xref: /freebsd/sys/dev/pci/pci.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_bus.h"
36 
37 #include <sys/param.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 
55 #include <sys/bus.h>
56 #include <machine/bus.h>
57 #include <sys/rman.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
60 
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
63 #endif
64 
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
69 
70 #ifdef PCI_IOV
71 #include <sys/nv.h>
72 #include <dev/pci/pci_iov_private.h>
73 #endif
74 
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
79 
80 #include "pcib_if.h"
81 #include "pci_if.h"
82 
83 #define	PCIR_IS_BIOS(cfg, reg)						\
84 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
85 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 
87 static int		pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t	pci_mapbase(uint64_t mapreg);
89 static const char	*pci_maptype(uint64_t mapreg);
90 static int		pci_maprange(uint64_t mapreg);
91 static pci_addr_t	pci_rombase(uint64_t mapreg);
92 static int		pci_romsize(uint64_t testval);
93 static void		pci_fixancient(pcicfgregs *cfg);
94 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 
96 static int		pci_porten(device_t dev);
97 static int		pci_memen(device_t dev);
98 static void		pci_assign_interrupt(device_t bus, device_t dev,
99 			    int force_route);
100 static int		pci_add_map(device_t bus, device_t dev, int reg,
101 			    struct resource_list *rl, int force, int prefetch);
102 static int		pci_probe(device_t dev);
103 static void		pci_load_vendor_data(void);
104 static int		pci_describe_parse_line(char **ptr, int *vendor,
105 			    int *device, char **desc);
106 static char		*pci_describe_device(device_t dev);
107 static int		pci_modevent(module_t mod, int what, void *arg);
108 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
109 			    pcicfgregs *cfg);
110 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
111 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
112 			    int reg, uint32_t *data);
113 #if 0
114 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
115 			    int reg, uint32_t data);
116 #endif
117 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
118 static void		pci_mask_msix(device_t dev, u_int index);
119 static void		pci_unmask_msix(device_t dev, u_int index);
120 static int		pci_msi_blacklisted(void);
121 static int		pci_msix_blacklisted(void);
122 static void		pci_resume_msi(device_t dev);
123 static void		pci_resume_msix(device_t dev);
124 static int		pci_remap_intr_method(device_t bus, device_t dev,
125 			    u_int irq);
126 static void		pci_hint_device_unit(device_t acdev, device_t child,
127 			    const char *name, int *unitp);
128 static int		pci_reset_post(device_t dev, device_t child);
129 static int		pci_reset_prepare(device_t dev, device_t child);
130 static int		pci_reset_child(device_t dev, device_t child,
131 			    int flags);
132 
133 static int		pci_get_id_method(device_t dev, device_t child,
134 			    enum pci_id_type type, uintptr_t *rid);
135 
136 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
137     int b, int s, int f, uint16_t vid, uint16_t did);
138 
139 static device_method_t pci_methods[] = {
140 	/* Device interface */
141 	DEVMETHOD(device_probe,		pci_probe),
142 	DEVMETHOD(device_attach,	pci_attach),
143 	DEVMETHOD(device_detach,	pci_detach),
144 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
145 	DEVMETHOD(device_suspend,	bus_generic_suspend),
146 	DEVMETHOD(device_resume,	pci_resume),
147 
148 	/* Bus interface */
149 	DEVMETHOD(bus_print_child,	pci_print_child),
150 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
151 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
152 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
153 	DEVMETHOD(bus_driver_added,	pci_driver_added),
154 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
155 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
156 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
157 	DEVMETHOD(bus_reset_post,	pci_reset_post),
158 	DEVMETHOD(bus_reset_child,	pci_reset_child),
159 
160 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
161 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
162 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
163 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
164 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
165 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
166 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
167 	DEVMETHOD(bus_release_resource,	pci_release_resource),
168 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
169 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
170 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
171 	DEVMETHOD(bus_child_detached,	pci_child_detached),
172 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
173 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
174 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
175 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
176 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
177 	DEVMETHOD(bus_resume_child,	pci_resume_child),
178 	DEVMETHOD(bus_rescan,		pci_rescan_method),
179 
180 	/* PCI interface */
181 	DEVMETHOD(pci_read_config,	pci_read_config_method),
182 	DEVMETHOD(pci_write_config,	pci_write_config_method),
183 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
184 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
185 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
186 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
187 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
188 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
189 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
190 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
191 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
192 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
193 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
194 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
195 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
196 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
197 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
198 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
199 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
200 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
201 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
202 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
203 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
204 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
205 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
206 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
207 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
208 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
209 	DEVMETHOD(pci_get_id,		pci_get_id_method),
210 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
211 	DEVMETHOD(pci_child_added,	pci_child_added_method),
212 #ifdef PCI_IOV
213 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
214 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
215 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
216 #endif
217 
218 	DEVMETHOD_END
219 };
220 
221 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
222 
223 static devclass_t pci_devclass;
224 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
225     BUS_PASS_BUS);
226 MODULE_VERSION(pci, 1);
227 
228 static char	*pci_vendordata;
229 static size_t	pci_vendordata_size;
230 
231 struct pci_quirk {
232 	uint32_t devid;	/* Vendor/device of the card */
233 	int	type;
234 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
235 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
236 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
237 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
238 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
239 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
240 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
241 	int	arg1;
242 	int	arg2;
243 };
244 
245 static const struct pci_quirk pci_quirks[] = {
246 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
247 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
248 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
249 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
250 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 
252 	/*
253 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
254 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
255 	 */
256 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
257 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
258 
259 	/*
260 	 * MSI doesn't work on earlier Intel chipsets including
261 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
262 	 */
263 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
264 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
265 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 
271 	/*
272 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
273 	 * bridge.
274 	 */
275 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
276 
277 	/*
278 	 * Some virtualization environments emulate an older chipset
279 	 * but support MSI just fine.  QEMU uses the Intel 82440.
280 	 */
281 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
282 
283 	/*
284 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
285 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
286 	 * It prevents us from attaching hpet(4) when the bit is unset.
287 	 * Note this quirk only affects SB600 revision A13 and earlier.
288 	 * For SB600 A21 and later, firmware must set the bit to hide it.
289 	 * For SB700 and later, it is unused and hardcoded to zero.
290 	 */
291 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
292 
293 	/*
294 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
295 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
296 	 * of the command register is set.
297 	 */
298 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
299 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
300 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 
304 	/*
305 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
306 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
307 	 */
308 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
309 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
310 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
311 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
312 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
313 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
314 
315 	/*
316 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
317 	 * expected place.
318 	 */
319 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
320 
321 	{ 0 }
322 };
323 
324 /* map register information */
325 #define	PCI_MAPMEM	0x01	/* memory map */
326 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
327 #define	PCI_MAPPORT	0x04	/* port map */
328 
329 struct devlist pci_devq;
330 uint32_t pci_generation;
331 uint32_t pci_numdevs = 0;
332 static int pcie_chipset, pcix_chipset;
333 
334 /* sysctl vars */
335 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
336 
337 static int pci_enable_io_modes = 1;
338 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
339     &pci_enable_io_modes, 1,
340     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
341     " enable these bits correctly.  We'd like to do this all the time, but"
342     " there are some peripherals that this causes problems with.");
343 
344 static int pci_do_realloc_bars = 1;
345 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
346     &pci_do_realloc_bars, 0,
347     "Attempt to allocate a new range for any BARs whose original "
348     "firmware-assigned ranges fail to allocate during the initial device scan.");
349 
350 static int pci_do_power_nodriver = 0;
351 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
352     &pci_do_power_nodriver, 0,
353     "Place a function into D3 state when no driver attaches to it.  0 means"
354     " disable.  1 means conservatively place devices into D3 state.  2 means"
355     " aggressively place devices into D3 state.  3 means put absolutely"
356     " everything in D3 state.");
357 
358 int pci_do_power_resume = 1;
359 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
360     &pci_do_power_resume, 1,
361   "Transition from D3 -> D0 on resume.");
362 
363 int pci_do_power_suspend = 1;
364 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
365     &pci_do_power_suspend, 1,
366   "Transition from D0 -> D3 on suspend.");
367 
368 static int pci_do_msi = 1;
369 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
370     "Enable support for MSI interrupts");
371 
372 static int pci_do_msix = 1;
373 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
374     "Enable support for MSI-X interrupts");
375 
376 static int pci_msix_rewrite_table = 0;
377 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
378     &pci_msix_rewrite_table, 0,
379     "Rewrite entire MSI-X table when updating MSI-X entries");
380 
381 static int pci_honor_msi_blacklist = 1;
382 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
383     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
384 
385 #if defined(__i386__) || defined(__amd64__)
386 static int pci_usb_takeover = 1;
387 #else
388 static int pci_usb_takeover = 0;
389 #endif
390 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
391     &pci_usb_takeover, 1,
392     "Enable early takeover of USB controllers. Disable this if you depend on"
393     " BIOS emulation of USB devices, that is you use USB devices (like"
394     " keyboard or mouse) but do not load USB drivers");
395 
396 static int pci_clear_bars;
397 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
398     "Ignore firmware-assigned resources for BARs.");
399 
400 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
401 static int pci_clear_buses;
402 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
403     "Ignore firmware-assigned bus numbers.");
404 #endif
405 
406 static int pci_enable_ari = 1;
407 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
408     0, "Enable support for PCIe Alternative RID Interpretation");
409 
410 static int pci_clear_aer_on_attach = 0;
411 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
412     &pci_clear_aer_on_attach, 0,
413     "Clear port and device AER state on driver attach");
414 
415 static int
416 pci_has_quirk(uint32_t devid, int quirk)
417 {
418 	const struct pci_quirk *q;
419 
420 	for (q = &pci_quirks[0]; q->devid; q++) {
421 		if (q->devid == devid && q->type == quirk)
422 			return (1);
423 	}
424 	return (0);
425 }
426 
427 /* Find a device_t by bus/slot/function in domain 0 */
428 
429 device_t
430 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
431 {
432 
433 	return (pci_find_dbsf(0, bus, slot, func));
434 }
435 
436 /* Find a device_t by domain/bus/slot/function */
437 
438 device_t
439 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
