sfxge.c revision 291843
1/*-
2 * Copyright (c) 2010-2015 Solarflare Communications Inc.
3 * All rights reserved.
4 *
5 * This software was developed in part by Philip Paeps under contract for
6 * Solarflare Communications, Inc.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright notice,
12 *    this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright notice,
14 *    this list of conditions and the following disclaimer in the documentation
15 *    and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
19 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
21 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
22 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
26 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
27 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * The views and conclusions contained in the software and documentation are
30 * those of the authors and should not be interpreted as representing official
31 * policies, either expressed or implied, of the FreeBSD Project.
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: head/sys/dev/sfxge/sfxge.c 291843 2015-12-05 07:04:11Z arybchik $");
36
37#include <sys/param.h>
38#include <sys/kernel.h>
39#include <sys/bus.h>
40#include <sys/rman.h>
41#include <sys/lock.h>
42#include <sys/module.h>
43#include <sys/mutex.h>
44#include <sys/smp.h>
45#include <sys/socket.h>
46#include <sys/taskqueue.h>
47#include <sys/sockio.h>
48#include <sys/sysctl.h>
49#include <sys/priv.h>
50#include <sys/syslog.h>
51
52#include <dev/pci/pcireg.h>
53#include <dev/pci/pcivar.h>
54
55#include <net/ethernet.h>
56#include <net/if.h>
57#include <net/if_var.h>
58#include <net/if_media.h>
59#include <net/if_types.h>
60
61#include "common/efx.h"
62
63#include "sfxge.h"
64#include "sfxge_rx.h"
65#include "sfxge_ioc.h"
66#include "sfxge_version.h"
67
68#define	SFXGE_CAP (IFCAP_VLAN_MTU | IFCAP_VLAN_HWCSUM |			\
69		   IFCAP_RXCSUM | IFCAP_TXCSUM |			\
70		   IFCAP_RXCSUM_IPV6 | IFCAP_TXCSUM_IPV6 |		\
71		   IFCAP_TSO4 | IFCAP_TSO6 |				\
72		   IFCAP_JUMBO_MTU |					\
73		   IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWSTATS)
74#define	SFXGE_CAP_ENABLE SFXGE_CAP
75#define	SFXGE_CAP_FIXED (IFCAP_VLAN_MTU |				\
76			 IFCAP_JUMBO_MTU | IFCAP_LINKSTATE | IFCAP_HWSTATS)
77
78MALLOC_DEFINE(M_SFXGE, "sfxge", "Solarflare 10GigE driver");
79
80
81SYSCTL_NODE(_hw, OID_AUTO, sfxge, CTLFLAG_RD, 0,
82	    "SFXGE driver parameters");
83
84#define	SFXGE_PARAM_RX_RING	SFXGE_PARAM(rx_ring)
85static int sfxge_rx_ring_entries = SFXGE_NDESCS;
86TUNABLE_INT(SFXGE_PARAM_RX_RING, &sfxge_rx_ring_entries);
87SYSCTL_INT(_hw_sfxge, OID_AUTO, rx_ring, CTLFLAG_RDTUN,
88	   &sfxge_rx_ring_entries, 0,
89	   "Maximum number of descriptors in a receive ring");
90
91#define	SFXGE_PARAM_TX_RING	SFXGE_PARAM(tx_ring)
92static int sfxge_tx_ring_entries = SFXGE_NDESCS;
93TUNABLE_INT(SFXGE_PARAM_TX_RING, &sfxge_tx_ring_entries);
94SYSCTL_INT(_hw_sfxge, OID_AUTO, tx_ring, CTLFLAG_RDTUN,
95	   &sfxge_tx_ring_entries, 0,
96	   "Maximum number of descriptors in a transmit ring");
97
98#if EFSYS_OPT_MCDI_LOGGING
99#define	SFXGE_PARAM_MCDI_LOGGING	SFXGE_PARAM(mcdi_logging)
100static int sfxge_mcdi_logging = 0;
101TUNABLE_INT(SFXGE_PARAM_MCDI_LOGGING, &sfxge_mcdi_logging);
102#endif
103
104static void
105sfxge_reset(void *arg, int npending);
106
107static int
108sfxge_estimate_rsrc_limits(struct sfxge_softc *sc)
109{
110	efx_drv_limits_t limits;
111	int rc;
112	unsigned int evq_max;
113	uint32_t evq_allocated;
114	uint32_t rxq_allocated;
115	uint32_t txq_allocated;
116
117	/*
118	 * Limit the number of event queues to:
119	 *  - number of CPUs
120	 *  - hardwire maximum RSS channels
121	 *  - administratively specified maximum RSS channels
122	 */
123	evq_max = MIN(mp_ncpus, EFX_MAXRSS);
124	if (sc->max_rss_channels > 0)
125		evq_max = MIN(evq_max, sc->max_rss_channels);
126
127	memset(&limits, 0, sizeof(limits));
128
129	limits.edl_min_evq_count = 1;
130	limits.edl_max_evq_count = evq_max;
131	limits.edl_min_txq_count = SFXGE_TXQ_NTYPES;
132	limits.edl_max_txq_count = evq_max + SFXGE_TXQ_NTYPES - 1;
133	limits.edl_min_rxq_count = 1;
134	limits.edl_max_rxq_count = evq_max;
135
136	efx_nic_set_drv_limits(sc->enp, &limits);
137
138	if ((rc = efx_nic_init(sc->enp)) != 0)
139		return (rc);
140
141	rc = efx_nic_get_vi_pool(sc->enp, &evq_allocated, &rxq_allocated,
142				 &txq_allocated);
143	if (rc != 0) {
144		efx_nic_fini(sc->enp);
145		return (rc);
146	}
147
148	KASSERT(txq_allocated >= SFXGE_TXQ_NTYPES,
149		("txq_allocated < SFXGE_TXQ_NTYPES"));
150
151	sc->evq_max = MIN(evq_allocated, evq_max);
152	sc->evq_max = MIN(rxq_allocated, sc->evq_max);
153	sc->evq_max = MIN(txq_allocated - (SFXGE_TXQ_NTYPES - 1),
154			  sc->evq_max);
155
156	KASSERT(sc->evq_max <= evq_max,
157		("allocated more than maximum requested"));
158
159	/*
160	 * NIC is kept initialized in the case of success to be able to
161	 * initialize port to find out media types.
