xref: /freebsd/sys/dev/pci/pci.c (revision 5ab1c5846ff41be24b1f6beb0317bf8258cd4409)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_bus.h"
36 
37 #include <sys/param.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 
55 #include <sys/bus.h>
56 #include <machine/bus.h>
57 #include <sys/rman.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
60 
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
63 #endif
64 
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
69 
70 #ifdef PCI_IOV
71 #include <sys/nv.h>
72 #include <dev/pci/pci_iov_private.h>
73 #endif
74 
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
79 
80 #include "pcib_if.h"
81 #include "pci_if.h"
82 
83 #define	PCIR_IS_BIOS(cfg, reg)						\
84 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
85 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 
87 static int		pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t	pci_mapbase(uint64_t mapreg);
89 static const char	*pci_maptype(uint64_t mapreg);
90 static int		pci_maprange(uint64_t mapreg);
91 static pci_addr_t	pci_rombase(uint64_t mapreg);
92 static int		pci_romsize(uint64_t testval);
93 static void		pci_fixancient(pcicfgregs *cfg);
94 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 
96 static int		pci_porten(device_t dev);
97 static int		pci_memen(device_t dev);
98 static void		pci_assign_interrupt(device_t bus, device_t dev,
99 			    int force_route);
100 static int		pci_add_map(device_t bus, device_t dev, int reg,
101 			    struct resource_list *rl, int force, int prefetch);
102 static int		pci_probe(device_t dev);
103 static void		pci_load_vendor_data(void);
104 static int		pci_describe_parse_line(char **ptr, int *vendor,
105 			    int *device, char **desc);
106 static char		*pci_describe_device(device_t dev);
107 static int		pci_modevent(module_t mod, int what, void *arg);
108 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
109 			    pcicfgregs *cfg);
110 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
111 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
112 			    int reg, uint32_t *data);
113 #if 0
114 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
115 			    int reg, uint32_t data);
116 #endif
117 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
118 static void		pci_mask_msix(device_t dev, u_int index);
119 static void		pci_unmask_msix(device_t dev, u_int index);
120 static int		pci_msi_blacklisted(void);
121 static int		pci_msix_blacklisted(void);
122 static void		pci_resume_msi(device_t dev);
123 static void		pci_resume_msix(device_t dev);
124 static int		pci_remap_intr_method(device_t bus, device_t dev,
125 			    u_int irq);
126 static void		pci_hint_device_unit(device_t acdev, device_t child,
127 			    const char *name, int *unitp);
128 static int		pci_reset_post(device_t dev, device_t child);
129 static int		pci_reset_prepare(device_t dev, device_t child);
130 static int		pci_reset_child(device_t dev, device_t child,
131 			    int flags);
132 
133 static int		pci_get_id_method(device_t dev, device_t child,
134 			    enum pci_id_type type, uintptr_t *rid);
135 
136 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
137     int b, int s, int f, uint16_t vid, uint16_t did);
138 
139 static device_method_t pci_methods[] = {
140 	/* Device interface */
141 	DEVMETHOD(device_probe,		pci_probe),
142 	DEVMETHOD(device_attach,	pci_attach),
143 	DEVMETHOD(device_detach,	pci_detach),
144 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
145 	DEVMETHOD(device_suspend,	bus_generic_suspend),
146 	DEVMETHOD(device_resume,	pci_resume),
147 
148 	/* Bus interface */
149 	DEVMETHOD(bus_print_child,	pci_print_child),
150 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
151 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
152 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
153 	DEVMETHOD(bus_driver_added,	pci_driver_added),
154 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
155 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
156 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
157 	DEVMETHOD(bus_reset_post,	pci_reset_post),
158 	DEVMETHOD(bus_reset_child,	pci_reset_child),
159 
160 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
161 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
162 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
163 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
164 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
165 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
166 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
167 	DEVMETHOD(bus_release_resource,	pci_release_resource),
168 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
169 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
170 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
171 	DEVMETHOD(bus_child_detached,	pci_child_detached),
172 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
173 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
174 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
175 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
176 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
177 	DEVMETHOD(bus_resume_child,	pci_resume_child),
178 	DEVMETHOD(bus_rescan,		pci_rescan_method),
179 
180 	/* PCI interface */
181 	DEVMETHOD(pci_read_config,	pci_read_config_method),
182 	DEVMETHOD(pci_write_config,	pci_write_config_method),
183 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
184 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
185 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
186 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
187 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
188 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
189 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
190 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
191 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
192 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
193 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
194 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
195 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
196 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
197 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
198 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
199 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
200 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
201 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
202 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
203 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
204 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
205 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
206 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
207 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
208 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
209 	DEVMETHOD(pci_get_id,		pci_get_id_method),
210 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
211 	DEVMETHOD(pci_child_added,	pci_child_added_method),
212 #ifdef PCI_IOV
213 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
214 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
215 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
216 #endif
217 
218 	DEVMETHOD_END
219 };
220 
221 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
222 
223 static devclass_t pci_devclass;
224 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
225     BUS_PASS_BUS);
226 MODULE_VERSION(pci, 1);
227 
228 static char	*pci_vendordata;
229 static size_t	pci_vendordata_size;
230 
231 struct pci_quirk {
232 	uint32_t devid;	/* Vendor/device of the card */
233 	int	type;
234 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
235 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
236 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
237 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
238 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
239 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
240 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
241 	int	arg1;
242 	int	arg2;
243 };
244 
245 static const struct pci_quirk pci_quirks[] = {
246 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
247 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
248 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
249 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
250 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 
252 	/*
253 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
254 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
255 	 */
256 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
257 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
258 
259 	/*
260 	 * MSI doesn't work on earlier Intel chipsets including
261 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
262 	 */
263 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
264 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
265 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 
271 	/*
272 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
273 	 * bridge.
274 	 */
275 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
276 
277 	/*
278 	 * MSI-X allocation doesn't work properly for devices passed through
279 	 * by VMware up to at least ESXi 5.1.
280 	 */
281 	{ 0x079015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCI/PCI-X */
282 	{ 0x07a015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCIe */
283 
284 	/*
285 	 * Some virtualization environments emulate an older chipset
286 	 * but support MSI just fine.  QEMU uses the Intel 82440.
287 	 */
288 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
289 
290 	/*
291 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
292 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
293 	 * It prevents us from attaching hpet(4) when the bit is unset.
294 	 * Note this quirk only affects SB600 revision A13 and earlier.
295 	 * For SB600 A21 and later, firmware must set the bit to hide it.
296 	 * For SB700 and later, it is unused and hardcoded to zero.
297 	 */
298 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
299 
300 	/*
301 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
302 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
303 	 * of the command register is set.
304 	 */
305 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
306 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
307 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
309 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
310 
311 	/*
312 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
313 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
314 	 */
315 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
316 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
317 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
318 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
319 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
320 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
321 
322 	/*
323 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
324 	 * expected place.
325 	 */
326 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
327 
328 	{ 0 }
329 };
330 
331 /* map register information */
332 #define	PCI_MAPMEM	0x01	/* memory map */
333 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
334 #define	PCI_MAPPORT	0x04	/* port map */
335 
336 struct devlist pci_devq;
337 uint32_t pci_generation;
338 uint32_t pci_numdevs = 0;
339 static int pcie_chipset, pcix_chipset;
340 
341 /* sysctl vars */
342 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
343 
344 static int pci_enable_io_modes = 1;
345 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
346     &pci_enable_io_modes, 1,
347     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
348     " enable these bits correctly.  We'd like to do this all the time, but"
349     " there are some peripherals that this causes problems with.");
350 
351 static int pci_do_realloc_bars = 1;
352 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
353     &pci_do_realloc_bars, 0,
354     "Attempt to allocate a new range for any BARs whose original "
355     "firmware-assigned ranges fail to allocate during the initial device scan.");
356 
357 static int pci_do_power_nodriver = 0;
358 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
359     &pci_do_power_nodriver, 0,
360     "Place a function into D3 state when no driver attaches to it.  0 means"
361     " disable.  1 means conservatively place devices into D3 state.  2 means"
362     " aggressively place devices into D3 state.  3 means put absolutely"
363     " everything in D3 state.");
364 
365 int pci_do_power_resume = 1;
366 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
367     &pci_do_power_resume, 1,
368   "Transition from D3 -> D0 on resume.");
369 
370 int pci_do_power_suspend = 1;
371 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
372     &pci_do_power_suspend, 1,
373   "Transition from D0 -> D3 on suspend.");
374 
375 static int pci_do_msi = 1;
376 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
377     "Enable support for MSI interrupts");
378 
379 static int pci_do_msix = 1;
380 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
381     "Enable support for MSI-X interrupts");
382 
383 static int pci_msix_rewrite_table = 0;
384 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
385     &pci_msix_rewrite_table, 0,
386     "Rewrite entire MSI-X table when updating MSI-X entries");
387 
388 static int pci_honor_msi_blacklist = 1;
389 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
390     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
391 
392 #if defined(__i386__) || defined(__amd64__)
393 static int pci_usb_takeover = 1;
394 #else
395 static int pci_usb_takeover = 0;
396 #endif
397 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
398     &pci_usb_takeover, 1,
399     "Enable early takeover of USB controllers. Disable this if you depend on"
400     " BIOS emulation of USB devices, that is you use USB devices (like"
401     " keyboard or mouse) but do not load USB drivers");
402 
403 static int pci_clear_bars;
404 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
405     "Ignore firmware-assigned resources for BARs.");
406 
407 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
408 static int pci_clear_buses;
409 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
410     "Ignore firmware-assigned bus numbers.");
411 #endif
412 
413 static int pci_enable_ari = 1;
414 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
415     0, "Enable support for PCIe Alternative RID Interpretation");
416 
417 static int pci_clear_aer_on_attach = 0;
418 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
419     &pci_clear_aer_on_attach, 0,
420     "Clear port and device AER state on driver attach");
421 
422 static int
423 pci_has_quirk(uint32_t devid, int quirk)
424 {
425 	const struct pci_quirk *q;
426 
427 	for (q = &pci_quirks[0]; q->devid; q++) {
428 		if (q->devid == devid && q->type == quirk)
429 			return (1);
430 	}
431 	return (0);
432 }
433 
434 /* Find a device_t by bus/slot/function in domain 0 */
435 
436 device_t
437 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
438 {
439 
440 	return (pci_find_dbsf(0, bus, slot, func));
441 }
442 
443 /* Find a device_t by domain/bus/slot/function */
444 
445 device_t
446 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
447 {
448 	struct pci_devinfo *dinfo = NULL;
449 
450 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
451 		if ((dinfo->cfg.domain == domain) &&
452 		    (dinfo->cfg.bus == bus) &&
453 		    (dinfo->cfg.slot == slot) &&
454 		    (dinfo->cfg.func == func)) {
455 			break;
456 		}
457 	}
458 
459 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
460 }
461 
462 /* Find a device_t by vendor/device ID */
463 
464 device_t
465 pci_find_device(uint16_t vendor, uint16_t device)
466 {
467 	struct pci_devinfo *dinfo;
468 
469 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
470 		if ((dinfo->cfg.vendor == vendor) &&
471 		    (dinfo->cfg.device == device)) {
472 			return (dinfo->cfg.dev);
473 		}
474 	}
475 
476 	return (NULL);
477 }
478 
479 device_t
480 pci_find_class(uint8_t class, uint8_t subclass)
481 {
482 	struct pci_devinfo *dinfo;
483 
484 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
485 		if (dinfo->cfg.baseclass == class &&
486 		    dinfo->cfg.subclass == subclass) {
487 			return (dinfo->cfg.dev);
488 		}
489 	}
490 
491 	return (NULL);
492 }
493 
494 static int
495 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
496 {
497 	va_list ap;
498 	int retval;
499 
500 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
501 	    cfg->func);
502 	va_start(ap, fmt);
503 	retval += vprintf(fmt, ap);
504 	va_end(ap);
505 	return (retval);
506 }
507 
508 /* return base address of memory or port map */
509 
510 static pci_addr_t
511 pci_mapbase(uint64_t mapreg)
512 {
513 
514 	if (PCI_BAR_MEM(mapreg))
515 		return (mapreg & PCIM_BAR_MEM_BASE);
516 	else
517 		return (mapreg & PCIM_BAR_IO_BASE);
518 }
519 
520 /* return map type of memory or port map */
521 
522 static const char *
523 pci_maptype(uint64_t mapreg)
524 {
525 
526 	if (PCI_BAR_IO(mapreg))
527 		return ("I/O Port");
528 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
529 		return ("Prefetchable Memory");
530 	return ("Memory");
531 }
532 
533 /* return log2 of map size decoded for memory or port map */
534 
535 int
536 pci_mapsize(uint64_t testval)
537 {
538 	int ln2size;
539 
540 	testval = pci_mapbase(testval);
541 	ln2size = 0;
542 	if (testval != 0) {
543 		while ((testval & 1) == 0)
544 		{
545 			ln2size++;
546 			testval >>= 1;
547 		}
548 	}
549 	return (ln2size);
550 }
551 
552 /* return base address of device ROM */
553 
554 static pci_addr_t
555 pci_rombase(uint64_t mapreg)
556 {
557 
558 	return (mapreg & PCIM_BIOS_ADDR_MASK);
559 }
560 
561 /* return log2 of map size decided for device ROM */
562 
563 static int
564 pci_romsize(uint64_t testval)
565 {
566 	int ln2size;
567 
568 	testval = pci_rombase(testval);
569 	ln2size = 0;
570 	if (testval != 0) {
571 		while ((testval & 1) == 0)
572 		{
573 			ln2size++;
574 			testval >>= 1;
575 		}
576 	}
577 	return (ln2size);
578 }
579 
580 /* return log2 of address range supported by map register */
581 
582 static int
583 pci_maprange(uint64_t mapreg)
584 {
585 	int ln2range = 0;
586 
587 	if (PCI_BAR_IO(mapreg))
588 		ln2range = 32;
589 	else
590 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
591 		case PCIM_BAR_MEM_32:
592 			ln2range = 32;
593 			break;
594 		case PCIM_BAR_MEM_1MB:
595 			ln2range = 20;
596 			break;
597 		case PCIM_BAR_MEM_64:
598 			ln2range = 64;
599 			break;
600 		}
601 	return (ln2range);
602 }
603 
604 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
605 
606 static void
607 pci_fixancient(pcicfgregs *cfg)
608 {
609 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
610 		return;
611 
612 	/* PCI to PCI bridges use header type 1 */
613 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
614 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
615 }
616 
617 /* extract header type specific config data */
618 
619 static void
620 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
621 {
622 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
623 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
624 	case PCIM_HDRTYPE_NORMAL:
625 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
626 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
627 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
628 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
629 		cfg->nummaps	    = PCI_MAXMAPS_0;
630 		break;
631 	case PCIM_HDRTYPE_BRIDGE:
632 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
633 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
634 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
635 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
636 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
637 		cfg->nummaps	    = PCI_MAXMAPS_1;
638 		break;
639 	case PCIM_HDRTYPE_CARDBUS:
640 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
641 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
642 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
643 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
644 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
645 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
646 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
647 		cfg->nummaps	    = PCI_MAXMAPS_2;
648 		break;
649 	}
650 #undef REG
651 }
652 
653 /* read configuration header into pcicfgregs structure */
654 struct pci_devinfo *
655 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
656 {
657 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
658 	uint16_t vid, did;
659 
660 	vid = REG(PCIR_VENDOR, 2);
661 	did = REG(PCIR_DEVICE, 2);
662 	if (vid != 0xffff)
663 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
664 
665 	return (NULL);
666 }
667 
668 struct pci_devinfo *
669 pci_alloc_devinfo_method(device_t dev)
670 {
671 
672 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
673 	    M_WAITOK | M_ZERO));
674 }
675 
676 static struct pci_devinfo *
677 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
678     uint16_t vid, uint16_t did)
679 {
680 	struct pci_devinfo *devlist_entry;
681 	pcicfgregs *cfg;
682 
683 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
684 
685 	cfg = &devlist_entry->cfg;
686 
687 	cfg->domain		= d;
688 	cfg->bus		= b;
689 	cfg->slot		= s;
690 	cfg->func		= f;
691 	cfg->vendor		= vid;
692 	cfg->device		= did;
693 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
694 	cfg->statreg		= REG(PCIR_STATUS, 2);
695 	cfg->baseclass		= REG(PCIR_CLASS, 1);
696 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
697 	cfg->progif		= REG(PCIR_PROGIF, 1);
698 	cfg->revid		= REG(PCIR_REVID, 1);
699 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
700 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
701 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
702 	cfg->intpin		= REG(PCIR_INTPIN, 1);
703 	cfg->intline		= REG(PCIR_INTLINE, 1);
704 
705 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
706 	cfg->hdrtype		&= ~PCIM_MFDEV;
707 	STAILQ_INIT(&cfg->maps);
708 
709 	cfg->iov		= NULL;
710 
711 	pci_fixancient(cfg);
712 	pci_hdrtypedata(pcib, b, s, f, cfg);
713 
714 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
715 		pci_read_cap(pcib, cfg);
716 
717 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
718 
719 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
720 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
721 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
722 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
723 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
724 
725 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
726 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
727 	devlist_entry->conf.pc_vendor = cfg->vendor;
728 	devlist_entry->conf.pc_device = cfg->device;
729 
730 	devlist_entry->conf.pc_class = cfg->baseclass;
731 	devlist_entry->conf.pc_subclass = cfg->subclass;
732 	devlist_entry->conf.pc_progif = cfg->progif;
733 	devlist_entry->conf.pc_revid = cfg->revid;
734 
735 	pci_numdevs++;
736 	pci_generation++;
737 
738 	return (devlist_entry);
739 }
740 #undef REG
741 
742 static void
743 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
744 {
745 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
746     cfg->ea.ea_location + (n), w)
747 	int num_ent;
748 	int ptr;
749 	int a, b;
750 	uint32_t val;
751 	int ent_size;
752 	uint32_t dw[4];
753 	uint64_t base, max_offset;
754 	struct pci_ea_entry *eae;
755 
756 	if (cfg->ea.ea_location == 0)
757 		return;
758 
759 	STAILQ_INIT(&cfg->ea.ea_entries);
760 
761 	/* Determine the number of entries */
762 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
763 	num_ent &= PCIM_EA_NUM_ENT_MASK;
764 
765 	/* Find the first entry to care of */
766 	ptr = PCIR_EA_FIRST_ENT;
767 
768 	/* Skip DWORD 2 for type 1 functions */
769 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
770 		ptr += 4;
771 
772 	for (a = 0; a < num_ent; a++) {
773 
774 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
775 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
776 
777 		/* Read a number of dwords in the entry */
778 		val = REG(ptr, 4);
779 		ptr += 4;
780 		ent_size = (val & PCIM_EA_ES);
781 
782 		for (b = 0; b < ent_size; b++) {
783 			dw[b] = REG(ptr, 4);
784 			ptr += 4;
785 		}
786 
787 		eae->eae_flags = val;
788 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
789 
790 		base = dw[0] & PCIM_EA_FIELD_MASK;
791 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
792 		b = 2;
793 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
794 			base |= (uint64_t)dw[b] << 32UL;
795 			b++;
796 		}
797 		if (((dw[1] & PCIM_EA_IS_64) != 0)
798 		    && (b < ent_size)) {
799 			max_offset |= (uint64_t)dw[b] << 32UL;
800 			b++;
801 		}
802 
803 		eae->eae_base = base;
804 		eae->eae_max_offset = max_offset;
805 
806 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
807 
808 		if (bootverbose) {
809 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
810 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
811 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
812 		}
813 	}
814 }
815 #undef REG
816 
817 static void
818 pci_read_cap(device_t pcib, pcicfgregs *cfg)
819 {
820 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
821 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
822 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
823 	uint64_t addr;
824 #endif
825 	uint32_t val;
826 	int	ptr, nextptr, ptrptr;
827 
828 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
829 	case PCIM_HDRTYPE_NORMAL:
830 	case PCIM_HDRTYPE_BRIDGE:
831 		ptrptr = PCIR_CAP_PTR;
832 		break;
833 	case PCIM_HDRTYPE_CARDBUS:
834 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
835 		break;
836 	default:
837 		return;		/* no extended capabilities support */
838 	}
839 	nextptr = REG(ptrptr, 1);	/* sanity check? */
840 
841 	/*
842 	 * Read capability entries.
843 	 */
844 	while (nextptr != 0) {
845 		/* Sanity check */
846 		if (nextptr > 255) {
847 			printf("illegal PCI extended capability offset %d\n",
848 			    nextptr);
849 			return;
850 		}
851 		/* Find the next entry */
852 		ptr = nextptr;
853 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
854 
855 		/* Process this entry */
856 		switch (REG(ptr + PCICAP_ID, 1)) {
857 		case PCIY_PMG:		/* PCI power management */
858 			if (cfg->pp.pp_cap == 0) {
859 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
860 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
861 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
862 				if ((nextptr - ptr) > PCIR_POWER_DATA)
863 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
864 			}
865 			break;
866 		case PCIY_HT:		/* HyperTransport */
867 			/* Determine HT-specific capability type. */
868 			val = REG(ptr + PCIR_HT_COMMAND, 2);
869 
870 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
871 				cfg->ht.ht_slave = ptr;
872 
873 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
874 			switch (val & PCIM_HTCMD_CAP_MASK) {
875 			case PCIM_HTCAP_MSI_MAPPING:
876 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
877 					/* Sanity check the mapping window. */
878 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
879 					    4);
880 					addr <<= 32;
881 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
882 					    4);
883 					if (addr != MSI_INTEL_ADDR_BASE)
884 						device_printf(pcib,
885 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
886 						    cfg->domain, cfg->bus,
887 						    cfg->slot, cfg->func,
888 						    (long long)addr);
889 				} else
890 					addr = MSI_INTEL_ADDR_BASE;
891 
892 				cfg->ht.ht_msimap = ptr;
893 				cfg->ht.ht_msictrl = val;
894 				cfg->ht.ht_msiaddr = addr;
895 				break;
896 			}
897 #endif
898 			break;
899 		case PCIY_MSI:		/* PCI MSI */
900 			cfg->msi.msi_location = ptr;
901 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
902 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
903 						     PCIM_MSICTRL_MMC_MASK)>>1);
904 			break;
905 		case PCIY_MSIX:		/* PCI MSI-X */
906 			cfg->msix.msix_location = ptr;
907 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
908 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
909 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
910 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
911 			cfg->msix.msix_table_bar = PCIR_BAR(val &
912 			    PCIM_MSIX_BIR_MASK);
913 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
914 			val = REG(ptr + PCIR_MSIX_PBA, 4);
915 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
916 			    PCIM_MSIX_BIR_MASK);
917 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
918 			break;
919 		case PCIY_VPD:		/* PCI Vital Product Data */
920 			cfg->vpd.vpd_reg = ptr;
921 			break;
922 		case PCIY_SUBVENDOR:
923 			/* Should always be true. */
924 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
925 			    PCIM_HDRTYPE_BRIDGE) {
926 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
927 				cfg->subvendor = val & 0xffff;
928 				cfg->subdevice = val >> 16;
929 			}
930 			break;
931 		case PCIY_PCIX:		/* PCI-X */
932 			/*
933 			 * Assume we have a PCI-X chipset if we have
934 			 * at least one PCI-PCI bridge with a PCI-X
935 			 * capability.  Note that some systems with
936 			 * PCI-express or HT chipsets might match on
937 			 * this check as well.
938 			 */
939 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
940 			    PCIM_HDRTYPE_BRIDGE)
941 				pcix_chipset = 1;
942 			cfg->pcix.pcix_location = ptr;
943 			break;
944 		case PCIY_EXPRESS:	/* PCI-express */
945 			/*
946 			 * Assume we have a PCI-express chipset if we have
947 			 * at least one PCI-express device.
