if_re.c revision 262151
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: stable/10/sys/dev/re/if_re.c 262151 2014-02-18 05:01:04Z luigi $");
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 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/G 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 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_8106E, RL_8169, "8106E", RL_MTU },
227	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
228	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
229	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
230	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
231	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
232	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
233	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
234	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
235	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
236	{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
237	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
238	{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
239	{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
240	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
241	{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
242	{ 0, 0, NULL, 0 }
243};
244
245static int re_probe		(device_t);
246static int re_attach		(device_t);
247static int re_detach		(device_t);
248
249static int re_encap		(struct rl_softc *, struct mbuf **);
250
251static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
252static int re_allocmem		(device_t, struct rl_softc *);
253static __inline void re_discard_rxbuf
254				(struct rl_softc *, int);
255static int re_newbuf		(struct rl_softc *, int);
256static int re_jumbo_newbuf	(struct rl_softc *, int);
257static int re_rx_list_init	(struct rl_softc *);
258static int re_jrx_list_init	(struct rl_softc *);
259static int re_tx_list_init	(struct rl_softc *);
260#ifdef RE_FIXUP_RX
261static __inline void re_fixup_rx
262				(struct mbuf *);
263#endif
264static int re_rxeof		(struct rl_softc *, int *);
265static void re_txeof		(struct rl_softc *);
266#ifdef DEVICE_POLLING
267static int re_poll		(struct ifnet *, enum poll_cmd, int);
268static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
269#endif
270static int re_intr		(void *);
271static void re_intr_msi		(void *);
272static void re_tick		(void *);
273static void re_int_task		(void *, int);
274static void re_start		(struct ifnet *);
275static void re_start_locked	(struct ifnet *);
276static int re_ioctl		(struct ifnet *, u_long, caddr_t);
277static void re_init		(void *);
278static void re_init_locked	(struct rl_softc *);
279static void re_stop		(struct rl_softc *);
280static void re_watchdog		(struct rl_softc *);
281static int re_suspend		(device_t);
282static int re_resume		(device_t);
283static int re_shutdown		(device_t);
284static int re_ifmedia_upd	(struct ifnet *);
285static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
286
287static void re_eeprom_putbyte	(struct rl_softc *, int);
288static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
289static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
290static int re_gmii_readreg	(device_t, int, int);
291static int re_gmii_writereg	(device_t, int, int, int);
292
293static int re_miibus_readreg	(device_t, int, int);
294static int re_miibus_writereg	(device_t, int, int, int);
295static void re_miibus_statchg	(device_t);
296
297static void re_set_jumbo	(struct rl_softc *, int);
298static void re_set_rxmode		(struct rl_softc *);
299static void re_reset		(struct rl_softc *);
300static void re_setwol		(struct rl_softc *);
301static void re_clrwol		(struct rl_softc *);
302static void re_set_linkspeed	(struct rl_softc *);
303
304#ifdef DEV_NETMAP	/* see ixgbe.c for details */
305#include <dev/netmap/if_re_netmap.h>
306#endif /* !DEV_NETMAP */
307
308#ifdef RE_DIAG
309static int re_diag		(struct rl_softc *);
310#endif
311
312static void re_add_sysctls	(struct rl_softc *);
313static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
314static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
315static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
316
317static device_method_t re_methods[] = {
318	/* Device interface */
319	DEVMETHOD(device_probe,		re_probe),
320	DEVMETHOD(device_attach,	re_attach),
321	DEVMETHOD(device_detach,	re_detach),
322	DEVMETHOD(device_suspend,	re_suspend),
323	DEVMETHOD(device_resume,	re_resume),
324	DEVMETHOD(device_shutdown,	re_shutdown),
325
326	/* MII interface */
327	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
328	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
329	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
330
331	DEVMETHOD_END
332};
333
334static driver_t re_driver = {
335	"re",
336	re_methods,
337	sizeof(struct rl_softc)
338};
339
340static devclass_t re_devclass;
341
342DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
343DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
344
345#define EE_SET(x)					\
346	CSR_WRITE_1(sc, RL_EECMD,			\
347		CSR_READ_1(sc, RL_EECMD) | x)
348
349#define EE_CLR(x)					\
350	CSR_WRITE_1(sc, RL_EECMD,			\
351		CSR_READ_1(sc, RL_EECMD) & ~x)
352
353/*
354 * Send a read command and address to the EEPROM, check for ACK.
355 */
356static void
357re_eeprom_putbyte(struct rl_softc *sc, int addr)
358{
359	int			d, i;
360
361	d = addr | (RL_9346_READ << sc->rl_eewidth);
362
363	/*
364	 * Feed in each bit and strobe the clock.
365	 */
366
367	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
368		if (d & i) {
369			EE_SET(RL_EE_DATAIN);
370		} else {
371			EE_CLR(RL_EE_DATAIN);
372		}
373		DELAY(100);
374		EE_SET(RL_EE_CLK);
375		DELAY(150);
376		EE_CLR(RL_EE_CLK);
377		DELAY(100);
378	}
379}
380
381/*
382 * Read a word of data stored in the EEPROM at address 'addr.'
383 */
384static void
385re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
386{
387	int			i;
388	u_int16_t		word = 0;
389
390	/*
391	 * Send address of word we want to read.
392	 */
393	re_eeprom_putbyte(sc, addr);
394
395	/*
396	 * Start reading bits from EEPROM.
397	 */
398	for (i = 0x8000; i; i >>= 1) {
399		EE_SET(RL_EE_CLK);
400		DELAY(100);
401		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
402			word |= i;
403		EE_CLR(RL_EE_CLK);
404		DELAY(100);
405	}
406
407	*dest = word;
408}
409
410/*
411 * Read a sequence of words from the EEPROM.
412 */
413static void
414re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
415{
416	int			i;
417	u_int16_t		word = 0, *ptr;
418
419	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
420
421        DELAY(100);
422
423	for (i = 0; i < cnt; i++) {
424		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
425		re_eeprom_getword(sc, off + i, &word);
426		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
427		ptr = (u_int16_t *)(dest + (i * 2));
428                *ptr = word;
429	}
430
431	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
432}
433
434static int
435re_gmii_readreg(device_t dev, int phy, int reg)
436{
437	struct rl_softc		*sc;
438	u_int32_t		rval;
439	int			i;
440
441	sc = device_get_softc(dev);
442
443	/* Let the rgephy driver read the GMEDIASTAT register */
444
445	if (reg == RL_GMEDIASTAT) {
446		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
447		return (rval);
448	}
449
450	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
451
452	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
453		rval = CSR_READ_4(sc, RL_PHYAR);
454		if (rval & RL_PHYAR_BUSY)
455			break;
456		DELAY(25);
457	}
458
459	if (i == RL_PHY_TIMEOUT) {
460		device_printf(sc->rl_dev, "PHY read failed\n");
461		return (0);
462	}
463
464	/*
465	 * Controller requires a 20us delay to process next MDIO request.
466	 */
467	DELAY(20);
468
469	return (rval & RL_PHYAR_PHYDATA);
470}
471
472static int
473re_gmii_writereg(device_t dev, int phy, int reg, int data)
474{
475	struct rl_softc		*sc;
476	u_int32_t		rval;
477	int			i;
478
479	sc = device_get_softc(dev);
480
481	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
482	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
483
484	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
485		rval = CSR_READ_4(sc, RL_PHYAR);
486		if (!(rval & RL_PHYAR_BUSY))
487			break;
488		DELAY(25);
489	}
490
491	if (i == RL_PHY_TIMEOUT) {
492		device_printf(sc->rl_dev, "PHY write failed\n");
493		return (0);
494	}
495
496	/*
497	 * Controller requires a 20us delay to process next MDIO request.
498	 */
499	DELAY(20);
500
501	return (0);
502}
503
504static int
505re_miibus_readreg(device_t dev, int phy, int reg)
506{
507	struct rl_softc		*sc;
508	u_int16_t		rval = 0;
509	u_int16_t		re8139_reg = 0;
510
511	sc = device_get_softc(dev);
512
513	if (sc->rl_type == RL_8169) {
514		rval = re_gmii_readreg(dev, phy, reg);
515		return (rval);
516	}
517
518	switch (reg) {
519	case MII_BMCR:
520		re8139_reg = RL_BMCR;
521		break;
522	case MII_BMSR:
523		re8139_reg = RL_BMSR;
524		break;
525	case MII_ANAR:
526		re8139_reg = RL_ANAR;
527		break;
528	case MII_ANER:
529		re8139_reg = RL_ANER;
530		break;
531	case MII_ANLPAR:
532		re8139_reg = RL_LPAR;
533		break;
534	case MII_PHYIDR1:
535	case MII_PHYIDR2:
536		return (0);
537	/*
538	 * Allow the rlphy driver to read the media status
539	 * register. If we have a link partner which does not
540	 * support NWAY, this is the register which will tell
541	 * us the results of parallel detection.
542	 */
543	case RL_MEDIASTAT:
544		rval = CSR_READ_1(sc, RL_MEDIASTAT);
545		return (rval);
546	default:
547		device_printf(sc->rl_dev, "bad phy register\n");
548		return (0);
549	}
550	rval = CSR_READ_2(sc, re8139_reg);
551	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
552		/* 8139C+ has different bit layout. */
553		rval &= ~(BMCR_LOOP | BMCR_ISO);
554	}
555	return (rval);
556}
557
558static int
559re_miibus_writereg(device_t dev, int phy, int reg, int data)
560{
561	struct rl_softc		*sc;
562	u_int16_t		re8139_reg = 0;
563	int			rval = 0;
564
565	sc = device_get_softc(dev);
566
567	if (sc->rl_type == RL_8169) {
568		rval = re_gmii_writereg(dev, phy, reg, data);
569		return (rval);
570	}
571
572	switch (reg) {
573	case MII_BMCR:
574		re8139_reg = RL_BMCR;
575		if (sc->rl_type == RL_8139CPLUS) {
576			/* 8139C+ has different bit layout. */
577			data &= ~(BMCR_LOOP | BMCR_ISO);
578		}
579		break;
580	case MII_BMSR:
581		re8139_reg = RL_BMSR;
582		break;
583	case MII_ANAR:
584		re8139_reg = RL_ANAR;
585		break;
586	case MII_ANER:
587		re8139_reg = RL_ANER;
588		break;
589	case MII_ANLPAR:
590		re8139_reg = RL_LPAR;
591		break;
592	case MII_PHYIDR1:
593	case MII_PHYIDR2:
594		return (0);
595		break;
596	default:
597		device_printf(sc->rl_dev, "bad phy register\n");
598		return (0);
599	}
600	CSR_WRITE_2(sc, re8139_reg, data);
601	return (0);
602}
603
604static void
605re_miibus_statchg(device_t dev)
606{
607	struct rl_softc		*sc;
608	struct ifnet		*ifp;
609	struct mii_data		*mii;
610
611	sc = device_get_softc(dev);
612	mii = device_get_softc(sc->rl_miibus);
613	ifp = sc->rl_ifp;
614	if (mii == NULL || ifp == NULL ||
615	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
616		return;
617
618	sc->rl_flags &= ~RL_FLAG_LINK;
619	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
620	    (IFM_ACTIVE | IFM_AVALID)) {
621		switch (IFM_SUBTYPE(mii->mii_media_active)) {
622		case IFM_10_T:
623		case IFM_100_TX:
624			sc->rl_flags |= RL_FLAG_LINK;
625			break;
626		case IFM_1000_T:
627			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
628				break;
629			sc->rl_flags |= RL_FLAG_LINK;
630			break;
631		default:
632			break;
633		}
634	}
635	/*
636	 * RealTek controllers does not provide any interface to
637	 * Tx/Rx MACs for resolved speed, duplex and flow-control
638	 * parameters.
639	 */
640}
641
642/*
643 * Set the RX configuration and 64-bit multicast hash filter.
644 */
645static void
646re_set_rxmode(struct rl_softc *sc)
647{
648	struct ifnet		*ifp;
649	struct ifmultiaddr	*ifma;
650	uint32_t		hashes[2] = { 0, 0 };
651	uint32_t		h, rxfilt;
652
653	RL_LOCK_ASSERT(sc);
654
655	ifp = sc->rl_ifp;
656
657	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
658
659	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
660		if (ifp->if_flags & IFF_PROMISC)
661			rxfilt |= RL_RXCFG_RX_ALLPHYS;
662		/*
663		 * Unlike other hardwares, we have to explicitly set
664		 * RL_RXCFG_RX_MULTI to receive multicast frames in
665		 * promiscuous mode.
666		 */
667		rxfilt |= RL_RXCFG_RX_MULTI;
668		hashes[0] = hashes[1] = 0xffffffff;
669		goto done;
670	}
671
672	if_maddr_rlock(ifp);
673	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
674		if (ifma->ifma_addr->sa_family != AF_LINK)
675			continue;
676		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
677		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
678		if (h < 32)
679			hashes[0] |= (1 << h);
680		else
681			hashes[1] |= (1 << (h - 32));
682	}
683	if_maddr_runlock(ifp);
684
685	if (hashes[0] != 0 || hashes[1] != 0) {
686		/*
687		 * For some unfathomable reason, RealTek decided to
688		 * reverse the order of the multicast hash registers
689		 * in the PCI Express parts.  This means we have to
690		 * write the hash pattern in reverse order for those
691		 * devices.
