xref: /freebsd/sys/dev/pci/pci.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_iommu.h"
36 #include "opt_bus.h"
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/endian.h>
41 #include <sys/eventhandler.h>
42 #include <sys/fcntl.h>
43 #include <sys/kernel.h>
44 #include <sys/limits.h>
45 #include <sys/linker.h>
46 #include <sys/malloc.h>
47 #include <sys/module.h>
48 #include <sys/queue.h>
49 #include <sys/sbuf.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/taskqueue.h>
53 #include <sys/tree.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_extern.h>
58 
59 #include <sys/bus.h>
60 #include <machine/bus.h>
61 #include <sys/rman.h>
62 #include <machine/resource.h>
63 #include <machine/stdarg.h>
64 
65 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
66 #include <machine/intr_machdep.h>
67 #endif
68 
69 #include <sys/pciio.h>
70 #include <dev/pci/pcireg.h>
71 #include <dev/pci/pcivar.h>
72 #include <dev/pci/pci_private.h>
73 
74 #ifdef PCI_IOV
75 #include <sys/nv.h>
76 #include <dev/pci/pci_iov_private.h>
77 #endif
78 
79 #include <dev/usb/controller/xhcireg.h>
80 #include <dev/usb/controller/ehcireg.h>
81 #include <dev/usb/controller/ohcireg.h>
82 #include <dev/usb/controller/uhcireg.h>
83 
84 #include <dev/iommu/iommu.h>
85 
86 #include "pcib_if.h"
87 #include "pci_if.h"
88 
89 #define	PCIR_IS_BIOS(cfg, reg)						\
90 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
91 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
92 
93 static int		pci_has_quirk(uint32_t devid, int quirk);
94 static pci_addr_t	pci_mapbase(uint64_t mapreg);
95 static const char	*pci_maptype(uint64_t mapreg);
96 static int		pci_maprange(uint64_t mapreg);
97 static pci_addr_t	pci_rombase(uint64_t mapreg);
98 static int		pci_romsize(uint64_t testval);
99 static void		pci_fixancient(pcicfgregs *cfg);
100 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
101 
102 static int		pci_porten(device_t dev);
103 static int		pci_memen(device_t dev);
104 static void		pci_assign_interrupt(device_t bus, device_t dev,
105 			    int force_route);
106 static int		pci_add_map(device_t bus, device_t dev, int reg,
107 			    struct resource_list *rl, int force, int prefetch);
108 static int		pci_probe(device_t dev);
109 static void		pci_load_vendor_data(void);
110 static int		pci_describe_parse_line(char **ptr, int *vendor,
111 			    int *device, char **desc);
112 static char		*pci_describe_device(device_t dev);
113 static int		pci_modevent(module_t mod, int what, void *arg);
114 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
115 			    pcicfgregs *cfg);
116 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
117 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
118 			    int reg, uint32_t *data);
119 #if 0
120 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
121 			    int reg, uint32_t data);
122 #endif
123 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
124 static void		pci_mask_msix(device_t dev, u_int index);
125 static void		pci_unmask_msix(device_t dev, u_int index);
126 static int		pci_msi_blacklisted(void);
127 static int		pci_msix_blacklisted(void);
128 static void		pci_resume_msi(device_t dev);
129 static void		pci_resume_msix(device_t dev);
130 static int		pci_remap_intr_method(device_t bus, device_t dev,
131 			    u_int irq);
132 static void		pci_hint_device_unit(device_t acdev, device_t child,
133 			    const char *name, int *unitp);
134 static int		pci_reset_post(device_t dev, device_t child);
135 static int		pci_reset_prepare(device_t dev, device_t child);
136 static int		pci_reset_child(device_t dev, device_t child,
137 			    int flags);
138 
139 static int		pci_get_id_method(device_t dev, device_t child,
140 			    enum pci_id_type type, uintptr_t *rid);
141 
142 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
143     int b, int s, int f, uint16_t vid, uint16_t did);
144 
145 static device_method_t pci_methods[] = {
146 	/* Device interface */
147 	DEVMETHOD(device_probe,		pci_probe),
148 	DEVMETHOD(device_attach,	pci_attach),
149 	DEVMETHOD(device_detach,	pci_detach),
150 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
151 	DEVMETHOD(device_suspend,	bus_generic_suspend),
152 	DEVMETHOD(device_resume,	pci_resume),
153 
154 	/* Bus interface */
155 	DEVMETHOD(bus_print_child,	pci_print_child),
156 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
157 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
158 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
159 	DEVMETHOD(bus_driver_added,	pci_driver_added),
160 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
161 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
162 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
163 	DEVMETHOD(bus_reset_post,	pci_reset_post),
164 	DEVMETHOD(bus_reset_child,	pci_reset_child),
165 
166 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
167 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
168 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
169 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
170 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
171 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
172 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
173 	DEVMETHOD(bus_release_resource,	pci_release_resource),
174 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
175 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
176 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
177 	DEVMETHOD(bus_child_detached,	pci_child_detached),
178 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
179 	DEVMETHOD(bus_child_location,	pci_child_location_method),
180 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
181 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
182 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
183 	DEVMETHOD(bus_resume_child,	pci_resume_child),
184 	DEVMETHOD(bus_rescan,		pci_rescan_method),
185 
186 	/* PCI interface */
187 	DEVMETHOD(pci_read_config,	pci_read_config_method),
188 	DEVMETHOD(pci_write_config,	pci_write_config_method),
189 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
190 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
191 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
192 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
193 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
194 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
195 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
196 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
197 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
198 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
199 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
200 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
201 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
202 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
203 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
204 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
205 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
206 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
207 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
208 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
209 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
210 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
211 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
212 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
213 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
214 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
215 	DEVMETHOD(pci_get_id,		pci_get_id_method),
216 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
217 	DEVMETHOD(pci_child_added,	pci_child_added_method),
218 #ifdef PCI_IOV
219 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
220 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
221 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
222 #endif
223 
224 	DEVMETHOD_END
225 };
226 
227 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
228 
229 static devclass_t pci_devclass;
230 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
231     BUS_PASS_BUS);
232 MODULE_VERSION(pci, 1);
233 
234 static char	*pci_vendordata;
235 static size_t	pci_vendordata_size;
236 
237 struct pci_quirk {
238 	uint32_t devid;	/* Vendor/device of the card */
239 	int	type;
240 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
241 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
242 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
243 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
244 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
245 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
246 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
247 	int	arg1;
248 	int	arg2;
249 };
250 
251 static const struct pci_quirk pci_quirks[] = {
252 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
253 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
254 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
255 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
256 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
257 
258 	/*
259 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
260 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
261 	 */
262 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
263 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
264 
265 	/*
266 	 * MSI doesn't work on earlier Intel chipsets including
267 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
268 	 */
269 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
273 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
274 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
275 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
276 
277 	/*
278 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
279 	 * bridge.
280 	 */
281 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
282 
283 	/*
284 	 * Some virtualization environments emulate an older chipset
285 	 * but support MSI just fine.  QEMU uses the Intel 82440.
286 	 */
287 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
288 
289 	/*
290 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
291 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
292 	 * It prevents us from attaching hpet(4) when the bit is unset.
293 	 * Note this quirk only affects SB600 revision A13 and earlier.
294 	 * For SB600 A21 and later, firmware must set the bit to hide it.
295 	 * For SB700 and later, it is unused and hardcoded to zero.
296 	 */
297 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
298 
299 	/*
300 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
301 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
302 	 * of the command register is set.
303 	 */
304 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
306 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
307 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
309 
310 	/*
311 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
312 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
313 	 */
314 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
315 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
316 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
317 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
318 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
319 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
320 
321 	/*
322 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
323 	 * expected place.
324 	 */
325 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
326 	{ 0 }
327 };
328 
329 /* map register information */
330 #define	PCI_MAPMEM	0x01	/* memory map */
331 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
332 #define	PCI_MAPPORT	0x04	/* port map */
333 
334 struct devlist pci_devq;
335 uint32_t pci_generation;
336 uint32_t pci_numdevs = 0;
337 static int pcie_chipset, pcix_chipset;
338 
339 /* sysctl vars */
340 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
341     "PCI bus tuning parameters");
342 
343 static int pci_enable_io_modes = 1;
344 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
345     &pci_enable_io_modes, 1,
346     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
347     " enable these bits correctly.  We'd like to do this all the time, but"
348     " there are some peripherals that this causes problems with.");
349 
350 static int pci_do_realloc_bars = 1;
351 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
352     &pci_do_realloc_bars, 0,
353     "Attempt to allocate a new range for any BARs whose original "
354     "firmware-assigned ranges fail to allocate during the initial device scan.");
355 
356 static int pci_do_power_nodriver = 0;
357 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
358     &pci_do_power_nodriver, 0,
359     "Place a function into D3 state when no driver attaches to it.  0 means"
360     " disable.  1 means conservatively place devices into D3 state.  2 means"
361     " aggressively place devices into D3 state.  3 means put absolutely"
362     " everything in D3 state.");
363 
364 int pci_do_power_resume = 1;
365 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
366     &pci_do_power_resume, 1,
367   "Transition from D3 -> D0 on resume.");
368 
369 int pci_do_power_suspend = 1;
370 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
371     &pci_do_power_suspend, 1,
372   "Transition from D0 -> D3 on suspend.");
373 
374 static int pci_do_msi = 1;
375 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
376     "Enable support for MSI interrupts");
377 
378 static int pci_do_msix = 1;
379 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
380     "Enable support for MSI-X interrupts");
381 
382 static int pci_msix_rewrite_table = 0;
383 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
384     &pci_msix_rewrite_table, 0,
385     "Rewrite entire MSI-X table when updating MSI-X entries");
386 
387 static int pci_honor_msi_blacklist = 1;
388 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
389     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
390 
391 #if defined(__i386__) || defined(__amd64__)
392 static int pci_usb_takeover = 1;
393 #else
394 static int pci_usb_takeover = 0;
395 #endif
396 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
397     &pci_usb_takeover, 1,
398     "Enable early takeover of USB controllers. Disable this if you depend on"
399     " BIOS emulation of USB devices, that is you use USB devices (like"
400     " keyboard or mouse) but do not load USB drivers");
401 
402 static int pci_clear_bars;
403 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
404     "Ignore firmware-assigned resources for BARs.");
405 
406 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
407 static int pci_clear_buses;
408 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
409     "Ignore firmware-assigned bus numbers.");
410 #endif
411 
412 static int pci_enable_ari = 1;
413 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
414     0, "Enable support for PCIe Alternative RID Interpretation");
415 
416 int pci_enable_aspm = 1;
417 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
418     0, "Enable support for PCIe Active State Power Management");
419 
420 static int pci_clear_aer_on_attach = 0;
421 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
422     &pci_clear_aer_on_attach, 0,
423     "Clear port and device AER state on driver attach");
424 
425 static int
426 pci_has_quirk(uint32_t devid, int quirk)
427 {
428 	const struct pci_quirk *q;
429 
430 	for (q = &pci_quirks[0]; q->devid; q++) {
431 		if (q->devid == devid && q->type == quirk)
432 			return (1);
433 	}
434 	return (0);
435 }
436 
437 /* Find a device_t by bus/slot/function in domain 0 */
438 
439 device_t
440 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
441 {
442 
443 	return (pci_find_dbsf(0, bus, slot, func));
444 }
445 
446 /* Find a device_t by domain/bus/slot/function */
447 
448 device_t
449 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
450 {
451 	struct pci_devinfo *dinfo = NULL;
452 
453 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
454 		if ((dinfo->cfg.domain == domain) &&
455 		    (dinfo->cfg.bus == bus) &&
456 		    (dinfo->cfg.slot == slot) &&
457 		    (dinfo->cfg.func == func)) {
458 			break;
459 		}
460 	}
461 
462 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
463 }
464 
465 /* Find a device_t by vendor/device ID */
466 
467 device_t
468 pci_find_device(uint16_t vendor, uint16_t device)
469 {
470 	struct pci_devinfo *dinfo;
471 
472 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
473 		if ((dinfo->cfg.vendor == vendor) &&
474 		    (dinfo->cfg.device == device)) {
475 			return (dinfo->cfg.dev);
476 		}
477 	}
478 
479 	return (NULL);
480 }
481 
482 device_t
483 pci_find_class(uint8_t class, uint8_t subclass)
484 {
485 	struct pci_devinfo *dinfo;
486 
487 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
488 		if (dinfo->cfg.baseclass == class &&
489 		    dinfo->cfg.subclass == subclass) {
490 			return (dinfo->cfg.dev);
491 		}
492 	}
493 
494 	return (NULL);
495 }
496 
497 device_t
498 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
499 {
500 	struct pci_devinfo *dinfo;
501 	bool found = false;
502 
503 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
504 		if (from != NULL && found == false) {
505 			if (from != dinfo->cfg.dev)
506 				continue;
507 			found = true;
508 			continue;
509 		}
510 		if (dinfo->cfg.baseclass == class &&
511 		    dinfo->cfg.subclass == subclass) {
512 			return (dinfo->cfg.dev);
513 		}
514 	}
515 
516 	return (NULL);
517 }
518 
519 static int
520 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
521 {
522 	va_list ap;
523 	int retval;
524 
525 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
526 	    cfg->func);
527 	va_start(ap, fmt);
528 	retval += vprintf(fmt, ap);
529 	va_end(ap);
530 	return (retval);
531 }
532 
533 /* return base address of memory or port map */
534 
535 static pci_addr_t
536 pci_mapbase(uint64_t mapreg)
537 {
538 
539 	if (PCI_BAR_MEM(mapreg))
540 		return (mapreg & PCIM_BAR_MEM_BASE);
541 	else
542 		return (mapreg & PCIM_BAR_IO_BASE);
543 }
544 
545 /* return map type of memory or port map */
546 
547 static const char *
548 pci_maptype(uint64_t mapreg)
549 {
550 
551 	if (PCI_BAR_IO(mapreg))
552 		return ("I/O Port");
553 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
554 		return ("Prefetchable Memory");
555 	return ("Memory");
556 }
557 
558 /* return log2 of map size decoded for memory or port map */
559 
560 int
561 pci_mapsize(uint64_t testval)
562 {
563 	int ln2size;
564 
565 	testval = pci_mapbase(testval);
566 	ln2size = 0;
567 	if (testval != 0) {
568 		while ((testval & 1) == 0)
569 		{
570 			ln2size++;
571 			testval >>= 1;
572 		}
573 	}
574 	return (ln2size);
575 }
576 
577 /* return base address of device ROM */
578 
579 static pci_addr_t
580 pci_rombase(uint64_t mapreg)
581 {
582 
583 	return (mapreg & PCIM_BIOS_ADDR_MASK);
584 }
585 
586 /* return log2 of map size decided for device ROM */
587 
588 static int
589 pci_romsize(uint64_t testval)
590 {
591 	int ln2size;
592 
593 	testval = pci_rombase(testval);
594 	ln2size = 0;
595 	if (testval != 0) {
596 		while ((testval & 1) == 0)
597 		{
598 			ln2size++;
599 			testval >>= 1;
600 		}
601 	}
602 	return (ln2size);
603 }
604 
605 /* return log2 of address range supported by map register */
606 
607 static int
608 pci_maprange(uint64_t mapreg)
609 {
610 	int ln2range = 0;
611 
612 	if (PCI_BAR_IO(mapreg))
613 		ln2range = 32;
614 	else
615 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
616 		case PCIM_BAR_MEM_32:
617 			ln2range = 32;
618 			break;
619 		case PCIM_BAR_MEM_1MB:
620 			ln2range = 20;
621 			break;
622 		case PCIM_BAR_MEM_64:
623 			ln2range = 64;
624 			break;
625 		}
626 	return (ln2range);
627 }
628 
629 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
630 
631 static void
632 pci_fixancient(pcicfgregs *cfg)
633 {
634 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
635 		return;
636 
637 	/* PCI to PCI bridges use header type 1 */
638 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
639 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
640 }
641 
642 /* extract header type specific config data */
643 
644 static void
645 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
646 {
647 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
648 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
649 	case PCIM_HDRTYPE_NORMAL:
650 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
651 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
652 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
653 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
654 		cfg->nummaps	    = PCI_MAXMAPS_0;
655 		break;
656 	case PCIM_HDRTYPE_BRIDGE:
657 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
658 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
659 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
660 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
661 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
662 		cfg->nummaps	    = PCI_MAXMAPS_1;
663 		break;
664 	case PCIM_HDRTYPE_CARDBUS:
665 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
666 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
667 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
668 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
669 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
670 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
671 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
672 		cfg->nummaps	    = PCI_MAXMAPS_2;
673 		break;
674 	}
675 #undef REG
676 }
677 
678 /* read configuration header into pcicfgregs structure */
679 struct pci_devinfo *
680 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
681 {
682 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
683 	uint16_t vid, did;
684 
685 	vid = REG(PCIR_VENDOR, 2);
686 	did = REG(PCIR_DEVICE, 2);
687 	if (vid != 0xffff)
688 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
689 
690 	return (NULL);
691 }
692 
693 struct pci_devinfo *
694 pci_alloc_devinfo_method(device_t dev)
695 {
696 
697 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
698 	    M_WAITOK | M_ZERO));
699 }
700 
701 static struct pci_devinfo *
702 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
703     uint16_t vid, uint16_t did)
704 {
705 	struct pci_devinfo *devlist_entry;
706 	pcicfgregs *cfg;
707 
708 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
709 
710 	cfg = &devlist_entry->cfg;
711 
712 	cfg->domain		= d;
713 	cfg->bus		= b;
714 	cfg->slot		= s;
715 	cfg->func		= f;
716 	cfg->vendor		= vid;
717 	cfg->device		= did;
718 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
719 	cfg->statreg		= REG(PCIR_STATUS, 2);
720 	cfg->baseclass		= REG(PCIR_CLASS, 1);
721 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
722 	cfg->progif		= REG(PCIR_PROGIF, 1);
723 	cfg->revid		= REG(PCIR_REVID, 1);
724 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
725 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
726 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
727 	cfg->intpin		= REG(PCIR_INTPIN, 1);
728 	cfg->intline		= REG(PCIR_INTLINE, 1);
729 
730 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
731 	cfg->hdrtype		&= ~PCIM_MFDEV;
732 	STAILQ_INIT(&cfg->maps);
733 
734 	cfg->iov		= NULL;
735 
736 	pci_fixancient(cfg);
737 	pci_hdrtypedata(pcib, b, s, f, cfg);
738 
739 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
740 		pci_read_cap(pcib, cfg);
741 
742 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
743 
744 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
745 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
746 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
747 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
748 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
749 
750 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
751 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
752 	devlist_entry->conf.pc_vendor = cfg->vendor;
753 	devlist_entry->conf.pc_device = cfg->device;
754 
755 	devlist_entry->conf.pc_class = cfg->baseclass;
756 	devlist_entry->conf.pc_subclass = cfg->subclass;
757 	devlist_entry->conf.pc_progif = cfg->progif;
758 	devlist_entry->conf.pc_revid = cfg->revid;
759 
760 	pci_numdevs++;
761 	pci_generation++;
762 
763 	return (devlist_entry);
764 }
765 #undef REG
766 
767 static void
768 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
769 {
770 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
771     cfg->ea.ea_location + (n), w)
772 	int num_ent;
773 	int ptr;
774 	int a, b;
775 	uint32_t val;
776 	int ent_size;
777 	uint32_t dw[4];
778 	uint64_t base, max_offset;
779 	struct pci_ea_entry *eae;
780 
781 	if (cfg->ea.ea_location == 0)
782 		return;
783 
784 	STAILQ_INIT(&cfg->ea.ea_entries);
785 
786 	/* Determine the number of entries */
787 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
788 	num_ent &= PCIM_EA_NUM_ENT_MASK;
789 
790 	/* Find the first entry to care of */
791 	ptr = PCIR_EA_FIRST_ENT;
792 
793 	/* Skip DWORD 2 for type 1 functions */
794 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
795 		ptr += 4;
796 
797 	for (a = 0; a < num_ent; a++) {
798 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
799 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
800 
801 		/* Read a number of dwords in the entry */
802 		val = REG(ptr, 4);
803 		ptr += 4;
804 		ent_size = (val & PCIM_EA_ES);
805 
806 		for (b = 0; b < ent_size; b++) {
807 			dw[b] = REG(ptr, 4);
808 			ptr += 4;
809 		}
810 
811 		eae->eae_flags = val;
812 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
813 
814 		base = dw[0] & PCIM_EA_FIELD_MASK;
815 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
816 		b = 2;
817 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
818 			base |= (uint64_t)dw[b] << 32UL;
819 			b++;
820 		}
821 		if (((dw[1] & PCIM_EA_IS_64) != 0)
822 		    && (b < ent_size)) {
823 			max_offset |= (uint64_t)dw[b] << 32UL;
824 			b++;
825 		}
826 
827 		eae->eae_base = base;
828 		eae->eae_max_offset = max_offset;
829 
830 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
831 
832 		if (bootverbose) {
833 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
834 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
835 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
836 		}
837 	}
838 }
839 #undef REG
840 
841 static void
842 pci_read_cap(device_t pcib, pcicfgregs *cfg)
843 {
844 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
845 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
846 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
847 	uint64_t addr;
848 #endif
849 	uint32_t val;
850 	int	ptr, nextptr, ptrptr;
851 
852 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
853 	case PCIM_HDRTYPE_NORMAL:
854 	case PCIM_HDRTYPE_BRIDGE:
855 		ptrptr = PCIR_CAP_PTR;
856 		break;
857 	case PCIM_HDRTYPE_CARDBUS:
858 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
859 		break;
860 	default:
861 		return;		/* no extended capabilities support */
862 	}
863 	nextptr = REG(ptrptr, 1);	/* sanity check? */
864 
865 	/*
866 	 * Read capability entries.
867 	 */
868 	while (nextptr != 0) {
869 		/* Sanity check */
870 		if (nextptr > 255) {
871 			printf("illegal PCI extended capability offset %d\n",
872 			    nextptr);
873 			return;
874 		}
875 		/* Find the next entry */
876 		ptr = nextptr;
877 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
878 
879 		/* Process this entry */
880 		switch (REG(ptr + PCICAP_ID, 1)) {
881 		case PCIY_PMG:		/* PCI power management */
882 			if (cfg->pp.pp_cap == 0) {
883 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
884 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
885 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
886 				if ((nextptr - ptr) > PCIR_POWER_DATA)
887 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
888 			}
889 			break;
890 		case PCIY_HT:		/* HyperTransport */
891 			/* Determine HT-specific capability type. */
892 			val = REG(ptr + PCIR_HT_COMMAND, 2);
893 
894 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
895 				cfg->ht.ht_slave = ptr;
896 
897 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
898 			switch (val & PCIM_HTCMD_CAP_MASK) {
899 			case PCIM_HTCAP_MSI_MAPPING:
900 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
901 					/* Sanity check the mapping window. */
902 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
903 					    4);
904 					addr <<= 32;
905 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
906 					    4);
907 					if (addr != MSI_INTEL_ADDR_BASE)
908 						device_printf(pcib,
909 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
910 						    cfg->domain, cfg->bus,
911 						    cfg->slot, cfg->func,
912 						    (long long)addr);
913 				} else
914 					addr = MSI_INTEL_ADDR_BASE;
915 
916 				cfg->ht.ht_msimap = ptr;
917 				cfg->ht.ht_msictrl = val;
918 				cfg->ht.ht_msiaddr = addr;
919 				break;
920 			}
921 #endif
922 			break;
923 		case PCIY_MSI:		/* PCI MSI */
924 			cfg->msi.msi_location = ptr;
925 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
926 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
927 						     PCIM_MSICTRL_MMC_MASK)>>1);
928 			break;
929 		case PCIY_MSIX:		/* PCI MSI-X */
930 			cfg->msix.msix_location = ptr;
931 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
932 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
933 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
934 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
935 			cfg->msix.msix_table_bar = PCIR_BAR(val &
936 			    PCIM_MSIX_BIR_MASK);
937 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
938 			val = REG(ptr + PCIR_MSIX_PBA, 4);
939 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
940 			    PCIM_MSIX_BIR_MASK);
941 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
942 			break;
943 		case PCIY_VPD:		/* PCI Vital Product Data */
944 			cfg->vpd.vpd_reg = ptr;
945 			break;
946 		case PCIY_SUBVENDOR:
947 			/* Should always be true. */
948 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
949 			    PCIM_HDRTYPE_BRIDGE) {
950 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
951 				cfg->subvendor = val & 0xffff;
952 				cfg->subdevice = val >> 16;
953 			}
954 			break;
955 		case PCIY_PCIX:		/* PCI-X */
956 			/*
957 			 * Assume we have a PCI-X chipset if we have
958 			 * at least one PCI-PCI bridge with a PCI-X
959 			 * capability.  Note that some systems with
960 			 * PCI-express or HT chipsets might match on
961 			 * this check as well.