948 			 */
949 			pcie_chipset = 1;
950 			cfg->pcie.pcie_location = ptr;
951 			val = REG(ptr + PCIER_FLAGS, 2);
952 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
953 			break;
954 		case PCIY_EA:		/* Enhanced Allocation */
955 			cfg->ea.ea_location = ptr;
956 			pci_ea_fill_info(pcib, cfg);
957 			break;
958 		default:
959 			break;
960 		}
961 	}
962 
963 #if defined(__powerpc__)
964 	/*
965 	 * Enable the MSI mapping window for all HyperTransport
966 	 * slaves.  PCI-PCI bridges have their windows enabled via
967 	 * PCIB_MAP_MSI().
968 	 */
969 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
970 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
971 		device_printf(pcib,
972 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
973 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
974 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
975 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
976 		     2);
977 	}
978 #endif
979 /* REG and WREG use carry through to next functions */
980 }
981 
982 /*
983  * PCI Vital Product Data
984  */
985 
986 #define	PCI_VPD_TIMEOUT		1000000
987 
988 static int
989 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
990 {
991 	int count = PCI_VPD_TIMEOUT;
992 
993 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
994 
995 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
996 
997 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
998 		if (--count < 0)
999 			return (ENXIO);
1000 		DELAY(1);	/* limit looping */
1001 	}
1002 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1003 
1004 	return (0);
1005 }
1006 
1007 #if 0
1008 static int
1009 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1010 {
1011 	int count = PCI_VPD_TIMEOUT;
1012 
1013 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1014 
1015 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1016 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1017 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1018 		if (--count < 0)
1019 			return (ENXIO);
1020 		DELAY(1);	/* limit looping */
1021 	}
1022 
1023 	return (0);
1024 }
1025 #endif
1026 
1027 #undef PCI_VPD_TIMEOUT
1028 
1029 struct vpd_readstate {
1030 	device_t	pcib;
1031 	pcicfgregs	*cfg;
1032 	uint32_t	val;
1033 	int		bytesinval;
1034 	int		off;
1035 	uint8_t		cksum;
1036 };
1037 
1038 static int
1039 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1040 {
1041 	uint32_t reg;
1042 	uint8_t byte;
1043 
1044 	if (vrs->bytesinval == 0) {
1045 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1046 			return (ENXIO);
1047 		vrs->val = le32toh(reg);
1048 		vrs->off += 4;
1049 		byte = vrs->val & 0xff;
1050 		vrs->bytesinval = 3;
1051 	} else {
1052 		vrs->val = vrs->val >> 8;
1053 		byte = vrs->val & 0xff;
1054 		vrs->bytesinval--;
1055 	}
1056 
1057 	vrs->cksum += byte;
1058 	*data = byte;
1059 	return (0);
1060 }
1061 
1062 static void
1063 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1064 {
1065 	struct vpd_readstate vrs;
1066 	int state;
1067 	int name;
1068 	int remain;
1069 	int i;
1070 	int alloc, off;		/* alloc/off for RO/W arrays */
1071 	int cksumvalid;
1072 	int dflen;
1073 	uint8_t byte;
1074 	uint8_t byte2;
1075 
1076 	/* init vpd reader */
1077 	vrs.bytesinval = 0;
1078 	vrs.off = 0;
1079 	vrs.pcib = pcib;
1080 	vrs.cfg = cfg;
1081 	vrs.cksum = 0;
1082 
1083 	state = 0;
1084 	name = remain = i = 0;	/* shut up stupid gcc */
1085 	alloc = off = 0;	/* shut up stupid gcc */
1086 	dflen = 0;		/* shut up stupid gcc */
1087 	cksumvalid = -1;
1088 	while (state >= 0) {
1089 		if (vpd_nextbyte(&vrs, &byte)) {
1090 			state = -2;
1091 			break;
1092 		}
1093 #if 0
1094 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1095 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1096 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1097 #endif
1098 		switch (state) {
1099 		case 0:		/* item name */
1100 			if (byte & 0x80) {
1101 				if (vpd_nextbyte(&vrs, &byte2)) {
1102 					state = -2;
1103 					break;
1104 				}
1105 				remain = byte2;
1106 				if (vpd_nextbyte(&vrs, &byte2)) {
1107 					state = -2;
1108 					break;
1109 				}
1110 				remain |= byte2 << 8;
1111 				if (remain > (0x7f*4 - vrs.off)) {
1112 					state = -1;
1113 					pci_printf(cfg,
1114 					    "invalid VPD data, remain %#x\n",
1115 					    remain);
1116 				}
1117 				name = byte & 0x7f;
1118 			} else {
1119 				remain = byte & 0x7;
1120 				name = (byte >> 3) & 0xf;
1121 			}
1122 			switch (name) {
1123 			case 0x2:	/* String */
1124 				cfg->vpd.vpd_ident = malloc(remain + 1,
1125 				    M_DEVBUF, M_WAITOK);
1126 				i = 0;
1127 				state = 1;
1128 				break;
1129 			case 0xf:	/* End */
1130 				state = -1;
1131 				break;
1132 			case 0x10:	/* VPD-R */
1133 				alloc = 8;
1134 				off = 0;
1135 				cfg->vpd.vpd_ros = malloc(alloc *
1136 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1137 				    M_WAITOK | M_ZERO);
1138 				state = 2;
1139 				break;
1140 			case 0x11:	/* VPD-W */
1141 				alloc = 8;
1142 				off = 0;
1143 				cfg->vpd.vpd_w = malloc(alloc *
1144 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1145 				    M_WAITOK | M_ZERO);
1146 				state = 5;
1147 				break;
1148 			default:	/* Invalid data, abort */
1149 				state = -1;
1150 				break;
1151 			}
1152 			break;
1153 
1154 		case 1:	/* Identifier String */
1155 			cfg->vpd.vpd_ident[i++] = byte;
1156 			remain--;
1157 			if (remain == 0)  {
1158 				cfg->vpd.vpd_ident[i] = '\0';
1159 				state = 0;
1160 			}
1161 			break;
1162 
1163 		case 2:	/* VPD-R Keyword Header */
1164 			if (off == alloc) {
1165 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1166 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1167 				    M_DEVBUF, M_WAITOK | M_ZERO);
1168 			}
1169 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1170 			if (vpd_nextbyte(&vrs, &byte2)) {
1171 				state = -2;
1172 				break;
1173 			}
1174 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1175 			if (vpd_nextbyte(&vrs, &byte2)) {
1176 				state = -2;
1177 				break;
1178 			}
1179 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1180 			if (dflen == 0 &&
1181 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1182 			    2) == 0) {
1183 				/*
1184 				 * if this happens, we can't trust the rest
1185 				 * of the VPD.
1186 				 */
1187 				pci_printf(cfg, "bad keyword length: %d\n",
1188 				    dflen);
1189 				cksumvalid = 0;
1190 				state = -1;
1191 				break;
1192 			} else if (dflen == 0) {
1193 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1194 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1195 				    M_DEVBUF, M_WAITOK);
1196 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1197 			} else
1198 				cfg->vpd.vpd_ros[off].value = malloc(
1199 				    (dflen + 1) *
1200 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1201 				    M_DEVBUF, M_WAITOK);
1202 			remain -= 3;
1203 			i = 0;
1204 			/* keep in sync w/ state 3's transistions */
1205 			if (dflen == 0 && remain == 0)
1206 				state = 0;
1207 			else if (dflen == 0)
1208 				state = 2;
1209 			else
1210 				state = 3;
1211 			break;
1212 
1213 		case 3:	/* VPD-R Keyword Value */
1214 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1215 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1216 			    "RV", 2) == 0 && cksumvalid == -1) {
1217 				if (vrs.cksum == 0)
1218 					cksumvalid = 1;
1219 				else {
1220 					if (bootverbose)
1221 						pci_printf(cfg,
1222 					    "bad VPD cksum, remain %hhu\n",
1223 						    vrs.cksum);
1224 					cksumvalid = 0;
1225 					state = -1;
1226 					break;
1227 				}
1228 			}
1229 			dflen--;
1230 			remain--;
1231 			/* keep in sync w/ state 2's transistions */
1232 			if (dflen == 0)
1233 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1234 			if (dflen == 0 && remain == 0) {
1235 				cfg->vpd.vpd_rocnt = off;
1236 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1237 				    off * sizeof(*cfg->vpd.vpd_ros),
1238 				    M_DEVBUF, M_WAITOK | M_ZERO);
1239 				state = 0;
1240 			} else if (dflen == 0)
1241 				state = 2;
1242 			break;
1243 
1244 		case 4:
1245 			remain--;
1246 			if (remain == 0)
1247 				state = 0;
1248 			break;
1249 
1250 		case 5:	/* VPD-W Keyword Header */
1251 			if (off == alloc) {
1252 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1253 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1254 				    M_DEVBUF, M_WAITOK | M_ZERO);
1255 			}
1256 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1257 			if (vpd_nextbyte(&vrs, &byte2)) {
1258 				state = -2;
1259 				break;
1260 			}
1261 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1262 			if (vpd_nextbyte(&vrs, &byte2)) {
1263 				state = -2;
1264 				break;
1265 			}
1266 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1267 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1268 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1269 			    sizeof(*cfg->vpd.vpd_w[off].value),
1270 			    M_DEVBUF, M_WAITOK);
1271 			remain -= 3;
1272 			i = 0;
1273 			/* keep in sync w/ state 6's transistions */
1274 			if (dflen == 0 && remain == 0)
1275 				state = 0;
1276 			else if (dflen == 0)
1277 				state = 5;
1278 			else
1279 				state = 6;
1280 			break;
1281 
1282 		case 6:	/* VPD-W Keyword Value */
1283 			cfg->vpd.vpd_w[off].value[i++] = byte;
1284 			dflen--;
1285 			remain--;
1286 			/* keep in sync w/ state 5's transistions */
1287 			if (dflen == 0)
1288 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1289 			if (dflen == 0 && remain == 0) {
1290 				cfg->vpd.vpd_wcnt = off;
1291 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1292 				    off * sizeof(*cfg->vpd.vpd_w),
1293 				    M_DEVBUF, M_WAITOK | M_ZERO);
1294 				state = 0;
1295 			} else if (dflen == 0)
1296 				state = 5;
1297 			break;
1298 
1299 		default:
1300 			pci_printf(cfg, "invalid state: %d\n", state);
1301 			state = -1;
1302 			break;
1303 		}
1304 	}
1305 
1306 	if (cksumvalid == 0 || state < -1) {
1307 		/* read-only data bad, clean up */
1308 		if (cfg->vpd.vpd_ros != NULL) {
1309 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1310 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1311 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1312 			cfg->vpd.vpd_ros = NULL;
1313 		}
1314 	}
1315 	if (state < -1) {
1316 		/* I/O error, clean up */
1317 		pci_printf(cfg, "failed to read VPD data.\n");
1318 		if (cfg->vpd.vpd_ident != NULL) {
1319 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1320 			cfg->vpd.vpd_ident = NULL;
1321 		}
1322 		if (cfg->vpd.vpd_w != NULL) {
1323 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1324 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1325 			free(cfg->vpd.vpd_w, M_DEVBUF);
1326 			cfg->vpd.vpd_w = NULL;
1327 		}
1328 	}
1329 	cfg->vpd.vpd_cached = 1;
1330 #undef REG
1331 #undef WREG
1332 }
1333 
1334 int
1335 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1336 {
1337 	struct pci_devinfo *dinfo = device_get_ivars(child);
1338 	pcicfgregs *cfg = &dinfo->cfg;
1339 
1340 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1341 		pci_read_vpd(device_get_parent(dev), cfg);
1342 
1343 	*identptr = cfg->vpd.vpd_ident;
1344 
1345 	if (*identptr == NULL)
1346 		return (ENXIO);
1347 
1348 	return (0);
1349 }
1350 
1351 int
1352 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1353 	const char **vptr)
1354 {
1355 	struct pci_devinfo *dinfo = device_get_ivars(child);
1356 	pcicfgregs *cfg = &dinfo->cfg;
1357 	int i;
1358 
1359 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1360 		pci_read_vpd(device_get_parent(dev), cfg);
1361 
1362 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1363 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1364 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1365 			*vptr = cfg->vpd.vpd_ros[i].value;
1366 			return (0);
1367 		}
1368 
1369 	*vptr = NULL;
1370 	return (ENXIO);
1371 }
1372 
1373 struct pcicfg_vpd *
1374 pci_fetch_vpd_list(device_t dev)
1375 {
1376 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1377 	pcicfgregs *cfg = &dinfo->cfg;
1378 
1379 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1380 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1381 	return (&cfg->vpd);
1382 }
1383 
1384 /*
1385  * Find the requested HyperTransport capability and return the offset
1386  * in configuration space via the pointer provided.  The function
1387  * returns 0 on success and an error code otherwise.
1388  */
1389 int
1390 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1391 {
1392 	int ptr, error;
1393 	uint16_t val;
1394 
1395 	error = pci_find_cap(child, PCIY_HT, &ptr);
1396 	if (error)
1397 		return (error);
1398 
1399 	/*
1400 	 * Traverse the capabilities list checking each HT capability
1401 	 * to see if it matches the requested HT capability.
1402 	 */
1403 	for (;;) {
1404 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1405 		if (capability == PCIM_HTCAP_SLAVE ||
1406 		    capability == PCIM_HTCAP_HOST)
1407 			val &= 0xe000;
1408 		else
1409 			val &= PCIM_HTCMD_CAP_MASK;
1410 		if (val == capability) {
1411 			if (capreg != NULL)
1412 				*capreg = ptr;
1413 			return (0);
1414 		}
1415 
1416 		/* Skip to the next HT capability. */
1417 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1418 			break;
1419 	}
1420 
1421 	return (ENOENT);
1422 }
1423 
1424 /*
1425  * Find the next requested HyperTransport capability after start and return
1426  * the offset in configuration space via the pointer provided.  The function
1427  * returns 0 on success and an error code otherwise.
1428  */
1429 int
1430 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1431     int start, int *capreg)
1432 {
1433 	int ptr;
1434 	uint16_t val;
1435 
1436 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1437 	    ("start capability is not HyperTransport capability"));
1438 	ptr = start;
1439 
1440 	/*
1441 	 * Traverse the capabilities list checking each HT capability
1442 	 * to see if it matches the requested HT capability.
1443 	 */
1444 	for (;;) {
1445 		/* Skip to the next HT capability. */
1446 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1447 			break;
1448 
1449 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1450 		if (capability == PCIM_HTCAP_SLAVE ||
1451 		    capability == PCIM_HTCAP_HOST)
1452 			val &= 0xe000;
1453 		else
1454 			val &= PCIM_HTCMD_CAP_MASK;
1455 		if (val == capability) {
1456 			if (capreg != NULL)
1457 				*capreg = ptr;
1458 			return (0);
1459 		}
1460 	}
1461 
1462 	return (ENOENT);
1463 }
1464 
1465 /*
1466  * Find the requested capability and return the offset in
1467  * configuration space via the pointer provided.  The function returns
1468  * 0 on success and an error code otherwise.
1469  */
1470 int
1471 pci_find_cap_method(device_t dev, device_t child, int capability,
1472     int *capreg)
1473 {
1474 	struct pci_devinfo *dinfo = device_get_ivars(child);
1475 	pcicfgregs *cfg = &dinfo->cfg;
1476 	uint32_t status;
1477 	uint8_t ptr;
1478 
1479 	/*
1480 	 * Check the CAP_LIST bit of the PCI status register first.
1481 	 */
1482 	status = pci_read_config(child, PCIR_STATUS, 2);
1483 	if (!(status & PCIM_STATUS_CAPPRESENT))
1484 		return (ENXIO);
1485 
1486 	/*
1487 	 * Determine the start pointer of the capabilities list.
1488 	 */
1489 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1490 	case PCIM_HDRTYPE_NORMAL:
1491 	case PCIM_HDRTYPE_BRIDGE:
1492 		ptr = PCIR_CAP_PTR;
1493 		break;
1494 	case PCIM_HDRTYPE_CARDBUS:
1495 		ptr = PCIR_CAP_PTR_2;
1496 		break;
1497 	default:
1498 		/* XXX: panic? */
1499 		return (ENXIO);		/* no extended capabilities support */
1500 	}
1501 	ptr = pci_read_config(child, ptr, 1);
1502 
1503 	/*
1504 	 * Traverse the capabilities list.
1505 	 */
1506 	while (ptr != 0) {
1507 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1508 			if (capreg != NULL)
1509 				*capreg = ptr;
1510 			return (0);
1511 		}
1512 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1513 	}
1514 
1515 	return (ENOENT);
1516 }
1517 
1518 /*
1519  * Find the next requested capability after start and return the offset in
1520  * configuration space via the pointer provided.  The function returns
1521  * 0 on success and an error code otherwise.
1522  */
1523 int
1524 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1525     int start, int *capreg)
1526 {
1527 	uint8_t ptr;
1528 
1529 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1530 	    ("start capability is not expected capability"));
1531 
1532 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1533 	while (ptr != 0) {
1534 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1535 			if (capreg != NULL)
1536 				*capreg = ptr;
1537 			return (0);
1538 		}
1539 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1540 	}
1541 
1542 	return (ENOENT);
1543 }
1544 
1545 /*
1546  * Find the requested extended capability and return the offset in
1547  * configuration space via the pointer provided.  The function returns
1548  * 0 on success and an error code otherwise.
1549  */
1550 int
1551 pci_find_extcap_method(device_t dev, device_t child, int capability,
1552     int *capreg)
1553 {
1554 	struct pci_devinfo *dinfo = device_get_ivars(child);
1555 	pcicfgregs *cfg = &dinfo->cfg;
1556 	uint32_t ecap;
1557 	uint16_t ptr;
1558 
1559 	/* Only supported for PCI-express devices. */
1560 	if (cfg->pcie.pcie_location == 0)
1561 		return (ENXIO);
1562 
1563 	ptr = PCIR_EXTCAP;
1564 	ecap = pci_read_config(child, ptr, 4);
1565 	if (ecap == 0xffffffff || ecap == 0)
1566 		return (ENOENT);
1567 	for (;;) {
1568 		if (PCI_EXTCAP_ID(ecap) == capability) {
1569 			if (capreg != NULL)
1570 				*capreg = ptr;
1571 			return (0);
1572 		}
1573 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1574 		if (ptr == 0)
1575 			break;
1576 		ecap = pci_read_config(child, ptr, 4);
1577 	}
1578 
1579 	return (ENOENT);
1580 }
1581 
1582 /*
1583  * Find the next requested extended capability after start and return the
1584  * offset in configuration space via the pointer provided.  The function
1585  * returns 0 on success and an error code otherwise.
1586  */
1587 int
1588 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1589     int start, int *capreg)
1590 {
1591 	struct pci_devinfo *dinfo = device_get_ivars(child);
1592 	pcicfgregs *cfg = &dinfo->cfg;
1593 	uint32_t ecap;
1594 	uint16_t ptr;
1595 
1596 	/* Only supported for PCI-express devices. */
1597 	if (cfg->pcie.pcie_location == 0)
1598 		return (ENXIO);
1599 
1600 	ecap = pci_read_config(child, start, 4);
1601 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1602 	    ("start extended capability is not expected capability"));
1603 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1604 	while (ptr != 0) {
1605 		ecap = pci_read_config(child, ptr, 4);
1606 		if (PCI_EXTCAP_ID(ecap) == capability) {
1607 			if (capreg != NULL)
1608 				*capreg = ptr;
1609 			return (0);
1610 		}
1611 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1612 	}
1613 
1614 	return (ENOENT);
1615 }
1616 
1617 /*
1618  * Support for MSI-X message interrupts.
1619  */
1620 static void
1621 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1622 {
1623 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1624 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1625 	uint32_t offset;
1626 
1627 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1628 	offset = msix->msix_table_offset + index * 16;
1629 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1630 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1631 	bus_write_4(msix->msix_table_res, offset + 8, data);
1632 }
1633 
1634 void
1635 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1636     uint64_t address, uint32_t data)
1637 {
1638 
1639 	if (pci_msix_rewrite_table) {
1640 		struct pci_devinfo *dinfo = device_get_ivars(child);
1641 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1642 
1643 		/*
1644 		 * Some VM hosts require MSIX to be disabled in the
1645 		 * control register before updating the MSIX table
1646 		 * entries are allowed. It is not enough to only
1647 		 * disable MSIX while updating a single entry. MSIX
1648 		 * must be disabled while updating all entries in the
1649 		 * table.
1650 		 */
1651 		pci_write_config(child,
1652 		    msix->msix_location + PCIR_MSIX_CTRL,
1653 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1654 		pci_resume_msix(child);
1655 	} else
1656 		pci_write_msix_entry(child, index, address, data);
1657 
1658 	/* Enable MSI -> HT mapping. */
1659 	pci_ht_map_msi(child, address);
1660 }
1661 
1662 void
1663 pci_mask_msix(device_t dev, u_int index)
1664 {
1665 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1666 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1667 	uint32_t offset, val;
1668 
1669 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1670 	offset = msix->msix_table_offset + index * 16 + 12;
1671 	val = bus_read_4(msix->msix_table_res, offset);
1672 	val |= PCIM_MSIX_VCTRL_MASK;
1673 
1674 	/*
1675 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1676 	 * register, so always write the new value.
1677 	 */
1678 	bus_write_4(msix->msix_table_res, offset, val);
1679 }
1680 
1681 void
1682 pci_unmask_msix(device_t dev, u_int index)
1683 {
1684 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1685 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1686 	uint32_t offset, val;
1687 
1688 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1689 	offset = msix->msix_table_offset + index * 16 + 12;
1690 	val = bus_read_4(msix->msix_table_res, offset);
1691 	val &= ~PCIM_MSIX_VCTRL_MASK;
1692 
1693 	/*
1694 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1695 	 * register, so always write the new value.
1696 	 */
1697 	bus_write_4(msix->msix_table_res, offset, val);
1698 }
1699 
1700 int
1701 pci_pending_msix(device_t dev, u_int index)
1702 {
1703 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1704 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1705 	uint32_t offset, bit;
1706 
1707 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1708 	offset = msix->msix_pba_offset + (index / 32) * 4;
1709 	bit = 1 << index % 32;
1710 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1711 }
1712 
1713 /*
1714  * Restore MSI-X registers and table during resume.  If MSI-X is
1715  * enabled then walk the virtual table to restore the actual MSI-X
1716  * table.
