Home | History | Annotate | Line # | Download | only in pci
pci.c revision 1.156
      1 /*	$NetBSD: pci.c,v 1.156 2019/10/15 13:27:11 jmcneill 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 1.156 2019/10/15 13:27:11 jmcneill 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 	bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    292 	if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    293 		return 0;
    294 
    295 	id = pci_conf_read(pc, tag, PCI_ID_REG);
    296 	/* csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG); */
    297 	pciclass = pci_conf_read(pc, tag, PCI_CLASS_REG);
    298 
    299 	/* Invalid vendor ID value? */
    300 	if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    301 		return 0;
    302 	/* XXX Not invalid, but we've done this ~forever. */
    303 	if (PCI_VENDOR(id) == 0)
    304 		return 0;
    305 
    306 	/* Collect memory range info */
    307 	memset(sc->PCI_SC_DEVICESC(device, function).c_range, 0,
    308 	    sizeof(sc->PCI_SC_DEVICESC(device, function).c_range));
    309 	i = 0;
    310 	switch (PCI_HDRTYPE_TYPE(bhlcr)) {
    311 	case PCI_HDRTYPE_PPB:
    312 		endbar = PCI_MAPREG_PPB_END;
    313 		break;
    314 	case PCI_HDRTYPE_PCB:
    315 		endbar = PCI_MAPREG_PCB_END;
    316 		break;
    317 	default:
    318 		endbar = PCI_MAPREG_END;
    319 		break;
    320 	}
    321 	for (bar = PCI_MAPREG_START; bar < endbar; bar += width) {
    322 		struct pci_range *r;
    323 		pcireg_t type;
    324 
    325 		width = 4;
    326 		if (pci_mapreg_probe(pc, tag, bar, &type) == 0)
    327 			continue;
    328 
    329 		if (PCI_MAPREG_TYPE(type) == PCI_MAPREG_TYPE_MEM) {
    330 			if (PCI_MAPREG_MEM_TYPE(type) ==
    331 			    PCI_MAPREG_MEM_TYPE_64BIT)
    332 				width = 8;
    333 
    334 			r = &sc->PCI_SC_DEVICESC(device, function).c_range[i++];
    335 			if (pci_mapreg_info(pc, tag, bar, type,
    336 			    &r->r_offset, &r->r_size, &r->r_flags) != 0)
    337 				break;
    338 			if ((PCI_VENDOR(id) == PCI_VENDOR_ATI) && (bar == 0x10)
    339 			    && (r->r_size == 0x1000000)) {
    340 				struct pci_range *nr;
    341 				/*
    342 				 * this has to be a mach64
    343 				 * split things up so each half-aperture can
    344 				 * be mapped PREFETCHABLE except the last page
    345 				 * which may contain registers
    346 				 */
    347 				r->r_size = 0x7ff000;
    348 				r->r_flags = BUS_SPACE_MAP_LINEAR |
    349 					     BUS_SPACE_MAP_PREFETCHABLE;
    350 				nr = &sc->PCI_SC_DEVICESC(device,
    351 				    function).c_range[i++];
    352 				nr->r_offset = r->r_offset + 0x800000;
    353 				nr->r_size = 0x7ff000;
    354 				nr->r_flags = BUS_SPACE_MAP_LINEAR |
    355 					      BUS_SPACE_MAP_PREFETCHABLE;
    356 			} else if ((PCI_VENDOR(id) == PCI_VENDOR_SILMOTION) &&
    357 			   (PCI_PRODUCT(id) == PCI_PRODUCT_SILMOTION_SM502) &&
    358 			   (bar == 0x10)) {
    359 			   	r->r_flags = BUS_SPACE_MAP_LINEAR |
    360 					     BUS_SPACE_MAP_PREFETCHABLE;
    361 			}
    362 		}
    363 	}
    364 
    365 	pa.pa_iot = sc->sc_iot;
    366 	pa.pa_memt = sc->sc_memt;
    367 	pa.pa_dmat = sc->sc_dmat;
    368 	pa.pa_dmat64 = sc->sc_dmat64;
    369 	pa.pa_pc = pc;
    370 	pa.pa_bus = bus;
    371 	pa.pa_device = device;
    372 	pa.pa_function = function;
    373 	pa.pa_tag = tag;
    374 	pa.pa_id = id;
    375 	pa.pa_class = pciclass;
    376 
    377 	/*
    378 	 * Set up memory, I/O enable, and PCI command flags
    379 	 * as appropriate.
    380 	 */
    381 	pa.pa_flags = sc->sc_flags;
    382 
    383 	/*
    384 	 * If the cache line size is not configured, then
    385 	 * clear the MRL/MRM/MWI command-ok flags.
    386 	 */
    387 	if (PCI_CACHELINE(bhlcr) == 0) {
    388 		pa.pa_flags &= ~(PCI_FLAGS_MRL_OKAY|
    389 		    PCI_FLAGS_MRM_OKAY|PCI_FLAGS_MWI_OKAY);
    390 	}
    391 
    392 	if (sc->sc_bridgetag == NULL) {
    393 		pa.pa_intrswiz = 0;
    394 		pa.pa_intrtag = tag;
    395 	} else {
    396 		pa.pa_intrswiz = sc->sc_intrswiz + device;
    397 		pa.pa_intrtag = sc->sc_intrtag;
    398 	}
    399 
    400 	intr = pci_conf_read(pc, tag, PCI_INTERRUPT_REG);
    401 
    402 	pin = PCI_INTERRUPT_PIN(intr);
    403 	pa.pa_rawintrpin = pin;
    404 	if (pin == PCI_INTERRUPT_PIN_NONE) {
    405 		/* no interrupt */
    406 		pa.pa_intrpin = 0;
    407 	} else {
    408 		/*
    409 		 * swizzle it based on the number of busses we're
    410 		 * behind and our device number.
