pci.c revision 182032
1/*-
2 * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
3 * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
4 * Copyright (c) 2000, BSDi
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice unmodified, this list of conditions, and the following
12 *    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 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/sys/dev/pci/pci.c 182032 2008-08-23 07:18:30Z imp $");
31
32#include "opt_bus.h"
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/malloc.h>
37#include <sys/module.h>
38#include <sys/linker.h>
39#include <sys/fcntl.h>
40#include <sys/conf.h>
41#include <sys/kernel.h>
42#include <sys/queue.h>
43#include <sys/sysctl.h>
44#include <sys/endian.h>
45
46#include <vm/vm.h>
47#include <vm/pmap.h>
48#include <vm/vm_extern.h>
49
50#include <sys/bus.h>
51#include <machine/bus.h>
52#include <sys/rman.h>
53#include <machine/resource.h>
54
55#if defined(__i386__) || defined(__amd64__)
56#include <machine/intr_machdep.h>
57#endif
58
59#include <sys/pciio.h>
60#include <dev/pci/pcireg.h>
61#include <dev/pci/pcivar.h>
62#include <dev/pci/pci_private.h>
63
64#include "pcib_if.h"
65#include "pci_if.h"
66
67#ifdef __HAVE_ACPI
68#include <contrib/dev/acpica/acpi.h>
69#include "acpi_if.h"
70#else
71#define	ACPI_PWR_FOR_SLEEP(x, y, z)
72#endif
73
74static uint32_t		pci_mapbase(unsigned mapreg);
75static const char	*pci_maptype(unsigned mapreg);
76static int		pci_mapsize(unsigned testval);
77static int		pci_maprange(unsigned mapreg);
78static void		pci_fixancient(pcicfgregs *cfg);
79
80static int		pci_porten(device_t pcib, int b, int s, int f);
81static int		pci_memen(device_t pcib, int b, int s, int f);
82static void		pci_assign_interrupt(device_t bus, device_t dev,
83			    int force_route);
84static int		pci_add_map(device_t pcib, device_t bus, device_t dev,
85			    int b, int s, int f, int reg,
86			    struct resource_list *rl, int force, int prefetch);
87static int		pci_probe(device_t dev);
88static int		pci_attach(device_t dev);
89static void		pci_load_vendor_data(void);
90static int		pci_describe_parse_line(char **ptr, int *vendor,
91			    int *device, char **desc);
92static char		*pci_describe_device(device_t dev);
93static int		pci_modevent(module_t mod, int what, void *arg);
94static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
95			    pcicfgregs *cfg);
96static void		pci_read_extcap(device_t pcib, pcicfgregs *cfg);
97static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
98			    int reg, uint32_t *data);
99#if 0
100static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
101			    int reg, uint32_t data);
102#endif
103static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
104static void		pci_disable_msi(device_t dev);
105static void		pci_enable_msi(device_t dev, uint64_t address,
106			    uint16_t data);
107static void		pci_enable_msix(device_t dev, u_int index,
108			    uint64_t address, uint32_t data);
109static void		pci_mask_msix(device_t dev, u_int index);
110static void		pci_unmask_msix(device_t dev, u_int index);
111static int		pci_msi_blacklisted(void);
112static void		pci_resume_msi(device_t dev);
113static void		pci_resume_msix(device_t dev);
114
115static device_method_t pci_methods[] = {
116	/* Device interface */
117	DEVMETHOD(device_probe,		pci_probe),
118	DEVMETHOD(device_attach,	pci_attach),
119	DEVMETHOD(device_detach,	bus_generic_detach),
120	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
121	DEVMETHOD(device_suspend,	pci_suspend),
122	DEVMETHOD(device_resume,	pci_resume),
123
124	/* Bus interface */
125	DEVMETHOD(bus_print_child,	pci_print_child),
126	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
127	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
128	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
129	DEVMETHOD(bus_driver_added,	pci_driver_added),
130	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
131	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
132
133	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
134	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
135	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
136	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
137	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
138	DEVMETHOD(bus_release_resource,	bus_generic_rl_release_resource),
139	DEVMETHOD(bus_activate_resource, bus_generic_activate_resource),
140	DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource),
141	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
142	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
143
144	/* PCI interface */
145	DEVMETHOD(pci_read_config,	pci_read_config_method),
146	DEVMETHOD(pci_write_config,	pci_write_config_method),
147	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
148	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
149	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
150	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
151	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
152	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
153	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
154	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
155	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
156	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
157	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
158	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
159	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
160	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
161	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
162	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
163
164	{ 0, 0 }
165};
166
167DEFINE_CLASS_0(pci, pci_driver, pci_methods, 0);
168
169static devclass_t pci_devclass;
170DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, 0);
171MODULE_VERSION(pci, 1);
172
173static char	*pci_vendordata;
174static size_t	pci_vendordata_size;
175
176
177struct pci_quirk {
178	uint32_t devid;	/* Vendor/device of the card */
179	int	type;
180#define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
181#define	PCI_QUIRK_DISABLE_MSI	2 /* MSI/MSI-X doesn't work */
182	int	arg1;
183	int	arg2;
184};
185
186struct pci_quirk pci_quirks[] = {
187	/* The Intel 82371AB and 82443MX has a map register at offset 0x90. */
188	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
189	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
190	/* As does the Serverworks OSB4 (the SMBus mapping register) */
191	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
192
193	/*
194	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
195	 * or the CMIC-SL (AKA ServerWorks GC_LE).
196	 */
197	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
198	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
199
200	/*
201	 * MSI doesn't work on earlier Intel chipsets including
202	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
203	 */
204	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
205	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
206	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
207	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
208	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
209	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
210	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
211
212	/*
213	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
214	 * bridge.
215	 */
216	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
217
218	{ 0 }
219};
220
221/* map register information */
222#define	PCI_MAPMEM	0x01	/* memory map */
223#define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
224#define	PCI_MAPPORT	0x04	/* port map */
225
226struct devlist pci_devq;
227uint32_t pci_generation;
228uint32_t pci_numdevs = 0;
229static int pcie_chipset, pcix_chipset;
230
231/* sysctl vars */
232SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
233
234static int pci_enable_io_modes = 1;
235TUNABLE_INT("hw.pci.enable_io_modes", &pci_enable_io_modes);
236SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RW,
237    &pci_enable_io_modes, 1,
238    "Enable I/O and memory bits in the config register.  Some BIOSes do not\n\
239enable these bits correctly.  We'd like to do this all the time, but there\n\
240are some peripherals that this causes problems with.");
241
242static int pci_do_power_nodriver = 0;
243TUNABLE_INT("hw.pci.do_power_nodriver", &pci_do_power_nodriver);
244SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RW,
245    &pci_do_power_nodriver, 0,
246  "Place a function into D3 state when no driver attaches to it.  0 means\n\
247disable.  1 means conservatively place devices into D3 state.  2 means\n\
248agressively place devices into D3 state.  3 means put absolutely everything\n\
249in D3 state.");
250
251static int pci_do_power_resume = 1;
252TUNABLE_INT("hw.pci.do_power_resume", &pci_do_power_resume);
253SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RW,
254    &pci_do_power_resume, 1,
255  "Transition from D3 -> D0 on resume.");
256
257static int pci_do_msi = 1;
258TUNABLE_INT("hw.pci.enable_msi", &pci_do_msi);
259SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RW, &pci_do_msi, 1,
260    "Enable support for MSI interrupts");
261
262static int pci_do_msix = 1;
263TUNABLE_INT("hw.pci.enable_msix", &pci_do_msix);
264SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RW, &pci_do_msix, 1,
265    "Enable support for MSI-X interrupts");
266
267static int pci_honor_msi_blacklist = 1;
268TUNABLE_INT("hw.pci.honor_msi_blacklist", &pci_honor_msi_blacklist);
269SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RD,
270    &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI");
271
272/* Find a device_t by bus/slot/function in domain 0 */
273
274device_t
275pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
276{
277
278	return (pci_find_dbsf(0, bus, slot, func));
279}
280
281/* Find a device_t by domain/bus/slot/function */
282
283device_t
284pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
285{
286	struct pci_devinfo *dinfo;
287
288	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
289		if ((dinfo->cfg.domain == domain) &&
290		    (dinfo->cfg.bus == bus) &&
291		    (dinfo->cfg.slot == slot) &&
292		    (dinfo->cfg.func == func)) {
293			return (dinfo->cfg.dev);
294		}
295	}
296
297	return (NULL);
298}
299
300/* Find a device_t by vendor/device ID */
301
302device_t
303pci_find_device(uint16_t vendor, uint16_t device)
304{
305	struct pci_devinfo *dinfo;
306
307	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
308		if ((dinfo->cfg.vendor == vendor) &&
309		    (dinfo->cfg.device == device)) {
310			return (dinfo->cfg.dev);
311		}
312	}
313
314	return (NULL);
315}
316
317/* return base address of memory or port map */
318
319static uint32_t
320pci_mapbase(uint32_t mapreg)
321{
322
323	if (PCI_BAR_MEM(mapreg))
324		return (mapreg & PCIM_BAR_MEM_BASE);
325	else
326		return (mapreg & PCIM_BAR_IO_BASE);
327}
328
329/* return map type of memory or port map */
330
331static const char *
332pci_maptype(unsigned mapreg)
333{
334
335	if (PCI_BAR_IO(mapreg))
336		return ("I/O Port");
337	if (mapreg & PCIM_BAR_MEM_PREFETCH)
338		return ("Prefetchable Memory");
339	return ("Memory");
340}
341
342/* return log2 of map size decoded for memory or port map */
343
344static int
345pci_mapsize(uint32_t testval)
346{
347	int ln2size;
348
349	testval = pci_mapbase(testval);
350	ln2size = 0;
351	if (testval != 0) {
352		while ((testval & 1) == 0)
353		{
354			ln2size++;
355			testval >>= 1;
356		}
357	}
358	return (ln2size);
359}
360
361/* return log2 of address range supported by map register */
362
363static int
364pci_maprange(unsigned mapreg)
365{
366	int ln2range = 0;
367
368	if (PCI_BAR_IO(mapreg))
369		ln2range = 32;
370	else
371		switch (mapreg & PCIM_BAR_MEM_TYPE) {
372		case PCIM_BAR_MEM_32:
373			ln2range = 32;
374			break;
375		case PCIM_BAR_MEM_1MB:
376			ln2range = 20;
377			break;
378		case PCIM_BAR_MEM_64:
379			ln2range = 64;
380			break;
381		}
382	return (ln2range);
383}
384
385/* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
386
387static void
388pci_fixancient(pcicfgregs *cfg)
389{
390	if (cfg->hdrtype != 0)
391		return;
392
393	/* PCI to PCI bridges use header type 1 */
394	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
395		cfg->hdrtype = 1;
396}
397
398/* extract header type specific config data */
399
400static void
401pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
402{
403#define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
404	switch (cfg->hdrtype) {
405	case 0:
406		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
407		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
408		cfg->nummaps	    = PCI_MAXMAPS_0;
409		break;
410	case 1:
411		cfg->nummaps	    = PCI_MAXMAPS_1;
412		break;
413	case 2:
414		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
415		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
416		cfg->nummaps	    = PCI_MAXMAPS_2;
417		break;
418	}
419#undef REG
420}
421
422/* read configuration header into pcicfgregs structure */
423struct pci_devinfo *
424pci_read_device(device_t pcib, int d, int b, int s, int f, size_t size)
425{
426#define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
427	pcicfgregs *cfg = NULL;
428	struct pci_devinfo *devlist_entry;
429	struct devlist *devlist_head;
430
431	devlist_head = &pci_devq;
432
433	devlist_entry = NULL;
434
435	if (REG(PCIR_DEVVENDOR, 4) != 0xfffffffful) {
436		devlist_entry = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
437		if (devlist_entry == NULL)
438			return (NULL);
439
440		cfg = &devlist_entry->cfg;
441
442		cfg->domain		= d;
443		cfg->bus		= b;
444		cfg->slot		= s;
445		cfg->func		= f;
446		cfg->vendor		= REG(PCIR_VENDOR, 2);
447		cfg->device		= REG(PCIR_DEVICE, 2);
448		cfg->cmdreg		= REG(PCIR_COMMAND, 2);
449		cfg->statreg		= REG(PCIR_STATUS, 2);
450		cfg->baseclass		= REG(PCIR_CLASS, 1);
451		cfg->subclass		= REG(PCIR_SUBCLASS, 1);
452		cfg->progif		= REG(PCIR_PROGIF, 1);
453		cfg->revid		= REG(PCIR_REVID, 1);
454		cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
455		cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
456		cfg->lattimer		= REG(PCIR_LATTIMER, 1);
457		cfg->intpin		= REG(PCIR_INTPIN, 1);
458		cfg->intline		= REG(PCIR_INTLINE, 1);
459
460		cfg->mingnt		= REG(PCIR_MINGNT, 1);
461		cfg->maxlat		= REG(PCIR_MAXLAT, 1);
462
463		cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
464		cfg->hdrtype		&= ~PCIM_MFDEV;
465
466		pci_fixancient(cfg);
467		pci_hdrtypedata(pcib, b, s, f, cfg);
468
469		if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
470			pci_read_extcap(pcib, cfg);
471
472		STAILQ_INSERT_TAIL(devlist_head, devlist_entry, pci_links);
473
474		devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
475		devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
476		devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
477		devlist_entry->conf.pc_sel.pc_func = cfg->func;
478		devlist_entry->conf.pc_hdr = cfg->hdrtype;
479
480		devlist_entry->conf.pc_subvendor = cfg->subvendor;
481		devlist_entry->conf.pc_subdevice = cfg->subdevice;
482		devlist_entry->conf.pc_vendor = cfg->vendor;
483		devlist_entry->conf.pc_device = cfg->device;
484
485		devlist_entry->conf.pc_class = cfg->baseclass;
486		devlist_entry->conf.pc_subclass = cfg->subclass;
487		devlist_entry->conf.pc_progif = cfg->progif;
488		devlist_entry->conf.pc_revid = cfg->revid;
489
490		pci_numdevs++;
491		pci_generation++;
492	}
493	return (devlist_entry);
494#undef REG
495}
496
497static void
498pci_read_extcap(device_t pcib, pcicfgregs *cfg)
499{
500#define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
501#define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
502#if defined(__i386__) || defined(__amd64__)
503	uint64_t addr;
504#endif
505	uint32_t val;
506	int	ptr, nextptr, ptrptr;
507
508	switch (cfg->hdrtype & PCIM_HDRTYPE) {
509	case 0:
510	case 1:
511		ptrptr = PCIR_CAP_PTR;
512		break;
513	case 2:
514		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
515		break;
516	default:
517		return;		/* no extended capabilities support */
518	}
519	nextptr = REG(ptrptr, 1);	/* sanity check? */
520
521	/*
522	 * Read capability entries.
