Home | History | Annotate | Line # | Download | only in pci
pci.c revision 1.162
      1 /*	$NetBSD: pci.c,v 1.162 2021/09/15 17:33:08 thorpej 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 1.162 2021/09/15 17:33:08 thorpej 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 #include <dev/pci/ppbvar.h>
     55 
     56 #include <dev/pci/pci_calls.h>
     57 
     58 #include <net/if.h>
     59 
     60 #include "locators.h"
     61 
     62 static bool pci_child_register(device_t);
     63 
     64 #ifdef PCI_CONFIG_DUMP
     65 int pci_config_dump = 1;
     66 #else
     67 int pci_config_dump = 0;
     68 #endif
     69 
     70 int	pciprint(void *, const char *);
     71 
     72 #ifdef PCI_MACHDEP_ENUMERATE_BUS
     73 #define pci_enumerate_bus PCI_MACHDEP_ENUMERATE_BUS
     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 static devhandle_t
    273 pci_bus_get_child_devhandle(struct pci_softc *sc, pcitag_t tag)
    274 {
    275 	struct pci_bus_get_child_devhandle_args args = {
    276 		.pc = sc->sc_pc,
    277 		.tag = tag,
    278 	};
    279 
    280 	if (device_call(sc->sc_dev, PCI_BUS_GET_CHILD_DEVHANDLE(&args)) != 0) {
    281 		/*
    282 		 * The call is either not supported or the requested
    283 		 * device was not found in the platform device tree.
    284 		 * Return an invalid handle.
    285 		 */
    286 		devhandle_invalidate(&args.devhandle);
    287 	}
    288 
    289 	return args.devhandle;
    290 }
    291 
    292 int
    293 pci_probe_device(struct pci_softc *sc, pcitag_t tag,
    294     int (*match)(const struct pci_attach_args *),
    295     struct pci_attach_args *pap)
    296 {
    297 	pci_chipset_tag_t pc = sc->sc_pc;
    298 	struct pci_attach_args pa;
    299 	pcireg_t id, /* csr, */ pciclass, intr, bhlcr, bar, endbar;
    300 #ifdef __HAVE_PCI_MSI_MSIX
    301 	pcireg_t cap;
    302 	int off;
    303 #endif
    304 	int ret, pin, bus, device, function, i, width;
    305 	int locs[PCICF_NLOCS];
    306 
    307 	pci_decompose_tag(pc, tag, &bus, &device, &function);
    308 
    309 	/* a driver already attached? */
    310 	if (sc->PCI_SC_DEVICESC(device, function).c_dev != NULL && !match)
    311 		return 0;
    312 
    313 	id = pci_conf_read(pc, tag, PCI_ID_REG);
    314 
    315 	/* Invalid vendor ID value? */
    316 	if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    317 		return 0;
    318 	/* XXX Not invalid, but we've done this ~forever. */
    319 	if (PCI_VENDOR(id) == 0)
    320 		return 0;
    321 
    322 	bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    323 	if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    324 		return 0;
    325 
    326 	/* csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG); */
    327 	pciclass = pci_conf_read(pc, tag, PCI_CLASS_REG);
    328 
    329 	/* Collect memory range info */
    330 	memset(sc->PCI_SC_DEVICESC(device, function).c_range, 0,
    331 	    sizeof(sc->PCI_SC_DEVICESC(device, function).c_range));
    332 	i = 0;
    333 	switch (PCI_HDRTYPE_TYPE(bhlcr)) {
    334 	case PCI_HDRTYPE_PPB:
    335 		endbar = PCI_MAPREG_PPB_END;
    336 		break;
    337 	case PCI_HDRTYPE_PCB:
    338 		endbar = PCI_MAPREG_PCB_END;
    339 		break;
    340 	default:
    341 		endbar = PCI_MAPREG_END;
    342 		break;
    343 	}
    344 	for (bar = PCI_MAPREG_START; bar < endbar; bar += width) {
    345 		struct pci_range *r;
    346 		pcireg_t type;
    347 
    348 		width = 4;
    349 		if (pci_mapreg_probe(pc, tag, bar, &type) == 0)
    350 			continue;
    351 
    352 		if (PCI_MAPREG_TYPE(type) == PCI_MAPREG_TYPE_MEM) {
    353 			if (PCI_MAPREG_MEM_TYPE(type) ==
    354 			    PCI_MAPREG_MEM_TYPE_64BIT)
    355 				width = 8;
    356 
    357 			r = &sc->PCI_SC_DEVICESC(device, function).c_range[i++];
    358 			if (pci_mapreg_info(pc, tag, bar, type,
    359 			    &r->r_offset, &r->r_size, &r->r_flags) != 0)
    360 				break;
    361 			if ((PCI_VENDOR(id) == PCI_VENDOR_ATI) && (bar == 0x10)
    362 			    && (r->r_size == 0x1000000)) {
    363 				struct pci_range *nr;
    364 				/*
    365 				 * this has to be a mach64
    366 				 * split things up so each half-aperture can
    367 				 * be mapped PREFETCHABLE except the last page
    368 				 * which may contain registers
    369 				 */
    370 				r->r_size = 0x7ff000;
    371 				r->r_flags = BUS_SPACE_MAP_LINEAR |
    372 					     BUS_SPACE_MAP_PREFETCHABLE;
    373 				nr = &sc->PCI_SC_DEVICESC(device,
    374 				    function).c_range[i++];
    375 				nr->r_offset = r->r_offset + 0x800000;
    376 				nr->r_size = 0x7ff000;
    377 				nr->r_flags = BUS_SPACE_MAP_LINEAR |
    378 					      BUS_SPACE_MAP_PREFETCHABLE;
    379 			} else if ((PCI_VENDOR(id) == PCI_VENDOR_SILMOTION) &&
    380 			   (PCI_PRODUCT(id) == PCI_PRODUCT_SILMOTION_SM502) &&
    381 			   (bar == 0x10)) {
    382 			   	r->r_flags = BUS_SPACE_MAP_LINEAR |
    383 					     BUS_SPACE_MAP_PREFETCHABLE;
    384 			}
    385 		}
    386 	}
    387 
    388 	pa.pa_iot = sc->sc_iot;
    389 	pa.pa_memt = sc->sc_memt;
    390 	pa.pa_dmat = sc->sc_dmat;
    391 	pa.pa_dmat64 = sc->sc_dmat64;
    392 	pa.pa_pc = pc;
    393 	pa.pa_bus = bus;
    394 	pa.pa_device = device;
    395 	pa.pa_function = function;
    396 	pa.pa_tag = tag;
    397 	pa.pa_id = id;
    398 	pa.pa_class = pciclass;
    399 
    400 	/*
    401 	 * Set up memory, I/O enable, and PCI command flags
    402 	 * as appropriate.
