Home | History | Annotate | Line # | Download | only in pci
      1 /*	$NetBSD: pciconf.c,v 1.64 2026/08/20 06:51:57 msaitoh Exp $	*/
      2 
      3 /*
      4  * Copyright 2001 Wasabi Systems, Inc.
      5  * All rights reserved.
      6  *
      7  * Written by Allen Briggs for Wasabi Systems, Inc.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *      This product includes software developed for the NetBSD Project by
     20  *      Wasabi Systems, Inc.
     21  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     22  *    or promote products derived from this software without specific prior
     23  *    written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     35  * POSSIBILITY OF SUCH DAMAGE.
     36  */
     37 /*
     38  * Derived in part from code from PMON/2000 (http://pmon.groupbsd.org/).
     39  */
     40 
     41 /*
     42  * To do:
     43  *    - Perform all data structure allocation dynamically, don't have
     44  *	statically-sized arrays ("oops, you lose because you have too
     45  *	many slots filled!")
     46  *    - Do this in 2 passes, with an MD hook to control the behavior:
     47  *		(1) Configure the bus (possibly including expansion
     48  *		    ROMs.
     49  *		(2) Another pass to disable expansion ROMs if they're
     50  *		    mapped (since you're not supposed to leave them
     51  *		    mapped when you're not using them).
     52  *	This would facilitate MD code executing the expansion ROMs
     53  *	if necessary (possibly with an x86 emulator) to configure
     54  *	devices (e.g. VGA cards).
     55  *    - Deal with "anything can be hot-plugged" -- i.e., carry configuration
     56  *	information around & be able to reconfigure on the fly
     57  *    - Deal with segments (See IA64 System Abstraction Layer)
     58  *    - Deal with subtractive bridges (& non-spec positive/subtractive decode)
     59  *    - Deal with ISA/VGA/VGA palette snooping
     60  *    - Deal with device capabilities on bridges
     61  *    - Worry about changing a bridge to/from transparency
     62  * From thorpej (05/25/01)
     63  *    - Try to handle devices that are already configured (perhaps using that
     64  *      as a hint to where we put other devices)
     65  */
     66 
     67 #include <sys/cdefs.h>
     68 __KERNEL_RCSID(0, "$NetBSD: pciconf.c,v 1.64 2026/08/20 06:51:57 msaitoh Exp $");
     69 
     70 #include "opt_pci.h"
     71 
     72 #include <sys/param.h>
     73 #include <sys/queue.h>
     74 #include <sys/systm.h>
     75 #include <sys/kmem.h>
     76 #include <sys/vmem.h>
     77 
     78 #include <dev/pci/pcivar.h>
     79 #include <dev/pci/pciconf.h>
     80 #include <dev/pci/pcidevs.h>
     81 #include <dev/pci/pccbbreg.h>
     82 
     83 int pci_conf_debug = 0;
     84 
     85 #if !defined(MIN)
     86 #define	MIN(a,b) (((a)<(b))?(a):(b))
     87 #define	MAX(a,b) (((a)>(b))?(a):(b))
     88 #endif
     89 
     90 /* per-bus constants. */
     91 #define MAX_CONF_DEV	32			/* Arbitrary */
     92 #define MAX_CONF_MEM	(3 * MAX_CONF_DEV)	/* Avg. 3 per device -- Arb. */
     93 #define MAX_CONF_IO	(3 * MAX_CONF_DEV)	/* Avg. 1 per device -- Arb. */
     94 
     95 #define	PCICONF_MAPREG_IS_ROM(reg)					\
     96 	(((reg) == PCI_MAPREG_ROM) || ((reg) == PCI_BRIDGE_EXPROMADDR_REG))
     97 
     98 struct _s_pciconf_bus_t;			/* Forward declaration */
     99 
    100 struct pciconf_resource {
    101 	vmem_t		*arena;
    102 	bus_addr_t	min_addr;
    103 	bus_addr_t	max_addr;
    104 	bus_size_t	total_size;
    105 };
    106 
    107 #define	PCICONF_RESOURCE_NTYPES	3
    108 CTASSERT(PCICONF_RESOURCE_IO < PCICONF_RESOURCE_NTYPES);
    109 CTASSERT(PCICONF_RESOURCE_MEM < PCICONF_RESOURCE_NTYPES);
    110 CTASSERT(PCICONF_RESOURCE_PREFETCHABLE_MEM < PCICONF_RESOURCE_NTYPES);
    111 
    112 static const char *pciconf_resource_names[] = {
    113 	[PCICONF_RESOURCE_IO]			=	"pci-io",
    114 	[PCICONF_RESOURCE_MEM]			=	"pci-mem",
    115 	[PCICONF_RESOURCE_PREFETCHABLE_MEM]	=	"pci-pmem",
    116 };
    117 
    118 struct pciconf_resources {
    119 	struct pciconf_resource resources[PCICONF_RESOURCE_NTYPES];
    120 };
    121 
    122 struct pciconf_resource_rsvd {
    123 	int		type;
    124 	uint64_t	start;
    125 	bus_size_t	size;
    126 	void		(*callback)(void *, uint64_t);
    127 	void		*callback_arg;
    128 	LIST_ENTRY(pciconf_resource_rsvd) next;
    129 };
    130 
    131 static LIST_HEAD(, pciconf_resource_rsvd) pciconf_resource_reservations =
    132     LIST_HEAD_INITIALIZER(pciconf_resource_reservations);
    133 
    134 typedef struct _s_pciconf_dev_t {
    135 	int		ipin;
    136 	int		iline;
    137 	int		min_gnt;
    138 	int		max_lat;
    139 	int		enable;
    140 	pcitag_t	tag;
    141 	pci_chipset_tag_t	pc;
    142 	struct _s_pciconf_bus_t	*ppb;		/* I am really a bridge */
    143 	pcireg_t	ea_cap_ptr;
    144 } pciconf_dev_t;
    145 
    146 typedef struct _s_pciconf_win_t {
    147 	pciconf_dev_t	*dev;
    148 	int		reg;			/* 0 for busses */
    149 	int		prefetch;
    150 	uint64_t	size;
    151 	uint64_t	address;
    152 	uint64_t	align;
    153 } pciconf_win_t;
    154 
    155 typedef struct _s_pciconf_bus_t {
    156 	int		busno;
    157 	int		next_busno;
    158 	int		last_busno;
    159 	int		max_mingnt;
    160 	int		min_maxlat;
    161 	int		cacheline_size;
    162 	int		prefetch;
    163 	int		fast_b2b;
    164 	int		freq_66;
    165 	int		def_ltim;
    166 	int		max_ltim;
    167 	int		bandwidth_used;
    168 	int		swiz;
    169 	int		io_32bit;
    170 	int		pmem_64bit;
    171 	int		mem_64bit;
    172 
    173 	int		ndevs;
    174 	pciconf_dev_t	device[MAX_CONF_DEV];
    175 
    176 	/* These should be sorted in order of decreasing alignment */
    177 	int		nmemwin;
    178 	pciconf_win_t	pcimemwin[MAX_CONF_MEM];
    179 	int		niowin;
    180 	pciconf_win_t	pciiowin[MAX_CONF_IO];
    181 
    182 	bus_size_t	io_total;
    183 	bus_size_t	mem_total;
    184 	bus_size_t	pmem_total;
    185 	bus_size_t	io_align;
    186 	bus_size_t	mem_align;
    187 	bus_size_t	pmem_align;
    188 
    189 	struct pciconf_resource io_res;
    190 	struct pciconf_resource mem_res;
    191 	struct pciconf_resource pmem_res;
    192 
    193 	pci_chipset_tag_t	pc;
    194 	struct _s_pciconf_bus_t *parent_bus;
    195 } pciconf_bus_t;
    196 
    197 static int	probe_bus(pciconf_bus_t *);
    198 static void	alloc_busno(pciconf_bus_t *, pciconf_bus_t *);
    199 static void	set_busreg(pci_chipset_tag_t, pcitag_t, int, int, int);
    200 static int	pci_do_device_query(pciconf_bus_t *, pcitag_t, int, int, int);
    201 static int	setup_iowins(pciconf_bus_t *);
    202 static int	setup_memwins(pciconf_bus_t *);
    203 static int	configure_bridge(pciconf_dev_t *);
    204 static int	configure_bus(pciconf_bus_t *);
    205 static uint64_t	pci_allocate_range(struct pciconf_resource *, uint64_t, int,
    206 		    bool);
    207 static pciconf_win_t	*get_io_desc(pciconf_bus_t *, bus_size_t);
    208 static pciconf_win_t	*get_mem_desc(pciconf_bus_t *, bus_size_t);
    209 static pciconf_bus_t	*query_bus(pciconf_bus_t *, pciconf_dev_t *, int);
    210 
    211 static void	print_tag(pci_chipset_tag_t, pcitag_t);
    212 
    213 static vmem_t *
    214 create_vmem_arena(const char *name, bus_addr_t start, bus_size_t size,
    215     int flags)
    216 {