440 {
441 	struct pci_devinfo *dinfo = NULL;
442 
443 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
444 		if ((dinfo->cfg.domain == domain) &&
445 		    (dinfo->cfg.bus == bus) &&
446 		    (dinfo->cfg.slot == slot) &&
447 		    (dinfo->cfg.func == func)) {
448 			break;
449 		}
450 	}
451 
452 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
453 }
454 
455 /* Find a device_t by vendor/device ID */
456 
457 device_t
458 pci_find_device(uint16_t vendor, uint16_t device)
459 {
460 	struct pci_devinfo *dinfo;
461 
462 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
463 		if ((dinfo->cfg.vendor == vendor) &&
464 		    (dinfo->cfg.device == device)) {
465 			return (dinfo->cfg.dev);
466 		}
467 	}
468 
469 	return (NULL);
470 }
471 
472 device_t
473 pci_find_class(uint8_t class, uint8_t subclass)
474 {
475 	struct pci_devinfo *dinfo;
476 
477 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
478 		if (dinfo->cfg.baseclass == class &&
479 		    dinfo->cfg.subclass == subclass) {
480 			return (dinfo->cfg.dev);
481 		}
482 	}
483 
484 	return (NULL);
485 }
486 
487 static int
488 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
489 {
490 	va_list ap;
491 	int retval;
492 
493 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
494 	    cfg->func);
495 	va_start(ap, fmt);
496 	retval += vprintf(fmt, ap);
497 	va_end(ap);
498 	return (retval);
499 }
500 
501 /* return base address of memory or port map */
502 
503 static pci_addr_t
504 pci_mapbase(uint64_t mapreg)
505 {
506 
507 	if (PCI_BAR_MEM(mapreg))
508 		return (mapreg & PCIM_BAR_MEM_BASE);
509 	else
510 		return (mapreg & PCIM_BAR_IO_BASE);
511 }
512 
513 /* return map type of memory or port map */
514 
515 static const char *
516 pci_maptype(uint64_t mapreg)
517 {
518 
519 	if (PCI_BAR_IO(mapreg))
520 		return ("I/O Port");
521 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
522 		return ("Prefetchable Memory");
523 	return ("Memory");
524 }
525 
526 /* return log2 of map size decoded for memory or port map */
527 
528 int
529 pci_mapsize(uint64_t testval)
530 {
531 	int ln2size;
532 
533 	testval = pci_mapbase(testval);
534 	ln2size = 0;
535 	if (testval != 0) {
536 		while ((testval & 1) == 0)
537 		{
538 			ln2size++;
539 			testval >>= 1;
540 		}
541 	}
542 	return (ln2size);
543 }
544 
545 /* return base address of device ROM */
546 
547 static pci_addr_t
548 pci_rombase(uint64_t mapreg)
549 {
550 
551 	return (mapreg & PCIM_BIOS_ADDR_MASK);
552 }
553 
554 /* return log2 of map size decided for device ROM */
555 
556 static int
557 pci_romsize(uint64_t testval)
558 {
559 	int ln2size;
560 
561 	testval = pci_rombase(testval);
562 	ln2size = 0;
563 	if (testval != 0) {
564 		while ((testval & 1) == 0)
565 		{
566 			ln2size++;
567 			testval >>= 1;
568 		}
569 	}
570 	return (ln2size);
571 }
572 
573 /* return log2 of address range supported by map register */
574 
575 static int
576 pci_maprange(uint64_t mapreg)
577 {
578 	int ln2range = 0;
579 
580 	if (PCI_BAR_IO(mapreg))
581 		ln2range = 32;
582 	else
583 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
584 		case PCIM_BAR_MEM_32:
585 			ln2range = 32;
586 			break;
587 		case PCIM_BAR_MEM_1MB:
588 			ln2range = 20;
589 			break;
590 		case PCIM_BAR_MEM_64:
591 			ln2range = 64;
592 			break;
593 		}
594 	return (ln2range);
595 }
596 
597 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
598 
599 static void
600 pci_fixancient(pcicfgregs *cfg)
601 {
602 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
603 		return;
604 
605 	/* PCI to PCI bridges use header type 1 */
606 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
607 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
608 }
609 
610 /* extract header type specific config data */
611 
612 static void
613 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
614 {
615 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
616 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
617 	case PCIM_HDRTYPE_NORMAL:
618 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
619 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
620 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
621 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
622 		cfg->nummaps	    = PCI_MAXMAPS_0;
623 		break;
624 	case PCIM_HDRTYPE_BRIDGE:
625 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
626 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
627 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
628 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
629 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
630 		cfg->nummaps	    = PCI_MAXMAPS_1;
631 		break;
632 	case PCIM_HDRTYPE_CARDBUS:
633 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
634 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
635 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
636 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
637 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
638 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
639 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
640 		cfg->nummaps	    = PCI_MAXMAPS_2;
641 		break;
642 	}
643 #undef REG
644 }
645 
646 /* read configuration header into pcicfgregs structure */
647 struct pci_devinfo *
648 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
649 {
650 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
651 	uint16_t vid, did;
652 
653 	vid = REG(PCIR_VENDOR, 2);
654 	did = REG(PCIR_DEVICE, 2);
655 	if (vid != 0xffff)
656 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
657 
658 	return (NULL);
659 }
660 
661 struct pci_devinfo *
662 pci_alloc_devinfo_method(device_t dev)
663 {
664 
665 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
666 	    M_WAITOK | M_ZERO));
667 }
668 
669 static struct pci_devinfo *
670 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
671     uint16_t vid, uint16_t did)
672 {
673 	struct pci_devinfo *devlist_entry;
674 	pcicfgregs *cfg;
675 
676 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
677 
678 	cfg = &devlist_entry->cfg;
679 
680 	cfg->domain		= d;
681 	cfg->bus		= b;
682 	cfg->slot		= s;
683 	cfg->func		= f;
684 	cfg->vendor		= vid;
685 	cfg->device		= did;
686 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
687 	cfg->statreg		= REG(PCIR_STATUS, 2);
688 	cfg->baseclass		= REG(PCIR_CLASS, 1);
689 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
690 	cfg->progif		= REG(PCIR_PROGIF, 1);
691 	cfg->revid		= REG(PCIR_REVID, 1);
692 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
693 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
694 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
695 	cfg->intpin		= REG(PCIR_INTPIN, 1);
696 	cfg->intline		= REG(PCIR_INTLINE, 1);
697 
698 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
699 	cfg->hdrtype		&= ~PCIM_MFDEV;
700 	STAILQ_INIT(&cfg->maps);
701 
702 	cfg->iov		= NULL;
703 
704 	pci_fixancient(cfg);
705 	pci_hdrtypedata(pcib, b, s, f, cfg);
706 
707 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
708 		pci_read_cap(pcib, cfg);
709 
710 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
711 
712 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
713 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
714 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
715 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
716 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
717 
718 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
719 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
720 	devlist_entry->conf.pc_vendor = cfg->vendor;
721 	devlist_entry->conf.pc_device = cfg->device;
722 
723 	devlist_entry->conf.pc_class = cfg->baseclass;
724 	devlist_entry->conf.pc_subclass = cfg->subclass;
725 	devlist_entry->conf.pc_progif = cfg->progif;
726 	devlist_entry->conf.pc_revid = cfg->revid;
727 
728 	pci_numdevs++;
729 	pci_generation++;
730 
731 	return (devlist_entry);
732 }
733 #undef REG
734 
735 static void
736 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
737 {
738 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
739     cfg->ea.ea_location + (n), w)
740 	int num_ent;
741 	int ptr;
742 	int a, b;
743 	uint32_t val;
744 	int ent_size;
745 	uint32_t dw[4];
746 	uint64_t base, max_offset;
747 	struct pci_ea_entry *eae;
748 
749 	if (cfg->ea.ea_location == 0)
750 		return;
751 
752 	STAILQ_INIT(&cfg->ea.ea_entries);
753 
754 	/* Determine the number of entries */
755 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
756 	num_ent &= PCIM_EA_NUM_ENT_MASK;
757 
758 	/* Find the first entry to care of */
759 	ptr = PCIR_EA_FIRST_ENT;
760 
761 	/* Skip DWORD 2 for type 1 functions */
762 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
763 		ptr += 4;
764 
765 	for (a = 0; a < num_ent; a++) {
766 
767 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
768 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
769 
770 		/* Read a number of dwords in the entry */
771 		val = REG(ptr, 4);
772 		ptr += 4;
773 		ent_size = (val & PCIM_EA_ES);
774 
775 		for (b = 0; b < ent_size; b++) {
776 			dw[b] = REG(ptr, 4);
777 			ptr += 4;
778 		}
779 
780 		eae->eae_flags = val;
781 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
782 
783 		base = dw[0] & PCIM_EA_FIELD_MASK;
784 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
785 		b = 2;
786 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
787 			base |= (uint64_t)dw[b] << 32UL;
788 			b++;
789 		}
790 		if (((dw[1] & PCIM_EA_IS_64) != 0)
791 		    && (b < ent_size)) {
792 			max_offset |= (uint64_t)dw[b] << 32UL;
793 			b++;
794 		}
795 
796 		eae->eae_base = base;
797 		eae->eae_max_offset = max_offset;
798 
799 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
800 
801 		if (bootverbose) {
802 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
803 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
804 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
805 		}
806 	}
807 }
808 #undef REG
809 
810 static void
811 pci_read_cap(device_t pcib, pcicfgregs *cfg)
812 {
813 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
814 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
815 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
816 	uint64_t addr;
817 #endif
818 	uint32_t val;
819 	int	ptr, nextptr, ptrptr;
820 
821 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
822 	case PCIM_HDRTYPE_NORMAL:
823 	case PCIM_HDRTYPE_BRIDGE:
824 		ptrptr = PCIR_CAP_PTR;
825 		break;
826 	case PCIM_HDRTYPE_CARDBUS:
827 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
828 		break;
829 	default:
830 		return;		/* no extended capabilities support */
831 	}
832 	nextptr = REG(ptrptr, 1);	/* sanity check? */
833 
834 	/*
835 	 * Read capability entries.
836 	 */
837 	while (nextptr != 0) {
838 		/* Sanity check */
839 		if (nextptr > 255) {
840 			printf("illegal PCI extended capability offset %d\n",
841 			    nextptr);
842 			return;
843 		}
844 		/* Find the next entry */
845 		ptr = nextptr;
846 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
847 
848 		/* Process this entry */
849 		switch (REG(ptr + PCICAP_ID, 1)) {
850 		case PCIY_PMG:		/* PCI power management */
851 			if (cfg->pp.pp_cap == 0) {
852 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
853 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
854 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
855 				if ((nextptr - ptr) > PCIR_POWER_DATA)
856 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
857 			}
858 			break;
859 		case PCIY_HT:		/* HyperTransport */
860 			/* Determine HT-specific capability type. */
861 			val = REG(ptr + PCIR_HT_COMMAND, 2);
862 
863 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
864 				cfg->ht.ht_slave = ptr;
865 
866 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
867 			switch (val & PCIM_HTCMD_CAP_MASK) {
868 			case PCIM_HTCAP_MSI_MAPPING:
869 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
870 					/* Sanity check the mapping window. */
871 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
872 					    4);
873 					addr <<= 32;
874 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
875 					    4);
876 					if (addr != MSI_INTEL_ADDR_BASE)
877 						device_printf(pcib,
878 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
879 						    cfg->domain, cfg->bus,
880 						    cfg->slot, cfg->func,
881 						    (long long)addr);
882 				} else
883 					addr = MSI_INTEL_ADDR_BASE;
884 
885 				cfg->ht.ht_msimap = ptr;
886 				cfg->ht.ht_msictrl = val;
887 				cfg->ht.ht_msiaddr = addr;
888 				break;
889 			}
890 #endif
891 			break;
892 		case PCIY_MSI:		/* PCI MSI */
893 			cfg->msi.msi_location = ptr;
894 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
895 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
896 						     PCIM_MSICTRL_MMC_MASK)>>1);
897 			break;
898 		case PCIY_MSIX:		/* PCI MSI-X */
899 			cfg->msix.msix_location = ptr;
900 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
901 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
902 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
903 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
904 			cfg->msix.msix_table_bar = PCIR_BAR(val &
905 			    PCIM_MSIX_BIR_MASK);
906 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
907 			val = REG(ptr + PCIR_MSIX_PBA, 4);
908 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
909 			    PCIM_MSIX_BIR_MASK);
910 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
911 			break;
912 		case PCIY_VPD:		/* PCI Vital Product Data */
913 			cfg->vpd.vpd_reg = ptr;
914 			break;
915 		case PCIY_SUBVENDOR:
916 			/* Should always be true. */
917 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
918 			    PCIM_HDRTYPE_BRIDGE) {
919 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
920 				cfg->subvendor = val & 0xffff;
921 				cfg->subdevice = val >> 16;
922 			}
923 			break;
924 		case PCIY_PCIX:		/* PCI-X */
925 			/*
926 			 * Assume we have a PCI-X chipset if we have
927 			 * at least one PCI-PCI bridge with a PCI-X
928 			 * capability.  Note that some systems with
929 			 * PCI-express or HT chipsets might match on
930 			 * this check as well.
931 			 */
932 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
933 			    PCIM_HDRTYPE_BRIDGE)
934 				pcix_chipset = 1;
935 			cfg->pcix.pcix_location = ptr;
936 			break;
937 		case PCIY_EXPRESS:	/* PCI-express */
938 			/*
939 			 * Assume we have a PCI-express chipset if we have
940 			 * at least one PCI-express device.
941 			 */
942 			pcie_chipset = 1;
943 			cfg->pcie.pcie_location = ptr;
944 			val = REG(ptr + PCIER_FLAGS, 2);
945 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
946 			break;
947 		case PCIY_EA:		/* Enhanced Allocation */
948 			cfg->ea.ea_location = ptr;
949 			pci_ea_fill_info(pcib, cfg);
950 			break;
951 		default:
952 			break;
953 		}
954 	}
955 
956 #if defined(__powerpc__)
957 	/*
958 	 * Enable the MSI mapping window for all HyperTransport
959 	 * slaves.  PCI-PCI bridges have their windows enabled via
960 	 * PCIB_MAP_MSI().