692		 */
693		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
694			h = bswap32(hashes[0]);
695			hashes[0] = bswap32(hashes[1]);
696			hashes[1] = h;
697		}
698		rxfilt |= RL_RXCFG_RX_MULTI;
699	}
700
701done:
702	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
703	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
704	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
705}
706
707static void
708re_reset(struct rl_softc *sc)
709{
710	int			i;
711
712	RL_LOCK_ASSERT(sc);
713
714	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
715
716	for (i = 0; i < RL_TIMEOUT; i++) {
717		DELAY(10);
718		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
719			break;
720	}
721	if (i == RL_TIMEOUT)
722		device_printf(sc->rl_dev, "reset never completed!\n");
723
724	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
725		CSR_WRITE_1(sc, 0x82, 1);
726	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
727		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
728}
729
730#ifdef RE_DIAG
731
732/*
733 * The following routine is designed to test for a defect on some
734 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
735 * lines connected to the bus, however for a 32-bit only card, they
736 * should be pulled high. The result of this defect is that the
737 * NIC will not work right if you plug it into a 64-bit slot: DMA
738 * operations will be done with 64-bit transfers, which will fail
739 * because the 64-bit data lines aren't connected.
740 *
741 * There's no way to work around this (short of talking a soldering
742 * iron to the board), however we can detect it. The method we use
743 * here is to put the NIC into digital loopback mode, set the receiver
744 * to promiscuous mode, and then try to send a frame. We then compare
745 * the frame data we sent to what was received. If the data matches,
746 * then the NIC is working correctly, otherwise we know the user has
747 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
748 * slot. In the latter case, there's no way the NIC can work correctly,
749 * so we print out a message on the console and abort the device attach.
750 */
751
752static int
753re_diag(struct rl_softc *sc)
754{
755	struct ifnet		*ifp = sc->rl_ifp;
756	struct mbuf		*m0;
757	struct ether_header	*eh;
758	struct rl_desc		*cur_rx;
759	u_int16_t		status;
760	u_int32_t		rxstat;
761	int			total_len, i, error = 0, phyaddr;
762	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
763	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
764
765	/* Allocate a single mbuf */
766	MGETHDR(m0, M_NOWAIT, MT_DATA);
767	if (m0 == NULL)
768		return (ENOBUFS);
769
770	RL_LOCK(sc);
771
772	/*
773	 * Initialize the NIC in test mode. This sets the chip up
774	 * so that it can send and receive frames, but performs the
775	 * following special functions:
776	 * - Puts receiver in promiscuous mode
777	 * - Enables digital loopback mode
778	 * - Leaves interrupts turned off
779	 */
780
781	ifp->if_flags |= IFF_PROMISC;
782	sc->rl_testmode = 1;
783	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
784	re_init_locked(sc);
785	sc->rl_flags |= RL_FLAG_LINK;
786	if (sc->rl_type == RL_8169)
787		phyaddr = 1;
788	else
789		phyaddr = 0;
790
791	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
792	for (i = 0; i < RL_TIMEOUT; i++) {
793		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
794		if (!(status & BMCR_RESET))
795			break;
796	}
797
798	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
799	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
800
801	DELAY(100000);
802
803	/* Put some data in the mbuf */
804
805	eh = mtod(m0, struct ether_header *);
806	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
807	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
808	eh->ether_type = htons(ETHERTYPE_IP);
809	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
810
811	/*
812	 * Queue the packet, start transmission.
813	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
814	 */
815
816	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
817	RL_UNLOCK(sc);
818	/* XXX: re_diag must not be called when in ALTQ mode */
819	IF_HANDOFF(&ifp->if_snd, m0, ifp);
820	RL_LOCK(sc);
821	m0 = NULL;
822
823	/* Wait for it to propagate through the chip */
824
825	DELAY(100000);
826	for (i = 0; i < RL_TIMEOUT; i++) {
827		status = CSR_READ_2(sc, RL_ISR);
828		CSR_WRITE_2(sc, RL_ISR, status);
829		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
830		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
831			break;
832		DELAY(10);
833	}
834
835	if (i == RL_TIMEOUT) {
836		device_printf(sc->rl_dev,
837		    "diagnostic failed, failed to receive packet in"
838		    " loopback mode\n");
839		error = EIO;
840		goto done;
841	}
842
843	/*
844	 * The packet should have been dumped into the first
845	 * entry in the RX DMA ring. Grab it from there.
846	 */
847
848	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
849	    sc->rl_ldata.rl_rx_list_map,
850	    BUS_DMASYNC_POSTREAD);
851	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
852	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
853	    BUS_DMASYNC_POSTREAD);
854	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
855	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
856
857	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
858	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
859	eh = mtod(m0, struct ether_header *);
860
861	cur_rx = &sc->rl_ldata.rl_rx_list[0];
862	total_len = RL_RXBYTES(cur_rx);
863	rxstat = le32toh(cur_rx->rl_cmdstat);
864
865	if (total_len != ETHER_MIN_LEN) {
866		device_printf(sc->rl_dev,
867		    "diagnostic failed, received short packet\n");
868		error = EIO;
869		goto done;
870	}
871
872	/* Test that the received packet data matches what we sent. */
873
874	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
875	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
876	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
877		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
878		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
879		    dst, ":", src, ":", ETHERTYPE_IP);
880		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
881		    eh->ether_dhost, ":", eh->ether_shost, ":",
882		    ntohs(eh->ether_type));
883		device_printf(sc->rl_dev, "You may have a defective 32-bit "
884		    "NIC plugged into a 64-bit PCI slot.\n");
885		device_printf(sc->rl_dev, "Please re-install the NIC in a "
886		    "32-bit slot for proper operation.\n");
887		device_printf(sc->rl_dev, "Read the re(4) man page for more "
888		    "details.\n");
889		error = EIO;
890	}
891
892done:
893	/* Turn interface off, release resources */
894
895	sc->rl_testmode = 0;
896	sc->rl_flags &= ~RL_FLAG_LINK;
897	ifp->if_flags &= ~IFF_PROMISC;
898	re_stop(sc);
899	if (m0 != NULL)
900		m_freem(m0);
901
902	RL_UNLOCK(sc);
903
904	return (error);
905}
906
907#endif
908
909/*
910 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
911 * IDs against our list and return a device name if we find a match.
912 */
913static int
914re_probe(device_t dev)
915{
916	const struct rl_type	*t;
917	uint16_t		devid, vendor;
918	uint16_t		revid, sdevid;
919	int			i;
920
921	vendor = pci_get_vendor(dev);
922	devid = pci_get_device(dev);
923	revid = pci_get_revid(dev);
924	sdevid = pci_get_subdevice(dev);
925
926	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
927		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
928			/*
929			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
930			 * Rev. 2 is supported by sk(4).
931			 */
932			return (ENXIO);
933		}
934	}
935
936	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
937		if (revid != 0x20) {
938			/* 8139, let rl(4) take care of this device. */
939			return (ENXIO);
940		}
941	}
942
943	t = re_devs;
944	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
945		if (vendor == t->rl_vid && devid == t->rl_did) {
946			device_set_desc(dev, t->rl_name);
947			return (BUS_PROBE_DEFAULT);
948		}
949	}
950
951	return (ENXIO);
952}
953
954/*
955 * Map a single buffer address.
956 */
957
958static void
959re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
960{
961	bus_addr_t		*addr;
962
963	if (error)
964		return;
965
966	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
967	addr = arg;
968	*addr = segs->ds_addr;
969}
970
971static int
972re_allocmem(device_t dev, struct rl_softc *sc)
973{
974	bus_addr_t		lowaddr;
975	bus_size_t		rx_list_size, tx_list_size;
976	int			error;
977	int			i;
978
979	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
980	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
981
982	/*
983	 * Allocate the parent bus DMA tag appropriate for PCI.
984	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
985	 * register should be set. However some RealTek chips are known
986	 * to be buggy on DAC handling, therefore disable DAC by limiting
987	 * DMA address space to 32bit. PCIe variants of RealTek chips
988	 * may not have the limitation.
989	 */
990	lowaddr = BUS_SPACE_MAXADDR;
991	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
992		lowaddr = BUS_SPACE_MAXADDR_32BIT;
993	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
994	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
995	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
996	    NULL, NULL, &sc->rl_parent_tag);
997	if (error) {
998		device_printf(dev, "could not allocate parent DMA tag\n");
999		return (error);
1000	}
1001
1002	/*
1003	 * Allocate map for TX mbufs.
1004	 */
1005	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
1006	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
1007	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
1008	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1009	if (error) {
1010		device_printf(dev, "could not allocate TX DMA tag\n");
1011		return (error);
1012	}
1013
1014	/*
1015	 * Allocate map for RX mbufs.
1016	 */
1017
1018	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1019		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1020		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1021		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1022		    &sc->rl_ldata.rl_jrx_mtag);
1023		if (error) {
1024			device_printf(dev,
1025			    "could not allocate jumbo RX DMA tag\n");
1026			return (error);
1027		}
1028	}
1029	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1030	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1031	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1032	if (error) {
1033		device_printf(dev, "could not allocate RX DMA tag\n");
1034		return (error);
1035	}
1036
1037	/*
1038	 * Allocate map for TX descriptor list.
1039	 */
1040	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1041	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1042	    NULL, tx_list_size, 1, tx_list_size, 0,
1043	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1044	if (error) {
1045		device_printf(dev, "could not allocate TX DMA ring tag\n");
1046		return (error);
1047	}
1048
1049	/* Allocate DMA'able memory for the TX ring */
1050
1051	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1052	    (void **)&sc->rl_ldata.rl_tx_list,
1053	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1054	    &sc->rl_ldata.rl_tx_list_map);
1055	if (error) {
1056		device_printf(dev, "could not allocate TX DMA ring\n");
1057		return (error);
1058	}
1059
1060	/* Load the map for the TX ring. */
1061
1062	sc->rl_ldata.rl_tx_list_addr = 0;
1063	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1064	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1065	     tx_list_size, re_dma_map_addr,
1066	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1067	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1068		device_printf(dev, "could not load TX DMA ring\n");
1069		return (ENOMEM);
1070	}
1071
1072	/* Create DMA maps for TX buffers */
1073
1074	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1075		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1076		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1077		if (error) {
1078			device_printf(dev, "could not create DMA map for TX\n");
1079			return (error);
1080		}
1081	}
1082
1083	/*
1084	 * Allocate map for RX descriptor list.
1085	 */
1086	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1087	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1088	    NULL, rx_list_size, 1, rx_list_size, 0,
1089	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1090	if (error) {
1091		device_printf(dev, "could not create RX DMA ring tag\n");
1092		return (error);
1093	}
1094
1095	/* Allocate DMA'able memory for the RX ring */
1096
1097	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1098	    (void **)&sc->rl_ldata.rl_rx_list,
1099	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1100	    &sc->rl_ldata.rl_rx_list_map);
1101	if (error) {
1102		device_printf(dev, "could not allocate RX DMA ring\n");
1103		return (error);
1104	}
1105
1106	/* Load the map for the RX ring. */
1107
1108	sc->rl_ldata.rl_rx_list_addr = 0;
1109	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1110	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1111	     rx_list_size, re_dma_map_addr,
1112	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1113	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1114		device_printf(dev, "could not load RX DMA ring\n");
1115		return (ENOMEM);
1116	}
1117
1118	/* Create DMA maps for RX buffers */
1119
1120	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1121		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1122		    &sc->rl_ldata.rl_jrx_sparemap);
1123		if (error) {
1124			device_printf(dev,
1125			    "could not create spare DMA map for jumbo RX\n");
1126			return (error);
1127		}
1128		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1129			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1130			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1131			if (error) {
1132				device_printf(dev,
1133				    "could not create DMA map for jumbo RX\n");
1134				return (error);
1135			}
1136		}
1137	}
1138	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1139	    &sc->rl_ldata.rl_rx_sparemap);
1140	if (error) {
1141		device_printf(dev, "could not create spare DMA map for RX\n");
1142		return (error);
1143	}
1144	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1145		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1146		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1147		if (error) {
1148			device_printf(dev, "could not create DMA map for RX\n");
1149			return (error);
1150		}
1151	}
1152
1153	/* Create DMA map for statistics. */
1154	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1155	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1156	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1157	    &sc->rl_ldata.rl_stag);
1158	if (error) {
1159		device_printf(dev, "could not create statistics DMA tag\n");
1160		return (error);
1161	}
1162	/* Allocate DMA'able memory for statistics. */
1163	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1164	    (void **)&sc->rl_ldata.rl_stats,
1165	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1166	    &sc->rl_ldata.rl_smap);
1167	if (error) {
1168		device_printf(dev,
1169		    "could not allocate statistics DMA memory\n");
1170		return (error);
1171	}
1172	/* Load the map for statistics. */
1173	sc->rl_ldata.rl_stats_addr = 0;
1174	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1175	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1176	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1177	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1178		device_printf(dev, "could not load statistics DMA memory\n");
1179		return (ENOMEM);
1180	}
1181
1182	return (0);
1183}
1184
1185/*
1186 * Attach the interface. Allocate softc structures, do ifmedia
1187 * setup and ethernet/BPF attach.
1188 */
1189static int
1190re_attach(device_t dev)
1191{
1192	u_char			eaddr[ETHER_ADDR_LEN];
1193	u_int16_t		as[ETHER_ADDR_LEN / 2];
1194	struct rl_softc		*sc;
1195	struct ifnet		*ifp;
1196	const struct rl_hwrev	*hw_rev;
1197	u_int32_t		cap, ctl;
1198	int			hwrev;
1199	u_int16_t		devid, re_did = 0;
1200	int			error = 0, i, phy, rid;
1201	int			msic, msixc, reg;
1202	uint8_t			cfg;
1203
1204	sc = device_get_softc(dev);
1205	sc->rl_dev = dev;
1206
1207	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1208	    MTX_DEF);
1209	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1210
1211	/*
1212	 * Map control/status registers.
1213	 */
1214	pci_enable_busmaster(dev);
1215
1216	devid = pci_get_device(dev);
1217	/*
1218	 * Prefer memory space register mapping over IO space.
1219	 * Because RTL8169SC does not seem to work when memory mapping
1220	 * is used always activate io mapping.
