xref: /freebsd/sys/dev/cxgb/cxgb_main.c (revision ce3adf4362fcca6a43e500b2531f0038adbfbd21)
1 /**************************************************************************
2 
3 Copyright (c) 2007-2009, Chelsio Inc.
4 All rights reserved.
5 
6 Redistribution and use in source and binary forms, with or without
7 modification, are permitted provided that the following conditions are met:
8 
9  1. Redistributions of source code must retain the above copyright notice,
10     this list of conditions and the following disclaimer.
11 
12  2. Neither the name of the Chelsio Corporation nor the names of its
13     contributors may be used to endorse or promote products derived from
14     this software without specific prior written permission.
15 
16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
20 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 POSSIBILITY OF SUCH DAMAGE.
27 
28 ***************************************************************************/
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_inet.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/bus.h>
39 #include <sys/module.h>
40 #include <sys/pciio.h>
41 #include <sys/conf.h>
42 #include <machine/bus.h>
43 #include <machine/resource.h>
44 #include <sys/bus_dma.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/ethernet.h>
62 #include <net/if.h>
63 #include <net/if_arp.h>
64 #include <net/if_dl.h>
65 #include <net/if_media.h>
66 #include <net/if_types.h>
67 #include <net/if_vlan_var.h>
68 
69 #include <netinet/in_systm.h>
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip.h>
74 #include <netinet/tcp.h>
75 #include <netinet/udp.h>
76 
77 #include <dev/pci/pcireg.h>
78 #include <dev/pci/pcivar.h>
79 #include <dev/pci/pci_private.h>
80 
81 #include <cxgb_include.h>
82 
83 #ifdef PRIV_SUPPORTED
84 #include <sys/priv.h>
85 #endif
86 
87 static int cxgb_setup_interrupts(adapter_t *);
88 static void cxgb_teardown_interrupts(adapter_t *);
89 static void cxgb_init(void *);
90 static int cxgb_init_locked(struct port_info *);
91 static int cxgb_uninit_locked(struct port_info *);
92 static int cxgb_uninit_synchronized(struct port_info *);
93 static int cxgb_ioctl(struct ifnet *, unsigned long, caddr_t);
94 static int cxgb_media_change(struct ifnet *);
95 static int cxgb_ifm_type(int);
96 static void cxgb_build_medialist(struct port_info *);
97 static void cxgb_media_status(struct ifnet *, struct ifmediareq *);
98 static int setup_sge_qsets(adapter_t *);
99 static void cxgb_async_intr(void *);
100 static void cxgb_tick_handler(void *, int);
101 static void cxgb_tick(void *);
102 static void link_check_callout(void *);
103 static void check_link_status(void *, int);
104 static void setup_rss(adapter_t *sc);
105 static int alloc_filters(struct adapter *);
106 static int setup_hw_filters(struct adapter *);
107 static int set_filter(struct adapter *, int, const struct filter_info *);
108 static inline void mk_set_tcb_field(struct cpl_set_tcb_field *, unsigned int,
109     unsigned int, u64, u64);
110 static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *, unsigned int,
111     unsigned int, u64, u64);
112 #ifdef TCP_OFFLOAD
113 static int cpl_not_handled(struct sge_qset *, struct rsp_desc *, struct mbuf *);
114 #endif
115 
116 /* Attachment glue for the PCI controller end of the device.  Each port of
117  * the device is attached separately, as defined later.
118  */
119 static int cxgb_controller_probe(device_t);
120 static int cxgb_controller_attach(device_t);
121 static int cxgb_controller_detach(device_t);
122 static void cxgb_free(struct adapter *);
123 static __inline void reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
124     unsigned int end);
125 static void cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf);
126 static int cxgb_get_regs_len(void);
127 static void touch_bars(device_t dev);
128 static void cxgb_update_mac_settings(struct port_info *p);
129 #ifdef TCP_OFFLOAD
130 static int toe_capability(struct port_info *, int);
131 #endif
132 
133 static device_method_t cxgb_controller_methods[] = {
134 	DEVMETHOD(device_probe,		cxgb_controller_probe),
135 	DEVMETHOD(device_attach,	cxgb_controller_attach),
136 	DEVMETHOD(device_detach,	cxgb_controller_detach),
137 
138 	DEVMETHOD_END
139 };
140 
141 static driver_t cxgb_controller_driver = {
142 	"cxgbc",
143 	cxgb_controller_methods,
144 	sizeof(struct adapter)
145 };
146 
147 static int cxgbc_mod_event(module_t, int, void *);
148 static devclass_t	cxgb_controller_devclass;
149 DRIVER_MODULE(cxgbc, pci, cxgb_controller_driver, cxgb_controller_devclass,
150     cxgbc_mod_event, 0);
151 MODULE_VERSION(cxgbc, 1);
152 MODULE_DEPEND(cxgbc, firmware, 1, 1, 1);
153 
154 /*
155  * Attachment glue for the ports.  Attachment is done directly to the
156  * controller device.
157  */
158 static int cxgb_port_probe(device_t);
159 static int cxgb_port_attach(device_t);
160 static int cxgb_port_detach(device_t);
161 
162 static device_method_t cxgb_port_methods[] = {
163 	DEVMETHOD(device_probe,		cxgb_port_probe),
164 	DEVMETHOD(device_attach,	cxgb_port_attach),
165 	DEVMETHOD(device_detach,	cxgb_port_detach),
166 	{ 0, 0 }
167 };
168 
169 static driver_t cxgb_port_driver = {
170 	"cxgb",
171 	cxgb_port_methods,
172 	0
173 };
174 
175 static d_ioctl_t cxgb_extension_ioctl;
176 static d_open_t cxgb_extension_open;
177 static d_close_t cxgb_extension_close;
178 
179 static struct cdevsw cxgb_cdevsw = {
180        .d_version =    D_VERSION,
181        .d_flags =      0,
182        .d_open =       cxgb_extension_open,
183        .d_close =      cxgb_extension_close,
184        .d_ioctl =      cxgb_extension_ioctl,
185        .d_name =       "cxgb",
186 };
187 
188 static devclass_t	cxgb_port_devclass;
189 DRIVER_MODULE(cxgb, cxgbc, cxgb_port_driver, cxgb_port_devclass, 0, 0);
190 MODULE_VERSION(cxgb, 1);
191 
192 static struct mtx t3_list_lock;
193 static SLIST_HEAD(, adapter) t3_list;
194 #ifdef TCP_OFFLOAD
195 static struct mtx t3_uld_list_lock;
196 static SLIST_HEAD(, uld_info) t3_uld_list;
197 #endif
198 
199 /*
200  * The driver uses the best interrupt scheme available on a platform in the
201  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
202  * of these schemes the driver may consider as follows:
203  *
204  * msi = 2: choose from among all three options
205  * msi = 1 : only consider MSI and pin interrupts
206  * msi = 0: force pin interrupts
207  */
208 static int msi_allowed = 2;
209 
210 TUNABLE_INT("hw.cxgb.msi_allowed", &msi_allowed);
211 SYSCTL_NODE(_hw, OID_AUTO, cxgb, CTLFLAG_RD, 0, "CXGB driver parameters");
212 SYSCTL_INT(_hw_cxgb, OID_AUTO, msi_allowed, CTLFLAG_RDTUN, &msi_allowed, 0,
213     "MSI-X, MSI, INTx selector");
214 
215 /*
216  * The driver uses an auto-queue algorithm by default.
217  * To disable it and force a single queue-set per port, use multiq = 0
218  */
219 static int multiq = 1;
220 TUNABLE_INT("hw.cxgb.multiq", &multiq);
221 SYSCTL_INT(_hw_cxgb, OID_AUTO, multiq, CTLFLAG_RDTUN, &multiq, 0,
222     "use min(ncpus/ports, 8) queue-sets per port");
223 
224 /*
225  * By default the driver will not update the firmware unless
226  * it was compiled against a newer version
227  *
228  */
229 static int force_fw_update = 0;
230 TUNABLE_INT("hw.cxgb.force_fw_update", &force_fw_update);
231 SYSCTL_INT(_hw_cxgb, OID_AUTO, force_fw_update, CTLFLAG_RDTUN, &force_fw_update, 0,
232     "update firmware even if up to date");
233 
234 int cxgb_use_16k_clusters = -1;
235 TUNABLE_INT("hw.cxgb.use_16k_clusters", &cxgb_use_16k_clusters);
236 SYSCTL_INT(_hw_cxgb, OID_AUTO, use_16k_clusters, CTLFLAG_RDTUN,
237     &cxgb_use_16k_clusters, 0, "use 16kB clusters for the jumbo queue ");
238 
239 static int nfilters = -1;
240 TUNABLE_INT("hw.cxgb.nfilters", &nfilters);
241 SYSCTL_INT(_hw_cxgb, OID_AUTO, nfilters, CTLFLAG_RDTUN,
242     &nfilters, 0, "max number of entries in the filter table");
243 
244 enum {
245 	MAX_TXQ_ENTRIES      = 16384,
246 	MAX_CTRL_TXQ_ENTRIES = 1024,
247 	MAX_RSPQ_ENTRIES     = 16384,
248 	MAX_RX_BUFFERS       = 16384,
249 	MAX_RX_JUMBO_BUFFERS = 16384,
250 	MIN_TXQ_ENTRIES      = 4,
251 	MIN_CTRL_TXQ_ENTRIES = 4,
252 	MIN_RSPQ_ENTRIES     = 32,
253 	MIN_FL_ENTRIES       = 32,
254 	MIN_FL_JUMBO_ENTRIES = 32
255 };
256 
257 struct filter_info {
258 	u32 sip;
259 	u32 sip_mask;
260 	u32 dip;
261 	u16 sport;
262 	u16 dport;
263 	u32 vlan:12;
264 	u32 vlan_prio:3;
265 	u32 mac_hit:1;
266 	u32 mac_idx:4;
267 	u32 mac_vld:1;
268 	u32 pkt_type:2;
269 	u32 report_filter_id:1;
270 	u32 pass:1;
271 	u32 rss:1;
272 	u32 qset:3;
273 	u32 locked:1;
274 	u32 valid:1;
275 };
276 
277 enum { FILTER_NO_VLAN_PRI = 7 };
278 
279 #define EEPROM_MAGIC 0x38E2F10C
280 
281 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
282 
283 /* Table for probing the cards.  The desc field isn't actually used */
284 struct cxgb_ident {
285 	uint16_t	vendor;
286 	uint16_t	device;
287 	int		index;
288 	char		*desc;
289 } cxgb_identifiers[] = {
290 	{PCI_VENDOR_ID_CHELSIO, 0x0020, 0, "PE9000"},
291 	{PCI_VENDOR_ID_CHELSIO, 0x0021, 1, "T302E"},
292 	{PCI_VENDOR_ID_CHELSIO, 0x0022, 2, "T310E"},
293 	{PCI_VENDOR_ID_CHELSIO, 0x0023, 3, "T320X"},
294 	{PCI_VENDOR_ID_CHELSIO, 0x0024, 1, "T302X"},
295 	{PCI_VENDOR_ID_CHELSIO, 0x0025, 3, "T320E"},
296 	{PCI_VENDOR_ID_CHELSIO, 0x0026, 2, "T310X"},
297 	{PCI_VENDOR_ID_CHELSIO, 0x0030, 2, "T3B10"},
298 	{PCI_VENDOR_ID_CHELSIO, 0x0031, 3, "T3B20"},
299 	{PCI_VENDOR_ID_CHELSIO, 0x0032, 1, "T3B02"},
300 	{PCI_VENDOR_ID_CHELSIO, 0x0033, 4, "T3B04"},
301 	{PCI_VENDOR_ID_CHELSIO, 0x0035, 6, "T3C10"},
302 	{PCI_VENDOR_ID_CHELSIO, 0x0036, 3, "S320E-CR"},
303 	{PCI_VENDOR_ID_CHELSIO, 0x0037, 7, "N320E-G2"},
304 	{0, 0, 0, NULL}
305 };
306 
307 static int set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset);
308 
309 
310 static __inline char
311 t3rev2char(struct adapter *adapter)
312 {
313 	char rev = 'z';
314 
315 	switch(adapter->params.rev) {
316 	case T3_REV_A:
317 		rev = 'a';
318 		break;
319 	case T3_REV_B:
320 	case T3_REV_B2:
321 		rev = 'b';
322 		break;
323 	case T3_REV_C:
324 		rev = 'c';
325 		break;
326 	}
327 	return rev;
328 }
329 
330 static struct cxgb_ident *
331 cxgb_get_ident(device_t dev)
332 {
333 	struct cxgb_ident *id;
334 
335 	for (id = cxgb_identifiers; id->desc != NULL; id++) {
336 		if ((id->vendor == pci_get_vendor(dev)) &&
337 		    (id->device == pci_get_device(dev))) {
338 			return (id);
339 		}
340 	}
341 	return (NULL);
342 }
343 
344 static const struct adapter_info *
345 cxgb_get_adapter_info(device_t dev)
346 {
347 	struct cxgb_ident *id;
348 	const struct adapter_info *ai;
349 
350 	id = cxgb_get_ident(dev);
351 	if (id == NULL)
352 		return (NULL);
353 
354 	ai = t3_get_adapter_info(id->index);
355 
356 	return (ai);
357 }
358 
359 static int
360 cxgb_controller_probe(device_t dev)
361 {
362 	const struct adapter_info *ai;
363 	char *ports, buf[80];
364 	int nports;
365 
366 	ai = cxgb_get_adapter_info(dev);
367 	if (ai == NULL)
368 		return (ENXIO);
369 
370 	nports = ai->nports0 + ai->nports1;
371 	if (nports == 1)
372 		ports = "port";
373 	else
374 		ports = "ports";
375 
376 	snprintf(buf, sizeof(buf), "%s, %d %s", ai->desc, nports, ports);
377 	device_set_desc_copy(dev, buf);
378 	return (BUS_PROBE_DEFAULT);
379 }
380 
381 #define FW_FNAME "cxgb_t3fw"
382 #define TPEEPROM_NAME "cxgb_t3%c_tp_eeprom"
383 #define TPSRAM_NAME "cxgb_t3%c_protocol_sram"
384 
385 static int
386 upgrade_fw(adapter_t *sc)
387 {
388 	const struct firmware *fw;
389 	int status;
390 	u32 vers;
391 
392 	if ((fw = firmware_get(FW_FNAME)) == NULL)  {
393 		device_printf(sc->dev, "Could not find firmware image %s\n", FW_FNAME);
394 		return (ENOENT);
395 	} else
396 		device_printf(sc->dev, "installing firmware on card\n");
397 	status = t3_load_fw(sc, (const uint8_t *)fw->data, fw->datasize);
398 
399 	if (status != 0) {
400 		device_printf(sc->dev, "failed to install firmware: %d\n",
401 		    status);
402 	} else {
403 		t3_get_fw_version(sc, &vers);
404 		snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
405 		    G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
406 		    G_FW_VERSION_MICRO(vers));
407 	}
408 
409 	firmware_put(fw, FIRMWARE_UNLOAD);
410 
411 	return (status);
412 }
413 
414 /*
415  * The cxgb_controller_attach function is responsible for the initial
416  * bringup of the device.  Its responsibilities include:
417  *
418  *  1. Determine if the device supports MSI or MSI-X.
419  *  2. Allocate bus resources so that we can access the Base Address Register
420  *  3. Create and initialize mutexes for the controller and its control
421  *     logic such as SGE and MDIO.
422  *  4. Call hardware specific setup routine for the adapter as a whole.
423  *  5. Allocate the BAR for doing MSI-X.
424  *  6. Setup the line interrupt iff MSI-X is not supported.
425  *  7. Create the driver's taskq.
426  *  8. Start one task queue service thread.
427  *  9. Check if the firmware and SRAM are up-to-date.  They will be
428  *     auto-updated later (before FULL_INIT_DONE), if required.
429  * 10. Create a child device for each MAC (port)
430  * 11. Initialize T3 private state.
431  * 12. Trigger the LED
432  * 13. Setup offload iff supported.
433  * 14. Reset/restart the tick callout.
434  * 15. Attach sysctls
435  *
436  * NOTE: Any modification or deviation from this list MUST be reflected in
437  * the above comment.  Failure to do so will result in problems on various
438  * error conditions including link flapping.