    403 	 */
    404 	pa.pa_flags = sc->sc_flags;
    405 
    406 	/*
    407 	 * If the cache line size is not configured, then
    408 	 * clear the MRL/MRM/MWI command-ok flags.
    409 	 */
    410 	if (PCI_CACHELINE(bhlcr) == 0) {
    411 		pa.pa_flags &= ~(PCI_FLAGS_MRL_OKAY|
    412 		    PCI_FLAGS_MRM_OKAY|PCI_FLAGS_MWI_OKAY);
    413 	}
    414 
    415 	if (sc->sc_bridgetag == NULL) {
    416 		pa.pa_intrswiz = 0;
    417 		pa.pa_intrtag = tag;
    418 	} else {
    419 		pa.pa_intrswiz = sc->sc_intrswiz + device;
    420 		pa.pa_intrtag = sc->sc_intrtag;
    421 	}
    422 
    423 	intr = pci_conf_read(pc, tag, PCI_INTERRUPT_REG);
    424 
    425 	pin = PCI_INTERRUPT_PIN(intr);
    426 	pa.pa_rawintrpin = pin;
    427 	if (pin == PCI_INTERRUPT_PIN_NONE) {
    428 		/* no interrupt */
    429 		pa.pa_intrpin = 0;
    430 	} else {
    431 		/*
    432 		 * swizzle it based on the number of busses we're
    433 		 * behind and our device number.
    434 		 */
    435 		pa.pa_intrpin = 	/* XXX */
    436 		    ((pin + pa.pa_intrswiz - 1) % 4) + 1;
    437 	}
    438 	pa.pa_intrline = PCI_INTERRUPT_LINE(intr);
    439 
    440 	devhandle_t devhandle = pci_bus_get_child_devhandle(sc, pa.pa_tag);
    441 
    442 #ifdef __HAVE_PCI_MSI_MSIX
    443 	if (pci_get_ht_capability(pc, tag, PCI_HT_CAP_MSIMAP, &off, &cap)) {
    444 		/*
    445 		 * XXX Should we enable MSI mapping ourselves on
    446 		 * systems that have it disabled?
    447 		 */
    448 		if (cap & PCI_HT_MSI_ENABLED) {
    449 			uint64_t addr;
    450 			if ((cap & PCI_HT_MSI_FIXED) == 0) {
    451 				addr = pci_conf_read(pc, tag,
    452 				    off + PCI_HT_MSI_ADDR_LO);
    453 				addr |= (uint64_t)pci_conf_read(pc, tag,
    454 				    off + PCI_HT_MSI_ADDR_HI) << 32;
    455 			} else
    456 				addr = PCI_HT_MSI_FIXED_ADDR;
    457 
    458 			/*
    459 			 * XXX This will fail to enable MSI on systems
    460 			 * that don't use the canonical address.