1221	 */
1222	if (devid == RT_DEVICEID_8169SC)
1223		prefer_iomap = 1;
1224	if (prefer_iomap == 0) {
1225		sc->rl_res_id = PCIR_BAR(1);
1226		sc->rl_res_type = SYS_RES_MEMORY;
1227		/* RTL8168/8101E seems to use different BARs. */
1228		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1229			sc->rl_res_id = PCIR_BAR(2);
1230	} else {
1231		sc->rl_res_id = PCIR_BAR(0);
1232		sc->rl_res_type = SYS_RES_IOPORT;
1233	}
1234	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1235	    &sc->rl_res_id, RF_ACTIVE);
1236	if (sc->rl_res == NULL && prefer_iomap == 0) {
1237		sc->rl_res_id = PCIR_BAR(0);
1238		sc->rl_res_type = SYS_RES_IOPORT;
1239		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1240		    &sc->rl_res_id, RF_ACTIVE);
1241	}
1242	if (sc->rl_res == NULL) {
1243		device_printf(dev, "couldn't map ports/memory\n");
1244		error = ENXIO;
1245		goto fail;
1246	}
1247
1248	sc->rl_btag = rman_get_bustag(sc->rl_res);
1249	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1250
1251	msic = pci_msi_count(dev);
1252	msixc = pci_msix_count(dev);
1253	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1254		sc->rl_flags |= RL_FLAG_PCIE;
1255		sc->rl_expcap = reg;
1256	}
1257	if (bootverbose) {
1258		device_printf(dev, "MSI count : %d\n", msic);
1259		device_printf(dev, "MSI-X count : %d\n", msixc);
1260	}
1261	if (msix_disable > 0)
1262		msixc = 0;
1263	if (msi_disable > 0)
1264		msic = 0;
1265	/* Prefer MSI-X to MSI. */
1266	if (msixc > 0) {
1267		msixc = 1;
1268		rid = PCIR_BAR(4);
1269		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1270		    &rid, RF_ACTIVE);
1271		if (sc->rl_res_pba == NULL) {
1272			device_printf(sc->rl_dev,
1273			    "could not allocate MSI-X PBA resource\n");
1274		}
1275		if (sc->rl_res_pba != NULL &&
1276		    pci_alloc_msix(dev, &msixc) == 0) {
1277			if (msixc == 1) {
1278				device_printf(dev, "Using %d MSI-X message\n",
1279				    msixc);
1280				sc->rl_flags |= RL_FLAG_MSIX;
1281			} else
1282				pci_release_msi(dev);
1283		}
1284		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1285			if (sc->rl_res_pba != NULL)
1286				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1287				    sc->rl_res_pba);
1288			sc->rl_res_pba = NULL;
1289			msixc = 0;
1290		}
1291	}
1292	/* Prefer MSI to INTx. */
1293	if (msixc == 0 && msic > 0) {
1294		msic = 1;
1295		if (pci_alloc_msi(dev, &msic) == 0) {
1296			if (msic == RL_MSI_MESSAGES) {
1297				device_printf(dev, "Using %d MSI message\n",
1298				    msic);
1299				sc->rl_flags |= RL_FLAG_MSI;
1300				/* Explicitly set MSI enable bit. */
1301				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1302				cfg = CSR_READ_1(sc, RL_CFG2);
1303				cfg |= RL_CFG2_MSI;
1304				CSR_WRITE_1(sc, RL_CFG2, cfg);
1305				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1306			} else
1307				pci_release_msi(dev);
1308		}
1309		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1310			msic = 0;
1311	}
1312
1313	/* Allocate interrupt */
1314	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1315		rid = 0;
1316		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1317		    RF_SHAREABLE | RF_ACTIVE);
1318		if (sc->rl_irq[0] == NULL) {
1319			device_printf(dev, "couldn't allocate IRQ resources\n");
1320			error = ENXIO;
1321			goto fail;
1322		}
1323	} else {
1324		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1325			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1326			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1327			if (sc->rl_irq[i] == NULL) {
1328				device_printf(dev,
1329				    "couldn't allocate IRQ resources for "
1330				    "message %d\n", rid);
1331				error = ENXIO;
1332				goto fail;
1333			}
1334		}
1335	}
1336
1337	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1338		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1339		cfg = CSR_READ_1(sc, RL_CFG2);
1340		if ((cfg & RL_CFG2_MSI) != 0) {
1341			device_printf(dev, "turning off MSI enable bit.\n");
1342			cfg &= ~RL_CFG2_MSI;
1343			CSR_WRITE_1(sc, RL_CFG2, cfg);
1344		}
1345		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1346	}
1347
1348	/* Disable ASPM L0S/L1. */
1349	if (sc->rl_expcap != 0) {
1350		cap = pci_read_config(dev, sc->rl_expcap +
1351		    PCIER_LINK_CAP, 2);
1352		if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
1353			ctl = pci_read_config(dev, sc->rl_expcap +
1354			    PCIER_LINK_CTL, 2);
1355			if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
1356				ctl &= ~PCIEM_LINK_CTL_ASPMC;
1357				pci_write_config(dev, sc->rl_expcap +
1358				    PCIER_LINK_CTL, ctl, 2);
1359				device_printf(dev, "ASPM disabled\n");
1360			}
1361		} else
1362			device_printf(dev, "no ASPM capability\n");
1363	}
1364
1365	hw_rev = re_hwrevs;
1366	hwrev = CSR_READ_4(sc, RL_TXCFG);
1367	switch (hwrev & 0x70000000) {
1368	case 0x00000000:
1369	case 0x10000000:
1370		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1371		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1372		break;
1373	default:
1374		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1375		sc->rl_macrev = hwrev & 0x00700000;
1376		hwrev &= RL_TXCFG_HWREV;
1377		break;
1378	}
1379	device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
1380	while (hw_rev->rl_desc != NULL) {
1381		if (hw_rev->rl_rev == hwrev) {
1382			sc->rl_type = hw_rev->rl_type;
1383			sc->rl_hwrev = hw_rev;
1384			break;
1385		}
1386		hw_rev++;
1387	}
1388	if (hw_rev->rl_desc == NULL) {
1389		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1390		error = ENXIO;
1391		goto fail;
1392	}
1393
1394	switch (hw_rev->rl_rev) {
1395	case RL_HWREV_8139CPLUS:
1396		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1397		break;
1398	case RL_HWREV_8100E:
1399	case RL_HWREV_8101E:
1400		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1401		break;
1402	case RL_HWREV_8102E:
1403	case RL_HWREV_8102EL:
1404	case RL_HWREV_8102EL_SPIN1:
1405		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1406		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1407		    RL_FLAG_AUTOPAD;
1408		break;
1409	case RL_HWREV_8103E:
1410		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1411		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1412		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1413		break;
1414	case RL_HWREV_8401E:
1415	case RL_HWREV_8105E:
1416	case RL_HWREV_8105E_SPIN1:
1417	case RL_HWREV_8106E:
1418		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1419		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1420		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1421		break;
1422	case RL_HWREV_8402:
1423		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1424		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1425		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1426		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1427		break;
1428	case RL_HWREV_8168B_SPIN1:
1429	case RL_HWREV_8168B_SPIN2:
1430		sc->rl_flags |= RL_FLAG_WOLRXENB;
1431		/* FALLTHROUGH */
1432	case RL_HWREV_8168B_SPIN3:
1433		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1434		break;
1435	case RL_HWREV_8168C_SPIN2:
1436		sc->rl_flags |= RL_FLAG_MACSLEEP;
1437		/* FALLTHROUGH */
1438	case RL_HWREV_8168C:
1439		if (sc->rl_macrev == 0x00200000)
1440			sc->rl_flags |= RL_FLAG_MACSLEEP;
1441		/* FALLTHROUGH */
1442	case RL_HWREV_8168CP:
1443		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1444		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1445		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1446		break;
1447	case RL_HWREV_8168D:
1448		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1449		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1450		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1451		    RL_FLAG_WOL_MANLINK;
1452		break;
1453	case RL_HWREV_8168DP:
1454		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1455		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1456		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1457		break;
1458	case RL_HWREV_8168E:
1459		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1460		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1461		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1462		    RL_FLAG_WOL_MANLINK;
1463		break;
1464	case RL_HWREV_8168E_VL:
1465	case RL_HWREV_8168EP:
1466	case RL_HWREV_8168F:
1467	case RL_HWREV_8168G:
1468	case RL_HWREV_8411:
1469	case RL_HWREV_8411B:
1470		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1471		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1472		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1473		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1474		break;
1475	case RL_HWREV_8168GU:
1476		if (pci_get_device(dev) == RT_DEVICEID_8101E) {
1477			/* RTL8106EUS */
1478			sc->rl_flags |= RL_FLAG_FASTETHER;
1479		} else
1480			sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1481
1482		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1483		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1484		    RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ;
1485		break;
1486	case RL_HWREV_8169_8110SB:
1487	case RL_HWREV_8169_8110SBL:
1488	case RL_HWREV_8169_8110SC:
1489	case RL_HWREV_8169_8110SCE:
1490		sc->rl_flags |= RL_FLAG_PHYWAKE;
1491		/* FALLTHROUGH */
1492	case RL_HWREV_8169:
1493	case RL_HWREV_8169S:
1494	case RL_HWREV_8110S:
1495		sc->rl_flags |= RL_FLAG_MACRESET;
1496		break;
1497	default:
1498		break;
1499	}
1500
1501	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1502		sc->rl_cfg0 = RL_8139_CFG0;
1503		sc->rl_cfg1 = RL_8139_CFG1;
1504		sc->rl_cfg2 = 0;
1505		sc->rl_cfg3 = RL_8139_CFG3;
1506		sc->rl_cfg4 = RL_8139_CFG4;
1507		sc->rl_cfg5 = RL_8139_CFG5;
1508	} else {
1509		sc->rl_cfg0 = RL_CFG0;
1510		sc->rl_cfg1 = RL_CFG1;
1511		sc->rl_cfg2 = RL_CFG2;
1512		sc->rl_cfg3 = RL_CFG3;
1513		sc->rl_cfg4 = RL_CFG4;
1514		sc->rl_cfg5 = RL_CFG5;
1515	}
1516
1517	/* Reset the adapter. */
1518	RL_LOCK(sc);
1519	re_reset(sc);
1520	RL_UNLOCK(sc);
1521
1522	/* Enable PME. */
1523	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1524	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1525	cfg |= RL_CFG1_PME;
1526	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1527	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1528	cfg &= RL_CFG5_PME_STS;
1529	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1530	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1531
1532	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1533		/*
1534		 * XXX Should have a better way to extract station
1535		 * address from EEPROM.
1536		 */
1537		for (i = 0; i < ETHER_ADDR_LEN; i++)
1538			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1539	} else {
1540		sc->rl_eewidth = RL_9356_ADDR_LEN;
1541		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1542		if (re_did != 0x8129)
1543			sc->rl_eewidth = RL_9346_ADDR_LEN;
1544
1545		/*
1546		 * Get station address from the EEPROM.
1547		 */
1548		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1549		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1550			as[i] = le16toh(as[i]);
1551		bcopy(as, eaddr, ETHER_ADDR_LEN);
1552	}
1553
1554	if (sc->rl_type == RL_8169) {
1555		/* Set RX length mask and number of descriptors. */
1556		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1557		sc->rl_txstart = RL_GTXSTART;
1558		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1559		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1560	} else {
1561		/* Set RX length mask and number of descriptors. */
1562		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1563		sc->rl_txstart = RL_TXSTART;
1564		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1565		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1566	}
1567
1568	error = re_allocmem(dev, sc);
1569	if (error)
1570		goto fail;
1571	re_add_sysctls(sc);
1572
1573	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1574	if (ifp == NULL) {
1575		device_printf(dev, "can not if_alloc()\n");
1576		error = ENOSPC;
1577		goto fail;
1578	}
1579
1580	/* Take controller out of deep sleep mode. */
1581	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1582		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1583			CSR_WRITE_1(sc, RL_GPIO,
1584			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1585		else
1586			CSR_WRITE_1(sc, RL_GPIO,
1587			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1588	}
1589
1590	/* Take PHY out of power down mode. */
1591	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1592		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1593		if (hw_rev->rl_rev == RL_HWREV_8401E)
1594			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1595	}
1596	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1597		re_gmii_writereg(dev, 1, 0x1f, 0);
1598		re_gmii_writereg(dev, 1, 0x0e, 0);
1599	}
1600
1601	ifp->if_softc = sc;
1602	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1603	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1604	ifp->if_ioctl = re_ioctl;
1605	ifp->if_start = re_start;
1606	/*
1607	 * RTL8168/8111C generates wrong IP checksummed frame if the
1608	 * packet has IP options so disable TX IP checksum offloading.
1609	 */
1610	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1611	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
1612	    sc->rl_hwrev->rl_rev == RL_HWREV_8168CP)
1613		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1614	else
1615		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1616	ifp->if_hwassist |= CSUM_TSO;
1617	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1618	ifp->if_capenable = ifp->if_capabilities;
1619	ifp->if_init = re_init;
1620	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1621	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1622	IFQ_SET_READY(&ifp->if_snd);
1623
1624	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1625
1626#define	RE_PHYAD_INTERNAL	 0
1627
1628	/* Do MII setup. */
1629	phy = RE_PHYAD_INTERNAL;
1630	if (sc->rl_type == RL_8169)
1631		phy = 1;
1632	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1633	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1634	if (error != 0) {
1635		device_printf(dev, "attaching PHYs failed\n");
1636		goto fail;
1637	}
1638
1639	/*
1640	 * Call MI attach routine.
1641	 */
1642	ether_ifattach(ifp, eaddr);
1643
1644	/* VLAN capability setup */
1645	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1646	if (ifp->if_capabilities & IFCAP_HWCSUM)
1647		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1648	/* Enable WOL if PM is supported. */
1649	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1650		ifp->if_capabilities |= IFCAP_WOL;
1651	ifp->if_capenable = ifp->if_capabilities;
1652	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1653	/*
1654	 * Don't enable TSO by default.  It is known to generate
1655	 * corrupted TCP segments(bad TCP options) under certain
1656	 * circumstances.