    217 	KASSERT(start < VMEM_ADDR_MAX);
    218 	KASSERT(size == 0 ||
    219 		(VMEM_ADDR_MAX - start) >= (size - 1));
    220 
    221 	return vmem_create(name, start, size,
    222 			   1,		/*quantum*/
    223 			   NULL,	/*importfn*/
    224 			   NULL,	/*releasefn*/
    225 			   NULL,	/*source*/
    226 			   0,		/*qcache_max*/
    227 			   flags,
    228 			   IPL_NONE);
    229 }
    230 
    231 static int
    232 init_range_resource(struct pciconf_resource *r, const char *name,
    233     bus_addr_t start, bus_addr_t size)
    234 {
    235 	r->arena = create_vmem_arena(name, start, size, VM_NOSLEEP);
    236 	if (r->arena == NULL)
    237 		return ENOMEM;
    238 
    239 	r->min_addr = start;
    240 	r->max_addr = start + (size - 1);
    241 	r->total_size = size;
    242 
    243 	return 0;
    244 }
    245 
    246 static void
    247 fini_range_resource(struct pciconf_resource *r)
    248 {
    249 	if (r->arena) {
    250 		vmem_xfreeall(r->arena);
    251 		vmem_destroy(r->arena);
    252 	}
    253 	memset(r, 0, sizeof(*r));
    254 }
    255 
    256 static void
    257 print_tag(pci_chipset_tag_t pc, pcitag_t tag)
    258 {
    259 	int	bus, dev, func;
    260 
    261 	pci_decompose_tag(pc, tag, &bus, &dev, &func);
    262 	printf("PCI: bus %d, device %d, function %d: ", bus, dev, func);
    263 }
    264 
    265 #ifdef _LP64
    266 #define	__used_only_lp64	__unused
    267 #else
    268 #define	__used_only_lp64	/* nothing */
    269 #endif /* _LP64 */
    270 
    271 /************************************************************************/
    272 /************************************************************************/
    273 /***********************   Bus probing routines   ***********************/
    274 /************************************************************************/
    275 /************************************************************************/
    276 static pciconf_win_t *
    277 get_io_desc(pciconf_bus_t *pb, bus_size_t align)
    278 {
    279 	int	i, n;
    280 
    281 	n = pb->niowin;
    282 	for (i = n; i > 0 && align > pb->pciiowin[i-1].align; i--)
    283 		pb->pciiowin[i] = pb->pciiowin[i-1]; /* struct copy */
    284 	return &pb->pciiowin[i];
    285 }
    286 
    287 static pciconf_win_t *
    288 get_mem_desc(pciconf_bus_t *pb, bus_size_t align)
    289 {
    290 	int	i, n;
    291 
    292 	n = pb->nmemwin;
    293 	for (i = n; i > 0 && align > pb->pcimemwin[i-1].align; i--)
    294 		pb->pcimemwin[i] = pb->pcimemwin[i-1]; /* struct copy */
    295 	return &pb->pcimemwin[i];
    296 }
    297 
    298 /*
    299  * Set up bus common stuff, then loop over devices & functions.
    300  * If we find something, call pci_do_device_query()).
    301  */
    302 static int
    303 probe_bus(pciconf_bus_t *pb)
    304 {
    305 	int device;
    306 	uint8_t devs[32];
    307 	int i, n;
    308 
    309 	pb->ndevs = 0;
    310 	pb->niowin = 0;
    311 	pb->nmemwin = 0;
    312 	pb->freq_66 = 1;
    313 #ifdef PCICONF_NO_FAST_B2B
    314 	pb->fast_b2b = 0;
    315 #else
    316 	pb->fast_b2b = 1;
    317 #endif
    318 	pb->prefetch = 1;
    319 	pb->max_mingnt = 0;	/* we are looking for the maximum */
    320 	pb->min_maxlat = 0x100;	/* we are looking for the minimum */
    321 	pb->bandwidth_used = 0;
    322 
    323 	n = pci_bus_devorder(pb->pc, pb->busno, devs, __arraycount(devs));
    324 	for (i = 0; i < n; i++) {
    325 		pcitag_t tag;
    326 		pcireg_t id, bhlcr;
    327 		int function, nfunction;
    328 		int confmode;
    329 
    330 		device = devs[i];
    331 
    332 		tag = pci_make_tag(pb->pc, pb->busno, device, 0);
    333 		if (pci_conf_debug) {
    334 			print_tag(pb->pc, tag);
    335 		}
    336 		id = pci_conf_read(pb->pc, tag, PCI_ID_REG);
    337 
    338 		if (pci_conf_debug) {
    339 			printf("id=%x: Vendor=%x, Product=%x\n",
    340 			    id, PCI_VENDOR(id), PCI_PRODUCT(id));
    341 		}
    342 		/* Invalid vendor ID value? */
    343 		if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    344 			continue;
    345 
    346 		bhlcr = pci_conf_read(pb->pc, tag, PCI_BHLC_REG);
    347 		nfunction = PCI_HDRTYPE_MULTIFN(bhlcr) ? 8 : 1;
    348 		for (function = 0; function < nfunction; function++) {
    349 			tag = pci_make_tag(pb->pc, pb->busno, device, function);
    350 			id = pci_conf_read(pb->pc, tag, PCI_ID_REG);
    351 			if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
    352 				continue;
    353 			if (pb->ndevs + 1 < MAX_CONF_DEV) {
    354 				if (pci_conf_debug) {
    355 					print_tag(pb->pc, tag);
    356 					printf("Found dev 0x%04x 0x%04x -- "
    357 					    "really probing.\n",
    358 					PCI_VENDOR(id), PCI_PRODUCT(id));
    359 				}
    360 #ifdef __HAVE_PCI_CONF_HOOK
    361 				confmode = pci_conf_hook(pb->pc, pb->busno,
    362 				    device, function, id);
    363 				if (confmode == 0)
    364 					continue;
    365 #else
    366 				/*
    367 				 * Don't enable expansion ROMS -- some cards
    368 				 * share address decoders between the EXPROM
    369 				 * and PCI memory space, and enabling the ROM
    370 				 * when not needed will cause all sorts of
    371 				 * lossage.
    372 				 */
    373 				confmode = PCI_CONF_DEFAULT;
    374 #endif
    375 				if (pci_do_device_query(pb, tag, device,
    376 				    function, confmode))
    377 					return -1;
    378 				pb->ndevs++;
    379 			}
    380 		}
    381 	}
    382 	return 0;
    383 }
    384 
    385 static void
    386 alloc_busno(pciconf_bus_t *parent, pciconf_bus_t *pb)
    387 {
    388 	pb->busno = parent->next_busno;
    389 	pb->next_busno = pb->busno + 1;
    390 }
    391 
    392 static void
    393 set_busreg(pci_chipset_tag_t pc, pcitag_t tag, int prim, int sec, int sub)
    394 {
    395 	pcireg_t	busreg;
    396 
    397 	busreg  = __SHIFTIN(prim, PCI_BRIDGE_BUS_PRIMARY);
    398 	busreg |= __SHIFTIN(sec,  PCI_BRIDGE_BUS_SECONDARY);
    399 	busreg |= __SHIFTIN(sub,  PCI_BRIDGE_BUS_SUBORDINATE);
    400 	pci_conf_write(pc, tag, PCI_BRIDGE_BUS_REG, busreg);
    401 }
    402 
    403 static pciconf_bus_t *
    404 query_bus(pciconf_bus_t *parent, pciconf_dev_t *pd, int dev)
    405 {
    406 	pciconf_bus_t	*pb;
    407 	pcireg_t	io, pmem;
    408 	pciconf_win_t	*pi, *pm;
    409 
    410 	pb = kmem_zalloc(sizeof (pciconf_bus_t), KM_SLEEP);
    411 	pb->cacheline_size = parent->cacheline_size;
    412 	pb->parent_bus = parent;
    413 	alloc_busno(parent, pb);
    414 
    415 	pb->mem_align = 0x100000;	/* 1M alignment */
    416 	pb->pmem_align = 0x100000;	/* 1M alignment */
    417 	pb->io_align = 0x1000;		/* 4K alignment */
    418 
    419 	set_busreg(parent->pc, pd->tag, parent->busno, pb->busno, 0xff);
    420 
    421 	pb->swiz = parent->swiz + dev;
    422 
    423 	memset(&pb->io_res, 0, sizeof(pb->io_res));
    424 	memset(&pb->mem_res, 0, sizeof(pb->mem_res));
    425 	memset(&pb->pmem_res, 0, sizeof(pb->pmem_res));
    426 
    427 	pb->pc = parent->pc;
    428 	pb->io_total = pb->mem_total = pb->pmem_total = 0;
    429 
    430 	pb->io_32bit = 0;
    431 	if (parent->io_32bit) {
    432 		io = pci_conf_read(parent->pc, pd->tag, PCI_BRIDGE_STATIO_REG);
    433 		if (PCI_BRIDGE_IO_32BITS(io))
    434 			pb->io_32bit = 1;
    435 	}
    436 
    437 	pb->pmem_64bit = 0;
    438 	if (parent->pmem_64bit) {
    439 		pmem = pci_conf_read(parent->pc, pd->tag,
