pci.c revision 187457
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 187457 2009-01-20 00:05:43Z nwhitehorn $");
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
2295	/*
2296	 * Disable decoding via the command register before
2297	 * determining the BARs length since we will be placing them
2298	 * in a weird state.
2299	 */
2300	cmd = PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2);
2301	PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND,
2302	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
2303
2304	/*
2305	 * Determine the BAR's length by writing all 1's.  The bottom
2306	 * log_2(size) bits of the BAR will stick as 0 when we read
2307	 * the value back.
2308	 */
2309	PCIB_WRITE_CONFIG(pcib, b, s, f, reg, 0xffffffff, 4);
2310	testval = PCIB_READ_CONFIG(pcib, b, s, f, reg, 4);
2311
2312	/* Restore the BAR and command register. */
2313	PCIB_WRITE_CONFIG(pcib, b, s, f, reg, map, 4);
2314	PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND, cmd, 2);
2315
2316	if (PCI_BAR_MEM(map))
2317		type = SYS_RES_MEMORY;
2318	else
2319		type = SYS_RES_IOPORT;
2320	ln2size = pci_mapsize(testval);
2321	ln2range = pci_maprange(testval);
2322	base = pci_mapbase(map);
2323	barlen = ln2range == 64 ? 2 : 1;
2324
2325	/*
2326	 * For I/O registers, if bottom bit is set, and the next bit up
2327	 * isn't clear, we know we have a BAR that doesn't conform to the
2328	 * spec, so ignore it.  Also, sanity check the size of the data
2329	 * areas to the type of memory involved.  Memory must be at least
2330	 * 16 bytes in size, while I/O ranges must be at least 4.
2331	 */
2332	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
2333		return (barlen);
2334	if ((type == SYS_RES_MEMORY && ln2size < 4) ||
2335	    (type == SYS_RES_IOPORT && ln2size < 2))
2336		return (barlen);
2337
2338	if (ln2range == 64)
2339		/* Read the other half of a 64bit map register */
2340		base |= (uint64_t) PCIB_READ_CONFIG(pcib, b, s, f, reg + 4, 4) << 32;
2341	if (bootverbose) {
2342		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
2343		    reg, pci_maptype(map), ln2range, (uintmax_t)base, ln2size);
2344		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f))
2345			printf(", port disabled\n");
2346		else if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f))
2347			printf(", memory disabled\n");
2348		else
2349			printf(", enabled\n");
2350	}
2351
2352	/*
2353	 * If base is 0, then we have problems.  It is best to ignore
2354	 * such entries for the moment.  These will be allocated later if
2355	 * the driver specifically requests them.  However, some
2356	 * removable busses look better when all resources are allocated,
2357	 * so allow '0' to be overriden.
2358	 *
2359	 * Similarly treat maps whose values is the same as the test value
2360	 * read back.  These maps have had all f's written to them by the
2361	 * BIOS in an attempt to disable the resources.
2362	 */
2363	if (!force && (base == 0 || map == testval))
2364		return (barlen);
2365	if ((u_long)base != base) {
2366		device_printf(bus,
2367		    "pci%d:%d:%d:%d bar %#x too many address bits",
2368		    pci_get_domain(dev), b, s, f, reg);
2369		return (barlen);
2370	}
2371
2372	/*
2373	 * This code theoretically does the right thing, but has
2374	 * undesirable side effects in some cases where peripherals
2375	 * respond oddly to having these bits enabled.  Let the user
2376	 * be able to turn them off (since pci_enable_io_modes is 1 by
2377	 * default).
2378	 */
2379	if (pci_enable_io_modes) {
2380		/* Turn on resources that have been left off by a lazy BIOS */
2381		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f)) {
2382			cmd = PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2);
2383			cmd |= PCIM_CMD_PORTEN;
2384			PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND, cmd, 2);
2385		}
2386		if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f)) {
2387			cmd = PCIB_READ_CONFIG(pcib, b, s, f, PCIR_COMMAND, 2);
2388			cmd |= PCIM_CMD_MEMEN;
2389			PCIB_WRITE_CONFIG(pcib, b, s, f, PCIR_COMMAND, cmd, 2);
2390		}
2391	} else {
2392		if (type == SYS_RES_IOPORT && !pci_porten(pcib, b, s, f))
2393			return (barlen);
2394		if (type == SYS_RES_MEMORY && !pci_memen(pcib, b, s, f))
2395			return (barlen);
2396	}
2397
2398	count = 1 << ln2size;
2399	if (base == 0 || base == pci_mapbase(testval)) {
2400		start = 0;	/* Let the parent decide. */
2401		end = ~0ULL;
2402	} else {
2403		start = base;
2404		end = base + (1 << ln2size) - 1;
2405	}
2406	resource_list_add(rl, type, reg, start, end, count);
2407
2408	/*
2409	 * Try to allocate the resource for this BAR from our parent
2410	 * so that this resource range is already reserved.  The
2411	 * driver for this device will later inherit this resource in
2412	 * pci_alloc_resource().
2413	 */
2414	res = resource_list_alloc(rl, bus, dev, type, &reg, start, end, count,
2415	    prefetch ? RF_PREFETCHABLE : 0);
2416	if (res == NULL) {
2417		/*
2418		 * If the allocation fails, clear the BAR and delete
2419		 * the resource list entry to force
2420		 * pci_alloc_resource() to allocate resources from the
2421		 * parent.
2422		 */
2423		resource_list_delete(rl, type, reg);
2424		start = 0;
2425	} else
2426		start = rman_get_start(res);
2427	pci_write_config(dev, reg, start, 4);
2428	if (ln2range == 64)
2429		pci_write_config(dev, reg + 4, start >> 32, 4);
2430	return (barlen);
2431}
2432
2433/*
2434 * For ATA devices we need to decide early what addressing mode to use.
2435 * Legacy demands that the primary and secondary ATA ports sits on the
2436 * same addresses that old ISA hardware did. This dictates that we use
2437 * those addresses and ignore the BAR's if we cannot set PCI native
2438 * addressing mode.
