xref: /freebsd/sys/dev/pci/pci.c (revision e17f5b1d)
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_bus.h"
36 
37 #include <sys/param.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 
55 #include <sys/bus.h>
56 #include <machine/bus.h>
57 #include <sys/rman.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
60 
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
63 #endif
64 
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
69 
70 #ifdef PCI_IOV
71 #include <sys/nv.h>
72 #include <dev/pci/pci_iov_private.h>
73 #endif
74 
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
79 
80 #include "pcib_if.h"
81 #include "pci_if.h"
82 
83 #define	PCIR_IS_BIOS(cfg, reg)						\
84 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
85 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 
87 static int		pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t	pci_mapbase(uint64_t mapreg);
89 static const char	*pci_maptype(uint64_t mapreg);
90 static int		pci_maprange(uint64_t mapreg);
91 static pci_addr_t	pci_rombase(uint64_t mapreg);
92 static int		pci_romsize(uint64_t testval);
93 static void		pci_fixancient(pcicfgregs *cfg);
94 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 
96 static int		pci_porten(device_t dev);
97 static int		pci_memen(device_t dev);
98 static void		pci_assign_interrupt(device_t bus, device_t dev,
99 			    int force_route);
100 static int		pci_add_map(device_t bus, device_t dev, int reg,
101 			    struct resource_list *rl, int force, int prefetch);
102 static int		pci_probe(device_t dev);
103 static void		pci_load_vendor_data(void);
104 static int		pci_describe_parse_line(char **ptr, int *vendor,
105 			    int *device, char **desc);
106 static char		*pci_describe_device(device_t dev);
107 static int		pci_modevent(module_t mod, int what, void *arg);
108 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
109 			    pcicfgregs *cfg);
110 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
111 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
112 			    int reg, uint32_t *data);
113 #if 0
114 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
115 			    int reg, uint32_t data);
116 #endif
117 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
118 static void		pci_mask_msix(device_t dev, u_int index);
119 static void		pci_unmask_msix(device_t dev, u_int index);
120 static int		pci_msi_blacklisted(void);
121 static int		pci_msix_blacklisted(void);
122 static void		pci_resume_msi(device_t dev);
123 static void		pci_resume_msix(device_t dev);
124 static int		pci_remap_intr_method(device_t bus, device_t dev,
125 			    u_int irq);
126 static void		pci_hint_device_unit(device_t acdev, device_t child,
127 			    const char *name, int *unitp);
128 static int		pci_reset_post(device_t dev, device_t child);
129 static int		pci_reset_prepare(device_t dev, device_t child);
130 static int		pci_reset_child(device_t dev, device_t child,
131 			    int flags);
132 
133 static int		pci_get_id_method(device_t dev, device_t child,
134 			    enum pci_id_type type, uintptr_t *rid);
135 
136 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
137     int b, int s, int f, uint16_t vid, uint16_t did);
138 
139 static device_method_t pci_methods[] = {
140 	/* Device interface */
141 	DEVMETHOD(device_probe,		pci_probe),
142 	DEVMETHOD(device_attach,	pci_attach),
143 	DEVMETHOD(device_detach,	pci_detach),
144 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
145 	DEVMETHOD(device_suspend,	bus_generic_suspend),
146 	DEVMETHOD(device_resume,	pci_resume),
147 
148 	/* Bus interface */
149 	DEVMETHOD(bus_print_child,	pci_print_child),
150 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
151 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
152 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
153 	DEVMETHOD(bus_driver_added,	pci_driver_added),
154 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
155 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
156 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
157 	DEVMETHOD(bus_reset_post,	pci_reset_post),
158 	DEVMETHOD(bus_reset_child,	pci_reset_child),
159 
160 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
161 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
162 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
163 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
164 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
165 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
166 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
167 	DEVMETHOD(bus_release_resource,	pci_release_resource),
168 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
169 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
170 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
171 	DEVMETHOD(bus_child_detached,	pci_child_detached),
172 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
173 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
174 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
175 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
176 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
177 	DEVMETHOD(bus_resume_child,	pci_resume_child),
178 	DEVMETHOD(bus_rescan,		pci_rescan_method),
179 
180 	/* PCI interface */
181 	DEVMETHOD(pci_read_config,	pci_read_config_method),
182 	DEVMETHOD(pci_write_config,	pci_write_config_method),
183 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
184 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
185 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
186 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
187 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
188 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
189 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
190 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
191 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
192 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
193 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
194 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
195 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
196 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
197 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
198 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
199 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
200 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
201 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
202 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
203 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
204 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
205 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
206 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
207 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
208 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
209 	DEVMETHOD(pci_get_id,		pci_get_id_method),
210 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
211 	DEVMETHOD(pci_child_added,	pci_child_added_method),
212 #ifdef PCI_IOV
213 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
214 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
215 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
216 #endif
217 
218 	DEVMETHOD_END
219 };
220 
221 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
222 
223 static devclass_t pci_devclass;
224 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
225     BUS_PASS_BUS);
226 MODULE_VERSION(pci, 1);
227 
228 static char	*pci_vendordata;
229 static size_t	pci_vendordata_size;
230 
231 struct pci_quirk {
232 	uint32_t devid;	/* Vendor/device of the card */
233 	int	type;
234 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
235 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
236 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
237 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
238 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
239 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
240 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
241 	int	arg1;
242 	int	arg2;
243 };
244 
245 static const struct pci_quirk pci_quirks[] = {
246 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
247 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
248 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
249 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
250 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 
252 	/*
253 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
254 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
255 	 */
256 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
257 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
258 
259 	/*
260 	 * MSI doesn't work on earlier Intel chipsets including
261 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
262 	 */
263 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
264 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
265 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 
271 	/*
272 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
273 	 * bridge.
274 	 */
275 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
276 
277 	/*
278 	 * Some virtualization environments emulate an older chipset
279 	 * but support MSI just fine.  QEMU uses the Intel 82440.
280 	 */
281 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
282 
283 	/*
284 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
285 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
286 	 * It prevents us from attaching hpet(4) when the bit is unset.
287 	 * Note this quirk only affects SB600 revision A13 and earlier.
288 	 * For SB600 A21 and later, firmware must set the bit to hide it.
289 	 * For SB700 and later, it is unused and hardcoded to zero.
290 	 */
291 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
292 
293 	/*
294 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
295 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
296 	 * of the command register is set.
297 	 */
298 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
299 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
300 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 
304 	/*
305 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
306 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
307 	 */
308 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
309 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
310 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
311 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
312 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
313 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
314 
315 	/*
316 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
317 	 * expected place.
318 	 */
319 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
320 
321 	{ 0 }
322 };
323 
324 /* map register information */
325 #define	PCI_MAPMEM	0x01	/* memory map */
326 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
327 #define	PCI_MAPPORT	0x04	/* port map */
328 
329 struct devlist pci_devq;
330 uint32_t pci_generation;
331 uint32_t pci_numdevs = 0;
332 static int pcie_chipset, pcix_chipset;
333 
334 /* sysctl vars */
335 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
336     "PCI bus tuning parameters");
337 
338 static int pci_enable_io_modes = 1;
339 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
340     &pci_enable_io_modes, 1,
341     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
342     " enable these bits correctly.  We'd like to do this all the time, but"
343     " there are some peripherals that this causes problems with.");
344 
345 static int pci_do_realloc_bars = 1;
346 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
347     &pci_do_realloc_bars, 0,
348     "Attempt to allocate a new range for any BARs whose original "
349     "firmware-assigned ranges fail to allocate during the initial device scan.");
350 
351 static int pci_do_power_nodriver = 0;
352 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
353     &pci_do_power_nodriver, 0,
354     "Place a function into D3 state when no driver attaches to it.  0 means"
355     " disable.  1 means conservatively place devices into D3 state.  2 means"
356     " aggressively place devices into D3 state.  3 means put absolutely"
357     " everything in D3 state.");
358 
359 int pci_do_power_resume = 1;
360 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
361     &pci_do_power_resume, 1,
362   "Transition from D3 -> D0 on resume.");
363 
364 int pci_do_power_suspend = 1;
365 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
366     &pci_do_power_suspend, 1,
367   "Transition from D0 -> D3 on suspend.");
368 
369 static int pci_do_msi = 1;
370 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
371     "Enable support for MSI interrupts");
372 
373 static int pci_do_msix = 1;
374 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
375     "Enable support for MSI-X interrupts");
376 
377 static int pci_msix_rewrite_table = 0;
378 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
379     &pci_msix_rewrite_table, 0,
380     "Rewrite entire MSI-X table when updating MSI-X entries");
381 
382 static int pci_honor_msi_blacklist = 1;
383 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
384     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
385 
386 #if defined(__i386__) || defined(__amd64__)
387 static int pci_usb_takeover = 1;
388 #else
389 static int pci_usb_takeover = 0;
390 #endif
391 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
392     &pci_usb_takeover, 1,
393     "Enable early takeover of USB controllers. Disable this if you depend on"
394     " BIOS emulation of USB devices, that is you use USB devices (like"
395     " keyboard or mouse) but do not load USB drivers");
396 
397 static int pci_clear_bars;
398 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
399     "Ignore firmware-assigned resources for BARs.");
400 
401 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
402 static int pci_clear_buses;
403 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
404     "Ignore firmware-assigned bus numbers.");
405 #endif
406 
407 static int pci_enable_ari = 1;
408 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
409     0, "Enable support for PCIe Alternative RID Interpretation");
410 
411 int pci_enable_aspm;
412 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
413     0, "Enable support for PCIe Active State Power Management");
414 
415 static int pci_clear_aer_on_attach = 0;
416 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
417     &pci_clear_aer_on_attach, 0,
418     "Clear port and device AER state on driver attach");
419 
420 static int
421 pci_has_quirk(uint32_t devid, int quirk)
422 {
423 	const struct pci_quirk *q;
424 
425 	for (q = &pci_quirks[0]; q->devid; q++) {
426 		if (q->devid == devid && q->type == quirk)
427 			return (1);
428 	}
429 	return (0);
430 }
431 
432 /* Find a device_t by bus/slot/function in domain 0 */
433 
434 device_t
435 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
436 {
437 
438 	return (pci_find_dbsf(0, bus, slot, func));
439 }
440 
441 /* Find a device_t by domain/bus/slot/function */
442 
443 device_t
444 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
445 {
446 	struct pci_devinfo *dinfo = NULL;
447 
448 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
449 		if ((dinfo->cfg.domain == domain) &&
450 		    (dinfo->cfg.bus == bus) &&
451 		    (dinfo->cfg.slot == slot) &&
452 		    (dinfo->cfg.func == func)) {
453 			break;
454 		}
455 	}
456 
457 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
458 }
459 
460 /* Find a device_t by vendor/device ID */
461 
462 device_t
463 pci_find_device(uint16_t vendor, uint16_t device)
464 {
465 	struct pci_devinfo *dinfo;
466 
467 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
468 		if ((dinfo->cfg.vendor == vendor) &&
469 		    (dinfo->cfg.device == device)) {
470 			return (dinfo->cfg.dev);
471 		}
472 	}
473 
474 	return (NULL);
475 }
476 
477 device_t
478 pci_find_class(uint8_t class, uint8_t subclass)
479 {
480 	struct pci_devinfo *dinfo;
481 
482 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
483 		if (dinfo->cfg.baseclass == class &&
484 		    dinfo->cfg.subclass == subclass) {
485 			return (dinfo->cfg.dev);
486 		}
487 	}
488 
489 	return (NULL);
490 }
491 
492 static int
493 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
494 {
495 	va_list ap;
496 	int retval;
497 
498 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
499 	    cfg->func);
500 	va_start(ap, fmt);
501 	retval += vprintf(fmt, ap);
502 	va_end(ap);
503 	return (retval);
504 }
505 
506 /* return base address of memory or port map */
507 
508 static pci_addr_t
509 pci_mapbase(uint64_t mapreg)
510 {
511 
512 	if (PCI_BAR_MEM(mapreg))
513 		return (mapreg & PCIM_BAR_MEM_BASE);
514 	else
515 		return (mapreg & PCIM_BAR_IO_BASE);
516 }
517 
518 /* return map type of memory or port map */
519 
520 static const char *
521 pci_maptype(uint64_t mapreg)
522 {
523 
524 	if (PCI_BAR_IO(mapreg))
525 		return ("I/O Port");
526 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
527 		return ("Prefetchable Memory");
528 	return ("Memory");
529 }
530 
531 /* return log2 of map size decoded for memory or port map */
532 
533 int
534 pci_mapsize(uint64_t testval)
535 {
536 	int ln2size;
537 
538 	testval = pci_mapbase(testval);
539 	ln2size = 0;
540 	if (testval != 0) {
541 		while ((testval & 1) == 0)
542 		{
543 			ln2size++;
544 			testval >>= 1;
545 		}
546 	}
547 	return (ln2size);
548 }
549 
550 /* return base address of device ROM */
551 
552 static pci_addr_t
553 pci_rombase(uint64_t mapreg)
554 {
555 
556 	return (mapreg & PCIM_BIOS_ADDR_MASK);
557 }
558 
559 /* return log2 of map size decided for device ROM */
560 
561 static int
562 pci_romsize(uint64_t testval)
563 {
564 	int ln2size;
565 
566 	testval = pci_rombase(testval);
567 	ln2size = 0;
568 	if (testval != 0) {
569 		while ((testval & 1) == 0)
570 		{
571 			ln2size++;
572 			testval >>= 1;
573 		}
574 	}
575 	return (ln2size);
576 }
577 
578 /* return log2 of address range supported by map register */
579 
580 static int
581 pci_maprange(uint64_t mapreg)
582 {
583 	int ln2range = 0;
584 
585 	if (PCI_BAR_IO(mapreg))
586 		ln2range = 32;
587 	else
588 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
589 		case PCIM_BAR_MEM_32:
590 			ln2range = 32;
591 			break;
592 		case PCIM_BAR_MEM_1MB:
593 			ln2range = 20;
594 			break;
595 		case PCIM_BAR_MEM_64:
596 			ln2range = 64;
597 			break;
598 		}
599 	return (ln2range);
600 }
601 
602 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
603 
604 static void
605 pci_fixancient(pcicfgregs *cfg)
606 {
607 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
608 		return;
609 
610 	/* PCI to PCI bridges use header type 1 */
611 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
612 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
613 }
614 
615 /* extract header type specific config data */
616 
617 static void
618 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
619 {
620 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
621 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
622 	case PCIM_HDRTYPE_NORMAL:
623 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
624 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
625 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
626 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
627 		cfg->nummaps	    = PCI_MAXMAPS_0;
628 		break;
629 	case PCIM_HDRTYPE_BRIDGE:
630 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
631 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
632 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
633 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
634 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
635 		cfg->nummaps	    = PCI_MAXMAPS_1;
636 		break;
637 	case PCIM_HDRTYPE_CARDBUS:
638 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
639 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
640 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
641 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
642 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
643 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
644 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
645 		cfg->nummaps	    = PCI_MAXMAPS_2;
646 		break;
647 	}
648 #undef REG
649 }
650 
651 /* read configuration header into pcicfgregs structure */
652 struct pci_devinfo *
653 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
654 {
655 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
656 	uint16_t vid, did;
657 
658 	vid = REG(PCIR_VENDOR, 2);
659 	did = REG(PCIR_DEVICE, 2);
660 	if (vid != 0xffff)
661 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
662 
663 	return (NULL);
664 }
665 
666 struct pci_devinfo *
667 pci_alloc_devinfo_method(device_t dev)
668 {
669 
670 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
671 	    M_WAITOK | M_ZERO));
672 }
673 
674 static struct pci_devinfo *
675 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
676     uint16_t vid, uint16_t did)
677 {
678 	struct pci_devinfo *devlist_entry;
679 	pcicfgregs *cfg;
680 
681 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
682 
683 	cfg = &devlist_entry->cfg;
684 
685 	cfg->domain		= d;
686 	cfg->bus		= b;
687 	cfg->slot		= s;
688 	cfg->func		= f;
689 	cfg->vendor		= vid;
690 	cfg->device		= did;
691 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
692 	cfg->statreg		= REG(PCIR_STATUS, 2);
693 	cfg->baseclass		= REG(PCIR_CLASS, 1);
694 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
695 	cfg->progif		= REG(PCIR_PROGIF, 1);
696 	cfg->revid		= REG(PCIR_REVID, 1);
697 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
698 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
699 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
700 	cfg->intpin		= REG(PCIR_INTPIN, 1);
701 	cfg->intline		= REG(PCIR_INTLINE, 1);
702 
703 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
704 	cfg->hdrtype		&= ~PCIM_MFDEV;
705 	STAILQ_INIT(&cfg->maps);
706 
707 	cfg->iov		= NULL;
708 
709 	pci_fixancient(cfg);
710 	pci_hdrtypedata(pcib, b, s, f, cfg);
711 
712 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
713 		pci_read_cap(pcib, cfg);
714 
715 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
716 
717 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
718 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
719 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
720 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
721 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
722 
723 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
724 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
725 	devlist_entry->conf.pc_vendor = cfg->vendor;
726 	devlist_entry->conf.pc_device = cfg->device;
727 
728 	devlist_entry->conf.pc_class = cfg->baseclass;
729 	devlist_entry->conf.pc_subclass = cfg->subclass;
730 	devlist_entry->conf.pc_progif = cfg->progif;
731 	devlist_entry->conf.pc_revid = cfg->revid;
732 
733 	pci_numdevs++;
734 	pci_generation++;
735 
736 	return (devlist_entry);
737 }
738 #undef REG
739 
740 static void
741 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
742 {
743 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
744     cfg->ea.ea_location + (n), w)
745 	int num_ent;
746 	int ptr;
747 	int a, b;
748 	uint32_t val;
749 	int ent_size;
750 	uint32_t dw[4];
751 	uint64_t base, max_offset;
752 	struct pci_ea_entry *eae;
753 
754 	if (cfg->ea.ea_location == 0)
755 		return;
756 
757 	STAILQ_INIT(&cfg->ea.ea_entries);
758 
759 	/* Determine the number of entries */
760 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
761 	num_ent &= PCIM_EA_NUM_ENT_MASK;
762 
763 	/* Find the first entry to care of */
764 	ptr = PCIR_EA_FIRST_ENT;
765 
766 	/* Skip DWORD 2 for type 1 functions */
767 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
768 		ptr += 4;
769 
770 	for (a = 0; a < num_ent; a++) {
771 
772 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
773 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
774 
775 		/* Read a number of dwords in the entry */
776 		val = REG(ptr, 4);
777 		ptr += 4;
778 		ent_size = (val & PCIM_EA_ES);
779 
780 		for (b = 0; b < ent_size; b++) {
781 			dw[b] = REG(ptr, 4);
782 			ptr += 4;
783 		}
784 
785 		eae->eae_flags = val;
786 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
787 
788 		base = dw[0] & PCIM_EA_FIELD_MASK;
789 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
790 		b = 2;
791 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
792 			base |= (uint64_t)dw[b] << 32UL;
793 			b++;
794 		}
795 		if (((dw[1] & PCIM_EA_IS_64) != 0)
796 		    && (b < ent_size)) {
797 			max_offset |= (uint64_t)dw[b] << 32UL;
798 			b++;
799 		}
800 
801 		eae->eae_base = base;
802 		eae->eae_max_offset = max_offset;
803 
804 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
805 
806 		if (bootverbose) {
807 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
808 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
809 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
810 		}
811 	}
812 }
813 #undef REG
814 
815 static void
816 pci_read_cap(device_t pcib, pcicfgregs *cfg)
817 {
818 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
819 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
820 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
821 	uint64_t addr;
822 #endif
823 	uint32_t val;
824 	int	ptr, nextptr, ptrptr;
825 
826 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
827 	case PCIM_HDRTYPE_NORMAL:
828 	case PCIM_HDRTYPE_BRIDGE:
829 		ptrptr = PCIR_CAP_PTR;
830 		break;
831 	case PCIM_HDRTYPE_CARDBUS:
832 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
833 		break;
834 	default:
835 		return;		/* no extended capabilities support */
836 	}
837 	nextptr = REG(ptrptr, 1);	/* sanity check? */
838 
839 	/*
840 	 * Read capability entries.
841 	 */
842 	while (nextptr != 0) {
843 		/* Sanity check */
844 		if (nextptr > 255) {
845 			printf("illegal PCI extended capability offset %d\n",
846 			    nextptr);
847 			return;
848 		}
849 		/* Find the next entry */
850 		ptr = nextptr;
851 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
852 
853 		/* Process this entry */
854 		switch (REG(ptr + PCICAP_ID, 1)) {
855 		case PCIY_PMG:		/* PCI power management */
856 			if (cfg->pp.pp_cap == 0) {
857 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
858 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
859 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
860 				if ((nextptr - ptr) > PCIR_POWER_DATA)
861 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
862 			}
863 			break;
864 		case PCIY_HT:		/* HyperTransport */
865 			/* Determine HT-specific capability type. */
866 			val = REG(ptr + PCIR_HT_COMMAND, 2);
867 
868 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
869 				cfg->ht.ht_slave = ptr;
870 
871 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
872 			switch (val & PCIM_HTCMD_CAP_MASK) {
873 			case PCIM_HTCAP_MSI_MAPPING:
874 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
875 					/* Sanity check the mapping window. */
876 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
877 					    4);
878 					addr <<= 32;
879 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
880 					    4);
881 					if (addr != MSI_INTEL_ADDR_BASE)
882 						device_printf(pcib,
883 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
884 						    cfg->domain, cfg->bus,
885 						    cfg->slot, cfg->func,
886 						    (long long)addr);
887 				} else
888 					addr = MSI_INTEL_ADDR_BASE;
889 
890 				cfg->ht.ht_msimap = ptr;
891 				cfg->ht.ht_msictrl = val;
892 				cfg->ht.ht_msiaddr = addr;
893 				break;
894 			}
895 #endif
896 			break;
897 		case PCIY_MSI:		/* PCI MSI */
898 			cfg->msi.msi_location = ptr;
899 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
900 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
901 						     PCIM_MSICTRL_MMC_MASK)>>1);
902 			break;
903 		case PCIY_MSIX:		/* PCI MSI-X */
904 			cfg->msix.msix_location = ptr;
905 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
906 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
907 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
908 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
909 			cfg->msix.msix_table_bar = PCIR_BAR(val &
910 			    PCIM_MSIX_BIR_MASK);
911 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
912 			val = REG(ptr + PCIR_MSIX_PBA, 4);
913 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
914 			    PCIM_MSIX_BIR_MASK);
915 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
916 			break;
917 		case PCIY_VPD:		/* PCI Vital Product Data */
918 			cfg->vpd.vpd_reg = ptr;
919 			break;
920 		case PCIY_SUBVENDOR:
921 			/* Should always be true. */
922 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
923 			    PCIM_HDRTYPE_BRIDGE) {
924 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
925 				cfg->subvendor = val & 0xffff;
926 				cfg->subdevice = val >> 16;
927 			}
928 			break;
929 		case PCIY_PCIX:		/* PCI-X */
930 			/*
931 			 * Assume we have a PCI-X chipset if we have
932 			 * at least one PCI-PCI bridge with a PCI-X
933 			 * capability.  Note that some systems with
934 			 * PCI-express or HT chipsets might match on
935 			 * this check as well.
936 			 */
937 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
938 			    PCIM_HDRTYPE_BRIDGE)
939 				pcix_chipset = 1;
940 			cfg->pcix.pcix_location = ptr;
941 			break;
942 		case PCIY_EXPRESS:	/* PCI-express */
943 			/*
944 			 * Assume we have a PCI-express chipset if we have
945 			 * at least one PCI-express device.