162	 */
163	return (0);
164}
165
166static int
167sfxge_set_drv_limits(struct sfxge_softc *sc)
168{
169	efx_drv_limits_t limits;
170
171	memset(&limits, 0, sizeof(limits));
172
173	/* Limits are strict since take into account initial estimation */
174	limits.edl_min_evq_count = limits.edl_max_evq_count =
175	    sc->intr.n_alloc;
176	limits.edl_min_txq_count = limits.edl_max_txq_count =
177	    sc->intr.n_alloc + SFXGE_TXQ_NTYPES - 1;
178	limits.edl_min_rxq_count = limits.edl_max_rxq_count =
179	    sc->intr.n_alloc;
180
181	return (efx_nic_set_drv_limits(sc->enp, &limits));
182}
183
184static int
185sfxge_start(struct sfxge_softc *sc)
186{
187	int rc;
188
189	SFXGE_ADAPTER_LOCK_ASSERT_OWNED(sc);
190
191	if (sc->init_state == SFXGE_STARTED)
192		return (0);
193
194	if (sc->init_state != SFXGE_REGISTERED) {
195		rc = EINVAL;
196		goto fail;
197	}
198
199	/* Set required resource limits */
200	if ((rc = sfxge_set_drv_limits(sc)) != 0)
201		goto fail;
202
203	if ((rc = efx_nic_init(sc->enp)) != 0)
204		goto fail;
205
206	/* Start processing interrupts. */
207	if ((rc = sfxge_intr_start(sc)) != 0)
208		goto fail2;
209
210	/* Start processing events. */
211	if ((rc = sfxge_ev_start(sc)) != 0)
212		goto fail3;
213
214	/* Fire up the port. */
215	if ((rc = sfxge_port_start(sc)) != 0)
216		goto fail4;
217
218	/* Start the receiver side. */
219	if ((rc = sfxge_rx_start(sc)) != 0)
220		goto fail5;
221
222	/* Start the transmitter side. */
223	if ((rc = sfxge_tx_start(sc)) != 0)
224		goto fail6;
225
226	sc->init_state = SFXGE_STARTED;
227
228	/* Tell the stack we're running. */
229	sc->ifnet->if_drv_flags |= IFF_DRV_RUNNING;
230	sc->ifnet->if_drv_flags &= ~IFF_DRV_OACTIVE;
231
232	return (0);
233
234fail6:
235	sfxge_rx_stop(sc);
236
237fail5:
238	sfxge_port_stop(sc);
239
240fail4:
241	sfxge_ev_stop(sc);
242
243fail3:
244	sfxge_intr_stop(sc);
245
246fail2:
247	efx_nic_fini(sc->enp);
248
249fail:
250	device_printf(sc->dev, "sfxge_start: %d\n", rc);
251
252	return (rc);
253}
254
255static void
256sfxge_if_init(void *arg)
257{
258	struct sfxge_softc *sc;
259
260	sc = (struct sfxge_softc *)arg;
261
262	SFXGE_ADAPTER_LOCK(sc);
263	(void)sfxge_start(sc);
264	SFXGE_ADAPTER_UNLOCK(sc);
265}
266
267static void
268sfxge_stop(struct sfxge_softc *sc)
269{
270	SFXGE_ADAPTER_LOCK_ASSERT_OWNED(sc);
271
272	if (sc->init_state != SFXGE_STARTED)
273		return;
274
275	sc->init_state = SFXGE_REGISTERED;
276
277	/* Stop the transmitter. */
278	sfxge_tx_stop(sc);
279
280	/* Stop the receiver. */
281	sfxge_rx_stop(sc);
282
283	/* Stop the port. */
284	sfxge_port_stop(sc);
285
286	/* Stop processing events. */
287	sfxge_ev_stop(sc);
288
289	/* Stop processing interrupts. */
290	sfxge_intr_stop(sc);
291
292	efx_nic_fini(sc->enp);
293
294	sc->ifnet->if_drv_flags &= ~IFF_DRV_RUNNING;
295}
296
297
298static int
299sfxge_vpd_ioctl(struct sfxge_softc *sc, sfxge_ioc_t *ioc)
300{
301	efx_vpd_value_t value;
302	int rc = 0;
303
304	switch (ioc->u.vpd.op) {
305	case SFXGE_VPD_OP_GET_KEYWORD:
306		value.evv_tag = ioc->u.vpd.tag;
307		value.evv_keyword = ioc->u.vpd.keyword;
308		rc = efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value);
309		if (rc != 0)
310			break;
311		ioc->u.vpd.len = MIN(ioc->u.vpd.len, value.evv_length);
