Home | History | Annotate | Line # | Download | only in pci
pci.c revision 1.152.14.2
      1 /*	$NetBSD: pci.c,v 1.152.14.2 2020/04/08 14:08:09 martin 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 1.152.14.2 2020/04/08 14:08:09 martin 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 <net/if.h>
     57 
     58 #include "locators.h"
     59 
     60 static bool pci_child_register(device_t);
     61 
     62 #ifdef PCI_CONFIG_DUMP
     63 int pci_config_dump = 1;
     64 #else
     65 int pci_config_dump = 0;
     66 #endif
     67 
     68 int	pciprint(void *, const char *);
     69 
     70 #ifdef PCI_MACHDEP_ENUMERATE_BUS
     71 #define pci_enumerate_bus PCI_MACHDEP_ENUMERATE_BUS
     72 #endif
     73 
     74 /*
     75  * Important note about PCI-ISA bridges:
     76  *
     77  * Callbacks are used to configure these devices so that ISA/EISA bridges
     78  * can attach their child busses after PCI configuration is done.
     79  *
     80  * This works because:
     81  *	(1) there can be at most one ISA/EISA bridge per PCI bus, and
     82  *	(2) any ISA/EISA bridges must be attached to primary PCI
     83  *	    busses (i.e. bus zero).
     84  *
     85  * That boils down to: there can only be one of these outstanding
     86  * at a time, it is cleared when configuring PCI bus 0 before any
     87  * subdevices have been found, and it is run after all subdevices
     88  * of PCI bus 0 have been found.
     89  *
     90  * This is needed because there are some (legacy) PCI devices which
     91  * can show up as ISA/EISA devices as well (the prime example of which
     92  * are VGA controllers).  If you attach ISA from a PCI-ISA/EISA bridge,
     93  * and the bridge is seen before the video board is, the board can show
     94  * up as an ISA device, and that can (bogusly) complicate the PCI device's
     95  * attach code, or make the PCI device not be properly attached at all.
     96  *
     97  * We use the generic config_defer() facility to achieve this.
     98  */
     99 
    100 int
    101 pcirescan(device_t self, const char *ifattr, const int *locators)
    102 {
    103 	struct pci_softc *sc = device_private(self);
    104 
    105 	KASSERT(ifattr && !strcmp(ifattr, "pci"));
    106 	KASSERT(locators);
    107 
    108 	pci_enumerate_bus(sc, locators, NULL, NULL);
    109 
    110 	return 0;
    111 }
    112 
    113 int
    114 pcimatch(device_t parent, cfdata_t cf, void *aux)
    115 {
    116 	struct pcibus_attach_args *pba = aux;
    117 
    118 	/* Check the locators */
    119 	if (cf->cf_loc[PCIBUSCF_BUS] != PCIBUSCF_BUS_DEFAULT &&
    120 	    cf->cf_loc[PCIBUSCF_BUS] != pba->pba_bus)
    121 		return 0;
    122 
    123 	/* sanity */
    124 	if (pba->pba_bus < 0 || pba->pba_bus > 255)
    125 		return 0;
    126 
    127 	/*
    128 	 * XXX check other (hardware?) indicators
    129 	 */
    130 
    131 	return 1;
    132 }
    133 
    134 void
    135 pciattach(device_t parent, device_t self, void *aux)
    136 {
    137 	struct pcibus_attach_args *pba = aux;
    138 	struct pci_softc *sc = device_private(self);
    139 	int io_enabled, mem_enabled, mrl_enabled, mrm_enabled, mwi_enabled;
    140 	const char *sep = "";
    141 	static const int wildcard[PCICF_NLOCS] = {
    142 		PCICF_DEV_DEFAULT, PCICF_FUNCTION_DEFAULT
    143 	};
    144 
    145 	sc->sc_dev = self;
    146 
    147 	pci_attach_hook(parent, self, pba);
    148 
    149 	aprint_naive("\n");
    150 	aprint_normal("\n");
    151 
    152 	io_enabled = (pba->pba_flags & PCI_FLAGS_IO_OKAY);
    153 	mem_enabled = (pba->pba_flags & PCI_FLAGS_MEM_OKAY);
    154 	mrl_enabled = (pba->pba_flags & PCI_FLAGS_MRL_OKAY);
    155 	mrm_enabled = (pba->pba_flags & PCI_FLAGS_MRM_OKAY);
    156 	mwi_enabled = (pba->pba_flags & PCI_FLAGS_MWI_OKAY);
    157 
    158 	if (io_enabled == 0 && mem_enabled == 0) {
    159 		aprint_error_dev(self, "no spaces enabled!\n");
    160 		goto fail;
    161 	}
    162 
    163 #define	PRINT(str)							\
    164 do {									\
    165 	aprint_verbose("%s%s", sep, str);				\
    166 	sep = ", ";							\
    167 } while (/*CONSTCOND*/0)
    168 
    169 	aprint_verbose_dev(self, "");
    170 
    171 	if (io_enabled)
    172 		PRINT("i/o space");
    173 	if (mem_enabled)
    174 		PRINT("memory space");
    175 	aprint_verbose(" enabled");
    176 
    177 	if (mrl_enabled || mrm_enabled || mwi_enabled) {
    178 		if (mrl_enabled)
    179 			PRINT("rd/line");
    180 		if (mrm_enabled)
    181 			PRINT("rd/mult");
    182 		if (mwi_enabled)
    183 			PRINT("wr/inv");
    184 		aprint_verbose(" ok");
    185 	}
    186 
    187 	aprint_verbose("\n");
    188 
    189 #undef PRINT
    190 
    191 	sc->sc_iot = pba->pba_iot;
    192 	sc->sc_memt = pba->pba_memt;
    193 	sc->sc_dmat = pba->pba_dmat;
    194 	sc->sc_dmat64 = pba->pba_dmat64;
    195 	sc->sc_pc = pba->pba_pc;
    196 	sc->sc_bus = pba->pba_bus;
    197 	sc->sc_bridgetag = pba->pba_bridgetag;
    198 	sc->sc_maxndevs = pci_bus_maxdevs(pba->pba_pc, pba->pba_bus);
    199 	sc->sc_intrswiz = pba->pba_intrswiz;
    200 	sc->sc_intrtag = pba->pba_intrtag;
    201 	sc->sc_flags = pba->pba_flags;
    202 
    203 	device_pmf_driver_set_child_register(sc->sc_dev, pci_child_register);
    204 
    205 	pcirescan(sc->sc_dev, "pci", wildcard);
    206 
    207 fail:
    208 	if (!pmf_device_register(self, NULL, NULL))
    209 		aprint_error_dev(self, "couldn't establish power handler\n");
    210 }
    211 
    212 int
    213 pcidetach(device_t self, int flags)
    214 {
    215 	int rc;
    216 
    217 	if ((rc = config_detach_children(self, flags)) != 0)
    218 		return rc;
    219 	pmf_device_deregister(self);
    220 	return 0;
    221 }
    222 
    223 int
    224 pciprint(void *aux, const char *pnp)