    440 		    PCI_BRIDGE_PREFETCHMEM_REG);
    441 		if (PCI_BRIDGE_PREFETCHMEM_64BITS(pmem))
    442 			pb->pmem_64bit = 1;
    443 	}
    444 
    445 	/* Bridges only forward a 32-bit range of non-prefetcable memory. */
    446 	pb->mem_64bit = 0;
    447 
    448 	if (probe_bus(pb)) {
    449 		printf("Failed to probe bus %d\n", pb->busno);
    450 		goto err;
    451 	}
    452 
    453 	/* We have found all subordinate busses now, reprogram busreg. */
    454 	pb->last_busno = pb->next_busno - 1;
    455 	parent->next_busno = pb->next_busno;
    456 	set_busreg(parent->pc, pd->tag, parent->busno, pb->busno,
    457 		   pb->last_busno);
    458 	if (pci_conf_debug)
    459 		printf("PCI bus bridge (parent %d) covers busses %d-%d\n",
    460 			parent->busno, pb->busno, pb->last_busno);
    461 
    462 	if (pb->io_total > 0) {
    463 		if (parent->niowin >= MAX_CONF_IO) {
    464 			printf("pciconf: too many (%d) I/O windows\n",
    465 			    parent->niowin);
    466 			goto err;
    467 		}
    468 		pb->io_total = roundup2(pb->io_total, pb->io_align);
    469 		pi = get_io_desc(parent, pb->io_align);
    470 		pi->dev = pd;
    471 		pi->reg = 0;
    472 		pi->size = pb->io_total;
    473 		pi->align = pb->io_align;	/* 4K min alignment */
    474 		if (parent->io_align < pb->io_align)
    475 			parent->io_align = pb->io_align;
    476 		pi->prefetch = 0;
    477 		parent->niowin++;
    478 		parent->io_total += pb->io_total;
    479 	}
    480 
    481 	if (pb->mem_total > 0) {
    482 		if (parent->nmemwin >= MAX_CONF_MEM) {
    483 			printf("pciconf: too many (%d) MEM windows\n",
    484 			    parent->nmemwin);
    485 			goto err;
    486 		}
    487 		pb->mem_total = roundup2(pb->mem_total, pb->mem_align);
    488 		pm = get_mem_desc(parent, pb->mem_align);
    489 		pm->dev = pd;
    490 		pm->reg = 0;
    491 		pm->size = pb->mem_total;
    492 		pm->align = pb->mem_align;	/* 1M min alignment */
    493 		if (parent->mem_align < pb->mem_align)
    494 			parent->mem_align = pb->mem_align;
    495 		pm->prefetch = 0;
    496 		parent->nmemwin++;
    497 		parent->mem_total += pb->mem_total;
    498 	}
    499 
    500 	if (pb->pmem_total > 0) {
    501 		if (parent->nmemwin >= MAX_CONF_MEM) {
    502 			printf("pciconf: too many MEM windows\n");
    503 			goto err;
    504 		}
    505 		pb->pmem_total = roundup2(pb->pmem_total, pb->pmem_align);
    506 		pm = get_mem_desc(parent, pb->pmem_align);
    507 		pm->dev = pd;
    508 		pm->reg = 0;
    509 		pm->size = pb->pmem_total;
    510 		pm->align = pb->pmem_align;	/* 1M min alignment */
    511 		if (parent->pmem_align < pb->pmem_align)
    512 			parent->pmem_align = pb->pmem_align;
    513 		pm->prefetch = 1;
    514 		parent->nmemwin++;
    515 		parent->pmem_total += pb->pmem_total;
    516 	}
    517 
    518 	return pb;
    519 err:
    520 	kmem_free(pb, sizeof(*pb));
    521 	return NULL;
    522 }
    523 
    524 static struct pciconf_resource_rsvd *
    525 pci_resource_is_reserved(int type, uint64_t addr, uint64_t size)
    526 {
    527 	struct pciconf_resource_rsvd *rsvd;
    528 
    529 	LIST_FOREACH(rsvd, &pciconf_resource_reservations, next) {
    530 		if (rsvd->type != type)
    531 			continue;
    532 		if (rsvd->start <= addr + size &&
    533 		    rsvd->start + rsvd->size >= addr)
    534 			return rsvd;
    535 	}
    536 
    537 	return NULL;
    538 }
    539 
    540 static struct pciconf_resource_rsvd *
    541 pci_bar_is_reserved(pciconf_bus_t *pb, pciconf_dev_t *pd, int br)
    542 {
    543 	pcireg_t base, base64, mask, mask64;
    544 	pcitag_t tag;
    545 	uint64_t addr, size;
    546 
    547 	/*
    548 	 * Resource reservation does not apply to bridges
    549 	 */
    550 	if (pd->ppb)
    551 		return NULL;
    552 
    553 	tag = pd->tag;
    554 
    555 	/*
    556 	 * Look to see if this device is enabled and one of the resources
    557 	 * is already in use (eg. firmware configured console device).
    558 	 */
    559 	base = pci_conf_read(pb->pc, tag, br);
    560 	pci_conf_write(pb->pc, tag, br, 0xffffffff);
    561 	mask = pci_conf_read(pb->pc, tag, br);
    562 	pci_conf_write(pb->pc, tag, br, base);
    563 
    564 	switch (PCI_MAPREG_TYPE(base)) {
    565 	case PCI_MAPREG_TYPE_IO:
    566 		addr = PCI_MAPREG_IO_ADDR(base);
    567 		size = PCI_MAPREG_IO_SIZE(mask);
    568 		return pci_resource_is_reserved(PCI_CONF_MAP_IO, addr, size);
    569 
    570 	case PCI_MAPREG_TYPE_MEM:
    571 		if (PCI_MAPREG_MEM_TYPE(base) == PCI_MAPREG_MEM_TYPE_64BIT) {
    572 			base64 = pci_conf_read(pb->pc, tag, br + 4);
    573 			pci_conf_write(pb->pc, tag, br + 4, 0xffffffff);
    574 			mask64 = pci_conf_read(pb->pc, tag, br + 4);
    575 			pci_conf_write(pb->pc, tag, br + 4, base64);
    576 			addr = (uint64_t)PCI_MAPREG_MEM64_ADDR(
    577 			      (((uint64_t)base64) << 32) | base);
    578 			size = (uint64_t)PCI_MAPREG_MEM64_SIZE(
    579 			      (((uint64_t)mask64) << 32) | mask);
    580 		} else {
    581 			addr = PCI_MAPREG_MEM_ADDR(base);
    582 			size = PCI_MAPREG_MEM_SIZE(mask);
    583 		}
    584 		return pci_resource_is_reserved(PCI_CONF_MAP_MEM, addr, size);
    585 
    586 	default:
    587 		return NULL;
    588 	}
    589 }
    590 
    591 static int
    592 pci_do_device_query(pciconf_bus_t *pb, pcitag_t tag, int dev, int func,
    593     int mode)
    594 {
    595 	pciconf_dev_t	*pd;
    596 	pciconf_win_t	*pi, *pm;
    597 	pcireg_t	classreg, cmd, icr, bhlc, bar, mask, bar64, mask64,
    598 	    busreg;
    599 	uint64_t	size;
    600 	u_int		hdrtype, br, width, reg_start, reg_end, reg_rom;
    601 
    602 	pd = &pb->device[pb->ndevs];
    603 	pd->pc = pb->pc;
    604 	pd->tag = tag;
    605 	pd->ppb = NULL;
    606 	pd->enable = mode;
    607 	pd->ea_cap_ptr = 0;
    608 
    609 	classreg = pci_conf_read(pb->pc, tag, PCI_CLASS_REG);
    610 
    611 	cmd = pci_conf_read(pb->pc, tag, PCI_COMMAND_STATUS_REG);
    612 	bhlc = pci_conf_read(pb->pc, tag, PCI_BHLC_REG);
    613 
    614 	if (pci_get_capability(pb->pc, tag, PCI_CAP_EA, &pd->ea_cap_ptr,
    615 	    NULL)) {
    616 		/* XXX Skip devices with EA for now. */
    617 		print_tag(pb->pc, tag);
    618 		printf("skipping devices with Enhanced Allocations\n");
    619 		return 0;
    620 	}
    621 
    622 	hdrtype = PCI_HDRTYPE_TYPE(bhlc);
    623 	if (PCI_CLASS(classreg) != PCI_CLASS_BRIDGE
    624 	    && hdrtype != PCI_HDRTYPE_PPB) {
    625 		cmd &= ~(PCI_COMMAND_MASTER_ENABLE |
    626 		    PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE);
    627 		pci_conf_write(pb->pc, tag, PCI_COMMAND_STATUS_REG, cmd);
    628 	} else if (pci_conf_debug) {
    629 		print_tag(pb->pc, tag);
    630 		printf("device is a bridge; not clearing enables\n");
    631 	}
    632 
    633 	if ((cmd & PCI_STATUS_BACKTOBACK_SUPPORT) == 0)
    634 		pb->fast_b2b = 0;
    635 
    636 	if ((cmd & PCI_STATUS_66MHZ_SUPPORT) == 0)
    637 		pb->freq_66 = 0;
    638 
    639 	reg_rom = PCI_MAPREG_ROM; /* Default */
    640 	switch (hdrtype) {
    641 	case PCI_HDRTYPE_DEVICE:
    642 		reg_start = PCI_MAPREG_START;
    643 		reg_end = PCI_MAPREG_END;
    644 		break;
    645 	case PCI_HDRTYPE_PPB:
    646 		pd->ppb = query_bus(pb, pd, dev);
    647 		if (pd->ppb == NULL)
    648 			return -1;
    649 		/*
    650 		 * PCI-to-PCI Bridge Spec rev. 1.2.