312		if (ioc->u.vpd.payload != 0) {
313			rc = copyout(value.evv_value, ioc->u.vpd.payload,
314				     ioc->u.vpd.len);
315		}
316		break;
317	case SFXGE_VPD_OP_SET_KEYWORD:
318		if (ioc->u.vpd.len > sizeof(value.evv_value))
319			return (EINVAL);
320		value.evv_tag = ioc->u.vpd.tag;
321		value.evv_keyword = ioc->u.vpd.keyword;
322		value.evv_length = ioc->u.vpd.len;
323		rc = copyin(ioc->u.vpd.payload, value.evv_value, value.evv_length);
324		if (rc != 0)
325			break;
326		rc = efx_vpd_set(sc->enp, sc->vpd_data, sc->vpd_size, &value);
327		if (rc != 0)
328			break;
329		rc = efx_vpd_verify(sc->enp, sc->vpd_data, sc->vpd_size);
330		if (rc != 0)
331			break;
332		rc = efx_vpd_write(sc->enp, sc->vpd_data, sc->vpd_size);
333		break;
334	default:
335		rc = EOPNOTSUPP;
336		break;
337	}
338
339	return (rc);
340}
341
342static int
343sfxge_private_ioctl(struct sfxge_softc *sc, sfxge_ioc_t *ioc)
344{
345	switch (ioc->op) {
346	case SFXGE_MCDI_IOC:
347		return (sfxge_mcdi_ioctl(sc, ioc));
348	case SFXGE_NVRAM_IOC:
349		return (sfxge_nvram_ioctl(sc, ioc));
350	case SFXGE_VPD_IOC:
351		return (sfxge_vpd_ioctl(sc, ioc));
352	default:
353		return (EOPNOTSUPP);
354	}
355}
356
357
358static int
359sfxge_if_ioctl(struct ifnet *ifp, unsigned long command, caddr_t data)
360{
361	struct sfxge_softc *sc;
362	struct ifreq *ifr;
363	sfxge_ioc_t ioc;
364	int error;
365
366	ifr = (struct ifreq *)data;
367	sc = ifp->if_softc;
368	error = 0;
369
370	switch (command) {
371	case SIOCSIFFLAGS:
372		SFXGE_ADAPTER_LOCK(sc);
373		if (ifp->if_flags & IFF_UP) {
374			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
375				if ((ifp->if_flags ^ sc->if_flags) &
376				    (IFF_PROMISC | IFF_ALLMULTI)) {
377					sfxge_mac_filter_set(sc);
378				}
379			} else
380				sfxge_start(sc);
381		} else
382			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
383				sfxge_stop(sc);
384		sc->if_flags = ifp->if_flags;
385		SFXGE_ADAPTER_UNLOCK(sc);
386		break;
387	case SIOCSIFMTU:
388		if (ifr->ifr_mtu == ifp->if_mtu) {
389			/* Nothing to do */
390			error = 0;
391		} else if (ifr->ifr_mtu > SFXGE_MAX_MTU) {
392			error = EINVAL;
393		} else if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
394			ifp->if_mtu = ifr->ifr_mtu;
395			error = 0;
396		} else {
397			/* Restart required */
398			SFXGE_ADAPTER_LOCK(sc);
399			sfxge_stop(sc);
400			ifp->if_mtu = ifr->ifr_mtu;
401			error = sfxge_start(sc);
402			SFXGE_ADAPTER_UNLOCK(sc);
403			if (error != 0) {
404				ifp->if_flags &= ~IFF_UP;
405				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
406				if_down(ifp);
407			}
408		}
409		break;
410	case SIOCADDMULTI:
411	case SIOCDELMULTI:
412		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
413			sfxge_mac_filter_set(sc);
414		break;
415	case SIOCSIFCAP:
416	{
417		int reqcap = ifr->ifr_reqcap;
418		int capchg_mask;
419
420		SFXGE_ADAPTER_LOCK(sc);
421
422		/* Capabilities to be changed in accordance with request */
423		capchg_mask = ifp->if_capenable ^ reqcap;
424
425		/*
426		 * The networking core already rejects attempts to
427		 * enable capabilities we don't have.  We still have
428		 * to reject attempts to disable capabilities that we
429		 * can't (yet) disable.
