xref: /freebsd/sys/dev/cxgb/cxgb_main.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /**************************************************************************
2 SPDX-License-Identifier: BSD-2-Clause
3 
4 Copyright (c) 2007-2009, Chelsio Inc.
5 All rights reserved.
6 
7 Redistribution and use in source and binary forms, with or without
8 modification, are permitted provided that the following conditions are met:
9 
10  1. Redistributions of source code must retain the above copyright notice,
11     this list of conditions and the following disclaimer.
12 
13  2. Neither the name of the Chelsio Corporation nor the names of its
14     contributors may be used to endorse or promote products derived from
15     this software without specific prior written permission.
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, THE
19 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 POSSIBILITY OF SUCH DAMAGE.
28 
29 ***************************************************************************/
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_inet.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/bus.h>
40 #include <sys/module.h>
41 #include <sys/pciio.h>
42 #include <sys/conf.h>
43 #include <machine/bus.h>
44 #include <machine/resource.h>
45 #include <sys/ktr.h>
46 #include <sys/rman.h>
47 #include <sys/ioccom.h>
48 #include <sys/mbuf.h>
49 #include <sys/linker.h>
50 #include <sys/firmware.h>
51 #include <sys/socket.h>
52 #include <sys/sockio.h>
53 #include <sys/smp.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/taskqueue.h>
58 #include <sys/proc.h>
59 
60 #include <net/bpf.h>
61 #include <net/debugnet.h>
62 #include <net/ethernet.h>
63 #include <net/if.h>
64 #include <net/if_var.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_media.h>
68 #include <net/if_types.h>
69 #include <net/if_vlan_var.h>
70 
71 #include <netinet/in_systm.h>
72 #include <netinet/in.h>
73 #include <netinet/if_ether.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip.h>
76 #include <netinet/tcp.h>
77 #include <netinet/udp.h>
78 
79 #include <dev/pci/pcireg.h>
80 #include <dev/pci/pcivar.h>
81 #include <dev/pci/pci_private.h>
82 
83 #include <cxgb_include.h>
84 
85 #ifdef PRIV_SUPPORTED
86 #include <sys/priv.h>
87 #endif
88 
89 static int cxgb_setup_interrupts(adapter_t *);
90 static void cxgb_teardown_interrupts(adapter_t *);
91 static void cxgb_init(void *);
92 static int cxgb_init_locked(struct port_info *);
93 static int cxgb_uninit_locked(struct port_info *);
94 static int cxgb_uninit_synchronized(struct port_info *);
95 static int cxgb_ioctl(if_t, unsigned long, caddr_t);
96 static int cxgb_media_change(if_t);
97 static int cxgb_ifm_type(int);
98 static void cxgb_build_medialist(struct port_info *);
99 static void cxgb_media_status(if_t, struct ifmediareq *);
100 static uint64_t cxgb_get_counter(if_t, ift_counter);
101 static int setup_sge_qsets(adapter_t *);
102 static void cxgb_async_intr(void *);
103 static void cxgb_tick_handler(void *, int);
104 static void cxgb_tick(void *);
105 static void link_check_callout(void *);
106 static void check_link_status(void *, int);
107 static void setup_rss(adapter_t *sc);
108 static int alloc_filters(struct adapter *);
109 static int setup_hw_filters(struct adapter *);
110 static int set_filter(struct adapter *, int, const struct filter_info *);
111 static inline void mk_set_tcb_field(struct cpl_set_tcb_field *, unsigned int,
112     unsigned int, u64, u64);
113 static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *, unsigned int,
114     unsigned int, u64, u64);
115 #ifdef TCP_OFFLOAD
116 static int cpl_not_handled(struct sge_qset *, struct rsp_desc *, struct mbuf *);
117 #endif
118 
119 /* Attachment glue for the PCI controller end of the device.  Each port of
120  * the device is attached separately, as defined later.
121  */
122 static int cxgb_controller_probe(device_t);
123 static int cxgb_controller_attach(device_t);
124 static int cxgb_controller_detach(device_t);
125 static void cxgb_free(struct adapter *);
126 static __inline void reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
127     unsigned int end);
128 static void cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf);
129 static int cxgb_get_regs_len(void);
130 static void touch_bars(device_t dev);
131 static void cxgb_update_mac_settings(struct port_info *p);
132 #ifdef TCP_OFFLOAD
133 static int toe_capability(struct port_info *, int);
134 #endif
135 
136 /* Table for probing the cards.  The desc field isn't actually used */
137 struct cxgb_ident {
138 	uint16_t	vendor;
139 	uint16_t	device;
140 	int		index;
141 	char		*desc;
142 } cxgb_identifiers[] = {
143 	{PCI_VENDOR_ID_CHELSIO, 0x0020, 0, "PE9000"},
144 	{PCI_VENDOR_ID_CHELSIO, 0x0021, 1, "T302E"},
145 	{PCI_VENDOR_ID_CHELSIO, 0x0022, 2, "T310E"},
146 	{PCI_VENDOR_ID_CHELSIO, 0x0023, 3, "T320X"},
147 	{PCI_VENDOR_ID_CHELSIO, 0x0024, 1, "T302X"},
148 	{PCI_VENDOR_ID_CHELSIO, 0x0025, 3, "T320E"},
149 	{PCI_VENDOR_ID_CHELSIO, 0x0026, 2, "T310X"},
150 	{PCI_VENDOR_ID_CHELSIO, 0x0030, 2, "T3B10"},
151 	{PCI_VENDOR_ID_CHELSIO, 0x0031, 3, "T3B20"},
152 	{PCI_VENDOR_ID_CHELSIO, 0x0032, 1, "T3B02"},
153 	{PCI_VENDOR_ID_CHELSIO, 0x0033, 4, "T3B04"},
154 	{PCI_VENDOR_ID_CHELSIO, 0x0035, 6, "T3C10"},
155 	{PCI_VENDOR_ID_CHELSIO, 0x0036, 3, "S320E-CR"},
156 	{PCI_VENDOR_ID_CHELSIO, 0x0037, 7, "N320E-G2"},
157 	{0, 0, 0, NULL}
158 };
159 
160 static device_method_t cxgb_controller_methods[] = {
161 	DEVMETHOD(device_probe,		cxgb_controller_probe),
162 	DEVMETHOD(device_attach,	cxgb_controller_attach),
163 	DEVMETHOD(device_detach,	cxgb_controller_detach),
164 
165 	DEVMETHOD_END
166 };
167 
168 static driver_t cxgb_controller_driver = {
169 	"cxgbc",
170 	cxgb_controller_methods,
171 	sizeof(struct adapter)
172 };
173 
174 static int cxgbc_mod_event(module_t, int, void *);
175 
176 DRIVER_MODULE(cxgbc, pci, cxgb_controller_driver, cxgbc_mod_event, NULL);
177 MODULE_PNP_INFO("U16:vendor;U16:device", pci, cxgbc, cxgb_identifiers,
178     nitems(cxgb_identifiers) - 1);
179 MODULE_VERSION(cxgbc, 1);
180 MODULE_DEPEND(cxgbc, firmware, 1, 1, 1);
181 
182 /*
183  * Attachment glue for the ports.  Attachment is done directly to the
184  * controller device.
185  */
186 static int cxgb_port_probe(device_t);
187 static int cxgb_port_attach(device_t);
188 static int cxgb_port_detach(device_t);
189 
190 static device_method_t cxgb_port_methods[] = {
191 	DEVMETHOD(device_probe,		cxgb_port_probe),
192 	DEVMETHOD(device_attach,	cxgb_port_attach),
193 	DEVMETHOD(device_detach,	cxgb_port_detach),
194 	{ 0, 0 }
195 };
196 
197 static driver_t cxgb_port_driver = {
198 	"cxgb",
199 	cxgb_port_methods,
200 	0
201 };
202 
203 static d_ioctl_t cxgb_extension_ioctl;
204 static d_open_t cxgb_extension_open;
205 static d_close_t cxgb_extension_close;
206 
207 static struct cdevsw cxgb_cdevsw = {
208        .d_version =    D_VERSION,
209        .d_flags =      0,
210        .d_open =       cxgb_extension_open,
211        .d_close =      cxgb_extension_close,
212        .d_ioctl =      cxgb_extension_ioctl,
213        .d_name =       "cxgb",
214 };
215 
216 DRIVER_MODULE(cxgb, cxgbc, cxgb_port_driver, 0, 0);
217 MODULE_VERSION(cxgb, 1);
218 
219 DEBUGNET_DEFINE(cxgb);
220 
221 static struct mtx t3_list_lock;
222 static SLIST_HEAD(, adapter) t3_list;
223 #ifdef TCP_OFFLOAD
224 static struct mtx t3_uld_list_lock;
225 static SLIST_HEAD(, uld_info) t3_uld_list;
226 #endif
227 
228 /*
229  * The driver uses the best interrupt scheme available on a platform in the
230  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
231  * of these schemes the driver may consider as follows:
232  *
233  * msi = 2: choose from among all three options
234  * msi = 1 : only consider MSI and pin interrupts
235  * msi = 0: force pin interrupts
236  */
237 static int msi_allowed = 2;
238 
239 SYSCTL_NODE(_hw, OID_AUTO, cxgb, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
240     "CXGB driver parameters");
241 SYSCTL_INT(_hw_cxgb, OID_AUTO, msi_allowed, CTLFLAG_RDTUN, &msi_allowed, 0,
242     "MSI-X, MSI, INTx selector");
243 
244 /*
245  * The driver uses an auto-queue algorithm by default.
246  * To disable it and force a single queue-set per port, use multiq = 0
247  */
248 static int multiq = 1;
249 SYSCTL_INT(_hw_cxgb, OID_AUTO, multiq, CTLFLAG_RDTUN, &multiq, 0,
250     "use min(ncpus/ports, 8) queue-sets per port");
251 
252 /*
253  * By default the driver will not update the firmware unless
254  * it was compiled against a newer version
255  *
256  */
257 static int force_fw_update = 0;
258 SYSCTL_INT(_hw_cxgb, OID_AUTO, force_fw_update, CTLFLAG_RDTUN, &force_fw_update, 0,
259     "update firmware even if up to date");
260 
261 int cxgb_use_16k_clusters = -1;
262 SYSCTL_INT(_hw_cxgb, OID_AUTO, use_16k_clusters, CTLFLAG_RDTUN,
263     &cxgb_use_16k_clusters, 0, "use 16kB clusters for the jumbo queue ");
264 
265 static int nfilters = -1;
266 SYSCTL_INT(_hw_cxgb, OID_AUTO, nfilters, CTLFLAG_RDTUN,
267     &nfilters, 0, "max number of entries in the filter table");
268 
269 enum {
270 	MAX_TXQ_ENTRIES      = 16384,
271 	MAX_CTRL_TXQ_ENTRIES = 1024,
272 	MAX_RSPQ_ENTRIES     = 16384,
273 	MAX_RX_BUFFERS       = 16384,
274 	MAX_RX_JUMBO_BUFFERS = 16384,
275 	MIN_TXQ_ENTRIES      = 4,
276 	MIN_CTRL_TXQ_ENTRIES = 4,
277 	MIN_RSPQ_ENTRIES     = 32,
278 	MIN_FL_ENTRIES       = 32,
279 	MIN_FL_JUMBO_ENTRIES = 32
280 };
281 
282 struct filter_info {
283 	u32 sip;
284 	u32 sip_mask;
285 	u32 dip;
286 	u16 sport;
287 	u16 dport;
288 	u32 vlan:12;
289 	u32 vlan_prio:3;
290 	u32 mac_hit:1;
291 	u32 mac_idx:4;
292 	u32 mac_vld:1;
293 	u32 pkt_type:2;
294 	u32 report_filter_id:1;
295 	u32 pass:1;
296 	u32 rss:1;
297 	u32 qset:3;
298 	u32 locked:1;
299 	u32 valid:1;
300 };
301 
302 enum { FILTER_NO_VLAN_PRI = 7 };
303 
304 #define EEPROM_MAGIC 0x38E2F10C
305 
306 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
307 
308 
309 static int set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset);
310 
311 
312 static __inline char
313 t3rev2char(struct adapter *adapter)
314 {
315 	char rev = 'z';
316 
317 	switch(adapter->params.rev) {
318 	case T3_REV_A:
319 		rev = 'a';
320 		break;
321 	case T3_REV_B:
322 	case T3_REV_B2:
323 		rev = 'b';
324 		break;
325 	case T3_REV_C:
326 		rev = 'c';
327 		break;
328 	}
329 	return rev;
330 }
331 
332 static struct cxgb_ident *
333 cxgb_get_ident(device_t dev)
334 {
335 	struct cxgb_ident *id;
336 
337 	for (id = cxgb_identifiers; id->desc != NULL; id++) {
338 		if ((id->vendor == pci_get_vendor(dev)) &&
339 		    (id->device == pci_get_device(dev))) {
340 			return (id);
341 		}
342 	}
343 	return (NULL);
344 }
345 
346 static const struct adapter_info *
347 cxgb_get_adapter_info(device_t dev)
348 {
349 	struct cxgb_ident *id;
350 	const struct adapter_info *ai;
351 
352 	id = cxgb_get_ident(dev);
353 	if (id == NULL)
354 		return (NULL);
355 
356 	ai = t3_get_adapter_info(id->index);
357 
358 	return (ai);
359 }
360 
361 static int
362 cxgb_controller_probe(device_t dev)
363 {
364 	const struct adapter_info *ai;
365 	char *ports, buf[80];
366 	int nports;
367 
368 	ai = cxgb_get_adapter_info(dev);
369 	if (ai == NULL)
370 		return (ENXIO);
371 
372 	nports = ai->nports0 + ai->nports1;
373 	if (nports == 1)
374 		ports = "port";
375 	else
376 		ports = "ports";
377 
378 	snprintf(buf, sizeof(buf), "%s, %d %s", ai->desc, nports, ports);
379 	device_set_desc_copy(dev, buf);
380 	return (BUS_PROBE_DEFAULT);
381 }
382 
383 #define FW_FNAME "cxgb_t3fw"
384 #define TPEEPROM_NAME "cxgb_t3%c_tp_eeprom"
385 #define TPSRAM_NAME "cxgb_t3%c_protocol_sram"
386 
387 static int
388 upgrade_fw(adapter_t *sc)
389 {
390 	const struct firmware *fw;
391 	int status;
392 	u32 vers;
393 
394 	if ((fw = firmware_get(FW_FNAME)) == NULL)  {
395 		device_printf(sc->dev, "Could not find firmware image %s\n", FW_FNAME);
396 		return (ENOENT);
397 	} else
398 		device_printf(sc->dev, "installing firmware on card\n");
399 	status = t3_load_fw(sc, (const uint8_t *)fw->data, fw->datasize);
400 
401 	if (status != 0) {
402 		device_printf(sc->dev, "failed to install firmware: %d\n",
403 		    status);
404 	} else {
405 		t3_get_fw_version(sc, &vers);
406 		snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
407 		    G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
408 		    G_FW_VERSION_MICRO(vers));
409 	}
410 
411 	firmware_put(fw, FIRMWARE_UNLOAD);
412 
413 	return (status);
414 }
415 
416 /*
417  * The cxgb_controller_attach function is responsible for the initial
418  * bringup of the device.  Its responsibilities include:
419  *
420  *  1. Determine if the device supports MSI or MSI-X.
421  *  2. Allocate bus resources so that we can access the Base Address Register
422  *  3. Create and initialize mutexes for the controller and its control
423  *     logic such as SGE and MDIO.
424  *  4. Call hardware specific setup routine for the adapter as a whole.
425  *  5. Allocate the BAR for doing MSI-X.
426  *  6. Setup the line interrupt iff MSI-X is not supported.
427  *  7. Create the driver's taskq.
428  *  8. Start one task queue service thread.
429  *  9. Check if the firmware and SRAM are up-to-date.  They will be
430  *     auto-updated later (before FULL_INIT_DONE), if required.
431  * 10. Create a child device for each MAC (port)
432  * 11. Initialize T3 private state.
433  * 12. Trigger the LED
434  * 13. Setup offload iff supported.
435  * 14. Reset/restart the tick callout.
436  * 15. Attach sysctls
437  *
438  * NOTE: Any modification or deviation from this list MUST be reflected in
439  * the above comment.  Failure to do so will result in problems on various
440  * error conditions including link flapping.
441  */
442 static int
443 cxgb_controller_attach(device_t dev)
444 {
445 	device_t child;
446 	const struct adapter_info *ai;
447 	struct adapter *sc;
448 	int i, error = 0;
449 	uint32_t vers;
450 	int port_qsets = 1;
451 	int msi_needed, reg;
452 	char buf[80];
453 
454 	sc = device_get_softc(dev);
455 	sc->dev = dev;
456 	sc->msi_count = 0;
457 	ai = cxgb_get_adapter_info(dev);
458 
459 	snprintf(sc->lockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb controller lock %d",
460 	    device_get_unit(dev));
461 	ADAPTER_LOCK_INIT(sc, sc->lockbuf);
462 
463 	snprintf(sc->reglockbuf, ADAPTER_LOCK_NAME_LEN, "SGE reg lock %d",
464 	    device_get_unit(dev));
465 	snprintf(sc->mdiolockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb mdio lock %d",
466 	    device_get_unit(dev));
467 	snprintf(sc->elmerlockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb elmer lock %d",
468 	    device_get_unit(dev));
469 
470 	MTX_INIT(&sc->sge.reg_lock, sc->reglockbuf, NULL, MTX_SPIN);
471 	MTX_INIT(&sc->mdio_lock, sc->mdiolockbuf, NULL, MTX_DEF);
472 	MTX_INIT(&sc->elmer_lock, sc->elmerlockbuf, NULL, MTX_DEF);
473 
474 	mtx_lock(&t3_list_lock);
475 	SLIST_INSERT_HEAD(&t3_list, sc, link);
476 	mtx_unlock(&t3_list_lock);
477 
478 	/* find the PCIe link width and set max read request to 4KB*/
479 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
480 		uint16_t lnk;
481 
482 		lnk = pci_read_config(dev, reg + PCIER_LINK_STA, 2);
483 		sc->link_width = (lnk & PCIEM_LINK_STA_WIDTH) >> 4;
484 		if (sc->link_width < 8 &&
485 		    (ai->caps & SUPPORTED_10000baseT_Full)) {
486 			device_printf(sc->dev,
487 			    "PCIe x%d Link, expect reduced performance\n",
488 			    sc->link_width);
489 		}
490 
491 		pci_set_max_read_req(dev, 4096);
492 	}
493 
494 	touch_bars(dev);
495 	pci_enable_busmaster(dev);
496 	/*
497 	 * Allocate the registers and make them available to the driver.
