xref: /freebsd/sys/dev/pci/pci.c (revision 731d06abf2105cc0873fa84e972178f9f37ca760)
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/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/limits.h>
42 #include <sys/linker.h>
43 #include <sys/fcntl.h>
44 #include <sys/conf.h>
45 #include <sys/kernel.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/endian.h>
49 
50 #include <vm/vm.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_extern.h>
53 
54 #include <sys/bus.h>
55 #include <machine/bus.h>
56 #include <sys/rman.h>
57 #include <machine/resource.h>
58 #include <machine/stdarg.h>
59 
60 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
61 #include <machine/intr_machdep.h>
62 #endif
63 
64 #include <sys/pciio.h>
65 #include <dev/pci/pcireg.h>
66 #include <dev/pci/pcivar.h>
67 #include <dev/pci/pci_private.h>
68 
69 #ifdef PCI_IOV
70 #include <sys/nv.h>
71 #include <dev/pci/pci_iov_private.h>
72 #endif
73 
74 #include <dev/usb/controller/xhcireg.h>
75 #include <dev/usb/controller/ehcireg.h>
76 #include <dev/usb/controller/ohcireg.h>
77 #include <dev/usb/controller/uhcireg.h>
78 
79 #include "pcib_if.h"
80 #include "pci_if.h"
81 
82 #define	PCIR_IS_BIOS(cfg, reg)						\
83 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
84 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
85 
86 static int		pci_has_quirk(uint32_t devid, int quirk);
87 static pci_addr_t	pci_mapbase(uint64_t mapreg);
88 static const char	*pci_maptype(uint64_t mapreg);
89 static int		pci_maprange(uint64_t mapreg);
90 static pci_addr_t	pci_rombase(uint64_t mapreg);
91 static int		pci_romsize(uint64_t testval);
92 static void		pci_fixancient(pcicfgregs *cfg);
93 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
94 
95 static int		pci_porten(device_t dev);
96 static int		pci_memen(device_t dev);
97 static void		pci_assign_interrupt(device_t bus, device_t dev,
98 			    int force_route);
99 static int		pci_add_map(device_t bus, device_t dev, int reg,
100 			    struct resource_list *rl, int force, int prefetch);
101 static int		pci_probe(device_t dev);
102 static int		pci_attach(device_t dev);
103 static int		pci_detach(device_t dev);
104 static void		pci_load_vendor_data(void);
105 static int		pci_describe_parse_line(char **ptr, int *vendor,
106 			    int *device, char **desc);
107 static char		*pci_describe_device(device_t dev);
108 static int		pci_modevent(module_t mod, int what, void *arg);
109 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
110 			    pcicfgregs *cfg);
111 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
112 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
113 			    int reg, uint32_t *data);
114 #if 0
115 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
116 			    int reg, uint32_t data);
117 #endif
118 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
119 static void		pci_mask_msix(device_t dev, u_int index);
120 static void		pci_unmask_msix(device_t dev, u_int index);
121 static int		pci_msi_blacklisted(void);
122 static int		pci_msix_blacklisted(void);
123 static void		pci_resume_msi(device_t dev);
124 static void		pci_resume_msix(device_t dev);
125 static int		pci_remap_intr_method(device_t bus, device_t dev,
126 			    u_int irq);
127 static void		pci_hint_device_unit(device_t acdev, device_t child,
128 			    const char *name, int *unitp);
129 static int		pci_reset_post(device_t dev, device_t child);
130 static int		pci_reset_prepare(device_t dev, device_t child);
131 static int		pci_reset_child(device_t dev, device_t child,
132 			    int flags);
133 
134 static int		pci_get_id_method(device_t dev, device_t child,
135 			    enum pci_id_type type, uintptr_t *rid);
136 
137 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
138     int b, int s, int f, uint16_t vid, uint16_t did);
139 
140 static device_method_t pci_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,		pci_probe),
143 	DEVMETHOD(device_attach,	pci_attach),
144 	DEVMETHOD(device_detach,	pci_detach),
145 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,	bus_generic_suspend),
147 	DEVMETHOD(device_resume,	pci_resume),
148 
149 	/* Bus interface */
150 	DEVMETHOD(bus_print_child,	pci_print_child),
151 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
152 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
153 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
154 	DEVMETHOD(bus_driver_added,	pci_driver_added),
155 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
156 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
157 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
158 	DEVMETHOD(bus_reset_post,	pci_reset_post),
159 	DEVMETHOD(bus_reset_child,	pci_reset_child),
160 
161 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
162 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
163 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
164 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
165 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
166 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
167 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
168 	DEVMETHOD(bus_release_resource,	pci_release_resource),
169 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
170 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
171 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
172 	DEVMETHOD(bus_child_detached,	pci_child_detached),
173 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
174 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
175 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
176 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
177 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
178 	DEVMETHOD(bus_resume_child,	pci_resume_child),
179 	DEVMETHOD(bus_rescan,		pci_rescan_method),
180 
181 	/* PCI interface */
182 	DEVMETHOD(pci_read_config,	pci_read_config_method),
183 	DEVMETHOD(pci_write_config,	pci_write_config_method),
184 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
185 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
186 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
187 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
188 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
189 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
190 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
191 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
192 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
193 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
194 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
195 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
196 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
197 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
198 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
199 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
200 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
201 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
202 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
203 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
204 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
205 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
206 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
207 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
208 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
209 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
210 	DEVMETHOD(pci_get_id,		pci_get_id_method),
211 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
212 	DEVMETHOD(pci_child_added,	pci_child_added_method),
213 #ifdef PCI_IOV
214 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
215 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
216 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
217 #endif
218 
219 	DEVMETHOD_END
220 };
221 
222 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
223 
224 static devclass_t pci_devclass;
225 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
226     BUS_PASS_BUS);
227 MODULE_VERSION(pci, 1);
228 
229 static char	*pci_vendordata;
230 static size_t	pci_vendordata_size;
231 
232 struct pci_quirk {
233 	uint32_t devid;	/* Vendor/device of the card */
234 	int	type;
235 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
236 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
237 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
238 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
239 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
240 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
241 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
242 	int	arg1;
243 	int	arg2;
244 };
245 
246 static const struct pci_quirk pci_quirks[] = {
247 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
248 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
249 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
250 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
251 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
252 
253 	/*
254 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
255 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
256 	 */
257 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
258 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
259 
260 	/*
261 	 * MSI doesn't work on earlier Intel chipsets including
262 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
263 	 */
264 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
265 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 
272 	/*
273 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
274 	 * bridge.
275 	 */
276 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
277 
278 	/*
279 	 * MSI-X allocation doesn't work properly for devices passed through
280 	 * by VMware up to at least ESXi 5.1.
281 	 */
282 	{ 0x079015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCI/PCI-X */
283 	{ 0x07a015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCIe */
284 
285 	/*
286 	 * Some virtualization environments emulate an older chipset
287 	 * but support MSI just fine.  QEMU uses the Intel 82440.
288 	 */
289 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
290 
291 	/*
292 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
293 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
294 	 * It prevents us from attaching hpet(4) when the bit is unset.
295 	 * Note this quirk only affects SB600 revision A13 and earlier.
296 	 * For SB600 A21 and later, firmware must set the bit to hide it.
297 	 * For SB700 and later, it is unused and hardcoded to zero.
298 	 */
299 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
300 
301 	/*
302 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
303 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
304 	 * of the command register is set.
305 	 */
306 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
307 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
309 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
310 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
311 
312 	/*
313 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
314 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
315 	 */
316 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
317 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
318 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
319 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
320 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
321 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
322 
323 	/*
324 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
325 	 * expected place.
326 	 */
327 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
328 
329 	{ 0 }
330 };
331 
332 /* map register information */
333 #define	PCI_MAPMEM	0x01	/* memory map */
334 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
335 #define	PCI_MAPPORT	0x04	/* port map */
336 
337 struct devlist pci_devq;
338 uint32_t pci_generation;
339 uint32_t pci_numdevs = 0;
340 static int pcie_chipset, pcix_chipset;
341 
342 /* sysctl vars */
343 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
344 
345 static int pci_enable_io_modes = 1;
346 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
347     &pci_enable_io_modes, 1,
348     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
349     " enable these bits correctly.  We'd like to do this all the time, but"
350     " there are some peripherals that this causes problems with.");
351 
352 static int pci_do_realloc_bars = 1;
353 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
354     &pci_do_realloc_bars, 0,
355     "Attempt to allocate a new range for any BARs whose original "
356     "firmware-assigned ranges fail to allocate during the initial device scan.");
357 
358 static int pci_do_power_nodriver = 0;
359 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
360     &pci_do_power_nodriver, 0,
361     "Place a function into D3 state when no driver attaches to it.  0 means"
362     " disable.  1 means conservatively place devices into D3 state.  2 means"
363     " aggressively place devices into D3 state.  3 means put absolutely"
364     " everything in D3 state.");
365 
366 int pci_do_power_resume = 1;
367 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
368     &pci_do_power_resume, 1,
369   "Transition from D3 -> D0 on resume.");
370 
371 int pci_do_power_suspend = 1;
372 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
373     &pci_do_power_suspend, 1,
374   "Transition from D0 -> D3 on suspend.");
375 
376 static int pci_do_msi = 1;
377 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
378     "Enable support for MSI interrupts");
379 
380 static int pci_do_msix = 1;
381 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
382     "Enable support for MSI-X interrupts");
383 
384 static int pci_msix_rewrite_table = 0;
385 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
386     &pci_msix_rewrite_table, 0,
387     "Rewrite entire MSI-X table when updating MSI-X entries");
388 
389 static int pci_honor_msi_blacklist = 1;
390 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
391     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
392 
393 #if defined(__i386__) || defined(__amd64__)
394 static int pci_usb_takeover = 1;
395 #else
396 static int pci_usb_takeover = 0;
397 #endif
398 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
399     &pci_usb_takeover, 1,
400     "Enable early takeover of USB controllers. Disable this if you depend on"
401     " BIOS emulation of USB devices, that is you use USB devices (like"
402     " keyboard or mouse) but do not load USB drivers");
403 
404 static int pci_clear_bars;
405 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
406     "Ignore firmware-assigned resources for BARs.");
407 
408 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
409 static int pci_clear_buses;
410 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
411     "Ignore firmware-assigned bus numbers.");
412 #endif
413 
414 static int pci_enable_ari = 1;
415 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
416     0, "Enable support for PCIe Alternative RID Interpretation");
417 
418 static int pci_clear_aer_on_attach = 0;
419 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
420     &pci_clear_aer_on_attach, 0,
421     "Clear port and device AER state on driver attach");
422 
423 static int
424 pci_has_quirk(uint32_t devid, int quirk)
425 {
426 	const struct pci_quirk *q;
427 
428 	for (q = &pci_quirks[0]; q->devid; q++) {
429 		if (q->devid == devid && q->type == quirk)
430 			return (1);
431 	}
432 	return (0);
433 }
434 
435 /* Find a device_t by bus/slot/function in domain 0 */
436 
437 device_t
438 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
439 {
440 
441 	return (pci_find_dbsf(0, bus, slot, func));
442 }
443 
444 /* Find a device_t by domain/bus/slot/function */
445 
446 device_t
447 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
448 {
449 	struct pci_devinfo *dinfo;
450 
451 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
452 		if ((dinfo->cfg.domain == domain) &&
453 		    (dinfo->cfg.bus == bus) &&
454 		    (dinfo->cfg.slot == slot) &&
455 		    (dinfo->cfg.func == func)) {
456 			return (dinfo->cfg.dev);
457 		}
458 	}
459 
460 	return (NULL);
461 }
462 
463 /* Find a device_t by vendor/device ID */
464 
465 device_t
466 pci_find_device(uint16_t vendor, uint16_t device)
467 {
468 	struct pci_devinfo *dinfo;
469 
470 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
471 		if ((dinfo->cfg.vendor == vendor) &&
472 		    (dinfo->cfg.device == device)) {
473 			return (dinfo->cfg.dev);
474 		}
475 	}
476 
477 	return (NULL);
478 }
479 
480 device_t
481 pci_find_class(uint8_t class, uint8_t subclass)
482 {
483 	struct pci_devinfo *dinfo;
484 
485 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
486 		if (dinfo->cfg.baseclass == class &&
487 		    dinfo->cfg.subclass == subclass) {
488 			return (dinfo->cfg.dev);
489 		}
490 	}
491 
492 	return (NULL);
493 }
494 
495 static int
496 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
497 {
498 	va_list ap;
499 	int retval;
500 
501 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
502 	    cfg->func);
503 	va_start(ap, fmt);
504 	retval += vprintf(fmt, ap);
505 	va_end(ap);
506 	return (retval);
507 }
508 
509 /* return base address of memory or port map */
510 
511 static pci_addr_t
512 pci_mapbase(uint64_t mapreg)
513 {
514 
515 	if (PCI_BAR_MEM(mapreg))
516 		return (mapreg & PCIM_BAR_MEM_BASE);
517 	else
518 		return (mapreg & PCIM_BAR_IO_BASE);
519 }
520 
521 /* return map type of memory or port map */
522 
523 static const char *
524 pci_maptype(uint64_t mapreg)
525 {
526 
527 	if (PCI_BAR_IO(mapreg))
528 		return ("I/O Port");
529 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
530 		return ("Prefetchable Memory");
531 	return ("Memory");
532 }
533 
534 /* return log2 of map size decoded for memory or port map */
535 
536 int
537 pci_mapsize(uint64_t testval)
538 {
539 	int ln2size;
540 
541 	testval = pci_mapbase(testval);
542 	ln2size = 0;
543 	if (testval != 0) {
544 		while ((testval & 1) == 0)
545 		{
546 			ln2size++;
547 			testval >>= 1;
548 		}
549 	}
550 	return (ln2size);
551 }
552 
553 /* return base address of device ROM */
554 
555 static pci_addr_t
556 pci_rombase(uint64_t mapreg)
557 {
558 
559 	return (mapreg & PCIM_BIOS_ADDR_MASK);
560 }
561 
562 /* return log2 of map size decided for device ROM */
563 
564 static int
565 pci_romsize(uint64_t testval)
566 {
567 	int ln2size;
568 
569 	testval = pci_rombase(testval);
570 	ln2size = 0;
571 	if (testval != 0) {
572 		while ((testval & 1) == 0)
573 		{
574 			ln2size++;
575 			testval >>= 1;
576 		}
577 	}
578 	return (ln2size);
579 }
580 
581 /* return log2 of address range supported by map register */
582 
583 static int
584 pci_maprange(uint64_t mapreg)
585 {
586 	int ln2range = 0;
587 
588 	if (PCI_BAR_IO(mapreg))
589 		ln2range = 32;
590 	else
591 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
592 		case PCIM_BAR_MEM_32:
593 			ln2range = 32;
594 			break;
595 		case PCIM_BAR_MEM_1MB:
596 			ln2range = 20;
597 			break;
598 		case PCIM_BAR_MEM_64:
599 			ln2range = 64;
600 			break;
601 		}
602 	return (ln2range);
603 }
604 
605 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
606 
607 static void
608 pci_fixancient(pcicfgregs *cfg)
609 {
610 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
611 		return;
612 
613 	/* PCI to PCI bridges use header type 1 */
614 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
615 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
616 }
617 
618 /* extract header type specific config data */
619 
620 static void
621 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
622 {
623 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
624 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
625 	case PCIM_HDRTYPE_NORMAL:
626 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
627 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
628 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
629 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
630 		cfg->nummaps	    = PCI_MAXMAPS_0;
631 		break;
632 	case PCIM_HDRTYPE_BRIDGE:
633 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
634 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
635 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
636 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
637 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
638 		cfg->nummaps	    = PCI_MAXMAPS_1;
639 		break;
640 	case PCIM_HDRTYPE_CARDBUS:
641 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
642 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
643 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
644 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
645 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
646 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
647 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
648 		cfg->nummaps	    = PCI_MAXMAPS_2;
649 		break;
650 	}
651 #undef REG
652 }
653 
654 /* read configuration header into pcicfgregs structure */
655 struct pci_devinfo *
656 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
657 {
658 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
659 	uint16_t vid, did;
660 
661 	vid = REG(PCIR_VENDOR, 2);
662 	did = REG(PCIR_DEVICE, 2);
663 	if (vid != 0xffff)
664 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
665 
666 	return (NULL);
667 }
668 
669 struct pci_devinfo *
670 pci_alloc_devinfo_method(device_t dev)
671 {
672 
673 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
674 	    M_WAITOK | M_ZERO));
675 }
676 
677 static struct pci_devinfo *
678 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
679     uint16_t vid, uint16_t did)
680 {
681 	struct pci_devinfo *devlist_entry;
682 	pcicfgregs *cfg;
683 
684 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
685 
686 	cfg = &devlist_entry->cfg;
687 
688 	cfg->domain		= d;
689 	cfg->bus		= b;
690 	cfg->slot		= s;
691 	cfg->func		= f;
692 	cfg->vendor		= vid;
693 	cfg->device		= did;
694 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
695 	cfg->statreg		= REG(PCIR_STATUS, 2);
696 	cfg->baseclass		= REG(PCIR_CLASS, 1);
697 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
698 	cfg->progif		= REG(PCIR_PROGIF, 1);
699 	cfg->revid		= REG(PCIR_REVID, 1);
700 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
701 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
702 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
703 	cfg->intpin		= REG(PCIR_INTPIN, 1);
704 	cfg->intline		= REG(PCIR_INTLINE, 1);
705 
706 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
707 	cfg->hdrtype		&= ~PCIM_MFDEV;
708 	STAILQ_INIT(&cfg->maps);
709 
710 	cfg->iov		= NULL;
711 
712 	pci_fixancient(cfg);
713 	pci_hdrtypedata(pcib, b, s, f, cfg);
714 
715 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
716 		pci_read_cap(pcib, cfg);
717 
718 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
719 
720 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
721 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
722 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
723 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
724 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
725 
726 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
727 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
728 	devlist_entry->conf.pc_vendor = cfg->vendor;
729 	devlist_entry->conf.pc_device = cfg->device;
730 
731 	devlist_entry->conf.pc_class = cfg->baseclass;
732 	devlist_entry->conf.pc_subclass = cfg->subclass;
733 	devlist_entry->conf.pc_progif = cfg->progif;
734 	devlist_entry->conf.pc_revid = cfg->revid;
735 
736 	pci_numdevs++;
737 	pci_generation++;
738 
739 	return (devlist_entry);
740 }
741 #undef REG
742 
743 static void
744 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
745 {
746 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
747     cfg->ea.ea_location + (n), w)
748 	int num_ent;
749 	int ptr;
750 	int a, b;
751 	uint32_t val;
752 	int ent_size;
753 	uint32_t dw[4];
754 	uint64_t base, max_offset;
755 	struct pci_ea_entry *eae;
756 
757 	if (cfg->ea.ea_location == 0)
758 		return;
759 
760 	STAILQ_INIT(&cfg->ea.ea_entries);
761 
762 	/* Determine the number of entries */
763 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
764 	num_ent &= PCIM_EA_NUM_ENT_MASK;
765 
766 	/* Find the first entry to care of */
767 	ptr = PCIR_EA_FIRST_ENT;
768 
769 	/* Skip DWORD 2 for type 1 functions */
770 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
771 		ptr += 4;
772 
773 	for (a = 0; a < num_ent; a++) {
774 
775 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
776 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
777 
778 		/* Read a number of dwords in the entry */
779 		val = REG(ptr, 4);
780 		ptr += 4;
781 		ent_size = (val & PCIM_EA_ES);
782 
783 		for (b = 0; b < ent_size; b++) {
784 			dw[b] = REG(ptr, 4);
785 			ptr += 4;
786 		}
787 
788 		eae->eae_flags = val;
789 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
790 
791 		base = dw[0] & PCIM_EA_FIELD_MASK;
792 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
793 		b = 2;
794 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
795 			base |= (uint64_t)dw[b] << 32UL;
796 			b++;
797 		}
798 		if (((dw[1] & PCIM_EA_IS_64) != 0)
799 		    && (b < ent_size)) {
800 			max_offset |= (uint64_t)dw[b] << 32UL;
801 			b++;
802 		}
803 
804 		eae->eae_base = base;
805 		eae->eae_max_offset = max_offset;
806 
807 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
808 
809 		if (bootverbose) {
810 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
811 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
812 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
813 		}
814 	}
815 }
816 #undef REG
817 
818 static void
819 pci_read_cap(device_t pcib, pcicfgregs *cfg)
820 {
821 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
822 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
823 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
824 	uint64_t addr;
825 #endif
826 	uint32_t val;
827 	int	ptr, nextptr, ptrptr;
828 
829 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
830 	case PCIM_HDRTYPE_NORMAL:
831 	case PCIM_HDRTYPE_BRIDGE:
832 		ptrptr = PCIR_CAP_PTR;
833 		break;
834 	case PCIM_HDRTYPE_CARDBUS:
835 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
836 		break;
837 	default:
838 		return;		/* no extended capabilities support */
839 	}
840 	nextptr = REG(ptrptr, 1);	/* sanity check? */
841 
842 	/*
843 	 * Read capability entries.
844 	 */
845 	while (nextptr != 0) {
846 		/* Sanity check */
847 		if (nextptr > 255) {
848 			printf("illegal PCI extended capability offset %d\n",
849 			    nextptr);
850 			return;
851 		}
852 		/* Find the next entry */
853 		ptr = nextptr;
854 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
855 
856 		/* Process this entry */
857 		switch (REG(ptr + PCICAP_ID, 1)) {
858 		case PCIY_PMG:		/* PCI power management */
859 			if (cfg->pp.pp_cap == 0) {
860 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
861 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
862 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
863 				if ((nextptr - ptr) > PCIR_POWER_DATA)
864 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
865 			}
866 			break;
867 		case PCIY_HT:		/* HyperTransport */
868 			/* Determine HT-specific capability type. */
869 			val = REG(ptr + PCIR_HT_COMMAND, 2);
870 
871 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
872 				cfg->ht.ht_slave = ptr;
873 
874 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
875 			switch (val & PCIM_HTCMD_CAP_MASK) {
876 			case PCIM_HTCAP_MSI_MAPPING:
877 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
878 					/* Sanity check the mapping window. */
879 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
880 					    4);
881 					addr <<= 32;
882 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
883 					    4);
884 					if (addr != MSI_INTEL_ADDR_BASE)
885 						device_printf(pcib,
886 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
887 						    cfg->domain, cfg->bus,
888 						    cfg->slot, cfg->func,
889 						    (long long)addr);
890 				} else
891 					addr = MSI_INTEL_ADDR_BASE;
892 
893 				cfg->ht.ht_msimap = ptr;
894 				cfg->ht.ht_msictrl = val;
895 				cfg->ht.ht_msiaddr = addr;
896 				break;
897 			}
898 #endif
899 			break;
900 		case PCIY_MSI:		/* PCI MSI */
901 			cfg->msi.msi_location = ptr;
902 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
903 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
904 						     PCIM_MSICTRL_MMC_MASK)>>1);
905 			break;
906 		case PCIY_MSIX:		/* PCI MSI-X */
907 			cfg->msix.msix_location = ptr;
908 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
909 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
910 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
911 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
912 			cfg->msix.msix_table_bar = PCIR_BAR(val &
913 			    PCIM_MSIX_BIR_MASK);
914 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
915 			val = REG(ptr + PCIR_MSIX_PBA, 4);
916 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
917 			    PCIM_MSIX_BIR_MASK);
918 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
919 			break;
920 		case PCIY_VPD:		/* PCI Vital Product Data */
921 			cfg->vpd.vpd_reg = ptr;
922 			break;
923 		case PCIY_SUBVENDOR:
924 			/* Should always be true. */
925 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
926 			    PCIM_HDRTYPE_BRIDGE) {
927 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
928 				cfg->subvendor = val & 0xffff;
929 				cfg->subdevice = val >> 16;
930 			}
931 			break;
932 		case PCIY_PCIX:		/* PCI-X */
933 			/*
934 			 * Assume we have a PCI-X chipset if we have
935 			 * at least one PCI-PCI bridge with a PCI-X
936 			 * capability.  Note that some systems with
937 			 * PCI-express or HT chipsets might match on
938 			 * this check as well.