1657	 */
1658	ifp->if_hwassist &= ~CSUM_TSO;
1659	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1660#ifdef DEVICE_POLLING
1661	ifp->if_capabilities |= IFCAP_POLLING;
1662#endif
1663	/*
1664	 * Tell the upper layer(s) we support long frames.
1665	 * Must appear after the call to ether_ifattach() because
1666	 * ether_ifattach() sets ifi_hdrlen to the default value.
1667	 */
1668	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1669
1670#ifdef DEV_NETMAP
1671	re_netmap_attach(sc);
1672#endif /* DEV_NETMAP */
1673#ifdef RE_DIAG
1674	/*
1675	 * Perform hardware diagnostic on the original RTL8169.
1676	 * Some 32-bit cards were incorrectly wired and would
1677	 * malfunction if plugged into a 64-bit slot.
1678	 */
1679
1680	if (hwrev == RL_HWREV_8169) {
1681		error = re_diag(sc);
1682		if (error) {
1683			device_printf(dev,
1684		    	"attach aborted due to hardware diag failure\n");
1685			ether_ifdetach(ifp);
1686			goto fail;
1687		}
1688	}
1689#endif
1690
1691#ifdef RE_TX_MODERATION
1692	intr_filter = 1;
1693#endif
1694	/* Hook interrupt last to avoid having to lock softc */
1695	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1696	    intr_filter == 0) {
1697		error = bus_setup_intr(dev, sc->rl_irq[0],
1698		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1699		    &sc->rl_intrhand[0]);
1700	} else {
1701		error = bus_setup_intr(dev, sc->rl_irq[0],
1702		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1703		    &sc->rl_intrhand[0]);
1704	}
1705	if (error) {
1706		device_printf(dev, "couldn't set up irq\n");
1707		ether_ifdetach(ifp);
1708	}
1709
1710fail:
1711
1712	if (error)
1713		re_detach(dev);
1714
1715	return (error);
1716}
1717
1718/*
1719 * Shutdown hardware and free up resources. This can be called any
1720 * time after the mutex has been initialized. It is called in both
1721 * the error case in attach and the normal detach case so it needs
1722 * to be careful about only freeing resources that have actually been
1723 * allocated.
1724 */
1725static int
1726re_detach(device_t dev)
1727{
1728	struct rl_softc		*sc;
1729	struct ifnet		*ifp;
1730	int			i, rid;
1731
1732	sc = device_get_softc(dev);
1733	ifp = sc->rl_ifp;
1734	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1735
1736	/* These should only be active if attach succeeded */
1737	if (device_is_attached(dev)) {
1738#ifdef DEVICE_POLLING
1739		if (ifp->if_capenable & IFCAP_POLLING)
1740			ether_poll_deregister(ifp);
1741#endif
1742		RL_LOCK(sc);
1743#if 0
1744		sc->suspended = 1;
1745#endif
1746		re_stop(sc);
1747		RL_UNLOCK(sc);
1748		callout_drain(&sc->rl_stat_callout);
1749		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1750		/*
1751		 * Force off the IFF_UP flag here, in case someone
1752		 * still had a BPF descriptor attached to this
1753		 * interface. If they do, ether_ifdetach() will cause
1754		 * the BPF code to try and clear the promisc mode
1755		 * flag, which will bubble down to re_ioctl(),
1756		 * which will try to call re_init() again. This will
1757		 * turn the NIC back on and restart the MII ticker,
1758		 * which will panic the system when the kernel tries
1759		 * to invoke the re_tick() function that isn't there
1760		 * anymore.
1761		 */
1762		ifp->if_flags &= ~IFF_UP;
1763		ether_ifdetach(ifp);
1764	}
1765	if (sc->rl_miibus)
1766		device_delete_child(dev, sc->rl_miibus);
1767	bus_generic_detach(dev);
1768
1769	/*
1770	 * The rest is resource deallocation, so we should already be
1771	 * stopped here.
1772	 */
1773
1774	if (sc->rl_intrhand[0] != NULL) {
1775		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1776		sc->rl_intrhand[0] = NULL;
1777	}
1778	if (ifp != NULL) {
1779#ifdef DEV_NETMAP
1780		netmap_detach(ifp);
1781#endif /* DEV_NETMAP */
1782		if_free(ifp);
1783	}
1784	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1785		rid = 0;
1786	else
1787		rid = 1;
1788	if (sc->rl_irq[0] != NULL) {
1789		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1790		sc->rl_irq[0] = NULL;
1791	}
1792	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1793		pci_release_msi(dev);
1794	if (sc->rl_res_pba) {
1795		rid = PCIR_BAR(4);
1796		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1797	}
1798	if (sc->rl_res)
1799		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1800		    sc->rl_res);
1801
1802	/* Unload and free the RX DMA ring memory and map */
1803
1804	if (sc->rl_ldata.rl_rx_list_tag) {
1805		if (sc->rl_ldata.rl_rx_list_map)
1806			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1807			    sc->rl_ldata.rl_rx_list_map);
1808		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1809			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1810			    sc->rl_ldata.rl_rx_list,
1811			    sc->rl_ldata.rl_rx_list_map);
1812		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1813	}
1814
1815	/* Unload and free the TX DMA ring memory and map */
1816
1817	if (sc->rl_ldata.rl_tx_list_tag) {
1818		if (sc->rl_ldata.rl_tx_list_map)
1819			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1820			    sc->rl_ldata.rl_tx_list_map);
1821		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1822			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1823			    sc->rl_ldata.rl_tx_list,
1824			    sc->rl_ldata.rl_tx_list_map);
1825		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1826	}
1827
1828	/* Destroy all the RX and TX buffer maps */
1829
1830	if (sc->rl_ldata.rl_tx_mtag) {
1831		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1832			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1833				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1834				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1835		}
1836		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1837	}
1838	if (sc->rl_ldata.rl_rx_mtag) {
1839		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1840			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1841				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1842				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1843		}
1844		if (sc->rl_ldata.rl_rx_sparemap)
1845			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1846			    sc->rl_ldata.rl_rx_sparemap);
1847		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1848	}
1849	if (sc->rl_ldata.rl_jrx_mtag) {
1850		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1851			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1852				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1853				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1854		}
1855		if (sc->rl_ldata.rl_jrx_sparemap)
1856			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1857			    sc->rl_ldata.rl_jrx_sparemap);
1858		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1859	}
1860	/* Unload and free the stats buffer and map */
1861
1862	if (sc->rl_ldata.rl_stag) {
1863		if (sc->rl_ldata.rl_smap)
1864			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1865			    sc->rl_ldata.rl_smap);
1866		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1867			bus_dmamem_free(sc->rl_ldata.rl_stag,
1868			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1869		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1870	}
1871
1872	if (sc->rl_parent_tag)
1873		bus_dma_tag_destroy(sc->rl_parent_tag);
1874
1875	mtx_destroy(&sc->rl_mtx);
1876
1877	return (0);
1878}
1879
1880static __inline void
1881re_discard_rxbuf(struct rl_softc *sc, int idx)
1882{
1883	struct rl_desc		*desc;
1884	struct rl_rxdesc	*rxd;
1885	uint32_t		cmdstat;
1886
1887	if (sc->rl_ifp->if_mtu > RL_MTU &&
1888	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1889		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1890	else
1891		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1892	desc = &sc->rl_ldata.rl_rx_list[idx];
1893	desc->rl_vlanctl = 0;
1894	cmdstat = rxd->rx_size;
1895	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1896		cmdstat |= RL_RDESC_CMD_EOR;
1897	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1898}
1899
1900static int
1901re_newbuf(struct rl_softc *sc, int idx)
1902{
1903	struct mbuf		*m;
1904	struct rl_rxdesc	*rxd;
1905	bus_dma_segment_t	segs[1];
1906	bus_dmamap_t		map;
1907	struct rl_desc		*desc;
1908	uint32_t		cmdstat;
1909	int			error, nsegs;
1910
1911	m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1912	if (m == NULL)
1913		return (ENOBUFS);
1914
1915	m->m_len = m->m_pkthdr.len = MCLBYTES;
1916#ifdef RE_FIXUP_RX
1917	/*
1918	 * This is part of an evil trick to deal with non-x86 platforms.
1919	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1920	 * boundaries, but that will hose non-x86 machines. To get around
1921	 * this, we leave some empty space at the start of each buffer
1922	 * and for non-x86 hosts, we copy the buffer back six bytes
1923	 * to achieve word alignment. This is slightly more efficient
1924	 * than allocating a new buffer, copying the contents, and
1925	 * discarding the old buffer.
1926	 */
1927	m_adj(m, RE_ETHER_ALIGN);
1928#endif
1929	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1930	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1931	if (error != 0) {
1932		m_freem(m);
1933		return (ENOBUFS);
1934	}
1935	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1936
1937	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1938	if (rxd->rx_m != NULL) {
1939		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1940		    BUS_DMASYNC_POSTREAD);
1941		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1942	}
1943
1944	rxd->rx_m = m;
1945	map = rxd->rx_dmamap;
1946	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1947	rxd->rx_size = segs[0].ds_len;
1948	sc->rl_ldata.rl_rx_sparemap = map;
1949	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1950	    BUS_DMASYNC_PREREAD);
1951
1952	desc = &sc->rl_ldata.rl_rx_list[idx];
1953	desc->rl_vlanctl = 0;
1954	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1955	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1956	cmdstat = segs[0].ds_len;
1957	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1958		cmdstat |= RL_RDESC_CMD_EOR;
1959	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1960
1961	return (0);
1962}
1963
1964static int
1965re_jumbo_newbuf(struct rl_softc *sc, int idx)
1966{
1967	struct mbuf		*m;
1968	struct rl_rxdesc	*rxd;
1969	bus_dma_segment_t	segs[1];
1970	bus_dmamap_t		map;
1971	struct rl_desc		*desc;
1972	uint32_t		cmdstat;
1973	int			error, nsegs;
1974
1975	m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1976	if (m == NULL)
1977		return (ENOBUFS);
1978	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1979#ifdef RE_FIXUP_RX
1980	m_adj(m, RE_ETHER_ALIGN);
1981#endif
1982	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1983	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1984	if (error != 0) {
1985		m_freem(m);
1986		return (ENOBUFS);
1987	}
1988	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1989
1990	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1991	if (rxd->rx_m != NULL) {
1992		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1993		    BUS_DMASYNC_POSTREAD);
1994		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1995	}
1996
1997	rxd->rx_m = m;
1998	map = rxd->rx_dmamap;
1999	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
2000	rxd->rx_size = segs[0].ds_len;
2001	sc->rl_ldata.rl_jrx_sparemap = map;
2002	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
2003	    BUS_DMASYNC_PREREAD);
2004
2005	desc = &sc->rl_ldata.rl_rx_list[idx];
2006	desc->rl_vlanctl = 0;
2007	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
2008	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
2009	cmdstat = segs[0].ds_len;
2010	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
2011		cmdstat |= RL_RDESC_CMD_EOR;
2012	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
2013
2014	return (0);
2015}
2016
2017#ifdef RE_FIXUP_RX
2018static __inline void
2019re_fixup_rx(struct mbuf *m)
2020{
2021	int                     i;
2022	uint16_t                *src, *dst;
2023
2024	src = mtod(m, uint16_t *);
2025	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
2026
2027	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
2028		*dst++ = *src++;
2029
2030	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
2031}
2032#endif
2033
2034static int
2035re_tx_list_init(struct rl_softc *sc)
2036{
2037	struct rl_desc		*desc;
2038	int			i;
2039
2040	RL_LOCK_ASSERT(sc);
2041
2042	bzero(sc->rl_ldata.rl_tx_list,
2043	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2044	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2045		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2046#ifdef DEV_NETMAP
2047	re_netmap_tx_init(sc);
2048#endif /* DEV_NETMAP */
2049	/* Set EOR. */
2050	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2051	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2052
2053	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2054	    sc->rl_ldata.rl_tx_list_map,
2055	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2056
2057	sc->rl_ldata.rl_tx_prodidx = 0;
2058	sc->rl_ldata.rl_tx_considx = 0;
2059	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2060
2061	return (0);
2062}
2063
2064static int
2065re_rx_list_init(struct rl_softc *sc)
2066{
2067	int			error, i;
2068
2069	bzero(sc->rl_ldata.rl_rx_list,
2070	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2071	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2072		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2073		if ((error = re_newbuf(sc, i)) != 0)
2074			return (error);
2075	}
2076#ifdef DEV_NETMAP
2077	re_netmap_rx_init(sc);
2078#endif /* DEV_NETMAP */
2079
2080	/* Flush the RX descriptors */
2081
2082	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2083	    sc->rl_ldata.rl_rx_list_map,
2084	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2085
2086	sc->rl_ldata.rl_rx_prodidx = 0;
2087	sc->rl_head = sc->rl_tail = NULL;
2088	sc->rl_int_rx_act = 0;
2089
2090	return (0);
2091}
2092
2093static int
2094re_jrx_list_init(struct rl_softc *sc)
2095{
2096	int			error, i;
2097
2098	bzero(sc->rl_ldata.rl_rx_list,
2099	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2100	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2101		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2102		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2103			return (error);
2104	}
2105
2106	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2107	    sc->rl_ldata.rl_rx_list_map,
2108	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2109
2110	sc->rl_ldata.rl_rx_prodidx = 0;
2111	sc->rl_head = sc->rl_tail = NULL;
2112	sc->rl_int_rx_act = 0;
2113
2114	return (0);
2115}
2116
2117/*
2118 * RX handler for C+ and 8169. For the gigE chips, we support
2119 * the reception of jumbo frames that have been fragmented
2120 * across multiple 2K mbuf cluster buffers.
