xref: /freebsd/sys/dev/cxgbe/t4_vf.c (revision 94cba8034ba53725c225c85e35724f0c2b13cea5)
1 /*-
2  * Copyright (c) 2016 Chelsio Communications, Inc.
3  * All rights reserved.
4  * Written by: John Baldwin <jhb@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/bus.h>
36 #include <sys/conf.h>
37 #include <sys/kernel.h>
38 #include <sys/module.h>
39 #include <sys/priv.h>
40 #include <dev/pci/pcivar.h>
41 #if defined(__i386__) || defined(__amd64__)
42 #include <vm/vm.h>
43 #include <vm/pmap.h>
44 #endif
45 
46 #include "common/common.h"
47 #include "common/t4_regs.h"
48 #include "t4_ioctl.h"
49 #include "t4_mp_ring.h"
50 
51 /*
52  * Some notes:
53  *
54  * The Virtual Interfaces are connected to an internal switch on the chip
55  * which allows VIs attached to the same port to talk to each other even when
56  * the port link is down.  As a result, we might want to always report a
57  * VF's link as being "up".
58  *
59  * XXX: Add a TUNABLE and possible per-device sysctl for this?
60  */
61 
62 struct intrs_and_queues {
63 	uint16_t intr_type;	/* MSI, or MSI-X */
64 	uint16_t nirq;		/* Total # of vectors */
65 	uint16_t ntxq;		/* # of NIC txq's for each port */
66 	uint16_t nrxq;		/* # of NIC rxq's for each port */
67 };
68 
69 struct {
70 	uint16_t device;
71 	char *desc;
72 } t4vf_pciids[] = {
73 	{0x4800, "Chelsio T440-dbg VF"},
74 	{0x4801, "Chelsio T420-CR VF"},
75 	{0x4802, "Chelsio T422-CR VF"},
76 	{0x4803, "Chelsio T440-CR VF"},
77 	{0x4804, "Chelsio T420-BCH VF"},
78 	{0x4805, "Chelsio T440-BCH VF"},
79 	{0x4806, "Chelsio T440-CH VF"},
80 	{0x4807, "Chelsio T420-SO VF"},
81 	{0x4808, "Chelsio T420-CX VF"},
82 	{0x4809, "Chelsio T420-BT VF"},
83 	{0x480a, "Chelsio T404-BT VF"},
84 	{0x480e, "Chelsio T440-LP-CR VF"},
85 }, t5vf_pciids[] = {
86 	{0x5800, "Chelsio T580-dbg VF"},
87 	{0x5801,  "Chelsio T520-CR VF"},	/* 2 x 10G */
88 	{0x5802,  "Chelsio T522-CR VF"},	/* 2 x 10G, 2 X 1G */
89 	{0x5803,  "Chelsio T540-CR VF"},	/* 4 x 10G */
90 	{0x5807,  "Chelsio T520-SO VF"},	/* 2 x 10G, nomem */
91 	{0x5809,  "Chelsio T520-BT VF"},	/* 2 x 10GBaseT */
92 	{0x580a,  "Chelsio T504-BT VF"},	/* 4 x 1G */
93 	{0x580d,  "Chelsio T580-CR VF"},	/* 2 x 40G */
94 	{0x580e,  "Chelsio T540-LP-CR VF"},	/* 4 x 10G */
95 	{0x5810,  "Chelsio T580-LP-CR VF"},	/* 2 x 40G */
96 	{0x5811,  "Chelsio T520-LL-CR VF"},	/* 2 x 10G */
97 	{0x5812,  "Chelsio T560-CR VF"},	/* 1 x 40G, 2 x 10G */
98 	{0x5814,  "Chelsio T580-LP-SO-CR VF"},	/* 2 x 40G, nomem */
99 	{0x5815,  "Chelsio T502-BT VF"},	/* 2 x 1G */
100 	{0x5818,  "Chelsio T540-BT VF"},	/* 4 x 10GBaseT */
101 	{0x5819,  "Chelsio T540-LP-BT VF"},	/* 4 x 10GBaseT */
102 	{0x581a,  "Chelsio T540-SO-BT VF"},	/* 4 x 10GBaseT, nomem */
103 	{0x581b,  "Chelsio T540-SO-CR VF"},	/* 4 x 10G, nomem */
104 }, t6vf_pciids[] = {
105 	{0x6800, "Chelsio T6-DBG-25 VF"},	/* 2 x 10/25G, debug */
106 	{0x6801, "Chelsio T6225-CR VF"},	/* 2 x 10/25G */
107 	{0x6802, "Chelsio T6225-SO-CR VF"},	/* 2 x 10/25G, nomem */
108 	{0x6803, "Chelsio T6425-CR VF"},	/* 4 x 10/25G */
109 	{0x6804, "Chelsio T6425-SO-CR VF"},	/* 4 x 10/25G, nomem */
110 	{0x6805, "Chelsio T6225-OCP-SO VF"},	/* 2 x 10/25G, nomem */
111 	{0x6806, "Chelsio T62100-OCP-SO VF"},	/* 2 x 40/50/100G, nomem */
112 	{0x6807, "Chelsio T62100-LP-CR VF"},	/* 2 x 40/50/100G */
113 	{0x6808, "Chelsio T62100-SO-CR VF"},	/* 2 x 40/50/100G, nomem */