962 			 */
963 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
964 			    PCIM_HDRTYPE_BRIDGE)
965 				pcix_chipset = 1;
966 			cfg->pcix.pcix_location = ptr;
967 			break;
968 		case PCIY_EXPRESS:	/* PCI-express */
969 			/*
970 			 * Assume we have a PCI-express chipset if we have
971 			 * at least one PCI-express device.
972 			 */
973 			pcie_chipset = 1;
974 			cfg->pcie.pcie_location = ptr;
975 			val = REG(ptr + PCIER_FLAGS, 2);
976 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
977 			break;
978 		case PCIY_EA:		/* Enhanced Allocation */
979 			cfg->ea.ea_location = ptr;
980 			pci_ea_fill_info(pcib, cfg);
981 			break;
982 		default:
983 			break;
984 		}
985 	}
986 
987 #if defined(__powerpc__)
988 	/*
989 	 * Enable the MSI mapping window for all HyperTransport
990 	 * slaves.  PCI-PCI bridges have their windows enabled via
991 	 * PCIB_MAP_MSI().
992 	 */
993 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
994 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
995 		device_printf(pcib,
996 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
997 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
998 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
999 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1000 		     2);
1001 	}
1002 #endif
1003 /* REG and WREG use carry through to next functions */
1004 }
1005 
1006 /*
1007  * PCI Vital Product Data
1008  */
1009 
1010 #define	PCI_VPD_TIMEOUT		1000000
1011 
1012 static int
1013 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1014 {
1015 	int count = PCI_VPD_TIMEOUT;
1016 
1017 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1018 
1019 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1020 
1021 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1022 		if (--count < 0)
1023 			return (ENXIO);
1024 		DELAY(1);	/* limit looping */
1025 	}
1026 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1027 
1028 	return (0);
1029 }
1030 
1031 #if 0
1032 static int
1033 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1034 {
1035 	int count = PCI_VPD_TIMEOUT;
1036 
1037 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1038 
1039 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1040 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1041 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1042 		if (--count < 0)
1043 			return (ENXIO);
1044 		DELAY(1);	/* limit looping */
1045 	}
1046 
1047 	return (0);
1048 }
1049 #endif
1050 
1051 #undef PCI_VPD_TIMEOUT
1052 
1053 struct vpd_readstate {
1054 	device_t	pcib;
1055 	pcicfgregs	*cfg;
1056 	uint32_t	val;
1057 	int		bytesinval;
1058 	int		off;
1059 	uint8_t		cksum;
1060 };
1061 
1062 static int
1063 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1064 {
1065 	uint32_t reg;
1066 	uint8_t byte;
1067 
1068 	if (vrs->bytesinval == 0) {
1069 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1070 			return (ENXIO);
1071 		vrs->val = le32toh(reg);
1072 		vrs->off += 4;
1073 		byte = vrs->val & 0xff;
1074 		vrs->bytesinval = 3;
1075 	} else {
1076 		vrs->val = vrs->val >> 8;
1077 		byte = vrs->val & 0xff;
1078 		vrs->bytesinval--;
1079 	}
1080 
1081 	vrs->cksum += byte;
1082 	*data = byte;
1083 	return (0);
1084 }
1085 
1086 static void
1087 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1088 {
1089 	struct vpd_readstate vrs;
1090 	int state;
1091 	int name;
1092 	int remain;
1093 	int i;
1094 	int alloc, off;		/* alloc/off for RO/W arrays */
1095 	int cksumvalid;
1096 	int dflen;
1097 	uint8_t byte;
1098 	uint8_t byte2;
1099 
1100 	/* init vpd reader */
1101 	vrs.bytesinval = 0;
1102 	vrs.off = 0;
1103 	vrs.pcib = pcib;
1104 	vrs.cfg = cfg;
1105 	vrs.cksum = 0;
1106 
1107 	state = 0;
1108 	name = remain = i = 0;	/* shut up stupid gcc */
1109 	alloc = off = 0;	/* shut up stupid gcc */
1110 	dflen = 0;		/* shut up stupid gcc */
1111 	cksumvalid = -1;
1112 	while (state >= 0) {
1113 		if (vpd_nextbyte(&vrs, &byte)) {
1114 			state = -2;
1115 			break;
1116 		}
1117 #if 0
1118 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1119 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1120 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1121 #endif
1122 		switch (state) {
1123 		case 0:		/* item name */
1124 			if (byte & 0x80) {
1125 				if (vpd_nextbyte(&vrs, &byte2)) {
1126 					state = -2;
1127 					break;
1128 				}
1129 				remain = byte2;
1130 				if (vpd_nextbyte(&vrs, &byte2)) {
1131 					state = -2;
1132 					break;
1133 				}
1134 				remain |= byte2 << 8;
1135 				name = byte & 0x7f;
1136 			} else {
1137 				remain = byte & 0x7;
1138 				name = (byte >> 3) & 0xf;
1139 			}
1140 			if (vrs.off + remain - vrs.bytesinval > 0x8000) {
1141 				pci_printf(cfg,
1142 				    "VPD data overflow, remain %#x\n", remain);
1143 				state = -1;
1144 				break;
1145 			}
1146 			switch (name) {
1147 			case 0x2:	/* String */
1148 				cfg->vpd.vpd_ident = malloc(remain + 1,
1149 				    M_DEVBUF, M_WAITOK);
1150 				i = 0;
1151 				state = 1;
1152 				break;
1153 			case 0xf:	/* End */
1154 				state = -1;
1155 				break;
1156 			case 0x10:	/* VPD-R */
1157 				alloc = 8;
1158 				off = 0;
1159 				cfg->vpd.vpd_ros = malloc(alloc *
1160 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1161 				    M_WAITOK | M_ZERO);
1162 				state = 2;
1163 				break;
1164 			case 0x11:	/* VPD-W */
1165 				alloc = 8;
1166 				off = 0;
1167 				cfg->vpd.vpd_w = malloc(alloc *
1168 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1169 				    M_WAITOK | M_ZERO);
1170 				state = 5;
1171 				break;
1172 			default:	/* Invalid data, abort */
1173 				state = -1;
1174 				break;
1175 			}
1176 			break;
1177 
1178 		case 1:	/* Identifier String */
1179 			cfg->vpd.vpd_ident[i++] = byte;
1180 			remain--;
1181 			if (remain == 0)  {
1182 				cfg->vpd.vpd_ident[i] = '\0';
1183 				state = 0;
1184 			}
1185 			break;
1186 
1187 		case 2:	/* VPD-R Keyword Header */
1188 			if (off == alloc) {
1189 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1190 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1191 				    M_DEVBUF, M_WAITOK | M_ZERO);
1192 			}
1193 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1194 			if (vpd_nextbyte(&vrs, &byte2)) {
1195 				state = -2;
1196 				break;
1197 			}
1198 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1199 			if (vpd_nextbyte(&vrs, &byte2)) {
1200 				state = -2;
1201 				break;
1202 			}
1203 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1204 			if (dflen == 0 &&
1205 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1206 			    2) == 0) {
1207 				/*
1208 				 * if this happens, we can't trust the rest
1209 				 * of the VPD.
1210 				 */
1211 				pci_printf(cfg, "bad keyword length: %d\n",
1212 				    dflen);
1213 				cksumvalid = 0;
1214 				state = -1;
1215 				break;
1216 			} else if (dflen == 0) {
1217 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1218 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1219 				    M_DEVBUF, M_WAITOK);
1220 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1221 			} else
1222 				cfg->vpd.vpd_ros[off].value = malloc(
1223 				    (dflen + 1) *
1224 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1225 				    M_DEVBUF, M_WAITOK);
1226 			remain -= 3;
1227 			i = 0;
1228 			/* keep in sync w/ state 3's transistions */
1229 			if (dflen == 0 && remain == 0)
1230 				state = 0;
1231 			else if (dflen == 0)
1232 				state = 2;
1233 			else
1234 				state = 3;
1235 			break;
1236 
1237 		case 3:	/* VPD-R Keyword Value */
1238 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1239 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1240 			    "RV", 2) == 0 && cksumvalid == -1) {
1241 				if (vrs.cksum == 0)
1242 					cksumvalid = 1;
1243 				else {
1244 					if (bootverbose)
1245 						pci_printf(cfg,
1246 					    "bad VPD cksum, remain %hhu\n",
1247 						    vrs.cksum);
1248 					cksumvalid = 0;
1249 					state = -1;
1250 					break;
1251 				}
1252 			}
1253 			dflen--;
1254 			remain--;
1255 			/* keep in sync w/ state 2's transistions */
1256 			if (dflen == 0)
1257 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1258 			if (dflen == 0 && remain == 0) {
1259 				cfg->vpd.vpd_rocnt = off;
1260 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1261 				    off * sizeof(*cfg->vpd.vpd_ros),
1262 				    M_DEVBUF, M_WAITOK | M_ZERO);
1263 				state = 0;
1264 			} else if (dflen == 0)
1265 				state = 2;
1266 			break;
1267 
1268 		case 4:
1269 			remain--;
1270 			if (remain == 0)
1271 				state = 0;
1272 			break;
1273 
1274 		case 5:	/* VPD-W Keyword Header */
1275 			if (off == alloc) {
1276 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1277 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1278 				    M_DEVBUF, M_WAITOK | M_ZERO);
1279 			}
1280 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1281 			if (vpd_nextbyte(&vrs, &byte2)) {
1282 				state = -2;
1283 				break;
1284 			}
1285 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1286 			if (vpd_nextbyte(&vrs, &byte2)) {
1287 				state = -2;
1288 				break;
1289 			}
1290 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1291 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1292 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1293 			    sizeof(*cfg->vpd.vpd_w[off].value),
1294 			    M_DEVBUF, M_WAITOK);
1295 			remain -= 3;
1296 			i = 0;
1297 			/* keep in sync w/ state 6's transistions */
1298 			if (dflen == 0 && remain == 0)
1299 				state = 0;
1300 			else if (dflen == 0)
1301 				state = 5;
1302 			else
1303 				state = 6;
1304 			break;
1305 
1306 		case 6:	/* VPD-W Keyword Value */
1307 			cfg->vpd.vpd_w[off].value[i++] = byte;
1308 			dflen--;
1309 			remain--;
1310 			/* keep in sync w/ state 5's transistions */
1311 			if (dflen == 0)
1312 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1313 			if (dflen == 0 && remain == 0) {
1314 				cfg->vpd.vpd_wcnt = off;
1315 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1316 				    off * sizeof(*cfg->vpd.vpd_w),
1317 				    M_DEVBUF, M_WAITOK | M_ZERO);
1318 				state = 0;
1319 			} else if (dflen == 0)
1320 				state = 5;
1321 			break;
1322 
1323 		default:
1324 			pci_printf(cfg, "invalid state: %d\n", state);
1325 			state = -1;
1326 			break;
1327 		}
1328 	}
1329 
1330 	if (cksumvalid == 0 || state < -1) {
1331 		/* read-only data bad, clean up */
1332 		if (cfg->vpd.vpd_ros != NULL) {
1333 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1334 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1335 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1336 			cfg->vpd.vpd_ros = NULL;
1337 		}
1338 	}
1339 	if (state < -1) {
1340 		/* I/O error, clean up */
1341 		pci_printf(cfg, "failed to read VPD data.\n");
1342 		if (cfg->vpd.vpd_ident != NULL) {
1343 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1344 			cfg->vpd.vpd_ident = NULL;
1345 		}
1346 		if (cfg->vpd.vpd_w != NULL) {
1347 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1348 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1349 			free(cfg->vpd.vpd_w, M_DEVBUF);
1350 			cfg->vpd.vpd_w = NULL;
1351 		}
1352 	}
1353 	cfg->vpd.vpd_cached = 1;
1354 #undef REG
1355 #undef WREG
1356 }
1357 
1358 int
1359 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1360 {
1361 	struct pci_devinfo *dinfo = device_get_ivars(child);
1362 	pcicfgregs *cfg = &dinfo->cfg;
1363 
1364 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1365 		pci_read_vpd(device_get_parent(dev), cfg);
1366 
1367 	*identptr = cfg->vpd.vpd_ident;
1368 
1369 	if (*identptr == NULL)
1370 		return (ENXIO);
1371 
1372 	return (0);
1373 }
1374 
1375 int
1376 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1377 	const char **vptr)
1378 {
1379 	struct pci_devinfo *dinfo = device_get_ivars(child);
1380 	pcicfgregs *cfg = &dinfo->cfg;
1381 	int i;
1382 
1383 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1384 		pci_read_vpd(device_get_parent(dev), cfg);
1385 
1386 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1387 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1388 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1389 			*vptr = cfg->vpd.vpd_ros[i].value;
1390 			return (0);
1391 		}
1392 
1393 	*vptr = NULL;
1394 	return (ENXIO);
1395 }
1396 
1397 struct pcicfg_vpd *
1398 pci_fetch_vpd_list(device_t dev)
1399 {
1400 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1401 	pcicfgregs *cfg = &dinfo->cfg;
1402 
1403 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1404 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1405 	return (&cfg->vpd);
1406 }
1407 
1408 /*
1409  * Find the requested HyperTransport capability and return the offset
1410  * in configuration space via the pointer provided.  The function
1411  * returns 0 on success and an error code otherwise.
1412  */
1413 int
1414 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1415 {
1416 	int ptr, error;
1417 	uint16_t val;
1418 
1419 	error = pci_find_cap(child, PCIY_HT, &ptr);
1420 	if (error)
1421 		return (error);
1422 
1423 	/*
1424 	 * Traverse the capabilities list checking each HT capability
1425 	 * to see if it matches the requested HT capability.
1426 	 */
1427 	for (;;) {
1428 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1429 		if (capability == PCIM_HTCAP_SLAVE ||
1430 		    capability == PCIM_HTCAP_HOST)
1431 			val &= 0xe000;
1432 		else
1433 			val &= PCIM_HTCMD_CAP_MASK;
1434 		if (val == capability) {
1435 			if (capreg != NULL)
1436 				*capreg = ptr;
1437 			return (0);
1438 		}
1439 
1440 		/* Skip to the next HT capability. */
1441 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1442 			break;
1443 	}
1444 
1445 	return (ENOENT);
1446 }
1447 
1448 /*
1449  * Find the next requested HyperTransport capability after start and return
1450  * the offset in configuration space via the pointer provided.  The function
1451  * returns 0 on success and an error code otherwise.
1452  */
1453 int
1454 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1455     int start, int *capreg)
1456 {
1457 	int ptr;
1458 	uint16_t val;
1459 
1460 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1461 	    ("start capability is not HyperTransport capability"));
1462 	ptr = start;
1463 
1464 	/*
1465 	 * Traverse the capabilities list checking each HT capability
1466 	 * to see if it matches the requested HT capability.
1467 	 */
1468 	for (;;) {
1469 		/* Skip to the next HT capability. */
1470 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1471 			break;
1472 
1473 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1474 		if (capability == PCIM_HTCAP_SLAVE ||
1475 		    capability == PCIM_HTCAP_HOST)
1476 			val &= 0xe000;
1477 		else
1478 			val &= PCIM_HTCMD_CAP_MASK;
1479 		if (val == capability) {
1480 			if (capreg != NULL)
1481 				*capreg = ptr;
1482 			return (0);
1483 		}
1484 	}
1485 
1486 	return (ENOENT);
1487 }
1488 
1489 /*
1490  * Find the requested capability and return the offset in
1491  * configuration space via the pointer provided.  The function returns
1492  * 0 on success and an error code otherwise.
1493  */
1494 int
1495 pci_find_cap_method(device_t dev, device_t child, int capability,
1496     int *capreg)
1497 {
1498 	struct pci_devinfo *dinfo = device_get_ivars(child);
1499 	pcicfgregs *cfg = &dinfo->cfg;
1500 	uint32_t status;
1501 	uint8_t ptr;
1502 
1503 	/*
1504 	 * Check the CAP_LIST bit of the PCI status register first.
1505 	 */
1506 	status = pci_read_config(child, PCIR_STATUS, 2);
1507 	if (!(status & PCIM_STATUS_CAPPRESENT))
1508 		return (ENXIO);
1509 
1510 	/*
1511 	 * Determine the start pointer of the capabilities list.
1512 	 */
1513 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1514 	case PCIM_HDRTYPE_NORMAL:
1515 	case PCIM_HDRTYPE_BRIDGE:
1516 		ptr = PCIR_CAP_PTR;
1517 		break;
1518 	case PCIM_HDRTYPE_CARDBUS:
1519 		ptr = PCIR_CAP_PTR_2;
1520 		break;
1521 	default:
1522 		/* XXX: panic? */
1523 		return (ENXIO);		/* no extended capabilities support */
1524 	}
1525 	ptr = pci_read_config(child, ptr, 1);
1526 
1527 	/*
1528 	 * Traverse the capabilities list.
1529 	 */
1530 	while (ptr != 0) {
1531 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1532 			if (capreg != NULL)
1533 				*capreg = ptr;
1534 			return (0);
1535 		}
1536 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1537 	}
1538 
1539 	return (ENOENT);
1540 }
1541 
1542 /*
1543  * Find the next requested capability after start and return the offset in
1544  * configuration space via the pointer provided.  The function returns
1545  * 0 on success and an error code otherwise.
1546  */
1547 int
1548 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1549     int start, int *capreg)
1550 {
1551 	uint8_t ptr;
1552 
1553 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1554 	    ("start capability is not expected capability"));
1555 
1556 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1557 	while (ptr != 0) {
1558 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1559 			if (capreg != NULL)
1560 				*capreg = ptr;
1561 			return (0);
1562 		}
1563 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1564 	}
1565 
1566 	return (ENOENT);
1567 }
1568 
1569 /*
1570  * Find the requested extended capability and return the offset in
1571  * configuration space via the pointer provided.  The function returns
1572  * 0 on success and an error code otherwise.
1573  */
1574 int
1575 pci_find_extcap_method(device_t dev, device_t child, int capability,
1576     int *capreg)
1577 {
1578 	struct pci_devinfo *dinfo = device_get_ivars(child);
1579 	pcicfgregs *cfg = &dinfo->cfg;
1580 	uint32_t ecap;
1581 	uint16_t ptr;
1582 
1583 	/* Only supported for PCI-express devices. */
1584 	if (cfg->pcie.pcie_location == 0)
1585 		return (ENXIO);
1586 
1587 	ptr = PCIR_EXTCAP;
1588 	ecap = pci_read_config(child, ptr, 4);
1589 	if (ecap == 0xffffffff || ecap == 0)
1590 		return (ENOENT);
1591 	for (;;) {
1592 		if (PCI_EXTCAP_ID(ecap) == capability) {
1593 			if (capreg != NULL)
1594 				*capreg = ptr;
1595 			return (0);
1596 		}
1597 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1598 		if (ptr == 0)
1599 			break;
1600 		ecap = pci_read_config(child, ptr, 4);
1601 	}
1602 
1603 	return (ENOENT);
1604 }
1605 
1606 /*
1607  * Find the next requested extended capability after start and return the
1608  * offset in configuration space via the pointer provided.  The function
1609  * returns 0 on success and an error code otherwise.
1610  */
1611 int
1612 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1613     int start, int *capreg)
1614 {
1615 	struct pci_devinfo *dinfo = device_get_ivars(child);
1616 	pcicfgregs *cfg = &dinfo->cfg;
1617 	uint32_t ecap;
1618 	uint16_t ptr;
1619 
1620 	/* Only supported for PCI-express devices. */
1621 	if (cfg->pcie.pcie_location == 0)
1622 		return (ENXIO);
1623 
1624 	ecap = pci_read_config(child, start, 4);
1625 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1626 	    ("start extended capability is not expected capability"));
1627 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1628 	while (ptr != 0) {
1629 		ecap = pci_read_config(child, ptr, 4);
1630 		if (PCI_EXTCAP_ID(ecap) == capability) {
1631 			if (capreg != NULL)
1632 				*capreg = ptr;
1633 			return (0);
1634 		}
1635 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1636 	}
1637 
1638 	return (ENOENT);
1639 }
1640 
1641 /*
1642  * Support for MSI-X message interrupts.
1643  */
1644 static void
1645 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1646 {
1647 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1648 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1649 	uint32_t offset;
1650 
1651 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1652 	offset = msix->msix_table_offset + index * 16;
1653 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1654 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1655 	bus_write_4(msix->msix_table_res, offset + 8, data);
1656 }
1657 
1658 void
1659 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1660     uint64_t address, uint32_t data)
1661 {
1662 
1663 	if (pci_msix_rewrite_table) {
1664 		struct pci_devinfo *dinfo = device_get_ivars(child);
1665 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1666 
1667 		/*
1668 		 * Some VM hosts require MSIX to be disabled in the
1669 		 * control register before updating the MSIX table
1670 		 * entries are allowed. It is not enough to only
1671 		 * disable MSIX while updating a single entry. MSIX
1672 		 * must be disabled while updating all entries in the
1673 		 * table.
1674 		 */
1675 		pci_write_config(child,
1676 		    msix->msix_location + PCIR_MSIX_CTRL,
1677 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1678 		pci_resume_msix(child);
1679 	} else
1680 		pci_write_msix_entry(child, index, address, data);
1681 
1682 	/* Enable MSI -> HT mapping. */
1683 	pci_ht_map_msi(child, address);
1684 }
1685 
1686 void
1687 pci_mask_msix(device_t dev, u_int index)
1688 {
1689 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1690 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1691 	uint32_t offset, val;
1692 
1693 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1694 	offset = msix->msix_table_offset + index * 16 + 12;
1695 	val = bus_read_4(msix->msix_table_res, offset);
1696 	val |= PCIM_MSIX_VCTRL_MASK;
1697 
1698 	/*
1699 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1700 	 * register, so always write the new value.
1701 	 */
1702 	bus_write_4(msix->msix_table_res, offset, val);
1703 }
1704 
1705 void
1706 pci_unmask_msix(device_t dev, u_int index)
1707 {
1708 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1709 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1710 	uint32_t offset, val;
1711 
1712 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1713 	offset = msix->msix_table_offset + index * 16 + 12;
1714 	val = bus_read_4(msix->msix_table_res, offset);
1715 	val &= ~PCIM_MSIX_VCTRL_MASK;
1716 
1717 	/*
1718 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1719 	 * register, so always write the new value.
1720 	 */
1721 	bus_write_4(msix->msix_table_res, offset, val);
1722 }
1723 
1724 int
1725 pci_pending_msix(device_t dev, u_int index)
1726 {
1727 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1728 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1729 	uint32_t offset, bit;
1730 
1731 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1732 	offset = msix->msix_pba_offset + (index / 32) * 4;
1733 	bit = 1 << index % 32;
1734 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1735 }
1736 
1737 /*
1738  * Restore MSI-X registers and table during resume.  If MSI-X is
1739  * enabled then walk the virtual table to restore the actual MSI-X
1740  * table.