961 	 */
962 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
963 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
964 		device_printf(pcib,
965 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
966 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
967 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
968 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
969 		     2);
970 	}
971 #endif
972 /* REG and WREG use carry through to next functions */
973 }
974 
975 /*
976  * PCI Vital Product Data
977  */
978 
979 #define	PCI_VPD_TIMEOUT		1000000
980 
981 static int
982 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
983 {
984 	int count = PCI_VPD_TIMEOUT;
985 
986 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
987 
988 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
989 
990 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
991 		if (--count < 0)
992 			return (ENXIO);
993 		DELAY(1);	/* limit looping */
994 	}
995 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
996 
997 	return (0);
998 }
999 
1000 #if 0
1001 static int
1002 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1003 {
1004 	int count = PCI_VPD_TIMEOUT;
1005 
1006 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1007 
1008 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1009 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1010 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1011 		if (--count < 0)
1012 			return (ENXIO);
1013 		DELAY(1);	/* limit looping */
1014 	}
1015 
1016 	return (0);
1017 }
1018 #endif
1019 
1020 #undef PCI_VPD_TIMEOUT
1021 
1022 struct vpd_readstate {
1023 	device_t	pcib;
1024 	pcicfgregs	*cfg;
1025 	uint32_t	val;
1026 	int		bytesinval;
1027 	int		off;
1028 	uint8_t		cksum;
1029 };
1030 
1031 static int
1032 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1033 {
1034 	uint32_t reg;
1035 	uint8_t byte;
1036 
1037 	if (vrs->bytesinval == 0) {
1038 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1039 			return (ENXIO);
1040 		vrs->val = le32toh(reg);
1041 		vrs->off += 4;
1042 		byte = vrs->val & 0xff;
1043 		vrs->bytesinval = 3;
1044 	} else {
1045 		vrs->val = vrs->val >> 8;
1046 		byte = vrs->val & 0xff;
1047 		vrs->bytesinval--;
1048 	}
1049 
1050 	vrs->cksum += byte;
1051 	*data = byte;
1052 	return (0);
1053 }
1054 
1055 static void
1056 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1057 {
1058 	struct vpd_readstate vrs;
1059 	int state;
1060 	int name;
1061 	int remain;
1062 	int i;
1063 	int alloc, off;		/* alloc/off for RO/W arrays */
1064 	int cksumvalid;
1065 	int dflen;
1066 	uint8_t byte;
1067 	uint8_t byte2;
1068 
1069 	/* init vpd reader */
1070 	vrs.bytesinval = 0;
1071 	vrs.off = 0;
1072 	vrs.pcib = pcib;
1073 	vrs.cfg = cfg;
1074 	vrs.cksum = 0;
1075 
1076 	state = 0;
1077 	name = remain = i = 0;	/* shut up stupid gcc */
1078 	alloc = off = 0;	/* shut up stupid gcc */
1079 	dflen = 0;		/* shut up stupid gcc */
1080 	cksumvalid = -1;
1081 	while (state >= 0) {
1082 		if (vpd_nextbyte(&vrs, &byte)) {
1083 			state = -2;
1084 			break;
1085 		}
1086 #if 0
1087 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1088 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1089 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1090 #endif
1091 		switch (state) {
1092 		case 0:		/* item name */
1093 			if (byte & 0x80) {
1094 				if (vpd_nextbyte(&vrs, &byte2)) {
1095 					state = -2;
1096 					break;
1097 				}
1098 				remain = byte2;
1099 				if (vpd_nextbyte(&vrs, &byte2)) {
1100 					state = -2;
1101 					break;
1102 				}
1103 				remain |= byte2 << 8;
1104 				if (remain > (0x7f*4 - vrs.off)) {
1105 					state = -1;
1106 					pci_printf(cfg,
1107 					    "invalid VPD data, remain %#x\n",
1108 					    remain);
1109 				}
1110 				name = byte & 0x7f;
1111 			} else {
1112 				remain = byte & 0x7;
1113 				name = (byte >> 3) & 0xf;
1114 			}
1115 			switch (name) {
1116 			case 0x2:	/* String */
1117 				cfg->vpd.vpd_ident = malloc(remain + 1,
1118 				    M_DEVBUF, M_WAITOK);
1119 				i = 0;
1120 				state = 1;
1121 				break;
1122 			case 0xf:	/* End */
1123 				state = -1;
1124 				break;
1125 			case 0x10:	/* VPD-R */
1126 				alloc = 8;
1127 				off = 0;
1128 				cfg->vpd.vpd_ros = malloc(alloc *
1129 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1130 				    M_WAITOK | M_ZERO);
1131 				state = 2;
1132 				break;
1133 			case 0x11:	/* VPD-W */
1134 				alloc = 8;
1135 				off = 0;
1136 				cfg->vpd.vpd_w = malloc(alloc *
1137 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1138 				    M_WAITOK | M_ZERO);
1139 				state = 5;
1140 				break;
1141 			default:	/* Invalid data, abort */
1142 				state = -1;
1143 				break;
1144 			}
1145 			break;
1146 
1147 		case 1:	/* Identifier String */
1148 			cfg->vpd.vpd_ident[i++] = byte;
1149 			remain--;
1150 			if (remain == 0)  {
1151 				cfg->vpd.vpd_ident[i] = '\0';
1152 				state = 0;
1153 			}
1154 			break;
1155 
1156 		case 2:	/* VPD-R Keyword Header */
1157 			if (off == alloc) {
1158 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1159 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1160 				    M_DEVBUF, M_WAITOK | M_ZERO);
1161 			}
1162 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1163 			if (vpd_nextbyte(&vrs, &byte2)) {
1164 				state = -2;
1165 				break;
1166 			}
1167 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1168 			if (vpd_nextbyte(&vrs, &byte2)) {
1169 				state = -2;
1170 				break;
1171 			}
1172 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1173 			if (dflen == 0 &&
1174 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1175 			    2) == 0) {
1176 				/*
1177 				 * if this happens, we can't trust the rest
1178 				 * of the VPD.
1179 				 */
1180 				pci_printf(cfg, "bad keyword length: %d\n",
1181 				    dflen);
1182 				cksumvalid = 0;
1183 				state = -1;
1184 				break;
1185 			} else if (dflen == 0) {
1186 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1187 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1188 				    M_DEVBUF, M_WAITOK);
1189 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1190 			} else
1191 				cfg->vpd.vpd_ros[off].value = malloc(
1192 				    (dflen + 1) *
1193 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1194 				    M_DEVBUF, M_WAITOK);
1195 			remain -= 3;
1196 			i = 0;
1197 			/* keep in sync w/ state 3's transistions */
1198 			if (dflen == 0 && remain == 0)
1199 				state = 0;
1200 			else if (dflen == 0)
1201 				state = 2;
1202 			else
1203 				state = 3;
1204 			break;
1205 
1206 		case 3:	/* VPD-R Keyword Value */
1207 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1208 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1209 			    "RV", 2) == 0 && cksumvalid == -1) {
1210 				if (vrs.cksum == 0)
1211 					cksumvalid = 1;
1212 				else {
1213 					if (bootverbose)
1214 						pci_printf(cfg,
1215 					    "bad VPD cksum, remain %hhu\n",
1216 						    vrs.cksum);
1217 					cksumvalid = 0;
1218 					state = -1;
1219 					break;
1220 				}
1221 			}
1222 			dflen--;
1223 			remain--;
1224 			/* keep in sync w/ state 2's transistions */
1225 			if (dflen == 0)
1226 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1227 			if (dflen == 0 && remain == 0) {
1228 				cfg->vpd.vpd_rocnt = off;
1229 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1230 				    off * sizeof(*cfg->vpd.vpd_ros),
1231 				    M_DEVBUF, M_WAITOK | M_ZERO);
1232 				state = 0;
1233 			} else if (dflen == 0)
1234 				state = 2;
1235 			break;
1236 
1237 		case 4:
1238 			remain--;
1239 			if (remain == 0)
1240 				state = 0;
1241 			break;
1242 
1243 		case 5:	/* VPD-W Keyword Header */
1244 			if (off == alloc) {
1245 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1246 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1247 				    M_DEVBUF, M_WAITOK | M_ZERO);
1248 			}
1249 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1250 			if (vpd_nextbyte(&vrs, &byte2)) {
1251 				state = -2;
1252 				break;
1253 			}
1254 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1255 			if (vpd_nextbyte(&vrs, &byte2)) {
1256 				state = -2;
1257 				break;
1258 			}
1259 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1260 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1261 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1262 			    sizeof(*cfg->vpd.vpd_w[off].value),
1263 			    M_DEVBUF, M_WAITOK);
1264 			remain -= 3;
1265 			i = 0;
1266 			/* keep in sync w/ state 6's transistions */
1267 			if (dflen == 0 && remain == 0)
1268 				state = 0;
1269 			else if (dflen == 0)
1270 				state = 5;
1271 			else
1272 				state = 6;
1273 			break;
1274 
1275 		case 6:	/* VPD-W Keyword Value */
1276 			cfg->vpd.vpd_w[off].value[i++] = byte;
1277 			dflen--;
1278 			remain--;
1279 			/* keep in sync w/ state 5's transistions */
1280 			if (dflen == 0)
1281 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1282 			if (dflen == 0 && remain == 0) {
1283 				cfg->vpd.vpd_wcnt = off;
1284 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1285 				    off * sizeof(*cfg->vpd.vpd_w),
1286 				    M_DEVBUF, M_WAITOK | M_ZERO);
1287 				state = 0;
1288 			} else if (dflen == 0)
1289 				state = 5;
1290 			break;
1291 
1292 		default:
1293 			pci_printf(cfg, "invalid state: %d\n", state);
1294 			state = -1;
1295 			break;
1296 		}
1297 	}
1298 
1299 	if (cksumvalid == 0 || state < -1) {
1300 		/* read-only data bad, clean up */
1301 		if (cfg->vpd.vpd_ros != NULL) {
1302 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1303 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1304 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1305 			cfg->vpd.vpd_ros = NULL;
1306 		}
1307 	}
1308 	if (state < -1) {
1309 		/* I/O error, clean up */
1310 		pci_printf(cfg, "failed to read VPD data.\n");
1311 		if (cfg->vpd.vpd_ident != NULL) {
1312 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1313 			cfg->vpd.vpd_ident = NULL;
1314 		}
1315 		if (cfg->vpd.vpd_w != NULL) {
1316 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1317 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1318 			free(cfg->vpd.vpd_w, M_DEVBUF);
1319 			cfg->vpd.vpd_w = NULL;
1320 		}
1321 	}
1322 	cfg->vpd.vpd_cached = 1;
1323 #undef REG
1324 #undef WREG
1325 }
1326 
1327 int
1328 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1329 {
1330 	struct pci_devinfo *dinfo = device_get_ivars(child);
1331 	pcicfgregs *cfg = &dinfo->cfg;
1332 
1333 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1334 		pci_read_vpd(device_get_parent(dev), cfg);
1335 
1336 	*identptr = cfg->vpd.vpd_ident;
1337 
1338 	if (*identptr == NULL)
1339 		return (ENXIO);
1340 
1341 	return (0);
1342 }
1343 
1344 int
1345 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1346 	const char **vptr)
1347 {
1348 	struct pci_devinfo *dinfo = device_get_ivars(child);
1349 	pcicfgregs *cfg = &dinfo->cfg;
1350 	int i;
1351 
1352 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1353 		pci_read_vpd(device_get_parent(dev), cfg);
1354 
1355 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1356 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1357 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1358 			*vptr = cfg->vpd.vpd_ros[i].value;
1359 			return (0);
1360 		}
1361 
1362 	*vptr = NULL;
1363 	return (ENXIO);
1364 }
1365 
1366 struct pcicfg_vpd *
1367 pci_fetch_vpd_list(device_t dev)
1368 {
1369 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1370 	pcicfgregs *cfg = &dinfo->cfg;
1371 
1372 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1373 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1374 	return (&cfg->vpd);
1375 }
1376 
1377 /*
1378  * Find the requested HyperTransport capability and return the offset
1379  * in configuration space via the pointer provided.  The function
1380  * returns 0 on success and an error code otherwise.
1381  */
1382 int
1383 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1384 {
1385 	int ptr, error;
1386 	uint16_t val;
1387 
1388 	error = pci_find_cap(child, PCIY_HT, &ptr);
1389 	if (error)
1390 		return (error);
1391 
1392 	/*
1393 	 * Traverse the capabilities list checking each HT capability
1394 	 * to see if it matches the requested HT capability.
1395 	 */
1396 	for (;;) {
1397 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1398 		if (capability == PCIM_HTCAP_SLAVE ||
1399 		    capability == PCIM_HTCAP_HOST)
1400 			val &= 0xe000;
1401 		else
1402 			val &= PCIM_HTCMD_CAP_MASK;
1403 		if (val == capability) {
1404 			if (capreg != NULL)
1405 				*capreg = ptr;
1406 			return (0);
1407 		}
1408 
1409 		/* Skip to the next HT capability. */
1410 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1411 			break;
1412 	}
1413 
1414 	return (ENOENT);
1415 }
1416 
1417 /*
1418  * Find the next requested HyperTransport capability after start and return
1419  * the offset in configuration space via the pointer provided.  The function
1420  * returns 0 on success and an error code otherwise.
1421  */
1422 int
1423 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1424     int start, int *capreg)
1425 {
1426 	int ptr;
1427 	uint16_t val;
1428 
1429 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1430 	    ("start capability is not HyperTransport capability"));
1431 	ptr = start;
1432 
1433 	/*
1434 	 * Traverse the capabilities list checking each HT capability
1435 	 * to see if it matches the requested HT capability.
1436 	 */
1437 	for (;;) {
1438 		/* Skip to the next HT capability. */
1439 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1440 			break;
1441 
1442 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1443 		if (capability == PCIM_HTCAP_SLAVE ||
1444 		    capability == PCIM_HTCAP_HOST)
1445 			val &= 0xe000;
1446 		else
1447 			val &= PCIM_HTCMD_CAP_MASK;
1448 		if (val == capability) {
1449 			if (capreg != NULL)
1450 				*capreg = ptr;
1451 			return (0);
1452 		}
1453 	}
1454 
1455 	return (ENOENT);
1456 }
1457 
1458 /*
1459  * Find the requested capability and return the offset in
1460  * configuration space via the pointer provided.  The function returns
1461  * 0 on success and an error code otherwise.
