (cons == sc->rl_ldata.rl_tx_prodidx)
2357		return;
2358
2359	ifp = sc->rl_ifp;
2360#ifdef DEV_NETMAP
2361	if (ifp->if_capenable & IFCAP_NETMAP) {
2362		selwakeuppri(&NA(ifp)->tx_rings[0].si, PI_NET);
2363		return;
2364	}
2365#endif /* DEV_NETMAP */
2366	/* Invalidate the TX descriptor list */
2367	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2368	    sc->rl_ldata.rl_tx_list_map,
2369	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2370
2371	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2372	    cons = RL_TX_DESC_NXT(sc, cons)) {
2373		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2374		if (txstat & RL_TDESC_STAT_OWN)
2375			break;
2376		/*
2377		 * We only stash mbufs in the last descriptor
2378		 * in a fragment chain, which also happens to
2379		 * be the only place where the TX status bits
2380		 * are valid.
2381		 */
2382		if (txstat & RL_TDESC_CMD_EOF) {
2383			txd = &sc->rl_ldata.rl_tx_desc[cons];
2384			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2385			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2386			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2387			    txd->tx_dmamap);
2388			KASSERT(txd->tx_m != NULL,
2389			    ("%s: freeing NULL mbufs!", __func__));
2390			m_freem(txd->tx_m);
2391			txd->tx_m = NULL;
2392			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2393			    RL_TDESC_STAT_COLCNT))
2394				ifp->if_collisions++;
2395			if (txstat & RL_TDESC_STAT_TXERRSUM)
2396				ifp->if_oerrors++;
2397			else
2398				ifp->if_opackets++;
2399		}
2400		sc->rl_ldata.rl_tx_free++;
2401		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2402	}
2403	sc->rl_ldata.rl_tx_considx = cons;
2404
2405	/* No changes made to the TX ring, so no flush needed */
2406
2407	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2408#ifdef RE_TX_MODERATION
2409		/*
2410		 * If not all descriptors have been reaped yet, reload
2411		 * the timer so that we will eventually get another
2412		 * interrupt that will cause us to re-enter this routine.
2413		 * This is done in case the transmitter has gone idle.
2414		 */
2415		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2416#endif
2417	} else
2418		sc->rl_watchdog_timer = 0;
2419}
2420
2421static void
2422re_tick(void *xsc)
2423{
2424	struct rl_softc		*sc;
2425	struct mii_data		*mii;
2426
2427	sc = xsc;
2428
2429	RL_LOCK_ASSERT(sc);
2430
2431	mii = device_get_softc(sc->rl_miibus);
2432	mii_tick(mii);
2433	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2434		re_miibus_statchg(sc->rl_dev);
2435	/*
2436	 * Reclaim transmitted frames here. Technically it is not
2437	 * necessary to do here but it ensures periodic reclamation
2438	 * regardless of Tx completion interrupt which seems to be
2439	 * lost on PCIe based controllers under certain situations.
2440	 */
2441	re_txeof(sc);
2442	re_watchdog(sc);
2443	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2444}
2445
2446#ifdef DEVICE_POLLING
2447static int
2448re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2449{
2450	struct rl_softc *sc = ifp->if_softc;
2451	int rx_npkts = 0;
2452
2453	RL_LOCK(sc);
2454	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2455		rx_npkts = re_poll_locked(ifp, cmd, count);
2456	RL_UNLOCK(sc);
2457	return (rx_npkts);
2458}
2459
2460static int
2461re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2462{
2463	struct rl_softc *sc = ifp->if_softc;
2464	int rx_npkts;
2465
2466	RL_LOCK_ASSERT(sc);
2467
2468	sc->rxcycles = count;
2469	re_rxeof(sc, &rx_npkts);
2470	re_txeof(sc);
2471
2472	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2473		re_start_locked(ifp);
2474
2475	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2476		u_int16_t       status;
2477
2478		status = CSR_READ_2(sc, RL_ISR);
2479		if (status == 0xffff)
2480			return (rx_npkts);
2481		if (status)
2482			CSR_WRITE_2(sc, RL_ISR, status);
2483		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2484		    (sc->rl_flags & RL_FLAG_PCIE))
2485			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2486
2487		/*
2488		 * XXX check behaviour on receiver stalls.
2489		 */
2490
2491		if (status & RL_ISR_SYSTEM_ERR) {
2492			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2493			re_init_locked(sc);
2494		}
2495	}
2496	return (rx_npkts);
2497}
2498#endif /* DEVICE_POLLING */
2499
2500static int
2501re_intr(void *arg)
2502{
2503	struct rl_softc		*sc;
2504	uint16_t		status;
2505
2506	sc = arg;
2507
2508	status = CSR_READ_2(sc, RL_ISR);
2509	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2510                return (FILTER_STRAY);
2511	CSR_WRITE_2(sc, RL_IMR, 0);
2512
2513	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2514
2515	return (FILTER_HANDLED);
2516}
2517
2518static void
2519re_int_task(void *arg, int npending)
2520{
2521	struct rl_softc		*sc;
2522	struct ifnet		*ifp;
2523	u_int16_t		status;
2524	int			rval = 0;
2525
2526	sc = arg;
2527	ifp = sc->rl_ifp;
2528
2529	RL_LOCK(sc);
2530
2531	status = CSR_READ_2(sc, RL_ISR);
2532        CSR_WRITE_2(sc, RL_ISR, status);
2533
2534	if (sc->suspended ||
2535	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2536		RL_UNLOCK(sc);
2537		return;
2538	}
2539
2540#ifdef DEVICE_POLLING
2541	if  (ifp->if_capenable & IFCAP_POLLING) {
2542		RL_UNLOCK(sc);
2543		return;
2544	}
2545#endif
2546
2547	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2548		rval = re_rxeof(sc, NULL);
2549
2550	/*
2551	 * Some chips will ignore a second TX request issued
2552	 * while an existing transmission is in progress. If
2553	 * the transmitter goes idle but there are still
2554	 * packets waiting to be sent, we need to restart the
2555	 * channel here to flush them out. This only seems to
2556	 * be required with the PCIe devices.
