xref: /freebsd/sys/dev/cxgbe/t4_vf.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
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 #include "opt_inet.h"
30 #include "opt_inet6.h"
31 
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/conf.h>
35 #include <sys/kernel.h>
36 #include <sys/module.h>
37 #include <sys/priv.h>
38 #include <dev/pci/pcivar.h>
39 #if defined(__i386__) || defined(__amd64__)
40 #include <vm/vm.h>
41 #include <vm/pmap.h>
42 #endif
43 
44 #include "common/common.h"
45 #include "common/t4_regs.h"
46 #include "t4_ioctl.h"
47 #include "t4_mp_ring.h"
48 
49 /*
50  * Some notes:
51  *
52  * The Virtual Interfaces are connected to an internal switch on the chip
53  * which allows VIs attached to the same port to talk to each other even when
54  * the port link is down.  As a result, we might want to always report a
55  * VF's link as being "up".
56  *
57  * XXX: Add a TUNABLE and possible per-device sysctl for this?
58  */
59 
60 struct intrs_and_queues {
61 	uint16_t intr_type;	/* MSI, or MSI-X */
62 	uint16_t nirq;		/* Total # of vectors */
63 	uint16_t ntxq;		/* # of NIC txq's for each port */
64 	uint16_t nrxq;		/* # of NIC rxq's for each port */
65 };
66 
67 struct {
68 	uint16_t device;
69 	char *desc;
70 } t4vf_pciids[] = {
71 	{0x4800, "Chelsio T440-dbg VF"},
72 	{0x4801, "Chelsio T420-CR VF"},
73 	{0x4802, "Chelsio T422-CR VF"},
74 	{0x4803, "Chelsio T440-CR VF"},
75 	{0x4804, "Chelsio T420-BCH VF"},
76 	{0x4805, "Chelsio T440-BCH VF"},
77 	{0x4806, "Chelsio T440-CH VF"},
78 	{0x4807, "Chelsio T420-SO VF"},
79 	{0x4808, "Chelsio T420-CX VF"},
80 	{0x4809, "Chelsio T420-BT VF"},
81 	{0x480a, "Chelsio T404-BT VF"},
82 	{0x480e, "Chelsio T440-LP-CR VF"},
83 }, t5vf_pciids[] = {
84 	{0x5800, "Chelsio T580-dbg VF"},
85 	{0x5801,  "Chelsio T520-CR VF"},	/* 2 x 10G */
86 	{0x5802,  "Chelsio T522-CR VF"},	/* 2 x 10G, 2 X 1G */
87 	{0x5803,  "Chelsio T540-CR VF"},	/* 4 x 10G */
88 	{0x5807,  "Chelsio T520-SO VF"},	/* 2 x 10G, nomem */
89 	{0x5809,  "Chelsio T520-BT VF"},	/* 2 x 10GBaseT */
90 	{0x580a,  "Chelsio T504-BT VF"},	/* 4 x 1G */
91 	{0x580d,  "Chelsio T580-CR VF"},	/* 2 x 40G */
92 	{0x580e,  "Chelsio T540-LP-CR VF"},	/* 4 x 10G */
93 	{0x5810,  "Chelsio T580-LP-CR VF"},	/* 2 x 40G */
94 	{0x5811,  "Chelsio T520-LL-CR VF"},	/* 2 x 10G */
95 	{0x5812,  "Chelsio T560-CR VF"},	/* 1 x 40G, 2 x 10G */
96 	{0x5814,  "Chelsio T580-LP-SO-CR VF"},	/* 2 x 40G, nomem */
97 	{0x5815,  "Chelsio T502-BT VF"},	/* 2 x 1G */
98 	{0x5818,  "Chelsio T540-BT VF"},	/* 4 x 10GBaseT */
99 	{0x5819,  "Chelsio T540-LP-BT VF"},	/* 4 x 10GBaseT */
100 	{0x581a,  "Chelsio T540-SO-BT VF"},	/* 4 x 10GBaseT, nomem */
101 	{0x581b,  "Chelsio T540-SO-CR VF"},	/* 4 x 10G, nomem */
102 }, t6vf_pciids[] = {
103 	{0x6800, "Chelsio T6-DBG-25 VF"},	/* 2 x 10/25G, debug */
104 	{0x6801, "Chelsio T6225-CR VF"},	/* 2 x 10/25G */
105 	{0x6802, "Chelsio T6225-SO-CR VF"},	/* 2 x 10/25G, nomem */
106 	{0x6803, "Chelsio T6425-CR VF"},	/* 4 x 10/25G */
107 	{0x6804, "Chelsio T6425-SO-CR VF"},	/* 4 x 10/25G, nomem */
108 	{0x6805, "Chelsio T6225-OCP-SO VF"},	/* 2 x 10/25G, nomem */
