if_re.c revision 233492
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/9/sys/dev/re/if_re.c 233492 2012-03-26 04:27:01Z yongari $");
35
36/*
37 * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
38 *
39 * Written by Bill Paul <wpaul@windriver.com>
40 * Senior Networking Software Engineer
41 * Wind River Systems
42 */
43
44/*
45 * This driver is designed to support RealTek's next generation of
46 * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
47 * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
48 * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
49 *
50 * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
51 * with the older 8139 family, however it also supports a special
52 * C+ mode of operation that provides several new performance enhancing
53 * features. These include:
54 *
55 *	o Descriptor based DMA mechanism. Each descriptor represents
56 *	  a single packet fragment. Data buffers may be aligned on
57 *	  any byte boundary.
58 *
59 *	o 64-bit DMA
60 *
61 *	o TCP/IP checksum offload for both RX and TX
62 *
63 *	o High and normal priority transmit DMA rings
64 *
65 *	o VLAN tag insertion and extraction
66 *
67 *	o TCP large send (segmentation offload)
68 *
69 * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
70 * programming API is fairly straightforward. The RX filtering, EEPROM
71 * access and PHY access is the same as it is on the older 8139 series
72 * chips.
73 *
74 * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
75 * same programming API and feature set as the 8139C+ with the following
76 * differences and additions:
77 *
78 *	o 1000Mbps mode
79 *
80 *	o Jumbo frames
81 *
82 *	o GMII and TBI ports/registers for interfacing with copper
83 *	  or fiber PHYs
84 *
85 *	o RX and TX DMA rings can have up to 1024 descriptors
86 *	  (the 8139C+ allows a maximum of 64)
87 *
88 *	o Slight differences in register layout from the 8139C+
89 *
90 * The TX start and timer interrupt registers are at different locations
91 * on the 8169 than they are on the 8139C+. Also, the status word in the
92 * RX descriptor has a slightly different bit layout. The 8169 does not
93 * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
94 * copper gigE PHY.
95 *
96 * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
97 * (the 'S' stands for 'single-chip'). These devices have the same
98 * programming API as the older 8169, but also have some vendor-specific
99 * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
100 * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
101 *
102 * This driver takes advantage of the RX and TX checksum offload and
103 * VLAN tag insertion/extraction features. It also implements TX
104 * interrupt moderation using the timer interrupt registers, which
105 * significantly reduces TX interrupt load. There is also support
106 * for jumbo frames, however the 8169/8169S/8110S can not transmit
107 * jumbo frames larger than 7440, so the max MTU possible with this
108 * driver is 7422 bytes.
109 */
110
111#ifdef HAVE_KERNEL_OPTION_HEADERS
112#include "opt_device_polling.h"
113#endif
114
115#include <sys/param.h>
116#include <sys/endian.h>
117#include <sys/systm.h>
118#include <sys/sockio.h>
119#include <sys/mbuf.h>
120#include <sys/malloc.h>
121#include <sys/module.h>
122#include <sys/kernel.h>
123#include <sys/socket.h>
124#include <sys/lock.h>
125#include <sys/mutex.h>
126#include <sys/sysctl.h>
127#include <sys/taskqueue.h>
128
129#include <net/if.h>
130#include <net/if_arp.h>
131#include <net/ethernet.h>
132#include <net/if_dl.h>
133#include <net/if_media.h>
134#include <net/if_types.h>
135#include <net/if_vlan_var.h>
136
137#include <net/bpf.h>
138
139#include <machine/bus.h>
140#include <machine/resource.h>
141#include <sys/bus.h>
142#include <sys/rman.h>
143
144#include <dev/mii/mii.h>
145#include <dev/mii/miivar.h>
146
147#include <dev/pci/pcireg.h>
148#include <dev/pci/pcivar.h>
149
150#include <pci/if_rlreg.h>
151
152MODULE_DEPEND(re, pci, 1, 1, 1);
153MODULE_DEPEND(re, ether, 1, 1, 1);
154MODULE_DEPEND(re, miibus, 1, 1, 1);
155
156/* "device miibus" required.  See GENERIC if you get errors here. */
157#include "miibus_if.h"
158
159/* Tunables. */
160static int intr_filter = 0;
161TUNABLE_INT("hw.re.intr_filter", &intr_filter);
162static int msi_disable = 0;
163TUNABLE_INT("hw.re.msi_disable", &msi_disable);
164static int msix_disable = 0;
165TUNABLE_INT("hw.re.msix_disable", &msix_disable);
166static int prefer_iomap = 0;
167TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
168
169#define RE_CSUM_FEATURES    (CSUM_IP | CSUM_TCP | CSUM_UDP)
170
171/*
172 * Various supported device vendors/types and their names.
173 */
174static const struct rl_type const re_devs[] = {
175	{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
176	    "D-Link DGE-528(T) Gigabit Ethernet Adapter" },
177	{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
178	    "D-Link DGE-530(T) Gigabit Ethernet Adapter" },
179	{ RT_VENDORID, RT_DEVICEID_8139, 0,
180	    "RealTek 8139C+ 10/100BaseTX" },
181	{ RT_VENDORID, RT_DEVICEID_8101E, 0,
182	    "RealTek 810xE PCIe 10/100baseTX" },
183	{ RT_VENDORID, RT_DEVICEID_8168, 0,
184	    "RealTek 8168/8111 B/C/CP/D/DP/E/F PCIe Gigabit Ethernet" },
185	{ RT_VENDORID, RT_DEVICEID_8169, 0,
186	    "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
187	{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
188	    "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
189	{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
190	    "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
191	{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
192	    "Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
193	{ USR_VENDORID, USR_DEVICEID_997902, 0,
194	    "US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
195};
196
197static const struct rl_hwrev const re_hwrevs[] = {
198	{ RL_HWREV_8139, RL_8139, "", RL_MTU },
199	{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
200	{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
201	{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
202	{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
203	{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
204	{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
205	{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
206	{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
207	{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
208	{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
209	{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
210	{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
211	{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
212	{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
213	{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
214	{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
215	{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
216	{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
217	{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
218	{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
219	{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
220	{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
221	{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
222	{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
223	{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
224	{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
225	{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
226	{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
227	{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
228	{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
229	{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
230	{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
231	{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
232	{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
233	{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
234	{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
235	{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
236	{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
237	{ 0, 0, NULL, 0 }
238};
239
240static int re_probe		(device_t);
241static int re_attach		(device_t);
242static int re_detach		(device_t);
243
244static int re_encap		(struct rl_softc *, struct mbuf **);
245
246static void re_dma_map_addr	(void *, bus_dma_segment_t *, int, int);
247static int re_allocmem		(device_t, struct rl_softc *);
248static __inline void re_discard_rxbuf
249				(struct rl_softc *, int);
250static int re_newbuf		(struct rl_softc *, int);
251static int re_jumbo_newbuf	(struct rl_softc *, int);
252static int re_rx_list_init	(struct rl_softc *);
253static int re_jrx_list_init	(struct rl_softc *);
254static int re_tx_list_init	(struct rl_softc *);
255#ifdef RE_FIXUP_RX
256static __inline void re_fixup_rx
257				(struct mbuf *);
258#endif
259static int re_rxeof		(struct rl_softc *, int *);
260static void re_txeof		(struct rl_softc *);
261#ifdef DEVICE_POLLING
262static int re_poll		(struct ifnet *, enum poll_cmd, int);
263static int re_poll_locked	(struct ifnet *, enum poll_cmd, int);
264#endif
265static int re_intr		(void *);
266static void re_intr_msi		(void *);
267static void re_tick		(void *);
268static void re_int_task		(void *, int);
269static void re_start		(struct ifnet *);
270static void re_start_locked	(struct ifnet *);
271static int re_ioctl		(struct ifnet *, u_long, caddr_t);
272static void re_init		(void *);
273static void re_init_locked	(struct rl_softc *);
274static void re_stop		(struct rl_softc *);
275static void re_watchdog		(struct rl_softc *);
276static int re_suspend		(device_t);
277static int re_resume		(device_t);
278static int re_shutdown		(device_t);
279static int re_ifmedia_upd	(struct ifnet *);
280static void re_ifmedia_sts	(struct ifnet *, struct ifmediareq *);
281
282static void re_eeprom_putbyte	(struct rl_softc *, int);
283static void re_eeprom_getword	(struct rl_softc *, int, u_int16_t *);
284static void re_read_eeprom	(struct rl_softc *, caddr_t, int, int);
285static int re_gmii_readreg	(device_t, int, int);
286static int re_gmii_writereg	(device_t, int, int, int);
287
288static int re_miibus_readreg	(device_t, int, int);
289static int re_miibus_writereg	(device_t, int, int, int);
290static void re_miibus_statchg	(device_t);
291
292static void re_set_jumbo	(struct rl_softc *, int);
293static void re_set_rxmode		(struct rl_softc *);
294static void re_reset		(struct rl_softc *);
295static void re_setwol		(struct rl_softc *);
296static void re_clrwol		(struct rl_softc *);
297static void re_set_linkspeed	(struct rl_softc *);
298
299#ifdef RE_DIAG
300static int re_diag		(struct rl_softc *);
301#endif
302
303static void re_add_sysctls	(struct rl_softc *);
304static int re_sysctl_stats	(SYSCTL_HANDLER_ARGS);
305static int sysctl_int_range	(SYSCTL_HANDLER_ARGS, int, int);
306static int sysctl_hw_re_int_mod	(SYSCTL_HANDLER_ARGS);
307
308static device_method_t re_methods[] = {
309	/* Device interface */
310	DEVMETHOD(device_probe,		re_probe),
311	DEVMETHOD(device_attach,	re_attach),
312	DEVMETHOD(device_detach,	re_detach),
313	DEVMETHOD(device_suspend,	re_suspend),
314	DEVMETHOD(device_resume,	re_resume),
315	DEVMETHOD(device_shutdown,	re_shutdown),
316
317	/* MII interface */
318	DEVMETHOD(miibus_readreg,	re_miibus_readreg),
319	DEVMETHOD(miibus_writereg,	re_miibus_writereg),
320	DEVMETHOD(miibus_statchg,	re_miibus_statchg),
321
322	DEVMETHOD_END
323};
324
325static driver_t re_driver = {
326	"re",
327	re_methods,
328	sizeof(struct rl_softc)
329};
330
331static devclass_t re_devclass;
332
333DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
334DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
335
336#define EE_SET(x)					\
337	CSR_WRITE_1(sc, RL_EECMD,			\
338		CSR_READ_1(sc, RL_EECMD) | x)
339
340#define EE_CLR(x)					\
341	CSR_WRITE_1(sc, RL_EECMD,			\
342		CSR_READ_1(sc, RL_EECMD) & ~x)
343
344/*
345 * Send a read command and address to the EEPROM, check for ACK.
346 */
347static void
348re_eeprom_putbyte(struct rl_softc *sc, int addr)
349{
350	int			d, i;
351
352	d = addr | (RL_9346_READ << sc->rl_eewidth);
353
354	/*
355	 * Feed in each bit and strobe the clock.
356	 */
357
358	for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
359		if (d & i) {
360			EE_SET(RL_EE_DATAIN);
361		} else {
362			EE_CLR(RL_EE_DATAIN);
363		}
364		DELAY(100);
365		EE_SET(RL_EE_CLK);
366		DELAY(150);
367		EE_CLR(RL_EE_CLK);
368		DELAY(100);
369	}
370}
371
372/*
373 * Read a word of data stored in the EEPROM at address 'addr.'
374 */
375static void
376re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
377{
378	int			i;
379	u_int16_t		word = 0;
380
381	/*
382	 * Send address of word we want to read.
383	 */
384	re_eeprom_putbyte(sc, addr);
385
386	/*
387	 * Start reading bits from EEPROM.
388	 */
389	for (i = 0x8000; i; i >>= 1) {
390		EE_SET(RL_EE_CLK);
391		DELAY(100);
392		if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
393			word |= i;
394		EE_CLR(RL_EE_CLK);
395		DELAY(100);
396	}
397
398	*dest = word;
399}
400
401/*
402 * Read a sequence of words from the EEPROM.
403 */
404static void
405re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
406{
407	int			i;
408	u_int16_t		word = 0, *ptr;
409
410	CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
411
412        DELAY(100);
413
414	for (i = 0; i < cnt; i++) {
415		CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
416		re_eeprom_getword(sc, off + i, &word);
417		CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
418		ptr = (u_int16_t *)(dest + (i * 2));
419                *ptr = word;
420	}
421
422	CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
423}
424
425static int
426re_gmii_readreg(device_t dev, int phy, int reg)
427{
428	struct rl_softc		*sc;
429	u_int32_t		rval;
430	int			i;
431
432	sc = device_get_softc(dev);
433
434	/* Let the rgephy driver read the GMEDIASTAT register */
435
436	if (reg == RL_GMEDIASTAT) {
437		rval = CSR_READ_1(sc, RL_GMEDIASTAT);
438		return (rval);
439	}
440
441	CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
442
443	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
444		rval = CSR_READ_4(sc, RL_PHYAR);
445		if (rval & RL_PHYAR_BUSY)
446			break;
447		DELAY(25);
448	}
449
450	if (i == RL_PHY_TIMEOUT) {
451		device_printf(sc->rl_dev, "PHY read failed\n");
452		return (0);
453	}
454
455	/*
456	 * Controller requires a 20us delay to process next MDIO request.
457	 */
458	DELAY(20);
459
460	return (rval & RL_PHYAR_PHYDATA);
461}
462
463static int
464re_gmii_writereg(device_t dev, int phy, int reg, int data)
465{
466	struct rl_softc		*sc;
467	u_int32_t		rval;
468	int			i;
469
470	sc = device_get_softc(dev);
471
472	CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
473	    (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
474
475	for (i = 0; i < RL_PHY_TIMEOUT; i++) {
476		rval = CSR_READ_4(sc, RL_PHYAR);
477		if (!(rval & RL_PHYAR_BUSY))
478			break;
479		DELAY(25);
480	}
481
482	if (i == RL_PHY_TIMEOUT) {
483		device_printf(sc->rl_dev, "PHY write failed\n");
484		return (0);
485	}
486
487	/*
488	 * Controller requires a 20us delay to process next MDIO request.
489	 */
490	DELAY(20);
491
492	return (0);
493}
494
495static int
496re_miibus_readreg(device_t dev, int phy, int reg)
497{
498	struct rl_softc		*sc;
499	u_int16_t		rval = 0;
500	u_int16_t		re8139_reg = 0;
501
502	sc = device_get_softc(dev);
503
504	if (sc->rl_type == RL_8169) {
505		rval = re_gmii_readreg(dev, phy, reg);
506		return (rval);
507	}
508
509	switch (reg) {
510	case MII_BMCR:
511		re8139_reg = RL_BMCR;
512		break;
513	case MII_BMSR:
514		re8139_reg = RL_BMSR;
515		break;
516	case MII_ANAR:
517		re8139_reg = RL_ANAR;
518		break;
519	case MII_ANER:
520		re8139_reg = RL_ANER;
521		break;
522	case MII_ANLPAR:
523		re8139_reg = RL_LPAR;
524		break;
525	case MII_PHYIDR1:
526	case MII_PHYIDR2:
527		return (0);
528	/*
529	 * Allow the rlphy driver to read the media status
530	 * register. If we have a link partner which does not
531	 * support NWAY, this is the register which will tell
532	 * us the results of parallel detection.
