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: stable/11/sys/dev/cxgbe/t4_vf.c 355253 2019-11-30 21:11:17Z np $");
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
62struct 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
69struct {
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
132static d_ioctl_t t4vf_ioctl;
133
134static struct cdevsw t4vf_cdevsw = {
135       .d_version = D_VERSION,
136       .d_ioctl = t4vf_ioctl,
137       .d_name = "t4vf",
138};
139
140static int
141t4vf_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
156static int
157t5vf_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
172static int
173t6vf_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
195static int
196get_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
230static int
231get_params__post_init(struct adapter *sc)
232{
233	int rc;
234
235	rc = -t4vf_get_sge_params(sc);
236	if (rc != 0) {
237		device_printf(sc->dev,
238		    "unable to retrieve adapter SGE parameters: %d\n", rc);
239		return (rc);
240	}
241
242	rc = -t4vf_get_rss_glb_config(sc);
243	if (rc != 0) {
244		device_printf(sc->dev,
245		    "unable to retrieve adapter RSS parameters: %d\n", rc);
246		return (rc);
247	}
248	if (sc->params.rss.mode != FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
249		device_printf(sc->dev,
250		    "unable to operate with global RSS mode %d\n",
251		    sc->params.rss.mode);
252		return (EINVAL);
253	}
254
255	rc = t4_read_chip_settings(sc);
256	if (rc != 0)
257		return (rc);
258
259	/*
260	 * Grab our Virtual Interface resource allocation, extract the
261	 * features that we're interested in and do a bit of sanity testing on
262	 * what we discover.
263	 */
264	rc = -t4vf_get_vfres(sc);
265	if (rc != 0) {
266		device_printf(sc->dev,
267		    "unable to get virtual interface resources: %d\n", rc);
268		return (rc);
269	}
270
271	/*
272	 * Check for various parameter sanity issues.
273	 */
274	if (sc->params.vfres.pmask == 0) {
275		device_printf(sc->dev, "no port access configured/usable!\n");
276		return (EINVAL);
277	}
278	if (sc->params.vfres.nvi == 0) {
279		device_printf(sc->dev,
280		    "no virtual interfaces configured/usable!\n");
281		return (EINVAL);
282	}
283	sc->params.portvec = sc->params.vfres.pmask;
284
285	return (0);
286}
287
288static int
289set_params__post_init(struct adapter *sc)
290{
291	uint32_t param, val;
292
293	/* ask for encapsulated CPLs */
294	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
295	val = 1;
296	(void)t4vf_set_params(sc, 1, &param, &val);
297
298	/* Enable 32b port caps if the firmware supports it. */
299	param = FW_PARAM_PFVF(PORT_CAPS32);
300	val = 1;
301	if (t4vf_set_params(sc, 1, &param, &val) == 0)
302		sc->params.port_caps32 = 1;
303
304	return (0);
305}
306
307#undef FW_PARAM_PFVF
308#undef FW_PARAM_DEV
309
310static int
311cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
312{
313	struct vf_resources *vfres;
314	int nrxq, ntxq, nports;
315	int itype, iq_avail, navail, rc;
316
317	/*
318	 * Figure out the layout of queues across our VIs and ensure
319	 * we can allocate enough interrupts for our layout.
320	 */
321	vfres = &sc->params.vfres;
322	nports = sc->params.nports;
323	bzero(iaq, sizeof(*iaq));
324
325	for (itype = INTR_MSIX; itype != 0; itype >>= 1) {
326		if (itype == INTR_INTX)
327			continue;
328
329		if (itype == INTR_MSIX)
330			navail = pci_msix_count(sc->dev);
331		else
332			navail = pci_msi_count(sc->dev);
333
334		if (navail == 0)
335			continue;
336
337		iaq->intr_type = itype;
338
339		/*
340		 * XXX: The Linux driver reserves an Ingress Queue for
341		 * forwarded interrupts when using MSI (but not MSI-X).
342		 * It seems it just always asks for 2 interrupts and
343		 * forwards all rxqs to the forwarded interrupt.
344		 *
345		 * We must reserve one IRQ for the for the firmware
346		 * event queue.