    651 		 * "3.2.5.1 Base Address registers" says that if the optional
    652 		 * BARs are implemented, the configuration software must map
    653 		 * address ranges.
    654 		 */
    655 		reg_start = PCI_MAPREG_START;
    656 		reg_end = PCI_MAPREG_PPB_END; /* Only BAR0 and BAR1 */
    657 		reg_rom = PCI_BRIDGE_EXPROMADDR_REG;
    658 		break;
    659 	case PCI_HDRTYPE_PCB:
    660 		reg_start = PCI_MAPREG_START;
    661 		reg_end = PCI_MAPREG_PCB_END;
    662 
    663 		busreg = pci_conf_read(pb->pc, tag, PCI_BUSNUM);
    664 		busreg = (busreg & 0xff000000) |
    665 		    __SHIFTIN(pb->busno, PCI_BRIDGE_BUS_PRIMARY) |
    666 		    __SHIFTIN(pb->next_busno, PCI_BRIDGE_BUS_SECONDARY) |
    667 		    __SHIFTIN(pb->next_busno, PCI_BRIDGE_BUS_SUBORDINATE);
    668 		pci_conf_write(pb->pc, tag, PCI_BUSNUM, busreg);
    669 
    670 		pb->next_busno++;
    671 		break;
    672 	default:
    673 		return -1;
    674 	}
    675 
    676 	icr = pci_conf_read(pb->pc, tag, PCI_INTERRUPT_REG);
    677 	pd->ipin = PCI_INTERRUPT_PIN(icr);
    678 	pd->iline = PCI_INTERRUPT_LINE(icr);
    679 	if (pd->iline || pd->ipin) {
    680 		pci_conf_interrupt(pb->pc, pb->busno, dev, pd->ipin, pb->swiz,
    681 		    &pd->iline);
    682 		icr &= ~(PCI_INTERRUPT_LINE_MASK << PCI_INTERRUPT_LINE_SHIFT);
    683 		icr |= (pd->iline << PCI_INTERRUPT_LINE_SHIFT);
    684 		pci_conf_write(pb->pc, tag, PCI_INTERRUPT_REG, icr);
    685 	}
    686 
    687 	/* For non-PPB devices */
    688 	if (hdrtype != PCI_HDRTYPE_PPB) {
    689 		pd->min_gnt = PCI_MIN_GNT(icr);
    690 		pd->max_lat = PCI_MAX_LAT(icr);
    691 		if (pd->min_gnt != 0 || pd->max_lat != 0) {
    692 			if (pd->min_gnt != 0 && pd->min_gnt > pb->max_mingnt)
    693 				pb->max_mingnt = pd->min_gnt;
    694 
    695 			if (pd->max_lat != 0 && pd->max_lat < pb->min_maxlat)
    696 				pb->min_maxlat = pd->max_lat;
    697 
    698 			pb->bandwidth_used += pd->min_gnt * 4000000 /
    699 			    (pd->min_gnt + pd->max_lat);
    700 		}
    701 	}
    702 
    703 	width = 4;
    704 	for (br = reg_start; br < reg_end; br += width) {
    705 #if 0
    706 /* XXX Should only ignore if IDE not in legacy mode? */
    707 		if (PCI_CLASS(classreg) == PCI_CLASS_MASS_STORAGE &&
    708 		    PCI_SUBCLASS(classreg) == PCI_SUBCLASS_MASS_STORAGE_IDE) {
    709 			break;
    710 		}
    711 #endif
    712 		bar = pci_conf_read(pb->pc, tag, br);
    713 		pci_conf_write(pb->pc, tag, br, 0xffffffff);
    714 		mask = pci_conf_read(pb->pc, tag, br);
    715 		pci_conf_write(pb->pc, tag, br, bar);
    716 		width = 4;
    717 
    718 		if (   (mode & PCI_CONF_MAP_IO)
    719 		    && (PCI_MAPREG_TYPE(mask) == PCI_MAPREG_TYPE_IO)) {
    720 			/*
    721 			 * Upper 16 bits must be one.  Devices may hardwire
    722 			 * them to zero, though, per PCI 2.2, 6.2.5.1, p 203.
    723 			 */
    724 			mask |= 0xffff0000;
    725 
    726 			size = PCI_MAPREG_IO_SIZE(mask);
    727 			if (size == 0) {
    728 				if (pci_conf_debug) {
    729 					print_tag(pb->pc, tag);
    730 					printf("I/O BAR 0x%x is void\n", br);
    731 				}
    732 				continue;
    733 			}
    734 
    735 			if (pb->niowin >= MAX_CONF_IO) {
    736 				printf("pciconf: too many I/O windows\n");
    737 				return -1;
    738 			}
    739 
    740 			pi = get_io_desc(pb, size /*align*/);
    741 			pi->dev = pd;
    742 			pi->reg = br;
    743 			pi->size = pi->align = size;
    744 			if (pb->io_align < pi->align)
    745 				pb->io_align = pi->align;
    746 			pi->prefetch = 0;
    747 			if (pci_conf_debug) {
    748 				print_tag(pb->pc, tag);
    749 				printf("Register 0x%x, I/O size %" PRIu64 "\n",
    750 				    br, pi->size);
    751 			}
    752 			pb->niowin++;
    753 			pb->io_total += size;
    754 		} else if ((mode & PCI_CONF_MAP_MEM)
    755 			   && (PCI_MAPREG_TYPE(mask) == PCI_MAPREG_TYPE_MEM)) {
    756 			switch (PCI_MAPREG_MEM_TYPE(mask)) {
    757 			case PCI_MAPREG_MEM_TYPE_32BIT:
    758 			case PCI_MAPREG_MEM_TYPE_32BIT_1M:
    759 				size = (uint64_t)PCI_MAPREG_MEM_SIZE(mask);
    760 				break;
    761 			case PCI_MAPREG_MEM_TYPE_64BIT:
    762 				bar64 = pci_conf_read(pb->pc, tag, br + 4);
    763 				pci_conf_write(pb->pc, tag, br + 4, 0xffffffff);
    764 				mask64 = pci_conf_read(pb->pc, tag, br + 4);
    765 				pci_conf_write(pb->pc, tag, br + 4, bar64);
    766 				size = (uint64_t)PCI_MAPREG_MEM64_SIZE(
    767 				      (((uint64_t)mask64) << 32) | mask);
    768 				width = 8;
    769 				break;
    770 			default:
    771 				print_tag(pb->pc, tag);
    772 				printf("reserved mapping type 0x%x\n",
    773 					PCI_MAPREG_MEM_TYPE(mask));
    774 				continue;
    775 			}
    776 
    777 			if (size == 0) {
    778 				if (pci_conf_debug) {
    779 					print_tag(pb->pc, tag);
    780 					printf("MEM%d BAR 0x%x is void\n",
    781 					    PCI_MAPREG_MEM_TYPE(mask) ==
    782 						PCI_MAPREG_MEM_TYPE_64BIT ?
    783 						64 : 32, br);
    784 				}
    785 				continue;
    786 			} else {
    787 				if (pci_conf_debug) {
    788 					print_tag(pb->pc, tag);