533	 */
534	case RL_MEDIASTAT:
535		rval = CSR_READ_1(sc, RL_MEDIASTAT);
536		return (rval);
537	default:
538		device_printf(sc->rl_dev, "bad phy register\n");
539		return (0);
540	}
541	rval = CSR_READ_2(sc, re8139_reg);
542	if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
543		/* 8139C+ has different bit layout. */
544		rval &= ~(BMCR_LOOP | BMCR_ISO);
545	}
546	return (rval);
547}
548
549static int
550re_miibus_writereg(device_t dev, int phy, int reg, int data)
551{
552	struct rl_softc		*sc;
553	u_int16_t		re8139_reg = 0;
554	int			rval = 0;
555
556	sc = device_get_softc(dev);
557
558	if (sc->rl_type == RL_8169) {
559		rval = re_gmii_writereg(dev, phy, reg, data);
560		return (rval);
561	}
562
563	switch (reg) {
564	case MII_BMCR:
565		re8139_reg = RL_BMCR;
566		if (sc->rl_type == RL_8139CPLUS) {
567			/* 8139C+ has different bit layout. */
568			data &= ~(BMCR_LOOP | BMCR_ISO);
569		}
570		break;
571	case MII_BMSR:
572		re8139_reg = RL_BMSR;
573		break;
574	case MII_ANAR:
575		re8139_reg = RL_ANAR;
576		break;
577	case MII_ANER:
578		re8139_reg = RL_ANER;
579		break;
580	case MII_ANLPAR:
581		re8139_reg = RL_LPAR;
582		break;
583	case MII_PHYIDR1:
584	case MII_PHYIDR2:
585		return (0);
586		break;
587	default:
588		device_printf(sc->rl_dev, "bad phy register\n");
589		return (0);
590	}
591	CSR_WRITE_2(sc, re8139_reg, data);
592	return (0);
593}
594
595static void
596re_miibus_statchg(device_t dev)
597{
598	struct rl_softc		*sc;
599	struct ifnet		*ifp;
600	struct mii_data		*mii;
601
602	sc = device_get_softc(dev);
603	mii = device_get_softc(sc->rl_miibus);
604	ifp = sc->rl_ifp;
605	if (mii == NULL || ifp == NULL ||
606	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
607		return;
608
609	sc->rl_flags &= ~RL_FLAG_LINK;
610	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
611	    (IFM_ACTIVE | IFM_AVALID)) {
612		switch (IFM_SUBTYPE(mii->mii_media_active)) {
613		case IFM_10_T:
614		case IFM_100_TX:
615			sc->rl_flags |= RL_FLAG_LINK;
616			break;
617		case IFM_1000_T:
618			if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
619				break;
620			sc->rl_flags |= RL_FLAG_LINK;
621			break;
622		default:
623			break;
624		}
625	}
626	/*
627	 * RealTek controllers does not provide any interface to
628	 * Tx/Rx MACs for resolved speed, duplex and flow-control
629	 * parameters.
630	 */
631}
632
633/*
634 * Set the RX configuration and 64-bit multicast hash filter.
635 */
636static void
637re_set_rxmode(struct rl_softc *sc)
638{
639	struct ifnet		*ifp;
640	struct ifmultiaddr	*ifma;
641	uint32_t		hashes[2] = { 0, 0 };
642	uint32_t		h, rxfilt;
643
644	RL_LOCK_ASSERT(sc);
645
646	ifp = sc->rl_ifp;
647
648	rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
649
650	if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
651		if (ifp->if_flags & IFF_PROMISC)
652			rxfilt |= RL_RXCFG_RX_ALLPHYS;
653		/*
654		 * Unlike other hardwares, we have to explicitly set
655		 * RL_RXCFG_RX_MULTI to receive multicast frames in
656		 * promiscuous mode.
657		 */
658		rxfilt |= RL_RXCFG_RX_MULTI;
659		hashes[0] = hashes[1] = 0xffffffff;
660		goto done;
661	}
662
663	if_maddr_rlock(ifp);
664	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
665		if (ifma->ifma_addr->sa_family != AF_LINK)
666			continue;
667		h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
668		    ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
669		if (h < 32)
670			hashes[0] |= (1 << h);
671		else
672			hashes[1] |= (1 << (h - 32));
673	}
674	if_maddr_runlock(ifp);
675
676	if (hashes[0] != 0 || hashes[1] != 0) {
677		/*
678		 * For some unfathomable reason, RealTek decided to
679		 * reverse the order of the multicast hash registers
680		 * in the PCI Express parts.  This means we have to
681		 * write the hash pattern in reverse order for those
682		 * devices.
683		 */
684		if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
685			h = bswap32(hashes[0]);
686			hashes[0] = bswap32(hashes[1]);
687			hashes[1] = h;
688		}
689		rxfilt |= RL_RXCFG_RX_MULTI;
690	}
691
692done:
693	CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
694	CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
695	CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
696}
697
698static void
699re_reset(struct rl_softc *sc)
700{
701	int			i;
702
703	RL_LOCK_ASSERT(sc);
704
705	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
706
707	for (i = 0; i < RL_TIMEOUT; i++) {
708		DELAY(10);
709		if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
710			break;
711	}
712	if (i == RL_TIMEOUT)
713		device_printf(sc->rl_dev, "reset never completed!\n");
714
715	if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
716		CSR_WRITE_1(sc, 0x82, 1);
717	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
718		re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
719}
720
721#ifdef RE_DIAG
722
723/*
724 * The following routine is designed to test for a defect on some
725 * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
726 * lines connected to the bus, however for a 32-bit only card, they
727 * should be pulled high. The result of this defect is that the
728 * NIC will not work right if you plug it into a 64-bit slot: DMA
729 * operations will be done with 64-bit transfers, which will fail
730 * because the 64-bit data lines aren't connected.
731 *
732 * There's no way to work around this (short of talking a soldering
733 * iron to the board), however we can detect it. The method we use
734 * here is to put the NIC into digital loopback mode, set the receiver
735 * to promiscuous mode, and then try to send a frame. We then compare
736 * the frame data we sent to what was received. If the data matches,
737 * then the NIC is working correctly, otherwise we know the user has
738 * a defective NIC which has been mistakenly plugged into a 64-bit PCI
739 * slot. In the latter case, there's no way the NIC can work correctly,
740 * so we print out a message on the console and abort the device attach.
741 */
742
743static int
744re_diag(struct rl_softc *sc)
745{
746	struct ifnet		*ifp = sc->rl_ifp;
747	struct mbuf		*m0;
748	struct ether_header	*eh;
749	struct rl_desc		*cur_rx;
750	u_int16_t		status;
751	u_int32_t		rxstat;
752	int			total_len, i, error = 0, phyaddr;
753	u_int8_t		dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
754	u_int8_t		src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
755
756	/* Allocate a single mbuf */
757	MGETHDR(m0, M_DONTWAIT, MT_DATA);
758	if (m0 == NULL)
759		return (ENOBUFS);
760
761	RL_LOCK(sc);
762
763	/*
764	 * Initialize the NIC in test mode. This sets the chip up
765	 * so that it can send and receive frames, but performs the
766	 * following special functions:
767	 * - Puts receiver in promiscuous mode
768	 * - Enables digital loopback mode
769	 * - Leaves interrupts turned off
770	 */
771
772	ifp->if_flags |= IFF_PROMISC;
773	sc->rl_testmode = 1;
774	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
775	re_init_locked(sc);
776	sc->rl_flags |= RL_FLAG_LINK;
777	if (sc->rl_type == RL_8169)
778		phyaddr = 1;
779	else
780		phyaddr = 0;
781
782	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
783	for (i = 0; i < RL_TIMEOUT; i++) {
784		status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
785		if (!(status & BMCR_RESET))
786			break;
787	}
788
789	re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
790	CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
791
792	DELAY(100000);
793
794	/* Put some data in the mbuf */
795
796	eh = mtod(m0, struct ether_header *);
797	bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
798	bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
799	eh->ether_type = htons(ETHERTYPE_IP);
800	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
801
802	/*
803	 * Queue the packet, start transmission.
804	 * Note: IF_HANDOFF() ultimately calls re_start() for us.
805	 */
806
807	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
808	RL_UNLOCK(sc);
809	/* XXX: re_diag must not be called when in ALTQ mode */
810	IF_HANDOFF(&ifp->if_snd, m0, ifp);
811	RL_LOCK(sc);
812	m0 = NULL;
813
814	/* Wait for it to propagate through the chip */
815
816	DELAY(100000);
817	for (i = 0; i < RL_TIMEOUT; i++) {
818		status = CSR_READ_2(sc, RL_ISR);
819		CSR_WRITE_2(sc, RL_ISR, status);
820		if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
821		    (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
822			break;
823		DELAY(10);
824	}
825
826	if (i == RL_TIMEOUT) {
827		device_printf(sc->rl_dev,
828		    "diagnostic failed, failed to receive packet in"
829		    " loopback mode\n");
830		error = EIO;
831		goto done;
832	}
833
834	/*
835	 * The packet should have been dumped into the first
836	 * entry in the RX DMA ring. Grab it from there.
837	 */
838
839	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
840	    sc->rl_ldata.rl_rx_list_map,
841	    BUS_DMASYNC_POSTREAD);
842	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
843	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
844	    BUS_DMASYNC_POSTREAD);
845	bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
846	    sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
847
848	m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
849	sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
850	eh = mtod(m0, struct ether_header *);
851
852	cur_rx = &sc->rl_ldata.rl_rx_list[0];
853	total_len = RL_RXBYTES(cur_rx);
854	rxstat = le32toh(cur_rx->rl_cmdstat);
855
856	if (total_len != ETHER_MIN_LEN) {
857		device_printf(sc->rl_dev,
858		    "diagnostic failed, received short packet\n");
859		error = EIO;
860		goto done;
861	}
862
863	/* Test that the received packet data matches what we sent. */
864
865	if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
866	    bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
867	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
868		device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
869		device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
870		    dst, ":", src, ":", ETHERTYPE_IP);
871		device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
872		    eh->ether_dhost, ":", eh->ether_shost, ":",
873		    ntohs(eh->ether_type));
874		device_printf(sc->rl_dev, "You may have a defective 32-bit "
875		    "NIC plugged into a 64-bit PCI slot.\n");
876		device_printf(sc->rl_dev, "Please re-install the NIC in a "
877		    "32-bit slot for proper operation.\n");
878		device_printf(sc->rl_dev, "Read the re(4) man page for more "
879		    "details.\n");
880		error = EIO;
881	}
882
883done:
884	/* Turn interface off, release resources */
885
886	sc->rl_testmode = 0;
887	sc->rl_flags &= ~RL_FLAG_LINK;
888	ifp->if_flags &= ~IFF_PROMISC;
889	re_stop(sc);
890	if (m0 != NULL)
891		m_freem(m0);
892
893	RL_UNLOCK(sc);
894
895	return (error);
896}
897
898#endif
899
900/*
901 * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
902 * IDs against our list and return a device name if we find a match.
903 */
904static int
905re_probe(device_t dev)
906{
907	const struct rl_type	*t;
908	uint16_t		devid, vendor;
909	uint16_t		revid, sdevid;
910	int			i;
911
912	vendor = pci_get_vendor(dev);
913	devid = pci_get_device(dev);
914	revid = pci_get_revid(dev);
915	sdevid = pci_get_subdevice(dev);
916
917	if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
918		if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
919			/*
920			 * Only attach to rev. 3 of the Linksys EG1032 adapter.
921			 * Rev. 2 is supported by sk(4).
922			 */
923			return (ENXIO);
924		}
925	}
926
927	if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
928		if (revid != 0x20) {
929			/* 8139, let rl(4) take care of this device. */
930			return (ENXIO);
931		}
932	}
933
934	t = re_devs;
935	for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
936		if (vendor == t->rl_vid && devid == t->rl_did) {
937			device_set_desc(dev, t->rl_name);
938			return (BUS_PROBE_DEFAULT);
939		}
940	}
941
942	return (ENXIO);
943}
944
945/*
946 * Map a single buffer address.
947 */
948
949static void
950re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
951{
952	bus_addr_t		*addr;
953
954	if (error)
955		return;
956
957	KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
958	addr = arg;
959	*addr = segs->ds_addr;
960}
961
962static int
963re_allocmem(device_t dev, struct rl_softc *sc)
964{
965	bus_addr_t		lowaddr;
966	bus_size_t		rx_list_size, tx_list_size;
967	int			error;
968	int			i;
969
970	rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
971	tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
972
973	/*
974	 * Allocate the parent bus DMA tag appropriate for PCI.
975	 * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
976	 * register should be set. However some RealTek chips are known
977	 * to be buggy on DAC handling, therefore disable DAC by limiting
978	 * DMA address space to 32bit. PCIe variants of RealTek chips
979	 * may not have the limitation.
980	 */
981	lowaddr = BUS_SPACE_MAXADDR;
982	if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
983		lowaddr = BUS_SPACE_MAXADDR_32BIT;
984	error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
985	    lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
986	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
987	    NULL, NULL, &sc->rl_parent_tag);
988	if (error) {
989		device_printf(dev, "could not allocate parent DMA tag\n");
990		return (error);
991	}
992
993	/*
994	 * Allocate map for TX mbufs.
995	 */
996	error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
997	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
998	    NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
999	    NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
1000	if (error) {
1001		device_printf(dev, "could not allocate TX DMA tag\n");
1002		return (error);
1003	}
1004
1005	/*
1006	 * Allocate map for RX mbufs.
1007	 */
1008
1009	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1010		error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
1011		    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1012		    MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
1013		    &sc->rl_ldata.rl_jrx_mtag);
1014		if (error) {
1015			device_printf(dev,
1016			    "could not allocate jumbo RX DMA tag\n");
1017			return (error);
1018		}
1019	}
1020	error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
1021	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1022	    MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
1023	if (error) {
1024		device_printf(dev, "could not allocate RX DMA tag\n");
1025		return (error);
1026	}
1027
1028	/*
1029	 * Allocate map for TX descriptor list.
1030	 */
1031	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1032	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1033	    NULL, tx_list_size, 1, tx_list_size, 0,
1034	    NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
1035	if (error) {
1036		device_printf(dev, "could not allocate TX DMA ring tag\n");
1037		return (error);
1038	}
1039
1040	/* Allocate DMA'able memory for the TX ring */
1041
1042	error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
1043	    (void **)&sc->rl_ldata.rl_tx_list,
1044	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1045	    &sc->rl_ldata.rl_tx_list_map);
1046	if (error) {
1047		device_printf(dev, "could not allocate TX DMA ring\n");
1048		return (error);
1049	}
1050
1051	/* Load the map for the TX ring. */
1052
1053	sc->rl_ldata.rl_tx_list_addr = 0;
1054	error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
1055	     sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
1056	     tx_list_size, re_dma_map_addr,
1057	     &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
1058	if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
1059		device_printf(dev, "could not load TX DMA ring\n");
1060		return (ENOMEM);
1061	}
1062
1063	/* Create DMA maps for TX buffers */
1064
1065	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1066		error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
1067		    &sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1068		if (error) {
1069			device_printf(dev, "could not create DMA map for TX\n");
1070			return (error);
1071		}
1072	}
1073
1074	/*
1075	 * Allocate map for RX descriptor list.