1462  */
1463 int
1464 pci_find_cap_method(device_t dev, device_t child, int capability,
1465     int *capreg)
1466 {
1467 	struct pci_devinfo *dinfo = device_get_ivars(child);
1468 	pcicfgregs *cfg = &dinfo->cfg;
1469 	uint32_t status;
1470 	uint8_t ptr;
1471 
1472 	/*
1473 	 * Check the CAP_LIST bit of the PCI status register first.
1474 	 */
1475 	status = pci_read_config(child, PCIR_STATUS, 2);
1476 	if (!(status & PCIM_STATUS_CAPPRESENT))
1477 		return (ENXIO);
1478 
1479 	/*
1480 	 * Determine the start pointer of the capabilities list.
1481 	 */
1482 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1483 	case PCIM_HDRTYPE_NORMAL:
1484 	case PCIM_HDRTYPE_BRIDGE:
1485 		ptr = PCIR_CAP_PTR;
1486 		break;
1487 	case PCIM_HDRTYPE_CARDBUS:
1488 		ptr = PCIR_CAP_PTR_2;
1489 		break;
1490 	default:
1491 		/* XXX: panic? */
1492 		return (ENXIO);		/* no extended capabilities support */
1493 	}
1494 	ptr = pci_read_config(child, ptr, 1);
1495 
1496 	/*
1497 	 * Traverse the capabilities list.
1498 	 */
1499 	while (ptr != 0) {
1500 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1501 			if (capreg != NULL)
1502 				*capreg = ptr;
1503 			return (0);
1504 		}
1505 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1506 	}
1507 
1508 	return (ENOENT);
1509 }
1510 
1511 /*
1512  * Find the next requested capability after start and return the offset in
1513  * configuration space via the pointer provided.  The function returns
1514  * 0 on success and an error code otherwise.
1515  */
1516 int
1517 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1518     int start, int *capreg)
1519 {
1520 	uint8_t ptr;
1521 
1522 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1523 	    ("start capability is not expected capability"));
1524 
1525 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1526 	while (ptr != 0) {
1527 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1528 			if (capreg != NULL)
1529 				*capreg = ptr;
1530 			return (0);
1531 		}
1532 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1533 	}
1534 
1535 	return (ENOENT);
1536 }
1537 
1538 /*
1539  * Find the requested extended capability and return the offset in
1540  * configuration space via the pointer provided.  The function returns
1541  * 0 on success and an error code otherwise.
1542  */
1543 int
1544 pci_find_extcap_method(device_t dev, device_t child, int capability,
1545     int *capreg)
1546 {
1547 	struct pci_devinfo *dinfo = device_get_ivars(child);
1548 	pcicfgregs *cfg = &dinfo->cfg;
1549 	uint32_t ecap;
1550 	uint16_t ptr;
1551 
1552 	/* Only supported for PCI-express devices. */
1553 	if (cfg->pcie.pcie_location == 0)
1554 		return (ENXIO);
1555 
1556 	ptr = PCIR_EXTCAP;
1557 	ecap = pci_read_config(child, ptr, 4);
1558 	if (ecap == 0xffffffff || ecap == 0)
1559 		return (ENOENT);
1560 	for (;;) {
1561 		if (PCI_EXTCAP_ID(ecap) == capability) {
1562 			if (capreg != NULL)
1563 				*capreg = ptr;
1564 			return (0);
1565 		}
1566 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1567 		if (ptr == 0)
1568 			break;
1569 		ecap = pci_read_config(child, ptr, 4);
1570 	}
1571 
1572 	return (ENOENT);
1573 }
1574 
1575 /*
1576  * Find the next requested extended capability after start and return the
1577  * offset in configuration space via the pointer provided.  The function
1578  * returns 0 on success and an error code otherwise.
1579  */
1580 int
1581 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1582     int start, int *capreg)
1583 {
1584 	struct pci_devinfo *dinfo = device_get_ivars(child);
1585 	pcicfgregs *cfg = &dinfo->cfg;
1586 	uint32_t ecap;
1587 	uint16_t ptr;
1588 
1589 	/* Only supported for PCI-express devices. */
1590 	if (cfg->pcie.pcie_location == 0)
1591 		return (ENXIO);
1592 
1593 	ecap = pci_read_config(child, start, 4);
1594 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1595 	    ("start extended capability is not expected capability"));
1596 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1597 	while (ptr != 0) {
1598 		ecap = pci_read_config(child, ptr, 4);
1599 		if (PCI_EXTCAP_ID(ecap) == capability) {
1600 			if (capreg != NULL)
1601 				*capreg = ptr;
1602 			return (0);
1603 		}
1604 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1605 	}
1606 
1607 	return (ENOENT);
1608 }
1609 
1610 /*
1611  * Support for MSI-X message interrupts.
1612  */
1613 static void
1614 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1615 {
1616 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1617 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1618 	uint32_t offset;
1619 
1620 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1621 	offset = msix->msix_table_offset + index * 16;
1622 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1623 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1624 	bus_write_4(msix->msix_table_res, offset + 8, data);
1625 }
1626 
1627 void
1628 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1629     uint64_t address, uint32_t data)
1630 {
1631 
1632 	if (pci_msix_rewrite_table) {
1633 		struct pci_devinfo *dinfo = device_get_ivars(child);
1634 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1635 
1636 		/*
1637 		 * Some VM hosts require MSIX to be disabled in the
1638 		 * control register before updating the MSIX table
1639 		 * entries are allowed. It is not enough to only
1640 		 * disable MSIX while updating a single entry. MSIX
1641 		 * must be disabled while updating all entries in the
1642 		 * table.
1643 		 */
1644 		pci_write_config(child,
1645 		    msix->msix_location + PCIR_MSIX_CTRL,
1646 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1647 		pci_resume_msix(child);
1648 	} else
1649 		pci_write_msix_entry(child, index, address, data);
1650 
1651 	/* Enable MSI -> HT mapping. */
1652 	pci_ht_map_msi(child, address);
1653 }
1654 
1655 void
1656 pci_mask_msix(device_t dev, u_int index)
1657 {
1658 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1659 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1660 	uint32_t offset, val;
1661 
1662 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1663 	offset = msix->msix_table_offset + index * 16 + 12;
1664 	val = bus_read_4(msix->msix_table_res, offset);
1665 	val |= PCIM_MSIX_VCTRL_MASK;
1666 
1667 	/*
1668 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1669 	 * register, so always write the new value.
1670 	 */
1671 	bus_write_4(msix->msix_table_res, offset, val);
1672 }
1673 
1674 void
1675 pci_unmask_msix(device_t dev, u_int index)
1676 {
1677 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1678 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1679 	uint32_t offset, val;
1680 
1681 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1682 	offset = msix->msix_table_offset + index * 16 + 12;
1683 	val = bus_read_4(msix->msix_table_res, offset);
1684 	val &= ~PCIM_MSIX_VCTRL_MASK;
1685 
1686 	/*
1687 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1688 	 * register, so always write the new value.
1689 	 */
1690 	bus_write_4(msix->msix_table_res, offset, val);
1691 }
1692 
1693 int
1694 pci_pending_msix(device_t dev, u_int index)
1695 {
1696 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1697 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1698 	uint32_t offset, bit;
1699 
1700 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1701 	offset = msix->msix_pba_offset + (index / 32) * 4;
1702 	bit = 1 << index % 32;
1703 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1704 }
1705 
1706 /*
1707  * Restore MSI-X registers and table during resume.  If MSI-X is
1708  * enabled then walk the virtual table to restore the actual MSI-X
1709  * table.
1710  */
1711 static void
1712 pci_resume_msix(device_t dev)
1713 {
1714 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1715 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1716 	struct msix_table_entry *mte;
1717 	struct msix_vector *mv;
1718 	int i;
1719 
1720 	if (msix->msix_alloc > 0) {
1721 		/* First, mask all vectors. */
1722 		for (i = 0; i < msix->msix_msgnum; i++)
1723 			pci_mask_msix(dev, i);
1724 
1725 		/* Second, program any messages with at least one handler. */
1726 		for (i = 0; i < msix->msix_table_len; i++) {
1727 			mte = &msix->msix_table[i];
1728 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1729 				continue;
1730 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1731 			pci_write_msix_entry(dev, i, mv->mv_address,
1732 			    mv->mv_data);
1733 			pci_unmask_msix(dev, i);
1734 		}
1735 	}
1736 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1737 	    msix->msix_ctrl, 2);
1738 }
1739 
1740 /*
1741  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1742  * returned in *count.  After this function returns, each message will be
1743  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1744  */
1745 int
1746 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1747 {
1748 	struct pci_devinfo *dinfo = device_get_ivars(child);
1749 	pcicfgregs *cfg = &dinfo->cfg;
1750 	struct resource_list_entry *rle;
1751 	int actual, error, i, irq, max;
1752 
1753 	/* Don't let count == 0 get us into trouble. */
1754 	if (*count == 0)
1755 		return (EINVAL);
1756 
1757 	/* If rid 0 is allocated, then fail. */
1758 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1759 	if (rle != NULL && rle->res != NULL)
1760 		return (ENXIO);
1761 
1762 	/* Already have allocated messages? */
1763 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1764 		return (ENXIO);
1765 
1766 	/* If MSI-X is blacklisted for this system, fail. */
1767 	if (pci_msix_blacklisted())
1768 		return (ENXIO);
1769 
1770 	/* MSI-X capability present? */
1771 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1772 		return (ENODEV);
1773 
1774 	/* Make sure the appropriate BARs are mapped. */
1775 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1776 	    cfg->msix.msix_table_bar);
1777 	if (rle == NULL || rle->res == NULL ||
1778 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1779 		return (ENXIO);
1780 	cfg->msix.msix_table_res = rle->res;
1781 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1782 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1783 		    cfg->msix.msix_pba_bar);
1784 		if (rle == NULL || rle->res == NULL ||
1785 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1786 			return (ENXIO);
1787 	}
1788 	cfg->msix.msix_pba_res = rle->res;
1789 
1790 	if (bootverbose)
1791 		device_printf(child,
1792 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1793 		    *count, cfg->msix.msix_msgnum);
1794 	max = min(*count, cfg->msix.msix_msgnum);
1795 	for (i = 0; i < max; i++) {
1796 		/* Allocate a message. */
1797 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1798 		if (error) {
1799 			if (i == 0)
1800 				return (error);
1801 			break;
1802 		}
1803 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1804 		    irq, 1);
1805 	}
1806 	actual = i;
1807 
1808 	if (bootverbose) {
1809 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1810 		if (actual == 1)
1811 			device_printf(child, "using IRQ %ju for MSI-X\n",
1812 			    rle->start);
1813 		else {
1814 			int run;
1815 
1816 			/*
1817 			 * Be fancy and try to print contiguous runs of
1818 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1819 			 * 'run' is true if we are in a range.
1820 			 */
1821 			device_printf(child, "using IRQs %ju", rle->start);
1822 			irq = rle->start;
1823 			run = 0;
1824 			for (i = 1; i < actual; i++) {
1825 				rle = resource_list_find(&dinfo->resources,
1826 				    SYS_RES_IRQ, i + 1);
1827 
1828 				/* Still in a run? */
1829 				if (rle->start == irq + 1) {
1830 					run = 1;
1831 					irq++;
1832 					continue;
1833 				}
1834 
1835 				/* Finish previous range. */
1836 				if (run) {
1837 					printf("-%d", irq);
1838 					run = 0;
1839 				}
1840 
1841 				/* Start new range. */
1842 				printf(",%ju", rle->start);
1843 				irq = rle->start;
1844 			}
1845 
1846 			/* Unfinished range? */
1847 			if (run)
1848 				printf("-%d", irq);
1849 			printf(" for MSI-X\n");
1850 		}
1851 	}
1852 
1853 	/* Mask all vectors. */
1854 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1855 		pci_mask_msix(child, i);
1856 
1857 	/* Allocate and initialize vector data and virtual table. */
1858 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1859 	    M_DEVBUF, M_WAITOK | M_ZERO);
1860 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1861 	    M_DEVBUF, M_WAITOK | M_ZERO);
1862 	for (i = 0; i < actual; i++) {
1863 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1864 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1865 		cfg->msix.msix_table[i].mte_vector = i + 1;
1866 	}
1867 
1868 	/* Update control register to enable MSI-X. */
1869 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1870 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1871 	    cfg->msix.msix_ctrl, 2);
1872 
1873 	/* Update counts of alloc'd messages. */
1874 	cfg->msix.msix_alloc = actual;
1875 	cfg->msix.msix_table_len = actual;
1876 	*count = actual;
1877 	return (0);
1878 }
1879 
1880 /*
1881  * By default, pci_alloc_msix() will assign the allocated IRQ
1882  * resources consecutively to the first N messages in the MSI-X table.
1883  * However, device drivers may want to use different layouts if they
1884  * either receive fewer messages than they asked for, or they wish to
1885  * populate the MSI-X table sparsely.  This method allows the driver
1886  * to specify what layout it wants.  It must be called after a
1887  * successful pci_alloc_msix() but before any of the associated
1888  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1889  *
1890  * The 'vectors' array contains 'count' message vectors.  The array
1891  * maps directly to the MSI-X table in that index 0 in the array
1892  * specifies the vector for the first message in the MSI-X table, etc.