523	 */
524	while (nextptr != 0) {
525		/* Sanity check */
526		if (nextptr > 255) {
527			printf("illegal PCI extended capability offset %d\n",
528			    nextptr);
529			return;
530		}
531		/* Find the next entry */
532		ptr = nextptr;
533		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
534
535		/* Process this entry */
536		switch (REG(ptr + PCICAP_ID, 1)) {
537		case PCIY_PMG:		/* PCI power management */
538			if (cfg->pp.pp_cap == 0) {
539				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
540				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
541				cfg->pp.pp_pmcsr = ptr + PCIR_POWER_PMCSR;
542				if ((nextptr - ptr) > PCIR_POWER_DATA)
543					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
544			}
545			break;
546#if defined(__i386__) || defined(__amd64__)
547		case PCIY_HT:		/* HyperTransport */
548			/* Determine HT-specific capability type. */
549			val = REG(ptr + PCIR_HT_COMMAND, 2);
550			switch (val & PCIM_HTCMD_CAP_MASK) {
551			case PCIM_HTCAP_MSI_MAPPING:
552				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
553					/* Sanity check the mapping window. */
554					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
555					    4);
556					addr <<= 32;
557					addr = REG(ptr + PCIR_HTMSI_ADDRESS_LO,
558					    4);
559					if (addr != MSI_INTEL_ADDR_BASE)
560						device_printf(pcib,
561	    "HT Bridge at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
562						    cfg->domain, cfg->bus,
563						    cfg->slot, cfg->func,
564						    (long long)addr);
565				} else
566					addr = MSI_INTEL_ADDR_BASE;
567
568				cfg->ht.ht_msimap = ptr;
569				cfg->ht.ht_msictrl = val;
570				cfg->ht.ht_msiaddr = addr;
571				break;
572			}
573			break;
574#endif
575		case PCIY_MSI:		/* PCI MSI */
576			cfg->msi.msi_location = ptr;
577			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
578			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
579						     PCIM_MSICTRL_MMC_MASK)>>1);
580			break;
581		case PCIY_MSIX:		/* PCI MSI-X */
582			cfg->msix.msix_location = ptr;
583			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
584			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
585			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
586			val = REG(ptr + PCIR_MSIX_TABLE, 4);
587			cfg->msix.msix_table_bar = PCIR_BAR(val &
588			    PCIM_MSIX_BIR_MASK);
589			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
590			val = REG(ptr + PCIR_MSIX_PBA, 4);
591			cfg->msix.msix_pba_bar = PCIR_BAR(val &
592			    PCIM_MSIX_BIR_MASK);
593			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
594			break;
595		case PCIY_VPD:		/* PCI Vital Product Data */
596			cfg->vpd.vpd_reg = ptr;
597			break;
598		case PCIY_SUBVENDOR:
599			/* Should always be true. */
600			if ((cfg->hdrtype & PCIM_HDRTYPE) == 1) {
601				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
602				cfg->subvendor = val & 0xffff;
603				cfg->subdevice = val >> 16;
604			}
605			break;
606		case PCIY_PCIX:		/* PCI-X */
607			/*
608			 * Assume we have a PCI-X chipset if we have
609			 * at least one PCI-PCI bridge with a PCI-X
610			 * capability.  Note that some systems with
611			 * PCI-express or HT chipsets might match on
612			 * this check as well.
613			 */
614			if ((cfg->hdrtype & PCIM_HDRTYPE) == 1)
615				pcix_chipset = 1;
616			break;
617		case PCIY_EXPRESS:	/* PCI-express */
618			/*
619			 * Assume we have a PCI-express chipset if we have
620			 * at least one PCI-express device.
621			 */
622			pcie_chipset = 1;
623			break;
624		default:
625			break;
626		}
627	}
628/* REG and WREG use carry through to next functions */
629}
630
631/*
632 * PCI Vital Product Data
633 */
634
635#define	PCI_VPD_TIMEOUT		1000000
636
637static int
638pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
639{
640	int count = PCI_VPD_TIMEOUT;
641
642	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
643
644	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
645
646	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
647		if (--count < 0)
648			return (ENXIO);
649		DELAY(1);	/* limit looping */
650	}
651	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
652
653	return (0);
654}
655
656#if 0
657static int
658pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
659{
660	int count = PCI_VPD_TIMEOUT;
661
662	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
663
664	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
665	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
666	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
667		if (--count < 0)
668			return (ENXIO);
669		DELAY(1);	/* limit looping */
670	}
671
672	return (0);
673}
674#endif
675
676#undef PCI_VPD_TIMEOUT
677
678struct vpd_readstate {
679	device_t	pcib;
680	pcicfgregs	*cfg;
681	uint32_t	val;
682	int		bytesinval;
683	int		off;
684	uint8_t		cksum;
685};
686
687static int
688vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
689{
690	uint32_t reg;
691	uint8_t byte;
692
693	if (vrs->bytesinval == 0) {
694		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
695			return (ENXIO);
696		vrs->val = le32toh(reg);
697		vrs->off += 4;
698		byte = vrs->val & 0xff;
699		vrs->bytesinval = 3;
700	} else {
701		vrs->val = vrs->val >> 8;
702		byte = vrs->val & 0xff;
703		vrs->bytesinval--;
704	}
705
706	vrs->cksum += byte;
707	*data = byte;
708	return (0);
709}
710
711static void
712pci_read_vpd(device_t pcib, pcicfgregs *cfg)
713{
714	struct vpd_readstate vrs;
715	int state;
716	int name;
717	int remain;
718	int i;
719	int alloc, off;		/* alloc/off for RO/W arrays */
720	int cksumvalid;
721	int dflen;
722	uint8_t byte;
723	uint8_t byte2;
724
725	/* init vpd reader */
726	vrs.bytesinval = 0;
727	vrs.off = 0;
728	vrs.pcib = pcib;
729	vrs.cfg = cfg;
730	vrs.cksum = 0;
731
732	state = 0;
733	name = remain = i = 0;	/* shut up stupid gcc */
734	alloc = off = 0;	/* shut up stupid gcc */
735	dflen = 0;		/* shut up stupid gcc */
736	cksumvalid = -1;
737	while (state >= 0) {
738		if (vpd_nextbyte(&vrs, &byte)) {
739			state = -2;
740			break;
741		}
742#if 0
743		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
744		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
745		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
746#endif
747		switch (state) {
748		case 0:		/* item name */
749			if (byte & 0x80) {
750				if (vpd_nextbyte(&vrs, &byte2)) {
751					state = -2;
752					break;
753				}
754				remain = byte2;
755				if (vpd_nextbyte(&vrs, &byte2)) {
756					state = -2;
757					break;
758				}
759				remain |= byte2 << 8;
760				if (remain > (0x7f*4 - vrs.off)) {
761					state = -1;
762					printf(
763			    "pci%d:%d:%d:%d: invalid VPD data, remain %#x\n",
764					    cfg->domain, cfg->bus, cfg->slot,
765					    cfg->func, remain);
766				}
767				name = byte & 0x7f;
768			} else {
769				remain = byte & 0x7;
770				name = (byte >> 3) & 0xf;
771			}
772			switch (name) {
773			case 0x2:	/* String */
774				cfg->vpd.vpd_ident = malloc(remain + 1,
775				    M_DEVBUF, M_WAITOK);
776				i = 0;
777				state = 1;
778				break;
779			case 0xf:	/* End */
780				state = -1;
781				break;
782			case 0x10:	/* VPD-R */
783				alloc = 8;
784				off = 0;
785				cfg->vpd.vpd_ros = malloc(alloc *
786				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
787				    M_WAITOK | M_ZERO);
788				state = 2;
789				break;
790			case 0x11:	/* VPD-W */
791				alloc = 8;
792				off = 0;
793				cfg->vpd.vpd_w = malloc(alloc *
794				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
795				    M_WAITOK | M_ZERO);
796				state = 5;
797				break;
798			default:	/* Invalid data, abort */
799				state = -1;
800				break;
801			}
802			break;
803
804		case 1:	/* Identifier String */
805			cfg->vpd.vpd_ident[i++] = byte;
806			remain--;
807			if (remain == 0)  {
808				cfg->vpd.vpd_ident[i] = '\0';
809				state = 0;
810			}
811			break;
812
813		case 2:	/* VPD-R Keyword Header */
814			if (off == alloc) {
815				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
816				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
817				    M_DEVBUF, M_WAITOK | M_ZERO);
818			}
819			cfg->vpd.vpd_ros[off].keyword[0] = byte;
820			if (vpd_nextbyte(&vrs, &byte2)) {
821				state = -2;
822				break;
823			}
824			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
825			if (vpd_nextbyte(&vrs, &byte2)) {
826				state = -2;
827				break;
828			}
829			dflen = byte2;
830			if (dflen == 0 &&
831			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
832			    2) == 0) {
833				/*
834				 * if this happens, we can't trust the rest
835				 * of the VPD.
836				 */
837				printf(
838				    "pci%d:%d:%d:%d: bad keyword length: %d\n",
839				    cfg->domain, cfg->bus, cfg->slot,
840				    cfg->func, dflen);
841				cksumvalid = 0;
842				state = -1;
843				break;
844			} else if (dflen == 0) {
845				cfg->vpd.vpd_ros[off].value = malloc(1 *
846				    sizeof(*cfg->vpd.vpd_ros[off].value),
847				    M_DEVBUF, M_WAITOK);
848				cfg->vpd.vpd_ros[off].value[0] = '\x00';
849			} else
850				cfg->vpd.vpd_ros[off].value = malloc(
851				    (dflen + 1) *
852				    sizeof(*cfg->vpd.vpd_ros[off].value),
853				    M_DEVBUF, M_WAITOK);
854			remain -= 3;
855			i = 0;
856			/* keep in sync w/ state 3's transistions */
857			if (dflen == 0 && remain == 0)
858				state = 0;
859			else if (dflen == 0)
860				state = 2;
861			else
862				state = 3;
863			break;
864
865		case 3:	/* VPD-R Keyword Value */
866			cfg->vpd.vpd_ros[off].value[i++] = byte;
867			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
868			    "RV", 2) == 0 && cksumvalid == -1) {
869				if (vrs.cksum == 0)
870					cksumvalid = 1;
871				else {
872					if (bootverbose)
873						printf(
874				"pci%d:%d:%d:%d: bad VPD cksum, remain %hhu\n",
875						    cfg->domain, cfg->bus,
876						    cfg->slot, cfg->func,
877						    vrs.cksum);
878					cksumvalid = 0;
879					state = -1;
880					break;
881				}
882			}
883			dflen--;
884			remain--;
885			/* keep in sync w/ state 2's transistions */
886			if (dflen == 0)
887				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
888			if (dflen == 0 && remain == 0) {
889				cfg->vpd.vpd_rocnt = off;
890				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
891				    off * sizeof(*cfg->vpd.vpd_ros),
892				    M_DEVBUF, M_WAITOK | M_ZERO);
893				state = 0;
894			} else if (dflen == 0)
895				state = 2;
896			break;
897
898		case 4:
899			remain--;
900			if (remain == 0)
901				state = 0;
902			break;
903
904		case 5:	/* VPD-W Keyword Header */
905			if (off == alloc) {
906				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
907				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
908				    M_DEVBUF, M_WAITOK | M_ZERO);
909			}
910			cfg->vpd.vpd_w[off].keyword[0] = byte;
911			if (vpd_nextbyte(&vrs, &byte2)) {
912				state = -2;
913				break;
914			}
915			cfg->vpd.vpd_w[off].keyword[1] = byte2;
916			if (vpd_nextbyte(&vrs, &byte2)) {
917				state = -2;
918				break;
919			}
920			cfg->vpd.vpd_w[off].len = dflen = byte2;
921			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
922			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
923			    sizeof(*cfg->vpd.vpd_w[off].value),
924			    M_DEVBUF, M_WAITOK);
925			remain -= 3;
926			i = 0;
927			/* keep in sync w/ state 6's transistions */
928			if (dflen == 0 && remain == 0)
929				state = 0;
930			else if (dflen == 0)
931				state = 5;
932			else
933				state = 6;
934			break;
935
936		case 6:	/* VPD-W Keyword Value */
937			cfg->vpd.vpd_w[off].value[i++] = byte;
938			dflen--;
939			remain--;
940			/* keep in sync w/ state 5's transistions */
941			if (dflen == 0)
942				cfg->vpd.vpd_w[off++].value[i++] = '\0';
943			if (dflen == 0 && remain == 0) {
944				cfg->vpd.vpd_wcnt = off;
945				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
946				    off * sizeof(*cfg->vpd.vpd_w),
947				    M_DEVBUF, M_WAITOK | M_ZERO);
948				state = 0;
949			} else if (dflen == 0)
950				state = 5;
951			break;
952
953		default:
954			printf("pci%d:%d:%d:%d: invalid state: %d\n",
955			    cfg->domain, cfg->bus, cfg->slot, cfg->func,
956			    state);
957			state = -1;
958			break;
959		}
960	}
961
962	if (cksumvalid == 0 || state < -1) {
963		/* read-only data bad, clean up */
964		if (cfg->vpd.vpd_ros != NULL) {
965			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
966				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
967			free(cfg->vpd.vpd_ros, M_DEVBUF);
968			cfg->vpd.vpd_ros = NULL;
969		}
970	}
971	if (state < -1) {
972		/* I/O error, clean up */
973		printf("pci%d:%d:%d:%d: failed to read VPD data.\n",
974		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
975		if (cfg->vpd.vpd_ident != NULL) {
976			free(cfg->vpd.vpd_ident, M_DEVBUF);
977			cfg->vpd.vpd_ident = NULL;
978		}
979		if (cfg->vpd.vpd_w != NULL) {
980			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
981				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
982			free(cfg->vpd.vpd_w, M_DEVBUF);
983			cfg->vpd.vpd_w = NULL;
984		}
985	}
986	cfg->vpd.vpd_cached = 1;
987#undef REG
988#undef WREG
989}
990
991int
992pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
993{
994	struct pci_devinfo *dinfo = device_get_ivars(child);
995	pcicfgregs *cfg = &dinfo->cfg;
996
997	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
998		pci_read_vpd(device_get_parent(dev), cfg);
999
1000	*identptr = cfg->vpd.vpd_ident;
1001
1002	if (*identptr == NULL)
1003		return (ENXIO);
1004
1005	return (0);
1006}
1007
1008int
1009pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1010	const char **vptr)
1011{
1012	struct pci_devinfo *dinfo = device_get_ivars(child);
1013	pcicfgregs *cfg = &dinfo->cfg;
1014	int i;
1015
1016	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1017		pci_read_vpd(device_get_parent(dev), cfg);
1018
1019	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1020		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1021		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1022			*vptr = cfg->vpd.vpd_ros[i].value;
1023		}
1024
1025	if (i != cfg->vpd.vpd_rocnt)
1026		return (0);
1027
1028	*vptr = NULL;
1029	return (ENXIO);
1030}
1031
1032/*
1033 * Return the offset in configuration space of the requested extended
1034 * capability entry or 0 if the specified capability was not found.
1035 */
1036int
1037pci_find_extcap_method(device_t dev, device_t child, int capability,
1038    int *capreg)
1039{
1040	struct pci_devinfo *dinfo = device_get_ivars(child);
1041	pcicfgregs *cfg = &dinfo->cfg;
1042	u_int32_t status;
1043	u_int8_t ptr;
1044
1045	/*
1046	 * Check the CAP_LIST bit of the PCI status register first.