    411 		 */
    412 		pa.pa_intrpin = 	/* XXX */
    413 		    ((pin + pa.pa_intrswiz - 1) % 4) + 1;
    414 	}
    415 	pa.pa_intrline = PCI_INTERRUPT_LINE(intr);
    416 
    417 #ifdef __HAVE_PCI_MSI_MSIX
    418 	if (pci_get_ht_capability(pc, tag, PCI_HT_CAP_MSIMAP, &off, &cap)) {
    419 		/*
    420 		 * XXX Should we enable MSI mapping ourselves on
    421 		 * systems that have it disabled?
    422 		 */
    423 		if (cap & PCI_HT_MSI_ENABLED) {
    424 			uint64_t addr;
    425 			if ((cap & PCI_HT_MSI_FIXED) == 0) {
    426 				addr = pci_conf_read(pc, tag,
    427 				    off + PCI_HT_MSI_ADDR_LO);
    428 				addr |= (uint64_t)pci_conf_read(pc, tag,
    429 				    off + PCI_HT_MSI_ADDR_HI) << 32;
    430 			} else
    431 				addr = PCI_HT_MSI_FIXED_ADDR;
    432 
    433 			/*
    434 			 * XXX This will fail to enable MSI on systems
    435 			 * that don't use the canonical address.
    436 			 */
    437 			if (addr == PCI_HT_MSI_FIXED_ADDR) {
    438 				pa.pa_flags |= PCI_FLAGS_MSI_OKAY;
    439 				pa.pa_flags |= PCI_FLAGS_MSIX_OKAY;
    440 			} else
    441 				aprint_verbose_dev(sc->sc_dev,
    442 				    "HyperTransport MSI mapping is not supported yet. Disable MSI/MSI-X.\n");
    443 		}
    444 	}
    445 #endif
    446 
    447 	if (match != NULL) {
    448 		ret = (*match)(&pa);
    449 		if (ret != 0 && pap != NULL)
    450 			*pap = pa;
    451 	} else {
    452 		struct pci_child *c;
    453 		locs[PCICF_DEV] = device;
    454 		locs[PCICF_FUNCTION] = function;
    455 
    456 		c = &sc->PCI_SC_DEVICESC(device, function);
    457 		pci_conf_capture(pc, tag, &c->c_conf);
    458 		if (pci_get_powerstate(pc, tag, &c->c_powerstate) == 0)
    459 			c->c_psok = true;
    460 		else
    461 			c->c_psok = false;
    462 
    463 		c->c_dev = config_found_sm_loc(sc->sc_dev, "pci", locs, &pa,
    464 					     pciprint, config_stdsubmatch);
    465 
    466 		ret = (c->c_dev != NULL);
    467 	}
    468 
    469 	return ret;
    470 }
    471 
    472 void
    473 pcidevdetached(device_t self, device_t child)
    474 {
    475 	struct pci_softc *sc = device_private(self);
    476 	int d, f;
    477 	pcitag_t tag;
    478 	struct pci_child *c;
    479 
    480 	d = device_locator(child, PCICF_DEV);
    481 	f = device_locator(child, PCICF_FUNCTION);
    482 
    483 	c = &sc->PCI_SC_DEVICESC(d, f);
    484 
    485 	KASSERT(c->c_dev == child);
    486 
    487 	tag = pci_make_tag(sc->sc_pc, sc->sc_bus, d, f);
    488 	if (c->c_psok)
    489 		pci_set_powerstate(sc->sc_pc, tag, c->c_powerstate);
    490 	pci_conf_restore(sc->sc_pc, tag, &c->c_conf);
    491 	c->c_dev = NULL;
    492 }
    493 
    494 CFATTACH_DECL3_NEW(pci, sizeof(struct pci_softc),
    495     pcimatch, pciattach, pcidetach, NULL, pcirescan, pcidevdetached,
    496     DVF_DETACH_SHUTDOWN);
    497 
    498 int
    499 pci_get_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    500     int *offset, pcireg_t *value)
    501 {
    502 	pcireg_t reg;
    503 	unsigned int ofs;
    504 
    505 	reg = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
    506 	if (!(reg & PCI_STATUS_CAPLIST_SUPPORT))
    507 		return 0;
    508 
    509 	/* Determine the Capability List Pointer register to start with. */
    510 	reg = pci_conf_read(pc, tag, PCI_BHLC_REG);
    511 	switch (PCI_HDRTYPE_TYPE(reg)) {
    512 	case 0:	/* standard device header */
    513 	case 1: /* PCI-PCI bridge header */
    514 		ofs = PCI_CAPLISTPTR_REG;
    515 		break;
    516 	case 2:	/* PCI-CardBus Bridge header */
    517 		ofs = PCI_CARDBUS_CAPLISTPTR_REG;
    518 		break;
    519 	default:
    520 		return 0;
    521 	}
    522 
    523 	ofs = PCI_CAPLIST_PTR(pci_conf_read(pc, tag, ofs));
    524 	while (ofs != 0) {
    525 		if ((ofs & 3) || (ofs < 0x40)) {
    526 			int bus, device, function;
    527 
    528 			pci_decompose_tag(pc, tag, &bus, &device, &function);
    529 
    530 			printf("Skipping broken PCI header on %d:%d:%d\n",
    531 			    bus, device, function);
    532 			break;
    533 		}
    534 		reg = pci_conf_read(pc, tag, ofs);
    535 		if (PCI_CAPLIST_CAP(reg) == capid) {