1076	 */
1077	error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
1078	    0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
1079	    NULL, rx_list_size, 1, rx_list_size, 0,
1080	    NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
1081	if (error) {
1082		device_printf(dev, "could not create RX DMA ring tag\n");
1083		return (error);
1084	}
1085
1086	/* Allocate DMA'able memory for the RX ring */
1087
1088	error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
1089	    (void **)&sc->rl_ldata.rl_rx_list,
1090	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1091	    &sc->rl_ldata.rl_rx_list_map);
1092	if (error) {
1093		device_printf(dev, "could not allocate RX DMA ring\n");
1094		return (error);
1095	}
1096
1097	/* Load the map for the RX ring. */
1098
1099	sc->rl_ldata.rl_rx_list_addr = 0;
1100	error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
1101	     sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
1102	     rx_list_size, re_dma_map_addr,
1103	     &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
1104	if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
1105		device_printf(dev, "could not load RX DMA ring\n");
1106		return (ENOMEM);
1107	}
1108
1109	/* Create DMA maps for RX buffers */
1110
1111	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
1112		error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1113		    &sc->rl_ldata.rl_jrx_sparemap);
1114		if (error) {
1115			device_printf(dev,
1116			    "could not create spare DMA map for jumbo RX\n");
1117			return (error);
1118		}
1119		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1120			error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
1121			    &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1122			if (error) {
1123				device_printf(dev,
1124				    "could not create DMA map for jumbo RX\n");
1125				return (error);
1126			}
1127		}
1128	}
1129	error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1130	    &sc->rl_ldata.rl_rx_sparemap);
1131	if (error) {
1132		device_printf(dev, "could not create spare DMA map for RX\n");
1133		return (error);
1134	}
1135	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1136		error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
1137		    &sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1138		if (error) {
1139			device_printf(dev, "could not create DMA map for RX\n");
1140			return (error);
1141		}
1142	}
1143
1144	/* Create DMA map for statistics. */
1145	error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
1146	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1147	    sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
1148	    &sc->rl_ldata.rl_stag);
1149	if (error) {
1150		device_printf(dev, "could not create statistics DMA tag\n");
1151		return (error);
1152	}
1153	/* Allocate DMA'able memory for statistics. */
1154	error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
1155	    (void **)&sc->rl_ldata.rl_stats,
1156	    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
1157	    &sc->rl_ldata.rl_smap);
1158	if (error) {
1159		device_printf(dev,
1160		    "could not allocate statistics DMA memory\n");
1161		return (error);
1162	}
1163	/* Load the map for statistics. */
1164	sc->rl_ldata.rl_stats_addr = 0;
1165	error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
1166	    sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
1167	     &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
1168	if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
1169		device_printf(dev, "could not load statistics DMA memory\n");
1170		return (ENOMEM);
1171	}
1172
1173	return (0);
1174}
1175
1176/*
1177 * Attach the interface. Allocate softc structures, do ifmedia
1178 * setup and ethernet/BPF attach.
1179 */
1180static int
1181re_attach(device_t dev)
1182{
1183	u_char			eaddr[ETHER_ADDR_LEN];
1184	u_int16_t		as[ETHER_ADDR_LEN / 2];
1185	struct rl_softc		*sc;
1186	struct ifnet		*ifp;
1187	const struct rl_hwrev	*hw_rev;
1188	u_int32_t		cap, ctl;
1189	int			hwrev;
1190	u_int16_t		devid, re_did = 0;
1191	int			error = 0, i, phy, rid;
1192	int			msic, msixc, reg;
1193	uint8_t			cfg;
1194
1195	sc = device_get_softc(dev);
1196	sc->rl_dev = dev;
1197
1198	mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
1199	    MTX_DEF);
1200	callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
1201
1202	/*
1203	 * Map control/status registers.
1204	 */
1205	pci_enable_busmaster(dev);
1206
1207	devid = pci_get_device(dev);
1208	/*
1209	 * Prefer memory space register mapping over IO space.
1210	 * Because RTL8169SC does not seem to work when memory mapping
1211	 * is used always activate io mapping.
1212	 */
1213	if (devid == RT_DEVICEID_8169SC)
1214		prefer_iomap = 1;
1215	if (prefer_iomap == 0) {
1216		sc->rl_res_id = PCIR_BAR(1);
1217		sc->rl_res_type = SYS_RES_MEMORY;
1218		/* RTL8168/8101E seems to use different BARs. */
1219		if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
1220			sc->rl_res_id = PCIR_BAR(2);
1221	} else {
1222		sc->rl_res_id = PCIR_BAR(0);
1223		sc->rl_res_type = SYS_RES_IOPORT;
1224	}
1225	sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1226	    &sc->rl_res_id, RF_ACTIVE);
1227	if (sc->rl_res == NULL && prefer_iomap == 0) {
1228		sc->rl_res_id = PCIR_BAR(0);
1229		sc->rl_res_type = SYS_RES_IOPORT;
1230		sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
1231		    &sc->rl_res_id, RF_ACTIVE);
1232	}
1233	if (sc->rl_res == NULL) {
1234		device_printf(dev, "couldn't map ports/memory\n");
1235		error = ENXIO;
1236		goto fail;
1237	}
1238
1239	sc->rl_btag = rman_get_bustag(sc->rl_res);
1240	sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
1241
1242	msic = pci_msi_count(dev);
1243	msixc = pci_msix_count(dev);
1244	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
1245		sc->rl_flags |= RL_FLAG_PCIE;
1246		sc->rl_expcap = reg;
1247	}
1248	if (bootverbose) {
1249		device_printf(dev, "MSI count : %d\n", msic);
1250		device_printf(dev, "MSI-X count : %d\n", msixc);
1251	}
1252	if (msix_disable > 0)
1253		msixc = 0;
1254	if (msi_disable > 0)
1255		msic = 0;
1256	/* Prefer MSI-X to MSI. */
1257	if (msixc > 0) {
1258		msixc = 1;
1259		rid = PCIR_BAR(4);
1260		sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1261		    &rid, RF_ACTIVE);
1262		if (sc->rl_res_pba == NULL) {
1263			device_printf(sc->rl_dev,
1264			    "could not allocate MSI-X PBA resource\n");
1265		}
1266		if (sc->rl_res_pba != NULL &&
1267		    pci_alloc_msix(dev, &msixc) == 0) {
1268			if (msixc == 1) {
1269				device_printf(dev, "Using %d MSI-X message\n",
1270				    msixc);
1271				sc->rl_flags |= RL_FLAG_MSIX;
1272			} else
1273				pci_release_msi(dev);
1274		}
1275		if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
1276			if (sc->rl_res_pba != NULL)
1277				bus_release_resource(dev, SYS_RES_MEMORY, rid,
1278				    sc->rl_res_pba);
1279			sc->rl_res_pba = NULL;
1280			msixc = 0;
1281		}
1282	}
1283	/* Prefer MSI to INTx. */
1284	if (msixc == 0 && msic > 0) {
1285		msic = 1;
1286		if (pci_alloc_msi(dev, &msic) == 0) {
1287			if (msic == RL_MSI_MESSAGES) {
1288				device_printf(dev, "Using %d MSI message\n",
1289				    msic);
1290				sc->rl_flags |= RL_FLAG_MSI;
1291				/* Explicitly set MSI enable bit. */
1292				CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1293				cfg = CSR_READ_1(sc, RL_CFG2);
1294				cfg |= RL_CFG2_MSI;
1295				CSR_WRITE_1(sc, RL_CFG2, cfg);
1296				CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1297			} else
1298				pci_release_msi(dev);
1299		}
1300		if ((sc->rl_flags & RL_FLAG_MSI) == 0)
1301			msic = 0;
1302	}
1303
1304	/* Allocate interrupt */
1305	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
1306		rid = 0;
1307		sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
1308		    RF_SHAREABLE | RF_ACTIVE);
1309		if (sc->rl_irq[0] == NULL) {
1310			device_printf(dev, "couldn't allocate IRQ resources\n");
1311			error = ENXIO;
1312			goto fail;
1313		}
1314	} else {
1315		for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
1316			sc->rl_irq[i] = bus_alloc_resource_any(dev,
1317			    SYS_RES_IRQ, &rid, RF_ACTIVE);
1318			if (sc->rl_irq[i] == NULL) {
1319				device_printf(dev,
1320				    "couldn't llocate IRQ resources for "
1321				    "message %d\n", rid);
1322				error = ENXIO;
1323				goto fail;
1324			}
1325		}
1326	}
1327
1328	if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
1329		CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1330		cfg = CSR_READ_1(sc, RL_CFG2);
1331		if ((cfg & RL_CFG2_MSI) != 0) {
1332			device_printf(dev, "turning off MSI enable bit.\n");
1333			cfg &= ~RL_CFG2_MSI;
1334			CSR_WRITE_1(sc, RL_CFG2, cfg);
1335		}
1336		CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1337	}
1338
1339	/* Disable ASPM L0S/L1. */
1340	if (sc->rl_expcap != 0) {
1341		cap = pci_read_config(dev, sc->rl_expcap +
1342		    PCIR_EXPRESS_LINK_CAP, 2);
1343		if ((cap & PCIM_LINK_CAP_ASPM) != 0) {
1344			ctl = pci_read_config(dev, sc->rl_expcap +
1345			    PCIR_EXPRESS_LINK_CTL, 2);
1346			if ((ctl & 0x0003) != 0) {
1347				ctl &= ~0x0003;
1348				pci_write_config(dev, sc->rl_expcap +
1349				    PCIR_EXPRESS_LINK_CTL, ctl, 2);
1350				device_printf(dev, "ASPM disabled\n");
1351			}
1352		} else
1353			device_printf(dev, "no ASPM capability\n");
1354	}
1355
1356	hw_rev = re_hwrevs;
1357	hwrev = CSR_READ_4(sc, RL_TXCFG);
1358	switch (hwrev & 0x70000000) {
1359	case 0x00000000:
1360	case 0x10000000:
1361		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
1362		hwrev &= (RL_TXCFG_HWREV | 0x80000000);
1363		break;
1364	default:
1365		device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
1366		hwrev &= RL_TXCFG_HWREV;
1367		break;
1368	}
1369	device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
1370	while (hw_rev->rl_desc != NULL) {
1371		if (hw_rev->rl_rev == hwrev) {
1372			sc->rl_type = hw_rev->rl_type;
1373			sc->rl_hwrev = hw_rev;
1374			break;
1375		}
1376		hw_rev++;
1377	}
1378	if (hw_rev->rl_desc == NULL) {
1379		device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
1380		error = ENXIO;
1381		goto fail;
1382	}
1383
1384	switch (hw_rev->rl_rev) {
1385	case RL_HWREV_8139CPLUS:
1386		sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
1387		break;
1388	case RL_HWREV_8100E:
1389	case RL_HWREV_8101E:
1390		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
1391		break;
1392	case RL_HWREV_8102E:
1393	case RL_HWREV_8102EL:
1394	case RL_HWREV_8102EL_SPIN1:
1395		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1396		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1397		    RL_FLAG_AUTOPAD;
1398		break;
1399	case RL_HWREV_8103E:
1400		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
1401		    RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
1402		    RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
1403		break;
1404	case RL_HWREV_8401E:
1405	case RL_HWREV_8105E:
1406	case RL_HWREV_8105E_SPIN1:
1407		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1408		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1409		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
1410		break;
1411	case RL_HWREV_8402:
1412		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1413		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1414		    RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
1415		    RL_FLAG_CMDSTOP_WAIT_TXQ;
1416		break;
1417	case RL_HWREV_8168B_SPIN1:
1418	case RL_HWREV_8168B_SPIN2:
1419		sc->rl_flags |= RL_FLAG_WOLRXENB;
1420		/* FALLTHROUGH */
1421	case RL_HWREV_8168B_SPIN3:
1422		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
1423		break;
1424	case RL_HWREV_8168C_SPIN2:
1425		sc->rl_flags |= RL_FLAG_MACSLEEP;
1426		/* FALLTHROUGH */
1427	case RL_HWREV_8168C:
1428		if ((hwrev & 0x00700000) == 0x00200000)
1429			sc->rl_flags |= RL_FLAG_MACSLEEP;
1430		/* FALLTHROUGH */
1431	case RL_HWREV_8168CP:
1432		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1433		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1434		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
1435		break;
1436	case RL_HWREV_8168D:
1437		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
1438		    RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
1439		    RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1440		    RL_FLAG_WOL_MANLINK;
1441		break;
1442	case RL_HWREV_8168DP:
1443		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1444		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
1445		    RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
1446		break;
1447	case RL_HWREV_8168E:
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_8168E_VL:
1454	case RL_HWREV_8168F:
1455	case RL_HWREV_8411:
1456		sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
1457		    RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
1458		    RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
1459		    RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
1460		break;
1461	case RL_HWREV_8169_8110SB:
1462	case RL_HWREV_8169_8110SBL:
1463	case RL_HWREV_8169_8110SC:
1464	case RL_HWREV_8169_8110SCE:
1465		sc->rl_flags |= RL_FLAG_PHYWAKE;
1466		/* FALLTHROUGH */
1467	case RL_HWREV_8169:
1468	case RL_HWREV_8169S:
1469	case RL_HWREV_8110S:
1470		sc->rl_flags |= RL_FLAG_MACRESET;
1471		break;
1472	default:
1473		break;
1474	}
1475
1476	if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
1477		sc->rl_cfg0 = RL_8139_CFG0;
1478		sc->rl_cfg1 = RL_8139_CFG1;
1479		sc->rl_cfg2 = 0;
1480		sc->rl_cfg3 = RL_8139_CFG3;
1481		sc->rl_cfg4 = RL_8139_CFG4;
1482		sc->rl_cfg5 = RL_8139_CFG5;
1483	} else {
1484		sc->rl_cfg0 = RL_CFG0;
1485		sc->rl_cfg1 = RL_CFG1;
1486		sc->rl_cfg2 = RL_CFG2;
1487		sc->rl_cfg3 = RL_CFG3;
1488		sc->rl_cfg4 = RL_CFG4;
1489		sc->rl_cfg5 = RL_CFG5;
1490	}
1491
1492	/* Reset the adapter. */
1493	RL_LOCK(sc);
1494	re_reset(sc);
1495	RL_UNLOCK(sc);
1496
1497	/* Enable PME. */
1498	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
1499	cfg = CSR_READ_1(sc, sc->rl_cfg1);
1500	cfg |= RL_CFG1_PME;
1501	CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
1502	cfg = CSR_READ_1(sc, sc->rl_cfg5);
1503	cfg &= RL_CFG5_PME_STS;
1504	CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
1505	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
1506
1507	if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
1508		/*
1509		 * XXX Should have a better way to extract station
1510		 * address from EEPROM.
1511		 */
1512		for (i = 0; i < ETHER_ADDR_LEN; i++)
1513			eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
1514	} else {
1515		sc->rl_eewidth = RL_9356_ADDR_LEN;
1516		re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
1517		if (re_did != 0x8129)
1518			sc->rl_eewidth = RL_9346_ADDR_LEN;
1519
1520		/*
1521		 * Get station address from the EEPROM.