430		 */
431		KASSERT((reqcap & ~ifp->if_capabilities) == 0,
432		    ("Unsupported capabilities 0x%x requested 0x%x vs "
433		     "supported 0x%x",
434		     reqcap & ~ifp->if_capabilities,
435		     reqcap , ifp->if_capabilities));
436		if (capchg_mask & SFXGE_CAP_FIXED) {
437			error = EINVAL;
438			SFXGE_ADAPTER_UNLOCK(sc);
439			break;
440		}
441
442		/* Check request before any changes */
443		if ((capchg_mask & IFCAP_TSO4) &&
444		    (reqcap & (IFCAP_TSO4 | IFCAP_TXCSUM)) == IFCAP_TSO4) {
445			error = EAGAIN;
446			SFXGE_ADAPTER_UNLOCK(sc);
447			if_printf(ifp, "enable txcsum before tso4\n");
448			break;
449		}
450		if ((capchg_mask & IFCAP_TSO6) &&
451		    (reqcap & (IFCAP_TSO6 | IFCAP_TXCSUM_IPV6)) == IFCAP_TSO6) {
452			error = EAGAIN;
453			SFXGE_ADAPTER_UNLOCK(sc);
454			if_printf(ifp, "enable txcsum6 before tso6\n");
455			break;
456		}
457
458		if (reqcap & IFCAP_TXCSUM) {
459			ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
460		} else {
461			ifp->if_hwassist &= ~(CSUM_IP | CSUM_TCP | CSUM_UDP);
462			if (reqcap & IFCAP_TSO4) {
463				reqcap &= ~IFCAP_TSO4;
464				if_printf(ifp,
465				    "tso4 disabled due to -txcsum\n");
466			}
467		}
468		if (reqcap & IFCAP_TXCSUM_IPV6) {
469			ifp->if_hwassist |= (CSUM_TCP_IPV6 | CSUM_UDP_IPV6);
470		} else {
471			ifp->if_hwassist &= ~(CSUM_TCP_IPV6 | CSUM_UDP_IPV6);
472			if (reqcap & IFCAP_TSO6) {
473				reqcap &= ~IFCAP_TSO6;
474				if_printf(ifp,
475				    "tso6 disabled due to -txcsum6\n");
476			}
477		}
478
479		/*
480		 * The kernel takes both IFCAP_TSOx and CSUM_TSO into
481		 * account before using TSO. So, we do not touch
482		 * checksum flags when IFCAP_TSOx is modified.
483		 * Note that CSUM_TSO is (CSUM_IP_TSO|CSUM_IP6_TSO),
484		 * but both bits are set in IPv4 and IPv6 mbufs.
485		 */
486
487		ifp->if_capenable = reqcap;
488
489		SFXGE_ADAPTER_UNLOCK(sc);
490		break;
491	}
492	case SIOCSIFMEDIA:
493	case SIOCGIFMEDIA:
494		error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
495		break;
496	case SIOCGPRIVATE_0:
497		error = priv_check(curthread, PRIV_DRIVER);
498		if (error != 0)
499			break;
500		error = copyin(ifr->ifr_data, &ioc, sizeof(ioc));
501		if (error != 0)
502			return (error);
503		error = sfxge_private_ioctl(sc, &ioc);
504		if (error == 0) {
505			error = copyout(&ioc, ifr->ifr_data, sizeof(ioc));
506		}
507		break;
508	default:
509		error = ether_ioctl(ifp, command, data);
510	}
511
512	return (error);
513}
514
515static void
516sfxge_ifnet_fini(struct ifnet *ifp)
517{
518	struct sfxge_softc *sc = ifp->if_softc;
519
520	SFXGE_ADAPTER_LOCK(sc);
521	sfxge_stop(sc);
522	SFXGE_ADAPTER_UNLOCK(sc);
523
524	ifmedia_removeall(&sc->media);
525	ether_ifdetach(ifp);
526	if_free(ifp);
527}
528
529static int
530sfxge_ifnet_init(struct ifnet *ifp, struct sfxge_softc *sc)
531{
532	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sc->enp);
533	device_t dev;
534	int rc;
535
536	dev = sc->dev;
537	sc->ifnet = ifp;
538
539	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
540	ifp->if_init = sfxge_if_init;
541	ifp->if_softc = sc;
542	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
543	ifp->if_ioctl = sfxge_if_ioctl;
544
545	ifp->if_capabilities = SFXGE_CAP;
546	ifp->if_capenable = SFXGE_CAP_ENABLE;
547	ifp->if_hw_tsomax = SFXGE_TSO_MAX_SIZE;
548	ifp->if_hw_tsomaxsegcount = SFXGE_TX_MAPPING_MAX_SEG;
549	ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
550
551#ifdef SFXGE_LRO
552	ifp->if_capabilities |= IFCAP_LRO;
553	ifp->if_capenable |= IFCAP_LRO;
554#endif
555
556	if (encp->enc_hw_tx_insert_vlan_enabled) {
557		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
558		ifp->if_capenable |= IFCAP_VLAN_HWTAGGING;
559	}
560	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
561			   CSUM_TCP_IPV6 | CSUM_UDP_IPV6;
562