946 			 */
947 			pcie_chipset = 1;
948 			cfg->pcie.pcie_location = ptr;
949 			val = REG(ptr + PCIER_FLAGS, 2);
950 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
951 			break;
952 		case PCIY_EA:		/* Enhanced Allocation */
953 			cfg->ea.ea_location = ptr;
954 			pci_ea_fill_info(pcib, cfg);
955 			break;
956 		default:
957 			break;
958 		}
959 	}
960 
961 #if defined(__powerpc__)
962 	/*
963 	 * Enable the MSI mapping window for all HyperTransport
964 	 * slaves.  PCI-PCI bridges have their windows enabled via
965 	 * PCIB_MAP_MSI().
966 	 */
967 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
968 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
969 		device_printf(pcib,
970 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
971 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
972 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
973 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
974 		     2);
975 	}
976 #endif
977 /* REG and WREG use carry through to next functions */
978 }
979 
980 /*
981  * PCI Vital Product Data
982  */
983 
984 #define	PCI_VPD_TIMEOUT		1000000
985 
986 static int
987 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
988 {
989 	int count = PCI_VPD_TIMEOUT;
990 
991 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
992 
993 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
994 
995 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
996 		if (--count < 0)
997 			return (ENXIO);
998 		DELAY(1);	/* limit looping */
999 	}
1000 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1001 
1002 	return (0);
1003 }
1004 
1005 #if 0
1006 static int
1007 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1008 {
1009 	int count = PCI_VPD_TIMEOUT;
1010 
1011 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1012 
1013 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1014 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1015 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1016 		if (--count < 0)
1017 			return (ENXIO);
1018 		DELAY(1);	/* limit looping */
1019 	}
1020 
1021 	return (0);
1022 }
1023 #endif
1024 
1025 #undef PCI_VPD_TIMEOUT
1026 
1027 struct vpd_readstate {
1028 	device_t	pcib;
1029 	pcicfgregs	*cfg;
1030 	uint32_t	val;
1031 	int		bytesinval;
1032 	int		off;
1033 	uint8_t		cksum;
1034 };
1035 
1036 static int
1037 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1038 {
1039 	uint32_t reg;
1040 	uint8_t byte;
1041 
1042 	if (vrs->bytesinval == 0) {
1043 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1044 			return (ENXIO);
1045 		vrs->val = le32toh(reg);
1046 		vrs->off += 4;
1047 		byte = vrs->val & 0xff;
1048 		vrs->bytesinval = 3;
1049 	} else {
1050 		vrs->val = vrs->val >> 8;
1051 		byte = vrs->val & 0xff;
1052 		vrs->bytesinval--;
1053 	}
1054 
1055 	vrs->cksum += byte;
1056 	*data = byte;
1057 	return (0);
1058 }
1059 
1060 static void
1061 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1062 {
1063 	struct vpd_readstate vrs;
1064 	int state;
1065 	int name;
1066 	int remain;
1067 	int i;
1068 	int alloc, off;		/* alloc/off for RO/W arrays */
1069 	int cksumvalid;
1070 	int dflen;
1071 	uint8_t byte;
1072 	uint8_t byte2;
1073 
1074 	/* init vpd reader */
1075 	vrs.bytesinval = 0;
1076 	vrs.off = 0;
1077 	vrs.pcib = pcib;
1078 	vrs.cfg = cfg;
1079 	vrs.cksum = 0;
1080 
1081 	state = 0;
1082 	name = remain = i = 0;	/* shut up stupid gcc */
1083 	alloc = off = 0;	/* shut up stupid gcc */
1084 	dflen = 0;		/* shut up stupid gcc */
1085 	cksumvalid = -1;
1086 	while (state >= 0) {
1087 		if (vpd_nextbyte(&vrs, &byte)) {
1088 			state = -2;
1089 			break;
1090 		}
1091 #if 0
1092 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1093 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1094 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1095 #endif
1096 		switch (state) {
1097 		case 0:		/* item name */
1098 			if (byte & 0x80) {
1099 				if (vpd_nextbyte(&vrs, &byte2)) {
1100 					state = -2;
1101 					break;
1102 				}
1103 				remain = byte2;
1104 				if (vpd_nextbyte(&vrs, &byte2)) {
1105 					state = -2;
1106 					break;
1107 				}
1108 				remain |= byte2 << 8;
1109 				name = byte & 0x7f;
1110 			} else {
1111 				remain = byte & 0x7;
1112 				name = (byte >> 3) & 0xf;
1113 			}
1114 			if (vrs.off + remain - vrs.bytesinval > 0x8000) {
1115 				pci_printf(cfg,
1116 				    "VPD data overflow, remain %#x\n", remain);
1117 				state = -1;
1118 				break;
1119 			}
1120 			switch (name) {
1121 			case 0x2:	/* String */
1122 				cfg->vpd.vpd_ident = malloc(remain + 1,
1123 				    M_DEVBUF, M_WAITOK);
1124 				i = 0;
1125 				state = 1;
1126 				break;
1127 			case 0xf:	/* End */
1128 				state = -1;
1129 				break;
1130 			case 0x10:	/* VPD-R */
1131 				alloc = 8;
1132 				off = 0;
1133 				cfg->vpd.vpd_ros = malloc(alloc *
1134 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1135 				    M_WAITOK | M_ZERO);
1136 				state = 2;
1137 				break;
1138 			case 0x11:	/* VPD-W */
1139 				alloc = 8;
1140 				off = 0;
1141 				cfg->vpd.vpd_w = malloc(alloc *
1142 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1143 				    M_WAITOK | M_ZERO);
1144 				state = 5;
1145 				break;
1146 			default:	/* Invalid data, abort */
1147 				state = -1;
1148 				break;
1149 			}
1150 			break;
1151 
1152 		case 1:	/* Identifier String */
1153 			cfg->vpd.vpd_ident[i++] = byte;
1154 			remain--;
1155 			if (remain == 0)  {
1156 				cfg->vpd.vpd_ident[i] = '\0';
1157 				state = 0;
1158 			}
1159 			break;
1160 
1161 		case 2:	/* VPD-R Keyword Header */
1162 			if (off == alloc) {
1163 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1164 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1165 				    M_DEVBUF, M_WAITOK | M_ZERO);
1166 			}
1167 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1168 			if (vpd_nextbyte(&vrs, &byte2)) {
1169 				state = -2;
1170 				break;
1171 			}
1172 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1173 			if (vpd_nextbyte(&vrs, &byte2)) {
1174 				state = -2;
1175 				break;
1176 			}
1177 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1178 			if (dflen == 0 &&
1179 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1180 			    2) == 0) {
1181 				/*
1182 				 * if this happens, we can't trust the rest
1183 				 * of the VPD.
1184 				 */
1185 				pci_printf(cfg, "bad keyword length: %d\n",
1186 				    dflen);
1187 				cksumvalid = 0;
1188 				state = -1;
1189 				break;
1190 			} else if (dflen == 0) {
1191 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1192 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1193 				    M_DEVBUF, M_WAITOK);
1194 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1195 			} else
1196 				cfg->vpd.vpd_ros[off].value = malloc(
1197 				    (dflen + 1) *
1198 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1199 				    M_DEVBUF, M_WAITOK);
1200 			remain -= 3;
1201 			i = 0;
1202 			/* keep in sync w/ state 3's transistions */
1203 			if (dflen == 0 && remain == 0)
1204 				state = 0;
1205 			else if (dflen == 0)
1206 				state = 2;
1207 			else
1208 				state = 3;
1209 			break;
1210 
1211 		case 3:	/* VPD-R Keyword Value */
1212 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1213 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1214 			    "RV", 2) == 0 && cksumvalid == -1) {
1215 				if (vrs.cksum == 0)
1216 					cksumvalid = 1;
1217 				else {
1218 					if (bootverbose)
1219 						pci_printf(cfg,
1220 					    "bad VPD cksum, remain %hhu\n",
1221 						    vrs.cksum);
1222 					cksumvalid = 0;
1223 					state = -1;
1224 					break;
1225 				}
1226 			}
1227 			dflen--;
1228 			remain--;
1229 			/* keep in sync w/ state 2's transistions */
1230 			if (dflen == 0)
1231 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1232 			if (dflen == 0 && remain == 0) {
1233 				cfg->vpd.vpd_rocnt = off;
1234 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1235 				    off * sizeof(*cfg->vpd.vpd_ros),
1236 				    M_DEVBUF, M_WAITOK | M_ZERO);
1237 				state = 0;
1238 			} else if (dflen == 0)
1239 				state = 2;
1240 			break;
1241 
1242 		case 4:
1243 			remain--;
1244 			if (remain == 0)
1245 				state = 0;
1246 			break;
1247 
1248 		case 5:	/* VPD-W Keyword Header */
1249 			if (off == alloc) {
1250 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1251 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1252 				    M_DEVBUF, M_WAITOK | M_ZERO);
1253 			}
1254 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1255 			if (vpd_nextbyte(&vrs, &byte2)) {
1256 				state = -2;
1257 				break;
1258 			}
1259 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1260 			if (vpd_nextbyte(&vrs, &byte2)) {
1261 				state = -2;
1262 				break;
1263 			}
1264 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1265 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1266 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1267 			    sizeof(*cfg->vpd.vpd_w[off].value),
1268 			    M_DEVBUF, M_WAITOK);
1269 			remain -= 3;
1270 			i = 0;
1271 			/* keep in sync w/ state 6's transistions */
1272 			if (dflen == 0 && remain == 0)
1273 				state = 0;
1274 			else if (dflen == 0)
1275 				state = 5;
1276 			else
1277 				state = 6;
1278 			break;
1279 
1280 		case 6:	/* VPD-W Keyword Value */
1281 			cfg->vpd.vpd_w[off].value[i++] = byte;
1282 			dflen--;
1283 			remain--;
1284 			/* keep in sync w/ state 5's transistions */
1285 			if (dflen == 0)
1286 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1287 			if (dflen == 0 && remain == 0) {
1288 				cfg->vpd.vpd_wcnt = off;
1289 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1290 				    off * sizeof(*cfg->vpd.vpd_w),
1291 				    M_DEVBUF, M_WAITOK | M_ZERO);
1292 				state = 0;
1293 			} else if (dflen == 0)
1294 				state = 5;
1295 			break;
1296 
1297 		default:
1298 			pci_printf(cfg, "invalid state: %d\n", state);
1299 			state = -1;
1300 			break;
1301 		}
1302 	}
1303 
1304 	if (cksumvalid == 0 || state < -1) {
1305 		/* read-only data bad, clean up */
1306 		if (cfg->vpd.vpd_ros != NULL) {
1307 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1308 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1309 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1310 			cfg->vpd.vpd_ros = NULL;
1311 		}
1312 	}
1313 	if (state < -1) {
1314 		/* I/O error, clean up */
1315 		pci_printf(cfg, "failed to read VPD data.\n");
1316 		if (cfg->vpd.vpd_ident != NULL) {
1317 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1318 			cfg->vpd.vpd_ident = NULL;
1319 		}
1320 		if (cfg->vpd.vpd_w != NULL) {
1321 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1322 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1323 			free(cfg->vpd.vpd_w, M_DEVBUF);
1324 			cfg->vpd.vpd_w = NULL;
1325 		}
1326 	}
1327 	cfg->vpd.vpd_cached = 1;
1328 #undef REG
1329 #undef WREG
1330 }
1331 
1332 int
1333 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1334 {
1335 	struct pci_devinfo *dinfo = device_get_ivars(child);
1336 	pcicfgregs *cfg = &dinfo->cfg;
1337 
1338 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1339 		pci_read_vpd(device_get_parent(dev), cfg);
1340 
1341 	*identptr = cfg->vpd.vpd_ident;
1342 
1343 	if (*identptr == NULL)
1344 		return (ENXIO);
1345 
1346 	return (0);
1347 }
1348 
1349 int
1350 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1351 	const char **vptr)
1352 {
1353 	struct pci_devinfo *dinfo = device_get_ivars(child);
1354 	pcicfgregs *cfg = &dinfo->cfg;
1355 	int i;
1356 
1357 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1358 		pci_read_vpd(device_get_parent(dev), cfg);
1359 
1360 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1361 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1362 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1363 			*vptr = cfg->vpd.vpd_ros[i].value;
1364 			return (0);
1365 		}
1366 
1367 	*vptr = NULL;
1368 	return (ENXIO);
1369 }
1370 
1371 struct pcicfg_vpd *
1372 pci_fetch_vpd_list(device_t dev)
1373 {
1374 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1375 	pcicfgregs *cfg = &dinfo->cfg;
1376 
1377 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1378 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1379 	return (&cfg->vpd);
1380 }
1381 
1382 /*
1383  * Find the requested HyperTransport capability and return the offset
1384  * in configuration space via the pointer provided.  The function
1385  * returns 0 on success and an error code otherwise.
1386  */
1387 int
1388 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1389 {
1390 	int ptr, error;
1391 	uint16_t val;
1392 
1393 	error = pci_find_cap(child, PCIY_HT, &ptr);
1394 	if (error)
1395 		return (error);
1396 
1397 	/*
1398 	 * Traverse the capabilities list checking each HT capability
1399 	 * to see if it matches the requested HT capability.
1400 	 */
1401 	for (;;) {
1402 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1403 		if (capability == PCIM_HTCAP_SLAVE ||
1404 		    capability == PCIM_HTCAP_HOST)
1405 			val &= 0xe000;
1406 		else
1407 			val &= PCIM_HTCMD_CAP_MASK;
1408 		if (val == capability) {
1409 			if (capreg != NULL)
1410 				*capreg = ptr;
1411 			return (0);
1412 		}
1413 
1414 		/* Skip to the next HT capability. */
1415 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1416 			break;
1417 	}
1418 
1419 	return (ENOENT);
1420 }
1421 
1422 /*
1423  * Find the next requested HyperTransport capability after start and return
1424  * the offset in configuration space via the pointer provided.  The function
1425  * returns 0 on success and an error code otherwise.
1426  */
1427 int
1428 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1429     int start, int *capreg)
1430 {
1431 	int ptr;
1432 	uint16_t val;
1433 
1434 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1435 	    ("start capability is not HyperTransport capability"));
1436 	ptr = start;
1437 
1438 	/*
1439 	 * Traverse the capabilities list checking each HT capability
1440 	 * to see if it matches the requested HT capability.
1441 	 */
1442 	for (;;) {
1443 		/* Skip to the next HT capability. */
1444 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1445 			break;
1446 
1447 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1448 		if (capability == PCIM_HTCAP_SLAVE ||
1449 		    capability == PCIM_HTCAP_HOST)
1450 			val &= 0xe000;
1451 		else
1452 			val &= PCIM_HTCMD_CAP_MASK;
1453 		if (val == capability) {
1454 			if (capreg != NULL)
1455 				*capreg = ptr;
1456 			return (0);
1457 		}
1458 	}
1459 
1460 	return (ENOENT);
1461 }
1462 
1463 /*
1464  * Find the requested capability and return the offset in
1465  * configuration space via the pointer provided.  The function returns
1466  * 0 on success and an error code otherwise.
1467  */
1468 int
1469 pci_find_cap_method(device_t dev, device_t child, int capability,
1470     int *capreg)
1471 {
1472 	struct pci_devinfo *dinfo = device_get_ivars(child);
1473 	pcicfgregs *cfg = &dinfo->cfg;
1474 	uint32_t status;
1475 	uint8_t ptr;
1476 
1477 	/*
1478 	 * Check the CAP_LIST bit of the PCI status register first.
1479 	 */
1480 	status = pci_read_config(child, PCIR_STATUS, 2);
1481 	if (!(status & PCIM_STATUS_CAPPRESENT))
1482 		return (ENXIO);
1483 
1484 	/*
1485 	 * Determine the start pointer of the capabilities list.
1486 	 */
1487 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1488 	case PCIM_HDRTYPE_NORMAL:
1489 	case PCIM_HDRTYPE_BRIDGE:
1490 		ptr = PCIR_CAP_PTR;
1491 		break;
1492 	case PCIM_HDRTYPE_CARDBUS:
1493 		ptr = PCIR_CAP_PTR_2;
1494 		break;
1495 	default:
1496 		/* XXX: panic? */
1497 		return (ENXIO);		/* no extended capabilities support */
1498 	}
1499 	ptr = pci_read_config(child, ptr, 1);
1500 
1501 	/*
1502 	 * Traverse the capabilities list.
1503 	 */
1504 	while (ptr != 0) {
1505 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1506 			if (capreg != NULL)
1507 				*capreg = ptr;
1508 			return (0);
1509 		}
1510 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1511 	}
1512 
1513 	return (ENOENT);
1514 }
1515 
1516 /*
1517  * Find the next requested capability after start and return the offset in
1518  * configuration space via the pointer provided.  The function returns
1519  * 0 on success and an error code otherwise.
1520  */
1521 int
1522 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1523     int start, int *capreg)
1524 {
1525 	uint8_t ptr;
1526 
1527 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1528 	    ("start capability is not expected capability"));
1529 
1530 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1531 	while (ptr != 0) {
1532 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1533 			if (capreg != NULL)
1534 				*capreg = ptr;
1535 			return (0);
1536 		}
1537 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1538 	}
1539 
1540 	return (ENOENT);
1541 }
1542 
1543 /*
1544  * Find the requested extended capability and return the offset in
1545  * configuration space via the pointer provided.  The function returns
1546  * 0 on success and an error code otherwise.
1547  */
1548 int
1549 pci_find_extcap_method(device_t dev, device_t child, int capability,
1550     int *capreg)
1551 {
1552 	struct pci_devinfo *dinfo = device_get_ivars(child);
1553 	pcicfgregs *cfg = &dinfo->cfg;
1554 	uint32_t ecap;
1555 	uint16_t ptr;
1556 
1557 	/* Only supported for PCI-express devices. */
1558 	if (cfg->pcie.pcie_location == 0)
1559 		return (ENXIO);
1560 
1561 	ptr = PCIR_EXTCAP;
1562 	ecap = pci_read_config(child, ptr, 4);
1563 	if (ecap == 0xffffffff || ecap == 0)
1564 		return (ENOENT);
1565 	for (;;) {
1566 		if (PCI_EXTCAP_ID(ecap) == capability) {
1567 			if (capreg != NULL)
1568 				*capreg = ptr;
1569 			return (0);
1570 		}
1571 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1572 		if (ptr == 0)
1573 			break;
1574 		ecap = pci_read_config(child, ptr, 4);
1575 	}
1576 
1577 	return (ENOENT);
1578 }
1579 
1580 /*
1581  * Find the next requested extended capability after start and return the
1582  * offset in configuration space via the pointer provided.  The function
1583  * returns 0 on success and an error code otherwise.
1584  */
1585 int
1586 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1587     int start, int *capreg)
1588 {
1589 	struct pci_devinfo *dinfo = device_get_ivars(child);
1590 	pcicfgregs *cfg = &dinfo->cfg;
1591 	uint32_t ecap;
1592 	uint16_t ptr;
1593 
1594 	/* Only supported for PCI-express devices. */
1595 	if (cfg->pcie.pcie_location == 0)
1596 		return (ENXIO);
1597 
1598 	ecap = pci_read_config(child, start, 4);
1599 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1600 	    ("start extended capability is not expected capability"));
1601 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1602 	while (ptr != 0) {
1603 		ecap = pci_read_config(child, ptr, 4);
1604 		if (PCI_EXTCAP_ID(ecap) == capability) {
1605 			if (capreg != NULL)
1606 				*capreg = ptr;
1607 			return (0);
1608 		}
1609 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1610 	}
1611 
1612 	return (ENOENT);
1613 }
1614 
1615 /*
1616  * Support for MSI-X message interrupts.
1617  */
1618 static void
1619 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1620 {
1621 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1622 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1623 	uint32_t offset;
1624 
1625 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1626 	offset = msix->msix_table_offset + index * 16;
1627 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1628 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1629 	bus_write_4(msix->msix_table_res, offset + 8, data);
1630 }
1631 
1632 void
1633 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1634     uint64_t address, uint32_t data)
1635 {
1636 
1637 	if (pci_msix_rewrite_table) {
1638 		struct pci_devinfo *dinfo = device_get_ivars(child);
1639 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1640 
1641 		/*
1642 		 * Some VM hosts require MSIX to be disabled in the
1643 		 * control register before updating the MSIX table
1644 		 * entries are allowed. It is not enough to only
1645 		 * disable MSIX while updating a single entry. MSIX
1646 		 * must be disabled while updating all entries in the
1647 		 * table.
1648 		 */
1649 		pci_write_config(child,
1650 		    msix->msix_location + PCIR_MSIX_CTRL,
1651 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1652 		pci_resume_msix(child);
1653 	} else
1654 		pci_write_msix_entry(child, index, address, data);
1655 
1656 	/* Enable MSI -> HT mapping. */
1657 	pci_ht_map_msi(child, address);
1658 }
1659 
1660 void
1661 pci_mask_msix(device_t dev, u_int index)
1662 {
1663 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1664 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1665 	uint32_t offset, val;
1666 
1667 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1668 	offset = msix->msix_table_offset + index * 16 + 12;
1669 	val = bus_read_4(msix->msix_table_res, offset);
1670 	val |= PCIM_MSIX_VCTRL_MASK;
1671 
1672 	/*
1673 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1674 	 * register, so always write the new value.
1675 	 */
1676 	bus_write_4(msix->msix_table_res, offset, val);
1677 }
1678 
1679 void
1680 pci_unmask_msix(device_t dev, u_int index)
1681 {
1682 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1683 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1684 	uint32_t offset, val;
1685 
1686 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1687 	offset = msix->msix_table_offset + index * 16 + 12;
1688 	val = bus_read_4(msix->msix_table_res, offset);
1689 	val &= ~PCIM_MSIX_VCTRL_MASK;
1690 
1691 	/*
1692 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1693 	 * register, so always write the new value.
1694 	 */
1695 	bus_write_4(msix->msix_table_res, offset, val);
1696 }
1697 
1698 int
1699 pci_pending_msix(device_t dev, u_int index)
1700 {
1701 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1702 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1703 	uint32_t offset, bit;
1704 
1705 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1706 	offset = msix->msix_pba_offset + (index / 32) * 4;
1707 	bit = 1 << index % 32;
1708 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1709 }
1710 
1711 /*
1712  * Restore MSI-X registers and table during resume.  If MSI-X is
1713  * enabled then walk the virtual table to restore the actual MSI-X
1714  * table.
1715  */
1716 static void
1717 pci_resume_msix(device_t dev)
1718 {
1719 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1720 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1721 	struct msix_table_entry *mte;
1722 	struct msix_vector *mv;
1723 	int i;
1724 
1725 	if (msix->msix_alloc > 0) {
1726 		/* First, mask all vectors. */
1727 		for (i = 0; i < msix->msix_msgnum; i++)
1728 			pci_mask_msix(dev, i);
1729 
1730 		/* Second, program any messages with at least one handler. */
1731 		for (i = 0; i < msix->msix_table_len; i++) {
1732 			mte = &msix->msix_table[i];
1733 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1734 				continue;
1735 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1736 			pci_write_msix_entry(dev, i, mv->mv_address,
1737 			    mv->mv_data);
1738 			pci_unmask_msix(dev, i);
1739 		}
1740 	}
1741 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1742 	    msix->msix_ctrl, 2);
1743 }
1744 
1745 /*
1746  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1747  * returned in *count.  After this function returns, each message will be
1748  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1749  */
1750 int
1751 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1752 {
1753 	struct pci_devinfo *dinfo = device_get_ivars(child);
1754 	pcicfgregs *cfg = &dinfo->cfg;
1755 	struct resource_list_entry *rle;
1756 	int actual, error, i, irq, max;
1757 
1758 	/* Don't let count == 0 get us into trouble. */
1759 	if (*count == 0)
1760 		return (EINVAL);
1761 
1762 	/* If rid 0 is allocated, then fail. */
1763 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1764 	if (rle != NULL && rle->res != NULL)
1765 		return (ENXIO);
1766 
1767 	/* Already have allocated messages? */
1768 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1769 		return (ENXIO);
1770 
1771 	/* If MSI-X is blacklisted for this system, fail. */
1772 	if (pci_msix_blacklisted())
1773 		return (ENXIO);
1774 
1775 	/* MSI-X capability present? */
1776 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1777 		return (ENODEV);
1778 
1779 	/* Make sure the appropriate BARs are mapped. */
1780 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1781 	    cfg->msix.msix_table_bar);
1782 	if (rle == NULL || rle->res == NULL ||
1783 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1784 		return (ENXIO);
1785 	cfg->msix.msix_table_res = rle->res;
1786 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1787 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1788 		    cfg->msix.msix_pba_bar);
1789 		if (rle == NULL || rle->res == NULL ||
1790 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1791 			return (ENXIO);
1792 	}
1793 	cfg->msix.msix_pba_res = rle->res;
1794 
1795 	if (bootverbose)
1796 		device_printf(child,
1797 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1798 		    *count, cfg->msix.msix_msgnum);
1799 	max = min(*count, cfg->msix.msix_msgnum);
1800 	for (i = 0; i < max; i++) {
1801 		/* Allocate a message. */
1802 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1803 		if (error) {
1804 			if (i == 0)
1805 				return (error);
1806 			break;
1807 		}
1808 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1809 		    irq, 1);
1810 	}
1811 	actual = i;
1812 
1813 	if (bootverbose) {
1814 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1815 		if (actual == 1)
1816 			device_printf(child, "using IRQ %ju for MSI-X\n",
1817 			    rle->start);
1818 		else {
1819 			int run;
1820 
1821 			/*
1822 			 * Be fancy and try to print contiguous runs of
1823 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1824 			 * 'run' is true if we are in a range.