1893  * The vector value in each array index can either be 0 to indicate
1894  * that no vector should be assigned to a message slot, or it can be a
1895  * number from 1 to N (where N is the count returned from a
1896  * succcessful call to pci_alloc_msix()) to indicate which message
1897  * vector (IRQ) to be used for the corresponding message.
1898  *
1899  * On successful return, each message with a non-zero vector will have
1900  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1901  * 1.  Additionally, if any of the IRQs allocated via the previous
1902  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1903  * will be freed back to the system automatically.
1904  *
1905  * For example, suppose a driver has a MSI-X table with 6 messages and
1906  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1907  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1908  * C.  After the call to pci_alloc_msix(), the device will be setup to
1909  * have an MSI-X table of ABC--- (where - means no vector assigned).
1910  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1911  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1912  * be freed back to the system.  This device will also have valid
1913  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1914  *
1915  * In any case, the SYS_RES_IRQ rid X will always map to the message
1916  * at MSI-X table index X - 1 and will only be valid if a vector is
1917  * assigned to that table entry.
1918  */
1919 int
1920 pci_remap_msix_method(device_t dev, device_t child, int count,
1921     const u_int *vectors)
1922 {
1923 	struct pci_devinfo *dinfo = device_get_ivars(child);
1924 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1925 	struct resource_list_entry *rle;
1926 	int i, irq, j, *used;
1927 
1928 	/*
1929 	 * Have to have at least one message in the table but the
1930 	 * table can't be bigger than the actual MSI-X table in the
1931 	 * device.
1932 	 */
1933 	if (count == 0 || count > msix->msix_msgnum)
1934 		return (EINVAL);
1935 
1936 	/* Sanity check the vectors. */
1937 	for (i = 0; i < count; i++)
1938 		if (vectors[i] > msix->msix_alloc)
1939 			return (EINVAL);
1940 
1941 	/*
1942 	 * Make sure there aren't any holes in the vectors to be used.
1943 	 * It's a big pain to support it, and it doesn't really make
1944 	 * sense anyway.  Also, at least one vector must be used.
1945 	 */
1946 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1947 	    M_ZERO);
1948 	for (i = 0; i < count; i++)
1949 		if (vectors[i] != 0)
1950 			used[vectors[i] - 1] = 1;
1951 	for (i = 0; i < msix->msix_alloc - 1; i++)
1952 		if (used[i] == 0 && used[i + 1] == 1) {
1953 			free(used, M_DEVBUF);
1954 			return (EINVAL);
1955 		}
1956 	if (used[0] != 1) {
1957 		free(used, M_DEVBUF);
1958 		return (EINVAL);
1959 	}
1960 
1961 	/* Make sure none of the resources are allocated. */
1962 	for (i = 0; i < msix->msix_table_len; i++) {
1963 		if (msix->msix_table[i].mte_vector == 0)
1964 			continue;
1965 		if (msix->msix_table[i].mte_handlers > 0) {
1966 			free(used, M_DEVBUF);
1967 			return (EBUSY);
1968 		}
1969 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1970 		KASSERT(rle != NULL, ("missing resource"));
1971 		if (rle->res != NULL) {
1972 			free(used, M_DEVBUF);
1973 			return (EBUSY);
1974 		}
1975 	}
1976 
1977 	/* Free the existing resource list entries. */
1978 	for (i = 0; i < msix->msix_table_len; i++) {
1979 		if (msix->msix_table[i].mte_vector == 0)
1980 			continue;
1981 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1982 	}
1983 
1984 	/*
1985 	 * Build the new virtual table keeping track of which vectors are
1986 	 * used.
1987 	 */
1988 	free(msix->msix_table, M_DEVBUF);
1989 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1990 	    M_DEVBUF, M_WAITOK | M_ZERO);
1991 	for (i = 0; i < count; i++)
1992 		msix->msix_table[i].mte_vector = vectors[i];
1993 	msix->msix_table_len = count;
1994 
1995 	/* Free any unused IRQs and resize the vectors array if necessary. */
1996 	j = msix->msix_alloc - 1;
1997 	if (used[j] == 0) {
1998 		struct msix_vector *vec;
1999 
2000 		while (used[j] == 0) {
2001 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2002 			    msix->msix_vectors[j].mv_irq);
2003 			j--;
2004 		}
2005 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2006 		    M_WAITOK);
2007 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2008 		    (j + 1));
2009 		free(msix->msix_vectors, M_DEVBUF);
2010 		msix->msix_vectors = vec;
2011 		msix->msix_alloc = j + 1;
2012 	}
2013 	free(used, M_DEVBUF);
2014 
2015 	/* Map the IRQs onto the rids. */
2016 	for (i = 0; i < count; i++) {
2017 		if (vectors[i] == 0)
2018 			continue;
2019 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2020 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2021 		    irq, 1);
2022 	}
2023 
2024 	if (bootverbose) {
2025 		device_printf(child, "Remapped MSI-X IRQs as: ");
2026 		for (i = 0; i < count; i++) {
2027 			if (i != 0)
2028 				printf(", ");
2029 			if (vectors[i] == 0)
2030 				printf("---");
2031 			else
2032 				printf("%d",
2033 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2034 		}
2035 		printf("\n");
2036 	}
2037 
2038 	return (0);
2039 }
2040 
2041 static int
2042 pci_release_msix(device_t dev, device_t child)
2043 {
2044 	struct pci_devinfo *dinfo = device_get_ivars(child);
2045 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2046 	struct resource_list_entry *rle;
2047 	int i;
2048 
2049 	/* Do we have any messages to release? */
2050 	if (msix->msix_alloc == 0)
2051 		return (ENODEV);
2052 
2053 	/* Make sure none of the resources are allocated. */
2054 	for (i = 0; i < msix->msix_table_len; i++) {
2055 		if (msix->msix_table[i].mte_vector == 0)
2056 			continue;
2057 		if (msix->msix_table[i].mte_handlers > 0)
2058 			return (EBUSY);
2059 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2060 		KASSERT(rle != NULL, ("missing resource"));
2061 		if (rle->res != NULL)
2062 			return (EBUSY);
2063 	}
2064 
2065 	/* Update control register to disable MSI-X. */
2066 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2067 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2068 	    msix->msix_ctrl, 2);
2069 
2070 	/* Free the resource list entries. */
2071 	for (i = 0; i < msix->msix_table_len; i++) {
2072 		if (msix->msix_table[i].mte_vector == 0)
2073 			continue;
2074 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2075 	}
2076 	free(msix->msix_table, M_DEVBUF);
2077 	msix->msix_table_len = 0;
2078 
2079 	/* Release the IRQs. */
2080 	for (i = 0; i < msix->msix_alloc; i++)
2081 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2082 		    msix->msix_vectors[i].mv_irq);
2083 	free(msix->msix_vectors, M_DEVBUF);
2084 	msix->msix_alloc = 0;
2085 	return (0);
2086 }
2087 
2088 /*
2089  * Return the max supported MSI-X messages this device supports.
2090  * Basically, assuming the MD code can alloc messages, this function
2091  * should return the maximum value that pci_alloc_msix() can return.
2092  * Thus, it is subject to the tunables, etc.
2093  */
2094 int
2095 pci_msix_count_method(device_t dev, device_t child)
2096 {
2097 	struct pci_devinfo *dinfo = device_get_ivars(child);
2098 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2099 
2100 	if (pci_do_msix && msix->msix_location != 0)
2101 		return (msix->msix_msgnum);
2102 	return (0);
2103 }
2104 
2105 int
2106 pci_msix_pba_bar_method(device_t dev, device_t child)
2107 {
2108 	struct pci_devinfo *dinfo = device_get_ivars(child);
2109 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2110 
2111 	if (pci_do_msix && msix->msix_location != 0)
2112 		return (msix->msix_pba_bar);
2113 	return (-1);
2114 }
2115 
2116 int
2117 pci_msix_table_bar_method(device_t dev, device_t child)
2118 {
2119 	struct pci_devinfo *dinfo = device_get_ivars(child);
2120 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2121 
2122 	if (pci_do_msix && msix->msix_location != 0)
2123 		return (msix->msix_table_bar);
2124 	return (-1);
2125 }
2126 
2127 /*
2128  * HyperTransport MSI mapping control
2129  */
2130 void
2131 pci_ht_map_msi(device_t dev, uint64_t addr)
2132 {
2133 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2134 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2135 
2136 	if (!ht->ht_msimap)
2137 		return;
2138 
2139 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2140 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2141 		/* Enable MSI -> HT mapping. */
2142 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2143 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2144 		    ht->ht_msictrl, 2);
2145 	}
2146 
2147 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2148 		/* Disable 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 
2155 int
2156 pci_get_max_payload(device_t dev)
2157 {
2158 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2159 	int cap;
2160 	uint16_t val;
2161 
2162 	cap = dinfo->cfg.pcie.pcie_location;
2163 	if (cap == 0)
2164 		return (0);
2165 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2166 	val &= PCIEM_CTL_MAX_PAYLOAD;
2167 	val >>= 5;
2168 	return (1 << (val + 7));
2169 }
2170 
2171 int
2172 pci_get_max_read_req(device_t dev)
2173 {
2174 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2175 	int cap;
2176 	uint16_t val;
2177 
2178 	cap = dinfo->cfg.pcie.pcie_location;
2179 	if (cap == 0)
2180 		return (0);
2181 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2182 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2183 	val >>= 12;
2184 	return (1 << (val + 7));
2185 }
2186 
2187 int
2188 pci_set_max_read_req(device_t dev, int size)
2189 {
2190 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2191 	int cap;
2192 	uint16_t val;
2193 
2194 	cap = dinfo->cfg.pcie.pcie_location;
2195 	if (cap == 0)
2196 		return (0);
2197 	if (size < 128)
2198 		size = 128;
2199 	if (size > 4096)
2200 		size = 4096;
2201 	size = (1 << (fls(size) - 1));
2202 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2203 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2204 	val |= (fls(size) - 8) << 12;
2205 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2206 	return (size);
2207 }
2208 
2209 uint32_t
2210 pcie_read_config(device_t dev, int reg, int width)
2211 {
2212 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2213 	int cap;
2214 
2215 	cap = dinfo->cfg.pcie.pcie_location;
2216 	if (cap == 0) {
2217 		if (width == 2)
2218 			return (0xffff);
2219 		return (0xffffffff);
2220 	}
2221 
2222 	return (pci_read_config(dev, cap + reg, width));
2223 }
2224 
2225 void
2226 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2227 {
2228 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2229 	int cap;
2230 
2231 	cap = dinfo->cfg.pcie.pcie_location;
2232 	if (cap == 0)
2233 		return;
2234 	pci_write_config(dev, cap + reg, value, width);
2235 }
2236 
2237 /*
2238  * Adjusts a PCI-e capability register by clearing the bits in mask
2239  * and setting the bits in (value & mask).  Bits not set in mask are
2240  * not adjusted.
2241  *
2242  * Returns the old value on success or all ones on failure.
2243  */
2244 uint32_t
2245 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2246     int width)
2247 {
2248 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2249 	uint32_t old, new;
2250 	int cap;
2251 
2252 	cap = dinfo->cfg.pcie.pcie_location;
2253 	if (cap == 0) {
2254 		if (width == 2)
2255 			return (0xffff);
2256 		return (0xffffffff);
2257 	}
2258 
2259 	old = pci_read_config(dev, cap + reg, width);
2260 	new = old & ~mask;
2261 	new |= (value & mask);
2262 	pci_write_config(dev, cap + reg, new, width);
2263 	return (old);
2264 }
2265 
2266 /*
2267  * Support for MSI message signalled interrupts.
2268  */
2269 void
2270 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2271     uint16_t data)
2272 {
2273 	struct pci_devinfo *dinfo = device_get_ivars(child);
2274 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2275 
2276 	/* Write data and address values. */
2277 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2278 	    address & 0xffffffff, 4);
2279 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2280 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2281 		    address >> 32, 4);
2282 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2283 		    data, 2);
2284 	} else
2285 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2286 		    2);
2287 
2288 	/* Enable MSI in the control register. */
2289 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2290 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2291 	    msi->msi_ctrl, 2);
2292 
2293 	/* Enable MSI -> HT mapping. */
2294 	pci_ht_map_msi(child, address);
2295 }
2296 
2297 void
2298 pci_disable_msi_method(device_t dev, device_t child)
2299 {
2300 	struct pci_devinfo *dinfo = device_get_ivars(child);
2301 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2302 
2303 	/* Disable MSI -> HT mapping. */
2304 	pci_ht_map_msi(child, 0);
2305 
2306 	/* Disable MSI in the control register. */
2307 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2308 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2309 	    msi->msi_ctrl, 2);
2310 }
2311 
2312 /*
2313  * Restore MSI registers during resume.  If MSI is enabled then
2314  * restore the data and address registers in addition to the control
2315  * register.