939 			 */
940 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
941 			    PCIM_HDRTYPE_BRIDGE)
942 				pcix_chipset = 1;
943 			cfg->pcix.pcix_location = ptr;
944 			break;
945 		case PCIY_EXPRESS:	/* PCI-express */
946 			/*
947 			 * Assume we have a PCI-express chipset if we have
948 			 * at least one PCI-express device.
949 			 */
950 			pcie_chipset = 1;
951 			cfg->pcie.pcie_location = ptr;
952 			val = REG(ptr + PCIER_FLAGS, 2);
953 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
954 			break;
955 		case PCIY_EA:		/* Enhanced Allocation */
956 			cfg->ea.ea_location = ptr;
957 			pci_ea_fill_info(pcib, cfg);
958 			break;
959 		default:
960 			break;
961 		}
962 	}
963 
964 #if defined(__powerpc__)
965 	/*
966 	 * Enable the MSI mapping window for all HyperTransport
967 	 * slaves.  PCI-PCI bridges have their windows enabled via
968 	 * PCIB_MAP_MSI().
969 	 */
970 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
971 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
972 		device_printf(pcib,
973 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
974 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
975 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
976 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
977 		     2);
978 	}
979 #endif
980 /* REG and WREG use carry through to next functions */
981 }
982 
983 /*
984  * PCI Vital Product Data
985  */
986 
987 #define	PCI_VPD_TIMEOUT		1000000
988 
989 static int
990 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
991 {
992 	int count = PCI_VPD_TIMEOUT;
993 
994 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
995 
996 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
997 
998 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
999 		if (--count < 0)
1000 			return (ENXIO);
1001 		DELAY(1);	/* limit looping */
1002 	}
1003 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1004 
1005 	return (0);
1006 }
1007 
1008 #if 0
1009 static int
1010 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1011 {
1012 	int count = PCI_VPD_TIMEOUT;
1013 
1014 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1015 
1016 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1017 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1018 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1019 		if (--count < 0)
1020 			return (ENXIO);
1021 		DELAY(1);	/* limit looping */
1022 	}
1023 
1024 	return (0);
1025 }
1026 #endif
1027 
1028 #undef PCI_VPD_TIMEOUT
1029 
1030 struct vpd_readstate {
1031 	device_t	pcib;
1032 	pcicfgregs	*cfg;
1033 	uint32_t	val;
1034 	int		bytesinval;
1035 	int		off;
1036 	uint8_t		cksum;
1037 };
1038 
1039 static int
1040 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1041 {
1042 	uint32_t reg;
1043 	uint8_t byte;
1044 
1045 	if (vrs->bytesinval == 0) {
1046 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1047 			return (ENXIO);
1048 		vrs->val = le32toh(reg);
1049 		vrs->off += 4;
1050 		byte = vrs->val & 0xff;
1051 		vrs->bytesinval = 3;
1052 	} else {
1053 		vrs->val = vrs->val >> 8;
1054 		byte = vrs->val & 0xff;
1055 		vrs->bytesinval--;
1056 	}
1057 
1058 	vrs->cksum += byte;
1059 	*data = byte;
1060 	return (0);
1061 }
1062 
1063 static void
1064 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1065 {
1066 	struct vpd_readstate vrs;
1067 	int state;
1068 	int name;
1069 	int remain;
1070 	int i;
1071 	int alloc, off;		/* alloc/off for RO/W arrays */
1072 	int cksumvalid;
1073 	int dflen;
1074 	uint8_t byte;
1075 	uint8_t byte2;
1076 
1077 	/* init vpd reader */
1078 	vrs.bytesinval = 0;
1079 	vrs.off = 0;
1080 	vrs.pcib = pcib;
1081 	vrs.cfg = cfg;
1082 	vrs.cksum = 0;
1083 
1084 	state = 0;
1085 	name = remain = i = 0;	/* shut up stupid gcc */
1086 	alloc = off = 0;	/* shut up stupid gcc */
1087 	dflen = 0;		/* shut up stupid gcc */
1088 	cksumvalid = -1;
1089 	while (state >= 0) {
1090 		if (vpd_nextbyte(&vrs, &byte)) {
1091 			state = -2;
1092 			break;
1093 		}
1094 #if 0
1095 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1096 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1097 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1098 #endif
1099 		switch (state) {
1100 		case 0:		/* item name */
1101 			if (byte & 0x80) {
1102 				if (vpd_nextbyte(&vrs, &byte2)) {
1103 					state = -2;
1104 					break;
1105 				}
1106 				remain = byte2;
1107 				if (vpd_nextbyte(&vrs, &byte2)) {
1108 					state = -2;
1109 					break;
1110 				}
1111 				remain |= byte2 << 8;
1112 				if (remain > (0x7f*4 - vrs.off)) {
1113 					state = -1;
1114 					pci_printf(cfg,
1115 					    "invalid VPD data, remain %#x\n",
1116 					    remain);
1117 				}
1118 				name = byte & 0x7f;
1119 			} else {
1120 				remain = byte & 0x7;
1121 				name = (byte >> 3) & 0xf;
1122 			}
1123 			switch (name) {
1124 			case 0x2:	/* String */
1125 				cfg->vpd.vpd_ident = malloc(remain + 1,
1126 				    M_DEVBUF, M_WAITOK);
1127 				i = 0;
1128 				state = 1;
1129 				break;
1130 			case 0xf:	/* End */
1131 				state = -1;
1132 				break;
1133 			case 0x10:	/* VPD-R */
1134 				alloc = 8;
1135 				off = 0;
1136 				cfg->vpd.vpd_ros = malloc(alloc *
1137 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1138 				    M_WAITOK | M_ZERO);
1139 				state = 2;
1140 				break;
1141 			case 0x11:	/* VPD-W */
1142 				alloc = 8;
1143 				off = 0;
1144 				cfg->vpd.vpd_w = malloc(alloc *
1145 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1146 				    M_WAITOK | M_ZERO);
1147 				state = 5;
1148 				break;
1149 			default:	/* Invalid data, abort */
1150 				state = -1;
1151 				break;
1152 			}
1153 			break;
1154 
1155 		case 1:	/* Identifier String */
1156 			cfg->vpd.vpd_ident[i++] = byte;
1157 			remain--;
1158 			if (remain == 0)  {
1159 				cfg->vpd.vpd_ident[i] = '\0';
1160 				state = 0;
1161 			}
1162 			break;
1163 
1164 		case 2:	/* VPD-R Keyword Header */
1165 			if (off == alloc) {
1166 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1167 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1168 				    M_DEVBUF, M_WAITOK | M_ZERO);
1169 			}
1170 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1171 			if (vpd_nextbyte(&vrs, &byte2)) {
1172 				state = -2;
1173 				break;
1174 			}
1175 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1176 			if (vpd_nextbyte(&vrs, &byte2)) {
1177 				state = -2;
1178 				break;
1179 			}
1180 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1181 			if (dflen == 0 &&
1182 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1183 			    2) == 0) {
1184 				/*
1185 				 * if this happens, we can't trust the rest
1186 				 * of the VPD.
1187 				 */
1188 				pci_printf(cfg, "bad keyword length: %d\n",
1189 				    dflen);
1190 				cksumvalid = 0;
1191 				state = -1;
1192 				break;
1193 			} else if (dflen == 0) {
1194 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1195 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1196 				    M_DEVBUF, M_WAITOK);
1197 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1198 			} else
1199 				cfg->vpd.vpd_ros[off].value = malloc(
1200 				    (dflen + 1) *
1201 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1202 				    M_DEVBUF, M_WAITOK);
1203 			remain -= 3;
1204 			i = 0;
1205 			/* keep in sync w/ state 3's transistions */
1206 			if (dflen == 0 && remain == 0)
1207 				state = 0;
1208 			else if (dflen == 0)
1209 				state = 2;
1210 			else
1211 				state = 3;
1212 			break;
1213 
1214 		case 3:	/* VPD-R Keyword Value */
1215 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1216 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1217 			    "RV", 2) == 0 && cksumvalid == -1) {
1218 				if (vrs.cksum == 0)
1219 					cksumvalid = 1;
1220 				else {
1221 					if (bootverbose)
1222 						pci_printf(cfg,
1223 					    "bad VPD cksum, remain %hhu\n",
1224 						    vrs.cksum);
1225 					cksumvalid = 0;
1226 					state = -1;
1227 					break;
1228 				}
1229 			}
1230 			dflen--;
1231 			remain--;
1232 			/* keep in sync w/ state 2's transistions */
1233 			if (dflen == 0)
1234 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1235 			if (dflen == 0 && remain == 0) {
1236 				cfg->vpd.vpd_rocnt = off;
1237 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1238 				    off * sizeof(*cfg->vpd.vpd_ros),
1239 				    M_DEVBUF, M_WAITOK | M_ZERO);
1240 				state = 0;
1241 			} else if (dflen == 0)
1242 				state = 2;
1243 			break;
1244 
1245 		case 4:
1246 			remain--;
1247 			if (remain == 0)
1248 				state = 0;
1249 			break;
1250 
1251 		case 5:	/* VPD-W Keyword Header */
1252 			if (off == alloc) {
1253 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1254 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1255 				    M_DEVBUF, M_WAITOK | M_ZERO);
1256 			}
1257 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1258 			if (vpd_nextbyte(&vrs, &byte2)) {
1259 				state = -2;
1260 				break;
1261 			}
1262 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1263 			if (vpd_nextbyte(&vrs, &byte2)) {
1264 				state = -2;
1265 				break;
1266 			}
1267 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1268 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1269 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1270 			    sizeof(*cfg->vpd.vpd_w[off].value),
1271 			    M_DEVBUF, M_WAITOK);
1272 			remain -= 3;
1273 			i = 0;
1274 			/* keep in sync w/ state 6's transistions */
1275 			if (dflen == 0 && remain == 0)
1276 				state = 0;
1277 			else if (dflen == 0)
1278 				state = 5;
1279 			else
1280 				state = 6;
1281 			break;
1282 
1283 		case 6:	/* VPD-W Keyword Value */
1284 			cfg->vpd.vpd_w[off].value[i++] = byte;
1285 			dflen--;
1286 			remain--;
1287 			/* keep in sync w/ state 5's transistions */
1288 			if (dflen == 0)
1289 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1290 			if (dflen == 0 && remain == 0) {
1291 				cfg->vpd.vpd_wcnt = off;
1292 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1293 				    off * sizeof(*cfg->vpd.vpd_w),
1294 				    M_DEVBUF, M_WAITOK | M_ZERO);
1295 				state = 0;
1296 			} else if (dflen == 0)
1297 				state = 5;
1298 			break;
1299 
1300 		default:
1301 			pci_printf(cfg, "invalid state: %d\n", state);
1302 			state = -1;
1303 			break;
1304 		}
1305 	}
1306 
1307 	if (cksumvalid == 0 || state < -1) {
1308 		/* read-only data bad, clean up */
1309 		if (cfg->vpd.vpd_ros != NULL) {
1310 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1311 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1312 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1313 			cfg->vpd.vpd_ros = NULL;
1314 		}
1315 	}
1316 	if (state < -1) {
1317 		/* I/O error, clean up */
1318 		pci_printf(cfg, "failed to read VPD data.\n");
1319 		if (cfg->vpd.vpd_ident != NULL) {
1320 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1321 			cfg->vpd.vpd_ident = NULL;
1322 		}
1323 		if (cfg->vpd.vpd_w != NULL) {
1324 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1325 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1326 			free(cfg->vpd.vpd_w, M_DEVBUF);
1327 			cfg->vpd.vpd_w = NULL;
1328 		}
1329 	}
1330 	cfg->vpd.vpd_cached = 1;
1331 #undef REG
1332 #undef WREG
1333 }
1334 
1335 int
1336 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1337 {
1338 	struct pci_devinfo *dinfo = device_get_ivars(child);
1339 	pcicfgregs *cfg = &dinfo->cfg;
1340 
1341 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1342 		pci_read_vpd(device_get_parent(dev), cfg);
1343 
1344 	*identptr = cfg->vpd.vpd_ident;
1345 
1346 	if (*identptr == NULL)
1347 		return (ENXIO);
1348 
1349 	return (0);
1350 }
1351 
1352 int
1353 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1354 	const char **vptr)
1355 {
1356 	struct pci_devinfo *dinfo = device_get_ivars(child);
1357 	pcicfgregs *cfg = &dinfo->cfg;
1358 	int i;
1359 
1360 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1361 		pci_read_vpd(device_get_parent(dev), cfg);
1362 
1363 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1364 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1365 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1366 			*vptr = cfg->vpd.vpd_ros[i].value;
1367 			return (0);
1368 		}
1369 
1370 	*vptr = NULL;
1371 	return (ENXIO);
1372 }
1373 
1374 struct pcicfg_vpd *
1375 pci_fetch_vpd_list(device_t dev)
1376 {
1377 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1378 	pcicfgregs *cfg = &dinfo->cfg;
1379 
1380 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1381 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1382 	return (&cfg->vpd);
1383 }
1384 
1385 /*
1386  * Find the requested HyperTransport capability and return the offset
1387  * in configuration space via the pointer provided.  The function
1388  * returns 0 on success and an error code otherwise.
1389  */
1390 int
1391 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1392 {
1393 	int ptr, error;
1394 	uint16_t val;
1395 
1396 	error = pci_find_cap(child, PCIY_HT, &ptr);
1397 	if (error)
1398 		return (error);
1399 
1400 	/*
1401 	 * Traverse the capabilities list checking each HT capability
1402 	 * to see if it matches the requested HT capability.
1403 	 */
1404 	for (;;) {
1405 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1406 		if (capability == PCIM_HTCAP_SLAVE ||
1407 		    capability == PCIM_HTCAP_HOST)
1408 			val &= 0xe000;
1409 		else
1410 			val &= PCIM_HTCMD_CAP_MASK;
1411 		if (val == capability) {
1412 			if (capreg != NULL)
1413 				*capreg = ptr;
1414 			return (0);
1415 		}
1416 
1417 		/* Skip to the next HT capability. */
1418 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1419 			break;
1420 	}
1421 
1422 	return (ENOENT);
1423 }
1424 
1425 /*
1426  * Find the next requested HyperTransport capability after start and return
1427  * the offset in configuration space via the pointer provided.  The function
1428  * returns 0 on success and an error code otherwise.
1429  */
1430 int
1431 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1432     int start, int *capreg)
1433 {
1434 	int ptr;
1435 	uint16_t val;
1436 
1437 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1438 	    ("start capability is not HyperTransport capability"));
1439 	ptr = start;
1440 
1441 	/*
1442 	 * Traverse the capabilities list checking each HT capability
1443 	 * to see if it matches the requested HT capability.
1444 	 */
1445 	for (;;) {
1446 		/* Skip to the next HT capability. */
1447 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1448 			break;
1449 
1450 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1451 		if (capability == PCIM_HTCAP_SLAVE ||
1452 		    capability == PCIM_HTCAP_HOST)
1453 			val &= 0xe000;
1454 		else
1455 			val &= PCIM_HTCMD_CAP_MASK;
1456 		if (val == capability) {
1457 			if (capreg != NULL)
1458 				*capreg = ptr;
1459 			return (0);
1460 		}
1461 	}
1462 
1463 	return (ENOENT);
1464 }
1465 
1466 /*
1467  * Find the requested capability and return the offset in
1468  * configuration space via the pointer provided.  The function returns
1469  * 0 on success and an error code otherwise.
1470  */
1471 int
1472 pci_find_cap_method(device_t dev, device_t child, int capability,
1473     int *capreg)
1474 {
1475 	struct pci_devinfo *dinfo = device_get_ivars(child);
1476 	pcicfgregs *cfg = &dinfo->cfg;
1477 	uint32_t status;
1478 	uint8_t ptr;
1479 
1480 	/*
1481 	 * Check the CAP_LIST bit of the PCI status register first.
1482 	 */
1483 	status = pci_read_config(child, PCIR_STATUS, 2);
1484 	if (!(status & PCIM_STATUS_CAPPRESENT))
1485 		return (ENXIO);
1486 
1487 	/*
1488 	 * Determine the start pointer of the capabilities list.
1489 	 */
1490 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1491 	case PCIM_HDRTYPE_NORMAL:
1492 	case PCIM_HDRTYPE_BRIDGE:
1493 		ptr = PCIR_CAP_PTR;
1494 		break;
1495 	case PCIM_HDRTYPE_CARDBUS:
1496 		ptr = PCIR_CAP_PTR_2;
1497 		break;
1498 	default:
1499 		/* XXX: panic? */
1500 		return (ENXIO);		/* no extended capabilities support */
1501 	}
1502 	ptr = pci_read_config(child, ptr, 1);
1503 
1504 	/*
1505 	 * Traverse the capabilities list.
1506 	 */
1507 	while (ptr != 0) {
1508 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1509 			if (capreg != NULL)
1510 				*capreg = ptr;
1511 			return (0);
1512 		}
1513 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1514 	}
1515 
1516 	return (ENOENT);
1517 }
1518 
1519 /*
1520  * Find the next requested capability after start and return the offset in
1521  * configuration space via the pointer provided.  The function returns
1522  * 0 on success and an error code otherwise.
1523  */
1524 int
1525 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1526     int start, int *capreg)
1527 {
1528 	uint8_t ptr;
1529 
1530 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1531 	    ("start capability is not expected capability"));
1532 
1533 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1534 	while (ptr != 0) {
1535 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1536 			if (capreg != NULL)
1537 				*capreg = ptr;
1538 			return (0);
1539 		}
1540 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1541 	}
1542 
1543 	return (ENOENT);
1544 }
1545 
1546 /*
1547  * Find the requested extended capability and return the offset in
1548  * configuration space via the pointer provided.  The function returns
1549  * 0 on success and an error code otherwise.
1550  */
1551 int
1552 pci_find_extcap_method(device_t dev, device_t child, int capability,
1553     int *capreg)
1554 {
1555 	struct pci_devinfo *dinfo = device_get_ivars(child);
1556 	pcicfgregs *cfg = &dinfo->cfg;
1557 	uint32_t ecap;
1558 	uint16_t ptr;
1559 
1560 	/* Only supported for PCI-express devices. */
1561 	if (cfg->pcie.pcie_location == 0)
1562 		return (ENXIO);
1563 
1564 	ptr = PCIR_EXTCAP;
1565 	ecap = pci_read_config(child, ptr, 4);
1566 	if (ecap == 0xffffffff || ecap == 0)
1567 		return (ENOENT);
1568 	for (;;) {
1569 		if (PCI_EXTCAP_ID(ecap) == capability) {
1570 			if (capreg != NULL)
1571 				*capreg = ptr;
1572 			return (0);
1573 		}
1574 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1575 		if (ptr == 0)
1576 			break;
1577 		ecap = pci_read_config(child, ptr, 4);
1578 	}
1579 
1580 	return (ENOENT);
1581 }
1582 
1583 /*
1584  * Find the next requested extended capability after start and return the
1585  * offset in configuration space via the pointer provided.  The function
1586  * returns 0 on success and an error code otherwise.
1587  */
1588 int
1589 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1590     int start, int *capreg)
1591 {
1592 	struct pci_devinfo *dinfo = device_get_ivars(child);
1593 	pcicfgregs *cfg = &dinfo->cfg;
1594 	uint32_t ecap;
1595 	uint16_t ptr;
1596 
1597 	/* Only supported for PCI-express devices. */
1598 	if (cfg->pcie.pcie_location == 0)
1599 		return (ENXIO);
1600 
1601 	ecap = pci_read_config(child, start, 4);
1602 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1603 	    ("start extended capability is not expected capability"));
1604 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1605 	while (ptr != 0) {
1606 		ecap = pci_read_config(child, ptr, 4);
1607 		if (PCI_EXTCAP_ID(ecap) == capability) {
1608 			if (capreg != NULL)
1609 				*capreg = ptr;
1610 			return (0);
1611 		}
1612 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1613 	}
1614 
1615 	return (ENOENT);
1616 }
1617 
1618 /*
1619  * Support for MSI-X message interrupts.
1620  */
1621 static void
1622 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1623 {
1624 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1625 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1626 	uint32_t offset;
1627 
1628 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1629 	offset = msix->msix_table_offset + index * 16;
1630 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1631 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1632 	bus_write_4(msix->msix_table_res, offset + 8, data);
1633 }
1634 
1635 void
1636 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1637     uint64_t address, uint32_t data)
1638 {
1639 
1640 	if (pci_msix_rewrite_table) {
1641 		struct pci_devinfo *dinfo = device_get_ivars(child);
1642 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1643 
1644 		/*
1645 		 * Some VM hosts require MSIX to be disabled in the
1646 		 * control register before updating the MSIX table
1647 		 * entries are allowed. It is not enough to only
1648 		 * disable MSIX while updating a single entry. MSIX
1649 		 * must be disabled while updating all entries in the
1650 		 * table.
1651 		 */
1652 		pci_write_config(child,
1653 		    msix->msix_location + PCIR_MSIX_CTRL,
1654 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1655 		pci_resume_msix(child);
1656 	} else
1657 		pci_write_msix_entry(child, index, address, data);
1658 
1659 	/* Enable MSI -> HT mapping. */
1660 	pci_ht_map_msi(child, address);
1661 }
1662 
1663 void
1664 pci_mask_msix(device_t dev, u_int index)
1665 {
1666 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1667 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1668 	uint32_t offset, val;
1669 
1670 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1671 	offset = msix->msix_table_offset + index * 16 + 12;
1672 	val = bus_read_4(msix->msix_table_res, offset);
1673 	if (!(val & PCIM_MSIX_VCTRL_MASK)) {
1674 		val |= PCIM_MSIX_VCTRL_MASK;
1675 		bus_write_4(msix->msix_table_res, offset, val);
1676 	}
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 	if (val & PCIM_MSIX_VCTRL_MASK) {
1690 		val &= ~PCIM_MSIX_VCTRL_MASK;
1691 		bus_write_4(msix->msix_table_res, offset, val);
1692 	}
1693 }
1694 
1695 int
1696 pci_pending_msix(device_t dev, u_int index)
1697 {
1698 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1699 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1700 	uint32_t offset, bit;
1701 
1702 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1703 	offset = msix->msix_pba_offset + (index / 32) * 4;
1704 	bit = 1 << index % 32;
1705 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1706 }
1707 
1708 /*
1709  * Restore MSI-X registers and table during resume.  If MSI-X is
1710  * enabled then walk the virtual table to restore the actual MSI-X
1711  * table.