563	ether_ifattach(ifp, encp->enc_mac_addr);
564
565	ifp->if_transmit = sfxge_if_transmit;
566	ifp->if_qflush = sfxge_if_qflush;
567
568	ifp->if_get_counter = sfxge_get_counter;
569
570	DBGPRINT(sc->dev, "ifmedia_init");
571	if ((rc = sfxge_port_ifmedia_init(sc)) != 0)
572		goto fail;
573
574	return (0);
575
576fail:
577	ether_ifdetach(sc->ifnet);
578	return (rc);
579}
580
581void
582sfxge_sram_buf_tbl_alloc(struct sfxge_softc *sc, size_t n, uint32_t *idp)
583{
584	KASSERT(sc->buffer_table_next + n <=
585		efx_nic_cfg_get(sc->enp)->enc_buftbl_limit,
586		("buffer table full"));
587
588	*idp = sc->buffer_table_next;
589	sc->buffer_table_next += n;
590}
591
592static int
593sfxge_bar_init(struct sfxge_softc *sc)
594{
595	efsys_bar_t *esbp = &sc->bar;
596
597	esbp->esb_rid = PCIR_BAR(EFX_MEM_BAR);
598	if ((esbp->esb_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
599	    &esbp->esb_rid, RF_ACTIVE)) == NULL) {
600		device_printf(sc->dev, "Cannot allocate BAR region %d\n",
601		    EFX_MEM_BAR);
602		return (ENXIO);
603	}
604	esbp->esb_tag = rman_get_bustag(esbp->esb_res);
605	esbp->esb_handle = rman_get_bushandle(esbp->esb_res);
606
607	SFXGE_BAR_LOCK_INIT(esbp, device_get_nameunit(sc->dev));
608
609	return (0);
610}
611
612static void
613sfxge_bar_fini(struct sfxge_softc *sc)
614{
615	efsys_bar_t *esbp = &sc->bar;
616
617	bus_release_resource(sc->dev, SYS_RES_MEMORY, esbp->esb_rid,
618	    esbp->esb_res);
619	SFXGE_BAR_LOCK_DESTROY(esbp);
620}
621
622static int
623sfxge_create(struct sfxge_softc *sc)
624{
625	device_t dev;
626	efx_nic_t *enp;
627	int error;
628	char rss_param_name[sizeof(SFXGE_PARAM(%d.max_rss_channels))];
629#if EFSYS_OPT_MCDI_LOGGING
630	char mcdi_log_param_name[sizeof(SFXGE_PARAM(%d.mcdi_logging))];
631#endif
632
633	dev = sc->dev;
634
635	SFXGE_ADAPTER_LOCK_INIT(sc, device_get_nameunit(sc->dev));
636
637	sc->max_rss_channels = 0;
638	snprintf(rss_param_name, sizeof(rss_param_name),
639		 SFXGE_PARAM(%d.max_rss_channels),
640		 (int)device_get_unit(dev));
641	TUNABLE_INT_FETCH(rss_param_name, &sc->max_rss_channels);
642#if EFSYS_OPT_MCDI_LOGGING
643	sc->mcdi_logging = sfxge_mcdi_logging;
644	snprintf(mcdi_log_param_name, sizeof(mcdi_log_param_name),
645		 SFXGE_PARAM(%d.mcdi_logging),
646		 (int)device_get_unit(dev));
647	TUNABLE_INT_FETCH(mcdi_log_param_name, &sc->mcdi_logging);
648#endif
649
650	sc->stats_node = SYSCTL_ADD_NODE(
651		device_get_sysctl_ctx(dev),
652		SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
653		OID_AUTO, "stats", CTLFLAG_RD, NULL, "Statistics");
654	if (sc->stats_node == NULL) {
655		error = ENOMEM;
656		goto fail;
657	}
658
659	TASK_INIT(&sc->task_reset, 0, sfxge_reset, sc);
660
661	(void) pci_enable_busmaster(dev);
662
663	/* Initialize DMA mappings. */
664	DBGPRINT(sc->dev, "dma_init...");
665	if ((error = sfxge_dma_init(sc)) != 0)
666		goto fail;
667
668	/* Map the device registers. */
669	DBGPRINT(sc->dev, "bar_init...");
670	if ((error = sfxge_bar_init(sc)) != 0)
671		goto fail;
672
673	error = efx_family(pci_get_vendor(dev), pci_get_device(dev),
674	    &sc->family);
675	KASSERT(error == 0, ("Family should be filtered by sfxge_probe()"));
676
677	DBGPRINT(sc->dev, "nic_create...");
678
679	/* Create the common code nic object. */
680	SFXGE_EFSYS_LOCK_INIT(&sc->enp_lock,
681			      device_get_nameunit(sc->dev), "nic");
682	if ((error = efx_nic_create(sc->family, (efsys_identifier_t *)sc,
683	    &sc->bar, &sc->enp_lock, &enp)) != 0)
684		goto fail3;
685	sc->enp = enp;
686
687	if (!ISP2(sfxge_rx_ring_entries) ||
688	    (sfxge_rx_ring_entries < EFX_RXQ_MINNDESCS) ||
689	    (sfxge_rx_ring_entries > EFX_RXQ_MAXNDESCS)) {
690		log(LOG_ERR, "%s=%d must be power of 2 from %u to %u",
691		    SFXGE_PARAM_RX_RING, sfxge_rx_ring_entries,
692		    EFX_RXQ_MINNDESCS, EFX_RXQ_MAXNDESCS);