114 	{0x6809, "Chelsio T6210-BT VF"},	/* 2 x 10GBASE-T */
115 	{0x680d, "Chelsio T62100-CR VF"},	/* 2 x 40/50/100G */
116 	{0x6810, "Chelsio T6-DBG-100 VF"},	/* 2 x 40/50/100G, debug */
117 	{0x6811, "Chelsio T6225-LL-CR VF"},	/* 2 x 10/25G */
118 	{0x6814, "Chelsio T61100-OCP-SO VF"},	/* 1 x 40/50/100G, nomem */
119 	{0x6815, "Chelsio T6201-BT VF"},	/* 2 x 1000BASE-T */
120 
121 	/* Custom */
122 	{0x6880, "Chelsio T6225 80 VF"},
123 	{0x6881, "Chelsio T62100 81 VF"},
124 	{0x6882, "Chelsio T6225-CR 82 VF"},
125 	{0x6883, "Chelsio T62100-CR 83 VF"},
126 	{0x6884, "Chelsio T64100-CR 84 VF"},
127 	{0x6885, "Chelsio T6240-SO 85 VF"},
128 	{0x6886, "Chelsio T6225-SO-CR 86 VF"},
129 	{0x6887, "Chelsio T6225-CR 87 VF"},
130 };
131 
132 static d_ioctl_t t4vf_ioctl;
133 
134 static struct cdevsw t4vf_cdevsw = {
135        .d_version = D_VERSION,
136        .d_ioctl = t4vf_ioctl,
137        .d_name = "t4vf",
138 };
139 
140 static int
141 t4vf_probe(device_t dev)
142 {
143 	uint16_t d;
144 	size_t i;
145 
146 	d = pci_get_device(dev);
147 	for (i = 0; i < nitems(t4vf_pciids); i++) {
148 		if (d == t4vf_pciids[i].device) {
149 			device_set_desc(dev, t4vf_pciids[i].desc);
150 			return (BUS_PROBE_DEFAULT);
151 		}
152 	}
153 	return (ENXIO);
154 }
155 
156 static int
157 t5vf_probe(device_t dev)
158 {
159 	uint16_t d;
160 	size_t i;
161 
162 	d = pci_get_device(dev);
163 	for (i = 0; i < nitems(t5vf_pciids); i++) {
164 		if (d == t5vf_pciids[i].device) {
165 			device_set_desc(dev, t5vf_pciids[i].desc);
166 			return (BUS_PROBE_DEFAULT);
167 		}
168 	}
169 	return (ENXIO);
170 }
171 
172 static int
173 t6vf_probe(device_t dev)
174 {
175 	uint16_t d;
176 	size_t i;
177 
178 	d = pci_get_device(dev);
179 	for (i = 0; i < nitems(t6vf_pciids); i++) {
180 		if (d == t6vf_pciids[i].device) {
181 			device_set_desc(dev, t6vf_pciids[i].desc);
182 			return (BUS_PROBE_DEFAULT);
183 		}
184 	}
185 	return (ENXIO);
186 }
187 
188 #define FW_PARAM_DEV(param) \
189 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
190 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
191 #define FW_PARAM_PFVF(param) \
192 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
193 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
194 
195 static int
196 get_params__pre_init(struct adapter *sc)
197 {
198 	int rc;
199 	uint32_t param[3], val[3];
200 
201 	param[0] = FW_PARAM_DEV(FWREV);
202 	param[1] = FW_PARAM_DEV(TPREV);
203 	param[2] = FW_PARAM_DEV(CCLK);
204 	rc = -t4vf_query_params(sc, nitems(param), param, val);
205 	if (rc != 0) {
206 		device_printf(sc->dev,
207 		    "failed to query parameters (pre_init): %d.\n", rc);
208 		return (rc);
209 	}
210 
211 	sc->params.fw_vers = val[0];
212 	sc->params.tp_vers = val[1];
213 	sc->params.vpd.cclk = val[2];
214 
215 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
216 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
217 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
218 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
219 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
220 
221 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
222 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
223 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
224 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