1047	 */
1048	status = pci_read_config(child, PCIR_STATUS, 2);
1049	if (!(status & PCIM_STATUS_CAPPRESENT))
1050		return (ENXIO);
1051
1052	/*
1053	 * Determine the start pointer of the capabilities list.
1054	 */
1055	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1056	case 0:
1057	case 1:
1058		ptr = PCIR_CAP_PTR;
1059		break;
1060	case 2:
1061		ptr = PCIR_CAP_PTR_2;
1062		break;
1063	default:
1064		/* XXX: panic? */
1065		return (ENXIO);		/* no extended capabilities support */
1066	}
1067	ptr = pci_read_config(child, ptr, 1);
1068
1069	/*
1070	 * Traverse the capabilities list.
1071	 */
1072	while (ptr != 0) {
1073		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1074			if (capreg != NULL)
1075				*capreg = ptr;
1076			return (0);
1077		}
1078		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1079	}
1080
1081	return (ENOENT);
1082}
1083
1084/*
1085 * Support for MSI-X message interrupts.
1086 */
1087void
1088pci_enable_msix(device_t dev, u_int index, uint64_t address, uint32_t data)
1089{
1090	struct pci_devinfo *dinfo = device_get_ivars(dev);
1091	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1092	uint32_t offset;
1093
1094	KASSERT(msix->msix_table_len > index, ("bogus index"));
1095	offset = msix->msix_table_offset + index * 16;
1096	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1097	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1098	bus_write_4(msix->msix_table_res, offset + 8, data);
1099
1100	/* Enable MSI -> HT mapping. */
1101	pci_ht_map_msi(dev, address);
1102}
1103
1104void
1105pci_mask_msix(device_t dev, u_int index)
1106{
1107	struct pci_devinfo *dinfo = device_get_ivars(dev);
1108	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1109	uint32_t offset, val;
1110
1111	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1112	offset = msix->msix_table_offset + index * 16 + 12;
1113	val = bus_read_4(msix->msix_table_res, offset);
1114	if (!(val & PCIM_MSIX_VCTRL_MASK)) {
1115		val |= PCIM_MSIX_VCTRL_MASK;
1116		bus_write_4(msix->msix_table_res, offset, val);
1117	}
1118}
1119
1120void
1121pci_unmask_msix(device_t dev, u_int index)
1122{
1123	struct pci_devinfo *dinfo = device_get_ivars(dev);
1124	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1125	uint32_t offset, val;
1126
1127	KASSERT(msix->msix_table_len > index, ("bogus index"));
1128	offset = msix->msix_table_offset + index * 16 + 12;
1129	val = bus_read_4(msix->msix_table_res, offset);
1130	if (val & PCIM_MSIX_VCTRL_MASK) {
1131		val &= ~PCIM_MSIX_VCTRL_MASK;
1132		bus_write_4(msix->msix_table_res, offset, val);
1133	}
1134}
1135
1136int
1137pci_pending_msix(device_t dev, u_int index)
1138{
1139	struct pci_devinfo *dinfo = device_get_ivars(dev);
1140	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1141	uint32_t offset, bit;
1142
1143	KASSERT(msix->msix_table_len > index, ("bogus index"));
1144	offset = msix->msix_pba_offset + (index / 32) * 4;
1145	bit = 1 << index % 32;
1146	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1147}
1148
1149/*
1150 * Restore MSI-X registers and table during resume.  If MSI-X is
1151 * enabled then walk the virtual table to restore the actual MSI-X
1152 * table.
1153 */
1154static void
1155pci_resume_msix(device_t dev)
1156{
1157	struct pci_devinfo *dinfo = device_get_ivars(dev);
1158	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1159	struct msix_table_entry *mte;
1160	struct msix_vector *mv;
1161	int i;
1162
1163	if (msix->msix_alloc > 0) {
1164		/* First, mask all vectors. */
1165		for (i = 0; i < msix->msix_msgnum; i++)
1166			pci_mask_msix(dev, i);
1167
1168		/* Second, program any messages with at least one handler. */
1169		for (i = 0; i < msix->msix_table_len; i++) {
1170			mte = &msix->msix_table[i];
1171			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1172				continue;
1173			mv = &msix->msix_vectors[mte->mte_vector - 1];
1174			pci_enable_msix(dev, i, mv->mv_address, mv->mv_data);
1175			pci_unmask_msix(dev, i);
1176		}
1177	}
1178	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1179	    msix->msix_ctrl, 2);
1180}
1181
1182/*
1183 * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1184 * returned in *count.  After this function returns, each message will be
1185 * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1186 */
1187int
1188pci_alloc_msix_method(device_t dev, device_t child, int *count)
1189{
1190	struct pci_devinfo *dinfo = device_get_ivars(child);
1191	pcicfgregs *cfg = &dinfo->cfg;
1192	struct resource_list_entry *rle;
1193	int actual, error, i, irq, max;
1194
1195	/* Don't let count == 0 get us into trouble. */
1196	if (*count == 0)
1197		return (EINVAL);
1198
1199	/* If rid 0 is allocated, then fail. */
1200	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1201	if (rle != NULL && rle->res != NULL)
1202		return (ENXIO);
1203
1204	/* Already have allocated messages? */
1205	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1206		return (ENXIO);
1207
1208	/* If MSI is blacklisted for this system, fail. */
1209	if (pci_msi_blacklisted())
1210		return (ENXIO);
1211
1212	/* MSI-X capability present? */
1213	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1214		return (ENODEV);
1215
1216	/* Make sure the appropriate BARs are mapped. */
1217	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1218	    cfg->msix.msix_table_bar);
1219	if (rle == NULL || rle->res == NULL ||
1220	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1221		return (ENXIO);
1222	cfg->msix.msix_table_res = rle->res;
1223	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1224		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1225		    cfg->msix.msix_pba_bar);
1226		if (rle == NULL || rle->res == NULL ||
1227		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1228			return (ENXIO);
1229	}
1230	cfg->msix.msix_pba_res = rle->res;
1231
1232	if (bootverbose)
1233		device_printf(child,
1234		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1235		    *count, cfg->msix.msix_msgnum);
1236	max = min(*count, cfg->msix.msix_msgnum);
1237	for (i = 0; i < max; i++) {
1238		/* Allocate a message. */
1239		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1240		if (error)
1241			break;
1242		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1243		    irq, 1);
1244	}
1245	actual = i;
1246
1247	if (bootverbose) {
1248		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1249		if (actual == 1)
1250			device_printf(child, "using IRQ %lu for MSI-X\n",
1251			    rle->start);
1252		else {
1253			int run;
1254
1255			/*
1256			 * Be fancy and try to print contiguous runs of
1257			 * IRQ values as ranges.  'irq' is the previous IRQ.
1258			 * 'run' is true if we are in a range.
1259			 */
1260			device_printf(child, "using IRQs %lu", rle->start);
1261			irq = rle->start;
1262			run = 0;
1263			for (i = 1; i < actual; i++) {
1264				rle = resource_list_find(&dinfo->resources,
1265				    SYS_RES_IRQ, i + 1);
1266
1267				/* Still in a run? */
1268				if (rle->start == irq + 1) {
1269					run = 1;
1270					irq++;
1271					continue;
1272				}
1273
1274				/* Finish previous range. */
1275				if (run) {
1276					printf("-%d", irq);
1277					run = 0;
1278				}
1279
1280				/* Start new range. */
1281				printf(",%lu", rle->start);
1282				irq = rle->start;
1283			}
1284
1285			/* Unfinished range? */
1286			if (run)
1287				printf("-%d", irq);
1288			printf(" for MSI-X\n");
1289		}
1290	}
1291
1292	/* Mask all vectors. */
1293	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1294		pci_mask_msix(child, i);
1295
1296	/* Allocate and initialize vector data and virtual table. */
1297	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1298	    M_DEVBUF, M_WAITOK | M_ZERO);
1299	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1300	    M_DEVBUF, M_WAITOK | M_ZERO);
1301	for (i = 0; i < actual; i++) {
1302		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1303		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1304		cfg->msix.msix_table[i].mte_vector = i + 1;
1305	}
1306
1307	/* Update control register to enable MSI-X. */
1308	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1309	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1310	    cfg->msix.msix_ctrl, 2);
1311
1312	/* Update counts of alloc'd messages. */
1313	cfg->msix.msix_alloc = actual;
1314	cfg->msix.msix_table_len = actual;
1315	*count = actual;
1316	return (0);
1317}
1318
1319/*
1320 * By default, pci_alloc_msix() will assign the allocated IRQ
1321 * resources consecutively to the first N messages in the MSI-X table.
1322 * However, device drivers may want to use different layouts if they
1323 * either receive fewer messages than they asked for, or they wish to
1324 * populate the MSI-X table sparsely.  This method allows the driver
1325 * to specify what layout it wants.  It must be called after a
1326 * successful pci_alloc_msix() but before any of the associated
1327 * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1328 *
1329 * The 'vectors' array contains 'count' message vectors.  The array
1330 * maps directly to the MSI-X table in that index 0 in the array
1331 * specifies the vector for the first message in the MSI-X table, etc.
1332 * The vector value in each array index can either be 0 to indicate
1333 * that no vector should be assigned to a message slot, or it can be a
1334 * number from 1 to N (where N is the count returned from a
1335 * succcessful call to pci_alloc_msix()) to indicate which message
1336 * vector (IRQ) to be used for the corresponding message.
1337 *
1338 * On successful return, each message with a non-zero vector will have
1339 * an associated SYS_RES_IRQ whose rid is equal to the array index +
1340 * 1.  Additionally, if any of the IRQs allocated via the previous
1341 * call to pci_alloc_msix() are not used in the mapping, those IRQs
1342 * will be freed back to the system automatically.
1343 *
1344 * For example, suppose a driver has a MSI-X table with 6 messages and
1345 * asks for 6 messages, but pci_alloc_msix() only returns a count of
1346 * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1347 * C.  After the call to pci_alloc_msix(), the device will be setup to
1348 * have an MSI-X table of ABC--- (where - means no vector assigned).
1349 * If the driver ten passes a vector array of { 1, 0, 1, 2, 0, 2 },
1350 * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1351 * be freed back to the system.  This device will also have valid
1352 * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1353 *
1354 * In any case, the SYS_RES_IRQ rid X will always map to the message
1355 * at MSI-X table index X - 1 and will only be valid if a vector is
1356 * assigned to that table entry.
1357 */
1358int
1359pci_remap_msix_method(device_t dev, device_t child, int count,
1360    const u_int *vectors)
1361{
1362	struct pci_devinfo *dinfo = device_get_ivars(child);
1363	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1364	struct resource_list_entry *rle;
1365	int i, irq, j, *used;
1366
1367	/*
1368	 * Have to have at least one message in the table but the
1369	 * table can't be bigger than the actual MSI-X table in the
1370	 * device.
1371	 */
1372	if (count == 0 || count > msix->msix_msgnum)
1373		return (EINVAL);
1374
1375	/* Sanity check the vectors. */
1376	for (i = 0; i < count; i++)
1377		if (vectors[i] > msix->msix_alloc)
1378			return (EINVAL);
1379
1380	/*
1381	 * Make sure there aren't any holes in the vectors to be used.
1382	 * It's a big pain to support it, and it doesn't really make
1383	 * sense anyway.  Also, at least one vector must be used.
1384	 */
1385	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1386	    M_ZERO);
1387	for (i = 0; i < count; i++)
1388		if (vectors[i] != 0)
1389			used[vectors[i] - 1] = 1;
1390	for (i = 0; i < msix->msix_alloc - 1; i++)
1391		if (used[i] == 0 && used[i + 1] == 1) {
1392			free(used, M_DEVBUF);
1393			return (EINVAL);
1394		}
1395	if (used[0] != 1) {
1396		free(used, M_DEVBUF);
1397		return (EINVAL);
1398	}
1399
1400	/* Make sure none of the resources are allocated. */
1401	for (i = 0; i < msix->msix_table_len; i++) {
1402		if (msix->msix_table[i].mte_vector == 0)
1403			continue;
1404		if (msix->msix_table[i].mte_handlers > 0)
1405			return (EBUSY);
1406		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1407		KASSERT(rle != NULL, ("missing resource"));
1408		if (rle->res != NULL)
1409			return (EBUSY);
1410	}
1411
1412	/* Free the existing resource list entries. */
1413	for (i = 0; i < msix->msix_table_len; i++) {
1414		if (msix->msix_table[i].mte_vector == 0)
1415			continue;
1416		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1417	}
1418
1419	/*
1420	 * Build the new virtual table keeping track of which vectors are
1421	 * used.
1422	 */
1423	free(msix->msix_table, M_DEVBUF);
1424	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1425	    M_DEVBUF, M_WAITOK | M_ZERO);
1426	for (i = 0; i < count; i++)
1427		msix->msix_table[i].mte_vector = vectors[i];
1428	msix->msix_table_len = count;
1429
1430	/* Free any unused IRQs and resize the vectors array if necessary. */
1431	j = msix->msix_alloc - 1;
1432	if (used[j] == 0) {
1433		struct msix_vector *vec;
1434
1435		while (used[j] == 0) {
1436			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
1437			    msix->msix_vectors[j].mv_irq);
1438			j--;
1439		}
1440		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
1441		    M_WAITOK);
1442		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
1443		    (j + 1));
1444		free(msix->msix_vectors, M_DEVBUF);
1445		msix->msix_vectors = vec;
1446		msix->msix_alloc = j + 1;
1447	}
1448	free(used, M_DEVBUF);
1449
1450	/* Map the IRQs onto the rids. */
1451	for (i = 0; i < count; i++) {
1452		if (vectors[i] == 0)
1453			continue;
1454		irq = msix->msix_vectors[vectors[i]].mv_irq;
1455		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1456		    irq, 1);
1457	}
1458
1459	if (bootverbose) {
1460		device_printf(child, "Remapped MSI-X IRQs as: ");
1461		for (i = 0; i < count; i++) {
1462			if (i != 0)
1463				printf(", ");
1464			if (vectors[i] == 0)
1465				printf("---");
1466			else
1467				printf("%d",
1468				    msix->msix_vectors[vectors[i]].mv_irq);
1469		}
1470		printf("\n");
1471	}
1472
1473	return (0);
1474}
1475
1476static int
1477pci_release_msix(device_t dev, device_t child)
1478{
1479	struct pci_devinfo *dinfo = device_get_ivars(child);
1480	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1481	struct resource_list_entry *rle;
1482	int i;
1483
1484	/* Do we have any messages to release? */
1485	if (msix->msix_alloc == 0)
1486		return (ENODEV);
1487
1488	/* Make sure none of the resources are allocated. */
1489	for (i = 0; i < msix->msix_table_len; i++) {
1490		if (msix->msix_table[i].mte_vector == 0)
1491			continue;
1492		if (msix->msix_table[i].mte_handlers > 0)
1493			return (EBUSY);
1494		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1495		KASSERT(rle != NULL, ("missing resource"));
1496		if (rle->res != NULL)
1497			return (EBUSY);
1498	}
1499
1500	/* Update control register to disable MSI-X. */
1501	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
1502	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
1503	    msix->msix_ctrl, 2);
1504
1505	/* Free the resource list entries. */
1506	for (i = 0; i < msix->msix_table_len; i++) {
1507		if (msix->msix_table[i].mte_vector == 0)
1508			continue;
1509		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1510	}
1511	free(msix->msix_table, M_DEVBUF);
1512	msix->msix_table_len = 0;
1513
1514	/* Release the IRQs. */
1515	for (i = 0; i < msix->msix_alloc; i++)
1516		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
1517		    msix->msix_vectors[i].mv_irq);
1518	free(msix->msix_vectors, M_DEVBUF);
1519	msix->msix_alloc = 0;
1520	return (0);
1521}
1522
1523/*
1524 * Return the max supported MSI-X messages this device supports.