    461 			 */
    462 			if (addr == PCI_HT_MSI_FIXED_ADDR) {
    463 				pa.pa_flags |= PCI_FLAGS_MSI_OKAY;
    464 				pa.pa_flags |= PCI_FLAGS_MSIX_OKAY;
    465 			} else
    466 				aprint_verbose_dev(sc->sc_dev,
    467 				    "HyperTransport MSI mapping is not supported yet. Disable MSI/MSI-X.\n");
    468 		}
    469 	}
    470 #endif
    471 
    472 	if (match != NULL) {
    473 		ret = (*match)(&pa);
    474 		if (ret != 0 && pap != NULL)
    475 			*pap = pa;
    476 	} else {
    477 		struct pci_child *c;
    478 		locs[PCICF_DEV] = device;
    479 		locs[PCICF_FUNCTION] = function;
    480 
    481 		c = &sc->PCI_SC_DEVICESC(device, function);
    482 		pci_conf_capture(pc, tag, &c->c_conf);
    483 		if (pci_get_powerstate(pc, tag, &c->c_powerstate) == 0)
    484 			c->c_psok = true;
    485 		else
    486 			c->c_psok = false;
    487 
    488 		c->c_dev = config_found(sc->sc_dev, &pa, pciprint,
    489 		    CFARGS(.submatch = config_stdsubmatch,
    490 			   .locators = locs,
    491 			   .devhandle = devhandle));
    492 
    493 		ret = (c->c_dev != NULL);
    494 	}
    495 
    496 	return ret;
    497 }
    498 
    499 void
    500 pcidevdetached(device_t self, device_t child)
    501 {
    502 	struct pci_softc *sc = device_private(self);
    503 	int d, f;
    504 	pcitag_t tag;
    505 	struct pci_child *c;
    506 
    507 	d = device_locator(child, PCICF_DEV);
    508 	f = device_locator(child, PCICF_FUNCTION);
    509 
    510 	c = &sc->PCI_SC_DEVICESC(d, f);
    511 
    512 	KASSERT(c->c_dev == child);
    513 
    514 	tag = pci_make_tag(sc->sc_pc, sc->sc_bus, d, f);
    515 	if (c->c_psok)
    516 		pci_set_powerstate(sc->sc_pc, tag, c->c_powerstate);
    517 	pci_conf_restore(sc->sc_pc, tag, &c->c_conf);
    518 	c->c_dev = NULL;
    519 }
    520 
    521 CFATTACH_DECL3_NEW(pci, sizeof(struct pci_softc),
    522     pcimatch, pciattach, pcidetach, NULL, pcirescan, pcidevdetached,
    523     DVF_DETACH_SHUTDOWN);
    524 
    525 int
    526 pci_get_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    527     int *offset, pcireg_t *value)
    528 {
    529 	pcireg_t reg;
    530 	unsigned int ofs;
    531 
    532 	reg = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
    533 	if (!(reg & PCI_STATUS_CAPLIST_SUPPORT))
    534 		return 0;
    535 
    536 	/* Determine the Capability List Pointer register to start with. */
    537 	reg = pci_conf_read(pc, tag, PCI_BHLC_REG);
    538 	switch (PCI_HDRTYPE_TYPE(reg)) {
    539 	case 0:	/* standard device header */
    540 	case 1: /* PCI-PCI bridge header */
    541 		ofs = PCI_CAPLISTPTR_REG;
    542 		break;
    543 	case 2:	/* PCI-CardBus Bridge header */
    544 		ofs = PCI_CARDBUS_CAPLISTPTR_REG;
    545 		break;
    546 	default:
    547 		return 0;
    548 	}
    549 
    550 	ofs = PCI_CAPLIST_PTR(pci_conf_read(pc, tag, ofs));
    551 	while (ofs != 0) {
    552 		if ((ofs & 3) || (ofs < 0x40)) {
    553 			int bus, device, function;
    554 
    555 			pci_decompose_tag(pc, tag, &bus, &device, &function);
    556 
    557 			printf("Skipping broken PCI header on %d:%d:%d\n",
    558 			    bus, device, function);
    559 			break;
    560 		}
    561 		reg = pci_conf_read(pc, tag, ofs);
    562 		if (PCI_CAPLIST_CAP(reg) == capid) {
    563 			if (offset)
    564 				*offset = ofs;
    565 			if (value)
    566 				*value = reg;
    567 			return 1;
    568 		}
    569 		ofs = PCI_CAPLIST_NEXT(reg);
    570 	}
    571 
    572 	return 0;
    573 }
    574 
    575 int
    576 pci_get_ht_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    577     int *offset, pcireg_t *value)
    578 {
    579 	pcireg_t reg;
    580 	unsigned int ofs;
    581 
    582 	if (pci_get_capability(pc, tag, PCI_CAP_LDT, &ofs, NULL) == 0)
    583 		return 0;
    584 
    585 	while (ofs != 0) {
    586 #ifdef DIAGNOSTIC
    587 		if ((ofs & 3) || (ofs < 0x40))
    588 			panic("pci_get_ht_capability");
    589 #endif
    590 		reg = pci_conf_read(pc, tag, ofs);
    591 		if (PCI_HT_CAP(reg) == capid) {
    592 			if (offset)
    593 				*offset = ofs;
    594 			if (value)
    595 				*value = reg;
    596 			return 1;
    597 		}
    598 		ofs = PCI_CAPLIST_NEXT(reg);
    599 	}
    600 
    601 	return 0;
    602 }
    603 
    604 /*
    605  * return number of the devices's MSI vectors
    606  * return 0 if the device does not support MSI
    607  */
    608 int
    609 pci_msi_count(pci_chipset_tag_t pc, pcitag_t tag)
    610 {
    611 	pcireg_t reg;
    612 	uint32_t mmc;
    613 	int count, offset;
    614 
    615 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &offset, NULL) == 0)