225 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
226 
227 	return (0);
228 }
229 
230 static int
231 get_params__post_init(struct adapter *sc)
232 {
233 	int rc;
234 	uint32_t param, val;
235 
236 	rc = -t4vf_get_sge_params(sc);
237 	if (rc != 0) {
238 		device_printf(sc->dev,
239 		    "unable to retrieve adapter SGE parameters: %d\n", rc);
240 		return (rc);
241 	}
242 
243 	rc = -t4vf_get_rss_glb_config(sc);
244 	if (rc != 0) {
245 		device_printf(sc->dev,
246 		    "unable to retrieve adapter RSS parameters: %d\n", rc);
247 		return (rc);
248 	}
249 	if (sc->params.rss.mode != FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
250 		device_printf(sc->dev,
251 		    "unable to operate with global RSS mode %d\n",
252 		    sc->params.rss.mode);
253 		return (EINVAL);
254 	}
255 
256 	rc = t4_read_chip_settings(sc);
257 	if (rc != 0)
258 		return (rc);
259 
260 	/*
261 	 * Grab our Virtual Interface resource allocation, extract the
262 	 * features that we're interested in and do a bit of sanity testing on
263 	 * what we discover.
264 	 */
265 	rc = -t4vf_get_vfres(sc);
266 	if (rc != 0) {
267 		device_printf(sc->dev,
268 		    "unable to get virtual interface resources: %d\n", rc);
269 		return (rc);
270 	}
271 
272 	/*
273 	 * Check for various parameter sanity issues.
274 	 */
275 	if (sc->params.vfres.pmask == 0) {
276 		device_printf(sc->dev, "no port access configured/usable!\n");
277 		return (EINVAL);
278 	}
279 	if (sc->params.vfres.nvi == 0) {
280 		device_printf(sc->dev,
281 		    "no virtual interfaces configured/usable!\n");
282 		return (EINVAL);
283 	}
284 	sc->params.portvec = sc->params.vfres.pmask;
285 
286 	param = FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
287 	rc = -t4vf_query_params(sc, 1, &param, &val);
288 	if (rc == 0)
289 		sc->params.max_pkts_per_eth_tx_pkts_wr = val;
290 	else
291 		sc->params.max_pkts_per_eth_tx_pkts_wr = 14;
292 
293 	return (0);
294 }
295 
296 static int
297 set_params__post_init(struct adapter *sc)
298 {
299 	uint32_t param, val;
300 
301 	/* ask for encapsulated CPLs */
302 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
303 	val = 1;
304 	(void)t4vf_set_params(sc, 1, &param, &val);
305 
306 	/* Enable 32b port caps if the firmware supports it. */
307 	param = FW_PARAM_PFVF(PORT_CAPS32);
308 	val = 1;
309 	if (t4vf_set_params(sc, 1, &param, &val) == 0)
310 		sc->params.port_caps32 = 1;
311 
312 	return (0);
313 }
314 
315 #undef FW_PARAM_PFVF
316 #undef FW_PARAM_DEV
317 
318 static int
319 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
320 {
321 	struct vf_resources *vfres;
322 	int nrxq, ntxq, nports;
323 	int itype, iq_avail, navail, rc;
324 
325 	/*
326 	 * Figure out the layout of queues across our VIs and ensure
327 	 * we can allocate enough interrupts for our layout.
328 	 */
329 	vfres = &sc->params.vfres;
330 	nports = sc->params.nports;
331 	bzero(iaq, sizeof(*iaq));
332 
333 	for (itype = INTR_MSIX; itype != 0; itype >>= 1) {
334 		if (itype == INTR_INTX)
335 			continue;
336 
337 		if (itype == INTR_MSIX)
338 			navail = pci_msix_count(sc->dev);
339 		else
340 			navail = pci_msi_count(sc->dev);
341 
342 		if (navail == 0)
343 			continue;
344 
345 		iaq->intr_type = itype;
346 
347 		/*
348 		 * XXX: The Linux driver reserves an Ingress Queue for
349 		 * forwarded interrupts when using MSI (but not MSI-X).