2121 */
2122static int
2123re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2124{
2125	struct mbuf		*m;
2126	struct ifnet		*ifp;
2127	int			i, rxerr, total_len;
2128	struct rl_desc		*cur_rx;
2129	u_int32_t		rxstat, rxvlan;
2130	int			jumbo, maxpkt = 16, rx_npkts = 0;
2131
2132	RL_LOCK_ASSERT(sc);
2133
2134	ifp = sc->rl_ifp;
2135#ifdef DEV_NETMAP
2136	if (netmap_rx_irq(ifp, 0, &rx_npkts))
2137		return 0;
2138#endif /* DEV_NETMAP */
2139	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2140		jumbo = 1;
2141	else
2142		jumbo = 0;
2143
2144	/* Invalidate the descriptor memory */
2145
2146	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2147	    sc->rl_ldata.rl_rx_list_map,
2148	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2149
2150	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2151	    i = RL_RX_DESC_NXT(sc, i)) {
2152		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2153			break;
2154		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2155		rxstat = le32toh(cur_rx->rl_cmdstat);
2156		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2157			break;
2158		total_len = rxstat & sc->rl_rxlenmask;
2159		rxvlan = le32toh(cur_rx->rl_vlanctl);
2160		if (jumbo != 0)
2161			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2162		else
2163			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2164
2165		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2166		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2167		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2168			/*
2169			 * RTL8168C or later controllers do not
2170			 * support multi-fragment packet.
2171			 */
2172			re_discard_rxbuf(sc, i);
2173			continue;
2174		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2175			if (re_newbuf(sc, i) != 0) {
2176				/*
2177				 * If this is part of a multi-fragment packet,
2178				 * discard all the pieces.
2179				 */
2180				if (sc->rl_head != NULL) {
2181					m_freem(sc->rl_head);
2182					sc->rl_head = sc->rl_tail = NULL;
2183				}
2184				re_discard_rxbuf(sc, i);
2185				continue;
2186			}
2187			m->m_len = RE_RX_DESC_BUFLEN;
2188			if (sc->rl_head == NULL)
2189				sc->rl_head = sc->rl_tail = m;
2190			else {
2191				m->m_flags &= ~M_PKTHDR;
2192				sc->rl_tail->m_next = m;
2193				sc->rl_tail = m;
2194			}
2195			continue;
2196		}
2197
2198		/*
2199		 * NOTE: for the 8139C+, the frame length field
2200		 * is always 12 bits in size, but for the gigE chips,
2201		 * it is 13 bits (since the max RX frame length is 16K).
2202		 * Unfortunately, all 32 bits in the status word
2203		 * were already used, so to make room for the extra
2204		 * length bit, RealTek took out the 'frame alignment
2205		 * error' bit and shifted the other status bits
2206		 * over one slot. The OWN, EOR, FS and LS bits are
2207		 * still in the same places. We have already extracted
2208		 * the frame length and checked the OWN bit, so rather
2209		 * than using an alternate bit mapping, we shift the
2210		 * status bits one space to the right so we can evaluate
2211		 * them using the 8169 status as though it was in the
2212		 * same format as that of the 8139C+.
2213		 */
2214		if (sc->rl_type == RL_8169)
2215			rxstat >>= 1;
2216
2217		/*
2218		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2219		 * set, but if CRC is clear, it will still be a valid frame.
2220		 */
2221		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2222			rxerr = 1;
2223			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2224			    total_len > 8191 &&
2225			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2226				rxerr = 0;
2227			if (rxerr != 0) {
2228				ifp->if_ierrors++;
2229				/*
2230				 * If this is part of a multi-fragment packet,
2231				 * discard all the pieces.
2232				 */
2233				if (sc->rl_head != NULL) {
2234					m_freem(sc->rl_head);
2235					sc->rl_head = sc->rl_tail = NULL;
2236				}
2237				re_discard_rxbuf(sc, i);
2238				continue;
2239			}
2240		}
2241
2242		/*
2243		 * If allocating a replacement mbuf fails,
2244		 * reload the current one.
2245		 */
2246		if (jumbo != 0)
2247			rxerr = re_jumbo_newbuf(sc, i);
2248		else
2249			rxerr = re_newbuf(sc, i);
2250		if (rxerr != 0) {
2251			ifp->if_iqdrops++;
2252			if (sc->rl_head != NULL) {
2253				m_freem(sc->rl_head);
2254				sc->rl_head = sc->rl_tail = NULL;
2255			}
2256			re_discard_rxbuf(sc, i);
2257			continue;
2258		}
2259
2260		if (sc->rl_head != NULL) {
2261			if (jumbo != 0)
2262				m->m_len = total_len;
2263			else {
2264				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2265				if (m->m_len == 0)
2266					m->m_len = RE_RX_DESC_BUFLEN;
2267			}
2268			/*
2269			 * Special case: if there's 4 bytes or less
2270			 * in this buffer, the mbuf can be discarded:
2271			 * the last 4 bytes is the CRC, which we don't
2272			 * care about anyway.
2273			 */
2274			if (m->m_len <= ETHER_CRC_LEN) {
2275				sc->rl_tail->m_len -=
2276				    (ETHER_CRC_LEN - m->m_len);
2277				m_freem(m);
2278			} else {
2279				m->m_len -= ETHER_CRC_LEN;
2280				m->m_flags &= ~M_PKTHDR;
2281				sc->rl_tail->m_next = m;
2282			}
2283			m = sc->rl_head;
2284			sc->rl_head = sc->rl_tail = NULL;
2285			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2286		} else
2287			m->m_pkthdr.len = m->m_len =
2288			    (total_len - ETHER_CRC_LEN);
2289
2290#ifdef RE_FIXUP_RX
2291		re_fixup_rx(m);
2292#endif
2293		ifp->if_ipackets++;
2294		m->m_pkthdr.rcvif = ifp;
2295
2296		/* Do RX checksumming if enabled */
2297
2298		if (ifp->if_capenable & IFCAP_RXCSUM) {
2299			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2300				/* Check IP header checksum */
2301				if (rxstat & RL_RDESC_STAT_PROTOID)
2302					m->m_pkthdr.csum_flags |=
2303					    CSUM_IP_CHECKED;
2304				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2305					m->m_pkthdr.csum_flags |=
2306					    CSUM_IP_VALID;
2307
2308				/* Check TCP/UDP checksum */
2309				if ((RL_TCPPKT(rxstat) &&
2310				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2311				    (RL_UDPPKT(rxstat) &&
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			} else {
2318				/*
2319				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2320				 */
2321				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2322				    (rxvlan & RL_RDESC_IPV4))
2323					m->m_pkthdr.csum_flags |=
2324					    CSUM_IP_CHECKED;
2325				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2326				    (rxvlan & RL_RDESC_IPV4))
2327					m->m_pkthdr.csum_flags |=
2328					    CSUM_IP_VALID;
2329				if (((rxstat & RL_RDESC_STAT_TCP) &&
2330				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2331				    ((rxstat & RL_RDESC_STAT_UDP) &&
2332				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2333					m->m_pkthdr.csum_flags |=
2334						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2335					m->m_pkthdr.csum_data = 0xffff;
2336				}
2337			}
2338		}
2339		maxpkt--;
2340		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2341			m->m_pkthdr.ether_vtag =
2342			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2343			m->m_flags |= M_VLANTAG;
2344		}
2345		RL_UNLOCK(sc);
2346		(*ifp->if_input)(ifp, m);
2347		RL_LOCK(sc);
2348		rx_npkts++;
2349	}
2350
2351	/* Flush the RX DMA ring */
2352
2353	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2354	    sc->rl_ldata.rl_rx_list_map,
2355	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2356
2357	sc->rl_ldata.rl_rx_prodidx = i;
2358
2359	if (rx_npktsp != NULL)
2360		*rx_npktsp = rx_npkts;
2361	if (maxpkt)
2362		return (EAGAIN);
2363
2364	return (0);
2365}
2366
2367static void
2368re_txeof(struct rl_softc *sc)
2369{
2370	struct ifnet		*ifp;
2371	struct rl_txdesc	*txd;
2372	u_int32_t		txstat;
2373	int			cons;
2374
2375	cons = sc->rl_ldata.rl_tx_considx;
2376	if (cons == sc->rl_ldata.rl_tx_prodidx)
2377		return;
2378
2379	ifp = sc->rl_ifp;
2380#ifdef DEV_NETMAP
2381	if (netmap_tx_irq(ifp, 0))
2382		return;
2383#endif /* DEV_NETMAP */
2384	/* Invalidate the TX descriptor list */
2385	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2386	    sc->rl_ldata.rl_tx_list_map,
2387	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2388
2389	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2390	    cons = RL_TX_DESC_NXT(sc, cons)) {
2391		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2392		if (txstat & RL_TDESC_STAT_OWN)
2393			break;
2394		/*
2395		 * We only stash mbufs in the last descriptor
2396		 * in a fragment chain, which also happens to
2397		 * be the only place where the TX status bits
2398		 * are valid.
2399		 */
2400		if (txstat & RL_TDESC_CMD_EOF) {
2401			txd = &sc->rl_ldata.rl_tx_desc[cons];
2402			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2403			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2404			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2405			    txd->tx_dmamap);
2406			KASSERT(txd->tx_m != NULL,
2407			    ("%s: freeing NULL mbufs!", __func__));
2408			m_freem(txd->tx_m);
2409			txd->tx_m = NULL;
2410			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2411			    RL_TDESC_STAT_COLCNT))
2412				ifp->if_collisions++;
2413			if (txstat & RL_TDESC_STAT_TXERRSUM)
2414				ifp->if_oerrors++;
2415			else
2416				ifp->if_opackets++;
2417		}
2418		sc->rl_ldata.rl_tx_free++;
2419		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2420	}
2421	sc->rl_ldata.rl_tx_considx = cons;
2422
2423	/* No changes made to the TX ring, so no flush needed */
2424
2425	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2426#ifdef RE_TX_MODERATION
2427		/*
2428		 * If not all descriptors have been reaped yet, reload
2429		 * the timer so that we will eventually get another
2430		 * interrupt that will cause us to re-enter this routine.
2431		 * This is done in case the transmitter has gone idle.
2432		 */
2433		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2434#endif
2435	} else
2436		sc->rl_watchdog_timer = 0;
2437}
2438
2439static void
2440re_tick(void *xsc)
2441{
2442	struct rl_softc		*sc;
2443	struct mii_data		*mii;
2444
2445	sc = xsc;
2446
2447	RL_LOCK_ASSERT(sc);
2448
2449	mii = device_get_softc(sc->rl_miibus);
2450	mii_tick(mii);
2451	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2452		re_miibus_statchg(sc->rl_dev);
2453	/*
2454	 * Reclaim transmitted frames here. Technically it is not
2455	 * necessary to do here but it ensures periodic reclamation
2456	 * regardless of Tx completion interrupt which seems to be
2457	 * lost on PCIe based controllers under certain situations.
2458	 */
2459	re_txeof(sc);
2460	re_watchdog(sc);
2461	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2462}
2463
2464#ifdef DEVICE_POLLING
2465static int
2466re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2467{
2468	struct rl_softc *sc = ifp->if_softc;
2469	int rx_npkts = 0;
2470
2471	RL_LOCK(sc);
2472	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2473		rx_npkts = re_poll_locked(ifp, cmd, count);
2474	RL_UNLOCK(sc);
2475	return (rx_npkts);
2476}
2477
2478static int
2479re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2480{
2481	struct rl_softc *sc = ifp->if_softc;
2482	int rx_npkts;
2483
2484	RL_LOCK_ASSERT(sc);
2485
2486	sc->rxcycles = count;
2487	re_rxeof(sc, &rx_npkts);
2488	re_txeof(sc);
2489
2490	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2491		re_start_locked(ifp);
2492
2493	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2494		u_int16_t       status;
2495
2496		status = CSR_READ_2(sc, RL_ISR);
2497		if (status == 0xffff)
2498			return (rx_npkts);
2499		if (status)
2500			CSR_WRITE_2(sc, RL_ISR, status);
2501		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2502		    (sc->rl_flags & RL_FLAG_PCIE))
2503			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2504
2505		/*
2506		 * XXX check behaviour on receiver stalls.
2507		 */
2508
2509		if (status & RL_ISR_SYSTEM_ERR) {
2510			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2511			re_init_locked(sc);
2512		}
2513	}
2514	return (rx_npkts);
2515}
2516#endif /* DEVICE_POLLING */
2517
2518static int
2519re_intr(void *arg)
2520{
2521	struct rl_softc		*sc;
2522	uint16_t		status;
2523
2524	sc = arg;
2525
2526	status = CSR_READ_2(sc, RL_ISR);
2527	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2528                return (FILTER_STRAY);
2529	CSR_WRITE_2(sc, RL_IMR, 0);
2530
2531	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2532
2533	return (FILTER_HANDLED);
2534}
2535
2536static void
2537re_int_task(void *arg, int npending)
2538{
2539	struct rl_softc		*sc;
2540	struct ifnet		*ifp;
2541	u_int16_t		status;
2542	int			rval = 0;
2543
2544	sc = arg;
2545	ifp = sc->rl_ifp;
2546
2547	RL_LOCK(sc);
2548
2549	status = CSR_READ_2(sc, RL_ISR);
2550        CSR_WRITE_2(sc, RL_ISR, status);
2551
2552	if (sc->suspended ||
2553	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2554		RL_UNLOCK(sc);
2555		return;
2556	}
2557
2558#ifdef DEVICE_POLLING
2559	if  (ifp->if_capenable & IFCAP_POLLING) {
2560		RL_UNLOCK(sc);
2561		return;
2562	}
2563#endif
2564
2565	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2566		rval = re_rxeof(sc, NULL);
2567
2568	/*
2569	 * Some chips will ignore a second TX request issued
2570	 * while an existing transmission is in progress. If
2571	 * the transmitter goes idle but there are still
2572	 * packets waiting to be sent, we need to restart the
2573	 * channel here to flush them out. This only seems to
2574	 * be required with the PCIe devices.