    225 {
    226 	struct pci_attach_args *pa = aux;
    227 	char devinfo[256];
    228 	const struct pci_quirkdata *qd;
    229 
    230 	if (pnp) {
    231 		pci_devinfo(pa->pa_id, pa->pa_class, 1, devinfo, sizeof(devinfo));
    232 		aprint_normal("%s at %s", devinfo, pnp);
    233 	}
    234 	aprint_normal(" dev %d function %d", pa->pa_device, pa->pa_function);
    235 	if (pci_config_dump) {
    236 		printf(": ");
    237 		pci_conf_print(pa->pa_pc, pa->pa_tag, NULL);
    238 		if (!pnp)
    239 			pci_devinfo(pa->pa_id, pa->pa_class, 1, devinfo, sizeof(devinfo));
    240 		printf("%s at %s", devinfo, pnp ? pnp : "?");
    241 		printf(" dev %d function %d (", pa->pa_device, pa->pa_function);
    242 #ifdef __i386__
    243 		printf("tag %#lx, intrtag %#lx, intrswiz %#lx, intrpin %#lx",
    244 		    *(long *)&pa->pa_tag, *(long *)&pa->pa_intrtag,
    245 		    (long)pa->pa_intrswiz, (long)pa->pa_intrpin);
    246 #else
    247 		printf("intrswiz %#lx, intrpin %#lx",
    248 		    (long)pa->pa_intrswiz, (long)pa->pa_intrpin);
    249 #endif
    250 		printf(", i/o %s, mem %s,",
    251 		    pa->pa_flags & PCI_FLAGS_IO_OKAY ? "on" : "off",
    252 		    pa->pa_flags & PCI_FLAGS_MEM_OKAY ? "on" : "off");
    253 		qd = pci_lookup_quirkdata(PCI_VENDOR(pa->pa_id),
    254 		    PCI_PRODUCT(pa->pa_id));
    255 		if (qd == NULL) {
    256 			printf(" no quirks");
    257 		} else {
    258 			snprintb(devinfo, sizeof (devinfo),
    259 			    "\002\001multifn\002singlefn\003skipfunc0"
    260 			    "\004skipfunc1\005skipfunc2\006skipfunc3"
    261 			    "\007skipfunc4\010skipfunc5\011skipfunc6"
    262 			    "\012skipfunc7", qd->quirks);
    263 			printf(" quirks %s", devinfo);
    264 		}
    265 		printf(")");
    266 	}
    267 	return UNCONF;
    268 }
    269 
    270 int
    271 pci_probe_device(struct pci_softc *sc, pcitag_t tag,
    272     int (*match)(const struct pci_attach_args *),
    273     struct pci_attach_args *pap)
    274 {
    275 	pci_chipset_tag_t pc = sc->sc_pc;
    276 	struct pci_attach_args pa;
    277 	pcireg_t id, /* csr, */ pciclass, intr, bhlcr, bar, endbar;
    278 #ifdef __HAVE_PCI_MSI_MSIX
    279 	pcireg_t cap;
    280 	int off;
    281 #endif
    282 	int ret, pin, bus, device, function, i, width;
    283 	int locs[PCICF_NLOCS];
    284 
    285 	pci_decompose_tag(pc, tag, &bus, &device, &function);
    286 
    287 	/* a driver already attached? */
    288 	if (sc->PCI_SC_DEVICESC(device, function).c_dev != NULL && !match)
    289 		return 0;
    290 
    291 	id = pci_conf_read(pc, tag, PCI_ID_REG);
    292 
    293 	/* Invalid vendor ID value? */
    294 	if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    295 		return 0;
    296 	/* XXX Not invalid, but we've done this ~forever. */
    297 	if (PCI_VENDOR(id) == 0)
    298 		return 0;
    299 
    300 	bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    301 	if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    302 		return 0;
    303 
    304 	/* csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG); */
    305 	pciclass = pci_conf_read(pc, tag, PCI_CLASS_REG);
    306 
    307 	/* Collect memory range info */
    308 	memset(sc->PCI_SC_DEVICESC(device, function).c_range, 0,
    309 	    sizeof(sc->PCI_SC_DEVICESC(device, function).c_range));
    310 	i = 0;
    311 	switch (PCI_HDRTYPE_TYPE(bhlcr)) {
    312 	case PCI_HDRTYPE_PPB:
    313 		endbar = PCI_MAPREG_PPB_END;
    314 		break;
    315 	case PCI_HDRTYPE_PCB:
    316 		endbar = PCI_MAPREG_PCB_END;
    317 		break;
    318 	default:
    319 		endbar = PCI_MAPREG_END;
    320 		break;
    321 	}
    322 	for (bar = PCI_MAPREG_START; bar < endbar; bar += width) {
    323 		struct pci_range *r;
    324 		pcireg_t type;
    325 
    326 		width = 4;
    327 		if (pci_mapreg_probe(pc, tag, bar, &type) == 0)
    328 			continue;
    329 
    330 		if (PCI_MAPREG_TYPE(type) == PCI_MAPREG_TYPE_MEM) {
    331 			if (PCI_MAPREG_MEM_TYPE(type) ==
    332 			    PCI_MAPREG_MEM_TYPE_64BIT)
    333 				width = 8;
    334 
    335 			r = &sc->PCI_SC_DEVICESC(device, function).c_range[i++];
    336 			if (pci_mapreg_info(pc, tag, bar, type,
    337 			    &r->r_offset, &r->r_size, &r->r_flags) != 0)
    338 				break;
    339 			if ((PCI_VENDOR(id) == PCI_VENDOR_ATI) && (bar == 0x10)
    340 			    && (r->r_size == 0x1000000)) {
    341 				struct pci_range *nr;
    342 				/*
    343 				 * this has to be a mach64
    344 				 * split things up so each half-aperture can
    345 				 * be mapped PREFETCHABLE except the last page
    346 				 * which may contain registers
    347 				 */
    348 				r->r_size = 0x7ff000;
    349 				r->r_flags = BUS_SPACE_MAP_LINEAR |
    350 					     BUS_SPACE_MAP_PREFETCHABLE;
    351 				nr = &sc->PCI_SC_DEVICESC(device,
    352 				    function).c_range[i++];
    353 				nr->r_offset = r->r_offset + 0x800000;
    354 				nr->r_size = 0x7ff000;
    355 				nr->r_flags = BUS_SPACE_MAP_LINEAR |
    356 					      BUS_SPACE_MAP_PREFETCHABLE;
    357 			} else if ((PCI_VENDOR(id) == PCI_VENDOR_SILMOTION) &&
    358 			   (PCI_PRODUCT(id) == PCI_PRODUCT_SILMOTION_SM502) &&
    359 			   (bar == 0x10)) {
    360 			   	r->r_flags = BUS_SPACE_MAP_LINEAR |
    361 					     BUS_SPACE_MAP_PREFETCHABLE;
    362 			}
    363 		}
    364 	}
    365 
    366 	pa.pa_iot = sc->sc_iot;
    367 	pa.pa_memt = sc->sc_memt;
    368 	pa.pa_dmat = sc->sc_dmat;
    369 	pa.pa_dmat64 = sc->sc_dmat64;
    370 	pa.pa_pc = pc;
    371 	pa.pa_bus = bus;
    372 	pa.pa_device = device;
    373 	pa.pa_function = function;
    374 	pa.pa_tag = tag;
    375 	pa.pa_id = id;
    376 	pa.pa_class = pciclass;
    377 
    378 	/*
    379 	 * Set up memory, I/O enable, and PCI command flags
    380 	 * as appropriate.