2439 */
2440static void
2441pci_ata_maps(device_t pcib, device_t bus, device_t dev, int b,
2442    int s, int f, struct resource_list *rl, int force, uint32_t prefetchmask)
2443{
2444	int rid, type, progif;
2445#if 0
2446	/* if this device supports PCI native addressing use it */
2447	progif = pci_read_config(dev, PCIR_PROGIF, 1);
2448	if ((progif & 0x8a) == 0x8a) {
2449		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
2450		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
2451			printf("Trying ATA native PCI addressing mode\n");
2452			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
2453		}
2454	}
2455#endif
2456	progif = pci_read_config(dev, PCIR_PROGIF, 1);
2457	type = SYS_RES_IOPORT;
2458	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
2459		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(0), rl, force,
2460		    prefetchmask & (1 << 0));
2461		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(1), rl, force,
2462		    prefetchmask & (1 << 1));
2463	} else {
2464		rid = PCIR_BAR(0);
2465		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
2466		resource_list_alloc(rl, bus, dev, type, &rid, 0x1f0, 0x1f7, 8,
2467		    0);
2468		rid = PCIR_BAR(1);
2469		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
2470		resource_list_alloc(rl, bus, dev, type, &rid, 0x3f6, 0x3f6, 1,
2471		    0);
2472	}
2473	if (progif & PCIP_STORAGE_IDE_MODESEC) {
2474		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(2), rl, force,
2475		    prefetchmask & (1 << 2));
2476		pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(3), rl, force,
2477		    prefetchmask & (1 << 3));
2478	} else {
2479		rid = PCIR_BAR(2);
2480		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
2481		resource_list_alloc(rl, bus, dev, type, &rid, 0x170, 0x177, 8,
2482		    0);
2483		rid = PCIR_BAR(3);
2484		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
2485		resource_list_alloc(rl, bus, dev, type, &rid, 0x376, 0x376, 1,
2486		    0);
2487	}
2488	pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(4), rl, force,
2489	    prefetchmask & (1 << 4));
2490	pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(5), rl, force,
2491	    prefetchmask & (1 << 5));
2492}
2493
2494static void
2495pci_assign_interrupt(device_t bus, device_t dev, int force_route)
2496{
2497	struct pci_devinfo *dinfo = device_get_ivars(dev);
2498	pcicfgregs *cfg = &dinfo->cfg;
2499	char tunable_name[64];
2500	int irq;
2501
2502	/* Has to have an intpin to have an interrupt. */
2503	if (cfg->intpin == 0)
2504		return;
2505
2506	/* Let the user override the IRQ with a tunable. */
2507	irq = PCI_INVALID_IRQ;
2508	snprintf(tunable_name, sizeof(tunable_name),
2509	    "hw.pci%d.%d.%d.INT%c.irq",
2510	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
2511	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
2512		irq = PCI_INVALID_IRQ;
2513
2514	/*
2515	 * If we didn't get an IRQ via the tunable, then we either use the
2516	 * IRQ value in the intline register or we ask the bus to route an
2517	 * interrupt for us.  If force_route is true, then we only use the
2518	 * value in the intline register if the bus was unable to assign an
2519	 * IRQ.
2520	 */
2521	if (!PCI_INTERRUPT_VALID(irq)) {
2522		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
2523			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
2524		if (!PCI_INTERRUPT_VALID(irq))
2525			irq = cfg->intline;
2526	}
2527
2528	/* If after all that we don't have an IRQ, just bail. */
2529	if (!PCI_INTERRUPT_VALID(irq))
2530		return;
2531
2532	/* Update the config register if it changed. */
2533	if (irq != cfg->intline) {
2534		cfg->intline = irq;
2535		pci_write_config(dev, PCIR_INTLINE, irq, 1);
2536	}
2537
2538	/* Add this IRQ as rid 0 interrupt resource. */
2539	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
2540}
2541
2542void
2543pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
2544{
2545	device_t pcib;
2546	struct pci_devinfo *dinfo = device_get_ivars(dev);
2547	pcicfgregs *cfg = &dinfo->cfg;
2548	struct resource_list *rl = &dinfo->resources;
2549	struct pci_quirk *q;
2550	int b, i, f, s;
2551
2552	pcib = device_get_parent(bus);
2553
2554	b = cfg->bus;
2555	s = cfg->slot;
2556	f = cfg->func;
2557
2558	/* ATA devices needs special map treatment */
2559	if ((pci_get_class(dev) == PCIC_STORAGE) &&
2560	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
2561	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
2562	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
2563	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
2564		pci_ata_maps(pcib, bus, dev, b, s, f, rl, force, prefetchmask);
2565	else
2566		for (i = 0; i < cfg->nummaps;)
2567			i += pci_add_map(pcib, bus, dev, b, s, f, PCIR_BAR(i),
2568			    rl, force, prefetchmask & (1 << i));
2569
2570	/*
2571	 * Add additional, quirked resources.
2572	 */
2573	for (q = &pci_quirks[0]; q->devid; q++) {
2574		if (q->devid == ((cfg->device << 16) | cfg->vendor)
2575		    && q->type == PCI_QUIRK_MAP_REG)
2576			pci_add_map(pcib, bus, dev, b, s, f, q->arg1, rl,
2577			  force, 0);
2578	}
2579
2580	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
2581#ifdef __PCI_REROUTE_INTERRUPT
2582		/*
2583		 * Try to re-route interrupts. Sometimes the BIOS or
2584		 * firmware may leave bogus values in these registers.
2585		 * If the re-route fails, then just stick with what we
2586		 * have.
2587		 */
2588		pci_assign_interrupt(bus, dev, 1);
2589#else
2590		pci_assign_interrupt(bus, dev, 0);
2591#endif
2592	}
2593}
2594
2595void
2596pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size)
2597{
2598#define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
2599	device_t pcib = device_get_parent(dev);
2600	struct pci_devinfo *dinfo;
2601	int maxslots;
2602	int s, f, pcifunchigh;
2603	uint8_t hdrtype;
2604
2605	KASSERT(dinfo_size >= sizeof(struct pci_devinfo),
2606	    ("dinfo_size too small"));
2607	maxslots = PCIB_MAXSLOTS(pcib);
2608	for (s = 0; s <= maxslots; s++) {
2609		pcifunchigh = 0;
2610		f = 0;
2611		DELAY(1);
2612		hdrtype = REG(PCIR_HDRTYPE, 1);
2613		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
2614			continue;
2615		if (hdrtype & PCIM_MFDEV)
2616			pcifunchigh = PCI_FUNCMAX;
2617		for (f = 0; f <= pcifunchigh; f++) {
2618			dinfo = pci_read_device(pcib, domain, busno, s, f,
2619			    dinfo_size);
2620			if (dinfo != NULL) {
2621				pci_add_child(dev, dinfo);
2622			}
2623		}
2624	}
2625#undef REG
2626}
2627
2628void
2629pci_add_child(device_t bus, struct pci_devinfo *dinfo)
2630{
2631	dinfo->cfg.dev = device_add_child(bus, NULL, -1);
2632	device_set_ivars(dinfo->cfg.dev, dinfo);
2633	resource_list_init(&dinfo->resources);
2634	pci_cfg_save(dinfo->cfg.dev, dinfo, 0);
2635	pci_cfg_restore(dinfo->cfg.dev, dinfo);
2636	pci_print_verbose(dinfo);
2637	pci_add_resources(bus, dinfo->cfg.dev, 0, 0);
2638}
2639
2640static int
2641pci_probe(device_t dev)
2642{
2643
2644	device_set_desc(dev, "PCI bus");
2645
2646	/* Allow other subclasses to override this driver. */
2647	return (BUS_PROBE_GENERIC);
2648}
2649
2650static int
2651pci_attach(device_t dev)
2652{
2653	int busno, domain;
2654
2655	/*
2656	 * Since there can be multiple independantly numbered PCI
2657	 * busses on systems with multiple PCI domains, we can't use
2658	 * the unit number to decide which bus we are probing. We ask
2659	 * the parent pcib what our domain and bus numbers are.
