Home | History | Annotate | Line # | Download | only in pci
pci.c revision 1.148
      1 /*	$NetBSD: pci.c,v 1.148 2015/08/24 23:55:04 pooka Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1995, 1996, 1997, 1998
      5  *     Christopher G. Demetriou.  All rights reserved.
      6  * Copyright (c) 1994 Charles M. Hannum.  All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. All advertising materials mentioning features or use of this software
     17  *    must display the following acknowledgement:
     18  *	This product includes software developed by Charles M. Hannum.
     19  * 4. The name of the author may not be used to endorse or promote products
     20  *    derived from this software without specific prior written permission.
     21  *
     22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     32  */
     33 
     34 /*
     35  * PCI bus autoconfiguration.
     36  */
     37 
     38 #include <sys/cdefs.h>
     39 __KERNEL_RCSID(0, "$NetBSD: pci.c,v 1.148 2015/08/24 23:55:04 pooka Exp $");
     40 
     41 #ifdef _KERNEL_OPT
     42 #include "opt_pci.h"
     43 #endif
     44 
     45 #include <sys/param.h>
     46 #include <sys/malloc.h>
     47 #include <sys/systm.h>
     48 #include <sys/device.h>
     49 #include <sys/module.h>
     50 
     51 #include <dev/pci/pcireg.h>
     52 #include <dev/pci/pcivar.h>
     53 #include <dev/pci/pcidevs.h>
     54 
     55 #include <net/if.h>
     56 
     57 #include "locators.h"
     58 
     59 static bool pci_child_register(device_t);
     60 
     61 #ifdef PCI_CONFIG_DUMP
     62 int pci_config_dump = 1;
     63 #else
     64 int pci_config_dump = 0;
     65 #endif
     66 
     67 int	pciprint(void *, const char *);
     68 
     69 #ifdef PCI_MACHDEP_ENUMERATE_BUS
     70 #define pci_enumerate_bus PCI_MACHDEP_ENUMERATE_BUS
     71 #else
     72 int pci_enumerate_bus(struct pci_softc *, const int *,
     73     int (*)(const struct pci_attach_args *), struct pci_attach_args *);
     74 #endif
     75 
     76 /*
     77  * Important note about PCI-ISA bridges:
     78  *
     79  * Callbacks are used to configure these devices so that ISA/EISA bridges
     80  * can attach their child busses after PCI configuration is done.
     81  *
     82  * This works because:
     83  *	(1) there can be at most one ISA/EISA bridge per PCI bus, and
     84  *	(2) any ISA/EISA bridges must be attached to primary PCI
     85  *	    busses (i.e. bus zero).
     86  *
     87  * That boils down to: there can only be one of these outstanding
     88  * at a time, it is cleared when configuring PCI bus 0 before any
     89  * subdevices have been found, and it is run after all subdevices
     90  * of PCI bus 0 have been found.
     91  *
     92  * This is needed because there are some (legacy) PCI devices which
     93  * can show up as ISA/EISA devices as well (the prime example of which
     94  * are VGA controllers).  If you attach ISA from a PCI-ISA/EISA bridge,
     95  * and the bridge is seen before the video board is, the board can show
     96  * up as an ISA device, and that can (bogusly) complicate the PCI device's
     97  * attach code, or make the PCI device not be properly attached at all.
     98  *
     99  * We use the generic config_defer() facility to achieve this.
    100  */
    101 
    102 int
    103 pcirescan(device_t self, const char *ifattr, const int *locators)
    104 {
    105 	struct pci_softc *sc = device_private(self);
    106 
    107 	KASSERT(ifattr && !strcmp(ifattr, "pci"));
    108 	KASSERT(locators);
    109 
    110 	pci_enumerate_bus(sc, locators, NULL, NULL);
    111 
    112 	return 0;
    113 }
    114 
    115 int
    116 pcimatch(device_t parent, cfdata_t cf, void *aux)
    117 {
    118 	struct pcibus_attach_args *pba = aux;
    119 
    120 	/* Check the locators */
    121 	if (cf->cf_loc[PCIBUSCF_BUS] != PCIBUSCF_BUS_DEFAULT &&
    122 	    cf->cf_loc[PCIBUSCF_BUS] != pba->pba_bus)
    123 		return 0;
    124 
    125 	/* sanity */
    126 	if (pba->pba_bus < 0 || pba->pba_bus > 255)
    127 		return 0;
    128 
    129 	/*
    130 	 * XXX check other (hardware?) indicators
    131 	 */
    132 
    133 	return 1;
    134 }
    135 
    136 void
    137 pciattach(device_t parent, device_t self, void *aux)