693		error = EINVAL;
694		goto fail_rx_ring_entries;
695	}
696	sc->rxq_entries = sfxge_rx_ring_entries;
697
698	if (!ISP2(sfxge_tx_ring_entries) ||
699	    (sfxge_tx_ring_entries < EFX_TXQ_MINNDESCS) ||
700	    (sfxge_tx_ring_entries > EFX_TXQ_MAXNDESCS(efx_nic_cfg_get(enp)))) {
701		log(LOG_ERR, "%s=%d must be power of 2 from %u to %u",
702		    SFXGE_PARAM_TX_RING, sfxge_tx_ring_entries,
703		    EFX_TXQ_MINNDESCS, EFX_TXQ_MAXNDESCS(efx_nic_cfg_get(enp)));
704		error = EINVAL;
705		goto fail_tx_ring_entries;
706	}
707	sc->txq_entries = sfxge_tx_ring_entries;
708
709	/* Initialize MCDI to talk to the microcontroller. */
710	DBGPRINT(sc->dev, "mcdi_init...");
711	if ((error = sfxge_mcdi_init(sc)) != 0)
712		goto fail4;
713
714	/* Probe the NIC and build the configuration data area. */
715	DBGPRINT(sc->dev, "nic_probe...");
716	if ((error = efx_nic_probe(enp)) != 0)
717		goto fail5;
718
719	SYSCTL_ADD_STRING(device_get_sysctl_ctx(dev),
720			  SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
721			  OID_AUTO, "version", CTLFLAG_RD,
722			  SFXGE_VERSION_STRING, 0,
723			  "Driver version");
724
725	SYSCTL_ADD_UINT(device_get_sysctl_ctx(dev),
726			SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
727			OID_AUTO, "phy_type", CTLFLAG_RD,
728			NULL, efx_nic_cfg_get(enp)->enc_phy_type,
729			"PHY type");
730
731	/* Initialize the NVRAM. */
732	DBGPRINT(sc->dev, "nvram_init...");
733	if ((error = efx_nvram_init(enp)) != 0)
734		goto fail6;
735
736	/* Initialize the VPD. */
737	DBGPRINT(sc->dev, "vpd_init...");
738	if ((error = efx_vpd_init(enp)) != 0)
739		goto fail7;
740
741	efx_mcdi_new_epoch(enp);
742
743	/* Reset the NIC. */
744	DBGPRINT(sc->dev, "nic_reset...");
745	if ((error = efx_nic_reset(enp)) != 0)
746		goto fail8;
747
748	/* Initialize buffer table allocation. */
749	sc->buffer_table_next = 0;
750
751	/*
752	 * Guarantee minimum and estimate maximum number of event queues
753	 * to take it into account when MSI-X interrupts are allocated.
754	 * It initializes NIC and keeps it initialized on success.
755	 */
756	if ((error = sfxge_estimate_rsrc_limits(sc)) != 0)
757		goto fail8;
758
759	/* Set up interrupts. */
760	DBGPRINT(sc->dev, "intr_init...");
761	if ((error = sfxge_intr_init(sc)) != 0)
762		goto fail9;
763
764	/* Initialize event processing state. */
765	DBGPRINT(sc->dev, "ev_init...");
766	if ((error = sfxge_ev_init(sc)) != 0)
767		goto fail11;
768
769	/* Initialize port state. */
770	DBGPRINT(sc->dev, "port_init...");
771	if ((error = sfxge_port_init(sc)) != 0)
772		goto fail12;
773
774	/* Initialize receive state. */
775	DBGPRINT(sc->dev, "rx_init...");
776	if ((error = sfxge_rx_init(sc)) != 0)
777		goto fail13;
778
779	/* Initialize transmit state. */
780	DBGPRINT(sc->dev, "tx_init...");
781	if ((error = sfxge_tx_init(sc)) != 0)
782		goto fail14;
783
784	sc->init_state = SFXGE_INITIALIZED;
785
786	DBGPRINT(sc->dev, "success");
787	return (0);
788
789fail14:
790	sfxge_rx_fini(sc);
791
792fail13:
793	sfxge_port_fini(sc);
794
795fail12:
796	sfxge_ev_fini(sc);
797
798fail11:
799	sfxge_intr_fini(sc);
800
801fail9:
802	efx_nic_fini(sc->enp);
803
804fail8:
805	efx_vpd_fini(enp);
806
807fail7:
808	efx_nvram_fini(enp);
809
810fail6:
811	efx_nic_unprobe(enp);
812
813fail5:
814	sfxge_mcdi_fini(sc);
815
816fail4:
817fail_tx_ring_entries:
818fail_rx_ring_entries:
819	sc->enp = NULL;
820	efx_nic_destroy(enp);
821	SFXGE_EFSYS_LOCK_DESTROY(&sc->enp_lock);
822
823fail3:
824	sfxge_bar_fini(sc);
825	(void) pci_disable_busmaster(sc->dev);
826
827fail:
828	DBGPRINT(sc->dev, "failed %d", error);
829	sc->dev = NULL;
830	SFXGE_ADAPTER_LOCK_DESTROY(sc);
831	return (error);
832}
833
834static void
835sfxge_destroy(struct sfxge_softc *sc)
836{
837	efx_nic_t *enp;
838
839	/* Clean up transmit state. */
840	sfxge_tx_fini(sc);
841