109 	{0x6806, "Chelsio T62100-OCP-SO VF"},	/* 2 x 40/50/100G, nomem */
110 	{0x6807, "Chelsio T62100-LP-CR VF"},	/* 2 x 40/50/100G */
111 	{0x6808, "Chelsio T62100-SO-CR VF"},	/* 2 x 40/50/100G, nomem */
112 	{0x6809, "Chelsio T6210-BT VF"},	/* 2 x 10GBASE-T */
113 	{0x680d, "Chelsio T62100-CR VF"},	/* 2 x 40/50/100G */
114 	{0x6810, "Chelsio T6-DBG-100 VF"},	/* 2 x 40/50/100G, debug */
115 	{0x6811, "Chelsio T6225-LL-CR VF"},	/* 2 x 10/25G */
116 	{0x6814, "Chelsio T61100-OCP-SO VF"},	/* 1 x 40/50/100G, nomem */
117 	{0x6815, "Chelsio T6201-BT VF"},	/* 2 x 1000BASE-T */
118 
119 	/* Custom */
120 	{0x6880, "Chelsio T6225 80 VF"},
121 	{0x6881, "Chelsio T62100 81 VF"},
122 	{0x6882, "Chelsio T6225-CR 82 VF"},
123 	{0x6883, "Chelsio T62100-CR 83 VF"},
124 	{0x6884, "Chelsio T64100-CR 84 VF"},
125 	{0x6885, "Chelsio T6240-SO 85 VF"},
126 	{0x6886, "Chelsio T6225-SO-CR 86 VF"},
127 	{0x6887, "Chelsio T6225-CR 87 VF"},
128 };
129 
130 static d_ioctl_t t4vf_ioctl;
131 
132 static struct cdevsw t4vf_cdevsw = {
133        .d_version = D_VERSION,
134        .d_ioctl = t4vf_ioctl,
135        .d_name = "t4vf",
136 };
137 
138 static int
139 t4vf_probe(device_t dev)
140 {
141 	uint16_t d;
142 	size_t i;
143 
144 	d = pci_get_device(dev);
145 	for (i = 0; i < nitems(t4vf_pciids); i++) {
146 		if (d == t4vf_pciids[i].device) {
147 			device_set_desc(dev, t4vf_pciids[i].desc);
148 			return (BUS_PROBE_DEFAULT);
149 		}
150 	}
151 	return (ENXIO);
152 }
153 
154 static int
155 t5vf_probe(device_t dev)
156 {
157 	uint16_t d;
158 	size_t i;
159 
160 	d = pci_get_device(dev);
161 	for (i = 0; i < nitems(t5vf_pciids); i++) {
162 		if (d == t5vf_pciids[i].device) {
163 			device_set_desc(dev, t5vf_pciids[i].desc);
164 			return (BUS_PROBE_DEFAULT);
165 		}
166 	}
167 	return (ENXIO);
168 }
169 
170 static int
171 t6vf_probe(device_t dev)
172 {
173 	uint16_t d;
174 	size_t i;
175 
176 	d = pci_get_device(dev);
177 	for (i = 0; i < nitems(t6vf_pciids); i++) {
178 		if (d == t6vf_pciids[i].device) {
179 			device_set_desc(dev, t6vf_pciids[i].desc);
180 			return (BUS_PROBE_DEFAULT);
181 		}
182 	}
183 	return (ENXIO);
184 }
185 
186 #define FW_PARAM_DEV(param) \
187 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
188 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
189 #define FW_PARAM_PFVF(param) \
190 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
191 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
192 
193 static int
194 get_params__pre_init(struct adapter *sc)
195 {
196 	int rc;
197 	uint32_t param[3], val[3];
198 
199 	param[0] = FW_PARAM_DEV(FWREV);
200 	param[1] = FW_PARAM_DEV(TPREV);
201 	param[2] = FW_PARAM_DEV(CCLK);
202 	rc = -t4vf_query_params(sc, nitems(param), param, val);
203 	if (rc != 0) {
204 		device_printf(sc->dev,
205 		    "failed to query parameters (pre_init): %d.\n", rc);
206 		return (rc);
207 	}
208 
209 	sc->params.fw_vers = val[0];
210 	sc->params.tp_vers = val[1];
211 	sc->params.vpd.cclk = val[2];
212 
213 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
214 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
215 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
216 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