347		 *
348		 * Every rxq requires an ingress queue with a free
349		 * list and interrupts and an egress queue.  Every txq
350		 * requires an ETH egress queue.
351		 */
352		iaq->nirq = T4VF_EXTRA_INTR;
353
354		/*
355		 * First, determine how many queues we can allocate.
356		 * Start by finding the upper bound on rxqs from the
357		 * limit on ingress queues.
358		 */
359		iq_avail = vfres->niqflint - iaq->nirq;
360		if (iq_avail < nports) {
361			device_printf(sc->dev,
362			    "Not enough ingress queues (%d) for %d ports\n",
363			    vfres->niqflint, nports);
364			return (ENXIO);
365		}
366
367		/*
368		 * Try to honor the cap on interrupts.  If there aren't
369		 * enough interrupts for at least one interrupt per
370		 * port, then don't bother, we will just forward all
371		 * interrupts to one interrupt in that case.
372		 */
373		if (iaq->nirq + nports <= navail) {
374			if (iq_avail > navail - iaq->nirq)
375				iq_avail = navail - iaq->nirq;
376		}
377
378		nrxq = nports * t4_nrxq;
379		if (nrxq > iq_avail) {
380			/*
381			 * Too many ingress queues.  Use what we can.
382			 */
383			nrxq = (iq_avail / nports) * nports;
384		}
385		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
386
387		/*
388		 * Next, determine the upper bound on txqs from the limit
389		 * on ETH queues.
390		 */
391		if (vfres->nethctrl < nports) {
392			device_printf(sc->dev,
393			    "Not enough ETH queues (%d) for %d ports\n",
394			    vfres->nethctrl, nports);
395			return (ENXIO);
396		}
397
398		ntxq = nports * t4_ntxq;
399		if (ntxq > vfres->nethctrl) {
400			/*
401			 * Too many ETH queues.  Use what we can.
402			 */
403			ntxq = (vfres->nethctrl / nports) * nports;
404		}
405		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
406
407		/*
408		 * Finally, ensure we have enough egress queues.
409		 */
410		if (vfres->neq < nports * 2) {
411			device_printf(sc->dev,
412			    "Not enough egress queues (%d) for %d ports\n",
413			    vfres->neq, nports);
414			return (ENXIO);
415		}
416		if (nrxq + ntxq > vfres->neq) {
417			/* Just punt and use 1 for everything. */
418			nrxq = ntxq = nports;
419		}
420		KASSERT(nrxq <= iq_avail, ("too many ingress queues"));
421		KASSERT(ntxq <= vfres->nethctrl, ("too many ETH queues"));
422		KASSERT(nrxq + ntxq <= vfres->neq, ("too many egress queues"));
423
424		/*
425		 * Do we have enough interrupts?  For MSI the interrupts
426		 * have to be a power of 2 as well.