    138 {
    139 	struct pcibus_attach_args *pba = aux;
    140 	struct pci_softc *sc = device_private(self);
    141 	int io_enabled, mem_enabled, mrl_enabled, mrm_enabled, mwi_enabled;
    142 	const char *sep = "";
    143 	static const int wildcard[PCICF_NLOCS] = {
    144 		PCICF_DEV_DEFAULT, PCICF_FUNCTION_DEFAULT
    145 	};
    146 
    147 	sc->sc_dev = self;
    148 
    149 	pci_attach_hook(parent, self, pba);
    150 
    151 	aprint_naive("\n");
    152 	aprint_normal("\n");
    153 
    154 	io_enabled = (pba->pba_flags & PCI_FLAGS_IO_OKAY);
    155 	mem_enabled = (pba->pba_flags & PCI_FLAGS_MEM_OKAY);
    156 	mrl_enabled = (pba->pba_flags & PCI_FLAGS_MRL_OKAY);
    157 	mrm_enabled = (pba->pba_flags & PCI_FLAGS_MRM_OKAY);
    158 	mwi_enabled = (pba->pba_flags & PCI_FLAGS_MWI_OKAY);
    159 
    160 	if (io_enabled == 0 && mem_enabled == 0) {
    161 		aprint_error_dev(self, "no spaces enabled!\n");
    162 		goto fail;
    163 	}
    164 
    165 #define	PRINT(str)							\
    166 do {									\
    167 	aprint_verbose("%s%s", sep, str);				\
    168 	sep = ", ";							\
    169 } while (/*CONSTCOND*/0)
    170 
    171 	aprint_verbose_dev(self, "");
    172 
    173 	if (io_enabled)
    174 		PRINT("i/o space");
    175 	if (mem_enabled)
    176 		PRINT("memory space");
    177 	aprint_verbose(" enabled");
    178 
    179 	if (mrl_enabled || mrm_enabled || mwi_enabled) {
    180 		if (mrl_enabled)
    181 			PRINT("rd/line");
    182 		if (mrm_enabled)
    183 			PRINT("rd/mult");
    184 		if (mwi_enabled)
    185 			PRINT("wr/inv");
    186 		aprint_verbose(" ok");
    187 	}
    188 
    189 	aprint_verbose("\n");
    190 
    191 #undef PRINT
    192 
    193 	sc->sc_iot = pba->pba_iot;
    194 	sc->sc_memt = pba->pba_memt;
    195 	sc->sc_dmat = pba->pba_dmat;
    196 	sc->sc_dmat64 = pba->pba_dmat64;
    197 	sc->sc_pc = pba->pba_pc;
    198 	sc->sc_bus = pba->pba_bus;
    199 	sc->sc_bridgetag = pba->pba_bridgetag;
    200 	sc->sc_maxndevs = pci_bus_maxdevs(pba->pba_pc, pba->pba_bus);
    201 	sc->sc_intrswiz = pba->pba_intrswiz;
    202 	sc->sc_intrtag = pba->pba_intrtag;
    203 	sc->sc_flags = pba->pba_flags;
    204 
    205 	device_pmf_driver_set_child_register(sc->sc_dev, pci_child_register);
    206 
    207 	pcirescan(sc->sc_dev, "pci", wildcard);
    208 
    209 fail:
    210 	if (!pmf_device_register(self, NULL, NULL))
    211 		aprint_error_dev(self, "couldn't establish power handler\n");
    212 }
    213 
    214 int
    215 pcidetach(device_t self, int flags)
    216 {
    217 	int rc;
    218 
    219 	if ((rc = config_detach_children(self, flags)) != 0)
    220 		return rc;
    221 	pmf_device_deregister(self);
    222 	return 0;
    223 }
    224 
    225 int
    226 pciprint(void *aux, const char *pnp)
    227 {
    228 	struct pci_attach_args *pa = aux;
    229 	char devinfo[256];
    230 	const struct pci_quirkdata *qd;
    231 
    232 	if (pnp) {
    233 		pci_devinfo(pa->pa_id, pa->pa_class, 1, devinfo, sizeof(devinfo));
    234 		aprint_normal("%s at %s", devinfo, pnp);
    235 	}
    236 	aprint_normal(" dev %d function %d", pa->pa_device, pa->pa_function);
    237 	if (pci_config_dump) {
    238 		printf(": ");
    239 		pci_conf_print(pa->pa_pc, pa->pa_tag, NULL);
    240 		if (!pnp)
    241 			pci_devinfo(pa->pa_id, pa->pa_class, 1, devinfo, sizeof(devinfo));
    242 		printf("%s at %s", devinfo, pnp ? pnp : "?");
    243 		printf(" dev %d function %d (", pa->pa_device, pa->pa_function);
    244 #ifdef __i386__
    245 		printf("tag %#lx, intrtag %#lx, intrswiz %#lx, intrpin %#lx",
    246 		    *(long *)&pa->pa_tag, *(long *)&pa->pa_intrtag,
    247 		    (long)pa->pa_intrswiz, (long)pa->pa_intrpin);
    248 #else
    249 		printf("intrswiz %#lx, intrpin %#lx",
    250 		    (long)pa->pa_intrswiz, (long)pa->pa_intrpin);
    251 #endif
    252 		printf(", i/o %s, mem %s,",
    253 		    pa->pa_flags & PCI_FLAGS_IO_OKAY ? "on" : "off",
    254 		    pa->pa_flags & PCI_FLAGS_MEM_OKAY ? "on" : "off");
    255 		qd = pci_lookup_quirkdata(PCI_VENDOR(pa->pa_id),
    256 		    PCI_PRODUCT(pa->pa_id));
    257 		if (qd == NULL) {
    258 			printf(" no quirks");
    259 		} else {
    260 			snprintb(devinfo, sizeof (devinfo),