2557	 */
2558	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2559	    (sc->rl_flags & RL_FLAG_PCIE))
2560		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2561	if (status & (
2562#ifdef RE_TX_MODERATION
2563	    RL_ISR_TIMEOUT_EXPIRED|
2564#else
2565	    RL_ISR_TX_OK|
2566#endif
2567	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2568		re_txeof(sc);
2569
2570	if (status & RL_ISR_SYSTEM_ERR) {
2571		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2572		re_init_locked(sc);
2573	}
2574
2575	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2576		re_start_locked(ifp);
2577
2578	RL_UNLOCK(sc);
2579
2580        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2581		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2582		return;
2583	}
2584
2585	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2586}
2587
2588static void
2589re_intr_msi(void *xsc)
2590{
2591	struct rl_softc		*sc;
2592	struct ifnet		*ifp;
2593	uint16_t		intrs, status;
2594
2595	sc = xsc;
2596	RL_LOCK(sc);
2597
2598	ifp = sc->rl_ifp;
2599#ifdef DEVICE_POLLING
2600	if (ifp->if_capenable & IFCAP_POLLING) {
2601		RL_UNLOCK(sc);
2602		return;
2603	}
2604#endif
2605	/* Disable interrupts. */
2606	CSR_WRITE_2(sc, RL_IMR, 0);
2607	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2608		RL_UNLOCK(sc);
2609		return;
2610	}
2611
2612	intrs = RL_INTRS_CPLUS;
2613	status = CSR_READ_2(sc, RL_ISR);
2614        CSR_WRITE_2(sc, RL_ISR, status);
2615	if (sc->rl_int_rx_act > 0) {
2616		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2617		    RL_ISR_RX_OVERRUN);
2618		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2619		    RL_ISR_RX_OVERRUN);
2620	}
2621
2622	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2623	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2624		re_rxeof(sc, NULL);
2625		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2626			if (sc->rl_int_rx_mod != 0 &&
2627			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2628			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2629				/* Rearm one-shot timer. */
2630				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2631				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2632				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2633				sc->rl_int_rx_act = 1;
2634			} else {
2635				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2636				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2637				sc->rl_int_rx_act = 0;
2638			}
2639		}
2640	}
2641
2642	/*
2643	 * Some chips will ignore a second TX request issued
2644	 * while an existing transmission is in progress. If
2645	 * the transmitter goes idle but there are still
2646	 * packets waiting to be sent, we need to restart the
2647	 * channel here to flush them out. This only seems to
2648	 * be required with the PCIe devices.
2649	 */
2650	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2651	    (sc->rl_flags & RL_FLAG_PCIE))
2652		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2653	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2654		re_txeof(sc);
2655
2656	if (status & RL_ISR_SYSTEM_ERR) {
2657		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2658		re_init_locked(sc);
2659	}
2660
2661	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2662		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2663			re_start_locked(ifp);
2664		CSR_WRITE_2(sc, RL_IMR, intrs);
2665	}
2666	RL_UNLOCK(sc);
2667}
2668
2669static int
2670re_encap(struct rl_softc *sc, struct mbuf **m_head)
2671{
2672	struct rl_txdesc	*txd, *txd_last;
2673	bus_dma_segment_t	segs[RL_NTXSEGS];
2674	bus_dmamap_t		map;
2675	struct mbuf		*m_new;
2676	struct rl_desc		*desc;
2677	int			nsegs, prod;
2678	int			i, error, ei, si;
2679	int			padlen;
2680	uint32_t		cmdstat, csum_flags, vlanctl;
2681
2682	RL_LOCK_ASSERT(sc);
2683	M_ASSERTPKTHDR((*m_head));
2684
2685	/*
2686	 * With some of the RealTek chips, using the checksum offload
2687	 * support in conjunction with the autopadding feature results
2688	 * in the transmission of corrupt frames. For example, if we
2689	 * need to send a really small IP fragment that's less than 60
2690	 * bytes in size, and IP header checksumming is enabled, the
2691	 * resulting ethernet frame that appears on the wire will
2692	 * have garbled payload. To work around this, if TX IP checksum
2693	 * offload is enabled, we always manually pad short frames out
2694	 * to the minimum ethernet frame size.
2695	 */
2696	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2697	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2698	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2699		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2700		if (M_WRITABLE(*m_head) == 0) {
2701			/* Get a writable copy. */
2702			m_new = m_dup(*m_head, M_DONTWAIT);
2703			m_freem(*m_head);
2704			if (m_new == NULL) {
2705				*m_head = NULL;
2706				return (ENOBUFS);
2707			}
2708			*m_head = m_new;
2709		}
2710		if ((*m_head)->m_next != NULL ||
2711		    M_TRAILINGSPACE(*m_head) < padlen) {
2712			m_new = m_defrag(*m_head, M_DONTWAIT);
2713			if (m_new == NULL) {
2714				m_freem(*m_head);
2715				*m_head = NULL;
2716				return (ENOBUFS);
2717			}
2718		} else
2719			m_new = *m_head;
2720
2721		/*
2722		 * Manually pad short frames, and zero the pad space
2723		 * to avoid leaking data.
2724		 */
2725		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2726		m_new->m_pkthdr.len += padlen;
2727		m_new->m_len = m_new->m_pkthdr.len;
2728		*m_head = m_new;
2729	}
2730
2731	prod = sc->rl_ldata.rl_tx_prodidx;
2732	txd = &sc->rl_ldata.rl_tx_desc[prod];
2733	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2734	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2735	if (error == EFBIG) {
2736		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2737		if (m_new == NULL) {
2738			m_freem(*m_head);
2739			*m_head = NULL;
2740			return (ENOBUFS);
2741		}
2742		*m_head = m_new;
2743		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2744		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2745		if (error != 0) {
2746			m_freem(*m_head);
2747			*m_head = NULL;
2748			return (error);
2749		}
2750	} else if (error != 0)
2751		return (error);
2752	if (nsegs == 0) {
2753		m_freem(*m_head);
2754		*m_head = NULL;
2755		return (EIO);
2756	}
2757
2758	/* Check for number of available descriptors. */
2759	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2760		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2761		return (ENOBUFS);
2762	}
2763
2764	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2765	    BUS_DMASYNC_PREWRITE);
2766
2767	/*
2768	 * Set up checksum offload. Note: checksum offload bits must
2769	 * appear in all descriptors of a multi-descriptor transmit
2770	 * attempt. This is according to testing done with an 8169
2771	 * chip. This is a requirement.