427		 */
428		iaq->nirq += nrxq;
429		iaq->ntxq = ntxq;
430		iaq->nrxq = nrxq;
431		if (iaq->nirq <= navail &&
432		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
433			navail = iaq->nirq;
434			if (itype == INTR_MSIX)
435				rc = pci_alloc_msix(sc->dev, &navail);
436			else
437				rc = pci_alloc_msi(sc->dev, &navail);
438			if (rc != 0) {
439				device_printf(sc->dev,
440		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
441				    itype, rc, iaq->nirq, navail);
442				return (rc);
443			}
444			if (navail == iaq->nirq) {
445				return (0);
446			}
447			pci_release_msi(sc->dev);
448		}
449
450		/* Fall back to a single interrupt. */
451		iaq->nirq = 1;
452		navail = iaq->nirq;
453		if (itype == INTR_MSIX)
454			rc = pci_alloc_msix(sc->dev, &navail);
455		else
456			rc = pci_alloc_msi(sc->dev, &navail);
457		if (rc != 0)
458			device_printf(sc->dev,
459		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
460			    itype, rc, iaq->nirq, navail);
461		return (rc);
462	}
463
464	device_printf(sc->dev,
465	    "failed to find a usable interrupt type.  "
466	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
467	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
468
469	return (ENXIO);
470}
471
472static int
473t4vf_attach(device_t dev)
474{
475	struct adapter *sc;
476	int rc = 0, i, j, rqidx, tqidx;
477	struct make_dev_args mda;
478	struct intrs_and_queues iaq;
479	struct sge *s;
480
481	sc = device_get_softc(dev);
482	sc->dev = dev;
483	pci_enable_busmaster(dev);
484	pci_set_max_read_req(dev, 4096);
485	sc->params.pci.mps = pci_get_max_payload(dev);
486
487	sc->flags |= IS_VF;
488	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
489
490	sc->sge_gts_reg = VF_SGE_REG(A_SGE_VF_GTS);
491	sc->sge_kdoorbell_reg = VF_SGE_REG(A_SGE_VF_KDOORBELL);
492	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
493	    device_get_nameunit(dev));
494	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
495	t4_add_adapter(sc);
496
497	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
498	TAILQ_INIT(&sc->sfl);
499	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
500
501	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
502
503	rc = t4_map_bars_0_and_4(sc);
504	if (rc != 0)
505		goto done; /* error message displayed already */
506
507	rc = -t4vf_prep_adapter(sc);
508	if (rc != 0)
509		goto done;
510
511	t4_init_devnames(sc);
512	if (sc->names == NULL) {
513		rc = ENOTSUP;
514		goto done; /* error message displayed already */
515	}
516
517	/*
518	 * Leave the 'pf' and 'mbox' values as zero.  This ensures
519	 * that various firmware messages do not set the fields which
520	 * is the correct thing to do for a VF.
521	 */
522
523	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
524
525	make_dev_args_init(&mda);
526	mda.mda_devsw = &t4vf_cdevsw;
527	mda.mda_uid = UID_ROOT;
528	mda.mda_gid = GID_WHEEL;
529	mda.mda_mode = 0600;
530	mda.mda_si_drv1 = sc;
531	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
532	if (rc != 0)
533		device_printf(dev, "failed to create nexus char device: %d.\n",
534		    rc);
535
536#if defined(__i386__)
537	if ((cpu_feature & CPUID_CX8) == 0) {
538		device_printf(dev, "64 bit atomics not available.\n");
539		rc = ENOTSUP;
540		goto done;
541	}
542#endif
543
544	/*
545	 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
546	 * 2.6.31 and later we can't call pci_reset_function() in order to
547	 * issue an FLR because of a self- deadlock on the device semaphore.
548	 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
549	 * cases where they're needed -- for instance, some versions of KVM
550	 * fail to reset "Assigned Devices" when the VM reboots.  Therefore we
551	 * use the firmware based reset in order to reset any per function
552	 * state.
553	 */
554	rc = -t4vf_fw_reset(sc);
555	if (rc != 0) {
556		device_printf(dev, "FW reset failed: %d\n", rc);
557		goto done;
558	}
559	sc->flags |= FW_OK;
560
561	/*
562	 * Grab basic operational parameters.  These will predominantly have
563	 * been set up by the Physical Function Driver or will be hard coded
564	 * into the adapter.  We just have to live with them ...  Note that
565	 * we _must_ get our VPD parameters before our SGE parameters because
566	 * we need to know the adapter's core clock from the VPD in order to
567	 * properly decode the SGE Timer Values.
568	 */
569	rc = get_params__pre_init(sc);
570	if (rc != 0)
571		goto done; /* error message displayed already */
572	rc = get_params__post_init(sc);
573	if (rc != 0)
574		goto done; /* error message displayed already */
575
576	rc = set_params__post_init(sc);
577	if (rc != 0)
578		goto done; /* error message displayed already */
579
580	rc = t4_map_bar_2(sc);
581	if (rc != 0)
582		goto done; /* error message displayed already */
583
584	rc = t4_create_dma_tag(sc);
585	if (rc != 0)
586		goto done; /* error message displayed already */
587
588	/*
589	 * The number of "ports" which we support is equal to the number of
590	 * Virtual Interfaces with which we've been provisioned.