    536 			if (offset)
    537 				*offset = ofs;
    538 			if (value)
    539 				*value = reg;
    540 			return 1;
    541 		}
    542 		ofs = PCI_CAPLIST_NEXT(reg);
    543 	}
    544 
    545 	return 0;
    546 }
    547 
    548 int
    549 pci_get_ht_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    550     int *offset, pcireg_t *value)
    551 {
    552 	pcireg_t reg;
    553 	unsigned int ofs;
    554 
    555 	if (pci_get_capability(pc, tag, PCI_CAP_LDT, &ofs, NULL) == 0)
    556 		return 0;
    557 
    558 	while (ofs != 0) {
    559 #ifdef DIAGNOSTIC
    560 		if ((ofs & 3) || (ofs < 0x40))
    561 			panic("pci_get_ht_capability");
    562 #endif
    563 		reg = pci_conf_read(pc, tag, ofs);
    564 		if (PCI_HT_CAP(reg) == capid) {
    565 			if (offset)
    566 				*offset = ofs;
    567 			if (value)
    568 				*value = reg;
    569 			return 1;
    570 		}
    571 		ofs = PCI_CAPLIST_NEXT(reg);
    572 	}
    573 
    574 	return 0;
    575 }
    576 
    577 /*
    578  * return number of the devices's MSI vectors
    579  * return 0 if the device does not support MSI
    580  */
    581 int
    582 pci_msi_count(pci_chipset_tag_t pc, pcitag_t tag)
    583 {
    584 	pcireg_t reg;
    585 	uint32_t mmc;
    586 	int count, offset;
    587 
    588 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &offset, NULL) == 0)
    589 		return 0;
    590 
    591 	reg = pci_conf_read(pc, tag, offset + PCI_MSI_CTL);
    592 	mmc = PCI_MSI_CTL_MMC(reg);
    593 	count = 1 << mmc;
    594 	if (count > PCI_MSI_MAX_VECTORS) {
    595 		aprint_error("detect an illegal device! The device use reserved MMC values.\n");
    596 		return 0;
    597 	}
    598 
    599 	return count;
    600 }
    601 
    602 /*
    603  * return number of the devices's MSI-X vectors
    604  * return 0 if the device does not support MSI-X
    605  */
    606 int
    607 pci_msix_count(pci_chipset_tag_t pc, pcitag_t tag)
    608 {
    609 	pcireg_t reg;
    610 	int offset;
    611 
    612 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &offset, NULL) == 0)
    613 		return 0;
    614 
    615 	reg = pci_conf_read(pc, tag, offset + PCI_MSIX_CTL);
    616 
    617 	return PCI_MSIX_CTL_TBLSIZE(reg);
    618 }
    619 
    620 int
    621 pci_get_ext_capability(pci_chipset_tag_t pc, pcitag_t tag, int capid,
    622     int *offset, pcireg_t *value)
    623 {
    624 	pcireg_t reg;
    625 	unsigned int ofs;
    626 
    627 	/* Only supported for PCI-express devices */
    628 	if (!pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, NULL, NULL))
    629 		return 0;
    630 
    631 	ofs = PCI_EXTCAPLIST_BASE;
    632 	reg = pci_conf_read(pc, tag, ofs);
    633 	if (reg == 0xffffffff || reg == 0)
    634 		return 0;
    635 
    636 	for (;;) {
    637 #ifdef DIAGNOSTIC
    638 		if ((ofs & 3) || ofs < PCI_EXTCAPLIST_BASE)
    639 			panic("%s: invalid offset %u", __func__, ofs);
    640 #endif
    641 		if (PCI_EXTCAPLIST_CAP(reg) == capid) {
    642 			if (offset != NULL)
    643 				*offset = ofs;
    644 			if (value != NULL)
    645 				*value = reg;
    646 			return 1;
    647 		}
    648 		ofs = PCI_EXTCAPLIST_NEXT(reg);
    649 		if (ofs == 0)
    650 			break;
    651 		reg = pci_conf_read(pc, tag, ofs);
    652 	}
    653 
    654 	return 0;
    655 }
    656 
    657 int
    658 pci_find_device(struct pci_attach_args *pa,
    659 		int (*match)(const struct pci_attach_args *))
    660 {
    661 	extern struct cfdriver pci_cd;
    662 	device_t pcidev;
    663 	int i;
    664 	static const int wildcard[2] = {
    665 		PCICF_DEV_DEFAULT,
    666 		PCICF_FUNCTION_DEFAULT
    667 	};
    668 
    669 	for (i = 0; i < pci_cd.cd_ndevs; i++) {
    670 		pcidev = device_lookup(&pci_cd, i);
    671 		if (pcidev != NULL &&
    672 		    pci_enumerate_bus(device_private(pcidev), wildcard,
    673 		    		      match, pa) != 0)
    674 			return 1;
    675 	}
    676 	return 0;
    677 }
    678 
    679 #ifndef PCI_MACHDEP_ENUMERATE_BUS
    680 /*
    681  * Generic PCI bus enumeration routine.  Used unless machine-dependent
    682  * code needs to provide something else.