1522		 */
1523		re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
1524		for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
1525			as[i] = le16toh(as[i]);
1526		bcopy(as, eaddr, sizeof(eaddr));
1527	}
1528
1529	if (sc->rl_type == RL_8169) {
1530		/* Set RX length mask and number of descriptors. */
1531		sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
1532		sc->rl_txstart = RL_GTXSTART;
1533		sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
1534		sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
1535	} else {
1536		/* Set RX length mask and number of descriptors. */
1537		sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
1538		sc->rl_txstart = RL_TXSTART;
1539		sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
1540		sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
1541	}
1542
1543	error = re_allocmem(dev, sc);
1544	if (error)
1545		goto fail;
1546	re_add_sysctls(sc);
1547
1548	ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
1549	if (ifp == NULL) {
1550		device_printf(dev, "can not if_alloc()\n");
1551		error = ENOSPC;
1552		goto fail;
1553	}
1554
1555	/* Take controller out of deep sleep mode. */
1556	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
1557		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
1558			CSR_WRITE_1(sc, RL_GPIO,
1559			    CSR_READ_1(sc, RL_GPIO) | 0x01);
1560		else
1561			CSR_WRITE_1(sc, RL_GPIO,
1562			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
1563	}
1564
1565	/* Take PHY out of power down mode. */
1566	if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
1567		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
1568		if (hw_rev->rl_rev == RL_HWREV_8401E)
1569			CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
1570	}
1571	if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
1572		re_gmii_writereg(dev, 1, 0x1f, 0);
1573		re_gmii_writereg(dev, 1, 0x0e, 0);
1574	}
1575
1576	ifp->if_softc = sc;
1577	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1578	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1579	ifp->if_ioctl = re_ioctl;
1580	ifp->if_start = re_start;
1581	/*
1582	 * RTL8168/8111C generates wrong IP checksummed frame if the
1583	 * packet has IP options so disable TX IP checksum offloading.
1584	 */
1585	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
1586	    sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
1587		ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
1588	else
1589		ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
1590	ifp->if_hwassist |= CSUM_TSO;
1591	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
1592	ifp->if_capenable = ifp->if_capabilities;
1593	ifp->if_init = re_init;
1594	IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
1595	ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
1596	IFQ_SET_READY(&ifp->if_snd);
1597
1598	TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
1599
1600#define	RE_PHYAD_INTERNAL	 0
1601
1602	/* Do MII setup. */
1603	phy = RE_PHYAD_INTERNAL;
1604	if (sc->rl_type == RL_8169)
1605		phy = 1;
1606	error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
1607	    re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
1608	if (error != 0) {
1609		device_printf(dev, "attaching PHYs failed\n");
1610		goto fail;
1611	}
1612
1613	/*
1614	 * Call MI attach routine.
1615	 */
1616	ether_ifattach(ifp, eaddr);
1617
1618	/* VLAN capability setup */
1619	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
1620	if (ifp->if_capabilities & IFCAP_HWCSUM)
1621		ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
1622	/* Enable WOL if PM is supported. */
1623	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
1624		ifp->if_capabilities |= IFCAP_WOL;
1625	ifp->if_capenable = ifp->if_capabilities;
1626	ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
1627	/*
1628	 * Don't enable TSO by default.  It is known to generate
1629	 * corrupted TCP segments(bad TCP options) under certain
1630	 * circumtances.
1631	 */
1632	ifp->if_hwassist &= ~CSUM_TSO;
1633	ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
1634#ifdef DEVICE_POLLING
1635	ifp->if_capabilities |= IFCAP_POLLING;
1636#endif
1637	/*
1638	 * Tell the upper layer(s) we support long frames.
1639	 * Must appear after the call to ether_ifattach() because
1640	 * ether_ifattach() sets ifi_hdrlen to the default value.
1641	 */
1642	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1643
1644#ifdef RE_DIAG
1645	/*
1646	 * Perform hardware diagnostic on the original RTL8169.
1647	 * Some 32-bit cards were incorrectly wired and would
1648	 * malfunction if plugged into a 64-bit slot.
1649	 */
1650
1651	if (hwrev == RL_HWREV_8169) {
1652		error = re_diag(sc);
1653		if (error) {
1654			device_printf(dev,
1655		    	"attach aborted due to hardware diag failure\n");
1656			ether_ifdetach(ifp);
1657			goto fail;
1658		}
1659	}
1660#endif
1661
1662#ifdef RE_TX_MODERATION
1663	intr_filter = 1;
1664#endif
1665	/* Hook interrupt last to avoid having to lock softc */
1666	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
1667	    intr_filter == 0) {
1668		error = bus_setup_intr(dev, sc->rl_irq[0],
1669		    INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
1670		    &sc->rl_intrhand[0]);
1671	} else {
1672		error = bus_setup_intr(dev, sc->rl_irq[0],
1673		    INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
1674		    &sc->rl_intrhand[0]);
1675	}
1676	if (error) {
1677		device_printf(dev, "couldn't set up irq\n");
1678		ether_ifdetach(ifp);
1679	}
1680
1681fail:
1682
1683	if (error)
1684		re_detach(dev);
1685
1686	return (error);
1687}
1688
1689/*
1690 * Shutdown hardware and free up resources. This can be called any
1691 * time after the mutex has been initialized. It is called in both
1692 * the error case in attach and the normal detach case so it needs
1693 * to be careful about only freeing resources that have actually been
1694 * allocated.
1695 */
1696static int
1697re_detach(device_t dev)
1698{
1699	struct rl_softc		*sc;
1700	struct ifnet		*ifp;
1701	int			i, rid;
1702
1703	sc = device_get_softc(dev);
1704	ifp = sc->rl_ifp;
1705	KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
1706
1707	/* These should only be active if attach succeeded */
1708	if (device_is_attached(dev)) {
1709#ifdef DEVICE_POLLING
1710		if (ifp->if_capenable & IFCAP_POLLING)
1711			ether_poll_deregister(ifp);
1712#endif
1713		RL_LOCK(sc);
1714#if 0
1715		sc->suspended = 1;
1716#endif
1717		re_stop(sc);
1718		RL_UNLOCK(sc);
1719		callout_drain(&sc->rl_stat_callout);
1720		taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
1721		/*
1722		 * Force off the IFF_UP flag here, in case someone
1723		 * still had a BPF descriptor attached to this
1724		 * interface. If they do, ether_ifdetach() will cause
1725		 * the BPF code to try and clear the promisc mode
1726		 * flag, which will bubble down to re_ioctl(),
1727		 * which will try to call re_init() again. This will
1728		 * turn the NIC back on and restart the MII ticker,
1729		 * which will panic the system when the kernel tries
1730		 * to invoke the re_tick() function that isn't there
1731		 * anymore.
1732		 */
1733		ifp->if_flags &= ~IFF_UP;
1734		ether_ifdetach(ifp);
1735	}
1736	if (sc->rl_miibus)
1737		device_delete_child(dev, sc->rl_miibus);
1738	bus_generic_detach(dev);
1739
1740	/*
1741	 * The rest is resource deallocation, so we should already be
1742	 * stopped here.
1743	 */
1744
1745	if (sc->rl_intrhand[0] != NULL) {
1746		bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
1747		sc->rl_intrhand[0] = NULL;
1748	}
1749	if (ifp != NULL)
1750		if_free(ifp);
1751	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
1752		rid = 0;
1753	else
1754		rid = 1;
1755	if (sc->rl_irq[0] != NULL) {
1756		bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
1757		sc->rl_irq[0] = NULL;
1758	}
1759	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
1760		pci_release_msi(dev);
1761	if (sc->rl_res_pba) {
1762		rid = PCIR_BAR(4);
1763		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
1764	}
1765	if (sc->rl_res)
1766		bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
1767		    sc->rl_res);
1768
1769	/* Unload and free the RX DMA ring memory and map */
1770
1771	if (sc->rl_ldata.rl_rx_list_tag) {
1772		if (sc->rl_ldata.rl_rx_list_map)
1773			bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
1774			    sc->rl_ldata.rl_rx_list_map);
1775		if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
1776			bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
1777			    sc->rl_ldata.rl_rx_list,
1778			    sc->rl_ldata.rl_rx_list_map);
1779		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
1780	}
1781
1782	/* Unload and free the TX DMA ring memory and map */
1783
1784	if (sc->rl_ldata.rl_tx_list_tag) {
1785		if (sc->rl_ldata.rl_tx_list_map)
1786			bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
1787			    sc->rl_ldata.rl_tx_list_map);
1788		if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
1789			bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
1790			    sc->rl_ldata.rl_tx_list,
1791			    sc->rl_ldata.rl_tx_list_map);
1792		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
1793	}
1794
1795	/* Destroy all the RX and TX buffer maps */
1796
1797	if (sc->rl_ldata.rl_tx_mtag) {
1798		for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
1799			if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
1800				bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
1801				    sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
1802		}
1803		bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
1804	}
1805	if (sc->rl_ldata.rl_rx_mtag) {
1806		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1807			if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
1808				bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1809				    sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
1810		}
1811		if (sc->rl_ldata.rl_rx_sparemap)
1812			bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
1813			    sc->rl_ldata.rl_rx_sparemap);
1814		bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
1815	}
1816	if (sc->rl_ldata.rl_jrx_mtag) {
1817		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
1818			if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
1819				bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1820				    sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
1821		}
1822		if (sc->rl_ldata.rl_jrx_sparemap)
1823			bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
1824			    sc->rl_ldata.rl_jrx_sparemap);
1825		bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
1826	}
1827	/* Unload and free the stats buffer and map */
1828
1829	if (sc->rl_ldata.rl_stag) {
1830		if (sc->rl_ldata.rl_smap)
1831			bus_dmamap_unload(sc->rl_ldata.rl_stag,
1832			    sc->rl_ldata.rl_smap);
1833		if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
1834			bus_dmamem_free(sc->rl_ldata.rl_stag,
1835			    sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
1836		bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
1837	}
1838
1839	if (sc->rl_parent_tag)
1840		bus_dma_tag_destroy(sc->rl_parent_tag);
1841
1842	mtx_destroy(&sc->rl_mtx);
1843
1844	return (0);
1845}
1846
1847static __inline void
1848re_discard_rxbuf(struct rl_softc *sc, int idx)
1849{
1850	struct rl_desc		*desc;
1851	struct rl_rxdesc	*rxd;
1852	uint32_t		cmdstat;
1853
1854	if (sc->rl_ifp->if_mtu > RL_MTU &&
1855	    (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
1856		rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1857	else
1858		rxd = &sc->rl_ldata.rl_rx_desc[idx];
1859	desc = &sc->rl_ldata.rl_rx_list[idx];
1860	desc->rl_vlanctl = 0;
1861	cmdstat = rxd->rx_size;
1862	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1863		cmdstat |= RL_RDESC_CMD_EOR;
1864	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1865}
1866
1867static int
1868re_newbuf(struct rl_softc *sc, int idx)
1869{
1870	struct mbuf		*m;
1871	struct rl_rxdesc	*rxd;
1872	bus_dma_segment_t	segs[1];
1873	bus_dmamap_t		map;
1874	struct rl_desc		*desc;
1875	uint32_t		cmdstat;
1876	int			error, nsegs;
1877
1878	m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1879	if (m == NULL)
1880		return (ENOBUFS);
1881
1882	m->m_len = m->m_pkthdr.len = MCLBYTES;
1883#ifdef RE_FIXUP_RX
1884	/*
1885	 * This is part of an evil trick to deal with non-x86 platforms.
1886	 * The RealTek chip requires RX buffers to be aligned on 64-bit
1887	 * boundaries, but that will hose non-x86 machines. To get around
1888	 * this, we leave some empty space at the start of each buffer
1889	 * and for non-x86 hosts, we copy the buffer back six bytes
1890	 * to achieve word alignment. This is slightly more efficient
1891	 * than allocating a new buffer, copying the contents, and
1892	 * discarding the old buffer.
1893	 */
1894	m_adj(m, RE_ETHER_ALIGN);
1895#endif
1896	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
1897	    sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1898	if (error != 0) {
1899		m_freem(m);
1900		return (ENOBUFS);
1901	}
1902	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1903
1904	rxd = &sc->rl_ldata.rl_rx_desc[idx];
1905	if (rxd->rx_m != NULL) {
1906		bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1907		    BUS_DMASYNC_POSTREAD);
1908		bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
1909	}
1910
1911	rxd->rx_m = m;
1912	map = rxd->rx_dmamap;
1913	rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
1914	rxd->rx_size = segs[0].ds_len;
1915	sc->rl_ldata.rl_rx_sparemap = map;
1916	bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
1917	    BUS_DMASYNC_PREREAD);
1918
1919	desc = &sc->rl_ldata.rl_rx_list[idx];
1920	desc->rl_vlanctl = 0;
1921	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1922	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1923	cmdstat = segs[0].ds_len;
1924	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1925		cmdstat |= RL_RDESC_CMD_EOR;
1926	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1927
1928	return (0);
1929}
1930
1931static int
1932re_jumbo_newbuf(struct rl_softc *sc, int idx)
1933{
1934	struct mbuf		*m;
1935	struct rl_rxdesc	*rxd;
1936	bus_dma_segment_t	segs[1];
1937	bus_dmamap_t		map;
1938	struct rl_desc		*desc;
1939	uint32_t		cmdstat;
1940	int			error, nsegs;
1941
1942	m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1943	if (m == NULL)
1944		return (ENOBUFS);
1945	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1946#ifdef RE_FIXUP_RX
1947	m_adj(m, RE_ETHER_ALIGN);
1948#endif
1949	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
1950	    sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
1951	if (error != 0) {
1952		m_freem(m);
1953		return (ENOBUFS);
1954	}
1955	KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
1956
1957	rxd = &sc->rl_ldata.rl_jrx_desc[idx];
1958	if (rxd->rx_m != NULL) {
1959		bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1960		    BUS_DMASYNC_POSTREAD);
1961		bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
1962	}
1963
1964	rxd->rx_m = m;
1965	map = rxd->rx_dmamap;
1966	rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
1967	rxd->rx_size = segs[0].ds_len;
1968	sc->rl_ldata.rl_jrx_sparemap = map;
1969	bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
1970	    BUS_DMASYNC_PREREAD);
1971
1972	desc = &sc->rl_ldata.rl_rx_list[idx];
1973	desc->rl_vlanctl = 0;
1974	desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
1975	desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
1976	cmdstat = segs[0].ds_len;
1977	if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
1978		cmdstat |= RL_RDESC_CMD_EOR;
1979	desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
1980
1981	return (0);
1982}
1983
1984#ifdef RE_FIXUP_RX
1985static __inline void
1986re_fixup_rx(struct mbuf *m)
1987{
1988	int                     i;
1989	uint16_t                *src, *dst;
1990
1991	src = mtod(m, uint16_t *);
1992	dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
1993
1994	for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
1995		*dst++ = *src++;
1996
1997	m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
1998}
1999#endif
2000
2001static int
2002re_tx_list_init(struct rl_softc *sc)
2003{
2004	struct rl_desc		*desc;
2005	int			i;
2006
2007	RL_LOCK_ASSERT(sc);
2008
2009	bzero(sc->rl_ldata.rl_tx_list,
2010	    sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
2011	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
2012		sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
2013	/* Set EOR. */
2014	desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
2015	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
2016
2017	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2018	    sc->rl_ldata.rl_tx_list_map,
2019	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2020
2021	sc->rl_ldata.rl_tx_prodidx = 0;
2022	sc->rl_ldata.rl_tx_considx = 0;
2023	sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
2024
2025	return (0);
2026}
2027
2028static int
2029re_rx_list_init(struct rl_softc *sc)
2030{
2031	int			error, i;
2032
2033	bzero(sc->rl_ldata.rl_rx_list,
2034	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2035	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2036		sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
2037		if ((error = re_newbuf(sc, i)) != 0)
2038			return (error);
2039	}
2040
2041	/* Flush the RX descriptors */
2042
2043	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2044	    sc->rl_ldata.rl_rx_list_map,
2045	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2046
2047	sc->rl_ldata.rl_rx_prodidx = 0;
2048	sc->rl_head = sc->rl_tail = NULL;
2049	sc->rl_int_rx_act = 0;
2050
2051	return (0);
2052}
2053
2054static int
2055re_jrx_list_init(struct rl_softc *sc)
2056{
2057	int			error, i;
2058
2059	bzero(sc->rl_ldata.rl_rx_list,
2060	    sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
2061	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
2062		sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
2063		if ((error = re_jumbo_newbuf(sc, i)) != 0)
2064			return (error);
2065	}
2066
2067	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2068	    sc->rl_ldata.rl_rx_list_map,
2069	    BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
2070
2071	sc->rl_ldata.rl_rx_prodidx = 0;
2072	sc->rl_head = sc->rl_tail = NULL;
2073	sc->rl_int_rx_act = 0;
2074
2075	return (0);
2076}
2077
2078/*
2079 * RX handler for C+ and 8169. For the gigE chips, we support
2080 * the reception of jumbo frames that have been fragmented
2081 * across multiple 2K mbuf cluster buffers.