    789 					printf("MEM%d BAR 0x%x has size %#lx\n",
    790 					    PCI_MAPREG_MEM_TYPE(mask) ==
    791 						PCI_MAPREG_MEM_TYPE_64BIT ?
    792 						64 : 32,
    793 					    br, (unsigned long)size);
    794 				}
    795 			}
    796 
    797 			if (pb->nmemwin >= MAX_CONF_MEM) {
    798 				printf("pciconf: too many memory windows\n");
    799 				return -1;
    800 			}
    801 
    802 			pm = get_mem_desc(pb, size /*align*/);
    803 			pm->dev = pd;
    804 			pm->reg = br;
    805 			pm->size = pm->align = size;
    806 			pm->prefetch = PCI_MAPREG_MEM_PREFETCHABLE(mask);
    807 			if (pci_conf_debug) {
    808 				print_tag(pb->pc, tag);
    809 				printf("Register 0x%x, memory size %"
    810 				    PRIu64 "\n", br, pm->size);
    811 			}
    812 			pb->nmemwin++;
    813 			if (pm->prefetch) {
    814 				pb->pmem_total += size;
    815 				if (pb->pmem_align < pm->align)
    816 					pb->pmem_align = pm->align;
    817 			} else {
    818 				pb->mem_total += size;
    819 				if (pb->mem_align < pm->align)
    820 					pb->mem_align = pm->align;
    821 			}
    822 		}
    823 	}
    824 
    825 	if (mode & PCI_CONF_MAP_ROM) {
    826 		bar = pci_conf_read(pb->pc, tag, reg_rom);
    827 		pci_conf_write(pb->pc, tag, reg_rom, 0xfffffffe);
    828 		mask = pci_conf_read(pb->pc, tag, reg_rom);
    829 		pci_conf_write(pb->pc, tag, reg_rom, bar);
    830 
    831 		if (mask != 0 && mask != 0xffffffff) {
    832 			if (pb->nmemwin >= MAX_CONF_MEM) {
    833 				printf("pciconf: too many memory windows\n");
    834 				return -1;
    835 			}
    836 			size = (uint64_t)PCI_MAPREG_MEM_SIZE(mask);
    837 
    838 			pm = get_mem_desc(pb, size /*align*/);
    839 			pm->dev = pd;
    840 			pm->reg = reg_rom;
    841 			pm->size = pm->align = size;
    842 			pm->prefetch = 0;
    843 			if (pci_conf_debug) {
    844 				print_tag(pb->pc, tag);
    845 				printf("Expansion ROM memory size %"
    846 				    PRIu64 "\n", pm->size);
    847 			}
    848 			pb->nmemwin++;
    849 			if (pm->prefetch) {
    850 				pb->pmem_total += size;
    851 				if (pb->pmem_align < pm->align)
    852 					pb->pmem_align = pm->align;
    853 			} else {
    854 				pb->mem_total += size;
    855 				if (pb->mem_align < pm->align)
    856 					pb->mem_align = pm->align;
    857 			}
    858 		}
    859 	} else {
    860 		/* Don't enable ROMs if we aren't going to map them. */
    861 		mode &= ~PCI_CONF_ENABLE_ROM;
    862 		pd->enable &= ~PCI_CONF_ENABLE_ROM;
    863 	}
    864 
    865 	if (!(mode & PCI_CONF_ENABLE_ROM)) {
    866 		/* Ensure ROM is disabled */
    867 		bar = pci_conf_read(pb->pc, tag, reg_rom);
    868 		pci_conf_write(pb->pc, tag, reg_rom,
    869 		    bar & ~PCI_MAPREG_ROM_ENABLE);
    870 	}
    871 
    872 	return 0;
    873 }
    874 
    875 /************************************************************************/
    876 /************************************************************************/
    877 /********************   Bus configuration routines   ********************/
    878 /************************************************************************/
    879 /************************************************************************/
    880 static uint64_t
    881 pci_allocate_range(struct pciconf_resource * const r, const uint64_t amt,
    882 		   const int align, const bool ok64 __used_only_lp64)
    883 {
    884 	vmem_size_t const size = (vmem_size_t) amt;
    885 	vmem_addr_t result;
    886 	int error;
    887 
    888 #ifdef _LP64
    889 	/*
    890 	 * If a 64-bit range IS OK, then we prefer allocating above 4GB.
    891 	 *
    892 	 * XXX We guard this with _LP64 because vmem uses uintptr_t
    893 	 * internally.
    894 	 */
    895 	if (!ok64) {
    896 		error = vmem_xalloc(r->arena, size, align, 0, 0,
    897 				    VMEM_ADDR_MIN, 0xffffffffUL,
    898 				    VM_BESTFIT | VM_NOSLEEP,
    899 				    &result);
    900 	} else {
    901 		error = vmem_xalloc(r->arena, size, align, 0, 0,
    902 				    (1UL << 32), VMEM_ADDR_MAX,
    903 				    VM_BESTFIT | VM_NOSLEEP,
    904 				    &result);
    905 		if (error) {
    906 			error = vmem_xalloc(r->arena, size, align, 0, 0,
    907 					    VMEM_ADDR_MIN, VMEM_ADDR_MAX,
    908 					    VM_BESTFIT | VM_NOSLEEP,
    909 					    &result);
    910 		}
    911 	}
    912 #else
    913 	error = vmem_xalloc(r->arena, size, align, 0, 0,
    914 			    VMEM_ADDR_MIN, 0xffffffffUL,
    915 			    VM_BESTFIT | VM_NOSLEEP,
    916 			    &result);
    917 #endif /* _L64 */
    918 
    919 	if (error)
    920 		return ~0ULL;
    921 
    922 	return result;
    923 }
    924 
    925 static int
    926 setup_iowins(pciconf_bus_t *pb)
    927 {
    928 	pciconf_win_t	*pi;
    929 	pciconf_dev_t	*pd;
    930 	struct pciconf_resource_rsvd *rsvd;
    931 	int		error;
    932 
    933 	for (pi = pb->pciiowin; pi < &pb->pciiowin[pb->niowin]; pi++) {
    934 		if (pi->size == 0)
    935 			continue;
    936 
    937 		pd = pi->dev;
    938 		rsvd = pci_bar_is_reserved(pb, pd, pi->reg);
    939 
    940 		if (pb->io_res.arena == NULL) {
    941 			/* Bus has no IO ranges, disable IO BAR */
    942 			pi->address = 0;
    943 			pd->enable &= ~PCI_CONF_ENABLE_IO;
    944 			goto write_ioaddr;
    945 		}
    946 
    947 		pi->address = pci_allocate_range(&pb->io_res, pi->size,
    948 		    pi->align, false);
    949 		if (~pi->address == 0) {
    950 			print_tag(pd->pc, pd->tag);
    951 			printf("Failed to allocate PCI I/O space (%"
    952 			    PRIu64 " req)\n", pi->size);
    953 			return -1;
    954 		}
    955 		if (pd->ppb && pi->reg == 0) {
    956 			error = init_range_resource(&pd->ppb->io_res,
    957 			    "ppb-io", pi->address, pi->size);
    958 			if (error) {
    959 				print_tag(pd->pc, pd->tag);
    960 				printf("Failed to alloc I/O arena for bus %d\n",
    961 				    pd->ppb->busno);
    962 				return -1;
    963 			}
    964 			continue;
    965 		}
    966 		if (!pb->io_32bit && pi->address > 0xFFFF) {
    967 			pi->address = 0;
    968 			pd->enable &= ~PCI_CONF_ENABLE_IO;
    969 		} else {
    970 			pd->enable |= PCI_CONF_ENABLE_IO;
    971 		}
    972 write_ioaddr:
    973 		if (pci_conf_debug) {
    974 			print_tag(pd->pc, pd->tag);
    975 			printf("Putting %" PRIu64 " I/O bytes @ %#" PRIx64
    976 			    " (reg %x)\n", pi->size, pi->address, pi->reg);
    977 		}
    978 		pci_conf_write(pd->pc, pd->tag, pi->reg,
    979 		    PCI_MAPREG_IO_ADDR(pi->address) | PCI_MAPREG_TYPE_IO);
    980 
    981 		if (rsvd != NULL && rsvd->start != pi->address)
    982 			rsvd->callback(rsvd->callback_arg, pi->address);
    983 	}
    984 	return 0;
    985 }
    986 
    987 static int
    988 setup_memwins(pciconf_bus_t *pb)
    989 {
    990 	pciconf_win_t	*pm;
    991 	pciconf_dev_t	*pd;
    992 	pcireg_t	base;
    993 	struct pciconf_resource *r;
    994 	struct pciconf_resource_rsvd *rsvd;
    995 	bool		ok64;
    996 	int		error;
    997 
    998 	for (pm = pb->pcimemwin; pm < &pb->pcimemwin[pb->nmemwin]; pm++) {
    999 		if (pm->size == 0)
   1000 			continue;
   1001 
   1002 		ok64 = false;
   1003 		pd = pm->dev;
   1004 		rsvd = pci_bar_is_reserved(pb, pd, pm->reg);
   1005 
   1006 		if (pm->prefetch) {
   1007 			r = &pb->pmem_res;
   1008 			ok64 = pb->pmem_64bit;
   1009 		} else {
   1010 			r = &pb->mem_res;
   1011 			ok64 = pb->mem_64bit && pd->ppb == NULL;
   1012 		}
   1013 
   1014 		/*
   1015 		 * We need to figure out if the memory BAR is 64-bit
   1016 		 * capable or not.  If it's not, then we need to constrain
   1017 		 * the address allocation.