1525 * Basically, assuming the MD code can alloc messages, this function
1526 * should return the maximum value that pci_alloc_msix() can return.
1527 * Thus, it is subject to the tunables, etc.
1528 */
1529int
1530pci_msix_count_method(device_t dev, device_t child)
1531{
1532	struct pci_devinfo *dinfo = device_get_ivars(child);
1533	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1534
1535	if (pci_do_msix && msix->msix_location != 0)
1536		return (msix->msix_msgnum);
1537	return (0);
1538}
1539
1540/*
1541 * HyperTransport MSI mapping control
1542 */
1543void
1544pci_ht_map_msi(device_t dev, uint64_t addr)
1545{
1546	struct pci_devinfo *dinfo = device_get_ivars(dev);
1547	struct pcicfg_ht *ht = &dinfo->cfg.ht;
1548
1549	if (!ht->ht_msimap)
1550		return;
1551
1552	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
1553	    ht->ht_msiaddr >> 20 == addr >> 20) {
1554		/* Enable MSI -> HT mapping. */
1555		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1556		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
1557		    ht->ht_msictrl, 2);
1558	}
1559
1560	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
1561		/* Disable MSI -> HT mapping. */
1562		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
1563		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
1564		    ht->ht_msictrl, 2);
1565	}
1566}
1567
1568/*
1569 * Support for MSI message signalled interrupts.
1570 */
1571void
1572pci_enable_msi(device_t dev, uint64_t address, uint16_t data)
1573{
1574	struct pci_devinfo *dinfo = device_get_ivars(dev);
1575	struct pcicfg_msi *msi = &dinfo->cfg.msi;
1576
1577	/* Write data and address values. */
1578	pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
1579	    address & 0xffffffff, 4);
1580	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
1581		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR_HIGH,
1582		    address >> 32, 4);
1583		pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA_64BIT,
1584		    data, 2);
1585	} else
1586		pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA, data,
1587		    2);
1588
1589	/* Enable MSI in the control register. */
1590	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
1591	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
1592	    2);
1593
1594	/* Enable MSI -> HT mapping. */
1595	pci_ht_map_msi(dev, address);
1596}
1597
1598void
1599pci_disable_msi(device_t dev)
1600{
1601	struct pci_devinfo *dinfo = device_get_ivars(dev);
1602	struct pcicfg_msi *msi = &dinfo->cfg.msi;
1603
1604	/* Disable MSI -> HT mapping. */
1605	pci_ht_map_msi(dev, 0);
1606
1607	/* Disable MSI in the control register. */
1608	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
1609	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
1610	    2);
1611}
1612
1613/*
1614 * Restore MSI registers during resume.  If MSI is enabled then
1615 * restore the data and address registers in addition to the control
1616 * register.
1617 */
1618static void
1619pci_resume_msi(device_t dev)
1620{
1621	struct pci_devinfo *dinfo = device_get_ivars(dev);
1622	struct pcicfg_msi *msi = &dinfo->cfg.msi;
1623	uint64_t address;
1624	uint16_t data;
1625
1626	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
1627		address = msi->msi_addr;
1628		data = msi->msi_data;
1629		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
1630		    address & 0xffffffff, 4);
1631		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
1632			pci_write_config(dev, msi->msi_location +
1633			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
1634			pci_write_config(dev, msi->msi_location +
1635			    PCIR_MSI_DATA_64BIT, data, 2);
1636		} else
1637			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
1638			    data, 2);
1639	}
1640	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
1641	    2);
1642}
1643
1644int
1645pci_remap_msi_irq(device_t dev, u_int irq)
1646{
1647	struct pci_devinfo *dinfo = device_get_ivars(dev);
1648	pcicfgregs *cfg = &dinfo->cfg;
1649	struct resource_list_entry *rle;
1650	struct msix_table_entry *mte;
1651	struct msix_vector *mv;
1652	device_t bus;
1653	uint64_t addr;
1654	uint32_t data;
1655	int error, i, j;
1656
1657	bus = device_get_parent(dev);
1658
1659	/*
1660	 * Handle MSI first.  We try to find this IRQ among our list
1661	 * of MSI IRQs.  If we find it, we request updated address and
1662	 * data registers and apply the results.
1663	 */
1664	if (cfg->msi.msi_alloc > 0) {
1665
1666		/* If we don't have any active handlers, nothing to do. */
1667		if (cfg->msi.msi_handlers == 0)
1668			return (0);
1669		for (i = 0; i < cfg->msi.msi_alloc; i++) {
1670			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
1671			    i + 1);
1672			if (rle->start == irq) {
1673				error = PCIB_MAP_MSI(device_get_parent(bus),
1674				    dev, irq, &addr, &data);
1675				if (error)
1676					return (error);
1677				pci_disable_msi(dev);
1678				dinfo->cfg.msi.msi_addr = addr;
1679				dinfo->cfg.msi.msi_data = data;
1680				pci_enable_msi(dev, addr, data);
1681				return (0);
1682			}
1683		}
1684		return (ENOENT);
1685	}
1686
1687	/*
1688	 * For MSI-X, we check to see if we have this IRQ.  If we do,
1689	 * we request the updated mapping info.  If that works, we go
1690	 * through all the slots that use this IRQ and update them.
1691	 */
1692	if (cfg->msix.msix_alloc > 0) {
1693		for (i = 0; i < cfg->msix.msix_alloc; i++) {
1694			mv = &cfg->msix.msix_vectors[i];
1695			if (mv->mv_irq == irq) {
1696				error = PCIB_MAP_MSI(device_get_parent(bus),
1697				    dev, irq, &addr, &data);
1698				if (error)
1699					return (error);
1700				mv->mv_address = addr;
1701				mv->mv_data = data;
1702				for (j = 0; j < cfg->msix.msix_table_len; j++) {
1703					mte = &cfg->msix.msix_table[j];
1704					if (mte->mte_vector != i + 1)
1705						continue;
1706					if (mte->mte_handlers == 0)
1707						continue;
1708					pci_mask_msix(dev, j);
1709					pci_enable_msix(dev, j, addr, data);
1710					pci_unmask_msix(dev, j);
1711				}
1712			}
1713		}
1714		return (ENOENT);
1715	}
1716
1717	return (ENOENT);
1718}
1719
1720/*
1721 * Returns true if the specified device is blacklisted because MSI
1722 * doesn't work.
1723 */
1724int
1725pci_msi_device_blacklisted(device_t dev)
1726{
1727	struct pci_quirk *q;
1728
1729	if (!pci_honor_msi_blacklist)
1730		return (0);
1731
1732	for (q = &pci_quirks[0]; q->devid; q++) {
1733		if (q->devid == pci_get_devid(dev) &&
1734		    q->type == PCI_QUIRK_DISABLE_MSI)
1735			return (1);
1736	}
1737	return (0);
1738}
1739
1740/*
1741 * Determine if MSI is blacklisted globally on this sytem.  Currently,
1742 * we just check for blacklisted chipsets as represented by the
1743 * host-PCI bridge at device 0:0:0.  In the future, it may become
1744 * necessary to check other system attributes, such as the kenv values
1745 * that give the motherboard manufacturer and model number.
1746 */
1747static int
1748pci_msi_blacklisted(void)
1749{
1750	device_t dev;
1751
1752	if (!pci_honor_msi_blacklist)
1753		return (0);
1754
1755	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
1756	if (!(pcie_chipset || pcix_chipset))
1757		return (1);
1758
1759	dev = pci_find_bsf(0, 0, 0);
1760	if (dev != NULL)
1761		return (pci_msi_device_blacklisted(dev));
1762	return (0);
1763}
1764
1765/*
1766 * Attempt to allocate *count MSI messages.  The actual number allocated is
1767 * returned in *count.  After this function returns, each message will be
1768 * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
1769 */
1770int
1771pci_alloc_msi_method(device_t dev, device_t child, int *count)
1772{
1773	struct pci_devinfo *dinfo = device_get_ivars(child);
1774	pcicfgregs *cfg = &dinfo->cfg;
1775	struct resource_list_entry *rle;
1776	int actual, error, i, irqs[32];
1777	uint16_t ctrl;
1778
1779	/* Don't let count == 0 get us into trouble. */
1780	if (*count == 0)
1781		return (EINVAL);
1782
1783	/* If rid 0 is allocated, then fail. */
1784	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1785	if (rle != NULL && rle->res != NULL)
1786		return (ENXIO);
1787
1788	/* Already have allocated messages? */
1789	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1790		return (ENXIO);
1791
1792	/* If MSI is blacklisted for this system, fail. */
1793	if (pci_msi_blacklisted())
1794		return (ENXIO);
1795
1796	/* MSI capability present? */
1797	if (cfg->msi.msi_location == 0 || !pci_do_msi)
1798		return (ENODEV);
1799
1800	if (bootverbose)
1801		device_printf(child,
1802		    "attempting to allocate %d MSI vectors (%d supported)\n",
1803		    *count, cfg->msi.msi_msgnum);
1804
1805	/* Don't ask for more than the device supports. */
1806	actual = min(*count, cfg->msi.msi_msgnum);
1807
1808	/* Don't ask for more than 32 messages. */
1809	actual = min(actual, 32);
1810
1811	/* MSI requires power of 2 number of messages. */
1812	if (!powerof2(actual))
1813		return (EINVAL);
1814
1815	for (;;) {
1816		/* Try to allocate N messages. */
1817		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
1818		    cfg->msi.msi_msgnum, irqs);
1819		if (error == 0)
1820			break;
1821		if (actual == 1)
1822			return (error);
1823
1824		/* Try N / 2. */
1825		actual >>= 1;
1826	}
1827
1828	/*
1829	 * We now have N actual messages mapped onto SYS_RES_IRQ
1830	 * resources in the irqs[] array, so add new resources
1831	 * starting at rid 1.
1832	 */
1833	for (i = 0; i < actual; i++)
1834		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
1835		    irqs[i], irqs[i], 1);
1836
1837	if (bootverbose) {
1838		if (actual == 1)
1839			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
1840		else {
1841			int run;
1842
1843			/*
1844			 * Be fancy and try to print contiguous runs
1845			 * of IRQ values as ranges.  'run' is true if
1846			 * we are in a range.
1847			 */
1848			device_printf(child, "using IRQs %d", irqs[0]);
1849			run = 0;
1850			for (i = 1; i < actual; i++) {
1851
1852				/* Still in a run? */
1853				if (irqs[i] == irqs[i - 1] + 1) {
1854					run = 1;
1855					continue;
1856				}
1857
1858				/* Finish previous range. */
1859				if (run) {
1860					printf("-%d", irqs[i - 1]);
1861					run = 0;
1862				}
1863
1864				/* Start new range. */
1865				printf(",%d", irqs[i]);
1866			}
1867
1868			/* Unfinished range? */
1869			if (run)
1870				printf("-%d", irqs[actual - 1]);
1871			printf(" for MSI\n");
1872		}
1873	}
1874
1875	/* Update control register with actual count. */
1876	ctrl = cfg->msi.msi_ctrl;
1877	ctrl &= ~PCIM_MSICTRL_MME_MASK;
1878	ctrl |= (ffs(actual) - 1) << 4;
1879	cfg->msi.msi_ctrl = ctrl;
1880	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
1881
1882	/* Update counts of alloc'd messages. */
1883	cfg->msi.msi_alloc = actual;
1884	cfg->msi.msi_handlers = 0;
1885	*count = actual;
1886	return (0);
1887}
1888
1889/* Release the MSI messages associated with this device. */
1890int
1891pci_release_msi_method(device_t dev, device_t child)
1892{
1893	struct pci_devinfo *dinfo = device_get_ivars(child);
1894	struct pcicfg_msi *msi = &dinfo->cfg.msi;
1895	struct resource_list_entry *rle;
1896	int error, i, irqs[32];
1897
1898	/* Try MSI-X first. */
1899	error = pci_release_msix(dev, child);
1900	if (error != ENODEV)
1901		return (error);
1902
1903	/* Do we have any messages to release? */
1904	if (msi->msi_alloc == 0)
1905		return (ENODEV);
1906	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
1907
1908	/* Make sure none of the resources are allocated. */
1909	if (msi->msi_handlers > 0)
1910		return (EBUSY);
1911	for (i = 0; i < msi->msi_alloc; i++) {
1912		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1913		KASSERT(rle != NULL, ("missing MSI resource"));
1914		if (rle->res != NULL)
1915			return (EBUSY);
1916		irqs[i] = rle->start;
1917	}
1918
1919	/* Update control register with 0 count. */
1920	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
1921	    ("%s: MSI still enabled", __func__));
1922	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
1923	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
1924	    msi->msi_ctrl, 2);
1925
1926	/* Release the messages. */
1927	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
1928	for (i = 0; i < msi->msi_alloc; i++)
1929		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1930
1931	/* Update alloc count. */
1932	msi->msi_alloc = 0;
1933	msi->msi_addr = 0;
1934	msi->msi_data = 0;
1935	return (0);
1936}
1937
1938/*
1939 * Return the max supported MSI messages this device supports.
1940 * Basically, assuming the MD code can alloc messages, this function
1941 * should return the maximum value that pci_alloc_msi() can return.
1942 * Thus, it is subject to the tunables, etc.
1943 */
1944int
1945pci_msi_count_method(device_t dev, device_t child)
1946{
1947	struct pci_devinfo *dinfo = device_get_ivars(child);
1948	struct pcicfg_msi *msi = &dinfo->cfg.msi;
1949
1950	if (pci_do_msi && msi->msi_location != 0)
1951		return (msi->msi_msgnum);
1952	return (0);
1953}
1954
1955/* free pcicfgregs structure and all depending data structures */
1956
1957int
1958pci_freecfg(struct pci_devinfo *dinfo)
1959{
1960	struct devlist *devlist_head;
1961	int i;
1962
1963	devlist_head = &pci_devq;
1964
1965	if (dinfo->cfg.vpd.vpd_reg) {
1966		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
1967		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
1968			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
1969		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
1970		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
1971			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
1972		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
1973	}
1974	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
1975	free(dinfo, M_DEVBUF);
1976
1977	/* increment the generation count */
1978	pci_generation++;
1979
1980	/* we're losing one device */
1981	pci_numdevs--;
1982	return (0);
1983}
1984
1985/*
1986 * PCI power manangement
1987 */
1988int
1989pci_set_powerstate_method(device_t dev, device_t child, int state)
1990{
1991	struct pci_devinfo *dinfo = device_get_ivars(child);
1992	pcicfgregs *cfg = &dinfo->cfg;
1993	uint16_t status;
1994	int result, oldstate, highest, delay;
1995
1996	if (cfg->pp.pp_cap == 0)
1997		return (EOPNOTSUPP);
1998
1999	/*
2000	 * Optimize a no state change request away.  While it would be OK to
2001	 * write to the hardware in theory, some devices have shown odd
2002	 * behavior when going from D3 -> D3.