    261 			    "\002\001multifn\002singlefn\003skipfunc0"
    262 			    "\004skipfunc1\005skipfunc2\006skipfunc3"
    263 			    "\007skipfunc4\010skipfunc5\011skipfunc6"
    264 			    "\012skipfunc7", qd->quirks);
    265 			printf(" quirks %s", devinfo);
    266 		}
    267 		printf(")");
    268 	}
    269 	return UNCONF;
    270 }
    271 
    272 int
    273 pci_probe_device(struct pci_softc *sc, pcitag_t tag,
    274     int (*match)(const struct pci_attach_args *),
    275     struct pci_attach_args *pap)
    276 {
    277 	pci_chipset_tag_t pc = sc->sc_pc;
    278 	struct pci_attach_args pa;
    279 	pcireg_t id, /* csr, */ pciclass, intr, bhlcr, bar, endbar;
    280 #ifdef __HAVE_PCI_MSI_MSIX
    281 	pcireg_t cap;
    282 	int off;
    283 #endif
    284 	int ret, pin, bus, device, function, i, width;
    285 	int locs[PCICF_NLOCS];
    286 
    287 	pci_decompose_tag(pc, tag, &bus, &device, &function);
    288 
    289 	/* a driver already attached? */
    290 	if (sc->PCI_SC_DEVICESC(device, function).c_dev != NULL && !match)
    291 		return 0;
    292 
    293 	bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    294 	if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    295 		return 0;
    296 
    297 	id = pci_conf_read(pc, tag, PCI_ID_REG);
    298 	/* csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG); */
    299 	pciclass = pci_conf_read(pc, tag, PCI_CLASS_REG);
    300 
    301 	/* Invalid vendor ID value? */
    302 	if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    303 		return 0;
    304 	/* XXX Not invalid, but we've done this ~forever. */
    305 	if (PCI_VENDOR(id) == 0)
    306 		return 0;
    307 
    308 	/* Collect memory range info */
    309 	memset(sc->PCI_SC_DEVICESC(device, function).c_range, 0,
    310 	    sizeof(sc->PCI_SC_DEVICESC(device, function).c_range));
    311 	i = 0;
    312 	switch (PCI_HDRTYPE_TYPE(bhlcr)) {
    313 	case PCI_HDRTYPE_PPB:
    314 		endbar = PCI_MAPREG_PPB_END;
    315 		break;
    316 	case PCI_HDRTYPE_PCB:
    317 		endbar = PCI_MAPREG_PCB_END;
    318 		break;
    319 	default:
    320 		endbar = PCI_MAPREG_END;
    321 		break;
    322 	}
    323 	for (bar = PCI_MAPREG_START; bar < endbar; bar += width) {
    324 		struct pci_range *r;
    325 		pcireg_t type;
    326 
    327 		width = 4;
    328 		if (pci_mapreg_probe(pc, tag, bar, &type) == 0)
    329 			continue;
    330 
    331 		if (PCI_MAPREG_TYPE(type) == PCI_MAPREG_TYPE_MEM) {
    332 			if (PCI_MAPREG_MEM_TYPE(type) ==
    333 			    PCI_MAPREG_MEM_TYPE_64BIT)
    334 				width = 8;
    335 
    336 			r = &sc->PCI_SC_DEVICESC(device, function).c_range[i++];
    337 			if (pci_mapreg_info(pc, tag, bar, type,
    338 			    &r->r_offset, &r->r_size, &r->r_flags) != 0)
    339 				break;
    340 			if ((PCI_VENDOR(id) == PCI_VENDOR_ATI) && (bar == 0x10)
    341 			    && (r->r_size == 0x1000000)) {
    342 				struct pci_range *nr;
    343 				/*
    344 				 * this has to be a mach64
    345 				 * split things up so each half-aperture can
    346 				 * be mapped PREFETCHABLE except the last page
    347 				 * which may contain registers
    348 				 */
    349 				r->r_size = 0x7ff000;
    350 				r->r_flags = BUS_SPACE_MAP_LINEAR |
    351 					     BUS_SPACE_MAP_PREFETCHABLE;
    352 				nr = &sc->PCI_SC_DEVICESC(device,
    353 				    function).c_range[i++];
    354 				nr->r_offset = r->r_offset + 0x800000;
    355 				nr->r_size = 0x7ff000;
    356 				nr->r_flags = BUS_SPACE_MAP_LINEAR |
    357 					      BUS_SPACE_MAP_PREFETCHABLE;
    358 			}
    359 
    360 		}
    361 	}
    362 
    363 	pa.pa_iot = sc->sc_iot;
    364 	pa.pa_memt = sc->sc_memt;
    365 	pa.pa_dmat = sc->sc_dmat;
    366 	pa.pa_dmat64 = sc->sc_dmat64;
    367 	pa.pa_pc = pc;
    368 	pa.pa_bus = bus;
    369 	pa.pa_device = device;
    370 	pa.pa_function = function;
    371 	pa.pa_tag = tag;
    372 	pa.pa_id = id;
    373 	pa.pa_class = pciclass;
    374 
    375 	/*
    376 	 * Set up memory, I/O enable, and PCI command flags
    377 	 * as appropriate.
    378 	 */
    379 	pa.pa_flags = sc->sc_flags;
    380 
    381 	/*
    382 	 * If the cache line size is not configured, then
    383 	 * clear the MRL/MRM/MWI command-ok flags.