591	 */
592	sc->params.nports = imin(sc->params.vfres.nvi, MAX_NPORTS);
593
594	/*
595	 * We may have been provisioned with more VIs than the number of
596	 * ports we're allowed to access (our Port Access Rights Mask).
597	 * Just use a single VI for each port.
598	 */
599	sc->params.nports = imin(sc->params.nports,
600	    bitcount32(sc->params.vfres.pmask));
601
602#ifdef notyet
603	/*
604	 * XXX: The Linux VF driver will lower nports if it thinks there
605	 * are too few resources in vfres (niqflint, nethctrl, neq).
606	 */
607#endif
608
609	/*
610	 * First pass over all the ports - allocate VIs and initialize some
611	 * basic parameters like mac address, port type, etc.
612	 */
613	for_each_port(sc, i) {
614		struct port_info *pi;
615
616		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
617		sc->port[i] = pi;
618
619		/* These must be set before t4_port_init */
620		pi->adapter = sc;
621		pi->port_id = i;
622		pi->nvi = 1;
623		pi->vi = malloc(sizeof(struct vi_info) * pi->nvi, M_CXGBE,
624		    M_ZERO | M_WAITOK);
625
626		/*
627		 * Allocate the "main" VI and initialize parameters
628		 * like mac addr.
629		 */
630		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
631		if (rc != 0) {
632			device_printf(dev, "unable to initialize port %d: %d\n",
633			    i, rc);
634			free(pi->vi, M_CXGBE);
635			free(pi, M_CXGBE);
636			sc->port[i] = NULL;
637			goto done;
638		}
639
640		/* No t4_link_start. */
641
642		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
643		    device_get_nameunit(dev), i);
644		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
645		sc->chan_map[pi->tx_chan] = i;
646
647		/* All VIs on this port share this media. */
648		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
649		    cxgbe_media_status);
650
651		pi->dev = device_add_child(dev, sc->names->vf_ifnet_name, -1);
652		if (pi->dev == NULL) {
653			device_printf(dev,
654			    "failed to add device for port %d.\n", i);
655			rc = ENXIO;
656			goto done;
657		}
658		pi->vi[0].dev = pi->dev;
659		device_set_softc(pi->dev, pi);
660	}
661
662	/*
663	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
664	 */
665	rc = cfg_itype_and_nqueues(sc, &iaq);
666	if (rc != 0)
667		goto done; /* error message displayed already */
668
669	sc->intr_type = iaq.intr_type;
670	sc->intr_count = iaq.nirq;
671
672	s = &sc->sge;
673	s->nrxq = sc->params.nports * iaq.nrxq;
674	s->ntxq = sc->params.nports * iaq.ntxq;
675	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
676	s->neq += sc->params.nports;	/* ctrl queues: 1 per port */
677	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
678
679	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
680	    M_ZERO | M_WAITOK);
681	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
682	    M_ZERO | M_WAITOK);
683	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
684	    M_ZERO | M_WAITOK);
685	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
686	    M_ZERO | M_WAITOK);
687
688	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
689	    M_ZERO | M_WAITOK);
690
691	/*
692	 * Second pass over the ports.  This time we know the number of rx and
693	 * tx queues that each port should get.
694	 */
695	rqidx = tqidx = 0;
696	for_each_port(sc, i) {
697		struct port_info *pi = sc->port[i];
698		struct vi_info *vi;
699
700		if (pi == NULL)
701			continue;
702
703		for_each_vi(pi, j, vi) {
704			vi->pi = pi;
705			vi->qsize_rxq = t4_qsize_rxq;
706			vi->qsize_txq = t4_qsize_txq;
707
708			vi->first_rxq = rqidx;
709			vi->first_txq = tqidx;
710			vi->tmr_idx = t4_tmr_idx;
711			vi->pktc_idx = t4_pktc_idx;
712			vi->nrxq = j == 0 ? iaq.nrxq: 1;
713			vi->ntxq = j == 0 ? iaq.ntxq: 1;
714
715			rqidx += vi->nrxq;
716			tqidx += vi->ntxq;
717
718			vi->rsrv_noflowq = 0;
719		}
720	}
721
722	rc = t4_setup_intr_handlers(sc);
723	if (rc != 0) {
724		device_printf(dev,
725		    "failed to setup interrupt handlers: %d\n", rc);
726		goto done;
727	}
728
729	rc = bus_generic_attach(dev);
730	if (rc != 0) {
731		device_printf(dev,
732		    "failed to attach all child ports: %d\n", rc);
733		goto done;
734	}
735
736	device_printf(dev,
737	    "%d ports, %d %s interrupt%s, %d eq, %d iq\n",
738	    sc->params.nports, sc->intr_count, sc->intr_type == INTR_MSIX ?