439  */
440 static int
441 cxgb_controller_attach(device_t dev)
442 {
443 	device_t child;
444 	const struct adapter_info *ai;
445 	struct adapter *sc;
446 	int i, error = 0;
447 	uint32_t vers;
448 	int port_qsets = 1;
449 	int msi_needed, reg;
450 	char buf[80];
451 
452 	sc = device_get_softc(dev);
453 	sc->dev = dev;
454 	sc->msi_count = 0;
455 	ai = cxgb_get_adapter_info(dev);
456 
457 	snprintf(sc->lockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb controller lock %d",
458 	    device_get_unit(dev));
459 	ADAPTER_LOCK_INIT(sc, sc->lockbuf);
460 
461 	snprintf(sc->reglockbuf, ADAPTER_LOCK_NAME_LEN, "SGE reg lock %d",
462 	    device_get_unit(dev));
463 	snprintf(sc->mdiolockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb mdio lock %d",
464 	    device_get_unit(dev));
465 	snprintf(sc->elmerlockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb elmer lock %d",
466 	    device_get_unit(dev));
467 
468 	MTX_INIT(&sc->sge.reg_lock, sc->reglockbuf, NULL, MTX_SPIN);
469 	MTX_INIT(&sc->mdio_lock, sc->mdiolockbuf, NULL, MTX_DEF);
470 	MTX_INIT(&sc->elmer_lock, sc->elmerlockbuf, NULL, MTX_DEF);
471 
472 	mtx_lock(&t3_list_lock);
473 	SLIST_INSERT_HEAD(&t3_list, sc, link);
474 	mtx_unlock(&t3_list_lock);
475 
476 	/* find the PCIe link width and set max read request to 4KB*/
477 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
478 		uint16_t lnk;
479 
480 		lnk = pci_read_config(dev, reg + PCIER_LINK_STA, 2);
481 		sc->link_width = (lnk & PCIEM_LINK_STA_WIDTH) >> 4;
482 		if (sc->link_width < 8 &&
483 		    (ai->caps & SUPPORTED_10000baseT_Full)) {
484 			device_printf(sc->dev,
485 			    "PCIe x%d Link, expect reduced performance\n",
486 			    sc->link_width);
487 		}
488 
489 		pci_set_max_read_req(dev, 4096);
490 	}
491 
492 	touch_bars(dev);
493 	pci_enable_busmaster(dev);
494 	/*
495 	 * Allocate the registers and make them available to the driver.
496 	 * The registers that we care about for NIC mode are in BAR 0
497 	 */
498 	sc->regs_rid = PCIR_BAR(0);
499 	if ((sc->regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
500 	    &sc->regs_rid, RF_ACTIVE)) == NULL) {
501 		device_printf(dev, "Cannot allocate BAR region 0\n");
502 		error = ENXIO;
503 		goto out;
504 	}
505 
506 	sc->bt = rman_get_bustag(sc->regs_res);
507 	sc->bh = rman_get_bushandle(sc->regs_res);
508 	sc->mmio_len = rman_get_size(sc->regs_res);
509 
510 	for (i = 0; i < MAX_NPORTS; i++)
511 		sc->port[i].adapter = sc;
512 
513 	if (t3_prep_adapter(sc, ai, 1) < 0) {
514 		printf("prep adapter failed\n");
515 		error = ENODEV;
516 		goto out;
517 	}
518 
519 	sc->udbs_rid = PCIR_BAR(2);
520 	sc->udbs_res = NULL;
521 	if (is_offload(sc) &&
522 	    ((sc->udbs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
523 		   &sc->udbs_rid, RF_ACTIVE)) == NULL)) {
524 		device_printf(dev, "Cannot allocate BAR region 1\n");
525 		error = ENXIO;
526 		goto out;
527 	}
528 
529         /* Allocate the BAR for doing MSI-X.  If it succeeds, try to allocate
530 	 * enough messages for the queue sets.  If that fails, try falling
531 	 * back to MSI.  If that fails, then try falling back to the legacy
532 	 * interrupt pin model.
533 	 */
534 	sc->msix_regs_rid = 0x20;
535 	if ((msi_allowed >= 2) &&
536 	    (sc->msix_regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
537 	    &sc->msix_regs_rid, RF_ACTIVE)) != NULL) {
538 
539 		if (multiq)
540 			port_qsets = min(SGE_QSETS/sc->params.nports, mp_ncpus);
541 		msi_needed = sc->msi_count = sc->params.nports * port_qsets + 1;
542 
543 		if (pci_msix_count(dev) == 0 ||
544 		    (error = pci_alloc_msix(dev, &sc->msi_count)) != 0 ||
545 		    sc->msi_count != msi_needed) {
546 			device_printf(dev, "alloc msix failed - "
547 				      "msi_count=%d, msi_needed=%d, err=%d; "
548 				      "will try MSI\n", sc->msi_count,
549 				      msi_needed, error);
550 			sc->msi_count = 0;
551 			port_qsets = 1;
552 			pci_release_msi(dev);
553 			bus_release_resource(dev, SYS_RES_MEMORY,
554 			    sc->msix_regs_rid, sc->msix_regs_res);
555 			sc->msix_regs_res = NULL;
556 		} else {
557 			sc->flags |= USING_MSIX;
558 			sc->cxgb_intr = cxgb_async_intr;
559 			device_printf(dev,
560 				      "using MSI-X interrupts (%u vectors)\n",
561 				      sc->msi_count);
562 		}
563 	}
564 
565 	if ((msi_allowed >= 1) && (sc->msi_count == 0)) {
566 		sc->msi_count = 1;
567 		if ((error = pci_alloc_msi(dev, &sc->msi_count)) != 0) {
568 			device_printf(dev, "alloc msi failed - "
569 				      "err=%d; will try INTx\n", error);
570 			sc->msi_count = 0;
571 			port_qsets = 1;
572 			pci_release_msi(dev);
573 		} else {
574 			sc->flags |= USING_MSI;
575 			sc->cxgb_intr = t3_intr_msi;
576 			device_printf(dev, "using MSI interrupts\n");
577 		}
578 	}
579 	if (sc->msi_count == 0) {
580 		device_printf(dev, "using line interrupts\n");
581 		sc->cxgb_intr = t3b_intr;
582 	}
583 
584 	/* Create a private taskqueue thread for handling driver events */
585 	sc->tq = taskqueue_create("cxgb_taskq", M_NOWAIT,
586 	    taskqueue_thread_enqueue, &sc->tq);
587 	if (sc->tq == NULL) {
588 		device_printf(dev, "failed to allocate controller task queue\n");
589 		goto out;
590 	}
591 
592 	taskqueue_start_threads(&sc->tq, 1, PI_NET, "%s taskq",
593 	    device_get_nameunit(dev));
594 	TASK_INIT(&sc->tick_task, 0, cxgb_tick_handler, sc);
595 
596 
597 	/* Create a periodic callout for checking adapter status */
598 	callout_init(&sc->cxgb_tick_ch, TRUE);
599 
600 	if (t3_check_fw_version(sc) < 0 || force_fw_update) {
601 		/*
602 		 * Warn user that a firmware update will be attempted in init.
603 		 */
604 		device_printf(dev, "firmware needs to be updated to version %d.%d.%d\n",
605 		    FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
606 		sc->flags &= ~FW_UPTODATE;
607 	} else {
608 		sc->flags |= FW_UPTODATE;
609 	}
610 
611 	if (t3_check_tpsram_version(sc) < 0) {
612 		/*
613 		 * Warn user that a firmware update will be attempted in init.
614 		 */
615 		device_printf(dev, "SRAM needs to be updated to version %c-%d.%d.%d\n",
616 		    t3rev2char(sc), TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
617 		sc->flags &= ~TPS_UPTODATE;
618 	} else {
619 		sc->flags |= TPS_UPTODATE;
620 	}
621 
622 	/*
623 	 * Create a child device for each MAC.  The ethernet attachment
624 	 * will be done in these children.
625 	 */
626 	for (i = 0; i < (sc)->params.nports; i++) {
627 		struct port_info *pi;
628 
629 		if ((child = device_add_child(dev, "cxgb", -1)) == NULL) {
630 			device_printf(dev, "failed to add child port\n");
631 			error = EINVAL;
632 			goto out;
633 		}
634 		pi = &sc->port[i];
635 		pi->adapter = sc;
636 		pi->nqsets = port_qsets;
637 		pi->first_qset = i*port_qsets;
638 		pi->port_id = i;
639 		pi->tx_chan = i >= ai->nports0;
640 		pi->txpkt_intf = pi->tx_chan ? 2 * (i - ai->nports0) + 1 : 2 * i;
641 		sc->rxpkt_map[pi->txpkt_intf] = i;
642 		sc->port[i].tx_chan = i >= ai->nports0;
643 		sc->portdev[i] = child;
644 		device_set_softc(child, pi);
645 	}
646 	if ((error = bus_generic_attach(dev)) != 0)
647 		goto out;
648 
649 	/* initialize sge private state */
650 	t3_sge_init_adapter(sc);
651 
652 	t3_led_ready(sc);
653 
654 	error = t3_get_fw_version(sc, &vers);
655 	if (error)
656 		goto out;
657 
658 	snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
659 	    G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
660 	    G_FW_VERSION_MICRO(vers));
661 
662 	snprintf(buf, sizeof(buf), "%s %sNIC\t E/C: %s S/N: %s",
663 		 ai->desc, is_offload(sc) ? "R" : "",
664 		 sc->params.vpd.ec, sc->params.vpd.sn);
665 	device_set_desc_copy(dev, buf);
666 
667 	snprintf(&sc->port_types[0], sizeof(sc->port_types), "%x%x%x%x",
668 		 sc->params.vpd.port_type[0], sc->params.vpd.port_type[1],
669 		 sc->params.vpd.port_type[2], sc->params.vpd.port_type[3]);
670 
671 	device_printf(sc->dev, "Firmware Version %s\n", &sc->fw_version[0]);
672 	callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
673 	t3_add_attach_sysctls(sc);
674 
675 #ifdef TCP_OFFLOAD
676 	for (i = 0; i < NUM_CPL_HANDLERS; i++)
677 		sc->cpl_handler[i] = cpl_not_handled;
678 #endif
679 
680 	t3_intr_clear(sc);
681 	error = cxgb_setup_interrupts(sc);
682 out:
683 	if (error)
684 		cxgb_free(sc);
685 
686 	return (error);
687 }
688 
689 /*
690  * The cxgb_controller_detach routine is called with the device is
691  * unloaded from the system.
692  */
693 
694 static int
695 cxgb_controller_detach(device_t dev)
696 {
697 	struct adapter *sc;
698 
699 	sc = device_get_softc(dev);
700 
701 	cxgb_free(sc);
702 
703 	return (0);
704 }
705 
706 /*
707  * The cxgb_free() is called by the cxgb_controller_detach() routine
708  * to tear down the structures that were built up in
709  * cxgb_controller_attach(), and should be the final piece of work
710  * done when fully unloading the driver.
711  *
712  *
713  *  1. Shutting down the threads started by the cxgb_controller_attach()
714  *     routine.
715  *  2. Stopping the lower level device and all callouts (cxgb_down_locked()).
716  *  3. Detaching all of the port devices created during the
717  *     cxgb_controller_attach() routine.
718  *  4. Removing the device children created via cxgb_controller_attach().
719  *  5. Releasing PCI resources associated with the device.
720  *  6. Turning off the offload support, iff it was turned on.
721  *  7. Destroying the mutexes created in cxgb_controller_attach().
722  *
723  */
724 static void
725 cxgb_free(struct adapter *sc)
726 {
727 	int i, nqsets = 0;
728 
729 	ADAPTER_LOCK(sc);
730 	sc->flags |= CXGB_SHUTDOWN;
731 	ADAPTER_UNLOCK(sc);
732 
733 	/*
734 	 * Make sure all child devices are gone.
735 	 */
736 	bus_generic_detach(sc->dev);
737 	for (i = 0; i < (sc)->params.nports; i++) {
738 		if (sc->portdev[i] &&
739 		    device_delete_child(sc->dev, sc->portdev[i]) != 0)
740 			device_printf(sc->dev, "failed to delete child port\n");
741 		nqsets += sc->port[i].nqsets;
742 	}
743 
744 	/*
745 	 * At this point, it is as if cxgb_port_detach has run on all ports, and
746 	 * cxgb_down has run on the adapter.  All interrupts have been silenced,
747 	 * all open devices have been closed.
748 	 */
749 	KASSERT(sc->open_device_map == 0, ("%s: device(s) still open (%x)",
750 					   __func__, sc->open_device_map));
751 	for (i = 0; i < sc->params.nports; i++) {
752 		KASSERT(sc->port[i].ifp == NULL, ("%s: port %i undead!",
753 						  __func__, i));
754 	}
755 
756 	/*
757 	 * Finish off the adapter's callouts.
758 	 */
759 	callout_drain(&sc->cxgb_tick_ch);
760 	callout_drain(&sc->sge_timer_ch);
761 
762 	/*
763 	 * Release resources grabbed under FULL_INIT_DONE by cxgb_up.  The
764 	 * sysctls are cleaned up by the kernel linker.
765 	 */
766 	if (sc->flags & FULL_INIT_DONE) {
767  		t3_free_sge_resources(sc, nqsets);
768  		sc->flags &= ~FULL_INIT_DONE;
769  	}
770 
771 	/*
772 	 * Release all interrupt resources.
773 	 */
774 	cxgb_teardown_interrupts(sc);
775 	if (sc->flags & (USING_MSI | USING_MSIX)) {
776 		device_printf(sc->dev, "releasing msi message(s)\n");
777 		pci_release_msi(sc->dev);
778 	} else {
779 		device_printf(sc->dev, "no msi message to release\n");
780 	}
781 
782 	if (sc->msix_regs_res != NULL) {
783 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->msix_regs_rid,
784 		    sc->msix_regs_res);
785 	}
786 
787 	/*
788 	 * Free the adapter's taskqueue.
789 	 */
790 	if (sc->tq != NULL) {
791 		taskqueue_free(sc->tq);
792 		sc->tq = NULL;
793 	}
794 
795 	free(sc->filters, M_DEVBUF);
796 	t3_sge_free(sc);
797 
798 	if (sc->udbs_res != NULL)
799 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->udbs_rid,
800 		    sc->udbs_res);
801 
802 	if (sc->regs_res != NULL)
803 		bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->regs_rid,
804 		    sc->regs_res);
805 
806 	MTX_DESTROY(&sc->mdio_lock);
807 	MTX_DESTROY(&sc->sge.reg_lock);
808 	MTX_DESTROY(&sc->elmer_lock);
809 	mtx_lock(&t3_list_lock);
810 	SLIST_REMOVE(&t3_list, sc, adapter, link);
811 	mtx_unlock(&t3_list_lock);
812 	ADAPTER_LOCK_DEINIT(sc);
813 }
814 
815 /**
816  *	setup_sge_qsets - configure SGE Tx/Rx/response queues
817  *	@sc: the controller softc
818  *
819  *	Determines how many sets of SGE queues to use and initializes them.
820  *	We support multiple queue sets per port if we have MSI-X, otherwise
821  *	just one queue set per port.