2772	 */
2773	vlanctl = 0;
2774	csum_flags = 0;
2775	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2776		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2777			csum_flags |= RL_TDESC_CMD_LGSEND;
2778			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2779			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2780		} else {
2781			csum_flags |= RL_TDESC_CMD_LGSEND |
2782			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2783			    RL_TDESC_CMD_MSSVAL_SHIFT);
2784		}
2785	} else {
2786		/*
2787		 * Unconditionally enable IP checksum if TCP or UDP
2788		 * checksum is required. Otherwise, TCP/UDP checksum
2789		 * does't make effects.
2790		 */
2791		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2792			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2793				csum_flags |= RL_TDESC_CMD_IPCSUM;
2794				if (((*m_head)->m_pkthdr.csum_flags &
2795				    CSUM_TCP) != 0)
2796					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2797				if (((*m_head)->m_pkthdr.csum_flags &
2798				    CSUM_UDP) != 0)
2799					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2800			} else {
2801				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2802				if (((*m_head)->m_pkthdr.csum_flags &
2803				    CSUM_TCP) != 0)
2804					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2805				if (((*m_head)->m_pkthdr.csum_flags &
2806				    CSUM_UDP) != 0)
2807					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2808			}
2809		}
2810	}
2811
2812	/*
2813	 * Set up hardware VLAN tagging. Note: vlan tag info must
2814	 * appear in all descriptors of a multi-descriptor
2815	 * transmission attempt.
2816	 */
2817	if ((*m_head)->m_flags & M_VLANTAG)
2818		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2819		    RL_TDESC_VLANCTL_TAG;
2820
2821	si = prod;
2822	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2823		desc = &sc->rl_ldata.rl_tx_list[prod];
2824		desc->rl_vlanctl = htole32(vlanctl);
2825		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2826		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2827		cmdstat = segs[i].ds_len;
2828		if (i != 0)
2829			cmdstat |= RL_TDESC_CMD_OWN;
2830		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2831			cmdstat |= RL_TDESC_CMD_EOR;
2832		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2833		sc->rl_ldata.rl_tx_free--;
2834	}
2835	/* Update producer index. */
2836	sc->rl_ldata.rl_tx_prodidx = prod;
2837
2838	/* Set EOF on the last descriptor. */
2839	ei = RL_TX_DESC_PRV(sc, prod);
2840	desc = &sc->rl_ldata.rl_tx_list[ei];
2841	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2842
2843	desc = &sc->rl_ldata.rl_tx_list[si];
2844	/* Set SOF and transfer ownership of packet to the chip. */
2845	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2846
2847	/*
2848	 * Insure that the map for this transmission
2849	 * is placed at the array index of the last descriptor
2850	 * in this chain.  (Swap last and first dmamaps.)
2851	 */
2852	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2853	map = txd->tx_dmamap;
2854	txd->tx_dmamap = txd_last->tx_dmamap;
2855	txd_last->tx_dmamap = map;
2856	txd_last->tx_m = *m_head;
2857
2858	return (0);
2859}
2860
2861static void
2862re_start(struct ifnet *ifp)
2863{
2864	struct rl_softc		*sc;
2865
2866	sc = ifp->if_softc;
2867	RL_LOCK(sc);
2868	re_start_locked(ifp);
2869	RL_UNLOCK(sc);
2870}
2871
2872/*
2873 * Main transmit routine for C+ and gigE NICs.
2874 */
2875static void
2876re_start_locked(struct ifnet *ifp)
2877{
2878	struct rl_softc		*sc;
2879	struct mbuf		*m_head;
2880	int			queued;
2881
2882	sc = ifp->if_softc;
2883
2884#ifdef DEV_NETMAP
2885	/* XXX is this necessary ? */
2886	if (ifp->if_capenable & IFCAP_NETMAP) {
2887		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2888		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2889			/* kick the tx unit */
2890			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2891#ifdef RE_TX_MODERATION
2892			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2893#endif
2894			sc->rl_watchdog_timer = 5;
2895		}
2896		return;
2897	}
2898#endif /* DEV_NETMAP */
2899	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2900	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2901		return;
2902
2903	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2904	    sc->rl_ldata.rl_tx_free > 1;) {
2905		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2906		if (m_head == NULL)
2907			break;
2908
2909		if (re_encap(sc, &m_head) != 0) {
2910			if (m_head == NULL)
2911				break;
2912			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2913			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2914			break;
2915		}
2916
2917		/*
2918		 * If there's a BPF listener, bounce a copy of this frame
2919		 * to him.
2920		 */
2921		ETHER_BPF_MTAP(ifp, m_head);
2922
2923		queued++;
2924	}
2925
2926	if (queued == 0) {
2927#ifdef RE_TX_MODERATION
2928		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2929			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2930#endif
2931		return;
2932	}
2933
2934	/* Flush the TX descriptors */
2935
2936	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2937	    sc->rl_ldata.rl_tx_list_map,
2938	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2939
2940	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2941
2942#ifdef RE_TX_MODERATION
2943	/*
2944	 * Use the countdown timer for interrupt moderation.
2945	 * 'TX done' interrupts are disabled. Instead, we reset the
2946	 * countdown timer, which will begin counting until it hits
2947	 * the value in the TIMERINT register, and then trigger an
2948	 * interrupt. Each time we write to the TIMERCNT register,
2949	 * the timer count is reset to 0.