350 		 * It seems it just always asks for 2 interrupts and
351 		 * forwards all rxqs to the forwarded interrupt.
352 		 *
353 		 * We must reserve one IRQ for the for the firmware
354 		 * event queue.
355 		 *
356 		 * Every rxq requires an ingress queue with a free
357 		 * list and interrupts and an egress queue.  Every txq
358 		 * requires an ETH egress queue.
359 		 */
360 		iaq->nirq = T4VF_EXTRA_INTR;
361 
362 		/*
363 		 * First, determine how many queues we can allocate.
364 		 * Start by finding the upper bound on rxqs from the
365 		 * limit on ingress queues.
366 		 */
367 		iq_avail = vfres->niqflint - iaq->nirq;
368 		if (iq_avail < nports) {
369 			device_printf(sc->dev,
370 			    "Not enough ingress queues (%d) for %d ports\n",
371 			    vfres->niqflint, nports);
372 			return (ENXIO);
373 		}
374 
375 		/*
376 		 * Try to honor the cap on interrupts.  If there aren't
377 		 * enough interrupts for at least one interrupt per
378 		 * port, then don't bother, we will just forward all
379 		 * interrupts to one interrupt in that case.
380 		 */
381 		if (iaq->nirq + nports <= navail) {
382 			if (iq_avail > navail - iaq->nirq)
383 				iq_avail = navail - iaq->nirq;
384 		}
385 
386 		nrxq = nports * t4_nrxq;
387 		if (nrxq > iq_avail) {
388 			/*
389 			 * Too many ingress queues.  Use what we can.
390 			 */
391 			nrxq = (iq_avail / nports) * nports;
392 		}
393 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
394 
395 		/*
396 		 * Next, determine the upper bound on txqs from the limit
397 		 * on ETH queues.
398 		 */
399 		if (vfres->nethctrl < nports) {
400 			device_printf(sc->dev,
401 			    "Not enough ETH queues (%d) for %d ports\n",
402 			    vfres->nethctrl, nports);
403 			return (ENXIO);
404 		}
405 
406 		ntxq = nports * t4_ntxq;
407 		if (ntxq > vfres->nethctrl) {
408 			/*
409 			 * Too many ETH queues.  Use what we can.
410 			 */
411 			ntxq = (vfres->nethctrl / nports) * nports;
412 		}
413 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
414 
415 		/*
416 		 * Finally, ensure we have enough egress queues.
417 		 */
418 		if (vfres->neq < nports * 2) {
419 			device_printf(sc->dev,
420 			    "Not enough egress queues (%d) for %d ports\n",
421 			    vfres->neq, nports);
422 			return (ENXIO);
423 		}
424 		if (nrxq + ntxq > vfres->neq) {
425 			/* Just punt and use 1 for everything. */
426 			nrxq = ntxq = nports;
427 		}
428 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
429 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
430 		KASSERT(nrxq + ntxq <= vfres->neq, ("too many egress queues"));
431 
432 		/*
433 		 * Do we have enough interrupts?  For MSI the interrupts
434 		 * have to be a power of 2 as well.