498 	 * The registers that we care about for NIC mode are in BAR 0
499 	 */
500 	sc->regs_rid = PCIR_BAR(0);
501 	if ((sc->regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
502 	    &sc->regs_rid, RF_ACTIVE)) == NULL) {
503 		device_printf(dev, "Cannot allocate BAR region 0\n");
504 		error = ENXIO;
505 		goto out;
506 	}
507 
508 	sc->bt = rman_get_bustag(sc->regs_res);
509 	sc->bh = rman_get_bushandle(sc->regs_res);
510 	sc->mmio_len = rman_get_size(sc->regs_res);
511 
512 	for (i = 0; i < MAX_NPORTS; i++)
513 		sc->port[i].adapter = sc;
514 
515 	if (t3_prep_adapter(sc, ai, 1) < 0) {
516 		printf("prep adapter failed\n");
517 		error = ENODEV;
518 		goto out;
519 	}
520 
521 	sc->udbs_rid = PCIR_BAR(2);
522 	sc->udbs_res = NULL;
523 	if (is_offload(sc) &&
524 	    ((sc->udbs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
525 		   &sc->udbs_rid, RF_ACTIVE)) == NULL)) {
526 		device_printf(dev, "Cannot allocate BAR region 1\n");
527 		error = ENXIO;
528 		goto out;
529 	}
530 
531         /* Allocate the BAR for doing MSI-X.  If it succeeds, try to allocate
532 	 * enough messages for the queue sets.  If that fails, try falling
533 	 * back to MSI.  If that fails, then try falling back to the legacy
534 	 * interrupt pin model.
535 	 */
536 	sc->msix_regs_rid = 0x20;
537 	if ((msi_allowed >= 2) &&
538 	    (sc->msix_regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
539 	    &sc->msix_regs_rid, RF_ACTIVE)) != NULL) {
540 
541 		if (multiq)
542 			port_qsets = min(SGE_QSETS/sc->params.nports, mp_ncpus);
543 		msi_needed = sc->msi_count = sc->params.nports * port_qsets + 1;
544 
545 		if (pci_msix_count(dev) == 0 ||
546 		    (error = pci_alloc_msix(dev, &sc->msi_count)) != 0 ||
547 		    sc->msi_count != msi_needed) {
548 			device_printf(dev, "alloc msix failed - "
549 				      "msi_count=%d, msi_needed=%d, err=%d; "
550 				      "will try MSI\n", sc->msi_count,
551 				      msi_needed, error);
552 			sc->msi_count = 0;
553 			port_qsets = 1;
554 			pci_release_msi(dev);
555 			bus_release_resource(dev, SYS_RES_MEMORY,
556 			    sc->msix_regs_rid, sc->msix_regs_res);
557 			sc->msix_regs_res = NULL;
558 		} else {
559 			sc->flags |= USING_MSIX;
560 			sc->cxgb_intr = cxgb_async_intr;
561 			device_printf(dev,
562 				      "using MSI-X interrupts (%u vectors)\n",
563 				      sc->msi_count);
564 		}
565 	}
566 
567 	if ((msi_allowed >= 1) && (sc->msi_count == 0)) {
568 		sc->msi_count = 1;
569 		if ((error = pci_alloc_msi(dev, &sc->msi_count)) != 0) {
570 			device_printf(dev, "alloc msi failed - "
571 				      "err=%d; will try INTx\n", error);
572 			sc->msi_count = 0;
573 			port_qsets = 1;
574 			pci_release_msi(dev);
575 		} else {
576 			sc->flags |= USING_MSI;
577 			sc->cxgb_intr = t3_intr_msi;
578 			device_printf(dev, "using MSI interrupts\n");
579 		}
580 	}
581 	if (sc->msi_count == 0) {
582 		device_printf(dev, "using line interrupts\n");
583 		sc->cxgb_intr = t3b_intr;
584 	}
585 
586 	/* Create a private taskqueue thread for handling driver events */
587 	sc->tq = taskqueue_create("cxgb_taskq", M_NOWAIT,
588 	    taskqueue_thread_enqueue, &sc->tq);
589 	if (sc->tq == NULL) {
590 		device_printf(dev, "failed to allocate controller task queue\n");
591 		goto out;
592 	}
593 
594 	taskqueue_start_threads(&sc->tq, 1, PI_NET, "%s taskq",
595 	    device_get_nameunit(dev));
596 	TASK_INIT(&sc->tick_task, 0, cxgb_tick_handler, sc);
597 
598 
599 	/* Create a periodic callout for checking adapter status */
600 	callout_init(&sc->cxgb_tick_ch, 1);
601 
602 	if (t3_check_fw_version(sc) < 0 || force_fw_update) {
603 		/*
604 		 * Warn user that a firmware update will be attempted in init.
605 		 */
606 		device_printf(dev, "firmware needs to be updated to version %d.%d.%d\n",
607 		    FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
608 		sc->flags &= ~FW_UPTODATE;
609 	} else {
610 		sc->flags |= FW_UPTODATE;
611 	}
612 
613 	if (t3_check_tpsram_version(sc) < 0) {
614 		/*
615 		 * Warn user that a firmware update will be attempted in init.
616 		 */
617 		device_printf(dev, "SRAM needs to be updated to version %c-%d.%d.%d\n",
618 		    t3rev2char(sc), TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
619 		sc->flags &= ~TPS_UPTODATE;
620 	} else {
621 		sc->flags |= TPS_UPTODATE;
622 	}
623 
624 	/*
625 	 * Create a child device for each MAC.  The ethernet attachment
626 	 * will be done in these children.
627 	 */
628 	for (i = 0; i < (sc)->params.nports; i++) {
629 		struct port_info *pi;
630 
631 		if ((child = device_add_child(dev, "cxgb", -1)) == NULL) {
632 			device_printf(dev, "failed to add child port\n");
633 			error = EINVAL;
634 			goto out;
635 		}
636 		pi = &sc->port[i];
637 		pi->adapter = sc;
638 		pi->nqsets = port_qsets;
639 		pi->first_qset = i*port_qsets;
640 		pi->port_id = i;
641 		pi->tx_chan = i >= ai->nports0;
642 		pi->txpkt_intf = pi->tx_chan ? 2 * (i - ai->nports0) + 1 : 2 * i;
643 		sc->rxpkt_map[pi->txpkt_intf] = i;
644 		sc->port[i].tx_chan = i >= ai->nports0;
645 		sc->portdev[i] = child;
646 		device_set_softc(child, pi);
647 	}
648 	if ((error = bus_generic_attach(dev)) != 0)
649 		goto out;
650 
651 	/* initialize sge private state */
652 	t3_sge_init_adapter(sc);
653 
654 	t3_led_ready(sc);
655 
656 	error = t3_get_fw_version(sc, &vers);
657 	if (error)
658 		goto out;
659 
660 	snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
661 	    G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
662 	    G_FW_VERSION_MICRO(vers));
663 
664 	snprintf(buf, sizeof(buf), "%s %sNIC\t E/C: %s S/N: %s",
665 		 ai->desc, is_offload(sc) ? "R" : "",
666 		 sc->params.vpd.ec, sc->params.vpd.sn);
667 	device_set_desc_copy(dev, buf);
668 
669 	snprintf(&sc->port_types[0], sizeof(sc->port_types), "%x%x%x%x",
670 		 sc->params.vpd.port_type[0], sc->params.vpd.port_type[1],
671 		 sc->params.vpd.port_type[2], sc->params.vpd.port_type[3]);
672 
673 	device_printf(sc->dev, "Firmware Version %s\n", &sc->fw_version[0]);
674 	callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
675 	t3_add_attach_sysctls(sc);
676 
677 #ifdef TCP_OFFLOAD
678 	for (i = 0; i < NUM_CPL_HANDLERS; i++)
679 		sc->cpl_handler[i] = cpl_not_handled;
680 #endif
681 
682 	t3_intr_clear(sc);
683 	error = cxgb_setup_interrupts(sc);
684 out:
685 	if (error)
686 		cxgb_free(sc);
687 
688 	return (error);
689 }
690 
691 /*
692  * The cxgb_controller_detach routine is called with the device is
693  * unloaded from the system.
694  */
695 
696 static int
697 cxgb_controller_detach(device_t dev)
698 {
699 	struct adapter *sc;
700 
701 	sc = device_get_softc(dev);
702 
703 	cxgb_free(sc);
704 
705 	return (0);
706 }
707 
708 /*
709  * The cxgb_free() is called by the cxgb_controller_detach() routine
710  * to tear down the structures that were built up in
711  * cxgb_controller_attach(), and should be the final piece of work
712  * done when fully unloading the driver.
713  *
714  *
715  *  1. Shutting down the threads started by the cxgb_controller_attach()
716  *     routine.
717  *  2. Stopping the lower level device and all callouts (cxgb_down_locked()).
718  *  3. Detaching all of the port devices created during the
719  *     cxgb_controller_attach() routine.
720  *  4. Removing the device children created via cxgb_controller_attach().
721  *  5. Releasing PCI resources associated with the device.
722  *  6. Turning off the offload support, iff it was turned on.
723  *  7. Destroying the mutexes created in cxgb_controller_attach().
724  *
725  */
726 static void
727 cxgb_free(struct adapter *sc)
728 {
729 	int i, nqsets = 0;
730 
731 	ADAPTER_LOCK(sc);
732 	sc->flags |= CXGB_SHUTDOWN;
733 	ADAPTER_UNLOCK(sc);
734 
735 	/*
736 	 * Make sure all child devices are gone.
737 	 */
738 	bus_generic_detach(sc->dev);
739 	for (i = 0; i < (sc)->params.nports; i++) {
740 		if (sc->portdev[i] &&
741 		    device_delete_child(sc->dev, sc->portdev[i]) != 0)
742 			device_printf(sc->dev, "failed to delete child port\n");
743 		nqsets += sc->port[i].nqsets;
744 	}
745 
746 	/*
747 	 * At this point, it is as if cxgb_port_detach has run on all ports, and
748 	 * cxgb_down has run on the adapter.  All interrupts have been silenced,
749 	 * all open devices have been closed.
750 	 */
751 	KASSERT(sc->open_device_map == 0, ("%s: device(s) still open (%x)",
752 					   __func__, sc->open_device_map));
753 	for (i = 0; i < sc->params.nports; i++) {
754 		KASSERT(sc->port[i].ifp == NULL, ("%s: port %i undead!",
755 						  __func__, i));
756 	}
757 
758 	/*
759 	 * Finish off the adapter's callouts.
760 	 */
761 	callout_drain(&sc->cxgb_tick_ch);
762 	callout_drain(&sc->sge_timer_ch);
763 
764 	/*
765 	 * Release resources grabbed under FULL_INIT_DONE by cxgb_up.  The
766 	 * sysctls are cleaned up by the kernel linker.
767 	 */
768 	if (sc->flags & FULL_INIT_DONE) {
769  		t3_free_sge_resources(sc, nqsets);
770  		sc->flags &= ~FULL_INIT_DONE;
771  	}
772 
773 	/*
774 	 * Release all interrupt resources.
775 	 */
776 	cxgb_teardown_interrupts(sc);
777 	if (sc->flags & (USING_MSI | USING_MSIX)) {
778 		device_printf(sc->dev, "releasing msi message(s)\n");
779 		pci_release_msi(sc->dev);
780 	} else {
781 		device_printf(sc->dev, "no msi message to release\n");
782 	}
783 
784 	if (sc->msix_regs_res != NULL) {
785 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->msix_regs_rid,
786 		    sc->msix_regs_res);
787 	}
788 
789 	/*
790 	 * Free the adapter's taskqueue.
791 	 */
792 	if (sc->tq != NULL) {
793 		taskqueue_free(sc->tq);
794 		sc->tq = NULL;
795 	}
796 
797 	free(sc->filters, M_DEVBUF);
798 	t3_sge_free(sc);
799 
800 	if (sc->udbs_res != NULL)
801 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->udbs_rid,
802 		    sc->udbs_res);
803 
804 	if (sc->regs_res != NULL)
805 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->regs_rid,
806 		    sc->regs_res);
807 
808 	MTX_DESTROY(&sc->mdio_lock);
809 	MTX_DESTROY(&sc->sge.reg_lock);
810 	MTX_DESTROY(&sc->elmer_lock);
811 	mtx_lock(&t3_list_lock);
812 	SLIST_REMOVE(&t3_list, sc, adapter, link);
813 	mtx_unlock(&t3_list_lock);
814 	ADAPTER_LOCK_DEINIT(sc);
815 }
816 
817 /**
818  *	setup_sge_qsets - configure SGE Tx/Rx/response queues
819  *	@sc: the controller softc
820  *
821  *	Determines how many sets of SGE queues to use and initializes them.
822  *	We support multiple queue sets per port if we have MSI-X, otherwise
823  *	just one queue set per port.