    381 	 */
    382 	pa.pa_flags = sc->sc_flags;
    383 
    384 	/*
    385 	 * If the cache line size is not configured, then
    386 	 * clear the MRL/MRM/MWI command-ok flags.
    387 	 */
    388 	if (PCI_CACHELINE(bhlcr) == 0) {
    389 		pa.pa_flags &= ~(PCI_FLAGS_MRL_OKAY|
    390 		    PCI_FLAGS_MRM_OKAY|PCI_FLAGS_MWI_OKAY);
    391 	}
    392 
    393 	if (sc->sc_bridgetag == NULL) {
    394 		pa.pa_intrswiz = 0;
    395 		pa.pa_intrtag = tag;
    396 	} else {
    397 		pa.pa_intrswiz = sc->sc_intrswiz + device;
    398 		pa.pa_intrtag = sc->sc_intrtag;
    399 	}
    400 
    401 	intr = pci_conf_read(pc, tag, PCI_INTERRUPT_REG);
    402 
    403 	pin = PCI_INTERRUPT_PIN(intr);
    404 	pa.pa_rawintrpin = pin;
    405 	if (pin == PCI_INTERRUPT_PIN_NONE) {
    406 		/* no interrupt */
    407 		pa.pa_intrpin = 0;
    408 	} else {
    409 		/*
    410 		 * swizzle it based on the number of busses we're
    411 		 * behind and our device number.
    412 		 */
    413 		pa.pa_intrpin = 	/* XXX */
    414 		    ((pin + pa.pa_intrswiz - 1) % 4) + 1;
    415 	}
    416 	pa.pa_intrline = PCI_INTERRUPT_LINE(intr);
    417 
    418 #ifdef __HAVE_PCI_MSI_MSIX
    419 	if (pci_get_ht_capability(pc, tag, PCI_HT_CAP_MSIMAP, &off, &cap)) {
    420 		/*
    421 		 * XXX Should we enable MSI mapping ourselves on
    422 		 * systems that have it disabled?
    423 		 */
    424 		if (cap & PCI_HT_MSI_ENABLED) {
    425 			uint64_t addr;
    426 			if ((cap & PCI_HT_MSI_FIXED) == 0) {
    427 				addr = pci_conf_read(pc, tag,
    428 				    off + PCI_HT_MSI_ADDR_LO);
    429 				addr |= (uint64_t)pci_conf_read(pc, tag,
    430 				    off + PCI_HT_MSI_ADDR_HI) << 32;
    431 			} else
    432 				addr = PCI_HT_MSI_FIXED_ADDR;
    433 
    434 			/*
    435 			 * XXX This will fail to enable MSI on systems
    436 			 * that don't use the canonical address.
    437 			 */
    438 			if (addr == PCI_HT_MSI_FIXED_ADDR) {
    439 				pa.pa_flags |= PCI_FLAGS_MSI_OKAY;
    440 				pa.pa_flags |= PCI_FLAGS_MSIX_OKAY;
    441 			} else
    442 				aprint_verbose_dev(sc->sc_dev,
    443 				    "HyperTransport MSI mapping is not supported yet. Disable MSI/MSI-X.\n");
    444 		}
    445 	}
    446 #endif
    447 
    448 	if (match != NULL) {
    449 		ret = (*match)(&pa);
    450 		if (ret != 0 && pap != NULL)
    451 			*pap = pa;
    452 	} else {
    453 		struct pci_child *c;
    454 		locs[PCICF_DEV] = device;
    455 		locs[PCICF_FUNCTION] = function;
    456 
    457 		c = &sc->PCI_SC_DEVICESC(device, function);
    458 		pci_conf_capture(pc, tag, &c->c_conf);
    459 		if (pci_get_powerstate(pc, tag, &c->c_powerstate) == 0)
    460 			c->c_psok = true;
    461 		else
    462 			c->c_psok = false;
    463 
    464 		c->c_dev = config_found_sm_loc(sc->sc_dev, "pci", locs, &pa,
    465 					     pciprint, config_stdsubmatch);
    466 
    467 		ret = (c->c_dev != NULL);
    468 	}
    469 
    470 	return ret;
    471 }
    472 
    473 void
    474 pcidevdetached(device_t self, device_t child)
    475 {
    476 	struct pci_softc *sc = device_private(self);
    477 	int d, f;
    478 	pcitag_t tag;
    479 	struct pci_child *c;
    480 
    481 	d = device_locator(child, PCICF_DEV);
    482 	f = device_locator(child, PCICF_FUNCTION);
    483 
    484 	c = &sc->PCI_SC_DEVICESC(d, f);
    485 
    486 	KASSERT(c->c_dev == child);
    487 
    488 	tag = pci_make_tag(sc->sc_pc, sc->sc_bus, d, f);
    489 	if (c->c_psok)
    490 		pci_set_powerstate(sc->sc_pc, tag, c->c_powerstate);
    491 	pci_conf_restore(sc->sc_pc, tag, &c->c_conf);
    492 	c->c_dev = NULL;
    493 }
    494 
    495 CFATTACH_DECL3_NEW(pci, sizeof(struct pci_softc),
    496     pcimatch, pciattach, pcidetach, NULL, pcirescan, pcidevdetached,
    497     DVF_DETACH_SHUTDOWN);
    498 
    499 int
    500 pci_get_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    501     int *offset, pcireg_t *value)
    502 {
    503 	pcireg_t reg;
    504 	unsigned int ofs;
    505 
    506 	reg = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
    507 	if (!(reg & PCI_STATUS_CAPLIST_SUPPORT))
    508 		return 0;
    509 
    510 	/* Determine the Capability List Pointer register to start with. */
    511 	reg = pci_conf_read(pc, tag, PCI_BHLC_REG);
    512 	switch (PCI_HDRTYPE_TYPE(reg)) {
    513 	case 0:	/* standard device header */
    514 	case 1: /* PCI-PCI bridge header */
    515 		ofs = PCI_CAPLISTPTR_REG;
    516 		break;
    517 	case 2:	/* PCI-CardBus Bridge header */
    518 		ofs = PCI_CARDBUS_CAPLISTPTR_REG;
    519 		break;
    520 	default:
    521 		return 0;
    522 	}