    683  */
    684 int
    685 pci_enumerate_bus(struct pci_softc *sc, const int *locators,
    686     int (*match)(const struct pci_attach_args *), struct pci_attach_args *pap)
    687 {
    688 	pci_chipset_tag_t pc = sc->sc_pc;
    689 	int device, function, nfunctions, ret;
    690 	const struct pci_quirkdata *qd;
    691 	pcireg_t id, bhlcr;
    692 	pcitag_t tag;
    693 	uint8_t devs[32];
    694 	int i, n;
    695 
    696 	device_t bridgedev;
    697 	bool arien = false;
    698 	bool downstream_port = false;
    699 
    700 	/* Check PCIe ARI and port type */
    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, capreg, reg;
    707 
    708 		if (pci_get_capability(ppbpc, ppbtag, PCI_CAP_PCIEXPRESS,
    709 		    &pciecap, &capreg) != 0) {
    710 			switch (PCIE_XCAP_TYPE(capreg)) {
    711 			case PCIE_XCAP_TYPE_ROOT:
    712 			case PCIE_XCAP_TYPE_DOWN:
    713 			case PCIE_XCAP_TYPE_PCI2PCIE:
    714 				downstream_port = true;
    715 				break;
    716 			}
    717 
    718 			reg = pci_conf_read(ppbpc, ppbtag, pciecap
    719 			    + PCIE_DCSR2);
    720 			if ((reg & PCIE_DCSR2_ARI_FWD) != 0)
    721 				arien = true;
    722 		}
    723 	}
    724 
    725 	n = pci_bus_devorder(sc->sc_pc, sc->sc_bus, devs, __arraycount(devs));
    726 	if (downstream_port) {
    727 		/* PCIe downstream ports only have a single child device */
    728 		n = 1;
    729 	}
    730 
    731 	for (i = 0; i < n; i++) {
    732 		device = devs[i];
    733 
    734 		if ((locators[PCICF_DEV] != PCICF_DEV_DEFAULT) &&
    735 		    (locators[PCICF_DEV] != device))
    736 			continue;
    737 
    738 		tag = pci_make_tag(pc, sc->sc_bus, device, 0);
    739 
    740 		id = pci_conf_read(pc, tag, PCI_ID_REG);
    741 
    742 		/* Invalid vendor ID value? */
    743 		if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    744 			continue;
    745 		/* XXX Not invalid, but we've done this ~forever. */
    746 		if (PCI_VENDOR(id) == 0)
    747 			continue;
    748 
    749 		bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
    750 		if (PCI_HDRTYPE_TYPE(bhlcr) > 2)
    751 			continue;
    752 
    753 		qd = pci_lookup_quirkdata(PCI_VENDOR(id), PCI_PRODUCT(id));
    754 
    755 		if (qd != NULL &&
    756 		      (qd->quirks & PCI_QUIRK_MULTIFUNCTION) != 0)
    757 			nfunctions = 8;
    758 		else if (qd != NULL &&
    759 		      (qd->quirks & PCI_QUIRK_MONOFUNCTION) != 0)
    760 			nfunctions = 1;
    761 		else if (arien)
    762 			nfunctions = 8; /* Scan all if ARI is enabled */
    763 		else
    764 			nfunctions = PCI_HDRTYPE_MULTIFN(bhlcr) ? 8 : 1;
    765 
    766 #ifdef __PCI_DEV_FUNCORDER
    767 		char funcs[8];
    768 		int j;
    769 		for (j = 0; j < nfunctions; j++) {
    770 			funcs[j] = j;
    771 		}
    772 		if (j < __arraycount(funcs))
    773 			funcs[j] = -1;
    774 		if (nfunctions > 1) {
    775 			pci_dev_funcorder(sc->sc_pc, sc->sc_bus, device,
    776 			    nfunctions, funcs);
    777 		}
    778 		for (j = 0;
    779 		     j < 8 && (function = funcs[j]) < 8 && function >= 0;
    780 		     j++) {
    781 #else
    782 		for (function = 0; function < nfunctions; function++) {
    783 #endif
    784 			if ((locators[PCICF_FUNCTION] != PCICF_FUNCTION_DEFAULT)
    785 			    && (locators[PCICF_FUNCTION] != function))
    786 				continue;
    787 
    788 			if (qd != NULL &&
    789 			    (qd->quirks & PCI_QUIRK_SKIP_FUNC(function)) != 0)
    790 				continue;
    791 			tag = pci_make_tag(pc, sc->sc_bus, device, function);
    792 			ret = pci_probe_device(sc, tag, match, pap);
    793 			if (match != NULL && ret != 0)
    794 				return ret;
    795 		}
    796 	}
    797 	return 0;
    798 }
    799 #endif /* PCI_MACHDEP_ENUMERATE_BUS */
    800 
    801 
    802 /*
    803  * Vital Product Data (PCI 2.2)
    804  */
    805 
    806 int
    807 pci_vpd_read(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    808     pcireg_t *data)
    809 {
    810 	uint32_t reg;
    811 	int ofs, i, j;
    812 
    813 	KASSERT(data != NULL);
    814 	KASSERT((offset + count) < 0x7fff);
    815 
    816 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    817 		return 1;
    818 
    819 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    820 		reg &= 0x0000ffff;
    821 		reg &= ~PCI_VPD_OPFLAG;
    822 		reg |= PCI_VPD_ADDRESS(offset);
    823 		pci_conf_write(pc, tag, ofs, reg);
    824 
    825 		/*
    826 		 * PCI 2.2 does not specify how long we should poll
    827 		 * for completion nor whether the operation can fail.