822  */
823 static int
824 setup_sge_qsets(adapter_t *sc)
825 {
826 	int i, j, err, irq_idx = 0, qset_idx = 0;
827 	u_int ntxq = SGE_TXQ_PER_SET;
828 
829 	if ((err = t3_sge_alloc(sc)) != 0) {
830 		device_printf(sc->dev, "t3_sge_alloc returned %d\n", err);
831 		return (err);
832 	}
833 
834 	if (sc->params.rev > 0 && !(sc->flags & USING_MSI))
835 		irq_idx = -1;
836 
837 	for (i = 0; i < (sc)->params.nports; i++) {
838 		struct port_info *pi = &sc->port[i];
839 
840 		for (j = 0; j < pi->nqsets; j++, qset_idx++) {
841 			err = t3_sge_alloc_qset(sc, qset_idx, (sc)->params.nports,
842 			    (sc->flags & USING_MSIX) ? qset_idx + 1 : irq_idx,
843 			    &sc->params.sge.qset[qset_idx], ntxq, pi);
844 			if (err) {
845 				t3_free_sge_resources(sc, qset_idx);
846 				device_printf(sc->dev,
847 				    "t3_sge_alloc_qset failed with %d\n", err);
848 				return (err);
849 			}
850 		}
851 	}
852 
853 	return (0);
854 }
855 
856 static void
857 cxgb_teardown_interrupts(adapter_t *sc)
858 {
859 	int i;
860 
861 	for (i = 0; i < SGE_QSETS; i++) {
862 		if (sc->msix_intr_tag[i] == NULL) {
863 
864 			/* Should have been setup fully or not at all */
865 			KASSERT(sc->msix_irq_res[i] == NULL &&
866 				sc->msix_irq_rid[i] == 0,
867 				("%s: half-done interrupt (%d).", __func__, i));
868 
869 			continue;
870 		}
871 
872 		bus_teardown_intr(sc->dev, sc->msix_irq_res[i],
873 				  sc->msix_intr_tag[i]);
874 		bus_release_resource(sc->dev, SYS_RES_IRQ, sc->msix_irq_rid[i],
875 				     sc->msix_irq_res[i]);
876 
877 		sc->msix_irq_res[i] = sc->msix_intr_tag[i] = NULL;
878 		sc->msix_irq_rid[i] = 0;
879 	}
880 
881 	if (sc->intr_tag) {
882 		KASSERT(sc->irq_res != NULL,
883 			("%s: half-done interrupt.", __func__));
884 
885 		bus_teardown_intr(sc->dev, sc->irq_res, sc->intr_tag);
886 		bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
887 				     sc->irq_res);
888 
889 		sc->irq_res = sc->intr_tag = NULL;
890 		sc->irq_rid = 0;
891 	}
892 }
893 
894 static int
895 cxgb_setup_interrupts(adapter_t *sc)
896 {
897 	struct resource *res;
898 	void *tag;
899 	int i, rid, err, intr_flag = sc->flags & (USING_MSI | USING_MSIX);
900 
901 	sc->irq_rid = intr_flag ? 1 : 0;
902 	sc->irq_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &sc->irq_rid,
903 					     RF_SHAREABLE | RF_ACTIVE);
904 	if (sc->irq_res == NULL) {
905 		device_printf(sc->dev, "Cannot allocate interrupt (%x, %u)\n",
906 			      intr_flag, sc->irq_rid);
907 		err = EINVAL;
908 		sc->irq_rid = 0;
909 	} else {
910 		err = bus_setup_intr(sc->dev, sc->irq_res,
911 		    INTR_MPSAFE | INTR_TYPE_NET, NULL,
912 		    sc->cxgb_intr, sc, &sc->intr_tag);
913 
914 		if (err) {
915 			device_printf(sc->dev,
916 				      "Cannot set up interrupt (%x, %u, %d)\n",
917 				      intr_flag, sc->irq_rid, err);
918 			bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
919 					     sc->irq_res);
920 			sc->irq_res = sc->intr_tag = NULL;
921 			sc->irq_rid = 0;
922 		}
923 	}
924 
925 	/* That's all for INTx or MSI */
926 	if (!(intr_flag & USING_MSIX) || err)
927 		return (err);
928 
929 	bus_describe_intr(sc->dev, sc->irq_res, sc->intr_tag, "err");
930 	for (i = 0; i < sc->msi_count - 1; i++) {
931 		rid = i + 2;
932 		res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &rid,
933 					     RF_SHAREABLE | RF_ACTIVE);
934 		if (res == NULL) {
935 			device_printf(sc->dev, "Cannot allocate interrupt "
936 				      "for message %d\n", rid);
937 			err = EINVAL;
938 			break;
939 		}
940 
941 		err = bus_setup_intr(sc->dev, res, INTR_MPSAFE | INTR_TYPE_NET,
942 				     NULL, t3_intr_msix, &sc->sge.qs[i], &tag);
943 		if (err) {
944 			device_printf(sc->dev, "Cannot set up interrupt "
945 				      "for message %d (%d)\n", rid, err);
946 			bus_release_resource(sc->dev, SYS_RES_IRQ, rid, res);
947 			break;
948 		}
949 
950 		sc->msix_irq_rid[i] = rid;
951 		sc->msix_irq_res[i] = res;
952 		sc->msix_intr_tag[i] = tag;
953 		bus_describe_intr(sc->dev, res, tag, "qs%d", i);
954 	}
955 
956 	if (err)
957 		cxgb_teardown_interrupts(sc);
958 
959 	return (err);
960 }
961 
962 
963 static int
964 cxgb_port_probe(device_t dev)
965 {
966 	struct port_info *p;
967 	char buf[80];
968 	const char *desc;
969 
970 	p = device_get_softc(dev);
971 	desc = p->phy.desc;
972 	snprintf(buf, sizeof(buf), "Port %d %s", p->port_id, desc);
973 	device_set_desc_copy(dev, buf);
974 	return (0);
975 }
976 
977 
978 static int
979 cxgb_makedev(struct port_info *pi)
980 {
981 
982 	pi->port_cdev = make_dev(&cxgb_cdevsw, pi->ifp->if_dunit,
983 	    UID_ROOT, GID_WHEEL, 0600, "%s", if_name(pi->ifp));
984 
985 	if (pi->port_cdev == NULL)
986 		return (ENOMEM);
987 
988 	pi->port_cdev->si_drv1 = (void *)pi;
989 
990 	return (0);
991 }
992 
993 #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
994     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
995     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6)
996 #define CXGB_CAP_ENABLE CXGB_CAP
997 
998 static int
999 cxgb_port_attach(device_t dev)
1000 {
1001 	struct port_info *p;
1002 	struct ifnet *ifp;
1003 	int err;
1004 	struct adapter *sc;
1005 
1006 	p = device_get_softc(dev);
1007 	sc = p->adapter;
1008 	snprintf(p->lockbuf, PORT_NAME_LEN, "cxgb port lock %d:%d",
1009 	    device_get_unit(device_get_parent(dev)), p->port_id);
1010 	PORT_LOCK_INIT(p, p->lockbuf);
1011 
1012 	callout_init(&p->link_check_ch, CALLOUT_MPSAFE);
1013 	TASK_INIT(&p->link_check_task, 0, check_link_status, p);
1014 
1015 	/* Allocate an ifnet object and set it up */
1016 	ifp = p->ifp = if_alloc(IFT_ETHER);
1017 	if (ifp == NULL) {
1018 		device_printf(dev, "Cannot allocate ifnet\n");
1019 		return (ENOMEM);
1020 	}
1021 
1022 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1023 	ifp->if_init = cxgb_init;
1024 	ifp->if_softc = p;
1025 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1026 	ifp->if_ioctl = cxgb_ioctl;
1027 	ifp->if_transmit = cxgb_transmit;
1028 	ifp->if_qflush = cxgb_qflush;
1029 
1030 	ifp->if_capabilities = CXGB_CAP;
1031 #ifdef TCP_OFFLOAD
1032 	if (is_offload(sc))
1033 		ifp->if_capabilities |= IFCAP_TOE4;
1034 #endif
1035 	ifp->if_capenable = CXGB_CAP_ENABLE;
1036 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1037 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1038 
1039 	/*
1040 	 * Disable TSO on 4-port - it isn't supported by the firmware.
1041 	 */
1042 	if (sc->params.nports > 2) {
1043 		ifp->if_capabilities &= ~(IFCAP_TSO | IFCAP_VLAN_HWTSO);
1044 		ifp->if_capenable &= ~(IFCAP_TSO | IFCAP_VLAN_HWTSO);
1045 		ifp->if_hwassist &= ~CSUM_TSO;
1046 	}
1047 
1048 	ether_ifattach(ifp, p->hw_addr);
1049 
1050 #ifdef DEFAULT_JUMBO
1051 	if (sc->params.nports <= 2)
1052 		ifp->if_mtu = ETHERMTU_JUMBO;
1053 #endif
1054 	if ((err = cxgb_makedev(p)) != 0) {
1055 		printf("makedev failed %d\n", err);
1056 		return (err);
1057 	}
1058 
1059 	/* Create a list of media supported by this port */
1060 	ifmedia_init(&p->media, IFM_IMASK, cxgb_media_change,
1061 	    cxgb_media_status);
1062 	cxgb_build_medialist(p);
1063 
1064 	t3_sge_init_port(p);
1065 
1066 	return (err);
1067 }
1068 
1069 /*
1070  * cxgb_port_detach() is called via the device_detach methods when
1071  * cxgb_free() calls the bus_generic_detach.  It is responsible for
1072  * removing the device from the view of the kernel, i.e. from all
1073  * interfaces lists etc.  This routine is only called when the driver is
1074  * being unloaded, not when the link goes down.
1075  */
1076 static int
1077 cxgb_port_detach(device_t dev)
1078 {
1079 	struct port_info *p;
1080 	struct adapter *sc;
1081 	int i;
1082 
1083 	p = device_get_softc(dev);
1084 	sc = p->adapter;
1085 
1086 	/* Tell cxgb_ioctl and if_init that the port is going away */
1087 	ADAPTER_LOCK(sc);
1088 	SET_DOOMED(p);
1089 	wakeup(&sc->flags);
1090 	while (IS_BUSY(sc))
1091 		mtx_sleep(&sc->flags, &sc->lock, 0, "cxgbdtch", 0);
1092 	SET_BUSY(sc);
1093 	ADAPTER_UNLOCK(sc);
1094 
1095 	if (p->port_cdev != NULL)
1096 		destroy_dev(p->port_cdev);
1097 
1098 	cxgb_uninit_synchronized(p);
1099 	ether_ifdetach(p->ifp);
1100 
1101 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1102 		struct sge_qset *qs = &sc->sge.qs[i];
1103 		struct sge_txq *txq = &qs->txq[TXQ_ETH];
1104 
1105 		callout_drain(&txq->txq_watchdog);
1106 		callout_drain(&txq->txq_timer);
1107 	}
1108 
1109 	PORT_LOCK_DEINIT(p);
1110 	if_free(p->ifp);
1111 	p->ifp = NULL;
1112 
1113 	ADAPTER_LOCK(sc);
1114 	CLR_BUSY(sc);
1115 	wakeup_one(&sc->flags);
1116 	ADAPTER_UNLOCK(sc);
1117 	return (0);
1118 }
1119 
1120 void
1121 t3_fatal_err(struct adapter *sc)
1122 {
1123 	u_int fw_status[4];
1124 
1125 	if (sc->flags & FULL_INIT_DONE) {
1126 		t3_sge_stop(sc);
1127 		t3_write_reg(sc, A_XGM_TX_CTRL, 0);
1128 		t3_write_reg(sc, A_XGM_RX_CTRL, 0);
1129 		t3_write_reg(sc, XGM_REG(A_XGM_TX_CTRL, 1), 0);
1130 		t3_write_reg(sc, XGM_REG(A_XGM_RX_CTRL, 1), 0);
1131 		t3_intr_disable(sc);
1132 	}
1133 	device_printf(sc->dev,"encountered fatal error, operation suspended\n");
1134 	if (!t3_cim_ctl_blk_read(sc, 0xa0, 4, fw_status))
1135 		device_printf(sc->dev, "FW_ status: 0x%x, 0x%x, 0x%x, 0x%x\n",
1136 		    fw_status[0], fw_status[1], fw_status[2], fw_status[3]);
1137 }
1138 
1139 int
1140 t3_os_find_pci_capability(adapter_t *sc, int cap)
1141 {
1142 	device_t dev;
1143 	struct pci_devinfo *dinfo;
1144 	pcicfgregs *cfg;
1145 	uint32_t status;
1146 	uint8_t ptr;
1147 
1148 	dev = sc->dev;
1149 	dinfo = device_get_ivars(dev);
1150 	cfg = &dinfo->cfg;
1151 
1152 	status = pci_read_config(dev, PCIR_STATUS, 2);
1153 	if (!(status & PCIM_STATUS_CAPPRESENT))
1154 		return (0);
1155 
1156 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1157 	case 0:
1158 	case 1:
1159 		ptr = PCIR_CAP_PTR;
1160 		break;
1161 	case 2:
1162 		ptr = PCIR_CAP_PTR_2;
1163 		break;
1164 	default:
1165 		return (0);
1166 		break;
1167 	}
1168 	ptr = pci_read_config(dev, ptr, 1);
1169 
1170 	while (ptr != 0) {
1171 		if (pci_read_config(dev, ptr + PCICAP_ID, 1) == cap)
1172 			return (ptr);
1173 		ptr = pci_read_config(dev, ptr + PCICAP_NEXTPTR, 1);
1174 	}
1175 
1176 	return (0);
1177 }
1178 
1179 int
1180 t3_os_pci_save_state(struct adapter *sc)
1181 {
1182 	device_t dev;
1183 	struct pci_devinfo *dinfo;
1184 
1185 	dev = sc->dev;
1186 	dinfo = device_get_ivars(dev);
1187 
1188 	pci_cfg_save(dev, dinfo, 0);
1189 	return (0);
1190 }
1191 
1192 int
1193 t3_os_pci_restore_state(struct adapter *sc)
1194 {
1195 	device_t dev;
1196 	struct pci_devinfo *dinfo;
1197 
1198 	dev = sc->dev;
1199 	dinfo = device_get_ivars(dev);
1200 
1201 	pci_cfg_restore(dev, dinfo);
1202 	return (0);
1203 }
1204 
1205 /**
1206  *	t3_os_link_changed - handle link status changes
1207  *	@sc: the adapter associated with the link change
1208  *	@port_id: the port index whose link status has changed
1209  *	@link_status: the new status of the link
1210  *	@speed: the new speed setting
1211  *	@duplex: the new duplex setting
1212  *	@fc: the new flow-control setting
1213  *
1214  *	This is the OS-dependent handler for link status changes.  The OS
1215  *	neutral handler takes care of most of the processing for these events,
1216  *	then calls this handler for any OS-specific processing.
1217  */
1218 void
1219 t3_os_link_changed(adapter_t *adapter, int port_id, int link_status, int speed,
1220      int duplex, int fc, int mac_was_reset)
1221 {
1222 	struct port_info *pi = &adapter->port[port_id];
1223 	struct ifnet *ifp = pi->ifp;
1224 
1225 	/* no race with detach, so ifp should always be good */
1226 	KASSERT(ifp, ("%s: if detached.", __func__));
1227 
1228 	/* Reapply mac settings if they were lost due to a reset */
1229 	if (mac_was_reset) {
1230 		PORT_LOCK(pi);
1231 		cxgb_update_mac_settings(pi);
1232 		PORT_UNLOCK(pi);
1233 	}
1234 
1235 	if (link_status) {
1236 		ifp->if_baudrate = IF_Mbps(speed);
1237 		if_link_state_change(ifp, LINK_STATE_UP);
1238 	} else
1239 		if_link_state_change(ifp, LINK_STATE_DOWN);
1240 }
1241 
1242 /**
1243  *	t3_os_phymod_changed - handle PHY module changes
1244  *	@phy: the PHY reporting the module change
1245  *	@mod_type: new module type
1246  *
1247  *	This is the OS-dependent handler for PHY module changes.  It is
1248  *	invoked when a PHY module is removed or inserted for any OS-specific
1249  *	processing.
1250  */
1251 void t3_os_phymod_changed(struct adapter *adap, int port_id)
1252 {
1253 	static const char *mod_str[] = {
1254 		NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX-L", "unknown"
1255 	};
1256 	struct port_info *pi = &adap->port[port_id];
1257 	int mod = pi->phy.modtype;
1258 
1259 	if (mod != pi->media.ifm_cur->ifm_data)
1260 		cxgb_build_medialist(pi);
1261 
1262 	if (mod == phy_modtype_none)
1263 		if_printf(pi->ifp, "PHY module unplugged\n");
1264 	else {
1265 		KASSERT(mod < ARRAY_SIZE(mod_str),
1266 			("invalid PHY module type %d", mod));
1267 		if_printf(pi->ifp, "%s PHY module inserted\n", mod_str[mod]);
1268 	}
1269 }
1270 
1271 void
1272 t3_os_set_hw_addr(adapter_t *adapter, int port_idx, u8 hw_addr[])
1273 {
1274 
1275 	/*
1276 	 * The ifnet might not be allocated before this gets called,
1277 	 * as this is called early on in attach by t3_prep_adapter
1278 	 * save the address off in the port structure
1279 	 */
1280 	if (cxgb_debug)
1281 		printf("set_hw_addr on idx %d addr %6D\n", port_idx, hw_addr, ":");
1282 	bcopy(hw_addr, adapter->port[port_idx].hw_addr, ETHER_ADDR_LEN);
1283 }
1284 
1285 /*
1286  * Programs the XGMAC based on the settings in the ifnet.  These settings
1287  * include MTU, MAC address, mcast addresses, etc.