739	    "MSI-X" : "MSI", sc->intr_count > 1 ? "s" : "", sc->sge.neq,
740	    sc->sge.niq);
741
742done:
743	if (rc != 0)
744		t4_detach_common(dev);
745	else
746		t4_sysctls(sc);
747
748	return (rc);
749}
750
751static void
752get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
753{
754
755	/* 0x3f is used as the revision for VFs. */
756	regs->version = chip_id(sc) | (0x3f << 10);
757	t4_get_regs(sc, buf, regs->len);
758}
759
760static void
761t4_clr_vi_stats(struct adapter *sc)
762{
763	int reg;
764
765	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
766	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
767		t4_write_reg(sc, VF_MPS_REG(reg), 0);
768}
769
770static int
771t4vf_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
772    struct thread *td)
773{
774	int rc;
775	struct adapter *sc = dev->si_drv1;
776
777	rc = priv_check(td, PRIV_DRIVER);
778	if (rc != 0)
779		return (rc);
780
781	switch (cmd) {
782	case CHELSIO_T4_GETREG: {
783		struct t4_reg *edata = (struct t4_reg *)data;
784
785		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
786			return (EFAULT);
787
788		if (edata->size == 4)
789			edata->val = t4_read_reg(sc, edata->addr);
790		else if (edata->size == 8)
791			edata->val = t4_read_reg64(sc, edata->addr);
792		else
793			return (EINVAL);
794
795		break;
796	}
797	case CHELSIO_T4_SETREG: {
798		struct t4_reg *edata = (struct t4_reg *)data;
799
800		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
801			return (EFAULT);
802
803		if (edata->size == 4) {
804			if (edata->val & 0xffffffff00000000)
805				return (EINVAL);
806			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
807		} else if (edata->size == 8)
808			t4_write_reg64(sc, edata->addr, edata->val);
809		else
810			return (EINVAL);
811		break;
812	}
813	case CHELSIO_T4_REGDUMP: {
814		struct t4_regdump *regs = (struct t4_regdump *)data;
815		int reglen = t4_get_regs_len(sc);
816		uint8_t *buf;
817
818		if (regs->len < reglen) {
819			regs->len = reglen; /* hint to the caller */
820			return (ENOBUFS);
821		}
822
823		regs->len = reglen;
824		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
825		get_regs(sc, regs, buf);
826		rc = copyout(buf, regs->data, reglen);
827		free(buf, M_CXGBE);
828		break;
829	}
830	case CHELSIO_T4_CLEAR_STATS: {
831		int i, v;
832		u_int port_id = *(uint32_t *)data;
833		struct port_info *pi;
834		struct vi_info *vi;
835
836		if (port_id >= sc->params.nports)
837			return (EINVAL);
838		pi = sc->port[port_id];
839
840		/* MAC stats */
841		pi->tx_parse_error = 0;
842		t4_clr_vi_stats(sc);
843
844		/*
845		 * Since this command accepts a port, clear stats for
846		 * all VIs on this port.