    523 
    524 	ofs = PCI_CAPLIST_PTR(pci_conf_read(pc, tag, ofs));
    525 	while (ofs != 0) {
    526 		if ((ofs & 3) || (ofs < 0x40)) {
    527 			int bus, device, function;
    528 
    529 			pci_decompose_tag(pc, tag, &bus, &device, &function);
    530 
    531 			printf("Skipping broken PCI header on %d:%d:%d\n",
    532 			    bus, device, function);
    533 			break;
    534 		}
    535 		reg = pci_conf_read(pc, tag, ofs);
    536 		if (PCI_CAPLIST_CAP(reg) == capid) {
    537 			if (offset)
    538 				*offset = ofs;
    539 			if (value)
    540 				*value = reg;
    541 			return 1;
    542 		}
    543 		ofs = PCI_CAPLIST_NEXT(reg);
    544 	}
    545 
    546 	return 0;
    547 }
    548 
    549 int
    550 pci_get_ht_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    551     int *offset, pcireg_t *value)
    552 {
    553 	pcireg_t reg;
    554 	unsigned int ofs;
    555 
    556 	if (pci_get_capability(pc, tag, PCI_CAP_LDT, &ofs, NULL) == 0)
    557 		return 0;
    558 
    559 	while (ofs != 0) {
    560 #ifdef DIAGNOSTIC
    561 		if ((ofs & 3) || (ofs < 0x40))
    562 			panic("pci_get_ht_capability");
    563 #endif
    564 		reg = pci_conf_read(pc, tag, ofs);
    565 		if (PCI_HT_CAP(reg) == capid) {
    566 			if (offset)
    567 				*offset = ofs;
    568 			if (value)
    569 				*value = reg;
    570 			return 1;
    571 		}
    572 		ofs = PCI_CAPLIST_NEXT(reg);
    573 	}
    574 
    575 	return 0;
    576 }
    577 
    578 /*
    579  * return number of the devices's MSI vectors
    580  * return 0 if the device does not support MSI
    581  */
    582 int
    583 pci_msi_count(pci_chipset_tag_t pc, pcitag_t tag)
    584 {
    585 	pcireg_t reg;
    586 	uint32_t mmc;
    587 	int count, offset;
    588 
    589 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &offset, NULL) == 0)
    590 		return 0;
    591 
    592 	reg = pci_conf_read(pc, tag, offset + PCI_MSI_CTL);
    593 	mmc = PCI_MSI_CTL_MMC(reg);
    594 	count = 1 << mmc;
    595 	if (count > PCI_MSI_MAX_VECTORS) {
    596 		aprint_error("detect an illegal device! The device use reserved MMC values.\n");
    597 		return 0;
    598 	}
    599 
    600 	return count;
    601 }
    602 
    603 /*
    604  * return number of the devices's MSI-X vectors
    605  * return 0 if the device does not support MSI-X
    606  */
    607 int
    608 pci_msix_count(pci_chipset_tag_t pc, pcitag_t tag)
    609 {
    610 	pcireg_t reg;
    611 	int offset;
    612 
    613 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &offset, NULL) == 0)
    614 		return 0;
    615 
    616 	reg = pci_conf_read(pc, tag, offset + PCI_MSIX_CTL);
    617 
    618 	return PCI_MSIX_CTL_TBLSIZE(reg);
    619 }
    620 
    621 int
    622 pci_get_ext_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    623     int *offset, pcireg_t *value)
    624 {
    625 	pcireg_t reg;
    626 	unsigned int ofs;
    627 
    628 	/* Only supported for PCI-express devices */
    629 	if (!pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, NULL, NULL))
    630 		return 0;
    631 
    632 	ofs = PCI_EXTCAPLIST_BASE;
    633 	reg = pci_conf_read(pc, tag, ofs);
    634 	if (reg == 0xffffffff || reg == 0)
    635 		return 0;
    636 
    637 	for (;;) {
    638 #ifdef DIAGNOSTIC
    639 		if ((ofs & 3) || ofs < PCI_EXTCAPLIST_BASE)
    640 			panic("%s: invalid offset %u", __func__, ofs);
    641 #endif
    642 		if (PCI_EXTCAPLIST_CAP(reg) == capid) {
    643 			if (offset != NULL)
    644 				*offset = ofs;
    645 			if (value != NULL)
    646 				*value = reg;
    647 			return 1;
    648 		}
    649 		ofs = PCI_EXTCAPLIST_NEXT(reg);
    650 		if (ofs == 0)
    651 			break;
    652 		reg = pci_conf_read(pc, tag, ofs);
    653 	}
    654 
    655 	return 0;
    656 }
    657 
    658 int
    659 pci_find_device(struct pci_attach_args *pa,
    660 		int (*match)(const struct pci_attach_args *))
    661 {
    662 	extern struct cfdriver pci_cd;
    663 	device_t pcidev;
    664 	int i;
    665 	static const int wildcard[2] = {
    666 		PCICF_DEV_DEFAULT,
    667 		PCICF_FUNCTION_DEFAULT
    668 	};
    669 
    670 	for (i = 0; i < pci_cd.cd_ndevs; i++) {
    671 		pcidev = device_lookup(&pci_cd, i);
    672 		if (pcidev != NULL &&
    673 		    pci_enumerate_bus(device_private(pcidev), wildcard,
    674 		    		      match, pa) != 0)
    675 			return 1;
    676 	}
    677 	return 0;
    678 }
    679 
    680 #ifndef PCI_MACHDEP_ENUMERATE_BUS
    681 /*
    682  * Generic PCI bus enumeration routine.  Used unless machine-dependent
    683  * code needs to provide something else.