2575	 */
2576	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2577	    (sc->rl_flags & RL_FLAG_PCIE))
2578		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2579	if (status & (
2580#ifdef RE_TX_MODERATION
2581	    RL_ISR_TIMEOUT_EXPIRED|
2582#else
2583	    RL_ISR_TX_OK|
2584#endif
2585	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2586		re_txeof(sc);
2587
2588	if (status & RL_ISR_SYSTEM_ERR) {
2589		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2590		re_init_locked(sc);
2591	}
2592
2593	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2594		re_start_locked(ifp);
2595
2596	RL_UNLOCK(sc);
2597
2598        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2599		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2600		return;
2601	}
2602
2603	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2604}
2605
2606static void
2607re_intr_msi(void *xsc)
2608{
2609	struct rl_softc		*sc;
2610	struct ifnet		*ifp;
2611	uint16_t		intrs, status;
2612
2613	sc = xsc;
2614	RL_LOCK(sc);
2615
2616	ifp = sc->rl_ifp;
2617#ifdef DEVICE_POLLING
2618	if (ifp->if_capenable & IFCAP_POLLING) {
2619		RL_UNLOCK(sc);
2620		return;
2621	}
2622#endif
2623	/* Disable interrupts. */
2624	CSR_WRITE_2(sc, RL_IMR, 0);
2625	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2626		RL_UNLOCK(sc);
2627		return;
2628	}
2629
2630	intrs = RL_INTRS_CPLUS;
2631	status = CSR_READ_2(sc, RL_ISR);
2632        CSR_WRITE_2(sc, RL_ISR, status);
2633	if (sc->rl_int_rx_act > 0) {
2634		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2635		    RL_ISR_RX_OVERRUN);
2636		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2637		    RL_ISR_RX_OVERRUN);
2638	}
2639
2640	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2641	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2642		re_rxeof(sc, NULL);
2643		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2644			if (sc->rl_int_rx_mod != 0 &&
2645			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2646			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2647				/* Rearm one-shot timer. */
2648				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2649				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2650				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2651				sc->rl_int_rx_act = 1;
2652			} else {
2653				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2654				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2655				sc->rl_int_rx_act = 0;
2656			}
2657		}
2658	}
2659
2660	/*
2661	 * Some chips will ignore a second TX request issued
2662	 * while an existing transmission is in progress. If
2663	 * the transmitter goes idle but there are still
2664	 * packets waiting to be sent, we need to restart the
2665	 * channel here to flush them out. This only seems to
2666	 * be required with the PCIe devices.
2667	 */
2668	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2669	    (sc->rl_flags & RL_FLAG_PCIE))
2670		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2671	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2672		re_txeof(sc);
2673
2674	if (status & RL_ISR_SYSTEM_ERR) {
2675		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2676		re_init_locked(sc);
2677	}
2678
2679	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2680		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2681			re_start_locked(ifp);
2682		CSR_WRITE_2(sc, RL_IMR, intrs);
2683	}
2684	RL_UNLOCK(sc);
2685}
2686
2687static int
2688re_encap(struct rl_softc *sc, struct mbuf **m_head)
2689{
2690	struct rl_txdesc	*txd, *txd_last;
2691	bus_dma_segment_t	segs[RL_NTXSEGS];
2692	bus_dmamap_t		map;
2693	struct mbuf		*m_new;
2694	struct rl_desc		*desc;
2695	int			nsegs, prod;
2696	int			i, error, ei, si;
2697	int			padlen;
2698	uint32_t		cmdstat, csum_flags, vlanctl;
2699
2700	RL_LOCK_ASSERT(sc);
2701	M_ASSERTPKTHDR((*m_head));
2702
2703	/*
2704	 * With some of the RealTek chips, using the checksum offload
2705	 * support in conjunction with the autopadding feature results
2706	 * in the transmission of corrupt frames. For example, if we
2707	 * need to send a really small IP fragment that's less than 60
2708	 * bytes in size, and IP header checksumming is enabled, the
2709	 * resulting ethernet frame that appears on the wire will
2710	 * have garbled payload. To work around this, if TX IP checksum
2711	 * offload is enabled, we always manually pad short frames out
2712	 * to the minimum ethernet frame size.
2713	 */
2714	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2715	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2716	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2717		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2718		if (M_WRITABLE(*m_head) == 0) {
2719			/* Get a writable copy. */
2720			m_new = m_dup(*m_head, M_NOWAIT);
2721			m_freem(*m_head);
2722			if (m_new == NULL) {
2723				*m_head = NULL;
2724				return (ENOBUFS);
2725			}
2726			*m_head = m_new;
2727		}
2728		if ((*m_head)->m_next != NULL ||
2729		    M_TRAILINGSPACE(*m_head) < padlen) {
2730			m_new = m_defrag(*m_head, M_NOWAIT);
2731			if (m_new == NULL) {
2732				m_freem(*m_head);
2733				*m_head = NULL;
2734				return (ENOBUFS);
2735			}
2736		} else
2737			m_new = *m_head;
2738
2739		/*
2740		 * Manually pad short frames, and zero the pad space
2741		 * to avoid leaking data.
2742		 */
2743		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2744		m_new->m_pkthdr.len += padlen;
2745		m_new->m_len = m_new->m_pkthdr.len;
2746		*m_head = m_new;
2747	}
2748
2749	prod = sc->rl_ldata.rl_tx_prodidx;
2750	txd = &sc->rl_ldata.rl_tx_desc[prod];
2751	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2752	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2753	if (error == EFBIG) {
2754		m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
2755		if (m_new == NULL) {
2756			m_freem(*m_head);
2757			*m_head = NULL;
2758			return (ENOBUFS);
2759		}
2760		*m_head = m_new;
2761		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2762		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2763		if (error != 0) {
2764			m_freem(*m_head);
2765			*m_head = NULL;
2766			return (error);
2767		}
2768	} else if (error != 0)
2769		return (error);
2770	if (nsegs == 0) {
2771		m_freem(*m_head);
2772		*m_head = NULL;
2773		return (EIO);
2774	}
2775
2776	/* Check for number of available descriptors. */
2777	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2778		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2779		return (ENOBUFS);
2780	}
2781
2782	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2783	    BUS_DMASYNC_PREWRITE);
2784
2785	/*
2786	 * Set up checksum offload. Note: checksum offload bits must
2787	 * appear in all descriptors of a multi-descriptor transmit
2788	 * attempt. This is according to testing done with an 8169
2789	 * chip. This is a requirement.
2790	 */
2791	vlanctl = 0;
2792	csum_flags = 0;
2793	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2794		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2795			csum_flags |= RL_TDESC_CMD_LGSEND;
2796			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2797			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2798		} else {
2799			csum_flags |= RL_TDESC_CMD_LGSEND |
2800			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2801			    RL_TDESC_CMD_MSSVAL_SHIFT);
2802		}
2803	} else {
2804		/*
2805		 * Unconditionally enable IP checksum if TCP or UDP
2806		 * checksum is required. Otherwise, TCP/UDP checksum
2807		 * doesn't make effects.
2808		 */
2809		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2810			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2811				csum_flags |= RL_TDESC_CMD_IPCSUM;
2812				if (((*m_head)->m_pkthdr.csum_flags &
2813				    CSUM_TCP) != 0)
2814					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2815				if (((*m_head)->m_pkthdr.csum_flags &
2816				    CSUM_UDP) != 0)
2817					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2818			} else {
2819				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2820				if (((*m_head)->m_pkthdr.csum_flags &
2821				    CSUM_TCP) != 0)
2822					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2823				if (((*m_head)->m_pkthdr.csum_flags &
2824				    CSUM_UDP) != 0)
2825					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2826			}
2827		}
2828	}
2829
2830	/*
2831	 * Set up hardware VLAN tagging. Note: vlan tag info must
2832	 * appear in all descriptors of a multi-descriptor
2833	 * transmission attempt.
2834	 */
2835	if ((*m_head)->m_flags & M_VLANTAG)
2836		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2837		    RL_TDESC_VLANCTL_TAG;
2838
2839	si = prod;
2840	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2841		desc = &sc->rl_ldata.rl_tx_list[prod];
2842		desc->rl_vlanctl = htole32(vlanctl);
2843		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2844		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2845		cmdstat = segs[i].ds_len;
2846		if (i != 0)
2847			cmdstat |= RL_TDESC_CMD_OWN;
2848		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2849			cmdstat |= RL_TDESC_CMD_EOR;
2850		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2851		sc->rl_ldata.rl_tx_free--;
2852	}
2853	/* Update producer index. */
2854	sc->rl_ldata.rl_tx_prodidx = prod;
2855
2856	/* Set EOF on the last descriptor. */
2857	ei = RL_TX_DESC_PRV(sc, prod);
2858	desc = &sc->rl_ldata.rl_tx_list[ei];
2859	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2860
2861	desc = &sc->rl_ldata.rl_tx_list[si];
2862	/* Set SOF and transfer ownership of packet to the chip. */
2863	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2864
2865	/*
2866	 * Insure that the map for this transmission
2867	 * is placed at the array index of the last descriptor
2868	 * in this chain.  (Swap last and first dmamaps.)
2869	 */
2870	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2871	map = txd->tx_dmamap;
2872	txd->tx_dmamap = txd_last->tx_dmamap;
2873	txd_last->tx_dmamap = map;
2874	txd_last->tx_m = *m_head;
2875
2876	return (0);
2877}
2878
2879static void
2880re_start(struct ifnet *ifp)
2881{
2882	struct rl_softc		*sc;
2883
2884	sc = ifp->if_softc;
2885	RL_LOCK(sc);
2886	re_start_locked(ifp);
2887	RL_UNLOCK(sc);
2888}
2889
2890/*
2891 * Main transmit routine for C+ and gigE NICs.
2892 */
2893static void
2894re_start_locked(struct ifnet *ifp)
2895{
2896	struct rl_softc		*sc;
2897	struct mbuf		*m_head;
2898	int			queued;
2899
2900	sc = ifp->if_softc;
2901
2902#ifdef DEV_NETMAP
2903	/* XXX is this necessary ? */
2904	if (ifp->if_capenable & IFCAP_NETMAP) {
2905		struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
2906		if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
2907			/* kick the tx unit */
2908			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2909#ifdef RE_TX_MODERATION
2910			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2911#endif
2912			sc->rl_watchdog_timer = 5;
2913		}
2914		return;
2915	}
2916#endif /* DEV_NETMAP */
2917	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2918	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2919		return;
2920
2921	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2922	    sc->rl_ldata.rl_tx_free > 1;) {
2923		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2924		if (m_head == NULL)
2925			break;
2926
2927		if (re_encap(sc, &m_head) != 0) {
2928			if (m_head == NULL)
2929				break;
2930			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2931			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2932			break;
2933		}
2934
2935		/*
2936		 * If there's a BPF listener, bounce a copy of this frame
2937		 * to him.
2938		 */
2939		ETHER_BPF_MTAP(ifp, m_head);
2940
2941		queued++;
2942	}
2943
2944	if (queued == 0) {
2945#ifdef RE_TX_MODERATION
2946		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2947			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2948#endif
2949		return;
2950	}
2951
2952	/* Flush the TX descriptors */
2953
2954	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2955	    sc->rl_ldata.rl_tx_list_map,
2956	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2957
2958	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2959
2960#ifdef RE_TX_MODERATION
2961	/*
2962	 * Use the countdown timer for interrupt moderation.
2963	 * 'TX done' interrupts are disabled. Instead, we reset the
2964	 * countdown timer, which will begin counting until it hits
2965	 * the value in the TIMERINT register, and then trigger an
2966	 * interrupt. Each time we write to the TIMERCNT register,
2967	 * the timer count is reset to 0.
2968	 */
2969	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2970#endif
2971
2972	/*
2973	 * Set a timeout in case the chip goes out to lunch.
2974	 */
2975	sc->rl_watchdog_timer = 5;
2976}
2977
2978static void
2979re_set_jumbo(struct rl_softc *sc, int jumbo)
2980{
2981
2982	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2983		pci_set_max_read_req(sc->rl_dev, 4096);
2984		return;
2985	}
2986
2987	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2988	if (jumbo != 0) {
2989		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
2990		    RL_CFG3_JUMBO_EN0);
2991		switch (sc->rl_hwrev->rl_rev) {
2992		case RL_HWREV_8168DP:
2993			break;
2994		case RL_HWREV_8168E:
2995			CSR_WRITE_1(sc, sc->rl_cfg4,
2996			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
2997			break;
2998		default:
2999			CSR_WRITE_1(sc, sc->rl_cfg4,
3000			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
3001		}
3002	} else {
3003		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
3004		    ~RL_CFG3_JUMBO_EN0);
3005		switch (sc->rl_hwrev->rl_rev) {
3006		case RL_HWREV_8168DP:
3007			break;
3008		case RL_HWREV_8168E:
3009			CSR_WRITE_1(sc, sc->rl_cfg4,
3010			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
3011			break;
3012		default:
3013			CSR_WRITE_1(sc, sc->rl_cfg4,
3014			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
3015		}
3016	}
3017	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3018
3019	switch (sc->rl_hwrev->rl_rev) {
3020	case RL_HWREV_8168DP:
3021		pci_set_max_read_req(sc->rl_dev, 4096);
3022		break;
3023	default:
3024		if (jumbo != 0)
3025			pci_set_max_read_req(sc->rl_dev, 512);
3026		else
3027			pci_set_max_read_req(sc->rl_dev, 4096);
3028	}
3029}
3030
3031static void
3032re_init(void *xsc)
3033{
3034	struct rl_softc		*sc = xsc;
3035
3036	RL_LOCK(sc);
3037	re_init_locked(sc);
3038	RL_UNLOCK(sc);
3039}
3040
3041static void
3042re_init_locked(struct rl_softc *sc)
3043{
3044	struct ifnet		*ifp = sc->rl_ifp;
3045	struct mii_data		*mii;
3046	uint32_t		reg;
3047	uint16_t		cfg;
3048	union {
3049		uint32_t align_dummy;
3050		u_char eaddr[ETHER_ADDR_LEN];
3051        } eaddr;
3052
3053	RL_LOCK_ASSERT(sc);
3054
3055	mii = device_get_softc(sc->rl_miibus);
3056
3057	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3058		return;
3059
3060	/*
3061	 * Cancel pending I/O and free all RX/TX buffers.
