xref: /freebsd/sys/dev/cxgbe/t4_vf.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
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 	/*
257 	 * Grab our Virtual Interface resource allocation, extract the
258 	 * features that we're interested in and do a bit of sanity testing on
259 	 * what we discover.
260 	 */
261 	rc = -t4vf_get_vfres(sc);
262 	if (rc != 0) {
263 		device_printf(sc->dev,
264 		    "unable to get virtual interface resources: %d\n", rc);
265 		return (rc);
266 	}
267 
268 	/*
269 	 * Check for various parameter sanity issues.
270 	 */
271 	if (sc->params.vfres.pmask == 0) {
272 		device_printf(sc->dev, "no port access configured/usable!\n");
273 		return (EINVAL);
274 	}
275 	if (sc->params.vfres.nvi == 0) {
276 		device_printf(sc->dev,
277 		    "no virtual interfaces configured/usable!\n");
278 		return (EINVAL);
279 	}
280 	sc->params.portvec = sc->params.vfres.pmask;
281 
282 	param = FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
283 	rc = -t4vf_query_params(sc, 1, &param, &val);
284 	if (rc == 0)
285 		sc->params.max_pkts_per_eth_tx_pkts_wr = val;
286 	else
287 		sc->params.max_pkts_per_eth_tx_pkts_wr = 14;
288 
289 	rc = t4_verify_chip_settings(sc);
290 	if (rc != 0)
291 		return (rc);
292 	t4_init_rx_buf_info(sc);
293 
294 	return (0);
295 }
296 
297 static int
298 set_params__post_init(struct adapter *sc)
299 {
300 	uint32_t param, val;
301 
302 	/* ask for encapsulated CPLs */
303 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
304 	val = 1;
305 	(void)t4vf_set_params(sc, 1, &param, &val);
306 
307 	/* Enable 32b port caps if the firmware supports it. */
308 	param = FW_PARAM_PFVF(PORT_CAPS32);
309 	val = 1;
310 	if (t4vf_set_params(sc, 1, &param, &val) == 0)
311 		sc->params.port_caps32 = 1;
312 
313 	return (0);
314 }
315 
316 #undef FW_PARAM_PFVF
317 #undef FW_PARAM_DEV
318 
319 static int
320 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
321 {
322 	struct vf_resources *vfres;
323 	int nrxq, ntxq, nports;
324 	int itype, iq_avail, navail, rc;
325 
326 	/*
327 	 * Figure out the layout of queues across our VIs and ensure
328 	 * we can allocate enough interrupts for our layout.
329 	 */
330 	vfres = &sc->params.vfres;
331 	nports = sc->params.nports;
332 	bzero(iaq, sizeof(*iaq));
333 
334 	for (itype = INTR_MSIX; itype != 0; itype >>= 1) {
335 		if (itype == INTR_INTX)
336 			continue;
337 
338 		if (itype == INTR_MSIX)
339 			navail = pci_msix_count(sc->dev);
340 		else
341 			navail = pci_msi_count(sc->dev);
342 
343 		if (navail == 0)
344 			continue;
345 
346 		iaq->intr_type = itype;
347 
348 		/*
349 		 * XXX: The Linux driver reserves an Ingress Queue for
350 		 * forwarded interrupts when using MSI (but not MSI-X).
351 		 * It seems it just always asks for 2 interrupts and
352 		 * forwards all rxqs to the forwarded interrupt.
353 		 *
354 		 * We must reserve one IRQ for the for the firmware
355 		 * event queue.
356 		 *
357 		 * Every rxq requires an ingress queue with a free
358 		 * list and interrupts and an egress queue.  Every txq
359 		 * requires an ETH egress queue.
360 		 */
361 		iaq->nirq = T4VF_EXTRA_INTR;
362 
363 		/*
364 		 * First, determine how many queues we can allocate.
365 		 * Start by finding the upper bound on rxqs from the
366 		 * limit on ingress queues.
367 		 */
368 		iq_avail = vfres->niqflint - iaq->nirq;
369 		if (iq_avail < nports) {
370 			device_printf(sc->dev,
371 			    "Not enough ingress queues (%d) for %d ports\n",
372 			    vfres->niqflint, nports);
373 			return (ENXIO);
374 		}
375 
376 		/*
377 		 * Try to honor the cap on interrupts.  If there aren't
378 		 * enough interrupts for at least one interrupt per
379 		 * port, then don't bother, we will just forward all
380 		 * interrupts to one interrupt in that case.