    684  */
    685 int
    686 pci_enumerate_bus(struct pci_softc *sc, const int *locators,
    687     int (*match)(const struct pci_attach_args *), struct pci_attach_args *pap)
    688 {
    689 	pci_chipset_tag_t pc = sc->sc_pc;
    690 	int device, function, nfunctions, ret;
    691 	const struct pci_quirkdata *qd;
    692 	pcireg_t id, bhlcr;
    693 	pcitag_t tag;
    694 	uint8_t devs[32];
    695 	int i, n;
    696 
    697 	device_t bridgedev;
    698 	bool arien = false;
    699 
    700 	/* Check PCIe ARI */
    701 	bridgedev = device_parent(sc->sc_dev);
    702 	if (device_is_a(bridgedev, "ppb")) {
    703 		struct ppb_softc *ppbsc = device_private(bridgedev);
    704 		pci_chipset_tag_t ppbpc = ppbsc->sc_pc;
    705 		pcitag_t ppbtag = ppbsc->sc_tag;
    706 		pcireg_t pciecap, reg;
    707 
    708 		if (pci_get_capability(ppbpc, ppbtag, PCI_CAP_PCIEXPRESS,
    709 		    &pciecap, NULL) != 0) {
    710 			reg = pci_conf_read(ppbpc, ppbtag, pciecap
    711 			    + PCIE_DCSR2);
    712 			if ((reg & PCIE_DCSR2_ARI_FWD) != 0)
    713 				arien = true;
    714 		}
    715 	}
    716 
    717 	n = pci_bus_devorder(sc->sc_pc, sc->sc_bus, devs, __arraycount(devs));
    718 	for (i = 0; i < n; i++) {
    719 		device = devs[i];
    720 
    721 		if ((locators[PCICF_DEV] != PCICF_DEV_DEFAULT) &&
    722 		    (locators[PCICF_DEV] != device))
    723 			continue;
    724 
    725 		tag = pci_make_tag(pc, sc->sc_bus, device, 0);
    726 
    727 		bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    728 		if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    729 			continue;
    730 
    731 		id = pci_conf_read(pc, tag, PCI_ID_REG);
    732 
    733 		/* Invalid vendor ID value? */
    734 		if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    735 			continue;
    736 		/* XXX Not invalid, but we've done this ~forever. */
    737 		if (PCI_VENDOR(id) == 0)
    738 			continue;
    739 
    740 		qd = pci_lookup_quirkdata(PCI_VENDOR(id), PCI_PRODUCT(id));
    741 
    742 		if (qd != NULL &&
    743 		      (qd->quirks & PCI_QUIRK_MULTIFUNCTION) != 0)
    744 			nfunctions = 8;
    745 		else if (qd != NULL &&
    746 		      (qd->quirks & PCI_QUIRK_MONOFUNCTION) != 0)
    747 			nfunctions = 1;
    748 		else if (arien)
    749 			nfunctions = 8; /* Scan all if ARI is enabled */
    750 		else
    751 			nfunctions = PCI_HDRTYPE_MULTIFN(bhlcr) ? 8 : 1;
    752 
    753 #ifdef __PCI_DEV_FUNCORDER
    754 		char funcs[8];
    755 		int j;
    756 		for (j = 0; j < nfunctions; j++) {
    757 			funcs[j] = j;
    758 		}
    759 		if (j < __arraycount(funcs))
    760 			funcs[j] = -1;
    761 		if (nfunctions > 1) {
    762 			pci_dev_funcorder(sc->sc_pc, sc->sc_bus, device,
    763 			    nfunctions, funcs);
    764 		}
    765 		for (j = 0;
    766 		     j < 8 && (function = funcs[j]) < 8 && function >= 0;
    767 		     j++) {
    768 #else
    769 		for (function = 0; function < nfunctions; function++) {
    770 #endif
    771 			if ((locators[PCICF_FUNCTION] != PCICF_FUNCTION_DEFAULT)
    772 			    && (locators[PCICF_FUNCTION] != function))
    773 				continue;
    774 
    775 			if (qd != NULL &&
    776 			    (qd->quirks & PCI_QUIRK_SKIP_FUNC(function)) != 0)
    777 				continue;
    778 			tag = pci_make_tag(pc, sc->sc_bus, device, function);
    779 			ret = pci_probe_device(sc, tag, match, pap);
    780 			if (match != NULL && ret != 0)
    781 				return ret;
    782 		}
    783 	}
    784 	return 0;
    785 }
    786 #endif /* PCI_MACHDEP_ENUMERATE_BUS */
    787 
    788 
    789 /*
    790  * Vital Product Data (PCI 2.2)
    791  */
    792 
    793 int
    794 pci_vpd_read(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    795     pcireg_t *data)
    796 {
    797 	uint32_t reg;
    798 	int ofs, i, j;
    799 
    800 	KASSERT(data != NULL);
    801 	KASSERT((offset + count) < 0x7fff);
    802 
    803 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    804 		return 1;
    805 
    806 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    807 		reg &= 0x0000ffff;
    808 		reg &= ~PCI_VPD_OPFLAG;
    809 		reg |= PCI_VPD_ADDRESS(offset);
    810 		pci_conf_write(pc, tag, ofs, reg);
    811 
    812 		/*
    813 		 * PCI 2.2 does not specify how long we should poll
    814 		 * for completion nor whether the operation can fail.