1288  */
1289 static void
1290 cxgb_update_mac_settings(struct port_info *p)
1291 {
1292 	struct ifnet *ifp = p->ifp;
1293 	struct t3_rx_mode rm;
1294 	struct cmac *mac = &p->mac;
1295 	int mtu, hwtagging;
1296 
1297 	PORT_LOCK_ASSERT_OWNED(p);
1298 
1299 	bcopy(IF_LLADDR(ifp), p->hw_addr, ETHER_ADDR_LEN);
1300 
1301 	mtu = ifp->if_mtu;
1302 	if (ifp->if_capenable & IFCAP_VLAN_MTU)
1303 		mtu += ETHER_VLAN_ENCAP_LEN;
1304 
1305 	hwtagging = (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0;
1306 
1307 	t3_mac_set_mtu(mac, mtu);
1308 	t3_set_vlan_accel(p->adapter, 1 << p->tx_chan, hwtagging);
1309 	t3_mac_set_address(mac, 0, p->hw_addr);
1310 	t3_init_rx_mode(&rm, p);
1311 	t3_mac_set_rx_mode(mac, &rm);
1312 }
1313 
1314 
1315 static int
1316 await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
1317 			      unsigned long n)
1318 {
1319 	int attempts = 5;
1320 
1321 	while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
1322 		if (!--attempts)
1323 			return (ETIMEDOUT);
1324 		t3_os_sleep(10);
1325 	}
1326 	return 0;
1327 }
1328 
1329 static int
1330 init_tp_parity(struct adapter *adap)
1331 {
1332 	int i;
1333 	struct mbuf *m;
1334 	struct cpl_set_tcb_field *greq;
1335 	unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
1336 
1337 	t3_tp_set_offload_mode(adap, 1);
1338 
1339 	for (i = 0; i < 16; i++) {
1340 		struct cpl_smt_write_req *req;
1341 
1342 		m = m_gethdr(M_WAITOK, MT_DATA);
1343 		req = mtod(m, struct cpl_smt_write_req *);
1344 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1345 		memset(req, 0, sizeof(*req));
1346 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1347 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
1348 		req->iff = i;
1349 		t3_mgmt_tx(adap, m);
1350 	}
1351 
1352 	for (i = 0; i < 2048; i++) {
1353 		struct cpl_l2t_write_req *req;
1354 
1355 		m = m_gethdr(M_WAITOK, MT_DATA);
1356 		req = mtod(m, struct cpl_l2t_write_req *);
1357 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1358 		memset(req, 0, sizeof(*req));
1359 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1360 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
1361 		req->params = htonl(V_L2T_W_IDX(i));
1362 		t3_mgmt_tx(adap, m);
1363 	}
1364 
1365 	for (i = 0; i < 2048; i++) {
1366 		struct cpl_rte_write_req *req;
1367 
1368 		m = m_gethdr(M_WAITOK, MT_DATA);
1369 		req = mtod(m, struct cpl_rte_write_req *);
1370 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1371 		memset(req, 0, sizeof(*req));
1372 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1373 		OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
1374 		req->l2t_idx = htonl(V_L2T_W_IDX(i));
1375 		t3_mgmt_tx(adap, m);
1376 	}
1377 
1378 	m = m_gethdr(M_WAITOK, MT_DATA);
1379 	greq = mtod(m, struct cpl_set_tcb_field *);
1380 	m->m_len = m->m_pkthdr.len = sizeof(*greq);
1381 	memset(greq, 0, sizeof(*greq));
1382 	greq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1383 	OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
1384 	greq->mask = htobe64(1);
1385 	t3_mgmt_tx(adap, m);
1386 
1387 	i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
1388 	t3_tp_set_offload_mode(adap, 0);
1389 	return (i);
1390 }
1391 
1392 /**
1393  *	setup_rss - configure Receive Side Steering (per-queue connection demux)
1394  *	@adap: the adapter
1395  *
1396  *	Sets up RSS to distribute packets to multiple receive queues.  We
1397  *	configure the RSS CPU lookup table to distribute to the number of HW
1398  *	receive queues, and the response queue lookup table to narrow that
1399  *	down to the response queues actually configured for each port.
1400  *	We always configure the RSS mapping for two ports since the mapping
1401  *	table has plenty of entries.
1402  */
1403 static void
1404 setup_rss(adapter_t *adap)
1405 {
1406 	int i;
1407 	u_int nq[2];
1408 	uint8_t cpus[SGE_QSETS + 1];
1409 	uint16_t rspq_map[RSS_TABLE_SIZE];
1410 
1411 	for (i = 0; i < SGE_QSETS; ++i)
1412 		cpus[i] = i;
1413 	cpus[SGE_QSETS] = 0xff;
1414 
1415 	nq[0] = nq[1] = 0;
1416 	for_each_port(adap, i) {
1417 		const struct port_info *pi = adap2pinfo(adap, i);
1418 
1419 		nq[pi->tx_chan] += pi->nqsets;
1420 	}
1421 	for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
1422 		rspq_map[i] = nq[0] ? i % nq[0] : 0;
1423 		rspq_map[i + RSS_TABLE_SIZE / 2] = nq[1] ? i % nq[1] + nq[0] : 0;
1424 	}
1425 
1426 	/* Calculate the reverse RSS map table */
1427 	for (i = 0; i < SGE_QSETS; ++i)
1428 		adap->rrss_map[i] = 0xff;
1429 	for (i = 0; i < RSS_TABLE_SIZE; ++i)
1430 		if (adap->rrss_map[rspq_map[i]] == 0xff)
1431 			adap->rrss_map[rspq_map[i]] = i;
1432 
1433 	t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
1434 		      F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN | F_OFDMAPEN |
1435 	              F_RRCPLMAPEN | V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ,
1436 	              cpus, rspq_map);
1437 
1438 }
1439 static void
1440 send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
1441 			      int hi, int port)
1442 {
1443 	struct mbuf *m;
1444 	struct mngt_pktsched_wr *req;
1445 
1446 	m = m_gethdr(M_NOWAIT, MT_DATA);
1447 	if (m) {
1448 		req = mtod(m, struct mngt_pktsched_wr *);
1449 		req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
1450 		req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
1451 		req->sched = sched;
1452 		req->idx = qidx;
1453 		req->min = lo;
1454 		req->max = hi;
1455 		req->binding = port;
1456 		m->m_len = m->m_pkthdr.len = sizeof(*req);
1457 		t3_mgmt_tx(adap, m);
1458 	}
1459 }
1460 
1461 static void
1462 bind_qsets(adapter_t *sc)
1463 {
1464 	int i, j;
1465 
1466 	for (i = 0; i < (sc)->params.nports; ++i) {
1467 		const struct port_info *pi = adap2pinfo(sc, i);
1468 
1469 		for (j = 0; j < pi->nqsets; ++j) {
1470 			send_pktsched_cmd(sc, 1, pi->first_qset + j, -1,
1471 					  -1, pi->tx_chan);
1472 
1473 		}
1474 	}
1475 }
1476 
1477 static void
1478 update_tpeeprom(struct adapter *adap)
1479 {
1480 	const struct firmware *tpeeprom;
1481 
1482 	uint32_t version;
1483 	unsigned int major, minor;
1484 	int ret, len;
1485 	char rev, name[32];
1486 
1487 	t3_seeprom_read(adap, TP_SRAM_OFFSET, &version);
1488 
1489 	major = G_TP_VERSION_MAJOR(version);
1490 	minor = G_TP_VERSION_MINOR(version);
1491 	if (major == TP_VERSION_MAJOR  && minor == TP_VERSION_MINOR)
1492 		return;
1493 
1494 	rev = t3rev2char(adap);
1495 	snprintf(name, sizeof(name), TPEEPROM_NAME, rev);
1496 
1497 	tpeeprom = firmware_get(name);
1498 	if (tpeeprom == NULL) {
1499 		device_printf(adap->dev,
1500 			      "could not load TP EEPROM: unable to load %s\n",
1501 			      name);
1502 		return;
1503 	}
1504 
1505 	len = tpeeprom->datasize - 4;
1506 
1507 	ret = t3_check_tpsram(adap, tpeeprom->data, tpeeprom->datasize);
1508 	if (ret)
1509 		goto release_tpeeprom;
1510 
1511 	if (len != TP_SRAM_LEN) {
1512 		device_printf(adap->dev,
1513 			      "%s length is wrong len=%d expected=%d\n", name,
1514 			      len, TP_SRAM_LEN);
1515 		return;
1516 	}
1517 
1518 	ret = set_eeprom(&adap->port[0], tpeeprom->data, tpeeprom->datasize,
1519 	    TP_SRAM_OFFSET);
1520 
1521 	if (!ret) {
1522 		device_printf(adap->dev,
1523 			"Protocol SRAM image updated in EEPROM to %d.%d.%d\n",
1524 			 TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1525 	} else
1526 		device_printf(adap->dev,
1527 			      "Protocol SRAM image update in EEPROM failed\n");
1528 
1529 release_tpeeprom:
1530 	firmware_put(tpeeprom, FIRMWARE_UNLOAD);
1531 
1532 	return;
1533 }
1534 
1535 static int
1536 update_tpsram(struct adapter *adap)
1537 {
1538 	const struct firmware *tpsram;
1539 	int ret;
1540 	char rev, name[32];
1541 
1542 	rev = t3rev2char(adap);
1543 	snprintf(name, sizeof(name), TPSRAM_NAME, rev);
1544 
1545 	update_tpeeprom(adap);
1546 
1547 	tpsram = firmware_get(name);
1548 	if (tpsram == NULL){
1549 		device_printf(adap->dev, "could not load TP SRAM\n");
1550 		return (EINVAL);
1551 	} else
1552 		device_printf(adap->dev, "updating TP SRAM\n");
1553 
1554 	ret = t3_check_tpsram(adap, tpsram->data, tpsram->datasize);
1555 	if (ret)
1556 		goto release_tpsram;
1557 
1558 	ret = t3_set_proto_sram(adap, tpsram->data);
1559 	if (ret)
1560 		device_printf(adap->dev, "loading protocol SRAM failed\n");
1561 
1562 release_tpsram:
1563 	firmware_put(tpsram, FIRMWARE_UNLOAD);
1564 
1565 	return ret;
1566 }
1567 
1568 /**
1569  *	cxgb_up - enable the adapter
1570  *	@adap: adapter being enabled
1571  *
1572  *	Called when the first port is enabled, this function performs the
1573  *	actions necessary to make an adapter operational, such as completing
1574  *	the initialization of HW modules, and enabling interrupts.
1575  */
1576 static int
1577 cxgb_up(struct adapter *sc)
1578 {
1579 	int err = 0;
1580 	unsigned int mxf = t3_mc5_size(&sc->mc5) - MC5_MIN_TIDS;
1581 
1582 	KASSERT(sc->open_device_map == 0, ("%s: device(s) already open (%x)",
1583 					   __func__, sc->open_device_map));
1584 
1585 	if ((sc->flags & FULL_INIT_DONE) == 0) {
1586 
1587 		ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1588 
1589 		if ((sc->flags & FW_UPTODATE) == 0)
1590 			if ((err = upgrade_fw(sc)))
1591 				goto out;
1592 
1593 		if ((sc->flags & TPS_UPTODATE) == 0)
1594 			if ((err = update_tpsram(sc)))
1595 				goto out;
1596 
1597 		if (is_offload(sc) && nfilters != 0) {
1598 			sc->params.mc5.nservers = 0;
1599 
1600 			if (nfilters < 0)
1601 				sc->params.mc5.nfilters = mxf;
1602 			else
1603 				sc->params.mc5.nfilters = min(nfilters, mxf);
1604 		}
1605 
1606 		err = t3_init_hw(sc, 0);
1607 		if (err)
1608 			goto out;
1609 
1610 		t3_set_reg_field(sc, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1611 		t3_write_reg(sc, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1612 
1613 		err = setup_sge_qsets(sc);
1614 		if (err)
1615 			goto out;
1616 
1617 		alloc_filters(sc);
1618 		setup_rss(sc);
1619 
1620 		t3_add_configured_sysctls(sc);
1621 		sc->flags |= FULL_INIT_DONE;
1622 	}
1623 
1624 	t3_intr_clear(sc);
1625 	t3_sge_start(sc);
1626 	t3_intr_enable(sc);
1627 
1628 	if (sc->params.rev >= T3_REV_C && !(sc->flags & TP_PARITY_INIT) &&
1629 	    is_offload(sc) && init_tp_parity(sc) == 0)
1630 		sc->flags |= TP_PARITY_INIT;
1631 
1632 	if (sc->flags & TP_PARITY_INIT) {
1633 		t3_write_reg(sc, A_TP_INT_CAUSE, F_CMCACHEPERR | F_ARPLUTPERR);
1634 		t3_write_reg(sc, A_TP_INT_ENABLE, 0x7fbfffff);
1635 	}
1636 
1637 	if (!(sc->flags & QUEUES_BOUND)) {
1638 		bind_qsets(sc);
1639 		setup_hw_filters(sc);
1640 		sc->flags |= QUEUES_BOUND;
1641 	}
1642 
1643 	t3_sge_reset_adapter(sc);
1644 out:
1645 	return (err);
1646 }
1647 
1648 /*
1649  * Called when the last open device is closed.  Does NOT undo all of cxgb_up's
1650  * work.  Specifically, the resources grabbed under FULL_INIT_DONE are released
1651  * during controller_detach, not here.
1652  */
1653 static void
1654 cxgb_down(struct adapter *sc)
1655 {
1656 	t3_sge_stop(sc);
1657 	t3_intr_disable(sc);
1658 }
1659 
1660 /*
1661  * if_init for cxgb ports.
1662  */
1663 static void
1664 cxgb_init(void *arg)
1665 {
1666 	struct port_info *p = arg;
1667 	struct adapter *sc = p->adapter;
1668 
1669 	ADAPTER_LOCK(sc);
1670 	cxgb_init_locked(p); /* releases adapter lock */
1671 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1672 }
1673 
1674 static int
1675 cxgb_init_locked(struct port_info *p)
1676 {
1677 	struct adapter *sc = p->adapter;
1678 	struct ifnet *ifp = p->ifp;
1679 	struct cmac *mac = &p->mac;
1680 	int i, rc = 0, may_sleep = 0, gave_up_lock = 0;
1681 
1682 	ADAPTER_LOCK_ASSERT_OWNED(sc);
1683 
1684 	while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1685 		gave_up_lock = 1;
1686 		if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbinit", 0)) {
1687 			rc = EINTR;
1688 			goto done;
1689 		}
1690 	}
1691 	if (IS_DOOMED(p)) {
1692 		rc = ENXIO;
1693 		goto done;
1694 	}
1695 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1696 
1697 	/*
1698 	 * The code that runs during one-time adapter initialization can sleep
1699 	 * so it's important not to hold any locks across it.
1700 	 */
1701 	may_sleep = sc->flags & FULL_INIT_DONE ? 0 : 1;
1702 
1703 	if (may_sleep) {
1704 		SET_BUSY(sc);
1705 		gave_up_lock = 1;
1706 		ADAPTER_UNLOCK(sc);
1707 	}
1708 
1709 	if (sc->open_device_map == 0 && ((rc = cxgb_up(sc)) != 0))
1710 			goto done;
1711 
1712 	PORT_LOCK(p);
1713 	if (isset(&sc->open_device_map, p->port_id) &&
1714 	    (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1715 		PORT_UNLOCK(p);
1716 		goto done;
1717 	}
1718 	t3_port_intr_enable(sc, p->port_id);
1719 	if (!mac->multiport)
1720 		t3_mac_init(mac);
1721 	cxgb_update_mac_settings(p);
1722 	t3_link_start(&p->phy, mac, &p->link_config);
1723 	t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
1724 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1725 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1726 	PORT_UNLOCK(p);
1727 
1728 	for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1729 		struct sge_qset *qs = &sc->sge.qs[i];
1730 		struct sge_txq *txq = &qs->txq[TXQ_ETH];
1731 
1732 		callout_reset_on(&txq->txq_watchdog, hz, cxgb_tx_watchdog, qs,
1733 				 txq->txq_watchdog.c_cpu);
1734 	}
1735 
1736 	/* all ok */
1737 	setbit(&sc->open_device_map, p->port_id);
1738 	callout_reset(&p->link_check_ch,
1739 	    p->phy.caps & SUPPORTED_LINK_IRQ ?  hz * 3 : hz / 4,
1740 	    link_check_callout, p);
1741 
1742 done:
1743 	if (may_sleep) {
1744 		ADAPTER_LOCK(sc);
1745 		KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1746 		CLR_BUSY(sc);
1747 	}
1748 	if (gave_up_lock)
1749 		wakeup_one(&sc->flags);
1750 	ADAPTER_UNLOCK(sc);
1751 	return (rc);
1752 }
1753 
1754 static int
1755 cxgb_uninit_locked(struct port_info *p)
1756 {
1757 	struct adapter *sc = p->adapter;
1758 	int rc;
1759 
1760 	ADAPTER_LOCK_ASSERT_OWNED(sc);
1761 
1762 	while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1763 		if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbunin", 0)) {
1764 			rc = EINTR;
1765 			goto done;
1766 		}
1767 	}
1768 	if (IS_DOOMED(p)) {
1769 		rc = ENXIO;
1770 		goto done;
1771 	}
1772 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1773 	SET_BUSY(sc);
1774 	ADAPTER_UNLOCK(sc);
1775 
1776 	rc = cxgb_uninit_synchronized(p);
1777 
1778 	ADAPTER_LOCK(sc);
1779 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1780 	CLR_BUSY(sc);
1781 	wakeup_one(&sc->flags);
1782 done:
1783 	ADAPTER_UNLOCK(sc);
1784 	return (rc);
1785 }
1786 
1787 /*
1788  * Called on "ifconfig down", and from port_detach
1789  */
1790 static int
1791 cxgb_uninit_synchronized(struct port_info *pi)
1792 {
1793 	struct adapter *sc = pi->adapter;
1794 	struct ifnet *ifp = pi->ifp;
1795 
1796 	/*
1797 	 * taskqueue_drain may cause a deadlock if the adapter lock is held.
1798 	 */
1799 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1800 
1801 	/*
1802 	 * Clear this port's bit from the open device map, and then drain all
1803 	 * the tasks that can access/manipulate this port's port_info or ifp.