2003	 */
2004	oldstate = pci_get_powerstate(child);
2005	if (oldstate == state)
2006		return (0);
2007
2008	/*
2009	 * The PCI power management specification states that after a state
2010	 * transition between PCI power states, system software must
2011	 * guarantee a minimal delay before the function accesses the device.
2012	 * Compute the worst case delay that we need to guarantee before we
2013	 * access the device.  Many devices will be responsive much more
2014	 * quickly than this delay, but there are some that don't respond
2015	 * instantly to state changes.  Transitions to/from D3 state require
2016	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2017	 * is done below with DELAY rather than a sleeper function because
2018	 * this function can be called from contexts where we cannot sleep.
2019	 */
2020	highest = (oldstate > state) ? oldstate : state;
2021	if (highest == PCI_POWERSTATE_D3)
2022	    delay = 10000;
2023	else if (highest == PCI_POWERSTATE_D2)
2024	    delay = 200;
2025	else
2026	    delay = 0;
2027	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2028	    & ~PCIM_PSTAT_DMASK;
2029	result = 0;
2030	switch (state) {
2031	case PCI_POWERSTATE_D0:
2032		status |= PCIM_PSTAT_D0;
2033		break;
2034	case PCI_POWERSTATE_D1:
2035		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2036			return (EOPNOTSUPP);
2037		status |= PCIM_PSTAT_D1;
2038		break;
2039	case PCI_POWERSTATE_D2:
2040		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2041			return (EOPNOTSUPP);
2042		status |= PCIM_PSTAT_D2;
2043		break;
2044	case PCI_POWERSTATE_D3:
2045		status |= PCIM_PSTAT_D3;
2046		break;
2047	default:
2048		return (EINVAL);
2049	}
2050
2051	if (bootverbose)
2052		printf(
2053		    "pci%d:%d:%d:%d: Transition from D%d to D%d\n",
2054		    dinfo->cfg.domain, dinfo->cfg.bus, dinfo->cfg.slot,
2055		    dinfo->cfg.func, oldstate, state);
2056
2057	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2058	if (delay)
2059		DELAY(delay);
2060	return (0);
2061}
2062
2063int
2064pci_get_powerstate_method(device_t dev, device_t child)
2065{
2066	struct pci_devinfo *dinfo = device_get_ivars(child);
2067	pcicfgregs *cfg = &dinfo->cfg;
2068	uint16_t status;
2069	int result;
2070
2071	if (cfg->pp.pp_cap != 0) {
2072		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2073		switch (status & PCIM_PSTAT_DMASK) {
2074		case PCIM_PSTAT_D0:
2075			result = PCI_POWERSTATE_D0;
2076			break;
2077		case PCIM_PSTAT_D1:
2078			result = PCI_POWERSTATE_D1;
2079			break;
2080		case PCIM_PSTAT_D2:
2081			result = PCI_POWERSTATE_D2;
2082			break;
2083		case PCIM_PSTAT_D3:
2084			result = PCI_POWERSTATE_D3;
2085			break;
2086		default:
2087			result = PCI_POWERSTATE_UNKNOWN;
2088			break;
2089		}
2090	} else {
2091		/* No support, device is always at D0 */
2092		result = PCI_POWERSTATE_D0;
2093	}
2094	return (result);
2095}
2096
2097/*
2098 * Some convenience functions for PCI device drivers.
2099 */
2100
2101static __inline void
2102pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2103{
2104	uint16_t	command;
2105
2106	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2107	command |= bit;
2108	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2109}
2110
2111static __inline void
2112pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2113{
2114	uint16_t	command;
2115
2116	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2117	command &= ~bit;
2118	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2119}
2120
2121int
2122pci_enable_busmaster_method(device_t dev, device_t child)
2123{
2124	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2125	return (0);
2126}
2127
2128int
2129pci_disable_busmaster_method(device_t dev, device_t child)
2130{
2131	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2132	return (0);
2133}
2134
2135int
2136pci_enable_io_method(device_t dev, device_t child, int space)
2137{
2138	uint16_t command;
2139	uint16_t bit;
2140	char *error;
2141
2142	bit = 0;
2143	error = NULL;
2144
2145	switch(space) {
2146	case SYS_RES_IOPORT:
2147		bit = PCIM_CMD_PORTEN;
2148		error = "port";
2149		break;
2150	case SYS_RES_MEMORY:
2151		bit = PCIM_CMD_MEMEN;
2152		error = "memory";
2153		break;
2154	default:
2155		return (EINVAL);
2156	}
2157	pci_set_command_bit(dev, child, bit);
2158	/* Some devices seem to need a brief stall here, what do to? */
2159	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2160	if (command & bit)
2161		return (0);
2162	device_printf(child, "failed to enable %s mapping!\n", error);
2163	return (ENXIO);
2164}
2165
2166int
2167pci_disable_io_method(device_t dev, device_t child, int space)
2168{
2169	uint16_t command;
2170	uint16_t bit;
2171	char *error;
2172
2173	bit = 0;
2174	error = NULL;
2175
2176	switch(space) {
2177	case SYS_RES_IOPORT:
2178		bit = PCIM_CMD_PORTEN;
2179		error = "port";
2180		break;
2181	case SYS_RES_MEMORY:
2182		bit = PCIM_CMD_MEMEN;
2183		error = "memory";
2184		break;
2185	default:
2186		return (EINVAL);
2187	}
2188	pci_clear_command_bit(dev, child, bit);
2189	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2190	if (command & bit) {
2191		device_printf(child, "failed to disable %s mapping!\n", error);
2192		return (ENXIO);
2193	}
2194	return (0);
2195}
2196
2197/*
2198 * New style pci driver.  Parent device is either a pci-host-bridge or a
2199 * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2200 */
2201
2202void
2203pci_print_verbose(struct pci_devinfo *dinfo)
2204{
2205
2206	if (bootverbose) {
2207		pcicfgregs *cfg = &dinfo->cfg;
2208
2209		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2210		    cfg->vendor, cfg->device, cfg->revid);
2211		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2212		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2213		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2214		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2215		    cfg->mfdev);
2216		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2217		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2218		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2219		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2220		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2221		if (cfg->intpin > 0)
2222			printf("\tintpin=%c, irq=%d\n",
2223			    cfg->intpin +'a' -1, cfg->intline);
2224		if (cfg->pp.pp_cap) {
2225			uint16_t status;
2226
2227			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2228			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2229			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2230			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2231			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2232			    status & PCIM_PSTAT_DMASK);
2233		}
2234		if (cfg->msi.msi_location) {
2235			int ctrl;
2236
2237			ctrl = cfg->msi.msi_ctrl;
2238			printf("\tMSI supports %d message%s%s%s\n",
2239			    cfg->msi.msi_msgnum,
2240			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2241			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2242			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2243		}
2244		if (cfg->msix.msix_location) {
2245			printf("\tMSI-X supports %d message%s ",
2246			    cfg->msix.msix_msgnum,
2247			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2248			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2249				printf("in map 0x%x\n",
2250				    cfg->msix.msix_table_bar);
2251			else
2252				printf("in maps 0x%x and 0x%x\n",
2253				    cfg->msix.msix_table_bar,
2254				    cfg->msix.msix_pba_bar);
2255		}
2256	}
2257}
2258
2259static int
2260pci_porten(device_t pcib, int b, int s, int f)
2261{
2262	return (PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2)
2263		& PCIM_CMD_PORTEN) != 0;
2264}
2265
2266static int
2267pci_memen(device_t pcib, int b, int s, int f)
2268{
2269	return (PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2)
2270		& PCIM_CMD_MEMEN) != 0;
2271}
2272
2273/*
2274 * Add a resource based on a pci map register. Return 1 if the map
2275 * register is a 32bit map register or 2 if it is a 64bit register.
2276 */
2277static int
2278pci_add_map(device_t pcib, device_t bus, device_t dev,
2279    int b, int s, int f, int reg, struct resource_list *rl, int force,
2280    int prefetch)
2281{
2282	uint32_t map;
2283	pci_addr_t base;
2284	pci_addr_t start, end, count;
2285	uint8_t ln2size;
2286	uint8_t ln2range;
2287	uint32_t testval;
2288	uint16_t cmd;
2289	int type;
2290	int barlen;
2291	struct resource *res;
2292
2293	map = PCIB_READ_CONFIG(pcib, b, s, f, reg, 4);
2294	PCIB_WRITE_CONFIG(pcib, b, s, f, reg, 0xffffffff, 4);
2295	testval = PCIB_READ_CONFIG(pcib, b, s, f, reg, 4);
2296	PCIB_WRITE_CONFIG(pcib, b, s, f, reg, map, 4);
2297
2298	if (PCI_BAR_MEM(map))
2299		type = SYS_RES_MEMORY;
2300	else
2301		type = SYS_RES_IOPORT;
2302	ln2size = pci_mapsize(testval);
2303	ln2range = pci_maprange(testval);
2304	base = pci_mapbase(map);
2305	barlen = ln2range == 64 ? 2 : 1;
2306
2307	/*
2308	 * For I/O registers, if bottom bit is set, and the next bit up
2309	 * isn't clear, we know we have a BAR that doesn't conform to the
2310	 * spec, so ignore it.  Also, sanity check the size of the data
2311	 * areas to the type of memory involved.  Memory must be at least
2312	 * 16 bytes in size, while I/O ranges must be at least 4.
2313	 */
2314	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
2315		return (barlen);
2316	if ((type == SYS_RES_MEMORY && ln2size < 4) ||
2317	    (type == SYS_RES_IOPORT && ln2size < 2))
2318		return (barlen);
2319
2320	if (ln2range == 64)
2321		/* Read the other half of a 64bit map register */
2322		base |= (uint64_t) PCIB_READ_CONFIG(pcib, b, s, f, reg + 4, 4) << 32;
2323	if (bootverbose) {
2324		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
2325		    reg, pci_maptype(map), ln2range, (uintmax_t)base, ln2size);
2326		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f))
2327			printf(", port disabled\n");
2328		else if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f))
2329			printf(", memory disabled\n");
2330		else
2331			printf(", enabled\n");
2332	}
2333
2334	/*
2335	 * If base is 0, then we have problems.  It is best to ignore
2336	 * such entries for the moment.  These will be allocated later if
2337	 * the driver specifically requests them.  However, some
2338	 * removable busses look better when all resources are allocated,
2339	 * so allow '0' to be overriden.
2340	 *
2341	 * Similarly treat maps whose values is the same as the test value
2342	 * read back.  These maps have had all f's written to them by the
2343	 * BIOS in an attempt to disable the resources.
2344	 */
2345	if (!force && (base == 0 || map == testval))
2346		return (barlen);
2347	if ((u_long)base != base) {
2348		device_printf(bus,
2349		    "pci%d:%d:%d:%d bar %#x too many address bits",
2350		    pci_get_domain(dev), b, s, f, reg);
2351		return (barlen);
2352	}
2353
2354	/*
2355	 * This code theoretically does the right thing, but has
2356	 * undesirable side effects in some cases where peripherals
2357	 * respond oddly to having these bits enabled.  Let the user
2358	 * be able to turn them off (since pci_enable_io_modes is 1 by
2359	 * default).
2360	 */
2361	if (pci_enable_io_modes) {
2362		/* Turn on resources that have been left off by a lazy BIOS */
2363		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f)) {
2364			cmd = PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2);
2365			cmd |= PCIM_CMD_PORTEN;
2366			PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND, cmd, 2);
2367		}
2368		if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f)) {
2369			cmd = PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2);
2370			cmd |= PCIM_CMD_MEMEN;
2371			PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND, cmd, 2);
2372		}
2373	} else {
2374		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f))
2375			return (barlen);
2376		if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f))
2377			return (barlen);
2378	}
2379
2380	count = 1 << ln2size;
2381	if (base == 0 || base == pci_mapbase(testval)) {
2382		start = 0;	/* Let the parent decide. */
2383		end = ~0ULL;
2384	} else {
2385		start = base;
2386		end = base + (1 << ln2size) - 1;
2387	}
2388	resource_list_add(rl, type, reg, start, end, count);
2389
2390	/*
2391	 * Try to allocate the resource for this BAR from our parent
2392	 * so that this resource range is already reserved.  The
2393	 * driver for this device will later inherit this resource in
2394	 * pci_alloc_resource().
2395	 */
2396	res = resource_list_alloc(rl, bus, dev, type, &reg, start, end, count,
2397	    prefetch ? RF_PREFETCHABLE : 0);
2398	if (res == NULL) {
2399		/*
2400		 * If the allocation fails, clear the BAR and delete
2401		 * the resource list entry to force
2402		 * pci_alloc_resource() to allocate resources from the
2403		 * parent.
2404		 */
2405		resource_list_delete(rl, type, reg);
2406		start = 0;
2407	} else
2408		start = rman_get_start(res);
2409	pci_write_config(dev, reg, start, 4);
2410	if (ln2range == 64)
2411		pci_write_config(dev, reg + 4, start >> 32, 4);
2412	return (barlen);
2413}
2414
2415/*
2416 * For ATA devices we need to decide early what addressing mode to use.
2417 * Legacy demands that the primary and secondary ATA ports sits on the
2418 * same addresses that old ISA hardware did. This dictates that we use
2419 * those addresses and ignore the BAR's if we cannot set PCI native
2420 * addressing mode.