1712  */
1713 static void
1714 pci_resume_msix(device_t dev)
1715 {
1716 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1717 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1718 	struct msix_table_entry *mte;
1719 	struct msix_vector *mv;
1720 	int i;
1721 
1722 	if (msix->msix_alloc > 0) {
1723 		/* First, mask all vectors. */
1724 		for (i = 0; i < msix->msix_msgnum; i++)
1725 			pci_mask_msix(dev, i);
1726 
1727 		/* Second, program any messages with at least one handler. */
1728 		for (i = 0; i < msix->msix_table_len; i++) {
1729 			mte = &msix->msix_table[i];
1730 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1731 				continue;
1732 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1733 			pci_write_msix_entry(dev, i, mv->mv_address,
1734 			    mv->mv_data);
1735 			pci_unmask_msix(dev, i);
1736 		}
1737 	}
1738 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1739 	    msix->msix_ctrl, 2);
1740 }
1741 
1742 /*
1743  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1744  * returned in *count.  After this function returns, each message will be
1745  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1746  */
1747 int
1748 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1749 {
1750 	struct pci_devinfo *dinfo = device_get_ivars(child);
1751 	pcicfgregs *cfg = &dinfo->cfg;
1752 	struct resource_list_entry *rle;
1753 	int actual, error, i, irq, max;
1754 
1755 	/* Don't let count == 0 get us into trouble. */
1756 	if (*count == 0)
1757 		return (EINVAL);
1758 
1759 	/* If rid 0 is allocated, then fail. */
1760 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1761 	if (rle != NULL && rle->res != NULL)
1762 		return (ENXIO);
1763 
1764 	/* Already have allocated messages? */
1765 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1766 		return (ENXIO);
1767 
1768 	/* If MSI-X is blacklisted for this system, fail. */
1769 	if (pci_msix_blacklisted())
1770 		return (ENXIO);
1771 
1772 	/* MSI-X capability present? */
1773 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1774 		return (ENODEV);
1775 
1776 	/* Make sure the appropriate BARs are mapped. */
1777 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1778 	    cfg->msix.msix_table_bar);
1779 	if (rle == NULL || rle->res == NULL ||
1780 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1781 		return (ENXIO);
1782 	cfg->msix.msix_table_res = rle->res;
1783 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1784 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1785 		    cfg->msix.msix_pba_bar);
1786 		if (rle == NULL || rle->res == NULL ||
1787 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1788 			return (ENXIO);
1789 	}
1790 	cfg->msix.msix_pba_res = rle->res;
1791 
1792 	if (bootverbose)
1793 		device_printf(child,
1794 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1795 		    *count, cfg->msix.msix_msgnum);
1796 	max = min(*count, cfg->msix.msix_msgnum);
1797 	for (i = 0; i < max; i++) {
1798 		/* Allocate a message. */
1799 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1800 		if (error) {
1801 			if (i == 0)
1802 				return (error);
1803 			break;
1804 		}
1805 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1806 		    irq, 1);
1807 	}
1808 	actual = i;
1809 
1810 	if (bootverbose) {
1811 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1812 		if (actual == 1)
1813 			device_printf(child, "using IRQ %ju for MSI-X\n",
1814 			    rle->start);
1815 		else {
1816 			int run;
1817 
1818 			/*
1819 			 * Be fancy and try to print contiguous runs of
1820 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1821 			 * 'run' is true if we are in a range.
1822 			 */
1823 			device_printf(child, "using IRQs %ju", rle->start);
1824 			irq = rle->start;
1825 			run = 0;
1826 			for (i = 1; i < actual; i++) {
1827 				rle = resource_list_find(&dinfo->resources,
1828 				    SYS_RES_IRQ, i + 1);
1829 
1830 				/* Still in a run? */
1831 				if (rle->start == irq + 1) {
1832 					run = 1;
1833 					irq++;
1834 					continue;
1835 				}
1836 
1837 				/* Finish previous range. */
1838 				if (run) {
1839 					printf("-%d", irq);
1840 					run = 0;
1841 				}
1842 
1843 				/* Start new range. */
1844 				printf(",%ju", rle->start);
1845 				irq = rle->start;
1846 			}
1847 
1848 			/* Unfinished range? */
1849 			if (run)
1850 				printf("-%d", irq);
1851 			printf(" for MSI-X\n");
1852 		}
1853 	}
1854 
1855 	/* Mask all vectors. */
1856 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1857 		pci_mask_msix(child, i);
1858 
1859 	/* Allocate and initialize vector data and virtual table. */
1860 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1861 	    M_DEVBUF, M_WAITOK | M_ZERO);
1862 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1863 	    M_DEVBUF, M_WAITOK | M_ZERO);
1864 	for (i = 0; i < actual; i++) {
1865 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1866 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1867 		cfg->msix.msix_table[i].mte_vector = i + 1;
1868 	}
1869 
1870 	/* Update control register to enable MSI-X. */
1871 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1872 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1873 	    cfg->msix.msix_ctrl, 2);
1874 
1875 	/* Update counts of alloc'd messages. */
1876 	cfg->msix.msix_alloc = actual;
1877 	cfg->msix.msix_table_len = actual;
1878 	*count = actual;
1879 	return (0);
1880 }
1881 
1882 /*
1883  * By default, pci_alloc_msix() will assign the allocated IRQ
1884  * resources consecutively to the first N messages in the MSI-X table.
1885  * However, device drivers may want to use different layouts if they
1886  * either receive fewer messages than they asked for, or they wish to
1887  * populate the MSI-X table sparsely.  This method allows the driver
1888  * to specify what layout it wants.  It must be called after a
1889  * successful pci_alloc_msix() but before any of the associated
1890  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1891  *
1892  * The 'vectors' array contains 'count' message vectors.  The array
1893  * maps directly to the MSI-X table in that index 0 in the array
1894  * specifies the vector for the first message in the MSI-X table, etc.
1895  * The vector value in each array index can either be 0 to indicate
1896  * that no vector should be assigned to a message slot, or it can be a
1897  * number from 1 to N (where N is the count returned from a
1898  * succcessful call to pci_alloc_msix()) to indicate which message
1899  * vector (IRQ) to be used for the corresponding message.
1900  *
1901  * On successful return, each message with a non-zero vector will have
1902  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1903  * 1.  Additionally, if any of the IRQs allocated via the previous
1904  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1905  * will be freed back to the system automatically.
1906  *
1907  * For example, suppose a driver has a MSI-X table with 6 messages and
1908  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1909  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1910  * C.  After the call to pci_alloc_msix(), the device will be setup to
1911  * have an MSI-X table of ABC--- (where - means no vector assigned).
1912  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1913  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1914  * be freed back to the system.  This device will also have valid
1915  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1916  *
1917  * In any case, the SYS_RES_IRQ rid X will always map to the message
1918  * at MSI-X table index X - 1 and will only be valid if a vector is
1919  * assigned to that table entry.
1920  */
1921 int
1922 pci_remap_msix_method(device_t dev, device_t child, int count,
1923     const u_int *vectors)
1924 {
1925 	struct pci_devinfo *dinfo = device_get_ivars(child);
1926 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1927 	struct resource_list_entry *rle;
1928 	int i, irq, j, *used;
1929 
1930 	/*
1931 	 * Have to have at least one message in the table but the
1932 	 * table can't be bigger than the actual MSI-X table in the
1933 	 * device.
1934 	 */
1935 	if (count == 0 || count > msix->msix_msgnum)
1936 		return (EINVAL);
1937 
1938 	/* Sanity check the vectors. */
1939 	for (i = 0; i < count; i++)
1940 		if (vectors[i] > msix->msix_alloc)
1941 			return (EINVAL);
1942 
1943 	/*
1944 	 * Make sure there aren't any holes in the vectors to be used.
1945 	 * It's a big pain to support it, and it doesn't really make
1946 	 * sense anyway.  Also, at least one vector must be used.
1947 	 */
1948 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1949 	    M_ZERO);
1950 	for (i = 0; i < count; i++)
1951 		if (vectors[i] != 0)
1952 			used[vectors[i] - 1] = 1;
1953 	for (i = 0; i < msix->msix_alloc - 1; i++)
1954 		if (used[i] == 0 && used[i + 1] == 1) {
1955 			free(used, M_DEVBUF);
1956 			return (EINVAL);
1957 		}
1958 	if (used[0] != 1) {
1959 		free(used, M_DEVBUF);
1960 		return (EINVAL);
1961 	}
1962 
1963 	/* Make sure none of the resources are allocated. */
1964 	for (i = 0; i < msix->msix_table_len; i++) {
1965 		if (msix->msix_table[i].mte_vector == 0)
1966 			continue;
1967 		if (msix->msix_table[i].mte_handlers > 0) {
1968 			free(used, M_DEVBUF);
1969 			return (EBUSY);
1970 		}
1971 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1972 		KASSERT(rle != NULL, ("missing resource"));
1973 		if (rle->res != NULL) {
1974 			free(used, M_DEVBUF);
1975 			return (EBUSY);
1976 		}
1977 	}
1978 
1979 	/* Free the existing resource list entries. */
1980 	for (i = 0; i < msix->msix_table_len; i++) {
1981 		if (msix->msix_table[i].mte_vector == 0)
1982 			continue;
1983 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1984 	}
1985 
1986 	/*
1987 	 * Build the new virtual table keeping track of which vectors are
1988 	 * used.
1989 	 */
1990 	free(msix->msix_table, M_DEVBUF);
1991 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1992 	    M_DEVBUF, M_WAITOK | M_ZERO);
1993 	for (i = 0; i < count; i++)
1994 		msix->msix_table[i].mte_vector = vectors[i];
1995 	msix->msix_table_len = count;
1996 
1997 	/* Free any unused IRQs and resize the vectors array if necessary. */
1998 	j = msix->msix_alloc - 1;
1999 	if (used[j] == 0) {
2000 		struct msix_vector *vec;
2001 
2002 		while (used[j] == 0) {
2003 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2004 			    msix->msix_vectors[j].mv_irq);
2005 			j--;
2006 		}
2007 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2008 		    M_WAITOK);
2009 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2010 		    (j + 1));
2011 		free(msix->msix_vectors, M_DEVBUF);
2012 		msix->msix_vectors = vec;
2013 		msix->msix_alloc = j + 1;
2014 	}
2015 	free(used, M_DEVBUF);
2016 
2017 	/* Map the IRQs onto the rids. */
2018 	for (i = 0; i < count; i++) {
2019 		if (vectors[i] == 0)
2020 			continue;
2021 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2022 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2023 		    irq, 1);
2024 	}
2025 
2026 	if (bootverbose) {
2027 		device_printf(child, "Remapped MSI-X IRQs as: ");
2028 		for (i = 0; i < count; i++) {
2029 			if (i != 0)
2030 				printf(", ");
2031 			if (vectors[i] == 0)
2032 				printf("---");
2033 			else
2034 				printf("%d",
2035 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2036 		}
2037 		printf("\n");
2038 	}
2039 
2040 	return (0);
2041 }
2042 
2043 static int
2044 pci_release_msix(device_t dev, device_t child)
2045 {
2046 	struct pci_devinfo *dinfo = device_get_ivars(child);
2047 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2048 	struct resource_list_entry *rle;
2049 	int i;
2050 
2051 	/* Do we have any messages to release? */
2052 	if (msix->msix_alloc == 0)
2053 		return (ENODEV);
2054 
2055 	/* Make sure none of the resources are allocated. */
2056 	for (i = 0; i < msix->msix_table_len; i++) {
2057 		if (msix->msix_table[i].mte_vector == 0)
2058 			continue;
2059 		if (msix->msix_table[i].mte_handlers > 0)
2060 			return (EBUSY);
2061 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2062 		KASSERT(rle != NULL, ("missing resource"));
2063 		if (rle->res != NULL)
2064 			return (EBUSY);
2065 	}
2066 
2067 	/* Update control register to disable MSI-X. */
2068 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2069 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2070 	    msix->msix_ctrl, 2);
2071 
2072 	/* Free the resource list entries. */
2073 	for (i = 0; i < msix->msix_table_len; i++) {
2074 		if (msix->msix_table[i].mte_vector == 0)
2075 			continue;
2076 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2077 	}
2078 	free(msix->msix_table, M_DEVBUF);
2079 	msix->msix_table_len = 0;
2080 
2081 	/* Release the IRQs. */
2082 	for (i = 0; i < msix->msix_alloc; i++)
2083 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2084 		    msix->msix_vectors[i].mv_irq);
2085 	free(msix->msix_vectors, M_DEVBUF);
2086 	msix->msix_alloc = 0;
2087 	return (0);
2088 }
2089 
2090 /*
2091  * Return the max supported MSI-X messages this device supports.
2092  * Basically, assuming the MD code can alloc messages, this function
2093  * should return the maximum value that pci_alloc_msix() can return.
2094  * Thus, it is subject to the tunables, etc.
2095  */
2096 int
2097 pci_msix_count_method(device_t dev, device_t child)
2098 {
2099 	struct pci_devinfo *dinfo = device_get_ivars(child);
2100 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2101 
2102 	if (pci_do_msix && msix->msix_location != 0)
2103 		return (msix->msix_msgnum);
2104 	return (0);
2105 }
2106 
2107 int
2108 pci_msix_pba_bar_method(device_t dev, device_t child)
2109 {
2110 	struct pci_devinfo *dinfo = device_get_ivars(child);
2111 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2112 
2113 	if (pci_do_msix && msix->msix_location != 0)
2114 		return (msix->msix_pba_bar);
2115 	return (-1);
2116 }
2117 
2118 int
2119 pci_msix_table_bar_method(device_t dev, device_t child)
2120 {
2121 	struct pci_devinfo *dinfo = device_get_ivars(child);
2122 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2123 
2124 	if (pci_do_msix && msix->msix_location != 0)
2125 		return (msix->msix_table_bar);
2126 	return (-1);
2127 }
2128 
2129 /*
2130  * HyperTransport MSI mapping control
2131  */
2132 void
2133 pci_ht_map_msi(device_t dev, uint64_t addr)
2134 {
2135 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2136 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2137 
2138 	if (!ht->ht_msimap)
2139 		return;
2140 
2141 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2142 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2143 		/* Enable MSI -> HT mapping. */
2144 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2145 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2146 		    ht->ht_msictrl, 2);
2147 	}
2148 
2149 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2150 		/* Disable MSI -> HT mapping. */
2151 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2152 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2153 		    ht->ht_msictrl, 2);
2154 	}
2155 }
2156 
2157 int
2158 pci_get_max_payload(device_t dev)
2159 {
2160 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2161 	int cap;
2162 	uint16_t val;
2163 
2164 	cap = dinfo->cfg.pcie.pcie_location;
2165 	if (cap == 0)
2166 		return (0);
2167 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2168 	val &= PCIEM_CTL_MAX_PAYLOAD;
2169 	val >>= 5;
2170 	return (1 << (val + 7));
2171 }
2172 
2173 int
2174 pci_get_max_read_req(device_t dev)
2175 {
2176 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2177 	int cap;
2178 	uint16_t val;
2179 
2180 	cap = dinfo->cfg.pcie.pcie_location;
2181 	if (cap == 0)
2182 		return (0);
2183 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2184 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2185 	val >>= 12;
2186 	return (1 << (val + 7));
2187 }
2188 
2189 int
2190 pci_set_max_read_req(device_t dev, int size)
2191 {
2192 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2193 	int cap;
2194 	uint16_t val;
2195 
2196 	cap = dinfo->cfg.pcie.pcie_location;
2197 	if (cap == 0)
2198 		return (0);
2199 	if (size < 128)
2200 		size = 128;
2201 	if (size > 4096)
2202 		size = 4096;
2203 	size = (1 << (fls(size) - 1));
2204 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2205 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2206 	val |= (fls(size) - 8) << 12;
2207 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2208 	return (size);
2209 }
2210 
2211 uint32_t
2212 pcie_read_config(device_t dev, int reg, int width)
2213 {
2214 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2215 	int cap;
2216 
2217 	cap = dinfo->cfg.pcie.pcie_location;
2218 	if (cap == 0) {
2219 		if (width == 2)
2220 			return (0xffff);
2221 		return (0xffffffff);
2222 	}
2223 
2224 	return (pci_read_config(dev, cap + reg, width));
2225 }
2226 
2227 void
2228 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2229 {
2230 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2231 	int cap;
2232 
2233 	cap = dinfo->cfg.pcie.pcie_location;
2234 	if (cap == 0)
2235 		return;
2236 	pci_write_config(dev, cap + reg, value, width);
2237 }
2238 
2239 /*
2240  * Adjusts a PCI-e capability register by clearing the bits in mask
2241  * and setting the bits in (value & mask).  Bits not set in mask are
2242  * not adjusted.
2243  *
2244  * Returns the old value on success or all ones on failure.
2245  */
2246 uint32_t
2247 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2248     int width)
2249 {
2250 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2251 	uint32_t old, new;
2252 	int cap;
2253 
2254 	cap = dinfo->cfg.pcie.pcie_location;
2255 	if (cap == 0) {
2256 		if (width == 2)
2257 			return (0xffff);
2258 		return (0xffffffff);
2259 	}
2260 
2261 	old = pci_read_config(dev, cap + reg, width);
2262 	new = old & ~mask;
2263 	new |= (value & mask);
2264 	pci_write_config(dev, cap + reg, new, width);
2265 	return (old);
2266 }
2267 
2268 /*
2269  * Support for MSI message signalled interrupts.
2270  */
2271 void
2272 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2273     uint16_t data)
2274 {
2275 	struct pci_devinfo *dinfo = device_get_ivars(child);
2276 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2277 
2278 	/* Write data and address values. */
2279 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2280 	    address & 0xffffffff, 4);
2281 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2282 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2283 		    address >> 32, 4);
2284 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2285 		    data, 2);
2286 	} else
2287 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2288 		    2);
2289 
2290 	/* Enable MSI in the control register. */
2291 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2292 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2293 	    msi->msi_ctrl, 2);
2294 
2295 	/* Enable MSI -> HT mapping. */
2296 	pci_ht_map_msi(child, address);
2297 }
2298 
2299 void
2300 pci_disable_msi_method(device_t dev, device_t child)
2301 {
2302 	struct pci_devinfo *dinfo = device_get_ivars(child);
2303 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2304 
2305 	/* Disable MSI -> HT mapping. */
2306 	pci_ht_map_msi(child, 0);
2307 
2308 	/* Disable MSI in the control register. */
2309 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2310 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2311 	    msi->msi_ctrl, 2);
2312 }
2313 
2314 /*
2315  * Restore MSI registers during resume.  If MSI is enabled then
2316  * restore the data and address registers in addition to the control
2317  * register.
2318  */
2319 static void
2320 pci_resume_msi(device_t dev)
2321 {
2322 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2323 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2324 	uint64_t address;
2325 	uint16_t data;
2326 
2327 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2328 		address = msi->msi_addr;
2329 		data = msi->msi_data;
2330 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2331 		    address & 0xffffffff, 4);
2332 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2333 			pci_write_config(dev, msi->msi_location +
2334 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2335 			pci_write_config(dev, msi->msi_location +
2336 			    PCIR_MSI_DATA_64BIT, data, 2);
2337 		} else
2338 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2339 			    data, 2);
2340 	}
2341 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2342 	    2);
2343 }
2344 
2345 static int
2346 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2347 {
2348 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2349 	pcicfgregs *cfg = &dinfo->cfg;
2350 	struct resource_list_entry *rle;
2351 	struct msix_table_entry *mte;
2352 	struct msix_vector *mv;
2353 	uint64_t addr;
2354 	uint32_t data;
2355 	int error, i, j;
2356 
2357 	/*
2358 	 * Handle MSI first.  We try to find this IRQ among our list
2359 	 * of MSI IRQs.  If we find it, we request updated address and
2360 	 * data registers and apply the results.
2361 	 */
2362 	if (cfg->msi.msi_alloc > 0) {
2363 
2364 		/* If we don't have any active handlers, nothing to do. */
2365 		if (cfg->msi.msi_handlers == 0)
2366 			return (0);
2367 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2368 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2369 			    i + 1);
2370 			if (rle->start == irq) {
2371 				error = PCIB_MAP_MSI(device_get_parent(bus),
2372 				    dev, irq, &addr, &data);
2373 				if (error)
2374 					return (error);
2375 				pci_disable_msi(dev);
2376 				dinfo->cfg.msi.msi_addr = addr;
2377 				dinfo->cfg.msi.msi_data = data;
2378 				pci_enable_msi(dev, addr, data);
2379 				return (0);
2380 			}
2381 		}
2382 		return (ENOENT);
2383 	}
2384 
2385 	/*
2386 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2387 	 * we request the updated mapping info.  If that works, we go
2388 	 * through all the slots that use this IRQ and update them.
2389 	 */
2390 	if (cfg->msix.msix_alloc > 0) {
2391 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2392 			mv = &cfg->msix.msix_vectors[i];
2393 			if (mv->mv_irq == irq) {
2394 				error = PCIB_MAP_MSI(device_get_parent(bus),
2395 				    dev, irq, &addr, &data);
2396 				if (error)
2397 					return (error);
2398 				mv->mv_address = addr;
2399 				mv->mv_data = data;
2400 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2401 					mte = &cfg->msix.msix_table[j];
2402 					if (mte->mte_vector != i + 1)
2403 						continue;
2404 					if (mte->mte_handlers == 0)
2405 						continue;
2406 					pci_mask_msix(dev, j);
2407 					pci_enable_msix(dev, j, addr, data);
2408 					pci_unmask_msix(dev, j);
2409 				}
2410 			}
2411 		}
2412 		return (ENOENT);
2413 	}
2414 
2415 	return (ENOENT);
2416 }
2417 
2418 /*
2419  * Returns true if the specified device is blacklisted because MSI
2420  * doesn't work.
2421  */
2422 int
2423 pci_msi_device_blacklisted(device_t dev)
2424 {
2425 
2426 	if (!pci_honor_msi_blacklist)
2427 		return (0);
2428 
2429 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2430 }
2431 
2432 /*
2433  * Determine if MSI is blacklisted globally on this system.  Currently,
2434  * we just check for blacklisted chipsets as represented by the
2435  * host-PCI bridge at device 0:0:0.  In the future, it may become
2436  * necessary to check other system attributes, such as the kenv values
2437  * that give the motherboard manufacturer and model number.
2438  */
2439 static int
2440 pci_msi_blacklisted(void)
2441 {
2442 	device_t dev;
2443 
2444 	if (!pci_honor_msi_blacklist)
2445 		return (0);
2446 
2447 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2448 	if (!(pcie_chipset || pcix_chipset)) {
2449 		if (vm_guest != VM_GUEST_NO) {
2450 			/*
2451 			 * Whitelist older chipsets in virtual
2452 			 * machines known to support MSI.
2453 			 */
2454 			dev = pci_find_bsf(0, 0, 0);
2455 			if (dev != NULL)
2456 				return (!pci_has_quirk(pci_get_devid(dev),
2457 					PCI_QUIRK_ENABLE_MSI_VM));
2458 		}
2459 		return (1);
2460 	}
2461 
2462 	dev = pci_find_bsf(0, 0, 0);
2463 	if (dev != NULL)
2464 		return (pci_msi_device_blacklisted(dev));
2465 	return (0);
2466 }
2467 
2468 /*
2469  * Returns true if the specified device is blacklisted because MSI-X
2470  * doesn't work.  Note that this assumes that if MSI doesn't work,
2471  * MSI-X doesn't either.
2472  */
2473 int
2474 pci_msix_device_blacklisted(device_t dev)
2475 {
2476 
2477 	if (!pci_honor_msi_blacklist)
2478 		return (0);
2479 
2480 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2481 		return (1);
2482 
2483 	return (pci_msi_device_blacklisted(dev));
2484 }
2485 
2486 /*
2487  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2488  * is blacklisted, assume that MSI-X is as well.  Check for additional
2489  * chipsets where MSI works but MSI-X does not.
2490  */
2491 static int
2492 pci_msix_blacklisted(void)
2493 {
2494 	device_t dev;
2495 
2496 	if (!pci_honor_msi_blacklist)
2497 		return (0);
2498 
2499 	dev = pci_find_bsf(0, 0, 0);
2500 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2501 	    PCI_QUIRK_DISABLE_MSIX))
2502 		return (1);
2503 
2504 	return (pci_msi_blacklisted());
2505 }
2506 
2507 /*
2508  * Attempt to allocate *count MSI messages.  The actual number allocated is
2509  * returned in *count.  After this function returns, each message will be
2510  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2511  */
2512 int
2513 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2514 {
2515 	struct pci_devinfo *dinfo = device_get_ivars(child);
2516 	pcicfgregs *cfg = &dinfo->cfg;
2517 	struct resource_list_entry *rle;
2518 	int actual, error, i, irqs[32];
2519 	uint16_t ctrl;
2520 
2521 	/* Don't let count == 0 get us into trouble. */
2522 	if (*count == 0)
2523 		return (EINVAL);
2524 
2525 	/* If rid 0 is allocated, then fail. */
2526 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2527 	if (rle != NULL && rle->res != NULL)
2528 		return (ENXIO);
2529 
2530 	/* Already have allocated messages? */
2531 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2532 		return (ENXIO);
2533 
2534 	/* If MSI is blacklisted for this system, fail. */
2535 	if (pci_msi_blacklisted())
2536 		return (ENXIO);
2537 
2538 	/* MSI capability present? */
2539 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2540 		return (ENODEV);
2541 
2542 	if (bootverbose)
2543 		device_printf(child,
2544 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2545 		    *count, cfg->msi.msi_msgnum);
2546 
2547 	/* Don't ask for more than the device supports. */
2548 	actual = min(*count, cfg->msi.msi_msgnum);
2549 
2550 	/* Don't ask for more than 32 messages. */
2551 	actual = min(actual, 32);
2552 
2553 	/* MSI requires power of 2 number of messages. */
2554 	if (!powerof2(actual))
2555 		return (EINVAL);
2556 
2557 	for (;;) {
2558 		/* Try to allocate N messages. */
2559 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2560 		    actual, irqs);
2561 		if (error == 0)
2562 			break;
2563 		if (actual == 1)
2564 			return (error);
2565 
2566 		/* Try N / 2. */
2567 		actual >>= 1;
2568 	}
2569 
2570 	/*
2571 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2572 	 * resources in the irqs[] array, so add new resources
2573 	 * starting at rid 1.