381 		 */
382 		if (iaq->nirq + nports <= navail) {
383 			if (iq_avail > navail - iaq->nirq)
384 				iq_avail = navail - iaq->nirq;
385 		}
386 
387 		nrxq = nports * t4_nrxq;
388 		if (nrxq > iq_avail) {
389 			/*
390 			 * Too many ingress queues.  Use what we can.
391 			 */
392 			nrxq = (iq_avail / nports) * nports;
393 		}
394 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
395 
396 		/*
397 		 * Next, determine the upper bound on txqs from the limit
398 		 * on ETH queues.
399 		 */
400 		if (vfres->nethctrl < nports) {
401 			device_printf(sc->dev,
402 			    "Not enough ETH queues (%d) for %d ports\n",
403 			    vfres->nethctrl, nports);
404 			return (ENXIO);
405 		}
406 
407 		ntxq = nports * t4_ntxq;
408 		if (ntxq > vfres->nethctrl) {
409 			/*
410 			 * Too many ETH queues.  Use what we can.
411 			 */
412 			ntxq = (vfres->nethctrl / nports) * nports;
413 		}
414 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
415 
416 		/*
417 		 * Finally, ensure we have enough egress queues.
418 		 */
419 		if (vfres->neq < nports * 2) {
420 			device_printf(sc->dev,
421 			    "Not enough egress queues (%d) for %d ports\n",
422 			    vfres->neq, nports);
423 			return (ENXIO);
424 		}
425 		if (nrxq + ntxq > vfres->neq) {
426 			/* Just punt and use 1 for everything. */
427 			nrxq = ntxq = nports;
428 		}
429 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
430 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
431 		KASSERT(nrxq + ntxq <= vfres->neq, ("too many egress queues"));
432 
433 		/*
434 		 * Do we have enough interrupts?  For MSI the interrupts
435 		 * have to be a power of 2 as well.
436 		 */
437 		iaq->nirq += nrxq;
438 		iaq->ntxq = ntxq;
439 		iaq->nrxq = nrxq;
440 		if (iaq->nirq <= navail &&
441 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
442 			navail = iaq->nirq;
443 			if (itype == INTR_MSIX)
444 				rc = pci_alloc_msix(sc->dev, &navail);
445 			else
446 				rc = pci_alloc_msi(sc->dev, &navail);
447 			if (rc != 0) {
448 				device_printf(sc->dev,
449 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
450 				    itype, rc, iaq->nirq, navail);
451 				return (rc);
452 			}
453 			if (navail == iaq->nirq) {
454 				return (0);
455 			}
456 			pci_release_msi(sc->dev);
457 		}
458 
459 		/* Fall back to a single interrupt. */
460 		iaq->nirq = 1;
461 		navail = iaq->nirq;
462 		if (itype == INTR_MSIX)
463 			rc = pci_alloc_msix(sc->dev, &navail);
464 		else
465 			rc = pci_alloc_msi(sc->dev, &navail);
466 		if (rc != 0)
467 			device_printf(sc->dev,
468 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
469 			    itype, rc, iaq->nirq, navail);
470 		return (rc);
471 	}
472 
473 	device_printf(sc->dev,
474 	    "failed to find a usable interrupt type.  "
475 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
476 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
477 
478 	return (ENXIO);
479 }
480 
481 static int
482 t4vf_attach(device_t dev)
483 {
484 	struct adapter *sc;
485 	int rc = 0, i, j, rqidx, tqidx, n, p, pmask;
486 	struct make_dev_args mda;
487 	struct intrs_and_queues iaq;
488 	struct sge *s;
489 
490 	sc = device_get_softc(dev);
491 	sc->dev = dev;
492 	pci_enable_busmaster(dev);
493 	pci_set_max_read_req(dev, 4096);
494 	sc->params.pci.mps = pci_get_max_payload(dev);
495 
496 	sc->flags |= IS_VF;
497 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
498 
499 	sc->sge_gts_reg = VF_SGE_REG(A_SGE_VF_GTS);
500 	sc->sge_kdoorbell_reg = VF_SGE_REG(A_SGE_VF_KDOORBELL);
501 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
502 	    device_get_nameunit(dev));
503 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
504 	t4_add_adapter(sc);
505 
506 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
507 	TAILQ_INIT(&sc->sfl);
508 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
509 
510 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
511 
512 	rc = t4_map_bars_0_and_4(sc);
513 	if (rc != 0)
514 		goto done; /* error message displayed already */
515 
516 	rc = -t4vf_prep_adapter(sc);
517 	if (rc != 0)
518 		goto done;
519 
520 	t4_init_devnames(sc);
521 	if (sc->names == NULL) {
522 		rc = ENOTSUP;
523 		goto done; /* error message displayed already */
524 	}
525 
526 	/*
527 	 * Leave the 'pf' and 'mbox' values as zero.  This ensures
528 	 * that various firmware messages do not set the fields which
529 	 * is the correct thing to do for a VF.