   1018 		 */
   1019 		if (PCICONF_MAPREG_IS_ROM(pm->reg)) {
   1020 			ok64 = false;
   1021 		} else if (ok64) {
   1022 			base = pci_conf_read(pd->pc, pd->tag, pm->reg);
   1023 			ok64 = PCI_MAPREG_MEM_TYPE(base) ==
   1024 			    PCI_MAPREG_MEM_TYPE_64BIT;
   1025 		}
   1026 
   1027 		pm->address = pci_allocate_range(r, pm->size, pm->align,
   1028 						 ok64);
   1029 		if (~pm->address == 0 && r == &pb->pmem_res) {
   1030 			r = &pb->mem_res;
   1031 			pm->address = pci_allocate_range(r, pm->size,
   1032 							 pm->align, ok64);
   1033 		}
   1034 		if (~pm->address == 0) {
   1035 			print_tag(pd->pc, pd->tag);
   1036 			printf(
   1037 			   "Failed to allocate PCI memory space (%" PRIu64
   1038 			   " req, prefetch=%d ok64=%d)\n", pm->size,
   1039 			   pm->prefetch, (int)ok64);
   1040 			return -1;
   1041 		}
   1042 		if (pd->ppb && pm->reg == 0) {
   1043 			const char *name = pm->prefetch ? "ppb-pmem"
   1044 							: "ppb-mem";
   1045 			r = pm->prefetch ? &pd->ppb->pmem_res
   1046 					 : &pd->ppb->mem_res;
   1047 			error = init_range_resource(r, name,
   1048 			    pm->address, pm->size);
   1049 			if (error) {
   1050 				print_tag(pd->pc, pd->tag);
   1051 				printf("Failed to alloc MEM arena for bus %d\n",
   1052 				    pd->ppb->busno);
   1053 				return -1;
   1054 			}
   1055 			continue;
   1056 		}
   1057 		if (!ok64 && pm->address > 0xFFFFFFFFULL) {
   1058 			pm->address = 0;
   1059 			pd->enable &= ~PCI_CONF_ENABLE_MEM;
   1060 		} else
   1061 			pd->enable |= PCI_CONF_ENABLE_MEM;
   1062 
   1063 		if (!PCICONF_MAPREG_IS_ROM(pm->reg)) {
   1064 			if (pci_conf_debug) {
   1065 				print_tag(pd->pc, pd->tag);
   1066 				printf(
   1067 				    "Putting %" PRIu64 " MEM bytes @ %#"
   1068 				    PRIx64 " (reg %x)\n", pm->size,
   1069 				    pm->address, pm->reg);
   1070 			}
   1071 			base = pci_conf_read(pd->pc, pd->tag, pm->reg);
   1072 			base = PCI_MAPREG_MEM_ADDR(pm->address) |
   1073 			    PCI_MAPREG_MEM_TYPE(base);
   1074 			pci_conf_write(pd->pc, pd->tag, pm->reg, base);
   1075 			if (PCI_MAPREG_MEM_TYPE(base) ==
   1076 			    PCI_MAPREG_MEM_TYPE_64BIT) {
   1077 				base = (pcireg_t)
   1078 				    (PCI_MAPREG_MEM64_ADDR(pm->address) >> 32);
   1079 				pci_conf_write(pd->pc, pd->tag, pm->reg + 4,
   1080 				    base);
   1081 			}
   1082 		}
   1083 
   1084 		if (rsvd != NULL && rsvd->start != pm->address) {
   1085 			/*
   1086 			 * Resource allocation will never reuse a reserved
   1087 			 * address. Check to see if the BAR is still reserved
   1088 			 * to cover the case where the new resource was not
   1089 			 * applied. In this case, there is no need to notify
   1090 			 * the device callback of a change.
   1091 			 */
   1092 			if (!pci_bar_is_reserved(pb, pd, pm->reg)) {
   1093 				rsvd->callback(rsvd->callback_arg, pm->address);
   1094 			}
   1095 		}
   1096 	}
   1097 	for (pm = pb->pcimemwin; pm < &pb->pcimemwin[pb->nmemwin]; pm++) {
   1098 		if (PCICONF_MAPREG_IS_ROM(pm->reg) && pm->address != -1) {
   1099 			pd = pm->dev;
   1100 			if (!(pd->enable & PCI_CONF_MAP_ROM))
   1101 				continue;
   1102 			if (pci_conf_debug) {
   1103 				print_tag(pd->pc, pd->tag);
   1104 				printf(
   1105 				    "Putting %" PRIu64 " ROM bytes @ %#"
   1106 				    PRIx64 " (reg %x)\n", pm->size,
   1107 				    pm->address, pm->reg);
   1108 			}
   1109 			base = (pcireg_t) pm->address;
   1110 			if (pd->enable & PCI_CONF_ENABLE_ROM)
   1111 				base |= PCI_MAPREG_ROM_ENABLE;
   1112 
   1113 			pci_conf_write(pd->pc, pd->tag, pm->reg, base);
   1114 		}
   1115 	}
   1116 	return 0;
   1117 }
   1118 
   1119 static bool
   1120 constrain_bridge_mem_range(struct pciconf_resource * const r,
   1121 			   u_long * const base,
   1122 			   u_long * const limit,
   1123 			   const bool ok64 __used_only_lp64)
   1124 {
   1125 
   1126 	*base = r->min_addr;
   1127 	*limit = r->max_addr;
   1128 
   1129 #ifdef _LP64
   1130 	if (!ok64) {
   1131 		if (r->min_addr >= (1UL << 32)) {
   1132 			return true;
   1133 		}
   1134 		if (r->max_addr > 0xffffffffUL) {
   1135 			*limit = 0xffffffffUL;
   1136 		}
   1137 	}
   1138 #endif /* _LP64 */
   1139 
   1140 	return false;
   1141 }
   1142 
   1143 /*
   1144  * Configure I/O, memory, and prefetcable memory spaces, then make
   1145  * a call to configure_bus().
   1146  */
   1147 static int
   1148 configure_bridge(pciconf_dev_t *pd)
   1149 {
   1150 	unsigned long	io_base, io_limit, mem_base, mem_limit;
   1151 	pciconf_bus_t	*pb;
   1152 	pcireg_t	io, iohigh, mem, cmd;
   1153 	int		rv;
   1154 	bool		isprefetchmem64;
   1155 	bool		bad_range;
   1156 
   1157 	pb = pd->ppb;
   1158 	/* Configure I/O base & limit*/
   1159 	if (pb->io_res.arena) {
   1160 		io_base = pb->io_res.min_addr;
   1161 		io_limit = pb->io_res.max_addr;
   1162 	} else {
   1163 		io_base  = 0x1000;	/* 4K */
   1164 		io_limit = 0x0000;
   1165 	}
   1166 	if (pb->io_32bit) {
   1167 		iohigh = __SHIFTIN(io_base >> 16, PCI_BRIDGE_IOHIGH_BASE) |
   1168 		    __SHIFTIN(io_limit >> 16, PCI_BRIDGE_IOHIGH_LIMIT);
   1169 	} else {
   1170 		if (io_limit > 0xFFFF) {
   1171 			printf("Bus %d bridge does not support 32-bit I/O.  ",
   1172 			    pb->busno);
   1173 			printf("Disabling I/O accesses\n");
   1174 			io_base  = 0x1000;	/* 4K */
   1175 			io_limit = 0x0000;
   1176 		}
   1177 		iohigh = 0;
   1178 	}
   1179 	io = pci_conf_read(pb->pc, pd->tag, PCI_BRIDGE_STATIO_REG) &
   1180 	    PCI_BRIDGE_STATIO_STATUS;
   1181 	io |= __SHIFTIN((io_base >> 8) & PCI_BRIDGE_STATIO_IOADDR,
   1182 	    PCI_BRIDGE_STATIO_IOBASE);
   1183 	io |= __SHIFTIN((io_limit >> 8) & PCI_BRIDGE_STATIO_IOADDR,
   1184 	    PCI_BRIDGE_STATIO_IOLIMIT);
   1185 	pci_conf_write(pb->pc, pd->tag, PCI_BRIDGE_STATIO_REG, io);
   1186 	pci_conf_write(pb->pc, pd->tag, PCI_BRIDGE_IOHIGH_REG, iohigh);
   1187 
   1188 	/* Configure mem base & limit */
   1189 	bad_range = false;
   1190 	if (pb->mem_res.arena) {
   1191 		bad_range = constrain_bridge_mem_range(&pb->mem_res,
   1192 						       &mem_base,
   1193 						       &mem_limit,
   1194 						       false);
   1195 	} else {
   1196 		mem_base  = 0x100000;	/* 1M */
   1197 		mem_limit = 0x000000;
   1198 	}
   1199 	if (bad_range) {
   1200 		printf("Bus %d bridge MEM range out of range.  ", pb->busno);
   1201 		printf("Disabling MEM accesses\n");
   1202 		mem_base  = 0x100000;	/* 1M */
   1203 		mem_limit = 0x000000;
   1204 	}
   1205 	mem = __SHIFTIN((mem_base >> 16) & PCI_BRIDGE_MEMORY_ADDR,
   1206 	    PCI_BRIDGE_MEMORY_BASE);
   1207 	mem |= __SHIFTIN((mem_limit >> 16) & PCI_BRIDGE_MEMORY_ADDR,
   1208 	    PCI_BRIDGE_MEMORY_LIMIT);
   1209 	pci_conf_write(pb->pc, pd->tag, PCI_BRIDGE_MEMORY_REG, mem);
   1210 
   1211 	/* Configure prefetchable mem base & limit */
   1212 	mem = pci_conf_read(pb->pc, pd->tag, PCI_BRIDGE_PREFETCHMEM_REG);
   1213 	isprefetchmem64 = PCI_BRIDGE_PREFETCHMEM_64BITS(mem);
   1214 	bad_range = false;
   1215 	if (pb->pmem_res.arena) {
   1216 		bad_range = constrain_bridge_mem_range(&pb->pmem_res,
   1217 						       &mem_base,
   1218 						       &mem_limit,
   1219 						       isprefetchmem64);
   1220 	} else {
   1221 		mem_base  = 0x100000;	/* 1M */
   1222 		mem_limit = 0x000000;
   1223 	}
   1224 	if (bad_range) {
   1225 		printf("Bus %d bridge does not support 64-bit PMEM.  ",
   1226 		    pb->busno);
   1227 		printf("Disabling prefetchable-MEM accesses\n");
   1228 		mem_base  = 0x100000;	/* 1M */
   1229 		mem_limit = 0x000000;
   1230 	}
   1231 	mem = __SHIFTIN((mem_base >> 16) & PCI_BRIDGE_PREFETCHMEM_ADDR,
   1232 	    PCI_BRIDGE_PREFETCHMEM_BASE);
   1233 	mem |= __SHIFTIN((mem_limit >> 16) & PCI_BRIDGE_PREFETCHMEM_ADDR,
   1234 	    PCI_BRIDGE_PREFETCHMEM_LIMIT);
   1235 	pci_conf_write(pb->pc, pd->tag, PCI_BRIDGE_PREFETCHMEM_REG, mem);
   1236 	/*
   1237 	 * XXX -- 64-bit systems need a lot more than just this...