1804 	 * We disable this port's interrupts here and so the slow/ext
1805 	 * interrupt tasks won't be enqueued.  The tick task will continue to
1806 	 * be enqueued every second but the runs after this drain will not see
1807 	 * this port in the open device map.
1808 	 *
1809 	 * A well behaved task must take open_device_map into account and ignore
1810 	 * ports that are not open.
1811 	 */
1812 	clrbit(&sc->open_device_map, pi->port_id);
1813 	t3_port_intr_disable(sc, pi->port_id);
1814 	taskqueue_drain(sc->tq, &sc->slow_intr_task);
1815 	taskqueue_drain(sc->tq, &sc->tick_task);
1816 
1817 	callout_drain(&pi->link_check_ch);
1818 	taskqueue_drain(sc->tq, &pi->link_check_task);
1819 
1820 	PORT_LOCK(pi);
1821 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1822 
1823 	/* disable pause frames */
1824 	t3_set_reg_field(sc, A_XGM_TX_CFG + pi->mac.offset, F_TXPAUSEEN, 0);
1825 
1826 	/* Reset RX FIFO HWM */
1827 	t3_set_reg_field(sc, A_XGM_RXFIFO_CFG +  pi->mac.offset,
1828 			 V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM), 0);
1829 
1830 	DELAY(100 * 1000);
1831 
1832 	/* Wait for TXFIFO empty */
1833 	t3_wait_op_done(sc, A_XGM_TXFIFO_CFG + pi->mac.offset,
1834 			F_TXFIFO_EMPTY, 1, 20, 5);
1835 
1836 	DELAY(100 * 1000);
1837 	t3_mac_disable(&pi->mac, MAC_DIRECTION_RX);
1838 
1839 	pi->phy.ops->power_down(&pi->phy, 1);
1840 
1841 	PORT_UNLOCK(pi);
1842 
1843 	pi->link_config.link_ok = 0;
1844 	t3_os_link_changed(sc, pi->port_id, 0, 0, 0, 0, 0);
1845 
1846 	if (sc->open_device_map == 0)
1847 		cxgb_down(pi->adapter);
1848 
1849 	return (0);
1850 }
1851 
1852 /*
1853  * Mark lro enabled or disabled in all qsets for this port
1854  */
1855 static int
1856 cxgb_set_lro(struct port_info *p, int enabled)
1857 {
1858 	int i;
1859 	struct adapter *adp = p->adapter;
1860 	struct sge_qset *q;
1861 
1862 	for (i = 0; i < p->nqsets; i++) {
1863 		q = &adp->sge.qs[p->first_qset + i];
1864 		q->lro.enabled = (enabled != 0);
1865 	}
1866 	return (0);
1867 }
1868 
1869 static int
1870 cxgb_ioctl(struct ifnet *ifp, unsigned long command, caddr_t data)
1871 {
1872 	struct port_info *p = ifp->if_softc;
1873 	struct adapter *sc = p->adapter;
1874 	struct ifreq *ifr = (struct ifreq *)data;
1875 	int flags, error = 0, mtu;
1876 	uint32_t mask;
1877 
1878 	switch (command) {
1879 	case SIOCSIFMTU:
1880 		ADAPTER_LOCK(sc);
1881 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1882 		if (error) {
1883 fail:
1884 			ADAPTER_UNLOCK(sc);
1885 			return (error);
1886 		}
1887 
1888 		mtu = ifr->ifr_mtu;
1889 		if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO)) {
1890 			error = EINVAL;
1891 		} else {
1892 			ifp->if_mtu = mtu;
1893 			PORT_LOCK(p);
1894 			cxgb_update_mac_settings(p);
1895 			PORT_UNLOCK(p);
1896 		}
1897 		ADAPTER_UNLOCK(sc);
1898 		break;
1899 	case SIOCSIFFLAGS:
1900 		ADAPTER_LOCK(sc);
1901 		if (IS_DOOMED(p)) {
1902 			error = ENXIO;
1903 			goto fail;
1904 		}
1905 		if (ifp->if_flags & IFF_UP) {
1906 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1907 				flags = p->if_flags;
1908 				if (((ifp->if_flags ^ flags) & IFF_PROMISC) ||
1909 				    ((ifp->if_flags ^ flags) & IFF_ALLMULTI)) {
1910 					if (IS_BUSY(sc)) {
1911 						error = EBUSY;
1912 						goto fail;
1913 					}
1914 					PORT_LOCK(p);
1915 					cxgb_update_mac_settings(p);
1916 					PORT_UNLOCK(p);
1917 				}
1918 				ADAPTER_UNLOCK(sc);
1919 			} else
1920 				error = cxgb_init_locked(p);
1921 			p->if_flags = ifp->if_flags;
1922 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1923 			error = cxgb_uninit_locked(p);
1924 		else
1925 			ADAPTER_UNLOCK(sc);
1926 
1927 		ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1928 		break;
1929 	case SIOCADDMULTI:
1930 	case SIOCDELMULTI:
1931 		ADAPTER_LOCK(sc);
1932 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1933 		if (error)
1934 			goto fail;
1935 
1936 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1937 			PORT_LOCK(p);
1938 			cxgb_update_mac_settings(p);
1939 			PORT_UNLOCK(p);
1940 		}
1941 		ADAPTER_UNLOCK(sc);
1942 
1943 		break;
1944 	case SIOCSIFCAP:
1945 		ADAPTER_LOCK(sc);
1946 		error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1947 		if (error)
1948 			goto fail;
1949 
1950 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1951 		if (mask & IFCAP_TXCSUM) {
1952 			ifp->if_capenable ^= IFCAP_TXCSUM;
1953 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1954 
1955 			if (IFCAP_TSO4 & ifp->if_capenable &&
1956 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1957 				ifp->if_capenable &= ~IFCAP_TSO4;
1958 				if_printf(ifp,
1959 				    "tso4 disabled due to -txcsum.\n");
1960 			}
1961 		}
1962 		if (mask & IFCAP_TXCSUM_IPV6) {
1963 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1964 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1965 
1966 			if (IFCAP_TSO6 & ifp->if_capenable &&
1967 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1968 				ifp->if_capenable &= ~IFCAP_TSO6;
1969 				if_printf(ifp,
1970 				    "tso6 disabled due to -txcsum6.\n");
1971 			}
1972 		}
1973 		if (mask & IFCAP_RXCSUM)
1974 			ifp->if_capenable ^= IFCAP_RXCSUM;
1975 		if (mask & IFCAP_RXCSUM_IPV6)
1976 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1977 
1978 		/*
1979 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1980 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1981 		 * sending a TSO request our way, so it's sufficient to toggle
1982 		 * IFCAP_TSOx only.
1983 		 */
1984 		if (mask & IFCAP_TSO4) {
1985 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1986 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1987 				if_printf(ifp, "enable txcsum first.\n");
1988 				error = EAGAIN;
1989 				goto fail;
1990 			}
1991 			ifp->if_capenable ^= IFCAP_TSO4;
1992 		}
1993 		if (mask & IFCAP_TSO6) {
1994 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1995 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1996 				if_printf(ifp, "enable txcsum6 first.\n");
1997 				error = EAGAIN;
1998 				goto fail;
1999 			}
2000 			ifp->if_capenable ^= IFCAP_TSO6;
2001 		}
2002 		if (mask & IFCAP_LRO) {
2003 			ifp->if_capenable ^= IFCAP_LRO;
2004 
2005 			/* Safe to do this even if cxgb_up not called yet */
2006 			cxgb_set_lro(p, ifp->if_capenable & IFCAP_LRO);
2007 		}
2008 #ifdef TCP_OFFLOAD
2009 		if (mask & IFCAP_TOE4) {
2010 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE4;
2011 
2012 			error = toe_capability(p, enable);
2013 			if (error == 0)
2014 				ifp->if_capenable ^= mask;
2015 		}
2016 #endif
2017 		if (mask & IFCAP_VLAN_HWTAGGING) {
2018 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2019 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2020 				PORT_LOCK(p);
2021 				cxgb_update_mac_settings(p);
2022 				PORT_UNLOCK(p);
2023 			}
2024 		}
2025 		if (mask & IFCAP_VLAN_MTU) {
2026 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
2027 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2028 				PORT_LOCK(p);
2029 				cxgb_update_mac_settings(p);
2030 				PORT_UNLOCK(p);
2031 			}
2032 		}
2033 		if (mask & IFCAP_VLAN_HWTSO)
2034 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2035 		if (mask & IFCAP_VLAN_HWCSUM)
2036 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
2037 
2038 #ifdef VLAN_CAPABILITIES
2039 		VLAN_CAPABILITIES(ifp);
2040 #endif
2041 		ADAPTER_UNLOCK(sc);
2042 		break;
2043 	case SIOCSIFMEDIA:
2044 	case SIOCGIFMEDIA:
2045 		error = ifmedia_ioctl(ifp, ifr, &p->media, command);
2046 		break;
2047 	default:
2048 		error = ether_ioctl(ifp, command, data);
2049 	}
2050 
2051 	return (error);
2052 }
2053 
2054 static int
2055 cxgb_media_change(struct ifnet *ifp)
2056 {
2057 	return (EOPNOTSUPP);
2058 }
2059 
2060 /*
2061  * Translates phy->modtype to the correct Ethernet media subtype.
2062  */
2063 static int
2064 cxgb_ifm_type(int mod)
2065 {
2066 	switch (mod) {
2067 	case phy_modtype_sr:
2068 		return (IFM_10G_SR);
2069 	case phy_modtype_lr:
2070 		return (IFM_10G_LR);
2071 	case phy_modtype_lrm:
2072 		return (IFM_10G_LRM);
2073 	case phy_modtype_twinax:
2074 		return (IFM_10G_TWINAX);
2075 	case phy_modtype_twinax_long:
2076 		return (IFM_10G_TWINAX_LONG);
2077 	case phy_modtype_none:
2078 		return (IFM_NONE);
2079 	case phy_modtype_unknown:
2080 		return (IFM_UNKNOWN);
2081 	}
2082 
2083 	KASSERT(0, ("%s: modtype %d unknown", __func__, mod));
2084 	return (IFM_UNKNOWN);
2085 }
2086 
2087 /*
2088  * Rebuilds the ifmedia list for this port, and sets the current media.
2089  */
2090 static void
2091 cxgb_build_medialist(struct port_info *p)
2092 {
2093 	struct cphy *phy = &p->phy;
2094 	struct ifmedia *media = &p->media;
2095 	int mod = phy->modtype;
2096 	int m = IFM_ETHER | IFM_FDX;
2097 
2098 	PORT_LOCK(p);
2099 
2100 	ifmedia_removeall(media);
2101 	if (phy->caps & SUPPORTED_TP && phy->caps & SUPPORTED_Autoneg) {
2102 		/* Copper (RJ45) */
2103 
2104 		if (phy->caps & SUPPORTED_10000baseT_Full)
2105 			ifmedia_add(media, m | IFM_10G_T, mod, NULL);
2106 
2107 		if (phy->caps & SUPPORTED_1000baseT_Full)
2108 			ifmedia_add(media, m | IFM_1000_T, mod, NULL);
2109 
2110 		if (phy->caps & SUPPORTED_100baseT_Full)
2111 			ifmedia_add(media, m | IFM_100_TX, mod, NULL);
2112 
2113 		if (phy->caps & SUPPORTED_10baseT_Full)
2114 			ifmedia_add(media, m | IFM_10_T, mod, NULL);
2115 
2116 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, mod, NULL);
2117 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
2118 
2119 	} else if (phy->caps & SUPPORTED_TP) {
2120 		/* Copper (CX4) */
2121 
2122 		KASSERT(phy->caps & SUPPORTED_10000baseT_Full,
2123 			("%s: unexpected cap 0x%x", __func__, phy->caps));
2124 
2125 		ifmedia_add(media, m | IFM_10G_CX4, mod, NULL);
2126 		ifmedia_set(media, m | IFM_10G_CX4);
2127 
2128 	} else if (phy->caps & SUPPORTED_FIBRE &&
2129 		   phy->caps & SUPPORTED_10000baseT_Full) {
2130 		/* 10G optical (but includes SFP+ twinax) */
2131 
2132 		m |= cxgb_ifm_type(mod);
2133 		if (IFM_SUBTYPE(m) == IFM_NONE)
2134 			m &= ~IFM_FDX;
2135 
2136 		ifmedia_add(media, m, mod, NULL);
2137 		ifmedia_set(media, m);
2138 
2139 	} else if (phy->caps & SUPPORTED_FIBRE &&
2140 		   phy->caps & SUPPORTED_1000baseT_Full) {
2141 		/* 1G optical */
2142 
2143 		/* XXX: Lie and claim to be SX, could actually be any 1G-X */
2144 		ifmedia_add(media, m | IFM_1000_SX, mod, NULL);
2145 		ifmedia_set(media, m | IFM_1000_SX);
2146 
2147 	} else {
2148 		KASSERT(0, ("%s: don't know how to handle 0x%x.", __func__,
2149 			    phy->caps));
2150 	}
2151 
2152 	PORT_UNLOCK(p);
2153 }
2154 
2155 static void
2156 cxgb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2157 {
2158 	struct port_info *p = ifp->if_softc;
2159 	struct ifmedia_entry *cur = p->media.ifm_cur;
2160 	int speed = p->link_config.speed;
2161 
2162 	if (cur->ifm_data != p->phy.modtype) {
2163 		cxgb_build_medialist(p);
2164 		cur = p->media.ifm_cur;
2165 	}
2166 
2167 	ifmr->ifm_status = IFM_AVALID;
2168 	if (!p->link_config.link_ok)
2169 		return;
2170 
2171 	ifmr->ifm_status |= IFM_ACTIVE;
2172 
2173 	/*
2174 	 * active and current will differ iff current media is autoselect.  That
2175 	 * can happen only for copper RJ45.
2176 	 */
2177 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2178 		return;
2179 	KASSERT(p->phy.caps & SUPPORTED_TP && p->phy.caps & SUPPORTED_Autoneg,
2180 		("%s: unexpected PHY caps 0x%x", __func__, p->phy.caps));
2181 
2182 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2183 	if (speed == SPEED_10000)
2184 		ifmr->ifm_active |= IFM_10G_T;
2185 	else if (speed == SPEED_1000)
2186 		ifmr->ifm_active |= IFM_1000_T;
2187 	else if (speed == SPEED_100)
2188 		ifmr->ifm_active |= IFM_100_TX;
2189 	else if (speed == SPEED_10)
2190 		ifmr->ifm_active |= IFM_10_T;
2191 	else
2192 		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
2193 			    speed));
2194 }
2195 
2196 static void
2197 cxgb_async_intr(void *data)
2198 {
2199 	adapter_t *sc = data;
2200 
2201 	t3_write_reg(sc, A_PL_INT_ENABLE0, 0);
2202 	(void) t3_read_reg(sc, A_PL_INT_ENABLE0);
2203 	taskqueue_enqueue(sc->tq, &sc->slow_intr_task);
2204 }
2205 
2206 static void
2207 link_check_callout(void *arg)
2208 {
2209 	struct port_info *pi = arg;
2210 	struct adapter *sc = pi->adapter;
2211 
2212 	if (!isset(&sc->open_device_map, pi->port_id))
2213 		return;
2214 
2215 	taskqueue_enqueue(sc->tq, &pi->link_check_task);
2216 }
2217 
2218 static void
2219 check_link_status(void *arg, int pending)
2220 {
2221 	struct port_info *pi = arg;
2222 	struct adapter *sc = pi->adapter;
2223 
2224 	if (!isset(&sc->open_device_map, pi->port_id))
2225 		return;
2226 
2227 	t3_link_changed(sc, pi->port_id);
2228 
2229 	if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ))
2230 		callout_reset(&pi->link_check_ch, hz, link_check_callout, pi);
2231 }
2232 
2233 void
2234 t3_os_link_intr(struct port_info *pi)
2235 {
2236 	/*
2237 	 * Schedule a link check in the near future.  If the link is flapping
2238 	 * rapidly we'll keep resetting the callout and delaying the check until
2239 	 * things stabilize a bit.