530 	 */
531 
532 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
533 
534 	make_dev_args_init(&mda);
535 	mda.mda_devsw = &t4vf_cdevsw;
536 	mda.mda_uid = UID_ROOT;
537 	mda.mda_gid = GID_WHEEL;
538 	mda.mda_mode = 0600;
539 	mda.mda_si_drv1 = sc;
540 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
541 	if (rc != 0)
542 		device_printf(dev, "failed to create nexus char device: %d.\n",
543 		    rc);
544 
545 #if defined(__i386__)
546 	if ((cpu_feature & CPUID_CX8) == 0) {
547 		device_printf(dev, "64 bit atomics not available.\n");
548 		rc = ENOTSUP;
549 		goto done;
550 	}
551 #endif
552 
553 	/*
554 	 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
555 	 * 2.6.31 and later we can't call pci_reset_function() in order to
556 	 * issue an FLR because of a self- deadlock on the device semaphore.
557 	 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
558 	 * cases where they're needed -- for instance, some versions of KVM
559 	 * fail to reset "Assigned Devices" when the VM reboots.  Therefore we
560 	 * use the firmware based reset in order to reset any per function
561 	 * state.
562 	 */
563 	rc = -t4vf_fw_reset(sc);
564 	if (rc != 0) {
565 		device_printf(dev, "FW reset failed: %d\n", rc);
566 		goto done;
567 	}
568 	sc->flags |= FW_OK;
569 
570 	/*
571 	 * Grab basic operational parameters.  These will predominantly have
572 	 * been set up by the Physical Function Driver or will be hard coded
573 	 * into the adapter.  We just have to live with them ...  Note that
574 	 * we _must_ get our VPD parameters before our SGE parameters because
575 	 * we need to know the adapter's core clock from the VPD in order to
576 	 * properly decode the SGE Timer Values.
577 	 */
578 	rc = get_params__pre_init(sc);
579 	if (rc != 0)
580 		goto done; /* error message displayed already */
581 	rc = get_params__post_init(sc);
582 	if (rc != 0)
583 		goto done; /* error message displayed already */
584 
585 	rc = set_params__post_init(sc);
586 	if (rc != 0)
587 		goto done; /* error message displayed already */
588 
589 	rc = t4_map_bar_2(sc);
590 	if (rc != 0)
591 		goto done; /* error message displayed already */
592 
593 	rc = t4_create_dma_tag(sc);
594 	if (rc != 0)
595 		goto done; /* error message displayed already */
596 
597 	/*
598 	 * The number of "ports" which we support is equal to the number of
599 	 * Virtual Interfaces with which we've been provisioned.
600 	 */
601 	sc->params.nports = imin(sc->params.vfres.nvi, MAX_NPORTS);
602 
603 	/*
604 	 * We may have been provisioned with more VIs than the number of
605 	 * ports we're allowed to access (our Port Access Rights Mask).
606 	 * Just use a single VI for each port.
607 	 */
608 	sc->params.nports = imin(sc->params.nports,
609 	    bitcount32(sc->params.vfres.pmask));
610 
611 #ifdef notyet
612 	/*
613 	 * XXX: The Linux VF driver will lower nports if it thinks there
614 	 * are too few resources in vfres (niqflint, nethctrl, neq).
615 	 */
616 #endif
617 
618 	/*
619 	 * First pass over all the ports - allocate VIs and initialize some
620 	 * basic parameters like mac address, port type, etc.