    828 		 */
    829 		j = 0;
    830 		do {
    831 			if (j++ == 20)
    832 				return 1;
    833 			delay(4);
    834 			reg = pci_conf_read(pc, tag, ofs);
    835 		} while ((reg & PCI_VPD_OPFLAG) == 0);
    836 		data[i] = pci_conf_read(pc, tag, PCI_VPD_DATAREG(ofs));
    837 	}
    838 
    839 	return 0;
    840 }
    841 
    842 int
    843 pci_vpd_write(pci_chipset_tag_t pc, pcitag_t tag, int offset, int count,
    844     pcireg_t *data)
    845 {
    846 	pcireg_t reg;
    847 	int ofs, i, j;
    848 
    849 	KASSERT(data != NULL);
    850 	KASSERT((offset + count) < 0x7fff);
    851 
    852 	if (pci_get_capability(pc, tag, PCI_CAP_VPD, &ofs, &reg) == 0)
    853 		return 1;
    854 
    855 	for (i = 0; i < count; offset += sizeof(*data), i++) {
    856 		pci_conf_write(pc, tag, PCI_VPD_DATAREG(ofs), data[i]);
    857 
    858 		reg &= 0x0000ffff;
    859 		reg |= PCI_VPD_OPFLAG;
    860 		reg |= PCI_VPD_ADDRESS(offset);
    861 		pci_conf_write(pc, tag, ofs, reg);
    862 
    863 		/*
    864 		 * PCI 2.2 does not specify how long we should poll
    865 		 * for completion nor whether the operation can fail.
    866 		 */
    867 		j = 0;
    868 		do {
    869 			if (j++ == 20)
    870 				return 1;
    871 			delay(1);
    872 			reg = pci_conf_read(pc, tag, ofs);
    873 		} while (reg & PCI_VPD_OPFLAG);
    874 	}
    875 
    876 	return 0;
    877 }
    878 
    879 int
    880 pci_dma64_available(const struct pci_attach_args *pa)
    881 {
    882 #ifdef _PCI_HAVE_DMA64
    883 	if (BUS_DMA_TAG_VALID(pa->pa_dmat64))
    884                         return 1;
    885 #endif
    886         return 0;
    887 }
    888 
    889 void
    890 pci_conf_capture(pci_chipset_tag_t pc, pcitag_t tag,
    891 		  struct pci_conf_state *pcs)
    892 {
    893 	int off;
    894 
    895 	for (off = 0; off < 16; off++)
    896 		pcs->reg[off] = pci_conf_read(pc, tag, (off * 4));
    897 
    898 	/* For PCI-X */
    899 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
    900 		pcs->x_csr = pci_conf_read(pc, tag, off + PCIX_CMD);
    901 
    902 	/* For PCIe */
    903 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
    904 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
    905 		unsigned int devtype;
    906 
    907 		devtype = PCIE_XCAP_TYPE(xcap);
    908 		pcs->e_dcr = (uint16_t)pci_conf_read(pc, tag, off + PCIE_DCSR);
    909 
    910 		if (PCIE_HAS_LINKREGS(devtype))
    911 			pcs->e_lcr = (uint16_t)pci_conf_read(pc, tag,
    912 			    off + PCIE_LCSR);
    913 
    914 		if ((xcap & PCIE_XCAP_SI) != 0)
    915 			pcs->e_slcr = (uint16_t)pci_conf_read(pc, tag,
    916 			    off + PCIE_SLCSR);
    917 
    918 		if (PCIE_HAS_ROOTREGS(devtype))
    919 			pcs->e_rcr = (uint16_t)pci_conf_read(pc, tag,
    920 			    off + PCIE_RCR);
    921 
    922 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
    923 			pcs->e_dcr2 = (uint16_t)pci_conf_read(pc, tag,
    924 			    off + PCIE_DCSR2);
    925 
    926 			if (PCIE_HAS_LINKREGS(devtype))
    927 				pcs->e_lcr2 = (uint16_t)pci_conf_read(pc, tag,
    928 			    off + PCIE_LCSR2);
    929 
    930 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
    931 		}
    932 	}
    933 
    934 	/* For MSI */
    935 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
    936 		bool bit64, pvmask;
    937 
    938 		pcs->msi_ctl = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
    939 
    940 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
    941 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
    942 
    943 		/* Address */
    944 		pcs->msi_maddr = pci_conf_read(pc, tag, off + PCI_MSI_MADDR);
    945 		if (bit64)
    946 			pcs->msi_maddr64_hi = pci_conf_read(pc, tag,
    947 			    off + PCI_MSI_MADDR64_HI);
    948 
    949 		/* Data */
    950 		pcs->msi_mdata = pci_conf_read(pc, tag,
    951 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA));
    952 
    953 		/* Per-vector masking */
    954 		if (pvmask)
    955 			pcs->msi_mask = pci_conf_read(pc, tag,
    956 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK));
    957 	}
    958 
    959 	/* For MSI-X */
    960 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
    961 		pcs->msix_ctl = pci_conf_read(pc, tag, off + PCI_MSIX_CTL);
    962 }
    963 
    964 void
    965 pci_conf_restore(pci_chipset_tag_t pc, pcitag_t tag,
    966 		  struct pci_conf_state *pcs)
    967 {
    968 	int off;
    969 	pcireg_t val;
    970 
    971 	for (off = 15; off >= 0; off--) {
    972 		val = pci_conf_read(pc, tag, (off * 4));
    973 		if (val != pcs->reg[off])
    974 			pci_conf_write(pc, tag, (off * 4), pcs->reg[off]);
    975 	}
    976 
    977 	/* For PCI-X */
    978 	if (pci_get_capability(pc, tag, PCI_CAP_PCIX, &off, NULL) != 0)
    979 		pci_conf_write(pc, tag, off + PCIX_CMD, pcs->x_csr);
    980 
    981 	/* For PCIe */
    982 	if (pci_get_capability(pc, tag, PCI_CAP_PCIEXPRESS, &off, NULL) != 0) {
    983 		pcireg_t xcap = pci_conf_read(pc, tag, off + PCIE_XCAP);
    984 		unsigned int devtype;
    985 
    986 		devtype = PCIE_XCAP_TYPE(xcap);
    987 		pci_conf_write(pc, tag, off + PCIE_DCSR, pcs->e_dcr);
    988 
    989 		/*
    990 		 * PCIe capability is variable sized. To not to write the next
    991 		 * area, check the existence of each register.