2574 	 */
2575 	for (i = 0; i < actual; i++)
2576 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2577 		    irqs[i], irqs[i], 1);
2578 
2579 	if (bootverbose) {
2580 		if (actual == 1)
2581 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2582 		else {
2583 			int run;
2584 
2585 			/*
2586 			 * Be fancy and try to print contiguous runs
2587 			 * of IRQ values as ranges.  'run' is true if
2588 			 * we are in a range.
2589 			 */
2590 			device_printf(child, "using IRQs %d", irqs[0]);
2591 			run = 0;
2592 			for (i = 1; i < actual; i++) {
2593 
2594 				/* Still in a run? */
2595 				if (irqs[i] == irqs[i - 1] + 1) {
2596 					run = 1;
2597 					continue;
2598 				}
2599 
2600 				/* Finish previous range. */
2601 				if (run) {
2602 					printf("-%d", irqs[i - 1]);
2603 					run = 0;
2604 				}
2605 
2606 				/* Start new range. */
2607 				printf(",%d", irqs[i]);
2608 			}
2609 
2610 			/* Unfinished range? */
2611 			if (run)
2612 				printf("-%d", irqs[actual - 1]);
2613 			printf(" for MSI\n");
2614 		}
2615 	}
2616 
2617 	/* Update control register with actual count. */
2618 	ctrl = cfg->msi.msi_ctrl;
2619 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2620 	ctrl |= (ffs(actual) - 1) << 4;
2621 	cfg->msi.msi_ctrl = ctrl;
2622 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2623 
2624 	/* Update counts of alloc'd messages. */
2625 	cfg->msi.msi_alloc = actual;
2626 	cfg->msi.msi_handlers = 0;
2627 	*count = actual;
2628 	return (0);
2629 }
2630 
2631 /* Release the MSI messages associated with this device. */
2632 int
2633 pci_release_msi_method(device_t dev, device_t child)
2634 {
2635 	struct pci_devinfo *dinfo = device_get_ivars(child);
2636 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2637 	struct resource_list_entry *rle;
2638 	int error, i, irqs[32];
2639 
2640 	/* Try MSI-X first. */
2641 	error = pci_release_msix(dev, child);
2642 	if (error != ENODEV)
2643 		return (error);
2644 
2645 	/* Do we have any messages to release? */
2646 	if (msi->msi_alloc == 0)
2647 		return (ENODEV);
2648 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2649 
2650 	/* Make sure none of the resources are allocated. */
2651 	if (msi->msi_handlers > 0)
2652 		return (EBUSY);
2653 	for (i = 0; i < msi->msi_alloc; i++) {
2654 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2655 		KASSERT(rle != NULL, ("missing MSI resource"));
2656 		if (rle->res != NULL)
2657 			return (EBUSY);
2658 		irqs[i] = rle->start;
2659 	}
2660 
2661 	/* Update control register with 0 count. */
2662 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2663 	    ("%s: MSI still enabled", __func__));
2664 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2665 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2666 	    msi->msi_ctrl, 2);
2667 
2668 	/* Release the messages. */
2669 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2670 	for (i = 0; i < msi->msi_alloc; i++)
2671 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2672 
2673 	/* Update alloc count. */
2674 	msi->msi_alloc = 0;
2675 	msi->msi_addr = 0;
2676 	msi->msi_data = 0;
2677 	return (0);
2678 }
2679 
2680 /*
2681  * Return the max supported MSI messages this device supports.
2682  * Basically, assuming the MD code can alloc messages, this function
2683  * should return the maximum value that pci_alloc_msi() can return.
2684  * Thus, it is subject to the tunables, etc.
2685  */
2686 int
2687 pci_msi_count_method(device_t dev, device_t child)
2688 {
2689 	struct pci_devinfo *dinfo = device_get_ivars(child);
2690 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2691 
2692 	if (pci_do_msi && msi->msi_location != 0)
2693 		return (msi->msi_msgnum);
2694 	return (0);
2695 }
2696 
2697 /* free pcicfgregs structure and all depending data structures */
2698 
2699 int
2700 pci_freecfg(struct pci_devinfo *dinfo)
2701 {
2702 	struct devlist *devlist_head;
2703 	struct pci_map *pm, *next;
2704 	int i;
2705 
2706 	devlist_head = &pci_devq;
2707 
2708 	if (dinfo->cfg.vpd.vpd_reg) {
2709 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2710 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2711 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2712 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2713 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2714 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2715 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2716 	}
2717 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2718 		free(pm, M_DEVBUF);
2719 	}
2720 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2721 	free(dinfo, M_DEVBUF);
2722 
2723 	/* increment the generation count */
2724 	pci_generation++;
2725 
2726 	/* we're losing one device */
2727 	pci_numdevs--;
2728 	return (0);
2729 }
2730 
2731 /*
2732  * PCI power manangement
2733  */
2734 int
2735 pci_set_powerstate_method(device_t dev, device_t child, int state)
2736 {
2737 	struct pci_devinfo *dinfo = device_get_ivars(child);
2738 	pcicfgregs *cfg = &dinfo->cfg;
2739 	uint16_t status;
2740 	int oldstate, highest, delay;
2741 
2742 	if (cfg->pp.pp_cap == 0)
2743 		return (EOPNOTSUPP);
2744 
2745 	/*
2746 	 * Optimize a no state change request away.  While it would be OK to
2747 	 * write to the hardware in theory, some devices have shown odd
2748 	 * behavior when going from D3 -> D3.
2749 	 */
2750 	oldstate = pci_get_powerstate(child);
2751 	if (oldstate == state)
2752 		return (0);
2753 
2754 	/*
2755 	 * The PCI power management specification states that after a state
2756 	 * transition between PCI power states, system software must
2757 	 * guarantee a minimal delay before the function accesses the device.
2758 	 * Compute the worst case delay that we need to guarantee before we
2759 	 * access the device.  Many devices will be responsive much more
2760 	 * quickly than this delay, but there are some that don't respond
2761 	 * instantly to state changes.  Transitions to/from D3 state require
2762 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2763 	 * is done below with DELAY rather than a sleeper function because
2764 	 * this function can be called from contexts where we cannot sleep.
2765 	 */
2766 	highest = (oldstate > state) ? oldstate : state;
2767 	if (highest == PCI_POWERSTATE_D3)
2768 	    delay = 10000;
2769 	else if (highest == PCI_POWERSTATE_D2)
2770 	    delay = 200;
2771 	else
2772 	    delay = 0;
2773 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2774 	    & ~PCIM_PSTAT_DMASK;
2775 	switch (state) {
2776 	case PCI_POWERSTATE_D0:
2777 		status |= PCIM_PSTAT_D0;
2778 		break;
2779 	case PCI_POWERSTATE_D1:
2780 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2781 			return (EOPNOTSUPP);
2782 		status |= PCIM_PSTAT_D1;
2783 		break;
2784 	case PCI_POWERSTATE_D2:
2785 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2786 			return (EOPNOTSUPP);
2787 		status |= PCIM_PSTAT_D2;
2788 		break;
2789 	case PCI_POWERSTATE_D3:
2790 		status |= PCIM_PSTAT_D3;
2791 		break;
2792 	default:
2793 		return (EINVAL);
2794 	}
2795 
2796 	if (bootverbose)
2797 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2798 		    state);
2799 
2800 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2801 	if (delay)
2802 		DELAY(delay);
2803 	return (0);
2804 }
2805 
2806 int
2807 pci_get_powerstate_method(device_t dev, device_t child)
2808 {
2809 	struct pci_devinfo *dinfo = device_get_ivars(child);
2810 	pcicfgregs *cfg = &dinfo->cfg;
2811 	uint16_t status;
2812 	int result;
2813 
2814 	if (cfg->pp.pp_cap != 0) {
2815 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2816 		switch (status & PCIM_PSTAT_DMASK) {
2817 		case PCIM_PSTAT_D0:
2818 			result = PCI_POWERSTATE_D0;
2819 			break;
2820 		case PCIM_PSTAT_D1:
2821 			result = PCI_POWERSTATE_D1;
2822 			break;
2823 		case PCIM_PSTAT_D2:
2824 			result = PCI_POWERSTATE_D2;
2825 			break;
2826 		case PCIM_PSTAT_D3:
2827 			result = PCI_POWERSTATE_D3;
2828 			break;
2829 		default:
2830 			result = PCI_POWERSTATE_UNKNOWN;
2831 			break;
2832 		}
2833 	} else {
2834 		/* No support, device is always at D0 */
2835 		result = PCI_POWERSTATE_D0;
2836 	}
2837 	return (result);
2838 }
2839 
2840 /*
2841  * Some convenience functions for PCI device drivers.
2842  */
2843 
2844 static __inline void
2845 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2846 {
2847 	uint16_t	command;
2848 
2849 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2850 	command |= bit;
2851 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2852 }
2853 
2854 static __inline void
2855 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2856 {
2857 	uint16_t	command;
2858 
2859 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2860 	command &= ~bit;
2861 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2862 }
2863 
2864 int
2865 pci_enable_busmaster_method(device_t dev, device_t child)
2866 {
2867 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2868 	return (0);
2869 }
2870 
2871 int
2872 pci_disable_busmaster_method(device_t dev, device_t child)
2873 {
2874 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2875 	return (0);
2876 }
2877 
2878 int
2879 pci_enable_io_method(device_t dev, device_t child, int space)
2880 {
2881 	uint16_t bit;
2882 
2883 	switch(space) {
2884 	case SYS_RES_IOPORT:
2885 		bit = PCIM_CMD_PORTEN;
2886 		break;
2887 	case SYS_RES_MEMORY:
2888 		bit = PCIM_CMD_MEMEN;
2889 		break;
2890 	default:
2891 		return (EINVAL);
2892 	}
2893 	pci_set_command_bit(dev, child, bit);
2894 	return (0);
2895 }
2896 
2897 int
2898 pci_disable_io_method(device_t dev, device_t child, int space)
2899 {
2900 	uint16_t bit;
2901 
2902 	switch(space) {
2903 	case SYS_RES_IOPORT:
2904 		bit = PCIM_CMD_PORTEN;
2905 		break;
2906 	case SYS_RES_MEMORY:
2907 		bit = PCIM_CMD_MEMEN;
2908 		break;
2909 	default:
2910 		return (EINVAL);
2911 	}
2912 	pci_clear_command_bit(dev, child, bit);
2913 	return (0);
2914 }
2915 
2916 /*
2917  * New style pci driver.  Parent device is either a pci-host-bridge or a
2918  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2919  */
2920 
2921 void
2922 pci_print_verbose(struct pci_devinfo *dinfo)
2923 {
2924 
2925 	if (bootverbose) {
2926 		pcicfgregs *cfg = &dinfo->cfg;
2927 
2928 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2929 		    cfg->vendor, cfg->device, cfg->revid);
2930 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2931 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2932 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2933 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2934 		    cfg->mfdev);
2935 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2936 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2937 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2938 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2939 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2940 		if (cfg->intpin > 0)
2941 			printf("\tintpin=%c, irq=%d\n",
2942 			    cfg->intpin +'a' -1, cfg->intline);
2943 		if (cfg->pp.pp_cap) {
2944 			uint16_t status;
2945 
2946 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2947 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2948 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2949 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2950 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2951 			    status & PCIM_PSTAT_DMASK);
2952 		}
2953 		if (cfg->msi.msi_location) {
2954 			int ctrl;
2955 
2956 			ctrl = cfg->msi.msi_ctrl;
2957 			printf("\tMSI supports %d message%s%s%s\n",
2958 			    cfg->msi.msi_msgnum,
2959 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2960 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2961 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2962 		}
2963 		if (cfg->msix.msix_location) {
2964 			printf("\tMSI-X supports %d message%s ",
2965 			    cfg->msix.msix_msgnum,
2966 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2967 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2968 				printf("in map 0x%x\n",
2969 				    cfg->msix.msix_table_bar);
2970 			else
2971 				printf("in maps 0x%x and 0x%x\n",
2972 				    cfg->msix.msix_table_bar,
2973 				    cfg->msix.msix_pba_bar);
2974 		}
2975 	}
2976 }
2977 
2978 static int
2979 pci_porten(device_t dev)
2980 {
2981 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2982 }
2983 
2984 static int
2985 pci_memen(device_t dev)
2986 {
2987 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2988 }
2989 
2990 void
2991 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2992     int *bar64)
2993 {
2994 	struct pci_devinfo *dinfo;
2995 	pci_addr_t map, testval;
2996 	int ln2range;
2997 	uint16_t cmd;
2998 
2999 	/*
3000 	 * The device ROM BAR is special.  It is always a 32-bit
3001 	 * memory BAR.  Bit 0 is special and should not be set when
3002 	 * sizing the BAR.
3003 	 */
3004 	dinfo = device_get_ivars(dev);
3005 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3006 		map = pci_read_config(dev, reg, 4);
3007 		pci_write_config(dev, reg, 0xfffffffe, 4);
3008 		testval = pci_read_config(dev, reg, 4);
3009 		pci_write_config(dev, reg, map, 4);
3010 		*mapp = map;
3011 		*testvalp = testval;
3012 		if (bar64 != NULL)
3013 			*bar64 = 0;
3014 		return;
3015 	}
3016 
3017 	map = pci_read_config(dev, reg, 4);
3018 	ln2range = pci_maprange(map);
3019 	if (ln2range == 64)
3020 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3021 
3022 	/*
3023 	 * Disable decoding via the command register before
3024 	 * determining the BAR's length since we will be placing it in
3025 	 * a weird state.
3026 	 */
3027 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3028 	pci_write_config(dev, PCIR_COMMAND,
3029 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3030 
3031 	/*
3032 	 * Determine the BAR's length by writing all 1's.  The bottom
3033 	 * log_2(size) bits of the BAR will stick as 0 when we read
3034 	 * the value back.
3035 	 *
3036 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3037 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3038 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3039 	 *
3040 	 * Writes to both registers must be performed before attempting to
3041 	 * read back the size value.
3042 	 */
3043 	testval = 0;
3044 	pci_write_config(dev, reg, 0xffffffff, 4);
3045 	if (ln2range == 64) {
3046 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3047 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3048 	}
3049 	testval |= pci_read_config(dev, reg, 4);
3050 
3051 	/*
3052 	 * Restore the original value of the BAR.  We may have reprogrammed
3053 	 * the BAR of the low-level console device and when booting verbose,
3054 	 * we need the console device addressable.
3055 	 */
3056 	pci_write_config(dev, reg, map, 4);
3057 	if (ln2range == 64)
3058 		pci_write_config(dev, reg + 4, map >> 32, 4);
3059 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3060 
3061 	*mapp = map;
3062 	*testvalp = testval;
3063 	if (bar64 != NULL)
3064 		*bar64 = (ln2range == 64);
3065 }
3066 
3067 static void
3068 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3069 {
3070 	struct pci_devinfo *dinfo;
3071 	int ln2range;
3072 
3073 	/* The device ROM BAR is always a 32-bit memory BAR. */
3074 	dinfo = device_get_ivars(dev);
3075 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3076 		ln2range = 32;
3077 	else
3078 		ln2range = pci_maprange(pm->pm_value);
3079 	pci_write_config(dev, pm->pm_reg, base, 4);
3080 	if (ln2range == 64)
3081 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3082 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3083 	if (ln2range == 64)
3084 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3085 		    pm->pm_reg + 4, 4) << 32;
3086 }
3087 
3088 struct pci_map *
3089 pci_find_bar(device_t dev, int reg)
3090 {
3091 	struct pci_devinfo *dinfo;
3092 	struct pci_map *pm;
3093 
3094 	dinfo = device_get_ivars(dev);
3095 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3096 		if (pm->pm_reg == reg)
3097 			return (pm);
3098 	}
3099 	return (NULL);
3100 }
3101 
3102 int
3103 pci_bar_enabled(device_t dev, struct pci_map *pm)
3104 {
3105 	struct pci_devinfo *dinfo;
3106 	uint16_t cmd;
3107 
3108 	dinfo = device_get_ivars(dev);
3109 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3110 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3111 		return (0);
3112 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3113 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3114 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3115 	else
3116 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3117 }
3118 
3119 struct pci_map *
3120 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3121 {
3122 	struct pci_devinfo *dinfo;
3123 	struct pci_map *pm, *prev;
3124 
3125 	dinfo = device_get_ivars(dev);
3126 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3127 	pm->pm_reg = reg;
3128 	pm->pm_value = value;
3129 	pm->pm_size = size;
3130 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3131 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3132 		    reg));
3133 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3134 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3135 			break;
3136 	}
3137 	if (prev != NULL)
3138 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3139 	else
3140 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3141 	return (pm);
3142 }
3143 
3144 static void
3145 pci_restore_bars(device_t dev)
3146 {
3147 	struct pci_devinfo *dinfo;
3148 	struct pci_map *pm;
3149 	int ln2range;
3150 
3151 	dinfo = device_get_ivars(dev);
3152 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3153 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3154 			ln2range = 32;
3155 		else
3156 			ln2range = pci_maprange(pm->pm_value);
3157 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3158 		if (ln2range == 64)
3159 			pci_write_config(dev, pm->pm_reg + 4,
3160 			    pm->pm_value >> 32, 4);
3161 	}
3162 }
3163 
3164 /*
3165  * Add a resource based on a pci map register. Return 1 if the map
3166  * register is a 32bit map register or 2 if it is a 64bit register.
3167  */
3168 static int
3169 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3170     int force, int prefetch)
3171 {
3172 	struct pci_map *pm;
3173 	pci_addr_t base, map, testval;
3174 	pci_addr_t start, end, count;
3175 	int barlen, basezero, flags, maprange, mapsize, type;
3176 	uint16_t cmd;
3177 	struct resource *res;
3178 
3179 	/*
3180 	 * The BAR may already exist if the device is a CardBus card
3181 	 * whose CIS is stored in this BAR.
3182 	 */
3183 	pm = pci_find_bar(dev, reg);
3184 	if (pm != NULL) {
3185 		maprange = pci_maprange(pm->pm_value);
3186 		barlen = maprange == 64 ? 2 : 1;
3187 		return (barlen);
3188 	}
3189 
3190 	pci_read_bar(dev, reg, &map, &testval, NULL);
3191 	if (PCI_BAR_MEM(map)) {
3192 		type = SYS_RES_MEMORY;
3193 		if (map & PCIM_BAR_MEM_PREFETCH)
3194 			prefetch = 1;
3195 	} else
3196 		type = SYS_RES_IOPORT;
3197 	mapsize = pci_mapsize(testval);
3198 	base = pci_mapbase(map);
3199 #ifdef __PCI_BAR_ZERO_VALID
3200 	basezero = 0;
3201 #else
3202 	basezero = base == 0;
3203 #endif
3204 	maprange = pci_maprange(map);
3205 	barlen = maprange == 64 ? 2 : 1;
3206 
3207 	/*
3208 	 * For I/O registers, if bottom bit is set, and the next bit up
3209 	 * isn't clear, we know we have a BAR that doesn't conform to the
3210 	 * spec, so ignore it.  Also, sanity check the size of the data
3211 	 * areas to the type of memory involved.  Memory must be at least
3212 	 * 16 bytes in size, while I/O ranges must be at least 4.
3213 	 */
3214 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3215 		return (barlen);
3216 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3217 	    (type == SYS_RES_IOPORT && mapsize < 2))
3218 		return (barlen);
3219 
3220 	/* Save a record of this BAR. */
3221 	pm = pci_add_bar(dev, reg, map, mapsize);
3222 	if (bootverbose) {
3223 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3224 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3225 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3226 			printf(", port disabled\n");
3227 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3228 			printf(", memory disabled\n");
3229 		else
3230 			printf(", enabled\n");
3231 	}
3232 
3233 	/*
3234 	 * If base is 0, then we have problems if this architecture does
3235 	 * not allow that.  It is best to ignore such entries for the
3236 	 * moment.  These will be allocated later if the driver specifically
3237 	 * requests them.  However, some removable buses look better when
3238 	 * all resources are allocated, so allow '0' to be overriden.
3239 	 *
3240 	 * Similarly treat maps whose values is the same as the test value
3241 	 * read back.  These maps have had all f's written to them by the
3242 	 * BIOS in an attempt to disable the resources.
3243 	 */
3244 	if (!force && (basezero || map == testval))
3245 		return (barlen);
3246 	if ((u_long)base != base) {
3247 		device_printf(bus,
3248 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3249 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3250 		    pci_get_function(dev), reg);
3251 		return (barlen);
3252 	}
3253 
3254 	/*
3255 	 * This code theoretically does the right thing, but has
3256 	 * undesirable side effects in some cases where peripherals
3257 	 * respond oddly to having these bits enabled.  Let the user
3258 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3259 	 * default).
3260 	 */
3261 	if (pci_enable_io_modes) {
3262 		/* Turn on resources that have been left off by a lazy BIOS */
3263 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3264 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3265 			cmd |= PCIM_CMD_PORTEN;
3266 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3267 		}
3268 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3269 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3270 			cmd |= PCIM_CMD_MEMEN;
3271 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3272 		}
3273 	} else {
3274 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3275 			return (barlen);
3276 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3277 			return (barlen);
3278 	}
3279 
3280 	count = (pci_addr_t)1 << mapsize;
3281 	flags = RF_ALIGNMENT_LOG2(mapsize);
3282 	if (prefetch)
3283 		flags |= RF_PREFETCHABLE;
3284 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3285 		start = 0;	/* Let the parent decide. */
3286 		end = ~0;
3287 	} else {
3288 		start = base;
3289 		end = base + count - 1;
3290 	}
3291 	resource_list_add(rl, type, reg, start, end, count);
3292 
3293 	/*
3294 	 * Try to allocate the resource for this BAR from our parent
3295 	 * so that this resource range is already reserved.  The
3296 	 * driver for this device will later inherit this resource in
3297 	 * pci_alloc_resource().
3298 	 */
3299 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3300 	    flags);
3301 	if ((pci_do_realloc_bars
3302 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3303 	    && res == NULL && (start != 0 || end != ~0)) {
3304 		/*
3305 		 * If the allocation fails, try to allocate a resource for
3306 		 * this BAR using any available range.  The firmware felt
3307 		 * it was important enough to assign a resource, so don't
3308 		 * disable decoding if we can help it.
3309 		 */
3310 		resource_list_delete(rl, type, reg);
3311 		resource_list_add(rl, type, reg, 0, ~0, count);
3312 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3313 		    count, flags);
3314 	}
3315 	if (res == NULL) {
3316 		/*
3317 		 * If the allocation fails, delete the resource list entry
3318 		 * and disable decoding for this device.
3319 		 *
3320 		 * If the driver requests this resource in the future,
3321 		 * pci_reserve_map() will try to allocate a fresh
3322 		 * resource range.