2082 */
2083static int
2084re_rxeof(struct rl_softc *sc, int *rx_npktsp)
2085{
2086	struct mbuf		*m;
2087	struct ifnet		*ifp;
2088	int			i, rxerr, total_len;
2089	struct rl_desc		*cur_rx;
2090	u_int32_t		rxstat, rxvlan;
2091	int			jumbo, maxpkt = 16, rx_npkts = 0;
2092
2093	RL_LOCK_ASSERT(sc);
2094
2095	ifp = sc->rl_ifp;
2096	if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
2097		jumbo = 1;
2098	else
2099		jumbo = 0;
2100
2101	/* Invalidate the descriptor memory */
2102
2103	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2104	    sc->rl_ldata.rl_rx_list_map,
2105	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2106
2107	for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
2108	    i = RL_RX_DESC_NXT(sc, i)) {
2109		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2110			break;
2111		cur_rx = &sc->rl_ldata.rl_rx_list[i];
2112		rxstat = le32toh(cur_rx->rl_cmdstat);
2113		if ((rxstat & RL_RDESC_STAT_OWN) != 0)
2114			break;
2115		total_len = rxstat & sc->rl_rxlenmask;
2116		rxvlan = le32toh(cur_rx->rl_vlanctl);
2117		if (jumbo != 0)
2118			m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
2119		else
2120			m = sc->rl_ldata.rl_rx_desc[i].rx_m;
2121
2122		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
2123		    (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
2124		    (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
2125			/*
2126			 * RTL8168C or later controllers do not
2127			 * support multi-fragment packet.
2128			 */
2129			re_discard_rxbuf(sc, i);
2130			continue;
2131		} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
2132			if (re_newbuf(sc, i) != 0) {
2133				/*
2134				 * If this is part of a multi-fragment packet,
2135				 * discard all the pieces.
2136				 */
2137				if (sc->rl_head != NULL) {
2138					m_freem(sc->rl_head);
2139					sc->rl_head = sc->rl_tail = NULL;
2140				}
2141				re_discard_rxbuf(sc, i);
2142				continue;
2143			}
2144			m->m_len = RE_RX_DESC_BUFLEN;
2145			if (sc->rl_head == NULL)
2146				sc->rl_head = sc->rl_tail = m;
2147			else {
2148				m->m_flags &= ~M_PKTHDR;
2149				sc->rl_tail->m_next = m;
2150				sc->rl_tail = m;
2151			}
2152			continue;
2153		}
2154
2155		/*
2156		 * NOTE: for the 8139C+, the frame length field
2157		 * is always 12 bits in size, but for the gigE chips,
2158		 * it is 13 bits (since the max RX frame length is 16K).
2159		 * Unfortunately, all 32 bits in the status word
2160		 * were already used, so to make room for the extra
2161		 * length bit, RealTek took out the 'frame alignment
2162		 * error' bit and shifted the other status bits
2163		 * over one slot. The OWN, EOR, FS and LS bits are
2164		 * still in the same places. We have already extracted
2165		 * the frame length and checked the OWN bit, so rather
2166		 * than using an alternate bit mapping, we shift the
2167		 * status bits one space to the right so we can evaluate
2168		 * them using the 8169 status as though it was in the
2169		 * same format as that of the 8139C+.
2170		 */
2171		if (sc->rl_type == RL_8169)
2172			rxstat >>= 1;
2173
2174		/*
2175		 * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
2176		 * set, but if CRC is clear, it will still be a valid frame.
2177		 */
2178		if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
2179			rxerr = 1;
2180			if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
2181			    total_len > 8191 &&
2182			    (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
2183				rxerr = 0;
2184			if (rxerr != 0) {
2185				ifp->if_ierrors++;
2186				/*
2187				 * If this is part of a multi-fragment packet,
2188				 * discard all the pieces.
2189				 */
2190				if (sc->rl_head != NULL) {
2191					m_freem(sc->rl_head);
2192					sc->rl_head = sc->rl_tail = NULL;
2193				}
2194				re_discard_rxbuf(sc, i);
2195				continue;
2196			}
2197		}
2198
2199		/*
2200		 * If allocating a replacement mbuf fails,
2201		 * reload the current one.
2202		 */
2203		if (jumbo != 0)
2204			rxerr = re_jumbo_newbuf(sc, i);
2205		else
2206			rxerr = re_newbuf(sc, i);
2207		if (rxerr != 0) {
2208			ifp->if_iqdrops++;
2209			if (sc->rl_head != NULL) {
2210				m_freem(sc->rl_head);
2211				sc->rl_head = sc->rl_tail = NULL;
2212			}
2213			re_discard_rxbuf(sc, i);
2214			continue;
2215		}
2216
2217		if (sc->rl_head != NULL) {
2218			if (jumbo != 0)
2219				m->m_len = total_len;
2220			else {
2221				m->m_len = total_len % RE_RX_DESC_BUFLEN;
2222				if (m->m_len == 0)
2223					m->m_len = RE_RX_DESC_BUFLEN;
2224			}
2225			/*
2226			 * Special case: if there's 4 bytes or less
2227			 * in this buffer, the mbuf can be discarded:
2228			 * the last 4 bytes is the CRC, which we don't
2229			 * care about anyway.
2230			 */
2231			if (m->m_len <= ETHER_CRC_LEN) {
2232				sc->rl_tail->m_len -=
2233				    (ETHER_CRC_LEN - m->m_len);
2234				m_freem(m);
2235			} else {
2236				m->m_len -= ETHER_CRC_LEN;
2237				m->m_flags &= ~M_PKTHDR;
2238				sc->rl_tail->m_next = m;
2239			}
2240			m = sc->rl_head;
2241			sc->rl_head = sc->rl_tail = NULL;
2242			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
2243		} else
2244			m->m_pkthdr.len = m->m_len =
2245			    (total_len - ETHER_CRC_LEN);
2246
2247#ifdef RE_FIXUP_RX
2248		re_fixup_rx(m);
2249#endif
2250		ifp->if_ipackets++;
2251		m->m_pkthdr.rcvif = ifp;
2252
2253		/* Do RX checksumming if enabled */
2254
2255		if (ifp->if_capenable & IFCAP_RXCSUM) {
2256			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2257				/* Check IP header checksum */
2258				if (rxstat & RL_RDESC_STAT_PROTOID)
2259					m->m_pkthdr.csum_flags |=
2260					    CSUM_IP_CHECKED;
2261				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
2262					m->m_pkthdr.csum_flags |=
2263					    CSUM_IP_VALID;
2264
2265				/* Check TCP/UDP checksum */
2266				if ((RL_TCPPKT(rxstat) &&
2267				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2268				    (RL_UDPPKT(rxstat) &&
2269				     !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2270					m->m_pkthdr.csum_flags |=
2271						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2272					m->m_pkthdr.csum_data = 0xffff;
2273				}
2274			} else {
2275				/*
2276				 * RTL8168C/RTL816CP/RTL8111C/RTL8111CP
2277				 */
2278				if ((rxstat & RL_RDESC_STAT_PROTOID) &&
2279				    (rxvlan & RL_RDESC_IPV4))
2280					m->m_pkthdr.csum_flags |=
2281					    CSUM_IP_CHECKED;
2282				if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
2283				    (rxvlan & RL_RDESC_IPV4))
2284					m->m_pkthdr.csum_flags |=
2285					    CSUM_IP_VALID;
2286				if (((rxstat & RL_RDESC_STAT_TCP) &&
2287				    !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
2288				    ((rxstat & RL_RDESC_STAT_UDP) &&
2289				    !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
2290					m->m_pkthdr.csum_flags |=
2291						CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
2292					m->m_pkthdr.csum_data = 0xffff;
2293				}
2294			}
2295		}
2296		maxpkt--;
2297		if (rxvlan & RL_RDESC_VLANCTL_TAG) {
2298			m->m_pkthdr.ether_vtag =
2299			    bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
2300			m->m_flags |= M_VLANTAG;
2301		}
2302		RL_UNLOCK(sc);
2303		(*ifp->if_input)(ifp, m);
2304		RL_LOCK(sc);
2305		rx_npkts++;
2306	}
2307
2308	/* Flush the RX DMA ring */
2309
2310	bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
2311	    sc->rl_ldata.rl_rx_list_map,
2312	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2313
2314	sc->rl_ldata.rl_rx_prodidx = i;
2315
2316	if (rx_npktsp != NULL)
2317		*rx_npktsp = rx_npkts;
2318	if (maxpkt)
2319		return (EAGAIN);
2320
2321	return (0);
2322}
2323
2324static void
2325re_txeof(struct rl_softc *sc)
2326{
2327	struct ifnet		*ifp;
2328	struct rl_txdesc	*txd;
2329	u_int32_t		txstat;
2330	int			cons;
2331
2332	cons = sc->rl_ldata.rl_tx_considx;
2333	if (cons == sc->rl_ldata.rl_tx_prodidx)
2334		return;
2335
2336	ifp = sc->rl_ifp;
2337	/* Invalidate the TX descriptor list */
2338	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2339	    sc->rl_ldata.rl_tx_list_map,
2340	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2341
2342	for (; cons != sc->rl_ldata.rl_tx_prodidx;
2343	    cons = RL_TX_DESC_NXT(sc, cons)) {
2344		txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
2345		if (txstat & RL_TDESC_STAT_OWN)
2346			break;
2347		/*
2348		 * We only stash mbufs in the last descriptor
2349		 * in a fragment chain, which also happens to
2350		 * be the only place where the TX status bits
2351		 * are valid.
2352		 */
2353		if (txstat & RL_TDESC_CMD_EOF) {
2354			txd = &sc->rl_ldata.rl_tx_desc[cons];
2355			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
2356			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2357			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
2358			    txd->tx_dmamap);
2359			KASSERT(txd->tx_m != NULL,
2360			    ("%s: freeing NULL mbufs!", __func__));
2361			m_freem(txd->tx_m);
2362			txd->tx_m = NULL;
2363			if (txstat & (RL_TDESC_STAT_EXCESSCOL|
2364			    RL_TDESC_STAT_COLCNT))
2365				ifp->if_collisions++;
2366			if (txstat & RL_TDESC_STAT_TXERRSUM)
2367				ifp->if_oerrors++;
2368			else
2369				ifp->if_opackets++;
2370		}
2371		sc->rl_ldata.rl_tx_free++;
2372		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2373	}
2374	sc->rl_ldata.rl_tx_considx = cons;
2375
2376	/* No changes made to the TX ring, so no flush needed */
2377
2378	if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
2379#ifdef RE_TX_MODERATION
2380		/*
2381		 * If not all descriptors have been reaped yet, reload
2382		 * the timer so that we will eventually get another
2383		 * interrupt that will cause us to re-enter this routine.
2384		 * This is done in case the transmitter has gone idle.
2385		 */
2386		CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2387#endif
2388	} else
2389		sc->rl_watchdog_timer = 0;
2390}
2391
2392static void
2393re_tick(void *xsc)
2394{
2395	struct rl_softc		*sc;
2396	struct mii_data		*mii;
2397
2398	sc = xsc;
2399
2400	RL_LOCK_ASSERT(sc);
2401
2402	mii = device_get_softc(sc->rl_miibus);
2403	mii_tick(mii);
2404	if ((sc->rl_flags & RL_FLAG_LINK) == 0)
2405		re_miibus_statchg(sc->rl_dev);
2406	/*
2407	 * Reclaim transmitted frames here. Technically it is not
2408	 * necessary to do here but it ensures periodic reclamation
2409	 * regardless of Tx completion interrupt which seems to be
2410	 * lost on PCIe based controllers under certain situations.
2411	 */
2412	re_txeof(sc);
2413	re_watchdog(sc);
2414	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
2415}
2416
2417#ifdef DEVICE_POLLING
2418static int
2419re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2420{
2421	struct rl_softc *sc = ifp->if_softc;
2422	int rx_npkts = 0;
2423
2424	RL_LOCK(sc);
2425	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2426		rx_npkts = re_poll_locked(ifp, cmd, count);
2427	RL_UNLOCK(sc);
2428	return (rx_npkts);
2429}
2430
2431static int
2432re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
2433{
2434	struct rl_softc *sc = ifp->if_softc;
2435	int rx_npkts;
2436
2437	RL_LOCK_ASSERT(sc);
2438
2439	sc->rxcycles = count;
2440	re_rxeof(sc, &rx_npkts);
2441	re_txeof(sc);
2442
2443	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2444		re_start_locked(ifp);
2445
2446	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2447		u_int16_t       status;
2448
2449		status = CSR_READ_2(sc, RL_ISR);
2450		if (status == 0xffff)
2451			return (rx_npkts);
2452		if (status)
2453			CSR_WRITE_2(sc, RL_ISR, status);
2454		if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2455		    (sc->rl_flags & RL_FLAG_PCIE))
2456			CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2457
2458		/*
2459		 * XXX check behaviour on receiver stalls.