824  */
825 static int
826 setup_sge_qsets(adapter_t *sc)
827 {
828 	int i, j, err, irq_idx = 0, qset_idx = 0;
829 	u_int ntxq = SGE_TXQ_PER_SET;
830 
831 	if ((err = t3_sge_alloc(sc)) != 0) {
832 		device_printf(sc->dev, "t3_sge_alloc returned %d\n", err);
833 		return (err);
834 	}
835 
836 	if (sc->params.rev > 0 && !(sc->flags & USING_MSI))
837 		irq_idx = -1;
838 
839 	for (i = 0; i < (sc)->params.nports; i++) {
840 		struct port_info *pi = &sc->port[i];
841 
842 		for (j = 0; j < pi->nqsets; j++, qset_idx++) {
843 			err = t3_sge_alloc_qset(sc, qset_idx, (sc)->params.nports,
844 			    (sc->flags & USING_MSIX) ? qset_idx + 1 : irq_idx,
845 			    &sc->params.sge.qset[qset_idx], ntxq, pi);
846 			if (err) {
847 				t3_free_sge_resources(sc, qset_idx);
848 				device_printf(sc->dev,
849 				    "t3_sge_alloc_qset failed with %d\n", err);
850 				return (err);
851 			}
852 		}
853 	}
854 
855 	sc->nqsets = qset_idx;
856 
857 	return (0);
858 }
859 
860 static void
861 cxgb_teardown_interrupts(adapter_t *sc)
862 {
863 	int i;
864 
865 	for (i = 0; i < SGE_QSETS; i++) {
866 		if (sc->msix_intr_tag[i] == NULL) {
867 
868 			/* Should have been setup fully or not at all */
869 			KASSERT(sc->msix_irq_res[i] == NULL &&
870 				sc->msix_irq_rid[i] == 0,
871 				("%s: half-done interrupt (%d).", __func__, i));
872 
873 			continue;
874 		}
875 
876 		bus_teardown_intr(sc->dev, sc->msix_irq_res[i],
877 				  sc->msix_intr_tag[i]);
878 		bus_release_resource(sc->dev, SYS_RES_IRQ, sc->msix_irq_rid[i],
879 				     sc->msix_irq_res[i]);
880 
881 		sc->msix_irq_res[i] = sc->msix_intr_tag[i] = NULL;
882 		sc->msix_irq_rid[i] = 0;
883 	}
884 
885 	if (sc->intr_tag) {
886 		KASSERT(sc->irq_res != NULL,
887 			("%s: half-done interrupt.", __func__));
888 
889 		bus_teardown_intr(sc->dev, sc->irq_res, sc->intr_tag);
890 		bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
891 				     sc->irq_res);
892 
893 		sc->irq_res = sc->intr_tag = NULL;
894 		sc->irq_rid = 0;
895 	}
896 }
897 
898 static int
899 cxgb_setup_interrupts(adapter_t *sc)
900 {
901 	struct resource *res;
902 	void *tag;
903 	int i, rid, err, intr_flag = sc->flags & (USING_MSI | USING_MSIX);
904 
905 	sc->irq_rid = intr_flag ? 1 : 0;
906 	sc->irq_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &sc->irq_rid,
907 					     RF_SHAREABLE | RF_ACTIVE);
908 	if (sc->irq_res == NULL) {
909 		device_printf(sc->dev, "Cannot allocate interrupt (%x, %u)\n",
910 			      intr_flag, sc->irq_rid);
911 		err = EINVAL;
912 		sc->irq_rid = 0;
913 	} else {
914 		err = bus_setup_intr(sc->dev, sc->irq_res,
915 		    INTR_MPSAFE | INTR_TYPE_NET, NULL,
916 		    sc->cxgb_intr, sc, &sc->intr_tag);
917 
918 		if (err) {
919 			device_printf(sc->dev,
920 				      "Cannot set up interrupt (%x, %u, %d)\n",
921 				      intr_flag, sc->irq_rid, err);
922 			bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
923 					     sc->irq_res);
924 			sc->irq_res = sc->intr_tag = NULL;
925 			sc->irq_rid = 0;
926 		}
927 	}
928 
929 	/* That's all for INTx or MSI */
930 	if (!(intr_flag & USING_MSIX) || err)
931 		return (err);
932 
933 	bus_describe_intr(sc->dev, sc->irq_res, sc->intr_tag, "err");
934 	for (i = 0; i < sc->msi_count - 1; i++) {
935 		rid = i + 2;
936 		res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &rid,
937 					     RF_SHAREABLE | RF_ACTIVE);
938 		if (res == NULL) {
939 			device_printf(sc->dev, "Cannot allocate interrupt "
940 				      "for message %d\n", rid);
941 			err = EINVAL;
942 			break;
943 		}
944 
945 		err = bus_setup_intr(sc->dev, res, INTR_MPSAFE | INTR_TYPE_NET,
946 				     NULL, t3_intr_msix, &sc->sge.qs[i], &tag);
947 		if (err) {
948 			device_printf(sc->dev, "Cannot set up interrupt "
949 				      "for message %d (%d)\n", rid, err);
950 			bus_release_resource(sc->dev, SYS_RES_IRQ, rid, res);
951 			break;
952 		}
953 
954 		sc->msix_irq_rid[i] = rid;
955 		sc->msix_irq_res[i] = res;
956 		sc->msix_intr_tag[i] = tag;
957 		bus_describe_intr(sc->dev, res, tag, "qs%d", i);
958 	}
959 
960 	if (err)
961 		cxgb_teardown_interrupts(sc);
962 
963 	return (err);
964 }
965 
966 
967 static int
968 cxgb_port_probe(device_t dev)
969 {
970 	struct port_info *p;
971 	char buf[80];
972 	const char *desc;
973 
974 	p = device_get_softc(dev);
975 	desc = p->phy.desc;
976 	snprintf(buf, sizeof(buf), "Port %d %s", p->port_id, desc);
977 	device_set_desc_copy(dev, buf);
978 	return (0);
979 }
980 
981 
982 static int
983 cxgb_makedev(struct port_info *pi)
984 {
985 
986 	pi->port_cdev = make_dev(&cxgb_cdevsw, if_getdunit(pi->ifp),
987 	    UID_ROOT, GID_WHEEL, 0600, "%s", if_name(pi->ifp));
988 
989 	if (pi->port_cdev == NULL)
990 		return (ENOMEM);
991 
992 	pi->port_cdev->si_drv1 = (void *)pi;
993 
994 	return (0);
995 }
996 
997 #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
998     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
999     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6)
1000 #define CXGB_CAP_ENABLE CXGB_CAP
1001 
1002 static int
1003 cxgb_port_attach(device_t dev)
1004 {
1005 	struct port_info *p;
1006 	if_t ifp;
1007 	int err;
1008 	struct adapter *sc;
1009 
1010 	p = device_get_softc(dev);
1011 	sc = p->adapter;
1012 	snprintf(p->lockbuf, PORT_NAME_LEN, "cxgb port lock %d:%d",
1013 	    device_get_unit(device_get_parent(dev)), p->port_id);
1014 	PORT_LOCK_INIT(p, p->lockbuf);
1015 
1016 	callout_init(&p->link_check_ch, 1);
1017 	TASK_INIT(&p->link_check_task, 0, check_link_status, p);
1018 
1019 	/* Allocate an ifnet object and set it up */
1020 	ifp = p->ifp = if_alloc(IFT_ETHER);
1021 	if (ifp == NULL) {
1022 		device_printf(dev, "Cannot allocate ifnet\n");
1023 		return (ENOMEM);
1024 	}
1025 
1026 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1027 	if_setinitfn(ifp, cxgb_init);
1028 	if_setsoftc(ifp, p);
1029 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
1030 	if_setioctlfn(ifp, cxgb_ioctl);
1031 	if_settransmitfn(ifp, cxgb_transmit);
1032 	if_setqflushfn(ifp, cxgb_qflush);
1033 	if_setgetcounterfn(ifp, cxgb_get_counter);
1034 
1035 	if_setcapabilities(ifp, CXGB_CAP);
1036 #ifdef TCP_OFFLOAD
1037 	if (is_offload(sc))
1038 		if_setcapabilitiesbit(ifp, IFCAP_TOE4, 0);
1039 #endif
1040 	if_setcapenable(ifp, CXGB_CAP_ENABLE);
1041 	if_sethwassist(ifp, CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1042 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1043 	if_sethwtsomax(ifp, IP_MAXPACKET);
1044 	if_sethwtsomaxsegcount(ifp, 36);
1045 	if_sethwtsomaxsegsize(ifp, 65536);
1046 
1047 	/*
1048 	 * Disable TSO on 4-port - it isn't supported by the firmware.
1049 	 */
1050 	if (sc->params.nports > 2) {
1051 		if_setcapabilitiesbit(ifp, 0, IFCAP_TSO | IFCAP_VLAN_HWTSO);
1052 		if_setcapenablebit(ifp, 0, IFCAP_TSO | IFCAP_VLAN_HWTSO);
1053 		if_sethwassistbits(ifp, 0, CSUM_TSO);
1054 	}
1055 
1056 	ether_ifattach(ifp, p->hw_addr);
1057 
1058 	/* Attach driver debugnet methods. */
1059 	DEBUGNET_SET(ifp, cxgb);
1060 
1061 #ifdef DEFAULT_JUMBO
1062 	if (sc->params.nports <= 2)
1063 		if_setmtu(ifp, ETHERMTU_JUMBO);
1064 #endif
1065 	if ((err = cxgb_makedev(p)) != 0) {
1066 		printf("makedev failed %d\n", err);
1067 		return (err);
1068 	}
1069 
1070 	/* Create a list of media supported by this port */
1071 	ifmedia_init(&p->media, IFM_IMASK, cxgb_media_change,
1072 	    cxgb_media_status);
1073 	cxgb_build_medialist(p);
1074 
1075 	t3_sge_init_port(p);
1076 
1077 	return (err);
1078 }
1079 
1080 /*
1081  * cxgb_port_detach() is called via the device_detach methods when
1082  * cxgb_free() calls the bus_generic_detach.  It is responsible for
1083  * removing the device from the view of the kernel, i.e. from all
1084  * interfaces lists etc.  This routine is only called when the driver is
1085  * being unloaded, not when the link goes down.
1086  */
1087 static int
1088 cxgb_port_detach(device_t dev)
1089 {
1090 	struct port_info *p;
1091 	struct adapter *sc;
1092 	int i;
1093 
1094 	p = device_get_softc(dev);
1095 	sc = p->adapter;
1096 
1097 	/* Tell cxgb_ioctl and if_init that the port is going away */
1098 	ADAPTER_LOCK(sc);
1099 	SET_DOOMED(p);
1100 	wakeup(&sc->flags);
1101 	while (IS_BUSY(sc))
1102 		mtx_sleep(&sc->flags, &sc->lock, 0, "cxgbdtch", 0);
1103 	SET_BUSY(sc);
1104 	ADAPTER_UNLOCK(sc);
1105 
1106 	if (p->port_cdev != NULL)
1107 		destroy_dev(p->port_cdev);
1108 
1109 	cxgb_uninit_synchronized(p);
1110 	ether_ifdetach(p->ifp);
1111 
1112 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1113 		struct sge_qset *qs = &sc->sge.qs[i];
1114 		struct sge_txq *txq = &qs->txq[TXQ_ETH];
1115 
1116 		callout_drain(&txq->txq_watchdog);
1117 		callout_drain(&txq->txq_timer);
1118 	}
1119 
1120 	PORT_LOCK_DEINIT(p);
1121 	if_free(p->ifp);
1122 	p->ifp = NULL;
1123 
1124 	ADAPTER_LOCK(sc);
1125 	CLR_BUSY(sc);
1126 	wakeup_one(&sc->flags);
1127 	ADAPTER_UNLOCK(sc);
1128 	return (0);
1129 }
1130 
1131 void
1132 t3_fatal_err(struct adapter *sc)
1133 {
1134 	u_int fw_status[4];
1135 
1136 	if (sc->flags & FULL_INIT_DONE) {
1137 		t3_sge_stop(sc);
1138 		t3_write_reg(sc, A_XGM_TX_CTRL, 0);
1139 		t3_write_reg(sc, A_XGM_RX_CTRL, 0);
1140 		t3_write_reg(sc, XGM_REG(A_XGM_TX_CTRL, 1), 0);
1141 		t3_write_reg(sc, XGM_REG(A_XGM_RX_CTRL, 1), 0);
1142 		t3_intr_disable(sc);
1143 	}
1144 	device_printf(sc->dev,"encountered fatal error, operation suspended\n");
1145 	if (!t3_cim_ctl_blk_read(sc, 0xa0, 4, fw_status))
1146 		device_printf(sc->dev, "FW_ status: 0x%x, 0x%x, 0x%x, 0x%x\n",
1147 		    fw_status[0], fw_status[1], fw_status[2], fw_status[3]);
1148 }
1149 
1150 int
1151 t3_os_find_pci_capability(adapter_t *sc, int cap)
1152 {
1153 	device_t dev;
1154 	struct pci_devinfo *dinfo;
1155 	pcicfgregs *cfg;
1156 	uint32_t status;
1157 	uint8_t ptr;
1158 
1159 	dev = sc->dev;
1160 	dinfo = device_get_ivars(dev);
1161 	cfg = &dinfo->cfg;
1162 
1163 	status = pci_read_config(dev, PCIR_STATUS, 2);
1164 	if (!(status & PCIM_STATUS_CAPPRESENT))
1165 		return (0);
1166 
1167 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1168 	case 0:
1169 	case 1:
1170 		ptr = PCIR_CAP_PTR;
1171 		break;
1172 	case 2:
1173 		ptr = PCIR_CAP_PTR_2;
1174 		break;
1175 	default:
1176 		return (0);
1177 		break;
1178 	}
1179 	ptr = pci_read_config(dev, ptr, 1);
1180 
1181 	while (ptr != 0) {
1182 		if (pci_read_config(dev, ptr + PCICAP_ID, 1) == cap)
1183 			return (ptr);
1184 		ptr = pci_read_config(dev, ptr + PCICAP_NEXTPTR, 1);
1185 	}
1186 
1187 	return (0);
1188 }
1189 
1190 int
1191 t3_os_pci_save_state(struct adapter *sc)
1192 {
1193 	device_t dev;
1194 	struct pci_devinfo *dinfo;
1195 
1196 	dev = sc->dev;
1197 	dinfo = device_get_ivars(dev);
1198 
1199 	pci_cfg_save(dev, dinfo, 0);
1200 	return (0);
1201 }
1202 
1203 int
1204 t3_os_pci_restore_state(struct adapter *sc)
1205 {
1206 	device_t dev;
1207 	struct pci_devinfo *dinfo;
1208 
1209 	dev = sc->dev;
1210 	dinfo = device_get_ivars(dev);
1211 
1212 	pci_cfg_restore(dev, dinfo);
1213 	return (0);
1214 }
1215 
1216 /**
1217  *	t3_os_link_changed - handle link status changes
1218  *	@sc: the adapter associated with the link change
1219  *	@port_id: the port index whose link status has changed
1220  *	@link_status: the new status of the link
1221  *	@speed: the new speed setting
1222  *	@duplex: the new duplex setting
1223  *	@fc: the new flow-control setting
1224  *
1225  *	This is the OS-dependent handler for link status changes.  The OS
1226  *	neutral handler takes care of most of the processing for these events,
1227  *	then calls this handler for any OS-specific processing.
1228  */
1229 void
1230 t3_os_link_changed(adapter_t *adapter, int port_id, int link_status, int speed,
1231      int duplex, int fc, int mac_was_reset)
1232 {
1233 	struct port_info *pi = &adapter->port[port_id];
1234 	if_t ifp = pi->ifp;
1235 
1236 	/* no race with detach, so ifp should always be good */
1237 	KASSERT(ifp, ("%s: if detached.", __func__));
1238 
1239 	/* Reapply mac settings if they were lost due to a reset */
1240 	if (mac_was_reset) {
1241 		PORT_LOCK(pi);
1242 		cxgb_update_mac_settings(pi);
1243 		PORT_UNLOCK(pi);
1244 	}
1245 
1246 	if (link_status) {
1247 		if_setbaudrate(ifp, IF_Mbps(speed));
1248 		if_link_state_change(ifp, LINK_STATE_UP);
1249 	} else
1250 		if_link_state_change(ifp, LINK_STATE_DOWN);
1251 }
1252 
1253 /**
1254  *	t3_os_phymod_changed - handle PHY module changes
1255  *	@phy: the PHY reporting the module change
1256  *	@mod_type: new module type
1257  *
1258  *	This is the OS-dependent handler for PHY module changes.  It is
1259  *	invoked when a PHY module is removed or inserted for any OS-specific
1260  *	processing.
1261  */
1262 void t3_os_phymod_changed(struct adapter *adap, int port_id)
1263 {
1264 	static const char *mod_str[] = {
1265 		NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX-L", "unknown"
1266 	};
1267 	struct port_info *pi = &adap->port[port_id];
1268 	int mod = pi->phy.modtype;
1269 
1270 	if (mod != pi->media.ifm_cur->ifm_data)
1271 		cxgb_build_medialist(pi);
1272 
1273 	if (mod == phy_modtype_none)
1274 		if_printf(pi->ifp, "PHY module unplugged\n");
1275 	else {
1276 		KASSERT(mod < ARRAY_SIZE(mod_str),
1277 			("invalid PHY module type %d", mod));
1278 		if_printf(pi->ifp, "%s PHY module inserted\n", mod_str[mod]);
1279 	}
1280 }
1281 
1282 void
1283 t3_os_set_hw_addr(adapter_t *adapter, int port_idx, u8 hw_addr[])
1284 {
1285 
1286 	/*
1287 	 * The ifnet might not be allocated before this gets called,
1288 	 * as this is called early on in attach by t3_prep_adapter
1289 	 * save the address off in the port structure
1290 	 */
1291 	if (cxgb_debug)
1292 		printf("set_hw_addr on idx %d addr %6D\n", port_idx, hw_addr, ":");
1293 	bcopy(hw_addr, adapter->port[port_idx].hw_addr, ETHER_ADDR_LEN);
1294 }
1295 
1296 /*
1297  * Programs the XGMAC based on the settings in the ifnet.  These settings
1298  * include MTU, MAC address, mcast addresses, etc.
1299  */
1300 static void
1301 cxgb_update_mac_settings(struct port_info *p)
1302 {
1303 	if_t ifp = p->ifp;
1304 	struct t3_rx_mode rm;
1305 	struct cmac *mac = &p->mac;
1306 	int mtu, hwtagging;
1307 
1308 	PORT_LOCK_ASSERT_OWNED(p);
1309 
1310 	bcopy(if_getlladdr(ifp), p->hw_addr, ETHER_ADDR_LEN);
1311 
1312 	mtu = if_getmtu(ifp);
1313 	if (if_getcapenable(ifp) & IFCAP_VLAN_MTU)
1314 		mtu += ETHER_VLAN_ENCAP_LEN;
1315 
1316 	hwtagging = (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0;
1317 
1318 	t3_mac_set_mtu(mac, mtu);
1319 	t3_set_vlan_accel(p->adapter, 1 << p->tx_chan, hwtagging);
1320 	t3_mac_set_address(mac, 0, p->hw_addr);
1321 	t3_init_rx_mode(&rm, p);
1322 	t3_mac_set_rx_mode(mac, &rm);
1323 }
1324 
1325 
1326 static int
1327 await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
1328 			      unsigned long n)
1329 {
1330 	int attempts = 5;
1331 
1332 	while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
1333 		if (!--attempts)
1334 			return (ETIMEDOUT);
1335 		t3_os_sleep(10);
1336 	}
1337 	return 0;
1338 }
1339 
1340 static int
1341 init_tp_parity(struct adapter *adap)
1342 {
1343 	int i;
1344 	struct mbuf *m;
1345 	struct cpl_set_tcb_field *greq;
1346 	unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
1347 
1348 	t3_tp_set_offload_mode(adap, 1);
1349 
1350 	for (i = 0; i < 16; i++) {
1351 		struct cpl_smt_write_req *req;
1352 
1353 		m = m_gethdr(M_WAITOK, MT_DATA);
1354 		req = mtod(m, struct cpl_smt_write_req *);
1355 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1356 		memset(req, 0, sizeof(*req));
1357 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1358 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
1359 		req->iff = i;
1360 		t3_mgmt_tx(adap, m);
1361 	}
1362 
1363 	for (i = 0; i < 2048; i++) {
1364 		struct cpl_l2t_write_req *req;
1365 
1366 		m = m_gethdr(M_WAITOK, MT_DATA);
1367 		req = mtod(m, struct cpl_l2t_write_req *);
1368 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1369 		memset(req, 0, sizeof(*req));
1370 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1371 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
1372 		req->params = htonl(V_L2T_W_IDX(i));
1373 		t3_mgmt_tx(adap, m);
1374 	}
1375 
1376 	for (i = 0; i < 2048; i++) {
1377 		struct cpl_rte_write_req *req;
1378 
1379 		m = m_gethdr(M_WAITOK, MT_DATA);
1380 		req = mtod(m, struct cpl_rte_write_req *);
1381 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1382 		memset(req, 0, sizeof(*req));
1383 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1384 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
1385 		req->l2t_idx = htonl(V_L2T_W_IDX(i));
1386 		t3_mgmt_tx(adap, m);
1387 	}
1388 
1389 	m = m_gethdr(M_WAITOK, MT_DATA);
1390 	greq = mtod(m, struct cpl_set_tcb_field *);
1391 	m->m_len = m->m_pkthdr.len = sizeof(*greq);
1392 	memset(greq, 0, sizeof(*greq));
1393 	greq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1394 	OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
1395 	greq->mask = htobe64(1);
1396 	t3_mgmt_tx(adap, m);
1397 
1398 	i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
1399 	t3_tp_set_offload_mode(adap, 0);
1400 	return (i);
1401 }
1402 
1403 /**
1404  *	setup_rss - configure Receive Side Steering (per-queue connection demux)
1405  *	@adap: the adapter
1406  *
1407  *	Sets up RSS to distribute packets to multiple receive queues.  We
1408  *	configure the RSS CPU lookup table to distribute to the number of HW
1409  *	receive queues, and the response queue lookup table to narrow that
1410  *	down to the response queues actually configured for each port.