    815 		 */
    816 		j = 0;
    817 		do {
    818 			if (j++ == 20)
    819 				return 1;
    820 			delay(4);
    821 			reg = pci_conf_read(pc, tag, ofs);
    822 		} while ((reg & PCI_VPD_OPFLAG) == 0);
    823 		data[i] = pci_conf_read(pc, tag, PCI_VPD_DATAREG(ofs));
    824 	}
    825 
    826 	return 0;
    827 }
    828 
    829 int
    830 pci_vpd_write(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    831     pcireg_t *data)
    832 {
    833 	pcireg_t reg;
    834 	int ofs, i, j;
    835 
    836 	KASSERT(data != NULL);
    837 	KASSERT((offset + count) < 0x7fff);
    838 
    839 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    840 		return 1;
    841 
    842 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    843 		pci_conf_write(pc, tag, PCI_VPD_DATAREG(ofs), data[i]);
    844 
    845 		reg &= 0x0000ffff;
    846 		reg |= PCI_VPD_OPFLAG;
    847 		reg |= PCI_VPD_ADDRESS(offset);
    848 		pci_conf_write(pc, tag, ofs, reg);
    849 
    850 		/*
    851 		 * PCI 2.2 does not specify how long we should poll
    852 		 * for completion nor whether the operation can fail.
    853 		 */
    854 		j = 0;
    855 		do {
    856 			if (j++ == 20)
    857 				return 1;
    858 			delay(1);
    859 			reg = pci_conf_read(pc, tag, ofs);
    860 		} while (reg & PCI_VPD_OPFLAG);
    861 	}
    862 
    863 	return 0;
    864 }
    865 
    866 int
    867 pci_dma64_available(const struct pci_attach_args *pa)
    868 {
    869 #ifdef _PCI_HAVE_DMA64
    870 	if (BUS_DMA_TAG_VALID(pa->pa_dmat64))
    871                         return 1;
    872 #endif
    873         return 0;
    874 }
    875 
    876 void
    877 pci_conf_capture(pci_chipset_tag_t pc, pcitag_t tag,
    878 		  struct pci_conf_state *pcs)
    879 {
    880 	int off;
    881 
    882 	for (off = 0; off < 16; off++)
    883 		pcs->reg[off] = pci_conf_read(pc, tag, (off * 4));
    884 
    885 	/* For PCI-X */
    886 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
    887 		pcs->x_csr = pci_conf_read(pc, tag, off + PCIX_CMD);
    888 
    889 	/* For PCIe */
    890 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
    891 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
    892 		unsigned int devtype;
    893 
    894 		devtype = PCIE_XCAP_TYPE(xcap);
    895 		pcs->e_dcr = (uint16_t)pci_conf_read(pc, tag, off + PCIE_DCSR);
    896 
    897 		if (PCIE_HAS_LINKREGS(devtype))
    898 			pcs->e_lcr = (uint16_t)pci_conf_read(pc, tag,
    899 			    off + PCIE_LCSR);
    900 
    901 		if ((xcap & PCIE_XCAP_SI) != 0)
    902 			pcs->e_slcr = (uint16_t)pci_conf_read(pc, tag,
    903 			    off + PCIE_SLCSR);
    904 
    905 		if (PCIE_HAS_ROOTREGS(devtype))
    906 			pcs->e_rcr = (uint16_t)pci_conf_read(pc, tag,
    907 			    off + PCIE_RCR);
    908 
    909 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
    910 			pcs->e_dcr2 = (uint16_t)pci_conf_read(pc, tag,
    911 			    off + PCIE_DCSR2);
    912 
    913 			if (PCIE_HAS_LINKREGS(devtype))
    914 				pcs->e_lcr2 = (uint16_t)pci_conf_read(pc, tag,
    915 			    off + PCIE_LCSR2);
    916 
    917 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
    918 		}
    919 	}
    920 
    921 	/* For MSI */
    922 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
    923 		bool bit64, pvmask;
    924 
    925 		pcs->msi_ctl = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
    926 
    927 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
    928 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
    929 
    930 		/* Address */
    931 		pcs->msi_maddr = pci_conf_read(pc, tag, off + PCI_MSI_MADDR);
    932 		if (bit64)
    933 			pcs->msi_maddr64_hi = pci_conf_read(pc, tag,
    934 			    off + PCI_MSI_MADDR64_HI);
    935 
    936 		/* Data */
    937 		pcs->msi_mdata = pci_conf_read(pc, tag,
    938 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA));
    939 
    940 		/* Per-vector masking */
    941 		if (pvmask)
    942 			pcs->msi_mask = pci_conf_read(pc, tag,
    943 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK));
    944 	}
    945 
    946 	/* For MSI-X */
    947 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
    948 		pcs->msix_ctl = pci_conf_read(pc, tag, off + PCI_MSIX_CTL);
    949 }
    950 
    951 void
    952 pci_conf_restore(pci_chipset_tag_t pc, pcitag_t tag,
    953 		  struct pci_conf_state *pcs)
    954 {
    955 	int off;
    956 	pcireg_t val;
    957 
    958 	for (off = 15; off >= 0; off--) {
    959 		val = pci_conf_read(pc, tag, (off * 4));
    960 		if (val != pcs->reg[off])
    961 			pci_conf_write(pc, tag, (off * 4), pcs->reg[off]);
    962 	}
    963 
    964 	/* For PCI-X */
    965 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
    966 		pci_conf_write(pc, tag, off + PCIX_CMD, pcs->x_csr);
    967 
    968 	/* For PCIe */
    969 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
    970 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
    971 		unsigned int devtype;
    972 
    973 		devtype = PCIE_XCAP_TYPE(xcap);
    974 		pci_conf_write(pc, tag, off + PCIE_DCSR, pcs->e_dcr);
    975 
    976 		/*
    977 		 * PCIe capability is variable sized. To not to write the next
    978 		 * area, check the existence of each register.