   1238 	 */
   1239 	if (isprefetchmem64) {
   1240 		mem_base  = (uint64_t)mem_base  >> 32;
   1241 		mem_limit = (uint64_t)mem_limit >> 32;
   1242 		pci_conf_write(pb->pc, pd->tag,
   1243 		    PCI_BRIDGE_PREFETCHBASEUP32_REG, mem_base & 0xffffffff);
   1244 		pci_conf_write(pb->pc, pd->tag,
   1245 		    PCI_BRIDGE_PREFETCHLIMITUP32_REG, mem_limit & 0xffffffff);
   1246 	}
   1247 
   1248 	rv = configure_bus(pb);
   1249 
   1250 	fini_range_resource(&pb->io_res);
   1251 	fini_range_resource(&pb->mem_res);
   1252 	fini_range_resource(&pb->pmem_res);
   1253 
   1254 	if (rv == 0) {
   1255 		cmd = pci_conf_read(pd->pc, pd->tag, PCI_BRIDGE_CONTROL_REG);
   1256 		cmd &= ~PCI_BRIDGE_CONTROL; /* Clear control bit first */
   1257 		cmd |= PCI_BRIDGE_CONTROL_PERE | PCI_BRIDGE_CONTROL_SERR;
   1258 		if (pb->fast_b2b)
   1259 			cmd |= PCI_BRIDGE_CONTROL_SECFASTB2B;
   1260 
   1261 		pci_conf_write(pd->pc, pd->tag, PCI_BRIDGE_CONTROL_REG, cmd);
   1262 		cmd = pci_conf_read(pd->pc, pd->tag, PCI_COMMAND_STATUS_REG);
   1263 		cmd |= PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE;
   1264 		pci_conf_write(pd->pc, pd->tag, PCI_COMMAND_STATUS_REG, cmd);
   1265 	}
   1266 
   1267 	return rv;
   1268 }
   1269 
   1270 /*
   1271  * Calculate latency values, allocate I/O and MEM segments, then set them
   1272  * up.  If a PCI-PCI bridge is found, configure the bridge separately,
   1273  * which will cause a recursive call back here.
   1274  */
   1275 static int
   1276 configure_bus(pciconf_bus_t *pb)
   1277 {
   1278 	pciconf_dev_t	*pd;
   1279 	int		def_ltim, max_ltim, band, bus_mhz;
   1280 
   1281 	if (pb->ndevs == 0) {
   1282 		if (pci_conf_debug)
   1283 			printf("PCI bus %d - no devices\n", pb->busno);
   1284 		return 1;
   1285 	}
   1286 	bus_mhz = pb->freq_66 ? 66 : 33;
   1287 	max_ltim = pb->max_mingnt * bus_mhz / 4;	/* cvt to cycle count */
   1288 	band = 4000000;					/* 0.25us cycles/sec */
   1289 	if (band < pb->bandwidth_used) {
   1290 		printf("PCI bus %d: Warning: Total bandwidth exceeded!? (%d)\n",
   1291 		    pb->busno, pb->bandwidth_used);
   1292 		def_ltim = -1;
   1293 	} else {
   1294 		def_ltim = (band - pb->bandwidth_used) / pb->ndevs;
   1295 		if (def_ltim > pb->min_maxlat)
   1296 			def_ltim = pb->min_maxlat;
   1297 		def_ltim = def_ltim * bus_mhz / 4;
   1298 	}
   1299 	def_ltim = (def_ltim + 7) & ~7;
   1300 	max_ltim = (max_ltim + 7) & ~7;
   1301 
   1302 	pb->def_ltim = MIN(def_ltim, 255);
   1303 	pb->max_ltim = MIN(MAX(max_ltim, def_ltim), 255);
   1304 
   1305 	/*
   1306 	 * Now we have what we need to initialize the devices.
   1307 	 * It would probably be better if we could allocate all of these
   1308 	 * for all busses at once, but "not right now".  First, get a list
   1309 	 * of free memory ranges from the m.d. system.
   1310 	 */
   1311 	if (setup_iowins(pb) || setup_memwins(pb)) {
   1312 		printf("PCI bus configuration failed: "
   1313 		"unable to assign all I/O and memory ranges.\n");
   1314 		return -1;
   1315 	}
   1316 
   1317 	/*
   1318 	 * Configure the latency for the devices, and enable them.
   1319 	 */
   1320 	for (pd = pb->device; pd < &pb->device[pb->ndevs]; pd++) {
   1321 		pcireg_t cmd, classreg, misc;
   1322 		int	ltim;
   1323 
   1324 		if (pci_conf_debug) {
   1325 			print_tag(pd->pc, pd->tag);
   1326 			printf("Configuring device.\n");
   1327 		}
   1328 		classreg = pci_conf_read(pd->pc, pd->tag, PCI_CLASS_REG);
   1329 		misc = pci_conf_read(pd->pc, pd->tag, PCI_BHLC_REG);
   1330 		cmd = pci_conf_read(pd->pc, pd->tag, PCI_COMMAND_STATUS_REG);
   1331 		if (pd->enable & PCI_CONF_ENABLE_PARITY)
   1332 			cmd |= PCI_COMMAND_PARITY_ENABLE;
   1333 		if (pd->enable & PCI_CONF_ENABLE_SERR)
   1334 			cmd |= PCI_COMMAND_SERR_ENABLE;
   1335 		if (pb->fast_b2b)
   1336 			cmd |= PCI_COMMAND_BACKTOBACK_ENABLE;
   1337 		if (PCI_CLASS(classreg) != PCI_CLASS_BRIDGE ||
   1338 		    PCI_SUBCLASS(classreg) != PCI_SUBCLASS_BRIDGE_PCI) {
   1339 			if (pd->enable & PCI_CONF_ENABLE_IO)
   1340 				cmd |= PCI_COMMAND_IO_ENABLE;
   1341 			if (pd->enable & PCI_CONF_ENABLE_MEM)
   1342 				cmd |= PCI_COMMAND_MEM_ENABLE;
   1343 			if (pd->enable & PCI_CONF_ENABLE_BM)
   1344 				cmd |= PCI_COMMAND_MASTER_ENABLE;
   1345 			ltim = pd->min_gnt * bus_mhz / 4;
   1346 			ltim = MIN (MAX (pb->def_ltim, ltim), pb->max_ltim);
   1347 		} else {
   1348 			cmd |= PCI_COMMAND_MASTER_ENABLE;
   1349 			ltim = MIN (pb->def_ltim, pb->max_ltim);
   1350 		}
   1351 		if ((pd->enable &
   1352 		    (PCI_CONF_ENABLE_MEM | PCI_CONF_ENABLE_IO)) == 0) {
   1353 			print_tag(pd->pc, pd->tag);
   1354 			printf("Disabled due to lack of resources.\n");
   1355 			cmd &= ~(PCI_COMMAND_MASTER_ENABLE |
   1356 			    PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE);
   1357 		}
   1358 		pci_conf_write(pd->pc, pd->tag, PCI_COMMAND_STATUS_REG, cmd);
   1359 
   1360 		misc &= ~((PCI_LATTIMER_MASK << PCI_LATTIMER_SHIFT) |
   1361 		    (PCI_CACHELINE_MASK << PCI_CACHELINE_SHIFT));
   1362 		misc |= (ltim & PCI_LATTIMER_MASK) << PCI_LATTIMER_SHIFT;
   1363 		misc |= ((pb->cacheline_size >> 2) & PCI_CACHELINE_MASK) <<
   1364 		    PCI_CACHELINE_SHIFT;
   1365 		pci_conf_write(pd->pc, pd->tag, PCI_BHLC_REG, misc);
   1366 
   1367 		if (pd->ppb) {
   1368 			if (configure_bridge(pd) < 0)
   1369 				return -1;
   1370 			continue;
   1371 		}
   1372 	}
   1373 
   1374 	if (pci_conf_debug)
   1375 		printf("PCI bus %d configured\n", pb->busno);
   1376 
   1377 	return 0;
   1378 }
   1379 
   1380 static bool
   1381 mem_region_ok64(struct pciconf_resource * const r __used_only_lp64)
   1382 {
   1383 	bool rv = false;
   1384 
   1385 #ifdef _LP64
   1386 	/*
   1387 	 * XXX We need to guard this with _LP64 because vmem uses
   1388 	 * uintptr_t internally.