2660	 */
2661	domain = pcib_get_domain(dev);
2662	busno = pcib_get_bus(dev);
2663	if (bootverbose)
2664		device_printf(dev, "domain=%d, physical bus=%d\n",
2665		    domain, busno);
2666	pci_add_children(dev, domain, busno, sizeof(struct pci_devinfo));
2667	return (bus_generic_attach(dev));
2668}
2669
2670int
2671pci_suspend(device_t dev)
2672{
2673	int dstate, error, i, numdevs;
2674	device_t acpi_dev, child, *devlist;
2675	struct pci_devinfo *dinfo;
2676
2677	/*
2678	 * Save the PCI configuration space for each child and set the
2679	 * device in the appropriate power state for this sleep state.
2680	 */
2681	acpi_dev = NULL;
2682	if (pci_do_power_resume)
2683		acpi_dev = devclass_get_device(devclass_find("acpi"), 0);
2684	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
2685		return (error);
2686	for (i = 0; i < numdevs; i++) {
2687		child = devlist[i];
2688		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2689		pci_cfg_save(child, dinfo, 0);
2690	}
2691
2692	/* Suspend devices before potentially powering them down. */
2693	error = bus_generic_suspend(dev);
2694	if (error) {
2695		free(devlist, M_TEMP);
2696		return (error);
2697	}
2698
2699	/*
2700	 * Always set the device to D3.  If ACPI suggests a different
2701	 * power state, use it instead.  If ACPI is not present, the
2702	 * firmware is responsible for managing device power.  Skip
2703	 * children who aren't attached since they are powered down
2704	 * separately.  Only manage type 0 devices for now.
2705	 */
2706	for (i = 0; acpi_dev && i < numdevs; i++) {
2707		child = devlist[i];
2708		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2709		if (device_is_attached(child) && dinfo->cfg.hdrtype == 0) {
2710			dstate = PCI_POWERSTATE_D3;
2711			ACPI_PWR_FOR_SLEEP(acpi_dev, child, &dstate);
2712			pci_set_powerstate(child, dstate);
2713		}
2714	}
2715	free(devlist, M_TEMP);
2716	return (0);
2717}
2718
2719int
2720pci_resume(device_t dev)
2721{
2722	int i, numdevs, error;
2723	device_t acpi_dev, child, *devlist;
2724	struct pci_devinfo *dinfo;
2725
2726	/*
2727	 * Set each child to D0 and restore its PCI configuration space.
2728	 */
2729	acpi_dev = NULL;
2730	if (pci_do_power_resume)
2731		acpi_dev = devclass_get_device(devclass_find("acpi"), 0);
2732	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
2733		return (error);
2734	for (i = 0; i < numdevs; i++) {
2735		/*
2736		 * Notify ACPI we're going to D0 but ignore the result.  If
2737		 * ACPI is not present, the firmware is responsible for
2738		 * managing device power.  Only manage type 0 devices for now.
2739		 */
2740		child = devlist[i];
2741		dinfo = (struct pci_devinfo *) device_get_ivars(child);
2742		if (acpi_dev && device_is_attached(child) &&
2743		    dinfo->cfg.hdrtype == 0) {
2744			ACPI_PWR_FOR_SLEEP(acpi_dev, child, NULL);
2745			pci_set_powerstate(child, PCI_POWERSTATE_D0);
2746		}
2747
2748		/* Now the device is powered up, restore its config space. */
2749		pci_cfg_restore(child, dinfo);
2750	}
2751	free(devlist, M_TEMP);
2752	return (bus_generic_resume(dev));
2753}
2754
2755static void
2756pci_load_vendor_data(void)
2757{
2758	caddr_t vendordata, info;
2759
2760	if ((vendordata = preload_search_by_type("pci_vendor_data")) != NULL) {
2761		info = preload_search_info(vendordata, MODINFO_ADDR);
2762		pci_vendordata = *(char **)info;
2763		info = preload_search_info(vendordata, MODINFO_SIZE);
2764		pci_vendordata_size = *(size_t *)info;
2765		/* terminate the database */
2766		pci_vendordata[pci_vendordata_size] = '\n';
2767	}
2768}
2769
2770void
2771pci_driver_added(device_t dev, driver_t *driver)
2772{
2773	int numdevs;
2774	device_t *devlist;
2775	device_t child;
2776	struct pci_devinfo *dinfo;
2777	int i;
2778
2779	if (bootverbose)
2780		device_printf(dev, "driver added\n");
2781	DEVICE_IDENTIFY(driver, dev);
2782	if (device_get_children(dev, &devlist, &numdevs) != 0)
2783		return;
2784	for (i = 0; i < numdevs; i++) {
2785		child = devlist[i];
2786		if (device_get_state(child) != DS_NOTPRESENT)
2787			continue;
2788		dinfo = device_get_ivars(child);
2789		pci_print_verbose(dinfo);
2790		if (bootverbose)
2791			printf("pci%d:%d:%d:%d: reprobing on driver added\n",
2792			    dinfo->cfg.domain, dinfo->cfg.bus, dinfo->cfg.slot,
2793			    dinfo->cfg.func);
2794		pci_cfg_restore(child, dinfo);
2795		if (device_probe_and_attach(child) != 0)
2796			pci_cfg_save(child, dinfo, 1);
2797	}
2798	free(devlist, M_TEMP);
2799}
2800
2801int
2802pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
2803    driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
2804{
2805	struct pci_devinfo *dinfo;
2806	struct msix_table_entry *mte;
2807	struct msix_vector *mv;
2808	uint64_t addr;
2809	uint32_t data;
2810	void *cookie;
2811	int error, rid;
2812
2813	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
2814	    arg, &cookie);
2815	if (error)
2816		return (error);
2817
2818	/*
2819	 * If this is a direct child, check to see if the interrupt is
2820	 * MSI or MSI-X.  If so, ask our parent to map the MSI and give
2821	 * us the address and data register values.  If we fail for some
2822	 * reason, teardown the interrupt handler.