2421 */
2422static void
2423pci_ata_maps(device_t pcib, device_t bus, device_t dev, int b,
2424    int s, int f, struct resource_list *rl, int force, uint32_t prefetchmask)
2425{
2426	int rid, type, progif;
2427#if 0
2428	/* if this device supports PCI native addressing use it */
2429	progif = pci_read_config(dev, PCIR_PROGIF, 1);
2430	if ((progif & 0x8a) == 0x8a) {
2431		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
2432		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
2433			printf("Trying ATA native PCI addressing mode\n");
2434			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
2435		}
2436	}
2437#endif
2438	progif = pci_read_config(dev, PCIR_PROGIF, 1);
2439	type = SYS_RES_IOPORT;
2440	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
2441		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(0), rl, force,
2442		    prefetchmask & (1 << 0));
2443		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(1), rl, force,
2444		    prefetchmask & (1 << 1));
2445	} else {
2446		rid = PCIR_BAR(0);
2447		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
2448		resource_list_alloc(rl, bus, dev, type, &rid, 0x1f0, 0x1f7, 8,
2449		    0);
2450		rid = PCIR_BAR(1);
2451		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
2452		resource_list_alloc(rl, bus, dev, type, &rid, 0x3f6, 0x3f6, 1,
2453		    0);
2454	}
2455	if (progif & PCIP_STORAGE_IDE_MODESEC) {
2456		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(2), rl, force,
2457		    prefetchmask & (1 << 2));
2458		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(3), rl, force,
2459		    prefetchmask & (1 << 3));
2460	} else {
2461		rid = PCIR_BAR(2);
2462		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
2463		resource_list_alloc(rl, bus, dev, type, &rid, 0x170, 0x177, 8,
2464		    0);
2465		rid = PCIR_BAR(3);
2466		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
2467		resource_list_alloc(rl, bus, dev, type, &rid, 0x376, 0x376, 1,
2468		    0);
2469	}
2470	pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(4), rl, force,
2471	    prefetchmask & (1 << 4));
2472	pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(5), rl, force,
2473	    prefetchmask & (1 << 5));
2474}
2475
2476static void
2477pci_assign_interrupt(device_t bus, device_t dev, int force_route)
2478{
2479	struct pci_devinfo *dinfo = device_get_ivars(dev);
2480	pcicfgregs *cfg = &dinfo->cfg;
2481	char tunable_name[64];
2482	int irq;
2483
2484	/* Has to have an intpin to have an interrupt. */
2485	if (cfg->intpin == 0)
2486		return;
2487
2488	/* Let the user override the IRQ with a tunable. */
2489	irq = PCI_INVALID_IRQ;
2490	snprintf(tunable_name, sizeof(tunable_name),
2491	    "hw.pci%d.%d.%d.INT%c.irq",
2492	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
2493	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
2494		irq = PCI_INVALID_IRQ;
2495
2496	/*
2497	 * If we didn't get an IRQ via the tunable, then we either use the
2498	 * IRQ value in the intline register or we ask the bus to route an
2499	 * interrupt for us.  If force_route is true, then we only use the
2500	 * value in the intline register if the bus was unable to assign an
2501	 * IRQ.
2502	 */
2503	if (!PCI_INTERRUPT_VALID(irq)) {
2504		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
2505			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
2506		if (!PCI_INTERRUPT_VALID(irq))
2507			irq = cfg->intline;
2508	}
2509
2510	/* If after all that we don't have an IRQ, just bail. */
2511	if (!PCI_INTERRUPT_VALID(irq))
2512		return;
2513
2514	/* Update the config register if it changed. */
2515	if (irq != cfg->intline) {
2516		cfg->intline = irq;
2517		pci_write_config(dev, PCIR_INTLINE, irq, 1);
2518	}
2519
2520	/* Add this IRQ as rid 0 interrupt resource. */
2521	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
2522}
2523
2524void
2525pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
2526{
2527	device_t pcib;
2528	struct pci_devinfo *dinfo = device_get_ivars(dev);
2529	pcicfgregs *cfg = &dinfo->cfg;
2530	struct resource_list *rl = &dinfo->resources;
2531	struct pci_quirk *q;
2532	int b, i, f, s;
2533
2534	pcib = device_get_parent(bus);
2535
2536	b = cfg->bus;
2537	s = cfg->slot;
2538	f = cfg->func;
2539
2540	/* ATA devices needs special map treatment */
2541	if ((pci_get_class(dev) == PCIC_STORAGE) &&
2542	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
2543	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
2544	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
2545	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
2546		pci_ata_maps(pcib, bus, dev, b, s, f, rl, force, prefetchmask);
2547	else
2548		for (i = 0; i < cfg->nummaps;)
2549			i += pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(i),
2550			    rl, force, prefetchmask & (1 << i));
2551
2552	/*
2553	 * Add additional, quirked resources.
2554	 */
2555	for (q = &pci_quirks[0]; q->devid; q++) {
2556		if (q->devid == ((cfg->device << 16) | cfg->vendor)
2557		    && q->type == PCI_QUIRK_MAP_REG)
2558			pci_add_map(pcib, bus, dev, b, s, f, q->arg1, rl,
2559			  force, 0);
2560	}
2561
2562	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
2563#ifdef __PCI_REROUTE_INTERRUPT
2564		/*
2565		 * Try to re-route interrupts. Sometimes the BIOS or
2566		 * firmware may leave bogus values in these registers.
2567		 * If the re-route fails, then just stick with what we
2568		 * have.
2569		 */
2570		pci_assign_interrupt(bus, dev, 1);
2571#else
2572		pci_assign_interrupt(bus, dev, 0);
2573#endif
2574	}
2575}
2576
2577void
2578pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size)
2579{
2580#define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
2581	device_t pcib = device_get_parent(dev);
2582	struct pci_devinfo *dinfo;
2583	int maxslots;
2584	int s, f, pcifunchigh;
2585	uint8_t hdrtype;
2586
2587	KASSERT(dinfo_size >= sizeof(struct pci_devinfo),
2588	    ("dinfo_size too small"));
2589	maxslots = PCIB_MAXSLOTS(pcib);
2590	for (s = 0; s <= maxslots; s++) {
2591		pcifunchigh = 0;
2592		f = 0;
2593		DELAY(1);
2594		hdrtype = REG(PCIR_HDRTYPE, 1);
2595		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
2596			continue;
2597		if (hdrtype & PCIM_MFDEV)
2598			pcifunchigh = PCI_FUNCMAX;
2599		for (f = 0; f <= pcifunchigh; f++) {
2600			dinfo = pci_read_device(pcib, domain, busno, s, f,
2601			    dinfo_size);
2602			if (dinfo != NULL) {
2603				pci_add_child(dev, dinfo);
2604			}
2605		}
2606	}
2607#undef REG
2608}
2609
2610void
2611pci_add_child(device_t bus, struct pci_devinfo *dinfo)
2612{
2613	dinfo->cfg.dev = device_add_child(bus, NULL, -1);
2614	device_set_ivars(dinfo->cfg.dev, dinfo);
2615	resource_list_init(&dinfo->resources);
2616	pci_cfg_save(dinfo->cfg.dev, dinfo, 0);
2617	pci_cfg_restore(dinfo->cfg.dev, dinfo);
2618	pci_print_verbose(dinfo);
2619	pci_add_resources(bus, dinfo->cfg.dev, 0, 0);
2620}
2621
2622static int
2623pci_probe(device_t dev)
2624{
2625
2626	device_set_desc(dev, "PCI bus");
2627
2628	/* Allow other subclasses to override this driver. */
2629	return (-1000);
2630}
2631
2632static int
2633pci_attach(device_t dev)
2634{
2635	int busno, domain;
2636
2637	/*
2638	 * Since there can be multiple independantly numbered PCI
2639	 * busses on systems with multiple PCI domains, we can't use
2640	 * the unit number to decide which bus we are probing. We ask
2641	 * the parent pcib what our domain and bus numbers are.
2642	 */
2643	domain = pcib_get_domain(dev);
2644	busno = pcib_get_bus(dev);
2645	if (bootverbose)
2646		device_printf(dev, "domain=%d, physical bus=%d\n",
2647		    domain, busno);
2648	pci_add_children(dev, domain, busno, sizeof(struct pci_devinfo));
2649	return (bus_generic_attach(dev));
2650}
2651
2652int
2653pci_suspend(device_t dev)
2654{
2655	int dstate, error, i, numdevs;
2656	device_t acpi_dev, child, *devlist;
2657	struct pci_devinfo *dinfo;
2658
2659	/*
2660	 * Save the PCI configuration space for each child and set the
2661	 * device in the appropriate power state for this sleep state.
2662	 */
2663	acpi_dev = NULL;
2664	if (pci_do_power_resume)
2665		acpi_dev = devclass_get_device(devclass_find("acpi"), 0);
2666	device_get_children(dev, &devlist, &numdevs);
2667	for (i = 0; i < numdevs; i++) {
2668		child = devlist[i];
2669		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2670		pci_cfg_save(child, dinfo, 0);
2671	}
2672
2673	/* Suspend devices before potentially powering them down. */
2674	error = bus_generic_suspend(dev);
2675	if (error) {
2676		free(devlist, M_TEMP);
2677		return (error);
2678	}
2679
2680	/*
2681	 * Always set the device to D3.  If ACPI suggests a different
2682	 * power state, use it instead.  If ACPI is not present, the
2683	 * firmware is responsible for managing device power.  Skip
2684	 * children who aren't attached since they are powered down
2685	 * separately.  Only manage type 0 devices for now.
2686	 */
2687	for (i = 0; acpi_dev && i < numdevs; i++) {
2688		child = devlist[i];
2689		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2690		if (device_is_attached(child) && dinfo->cfg.hdrtype == 0) {
2691			dstate = PCI_POWERSTATE_D3;
2692			ACPI_PWR_FOR_SLEEP(acpi_dev, child, &dstate);
2693			pci_set_powerstate(child, dstate);
2694		}
2695	}
2696	free(devlist, M_TEMP);
2697	return (0);
2698}
2699
2700int
2701pci_resume(device_t dev)
2702{
2703	int i, numdevs;
2704	device_t acpi_dev, child, *devlist;
2705	struct pci_devinfo *dinfo;
2706
2707	/*
2708	 * Set each child to D0 and restore its PCI configuration space.
2709	 */
2710	acpi_dev = NULL;
2711	if (pci_do_power_resume)
2712		acpi_dev = devclass_get_device(devclass_find("acpi"), 0);
2713	device_get_children(dev, &devlist, &numdevs);
2714	for (i = 0; i < numdevs; i++) {
2715		/*
2716		 * Notify ACPI we're going to D0 but ignore the result.  If
2717		 * ACPI is not present, the firmware is responsible for
2718		 * managing device power.  Only manage type 0 devices for now.
2719		 */
2720		child = devlist[i];
2721		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2722		if (acpi_dev && device_is_attached(child) &&
2723		    dinfo->cfg.hdrtype == 0) {
2724			ACPI_PWR_FOR_SLEEP(acpi_dev, child, NULL);
2725			pci_set_powerstate(child, PCI_POWERSTATE_D0);
2726		}
2727
2728		/* Now the device is powered up, restore its config space. */
2729		pci_cfg_restore(child, dinfo);
2730	}
2731	free(devlist, M_TEMP);
2732	return (bus_generic_resume(dev));
2733}
2734
2735static void
2736pci_load_vendor_data(void)
2737{
2738	caddr_t vendordata, info;
2739
2740	if ((vendordata = preload_search_by_type("pci_vendor_data")) != NULL) {
2741		info = preload_search_info(vendordata, MODINFO_ADDR);
2742		pci_vendordata = *(char **)info;
2743		info = preload_search_info(vendordata, MODINFO_SIZE);
2744		pci_vendordata_size = *(size_t *)info;
2745		/* terminate the database */
2746		pci_vendordata[pci_vendordata_size] = '\n';
2747	}
2748}
2749
2750void
2751pci_driver_added(device_t dev, driver_t *driver)
2752{
2753	int numdevs;
2754	device_t *devlist;
2755	device_t child;
2756	struct pci_devinfo *dinfo;
2757	int i;
2758
2759	if (bootverbose)
2760		device_printf(dev, "driver added\n");
2761	DEVICE_IDENTIFY(driver, dev);
2762	device_get_children(dev, &devlist, &numdevs);
2763	for (i = 0; i < numdevs; i++) {
2764		child = devlist[i];
2765		if (device_get_state(child) != DS_NOTPRESENT)
2766			continue;
2767		dinfo = device_get_ivars(child);
2768		pci_print_verbose(dinfo);
2769		if (bootverbose)
2770			printf("pci%d:%d:%d:%d: reprobing on driver added\n",
2771			    dinfo->cfg.domain, dinfo->cfg.bus, dinfo->cfg.slot,
2772			    dinfo->cfg.func);
2773		pci_cfg_restore(child, dinfo);
2774		if (device_probe_and_attach(child) != 0)
2775			pci_cfg_save(child, dinfo, 1);
2776	}
2777	free(devlist, M_TEMP);
2778}
2779
2780int
2781pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
2782    driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
2783{
2784	struct pci_devinfo *dinfo;
2785	struct msix_table_entry *mte;
2786	struct msix_vector *mv;
2787	uint64_t addr;
2788	uint32_t data;
2789	void *cookie;
2790	int error, rid;
2791
2792	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
2793	    arg, &cookie);
2794	if (error)
2795		return (error);
2796
2797	/*
2798	 * If this is a direct child, check to see if the interrupt is
2799	 * MSI or MSI-X.  If so, ask our parent to map the MSI and give
2800	 * us the address and data register values.  If we fail for some
2801	 * reason, teardown the interrupt handler.
2802	 */
2803	rid = rman_get_rid(irq);
2804	if (device_get_parent(child) == dev && rid > 0) {
2805		dinfo = device_get_ivars(child);
2806		if (dinfo->cfg.msi.msi_alloc > 0) {
2807			if (dinfo->cfg.msi.msi_addr == 0) {
2808				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
2809			    ("MSI has handlers, but vectors not mapped"));
2810				error = PCIB_MAP_MSI(device_get_parent(dev),
2811				    child, rman_get_start(irq), &addr, &data);
2812				if (error)
2813					goto bad;
2814				dinfo->cfg.msi.msi_addr = addr;
2815				dinfo->cfg.msi.msi_data = data;
2816				pci_enable_msi(child, addr, data);
2817			}
2818			dinfo->cfg.msi.msi_handlers++;
2819		} else {
2820			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
2821			    ("No MSI or MSI-X interrupts allocated"));
2822			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
2823			    ("MSI-X index too high"));
2824			mte = &dinfo->cfg.msix.msix_table[rid - 1];
2825			KASSERT(mte->mte_vector != 0, ("no message vector"));
2826			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
2827			KASSERT(mv->mv_irq == rman_get_start(irq),
2828			    ("IRQ mismatch"));
2829			if (mv->mv_address == 0) {
2830				KASSERT(mte->mte_handlers == 0,
2831		    ("MSI-X table entry has handlers, but vector not mapped"));
2832				error = PCIB_MAP_MSI(device_get_parent(dev),
2833				    child, rman_get_start(irq), &addr, &data);
2834				if (error)
2835					goto bad;
2836				mv->mv_address = addr;
2837				mv->mv_data = data;
2838			}
2839			if (mte->mte_handlers == 0) {
2840				pci_enable_msix(child, rid - 1, mv->mv_address,
2841				    mv->mv_data);
2842				pci_unmask_msix(child, rid - 1);
2843			}
2844			mte->mte_handlers++;
2845		}
2846	bad:
2847		if (error) {
2848			(void)bus_generic_teardown_intr(dev, child, irq,
2849			    cookie);
2850			return (error);
2851		}
2852	}
2853	*cookiep = cookie;
2854	return (0);
2855}
2856
2857int
2858pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
2859    void *cookie)
2860{
2861	struct msix_table_entry *mte;
2862	struct resource_list_entry *rle;
2863	struct pci_devinfo *dinfo;
2864	int error, rid;
2865
2866	/*
2867	 * If this is a direct child, check to see if the interrupt is
2868	 * MSI or MSI-X.  If so, decrement the appropriate handlers
2869	 * count and mask the MSI-X message, or disable MSI messages
2870	 * if the count drops to 0.