1741  */
1742 static void
1743 pci_resume_msix(device_t dev)
1744 {
1745 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1746 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1747 	struct msix_table_entry *mte;
1748 	struct msix_vector *mv;
1749 	int i;
1750 
1751 	if (msix->msix_alloc > 0) {
1752 		/* First, mask all vectors. */
1753 		for (i = 0; i < msix->msix_msgnum; i++)
1754 			pci_mask_msix(dev, i);
1755 
1756 		/* Second, program any messages with at least one handler. */
1757 		for (i = 0; i < msix->msix_table_len; i++) {
1758 			mte = &msix->msix_table[i];
1759 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1760 				continue;
1761 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1762 			pci_write_msix_entry(dev, i, mv->mv_address,
1763 			    mv->mv_data);
1764 			pci_unmask_msix(dev, i);
1765 		}
1766 	}
1767 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1768 	    msix->msix_ctrl, 2);
1769 }
1770 
1771 /*
1772  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1773  * returned in *count.  After this function returns, each message will be
1774  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1775  */
1776 int
1777 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1778 {
1779 	struct pci_devinfo *dinfo = device_get_ivars(child);
1780 	pcicfgregs *cfg = &dinfo->cfg;
1781 	struct resource_list_entry *rle;
1782 	int actual, error, i, irq, max;
1783 
1784 	/* Don't let count == 0 get us into trouble. */
1785 	if (*count == 0)
1786 		return (EINVAL);
1787 
1788 	/* If rid 0 is allocated, then fail. */
1789 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1790 	if (rle != NULL && rle->res != NULL)
1791 		return (ENXIO);
1792 
1793 	/* Already have allocated messages? */
1794 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1795 		return (ENXIO);
1796 
1797 	/* If MSI-X is blacklisted for this system, fail. */
1798 	if (pci_msix_blacklisted())
1799 		return (ENXIO);
1800 
1801 	/* MSI-X capability present? */
1802 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1803 		return (ENODEV);
1804 
1805 	/* Make sure the appropriate BARs are mapped. */
1806 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1807 	    cfg->msix.msix_table_bar);
1808 	if (rle == NULL || rle->res == NULL ||
1809 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1810 		return (ENXIO);
1811 	cfg->msix.msix_table_res = rle->res;
1812 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1813 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1814 		    cfg->msix.msix_pba_bar);
1815 		if (rle == NULL || rle->res == NULL ||
1816 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1817 			return (ENXIO);
1818 	}
1819 	cfg->msix.msix_pba_res = rle->res;
1820 
1821 	if (bootverbose)
1822 		device_printf(child,
1823 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1824 		    *count, cfg->msix.msix_msgnum);
1825 	max = min(*count, cfg->msix.msix_msgnum);
1826 	for (i = 0; i < max; i++) {
1827 		/* Allocate a message. */
1828 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1829 		if (error) {
1830 			if (i == 0)
1831 				return (error);
1832 			break;
1833 		}
1834 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1835 		    irq, 1);
1836 	}
1837 	actual = i;
1838 
1839 	if (bootverbose) {
1840 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1841 		if (actual == 1)
1842 			device_printf(child, "using IRQ %ju for MSI-X\n",
1843 			    rle->start);
1844 		else {
1845 			int run;
1846 
1847 			/*
1848 			 * Be fancy and try to print contiguous runs of
1849 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1850 			 * 'run' is true if we are in a range.
1851 			 */
1852 			device_printf(child, "using IRQs %ju", rle->start);
1853 			irq = rle->start;
1854 			run = 0;
1855 			for (i = 1; i < actual; i++) {
1856 				rle = resource_list_find(&dinfo->resources,
1857 				    SYS_RES_IRQ, i + 1);
1858 
1859 				/* Still in a run? */
1860 				if (rle->start == irq + 1) {
1861 					run = 1;
1862 					irq++;
1863 					continue;
1864 				}
1865 
1866 				/* Finish previous range. */
1867 				if (run) {
1868 					printf("-%d", irq);
1869 					run = 0;
1870 				}
1871 
1872 				/* Start new range. */
1873 				printf(",%ju", rle->start);
1874 				irq = rle->start;
1875 			}
1876 
1877 			/* Unfinished range? */
1878 			if (run)
1879 				printf("-%d", irq);
1880 			printf(" for MSI-X\n");
1881 		}
1882 	}
1883 
1884 	/* Mask all vectors. */
1885 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1886 		pci_mask_msix(child, i);
1887 
1888 	/* Allocate and initialize vector data and virtual table. */
1889 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1890 	    M_DEVBUF, M_WAITOK | M_ZERO);
1891 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1892 	    M_DEVBUF, M_WAITOK | M_ZERO);
1893 	for (i = 0; i < actual; i++) {
1894 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1895 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1896 		cfg->msix.msix_table[i].mte_vector = i + 1;
1897 	}
1898 
1899 	/* Update control register to enable MSI-X. */
1900 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1901 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1902 	    cfg->msix.msix_ctrl, 2);
1903 
1904 	/* Update counts of alloc'd messages. */
1905 	cfg->msix.msix_alloc = actual;
1906 	cfg->msix.msix_table_len = actual;
1907 	*count = actual;
1908 	return (0);
1909 }
1910 
1911 /*
1912  * By default, pci_alloc_msix() will assign the allocated IRQ
1913  * resources consecutively to the first N messages in the MSI-X table.
1914  * However, device drivers may want to use different layouts if they
1915  * either receive fewer messages than they asked for, or they wish to
1916  * populate the MSI-X table sparsely.  This method allows the driver
1917  * to specify what layout it wants.  It must be called after a
1918  * successful pci_alloc_msix() but before any of the associated
1919  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1920  *
1921  * The 'vectors' array contains 'count' message vectors.  The array
1922  * maps directly to the MSI-X table in that index 0 in the array
1923  * specifies the vector for the first message in the MSI-X table, etc.
1924  * The vector value in each array index can either be 0 to indicate
1925  * that no vector should be assigned to a message slot, or it can be a
1926  * number from 1 to N (where N is the count returned from a
1927  * succcessful call to pci_alloc_msix()) to indicate which message
1928  * vector (IRQ) to be used for the corresponding message.
1929  *
1930  * On successful return, each message with a non-zero vector will have
1931  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1932  * 1.  Additionally, if any of the IRQs allocated via the previous
1933  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1934  * will be freed back to the system automatically.
1935  *
1936  * For example, suppose a driver has a MSI-X table with 6 messages and
1937  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1938  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1939  * C.  After the call to pci_alloc_msix(), the device will be setup to
1940  * have an MSI-X table of ABC--- (where - means no vector assigned).
1941  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1942  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1943  * be freed back to the system.  This device will also have valid
1944  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1945  *
1946  * In any case, the SYS_RES_IRQ rid X will always map to the message
1947  * at MSI-X table index X - 1 and will only be valid if a vector is
1948  * assigned to that table entry.
1949  */
1950 int
1951 pci_remap_msix_method(device_t dev, device_t child, int count,
1952     const u_int *vectors)
1953 {
1954 	struct pci_devinfo *dinfo = device_get_ivars(child);
1955 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1956 	struct resource_list_entry *rle;
1957 	int i, irq, j, *used;
1958 
1959 	/*
1960 	 * Have to have at least one message in the table but the
1961 	 * table can't be bigger than the actual MSI-X table in the
1962 	 * device.
1963 	 */
1964 	if (count == 0 || count > msix->msix_msgnum)
1965 		return (EINVAL);
1966 
1967 	/* Sanity check the vectors. */
1968 	for (i = 0; i < count; i++)
1969 		if (vectors[i] > msix->msix_alloc)
1970 			return (EINVAL);
1971 
1972 	/*
1973 	 * Make sure there aren't any holes in the vectors to be used.
1974 	 * It's a big pain to support it, and it doesn't really make
1975 	 * sense anyway.  Also, at least one vector must be used.
1976 	 */
1977 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1978 	    M_ZERO);
1979 	for (i = 0; i < count; i++)
1980 		if (vectors[i] != 0)
1981 			used[vectors[i] - 1] = 1;
1982 	for (i = 0; i < msix->msix_alloc - 1; i++)
1983 		if (used[i] == 0 && used[i + 1] == 1) {
1984 			free(used, M_DEVBUF);
1985 			return (EINVAL);
1986 		}
1987 	if (used[0] != 1) {
1988 		free(used, M_DEVBUF);
1989 		return (EINVAL);
1990 	}
1991 
1992 	/* Make sure none of the resources are allocated. */
1993 	for (i = 0; i < msix->msix_table_len; i++) {
1994 		if (msix->msix_table[i].mte_vector == 0)
1995 			continue;
1996 		if (msix->msix_table[i].mte_handlers > 0) {
1997 			free(used, M_DEVBUF);
1998 			return (EBUSY);
1999 		}
2000 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2001 		KASSERT(rle != NULL, ("missing resource"));
2002 		if (rle->res != NULL) {
2003 			free(used, M_DEVBUF);
2004 			return (EBUSY);
2005 		}
2006 	}
2007 
2008 	/* Free the existing resource list entries. */
2009 	for (i = 0; i < msix->msix_table_len; i++) {
2010 		if (msix->msix_table[i].mte_vector == 0)
2011 			continue;
2012 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2013 	}
2014 
2015 	/*
2016 	 * Build the new virtual table keeping track of which vectors are
2017 	 * used.
2018 	 */
2019 	free(msix->msix_table, M_DEVBUF);
2020 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
2021 	    M_DEVBUF, M_WAITOK | M_ZERO);
2022 	for (i = 0; i < count; i++)
2023 		msix->msix_table[i].mte_vector = vectors[i];
2024 	msix->msix_table_len = count;
2025 
2026 	/* Free any unused IRQs and resize the vectors array if necessary. */
2027 	j = msix->msix_alloc - 1;
2028 	if (used[j] == 0) {
2029 		struct msix_vector *vec;
2030 
2031 		while (used[j] == 0) {
2032 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2033 			    msix->msix_vectors[j].mv_irq);
2034 			j--;
2035 		}
2036 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2037 		    M_WAITOK);
2038 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2039 		    (j + 1));
2040 		free(msix->msix_vectors, M_DEVBUF);
2041 		msix->msix_vectors = vec;
2042 		msix->msix_alloc = j + 1;
2043 	}
2044 	free(used, M_DEVBUF);
2045 
2046 	/* Map the IRQs onto the rids. */
2047 	for (i = 0; i < count; i++) {
2048 		if (vectors[i] == 0)
2049 			continue;
2050 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2051 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2052 		    irq, 1);
2053 	}
2054 
2055 	if (bootverbose) {
2056 		device_printf(child, "Remapped MSI-X IRQs as: ");
2057 		for (i = 0; i < count; i++) {
2058 			if (i != 0)
2059 				printf(", ");
2060 			if (vectors[i] == 0)
2061 				printf("---");
2062 			else
2063 				printf("%d",
2064 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2065 		}
2066 		printf("\n");
2067 	}
2068 
2069 	return (0);
2070 }
2071 
2072 static int
2073 pci_release_msix(device_t dev, device_t child)
2074 {
2075 	struct pci_devinfo *dinfo = device_get_ivars(child);
2076 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2077 	struct resource_list_entry *rle;
2078 	int i;
2079 
2080 	/* Do we have any messages to release? */
2081 	if (msix->msix_alloc == 0)
2082 		return (ENODEV);
2083 
2084 	/* Make sure none of the resources are allocated. */
2085 	for (i = 0; i < msix->msix_table_len; i++) {
2086 		if (msix->msix_table[i].mte_vector == 0)
2087 			continue;
2088 		if (msix->msix_table[i].mte_handlers > 0)
2089 			return (EBUSY);
2090 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2091 		KASSERT(rle != NULL, ("missing resource"));
2092 		if (rle->res != NULL)
2093 			return (EBUSY);
2094 	}
2095 
2096 	/* Update control register to disable MSI-X. */
2097 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2098 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2099 	    msix->msix_ctrl, 2);
2100 
2101 	/* Free the resource list entries. */
2102 	for (i = 0; i < msix->msix_table_len; i++) {
2103 		if (msix->msix_table[i].mte_vector == 0)
2104 			continue;
2105 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2106 	}
2107 	free(msix->msix_table, M_DEVBUF);
2108 	msix->msix_table_len = 0;
2109 
2110 	/* Release the IRQs. */
2111 	for (i = 0; i < msix->msix_alloc; i++)
2112 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2113 		    msix->msix_vectors[i].mv_irq);
2114 	free(msix->msix_vectors, M_DEVBUF);
2115 	msix->msix_alloc = 0;
2116 	return (0);
2117 }
2118 
2119 /*
2120  * Return the max supported MSI-X messages this device supports.
2121  * Basically, assuming the MD code can alloc messages, this function
2122  * should return the maximum value that pci_alloc_msix() can return.
2123  * Thus, it is subject to the tunables, etc.
2124  */
2125 int
2126 pci_msix_count_method(device_t dev, device_t child)
2127 {
2128 	struct pci_devinfo *dinfo = device_get_ivars(child);
2129 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2130 
2131 	if (pci_do_msix && msix->msix_location != 0)
2132 		return (msix->msix_msgnum);
2133 	return (0);
2134 }
2135 
2136 int
2137 pci_msix_pba_bar_method(device_t dev, device_t child)
2138 {
2139 	struct pci_devinfo *dinfo = device_get_ivars(child);
2140 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2141 
2142 	if (pci_do_msix && msix->msix_location != 0)
2143 		return (msix->msix_pba_bar);
2144 	return (-1);
2145 }
2146 
2147 int
2148 pci_msix_table_bar_method(device_t dev, device_t child)
2149 {
2150 	struct pci_devinfo *dinfo = device_get_ivars(child);
2151 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2152 
2153 	if (pci_do_msix && msix->msix_location != 0)
2154 		return (msix->msix_table_bar);
2155 	return (-1);
2156 }
2157 
2158 /*
2159  * HyperTransport MSI mapping control
2160  */
2161 void
2162 pci_ht_map_msi(device_t dev, uint64_t addr)
2163 {
2164 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2165 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2166 
2167 	if (!ht->ht_msimap)
2168 		return;
2169 
2170 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2171 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2172 		/* Enable MSI -> HT mapping. */
2173 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2174 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2175 		    ht->ht_msictrl, 2);
2176 	}
2177 
2178 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2179 		/* Disable MSI -> HT mapping. */
2180 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2181 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2182 		    ht->ht_msictrl, 2);
2183 	}
2184 }
2185 
2186 int
2187 pci_get_relaxed_ordering_enabled(device_t dev)
2188 {
2189 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2190 	int cap;
2191 	uint16_t val;
2192 
2193 	cap = dinfo->cfg.pcie.pcie_location;
2194 	if (cap == 0)
2195 		return (0);
2196 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2197 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2198 	return (val != 0);
2199 }
2200 
2201 int
2202 pci_get_max_payload(device_t dev)
2203 {
2204 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2205 	int cap;
2206 	uint16_t val;
2207 
2208 	cap = dinfo->cfg.pcie.pcie_location;
2209 	if (cap == 0)
2210 		return (0);
2211 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2212 	val &= PCIEM_CTL_MAX_PAYLOAD;
2213 	val >>= 5;
2214 	return (1 << (val + 7));
2215 }
2216 
2217 int
2218 pci_get_max_read_req(device_t dev)
2219 {
2220 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2221 	int cap;
2222 	uint16_t val;
2223 
2224 	cap = dinfo->cfg.pcie.pcie_location;
2225 	if (cap == 0)
2226 		return (0);
2227 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2228 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2229 	val >>= 12;
2230 	return (1 << (val + 7));
2231 }
2232 
2233 int
2234 pci_set_max_read_req(device_t dev, int size)
2235 {
2236 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2237 	int cap;
2238 	uint16_t val;
2239 
2240 	cap = dinfo->cfg.pcie.pcie_location;
2241 	if (cap == 0)
2242 		return (0);
2243 	if (size < 128)
2244 		size = 128;
2245 	if (size > 4096)
2246 		size = 4096;
2247 	size = (1 << (fls(size) - 1));
2248 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2249 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2250 	val |= (fls(size) - 8) << 12;
2251 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2252 	return (size);
2253 }
2254 
2255 uint32_t
2256 pcie_read_config(device_t dev, int reg, int width)
2257 {
2258 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2259 	int cap;
2260 
2261 	cap = dinfo->cfg.pcie.pcie_location;
2262 	if (cap == 0) {
2263 		if (width == 2)
2264 			return (0xffff);
2265 		return (0xffffffff);
2266 	}
2267 
2268 	return (pci_read_config(dev, cap + reg, width));
2269 }
2270 
2271 void
2272 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2273 {
2274 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2275 	int cap;
2276 
2277 	cap = dinfo->cfg.pcie.pcie_location;
2278 	if (cap == 0)
2279 		return;
2280 	pci_write_config(dev, cap + reg, value, width);
2281 }
2282 
2283 /*
2284  * Adjusts a PCI-e capability register by clearing the bits in mask
2285  * and setting the bits in (value & mask).  Bits not set in mask are
2286  * not adjusted.
2287  *
2288  * Returns the old value on success or all ones on failure.
2289  */
2290 uint32_t
2291 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2292     int width)
2293 {
2294 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2295 	uint32_t old, new;
2296 	int cap;
2297 
2298 	cap = dinfo->cfg.pcie.pcie_location;
2299 	if (cap == 0) {
2300 		if (width == 2)
2301 			return (0xffff);
2302 		return (0xffffffff);
2303 	}
2304 
2305 	old = pci_read_config(dev, cap + reg, width);
2306 	new = old & ~mask;
2307 	new |= (value & mask);
2308 	pci_write_config(dev, cap + reg, new, width);
2309 	return (old);
2310 }
2311 
2312 /*
2313  * Support for MSI message signalled interrupts.
2314  */
2315 void
2316 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2317     uint16_t data)
2318 {
2319 	struct pci_devinfo *dinfo = device_get_ivars(child);
2320 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2321 
2322 	/* Write data and address values. */
2323 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2324 	    address & 0xffffffff, 4);
2325 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2326 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2327 		    address >> 32, 4);
2328 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2329 		    data, 2);
2330 	} else
2331 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2332 		    2);
2333 
2334 	/* Enable MSI in the control register. */
2335 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2336 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2337 	    msi->msi_ctrl, 2);
2338 
2339 	/* Enable MSI -> HT mapping. */
2340 	pci_ht_map_msi(child, address);
2341 }
2342 
2343 void
2344 pci_disable_msi_method(device_t dev, device_t child)
2345 {
2346 	struct pci_devinfo *dinfo = device_get_ivars(child);
2347 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2348 
2349 	/* Disable MSI -> HT mapping. */
2350 	pci_ht_map_msi(child, 0);
2351 
2352 	/* Disable MSI in the control register. */
2353 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2354 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2355 	    msi->msi_ctrl, 2);
2356 }
2357 
2358 /*
2359  * Restore MSI registers during resume.  If MSI is enabled then
2360  * restore the data and address registers in addition to the control
2361  * register.
2362  */
2363 static void
2364 pci_resume_msi(device_t dev)
2365 {
2366 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2367 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2368 	uint64_t address;
2369 	uint16_t data;
2370 
2371 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2372 		address = msi->msi_addr;
2373 		data = msi->msi_data;
2374 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2375 		    address & 0xffffffff, 4);
2376 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2377 			pci_write_config(dev, msi->msi_location +
2378 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2379 			pci_write_config(dev, msi->msi_location +
2380 			    PCIR_MSI_DATA_64BIT, data, 2);
2381 		} else
2382 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2383 			    data, 2);
2384 	}
2385 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2386 	    2);
2387 }
2388 
2389 static int
2390 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2391 {
2392 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2393 	pcicfgregs *cfg = &dinfo->cfg;
2394 	struct resource_list_entry *rle;
2395 	struct msix_table_entry *mte;
2396 	struct msix_vector *mv;
2397 	uint64_t addr;
2398 	uint32_t data;
2399 	int error, i, j;
2400 
2401 	/*
2402 	 * Handle MSI first.  We try to find this IRQ among our list
2403 	 * of MSI IRQs.  If we find it, we request updated address and
2404 	 * data registers and apply the results.
2405 	 */
2406 	if (cfg->msi.msi_alloc > 0) {
2407 		/* If we don't have any active handlers, nothing to do. */
2408 		if (cfg->msi.msi_handlers == 0)
2409 			return (0);
2410 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2411 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2412 			    i + 1);
2413 			if (rle->start == irq) {
2414 				error = PCIB_MAP_MSI(device_get_parent(bus),
2415 				    dev, irq, &addr, &data);
2416 				if (error)
2417 					return (error);
2418 				pci_disable_msi(dev);
2419 				dinfo->cfg.msi.msi_addr = addr;
2420 				dinfo->cfg.msi.msi_data = data;
2421 				pci_enable_msi(dev, addr, data);
2422 				return (0);
2423 			}
2424 		}
2425 		return (ENOENT);
2426 	}
2427 
2428 	/*
2429 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2430 	 * we request the updated mapping info.  If that works, we go
2431 	 * through all the slots that use this IRQ and update them.
2432 	 */
2433 	if (cfg->msix.msix_alloc > 0) {
2434 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2435 			mv = &cfg->msix.msix_vectors[i];
2436 			if (mv->mv_irq == irq) {
2437 				error = PCIB_MAP_MSI(device_get_parent(bus),
2438 				    dev, irq, &addr, &data);
2439 				if (error)
2440 					return (error);
2441 				mv->mv_address = addr;
2442 				mv->mv_data = data;
2443 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2444 					mte = &cfg->msix.msix_table[j];
2445 					if (mte->mte_vector != i + 1)
2446 						continue;
2447 					if (mte->mte_handlers == 0)
2448 						continue;
2449 					pci_mask_msix(dev, j);
2450 					pci_enable_msix(dev, j, addr, data);
2451 					pci_unmask_msix(dev, j);
2452 				}
2453 			}
2454 		}
2455 		return (ENOENT);
2456 	}
2457 
2458 	return (ENOENT);
2459 }
2460 
2461 /*
2462  * Returns true if the specified device is blacklisted because MSI
2463  * doesn't work.
2464  */
2465 int
2466 pci_msi_device_blacklisted(device_t dev)
2467 {
2468 
2469 	if (!pci_honor_msi_blacklist)
2470 		return (0);
2471 
2472 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2473 }
2474 
2475 /*
2476  * Determine if MSI is blacklisted globally on this system.  Currently,
2477  * we just check for blacklisted chipsets as represented by the
2478  * host-PCI bridge at device 0:0:0.  In the future, it may become
2479  * necessary to check other system attributes, such as the kenv values
2480  * that give the motherboard manufacturer and model number.
2481  */
2482 static int
2483 pci_msi_blacklisted(void)
2484 {
2485 	device_t dev;
2486 
2487 	if (!pci_honor_msi_blacklist)
2488 		return (0);
2489 
2490 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2491 	if (!(pcie_chipset || pcix_chipset)) {
2492 		if (vm_guest != VM_GUEST_NO) {
2493 			/*
2494 			 * Whitelist older chipsets in virtual
2495 			 * machines known to support MSI.
2496 			 */
2497 			dev = pci_find_bsf(0, 0, 0);
2498 			if (dev != NULL)
2499 				return (!pci_has_quirk(pci_get_devid(dev),
2500 					PCI_QUIRK_ENABLE_MSI_VM));
2501 		}
2502 		return (1);
2503 	}
2504 
2505 	dev = pci_find_bsf(0, 0, 0);
2506 	if (dev != NULL)
2507 		return (pci_msi_device_blacklisted(dev));
2508 	return (0);
2509 }
2510 
2511 /*
2512  * Returns true if the specified device is blacklisted because MSI-X
2513  * doesn't work.  Note that this assumes that if MSI doesn't work,
2514  * MSI-X doesn't either.
2515  */
2516 int
2517 pci_msix_device_blacklisted(device_t dev)
2518 {
2519 
2520 	if (!pci_honor_msi_blacklist)
2521 		return (0);
2522 
2523 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2524 		return (1);
2525 
2526 	return (pci_msi_device_blacklisted(dev));
2527 }
2528 
2529 /*
2530  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2531  * is blacklisted, assume that MSI-X is as well.  Check for additional
2532  * chipsets where MSI works but MSI-X does not.
2533  */
2534 static int
2535 pci_msix_blacklisted(void)
2536 {
2537 	device_t dev;
2538 
2539 	if (!pci_honor_msi_blacklist)
2540 		return (0);
2541 
2542 	dev = pci_find_bsf(0, 0, 0);
2543 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2544 	    PCI_QUIRK_DISABLE_MSIX))
2545 		return (1);
2546 
2547 	return (pci_msi_blacklisted());
2548 }
2549 
2550 /*
2551  * Attempt to allocate *count MSI messages.  The actual number allocated is
2552  * returned in *count.  After this function returns, each message will be
2553  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2554  */
2555 int
2556 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2557 {
2558 	struct pci_devinfo *dinfo = device_get_ivars(child);
2559 	pcicfgregs *cfg = &dinfo->cfg;
2560 	struct resource_list_entry *rle;
2561 	int actual, error, i, irqs[32];
2562 	uint16_t ctrl;
2563 
2564 	/* Don't let count == 0 get us into trouble. */
2565 	if (*count == 0)
2566 		return (EINVAL);
2567 
2568 	/* If rid 0 is allocated, then fail. */
2569 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2570 	if (rle != NULL && rle->res != NULL)
2571 		return (ENXIO);
2572 
2573 	/* Already have allocated messages? */
2574 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2575 		return (ENXIO);
2576 
2577 	/* If MSI is blacklisted for this system, fail. */
2578 	if (pci_msi_blacklisted())
2579 		return (ENXIO);
2580 
2581 	/* MSI capability present? */
2582 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2583 		return (ENODEV);
2584 
2585 	if (bootverbose)
2586 		device_printf(child,
2587 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2588 		    *count, cfg->msi.msi_msgnum);
2589 
2590 	/* Don't ask for more than the device supports. */
2591 	actual = min(*count, cfg->msi.msi_msgnum);
2592 
2593 	/* Don't ask for more than 32 messages. */
2594 	actual = min(actual, 32);
2595 
2596 	/* MSI requires power of 2 number of messages. */
2597 	if (!powerof2(actual))
2598 		return (EINVAL);
2599 
2600 	for (;;) {
2601 		/* Try to allocate N messages. */
2602 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2603 		    actual, irqs);
2604 		if (error == 0)
2605 			break;
2606 		if (actual == 1)
2607 			return (error);
2608 
2609 		/* Try N / 2. */
2610 		actual >>= 1;
2611 	}
2612 
2613 	/*
2614 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2615 	 * resources in the irqs[] array, so add new resources
2616 	 * starting at rid 1.