    616 		return 0;
    617 
    618 	reg = pci_conf_read(pc, tag, offset + PCI_MSI_CTL);
    619 	mmc = PCI_MSI_CTL_MMC(reg);
    620 	count = 1 << mmc;
    621 	if (count > PCI_MSI_MAX_VECTORS) {
    622 		aprint_error("detect an illegal device! The device use reserved MMC values.\n");
    623 		return 0;
    624 	}
    625 
    626 	return count;
    627 }
    628 
    629 /*
    630  * return number of the devices's MSI-X vectors
    631  * return 0 if the device does not support MSI-X
    632  */
    633 int
    634 pci_msix_count(pci_chipset_tag_t pc, pcitag_t tag)
    635 {
    636 	pcireg_t reg;
    637 	int offset;
    638 
    639 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &offset, NULL) == 0)
    640 		return 0;
    641 
    642 	reg = pci_conf_read(pc, tag, offset + PCI_MSIX_CTL);
    643 
    644 	return PCI_MSIX_CTL_TBLSIZE(reg);
    645 }
    646 
    647 int
    648 pci_get_ext_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    649     int *offset, pcireg_t *value)
    650 {
    651 	pcireg_t reg;
    652 	unsigned int ofs;
    653 
    654 	/* Only supported for PCI-express devices */
    655 	if (!pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, NULL, NULL))
    656 		return 0;
    657 
    658 	ofs = PCI_EXTCAPLIST_BASE;
    659 	reg = pci_conf_read(pc, tag, ofs);
    660 	if (reg == 0xffffffff || reg == 0)
    661 		return 0;
    662 
    663 	for (;;) {
    664 #ifdef DIAGNOSTIC
    665 		if ((ofs & 3) || ofs < PCI_EXTCAPLIST_BASE)
    666 			panic("%s: invalid offset %u", __func__, ofs);
    667 #endif
    668 		if (PCI_EXTCAPLIST_CAP(reg) == capid) {
    669 			if (offset != NULL)
    670 				*offset = ofs;
    671 			if (value != NULL)
    672 				*value = reg;
    673 			return 1;
    674 		}
    675 		ofs = PCI_EXTCAPLIST_NEXT(reg);
    676 		if (ofs == 0)
    677 			break;
    678 		reg = pci_conf_read(pc, tag, ofs);
    679 	}
    680 
    681 	return 0;
    682 }
    683 
    684 int
    685 pci_find_device(struct pci_attach_args *pa,
    686 		int (*match)(const struct pci_attach_args *))
    687 {
    688 	extern struct cfdriver pci_cd;
    689 	device_t pcidev;
    690 	int i;
    691 	static const int wildcard[2] = {
    692 		PCICF_DEV_DEFAULT,
    693 		PCICF_FUNCTION_DEFAULT
    694 	};
    695 
    696 	for (i = 0; i < pci_cd.cd_ndevs; i++) {
    697 		pcidev = device_lookup(&pci_cd, i);
    698 		if (pcidev != NULL &&
    699 		    pci_enumerate_bus(device_private(pcidev), wildcard,
    700 		    		      match, pa) != 0)
    701 			return 1;
    702 	}
    703 	return 0;
    704 }
    705 
    706 #ifndef PCI_MACHDEP_ENUMERATE_BUS
    707 /*
    708  * Generic PCI bus enumeration routine.  Used unless machine-dependent
    709  * code needs to provide something else.
    710  */
    711 int
    712 pci_enumerate_bus(struct pci_softc *sc, const int *locators,
    713     int (*match)(const struct pci_attach_args *), struct pci_attach_args *pap)
    714 {
    715 	pci_chipset_tag_t pc = sc->sc_pc;
    716 	int device, function, nfunctions, ret;
    717 	const struct pci_quirkdata *qd;
    718 	pcireg_t id, bhlcr;
    719 	pcitag_t tag;
    720 	uint8_t devs[32];
    721 	int i, n;
    722 
    723 	device_t bridgedev;
    724 	bool arien = false;
    725 	bool downstream_port = false;
    726 
    727 	/* Check PCIe ARI and port type */
    728 	bridgedev = device_parent(sc->sc_dev);
    729 	if (device_is_a(bridgedev, "ppb")) {
    730 		struct ppb_softc *ppbsc = device_private(bridgedev);
    731 		pci_chipset_tag_t ppbpc = ppbsc->sc_pc;
    732 		pcitag_t ppbtag = ppbsc->sc_tag;
    733 		pcireg_t pciecap, capreg, reg;
    734 
    735 		if (pci_get_capability(ppbpc, ppbtag, PCI_CAP_PCIEXPRESS,
    736 		    &pciecap, &capreg) != 0) {
    737 			switch (PCIE_XCAP_TYPE(capreg)) {
    738 			case PCIE_XCAP_TYPE_ROOT:
    739 			case PCIE_XCAP_TYPE_DOWN:
    740 			case PCIE_XCAP_TYPE_PCI2PCIE:
    741 				downstream_port = true;
    742 				break;
    743 			}
    744 
    745 			reg = pci_conf_read(ppbpc, ppbtag, pciecap
    746 			    + PCIE_DCSR2);
    747 			if ((reg & PCIE_DCSR2_ARI_FWD) != 0)
    748 				arien = true;
    749 		}
    750 	}
    751 
    752 	n = pci_bus_devorder(sc->sc_pc, sc->sc_bus, devs, __arraycount(devs));
    753 	if (downstream_port) {
    754 		/* PCIe downstream ports only have a single child device */
    755 		n = 1;
    756 	}
    757 
    758 	for (i = 0; i < n; i++) {
    759 		device = devs[i];
    760 
    761 		if ((locators[PCICF_DEV] != PCICF_DEV_DEFAULT) &&
    762 		    (locators[PCICF_DEV] != device))
    763 			continue;
    764 
    765 		tag = pci_make_tag(pc, sc->sc_bus, device, 0);