    992 		 */
    993 		if (PCIE_HAS_LINKREGS(devtype))
    994 			pci_conf_write(pc, tag, off + PCIE_LCSR, pcs->e_lcr);
    995 
    996 		if ((xcap & PCIE_XCAP_SI) != 0)
    997 			pci_conf_write(pc, tag, off + PCIE_SLCSR, pcs->e_slcr);
    998 
    999 		if (PCIE_HAS_ROOTREGS(devtype))
   1000 			pci_conf_write(pc, tag, off + PCIE_RCR, pcs->e_rcr);
   1001 
   1002 		if (__SHIFTOUT(xcap, PCIE_XCAP_VER_MASK) >= 2) {
   1003 			pci_conf_write(pc, tag, off + PCIE_DCSR2, pcs->e_dcr2);
   1004 
   1005 			if (PCIE_HAS_LINKREGS(devtype))
   1006 				pci_conf_write(pc, tag, off + PCIE_LCSR2,
   1007 				    pcs->e_lcr2);
   1008 
   1009 			/* XXX PCIE_SLCSR2 (It's reserved by the PCIe spec) */
   1010 		}
   1011 	}
   1012 
   1013 	/* For MSI */
   1014 	if (pci_get_capability(pc, tag, PCI_CAP_MSI, &off, NULL) != 0) {
   1015 		pcireg_t reg;
   1016 		bool bit64, pvmask;
   1017 
   1018 		/* First, drop Enable bit in case it's already set. */
   1019 		reg = pci_conf_read(pc, tag, off + PCI_MSI_CTL);
   1020 		pci_conf_write(pc, tag, off + PCI_MSI_CTL,
   1021 		    reg & ~PCI_MSI_CTL_MSI_ENABLE);
   1022 
   1023 		bit64 = pcs->msi_ctl & PCI_MSI_CTL_64BIT_ADDR;
   1024 		pvmask = pcs->msi_ctl & PCI_MSI_CTL_PERVEC_MASK;
   1025 
   1026 		/* Address */
   1027 		pci_conf_write(pc, tag, off + PCI_MSI_MADDR, pcs->msi_maddr);
   1028 
   1029 		if (bit64)
   1030 			pci_conf_write(pc, tag,
   1031 			    off + PCI_MSI_MADDR64_HI, pcs->msi_maddr64_hi);
   1032 
   1033 		/* Data */
   1034 		pci_conf_write(pc, tag,
   1035 		    off + (bit64 ? PCI_MSI_MDATA64 : PCI_MSI_MDATA),
   1036 		    pcs->msi_mdata);
   1037 
   1038 		/* Per-vector masking */
   1039 		if (pvmask)
   1040 			pci_conf_write(pc, tag,
   1041 			    off + (bit64 ? PCI_MSI_MASK64 : PCI_MSI_MASK),
   1042 			    pcs->msi_mask);
   1043 
   1044 		/* Write CTRL register in the end */
   1045 		pci_conf_write(pc, tag, off + PCI_MSI_CTL, pcs->msi_ctl);
   1046 	}
   1047 
   1048 	/* For MSI-X */
   1049 	if (pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL) != 0)
   1050 		pci_conf_write(pc, tag, off + PCI_MSIX_CTL, pcs->msix_ctl);
   1051 }
   1052 
   1053 /*
   1054  * Power Management Capability (Rev 2.2)
   1055  */
   1056 static int
   1057 pci_get_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state,
   1058     int offset)
   1059 {
   1060 	pcireg_t value, now;
   1061 
   1062 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1063 	now = value & PCI_PMCSR_STATE_MASK;
   1064 	switch (now) {
   1065 	case PCI_PMCSR_STATE_D0:
   1066 	case PCI_PMCSR_STATE_D1:
   1067 	case PCI_PMCSR_STATE_D2:
   1068 	case PCI_PMCSR_STATE_D3:
   1069 		*state = now;
   1070 		return 0;
   1071 	default:
   1072 		return EINVAL;
   1073 	}
   1074 }
   1075 
   1076 int
   1077 pci_get_powerstate(pci_chipset_tag_t pc, pcitag_t tag , pcireg_t *state)
   1078 {
   1079 	int offset;
   1080 	pcireg_t value;
   1081 
   1082 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value))
   1083 		return EOPNOTSUPP;
   1084 
   1085 	return pci_get_powerstate_int(pc, tag, state, offset);
   1086 }
   1087 
   1088 static int
   1089 pci_set_powerstate_int(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state,
   1090     int offset, pcireg_t cap_reg)
   1091 {
   1092 	pcireg_t value, cap, now;
   1093 
   1094 	cap = cap_reg >> PCI_PMCR_SHIFT;
   1095 	value = pci_conf_read(pc, tag, offset + PCI_PMCSR);
   1096 	now = value & PCI_PMCSR_STATE_MASK;
   1097 	value &= ~PCI_PMCSR_STATE_MASK;
   1098 
   1099 	if (now == state)
   1100 		return 0;
   1101 	switch (state) {
   1102 	case PCI_PMCSR_STATE_D0:
   1103 		break;
   1104 	case PCI_PMCSR_STATE_D1:
   1105 		if (now == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D3) {
   1106 			printf("invalid transition from %d to D1\n", (int)now);