217 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
218 
219 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
220 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
221 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
222 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
223 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
224 
225 	return (0);
226 }
227 
228 static int
229 get_params__post_init(struct adapter *sc)
230 {
231 	int rc;
232 	uint32_t param, val;
233 
234 	rc = -t4vf_get_sge_params(sc);
235 	if (rc != 0) {
236 		device_printf(sc->dev,
237 		    "unable to retrieve adapter SGE parameters: %d\n", rc);
238 		return (rc);
239 	}
240 
241 	rc = -t4vf_get_rss_glb_config(sc);
242 	if (rc != 0) {
243 		device_printf(sc->dev,
244 		    "unable to retrieve adapter RSS parameters: %d\n", rc);
245 		return (rc);
246 	}
247 	if (sc->params.rss.mode != FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
248 		device_printf(sc->dev,
249 		    "unable to operate with global RSS mode %d\n",
250 		    sc->params.rss.mode);
251 		return (EINVAL);
252 	}
253 
254 	/*
255 	 * Grab our Virtual Interface resource allocation, extract the
256 	 * features that we're interested in and do a bit of sanity testing on
257 	 * what we discover.
258 	 */
259 	rc = -t4vf_get_vfres(sc);
260 	if (rc != 0) {
261 		device_printf(sc->dev,
262 		    "unable to get virtual interface resources: %d\n", rc);
263 		return (rc);
264 	}
265 
266 	/*
267 	 * Check for various parameter sanity issues.
268 	 */
269 	if (sc->params.vfres.pmask == 0) {
270 		device_printf(sc->dev, "no port access configured/usable!\n");
271 		return (EINVAL);
272 	}
273 	if (sc->params.vfres.nvi == 0) {
274 		device_printf(sc->dev,
275 		    "no virtual interfaces configured/usable!\n");
276 		return (EINVAL);
277 	}
278 	sc->params.portvec = sc->params.vfres.pmask;
279 
280 	param = FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
281 	rc = -t4vf_query_params(sc, 1, &param, &val);
282 	if (rc == 0)
283 		sc->params.max_pkts_per_eth_tx_pkts_wr = val;
284 	else
285 		sc->params.max_pkts_per_eth_tx_pkts_wr = 14;
286 
287 	rc = t4_verify_chip_settings(sc);
288 	if (rc != 0)
289 		return (rc);
290 	t4_init_rx_buf_info(sc);
291 
292 	return (0);
293 }
294 
295 static int
296 set_params__post_init(struct adapter *sc)
297 {
298 	uint32_t param, val;
299 
300 	/* ask for encapsulated CPLs */
301 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
302 	val = 1;
303 	(void)t4vf_set_params(sc, 1, &param, &val);
304 
305 	/* Enable 32b port caps if the firmware supports it. */
306 	param = FW_PARAM_PFVF(PORT_CAPS32);
307 	val = 1;
308 	if (t4vf_set_params(sc, 1, &param, &val) == 0)
309 		sc->params.port_caps32 = 1;
310 
311 	return (0);
312 }
313 
314 #undef FW_PARAM_PFVF
315 #undef FW_PARAM_DEV
316 
317 static int
318 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
319 {
320 	struct vf_resources *vfres;
321 	int nrxq, ntxq, nports;
322 	int itype, iq_avail, navail, rc;
323 
324 	/*
325 	 * Figure out the layout of queues across our VIs and ensure
326 	 * we can allocate enough interrupts for our layout.