1717  */
1718 static void
1719 pci_resume_msix(device_t dev)
1720 {
1721 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1722 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1723 	struct msix_table_entry *mte;
1724 	struct msix_vector *mv;
1725 	int i;
1726 
1727 	if (msix->msix_alloc > 0) {
1728 		/* First, mask all vectors. */
1729 		for (i = 0; i < msix->msix_msgnum; i++)
1730 			pci_mask_msix(dev, i);
1731 
1732 		/* Second, program any messages with at least one handler. */
1733 		for (i = 0; i < msix->msix_table_len; i++) {
1734 			mte = &msix->msix_table[i];
1735 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1736 				continue;
1737 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1738 			pci_write_msix_entry(dev, i, mv->mv_address,
1739 			    mv->mv_data);
1740 			pci_unmask_msix(dev, i);
1741 		}
1742 	}
1743 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1744 	    msix->msix_ctrl, 2);
1745 }
1746 
1747 /*
1748  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1749  * returned in *count.  After this function returns, each message will be
1750  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1751  */
1752 int
1753 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1754 {
1755 	struct pci_devinfo *dinfo = device_get_ivars(child);
1756 	pcicfgregs *cfg = &dinfo->cfg;
1757 	struct resource_list_entry *rle;
1758 	int actual, error, i, irq, max;
1759 
1760 	/* Don't let count == 0 get us into trouble. */
1761 	if (*count == 0)
1762 		return (EINVAL);
1763 
1764 	/* If rid 0 is allocated, then fail. */
1765 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1766 	if (rle != NULL && rle->res != NULL)
1767 		return (ENXIO);
1768 
1769 	/* Already have allocated messages? */
1770 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1771 		return (ENXIO);
1772 
1773 	/* If MSI-X is blacklisted for this system, fail. */
1774 	if (pci_msix_blacklisted())
1775 		return (ENXIO);
1776 
1777 	/* MSI-X capability present? */
1778 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1779 		return (ENODEV);
1780 
1781 	/* Make sure the appropriate BARs are mapped. */
1782 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1783 	    cfg->msix.msix_table_bar);
1784 	if (rle == NULL || rle->res == NULL ||
1785 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1786 		return (ENXIO);
1787 	cfg->msix.msix_table_res = rle->res;
1788 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1789 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1790 		    cfg->msix.msix_pba_bar);
1791 		if (rle == NULL || rle->res == NULL ||
1792 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1793 			return (ENXIO);
1794 	}
1795 	cfg->msix.msix_pba_res = rle->res;
1796 
1797 	if (bootverbose)
1798 		device_printf(child,
1799 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1800 		    *count, cfg->msix.msix_msgnum);
1801 	max = min(*count, cfg->msix.msix_msgnum);
1802 	for (i = 0; i < max; i++) {
1803 		/* Allocate a message. */
1804 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1805 		if (error) {
1806 			if (i == 0)
1807 				return (error);
1808 			break;
1809 		}
1810 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1811 		    irq, 1);
1812 	}
1813 	actual = i;
1814 
1815 	if (bootverbose) {
1816 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1817 		if (actual == 1)
1818 			device_printf(child, "using IRQ %ju for MSI-X\n",
1819 			    rle->start);
1820 		else {
1821 			int run;
1822 
1823 			/*
1824 			 * Be fancy and try to print contiguous runs of
1825 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1826 			 * 'run' is true if we are in a range.
1827 			 */
1828 			device_printf(child, "using IRQs %ju", rle->start);
1829 			irq = rle->start;
1830 			run = 0;
1831 			for (i = 1; i < actual; i++) {
1832 				rle = resource_list_find(&dinfo->resources,
1833 				    SYS_RES_IRQ, i + 1);
1834 
1835 				/* Still in a run? */
1836 				if (rle->start == irq + 1) {
1837 					run = 1;
1838 					irq++;
1839 					continue;
1840 				}
1841 
1842 				/* Finish previous range. */
1843 				if (run) {
1844 					printf("-%d", irq);
1845 					run = 0;
1846 				}
1847 
1848 				/* Start new range. */
1849 				printf(",%ju", rle->start);
1850 				irq = rle->start;
1851 			}
1852 
1853 			/* Unfinished range? */
1854 			if (run)
1855 				printf("-%d", irq);
1856 			printf(" for MSI-X\n");
1857 		}
1858 	}
1859 
1860 	/* Mask all vectors. */
1861 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1862 		pci_mask_msix(child, i);
1863 
1864 	/* Allocate and initialize vector data and virtual table. */
1865 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1866 	    M_DEVBUF, M_WAITOK | M_ZERO);
1867 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1868 	    M_DEVBUF, M_WAITOK | M_ZERO);
1869 	for (i = 0; i < actual; i++) {
1870 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1871 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1872 		cfg->msix.msix_table[i].mte_vector = i + 1;
1873 	}
1874 
1875 	/* Update control register to enable MSI-X. */
1876 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1877 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1878 	    cfg->msix.msix_ctrl, 2);
1879 
1880 	/* Update counts of alloc'd messages. */
1881 	cfg->msix.msix_alloc = actual;
1882 	cfg->msix.msix_table_len = actual;
1883 	*count = actual;
1884 	return (0);
1885 }
1886 
1887 /*
1888  * By default, pci_alloc_msix() will assign the allocated IRQ
1889  * resources consecutively to the first N messages in the MSI-X table.
1890  * However, device drivers may want to use different layouts if they
1891  * either receive fewer messages than they asked for, or they wish to
1892  * populate the MSI-X table sparsely.  This method allows the driver
1893  * to specify what layout it wants.  It must be called after a
1894  * successful pci_alloc_msix() but before any of the associated
1895  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1896  *
1897  * The 'vectors' array contains 'count' message vectors.  The array
1898  * maps directly to the MSI-X table in that index 0 in the array
1899  * specifies the vector for the first message in the MSI-X table, etc.
1900  * The vector value in each array index can either be 0 to indicate
1901  * that no vector should be assigned to a message slot, or it can be a
1902  * number from 1 to N (where N is the count returned from a
1903  * succcessful call to pci_alloc_msix()) to indicate which message
1904  * vector (IRQ) to be used for the corresponding message.
1905  *
1906  * On successful return, each message with a non-zero vector will have
1907  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1908  * 1.  Additionally, if any of the IRQs allocated via the previous
1909  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1910  * will be freed back to the system automatically.
1911  *
1912  * For example, suppose a driver has a MSI-X table with 6 messages and
1913  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1914  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1915  * C.  After the call to pci_alloc_msix(), the device will be setup to
1916  * have an MSI-X table of ABC--- (where - means no vector assigned).
1917  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1918  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1919  * be freed back to the system.  This device will also have valid
1920  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1921  *
1922  * In any case, the SYS_RES_IRQ rid X will always map to the message
1923  * at MSI-X table index X - 1 and will only be valid if a vector is
1924  * assigned to that table entry.
1925  */
1926 int
1927 pci_remap_msix_method(device_t dev, device_t child, int count,
1928     const u_int *vectors)
1929 {
1930 	struct pci_devinfo *dinfo = device_get_ivars(child);
1931 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1932 	struct resource_list_entry *rle;
1933 	int i, irq, j, *used;
1934 
1935 	/*
1936 	 * Have to have at least one message in the table but the
1937 	 * table can't be bigger than the actual MSI-X table in the
1938 	 * device.
1939 	 */
1940 	if (count == 0 || count > msix->msix_msgnum)
1941 		return (EINVAL);
1942 
1943 	/* Sanity check the vectors. */
1944 	for (i = 0; i < count; i++)
1945 		if (vectors[i] > msix->msix_alloc)
1946 			return (EINVAL);
1947 
1948 	/*
1949 	 * Make sure there aren't any holes in the vectors to be used.
1950 	 * It's a big pain to support it, and it doesn't really make
1951 	 * sense anyway.  Also, at least one vector must be used.
1952 	 */
1953 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1954 	    M_ZERO);
1955 	for (i = 0; i < count; i++)
1956 		if (vectors[i] != 0)
1957 			used[vectors[i] - 1] = 1;
1958 	for (i = 0; i < msix->msix_alloc - 1; i++)
1959 		if (used[i] == 0 && used[i + 1] == 1) {
1960 			free(used, M_DEVBUF);
1961 			return (EINVAL);
1962 		}
1963 	if (used[0] != 1) {
1964 		free(used, M_DEVBUF);
1965 		return (EINVAL);
1966 	}
1967 
1968 	/* Make sure none of the resources are allocated. */
1969 	for (i = 0; i < msix->msix_table_len; i++) {
1970 		if (msix->msix_table[i].mte_vector == 0)
1971 			continue;
1972 		if (msix->msix_table[i].mte_handlers > 0) {
1973 			free(used, M_DEVBUF);
1974 			return (EBUSY);
1975 		}
1976 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1977 		KASSERT(rle != NULL, ("missing resource"));
1978 		if (rle->res != NULL) {
1979 			free(used, M_DEVBUF);
1980 			return (EBUSY);
1981 		}
1982 	}
1983 
1984 	/* Free the existing resource list entries. */
1985 	for (i = 0; i < msix->msix_table_len; i++) {
1986 		if (msix->msix_table[i].mte_vector == 0)
1987 			continue;
1988 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1989 	}
1990 
1991 	/*
1992 	 * Build the new virtual table keeping track of which vectors are
1993 	 * used.
1994 	 */
1995 	free(msix->msix_table, M_DEVBUF);
1996 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1997 	    M_DEVBUF, M_WAITOK | M_ZERO);
1998 	for (i = 0; i < count; i++)
1999 		msix->msix_table[i].mte_vector = vectors[i];
2000 	msix->msix_table_len = count;
2001 
2002 	/* Free any unused IRQs and resize the vectors array if necessary. */
2003 	j = msix->msix_alloc - 1;
2004 	if (used[j] == 0) {
2005 		struct msix_vector *vec;
2006 
2007 		while (used[j] == 0) {
2008 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2009 			    msix->msix_vectors[j].mv_irq);
2010 			j--;
2011 		}
2012 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2013 		    M_WAITOK);
2014 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2015 		    (j + 1));
2016 		free(msix->msix_vectors, M_DEVBUF);
2017 		msix->msix_vectors = vec;
2018 		msix->msix_alloc = j + 1;
2019 	}
2020 	free(used, M_DEVBUF);
2021 
2022 	/* Map the IRQs onto the rids. */
2023 	for (i = 0; i < count; i++) {
2024 		if (vectors[i] == 0)
2025 			continue;
2026 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2027 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2028 		    irq, 1);
2029 	}
2030 
2031 	if (bootverbose) {
2032 		device_printf(child, "Remapped MSI-X IRQs as: ");
2033 		for (i = 0; i < count; i++) {
2034 			if (i != 0)
2035 				printf(", ");
2036 			if (vectors[i] == 0)
2037 				printf("---");
2038 			else
2039 				printf("%d",
2040 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2041 		}
2042 		printf("\n");
2043 	}
2044 
2045 	return (0);
2046 }
2047 
2048 static int
2049 pci_release_msix(device_t dev, device_t child)
2050 {
2051 	struct pci_devinfo *dinfo = device_get_ivars(child);
2052 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2053 	struct resource_list_entry *rle;
2054 	int i;
2055 
2056 	/* Do we have any messages to release? */
2057 	if (msix->msix_alloc == 0)
2058 		return (ENODEV);
2059 
2060 	/* Make sure none of the resources are allocated. */
2061 	for (i = 0; i < msix->msix_table_len; i++) {
2062 		if (msix->msix_table[i].mte_vector == 0)
2063 			continue;
2064 		if (msix->msix_table[i].mte_handlers > 0)
2065 			return (EBUSY);
2066 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2067 		KASSERT(rle != NULL, ("missing resource"));
2068 		if (rle->res != NULL)
2069 			return (EBUSY);
2070 	}
2071 
2072 	/* Update control register to disable MSI-X. */
2073 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2074 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2075 	    msix->msix_ctrl, 2);
2076 
2077 	/* Free the resource list entries. */
2078 	for (i = 0; i < msix->msix_table_len; i++) {
2079 		if (msix->msix_table[i].mte_vector == 0)
2080 			continue;
2081 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2082 	}
2083 	free(msix->msix_table, M_DEVBUF);
2084 	msix->msix_table_len = 0;
2085 
2086 	/* Release the IRQs. */
2087 	for (i = 0; i < msix->msix_alloc; i++)
2088 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2089 		    msix->msix_vectors[i].mv_irq);
2090 	free(msix->msix_vectors, M_DEVBUF);
2091 	msix->msix_alloc = 0;
2092 	return (0);
2093 }
2094 
2095 /*
2096  * Return the max supported MSI-X messages this device supports.
2097  * Basically, assuming the MD code can alloc messages, this function
2098  * should return the maximum value that pci_alloc_msix() can return.
2099  * Thus, it is subject to the tunables, etc.
2100  */
2101 int
2102 pci_msix_count_method(device_t dev, device_t child)
2103 {
2104 	struct pci_devinfo *dinfo = device_get_ivars(child);
2105 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2106 
2107 	if (pci_do_msix && msix->msix_location != 0)
2108 		return (msix->msix_msgnum);
2109 	return (0);
2110 }
2111 
2112 int
2113 pci_msix_pba_bar_method(device_t dev, device_t child)
2114 {
2115 	struct pci_devinfo *dinfo = device_get_ivars(child);
2116 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2117 
2118 	if (pci_do_msix && msix->msix_location != 0)
2119 		return (msix->msix_pba_bar);
2120 	return (-1);
2121 }
2122 
2123 int
2124 pci_msix_table_bar_method(device_t dev, device_t child)
2125 {
2126 	struct pci_devinfo *dinfo = device_get_ivars(child);
2127 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2128 
2129 	if (pci_do_msix && msix->msix_location != 0)
2130 		return (msix->msix_table_bar);
2131 	return (-1);
2132 }
2133 
2134 /*
2135  * HyperTransport MSI mapping control
2136  */
2137 void
2138 pci_ht_map_msi(device_t dev, uint64_t addr)
2139 {
2140 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2141 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2142 
2143 	if (!ht->ht_msimap)
2144 		return;
2145 
2146 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2147 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2148 		/* Enable MSI -> HT mapping. */
2149 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2150 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2151 		    ht->ht_msictrl, 2);
2152 	}
2153 
2154 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2155 		/* Disable MSI -> HT mapping. */
2156 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2157 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2158 		    ht->ht_msictrl, 2);
2159 	}
2160 }
2161 
2162 int
2163 pci_get_max_payload(device_t dev)
2164 {
2165 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2166 	int cap;
2167 	uint16_t val;
2168 
2169 	cap = dinfo->cfg.pcie.pcie_location;
2170 	if (cap == 0)
2171 		return (0);
2172 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2173 	val &= PCIEM_CTL_MAX_PAYLOAD;
2174 	val >>= 5;
2175 	return (1 << (val + 7));
2176 }
2177 
2178 int
2179 pci_get_max_read_req(device_t dev)
2180 {
2181 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2182 	int cap;
2183 	uint16_t val;
2184 
2185 	cap = dinfo->cfg.pcie.pcie_location;
2186 	if (cap == 0)
2187 		return (0);
2188 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2189 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2190 	val >>= 12;
2191 	return (1 << (val + 7));
2192 }
2193 
2194 int
2195 pci_set_max_read_req(device_t dev, int size)
2196 {
2197 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2198 	int cap;
2199 	uint16_t val;
2200 
2201 	cap = dinfo->cfg.pcie.pcie_location;
2202 	if (cap == 0)
2203 		return (0);
2204 	if (size < 128)
2205 		size = 128;
2206 	if (size > 4096)
2207 		size = 4096;
2208 	size = (1 << (fls(size) - 1));
2209 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2210 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2211 	val |= (fls(size) - 8) << 12;
2212 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2213 	return (size);
2214 }
2215 
2216 uint32_t
2217 pcie_read_config(device_t dev, int reg, int width)
2218 {
2219 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2220 	int cap;
2221 
2222 	cap = dinfo->cfg.pcie.pcie_location;
2223 	if (cap == 0) {
2224 		if (width == 2)
2225 			return (0xffff);
2226 		return (0xffffffff);
2227 	}
2228 
2229 	return (pci_read_config(dev, cap + reg, width));
2230 }
2231 
2232 void
2233 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2234 {
2235 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2236 	int cap;
2237 
2238 	cap = dinfo->cfg.pcie.pcie_location;
2239 	if (cap == 0)
2240 		return;
2241 	pci_write_config(dev, cap + reg, value, width);
2242 }
2243 
2244 /*
2245  * Adjusts a PCI-e capability register by clearing the bits in mask
2246  * and setting the bits in (value & mask).  Bits not set in mask are
2247  * not adjusted.
2248  *
2249  * Returns the old value on success or all ones on failure.
2250  */
2251 uint32_t
2252 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2253     int width)
2254 {
2255 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2256 	uint32_t old, new;
2257 	int cap;
2258 
2259 	cap = dinfo->cfg.pcie.pcie_location;
2260 	if (cap == 0) {
2261 		if (width == 2)
2262 			return (0xffff);
2263 		return (0xffffffff);
2264 	}
2265 
2266 	old = pci_read_config(dev, cap + reg, width);
2267 	new = old & ~mask;
2268 	new |= (value & mask);
2269 	pci_write_config(dev, cap + reg, new, width);
2270 	return (old);
2271 }
2272 
2273 /*
2274  * Support for MSI message signalled interrupts.
2275  */
2276 void
2277 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2278     uint16_t data)
2279 {
2280 	struct pci_devinfo *dinfo = device_get_ivars(child);
2281 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2282 
2283 	/* Write data and address values. */
2284 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2285 	    address & 0xffffffff, 4);
2286 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2287 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2288 		    address >> 32, 4);
2289 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2290 		    data, 2);
2291 	} else
2292 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2293 		    2);
2294 
2295 	/* Enable MSI in the control register. */
2296 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2297 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2298 	    msi->msi_ctrl, 2);
2299 
2300 	/* Enable MSI -> HT mapping. */
2301 	pci_ht_map_msi(child, address);
2302 }
2303 
2304 void
2305 pci_disable_msi_method(device_t dev, device_t child)
2306 {
2307 	struct pci_devinfo *dinfo = device_get_ivars(child);
2308 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2309 
2310 	/* Disable MSI -> HT mapping. */
2311 	pci_ht_map_msi(child, 0);
2312 
2313 	/* Disable MSI in the control register. */
2314 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2315 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2316 	    msi->msi_ctrl, 2);
2317 }
2318 
2319 /*
2320  * Restore MSI registers during resume.  If MSI is enabled then
2321  * restore the data and address registers in addition to the control
2322  * register.
2323  */
2324 static void
2325 pci_resume_msi(device_t dev)
2326 {
2327 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2328 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2329 	uint64_t address;
2330 	uint16_t data;
2331 
2332 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2333 		address = msi->msi_addr;
2334 		data = msi->msi_data;
2335 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2336 		    address & 0xffffffff, 4);
2337 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2338 			pci_write_config(dev, msi->msi_location +
2339 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2340 			pci_write_config(dev, msi->msi_location +
2341 			    PCIR_MSI_DATA_64BIT, data, 2);
2342 		} else
2343 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2344 			    data, 2);
2345 	}
2346 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2347 	    2);
2348 }
2349 
2350 static int
2351 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2352 {
2353 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2354 	pcicfgregs *cfg = &dinfo->cfg;
2355 	struct resource_list_entry *rle;
2356 	struct msix_table_entry *mte;
2357 	struct msix_vector *mv;
2358 	uint64_t addr;
2359 	uint32_t data;
2360 	int error, i, j;
2361 
2362 	/*
2363 	 * Handle MSI first.  We try to find this IRQ among our list
2364 	 * of MSI IRQs.  If we find it, we request updated address and
2365 	 * data registers and apply the results.
2366 	 */
2367 	if (cfg->msi.msi_alloc > 0) {
2368 
2369 		/* If we don't have any active handlers, nothing to do. */
2370 		if (cfg->msi.msi_handlers == 0)
2371 			return (0);
2372 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2373 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2374 			    i + 1);
2375 			if (rle->start == irq) {
2376 				error = PCIB_MAP_MSI(device_get_parent(bus),
2377 				    dev, irq, &addr, &data);
2378 				if (error)
2379 					return (error);
2380 				pci_disable_msi(dev);
2381 				dinfo->cfg.msi.msi_addr = addr;
2382 				dinfo->cfg.msi.msi_data = data;
2383 				pci_enable_msi(dev, addr, data);
2384 				return (0);
2385 			}
2386 		}
2387 		return (ENOENT);
2388 	}
2389 
2390 	/*
2391 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2392 	 * we request the updated mapping info.  If that works, we go
2393 	 * through all the slots that use this IRQ and update them.
2394 	 */
2395 	if (cfg->msix.msix_alloc > 0) {
2396 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2397 			mv = &cfg->msix.msix_vectors[i];
2398 			if (mv->mv_irq == irq) {
2399 				error = PCIB_MAP_MSI(device_get_parent(bus),
2400 				    dev, irq, &addr, &data);
2401 				if (error)
2402 					return (error);
2403 				mv->mv_address = addr;
2404 				mv->mv_data = data;
2405 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2406 					mte = &cfg->msix.msix_table[j];
2407 					if (mte->mte_vector != i + 1)
2408 						continue;
2409 					if (mte->mte_handlers == 0)
2410 						continue;
2411 					pci_mask_msix(dev, j);
2412 					pci_enable_msix(dev, j, addr, data);
2413 					pci_unmask_msix(dev, j);
2414 				}
2415 			}
2416 		}
2417 		return (ENOENT);
2418 	}
2419 
2420 	return (ENOENT);
2421 }
2422 
2423 /*
2424  * Returns true if the specified device is blacklisted because MSI
2425  * doesn't work.
2426  */
2427 int
2428 pci_msi_device_blacklisted(device_t dev)
2429 {
2430 
2431 	if (!pci_honor_msi_blacklist)
2432 		return (0);
2433 
2434 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2435 }
2436 
2437 /*
2438  * Determine if MSI is blacklisted globally on this system.  Currently,
2439  * we just check for blacklisted chipsets as represented by the
2440  * host-PCI bridge at device 0:0:0.  In the future, it may become
2441  * necessary to check other system attributes, such as the kenv values
2442  * that give the motherboard manufacturer and model number.
2443  */
2444 static int
2445 pci_msi_blacklisted(void)
2446 {
2447 	device_t dev;
2448 
2449 	if (!pci_honor_msi_blacklist)
2450 		return (0);
2451 
2452 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2453 	if (!(pcie_chipset || pcix_chipset)) {
2454 		if (vm_guest != VM_GUEST_NO) {
2455 			/*
2456 			 * Whitelist older chipsets in virtual
2457 			 * machines known to support MSI.
2458 			 */
2459 			dev = pci_find_bsf(0, 0, 0);
2460 			if (dev != NULL)
2461 				return (!pci_has_quirk(pci_get_devid(dev),
2462 					PCI_QUIRK_ENABLE_MSI_VM));
2463 		}
2464 		return (1);
2465 	}
2466 
2467 	dev = pci_find_bsf(0, 0, 0);
2468 	if (dev != NULL)
2469 		return (pci_msi_device_blacklisted(dev));
2470 	return (0);
2471 }
2472 
2473 /*
2474  * Returns true if the specified device is blacklisted because MSI-X
2475  * doesn't work.  Note that this assumes that if MSI doesn't work,
2476  * MSI-X doesn't either.
2477  */
2478 int
2479 pci_msix_device_blacklisted(device_t dev)
2480 {
2481 
2482 	if (!pci_honor_msi_blacklist)
2483 		return (0);
2484 
2485 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2486 		return (1);
2487 
2488 	return (pci_msi_device_blacklisted(dev));
2489 }
2490 
2491 /*
2492  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2493  * is blacklisted, assume that MSI-X is as well.  Check for additional
2494  * chipsets where MSI works but MSI-X does not.
2495  */
2496 static int
2497 pci_msix_blacklisted(void)
2498 {
2499 	device_t dev;
2500 
2501 	if (!pci_honor_msi_blacklist)
2502 		return (0);
2503 
2504 	dev = pci_find_bsf(0, 0, 0);
2505 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2506 	    PCI_QUIRK_DISABLE_MSIX))
2507 		return (1);
2508 
2509 	return (pci_msi_blacklisted());
2510 }
2511 
2512 /*
2513  * Attempt to allocate *count MSI messages.  The actual number allocated is
2514  * returned in *count.  After this function returns, each message will be
2515  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2516  */
2517 int
2518 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2519 {
2520 	struct pci_devinfo *dinfo = device_get_ivars(child);
2521 	pcicfgregs *cfg = &dinfo->cfg;
2522 	struct resource_list_entry *rle;
2523 	int actual, error, i, irqs[32];
2524 	uint16_t ctrl;
2525 
2526 	/* Don't let count == 0 get us into trouble. */
2527 	if (*count == 0)
2528 		return (EINVAL);
2529 
2530 	/* If rid 0 is allocated, then fail. */
2531 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2532 	if (rle != NULL && rle->res != NULL)
2533 		return (ENXIO);
2534 
2535 	/* Already have allocated messages? */
2536 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2537 		return (ENXIO);
2538 
2539 	/* If MSI is blacklisted for this system, fail. */
2540 	if (pci_msi_blacklisted())
2541 		return (ENXIO);
2542 
2543 	/* MSI capability present? */
2544 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2545 		return (ENODEV);
2546 
2547 	if (bootverbose)
2548 		device_printf(child,
2549 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2550 		    *count, cfg->msi.msi_msgnum);
2551 
2552 	/* Don't ask for more than the device supports. */
2553 	actual = min(*count, cfg->msi.msi_msgnum);
2554 
2555 	/* Don't ask for more than 32 messages. */
2556 	actual = min(actual, 32);
2557 
2558 	/* MSI requires power of 2 number of messages. */
2559 	if (!powerof2(actual))
2560 		return (EINVAL);
2561 
2562 	for (;;) {
2563 		/* Try to allocate N messages. */
2564 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2565 		    actual, irqs);
2566 		if (error == 0)
2567 			break;
2568 		if (actual == 1)
2569 			return (error);
2570 
2571 		/* Try N / 2. */
2572 		actual >>= 1;
2573 	}
2574 
2575 	/*
2576 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2577 	 * resources in the irqs[] array, so add new resources
2578 	 * starting at rid 1.