2617 	 */
2618 	for (i = 0; i < actual; i++)
2619 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2620 		    irqs[i], irqs[i], 1);
2621 
2622 	if (bootverbose) {
2623 		if (actual == 1)
2624 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2625 		else {
2626 			int run;
2627 
2628 			/*
2629 			 * Be fancy and try to print contiguous runs
2630 			 * of IRQ values as ranges.  'run' is true if
2631 			 * we are in a range.
2632 			 */
2633 			device_printf(child, "using IRQs %d", irqs[0]);
2634 			run = 0;
2635 			for (i = 1; i < actual; i++) {
2636 				/* Still in a run? */
2637 				if (irqs[i] == irqs[i - 1] + 1) {
2638 					run = 1;
2639 					continue;
2640 				}
2641 
2642 				/* Finish previous range. */
2643 				if (run) {
2644 					printf("-%d", irqs[i - 1]);
2645 					run = 0;
2646 				}
2647 
2648 				/* Start new range. */
2649 				printf(",%d", irqs[i]);
2650 			}
2651 
2652 			/* Unfinished range? */
2653 			if (run)
2654 				printf("-%d", irqs[actual - 1]);
2655 			printf(" for MSI\n");
2656 		}
2657 	}
2658 
2659 	/* Update control register with actual count. */
2660 	ctrl = cfg->msi.msi_ctrl;
2661 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2662 	ctrl |= (ffs(actual) - 1) << 4;
2663 	cfg->msi.msi_ctrl = ctrl;
2664 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2665 
2666 	/* Update counts of alloc'd messages. */
2667 	cfg->msi.msi_alloc = actual;
2668 	cfg->msi.msi_handlers = 0;
2669 	*count = actual;
2670 	return (0);
2671 }
2672 
2673 /* Release the MSI messages associated with this device. */
2674 int
2675 pci_release_msi_method(device_t dev, device_t child)
2676 {
2677 	struct pci_devinfo *dinfo = device_get_ivars(child);
2678 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2679 	struct resource_list_entry *rle;
2680 	int error, i, irqs[32];
2681 
2682 	/* Try MSI-X first. */
2683 	error = pci_release_msix(dev, child);
2684 	if (error != ENODEV)
2685 		return (error);
2686 
2687 	/* Do we have any messages to release? */
2688 	if (msi->msi_alloc == 0)
2689 		return (ENODEV);
2690 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2691 
2692 	/* Make sure none of the resources are allocated. */
2693 	if (msi->msi_handlers > 0)
2694 		return (EBUSY);
2695 	for (i = 0; i < msi->msi_alloc; i++) {
2696 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2697 		KASSERT(rle != NULL, ("missing MSI resource"));
2698 		if (rle->res != NULL)
2699 			return (EBUSY);
2700 		irqs[i] = rle->start;
2701 	}
2702 
2703 	/* Update control register with 0 count. */
2704 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2705 	    ("%s: MSI still enabled", __func__));
2706 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2707 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2708 	    msi->msi_ctrl, 2);
2709 
2710 	/* Release the messages. */
2711 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2712 	for (i = 0; i < msi->msi_alloc; i++)
2713 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2714 
2715 	/* Update alloc count. */
2716 	msi->msi_alloc = 0;
2717 	msi->msi_addr = 0;
2718 	msi->msi_data = 0;
2719 	return (0);
2720 }
2721 
2722 /*
2723  * Return the max supported MSI messages this device supports.
2724  * Basically, assuming the MD code can alloc messages, this function
2725  * should return the maximum value that pci_alloc_msi() can return.
2726  * Thus, it is subject to the tunables, etc.
2727  */
2728 int
2729 pci_msi_count_method(device_t dev, device_t child)
2730 {
2731 	struct pci_devinfo *dinfo = device_get_ivars(child);
2732 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2733 
2734 	if (pci_do_msi && msi->msi_location != 0)
2735 		return (msi->msi_msgnum);
2736 	return (0);
2737 }
2738 
2739 /* free pcicfgregs structure and all depending data structures */
2740 
2741 int
2742 pci_freecfg(struct pci_devinfo *dinfo)
2743 {
2744 	struct devlist *devlist_head;
2745 	struct pci_map *pm, *next;
2746 	int i;
2747 
2748 	devlist_head = &pci_devq;
2749 
2750 	if (dinfo->cfg.vpd.vpd_reg) {
2751 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2752 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2753 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2754 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2755 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2756 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2757 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2758 	}
2759 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2760 		free(pm, M_DEVBUF);
2761 	}
2762 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2763 	free(dinfo, M_DEVBUF);
2764 
2765 	/* increment the generation count */
2766 	pci_generation++;
2767 
2768 	/* we're losing one device */
2769 	pci_numdevs--;
2770 	return (0);
2771 }
2772 
2773 /*
2774  * PCI power manangement
2775  */
2776 int
2777 pci_set_powerstate_method(device_t dev, device_t child, int state)
2778 {
2779 	struct pci_devinfo *dinfo = device_get_ivars(child);
2780 	pcicfgregs *cfg = &dinfo->cfg;
2781 	uint16_t status;
2782 	int oldstate, highest, delay;
2783 
2784 	if (cfg->pp.pp_cap == 0)
2785 		return (EOPNOTSUPP);
2786 
2787 	/*
2788 	 * Optimize a no state change request away.  While it would be OK to
2789 	 * write to the hardware in theory, some devices have shown odd
2790 	 * behavior when going from D3 -> D3.
2791 	 */
2792 	oldstate = pci_get_powerstate(child);
2793 	if (oldstate == state)
2794 		return (0);
2795 
2796 	/*
2797 	 * The PCI power management specification states that after a state
2798 	 * transition between PCI power states, system software must
2799 	 * guarantee a minimal delay before the function accesses the device.
2800 	 * Compute the worst case delay that we need to guarantee before we
2801 	 * access the device.  Many devices will be responsive much more
2802 	 * quickly than this delay, but there are some that don't respond
2803 	 * instantly to state changes.  Transitions to/from D3 state require
2804 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2805 	 * is done below with DELAY rather than a sleeper function because
2806 	 * this function can be called from contexts where we cannot sleep.
2807 	 */
2808 	highest = (oldstate > state) ? oldstate : state;
2809 	if (highest == PCI_POWERSTATE_D3)
2810 	    delay = 10000;
2811 	else if (highest == PCI_POWERSTATE_D2)
2812 	    delay = 200;
2813 	else
2814 	    delay = 0;
2815 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2816 	    & ~PCIM_PSTAT_DMASK;
2817 	switch (state) {
2818 	case PCI_POWERSTATE_D0:
2819 		status |= PCIM_PSTAT_D0;
2820 		break;
2821 	case PCI_POWERSTATE_D1:
2822 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2823 			return (EOPNOTSUPP);
2824 		status |= PCIM_PSTAT_D1;
2825 		break;
2826 	case PCI_POWERSTATE_D2:
2827 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2828 			return (EOPNOTSUPP);
2829 		status |= PCIM_PSTAT_D2;
2830 		break;
2831 	case PCI_POWERSTATE_D3:
2832 		status |= PCIM_PSTAT_D3;
2833 		break;
2834 	default:
2835 		return (EINVAL);
2836 	}
2837 
2838 	if (bootverbose)
2839 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2840 		    state);
2841 
2842 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2843 	if (delay)
2844 		DELAY(delay);
2845 	return (0);
2846 }
2847 
2848 int
2849 pci_get_powerstate_method(device_t dev, device_t child)
2850 {
2851 	struct pci_devinfo *dinfo = device_get_ivars(child);
2852 	pcicfgregs *cfg = &dinfo->cfg;
2853 	uint16_t status;
2854 	int result;
2855 
2856 	if (cfg->pp.pp_cap != 0) {
2857 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2858 		switch (status & PCIM_PSTAT_DMASK) {
2859 		case PCIM_PSTAT_D0:
2860 			result = PCI_POWERSTATE_D0;
2861 			break;
2862 		case PCIM_PSTAT_D1:
2863 			result = PCI_POWERSTATE_D1;
2864 			break;
2865 		case PCIM_PSTAT_D2:
2866 			result = PCI_POWERSTATE_D2;
2867 			break;
2868 		case PCIM_PSTAT_D3:
2869 			result = PCI_POWERSTATE_D3;
2870 			break;
2871 		default:
2872 			result = PCI_POWERSTATE_UNKNOWN;
2873 			break;
2874 		}
2875 	} else {
2876 		/* No support, device is always at D0 */
2877 		result = PCI_POWERSTATE_D0;
2878 	}
2879 	return (result);
2880 }
2881 
2882 /*
2883  * Some convenience functions for PCI device drivers.
2884  */
2885 
2886 static __inline void
2887 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2888 {
2889 	uint16_t	command;
2890 
2891 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2892 	command |= bit;
2893 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2894 }
2895 
2896 static __inline void
2897 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2898 {
2899 	uint16_t	command;
2900 
2901 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2902 	command &= ~bit;
2903 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2904 }
2905 
2906 int
2907 pci_enable_busmaster_method(device_t dev, device_t child)
2908 {
2909 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2910 	return (0);
2911 }
2912 
2913 int
2914 pci_disable_busmaster_method(device_t dev, device_t child)
2915 {
2916 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2917 	return (0);
2918 }
2919 
2920 int
2921 pci_enable_io_method(device_t dev, device_t child, int space)
2922 {
2923 	uint16_t bit;
2924 
2925 	switch(space) {
2926 	case SYS_RES_IOPORT:
2927 		bit = PCIM_CMD_PORTEN;
2928 		break;
2929 	case SYS_RES_MEMORY:
2930 		bit = PCIM_CMD_MEMEN;
2931 		break;
2932 	default:
2933 		return (EINVAL);
2934 	}
2935 	pci_set_command_bit(dev, child, bit);
2936 	return (0);
2937 }
2938 
2939 int
2940 pci_disable_io_method(device_t dev, device_t child, int space)
2941 {
2942 	uint16_t bit;
2943 
2944 	switch(space) {
2945 	case SYS_RES_IOPORT:
2946 		bit = PCIM_CMD_PORTEN;
2947 		break;
2948 	case SYS_RES_MEMORY:
2949 		bit = PCIM_CMD_MEMEN;
2950 		break;
2951 	default:
2952 		return (EINVAL);
2953 	}
2954 	pci_clear_command_bit(dev, child, bit);
2955 	return (0);
2956 }
2957 
2958 /*
2959  * New style pci driver.  Parent device is either a pci-host-bridge or a
2960  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2961  */
2962 
2963 void
2964 pci_print_verbose(struct pci_devinfo *dinfo)
2965 {
2966 
2967 	if (bootverbose) {
2968 		pcicfgregs *cfg = &dinfo->cfg;
2969 
2970 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2971 		    cfg->vendor, cfg->device, cfg->revid);
2972 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2973 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2974 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2975 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2976 		    cfg->mfdev);
2977 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2978 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2979 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2980 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2981 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2982 		if (cfg->intpin > 0)
2983 			printf("\tintpin=%c, irq=%d\n",
2984 			    cfg->intpin +'a' -1, cfg->intline);
2985 		if (cfg->pp.pp_cap) {
2986 			uint16_t status;
2987 
2988 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2989 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2990 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2991 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2992 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2993 			    status & PCIM_PSTAT_DMASK);
2994 		}
2995 		if (cfg->msi.msi_location) {
2996 			int ctrl;
2997 
2998 			ctrl = cfg->msi.msi_ctrl;
2999 			printf("\tMSI supports %d message%s%s%s\n",
3000 			    cfg->msi.msi_msgnum,
3001 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
3002 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3003 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3004 		}
3005 		if (cfg->msix.msix_location) {
3006 			printf("\tMSI-X supports %d message%s ",
3007 			    cfg->msix.msix_msgnum,
3008 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
3009 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3010 				printf("in map 0x%x\n",
3011 				    cfg->msix.msix_table_bar);
3012 			else
3013 				printf("in maps 0x%x and 0x%x\n",
3014 				    cfg->msix.msix_table_bar,
3015 				    cfg->msix.msix_pba_bar);
3016 		}
3017 	}
3018 }
3019 
3020 static int
3021 pci_porten(device_t dev)
3022 {
3023 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3024 }
3025 
3026 static int
3027 pci_memen(device_t dev)
3028 {
3029 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3030 }
3031 
3032 void
3033 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3034     int *bar64)
3035 {
3036 	struct pci_devinfo *dinfo;
3037 	pci_addr_t map, testval;
3038 	int ln2range;
3039 	uint16_t cmd;
3040 
3041 	/*
3042 	 * The device ROM BAR is special.  It is always a 32-bit
3043 	 * memory BAR.  Bit 0 is special and should not be set when
3044 	 * sizing the BAR.
3045 	 */
3046 	dinfo = device_get_ivars(dev);
3047 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3048 		map = pci_read_config(dev, reg, 4);
3049 		pci_write_config(dev, reg, 0xfffffffe, 4);
3050 		testval = pci_read_config(dev, reg, 4);
3051 		pci_write_config(dev, reg, map, 4);
3052 		*mapp = map;
3053 		*testvalp = testval;
3054 		if (bar64 != NULL)
3055 			*bar64 = 0;
3056 		return;
3057 	}
3058 
3059 	map = pci_read_config(dev, reg, 4);
3060 	ln2range = pci_maprange(map);
3061 	if (ln2range == 64)
3062 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3063 
3064 	/*
3065 	 * Disable decoding via the command register before
3066 	 * determining the BAR's length since we will be placing it in
3067 	 * a weird state.
3068 	 */
3069 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3070 	pci_write_config(dev, PCIR_COMMAND,
3071 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3072 
3073 	/*
3074 	 * Determine the BAR's length by writing all 1's.  The bottom
3075 	 * log_2(size) bits of the BAR will stick as 0 when we read
3076 	 * the value back.
3077 	 *
3078 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3079 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3080 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3081 	 *
3082 	 * Writes to both registers must be performed before attempting to
3083 	 * read back the size value.
3084 	 */
3085 	testval = 0;
3086 	pci_write_config(dev, reg, 0xffffffff, 4);
3087 	if (ln2range == 64) {
3088 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3089 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3090 	}
3091 	testval |= pci_read_config(dev, reg, 4);
3092 
3093 	/*
3094 	 * Restore the original value of the BAR.  We may have reprogrammed
3095 	 * the BAR of the low-level console device and when booting verbose,
3096 	 * we need the console device addressable.
3097 	 */
3098 	pci_write_config(dev, reg, map, 4);
3099 	if (ln2range == 64)
3100 		pci_write_config(dev, reg + 4, map >> 32, 4);
3101 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3102 
3103 	*mapp = map;
3104 	*testvalp = testval;
3105 	if (bar64 != NULL)
3106 		*bar64 = (ln2range == 64);
3107 }
3108 
3109 static void
3110 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3111 {
3112 	struct pci_devinfo *dinfo;
3113 	int ln2range;
3114 
3115 	/* The device ROM BAR is always a 32-bit memory BAR. */
3116 	dinfo = device_get_ivars(dev);
3117 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3118 		ln2range = 32;
3119 	else
3120 		ln2range = pci_maprange(pm->pm_value);
3121 	pci_write_config(dev, pm->pm_reg, base, 4);
3122 	if (ln2range == 64)
3123 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3124 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3125 	if (ln2range == 64)
3126 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3127 		    pm->pm_reg + 4, 4) << 32;
3128 }
3129 
3130 struct pci_map *
3131 pci_find_bar(device_t dev, int reg)
3132 {
3133 	struct pci_devinfo *dinfo;
3134 	struct pci_map *pm;
3135 
3136 	dinfo = device_get_ivars(dev);
3137 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3138 		if (pm->pm_reg == reg)
3139 			return (pm);
3140 	}
3141 	return (NULL);
3142 }
3143 
3144 int
3145 pci_bar_enabled(device_t dev, struct pci_map *pm)
3146 {
3147 	struct pci_devinfo *dinfo;
3148 	uint16_t cmd;
3149 
3150 	dinfo = device_get_ivars(dev);
3151 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3152 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3153 		return (0);
3154 #ifdef PCI_IOV
3155 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3156 		struct pcicfg_iov *iov;
3157 
3158 		iov = dinfo->cfg.iov;
3159 		cmd = pci_read_config(iov->iov_pf,
3160 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3161 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3162 	}
3163 #endif
3164 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3165 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3166 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3167 	else
3168 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3169 }
3170 
3171 struct pci_map *
3172 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3173 {
3174 	struct pci_devinfo *dinfo;
3175 	struct pci_map *pm, *prev;
3176 
3177 	dinfo = device_get_ivars(dev);
3178 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3179 	pm->pm_reg = reg;
3180 	pm->pm_value = value;
3181 	pm->pm_size = size;
3182 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3183 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3184 		    reg));
3185 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3186 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3187 			break;
3188 	}
3189 	if (prev != NULL)
3190 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3191 	else
3192 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3193 	return (pm);
3194 }
3195 
3196 static void
3197 pci_restore_bars(device_t dev)
3198 {
3199 	struct pci_devinfo *dinfo;
3200 	struct pci_map *pm;
3201 	int ln2range;
3202 
3203 	dinfo = device_get_ivars(dev);
3204 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3205 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3206 			ln2range = 32;
3207 		else
3208 			ln2range = pci_maprange(pm->pm_value);
3209 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3210 		if (ln2range == 64)
3211 			pci_write_config(dev, pm->pm_reg + 4,
3212 			    pm->pm_value >> 32, 4);
3213 	}
3214 }
3215 
3216 /*
3217  * Add a resource based on a pci map register. Return 1 if the map
3218  * register is a 32bit map register or 2 if it is a 64bit register.
3219  */
3220 static int
3221 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3222     int force, int prefetch)
3223 {
3224 	struct pci_map *pm;
3225 	pci_addr_t base, map, testval;
3226 	pci_addr_t start, end, count;
3227 	int barlen, basezero, flags, maprange, mapsize, type;
3228 	uint16_t cmd;
3229 	struct resource *res;
3230 
3231 	/*
3232 	 * The BAR may already exist if the device is a CardBus card
3233 	 * whose CIS is stored in this BAR.
3234 	 */
3235 	pm = pci_find_bar(dev, reg);
3236 	if (pm != NULL) {
3237 		maprange = pci_maprange(pm->pm_value);
3238 		barlen = maprange == 64 ? 2 : 1;
3239 		return (barlen);
3240 	}
3241 
3242 	pci_read_bar(dev, reg, &map, &testval, NULL);
3243 	if (PCI_BAR_MEM(map)) {
3244 		type = SYS_RES_MEMORY;
3245 		if (map & PCIM_BAR_MEM_PREFETCH)
3246 			prefetch = 1;
3247 	} else
3248 		type = SYS_RES_IOPORT;
3249 	mapsize = pci_mapsize(testval);
3250 	base = pci_mapbase(map);
3251 #ifdef __PCI_BAR_ZERO_VALID
3252 	basezero = 0;
3253 #else
3254 	basezero = base == 0;
3255 #endif
3256 	maprange = pci_maprange(map);
3257 	barlen = maprange == 64 ? 2 : 1;
3258 
3259 	/*
3260 	 * For I/O registers, if bottom bit is set, and the next bit up
3261 	 * isn't clear, we know we have a BAR that doesn't conform to the
3262 	 * spec, so ignore it.  Also, sanity check the size of the data
3263 	 * areas to the type of memory involved.  Memory must be at least
3264 	 * 16 bytes in size, while I/O ranges must be at least 4.
3265 	 */
3266 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3267 		return (barlen);
3268 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3269 	    (type == SYS_RES_IOPORT && mapsize < 2))
3270 		return (barlen);
3271 
3272 	/* Save a record of this BAR. */
3273 	pm = pci_add_bar(dev, reg, map, mapsize);
3274 	if (bootverbose) {
3275 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3276 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3277 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3278 			printf(", port disabled\n");
3279 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3280 			printf(", memory disabled\n");
3281 		else
3282 			printf(", enabled\n");
3283 	}
3284 
3285 	/*
3286 	 * If base is 0, then we have problems if this architecture does
3287 	 * not allow that.  It is best to ignore such entries for the
3288 	 * moment.  These will be allocated later if the driver specifically
3289 	 * requests them.  However, some removable buses look better when
3290 	 * all resources are allocated, so allow '0' to be overriden.
3291 	 *
3292 	 * Similarly treat maps whose values is the same as the test value
3293 	 * read back.  These maps have had all f's written to them by the
3294 	 * BIOS in an attempt to disable the resources.
3295 	 */
3296 	if (!force && (basezero || map == testval))
3297 		return (barlen);
3298 	if ((u_long)base != base) {
3299 		device_printf(bus,
3300 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3301 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3302 		    pci_get_function(dev), reg);
3303 		return (barlen);
3304 	}
3305 
3306 	/*
3307 	 * This code theoretically does the right thing, but has
3308 	 * undesirable side effects in some cases where peripherals
3309 	 * respond oddly to having these bits enabled.  Let the user
3310 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3311 	 * default).
3312 	 */
3313 	if (pci_enable_io_modes) {
3314 		/* Turn on resources that have been left off by a lazy BIOS */
3315 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3316 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3317 			cmd |= PCIM_CMD_PORTEN;
3318 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3319 		}
3320 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3321 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3322 			cmd |= PCIM_CMD_MEMEN;
3323 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3324 		}
3325 	} else {
3326 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3327 			return (barlen);
3328 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3329 			return (barlen);
3330 	}
3331 
3332 	count = (pci_addr_t)1 << mapsize;
3333 	flags = RF_ALIGNMENT_LOG2(mapsize);
3334 	if (prefetch)
3335 		flags |= RF_PREFETCHABLE;
3336 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3337 		start = 0;	/* Let the parent decide. */
3338 		end = ~0;
3339 	} else {
3340 		start = base;
3341 		end = base + count - 1;
3342 	}
3343 	resource_list_add(rl, type, reg, start, end, count);
3344 
3345 	/*
3346 	 * Try to allocate the resource for this BAR from our parent
3347 	 * so that this resource range is already reserved.  The
3348 	 * driver for this device will later inherit this resource in
3349 	 * pci_alloc_resource().
3350 	 */
3351 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3352 	    flags);
3353 	if ((pci_do_realloc_bars
3354 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3355 	    && res == NULL && (start != 0 || end != ~0)) {
3356 		/*
3357 		 * If the allocation fails, try to allocate a resource for
3358 		 * this BAR using any available range.  The firmware felt
3359 		 * it was important enough to assign a resource, so don't
3360 		 * disable decoding if we can help it.
3361 		 */
3362 		resource_list_delete(rl, type, reg);
3363 		resource_list_add(rl, type, reg, 0, ~0, count);
3364 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3365 		    count, flags);
3366 	}
3367 	if (res == NULL) {
3368 		/*
3369 		 * If the allocation fails, delete the resource list entry
3370 		 * and disable decoding for this device.
3371 		 *
3372 		 * If the driver requests this resource in the future,
3373 		 * pci_reserve_map() will try to allocate a fresh
3374 		 * resource range.
3375 		 */
3376 		resource_list_delete(rl, type, reg);
3377 		pci_disable_io(dev, type);
3378 		if (bootverbose)
3379 			device_printf(bus,
3380 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3381 			    pci_get_domain(dev), pci_get_bus(dev),
3382 			    pci_get_slot(dev), pci_get_function(dev), reg);
3383 	} else {
3384 		start = rman_get_start(res);
3385 		pci_write_bar(dev, pm, start);
3386 	}
3387 	return (barlen);
3388 }
3389 
3390 /*
3391  * For ATA devices we need to decide early what addressing mode to use.