847		 */
848		for_each_vi(pi, v, vi) {
849			if (vi->flags & VI_INIT_DONE) {
850				struct sge_rxq *rxq;
851				struct sge_txq *txq;
852
853				for_each_rxq(vi, i, rxq) {
854#if defined(INET) || defined(INET6)
855					rxq->lro.lro_queued = 0;
856					rxq->lro.lro_flushed = 0;
857#endif
858					rxq->rxcsum = 0;
859					rxq->vlan_extraction = 0;
860				}
861
862				for_each_txq(vi, i, txq) {
863					txq->txcsum = 0;
864					txq->tso_wrs = 0;
865					txq->vlan_insertion = 0;
866					txq->imm_wrs = 0;
867					txq->sgl_wrs = 0;
868					txq->txpkt_wrs = 0;
869					txq->txpkts0_wrs = 0;
870					txq->txpkts1_wrs = 0;
871					txq->txpkts0_pkts = 0;
872					txq->txpkts1_pkts = 0;
873					mp_ring_reset_stats(txq->r);
874				}
875			}
876		}
877		break;
878	}
879	case CHELSIO_T4_SCHED_CLASS:
880		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
881		break;
882	case CHELSIO_T4_SCHED_QUEUE:
883		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
884		break;
885	default:
886		rc = ENOTTY;
887	}
888
889	return (rc);
890}
891
892static device_method_t t4vf_methods[] = {
893	DEVMETHOD(device_probe,		t4vf_probe),
894	DEVMETHOD(device_attach,	t4vf_attach),
895	DEVMETHOD(device_detach,	t4_detach_common),
896
897	DEVMETHOD_END
898};
899
900static driver_t t4vf_driver = {
901	"t4vf",
902	t4vf_methods,
903	sizeof(struct adapter)
904};
905
906static device_method_t t5vf_methods[] = {
907	DEVMETHOD(device_probe,		t5vf_probe),
908	DEVMETHOD(device_attach,	t4vf_attach),
909	DEVMETHOD(device_detach,	t4_detach_common),
910
911	DEVMETHOD_END
912};
913
914static driver_t t5vf_driver = {
915	"t5vf",
916	t5vf_methods,
917	sizeof(struct adapter)
918};
919
920static device_method_t t6vf_methods[] = {
921	DEVMETHOD(device_probe,		t6vf_probe),
922	DEVMETHOD(device_attach,	t4vf_attach),
923	DEVMETHOD(device_detach,	t4_detach_common),
924
925	DEVMETHOD_END
926};
927
928static driver_t t6vf_driver = {
929	"t6vf",
930	t6vf_methods,
931	sizeof(struct adapter)
932};
933
934static driver_t cxgbev_driver = {
935	"cxgbev",
936	cxgbe_methods,
937	sizeof(struct port_info)
938};
939
940static driver_t cxlv_driver = {
941	"cxlv",
942	cxgbe_methods,
943	sizeof(struct port_info)
944};
945
946static driver_t ccv_driver = {
947	"ccv",
948	cxgbe_methods,
949	sizeof(struct port_info)
950};
951
952static devclass_t t4vf_devclass, t5vf_devclass, t6vf_devclass;
953static devclass_t cxgbev_devclass, cxlv_devclass, ccv_devclass;
954
955DRIVER_MODULE(t4vf, pci, t4vf_driver, t4vf_devclass, 0, 0);
956MODULE_VERSION(t4vf, 1);
957MODULE_DEPEND(t4vf, t4nex, 1, 1, 1);
958
959DRIVER_MODULE(t5vf, pci, t5vf_driver, t5vf_devclass, 0, 0);
960MODULE_VERSION(t5vf, 1);
961MODULE_DEPEND(t5vf, t5nex, 1, 1, 1);
962
963DRIVER_MODULE(t6vf, pci, t6vf_driver, t6vf_devclass, 0, 0);
964MODULE_VERSION(t6vf, 1);
965MODULE_DEPEND(t6vf, t6nex, 1, 1, 1);
966
967DRIVER_MODULE(cxgbev, t4vf, cxgbev_driver, cxgbev_devclass, 0, 0);
968MODULE_VERSION(cxgbev, 1);
969
970DRIVER_MODULE(cxlv, t5vf, cxlv_driver, cxlv_devclass, 0, 0);
971MODULE_VERSION(cxlv, 1);
972
973DRIVER_MODULE(ccv, t6vf, ccv_driver, ccv_devclass, 0, 0);
974MODULE_VERSION(ccv, 1);
975