2579 	 */
2580 	for (i = 0; i < actual; i++)
2581 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2582 		    irqs[i], irqs[i], 1);
2583 
2584 	if (bootverbose) {
2585 		if (actual == 1)
2586 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2587 		else {
2588 			int run;
2589 
2590 			/*
2591 			 * Be fancy and try to print contiguous runs
2592 			 * of IRQ values as ranges.  'run' is true if
2593 			 * we are in a range.
2594 			 */
2595 			device_printf(child, "using IRQs %d", irqs[0]);
2596 			run = 0;
2597 			for (i = 1; i < actual; i++) {
2598 
2599 				/* Still in a run? */
2600 				if (irqs[i] == irqs[i - 1] + 1) {
2601 					run = 1;
2602 					continue;
2603 				}
2604 
2605 				/* Finish previous range. */
2606 				if (run) {
2607 					printf("-%d", irqs[i - 1]);
2608 					run = 0;
2609 				}
2610 
2611 				/* Start new range. */
2612 				printf(",%d", irqs[i]);
2613 			}
2614 
2615 			/* Unfinished range? */
2616 			if (run)
2617 				printf("-%d", irqs[actual - 1]);
2618 			printf(" for MSI\n");
2619 		}
2620 	}
2621 
2622 	/* Update control register with actual count. */
2623 	ctrl = cfg->msi.msi_ctrl;
2624 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2625 	ctrl |= (ffs(actual) - 1) << 4;
2626 	cfg->msi.msi_ctrl = ctrl;
2627 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2628 
2629 	/* Update counts of alloc'd messages. */
2630 	cfg->msi.msi_alloc = actual;
2631 	cfg->msi.msi_handlers = 0;
2632 	*count = actual;
2633 	return (0);
2634 }
2635 
2636 /* Release the MSI messages associated with this device. */
2637 int
2638 pci_release_msi_method(device_t dev, device_t child)
2639 {
2640 	struct pci_devinfo *dinfo = device_get_ivars(child);
2641 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2642 	struct resource_list_entry *rle;
2643 	int error, i, irqs[32];
2644 
2645 	/* Try MSI-X first. */
2646 	error = pci_release_msix(dev, child);
2647 	if (error != ENODEV)
2648 		return (error);
2649 
2650 	/* Do we have any messages to release? */
2651 	if (msi->msi_alloc == 0)
2652 		return (ENODEV);
2653 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2654 
2655 	/* Make sure none of the resources are allocated. */
2656 	if (msi->msi_handlers > 0)
2657 		return (EBUSY);
2658 	for (i = 0; i < msi->msi_alloc; i++) {
2659 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2660 		KASSERT(rle != NULL, ("missing MSI resource"));
2661 		if (rle->res != NULL)
2662 			return (EBUSY);
2663 		irqs[i] = rle->start;
2664 	}
2665 
2666 	/* Update control register with 0 count. */
2667 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2668 	    ("%s: MSI still enabled", __func__));
2669 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2670 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2671 	    msi->msi_ctrl, 2);
2672 
2673 	/* Release the messages. */
2674 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2675 	for (i = 0; i < msi->msi_alloc; i++)
2676 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2677 
2678 	/* Update alloc count. */
2679 	msi->msi_alloc = 0;
2680 	msi->msi_addr = 0;
2681 	msi->msi_data = 0;
2682 	return (0);
2683 }
2684 
2685 /*
2686  * Return the max supported MSI messages this device supports.
2687  * Basically, assuming the MD code can alloc messages, this function
2688  * should return the maximum value that pci_alloc_msi() can return.
2689  * Thus, it is subject to the tunables, etc.
2690  */
2691 int
2692 pci_msi_count_method(device_t dev, device_t child)
2693 {
2694 	struct pci_devinfo *dinfo = device_get_ivars(child);
2695 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2696 
2697 	if (pci_do_msi && msi->msi_location != 0)
2698 		return (msi->msi_msgnum);
2699 	return (0);
2700 }
2701 
2702 /* free pcicfgregs structure and all depending data structures */
2703 
2704 int
2705 pci_freecfg(struct pci_devinfo *dinfo)
2706 {
2707 	struct devlist *devlist_head;
2708 	struct pci_map *pm, *next;
2709 	int i;
2710 
2711 	devlist_head = &pci_devq;
2712 
2713 	if (dinfo->cfg.vpd.vpd_reg) {
2714 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2715 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2716 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2717 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2718 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2719 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2720 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2721 	}
2722 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2723 		free(pm, M_DEVBUF);
2724 	}
2725 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2726 	free(dinfo, M_DEVBUF);
2727 
2728 	/* increment the generation count */
2729 	pci_generation++;
2730 
2731 	/* we're losing one device */
2732 	pci_numdevs--;
2733 	return (0);
2734 }
2735 
2736 /*
2737  * PCI power manangement
2738  */
2739 int
2740 pci_set_powerstate_method(device_t dev, device_t child, int state)
2741 {
2742 	struct pci_devinfo *dinfo = device_get_ivars(child);
2743 	pcicfgregs *cfg = &dinfo->cfg;
2744 	uint16_t status;
2745 	int oldstate, highest, delay;
2746 
2747 	if (cfg->pp.pp_cap == 0)
2748 		return (EOPNOTSUPP);
2749 
2750 	/*
2751 	 * Optimize a no state change request away.  While it would be OK to
2752 	 * write to the hardware in theory, some devices have shown odd
2753 	 * behavior when going from D3 -> D3.
2754 	 */
2755 	oldstate = pci_get_powerstate(child);
2756 	if (oldstate == state)
2757 		return (0);
2758 
2759 	/*
2760 	 * The PCI power management specification states that after a state
2761 	 * transition between PCI power states, system software must
2762 	 * guarantee a minimal delay before the function accesses the device.
2763 	 * Compute the worst case delay that we need to guarantee before we
2764 	 * access the device.  Many devices will be responsive much more
2765 	 * quickly than this delay, but there are some that don't respond
2766 	 * instantly to state changes.  Transitions to/from D3 state require
2767 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2768 	 * is done below with DELAY rather than a sleeper function because
2769 	 * this function can be called from contexts where we cannot sleep.
2770 	 */
2771 	highest = (oldstate > state) ? oldstate : state;
2772 	if (highest == PCI_POWERSTATE_D3)
2773 	    delay = 10000;
2774 	else if (highest == PCI_POWERSTATE_D2)
2775 	    delay = 200;
2776 	else
2777 	    delay = 0;
2778 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2779 	    & ~PCIM_PSTAT_DMASK;
2780 	switch (state) {
2781 	case PCI_POWERSTATE_D0:
2782 		status |= PCIM_PSTAT_D0;
2783 		break;
2784 	case PCI_POWERSTATE_D1:
2785 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2786 			return (EOPNOTSUPP);
2787 		status |= PCIM_PSTAT_D1;
2788 		break;
2789 	case PCI_POWERSTATE_D2:
2790 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2791 			return (EOPNOTSUPP);
2792 		status |= PCIM_PSTAT_D2;
2793 		break;
2794 	case PCI_POWERSTATE_D3:
2795 		status |= PCIM_PSTAT_D3;
2796 		break;
2797 	default:
2798 		return (EINVAL);
2799 	}
2800 
2801 	if (bootverbose)
2802 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2803 		    state);
2804 
2805 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2806 	if (delay)
2807 		DELAY(delay);
2808 	return (0);
2809 }
2810 
2811 int
2812 pci_get_powerstate_method(device_t dev, device_t child)
2813 {
2814 	struct pci_devinfo *dinfo = device_get_ivars(child);
2815 	pcicfgregs *cfg = &dinfo->cfg;
2816 	uint16_t status;
2817 	int result;
2818 
2819 	if (cfg->pp.pp_cap != 0) {
2820 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2821 		switch (status & PCIM_PSTAT_DMASK) {
2822 		case PCIM_PSTAT_D0:
2823 			result = PCI_POWERSTATE_D0;
2824 			break;
2825 		case PCIM_PSTAT_D1:
2826 			result = PCI_POWERSTATE_D1;
2827 			break;
2828 		case PCIM_PSTAT_D2:
2829 			result = PCI_POWERSTATE_D2;
2830 			break;
2831 		case PCIM_PSTAT_D3:
2832 			result = PCI_POWERSTATE_D3;
2833 			break;
2834 		default:
2835 			result = PCI_POWERSTATE_UNKNOWN;
2836 			break;
2837 		}
2838 	} else {
2839 		/* No support, device is always at D0 */
2840 		result = PCI_POWERSTATE_D0;
2841 	}
2842 	return (result);
2843 }
2844 
2845 /*
2846  * Some convenience functions for PCI device drivers.
2847  */
2848 
2849 static __inline void
2850 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2851 {
2852 	uint16_t	command;
2853 
2854 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2855 	command |= bit;
2856 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2857 }
2858 
2859 static __inline void
2860 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2861 {
2862 	uint16_t	command;
2863 
2864 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2865 	command &= ~bit;
2866 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2867 }
2868 
2869 int
2870 pci_enable_busmaster_method(device_t dev, device_t child)
2871 {
2872 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2873 	return (0);
2874 }
2875 
2876 int
2877 pci_disable_busmaster_method(device_t dev, device_t child)
2878 {
2879 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2880 	return (0);
2881 }
2882 
2883 int
2884 pci_enable_io_method(device_t dev, device_t child, int space)
2885 {
2886 	uint16_t bit;
2887 
2888 	switch(space) {
2889 	case SYS_RES_IOPORT:
2890 		bit = PCIM_CMD_PORTEN;
2891 		break;
2892 	case SYS_RES_MEMORY:
2893 		bit = PCIM_CMD_MEMEN;
2894 		break;
2895 	default:
2896 		return (EINVAL);
2897 	}
2898 	pci_set_command_bit(dev, child, bit);
2899 	return (0);
2900 }
2901 
2902 int
2903 pci_disable_io_method(device_t dev, device_t child, int space)
2904 {
2905 	uint16_t bit;
2906 
2907 	switch(space) {
2908 	case SYS_RES_IOPORT:
2909 		bit = PCIM_CMD_PORTEN;
2910 		break;
2911 	case SYS_RES_MEMORY:
2912 		bit = PCIM_CMD_MEMEN;
2913 		break;
2914 	default:
2915 		return (EINVAL);
2916 	}
2917 	pci_clear_command_bit(dev, child, bit);
2918 	return (0);
2919 }
2920 
2921 /*
2922  * New style pci driver.  Parent device is either a pci-host-bridge or a
2923  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2924  */
2925 
2926 void
2927 pci_print_verbose(struct pci_devinfo *dinfo)
2928 {
2929 
2930 	if (bootverbose) {
2931 		pcicfgregs *cfg = &dinfo->cfg;
2932 
2933 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2934 		    cfg->vendor, cfg->device, cfg->revid);
2935 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2936 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2937 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2938 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2939 		    cfg->mfdev);
2940 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2941 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2942 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2943 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2944 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2945 		if (cfg->intpin > 0)
2946 			printf("\tintpin=%c, irq=%d\n",
2947 			    cfg->intpin +'a' -1, cfg->intline);
2948 		if (cfg->pp.pp_cap) {
2949 			uint16_t status;
2950 
2951 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2952 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2953 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2954 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2955 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2956 			    status & PCIM_PSTAT_DMASK);
2957 		}
2958 		if (cfg->msi.msi_location) {
2959 			int ctrl;
2960 
2961 			ctrl = cfg->msi.msi_ctrl;
2962 			printf("\tMSI supports %d message%s%s%s\n",
2963 			    cfg->msi.msi_msgnum,
2964 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2965 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2966 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2967 		}
2968 		if (cfg->msix.msix_location) {
2969 			printf("\tMSI-X supports %d message%s ",
2970 			    cfg->msix.msix_msgnum,
2971 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2972 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2973 				printf("in map 0x%x\n",
2974 				    cfg->msix.msix_table_bar);
2975 			else
2976 				printf("in maps 0x%x and 0x%x\n",
2977 				    cfg->msix.msix_table_bar,
2978 				    cfg->msix.msix_pba_bar);
2979 		}
2980 	}
2981 }
2982 
2983 static int
2984 pci_porten(device_t dev)
2985 {
2986 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2987 }
2988 
2989 static int
2990 pci_memen(device_t dev)
2991 {
2992 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2993 }
2994 
2995 void
2996 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2997     int *bar64)
2998 {
2999 	struct pci_devinfo *dinfo;
3000 	pci_addr_t map, testval;
3001 	int ln2range;
3002 	uint16_t cmd;
3003 
3004 	/*
3005 	 * The device ROM BAR is special.  It is always a 32-bit
3006 	 * memory BAR.  Bit 0 is special and should not be set when
3007 	 * sizing the BAR.
3008 	 */
3009 	dinfo = device_get_ivars(dev);
3010 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3011 		map = pci_read_config(dev, reg, 4);
3012 		pci_write_config(dev, reg, 0xfffffffe, 4);
3013 		testval = pci_read_config(dev, reg, 4);
3014 		pci_write_config(dev, reg, map, 4);
3015 		*mapp = map;
3016 		*testvalp = testval;
3017 		if (bar64 != NULL)
3018 			*bar64 = 0;
3019 		return;
3020 	}
3021 
3022 	map = pci_read_config(dev, reg, 4);
3023 	ln2range = pci_maprange(map);
3024 	if (ln2range == 64)
3025 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3026 
3027 	/*
3028 	 * Disable decoding via the command register before
3029 	 * determining the BAR's length since we will be placing it in
3030 	 * a weird state.
3031 	 */
3032 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3033 	pci_write_config(dev, PCIR_COMMAND,
3034 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3035 
3036 	/*
3037 	 * Determine the BAR's length by writing all 1's.  The bottom
3038 	 * log_2(size) bits of the BAR will stick as 0 when we read
3039 	 * the value back.
3040 	 *
3041 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3042 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3043 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3044 	 *
3045 	 * Writes to both registers must be performed before attempting to
3046 	 * read back the size value.
3047 	 */
3048 	testval = 0;
3049 	pci_write_config(dev, reg, 0xffffffff, 4);
3050 	if (ln2range == 64) {
3051 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3052 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3053 	}
3054 	testval |= pci_read_config(dev, reg, 4);
3055 
3056 	/*
3057 	 * Restore the original value of the BAR.  We may have reprogrammed
3058 	 * the BAR of the low-level console device and when booting verbose,
3059 	 * we need the console device addressable.
3060 	 */
3061 	pci_write_config(dev, reg, map, 4);
3062 	if (ln2range == 64)
3063 		pci_write_config(dev, reg + 4, map >> 32, 4);
3064 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3065 
3066 	*mapp = map;
3067 	*testvalp = testval;
3068 	if (bar64 != NULL)
3069 		*bar64 = (ln2range == 64);
3070 }
3071 
3072 static void
3073 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3074 {
3075 	struct pci_devinfo *dinfo;
3076 	int ln2range;
3077 
3078 	/* The device ROM BAR is always a 32-bit memory BAR. */
3079 	dinfo = device_get_ivars(dev);
3080 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3081 		ln2range = 32;
3082 	else
3083 		ln2range = pci_maprange(pm->pm_value);
3084 	pci_write_config(dev, pm->pm_reg, base, 4);
3085 	if (ln2range == 64)
3086 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3087 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3088 	if (ln2range == 64)
3089 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3090 		    pm->pm_reg + 4, 4) << 32;
3091 }
3092 
3093 struct pci_map *
3094 pci_find_bar(device_t dev, int reg)
3095 {
3096 	struct pci_devinfo *dinfo;
3097 	struct pci_map *pm;
3098 
3099 	dinfo = device_get_ivars(dev);
3100 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3101 		if (pm->pm_reg == reg)
3102 			return (pm);
3103 	}
3104 	return (NULL);
3105 }
3106 
3107 int
3108 pci_bar_enabled(device_t dev, struct pci_map *pm)
3109 {
3110 	struct pci_devinfo *dinfo;
3111 	uint16_t cmd;
3112 
3113 	dinfo = device_get_ivars(dev);
3114 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3115 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3116 		return (0);
3117 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3118 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3119 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3120 	else
3121 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3122 }
3123 
3124 struct pci_map *
3125 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3126 {
3127 	struct pci_devinfo *dinfo;
3128 	struct pci_map *pm, *prev;
3129 
3130 	dinfo = device_get_ivars(dev);
3131 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3132 	pm->pm_reg = reg;
3133 	pm->pm_value = value;
3134 	pm->pm_size = size;
3135 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3136 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3137 		    reg));
3138 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3139 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3140 			break;
3141 	}
3142 	if (prev != NULL)
3143 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3144 	else
3145 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3146 	return (pm);
3147 }
3148 
3149 static void
3150 pci_restore_bars(device_t dev)
3151 {
3152 	struct pci_devinfo *dinfo;
3153 	struct pci_map *pm;
3154 	int ln2range;
3155 
3156 	dinfo = device_get_ivars(dev);
3157 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3158 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3159 			ln2range = 32;
3160 		else
3161 			ln2range = pci_maprange(pm->pm_value);
3162 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3163 		if (ln2range == 64)
3164 			pci_write_config(dev, pm->pm_reg + 4,
3165 			    pm->pm_value >> 32, 4);
3166 	}
3167 }
3168 
3169 /*
3170  * Add a resource based on a pci map register. Return 1 if the map
3171  * register is a 32bit map register or 2 if it is a 64bit register.
3172  */
3173 static int
3174 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3175     int force, int prefetch)
3176 {
3177 	struct pci_map *pm;
3178 	pci_addr_t base, map, testval;
3179 	pci_addr_t start, end, count;
3180 	int barlen, basezero, flags, maprange, mapsize, type;
3181 	uint16_t cmd;
3182 	struct resource *res;
3183 
3184 	/*
3185 	 * The BAR may already exist if the device is a CardBus card
3186 	 * whose CIS is stored in this BAR.
3187 	 */
3188 	pm = pci_find_bar(dev, reg);
3189 	if (pm != NULL) {
3190 		maprange = pci_maprange(pm->pm_value);
3191 		barlen = maprange == 64 ? 2 : 1;
3192 		return (barlen);
3193 	}
3194 
3195 	pci_read_bar(dev, reg, &map, &testval, NULL);
3196 	if (PCI_BAR_MEM(map)) {
3197 		type = SYS_RES_MEMORY;
3198 		if (map & PCIM_BAR_MEM_PREFETCH)
3199 			prefetch = 1;
3200 	} else
3201 		type = SYS_RES_IOPORT;
3202 	mapsize = pci_mapsize(testval);
3203 	base = pci_mapbase(map);
3204 #ifdef __PCI_BAR_ZERO_VALID
3205 	basezero = 0;
3206 #else
3207 	basezero = base == 0;
3208 #endif
3209 	maprange = pci_maprange(map);
3210 	barlen = maprange == 64 ? 2 : 1;
3211 
3212 	/*
3213 	 * For I/O registers, if bottom bit is set, and the next bit up
3214 	 * isn't clear, we know we have a BAR that doesn't conform to the
3215 	 * spec, so ignore it.  Also, sanity check the size of the data
3216 	 * areas to the type of memory involved.  Memory must be at least
3217 	 * 16 bytes in size, while I/O ranges must be at least 4.
3218 	 */
3219 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3220 		return (barlen);
3221 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3222 	    (type == SYS_RES_IOPORT && mapsize < 2))
3223 		return (barlen);
3224 
3225 	/* Save a record of this BAR. */
3226 	pm = pci_add_bar(dev, reg, map, mapsize);
3227 	if (bootverbose) {
3228 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3229 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3230 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3231 			printf(", port disabled\n");
3232 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3233 			printf(", memory disabled\n");
3234 		else
3235 			printf(", enabled\n");
3236 	}
3237 
3238 	/*
3239 	 * If base is 0, then we have problems if this architecture does
3240 	 * not allow that.  It is best to ignore such entries for the
3241 	 * moment.  These will be allocated later if the driver specifically
3242 	 * requests them.  However, some removable buses look better when
3243 	 * all resources are allocated, so allow '0' to be overriden.
3244 	 *
3245 	 * Similarly treat maps whose values is the same as the test value
3246 	 * read back.  These maps have had all f's written to them by the
3247 	 * BIOS in an attempt to disable the resources.
3248 	 */
3249 	if (!force && (basezero || map == testval))
3250 		return (barlen);
3251 	if ((u_long)base != base) {
3252 		device_printf(bus,
3253 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3254 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3255 		    pci_get_function(dev), reg);
3256 		return (barlen);
3257 	}
3258 
3259 	/*
3260 	 * This code theoretically does the right thing, but has
3261 	 * undesirable side effects in some cases where peripherals
3262 	 * respond oddly to having these bits enabled.  Let the user
3263 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3264 	 * default).
3265 	 */
3266 	if (pci_enable_io_modes) {
3267 		/* Turn on resources that have been left off by a lazy BIOS */
3268 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3269 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3270 			cmd |= PCIM_CMD_PORTEN;
3271 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3272 		}
3273 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3274 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3275 			cmd |= PCIM_CMD_MEMEN;
3276 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3277 		}
3278 	} else {
3279 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3280 			return (barlen);
3281 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3282 			return (barlen);
3283 	}
3284 
3285 	count = (pci_addr_t)1 << mapsize;
3286 	flags = RF_ALIGNMENT_LOG2(mapsize);
3287 	if (prefetch)
3288 		flags |= RF_PREFETCHABLE;
3289 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3290 		start = 0;	/* Let the parent decide. */
3291 		end = ~0;
3292 	} else {
3293 		start = base;
3294 		end = base + count - 1;
3295 	}
3296 	resource_list_add(rl, type, reg, start, end, count);
3297 
3298 	/*
3299 	 * Try to allocate the resource for this BAR from our parent
3300 	 * so that this resource range is already reserved.  The
3301 	 * driver for this device will later inherit this resource in
3302 	 * pci_alloc_resource().
3303 	 */
3304 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3305 	    flags);
3306 	if ((pci_do_realloc_bars
3307 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3308 	    && res == NULL && (start != 0 || end != ~0)) {
3309 		/*
3310 		 * If the allocation fails, try to allocate a resource for
3311 		 * this BAR using any available range.  The firmware felt
3312 		 * it was important enough to assign a resource, so don't
3313 		 * disable decoding if we can help it.
3314 		 */
3315 		resource_list_delete(rl, type, reg);
3316 		resource_list_add(rl, type, reg, 0, ~0, count);
3317 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3318 		    count, flags);
3319 	}
3320 	if (res == NULL) {
3321 		/*
3322 		 * If the allocation fails, delete the resource list entry
3323 		 * and disable decoding for this device.
3324 		 *
3325 		 * If the driver requests this resource in the future,
3326 		 * pci_reserve_map() will try to allocate a fresh
3327 		 * resource range.