    384 	 */
    385 	if (PCI_CACHELINE(bhlcr) == 0) {
    386 		pa.pa_flags &= ~(PCI_FLAGS_MRL_OKAY|
    387 		    PCI_FLAGS_MRM_OKAY|PCI_FLAGS_MWI_OKAY);
    388 	}
    389 
    390 	if (sc->sc_bridgetag == NULL) {
    391 		pa.pa_intrswiz = 0;
    392 		pa.pa_intrtag = tag;
    393 	} else {
    394 		pa.pa_intrswiz = sc->sc_intrswiz + device;
    395 		pa.pa_intrtag = sc->sc_intrtag;
    396 	}
    397 
    398 	intr = pci_conf_read(pc, tag, PCI_INTERRUPT_REG);
    399 
    400 	pin = PCI_INTERRUPT_PIN(intr);
    401 	pa.pa_rawintrpin = pin;
    402 	if (pin == PCI_INTERRUPT_PIN_NONE) {
    403 		/* no interrupt */
    404 		pa.pa_intrpin = 0;
    405 	} else {
    406 		/*
    407 		 * swizzle it based on the number of busses we're
    408 		 * behind and our device number.
    409 		 */
    410 		pa.pa_intrpin = 	/* XXX */
    411 		    ((pin + pa.pa_intrswiz - 1) % 4) + 1;
    412 	}
    413 	pa.pa_intrline = PCI_INTERRUPT_LINE(intr);
    414 
    415 #ifdef __HAVE_PCI_MSI_MSIX
    416 	if (pci_get_ht_capability(pc, tag, PCI_HT_CAP_MSIMAP, &off, &cap)) {
    417 		/*
    418 		 * XXX Should we enable MSI mapping ourselves on
    419 		 * systems that have it disabled?
    420 		 */
    421 		if (cap & PCI_HT_MSI_ENABLED) {
    422 			uint64_t addr;
    423 			if ((cap & PCI_HT_MSI_FIXED) == 0) {
    424 				addr = pci_conf_read(pc, tag,
    425 				    off + PCI_HT_MSI_ADDR_LO);
    426 				addr |= (uint64_t)pci_conf_read(pc, tag,
    427 				    off + PCI_HT_MSI_ADDR_HI) << 32;
    428 			} else
    429 				addr = PCI_HT_MSI_FIXED_ADDR;
    430 
    431 			/*
    432 			 * XXX This will fail to enable MSI on systems
    433 			 * that don't use the canonical address.
    434 			 */
    435 			if (addr == PCI_HT_MSI_FIXED_ADDR) {
    436 				pa.pa_flags |= PCI_FLAGS_MSI_OKAY;
    437 				pa.pa_flags |= PCI_FLAGS_MSIX_OKAY;
    438 			}
    439 		}
    440 	}
    441 #endif
    442 
    443 	if (match != NULL) {
    444 		ret = (*match)(&pa);
    445 		if (ret != 0 && pap != NULL)
    446 			*pap = pa;
    447 	} else {
    448 		struct pci_child *c;
    449 		locs[PCICF_DEV] = device;
    450 		locs[PCICF_FUNCTION] = function;
    451 
    452 		c = &sc->PCI_SC_DEVICESC(device, function);
    453 		pci_conf_capture(pc, tag, &c->c_conf);
    454 		if (pci_get_powerstate(pc, tag, &c->c_powerstate) == 0)
    455 			c->c_psok = true;
    456 		else
    457 			c->c_psok = false;
    458 
    459 		c->c_dev = config_found_sm_loc(sc->sc_dev, "pci", locs, &pa,
    460 					     pciprint, config_stdsubmatch);
    461 
    462 		ret = (c->c_dev != NULL);
    463 	}
    464 
    465 	return ret;
    466 }
    467 
    468 void
    469 pcidevdetached(device_t self, device_t child)
    470 {
    471 	struct pci_softc *sc = device_private(self);
    472 	int d, f;
    473 	pcitag_t tag;
    474 	struct pci_child *c;
    475 
    476 	d = device_locator(child, PCICF_DEV);
    477 	f = device_locator(child, PCICF_FUNCTION);
    478 
    479 	c = &sc->PCI_SC_DEVICESC(d, f);
    480 
    481 	KASSERT(c->c_dev == child);
    482 
    483 	tag = pci_make_tag(sc->sc_pc, sc->sc_bus, d, f);
    484 	if (c->c_psok)
    485 		pci_set_powerstate(sc->sc_pc, tag, c->c_powerstate);
    486 	pci_conf_restore(sc->sc_pc, tag, &c->c_conf);
    487 	c->c_dev = NULL;
    488 }
    489 
    490 CFATTACH_DECL3_NEW(pci, sizeof(struct pci_softc),
    491     pcimatch, pciattach, pcidetach, NULL, pcirescan, pcidevdetached,
    492     DVF_DETACH_SHUTDOWN);
    493 
    494 int
    495 pci_get_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    496     int *offset, pcireg_t *value)
    497 {
    498 	pcireg_t reg;
    499 	unsigned int ofs;
    500 
    501 	reg = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
    502 	if (!(reg & PCI_STATUS_CAPLIST_SUPPORT))
    503 		return 0;
    504 
    505 	/* Determine the Capability List Pointer register to start with. */