1825 			 */
1826 			device_printf(child, "using IRQs %ju", rle->start);
1827 			irq = rle->start;
1828 			run = 0;
1829 			for (i = 1; i < actual; i++) {
1830 				rle = resource_list_find(&dinfo->resources,
1831 				    SYS_RES_IRQ, i + 1);
1832 
1833 				/* Still in a run? */
1834 				if (rle->start == irq + 1) {
1835 					run = 1;
1836 					irq++;
1837 					continue;
1838 				}
1839 
1840 				/* Finish previous range. */
1841 				if (run) {
1842 					printf("-%d", irq);
1843 					run = 0;
1844 				}
1845 
1846 				/* Start new range. */
1847 				printf(",%ju", rle->start);
1848 				irq = rle->start;
1849 			}
1850 
1851 			/* Unfinished range? */
1852 			if (run)
1853 				printf("-%d", irq);
1854 			printf(" for MSI-X\n");
1855 		}
1856 	}
1857 
1858 	/* Mask all vectors. */
1859 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1860 		pci_mask_msix(child, i);
1861 
1862 	/* Allocate and initialize vector data and virtual table. */
1863 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1864 	    M_DEVBUF, M_WAITOK | M_ZERO);
1865 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1866 	    M_DEVBUF, M_WAITOK | M_ZERO);
1867 	for (i = 0; i < actual; i++) {
1868 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1869 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1870 		cfg->msix.msix_table[i].mte_vector = i + 1;
1871 	}
1872 
1873 	/* Update control register to enable MSI-X. */
1874 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1875 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1876 	    cfg->msix.msix_ctrl, 2);
1877 
1878 	/* Update counts of alloc'd messages. */
1879 	cfg->msix.msix_alloc = actual;
1880 	cfg->msix.msix_table_len = actual;
1881 	*count = actual;
1882 	return (0);
1883 }
1884 
1885 /*
1886  * By default, pci_alloc_msix() will assign the allocated IRQ
1887  * resources consecutively to the first N messages in the MSI-X table.
1888  * However, device drivers may want to use different layouts if they
1889  * either receive fewer messages than they asked for, or they wish to
1890  * populate the MSI-X table sparsely.  This method allows the driver
1891  * to specify what layout it wants.  It must be called after a
1892  * successful pci_alloc_msix() but before any of the associated
1893  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1894  *
1895  * The 'vectors' array contains 'count' message vectors.  The array
1896  * maps directly to the MSI-X table in that index 0 in the array
1897  * specifies the vector for the first message in the MSI-X table, etc.
1898  * The vector value in each array index can either be 0 to indicate
1899  * that no vector should be assigned to a message slot, or it can be a
1900  * number from 1 to N (where N is the count returned from a
1901  * succcessful call to pci_alloc_msix()) to indicate which message
1902  * vector (IRQ) to be used for the corresponding message.
1903  *
1904  * On successful return, each message with a non-zero vector will have
1905  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1906  * 1.  Additionally, if any of the IRQs allocated via the previous
1907  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1908  * will be freed back to the system automatically.
1909  *
1910  * For example, suppose a driver has a MSI-X table with 6 messages and
1911  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1912  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1913  * C.  After the call to pci_alloc_msix(), the device will be setup to
1914  * have an MSI-X table of ABC--- (where - means no vector assigned).
1915  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1916  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1917  * be freed back to the system.  This device will also have valid
1918  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1919  *
1920  * In any case, the SYS_RES_IRQ rid X will always map to the message
1921  * at MSI-X table index X - 1 and will only be valid if a vector is
1922  * assigned to that table entry.
1923  */
1924 int
1925 pci_remap_msix_method(device_t dev, device_t child, int count,
1926     const u_int *vectors)
1927 {
1928 	struct pci_devinfo *dinfo = device_get_ivars(child);
1929 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1930 	struct resource_list_entry *rle;
1931 	int i, irq, j, *used;
1932 
1933 	/*
1934 	 * Have to have at least one message in the table but the
1935 	 * table can't be bigger than the actual MSI-X table in the
1936 	 * device.
1937 	 */
1938 	if (count == 0 || count > msix->msix_msgnum)
1939 		return (EINVAL);
1940 
1941 	/* Sanity check the vectors. */
1942 	for (i = 0; i < count; i++)
1943 		if (vectors[i] > msix->msix_alloc)
1944 			return (EINVAL);
1945 
1946 	/*
1947 	 * Make sure there aren't any holes in the vectors to be used.
1948 	 * It's a big pain to support it, and it doesn't really make
1949 	 * sense anyway.  Also, at least one vector must be used.
1950 	 */
1951 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1952 	    M_ZERO);
1953 	for (i = 0; i < count; i++)
1954 		if (vectors[i] != 0)
1955 			used[vectors[i] - 1] = 1;
1956 	for (i = 0; i < msix->msix_alloc - 1; i++)
1957 		if (used[i] == 0 && used[i + 1] == 1) {
1958 			free(used, M_DEVBUF);
1959 			return (EINVAL);
1960 		}
1961 	if (used[0] != 1) {
1962 		free(used, M_DEVBUF);
1963 		return (EINVAL);
1964 	}
1965 
1966 	/* Make sure none of the resources are allocated. */
1967 	for (i = 0; i < msix->msix_table_len; i++) {
1968 		if (msix->msix_table[i].mte_vector == 0)
1969 			continue;
1970 		if (msix->msix_table[i].mte_handlers > 0) {
1971 			free(used, M_DEVBUF);
1972 			return (EBUSY);
1973 		}
1974 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1975 		KASSERT(rle != NULL, ("missing resource"));
1976 		if (rle->res != NULL) {
1977 			free(used, M_DEVBUF);
1978 			return (EBUSY);
1979 		}
1980 	}
1981 
1982 	/* Free the existing resource list entries. */
1983 	for (i = 0; i < msix->msix_table_len; i++) {
1984 		if (msix->msix_table[i].mte_vector == 0)
1985 			continue;
1986 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1987 	}
1988 
1989 	/*
1990 	 * Build the new virtual table keeping track of which vectors are
1991 	 * used.
1992 	 */
1993 	free(msix->msix_table, M_DEVBUF);
1994 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1995 	    M_DEVBUF, M_WAITOK | M_ZERO);
1996 	for (i = 0; i < count; i++)
1997 		msix->msix_table[i].mte_vector = vectors[i];
1998 	msix->msix_table_len = count;
1999 
2000 	/* Free any unused IRQs and resize the vectors array if necessary. */
2001 	j = msix->msix_alloc - 1;
2002 	if (used[j] == 0) {
2003 		struct msix_vector *vec;
2004 
2005 		while (used[j] == 0) {
2006 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2007 			    msix->msix_vectors[j].mv_irq);
2008 			j--;
2009 		}
2010 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2011 		    M_WAITOK);
2012 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2013 		    (j + 1));
2014 		free(msix->msix_vectors, M_DEVBUF);
2015 		msix->msix_vectors = vec;
2016 		msix->msix_alloc = j + 1;
2017 	}
2018 	free(used, M_DEVBUF);
2019 
2020 	/* Map the IRQs onto the rids. */
2021 	for (i = 0; i < count; i++) {
2022 		if (vectors[i] == 0)
2023 			continue;
2024 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2025 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2026 		    irq, 1);
2027 	}
2028 
2029 	if (bootverbose) {
2030 		device_printf(child, "Remapped MSI-X IRQs as: ");
2031 		for (i = 0; i < count; i++) {
2032 			if (i != 0)
2033 				printf(", ");
2034 			if (vectors[i] == 0)
2035 				printf("---");
2036 			else
2037 				printf("%d",
2038 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2039 		}
2040 		printf("\n");
2041 	}
2042 
2043 	return (0);
2044 }
2045 
2046 static int
2047 pci_release_msix(device_t dev, device_t child)
2048 {
2049 	struct pci_devinfo *dinfo = device_get_ivars(child);
2050 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2051 	struct resource_list_entry *rle;
2052 	int i;
2053 
2054 	/* Do we have any messages to release? */
2055 	if (msix->msix_alloc == 0)
2056 		return (ENODEV);
2057 
2058 	/* Make sure none of the resources are allocated. */
2059 	for (i = 0; i < msix->msix_table_len; i++) {
2060 		if (msix->msix_table[i].mte_vector == 0)
2061 			continue;
2062 		if (msix->msix_table[i].mte_handlers > 0)
2063 			return (EBUSY);
2064 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2065 		KASSERT(rle != NULL, ("missing resource"));
2066 		if (rle->res != NULL)
2067 			return (EBUSY);
2068 	}
2069 
2070 	/* Update control register to disable MSI-X. */
2071 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2072 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2073 	    msix->msix_ctrl, 2);
2074 
2075 	/* Free the resource list entries. */
2076 	for (i = 0; i < msix->msix_table_len; i++) {
2077 		if (msix->msix_table[i].mte_vector == 0)
2078 			continue;
2079 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2080 	}
2081 	free(msix->msix_table, M_DEVBUF);
2082 	msix->msix_table_len = 0;
2083 
2084 	/* Release the IRQs. */
2085 	for (i = 0; i < msix->msix_alloc; i++)
2086 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2087 		    msix->msix_vectors[i].mv_irq);
2088 	free(msix->msix_vectors, M_DEVBUF);
2089 	msix->msix_alloc = 0;
2090 	return (0);
2091 }
2092 
2093 /*
2094  * Return the max supported MSI-X messages this device supports.
2095  * Basically, assuming the MD code can alloc messages, this function
2096  * should return the maximum value that pci_alloc_msix() can return.
2097  * Thus, it is subject to the tunables, etc.
2098  */
2099 int
2100 pci_msix_count_method(device_t dev, device_t child)
2101 {
2102 	struct pci_devinfo *dinfo = device_get_ivars(child);
2103 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2104 
2105 	if (pci_do_msix && msix->msix_location != 0)
2106 		return (msix->msix_msgnum);
2107 	return (0);
2108 }
2109 
2110 int
2111 pci_msix_pba_bar_method(device_t dev, device_t child)
2112 {
2113 	struct pci_devinfo *dinfo = device_get_ivars(child);
2114 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2115 
2116 	if (pci_do_msix && msix->msix_location != 0)
2117 		return (msix->msix_pba_bar);
2118 	return (-1);
2119 }
2120 
2121 int
2122 pci_msix_table_bar_method(device_t dev, device_t child)
2123 {
2124 	struct pci_devinfo *dinfo = device_get_ivars(child);
2125 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2126 
2127 	if (pci_do_msix && msix->msix_location != 0)
2128 		return (msix->msix_table_bar);
2129 	return (-1);
2130 }
2131 
2132 /*
2133  * HyperTransport MSI mapping control
2134  */
2135 void
2136 pci_ht_map_msi(device_t dev, uint64_t addr)
2137 {
2138 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2139 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2140 
2141 	if (!ht->ht_msimap)
2142 		return;
2143 
2144 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2145 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2146 		/* Enable MSI -> HT mapping. */
2147 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2148 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2149 		    ht->ht_msictrl, 2);
2150 	}
2151 
2152 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2153 		/* Disable MSI -> HT mapping. */
2154 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2155 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2156 		    ht->ht_msictrl, 2);
2157 	}
2158 }
2159 
2160 int
2161 pci_get_max_payload(device_t dev)
2162 {
2163 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2164 	int cap;
2165 	uint16_t val;
2166 
2167 	cap = dinfo->cfg.pcie.pcie_location;
2168 	if (cap == 0)
2169 		return (0);
2170 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2171 	val &= PCIEM_CTL_MAX_PAYLOAD;
2172 	val >>= 5;
2173 	return (1 << (val + 7));
2174 }
2175 
2176 int
2177 pci_get_max_read_req(device_t dev)
2178 {
2179 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2180 	int cap;
2181 	uint16_t val;
2182 
2183 	cap = dinfo->cfg.pcie.pcie_location;
2184 	if (cap == 0)
2185 		return (0);
2186 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2187 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2188 	val >>= 12;
2189 	return (1 << (val + 7));
2190 }
2191 
2192 int
2193 pci_set_max_read_req(device_t dev, int size)
2194 {
2195 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2196 	int cap;
2197 	uint16_t val;
2198 
2199 	cap = dinfo->cfg.pcie.pcie_location;
2200 	if (cap == 0)
2201 		return (0);
2202 	if (size < 128)
2203 		size = 128;
2204 	if (size > 4096)
2205 		size = 4096;
2206 	size = (1 << (fls(size) - 1));
2207 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2208 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2209 	val |= (fls(size) - 8) << 12;
2210 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2211 	return (size);
2212 }
2213 
2214 uint32_t
2215 pcie_read_config(device_t dev, int reg, int width)
2216 {
2217 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2218 	int cap;
2219 
2220 	cap = dinfo->cfg.pcie.pcie_location;
2221 	if (cap == 0) {
2222 		if (width == 2)
2223 			return (0xffff);
2224 		return (0xffffffff);
2225 	}
2226 
2227 	return (pci_read_config(dev, cap + reg, width));
2228 }
2229 
2230 void
2231 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2232 {
2233 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2234 	int cap;
2235 
2236 	cap = dinfo->cfg.pcie.pcie_location;
2237 	if (cap == 0)
2238 		return;
2239 	pci_write_config(dev, cap + reg, value, width);
2240 }
2241 
2242 /*
2243  * Adjusts a PCI-e capability register by clearing the bits in mask
2244  * and setting the bits in (value & mask).  Bits not set in mask are
2245  * not adjusted.
2246  *
2247  * Returns the old value on success or all ones on failure.
2248  */
2249 uint32_t
2250 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2251     int width)
2252 {
2253 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2254 	uint32_t old, new;
2255 	int cap;
2256 
2257 	cap = dinfo->cfg.pcie.pcie_location;
2258 	if (cap == 0) {
2259 		if (width == 2)
2260 			return (0xffff);
2261 		return (0xffffffff);
2262 	}
2263 
2264 	old = pci_read_config(dev, cap + reg, width);
2265 	new = old & ~mask;
2266 	new |= (value & mask);
2267 	pci_write_config(dev, cap + reg, new, width);
2268 	return (old);
2269 }
2270 
2271 /*
2272  * Support for MSI message signalled interrupts.
2273  */
2274 void
2275 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2276     uint16_t data)
2277 {
2278 	struct pci_devinfo *dinfo = device_get_ivars(child);
2279 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2280 
2281 	/* Write data and address values. */
2282 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2283 	    address & 0xffffffff, 4);
2284 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2285 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2286 		    address >> 32, 4);
2287 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2288 		    data, 2);
2289 	} else
2290 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2291 		    2);
2292 
2293 	/* Enable MSI in the control register. */
2294 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2295 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2296 	    msi->msi_ctrl, 2);
2297 
2298 	/* Enable MSI -> HT mapping. */
2299 	pci_ht_map_msi(child, address);
2300 }
2301 
2302 void
2303 pci_disable_msi_method(device_t dev, device_t child)
2304 {
2305 	struct pci_devinfo *dinfo = device_get_ivars(child);
2306 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2307 
2308 	/* Disable MSI -> HT mapping. */
2309 	pci_ht_map_msi(child, 0);
2310 
2311 	/* Disable MSI in the control register. */
2312 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2313 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2314 	    msi->msi_ctrl, 2);
2315 }
2316 
2317 /*
2318  * Restore MSI registers during resume.  If MSI is enabled then
2319  * restore the data and address registers in addition to the control
2320  * register.
2321  */
2322 static void
2323 pci_resume_msi(device_t dev)
2324 {
2325 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2326 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2327 	uint64_t address;
2328 	uint16_t data;
2329 
2330 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2331 		address = msi->msi_addr;
2332 		data = msi->msi_data;
2333 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2334 		    address & 0xffffffff, 4);
2335 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2336 			pci_write_config(dev, msi->msi_location +
2337 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2338 			pci_write_config(dev, msi->msi_location +
2339 			    PCIR_MSI_DATA_64BIT, data, 2);
2340 		} else
2341 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2342 			    data, 2);
2343 	}
2344 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2345 	    2);
2346 }
2347 
2348 static int
2349 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2350 {
2351 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2352 	pcicfgregs *cfg = &dinfo->cfg;
2353 	struct resource_list_entry *rle;
2354 	struct msix_table_entry *mte;
2355 	struct msix_vector *mv;
2356 	uint64_t addr;
2357 	uint32_t data;
2358 	int error, i, j;
2359 
2360 	/*
2361 	 * Handle MSI first.  We try to find this IRQ among our list
2362 	 * of MSI IRQs.  If we find it, we request updated address and
2363 	 * data registers and apply the results.
2364 	 */
2365 	if (cfg->msi.msi_alloc > 0) {
2366 
2367 		/* If we don't have any active handlers, nothing to do. */
2368 		if (cfg->msi.msi_handlers == 0)
2369 			return (0);
2370 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2371 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2372 			    i + 1);
2373 			if (rle->start == irq) {
2374 				error = PCIB_MAP_MSI(device_get_parent(bus),
2375 				    dev, irq, &addr, &data);
2376 				if (error)
2377 					return (error);
2378 				pci_disable_msi(dev);
2379 				dinfo->cfg.msi.msi_addr = addr;
2380 				dinfo->cfg.msi.msi_data = data;
2381 				pci_enable_msi(dev, addr, data);
2382 				return (0);
2383 			}
2384 		}
2385 		return (ENOENT);
2386 	}
2387 
2388 	/*
2389 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2390 	 * we request the updated mapping info.  If that works, we go
2391 	 * through all the slots that use this IRQ and update them.
2392 	 */
2393 	if (cfg->msix.msix_alloc > 0) {
2394 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2395 			mv = &cfg->msix.msix_vectors[i];
2396 			if (mv->mv_irq == irq) {
2397 				error = PCIB_MAP_MSI(device_get_parent(bus),
2398 				    dev, irq, &addr, &data);
2399 				if (error)
2400 					return (error);
2401 				mv->mv_address = addr;
2402 				mv->mv_data = data;
2403 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2404 					mte = &cfg->msix.msix_table[j];
2405 					if (mte->mte_vector != i + 1)
2406 						continue;
2407 					if (mte->mte_handlers == 0)
2408 						continue;
2409 					pci_mask_msix(dev, j);
2410 					pci_enable_msix(dev, j, addr, data);
2411 					pci_unmask_msix(dev, j);
2412 				}
2413 			}
2414 		}
2415 		return (ENOENT);
2416 	}
2417 
2418 	return (ENOENT);
2419 }
2420 
2421 /*
2422  * Returns true if the specified device is blacklisted because MSI
2423  * doesn't work.
2424  */
2425 int
2426 pci_msi_device_blacklisted(device_t dev)
2427 {
2428 
2429 	if (!pci_honor_msi_blacklist)
2430 		return (0);
2431 
2432 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2433 }
2434 
2435 /*
2436  * Determine if MSI is blacklisted globally on this system.  Currently,
2437  * we just check for blacklisted chipsets as represented by the
2438  * host-PCI bridge at device 0:0:0.  In the future, it may become
2439  * necessary to check other system attributes, such as the kenv values
2440  * that give the motherboard manufacturer and model number.
2441  */
2442 static int
2443 pci_msi_blacklisted(void)
2444 {
2445 	device_t dev;
2446 
2447 	if (!pci_honor_msi_blacklist)
2448 		return (0);
2449 
2450 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2451 	if (!(pcie_chipset || pcix_chipset)) {
2452 		if (vm_guest != VM_GUEST_NO) {
2453 			/*
2454 			 * Whitelist older chipsets in virtual
2455 			 * machines known to support MSI.
2456 			 */
2457 			dev = pci_find_bsf(0, 0, 0);
2458 			if (dev != NULL)
2459 				return (!pci_has_quirk(pci_get_devid(dev),
2460 					PCI_QUIRK_ENABLE_MSI_VM));
2461 		}
2462 		return (1);
2463 	}
2464 
2465 	dev = pci_find_bsf(0, 0, 0);
2466 	if (dev != NULL)
2467 		return (pci_msi_device_blacklisted(dev));
2468 	return (0);
2469 }
2470 
2471 /*
2472  * Returns true if the specified device is blacklisted because MSI-X
2473  * doesn't work.  Note that this assumes that if MSI doesn't work,
2474  * MSI-X doesn't either.
2475  */
2476 int
2477 pci_msix_device_blacklisted(device_t dev)
2478 {
2479 
2480 	if (!pci_honor_msi_blacklist)
2481 		return (0);
2482 
2483 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2484 		return (1);
2485 
2486 	return (pci_msi_device_blacklisted(dev));
2487 }
2488 
2489 /*
2490  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2491  * is blacklisted, assume that MSI-X is as well.  Check for additional
2492  * chipsets where MSI works but MSI-X does not.
2493  */
2494 static int
2495 pci_msix_blacklisted(void)
2496 {
2497 	device_t dev;
2498 
2499 	if (!pci_honor_msi_blacklist)
2500 		return (0);
2501 
2502 	dev = pci_find_bsf(0, 0, 0);
2503 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2504 	    PCI_QUIRK_DISABLE_MSIX))
2505 		return (1);
2506 
2507 	return (pci_msi_blacklisted());
2508 }
2509 
2510 /*
2511  * Attempt to allocate *count MSI messages.  The actual number allocated is
2512  * returned in *count.  After this function returns, each message will be
2513  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2514  */
2515 int
2516 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2517 {
2518 	struct pci_devinfo *dinfo = device_get_ivars(child);
2519 	pcicfgregs *cfg = &dinfo->cfg;
2520 	struct resource_list_entry *rle;
2521 	int actual, error, i, irqs[32];
2522 	uint16_t ctrl;
2523 
2524 	/* Don't let count == 0 get us into trouble. */
2525 	if (*count == 0)
2526 		return (EINVAL);
2527 
2528 	/* If rid 0 is allocated, then fail. */
2529 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2530 	if (rle != NULL && rle->res != NULL)
2531 		return (ENXIO);
2532 
2533 	/* Already have allocated messages? */
2534 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2535 		return (ENXIO);
2536 
2537 	/* If MSI is blacklisted for this system, fail. */
2538 	if (pci_msi_blacklisted())
2539 		return (ENXIO);
2540 
2541 	/* MSI capability present? */
2542 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2543 		return (ENODEV);
2544 
2545 	if (bootverbose)
2546 		device_printf(child,
2547 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2548 		    *count, cfg->msi.msi_msgnum);
2549 
2550 	/* Don't ask for more than the device supports. */
2551 	actual = min(*count, cfg->msi.msi_msgnum);
2552 
2553 	/* Don't ask for more than 32 messages. */
2554 	actual = min(actual, 32);
2555 
2556 	/* MSI requires power of 2 number of messages. */
2557 	if (!powerof2(actual))
2558 		return (EINVAL);
2559 
2560 	for (;;) {
2561 		/* Try to allocate N messages. */
2562 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2563 		    actual, irqs);
2564 		if (error == 0)
2565 			break;
2566 		if (actual == 1)
2567 			return (error);
2568 
2569 		/* Try N / 2. */
2570 		actual >>= 1;
2571 	}
2572 
2573 	/*
2574 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2575 	 * resources in the irqs[] array, so add new resources
2576 	 * starting at rid 1.