3328 		 */
3329 		resource_list_delete(rl, type, reg);
3330 		pci_disable_io(dev, type);
3331 		if (bootverbose)
3332 			device_printf(bus,
3333 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3334 			    pci_get_domain(dev), pci_get_bus(dev),
3335 			    pci_get_slot(dev), pci_get_function(dev), reg);
3336 	} else {
3337 		start = rman_get_start(res);
3338 		pci_write_bar(dev, pm, start);
3339 	}
3340 	return (barlen);
3341 }
3342 
3343 /*
3344  * For ATA devices we need to decide early what addressing mode to use.
3345  * Legacy demands that the primary and secondary ATA ports sits on the
3346  * same addresses that old ISA hardware did. This dictates that we use
3347  * those addresses and ignore the BAR's if we cannot set PCI native
3348  * addressing mode.
3349  */
3350 static void
3351 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3352     uint32_t prefetchmask)
3353 {
3354 	int rid, type, progif;
3355 #if 0
3356 	/* if this device supports PCI native addressing use it */
3357 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3358 	if ((progif & 0x8a) == 0x8a) {
3359 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3360 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3361 			printf("Trying ATA native PCI addressing mode\n");
3362 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3363 		}
3364 	}
3365 #endif
3366 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3367 	type = SYS_RES_IOPORT;
3368 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3369 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3370 		    prefetchmask & (1 << 0));
3371 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3372 		    prefetchmask & (1 << 1));
3373 	} else {
3374 		rid = PCIR_BAR(0);
3375 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3376 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3377 		    0x1f7, 8, 0);
3378 		rid = PCIR_BAR(1);
3379 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3380 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3381 		    0x3f6, 1, 0);
3382 	}
3383 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3384 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3385 		    prefetchmask & (1 << 2));
3386 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3387 		    prefetchmask & (1 << 3));
3388 	} else {
3389 		rid = PCIR_BAR(2);
3390 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3391 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3392 		    0x177, 8, 0);
3393 		rid = PCIR_BAR(3);
3394 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3395 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3396 		    0x376, 1, 0);
3397 	}
3398 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3399 	    prefetchmask & (1 << 4));
3400 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3401 	    prefetchmask & (1 << 5));
3402 }
3403 
3404 static void
3405 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3406 {
3407 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3408 	pcicfgregs *cfg = &dinfo->cfg;
3409 	char tunable_name[64];
3410 	int irq;
3411 
3412 	/* Has to have an intpin to have an interrupt. */
3413 	if (cfg->intpin == 0)
3414 		return;
3415 
3416 	/* Let the user override the IRQ with a tunable. */
3417 	irq = PCI_INVALID_IRQ;
3418 	snprintf(tunable_name, sizeof(tunable_name),
3419 	    "hw.pci%d.%d.%d.INT%c.irq",
3420 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3421 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3422 		irq = PCI_INVALID_IRQ;
3423 
3424 	/*
3425 	 * If we didn't get an IRQ via the tunable, then we either use the
3426 	 * IRQ value in the intline register or we ask the bus to route an
3427 	 * interrupt for us.  If force_route is true, then we only use the
3428 	 * value in the intline register if the bus was unable to assign an
3429 	 * IRQ.
3430 	 */
3431 	if (!PCI_INTERRUPT_VALID(irq)) {
3432 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3433 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3434 		if (!PCI_INTERRUPT_VALID(irq))
3435 			irq = cfg->intline;
3436 	}
3437 
3438 	/* If after all that we don't have an IRQ, just bail. */
3439 	if (!PCI_INTERRUPT_VALID(irq))
3440 		return;
3441 
3442 	/* Update the config register if it changed. */
3443 	if (irq != cfg->intline) {
3444 		cfg->intline = irq;
3445 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3446 	}
3447 
3448 	/* Add this IRQ as rid 0 interrupt resource. */
3449 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3450 }
3451 
3452 /* Perform early OHCI takeover from SMM. */
3453 static void
3454 ohci_early_takeover(device_t self)
3455 {
3456 	struct resource *res;
3457 	uint32_t ctl;
3458 	int rid;
3459 	int i;
3460 
3461 	rid = PCIR_BAR(0);
3462 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3463 	if (res == NULL)
3464 		return;
3465 
3466 	ctl = bus_read_4(res, OHCI_CONTROL);
3467 	if (ctl & OHCI_IR) {
3468 		if (bootverbose)
3469 			printf("ohci early: "
3470 			    "SMM active, request owner change\n");
3471 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3472 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3473 			DELAY(1000);
3474 			ctl = bus_read_4(res, OHCI_CONTROL);
3475 		}
3476 		if (ctl & OHCI_IR) {
3477 			if (bootverbose)
3478 				printf("ohci early: "
3479 				    "SMM does not respond, resetting\n");
3480 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3481 		}
3482 		/* Disable interrupts */
3483 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3484 	}
3485 
3486 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3487 }
3488 
3489 /* Perform early UHCI takeover from SMM. */
3490 static void
3491 uhci_early_takeover(device_t self)
3492 {
3493 	struct resource *res;
3494 	int rid;
3495 
3496 	/*
3497 	 * Set the PIRQD enable bit and switch off all the others. We don't
3498 	 * want legacy support to interfere with us XXX Does this also mean
3499 	 * that the BIOS won't touch the keyboard anymore if it is connected
3500 	 * to the ports of the root hub?
3501 	 */
3502 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3503 
3504 	/* Disable interrupts */
3505 	rid = PCI_UHCI_BASE_REG;
3506 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3507 	if (res != NULL) {
3508 		bus_write_2(res, UHCI_INTR, 0);
3509 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3510 	}
3511 }
3512 
3513 /* Perform early EHCI takeover from SMM. */
3514 static void
3515 ehci_early_takeover(device_t self)
3516 {
3517 	struct resource *res;
3518 	uint32_t cparams;
3519 	uint32_t eec;
3520 	uint8_t eecp;
3521 	uint8_t bios_sem;
3522 	uint8_t offs;
3523 	int rid;
3524 	int i;
3525 
3526 	rid = PCIR_BAR(0);
3527 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3528 	if (res == NULL)
3529 		return;
3530 
3531 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3532 
3533 	/* Synchronise with the BIOS if it owns the controller. */
3534 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3535 	    eecp = EHCI_EECP_NEXT(eec)) {
3536 		eec = pci_read_config(self, eecp, 4);
3537 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3538 			continue;
3539 		}
3540 		bios_sem = pci_read_config(self, eecp +
3541 		    EHCI_LEGSUP_BIOS_SEM, 1);
3542 		if (bios_sem == 0) {
3543 			continue;
3544 		}
3545 		if (bootverbose)
3546 			printf("ehci early: "
3547 			    "SMM active, request owner change\n");
3548 
3549 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3550 
3551 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3552 			DELAY(1000);
3553 			bios_sem = pci_read_config(self, eecp +
3554 			    EHCI_LEGSUP_BIOS_SEM, 1);
3555 		}
3556 
3557 		if (bios_sem != 0) {
3558 			if (bootverbose)
3559 				printf("ehci early: "
3560 				    "SMM does not respond\n");
3561 		}
3562 		/* Disable interrupts */
3563 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3564 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3565 	}
3566 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3567 }
3568 
3569 /* Perform early XHCI takeover from SMM. */
3570 static void
3571 xhci_early_takeover(device_t self)
3572 {
3573 	struct resource *res;
3574 	uint32_t cparams;
3575 	uint32_t eec;
3576 	uint8_t eecp;
3577 	uint8_t bios_sem;
3578 	uint8_t offs;
3579 	int rid;
3580 	int i;
3581 
3582 	rid = PCIR_BAR(0);
3583 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3584 	if (res == NULL)
3585 		return;
3586 
3587 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3588 
3589 	eec = -1;
3590 
3591 	/* Synchronise with the BIOS if it owns the controller. */
3592 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3593 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3594 		eec = bus_read_4(res, eecp);
3595 
3596 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3597 			continue;
3598 
3599 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3600 		if (bios_sem == 0)
3601 			continue;
3602 
3603 		if (bootverbose)
3604 			printf("xhci early: "
3605 			    "SMM active, request owner change\n");
3606 
3607 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3608 
3609 		/* wait a maximum of 5 second */
3610 
3611 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3612 			DELAY(1000);
3613 			bios_sem = bus_read_1(res, eecp +
3614 			    XHCI_XECP_BIOS_SEM);
3615 		}
3616 
3617 		if (bios_sem != 0) {
3618 			if (bootverbose)
3619 				printf("xhci early: "
3620 				    "SMM does not respond\n");
3621 		}
3622 
3623 		/* Disable interrupts */
3624 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3625 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3626 		bus_read_4(res, offs + XHCI_USBSTS);
3627 	}
3628 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3629 }
3630 
3631 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3632 static void
3633 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3634     struct resource_list *rl)
3635 {
3636 	struct resource *res;
3637 	char *cp;
3638 	rman_res_t start, end, count;
3639 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3640 
3641 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3642 	case PCIM_HDRTYPE_BRIDGE:
3643 		sec_reg = PCIR_SECBUS_1;
3644 		sub_reg = PCIR_SUBBUS_1;
3645 		break;
3646 	case PCIM_HDRTYPE_CARDBUS:
3647 		sec_reg = PCIR_SECBUS_2;
3648 		sub_reg = PCIR_SUBBUS_2;
3649 		break;
3650 	default:
3651 		return;
3652 	}
3653 
3654 	/*
3655 	 * If the existing bus range is valid, attempt to reserve it
3656 	 * from our parent.  If this fails for any reason, clear the
3657 	 * secbus and subbus registers.
3658 	 *
3659 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3660 	 * This would at least preserve the existing sec_bus if it is
3661 	 * valid.
3662 	 */
3663 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3664 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3665 
3666 	/* Quirk handling. */
3667 	switch (pci_get_devid(dev)) {
3668 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3669 		sup_bus = pci_read_config(dev, 0x41, 1);
3670 		if (sup_bus != 0xff) {
3671 			sec_bus = sup_bus + 1;
3672 			sub_bus = sup_bus + 1;
3673 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3674 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3675 		}
3676 		break;
3677 
3678 	case 0x00dd10de:
3679 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3680 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3681 			break;
3682 		if (strncmp(cp, "Compal", 6) != 0) {
3683 			freeenv(cp);
3684 			break;
3685 		}
3686 		freeenv(cp);
3687 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3688 			break;
3689 		if (strncmp(cp, "08A0", 4) != 0) {
3690 			freeenv(cp);
3691 			break;
3692 		}
3693 		freeenv(cp);
3694 		if (sub_bus < 0xa) {
3695 			sub_bus = 0xa;
3696 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3697 		}
3698 		break;
3699 	}
3700 
3701 	if (bootverbose)
3702 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3703 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3704 		start = sec_bus;
3705 		end = sub_bus;
3706 		count = end - start + 1;
3707 
3708 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3709 
3710 		/*
3711 		 * If requested, clear secondary bus registers in
3712 		 * bridge devices to force a complete renumbering
3713 		 * rather than reserving the existing range.  However,
3714 		 * preserve the existing size.
3715 		 */
3716 		if (pci_clear_buses)
3717 			goto clear;
3718 
3719 		rid = 0;
3720 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3721 		    start, end, count, 0);
3722 		if (res != NULL)
3723 			return;
3724 
3725 		if (bootverbose)
3726 			device_printf(bus,
3727 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3728 			    pci_get_domain(dev), pci_get_bus(dev),
3729 			    pci_get_slot(dev), pci_get_function(dev));
3730 	}
3731 
3732 clear:
3733 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3734 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3735 }
3736 
3737 static struct resource *
3738 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3739     rman_res_t end, rman_res_t count, u_int flags)
3740 {
3741 	struct pci_devinfo *dinfo;
3742 	pcicfgregs *cfg;
3743 	struct resource_list *rl;
3744 	struct resource *res;
3745 	int sec_reg, sub_reg;
3746 
3747 	dinfo = device_get_ivars(child);
3748 	cfg = &dinfo->cfg;
3749 	rl = &dinfo->resources;
3750 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3751 	case PCIM_HDRTYPE_BRIDGE:
3752 		sec_reg = PCIR_SECBUS_1;
3753 		sub_reg = PCIR_SUBBUS_1;
3754 		break;
3755 	case PCIM_HDRTYPE_CARDBUS:
3756 		sec_reg = PCIR_SECBUS_2;
3757 		sub_reg = PCIR_SUBBUS_2;
3758 		break;
3759 	default:
3760 		return (NULL);
3761 	}
3762 
3763 	if (*rid != 0)
3764 		return (NULL);
3765 
3766 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3767 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3768 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3769 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3770 		    start, end, count, flags & ~RF_ACTIVE);
3771 		if (res == NULL) {
3772 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3773 			device_printf(child, "allocating %ju bus%s failed\n",
3774 			    count, count == 1 ? "" : "es");
3775 			return (NULL);
3776 		}
3777 		if (bootverbose)
3778 			device_printf(child,
3779 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3780 			    count == 1 ? "" : "es", rman_get_start(res));
3781 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3782 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3783 	}
3784 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3785 	    end, count, flags));
3786 }
3787 #endif
3788 
3789 static int
3790 pci_ea_bei_to_rid(device_t dev, int bei)
3791 {
3792 #ifdef PCI_IOV
3793 	struct pci_devinfo *dinfo;
3794 	int iov_pos;
3795 	struct pcicfg_iov *iov;
3796 
3797 	dinfo = device_get_ivars(dev);
3798 	iov = dinfo->cfg.iov;
3799 	if (iov != NULL)
3800 		iov_pos = iov->iov_pos;
3801 	else
3802 		iov_pos = 0;
3803 #endif
3804 
3805 	/* Check if matches BAR */
3806 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3807 	    (bei <= PCIM_EA_BEI_BAR_5))
3808 		return (PCIR_BAR(bei));
3809 
3810 	/* Check ROM */
3811 	if (bei == PCIM_EA_BEI_ROM)
3812 		return (PCIR_BIOS);
3813 
3814 #ifdef PCI_IOV
3815 	/* Check if matches VF_BAR */
3816 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3817 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3818 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3819 		    iov_pos);
3820 #endif
3821 
3822 	return (-1);
3823 }
3824 
3825 int
3826 pci_ea_is_enabled(device_t dev, int rid)
3827 {
3828 	struct pci_ea_entry *ea;
3829 	struct pci_devinfo *dinfo;
3830 
3831 	dinfo = device_get_ivars(dev);
3832 
3833 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3834 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3835 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3836 	}
3837 
3838 	return (0);
3839 }
3840 
3841 void
3842 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3843 {
3844 	struct pci_ea_entry *ea;
3845 	struct pci_devinfo *dinfo;
3846 	pci_addr_t start, end, count;
3847 	struct resource_list *rl;
3848 	int type, flags, rid;
3849 	struct resource *res;
3850 	uint32_t tmp;
3851 #ifdef PCI_IOV
3852 	struct pcicfg_iov *iov;
3853 #endif
3854 
3855 	dinfo = device_get_ivars(dev);
3856 	rl = &dinfo->resources;
3857 	flags = 0;
3858 
3859 #ifdef PCI_IOV
3860 	iov = dinfo->cfg.iov;
3861 #endif
3862 
3863 	if (dinfo->cfg.ea.ea_location == 0)
3864 		return;
3865 
3866 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3867 
3868 		/*
3869 		 * TODO: Ignore EA-BAR if is not enabled.
3870 		 *   Currently the EA implementation supports
3871 		 *   only situation, where EA structure contains
3872 		 *   predefined entries. In case they are not enabled
3873 		 *   leave them unallocated and proceed with
3874 		 *   a legacy-BAR mechanism.
3875 		 */
3876 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3877 			continue;
3878 
3879 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3880 		case PCIM_EA_P_MEM_PREFETCH:
3881 		case PCIM_EA_P_VF_MEM_PREFETCH:
3882 			flags = RF_PREFETCHABLE;
3883 			/* FALLTHROUGH */
3884 		case PCIM_EA_P_VF_MEM:
3885 		case PCIM_EA_P_MEM:
3886 			type = SYS_RES_MEMORY;
3887 			break;
3888 		case PCIM_EA_P_IO:
3889 			type = SYS_RES_IOPORT;
3890 			break;
3891 		default:
3892 			continue;
3893 		}
3894 
3895 		if (alloc_iov != 0) {
3896 #ifdef PCI_IOV
3897 			/* Allocating IOV, confirm BEI matches */
3898 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3899 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3900 				continue;
3901 #else
3902 			continue;
3903 #endif
3904 		} else {
3905 			/* Allocating BAR, confirm BEI matches */
3906 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3907 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3908 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3909 				continue;
3910 		}
3911 
3912 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3913 		if (rid < 0)
3914 			continue;
3915 
3916 		/* Skip resources already allocated by EA */
3917 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3918 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3919 			continue;
3920 
3921 		start = ea->eae_base;
3922 		count = ea->eae_max_offset + 1;
3923 #ifdef PCI_IOV
3924 		if (iov != NULL)
3925 			count = count * iov->iov_num_vfs;
3926 #endif
3927 		end = start + count - 1;
3928 		if (count == 0)
3929 			continue;
3930 
3931 		resource_list_add(rl, type, rid, start, end, count);
3932 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3933 		    flags);
3934 		if (res == NULL) {
3935 			resource_list_delete(rl, type, rid);
3936 
3937 			/*
3938 			 * Failed to allocate using EA, disable entry.
3939 			 * Another attempt to allocation will be performed
3940 			 * further, but this time using legacy BAR registers
3941 			 */
3942 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3943 			tmp &= ~PCIM_EA_ENABLE;
3944 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3945 
3946 			/*
3947 			 * Disabling entry might fail in case it is hardwired.
3948 			 * Read flags again to match current status.
3949 			 */
3950 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3951 
3952 			continue;
3953 		}
3954 
3955 		/* As per specification, fill BAR with zeros */
3956 		pci_write_config(dev, rid, 0, 4);
3957 	}
3958 }
3959 
3960 void
3961 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3962 {
3963 	struct pci_devinfo *dinfo;
3964 	pcicfgregs *cfg;
3965 	struct resource_list *rl;
3966 	const struct pci_quirk *q;
3967 	uint32_t devid;
3968 	int i;
3969 
3970 	dinfo = device_get_ivars(dev);
3971 	cfg = &dinfo->cfg;
3972 	rl = &dinfo->resources;
3973 	devid = (cfg->device << 16) | cfg->vendor;
3974 
3975 	/* Allocate resources using Enhanced Allocation */
3976 	pci_add_resources_ea(bus, dev, 0);
3977 
3978 	/* ATA devices needs special map treatment */
3979 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3980 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3981 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3982 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3983 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3984 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3985 	else
3986 		for (i = 0; i < cfg->nummaps;) {
3987 			/* Skip resources already managed by EA */
3988 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3989 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3990 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3991 				i++;
3992 				continue;
3993 			}
3994 
3995 			/*
3996 			 * Skip quirked resources.
3997 			 */
3998 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3999 				if (q->devid == devid &&
4000 				    q->type == PCI_QUIRK_UNMAP_REG &&
4001 				    q->arg1 == PCIR_BAR(i))
4002 					break;
4003 			if (q->devid != 0) {
4004 				i++;
4005 				continue;
4006 			}
4007 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4008 			    prefetchmask & (1 << i));
4009 		}
4010 
4011 	/*
4012 	 * Add additional, quirked resources.
4013 	 */
4014 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4015 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4016 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4017 
4018 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4019 #ifdef __PCI_REROUTE_INTERRUPT
4020 		/*
4021 		 * Try to re-route interrupts. Sometimes the BIOS or
4022 		 * firmware may leave bogus values in these registers.
4023 		 * If the re-route fails, then just stick with what we
4024 		 * have.
4025 		 */
4026 		pci_assign_interrupt(bus, dev, 1);
4027 #else
4028 		pci_assign_interrupt(bus, dev, 0);
4029 #endif
4030 	}
4031 
4032 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4033 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4034 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4035 			xhci_early_takeover(dev);
4036 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4037 			ehci_early_takeover(dev);
4038 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4039 			ohci_early_takeover(dev);
4040 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4041 			uhci_early_takeover(dev);
4042 	}
4043 
4044 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4045 	/*
4046 	 * Reserve resources for secondary bus ranges behind bridge
4047 	 * devices.
4048 	 */
4049 	pci_reserve_secbus(bus, dev, cfg, rl);
4050 #endif
4051 }
4052 
4053 static struct pci_devinfo *
4054 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4055     int slot, int func)
4056 {
4057 	struct pci_devinfo *dinfo;
4058 
4059 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4060 	if (dinfo != NULL)
4061 		pci_add_child(dev, dinfo);
4062 
4063 	return (dinfo);
4064 }
4065 
4066 void
4067 pci_add_children(device_t dev, int domain, int busno)
4068 {
4069 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4070 	device_t pcib = device_get_parent(dev);
4071 	struct pci_devinfo *dinfo;
4072 	int maxslots;
4073 	int s, f, pcifunchigh;
4074 	uint8_t hdrtype;
4075 	int first_func;
4076 
4077 	/*
4078 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4079 	 * enable ARI.  We must enable ARI before detecting the rest of the
4080 	 * functions on this bus as ARI changes the set of slots and functions
4081 	 * that are legal on this bus.
4082 	 */
4083 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4084 	if (dinfo != NULL && pci_enable_ari)
4085 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4086 
4087 	/*
4088 	 * Start looking for new devices on slot 0 at function 1 because we
4089 	 * just identified the device at slot 0, function 0.
4090 	 */
4091 	first_func = 1;
4092 
4093 	maxslots = PCIB_MAXSLOTS(pcib);
4094 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4095 		pcifunchigh = 0;
4096 		f = 0;
4097 		DELAY(1);
4098 		hdrtype = REG(PCIR_HDRTYPE, 1);
4099 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4100 			continue;
4101 		if (hdrtype & PCIM_MFDEV)
4102 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4103 		for (f = first_func; f <= pcifunchigh; f++)
4104 			pci_identify_function(pcib, dev, domain, busno, s, f);
4105 	}
4106 #undef REG
4107 }
4108 
4109 int
4110 pci_rescan_method(device_t dev)
4111 {
4112 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4113 	device_t pcib = device_get_parent(dev);
4114 	device_t child, *devlist, *unchanged;
4115 	int devcount, error, i, j, maxslots, oldcount;
4116 	int busno, domain, s, f, pcifunchigh;
4117 	uint8_t hdrtype;
4118 
4119 	/* No need to check for ARI on a rescan. */
4120 	error = device_get_children(dev, &devlist, &devcount);
4121 	if (error)
4122 		return (error);
4123 	if (devcount != 0) {
4124 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4125 		    M_NOWAIT | M_ZERO);
4126 		if (unchanged == NULL) {
4127 			free(devlist, M_TEMP);
4128 			return (ENOMEM);
4129 		}
4130 	} else
4131 		unchanged = NULL;
4132 
4133 	domain = pcib_get_domain(dev);
4134 	busno = pcib_get_bus(dev);
4135 	maxslots = PCIB_MAXSLOTS(pcib);
4136 	for (s = 0; s <= maxslots; s++) {
4137 		/* If function 0 is not present, skip to the next slot. */
4138 		f = 0;
4139 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4140 			continue;
4141 		pcifunchigh = 0;
4142 		hdrtype = REG(PCIR_HDRTYPE, 1);
4143 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4144 			continue;
4145 		if (hdrtype & PCIM_MFDEV)
4146 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4147 		for (f = 0; f <= pcifunchigh; f++) {
4148 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4149 				continue;
4150 
4151 			/*
4152 			 * Found a valid function.  Check if a
4153 			 * device_t for this device already exists.