3323 		 */
3324 		resource_list_delete(rl, type, reg);
3325 		pci_disable_io(dev, type);
3326 		if (bootverbose)
3327 			device_printf(bus,
3328 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3329 			    pci_get_domain(dev), pci_get_bus(dev),
3330 			    pci_get_slot(dev), pci_get_function(dev), reg);
3331 	} else {
3332 		start = rman_get_start(res);
3333 		pci_write_bar(dev, pm, start);
3334 	}
3335 	return (barlen);
3336 }
3337 
3338 /*
3339  * For ATA devices we need to decide early what addressing mode to use.
3340  * Legacy demands that the primary and secondary ATA ports sits on the
3341  * same addresses that old ISA hardware did. This dictates that we use
3342  * those addresses and ignore the BAR's if we cannot set PCI native
3343  * addressing mode.
3344  */
3345 static void
3346 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3347     uint32_t prefetchmask)
3348 {
3349 	int rid, type, progif;
3350 #if 0
3351 	/* if this device supports PCI native addressing use it */
3352 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3353 	if ((progif & 0x8a) == 0x8a) {
3354 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3355 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3356 			printf("Trying ATA native PCI addressing mode\n");
3357 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3358 		}
3359 	}
3360 #endif
3361 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3362 	type = SYS_RES_IOPORT;
3363 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3364 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3365 		    prefetchmask & (1 << 0));
3366 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3367 		    prefetchmask & (1 << 1));
3368 	} else {
3369 		rid = PCIR_BAR(0);
3370 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3371 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3372 		    0x1f7, 8, 0);
3373 		rid = PCIR_BAR(1);
3374 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3375 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3376 		    0x3f6, 1, 0);
3377 	}
3378 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3379 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3380 		    prefetchmask & (1 << 2));
3381 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3382 		    prefetchmask & (1 << 3));
3383 	} else {
3384 		rid = PCIR_BAR(2);
3385 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3386 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3387 		    0x177, 8, 0);
3388 		rid = PCIR_BAR(3);
3389 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3390 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3391 		    0x376, 1, 0);
3392 	}
3393 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3394 	    prefetchmask & (1 << 4));
3395 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3396 	    prefetchmask & (1 << 5));
3397 }
3398 
3399 static void
3400 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3401 {
3402 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3403 	pcicfgregs *cfg = &dinfo->cfg;
3404 	char tunable_name[64];
3405 	int irq;
3406 
3407 	/* Has to have an intpin to have an interrupt. */
3408 	if (cfg->intpin == 0)
3409 		return;
3410 
3411 	/* Let the user override the IRQ with a tunable. */
3412 	irq = PCI_INVALID_IRQ;
3413 	snprintf(tunable_name, sizeof(tunable_name),
3414 	    "hw.pci%d.%d.%d.INT%c.irq",
3415 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3416 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3417 		irq = PCI_INVALID_IRQ;
3418 
3419 	/*
3420 	 * If we didn't get an IRQ via the tunable, then we either use the
3421 	 * IRQ value in the intline register or we ask the bus to route an
3422 	 * interrupt for us.  If force_route is true, then we only use the
3423 	 * value in the intline register if the bus was unable to assign an
3424 	 * IRQ.
3425 	 */
3426 	if (!PCI_INTERRUPT_VALID(irq)) {
3427 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3428 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3429 		if (!PCI_INTERRUPT_VALID(irq))
3430 			irq = cfg->intline;
3431 	}
3432 
3433 	/* If after all that we don't have an IRQ, just bail. */
3434 	if (!PCI_INTERRUPT_VALID(irq))
3435 		return;
3436 
3437 	/* Update the config register if it changed. */
3438 	if (irq != cfg->intline) {
3439 		cfg->intline = irq;
3440 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3441 	}
3442 
3443 	/* Add this IRQ as rid 0 interrupt resource. */
3444 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3445 }
3446 
3447 /* Perform early OHCI takeover from SMM. */
3448 static void
3449 ohci_early_takeover(device_t self)
3450 {
3451 	struct resource *res;
3452 	uint32_t ctl;
3453 	int rid;
3454 	int i;
3455 
3456 	rid = PCIR_BAR(0);
3457 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3458 	if (res == NULL)
3459 		return;
3460 
3461 	ctl = bus_read_4(res, OHCI_CONTROL);
3462 	if (ctl & OHCI_IR) {
3463 		if (bootverbose)
3464 			printf("ohci early: "
3465 			    "SMM active, request owner change\n");
3466 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3467 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3468 			DELAY(1000);
3469 			ctl = bus_read_4(res, OHCI_CONTROL);
3470 		}
3471 		if (ctl & OHCI_IR) {
3472 			if (bootverbose)
3473 				printf("ohci early: "
3474 				    "SMM does not respond, resetting\n");
3475 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3476 		}
3477 		/* Disable interrupts */
3478 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3479 	}
3480 
3481 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3482 }
3483 
3484 /* Perform early UHCI takeover from SMM. */
3485 static void
3486 uhci_early_takeover(device_t self)
3487 {
3488 	struct resource *res;
3489 	int rid;
3490 
3491 	/*
3492 	 * Set the PIRQD enable bit and switch off all the others. We don't
3493 	 * want legacy support to interfere with us XXX Does this also mean
3494 	 * that the BIOS won't touch the keyboard anymore if it is connected
3495 	 * to the ports of the root hub?
3496 	 */
3497 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3498 
3499 	/* Disable interrupts */
3500 	rid = PCI_UHCI_BASE_REG;
3501 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3502 	if (res != NULL) {
3503 		bus_write_2(res, UHCI_INTR, 0);
3504 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3505 	}
3506 }
3507 
3508 /* Perform early EHCI takeover from SMM. */
3509 static void
3510 ehci_early_takeover(device_t self)
3511 {
3512 	struct resource *res;
3513 	uint32_t cparams;
3514 	uint32_t eec;
3515 	uint8_t eecp;
3516 	uint8_t bios_sem;
3517 	uint8_t offs;
3518 	int rid;
3519 	int i;
3520 
3521 	rid = PCIR_BAR(0);
3522 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3523 	if (res == NULL)
3524 		return;
3525 
3526 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3527 
3528 	/* Synchronise with the BIOS if it owns the controller. */
3529 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3530 	    eecp = EHCI_EECP_NEXT(eec)) {
3531 		eec = pci_read_config(self, eecp, 4);
3532 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3533 			continue;
3534 		}
3535 		bios_sem = pci_read_config(self, eecp +
3536 		    EHCI_LEGSUP_BIOS_SEM, 1);
3537 		if (bios_sem == 0) {
3538 			continue;
3539 		}
3540 		if (bootverbose)
3541 			printf("ehci early: "
3542 			    "SMM active, request owner change\n");
3543 
3544 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3545 
3546 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3547 			DELAY(1000);
3548 			bios_sem = pci_read_config(self, eecp +
3549 			    EHCI_LEGSUP_BIOS_SEM, 1);
3550 		}
3551 
3552 		if (bios_sem != 0) {
3553 			if (bootverbose)
3554 				printf("ehci early: "
3555 				    "SMM does not respond\n");
3556 		}
3557 		/* Disable interrupts */
3558 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3559 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3560 	}
3561 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3562 }
3563 
3564 /* Perform early XHCI takeover from SMM. */
3565 static void
3566 xhci_early_takeover(device_t self)
3567 {
3568 	struct resource *res;
3569 	uint32_t cparams;
3570 	uint32_t eec;
3571 	uint8_t eecp;
3572 	uint8_t bios_sem;
3573 	uint8_t offs;
3574 	int rid;
3575 	int i;
3576 
3577 	rid = PCIR_BAR(0);
3578 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3579 	if (res == NULL)
3580 		return;
3581 
3582 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3583 
3584 	eec = -1;
3585 
3586 	/* Synchronise with the BIOS if it owns the controller. */
3587 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3588 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3589 		eec = bus_read_4(res, eecp);
3590 
3591 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3592 			continue;
3593 
3594 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3595 		if (bios_sem == 0)
3596 			continue;
3597 
3598 		if (bootverbose)
3599 			printf("xhci early: "
3600 			    "SMM active, request owner change\n");
3601 
3602 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3603 
3604 		/* wait a maximum of 5 second */
3605 
3606 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3607 			DELAY(1000);
3608 			bios_sem = bus_read_1(res, eecp +
3609 			    XHCI_XECP_BIOS_SEM);
3610 		}
3611 
3612 		if (bios_sem != 0) {
3613 			if (bootverbose)
3614 				printf("xhci early: "
3615 				    "SMM does not respond\n");
3616 		}
3617 
3618 		/* Disable interrupts */
3619 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3620 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3621 		bus_read_4(res, offs + XHCI_USBSTS);
3622 	}
3623 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3624 }
3625 
3626 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3627 static void
3628 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3629     struct resource_list *rl)
3630 {
3631 	struct resource *res;
3632 	char *cp;
3633 	rman_res_t start, end, count;
3634 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3635 
3636 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3637 	case PCIM_HDRTYPE_BRIDGE:
3638 		sec_reg = PCIR_SECBUS_1;
3639 		sub_reg = PCIR_SUBBUS_1;
3640 		break;
3641 	case PCIM_HDRTYPE_CARDBUS:
3642 		sec_reg = PCIR_SECBUS_2;
3643 		sub_reg = PCIR_SUBBUS_2;
3644 		break;
3645 	default:
3646 		return;
3647 	}
3648 
3649 	/*
3650 	 * If the existing bus range is valid, attempt to reserve it
3651 	 * from our parent.  If this fails for any reason, clear the
3652 	 * secbus and subbus registers.
3653 	 *
3654 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3655 	 * This would at least preserve the existing sec_bus if it is
3656 	 * valid.
3657 	 */
3658 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3659 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3660 
3661 	/* Quirk handling. */
3662 	switch (pci_get_devid(dev)) {
3663 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3664 		sup_bus = pci_read_config(dev, 0x41, 1);
3665 		if (sup_bus != 0xff) {
3666 			sec_bus = sup_bus + 1;
3667 			sub_bus = sup_bus + 1;
3668 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3669 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3670 		}
3671 		break;
3672 
3673 	case 0x00dd10de:
3674 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3675 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3676 			break;
3677 		if (strncmp(cp, "Compal", 6) != 0) {
3678 			freeenv(cp);
3679 			break;
3680 		}
3681 		freeenv(cp);
3682 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3683 			break;
3684 		if (strncmp(cp, "08A0", 4) != 0) {
3685 			freeenv(cp);
3686 			break;
3687 		}
3688 		freeenv(cp);
3689 		if (sub_bus < 0xa) {
3690 			sub_bus = 0xa;
3691 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3692 		}
3693 		break;
3694 	}
3695 
3696 	if (bootverbose)
3697 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3698 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3699 		start = sec_bus;
3700 		end = sub_bus;
3701 		count = end - start + 1;
3702 
3703 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3704 
3705 		/*
3706 		 * If requested, clear secondary bus registers in
3707 		 * bridge devices to force a complete renumbering
3708 		 * rather than reserving the existing range.  However,
3709 		 * preserve the existing size.
3710 		 */
3711 		if (pci_clear_buses)
3712 			goto clear;
3713 
3714 		rid = 0;
3715 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3716 		    start, end, count, 0);
3717 		if (res != NULL)
3718 			return;
3719 
3720 		if (bootverbose)
3721 			device_printf(bus,
3722 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3723 			    pci_get_domain(dev), pci_get_bus(dev),
3724 			    pci_get_slot(dev), pci_get_function(dev));
3725 	}
3726 
3727 clear:
3728 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3729 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3730 }
3731 
3732 static struct resource *
3733 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3734     rman_res_t end, rman_res_t count, u_int flags)
3735 {
3736 	struct pci_devinfo *dinfo;
3737 	pcicfgregs *cfg;
3738 	struct resource_list *rl;
3739 	struct resource *res;
3740 	int sec_reg, sub_reg;
3741 
3742 	dinfo = device_get_ivars(child);
3743 	cfg = &dinfo->cfg;
3744 	rl = &dinfo->resources;
3745 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3746 	case PCIM_HDRTYPE_BRIDGE:
3747 		sec_reg = PCIR_SECBUS_1;
3748 		sub_reg = PCIR_SUBBUS_1;
3749 		break;
3750 	case PCIM_HDRTYPE_CARDBUS:
3751 		sec_reg = PCIR_SECBUS_2;
3752 		sub_reg = PCIR_SUBBUS_2;
3753 		break;
3754 	default:
3755 		return (NULL);
3756 	}
3757 
3758 	if (*rid != 0)
3759 		return (NULL);
3760 
3761 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3762 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3763 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3764 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3765 		    start, end, count, flags & ~RF_ACTIVE);
3766 		if (res == NULL) {
3767 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3768 			device_printf(child, "allocating %ju bus%s failed\n",
3769 			    count, count == 1 ? "" : "es");
3770 			return (NULL);
3771 		}
3772 		if (bootverbose)
3773 			device_printf(child,
3774 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3775 			    count == 1 ? "" : "es", rman_get_start(res));
3776 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3777 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3778 	}
3779 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3780 	    end, count, flags));
3781 }
3782 #endif
3783 
3784 static int
3785 pci_ea_bei_to_rid(device_t dev, int bei)
3786 {
3787 #ifdef PCI_IOV
3788 	struct pci_devinfo *dinfo;
3789 	int iov_pos;
3790 	struct pcicfg_iov *iov;
3791 
3792 	dinfo = device_get_ivars(dev);
3793 	iov = dinfo->cfg.iov;
3794 	if (iov != NULL)
3795 		iov_pos = iov->iov_pos;
3796 	else
3797 		iov_pos = 0;
3798 #endif
3799 
3800 	/* Check if matches BAR */
3801 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3802 	    (bei <= PCIM_EA_BEI_BAR_5))
3803 		return (PCIR_BAR(bei));
3804 
3805 	/* Check ROM */
3806 	if (bei == PCIM_EA_BEI_ROM)
3807 		return (PCIR_BIOS);
3808 
3809 #ifdef PCI_IOV
3810 	/* Check if matches VF_BAR */
3811 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3812 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3813 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3814 		    iov_pos);
3815 #endif
3816 
3817 	return (-1);
3818 }
3819 
3820 int
3821 pci_ea_is_enabled(device_t dev, int rid)
3822 {
3823 	struct pci_ea_entry *ea;
3824 	struct pci_devinfo *dinfo;
3825 
3826 	dinfo = device_get_ivars(dev);
3827 
3828 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3829 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3830 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3831 	}
3832 
3833 	return (0);
3834 }
3835 
3836 void
3837 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3838 {
3839 	struct pci_ea_entry *ea;
3840 	struct pci_devinfo *dinfo;
3841 	pci_addr_t start, end, count;
3842 	struct resource_list *rl;
3843 	int type, flags, rid;
3844 	struct resource *res;
3845 	uint32_t tmp;
3846 #ifdef PCI_IOV
3847 	struct pcicfg_iov *iov;
3848 #endif
3849 
3850 	dinfo = device_get_ivars(dev);
3851 	rl = &dinfo->resources;
3852 	flags = 0;
3853 
3854 #ifdef PCI_IOV
3855 	iov = dinfo->cfg.iov;
3856 #endif
3857 
3858 	if (dinfo->cfg.ea.ea_location == 0)
3859 		return;
3860 
3861 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3862 
3863 		/*
3864 		 * TODO: Ignore EA-BAR if is not enabled.
3865 		 *   Currently the EA implementation supports
3866 		 *   only situation, where EA structure contains
3867 		 *   predefined entries. In case they are not enabled
3868 		 *   leave them unallocated and proceed with
3869 		 *   a legacy-BAR mechanism.
3870 		 */
3871 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3872 			continue;
3873 
3874 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3875 		case PCIM_EA_P_MEM_PREFETCH:
3876 		case PCIM_EA_P_VF_MEM_PREFETCH:
3877 			flags = RF_PREFETCHABLE;
3878 			/* FALLTHROUGH */
3879 		case PCIM_EA_P_VF_MEM:
3880 		case PCIM_EA_P_MEM:
3881 			type = SYS_RES_MEMORY;
3882 			break;
3883 		case PCIM_EA_P_IO:
3884 			type = SYS_RES_IOPORT;
3885 			break;
3886 		default:
3887 			continue;
3888 		}
3889 
3890 		if (alloc_iov != 0) {
3891 #ifdef PCI_IOV
3892 			/* Allocating IOV, confirm BEI matches */
3893 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3894 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3895 				continue;
3896 #else
3897 			continue;
3898 #endif
3899 		} else {
3900 			/* Allocating BAR, confirm BEI matches */
3901 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3902 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3903 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3904 				continue;
3905 		}
3906 
3907 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3908 		if (rid < 0)
3909 			continue;
3910 
3911 		/* Skip resources already allocated by EA */
3912 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3913 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3914 			continue;
3915 
3916 		start = ea->eae_base;
3917 		count = ea->eae_max_offset + 1;
3918 #ifdef PCI_IOV
3919 		if (iov != NULL)
3920 			count = count * iov->iov_num_vfs;
3921 #endif
3922 		end = start + count - 1;
3923 		if (count == 0)
3924 			continue;
3925 
3926 		resource_list_add(rl, type, rid, start, end, count);
3927 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3928 		    flags);
3929 		if (res == NULL) {
3930 			resource_list_delete(rl, type, rid);
3931 
3932 			/*
3933 			 * Failed to allocate using EA, disable entry.
3934 			 * Another attempt to allocation will be performed
3935 			 * further, but this time using legacy BAR registers
3936 			 */
3937 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3938 			tmp &= ~PCIM_EA_ENABLE;
3939 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3940 
3941 			/*
3942 			 * Disabling entry might fail in case it is hardwired.
3943 			 * Read flags again to match current status.
3944 			 */
3945 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3946 
3947 			continue;
3948 		}
3949 
3950 		/* As per specification, fill BAR with zeros */
3951 		pci_write_config(dev, rid, 0, 4);
3952 	}
3953 }
3954 
3955 void
3956 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3957 {
3958 	struct pci_devinfo *dinfo;
3959 	pcicfgregs *cfg;
3960 	struct resource_list *rl;
3961 	const struct pci_quirk *q;
3962 	uint32_t devid;
3963 	int i;
3964 
3965 	dinfo = device_get_ivars(dev);
3966 	cfg = &dinfo->cfg;
3967 	rl = &dinfo->resources;
3968 	devid = (cfg->device << 16) | cfg->vendor;
3969 
3970 	/* Allocate resources using Enhanced Allocation */
3971 	pci_add_resources_ea(bus, dev, 0);
3972 
3973 	/* ATA devices needs special map treatment */
3974 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3975 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3976 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3977 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3978 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3979 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3980 	else
3981 		for (i = 0; i < cfg->nummaps;) {
3982 			/* Skip resources already managed by EA */
3983 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3984 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3985 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3986 				i++;
3987 				continue;
3988 			}
3989 
3990 			/*
3991 			 * Skip quirked resources.
3992 			 */
3993 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3994 				if (q->devid == devid &&
3995 				    q->type == PCI_QUIRK_UNMAP_REG &&
3996 				    q->arg1 == PCIR_BAR(i))
3997 					break;
3998 			if (q->devid != 0) {
3999 				i++;
4000 				continue;
4001 			}
4002 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4003 			    prefetchmask & (1 << i));
4004 		}
4005 
4006 	/*
4007 	 * Add additional, quirked resources.
4008 	 */
4009 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4010 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4011 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4012 
4013 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4014 #ifdef __PCI_REROUTE_INTERRUPT
4015 		/*
4016 		 * Try to re-route interrupts. Sometimes the BIOS or
4017 		 * firmware may leave bogus values in these registers.
4018 		 * If the re-route fails, then just stick with what we
4019 		 * have.
4020 		 */
4021 		pci_assign_interrupt(bus, dev, 1);
4022 #else
4023 		pci_assign_interrupt(bus, dev, 0);
4024 #endif
4025 	}
4026 
4027 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4028 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4029 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4030 			xhci_early_takeover(dev);
4031 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4032 			ehci_early_takeover(dev);
4033 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4034 			ohci_early_takeover(dev);
4035 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4036 			uhci_early_takeover(dev);
4037 	}
4038 
4039 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4040 	/*
4041 	 * Reserve resources for secondary bus ranges behind bridge
4042 	 * devices.
4043 	 */
4044 	pci_reserve_secbus(bus, dev, cfg, rl);
4045 #endif
4046 }
4047 
4048 static struct pci_devinfo *
4049 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4050     int slot, int func)
4051 {
4052 	struct pci_devinfo *dinfo;
4053 
4054 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4055 	if (dinfo != NULL)
4056 		pci_add_child(dev, dinfo);
4057 
4058 	return (dinfo);
4059 }
4060 
4061 void
4062 pci_add_children(device_t dev, int domain, int busno)
4063 {
4064 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4065 	device_t pcib = device_get_parent(dev);
4066 	struct pci_devinfo *dinfo;
4067 	int maxslots;
4068 	int s, f, pcifunchigh;
4069 	uint8_t hdrtype;
4070 	int first_func;
4071 
4072 	/*
4073 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4074 	 * enable ARI.  We must enable ARI before detecting the rest of the
4075 	 * functions on this bus as ARI changes the set of slots and functions
4076 	 * that are legal on this bus.
4077 	 */
4078 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4079 	if (dinfo != NULL && pci_enable_ari)
4080 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4081 
4082 	/*
4083 	 * Start looking for new devices on slot 0 at function 1 because we
4084 	 * just identified the device at slot 0, function 0.
4085 	 */
4086 	first_func = 1;
4087 
4088 	maxslots = PCIB_MAXSLOTS(pcib);
4089 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4090 		pcifunchigh = 0;
4091 		f = 0;
4092 		DELAY(1);
4093 		hdrtype = REG(PCIR_HDRTYPE, 1);
4094 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4095 			continue;
4096 		if (hdrtype & PCIM_MFDEV)
4097 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4098 		for (f = first_func; f <= pcifunchigh; f++)
4099 			pci_identify_function(pcib, dev, domain, busno, s, f);
4100 	}
4101 #undef REG
4102 }
4103 
4104 int
4105 pci_rescan_method(device_t dev)
4106 {
4107 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4108 	device_t pcib = device_get_parent(dev);
4109 	device_t child, *devlist, *unchanged;
4110 	int devcount, error, i, j, maxslots, oldcount;
4111 	int busno, domain, s, f, pcifunchigh;
4112 	uint8_t hdrtype;
4113 
4114 	/* No need to check for ARI on a rescan. */
4115 	error = device_get_children(dev, &devlist, &devcount);
4116 	if (error)
4117 		return (error);
4118 	if (devcount != 0) {
4119 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4120 		    M_NOWAIT | M_ZERO);
4121 		if (unchanged == NULL) {
4122 			free(devlist, M_TEMP);
4123 			return (ENOMEM);
4124 		}
4125 	} else
4126 		unchanged = NULL;
4127 
4128 	domain = pcib_get_domain(dev);
4129 	busno = pcib_get_bus(dev);
4130 	maxslots = PCIB_MAXSLOTS(pcib);
4131 	for (s = 0; s <= maxslots; s++) {
4132 		/* If function 0 is not present, skip to the next slot. */
4133 		f = 0;
4134 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4135 			continue;
4136 		pcifunchigh = 0;
4137 		hdrtype = REG(PCIR_HDRTYPE, 1);
4138 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4139 			continue;
4140 		if (hdrtype & PCIM_MFDEV)
4141 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4142 		for (f = 0; f <= pcifunchigh; f++) {
4143 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4144 				continue;
4145 
4146 			/*
4147 			 * Found a valid function.  Check if a
4148 			 * device_t for this device already exists.