2577 	 */
2578 	for (i = 0; i < actual; i++)
2579 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2580 		    irqs[i], irqs[i], 1);
2581 
2582 	if (bootverbose) {
2583 		if (actual == 1)
2584 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2585 		else {
2586 			int run;
2587 
2588 			/*
2589 			 * Be fancy and try to print contiguous runs
2590 			 * of IRQ values as ranges.  'run' is true if
2591 			 * we are in a range.
2592 			 */
2593 			device_printf(child, "using IRQs %d", irqs[0]);
2594 			run = 0;
2595 			for (i = 1; i < actual; i++) {
2596 
2597 				/* Still in a run? */
2598 				if (irqs[i] == irqs[i - 1] + 1) {
2599 					run = 1;
2600 					continue;
2601 				}
2602 
2603 				/* Finish previous range. */
2604 				if (run) {
2605 					printf("-%d", irqs[i - 1]);
2606 					run = 0;
2607 				}
2608 
2609 				/* Start new range. */
2610 				printf(",%d", irqs[i]);
2611 			}
2612 
2613 			/* Unfinished range? */
2614 			if (run)
2615 				printf("-%d", irqs[actual - 1]);
2616 			printf(" for MSI\n");
2617 		}
2618 	}
2619 
2620 	/* Update control register with actual count. */
2621 	ctrl = cfg->msi.msi_ctrl;
2622 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2623 	ctrl |= (ffs(actual) - 1) << 4;
2624 	cfg->msi.msi_ctrl = ctrl;
2625 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2626 
2627 	/* Update counts of alloc'd messages. */
2628 	cfg->msi.msi_alloc = actual;
2629 	cfg->msi.msi_handlers = 0;
2630 	*count = actual;
2631 	return (0);
2632 }
2633 
2634 /* Release the MSI messages associated with this device. */
2635 int
2636 pci_release_msi_method(device_t dev, device_t child)
2637 {
2638 	struct pci_devinfo *dinfo = device_get_ivars(child);
2639 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2640 	struct resource_list_entry *rle;
2641 	int error, i, irqs[32];
2642 
2643 	/* Try MSI-X first. */
2644 	error = pci_release_msix(dev, child);
2645 	if (error != ENODEV)
2646 		return (error);
2647 
2648 	/* Do we have any messages to release? */
2649 	if (msi->msi_alloc == 0)
2650 		return (ENODEV);
2651 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2652 
2653 	/* Make sure none of the resources are allocated. */
2654 	if (msi->msi_handlers > 0)
2655 		return (EBUSY);
2656 	for (i = 0; i < msi->msi_alloc; i++) {
2657 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2658 		KASSERT(rle != NULL, ("missing MSI resource"));
2659 		if (rle->res != NULL)
2660 			return (EBUSY);
2661 		irqs[i] = rle->start;
2662 	}
2663 
2664 	/* Update control register with 0 count. */
2665 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2666 	    ("%s: MSI still enabled", __func__));
2667 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2668 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2669 	    msi->msi_ctrl, 2);
2670 
2671 	/* Release the messages. */
2672 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2673 	for (i = 0; i < msi->msi_alloc; i++)
2674 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2675 
2676 	/* Update alloc count. */
2677 	msi->msi_alloc = 0;
2678 	msi->msi_addr = 0;
2679 	msi->msi_data = 0;
2680 	return (0);
2681 }
2682 
2683 /*
2684  * Return the max supported MSI messages this device supports.
2685  * Basically, assuming the MD code can alloc messages, this function
2686  * should return the maximum value that pci_alloc_msi() can return.
2687  * Thus, it is subject to the tunables, etc.
2688  */
2689 int
2690 pci_msi_count_method(device_t dev, device_t child)
2691 {
2692 	struct pci_devinfo *dinfo = device_get_ivars(child);
2693 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2694 
2695 	if (pci_do_msi && msi->msi_location != 0)
2696 		return (msi->msi_msgnum);
2697 	return (0);
2698 }
2699 
2700 /* free pcicfgregs structure and all depending data structures */
2701 
2702 int
2703 pci_freecfg(struct pci_devinfo *dinfo)
2704 {
2705 	struct devlist *devlist_head;
2706 	struct pci_map *pm, *next;
2707 	int i;
2708 
2709 	devlist_head = &pci_devq;
2710 
2711 	if (dinfo->cfg.vpd.vpd_reg) {
2712 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2713 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2714 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2715 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2716 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2717 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2718 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2719 	}
2720 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2721 		free(pm, M_DEVBUF);
2722 	}
2723 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2724 	free(dinfo, M_DEVBUF);
2725 
2726 	/* increment the generation count */
2727 	pci_generation++;
2728 
2729 	/* we're losing one device */
2730 	pci_numdevs--;
2731 	return (0);
2732 }
2733 
2734 /*
2735  * PCI power manangement
2736  */
2737 int
2738 pci_set_powerstate_method(device_t dev, device_t child, int state)
2739 {
2740 	struct pci_devinfo *dinfo = device_get_ivars(child);
2741 	pcicfgregs *cfg = &dinfo->cfg;
2742 	uint16_t status;
2743 	int oldstate, highest, delay;
2744 
2745 	if (cfg->pp.pp_cap == 0)
2746 		return (EOPNOTSUPP);
2747 
2748 	/*
2749 	 * Optimize a no state change request away.  While it would be OK to
2750 	 * write to the hardware in theory, some devices have shown odd
2751 	 * behavior when going from D3 -> D3.
2752 	 */
2753 	oldstate = pci_get_powerstate(child);
2754 	if (oldstate == state)
2755 		return (0);
2756 
2757 	/*
2758 	 * The PCI power management specification states that after a state
2759 	 * transition between PCI power states, system software must
2760 	 * guarantee a minimal delay before the function accesses the device.
2761 	 * Compute the worst case delay that we need to guarantee before we
2762 	 * access the device.  Many devices will be responsive much more
2763 	 * quickly than this delay, but there are some that don't respond
2764 	 * instantly to state changes.  Transitions to/from D3 state require
2765 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2766 	 * is done below with DELAY rather than a sleeper function because
2767 	 * this function can be called from contexts where we cannot sleep.
2768 	 */
2769 	highest = (oldstate > state) ? oldstate : state;
2770 	if (highest == PCI_POWERSTATE_D3)
2771 	    delay = 10000;
2772 	else if (highest == PCI_POWERSTATE_D2)
2773 	    delay = 200;
2774 	else
2775 	    delay = 0;
2776 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2777 	    & ~PCIM_PSTAT_DMASK;
2778 	switch (state) {
2779 	case PCI_POWERSTATE_D0:
2780 		status |= PCIM_PSTAT_D0;
2781 		break;
2782 	case PCI_POWERSTATE_D1:
2783 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2784 			return (EOPNOTSUPP);
2785 		status |= PCIM_PSTAT_D1;
2786 		break;
2787 	case PCI_POWERSTATE_D2:
2788 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2789 			return (EOPNOTSUPP);
2790 		status |= PCIM_PSTAT_D2;
2791 		break;
2792 	case PCI_POWERSTATE_D3:
2793 		status |= PCIM_PSTAT_D3;
2794 		break;
2795 	default:
2796 		return (EINVAL);
2797 	}
2798 
2799 	if (bootverbose)
2800 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2801 		    state);
2802 
2803 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2804 	if (delay)
2805 		DELAY(delay);
2806 	return (0);
2807 }
2808 
2809 int
2810 pci_get_powerstate_method(device_t dev, device_t child)
2811 {
2812 	struct pci_devinfo *dinfo = device_get_ivars(child);
2813 	pcicfgregs *cfg = &dinfo->cfg;
2814 	uint16_t status;
2815 	int result;
2816 
2817 	if (cfg->pp.pp_cap != 0) {
2818 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2819 		switch (status & PCIM_PSTAT_DMASK) {
2820 		case PCIM_PSTAT_D0:
2821 			result = PCI_POWERSTATE_D0;
2822 			break;
2823 		case PCIM_PSTAT_D1:
2824 			result = PCI_POWERSTATE_D1;
2825 			break;
2826 		case PCIM_PSTAT_D2:
2827 			result = PCI_POWERSTATE_D2;
2828 			break;
2829 		case PCIM_PSTAT_D3:
2830 			result = PCI_POWERSTATE_D3;
2831 			break;
2832 		default:
2833 			result = PCI_POWERSTATE_UNKNOWN;
2834 			break;
2835 		}
2836 	} else {
2837 		/* No support, device is always at D0 */
2838 		result = PCI_POWERSTATE_D0;
2839 	}
2840 	return (result);
2841 }
2842 
2843 /*
2844  * Some convenience functions for PCI device drivers.
2845  */
2846 
2847 static __inline void
2848 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2849 {
2850 	uint16_t	command;
2851 
2852 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2853 	command |= bit;
2854 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2855 }
2856 
2857 static __inline void
2858 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2859 {
2860 	uint16_t	command;
2861 
2862 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2863 	command &= ~bit;
2864 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2865 }
2866 
2867 int
2868 pci_enable_busmaster_method(device_t dev, device_t child)
2869 {
2870 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2871 	return (0);
2872 }
2873 
2874 int
2875 pci_disable_busmaster_method(device_t dev, device_t child)
2876 {
2877 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2878 	return (0);
2879 }
2880 
2881 int
2882 pci_enable_io_method(device_t dev, device_t child, int space)
2883 {
2884 	uint16_t bit;
2885 
2886 	switch(space) {
2887 	case SYS_RES_IOPORT:
2888 		bit = PCIM_CMD_PORTEN;
2889 		break;
2890 	case SYS_RES_MEMORY:
2891 		bit = PCIM_CMD_MEMEN;
2892 		break;
2893 	default:
2894 		return (EINVAL);
2895 	}
2896 	pci_set_command_bit(dev, child, bit);
2897 	return (0);
2898 }
2899 
2900 int
2901 pci_disable_io_method(device_t dev, device_t child, int space)
2902 {
2903 	uint16_t bit;
2904 
2905 	switch(space) {
2906 	case SYS_RES_IOPORT:
2907 		bit = PCIM_CMD_PORTEN;
2908 		break;
2909 	case SYS_RES_MEMORY:
2910 		bit = PCIM_CMD_MEMEN;
2911 		break;
2912 	default:
2913 		return (EINVAL);
2914 	}
2915 	pci_clear_command_bit(dev, child, bit);
2916 	return (0);
2917 }
2918 
2919 /*
2920  * New style pci driver.  Parent device is either a pci-host-bridge or a
2921  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2922  */
2923 
2924 void
2925 pci_print_verbose(struct pci_devinfo *dinfo)
2926 {
2927 
2928 	if (bootverbose) {
2929 		pcicfgregs *cfg = &dinfo->cfg;
2930 
2931 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2932 		    cfg->vendor, cfg->device, cfg->revid);
2933 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2934 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2935 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2936 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2937 		    cfg->mfdev);
2938 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2939 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2940 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2941 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2942 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2943 		if (cfg->intpin > 0)
2944 			printf("\tintpin=%c, irq=%d\n",
2945 			    cfg->intpin +'a' -1, cfg->intline);
2946 		if (cfg->pp.pp_cap) {
2947 			uint16_t status;
2948 
2949 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2950 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2951 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2952 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2953 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2954 			    status & PCIM_PSTAT_DMASK);
2955 		}
2956 		if (cfg->msi.msi_location) {
2957 			int ctrl;
2958 
2959 			ctrl = cfg->msi.msi_ctrl;
2960 			printf("\tMSI supports %d message%s%s%s\n",
2961 			    cfg->msi.msi_msgnum,
2962 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2963 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2964 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2965 		}
2966 		if (cfg->msix.msix_location) {
2967 			printf("\tMSI-X supports %d message%s ",
2968 			    cfg->msix.msix_msgnum,
2969 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2970 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2971 				printf("in map 0x%x\n",
2972 				    cfg->msix.msix_table_bar);
2973 			else
2974 				printf("in maps 0x%x and 0x%x\n",
2975 				    cfg->msix.msix_table_bar,
2976 				    cfg->msix.msix_pba_bar);
2977 		}
2978 	}
2979 }
2980 
2981 static int
2982 pci_porten(device_t dev)
2983 {
2984 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2985 }
2986 
2987 static int
2988 pci_memen(device_t dev)
2989 {
2990 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2991 }
2992 
2993 void
2994 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2995     int *bar64)
2996 {
2997 	struct pci_devinfo *dinfo;
2998 	pci_addr_t map, testval;
2999 	int ln2range;
3000 	uint16_t cmd;
3001 
3002 	/*
3003 	 * The device ROM BAR is special.  It is always a 32-bit
3004 	 * memory BAR.  Bit 0 is special and should not be set when
3005 	 * sizing the BAR.
3006 	 */
3007 	dinfo = device_get_ivars(dev);
3008 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3009 		map = pci_read_config(dev, reg, 4);
3010 		pci_write_config(dev, reg, 0xfffffffe, 4);
3011 		testval = pci_read_config(dev, reg, 4);
3012 		pci_write_config(dev, reg, map, 4);
3013 		*mapp = map;
3014 		*testvalp = testval;
3015 		if (bar64 != NULL)
3016 			*bar64 = 0;
3017 		return;
3018 	}
3019 
3020 	map = pci_read_config(dev, reg, 4);
3021 	ln2range = pci_maprange(map);
3022 	if (ln2range == 64)
3023 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3024 
3025 	/*
3026 	 * Disable decoding via the command register before
3027 	 * determining the BAR's length since we will be placing it in
3028 	 * a weird state.
3029 	 */
3030 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3031 	pci_write_config(dev, PCIR_COMMAND,
3032 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3033 
3034 	/*
3035 	 * Determine the BAR's length by writing all 1's.  The bottom
3036 	 * log_2(size) bits of the BAR will stick as 0 when we read
3037 	 * the value back.
3038 	 *
3039 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3040 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3041 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3042 	 *
3043 	 * Writes to both registers must be performed before attempting to
3044 	 * read back the size value.
3045 	 */
3046 	testval = 0;
3047 	pci_write_config(dev, reg, 0xffffffff, 4);
3048 	if (ln2range == 64) {
3049 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3050 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3051 	}
3052 	testval |= pci_read_config(dev, reg, 4);
3053 
3054 	/*
3055 	 * Restore the original value of the BAR.  We may have reprogrammed
3056 	 * the BAR of the low-level console device and when booting verbose,
3057 	 * we need the console device addressable.
3058 	 */
3059 	pci_write_config(dev, reg, map, 4);
3060 	if (ln2range == 64)
3061 		pci_write_config(dev, reg + 4, map >> 32, 4);
3062 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3063 
3064 	*mapp = map;
3065 	*testvalp = testval;
3066 	if (bar64 != NULL)
3067 		*bar64 = (ln2range == 64);
3068 }
3069 
3070 static void
3071 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3072 {
3073 	struct pci_devinfo *dinfo;
3074 	int ln2range;
3075 
3076 	/* The device ROM BAR is always a 32-bit memory BAR. */
3077 	dinfo = device_get_ivars(dev);
3078 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3079 		ln2range = 32;
3080 	else
3081 		ln2range = pci_maprange(pm->pm_value);
3082 	pci_write_config(dev, pm->pm_reg, base, 4);
3083 	if (ln2range == 64)
3084 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3085 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3086 	if (ln2range == 64)
3087 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3088 		    pm->pm_reg + 4, 4) << 32;
3089 }
3090 
3091 struct pci_map *
3092 pci_find_bar(device_t dev, int reg)
3093 {
3094 	struct pci_devinfo *dinfo;
3095 	struct pci_map *pm;
3096 
3097 	dinfo = device_get_ivars(dev);
3098 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3099 		if (pm->pm_reg == reg)
3100 			return (pm);
3101 	}
3102 	return (NULL);
3103 }
3104 
3105 int
3106 pci_bar_enabled(device_t dev, struct pci_map *pm)
3107 {
3108 	struct pci_devinfo *dinfo;
3109 	uint16_t cmd;
3110 
3111 	dinfo = device_get_ivars(dev);
3112 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3113 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3114 		return (0);
3115 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3116 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3117 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3118 	else
3119 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3120 }
3121 
3122 struct pci_map *
3123 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3124 {
3125 	struct pci_devinfo *dinfo;
3126 	struct pci_map *pm, *prev;
3127 
3128 	dinfo = device_get_ivars(dev);
3129 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3130 	pm->pm_reg = reg;
3131 	pm->pm_value = value;
3132 	pm->pm_size = size;
3133 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3134 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3135 		    reg));
3136 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3137 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3138 			break;
3139 	}
3140 	if (prev != NULL)
3141 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3142 	else
3143 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3144 	return (pm);
3145 }
3146 
3147 static void
3148 pci_restore_bars(device_t dev)
3149 {
3150 	struct pci_devinfo *dinfo;
3151 	struct pci_map *pm;
3152 	int ln2range;
3153 
3154 	dinfo = device_get_ivars(dev);
3155 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3156 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3157 			ln2range = 32;
3158 		else
3159 			ln2range = pci_maprange(pm->pm_value);
3160 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3161 		if (ln2range == 64)
3162 			pci_write_config(dev, pm->pm_reg + 4,
3163 			    pm->pm_value >> 32, 4);
3164 	}
3165 }
3166 
3167 /*
3168  * Add a resource based on a pci map register. Return 1 if the map
3169  * register is a 32bit map register or 2 if it is a 64bit register.
3170  */
3171 static int
3172 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3173     int force, int prefetch)
3174 {
3175 	struct pci_map *pm;
3176 	pci_addr_t base, map, testval;
3177 	pci_addr_t start, end, count;
3178 	int barlen, basezero, flags, maprange, mapsize, type;
3179 	uint16_t cmd;
3180 	struct resource *res;
3181 
3182 	/*
3183 	 * The BAR may already exist if the device is a CardBus card
3184 	 * whose CIS is stored in this BAR.
3185 	 */
3186 	pm = pci_find_bar(dev, reg);
3187 	if (pm != NULL) {
3188 		maprange = pci_maprange(pm->pm_value);
3189 		barlen = maprange == 64 ? 2 : 1;
3190 		return (barlen);
3191 	}
3192 
3193 	pci_read_bar(dev, reg, &map, &testval, NULL);
3194 	if (PCI_BAR_MEM(map)) {
3195 		type = SYS_RES_MEMORY;
3196 		if (map & PCIM_BAR_MEM_PREFETCH)
3197 			prefetch = 1;
3198 	} else
3199 		type = SYS_RES_IOPORT;
3200 	mapsize = pci_mapsize(testval);
3201 	base = pci_mapbase(map);
3202 #ifdef __PCI_BAR_ZERO_VALID
3203 	basezero = 0;
3204 #else
3205 	basezero = base == 0;
3206 #endif
3207 	maprange = pci_maprange(map);
3208 	barlen = maprange == 64 ? 2 : 1;
3209 
3210 	/*
3211 	 * For I/O registers, if bottom bit is set, and the next bit up
3212 	 * isn't clear, we know we have a BAR that doesn't conform to the
3213 	 * spec, so ignore it.  Also, sanity check the size of the data
3214 	 * areas to the type of memory involved.  Memory must be at least
3215 	 * 16 bytes in size, while I/O ranges must be at least 4.
3216 	 */
3217 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3218 		return (barlen);
3219 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3220 	    (type == SYS_RES_IOPORT && mapsize < 2))
3221 		return (barlen);
3222 
3223 	/* Save a record of this BAR. */
3224 	pm = pci_add_bar(dev, reg, map, mapsize);
3225 	if (bootverbose) {
3226 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3227 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3228 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3229 			printf(", port disabled\n");
3230 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3231 			printf(", memory disabled\n");
3232 		else
3233 			printf(", enabled\n");
3234 	}
3235 
3236 	/*
3237 	 * If base is 0, then we have problems if this architecture does
3238 	 * not allow that.  It is best to ignore such entries for the
3239 	 * moment.  These will be allocated later if the driver specifically
3240 	 * requests them.  However, some removable buses look better when
3241 	 * all resources are allocated, so allow '0' to be overriden.
3242 	 *
3243 	 * Similarly treat maps whose values is the same as the test value
3244 	 * read back.  These maps have had all f's written to them by the
3245 	 * BIOS in an attempt to disable the resources.
3246 	 */
3247 	if (!force && (basezero || map == testval))
3248 		return (barlen);
3249 	if ((u_long)base != base) {
3250 		device_printf(bus,
3251 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3252 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3253 		    pci_get_function(dev), reg);
3254 		return (barlen);
3255 	}
3256 
3257 	/*
3258 	 * This code theoretically does the right thing, but has
3259 	 * undesirable side effects in some cases where peripherals
3260 	 * respond oddly to having these bits enabled.  Let the user
3261 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3262 	 * default).
3263 	 */
3264 	if (pci_enable_io_modes) {
3265 		/* Turn on resources that have been left off by a lazy BIOS */
3266 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3267 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3268 			cmd |= PCIM_CMD_PORTEN;
3269 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3270 		}
3271 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3272 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3273 			cmd |= PCIM_CMD_MEMEN;
3274 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3275 		}
3276 	} else {
3277 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3278 			return (barlen);
3279 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3280 			return (barlen);
3281 	}
3282 
3283 	count = (pci_addr_t)1 << mapsize;
3284 	flags = RF_ALIGNMENT_LOG2(mapsize);
3285 	if (prefetch)
3286 		flags |= RF_PREFETCHABLE;
3287 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3288 		start = 0;	/* Let the parent decide. */
3289 		end = ~0;
3290 	} else {
3291 		start = base;
3292 		end = base + count - 1;
3293 	}
3294 	resource_list_add(rl, type, reg, start, end, count);
3295 
3296 	/*
3297 	 * Try to allocate the resource for this BAR from our parent
3298 	 * so that this resource range is already reserved.  The
3299 	 * driver for this device will later inherit this resource in
3300 	 * pci_alloc_resource().
3301 	 */
3302 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3303 	    flags);
3304 	if ((pci_do_realloc_bars
3305 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3306 	    && res == NULL && (start != 0 || end != ~0)) {
3307 		/*
3308 		 * If the allocation fails, try to allocate a resource for
3309 		 * this BAR using any available range.  The firmware felt
3310 		 * it was important enough to assign a resource, so don't
3311 		 * disable decoding if we can help it.
3312 		 */
3313 		resource_list_delete(rl, type, reg);
3314 		resource_list_add(rl, type, reg, 0, ~0, count);
3315 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3316 		    count, flags);
3317 	}
3318 	if (res == NULL) {
3319 		/*
3320 		 * If the allocation fails, delete the resource list entry
3321 		 * and disable decoding for this device.
3322 		 *
3323 		 * If the driver requests this resource in the future,
3324 		 * pci_reserve_map() will try to allocate a fresh
3325 		 * resource range.
3326 		 */
3327 		resource_list_delete(rl, type, reg);
3328 		pci_disable_io(dev, type);
3329 		if (bootverbose)
3330 			device_printf(bus,
3331 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3332 			    pci_get_domain(dev), pci_get_bus(dev),
3333 			    pci_get_slot(dev), pci_get_function(dev), reg);
3334 	} else {
3335 		start = rman_get_start(res);
3336 		pci_write_bar(dev, pm, start);
3337 	}
3338 	return (barlen);
3339 }
3340 
3341 /*
3342  * For ATA devices we need to decide early what addressing mode to use.