2950	 */
2951	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2952#endif
2953
2954	/*
2955	 * Set a timeout in case the chip goes out to lunch.
2956	 */
2957	sc->rl_watchdog_timer = 5;
2958}
2959
2960static void
2961re_set_jumbo(struct rl_softc *sc, int jumbo)
2962{
2963
2964	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2965		pci_set_max_read_req(sc->rl_dev, 4096);
2966		return;
2967	}
2968
2969	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2970	if (jumbo != 0) {
2971		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
2972		    RL_CFG3_JUMBO_EN0);
2973		switch (sc->rl_hwrev->rl_rev) {
2974		case RL_HWREV_8168DP:
2975			break;
2976		case RL_HWREV_8168E:
2977			CSR_WRITE_1(sc, sc->rl_cfg4,
2978			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
2979			break;
2980		default:
2981			CSR_WRITE_1(sc, sc->rl_cfg4,
2982			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
2983		}
2984	} else {
2985		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
2986		    ~RL_CFG3_JUMBO_EN0);
2987		switch (sc->rl_hwrev->rl_rev) {
2988		case RL_HWREV_8168DP:
2989			break;
2990		case RL_HWREV_8168E:
2991			CSR_WRITE_1(sc, sc->rl_cfg4,
2992			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
2993			break;
2994		default:
2995			CSR_WRITE_1(sc, sc->rl_cfg4,
2996			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
2997		}
2998	}
2999	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3000
3001	switch (sc->rl_hwrev->rl_rev) {
3002	case RL_HWREV_8168DP:
3003		pci_set_max_read_req(sc->rl_dev, 4096);
3004		break;
3005	default:
3006		if (jumbo != 0)
3007			pci_set_max_read_req(sc->rl_dev, 512);
3008		else
3009			pci_set_max_read_req(sc->rl_dev, 4096);
3010	}
3011}
3012
3013static void
3014re_init(void *xsc)
3015{
3016	struct rl_softc		*sc = xsc;
3017
3018	RL_LOCK(sc);
3019	re_init_locked(sc);
3020	RL_UNLOCK(sc);
3021}
3022
3023static void
3024re_init_locked(struct rl_softc *sc)
3025{
3026	struct ifnet		*ifp = sc->rl_ifp;
3027	struct mii_data		*mii;
3028	uint32_t		reg;
3029	uint16_t		cfg;
3030	union {
3031		uint32_t align_dummy;
3032		u_char eaddr[ETHER_ADDR_LEN];
3033        } eaddr;
3034
3035	RL_LOCK_ASSERT(sc);
3036
3037	mii = device_get_softc(sc->rl_miibus);
3038
3039	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3040		return;
3041
3042	/*
3043	 * Cancel pending I/O and free all RX/TX buffers.
3044	 */
3045	re_stop(sc);
3046
3047	/* Put controller into known state. */
3048	re_reset(sc);
3049
3050	/*
3051	 * For C+ mode, initialize the RX descriptors and mbufs.
3052	 */
3053	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3054		if (ifp->if_mtu > RL_MTU) {
3055			if (re_jrx_list_init(sc) != 0) {
3056				device_printf(sc->rl_dev,
3057				    "no memory for jumbo RX buffers\n");
3058				re_stop(sc);
3059				return;
3060			}
3061			/* Disable checksum offloading for jumbo frames. */
3062			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3063			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3064		} else {
3065			if (re_rx_list_init(sc) != 0) {
3066				device_printf(sc->rl_dev,
3067				    "no memory for RX buffers\n");
3068				re_stop(sc);
3069				return;
3070			}
3071		}
3072		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3073	} else {
3074		if (re_rx_list_init(sc) != 0) {
3075			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3076			re_stop(sc);
3077			return;
3078		}
3079		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3080		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3081			if (ifp->if_mtu > RL_MTU)
3082				pci_set_max_read_req(sc->rl_dev, 512);
3083			else
3084				pci_set_max_read_req(sc->rl_dev, 4096);
3085		}
3086	}
3087	re_tx_list_init(sc);
3088
3089	/*
3090	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3091	 * RX checksum offload. We must configure the C+ register
3092	 * before all others.
3093	 */
3094	cfg = RL_CPLUSCMD_PCI_MRW;
3095	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3096		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3097	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3098		cfg |= RL_CPLUSCMD_VLANSTRIP;
3099	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3100		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3101		/* XXX magic. */
3102		cfg |= 0x0001;
3103	} else
3104		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3105	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3106	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3107	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3108		reg = 0x000fff00;
3109		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3110			reg |= 0x000000ff;
3111		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3112			reg |= 0x00f00000;
3113		CSR_WRITE_4(sc, 0x7c, reg);
3114		/* Disable interrupt mitigation. */
3115		CSR_WRITE_2(sc, 0xe2, 0);
3116	}
3117	/*
3118	 * Disable TSO if interface MTU size is greater than MSS
3119	 * allowed in controller.
3120	 */
3121	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3122		ifp->if_capenable &= ~IFCAP_TSO4;
3123		ifp->if_hwassist &= ~CSUM_TSO;
3124	}
3125
3126	/*
3127	 * Init our MAC address.  Even though the chipset
3128	 * documentation doesn't mention it, we need to enter "Config
3129	 * register write enable" mode to modify the ID registers.
3130	 */
3131	/* Copy MAC address on stack to align. */
3132	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3133	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3134	CSR_WRITE_4(sc, RL_IDR0,
3135	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3136	CSR_WRITE_4(sc, RL_IDR4,
3137	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3138	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3139
3140	/*
3141	 * Load the addresses of the RX and TX lists into the chip.
3142	 */
3143
3144	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3145	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3146	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3147	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3148
3149	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3150	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3151	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3152	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3153
3154	/*
3155	 * Enable transmit and receive.
3156	 */
3157	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3158
3159	/*
3160	 * Set the initial TX configuration.