621 	 */
622 	pmask = sc->params.vfres.pmask;
623 	for_each_port(sc, i) {
624 		struct port_info *pi;
625 		uint8_t mac[ETHER_ADDR_LEN];
626 
627 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
628 		sc->port[i] = pi;
629 
630 		/* These must be set before t4_port_init */
631 		pi->adapter = sc;
632 		pi->port_id = i;
633 		pi->nvi = 1;
634 		pi->vi = malloc(sizeof(struct vi_info) * pi->nvi, M_CXGBE,
635 		    M_ZERO | M_WAITOK);
636 
637 		/*
638 		 * Allocate the "main" VI and initialize parameters
639 		 * like mac addr.
640 		 */
641 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
642 		if (rc != 0) {
643 			device_printf(dev, "unable to initialize port %d: %d\n",
644 			    i, rc);
645 			free(pi->vi, M_CXGBE);
646 			free(pi, M_CXGBE);
647 			sc->port[i] = NULL;
648 			goto done;
649 		}
650 
651 		/* Prefer the MAC address set by the PF, if there is one. */
652 		n = 1;
653 		p = ffs(pmask) - 1;
654 		MPASS(p >= 0);
655 		rc = t4vf_get_vf_mac(sc, p, &n, mac);
656 		if (rc == 0 && n == 1)
657 			t4_os_set_hw_addr(pi, mac);
658 		pmask &= ~(1 << p);
659 
660 		/* No t4_link_start. */
661 
662 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
663 		    device_get_nameunit(dev), i);
664 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
665 		sc->chan_map[pi->tx_chan] = i;
666 
667 		/* All VIs on this port share this media. */
668 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
669 		    cxgbe_media_status);
670 
671 		pi->dev = device_add_child(dev, sc->names->vf_ifnet_name, -1);
672 		if (pi->dev == NULL) {
673 			device_printf(dev,
674 			    "failed to add device for port %d.\n", i);
675 			rc = ENXIO;
676 			goto done;
677 		}
678 		pi->vi[0].dev = pi->dev;
679 		device_set_softc(pi->dev, pi);
680 	}
681 
682 	/*
683 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
684 	 */
685 	rc = cfg_itype_and_nqueues(sc, &iaq);
686 	if (rc != 0)
687 		goto done; /* error message displayed already */
688 
689 	sc->intr_type = iaq.intr_type;
690 	sc->intr_count = iaq.nirq;
691 
692 	s = &sc->sge;
693 	s->nrxq = sc->params.nports * iaq.nrxq;
694 	s->ntxq = sc->params.nports * iaq.ntxq;
695 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
696 	s->neq += sc->params.nports;	/* ctrl queues: 1 per port */
697 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
698 
699 	s->iqmap_sz = s->niq;
700 	s->eqmap_sz = s->neq;
701 
702 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
703 	    M_ZERO | M_WAITOK);
704 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
705 	    M_ZERO | M_WAITOK);
706 	s->iqmap = malloc(s->iqmap_sz * sizeof(struct sge_iq *), M_CXGBE,
707 	    M_ZERO | M_WAITOK);
708 	s->eqmap = malloc(s->eqmap_sz * sizeof(struct sge_eq *), M_CXGBE,
709 	    M_ZERO | M_WAITOK);
710 
711 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
712 	    M_ZERO | M_WAITOK);
713 
714 	/*
715 	 * Second pass over the ports.  This time we know the number of rx and
716 	 * tx queues that each port should get.
717 	 */
718 	rqidx = tqidx = 0;
719 	for_each_port(sc, i) {
720 		struct port_info *pi = sc->port[i];
721 		struct vi_info *vi;
722 
723 		if (pi == NULL)
724 			continue;
725 
726 		for_each_vi(pi, j, vi) {
727 			vi->pi = pi;
728 			vi->adapter = sc;
729 			vi->qsize_rxq = t4_qsize_rxq;
730 			vi->qsize_txq = t4_qsize_txq;
731 
732 			vi->first_rxq = rqidx;
733 			vi->first_txq = tqidx;
734 			vi->tmr_idx = t4_tmr_idx;
735 			vi->pktc_idx = t4_pktc_idx;
736 			vi->nrxq = j == 0 ? iaq.nrxq: 1;
737 			vi->ntxq = j == 0 ? iaq.ntxq: 1;
738 
739 			rqidx += vi->nrxq;
740 			tqidx += vi->ntxq;
741 
742 			vi->rsrv_noflowq = 0;
743 		}
744 	}
745 
746 	rc = t4_setup_intr_handlers(sc);
747 	if (rc != 0) {
748 		device_printf(dev,
749 		    "failed to setup interrupt handlers: %d\n", rc);
750 		goto done;
751 	}
752 
753 	rc = bus_generic_attach(dev);
754 	if (rc != 0) {
755 		device_printf(dev,
756 		    "failed to attach all child ports: %d\n", rc);
757 		goto done;
758 	}
759 
760 	device_printf(dev,
761 	    "%d ports, %d %s interrupt%s, %d eq, %d iq\n",
762 	    sc->params.nports, sc->intr_count, sc->intr_type == INTR_MSIX ?