    766 
    767 		id = pci_conf_read(pc, tag, PCI_ID_REG);
    768 
    769 		/* Invalid vendor ID value? */
    770 		if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    771 			continue;
    772 		/* XXX Not invalid, but we've done this ~forever. */
    773 		if (PCI_VENDOR(id) == 0)
    774 			continue;
    775 
    776 		bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    777 		if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    778 			continue;
    779 
    780 		qd = pci_lookup_quirkdata(PCI_VENDOR(id), PCI_PRODUCT(id));
    781 
    782 		if (qd != NULL &&
    783 		      (qd->quirks & PCI_QUIRK_MULTIFUNCTION) != 0)
    784 			nfunctions = 8;
    785 		else if (qd != NULL &&
    786 		      (qd->quirks & PCI_QUIRK_MONOFUNCTION) != 0)
    787 			nfunctions = 1;
    788 		else if (arien)
    789 			nfunctions = 8; /* Scan all if ARI is enabled */
    790 		else
    791 			nfunctions = PCI_HDRTYPE_MULTIFN(bhlcr) ? 8 : 1;
    792 
    793 #ifdef __PCI_DEV_FUNCORDER
    794 		char funcs[8];
    795 		int j;
    796 		for (j = 0; j < nfunctions; j++) {
    797 			funcs[j] = j;
    798 		}
    799 		if (j < __arraycount(funcs))
    800 			funcs[j] = -1;
    801 		if (nfunctions > 1) {
    802 			pci_dev_funcorder(sc->sc_pc, sc->sc_bus, device,
    803 			    nfunctions, funcs);
    804 		}
    805 		for (j = 0;
    806 		     j < 8 && (function = funcs[j]) < 8 && function >= 0;
    807 		     j++) {
    808 #else
    809 		for (function = 0; function < nfunctions; function++) {
    810 #endif
    811 			if ((locators[PCICF_FUNCTION] != PCICF_FUNCTION_DEFAULT)
    812 			    && (locators[PCICF_FUNCTION] != function))
    813 				continue;
    814 
    815 			if (qd != NULL &&
    816 			    (qd->quirks & PCI_QUIRK_SKIP_FUNC(function)) != 0)
    817 				continue;
    818 			tag = pci_make_tag(pc, sc->sc_bus, device, function);
    819 			ret = pci_probe_device(sc, tag, match, pap);
    820 			if (match != NULL && ret != 0)
    821 				return ret;
    822 		}
    823 	}
    824 	return 0;
    825 }
    826 #endif /* PCI_MACHDEP_ENUMERATE_BUS */
    827 
    828 
    829 /*
    830  * Vital Product Data (PCI 2.2)
    831  */
    832 
    833 int
    834 pci_vpd_read(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    835     pcireg_t *data)
    836 {
    837 	uint32_t reg;
    838 	int ofs, i, j;
    839 
    840 	KASSERT(data != NULL);
    841 	KASSERT((offset + count) < 0x7fff);
    842 
    843 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    844 		return 1;
    845 
    846 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    847 		reg &= 0x0000ffff;
    848 		reg &= ~PCI_VPD_OPFLAG;
    849 		reg |= PCI_VPD_ADDRESS(offset);
    850 		pci_conf_write(pc, tag, ofs, reg);
    851 
    852 		/*
    853 		 * PCI 2.2 does not specify how long we should poll
    854 		 * for completion nor whether the operation can fail.
    855 		 */
    856 		j = 0;
    857 		do {
    858 			if (j++ == 20)
    859 				return 1;
    860 			delay(4);
    861 			reg = pci_conf_read(pc, tag, ofs);
    862 		} while ((reg & PCI_VPD_OPFLAG) == 0);
    863 		data[i] = pci_conf_read(pc, tag, PCI_VPD_DATAREG(ofs));
    864 	}
    865 
    866 	return 0;
    867 }
    868 
    869 int
    870 pci_vpd_write(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    871     pcireg_t *data)
    872 {
    873 	pcireg_t reg;
    874 	int ofs, i, j;
    875 
    876 	KASSERT(data != NULL);
    877 	KASSERT((offset + count) < 0x7fff);
    878 
    879 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    880 		return 1;
    881 
    882 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    883 		pci_conf_write(pc, tag, PCI_VPD_DATAREG(ofs), data[i]);
    884 
    885 		reg &= 0x0000ffff;
    886 		reg |= PCI_VPD_OPFLAG;
    887 		reg |= PCI_VPD_ADDRESS(offset);
    888 		pci_conf_write(pc, tag, ofs, reg);
    889 
    890 		/*
    891 		 * PCI 2.2 does not specify how long we should poll
    892 		 * for completion nor whether the operation can fail.
    893 		 */
    894 		j = 0;
    895 		do {
    896 			if (j++ == 20)
    897 				return 1;
    898 			delay(1);
    899 			reg = pci_conf_read(pc, tag, ofs);
    900 		} while (reg & PCI_VPD_OPFLAG);
    901 	}
    902 
    903 	return 0;
    904 }
    905 
    906 int
    907 pci_dma64_available(const struct pci_attach_args *pa)
    908 {
    909 #ifdef _PCI_HAVE_DMA64
    910 	if (BUS_DMA_TAG_VALID(pa->pa_dmat64))
    911                         return 1;
    912 #endif
    913         return 0;
    914 }
    915 
    916 void
    917 pci_conf_capture(pci_chipset_tag_t pc, pcitag_t tag,
    918 		  struct pci_conf_state *pcs)