327 	 */
328 	vfres = &sc->params.vfres;
329 	nports = sc->params.nports;
330 	bzero(iaq, sizeof(*iaq));
331 
332 	for (itype = INTR_MSIX; itype != 0; itype >>= 1) {
333 		if (itype == INTR_INTX)
334 			continue;
335 
336 		if (itype == INTR_MSIX)
337 			navail = pci_msix_count(sc->dev);
338 		else
339 			navail = pci_msi_count(sc->dev);
340 
341 		if (navail == 0)
342 			continue;
343 
344 		iaq->intr_type = itype;
345 
346 		/*
347 		 * XXX: The Linux driver reserves an Ingress Queue for
348 		 * forwarded interrupts when using MSI (but not MSI-X).
349 		 * It seems it just always asks for 2 interrupts and
350 		 * forwards all rxqs to the forwarded interrupt.
351 		 *
352 		 * We must reserve one IRQ for the for the firmware
353 		 * event queue.
354 		 *
355 		 * Every rxq requires an ingress queue with a free
356 		 * list and interrupts and an egress queue.  Every txq
357 		 * requires an ETH egress queue.
358 		 */
359 		iaq->nirq = T4VF_EXTRA_INTR;
360 
361 		/*
362 		 * First, determine how many queues we can allocate.
363 		 * Start by finding the upper bound on rxqs from the
364 		 * limit on ingress queues.
365 		 */
366 		iq_avail = vfres->niqflint - iaq->nirq;
367 		if (iq_avail < nports) {
368 			device_printf(sc->dev,
369 			    "Not enough ingress queues (%d) for %d ports\n",
370 			    vfres->niqflint, nports);
371 			return (ENXIO);
372 		}
373 
374 		/*
375 		 * Try to honor the cap on interrupts.  If there aren't
376 		 * enough interrupts for at least one interrupt per
377 		 * port, then don't bother, we will just forward all
378 		 * interrupts to one interrupt in that case.
379 		 */
380 		if (iaq->nirq + nports <= navail) {
381 			if (iq_avail > navail - iaq->nirq)
382 				iq_avail = navail - iaq->nirq;
383 		}
384 
385 		nrxq = nports * t4_nrxq;
386 		if (nrxq > iq_avail) {
387 			/*
388 			 * Too many ingress queues.  Use what we can.
389 			 */
390 			nrxq = (iq_avail / nports) * nports;
391 		}
392 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
393 
394 		/*
395 		 * Next, determine the upper bound on txqs from the limit
396 		 * on ETH queues.
397 		 */
398 		if (vfres->nethctrl < nports) {
399 			device_printf(sc->dev,
400 			    "Not enough ETH queues (%d) for %d ports\n",
401 			    vfres->nethctrl, nports);
402 			return (ENXIO);
403 		}
404 
405 		ntxq = nports * t4_ntxq;
406 		if (ntxq > vfres->nethctrl) {
407 			/*
408 			 * Too many ETH queues.  Use what we can.