   1107 			return EINVAL;
   1108 		}
   1109 		if (!(cap & PCI_PMCR_D1SUPP)) {
   1110 			printf("D1 not supported\n");
   1111 			return EOPNOTSUPP;
   1112 		}
   1113 		break;
   1114 	case PCI_PMCSR_STATE_D2:
   1115 		if (now == PCI_PMCSR_STATE_D3) {
   1116 			printf("invalid transition from %d to D2\n", (int)now);
   1117 			return EINVAL;
   1118 		}
   1119 		if (!(cap & PCI_PMCR_D2SUPP)) {
   1120 			printf("D2 not supported\n");
   1121 			return EOPNOTSUPP;
   1122 		}
   1123 		break;
   1124 	case PCI_PMCSR_STATE_D3:
   1125 		break;
   1126 	default:
   1127 		return EINVAL;
   1128 	}
   1129 	value |= state;
   1130 	pci_conf_write(pc, tag, offset + PCI_PMCSR, value);
   1131 	/* delay according to pcipm1.2, ch. 5.6.1 */
   1132 	if (state == PCI_PMCSR_STATE_D3 || now == PCI_PMCSR_STATE_D3)
   1133 		DELAY(10000);
   1134 	else if (state == PCI_PMCSR_STATE_D2 || now == PCI_PMCSR_STATE_D2)
   1135 		DELAY(200);
   1136 
   1137 	return 0;
   1138 }
   1139 
   1140 int
   1141 pci_set_powerstate(pci_chipset_tag_t pc, pcitag_t tag, pcireg_t state)
   1142 {
   1143 	int offset;
   1144 	pcireg_t value;
   1145 
   1146 	if (!pci_get_capability(pc, tag, PCI_CAP_PWRMGMT, &offset, &value)) {
   1147 		printf("pci_set_powerstate not supported\n");
   1148 		return EOPNOTSUPP;
   1149 	}
   1150 
   1151 	return pci_set_powerstate_int(pc, tag, state, offset, value);
   1152 }
   1153 
   1154 int
   1155 pci_activate(pci_chipset_tag_t pc, pcitag_t tag, device_t dev,
   1156     int (*wakefun)(pci_chipset_tag_t, pcitag_t, device_t, pcireg_t))
   1157 {
   1158 	pcireg_t pmode;
   1159 	int error;
   1160 
   1161 	if ((error = pci_get_powerstate(pc, tag, &pmode)))
   1162 		return error;
   1163 
   1164 	switch (pmode) {
   1165 	case PCI_PMCSR_STATE_D0:
   1166 		break;
   1167 	case PCI_PMCSR_STATE_D3:
   1168 		if (wakefun == NULL) {
   1169 			/*
   1170 			 * The card has lost all configuration data in
   1171 			 * this state, so punt.
   1172 			 */
   1173 			aprint_error_dev(dev,
   1174 			    "unable to wake up from power state D3\n");
   1175 			return EOPNOTSUPP;
   1176 		}
   1177 		/*FALLTHROUGH*/
   1178 	default:
   1179 		if (wakefun) {
   1180 			error = (*wakefun)(pc, tag, dev, pmode);
   1181 			if (error)
   1182 				return error;
   1183 		}
   1184 		aprint_normal_dev(dev, "waking up from power state D%d\n",
   1185 		    pmode);
   1186 		if ((error = pci_set_powerstate(pc, tag, PCI_PMCSR_STATE_D0)))
   1187 			return error;
   1188 	}
   1189 	return 0;
   1190 }
   1191 
   1192 int
   1193 pci_activate_null(pci_chipset_tag_t pc, pcitag_t tag,
   1194     device_t dev, pcireg_t state)
   1195 {
   1196 	return 0;
   1197 }
   1198 
   1199 struct pci_child_power {
   1200 	struct pci_conf_state p_pciconf;
   1201 	pci_chipset_tag_t p_pc;
   1202 	pcitag_t p_tag;
   1203 	bool p_has_pm;
   1204 	int p_pm_offset;
   1205 	pcireg_t p_pm_cap;
   1206 	pcireg_t p_class;
   1207 	pcireg_t p_csr;
   1208 };
   1209 
   1210 static bool
   1211 pci_child_suspend(device_t dv, const pmf_qual_t *qual)
   1212 {
   1213 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1214 	pcireg_t ocsr, csr;
   1215 
   1216 	pci_conf_capture(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1217 
   1218 	if (!priv->p_has_pm)
   1219 		return true; /* ??? hopefully handled by ACPI */
   1220 	if (PCI_CLASS(priv->p_class) == PCI_CLASS_DISPLAY)
   1221 		return true; /* XXX */
   1222 
   1223 	/* disable decoding and busmastering, see pcipm1.2 ch. 8.2.1 */
   1224 	ocsr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1225 	csr = ocsr & ~(PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE
   1226 		       | PCI_COMMAND_MASTER_ENABLE);
   1227 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1228 	if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1229 	    PCI_PMCSR_STATE_D3, priv->p_pm_offset, priv->p_pm_cap)) {