2823	 */
2824	rid = rman_get_rid(irq);
2825	if (device_get_parent(child) == dev && rid > 0) {
2826		dinfo = device_get_ivars(child);
2827		if (dinfo->cfg.msi.msi_alloc > 0) {
2828			if (dinfo->cfg.msi.msi_addr == 0) {
2829				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
2830			    ("MSI has handlers, but vectors not mapped"));
2831				error = PCIB_MAP_MSI(device_get_parent(dev),
2832				    child, rman_get_start(irq), &addr, &data);
2833				if (error)
2834					goto bad;
2835				dinfo->cfg.msi.msi_addr = addr;
2836				dinfo->cfg.msi.msi_data = data;
2837				pci_enable_msi(child, addr, data);
2838			}
2839			dinfo->cfg.msi.msi_handlers++;
2840		} else {
2841			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
2842			    ("No MSI or MSI-X interrupts allocated"));
2843			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
2844			    ("MSI-X index too high"));
2845			mte = &dinfo->cfg.msix.msix_table[rid - 1];
2846			KASSERT(mte->mte_vector != 0, ("no message vector"));
2847			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
2848			KASSERT(mv->mv_irq == rman_get_start(irq),
2849			    ("IRQ mismatch"));
2850			if (mv->mv_address == 0) {
2851				KASSERT(mte->mte_handlers == 0,
2852		    ("MSI-X table entry has handlers, but vector not mapped"));
2853				error = PCIB_MAP_MSI(device_get_parent(dev),
2854				    child, rman_get_start(irq), &addr, &data);
2855				if (error)
2856					goto bad;
2857				mv->mv_address = addr;
2858				mv->mv_data = data;
2859			}
2860			if (mte->mte_handlers == 0) {
2861				pci_enable_msix(child, rid - 1, mv->mv_address,
2862				    mv->mv_data);
2863				pci_unmask_msix(child, rid - 1);
2864			}
2865			mte->mte_handlers++;
2866		}
2867	bad:
2868		if (error) {
2869			(void)bus_generic_teardown_intr(dev, child, irq,
2870			    cookie);
2871			return (error);
2872		}
2873	}
2874	*cookiep = cookie;
2875	return (0);
2876}
2877
2878int
2879pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
2880    void *cookie)
2881{
2882	struct msix_table_entry *mte;
2883	struct resource_list_entry *rle;
2884	struct pci_devinfo *dinfo;
2885	int error, rid;
2886
2887	/*
2888	 * If this is a direct child, check to see if the interrupt is
2889	 * MSI or MSI-X.  If so, decrement the appropriate handlers
2890	 * count and mask the MSI-X message, or disable MSI messages
2891	 * if the count drops to 0.
2892	 */
2893	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
2894		return (EINVAL);
2895	rid = rman_get_rid(irq);
2896	if (device_get_parent(child) == dev && rid > 0) {
2897		dinfo = device_get_ivars(child);
2898		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
2899		if (rle->res != irq)
2900			return (EINVAL);
2901		if (dinfo->cfg.msi.msi_alloc > 0) {
2902			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
2903			    ("MSI-X index too high"));
2904			if (dinfo->cfg.msi.msi_handlers == 0)
2905				return (EINVAL);
2906			dinfo->cfg.msi.msi_handlers--;
2907			if (dinfo->cfg.msi.msi_handlers == 0)
2908				pci_disable_msi(child);
2909		} else {
2910			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
2911			    ("No MSI or MSI-X interrupts allocated"));
2912			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
2913			    ("MSI-X index too high"));
2914			mte = &dinfo->cfg.msix.msix_table[rid - 1];
2915			if (mte->mte_handlers == 0)
2916				return (EINVAL);
2917			mte->mte_handlers--;
2918			if (mte->mte_handlers == 0)
2919				pci_mask_msix(child, rid - 1);
2920		}
2921	}
2922	error = bus_generic_teardown_intr(dev, child, irq, cookie);
2923	if (device_get_parent(child) == dev && rid > 0)
2924		KASSERT(error == 0,
2925		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
2926	return (error);
2927}
2928
2929int
2930pci_print_child(device_t dev, device_t child)
2931{
2932	struct pci_devinfo *dinfo;
2933	struct resource_list *rl;
2934	int retval = 0;
2935
2936	dinfo = device_get_ivars(child);
2937	rl = &dinfo->resources;
2938
2939	retval += bus_print_child_header(dev, child);
2940
2941	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#lx");
2942	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#lx");
2943	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%ld");
2944	if (device_get_flags(dev))
2945		retval += printf(" flags %#x", device_get_flags(dev));
2946
2947	retval += printf(" at device %d.%d", pci_get_slot(child),
2948	    pci_get_function(child));
2949
2950	retval += bus_print_child_footer(dev, child);
2951
2952	return (retval);
2953}
2954
2955static struct
2956{
2957	int	class;
2958	int	subclass;
2959	char	*desc;
2960} pci_nomatch_tab[] = {
2961	{PCIC_OLD,		-1,			"old"},
2962	{PCIC_OLD,		PCIS_OLD_NONVGA,	"non-VGA display device"},
2963	{PCIC_OLD,		PCIS_OLD_VGA,		"VGA-compatible display device"},
2964	{PCIC_STORAGE,		-1,			"mass storage"},
2965	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	"SCSI"},
2966	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	"ATA"},
2967	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	"floppy disk"},
2968	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	"IPI"},
2969	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	"RAID"},
2970	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	"ATA (ADMA)"},
2971	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	"SATA"},
2972	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	"SAS"},
2973	{PCIC_NETWORK,		-1,			"network"},
2974	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	"ethernet"},
2975	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	"token ring"},
2976	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	"fddi"},
2977	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	"ATM"},
2978	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	"ISDN"},
2979	{PCIC_DISPLAY,		-1,			"display"},
2980	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	"VGA"},
2981	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	"XGA"},
2982	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	"3D"},
2983	{PCIC_MULTIMEDIA,	-1,			"multimedia"},
2984	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	"video"},
2985	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	"audio"},
2986	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	"telephony"},
2987	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	"HDA"},
2988	{PCIC_MEMORY,		-1,			"memory"},
2989	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	"RAM"},
2990	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	"flash"},
2991	{PCIC_BRIDGE,		-1,			"bridge"},
2992	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	"HOST-PCI"},
2993	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	"PCI-ISA"},
2994	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	"PCI-EISA"},
2995	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	"PCI-MCA"},
2996	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	"PCI-PCI"},
2997	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	"PCI-PCMCIA"},
2998	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	"PCI-NuBus"},
2999	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	"PCI-CardBus"},
3000	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	"PCI-RACEway"},
3001	{PCIC_SIMPLECOMM,	-1,			"simple comms"},
3002	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	"UART"},	/* could detect 16550 */
3003	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	"parallel port"},
3004	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	"multiport serial"},
3005	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	"generic modem"},
3006	{PCIC_BASEPERIPH,	-1,			"base peripheral"},
3007	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	"interrupt controller"},
3008	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	"DMA controller"},
3009	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	"timer"},
3010	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	"realtime clock"},
3011	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	"PCI hot-plug controller"},
3012	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	"SD host controller"},
3013	{PCIC_INPUTDEV,		-1,			"input device"},
3014	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	"keyboard"},
3015	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,"digitizer"},
3016	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	"mouse"},
3017	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	"scanner"},
3018	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	"gameport"},
3019	{PCIC_DOCKING,		-1,			"docking station"},
3020	{PCIC_PROCESSOR,	-1,			"processor"},
3021	{PCIC_SERIALBUS,	-1,			"serial bus"},
3022	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	"FireWire"},
3023	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	"AccessBus"},
3024	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	"SSA"},
3025	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	"USB"},
3026	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	"Fibre Channel"},
3027	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	"SMBus"},
3028	{PCIC_WIRELESS,		-1,			"wireless controller"},
3029	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	"iRDA"},
3030	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	"IR"},
3031	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	"RF"},
3032	{PCIC_INTELLIIO,	-1,			"intelligent I/O controller"},
3033	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	"I2O"},
3034	{PCIC_SATCOM,		-1,			"satellite communication"},
3035	{PCIC_SATCOM,		PCIS_SATCOM_TV,		"sat TV"},
3036	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	"sat audio"},
3037	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	"sat voice"},
3038	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	"sat data"},
3039	{PCIC_CRYPTO,		-1,			"encrypt/decrypt"},
3040	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	"network/computer crypto"},
3041	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	"entertainment crypto"},
3042	{PCIC_DASP,		-1,			"dasp"},
3043	{PCIC_DASP,		PCIS_DASP_DPIO,		"DPIO module"},
3044	{0, 0,		NULL}
3045};
3046
3047void
3048pci_probe_nomatch(device_t dev, device_t child)
3049{
3050	int	i;
3051	char	*cp, *scp, *device;
3052
3053	/*
3054	 * Look for a listing for this device in a loaded device database.