763 	    "MSI-X" : "MSI", sc->intr_count > 1 ? "s" : "", sc->sge.neq,
764 	    sc->sge.niq);
765 
766 done:
767 	if (rc != 0)
768 		t4_detach_common(dev);
769 	else
770 		t4_sysctls(sc);
771 
772 	return (rc);
773 }
774 
775 static void
776 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
777 {
778 
779 	/* 0x3f is used as the revision for VFs. */
780 	regs->version = chip_id(sc) | (0x3f << 10);
781 	t4_get_regs(sc, buf, regs->len);
782 }
783 
784 static void
785 t4_clr_vi_stats(struct adapter *sc)
786 {
787 	int reg;
788 
789 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
790 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
791 		t4_write_reg(sc, VF_MPS_REG(reg), 0);
792 }
793 
794 static int
795 t4vf_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
796     struct thread *td)
797 {
798 	int rc;
799 	struct adapter *sc = dev->si_drv1;
800 
801 	rc = priv_check(td, PRIV_DRIVER);
802 	if (rc != 0)
803 		return (rc);
804 
805 	switch (cmd) {
806 	case CHELSIO_T4_GETREG: {
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 			edata->val = t4_read_reg(sc, edata->addr);
814 		else if (edata->size == 8)
815 			edata->val = t4_read_reg64(sc, edata->addr);
816 		else
817 			return (EINVAL);
818 
819 		break;
820 	}
821 	case CHELSIO_T4_SETREG: {
822 		struct t4_reg *edata = (struct t4_reg *)data;
823 
824 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
825 			return (EFAULT);
826 
827 		if (edata->size == 4) {
828 			if (edata->val & 0xffffffff00000000)
829 				return (EINVAL);
830 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
831 		} else if (edata->size == 8)
832 			t4_write_reg64(sc, edata->addr, edata->val);
833 		else
834 			return (EINVAL);
835 		break;
836 	}
837 	case CHELSIO_T4_REGDUMP: {
838 		struct t4_regdump *regs = (struct t4_regdump *)data;
839 		int reglen = t4_get_regs_len(sc);
840 		uint8_t *buf;
841 
842 		if (regs->len < reglen) {
843 			regs->len = reglen; /* hint to the caller */
844 			return (ENOBUFS);
845 		}
846 
847 		regs->len = reglen;
848 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
849 		get_regs(sc, regs, buf);
850 		rc = copyout(buf, regs->data, reglen);
851 		free(buf, M_CXGBE);
852 		break;
853 	}
854 	case CHELSIO_T4_CLEAR_STATS: {
855 		int i, v;
856 		u_int port_id = *(uint32_t *)data;
857 		struct port_info *pi;
858 		struct vi_info *vi;
859 
860 		if (port_id >= sc->params.nports)
861 			return (EINVAL);
862 		pi = sc->port[port_id];
863 
864 		/* MAC stats */
865 		pi->tx_parse_error = 0;
866 		t4_clr_vi_stats(sc);
867 
868 		/*
869 		 * Since this command accepts a port, clear stats for
870 		 * all VIs on this port.