4149 			 */
4150 			for (i = 0; i < devcount; i++) {
4151 				child = devlist[i];
4152 				if (child == NULL)
4153 					continue;
4154 				if (pci_get_slot(child) == s &&
4155 				    pci_get_function(child) == f) {
4156 					unchanged[i] = child;
4157 					goto next_func;
4158 				}
4159 			}
4160 
4161 			pci_identify_function(pcib, dev, domain, busno, s, f);
4162 		next_func:;
4163 		}
4164 	}
4165 
4166 	/* Remove devices that are no longer present. */
4167 	for (i = 0; i < devcount; i++) {
4168 		if (unchanged[i] != NULL)
4169 			continue;
4170 		device_delete_child(dev, devlist[i]);
4171 	}
4172 
4173 	free(devlist, M_TEMP);
4174 	oldcount = devcount;
4175 
4176 	/* Try to attach the devices just added. */
4177 	error = device_get_children(dev, &devlist, &devcount);
4178 	if (error) {
4179 		free(unchanged, M_TEMP);
4180 		return (error);
4181 	}
4182 
4183 	for (i = 0; i < devcount; i++) {
4184 		for (j = 0; j < oldcount; j++) {
4185 			if (devlist[i] == unchanged[j])
4186 				goto next_device;
4187 		}
4188 
4189 		device_probe_and_attach(devlist[i]);
4190 	next_device:;
4191 	}
4192 
4193 	free(unchanged, M_TEMP);
4194 	free(devlist, M_TEMP);
4195 	return (0);
4196 #undef REG
4197 }
4198 
4199 #ifdef PCI_IOV
4200 device_t
4201 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4202     uint16_t did)
4203 {
4204 	struct pci_devinfo *vf_dinfo;
4205 	device_t pcib;
4206 	int busno, slot, func;
4207 
4208 	pcib = device_get_parent(bus);
4209 
4210 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4211 
4212 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4213 	    slot, func, vid, did);
4214 
4215 	vf_dinfo->cfg.flags |= PCICFG_VF;
4216 	pci_add_child(bus, vf_dinfo);
4217 
4218 	return (vf_dinfo->cfg.dev);
4219 }
4220 
4221 device_t
4222 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4223     uint16_t vid, uint16_t did)
4224 {
4225 
4226 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4227 }
4228 #endif
4229 
4230 static void
4231 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4232 {
4233 	int aer;
4234 	uint32_t r;
4235 	uint16_t r2;
4236 
4237 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4238 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4239 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4240 		    PCIER_ROOT_CTL, 2);
4241 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4242 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4243 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4244 		    PCIER_ROOT_CTL, r2, 2);
4245 	}
4246 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4247 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4248 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4249 		if (r != 0 && bootverbose) {
4250 			pci_printf(&dinfo->cfg,
4251 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4252 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4253 			    4));
4254 		}
4255 
4256 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4257 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4258 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4259 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4260 		    PCIM_AER_UC_POISONED_TLP |
4261 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4262 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4263 		    PCIM_AER_UC_COMPLETER_ABORT |
4264 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4265 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4266 		    PCIM_AER_UC_MALFORMED_TLP |
4267 		    PCIM_AER_UC_ECRC_ERROR |
4268 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4269 		    PCIM_AER_UC_ACS_VIOLATION |
4270 		    PCIM_AER_UC_INTERNAL_ERROR |
4271 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4272 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4273 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4274 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4275 
4276 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4277 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4278 		if (r != 0 && bootverbose) {
4279 			pci_printf(&dinfo->cfg,
4280 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4281 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4282 			    4));
4283 		}
4284 
4285 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4286 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4287 		    PCIM_AER_COR_BAD_TLP |
4288 		    PCIM_AER_COR_BAD_DLLP |
4289 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4290 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4291 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4292 		    PCIM_AER_COR_INTERNAL_ERROR |
4293 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4294 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4295 
4296 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4297 		    PCIER_DEVICE_CTL, 2);
4298 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4299 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4300 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4301 		    PCIER_DEVICE_CTL, r, 2);
4302 	}
4303 }
4304 
4305 void
4306 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4307 {
4308 	device_t dev;
4309 
4310 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4311 	device_set_ivars(dev, dinfo);
4312 	resource_list_init(&dinfo->resources);
4313 	pci_cfg_save(dev, dinfo, 0);
4314 	pci_cfg_restore(dev, dinfo);
4315 	pci_print_verbose(dinfo);
4316 	pci_add_resources(bus, dev, 0, 0);
4317 	pci_child_added(dinfo->cfg.dev);
4318 
4319 	if (pci_clear_aer_on_attach)
4320 		pci_add_child_clear_aer(dev, dinfo);
4321 
4322 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4323 }
4324 
4325 void
4326 pci_child_added_method(device_t dev, device_t child)
4327 {
4328 
4329 }
4330 
4331 static int
4332 pci_probe(device_t dev)
4333 {
4334 
4335 	device_set_desc(dev, "PCI bus");
4336 
4337 	/* Allow other subclasses to override this driver. */
4338 	return (BUS_PROBE_GENERIC);
4339 }
4340 
4341 int
4342 pci_attach_common(device_t dev)
4343 {
4344 	struct pci_softc *sc;
4345 	int busno, domain;
4346 #ifdef PCI_DMA_BOUNDARY
4347 	int error, tag_valid;
4348 #endif
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 #ifdef PCI_DMA_BOUNDARY
4369 	tag_valid = 0;
4370 	if (device_get_devclass(device_get_parent(device_get_parent(dev))) !=
4371 	    devclass_find("pci")) {
4372 		error = bus_dma_tag_create(bus_get_dma_tag(dev), 1,
4373 		    PCI_DMA_BOUNDARY, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4374 		    NULL, NULL, BUS_SPACE_MAXSIZE, BUS_SPACE_UNRESTRICTED,
4375 		    BUS_SPACE_MAXSIZE, 0, NULL, NULL, &sc->sc_dma_tag);
4376 		if (error)
4377 			device_printf(dev, "Failed to create DMA tag: %d\n",
4378 			    error);
4379 		else
4380 			tag_valid = 1;
4381 	}
4382 	if (!tag_valid)
4383 #endif
4384 		sc->sc_dma_tag = bus_get_dma_tag(dev);
4385 	return (0);
4386 }
4387 
4388 static int
4389 pci_attach(device_t dev)
4390 {
4391 	int busno, domain, error;
4392 
4393 	error = pci_attach_common(dev);
4394 	if (error)
4395 		return (error);
4396 
4397 	/*
4398 	 * Since there can be multiple independently numbered PCI
4399 	 * buses on systems with multiple PCI domains, we can't use
4400 	 * the unit number to decide which bus we are probing. We ask
4401 	 * the parent pcib what our domain and bus numbers are.
4402 	 */
4403 	domain = pcib_get_domain(dev);
4404 	busno = pcib_get_bus(dev);
4405 	pci_add_children(dev, domain, busno);
4406 	return (bus_generic_attach(dev));
4407 }
4408 
4409 static int
4410 pci_detach(device_t dev)
4411 {
4412 #ifdef PCI_RES_BUS
4413 	struct pci_softc *sc;
4414 #endif
4415 	int error;
4416 
4417 	error = bus_generic_detach(dev);
4418 	if (error)
4419 		return (error);
4420 #ifdef PCI_RES_BUS
4421 	sc = device_get_softc(dev);
4422 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4423 	if (error)
4424 		return (error);
4425 #endif
4426 	return (device_delete_children(dev));
4427 }
4428 
4429 static void
4430 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4431 {
4432 	int line, unit;
4433 	const char *at;
4434 	char me1[24], me2[32];
4435 	uint8_t b, s, f;
4436 	uint32_t d;
4437 
4438 	d = pci_get_domain(child);
4439 	b = pci_get_bus(child);
4440 	s = pci_get_slot(child);
4441 	f = pci_get_function(child);
4442 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4443 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4444 	line = 0;
4445 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4446 		resource_string_value(name, unit, "at", &at);
4447 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4448 			continue; /* No match, try next candidate */
4449 		*unitp = unit;
4450 		return;
4451 	}
4452 }
4453 
4454 static void
4455 pci_set_power_child(device_t dev, device_t child, int state)
4456 {
4457 	device_t pcib;
4458 	int dstate;
4459 
4460 	/*
4461 	 * Set the device to the given state.  If the firmware suggests
4462 	 * a different power state, use it instead.  If power management
4463 	 * is not present, the firmware is responsible for managing
4464 	 * device power.  Skip children who aren't attached since they
4465 	 * are handled separately.
4466 	 */
4467 	pcib = device_get_parent(dev);
4468 	dstate = state;
4469 	if (device_is_attached(child) &&
4470 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4471 		pci_set_powerstate(child, dstate);
4472 }
4473 
4474 int
4475 pci_suspend_child(device_t dev, device_t child)
4476 {
4477 	struct pci_devinfo *dinfo;
4478 	struct resource_list_entry *rle;
4479 	int error;
4480 
4481 	dinfo = device_get_ivars(child);
4482 
4483 	/*
4484 	 * Save the PCI configuration space for the child and set the
4485 	 * device in the appropriate power state for this sleep state.
4486 	 */
4487 	pci_cfg_save(child, dinfo, 0);
4488 
4489 	/* Suspend devices before potentially powering them down. */
4490 	error = bus_generic_suspend_child(dev, child);
4491 
4492 	if (error)
4493 		return (error);
4494 
4495 	if (pci_do_power_suspend) {
4496 		/*
4497 		 * Make sure this device's interrupt handler is not invoked
4498 		 * in the case the device uses a shared interrupt that can
4499 		 * be raised by some other device.
4500 		 * This is applicable only to regular (legacy) PCI interrupts
4501 		 * as MSI/MSI-X interrupts are never shared.
4502 		 */
4503 		rle = resource_list_find(&dinfo->resources,
4504 		    SYS_RES_IRQ, 0);
4505 		if (rle != NULL && rle->res != NULL)
4506 			(void)bus_suspend_intr(child, rle->res);
4507 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4508 	}
4509 
4510 	return (0);
4511 }
4512 
4513 int
4514 pci_resume_child(device_t dev, device_t child)
4515 {
4516 	struct pci_devinfo *dinfo;
4517 	struct resource_list_entry *rle;
4518 
4519 	if (pci_do_power_resume)
4520 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4521 
4522 	dinfo = device_get_ivars(child);
4523 	pci_cfg_restore(child, dinfo);
4524 	if (!device_is_attached(child))
4525 		pci_cfg_save(child, dinfo, 1);
4526 
4527 	bus_generic_resume_child(dev, child);
4528 
4529 	/*
4530 	 * Allow interrupts only after fully resuming the driver and hardware.
4531 	 */
4532 	if (pci_do_power_suspend) {
4533 		/* See pci_suspend_child for details. */
4534 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4535 		if (rle != NULL && rle->res != NULL)
4536 			(void)bus_resume_intr(child, rle->res);
4537 	}
4538 
4539 	return (0);
4540 }
4541 
4542 int
4543 pci_resume(device_t dev)
4544 {
4545 	device_t child, *devlist;
4546 	int error, i, numdevs;
4547 
4548 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4549 		return (error);
4550 
4551 	/*
4552 	 * Resume critical devices first, then everything else later.
4553 	 */
4554 	for (i = 0; i < numdevs; i++) {
4555 		child = devlist[i];
4556 		switch (pci_get_class(child)) {
4557 		case PCIC_DISPLAY:
4558 		case PCIC_MEMORY:
4559 		case PCIC_BRIDGE:
4560 		case PCIC_BASEPERIPH:
4561 			BUS_RESUME_CHILD(dev, child);
4562 			break;
4563 		}
4564 	}
4565 	for (i = 0; i < numdevs; i++) {
4566 		child = devlist[i];
4567 		switch (pci_get_class(child)) {
4568 		case PCIC_DISPLAY:
4569 		case PCIC_MEMORY:
4570 		case PCIC_BRIDGE:
4571 		case PCIC_BASEPERIPH:
4572 			break;
4573 		default:
4574 			BUS_RESUME_CHILD(dev, child);
4575 		}
4576 	}
4577 	free(devlist, M_TEMP);
4578 	return (0);
4579 }
4580 
4581 static void
4582 pci_load_vendor_data(void)
4583 {
4584 	caddr_t data;
4585 	void *ptr;
4586 	size_t sz;
4587 
4588 	data = preload_search_by_type("pci_vendor_data");
4589 	if (data != NULL) {
4590 		ptr = preload_fetch_addr(data);
4591 		sz = preload_fetch_size(data);
4592 		if (ptr != NULL && sz != 0) {
4593 			pci_vendordata = ptr;
4594 			pci_vendordata_size = sz;
4595 			/* terminate the database */
4596 			pci_vendordata[pci_vendordata_size] = '\n';
4597 		}
4598 	}
4599 }
4600 
4601 void
4602 pci_driver_added(device_t dev, driver_t *driver)
4603 {
4604 	int numdevs;
4605 	device_t *devlist;
4606 	device_t child;
4607 	struct pci_devinfo *dinfo;
4608 	int i;
4609 
4610 	if (bootverbose)
4611 		device_printf(dev, "driver added\n");
4612 	DEVICE_IDENTIFY(driver, dev);
4613 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4614 		return;
4615 	for (i = 0; i < numdevs; i++) {
4616 		child = devlist[i];
4617 		if (device_get_state(child) != DS_NOTPRESENT)
4618 			continue;
4619 		dinfo = device_get_ivars(child);
4620 		pci_print_verbose(dinfo);
4621 		if (bootverbose)
4622 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4623 		pci_cfg_restore(child, dinfo);
4624 		if (device_probe_and_attach(child) != 0)
4625 			pci_child_detached(dev, child);
4626 	}
4627 	free(devlist, M_TEMP);
4628 }
4629 
4630 int
4631 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4632     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4633 {
4634 	struct pci_devinfo *dinfo;
4635 	struct msix_table_entry *mte;
4636 	struct msix_vector *mv;
4637 	uint64_t addr;
4638 	uint32_t data;
4639 	void *cookie;
4640 	int error, rid;
4641 
4642 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4643 	    arg, &cookie);
4644 	if (error)
4645 		return (error);
4646 
4647 	/* If this is not a direct child, just bail out. */
4648 	if (device_get_parent(child) != dev) {
4649 		*cookiep = cookie;
4650 		return(0);
4651 	}
4652 
4653 	rid = rman_get_rid(irq);
4654 	if (rid == 0) {
4655 		/* Make sure that INTx is enabled */
4656 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4657 	} else {
4658 		/*
4659 		 * Check to see if the interrupt is MSI or MSI-X.
4660 		 * Ask our parent to map the MSI and give
4661 		 * us the address and data register values.
4662 		 * If we fail for some reason, teardown the
4663 		 * interrupt handler.
4664 		 */
4665 		dinfo = device_get_ivars(child);
4666 		if (dinfo->cfg.msi.msi_alloc > 0) {
4667 			if (dinfo->cfg.msi.msi_addr == 0) {
4668 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4669 			    ("MSI has handlers, but vectors not mapped"));
4670 				error = PCIB_MAP_MSI(device_get_parent(dev),
4671 				    child, rman_get_start(irq), &addr, &data);
4672 				if (error)
4673 					goto bad;
4674 				dinfo->cfg.msi.msi_addr = addr;
4675 				dinfo->cfg.msi.msi_data = data;
4676 			}
4677 			if (dinfo->cfg.msi.msi_handlers == 0)
4678 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4679 				    dinfo->cfg.msi.msi_data);
4680 			dinfo->cfg.msi.msi_handlers++;
4681 		} else {
4682 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4683 			    ("No MSI or MSI-X interrupts allocated"));
4684 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4685 			    ("MSI-X index too high"));
4686 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4687 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4688 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4689 			KASSERT(mv->mv_irq == rman_get_start(irq),
4690 			    ("IRQ mismatch"));
4691 			if (mv->mv_address == 0) {
4692 				KASSERT(mte->mte_handlers == 0,
4693 		    ("MSI-X table entry has handlers, but vector not mapped"));
4694 				error = PCIB_MAP_MSI(device_get_parent(dev),
4695 				    child, rman_get_start(irq), &addr, &data);
4696 				if (error)
4697 					goto bad;
4698 				mv->mv_address = addr;
4699 				mv->mv_data = data;
4700 			}
4701 
4702 			/*
4703 			 * The MSIX table entry must be made valid by
4704 			 * incrementing the mte_handlers before
4705 			 * calling pci_enable_msix() and
4706 			 * pci_resume_msix(). Else the MSIX rewrite
4707 			 * table quirk will not work as expected.
4708 			 */
4709 			mte->mte_handlers++;
4710 			if (mte->mte_handlers == 1) {
4711 				pci_enable_msix(child, rid - 1, mv->mv_address,
4712 				    mv->mv_data);
4713 				pci_unmask_msix(child, rid - 1);
4714 			}
4715 		}
4716 
4717 		/*
4718 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4719 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4720 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4721 		 */
4722 		if (!pci_has_quirk(pci_get_devid(child),
4723 		    PCI_QUIRK_MSI_INTX_BUG))
4724 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4725 		else
4726 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4727 	bad:
4728 		if (error) {
4729 			(void)bus_generic_teardown_intr(dev, child, irq,
4730 			    cookie);
4731 			return (error);
4732 		}
4733 	}
4734 	*cookiep = cookie;
4735 	return (0);
4736 }
4737 
4738 int
4739 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4740     void *cookie)
4741 {
4742 	struct msix_table_entry *mte;
4743 	struct resource_list_entry *rle;
4744 	struct pci_devinfo *dinfo;
4745 	int error, rid;
4746 
4747 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4748 		return (EINVAL);
4749 
4750 	/* If this isn't a direct child, just bail out */
4751 	if (device_get_parent(child) != dev)
4752 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4753 
4754 	rid = rman_get_rid(irq);
4755 	if (rid == 0) {
4756 		/* Mask INTx */
4757 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4758 	} else {
4759 		/*
4760 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4761 		 * decrement the appropriate handlers count and mask the
4762 		 * MSI-X message, or disable MSI messages if the count
4763 		 * drops to 0.
4764 		 */
4765 		dinfo = device_get_ivars(child);
4766 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4767 		if (rle->res != irq)
4768 			return (EINVAL);
4769 		if (dinfo->cfg.msi.msi_alloc > 0) {
4770 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4771 			    ("MSI-X index too high"));
4772 			if (dinfo->cfg.msi.msi_handlers == 0)
4773 				return (EINVAL);
4774 			dinfo->cfg.msi.msi_handlers--;
4775 			if (dinfo->cfg.msi.msi_handlers == 0)
4776 				pci_disable_msi(child);
4777 		} else {
4778 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4779 			    ("No MSI or MSI-X interrupts allocated"));
4780 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4781 			    ("MSI-X index too high"));
4782 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4783 			if (mte->mte_handlers == 0)
4784 				return (EINVAL);
4785 			mte->mte_handlers--;
4786 			if (mte->mte_handlers == 0)
4787 				pci_mask_msix(child, rid - 1);
4788 		}
4789 	}
4790 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4791 	if (rid > 0)
4792 		KASSERT(error == 0,
4793 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4794 	return (error);
4795 }
4796 
4797 int
4798 pci_print_child(device_t dev, device_t child)
4799 {
4800 	struct pci_devinfo *dinfo;
4801 	struct resource_list *rl;
4802 	int retval = 0;
4803 
4804 	dinfo = device_get_ivars(child);
4805 	rl = &dinfo->resources;
4806 
4807 	retval += bus_print_child_header(dev, child);
4808 
4809 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4810 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4811 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4812 	if (device_get_flags(dev))
4813 		retval += printf(" flags %#x", device_get_flags(dev));
4814 
4815 	retval += printf(" at device %d.%d", pci_get_slot(child),
4816 	    pci_get_function(child));
4817 
4818 	retval += bus_print_child_domain(dev, child);
4819 	retval += bus_print_child_footer(dev, child);
4820 
4821 	return (retval);
4822 }
4823 
4824 static const struct
4825 {
4826 	int		class;
4827 	int		subclass;
4828 	int		report; /* 0 = bootverbose, 1 = always */
4829 	const char	*desc;
4830 } pci_nomatch_tab[] = {
4831 	{PCIC_OLD,		-1,			1, "old"},
4832 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4833 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4834 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4835 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4836 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4837 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4838 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4839 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4840 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4841 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4842 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4843 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4844 	{PCIC_NETWORK,		-1,			1, "network"},
4845 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4846 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4847 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4848 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4849 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4850 	{PCIC_DISPLAY,		-1,			1, "display"},
4851 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4852 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4853 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4854 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4855 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4856 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4857 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4858 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4859 	{PCIC_MEMORY,		-1,			1, "memory"},
4860 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4861 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4862 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4863 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4864 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4865 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4866 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4867 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4868 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4869 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4870 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4871 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4872 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4873 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4874 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4875 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4876 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4877 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4878 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4879 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4880 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4881 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4882 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4883 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4884 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4885 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4886 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4887 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4888 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4889 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4890 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4891 	{PCIC_DOCKING,		-1,			1, "docking station"},
4892 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4893 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4894 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4895 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4896 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4897 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4898 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4899 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4900 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4901 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4902 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4903 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4904 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4905 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4906 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4907 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4908 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4909 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4910 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4911 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4912 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4913 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4914 	{PCIC_DASP,		-1,			0, "dasp"},
4915 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4916 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4917 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4918 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4919 	{0, 0, 0,		NULL}
4920 };
4921 
4922 void
4923 pci_probe_nomatch(device_t dev, device_t child)
4924 {
4925 	int i, report;
4926 	const char *cp, *scp;
4927 	char *device;
4928 
4929 	/*
4930 	 * Look for a listing for this device in a loaded device database.
4931 	 */
4932 	report = 1;
4933 	if ((device = pci_describe_device(child)) != NULL) {
4934 		device_printf(dev, "<%s>", device);
4935 		free(device, M_DEVBUF);
4936 	} else {
4937 		/*
4938 		 * Scan the class/subclass descriptions for a general
4939 		 * description.
4940 		 */
4941 		cp = "unknown";
4942 		scp = NULL;
4943 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4944 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4945 				if (pci_nomatch_tab[i].subclass == -1) {
4946 					cp = pci_nomatch_tab[i].desc;
4947 					report = pci_nomatch_tab[i].report;
4948 				} else if (pci_nomatch_tab[i].subclass ==
4949 				    pci_get_subclass(child)) {
4950 					scp = pci_nomatch_tab[i].desc;
4951 					report = pci_nomatch_tab[i].report;
4952 				}
4953 			}
4954 		}
4955 		if (report || bootverbose) {
4956 			device_printf(dev, "<%s%s%s>",
4957 			    cp ? cp : "",
4958 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4959 			    scp ? scp : "");
4960 		}
4961 	}
4962 	if (report || bootverbose) {
4963 		printf(" at device %d.%d (no driver attached)\n",
4964 		    pci_get_slot(child), pci_get_function(child));
4965 	}
4966 	pci_cfg_save(child, device_get_ivars(child), 1);
4967 }
4968 
4969 void
4970 pci_child_detached(device_t dev, device_t child)
4971 {
4972 	struct pci_devinfo *dinfo;
4973 	struct resource_list *rl;
4974 
4975 	dinfo = device_get_ivars(child);
4976 	rl = &dinfo->resources;
4977 
4978 	/*
4979 	 * Have to deallocate IRQs before releasing any MSI messages and
4980 	 * have to release MSI messages before deallocating any memory
4981 	 * BARs.
4982 	 */
4983 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4984 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4985 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4986 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4987 		(void)pci_release_msi(child);
4988 	}
4989 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4990 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4991 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4992 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4993 #ifdef PCI_RES_BUS
4994 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4995 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4996 #endif
4997 
4998 	pci_cfg_save(child, dinfo, 1);
4999 }
5000 
5001 /*
5002  * Parse the PCI device database, if loaded, and return a pointer to a
5003  * description of the device.
5004  *
5005  * The database is flat text formatted as follows:
5006  *
5007  * Any line not in a valid format is ignored.
5008  * Lines are terminated with newline '\n' characters.
5009  *
5010  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5011  * the vendor name.
5012  *
5013  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5014  * - devices cannot be listed without a corresponding VENDOR line.
5015  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5016  * another TAB, then the device name.
5017  */
5018 
5019 /*
5020  * Assuming (ptr) points to the beginning of a line in the database,
5021  * return the vendor or device and description of the next entry.
5022  * The value of (vendor) or (device) inappropriate for the entry type
5023  * is set to -1.  Returns nonzero at the end of the database.