3161	 */
3162	if (sc->rl_testmode) {
3163		if (sc->rl_type == RL_8169)
3164			CSR_WRITE_4(sc, RL_TXCFG,
3165			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3166		else
3167			CSR_WRITE_4(sc, RL_TXCFG,
3168			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3169	} else
3170		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3171
3172	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3173
3174	/*
3175	 * Set the initial RX configuration.
3176	 */
3177	re_set_rxmode(sc);
3178
3179	/* Configure interrupt moderation. */
3180	if (sc->rl_type == RL_8169) {
3181		/* Magic from vendor. */
3182		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3183	}
3184
3185#ifdef DEVICE_POLLING
3186	/*
3187	 * Disable interrupts if we are polling.
3188	 */
3189	if (ifp->if_capenable & IFCAP_POLLING)
3190		CSR_WRITE_2(sc, RL_IMR, 0);
3191	else	/* otherwise ... */
3192#endif
3193
3194	/*
3195	 * Enable interrupts.
3196	 */
3197	if (sc->rl_testmode)
3198		CSR_WRITE_2(sc, RL_IMR, 0);
3199	else
3200		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3201	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3202
3203	/* Set initial TX threshold */
3204	sc->rl_txthresh = RL_TX_THRESH_INIT;
3205
3206	/* Start RX/TX process. */
3207	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3208#ifdef notdef
3209	/* Enable receiver and transmitter. */
3210	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3211#endif
3212
3213	/*
3214	 * Initialize the timer interrupt register so that
3215	 * a timer interrupt will be generated once the timer
3216	 * reaches a certain number of ticks. The timer is
3217	 * reloaded on each transmit.
3218	 */
3219#ifdef RE_TX_MODERATION
3220	/*
3221	 * Use timer interrupt register to moderate TX interrupt
3222	 * moderation, which dramatically improves TX frame rate.
3223	 */
3224	if (sc->rl_type == RL_8169)
3225		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3226	else
3227		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3228#else
3229	/*
3230	 * Use timer interrupt register to moderate RX interrupt
3231	 * moderation.
3232	 */
3233	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3234	    intr_filter == 0) {
3235		if (sc->rl_type == RL_8169)
3236			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3237			    RL_USECS(sc->rl_int_rx_mod));
3238	} else {
3239		if (sc->rl_type == RL_8169)
3240			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3241	}
3242#endif
3243
3244	/*
3245	 * For 8169 gigE NICs, set the max allowed RX packet
3246	 * size so we can receive jumbo frames.
3247	 */
3248	if (sc->rl_type == RL_8169) {
3249		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3250			/*
3251			 * For controllers that use new jumbo frame scheme,
3252			 * set maximum size of jumbo frame depedning on
3253			 * controller revisions.
3254			 */
3255			if (ifp->if_mtu > RL_MTU)
3256				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3257				    sc->rl_hwrev->rl_max_mtu +
3258				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3259				    ETHER_CRC_LEN);
3260			else
3261				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3262				    RE_RX_DESC_BUFLEN);
3263		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3264		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3265			/* RTL810x has no jumbo frame support. */
3266			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3267		} else
3268			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3269	}
3270
3271	if (sc->rl_testmode)
3272		return;
3273
3274	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3275	    RL_CFG1_DRVLOAD);
3276
3277	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3278	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3279
3280	sc->rl_flags &= ~RL_FLAG_LINK;
3281	mii_mediachg(mii);
3282
3283	sc->rl_watchdog_timer = 0;
3284	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3285}
3286
3287/*
3288 * Set media options.
3289 */
3290static int
3291re_ifmedia_upd(struct ifnet *ifp)
3292{
3293	struct rl_softc		*sc;
3294	struct mii_data		*mii;
3295	int			error;
3296
3297	sc = ifp->if_softc;
3298	mii = device_get_softc(sc->rl_miibus);
3299	RL_LOCK(sc);
3300	error = mii_mediachg(mii);
3301	RL_UNLOCK(sc);
3302
3303	return (error);
3304}
3305
3306/*
3307 * Report current media status.
3308 */
3309static void
3310re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3311{
3312	struct rl_softc		*sc;
3313	struct mii_data		*mii;
3314
3315	sc = ifp->if_softc;
3316	mii = device_get_softc(sc->rl_miibus);
3317
3318	RL_LOCK(sc);
3319	mii_pollstat(mii);
3320	ifmr->ifm_active = mii->mii_media_active;
3321	ifmr->ifm_status = mii->mii_media_status;
3322	RL_UNLOCK(sc);
3323}
3324
3325static int
3326re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3327{
3328	struct rl_softc		*sc = ifp->if_softc;
3329	struct ifreq		*ifr = (struct ifreq *) data;
3330	struct mii_data		*mii;
3331	uint32_t		rev;
3332	int			error = 0;
3333
3334	switch (command) {
3335	case SIOCSIFMTU:
3336		if (ifr->ifr_mtu < ETHERMIN ||
3337		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3338			error = EINVAL;
3339			break;
3340		}
3341		RL_LOCK(sc);
3342		if (ifp->if_mtu != ifr->ifr_mtu) {
3343			ifp->if_mtu = ifr->ifr_mtu;
3344			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3345			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3346				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3347				re_init_locked(sc);
3348			}
3349			if (ifp->if_mtu > RL_TSO_MTU &&
3350			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3351				ifp->if_capenable &= ~(IFCAP_TSO4 |
3352				    IFCAP_VLAN_HWTSO);
3353				ifp->if_hwassist &= ~CSUM_TSO;
3354			}
3355			VLAN_CAPABILITIES(ifp);
3356		}
3357		RL_UNLOCK(sc);
3358		break;
3359	case SIOCSIFFLAGS:
3360		RL_LOCK(sc);
3361		if ((ifp->if_flags & IFF_UP) != 0) {