3343  * Legacy demands that the primary and secondary ATA ports sits on the
3344  * same addresses that old ISA hardware did. This dictates that we use
3345  * those addresses and ignore the BAR's if we cannot set PCI native
3346  * addressing mode.
3347  */
3348 static void
3349 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3350     uint32_t prefetchmask)
3351 {
3352 	int rid, type, progif;
3353 #if 0
3354 	/* if this device supports PCI native addressing use it */
3355 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3356 	if ((progif & 0x8a) == 0x8a) {
3357 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3358 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3359 			printf("Trying ATA native PCI addressing mode\n");
3360 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3361 		}
3362 	}
3363 #endif
3364 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3365 	type = SYS_RES_IOPORT;
3366 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3367 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3368 		    prefetchmask & (1 << 0));
3369 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3370 		    prefetchmask & (1 << 1));
3371 	} else {
3372 		rid = PCIR_BAR(0);
3373 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3374 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3375 		    0x1f7, 8, 0);
3376 		rid = PCIR_BAR(1);
3377 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3378 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3379 		    0x3f6, 1, 0);
3380 	}
3381 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3382 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3383 		    prefetchmask & (1 << 2));
3384 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3385 		    prefetchmask & (1 << 3));
3386 	} else {
3387 		rid = PCIR_BAR(2);
3388 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3389 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3390 		    0x177, 8, 0);
3391 		rid = PCIR_BAR(3);
3392 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3393 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3394 		    0x376, 1, 0);
3395 	}
3396 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3397 	    prefetchmask & (1 << 4));
3398 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3399 	    prefetchmask & (1 << 5));
3400 }
3401 
3402 static void
3403 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3404 {
3405 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3406 	pcicfgregs *cfg = &dinfo->cfg;
3407 	char tunable_name[64];
3408 	int irq;
3409 
3410 	/* Has to have an intpin to have an interrupt. */
3411 	if (cfg->intpin == 0)
3412 		return;
3413 
3414 	/* Let the user override the IRQ with a tunable. */
3415 	irq = PCI_INVALID_IRQ;
3416 	snprintf(tunable_name, sizeof(tunable_name),
3417 	    "hw.pci%d.%d.%d.INT%c.irq",
3418 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3419 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3420 		irq = PCI_INVALID_IRQ;
3421 
3422 	/*
3423 	 * If we didn't get an IRQ via the tunable, then we either use the
3424 	 * IRQ value in the intline register or we ask the bus to route an
3425 	 * interrupt for us.  If force_route is true, then we only use the
3426 	 * value in the intline register if the bus was unable to assign an
3427 	 * IRQ.
3428 	 */
3429 	if (!PCI_INTERRUPT_VALID(irq)) {
3430 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3431 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3432 		if (!PCI_INTERRUPT_VALID(irq))
3433 			irq = cfg->intline;
3434 	}
3435 
3436 	/* If after all that we don't have an IRQ, just bail. */
3437 	if (!PCI_INTERRUPT_VALID(irq))
3438 		return;
3439 
3440 	/* Update the config register if it changed. */
3441 	if (irq != cfg->intline) {
3442 		cfg->intline = irq;
3443 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3444 	}
3445 
3446 	/* Add this IRQ as rid 0 interrupt resource. */
3447 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3448 }
3449 
3450 /* Perform early OHCI takeover from SMM. */
3451 static void
3452 ohci_early_takeover(device_t self)
3453 {
3454 	struct resource *res;
3455 	uint32_t ctl;
3456 	int rid;
3457 	int i;
3458 
3459 	rid = PCIR_BAR(0);
3460 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3461 	if (res == NULL)
3462 		return;
3463 
3464 	ctl = bus_read_4(res, OHCI_CONTROL);
3465 	if (ctl & OHCI_IR) {
3466 		if (bootverbose)
3467 			printf("ohci early: "
3468 			    "SMM active, request owner change\n");
3469 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3470 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3471 			DELAY(1000);
3472 			ctl = bus_read_4(res, OHCI_CONTROL);
3473 		}
3474 		if (ctl & OHCI_IR) {
3475 			if (bootverbose)
3476 				printf("ohci early: "
3477 				    "SMM does not respond, resetting\n");
3478 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3479 		}
3480 		/* Disable interrupts */
3481 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3482 	}
3483 
3484 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3485 }
3486 
3487 /* Perform early UHCI takeover from SMM. */
3488 static void
3489 uhci_early_takeover(device_t self)
3490 {
3491 	struct resource *res;
3492 	int rid;
3493 
3494 	/*
3495 	 * Set the PIRQD enable bit and switch off all the others. We don't
3496 	 * want legacy support to interfere with us XXX Does this also mean
3497 	 * that the BIOS won't touch the keyboard anymore if it is connected
3498 	 * to the ports of the root hub?
3499 	 */
3500 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3501 
3502 	/* Disable interrupts */
3503 	rid = PCI_UHCI_BASE_REG;
3504 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3505 	if (res != NULL) {
3506 		bus_write_2(res, UHCI_INTR, 0);
3507 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3508 	}
3509 }
3510 
3511 /* Perform early EHCI takeover from SMM. */
3512 static void
3513 ehci_early_takeover(device_t self)
3514 {
3515 	struct resource *res;
3516 	uint32_t cparams;
3517 	uint32_t eec;
3518 	uint8_t eecp;
3519 	uint8_t bios_sem;
3520 	uint8_t offs;
3521 	int rid;
3522 	int i;
3523 
3524 	rid = PCIR_BAR(0);
3525 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3526 	if (res == NULL)
3527 		return;
3528 
3529 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3530 
3531 	/* Synchronise with the BIOS if it owns the controller. */
3532 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3533 	    eecp = EHCI_EECP_NEXT(eec)) {
3534 		eec = pci_read_config(self, eecp, 4);
3535 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3536 			continue;
3537 		}
3538 		bios_sem = pci_read_config(self, eecp +
3539 		    EHCI_LEGSUP_BIOS_SEM, 1);
3540 		if (bios_sem == 0) {
3541 			continue;
3542 		}
3543 		if (bootverbose)
3544 			printf("ehci early: "
3545 			    "SMM active, request owner change\n");
3546 
3547 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3548 
3549 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3550 			DELAY(1000);
3551 			bios_sem = pci_read_config(self, eecp +
3552 			    EHCI_LEGSUP_BIOS_SEM, 1);
3553 		}
3554 
3555 		if (bios_sem != 0) {
3556 			if (bootverbose)
3557 				printf("ehci early: "
3558 				    "SMM does not respond\n");
3559 		}
3560 		/* Disable interrupts */
3561 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3562 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3563 	}
3564 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3565 }
3566 
3567 /* Perform early XHCI takeover from SMM. */
3568 static void
3569 xhci_early_takeover(device_t self)
3570 {
3571 	struct resource *res;
3572 	uint32_t cparams;
3573 	uint32_t eec;
3574 	uint8_t eecp;
3575 	uint8_t bios_sem;
3576 	uint8_t offs;
3577 	int rid;
3578 	int i;
3579 
3580 	rid = PCIR_BAR(0);
3581 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3582 	if (res == NULL)
3583 		return;
3584 
3585 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3586 
3587 	eec = -1;
3588 
3589 	/* Synchronise with the BIOS if it owns the controller. */
3590 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3591 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3592 		eec = bus_read_4(res, eecp);
3593 
3594 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3595 			continue;
3596 
3597 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3598 		if (bios_sem == 0)
3599 			continue;
3600 
3601 		if (bootverbose)
3602 			printf("xhci early: "
3603 			    "SMM active, request owner change\n");
3604 
3605 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3606 
3607 		/* wait a maximum of 5 second */
3608 
3609 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3610 			DELAY(1000);
3611 			bios_sem = bus_read_1(res, eecp +
3612 			    XHCI_XECP_BIOS_SEM);
3613 		}
3614 
3615 		if (bios_sem != 0) {
3616 			if (bootverbose)
3617 				printf("xhci early: "
3618 				    "SMM does not respond\n");
3619 		}
3620 
3621 		/* Disable interrupts */
3622 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3623 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3624 		bus_read_4(res, offs + XHCI_USBSTS);
3625 	}
3626 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3627 }
3628 
3629 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3630 static void
3631 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3632     struct resource_list *rl)
3633 {
3634 	struct resource *res;
3635 	char *cp;
3636 	rman_res_t start, end, count;
3637 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3638 
3639 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3640 	case PCIM_HDRTYPE_BRIDGE:
3641 		sec_reg = PCIR_SECBUS_1;
3642 		sub_reg = PCIR_SUBBUS_1;
3643 		break;
3644 	case PCIM_HDRTYPE_CARDBUS:
3645 		sec_reg = PCIR_SECBUS_2;
3646 		sub_reg = PCIR_SUBBUS_2;
3647 		break;
3648 	default:
3649 		return;
3650 	}
3651 
3652 	/*
3653 	 * If the existing bus range is valid, attempt to reserve it
3654 	 * from our parent.  If this fails for any reason, clear the
3655 	 * secbus and subbus registers.
3656 	 *
3657 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3658 	 * This would at least preserve the existing sec_bus if it is
3659 	 * valid.
3660 	 */
3661 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3662 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3663 
3664 	/* Quirk handling. */
3665 	switch (pci_get_devid(dev)) {
3666 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3667 		sup_bus = pci_read_config(dev, 0x41, 1);
3668 		if (sup_bus != 0xff) {
3669 			sec_bus = sup_bus + 1;
3670 			sub_bus = sup_bus + 1;
3671 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3672 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3673 		}
3674 		break;
3675 
3676 	case 0x00dd10de:
3677 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3678 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3679 			break;
3680 		if (strncmp(cp, "Compal", 6) != 0) {
3681 			freeenv(cp);
3682 			break;
3683 		}
3684 		freeenv(cp);
3685 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3686 			break;
3687 		if (strncmp(cp, "08A0", 4) != 0) {
3688 			freeenv(cp);
3689 			break;
3690 		}
3691 		freeenv(cp);
3692 		if (sub_bus < 0xa) {
3693 			sub_bus = 0xa;
3694 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3695 		}
3696 		break;
3697 	}
3698 
3699 	if (bootverbose)
3700 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3701 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3702 		start = sec_bus;
3703 		end = sub_bus;
3704 		count = end - start + 1;
3705 
3706 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3707 
3708 		/*
3709 		 * If requested, clear secondary bus registers in
3710 		 * bridge devices to force a complete renumbering
3711 		 * rather than reserving the existing range.  However,
3712 		 * preserve the existing size.
3713 		 */
3714 		if (pci_clear_buses)
3715 			goto clear;
3716 
3717 		rid = 0;
3718 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3719 		    start, end, count, 0);
3720 		if (res != NULL)
3721 			return;
3722 
3723 		if (bootverbose)
3724 			device_printf(bus,
3725 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3726 			    pci_get_domain(dev), pci_get_bus(dev),
3727 			    pci_get_slot(dev), pci_get_function(dev));
3728 	}
3729 
3730 clear:
3731 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3732 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3733 }
3734 
3735 static struct resource *
3736 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3737     rman_res_t end, rman_res_t count, u_int flags)
3738 {
3739 	struct pci_devinfo *dinfo;
3740 	pcicfgregs *cfg;
3741 	struct resource_list *rl;
3742 	struct resource *res;
3743 	int sec_reg, sub_reg;
3744 
3745 	dinfo = device_get_ivars(child);
3746 	cfg = &dinfo->cfg;
3747 	rl = &dinfo->resources;
3748 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3749 	case PCIM_HDRTYPE_BRIDGE:
3750 		sec_reg = PCIR_SECBUS_1;
3751 		sub_reg = PCIR_SUBBUS_1;
3752 		break;
3753 	case PCIM_HDRTYPE_CARDBUS:
3754 		sec_reg = PCIR_SECBUS_2;
3755 		sub_reg = PCIR_SUBBUS_2;
3756 		break;
3757 	default:
3758 		return (NULL);
3759 	}
3760 
3761 	if (*rid != 0)
3762 		return (NULL);
3763 
3764 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3765 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3766 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3767 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3768 		    start, end, count, flags & ~RF_ACTIVE);
3769 		if (res == NULL) {
3770 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3771 			device_printf(child, "allocating %ju bus%s failed\n",
3772 			    count, count == 1 ? "" : "es");
3773 			return (NULL);
3774 		}
3775 		if (bootverbose)
3776 			device_printf(child,
3777 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3778 			    count == 1 ? "" : "es", rman_get_start(res));
3779 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3780 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3781 	}
3782 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3783 	    end, count, flags));
3784 }
3785 #endif
3786 
3787 static int
3788 pci_ea_bei_to_rid(device_t dev, int bei)
3789 {
3790 #ifdef PCI_IOV
3791 	struct pci_devinfo *dinfo;
3792 	int iov_pos;
3793 	struct pcicfg_iov *iov;
3794 
3795 	dinfo = device_get_ivars(dev);
3796 	iov = dinfo->cfg.iov;
3797 	if (iov != NULL)
3798 		iov_pos = iov->iov_pos;
3799 	else
3800 		iov_pos = 0;
3801 #endif
3802 
3803 	/* Check if matches BAR */
3804 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3805 	    (bei <= PCIM_EA_BEI_BAR_5))
3806 		return (PCIR_BAR(bei));
3807 
3808 	/* Check ROM */
3809 	if (bei == PCIM_EA_BEI_ROM)
3810 		return (PCIR_BIOS);
3811 
3812 #ifdef PCI_IOV
3813 	/* Check if matches VF_BAR */
3814 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3815 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3816 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3817 		    iov_pos);
3818 #endif
3819 
3820 	return (-1);
3821 }
3822 
3823 int
3824 pci_ea_is_enabled(device_t dev, int rid)
3825 {
3826 	struct pci_ea_entry *ea;
3827 	struct pci_devinfo *dinfo;
3828 
3829 	dinfo = device_get_ivars(dev);
3830 
3831 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3832 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3833 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3834 	}
3835 
3836 	return (0);
3837 }
3838 
3839 void
3840 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3841 {
3842 	struct pci_ea_entry *ea;
3843 	struct pci_devinfo *dinfo;
3844 	pci_addr_t start, end, count;
3845 	struct resource_list *rl;
3846 	int type, flags, rid;
3847 	struct resource *res;
3848 	uint32_t tmp;
3849 #ifdef PCI_IOV
3850 	struct pcicfg_iov *iov;
3851 #endif
3852 
3853 	dinfo = device_get_ivars(dev);
3854 	rl = &dinfo->resources;
3855 	flags = 0;
3856 
3857 #ifdef PCI_IOV
3858 	iov = dinfo->cfg.iov;
3859 #endif
3860 
3861 	if (dinfo->cfg.ea.ea_location == 0)
3862 		return;
3863 
3864 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3865 
3866 		/*
3867 		 * TODO: Ignore EA-BAR if is not enabled.
3868 		 *   Currently the EA implementation supports
3869 		 *   only situation, where EA structure contains
3870 		 *   predefined entries. In case they are not enabled
3871 		 *   leave them unallocated and proceed with
3872 		 *   a legacy-BAR mechanism.
3873 		 */
3874 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3875 			continue;
3876 
3877 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3878 		case PCIM_EA_P_MEM_PREFETCH:
3879 		case PCIM_EA_P_VF_MEM_PREFETCH:
3880 			flags = RF_PREFETCHABLE;
3881 			/* FALLTHROUGH */
3882 		case PCIM_EA_P_VF_MEM:
3883 		case PCIM_EA_P_MEM:
3884 			type = SYS_RES_MEMORY;
3885 			break;
3886 		case PCIM_EA_P_IO:
3887 			type = SYS_RES_IOPORT;
3888 			break;
3889 		default:
3890 			continue;
3891 		}
3892 
3893 		if (alloc_iov != 0) {
3894 #ifdef PCI_IOV
3895 			/* Allocating IOV, confirm BEI matches */
3896 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3897 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3898 				continue;
3899 #else
3900 			continue;
3901 #endif
3902 		} else {
3903 			/* Allocating BAR, confirm BEI matches */
3904 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3905 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3906 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3907 				continue;
3908 		}
3909 
3910 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3911 		if (rid < 0)
3912 			continue;
3913 
3914 		/* Skip resources already allocated by EA */
3915 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3916 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3917 			continue;
3918 
3919 		start = ea->eae_base;
3920 		count = ea->eae_max_offset + 1;
3921 #ifdef PCI_IOV
3922 		if (iov != NULL)
3923 			count = count * iov->iov_num_vfs;
3924 #endif
3925 		end = start + count - 1;
3926 		if (count == 0)
3927 			continue;
3928 
3929 		resource_list_add(rl, type, rid, start, end, count);
3930 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3931 		    flags);
3932 		if (res == NULL) {
3933 			resource_list_delete(rl, type, rid);
3934 
3935 			/*
3936 			 * Failed to allocate using EA, disable entry.
3937 			 * Another attempt to allocation will be performed
3938 			 * further, but this time using legacy BAR registers
3939 			 */
3940 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3941 			tmp &= ~PCIM_EA_ENABLE;
3942 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3943 
3944 			/*
3945 			 * Disabling entry might fail in case it is hardwired.
3946 			 * Read flags again to match current status.
3947 			 */
3948 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3949 
3950 			continue;
3951 		}
3952 
3953 		/* As per specification, fill BAR with zeros */
3954 		pci_write_config(dev, rid, 0, 4);
3955 	}
3956 }
3957 
3958 void
3959 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3960 {
3961 	struct pci_devinfo *dinfo;
3962 	pcicfgregs *cfg;
3963 	struct resource_list *rl;
3964 	const struct pci_quirk *q;
3965 	uint32_t devid;
3966 	int i;
3967 
3968 	dinfo = device_get_ivars(dev);
3969 	cfg = &dinfo->cfg;
3970 	rl = &dinfo->resources;
3971 	devid = (cfg->device << 16) | cfg->vendor;
3972 
3973 	/* Allocate resources using Enhanced Allocation */
3974 	pci_add_resources_ea(bus, dev, 0);
3975 
3976 	/* ATA devices needs special map treatment */
3977 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3978 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3979 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3980 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3981 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3982 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3983 	else
3984 		for (i = 0; i < cfg->nummaps;) {
3985 			/* Skip resources already managed by EA */
3986 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3987 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3988 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3989 				i++;
3990 				continue;
3991 			}
3992 
3993 			/*
3994 			 * Skip quirked resources.
3995 			 */
3996 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3997 				if (q->devid == devid &&
3998 				    q->type == PCI_QUIRK_UNMAP_REG &&
3999 				    q->arg1 == PCIR_BAR(i))
4000 					break;
4001 			if (q->devid != 0) {
4002 				i++;
4003 				continue;
4004 			}
4005 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4006 			    prefetchmask & (1 << i));
4007 		}
4008 
4009 	/*
4010 	 * Add additional, quirked resources.
4011 	 */
4012 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4013 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4014 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4015 
4016 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4017 #ifdef __PCI_REROUTE_INTERRUPT
4018 		/*
4019 		 * Try to re-route interrupts. Sometimes the BIOS or
4020 		 * firmware may leave bogus values in these registers.
4021 		 * If the re-route fails, then just stick with what we
4022 		 * have.
4023 		 */
4024 		pci_assign_interrupt(bus, dev, 1);
4025 #else
4026 		pci_assign_interrupt(bus, dev, 0);
4027 #endif
4028 	}
4029 
4030 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4031 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4032 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4033 			xhci_early_takeover(dev);
4034 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4035 			ehci_early_takeover(dev);
4036 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4037 			ohci_early_takeover(dev);
4038 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4039 			uhci_early_takeover(dev);
4040 	}
4041 
4042 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4043 	/*
4044 	 * Reserve resources for secondary bus ranges behind bridge
4045 	 * devices.
4046 	 */
4047 	pci_reserve_secbus(bus, dev, cfg, rl);
4048 #endif
4049 }
4050 
4051 static struct pci_devinfo *
4052 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4053     int slot, int func)
4054 {
4055 	struct pci_devinfo *dinfo;
4056 
4057 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4058 	if (dinfo != NULL)
4059 		pci_add_child(dev, dinfo);
4060 
4061 	return (dinfo);
4062 }
4063 
4064 void
4065 pci_add_children(device_t dev, int domain, int busno)
4066 {
4067 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4068 	device_t pcib = device_get_parent(dev);
4069 	struct pci_devinfo *dinfo;
4070 	int maxslots;
4071 	int s, f, pcifunchigh;
4072 	uint8_t hdrtype;
4073 	int first_func;
4074 
4075 	/*
4076 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4077 	 * enable ARI.  We must enable ARI before detecting the rest of the
4078 	 * functions on this bus as ARI changes the set of slots and functions
4079 	 * that are legal on this bus.
4080 	 */
4081 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4082 	if (dinfo != NULL && pci_enable_ari)
4083 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4084 
4085 	/*
4086 	 * Start looking for new devices on slot 0 at function 1 because we
4087 	 * just identified the device at slot 0, function 0.
4088 	 */
4089 	first_func = 1;
4090 
4091 	maxslots = PCIB_MAXSLOTS(pcib);
4092 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4093 		pcifunchigh = 0;
4094 		f = 0;
4095 		DELAY(1);
4096 		hdrtype = REG(PCIR_HDRTYPE, 1);
4097 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4098 			continue;
4099 		if (hdrtype & PCIM_MFDEV)
4100 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4101 		for (f = first_func; f <= pcifunchigh; f++)
4102 			pci_identify_function(pcib, dev, domain, busno, s, f);
4103 	}
4104 #undef REG
4105 }
4106 
4107 int
4108 pci_rescan_method(device_t dev)
4109 {
4110 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4111 	device_t pcib = device_get_parent(dev);
4112 	device_t child, *devlist, *unchanged;
4113 	int devcount, error, i, j, maxslots, oldcount;
4114 	int busno, domain, s, f, pcifunchigh;
4115 	uint8_t hdrtype;
4116 
4117 	/* No need to check for ARI on a rescan. */
4118 	error = device_get_children(dev, &devlist, &devcount);
4119 	if (error)
4120 		return (error);
4121 	if (devcount != 0) {
4122 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4123 		    M_NOWAIT | M_ZERO);
4124 		if (unchanged == NULL) {
4125 			free(devlist, M_TEMP);
4126 			return (ENOMEM);
4127 		}
4128 	} else
4129 		unchanged = NULL;
4130 
4131 	domain = pcib_get_domain(dev);
4132 	busno = pcib_get_bus(dev);
4133 	maxslots = PCIB_MAXSLOTS(pcib);
4134 	for (s = 0; s <= maxslots; s++) {
4135 		/* If function 0 is not present, skip to the next slot. */
4136 		f = 0;
4137 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4138 			continue;
4139 		pcifunchigh = 0;
4140 		hdrtype = REG(PCIR_HDRTYPE, 1);
4141 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4142 			continue;
4143 		if (hdrtype & PCIM_MFDEV)
4144 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4145 		for (f = 0; f <= pcifunchigh; f++) {
4146 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4147 				continue;
4148 
4149 			/*
4150 			 * Found a valid function.  Check if a
4151 			 * device_t for this device already exists.