409 			 */
410 			ntxq = (vfres->nethctrl / nports) * nports;
411 		}
412 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
413 
414 		/*
415 		 * Finally, ensure we have enough egress queues.
416 		 */
417 		if (vfres->neq < nports * 2) {
418 			device_printf(sc->dev,
419 			    "Not enough egress queues (%d) for %d ports\n",
420 			    vfres->neq, nports);
421 			return (ENXIO);
422 		}
423 		if (nrxq + ntxq > vfres->neq) {
424 			/* Just punt and use 1 for everything. */
425 			nrxq = ntxq = nports;
426 		}
427 		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
428 		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
429 		KASSERT(nrxq + ntxq <= vfres->neq, ("too many egress queues"));
430 
431 		/*
432 		 * Do we have enough interrupts?  For MSI the interrupts
433 		 * have to be a power of 2 as well.
434 		 */
435 		iaq->nirq += nrxq;
436 		iaq->ntxq = ntxq;
437 		iaq->nrxq = nrxq;
438 		if (iaq->nirq <= navail &&
439 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
440 			navail = iaq->nirq;
441 			if (itype == INTR_MSIX)
442 				rc = pci_alloc_msix(sc->dev, &navail);
443 			else
444 				rc = pci_alloc_msi(sc->dev, &navail);
445 			if (rc != 0) {
446 				device_printf(sc->dev,
447 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
448 				    itype, rc, iaq->nirq, navail);
449 				return (rc);
450 			}
451 			if (navail == iaq->nirq) {
452 				return (0);
453 			}
454 			pci_release_msi(sc->dev);
455 		}
456 
457 		/* Fall back to a single interrupt. */
458 		iaq->nirq = 1;
459 		navail = iaq->nirq;
460 		if (itype == INTR_MSIX)
461 			rc = pci_alloc_msix(sc->dev, &navail);
462 		else
463 			rc = pci_alloc_msi(sc->dev, &navail);
464 		if (rc != 0)
465 			device_printf(sc->dev,
466 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
467 			    itype, rc, iaq->nirq, navail);
468 		return (rc);
469 	}
470 
471 	device_printf(sc->dev,
472 	    "failed to find a usable interrupt type.  "
473 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
474 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
475 
476 	return (ENXIO);
477 }
478 
479 static int
480 t4vf_attach(device_t dev)
481 {
482 	struct adapter *sc;
483 	int rc = 0, i, j, rqidx, tqidx, n, p, pmask;
484 	struct make_dev_args mda;
485 	struct intrs_and_queues iaq;
486 	struct sge *s;
487 
488 	sc = device_get_softc(dev);
489 	sc->dev = dev;
490 	sysctl_ctx_init(&sc->ctx);
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 	pmask = sc->params.vfres.pmask;
622 	for_each_port(sc, i) {
623 		struct port_info *pi;
624 		uint8_t mac[ETHER_ADDR_LEN];
625 
626 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
627 		sc->port[i] = pi;
628 
629 		/* These must be set before t4_port_init */
630 		pi->adapter = sc;
631 		pi->port_id = i;
632 		pi->nvi = 1;
633 		pi->vi = malloc(sizeof(struct vi_info) * pi->nvi, M_CXGBE,
634 		    M_ZERO | M_WAITOK);
635 
636 		/*
637 		 * Allocate the "main" VI and initialize parameters
638 		 * like mac addr.