3362			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3363				if (((ifp->if_flags ^ sc->rl_if_flags)
3364				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3365					re_set_rxmode(sc);
3366			} else
3367				re_init_locked(sc);
3368		} else {
3369			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3370				re_stop(sc);
3371		}
3372		sc->rl_if_flags = ifp->if_flags;
3373		RL_UNLOCK(sc);
3374		break;
3375	case SIOCADDMULTI:
3376	case SIOCDELMULTI:
3377		RL_LOCK(sc);
3378		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3379			re_set_rxmode(sc);
3380		RL_UNLOCK(sc);
3381		break;
3382	case SIOCGIFMEDIA:
3383	case SIOCSIFMEDIA:
3384		mii = device_get_softc(sc->rl_miibus);
3385		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3386		break;
3387	case SIOCSIFCAP:
3388	    {
3389		int mask, reinit;
3390
3391		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3392		reinit = 0;
3393#ifdef DEVICE_POLLING
3394		if (mask & IFCAP_POLLING) {
3395			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3396				error = ether_poll_register(re_poll, ifp);
3397				if (error)
3398					return (error);
3399				RL_LOCK(sc);
3400				/* Disable interrupts */
3401				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3402				ifp->if_capenable |= IFCAP_POLLING;
3403				RL_UNLOCK(sc);
3404			} else {
3405				error = ether_poll_deregister(ifp);
3406				/* Enable interrupts. */
3407				RL_LOCK(sc);
3408				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3409				ifp->if_capenable &= ~IFCAP_POLLING;
3410				RL_UNLOCK(sc);
3411			}
3412		}
3413#endif /* DEVICE_POLLING */
3414		RL_LOCK(sc);
3415		if ((mask & IFCAP_TXCSUM) != 0 &&
3416		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3417			ifp->if_capenable ^= IFCAP_TXCSUM;
3418			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3419				rev = sc->rl_hwrev->rl_rev;
3420				if (rev == RL_HWREV_8168C ||
3421				    rev == RL_HWREV_8168C_SPIN2)
3422					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3423				else
3424					ifp->if_hwassist |= RE_CSUM_FEATURES;
3425			} else
3426				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3427			reinit = 1;
3428		}
3429		if ((mask & IFCAP_RXCSUM) != 0 &&
3430		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3431			ifp->if_capenable ^= IFCAP_RXCSUM;
3432			reinit = 1;
3433		}
3434		if ((mask & IFCAP_TSO4) != 0 &&
3435		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3436			ifp->if_capenable ^= IFCAP_TSO4;
3437			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3438				ifp->if_hwassist |= CSUM_TSO;
3439			else
3440				ifp->if_hwassist &= ~CSUM_TSO;
3441			if (ifp->if_mtu > RL_TSO_MTU &&
3442			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3443				ifp->if_capenable &= ~IFCAP_TSO4;
3444				ifp->if_hwassist &= ~CSUM_TSO;
3445			}
3446		}
3447		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3448		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3449			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3450		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3451		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3452			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3453			/* TSO over VLAN requires VLAN hardware tagging. */
3454			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3455				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3456			reinit = 1;
3457		}
3458		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3459		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3460		    IFCAP_VLAN_HWTSO)) != 0)
3461				reinit = 1;
3462		if ((mask & IFCAP_WOL) != 0 &&
3463		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3464			if ((mask & IFCAP_WOL_UCAST) != 0)
3465				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3466			if ((mask & IFCAP_WOL_MCAST) != 0)
3467				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3468			if ((mask & IFCAP_WOL_MAGIC) != 0)
3469				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3470		}
3471		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3472			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3473			re_init_locked(sc);
3474		}
3475		RL_UNLOCK(sc);
3476		VLAN_CAPABILITIES(ifp);
3477	    }
3478		break;
3479	default:
3480		error = ether_ioctl(ifp, command, data);
3481		break;
3482	}
3483
3484	return (error);
3485}
3486
3487static void
3488re_watchdog(struct rl_softc *sc)
3489{
3490	struct ifnet		*ifp;
3491
3492	RL_LOCK_ASSERT(sc);
3493
3494	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3495		return;
3496
3497	ifp = sc->rl_ifp;
3498	re_txeof(sc);
3499	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3500		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3501		    "-- recovering\n");
3502		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3503			re_start_locked(ifp);
3504		return;
3505	}
3506
3507	if_printf(ifp, "watchdog timeout\n");
3508	ifp->if_oerrors++;
3509
3510	re_rxeof(sc, NULL);
3511	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3512	re_init_locked(sc);
3513	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3514		re_start_locked(ifp);
3515}
3516
3517/*
3518 * Stop the adapter and free any mbufs allocated to the
3519 * RX and TX lists.
3520 */
3521static void
3522re_stop(struct rl_softc *sc)
3523{
3524	int			i;
3525	struct ifnet		*ifp;
3526	struct rl_txdesc	*txd;
3527	struct rl_rxdesc	*rxd;
3528
3529	RL_LOCK_ASSERT(sc);
3530
3531	ifp = sc->rl_ifp;
3532
3533	sc->rl_watchdog_timer = 0;
3534	callout_stop(&sc->rl_stat_callout);
3535	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3536
3537	/*
3538	 * Disable accepting frames to put RX MAC into idle state.
3539	 * Otherwise it's possible to get frames while stop command
3540	 * execution is in progress and controller can DMA the frame
3541	 * to already freed RX buffer during that period.