2871	 */
2872	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
2873		return (EINVAL);
2874	rid = rman_get_rid(irq);
2875	if (device_get_parent(child) == dev && rid > 0) {
2876		dinfo = device_get_ivars(child);
2877		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
2878		if (rle->res != irq)
2879			return (EINVAL);
2880		if (dinfo->cfg.msi.msi_alloc > 0) {
2881			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
2882			    ("MSI-X index too high"));
2883			if (dinfo->cfg.msi.msi_handlers == 0)
2884				return (EINVAL);
2885			dinfo->cfg.msi.msi_handlers--;
2886			if (dinfo->cfg.msi.msi_handlers == 0)
2887				pci_disable_msi(child);
2888		} else {
2889			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
2890			    ("No MSI or MSI-X interrupts allocated"));
2891			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
2892			    ("MSI-X index too high"));
2893			mte = &dinfo->cfg.msix.msix_table[rid - 1];
2894			if (mte->mte_handlers == 0)
2895				return (EINVAL);
2896			mte->mte_handlers--;
2897			if (mte->mte_handlers == 0)
2898				pci_mask_msix(child, rid - 1);
2899		}
2900	}
2901	error = bus_generic_teardown_intr(dev, child, irq, cookie);
2902	if (device_get_parent(child) == dev && rid > 0)
2903		KASSERT(error == 0,
2904		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
2905	return (error);
2906}
2907
2908int
2909pci_print_child(device_t dev, device_t child)
2910{
2911	struct pci_devinfo *dinfo;
2912	struct resource_list *rl;
2913	int retval = 0;
2914
2915	dinfo = device_get_ivars(child);
2916	rl = &dinfo->resources;
2917
2918	retval += bus_print_child_header(dev, child);
2919
2920	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#lx");
2921	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#lx");
2922	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%ld");
2923	if (device_get_flags(dev))
2924		retval += printf(" flags %#x", device_get_flags(dev));
2925
2926	retval += printf(" at device %d.%d", pci_get_slot(child),
2927	    pci_get_function(child));
2928
2929	retval += bus_print_child_footer(dev, child);
2930
2931	return (retval);
2932}
2933
2934static struct
2935{
2936	int	class;
2937	int	subclass;
2938	char	*desc;
2939} pci_nomatch_tab[] = {
2940	{PCIC_OLD,		-1,			"old"},
2941	{PCIC_OLD,		PCIS_OLD_NONVGA,	"non-VGA display device"},
2942	{PCIC_OLD,		PCIS_OLD_VGA,		"VGA-compatible display device"},
2943	{PCIC_STORAGE,		-1,			"mass storage"},
2944	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	"SCSI"},
2945	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	"ATA"},
2946	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	"floppy disk"},
2947	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	"IPI"},
2948	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	"RAID"},
2949	{PCIC_NETWORK,		-1,			"network"},
2950	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	"ethernet"},
2951	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	"token ring"},
2952	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	"fddi"},
2953	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	"ATM"},
2954	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	"ISDN"},
2955	{PCIC_DISPLAY,		-1,			"display"},
2956	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	"VGA"},
2957	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	"XGA"},
2958	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	"3D"},
2959	{PCIC_MULTIMEDIA,	-1,			"multimedia"},
2960	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	"video"},
2961	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	"audio"},
2962	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	"telephony"},
2963	{PCIC_MEMORY,		-1,			"memory"},
2964	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	"RAM"},
2965	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	"flash"},
2966	{PCIC_BRIDGE,		-1,			"bridge"},
2967	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	"HOST-PCI"},
2968	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	"PCI-ISA"},
2969	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	"PCI-EISA"},
2970	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	"PCI-MCA"},
2971	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	"PCI-PCI"},
2972	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	"PCI-PCMCIA"},
2973	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	"PCI-NuBus"},
2974	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	"PCI-CardBus"},
2975	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	"PCI-RACEway"},
2976	{PCIC_SIMPLECOMM,	-1,			"simple comms"},
2977	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	"UART"},	/* could detect 16550 */
2978	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	"parallel port"},
2979	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	"multiport serial"},
2980	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	"generic modem"},
2981	{PCIC_BASEPERIPH,	-1,			"base peripheral"},
2982	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	"interrupt controller"},
2983	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	"DMA controller"},
2984	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	"timer"},
2985	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	"realtime clock"},
2986	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	"PCI hot-plug controller"},
2987	{PCIC_INPUTDEV,		-1,			"input device"},
2988	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	"keyboard"},
2989	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,"digitizer"},
2990	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	"mouse"},
2991	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	"scanner"},
2992	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	"gameport"},
2993	{PCIC_DOCKING,		-1,			"docking station"},
2994	{PCIC_PROCESSOR,	-1,			"processor"},
2995	{PCIC_SERIALBUS,	-1,			"serial bus"},
2996	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	"FireWire"},
2997	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	"AccessBus"},
2998	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	"SSA"},
2999	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	"USB"},
3000	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	"Fibre Channel"},
3001	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	"SMBus"},
3002	{PCIC_WIRELESS,		-1,			"wireless controller"},
3003	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	"iRDA"},
3004	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	"IR"},
3005	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	"RF"},
3006	{PCIC_INTELLIIO,	-1,			"intelligent I/O controller"},
3007	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	"I2O"},
3008	{PCIC_SATCOM,		-1,			"satellite communication"},
3009	{PCIC_SATCOM,		PCIS_SATCOM_TV,		"sat TV"},
3010	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	"sat audio"},
3011	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	"sat voice"},
3012	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	"sat data"},
3013	{PCIC_CRYPTO,		-1,			"encrypt/decrypt"},
3014	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	"network/computer crypto"},
3015	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	"entertainment crypto"},
3016	{PCIC_DASP,		-1,			"dasp"},
3017	{PCIC_DASP,		PCIS_DASP_DPIO,		"DPIO module"},
3018	{0, 0,		NULL}
3019};
3020
3021void
3022pci_probe_nomatch(device_t dev, device_t child)
3023{
3024	int	i;
3025	char	*cp, *scp, *device;
3026
3027	/*
3028	 * Look for a listing for this device in a loaded device database.
3029	 */
3030	if ((device = pci_describe_device(child)) != NULL) {
3031		device_printf(dev, "<%s>", device);
3032		free(device, M_DEVBUF);
3033	} else {
3034		/*
3035		 * Scan the class/subclass descriptions for a general
3036		 * description.
3037		 */
3038		cp = "unknown";
3039		scp = NULL;
3040		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
3041			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
3042				if (pci_nomatch_tab[i].subclass == -1) {
3043					cp = pci_nomatch_tab[i].desc;
3044				} else if (pci_nomatch_tab[i].subclass ==
3045				    pci_get_subclass(child)) {
3046					scp = pci_nomatch_tab[i].desc;
3047				}
3048			}
3049		}
3050		device_printf(dev, "<%s%s%s>",
3051		    cp ? cp : "",
3052		    ((cp != NULL) && (scp != NULL)) ? ", " : "",
3053		    scp ? scp : "");
3054	}
3055	printf(" at device %d.%d (no driver attached)\n",
3056	    pci_get_slot(child), pci_get_function(child));
3057	pci_cfg_save(child, (struct pci_devinfo *)device_get_ivars(child), 1);
3058	return;
3059}
3060
3061/*
3062 * Parse the PCI device database, if loaded, and return a pointer to a
3063 * description of the device.
3064 *
3065 * The database is flat text formatted as follows:
3066 *
3067 * Any line not in a valid format is ignored.
3068 * Lines are terminated with newline '\n' characters.
3069 *
3070 * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
3071 * the vendor name.
3072 *
3073 * A DEVICE line is entered immediately below the corresponding VENDOR ID.
3074 * - devices cannot be listed without a corresponding VENDOR line.
3075 * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
3076 * another TAB, then the device name.
3077 */
3078
3079/*
3080 * Assuming (ptr) points to the beginning of a line in the database,
3081 * return the vendor or device and description of the next entry.
3082 * The value of (vendor) or (device) inappropriate for the entry type
3083 * is set to -1.  Returns nonzero at the end of the database.
3084 *
3085 * Note that this is slightly unrobust in the face of corrupt data;
3086 * we attempt to safeguard against this by spamming the end of the
3087 * database with a newline when we initialise.
3088 */
3089static int
3090pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
3091{
3092	char	*cp = *ptr;
3093	int	left;
3094
3095	*device = -1;
3096	*vendor = -1;
3097	**desc = '\0';
3098	for (;;) {
3099		left = pci_vendordata_size - (cp - pci_vendordata);
3100		if (left <= 0) {
3101			*ptr = cp;
3102			return(1);
3103		}
3104
3105		/* vendor entry? */
3106		if (*cp != '\t' &&
3107		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
3108			break;
3109		/* device entry? */
3110		if (*cp == '\t' &&
3111		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
3112			break;
3113
3114		/* skip to next line */
3115		while (*cp != '\n' && left > 0) {
3116			cp++;
3117			left--;
3118		}
3119		if (*cp == '\n') {
3120			cp++;
3121			left--;
3122		}
3123	}
3124	/* skip to next line */
3125	while (*cp != '\n' && left > 0) {
3126		cp++;
3127		left--;
3128	}
3129	if (*cp == '\n' && left > 0)
3130		cp++;
3131	*ptr = cp;
3132	return(0);
3133}
3134
3135static char *
3136pci_describe_device(device_t dev)
3137{
3138	int	vendor, device;
3139	char	*desc, *vp, *dp, *line;
3140
3141	desc = vp = dp = NULL;
3142
3143	/*
3144	 * If we have no vendor data, we can't do anything.
3145	 */
3146	if (pci_vendordata == NULL)
3147		goto out;
3148
3149	/*
3150	 * Scan the vendor data looking for this device
3151	 */
3152	line = pci_vendordata;
3153	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
3154		goto out;
3155	for (;;) {
3156		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
3157			goto out;
3158		if (vendor == pci_get_vendor(dev))
3159			break;
3160	}
3161	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
3162		goto out;
3163	for (;;) {
3164		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
3165			*dp = 0;
3166			break;
3167		}
3168		if (vendor != -1) {
3169			*dp = 0;
3170			break;
3171		}
3172		if (device == pci_get_device(dev))
3173			break;
3174	}
3175	if (dp[0] == '\0')
3176		snprintf(dp, 80, "0x%x", pci_get_device(dev));
3177	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
3178	    NULL)
3179		sprintf(desc, "%s, %s", vp, dp);
3180 out:
3181	if (vp != NULL)
3182		free(vp, M_DEVBUF);
3183	if (dp != NULL)
3184		free(dp, M_DEVBUF);
3185	return(desc);
3186}
3187
3188int
3189pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
3190{
3191	struct pci_devinfo *dinfo;
3192	pcicfgregs *cfg;
3193
3194	dinfo = device_get_ivars(child);
3195	cfg = &dinfo->cfg;
3196
3197	switch (which) {
3198	case PCI_IVAR_ETHADDR:
3199		/*
3200		 * The generic accessor doesn't deal with failure, so
3201		 * we set the return value, then return an error.
3202		 */
3203		*((uint8_t **) result) = NULL;
3204		return (EINVAL);
3205	case PCI_IVAR_SUBVENDOR:
3206		*result = cfg->subvendor;
3207		break;
3208	case PCI_IVAR_SUBDEVICE:
3209		*result = cfg->subdevice;
3210		break;
3211	case PCI_IVAR_VENDOR:
3212		*result = cfg->vendor;
3213		break;
3214	case PCI_IVAR_DEVICE:
3215		*result = cfg->device;
3216		break;
3217	case PCI_IVAR_DEVID:
3218		*result = (cfg->device << 16) | cfg->vendor;
3219		break;
3220	case PCI_IVAR_CLASS:
3221		*result = cfg->baseclass;
3222		break;
3223	case PCI_IVAR_SUBCLASS:
3224		*result = cfg->subclass;
3225		break;
3226	case PCI_IVAR_PROGIF:
3227		*result = cfg->progif;
3228		break;
3229	case PCI_IVAR_REVID:
3230		*result = cfg->revid;
3231		break;
3232	case PCI_IVAR_INTPIN:
3233		*result = cfg->intpin;
3234		break;
3235	case PCI_IVAR_IRQ:
3236		*result = cfg->intline;
3237		break;
3238	case PCI_IVAR_DOMAIN:
3239		*result = cfg->domain;
3240		break;
3241	case PCI_IVAR_BUS:
3242		*result = cfg->bus;
3243		break;
3244	case PCI_IVAR_SLOT:
3245		*result = cfg->slot;
3246		break;
3247	case PCI_IVAR_FUNCTION:
3248		*result = cfg->func;
3249		break;
3250	case PCI_IVAR_CMDREG:
3251		*result = cfg->cmdreg;
3252		break;
3253	case PCI_IVAR_CACHELNSZ:
3254		*result = cfg->cachelnsz;
3255		break;
3256	case PCI_IVAR_MINGNT:
3257		*result = cfg->mingnt;
3258		break;
3259	case PCI_IVAR_MAXLAT:
3260		*result = cfg->maxlat;
3261		break;
3262	case PCI_IVAR_LATTIMER:
3263		*result = cfg->lattimer;
3264		break;
3265	default:
3266		return (ENOENT);
3267	}
3268	return (0);
3269}
3270
3271int
3272pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
3273{
3274	struct pci_devinfo *dinfo;
3275
3276	dinfo = device_get_ivars(child);
3277
3278	switch (which) {
3279	case PCI_IVAR_INTPIN:
3280		dinfo->cfg.intpin = value;
3281		return (0);
3282	case PCI_IVAR_ETHADDR:
3283	case PCI_IVAR_SUBVENDOR:
3284	case PCI_IVAR_SUBDEVICE:
3285	case PCI_IVAR_VENDOR:
3286	case PCI_IVAR_DEVICE:
3287	case PCI_IVAR_DEVID:
3288	case PCI_IVAR_CLASS:
3289	case PCI_IVAR_SUBCLASS:
3290	case PCI_IVAR_PROGIF:
3291	case PCI_IVAR_REVID:
3292	case PCI_IVAR_IRQ:
3293	case PCI_IVAR_DOMAIN:
3294	case PCI_IVAR_BUS:
3295	case PCI_IVAR_SLOT:
3296	case PCI_IVAR_FUNCTION:
3297		return (EINVAL);	/* disallow for now */
3298
3299	default:
3300		return (ENOENT);
3301	}
3302}
3303
3304
3305#include "opt_ddb.h"
3306#ifdef DDB
3307#include <ddb/ddb.h>
3308#include <sys/cons.h>
3309
3310/*
3311 * List resources based on pci map registers, used for within ddb
3312 */
3313
3314DB_SHOW_COMMAND(pciregs, db_pci_dump)
3315{
3316	struct pci_devinfo *dinfo;
3317	struct devlist *devlist_head;
3318	struct pci_conf *p;
3319	const char *name;
3320	int i, error, none_count;
3321
3322	none_count = 0;
3323	/* get the head of the device queue */
3324	devlist_head = &pci_devq;
3325
3326	/*
3327	 * Go through the list of devices and print out devices
3328	 */
3329	for (error = 0, i = 0,
3330	     dinfo = STAILQ_FIRST(devlist_head);
3331	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
3332	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
3333
3334		/* Populate pd_name and pd_unit */
3335		name = NULL;
3336		if (dinfo->cfg.dev)
3337			name = device_get_name(dinfo->cfg.dev);
3338
3339		p = &dinfo->conf;
3340		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
3341			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
3342			(name && *name) ? name : "none",
3343			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
3344			none_count++,
3345			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
3346			p->pc_sel.pc_func, (p->pc_class << 16) |
3347			(p->pc_subclass << 8) | p->pc_progif,
3348			(p->pc_subdevice << 16) | p->pc_subvendor,
3349			(p->pc_device << 16) | p->pc_vendor,
3350			p->pc_revid, p->pc_hdr);
3351	}
3352}
3353#endif /* DDB */
3354
3355static struct resource *
3356pci_alloc_map(device_t dev, device_t child, int type, int *rid,
3357    u_long start, u_long end, u_long count, u_int flags)
3358{
3359	struct pci_devinfo *dinfo = device_get_ivars(child);
3360	struct resource_list *rl = &dinfo->resources;
3361	struct resource_list_entry *rle;
3362	struct resource *res;
3363	pci_addr_t map, testval;
3364	int mapsize;
3365
3366	/*
3367	 * Weed out the bogons, and figure out how large the BAR/map
3368	 * is.  Bars that read back 0 here are bogus and unimplemented.