5024  *
5025  * Note that this is slightly unrobust in the face of corrupt data;
5026  * we attempt to safeguard against this by spamming the end of the
5027  * database with a newline when we initialise.
5028  */
5029 static int
5030 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5031 {
5032 	char	*cp = *ptr;
5033 	int	left;
5034 
5035 	*device = -1;
5036 	*vendor = -1;
5037 	**desc = '\0';
5038 	for (;;) {
5039 		left = pci_vendordata_size - (cp - pci_vendordata);
5040 		if (left <= 0) {
5041 			*ptr = cp;
5042 			return(1);
5043 		}
5044 
5045 		/* vendor entry? */
5046 		if (*cp != '\t' &&
5047 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5048 			break;
5049 		/* device entry? */
5050 		if (*cp == '\t' &&
5051 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5052 			break;
5053 
5054 		/* skip to next line */
5055 		while (*cp != '\n' && left > 0) {
5056 			cp++;
5057 			left--;
5058 		}
5059 		if (*cp == '\n') {
5060 			cp++;
5061 			left--;
5062 		}
5063 	}
5064 	/* skip to next line */
5065 	while (*cp != '\n' && left > 0) {
5066 		cp++;
5067 		left--;
5068 	}
5069 	if (*cp == '\n' && left > 0)
5070 		cp++;
5071 	*ptr = cp;
5072 	return(0);
5073 }
5074 
5075 static char *
5076 pci_describe_device(device_t dev)
5077 {
5078 	int	vendor, device;
5079 	char	*desc, *vp, *dp, *line;
5080 
5081 	desc = vp = dp = NULL;
5082 
5083 	/*
5084 	 * If we have no vendor data, we can't do anything.
5085 	 */
5086 	if (pci_vendordata == NULL)
5087 		goto out;
5088 
5089 	/*
5090 	 * Scan the vendor data looking for this device
5091 	 */
5092 	line = pci_vendordata;
5093 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5094 		goto out;
5095 	for (;;) {
5096 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5097 			goto out;
5098 		if (vendor == pci_get_vendor(dev))
5099 			break;
5100 	}
5101 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5102 		goto out;
5103 	for (;;) {
5104 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5105 			*dp = 0;
5106 			break;
5107 		}
5108 		if (vendor != -1) {
5109 			*dp = 0;
5110 			break;
5111 		}
5112 		if (device == pci_get_device(dev))
5113 			break;
5114 	}
5115 	if (dp[0] == '\0')
5116 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5117 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5118 	    NULL)
5119 		sprintf(desc, "%s, %s", vp, dp);
5120 out:
5121 	if (vp != NULL)
5122 		free(vp, M_DEVBUF);
5123 	if (dp != NULL)
5124 		free(dp, M_DEVBUF);
5125 	return(desc);
5126 }
5127 
5128 int
5129 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5130 {
5131 	struct pci_devinfo *dinfo;
5132 	pcicfgregs *cfg;
5133 
5134 	dinfo = device_get_ivars(child);
5135 	cfg = &dinfo->cfg;
5136 
5137 	switch (which) {
5138 	case PCI_IVAR_ETHADDR:
5139 		/*
5140 		 * The generic accessor doesn't deal with failure, so
5141 		 * we set the return value, then return an error.
5142 		 */
5143 		*((uint8_t **) result) = NULL;
5144 		return (EINVAL);
5145 	case PCI_IVAR_SUBVENDOR:
5146 		*result = cfg->subvendor;
5147 		break;
5148 	case PCI_IVAR_SUBDEVICE:
5149 		*result = cfg->subdevice;
5150 		break;
5151 	case PCI_IVAR_VENDOR:
5152 		*result = cfg->vendor;
5153 		break;
5154 	case PCI_IVAR_DEVICE:
5155 		*result = cfg->device;
5156 		break;
5157 	case PCI_IVAR_DEVID:
5158 		*result = (cfg->device << 16) | cfg->vendor;
5159 		break;
5160 	case PCI_IVAR_CLASS:
5161 		*result = cfg->baseclass;
5162 		break;
5163 	case PCI_IVAR_SUBCLASS:
5164 		*result = cfg->subclass;
5165 		break;
5166 	case PCI_IVAR_PROGIF:
5167 		*result = cfg->progif;
5168 		break;
5169 	case PCI_IVAR_REVID:
5170 		*result = cfg->revid;
5171 		break;
5172 	case PCI_IVAR_INTPIN:
5173 		*result = cfg->intpin;
5174 		break;
5175 	case PCI_IVAR_IRQ:
5176 		*result = cfg->intline;
5177 		break;
5178 	case PCI_IVAR_DOMAIN:
5179 		*result = cfg->domain;
5180 		break;
5181 	case PCI_IVAR_BUS:
5182 		*result = cfg->bus;
5183 		break;
5184 	case PCI_IVAR_SLOT:
5185 		*result = cfg->slot;
5186 		break;
5187 	case PCI_IVAR_FUNCTION:
5188 		*result = cfg->func;
5189 		break;
5190 	case PCI_IVAR_CMDREG:
5191 		*result = cfg->cmdreg;
5192 		break;
5193 	case PCI_IVAR_CACHELNSZ:
5194 		*result = cfg->cachelnsz;
5195 		break;
5196 	case PCI_IVAR_MINGNT:
5197 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5198 			*result = -1;
5199 			return (EINVAL);
5200 		}
5201 		*result = cfg->mingnt;
5202 		break;
5203 	case PCI_IVAR_MAXLAT:
5204 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5205 			*result = -1;
5206 			return (EINVAL);
5207 		}
5208 		*result = cfg->maxlat;
5209 		break;
5210 	case PCI_IVAR_LATTIMER:
5211 		*result = cfg->lattimer;
5212 		break;
5213 	default:
5214 		return (ENOENT);
5215 	}
5216 	return (0);
5217 }
5218 
5219 int
5220 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5221 {
5222 	struct pci_devinfo *dinfo;
5223 
5224 	dinfo = device_get_ivars(child);
5225 
5226 	switch (which) {
5227 	case PCI_IVAR_INTPIN:
5228 		dinfo->cfg.intpin = value;
5229 		return (0);
5230 	case PCI_IVAR_ETHADDR:
5231 	case PCI_IVAR_SUBVENDOR:
5232 	case PCI_IVAR_SUBDEVICE:
5233 	case PCI_IVAR_VENDOR:
5234 	case PCI_IVAR_DEVICE:
5235 	case PCI_IVAR_DEVID:
5236 	case PCI_IVAR_CLASS:
5237 	case PCI_IVAR_SUBCLASS:
5238 	case PCI_IVAR_PROGIF:
5239 	case PCI_IVAR_REVID:
5240 	case PCI_IVAR_IRQ:
5241 	case PCI_IVAR_DOMAIN:
5242 	case PCI_IVAR_BUS:
5243 	case PCI_IVAR_SLOT:
5244 	case PCI_IVAR_FUNCTION:
5245 		return (EINVAL);	/* disallow for now */
5246 
5247 	default:
5248 		return (ENOENT);
5249 	}
5250 }
5251 
5252 #include "opt_ddb.h"
5253 #ifdef DDB
5254 #include <ddb/ddb.h>
5255 #include <sys/cons.h>
5256 
5257 /*
5258  * List resources based on pci map registers, used for within ddb
5259  */
5260 
5261 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5262 {
5263 	struct pci_devinfo *dinfo;
5264 	struct devlist *devlist_head;
5265 	struct pci_conf *p;
5266 	const char *name;
5267 	int i, error, none_count;
5268 
5269 	none_count = 0;
5270 	/* get the head of the device queue */
5271 	devlist_head = &pci_devq;
5272 
5273 	/*
5274 	 * Go through the list of devices and print out devices
5275 	 */
5276 	for (error = 0, i = 0,
5277 	     dinfo = STAILQ_FIRST(devlist_head);
5278 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5279 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5280 
5281 		/* Populate pd_name and pd_unit */
5282 		name = NULL;
5283 		if (dinfo->cfg.dev)
5284 			name = device_get_name(dinfo->cfg.dev);
5285 
5286 		p = &dinfo->conf;
5287 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5288 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5289 			(name && *name) ? name : "none",
5290 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5291 			none_count++,
5292 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5293 			p->pc_sel.pc_func, (p->pc_class << 16) |
5294 			(p->pc_subclass << 8) | p->pc_progif,
5295 			(p->pc_subdevice << 16) | p->pc_subvendor,
5296 			(p->pc_device << 16) | p->pc_vendor,
5297 			p->pc_revid, p->pc_hdr);
5298 	}
5299 }
5300 #endif /* DDB */
5301 
5302 static struct resource *
5303 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5304     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5305     u_int flags)
5306 {
5307 	struct pci_devinfo *dinfo = device_get_ivars(child);
5308 	struct resource_list *rl = &dinfo->resources;
5309 	struct resource *res;
5310 	struct pci_map *pm;
5311 	uint16_t cmd;
5312 	pci_addr_t map, testval;
5313 	int mapsize;
5314 
5315 	res = NULL;
5316 
5317 	/* If rid is managed by EA, ignore it */
5318 	if (pci_ea_is_enabled(child, *rid))
5319 		goto out;
5320 
5321 	pm = pci_find_bar(child, *rid);
5322 	if (pm != NULL) {
5323 		/* This is a BAR that we failed to allocate earlier. */
5324 		mapsize = pm->pm_size;
5325 		map = pm->pm_value;
5326 	} else {
5327 		/*
5328 		 * Weed out the bogons, and figure out how large the
5329 		 * BAR/map is.  BARs that read back 0 here are bogus
5330 		 * and unimplemented.  Note: atapci in legacy mode are
5331 		 * special and handled elsewhere in the code.  If you
5332 		 * have a atapci device in legacy mode and it fails
5333 		 * here, that other code is broken.
5334 		 */
5335 		pci_read_bar(child, *rid, &map, &testval, NULL);
5336 
5337 		/*
5338 		 * Determine the size of the BAR and ignore BARs with a size
5339 		 * of 0.  Device ROM BARs use a different mask value.
5340 		 */
5341 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5342 			mapsize = pci_romsize(testval);
5343 		else
5344 			mapsize = pci_mapsize(testval);
5345 		if (mapsize == 0)
5346 			goto out;
5347 		pm = pci_add_bar(child, *rid, map, mapsize);
5348 	}
5349 
5350 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5351 		if (type != SYS_RES_MEMORY) {
5352 			if (bootverbose)
5353 				device_printf(dev,
5354 				    "child %s requested type %d for rid %#x,"
5355 				    " but the BAR says it is an memio\n",
5356 				    device_get_nameunit(child), type, *rid);
5357 			goto out;
5358 		}
5359 	} else {
5360 		if (type != SYS_RES_IOPORT) {
5361 			if (bootverbose)
5362 				device_printf(dev,
5363 				    "child %s requested type %d for rid %#x,"
5364 				    " but the BAR says it is an ioport\n",
5365 				    device_get_nameunit(child), type, *rid);
5366 			goto out;
5367 		}
5368 	}
5369 
5370 	/*
5371 	 * For real BARs, we need to override the size that
5372 	 * the driver requests, because that's what the BAR
5373 	 * actually uses and we would otherwise have a
5374 	 * situation where we might allocate the excess to
5375 	 * another driver, which won't work.
5376 	 */
5377 	count = ((pci_addr_t)1 << mapsize) * num;
5378 	if (RF_ALIGNMENT(flags) < mapsize)
5379 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5380 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5381 		flags |= RF_PREFETCHABLE;
5382 
5383 	/*
5384 	 * Allocate enough resource, and then write back the
5385 	 * appropriate BAR for that resource.
5386 	 */
5387 	resource_list_add(rl, type, *rid, start, end, count);
5388 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5389 	    count, flags & ~RF_ACTIVE);
5390 	if (res == NULL) {
5391 		resource_list_delete(rl, type, *rid);
5392 		device_printf(child,
5393 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5394 		    count, *rid, type, start, end);
5395 		goto out;
5396 	}
5397 	if (bootverbose)
5398 		device_printf(child,
5399 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5400 		    count, *rid, type, rman_get_start(res));
5401 
5402 	/* Disable decoding via the CMD register before updating the BAR */
5403 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5404 	pci_write_config(child, PCIR_COMMAND,
5405 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5406 
5407 	map = rman_get_start(res);
5408 	pci_write_bar(child, pm, map);
5409 
5410 	/* Restore the original value of the CMD register */
5411 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5412 out:
5413 	return (res);
5414 }
5415 
5416 struct resource *
5417 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5418     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5419     u_int flags)
5420 {
5421 	struct pci_devinfo *dinfo;
5422 	struct resource_list *rl;
5423 	struct resource_list_entry *rle;
5424 	struct resource *res;
5425 	pcicfgregs *cfg;
5426 
5427 	/*
5428 	 * Perform lazy resource allocation
5429 	 */
5430 	dinfo = device_get_ivars(child);
5431 	rl = &dinfo->resources;
5432 	cfg = &dinfo->cfg;
5433 	switch (type) {
5434 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5435 	case PCI_RES_BUS:
5436 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5437 		    flags));
5438 #endif
5439 	case SYS_RES_IRQ:
5440 		/*
5441 		 * Can't alloc legacy interrupt once MSI messages have
5442 		 * been allocated.
5443 		 */
5444 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5445 		    cfg->msix.msix_alloc > 0))
5446 			return (NULL);
5447 
5448 		/*
5449 		 * If the child device doesn't have an interrupt
5450 		 * routed and is deserving of an interrupt, try to
5451 		 * assign it one.
5452 		 */
5453 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5454 		    (cfg->intpin != 0))
5455 			pci_assign_interrupt(dev, child, 0);
5456 		break;
5457 	case SYS_RES_IOPORT:
5458 	case SYS_RES_MEMORY:
5459 #ifdef NEW_PCIB
5460 		/*
5461 		 * PCI-PCI bridge I/O window resources are not BARs.
5462 		 * For those allocations just pass the request up the
5463 		 * tree.
5464 		 */
5465 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5466 			switch (*rid) {
5467 			case PCIR_IOBASEL_1:
5468 			case PCIR_MEMBASE_1:
5469 			case PCIR_PMBASEL_1:
5470 				/*
5471 				 * XXX: Should we bother creating a resource
5472 				 * list entry?
5473 				 */
5474 				return (bus_generic_alloc_resource(dev, child,
5475 				    type, rid, start, end, count, flags));
5476 			}
5477 		}
5478 #endif
5479 		/* Reserve resources for this BAR if needed. */
5480 		rle = resource_list_find(rl, type, *rid);
5481 		if (rle == NULL) {
5482 			res = pci_reserve_map(dev, child, type, rid, start, end,
5483 			    count, num, flags);
5484 			if (res == NULL)
5485 				return (NULL);
5486 		}
5487 	}
5488 	return (resource_list_alloc(rl, dev, child, type, rid,
5489 	    start, end, count, flags));
5490 }
5491 
5492 struct resource *
5493 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5494     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5495 {
5496 #ifdef PCI_IOV
5497 	struct pci_devinfo *dinfo;
5498 #endif
5499 
5500 	if (device_get_parent(child) != dev)
5501 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5502 		    type, rid, start, end, count, flags));
5503 
5504 #ifdef PCI_IOV
5505 	dinfo = device_get_ivars(child);
5506 	if (dinfo->cfg.flags & PCICFG_VF) {
5507 		switch (type) {
5508 		/* VFs can't have I/O BARs. */
5509 		case SYS_RES_IOPORT:
5510 			return (NULL);
5511 		case SYS_RES_MEMORY:
5512 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5513 			    start, end, count, flags));
5514 		}
5515 
5516 		/* Fall through for other types of resource allocations. */
5517 	}
5518 #endif
5519 
5520 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5521 	    count, 1, flags));
5522 }
5523 
5524 int
5525 pci_release_resource(device_t dev, device_t child, int type, int rid,
5526     struct resource *r)
5527 {
5528 	struct pci_devinfo *dinfo;
5529 	struct resource_list *rl;
5530 	pcicfgregs *cfg;
5531 
5532 	if (device_get_parent(child) != dev)
5533 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5534 		    type, rid, r));
5535 
5536 	dinfo = device_get_ivars(child);
5537 	cfg = &dinfo->cfg;
5538 
5539 #ifdef PCI_IOV
5540 	if (dinfo->cfg.flags & PCICFG_VF) {
5541 		switch (type) {
5542 		/* VFs can't have I/O BARs. */
5543 		case SYS_RES_IOPORT:
5544 			return (EDOOFUS);
5545 		case SYS_RES_MEMORY:
5546 			return (pci_vf_release_mem_resource(dev, child, rid,
5547 			    r));
5548 		}
5549 
5550 		/* Fall through for other types of resource allocations. */
5551 	}
5552 #endif
5553 
5554 #ifdef NEW_PCIB
5555 	/*
5556 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5557 	 * those allocations just pass the request up the tree.
5558 	 */
5559 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5560 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5561 		switch (rid) {
5562 		case PCIR_IOBASEL_1:
5563 		case PCIR_MEMBASE_1:
5564 		case PCIR_PMBASEL_1:
5565 			return (bus_generic_release_resource(dev, child, type,
5566 			    rid, r));
5567 		}
5568 	}
5569 #endif
5570 
5571 	rl = &dinfo->resources;
5572 	return (resource_list_release(rl, dev, child, type, rid, r));
5573 }
5574 
5575 int
5576 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5577     struct resource *r)
5578 {
5579 	struct pci_devinfo *dinfo;
5580 	int error;
5581 
5582 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5583 	if (error)
5584 		return (error);
5585 
5586 	/* Enable decoding in the command register when activating BARs. */
5587 	if (device_get_parent(child) == dev) {
5588 		/* Device ROMs need their decoding explicitly enabled. */
5589 		dinfo = device_get_ivars(child);
5590 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5591 			pci_write_bar(child, pci_find_bar(child, rid),
5592 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5593 		switch (type) {
5594 		case SYS_RES_IOPORT:
5595 		case SYS_RES_MEMORY:
5596 			error = PCI_ENABLE_IO(dev, child, type);
5597 			break;
5598 		}
5599 	}
5600 	return (error);
5601 }
5602 
5603 int
5604 pci_deactivate_resource(device_t dev, device_t child, int type,
5605     int rid, struct resource *r)
5606 {
5607 	struct pci_devinfo *dinfo;
5608 	int error;
5609 
5610 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5611 	if (error)
5612 		return (error);
5613 
5614 	/* Disable decoding for device ROMs. */
5615 	if (device_get_parent(child) == dev) {
5616 		dinfo = device_get_ivars(child);
5617 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5618 			pci_write_bar(child, pci_find_bar(child, rid),
5619 			    rman_get_start(r));
5620 	}
5621 	return (0);
5622 }
5623 
5624 void
5625 pci_child_deleted(device_t dev, device_t child)
5626 {
5627 	struct resource_list_entry *rle;
5628 	struct resource_list *rl;
5629 	struct pci_devinfo *dinfo;
5630 
5631 	dinfo = device_get_ivars(child);
5632 	rl = &dinfo->resources;
5633 
5634 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5635 
5636 	/* Turn off access to resources we're about to free */
5637 	if (bus_child_present(child) != 0) {
5638 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5639 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5640 
5641 		pci_disable_busmaster(child);
5642 	}
5643 
5644 	/* Free all allocated resources */
5645 	STAILQ_FOREACH(rle, rl, link) {
5646 		if (rle->res) {
5647 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5648 			    resource_list_busy(rl, rle->type, rle->rid)) {
5649 				pci_printf(&dinfo->cfg,
5650 				    "Resource still owned, oops. "
5651 				    "(type=%d, rid=%d, addr=%lx)\n",
5652 				    rle->type, rle->rid,
5653 				    rman_get_start(rle->res));
5654 				bus_release_resource(child, rle->type, rle->rid,
5655 				    rle->res);
5656 			}
5657 			resource_list_unreserve(rl, dev, child, rle->type,
5658 			    rle->rid);
5659 		}
5660 	}
5661 	resource_list_free(rl);
5662 
5663 	pci_freecfg(dinfo);
5664 }
5665 
5666 void
5667 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5668 {
5669 	struct pci_devinfo *dinfo;
5670 	struct resource_list *rl;
5671 	struct resource_list_entry *rle;
5672 
5673 	if (device_get_parent(child) != dev)
5674 		return;
5675 
5676 	dinfo = device_get_ivars(child);
5677 	rl = &dinfo->resources;
5678 	rle = resource_list_find(rl, type, rid);
5679 	if (rle == NULL)
5680 		return;
5681 
5682 	if (rle->res) {
5683 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5684 		    resource_list_busy(rl, type, rid)) {
5685 			device_printf(dev, "delete_resource: "
5686 			    "Resource still owned by child, oops. "
5687 			    "(type=%d, rid=%d, addr=%jx)\n",
5688 			    type, rid, rman_get_start(rle->res));
5689 			return;
5690 		}
5691 		resource_list_unreserve(rl, dev, child, type, rid);
5692 	}
5693 	resource_list_delete(rl, type, rid);
5694 }
5695 
5696 struct resource_list *
5697 pci_get_resource_list (device_t dev, device_t child)
5698 {
5699 	struct pci_devinfo *dinfo = device_get_ivars(child);
5700 
5701 	return (&dinfo->resources);
5702 }
5703 
5704 #ifdef ACPI_DMAR
5705 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5706 bus_dma_tag_t
5707 pci_get_dma_tag(device_t bus, device_t dev)
5708 {
5709 	bus_dma_tag_t tag;
5710 	struct pci_softc *sc;
5711 
5712 	if (device_get_parent(dev) == bus) {
5713 		/* try dmar and return if it works */
5714 		tag = dmar_get_dma_tag(bus, dev);
5715 	} else
5716 		tag = NULL;
5717 	if (tag == NULL) {
5718 		sc = device_get_softc(bus);
5719 		tag = sc->sc_dma_tag;
5720 	}
5721 	return (tag);
5722 }
5723 #else
5724 bus_dma_tag_t
5725 pci_get_dma_tag(device_t bus, device_t dev)
5726 {
5727 	struct pci_softc *sc = device_get_softc(bus);
5728 
5729 	return (sc->sc_dma_tag);
5730 }
5731 #endif
5732 
5733 uint32_t
5734 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5735 {
5736 	struct pci_devinfo *dinfo = device_get_ivars(child);
5737 	pcicfgregs *cfg = &dinfo->cfg;
5738 
5739 #ifdef PCI_IOV
5740 	/*
5741 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5742 	 * emulate them here.
5743 	 */
5744 	if (cfg->flags & PCICFG_VF) {
5745 		if (reg == PCIR_VENDOR) {
5746 			switch (width) {
5747 			case 4:
5748 				return (cfg->device << 16 | cfg->vendor);
5749 			case 2:
5750 				return (cfg->vendor);
5751 			case 1:
5752 				return (cfg->vendor & 0xff);
5753 			default:
5754 				return (0xffffffff);
5755 			}
5756 		} else if (reg == PCIR_DEVICE) {
5757 			switch (width) {
5758 			/* Note that an unaligned 4-byte read is an error. */
5759 			case 2:
5760 				return (cfg->device);
5761 			case 1:
5762 				return (cfg->device & 0xff);
5763 			default:
5764 				return (0xffffffff);
5765 			}
5766 		}
5767 	}
5768 #endif
5769 
5770 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5771 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5772 }
5773 
5774 void
5775 pci_write_config_method(device_t dev, device_t child, int reg,
5776     uint32_t val, int width)
5777 {
5778 	struct pci_devinfo *dinfo = device_get_ivars(child);
5779 	pcicfgregs *cfg = &dinfo->cfg;
5780 
5781 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5782 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5783 }
5784 
5785 int
5786 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5787     size_t buflen)
5788 {
5789 
5790 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5791 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5792 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5793 	return (0);
5794 }
5795 
5796 int
5797 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5798     size_t buflen)
5799 {
5800 	struct pci_devinfo *dinfo;
5801 	pcicfgregs *cfg;
5802 
5803 	dinfo = device_get_ivars(child);
5804 	cfg = &dinfo->cfg;
5805 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5806 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5807 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5808 	    cfg->progif);
5809 	return (0);
5810 }
5811 
5812 int
5813 pci_assign_interrupt_method(device_t dev, device_t child)
5814 {
5815 	struct pci_devinfo *dinfo = device_get_ivars(child);
5816 	pcicfgregs *cfg = &dinfo->cfg;
5817 
5818 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5819 	    cfg->intpin));
5820 }
5821 
5822 static void
5823 pci_lookup(void *arg, const char *name, device_t *dev)
5824 {
5825 	long val;
5826 	char *end;
5827 	int domain, bus, slot, func;
5828 
5829 	if (*dev != NULL)
5830 		return;
5831 
5832 	/*
5833 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5834 	 * pciB:S:F.  In the latter case, the domain is assumed to
5835 	 * be zero.