639 		 */
640 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
641 		if (rc != 0) {
642 			device_printf(dev, "unable to initialize port %d: %d\n",
643 			    i, rc);
644 			free(pi->vi, M_CXGBE);
645 			free(pi, M_CXGBE);
646 			sc->port[i] = NULL;
647 			goto done;
648 		}
649 
650 		/* Prefer the MAC address set by the PF, if there is one. */
651 		n = 1;
652 		p = ffs(pmask) - 1;
653 		MPASS(p >= 0);
654 		rc = t4vf_get_vf_mac(sc, p, &n, mac);
655 		if (rc == 0 && n == 1)
656 			t4_os_set_hw_addr(pi, mac);
657 		pmask &= ~(1 << p);
658 
659 		/* No t4_link_start. */
660 
661 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
662 		    device_get_nameunit(dev), i);
663 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
664 		sc->chan_map[pi->tx_chan] = i;
665 
666 		/* All VIs on this port share this media. */
667 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
668 		    cxgbe_media_status);
669 
670 		pi->dev = device_add_child(dev, sc->names->vf_ifnet_name, -1);
671 		if (pi->dev == NULL) {
672 			device_printf(dev,
673 			    "failed to add device for port %d.\n", i);
674 			rc = ENXIO;
675 			goto done;
676 		}
677 		pi->vi[0].dev = pi->dev;
678 		device_set_softc(pi->dev, pi);
679 	}
680 
681 	/*
682 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
683 	 */
684 	rc = cfg_itype_and_nqueues(sc, &iaq);
685 	if (rc != 0)
686 		goto done; /* error message displayed already */
687 
688 	sc->intr_type = iaq.intr_type;
689 	sc->intr_count = iaq.nirq;
690 
691 	s = &sc->sge;
692 	s->nrxq = sc->params.nports * iaq.nrxq;
693 	s->ntxq = sc->params.nports * iaq.ntxq;
694 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
695 	s->neq += sc->params.nports;	/* ctrl queues: 1 per port */
696 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
697 
698 	s->iqmap_sz = s->niq;
699 	s->eqmap_sz = s->neq;
700 
701 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
702 	    M_ZERO | M_WAITOK);
703 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
704 	    M_ZERO | M_WAITOK);
705 	s->iqmap = malloc(s->iqmap_sz * sizeof(struct sge_iq *), M_CXGBE,
706 	    M_ZERO | M_WAITOK);
707 	s->eqmap = malloc(s->eqmap_sz * sizeof(struct sge_eq *), M_CXGBE,
708 	    M_ZERO | M_WAITOK);
709 
710 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
711 	    M_ZERO | M_WAITOK);
712 
713 	/*
714 	 * Second pass over the ports.  This time we know the number of rx and
715 	 * tx queues that each port should get.
716 	 */
717 	rqidx = tqidx = 0;
718 	for_each_port(sc, i) {
719 		struct port_info *pi = sc->port[i];
720 		struct vi_info *vi;
721 
722 		if (pi == NULL)
723 			continue;
724 
725 		for_each_vi(pi, j, vi) {
726 			vi->pi = pi;
727 			vi->adapter = sc;
728 			vi->qsize_rxq = t4_qsize_rxq;
729 			vi->qsize_txq = t4_qsize_txq;
730 
731 			vi->first_rxq = rqidx;
732 			vi->first_txq = tqidx;
733 			vi->tmr_idx = t4_tmr_idx;
734 			vi->pktc_idx = t4_pktc_idx;
735 			vi->nrxq = j == 0 ? iaq.nrxq: 1;
736 			vi->ntxq = j == 0 ? iaq.ntxq: 1;
737 
738 			rqidx += vi->nrxq;
739 			tqidx += vi->ntxq;
740 
741 			vi->rsrv_noflowq = 0;
742 		}
743 	}
744 
745 	rc = t4_setup_intr_handlers(sc);
746 	if (rc != 0) {
747 		device_printf(dev,
748 		    "failed to setup interrupt handlers: %d\n", rc);
749 		goto done;
750 	}
751 
752 	rc = bus_generic_attach(dev);
753 	if (rc != 0) {
754 		device_printf(dev,
755 		    "failed to attach all child ports: %d\n", rc);
756 		goto done;
757 	}
758 
759 	device_printf(dev,
760 	    "%d ports, %d %s interrupt%s, %d eq, %d iq\n",
761 	    sc->params.nports, sc->intr_count, sc->intr_type == INTR_MSIX ?