3055	 */
3056	if ((device = pci_describe_device(child)) != NULL) {
3057		device_printf(dev, "<%s>", device);
3058		free(device, M_DEVBUF);
3059	} else {
3060		/*
3061		 * Scan the class/subclass descriptions for a general
3062		 * description.
3063		 */
3064		cp = "unknown";
3065		scp = NULL;
3066		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
3067			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
3068				if (pci_nomatch_tab[i].subclass == -1) {
3069					cp = pci_nomatch_tab[i].desc;
3070				} else if (pci_nomatch_tab[i].subclass ==
3071				    pci_get_subclass(child)) {
3072					scp = pci_nomatch_tab[i].desc;
3073				}
3074			}
3075		}
3076		device_printf(dev, "<%s%s%s>",
3077		    cp ? cp : "",
3078		    ((cp != NULL) && (scp != NULL)) ? ", " : "",
3079		    scp ? scp : "");
3080	}
3081	printf(" at device %d.%d (no driver attached)\n",
3082	    pci_get_slot(child), pci_get_function(child));
3083	pci_cfg_save(child, (struct pci_devinfo *)device_get_ivars(child), 1);
3084	return;
3085}
3086
3087/*
3088 * Parse the PCI device database, if loaded, and return a pointer to a
3089 * description of the device.
3090 *
3091 * The database is flat text formatted as follows:
3092 *
3093 * Any line not in a valid format is ignored.
3094 * Lines are terminated with newline '\n' characters.
3095 *
3096 * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
3097 * the vendor name.
3098 *
3099 * A DEVICE line is entered immediately below the corresponding VENDOR ID.
3100 * - devices cannot be listed without a corresponding VENDOR line.
3101 * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
3102 * another TAB, then the device name.
3103 */
3104
3105/*
3106 * Assuming (ptr) points to the beginning of a line in the database,
3107 * return the vendor or device and description of the next entry.
3108 * The value of (vendor) or (device) inappropriate for the entry type
3109 * is set to -1.  Returns nonzero at the end of the database.
3110 *
3111 * Note that this is slightly unrobust in the face of corrupt data;
3112 * we attempt to safeguard against this by spamming the end of the
3113 * database with a newline when we initialise.
3114 */
3115static int
3116pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
3117{
3118	char	*cp = *ptr;
3119	int	left;
3120
3121	*device = -1;
3122	*vendor = -1;
3123	**desc = '\0';
3124	for (;;) {
3125		left = pci_vendordata_size - (cp - pci_vendordata);
3126		if (left <= 0) {
3127			*ptr = cp;
3128			return(1);
3129		}
3130
3131		/* vendor entry? */
3132		if (*cp != '\t' &&
3133		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
3134			break;
3135		/* device entry? */
3136		if (*cp == '\t' &&
3137		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
3138			break;
3139
3140		/* skip to next line */
3141		while (*cp != '\n' && left > 0) {
3142			cp++;
3143			left--;
3144		}
3145		if (*cp == '\n') {
3146			cp++;
3147			left--;
3148		}
3149	}
3150	/* skip to next line */
3151	while (*cp != '\n' && left > 0) {
3152		cp++;
3153		left--;
3154	}
3155	if (*cp == '\n' && left > 0)
3156		cp++;
3157	*ptr = cp;
3158	return(0);
3159}
3160
3161static char *
3162pci_describe_device(device_t dev)
3163{
3164	int	vendor, device;
3165	char	*desc, *vp, *dp, *line;
3166
3167	desc = vp = dp = NULL;
3168
3169	/*
3170	 * If we have no vendor data, we can't do anything.
3171	 */
3172	if (pci_vendordata == NULL)
3173		goto out;
3174
3175	/*
3176	 * Scan the vendor data looking for this device
3177	 */
3178	line = pci_vendordata;
3179	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
3180		goto out;
3181	for (;;) {
3182		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
3183			goto out;
3184		if (vendor == pci_get_vendor(dev))
3185			break;
3186	}
3187	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
3188		goto out;
3189	for (;;) {
3190		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
3191			*dp = 0;
3192			break;
3193		}
3194		if (vendor != -1) {
3195			*dp = 0;
3196			break;
3197		}
3198		if (device == pci_get_device(dev))
3199			break;
3200	}
3201	if (dp[0] == '\0')
3202		snprintf(dp, 80, "0x%x", pci_get_device(dev));
3203	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
3204	    NULL)
3205		sprintf(desc, "%s, %s", vp, dp);
3206 out:
3207	if (vp != NULL)
3208		free(vp, M_DEVBUF);
3209	if (dp != NULL)
3210		free(dp, M_DEVBUF);
3211	return(desc);
3212}
3213
3214int
3215pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
3216{
3217	struct pci_devinfo *dinfo;
3218	pcicfgregs *cfg;
3219
3220	dinfo = device_get_ivars(child);
3221	cfg = &dinfo->cfg;
3222
3223	switch (which) {
3224	case PCI_IVAR_ETHADDR:
3225		/*
3226		 * The generic accessor doesn't deal with failure, so
3227		 * we set the return value, then return an error.