842	/* Clean up receive state. */
843	sfxge_rx_fini(sc);
844
845	/* Clean up port state. */
846	sfxge_port_fini(sc);
847
848	/* Clean up event processing state. */
849	sfxge_ev_fini(sc);
850
851	/* Clean up interrupts. */
852	sfxge_intr_fini(sc);
853
854	/* Tear down common code subsystems. */
855	efx_nic_reset(sc->enp);
856	efx_vpd_fini(sc->enp);
857	efx_nvram_fini(sc->enp);
858	efx_nic_unprobe(sc->enp);
859
860	/* Tear down MCDI. */
861	sfxge_mcdi_fini(sc);
862
863	/* Destroy common code context. */
864	enp = sc->enp;
865	sc->enp = NULL;
866	efx_nic_destroy(enp);
867
868	/* Free DMA memory. */
869	sfxge_dma_fini(sc);
870
871	/* Free mapped BARs. */
872	sfxge_bar_fini(sc);
873
874	(void) pci_disable_busmaster(sc->dev);
875
876	taskqueue_drain(taskqueue_thread, &sc->task_reset);
877
878	/* Destroy the softc lock. */
879	SFXGE_ADAPTER_LOCK_DESTROY(sc);
880}
881
882static int
883sfxge_vpd_handler(SYSCTL_HANDLER_ARGS)
884{
885	struct sfxge_softc *sc = arg1;
886	efx_vpd_value_t value;
887	int rc;
888
889	value.evv_tag = arg2 >> 16;
890	value.evv_keyword = arg2 & 0xffff;
891	if ((rc = efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value))
892	    != 0)
893		return (rc);
894
895	return (SYSCTL_OUT(req, value.evv_value, value.evv_length));
896}
897
898static void
899sfxge_vpd_try_add(struct sfxge_softc *sc, struct sysctl_oid_list *list,
900		  efx_vpd_tag_t tag, const char *keyword)
901{
902	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
903	efx_vpd_value_t value;
904
905	/* Check whether VPD tag/keyword is present */
906	value.evv_tag = tag;
907	value.evv_keyword = EFX_VPD_KEYWORD(keyword[0], keyword[1]);
908	if (efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value) != 0)
909		return;
910
911	SYSCTL_ADD_PROC(
912		ctx, list, OID_AUTO, keyword, CTLTYPE_STRING|CTLFLAG_RD,
913		sc, tag << 16 | EFX_VPD_KEYWORD(keyword[0], keyword[1]),
914		sfxge_vpd_handler, "A", "");
915}
916
917static int
918sfxge_vpd_init(struct sfxge_softc *sc)
919{
920	struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(sc->dev);
921	struct sysctl_oid *vpd_node;
922	struct sysctl_oid_list *vpd_list;
923	char keyword[3];
924	efx_vpd_value_t value;
925	int rc;
926
927	if ((rc = efx_vpd_size(sc->enp, &sc->vpd_size)) != 0) {
928		/*
929		 * Unpriviledged functions deny VPD access.
930		 * Simply skip VPD in this case.
931		 */
932		if (rc == EACCES)
933			goto done;
934		goto fail;
935	}
936	sc->vpd_data = malloc(sc->vpd_size, M_SFXGE, M_WAITOK);
937	if ((rc = efx_vpd_read(sc->enp, sc->vpd_data, sc->vpd_size)) != 0)
938		goto fail2;
939
940	/* Copy ID (product name) into device description, and log it. */
941	value.evv_tag = EFX_VPD_ID;
942	if (efx_vpd_get(sc->enp, sc->vpd_data, sc->vpd_size, &value) == 0) {
943		value.evv_value[value.evv_length] = 0;
944		device_set_desc_copy(sc->dev, value.evv_value);
945		device_printf(sc->dev, "%s\n", value.evv_value);
946	}
947
948	vpd_node = SYSCTL_ADD_NODE(
949		ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)),
950		OID_AUTO, "vpd", CTLFLAG_RD, NULL, "Vital Product Data");
951	vpd_list = SYSCTL_CHILDREN(vpd_node);
952
953	/* Add sysctls for all expected and any vendor-defined keywords. */
954	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "PN");
955	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "EC");
956	sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, "SN");
957	keyword[0] = 'V';
958	keyword[2] = 0;
959	for (keyword[1] = '0'; keyword[1] <= '9'; keyword[1]++)
960		sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, keyword);
961	for (keyword[1] = 'A'; keyword[1] <= 'Z'; keyword[1]++)
962		sfxge_vpd_try_add(sc, vpd_list, EFX_VPD_RO, keyword);
963
964done:
965	return (0);
966
967fail2:
968	free(sc->vpd_data, M_SFXGE);
969fail:
970	return (rc);
971}
972
973static void