2240 	 */
2241 	callout_reset(&pi->link_check_ch, hz / 4, link_check_callout, pi);
2242 }
2243 
2244 static void
2245 check_t3b2_mac(struct adapter *sc)
2246 {
2247 	int i;
2248 
2249 	if (sc->flags & CXGB_SHUTDOWN)
2250 		return;
2251 
2252 	for_each_port(sc, i) {
2253 		struct port_info *p = &sc->port[i];
2254 		int status;
2255 #ifdef INVARIANTS
2256 		struct ifnet *ifp = p->ifp;
2257 #endif
2258 
2259 		if (!isset(&sc->open_device_map, p->port_id) || p->link_fault ||
2260 		    !p->link_config.link_ok)
2261 			continue;
2262 
2263 		KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING,
2264 			("%s: state mismatch (drv_flags %x, device_map %x)",
2265 			 __func__, ifp->if_drv_flags, sc->open_device_map));
2266 
2267 		PORT_LOCK(p);
2268 		status = t3b2_mac_watchdog_task(&p->mac);
2269 		if (status == 1)
2270 			p->mac.stats.num_toggled++;
2271 		else if (status == 2) {
2272 			struct cmac *mac = &p->mac;
2273 
2274 			cxgb_update_mac_settings(p);
2275 			t3_link_start(&p->phy, mac, &p->link_config);
2276 			t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2277 			t3_port_intr_enable(sc, p->port_id);
2278 			p->mac.stats.num_resets++;
2279 		}
2280 		PORT_UNLOCK(p);
2281 	}
2282 }
2283 
2284 static void
2285 cxgb_tick(void *arg)
2286 {
2287 	adapter_t *sc = (adapter_t *)arg;
2288 
2289 	if (sc->flags & CXGB_SHUTDOWN)
2290 		return;
2291 
2292 	taskqueue_enqueue(sc->tq, &sc->tick_task);
2293 	callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
2294 }
2295 
2296 static void
2297 cxgb_tick_handler(void *arg, int count)
2298 {
2299 	adapter_t *sc = (adapter_t *)arg;
2300 	const struct adapter_params *p = &sc->params;
2301 	int i;
2302 	uint32_t cause, reset;
2303 
2304 	if (sc->flags & CXGB_SHUTDOWN || !(sc->flags & FULL_INIT_DONE))
2305 		return;
2306 
2307 	if (p->rev == T3_REV_B2 && p->nports < 4 && sc->open_device_map)
2308 		check_t3b2_mac(sc);
2309 
2310 	cause = t3_read_reg(sc, A_SG_INT_CAUSE) & (F_RSPQSTARVE | F_FLEMPTY);
2311 	if (cause) {
2312 		struct sge_qset *qs = &sc->sge.qs[0];
2313 		uint32_t mask, v;
2314 
2315 		v = t3_read_reg(sc, A_SG_RSPQ_FL_STATUS) & ~0xff00;
2316 
2317 		mask = 1;
2318 		for (i = 0; i < SGE_QSETS; i++) {
2319 			if (v & mask)
2320 				qs[i].rspq.starved++;
2321 			mask <<= 1;
2322 		}
2323 
2324 		mask <<= SGE_QSETS; /* skip RSPQXDISABLED */
2325 
2326 		for (i = 0; i < SGE_QSETS * 2; i++) {
2327 			if (v & mask) {
2328 				qs[i / 2].fl[i % 2].empty++;
2329 			}
2330 			mask <<= 1;
2331 		}
2332 
2333 		/* clear */
2334 		t3_write_reg(sc, A_SG_RSPQ_FL_STATUS, v);
2335 		t3_write_reg(sc, A_SG_INT_CAUSE, cause);
2336 	}
2337 
2338 	for (i = 0; i < sc->params.nports; i++) {
2339 		struct port_info *pi = &sc->port[i];
2340 		struct ifnet *ifp = pi->ifp;
2341 		struct cmac *mac = &pi->mac;
2342 		struct mac_stats *mstats = &mac->stats;
2343 		int drops, j;
2344 
2345 		if (!isset(&sc->open_device_map, pi->port_id))
2346 			continue;
2347 
2348 		PORT_LOCK(pi);
2349 		t3_mac_update_stats(mac);
2350 		PORT_UNLOCK(pi);
2351 
2352 		ifp->if_opackets = mstats->tx_frames;
2353 		ifp->if_ipackets = mstats->rx_frames;
2354 		ifp->if_obytes = mstats->tx_octets;
2355 		ifp->if_ibytes = mstats->rx_octets;
2356 		ifp->if_omcasts = mstats->tx_mcast_frames;
2357 		ifp->if_imcasts = mstats->rx_mcast_frames;
2358 		ifp->if_collisions = mstats->tx_total_collisions;
2359 		ifp->if_iqdrops = mstats->rx_cong_drops;
2360 
2361 		drops = 0;
2362 		for (j = pi->first_qset; j < pi->first_qset + pi->nqsets; j++)
2363 			drops += sc->sge.qs[j].txq[TXQ_ETH].txq_mr->br_drops;
2364 		ifp->if_snd.ifq_drops = drops;
2365 
2366 		ifp->if_oerrors =
2367 		    mstats->tx_excess_collisions +
2368 		    mstats->tx_underrun +
2369 		    mstats->tx_len_errs +
2370 		    mstats->tx_mac_internal_errs +
2371 		    mstats->tx_excess_deferral +
2372 		    mstats->tx_fcs_errs;
2373 		ifp->if_ierrors =
2374 		    mstats->rx_jabber +
2375 		    mstats->rx_data_errs +
2376 		    mstats->rx_sequence_errs +
2377 		    mstats->rx_runt +
2378 		    mstats->rx_too_long +
2379 		    mstats->rx_mac_internal_errs +
2380 		    mstats->rx_short +
2381 		    mstats->rx_fcs_errs;
2382 
2383 		if (mac->multiport)
2384 			continue;
2385 
2386 		/* Count rx fifo overflows, once per second */
2387 		cause = t3_read_reg(sc, A_XGM_INT_CAUSE + mac->offset);
2388 		reset = 0;
2389 		if (cause & F_RXFIFO_OVERFLOW) {
2390 			mac->stats.rx_fifo_ovfl++;
2391 			reset |= F_RXFIFO_OVERFLOW;
2392 		}
2393 		t3_write_reg(sc, A_XGM_INT_CAUSE + mac->offset, reset);
2394 	}
2395 }
2396 
2397 static void
2398 touch_bars(device_t dev)
2399 {
2400 	/*
2401 	 * Don't enable yet
2402 	 */
2403 #if !defined(__LP64__) && 0
2404 	u32 v;
2405 
2406 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_1, &v);
2407 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_1, v);
2408 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_3, &v);
2409 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_3, v);
2410 	pci_read_config_dword(pdev, PCI_BASE_ADDRESS_5, &v);
2411 	pci_write_config_dword(pdev, PCI_BASE_ADDRESS_5, v);
2412 #endif
2413 }
2414 
2415 static int
2416 set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset)
2417 {
2418 	uint8_t *buf;
2419 	int err = 0;
2420 	u32 aligned_offset, aligned_len, *p;
2421 	struct adapter *adapter = pi->adapter;
2422 
2423 
2424 	aligned_offset = offset & ~3;
2425 	aligned_len = (len + (offset & 3) + 3) & ~3;
2426 
2427 	if (aligned_offset != offset || aligned_len != len) {
2428 		buf = malloc(aligned_len, M_DEVBUF, M_WAITOK|M_ZERO);
2429 		if (!buf)
2430 			return (ENOMEM);
2431 		err = t3_seeprom_read(adapter, aligned_offset, (u32 *)buf);
2432 		if (!err && aligned_len > 4)
2433 			err = t3_seeprom_read(adapter,
2434 					      aligned_offset + aligned_len - 4,
2435 					      (u32 *)&buf[aligned_len - 4]);
2436 		if (err)
2437 			goto out;
2438 		memcpy(buf + (offset & 3), data, len);
2439 	} else
2440 		buf = (uint8_t *)(uintptr_t)data;
2441 
2442 	err = t3_seeprom_wp(adapter, 0);
2443 	if (err)
2444 		goto out;
2445 
2446 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2447 		err = t3_seeprom_write(adapter, aligned_offset, *p);
2448 		aligned_offset += 4;
2449 	}
2450 
2451 	if (!err)
2452 		err = t3_seeprom_wp(adapter, 1);
2453 out:
2454 	if (buf != data)
2455 		free(buf, M_DEVBUF);
2456 	return err;
2457 }
2458 
2459 
2460 static int
2461 in_range(int val, int lo, int hi)
2462 {
2463 	return val < 0 || (val <= hi && val >= lo);
2464 }
2465 
2466 static int
2467 cxgb_extension_open(struct cdev *dev, int flags, int fmp, struct thread *td)
2468 {
2469        return (0);
2470 }
2471 
2472 static int
2473 cxgb_extension_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2474 {
2475        return (0);
2476 }
2477 
2478 static int
2479 cxgb_extension_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data,
2480     int fflag, struct thread *td)
2481 {
2482 	int mmd, error = 0;
2483 	struct port_info *pi = dev->si_drv1;
2484 	adapter_t *sc = pi->adapter;
2485 
2486 #ifdef PRIV_SUPPORTED
2487 	if (priv_check(td, PRIV_DRIVER)) {
2488 		if (cxgb_debug)
2489 			printf("user does not have access to privileged ioctls\n");
2490 		return (EPERM);
2491 	}
2492 #else
2493 	if (suser(td)) {
2494 		if (cxgb_debug)
2495 			printf("user does not have access to privileged ioctls\n");
2496 		return (EPERM);
2497 	}
2498 #endif
2499 
2500 	switch (cmd) {
2501 	case CHELSIO_GET_MIIREG: {
2502 		uint32_t val;
2503 		struct cphy *phy = &pi->phy;
2504 		struct ch_mii_data *mid = (struct ch_mii_data *)data;
2505 
2506 		if (!phy->mdio_read)
2507 			return (EOPNOTSUPP);
2508 		if (is_10G(sc)) {
2509 			mmd = mid->phy_id >> 8;
2510 			if (!mmd)
2511 				mmd = MDIO_DEV_PCS;
2512 			else if (mmd > MDIO_DEV_VEND2)
2513 				return (EINVAL);
2514 
2515 			error = phy->mdio_read(sc, mid->phy_id & 0x1f, mmd,
2516 					     mid->reg_num, &val);
2517 		} else
2518 		        error = phy->mdio_read(sc, mid->phy_id & 0x1f, 0,
2519 					     mid->reg_num & 0x1f, &val);
2520 		if (error == 0)
2521 			mid->val_out = val;
2522 		break;
2523 	}
2524 	case CHELSIO_SET_MIIREG: {
2525 		struct cphy *phy = &pi->phy;
2526 		struct ch_mii_data *mid = (struct ch_mii_data *)data;
2527 
2528 		if (!phy->mdio_write)
2529 			return (EOPNOTSUPP);
2530 		if (is_10G(sc)) {
2531 			mmd = mid->phy_id >> 8;
2532 			if (!mmd)
2533 				mmd = MDIO_DEV_PCS;
2534 			else if (mmd > MDIO_DEV_VEND2)
2535 				return (EINVAL);
2536 
2537 			error = phy->mdio_write(sc, mid->phy_id & 0x1f,
2538 					      mmd, mid->reg_num, mid->val_in);
2539 		} else
2540 			error = phy->mdio_write(sc, mid->phy_id & 0x1f, 0,
2541 					      mid->reg_num & 0x1f,
2542 					      mid->val_in);
2543 		break;
2544 	}
2545 	case CHELSIO_SETREG: {
2546 		struct ch_reg *edata = (struct ch_reg *)data;
2547 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2548 			return (EFAULT);
2549 		t3_write_reg(sc, edata->addr, edata->val);
2550 		break;
2551 	}
2552 	case CHELSIO_GETREG: {
2553 		struct ch_reg *edata = (struct ch_reg *)data;
2554 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2555 			return (EFAULT);
2556 		edata->val = t3_read_reg(sc, edata->addr);
2557 		break;
2558 	}
2559 	case CHELSIO_GET_SGE_CONTEXT: {
2560 		struct ch_cntxt *ecntxt = (struct ch_cntxt *)data;
2561 		mtx_lock_spin(&sc->sge.reg_lock);
2562 		switch (ecntxt->cntxt_type) {
2563 		case CNTXT_TYPE_EGRESS:
2564 			error = -t3_sge_read_ecntxt(sc, ecntxt->cntxt_id,
2565 			    ecntxt->data);
2566 			break;
2567 		case CNTXT_TYPE_FL:
2568 			error = -t3_sge_read_fl(sc, ecntxt->cntxt_id,
2569 			    ecntxt->data);
2570 			break;
2571 		case CNTXT_TYPE_RSP:
2572 			error = -t3_sge_read_rspq(sc, ecntxt->cntxt_id,
2573 			    ecntxt->data);
2574 			break;
2575 		case CNTXT_TYPE_CQ:
2576 			error = -t3_sge_read_cq(sc, ecntxt->cntxt_id,
2577 			    ecntxt->data);
2578 			break;
2579 		default:
2580 			error = EINVAL;
2581 			break;
2582 		}
2583 		mtx_unlock_spin(&sc->sge.reg_lock);
2584 		break;
2585 	}
2586 	case CHELSIO_GET_SGE_DESC: {
2587 		struct ch_desc *edesc = (struct ch_desc *)data;
2588 		int ret;
2589 		if (edesc->queue_num >= SGE_QSETS * 6)
2590 			return (EINVAL);
2591 		ret = t3_get_desc(&sc->sge.qs[edesc->queue_num / 6],
2592 		    edesc->queue_num % 6, edesc->idx, edesc->data);
2593 		if (ret < 0)
2594 			return (EINVAL);
2595 		edesc->size = ret;
2596 		break;
2597 	}
2598 	case CHELSIO_GET_QSET_PARAMS: {
2599 		struct qset_params *q;
2600 		struct ch_qset_params *t = (struct ch_qset_params *)data;
2601 		int q1 = pi->first_qset;
2602 		int nqsets = pi->nqsets;
2603 		int i;
2604 
2605 		if (t->qset_idx >= nqsets)
2606 			return EINVAL;
2607 
2608 		i = q1 + t->qset_idx;
2609 		q = &sc->params.sge.qset[i];
2610 		t->rspq_size   = q->rspq_size;
2611 		t->txq_size[0] = q->txq_size[0];
2612 		t->txq_size[1] = q->txq_size[1];
2613 		t->txq_size[2] = q->txq_size[2];
2614 		t->fl_size[0]  = q->fl_size;
2615 		t->fl_size[1]  = q->jumbo_size;
2616 		t->polling     = q->polling;
2617 		t->lro         = q->lro;
2618 		t->intr_lat    = q->coalesce_usecs;
2619 		t->cong_thres  = q->cong_thres;
2620 		t->qnum        = i;
2621 
2622 		if ((sc->flags & FULL_INIT_DONE) == 0)
2623 			t->vector = 0;
2624 		else if (sc->flags & USING_MSIX)
2625 			t->vector = rman_get_start(sc->msix_irq_res[i]);
2626 		else
2627 			t->vector = rman_get_start(sc->irq_res);
2628 
2629 		break;
2630 	}
2631 	case CHELSIO_GET_QSET_NUM: {
2632 		struct ch_reg *edata = (struct ch_reg *)data;
2633 		edata->val = pi->nqsets;
2634 		break;
2635 	}
2636 	case CHELSIO_LOAD_FW: {
2637 		uint8_t *fw_data;
2638 		uint32_t vers;
2639 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2640 
2641 		/*
2642 		 * You're allowed to load a firmware only before FULL_INIT_DONE
2643 		 *
2644 		 * FW_UPTODATE is also set so the rest of the initialization
2645 		 * will not overwrite what was loaded here.  This gives you the
2646 		 * flexibility to load any firmware (and maybe shoot yourself in
2647 		 * the foot).