762 	    "MSI-X" : "MSI", sc->intr_count > 1 ? "s" : "", sc->sge.neq,
763 	    sc->sge.niq);
764 
765 done:
766 	if (rc != 0)
767 		t4_detach_common(dev);
768 	else
769 		t4_sysctls(sc);
770 
771 	return (rc);
772 }
773 
774 static void
775 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
776 {
777 
778 	/* 0x3f is used as the revision for VFs. */
779 	regs->version = chip_id(sc) | (0x3f << 10);
780 	t4_get_regs(sc, buf, regs->len);
781 }
782 
783 static void
784 t4_clr_vi_stats(struct adapter *sc)
785 {
786 	int reg;
787 
788 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
789 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
790 		t4_write_reg(sc, VF_MPS_REG(reg), 0);
791 }
792 
793 static int
794 t4vf_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
795     struct thread *td)
796 {
797 	int rc;
798 	struct adapter *sc = dev->si_drv1;
799 
800 	rc = priv_check(td, PRIV_DRIVER);
801 	if (rc != 0)
802 		return (rc);
803 
804 	switch (cmd) {
805 	case CHELSIO_T4_GETREG: {
806 		struct t4_reg *edata = (struct t4_reg *)data;
807 
808 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
809 			return (EFAULT);
810 
811 		if (edata->size == 4)
812 			edata->val = t4_read_reg(sc, edata->addr);
813 		else if (edata->size == 8)
814 			edata->val = t4_read_reg64(sc, edata->addr);
815 		else
816 			return (EINVAL);
817 
818 		break;
819 	}
820 	case CHELSIO_T4_SETREG: {
821 		struct t4_reg *edata = (struct t4_reg *)data;
822 
823 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
824 			return (EFAULT);
825 
826 		if (edata->size == 4) {
827 			if (edata->val & 0xffffffff00000000)
828 				return (EINVAL);
829 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
830 		} else if (edata->size == 8)
831 			t4_write_reg64(sc, edata->addr, edata->val);
832 		else
833 			return (EINVAL);
834 		break;
835 	}
836 	case CHELSIO_T4_REGDUMP: {
837 		struct t4_regdump *regs = (struct t4_regdump *)data;
838 		int reglen = t4_get_regs_len(sc);
839 		uint8_t *buf;
840 
841 		if (regs->len < reglen) {
842 			regs->len = reglen; /* hint to the caller */
843 			return (ENOBUFS);
844 		}
845 
846 		regs->len = reglen;
847 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
848 		get_regs(sc, regs, buf);
849 		rc = copyout(buf, regs->data, reglen);
850 		free(buf, M_CXGBE);
851 		break;
852 	}
853 	case CHELSIO_T4_CLEAR_STATS: {
854 		int i, v;
855 		u_int port_id = *(uint32_t *)data;
856 		struct port_info *pi;
857 		struct vi_info *vi;
858 
859 		if (port_id >= sc->params.nports)
860 			return (EINVAL);
861 		pi = sc->port[port_id];
862 
863 		/* MAC stats */
864 		pi->tx_parse_error = 0;
865 		t4_clr_vi_stats(sc);
866 
867 		/*
868 		 * Since this command accepts a port, clear stats for
869 		 * all VIs on this port.