    506 	reg = pci_conf_read(pc, tag, PCI_BHLC_REG);
    507 	switch (PCI_HDRTYPE_TYPE(reg)) {
    508 	case 0:	/* standard device header */
    509 	case 1: /* PCI-PCI bridge header */
    510 		ofs = PCI_CAPLISTPTR_REG;
    511 		break;
    512 	case 2:	/* PCI-CardBus Bridge header */
    513 		ofs = PCI_CARDBUS_CAPLISTPTR_REG;
    514 		break;
    515 	default:
    516 		return 0;
    517 	}
    518 
    519 	ofs = PCI_CAPLIST_PTR(pci_conf_read(pc, tag, ofs));
    520 	while (ofs != 0) {
    521 		if ((ofs & 3) || (ofs < 0x40)) {
    522 			int bus, device, function;
    523 
    524 			pci_decompose_tag(pc, tag, &bus, &device, &function);
    525 
    526 			printf("Skipping broken PCI header on %d:%d:%d\n",
    527 			    bus, device, function);
    528 			break;
    529 		}
    530 		reg = pci_conf_read(pc, tag, ofs);
    531 		if (PCI_CAPLIST_CAP(reg) == capid) {
    532 			if (offset)
    533 				*offset = ofs;
    534 			if (value)
    535 				*value = reg;
    536 			return 1;
    537 		}
    538 		ofs = PCI_CAPLIST_NEXT(reg);
    539 	}
    540 
    541 	return 0;
    542 }
    543 
    544 int
    545 pci_get_ht_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    546     int *offset, pcireg_t *value)
    547 {
    548 	pcireg_t reg;
    549 	unsigned int ofs;
    550 
    551 	if (pci_get_capability(pc, tag, PCI_CAP_LDT, &ofs, NULL) == 0)
    552 		return 0;
    553 
    554 	while (ofs != 0) {
    555 #ifdef DIAGNOSTIC
    556 		if ((ofs & 3) || (ofs < 0x40))
    557 			panic("pci_get_ht_capability");
    558 #endif
    559 		reg = pci_conf_read(pc, tag, ofs);
    560 		if (PCI_HT_CAP(reg) == capid) {
    561 			if (offset)
    562 				*offset = ofs;
    563 			if (value)
    564 				*value = reg;
    565 			return 1;
    566 		}
    567 		ofs = PCI_CAPLIST_NEXT(reg);
    568 	}
    569 
    570 	return 0;
    571 }
    572 
    573 /*
    574  * return number of the devices's MSI vectors
    575  * return 0 if the device does not support MSI
    576  */
    577 int
    578 pci_msi_count(pci_chipset_tag_t pc, pcitag_t tag)
    579 {
    580 	pcireg_t reg;
    581 	uint32_t mmc;
    582 	int count, offset;
    583 
    584 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &offset, NULL) == 0)
    585 		return 0;
    586 
    587 	reg = pci_conf_read(pc, tag, offset + PCI_MSI_CTL);
    588 	mmc = PCI_MSI_CTL_MMC(reg);
    589 	count = 1 << mmc;
    590 	if (count > PCI_MSI_MAX_VECTORS) {
    591 		aprint_error("detect an illegal device! The device use reserved MMC values.\n");
    592 		return 0;
    593 	}
    594 
    595 	return count;
    596 }
    597 
    598 /*
    599  * return number of the devices's MSI-X vectors
    600  * return 0 if the device does not support MSI-X
    601  */
    602 int
    603 pci_msix_count(pci_chipset_tag_t pc, pcitag_t tag)
    604 {
    605 	pcireg_t reg;
    606 	int offset;
    607 
    608 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &offset, NULL) == 0)
    609 		return 0;
    610 
    611 	reg = pci_conf_read(pc, tag, offset + PCI_MSIX_CTL);
    612 
    613 	return PCI_MSIX_CTL_TBLSIZE(reg);
    614 }
    615 
    616 int
    617 pci_find_device(struct pci_attach_args *pa,
    618 		int (*match)(const struct pci_attach_args *))
    619 {
    620 	extern struct cfdriver pci_cd;
    621 	device_t pcidev;
    622 	int i;
    623 	static const int wildcard[2] = {
    624 		PCICF_DEV_DEFAULT,
    625 		PCICF_FUNCTION_DEFAULT
    626 	};
    627 
    628 	for (i = 0; i < pci_cd.cd_ndevs; i++) {
    629 		pcidev = device_lookup(&pci_cd, i);
    630 		if (pcidev != NULL &&
    631 		    pci_enumerate_bus(device_private(pcidev), wildcard,
    632 		    		      match, pa) != 0)
    633 			return 1;
    634 	}
    635 	return 0;
    636 }
    637 
    638 #ifndef PCI_MACHDEP_ENUMERATE_BUS
    639 /*
    640  * Generic PCI bus enumeration routine.  Used unless machine-dependent
    641  * code needs to provide something else.