1411  *	We always configure the RSS mapping for two ports since the mapping
1412  *	table has plenty of entries.
1413  */
1414 static void
1415 setup_rss(adapter_t *adap)
1416 {
1417 	int i;
1418 	u_int nq[2];
1419 	uint8_t cpus[SGE_QSETS + 1];
1420 	uint16_t rspq_map[RSS_TABLE_SIZE];
1421 
1422 	for (i = 0; i < SGE_QSETS; ++i)
1423 		cpus[i] = i;
1424 	cpus[SGE_QSETS] = 0xff;
1425 
1426 	nq[0] = nq[1] = 0;
1427 	for_each_port(adap, i) {
1428 		const struct port_info *pi = adap2pinfo(adap, i);
1429 
1430 		nq[pi->tx_chan] += pi->nqsets;
1431 	}
1432 	for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
1433 		rspq_map[i] = nq[0] ? i % nq[0] : 0;
1434 		rspq_map[i + RSS_TABLE_SIZE / 2] = nq[1] ? i % nq[1] + nq[0] : 0;
1435 	}
1436 
1437 	/* Calculate the reverse RSS map table */
1438 	for (i = 0; i < SGE_QSETS; ++i)
1439 		adap->rrss_map[i] = 0xff;
1440 	for (i = 0; i < RSS_TABLE_SIZE; ++i)
1441 		if (adap->rrss_map[rspq_map[i]] == 0xff)
1442 			adap->rrss_map[rspq_map[i]] = i;
1443 
1444 	t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
1445 		      F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN | F_OFDMAPEN |
1446 	              F_RRCPLMAPEN | V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ,
1447 	              cpus, rspq_map);
1448 
1449 }
1450 static void
1451 send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
1452 			      int hi, int port)
1453 {
1454 	struct mbuf *m;
1455 	struct mngt_pktsched_wr *req;
1456 
1457 	m = m_gethdr(M_NOWAIT, MT_DATA);
1458 	if (m) {
1459 		req = mtod(m, struct mngt_pktsched_wr *);
1460 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
1461 		req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
1462 		req->sched = sched;
1463 		req->idx = qidx;
1464 		req->min = lo;
1465 		req->max = hi;
1466 		req->binding = port;
1467 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1468 		t3_mgmt_tx(adap, m);
1469 	}
1470 }
1471 
1472 static void
1473 bind_qsets(adapter_t *sc)
1474 {
1475 	int i, j;
1476 
1477 	for (i = 0; i < (sc)->params.nports; ++i) {
1478 		const struct port_info *pi = adap2pinfo(sc, i);
1479 
1480 		for (j = 0; j < pi->nqsets; ++j) {
1481 			send_pktsched_cmd(sc, 1, pi->first_qset + j, -1,
1482 					  -1, pi->tx_chan);
1483 
1484 		}
1485 	}
1486 }
1487 
1488 static void
1489 update_tpeeprom(struct adapter *adap)
1490 {
1491 	const struct firmware *tpeeprom;
1492 
1493 	uint32_t version;
1494 	unsigned int major, minor;
1495 	int ret, len;
1496 	char rev, name[32];
1497 
1498 	t3_seeprom_read(adap, TP_SRAM_OFFSET, &version);
1499 
1500 	major = G_TP_VERSION_MAJOR(version);
1501 	minor = G_TP_VERSION_MINOR(version);
1502 	if (major == TP_VERSION_MAJOR  && minor == TP_VERSION_MINOR)
1503 		return;
1504 
1505 	rev = t3rev2char(adap);
1506 	snprintf(name, sizeof(name), TPEEPROM_NAME, rev);
1507 
1508 	tpeeprom = firmware_get(name);
1509 	if (tpeeprom == NULL) {
1510 		device_printf(adap->dev,
1511 			      "could not load TP EEPROM: unable to load %s\n",
1512 			      name);
1513 		return;
1514 	}
1515 
1516 	len = tpeeprom->datasize - 4;
1517 
1518 	ret = t3_check_tpsram(adap, tpeeprom->data, tpeeprom->datasize);
1519 	if (ret)
1520 		goto release_tpeeprom;
1521 
1522 	if (len != TP_SRAM_LEN) {
1523 		device_printf(adap->dev,
1524 			      "%s length is wrong len=%d expected=%d\n", name,
1525 			      len, TP_SRAM_LEN);
1526 		return;
1527 	}
1528 
1529 	ret = set_eeprom(&adap->port[0], tpeeprom->data, tpeeprom->datasize,
1530 	    TP_SRAM_OFFSET);
1531 
1532 	if (!ret) {
1533 		device_printf(adap->dev,
1534 			"Protocol SRAM image updated in EEPROM to %d.%d.%d\n",
1535 			 TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1536 	} else
1537 		device_printf(adap->dev,
1538 			      "Protocol SRAM image update in EEPROM failed\n");
1539 
1540 release_tpeeprom:
1541 	firmware_put(tpeeprom, FIRMWARE_UNLOAD);
1542 
1543 	return;
1544 }
1545 
1546 static int
1547 update_tpsram(struct adapter *adap)
1548 {
1549 	const struct firmware *tpsram;
1550 	int ret;
1551 	char rev, name[32];
1552 
1553 	rev = t3rev2char(adap);
1554 	snprintf(name, sizeof(name), TPSRAM_NAME, rev);
1555 
1556 	update_tpeeprom(adap);
1557 
1558 	tpsram = firmware_get(name);
1559 	if (tpsram == NULL){
1560 		device_printf(adap->dev, "could not load TP SRAM\n");
1561 		return (EINVAL);
1562 	} else
1563 		device_printf(adap->dev, "updating TP SRAM\n");
1564 
1565 	ret = t3_check_tpsram(adap, tpsram->data, tpsram->datasize);
1566 	if (ret)
1567 		goto release_tpsram;
1568 
1569 	ret = t3_set_proto_sram(adap, tpsram->data);
1570 	if (ret)
1571 		device_printf(adap->dev, "loading protocol SRAM failed\n");
1572 
1573 release_tpsram:
1574 	firmware_put(tpsram, FIRMWARE_UNLOAD);
1575 
1576 	return ret;
1577 }
1578 
1579 /**
1580  *	cxgb_up - enable the adapter
1581  *	@adap: adapter being enabled
1582  *
1583  *	Called when the first port is enabled, this function performs the
1584  *	actions necessary to make an adapter operational, such as completing
1585  *	the initialization of HW modules, and enabling interrupts.
1586  */
1587 static int
1588 cxgb_up(struct adapter *sc)
1589 {
1590 	int err = 0;
1591 	unsigned int mxf = t3_mc5_size(&sc->mc5) - MC5_MIN_TIDS;
1592 
1593 	KASSERT(sc->open_device_map == 0, ("%s: device(s) already open (%x)",
1594 					   __func__, sc->open_device_map));
1595 
1596 	if ((sc->flags & FULL_INIT_DONE) == 0) {
1597 
1598 		ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1599 
1600 		if ((sc->flags & FW_UPTODATE) == 0)
1601 			if ((err = upgrade_fw(sc)))
1602 				goto out;
1603 
1604 		if ((sc->flags & TPS_UPTODATE) == 0)
1605 			if ((err = update_tpsram(sc)))
1606 				goto out;
1607 
1608 		if (is_offload(sc) && nfilters != 0) {
1609 			sc->params.mc5.nservers = 0;
1610 
1611 			if (nfilters < 0)
1612 				sc->params.mc5.nfilters = mxf;
1613 			else
1614 				sc->params.mc5.nfilters = min(nfilters, mxf);
1615 		}
1616 
1617 		err = t3_init_hw(sc, 0);
1618 		if (err)
1619 			goto out;
1620 
1621 		t3_set_reg_field(sc, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1622 		t3_write_reg(sc, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1623 
1624 		err = setup_sge_qsets(sc);
1625 		if (err)
1626 			goto out;
1627 
1628 		alloc_filters(sc);
1629 		setup_rss(sc);
1630 
1631 		t3_add_configured_sysctls(sc);
1632 		sc->flags |= FULL_INIT_DONE;
1633 	}
1634 
1635 	t3_intr_clear(sc);
1636 	t3_sge_start(sc);
1637 	t3_intr_enable(sc);
1638 
1639 	if (sc->params.rev >= T3_REV_C && !(sc->flags & TP_PARITY_INIT) &&
1640 	    is_offload(sc) && init_tp_parity(sc) == 0)
1641 		sc->flags |= TP_PARITY_INIT;
1642 
1643 	if (sc->flags & TP_PARITY_INIT) {
1644 		t3_write_reg(sc, A_TP_INT_CAUSE, F_CMCACHEPERR | F_ARPLUTPERR);
1645 		t3_write_reg(sc, A_TP_INT_ENABLE, 0x7fbfffff);
1646 	}
1647 
1648 	if (!(sc->flags & QUEUES_BOUND)) {
1649 		bind_qsets(sc);
1650 		setup_hw_filters(sc);
1651 		sc->flags |= QUEUES_BOUND;
1652 	}
1653 
1654 	t3_sge_reset_adapter(sc);
1655 out:
1656 	return (err);
1657 }
1658 
1659 /*
1660  * Called when the last open device is closed.  Does NOT undo all of cxgb_up's
1661  * work.  Specifically, the resources grabbed under FULL_INIT_DONE are released
1662  * during controller_detach, not here.
1663  */
1664 static void
1665 cxgb_down(struct adapter *sc)
1666 {
1667 	t3_sge_stop(sc);
1668 	t3_intr_disable(sc);
1669 }
1670 
1671 /*
1672  * if_init for cxgb ports.
1673  */
1674 static void
1675 cxgb_init(void *arg)
1676 {
1677 	struct port_info *p = arg;
1678 	struct adapter *sc = p->adapter;
1679 
1680 	ADAPTER_LOCK(sc);
1681 	cxgb_init_locked(p); /* releases adapter lock */
1682 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1683 }
1684 
1685 static int
1686 cxgb_init_locked(struct port_info *p)
1687 {
1688 	struct adapter *sc = p->adapter;
1689 	if_t ifp = p->ifp;
1690 	struct cmac *mac = &p->mac;
1691 	int i, rc = 0, may_sleep = 0, gave_up_lock = 0;
1692 
1693 	ADAPTER_LOCK_ASSERT_OWNED(sc);
1694 
1695 	while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1696 		gave_up_lock = 1;
1697 		if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbinit", 0)) {
1698 			rc = EINTR;
1699 			goto done;
1700 		}
1701 	}
1702 	if (IS_DOOMED(p)) {
1703 		rc = ENXIO;
1704 		goto done;
1705 	}
1706 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1707 
1708 	/*
1709 	 * The code that runs during one-time adapter initialization can sleep
1710 	 * so it's important not to hold any locks across it.
1711 	 */
1712 	may_sleep = sc->flags & FULL_INIT_DONE ? 0 : 1;
1713 
1714 	if (may_sleep) {
1715 		SET_BUSY(sc);
1716 		gave_up_lock = 1;
1717 		ADAPTER_UNLOCK(sc);
1718 	}
1719 
1720 	if (sc->open_device_map == 0 && ((rc = cxgb_up(sc)) != 0))
1721 			goto done;
1722 
1723 	PORT_LOCK(p);
1724 	if (isset(&sc->open_device_map, p->port_id) &&
1725 	    (if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
1726 		PORT_UNLOCK(p);
1727 		goto done;
1728 	}
1729 	t3_port_intr_enable(sc, p->port_id);
1730 	if (!mac->multiport)
1731 		t3_mac_init(mac);
1732 	cxgb_update_mac_settings(p);
1733 	t3_link_start(&p->phy, mac, &p->link_config);
1734 	t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
1735 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
1736 	PORT_UNLOCK(p);
1737 
1738 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1739 		struct sge_qset *qs = &sc->sge.qs[i];
1740 		struct sge_txq *txq = &qs->txq[TXQ_ETH];
1741 
1742 		callout_reset_on(&txq->txq_watchdog, hz, cxgb_tx_watchdog, qs,
1743 				 txq->txq_watchdog.c_cpu);
1744 	}
1745 
1746 	/* all ok */
1747 	setbit(&sc->open_device_map, p->port_id);
1748 	callout_reset(&p->link_check_ch,
1749 	    p->phy.caps & SUPPORTED_LINK_IRQ ?  hz * 3 : hz / 4,
1750 	    link_check_callout, p);
1751 
1752 done:
1753 	if (may_sleep) {
1754 		ADAPTER_LOCK(sc);
1755 		KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1756 		CLR_BUSY(sc);
1757 	}
1758 	if (gave_up_lock)
1759 		wakeup_one(&sc->flags);
1760 	ADAPTER_UNLOCK(sc);
1761 	return (rc);
1762 }
1763 
1764 static int
1765 cxgb_uninit_locked(struct port_info *p)
1766 {
1767 	struct adapter *sc = p->adapter;
1768 	int rc;
1769 
1770 	ADAPTER_LOCK_ASSERT_OWNED(sc);
1771 
1772 	while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1773 		if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbunin", 0)) {
1774 			rc = EINTR;
1775 			goto done;
1776 		}
1777 	}
1778 	if (IS_DOOMED(p)) {
1779 		rc = ENXIO;
1780 		goto done;
1781 	}
1782 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1783 	SET_BUSY(sc);
1784 	ADAPTER_UNLOCK(sc);
1785 
1786 	rc = cxgb_uninit_synchronized(p);
1787 
1788 	ADAPTER_LOCK(sc);
1789 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1790 	CLR_BUSY(sc);
1791 	wakeup_one(&sc->flags);
1792 done:
1793 	ADAPTER_UNLOCK(sc);
1794 	return (rc);
1795 }
1796 
1797 /*
1798  * Called on "ifconfig down", and from port_detach
1799  */
1800 static int
1801 cxgb_uninit_synchronized(struct port_info *pi)
1802 {
1803 	struct adapter *sc = pi->adapter;
1804 	if_t ifp = pi->ifp;
1805 
1806 	/*
1807 	 * taskqueue_drain may cause a deadlock if the adapter lock is held.
1808 	 */
1809 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1810 
1811 	/*
1812 	 * Clear this port's bit from the open device map, and then drain all
1813 	 * the tasks that can access/manipulate this port's port_info or ifp.
1814 	 * We disable this port's interrupts here and so the slow/ext
1815 	 * interrupt tasks won't be enqueued.  The tick task will continue to
1816 	 * be enqueued every second but the runs after this drain will not see
1817 	 * this port in the open device map.
1818 	 *
1819 	 * A well behaved task must take open_device_map into account and ignore
1820 	 * ports that are not open.