2460		 */
2461
2462		if (status & RL_ISR_SYSTEM_ERR) {
2463			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2464			re_init_locked(sc);
2465		}
2466	}
2467	return (rx_npkts);
2468}
2469#endif /* DEVICE_POLLING */
2470
2471static int
2472re_intr(void *arg)
2473{
2474	struct rl_softc		*sc;
2475	uint16_t		status;
2476
2477	sc = arg;
2478
2479	status = CSR_READ_2(sc, RL_ISR);
2480	if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
2481                return (FILTER_STRAY);
2482	CSR_WRITE_2(sc, RL_IMR, 0);
2483
2484	taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2485
2486	return (FILTER_HANDLED);
2487}
2488
2489static void
2490re_int_task(void *arg, int npending)
2491{
2492	struct rl_softc		*sc;
2493	struct ifnet		*ifp;
2494	u_int16_t		status;
2495	int			rval = 0;
2496
2497	sc = arg;
2498	ifp = sc->rl_ifp;
2499
2500	RL_LOCK(sc);
2501
2502	status = CSR_READ_2(sc, RL_ISR);
2503        CSR_WRITE_2(sc, RL_ISR, status);
2504
2505	if (sc->suspended ||
2506	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2507		RL_UNLOCK(sc);
2508		return;
2509	}
2510
2511#ifdef DEVICE_POLLING
2512	if  (ifp->if_capenable & IFCAP_POLLING) {
2513		RL_UNLOCK(sc);
2514		return;
2515	}
2516#endif
2517
2518	if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
2519		rval = re_rxeof(sc, NULL);
2520
2521	/*
2522	 * Some chips will ignore a second TX request issued
2523	 * while an existing transmission is in progress. If
2524	 * the transmitter goes idle but there are still
2525	 * packets waiting to be sent, we need to restart the
2526	 * channel here to flush them out. This only seems to
2527	 * be required with the PCIe devices.
2528	 */
2529	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2530	    (sc->rl_flags & RL_FLAG_PCIE))
2531		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2532	if (status & (
2533#ifdef RE_TX_MODERATION
2534	    RL_ISR_TIMEOUT_EXPIRED|
2535#else
2536	    RL_ISR_TX_OK|
2537#endif
2538	    RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
2539		re_txeof(sc);
2540
2541	if (status & RL_ISR_SYSTEM_ERR) {
2542		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2543		re_init_locked(sc);
2544	}
2545
2546	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2547		re_start_locked(ifp);
2548
2549	RL_UNLOCK(sc);
2550
2551        if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
2552		taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
2553		return;
2554	}
2555
2556	CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
2557}
2558
2559static void
2560re_intr_msi(void *xsc)
2561{
2562	struct rl_softc		*sc;
2563	struct ifnet		*ifp;
2564	uint16_t		intrs, status;
2565
2566	sc = xsc;
2567	RL_LOCK(sc);
2568
2569	ifp = sc->rl_ifp;
2570#ifdef DEVICE_POLLING
2571	if (ifp->if_capenable & IFCAP_POLLING) {
2572		RL_UNLOCK(sc);
2573		return;
2574	}
2575#endif
2576	/* Disable interrupts. */
2577	CSR_WRITE_2(sc, RL_IMR, 0);
2578	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2579		RL_UNLOCK(sc);
2580		return;
2581	}
2582
2583	intrs = RL_INTRS_CPLUS;
2584	status = CSR_READ_2(sc, RL_ISR);
2585        CSR_WRITE_2(sc, RL_ISR, status);
2586	if (sc->rl_int_rx_act > 0) {
2587		intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2588		    RL_ISR_RX_OVERRUN);
2589		status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
2590		    RL_ISR_RX_OVERRUN);
2591	}
2592
2593	if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
2594	    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
2595		re_rxeof(sc, NULL);
2596		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2597			if (sc->rl_int_rx_mod != 0 &&
2598			    (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
2599			    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
2600				/* Rearm one-shot timer. */
2601				CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2602				intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
2603				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
2604				sc->rl_int_rx_act = 1;
2605			} else {
2606				intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
2607				    RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
2608				sc->rl_int_rx_act = 0;
2609			}
2610		}
2611	}
2612
2613	/*
2614	 * Some chips will ignore a second TX request issued
2615	 * while an existing transmission is in progress. If
2616	 * the transmitter goes idle but there are still
2617	 * packets waiting to be sent, we need to restart the
2618	 * channel here to flush them out. This only seems to
2619	 * be required with the PCIe devices.
2620	 */
2621	if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
2622	    (sc->rl_flags & RL_FLAG_PCIE))
2623		CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2624	if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
2625		re_txeof(sc);
2626
2627	if (status & RL_ISR_SYSTEM_ERR) {
2628		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2629		re_init_locked(sc);
2630	}
2631
2632	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2633		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2634			re_start_locked(ifp);
2635		CSR_WRITE_2(sc, RL_IMR, intrs);
2636	}
2637	RL_UNLOCK(sc);
2638}
2639
2640static int
2641re_encap(struct rl_softc *sc, struct mbuf **m_head)
2642{
2643	struct rl_txdesc	*txd, *txd_last;
2644	bus_dma_segment_t	segs[RL_NTXSEGS];
2645	bus_dmamap_t		map;
2646	struct mbuf		*m_new;
2647	struct rl_desc		*desc;
2648	int			nsegs, prod;
2649	int			i, error, ei, si;
2650	int			padlen;
2651	uint32_t		cmdstat, csum_flags, vlanctl;
2652
2653	RL_LOCK_ASSERT(sc);
2654	M_ASSERTPKTHDR((*m_head));
2655
2656	/*
2657	 * With some of the RealTek chips, using the checksum offload
2658	 * support in conjunction with the autopadding feature results
2659	 * in the transmission of corrupt frames. For example, if we
2660	 * need to send a really small IP fragment that's less than 60
2661	 * bytes in size, and IP header checksumming is enabled, the
2662	 * resulting ethernet frame that appears on the wire will
2663	 * have garbled payload. To work around this, if TX IP checksum
2664	 * offload is enabled, we always manually pad short frames out
2665	 * to the minimum ethernet frame size.
2666	 */
2667	if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
2668	    (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
2669	    ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2670		padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
2671		if (M_WRITABLE(*m_head) == 0) {
2672			/* Get a writable copy. */
2673			m_new = m_dup(*m_head, M_DONTWAIT);
2674			m_freem(*m_head);
2675			if (m_new == NULL) {
2676				*m_head = NULL;
2677				return (ENOBUFS);
2678			}
2679			*m_head = m_new;
2680		}
2681		if ((*m_head)->m_next != NULL ||
2682		    M_TRAILINGSPACE(*m_head) < padlen) {
2683			m_new = m_defrag(*m_head, M_DONTWAIT);
2684			if (m_new == NULL) {
2685				m_freem(*m_head);
2686				*m_head = NULL;
2687				return (ENOBUFS);
2688			}
2689		} else
2690			m_new = *m_head;
2691
2692		/*
2693		 * Manually pad short frames, and zero the pad space
2694		 * to avoid leaking data.
2695		 */
2696		bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
2697		m_new->m_pkthdr.len += padlen;
2698		m_new->m_len = m_new->m_pkthdr.len;
2699		*m_head = m_new;
2700	}
2701
2702	prod = sc->rl_ldata.rl_tx_prodidx;
2703	txd = &sc->rl_ldata.rl_tx_desc[prod];
2704	error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2705	    *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2706	if (error == EFBIG) {
2707		m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS);
2708		if (m_new == NULL) {
2709			m_freem(*m_head);
2710			*m_head = NULL;
2711			return (ENOBUFS);
2712		}
2713		*m_head = m_new;
2714		error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
2715		    txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
2716		if (error != 0) {
2717			m_freem(*m_head);
2718			*m_head = NULL;
2719			return (error);
2720		}
2721	} else if (error != 0)
2722		return (error);
2723	if (nsegs == 0) {
2724		m_freem(*m_head);
2725		*m_head = NULL;
2726		return (EIO);
2727	}
2728
2729	/* Check for number of available descriptors. */
2730	if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
2731		bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
2732		return (ENOBUFS);
2733	}
2734
2735	bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
2736	    BUS_DMASYNC_PREWRITE);
2737
2738	/*
2739	 * Set up checksum offload. Note: checksum offload bits must
2740	 * appear in all descriptors of a multi-descriptor transmit
2741	 * attempt. This is according to testing done with an 8169
2742	 * chip. This is a requirement.
2743	 */
2744	vlanctl = 0;
2745	csum_flags = 0;
2746	if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2747		if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
2748			csum_flags |= RL_TDESC_CMD_LGSEND;
2749			vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2750			    RL_TDESC_CMD_MSSVALV2_SHIFT);
2751		} else {
2752			csum_flags |= RL_TDESC_CMD_LGSEND |
2753			    ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
2754			    RL_TDESC_CMD_MSSVAL_SHIFT);
2755		}
2756	} else {
2757		/*
2758		 * Unconditionally enable IP checksum if TCP or UDP
2759		 * checksum is required. Otherwise, TCP/UDP checksum
2760		 * does't make effects.
2761		 */
2762		if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
2763			if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
2764				csum_flags |= RL_TDESC_CMD_IPCSUM;
2765				if (((*m_head)->m_pkthdr.csum_flags &
2766				    CSUM_TCP) != 0)
2767					csum_flags |= RL_TDESC_CMD_TCPCSUM;
2768				if (((*m_head)->m_pkthdr.csum_flags &
2769				    CSUM_UDP) != 0)
2770					csum_flags |= RL_TDESC_CMD_UDPCSUM;
2771			} else {
2772				vlanctl |= RL_TDESC_CMD_IPCSUMV2;
2773				if (((*m_head)->m_pkthdr.csum_flags &
2774				    CSUM_TCP) != 0)
2775					vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
2776				if (((*m_head)->m_pkthdr.csum_flags &
2777				    CSUM_UDP) != 0)
2778					vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
2779			}
2780		}
2781	}
2782
2783	/*
2784	 * Set up hardware VLAN tagging. Note: vlan tag info must
2785	 * appear in all descriptors of a multi-descriptor
2786	 * transmission attempt.
2787	 */
2788	if ((*m_head)->m_flags & M_VLANTAG)
2789		vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
2790		    RL_TDESC_VLANCTL_TAG;
2791
2792	si = prod;
2793	for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
2794		desc = &sc->rl_ldata.rl_tx_list[prod];
2795		desc->rl_vlanctl = htole32(vlanctl);
2796		desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
2797		desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
2798		cmdstat = segs[i].ds_len;
2799		if (i != 0)
2800			cmdstat |= RL_TDESC_CMD_OWN;
2801		if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
2802			cmdstat |= RL_TDESC_CMD_EOR;
2803		desc->rl_cmdstat = htole32(cmdstat | csum_flags);
2804		sc->rl_ldata.rl_tx_free--;
2805	}
2806	/* Update producer index. */
2807	sc->rl_ldata.rl_tx_prodidx = prod;
2808
2809	/* Set EOF on the last descriptor. */
2810	ei = RL_TX_DESC_PRV(sc, prod);
2811	desc = &sc->rl_ldata.rl_tx_list[ei];
2812	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
2813
2814	desc = &sc->rl_ldata.rl_tx_list[si];
2815	/* Set SOF and transfer ownership of packet to the chip. */
2816	desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
2817
2818	/*
2819	 * Insure that the map for this transmission
2820	 * is placed at the array index of the last descriptor
2821	 * in this chain.  (Swap last and first dmamaps.)
2822	 */
2823	txd_last = &sc->rl_ldata.rl_tx_desc[ei];
2824	map = txd->tx_dmamap;
2825	txd->tx_dmamap = txd_last->tx_dmamap;
2826	txd_last->tx_dmamap = map;
2827	txd_last->tx_m = *m_head;
2828
2829	return (0);
2830}
2831
2832static void
2833re_start(struct ifnet *ifp)
2834{
2835	struct rl_softc		*sc;
2836
2837	sc = ifp->if_softc;
2838	RL_LOCK(sc);
2839	re_start_locked(ifp);
2840	RL_UNLOCK(sc);
2841}
2842
2843/*
2844 * Main transmit routine for C+ and gigE NICs.
2845 */
2846static void
2847re_start_locked(struct ifnet *ifp)
2848{
2849	struct rl_softc		*sc;
2850	struct mbuf		*m_head;
2851	int			queued;
2852
2853	sc = ifp->if_softc;
2854
2855	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
2856	    IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
2857		return;
2858
2859	for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
2860	    sc->rl_ldata.rl_tx_free > 1;) {
2861		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
2862		if (m_head == NULL)
2863			break;
2864
2865		if (re_encap(sc, &m_head) != 0) {
2866			if (m_head == NULL)
2867				break;
2868			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
2869			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
2870			break;
2871		}
2872
2873		/*
2874		 * If there's a BPF listener, bounce a copy of this frame
2875		 * to him.
2876		 */
2877		ETHER_BPF_MTAP(ifp, m_head);
2878
2879		queued++;
2880	}
2881
2882	if (queued == 0) {
2883#ifdef RE_TX_MODERATION
2884		if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
2885			CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2886#endif
2887		return;
2888	}
2889
2890	/* Flush the TX descriptors */
2891
2892	bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
2893	    sc->rl_ldata.rl_tx_list_map,
2894	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
2895
2896	CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
2897
2898#ifdef RE_TX_MODERATION
2899	/*
2900	 * Use the countdown timer for interrupt moderation.
2901	 * 'TX done' interrupts are disabled. Instead, we reset the
2902	 * countdown timer, which will begin counting until it hits
2903	 * the value in the TIMERINT register, and then trigger an
2904	 * interrupt. Each time we write to the TIMERCNT register,
2905	 * the timer count is reset to 0.
2906	 */
2907	CSR_WRITE_4(sc, RL_TIMERCNT, 1);
2908#endif
2909
2910	/*
2911	 * Set a timeout in case the chip goes out to lunch.
2912	 */
2913	sc->rl_watchdog_timer = 5;
2914}
2915
2916static void
2917re_set_jumbo(struct rl_softc *sc, int jumbo)
2918{
2919
2920	if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
2921		pci_set_max_read_req(sc->rl_dev, 4096);
2922		return;
2923	}
2924
2925	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
2926	if (jumbo != 0) {
2927		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
2928		    RL_CFG3_JUMBO_EN0);
2929		switch (sc->rl_hwrev->rl_rev) {
2930		case RL_HWREV_8168DP:
2931			break;
2932		case RL_HWREV_8168E:
2933			CSR_WRITE_1(sc, sc->rl_cfg4,
2934			    CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
2935			break;
2936		default:
2937			CSR_WRITE_1(sc, sc->rl_cfg4,
2938			    CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
2939		}
2940	} else {
2941		CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
2942		    ~RL_CFG3_JUMBO_EN0);
2943		switch (sc->rl_hwrev->rl_rev) {
2944		case RL_HWREV_8168DP:
2945			break;
2946		case RL_HWREV_8168E:
2947			CSR_WRITE_1(sc, sc->rl_cfg4,
2948			    CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
2949			break;
2950		default:
2951			CSR_WRITE_1(sc, sc->rl_cfg4,
2952			    CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
2953		}
2954	}
2955	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
2956
2957	switch (sc->rl_hwrev->rl_rev) {
2958	case RL_HWREV_8168DP:
2959		pci_set_max_read_req(sc->rl_dev, 4096);
2960		break;
2961	default:
2962		if (jumbo != 0)
2963			pci_set_max_read_req(sc->rl_dev, 512);
2964		else
2965			pci_set_max_read_req(sc->rl_dev, 4096);
2966	}
2967}
2968
2969static void
2970re_init(void *xsc)
2971{
2972	struct rl_softc		*sc = xsc;
2973
2974	RL_LOCK(sc);
2975	re_init_locked(sc);
2976	RL_UNLOCK(sc);
2977}
2978
2979static void
2980re_init_locked(struct rl_softc *sc)
2981{
2982	struct ifnet		*ifp = sc->rl_ifp;
2983	struct mii_data		*mii;
2984	uint32_t		reg;
2985	uint16_t		cfg;
2986	union {
2987		uint32_t align_dummy;
2988		u_char eaddr[ETHER_ADDR_LEN];
2989        } eaddr;
2990
2991	RL_LOCK_ASSERT(sc);
2992
2993	mii = device_get_softc(sc->rl_miibus);
2994
2995	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2996		return;
2997
2998	/*
2999	 * Cancel pending I/O and free all RX/TX buffers.