4154 			 */
4155 			for (i = 0; i < devcount; i++) {
4156 				child = devlist[i];
4157 				if (child == NULL)
4158 					continue;
4159 				if (pci_get_slot(child) == s &&
4160 				    pci_get_function(child) == f) {
4161 					unchanged[i] = child;
4162 					goto next_func;
4163 				}
4164 			}
4165 
4166 			pci_identify_function(pcib, dev, domain, busno, s, f);
4167 		next_func:;
4168 		}
4169 	}
4170 
4171 	/* Remove devices that are no longer present. */
4172 	for (i = 0; i < devcount; i++) {
4173 		if (unchanged[i] != NULL)
4174 			continue;
4175 		device_delete_child(dev, devlist[i]);
4176 	}
4177 
4178 	free(devlist, M_TEMP);
4179 	oldcount = devcount;
4180 
4181 	/* Try to attach the devices just added. */
4182 	error = device_get_children(dev, &devlist, &devcount);
4183 	if (error) {
4184 		free(unchanged, M_TEMP);
4185 		return (error);
4186 	}
4187 
4188 	for (i = 0; i < devcount; i++) {
4189 		for (j = 0; j < oldcount; j++) {
4190 			if (devlist[i] == unchanged[j])
4191 				goto next_device;
4192 		}
4193 
4194 		device_probe_and_attach(devlist[i]);
4195 	next_device:;
4196 	}
4197 
4198 	free(unchanged, M_TEMP);
4199 	free(devlist, M_TEMP);
4200 	return (0);
4201 #undef REG
4202 }
4203 
4204 #ifdef PCI_IOV
4205 device_t
4206 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4207     uint16_t did)
4208 {
4209 	struct pci_devinfo *vf_dinfo;
4210 	device_t pcib;
4211 	int busno, slot, func;
4212 
4213 	pcib = device_get_parent(bus);
4214 
4215 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4216 
4217 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4218 	    slot, func, vid, did);
4219 
4220 	vf_dinfo->cfg.flags |= PCICFG_VF;
4221 	pci_add_child(bus, vf_dinfo);
4222 
4223 	return (vf_dinfo->cfg.dev);
4224 }
4225 
4226 device_t
4227 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4228     uint16_t vid, uint16_t did)
4229 {
4230 
4231 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4232 }
4233 #endif
4234 
4235 static void
4236 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4237 {
4238 	int aer;
4239 	uint32_t r;
4240 	uint16_t r2;
4241 
4242 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4243 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4244 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4245 		    PCIER_ROOT_CTL, 2);
4246 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4247 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4248 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4249 		    PCIER_ROOT_CTL, r2, 2);
4250 	}
4251 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4252 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4253 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4254 		if (r != 0 && bootverbose) {
4255 			pci_printf(&dinfo->cfg,
4256 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4257 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4258 			    4));
4259 		}
4260 
4261 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4262 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4263 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4264 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4265 		    PCIM_AER_UC_POISONED_TLP |
4266 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4267 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4268 		    PCIM_AER_UC_COMPLETER_ABORT |
4269 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4270 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4271 		    PCIM_AER_UC_MALFORMED_TLP |
4272 		    PCIM_AER_UC_ECRC_ERROR |
4273 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4274 		    PCIM_AER_UC_ACS_VIOLATION |
4275 		    PCIM_AER_UC_INTERNAL_ERROR |
4276 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4277 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4278 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4279 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4280 
4281 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4282 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4283 		if (r != 0 && bootverbose) {
4284 			pci_printf(&dinfo->cfg,
4285 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4286 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4287 			    4));
4288 		}
4289 
4290 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4291 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4292 		    PCIM_AER_COR_BAD_TLP |
4293 		    PCIM_AER_COR_BAD_DLLP |
4294 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4295 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4296 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4297 		    PCIM_AER_COR_INTERNAL_ERROR |
4298 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4299 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4300 
4301 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4302 		    PCIER_DEVICE_CTL, 2);
4303 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4304 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4305 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4306 		    PCIER_DEVICE_CTL, r, 2);
4307 	}
4308 }
4309 
4310 void
4311 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4312 {
4313 	device_t dev;
4314 
4315 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4316 	device_set_ivars(dev, dinfo);
4317 	resource_list_init(&dinfo->resources);
4318 	pci_cfg_save(dev, dinfo, 0);
4319 	pci_cfg_restore(dev, dinfo);
4320 	pci_print_verbose(dinfo);
4321 	pci_add_resources(bus, dev, 0, 0);
4322 	pci_child_added(dinfo->cfg.dev);
4323 
4324 	if (pci_clear_aer_on_attach)
4325 		pci_add_child_clear_aer(dev, dinfo);
4326 
4327 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4328 }
4329 
4330 void
4331 pci_child_added_method(device_t dev, device_t child)
4332 {
4333 
4334 }
4335 
4336 static int
4337 pci_probe(device_t dev)
4338 {
4339 
4340 	device_set_desc(dev, "PCI bus");
4341 
4342 	/* Allow other subclasses to override this driver. */
4343 	return (BUS_PROBE_GENERIC);
4344 }
4345 
4346 int
4347 pci_attach_common(device_t dev)
4348 {
4349 	struct pci_softc *sc;
4350 	int busno, domain;
4351 #ifdef PCI_RES_BUS
4352 	int rid;
4353 #endif
4354 
4355 	sc = device_get_softc(dev);
4356 	domain = pcib_get_domain(dev);
4357 	busno = pcib_get_bus(dev);
4358 #ifdef PCI_RES_BUS
4359 	rid = 0;
4360 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4361 	    1, 0);
4362 	if (sc->sc_bus == NULL) {
4363 		device_printf(dev, "failed to allocate bus number\n");
4364 		return (ENXIO);
4365 	}
4366 #endif
4367 	if (bootverbose)
4368 		device_printf(dev, "domain=%d, physical bus=%d\n",
4369 		    domain, busno);
4370 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4371 	return (0);
4372 }
4373 
4374 int
4375 pci_attach(device_t dev)
4376 {
4377 	int busno, domain, error;
4378 
4379 	error = pci_attach_common(dev);
4380 	if (error)
4381 		return (error);
4382 
4383 	/*
4384 	 * Since there can be multiple independently numbered PCI
4385 	 * buses on systems with multiple PCI domains, we can't use
4386 	 * the unit number to decide which bus we are probing. We ask
4387 	 * the parent pcib what our domain and bus numbers are.
4388 	 */
4389 	domain = pcib_get_domain(dev);
4390 	busno = pcib_get_bus(dev);
4391 	pci_add_children(dev, domain, busno);
4392 	return (bus_generic_attach(dev));
4393 }
4394 
4395 int
4396 pci_detach(device_t dev)
4397 {
4398 #ifdef PCI_RES_BUS
4399 	struct pci_softc *sc;
4400 #endif
4401 	int error;
4402 
4403 	error = bus_generic_detach(dev);
4404 	if (error)
4405 		return (error);
4406 #ifdef PCI_RES_BUS
4407 	sc = device_get_softc(dev);
4408 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4409 	if (error)
4410 		return (error);
4411 #endif
4412 	return (device_delete_children(dev));
4413 }
4414 
4415 static void
4416 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4417 {
4418 	int line, unit;
4419 	const char *at;
4420 	char me1[24], me2[32];
4421 	uint8_t b, s, f;
4422 	uint32_t d;
4423 
4424 	d = pci_get_domain(child);
4425 	b = pci_get_bus(child);
4426 	s = pci_get_slot(child);
4427 	f = pci_get_function(child);
4428 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4429 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4430 	line = 0;
4431 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4432 		resource_string_value(name, unit, "at", &at);
4433 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4434 			continue; /* No match, try next candidate */
4435 		*unitp = unit;
4436 		return;
4437 	}
4438 }
4439 
4440 static void
4441 pci_set_power_child(device_t dev, device_t child, int state)
4442 {
4443 	device_t pcib;
4444 	int dstate;
4445 
4446 	/*
4447 	 * Set the device to the given state.  If the firmware suggests
4448 	 * a different power state, use it instead.  If power management
4449 	 * is not present, the firmware is responsible for managing
4450 	 * device power.  Skip children who aren't attached since they
4451 	 * are handled separately.
4452 	 */
4453 	pcib = device_get_parent(dev);
4454 	dstate = state;
4455 	if (device_is_attached(child) &&
4456 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4457 		pci_set_powerstate(child, dstate);
4458 }
4459 
4460 int
4461 pci_suspend_child(device_t dev, device_t child)
4462 {
4463 	struct pci_devinfo *dinfo;
4464 	struct resource_list_entry *rle;
4465 	int error;
4466 
4467 	dinfo = device_get_ivars(child);
4468 
4469 	/*
4470 	 * Save the PCI configuration space for the child and set the
4471 	 * device in the appropriate power state for this sleep state.
4472 	 */
4473 	pci_cfg_save(child, dinfo, 0);
4474 
4475 	/* Suspend devices before potentially powering them down. */
4476 	error = bus_generic_suspend_child(dev, child);
4477 
4478 	if (error)
4479 		return (error);
4480 
4481 	if (pci_do_power_suspend) {
4482 		/*
4483 		 * Make sure this device's interrupt handler is not invoked
4484 		 * in the case the device uses a shared interrupt that can
4485 		 * be raised by some other device.
4486 		 * This is applicable only to regular (legacy) PCI interrupts
4487 		 * as MSI/MSI-X interrupts are never shared.
4488 		 */
4489 		rle = resource_list_find(&dinfo->resources,
4490 		    SYS_RES_IRQ, 0);
4491 		if (rle != NULL && rle->res != NULL)
4492 			(void)bus_suspend_intr(child, rle->res);
4493 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4494 	}
4495 
4496 	return (0);
4497 }
4498 
4499 int
4500 pci_resume_child(device_t dev, device_t child)
4501 {
4502 	struct pci_devinfo *dinfo;
4503 	struct resource_list_entry *rle;
4504 
4505 	if (pci_do_power_resume)
4506 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4507 
4508 	dinfo = device_get_ivars(child);
4509 	pci_cfg_restore(child, dinfo);
4510 	if (!device_is_attached(child))
4511 		pci_cfg_save(child, dinfo, 1);
4512 
4513 	bus_generic_resume_child(dev, child);
4514 
4515 	/*
4516 	 * Allow interrupts only after fully resuming the driver and hardware.
4517 	 */
4518 	if (pci_do_power_suspend) {
4519 		/* See pci_suspend_child for details. */
4520 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4521 		if (rle != NULL && rle->res != NULL)
4522 			(void)bus_resume_intr(child, rle->res);
4523 	}
4524 
4525 	return (0);
4526 }
4527 
4528 int
4529 pci_resume(device_t dev)
4530 {
4531 	device_t child, *devlist;
4532 	int error, i, numdevs;
4533 
4534 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4535 		return (error);
4536 
4537 	/*
4538 	 * Resume critical devices first, then everything else later.
4539 	 */
4540 	for (i = 0; i < numdevs; i++) {
4541 		child = devlist[i];
4542 		switch (pci_get_class(child)) {
4543 		case PCIC_DISPLAY:
4544 		case PCIC_MEMORY:
4545 		case PCIC_BRIDGE:
4546 		case PCIC_BASEPERIPH:
4547 			BUS_RESUME_CHILD(dev, child);
4548 			break;
4549 		}
4550 	}
4551 	for (i = 0; i < numdevs; i++) {
4552 		child = devlist[i];
4553 		switch (pci_get_class(child)) {
4554 		case PCIC_DISPLAY:
4555 		case PCIC_MEMORY:
4556 		case PCIC_BRIDGE:
4557 		case PCIC_BASEPERIPH:
4558 			break;
4559 		default:
4560 			BUS_RESUME_CHILD(dev, child);
4561 		}
4562 	}
4563 	free(devlist, M_TEMP);
4564 	return (0);
4565 }
4566 
4567 static void
4568 pci_load_vendor_data(void)
4569 {
4570 	caddr_t data;
4571 	void *ptr;
4572 	size_t sz;
4573 
4574 	data = preload_search_by_type("pci_vendor_data");
4575 	if (data != NULL) {
4576 		ptr = preload_fetch_addr(data);
4577 		sz = preload_fetch_size(data);
4578 		if (ptr != NULL && sz != 0) {
4579 			pci_vendordata = ptr;
4580 			pci_vendordata_size = sz;
4581 			/* terminate the database */
4582 			pci_vendordata[pci_vendordata_size] = '\n';
4583 		}
4584 	}
4585 }
4586 
4587 void
4588 pci_driver_added(device_t dev, driver_t *driver)
4589 {
4590 	int numdevs;
4591 	device_t *devlist;
4592 	device_t child;
4593 	struct pci_devinfo *dinfo;
4594 	int i;
4595 
4596 	if (bootverbose)
4597 		device_printf(dev, "driver added\n");
4598 	DEVICE_IDENTIFY(driver, dev);
4599 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4600 		return;
4601 	for (i = 0; i < numdevs; i++) {
4602 		child = devlist[i];
4603 		if (device_get_state(child) != DS_NOTPRESENT)
4604 			continue;
4605 		dinfo = device_get_ivars(child);
4606 		pci_print_verbose(dinfo);
4607 		if (bootverbose)
4608 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4609 		pci_cfg_restore(child, dinfo);
4610 		if (device_probe_and_attach(child) != 0)
4611 			pci_child_detached(dev, child);
4612 	}
4613 	free(devlist, M_TEMP);
4614 }
4615 
4616 int
4617 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4618     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4619 {
4620 	struct pci_devinfo *dinfo;
4621 	struct msix_table_entry *mte;
4622 	struct msix_vector *mv;
4623 	uint64_t addr;
4624 	uint32_t data;
4625 	void *cookie;
4626 	int error, rid;
4627 
4628 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4629 	    arg, &cookie);
4630 	if (error)
4631 		return (error);
4632 
4633 	/* If this is not a direct child, just bail out. */
4634 	if (device_get_parent(child) != dev) {
4635 		*cookiep = cookie;
4636 		return(0);
4637 	}
4638 
4639 	rid = rman_get_rid(irq);
4640 	if (rid == 0) {
4641 		/* Make sure that INTx is enabled */
4642 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4643 	} else {
4644 		/*
4645 		 * Check to see if the interrupt is MSI or MSI-X.
4646 		 * Ask our parent to map the MSI and give
4647 		 * us the address and data register values.
4648 		 * If we fail for some reason, teardown the
4649 		 * interrupt handler.
4650 		 */
4651 		dinfo = device_get_ivars(child);
4652 		if (dinfo->cfg.msi.msi_alloc > 0) {
4653 			if (dinfo->cfg.msi.msi_addr == 0) {
4654 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4655 			    ("MSI has handlers, but vectors not mapped"));
4656 				error = PCIB_MAP_MSI(device_get_parent(dev),
4657 				    child, rman_get_start(irq), &addr, &data);
4658 				if (error)
4659 					goto bad;
4660 				dinfo->cfg.msi.msi_addr = addr;
4661 				dinfo->cfg.msi.msi_data = data;
4662 			}
4663 			if (dinfo->cfg.msi.msi_handlers == 0)
4664 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4665 				    dinfo->cfg.msi.msi_data);
4666 			dinfo->cfg.msi.msi_handlers++;
4667 		} else {
4668 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4669 			    ("No MSI or MSI-X interrupts allocated"));
4670 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4671 			    ("MSI-X index too high"));
4672 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4673 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4674 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4675 			KASSERT(mv->mv_irq == rman_get_start(irq),
4676 			    ("IRQ mismatch"));
4677 			if (mv->mv_address == 0) {
4678 				KASSERT(mte->mte_handlers == 0,
4679 		    ("MSI-X table entry has handlers, but vector not mapped"));
4680 				error = PCIB_MAP_MSI(device_get_parent(dev),
4681 				    child, rman_get_start(irq), &addr, &data);
4682 				if (error)
4683 					goto bad;
4684 				mv->mv_address = addr;
4685 				mv->mv_data = data;
4686 			}
4687 
4688 			/*
4689 			 * The MSIX table entry must be made valid by
4690 			 * incrementing the mte_handlers before
4691 			 * calling pci_enable_msix() and
4692 			 * pci_resume_msix(). Else the MSIX rewrite
4693 			 * table quirk will not work as expected.
4694 			 */
4695 			mte->mte_handlers++;
4696 			if (mte->mte_handlers == 1) {
4697 				pci_enable_msix(child, rid - 1, mv->mv_address,
4698 				    mv->mv_data);
4699 				pci_unmask_msix(child, rid - 1);
4700 			}
4701 		}
4702 
4703 		/*
4704 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4705 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4706 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4707 		 */
4708 		if (!pci_has_quirk(pci_get_devid(child),
4709 		    PCI_QUIRK_MSI_INTX_BUG))
4710 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4711 		else
4712 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4713 	bad:
4714 		if (error) {
4715 			(void)bus_generic_teardown_intr(dev, child, irq,
4716 			    cookie);
4717 			return (error);
4718 		}
4719 	}
4720 	*cookiep = cookie;
4721 	return (0);
4722 }
4723 
4724 int
4725 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4726     void *cookie)
4727 {
4728 	struct msix_table_entry *mte;
4729 	struct resource_list_entry *rle;
4730 	struct pci_devinfo *dinfo;
4731 	int error, rid;
4732 
4733 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4734 		return (EINVAL);
4735 
4736 	/* If this isn't a direct child, just bail out */
4737 	if (device_get_parent(child) != dev)
4738 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4739 
4740 	rid = rman_get_rid(irq);
4741 	if (rid == 0) {
4742 		/* Mask INTx */
4743 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4744 	} else {
4745 		/*
4746 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4747 		 * decrement the appropriate handlers count and mask the
4748 		 * MSI-X message, or disable MSI messages if the count
4749 		 * drops to 0.
4750 		 */
4751 		dinfo = device_get_ivars(child);
4752 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4753 		if (rle->res != irq)
4754 			return (EINVAL);
4755 		if (dinfo->cfg.msi.msi_alloc > 0) {
4756 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4757 			    ("MSI-X index too high"));
4758 			if (dinfo->cfg.msi.msi_handlers == 0)
4759 				return (EINVAL);
4760 			dinfo->cfg.msi.msi_handlers--;
4761 			if (dinfo->cfg.msi.msi_handlers == 0)
4762 				pci_disable_msi(child);
4763 		} else {
4764 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4765 			    ("No MSI or MSI-X interrupts allocated"));
4766 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4767 			    ("MSI-X index too high"));
4768 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4769 			if (mte->mte_handlers == 0)
4770 				return (EINVAL);
4771 			mte->mte_handlers--;
4772 			if (mte->mte_handlers == 0)
4773 				pci_mask_msix(child, rid - 1);
4774 		}
4775 	}
4776 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4777 	if (rid > 0)
4778 		KASSERT(error == 0,
4779 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4780 	return (error);
4781 }
4782 
4783 int
4784 pci_print_child(device_t dev, device_t child)
4785 {
4786 	struct pci_devinfo *dinfo;
4787 	struct resource_list *rl;
4788 	int retval = 0;
4789 
4790 	dinfo = device_get_ivars(child);
4791 	rl = &dinfo->resources;
4792 
4793 	retval += bus_print_child_header(dev, child);
4794 
4795 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4796 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4797 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4798 	if (device_get_flags(dev))
4799 		retval += printf(" flags %#x", device_get_flags(dev));
4800 
4801 	retval += printf(" at device %d.%d", pci_get_slot(child),
4802 	    pci_get_function(child));
4803 
4804 	retval += bus_print_child_domain(dev, child);
4805 	retval += bus_print_child_footer(dev, child);
4806 
4807 	return (retval);
4808 }
4809 
4810 static const struct
4811 {
4812 	int		class;
4813 	int		subclass;
4814 	int		report; /* 0 = bootverbose, 1 = always */
4815 	const char	*desc;
4816 } pci_nomatch_tab[] = {
4817 	{PCIC_OLD,		-1,			1, "old"},
4818 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4819 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4820 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4821 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4822 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4823 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4824 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4825 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4826 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4827 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4828 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4829 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4830 	{PCIC_NETWORK,		-1,			1, "network"},
4831 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4832 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4833 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4834 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4835 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4836 	{PCIC_DISPLAY,		-1,			1, "display"},
4837 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4838 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4839 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4840 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4841 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4842 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4843 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4844 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4845 	{PCIC_MEMORY,		-1,			1, "memory"},
4846 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4847 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4848 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4849 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4850 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4851 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4852 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4853 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4854 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4855 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4856 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4857 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4858 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4859 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4860 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4861 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4862 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4863 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4864 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4865 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4866 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4867 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4868 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4869 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4870 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4871 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4872 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4873 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4874 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4875 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4876 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4877 	{PCIC_DOCKING,		-1,			1, "docking station"},
4878 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4879 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4880 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4881 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4882 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4883 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4884 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4885 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4886 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4887 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4888 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4889 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4890 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4891 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4892 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4893 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4894 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4895 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4896 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4897 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4898 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4899 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4900 	{PCIC_DASP,		-1,			0, "dasp"},
4901 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4902 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4903 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4904 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4905 	{0, 0, 0,		NULL}
4906 };
4907 
4908 void
4909 pci_probe_nomatch(device_t dev, device_t child)
4910 {
4911 	int i, report;
4912 	const char *cp, *scp;
4913 	char *device;
4914 
4915 	/*
4916 	 * Look for a listing for this device in a loaded device database.
4917 	 */
4918 	report = 1;
4919 	if ((device = pci_describe_device(child)) != NULL) {
4920 		device_printf(dev, "<%s>", device);
4921 		free(device, M_DEVBUF);
4922 	} else {
4923 		/*
4924 		 * Scan the class/subclass descriptions for a general
4925 		 * description.
4926 		 */
4927 		cp = "unknown";
4928 		scp = NULL;
4929 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4930 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4931 				if (pci_nomatch_tab[i].subclass == -1) {
4932 					cp = pci_nomatch_tab[i].desc;
4933 					report = pci_nomatch_tab[i].report;
4934 				} else if (pci_nomatch_tab[i].subclass ==
4935 				    pci_get_subclass(child)) {
4936 					scp = pci_nomatch_tab[i].desc;
4937 					report = pci_nomatch_tab[i].report;
4938 				}
4939 			}
4940 		}
4941 		if (report || bootverbose) {
4942 			device_printf(dev, "<%s%s%s>",
4943 			    cp ? cp : "",
4944 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4945 			    scp ? scp : "");
4946 		}
4947 	}
4948 	if (report || bootverbose) {
4949 		printf(" at device %d.%d (no driver attached)\n",
4950 		    pci_get_slot(child), pci_get_function(child));
4951 	}
4952 	pci_cfg_save(child, device_get_ivars(child), 1);
4953 }
4954 
4955 void
4956 pci_child_detached(device_t dev, device_t child)
4957 {
4958 	struct pci_devinfo *dinfo;
4959 	struct resource_list *rl;
4960 
4961 	dinfo = device_get_ivars(child);
4962 	rl = &dinfo->resources;
4963 
4964 	/*
4965 	 * Have to deallocate IRQs before releasing any MSI messages and
4966 	 * have to release MSI messages before deallocating any memory
4967 	 * BARs.
4968 	 */
4969 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4970 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4971 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4972 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4973 		(void)pci_release_msi(child);
4974 	}
4975 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4976 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4977 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4978 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4979 #ifdef PCI_RES_BUS
4980 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4981 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4982 #endif
4983 
4984 	pci_cfg_save(child, dinfo, 1);
4985 }
4986 
4987 /*
4988  * Parse the PCI device database, if loaded, and return a pointer to a
4989  * description of the device.
4990  *
4991  * The database is flat text formatted as follows:
4992  *
4993  * Any line not in a valid format is ignored.
4994  * Lines are terminated with newline '\n' characters.
4995  *
4996  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4997  * the vendor name.
4998  *
4999  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5000  * - devices cannot be listed without a corresponding VENDOR line.
5001  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5002  * another TAB, then the device name.
5003  */
5004 
5005 /*
5006  * Assuming (ptr) points to the beginning of a line in the database,
5007  * return the vendor or device and description of the next entry.
5008  * The value of (vendor) or (device) inappropriate for the entry type
5009  * is set to -1.  Returns nonzero at the end of the database.