4152 			 */
4153 			for (i = 0; i < devcount; i++) {
4154 				child = devlist[i];
4155 				if (child == NULL)
4156 					continue;
4157 				if (pci_get_slot(child) == s &&
4158 				    pci_get_function(child) == f) {
4159 					unchanged[i] = child;
4160 					goto next_func;
4161 				}
4162 			}
4163 
4164 			pci_identify_function(pcib, dev, domain, busno, s, f);
4165 		next_func:;
4166 		}
4167 	}
4168 
4169 	/* Remove devices that are no longer present. */
4170 	for (i = 0; i < devcount; i++) {
4171 		if (unchanged[i] != NULL)
4172 			continue;
4173 		device_delete_child(dev, devlist[i]);
4174 	}
4175 
4176 	free(devlist, M_TEMP);
4177 	oldcount = devcount;
4178 
4179 	/* Try to attach the devices just added. */
4180 	error = device_get_children(dev, &devlist, &devcount);
4181 	if (error) {
4182 		free(unchanged, M_TEMP);
4183 		return (error);
4184 	}
4185 
4186 	for (i = 0; i < devcount; i++) {
4187 		for (j = 0; j < oldcount; j++) {
4188 			if (devlist[i] == unchanged[j])
4189 				goto next_device;
4190 		}
4191 
4192 		device_probe_and_attach(devlist[i]);
4193 	next_device:;
4194 	}
4195 
4196 	free(unchanged, M_TEMP);
4197 	free(devlist, M_TEMP);
4198 	return (0);
4199 #undef REG
4200 }
4201 
4202 #ifdef PCI_IOV
4203 device_t
4204 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4205     uint16_t did)
4206 {
4207 	struct pci_devinfo *vf_dinfo;
4208 	device_t pcib;
4209 	int busno, slot, func;
4210 
4211 	pcib = device_get_parent(bus);
4212 
4213 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4214 
4215 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4216 	    slot, func, vid, did);
4217 
4218 	vf_dinfo->cfg.flags |= PCICFG_VF;
4219 	pci_add_child(bus, vf_dinfo);
4220 
4221 	return (vf_dinfo->cfg.dev);
4222 }
4223 
4224 device_t
4225 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4226     uint16_t vid, uint16_t did)
4227 {
4228 
4229 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4230 }
4231 #endif
4232 
4233 static void
4234 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4235 {
4236 	int aer;
4237 	uint32_t r;
4238 	uint16_t r2;
4239 
4240 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4241 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4242 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4243 		    PCIER_ROOT_CTL, 2);
4244 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4245 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4246 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4247 		    PCIER_ROOT_CTL, r2, 2);
4248 	}
4249 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4250 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4251 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4252 		if (r != 0 && bootverbose) {
4253 			pci_printf(&dinfo->cfg,
4254 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4255 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4256 			    4));
4257 		}
4258 
4259 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4260 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4261 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4262 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4263 		    PCIM_AER_UC_POISONED_TLP |
4264 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4265 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4266 		    PCIM_AER_UC_COMPLETER_ABORT |
4267 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4268 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4269 		    PCIM_AER_UC_MALFORMED_TLP |
4270 		    PCIM_AER_UC_ECRC_ERROR |
4271 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4272 		    PCIM_AER_UC_ACS_VIOLATION |
4273 		    PCIM_AER_UC_INTERNAL_ERROR |
4274 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4275 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4276 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4277 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4278 
4279 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4280 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4281 		if (r != 0 && bootverbose) {
4282 			pci_printf(&dinfo->cfg,
4283 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4284 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4285 			    4));
4286 		}
4287 
4288 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4289 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4290 		    PCIM_AER_COR_BAD_TLP |
4291 		    PCIM_AER_COR_BAD_DLLP |
4292 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4293 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4294 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4295 		    PCIM_AER_COR_INTERNAL_ERROR |
4296 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4297 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4298 
4299 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4300 		    PCIER_DEVICE_CTL, 2);
4301 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4302 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4303 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4304 		    PCIER_DEVICE_CTL, r, 2);
4305 	}
4306 }
4307 
4308 void
4309 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4310 {
4311 	device_t dev;
4312 
4313 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4314 	device_set_ivars(dev, dinfo);
4315 	resource_list_init(&dinfo->resources);
4316 	pci_cfg_save(dev, dinfo, 0);
4317 	pci_cfg_restore(dev, dinfo);
4318 	pci_print_verbose(dinfo);
4319 	pci_add_resources(bus, dev, 0, 0);
4320 	pci_child_added(dinfo->cfg.dev);
4321 
4322 	if (pci_clear_aer_on_attach)
4323 		pci_add_child_clear_aer(dev, dinfo);
4324 
4325 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4326 }
4327 
4328 void
4329 pci_child_added_method(device_t dev, device_t child)
4330 {
4331 
4332 }
4333 
4334 static int
4335 pci_probe(device_t dev)
4336 {
4337 
4338 	device_set_desc(dev, "PCI bus");
4339 
4340 	/* Allow other subclasses to override this driver. */
4341 	return (BUS_PROBE_GENERIC);
4342 }
4343 
4344 int
4345 pci_attach_common(device_t dev)
4346 {
4347 	struct pci_softc *sc;
4348 	int busno, domain;
4349 #ifdef PCI_RES_BUS
4350 	int rid;
4351 #endif
4352 
4353 	sc = device_get_softc(dev);
4354 	domain = pcib_get_domain(dev);
4355 	busno = pcib_get_bus(dev);
4356 #ifdef PCI_RES_BUS
4357 	rid = 0;
4358 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4359 	    1, 0);
4360 	if (sc->sc_bus == NULL) {
4361 		device_printf(dev, "failed to allocate bus number\n");
4362 		return (ENXIO);
4363 	}
4364 #endif
4365 	if (bootverbose)
4366 		device_printf(dev, "domain=%d, physical bus=%d\n",
4367 		    domain, busno);
4368 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4369 	return (0);
4370 }
4371 
4372 int
4373 pci_attach(device_t dev)
4374 {
4375 	int busno, domain, error;
4376 
4377 	error = pci_attach_common(dev);
4378 	if (error)
4379 		return (error);
4380 
4381 	/*
4382 	 * Since there can be multiple independently numbered PCI
4383 	 * buses on systems with multiple PCI domains, we can't use
4384 	 * the unit number to decide which bus we are probing. We ask
4385 	 * the parent pcib what our domain and bus numbers are.
4386 	 */
4387 	domain = pcib_get_domain(dev);
4388 	busno = pcib_get_bus(dev);
4389 	pci_add_children(dev, domain, busno);
4390 	return (bus_generic_attach(dev));
4391 }
4392 
4393 int
4394 pci_detach(device_t dev)
4395 {
4396 #ifdef PCI_RES_BUS
4397 	struct pci_softc *sc;
4398 #endif
4399 	int error;
4400 
4401 	error = bus_generic_detach(dev);
4402 	if (error)
4403 		return (error);
4404 #ifdef PCI_RES_BUS
4405 	sc = device_get_softc(dev);
4406 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4407 	if (error)
4408 		return (error);
4409 #endif
4410 	return (device_delete_children(dev));
4411 }
4412 
4413 static void
4414 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4415 {
4416 	int line, unit;
4417 	const char *at;
4418 	char me1[24], me2[32];
4419 	uint8_t b, s, f;
4420 	uint32_t d;
4421 
4422 	d = pci_get_domain(child);
4423 	b = pci_get_bus(child);
4424 	s = pci_get_slot(child);
4425 	f = pci_get_function(child);
4426 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4427 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4428 	line = 0;
4429 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4430 		resource_string_value(name, unit, "at", &at);
4431 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4432 			continue; /* No match, try next candidate */
4433 		*unitp = unit;
4434 		return;
4435 	}
4436 }
4437 
4438 static void
4439 pci_set_power_child(device_t dev, device_t child, int state)
4440 {
4441 	device_t pcib;
4442 	int dstate;
4443 
4444 	/*
4445 	 * Set the device to the given state.  If the firmware suggests
4446 	 * a different power state, use it instead.  If power management
4447 	 * is not present, the firmware is responsible for managing
4448 	 * device power.  Skip children who aren't attached since they
4449 	 * are handled separately.
4450 	 */
4451 	pcib = device_get_parent(dev);
4452 	dstate = state;
4453 	if (device_is_attached(child) &&
4454 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4455 		pci_set_powerstate(child, dstate);
4456 }
4457 
4458 int
4459 pci_suspend_child(device_t dev, device_t child)
4460 {
4461 	struct pci_devinfo *dinfo;
4462 	struct resource_list_entry *rle;
4463 	int error;
4464 
4465 	dinfo = device_get_ivars(child);
4466 
4467 	/*
4468 	 * Save the PCI configuration space for the child and set the
4469 	 * device in the appropriate power state for this sleep state.
4470 	 */
4471 	pci_cfg_save(child, dinfo, 0);
4472 
4473 	/* Suspend devices before potentially powering them down. */
4474 	error = bus_generic_suspend_child(dev, child);
4475 
4476 	if (error)
4477 		return (error);
4478 
4479 	if (pci_do_power_suspend) {
4480 		/*
4481 		 * Make sure this device's interrupt handler is not invoked
4482 		 * in the case the device uses a shared interrupt that can
4483 		 * be raised by some other device.
4484 		 * This is applicable only to regular (legacy) PCI interrupts
4485 		 * as MSI/MSI-X interrupts are never shared.
4486 		 */
4487 		rle = resource_list_find(&dinfo->resources,
4488 		    SYS_RES_IRQ, 0);
4489 		if (rle != NULL && rle->res != NULL)
4490 			(void)bus_suspend_intr(child, rle->res);
4491 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4492 	}
4493 
4494 	return (0);
4495 }
4496 
4497 int
4498 pci_resume_child(device_t dev, device_t child)
4499 {
4500 	struct pci_devinfo *dinfo;
4501 	struct resource_list_entry *rle;
4502 
4503 	if (pci_do_power_resume)
4504 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4505 
4506 	dinfo = device_get_ivars(child);
4507 	pci_cfg_restore(child, dinfo);
4508 	if (!device_is_attached(child))
4509 		pci_cfg_save(child, dinfo, 1);
4510 
4511 	bus_generic_resume_child(dev, child);
4512 
4513 	/*
4514 	 * Allow interrupts only after fully resuming the driver and hardware.
4515 	 */
4516 	if (pci_do_power_suspend) {
4517 		/* See pci_suspend_child for details. */
4518 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4519 		if (rle != NULL && rle->res != NULL)
4520 			(void)bus_resume_intr(child, rle->res);
4521 	}
4522 
4523 	return (0);
4524 }
4525 
4526 int
4527 pci_resume(device_t dev)
4528 {
4529 	device_t child, *devlist;
4530 	int error, i, numdevs;
4531 
4532 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4533 		return (error);
4534 
4535 	/*
4536 	 * Resume critical devices first, then everything else later.
4537 	 */
4538 	for (i = 0; i < numdevs; i++) {
4539 		child = devlist[i];
4540 		switch (pci_get_class(child)) {
4541 		case PCIC_DISPLAY:
4542 		case PCIC_MEMORY:
4543 		case PCIC_BRIDGE:
4544 		case PCIC_BASEPERIPH:
4545 			BUS_RESUME_CHILD(dev, child);
4546 			break;
4547 		}
4548 	}
4549 	for (i = 0; i < numdevs; i++) {
4550 		child = devlist[i];
4551 		switch (pci_get_class(child)) {
4552 		case PCIC_DISPLAY:
4553 		case PCIC_MEMORY:
4554 		case PCIC_BRIDGE:
4555 		case PCIC_BASEPERIPH:
4556 			break;
4557 		default:
4558 			BUS_RESUME_CHILD(dev, child);
4559 		}
4560 	}
4561 	free(devlist, M_TEMP);
4562 	return (0);
4563 }
4564 
4565 static void
4566 pci_load_vendor_data(void)
4567 {
4568 	caddr_t data;
4569 	void *ptr;
4570 	size_t sz;
4571 
4572 	data = preload_search_by_type("pci_vendor_data");
4573 	if (data != NULL) {
4574 		ptr = preload_fetch_addr(data);
4575 		sz = preload_fetch_size(data);
4576 		if (ptr != NULL && sz != 0) {
4577 			pci_vendordata = ptr;
4578 			pci_vendordata_size = sz;
4579 			/* terminate the database */
4580 			pci_vendordata[pci_vendordata_size] = '\n';
4581 		}
4582 	}
4583 }
4584 
4585 void
4586 pci_driver_added(device_t dev, driver_t *driver)
4587 {
4588 	int numdevs;
4589 	device_t *devlist;
4590 	device_t child;
4591 	struct pci_devinfo *dinfo;
4592 	int i;
4593 
4594 	if (bootverbose)
4595 		device_printf(dev, "driver added\n");
4596 	DEVICE_IDENTIFY(driver, dev);
4597 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4598 		return;
4599 	for (i = 0; i < numdevs; i++) {
4600 		child = devlist[i];
4601 		if (device_get_state(child) != DS_NOTPRESENT)
4602 			continue;
4603 		dinfo = device_get_ivars(child);
4604 		pci_print_verbose(dinfo);
4605 		if (bootverbose)
4606 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4607 		pci_cfg_restore(child, dinfo);
4608 		if (device_probe_and_attach(child) != 0)
4609 			pci_child_detached(dev, child);
4610 	}
4611 	free(devlist, M_TEMP);
4612 }
4613 
4614 int
4615 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4616     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4617 {
4618 	struct pci_devinfo *dinfo;
4619 	struct msix_table_entry *mte;
4620 	struct msix_vector *mv;
4621 	uint64_t addr;
4622 	uint32_t data;
4623 	void *cookie;
4624 	int error, rid;
4625 
4626 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4627 	    arg, &cookie);
4628 	if (error)
4629 		return (error);
4630 
4631 	/* If this is not a direct child, just bail out. */
4632 	if (device_get_parent(child) != dev) {
4633 		*cookiep = cookie;
4634 		return(0);
4635 	}
4636 
4637 	rid = rman_get_rid(irq);
4638 	if (rid == 0) {
4639 		/* Make sure that INTx is enabled */
4640 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4641 	} else {
4642 		/*
4643 		 * Check to see if the interrupt is MSI or MSI-X.
4644 		 * Ask our parent to map the MSI and give
4645 		 * us the address and data register values.
4646 		 * If we fail for some reason, teardown the
4647 		 * interrupt handler.
4648 		 */
4649 		dinfo = device_get_ivars(child);
4650 		if (dinfo->cfg.msi.msi_alloc > 0) {
4651 			if (dinfo->cfg.msi.msi_addr == 0) {
4652 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4653 			    ("MSI has handlers, but vectors not mapped"));
4654 				error = PCIB_MAP_MSI(device_get_parent(dev),
4655 				    child, rman_get_start(irq), &addr, &data);
4656 				if (error)
4657 					goto bad;
4658 				dinfo->cfg.msi.msi_addr = addr;
4659 				dinfo->cfg.msi.msi_data = data;
4660 			}
4661 			if (dinfo->cfg.msi.msi_handlers == 0)
4662 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4663 				    dinfo->cfg.msi.msi_data);
4664 			dinfo->cfg.msi.msi_handlers++;
4665 		} else {
4666 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4667 			    ("No MSI or MSI-X interrupts allocated"));
4668 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4669 			    ("MSI-X index too high"));
4670 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4671 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4672 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4673 			KASSERT(mv->mv_irq == rman_get_start(irq),
4674 			    ("IRQ mismatch"));
4675 			if (mv->mv_address == 0) {
4676 				KASSERT(mte->mte_handlers == 0,
4677 		    ("MSI-X table entry has handlers, but vector not mapped"));
4678 				error = PCIB_MAP_MSI(device_get_parent(dev),
4679 				    child, rman_get_start(irq), &addr, &data);
4680 				if (error)
4681 					goto bad;
4682 				mv->mv_address = addr;
4683 				mv->mv_data = data;
4684 			}
4685 
4686 			/*
4687 			 * The MSIX table entry must be made valid by
4688 			 * incrementing the mte_handlers before
4689 			 * calling pci_enable_msix() and
4690 			 * pci_resume_msix(). Else the MSIX rewrite
4691 			 * table quirk will not work as expected.
4692 			 */
4693 			mte->mte_handlers++;
4694 			if (mte->mte_handlers == 1) {
4695 				pci_enable_msix(child, rid - 1, mv->mv_address,
4696 				    mv->mv_data);
4697 				pci_unmask_msix(child, rid - 1);
4698 			}
4699 		}
4700 
4701 		/*
4702 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4703 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4704 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4705 		 */
4706 		if (!pci_has_quirk(pci_get_devid(child),
4707 		    PCI_QUIRK_MSI_INTX_BUG))
4708 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4709 		else
4710 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4711 	bad:
4712 		if (error) {
4713 			(void)bus_generic_teardown_intr(dev, child, irq,
4714 			    cookie);
4715 			return (error);
4716 		}
4717 	}
4718 	*cookiep = cookie;
4719 	return (0);
4720 }
4721 
4722 int
4723 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4724     void *cookie)
4725 {
4726 	struct msix_table_entry *mte;
4727 	struct resource_list_entry *rle;
4728 	struct pci_devinfo *dinfo;
4729 	int error, rid;
4730 
4731 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4732 		return (EINVAL);
4733 
4734 	/* If this isn't a direct child, just bail out */
4735 	if (device_get_parent(child) != dev)
4736 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4737 
4738 	rid = rman_get_rid(irq);
4739 	if (rid == 0) {
4740 		/* Mask INTx */
4741 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4742 	} else {
4743 		/*
4744 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4745 		 * decrement the appropriate handlers count and mask the
4746 		 * MSI-X message, or disable MSI messages if the count
4747 		 * drops to 0.
4748 		 */
4749 		dinfo = device_get_ivars(child);
4750 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4751 		if (rle->res != irq)
4752 			return (EINVAL);
4753 		if (dinfo->cfg.msi.msi_alloc > 0) {
4754 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4755 			    ("MSI-X index too high"));
4756 			if (dinfo->cfg.msi.msi_handlers == 0)
4757 				return (EINVAL);
4758 			dinfo->cfg.msi.msi_handlers--;
4759 			if (dinfo->cfg.msi.msi_handlers == 0)
4760 				pci_disable_msi(child);
4761 		} else {
4762 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4763 			    ("No MSI or MSI-X interrupts allocated"));
4764 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4765 			    ("MSI-X index too high"));
4766 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4767 			if (mte->mte_handlers == 0)
4768 				return (EINVAL);
4769 			mte->mte_handlers--;
4770 			if (mte->mte_handlers == 0)
4771 				pci_mask_msix(child, rid - 1);
4772 		}
4773 	}
4774 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4775 	if (rid > 0)
4776 		KASSERT(error == 0,
4777 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4778 	return (error);
4779 }
4780 
4781 int
4782 pci_print_child(device_t dev, device_t child)
4783 {
4784 	struct pci_devinfo *dinfo;
4785 	struct resource_list *rl;
4786 	int retval = 0;
4787 
4788 	dinfo = device_get_ivars(child);
4789 	rl = &dinfo->resources;
4790 
4791 	retval += bus_print_child_header(dev, child);
4792 
4793 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4794 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4795 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4796 	if (device_get_flags(dev))
4797 		retval += printf(" flags %#x", device_get_flags(dev));
4798 
4799 	retval += printf(" at device %d.%d", pci_get_slot(child),
4800 	    pci_get_function(child));
4801 
4802 	retval += bus_print_child_domain(dev, child);
4803 	retval += bus_print_child_footer(dev, child);
4804 
4805 	return (retval);
4806 }
4807 
4808 static const struct
4809 {
4810 	int		class;
4811 	int		subclass;
4812 	int		report; /* 0 = bootverbose, 1 = always */
4813 	const char	*desc;
4814 } pci_nomatch_tab[] = {
4815 	{PCIC_OLD,		-1,			1, "old"},
4816 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4817 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4818 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4819 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4820 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4821 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4822 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4823 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4824 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4825 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4826 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4827 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4828 	{PCIC_NETWORK,		-1,			1, "network"},
4829 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4830 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4831 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4832 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4833 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4834 	{PCIC_DISPLAY,		-1,			1, "display"},
4835 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4836 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4837 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4838 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4839 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4840 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4841 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4842 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4843 	{PCIC_MEMORY,		-1,			1, "memory"},
4844 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4845 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4846 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4847 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4848 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4849 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4850 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4851 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4852 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4853 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4854 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4855 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4856 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4857 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4858 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4859 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4860 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4861 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4862 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4863 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4864 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4865 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4866 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4867 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4868 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4869 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4870 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4871 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4872 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4873 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4874 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4875 	{PCIC_DOCKING,		-1,			1, "docking station"},
4876 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4877 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4878 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4879 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4880 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4881 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4882 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4883 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4884 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4885 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4886 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4887 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4888 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4889 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4890 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4891 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4892 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4893 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4894 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4895 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4896 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4897 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4898 	{PCIC_DASP,		-1,			0, "dasp"},
4899 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4900 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4901 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4902 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4903 	{0, 0, 0,		NULL}
4904 };
4905 
4906 void
4907 pci_probe_nomatch(device_t dev, device_t child)
4908 {
4909 	int i, report;
4910 	const char *cp, *scp;
4911 	char *device;
4912 
4913 	/*
4914 	 * Look for a listing for this device in a loaded device database.
4915 	 */
4916 	report = 1;
4917 	if ((device = pci_describe_device(child)) != NULL) {
4918 		device_printf(dev, "<%s>", device);
4919 		free(device, M_DEVBUF);
4920 	} else {
4921 		/*
4922 		 * Scan the class/subclass descriptions for a general
4923 		 * description.
4924 		 */
4925 		cp = "unknown";
4926 		scp = NULL;
4927 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4928 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4929 				if (pci_nomatch_tab[i].subclass == -1) {
4930 					cp = pci_nomatch_tab[i].desc;
4931 					report = pci_nomatch_tab[i].report;
4932 				} else if (pci_nomatch_tab[i].subclass ==
4933 				    pci_get_subclass(child)) {
4934 					scp = pci_nomatch_tab[i].desc;
4935 					report = pci_nomatch_tab[i].report;
4936 				}
4937 			}
4938 		}
4939 		if (report || bootverbose) {
4940 			device_printf(dev, "<%s%s%s>",
4941 			    cp ? cp : "",
4942 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4943 			    scp ? scp : "");
4944 		}
4945 	}
4946 	if (report || bootverbose) {
4947 		printf(" at device %d.%d (no driver attached)\n",
4948 		    pci_get_slot(child), pci_get_function(child));
4949 	}
4950 	pci_cfg_save(child, device_get_ivars(child), 1);
4951 }
4952 
4953 void
4954 pci_child_detached(device_t dev, device_t child)
4955 {
4956 	struct pci_devinfo *dinfo;
4957 	struct resource_list *rl;
4958 
4959 	dinfo = device_get_ivars(child);
4960 	rl = &dinfo->resources;
4961 
4962 	/*
4963 	 * Have to deallocate IRQs before releasing any MSI messages and
4964 	 * have to release MSI messages before deallocating any memory
4965 	 * BARs.
4966 	 */
4967 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4968 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4969 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4970 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4971 		(void)pci_release_msi(child);
4972 	}
4973 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4974 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4975 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4976 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4977 #ifdef PCI_RES_BUS
4978 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4979 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4980 #endif
4981 
4982 	pci_cfg_save(child, dinfo, 1);
4983 }
4984 
4985 /*
4986  * Parse the PCI device database, if loaded, and return a pointer to a
4987  * description of the device.
4988  *
4989  * The database is flat text formatted as follows:
4990  *
4991  * Any line not in a valid format is ignored.
4992  * Lines are terminated with newline '\n' characters.
4993  *
4994  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4995  * the vendor name.
4996  *
4997  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
4998  * - devices cannot be listed without a corresponding VENDOR line.
4999  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5000  * another TAB, then the device name.
5001  */
5002 
5003 /*
5004  * Assuming (ptr) points to the beginning of a line in the database,
5005  * return the vendor or device and description of the next entry.
5006  * The value of (vendor) or (device) inappropriate for the entry type
5007  * is set to -1.  Returns nonzero at the end of the database.
5008  *
5009  * Note that this is slightly unrobust in the face of corrupt data;
5010  * we attempt to safeguard against this by spamming the end of the
5011  * database with a newline when we initialise.