3062	 */
3063	re_stop(sc);
3064
3065	/* Put controller into known state. */
3066	re_reset(sc);
3067
3068	/*
3069	 * For C+ mode, initialize the RX descriptors and mbufs.
3070	 */
3071	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3072		if (ifp->if_mtu > RL_MTU) {
3073			if (re_jrx_list_init(sc) != 0) {
3074				device_printf(sc->rl_dev,
3075				    "no memory for jumbo RX buffers\n");
3076				re_stop(sc);
3077				return;
3078			}
3079			/* Disable checksum offloading for jumbo frames. */
3080			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3081			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3082		} else {
3083			if (re_rx_list_init(sc) != 0) {
3084				device_printf(sc->rl_dev,
3085				    "no memory for RX buffers\n");
3086				re_stop(sc);
3087				return;
3088			}
3089		}
3090		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3091	} else {
3092		if (re_rx_list_init(sc) != 0) {
3093			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3094			re_stop(sc);
3095			return;
3096		}
3097		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3098		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3099			if (ifp->if_mtu > RL_MTU)
3100				pci_set_max_read_req(sc->rl_dev, 512);
3101			else
3102				pci_set_max_read_req(sc->rl_dev, 4096);
3103		}
3104	}
3105	re_tx_list_init(sc);
3106
3107	/*
3108	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3109	 * RX checksum offload. We must configure the C+ register
3110	 * before all others.
3111	 */
3112	cfg = RL_CPLUSCMD_PCI_MRW;
3113	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3114		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3115	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3116		cfg |= RL_CPLUSCMD_VLANSTRIP;
3117	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3118		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3119		/* XXX magic. */
3120		cfg |= 0x0001;
3121	} else
3122		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3123	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3124	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3125	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3126		reg = 0x000fff00;
3127		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3128			reg |= 0x000000ff;
3129		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3130			reg |= 0x00f00000;
3131		CSR_WRITE_4(sc, 0x7c, reg);
3132		/* Disable interrupt mitigation. */
3133		CSR_WRITE_2(sc, 0xe2, 0);
3134	}
3135	/*
3136	 * Disable TSO if interface MTU size is greater than MSS
3137	 * allowed in controller.
3138	 */
3139	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3140		ifp->if_capenable &= ~IFCAP_TSO4;
3141		ifp->if_hwassist &= ~CSUM_TSO;
3142	}
3143
3144	/*
3145	 * Init our MAC address.  Even though the chipset
3146	 * documentation doesn't mention it, we need to enter "Config
3147	 * register write enable" mode to modify the ID registers.
3148	 */
3149	/* Copy MAC address on stack to align. */
3150	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3151	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3152	CSR_WRITE_4(sc, RL_IDR0,
3153	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3154	CSR_WRITE_4(sc, RL_IDR4,
3155	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3156	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3157
3158	/*
3159	 * Load the addresses of the RX and TX lists into the chip.
3160	 */
3161
3162	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3163	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3164	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3165	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3166
3167	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3168	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3169	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3170	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3171
3172	/*
3173	 * Enable transmit and receive.
3174	 */
3175	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3176
3177	/*
3178	 * Set the initial TX configuration.
3179	 */
3180	if (sc->rl_testmode) {
3181		if (sc->rl_type == RL_8169)
3182			CSR_WRITE_4(sc, RL_TXCFG,
3183			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3184		else
3185			CSR_WRITE_4(sc, RL_TXCFG,
3186			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3187	} else
3188		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3189
3190	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3191
3192	/*
3193	 * Set the initial RX configuration.
3194	 */
3195	re_set_rxmode(sc);
3196
3197	/* Configure interrupt moderation. */
3198	if (sc->rl_type == RL_8169) {
3199		/* Magic from vendor. */
3200		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3201	}
3202
3203#ifdef DEVICE_POLLING
3204	/*
3205	 * Disable interrupts if we are polling.
3206	 */
3207	if (ifp->if_capenable & IFCAP_POLLING)
3208		CSR_WRITE_2(sc, RL_IMR, 0);
3209	else	/* otherwise ... */
3210#endif
3211
3212	/*
3213	 * Enable interrupts.
3214	 */
3215	if (sc->rl_testmode)
3216		CSR_WRITE_2(sc, RL_IMR, 0);
3217	else
3218		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3219	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3220
3221	/* Set initial TX threshold */
3222	sc->rl_txthresh = RL_TX_THRESH_INIT;
3223
3224	/* Start RX/TX process. */
3225	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3226#ifdef notdef
3227	/* Enable receiver and transmitter. */
3228	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3229#endif
3230
3231	/*
3232	 * Initialize the timer interrupt register so that
3233	 * a timer interrupt will be generated once the timer
3234	 * reaches a certain number of ticks. The timer is
3235	 * reloaded on each transmit.
3236	 */
3237#ifdef RE_TX_MODERATION
3238	/*
3239	 * Use timer interrupt register to moderate TX interrupt
3240	 * moderation, which dramatically improves TX frame rate.
3241	 */
3242	if (sc->rl_type == RL_8169)
3243		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3244	else
3245		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3246#else
3247	/*
3248	 * Use timer interrupt register to moderate RX interrupt
3249	 * moderation.
3250	 */
3251	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3252	    intr_filter == 0) {
3253		if (sc->rl_type == RL_8169)
3254			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3255			    RL_USECS(sc->rl_int_rx_mod));
3256	} else {
3257		if (sc->rl_type == RL_8169)
3258			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3259	}
3260#endif
3261
3262	/*
3263	 * For 8169 gigE NICs, set the max allowed RX packet
3264	 * size so we can receive jumbo frames.
3265	 */
3266	if (sc->rl_type == RL_8169) {
3267		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3268			/*
3269			 * For controllers that use new jumbo frame scheme,
3270			 * set maximum size of jumbo frame depending on
3271			 * controller revisions.
3272			 */
3273			if (ifp->if_mtu > RL_MTU)
3274				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3275				    sc->rl_hwrev->rl_max_mtu +
3276				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3277				    ETHER_CRC_LEN);
3278			else
3279				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3280				    RE_RX_DESC_BUFLEN);
3281		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3282		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3283			/* RTL810x has no jumbo frame support. */
3284			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3285		} else
3286			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3287	}
3288
3289	if (sc->rl_testmode)
3290		return;
3291
3292	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3293	    RL_CFG1_DRVLOAD);
3294
3295	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3296	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3297
3298	sc->rl_flags &= ~RL_FLAG_LINK;
3299	mii_mediachg(mii);
3300
3301	sc->rl_watchdog_timer = 0;
3302	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3303}
3304
3305/*
3306 * Set media options.
3307 */
3308static int
3309re_ifmedia_upd(struct ifnet *ifp)
3310{
3311	struct rl_softc		*sc;
3312	struct mii_data		*mii;
3313	int			error;
3314
3315	sc = ifp->if_softc;
3316	mii = device_get_softc(sc->rl_miibus);
3317	RL_LOCK(sc);
3318	error = mii_mediachg(mii);
3319	RL_UNLOCK(sc);
3320
3321	return (error);
3322}
3323
3324/*
3325 * Report current media status.
3326 */
3327static void
3328re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3329{
3330	struct rl_softc		*sc;
3331	struct mii_data		*mii;
3332
3333	sc = ifp->if_softc;
3334	mii = device_get_softc(sc->rl_miibus);
3335
3336	RL_LOCK(sc);
3337	mii_pollstat(mii);
3338	ifmr->ifm_active = mii->mii_media_active;
3339	ifmr->ifm_status = mii->mii_media_status;
3340	RL_UNLOCK(sc);
3341}
3342
3343static int
3344re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3345{
3346	struct rl_softc		*sc = ifp->if_softc;
3347	struct ifreq		*ifr = (struct ifreq *) data;
3348	struct mii_data		*mii;
3349	uint32_t		rev;
3350	int			error = 0;
3351
3352	switch (command) {
3353	case SIOCSIFMTU:
3354		if (ifr->ifr_mtu < ETHERMIN ||
3355		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
3356		    ((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
3357		    ifr->ifr_mtu > RL_MTU)) {
3358			error = EINVAL;
3359			break;
3360		}
3361		RL_LOCK(sc);
3362		if (ifp->if_mtu != ifr->ifr_mtu) {
3363			ifp->if_mtu = ifr->ifr_mtu;
3364			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3365			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3366				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3367				re_init_locked(sc);
3368			}
3369			if (ifp->if_mtu > RL_TSO_MTU &&
3370			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3371				ifp->if_capenable &= ~(IFCAP_TSO4 |
3372				    IFCAP_VLAN_HWTSO);
3373				ifp->if_hwassist &= ~CSUM_TSO;
3374			}
3375			VLAN_CAPABILITIES(ifp);
3376		}
3377		RL_UNLOCK(sc);
3378		break;
3379	case SIOCSIFFLAGS:
3380		RL_LOCK(sc);
3381		if ((ifp->if_flags & IFF_UP) != 0) {
3382			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3383				if (((ifp->if_flags ^ sc->rl_if_flags)
3384				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3385					re_set_rxmode(sc);
3386			} else
3387				re_init_locked(sc);
3388		} else {
3389			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3390				re_stop(sc);
3391		}
3392		sc->rl_if_flags = ifp->if_flags;
3393		RL_UNLOCK(sc);
3394		break;
3395	case SIOCADDMULTI:
3396	case SIOCDELMULTI:
3397		RL_LOCK(sc);
3398		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3399			re_set_rxmode(sc);
3400		RL_UNLOCK(sc);
3401		break;
3402	case SIOCGIFMEDIA:
3403	case SIOCSIFMEDIA:
3404		mii = device_get_softc(sc->rl_miibus);
3405		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3406		break;
3407	case SIOCSIFCAP:
3408	    {
3409		int mask, reinit;
3410
3411		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3412		reinit = 0;
3413#ifdef DEVICE_POLLING
3414		if (mask & IFCAP_POLLING) {
3415			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3416				error = ether_poll_register(re_poll, ifp);
3417				if (error)
3418					return (error);
3419				RL_LOCK(sc);
3420				/* Disable interrupts */
3421				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3422				ifp->if_capenable |= IFCAP_POLLING;
3423				RL_UNLOCK(sc);
3424			} else {
3425				error = ether_poll_deregister(ifp);
3426				/* Enable interrupts. */
3427				RL_LOCK(sc);
3428				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3429				ifp->if_capenable &= ~IFCAP_POLLING;
3430				RL_UNLOCK(sc);
3431			}
3432		}
3433#endif /* DEVICE_POLLING */
3434		RL_LOCK(sc);
3435		if ((mask & IFCAP_TXCSUM) != 0 &&
3436		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3437			ifp->if_capenable ^= IFCAP_TXCSUM;
3438			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3439				rev = sc->rl_hwrev->rl_rev;
3440				if (rev == RL_HWREV_8168C ||
3441				    rev == RL_HWREV_8168C_SPIN2 ||
3442				    rev == RL_HWREV_8168CP)
3443					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3444				else
3445					ifp->if_hwassist |= RE_CSUM_FEATURES;
3446			} else
3447				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3448			reinit = 1;
3449		}
3450		if ((mask & IFCAP_RXCSUM) != 0 &&
3451		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3452			ifp->if_capenable ^= IFCAP_RXCSUM;
3453			reinit = 1;
3454		}
3455		if ((mask & IFCAP_TSO4) != 0 &&
3456		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
3457			ifp->if_capenable ^= IFCAP_TSO4;
3458			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3459				ifp->if_hwassist |= CSUM_TSO;
3460			else
3461				ifp->if_hwassist &= ~CSUM_TSO;
3462			if (ifp->if_mtu > RL_TSO_MTU &&
3463			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3464				ifp->if_capenable &= ~IFCAP_TSO4;
3465				ifp->if_hwassist &= ~CSUM_TSO;
3466			}
3467		}
3468		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3469		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3470			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3471		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3472		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3473			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3474			/* TSO over VLAN requires VLAN hardware tagging. */
3475			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3476				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3477			reinit = 1;
3478		}
3479		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3480		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3481		    IFCAP_VLAN_HWTSO)) != 0)
3482				reinit = 1;
3483		if ((mask & IFCAP_WOL) != 0 &&
3484		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3485			if ((mask & IFCAP_WOL_UCAST) != 0)
3486				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3487			if ((mask & IFCAP_WOL_MCAST) != 0)
3488				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3489			if ((mask & IFCAP_WOL_MAGIC) != 0)
3490				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3491		}
3492		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3493			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3494			re_init_locked(sc);
3495		}
3496		RL_UNLOCK(sc);
3497		VLAN_CAPABILITIES(ifp);
3498	    }
3499		break;
3500	default:
3501		error = ether_ioctl(ifp, command, data);
3502		break;
3503	}
3504
3505	return (error);
3506}
3507
3508static void
3509re_watchdog(struct rl_softc *sc)
3510{
3511	struct ifnet		*ifp;
3512
3513	RL_LOCK_ASSERT(sc);
3514
3515	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3516		return;
3517
3518	ifp = sc->rl_ifp;
3519	re_txeof(sc);
3520	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3521		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3522		    "-- recovering\n");
3523		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3524			re_start_locked(ifp);
3525		return;
3526	}
3527
3528	if_printf(ifp, "watchdog timeout\n");
3529	ifp->if_oerrors++;
3530
3531	re_rxeof(sc, NULL);
3532	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3533	re_init_locked(sc);
3534	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3535		re_start_locked(ifp);
3536}
3537
3538/*
3539 * Stop the adapter and free any mbufs allocated to the
3540 * RX and TX lists.