    919 {
    920 	int off;
    921 
    922 	for (off = 0; off < 16; off++)
    923 		pcs->reg[off] = pci_conf_read(pc, tag, (off * 4));
    924 
    925 	/* For PCI-X */
    926 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
    927 		pcs->x_csr = pci_conf_read(pc, tag, off + PCIX_CMD);
    928 
    929 	/* For PCIe */
    930 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
    931 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
    932 		unsigned int devtype;
    933 
    934 		devtype = PCIE_XCAP_TYPE(xcap);
    935 		pcs->e_dcr = (uint16_t)pci_conf_read(pc, tag, off + PCIE_DCSR);
    936 
    937 		if (PCIE_HAS_LINKREGS(devtype))
    938 			pcs->e_lcr = (uint16_t)pci_conf_read(pc, tag,
    939 			    off + PCIE_LCSR);
    940 
    941 		if ((xcap & PCIE_XCAP_SI) != 0)
    942 			pcs->e_slcr = (uint16_t)pci_conf_read(pc, tag,
    943 			    off + PCIE_SLCSR);
    944 
    945 		if (PCIE_HAS_ROOTREGS(devtype))
    946 			pcs->e_rcr = (uint16_t)pci_conf_read(pc, tag,
    947 			    off + PCIE_RCR);
    948 
    949 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
    950 			pcs->e_dcr2 = (uint16_t)pci_conf_read(pc, tag,
    951 			    off + PCIE_DCSR2);
    952 
    953 			if (PCIE_HAS_LINKREGS(devtype))
    954 				pcs->e_lcr2 = (uint16_t)pci_conf_read(pc, tag,
    955 			    off + PCIE_LCSR2);
    956 
    957 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
    958 		}
    959 	}
    960 
    961 	/* For MSI */
    962 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
    963 		bool bit64, pvmask;
    964 
    965 		pcs->msi_ctl = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
    966 
    967 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
    968 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
    969 
    970 		/* Address */
    971 		pcs->msi_maddr = pci_conf_read(pc, tag, off + PCI_MSI_MADDR);
    972 		if (bit64)
    973 			pcs->msi_maddr64_hi = pci_conf_read(pc, tag,
    974 			    off + PCI_MSI_MADDR64_HI);
    975 
    976 		/* Data */
    977 		pcs->msi_mdata = pci_conf_read(pc, tag,
    978 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA));
    979 
    980 		/* Per-vector masking */
    981 		if (pvmask)
    982 			pcs->msi_mask = pci_conf_read(pc, tag,
    983 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK));
    984 	}
    985 
    986 	/* For MSI-X */
    987 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
    988 		pcs->msix_ctl = pci_conf_read(pc, tag, off + PCI_MSIX_CTL);
    989 }
    990 
    991 void
    992 pci_conf_restore(pci_chipset_tag_t pc, pcitag_t tag,
    993 		  struct pci_conf_state *pcs)
    994 {
    995 	int off;
    996 	pcireg_t val;
    997 
    998 	for (off = 15; off >= 0; off--) {
    999 		val = pci_conf_read(pc, tag, (off * 4));
   1000 		if (val != pcs->reg[off])
   1001 			pci_conf_write(pc, tag, (off * 4), pcs->reg[off]);
   1002 	}
   1003 
   1004 	/* For PCI-X */
   1005 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
   1006 		pci_conf_write(pc, tag, off + PCIX_CMD, pcs->x_csr);
   1007 
   1008 	/* For PCIe */
   1009 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
   1010 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
   1011 		unsigned int devtype;
   1012 
   1013 		devtype = PCIE_XCAP_TYPE(xcap);
   1014 		pci_conf_write(pc, tag, off + PCIE_DCSR, pcs->e_dcr);
   1015 
   1016 		/*
   1017 		 * PCIe capability is variable sized. To not to write the next
   1018 		 * area, check the existence of each register.
   1019 		 */
   1020 		if (PCIE_HAS_LINKREGS(devtype))
   1021 			pci_conf_write(pc, tag, off + PCIE_LCSR, pcs->e_lcr);
   1022 
   1023 		if ((xcap & PCIE_XCAP_SI) != 0)
   1024 			pci_conf_write(pc, tag, off + PCIE_SLCSR, pcs->e_slcr);
   1025 
   1026 		if (PCIE_HAS_ROOTREGS(devtype))
   1027 			pci_conf_write(pc, tag, off + PCIE_RCR, pcs->e_rcr);
   1028 
   1029 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
   1030 			pci_conf_write(pc, tag, off + PCIE_DCSR2, pcs->e_dcr2);
   1031 
   1032 			if (PCIE_HAS_LINKREGS(devtype))
   1033 				pci_conf_write(pc, tag, off + PCIE_LCSR2,
   1034 				    pcs->e_lcr2);
   1035 
   1036 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
   1037 		}
   1038 	}
   1039 
   1040 	/* For MSI */
   1041 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
   1042 		pcireg_t reg;
   1043 		bool bit64, pvmask;
   1044 
   1045 		/* First, drop Enable bit in case it's already set. */
   1046 		reg = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
   1047 		pci_conf_write(pc, tag, off + PCI_MSI_CTL,
   1048 		    reg & ~PCI_MSI_CTL_MSI_ENABLE);
   1049 