3228		 */
3229		*((uint8_t **) result) = NULL;
3230		return (EINVAL);
3231	case PCI_IVAR_SUBVENDOR:
3232		*result = cfg->subvendor;
3233		break;
3234	case PCI_IVAR_SUBDEVICE:
3235		*result = cfg->subdevice;
3236		break;
3237	case PCI_IVAR_VENDOR:
3238		*result = cfg->vendor;
3239		break;
3240	case PCI_IVAR_DEVICE:
3241		*result = cfg->device;
3242		break;
3243	case PCI_IVAR_DEVID:
3244		*result = (cfg->device << 16) | cfg->vendor;
3245		break;
3246	case PCI_IVAR_CLASS:
3247		*result = cfg->baseclass;
3248		break;
3249	case PCI_IVAR_SUBCLASS:
3250		*result = cfg->subclass;
3251		break;
3252	case PCI_IVAR_PROGIF:
3253		*result = cfg->progif;
3254		break;
3255	case PCI_IVAR_REVID:
3256		*result = cfg->revid;
3257		break;
3258	case PCI_IVAR_INTPIN:
3259		*result = cfg->intpin;
3260		break;
3261	case PCI_IVAR_IRQ:
3262		*result = cfg->intline;
3263		break;
3264	case PCI_IVAR_DOMAIN:
3265		*result = cfg->domain;
3266		break;
3267	case PCI_IVAR_BUS:
3268		*result = cfg->bus;
3269		break;
3270	case PCI_IVAR_SLOT:
3271		*result = cfg->slot;
3272		break;
3273	case PCI_IVAR_FUNCTION:
3274		*result = cfg->func;
3275		break;
3276	case PCI_IVAR_CMDREG:
3277		*result = cfg->cmdreg;
3278		break;
3279	case PCI_IVAR_CACHELNSZ:
3280		*result = cfg->cachelnsz;
3281		break;
3282	case PCI_IVAR_MINGNT:
3283		*result = cfg->mingnt;
3284		break;
3285	case PCI_IVAR_MAXLAT:
3286		*result = cfg->maxlat;
3287		break;
3288	case PCI_IVAR_LATTIMER:
3289		*result = cfg->lattimer;
3290		break;
3291	default:
3292		return (ENOENT);
3293	}
3294	return (0);
3295}
3296
3297int
3298pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
3299{
3300	struct pci_devinfo *dinfo;
3301
3302	dinfo = device_get_ivars(child);
3303
3304	switch (which) {
3305	case PCI_IVAR_INTPIN:
3306		dinfo->cfg.intpin = value;
3307		return (0);
3308	case PCI_IVAR_ETHADDR:
3309	case PCI_IVAR_SUBVENDOR:
3310	case PCI_IVAR_SUBDEVICE:
3311	case PCI_IVAR_VENDOR:
3312	case PCI_IVAR_DEVICE:
3313	case PCI_IVAR_DEVID:
3314	case PCI_IVAR_CLASS:
3315	case PCI_IVAR_SUBCLASS:
3316	case PCI_IVAR_PROGIF:
3317	case PCI_IVAR_REVID:
3318	case PCI_IVAR_IRQ:
3319	case PCI_IVAR_DOMAIN:
3320	case PCI_IVAR_BUS:
3321	case PCI_IVAR_SLOT:
3322	case PCI_IVAR_FUNCTION:
3323		return (EINVAL);	/* disallow for now */
3324
3325	default:
3326		return (ENOENT);
3327	}
3328}
3329
3330
3331#include "opt_ddb.h"
3332#ifdef DDB
3333#include <ddb/ddb.h>
3334#include <sys/cons.h>
3335
3336/*
3337 * List resources based on pci map registers, used for within ddb
3338 */
3339
3340DB_SHOW_COMMAND(pciregs, db_pci_dump)
3341{
3342	struct pci_devinfo *dinfo;
3343	struct devlist *devlist_head;
3344	struct pci_conf *p;
3345	const char *name;
3346	int i, error, none_count;
3347
3348	none_count = 0;
3349	/* get the head of the device queue */
3350	devlist_head = &pci_devq;
3351
3352	/*
3353	 * Go through the list of devices and print out devices
3354	 */
3355	for (error = 0, i = 0,
3356	     dinfo = STAILQ_FIRST(devlist_head);
3357	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
3358	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
3359
3360		/* Populate pd_name and pd_unit */
3361		name = NULL;
3362		if (dinfo->cfg.dev)
3363			name = device_get_name(dinfo->cfg.dev);
3364
3365		p = &dinfo->conf;
3366		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
3367			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
3368			(name && *name) ? name : "none",
3369			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
3370			none_count++,
3371			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
3372			p->pc_sel.pc_func, (p->pc_class << 16) |
3373			(p->pc_subclass << 8) | p->pc_progif,
3374			(p->pc_subdevice << 16) | p->pc_subvendor,
3375			(p->pc_device << 16) | p->pc_vendor,
3376			p->pc_revid, p->pc_hdr);
3377	}
3378}
3379#endif /* DDB */
3380
3381static struct resource *
3382pci_alloc_map(device_t dev, device_t child, int type, int *rid,
3383    u_long start, u_long end, u_long count, u_int flags)
3384{
3385	struct pci_devinfo *dinfo = device_get_ivars(child);
3386	struct resource_list *rl = &dinfo->resources;
3387	struct resource_list_entry *rle;
3388	struct resource *res;
3389	pci_addr_t map, testval;
3390	int mapsize;
3391
3392	/*
3393	 * Weed out the bogons, and figure out how large the BAR/map
3394	 * is.  Bars that read back 0 here are bogus and unimplemented.
3395	 * Note: atapci in legacy mode are special and handled elsewhere
3396	 * in the code.  If you have a atapci device in legacy mode and
3397	 * it fails here, that other code is broken.
3398	 */
3399	res = NULL;
3400	map = pci_read_config(child, *rid, 4);
3401	pci_write_config(child, *rid, 0xffffffff, 4);
3402	testval = pci_read_config(child, *rid, 4);
3403	if (pci_maprange(testval) == 64)
3404		map |= (pci_addr_t)pci_read_config(child, *rid + 4, 4) << 32;
3405	if (pci_mapbase(testval) == 0)
3406		goto out;
3407
3408	/*
3409	 * Restore the original value of the BAR.  We may have reprogrammed
3410	 * the BAR of the low-level console device and when booting verbose,
3411	 * we need the console device addressable.