871 		 */
872 		for_each_vi(pi, v, vi) {
873 			if (vi->flags & VI_INIT_DONE) {
874 				struct sge_rxq *rxq;
875 				struct sge_txq *txq;
876 
877 				for_each_rxq(vi, i, rxq) {
878 #if defined(INET) || defined(INET6)
879 					rxq->lro.lro_queued = 0;
880 					rxq->lro.lro_flushed = 0;
881 #endif
882 					rxq->rxcsum = 0;
883 					rxq->vlan_extraction = 0;
884 				}
885 
886 				for_each_txq(vi, i, txq) {
887 					txq->txcsum = 0;
888 					txq->tso_wrs = 0;
889 					txq->vlan_insertion = 0;
890 					txq->imm_wrs = 0;
891 					txq->sgl_wrs = 0;
892 					txq->txpkt_wrs = 0;
893 					txq->txpkts0_wrs = 0;
894 					txq->txpkts1_wrs = 0;
895 					txq->txpkts0_pkts = 0;
896 					txq->txpkts1_pkts = 0;
897 					txq->txpkts_flush = 0;
898 					mp_ring_reset_stats(txq->r);
899 				}
900 			}
901 		}
902 		break;
903 	}
904 	case CHELSIO_T4_SCHED_CLASS:
905 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
906 		break;
907 	case CHELSIO_T4_SCHED_QUEUE:
908 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
909 		break;
910 	default:
911 		rc = ENOTTY;
912 	}
913 
914 	return (rc);
915 }
916 
917 static device_method_t t4vf_methods[] = {
918 	DEVMETHOD(device_probe,		t4vf_probe),
919 	DEVMETHOD(device_attach,	t4vf_attach),
920 	DEVMETHOD(device_detach,	t4_detach_common),
921 
922 	DEVMETHOD_END
923 };
924 
925 static driver_t t4vf_driver = {
926 	"t4vf",
927 	t4vf_methods,
928 	sizeof(struct adapter)
929 };
930 
931 static device_method_t t5vf_methods[] = {
932 	DEVMETHOD(device_probe,		t5vf_probe),
933 	DEVMETHOD(device_attach,	t4vf_attach),
934 	DEVMETHOD(device_detach,	t4_detach_common),
935 
936 	DEVMETHOD_END
937 };
938 
939 static driver_t t5vf_driver = {
940 	"t5vf",
941 	t5vf_methods,
942 	sizeof(struct adapter)
943 };
944 
945 static device_method_t t6vf_methods[] = {
946 	DEVMETHOD(device_probe,		t6vf_probe),
947 	DEVMETHOD(device_attach,	t4vf_attach),
948 	DEVMETHOD(device_detach,	t4_detach_common),
949 
950 	DEVMETHOD_END
951 };
952 
953 static driver_t t6vf_driver = {
954 	"t6vf",
955 	t6vf_methods,
956 	sizeof(struct adapter)
957 };
958 
959 static driver_t cxgbev_driver = {
960 	"cxgbev",
961 	cxgbe_methods,
962 	sizeof(struct port_info)
963 };
964 
965 static driver_t cxlv_driver = {
966 	"cxlv",
967 	cxgbe_methods,
968 	sizeof(struct port_info)
969 };
970 
971 static driver_t ccv_driver = {
972 	"ccv",
973 	cxgbe_methods,
974 	sizeof(struct port_info)
975 };
976 
977 static devclass_t t4vf_devclass, t5vf_devclass, t6vf_devclass;
978 static devclass_t cxgbev_devclass, cxlv_devclass, ccv_devclass;
979 
980 DRIVER_MODULE(t4vf, pci, t4vf_driver, t4vf_devclass, 0, 0);
981 MODULE_VERSION(t4vf, 1);
982 MODULE_DEPEND(t4vf, t4nex, 1, 1, 1);
983 
984 DRIVER_MODULE(t5vf, pci, t5vf_driver, t5vf_devclass, 0, 0);
985 MODULE_VERSION(t5vf, 1);
986 MODULE_DEPEND(t5vf, t5nex, 1, 1, 1);
987 
988 DRIVER_MODULE(t6vf, pci, t6vf_driver, t6vf_devclass, 0, 0);
989 MODULE_VERSION(t6vf, 1);
990 MODULE_DEPEND(t6vf, t6nex, 1, 1, 1);
991 
992 DRIVER_MODULE(cxgbev, t4vf, cxgbev_driver, cxgbev_devclass, 0, 0);
993 MODULE_VERSION(cxgbev, 1);
994 
995 DRIVER_MODULE(cxlv, t5vf, cxlv_driver, cxlv_devclass, 0, 0);
996 MODULE_VERSION(cxlv, 1);
997 
998 DRIVER_MODULE(ccv, t6vf, ccv_driver, ccv_devclass, 0, 0);
999 MODULE_VERSION(ccv, 1);
1000