   1230 		pci_conf_write(priv->p_pc, priv->p_tag,
   1231 			       PCI_COMMAND_STATUS_REG, ocsr);
   1232 		aprint_error_dev(dv, "unsupported state, continuing.\n");
   1233 		return false;
   1234 	}
   1235 	return true;
   1236 }
   1237 
   1238 static void
   1239 pci_pme_check_and_clear(device_t dv, pci_chipset_tag_t pc, pcitag_t tag,
   1240     int off)
   1241 {
   1242 	pcireg_t pmcsr;
   1243 
   1244 	pmcsr = pci_conf_read(pc, tag, off + PCI_PMCSR);
   1245 
   1246 	if (pmcsr & PCI_PMCSR_PME_STS) {
   1247 		/* Clear W1C bit */
   1248 		pmcsr |= PCI_PMCSR_PME_STS;
   1249 		pci_conf_write(pc, tag, off + PCI_PMCSR, pmcsr);
   1250 		aprint_verbose_dev(dv, "Clear PME# now\n");
   1251 	}
   1252 }
   1253 
   1254 static bool
   1255 pci_child_resume(device_t dv, const pmf_qual_t *qual)
   1256 {
   1257 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1258 
   1259 	if (priv->p_has_pm) {
   1260 		if (pci_set_powerstate_int(priv->p_pc, priv->p_tag,
   1261 		    PCI_PMCSR_STATE_D0, priv->p_pm_offset, priv->p_pm_cap)) {
   1262 			aprint_error_dev(dv,
   1263 			    "unsupported state, continuing.\n");
   1264 			return false;
   1265 		}
   1266 		pci_pme_check_and_clear(dv, priv->p_pc, priv->p_tag,
   1267 		    priv->p_pm_offset);
   1268 	}
   1269 
   1270 	pci_conf_restore(priv->p_pc, priv->p_tag, &priv->p_pciconf);
   1271 
   1272 	return true;
   1273 }
   1274 
   1275 static bool
   1276 pci_child_shutdown(device_t dv, int how)
   1277 {
   1278 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1279 	pcireg_t csr;
   1280 
   1281 	/* restore original bus-mastering state */
   1282 	csr = pci_conf_read(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG);
   1283 	csr &= ~PCI_COMMAND_MASTER_ENABLE;
   1284 	csr |= priv->p_csr & PCI_COMMAND_MASTER_ENABLE;
   1285 	pci_conf_write(priv->p_pc, priv->p_tag, PCI_COMMAND_STATUS_REG, csr);
   1286 	return true;
   1287 }
   1288 
   1289 static void
   1290 pci_child_deregister(device_t dv)
   1291 {
   1292 	struct pci_child_power *priv = device_pmf_bus_private(dv);
   1293 
   1294 	free(priv, M_DEVBUF);
   1295 }
   1296 
   1297 static bool
   1298 pci_child_register(device_t child)
   1299 {
   1300 	device_t self = device_parent(child);
   1301 	struct pci_softc *sc = device_private(self);
   1302 	struct pci_child_power *priv;
   1303 	int device, function, off;
   1304 	pcireg_t reg;
   1305 
   1306 	priv = malloc(sizeof(*priv), M_DEVBUF, M_WAITOK);
   1307 
   1308 	device = device_locator(child, PCICF_DEV);
   1309 	function = device_locator(child, PCICF_FUNCTION);
   1310 
   1311 	priv->p_pc = sc->sc_pc;
   1312 	priv->p_tag = pci_make_tag(priv->p_pc, sc->sc_bus, device,
   1313 	    function);
   1314 	priv->p_class = pci_conf_read(priv->p_pc, priv->p_tag, PCI_CLASS_REG);
   1315 	priv->p_csr = pci_conf_read(priv->p_pc, priv->p_tag,
   1316 	    PCI_COMMAND_STATUS_REG);
   1317 
   1318 	if (pci_get_capability(priv->p_pc, priv->p_tag,
   1319 			       PCI_CAP_PWRMGMT, &off, &reg)) {
   1320 		priv->p_has_pm = true;
   1321 		priv->p_pm_offset = off;
   1322 		priv->p_pm_cap = reg;
   1323 		pci_pme_check_and_clear(child, priv->p_pc, priv->p_tag, off);
   1324 	} else {
   1325 		priv->p_has_pm = false;
   1326 		priv->p_pm_offset = -1;
   1327 	}
   1328 
   1329 	device_pmf_bus_register(child, priv, pci_child_suspend,
   1330 	    pci_child_resume, pci_child_shutdown, pci_child_deregister);
   1331 
   1332 	return true;
   1333 }
   1334 
   1335 MODULE(MODULE_CLASS_DRIVER, pci, NULL);
   1336 
   1337 static int
   1338 pci_modcmd(modcmd_t cmd, void *priv)
   1339 {
   1340 	if (cmd == MODULE_CMD_INIT || cmd == MODULE_CMD_FINI)
   1341 		return 0;
   1342 	return ENOTTY;
   1343 }
   1344