3392  * Legacy demands that the primary and secondary ATA ports sits on the
3393  * same addresses that old ISA hardware did. This dictates that we use
3394  * those addresses and ignore the BAR's if we cannot set PCI native
3395  * addressing mode.
3396  */
3397 static void
3398 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3399     uint32_t prefetchmask)
3400 {
3401 	int rid, type, progif;
3402 #if 0
3403 	/* if this device supports PCI native addressing use it */
3404 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3405 	if ((progif & 0x8a) == 0x8a) {
3406 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3407 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3408 			printf("Trying ATA native PCI addressing mode\n");
3409 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3410 		}
3411 	}
3412 #endif
3413 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3414 	type = SYS_RES_IOPORT;
3415 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3416 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3417 		    prefetchmask & (1 << 0));
3418 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3419 		    prefetchmask & (1 << 1));
3420 	} else {
3421 		rid = PCIR_BAR(0);
3422 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3423 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3424 		    0x1f7, 8, 0);
3425 		rid = PCIR_BAR(1);
3426 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3427 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3428 		    0x3f6, 1, 0);
3429 	}
3430 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3431 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3432 		    prefetchmask & (1 << 2));
3433 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3434 		    prefetchmask & (1 << 3));
3435 	} else {
3436 		rid = PCIR_BAR(2);
3437 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3438 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3439 		    0x177, 8, 0);
3440 		rid = PCIR_BAR(3);
3441 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3442 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3443 		    0x376, 1, 0);
3444 	}
3445 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3446 	    prefetchmask & (1 << 4));
3447 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3448 	    prefetchmask & (1 << 5));
3449 }
3450 
3451 static void
3452 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3453 {
3454 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3455 	pcicfgregs *cfg = &dinfo->cfg;
3456 	char tunable_name[64];
3457 	int irq;
3458 
3459 	/* Has to have an intpin to have an interrupt. */
3460 	if (cfg->intpin == 0)
3461 		return;
3462 
3463 	/* Let the user override the IRQ with a tunable. */
3464 	irq = PCI_INVALID_IRQ;
3465 	snprintf(tunable_name, sizeof(tunable_name),
3466 	    "hw.pci%d.%d.%d.INT%c.irq",
3467 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3468 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3469 		irq = PCI_INVALID_IRQ;
3470 
3471 	/*
3472 	 * If we didn't get an IRQ via the tunable, then we either use the
3473 	 * IRQ value in the intline register or we ask the bus to route an
3474 	 * interrupt for us.  If force_route is true, then we only use the
3475 	 * value in the intline register if the bus was unable to assign an
3476 	 * IRQ.
3477 	 */
3478 	if (!PCI_INTERRUPT_VALID(irq)) {
3479 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3480 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3481 		if (!PCI_INTERRUPT_VALID(irq))
3482 			irq = cfg->intline;
3483 	}
3484 
3485 	/* If after all that we don't have an IRQ, just bail. */
3486 	if (!PCI_INTERRUPT_VALID(irq))
3487 		return;
3488 
3489 	/* Update the config register if it changed. */
3490 	if (irq != cfg->intline) {
3491 		cfg->intline = irq;
3492 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3493 	}
3494 
3495 	/* Add this IRQ as rid 0 interrupt resource. */
3496 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3497 }
3498 
3499 /* Perform early OHCI takeover from SMM. */
3500 static void
3501 ohci_early_takeover(device_t self)
3502 {
3503 	struct resource *res;
3504 	uint32_t ctl;
3505 	int rid;
3506 	int i;
3507 
3508 	rid = PCIR_BAR(0);
3509 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3510 	if (res == NULL)
3511 		return;
3512 
3513 	ctl = bus_read_4(res, OHCI_CONTROL);
3514 	if (ctl & OHCI_IR) {
3515 		if (bootverbose)
3516 			printf("ohci early: "
3517 			    "SMM active, request owner change\n");
3518 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3519 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3520 			DELAY(1000);
3521 			ctl = bus_read_4(res, OHCI_CONTROL);
3522 		}
3523 		if (ctl & OHCI_IR) {
3524 			if (bootverbose)
3525 				printf("ohci early: "
3526 				    "SMM does not respond, resetting\n");
3527 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3528 		}
3529 		/* Disable interrupts */
3530 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3531 	}
3532 
3533 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3534 }
3535 
3536 /* Perform early UHCI takeover from SMM. */
3537 static void
3538 uhci_early_takeover(device_t self)
3539 {
3540 	struct resource *res;
3541 	int rid;
3542 
3543 	/*
3544 	 * Set the PIRQD enable bit and switch off all the others. We don't
3545 	 * want legacy support to interfere with us XXX Does this also mean
3546 	 * that the BIOS won't touch the keyboard anymore if it is connected
3547 	 * to the ports of the root hub?
3548 	 */
3549 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3550 
3551 	/* Disable interrupts */
3552 	rid = PCI_UHCI_BASE_REG;
3553 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3554 	if (res != NULL) {
3555 		bus_write_2(res, UHCI_INTR, 0);
3556 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3557 	}
3558 }
3559 
3560 /* Perform early EHCI takeover from SMM. */
3561 static void
3562 ehci_early_takeover(device_t self)
3563 {
3564 	struct resource *res;
3565 	uint32_t cparams;
3566 	uint32_t eec;
3567 	uint8_t eecp;
3568 	uint8_t bios_sem;
3569 	uint8_t offs;
3570 	int rid;
3571 	int i;
3572 
3573 	rid = PCIR_BAR(0);
3574 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3575 	if (res == NULL)
3576 		return;
3577 
3578 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3579 
3580 	/* Synchronise with the BIOS if it owns the controller. */
3581 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3582 	    eecp = EHCI_EECP_NEXT(eec)) {
3583 		eec = pci_read_config(self, eecp, 4);
3584 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3585 			continue;
3586 		}
3587 		bios_sem = pci_read_config(self, eecp +
3588 		    EHCI_LEGSUP_BIOS_SEM, 1);
3589 		if (bios_sem == 0) {
3590 			continue;
3591 		}
3592 		if (bootverbose)
3593 			printf("ehci early: "
3594 			    "SMM active, request owner change\n");
3595 
3596 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3597 
3598 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3599 			DELAY(1000);
3600 			bios_sem = pci_read_config(self, eecp +
3601 			    EHCI_LEGSUP_BIOS_SEM, 1);
3602 		}
3603 
3604 		if (bios_sem != 0) {
3605 			if (bootverbose)
3606 				printf("ehci early: "
3607 				    "SMM does not respond\n");
3608 		}
3609 		/* Disable interrupts */
3610 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3611 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3612 	}
3613 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3614 }
3615 
3616 /* Perform early XHCI takeover from SMM. */
3617 static void
3618 xhci_early_takeover(device_t self)
3619 {
3620 	struct resource *res;
3621 	uint32_t cparams;
3622 	uint32_t eec;
3623 	uint8_t eecp;
3624 	uint8_t bios_sem;
3625 	uint8_t offs;
3626 	int rid;
3627 	int i;
3628 
3629 	rid = PCIR_BAR(0);
3630 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3631 	if (res == NULL)
3632 		return;
3633 
3634 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3635 
3636 	eec = -1;
3637 
3638 	/* Synchronise with the BIOS if it owns the controller. */
3639 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3640 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3641 		eec = bus_read_4(res, eecp);
3642 
3643 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3644 			continue;
3645 
3646 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3647 		if (bios_sem == 0)
3648 			continue;
3649 
3650 		if (bootverbose)
3651 			printf("xhci early: "
3652 			    "SMM active, request owner change\n");
3653 
3654 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3655 
3656 		/* wait a maximum of 5 second */
3657 
3658 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3659 			DELAY(1000);
3660 			bios_sem = bus_read_1(res, eecp +
3661 			    XHCI_XECP_BIOS_SEM);
3662 		}
3663 
3664 		if (bios_sem != 0) {
3665 			if (bootverbose)
3666 				printf("xhci early: "
3667 				    "SMM does not respond\n");
3668 		}
3669 
3670 		/* Disable interrupts */
3671 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3672 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3673 		bus_read_4(res, offs + XHCI_USBSTS);
3674 	}
3675 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3676 }
3677 
3678 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3679 static void
3680 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3681     struct resource_list *rl)
3682 {
3683 	struct resource *res;
3684 	char *cp;
3685 	rman_res_t start, end, count;
3686 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3687 
3688 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3689 	case PCIM_HDRTYPE_BRIDGE:
3690 		sec_reg = PCIR_SECBUS_1;
3691 		sub_reg = PCIR_SUBBUS_1;
3692 		break;
3693 	case PCIM_HDRTYPE_CARDBUS:
3694 		sec_reg = PCIR_SECBUS_2;
3695 		sub_reg = PCIR_SUBBUS_2;
3696 		break;
3697 	default:
3698 		return;
3699 	}
3700 
3701 	/*
3702 	 * If the existing bus range is valid, attempt to reserve it
3703 	 * from our parent.  If this fails for any reason, clear the
3704 	 * secbus and subbus registers.
3705 	 *
3706 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3707 	 * This would at least preserve the existing sec_bus if it is
3708 	 * valid.
3709 	 */
3710 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3711 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3712 
3713 	/* Quirk handling. */
3714 	switch (pci_get_devid(dev)) {
3715 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3716 		sup_bus = pci_read_config(dev, 0x41, 1);
3717 		if (sup_bus != 0xff) {
3718 			sec_bus = sup_bus + 1;
3719 			sub_bus = sup_bus + 1;
3720 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3721 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3722 		}
3723 		break;
3724 
3725 	case 0x00dd10de:
3726 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3727 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3728 			break;
3729 		if (strncmp(cp, "Compal", 6) != 0) {
3730 			freeenv(cp);
3731 			break;
3732 		}
3733 		freeenv(cp);
3734 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3735 			break;
3736 		if (strncmp(cp, "08A0", 4) != 0) {
3737 			freeenv(cp);
3738 			break;
3739 		}
3740 		freeenv(cp);
3741 		if (sub_bus < 0xa) {
3742 			sub_bus = 0xa;
3743 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3744 		}
3745 		break;
3746 	}
3747 
3748 	if (bootverbose)
3749 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3750 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3751 		start = sec_bus;
3752 		end = sub_bus;
3753 		count = end - start + 1;
3754 
3755 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3756 
3757 		/*
3758 		 * If requested, clear secondary bus registers in
3759 		 * bridge devices to force a complete renumbering
3760 		 * rather than reserving the existing range.  However,
3761 		 * preserve the existing size.
3762 		 */
3763 		if (pci_clear_buses)
3764 			goto clear;
3765 
3766 		rid = 0;
3767 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3768 		    start, end, count, 0);
3769 		if (res != NULL)
3770 			return;
3771 
3772 		if (bootverbose)
3773 			device_printf(bus,
3774 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3775 			    pci_get_domain(dev), pci_get_bus(dev),
3776 			    pci_get_slot(dev), pci_get_function(dev));
3777 	}
3778 
3779 clear:
3780 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3781 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3782 }
3783 
3784 static struct resource *
3785 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3786     rman_res_t end, rman_res_t count, u_int flags)
3787 {
3788 	struct pci_devinfo *dinfo;
3789 	pcicfgregs *cfg;
3790 	struct resource_list *rl;
3791 	struct resource *res;
3792 	int sec_reg, sub_reg;
3793 
3794 	dinfo = device_get_ivars(child);
3795 	cfg = &dinfo->cfg;
3796 	rl = &dinfo->resources;
3797 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3798 	case PCIM_HDRTYPE_BRIDGE:
3799 		sec_reg = PCIR_SECBUS_1;
3800 		sub_reg = PCIR_SUBBUS_1;
3801 		break;
3802 	case PCIM_HDRTYPE_CARDBUS:
3803 		sec_reg = PCIR_SECBUS_2;
3804 		sub_reg = PCIR_SUBBUS_2;
3805 		break;
3806 	default:
3807 		return (NULL);
3808 	}
3809 
3810 	if (*rid != 0)
3811 		return (NULL);
3812 
3813 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3814 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3815 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3816 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3817 		    start, end, count, flags & ~RF_ACTIVE);
3818 		if (res == NULL) {
3819 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3820 			device_printf(child, "allocating %ju bus%s failed\n",
3821 			    count, count == 1 ? "" : "es");
3822 			return (NULL);
3823 		}
3824 		if (bootverbose)
3825 			device_printf(child,
3826 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3827 			    count == 1 ? "" : "es", rman_get_start(res));
3828 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3829 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3830 	}
3831 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3832 	    end, count, flags));
3833 }
3834 #endif
3835 
3836 static int
3837 pci_ea_bei_to_rid(device_t dev, int bei)
3838 {
3839 #ifdef PCI_IOV
3840 	struct pci_devinfo *dinfo;
3841 	int iov_pos;
3842 	struct pcicfg_iov *iov;
3843 
3844 	dinfo = device_get_ivars(dev);
3845 	iov = dinfo->cfg.iov;
3846 	if (iov != NULL)
3847 		iov_pos = iov->iov_pos;
3848 	else
3849 		iov_pos = 0;
3850 #endif
3851 
3852 	/* Check if matches BAR */
3853 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3854 	    (bei <= PCIM_EA_BEI_BAR_5))
3855 		return (PCIR_BAR(bei));
3856 
3857 	/* Check ROM */
3858 	if (bei == PCIM_EA_BEI_ROM)
3859 		return (PCIR_BIOS);
3860 
3861 #ifdef PCI_IOV
3862 	/* Check if matches VF_BAR */
3863 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3864 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3865 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3866 		    iov_pos);
3867 #endif
3868 
3869 	return (-1);
3870 }
3871 
3872 int
3873 pci_ea_is_enabled(device_t dev, int rid)
3874 {
3875 	struct pci_ea_entry *ea;
3876 	struct pci_devinfo *dinfo;
3877 
3878 	dinfo = device_get_ivars(dev);
3879 
3880 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3881 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3882 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3883 	}
3884 
3885 	return (0);
3886 }
3887 
3888 void
3889 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3890 {
3891 	struct pci_ea_entry *ea;
3892 	struct pci_devinfo *dinfo;
3893 	pci_addr_t start, end, count;
3894 	struct resource_list *rl;
3895 	int type, flags, rid;
3896 	struct resource *res;
3897 	uint32_t tmp;
3898 #ifdef PCI_IOV
3899 	struct pcicfg_iov *iov;
3900 #endif
3901 
3902 	dinfo = device_get_ivars(dev);
3903 	rl = &dinfo->resources;
3904 	flags = 0;
3905 
3906 #ifdef PCI_IOV
3907 	iov = dinfo->cfg.iov;
3908 #endif
3909 
3910 	if (dinfo->cfg.ea.ea_location == 0)
3911 		return;
3912 
3913 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3914 		/*
3915 		 * TODO: Ignore EA-BAR if is not enabled.
3916 		 *   Currently the EA implementation supports
3917 		 *   only situation, where EA structure contains
3918 		 *   predefined entries. In case they are not enabled
3919 		 *   leave them unallocated and proceed with
3920 		 *   a legacy-BAR mechanism.
3921 		 */
3922 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3923 			continue;
3924 
3925 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3926 		case PCIM_EA_P_MEM_PREFETCH:
3927 		case PCIM_EA_P_VF_MEM_PREFETCH:
3928 			flags = RF_PREFETCHABLE;
3929 			/* FALLTHROUGH */
3930 		case PCIM_EA_P_VF_MEM:
3931 		case PCIM_EA_P_MEM:
3932 			type = SYS_RES_MEMORY;
3933 			break;
3934 		case PCIM_EA_P_IO:
3935 			type = SYS_RES_IOPORT;
3936 			break;
3937 		default:
3938 			continue;
3939 		}
3940 
3941 		if (alloc_iov != 0) {
3942 #ifdef PCI_IOV
3943 			/* Allocating IOV, confirm BEI matches */
3944 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3945 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3946 				continue;
3947 #else
3948 			continue;
3949 #endif
3950 		} else {
3951 			/* Allocating BAR, confirm BEI matches */
3952 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3953 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3954 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3955 				continue;
3956 		}
3957 
3958 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3959 		if (rid < 0)
3960 			continue;
3961 
3962 		/* Skip resources already allocated by EA */
3963 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3964 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3965 			continue;
3966 
3967 		start = ea->eae_base;
3968 		count = ea->eae_max_offset + 1;
3969 #ifdef PCI_IOV
3970 		if (iov != NULL)
3971 			count = count * iov->iov_num_vfs;
3972 #endif
3973 		end = start + count - 1;
3974 		if (count == 0)
3975 			continue;
3976 
3977 		resource_list_add(rl, type, rid, start, end, count);
3978 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3979 		    flags);
3980 		if (res == NULL) {
3981 			resource_list_delete(rl, type, rid);
3982 
3983 			/*
3984 			 * Failed to allocate using EA, disable entry.
3985 			 * Another attempt to allocation will be performed
3986 			 * further, but this time using legacy BAR registers
3987 			 */
3988 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3989 			tmp &= ~PCIM_EA_ENABLE;
3990 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3991 
3992 			/*
3993 			 * Disabling entry might fail in case it is hardwired.
3994 			 * Read flags again to match current status.
3995 			 */
3996 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3997 
3998 			continue;
3999 		}
4000 
4001 		/* As per specification, fill BAR with zeros */
4002 		pci_write_config(dev, rid, 0, 4);
4003 	}
4004 }
4005 
4006 void
4007 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4008 {
4009 	struct pci_devinfo *dinfo;
4010 	pcicfgregs *cfg;
4011 	struct resource_list *rl;
4012 	const struct pci_quirk *q;
4013 	uint32_t devid;
4014 	int i;
4015 
4016 	dinfo = device_get_ivars(dev);
4017 	cfg = &dinfo->cfg;
4018 	rl = &dinfo->resources;
4019 	devid = (cfg->device << 16) | cfg->vendor;
4020 
4021 	/* Allocate resources using Enhanced Allocation */
4022 	pci_add_resources_ea(bus, dev, 0);
4023 
4024 	/* ATA devices needs special map treatment */
4025 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4026 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4027 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4028 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4029 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4030 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4031 	else
4032 		for (i = 0; i < cfg->nummaps;) {
4033 			/* Skip resources already managed by EA */
4034 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4035 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4036 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4037 				i++;
4038 				continue;
4039 			}
4040 
4041 			/*
4042 			 * Skip quirked resources.
4043 			 */
4044 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4045 				if (q->devid == devid &&
4046 				    q->type == PCI_QUIRK_UNMAP_REG &&
4047 				    q->arg1 == PCIR_BAR(i))
4048 					break;
4049 			if (q->devid != 0) {
4050 				i++;
4051 				continue;
4052 			}
4053 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4054 			    prefetchmask & (1 << i));
4055 		}
4056 
4057 	/*
4058 	 * Add additional, quirked resources.
4059 	 */
4060 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4061 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4062 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4063 
4064 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4065 #ifdef __PCI_REROUTE_INTERRUPT
4066 		/*
4067 		 * Try to re-route interrupts. Sometimes the BIOS or
4068 		 * firmware may leave bogus values in these registers.
4069 		 * If the re-route fails, then just stick with what we
4070 		 * have.
4071 		 */
4072 		pci_assign_interrupt(bus, dev, 1);
4073 #else
4074 		pci_assign_interrupt(bus, dev, 0);
4075 #endif
4076 	}
4077 
4078 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4079 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4080 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4081 			xhci_early_takeover(dev);
4082 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4083 			ehci_early_takeover(dev);
4084 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4085 			ohci_early_takeover(dev);
4086 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4087 			uhci_early_takeover(dev);
4088 	}
4089 
4090 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4091 	/*
4092 	 * Reserve resources for secondary bus ranges behind bridge
4093 	 * devices.
4094 	 */
4095 	pci_reserve_secbus(bus, dev, cfg, rl);
4096 #endif
4097 }
4098 
4099 static struct pci_devinfo *
4100 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4101     int slot, int func)
4102 {
4103 	struct pci_devinfo *dinfo;
4104 
4105 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4106 	if (dinfo != NULL)
4107 		pci_add_child(dev, dinfo);
4108 
4109 	return (dinfo);
4110 }
4111 
4112 void
4113 pci_add_children(device_t dev, int domain, int busno)
4114 {
4115 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4116 	device_t pcib = device_get_parent(dev);
4117 	struct pci_devinfo *dinfo;
4118 	int maxslots;
4119 	int s, f, pcifunchigh;
4120 	uint8_t hdrtype;
4121 	int first_func;
4122 
4123 	/*
4124 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4125 	 * enable ARI.  We must enable ARI before detecting the rest of the
4126 	 * functions on this bus as ARI changes the set of slots and functions
4127 	 * that are legal on this bus.
4128 	 */
4129 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4130 	if (dinfo != NULL && pci_enable_ari)
4131 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4132 
4133 	/*
4134 	 * Start looking for new devices on slot 0 at function 1 because we
4135 	 * just identified the device at slot 0, function 0.
4136 	 */
4137 	first_func = 1;
4138 
4139 	maxslots = PCIB_MAXSLOTS(pcib);
4140 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4141 		pcifunchigh = 0;
4142 		f = 0;
4143 		DELAY(1);
4144 		hdrtype = REG(PCIR_HDRTYPE, 1);
4145 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4146 			continue;
4147 		if (hdrtype & PCIM_MFDEV)
4148 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4149 		for (f = first_func; f <= pcifunchigh; f++)
4150 			pci_identify_function(pcib, dev, domain, busno, s, f);
4151 	}
4152 #undef REG
4153 }
4154 
4155 int
4156 pci_rescan_method(device_t dev)
4157 {
4158 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4159 	device_t pcib = device_get_parent(dev);
4160 	device_t child, *devlist, *unchanged;
4161 	int devcount, error, i, j, maxslots, oldcount;
4162 	int busno, domain, s, f, pcifunchigh;
4163 	uint8_t hdrtype;
4164 
4165 	/* No need to check for ARI on a rescan. */
4166 	error = device_get_children(dev, &devlist, &devcount);
4167 	if (error)
4168 		return (error);
4169 	if (devcount != 0) {
4170 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4171 		    M_NOWAIT | M_ZERO);
4172 		if (unchanged == NULL) {
4173 			free(devlist, M_TEMP);
4174 			return (ENOMEM);
4175 		}
4176 	} else
4177 		unchanged = NULL;
4178 
4179 	domain = pcib_get_domain(dev);
4180 	busno = pcib_get_bus(dev);
4181 	maxslots = PCIB_MAXSLOTS(pcib);
4182 	for (s = 0; s <= maxslots; s++) {
4183 		/* If function 0 is not present, skip to the next slot. */
4184 		f = 0;
4185 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4186 			continue;
4187 		pcifunchigh = 0;
4188 		hdrtype = REG(PCIR_HDRTYPE, 1);
4189 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4190 			continue;
4191 		if (hdrtype & PCIM_MFDEV)
4192 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4193 		for (f = 0; f <= pcifunchigh; f++) {
4194 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4195 				continue;
4196 
4197 			/*
4198 			 * Found a valid function.  Check if a
4199 			 * device_t for this device already exists.
4200 			 */
4201 			for (i = 0; i < devcount; i++) {
4202 				child = devlist[i];
4203 				if (child == NULL)
4204 					continue;
4205 				if (pci_get_slot(child) == s &&
4206 				    pci_get_function(child) == f) {
4207 					unchanged[i] = child;
4208 					goto next_func;
4209 				}
4210 			}
4211 
4212 			pci_identify_function(pcib, dev, domain, busno, s, f);
4213 		next_func:;
4214 		}
4215 	}
4216 
4217 	/* Remove devices that are no longer present. */
4218 	for (i = 0; i < devcount; i++) {
4219 		if (unchanged[i] != NULL)
4220 			continue;
4221 		device_delete_child(dev, devlist[i]);
4222 	}
4223 
4224 	free(devlist, M_TEMP);
4225 	oldcount = devcount;
4226 
4227 	/* Try to attach the devices just added. */
4228 	error = device_get_children(dev, &devlist, &devcount);
4229 	if (error) {
4230 		free(unchanged, M_TEMP);
4231 		return (error);
4232 	}
4233 
4234 	for (i = 0; i < devcount; i++) {
4235 		for (j = 0; j < oldcount; j++) {
4236 			if (devlist[i] == unchanged[j])
4237 				goto next_device;
4238 		}
4239 
4240 		device_probe_and_attach(devlist[i]);
4241 	next_device:;
4242 	}
4243 
4244 	free(unchanged, M_TEMP);
4245 	free(devlist, M_TEMP);
4246 	return (0);
4247 #undef REG
4248 }
4249 
4250 #ifdef PCI_IOV
4251 device_t
4252 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4253     uint16_t did)
4254 {
4255 	struct pci_devinfo *vf_dinfo;
4256 	device_t pcib;
4257 	int busno, slot, func;
4258 
4259 	pcib = device_get_parent(bus);
4260 
4261 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4262 
4263 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4264 	    slot, func, vid, did);
4265 
4266 	vf_dinfo->cfg.flags |= PCICFG_VF;
4267 	pci_add_child(bus, vf_dinfo);
4268 
4269 	return (vf_dinfo->cfg.dev);
4270 }
4271 
4272 device_t
4273 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4274     uint16_t vid, uint16_t did)
4275 {
4276 
4277 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4278 }
4279 #endif
4280 
4281 /*
4282  * For PCIe device set Max_Payload_Size to match PCIe root's.