435 		 */
436 		iaq->nirq += nrxq;
437 		iaq->ntxq = ntxq;
438 		iaq->nrxq = nrxq;
439 		if (iaq->nirq <= navail &&
440 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
441 			navail = iaq->nirq;
442 			if (itype == INTR_MSIX)
443 				rc = pci_alloc_msix(sc->dev, &navail);
444 			else
445 				rc = pci_alloc_msi(sc->dev, &navail);
446 			if (rc != 0) {
447 				device_printf(sc->dev,
448 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
449 				    itype, rc, iaq->nirq, navail);
450 				return (rc);
451 			}
452 			if (navail == iaq->nirq) {
453 				return (0);
454 			}
455 			pci_release_msi(sc->dev);
456 		}
457 
458 		/* Fall back to a single interrupt. */
459 		iaq->nirq = 1;
460 		navail = iaq->nirq;
461 		if (itype == INTR_MSIX)
462 			rc = pci_alloc_msix(sc->dev, &navail);
463 		else
464 			rc = pci_alloc_msi(sc->dev, &navail);
465 		if (rc != 0)
466 			device_printf(sc->dev,
467 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
468 			    itype, rc, iaq->nirq, navail);
469 		return (rc);
470 	}
471 
472 	device_printf(sc->dev,
473 	    "failed to find a usable interrupt type.  "
474 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
475 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
476 
477 	return (ENXIO);
478 }
479 
480 static int
481 t4vf_attach(device_t dev)
482 {
483 	struct adapter *sc;
484 	int rc = 0, i, j, rqidx, tqidx;
485 	struct make_dev_args mda;
486 	struct intrs_and_queues iaq;
487 	struct sge *s;
488 
489 	sc = device_get_softc(dev);
490 	sc->dev = dev;
491 	pci_enable_busmaster(dev);
492 	pci_set_max_read_req(dev, 4096);
493 	sc->params.pci.mps = pci_get_max_payload(dev);
494 
495 	sc->flags |= IS_VF;
496 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
497 
498 	sc->sge_gts_reg = VF_SGE_REG(A_SGE_VF_GTS);
499 	sc->sge_kdoorbell_reg = VF_SGE_REG(A_SGE_VF_KDOORBELL);
500 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
501 	    device_get_nameunit(dev));
502 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
503 	t4_add_adapter(sc);
504 
505 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
506 	TAILQ_INIT(&sc->sfl);
507 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
508 
509 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
510 
511 	rc = t4_map_bars_0_and_4(sc);
512 	if (rc != 0)
513 		goto done; /* error message displayed already */
514 
515 	rc = -t4vf_prep_adapter(sc);
516 	if (rc != 0)
517 		goto done;
518 
519 	t4_init_devnames(sc);
520 	if (sc->names == NULL) {
521 		rc = ENOTSUP;
522 		goto done; /* error message displayed already */
523 	}
524 
525 	/*
526 	 * Leave the 'pf' and 'mbox' values as zero.  This ensures
527 	 * that various firmware messages do not set the fields which
528 	 * is the correct thing to do for a VF.
529 	 */
530 
531 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
532 
533 	make_dev_args_init(&mda);
534 	mda.mda_devsw = &t4vf_cdevsw;
535 	mda.mda_uid = UID_ROOT;
536 	mda.mda_gid = GID_WHEEL;
537 	mda.mda_mode = 0600;
538 	mda.mda_si_drv1 = sc;
539 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
540 	if (rc != 0)
541 		device_printf(dev, "failed to create nexus char device: %d.\n",
542 		    rc);
543 
544 #if defined(__i386__)
545 	if ((cpu_feature & CPUID_CX8) == 0) {
546 		device_printf(dev, "64 bit atomics not available.\n");
547 		rc = ENOTSUP;
548 		goto done;
549 	}
550 #endif
551 
552 	/*
553 	 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
554 	 * 2.6.31 and later we can't call pci_reset_function() in order to
555 	 * issue an FLR because of a self- deadlock on the device semaphore.
556 	 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
557 	 * cases where they're needed -- for instance, some versions of KVM
558 	 * fail to reset "Assigned Devices" when the VM reboots.  Therefore we
559 	 * use the firmware based reset in order to reset any per function
560 	 * state.
561 	 */
562 	rc = -t4vf_fw_reset(sc);
563 	if (rc != 0) {
564 		device_printf(dev, "FW reset failed: %d\n", rc);
565 		goto done;
566 	}
567 	sc->flags |= FW_OK;
568 
569 	/*
570 	 * Grab basic operational parameters.  These will predominantly have
571 	 * been set up by the Physical Function Driver or will be hard coded
572 	 * into the adapter.  We just have to live with them ...  Note that
573 	 * we _must_ get our VPD parameters before our SGE parameters because
574 	 * we need to know the adapter's core clock from the VPD in order to
575 	 * properly decode the SGE Timer Values.
576 	 */
577 	rc = get_params__pre_init(sc);
578 	if (rc != 0)
579 		goto done; /* error message displayed already */
580 	rc = get_params__post_init(sc);
581 	if (rc != 0)
582 		goto done; /* error message displayed already */
583 
584 	rc = set_params__post_init(sc);
585 	if (rc != 0)
586 		goto done; /* error message displayed already */
587 
588 	rc = t4_map_bar_2(sc);
589 	if (rc != 0)
590 		goto done; /* error message displayed already */
591 
592 	rc = t4_create_dma_tag(sc);
593 	if (rc != 0)
594 		goto done; /* error message displayed already */
595 
596 	/*
597 	 * The number of "ports" which we support is equal to the number of
598 	 * Virtual Interfaces with which we've been provisioned.