3369	 * Note: atapci in legacy mode are special and handled elsewhere
3370	 * in the code.  If you have a atapci device in legacy mode and
3371	 * it fails here, that other code is broken.
3372	 */
3373	res = NULL;
3374	map = pci_read_config(child, *rid, 4);
3375	pci_write_config(child, *rid, 0xffffffff, 4);
3376	testval = pci_read_config(child, *rid, 4);
3377	if (pci_maprange(testval) == 64)
3378		map |= (pci_addr_t)pci_read_config(child, *rid + 4, 4) << 32;
3379	if (pci_mapbase(testval) == 0)
3380		goto out;
3381
3382	/*
3383	 * Restore the original value of the BAR.  We may have reprogrammed
3384	 * the BAR of the low-level console device and when booting verbose,
3385	 * we need the console device addressable.
3386	 */
3387	pci_write_config(child, *rid, map, 4);
3388
3389	if (PCI_BAR_MEM(testval)) {
3390		if (type != SYS_RES_MEMORY) {
3391			if (bootverbose)
3392				device_printf(dev,
3393				    "child %s requested type %d for rid %#x,"
3394				    " but the BAR says it is an memio\n",
3395				    device_get_nameunit(child), type, *rid);
3396			goto out;
3397		}
3398	} else {
3399		if (type != SYS_RES_IOPORT) {
3400			if (bootverbose)
3401				device_printf(dev,
3402				    "child %s requested type %d for rid %#x,"
3403				    " but the BAR says it is an ioport\n",
3404				    device_get_nameunit(child), type, *rid);
3405			goto out;
3406		}
3407	}
3408	/*
3409	 * For real BARs, we need to override the size that
3410	 * the driver requests, because that's what the BAR
3411	 * actually uses and we would otherwise have a
3412	 * situation where we might allocate the excess to
3413	 * another driver, which won't work.
3414	 */
3415	mapsize = pci_mapsize(testval);
3416	count = 1UL << mapsize;
3417	if (RF_ALIGNMENT(flags) < mapsize)
3418		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
3419
3420	/*
3421	 * Allocate enough resource, and then write back the
3422	 * appropriate bar for that resource.
3423	 */
3424	res = BUS_ALLOC_RESOURCE(device_get_parent(dev), child, type, rid,
3425	    start, end, count, flags);
3426	if (res == NULL) {
3427		device_printf(child,
3428		    "%#lx bytes of rid %#x res %d failed (%#lx, %#lx).\n",
3429		    count, *rid, type, start, end);
3430		goto out;
3431	}
3432	resource_list_add(rl, type, *rid, start, end, count);
3433	rle = resource_list_find(rl, type, *rid);
3434	if (rle == NULL)
3435		panic("pci_alloc_map: unexpectedly can't find resource.");
3436	rle->res = res;
3437	rle->start = rman_get_start(res);
3438	rle->end = rman_get_end(res);
3439	rle->count = count;
3440	if (bootverbose)
3441		device_printf(child,
3442		    "Lazy allocation of %#lx bytes rid %#x type %d at %#lx\n",
3443		    count, *rid, type, rman_get_start(res));
3444	map = rman_get_start(res);
3445out:;
3446	pci_write_config(child, *rid, map, 4);
3447	if (pci_maprange(testval) == 64)
3448		pci_write_config(child, *rid + 4, map >> 32, 4);
3449	return (res);
3450}
3451
3452
3453struct resource *
3454pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
3455		   u_long start, u_long end, u_long count, u_int flags)
3456{
3457	struct pci_devinfo *dinfo = device_get_ivars(child);
3458	struct resource_list *rl = &dinfo->resources;
3459	struct resource_list_entry *rle;
3460	pcicfgregs *cfg = &dinfo->cfg;
3461
3462	/*
3463	 * Perform lazy resource allocation
3464	 */
3465	if (device_get_parent(child) == dev) {
3466		switch (type) {
3467		case SYS_RES_IRQ:
3468			/*
3469			 * Can't alloc legacy interrupt once MSI messages
3470			 * have been allocated.
3471			 */
3472			if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
3473			    cfg->msix.msix_alloc > 0))
3474				return (NULL);
3475			/*
3476			 * If the child device doesn't have an
3477			 * interrupt routed and is deserving of an
3478			 * interrupt, try to assign it one.
3479			 */
3480			if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
3481			    (cfg->intpin != 0))
3482				pci_assign_interrupt(dev, child, 0);
3483			break;
3484		case SYS_RES_IOPORT:
3485		case SYS_RES_MEMORY:
3486			if (*rid < PCIR_BAR(cfg->nummaps)) {
3487				/*
3488				 * Enable the I/O mode.  We should
3489				 * also be assigning resources too
3490				 * when none are present.  The
3491				 * resource_list_alloc kind of sorta does
3492				 * this...
3493				 */
3494				if (PCI_ENABLE_IO(dev, child, type))
3495					return (NULL);
3496			}
3497			rle = resource_list_find(rl, type, *rid);
3498			if (rle == NULL)
3499				return (pci_alloc_map(dev, child, type, rid,
3500				    start, end, count, flags));
3501			break;
3502		}
3503		/*
3504		 * If we've already allocated the resource, then
3505		 * return it now.  But first we may need to activate
3506		 * it, since we don't allocate the resource as active
3507		 * above.  Normally this would be done down in the
3508		 * nexus, but since we short-circuit that path we have
3509		 * to do its job here.  Not sure if we should free the
3510		 * resource if it fails to activate.
3511		 */
3512		rle = resource_list_find(rl, type, *rid);
3513		if (rle != NULL && rle->res != NULL) {
3514			if (bootverbose)
3515				device_printf(child,
3516			    "Reserved %#lx bytes for rid %#x type %d at %#lx\n",
3517				    rman_get_size(rle->res), *rid, type,
3518				    rman_get_start(rle->res));
3519			if ((flags & RF_ACTIVE) &&
3520			    bus_generic_activate_resource(dev, child, type,
3521			    *rid, rle->res) != 0)
3522				return (NULL);
3523			return (rle->res);
3524		}
3525	}
3526	return (resource_list_alloc(rl, dev, child, type, rid,
3527	    start, end, count, flags));
3528}
3529
3530void
3531pci_delete_resource(device_t dev, device_t child, int type, int rid)
3532{
3533	struct pci_devinfo *dinfo;
3534	struct resource_list *rl;
3535	struct resource_list_entry *rle;
3536
3537	if (device_get_parent(child) != dev)
3538		return;
3539
3540	dinfo = device_get_ivars(child);
3541	rl = &dinfo->resources;
3542	rle = resource_list_find(rl, type, rid);
3543	if (rle) {
3544		if (rle->res) {
3545			if (rman_get_device(rle->res) != dev ||
3546			    rman_get_flags(rle->res) & RF_ACTIVE) {
3547				device_printf(dev, "delete_resource: "
3548				    "Resource still owned by child, oops. "
3549				    "(type=%d, rid=%d, addr=%lx)\n",
3550				    rle->type, rle->rid,
3551				    rman_get_start(rle->res));
3552				return;
3553			}
3554			bus_release_resource(dev, type, rid, rle->res);
3555		}
3556		resource_list_delete(rl, type, rid);
3557	}
3558	/*
3559	 * Why do we turn off the PCI configuration BAR when we delete a
3560	 * resource? -- imp
3561	 */
3562	pci_write_config(child, rid, 0, 4);
3563	BUS_DELETE_RESOURCE(device_get_parent(dev), child, type, rid);
3564}
3565
3566struct resource_list *
3567pci_get_resource_list (device_t dev, device_t child)
3568{
3569	struct pci_devinfo *dinfo = device_get_ivars(child);
3570
3571	return (&dinfo->resources);
3572}
3573
3574uint32_t
3575pci_read_config_method(device_t dev, device_t child, int reg, int width)
3576{
3577	struct pci_devinfo *dinfo = device_get_ivars(child);
3578	pcicfgregs *cfg = &dinfo->cfg;
3579
3580	return (PCIB_READ_CONFIG(device_get_parent(dev),
3581	    cfg->bus, cfg->slot, cfg->func, reg, width));
3582}
3583
3584void
3585pci_write_config_method(device_t dev, device_t child, int reg,
3586    uint32_t val, int width)
3587{
3588	struct pci_devinfo *dinfo = device_get_ivars(child);
3589	pcicfgregs *cfg = &dinfo->cfg;
3590
3591	PCIB_WRITE_CONFIG(device_get_parent(dev),
3592	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
3593}
3594
3595int
3596pci_child_location_str_method(device_t dev, device_t child, char *buf,
3597    size_t buflen)
3598{
3599
3600	snprintf(buf, buflen, "slot=%d function=%d", pci_get_slot(child),
3601	    pci_get_function(child));
3602	return (0);
3603}
3604
3605int
3606pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
3607    size_t buflen)
3608{
3609	struct pci_devinfo *dinfo;
3610	pcicfgregs *cfg;
3611
3612	dinfo = device_get_ivars(child);
3613	cfg = &dinfo->cfg;
3614	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
3615	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
3616	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
3617	    cfg->progif);
3618	return (0);
3619}
3620
3621int
3622pci_assign_interrupt_method(device_t dev, device_t child)
3623{
3624	struct pci_devinfo *dinfo = device_get_ivars(child);
3625	pcicfgregs *cfg = &dinfo->cfg;
3626
3627	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
3628	    cfg->intpin));
3629}
3630
3631static int
3632pci_modevent(module_t mod, int what, void *arg)
3633{
3634	static struct cdev *pci_cdev;
3635
3636	switch (what) {
3637	case MOD_LOAD:
3638		STAILQ_INIT(&pci_devq);
3639		pci_generation = 0;
3640		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
3641		    "pci");
3642		pci_load_vendor_data();
3643		break;
3644
3645	case MOD_UNLOAD:
3646		destroy_dev(pci_cdev);
3647		break;
3648	}
3649
3650	return (0);
3651}
3652
3653void
3654pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
3655{
3656	int i;
3657
3658	/*
3659	 * Only do header type 0 devices.  Type 1 devices are bridges,
3660	 * which we know need special treatment.  Type 2 devices are
3661	 * cardbus bridges which also require special treatment.
3662	 * Other types are unknown, and we err on the side of safety
3663	 * by ignoring them.
3664	 */
3665	if (dinfo->cfg.hdrtype != 0)
3666		return;
3667
3668	/*
3669	 * Restore the device to full power mode.  We must do this
3670	 * before we restore the registers because moving from D3 to
3671	 * D0 will cause the chip's BARs and some other registers to
3672	 * be reset to some unknown power on reset values.  Cut down
3673	 * the noise on boot by doing nothing if we are already in
3674	 * state D0.
3675	 */
3676	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
3677		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
3678	}
3679	for (i = 0; i < dinfo->cfg.nummaps; i++)
3680		pci_write_config(dev, PCIR_BAR(i), dinfo->cfg.bar[i], 4);
3681	pci_write_config(dev, PCIR_BIOS, dinfo->cfg.bios, 4);
3682	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
3683	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
3684	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
3685	pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
3686	pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
3687	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
3688	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
3689	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
3690	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
3691
3692	/* Restore MSI and MSI-X configurations if they are present. */
3693	if (dinfo->cfg.msi.msi_location != 0)
3694		pci_resume_msi(dev);
3695	if (dinfo->cfg.msix.msix_location != 0)
3696		pci_resume_msix(dev);
3697}
3698
3699void
3700pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
3701{
3702	int i;
3703	uint32_t cls;
3704	int ps;
3705
3706	/*
3707	 * Only do header type 0 devices.  Type 1 devices are bridges, which
3708	 * we know need special treatment.  Type 2 devices are cardbus bridges
3709	 * which also require special treatment.  Other types are unknown, and
3710	 * we err on the side of safety by ignoring them.  Powering down
3711	 * bridges should not be undertaken lightly.
3712	 */
3713	if (dinfo->cfg.hdrtype != 0)
3714		return;
3715	for (i = 0; i < dinfo->cfg.nummaps; i++)
3716		dinfo->cfg.bar[i] = pci_read_config(dev, PCIR_BAR(i), 4);
3717	dinfo->cfg.bios = pci_read_config(dev, PCIR_BIOS, 4);
3718
3719	/*
3720	 * Some drivers apparently write to these registers w/o updating our
3721	 * cached copy.  No harm happens if we update the copy, so do so here
3722	 * so we can restore them.  The COMMAND register is modified by the
3723	 * bus w/o updating the cache.  This should represent the normally
3724	 * writable portion of the 'defined' part of type 0 headers.  In
3725	 * theory we also need to save/restore the PCI capability structures
3726	 * we know about, but apart from power we don't know any that are
3727	 * writable.
3728	 */
3729	dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
3730	dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
3731	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
3732	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
3733	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
3734	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
3735	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
3736	dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
3737	dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
3738	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3739	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3740	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
3741	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
3742	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
3743	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
3744
3745	/*
3746	 * don't set the state for display devices, base peripherals and
3747	 * memory devices since bad things happen when they are powered down.
3748	 * We should (a) have drivers that can easily detach and (b) use
3749	 * generic drivers for these devices so that some device actually
3750	 * attaches.  We need to make sure that when we implement (a) we don't
3751	 * power the device down on a reattach.
3752	 */
3753	cls = pci_get_class(dev);
3754	if (!setstate)
3755		return;
3756	switch (pci_do_power_nodriver)
3757	{
3758		case 0:		/* NO powerdown at all */
3759			return;
3760		case 1:		/* Conservative about what to power down */
3761			if (cls == PCIC_STORAGE)
3762				return;
3763			/*FALLTHROUGH*/
3764		case 2:		/* Agressive about what to power down */
3765			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
3766			    cls == PCIC_BASEPERIPH)
3767				return;
3768			/*FALLTHROUGH*/
3769		case 3:		/* Power down everything */
3770			break;
3771	}
3772	/*
3773	 * PCI spec says we can only go into D3 state from D0 state.
3774	 * Transition from D[12] into D0 before going to D3 state.
3775	 */
3776	ps = pci_get_powerstate(dev);
3777	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
3778		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
3779	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
3780		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
3781}
3782