4283  */
4284 static void
4285 pcie_setup_mps(device_t dev)
4286 {
4287 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4288 	device_t root;
4289 	uint16_t rmps, mmps, mps;
4290 
4291 	if (dinfo->cfg.pcie.pcie_location == 0)
4292 		return;
4293 	root = pci_find_pcie_root_port(dev);
4294 	if (root == NULL)
4295 		return;
4296 	/* Check whether the MPS is already configured. */
4297 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4298 	    PCIEM_CTL_MAX_PAYLOAD;
4299 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4300 	    PCIEM_CTL_MAX_PAYLOAD;
4301 	if (mps == rmps)
4302 		return;
4303 	/* Check whether the device is capable of the root's MPS. */
4304 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4305 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4306 	if (rmps > mmps) {
4307 		/*
4308 		 * The device is unable to handle root's MPS.  Limit root.
4309 		 * XXX: We should traverse through all the tree, applying
4310 		 * it to all the devices.
4311 		 */
4312 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4313 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4314 	} else {
4315 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4316 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4317 	}
4318 }
4319 
4320 static void
4321 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4322 {
4323 	int aer;
4324 	uint32_t r;
4325 	uint16_t r2;
4326 
4327 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4328 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4329 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4330 		    PCIER_ROOT_CTL, 2);
4331 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4332 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4333 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4334 		    PCIER_ROOT_CTL, r2, 2);
4335 	}
4336 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4337 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4338 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4339 		if (r != 0 && bootverbose) {
4340 			pci_printf(&dinfo->cfg,
4341 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4342 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4343 			    4));
4344 		}
4345 
4346 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4347 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4348 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4349 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4350 		    PCIM_AER_UC_POISONED_TLP |
4351 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4352 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4353 		    PCIM_AER_UC_COMPLETER_ABORT |
4354 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4355 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4356 		    PCIM_AER_UC_MALFORMED_TLP |
4357 		    PCIM_AER_UC_ECRC_ERROR |
4358 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4359 		    PCIM_AER_UC_ACS_VIOLATION |
4360 		    PCIM_AER_UC_INTERNAL_ERROR |
4361 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4362 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4363 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4364 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4365 
4366 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4367 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4368 		if (r != 0 && bootverbose) {
4369 			pci_printf(&dinfo->cfg,
4370 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4371 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4372 			    4));
4373 		}
4374 
4375 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4376 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4377 		    PCIM_AER_COR_BAD_TLP |
4378 		    PCIM_AER_COR_BAD_DLLP |
4379 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4380 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4381 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4382 		    PCIM_AER_COR_INTERNAL_ERROR |
4383 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4384 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4385 
4386 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4387 		    PCIER_DEVICE_CTL, 2);
4388 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4389 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4390 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4391 		    PCIER_DEVICE_CTL, r, 2);
4392 	}
4393 }
4394 
4395 void
4396 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4397 {
4398 	device_t dev;
4399 
4400 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4401 	device_set_ivars(dev, dinfo);
4402 	resource_list_init(&dinfo->resources);
4403 	pci_cfg_save(dev, dinfo, 0);
4404 	pci_cfg_restore(dev, dinfo);
4405 	pci_print_verbose(dinfo);
4406 	pci_add_resources(bus, dev, 0, 0);
4407 	pcie_setup_mps(dev);
4408 	pci_child_added(dinfo->cfg.dev);
4409 
4410 	if (pci_clear_aer_on_attach)
4411 		pci_add_child_clear_aer(dev, dinfo);
4412 
4413 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4414 }
4415 
4416 void
4417 pci_child_added_method(device_t dev, device_t child)
4418 {
4419 
4420 }
4421 
4422 static int
4423 pci_probe(device_t dev)
4424 {
4425 
4426 	device_set_desc(dev, "PCI bus");
4427 
4428 	/* Allow other subclasses to override this driver. */
4429 	return (BUS_PROBE_GENERIC);
4430 }
4431 
4432 int
4433 pci_attach_common(device_t dev)
4434 {
4435 	struct pci_softc *sc;
4436 	int busno, domain;
4437 #ifdef PCI_RES_BUS
4438 	int rid;
4439 #endif
4440 
4441 	sc = device_get_softc(dev);
4442 	domain = pcib_get_domain(dev);
4443 	busno = pcib_get_bus(dev);
4444 #ifdef PCI_RES_BUS
4445 	rid = 0;
4446 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4447 	    1, 0);
4448 	if (sc->sc_bus == NULL) {
4449 		device_printf(dev, "failed to allocate bus number\n");
4450 		return (ENXIO);
4451 	}
4452 #endif
4453 	if (bootverbose)
4454 		device_printf(dev, "domain=%d, physical bus=%d\n",
4455 		    domain, busno);
4456 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4457 	return (0);
4458 }
4459 
4460 int
4461 pci_attach(device_t dev)
4462 {
4463 	int busno, domain, error;
4464 
4465 	error = pci_attach_common(dev);
4466 	if (error)
4467 		return (error);
4468 
4469 	/*
4470 	 * Since there can be multiple independently numbered PCI
4471 	 * buses on systems with multiple PCI domains, we can't use
4472 	 * the unit number to decide which bus we are probing. We ask
4473 	 * the parent pcib what our domain and bus numbers are.
4474 	 */
4475 	domain = pcib_get_domain(dev);
4476 	busno = pcib_get_bus(dev);
4477 	pci_add_children(dev, domain, busno);
4478 	return (bus_generic_attach(dev));
4479 }
4480 
4481 int
4482 pci_detach(device_t dev)
4483 {
4484 #ifdef PCI_RES_BUS
4485 	struct pci_softc *sc;
4486 #endif
4487 	int error;
4488 
4489 	error = bus_generic_detach(dev);
4490 	if (error)
4491 		return (error);
4492 #ifdef PCI_RES_BUS
4493 	sc = device_get_softc(dev);
4494 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4495 	if (error)
4496 		return (error);
4497 #endif
4498 	return (device_delete_children(dev));
4499 }
4500 
4501 static void
4502 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4503 {
4504 	int line, unit;
4505 	const char *at;
4506 	char me1[24], me2[32];
4507 	uint8_t b, s, f;
4508 	uint32_t d;
4509 
4510 	d = pci_get_domain(child);
4511 	b = pci_get_bus(child);
4512 	s = pci_get_slot(child);
4513 	f = pci_get_function(child);
4514 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4515 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4516 	line = 0;
4517 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4518 		resource_string_value(name, unit, "at", &at);
4519 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4520 			continue; /* No match, try next candidate */
4521 		*unitp = unit;
4522 		return;
4523 	}
4524 }
4525 
4526 static void
4527 pci_set_power_child(device_t dev, device_t child, int state)
4528 {
4529 	device_t pcib;
4530 	int dstate;
4531 
4532 	/*
4533 	 * Set the device to the given state.  If the firmware suggests
4534 	 * a different power state, use it instead.  If power management
4535 	 * is not present, the firmware is responsible for managing
4536 	 * device power.  Skip children who aren't attached since they
4537 	 * are handled separately.
4538 	 */
4539 	pcib = device_get_parent(dev);
4540 	dstate = state;
4541 	if (device_is_attached(child) &&
4542 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4543 		pci_set_powerstate(child, dstate);
4544 }
4545 
4546 int
4547 pci_suspend_child(device_t dev, device_t child)
4548 {
4549 	struct pci_devinfo *dinfo;
4550 	struct resource_list_entry *rle;
4551 	int error;
4552 
4553 	dinfo = device_get_ivars(child);
4554 
4555 	/*
4556 	 * Save the PCI configuration space for the child and set the
4557 	 * device in the appropriate power state for this sleep state.
4558 	 */
4559 	pci_cfg_save(child, dinfo, 0);
4560 
4561 	/* Suspend devices before potentially powering them down. */
4562 	error = bus_generic_suspend_child(dev, child);
4563 
4564 	if (error)
4565 		return (error);
4566 
4567 	if (pci_do_power_suspend) {
4568 		/*
4569 		 * Make sure this device's interrupt handler is not invoked
4570 		 * in the case the device uses a shared interrupt that can
4571 		 * be raised by some other device.
4572 		 * This is applicable only to regular (legacy) PCI interrupts
4573 		 * as MSI/MSI-X interrupts are never shared.
4574 		 */
4575 		rle = resource_list_find(&dinfo->resources,
4576 		    SYS_RES_IRQ, 0);
4577 		if (rle != NULL && rle->res != NULL)
4578 			(void)bus_suspend_intr(child, rle->res);
4579 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4580 	}
4581 
4582 	return (0);
4583 }
4584 
4585 int
4586 pci_resume_child(device_t dev, device_t child)
4587 {
4588 	struct pci_devinfo *dinfo;
4589 	struct resource_list_entry *rle;
4590 
4591 	if (pci_do_power_resume)
4592 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4593 
4594 	dinfo = device_get_ivars(child);
4595 	pci_cfg_restore(child, dinfo);
4596 	if (!device_is_attached(child))
4597 		pci_cfg_save(child, dinfo, 1);
4598 
4599 	bus_generic_resume_child(dev, child);
4600 
4601 	/*
4602 	 * Allow interrupts only after fully resuming the driver and hardware.
4603 	 */
4604 	if (pci_do_power_suspend) {
4605 		/* See pci_suspend_child for details. */
4606 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4607 		if (rle != NULL && rle->res != NULL)
4608 			(void)bus_resume_intr(child, rle->res);
4609 	}
4610 
4611 	return (0);
4612 }
4613 
4614 int
4615 pci_resume(device_t dev)
4616 {
4617 	device_t child, *devlist;
4618 	int error, i, numdevs;
4619 
4620 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4621 		return (error);
4622 
4623 	/*
4624 	 * Resume critical devices first, then everything else later.
4625 	 */
4626 	for (i = 0; i < numdevs; i++) {
4627 		child = devlist[i];
4628 		switch (pci_get_class(child)) {
4629 		case PCIC_DISPLAY:
4630 		case PCIC_MEMORY:
4631 		case PCIC_BRIDGE:
4632 		case PCIC_BASEPERIPH:
4633 			BUS_RESUME_CHILD(dev, child);
4634 			break;
4635 		}
4636 	}
4637 	for (i = 0; i < numdevs; i++) {
4638 		child = devlist[i];
4639 		switch (pci_get_class(child)) {
4640 		case PCIC_DISPLAY:
4641 		case PCIC_MEMORY:
4642 		case PCIC_BRIDGE:
4643 		case PCIC_BASEPERIPH:
4644 			break;
4645 		default:
4646 			BUS_RESUME_CHILD(dev, child);
4647 		}
4648 	}
4649 	free(devlist, M_TEMP);
4650 	return (0);
4651 }
4652 
4653 static void
4654 pci_load_vendor_data(void)
4655 {
4656 	caddr_t data;
4657 	void *ptr;
4658 	size_t sz;
4659 
4660 	data = preload_search_by_type("pci_vendor_data");
4661 	if (data != NULL) {
4662 		ptr = preload_fetch_addr(data);
4663 		sz = preload_fetch_size(data);
4664 		if (ptr != NULL && sz != 0) {
4665 			pci_vendordata = ptr;
4666 			pci_vendordata_size = sz;
4667 			/* terminate the database */
4668 			pci_vendordata[pci_vendordata_size] = '\n';
4669 		}
4670 	}
4671 }
4672 
4673 void
4674 pci_driver_added(device_t dev, driver_t *driver)
4675 {
4676 	int numdevs;
4677 	device_t *devlist;
4678 	device_t child;
4679 	struct pci_devinfo *dinfo;
4680 	int i;
4681 
4682 	if (bootverbose)
4683 		device_printf(dev, "driver added\n");
4684 	DEVICE_IDENTIFY(driver, dev);
4685 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4686 		return;
4687 	for (i = 0; i < numdevs; i++) {
4688 		child = devlist[i];
4689 		if (device_get_state(child) != DS_NOTPRESENT)
4690 			continue;
4691 		dinfo = device_get_ivars(child);
4692 		pci_print_verbose(dinfo);
4693 		if (bootverbose)
4694 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4695 		pci_cfg_restore(child, dinfo);
4696 		if (device_probe_and_attach(child) != 0)
4697 			pci_child_detached(dev, child);
4698 	}
4699 	free(devlist, M_TEMP);
4700 }
4701 
4702 int
4703 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4704     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4705 {
4706 	struct pci_devinfo *dinfo;
4707 	struct msix_table_entry *mte;
4708 	struct msix_vector *mv;
4709 	uint64_t addr;
4710 	uint32_t data;
4711 	void *cookie;
4712 	int error, rid;
4713 
4714 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4715 	    arg, &cookie);
4716 	if (error)
4717 		return (error);
4718 
4719 	/* If this is not a direct child, just bail out. */
4720 	if (device_get_parent(child) != dev) {
4721 		*cookiep = cookie;
4722 		return(0);
4723 	}
4724 
4725 	rid = rman_get_rid(irq);
4726 	if (rid == 0) {
4727 		/* Make sure that INTx is enabled */
4728 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4729 	} else {
4730 		/*
4731 		 * Check to see if the interrupt is MSI or MSI-X.
4732 		 * Ask our parent to map the MSI and give
4733 		 * us the address and data register values.
4734 		 * If we fail for some reason, teardown the
4735 		 * interrupt handler.
4736 		 */
4737 		dinfo = device_get_ivars(child);
4738 		if (dinfo->cfg.msi.msi_alloc > 0) {
4739 			if (dinfo->cfg.msi.msi_addr == 0) {
4740 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4741 			    ("MSI has handlers, but vectors not mapped"));
4742 				error = PCIB_MAP_MSI(device_get_parent(dev),
4743 				    child, rman_get_start(irq), &addr, &data);
4744 				if (error)
4745 					goto bad;
4746 				dinfo->cfg.msi.msi_addr = addr;
4747 				dinfo->cfg.msi.msi_data = data;
4748 			}
4749 			if (dinfo->cfg.msi.msi_handlers == 0)
4750 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4751 				    dinfo->cfg.msi.msi_data);
4752 			dinfo->cfg.msi.msi_handlers++;
4753 		} else {
4754 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4755 			    ("No MSI or MSI-X interrupts allocated"));
4756 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4757 			    ("MSI-X index too high"));
4758 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4759 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4760 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4761 			KASSERT(mv->mv_irq == rman_get_start(irq),
4762 			    ("IRQ mismatch"));
4763 			if (mv->mv_address == 0) {
4764 				KASSERT(mte->mte_handlers == 0,
4765 		    ("MSI-X table entry has handlers, but vector not mapped"));
4766 				error = PCIB_MAP_MSI(device_get_parent(dev),
4767 				    child, rman_get_start(irq), &addr, &data);
4768 				if (error)
4769 					goto bad;
4770 				mv->mv_address = addr;
4771 				mv->mv_data = data;
4772 			}
4773 
4774 			/*
4775 			 * The MSIX table entry must be made valid by
4776 			 * incrementing the mte_handlers before
4777 			 * calling pci_enable_msix() and
4778 			 * pci_resume_msix(). Else the MSIX rewrite
4779 			 * table quirk will not work as expected.
4780 			 */
4781 			mte->mte_handlers++;
4782 			if (mte->mte_handlers == 1) {
4783 				pci_enable_msix(child, rid - 1, mv->mv_address,
4784 				    mv->mv_data);
4785 				pci_unmask_msix(child, rid - 1);
4786 			}
4787 		}
4788 
4789 		/*
4790 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4791 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4792 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4793 		 */
4794 		if (!pci_has_quirk(pci_get_devid(child),
4795 		    PCI_QUIRK_MSI_INTX_BUG))
4796 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4797 		else
4798 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4799 	bad:
4800 		if (error) {
4801 			(void)bus_generic_teardown_intr(dev, child, irq,
4802 			    cookie);
4803 			return (error);
4804 		}
4805 	}
4806 	*cookiep = cookie;
4807 	return (0);
4808 }
4809 
4810 int
4811 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4812     void *cookie)
4813 {
4814 	struct msix_table_entry *mte;
4815 	struct resource_list_entry *rle;
4816 	struct pci_devinfo *dinfo;
4817 	int error, rid;
4818 
4819 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4820 		return (EINVAL);
4821 
4822 	/* If this isn't a direct child, just bail out */
4823 	if (device_get_parent(child) != dev)
4824 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4825 
4826 	rid = rman_get_rid(irq);
4827 	if (rid == 0) {
4828 		/* Mask INTx */
4829 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4830 	} else {
4831 		/*
4832 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4833 		 * decrement the appropriate handlers count and mask the
4834 		 * MSI-X message, or disable MSI messages if the count
4835 		 * drops to 0.
4836 		 */
4837 		dinfo = device_get_ivars(child);
4838 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4839 		if (rle->res != irq)
4840 			return (EINVAL);
4841 		if (dinfo->cfg.msi.msi_alloc > 0) {
4842 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4843 			    ("MSI-X index too high"));
4844 			if (dinfo->cfg.msi.msi_handlers == 0)
4845 				return (EINVAL);
4846 			dinfo->cfg.msi.msi_handlers--;
4847 			if (dinfo->cfg.msi.msi_handlers == 0)
4848 				pci_disable_msi(child);
4849 		} else {
4850 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4851 			    ("No MSI or MSI-X interrupts allocated"));
4852 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4853 			    ("MSI-X index too high"));
4854 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4855 			if (mte->mte_handlers == 0)
4856 				return (EINVAL);
4857 			mte->mte_handlers--;
4858 			if (mte->mte_handlers == 0)
4859 				pci_mask_msix(child, rid - 1);
4860 		}
4861 	}
4862 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4863 	if (rid > 0)
4864 		KASSERT(error == 0,
4865 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4866 	return (error);
4867 }
4868 
4869 int
4870 pci_print_child(device_t dev, device_t child)
4871 {
4872 	struct pci_devinfo *dinfo;
4873 	struct resource_list *rl;
4874 	int retval = 0;
4875 
4876 	dinfo = device_get_ivars(child);
4877 	rl = &dinfo->resources;
4878 
4879 	retval += bus_print_child_header(dev, child);
4880 
4881 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4882 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4883 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4884 	if (device_get_flags(dev))
4885 		retval += printf(" flags %#x", device_get_flags(dev));
4886 
4887 	retval += printf(" at device %d.%d", pci_get_slot(child),
4888 	    pci_get_function(child));
4889 
4890 	retval += bus_print_child_domain(dev, child);
4891 	retval += bus_print_child_footer(dev, child);
4892 
4893 	return (retval);
4894 }
4895 
4896 static const struct
4897 {
4898 	int		class;
4899 	int		subclass;
4900 	int		report; /* 0 = bootverbose, 1 = always */
4901 	const char	*desc;
4902 } pci_nomatch_tab[] = {
4903 	{PCIC_OLD,		-1,			1, "old"},
4904 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4905 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4906 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4907 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4908 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4909 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4910 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4911 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4912 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4913 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4914 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4915 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4916 	{PCIC_NETWORK,		-1,			1, "network"},
4917 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4918 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4919 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4920 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4921 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4922 	{PCIC_DISPLAY,		-1,			1, "display"},
4923 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4924 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4925 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4926 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4927 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4928 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4929 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4930 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4931 	{PCIC_MEMORY,		-1,			1, "memory"},
4932 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4933 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4934 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4935 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4936 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4937 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4938 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4939 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4940 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4941 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4942 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4943 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4944 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4945 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4946 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4947 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4948 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4949 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4950 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4951 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4952 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4953 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4954 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4955 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4956 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4957 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4958 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4959 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4960 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4961 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4962 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4963 	{PCIC_DOCKING,		-1,			1, "docking station"},
4964 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4965 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4966 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4967 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4968 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4969 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4970 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4971 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4972 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4973 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4974 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4975 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4976 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4977 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4978 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4979 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4980 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4981 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4982 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4983 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4984 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4985 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4986 	{PCIC_DASP,		-1,			0, "dasp"},
4987 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4988 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4989 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4990 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4991 	{0, 0, 0,		NULL}
4992 };
4993 
4994 void
4995 pci_probe_nomatch(device_t dev, device_t child)
4996 {
4997 	int i, report;
4998 	const char *cp, *scp;
4999 	char *device;
5000 
5001 	/*
5002 	 * Look for a listing for this device in a loaded device database.
5003 	 */
5004 	report = 1;
5005 	if ((device = pci_describe_device(child)) != NULL) {
5006 		device_printf(dev, "<%s>", device);
5007 		free(device, M_DEVBUF);
5008 	} else {
5009 		/*
5010 		 * Scan the class/subclass descriptions for a general
5011 		 * description.
5012 		 */
5013 		cp = "unknown";
5014 		scp = NULL;
5015 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5016 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5017 				if (pci_nomatch_tab[i].subclass == -1) {
5018 					cp = pci_nomatch_tab[i].desc;
5019 					report = pci_nomatch_tab[i].report;
5020 				} else if (pci_nomatch_tab[i].subclass ==
5021 				    pci_get_subclass(child)) {
5022 					scp = pci_nomatch_tab[i].desc;
5023 					report = pci_nomatch_tab[i].report;
5024 				}
5025 			}
5026 		}
5027 		if (report || bootverbose) {
5028 			device_printf(dev, "<%s%s%s>",
5029 			    cp ? cp : "",
5030 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5031 			    scp ? scp : "");
5032 		}
5033 	}
5034 	if (report || bootverbose) {
5035 		printf(" at device %d.%d (no driver attached)\n",
5036 		    pci_get_slot(child), pci_get_function(child));
5037 	}
5038 	pci_cfg_save(child, device_get_ivars(child), 1);
5039 }
5040 
5041 void
5042 pci_child_detached(device_t dev, device_t child)
5043 {
5044 	struct pci_devinfo *dinfo;
5045 	struct resource_list *rl;
5046 
5047 	dinfo = device_get_ivars(child);
5048 	rl = &dinfo->resources;
5049 
5050 	/*
5051 	 * Have to deallocate IRQs before releasing any MSI messages and
5052 	 * have to release MSI messages before deallocating any memory
5053 	 * BARs.
5054 	 */
5055 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5056 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5057 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5058 		if (dinfo->cfg.msi.msi_alloc != 0)
5059 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5060 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5061 		else
5062 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5063 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5064 		(void)pci_release_msi(child);
5065 	}
5066 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5067 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5068 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5069 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5070 #ifdef PCI_RES_BUS
5071 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5072 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5073 #endif
5074 
5075 	pci_cfg_save(child, dinfo, 1);
5076 }
5077 
5078 /*
5079  * Parse the PCI device database, if loaded, and return a pointer to a
5080  * description of the device.
5081  *
5082  * The database is flat text formatted as follows:
5083  *
5084  * Any line not in a valid format is ignored.
5085  * Lines are terminated with newline '\n' characters.
5086  *
5087  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5088  * the vendor name.
5089  *
5090  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5091  * - devices cannot be listed without a corresponding VENDOR line.
5092  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5093  * another TAB, then the device name.
5094  */
5095 
5096 /*
5097  * Assuming (ptr) points to the beginning of a line in the database,
5098  * return the vendor or device and description of the next entry.
5099  * The value of (vendor) or (device) inappropriate for the entry type
5100  * is set to -1.  Returns nonzero at the end of the database.
5101  *
5102  * Note that this is slightly unrobust in the face of corrupt data;
5103  * we attempt to safeguard against this by spamming the end of the
5104  * database with a newline when we initialise.