   1050 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
   1051 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
   1052 
   1053 		/* Address */
   1054 		pci_conf_write(pc, tag, off + PCI_MSI_MADDR, pcs->msi_maddr);
   1055 
   1056 		if (bit64)
   1057 			pci_conf_write(pc, tag,
   1058 			    off + PCI_MSI_MADDR64_HI, pcs->msi_maddr64_hi);
   1059 
   1060 		/* Data */
   1061 		pci_conf_write(pc, tag,
   1062 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA),
   1063 		    pcs->msi_mdata);
   1064 
   1065 		/* Per-vector masking */
   1066 		if (pvmask)
   1067 			pci_conf_write(pc, tag,
   1068 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK),
   1069 			    pcs->msi_mask);
   1070 
   1071 		/* Write CTRL register in the end */
   1072 		pci_conf_write(pc, tag, off + PCI_MSI_CTL, pcs->msi_ctl);
   1073 	}
   1074 
   1075 	/* For MSI-X */
   1076 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
   1077 		pci_conf_write(pc, tag, off + PCI_MSIX_CTL, pcs->msix_ctl);
   1078 }
   1079 
   1080 /*
   1081  * Power Management Capability (Rev 2.2)
   1082  */
   1083 static int
   1084 pci_get_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state,
   1085     int offset)
   1086 {
   1087 	pcireg_t value, now;
   1088 
   1089 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1090 	now = value & PCI_PMCSR_STATE_MASK;
   1091 	switch (now) {
   1092 	case PCI_PMCSR_STATE_D0:
   1093 	case PCI_PMCSR_STATE_D1:
   1094 	case PCI_PMCSR_STATE_D2:
   1095 	case PCI_PMCSR_STATE_D3:
   1096 		*state = now;
   1097 		return 0;
   1098 	default:
   1099 		return EINVAL;
   1100 	}
   1101 }
   1102 
   1103 int
   1104 pci_get_powerstate(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state)
   1105 {
   1106 	int offset;
   1107 	pcireg_t value;
   1108 
   1109 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value))
   1110 		return EOPNOTSUPP;
   1111 
   1112 	return pci_get_powerstate_int(pc, tag, state, offset);
   1113 }
   1114 
   1115 static int
   1116 pci_set_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state,
   1117     int offset, pcireg_t cap_reg)
   1118 {
   1119 	pcireg_t value, cap, now;
   1120 
   1121 	cap = cap_reg >> PCI_PMCR_SHIFT;
   1122 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1123 	now = value & PCI_PMCSR_STATE_MASK;
   1124 	value &= ~PCI_PMCSR_STATE_MASK;
   1125 
   1126 	if (now == state)
   1127 		return 0;
   1128 	switch (state) {
   1129 	case PCI_PMCSR_STATE_D0:
   1130 		break;
   1131 	case PCI_PMCSR_STATE_D1:
   1132 		if (now == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D3) {
   1133 			printf("invalid transition from %d to D1\n", (int)now);
   1134 			return EINVAL;
   1135 		}
   1136 		if (!(cap & PCI_PMCR_D1SUPP)) {
   1137 			printf("D1 not supported\n");
   1138 			return EOPNOTSUPP;
   1139 		}
   1140 		break;
   1141 	case PCI_PMCSR_STATE_D2:
   1142 		if (now == PCI_PMCSR_STATE_D3) {
   1143 			printf("invalid transition from %d to D2\n", (int)now);
   1144 			return EINVAL;
   1145 		}
   1146 		if (!(cap & PCI_PMCR_D2SUPP)) {
   1147 			printf("D2 not supported\n");
   1148 			return EOPNOTSUPP;
   1149 		}
   1150 		break;
   1151 	case PCI_PMCSR_STATE_D3:
   1152 		break;
   1153 	default:
   1154 		return EINVAL;
   1155 	}
   1156 	value |= state;
   1157 	pci_conf_write(pc, tag, offset + PCI_PMCSR, value);
   1158 	/* delay according to pcipm1.2, ch. 5.6.1 */
   1159 	if (state == PCI_PMCSR_STATE_D3 || now == PCI_PMCSR_STATE_D3)
   1160 		DELAY(10000);
   1161 	else if (state == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D2)
   1162 		DELAY(200);
   1163 
   1164 	return 0;
   1165 }
   1166 
   1167 int
   1168 pci_set_powerstate(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state)
   1169 {
   1170 	int offset;
   1171 	pcireg_t value;
   1172 
   1173 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value)) {
   1174 		printf("pci_set_powerstate not supported\n");
   1175 		return EOPNOTSUPP;
   1176 	}
   1177 
   1178 	return pci_set_powerstate_int(pc, tag, state, offset, value);
   1179 }
   1180 
   1181 int
   1182 pci_activate(pci_chipset_tag_t pc, pcitag_t tag, device_t dev,
   1183     int (*wakefun)(pci_chipset_tag_t, pcitag_t, device_t, pcireg_t))
   1184 {
   1185 	pcireg_t pmode;
   1186 	int error;
   1187 
   1188 	if ((error = pci_get_powerstate(pc, tag, &pmode)))
   1189 		return error;
   1190 
   1191 	switch (pmode) {
   1192 	case PCI_PMCSR_STATE_D0:
   1193 		break;
   1194 	case PCI_PMCSR_STATE_D3:
   1195 		if (wakefun == NULL) {
   1196 			/*
   1197 			 * The card has lost all configuration data in
   1198 			 * this state, so punt.