5010  *
5011  * Note that this is slightly unrobust in the face of corrupt data;
5012  * we attempt to safeguard against this by spamming the end of the
5013  * database with a newline when we initialise.
5014  */
5015 static int
5016 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5017 {
5018 	char	*cp = *ptr;
5019 	int	left;
5020 
5021 	*device = -1;
5022 	*vendor = -1;
5023 	**desc = '\0';
5024 	for (;;) {
5025 		left = pci_vendordata_size - (cp - pci_vendordata);
5026 		if (left <= 0) {
5027 			*ptr = cp;
5028 			return(1);
5029 		}
5030 
5031 		/* vendor entry? */
5032 		if (*cp != '\t' &&
5033 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5034 			break;
5035 		/* device entry? */
5036 		if (*cp == '\t' &&
5037 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5038 			break;
5039 
5040 		/* skip to next line */
5041 		while (*cp != '\n' && left > 0) {
5042 			cp++;
5043 			left--;
5044 		}
5045 		if (*cp == '\n') {
5046 			cp++;
5047 			left--;
5048 		}
5049 	}
5050 	/* skip to next line */
5051 	while (*cp != '\n' && left > 0) {
5052 		cp++;
5053 		left--;
5054 	}
5055 	if (*cp == '\n' && left > 0)
5056 		cp++;
5057 	*ptr = cp;
5058 	return(0);
5059 }
5060 
5061 static char *
5062 pci_describe_device(device_t dev)
5063 {
5064 	int	vendor, device;
5065 	char	*desc, *vp, *dp, *line;
5066 
5067 	desc = vp = dp = NULL;
5068 
5069 	/*
5070 	 * If we have no vendor data, we can't do anything.
5071 	 */
5072 	if (pci_vendordata == NULL)
5073 		goto out;
5074 
5075 	/*
5076 	 * Scan the vendor data looking for this device
5077 	 */
5078 	line = pci_vendordata;
5079 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5080 		goto out;
5081 	for (;;) {
5082 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5083 			goto out;
5084 		if (vendor == pci_get_vendor(dev))
5085 			break;
5086 	}
5087 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5088 		goto out;
5089 	for (;;) {
5090 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5091 			*dp = 0;
5092 			break;
5093 		}
5094 		if (vendor != -1) {
5095 			*dp = 0;
5096 			break;
5097 		}
5098 		if (device == pci_get_device(dev))
5099 			break;
5100 	}
5101 	if (dp[0] == '\0')
5102 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5103 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5104 	    NULL)
5105 		sprintf(desc, "%s, %s", vp, dp);
5106 out:
5107 	if (vp != NULL)
5108 		free(vp, M_DEVBUF);
5109 	if (dp != NULL)
5110 		free(dp, M_DEVBUF);
5111 	return(desc);
5112 }
5113 
5114 int
5115 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5116 {
5117 	struct pci_devinfo *dinfo;
5118 	pcicfgregs *cfg;
5119 
5120 	dinfo = device_get_ivars(child);
5121 	cfg = &dinfo->cfg;
5122 
5123 	switch (which) {
5124 	case PCI_IVAR_ETHADDR:
5125 		/*
5126 		 * The generic accessor doesn't deal with failure, so
5127 		 * we set the return value, then return an error.
5128 		 */
5129 		*((uint8_t **) result) = NULL;
5130 		return (EINVAL);
5131 	case PCI_IVAR_SUBVENDOR:
5132 		*result = cfg->subvendor;
5133 		break;
5134 	case PCI_IVAR_SUBDEVICE:
5135 		*result = cfg->subdevice;
5136 		break;
5137 	case PCI_IVAR_VENDOR:
5138 		*result = cfg->vendor;
5139 		break;
5140 	case PCI_IVAR_DEVICE:
5141 		*result = cfg->device;
5142 		break;
5143 	case PCI_IVAR_DEVID:
5144 		*result = (cfg->device << 16) | cfg->vendor;
5145 		break;
5146 	case PCI_IVAR_CLASS:
5147 		*result = cfg->baseclass;
5148 		break;
5149 	case PCI_IVAR_SUBCLASS:
5150 		*result = cfg->subclass;
5151 		break;
5152 	case PCI_IVAR_PROGIF:
5153 		*result = cfg->progif;
5154 		break;
5155 	case PCI_IVAR_REVID:
5156 		*result = cfg->revid;
5157 		break;
5158 	case PCI_IVAR_INTPIN:
5159 		*result = cfg->intpin;
5160 		break;
5161 	case PCI_IVAR_IRQ:
5162 		*result = cfg->intline;
5163 		break;
5164 	case PCI_IVAR_DOMAIN:
5165 		*result = cfg->domain;
5166 		break;
5167 	case PCI_IVAR_BUS:
5168 		*result = cfg->bus;
5169 		break;
5170 	case PCI_IVAR_SLOT:
5171 		*result = cfg->slot;
5172 		break;
5173 	case PCI_IVAR_FUNCTION:
5174 		*result = cfg->func;
5175 		break;
5176 	case PCI_IVAR_CMDREG:
5177 		*result = cfg->cmdreg;
5178 		break;
5179 	case PCI_IVAR_CACHELNSZ:
5180 		*result = cfg->cachelnsz;
5181 		break;
5182 	case PCI_IVAR_MINGNT:
5183 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5184 			*result = -1;
5185 			return (EINVAL);
5186 		}
5187 		*result = cfg->mingnt;
5188 		break;
5189 	case PCI_IVAR_MAXLAT:
5190 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5191 			*result = -1;
5192 			return (EINVAL);
5193 		}
5194 		*result = cfg->maxlat;
5195 		break;
5196 	case PCI_IVAR_LATTIMER:
5197 		*result = cfg->lattimer;
5198 		break;
5199 	default:
5200 		return (ENOENT);
5201 	}
5202 	return (0);
5203 }
5204 
5205 int
5206 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5207 {
5208 	struct pci_devinfo *dinfo;
5209 
5210 	dinfo = device_get_ivars(child);
5211 
5212 	switch (which) {
5213 	case PCI_IVAR_INTPIN:
5214 		dinfo->cfg.intpin = value;
5215 		return (0);
5216 	case PCI_IVAR_ETHADDR:
5217 	case PCI_IVAR_SUBVENDOR:
5218 	case PCI_IVAR_SUBDEVICE:
5219 	case PCI_IVAR_VENDOR:
5220 	case PCI_IVAR_DEVICE:
5221 	case PCI_IVAR_DEVID:
5222 	case PCI_IVAR_CLASS:
5223 	case PCI_IVAR_SUBCLASS:
5224 	case PCI_IVAR_PROGIF:
5225 	case PCI_IVAR_REVID:
5226 	case PCI_IVAR_IRQ:
5227 	case PCI_IVAR_DOMAIN:
5228 	case PCI_IVAR_BUS:
5229 	case PCI_IVAR_SLOT:
5230 	case PCI_IVAR_FUNCTION:
5231 		return (EINVAL);	/* disallow for now */
5232 
5233 	default:
5234 		return (ENOENT);
5235 	}
5236 }
5237 
5238 #include "opt_ddb.h"
5239 #ifdef DDB
5240 #include <ddb/ddb.h>
5241 #include <sys/cons.h>
5242 
5243 /*
5244  * List resources based on pci map registers, used for within ddb
5245  */
5246 
5247 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5248 {
5249 	struct pci_devinfo *dinfo;
5250 	struct devlist *devlist_head;
5251 	struct pci_conf *p;
5252 	const char *name;
5253 	int i, error, none_count;
5254 
5255 	none_count = 0;
5256 	/* get the head of the device queue */
5257 	devlist_head = &pci_devq;
5258 
5259 	/*
5260 	 * Go through the list of devices and print out devices
5261 	 */
5262 	for (error = 0, i = 0,
5263 	     dinfo = STAILQ_FIRST(devlist_head);
5264 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5265 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5266 
5267 		/* Populate pd_name and pd_unit */
5268 		name = NULL;
5269 		if (dinfo->cfg.dev)
5270 			name = device_get_name(dinfo->cfg.dev);
5271 
5272 		p = &dinfo->conf;
5273 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5274 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5275 			(name && *name) ? name : "none",
5276 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5277 			none_count++,
5278 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5279 			p->pc_sel.pc_func, (p->pc_class << 16) |
5280 			(p->pc_subclass << 8) | p->pc_progif,
5281 			(p->pc_subdevice << 16) | p->pc_subvendor,
5282 			(p->pc_device << 16) | p->pc_vendor,
5283 			p->pc_revid, p->pc_hdr);
5284 	}
5285 }
5286 #endif /* DDB */
5287 
5288 static struct resource *
5289 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5290     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5291     u_int flags)
5292 {
5293 	struct pci_devinfo *dinfo = device_get_ivars(child);
5294 	struct resource_list *rl = &dinfo->resources;
5295 	struct resource *res;
5296 	struct pci_map *pm;
5297 	uint16_t cmd;
5298 	pci_addr_t map, testval;
5299 	int mapsize;
5300 
5301 	res = NULL;
5302 
5303 	/* If rid is managed by EA, ignore it */
5304 	if (pci_ea_is_enabled(child, *rid))
5305 		goto out;
5306 
5307 	pm = pci_find_bar(child, *rid);
5308 	if (pm != NULL) {
5309 		/* This is a BAR that we failed to allocate earlier. */
5310 		mapsize = pm->pm_size;
5311 		map = pm->pm_value;
5312 	} else {
5313 		/*
5314 		 * Weed out the bogons, and figure out how large the
5315 		 * BAR/map is.  BARs that read back 0 here are bogus
5316 		 * and unimplemented.  Note: atapci in legacy mode are
5317 		 * special and handled elsewhere in the code.  If you
5318 		 * have a atapci device in legacy mode and it fails
5319 		 * here, that other code is broken.
5320 		 */
5321 		pci_read_bar(child, *rid, &map, &testval, NULL);
5322 
5323 		/*
5324 		 * Determine the size of the BAR and ignore BARs with a size
5325 		 * of 0.  Device ROM BARs use a different mask value.
5326 		 */
5327 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5328 			mapsize = pci_romsize(testval);
5329 		else
5330 			mapsize = pci_mapsize(testval);
5331 		if (mapsize == 0)
5332 			goto out;
5333 		pm = pci_add_bar(child, *rid, map, mapsize);
5334 	}
5335 
5336 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5337 		if (type != SYS_RES_MEMORY) {
5338 			if (bootverbose)
5339 				device_printf(dev,
5340 				    "child %s requested type %d for rid %#x,"
5341 				    " but the BAR says it is an memio\n",
5342 				    device_get_nameunit(child), type, *rid);
5343 			goto out;
5344 		}
5345 	} else {
5346 		if (type != SYS_RES_IOPORT) {
5347 			if (bootverbose)
5348 				device_printf(dev,
5349 				    "child %s requested type %d for rid %#x,"
5350 				    " but the BAR says it is an ioport\n",
5351 				    device_get_nameunit(child), type, *rid);
5352 			goto out;
5353 		}
5354 	}
5355 
5356 	/*
5357 	 * For real BARs, we need to override the size that
5358 	 * the driver requests, because that's what the BAR
5359 	 * actually uses and we would otherwise have a
5360 	 * situation where we might allocate the excess to
5361 	 * another driver, which won't work.
5362 	 */
5363 	count = ((pci_addr_t)1 << mapsize) * num;
5364 	if (RF_ALIGNMENT(flags) < mapsize)
5365 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5366 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5367 		flags |= RF_PREFETCHABLE;
5368 
5369 	/*
5370 	 * Allocate enough resource, and then write back the
5371 	 * appropriate BAR for that resource.
5372 	 */
5373 	resource_list_add(rl, type, *rid, start, end, count);
5374 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5375 	    count, flags & ~RF_ACTIVE);
5376 	if (res == NULL) {
5377 		resource_list_delete(rl, type, *rid);
5378 		device_printf(child,
5379 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5380 		    count, *rid, type, start, end);
5381 		goto out;
5382 	}
5383 	if (bootverbose)
5384 		device_printf(child,
5385 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5386 		    count, *rid, type, rman_get_start(res));
5387 
5388 	/* Disable decoding via the CMD register before updating the BAR */
5389 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5390 	pci_write_config(child, PCIR_COMMAND,
5391 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5392 
5393 	map = rman_get_start(res);
5394 	pci_write_bar(child, pm, map);
5395 
5396 	/* Restore the original value of the CMD register */
5397 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5398 out:
5399 	return (res);
5400 }
5401 
5402 struct resource *
5403 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5404     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5405     u_int flags)
5406 {
5407 	struct pci_devinfo *dinfo;
5408 	struct resource_list *rl;
5409 	struct resource_list_entry *rle;
5410 	struct resource *res;
5411 	pcicfgregs *cfg;
5412 
5413 	/*
5414 	 * Perform lazy resource allocation
5415 	 */
5416 	dinfo = device_get_ivars(child);
5417 	rl = &dinfo->resources;
5418 	cfg = &dinfo->cfg;
5419 	switch (type) {
5420 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5421 	case PCI_RES_BUS:
5422 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5423 		    flags));
5424 #endif
5425 	case SYS_RES_IRQ:
5426 		/*
5427 		 * Can't alloc legacy interrupt once MSI messages have
5428 		 * been allocated.
5429 		 */
5430 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5431 		    cfg->msix.msix_alloc > 0))
5432 			return (NULL);
5433 
5434 		/*
5435 		 * If the child device doesn't have an interrupt
5436 		 * routed and is deserving of an interrupt, try to
5437 		 * assign it one.
5438 		 */
5439 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5440 		    (cfg->intpin != 0))
5441 			pci_assign_interrupt(dev, child, 0);
5442 		break;
5443 	case SYS_RES_IOPORT:
5444 	case SYS_RES_MEMORY:
5445 #ifdef NEW_PCIB
5446 		/*
5447 		 * PCI-PCI bridge I/O window resources are not BARs.
5448 		 * For those allocations just pass the request up the
5449 		 * tree.
5450 		 */
5451 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5452 			switch (*rid) {
5453 			case PCIR_IOBASEL_1:
5454 			case PCIR_MEMBASE_1:
5455 			case PCIR_PMBASEL_1:
5456 				/*
5457 				 * XXX: Should we bother creating a resource
5458 				 * list entry?
5459 				 */
5460 				return (bus_generic_alloc_resource(dev, child,
5461 				    type, rid, start, end, count, flags));
5462 			}
5463 		}
5464 #endif
5465 		/* Reserve resources for this BAR if needed. */
5466 		rle = resource_list_find(rl, type, *rid);
5467 		if (rle == NULL) {
5468 			res = pci_reserve_map(dev, child, type, rid, start, end,
5469 			    count, num, flags);
5470 			if (res == NULL)
5471 				return (NULL);
5472 		}
5473 	}
5474 	return (resource_list_alloc(rl, dev, child, type, rid,
5475 	    start, end, count, flags));
5476 }
5477 
5478 struct resource *
5479 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5480     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5481 {
5482 #ifdef PCI_IOV
5483 	struct pci_devinfo *dinfo;
5484 #endif
5485 
5486 	if (device_get_parent(child) != dev)
5487 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5488 		    type, rid, start, end, count, flags));
5489 
5490 #ifdef PCI_IOV
5491 	dinfo = device_get_ivars(child);
5492 	if (dinfo->cfg.flags & PCICFG_VF) {
5493 		switch (type) {
5494 		/* VFs can't have I/O BARs. */
5495 		case SYS_RES_IOPORT:
5496 			return (NULL);
5497 		case SYS_RES_MEMORY:
5498 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5499 			    start, end, count, flags));
5500 		}
5501 
5502 		/* Fall through for other types of resource allocations. */
5503 	}
5504 #endif
5505 
5506 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5507 	    count, 1, flags));
5508 }
5509 
5510 int
5511 pci_release_resource(device_t dev, device_t child, int type, int rid,
5512     struct resource *r)
5513 {
5514 	struct pci_devinfo *dinfo;
5515 	struct resource_list *rl;
5516 	pcicfgregs *cfg;
5517 
5518 	if (device_get_parent(child) != dev)
5519 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5520 		    type, rid, r));
5521 
5522 	dinfo = device_get_ivars(child);
5523 	cfg = &dinfo->cfg;
5524 
5525 #ifdef PCI_IOV
5526 	if (dinfo->cfg.flags & PCICFG_VF) {
5527 		switch (type) {
5528 		/* VFs can't have I/O BARs. */
5529 		case SYS_RES_IOPORT:
5530 			return (EDOOFUS);
5531 		case SYS_RES_MEMORY:
5532 			return (pci_vf_release_mem_resource(dev, child, rid,
5533 			    r));
5534 		}
5535 
5536 		/* Fall through for other types of resource allocations. */
5537 	}
5538 #endif
5539 
5540 #ifdef NEW_PCIB
5541 	/*
5542 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5543 	 * those allocations just pass the request up the tree.
5544 	 */
5545 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5546 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5547 		switch (rid) {
5548 		case PCIR_IOBASEL_1:
5549 		case PCIR_MEMBASE_1:
5550 		case PCIR_PMBASEL_1:
5551 			return (bus_generic_release_resource(dev, child, type,
5552 			    rid, r));
5553 		}
5554 	}
5555 #endif
5556 
5557 	rl = &dinfo->resources;
5558 	return (resource_list_release(rl, dev, child, type, rid, r));
5559 }
5560 
5561 int
5562 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5563     struct resource *r)
5564 {
5565 	struct pci_devinfo *dinfo;
5566 	int error;
5567 
5568 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5569 	if (error)
5570 		return (error);
5571 
5572 	/* Enable decoding in the command register when activating BARs. */
5573 	if (device_get_parent(child) == dev) {
5574 		/* Device ROMs need their decoding explicitly enabled. */
5575 		dinfo = device_get_ivars(child);
5576 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5577 			pci_write_bar(child, pci_find_bar(child, rid),
5578 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5579 		switch (type) {
5580 		case SYS_RES_IOPORT:
5581 		case SYS_RES_MEMORY:
5582 			error = PCI_ENABLE_IO(dev, child, type);
5583 			break;
5584 		}
5585 	}
5586 	return (error);
5587 }
5588 
5589 int
5590 pci_deactivate_resource(device_t dev, device_t child, int type,
5591     int rid, struct resource *r)
5592 {
5593 	struct pci_devinfo *dinfo;
5594 	int error;
5595 
5596 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5597 	if (error)
5598 		return (error);
5599 
5600 	/* Disable decoding for device ROMs. */
5601 	if (device_get_parent(child) == dev) {
5602 		dinfo = device_get_ivars(child);
5603 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5604 			pci_write_bar(child, pci_find_bar(child, rid),
5605 			    rman_get_start(r));
5606 	}
5607 	return (0);
5608 }
5609 
5610 void
5611 pci_child_deleted(device_t dev, device_t child)
5612 {
5613 	struct resource_list_entry *rle;
5614 	struct resource_list *rl;
5615 	struct pci_devinfo *dinfo;
5616 
5617 	dinfo = device_get_ivars(child);
5618 	rl = &dinfo->resources;
5619 
5620 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5621 
5622 	/* Turn off access to resources we're about to free */
5623 	if (bus_child_present(child) != 0) {
5624 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5625 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5626 
5627 		pci_disable_busmaster(child);
5628 	}
5629 
5630 	/* Free all allocated resources */
5631 	STAILQ_FOREACH(rle, rl, link) {
5632 		if (rle->res) {
5633 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5634 			    resource_list_busy(rl, rle->type, rle->rid)) {
5635 				pci_printf(&dinfo->cfg,
5636 				    "Resource still owned, oops. "
5637 				    "(type=%d, rid=%d, addr=%lx)\n",
5638 				    rle->type, rle->rid,
5639 				    rman_get_start(rle->res));
5640 				bus_release_resource(child, rle->type, rle->rid,
5641 				    rle->res);
5642 			}
5643 			resource_list_unreserve(rl, dev, child, rle->type,
5644 			    rle->rid);
5645 		}
5646 	}
5647 	resource_list_free(rl);
5648 
5649 	pci_freecfg(dinfo);
5650 }
5651 
5652 void
5653 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5654 {
5655 	struct pci_devinfo *dinfo;
5656 	struct resource_list *rl;
5657 	struct resource_list_entry *rle;
5658 
5659 	if (device_get_parent(child) != dev)
5660 		return;
5661 
5662 	dinfo = device_get_ivars(child);
5663 	rl = &dinfo->resources;
5664 	rle = resource_list_find(rl, type, rid);
5665 	if (rle == NULL)
5666 		return;
5667 
5668 	if (rle->res) {
5669 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5670 		    resource_list_busy(rl, type, rid)) {
5671 			device_printf(dev, "delete_resource: "
5672 			    "Resource still owned by child, oops. "
5673 			    "(type=%d, rid=%d, addr=%jx)\n",
5674 			    type, rid, rman_get_start(rle->res));
5675 			return;
5676 		}
5677 		resource_list_unreserve(rl, dev, child, type, rid);
5678 	}
5679 	resource_list_delete(rl, type, rid);
5680 }
5681 
5682 struct resource_list *
5683 pci_get_resource_list (device_t dev, device_t child)
5684 {
5685 	struct pci_devinfo *dinfo = device_get_ivars(child);
5686 
5687 	return (&dinfo->resources);
5688 }
5689 
5690 #ifdef ACPI_DMAR
5691 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5692 bus_dma_tag_t
5693 pci_get_dma_tag(device_t bus, device_t dev)
5694 {
5695 	bus_dma_tag_t tag;
5696 	struct pci_softc *sc;
5697 
5698 	if (device_get_parent(dev) == bus) {
5699 		/* try dmar and return if it works */
5700 		tag = dmar_get_dma_tag(bus, dev);
5701 	} else
5702 		tag = NULL;
5703 	if (tag == NULL) {
5704 		sc = device_get_softc(bus);
5705 		tag = sc->sc_dma_tag;
5706 	}
5707 	return (tag);
5708 }
5709 #else
5710 bus_dma_tag_t
5711 pci_get_dma_tag(device_t bus, device_t dev)
5712 {
5713 	struct pci_softc *sc = device_get_softc(bus);
5714 
5715 	return (sc->sc_dma_tag);
5716 }
5717 #endif
5718 
5719 uint32_t
5720 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5721 {
5722 	struct pci_devinfo *dinfo = device_get_ivars(child);
5723 	pcicfgregs *cfg = &dinfo->cfg;
5724 
5725 #ifdef PCI_IOV
5726 	/*
5727 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5728 	 * emulate them here.
5729 	 */
5730 	if (cfg->flags & PCICFG_VF) {
5731 		if (reg == PCIR_VENDOR) {
5732 			switch (width) {
5733 			case 4:
5734 				return (cfg->device << 16 | cfg->vendor);
5735 			case 2:
5736 				return (cfg->vendor);
5737 			case 1:
5738 				return (cfg->vendor & 0xff);
5739 			default:
5740 				return (0xffffffff);
5741 			}
5742 		} else if (reg == PCIR_DEVICE) {
5743 			switch (width) {
5744 			/* Note that an unaligned 4-byte read is an error. */
5745 			case 2:
5746 				return (cfg->device);
5747 			case 1:
5748 				return (cfg->device & 0xff);
5749 			default:
5750 				return (0xffffffff);
5751 			}
5752 		}
5753 	}
5754 #endif
5755 
5756 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5757 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5758 }
5759 
5760 void
5761 pci_write_config_method(device_t dev, device_t child, int reg,
5762     uint32_t val, int width)
5763 {
5764 	struct pci_devinfo *dinfo = device_get_ivars(child);
5765 	pcicfgregs *cfg = &dinfo->cfg;
5766 
5767 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5768 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5769 }
5770 
5771 int
5772 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5773     size_t buflen)
5774 {
5775 
5776 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5777 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5778 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5779 	return (0);
5780 }
5781 
5782 int
5783 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5784     size_t buflen)
5785 {
5786 	struct pci_devinfo *dinfo;
5787 	pcicfgregs *cfg;
5788 
5789 	dinfo = device_get_ivars(child);
5790 	cfg = &dinfo->cfg;
5791 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5792 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5793 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5794 	    cfg->progif);
5795 	return (0);
5796 }
5797 
5798 int
5799 pci_assign_interrupt_method(device_t dev, device_t child)
5800 {
5801 	struct pci_devinfo *dinfo = device_get_ivars(child);
5802 	pcicfgregs *cfg = &dinfo->cfg;
5803 
5804 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5805 	    cfg->intpin));
5806 }
5807 
5808 static void
5809 pci_lookup(void *arg, const char *name, device_t *dev)
5810 {
5811 	long val;
5812 	char *end;
5813 	int domain, bus, slot, func;
5814 
5815 	if (*dev != NULL)
5816 		return;
5817 
5818 	/*
5819 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5820 	 * pciB:S:F.  In the latter case, the domain is assumed to
5821 	 * be zero.