5105  */
5106 static int
5107 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5108 {
5109 	char	*cp = *ptr;
5110 	int	left;
5111 
5112 	*device = -1;
5113 	*vendor = -1;
5114 	**desc = '\0';
5115 	for (;;) {
5116 		left = pci_vendordata_size - (cp - pci_vendordata);
5117 		if (left <= 0) {
5118 			*ptr = cp;
5119 			return(1);
5120 		}
5121 
5122 		/* vendor entry? */
5123 		if (*cp != '\t' &&
5124 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5125 			break;
5126 		/* device entry? */
5127 		if (*cp == '\t' &&
5128 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5129 			break;
5130 
5131 		/* skip to next line */
5132 		while (*cp != '\n' && left > 0) {
5133 			cp++;
5134 			left--;
5135 		}
5136 		if (*cp == '\n') {
5137 			cp++;
5138 			left--;
5139 		}
5140 	}
5141 	/* skip to next line */
5142 	while (*cp != '\n' && left > 0) {
5143 		cp++;
5144 		left--;
5145 	}
5146 	if (*cp == '\n' && left > 0)
5147 		cp++;
5148 	*ptr = cp;
5149 	return(0);
5150 }
5151 
5152 static char *
5153 pci_describe_device(device_t dev)
5154 {
5155 	int	vendor, device;
5156 	char	*desc, *vp, *dp, *line;
5157 
5158 	desc = vp = dp = NULL;
5159 
5160 	/*
5161 	 * If we have no vendor data, we can't do anything.
5162 	 */
5163 	if (pci_vendordata == NULL)
5164 		goto out;
5165 
5166 	/*
5167 	 * Scan the vendor data looking for this device
5168 	 */
5169 	line = pci_vendordata;
5170 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5171 		goto out;
5172 	for (;;) {
5173 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5174 			goto out;
5175 		if (vendor == pci_get_vendor(dev))
5176 			break;
5177 	}
5178 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5179 		goto out;
5180 	for (;;) {
5181 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5182 			*dp = 0;
5183 			break;
5184 		}
5185 		if (vendor != -1) {
5186 			*dp = 0;
5187 			break;
5188 		}
5189 		if (device == pci_get_device(dev))
5190 			break;
5191 	}
5192 	if (dp[0] == '\0')
5193 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5194 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5195 	    NULL)
5196 		sprintf(desc, "%s, %s", vp, dp);
5197 out:
5198 	if (vp != NULL)
5199 		free(vp, M_DEVBUF);
5200 	if (dp != NULL)
5201 		free(dp, M_DEVBUF);
5202 	return(desc);
5203 }
5204 
5205 int
5206 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5207 {
5208 	struct pci_devinfo *dinfo;
5209 	pcicfgregs *cfg;
5210 
5211 	dinfo = device_get_ivars(child);
5212 	cfg = &dinfo->cfg;
5213 
5214 	switch (which) {
5215 	case PCI_IVAR_ETHADDR:
5216 		/*
5217 		 * The generic accessor doesn't deal with failure, so
5218 		 * we set the return value, then return an error.
5219 		 */
5220 		*((uint8_t **) result) = NULL;
5221 		return (EINVAL);
5222 	case PCI_IVAR_SUBVENDOR:
5223 		*result = cfg->subvendor;
5224 		break;
5225 	case PCI_IVAR_SUBDEVICE:
5226 		*result = cfg->subdevice;
5227 		break;
5228 	case PCI_IVAR_VENDOR:
5229 		*result = cfg->vendor;
5230 		break;
5231 	case PCI_IVAR_DEVICE:
5232 		*result = cfg->device;
5233 		break;
5234 	case PCI_IVAR_DEVID:
5235 		*result = (cfg->device << 16) | cfg->vendor;
5236 		break;
5237 	case PCI_IVAR_CLASS:
5238 		*result = cfg->baseclass;
5239 		break;
5240 	case PCI_IVAR_SUBCLASS:
5241 		*result = cfg->subclass;
5242 		break;
5243 	case PCI_IVAR_PROGIF:
5244 		*result = cfg->progif;
5245 		break;
5246 	case PCI_IVAR_REVID:
5247 		*result = cfg->revid;
5248 		break;
5249 	case PCI_IVAR_INTPIN:
5250 		*result = cfg->intpin;
5251 		break;
5252 	case PCI_IVAR_IRQ:
5253 		*result = cfg->intline;
5254 		break;
5255 	case PCI_IVAR_DOMAIN:
5256 		*result = cfg->domain;
5257 		break;
5258 	case PCI_IVAR_BUS:
5259 		*result = cfg->bus;
5260 		break;
5261 	case PCI_IVAR_SLOT:
5262 		*result = cfg->slot;
5263 		break;
5264 	case PCI_IVAR_FUNCTION:
5265 		*result = cfg->func;
5266 		break;
5267 	case PCI_IVAR_CMDREG:
5268 		*result = cfg->cmdreg;
5269 		break;
5270 	case PCI_IVAR_CACHELNSZ:
5271 		*result = cfg->cachelnsz;
5272 		break;
5273 	case PCI_IVAR_MINGNT:
5274 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5275 			*result = -1;
5276 			return (EINVAL);
5277 		}
5278 		*result = cfg->mingnt;
5279 		break;
5280 	case PCI_IVAR_MAXLAT:
5281 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5282 			*result = -1;
5283 			return (EINVAL);
5284 		}
5285 		*result = cfg->maxlat;
5286 		break;
5287 	case PCI_IVAR_LATTIMER:
5288 		*result = cfg->lattimer;
5289 		break;
5290 	default:
5291 		return (ENOENT);
5292 	}
5293 	return (0);
5294 }
5295 
5296 int
5297 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5298 {
5299 	struct pci_devinfo *dinfo;
5300 
5301 	dinfo = device_get_ivars(child);
5302 
5303 	switch (which) {
5304 	case PCI_IVAR_INTPIN:
5305 		dinfo->cfg.intpin = value;
5306 		return (0);
5307 	case PCI_IVAR_ETHADDR:
5308 	case PCI_IVAR_SUBVENDOR:
5309 	case PCI_IVAR_SUBDEVICE:
5310 	case PCI_IVAR_VENDOR:
5311 	case PCI_IVAR_DEVICE:
5312 	case PCI_IVAR_DEVID:
5313 	case PCI_IVAR_CLASS:
5314 	case PCI_IVAR_SUBCLASS:
5315 	case PCI_IVAR_PROGIF:
5316 	case PCI_IVAR_REVID:
5317 	case PCI_IVAR_IRQ:
5318 	case PCI_IVAR_DOMAIN:
5319 	case PCI_IVAR_BUS:
5320 	case PCI_IVAR_SLOT:
5321 	case PCI_IVAR_FUNCTION:
5322 		return (EINVAL);	/* disallow for now */
5323 
5324 	default:
5325 		return (ENOENT);
5326 	}
5327 }
5328 
5329 #include "opt_ddb.h"
5330 #ifdef DDB
5331 #include <ddb/ddb.h>
5332 #include <sys/cons.h>
5333 
5334 /*
5335  * List resources based on pci map registers, used for within ddb
5336  */
5337 
5338 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5339 {
5340 	struct pci_devinfo *dinfo;
5341 	struct devlist *devlist_head;
5342 	struct pci_conf *p;
5343 	const char *name;
5344 	int i, error, none_count;
5345 
5346 	none_count = 0;
5347 	/* get the head of the device queue */
5348 	devlist_head = &pci_devq;
5349 
5350 	/*
5351 	 * Go through the list of devices and print out devices
5352 	 */
5353 	for (error = 0, i = 0,
5354 	     dinfo = STAILQ_FIRST(devlist_head);
5355 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5356 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5357 		/* Populate pd_name and pd_unit */
5358 		name = NULL;
5359 		if (dinfo->cfg.dev)
5360 			name = device_get_name(dinfo->cfg.dev);
5361 
5362 		p = &dinfo->conf;
5363 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5364 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5365 			(name && *name) ? name : "none",
5366 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5367 			none_count++,
5368 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5369 			p->pc_sel.pc_func, (p->pc_class << 16) |
5370 			(p->pc_subclass << 8) | p->pc_progif,
5371 			(p->pc_subdevice << 16) | p->pc_subvendor,
5372 			(p->pc_device << 16) | p->pc_vendor,
5373 			p->pc_revid, p->pc_hdr);
5374 	}
5375 }
5376 #endif /* DDB */
5377 
5378 struct resource *
5379 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5380     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5381     u_int flags)
5382 {
5383 	struct pci_devinfo *dinfo = device_get_ivars(child);
5384 	struct resource_list *rl = &dinfo->resources;
5385 	struct resource *res;
5386 	struct pci_map *pm;
5387 	uint16_t cmd;
5388 	pci_addr_t map, testval;
5389 	int mapsize;
5390 
5391 	res = NULL;
5392 
5393 	/* If rid is managed by EA, ignore it */
5394 	if (pci_ea_is_enabled(child, *rid))
5395 		goto out;
5396 
5397 	pm = pci_find_bar(child, *rid);
5398 	if (pm != NULL) {
5399 		/* This is a BAR that we failed to allocate earlier. */
5400 		mapsize = pm->pm_size;
5401 		map = pm->pm_value;
5402 	} else {
5403 		/*
5404 		 * Weed out the bogons, and figure out how large the
5405 		 * BAR/map is.  BARs that read back 0 here are bogus
5406 		 * and unimplemented.  Note: atapci in legacy mode are
5407 		 * special and handled elsewhere in the code.  If you
5408 		 * have a atapci device in legacy mode and it fails
5409 		 * here, that other code is broken.
5410 		 */
5411 		pci_read_bar(child, *rid, &map, &testval, NULL);
5412 
5413 		/*
5414 		 * Determine the size of the BAR and ignore BARs with a size
5415 		 * of 0.  Device ROM BARs use a different mask value.
5416 		 */
5417 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5418 			mapsize = pci_romsize(testval);
5419 		else
5420 			mapsize = pci_mapsize(testval);
5421 		if (mapsize == 0)
5422 			goto out;
5423 		pm = pci_add_bar(child, *rid, map, mapsize);
5424 	}
5425 
5426 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5427 		if (type != SYS_RES_MEMORY) {
5428 			if (bootverbose)
5429 				device_printf(dev,
5430 				    "child %s requested type %d for rid %#x,"
5431 				    " but the BAR says it is an memio\n",
5432 				    device_get_nameunit(child), type, *rid);
5433 			goto out;
5434 		}
5435 	} else {
5436 		if (type != SYS_RES_IOPORT) {
5437 			if (bootverbose)
5438 				device_printf(dev,
5439 				    "child %s requested type %d for rid %#x,"
5440 				    " but the BAR says it is an ioport\n",
5441 				    device_get_nameunit(child), type, *rid);
5442 			goto out;
5443 		}
5444 	}
5445 
5446 	/*
5447 	 * For real BARs, we need to override the size that
5448 	 * the driver requests, because that's what the BAR
5449 	 * actually uses and we would otherwise have a
5450 	 * situation where we might allocate the excess to
5451 	 * another driver, which won't work.
5452 	 */
5453 	count = ((pci_addr_t)1 << mapsize) * num;
5454 	if (RF_ALIGNMENT(flags) < mapsize)
5455 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5456 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5457 		flags |= RF_PREFETCHABLE;
5458 
5459 	/*
5460 	 * Allocate enough resource, and then write back the
5461 	 * appropriate BAR for that resource.
5462 	 */
5463 	resource_list_add(rl, type, *rid, start, end, count);
5464 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5465 	    count, flags & ~RF_ACTIVE);
5466 	if (res == NULL) {
5467 		resource_list_delete(rl, type, *rid);
5468 		device_printf(child,
5469 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5470 		    count, *rid, type, start, end);
5471 		goto out;
5472 	}
5473 	if (bootverbose)
5474 		device_printf(child,
5475 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5476 		    count, *rid, type, rman_get_start(res));
5477 
5478 	/* Disable decoding via the CMD register before updating the BAR */
5479 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5480 	pci_write_config(child, PCIR_COMMAND,
5481 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5482 
5483 	map = rman_get_start(res);
5484 	pci_write_bar(child, pm, map);
5485 
5486 	/* Restore the original value of the CMD register */
5487 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5488 out:
5489 	return (res);
5490 }
5491 
5492 struct resource *
5493 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5494     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5495     u_int flags)
5496 {
5497 	struct pci_devinfo *dinfo;
5498 	struct resource_list *rl;
5499 	struct resource_list_entry *rle;
5500 	struct resource *res;
5501 	pcicfgregs *cfg;
5502 
5503 	/*
5504 	 * Perform lazy resource allocation
5505 	 */
5506 	dinfo = device_get_ivars(child);
5507 	rl = &dinfo->resources;
5508 	cfg = &dinfo->cfg;
5509 	switch (type) {
5510 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5511 	case PCI_RES_BUS:
5512 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5513 		    flags));
5514 #endif
5515 	case SYS_RES_IRQ:
5516 		/*
5517 		 * Can't alloc legacy interrupt once MSI messages have
5518 		 * been allocated.
5519 		 */
5520 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5521 		    cfg->msix.msix_alloc > 0))
5522 			return (NULL);
5523 
5524 		/*
5525 		 * If the child device doesn't have an interrupt
5526 		 * routed and is deserving of an interrupt, try to
5527 		 * assign it one.
5528 		 */
5529 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5530 		    (cfg->intpin != 0))
5531 			pci_assign_interrupt(dev, child, 0);
5532 		break;
5533 	case SYS_RES_IOPORT:
5534 	case SYS_RES_MEMORY:
5535 #ifdef NEW_PCIB
5536 		/*
5537 		 * PCI-PCI bridge I/O window resources are not BARs.
5538 		 * For those allocations just pass the request up the
5539 		 * tree.
5540 		 */
5541 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5542 			switch (*rid) {
5543 			case PCIR_IOBASEL_1:
5544 			case PCIR_MEMBASE_1:
5545 			case PCIR_PMBASEL_1:
5546 				/*
5547 				 * XXX: Should we bother creating a resource
5548 				 * list entry?
5549 				 */
5550 				return (bus_generic_alloc_resource(dev, child,
5551 				    type, rid, start, end, count, flags));
5552 			}
5553 		}
5554 #endif
5555 		/* Reserve resources for this BAR if needed. */
5556 		rle = resource_list_find(rl, type, *rid);
5557 		if (rle == NULL) {
5558 			res = pci_reserve_map(dev, child, type, rid, start, end,
5559 			    count, num, flags);
5560 			if (res == NULL)
5561 				return (NULL);
5562 		}
5563 	}
5564 	return (resource_list_alloc(rl, dev, child, type, rid,
5565 	    start, end, count, flags));
5566 }
5567 
5568 struct resource *
5569 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5570     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5571 {
5572 #ifdef PCI_IOV
5573 	struct pci_devinfo *dinfo;
5574 #endif
5575 
5576 	if (device_get_parent(child) != dev)
5577 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5578 		    type, rid, start, end, count, flags));
5579 
5580 #ifdef PCI_IOV
5581 	dinfo = device_get_ivars(child);
5582 	if (dinfo->cfg.flags & PCICFG_VF) {
5583 		switch (type) {
5584 		/* VFs can't have I/O BARs. */
5585 		case SYS_RES_IOPORT:
5586 			return (NULL);
5587 		case SYS_RES_MEMORY:
5588 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5589 			    start, end, count, flags));
5590 		}
5591 
5592 		/* Fall through for other types of resource allocations. */
5593 	}
5594 #endif
5595 
5596 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5597 	    count, 1, flags));
5598 }
5599 
5600 int
5601 pci_release_resource(device_t dev, device_t child, int type, int rid,
5602     struct resource *r)
5603 {
5604 	struct pci_devinfo *dinfo;
5605 	struct resource_list *rl;
5606 	pcicfgregs *cfg;
5607 
5608 	if (device_get_parent(child) != dev)
5609 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5610 		    type, rid, r));
5611 
5612 	dinfo = device_get_ivars(child);
5613 	cfg = &dinfo->cfg;
5614 
5615 #ifdef PCI_IOV
5616 	if (dinfo->cfg.flags & PCICFG_VF) {
5617 		switch (type) {
5618 		/* VFs can't have I/O BARs. */
5619 		case SYS_RES_IOPORT:
5620 			return (EDOOFUS);
5621 		case SYS_RES_MEMORY:
5622 			return (pci_vf_release_mem_resource(dev, child, rid,
5623 			    r));
5624 		}
5625 
5626 		/* Fall through for other types of resource allocations. */
5627 	}
5628 #endif
5629 
5630 #ifdef NEW_PCIB
5631 	/*
5632 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5633 	 * those allocations just pass the request up the tree.
5634 	 */
5635 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5636 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5637 		switch (rid) {
5638 		case PCIR_IOBASEL_1:
5639 		case PCIR_MEMBASE_1:
5640 		case PCIR_PMBASEL_1:
5641 			return (bus_generic_release_resource(dev, child, type,
5642 			    rid, r));
5643 		}
5644 	}
5645 #endif
5646 
5647 	rl = &dinfo->resources;
5648 	return (resource_list_release(rl, dev, child, type, rid, r));
5649 }
5650 
5651 int
5652 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5653     struct resource *r)
5654 {
5655 	struct pci_devinfo *dinfo;
5656 	int error;
5657 
5658 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5659 	if (error)
5660 		return (error);
5661 
5662 	/* Enable decoding in the command register when activating BARs. */
5663 	if (device_get_parent(child) == dev) {
5664 		/* Device ROMs need their decoding explicitly enabled. */
5665 		dinfo = device_get_ivars(child);
5666 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5667 			pci_write_bar(child, pci_find_bar(child, rid),
5668 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5669 		switch (type) {
5670 		case SYS_RES_IOPORT:
5671 		case SYS_RES_MEMORY:
5672 			error = PCI_ENABLE_IO(dev, child, type);
5673 			break;
5674 		}
5675 	}
5676 	return (error);
5677 }
5678 
5679 int
5680 pci_deactivate_resource(device_t dev, device_t child, int type,
5681     int rid, struct resource *r)
5682 {
5683 	struct pci_devinfo *dinfo;
5684 	int error;
5685 
5686 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5687 	if (error)
5688 		return (error);
5689 
5690 	/* Disable decoding for device ROMs. */
5691 	if (device_get_parent(child) == dev) {
5692 		dinfo = device_get_ivars(child);
5693 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5694 			pci_write_bar(child, pci_find_bar(child, rid),
5695 			    rman_get_start(r));
5696 	}
5697 	return (0);
5698 }
5699 
5700 void
5701 pci_child_deleted(device_t dev, device_t child)
5702 {
5703 	struct resource_list_entry *rle;
5704 	struct resource_list *rl;
5705 	struct pci_devinfo *dinfo;
5706 
5707 	dinfo = device_get_ivars(child);
5708 	rl = &dinfo->resources;
5709 
5710 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5711 
5712 	/* Turn off access to resources we're about to free */
5713 	if (bus_child_present(child) != 0) {
5714 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5715 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5716 
5717 		pci_disable_busmaster(child);
5718 	}
5719 
5720 	/* Free all allocated resources */
5721 	STAILQ_FOREACH(rle, rl, link) {
5722 		if (rle->res) {
5723 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5724 			    resource_list_busy(rl, rle->type, rle->rid)) {
5725 				pci_printf(&dinfo->cfg,
5726 				    "Resource still owned, oops. "
5727 				    "(type=%d, rid=%d, addr=%lx)\n",
5728 				    rle->type, rle->rid,
5729 				    rman_get_start(rle->res));
5730 				bus_release_resource(child, rle->type, rle->rid,
5731 				    rle->res);
5732 			}
5733 			resource_list_unreserve(rl, dev, child, rle->type,
5734 			    rle->rid);
5735 		}
5736 	}
5737 	resource_list_free(rl);
5738 
5739 	pci_freecfg(dinfo);
5740 }
5741 
5742 void
5743 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5744 {
5745 	struct pci_devinfo *dinfo;
5746 	struct resource_list *rl;
5747 	struct resource_list_entry *rle;
5748 
5749 	if (device_get_parent(child) != dev)
5750 		return;
5751 
5752 	dinfo = device_get_ivars(child);
5753 	rl = &dinfo->resources;
5754 	rle = resource_list_find(rl, type, rid);
5755 	if (rle == NULL)
5756 		return;
5757 
5758 	if (rle->res) {
5759 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5760 		    resource_list_busy(rl, type, rid)) {
5761 			device_printf(dev, "delete_resource: "
5762 			    "Resource still owned by child, oops. "
5763 			    "(type=%d, rid=%d, addr=%jx)\n",
5764 			    type, rid, rman_get_start(rle->res));
5765 			return;
5766 		}
5767 		resource_list_unreserve(rl, dev, child, type, rid);
5768 	}
5769 	resource_list_delete(rl, type, rid);
5770 }
5771 
5772 struct resource_list *
5773 pci_get_resource_list (device_t dev, device_t child)
5774 {
5775 	struct pci_devinfo *dinfo = device_get_ivars(child);
5776 
5777 	return (&dinfo->resources);
5778 }
5779 
5780 #ifdef IOMMU
5781 bus_dma_tag_t
5782 pci_get_dma_tag(device_t bus, device_t dev)
5783 {
5784 	bus_dma_tag_t tag;
5785 	struct pci_softc *sc;
5786 
5787 	if (device_get_parent(dev) == bus) {
5788 		/* try iommu and return if it works */
5789 		tag = iommu_get_dma_tag(bus, dev);
5790 	} else
5791 		tag = NULL;
5792 	if (tag == NULL) {
5793 		sc = device_get_softc(bus);
5794 		tag = sc->sc_dma_tag;
5795 	}
5796 	return (tag);
5797 }
5798 #else
5799 bus_dma_tag_t
5800 pci_get_dma_tag(device_t bus, device_t dev)
5801 {
5802 	struct pci_softc *sc = device_get_softc(bus);
5803 
5804 	return (sc->sc_dma_tag);
5805 }
5806 #endif
5807 
5808 uint32_t
5809 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5810 {
5811 	struct pci_devinfo *dinfo = device_get_ivars(child);
5812 	pcicfgregs *cfg = &dinfo->cfg;
5813 
5814 #ifdef PCI_IOV
5815 	/*
5816 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5817 	 * emulate them here.
5818 	 */
5819 	if (cfg->flags & PCICFG_VF) {
5820 		if (reg == PCIR_VENDOR) {
5821 			switch (width) {
5822 			case 4:
5823 				return (cfg->device << 16 | cfg->vendor);
5824 			case 2:
5825 				return (cfg->vendor);
5826 			case 1:
5827 				return (cfg->vendor & 0xff);
5828 			default:
5829 				return (0xffffffff);
5830 			}
5831 		} else if (reg == PCIR_DEVICE) {
5832 			switch (width) {
5833 			/* Note that an unaligned 4-byte read is an error. */
5834 			case 2:
5835 				return (cfg->device);
5836 			case 1:
5837 				return (cfg->device & 0xff);
5838 			default:
5839 				return (0xffffffff);
5840 			}
5841 		}
5842 	}
5843 #endif
5844 
5845 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5846 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5847 }
5848 
5849 void
5850 pci_write_config_method(device_t dev, device_t child, int reg,
5851     uint32_t val, int width)
5852 {
5853 	struct pci_devinfo *dinfo = device_get_ivars(child);
5854 	pcicfgregs *cfg = &dinfo->cfg;
5855 
5856 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5857 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5858 }
5859 
5860 int
5861 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
5862 {
5863 
5864 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5865 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5866 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5867 	return (0);
5868 }
5869 
5870 int
5871 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
5872 {
5873 	struct pci_devinfo *dinfo;
5874 	pcicfgregs *cfg;
5875 
5876 	dinfo = device_get_ivars(child);
5877 	cfg = &dinfo->cfg;
5878 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5879 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5880 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5881 	    cfg->progif);
5882 	return (0);
5883 }
5884 
5885 int
5886 pci_assign_interrupt_method(device_t dev, device_t child)
5887 {
5888 	struct pci_devinfo *dinfo = device_get_ivars(child);
5889 	pcicfgregs *cfg = &dinfo->cfg;
5890 
5891 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5892 	    cfg->intpin));
5893 }
5894 
5895 static void
5896 pci_lookup(void *arg, const char *name, device_t *dev)
5897 {
5898 	long val;
5899 	char *end;
5900 	int domain, bus, slot, func;
5901 
5902 	if (*dev != NULL)
5903 		return;
5904 
5905 	/*
5906 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5907 	 * pciB:S:F.  In the latter case, the domain is assumed to
5908 	 * be zero.