5012  */
5013 static int
5014 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5015 {
5016 	char	*cp = *ptr;
5017 	int	left;
5018 
5019 	*device = -1;
5020 	*vendor = -1;
5021 	**desc = '\0';
5022 	for (;;) {
5023 		left = pci_vendordata_size - (cp - pci_vendordata);
5024 		if (left <= 0) {
5025 			*ptr = cp;
5026 			return(1);
5027 		}
5028 
5029 		/* vendor entry? */
5030 		if (*cp != '\t' &&
5031 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5032 			break;
5033 		/* device entry? */
5034 		if (*cp == '\t' &&
5035 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5036 			break;
5037 
5038 		/* skip to next line */
5039 		while (*cp != '\n' && left > 0) {
5040 			cp++;
5041 			left--;
5042 		}
5043 		if (*cp == '\n') {
5044 			cp++;
5045 			left--;
5046 		}
5047 	}
5048 	/* skip to next line */
5049 	while (*cp != '\n' && left > 0) {
5050 		cp++;
5051 		left--;
5052 	}
5053 	if (*cp == '\n' && left > 0)
5054 		cp++;
5055 	*ptr = cp;
5056 	return(0);
5057 }
5058 
5059 static char *
5060 pci_describe_device(device_t dev)
5061 {
5062 	int	vendor, device;
5063 	char	*desc, *vp, *dp, *line;
5064 
5065 	desc = vp = dp = NULL;
5066 
5067 	/*
5068 	 * If we have no vendor data, we can't do anything.
5069 	 */
5070 	if (pci_vendordata == NULL)
5071 		goto out;
5072 
5073 	/*
5074 	 * Scan the vendor data looking for this device
5075 	 */
5076 	line = pci_vendordata;
5077 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5078 		goto out;
5079 	for (;;) {
5080 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5081 			goto out;
5082 		if (vendor == pci_get_vendor(dev))
5083 			break;
5084 	}
5085 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5086 		goto out;
5087 	for (;;) {
5088 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5089 			*dp = 0;
5090 			break;
5091 		}
5092 		if (vendor != -1) {
5093 			*dp = 0;
5094 			break;
5095 		}
5096 		if (device == pci_get_device(dev))
5097 			break;
5098 	}
5099 	if (dp[0] == '\0')
5100 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5101 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5102 	    NULL)
5103 		sprintf(desc, "%s, %s", vp, dp);
5104 out:
5105 	if (vp != NULL)
5106 		free(vp, M_DEVBUF);
5107 	if (dp != NULL)
5108 		free(dp, M_DEVBUF);
5109 	return(desc);
5110 }
5111 
5112 int
5113 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5114 {
5115 	struct pci_devinfo *dinfo;
5116 	pcicfgregs *cfg;
5117 
5118 	dinfo = device_get_ivars(child);
5119 	cfg = &dinfo->cfg;
5120 
5121 	switch (which) {
5122 	case PCI_IVAR_ETHADDR:
5123 		/*
5124 		 * The generic accessor doesn't deal with failure, so
5125 		 * we set the return value, then return an error.
5126 		 */
5127 		*((uint8_t **) result) = NULL;
5128 		return (EINVAL);
5129 	case PCI_IVAR_SUBVENDOR:
5130 		*result = cfg->subvendor;
5131 		break;
5132 	case PCI_IVAR_SUBDEVICE:
5133 		*result = cfg->subdevice;
5134 		break;
5135 	case PCI_IVAR_VENDOR:
5136 		*result = cfg->vendor;
5137 		break;
5138 	case PCI_IVAR_DEVICE:
5139 		*result = cfg->device;
5140 		break;
5141 	case PCI_IVAR_DEVID:
5142 		*result = (cfg->device << 16) | cfg->vendor;
5143 		break;
5144 	case PCI_IVAR_CLASS:
5145 		*result = cfg->baseclass;
5146 		break;
5147 	case PCI_IVAR_SUBCLASS:
5148 		*result = cfg->subclass;
5149 		break;
5150 	case PCI_IVAR_PROGIF:
5151 		*result = cfg->progif;
5152 		break;
5153 	case PCI_IVAR_REVID:
5154 		*result = cfg->revid;
5155 		break;
5156 	case PCI_IVAR_INTPIN:
5157 		*result = cfg->intpin;
5158 		break;
5159 	case PCI_IVAR_IRQ:
5160 		*result = cfg->intline;
5161 		break;
5162 	case PCI_IVAR_DOMAIN:
5163 		*result = cfg->domain;
5164 		break;
5165 	case PCI_IVAR_BUS:
5166 		*result = cfg->bus;
5167 		break;
5168 	case PCI_IVAR_SLOT:
5169 		*result = cfg->slot;
5170 		break;
5171 	case PCI_IVAR_FUNCTION:
5172 		*result = cfg->func;
5173 		break;
5174 	case PCI_IVAR_CMDREG:
5175 		*result = cfg->cmdreg;
5176 		break;
5177 	case PCI_IVAR_CACHELNSZ:
5178 		*result = cfg->cachelnsz;
5179 		break;
5180 	case PCI_IVAR_MINGNT:
5181 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5182 			*result = -1;
5183 			return (EINVAL);
5184 		}
5185 		*result = cfg->mingnt;
5186 		break;
5187 	case PCI_IVAR_MAXLAT:
5188 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5189 			*result = -1;
5190 			return (EINVAL);
5191 		}
5192 		*result = cfg->maxlat;
5193 		break;
5194 	case PCI_IVAR_LATTIMER:
5195 		*result = cfg->lattimer;
5196 		break;
5197 	default:
5198 		return (ENOENT);
5199 	}
5200 	return (0);
5201 }
5202 
5203 int
5204 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5205 {
5206 	struct pci_devinfo *dinfo;
5207 
5208 	dinfo = device_get_ivars(child);
5209 
5210 	switch (which) {
5211 	case PCI_IVAR_INTPIN:
5212 		dinfo->cfg.intpin = value;
5213 		return (0);
5214 	case PCI_IVAR_ETHADDR:
5215 	case PCI_IVAR_SUBVENDOR:
5216 	case PCI_IVAR_SUBDEVICE:
5217 	case PCI_IVAR_VENDOR:
5218 	case PCI_IVAR_DEVICE:
5219 	case PCI_IVAR_DEVID:
5220 	case PCI_IVAR_CLASS:
5221 	case PCI_IVAR_SUBCLASS:
5222 	case PCI_IVAR_PROGIF:
5223 	case PCI_IVAR_REVID:
5224 	case PCI_IVAR_IRQ:
5225 	case PCI_IVAR_DOMAIN:
5226 	case PCI_IVAR_BUS:
5227 	case PCI_IVAR_SLOT:
5228 	case PCI_IVAR_FUNCTION:
5229 		return (EINVAL);	/* disallow for now */
5230 
5231 	default:
5232 		return (ENOENT);
5233 	}
5234 }
5235 
5236 #include "opt_ddb.h"
5237 #ifdef DDB
5238 #include <ddb/ddb.h>
5239 #include <sys/cons.h>
5240 
5241 /*
5242  * List resources based on pci map registers, used for within ddb
5243  */
5244 
5245 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5246 {
5247 	struct pci_devinfo *dinfo;
5248 	struct devlist *devlist_head;
5249 	struct pci_conf *p;
5250 	const char *name;
5251 	int i, error, none_count;
5252 
5253 	none_count = 0;
5254 	/* get the head of the device queue */
5255 	devlist_head = &pci_devq;
5256 
5257 	/*
5258 	 * Go through the list of devices and print out devices
5259 	 */
5260 	for (error = 0, i = 0,
5261 	     dinfo = STAILQ_FIRST(devlist_head);
5262 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5263 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5264 
5265 		/* Populate pd_name and pd_unit */
5266 		name = NULL;
5267 		if (dinfo->cfg.dev)
5268 			name = device_get_name(dinfo->cfg.dev);
5269 
5270 		p = &dinfo->conf;
5271 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5272 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5273 			(name && *name) ? name : "none",
5274 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5275 			none_count++,
5276 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5277 			p->pc_sel.pc_func, (p->pc_class << 16) |
5278 			(p->pc_subclass << 8) | p->pc_progif,
5279 			(p->pc_subdevice << 16) | p->pc_subvendor,
5280 			(p->pc_device << 16) | p->pc_vendor,
5281 			p->pc_revid, p->pc_hdr);
5282 	}
5283 }
5284 #endif /* DDB */
5285 
5286 static struct resource *
5287 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5288     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5289     u_int flags)
5290 {
5291 	struct pci_devinfo *dinfo = device_get_ivars(child);
5292 	struct resource_list *rl = &dinfo->resources;
5293 	struct resource *res;
5294 	struct pci_map *pm;
5295 	uint16_t cmd;
5296 	pci_addr_t map, testval;
5297 	int mapsize;
5298 
5299 	res = NULL;
5300 
5301 	/* If rid is managed by EA, ignore it */
5302 	if (pci_ea_is_enabled(child, *rid))
5303 		goto out;
5304 
5305 	pm = pci_find_bar(child, *rid);
5306 	if (pm != NULL) {
5307 		/* This is a BAR that we failed to allocate earlier. */
5308 		mapsize = pm->pm_size;
5309 		map = pm->pm_value;
5310 	} else {
5311 		/*
5312 		 * Weed out the bogons, and figure out how large the
5313 		 * BAR/map is.  BARs that read back 0 here are bogus
5314 		 * and unimplemented.  Note: atapci in legacy mode are
5315 		 * special and handled elsewhere in the code.  If you
5316 		 * have a atapci device in legacy mode and it fails
5317 		 * here, that other code is broken.
5318 		 */
5319 		pci_read_bar(child, *rid, &map, &testval, NULL);
5320 
5321 		/*
5322 		 * Determine the size of the BAR and ignore BARs with a size
5323 		 * of 0.  Device ROM BARs use a different mask value.
5324 		 */
5325 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5326 			mapsize = pci_romsize(testval);
5327 		else
5328 			mapsize = pci_mapsize(testval);
5329 		if (mapsize == 0)
5330 			goto out;
5331 		pm = pci_add_bar(child, *rid, map, mapsize);
5332 	}
5333 
5334 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5335 		if (type != SYS_RES_MEMORY) {
5336 			if (bootverbose)
5337 				device_printf(dev,
5338 				    "child %s requested type %d for rid %#x,"
5339 				    " but the BAR says it is an memio\n",
5340 				    device_get_nameunit(child), type, *rid);
5341 			goto out;
5342 		}
5343 	} else {
5344 		if (type != SYS_RES_IOPORT) {
5345 			if (bootverbose)
5346 				device_printf(dev,
5347 				    "child %s requested type %d for rid %#x,"
5348 				    " but the BAR says it is an ioport\n",
5349 				    device_get_nameunit(child), type, *rid);
5350 			goto out;
5351 		}
5352 	}
5353 
5354 	/*
5355 	 * For real BARs, we need to override the size that
5356 	 * the driver requests, because that's what the BAR
5357 	 * actually uses and we would otherwise have a
5358 	 * situation where we might allocate the excess to
5359 	 * another driver, which won't work.
5360 	 */
5361 	count = ((pci_addr_t)1 << mapsize) * num;
5362 	if (RF_ALIGNMENT(flags) < mapsize)
5363 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5364 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5365 		flags |= RF_PREFETCHABLE;
5366 
5367 	/*
5368 	 * Allocate enough resource, and then write back the
5369 	 * appropriate BAR for that resource.
5370 	 */
5371 	resource_list_add(rl, type, *rid, start, end, count);
5372 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5373 	    count, flags & ~RF_ACTIVE);
5374 	if (res == NULL) {
5375 		resource_list_delete(rl, type, *rid);
5376 		device_printf(child,
5377 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5378 		    count, *rid, type, start, end);
5379 		goto out;
5380 	}
5381 	if (bootverbose)
5382 		device_printf(child,
5383 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5384 		    count, *rid, type, rman_get_start(res));
5385 
5386 	/* Disable decoding via the CMD register before updating the BAR */
5387 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5388 	pci_write_config(child, PCIR_COMMAND,
5389 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5390 
5391 	map = rman_get_start(res);
5392 	pci_write_bar(child, pm, map);
5393 
5394 	/* Restore the original value of the CMD register */
5395 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5396 out:
5397 	return (res);
5398 }
5399 
5400 struct resource *
5401 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5402     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5403     u_int flags)
5404 {
5405 	struct pci_devinfo *dinfo;
5406 	struct resource_list *rl;
5407 	struct resource_list_entry *rle;
5408 	struct resource *res;
5409 	pcicfgregs *cfg;
5410 
5411 	/*
5412 	 * Perform lazy resource allocation
5413 	 */
5414 	dinfo = device_get_ivars(child);
5415 	rl = &dinfo->resources;
5416 	cfg = &dinfo->cfg;
5417 	switch (type) {
5418 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5419 	case PCI_RES_BUS:
5420 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5421 		    flags));
5422 #endif
5423 	case SYS_RES_IRQ:
5424 		/*
5425 		 * Can't alloc legacy interrupt once MSI messages have
5426 		 * been allocated.
5427 		 */
5428 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5429 		    cfg->msix.msix_alloc > 0))
5430 			return (NULL);
5431 
5432 		/*
5433 		 * If the child device doesn't have an interrupt
5434 		 * routed and is deserving of an interrupt, try to
5435 		 * assign it one.
5436 		 */
5437 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5438 		    (cfg->intpin != 0))
5439 			pci_assign_interrupt(dev, child, 0);
5440 		break;
5441 	case SYS_RES_IOPORT:
5442 	case SYS_RES_MEMORY:
5443 #ifdef NEW_PCIB
5444 		/*
5445 		 * PCI-PCI bridge I/O window resources are not BARs.
5446 		 * For those allocations just pass the request up the
5447 		 * tree.
5448 		 */
5449 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5450 			switch (*rid) {
5451 			case PCIR_IOBASEL_1:
5452 			case PCIR_MEMBASE_1:
5453 			case PCIR_PMBASEL_1:
5454 				/*
5455 				 * XXX: Should we bother creating a resource
5456 				 * list entry?
5457 				 */
5458 				return (bus_generic_alloc_resource(dev, child,
5459 				    type, rid, start, end, count, flags));
5460 			}
5461 		}
5462 #endif
5463 		/* Reserve resources for this BAR if needed. */
5464 		rle = resource_list_find(rl, type, *rid);
5465 		if (rle == NULL) {
5466 			res = pci_reserve_map(dev, child, type, rid, start, end,
5467 			    count, num, flags);
5468 			if (res == NULL)
5469 				return (NULL);
5470 		}
5471 	}
5472 	return (resource_list_alloc(rl, dev, child, type, rid,
5473 	    start, end, count, flags));
5474 }
5475 
5476 struct resource *
5477 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5478     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5479 {
5480 #ifdef PCI_IOV
5481 	struct pci_devinfo *dinfo;
5482 #endif
5483 
5484 	if (device_get_parent(child) != dev)
5485 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5486 		    type, rid, start, end, count, flags));
5487 
5488 #ifdef PCI_IOV
5489 	dinfo = device_get_ivars(child);
5490 	if (dinfo->cfg.flags & PCICFG_VF) {
5491 		switch (type) {
5492 		/* VFs can't have I/O BARs. */
5493 		case SYS_RES_IOPORT:
5494 			return (NULL);
5495 		case SYS_RES_MEMORY:
5496 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5497 			    start, end, count, flags));
5498 		}
5499 
5500 		/* Fall through for other types of resource allocations. */
5501 	}
5502 #endif
5503 
5504 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5505 	    count, 1, flags));
5506 }
5507 
5508 int
5509 pci_release_resource(device_t dev, device_t child, int type, int rid,
5510     struct resource *r)
5511 {
5512 	struct pci_devinfo *dinfo;
5513 	struct resource_list *rl;
5514 	pcicfgregs *cfg;
5515 
5516 	if (device_get_parent(child) != dev)
5517 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5518 		    type, rid, r));
5519 
5520 	dinfo = device_get_ivars(child);
5521 	cfg = &dinfo->cfg;
5522 
5523 #ifdef PCI_IOV
5524 	if (dinfo->cfg.flags & PCICFG_VF) {
5525 		switch (type) {
5526 		/* VFs can't have I/O BARs. */
5527 		case SYS_RES_IOPORT:
5528 			return (EDOOFUS);
5529 		case SYS_RES_MEMORY:
5530 			return (pci_vf_release_mem_resource(dev, child, rid,
5531 			    r));
5532 		}
5533 
5534 		/* Fall through for other types of resource allocations. */
5535 	}
5536 #endif
5537 
5538 #ifdef NEW_PCIB
5539 	/*
5540 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5541 	 * those allocations just pass the request up the tree.
5542 	 */
5543 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5544 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5545 		switch (rid) {
5546 		case PCIR_IOBASEL_1:
5547 		case PCIR_MEMBASE_1:
5548 		case PCIR_PMBASEL_1:
5549 			return (bus_generic_release_resource(dev, child, type,
5550 			    rid, r));
5551 		}
5552 	}
5553 #endif
5554 
5555 	rl = &dinfo->resources;
5556 	return (resource_list_release(rl, dev, child, type, rid, r));
5557 }
5558 
5559 int
5560 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5561     struct resource *r)
5562 {
5563 	struct pci_devinfo *dinfo;
5564 	int error;
5565 
5566 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5567 	if (error)
5568 		return (error);
5569 
5570 	/* Enable decoding in the command register when activating BARs. */
5571 	if (device_get_parent(child) == dev) {
5572 		/* Device ROMs need their decoding explicitly enabled. */
5573 		dinfo = device_get_ivars(child);
5574 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5575 			pci_write_bar(child, pci_find_bar(child, rid),
5576 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5577 		switch (type) {
5578 		case SYS_RES_IOPORT:
5579 		case SYS_RES_MEMORY:
5580 			error = PCI_ENABLE_IO(dev, child, type);
5581 			break;
5582 		}
5583 	}
5584 	return (error);
5585 }
5586 
5587 int
5588 pci_deactivate_resource(device_t dev, device_t child, int type,
5589     int rid, struct resource *r)
5590 {
5591 	struct pci_devinfo *dinfo;
5592 	int error;
5593 
5594 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5595 	if (error)
5596 		return (error);
5597 
5598 	/* Disable decoding for device ROMs. */
5599 	if (device_get_parent(child) == dev) {
5600 		dinfo = device_get_ivars(child);
5601 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5602 			pci_write_bar(child, pci_find_bar(child, rid),
5603 			    rman_get_start(r));
5604 	}
5605 	return (0);
5606 }
5607 
5608 void
5609 pci_child_deleted(device_t dev, device_t child)
5610 {
5611 	struct resource_list_entry *rle;
5612 	struct resource_list *rl;
5613 	struct pci_devinfo *dinfo;
5614 
5615 	dinfo = device_get_ivars(child);
5616 	rl = &dinfo->resources;
5617 
5618 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5619 
5620 	/* Turn off access to resources we're about to free */
5621 	if (bus_child_present(child) != 0) {
5622 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5623 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5624 
5625 		pci_disable_busmaster(child);
5626 	}
5627 
5628 	/* Free all allocated resources */
5629 	STAILQ_FOREACH(rle, rl, link) {
5630 		if (rle->res) {
5631 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5632 			    resource_list_busy(rl, rle->type, rle->rid)) {
5633 				pci_printf(&dinfo->cfg,
5634 				    "Resource still owned, oops. "
5635 				    "(type=%d, rid=%d, addr=%lx)\n",
5636 				    rle->type, rle->rid,
5637 				    rman_get_start(rle->res));
5638 				bus_release_resource(child, rle->type, rle->rid,
5639 				    rle->res);
5640 			}
5641 			resource_list_unreserve(rl, dev, child, rle->type,
5642 			    rle->rid);
5643 		}
5644 	}
5645 	resource_list_free(rl);
5646 
5647 	pci_freecfg(dinfo);
5648 }
5649 
5650 void
5651 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5652 {
5653 	struct pci_devinfo *dinfo;
5654 	struct resource_list *rl;
5655 	struct resource_list_entry *rle;
5656 
5657 	if (device_get_parent(child) != dev)
5658 		return;
5659 
5660 	dinfo = device_get_ivars(child);
5661 	rl = &dinfo->resources;
5662 	rle = resource_list_find(rl, type, rid);
5663 	if (rle == NULL)
5664 		return;
5665 
5666 	if (rle->res) {
5667 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5668 		    resource_list_busy(rl, type, rid)) {
5669 			device_printf(dev, "delete_resource: "
5670 			    "Resource still owned by child, oops. "
5671 			    "(type=%d, rid=%d, addr=%jx)\n",
5672 			    type, rid, rman_get_start(rle->res));
5673 			return;
5674 		}
5675 		resource_list_unreserve(rl, dev, child, type, rid);
5676 	}
5677 	resource_list_delete(rl, type, rid);
5678 }
5679 
5680 struct resource_list *
5681 pci_get_resource_list (device_t dev, device_t child)
5682 {
5683 	struct pci_devinfo *dinfo = device_get_ivars(child);
5684 
5685 	return (&dinfo->resources);
5686 }
5687 
5688 #ifdef ACPI_DMAR
5689 bus_dma_tag_t acpi_iommu_get_dma_tag(device_t dev, device_t child);
5690 bus_dma_tag_t
5691 pci_get_dma_tag(device_t bus, device_t dev)
5692 {
5693 	bus_dma_tag_t tag;
5694 	struct pci_softc *sc;
5695 
5696 	if (device_get_parent(dev) == bus) {
5697 		/* try iommu and return if it works */
5698 		tag = acpi_iommu_get_dma_tag(bus, dev);
5699 	} else
5700 		tag = NULL;
5701 	if (tag == NULL) {
5702 		sc = device_get_softc(bus);
5703 		tag = sc->sc_dma_tag;
5704 	}
5705 	return (tag);
5706 }
5707 #else
5708 bus_dma_tag_t
5709 pci_get_dma_tag(device_t bus, device_t dev)
5710 {
5711 	struct pci_softc *sc = device_get_softc(bus);
5712 
5713 	return (sc->sc_dma_tag);
5714 }
5715 #endif
5716 
5717 uint32_t
5718 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5719 {
5720 	struct pci_devinfo *dinfo = device_get_ivars(child);
5721 	pcicfgregs *cfg = &dinfo->cfg;
5722 
5723 #ifdef PCI_IOV
5724 	/*
5725 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5726 	 * emulate them here.
5727 	 */
5728 	if (cfg->flags & PCICFG_VF) {
5729 		if (reg == PCIR_VENDOR) {
5730 			switch (width) {
5731 			case 4:
5732 				return (cfg->device << 16 | cfg->vendor);
5733 			case 2:
5734 				return (cfg->vendor);
5735 			case 1:
5736 				return (cfg->vendor & 0xff);
5737 			default:
5738 				return (0xffffffff);
5739 			}
5740 		} else if (reg == PCIR_DEVICE) {
5741 			switch (width) {
5742 			/* Note that an unaligned 4-byte read is an error. */
5743 			case 2:
5744 				return (cfg->device);
5745 			case 1:
5746 				return (cfg->device & 0xff);
5747 			default:
5748 				return (0xffffffff);
5749 			}
5750 		}
5751 	}
5752 #endif
5753 
5754 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5755 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5756 }
5757 
5758 void
5759 pci_write_config_method(device_t dev, device_t child, int reg,
5760     uint32_t val, int width)
5761 {
5762 	struct pci_devinfo *dinfo = device_get_ivars(child);
5763 	pcicfgregs *cfg = &dinfo->cfg;
5764 
5765 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5766 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5767 }
5768 
5769 int
5770 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5771     size_t buflen)
5772 {
5773 
5774 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5775 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5776 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5777 	return (0);
5778 }
5779 
5780 int
5781 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5782     size_t buflen)
5783 {
5784 	struct pci_devinfo *dinfo;
5785 	pcicfgregs *cfg;
5786 
5787 	dinfo = device_get_ivars(child);
5788 	cfg = &dinfo->cfg;
5789 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5790 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5791 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5792 	    cfg->progif);
5793 	return (0);
5794 }
5795 
5796 int
5797 pci_assign_interrupt_method(device_t dev, device_t child)
5798 {
5799 	struct pci_devinfo *dinfo = device_get_ivars(child);
5800 	pcicfgregs *cfg = &dinfo->cfg;
5801 
5802 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5803 	    cfg->intpin));
5804 }
5805 
5806 static void
5807 pci_lookup(void *arg, const char *name, device_t *dev)
5808 {
5809 	long val;
5810 	char *end;
5811 	int domain, bus, slot, func;
5812 
5813 	if (*dev != NULL)
5814 		return;
5815 
5816 	/*
5817 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5818 	 * pciB:S:F.  In the latter case, the domain is assumed to
5819 	 * be zero.