1821 	 */
1822 	clrbit(&sc->open_device_map, pi->port_id);
1823 	t3_port_intr_disable(sc, pi->port_id);
1824 	taskqueue_drain(sc->tq, &sc->slow_intr_task);
1825 	taskqueue_drain(sc->tq, &sc->tick_task);
1826 
1827 	callout_drain(&pi->link_check_ch);
1828 	taskqueue_drain(sc->tq, &pi->link_check_task);
1829 
1830 	PORT_LOCK(pi);
1831 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1832 
1833 	/* disable pause frames */
1834 	t3_set_reg_field(sc, A_XGM_TX_CFG + pi->mac.offset, F_TXPAUSEEN, 0);
1835 
1836 	/* Reset RX FIFO HWM */
1837 	t3_set_reg_field(sc, A_XGM_RXFIFO_CFG +  pi->mac.offset,
1838 			 V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM), 0);
1839 
1840 	DELAY(100 * 1000);
1841 
1842 	/* Wait for TXFIFO empty */
1843 	t3_wait_op_done(sc, A_XGM_TXFIFO_CFG + pi->mac.offset,
1844 			F_TXFIFO_EMPTY, 1, 20, 5);
1845 
1846 	DELAY(100 * 1000);
1847 	t3_mac_disable(&pi->mac, MAC_DIRECTION_RX);
1848 
1849 	pi->phy.ops->power_down(&pi->phy, 1);
1850 
1851 	PORT_UNLOCK(pi);
1852 
1853 	pi->link_config.link_ok = 0;
1854 	t3_os_link_changed(sc, pi->port_id, 0, 0, 0, 0, 0);
1855 
1856 	if (sc->open_device_map == 0)
1857 		cxgb_down(pi->adapter);
1858 
1859 	return (0);
1860 }
1861 
1862 /*
1863  * Mark lro enabled or disabled in all qsets for this port
1864  */
1865 static int
1866 cxgb_set_lro(struct port_info *p, int enabled)
1867 {
1868 	int i;
1869 	struct adapter *adp = p->adapter;
1870 	struct sge_qset *q;
1871 
1872 	for (i = 0; i < p->nqsets; i++) {
1873 		q = &adp->sge.qs[p->first_qset + i];
1874 		q->lro.enabled = (enabled != 0);
1875 	}
1876 	return (0);
1877 }
1878 
1879 static int
1880 cxgb_ioctl(if_t ifp, unsigned long command, caddr_t data)
1881 {
1882 	struct port_info *p = if_getsoftc(ifp);
1883 	struct adapter *sc = p->adapter;
1884 	struct ifreq *ifr = (struct ifreq *)data;
1885 	int flags, error = 0, mtu;
1886 	uint32_t mask;
1887 
1888 	switch (command) {
1889 	case SIOCSIFMTU:
1890 		ADAPTER_LOCK(sc);
1891 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1892 		if (error) {
1893 fail:
1894 			ADAPTER_UNLOCK(sc);
1895 			return (error);
1896 		}
1897 
1898 		mtu = ifr->ifr_mtu;
1899 		if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO)) {
1900 			error = EINVAL;
1901 		} else {
1902 			if_setmtu(ifp, mtu);
1903 			PORT_LOCK(p);
1904 			cxgb_update_mac_settings(p);
1905 			PORT_UNLOCK(p);
1906 		}
1907 		ADAPTER_UNLOCK(sc);
1908 		break;
1909 	case SIOCSIFFLAGS:
1910 		ADAPTER_LOCK(sc);
1911 		if (IS_DOOMED(p)) {
1912 			error = ENXIO;
1913 			goto fail;
1914 		}
1915 		if (if_getflags(ifp) & IFF_UP) {
1916 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1917 				flags = p->if_flags;
1918 				if (((if_getflags(ifp) ^ flags) & IFF_PROMISC) ||
1919 				    ((if_getflags(ifp) ^ flags) & IFF_ALLMULTI)) {
1920 					if (IS_BUSY(sc)) {
1921 						error = EBUSY;
1922 						goto fail;
1923 					}
1924 					PORT_LOCK(p);
1925 					cxgb_update_mac_settings(p);
1926 					PORT_UNLOCK(p);
1927 				}
1928 				ADAPTER_UNLOCK(sc);
1929 			} else
1930 				error = cxgb_init_locked(p);
1931 			p->if_flags = if_getflags(ifp);
1932 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1933 			error = cxgb_uninit_locked(p);
1934 		else
1935 			ADAPTER_UNLOCK(sc);
1936 
1937 		ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1938 		break;
1939 	case SIOCADDMULTI:
1940 	case SIOCDELMULTI:
1941 		ADAPTER_LOCK(sc);
1942 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1943 		if (error)
1944 			goto fail;
1945 
1946 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
1947 			PORT_LOCK(p);
1948 			cxgb_update_mac_settings(p);
1949 			PORT_UNLOCK(p);
1950 		}
1951 		ADAPTER_UNLOCK(sc);
1952 
1953 		break;
1954 	case SIOCSIFCAP:
1955 		ADAPTER_LOCK(sc);
1956 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1957 		if (error)
1958 			goto fail;
1959 
1960 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
1961 		if (mask & IFCAP_TXCSUM) {
1962 			if_togglecapenable(ifp, IFCAP_TXCSUM);
1963 			if_togglehwassist(ifp, CSUM_TCP | CSUM_UDP | CSUM_IP);
1964 
1965 			if (IFCAP_TSO4 & if_getcapenable(ifp) &&
1966 			    !(IFCAP_TXCSUM & if_getcapenable(ifp))) {
1967 				mask &= ~IFCAP_TSO4;
1968 				if_setcapenablebit(ifp, 0, IFCAP_TSO4);
1969 				if_printf(ifp,
1970 				    "tso4 disabled due to -txcsum.\n");
1971 			}
1972 		}
1973 		if (mask & IFCAP_TXCSUM_IPV6) {
1974 			if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
1975 			if_togglehwassist(ifp, CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1976 
1977 			if (IFCAP_TSO6 & if_getcapenable(ifp) &&
1978 			    !(IFCAP_TXCSUM_IPV6 & if_getcapenable(ifp))) {
1979 				mask &= ~IFCAP_TSO6;
1980 				if_setcapenablebit(ifp, 0, IFCAP_TSO6);
1981 				if_printf(ifp,
1982 				    "tso6 disabled due to -txcsum6.\n");
1983 			}
1984 		}
1985 		if (mask & IFCAP_RXCSUM)
1986 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1987 		if (mask & IFCAP_RXCSUM_IPV6)
1988 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
1989 
1990 		/*
1991 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1992 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1993 		 * sending a TSO request our way, so it's sufficient to toggle
1994 		 * IFCAP_TSOx only.
1995 		 */
1996 		if (mask & IFCAP_TSO4) {
1997 			if (!(IFCAP_TSO4 & if_getcapenable(ifp)) &&
1998 			    !(IFCAP_TXCSUM & if_getcapenable(ifp))) {
1999 				if_printf(ifp, "enable txcsum first.\n");
2000 				error = EAGAIN;
2001 				goto fail;
2002 			}
2003 			if_togglecapenable(ifp, IFCAP_TSO4);
2004 		}
2005 		if (mask & IFCAP_TSO6) {
2006 			if (!(IFCAP_TSO6 & if_getcapenable(ifp)) &&
2007 			    !(IFCAP_TXCSUM_IPV6 & if_getcapenable(ifp))) {
2008 				if_printf(ifp, "enable txcsum6 first.\n");
2009 				error = EAGAIN;
2010 				goto fail;
2011 			}
2012 			if_togglecapenable(ifp, IFCAP_TSO6);
2013 		}
2014 		if (mask & IFCAP_LRO) {
2015 			if_togglecapenable(ifp, IFCAP_LRO);
2016 
2017 			/* Safe to do this even if cxgb_up not called yet */
2018 			cxgb_set_lro(p, if_getcapenable(ifp) & IFCAP_LRO);
2019 		}
2020 #ifdef TCP_OFFLOAD
2021 		if (mask & IFCAP_TOE4) {
2022 			int enable = (if_getcapenable(ifp) ^ mask) & IFCAP_TOE4;
2023 
2024 			error = toe_capability(p, enable);
2025 			if (error == 0)
2026 				if_togglecapenable(ifp, mask);
2027 		}
2028 #endif
2029 		if (mask & IFCAP_VLAN_HWTAGGING) {
2030 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
2031 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
2032 				PORT_LOCK(p);
2033 				cxgb_update_mac_settings(p);
2034 				PORT_UNLOCK(p);
2035 			}
2036 		}
2037 		if (mask & IFCAP_VLAN_MTU) {
2038 			if_togglecapenable(ifp, IFCAP_VLAN_MTU);
2039 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
2040 				PORT_LOCK(p);
2041 				cxgb_update_mac_settings(p);
2042 				PORT_UNLOCK(p);
2043 			}
2044 		}
2045 		if (mask & IFCAP_VLAN_HWTSO)
2046 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
2047 		if (mask & IFCAP_VLAN_HWCSUM)
2048 			if_togglecapenable(ifp, IFCAP_VLAN_HWCSUM);
2049 
2050 #ifdef VLAN_CAPABILITIES
2051 		VLAN_CAPABILITIES(ifp);
2052 #endif
2053 		ADAPTER_UNLOCK(sc);
2054 		break;
2055 	case SIOCSIFMEDIA:
2056 	case SIOCGIFMEDIA:
2057 		error = ifmedia_ioctl(ifp, ifr, &p->media, command);
2058 		break;
2059 	default:
2060 		error = ether_ioctl(ifp, command, data);
2061 	}
2062 
2063 	return (error);
2064 }
2065 
2066 static int
2067 cxgb_media_change(if_t ifp)
2068 {
2069 	return (EOPNOTSUPP);
2070 }
2071 
2072 /*
2073  * Translates phy->modtype to the correct Ethernet media subtype.
2074  */
2075 static int
2076 cxgb_ifm_type(int mod)
2077 {
2078 	switch (mod) {
2079 	case phy_modtype_sr:
2080 		return (IFM_10G_SR);
2081 	case phy_modtype_lr:
2082 		return (IFM_10G_LR);
2083 	case phy_modtype_lrm:
2084 		return (IFM_10G_LRM);
2085 	case phy_modtype_twinax:
2086 		return (IFM_10G_TWINAX);
2087 	case phy_modtype_twinax_long:
2088 		return (IFM_10G_TWINAX_LONG);
2089 	case phy_modtype_none:
2090 		return (IFM_NONE);
2091 	case phy_modtype_unknown:
2092 		return (IFM_UNKNOWN);
2093 	}
2094 
2095 	KASSERT(0, ("%s: modtype %d unknown", __func__, mod));
2096 	return (IFM_UNKNOWN);
2097 }
2098 
2099 /*
2100  * Rebuilds the ifmedia list for this port, and sets the current media.
2101  */
2102 static void
2103 cxgb_build_medialist(struct port_info *p)
2104 {
2105 	struct cphy *phy = &p->phy;
2106 	struct ifmedia *media = &p->media;
2107 	int mod = phy->modtype;
2108 	int m = IFM_ETHER | IFM_FDX;
2109 
2110 	PORT_LOCK(p);
2111 
2112 	ifmedia_removeall(media);
2113 	if (phy->caps & SUPPORTED_TP && phy->caps & SUPPORTED_Autoneg) {
2114 		/* Copper (RJ45) */
2115 
2116 		if (phy->caps & SUPPORTED_10000baseT_Full)
2117 			ifmedia_add(media, m | IFM_10G_T, mod, NULL);
2118 
2119 		if (phy->caps & SUPPORTED_1000baseT_Full)
2120 			ifmedia_add(media, m | IFM_1000_T, mod, NULL);
2121 
2122 		if (phy->caps & SUPPORTED_100baseT_Full)
2123 			ifmedia_add(media, m | IFM_100_TX, mod, NULL);
2124 
2125 		if (phy->caps & SUPPORTED_10baseT_Full)
2126 			ifmedia_add(media, m | IFM_10_T, mod, NULL);
2127 
2128 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, mod, NULL);
2129 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
2130 
2131 	} else if (phy->caps & SUPPORTED_TP) {
2132 		/* Copper (CX4) */
2133 
2134 		KASSERT(phy->caps & SUPPORTED_10000baseT_Full,
2135 			("%s: unexpected cap 0x%x", __func__, phy->caps));
2136 
2137 		ifmedia_add(media, m | IFM_10G_CX4, mod, NULL);
2138 		ifmedia_set(media, m | IFM_10G_CX4);
2139 
2140 	} else if (phy->caps & SUPPORTED_FIBRE &&
2141 		   phy->caps & SUPPORTED_10000baseT_Full) {
2142 		/* 10G optical (but includes SFP+ twinax) */
2143 
2144 		m |= cxgb_ifm_type(mod);
2145 		if (IFM_SUBTYPE(m) == IFM_NONE)
2146 			m &= ~IFM_FDX;
2147 
2148 		ifmedia_add(media, m, mod, NULL);
2149 		ifmedia_set(media, m);
2150 
2151 	} else if (phy->caps & SUPPORTED_FIBRE &&
2152 		   phy->caps & SUPPORTED_1000baseT_Full) {
2153 		/* 1G optical */
2154 
2155 		/* XXX: Lie and claim to be SX, could actually be any 1G-X */
2156 		ifmedia_add(media, m | IFM_1000_SX, mod, NULL);
2157 		ifmedia_set(media, m | IFM_1000_SX);
2158 
2159 	} else {
2160 		KASSERT(0, ("%s: don't know how to handle 0x%x.", __func__,
2161 			    phy->caps));
2162 	}
2163 
2164 	PORT_UNLOCK(p);
2165 }
2166 
2167 static void
2168 cxgb_media_status(if_t ifp, struct ifmediareq *ifmr)
2169 {
2170 	struct port_info *p = if_getsoftc(ifp);
2171 	struct ifmedia_entry *cur = p->media.ifm_cur;
2172 	int speed = p->link_config.speed;
2173 
2174 	if (cur->ifm_data != p->phy.modtype) {
2175 		cxgb_build_medialist(p);
2176 		cur = p->media.ifm_cur;
2177 	}
2178 
2179 	ifmr->ifm_status = IFM_AVALID;
2180 	if (!p->link_config.link_ok)
2181 		return;
2182 
2183 	ifmr->ifm_status |= IFM_ACTIVE;
2184 
2185 	/*
2186 	 * active and current will differ iff current media is autoselect.  That
2187 	 * can happen only for copper RJ45.
2188 	 */
2189 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2190 		return;
2191 	KASSERT(p->phy.caps & SUPPORTED_TP && p->phy.caps & SUPPORTED_Autoneg,
2192 		("%s: unexpected PHY caps 0x%x", __func__, p->phy.caps));
2193 
2194 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2195 	if (speed == SPEED_10000)
2196 		ifmr->ifm_active |= IFM_10G_T;
2197 	else if (speed == SPEED_1000)
2198 		ifmr->ifm_active |= IFM_1000_T;
2199 	else if (speed == SPEED_100)
2200 		ifmr->ifm_active |= IFM_100_TX;
2201 	else if (speed == SPEED_10)
2202 		ifmr->ifm_active |= IFM_10_T;
2203 	else
2204 		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
2205 			    speed));
2206 }
2207 
2208 static uint64_t
2209 cxgb_get_counter(if_t ifp, ift_counter c)
2210 {
2211 	struct port_info *pi = if_getsoftc(ifp);
2212 	struct adapter *sc = pi->adapter;
2213 	struct cmac *mac = &pi->mac;
2214 	struct mac_stats *mstats = &mac->stats;
2215 
2216 	cxgb_refresh_stats(pi);
2217 
2218 	switch (c) {
2219 	case IFCOUNTER_IPACKETS:
2220 		return (mstats->rx_frames);
2221 
2222 	case IFCOUNTER_IERRORS:
2223 		return (mstats->rx_jabber + mstats->rx_data_errs +
2224 		    mstats->rx_sequence_errs + mstats->rx_runt +
2225 		    mstats->rx_too_long + mstats->rx_mac_internal_errs +
2226 		    mstats->rx_short + mstats->rx_fcs_errs);
2227 
2228 	case IFCOUNTER_OPACKETS:
2229 		return (mstats->tx_frames);
2230 
2231 	case IFCOUNTER_OERRORS:
2232 		return (mstats->tx_excess_collisions + mstats->tx_underrun +
2233 		    mstats->tx_len_errs + mstats->tx_mac_internal_errs +
2234 		    mstats->tx_excess_deferral + mstats->tx_fcs_errs);
2235 
2236 	case IFCOUNTER_COLLISIONS:
2237 		return (mstats->tx_total_collisions);
2238 
2239 	case IFCOUNTER_IBYTES:
2240 		return (mstats->rx_octets);
2241 
2242 	case IFCOUNTER_OBYTES:
2243 		return (mstats->tx_octets);
2244 
2245 	case IFCOUNTER_IMCASTS:
2246 		return (mstats->rx_mcast_frames);
2247 
2248 	case IFCOUNTER_OMCASTS:
2249 		return (mstats->tx_mcast_frames);
2250 
2251 	case IFCOUNTER_IQDROPS:
2252 		return (mstats->rx_cong_drops);
2253 
2254 	case IFCOUNTER_OQDROPS: {
2255 		int i;
2256 		uint64_t drops;
2257 
2258 		drops = 0;
2259 		if (sc->flags & FULL_INIT_DONE) {
2260 			for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++)
2261 				drops += sc->sge.qs[i].txq[TXQ_ETH].txq_mr->br_drops;
2262 		}
2263 
2264 		return (drops);
2265 
2266 	}
2267 
2268 	default:
2269 		return (if_get_counter_default(ifp, c));
2270 	}
2271 }
2272 
2273 static void
2274 cxgb_async_intr(void *data)
2275 {
2276 	adapter_t *sc = data;
2277 
2278 	t3_write_reg(sc, A_PL_INT_ENABLE0, 0);
2279 	(void) t3_read_reg(sc, A_PL_INT_ENABLE0);
2280 	taskqueue_enqueue(sc->tq, &sc->slow_intr_task);
2281 }
2282 
2283 static void
2284 link_check_callout(void *arg)
2285 {
2286 	struct port_info *pi = arg;
2287 	struct adapter *sc = pi->adapter;
2288 
2289 	if (!isset(&sc->open_device_map, pi->port_id))
2290 		return;
2291 
2292 	taskqueue_enqueue(sc->tq, &pi->link_check_task);
2293 }
2294 
2295 static void
2296 check_link_status(void *arg, int pending)
2297 {
2298 	struct port_info *pi = arg;
2299 	struct adapter *sc = pi->adapter;
2300 
2301 	if (!isset(&sc->open_device_map, pi->port_id))
2302 		return;
2303 
2304 	t3_link_changed(sc, pi->port_id);
2305 
2306 	if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ) ||
2307 	    pi->link_config.link_ok == 0)
2308 		callout_reset(&pi->link_check_ch, hz, link_check_callout, pi);
2309 }
2310 
2311 void
2312 t3_os_link_intr(struct port_info *pi)
2313 {
2314 	/*
2315 	 * Schedule a link check in the near future.  If the link is flapping
2316 	 * rapidly we'll keep resetting the callout and delaying the check until
2317 	 * things stabilize a bit.