    642  */
    643 int
    644 pci_enumerate_bus(struct pci_softc *sc, const int *locators,
    645     int (*match)(const struct pci_attach_args *), struct pci_attach_args *pap)
    646 {
    647 	pci_chipset_tag_t pc = sc->sc_pc;
    648 	int device, function, nfunctions, ret;
    649 	const struct pci_quirkdata *qd;
    650 	pcireg_t id, bhlcr;
    651 	pcitag_t tag;
    652 	uint8_t devs[32];
    653 	int i, n;
    654 
    655 	n = pci_bus_devorder(sc->sc_pc, sc->sc_bus, devs, __arraycount(devs));
    656 	for (i = 0; i < n; i++) {
    657 		device = devs[i];
    658 
    659 		if ((locators[PCICF_DEV] != PCICF_DEV_DEFAULT) &&
    660 		    (locators[PCICF_DEV] != device))
    661 			continue;
    662 
    663 		tag = pci_make_tag(pc, sc->sc_bus, device, 0);
    664 
    665 		bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    666 		if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    667 			continue;
    668 
    669 		id = pci_conf_read(pc, tag, PCI_ID_REG);
    670 
    671 		/* Invalid vendor ID value? */
    672 		if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    673 			continue;
    674 		/* XXX Not invalid, but we've done this ~forever. */
    675 		if (PCI_VENDOR(id) == 0)
    676 			continue;
    677 
    678 		qd = pci_lookup_quirkdata(PCI_VENDOR(id), PCI_PRODUCT(id));
    679 
    680 		if (qd != NULL &&
    681 		      (qd->quirks & PCI_QUIRK_MULTIFUNCTION) != 0)
    682 			nfunctions = 8;
    683 		else if (qd != NULL &&
    684 		      (qd->quirks & PCI_QUIRK_MONOFUNCTION) != 0)
    685 			nfunctions = 1;
    686 		else
    687 			nfunctions = PCI_HDRTYPE_MULTIFN(bhlcr) ? 8 : 1;
    688 
    689 #ifdef __PCI_DEV_FUNCORDER
    690 		char funcs[8];
    691 		int j;
    692 		for (j = 0; j < nfunctions; j++) {
    693 			funcs[j] = j;
    694 		}
    695 		if (j < __arraycount(funcs))
    696 			funcs[j] = -1;
    697 		if (nfunctions > 1) {
    698 			pci_dev_funcorder(sc->sc_pc, sc->sc_bus, device,
    699 			    nfunctions, funcs);
    700 		}
    701 		for (j = 0;
    702 		     j < 8 && (function = funcs[j]) < 8 && function >= 0;
    703 		     j++) {
    704 #else
    705 		for (function = 0; function < nfunctions; function++) {
    706 #endif
    707 			if ((locators[PCICF_FUNCTION] != PCICF_FUNCTION_DEFAULT)
    708 			    && (locators[PCICF_FUNCTION] != function))
    709 				continue;
    710 
    711 			if (qd != NULL &&
    712 			    (qd->quirks & PCI_QUIRK_SKIP_FUNC(function)) != 0)
    713 				continue;
    714 			tag = pci_make_tag(pc, sc->sc_bus, device, function);
    715 			ret = pci_probe_device(sc, tag, match, pap);
    716 			if (match != NULL && ret != 0)
    717 				return ret;
    718 		}
    719 	}
    720 	return 0;
    721 }
    722 #endif /* PCI_MACHDEP_ENUMERATE_BUS */
    723 
    724 
    725 /*
    726  * Vital Product Data (PCI 2.2)
    727  */
    728 
    729 int
    730 pci_vpd_read(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    731     pcireg_t *data)
    732 {
    733 	uint32_t reg;
    734 	int ofs, i, j;
    735 
    736 	KASSERT(data != NULL);
    737 	KASSERT((offset + count) < 0x7fff);
    738 
    739 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    740 		return 1;
    741 
    742 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    743 		reg &= 0x0000ffff;
    744 		reg &= ~PCI_VPD_OPFLAG;
    745 		reg |= PCI_VPD_ADDRESS(offset);
    746 		pci_conf_write(pc, tag, ofs, reg);
    747 
    748 		/*
    749 		 * PCI 2.2 does not specify how long we should poll
    750 		 * for completion nor whether the operation can fail.
    751 		 */
    752 		j = 0;
    753 		do {
    754 			if (j++ == 20)
    755 				return 1;
    756 			delay(4);
    757 			reg = pci_conf_read(pc, tag, ofs);
    758 		} while ((reg & PCI_VPD_OPFLAG) == 0);
    759 		data[i] = pci_conf_read(pc, tag, PCI_VPD_DATAREG(ofs));
    760 	}
    761 
    762 	return 0;
    763 }
    764 
    765 int
    766 pci_vpd_write(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    767     pcireg_t *data)
    768 {
    769 	pcireg_t reg;
    770 	int ofs, i, j;
    771 
    772 	KASSERT(data != NULL);
    773 	KASSERT((offset + count) < 0x7fff);
    774 
    775 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    776 		return 1;
    777 
    778 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    779 		pci_conf_write(pc, tag, PCI_VPD_DATAREG(ofs), data[i]);
    780 
    781 		reg &= 0x0000ffff;
    782 		reg |= PCI_VPD_OPFLAG;
    783 		reg |= PCI_VPD_ADDRESS(offset);
    784 		pci_conf_write(pc, tag, ofs, reg);
    785 
    786 		/*
    787 		 * PCI 2.2 does not specify how long we should poll
    788 		 * for completion nor whether the operation can fail.