3000	 */
3001	re_stop(sc);
3002
3003	/* Put controller into known state. */
3004	re_reset(sc);
3005
3006	/*
3007	 * For C+ mode, initialize the RX descriptors and mbufs.
3008	 */
3009	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3010		if (ifp->if_mtu > RL_MTU) {
3011			if (re_jrx_list_init(sc) != 0) {
3012				device_printf(sc->rl_dev,
3013				    "no memory for jumbo RX buffers\n");
3014				re_stop(sc);
3015				return;
3016			}
3017			/* Disable checksum offloading for jumbo frames. */
3018			ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
3019			ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
3020		} else {
3021			if (re_rx_list_init(sc) != 0) {
3022				device_printf(sc->rl_dev,
3023				    "no memory for RX buffers\n");
3024				re_stop(sc);
3025				return;
3026			}
3027		}
3028		re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
3029	} else {
3030		if (re_rx_list_init(sc) != 0) {
3031			device_printf(sc->rl_dev, "no memory for RX buffers\n");
3032			re_stop(sc);
3033			return;
3034		}
3035		if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3036		    pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
3037			if (ifp->if_mtu > RL_MTU)
3038				pci_set_max_read_req(sc->rl_dev, 512);
3039			else
3040				pci_set_max_read_req(sc->rl_dev, 4096);
3041		}
3042	}
3043	re_tx_list_init(sc);
3044
3045	/*
3046	 * Enable C+ RX and TX mode, as well as VLAN stripping and
3047	 * RX checksum offload. We must configure the C+ register
3048	 * before all others.
3049	 */
3050	cfg = RL_CPLUSCMD_PCI_MRW;
3051	if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
3052		cfg |= RL_CPLUSCMD_RXCSUM_ENB;
3053	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3054		cfg |= RL_CPLUSCMD_VLANSTRIP;
3055	if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
3056		cfg |= RL_CPLUSCMD_MACSTAT_DIS;
3057		/* XXX magic. */
3058		cfg |= 0x0001;
3059	} else
3060		cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
3061	CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
3062	if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
3063	    sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
3064		reg = 0x000fff00;
3065		if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
3066			reg |= 0x000000ff;
3067		if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
3068			reg |= 0x00f00000;
3069		CSR_WRITE_4(sc, 0x7c, reg);
3070		/* Disable interrupt mitigation. */
3071		CSR_WRITE_2(sc, 0xe2, 0);
3072	}
3073	/*
3074	 * Disable TSO if interface MTU size is greater than MSS
3075	 * allowed in controller.
3076	 */
3077	if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
3078		ifp->if_capenable &= ~IFCAP_TSO4;
3079		ifp->if_hwassist &= ~CSUM_TSO;
3080	}
3081
3082	/*
3083	 * Init our MAC address.  Even though the chipset
3084	 * documentation doesn't mention it, we need to enter "Config
3085	 * register write enable" mode to modify the ID registers.
3086	 */
3087	/* Copy MAC address on stack to align. */
3088	bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
3089	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
3090	CSR_WRITE_4(sc, RL_IDR0,
3091	    htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
3092	CSR_WRITE_4(sc, RL_IDR4,
3093	    htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
3094	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3095
3096	/*
3097	 * Load the addresses of the RX and TX lists into the chip.
3098	 */
3099
3100	CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
3101	    RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
3102	CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
3103	    RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
3104
3105	CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
3106	    RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
3107	CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
3108	    RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
3109
3110	/*
3111	 * Enable transmit and receive.
3112	 */
3113	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3114
3115	/*
3116	 * Set the initial TX configuration.
3117	 */
3118	if (sc->rl_testmode) {
3119		if (sc->rl_type == RL_8169)
3120			CSR_WRITE_4(sc, RL_TXCFG,
3121			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
3122		else
3123			CSR_WRITE_4(sc, RL_TXCFG,
3124			    RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
3125	} else
3126		CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
3127
3128	CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
3129
3130	/*
3131	 * Set the initial RX configuration.
3132	 */
3133	re_set_rxmode(sc);
3134
3135	/* Configure interrupt moderation. */
3136	if (sc->rl_type == RL_8169) {
3137		/* Magic from vendor. */
3138		CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
3139	}
3140
3141#ifdef DEVICE_POLLING
3142	/*
3143	 * Disable interrupts if we are polling.
3144	 */
3145	if (ifp->if_capenable & IFCAP_POLLING)
3146		CSR_WRITE_2(sc, RL_IMR, 0);
3147	else	/* otherwise ... */
3148#endif
3149
3150	/*
3151	 * Enable interrupts.
3152	 */
3153	if (sc->rl_testmode)
3154		CSR_WRITE_2(sc, RL_IMR, 0);
3155	else
3156		CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3157	CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
3158
3159	/* Set initial TX threshold */
3160	sc->rl_txthresh = RL_TX_THRESH_INIT;
3161
3162	/* Start RX/TX process. */
3163	CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
3164#ifdef notdef
3165	/* Enable receiver and transmitter. */
3166	CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
3167#endif
3168
3169	/*
3170	 * Initialize the timer interrupt register so that
3171	 * a timer interrupt will be generated once the timer
3172	 * reaches a certain number of ticks. The timer is
3173	 * reloaded on each transmit.
3174	 */
3175#ifdef RE_TX_MODERATION
3176	/*
3177	 * Use timer interrupt register to moderate TX interrupt
3178	 * moderation, which dramatically improves TX frame rate.
3179	 */
3180	if (sc->rl_type == RL_8169)
3181		CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
3182	else
3183		CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
3184#else
3185	/*
3186	 * Use timer interrupt register to moderate RX interrupt
3187	 * moderation.
3188	 */
3189	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
3190	    intr_filter == 0) {
3191		if (sc->rl_type == RL_8169)
3192			CSR_WRITE_4(sc, RL_TIMERINT_8169,
3193			    RL_USECS(sc->rl_int_rx_mod));
3194	} else {
3195		if (sc->rl_type == RL_8169)
3196			CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
3197	}
3198#endif
3199
3200	/*
3201	 * For 8169 gigE NICs, set the max allowed RX packet
3202	 * size so we can receive jumbo frames.
3203	 */
3204	if (sc->rl_type == RL_8169) {
3205		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3206			/*
3207			 * For controllers that use new jumbo frame scheme,
3208			 * set maximum size of jumbo frame depedning on
3209			 * controller revisions.
3210			 */
3211			if (ifp->if_mtu > RL_MTU)
3212				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3213				    sc->rl_hwrev->rl_max_mtu +
3214				    ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
3215				    ETHER_CRC_LEN);
3216			else
3217				CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
3218				    RE_RX_DESC_BUFLEN);
3219		} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
3220		    sc->rl_hwrev->rl_max_mtu == RL_MTU) {
3221			/* RTL810x has no jumbo frame support. */
3222			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
3223		} else
3224			CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
3225	}
3226
3227	if (sc->rl_testmode)
3228		return;
3229
3230	CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
3231	    RL_CFG1_DRVLOAD);
3232
3233	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3234	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
3235
3236	sc->rl_flags &= ~RL_FLAG_LINK;
3237	mii_mediachg(mii);
3238
3239	sc->rl_watchdog_timer = 0;
3240	callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
3241}
3242
3243/*
3244 * Set media options.
3245 */
3246static int
3247re_ifmedia_upd(struct ifnet *ifp)
3248{
3249	struct rl_softc		*sc;
3250	struct mii_data		*mii;
3251	int			error;
3252
3253	sc = ifp->if_softc;
3254	mii = device_get_softc(sc->rl_miibus);
3255	RL_LOCK(sc);
3256	error = mii_mediachg(mii);
3257	RL_UNLOCK(sc);
3258
3259	return (error);
3260}
3261
3262/*
3263 * Report current media status.
3264 */
3265static void
3266re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3267{
3268	struct rl_softc		*sc;
3269	struct mii_data		*mii;
3270
3271	sc = ifp->if_softc;
3272	mii = device_get_softc(sc->rl_miibus);
3273
3274	RL_LOCK(sc);
3275	mii_pollstat(mii);
3276	ifmr->ifm_active = mii->mii_media_active;
3277	ifmr->ifm_status = mii->mii_media_status;
3278	RL_UNLOCK(sc);
3279}
3280
3281static int
3282re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
3283{
3284	struct rl_softc		*sc = ifp->if_softc;
3285	struct ifreq		*ifr = (struct ifreq *) data;
3286	struct mii_data		*mii;
3287	uint32_t		rev;
3288	int			error = 0;
3289
3290	switch (command) {
3291	case SIOCSIFMTU:
3292		if (ifr->ifr_mtu < ETHERMIN ||
3293		    ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) {
3294			error = EINVAL;
3295			break;
3296		}
3297		RL_LOCK(sc);
3298		if (ifp->if_mtu != ifr->ifr_mtu) {
3299			ifp->if_mtu = ifr->ifr_mtu;
3300			if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3301			    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3302				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3303				re_init_locked(sc);
3304			}
3305			if (ifp->if_mtu > RL_TSO_MTU &&
3306			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3307				ifp->if_capenable &= ~(IFCAP_TSO4 |
3308				    IFCAP_VLAN_HWTSO);
3309				ifp->if_hwassist &= ~CSUM_TSO;
3310			}
3311			VLAN_CAPABILITIES(ifp);
3312		}
3313		RL_UNLOCK(sc);
3314		break;
3315	case SIOCSIFFLAGS:
3316		RL_LOCK(sc);
3317		if ((ifp->if_flags & IFF_UP) != 0) {
3318			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
3319				if (((ifp->if_flags ^ sc->rl_if_flags)
3320				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
3321					re_set_rxmode(sc);
3322			} else
3323				re_init_locked(sc);
3324		} else {
3325			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3326				re_stop(sc);
3327		}
3328		sc->rl_if_flags = ifp->if_flags;
3329		RL_UNLOCK(sc);
3330		break;
3331	case SIOCADDMULTI:
3332	case SIOCDELMULTI:
3333		RL_LOCK(sc);
3334		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
3335			re_set_rxmode(sc);
3336		RL_UNLOCK(sc);
3337		break;
3338	case SIOCGIFMEDIA:
3339	case SIOCSIFMEDIA:
3340		mii = device_get_softc(sc->rl_miibus);
3341		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
3342		break;
3343	case SIOCSIFCAP:
3344	    {
3345		int mask, reinit;
3346
3347		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3348		reinit = 0;
3349#ifdef DEVICE_POLLING
3350		if (mask & IFCAP_POLLING) {
3351			if (ifr->ifr_reqcap & IFCAP_POLLING) {
3352				error = ether_poll_register(re_poll, ifp);
3353				if (error)
3354					return (error);
3355				RL_LOCK(sc);
3356				/* Disable interrupts */
3357				CSR_WRITE_2(sc, RL_IMR, 0x0000);
3358				ifp->if_capenable |= IFCAP_POLLING;
3359				RL_UNLOCK(sc);
3360			} else {
3361				error = ether_poll_deregister(ifp);
3362				/* Enable interrupts. */
3363				RL_LOCK(sc);
3364				CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
3365				ifp->if_capenable &= ~IFCAP_POLLING;
3366				RL_UNLOCK(sc);
3367			}
3368		}
3369#endif /* DEVICE_POLLING */
3370		RL_LOCK(sc);
3371		if ((mask & IFCAP_TXCSUM) != 0 &&
3372		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
3373			ifp->if_capenable ^= IFCAP_TXCSUM;
3374			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
3375				rev = sc->rl_hwrev->rl_rev;
3376				if (rev == RL_HWREV_8168C ||
3377				    rev == RL_HWREV_8168C_SPIN2)
3378					ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
3379				else
3380					ifp->if_hwassist |= RE_CSUM_FEATURES;
3381			} else
3382				ifp->if_hwassist &= ~RE_CSUM_FEATURES;
3383			reinit = 1;
3384		}
3385		if ((mask & IFCAP_RXCSUM) != 0 &&
3386		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
3387			ifp->if_capenable ^= IFCAP_RXCSUM;
3388			reinit = 1;
3389		}
3390		if ((mask & IFCAP_TSO4) != 0 &&
3391		    (ifp->if_capabilities & IFCAP_TSO) != 0) {
3392			ifp->if_capenable ^= IFCAP_TSO4;
3393			if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
3394				ifp->if_hwassist |= CSUM_TSO;
3395			else
3396				ifp->if_hwassist &= ~CSUM_TSO;
3397			if (ifp->if_mtu > RL_TSO_MTU &&
3398			    (ifp->if_capenable & IFCAP_TSO4) != 0) {
3399				ifp->if_capenable &= ~IFCAP_TSO4;
3400				ifp->if_hwassist &= ~CSUM_TSO;
3401			}
3402		}
3403		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
3404		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
3405			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
3406		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
3407		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
3408			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
3409			/* TSO over VLAN requires VLAN hardware tagging. */
3410			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
3411				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
3412			reinit = 1;
3413		}
3414		if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
3415		    (mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
3416		    IFCAP_VLAN_HWTSO)) != 0)
3417				reinit = 1;
3418		if ((mask & IFCAP_WOL) != 0 &&
3419		    (ifp->if_capabilities & IFCAP_WOL) != 0) {
3420			if ((mask & IFCAP_WOL_UCAST) != 0)
3421				ifp->if_capenable ^= IFCAP_WOL_UCAST;
3422			if ((mask & IFCAP_WOL_MCAST) != 0)
3423				ifp->if_capenable ^= IFCAP_WOL_MCAST;
3424			if ((mask & IFCAP_WOL_MAGIC) != 0)
3425				ifp->if_capenable ^= IFCAP_WOL_MAGIC;
3426		}
3427		if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
3428			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3429			re_init_locked(sc);
3430		}
3431		RL_UNLOCK(sc);
3432		VLAN_CAPABILITIES(ifp);
3433	    }
3434		break;
3435	default:
3436		error = ether_ioctl(ifp, command, data);
3437		break;
3438	}
3439
3440	return (error);
3441}
3442
3443static void
3444re_watchdog(struct rl_softc *sc)
3445{
3446	struct ifnet		*ifp;
3447
3448	RL_LOCK_ASSERT(sc);
3449
3450	if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
3451		return;
3452
3453	ifp = sc->rl_ifp;
3454	re_txeof(sc);
3455	if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
3456		if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
3457		    "-- recovering\n");
3458		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3459			re_start_locked(ifp);
3460		return;
3461	}
3462
3463	if_printf(ifp, "watchdog timeout\n");
3464	ifp->if_oerrors++;
3465
3466	re_rxeof(sc, NULL);
3467	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3468	re_init_locked(sc);
3469	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
3470		re_start_locked(ifp);
3471}
3472
3473/*
3474 * Stop the adapter and free any mbufs allocated to the
3475 * RX and TX lists.