2318 	 */
2319 	callout_reset(&pi->link_check_ch, hz / 4, link_check_callout, pi);
2320 }
2321 
2322 static void
2323 check_t3b2_mac(struct adapter *sc)
2324 {
2325 	int i;
2326 
2327 	if (sc->flags & CXGB_SHUTDOWN)
2328 		return;
2329 
2330 	for_each_port(sc, i) {
2331 		struct port_info *p = &sc->port[i];
2332 		int status;
2333 #ifdef INVARIANTS
2334 		if_t ifp = p->ifp;
2335 #endif
2336 
2337 		if (!isset(&sc->open_device_map, p->port_id) || p->link_fault ||
2338 		    !p->link_config.link_ok)
2339 			continue;
2340 
2341 		KASSERT(if_getdrvflags(ifp) & IFF_DRV_RUNNING,
2342 			("%s: state mismatch (drv_flags %x, device_map %x)",
2343 			 __func__, if_getdrvflags(ifp), sc->open_device_map));
2344 
2345 		PORT_LOCK(p);
2346 		status = t3b2_mac_watchdog_task(&p->mac);
2347 		if (status == 1)
2348 			p->mac.stats.num_toggled++;
2349 		else if (status == 2) {
2350 			struct cmac *mac = &p->mac;
2351 
2352 			cxgb_update_mac_settings(p);
2353 			t3_link_start(&p->phy, mac, &p->link_config);
2354 			t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2355 			t3_port_intr_enable(sc, p->port_id);
2356 			p->mac.stats.num_resets++;
2357 		}
2358 		PORT_UNLOCK(p);
2359 	}
2360 }
2361 
2362 static void
2363 cxgb_tick(void *arg)
2364 {
2365 	adapter_t *sc = (adapter_t *)arg;
2366 
2367 	if (sc->flags & CXGB_SHUTDOWN)
2368 		return;
2369 
2370 	taskqueue_enqueue(sc->tq, &sc->tick_task);
2371 	callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
2372 }
2373 
2374 void
2375 cxgb_refresh_stats(struct port_info *pi)
2376 {
2377 	struct timeval tv;
2378 	const struct timeval interval = {0, 250000};    /* 250ms */
2379 
2380 	getmicrotime(&tv);
2381 	timevalsub(&tv, &interval);
2382 	if (timevalcmp(&tv, &pi->last_refreshed, <))
2383 		return;
2384 
2385 	PORT_LOCK(pi);
2386 	t3_mac_update_stats(&pi->mac);
2387 	PORT_UNLOCK(pi);
2388 	getmicrotime(&pi->last_refreshed);
2389 }
2390 
2391 static void
2392 cxgb_tick_handler(void *arg, int count)
2393 {
2394 	adapter_t *sc = (adapter_t *)arg;
2395 	const struct adapter_params *p = &sc->params;
2396 	int i;
2397 	uint32_t cause, reset;
2398 
2399 	if (sc->flags & CXGB_SHUTDOWN || !(sc->flags & FULL_INIT_DONE))
2400 		return;
2401 
2402 	if (p->rev == T3_REV_B2 && p->nports < 4 && sc->open_device_map)
2403 		check_t3b2_mac(sc);
2404 
2405 	cause = t3_read_reg(sc, A_SG_INT_CAUSE) & (F_RSPQSTARVE | F_FLEMPTY);
2406 	if (cause) {
2407 		struct sge_qset *qs = &sc->sge.qs[0];
2408 		uint32_t mask, v;
2409 
2410 		v = t3_read_reg(sc, A_SG_RSPQ_FL_STATUS) & ~0xff00;
2411 
2412 		mask = 1;
2413 		for (i = 0; i < SGE_QSETS; i++) {
2414 			if (v & mask)
2415 				qs[i].rspq.starved++;
2416 			mask <<= 1;
2417 		}
2418 
2419 		mask <<= SGE_QSETS; /* skip RSPQXDISABLED */
2420 
2421 		for (i = 0; i < SGE_QSETS * 2; i++) {
2422 			if (v & mask) {
2423 				qs[i / 2].fl[i % 2].empty++;
2424 			}
2425 			mask <<= 1;
2426 		}
2427 
2428 		/* clear */
2429 		t3_write_reg(sc, A_SG_RSPQ_FL_STATUS, v);
2430 		t3_write_reg(sc, A_SG_INT_CAUSE, cause);
2431 	}
2432 
2433 	for (i = 0; i < sc->params.nports; i++) {
2434 		struct port_info *pi = &sc->port[i];
2435 		struct cmac *mac = &pi->mac;
2436 
2437 		if (!isset(&sc->open_device_map, pi->port_id))
2438 			continue;
2439 
2440 		cxgb_refresh_stats(pi);
2441 
2442 		if (mac->multiport)
2443 			continue;
2444 
2445 		/* Count rx fifo overflows, once per second */
2446 		cause = t3_read_reg(sc, A_XGM_INT_CAUSE + mac->offset);
2447 		reset = 0;
2448 		if (cause & F_RXFIFO_OVERFLOW) {
2449 			mac->stats.rx_fifo_ovfl++;
2450 			reset |= F_RXFIFO_OVERFLOW;
2451 		}
2452 		t3_write_reg(sc, A_XGM_INT_CAUSE + mac->offset, reset);
2453 	}
2454 }
2455 
2456 static void
2457 touch_bars(device_t dev)
2458 {
2459 	/*
2460 	 * Don't enable yet
2461 	 */
2462 #if !defined(__LP64__) && 0
2463 	u32 v;
2464 
2465 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_1, &v);
2466 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_1, v);
2467 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_3, &v);
2468 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_3, v);
2469 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_5, &v);
2470 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_5, v);
2471 #endif
2472 }
2473 
2474 static int
2475 set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset)
2476 {
2477 	uint8_t *buf;
2478 	int err = 0;
2479 	u32 aligned_offset, aligned_len, *p;
2480 	struct adapter *adapter = pi->adapter;
2481 
2482 
2483 	aligned_offset = offset & ~3;
2484 	aligned_len = (len + (offset & 3) + 3) & ~3;
2485 
2486 	if (aligned_offset != offset || aligned_len != len) {
2487 		buf = malloc(aligned_len, M_DEVBUF, M_WAITOK|M_ZERO);
2488 		if (!buf)
2489 			return (ENOMEM);
2490 		err = t3_seeprom_read(adapter, aligned_offset, (u32 *)buf);
2491 		if (!err && aligned_len > 4)
2492 			err = t3_seeprom_read(adapter,
2493 					      aligned_offset + aligned_len - 4,
2494 					      (u32 *)&buf[aligned_len - 4]);
2495 		if (err)
2496 			goto out;
2497 		memcpy(buf + (offset & 3), data, len);
2498 	} else
2499 		buf = (uint8_t *)(uintptr_t)data;
2500 
2501 	err = t3_seeprom_wp(adapter, 0);
2502 	if (err)
2503 		goto out;
2504 
2505 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2506 		err = t3_seeprom_write(adapter, aligned_offset, *p);
2507 		aligned_offset += 4;
2508 	}
2509 
2510 	if (!err)
2511 		err = t3_seeprom_wp(adapter, 1);
2512 out:
2513 	if (buf != data)
2514 		free(buf, M_DEVBUF);
2515 	return err;
2516 }
2517 
2518 
2519 static int
2520 in_range(int val, int lo, int hi)
2521 {
2522 	return val < 0 || (val <= hi && val >= lo);
2523 }
2524 
2525 static int
2526 cxgb_extension_open(struct cdev *dev, int flags, int fmp, struct thread *td)
2527 {
2528        return (0);
2529 }
2530 
2531 static int
2532 cxgb_extension_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2533 {
2534        return (0);
2535 }
2536 
2537 static int
2538 cxgb_extension_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data,
2539     int fflag, struct thread *td)
2540 {
2541 	int mmd, error = 0;
2542 	struct port_info *pi = dev->si_drv1;
2543 	adapter_t *sc = pi->adapter;
2544 
2545 #ifdef PRIV_SUPPORTED
2546 	if (priv_check(td, PRIV_DRIVER)) {
2547 		if (cxgb_debug)
2548 			printf("user does not have access to privileged ioctls\n");
2549 		return (EPERM);
2550 	}
2551 #else
2552 	if (suser(td)) {
2553 		if (cxgb_debug)
2554 			printf("user does not have access to privileged ioctls\n");
2555 		return (EPERM);
2556 	}
2557 #endif
2558 
2559 	switch (cmd) {
2560 	case CHELSIO_GET_MIIREG: {
2561 		uint32_t val;
2562 		struct cphy *phy = &pi->phy;
2563 		struct ch_mii_data *mid = (struct ch_mii_data *)data;
2564 
2565 		if (!phy->mdio_read)
2566 			return (EOPNOTSUPP);
2567 		if (is_10G(sc)) {
2568 			mmd = mid->phy_id >> 8;
2569 			if (!mmd)
2570 				mmd = MDIO_DEV_PCS;
2571 			else if (mmd > MDIO_DEV_VEND2)
2572 				return (EINVAL);
2573 
2574 			error = phy->mdio_read(sc, mid->phy_id & 0x1f, mmd,
2575 					     mid->reg_num, &val);
2576 		} else
2577 		        error = phy->mdio_read(sc, mid->phy_id & 0x1f, 0,
2578 					     mid->reg_num & 0x1f, &val);
2579 		if (error == 0)
2580 			mid->val_out = val;
2581 		break;
2582 	}
2583 	case CHELSIO_SET_MIIREG: {
2584 		struct cphy *phy = &pi->phy;
2585 		struct ch_mii_data *mid = (struct ch_mii_data *)data;
2586 
2587 		if (!phy->mdio_write)
2588 			return (EOPNOTSUPP);
2589 		if (is_10G(sc)) {
2590 			mmd = mid->phy_id >> 8;
2591 			if (!mmd)
2592 				mmd = MDIO_DEV_PCS;
2593 			else if (mmd > MDIO_DEV_VEND2)
2594 				return (EINVAL);
2595 
2596 			error = phy->mdio_write(sc, mid->phy_id & 0x1f,
2597 					      mmd, mid->reg_num, mid->val_in);
2598 		} else
2599 			error = phy->mdio_write(sc, mid->phy_id & 0x1f, 0,
2600 					      mid->reg_num & 0x1f,
2601 					      mid->val_in);
2602 		break;
2603 	}
2604 	case CHELSIO_SETREG: {
2605 		struct ch_reg *edata = (struct ch_reg *)data;
2606 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2607 			return (EFAULT);
2608 		t3_write_reg(sc, edata->addr, edata->val);
2609 		break;
2610 	}
2611 	case CHELSIO_GETREG: {
2612 		struct ch_reg *edata = (struct ch_reg *)data;
2613 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2614 			return (EFAULT);
2615 		edata->val = t3_read_reg(sc, edata->addr);
2616 		break;
2617 	}
2618 	case CHELSIO_GET_SGE_CONTEXT: {
2619 		struct ch_cntxt *ecntxt = (struct ch_cntxt *)data;
2620 		mtx_lock_spin(&sc->sge.reg_lock);
2621 		switch (ecntxt->cntxt_type) {
2622 		case CNTXT_TYPE_EGRESS:
2623 			error = -t3_sge_read_ecntxt(sc, ecntxt->cntxt_id,
2624 			    ecntxt->data);
2625 			break;
2626 		case CNTXT_TYPE_FL:
2627 			error = -t3_sge_read_fl(sc, ecntxt->cntxt_id,
2628 			    ecntxt->data);
2629 			break;
2630 		case CNTXT_TYPE_RSP:
2631 			error = -t3_sge_read_rspq(sc, ecntxt->cntxt_id,
2632 			    ecntxt->data);
2633 			break;
2634 		case CNTXT_TYPE_CQ:
2635 			error = -t3_sge_read_cq(sc, ecntxt->cntxt_id,
2636 			    ecntxt->data);
2637 			break;
2638 		default:
2639 			error = EINVAL;
2640 			break;
2641 		}
2642 		mtx_unlock_spin(&sc->sge.reg_lock);
2643 		break;
2644 	}
2645 	case CHELSIO_GET_SGE_DESC: {
2646 		struct ch_desc *edesc = (struct ch_desc *)data;
2647 		int ret;
2648 		if (edesc->queue_num >= SGE_QSETS * 6)
2649 			return (EINVAL);
2650 		ret = t3_get_desc(&sc->sge.qs[edesc->queue_num / 6],
2651 		    edesc->queue_num % 6, edesc->idx, edesc->data);
2652 		if (ret < 0)
2653 			return (EINVAL);
2654 		edesc->size = ret;
2655 		break;
2656 	}
2657 	case CHELSIO_GET_QSET_PARAMS: {
2658 		struct qset_params *q;
2659 		struct ch_qset_params *t = (struct ch_qset_params *)data;
2660 		int q1 = pi->first_qset;
2661 		int nqsets = pi->nqsets;
2662 		int i;
2663 
2664 		if (t->qset_idx >= nqsets)
2665 			return EINVAL;
2666 
2667 		i = q1 + t->qset_idx;
2668 		q = &sc->params.sge.qset[i];
2669 		t->rspq_size   = q->rspq_size;
2670 		t->txq_size[0] = q->txq_size[0];
2671 		t->txq_size[1] = q->txq_size[1];
2672 		t->txq_size[2] = q->txq_size[2];
2673 		t->fl_size[0]  = q->fl_size;
2674 		t->fl_size[1]  = q->jumbo_size;
2675 		t->polling     = q->polling;
2676 		t->lro         = q->lro;
2677 		t->intr_lat    = q->coalesce_usecs;
2678 		t->cong_thres  = q->cong_thres;
2679 		t->qnum        = i;
2680 
2681 		if ((sc->flags & FULL_INIT_DONE) == 0)
2682 			t->vector = 0;
2683 		else if (sc->flags & USING_MSIX)
2684 			t->vector = rman_get_start(sc->msix_irq_res[i]);
2685 		else
2686 			t->vector = rman_get_start(sc->irq_res);
2687 
2688 		break;
2689 	}
2690 	case CHELSIO_GET_QSET_NUM: {
2691 		struct ch_reg *edata = (struct ch_reg *)data;
2692 		edata->val = pi->nqsets;
2693 		break;
2694 	}
2695 	case CHELSIO_LOAD_FW: {
2696 		uint8_t *fw_data;
2697 		uint32_t vers;
2698 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2699 
2700 		/*
2701 		 * You're allowed to load a firmware only before FULL_INIT_DONE
2702 		 *
2703 		 * FW_UPTODATE is also set so the rest of the initialization
2704 		 * will not overwrite what was loaded here.  This gives you the
2705 		 * flexibility to load any firmware (and maybe shoot yourself in
2706 		 * the foot).