3542	 */
3543	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3544	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3545	    RL_RXCFG_RX_BROAD));
3546
3547	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3548		for (i = RL_TIMEOUT; i > 0; i--) {
3549			if ((CSR_READ_1(sc, sc->rl_txstart) &
3550			    RL_TXSTART_START) == 0)
3551				break;
3552			DELAY(20);
3553		}
3554		if (i == 0)
3555			device_printf(sc->rl_dev,
3556			    "stopping TX poll timed out!\n");
3557		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3558	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3559		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3560		    RL_CMD_RX_ENB);
3561		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3562			for (i = RL_TIMEOUT; i > 0; i--) {
3563				if ((CSR_READ_4(sc, RL_TXCFG) &
3564				    RL_TXCFG_QUEUE_EMPTY) != 0)
3565					break;
3566				DELAY(100);
3567			}
3568			if (i == 0)
3569				device_printf(sc->rl_dev,
3570				   "stopping TXQ timed out!\n");
3571		}
3572	} else
3573		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3574	DELAY(1000);
3575	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3576	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3577
3578	if (sc->rl_head != NULL) {
3579		m_freem(sc->rl_head);
3580		sc->rl_head = sc->rl_tail = NULL;
3581	}
3582
3583	/* Free the TX list buffers. */
3584	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3585		txd = &sc->rl_ldata.rl_tx_desc[i];
3586		if (txd->tx_m != NULL) {
3587			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3588			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3589			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3590			    txd->tx_dmamap);
3591			m_freem(txd->tx_m);
3592			txd->tx_m = NULL;
3593		}
3594	}
3595
3596	/* Free the RX list buffers. */
3597	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3598		rxd = &sc->rl_ldata.rl_rx_desc[i];
3599		if (rxd->rx_m != NULL) {
3600			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3601			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3602			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3603			    rxd->rx_dmamap);
3604			m_freem(rxd->rx_m);
3605			rxd->rx_m = NULL;
3606		}
3607	}
3608
3609	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3610		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3611			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3612			if (rxd->rx_m != NULL) {
3613				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3614				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3615				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3616				    rxd->rx_dmamap);
3617				m_freem(rxd->rx_m);
3618				rxd->rx_m = NULL;
3619			}
3620		}
3621	}
3622}
3623
3624/*
3625 * Device suspend routine.  Stop the interface and save some PCI
3626 * settings in case the BIOS doesn't restore them properly on
3627 * resume.
3628 */
3629static int
3630re_suspend(device_t dev)
3631{
3632	struct rl_softc		*sc;
3633
3634	sc = device_get_softc(dev);
3635
3636	RL_LOCK(sc);
3637	re_stop(sc);
3638	re_setwol(sc);
3639	sc->suspended = 1;
3640	RL_UNLOCK(sc);
3641
3642	return (0);
3643}
3644
3645/*
3646 * Device resume routine.  Restore some PCI settings in case the BIOS
3647 * doesn't, re-enable busmastering, and restart the interface if
3648 * appropriate.
3649 */
3650static int
3651re_resume(device_t dev)
3652{
3653	struct rl_softc		*sc;
3654	struct ifnet		*ifp;
3655
3656	sc = device_get_softc(dev);
3657
3658	RL_LOCK(sc);
3659
3660	ifp = sc->rl_ifp;
3661	/* Take controller out of sleep mode. */
3662	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3663		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3664			CSR_WRITE_1(sc, RL_GPIO,
3665			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3666	}
3667
3668	/*
3669	 * Clear WOL matching such that normal Rx filtering
3670	 * wouldn't interfere with WOL patterns.
3671	 */
3672	re_clrwol(sc);
3673
3674	/* reinitialize interface if necessary */
3675	if (ifp->if_flags & IFF_UP)
3676		re_init_locked(sc);
3677
3678	sc->suspended = 0;
3679	RL_UNLOCK(sc);
3680
3681	return (0);
3682}
3683
3684/*
3685 * Stop all chip I/O so that the kernel's probe routines don't
3686 * get confused by errant DMAs when rebooting.
3687 */
3688static int
3689re_shutdown(device_t dev)
3690{
3691	struct rl_softc		*sc;
3692
3693	sc = device_get_softc(dev);
3694
3695	RL_LOCK(sc);
3696	re_stop(sc);
3697	/*
3698	 * Mark interface as down since otherwise we will panic if
3699	 * interrupt comes in later on, which can happen in some
3700	 * cases.
3701	 */
3702	sc->rl_ifp->if_flags &= ~IFF_UP;
3703	re_setwol(sc);
3704	RL_UNLOCK(sc);
3705
3706	return (0);
3707}
3708
3709static void
3710re_set_linkspeed(struct rl_softc *sc)
3711{
3712	struct mii_softc *miisc;
3713	struct mii_data *mii;
3714	int aneg, i, phyno;
3715
3716	RL_LOCK_ASSERT(sc);
3717
3718	mii = device_get_softc(sc->rl_miibus);
3719	mii_pollstat(mii);
3720	aneg = 0;
3721	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3722	    (IFM_ACTIVE | IFM_AVALID)) {
3723		switch IFM_SUBTYPE(mii->mii_media_active) {
3724		case IFM_10_T:
3725		case IFM_100_TX:
3726			return;
3727		case IFM_1000_T:
3728			aneg++;
3729			break;
3730		default:
3731			break;
3732		}
3733	}
3734	miisc = LIST_FIRST(&mii->mii_phys);
3735	phyno = miisc->mii_phy;
3736	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3737		PHY_RESET(miisc);
3738	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3739	re_miibus_writereg(sc->rl_dev, phyno,
3740	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3741	re_miibus_writereg(sc->rl_dev, phyno,
3742	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3743	DELAY(1000);
3744	if (aneg != 0) {
3745		/*
3746		 * Poll link state until re(4) get a 10/100Mbps link.
3747		 */
3748		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3749			mii_pollstat(mii);
3750			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3751			    == (IFM_ACTIVE | IFM_AVALID)) {
3752				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3753				case IFM_10_T:
3754				case IFM_100_TX:
3755					return;
3756				default:
3757					break;
3758				}
3759			}
3760			RL_UNLOCK(sc);
3761			pause("relnk", hz);
3762			RL_LOCK(sc);
3763		}
3764		if (i == MII_ANEGTICKS_GIGE)
3765			device_printf(sc->rl_dev,
3766			    "establishing a link failed, WOL may not work!");
3767	}
3768	/*
3769	 * No link, force MAC to have 100Mbps, full-duplex link.
3770	 * MAC does not require reprogramming on resolved speed/duplex,
3771	 * so this is just for completeness.