    979 		 */
    980 		if (PCIE_HAS_LINKREGS(devtype))
    981 			pci_conf_write(pc, tag, off + PCIE_LCSR, pcs->e_lcr);
    982 
    983 		if ((xcap & PCIE_XCAP_SI) != 0)
    984 			pci_conf_write(pc, tag, off + PCIE_SLCSR, pcs->e_slcr);
    985 
    986 		if (PCIE_HAS_ROOTREGS(devtype))
    987 			pci_conf_write(pc, tag, off + PCIE_RCR, pcs->e_rcr);
    988 
    989 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
    990 			pci_conf_write(pc, tag, off + PCIE_DCSR2, pcs->e_dcr2);
    991 
    992 			if (PCIE_HAS_LINKREGS(devtype))
    993 				pci_conf_write(pc, tag, off + PCIE_LCSR2,
    994 				    pcs->e_lcr2);
    995 
    996 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
    997 		}
    998 	}
    999 
   1000 	/* For MSI */
   1001 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
   1002 		pcireg_t reg;
   1003 		bool bit64, pvmask;
   1004 
   1005 		/* First, drop Enable bit in case it's already set. */
   1006 		reg = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
   1007 		pci_conf_write(pc, tag, off + PCI_MSI_CTL,
   1008 		    reg & ~PCI_MSI_CTL_MSI_ENABLE);
   1009 
   1010 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
   1011 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
   1012 
   1013 		/* Address */
   1014 		pci_conf_write(pc, tag, off + PCI_MSI_MADDR, pcs->msi_maddr);
   1015 
   1016 		if (bit64)
   1017 			pci_conf_write(pc, tag,
   1018 			    off + PCI_MSI_MADDR64_HI, pcs->msi_maddr64_hi);
   1019 
   1020 		/* Data */
   1021 		pci_conf_write(pc, tag,
   1022 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA),
   1023 		    pcs->msi_mdata);
   1024 
   1025 		/* Per-vector masking */
   1026 		if (pvmask)
   1027 			pci_conf_write(pc, tag,
   1028 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK),
   1029 			    pcs->msi_mask);
   1030 
   1031 		/* Write CTRL register in the end */
   1032 		pci_conf_write(pc, tag, off + PCI_MSI_CTL, pcs->msi_ctl);
   1033 	}
   1034 
   1035 	/* For MSI-X */
   1036 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
   1037 		pci_conf_write(pc, tag, off + PCI_MSIX_CTL, pcs->msix_ctl);
   1038 }
   1039 
   1040 /*
   1041  * Power Management Capability (Rev 2.2)
   1042  */
   1043 static int
   1044 pci_get_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state,
   1045     int offset)
   1046 {
   1047 	pcireg_t value, now;
   1048 
   1049 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1050 	now = value & PCI_PMCSR_STATE_MASK;
   1051 	switch (now) {
   1052 	case PCI_PMCSR_STATE_D0:
   1053 	case PCI_PMCSR_STATE_D1:
   1054 	case PCI_PMCSR_STATE_D2:
   1055 	case PCI_PMCSR_STATE_D3:
   1056 		*state = now;
   1057 		return 0;
   1058 	default:
   1059 		return EINVAL;
   1060 	}
   1061 }
   1062 
   1063 int
   1064 pci_get_powerstate(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state)
   1065 {
   1066 	int offset;
   1067 	pcireg_t value;
   1068 
   1069 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value))
   1070 		return EOPNOTSUPP;
   1071 
   1072 	return pci_get_powerstate_int(pc, tag, state, offset);
   1073 }
   1074 
   1075 static int
   1076 pci_set_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state,
   1077     int offset, pcireg_t cap_reg)
   1078 {
   1079 	pcireg_t value, cap, now;
   1080 
   1081 	cap = cap_reg >> PCI_PMCR_SHIFT;
   1082 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1083 	now = value & PCI_PMCSR_STATE_MASK;
   1084 	value &= ~PCI_PMCSR_STATE_MASK;
   1085 
   1086 	if (now == state)
   1087 		return 0;
   1088 	switch (state) {
   1089 	case PCI_PMCSR_STATE_D0:
   1090 		break;
   1091 	case PCI_PMCSR_STATE_D1:
   1092 		if (now == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D3) {
   1093 			printf("invalid transition from %d to D1\n", (int)now);
   1094 			return EINVAL;
   1095 		}
   1096 		if (!(cap & PCI_PMCR_D1SUPP)) {
   1097 			printf("D1 not supported\n");
   1098 			return EOPNOTSUPP;
   1099 		}
   1100 		break;
   1101 	case PCI_PMCSR_STATE_D2:
   1102 		if (now == PCI_PMCSR_STATE_D3) {
   1103 			printf("invalid transition from %d to D2\n", (int)now);
   1104 			return EINVAL;
   1105 		}
   1106 		if (!(cap & PCI_PMCR_D2SUPP)) {
   1107 			printf("D2 not supported\n");
   1108 			return EOPNOTSUPP;
   1109 		}
   1110 		break;
   1111 	case PCI_PMCSR_STATE_D3:
   1112 		break;
   1113 	default:
   1114 		return EINVAL;
   1115 	}
   1116 	value |= state;
   1117 	pci_conf_write(pc, tag, offset + PCI_PMCSR, value);
   1118 	/* delay according to pcipm1.2, ch. 5.6.1 */
   1119 	if (state == PCI_PMCSR_STATE_D3 || now == PCI_PMCSR_STATE_D3)
   1120 		DELAY(10000);
   1121 	else if (state == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D2)
   1122 		DELAY(200);
   1123 
   1124 	return 0;
   1125 }
   1126 
   1127 int
   1128 pci_set_powerstate(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state)
   1129 {
   1130 	int offset;
   1131 	pcireg_t value;
   1132 
   1133 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value)) {
   1134 		printf("pci_set_powerstate not supported\n");
   1135 		return EOPNOTSUPP;
   1136 	}
   1137 
   1138 	return pci_set_powerstate_int(pc, tag, state, offset, value);
   1139 }
   1140 
   1141 int
   1142 pci_activate(pci_chipset_tag_t pc, pcitag_t tag, device_t dev,
   1143     int (*wakefun)(pci_chipset_tag_t, pcitag_t, device_t, pcireg_t))
   1144 {
   1145 	pcireg_t pmode;
   1146 	int error;
   1147 
   1148 	if ((error = pci_get_powerstate(pc, tag, &pmode)))
   1149 		return error;
   1150 
   1151 	switch (pmode) {
   1152 	case PCI_PMCSR_STATE_D0:
   1153 		break;
   1154 	case PCI_PMCSR_STATE_D3:
   1155 		if (wakefun == NULL) {
   1156 			/*
   1157 			 * The card has lost all configuration data in
   1158 			 * this state, so punt.