870 		 */
871 		for_each_vi(pi, v, vi) {
872 			if (vi->flags & VI_INIT_DONE) {
873 				struct sge_rxq *rxq;
874 				struct sge_txq *txq;
875 
876 				for_each_rxq(vi, i, rxq) {
877 #if defined(INET) || defined(INET6)
878 					rxq->lro.lro_queued = 0;
879 					rxq->lro.lro_flushed = 0;
880 #endif
881 					rxq->rxcsum = 0;
882 					rxq->vlan_extraction = 0;
883 				}
884 
885 				for_each_txq(vi, i, txq) {
886 					txq->txcsum = 0;
887 					txq->tso_wrs = 0;
888 					txq->vlan_insertion = 0;
889 					txq->imm_wrs = 0;
890 					txq->sgl_wrs = 0;
891 					txq->txpkt_wrs = 0;
892 					txq->txpkts0_wrs = 0;
893 					txq->txpkts1_wrs = 0;
894 					txq->txpkts0_pkts = 0;
895 					txq->txpkts1_pkts = 0;
896 					txq->txpkts_flush = 0;
897 					mp_ring_reset_stats(txq->r);
898 				}
899 			}
900 		}
901 		break;
902 	}
903 	case CHELSIO_T4_SCHED_CLASS:
904 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
905 		break;
906 	case CHELSIO_T4_SCHED_QUEUE:
907 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
908 		break;
909 	default:
910 		rc = ENOTTY;
911 	}
912 
913 	return (rc);
914 }
915 
916 static device_method_t t4vf_methods[] = {
917 	DEVMETHOD(device_probe,		t4vf_probe),
918 	DEVMETHOD(device_attach,	t4vf_attach),
919 	DEVMETHOD(device_detach,	t4_detach_common),
920 
921 	DEVMETHOD_END
922 };
923 
924 static driver_t t4vf_driver = {
925 	"t4vf",
926 	t4vf_methods,
927 	sizeof(struct adapter)
928 };
929 
930 static device_method_t t5vf_methods[] = {
931 	DEVMETHOD(device_probe,		t5vf_probe),
932 	DEVMETHOD(device_attach,	t4vf_attach),
933 	DEVMETHOD(device_detach,	t4_detach_common),
934 
935 	DEVMETHOD_END
936 };
937 
938 static driver_t t5vf_driver = {
939 	"t5vf",
940 	t5vf_methods,
941 	sizeof(struct adapter)
942 };
943 
944 static device_method_t t6vf_methods[] = {
945 	DEVMETHOD(device_probe,		t6vf_probe),
946 	DEVMETHOD(device_attach,	t4vf_attach),
947 	DEVMETHOD(device_detach,	t4_detach_common),
948 
949 	DEVMETHOD_END
950 };
951 
952 static driver_t t6vf_driver = {
953 	"t6vf",
954 	t6vf_methods,
955 	sizeof(struct adapter)
956 };
957 
958 static driver_t cxgbev_driver = {
959 	"cxgbev",
960 	cxgbe_methods,
961 	sizeof(struct port_info)
962 };
963 
964 static driver_t cxlv_driver = {
965 	"cxlv",
966 	cxgbe_methods,
967 	sizeof(struct port_info)
968 };
969 
970 static driver_t ccv_driver = {
971 	"ccv",
972 	cxgbe_methods,
973 	sizeof(struct port_info)
974 };
975 
976 DRIVER_MODULE(t4vf, pci, t4vf_driver, 0, 0);
977 MODULE_VERSION(t4vf, 1);
978 MODULE_DEPEND(t4vf, t4nex, 1, 1, 1);
979 
980 DRIVER_MODULE(t5vf, pci, t5vf_driver, 0, 0);
981 MODULE_VERSION(t5vf, 1);
982 MODULE_DEPEND(t5vf, t5nex, 1, 1, 1);
983 
984 DRIVER_MODULE(t6vf, pci, t6vf_driver, 0, 0);
985 MODULE_VERSION(t6vf, 1);
986 MODULE_DEPEND(t6vf, t6nex, 1, 1, 1);
987 
988 DRIVER_MODULE(cxgbev, t4vf, cxgbev_driver, 0, 0);
989 MODULE_VERSION(cxgbev, 1);
990 
991 DRIVER_MODULE(cxlv, t5vf, cxlv_driver, 0, 0);
992 MODULE_VERSION(cxlv, 1);
993 
994 DRIVER_MODULE(ccv, t6vf, ccv_driver, 0, 0);
995 MODULE_VERSION(ccv, 1);
996