3412	 */
3413	pci_write_config(child, *rid, map, 4);
3414
3415	if (PCI_BAR_MEM(testval)) {
3416		if (type != SYS_RES_MEMORY) {
3417			if (bootverbose)
3418				device_printf(dev,
3419				    "child %s requested type %d for rid %#x,"
3420				    " but the BAR says it is an memio\n",
3421				    device_get_nameunit(child), type, *rid);
3422			goto out;
3423		}
3424	} else {
3425		if (type != SYS_RES_IOPORT) {
3426			if (bootverbose)
3427				device_printf(dev,
3428				    "child %s requested type %d for rid %#x,"
3429				    " but the BAR says it is an ioport\n",
3430				    device_get_nameunit(child), type, *rid);
3431			goto out;
3432		}
3433	}
3434	/*
3435	 * For real BARs, we need to override the size that
3436	 * the driver requests, because that's what the BAR
3437	 * actually uses and we would otherwise have a
3438	 * situation where we might allocate the excess to
3439	 * another driver, which won't work.
3440	 */
3441	mapsize = pci_mapsize(testval);
3442	count = 1UL << mapsize;
3443	if (RF_ALIGNMENT(flags) < mapsize)
3444		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
3445
3446	/*
3447	 * Allocate enough resource, and then write back the
3448	 * appropriate bar for that resource.
3449	 */
3450	res = BUS_ALLOC_RESOURCE(device_get_parent(dev), child, type, rid,
3451	    start, end, count, flags);
3452	if (res == NULL) {
3453		device_printf(child,
3454		    "%#lx bytes of rid %#x res %d failed (%#lx, %#lx).\n",
3455		    count, *rid, type, start, end);
3456		goto out;
3457	}
3458	resource_list_add(rl, type, *rid, start, end, count);
3459	rle = resource_list_find(rl, type, *rid);
3460	if (rle == NULL)
3461		panic("pci_alloc_map: unexpectedly can't find resource.");
3462	rle->res = res;
3463	rle->start = rman_get_start(res);
3464	rle->end = rman_get_end(res);
3465	rle->count = count;
3466	if (bootverbose)
3467		device_printf(child,
3468		    "Lazy allocation of %#lx bytes rid %#x type %d at %#lx\n",
3469		    count, *rid, type, rman_get_start(res));
3470	map = rman_get_start(res);
3471out:;
3472	pci_write_config(child, *rid, map, 4);
3473	if (pci_maprange(testval) == 64)
3474		pci_write_config(child, *rid + 4, map >> 32, 4);
3475	return (res);
3476}
3477
3478
3479struct resource *
3480pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
3481		   u_long start, u_long end, u_long count, u_int flags)
3482{
3483	struct pci_devinfo *dinfo = device_get_ivars(child);
3484	struct resource_list *rl = &dinfo->resources;
3485	struct resource_list_entry *rle;
3486	pcicfgregs *cfg = &dinfo->cfg;
3487
3488	/*
3489	 * Perform lazy resource allocation
3490	 */
3491	if (device_get_parent(child) == dev) {
3492		switch (type) {
3493		case SYS_RES_IRQ:
3494			/*
3495			 * Can't alloc legacy interrupt once MSI messages
3496			 * have been allocated.
3497			 */
3498			if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
3499			    cfg->msix.msix_alloc > 0))
3500				return (NULL);
3501			/*
3502			 * If the child device doesn't have an
3503			 * interrupt routed and is deserving of an
3504			 * interrupt, try to assign it one.
3505			 */
3506			if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
3507			    (cfg->intpin != 0))
3508				pci_assign_interrupt(dev, child, 0);
3509			break;
3510		case SYS_RES_IOPORT:
3511		case SYS_RES_MEMORY:
3512			if (*rid < PCIR_BAR(cfg->nummaps)) {
3513				/*
3514				 * Enable the I/O mode.  We should
3515				 * also be assigning resources too
3516				 * when none are present.  The
3517				 * resource_list_alloc kind of sorta does
3518				 * this...
3519				 */
3520				if (PCI_ENABLE_IO(dev, child, type))
3521					return (NULL);
3522			}
3523			rle = resource_list_find(rl, type, *rid);
3524			if (rle == NULL)
3525				return (pci_alloc_map(dev, child, type, rid,
3526				    start, end, count, flags));
3527			break;
3528		}
3529		/*
3530		 * If we've already allocated the resource, then
3531		 * return it now.  But first we may need to activate
3532		 * it, since we don't allocate the resource as active
3533		 * above.  Normally this would be done down in the
3534		 * nexus, but since we short-circuit that path we have
3535		 * to do its job here.  Not sure if we should free the
3536		 * resource if it fails to activate.
3537		 */
3538		rle = resource_list_find(rl, type, *rid);
3539		if (rle != NULL && rle->res != NULL) {
3540			if (bootverbose)
3541				device_printf(child,
3542			    "Reserved %#lx bytes for rid %#x type %d at %#lx\n",
3543				    rman_get_size(rle->res), *rid, type,
3544				    rman_get_start(rle->res));
3545			if ((flags & RF_ACTIVE) &&
3546			    bus_generic_activate_resource(dev, child, type,
3547			    *rid, rle->res) != 0)
3548				return (NULL);
3549			return (rle->res);
3550		}
3551	}
3552	return (resource_list_alloc(rl, dev, child, type, rid,
3553	    start, end, count, flags));
3554}
3555
3556void
3557pci_delete_resource(device_t dev, device_t child, int type, int rid)
3558{
3559	struct pci_devinfo *dinfo;
3560	struct resource_list *rl;
3561	struct resource_list_entry *rle;
3562
3563	if (device_get_parent(child) != dev)
3564		return;
3565
3566	dinfo = device_get_ivars(child);
3567	rl = &dinfo->resources;
3568	rle = resource_list_find(rl, type, rid);
3569	if (rle) {
3570		if (rle->res) {
3571			if (rman_get_device(rle->res) != dev ||
3572			    rman_get_flags(rle->res) & RF_ACTIVE) {
3573				device_printf(dev, "delete_resource: "
3574				    "Resource still owned by child, oops. "
3575				    "(type=%d, rid=%d, addr=%lx)\n",
3576				    rle->type, rle->rid,
3577				    rman_get_start(rle->res));
3578				return;
3579			}
3580			bus_release_resource(dev, type, rid, rle->res);
3581		}
3582		resource_list_delete(rl, type, rid);
3583	}
3584	/*
3585	 * Why do we turn off the PCI configuration BAR when we delete a
3586	 * resource? -- imp
3587	 */
3588	pci_write_config(child, rid, 0, 4);
3589	BUS_DELETE_RESOURCE(device_get_parent(dev), child, type, rid);
3590}
3591
3592struct resource_list *
3593pci_get_resource_list (device_t dev, device_t child)
3594{
3595	struct pci_devinfo *dinfo = device_get_ivars(child);
3596
3597	return (&dinfo->resources);
3598}
3599
3600uint32_t
3601pci_read_config_method(device_t dev, device_t child, int reg, int width)
3602{
3603	struct pci_devinfo *dinfo = device_get_ivars(child);
3604	pcicfgregs *cfg = &dinfo->cfg;
3605
3606	return (PCIB_READ_CONFIG(device_get_parent(dev),
3607	    cfg->bus, cfg->slot, cfg->func, reg, width));
3608}
3609
3610void
3611pci_write_config_method(device_t dev, device_t child, int reg,
3612    uint32_t val, int width)
3613{
3614	struct pci_devinfo *dinfo = device_get_ivars(child);
3615	pcicfgregs *cfg = &dinfo->cfg;
3616
3617	PCIB_WRITE_CONFIG(device_get_parent(dev),
3618	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
3619}
3620
3621int