    789 		 */
    790 		j = 0;
    791 		do {
    792 			if (j++ == 20)
    793 				return 1;
    794 			delay(1);
    795 			reg = pci_conf_read(pc, tag, ofs);
    796 		} while (reg & PCI_VPD_OPFLAG);
    797 	}
    798 
    799 	return 0;
    800 }
    801 
    802 int
    803 pci_dma64_available(const struct pci_attach_args *pa)
    804 {
    805 #ifdef _PCI_HAVE_DMA64
    806 	if (BUS_DMA_TAG_VALID(pa->pa_dmat64))
    807                         return 1;
    808 #endif
    809         return 0;
    810 }
    811 
    812 void
    813 pci_conf_capture(pci_chipset_tag_t pc, pcitag_t tag,
    814 		  struct pci_conf_state *pcs)
    815 {
    816 	int off;
    817 
    818 	for (off = 0; off < 16; off++)
    819 		pcs->reg[off] = pci_conf_read(pc, tag, (off * 4));
    820 
    821 	return;
    822 }
    823 
    824 void
    825 pci_conf_restore(pci_chipset_tag_t pc, pcitag_t tag,
    826 		  struct pci_conf_state *pcs)
    827 {
    828 	int off;
    829 	pcireg_t val;
    830 
    831 	for (off = 15; off >= 0; off--) {
    832 		val = pci_conf_read(pc, tag, (off * 4));
    833 		if (val != pcs->reg[off])
    834 			pci_conf_write(pc, tag, (off * 4), pcs->reg[off]);
    835 	}
    836 
    837 	return;
    838 }
    839 
    840 /*
    841  * Power Management Capability (Rev 2.2)
    842  */
    843 static int
    844 pci_get_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state,
    845     int offset)
    846 {
    847 	pcireg_t value, now;
    848 
    849 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
    850 	now = value & PCI_PMCSR_STATE_MASK;
    851 	switch (now) {
    852 	case PCI_PMCSR_STATE_D0:
    853 	case PCI_PMCSR_STATE_D1:
    854 	case PCI_PMCSR_STATE_D2:
    855 	case PCI_PMCSR_STATE_D3:
    856 		*state = now;
    857 		return 0;
    858 	default:
    859 		return EINVAL;
    860 	}
    861 }
    862 
    863 int
    864 pci_get_powerstate(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state)
    865 {
    866 	int offset;
    867 	pcireg_t value;
    868 
    869 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value))
    870 		return EOPNOTSUPP;
    871 
    872 	return pci_get_powerstate_int(pc, tag, state, offset);
    873 }
    874 
    875 static int
    876 pci_set_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state,
    877     int offset, pcireg_t cap_reg)
    878 {
    879 	pcireg_t value, cap, now;
    880 
    881 	cap = cap_reg >> PCI_PMCR_SHIFT;
    882 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
    883 	now = value & PCI_PMCSR_STATE_MASK;
    884 	value &= ~PCI_PMCSR_STATE_MASK;
    885 
    886 	if (now == state)
    887 		return 0;
    888 	switch (state) {
    889 	case PCI_PMCSR_STATE_D0:
    890 		break;
    891 	case PCI_PMCSR_STATE_D1:
    892 		if (now == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D3) {
    893 			printf("invalid transition from %d to D1\n", (int)now);
    894 			return EINVAL;
    895 		}
    896 		if (!(cap & PCI_PMCR_D1SUPP)) {
    897 			printf("D1 not supported\n");
    898 			return EOPNOTSUPP;
    899 		}
    900 		break;
    901 	case PCI_PMCSR_STATE_D2:
    902 		if (now == PCI_PMCSR_STATE_D3) {
    903 			printf("invalid transition from %d to D2\n", (int)now);
    904 			return EINVAL;
    905 		}
    906 		if (!(cap & PCI_PMCR_D2SUPP)) {
    907 			printf("D2 not supported\n");
    908 			return EOPNOTSUPP;
    909 		}
    910 		break;
    911 	case PCI_PMCSR_STATE_D3:
    912 		break;
    913 	default:
    914 		return EINVAL;
    915 	}
    916 	value |= state;
    917 	pci_conf_write(pc, tag, offset + PCI_PMCSR, value);
    918 	/* delay according to pcipm1.2, ch. 5.6.1 */
    919 	if (state == PCI_PMCSR_STATE_D3 || now == PCI_PMCSR_STATE_D3)
    920 		DELAY(10000);
    921 	else if (state == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D2)
    922 		DELAY(200);
    923 
    924 	return 0;
    925 }
    926 
    927 int
    928 pci_set_powerstate(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state)
    929 {
    930 	int offset;
    931 	pcireg_t value;
    932 
    933 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value)) {
    934 		printf("pci_set_powerstate not supported\n");
    935 		return EOPNOTSUPP;
    936 	}
    937 
    938 	return pci_set_powerstate_int(pc, tag, state, offset, value);
    939 }
    940 
    941 int
    942 pci_activate(pci_chipset_tag_t pc, pcitag_t tag, device_t dev,
    943     int (*wakefun)(pci_chipset_tag_t, pcitag_t, device_t, pcireg_t))
    944 {
    945 	pcireg_t pmode;
    946 	int error;
    947 
    948 	if ((error = pci_get_powerstate(pc, tag, &pmode)))
    949 		return error;
    950 
    951 	switch (pmode) {
    952 	case PCI_PMCSR_STATE_D0:
    953 		break;
    954 	case PCI_PMCSR_STATE_D3:
    955 		if (wakefun == NULL) {
    956 			/*
    957 			 * The card has lost all configuration data in
    958 			 * this state, so punt.