2648 		 */
2649 
2650 		ADAPTER_LOCK(sc);
2651 		if (sc->open_device_map || sc->flags & FULL_INIT_DONE) {
2652 			ADAPTER_UNLOCK(sc);
2653 			return (EBUSY);
2654 		}
2655 
2656 		fw_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2657 		if (!fw_data)
2658 			error = ENOMEM;
2659 		else
2660 			error = copyin(t->buf, fw_data, t->len);
2661 
2662 		if (!error)
2663 			error = -t3_load_fw(sc, fw_data, t->len);
2664 
2665 		if (t3_get_fw_version(sc, &vers) == 0) {
2666 			snprintf(&sc->fw_version[0], sizeof(sc->fw_version),
2667 			    "%d.%d.%d", G_FW_VERSION_MAJOR(vers),
2668 			    G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers));
2669 		}
2670 
2671 		if (!error)
2672 			sc->flags |= FW_UPTODATE;
2673 
2674 		free(fw_data, M_DEVBUF);
2675 		ADAPTER_UNLOCK(sc);
2676 		break;
2677 	}
2678 	case CHELSIO_LOAD_BOOT: {
2679 		uint8_t *boot_data;
2680 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2681 
2682 		boot_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2683 		if (!boot_data)
2684 			return ENOMEM;
2685 
2686 		error = copyin(t->buf, boot_data, t->len);
2687 		if (!error)
2688 			error = -t3_load_boot(sc, boot_data, t->len);
2689 
2690 		free(boot_data, M_DEVBUF);
2691 		break;
2692 	}
2693 	case CHELSIO_GET_PM: {
2694 		struct ch_pm *m = (struct ch_pm *)data;
2695 		struct tp_params *p = &sc->params.tp;
2696 
2697 		if (!is_offload(sc))
2698 			return (EOPNOTSUPP);
2699 
2700 		m->tx_pg_sz = p->tx_pg_size;
2701 		m->tx_num_pg = p->tx_num_pgs;
2702 		m->rx_pg_sz  = p->rx_pg_size;
2703 		m->rx_num_pg = p->rx_num_pgs;
2704 		m->pm_total  = p->pmtx_size + p->chan_rx_size * p->nchan;
2705 
2706 		break;
2707 	}
2708 	case CHELSIO_SET_PM: {
2709 		struct ch_pm *m = (struct ch_pm *)data;
2710 		struct tp_params *p = &sc->params.tp;
2711 
2712 		if (!is_offload(sc))
2713 			return (EOPNOTSUPP);
2714 		if (sc->flags & FULL_INIT_DONE)
2715 			return (EBUSY);
2716 
2717 		if (!m->rx_pg_sz || (m->rx_pg_sz & (m->rx_pg_sz - 1)) ||
2718 		    !m->tx_pg_sz || (m->tx_pg_sz & (m->tx_pg_sz - 1)))
2719 			return (EINVAL);	/* not power of 2 */
2720 		if (!(m->rx_pg_sz & 0x14000))
2721 			return (EINVAL);	/* not 16KB or 64KB */
2722 		if (!(m->tx_pg_sz & 0x1554000))
2723 			return (EINVAL);
2724 		if (m->tx_num_pg == -1)
2725 			m->tx_num_pg = p->tx_num_pgs;
2726 		if (m->rx_num_pg == -1)
2727 			m->rx_num_pg = p->rx_num_pgs;
2728 		if (m->tx_num_pg % 24 || m->rx_num_pg % 24)
2729 			return (EINVAL);
2730 		if (m->rx_num_pg * m->rx_pg_sz > p->chan_rx_size ||
2731 		    m->tx_num_pg * m->tx_pg_sz > p->chan_tx_size)
2732 			return (EINVAL);
2733 
2734 		p->rx_pg_size = m->rx_pg_sz;
2735 		p->tx_pg_size = m->tx_pg_sz;
2736 		p->rx_num_pgs = m->rx_num_pg;
2737 		p->tx_num_pgs = m->tx_num_pg;
2738 		break;
2739 	}
2740 	case CHELSIO_SETMTUTAB: {
2741 		struct ch_mtus *m = (struct ch_mtus *)data;
2742 		int i;
2743 
2744 		if (!is_offload(sc))
2745 			return (EOPNOTSUPP);
2746 		if (offload_running(sc))
2747 			return (EBUSY);
2748 		if (m->nmtus != NMTUS)
2749 			return (EINVAL);
2750 		if (m->mtus[0] < 81)         /* accommodate SACK */
2751 			return (EINVAL);
2752 
2753 		/*
2754 		 * MTUs must be in ascending order
2755 		 */
2756 		for (i = 1; i < NMTUS; ++i)
2757 			if (m->mtus[i] < m->mtus[i - 1])
2758 				return (EINVAL);
2759 
2760 		memcpy(sc->params.mtus, m->mtus, sizeof(sc->params.mtus));
2761 		break;
2762 	}
2763 	case CHELSIO_GETMTUTAB: {
2764 		struct ch_mtus *m = (struct ch_mtus *)data;
2765 
2766 		if (!is_offload(sc))
2767 			return (EOPNOTSUPP);
2768 
2769 		memcpy(m->mtus, sc->params.mtus, sizeof(m->mtus));
2770 		m->nmtus = NMTUS;
2771 		break;
2772 	}
2773 	case CHELSIO_GET_MEM: {
2774 		struct ch_mem_range *t = (struct ch_mem_range *)data;
2775 		struct mc7 *mem;
2776 		uint8_t *useraddr;
2777 		u64 buf[32];
2778 
2779 		/*
2780 		 * Use these to avoid modifying len/addr in the return
2781 		 * struct
2782 		 */
2783 		uint32_t len = t->len, addr = t->addr;
2784 
2785 		if (!is_offload(sc))
2786 			return (EOPNOTSUPP);
2787 		if (!(sc->flags & FULL_INIT_DONE))
2788 			return (EIO);         /* need the memory controllers */
2789 		if ((addr & 0x7) || (len & 0x7))
2790 			return (EINVAL);
2791 		if (t->mem_id == MEM_CM)
2792 			mem = &sc->cm;
2793 		else if (t->mem_id == MEM_PMRX)
2794 			mem = &sc->pmrx;
2795 		else if (t->mem_id == MEM_PMTX)
2796 			mem = &sc->pmtx;
2797 		else
2798 			return (EINVAL);
2799 
2800 		/*
2801 		 * Version scheme:
2802 		 * bits 0..9: chip version
2803 		 * bits 10..15: chip revision
2804 		 */
2805 		t->version = 3 | (sc->params.rev << 10);
2806 
2807 		/*
2808 		 * Read 256 bytes at a time as len can be large and we don't
2809 		 * want to use huge intermediate buffers.
2810 		 */
2811 		useraddr = (uint8_t *)t->buf;
2812 		while (len) {
2813 			unsigned int chunk = min(len, sizeof(buf));
2814 
2815 			error = t3_mc7_bd_read(mem, addr / 8, chunk / 8, buf);
2816 			if (error)
2817 				return (-error);
2818 			if (copyout(buf, useraddr, chunk))
2819 				return (EFAULT);
2820 			useraddr += chunk;
2821 			addr += chunk;
2822 			len -= chunk;
2823 		}
2824 		break;
2825 	}
2826 	case CHELSIO_READ_TCAM_WORD: {
2827 		struct ch_tcam_word *t = (struct ch_tcam_word *)data;
2828 
2829 		if (!is_offload(sc))
2830 			return (EOPNOTSUPP);
2831 		if (!(sc->flags & FULL_INIT_DONE))
2832 			return (EIO);         /* need MC5 */
2833 		return -t3_read_mc5_range(&sc->mc5, t->addr, 1, t->buf);
2834 		break;
2835 	}
2836 	case CHELSIO_SET_TRACE_FILTER: {
2837 		struct ch_trace *t = (struct ch_trace *)data;
2838 		const struct trace_params *tp;
2839 
2840 		tp = (const struct trace_params *)&t->sip;
2841 		if (t->config_tx)
2842 			t3_config_trace_filter(sc, tp, 0, t->invert_match,
2843 					       t->trace_tx);
2844 		if (t->config_rx)
2845 			t3_config_trace_filter(sc, tp, 1, t->invert_match,
2846 					       t->trace_rx);
2847 		break;
2848 	}
2849 	case CHELSIO_SET_PKTSCHED: {
2850 		struct ch_pktsched_params *p = (struct ch_pktsched_params *)data;
2851 		if (sc->open_device_map == 0)
2852 			return (EAGAIN);
2853 		send_pktsched_cmd(sc, p->sched, p->idx, p->min, p->max,
2854 		    p->binding);
2855 		break;
2856 	}
2857 	case CHELSIO_IFCONF_GETREGS: {
2858 		struct ch_ifconf_regs *regs = (struct ch_ifconf_regs *)data;
2859 		int reglen = cxgb_get_regs_len();
2860 		uint8_t *buf = malloc(reglen, M_DEVBUF, M_NOWAIT);
2861 		if (buf == NULL) {
2862 			return (ENOMEM);
2863 		}
2864 		if (regs->len > reglen)
2865 			regs->len = reglen;
2866 		else if (regs->len < reglen)
2867 			error = ENOBUFS;
2868 
2869 		if (!error) {
2870 			cxgb_get_regs(sc, regs, buf);
2871 			error = copyout(buf, regs->data, reglen);
2872 		}
2873 		free(buf, M_DEVBUF);
2874 
2875 		break;
2876 	}
2877 	case CHELSIO_SET_HW_SCHED: {
2878 		struct ch_hw_sched *t = (struct ch_hw_sched *)data;
2879 		unsigned int ticks_per_usec = core_ticks_per_usec(sc);
2880 
2881 		if ((sc->flags & FULL_INIT_DONE) == 0)
2882 			return (EAGAIN);       /* need TP to be initialized */
2883 		if (t->sched >= NTX_SCHED || !in_range(t->mode, 0, 1) ||
2884 		    !in_range(t->channel, 0, 1) ||
2885 		    !in_range(t->kbps, 0, 10000000) ||
2886 		    !in_range(t->class_ipg, 0, 10000 * 65535 / ticks_per_usec) ||
2887 		    !in_range(t->flow_ipg, 0,
2888 			      dack_ticks_to_usec(sc, 0x7ff)))
2889 			return (EINVAL);
2890 
2891 		if (t->kbps >= 0) {
2892 			error = t3_config_sched(sc, t->kbps, t->sched);
2893 			if (error < 0)
2894 				return (-error);
2895 		}
2896 		if (t->class_ipg >= 0)
2897 			t3_set_sched_ipg(sc, t->sched, t->class_ipg);
2898 		if (t->flow_ipg >= 0) {
2899 			t->flow_ipg *= 1000;     /* us -> ns */
2900 			t3_set_pace_tbl(sc, &t->flow_ipg, t->sched, 1);
2901 		}
2902 		if (t->mode >= 0) {
2903 			int bit = 1 << (S_TX_MOD_TIMER_MODE + t->sched);
2904 
2905 			t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2906 					 bit, t->mode ? bit : 0);
2907 		}
2908 		if (t->channel >= 0)
2909 			t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2910 					 1 << t->sched, t->channel << t->sched);
2911 		break;
2912 	}
2913 	case CHELSIO_GET_EEPROM: {
2914 		int i;
2915 		struct ch_eeprom *e = (struct ch_eeprom *)data;
2916 		uint8_t *buf = malloc(EEPROMSIZE, M_DEVBUF, M_NOWAIT);
2917 
2918 		if (buf == NULL) {
2919 			return (ENOMEM);
2920 		}
2921 		e->magic = EEPROM_MAGIC;
2922 		for (i = e->offset & ~3; !error && i < e->offset + e->len; i += 4)
2923 			error = -t3_seeprom_read(sc, i, (uint32_t *)&buf[i]);
2924 
2925 		if (!error)
2926 			error = copyout(buf + e->offset, e->data, e->len);
2927 
2928 		free(buf, M_DEVBUF);
2929 		break;
2930 	}
2931 	case CHELSIO_CLEAR_STATS: {
2932 		if (!(sc->flags & FULL_INIT_DONE))
2933 			return EAGAIN;
2934 
2935 		PORT_LOCK(pi);
2936 		t3_mac_update_stats(&pi->mac);
2937 		memset(&pi->mac.stats, 0, sizeof(pi->mac.stats));
2938 		PORT_UNLOCK(pi);
2939 		break;
2940 	}
2941 	case CHELSIO_GET_UP_LA: {
2942 		struct ch_up_la *la = (struct ch_up_la *)data;
2943 		uint8_t *buf = malloc(LA_BUFSIZE, M_DEVBUF, M_NOWAIT);
2944 		if (buf == NULL) {
2945 			return (ENOMEM);
2946 		}
2947 		if (la->bufsize < LA_BUFSIZE)
2948 			error = ENOBUFS;
2949 
2950 		if (!error)
2951 			error = -t3_get_up_la(sc, &la->stopped, &la->idx,
2952 					      &la->bufsize, buf);
2953 		if (!error)
2954 			error = copyout(buf, la->data, la->bufsize);
2955 
2956 		free(buf, M_DEVBUF);
2957 		break;
2958 	}
2959 	case CHELSIO_GET_UP_IOQS: {
2960 		struct ch_up_ioqs *ioqs = (struct ch_up_ioqs *)data;
2961 		uint8_t *buf = malloc(IOQS_BUFSIZE, M_DEVBUF, M_NOWAIT);
2962 		uint32_t *v;
2963 
2964 		if (buf == NULL) {
2965 			return (ENOMEM);
2966 		}
2967 		if (ioqs->bufsize < IOQS_BUFSIZE)
2968 			error = ENOBUFS;
2969 
2970 		if (!error)
2971 			error = -t3_get_up_ioqs(sc, &ioqs->bufsize, buf);
2972 
2973 		if (!error) {
2974 			v = (uint32_t *)buf;
2975 
2976 			ioqs->ioq_rx_enable = *v++;
2977 			ioqs->ioq_tx_enable = *v++;
2978 			ioqs->ioq_rx_status = *v++;
2979 			ioqs->ioq_tx_status = *v++;
2980 
2981 			error = copyout(v, ioqs->data, ioqs->bufsize);
2982 		}
2983 
2984 		free(buf, M_DEVBUF);
2985 		break;
2986 	}
2987 	case CHELSIO_SET_FILTER: {
2988 		struct ch_filter *f = (struct ch_filter *)data;
2989 		struct filter_info *p;
2990 		unsigned int nfilters = sc->params.mc5.nfilters;
2991 
2992 		if (!is_offload(sc))
2993 			return (EOPNOTSUPP);	/* No TCAM */
2994 		if (!(sc->flags & FULL_INIT_DONE))
2995 			return (EAGAIN);	/* mc5 not setup yet */
2996 		if (nfilters == 0)
2997 			return (EBUSY);		/* TOE will use TCAM */
2998 
2999 		/* sanity checks */
3000 		if (f->filter_id >= nfilters ||
3001 		    (f->val.dip && f->mask.dip != 0xffffffff) ||
3002 		    (f->val.sport && f->mask.sport != 0xffff) ||
3003 		    (f->val.dport && f->mask.dport != 0xffff) ||
3004 		    (f->val.vlan && f->mask.vlan != 0xfff) ||
3005 		    (f->val.vlan_prio &&
3006 			f->mask.vlan_prio != FILTER_NO_VLAN_PRI) ||
3007 		    (f->mac_addr_idx != 0xffff && f->mac_addr_idx > 15) ||
3008 		    f->qset >= SGE_QSETS ||
3009 		    sc->rrss_map[f->qset] >= RSS_TABLE_SIZE)
3010 			return (EINVAL);
3011 
3012 		/* Was allocated with M_WAITOK */
3013 		KASSERT(sc->filters, ("filter table NULL\n"));
3014 
3015 		p = &sc->filters[f->filter_id];
3016 		if (p->locked)
3017 			return (EPERM);
3018 
3019 		bzero(p, sizeof(*p));
3020 		p->sip = f->val.sip;
3021 		p->sip_mask = f->mask.sip;
3022 		p->dip = f->val.dip;
3023 		p->sport = f->val.sport;
3024 		p->dport = f->val.dport;
3025 		p->vlan = f->mask.vlan ? f->val.vlan : 0xfff;
3026 		p->vlan_prio = f->mask.vlan_prio ? (f->val.vlan_prio & 6) :
3027 		    FILTER_NO_VLAN_PRI;
3028 		p->mac_hit = f->mac_hit;
3029 		p->mac_vld = f->mac_addr_idx != 0xffff;
3030 		p->mac_idx = f->mac_addr_idx;
3031 		p->pkt_type = f->proto;
3032 		p->report_filter_id = f->want_filter_id;
3033 		p->pass = f->pass;
3034 		p->rss = f->rss;
3035 		p->qset = f->qset;
3036 
3037 		error = set_filter(sc, f->filter_id, p);
3038 		if (error == 0)
3039 			p->valid = 1;
3040 		break;
3041 	}
3042 	case CHELSIO_DEL_FILTER: {
3043 		struct ch_filter *f = (struct ch_filter *)data;
3044 		struct filter_info *p;
3045 		unsigned int nfilters = sc->params.mc5.nfilters;
3046 
3047 		if (!is_offload(sc))
3048 			return (EOPNOTSUPP);
3049 		if (!(sc->flags & FULL_INIT_DONE))
3050 			return (EAGAIN);
3051 		if (nfilters == 0 || sc->filters == NULL)
3052 			return (EINVAL);
3053 		if (f->filter_id >= nfilters)
3054 		       return (EINVAL);
3055 
3056 		p = &sc->filters[f->filter_id];
3057 		if (p->locked)
3058 			return (EPERM);
3059 		if (!p->valid)
3060 			return (EFAULT); /* Read "Bad address" as "Bad index" */
3061 
3062 		bzero(p, sizeof(*p));
3063 		p->sip = p->sip_mask = 0xffffffff;
3064 		p->vlan = 0xfff;
3065 		p->vlan_prio = FILTER_NO_VLAN_PRI;
3066 		p->pkt_type = 1;
3067 		error = set_filter(sc, f->filter_id, p);
3068 		break;
3069 	}
3070 	case CHELSIO_GET_FILTER: {
3071 		struct ch_filter *f = (struct ch_filter *)data;
3072 		struct filter_info *p;
3073 		unsigned int i, nfilters = sc->params.mc5.nfilters;
3074 
3075 		if (!is_offload(sc))
3076 			return (EOPNOTSUPP);
3077 		if (!(sc->flags & FULL_INIT_DONE))
3078 			return (EAGAIN);
3079 		if (nfilters == 0 || sc->filters == NULL)
3080 			return (EINVAL);
3081 
3082 		i = f->filter_id == 0xffffffff ? 0 : f->filter_id + 1;
3083 		for (; i < nfilters; i++) {
3084 			p = &sc->filters[i];
3085 			if (!p->valid)
3086 				continue;
3087 
3088 			bzero(f, sizeof(*f));
3089 
3090 			f->filter_id = i;
3091 			f->val.sip = p->sip;
3092 			f->mask.sip = p->sip_mask;
3093 			f->val.dip = p->dip;
3094 			f->mask.dip = p->dip ? 0xffffffff : 0;
3095 			f->val.sport = p->sport;
3096 			f->mask.sport = p->sport ? 0xffff : 0;
3097 			f->val.dport = p->dport;
3098 			f->mask.dport = p->dport ? 0xffff : 0;
3099 			f->val.vlan = p->vlan == 0xfff ? 0 : p->vlan;
3100 			f->mask.vlan = p->vlan == 0xfff ? 0 : 0xfff;
3101 			f->val.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3102 			    0 : p->vlan_prio;
3103 			f->mask.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3104 			    0 : FILTER_NO_VLAN_PRI;
3105 			f->mac_hit = p->mac_hit;
3106 			f->mac_addr_idx = p->mac_vld ? p->mac_idx : 0xffff;
3107 			f->proto = p->pkt_type;
3108 			f->want_filter_id = p->report_filter_id;
3109 			f->pass = p->pass;
3110 			f->rss = p->rss;
3111 			f->qset = p->qset;
3112 
3113 			break;
3114 		}
3115 
3116 		if (i == nfilters)
3117 			f->filter_id = 0xffffffff;
3118 		break;
3119 	}
3120 	default:
3121 		return (EOPNOTSUPP);
3122 		break;
3123 	}
3124 
3125 	return (error);
3126 }
3127 
3128 static __inline void
3129 reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
3130     unsigned int end)
3131 {
3132 	uint32_t *p = (uint32_t *)(buf + start);
3133 
3134 	for ( ; start <= end; start += sizeof(uint32_t))
3135 		*p++ = t3_read_reg(ap, start);
3136 }
3137 
3138 #define T3_REGMAP_SIZE (3 * 1024)
3139 static int
3140 cxgb_get_regs_len(void)
3141 {
3142 	return T3_REGMAP_SIZE;
3143 }
3144 
3145 static void
3146 cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf)
3147 {
3148 
3149 	/*
3150 	 * Version scheme:
3151 	 * bits 0..9: chip version
3152 	 * bits 10..15: chip revision
3153 	 * bit 31: set for PCIe cards
3154 	 */
3155 	regs->version = 3 | (sc->params.rev << 10) | (is_pcie(sc) << 31);
3156 
3157 	/*
3158 	 * We skip the MAC statistics registers because they are clear-on-read.