5820 	 */
5821 	if (strncmp(name, "pci", 3) != 0)
5822 		return;
5823 	val = strtol(name + 3, &end, 10);
5824 	if (val < 0 || val > INT_MAX || *end != ':')
5825 		return;
5826 	domain = val;
5827 	val = strtol(end + 1, &end, 10);
5828 	if (val < 0 || val > INT_MAX || *end != ':')
5829 		return;
5830 	bus = val;
5831 	val = strtol(end + 1, &end, 10);
5832 	if (val < 0 || val > INT_MAX)
5833 		return;
5834 	slot = val;
5835 	if (*end == ':') {
5836 		val = strtol(end + 1, &end, 10);
5837 		if (val < 0 || val > INT_MAX || *end != '\0')
5838 			return;
5839 		func = val;
5840 	} else if (*end == '\0') {
5841 		func = slot;
5842 		slot = bus;
5843 		bus = domain;
5844 		domain = 0;
5845 	} else
5846 		return;
5847 
5848 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5849 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5850 		return;
5851 
5852 	*dev = pci_find_dbsf(domain, bus, slot, func);
5853 }
5854 
5855 static int
5856 pci_modevent(module_t mod, int what, void *arg)
5857 {
5858 	static struct cdev *pci_cdev;
5859 	static eventhandler_tag tag;
5860 
5861 	switch (what) {
5862 	case MOD_LOAD:
5863 		STAILQ_INIT(&pci_devq);
5864 		pci_generation = 0;
5865 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5866 		    "pci");
5867 		pci_load_vendor_data();
5868 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5869 		    1000);
5870 		break;
5871 
5872 	case MOD_UNLOAD:
5873 		if (tag != NULL)
5874 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5875 		destroy_dev(pci_cdev);
5876 		break;
5877 	}
5878 
5879 	return (0);
5880 }
5881 
5882 static void
5883 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5884 {
5885 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5886 	struct pcicfg_pcie *cfg;
5887 	int version, pos;
5888 
5889 	cfg = &dinfo->cfg.pcie;
5890 	pos = cfg->pcie_location;
5891 
5892 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5893 
5894 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5895 
5896 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5897 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5898 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5899 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5900 
5901 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5902 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5903 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5904 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5905 
5906 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5907 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5908 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5909 
5910 	if (version > 1) {
5911 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5912 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5913 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5914 	}
5915 #undef WREG
5916 }
5917 
5918 static void
5919 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5920 {
5921 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5922 	    dinfo->cfg.pcix.pcix_command,  2);
5923 }
5924 
5925 void
5926 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5927 {
5928 
5929 	/*
5930 	 * Restore the device to full power mode.  We must do this
5931 	 * before we restore the registers because moving from D3 to
5932 	 * D0 will cause the chip's BARs and some other registers to
5933 	 * be reset to some unknown power on reset values.  Cut down
5934 	 * the noise on boot by doing nothing if we are already in
5935 	 * state D0.
5936 	 */
5937 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5938 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5939 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5940 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5941 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5942 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5943 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5944 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5945 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5946 	case PCIM_HDRTYPE_NORMAL:
5947 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5948 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5949 		break;
5950 	case PCIM_HDRTYPE_BRIDGE:
5951 		pci_write_config(dev, PCIR_SECLAT_1,
5952 		    dinfo->cfg.bridge.br_seclat, 1);
5953 		pci_write_config(dev, PCIR_SUBBUS_1,
5954 		    dinfo->cfg.bridge.br_subbus, 1);
5955 		pci_write_config(dev, PCIR_SECBUS_1,
5956 		    dinfo->cfg.bridge.br_secbus, 1);
5957 		pci_write_config(dev, PCIR_PRIBUS_1,
5958 		    dinfo->cfg.bridge.br_pribus, 1);
5959 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5960 		    dinfo->cfg.bridge.br_control, 2);
5961 		break;
5962 	case PCIM_HDRTYPE_CARDBUS:
5963 		pci_write_config(dev, PCIR_SECLAT_2,
5964 		    dinfo->cfg.bridge.br_seclat, 1);
5965 		pci_write_config(dev, PCIR_SUBBUS_2,
5966 		    dinfo->cfg.bridge.br_subbus, 1);
5967 		pci_write_config(dev, PCIR_SECBUS_2,
5968 		    dinfo->cfg.bridge.br_secbus, 1);
5969 		pci_write_config(dev, PCIR_PRIBUS_2,
5970 		    dinfo->cfg.bridge.br_pribus, 1);
5971 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5972 		    dinfo->cfg.bridge.br_control, 2);
5973 		break;
5974 	}
5975 	pci_restore_bars(dev);
5976 
5977 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
5978 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5979 
5980 	/*
5981 	 * Restore extended capabilities for PCI-Express and PCI-X
5982 	 */
5983 	if (dinfo->cfg.pcie.pcie_location != 0)
5984 		pci_cfg_restore_pcie(dev, dinfo);
5985 	if (dinfo->cfg.pcix.pcix_location != 0)
5986 		pci_cfg_restore_pcix(dev, dinfo);
5987 
5988 	/* Restore MSI and MSI-X configurations if they are present. */
5989 	if (dinfo->cfg.msi.msi_location != 0)
5990 		pci_resume_msi(dev);
5991 	if (dinfo->cfg.msix.msix_location != 0)
5992 		pci_resume_msix(dev);
5993 
5994 #ifdef PCI_IOV
5995 	if (dinfo->cfg.iov != NULL)
5996 		pci_iov_cfg_restore(dev, dinfo);
5997 #endif
5998 }
5999 
6000 static void
6001 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6002 {
6003 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6004 	struct pcicfg_pcie *cfg;
6005 	int version, pos;
6006 
6007 	cfg = &dinfo->cfg.pcie;
6008 	pos = cfg->pcie_location;
6009 
6010 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6011 
6012 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6013 
6014 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6015 
6016 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6017 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6018 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6019 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6020 
6021 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6022 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6023 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6024 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6025 
6026 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6027 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6028 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6029 
6030 	if (version > 1) {
6031 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6032 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6033 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6034 	}
6035 #undef RREG
6036 }
6037 
6038 static void
6039 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6040 {
6041 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6042 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6043 }
6044 
6045 void
6046 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6047 {
6048 	uint32_t cls;
6049 	int ps;
6050 
6051 	/*
6052 	 * Some drivers apparently write to these registers w/o updating our
6053 	 * cached copy.  No harm happens if we update the copy, so do so here
6054 	 * so we can restore them.  The COMMAND register is modified by the
6055 	 * bus w/o updating the cache.  This should represent the normally
6056 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6057 	 */
6058 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6059 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6060 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6061 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6062 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6063 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6064 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6065 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6066 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6067 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6068 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6069 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6070 	case PCIM_HDRTYPE_NORMAL:
6071 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6072 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6073 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6074 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6075 		break;
6076 	case PCIM_HDRTYPE_BRIDGE:
6077 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6078 		    PCIR_SECLAT_1, 1);
6079 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6080 		    PCIR_SUBBUS_1, 1);
6081 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6082 		    PCIR_SECBUS_1, 1);
6083 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6084 		    PCIR_PRIBUS_1, 1);
6085 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6086 		    PCIR_BRIDGECTL_1, 2);
6087 		break;
6088 	case PCIM_HDRTYPE_CARDBUS:
6089 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6090 		    PCIR_SECLAT_2, 1);
6091 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6092 		    PCIR_SUBBUS_2, 1);
6093 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6094 		    PCIR_SECBUS_2, 1);
6095 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6096 		    PCIR_PRIBUS_2, 1);
6097 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6098 		    PCIR_BRIDGECTL_2, 2);
6099 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6100 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6101 		break;
6102 	}
6103 
6104 	if (dinfo->cfg.pcie.pcie_location != 0)
6105 		pci_cfg_save_pcie(dev, dinfo);
6106 
6107 	if (dinfo->cfg.pcix.pcix_location != 0)
6108 		pci_cfg_save_pcix(dev, dinfo);
6109 
6110 #ifdef PCI_IOV
6111 	if (dinfo->cfg.iov != NULL)
6112 		pci_iov_cfg_save(dev, dinfo);
6113 #endif
6114 
6115 	/*
6116 	 * don't set the state for display devices, base peripherals and
6117 	 * memory devices since bad things happen when they are powered down.
6118 	 * We should (a) have drivers that can easily detach and (b) use
6119 	 * generic drivers for these devices so that some device actually
6120 	 * attaches.  We need to make sure that when we implement (a) we don't
6121 	 * power the device down on a reattach.
6122 	 */
6123 	cls = pci_get_class(dev);
6124 	if (!setstate)
6125 		return;
6126 	switch (pci_do_power_nodriver)
6127 	{
6128 		case 0:		/* NO powerdown at all */
6129 			return;
6130 		case 1:		/* Conservative about what to power down */
6131 			if (cls == PCIC_STORAGE)
6132 				return;
6133 			/*FALLTHROUGH*/
6134 		case 2:		/* Aggressive about what to power down */
6135 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6136 			    cls == PCIC_BASEPERIPH)
6137 				return;
6138 			/*FALLTHROUGH*/
6139 		case 3:		/* Power down everything */
6140 			break;
6141 	}
6142 	/*
6143 	 * PCI spec says we can only go into D3 state from D0 state.
6144 	 * Transition from D[12] into D0 before going to D3 state.
6145 	 */
6146 	ps = pci_get_powerstate(dev);
6147 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6148 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6149 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6150 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6151 }
6152 
6153 /* Wrapper APIs suitable for device driver use. */
6154 void
6155 pci_save_state(device_t dev)
6156 {
6157 	struct pci_devinfo *dinfo;
6158 
6159 	dinfo = device_get_ivars(dev);
6160 	pci_cfg_save(dev, dinfo, 0);
6161 }
6162 
6163 void
6164 pci_restore_state(device_t dev)
6165 {
6166 	struct pci_devinfo *dinfo;
6167 
6168 	dinfo = device_get_ivars(dev);
6169 	pci_cfg_restore(dev, dinfo);
6170 }
6171 
6172 static int
6173 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6174     uintptr_t *id)
6175 {
6176 
6177 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6178 }
6179 
6180 /* Find the upstream port of a given PCI device in a root complex. */
6181 device_t
6182 pci_find_pcie_root_port(device_t dev)
6183 {
6184 	struct pci_devinfo *dinfo;
6185 	devclass_t pci_class;
6186 	device_t pcib, bus;
6187 
6188 	pci_class = devclass_find("pci");
6189 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6190 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6191 
6192 	/*
6193 	 * Walk the bridge hierarchy until we find a PCI-e root
6194 	 * port or a non-PCI device.
6195 	 */
6196 	for (;;) {
6197 		bus = device_get_parent(dev);
6198 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6199 		    device_get_nameunit(dev)));
6200 
6201 		pcib = device_get_parent(bus);
6202 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6203 		    device_get_nameunit(bus)));
6204 
6205 		/*
6206 		 * pcib's parent must be a PCI bus for this to be a
6207 		 * PCI-PCI bridge.
6208 		 */
6209 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6210 			return (NULL);
6211 
6212 		dinfo = device_get_ivars(pcib);
6213 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6214 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6215 			return (pcib);
6216 
6217 		dev = pcib;
6218 	}
6219 }
6220 
6221 /*
6222  * Wait for pending transactions to complete on a PCI-express function.
6223  *
6224  * The maximum delay is specified in milliseconds in max_delay.  Note
6225  * that this function may sleep.
6226  *
6227  * Returns true if the function is idle and false if the timeout is
6228  * exceeded.  If dev is not a PCI-express function, this returns true.
6229  */
6230 bool
6231 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6232 {
6233 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6234 	uint16_t sta;
6235 	int cap;
6236 
6237 	cap = dinfo->cfg.pcie.pcie_location;
6238 	if (cap == 0)
6239 		return (true);
6240 
6241 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6242 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6243 		if (max_delay == 0)
6244 			return (false);
6245 
6246 		/* Poll once every 100 milliseconds up to the timeout. */
6247 		if (max_delay > 100) {
6248 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6249 			max_delay -= 100;
6250 		} else {
6251 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6252 			    C_HARDCLOCK);
6253 			max_delay = 0;
6254 		}
6255 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6256 	}
6257 
6258 	return (true);
6259 }
6260 
6261 /*
6262  * Determine the maximum Completion Timeout in microseconds.
6263  *
6264  * For non-PCI-express functions this returns 0.
6265  */
6266 int
6267 pcie_get_max_completion_timeout(device_t dev)
6268 {
6269 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6270 	int cap;
6271 
6272 	cap = dinfo->cfg.pcie.pcie_location;
6273 	if (cap == 0)
6274 		return (0);
6275 
6276 	/*
6277 	 * Functions using the 1.x spec use the default timeout range of
6278 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6279 	 * support programmable timeouts also use this range.
6280 	 */
6281 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6282 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6283 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6284 		return (50 * 1000);
6285 
6286 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6287 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6288 	case PCIEM_CTL2_COMP_TIMO_100US:
6289 		return (100);
6290 	case PCIEM_CTL2_COMP_TIMO_10MS:
6291 		return (10 * 1000);
6292 	case PCIEM_CTL2_COMP_TIMO_55MS:
6293 		return (55 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_210MS:
6295 		return (210 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_900MS:
6297 		return (900 * 1000);
6298 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6299 		return (3500 * 1000);
6300 	case PCIEM_CTL2_COMP_TIMO_13S:
6301 		return (13 * 1000 * 1000);
6302 	case PCIEM_CTL2_COMP_TIMO_64S:
6303 		return (64 * 1000 * 1000);
6304 	default:
6305 		return (50 * 1000);
6306 	}
6307 }
6308 
6309 /*
6310  * Perform a Function Level Reset (FLR) on a device.
6311  *
6312  * This function first waits for any pending transactions to complete
6313  * within the timeout specified by max_delay.  If transactions are
6314  * still pending, the function will return false without attempting a
6315  * reset.
6316  *
6317  * If dev is not a PCI-express function or does not support FLR, this
6318  * function returns false.
6319  *
6320  * Note that no registers are saved or restored.  The caller is
6321  * responsible for saving and restoring any registers including
6322  * PCI-standard registers via pci_save_state() and
6323  * pci_restore_state().
6324  */
6325 bool
6326 pcie_flr(device_t dev, u_int max_delay, bool force)
6327 {
6328 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6329 	uint16_t cmd, ctl;
6330 	int compl_delay;
6331 	int cap;
6332 
6333 	cap = dinfo->cfg.pcie.pcie_location;
6334 	if (cap == 0)
6335 		return (false);
6336 
6337 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6338 		return (false);
6339 
6340 	/*
6341 	 * Disable busmastering to prevent generation of new
6342 	 * transactions while waiting for the device to go idle.  If
6343 	 * the idle timeout fails, the command register is restored
6344 	 * which will re-enable busmastering.
6345 	 */
6346 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6347 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6348 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6349 		if (!force) {
6350 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6351 			return (false);
6352 		}
6353 		pci_printf(&dinfo->cfg,
6354 		    "Resetting with transactions pending after %d ms\n",
6355 		    max_delay);
6356 
6357 		/*
6358 		 * Extend the post-FLR delay to cover the maximum
6359 		 * Completion Timeout delay of anything in flight
6360 		 * during the FLR delay.  Enforce a minimum delay of
6361 		 * at least 10ms.
6362 		 */
6363 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6364 		if (compl_delay < 10)
6365 			compl_delay = 10;
6366 	} else
6367 		compl_delay = 0;
6368 
6369 	/* Initiate the reset. */
6370 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6371 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6372 	    PCIEM_CTL_INITIATE_FLR, 2);
6373 
6374 	/* Wait for 100ms. */
6375 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6376 
6377 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6378 	    PCIEM_STA_TRANSACTION_PND)
6379 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6380 	return (true);
6381 }
6382 
6383 /*
6384  * Attempt a power-management reset by cycling the device in/out of D3
6385  * state.  PCI spec says we can only go into D3 state from D0 state.
6386  * Transition from D[12] into D0 before going to D3 state.
6387  */
6388 int
6389 pci_power_reset(device_t dev)
6390 {
6391 	int ps;
6392 
6393 	ps = pci_get_powerstate(dev);
6394 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6395 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6396 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6397 	pci_set_powerstate(dev, ps);
6398 	return (0);
6399 }
6400 
6401 /*
6402  * Try link drop and retrain of the downstream port of upstream
6403  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6404  * cause Conventional Hot reset of the device in the slot.
6405  * Alternative, for PCIe, could be the secondary bus reset initiatied
6406  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6407  */
6408 int
6409 pcie_link_reset(device_t port, int pcie_location)
6410 {
6411 	uint16_t v;
6412 
6413 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6414 	v |= PCIEM_LINK_CTL_LINK_DIS;
6415 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6416 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6417 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6418 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6419 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6420 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6421 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6422 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6423 }
6424 
6425 static int
6426 pci_reset_post(device_t dev, device_t child)
6427 {
6428 
6429 	if (dev == device_get_parent(child))
6430 		pci_restore_state(child);
6431 	return (0);
6432 }
6433 
6434 static int
6435 pci_reset_prepare(device_t dev, device_t child)
6436 {
6437 
6438 	if (dev == device_get_parent(child))
6439 		pci_save_state(child);
6440 	return (0);
6441 }
6442 
6443 static int
6444 pci_reset_child(device_t dev, device_t child, int flags)
6445 {
6446 	int error;
6447 
6448 	if (dev == NULL || device_get_parent(child) != dev)
6449 		return (0);
6450 	if ((flags & DEVF_RESET_DETACH) != 0) {
6451 		error = device_get_state(child) == DS_ATTACHED ?
6452 		    device_detach(child) : 0;
6453 	} else {
6454 		error = BUS_SUSPEND_CHILD(dev, child);
6455 	}
6456 	if (error == 0) {
6457 		if (!pcie_flr(child, 1000, false)) {
6458 			error = BUS_RESET_PREPARE(dev, child);
6459 			if (error == 0)
6460 				pci_power_reset(child);
6461 			BUS_RESET_POST(dev, child);
6462 		}
6463 		if ((flags & DEVF_RESET_DETACH) != 0)
6464 			device_probe_and_attach(child);
6465 		else
6466 			BUS_RESUME_CHILD(dev, child);
6467 	}
6468 	return (error);
6469 }
6470 
6471 const struct pci_device_table *
6472 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6473 {
6474 	bool match;
6475 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6476 
6477 	vendor = pci_get_vendor(child);
6478 	device = pci_get_device(child);
6479 	subvendor = pci_get_subvendor(child);
6480 	subdevice = pci_get_subdevice(child);
6481 	class = pci_get_class(child);
6482 	subclass = pci_get_subclass(child);
6483 	revid = pci_get_revid(child);
6484 	while (nelt-- > 0) {
6485 		match = true;
6486 		if (id->match_flag_vendor)
6487 			match &= vendor == id->vendor;
6488 		if (id->match_flag_device)
6489 			match &= device == id->device;
6490 		if (id->match_flag_subvendor)
6491 			match &= subvendor == id->subvendor;
6492 		if (id->match_flag_subdevice)
6493 			match &= subdevice == id->subdevice;
6494 		if (id->match_flag_class)
6495 			match &= class == id->class_id;
6496 		if (id->match_flag_subclass)
6497 			match &= subclass == id->subclass;
6498 		if (id->match_flag_revid)
6499 			match &= revid == id->revid;
6500 		if (match)
6501 			return (id);
6502 		id++;
6503 	}
6504 	return (NULL);
6505 }
6506 
6507 static void
6508 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6509 {
6510 	const char *dev_name;
6511 	device_t dev;
6512 
6513 	dev = dinfo->cfg.dev;
6514 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6515 	    dinfo->cfg.slot, dinfo->cfg.func);
6516 	dev_name = device_get_name(dev);
6517 	if (dev_name != NULL)
6518 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6519 }
6520 
6521 void
6522 pci_print_faulted_dev(void)
6523 {
6524 	struct pci_devinfo *dinfo;
6525 	device_t dev;
6526 	int aer, i;
6527 	uint32_t r1, r2;
6528 	uint16_t status;
6529 
6530 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6531 		dev = dinfo->cfg.dev;
6532 		status = pci_read_config(dev, PCIR_STATUS, 2);
6533 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6534 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6535 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6536 		if (status != 0) {
6537 			pci_print_faulted_dev_name(dinfo);
6538 			printf(" error 0x%04x\n", status);
6539 		}
6540 		if (dinfo->cfg.pcie.pcie_location != 0) {
6541 			status = pci_read_config(dev,
6542 			    dinfo->cfg.pcie.pcie_location +
6543 			    PCIER_DEVICE_STA, 2);
6544 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6545 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6546 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6547 				pci_print_faulted_dev_name(dinfo);
6548 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6549 				    pci_read_config(dev,
6550 				    dinfo->cfg.pcie.pcie_location +
6551 				    PCIER_DEVICE_CTL, 2),
6552 				    status);
6553 			}
6554 		}
6555 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6556 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6557 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6558 			if (r1 != 0 || r2 != 0) {
6559 				pci_print_faulted_dev_name(dinfo);
6560 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6561 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6562 				    r1, pci_read_config(dev, aer +
6563 				    PCIR_AER_UC_MASK, 4),
6564 				    pci_read_config(dev, aer +
6565 				    PCIR_AER_UC_SEVERITY, 4),
6566 				    r2, pci_read_config(dev, aer +
6567 				    PCIR_AER_COR_MASK, 4),
6568 				    pci_read_config(dev, aer +
6569 				    PCIR_AER_CAP_CONTROL, 4));
6570 				for (i = 0; i < 4; i++) {
6571 					r1 = pci_read_config(dev, aer +
6572 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6573 					printf("    HL%d: 0x%08x\n", i, r1);
6574 				}
6575 			}
6576 		}
6577 	}
6578 }
6579 
6580 #ifdef DDB
6581 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6582 {
6583 
6584 	pci_print_faulted_dev();
6585 }
6586 
6587 static void
6588 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6589 {
6590 	device_t dev;
6591 	int aer;
6592 	uint32_t r;
6593 
6594 	dev = dinfo->cfg.dev;
6595 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6596 	    PCIER_DEVICE_STA, 2);
6597 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6598 	    PCIER_DEVICE_STA, r, 2);
6599 
6600 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6601 		return;
6602 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6603 	if (r != 0)
6604 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6605 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6606 	if (r != 0)
6607 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6608 }
6609 
6610 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6611 {
6612 	struct pci_devinfo *dinfo;
6613 	device_t dev;
6614 	uint16_t status, status1;
6615 
6616 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6617 		dev = dinfo->cfg.dev;
6618 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6619 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6620 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6621 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6622 		if (status1 != 0) {
6623 			status &= ~status1;
6624 			pci_write_config(dev, PCIR_STATUS, status, 2);
6625 		}
6626 		if (dinfo->cfg.pcie.pcie_location != 0)
6627 			db_clear_pcie_errors(dinfo);
6628 	}
6629 }
6630 #endif
6631