3476 */
3477static void
3478re_stop(struct rl_softc *sc)
3479{
3480	int			i;
3481	struct ifnet		*ifp;
3482	struct rl_txdesc	*txd;
3483	struct rl_rxdesc	*rxd;
3484
3485	RL_LOCK_ASSERT(sc);
3486
3487	ifp = sc->rl_ifp;
3488
3489	sc->rl_watchdog_timer = 0;
3490	callout_stop(&sc->rl_stat_callout);
3491	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
3492
3493	/*
3494	 * Disable accepting frames to put RX MAC into idle state.
3495	 * Otherwise it's possible to get frames while stop command
3496	 * execution is in progress and controller can DMA the frame
3497	 * to already freed RX buffer during that period.
3498	 */
3499	CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
3500	    ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
3501	    RL_RXCFG_RX_BROAD));
3502
3503	if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
3504		for (i = RL_TIMEOUT; i > 0; i--) {
3505			if ((CSR_READ_1(sc, sc->rl_txstart) &
3506			    RL_TXSTART_START) == 0)
3507				break;
3508			DELAY(20);
3509		}
3510		if (i == 0)
3511			device_printf(sc->rl_dev,
3512			    "stopping TX poll timed out!\n");
3513		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3514	} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
3515		CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
3516		    RL_CMD_RX_ENB);
3517		if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
3518			for (i = RL_TIMEOUT; i > 0; i--) {
3519				if ((CSR_READ_4(sc, RL_TXCFG) &
3520				    RL_TXCFG_QUEUE_EMPTY) != 0)
3521					break;
3522				DELAY(100);
3523			}
3524			if (i == 0)
3525				device_printf(sc->rl_dev,
3526				   "stopping TXQ timed out!\n");
3527		}
3528	} else
3529		CSR_WRITE_1(sc, RL_COMMAND, 0x00);
3530	DELAY(1000);
3531	CSR_WRITE_2(sc, RL_IMR, 0x0000);
3532	CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
3533
3534	if (sc->rl_head != NULL) {
3535		m_freem(sc->rl_head);
3536		sc->rl_head = sc->rl_tail = NULL;
3537	}
3538
3539	/* Free the TX list buffers. */
3540	for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
3541		txd = &sc->rl_ldata.rl_tx_desc[i];
3542		if (txd->tx_m != NULL) {
3543			bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
3544			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
3545			bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
3546			    txd->tx_dmamap);
3547			m_freem(txd->tx_m);
3548			txd->tx_m = NULL;
3549		}
3550	}
3551
3552	/* Free the RX list buffers. */
3553	for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3554		rxd = &sc->rl_ldata.rl_rx_desc[i];
3555		if (rxd->rx_m != NULL) {
3556			bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
3557			    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3558			bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
3559			    rxd->rx_dmamap);
3560			m_freem(rxd->rx_m);
3561			rxd->rx_m = NULL;
3562		}
3563	}
3564
3565	if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
3566		for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
3567			rxd = &sc->rl_ldata.rl_jrx_desc[i];
3568			if (rxd->rx_m != NULL) {
3569				bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
3570				    rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
3571				bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
3572				    rxd->rx_dmamap);
3573				m_freem(rxd->rx_m);
3574				rxd->rx_m = NULL;
3575			}
3576		}
3577	}
3578}
3579
3580/*
3581 * Device suspend routine.  Stop the interface and save some PCI
3582 * settings in case the BIOS doesn't restore them properly on
3583 * resume.
3584 */
3585static int
3586re_suspend(device_t dev)
3587{
3588	struct rl_softc		*sc;
3589
3590	sc = device_get_softc(dev);
3591
3592	RL_LOCK(sc);
3593	re_stop(sc);
3594	re_setwol(sc);
3595	sc->suspended = 1;
3596	RL_UNLOCK(sc);
3597
3598	return (0);
3599}
3600
3601/*
3602 * Device resume routine.  Restore some PCI settings in case the BIOS
3603 * doesn't, re-enable busmastering, and restart the interface if
3604 * appropriate.
3605 */
3606static int
3607re_resume(device_t dev)
3608{
3609	struct rl_softc		*sc;
3610	struct ifnet		*ifp;
3611
3612	sc = device_get_softc(dev);
3613
3614	RL_LOCK(sc);
3615
3616	ifp = sc->rl_ifp;
3617	/* Take controller out of sleep mode. */
3618	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3619		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3620			CSR_WRITE_1(sc, RL_GPIO,
3621			    CSR_READ_1(sc, RL_GPIO) | 0x01);
3622	}
3623
3624	/*
3625	 * Clear WOL matching such that normal Rx filtering
3626	 * wouldn't interfere with WOL patterns.
3627	 */
3628	re_clrwol(sc);
3629
3630	/* reinitialize interface if necessary */
3631	if (ifp->if_flags & IFF_UP)
3632		re_init_locked(sc);
3633
3634	sc->suspended = 0;
3635	RL_UNLOCK(sc);
3636
3637	return (0);
3638}
3639
3640/*
3641 * Stop all chip I/O so that the kernel's probe routines don't
3642 * get confused by errant DMAs when rebooting.
3643 */
3644static int
3645re_shutdown(device_t dev)
3646{
3647	struct rl_softc		*sc;
3648
3649	sc = device_get_softc(dev);
3650
3651	RL_LOCK(sc);
3652	re_stop(sc);
3653	/*
3654	 * Mark interface as down since otherwise we will panic if
3655	 * interrupt comes in later on, which can happen in some
3656	 * cases.
3657	 */
3658	sc->rl_ifp->if_flags &= ~IFF_UP;
3659	re_setwol(sc);
3660	RL_UNLOCK(sc);
3661
3662	return (0);
3663}
3664
3665static void
3666re_set_linkspeed(struct rl_softc *sc)
3667{
3668	struct mii_softc *miisc;
3669	struct mii_data *mii;
3670	int aneg, i, phyno;
3671
3672	RL_LOCK_ASSERT(sc);
3673
3674	mii = device_get_softc(sc->rl_miibus);
3675	mii_pollstat(mii);
3676	aneg = 0;
3677	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
3678	    (IFM_ACTIVE | IFM_AVALID)) {
3679		switch IFM_SUBTYPE(mii->mii_media_active) {
3680		case IFM_10_T:
3681		case IFM_100_TX:
3682			return;
3683		case IFM_1000_T:
3684			aneg++;
3685			break;
3686		default:
3687			break;
3688		}
3689	}
3690	miisc = LIST_FIRST(&mii->mii_phys);
3691	phyno = miisc->mii_phy;
3692	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3693		PHY_RESET(miisc);
3694	re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
3695	re_miibus_writereg(sc->rl_dev, phyno,
3696	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
3697	re_miibus_writereg(sc->rl_dev, phyno,
3698	    MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
3699	DELAY(1000);
3700	if (aneg != 0) {
3701		/*
3702		 * Poll link state until re(4) get a 10/100Mbps link.
3703		 */
3704		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
3705			mii_pollstat(mii);
3706			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
3707			    == (IFM_ACTIVE | IFM_AVALID)) {
3708				switch (IFM_SUBTYPE(mii->mii_media_active)) {
3709				case IFM_10_T:
3710				case IFM_100_TX:
3711					return;
3712				default:
3713					break;
3714				}
3715			}
3716			RL_UNLOCK(sc);
3717			pause("relnk", hz);
3718			RL_LOCK(sc);
3719		}
3720		if (i == MII_ANEGTICKS_GIGE)
3721			device_printf(sc->rl_dev,
3722			    "establishing a link failed, WOL may not work!");
3723	}
3724	/*
3725	 * No link, force MAC to have 100Mbps, full-duplex link.
3726	 * MAC does not require reprogramming on resolved speed/duplex,
3727	 * so this is just for completeness.
3728	 */
3729	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
3730	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
3731}
3732
3733static void
3734re_setwol(struct rl_softc *sc)
3735{
3736	struct ifnet		*ifp;
3737	int			pmc;
3738	uint16_t		pmstat;
3739	uint8_t			v;
3740
3741	RL_LOCK_ASSERT(sc);
3742
3743	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3744		return;
3745
3746	ifp = sc->rl_ifp;
3747	/* Put controller into sleep mode. */
3748	if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
3749		if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
3750			CSR_WRITE_1(sc, RL_GPIO,
3751			    CSR_READ_1(sc, RL_GPIO) & ~0x01);
3752	}
3753	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
3754		re_set_rxmode(sc);
3755		if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
3756			re_set_linkspeed(sc);
3757		if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
3758			CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
3759	}
3760	/* Enable config register write. */
3761	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3762
3763	/* Enable PME. */
3764	v = CSR_READ_1(sc, sc->rl_cfg1);
3765	v &= ~RL_CFG1_PME;
3766	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3767		v |= RL_CFG1_PME;
3768	CSR_WRITE_1(sc, sc->rl_cfg1, v);
3769
3770	v = CSR_READ_1(sc, sc->rl_cfg3);
3771	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3772	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
3773		v |= RL_CFG3_WOL_MAGIC;
3774	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3775
3776	v = CSR_READ_1(sc, sc->rl_cfg5);
3777	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
3778	    RL_CFG5_WOL_LANWAKE);
3779	if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
3780		v |= RL_CFG5_WOL_UCAST;
3781	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
3782		v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
3783	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3784		v |= RL_CFG5_WOL_LANWAKE;
3785	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3786
3787	/* Config register write done. */
3788	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3789
3790	if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
3791	    (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
3792		CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
3793	/*
3794	 * It seems that hardware resets its link speed to 100Mbps in
3795	 * power down mode so switching to 100Mbps in driver is not
3796	 * needed.
3797	 */
3798
3799	/* Request PME if WOL is requested. */
3800	pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
3801	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3802	if ((ifp->if_capenable & IFCAP_WOL) != 0)
3803		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3804	pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
3805}
3806
3807static void
3808re_clrwol(struct rl_softc *sc)
3809{
3810	int			pmc;
3811	uint8_t			v;
3812
3813	RL_LOCK_ASSERT(sc);
3814
3815	if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
3816		return;
3817
3818	/* Enable config register write. */
3819	CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
3820
3821	v = CSR_READ_1(sc, sc->rl_cfg3);
3822	v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
3823	CSR_WRITE_1(sc, sc->rl_cfg3, v);
3824
3825	/* Config register write done. */
3826	CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
3827
3828	v = CSR_READ_1(sc, sc->rl_cfg5);
3829	v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
3830	v &= ~RL_CFG5_WOL_LANWAKE;
3831	CSR_WRITE_1(sc, sc->rl_cfg5, v);
3832}
3833
3834static void
3835re_add_sysctls(struct rl_softc *sc)
3836{
3837	struct sysctl_ctx_list	*ctx;
3838	struct sysctl_oid_list	*children;
3839	int			error;
3840
3841	ctx = device_get_sysctl_ctx(sc->rl_dev);
3842	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
3843
3844	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
3845	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
3846	    "Statistics Information");
3847	if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
3848		return;
3849
3850	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
3851	    CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
3852	    sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
3853	/* Pull in device tunables. */
3854	sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3855	error = resource_int_value(device_get_name(sc->rl_dev),
3856	    device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
3857	if (error == 0) {
3858		if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
3859		    sc->rl_int_rx_mod > RL_TIMER_MAX) {
3860			device_printf(sc->rl_dev, "int_rx_mod value out of "
3861			    "range; using default: %d\n",
3862			    RL_TIMER_DEFAULT);
3863			sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
3864		}
3865	}
3866
3867}
3868
3869static int
3870re_sysctl_stats(SYSCTL_HANDLER_ARGS)
3871{
3872	struct rl_softc		*sc;
3873	struct rl_stats		*stats;
3874	int			error, i, result;
3875
3876	result = -1;
3877	error = sysctl_handle_int(oidp, &result, 0, req);
3878	if (error || req->newptr == NULL)
3879		return (error);
3880
3881	if (result == 1) {
3882		sc = (struct rl_softc *)arg1;
3883		RL_LOCK(sc);
3884		if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
3885			RL_UNLOCK(sc);
3886			goto done;
3887		}
3888		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3889		    sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
3890		CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
3891		    RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
3892		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3893		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
3894		CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
3895		    RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
3896		    RL_DUMPSTATS_START));
3897		for (i = RL_TIMEOUT; i > 0; i--) {
3898			if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
3899			    RL_DUMPSTATS_START) == 0)
3900				break;
3901			DELAY(1000);
3902		}
3903		bus_dmamap_sync(sc->rl_ldata.rl_stag,
3904		    sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
3905		RL_UNLOCK(sc);
3906		if (i == 0) {
3907			device_printf(sc->rl_dev,
3908			    "DUMP statistics request timedout\n");
3909			return (ETIMEDOUT);
3910		}
3911done:
3912		stats = sc->rl_ldata.rl_stats;
3913		printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
3914		printf("Tx frames : %ju\n",
3915		    (uintmax_t)le64toh(stats->rl_tx_pkts));
3916		printf("Rx frames : %ju\n",
3917		    (uintmax_t)le64toh(stats->rl_rx_pkts));
3918		printf("Tx errors : %ju\n",
3919		    (uintmax_t)le64toh(stats->rl_tx_errs));
3920		printf("Rx errors : %u\n",
3921		    le32toh(stats->rl_rx_errs));
3922		printf("Rx missed frames : %u\n",
3923		    (uint32_t)le16toh(stats->rl_missed_pkts));
3924		printf("Rx frame alignment errs : %u\n",
3925		    (uint32_t)le16toh(stats->rl_rx_framealign_errs));
3926		printf("Tx single collisions : %u\n",
3927		    le32toh(stats->rl_tx_onecoll));
3928		printf("Tx multiple collisions : %u\n",
3929		    le32toh(stats->rl_tx_multicolls));
3930		printf("Rx unicast frames : %ju\n",
3931		    (uintmax_t)le64toh(stats->rl_rx_ucasts));
3932		printf("Rx broadcast frames : %ju\n",
3933		    (uintmax_t)le64toh(stats->rl_rx_bcasts));
3934		printf("Rx multicast frames : %u\n",
3935		    le32toh(stats->rl_rx_mcasts));
3936		printf("Tx aborts : %u\n",
3937		    (uint32_t)le16toh(stats->rl_tx_aborts));
3938		printf("Tx underruns : %u\n",
3939		    (uint32_t)le16toh(stats->rl_rx_underruns));
3940	}
3941
3942	return (error);
3943}
3944
3945static int
3946sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3947{
3948	int error, value;
3949
3950	if (arg1 == NULL)
3951		return (EINVAL);
3952	value = *(int *)arg1;
3953	error = sysctl_handle_int(oidp, &value, 0, req);
3954	if (error || req->newptr == NULL)
3955		return (error);
3956	if (value < low || value > high)
3957		return (EINVAL);
3958	*(int *)arg1 = value;
3959
3960	return (0);
3961}
3962
3963static int
3964sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
3965{
3966
3967	return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
3968	    RL_TIMER_MAX));
3969}
3970