5836 	 */
5837 	if (strncmp(name, "pci", 3) != 0)
5838 		return;
5839 	val = strtol(name + 3, &end, 10);
5840 	if (val < 0 || val > INT_MAX || *end != ':')
5841 		return;
5842 	domain = val;
5843 	val = strtol(end + 1, &end, 10);
5844 	if (val < 0 || val > INT_MAX || *end != ':')
5845 		return;
5846 	bus = val;
5847 	val = strtol(end + 1, &end, 10);
5848 	if (val < 0 || val > INT_MAX)
5849 		return;
5850 	slot = val;
5851 	if (*end == ':') {
5852 		val = strtol(end + 1, &end, 10);
5853 		if (val < 0 || val > INT_MAX || *end != '\0')
5854 			return;
5855 		func = val;
5856 	} else if (*end == '\0') {
5857 		func = slot;
5858 		slot = bus;
5859 		bus = domain;
5860 		domain = 0;
5861 	} else
5862 		return;
5863 
5864 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5865 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5866 		return;
5867 
5868 	*dev = pci_find_dbsf(domain, bus, slot, func);
5869 }
5870 
5871 static int
5872 pci_modevent(module_t mod, int what, void *arg)
5873 {
5874 	static struct cdev *pci_cdev;
5875 	static eventhandler_tag tag;
5876 
5877 	switch (what) {
5878 	case MOD_LOAD:
5879 		STAILQ_INIT(&pci_devq);
5880 		pci_generation = 0;
5881 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5882 		    "pci");
5883 		pci_load_vendor_data();
5884 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5885 		    1000);
5886 		break;
5887 
5888 	case MOD_UNLOAD:
5889 		if (tag != NULL)
5890 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5891 		destroy_dev(pci_cdev);
5892 		break;
5893 	}
5894 
5895 	return (0);
5896 }
5897 
5898 static void
5899 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5900 {
5901 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5902 	struct pcicfg_pcie *cfg;
5903 	int version, pos;
5904 
5905 	cfg = &dinfo->cfg.pcie;
5906 	pos = cfg->pcie_location;
5907 
5908 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5909 
5910 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5911 
5912 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5913 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5914 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5915 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5916 
5917 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5918 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5919 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5920 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5921 
5922 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5923 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5924 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5925 
5926 	if (version > 1) {
5927 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5928 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5929 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5930 	}
5931 #undef WREG
5932 }
5933 
5934 static void
5935 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5936 {
5937 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5938 	    dinfo->cfg.pcix.pcix_command,  2);
5939 }
5940 
5941 void
5942 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5943 {
5944 
5945 	/*
5946 	 * Restore the device to full power mode.  We must do this
5947 	 * before we restore the registers because moving from D3 to
5948 	 * D0 will cause the chip's BARs and some other registers to
5949 	 * be reset to some unknown power on reset values.  Cut down
5950 	 * the noise on boot by doing nothing if we are already in
5951 	 * state D0.
5952 	 */
5953 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5954 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5955 	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5956 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5957 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5958 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5959 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5960 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5961 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5962 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5963 	case PCIM_HDRTYPE_NORMAL:
5964 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5965 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5966 		break;
5967 	case PCIM_HDRTYPE_BRIDGE:
5968 		pci_write_config(dev, PCIR_SECLAT_1,
5969 		    dinfo->cfg.bridge.br_seclat, 1);
5970 		pci_write_config(dev, PCIR_SUBBUS_1,
5971 		    dinfo->cfg.bridge.br_subbus, 1);
5972 		pci_write_config(dev, PCIR_SECBUS_1,
5973 		    dinfo->cfg.bridge.br_secbus, 1);
5974 		pci_write_config(dev, PCIR_PRIBUS_1,
5975 		    dinfo->cfg.bridge.br_pribus, 1);
5976 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5977 		    dinfo->cfg.bridge.br_control, 2);
5978 		break;
5979 	case PCIM_HDRTYPE_CARDBUS:
5980 		pci_write_config(dev, PCIR_SECLAT_2,
5981 		    dinfo->cfg.bridge.br_seclat, 1);
5982 		pci_write_config(dev, PCIR_SUBBUS_2,
5983 		    dinfo->cfg.bridge.br_subbus, 1);
5984 		pci_write_config(dev, PCIR_SECBUS_2,
5985 		    dinfo->cfg.bridge.br_secbus, 1);
5986 		pci_write_config(dev, PCIR_PRIBUS_2,
5987 		    dinfo->cfg.bridge.br_pribus, 1);
5988 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5989 		    dinfo->cfg.bridge.br_control, 2);
5990 		break;
5991 	}
5992 	pci_restore_bars(dev);
5993 
5994 	/*
5995 	 * Restore extended capabilities for PCI-Express and PCI-X
5996 	 */
5997 	if (dinfo->cfg.pcie.pcie_location != 0)
5998 		pci_cfg_restore_pcie(dev, dinfo);
5999 	if (dinfo->cfg.pcix.pcix_location != 0)
6000 		pci_cfg_restore_pcix(dev, dinfo);
6001 
6002 	/* Restore MSI and MSI-X configurations if they are present. */
6003 	if (dinfo->cfg.msi.msi_location != 0)
6004 		pci_resume_msi(dev);
6005 	if (dinfo->cfg.msix.msix_location != 0)
6006 		pci_resume_msix(dev);
6007 
6008 #ifdef PCI_IOV
6009 	if (dinfo->cfg.iov != NULL)
6010 		pci_iov_cfg_restore(dev, dinfo);
6011 #endif
6012 }
6013 
6014 static void
6015 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6016 {
6017 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6018 	struct pcicfg_pcie *cfg;
6019 	int version, pos;
6020 
6021 	cfg = &dinfo->cfg.pcie;
6022 	pos = cfg->pcie_location;
6023 
6024 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6025 
6026 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6027 
6028 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6029 
6030 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6031 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6032 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6033 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6034 
6035 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6036 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6037 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6038 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6039 
6040 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6041 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6042 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6043 
6044 	if (version > 1) {
6045 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6046 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6047 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6048 	}
6049 #undef RREG
6050 }
6051 
6052 static void
6053 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6054 {
6055 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6056 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6057 }
6058 
6059 void
6060 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6061 {
6062 	uint32_t cls;
6063 	int ps;
6064 
6065 	/*
6066 	 * Some drivers apparently write to these registers w/o updating our
6067 	 * cached copy.  No harm happens if we update the copy, so do so here
6068 	 * so we can restore them.  The COMMAND register is modified by the
6069 	 * bus w/o updating the cache.  This should represent the normally
6070 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6071 	 */
6072 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6073 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6074 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6075 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6076 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6077 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6078 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6079 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6080 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6081 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6082 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6083 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6084 	case PCIM_HDRTYPE_NORMAL:
6085 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6086 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6087 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6088 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6089 		break;
6090 	case PCIM_HDRTYPE_BRIDGE:
6091 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6092 		    PCIR_SECLAT_1, 1);
6093 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6094 		    PCIR_SUBBUS_1, 1);
6095 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6096 		    PCIR_SECBUS_1, 1);
6097 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6098 		    PCIR_PRIBUS_1, 1);
6099 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6100 		    PCIR_BRIDGECTL_1, 2);
6101 		break;
6102 	case PCIM_HDRTYPE_CARDBUS:
6103 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6104 		    PCIR_SECLAT_2, 1);
6105 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6106 		    PCIR_SUBBUS_2, 1);
6107 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6108 		    PCIR_SECBUS_2, 1);
6109 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6110 		    PCIR_PRIBUS_2, 1);
6111 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6112 		    PCIR_BRIDGECTL_2, 2);
6113 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6114 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6115 		break;
6116 	}
6117 
6118 	if (dinfo->cfg.pcie.pcie_location != 0)
6119 		pci_cfg_save_pcie(dev, dinfo);
6120 
6121 	if (dinfo->cfg.pcix.pcix_location != 0)
6122 		pci_cfg_save_pcix(dev, dinfo);
6123 
6124 #ifdef PCI_IOV
6125 	if (dinfo->cfg.iov != NULL)
6126 		pci_iov_cfg_save(dev, dinfo);
6127 #endif
6128 
6129 	/*
6130 	 * don't set the state for display devices, base peripherals and
6131 	 * memory devices since bad things happen when they are powered down.
6132 	 * We should (a) have drivers that can easily detach and (b) use
6133 	 * generic drivers for these devices so that some device actually
6134 	 * attaches.  We need to make sure that when we implement (a) we don't
6135 	 * power the device down on a reattach.
6136 	 */
6137 	cls = pci_get_class(dev);
6138 	if (!setstate)
6139 		return;
6140 	switch (pci_do_power_nodriver)
6141 	{
6142 		case 0:		/* NO powerdown at all */
6143 			return;
6144 		case 1:		/* Conservative about what to power down */
6145 			if (cls == PCIC_STORAGE)
6146 				return;
6147 			/*FALLTHROUGH*/
6148 		case 2:		/* Aggressive about what to power down */
6149 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6150 			    cls == PCIC_BASEPERIPH)
6151 				return;
6152 			/*FALLTHROUGH*/
6153 		case 3:		/* Power down everything */
6154 			break;
6155 	}
6156 	/*
6157 	 * PCI spec says we can only go into D3 state from D0 state.
6158 	 * Transition from D[12] into D0 before going to D3 state.
6159 	 */
6160 	ps = pci_get_powerstate(dev);
6161 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6162 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6163 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6164 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6165 }
6166 
6167 /* Wrapper APIs suitable for device driver use. */
6168 void
6169 pci_save_state(device_t dev)
6170 {
6171 	struct pci_devinfo *dinfo;
6172 
6173 	dinfo = device_get_ivars(dev);
6174 	pci_cfg_save(dev, dinfo, 0);
6175 }
6176 
6177 void
6178 pci_restore_state(device_t dev)
6179 {
6180 	struct pci_devinfo *dinfo;
6181 
6182 	dinfo = device_get_ivars(dev);
6183 	pci_cfg_restore(dev, dinfo);
6184 }
6185 
6186 static int
6187 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6188     uintptr_t *id)
6189 {
6190 
6191 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6192 }
6193 
6194 /* Find the upstream port of a given PCI device in a root complex. */
6195 device_t
6196 pci_find_pcie_root_port(device_t dev)
6197 {
6198 	struct pci_devinfo *dinfo;
6199 	devclass_t pci_class;
6200 	device_t pcib, bus;
6201 
6202 	pci_class = devclass_find("pci");
6203 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6204 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6205 
6206 	/*
6207 	 * Walk the bridge hierarchy until we find a PCI-e root
6208 	 * port or a non-PCI device.
6209 	 */
6210 	for (;;) {
6211 		bus = device_get_parent(dev);
6212 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6213 		    device_get_nameunit(dev)));
6214 
6215 		pcib = device_get_parent(bus);
6216 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6217 		    device_get_nameunit(bus)));
6218 
6219 		/*
6220 		 * pcib's parent must be a PCI bus for this to be a
6221 		 * PCI-PCI bridge.
6222 		 */
6223 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6224 			return (NULL);
6225 
6226 		dinfo = device_get_ivars(pcib);
6227 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6228 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6229 			return (pcib);
6230 
6231 		dev = pcib;
6232 	}
6233 }
6234 
6235 /*
6236  * Wait for pending transactions to complete on a PCI-express function.
6237  *
6238  * The maximum delay is specified in milliseconds in max_delay.  Note
6239  * that this function may sleep.
6240  *
6241  * Returns true if the function is idle and false if the timeout is
6242  * exceeded.  If dev is not a PCI-express function, this returns true.
6243  */
6244 bool
6245 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6246 {
6247 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6248 	uint16_t sta;
6249 	int cap;
6250 
6251 	cap = dinfo->cfg.pcie.pcie_location;
6252 	if (cap == 0)
6253 		return (true);
6254 
6255 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6256 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6257 		if (max_delay == 0)
6258 			return (false);
6259 
6260 		/* Poll once every 100 milliseconds up to the timeout. */
6261 		if (max_delay > 100) {
6262 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6263 			max_delay -= 100;
6264 		} else {
6265 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6266 			    C_HARDCLOCK);
6267 			max_delay = 0;
6268 		}
6269 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6270 	}
6271 
6272 	return (true);
6273 }
6274 
6275 /*
6276  * Determine the maximum Completion Timeout in microseconds.
6277  *
6278  * For non-PCI-express functions this returns 0.
6279  */
6280 int
6281 pcie_get_max_completion_timeout(device_t dev)
6282 {
6283 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6284 	int cap;
6285 
6286 	cap = dinfo->cfg.pcie.pcie_location;
6287 	if (cap == 0)
6288 		return (0);
6289 
6290 	/*
6291 	 * Functions using the 1.x spec use the default timeout range of
6292 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6293 	 * support programmable timeouts also use this range.
6294 	 */
6295 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6296 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6297 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6298 		return (50 * 1000);
6299 
6300 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6301 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6302 	case PCIEM_CTL2_COMP_TIMO_100US:
6303 		return (100);
6304 	case PCIEM_CTL2_COMP_TIMO_10MS:
6305 		return (10 * 1000);
6306 	case PCIEM_CTL2_COMP_TIMO_55MS:
6307 		return (55 * 1000);
6308 	case PCIEM_CTL2_COMP_TIMO_210MS:
6309 		return (210 * 1000);
6310 	case PCIEM_CTL2_COMP_TIMO_900MS:
6311 		return (900 * 1000);
6312 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6313 		return (3500 * 1000);
6314 	case PCIEM_CTL2_COMP_TIMO_13S:
6315 		return (13 * 1000 * 1000);
6316 	case PCIEM_CTL2_COMP_TIMO_64S:
6317 		return (64 * 1000 * 1000);
6318 	default:
6319 		return (50 * 1000);
6320 	}
6321 }
6322 
6323 /*
6324  * Perform a Function Level Reset (FLR) on a device.
6325  *
6326  * This function first waits for any pending transactions to complete
6327  * within the timeout specified by max_delay.  If transactions are
6328  * still pending, the function will return false without attempting a
6329  * reset.
6330  *
6331  * If dev is not a PCI-express function or does not support FLR, this
6332  * function returns false.
6333  *
6334  * Note that no registers are saved or restored.  The caller is
6335  * responsible for saving and restoring any registers including
6336  * PCI-standard registers via pci_save_state() and
6337  * pci_restore_state().
6338  */
6339 bool
6340 pcie_flr(device_t dev, u_int max_delay, bool force)
6341 {
6342 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6343 	uint16_t cmd, ctl;
6344 	int compl_delay;
6345 	int cap;
6346 
6347 	cap = dinfo->cfg.pcie.pcie_location;
6348 	if (cap == 0)
6349 		return (false);
6350 
6351 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6352 		return (false);
6353 
6354 	/*
6355 	 * Disable busmastering to prevent generation of new
6356 	 * transactions while waiting for the device to go idle.  If
6357 	 * the idle timeout fails, the command register is restored
6358 	 * which will re-enable busmastering.
6359 	 */
6360 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6361 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6362 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6363 		if (!force) {
6364 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6365 			return (false);
6366 		}
6367 		pci_printf(&dinfo->cfg,
6368 		    "Resetting with transactions pending after %d ms\n",
6369 		    max_delay);
6370 
6371 		/*
6372 		 * Extend the post-FLR delay to cover the maximum
6373 		 * Completion Timeout delay of anything in flight
6374 		 * during the FLR delay.  Enforce a minimum delay of
6375 		 * at least 10ms.
6376 		 */
6377 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6378 		if (compl_delay < 10)
6379 			compl_delay = 10;
6380 	} else
6381 		compl_delay = 0;
6382 
6383 	/* Initiate the reset. */
6384 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6385 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6386 	    PCIEM_CTL_INITIATE_FLR, 2);
6387 
6388 	/* Wait for 100ms. */
6389 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6390 
6391 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6392 	    PCIEM_STA_TRANSACTION_PND)
6393 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6394 	return (true);
6395 }
6396 
6397 /*
6398  * Attempt a power-management reset by cycling the device in/out of D3
6399  * state.  PCI spec says we can only go into D3 state from D0 state.
6400  * Transition from D[12] into D0 before going to D3 state.
6401  */
6402 int
6403 pci_power_reset(device_t dev)
6404 {
6405 	int ps;
6406 
6407 	ps = pci_get_powerstate(dev);
6408 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6409 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6410 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6411 	pci_set_powerstate(dev, ps);
6412 	return (0);
6413 }
6414 
6415 /*
6416  * Try link drop and retrain of the downstream port of upstream
6417  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6418  * cause Conventional Hot reset of the device in the slot.
6419  * Alternative, for PCIe, could be the secondary bus reset initiatied
6420  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6421  */
6422 int
6423 pcie_link_reset(device_t port, int pcie_location)
6424 {
6425 	uint16_t v;
6426 
6427 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6428 	v |= PCIEM_LINK_CTL_LINK_DIS;
6429 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6430 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6431 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6432 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6433 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6434 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6435 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6436 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6437 }
6438 
6439 static int
6440 pci_reset_post(device_t dev, device_t child)
6441 {
6442 
6443 	if (dev == device_get_parent(child))
6444 		pci_restore_state(child);
6445 	return (0);
6446 }
6447 
6448 static int
6449 pci_reset_prepare(device_t dev, device_t child)
6450 {
6451 
6452 	if (dev == device_get_parent(child))
6453 		pci_save_state(child);
6454 	return (0);
6455 }
6456 
6457 static int
6458 pci_reset_child(device_t dev, device_t child, int flags)
6459 {
6460 	int error;
6461 
6462 	if (dev == NULL || device_get_parent(child) != dev)
6463 		return (0);
6464 	if ((flags & DEVF_RESET_DETACH) != 0) {
6465 		error = device_get_state(child) == DS_ATTACHED ?
6466 		    device_detach(child) : 0;
6467 	} else {
6468 		error = BUS_SUSPEND_CHILD(dev, child);
6469 	}
6470 	if (error == 0) {
6471 		if (!pcie_flr(child, 1000, false)) {
6472 			error = BUS_RESET_PREPARE(dev, child);
6473 			if (error == 0)
6474 				pci_power_reset(child);
6475 			BUS_RESET_POST(dev, child);
6476 		}
6477 		if ((flags & DEVF_RESET_DETACH) != 0)
6478 			device_probe_and_attach(child);
6479 		else
6480 			BUS_RESUME_CHILD(dev, child);
6481 	}
6482 	return (error);
6483 }
6484 
6485 const struct pci_device_table *
6486 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6487 {
6488 	bool match;
6489 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6490 
6491 	vendor = pci_get_vendor(child);
6492 	device = pci_get_device(child);
6493 	subvendor = pci_get_subvendor(child);
6494 	subdevice = pci_get_subdevice(child);
6495 	class = pci_get_class(child);
6496 	subclass = pci_get_subclass(child);
6497 	revid = pci_get_revid(child);
6498 	while (nelt-- > 0) {
6499 		match = true;
6500 		if (id->match_flag_vendor)
6501 			match &= vendor == id->vendor;
6502 		if (id->match_flag_device)
6503 			match &= device == id->device;
6504 		if (id->match_flag_subvendor)
6505 			match &= subvendor == id->subvendor;
6506 		if (id->match_flag_subdevice)
6507 			match &= subdevice == id->subdevice;
6508 		if (id->match_flag_class)
6509 			match &= class == id->class_id;
6510 		if (id->match_flag_subclass)
6511 			match &= subclass == id->subclass;
6512 		if (id->match_flag_revid)
6513 			match &= revid == id->revid;
6514 		if (match)
6515 			return (id);
6516 		id++;
6517 	}
6518 	return (NULL);
6519 }
6520 
6521 static void
6522 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6523 {
6524 	const char *dev_name;
6525 	device_t dev;
6526 
6527 	dev = dinfo->cfg.dev;
6528 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6529 	    dinfo->cfg.slot, dinfo->cfg.func);
6530 	dev_name = device_get_name(dev);
6531 	if (dev_name != NULL)
6532 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6533 }
6534 
6535 void
6536 pci_print_faulted_dev(void)
6537 {
6538 	struct pci_devinfo *dinfo;
6539 	device_t dev;
6540 	int aer, i;
6541 	uint32_t r1, r2;
6542 	uint16_t status;
6543 
6544 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6545 		dev = dinfo->cfg.dev;
6546 		status = pci_read_config(dev, PCIR_STATUS, 2);
6547 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6548 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6549 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6550 		if (status != 0) {
6551 			pci_print_faulted_dev_name(dinfo);
6552 			printf(" error 0x%04x\n", status);
6553 		}
6554 		if (dinfo->cfg.pcie.pcie_location != 0) {
6555 			status = pci_read_config(dev,
6556 			    dinfo->cfg.pcie.pcie_location +
6557 			    PCIER_DEVICE_STA, 2);
6558 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6559 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6560 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6561 				pci_print_faulted_dev_name(dinfo);
6562 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6563 				    pci_read_config(dev,
6564 				    dinfo->cfg.pcie.pcie_location +
6565 				    PCIER_DEVICE_CTL, 2),
6566 				    status);
6567 			}
6568 		}
6569 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6570 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6571 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6572 			if (r1 != 0 || r2 != 0) {
6573 				pci_print_faulted_dev_name(dinfo);
6574 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6575 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6576 				    r1, pci_read_config(dev, aer +
6577 				    PCIR_AER_UC_MASK, 4),
6578 				    pci_read_config(dev, aer +
6579 				    PCIR_AER_UC_SEVERITY, 4),
6580 				    r2, pci_read_config(dev, aer +
6581 				    PCIR_AER_COR_MASK, 4),
6582 				    pci_read_config(dev, aer +
6583 				    PCIR_AER_CAP_CONTROL, 4));
6584 				for (i = 0; i < 4; i++) {
6585 					r1 = pci_read_config(dev, aer +
6586 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6587 					printf("    HL%d: 0x%08x\n", i, r1);
6588 				}
6589 			}
6590 		}
6591 	}
6592 }
6593 
6594 #ifdef DDB
6595 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6596 {
6597 
6598 	pci_print_faulted_dev();
6599 }
6600 
6601 static void
6602 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6603 {
6604 	device_t dev;
6605 	int aer;
6606 	uint32_t r;
6607 
6608 	dev = dinfo->cfg.dev;
6609 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6610 	    PCIER_DEVICE_STA, 2);
6611 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6612 	    PCIER_DEVICE_STA, r, 2);
6613 
6614 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6615 		return;
6616 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6617 	if (r != 0)
6618 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6619 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6620 	if (r != 0)
6621 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6622 }
6623 
6624 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6625 {
6626 	struct pci_devinfo *dinfo;
6627 	device_t dev;
6628 	uint16_t status, status1;
6629 
6630 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6631 		dev = dinfo->cfg.dev;
6632 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6633 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6634 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6635 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6636 		if (status1 != 0) {
6637 			status &= ~status1;
6638 			pci_write_config(dev, PCIR_STATUS, status, 2);
6639 		}
6640 		if (dinfo->cfg.pcie.pcie_location != 0)
6641 			db_clear_pcie_errors(dinfo);
6642 	}
6643 }
6644 #endif
6645