2707 		 */
2708 
2709 		ADAPTER_LOCK(sc);
2710 		if (sc->open_device_map || sc->flags & FULL_INIT_DONE) {
2711 			ADAPTER_UNLOCK(sc);
2712 			return (EBUSY);
2713 		}
2714 
2715 		fw_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2716 		if (!fw_data)
2717 			error = ENOMEM;
2718 		else
2719 			error = copyin(t->buf, fw_data, t->len);
2720 
2721 		if (!error)
2722 			error = -t3_load_fw(sc, fw_data, t->len);
2723 
2724 		if (t3_get_fw_version(sc, &vers) == 0) {
2725 			snprintf(&sc->fw_version[0], sizeof(sc->fw_version),
2726 			    "%d.%d.%d", G_FW_VERSION_MAJOR(vers),
2727 			    G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers));
2728 		}
2729 
2730 		if (!error)
2731 			sc->flags |= FW_UPTODATE;
2732 
2733 		free(fw_data, M_DEVBUF);
2734 		ADAPTER_UNLOCK(sc);
2735 		break;
2736 	}
2737 	case CHELSIO_LOAD_BOOT: {
2738 		uint8_t *boot_data;
2739 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2740 
2741 		boot_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2742 		if (!boot_data)
2743 			return ENOMEM;
2744 
2745 		error = copyin(t->buf, boot_data, t->len);
2746 		if (!error)
2747 			error = -t3_load_boot(sc, boot_data, t->len);
2748 
2749 		free(boot_data, M_DEVBUF);
2750 		break;
2751 	}
2752 	case CHELSIO_GET_PM: {
2753 		struct ch_pm *m = (struct ch_pm *)data;
2754 		struct tp_params *p = &sc->params.tp;
2755 
2756 		if (!is_offload(sc))
2757 			return (EOPNOTSUPP);
2758 
2759 		m->tx_pg_sz = p->tx_pg_size;
2760 		m->tx_num_pg = p->tx_num_pgs;
2761 		m->rx_pg_sz  = p->rx_pg_size;
2762 		m->rx_num_pg = p->rx_num_pgs;
2763 		m->pm_total  = p->pmtx_size + p->chan_rx_size * p->nchan;
2764 
2765 		break;
2766 	}
2767 	case CHELSIO_SET_PM: {
2768 		struct ch_pm *m = (struct ch_pm *)data;
2769 		struct tp_params *p = &sc->params.tp;
2770 
2771 		if (!is_offload(sc))
2772 			return (EOPNOTSUPP);
2773 		if (sc->flags & FULL_INIT_DONE)
2774 			return (EBUSY);
2775 
2776 		if (!m->rx_pg_sz || (m->rx_pg_sz & (m->rx_pg_sz - 1)) ||
2777 		    !m->tx_pg_sz || (m->tx_pg_sz & (m->tx_pg_sz - 1)))
2778 			return (EINVAL);	/* not power of 2 */
2779 		if (!(m->rx_pg_sz & 0x14000))
2780 			return (EINVAL);	/* not 16KB or 64KB */
2781 		if (!(m->tx_pg_sz & 0x1554000))
2782 			return (EINVAL);
2783 		if (m->tx_num_pg == -1)
2784 			m->tx_num_pg = p->tx_num_pgs;
2785 		if (m->rx_num_pg == -1)
2786 			m->rx_num_pg = p->rx_num_pgs;
2787 		if (m->tx_num_pg % 24 || m->rx_num_pg % 24)
2788 			return (EINVAL);
2789 		if (m->rx_num_pg * m->rx_pg_sz > p->chan_rx_size ||
2790 		    m->tx_num_pg * m->tx_pg_sz > p->chan_tx_size)
2791 			return (EINVAL);
2792 
2793 		p->rx_pg_size = m->rx_pg_sz;
2794 		p->tx_pg_size = m->tx_pg_sz;
2795 		p->rx_num_pgs = m->rx_num_pg;
2796 		p->tx_num_pgs = m->tx_num_pg;
2797 		break;
2798 	}
2799 	case CHELSIO_SETMTUTAB: {
2800 		struct ch_mtus *m = (struct ch_mtus *)data;
2801 		int i;
2802 
2803 		if (!is_offload(sc))
2804 			return (EOPNOTSUPP);
2805 		if (offload_running(sc))
2806 			return (EBUSY);
2807 		if (m->nmtus != NMTUS)
2808 			return (EINVAL);
2809 		if (m->mtus[0] < 81)         /* accommodate SACK */
2810 			return (EINVAL);
2811 
2812 		/*
2813 		 * MTUs must be in ascending order
2814 		 */
2815 		for (i = 1; i < NMTUS; ++i)
2816 			if (m->mtus[i] < m->mtus[i - 1])
2817 				return (EINVAL);
2818 
2819 		memcpy(sc->params.mtus, m->mtus, sizeof(sc->params.mtus));
2820 		break;
2821 	}
2822 	case CHELSIO_GETMTUTAB: {
2823 		struct ch_mtus *m = (struct ch_mtus *)data;
2824 
2825 		if (!is_offload(sc))
2826 			return (EOPNOTSUPP);
2827 
2828 		memcpy(m->mtus, sc->params.mtus, sizeof(m->mtus));
2829 		m->nmtus = NMTUS;
2830 		break;
2831 	}
2832 	case CHELSIO_GET_MEM: {
2833 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2834 		struct mc7 *mem;
2835 		uint8_t *useraddr;
2836 		u64 buf[32];
2837 
2838 		/*
2839 		 * Use these to avoid modifying len/addr in the return
2840 		 * struct
2841 		 */
2842 		uint32_t len = t->len, addr = t->addr;
2843 
2844 		if (!is_offload(sc))
2845 			return (EOPNOTSUPP);
2846 		if (!(sc->flags & FULL_INIT_DONE))
2847 			return (EIO);         /* need the memory controllers */
2848 		if ((addr & 0x7) || (len & 0x7))
2849 			return (EINVAL);
2850 		if (t->mem_id == MEM_CM)
2851 			mem = &sc->cm;
2852 		else if (t->mem_id == MEM_PMRX)
2853 			mem = &sc->pmrx;
2854 		else if (t->mem_id == MEM_PMTX)
2855 			mem = &sc->pmtx;
2856 		else
2857 			return (EINVAL);
2858 
2859 		/*
2860 		 * Version scheme:
2861 		 * bits 0..9: chip version
2862 		 * bits 10..15: chip revision
2863 		 */
2864 		t->version = 3 | (sc->params.rev << 10);
2865 
2866 		/*
2867 		 * Read 256 bytes at a time as len can be large and we don't
2868 		 * want to use huge intermediate buffers.
2869 		 */
2870 		useraddr = (uint8_t *)t->buf;
2871 		while (len) {
2872 			unsigned int chunk = min(len, sizeof(buf));
2873 
2874 			error = t3_mc7_bd_read(mem, addr / 8, chunk / 8, buf);
2875 			if (error)
2876 				return (-error);
2877 			if (copyout(buf, useraddr, chunk))
2878 				return (EFAULT);
2879 			useraddr += chunk;
2880 			addr += chunk;
2881 			len -= chunk;
2882 		}
2883 		break;
2884 	}
2885 	case CHELSIO_READ_TCAM_WORD: {
2886 		struct ch_tcam_word *t = (struct ch_tcam_word *)data;
2887 
2888 		if (!is_offload(sc))
2889 			return (EOPNOTSUPP);
2890 		if (!(sc->flags & FULL_INIT_DONE))
2891 			return (EIO);         /* need MC5 */
2892 		return -t3_read_mc5_range(&sc->mc5, t->addr, 1, t->buf);
2893 		break;
2894 	}
2895 	case CHELSIO_SET_TRACE_FILTER: {
2896 		struct ch_trace *t = (struct ch_trace *)data;
2897 		const struct trace_params *tp;
2898 
2899 		tp = (const struct trace_params *)&t->sip;
2900 		if (t->config_tx)
2901 			t3_config_trace_filter(sc, tp, 0, t->invert_match,
2902 					       t->trace_tx);
2903 		if (t->config_rx)
2904 			t3_config_trace_filter(sc, tp, 1, t->invert_match,
2905 					       t->trace_rx);
2906 		break;
2907 	}
2908 	case CHELSIO_SET_PKTSCHED: {
2909 		struct ch_pktsched_params *p = (struct ch_pktsched_params *)data;
2910 		if (sc->open_device_map == 0)
2911 			return (EAGAIN);
2912 		send_pktsched_cmd(sc, p->sched, p->idx, p->min, p->max,
2913 		    p->binding);
2914 		break;
2915 	}
2916 	case CHELSIO_IFCONF_GETREGS: {
2917 		struct ch_ifconf_regs *regs = (struct ch_ifconf_regs *)data;
2918 		int reglen = cxgb_get_regs_len();
2919 		uint8_t *buf = malloc(reglen, M_DEVBUF, M_NOWAIT);
2920 		if (buf == NULL) {
2921 			return (ENOMEM);
2922 		}
2923 		if (regs->len > reglen)
2924 			regs->len = reglen;
2925 		else if (regs->len < reglen)
2926 			error = ENOBUFS;
2927 
2928 		if (!error) {
2929 			cxgb_get_regs(sc, regs, buf);
2930 			error = copyout(buf, regs->data, reglen);
2931 		}
2932 		free(buf, M_DEVBUF);
2933 
2934 		break;
2935 	}
2936 	case CHELSIO_SET_HW_SCHED: {
2937 		struct ch_hw_sched *t = (struct ch_hw_sched *)data;
2938 		unsigned int ticks_per_usec = core_ticks_per_usec(sc);
2939 
2940 		if ((sc->flags & FULL_INIT_DONE) == 0)
2941 			return (EAGAIN);       /* need TP to be initialized */
2942 		if (t->sched >= NTX_SCHED || !in_range(t->mode, 0, 1) ||
2943 		    !in_range(t->channel, 0, 1) ||
2944 		    !in_range(t->kbps, 0, 10000000) ||
2945 		    !in_range(t->class_ipg, 0, 10000 * 65535 / ticks_per_usec) ||
2946 		    !in_range(t->flow_ipg, 0,
2947 			      dack_ticks_to_usec(sc, 0x7ff)))
2948 			return (EINVAL);
2949 
2950 		if (t->kbps >= 0) {
2951 			error = t3_config_sched(sc, t->kbps, t->sched);
2952 			if (error < 0)
2953 				return (-error);
2954 		}
2955 		if (t->class_ipg >= 0)
2956 			t3_set_sched_ipg(sc, t->sched, t->class_ipg);
2957 		if (t->flow_ipg >= 0) {
2958 			t->flow_ipg *= 1000;     /* us -> ns */
2959 			t3_set_pace_tbl(sc, &t->flow_ipg, t->sched, 1);
2960 		}
2961 		if (t->mode >= 0) {
2962 			int bit = 1 << (S_TX_MOD_TIMER_MODE + t->sched);
2963 
2964 			t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2965 					 bit, t->mode ? bit : 0);
2966 		}
2967 		if (t->channel >= 0)
2968 			t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2969 					 1 << t->sched, t->channel << t->sched);
2970 		break;
2971 	}
2972 	case CHELSIO_GET_EEPROM: {
2973 		int i;
2974 		struct ch_eeprom *e = (struct ch_eeprom *)data;
2975 		uint8_t *buf;
2976 
2977 		if (e->offset & 3 || e->offset >= EEPROMSIZE ||
2978 		    e->len > EEPROMSIZE || e->offset + e->len > EEPROMSIZE) {
2979 			return (EINVAL);
2980 		}
2981 
2982 		buf = malloc(EEPROMSIZE, M_DEVBUF, M_NOWAIT);
2983 		if (buf == NULL) {
2984 			return (ENOMEM);
2985 		}
2986 		e->magic = EEPROM_MAGIC;
2987 		for (i = e->offset & ~3; !error && i < e->offset + e->len; i += 4)
2988 			error = -t3_seeprom_read(sc, i, (uint32_t *)&buf[i]);
2989 
2990 		if (!error)
2991 			error = copyout(buf + e->offset, e->data, e->len);
2992 
2993 		free(buf, M_DEVBUF);
2994 		break;
2995 	}
2996 	case CHELSIO_CLEAR_STATS: {
2997 		if (!(sc->flags & FULL_INIT_DONE))
2998 			return EAGAIN;
2999 
3000 		PORT_LOCK(pi);
3001 		t3_mac_update_stats(&pi->mac);
3002 		memset(&pi->mac.stats, 0, sizeof(pi->mac.stats));
3003 		PORT_UNLOCK(pi);
3004 		break;
3005 	}
3006 	case CHELSIO_GET_UP_LA: {
3007 		struct ch_up_la *la = (struct ch_up_la *)data;
3008 		uint8_t *buf = malloc(LA_BUFSIZE, M_DEVBUF, M_NOWAIT);
3009 		if (buf == NULL) {
3010 			return (ENOMEM);
3011 		}
3012 		if (la->bufsize < LA_BUFSIZE)
3013 			error = ENOBUFS;
3014 
3015 		if (!error)
3016 			error = -t3_get_up_la(sc, &la->stopped, &la->idx,
3017 					      &la->bufsize, buf);
3018 		if (!error)
3019 			error = copyout(buf, la->data, la->bufsize);
3020 
3021 		free(buf, M_DEVBUF);
3022 		break;
3023 	}
3024 	case CHELSIO_GET_UP_IOQS: {
3025 		struct ch_up_ioqs *ioqs = (struct ch_up_ioqs *)data;
3026 		uint8_t *buf = malloc(IOQS_BUFSIZE, M_DEVBUF, M_NOWAIT);
3027 		uint32_t *v;
3028 
3029 		if (buf == NULL) {
3030 			return (ENOMEM);
3031 		}
3032 		if (ioqs->bufsize < IOQS_BUFSIZE)
3033 			error = ENOBUFS;
3034 
3035 		if (!error)
3036 			error = -t3_get_up_ioqs(sc, &ioqs->bufsize, buf);
3037 
3038 		if (!error) {
3039 			v = (uint32_t *)buf;
3040 
3041 			ioqs->ioq_rx_enable = *v++;
3042 			ioqs->ioq_tx_enable = *v++;
3043 			ioqs->ioq_rx_status = *v++;
3044 			ioqs->ioq_tx_status = *v++;
3045 
3046 			error = copyout(v, ioqs->data, ioqs->bufsize);
3047 		}
3048 
3049 		free(buf, M_DEVBUF);
3050 		break;
3051 	}
3052 	case CHELSIO_SET_FILTER: {
3053 		struct ch_filter *f = (struct ch_filter *)data;
3054 		struct filter_info *p;
3055 		unsigned int nfilters = sc->params.mc5.nfilters;
3056 
3057 		if (!is_offload(sc))
3058 			return (EOPNOTSUPP);	/* No TCAM */
3059 		if (!(sc->flags & FULL_INIT_DONE))
3060 			return (EAGAIN);	/* mc5 not setup yet */
3061 		if (nfilters == 0)
3062 			return (EBUSY);		/* TOE will use TCAM */
3063 
3064 		/* sanity checks */
3065 		if (f->filter_id >= nfilters ||
3066 		    (f->val.dip && f->mask.dip != 0xffffffff) ||
3067 		    (f->val.sport && f->mask.sport != 0xffff) ||
3068 		    (f->val.dport && f->mask.dport != 0xffff) ||
3069 		    (f->val.vlan && f->mask.vlan != 0xfff) ||
3070 		    (f->val.vlan_prio &&
3071 			f->mask.vlan_prio != FILTER_NO_VLAN_PRI) ||
3072 		    (f->mac_addr_idx != 0xffff && f->mac_addr_idx > 15) ||
3073 		    f->qset >= SGE_QSETS ||
3074 		    sc->rrss_map[f->qset] >= RSS_TABLE_SIZE)
3075 			return (EINVAL);
3076 
3077 		/* Was allocated with M_WAITOK */
3078 		KASSERT(sc->filters, ("filter table NULL\n"));
3079 
3080 		p = &sc->filters[f->filter_id];
3081 		if (p->locked)
3082 			return (EPERM);
3083 
3084 		bzero(p, sizeof(*p));
3085 		p->sip = f->val.sip;
3086 		p->sip_mask = f->mask.sip;
3087 		p->dip = f->val.dip;
3088 		p->sport = f->val.sport;
3089 		p->dport = f->val.dport;
3090 		p->vlan = f->mask.vlan ? f->val.vlan : 0xfff;
3091 		p->vlan_prio = f->mask.vlan_prio ? (f->val.vlan_prio & 6) :
3092 		    FILTER_NO_VLAN_PRI;
3093 		p->mac_hit = f->mac_hit;
3094 		p->mac_vld = f->mac_addr_idx != 0xffff;
3095 		p->mac_idx = f->mac_addr_idx;
3096 		p->pkt_type = f->proto;
3097 		p->report_filter_id = f->want_filter_id;
3098 		p->pass = f->pass;
3099 		p->rss = f->rss;
3100 		p->qset = f->qset;
3101 
3102 		error = set_filter(sc, f->filter_id, p);
3103 		if (error == 0)
3104 			p->valid = 1;
3105 		break;
3106 	}
3107 	case CHELSIO_DEL_FILTER: {
3108 		struct ch_filter *f = (struct ch_filter *)data;
3109 		struct filter_info *p;
3110 		unsigned int nfilters = sc->params.mc5.nfilters;
3111 
3112 		if (!is_offload(sc))
3113 			return (EOPNOTSUPP);
3114 		if (!(sc->flags & FULL_INIT_DONE))
3115 			return (EAGAIN);
3116 		if (nfilters == 0 || sc->filters == NULL)
3117 			return (EINVAL);
3118 		if (f->filter_id >= nfilters)
3119 		       return (EINVAL);
3120 
3121 		p = &sc->filters[f->filter_id];
3122 		if (p->locked)
3123 			return (EPERM);
3124 		if (!p->valid)
3125 			return (EFAULT); /* Read "Bad address" as "Bad index" */
3126 
3127 		bzero(p, sizeof(*p));
3128 		p->sip = p->sip_mask = 0xffffffff;
3129 		p->vlan = 0xfff;
3130 		p->vlan_prio = FILTER_NO_VLAN_PRI;
3131 		p->pkt_type = 1;
3132 		error = set_filter(sc, f->filter_id, p);
3133 		break;
3134 	}
3135 	case CHELSIO_GET_FILTER: {
3136 		struct ch_filter *f = (struct ch_filter *)data;
3137 		struct filter_info *p;
3138 		unsigned int i, nfilters = sc->params.mc5.nfilters;
3139 
3140 		if (!is_offload(sc))
3141 			return (EOPNOTSUPP);
3142 		if (!(sc->flags & FULL_INIT_DONE))
3143 			return (EAGAIN);
3144 		if (nfilters == 0 || sc->filters == NULL)
3145 			return (EINVAL);
3146 
3147 		i = f->filter_id == 0xffffffff ? 0 : f->filter_id + 1;
3148 		for (; i < nfilters; i++) {
3149 			p = &sc->filters[i];
3150 			if (!p->valid)
3151 				continue;
3152 
3153 			bzero(f, sizeof(*f));
3154 
3155 			f->filter_id = i;
3156 			f->val.sip = p->sip;
3157 			f->mask.sip = p->sip_mask;
3158 			f->val.dip = p->dip;
3159 			f->mask.dip = p->dip ? 0xffffffff : 0;
3160 			f->val.sport = p->sport;
3161 			f->mask.sport = p->sport ? 0xffff : 0;
3162 			f->val.dport = p->dport;
3163 			f->mask.dport = p->dport ? 0xffff : 0;
3164 			f->val.vlan = p->vlan == 0xfff ? 0 : p->vlan;
3165 			f->mask.vlan = p->vlan == 0xfff ? 0 : 0xfff;
3166 			f->val.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3167 			    0 : p->vlan_prio;
3168 			f->mask.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3169 			    0 : FILTER_NO_VLAN_PRI;
3170 			f->mac_hit = p->mac_hit;
3171 			f->mac_addr_idx = p->mac_vld ? p->mac_idx : 0xffff;
3172 			f->proto = p->pkt_type;
3173 			f->want_filter_id = p->report_filter_id;
3174 			f->pass = p->pass;
3175 			f->rss = p->rss;
3176 			f->qset = p->qset;
3177 
3178 			break;
3179 		}
3180 
3181 		if (i == nfilters)
3182 			f->filter_id = 0xffffffff;
3183 		break;
3184 	}
3185 	default:
3186 		return (EOPNOTSUPP);
3187 		break;
3188 	}
3189 
3190 	return (error);
3191 }
3192 
3193 static __inline void
3194 reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
3195     unsigned int end)
3196 {
3197 	uint32_t *p = (uint32_t *)(buf + start);
3198 
3199 	for ( ; start <= end; start += sizeof(uint32_t))
3200 		*p++ = t3_read_reg(ap, start);
3201 }
3202 
3203 #define T3_REGMAP_SIZE (3 * 1024)
3204 static int
3205 cxgb_get_regs_len(void)
3206 {
3207 	return T3_REGMAP_SIZE;
3208 }
3209 
3210 static void
3211 cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf)
3212 {
3213 
3214 	/*
3215 	 * Version scheme:
3216 	 * bits 0..9: chip version
3217 	 * bits 10..15: chip revision
3218 	 * bit 31: set for PCIe cards
3219 	 */
3220 	regs->version = 3 | (sc->params.rev << 10) | (is_pcie(sc) << 31);
3221 
3222 	/*
3223 	 * We skip the MAC statistics registers because they are clear-on-read.