5909 	 */
5910 	if (strncmp(name, "pci", 3) != 0)
5911 		return;
5912 	val = strtol(name + 3, &end, 10);
5913 	if (val < 0 || val > INT_MAX || *end != ':')
5914 		return;
5915 	domain = val;
5916 	val = strtol(end + 1, &end, 10);
5917 	if (val < 0 || val > INT_MAX || *end != ':')
5918 		return;
5919 	bus = val;
5920 	val = strtol(end + 1, &end, 10);
5921 	if (val < 0 || val > INT_MAX)
5922 		return;
5923 	slot = val;
5924 	if (*end == ':') {
5925 		val = strtol(end + 1, &end, 10);
5926 		if (val < 0 || val > INT_MAX || *end != '\0')
5927 			return;
5928 		func = val;
5929 	} else if (*end == '\0') {
5930 		func = slot;
5931 		slot = bus;
5932 		bus = domain;
5933 		domain = 0;
5934 	} else
5935 		return;
5936 
5937 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5938 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5939 		return;
5940 
5941 	*dev = pci_find_dbsf(domain, bus, slot, func);
5942 }
5943 
5944 static int
5945 pci_modevent(module_t mod, int what, void *arg)
5946 {
5947 	static struct cdev *pci_cdev;
5948 	static eventhandler_tag tag;
5949 
5950 	switch (what) {
5951 	case MOD_LOAD:
5952 		STAILQ_INIT(&pci_devq);
5953 		pci_generation = 0;
5954 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5955 		    "pci");
5956 		pci_load_vendor_data();
5957 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5958 		    1000);
5959 		break;
5960 
5961 	case MOD_UNLOAD:
5962 		if (tag != NULL)
5963 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5964 		destroy_dev(pci_cdev);
5965 		break;
5966 	}
5967 
5968 	return (0);
5969 }
5970 
5971 static void
5972 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5973 {
5974 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5975 	struct pcicfg_pcie *cfg;
5976 	int version, pos;
5977 
5978 	cfg = &dinfo->cfg.pcie;
5979 	pos = cfg->pcie_location;
5980 
5981 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5982 
5983 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5984 
5985 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5986 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5987 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5988 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5989 
5990 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5991 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5992 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5993 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5994 
5995 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5996 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5997 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5998 
5999 	if (version > 1) {
6000 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6001 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6002 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6003 	}
6004 #undef WREG
6005 }
6006 
6007 static void
6008 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6009 {
6010 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6011 	    dinfo->cfg.pcix.pcix_command,  2);
6012 }
6013 
6014 void
6015 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6016 {
6017 
6018 	/*
6019 	 * Restore the device to full power mode.  We must do this
6020 	 * before we restore the registers because moving from D3 to
6021 	 * D0 will cause the chip's BARs and some other registers to
6022 	 * be reset to some unknown power on reset values.  Cut down
6023 	 * the noise on boot by doing nothing if we are already in
6024 	 * state D0.
6025 	 */
6026 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6027 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6028 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6029 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6030 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6031 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6032 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6033 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6034 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6035 	case PCIM_HDRTYPE_NORMAL:
6036 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6037 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6038 		break;
6039 	case PCIM_HDRTYPE_BRIDGE:
6040 		pci_write_config(dev, PCIR_SECLAT_1,
6041 		    dinfo->cfg.bridge.br_seclat, 1);
6042 		pci_write_config(dev, PCIR_SUBBUS_1,
6043 		    dinfo->cfg.bridge.br_subbus, 1);
6044 		pci_write_config(dev, PCIR_SECBUS_1,
6045 		    dinfo->cfg.bridge.br_secbus, 1);
6046 		pci_write_config(dev, PCIR_PRIBUS_1,
6047 		    dinfo->cfg.bridge.br_pribus, 1);
6048 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6049 		    dinfo->cfg.bridge.br_control, 2);
6050 		break;
6051 	case PCIM_HDRTYPE_CARDBUS:
6052 		pci_write_config(dev, PCIR_SECLAT_2,
6053 		    dinfo->cfg.bridge.br_seclat, 1);
6054 		pci_write_config(dev, PCIR_SUBBUS_2,
6055 		    dinfo->cfg.bridge.br_subbus, 1);
6056 		pci_write_config(dev, PCIR_SECBUS_2,
6057 		    dinfo->cfg.bridge.br_secbus, 1);
6058 		pci_write_config(dev, PCIR_PRIBUS_2,
6059 		    dinfo->cfg.bridge.br_pribus, 1);
6060 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6061 		    dinfo->cfg.bridge.br_control, 2);
6062 		break;
6063 	}
6064 	pci_restore_bars(dev);
6065 
6066 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6067 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6068 
6069 	/*
6070 	 * Restore extended capabilities for PCI-Express and PCI-X
6071 	 */
6072 	if (dinfo->cfg.pcie.pcie_location != 0)
6073 		pci_cfg_restore_pcie(dev, dinfo);
6074 	if (dinfo->cfg.pcix.pcix_location != 0)
6075 		pci_cfg_restore_pcix(dev, dinfo);
6076 
6077 	/* Restore MSI and MSI-X configurations if they are present. */
6078 	if (dinfo->cfg.msi.msi_location != 0)
6079 		pci_resume_msi(dev);
6080 	if (dinfo->cfg.msix.msix_location != 0)
6081 		pci_resume_msix(dev);
6082 
6083 #ifdef PCI_IOV
6084 	if (dinfo->cfg.iov != NULL)
6085 		pci_iov_cfg_restore(dev, dinfo);
6086 #endif
6087 }
6088 
6089 static void
6090 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6091 {
6092 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6093 	struct pcicfg_pcie *cfg;
6094 	int version, pos;
6095 
6096 	cfg = &dinfo->cfg.pcie;
6097 	pos = cfg->pcie_location;
6098 
6099 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6100 
6101 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6102 
6103 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6104 
6105 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6106 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6107 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6108 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6109 
6110 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6111 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6112 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6113 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6114 
6115 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6116 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6117 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6118 
6119 	if (version > 1) {
6120 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6121 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6122 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6123 	}
6124 #undef RREG
6125 }
6126 
6127 static void
6128 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6129 {
6130 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6131 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6132 }
6133 
6134 void
6135 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6136 {
6137 	uint32_t cls;
6138 	int ps;
6139 
6140 	/*
6141 	 * Some drivers apparently write to these registers w/o updating our
6142 	 * cached copy.  No harm happens if we update the copy, so do so here
6143 	 * so we can restore them.  The COMMAND register is modified by the
6144 	 * bus w/o updating the cache.  This should represent the normally
6145 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6146 	 */
6147 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6148 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6149 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6150 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6151 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6152 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6153 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6154 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6155 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6156 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6157 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6158 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6159 	case PCIM_HDRTYPE_NORMAL:
6160 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6161 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6162 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6163 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6164 		break;
6165 	case PCIM_HDRTYPE_BRIDGE:
6166 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6167 		    PCIR_SECLAT_1, 1);
6168 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6169 		    PCIR_SUBBUS_1, 1);
6170 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6171 		    PCIR_SECBUS_1, 1);
6172 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6173 		    PCIR_PRIBUS_1, 1);
6174 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6175 		    PCIR_BRIDGECTL_1, 2);
6176 		break;
6177 	case PCIM_HDRTYPE_CARDBUS:
6178 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6179 		    PCIR_SECLAT_2, 1);
6180 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6181 		    PCIR_SUBBUS_2, 1);
6182 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6183 		    PCIR_SECBUS_2, 1);
6184 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6185 		    PCIR_PRIBUS_2, 1);
6186 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6187 		    PCIR_BRIDGECTL_2, 2);
6188 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6189 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6190 		break;
6191 	}
6192 
6193 	if (dinfo->cfg.pcie.pcie_location != 0)
6194 		pci_cfg_save_pcie(dev, dinfo);
6195 
6196 	if (dinfo->cfg.pcix.pcix_location != 0)
6197 		pci_cfg_save_pcix(dev, dinfo);
6198 
6199 #ifdef PCI_IOV
6200 	if (dinfo->cfg.iov != NULL)
6201 		pci_iov_cfg_save(dev, dinfo);
6202 #endif
6203 
6204 	/*
6205 	 * don't set the state for display devices, base peripherals and
6206 	 * memory devices since bad things happen when they are powered down.
6207 	 * We should (a) have drivers that can easily detach and (b) use
6208 	 * generic drivers for these devices so that some device actually
6209 	 * attaches.  We need to make sure that when we implement (a) we don't
6210 	 * power the device down on a reattach.
6211 	 */
6212 	cls = pci_get_class(dev);
6213 	if (!setstate)
6214 		return;
6215 	switch (pci_do_power_nodriver)
6216 	{
6217 		case 0:		/* NO powerdown at all */
6218 			return;
6219 		case 1:		/* Conservative about what to power down */
6220 			if (cls == PCIC_STORAGE)
6221 				return;
6222 			/*FALLTHROUGH*/
6223 		case 2:		/* Aggressive about what to power down */
6224 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6225 			    cls == PCIC_BASEPERIPH)
6226 				return;
6227 			/*FALLTHROUGH*/
6228 		case 3:		/* Power down everything */
6229 			break;
6230 	}
6231 	/*
6232 	 * PCI spec says we can only go into D3 state from D0 state.
6233 	 * Transition from D[12] into D0 before going to D3 state.
6234 	 */
6235 	ps = pci_get_powerstate(dev);
6236 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6237 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6238 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6239 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6240 }
6241 
6242 /* Wrapper APIs suitable for device driver use. */
6243 void
6244 pci_save_state(device_t dev)
6245 {
6246 	struct pci_devinfo *dinfo;
6247 
6248 	dinfo = device_get_ivars(dev);
6249 	pci_cfg_save(dev, dinfo, 0);
6250 }
6251 
6252 void
6253 pci_restore_state(device_t dev)
6254 {
6255 	struct pci_devinfo *dinfo;
6256 
6257 	dinfo = device_get_ivars(dev);
6258 	pci_cfg_restore(dev, dinfo);
6259 }
6260 
6261 static int
6262 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6263     uintptr_t *id)
6264 {
6265 
6266 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6267 }
6268 
6269 /* Find the upstream port of a given PCI device in a root complex. */
6270 device_t
6271 pci_find_pcie_root_port(device_t dev)
6272 {
6273 	struct pci_devinfo *dinfo;
6274 	devclass_t pci_class;
6275 	device_t pcib, bus;
6276 
6277 	pci_class = devclass_find("pci");
6278 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6279 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6280 
6281 	/*
6282 	 * Walk the bridge hierarchy until we find a PCI-e root
6283 	 * port or a non-PCI device.
6284 	 */
6285 	for (;;) {
6286 		bus = device_get_parent(dev);
6287 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6288 		    device_get_nameunit(dev)));
6289 
6290 		pcib = device_get_parent(bus);
6291 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6292 		    device_get_nameunit(bus)));
6293 
6294 		/*
6295 		 * pcib's parent must be a PCI bus for this to be a
6296 		 * PCI-PCI bridge.
6297 		 */
6298 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6299 			return (NULL);
6300 
6301 		dinfo = device_get_ivars(pcib);
6302 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6303 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6304 			return (pcib);
6305 
6306 		dev = pcib;
6307 	}
6308 }
6309 
6310 /*
6311  * Wait for pending transactions to complete on a PCI-express function.
6312  *
6313  * The maximum delay is specified in milliseconds in max_delay.  Note
6314  * that this function may sleep.
6315  *
6316  * Returns true if the function is idle and false if the timeout is
6317  * exceeded.  If dev is not a PCI-express function, this returns true.
6318  */
6319 bool
6320 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6321 {
6322 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6323 	uint16_t sta;
6324 	int cap;
6325 
6326 	cap = dinfo->cfg.pcie.pcie_location;
6327 	if (cap == 0)
6328 		return (true);
6329 
6330 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6331 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6332 		if (max_delay == 0)
6333 			return (false);
6334 
6335 		/* Poll once every 100 milliseconds up to the timeout. */
6336 		if (max_delay > 100) {
6337 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6338 			max_delay -= 100;
6339 		} else {
6340 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6341 			    C_HARDCLOCK);
6342 			max_delay = 0;
6343 		}
6344 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6345 	}
6346 
6347 	return (true);
6348 }
6349 
6350 /*
6351  * Determine the maximum Completion Timeout in microseconds.
6352  *
6353  * For non-PCI-express functions this returns 0.
6354  */
6355 int
6356 pcie_get_max_completion_timeout(device_t dev)
6357 {
6358 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6359 	int cap;
6360 
6361 	cap = dinfo->cfg.pcie.pcie_location;
6362 	if (cap == 0)
6363 		return (0);
6364 
6365 	/*
6366 	 * Functions using the 1.x spec use the default timeout range of
6367 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6368 	 * support programmable timeouts also use this range.
6369 	 */
6370 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6371 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6372 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6373 		return (50 * 1000);
6374 
6375 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6376 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6377 	case PCIEM_CTL2_COMP_TIMO_100US:
6378 		return (100);
6379 	case PCIEM_CTL2_COMP_TIMO_10MS:
6380 		return (10 * 1000);
6381 	case PCIEM_CTL2_COMP_TIMO_55MS:
6382 		return (55 * 1000);
6383 	case PCIEM_CTL2_COMP_TIMO_210MS:
6384 		return (210 * 1000);
6385 	case PCIEM_CTL2_COMP_TIMO_900MS:
6386 		return (900 * 1000);
6387 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6388 		return (3500 * 1000);
6389 	case PCIEM_CTL2_COMP_TIMO_13S:
6390 		return (13 * 1000 * 1000);
6391 	case PCIEM_CTL2_COMP_TIMO_64S:
6392 		return (64 * 1000 * 1000);
6393 	default:
6394 		return (50 * 1000);
6395 	}
6396 }
6397 
6398 void
6399 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6400 {
6401 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6402 	const char *s;
6403 	int aer;
6404 	uint32_t r, r1;
6405 	uint16_t rs;
6406 
6407 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6408 		s = "Correctable";
6409 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6410 		s = "Uncorrectable (Non-Fatal)";
6411 	else
6412 		s = "Uncorrectable (Fatal)";
6413 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6414 	if (aerp) {
6415 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6416 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6417 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6418 		} else {
6419 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6420 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6421 		}
6422 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6423 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6424 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6425 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6426 			printf(" severity 0x%08x first %d\n",
6427 			    r, rs & 0x1f);
6428 		} else
6429 			printf("\n");
6430 	}
6431 
6432 	/* As kind of recovery just report and clear the error statuses. */
6433 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6434 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6435 		if (r != 0) {
6436 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6437 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6438 		}
6439 
6440 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6441 		if (r != 0) {
6442 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6443 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6444 		}
6445 	}
6446 	if (dinfo->cfg.pcie.pcie_location != 0) {
6447 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6448 		    PCIER_DEVICE_STA, 2);
6449 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6450 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6451 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6452 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6453 			    PCIER_DEVICE_STA, rs, 2);
6454 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6455 		}
6456 	}
6457 }
6458 
6459 /*
6460  * Perform a Function Level Reset (FLR) on a device.
6461  *
6462  * This function first waits for any pending transactions to complete
6463  * within the timeout specified by max_delay.  If transactions are
6464  * still pending, the function will return false without attempting a
6465  * reset.
6466  *
6467  * If dev is not a PCI-express function or does not support FLR, this
6468  * function returns false.
6469  *
6470  * Note that no registers are saved or restored.  The caller is
6471  * responsible for saving and restoring any registers including
6472  * PCI-standard registers via pci_save_state() and
6473  * pci_restore_state().
6474  */
6475 bool
6476 pcie_flr(device_t dev, u_int max_delay, bool force)
6477 {
6478 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6479 	uint16_t cmd, ctl;
6480 	int compl_delay;
6481 	int cap;
6482 
6483 	cap = dinfo->cfg.pcie.pcie_location;
6484 	if (cap == 0)
6485 		return (false);
6486 
6487 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6488 		return (false);
6489 
6490 	/*
6491 	 * Disable busmastering to prevent generation of new
6492 	 * transactions while waiting for the device to go idle.  If
6493 	 * the idle timeout fails, the command register is restored
6494 	 * which will re-enable busmastering.
6495 	 */
6496 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6497 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6498 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6499 		if (!force) {
6500 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6501 			return (false);
6502 		}
6503 		pci_printf(&dinfo->cfg,
6504 		    "Resetting with transactions pending after %d ms\n",
6505 		    max_delay);
6506 
6507 		/*
6508 		 * Extend the post-FLR delay to cover the maximum
6509 		 * Completion Timeout delay of anything in flight
6510 		 * during the FLR delay.  Enforce a minimum delay of
6511 		 * at least 10ms.
6512 		 */
6513 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6514 		if (compl_delay < 10)
6515 			compl_delay = 10;
6516 	} else
6517 		compl_delay = 0;
6518 
6519 	/* Initiate the reset. */
6520 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6521 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6522 	    PCIEM_CTL_INITIATE_FLR, 2);
6523 
6524 	/* Wait for 100ms. */
6525 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6526 
6527 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6528 	    PCIEM_STA_TRANSACTION_PND)
6529 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6530 	return (true);
6531 }
6532 
6533 /*
6534  * Attempt a power-management reset by cycling the device in/out of D3
6535  * state.  PCI spec says we can only go into D3 state from D0 state.
6536  * Transition from D[12] into D0 before going to D3 state.
6537  */
6538 int
6539 pci_power_reset(device_t dev)
6540 {
6541 	int ps;
6542 
6543 	ps = pci_get_powerstate(dev);
6544 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6545 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6546 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6547 	pci_set_powerstate(dev, ps);
6548 	return (0);
6549 }
6550 
6551 /*
6552  * Try link drop and retrain of the downstream port of upstream
6553  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6554  * cause Conventional Hot reset of the device in the slot.
6555  * Alternative, for PCIe, could be the secondary bus reset initiatied
6556  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6557  */
6558 int
6559 pcie_link_reset(device_t port, int pcie_location)
6560 {
6561 	uint16_t v;
6562 
6563 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6564 	v |= PCIEM_LINK_CTL_LINK_DIS;
6565 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6566 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6567 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6568 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6569 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6570 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6571 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6572 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6573 }
6574 
6575 static int
6576 pci_reset_post(device_t dev, device_t child)
6577 {
6578 
6579 	if (dev == device_get_parent(child))
6580 		pci_restore_state(child);
6581 	return (0);
6582 }
6583 
6584 static int
6585 pci_reset_prepare(device_t dev, device_t child)
6586 {
6587 
6588 	if (dev == device_get_parent(child))
6589 		pci_save_state(child);
6590 	return (0);
6591 }
6592 
6593 static int
6594 pci_reset_child(device_t dev, device_t child, int flags)
6595 {
6596 	int error;
6597 
6598 	if (dev == NULL || device_get_parent(child) != dev)
6599 		return (0);
6600 	if ((flags & DEVF_RESET_DETACH) != 0) {
6601 		error = device_get_state(child) == DS_ATTACHED ?
6602 		    device_detach(child) : 0;
6603 	} else {
6604 		error = BUS_SUSPEND_CHILD(dev, child);
6605 	}
6606 	if (error == 0) {
6607 		if (!pcie_flr(child, 1000, false)) {
6608 			error = BUS_RESET_PREPARE(dev, child);
6609 			if (error == 0)
6610 				pci_power_reset(child);
6611 			BUS_RESET_POST(dev, child);
6612 		}
6613 		if ((flags & DEVF_RESET_DETACH) != 0)
6614 			device_probe_and_attach(child);
6615 		else
6616 			BUS_RESUME_CHILD(dev, child);
6617 	}
6618 	return (error);
6619 }
6620 
6621 const struct pci_device_table *
6622 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6623 {
6624 	bool match;
6625 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6626 
6627 	vendor = pci_get_vendor(child);
6628 	device = pci_get_device(child);
6629 	subvendor = pci_get_subvendor(child);
6630 	subdevice = pci_get_subdevice(child);
6631 	class = pci_get_class(child);
6632 	subclass = pci_get_subclass(child);
6633 	revid = pci_get_revid(child);
6634 	while (nelt-- > 0) {
6635 		match = true;
6636 		if (id->match_flag_vendor)
6637 			match &= vendor == id->vendor;
6638 		if (id->match_flag_device)
6639 			match &= device == id->device;
6640 		if (id->match_flag_subvendor)
6641 			match &= subvendor == id->subvendor;
6642 		if (id->match_flag_subdevice)
6643 			match &= subdevice == id->subdevice;
6644 		if (id->match_flag_class)
6645 			match &= class == id->class_id;
6646 		if (id->match_flag_subclass)
6647 			match &= subclass == id->subclass;
6648 		if (id->match_flag_revid)
6649 			match &= revid == id->revid;
6650 		if (match)
6651 			return (id);
6652 		id++;
6653 	}
6654 	return (NULL);
6655 }
6656 
6657 static void
6658 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6659 {
6660 	const char *dev_name;
6661 	device_t dev;
6662 
6663 	dev = dinfo->cfg.dev;
6664 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6665 	    dinfo->cfg.slot, dinfo->cfg.func);
6666 	dev_name = device_get_name(dev);
6667 	if (dev_name != NULL)
6668 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6669 }
6670 
6671 void
6672 pci_print_faulted_dev(void)
6673 {
6674 	struct pci_devinfo *dinfo;
6675 	device_t dev;
6676 	int aer, i;
6677 	uint32_t r1, r2;
6678 	uint16_t status;
6679 
6680 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6681 		dev = dinfo->cfg.dev;
6682 		status = pci_read_config(dev, PCIR_STATUS, 2);
6683 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6684 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6685 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6686 		if (status != 0) {
6687 			pci_print_faulted_dev_name(dinfo);
6688 			printf(" error 0x%04x\n", status);
6689 		}
6690 		if (dinfo->cfg.pcie.pcie_location != 0) {
6691 			status = pci_read_config(dev,
6692 			    dinfo->cfg.pcie.pcie_location +
6693 			    PCIER_DEVICE_STA, 2);
6694 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6695 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6696 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6697 				pci_print_faulted_dev_name(dinfo);
6698 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6699 				    pci_read_config(dev,
6700 				    dinfo->cfg.pcie.pcie_location +
6701 				    PCIER_DEVICE_CTL, 2),
6702 				    status);
6703 			}
6704 		}
6705 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6706 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6707 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6708 			if (r1 != 0 || r2 != 0) {
6709 				pci_print_faulted_dev_name(dinfo);
6710 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6711 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6712 				    r1, pci_read_config(dev, aer +
6713 				    PCIR_AER_UC_MASK, 4),
6714 				    pci_read_config(dev, aer +
6715 				    PCIR_AER_UC_SEVERITY, 4),
6716 				    r2, pci_read_config(dev, aer +
6717 				    PCIR_AER_COR_MASK, 4),
6718 				    pci_read_config(dev, aer +
6719 				    PCIR_AER_CAP_CONTROL, 4));
6720 				for (i = 0; i < 4; i++) {
6721 					r1 = pci_read_config(dev, aer +
6722 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6723 					printf("    HL%d: 0x%08x\n", i, r1);
6724 				}
6725 			}
6726 		}
6727 	}
6728 }
6729 
6730 #ifdef DDB
6731 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6732 {
6733 
6734 	pci_print_faulted_dev();
6735 }
6736 
6737 static void
6738 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6739 {
6740 	device_t dev;
6741 	int aer;
6742 	uint32_t r;
6743 
6744 	dev = dinfo->cfg.dev;
6745 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6746 	    PCIER_DEVICE_STA, 2);
6747 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6748 	    PCIER_DEVICE_STA, r, 2);
6749 
6750 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6751 		return;
6752 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6753 	if (r != 0)
6754 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6755 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6756 	if (r != 0)
6757 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6758 }
6759 
6760 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6761 {
6762 	struct pci_devinfo *dinfo;
6763 	device_t dev;
6764 	uint16_t status, status1;
6765 
6766 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6767 		dev = dinfo->cfg.dev;
6768 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6769 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6770 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6771 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6772 		if (status1 != 0) {
6773 			status &= ~status1;
6774 			pci_write_config(dev, PCIR_STATUS, status, 2);
6775 		}
6776 		if (dinfo->cfg.pcie.pcie_location != 0)
6777 			db_clear_pcie_errors(dinfo);
6778 	}
6779 }
6780 #endif
6781