3772	 */
3773	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3774	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3775}
3776
3777static void
3778re_setwol(struct rl_softc *sc)
3779{
3780	struct ifnet		*ifp;
3781	int			pmc;
3782	uint16_t		pmstat;
3783	uint8_t			v;
3784
3785	RL_LOCK_ASSERT(sc);
3786
3787	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3788		return;
3789
3790	ifp = sc->rl_ifp;
3791	/* Put controller into sleep mode. */
3792	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3793		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3794			CSR_WRITE_1(sc, RL_GPIO,
3795			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3796	}
3797	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3798		re_set_rxmode(sc);
3799		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3800			re_set_linkspeed(sc);
3801		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3802			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3803	}
3804	/* Enable config register write. */
3805	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3806
3807	/* Enable PME. */
3808	v = CSR_READ_1(sc, sc->rl_cfg1);
3809	v &= ~RL_CFG1_PME;
3810	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3811		v |= RL_CFG1_PME;
3812	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3813
3814	v = CSR_READ_1(sc, sc->rl_cfg3);
3815	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3816	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3817		v |= RL_CFG3_WOL_MAGIC;
3818	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3819
3820	v = CSR_READ_1(sc, sc->rl_cfg5);
3821	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3822	    RL_CFG5_WOL_LANWAKE);
3823	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3824		v |= RL_CFG5_WOL_UCAST;
3825	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3826		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3827	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3828		v |= RL_CFG5_WOL_LANWAKE;
3829	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3830
3831	/* Config register write done. */
3832	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3833
3834	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3835	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3836		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3837	/*
3838	 * It seems that hardware resets its link speed to 100Mbps in
3839	 * power down mode so switching to 100Mbps in driver is not
3840	 * needed.
3841	 */
3842
3843	/* Request PME if WOL is requested. */
3844	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3845	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3846	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3847		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3848	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3849}
3850
3851static void
3852re_clrwol(struct rl_softc *sc)
3853{
3854	int			pmc;
3855	uint8_t			v;
3856
3857	RL_LOCK_ASSERT(sc);
3858
3859	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3860		return;
3861
3862	/* Enable config register write. */
3863	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3864
3865	v = CSR_READ_1(sc, sc->rl_cfg3);
3866	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3867	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3868
3869	/* Config register write done. */
3870	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3871
3872	v = CSR_READ_1(sc, sc->rl_cfg5);
3873	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3874	v &= ~RL_CFG5_WOL_LANWAKE;
3875	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3876}
3877
3878static void
3879re_add_sysctls(struct rl_softc *sc)
3880{
3881	struct sysctl_ctx_list	*ctx;
3882	struct sysctl_oid_list	*children;
3883	int			error;
3884
3885	ctx = device_get_sysctl_ctx(sc->rl_dev);
3886	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3887
3888	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3889	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3890	    "Statistics Information");
3891	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3892		return;
3893
3894	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3895	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3896	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3897	/* Pull in device tunables. */
3898	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3899	error = resource_int_value(device_get_name(sc->rl_dev),
3900	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3901	if (error == 0) {
3902		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3903		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3904			device_printf(sc->rl_dev, "int_rx_mod value out of "
3905			    "range; using default: %d\n",
3906			    RL_TIMER_DEFAULT);
3907			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3908		}
3909	}
3910
3911}
3912
3913static int
3914re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3915{
3916	struct rl_softc		*sc;
3917	struct rl_stats		*stats;
3918	int			error, i, result;
3919
3920	result = -1;
3921	error = sysctl_handle_int(oidp, &result, 0, req);
3922	if (error || req->newptr == NULL)
3923		return (error);
3924
3925	if (result == 1) {
3926		sc = (struct rl_softc *)arg1;
3927		RL_LOCK(sc);
3928		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3929			RL_UNLOCK(sc);
3930			goto done;
3931		}
3932		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3933		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3934		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3935		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3936		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3937		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3938		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3939		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3940		    RL_DUMPSTATS_START));
3941		for (i = RL_TIMEOUT; i > 0; i--) {
3942			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3943			    RL_DUMPSTATS_START) == 0)
3944				break;
3945			DELAY(1000);
3946		}
3947		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3948		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3949		RL_UNLOCK(sc);
3950		if (i == 0) {
3951			device_printf(sc->rl_dev,
3952			    "DUMP statistics request timedout\n");
3953			return (ETIMEDOUT);
3954		}
3955done:
3956		stats = sc->rl_ldata.rl_stats;
3957		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3958		printf("Tx frames : %ju\n",
3959		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3960		printf("Rx frames : %ju\n",
3961		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3962		printf("Tx errors : %ju\n",
3963		    (uintmax_t)le64toh(stats->rl_tx_errs));
3964		printf("Rx errors : %u\n",
3965		    le32toh(stats->rl_rx_errs));
3966		printf("Rx missed frames : %u\n",
3967		    (uint32_t)le16toh(stats->rl_missed_pkts));
3968		printf("Rx frame alignment errs : %u\n",
3969		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3970		printf("Tx single collisions : %u\n",
3971		    le32toh(stats->rl_tx_onecoll));
3972		printf("Tx multiple collisions : %u\n",
3973		    le32toh(stats->rl_tx_multicolls));
3974		printf("Rx unicast frames : %ju\n",
3975		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3976		printf("Rx broadcast frames : %ju\n",
3977		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3978		printf("Rx multicast frames : %u\n",
3979		    le32toh(stats->rl_rx_mcasts));
3980		printf("Tx aborts : %u\n",
3981		    (uint32_t)le16toh(stats->rl_tx_aborts));
3982		printf("Tx underruns : %u\n",
3983		    (uint32_t)le16toh(stats->rl_rx_underruns));
3984	}
3985
3986	return (error);
3987}
3988
3989static int
3990sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3991{
3992	int error, value;
3993
3994	if (arg1 == NULL)
3995		return (EINVAL);
3996	value = *(int *)arg1;
3997	error = sysctl_handle_int(oidp, &value, 0, req);
3998	if (error || req->newptr == NULL)
3999		return (error);
4000	if (value < low || value > high)
4001		return (EINVAL);
4002	*(int *)arg1 = value;
4003
4004	return (0);
4005}
4006
4007static int
4008sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4009{
4010
4011	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4012	    RL_TIMER_MAX));
4013}
4014