3224 	 * Also reading multi-register stats would need to synchronize with the
3225 	 * periodic mac stats accumulation.  Hard to justify the complexity.
3226 	 */
3227 	memset(buf, 0, cxgb_get_regs_len());
3228 	reg_block_dump(sc, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
3229 	reg_block_dump(sc, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
3230 	reg_block_dump(sc, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
3231 	reg_block_dump(sc, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
3232 	reg_block_dump(sc, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
3233 	reg_block_dump(sc, buf, A_XGM_SERDES_STATUS0,
3234 		       XGM_REG(A_XGM_SERDES_STAT3, 1));
3235 	reg_block_dump(sc, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
3236 		       XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
3237 }
3238 
3239 static int
3240 alloc_filters(struct adapter *sc)
3241 {
3242 	struct filter_info *p;
3243 	unsigned int nfilters = sc->params.mc5.nfilters;
3244 
3245 	if (nfilters == 0)
3246 		return (0);
3247 
3248 	p = malloc(sizeof(*p) * nfilters, M_DEVBUF, M_WAITOK | M_ZERO);
3249 	sc->filters = p;
3250 
3251 	p = &sc->filters[nfilters - 1];
3252 	p->vlan = 0xfff;
3253 	p->vlan_prio = FILTER_NO_VLAN_PRI;
3254 	p->pass = p->rss = p->valid = p->locked = 1;
3255 
3256 	return (0);
3257 }
3258 
3259 static int
3260 setup_hw_filters(struct adapter *sc)
3261 {
3262 	int i, rc;
3263 	unsigned int nfilters = sc->params.mc5.nfilters;
3264 
3265 	if (!sc->filters)
3266 		return (0);
3267 
3268 	t3_enable_filters(sc);
3269 
3270 	for (i = rc = 0; i < nfilters && !rc; i++) {
3271 		if (sc->filters[i].locked)
3272 			rc = set_filter(sc, i, &sc->filters[i]);
3273 	}
3274 
3275 	return (rc);
3276 }
3277 
3278 static int
3279 set_filter(struct adapter *sc, int id, const struct filter_info *f)
3280 {
3281 	int len;
3282 	struct mbuf *m;
3283 	struct ulp_txpkt *txpkt;
3284 	struct work_request_hdr *wr;
3285 	struct cpl_pass_open_req *oreq;
3286 	struct cpl_set_tcb_field *sreq;
3287 
3288 	len = sizeof(*wr) + sizeof(*oreq) + 2 * sizeof(*sreq);
3289 	KASSERT(len <= MHLEN, ("filter request too big for an mbuf"));
3290 
3291 	id += t3_mc5_size(&sc->mc5) - sc->params.mc5.nroutes -
3292 	      sc->params.mc5.nfilters;
3293 
3294 	m = m_gethdr(M_WAITOK, MT_DATA);
3295 	m->m_len = m->m_pkthdr.len = len;
3296 	bzero(mtod(m, char *), len);
3297 
3298 	wr = mtod(m, struct work_request_hdr *);
3299 	wr->wrh_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS) | F_WR_ATOMIC);
3300 
3301 	oreq = (struct cpl_pass_open_req *)(wr + 1);
3302 	txpkt = (struct ulp_txpkt *)oreq;
3303 	txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3304 	txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*oreq) / 8));
3305 	OPCODE_TID(oreq) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, id));
3306 	oreq->local_port = htons(f->dport);
3307 	oreq->peer_port = htons(f->sport);
3308 	oreq->local_ip = htonl(f->dip);
3309 	oreq->peer_ip = htonl(f->sip);
3310 	oreq->peer_netmask = htonl(f->sip_mask);
3311 	oreq->opt0h = 0;
3312 	oreq->opt0l = htonl(F_NO_OFFLOAD);
3313 	oreq->opt1 = htonl(V_MAC_MATCH_VALID(f->mac_vld) |
3314 			 V_CONN_POLICY(CPL_CONN_POLICY_FILTER) |
3315 			 V_VLAN_PRI(f->vlan_prio >> 1) |
3316 			 V_VLAN_PRI_VALID(f->vlan_prio != FILTER_NO_VLAN_PRI) |
3317 			 V_PKT_TYPE(f->pkt_type) | V_OPT1_VLAN(f->vlan) |
3318 			 V_MAC_MATCH(f->mac_idx | (f->mac_hit << 4)));
3319 
3320 	sreq = (struct cpl_set_tcb_field *)(oreq + 1);
3321 	set_tcb_field_ulp(sreq, id, 1, 0x1800808000ULL,
3322 			  (f->report_filter_id << 15) | (1 << 23) |
3323 			  ((u64)f->pass << 35) | ((u64)!f->rss << 36));
3324 	set_tcb_field_ulp(sreq + 1, id, 0, 0xffffffff, (2 << 19) | 1);
3325 	t3_mgmt_tx(sc, m);
3326 
3327 	if (f->pass && !f->rss) {
3328 		len = sizeof(*sreq);
3329 		m = m_gethdr(M_WAITOK, MT_DATA);
3330 		m->m_len = m->m_pkthdr.len = len;
3331 		bzero(mtod(m, char *), len);
3332 		sreq = mtod(m, struct cpl_set_tcb_field *);
3333 		sreq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
3334 		mk_set_tcb_field(sreq, id, 25, 0x3f80000,
3335 				 (u64)sc->rrss_map[f->qset] << 19);
3336 		t3_mgmt_tx(sc, m);
3337 	}
3338 	return 0;
3339 }
3340 
3341 static inline void
3342 mk_set_tcb_field(struct cpl_set_tcb_field *req, unsigned int tid,
3343     unsigned int word, u64 mask, u64 val)
3344 {
3345 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
3346 	req->reply = V_NO_REPLY(1);
3347 	req->cpu_idx = 0;
3348 	req->word = htons(word);
3349 	req->mask = htobe64(mask);
3350 	req->val = htobe64(val);
3351 }
3352 
3353 static inline void
3354 set_tcb_field_ulp(struct cpl_set_tcb_field *req, unsigned int tid,
3355     unsigned int word, u64 mask, u64 val)
3356 {
3357 	struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req;
3358 
3359 	txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3360 	txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*req) / 8));
3361 	mk_set_tcb_field(req, tid, word, mask, val);
3362 }
3363 
3364 void
3365 t3_iterate(void (*func)(struct adapter *, void *), void *arg)
3366 {
3367 	struct adapter *sc;
3368 
3369 	mtx_lock(&t3_list_lock);
3370 	SLIST_FOREACH(sc, &t3_list, link) {
3371 		/*
3372 		 * func should not make any assumptions about what state sc is
3373 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
3374 		 */
3375 		func(sc, arg);
3376 	}
3377 	mtx_unlock(&t3_list_lock);
3378 }
3379 
3380 #ifdef TCP_OFFLOAD
3381 static int
3382 toe_capability(struct port_info *pi, int enable)
3383 {
3384 	int rc;
3385 	struct adapter *sc = pi->adapter;
3386 
3387 	ADAPTER_LOCK_ASSERT_OWNED(sc);
3388 
3389 	if (!is_offload(sc))
3390 		return (ENODEV);
3391 
3392 	if (enable) {
3393 		if (!(sc->flags & FULL_INIT_DONE)) {
3394 			log(LOG_WARNING,
3395 			    "You must enable a cxgb interface first\n");
3396 			return (EAGAIN);
3397 		}
3398 
3399 		if (isset(&sc->offload_map, pi->port_id))
3400 			return (0);
3401 
3402 		if (!(sc->flags & TOM_INIT_DONE)) {
3403 			rc = t3_activate_uld(sc, ULD_TOM);
3404 			if (rc == EAGAIN) {
3405 				log(LOG_WARNING,
3406 				    "You must kldload t3_tom.ko before trying "
3407 				    "to enable TOE on a cxgb interface.\n");
3408 			}
3409 			if (rc != 0)
3410 				return (rc);
3411 			KASSERT(sc->tom_softc != NULL,
3412 			    ("%s: TOM activated but softc NULL", __func__));
3413 			KASSERT(sc->flags & TOM_INIT_DONE,
3414 			    ("%s: TOM activated but flag not set", __func__));
3415 		}
3416 
3417 		setbit(&sc->offload_map, pi->port_id);
3418 
3419 		/*
3420 		 * XXX: Temporary code to allow iWARP to be enabled when TOE is
3421 		 * enabled on any port.  Need to figure out how to enable,
3422 		 * disable, load, and unload iWARP cleanly.
3423 		 */
3424 		if (!isset(&sc->offload_map, MAX_NPORTS) &&
3425 		    t3_activate_uld(sc, ULD_IWARP) == 0)
3426 			setbit(&sc->offload_map, MAX_NPORTS);
3427 	} else {
3428 		if (!isset(&sc->offload_map, pi->port_id))
3429 			return (0);
3430 
3431 		KASSERT(sc->flags & TOM_INIT_DONE,
3432 		    ("%s: TOM never initialized?", __func__));
3433 		clrbit(&sc->offload_map, pi->port_id);
3434 	}
3435 
3436 	return (0);
3437 }
3438 
3439 /*
3440  * Add an upper layer driver to the global list.
3441  */
3442 int
3443 t3_register_uld(struct uld_info *ui)
3444 {
3445 	int rc = 0;
3446 	struct uld_info *u;
3447 
3448 	mtx_lock(&t3_uld_list_lock);
3449 	SLIST_FOREACH(u, &t3_uld_list, link) {
3450 	    if (u->uld_id == ui->uld_id) {
3451 		    rc = EEXIST;
3452 		    goto done;
3453 	    }
3454 	}
3455 
3456 	SLIST_INSERT_HEAD(&t3_uld_list, ui, link);
3457 	ui->refcount = 0;
3458 done:
3459 	mtx_unlock(&t3_uld_list_lock);
3460 	return (rc);
3461 }
3462 
3463 int
3464 t3_unregister_uld(struct uld_info *ui)
3465 {
3466 	int rc = EINVAL;
3467 	struct uld_info *u;
3468 
3469 	mtx_lock(&t3_uld_list_lock);
3470 
3471 	SLIST_FOREACH(u, &t3_uld_list, link) {
3472 	    if (u == ui) {
3473 		    if (ui->refcount > 0) {
3474 			    rc = EBUSY;
3475 			    goto done;
3476 		    }
3477 
3478 		    SLIST_REMOVE(&t3_uld_list, ui, uld_info, link);
3479 		    rc = 0;
3480 		    goto done;
3481 	    }
3482 	}
3483 done:
3484 	mtx_unlock(&t3_uld_list_lock);
3485 	return (rc);
3486 }
3487 
3488 int
3489 t3_activate_uld(struct adapter *sc, int id)
3490 {
3491 	int rc = EAGAIN;
3492 	struct uld_info *ui;
3493 
3494 	mtx_lock(&t3_uld_list_lock);
3495 
3496 	SLIST_FOREACH(ui, &t3_uld_list, link) {
3497 		if (ui->uld_id == id) {
3498 			rc = ui->activate(sc);
3499 			if (rc == 0)
3500 				ui->refcount++;
3501 			goto done;
3502 		}
3503 	}
3504 done:
3505 	mtx_unlock(&t3_uld_list_lock);
3506 
3507 	return (rc);
3508 }
3509 
3510 int
3511 t3_deactivate_uld(struct adapter *sc, int id)
3512 {
3513 	int rc = EINVAL;
3514 	struct uld_info *ui;
3515 
3516 	mtx_lock(&t3_uld_list_lock);
3517 
3518 	SLIST_FOREACH(ui, &t3_uld_list, link) {
3519 		if (ui->uld_id == id) {
3520 			rc = ui->deactivate(sc);
3521 			if (rc == 0)
3522 				ui->refcount--;
3523 			goto done;
3524 		}
3525 	}
3526 done:
3527 	mtx_unlock(&t3_uld_list_lock);
3528 
3529 	return (rc);
3530 }
3531 
3532 static int
3533 cpl_not_handled(struct sge_qset *qs __unused, struct rsp_desc *r __unused,
3534     struct mbuf *m)
3535 {
3536 	m_freem(m);
3537 	return (EDOOFUS);
3538 }
3539 
3540 int
3541 t3_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
3542 {
3543 	uintptr_t *loc, new;
3544 
3545 	if (opcode >= NUM_CPL_HANDLERS)
3546 		return (EINVAL);
3547 
3548 	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
3549 	loc = (uintptr_t *) &sc->cpl_handler[opcode];
3550 	atomic_store_rel_ptr(loc, new);
3551 
3552 	return (0);
3553 }
3554 #endif
3555 
3556 static int
3557 cxgbc_mod_event(module_t mod, int cmd, void *arg)
3558 {
3559 	int rc = 0;
3560 
3561 	switch (cmd) {
3562 	case MOD_LOAD:
3563 		mtx_init(&t3_list_lock, "T3 adapters", 0, MTX_DEF);
3564 		SLIST_INIT(&t3_list);
3565 #ifdef TCP_OFFLOAD
3566 		mtx_init(&t3_uld_list_lock, "T3 ULDs", 0, MTX_DEF);
3567 		SLIST_INIT(&t3_uld_list);
3568 #endif
3569 		break;
3570 
3571 	case MOD_UNLOAD:
3572 #ifdef TCP_OFFLOAD
3573 		mtx_lock(&t3_uld_list_lock);
3574 		if (!SLIST_EMPTY(&t3_uld_list)) {
3575 			rc = EBUSY;
3576 			mtx_unlock(&t3_uld_list_lock);
3577 			break;
3578 		}
3579 		mtx_unlock(&t3_uld_list_lock);
3580 		mtx_destroy(&t3_uld_list_lock);
3581 #endif
3582 		mtx_lock(&t3_list_lock);
3583 		if (!SLIST_EMPTY(&t3_list)) {
3584 			rc = EBUSY;
3585 			mtx_unlock(&t3_list_lock);
3586 			break;
3587 		}
3588 		mtx_unlock(&t3_list_lock);
3589 		mtx_destroy(&t3_list_lock);
3590 		break;
3591 	}
3592 
3593 	return (rc);
3594 }
3595 
3596 #ifdef DEBUGNET
3597 static void
3598 cxgb_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
3599 {
3600 	struct port_info *pi;
3601 	adapter_t *adap;
3602 
3603 	pi = if_getsoftc(ifp);
3604 	adap = pi->adapter;
3605 	ADAPTER_LOCK(adap);
3606 	*nrxr = adap->nqsets;
3607 	*ncl = adap->sge.qs[0].fl[1].size;
3608 	*clsize = adap->sge.qs[0].fl[1].buf_size;
3609 	ADAPTER_UNLOCK(adap);
3610 }
3611 
3612 static void
3613 cxgb_debugnet_event(if_t ifp, enum debugnet_ev event)
3614 {
3615 	struct port_info *pi;
3616 	struct sge_qset *qs;
3617 	int i;
3618 
3619 	pi = if_getsoftc(ifp);
3620 	if (event == DEBUGNET_START)
3621 		for (i = 0; i < pi->adapter->nqsets; i++) {
3622 			qs = &pi->adapter->sge.qs[i];
3623 
3624 			/* Need to reinit after debugnet_mbuf_start(). */
3625 			qs->fl[0].zone = zone_pack;
3626 			qs->fl[1].zone = zone_clust;
3627 			qs->lro.enabled = 0;
3628 		}
3629 }
3630 
3631 static int
3632 cxgb_debugnet_transmit(if_t ifp, struct mbuf *m)
3633 {
3634 	struct port_info *pi;
3635 	struct sge_qset *qs;
3636 
3637 	pi = if_getsoftc(ifp);
3638 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
3639 	    IFF_DRV_RUNNING)
3640 		return (ENOENT);
3641 
3642 	qs = &pi->adapter->sge.qs[pi->first_qset];
3643 	return (cxgb_debugnet_encap(qs, &m));
3644 }
3645 
3646 static int
3647 cxgb_debugnet_poll(if_t ifp, int count)
3648 {
3649 	struct port_info *pi;
3650 	adapter_t *adap;
3651 	int i;
3652 
3653 	pi = if_getsoftc(ifp);
3654 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
3655 		return (ENOENT);
3656 
3657 	adap = pi->adapter;
3658 	for (i = 0; i < adap->nqsets; i++)
3659 		(void)cxgb_debugnet_poll_rx(adap, &adap->sge.qs[i]);
3660 	(void)cxgb_debugnet_poll_tx(&adap->sge.qs[pi->first_qset]);
3661 	return (0);
3662 }
3663 #endif /* DEBUGNET */
3664