    959 			 */
    960 			aprint_error_dev(dev,
    961 			    "unable to wake up from power state D3\n");
    962 			return EOPNOTSUPP;
    963 		}
    964 		/*FALLTHROUGH*/
    965 	default:
    966 		if (wakefun) {
    967 			error = (*wakefun)(pc, tag, dev, pmode);
    968 			if (error)
    969 				return error;
    970 		}
    971 		aprint_normal_dev(dev, "waking up from power state D%d\n",
    972 		    pmode);
    973 		if ((error = pci_set_powerstate(pc, tag, PCI_PMCSR_STATE_D0)))
    974 			return error;
    975 	}
    976 	return 0;
    977 }
    978 
    979 int
    980 pci_activate_null(pci_chipset_tag_t pc, pcitag_t tag,
    981     device_t dev, pcireg_t state)
    982 {
    983 	return 0;
    984 }
    985 
    986 struct pci_child_power {
    987 	struct pci_conf_state p_pciconf;
    988 	pci_chipset_tag_t p_pc;
    989 	pcitag_t p_tag;
    990 	bool p_has_pm;
    991 	int p_pm_offset;
    992 	pcireg_t p_pm_cap;
    993 	pcireg_t p_class;
    994 	pcireg_t p_csr;
    995 };
    996 
    997 static bool
    998 pci_child_suspend(device_t dv, const pmf_qual_t *qual)
    999 {
   1000 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1001 	pcireg_t ocsr, csr;
   1002 
   1003 	pci_conf_capture(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1004 
   1005 	if (!priv->p_has_pm)
   1006 		return true; /* ??? hopefully handled by ACPI */
   1007 	if (PCI_CLASS(priv->p_class) == PCI_CLASS_DISPLAY)
   1008 		return true; /* XXX */
   1009 
   1010 	/* disable decoding and busmastering, see pcipm1.2 ch. 8.2.1 */
   1011 	ocsr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1012 	csr = ocsr & ~(PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE
   1013 		       | PCI_COMMAND_MASTER_ENABLE);
   1014 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1015 	if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1016 	    PCI_PMCSR_STATE_D3, priv->p_pm_offset, priv->p_pm_cap)) {
   1017 		pci_conf_write(priv->p_pc, priv->p_tag,
   1018 			       PCI_COMMAND_STATUS_REG, ocsr);
   1019 		aprint_error_dev(dv, "unsupported state, continuing.\n");
   1020 		return false;
   1021 	}
   1022 	return true;
   1023 }
   1024 
   1025 static bool
   1026 pci_child_resume(device_t dv, const pmf_qual_t *qual)
   1027 {
   1028 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1029 
   1030 	if (priv->p_has_pm &&
   1031 	    pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1032 	    PCI_PMCSR_STATE_D0, priv->p_pm_offset, priv->p_pm_cap)) {
   1033 		aprint_error_dev(dv, "unsupported state, continuing.\n");
   1034 		return false;
   1035 	}
   1036 
   1037 	pci_conf_restore(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1038 
   1039 	return true;
   1040 }
   1041 
   1042 static bool
   1043 pci_child_shutdown(device_t dv, int how)
   1044 {
   1045 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1046 	pcireg_t csr;
   1047 
   1048 	/* restore original bus-mastering state */
   1049 	csr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1050 	csr &= ~PCI_COMMAND_MASTER_ENABLE;
   1051 	csr |= priv->p_csr & PCI_COMMAND_MASTER_ENABLE;
   1052 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1053 	return true;
   1054 }
   1055 
   1056 static void
   1057 pci_child_deregister(device_t dv)
   1058 {
   1059 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1060 
   1061 	free(priv, M_DEVBUF);
   1062 }
   1063 
   1064 static bool
   1065 pci_child_register(device_t child)
   1066 {
   1067 	device_t self = device_parent(child);
   1068 	struct pci_softc *sc = device_private(self);
   1069 	struct pci_child_power *priv;
   1070 	int device, function, off;
   1071 	pcireg_t reg;
   1072 
   1073 	priv = malloc(sizeof(*priv), M_DEVBUF, M_WAITOK);
   1074 
   1075 	device = device_locator(child, PCICF_DEV);
   1076 	function = device_locator(child, PCICF_FUNCTION);
   1077 
   1078 	priv->p_pc = sc->sc_pc;
   1079 	priv->p_tag = pci_make_tag(priv->p_pc, sc->sc_bus, device,
   1080 	    function);
   1081 	priv->p_class = pci_conf_read(priv->p_pc, priv->p_tag, PCI_CLASS_REG);
   1082 	priv->p_csr = pci_conf_read(priv->p_pc, priv->p_tag,
   1083 	    PCI_COMMAND_STATUS_REG);
   1084 
   1085 	if (pci_get_capability(priv->p_pc, priv->p_tag,
   1086 			       PCI_CAP_PWRMGMT, &off, &reg)) {
   1087 		priv->p_has_pm = true;
   1088 		priv->p_pm_offset = off;
   1089 		priv->p_pm_cap = reg;
   1090 	} else {
   1091 		priv->p_has_pm = false;
   1092 		priv->p_pm_offset = -1;
   1093 	}
   1094 
   1095 	device_pmf_bus_register(child, priv, pci_child_suspend,
   1096 	    pci_child_resume, pci_child_shutdown, pci_child_deregister);
   1097 
   1098 	return true;
   1099 }
   1100 
   1101 MODULE(MODULE_CLASS_DRIVER, pci, NULL);
   1102 
   1103 static int
   1104 pci_modcmd(modcmd_t cmd, void *priv)
   1105 {
   1106 	if (cmd == MODULE_CMD_INIT || cmd == MODULE_CMD_FINI)
   1107 		return 0;
   1108 	return ENOTTY;
   1109 }
   1110