   1389 	 */
   1390 	vmem_size_t result;
   1391 	if (vmem_xalloc(r->arena, 1/*size*/, 1/*align*/, 0/*phase*/,
   1392 			0/*nocross*/, (1UL << 32), VMEM_ADDR_MAX,
   1393 			VM_INSTANTFIT | VM_NOSLEEP, &result) == 0) {
   1394 		vmem_free(r->arena, result, 1);
   1395 		rv = true;
   1396 	}
   1397 #endif /* _LP64 */
   1398 
   1399 	return rv;
   1400 }
   1401 
   1402 /*
   1403  * pciconf_resource_init:
   1404  *
   1405  *	Allocate and initialize a pci configuration resources container.
   1406  */
   1407 struct pciconf_resources *
   1408 pciconf_resource_init(void)
   1409 {
   1410 	struct pciconf_resources *rs;
   1411 
   1412 	rs = kmem_zalloc(sizeof(*rs), KM_SLEEP);
   1413 
   1414 	return rs;
   1415 }
   1416 
   1417 /*
   1418  * pciconf_resource_fini:
   1419  *
   1420  *	Dispose of a pci configuration resources container.
   1421  */
   1422 void
   1423 pciconf_resource_fini(struct pciconf_resources *rs)
   1424 {
   1425 	int i;
   1426 
   1427 	for (i = 0; i < PCICONF_RESOURCE_NTYPES; i++) {
   1428 		fini_range_resource(&rs->resources[i]);
   1429 	}
   1430 
   1431 	kmem_free(rs, sizeof(*rs));
   1432 }
   1433 
   1434 /*
   1435  * pciconf_resource_add:
   1436  *
   1437  *	Add a pci configuration resource to a container.
   1438  */
   1439 int
   1440 pciconf_resource_add(struct pciconf_resources *rs, int type,
   1441     bus_addr_t start, bus_size_t size)
   1442 {
   1443 	bus_addr_t end = start + (size - 1);
   1444 	struct pciconf_resource *r;
   1445 	struct pciconf_resource_rsvd *rsvd;
   1446 	int error, rsvd_type, align;
   1447 	vmem_addr_t result;
   1448 	bool first;
   1449 
   1450 	if (size == 0 || end <= start)
   1451 		return EINVAL;
   1452 
   1453 	if (type < 0 || type >= PCICONF_RESOURCE_NTYPES)
   1454 		return EINVAL;
   1455 
   1456 	r = &rs->resources[type];
   1457 
   1458 	first = r->arena == NULL;
   1459 	if (first) {
   1460 		r->arena = create_vmem_arena(pciconf_resource_names[type],
   1461 		    0, 0, VM_SLEEP);
   1462 		r->min_addr = VMEM_ADDR_MAX;
   1463 		r->max_addr = VMEM_ADDR_MIN;
   1464 	}
   1465 
   1466 	error = vmem_add(r->arena, start, size, VM_SLEEP);
   1467 	if (error == 0) {
   1468 		if (start < r->min_addr)
   1469 			r->min_addr = start;
   1470 		if (end > r->max_addr)
   1471 			r->max_addr = end;
   1472 	}
   1473 
   1474 	r->total_size += size;
   1475 
   1476 	switch (type) {
   1477 	case PCICONF_RESOURCE_IO:
   1478 		rsvd_type = PCI_CONF_MAP_IO;
   1479 		align = 0x1000;
   1480 		break;
   1481 	case PCICONF_RESOURCE_MEM:
   1482 	case PCICONF_RESOURCE_PREFETCHABLE_MEM:
   1483 		rsvd_type = PCI_CONF_MAP_MEM;
   1484 		align = 0x100000;
   1485 		break;
   1486 	default:
   1487 		rsvd_type = 0;
   1488 		align = 0;
   1489 		break;
   1490 	}
   1491 
   1492 	/*
   1493 	 * Exclude reserved ranges from available resources
   1494 	 */
   1495 	LIST_FOREACH(rsvd, &pciconf_resource_reservations, next) {
   1496 		if (rsvd->type != rsvd_type)
   1497 			continue;
   1498 		/*
   1499 		 * The reserved range may not be within our resource window.
   1500 		 * That's fine, so ignore the error.
   1501 		 */
   1502 		(void)vmem_xalloc(r->arena, rsvd->size, align, 0, 0,
   1503 				  rsvd->start, rsvd->start + rsvd->size,
   1504 				  VM_BESTFIT | VM_NOSLEEP,
   1505 				  &result);
   1506 	}
   1507 
   1508 	return 0;
   1509 }
   1510 
   1511 /*
   1512  * pciconf_resource_reserve:
   1513  *
   1514  *	Mark a pci configuration resource as in-use. Devices
   1515  *	already configured to use these resources are notified
   1516  *	during resource assignment if their resources are changed.
   1517  */
   1518 void
   1519 pciconf_resource_reserve(int type, bus_addr_t start, bus_size_t size,
   1520     void (*callback)(void *, uint64_t), void *callback_arg)
   1521 {
   1522 	struct pciconf_resource_rsvd *rsvd;
   1523 
   1524 	rsvd = kmem_zalloc(sizeof(*rsvd), KM_SLEEP);
   1525 	rsvd->type = type;
   1526 	rsvd->start = start;
   1527 	rsvd->size = size;
   1528 	rsvd->callback = callback;
   1529 	rsvd->callback_arg = callback_arg;
   1530 	LIST_INSERT_HEAD(&pciconf_resource_reservations, rsvd, next);
   1531 }
   1532 
   1533 /*
   1534  * Let's configure the PCI bus.
   1535  * This consists of basically scanning for all existing devices,
   1536  * identifying their needs, and then making another pass over them
   1537  * to set:
   1538  *	1. I/O addresses
   1539  *	2. Memory addresses (Prefetchable and not)
   1540  *	3. PCI command register
   1541  *	4. The latency part of the PCI BHLC (BIST (Built-In Self Test),
   1542  *	    Header type, Latency timer, Cache line size) register
   1543  *
   1544  * The command register is set to enable fast back-to-back transactions
   1545  * if the host bridge says it can handle it.  We also configure
   1546  * Master Enable, SERR enable, parity enable, and (if this is not a
   1547  * PCI-PCI bridge) the I/O and Memory spaces.  Apparently some devices
   1548  * will not report some I/O space.
   1549  *
   1550  * The latency is computed to be a "fair share" of the bus bandwidth.
   1551  * The bus bandwidth variable is initialized to the number of PCI cycles
   1552  * in one second.  The number of cycles taken for one transaction by each
   1553  * device (MAX_LAT + MIN_GNT) is then subtracted from the bandwidth.
   1554  * Care is taken to ensure that the latency timer won't be set such that
   1555  * it would exceed the critical time for any device.
   1556  *
   1557  * This is complicated somewhat due to the presence of bridges.  PCI-PCI
   1558  * bridges are probed and configured recursively.
   1559  */
   1560 int
   1561 pci_configure_bus(pci_chipset_tag_t pc, struct pciconf_resources *rs,
   1562     int firstbus, int cacheline_size)
   1563 {
   1564 	pciconf_bus_t	*pb;
   1565 	int		rv;
   1566 
   1567 	pb = kmem_zalloc(sizeof (pciconf_bus_t), KM_SLEEP);
   1568 	pb->busno = firstbus;
   1569 	pb->next_busno = pb->busno + 1;
   1570 	pb->last_busno = 255;
   1571 	pb->cacheline_size = cacheline_size;
   1572 	pb->parent_bus = NULL;
   1573 	pb->swiz = 0;
   1574 	pb->io_32bit = 1;
   1575 	pb->io_res = rs->resources[PCICONF_RESOURCE_IO];
   1576 
   1577 	pb->mem_res = rs->resources[PCICONF_RESOURCE_MEM];
   1578 	if (pb->mem_res.arena == NULL)
   1579 		pb->mem_res = rs->resources[PCICONF_RESOURCE_PREFETCHABLE_MEM];
   1580 
   1581 	pb->pmem_res = rs->resources[PCICONF_RESOURCE_PREFETCHABLE_MEM];
   1582 	if (pb->pmem_res.arena == NULL)
   1583 		pb->pmem_res = rs->resources[PCICONF_RESOURCE_MEM];
   1584 
   1585 	/*
   1586 	 * Probe the memory region arenas to see if allocation of
   1587 	 * 64-bit addresses is possible.
   1588 	 */
   1589 	pb->mem_64bit = mem_region_ok64(&pb->mem_res);
   1590 	pb->pmem_64bit = mem_region_ok64(&pb->pmem_res);
   1591 
   1592 	pb->pc = pc;
   1593 	pb->io_total = pb->mem_total = pb->pmem_total = 0;
   1594 
   1595 	rv = probe_bus(pb);
   1596 	pb->last_busno = pb->next_busno - 1;
   1597 	if (rv == 0)
   1598 		rv = configure_bus(pb);
   1599 
   1600 	/*
   1601 	 * All done!
   1602 	 */
   1603 	kmem_free(pb, sizeof(*pb));
   1604 	return rv;
   1605 }
   1606