2316  */
2317 static void
2318 pci_resume_msi(device_t dev)
2319 {
2320 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2321 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2322 	uint64_t address;
2323 	uint16_t data;
2324 
2325 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2326 		address = msi->msi_addr;
2327 		data = msi->msi_data;
2328 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2329 		    address & 0xffffffff, 4);
2330 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2331 			pci_write_config(dev, msi->msi_location +
2332 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2333 			pci_write_config(dev, msi->msi_location +
2334 			    PCIR_MSI_DATA_64BIT, data, 2);
2335 		} else
2336 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2337 			    data, 2);
2338 	}
2339 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2340 	    2);
2341 }
2342 
2343 static int
2344 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2345 {
2346 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2347 	pcicfgregs *cfg = &dinfo->cfg;
2348 	struct resource_list_entry *rle;
2349 	struct msix_table_entry *mte;
2350 	struct msix_vector *mv;
2351 	uint64_t addr;
2352 	uint32_t data;
2353 	int error, i, j;
2354 
2355 	/*
2356 	 * Handle MSI first.  We try to find this IRQ among our list
2357 	 * of MSI IRQs.  If we find it, we request updated address and
2358 	 * data registers and apply the results.
2359 	 */
2360 	if (cfg->msi.msi_alloc > 0) {
2361 
2362 		/* If we don't have any active handlers, nothing to do. */
2363 		if (cfg->msi.msi_handlers == 0)
2364 			return (0);
2365 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2366 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2367 			    i + 1);
2368 			if (rle->start == irq) {
2369 				error = PCIB_MAP_MSI(device_get_parent(bus),
2370 				    dev, irq, &addr, &data);
2371 				if (error)
2372 					return (error);
2373 				pci_disable_msi(dev);
2374 				dinfo->cfg.msi.msi_addr = addr;
2375 				dinfo->cfg.msi.msi_data = data;
2376 				pci_enable_msi(dev, addr, data);
2377 				return (0);
2378 			}
2379 		}
2380 		return (ENOENT);
2381 	}
2382 
2383 	/*
2384 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2385 	 * we request the updated mapping info.  If that works, we go
2386 	 * through all the slots that use this IRQ and update them.
2387 	 */
2388 	if (cfg->msix.msix_alloc > 0) {
2389 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2390 			mv = &cfg->msix.msix_vectors[i];
2391 			if (mv->mv_irq == irq) {
2392 				error = PCIB_MAP_MSI(device_get_parent(bus),
2393 				    dev, irq, &addr, &data);
2394 				if (error)
2395 					return (error);
2396 				mv->mv_address = addr;
2397 				mv->mv_data = data;
2398 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2399 					mte = &cfg->msix.msix_table[j];
2400 					if (mte->mte_vector != i + 1)
2401 						continue;
2402 					if (mte->mte_handlers == 0)
2403 						continue;
2404 					pci_mask_msix(dev, j);
2405 					pci_enable_msix(dev, j, addr, data);
2406 					pci_unmask_msix(dev, j);
2407 				}
2408 			}
2409 		}
2410 		return (ENOENT);
2411 	}
2412 
2413 	return (ENOENT);
2414 }
2415 
2416 /*
2417  * Returns true if the specified device is blacklisted because MSI
2418  * doesn't work.
2419  */
2420 int
2421 pci_msi_device_blacklisted(device_t dev)
2422 {
2423 
2424 	if (!pci_honor_msi_blacklist)
2425 		return (0);
2426 
2427 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2428 }
2429 
2430 /*
2431  * Determine if MSI is blacklisted globally on this system.  Currently,
2432  * we just check for blacklisted chipsets as represented by the
2433  * host-PCI bridge at device 0:0:0.  In the future, it may become
2434  * necessary to check other system attributes, such as the kenv values
2435  * that give the motherboard manufacturer and model number.
2436  */
2437 static int
2438 pci_msi_blacklisted(void)
2439 {
2440 	device_t dev;
2441 
2442 	if (!pci_honor_msi_blacklist)
2443 		return (0);
2444 
2445 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2446 	if (!(pcie_chipset || pcix_chipset)) {
2447 		if (vm_guest != VM_GUEST_NO) {
2448 			/*
2449 			 * Whitelist older chipsets in virtual
2450 			 * machines known to support MSI.
2451 			 */
2452 			dev = pci_find_bsf(0, 0, 0);
2453 			if (dev != NULL)
2454 				return (!pci_has_quirk(pci_get_devid(dev),
2455 					PCI_QUIRK_ENABLE_MSI_VM));
2456 		}
2457 		return (1);
2458 	}
2459 
2460 	dev = pci_find_bsf(0, 0, 0);
2461 	if (dev != NULL)
2462 		return (pci_msi_device_blacklisted(dev));
2463 	return (0);
2464 }
2465 
2466 /*
2467  * Returns true if the specified device is blacklisted because MSI-X
2468  * doesn't work.  Note that this assumes that if MSI doesn't work,
2469  * MSI-X doesn't either.
2470  */
2471 int
2472 pci_msix_device_blacklisted(device_t dev)
2473 {
2474 
2475 	if (!pci_honor_msi_blacklist)
2476 		return (0);
2477 
2478 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2479 		return (1);
2480 
2481 	return (pci_msi_device_blacklisted(dev));
2482 }
2483 
2484 /*
2485  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2486  * is blacklisted, assume that MSI-X is as well.  Check for additional
2487  * chipsets where MSI works but MSI-X does not.
2488  */
2489 static int
2490 pci_msix_blacklisted(void)
2491 {
2492 	device_t dev;
2493 
2494 	if (!pci_honor_msi_blacklist)
2495 		return (0);
2496 
2497 	dev = pci_find_bsf(0, 0, 0);
2498 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2499 	    PCI_QUIRK_DISABLE_MSIX))
2500 		return (1);
2501 
2502 	return (pci_msi_blacklisted());
2503 }
2504 
2505 /*
2506  * Attempt to allocate *count MSI messages.  The actual number allocated is
2507  * returned in *count.  After this function returns, each message will be
2508  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2509  */
2510 int
2511 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2512 {
2513 	struct pci_devinfo *dinfo = device_get_ivars(child);
2514 	pcicfgregs *cfg = &dinfo->cfg;
2515 	struct resource_list_entry *rle;
2516 	int actual, error, i, irqs[32];
2517 	uint16_t ctrl;
2518 
2519 	/* Don't let count == 0 get us into trouble. */
2520 	if (*count == 0)
2521 		return (EINVAL);
2522 
2523 	/* If rid 0 is allocated, then fail. */
2524 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2525 	if (rle != NULL && rle->res != NULL)
2526 		return (ENXIO);
2527 
2528 	/* Already have allocated messages? */
2529 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2530 		return (ENXIO);
2531 
2532 	/* If MSI is blacklisted for this system, fail. */
2533 	if (pci_msi_blacklisted())
2534 		return (ENXIO);
2535 
2536 	/* MSI capability present? */
2537 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2538 		return (ENODEV);
2539 
2540 	if (bootverbose)
2541 		device_printf(child,
2542 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2543 		    *count, cfg->msi.msi_msgnum);
2544 
2545 	/* Don't ask for more than the device supports. */
2546 	actual = min(*count, cfg->msi.msi_msgnum);
2547 
2548 	/* Don't ask for more than 32 messages. */
2549 	actual = min(actual, 32);
2550 
2551 	/* MSI requires power of 2 number of messages. */
2552 	if (!powerof2(actual))
2553 		return (EINVAL);
2554 
2555 	for (;;) {
2556 		/* Try to allocate N messages. */
2557 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2558 		    actual, irqs);
2559 		if (error == 0)
2560 			break;
2561 		if (actual == 1)
2562 			return (error);
2563 
2564 		/* Try N / 2. */
2565 		actual >>= 1;
2566 	}
2567 
2568 	/*
2569 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2570 	 * resources in the irqs[] array, so add new resources
2571 	 * starting at rid 1.
2572 	 */
2573 	for (i = 0; i < actual; i++)
2574 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2575 		    irqs[i], irqs[i], 1);
2576 
2577 	if (bootverbose) {
2578 		if (actual == 1)
2579 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2580 		else {
2581 			int run;
2582 
2583 			/*
2584 			 * Be fancy and try to print contiguous runs
2585 			 * of IRQ values as ranges.  'run' is true if
2586 			 * we are in a range.
2587 			 */
2588 			device_printf(child, "using IRQs %d", irqs[0]);
2589 			run = 0;
2590 			for (i = 1; i < actual; i++) {
2591 
2592 				/* Still in a run? */
2593 				if (irqs[i] == irqs[i - 1] + 1) {
2594 					run = 1;
2595 					continue;
2596 				}
2597 
2598 				/* Finish previous range. */
2599 				if (run) {
2600 					printf("-%d", irqs[i - 1]);
2601 					run = 0;
2602 				}
2603 
2604 				/* Start new range. */
2605 				printf(",%d", irqs[i]);
2606 			}
2607 
2608 			/* Unfinished range? */
2609 			if (run)
2610 				printf("-%d", irqs[actual - 1]);
2611 			printf(" for MSI\n");
2612 		}
2613 	}
2614 
2615 	/* Update control register with actual count. */
2616 	ctrl = cfg->msi.msi_ctrl;
2617 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2618 	ctrl |= (ffs(actual) - 1) << 4;
2619 	cfg->msi.msi_ctrl = ctrl;
2620 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2621 
2622 	/* Update counts of alloc'd messages. */
2623 	cfg->msi.msi_alloc = actual;
2624 	cfg->msi.msi_handlers = 0;
2625 	*count = actual;
2626 	return (0);
2627 }
2628 
2629 /* Release the MSI messages associated with this device. */
2630 int
2631 pci_release_msi_method(device_t dev, device_t child)
2632 {
2633 	struct pci_devinfo *dinfo = device_get_ivars(child);
2634 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2635 	struct resource_list_entry *rle;
2636 	int error, i, irqs[32];
2637 
2638 	/* Try MSI-X first. */
2639 	error = pci_release_msix(dev, child);
2640 	if (error != ENODEV)
2641 		return (error);
2642 
2643 	/* Do we have any messages to release? */
2644 	if (msi->msi_alloc == 0)
2645 		return (ENODEV);
2646 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2647 
2648 	/* Make sure none of the resources are allocated. */
2649 	if (msi->msi_handlers > 0)
2650 		return (EBUSY);
2651 	for (i = 0; i < msi->msi_alloc; i++) {
2652 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2653 		KASSERT(rle != NULL, ("missing MSI resource"));
2654 		if (rle->res != NULL)
2655 			return (EBUSY);
2656 		irqs[i] = rle->start;
2657 	}
2658 
2659 	/* Update control register with 0 count. */
2660 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2661 	    ("%s: MSI still enabled", __func__));
2662 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2663 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2664 	    msi->msi_ctrl, 2);
2665 
2666 	/* Release the messages. */
2667 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2668 	for (i = 0; i < msi->msi_alloc; i++)
2669 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2670 
2671 	/* Update alloc count. */
2672 	msi->msi_alloc = 0;
2673 	msi->msi_addr = 0;
2674 	msi->msi_data = 0;
2675 	return (0);
2676 }
2677 
2678 /*
2679  * Return the max supported MSI messages this device supports.
2680  * Basically, assuming the MD code can alloc messages, this function
2681  * should return the maximum value that pci_alloc_msi() can return.
2682  * Thus, it is subject to the tunables, etc.
2683  */
2684 int
2685 pci_msi_count_method(device_t dev, device_t child)
2686 {
2687 	struct pci_devinfo *dinfo = device_get_ivars(child);
2688 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2689 
2690 	if (pci_do_msi && msi->msi_location != 0)
2691 		return (msi->msi_msgnum);
2692 	return (0);
2693 }
2694 
2695 /* free pcicfgregs structure and all depending data structures */
2696 
2697 int
2698 pci_freecfg(struct pci_devinfo *dinfo)
2699 {
2700 	struct devlist *devlist_head;
2701 	struct pci_map *pm, *next;
2702 	int i;
2703 
2704 	devlist_head = &pci_devq;
2705 
2706 	if (dinfo->cfg.vpd.vpd_reg) {
2707 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2708 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2709 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2710 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2711 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2712 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2713 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2714 	}
2715 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2716 		free(pm, M_DEVBUF);
2717 	}
2718 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2719 	free(dinfo, M_DEVBUF);
2720 
2721 	/* increment the generation count */
2722 	pci_generation++;
2723 
2724 	/* we're losing one device */
2725 	pci_numdevs--;
2726 	return (0);
2727 }
2728 
2729 /*
2730  * PCI power manangement
2731  */
2732 int
2733 pci_set_powerstate_method(device_t dev, device_t child, int state)
2734 {
2735 	struct pci_devinfo *dinfo = device_get_ivars(child);
2736 	pcicfgregs *cfg = &dinfo->cfg;
2737 	uint16_t status;
2738 	int oldstate, highest, delay;
2739 
2740 	if (cfg->pp.pp_cap == 0)
2741 		return (EOPNOTSUPP);
2742 
2743 	/*
2744 	 * Optimize a no state change request away.  While it would be OK to
2745 	 * write to the hardware in theory, some devices have shown odd
2746 	 * behavior when going from D3 -> D3.