   1159 			 */
   1160 			aprint_error_dev(dev,
   1161 			    "unable to wake up from power state D3\n");
   1162 			return EOPNOTSUPP;
   1163 		}
   1164 		/*FALLTHROUGH*/
   1165 	default:
   1166 		if (wakefun) {
   1167 			error = (*wakefun)(pc, tag, dev, pmode);
   1168 			if (error)
   1169 				return error;
   1170 		}
   1171 		aprint_normal_dev(dev, "waking up from power state D%d\n",
   1172 		    pmode);
   1173 		if ((error = pci_set_powerstate(pc, tag, PCI_PMCSR_STATE_D0)))
   1174 			return error;
   1175 	}
   1176 	return 0;
   1177 }
   1178 
   1179 int
   1180 pci_activate_null(pci_chipset_tag_t pc, pcitag_t tag,
   1181     device_t dev, pcireg_t state)
   1182 {
   1183 	return 0;
   1184 }
   1185 
   1186 struct pci_child_power {
   1187 	struct pci_conf_state p_pciconf;
   1188 	pci_chipset_tag_t p_pc;
   1189 	pcitag_t p_tag;
   1190 	bool p_has_pm;
   1191 	int p_pm_offset;
   1192 	pcireg_t p_pm_cap;
   1193 	pcireg_t p_class;
   1194 	pcireg_t p_csr;
   1195 };
   1196 
   1197 static bool
   1198 pci_child_suspend(device_t dv, const pmf_qual_t *qual)
   1199 {
   1200 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1201 	pcireg_t ocsr, csr;
   1202 
   1203 	pci_conf_capture(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1204 
   1205 	if (!priv->p_has_pm)
   1206 		return true; /* ??? hopefully handled by ACPI */
   1207 	if (PCI_CLASS(priv->p_class) == PCI_CLASS_DISPLAY)
   1208 		return true; /* XXX */
   1209 
   1210 	/* disable decoding and busmastering, see pcipm1.2 ch. 8.2.1 */
   1211 	ocsr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1212 	csr = ocsr & ~(PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE
   1213 		       | PCI_COMMAND_MASTER_ENABLE);
   1214 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1215 	if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1216 	    PCI_PMCSR_STATE_D3, priv->p_pm_offset, priv->p_pm_cap)) {
   1217 		pci_conf_write(priv->p_pc, priv->p_tag,
   1218 			       PCI_COMMAND_STATUS_REG, ocsr);
   1219 		aprint_error_dev(dv, "unsupported state, continuing.\n");
   1220 		return false;
   1221 	}
   1222 	return true;
   1223 }
   1224 
   1225 static void
   1226 pci_pme_check_and_clear(device_t dv, pci_chipset_tag_t pc, pcitag_t tag,
   1227     int off)
   1228 {
   1229 	pcireg_t pmcsr;
   1230 
   1231 	pmcsr = pci_conf_read(pc, tag, off + PCI_PMCSR);
   1232 
   1233 	if (pmcsr & PCI_PMCSR_PME_STS) {
   1234 		/* Clear W1C bit */
   1235 		pmcsr |= PCI_PMCSR_PME_STS;
   1236 		pci_conf_write(pc, tag, off + PCI_PMCSR, pmcsr);
   1237 		aprint_verbose_dev(dv, "Clear PME# now\n");
   1238 	}
   1239 }
   1240 
   1241 static bool
   1242 pci_child_resume(device_t dv, const pmf_qual_t *qual)
   1243 {
   1244 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1245 
   1246 	if (priv->p_has_pm) {
   1247 		if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1248 		    PCI_PMCSR_STATE_D0, priv->p_pm_offset, priv->p_pm_cap)) {
   1249 			aprint_error_dev(dv,
   1250 			    "unsupported state, continuing.\n");
   1251 			return false;
   1252 		}
   1253 		pci_pme_check_and_clear(dv, priv->p_pc, priv->p_tag,
   1254 		    priv->p_pm_offset);
   1255 	}
   1256 
   1257 	pci_conf_restore(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1258 
   1259 	return true;
   1260 }
   1261 
   1262 static bool
   1263 pci_child_shutdown(device_t dv, int how)
   1264 {
   1265 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1266 	pcireg_t csr;
   1267 
   1268 	/* restore original bus-mastering state */
   1269 	csr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1270 	csr &= ~PCI_COMMAND_MASTER_ENABLE;
   1271 	csr |= priv->p_csr & PCI_COMMAND_MASTER_ENABLE;
   1272 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1273 	return true;
   1274 }
   1275 
   1276 static void
   1277 pci_child_deregister(device_t dv)
   1278 {
   1279 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1280 
   1281 	free(priv, M_DEVBUF);
   1282 }
   1283 
   1284 static bool
   1285 pci_child_register(device_t child)
   1286 {
   1287 	device_t self = device_parent(child);
   1288 	struct pci_softc *sc = device_private(self);
   1289 	struct pci_child_power *priv;
   1290 	int device, function, off;
   1291 	pcireg_t reg;
   1292 
   1293 	priv = malloc(sizeof(*priv), M_DEVBUF, M_WAITOK);
   1294 
   1295 	device = device_locator(child, PCICF_DEV);
   1296 	function = device_locator(child, PCICF_FUNCTION);
   1297 
   1298 	priv->p_pc = sc->sc_pc;
   1299 	priv->p_tag = pci_make_tag(priv->p_pc, sc->sc_bus, device,
   1300 	    function);
   1301 	priv->p_class = pci_conf_read(priv->p_pc, priv->p_tag, PCI_CLASS_REG);
   1302 	priv->p_csr = pci_conf_read(priv->p_pc, priv->p_tag,
   1303 	    PCI_COMMAND_STATUS_REG);
   1304 
   1305 	if (pci_get_capability(priv->p_pc, priv->p_tag,
   1306 			       PCI_CAP_PWRMGMT, &off, &reg)) {
   1307 		priv->p_has_pm = true;
   1308 		priv->p_pm_offset = off;
   1309 		priv->p_pm_cap = reg;
   1310 		pci_pme_check_and_clear(child, priv->p_pc, priv->p_tag, off);
   1311 	} else {
   1312 		priv->p_has_pm = false;
   1313 		priv->p_pm_offset = -1;
   1314 	}
   1315 
   1316 	device_pmf_bus_register(child, priv, pci_child_suspend,
   1317 	    pci_child_resume, pci_child_shutdown, pci_child_deregister);
   1318 
   1319 	return true;
   1320 }
   1321 
   1322 MODULE(MODULE_CLASS_DRIVER, pci, NULL);
   1323 
   1324 static int
   1325 pci_modcmd(modcmd_t cmd, void *priv)
   1326 {
   1327 	if (cmd == MODULE_CMD_INIT || cmd == MODULE_CMD_FINI)
   1328 		return 0;
   1329 	return ENOTTY;
   1330 }
   1331