3541 */
3542static void
3543re_stop(struct rl_softc *sc)
3544{
3545	int			i;
3546	struct ifnet		*ifp;
3547	struct rl_txdesc	*txd;
3548	struct rl_rxdesc	*rxd;
3549
3550	RL_LOCK_ASSERT(sc);
3551
3552	ifp = sc->rl_ifp;
3553
3554	sc->rl_watchdog_timer = 0;
3555	callout_stop(&sc->rl_stat_callout);
3556	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3557
3558	/*
3559	 * Disable accepting frames to put RX MAC into idle state.
3560	 * Otherwise it's possible to get frames while stop command
3561	 * execution is in progress and controller can DMA the frame
3562	 * to already freed RX buffer during that period.
3563	 */
3564	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3565	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3566	    RL_RXCFG_RX_BROAD));
3567
3568	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3569		for (i = RL_TIMEOUT; i > 0; i--) {
3570			if ((CSR_READ_1(sc, sc->rl_txstart) &
3571			    RL_TXSTART_START) == 0)
3572				break;
3573			DELAY(20);
3574		}
3575		if (i == 0)
3576			device_printf(sc->rl_dev,
3577			    "stopping TX poll timed out!\n");
3578		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3579	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3580		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3581		    RL_CMD_RX_ENB);
3582		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3583			for (i = RL_TIMEOUT; i > 0; i--) {
3584				if ((CSR_READ_4(sc, RL_TXCFG) &
3585				    RL_TXCFG_QUEUE_EMPTY) != 0)
3586					break;
3587				DELAY(100);
3588			}
3589			if (i == 0)
3590				device_printf(sc->rl_dev,
3591				   "stopping TXQ timed out!\n");
3592		}
3593	} else
3594		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3595	DELAY(1000);
3596	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3597	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3598
3599	if (sc->rl_head != NULL) {
3600		m_freem(sc->rl_head);
3601		sc->rl_head = sc->rl_tail = NULL;
3602	}
3603
3604	/* Free the TX list buffers. */
3605	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3606		txd = &sc->rl_ldata.rl_tx_desc[i];
3607		if (txd->tx_m != NULL) {
3608			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3609			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3610			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3611			    txd->tx_dmamap);
3612			m_freem(txd->tx_m);
3613			txd->tx_m = NULL;
3614		}
3615	}
3616
3617	/* Free the RX list buffers. */
3618	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3619		rxd = &sc->rl_ldata.rl_rx_desc[i];
3620		if (rxd->rx_m != NULL) {
3621			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3622			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3623			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3624			    rxd->rx_dmamap);
3625			m_freem(rxd->rx_m);
3626			rxd->rx_m = NULL;
3627		}
3628	}
3629
3630	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3631		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3632			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3633			if (rxd->rx_m != NULL) {
3634				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3635				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3636				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3637				    rxd->rx_dmamap);
3638				m_freem(rxd->rx_m);
3639				rxd->rx_m = NULL;
3640			}
3641		}
3642	}
3643}
3644
3645/*
3646 * Device suspend routine.  Stop the interface and save some PCI
3647 * settings in case the BIOS doesn't restore them properly on
3648 * resume.
3649 */
3650static int
3651re_suspend(device_t dev)
3652{
3653	struct rl_softc		*sc;
3654
3655	sc = device_get_softc(dev);
3656
3657	RL_LOCK(sc);
3658	re_stop(sc);
3659	re_setwol(sc);
3660	sc->suspended = 1;
3661	RL_UNLOCK(sc);
3662
3663	return (0);
3664}
3665
3666/*
3667 * Device resume routine.  Restore some PCI settings in case the BIOS
3668 * doesn't, re-enable busmastering, and restart the interface if
3669 * appropriate.
3670 */
3671static int
3672re_resume(device_t dev)
3673{
3674	struct rl_softc		*sc;
3675	struct ifnet		*ifp;
3676
3677	sc = device_get_softc(dev);
3678
3679	RL_LOCK(sc);
3680
3681	ifp = sc->rl_ifp;
3682	/* Take controller out of sleep mode. */
3683	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3684		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3685			CSR_WRITE_1(sc, RL_GPIO,
3686			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3687	}
3688
3689	/*
3690	 * Clear WOL matching such that normal Rx filtering
3691	 * wouldn't interfere with WOL patterns.
3692	 */
3693	re_clrwol(sc);
3694
3695	/* reinitialize interface if necessary */
3696	if (ifp->if_flags & IFF_UP)
3697		re_init_locked(sc);
3698
3699	sc->suspended = 0;
3700	RL_UNLOCK(sc);
3701
3702	return (0);
3703}
3704
3705/*
3706 * Stop all chip I/O so that the kernel's probe routines don't
3707 * get confused by errant DMAs when rebooting.
3708 */
3709static int
3710re_shutdown(device_t dev)
3711{
3712	struct rl_softc		*sc;
3713
3714	sc = device_get_softc(dev);
3715
3716	RL_LOCK(sc);
3717	re_stop(sc);
3718	/*
3719	 * Mark interface as down since otherwise we will panic if
3720	 * interrupt comes in later on, which can happen in some
3721	 * cases.
3722	 */
3723	sc->rl_ifp->if_flags &= ~IFF_UP;
3724	re_setwol(sc);
3725	RL_UNLOCK(sc);
3726
3727	return (0);
3728}
3729
3730static void
3731re_set_linkspeed(struct rl_softc *sc)
3732{
3733	struct mii_softc *miisc;
3734	struct mii_data *mii;
3735	int aneg, i, phyno;
3736
3737	RL_LOCK_ASSERT(sc);
3738
3739	mii = device_get_softc(sc->rl_miibus);
3740	mii_pollstat(mii);
3741	aneg = 0;
3742	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3743	    (IFM_ACTIVE | IFM_AVALID)) {
3744		switch IFM_SUBTYPE(mii->mii_media_active) {
3745		case IFM_10_T:
3746		case IFM_100_TX:
3747			return;
3748		case IFM_1000_T:
3749			aneg++;
3750			break;
3751		default:
3752			break;
3753		}
3754	}
3755	miisc = LIST_FIRST(&mii->mii_phys);
3756	phyno = miisc->mii_phy;
3757	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3758		PHY_RESET(miisc);
3759	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3760	re_miibus_writereg(sc->rl_dev, phyno,
3761	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3762	re_miibus_writereg(sc->rl_dev, phyno,
3763	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3764	DELAY(1000);
3765	if (aneg != 0) {
3766		/*
3767		 * Poll link state until re(4) get a 10/100Mbps link.
3768		 */
3769		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3770			mii_pollstat(mii);
3771			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3772			    == (IFM_ACTIVE | IFM_AVALID)) {
3773				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3774				case IFM_10_T:
3775				case IFM_100_TX:
3776					return;
3777				default:
3778					break;
3779				}
3780			}
3781			RL_UNLOCK(sc);
3782			pause("relnk", hz);
3783			RL_LOCK(sc);
3784		}
3785		if (i == MII_ANEGTICKS_GIGE)
3786			device_printf(sc->rl_dev,
3787			    "establishing a link failed, WOL may not work!");
3788	}
3789	/*
3790	 * No link, force MAC to have 100Mbps, full-duplex link.
3791	 * MAC does not require reprogramming on resolved speed/duplex,
3792	 * so this is just for completeness.
3793	 */
3794	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3795	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3796}
3797
3798static void
3799re_setwol(struct rl_softc *sc)
3800{
3801	struct ifnet		*ifp;
3802	int			pmc;
3803	uint16_t		pmstat;
3804	uint8_t			v;
3805
3806	RL_LOCK_ASSERT(sc);
3807
3808	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3809		return;
3810
3811	ifp = sc->rl_ifp;
3812	/* Put controller into sleep mode. */
3813	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3814		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3815			CSR_WRITE_1(sc, RL_GPIO,
3816			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3817	}
3818	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3819		re_set_rxmode(sc);
3820		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3821			re_set_linkspeed(sc);
3822		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3823			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3824	}
3825	/* Enable config register write. */
3826	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3827
3828	/* Enable PME. */
3829	v = CSR_READ_1(sc, sc->rl_cfg1);
3830	v &= ~RL_CFG1_PME;
3831	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3832		v |= RL_CFG1_PME;
3833	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3834
3835	v = CSR_READ_1(sc, sc->rl_cfg3);
3836	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3837	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3838		v |= RL_CFG3_WOL_MAGIC;
3839	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3840
3841	v = CSR_READ_1(sc, sc->rl_cfg5);
3842	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3843	    RL_CFG5_WOL_LANWAKE);
3844	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3845		v |= RL_CFG5_WOL_UCAST;
3846	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3847		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3848	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3849		v |= RL_CFG5_WOL_LANWAKE;
3850	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3851
3852	/* Config register write done. */
3853	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3854
3855	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3856	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3857		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3858	/*
3859	 * It seems that hardware resets its link speed to 100Mbps in
3860	 * power down mode so switching to 100Mbps in driver is not
3861	 * needed.
3862	 */
3863
3864	/* Request PME if WOL is requested. */
3865	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3866	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3867	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3868		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3869	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3870}
3871
3872static void
3873re_clrwol(struct rl_softc *sc)
3874{
3875	int			pmc;
3876	uint8_t			v;
3877
3878	RL_LOCK_ASSERT(sc);
3879
3880	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3881		return;
3882
3883	/* Enable config register write. */
3884	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3885
3886	v = CSR_READ_1(sc, sc->rl_cfg3);
3887	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3888	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3889
3890	/* Config register write done. */
3891	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3892
3893	v = CSR_READ_1(sc, sc->rl_cfg5);
3894	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3895	v &= ~RL_CFG5_WOL_LANWAKE;
3896	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3897}
3898
3899static void
3900re_add_sysctls(struct rl_softc *sc)
3901{
3902	struct sysctl_ctx_list	*ctx;
3903	struct sysctl_oid_list	*children;
3904	int			error;
3905
3906	ctx = device_get_sysctl_ctx(sc->rl_dev);
3907	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3908
3909	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3910	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3911	    "Statistics Information");
3912	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3913		return;
3914
3915	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3916	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3917	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3918	/* Pull in device tunables. */
3919	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3920	error = resource_int_value(device_get_name(sc->rl_dev),
3921	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3922	if (error == 0) {
3923		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3924		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3925			device_printf(sc->rl_dev, "int_rx_mod value out of "
3926			    "range; using default: %d\n",
3927			    RL_TIMER_DEFAULT);
3928			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3929		}
3930	}
3931
3932}
3933
3934static int
3935re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3936{
3937	struct rl_softc		*sc;
3938	struct rl_stats		*stats;
3939	int			error, i, result;
3940
3941	result = -1;
3942	error = sysctl_handle_int(oidp, &result, 0, req);
3943	if (error || req->newptr == NULL)
3944		return (error);
3945
3946	if (result == 1) {
3947		sc = (struct rl_softc *)arg1;
3948		RL_LOCK(sc);
3949		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3950			RL_UNLOCK(sc);
3951			goto done;
3952		}
3953		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3954		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3955		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3956		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3957		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3958		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3959		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3960		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3961		    RL_DUMPSTATS_START));
3962		for (i = RL_TIMEOUT; i > 0; i--) {
3963			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3964			    RL_DUMPSTATS_START) == 0)
3965				break;
3966			DELAY(1000);
3967		}
3968		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3969		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3970		RL_UNLOCK(sc);
3971		if (i == 0) {
3972			device_printf(sc->rl_dev,
3973			    "DUMP statistics request timed out\n");
3974			return (ETIMEDOUT);
3975		}
3976done:
3977		stats = sc->rl_ldata.rl_stats;
3978		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3979		printf("Tx frames : %ju\n",
3980		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3981		printf("Rx frames : %ju\n",
3982		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3983		printf("Tx errors : %ju\n",
3984		    (uintmax_t)le64toh(stats->rl_tx_errs));
3985		printf("Rx errors : %u\n",
3986		    le32toh(stats->rl_rx_errs));
3987		printf("Rx missed frames : %u\n",
3988		    (uint32_t)le16toh(stats->rl_missed_pkts));
3989		printf("Rx frame alignment errs : %u\n",
3990		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3991		printf("Tx single collisions : %u\n",
3992		    le32toh(stats->rl_tx_onecoll));
3993		printf("Tx multiple collisions : %u\n",
3994		    le32toh(stats->rl_tx_multicolls));
3995		printf("Rx unicast frames : %ju\n",
3996		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3997		printf("Rx broadcast frames : %ju\n",
3998		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3999		printf("Rx multicast frames : %u\n",
4000		    le32toh(stats->rl_rx_mcasts));
4001		printf("Tx aborts : %u\n",
4002		    (uint32_t)le16toh(stats->rl_tx_aborts));
4003		printf("Tx underruns : %u\n",
4004		    (uint32_t)le16toh(stats->rl_rx_underruns));
4005	}
4006
4007	return (error);
4008}
4009
4010static int
4011sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
4012{
4013	int error, value;
4014
4015	if (arg1 == NULL)
4016		return (EINVAL);
4017	value = *(int *)arg1;
4018	error = sysctl_handle_int(oidp, &value, 0, req);
4019	if (error || req->newptr == NULL)
4020		return (error);
4021	if (value < low || value > high)
4022		return (EINVAL);
4023	*(int *)arg1 = value;
4024
4025	return (0);
4026}
4027
4028static int
4029sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
4030{
4031
4032	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
4033	    RL_TIMER_MAX));
4034}
4035