5822 	 */
5823 	if (strncmp(name, "pci", 3) != 0)
5824 		return;
5825 	val = strtol(name + 3, &end, 10);
5826 	if (val < 0 || val > INT_MAX || *end != ':')
5827 		return;
5828 	domain = val;
5829 	val = strtol(end + 1, &end, 10);
5830 	if (val < 0 || val > INT_MAX || *end != ':')
5831 		return;
5832 	bus = val;
5833 	val = strtol(end + 1, &end, 10);
5834 	if (val < 0 || val > INT_MAX)
5835 		return;
5836 	slot = val;
5837 	if (*end == ':') {
5838 		val = strtol(end + 1, &end, 10);
5839 		if (val < 0 || val > INT_MAX || *end != '\0')
5840 			return;
5841 		func = val;
5842 	} else if (*end == '\0') {
5843 		func = slot;
5844 		slot = bus;
5845 		bus = domain;
5846 		domain = 0;
5847 	} else
5848 		return;
5849 
5850 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5851 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5852 		return;
5853 
5854 	*dev = pci_find_dbsf(domain, bus, slot, func);
5855 }
5856 
5857 static int
5858 pci_modevent(module_t mod, int what, void *arg)
5859 {
5860 	static struct cdev *pci_cdev;
5861 	static eventhandler_tag tag;
5862 
5863 	switch (what) {
5864 	case MOD_LOAD:
5865 		STAILQ_INIT(&pci_devq);
5866 		pci_generation = 0;
5867 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5868 		    "pci");
5869 		pci_load_vendor_data();
5870 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5871 		    1000);
5872 		break;
5873 
5874 	case MOD_UNLOAD:
5875 		if (tag != NULL)
5876 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5877 		destroy_dev(pci_cdev);
5878 		break;
5879 	}
5880 
5881 	return (0);
5882 }
5883 
5884 static void
5885 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5886 {
5887 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5888 	struct pcicfg_pcie *cfg;
5889 	int version, pos;
5890 
5891 	cfg = &dinfo->cfg.pcie;
5892 	pos = cfg->pcie_location;
5893 
5894 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5895 
5896 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5897 
5898 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5899 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5900 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5901 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5902 
5903 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5904 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5905 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5906 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5907 
5908 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5909 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5910 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5911 
5912 	if (version > 1) {
5913 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5914 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5915 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5916 	}
5917 #undef WREG
5918 }
5919 
5920 static void
5921 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5922 {
5923 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5924 	    dinfo->cfg.pcix.pcix_command,  2);
5925 }
5926 
5927 void
5928 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5929 {
5930 
5931 	/*
5932 	 * Restore the device to full power mode.  We must do this
5933 	 * before we restore the registers because moving from D3 to
5934 	 * D0 will cause the chip's BARs and some other registers to
5935 	 * be reset to some unknown power on reset values.  Cut down
5936 	 * the noise on boot by doing nothing if we are already in
5937 	 * state D0.
5938 	 */
5939 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5940 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5941 	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5942 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5943 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5944 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5945 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5946 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5947 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5948 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5949 	case PCIM_HDRTYPE_NORMAL:
5950 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5951 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5952 		break;
5953 	case PCIM_HDRTYPE_BRIDGE:
5954 		pci_write_config(dev, PCIR_SECLAT_1,
5955 		    dinfo->cfg.bridge.br_seclat, 1);
5956 		pci_write_config(dev, PCIR_SUBBUS_1,
5957 		    dinfo->cfg.bridge.br_subbus, 1);
5958 		pci_write_config(dev, PCIR_SECBUS_1,
5959 		    dinfo->cfg.bridge.br_secbus, 1);
5960 		pci_write_config(dev, PCIR_PRIBUS_1,
5961 		    dinfo->cfg.bridge.br_pribus, 1);
5962 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5963 		    dinfo->cfg.bridge.br_control, 2);
5964 		break;
5965 	case PCIM_HDRTYPE_CARDBUS:
5966 		pci_write_config(dev, PCIR_SECLAT_2,
5967 		    dinfo->cfg.bridge.br_seclat, 1);
5968 		pci_write_config(dev, PCIR_SUBBUS_2,
5969 		    dinfo->cfg.bridge.br_subbus, 1);
5970 		pci_write_config(dev, PCIR_SECBUS_2,
5971 		    dinfo->cfg.bridge.br_secbus, 1);
5972 		pci_write_config(dev, PCIR_PRIBUS_2,
5973 		    dinfo->cfg.bridge.br_pribus, 1);
5974 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5975 		    dinfo->cfg.bridge.br_control, 2);
5976 		break;
5977 	}
5978 	pci_restore_bars(dev);
5979 
5980 	/*
5981 	 * Restore extended capabilities for PCI-Express and PCI-X
5982 	 */
5983 	if (dinfo->cfg.pcie.pcie_location != 0)
5984 		pci_cfg_restore_pcie(dev, dinfo);
5985 	if (dinfo->cfg.pcix.pcix_location != 0)
5986 		pci_cfg_restore_pcix(dev, dinfo);
5987 
5988 	/* Restore MSI and MSI-X configurations if they are present. */
5989 	if (dinfo->cfg.msi.msi_location != 0)
5990 		pci_resume_msi(dev);
5991 	if (dinfo->cfg.msix.msix_location != 0)
5992 		pci_resume_msix(dev);
5993 
5994 #ifdef PCI_IOV
5995 	if (dinfo->cfg.iov != NULL)
5996 		pci_iov_cfg_restore(dev, dinfo);
5997 #endif
5998 }
5999 
6000 static void
6001 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6002 {
6003 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6004 	struct pcicfg_pcie *cfg;
6005 	int version, pos;
6006 
6007 	cfg = &dinfo->cfg.pcie;
6008 	pos = cfg->pcie_location;
6009 
6010 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6011 
6012 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6013 
6014 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6015 
6016 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6017 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6018 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6019 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6020 
6021 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6022 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6023 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6024 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6025 
6026 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6027 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6028 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6029 
6030 	if (version > 1) {
6031 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6032 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6033 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6034 	}
6035 #undef RREG
6036 }
6037 
6038 static void
6039 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6040 {
6041 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6042 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6043 }
6044 
6045 void
6046 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6047 {
6048 	uint32_t cls;
6049 	int ps;
6050 
6051 	/*
6052 	 * Some drivers apparently write to these registers w/o updating our
6053 	 * cached copy.  No harm happens if we update the copy, so do so here
6054 	 * so we can restore them.  The COMMAND register is modified by the
6055 	 * bus w/o updating the cache.  This should represent the normally
6056 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6057 	 */
6058 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6059 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6060 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6061 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6062 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6063 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6064 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6065 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6066 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6067 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6068 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6069 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6070 	case PCIM_HDRTYPE_NORMAL:
6071 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6072 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6073 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6074 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6075 		break;
6076 	case PCIM_HDRTYPE_BRIDGE:
6077 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6078 		    PCIR_SECLAT_1, 1);
6079 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6080 		    PCIR_SUBBUS_1, 1);
6081 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6082 		    PCIR_SECBUS_1, 1);
6083 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6084 		    PCIR_PRIBUS_1, 1);
6085 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6086 		    PCIR_BRIDGECTL_1, 2);
6087 		break;
6088 	case PCIM_HDRTYPE_CARDBUS:
6089 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6090 		    PCIR_SECLAT_2, 1);
6091 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6092 		    PCIR_SUBBUS_2, 1);
6093 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6094 		    PCIR_SECBUS_2, 1);
6095 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6096 		    PCIR_PRIBUS_2, 1);
6097 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6098 		    PCIR_BRIDGECTL_2, 2);
6099 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6100 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6101 		break;
6102 	}
6103 
6104 	if (dinfo->cfg.pcie.pcie_location != 0)
6105 		pci_cfg_save_pcie(dev, dinfo);
6106 
6107 	if (dinfo->cfg.pcix.pcix_location != 0)
6108 		pci_cfg_save_pcix(dev, dinfo);
6109 
6110 #ifdef PCI_IOV
6111 	if (dinfo->cfg.iov != NULL)
6112 		pci_iov_cfg_save(dev, dinfo);
6113 #endif
6114 
6115 	/*
6116 	 * don't set the state for display devices, base peripherals and
6117 	 * memory devices since bad things happen when they are powered down.
6118 	 * We should (a) have drivers that can easily detach and (b) use
6119 	 * generic drivers for these devices so that some device actually
6120 	 * attaches.  We need to make sure that when we implement (a) we don't
6121 	 * power the device down on a reattach.
6122 	 */
6123 	cls = pci_get_class(dev);
6124 	if (!setstate)
6125 		return;
6126 	switch (pci_do_power_nodriver)
6127 	{
6128 		case 0:		/* NO powerdown at all */
6129 			return;
6130 		case 1:		/* Conservative about what to power down */
6131 			if (cls == PCIC_STORAGE)
6132 				return;
6133 			/*FALLTHROUGH*/
6134 		case 2:		/* Aggressive about what to power down */
6135 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6136 			    cls == PCIC_BASEPERIPH)
6137 				return;
6138 			/*FALLTHROUGH*/
6139 		case 3:		/* Power down everything */
6140 			break;
6141 	}
6142 	/*
6143 	 * PCI spec says we can only go into D3 state from D0 state.
6144 	 * Transition from D[12] into D0 before going to D3 state.
6145 	 */
6146 	ps = pci_get_powerstate(dev);
6147 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6148 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6149 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6150 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6151 }
6152 
6153 /* Wrapper APIs suitable for device driver use. */
6154 void
6155 pci_save_state(device_t dev)
6156 {
6157 	struct pci_devinfo *dinfo;
6158 
6159 	dinfo = device_get_ivars(dev);
6160 	pci_cfg_save(dev, dinfo, 0);
6161 }
6162 
6163 void
6164 pci_restore_state(device_t dev)
6165 {
6166 	struct pci_devinfo *dinfo;
6167 
6168 	dinfo = device_get_ivars(dev);
6169 	pci_cfg_restore(dev, dinfo);
6170 }
6171 
6172 static int
6173 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6174     uintptr_t *id)
6175 {
6176 
6177 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6178 }
6179 
6180 /* Find the upstream port of a given PCI device in a root complex. */
6181 device_t
6182 pci_find_pcie_root_port(device_t dev)
6183 {
6184 	struct pci_devinfo *dinfo;
6185 	devclass_t pci_class;
6186 	device_t pcib, bus;
6187 
6188 	pci_class = devclass_find("pci");
6189 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6190 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6191 
6192 	/*
6193 	 * Walk the bridge hierarchy until we find a PCI-e root
6194 	 * port or a non-PCI device.
6195 	 */
6196 	for (;;) {
6197 		bus = device_get_parent(dev);
6198 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6199 		    device_get_nameunit(dev)));
6200 
6201 		pcib = device_get_parent(bus);
6202 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6203 		    device_get_nameunit(bus)));
6204 
6205 		/*
6206 		 * pcib's parent must be a PCI bus for this to be a
6207 		 * PCI-PCI bridge.
6208 		 */
6209 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6210 			return (NULL);
6211 
6212 		dinfo = device_get_ivars(pcib);
6213 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6214 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6215 			return (pcib);
6216 
6217 		dev = pcib;
6218 	}
6219 }
6220 
6221 /*
6222  * Wait for pending transactions to complete on a PCI-express function.
6223  *
6224  * The maximum delay is specified in milliseconds in max_delay.  Note
6225  * that this function may sleep.
6226  *
6227  * Returns true if the function is idle and false if the timeout is
6228  * exceeded.  If dev is not a PCI-express function, this returns true.
6229  */
6230 bool
6231 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6232 {
6233 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6234 	uint16_t sta;
6235 	int cap;
6236 
6237 	cap = dinfo->cfg.pcie.pcie_location;
6238 	if (cap == 0)
6239 		return (true);
6240 
6241 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6242 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6243 		if (max_delay == 0)
6244 			return (false);
6245 
6246 		/* Poll once every 100 milliseconds up to the timeout. */
6247 		if (max_delay > 100) {
6248 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6249 			max_delay -= 100;
6250 		} else {
6251 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6252 			    C_HARDCLOCK);
6253 			max_delay = 0;
6254 		}
6255 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6256 	}
6257 
6258 	return (true);
6259 }
6260 
6261 /*
6262  * Determine the maximum Completion Timeout in microseconds.
6263  *
6264  * For non-PCI-express functions this returns 0.
6265  */
6266 int
6267 pcie_get_max_completion_timeout(device_t dev)
6268 {
6269 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6270 	int cap;
6271 
6272 	cap = dinfo->cfg.pcie.pcie_location;
6273 	if (cap == 0)
6274 		return (0);
6275 
6276 	/*
6277 	 * Functions using the 1.x spec use the default timeout range of
6278 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6279 	 * support programmable timeouts also use this range.
6280 	 */
6281 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6282 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6283 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6284 		return (50 * 1000);
6285 
6286 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6287 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6288 	case PCIEM_CTL2_COMP_TIMO_100US:
6289 		return (100);
6290 	case PCIEM_CTL2_COMP_TIMO_10MS:
6291 		return (10 * 1000);
6292 	case PCIEM_CTL2_COMP_TIMO_55MS:
6293 		return (55 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_210MS:
6295 		return (210 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_900MS:
6297 		return (900 * 1000);
6298 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6299 		return (3500 * 1000);
6300 	case PCIEM_CTL2_COMP_TIMO_13S:
6301 		return (13 * 1000 * 1000);
6302 	case PCIEM_CTL2_COMP_TIMO_64S:
6303 		return (64 * 1000 * 1000);
6304 	default:
6305 		return (50 * 1000);
6306 	}
6307 }
6308 
6309 /*
6310  * Perform a Function Level Reset (FLR) on a device.
6311  *
6312  * This function first waits for any pending transactions to complete
6313  * within the timeout specified by max_delay.  If transactions are
6314  * still pending, the function will return false without attempting a
6315  * reset.
6316  *
6317  * If dev is not a PCI-express function or does not support FLR, this
6318  * function returns false.
6319  *
6320  * Note that no registers are saved or restored.  The caller is
6321  * responsible for saving and restoring any registers including
6322  * PCI-standard registers via pci_save_state() and
6323  * pci_restore_state().
6324  */
6325 bool
6326 pcie_flr(device_t dev, u_int max_delay, bool force)
6327 {
6328 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6329 	uint16_t cmd, ctl;
6330 	int compl_delay;
6331 	int cap;
6332 
6333 	cap = dinfo->cfg.pcie.pcie_location;
6334 	if (cap == 0)
6335 		return (false);
6336 
6337 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6338 		return (false);
6339 
6340 	/*
6341 	 * Disable busmastering to prevent generation of new
6342 	 * transactions while waiting for the device to go idle.  If
6343 	 * the idle timeout fails, the command register is restored
6344 	 * which will re-enable busmastering.
6345 	 */
6346 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6347 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6348 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6349 		if (!force) {
6350 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6351 			return (false);
6352 		}
6353 		pci_printf(&dinfo->cfg,
6354 		    "Resetting with transactions pending after %d ms\n",
6355 		    max_delay);
6356 
6357 		/*
6358 		 * Extend the post-FLR delay to cover the maximum
6359 		 * Completion Timeout delay of anything in flight
6360 		 * during the FLR delay.  Enforce a minimum delay of
6361 		 * at least 10ms.
6362 		 */
6363 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6364 		if (compl_delay < 10)
6365 			compl_delay = 10;
6366 	} else
6367 		compl_delay = 0;
6368 
6369 	/* Initiate the reset. */
6370 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6371 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6372 	    PCIEM_CTL_INITIATE_FLR, 2);
6373 
6374 	/* Wait for 100ms. */
6375 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6376 
6377 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6378 	    PCIEM_STA_TRANSACTION_PND)
6379 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6380 	return (true);
6381 }
6382 
6383 /*
6384  * Attempt a power-management reset by cycling the device in/out of D3
6385  * state.  PCI spec says we can only go into D3 state from D0 state.
6386  * Transition from D[12] into D0 before going to D3 state.
6387  */
6388 int
6389 pci_power_reset(device_t dev)
6390 {
6391 	int ps;
6392 
6393 	ps = pci_get_powerstate(dev);
6394 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6395 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6396 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6397 	pci_set_powerstate(dev, ps);
6398 	return (0);
6399 }
6400 
6401 /*
6402  * Try link drop and retrain of the downstream port of upstream
6403  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6404  * cause Conventional Hot reset of the device in the slot.
6405  * Alternative, for PCIe, could be the secondary bus reset initiatied
6406  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6407  */
6408 int
6409 pcie_link_reset(device_t port, int pcie_location)
6410 {
6411 	uint16_t v;
6412 
6413 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6414 	v |= PCIEM_LINK_CTL_LINK_DIS;
6415 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6416 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6417 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6418 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6419 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6420 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6421 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6422 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6423 }
6424 
6425 static int
6426 pci_reset_post(device_t dev, device_t child)
6427 {
6428 
6429 	if (dev == device_get_parent(child))
6430 		pci_restore_state(child);
6431 	return (0);
6432 }
6433 
6434 static int
6435 pci_reset_prepare(device_t dev, device_t child)
6436 {
6437 
6438 	if (dev == device_get_parent(child))
6439 		pci_save_state(child);
6440 	return (0);
6441 }
6442 
6443 static int
6444 pci_reset_child(device_t dev, device_t child, int flags)
6445 {
6446 	int error;
6447 
6448 	if (dev == NULL || device_get_parent(child) != dev)
6449 		return (0);
6450 	if ((flags & DEVF_RESET_DETACH) != 0) {
6451 		error = device_get_state(child) == DS_ATTACHED ?
6452 		    device_detach(child) : 0;
6453 	} else {
6454 		error = BUS_SUSPEND_CHILD(dev, child);
6455 	}
6456 	if (error == 0) {
6457 		if (!pcie_flr(child, 1000, false)) {
6458 			error = BUS_RESET_PREPARE(dev, child);
6459 			if (error == 0)
6460 				pci_power_reset(child);
6461 			BUS_RESET_POST(dev, child);
6462 		}
6463 		if ((flags & DEVF_RESET_DETACH) != 0)
6464 			device_probe_and_attach(child);
6465 		else
6466 			BUS_RESUME_CHILD(dev, child);
6467 	}
6468 	return (error);
6469 }
6470 
6471 const struct pci_device_table *
6472 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6473 {
6474 	bool match;
6475 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6476 
6477 	vendor = pci_get_vendor(child);
6478 	device = pci_get_device(child);
6479 	subvendor = pci_get_subvendor(child);
6480 	subdevice = pci_get_subdevice(child);
6481 	class = pci_get_class(child);
6482 	subclass = pci_get_subclass(child);
6483 	revid = pci_get_revid(child);
6484 	while (nelt-- > 0) {
6485 		match = true;
6486 		if (id->match_flag_vendor)
6487 			match &= vendor == id->vendor;
6488 		if (id->match_flag_device)
6489 			match &= device == id->device;
6490 		if (id->match_flag_subvendor)
6491 			match &= subvendor == id->subvendor;
6492 		if (id->match_flag_subdevice)
6493 			match &= subdevice == id->subdevice;
6494 		if (id->match_flag_class)
6495 			match &= class == id->class_id;
6496 		if (id->match_flag_subclass)
6497 			match &= subclass == id->subclass;
6498 		if (id->match_flag_revid)
6499 			match &= revid == id->revid;
6500 		if (match)
6501 			return (id);
6502 		id++;
6503 	}
6504 	return (NULL);
6505 }
6506 
6507 static void
6508 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6509 {
6510 	const char *dev_name;
6511 	device_t dev;
6512 
6513 	dev = dinfo->cfg.dev;
6514 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6515 	    dinfo->cfg.slot, dinfo->cfg.func);
6516 	dev_name = device_get_name(dev);
6517 	if (dev_name != NULL)
6518 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6519 }
6520 
6521 void
6522 pci_print_faulted_dev(void)
6523 {
6524 	struct pci_devinfo *dinfo;
6525 	device_t dev;
6526 	int aer, i;
6527 	uint32_t r1, r2;
6528 	uint16_t status;
6529 
6530 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6531 		dev = dinfo->cfg.dev;
6532 		status = pci_read_config(dev, PCIR_STATUS, 2);
6533 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6534 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6535 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6536 		if (status != 0) {
6537 			pci_print_faulted_dev_name(dinfo);
6538 			printf(" error 0x%04x\n", status);
6539 		}
6540 		if (dinfo->cfg.pcie.pcie_location != 0) {
6541 			status = pci_read_config(dev,
6542 			    dinfo->cfg.pcie.pcie_location +
6543 			    PCIER_DEVICE_STA, 2);
6544 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6545 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6546 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6547 				pci_print_faulted_dev_name(dinfo);
6548 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6549 				    pci_read_config(dev,
6550 				    dinfo->cfg.pcie.pcie_location +
6551 				    PCIER_DEVICE_CTL, 2),
6552 				    status);
6553 			}
6554 		}
6555 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6556 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6557 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6558 			if (r1 != 0 || r2 != 0) {
6559 				pci_print_faulted_dev_name(dinfo);
6560 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6561 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6562 				    r1, pci_read_config(dev, aer +
6563 				    PCIR_AER_UC_MASK, 4),
6564 				    pci_read_config(dev, aer +
6565 				    PCIR_AER_UC_SEVERITY, 4),
6566 				    r2, pci_read_config(dev, aer +
6567 				    PCIR_AER_COR_MASK, 4),
6568 				    pci_read_config(dev, aer +
6569 				    PCIR_AER_CAP_CONTROL, 4));
6570 				for (i = 0; i < 4; i++) {
6571 					r1 = pci_read_config(dev, aer +
6572 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6573 					printf("    HL%d: 0x%08x\n", i, r1);
6574 				}
6575 			}
6576 		}
6577 	}
6578 }
6579 
6580 #ifdef DDB
6581 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6582 {
6583 
6584 	pci_print_faulted_dev();
6585 }
6586 
6587 static void
6588 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6589 {
6590 	device_t dev;
6591 	int aer;
6592 	uint32_t r;
6593 
6594 	dev = dinfo->cfg.dev;
6595 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6596 	    PCIER_DEVICE_STA, 2);
6597 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6598 	    PCIER_DEVICE_STA, r, 2);
6599 
6600 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6601 		return;
6602 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6603 	if (r != 0)
6604 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6605 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6606 	if (r != 0)
6607 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6608 }
6609 
6610 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6611 {
6612 	struct pci_devinfo *dinfo;
6613 	device_t dev;
6614 	uint16_t status, status1;
6615 
6616 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6617 		dev = dinfo->cfg.dev;
6618 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6619 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6620 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6621 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6622 		if (status1 != 0) {
6623 			status &= ~status1;
6624 			pci_write_config(dev, PCIR_STATUS, status, 2);
6625 		}
6626 		if (dinfo->cfg.pcie.pcie_location != 0)
6627 			db_clear_pcie_errors(dinfo);
6628 	}
6629 }
6630 #endif
6631