599 	 */
600 	sc->params.nports = imin(sc->params.vfres.nvi, MAX_NPORTS);
601 
602 	/*
603 	 * We may have been provisioned with more VIs than the number of
604 	 * ports we're allowed to access (our Port Access Rights Mask).
605 	 * Just use a single VI for each port.
606 	 */
607 	sc->params.nports = imin(sc->params.nports,
608 	    bitcount32(sc->params.vfres.pmask));
609 
610 #ifdef notyet
611 	/*
612 	 * XXX: The Linux VF driver will lower nports if it thinks there
613 	 * are too few resources in vfres (niqflint, nethctrl, neq).
614 	 */
615 #endif
616 
617 	/*
618 	 * First pass over all the ports - allocate VIs and initialize some
619 	 * basic parameters like mac address, port type, etc.
620 	 */
621 	for_each_port(sc, i) {
622 		struct port_info *pi;
623 
624 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
625 		sc->port[i] = pi;
626 
627 		/* These must be set before t4_port_init */
628 		pi->adapter = sc;
629 		pi->port_id = i;
630 		pi->nvi = 1;
631 		pi->vi = malloc(sizeof(struct vi_info) * pi->nvi, M_CXGBE,
632 		    M_ZERO | M_WAITOK);
633 
634 		/*
635 		 * Allocate the "main" VI and initialize parameters
636 		 * like mac addr.
637 		 */
638 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
639 		if (rc != 0) {
640 			device_printf(dev, "unable to initialize port %d: %d\n",
641 			    i, rc);
642 			free(pi->vi, M_CXGBE);
643 			free(pi, M_CXGBE);
644 			sc->port[i] = NULL;
645 			goto done;
646 		}
647 
648 		/* No t4_link_start. */
649 
650 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
651 		    device_get_nameunit(dev), i);
652 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
653 		sc->chan_map[pi->tx_chan] = i;
654 
655 		/* All VIs on this port share this media. */
656 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
657 		    cxgbe_media_status);
658 
659 		pi->dev = device_add_child(dev, sc->names->vf_ifnet_name, -1);
660 		if (pi->dev == NULL) {
661 			device_printf(dev,
662 			    "failed to add device for port %d.\n", i);
663 			rc = ENXIO;
664 			goto done;
665 		}
666 		pi->vi[0].dev = pi->dev;
667 		device_set_softc(pi->dev, pi);
668 	}
669 
670 	/*
671 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
672 	 */
673 	rc = cfg_itype_and_nqueues(sc, &iaq);
674 	if (rc != 0)
675 		goto done; /* error message displayed already */
676 
677 	sc->intr_type = iaq.intr_type;
678 	sc->intr_count = iaq.nirq;
679 
680 	s = &sc->sge;
681 	s->nrxq = sc->params.nports * iaq.nrxq;
682 	s->ntxq = sc->params.nports * iaq.ntxq;
683 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
684 	s->neq += sc->params.nports;	/* ctrl queues: 1 per port */
685 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
686 
687 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
688 	    M_ZERO | M_WAITOK);
689 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
690 	    M_ZERO | M_WAITOK);
691 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
692 	    M_ZERO | M_WAITOK);
693 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
694 	    M_ZERO | M_WAITOK);
695 
696 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
697 	    M_ZERO | M_WAITOK);
698 
699 	/*
700 	 * Second pass over the ports.  This time we know the number of rx and
701 	 * tx queues that each port should get.
702 	 */
703 	rqidx = tqidx = 0;
704 	for_each_port(sc, i) {
705 		struct port_info *pi = sc->port[i];
706 		struct vi_info *vi;
707 
708 		if (pi == NULL)
709 			continue;
710 
711 		for_each_vi(pi, j, vi) {
712 			vi->pi = pi;
713 			vi->adapter = sc;
714 			vi->qsize_rxq = t4_qsize_rxq;
715 			vi->qsize_txq = t4_qsize_txq;
716 
717 			vi->first_rxq = rqidx;
718 			vi->first_txq = tqidx;
719 			vi->tmr_idx = t4_tmr_idx;
720 			vi->pktc_idx = t4_pktc_idx;
721 			vi->nrxq = j == 0 ? iaq.nrxq: 1;
722 			vi->ntxq = j == 0 ? iaq.ntxq: 1;
723 
724 			rqidx += vi->nrxq;
725 			tqidx += vi->ntxq;
726 
727 			vi->rsrv_noflowq = 0;
728 		}
729 	}
730 
731 	rc = t4_setup_intr_handlers(sc);
732 	if (rc != 0) {
733 		device_printf(dev,
734 		    "failed to setup interrupt handlers: %d\n", rc);
735 		goto done;
736 	}
737 
738 	rc = bus_generic_attach(dev);
739 	if (rc != 0) {
740 		device_printf(dev,
741 		    "failed to attach all child ports: %d\n", rc);
742 		goto done;
743 	}
744 
745 	device_printf(dev,
746 	    "%d ports, %d %s interrupt%s, %d eq, %d iq\n",
747 	    sc->params.nports, sc->intr_count, sc->intr_type == INTR_MSIX ?