3159 	 * Also reading multi-register stats would need to synchronize with the
3160 	 * periodic mac stats accumulation.  Hard to justify the complexity.
3161 	 */
3162 	memset(buf, 0, cxgb_get_regs_len());
3163 	reg_block_dump(sc, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
3164 	reg_block_dump(sc, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
3165 	reg_block_dump(sc, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
3166 	reg_block_dump(sc, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
3167 	reg_block_dump(sc, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
3168 	reg_block_dump(sc, buf, A_XGM_SERDES_STATUS0,
3169 		       XGM_REG(A_XGM_SERDES_STAT3, 1));
3170 	reg_block_dump(sc, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
3171 		       XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
3172 }
3173 
3174 static int
3175 alloc_filters(struct adapter *sc)
3176 {
3177 	struct filter_info *p;
3178 	unsigned int nfilters = sc->params.mc5.nfilters;
3179 
3180 	if (nfilters == 0)
3181 		return (0);
3182 
3183 	p = malloc(sizeof(*p) * nfilters, M_DEVBUF, M_WAITOK | M_ZERO);
3184 	sc->filters = p;
3185 
3186 	p = &sc->filters[nfilters - 1];
3187 	p->vlan = 0xfff;
3188 	p->vlan_prio = FILTER_NO_VLAN_PRI;
3189 	p->pass = p->rss = p->valid = p->locked = 1;
3190 
3191 	return (0);
3192 }
3193 
3194 static int
3195 setup_hw_filters(struct adapter *sc)
3196 {
3197 	int i, rc;
3198 	unsigned int nfilters = sc->params.mc5.nfilters;
3199 
3200 	if (!sc->filters)
3201 		return (0);
3202 
3203 	t3_enable_filters(sc);
3204 
3205 	for (i = rc = 0; i < nfilters && !rc; i++) {
3206 		if (sc->filters[i].locked)
3207 			rc = set_filter(sc, i, &sc->filters[i]);
3208 	}
3209 
3210 	return (rc);
3211 }
3212 
3213 static int
3214 set_filter(struct adapter *sc, int id, const struct filter_info *f)
3215 {
3216 	int len;
3217 	struct mbuf *m;
3218 	struct ulp_txpkt *txpkt;
3219 	struct work_request_hdr *wr;
3220 	struct cpl_pass_open_req *oreq;
3221 	struct cpl_set_tcb_field *sreq;
3222 
3223 	len = sizeof(*wr) + sizeof(*oreq) + 2 * sizeof(*sreq);
3224 	KASSERT(len <= MHLEN, ("filter request too big for an mbuf"));
3225 
3226 	id += t3_mc5_size(&sc->mc5) - sc->params.mc5.nroutes -
3227 	      sc->params.mc5.nfilters;
3228 
3229 	m = m_gethdr(M_WAITOK, MT_DATA);
3230 	m->m_len = m->m_pkthdr.len = len;
3231 	bzero(mtod(m, char *), len);
3232 
3233 	wr = mtod(m, struct work_request_hdr *);
3234 	wr->wrh_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS) | F_WR_ATOMIC);
3235 
3236 	oreq = (struct cpl_pass_open_req *)(wr + 1);
3237 	txpkt = (struct ulp_txpkt *)oreq;
3238 	txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3239 	txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*oreq) / 8));
3240 	OPCODE_TID(oreq) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, id));
3241 	oreq->local_port = htons(f->dport);
3242 	oreq->peer_port = htons(f->sport);
3243 	oreq->local_ip = htonl(f->dip);
3244 	oreq->peer_ip = htonl(f->sip);
3245 	oreq->peer_netmask = htonl(f->sip_mask);
3246 	oreq->opt0h = 0;
3247 	oreq->opt0l = htonl(F_NO_OFFLOAD);
3248 	oreq->opt1 = htonl(V_MAC_MATCH_VALID(f->mac_vld) |
3249 			 V_CONN_POLICY(CPL_CONN_POLICY_FILTER) |
3250 			 V_VLAN_PRI(f->vlan_prio >> 1) |
3251 			 V_VLAN_PRI_VALID(f->vlan_prio != FILTER_NO_VLAN_PRI) |
3252 			 V_PKT_TYPE(f->pkt_type) | V_OPT1_VLAN(f->vlan) |
3253 			 V_MAC_MATCH(f->mac_idx | (f->mac_hit << 4)));
3254 
3255 	sreq = (struct cpl_set_tcb_field *)(oreq + 1);
3256 	set_tcb_field_ulp(sreq, id, 1, 0x1800808000ULL,
3257 			  (f->report_filter_id << 15) | (1 << 23) |
3258 			  ((u64)f->pass << 35) | ((u64)!f->rss << 36));
3259 	set_tcb_field_ulp(sreq + 1, id, 0, 0xffffffff, (2 << 19) | 1);
3260 	t3_mgmt_tx(sc, m);
3261 
3262 	if (f->pass && !f->rss) {
3263 		len = sizeof(*sreq);
3264 		m = m_gethdr(M_WAITOK, MT_DATA);
3265 		m->m_len = m->m_pkthdr.len = len;
3266 		bzero(mtod(m, char *), len);
3267 		sreq = mtod(m, struct cpl_set_tcb_field *);
3268 		sreq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
3269 		mk_set_tcb_field(sreq, id, 25, 0x3f80000,
3270 				 (u64)sc->rrss_map[f->qset] << 19);
3271 		t3_mgmt_tx(sc, m);
3272 	}
3273 	return 0;
3274 }
3275 
3276 static inline void
3277 mk_set_tcb_field(struct cpl_set_tcb_field *req, unsigned int tid,
3278     unsigned int word, u64 mask, u64 val)
3279 {
3280 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
3281 	req->reply = V_NO_REPLY(1);
3282 	req->cpu_idx = 0;
3283 	req->word = htons(word);
3284 	req->mask = htobe64(mask);
3285 	req->val = htobe64(val);
3286 }
3287 
3288 static inline void
3289 set_tcb_field_ulp(struct cpl_set_tcb_field *req, unsigned int tid,
3290     unsigned int word, u64 mask, u64 val)
3291 {
3292 	struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req;
3293 
3294 	txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3295 	txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*req) / 8));
3296 	mk_set_tcb_field(req, tid, word, mask, val);
3297 }
3298 
3299 void
3300 t3_iterate(void (*func)(struct adapter *, void *), void *arg)
3301 {
3302 	struct adapter *sc;
3303 
3304 	mtx_lock(&t3_list_lock);
3305 	SLIST_FOREACH(sc, &t3_list, link) {
3306 		/*
3307 		 * func should not make any assumptions about what state sc is
3308 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
3309 		 */
3310 		func(sc, arg);
3311 	}
3312 	mtx_unlock(&t3_list_lock);
3313 }
3314 
3315 #ifdef TCP_OFFLOAD
3316 static int
3317 toe_capability(struct port_info *pi, int enable)
3318 {
3319 	int rc;
3320 	struct adapter *sc = pi->adapter;
3321 
3322 	ADAPTER_LOCK_ASSERT_OWNED(sc);
3323 
3324 	if (!is_offload(sc))
3325 		return (ENODEV);
3326 
3327 	if (enable) {
3328 		if (!(sc->flags & FULL_INIT_DONE)) {
3329 			log(LOG_WARNING,
3330 			    "You must enable a cxgb interface first\n");
3331 			return (EAGAIN);
3332 		}
3333 
3334 		if (isset(&sc->offload_map, pi->port_id))
3335 			return (0);
3336 
3337 		if (!(sc->flags & TOM_INIT_DONE)) {
3338 			rc = t3_activate_uld(sc, ULD_TOM);
3339 			if (rc == EAGAIN) {
3340 				log(LOG_WARNING,
3341 				    "You must kldload t3_tom.ko before trying "
3342 				    "to enable TOE on a cxgb interface.\n");
3343 			}
3344 			if (rc != 0)
3345 				return (rc);
3346 			KASSERT(sc->tom_softc != NULL,
3347 			    ("%s: TOM activated but softc NULL", __func__));
3348 			KASSERT(sc->flags & TOM_INIT_DONE,
3349 			    ("%s: TOM activated but flag not set", __func__));
3350 		}
3351 
3352 		setbit(&sc->offload_map, pi->port_id);
3353 
3354 		/*
3355 		 * XXX: Temporary code to allow iWARP to be enabled when TOE is
3356 		 * enabled on any port.  Need to figure out how to enable,
3357 		 * disable, load, and unload iWARP cleanly.
3358 		 */
3359 		if (!isset(&sc->offload_map, MAX_NPORTS) &&
3360 		    t3_activate_uld(sc, ULD_IWARP) == 0)
3361 			setbit(&sc->offload_map, MAX_NPORTS);
3362 	} else {
3363 		if (!isset(&sc->offload_map, pi->port_id))
3364 			return (0);
3365 
3366 		KASSERT(sc->flags & TOM_INIT_DONE,
3367 		    ("%s: TOM never initialized?", __func__));
3368 		clrbit(&sc->offload_map, pi->port_id);
3369 	}
3370 
3371 	return (0);
3372 }
3373 
3374 /*
3375  * Add an upper layer driver to the global list.
3376  */
3377 int
3378 t3_register_uld(struct uld_info *ui)
3379 {
3380 	int rc = 0;
3381 	struct uld_info *u;
3382 
3383 	mtx_lock(&t3_uld_list_lock);
3384 	SLIST_FOREACH(u, &t3_uld_list, link) {
3385 	    if (u->uld_id == ui->uld_id) {
3386 		    rc = EEXIST;
3387 		    goto done;
3388 	    }
3389 	}
3390 
3391 	SLIST_INSERT_HEAD(&t3_uld_list, ui, link);
3392 	ui->refcount = 0;
3393 done:
3394 	mtx_unlock(&t3_uld_list_lock);
3395 	return (rc);
3396 }
3397 
3398 int
3399 t3_unregister_uld(struct uld_info *ui)
3400 {
3401 	int rc = EINVAL;
3402 	struct uld_info *u;
3403 
3404 	mtx_lock(&t3_uld_list_lock);
3405 
3406 	SLIST_FOREACH(u, &t3_uld_list, link) {
3407 	    if (u == ui) {
3408 		    if (ui->refcount > 0) {
3409 			    rc = EBUSY;
3410 			    goto done;
3411 		    }
3412 
3413 		    SLIST_REMOVE(&t3_uld_list, ui, uld_info, link);
3414 		    rc = 0;
3415 		    goto done;
3416 	    }
3417 	}
3418 done:
3419 	mtx_unlock(&t3_uld_list_lock);
3420 	return (rc);
3421 }
3422 
3423 int
3424 t3_activate_uld(struct adapter *sc, int id)
3425 {
3426 	int rc = EAGAIN;
3427 	struct uld_info *ui;
3428 
3429 	mtx_lock(&t3_uld_list_lock);
3430 
3431 	SLIST_FOREACH(ui, &t3_uld_list, link) {
3432 		if (ui->uld_id == id) {
3433 			rc = ui->activate(sc);
3434 			if (rc == 0)
3435 				ui->refcount++;
3436 			goto done;
3437 		}
3438 	}
3439 done:
3440 	mtx_unlock(&t3_uld_list_lock);
3441 
3442 	return (rc);
3443 }
3444 
3445 int
3446 t3_deactivate_uld(struct adapter *sc, int id)
3447 {
3448 	int rc = EINVAL;
3449 	struct uld_info *ui;
3450 
3451 	mtx_lock(&t3_uld_list_lock);
3452 
3453 	SLIST_FOREACH(ui, &t3_uld_list, link) {
3454 		if (ui->uld_id == id) {
3455 			rc = ui->deactivate(sc);
3456 			if (rc == 0)
3457 				ui->refcount--;
3458 			goto done;
3459 		}
3460 	}
3461 done:
3462 	mtx_unlock(&t3_uld_list_lock);
3463 
3464 	return (rc);
3465 }
3466 
3467 static int
3468 cpl_not_handled(struct sge_qset *qs __unused, struct rsp_desc *r __unused,
3469     struct mbuf *m)
3470 {
3471 	m_freem(m);
3472 	return (EDOOFUS);
3473 }
3474 
3475 int
3476 t3_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
3477 {
3478 	uintptr_t *loc, new;
3479 
3480 	if (opcode >= NUM_CPL_HANDLERS)
3481 		return (EINVAL);
3482 
3483 	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
3484 	loc = (uintptr_t *) &sc->cpl_handler[opcode];
3485 	atomic_store_rel_ptr(loc, new);
3486 
3487 	return (0);
3488 }
3489 #endif
3490 
3491 static int
3492 cxgbc_mod_event(module_t mod, int cmd, void *arg)
3493 {
3494 	int rc = 0;
3495 
3496 	switch (cmd) {
3497 	case MOD_LOAD:
3498 		mtx_init(&t3_list_lock, "T3 adapters", 0, MTX_DEF);
3499 		SLIST_INIT(&t3_list);
3500 #ifdef TCP_OFFLOAD
3501 		mtx_init(&t3_uld_list_lock, "T3 ULDs", 0, MTX_DEF);
3502 		SLIST_INIT(&t3_uld_list);
3503 #endif
3504 		break;
3505 
3506 	case MOD_UNLOAD:
3507 #ifdef TCP_OFFLOAD
3508 		mtx_lock(&t3_uld_list_lock);
3509 		if (!SLIST_EMPTY(&t3_uld_list)) {
3510 			rc = EBUSY;
3511 			mtx_unlock(&t3_uld_list_lock);
3512 			break;
3513 		}
3514 		mtx_unlock(&t3_uld_list_lock);
3515 		mtx_destroy(&t3_uld_list_lock);
3516 #endif
3517 		mtx_lock(&t3_list_lock);
3518 		if (!SLIST_EMPTY(&t3_list)) {
3519 			rc = EBUSY;
3520 			mtx_unlock(&t3_list_lock);
3521 			break;
3522 		}
3523 		mtx_unlock(&t3_list_lock);
3524 		mtx_destroy(&t3_list_lock);
3525 		break;
3526 	}
3527 
3528 	return (rc);
3529 }
3530