2747 	 */
2748 	oldstate = pci_get_powerstate(child);
2749 	if (oldstate == state)
2750 		return (0);
2751 
2752 	/*
2753 	 * The PCI power management specification states that after a state
2754 	 * transition between PCI power states, system software must
2755 	 * guarantee a minimal delay before the function accesses the device.
2756 	 * Compute the worst case delay that we need to guarantee before we
2757 	 * access the device.  Many devices will be responsive much more
2758 	 * quickly than this delay, but there are some that don't respond
2759 	 * instantly to state changes.  Transitions to/from D3 state require
2760 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2761 	 * is done below with DELAY rather than a sleeper function because
2762 	 * this function can be called from contexts where we cannot sleep.
2763 	 */
2764 	highest = (oldstate > state) ? oldstate : state;
2765 	if (highest == PCI_POWERSTATE_D3)
2766 	    delay = 10000;
2767 	else if (highest == PCI_POWERSTATE_D2)
2768 	    delay = 200;
2769 	else
2770 	    delay = 0;
2771 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2772 	    & ~PCIM_PSTAT_DMASK;
2773 	switch (state) {
2774 	case PCI_POWERSTATE_D0:
2775 		status |= PCIM_PSTAT_D0;
2776 		break;
2777 	case PCI_POWERSTATE_D1:
2778 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2779 			return (EOPNOTSUPP);
2780 		status |= PCIM_PSTAT_D1;
2781 		break;
2782 	case PCI_POWERSTATE_D2:
2783 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2784 			return (EOPNOTSUPP);
2785 		status |= PCIM_PSTAT_D2;
2786 		break;
2787 	case PCI_POWERSTATE_D3:
2788 		status |= PCIM_PSTAT_D3;
2789 		break;
2790 	default:
2791 		return (EINVAL);
2792 	}
2793 
2794 	if (bootverbose)
2795 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2796 		    state);
2797 
2798 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2799 	if (delay)
2800 		DELAY(delay);
2801 	return (0);
2802 }
2803 
2804 int
2805 pci_get_powerstate_method(device_t dev, device_t child)
2806 {
2807 	struct pci_devinfo *dinfo = device_get_ivars(child);
2808 	pcicfgregs *cfg = &dinfo->cfg;
2809 	uint16_t status;
2810 	int result;
2811 
2812 	if (cfg->pp.pp_cap != 0) {
2813 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2814 		switch (status & PCIM_PSTAT_DMASK) {
2815 		case PCIM_PSTAT_D0:
2816 			result = PCI_POWERSTATE_D0;
2817 			break;
2818 		case PCIM_PSTAT_D1:
2819 			result = PCI_POWERSTATE_D1;
2820 			break;
2821 		case PCIM_PSTAT_D2:
2822 			result = PCI_POWERSTATE_D2;
2823 			break;
2824 		case PCIM_PSTAT_D3:
2825 			result = PCI_POWERSTATE_D3;
2826 			break;
2827 		default:
2828 			result = PCI_POWERSTATE_UNKNOWN;
2829 			break;
2830 		}
2831 	} else {
2832 		/* No support, device is always at D0 */
2833 		result = PCI_POWERSTATE_D0;
2834 	}
2835 	return (result);
2836 }
2837 
2838 /*
2839  * Some convenience functions for PCI device drivers.
2840  */
2841 
2842 static __inline void
2843 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2844 {
2845 	uint16_t	command;
2846 
2847 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2848 	command |= bit;
2849 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2850 }
2851 
2852 static __inline void
2853 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2854 {
2855 	uint16_t	command;
2856 
2857 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2858 	command &= ~bit;
2859 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2860 }
2861 
2862 int
2863 pci_enable_busmaster_method(device_t dev, device_t child)
2864 {
2865 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2866 	return (0);
2867 }
2868 
2869 int
2870 pci_disable_busmaster_method(device_t dev, device_t child)
2871 {
2872 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2873 	return (0);
2874 }
2875 
2876 int
2877 pci_enable_io_method(device_t dev, device_t child, int space)
2878 {
2879 	uint16_t bit;
2880 
2881 	switch(space) {
2882 	case SYS_RES_IOPORT:
2883 		bit = PCIM_CMD_PORTEN;
2884 		break;
2885 	case SYS_RES_MEMORY:
2886 		bit = PCIM_CMD_MEMEN;
2887 		break;
2888 	default:
2889 		return (EINVAL);
2890 	}
2891 	pci_set_command_bit(dev, child, bit);
2892 	return (0);
2893 }
2894 
2895 int
2896 pci_disable_io_method(device_t dev, device_t child, int space)
2897 {
2898 	uint16_t bit;
2899 
2900 	switch(space) {
2901 	case SYS_RES_IOPORT:
2902 		bit = PCIM_CMD_PORTEN;
2903 		break;
2904 	case SYS_RES_MEMORY:
2905 		bit = PCIM_CMD_MEMEN;
2906 		break;
2907 	default:
2908 		return (EINVAL);
2909 	}
2910 	pci_clear_command_bit(dev, child, bit);
2911 	return (0);
2912 }
2913 
2914 /*
2915  * New style pci driver.  Parent device is either a pci-host-bridge or a
2916  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2917  */
2918 
2919 void
2920 pci_print_verbose(struct pci_devinfo *dinfo)
2921 {
2922 
2923 	if (bootverbose) {
2924 		pcicfgregs *cfg = &dinfo->cfg;
2925 
2926 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2927 		    cfg->vendor, cfg->device, cfg->revid);
2928 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2929 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2930 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2931 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2932 		    cfg->mfdev);
2933 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2934 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2935 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2936 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2937 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2938 		if (cfg->intpin > 0)
2939 			printf("\tintpin=%c, irq=%d\n",
2940 			    cfg->intpin +'a' -1, cfg->intline);
2941 		if (cfg->pp.pp_cap) {
2942 			uint16_t status;
2943 
2944 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2945 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2946 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2947 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2948 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2949 			    status & PCIM_PSTAT_DMASK);
2950 		}
2951 		if (cfg->msi.msi_location) {
2952 			int ctrl;
2953 
2954 			ctrl = cfg->msi.msi_ctrl;
2955 			printf("\tMSI supports %d message%s%s%s\n",
2956 			    cfg->msi.msi_msgnum,
2957 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2958 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2959 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2960 		}
2961 		if (cfg->msix.msix_location) {
2962 			printf("\tMSI-X supports %d message%s ",
2963 			    cfg->msix.msix_msgnum,
2964 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2965 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2966 				printf("in map 0x%x\n",
2967 				    cfg->msix.msix_table_bar);
2968 			else
2969 				printf("in maps 0x%x and 0x%x\n",
2970 				    cfg->msix.msix_table_bar,
2971 				    cfg->msix.msix_pba_bar);
2972 		}
2973 	}
2974 }
2975 
2976 static int
2977 pci_porten(device_t dev)
2978 {
2979 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2980 }
2981 
2982 static int
2983 pci_memen(device_t dev)
2984 {
2985 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2986 }
2987 
2988 void
2989 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2990     int *bar64)
2991 {
2992 	struct pci_devinfo *dinfo;
2993 	pci_addr_t map, testval;
2994 	int ln2range;
2995 	uint16_t cmd;
2996 
2997 	/*
2998 	 * The device ROM BAR is special.  It is always a 32-bit
2999 	 * memory BAR.  Bit 0 is special and should not be set when
3000 	 * sizing the BAR.
3001 	 */
3002 	dinfo = device_get_ivars(dev);
3003 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3004 		map = pci_read_config(dev, reg, 4);
3005 		pci_write_config(dev, reg, 0xfffffffe, 4);
3006 		testval = pci_read_config(dev, reg, 4);
3007 		pci_write_config(dev, reg, map, 4);
3008 		*mapp = map;
3009 		*testvalp = testval;
3010 		if (bar64 != NULL)
3011 			*bar64 = 0;
3012 		return;
3013 	}
3014 
3015 	map = pci_read_config(dev, reg, 4);
3016 	ln2range = pci_maprange(map);
3017 	if (ln2range == 64)
3018 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3019 
3020 	/*
3021 	 * Disable decoding via the command register before
3022 	 * determining the BAR's length since we will be placing it in
3023 	 * a weird state.
3024 	 */
3025 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3026 	pci_write_config(dev, PCIR_COMMAND,
3027 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3028 
3029 	/*
3030 	 * Determine the BAR's length by writing all 1's.  The bottom
3031 	 * log_2(size) bits of the BAR will stick as 0 when we read
3032 	 * the value back.
3033 	 *
3034 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3035 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3036 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3037 	 *
3038 	 * Writes to both registers must be performed before attempting to
3039 	 * read back the size value.
3040 	 */
3041 	testval = 0;
3042 	pci_write_config(dev, reg, 0xffffffff, 4);
3043 	if (ln2range == 64) {
3044 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3045 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3046 	}
3047 	testval |= pci_read_config(dev, reg, 4);
3048 
3049 	/*
3050 	 * Restore the original value of the BAR.  We may have reprogrammed
3051 	 * the BAR of the low-level console device and when booting verbose,
3052 	 * we need the console device addressable.
3053 	 */
3054 	pci_write_config(dev, reg, map, 4);
3055 	if (ln2range == 64)
3056 		pci_write_config(dev, reg + 4, map >> 32, 4);
3057 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3058 
3059 	*mapp = map;
3060 	*testvalp = testval;
3061 	if (bar64 != NULL)
3062 		*bar64 = (ln2range == 64);
3063 }
3064 
3065 static void
3066 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3067 {
3068 	struct pci_devinfo *dinfo;
3069 	int ln2range;
3070 
3071 	/* The device ROM BAR is always a 32-bit memory BAR. */
3072 	dinfo = device_get_ivars(dev);
3073 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3074 		ln2range = 32;
3075 	else
3076 		ln2range = pci_maprange(pm->pm_value);
3077 	pci_write_config(dev, pm->pm_reg, base, 4);
3078 	if (ln2range == 64)
3079 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3080 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3081 	if (ln2range == 64)
3082 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3083 		    pm->pm_reg + 4, 4) << 32;
3084 }
3085 
3086 struct pci_map *
3087 pci_find_bar(device_t dev, int reg)
3088 {
3089 	struct pci_devinfo *dinfo;
3090 	struct pci_map *pm;
3091 
3092 	dinfo = device_get_ivars(dev);
3093 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3094 		if (pm->pm_reg == reg)
3095 			return (pm);
3096 	}
3097 	return (NULL);
3098 }
3099 
3100 int
3101 pci_bar_enabled(device_t dev, struct pci_map *pm)
3102 {
3103 	struct pci_devinfo *dinfo;
3104 	uint16_t cmd;
3105 
3106 	dinfo = device_get_ivars(dev);
3107 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3108 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3109 		return (0);
3110 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3111 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3112 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3113 	else
3114 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3115 }
3116 
3117 struct pci_map *
3118 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3119 {
3120 	struct pci_devinfo *dinfo;
3121 	struct pci_map *pm, *prev;
3122 
3123 	dinfo = device_get_ivars(dev);
3124 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3125 	pm->pm_reg = reg;
3126 	pm->pm_value = value;
3127 	pm->pm_size = size;
3128 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3129 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3130 		    reg));
3131 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3132 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3133 			break;
3134 	}
3135 	if (prev != NULL)
3136 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3137 	else
3138 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3139 	return (pm);
3140 }
3141 
3142 static void
3143 pci_restore_bars(device_t dev)
3144 {
3145 	struct pci_devinfo *dinfo;
3146 	struct pci_map *pm;
3147 	int ln2range;
3148 
3149 	dinfo = device_get_ivars(dev);
3150 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3151 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3152 			ln2range = 32;
3153 		else
3154 			ln2range = pci_maprange(pm->pm_value);
3155 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3156 		if (ln2range == 64)
3157 			pci_write_config(dev, pm->pm_reg + 4,
3158 			    pm->pm_value >> 32, 4);
3159 	}
3160 }
3161 
3162 /*
3163  * Add a resource based on a pci map register. Return 1 if the map
3164  * register is a 32bit map register or 2 if it is a 64bit register.
3165  */
3166 static int
3167 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3168     int force, int prefetch)
3169 {
3170 	struct pci_map *pm;
3171 	pci_addr_t base, map, testval;
3172 	pci_addr_t start, end, count;
3173 	int barlen, basezero, flags, maprange, mapsize, type;
3174 	uint16_t cmd;
3175 	struct resource *res;
3176 
3177 	/*
3178 	 * The BAR may already exist if the device is a CardBus card
3179 	 * whose CIS is stored in this BAR.