748 	    "MSI-X" : "MSI", sc->intr_count > 1 ? "s" : "", sc->sge.neq,
749 	    sc->sge.niq);
750 
751 done:
752 	if (rc != 0)
753 		t4_detach_common(dev);
754 	else
755 		t4_sysctls(sc);
756 
757 	return (rc);
758 }
759 
760 static void
761 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
762 {
763 
764 	/* 0x3f is used as the revision for VFs. */
765 	regs->version = chip_id(sc) | (0x3f << 10);
766 	t4_get_regs(sc, buf, regs->len);
767 }
768 
769 static void
770 t4_clr_vi_stats(struct adapter *sc)
771 {
772 	int reg;
773 
774 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
775 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
776 		t4_write_reg(sc, VF_MPS_REG(reg), 0);
777 }
778 
779 static int
780 t4vf_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
781     struct thread *td)
782 {
783 	int rc;
784 	struct adapter *sc = dev->si_drv1;
785 
786 	rc = priv_check(td, PRIV_DRIVER);
787 	if (rc != 0)
788 		return (rc);
789 
790 	switch (cmd) {
791 	case CHELSIO_T4_GETREG: {
792 		struct t4_reg *edata = (struct t4_reg *)data;
793 
794 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
795 			return (EFAULT);
796 
797 		if (edata->size == 4)
798 			edata->val = t4_read_reg(sc, edata->addr);
799 		else if (edata->size == 8)
800 			edata->val = t4_read_reg64(sc, edata->addr);
801 		else
802 			return (EINVAL);
803 
804 		break;
805 	}
806 	case CHELSIO_T4_SETREG: {
807 		struct t4_reg *edata = (struct t4_reg *)data;
808 
809 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
810 			return (EFAULT);
811 
812 		if (edata->size == 4) {
813 			if (edata->val & 0xffffffff00000000)
814 				return (EINVAL);
815 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
816 		} else if (edata->size == 8)
817 			t4_write_reg64(sc, edata->addr, edata->val);
818 		else
819 			return (EINVAL);
820 		break;
821 	}
822 	case CHELSIO_T4_REGDUMP: {
823 		struct t4_regdump *regs = (struct t4_regdump *)data;
824 		int reglen = t4_get_regs_len(sc);
825 		uint8_t *buf;
826 
827 		if (regs->len < reglen) {
828 			regs->len = reglen; /* hint to the caller */
829 			return (ENOBUFS);
830 		}
831 
832 		regs->len = reglen;
833 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
834 		get_regs(sc, regs, buf);
835 		rc = copyout(buf, regs->data, reglen);
836 		free(buf, M_CXGBE);
837 		break;
838 	}
839 	case CHELSIO_T4_CLEAR_STATS: {
840 		int i, v;
841 		u_int port_id = *(uint32_t *)data;
842 		struct port_info *pi;
843 		struct vi_info *vi;
844 
845 		if (port_id >= sc->params.nports)
846 			return (EINVAL);
847 		pi = sc->port[port_id];
848 
849 		/* MAC stats */
850 		pi->tx_parse_error = 0;
851 		t4_clr_vi_stats(sc);
852 
853 		/*
854 		 * Since this command accepts a port, clear stats for
855 		 * all VIs on this port.