   1199 			 */
   1200 			aprint_error_dev(dev,
   1201 			    "unable to wake up from power state D3\n");
   1202 			return EOPNOTSUPP;
   1203 		}
   1204 		/*FALLTHROUGH*/
   1205 	default:
   1206 		if (wakefun) {
   1207 			error = (*wakefun)(pc, tag, dev, pmode);
   1208 			if (error)
   1209 				return error;
   1210 		}
   1211 		aprint_normal_dev(dev, "waking up from power state D%d\n",
   1212 		    pmode);
   1213 		if ((error = pci_set_powerstate(pc, tag, PCI_PMCSR_STATE_D0)))
   1214 			return error;
   1215 	}
   1216 	return 0;
   1217 }
   1218 
   1219 int
   1220 pci_activate_null(pci_chipset_tag_t pc, pcitag_t tag,
   1221     device_t dev, pcireg_t state)
   1222 {
   1223 	return 0;
   1224 }
   1225 
   1226 struct pci_child_power {
   1227 	struct pci_conf_state p_pciconf;
   1228 	pci_chipset_tag_t p_pc;
   1229 	pcitag_t p_tag;
   1230 	bool p_has_pm;
   1231 	int p_pm_offset;
   1232 	pcireg_t p_pm_cap;
   1233 	pcireg_t p_class;
   1234 	pcireg_t p_csr;
   1235 };
   1236 
   1237 static bool
   1238 pci_child_suspend(device_t dv, const pmf_qual_t *qual)
   1239 {
   1240 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1241 	pcireg_t ocsr, csr;
   1242 
   1243 	pci_conf_capture(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1244 
   1245 	if (!priv->p_has_pm)
   1246 		return true; /* ??? hopefully handled by ACPI */
   1247 	if (PCI_CLASS(priv->p_class) == PCI_CLASS_DISPLAY)
   1248 		return true; /* XXX */
   1249 
   1250 	/* disable decoding and busmastering, see pcipm1.2 ch. 8.2.1 */
   1251 	ocsr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1252 	csr = ocsr & ~(PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE
   1253 		       | PCI_COMMAND_MASTER_ENABLE);
   1254 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1255 	if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1256 	    PCI_PMCSR_STATE_D3, priv->p_pm_offset, priv->p_pm_cap)) {
   1257 		pci_conf_write(priv->p_pc, priv->p_tag,
   1258 			       PCI_COMMAND_STATUS_REG, ocsr);
   1259 		aprint_error_dev(dv, "unsupported state, continuing.\n");
   1260 		return false;
   1261 	}
   1262 	return true;
   1263 }
   1264 
   1265 static void
   1266 pci_pme_check_and_clear(device_t dv, pci_chipset_tag_t pc, pcitag_t tag,
   1267     int off)
   1268 {
   1269 	pcireg_t pmcsr;
   1270 
   1271 	pmcsr = pci_conf_read(pc, tag, off + PCI_PMCSR);
   1272 
   1273 	if (pmcsr & PCI_PMCSR_PME_STS) {
   1274 		/* Clear W1C bit */
   1275 		pmcsr |= PCI_PMCSR_PME_STS;
   1276 		pci_conf_write(pc, tag, off + PCI_PMCSR, pmcsr);
   1277 		aprint_verbose_dev(dv, "Clear PME# now\n");
   1278 	}
   1279 }
   1280 
   1281 static bool
   1282 pci_child_resume(device_t dv, const pmf_qual_t *qual)
   1283 {
   1284 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1285 
   1286 	if (priv->p_has_pm) {
   1287 		if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1288 		    PCI_PMCSR_STATE_D0, priv->p_pm_offset, priv->p_pm_cap)) {
   1289 			aprint_error_dev(dv,
   1290 			    "unsupported state, continuing.\n");
   1291 			return false;
   1292 		}
   1293 		pci_pme_check_and_clear(dv, priv->p_pc, priv->p_tag,
   1294 		    priv->p_pm_offset);
   1295 	}
   1296 
   1297 	pci_conf_restore(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1298 
   1299 	return true;
   1300 }
   1301 
   1302 static bool
   1303 pci_child_shutdown(device_t dv, int how)
   1304 {
   1305 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1306 	pcireg_t csr;
   1307 
   1308 	/* restore original bus-mastering state */
   1309 	csr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1310 	csr &= ~PCI_COMMAND_MASTER_ENABLE;
   1311 	csr |= priv->p_csr & PCI_COMMAND_MASTER_ENABLE;
   1312 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1313 	return true;
   1314 }
   1315 
   1316 static void
   1317 pci_child_deregister(device_t dv)
   1318 {
   1319 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1320 
   1321 	free(priv, M_DEVBUF);
   1322 }
   1323 
   1324 static bool
   1325 pci_child_register(device_t child)
   1326 {
   1327 	device_t self = device_parent(child);
   1328 	struct pci_softc *sc = device_private(self);
   1329 	struct pci_child_power *priv;
   1330 	int device, function, off;
   1331 	pcireg_t reg;
   1332 
   1333 	priv = malloc(sizeof(*priv), M_DEVBUF, M_WAITOK);
   1334 
   1335 	device = device_locator(child, PCICF_DEV);
   1336 	function = device_locator(child, PCICF_FUNCTION);
   1337 
   1338 	priv->p_pc = sc->sc_pc;
   1339 	priv->p_tag = pci_make_tag(priv->p_pc, sc->sc_bus, device,
   1340 	    function);
   1341 	priv->p_class = pci_conf_read(priv->p_pc, priv->p_tag, PCI_CLASS_REG);
   1342 	priv->p_csr = pci_conf_read(priv->p_pc, priv->p_tag,
   1343 	    PCI_COMMAND_STATUS_REG);
   1344 
   1345 	if (pci_get_capability(priv->p_pc, priv->p_tag,
   1346 			       PCI_CAP_PWRMGMT, &off, &reg)) {
   1347 		priv->p_has_pm = true;
   1348 		priv->p_pm_offset = off;
   1349 		priv->p_pm_cap = reg;
   1350 		pci_pme_check_and_clear(child, priv->p_pc, priv->p_tag, off);
   1351 	} else {
   1352 		priv->p_has_pm = false;
   1353 		priv->p_pm_offset = -1;
   1354 	}
   1355 
   1356 	device_pmf_bus_register(child, priv, pci_child_suspend,
   1357 	    pci_child_resume, pci_child_shutdown, pci_child_deregister);
   1358 
   1359 	return true;
   1360 }
   1361 
   1362 MODULE(MODULE_CLASS_DRIVER, pci, NULL);
   1363 
   1364 static int
   1365 pci_modcmd(modcmd_t cmd, void *priv)
   1366 {
   1367 	if (cmd == MODULE_CMD_INIT || cmd == MODULE_CMD_FINI)
   1368 		return 0;
   1369 	return ENOTTY;
   1370 }
   1371