3622pci_child_location_str_method(device_t dev, device_t child, char *buf,
3623    size_t buflen)
3624{
3625
3626	snprintf(buf, buflen, "slot=%d function=%d", pci_get_slot(child),
3627	    pci_get_function(child));
3628	return (0);
3629}
3630
3631int
3632pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
3633    size_t buflen)
3634{
3635	struct pci_devinfo *dinfo;
3636	pcicfgregs *cfg;
3637
3638	dinfo = device_get_ivars(child);
3639	cfg = &dinfo->cfg;
3640	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
3641	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
3642	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
3643	    cfg->progif);
3644	return (0);
3645}
3646
3647int
3648pci_assign_interrupt_method(device_t dev, device_t child)
3649{
3650	struct pci_devinfo *dinfo = device_get_ivars(child);
3651	pcicfgregs *cfg = &dinfo->cfg;
3652
3653	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
3654	    cfg->intpin));
3655}
3656
3657static int
3658pci_modevent(module_t mod, int what, void *arg)
3659{
3660	static struct cdev *pci_cdev;
3661
3662	switch (what) {
3663	case MOD_LOAD:
3664		STAILQ_INIT(&pci_devq);
3665		pci_generation = 0;
3666		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
3667		    "pci");
3668		pci_load_vendor_data();
3669		break;
3670
3671	case MOD_UNLOAD:
3672		destroy_dev(pci_cdev);
3673		break;
3674	}
3675
3676	return (0);
3677}
3678
3679void
3680pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
3681{
3682	int i;
3683
3684	/*
3685	 * Only do header type 0 devices.  Type 1 devices are bridges,
3686	 * which we know need special treatment.  Type 2 devices are
3687	 * cardbus bridges which also require special treatment.
3688	 * Other types are unknown, and we err on the side of safety
3689	 * by ignoring them.
3690	 */
3691	if (dinfo->cfg.hdrtype != 0)
3692		return;
3693
3694	/*
3695	 * Restore the device to full power mode.  We must do this
3696	 * before we restore the registers because moving from D3 to
3697	 * D0 will cause the chip's BARs and some other registers to
3698	 * be reset to some unknown power on reset values.  Cut down
3699	 * the noise on boot by doing nothing if we are already in
3700	 * state D0.
3701	 */
3702	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
3703		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
3704	}
3705	for (i = 0; i < dinfo->cfg.nummaps; i++)
3706		pci_write_config(dev, PCIR_BAR(i), dinfo->cfg.bar[i], 4);
3707	pci_write_config(dev, PCIR_BIOS, dinfo->cfg.bios, 4);
3708	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
3709	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
3710	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
3711	pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
3712	pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
3713	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
3714	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
3715	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
3716	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
3717
3718	/* Restore MSI and MSI-X configurations if they are present. */
3719	if (dinfo->cfg.msi.msi_location != 0)
3720		pci_resume_msi(dev);
3721	if (dinfo->cfg.msix.msix_location != 0)
3722		pci_resume_msix(dev);
3723}
3724
3725void
3726pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
3727{
3728	int i;
3729	uint32_t cls;
3730	int ps;
3731
3732	/*
3733	 * Only do header type 0 devices.  Type 1 devices are bridges, which
3734	 * we know need special treatment.  Type 2 devices are cardbus bridges
3735	 * which also require special treatment.  Other types are unknown, and
3736	 * we err on the side of safety by ignoring them.  Powering down
3737	 * bridges should not be undertaken lightly.
3738	 */
3739	if (dinfo->cfg.hdrtype != 0)
3740		return;
3741	for (i = 0; i < dinfo->cfg.nummaps; i++)
3742		dinfo->cfg.bar[i] = pci_read_config(dev, PCIR_BAR(i), 4);
3743	dinfo->cfg.bios = pci_read_config(dev, PCIR_BIOS, 4);
3744
3745	/*
3746	 * Some drivers apparently write to these registers w/o updating our
3747	 * cached copy.  No harm happens if we update the copy, so do so here
3748	 * so we can restore them.  The COMMAND register is modified by the
3749	 * bus w/o updating the cache.  This should represent the normally
3750	 * writable portion of the 'defined' part of type 0 headers.  In
3751	 * theory we also need to save/restore the PCI capability structures
3752	 * we know about, but apart from power we don't know any that are
3753	 * writable.
3754	 */
3755	dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
3756	dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
3757	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
3758	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
3759	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
3760	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
3761	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
3762	dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
3763	dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
3764	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
3765	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
3766	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
3767	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
3768	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
3769	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
3770
3771	/*
3772	 * don't set the state for display devices, base peripherals and
3773	 * memory devices since bad things happen when they are powered down.
3774	 * We should (a) have drivers that can easily detach and (b) use
3775	 * generic drivers for these devices so that some device actually
3776	 * attaches.  We need to make sure that when we implement (a) we don't
3777	 * power the device down on a reattach.
3778	 */
3779	cls = pci_get_class(dev);
3780	if (!setstate)
3781		return;
3782	switch (pci_do_power_nodriver)
3783	{
3784		case 0:		/* NO powerdown at all */
3785			return;
3786		case 1:		/* Conservative about what to power down */
3787			if (cls == PCIC_STORAGE)
3788				return;
3789			/*FALLTHROUGH*/
3790		case 2:		/* Agressive about what to power down */
3791			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
3792			    cls == PCIC_BASEPERIPH)
3793				return;
3794			/*FALLTHROUGH*/
3795		case 3:		/* Power down everything */
3796			break;
3797	}
3798	/*
3799	 * PCI spec says we can only go into D3 state from D0 state.
3800	 * Transition from D[12] into D0 before going to D3 state.
3801	 */
3802	ps = pci_get_powerstate(dev);
3803	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
3804		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
3805	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
3806		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
3807}
3808