856 		 */
857 		for_each_vi(pi, v, vi) {
858 			if (vi->flags & VI_INIT_DONE) {
859 				struct sge_rxq *rxq;
860 				struct sge_txq *txq;
861 
862 				for_each_rxq(vi, i, rxq) {
863 #if defined(INET) || defined(INET6)
864 					rxq->lro.lro_queued = 0;
865 					rxq->lro.lro_flushed = 0;
866 #endif
867 					rxq->rxcsum = 0;
868 					rxq->vlan_extraction = 0;
869 				}
870 
871 				for_each_txq(vi, i, txq) {
872 					txq->txcsum = 0;
873 					txq->tso_wrs = 0;
874 					txq->vlan_insertion = 0;
875 					txq->imm_wrs = 0;
876 					txq->sgl_wrs = 0;
877 					txq->txpkt_wrs = 0;
878 					txq->txpkts0_wrs = 0;
879 					txq->txpkts1_wrs = 0;
880 					txq->txpkts0_pkts = 0;
881 					txq->txpkts1_pkts = 0;
882 					mp_ring_reset_stats(txq->r);
883 				}
884 			}
885 		}
886 		break;
887 	}
888 	case CHELSIO_T4_SCHED_CLASS:
889 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
890 		break;
891 	case CHELSIO_T4_SCHED_QUEUE:
892 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
893 		break;
894 	default:
895 		rc = ENOTTY;
896 	}
897 
898 	return (rc);
899 }
900 
901 static device_method_t t4vf_methods[] = {
902 	DEVMETHOD(device_probe,		t4vf_probe),
903 	DEVMETHOD(device_attach,	t4vf_attach),
904 	DEVMETHOD(device_detach,	t4_detach_common),
905 
906 	DEVMETHOD_END
907 };
908 
909 static driver_t t4vf_driver = {
910 	"t4vf",
911 	t4vf_methods,
912 	sizeof(struct adapter)
913 };
914 
915 static device_method_t t5vf_methods[] = {
916 	DEVMETHOD(device_probe,		t5vf_probe),
917 	DEVMETHOD(device_attach,	t4vf_attach),
918 	DEVMETHOD(device_detach,	t4_detach_common),
919 
920 	DEVMETHOD_END
921 };
922 
923 static driver_t t5vf_driver = {
924 	"t5vf",
925 	t5vf_methods,
926 	sizeof(struct adapter)
927 };
928 
929 static device_method_t t6vf_methods[] = {
930 	DEVMETHOD(device_probe,		t6vf_probe),
931 	DEVMETHOD(device_attach,	t4vf_attach),
932 	DEVMETHOD(device_detach,	t4_detach_common),
933 
934 	DEVMETHOD_END
935 };
936 
937 static driver_t t6vf_driver = {
938 	"t6vf",
939 	t6vf_methods,
940 	sizeof(struct adapter)
941 };
942 
943 static driver_t cxgbev_driver = {
944 	"cxgbev",
945 	cxgbe_methods,
946 	sizeof(struct port_info)
947 };
948 
949 static driver_t cxlv_driver = {
950 	"cxlv",
951 	cxgbe_methods,
952 	sizeof(struct port_info)
953 };
954 
955 static driver_t ccv_driver = {
956 	"ccv",
957 	cxgbe_methods,
958 	sizeof(struct port_info)
959 };
960 
961 static devclass_t t4vf_devclass, t5vf_devclass, t6vf_devclass;
962 static devclass_t cxgbev_devclass, cxlv_devclass, ccv_devclass;
963 
964 DRIVER_MODULE(t4vf, pci, t4vf_driver, t4vf_devclass, 0, 0);
965 MODULE_VERSION(t4vf, 1);
966 MODULE_DEPEND(t4vf, t4nex, 1, 1, 1);
967 
968 DRIVER_MODULE(t5vf, pci, t5vf_driver, t5vf_devclass, 0, 0);
969 MODULE_VERSION(t5vf, 1);
970 MODULE_DEPEND(t5vf, t5nex, 1, 1, 1);
971 
972 DRIVER_MODULE(t6vf, pci, t6vf_driver, t6vf_devclass, 0, 0);
973 MODULE_VERSION(t6vf, 1);
974 MODULE_DEPEND(t6vf, t6nex, 1, 1, 1);
975 
976 DRIVER_MODULE(cxgbev, t4vf, cxgbev_driver, cxgbev_devclass, 0, 0);
977 MODULE_VERSION(cxgbev, 1);
978 
979 DRIVER_MODULE(cxlv, t5vf, cxlv_driver, cxlv_devclass, 0, 0);
980 MODULE_VERSION(cxlv, 1);
981 
982 DRIVER_MODULE(ccv, t6vf, ccv_driver, ccv_devclass, 0, 0);
983 MODULE_VERSION(ccv, 1);
984