974sfxge_vpd_fini(struct sfxge_softc *sc)
975{
976	free(sc->vpd_data, M_SFXGE);
977}
978
979static void
980sfxge_reset(void *arg, int npending)
981{
982	struct sfxge_softc *sc;
983	int rc;
984	unsigned attempt;
985
986	(void)npending;
987
988	sc = (struct sfxge_softc *)arg;
989
990	SFXGE_ADAPTER_LOCK(sc);
991
992	if (sc->init_state != SFXGE_STARTED)
993		goto done;
994
995	sfxge_stop(sc);
996	efx_nic_reset(sc->enp);
997	for (attempt = 0; attempt < 3; ++attempt) {
998		if ((rc = sfxge_start(sc)) == 0)
999			goto done;
1000
1001		device_printf(sc->dev, "start on reset failed (%d)\n", rc);
1002		DELAY(100000);
1003	}
1004
1005	device_printf(sc->dev, "reset failed; interface is now stopped\n");
1006
1007done:
1008	SFXGE_ADAPTER_UNLOCK(sc);
1009}
1010
1011void
1012sfxge_schedule_reset(struct sfxge_softc *sc)
1013{
1014	taskqueue_enqueue(taskqueue_thread, &sc->task_reset);
1015}
1016
1017static int
1018sfxge_attach(device_t dev)
1019{
1020	struct sfxge_softc *sc;
1021	struct ifnet *ifp;
1022	int error;
1023
1024	sc = device_get_softc(dev);
1025	sc->dev = dev;
1026
1027	/* Allocate ifnet. */
1028	ifp = if_alloc(IFT_ETHER);
1029	if (ifp == NULL) {
1030		device_printf(dev, "Couldn't allocate ifnet\n");
1031		error = ENOMEM;
1032		goto fail;
1033	}
1034	sc->ifnet = ifp;
1035
1036	/* Initialize hardware. */
1037	DBGPRINT(sc->dev, "create nic");
1038	if ((error = sfxge_create(sc)) != 0)
1039		goto fail2;
1040
1041	/* Create the ifnet for the port. */
1042	DBGPRINT(sc->dev, "init ifnet");
1043	if ((error = sfxge_ifnet_init(ifp, sc)) != 0)
1044		goto fail3;
1045
1046	DBGPRINT(sc->dev, "init vpd");
1047	if ((error = sfxge_vpd_init(sc)) != 0)
1048		goto fail4;
1049
1050	/*
1051	 * NIC is initialized inside sfxge_create() and kept inialized
1052	 * to be able to initialize port to discover media types in
1053	 * sfxge_ifnet_init().
1054	 */
1055	efx_nic_fini(sc->enp);
1056
1057	sc->init_state = SFXGE_REGISTERED;
1058
1059	DBGPRINT(sc->dev, "success");
1060	return (0);
1061
1062fail4:
1063	sfxge_ifnet_fini(ifp);
1064fail3:
1065	efx_nic_fini(sc->enp);
1066	sfxge_destroy(sc);
1067
1068fail2:
1069	if_free(sc->ifnet);
1070
1071fail:
1072	DBGPRINT(sc->dev, "failed %d", error);
1073	return (error);
1074}
1075
1076static int
1077sfxge_detach(device_t dev)
1078{
1079	struct sfxge_softc *sc;
1080
1081	sc = device_get_softc(dev);
1082
1083	sfxge_vpd_fini(sc);
1084
1085	/* Destroy the ifnet. */
1086	sfxge_ifnet_fini(sc->ifnet);
1087
1088	/* Tear down hardware. */
1089	sfxge_destroy(sc);
1090
1091	return (0);
1092}
1093
1094static int
1095sfxge_probe(device_t dev)
1096{
1097	uint16_t pci_vendor_id;
1098	uint16_t pci_device_id;
1099	efx_family_t family;
1100	int rc;
1101
1102	pci_vendor_id = pci_get_vendor(dev);
1103	pci_device_id = pci_get_device(dev);
1104
1105	DBGPRINT(dev, "PCI ID %04x:%04x", pci_vendor_id, pci_device_id);
1106	rc = efx_family(pci_vendor_id, pci_device_id, &family);
1107	if (rc != 0) {
1108		DBGPRINT(dev, "efx_family fail %d", rc);
1109		return (ENXIO);
1110	}
1111
1112	if (family == EFX_FAMILY_SIENA) {
1113		device_set_desc(dev, "Solarflare SFC9000 family");
1114		return (0);
1115	}
1116
1117	if (family == EFX_FAMILY_HUNTINGTON) {
1118		device_set_desc(dev, "Solarflare SFC9100 family");
1119		return (0);
1120	}
1121
1122	DBGPRINT(dev, "impossible controller family %d", family);
1123	return (ENXIO);
1124}
1125
1126static device_method_t sfxge_methods[] = {
1127	DEVMETHOD(device_probe,		sfxge_probe),
1128	DEVMETHOD(device_attach,	sfxge_attach),
1129	DEVMETHOD(device_detach,	sfxge_detach),
1130
1131	DEVMETHOD_END
1132};
1133
1134static devclass_t sfxge_devclass;
1135
1136static driver_t sfxge_driver = {
1137	"sfxge",
1138	sfxge_methods,
1139	sizeof(struct sfxge_softc)
1140};
1141
1142DRIVER_MODULE(sfxge, pci, sfxge_driver, sfxge_devclass, 0, 0);
1143