3180 	 */
3181 	pm = pci_find_bar(dev, reg);
3182 	if (pm != NULL) {
3183 		maprange = pci_maprange(pm->pm_value);
3184 		barlen = maprange == 64 ? 2 : 1;
3185 		return (barlen);
3186 	}
3187 
3188 	pci_read_bar(dev, reg, &map, &testval, NULL);
3189 	if (PCI_BAR_MEM(map)) {
3190 		type = SYS_RES_MEMORY;
3191 		if (map & PCIM_BAR_MEM_PREFETCH)
3192 			prefetch = 1;
3193 	} else
3194 		type = SYS_RES_IOPORT;
3195 	mapsize = pci_mapsize(testval);
3196 	base = pci_mapbase(map);
3197 #ifdef __PCI_BAR_ZERO_VALID
3198 	basezero = 0;
3199 #else
3200 	basezero = base == 0;
3201 #endif
3202 	maprange = pci_maprange(map);
3203 	barlen = maprange == 64 ? 2 : 1;
3204 
3205 	/*
3206 	 * For I/O registers, if bottom bit is set, and the next bit up
3207 	 * isn't clear, we know we have a BAR that doesn't conform to the
3208 	 * spec, so ignore it.  Also, sanity check the size of the data
3209 	 * areas to the type of memory involved.  Memory must be at least
3210 	 * 16 bytes in size, while I/O ranges must be at least 4.
3211 	 */
3212 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3213 		return (barlen);
3214 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3215 	    (type == SYS_RES_IOPORT && mapsize < 2))
3216 		return (barlen);
3217 
3218 	/* Save a record of this BAR. */
3219 	pm = pci_add_bar(dev, reg, map, mapsize);
3220 	if (bootverbose) {
3221 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3222 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3223 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3224 			printf(", port disabled\n");
3225 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3226 			printf(", memory disabled\n");
3227 		else
3228 			printf(", enabled\n");
3229 	}
3230 
3231 	/*
3232 	 * If base is 0, then we have problems if this architecture does
3233 	 * not allow that.  It is best to ignore such entries for the
3234 	 * moment.  These will be allocated later if the driver specifically
3235 	 * requests them.  However, some removable buses look better when
3236 	 * all resources are allocated, so allow '0' to be overriden.
3237 	 *
3238 	 * Similarly treat maps whose values is the same as the test value
3239 	 * read back.  These maps have had all f's written to them by the
3240 	 * BIOS in an attempt to disable the resources.
3241 	 */
3242 	if (!force && (basezero || map == testval))
3243 		return (barlen);
3244 	if ((u_long)base != base) {
3245 		device_printf(bus,
3246 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3247 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3248 		    pci_get_function(dev), reg);
3249 		return (barlen);
3250 	}
3251 
3252 	/*
3253 	 * This code theoretically does the right thing, but has
3254 	 * undesirable side effects in some cases where peripherals
3255 	 * respond oddly to having these bits enabled.  Let the user
3256 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3257 	 * default).
3258 	 */
3259 	if (pci_enable_io_modes) {
3260 		/* Turn on resources that have been left off by a lazy BIOS */
3261 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3262 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3263 			cmd |= PCIM_CMD_PORTEN;
3264 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3265 		}
3266 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3267 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3268 			cmd |= PCIM_CMD_MEMEN;
3269 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3270 		}
3271 	} else {
3272 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3273 			return (barlen);
3274 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3275 			return (barlen);
3276 	}
3277 
3278 	count = (pci_addr_t)1 << mapsize;
3279 	flags = RF_ALIGNMENT_LOG2(mapsize);
3280 	if (prefetch)
3281 		flags |= RF_PREFETCHABLE;
3282 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3283 		start = 0;	/* Let the parent decide. */
3284 		end = ~0;
3285 	} else {
3286 		start = base;
3287 		end = base + count - 1;
3288 	}
3289 	resource_list_add(rl, type, reg, start, end, count);
3290 
3291 	/*
3292 	 * Try to allocate the resource for this BAR from our parent
3293 	 * so that this resource range is already reserved.  The
3294 	 * driver for this device will later inherit this resource in
3295 	 * pci_alloc_resource().
3296 	 */
3297 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3298 	    flags);
3299 	if ((pci_do_realloc_bars
3300 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3301 	    && res == NULL && (start != 0 || end != ~0)) {
3302 		/*
3303 		 * If the allocation fails, try to allocate a resource for
3304 		 * this BAR using any available range.  The firmware felt
3305 		 * it was important enough to assign a resource, so don't
3306 		 * disable decoding if we can help it.
3307 		 */
3308 		resource_list_delete(rl, type, reg);
3309 		resource_list_add(rl, type, reg, 0, ~0, count);
3310 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3311 		    count, flags);
3312 	}
3313 	if (res == NULL) {
3314 		/*
3315 		 * If the allocation fails, delete the resource list entry
3316 		 * and disable decoding for this device.
3317 		 *
3318 		 * If the driver requests this resource in the future,
3319 		 * pci_reserve_map() will try to allocate a fresh
3320 		 * resource range.
3321 		 */
3322 		resource_list_delete(rl, type, reg);
3323 		pci_disable_io(dev, type);
3324 		if (bootverbose)
3325 			device_printf(bus,
3326 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3327 			    pci_get_domain(dev), pci_get_bus(dev),
3328 			    pci_get_slot(dev), pci_get_function(dev), reg);
3329 	} else {
3330 		start = rman_get_start(res);
3331 		pci_write_bar(dev, pm, start);
3332 	}
3333 	return (barlen);
3334 }
3335 
3336 /*
3337  * For ATA devices we need to decide early what addressing mode to use.
3338  * Legacy demands that the primary and secondary ATA ports sits on the
3339  * same addresses that old ISA hardware did. This dictates that we use
3340  * those addresses and ignore the BAR's if we cannot set PCI native
3341  * addressing mode.
3342  */
3343 static void
3344 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3345     uint32_t prefetchmask)
3346 {
3347 	int rid, type, progif;
3348 #if 0
3349 	/* if this device supports PCI native addressing use it */
3350 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3351 	if ((progif & 0x8a) == 0x8a) {
3352 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3353 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3354 			printf("Trying ATA native PCI addressing mode\n");
3355 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3356 		}
3357 	}
3358 #endif
3359 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3360 	type = SYS_RES_IOPORT;
3361 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3362 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3363 		    prefetchmask & (1 << 0));
3364 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3365 		    prefetchmask & (1 << 1));
3366 	} else {
3367 		rid = PCIR_BAR(0);
3368 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3369 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3370 		    0x1f7, 8, 0);
3371 		rid = PCIR_BAR(1);
3372 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3373 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3374 		    0x3f6, 1, 0);
3375 	}
3376 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3377 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3378 		    prefetchmask & (1 << 2));
3379 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3380 		    prefetchmask & (1 << 3));
3381 	} else {
3382 		rid = PCIR_BAR(2);
3383 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3384 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3385 		    0x177, 8, 0);
3386 		rid = PCIR_BAR(3);
3387 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3388 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3389 		    0x376, 1, 0);
3390 	}
3391 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3392 	    prefetchmask & (1 << 4));
3393 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3394 	    prefetchmask & (1 << 5));
3395 }
3396 
3397 static void
3398 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3399 {
3400 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3401 	pcicfgregs *cfg = &dinfo->cfg;
3402 	char tunable_name[64];
3403 	int irq;
3404 
3405 	/* Has to have an intpin to have an interrupt. */
3406 	if (cfg->intpin == 0)
3407 		return;
3408 
3409 	/* Let the user override the IRQ with a tunable. */
3410 	irq = PCI_INVALID_IRQ;
3411 	snprintf(tunable_name, sizeof(tunable_name),
3412 	    "hw.pci%d.%d.%d.INT%c.irq",
3413 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3414 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3415 		irq = PCI_INVALID_IRQ;
3416 
3417 	/*
3418 	 * If we didn't get an IRQ via the tunable, then we either use the
3419 	 * IRQ value in the intline register or we ask the bus to route an
3420 	 * interrupt for us.  If force_route is true, then we only use the
3421 	 * value in the intline register if the bus was unable to assign an
3422 	 * IRQ.
3423 	 */
3424 	if (!PCI_INTERRUPT_VALID(irq)) {
3425 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3426 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3427 		if (!PCI_INTERRUPT_VALID(irq))
3428 			irq = cfg->intline;
3429 	}
3430 
3431 	/* If after all that we don't have an IRQ, just bail. */
3432 	if (!PCI_INTERRUPT_VALID(irq))
3433 		return;
3434 
3435 	/* Update the config register if it changed. */
3436 	if (irq != cfg->intline) {
3437 		cfg->intline = irq;
3438 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3439 	}
3440 
3441 	/* Add this IRQ as rid 0 interrupt resource. */
3442 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3443 }
3444 
3445 /* Perform early OHCI takeover from SMM. */
3446 static void
3447 ohci_early_takeover(device_t self)
3448 {
3449 	struct resource *res;
3450 	uint32_t ctl;
3451 	int rid;
3452 	int i;
3453 
3454 	rid = PCIR_BAR(0);
3455 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3456 	if (res == NULL)
3457 		return;
3458 
3459 	ctl = bus_read_4(res, OHCI_CONTROL);
3460 	if (ctl & OHCI_IR) {
3461 		if (bootverbose)
3462 			printf("ohci early: "
3463 			    "SMM active, request owner change\n");
3464 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3465 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3466 			DELAY(1000);
3467 			ctl = bus_read_4(res, OHCI_CONTROL);
3468 		}
3469 		if (ctl & OHCI_IR) {
3470 			if (bootverbose)
3471 				printf("ohci early: "
3472 				    "SMM does not respond, resetting\n");
3473 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3474 		}
3475 		/* Disable interrupts */
3476 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3477 	}
3478 
3479 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3480 }
3481 
3482 /* Perform early UHCI takeover from SMM. */
3483 static void
3484 uhci_early_takeover(device_t self)
3485 {
3486 	struct resource *res;
3487 	int rid;
3488 
3489 	/*
3490 	 * Set the PIRQD enable bit and switch off all the others. We don't
3491 	 * want legacy support to interfere with us XXX Does this also mean
3492 	 * that the BIOS won't touch the keyboard anymore if it is connected
3493 	 * to the ports of the root hub?
3494 	 */
3495 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3496 
3497 	/* Disable interrupts */
3498 	rid = PCI_UHCI_BASE_REG;
3499 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3500 	if (res != NULL) {
3501 		bus_write_2(res, UHCI_INTR, 0);
3502 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3503 	}
3504 }
3505 
3506 /* Perform early EHCI takeover from SMM. */
3507 static void
3508 ehci_early_takeover(device_t self)
3509 {
3510 	struct resource *res;
3511 	uint32_t cparams;
3512 	uint32_t eec;
3513 	uint8_t eecp;
3514 	uint8_t bios_sem;
3515 	uint8_t offs;
3516 	int rid;
3517 	int i;
3518 
3519 	rid = PCIR_BAR(0);
3520 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3521 	if (res == NULL)
3522 		return;
3523 
3524 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3525 
3526 	/* Synchronise with the BIOS if it owns the controller. */
3527 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3528 	    eecp = EHCI_EECP_NEXT(eec)) {
3529 		eec = pci_read_config(self, eecp, 4);
3530 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3531 			continue;
3532 		}
3533 		bios_sem = pci_read_config(self, eecp +
3534 		    EHCI_LEGSUP_BIOS_SEM, 1);
3535 		if (bios_sem == 0) {
3536 			continue;
3537 		}
3538 		if (bootverbose)
3539 			printf("ehci early: "
3540 			    "SMM active, request owner change\n");
3541 
3542 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3543 
3544 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3545 			DELAY(1000);
3546 			bios_sem = pci_read_config(self, eecp +
3547 			    EHCI_LEGSUP_BIOS_SEM, 1);
3548 		}
3549 
3550 		if (bios_sem != 0) {
3551 			if (bootverbose)
3552 				printf("ehci early: "
3553 				    "SMM does not respond\n");
3554 		}
3555 		/* Disable interrupts */
3556 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3557 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3558 	}
3559 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3560 }
3561 
3562 /* Perform early XHCI takeover from SMM. */
3563 static void
3564 xhci_early_takeover(device_t self)
3565 {
3566 	struct resource *res;
3567 	uint32_t cparams;
3568 	uint32_t eec;
3569 	uint8_t eecp;
3570 	uint8_t bios_sem;
3571 	uint8_t offs;
3572 	int rid;
3573 	int i;
3574 
3575 	rid = PCIR_BAR(0);
3576 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3577 	if (res == NULL)
3578 		return;
3579 
3580 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3581 
3582 	eec = -1;
3583 
3584 	/* Synchronise with the BIOS if it owns the controller. */
3585 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3586 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3587 		eec = bus_read_4(res, eecp);
3588 
3589 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3590 			continue;
3591 
3592 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3593 		if (bios_sem == 0)
3594 			continue;
3595 
3596 		if (bootverbose)
3597 			printf("xhci early: "
3598 			    "SMM active, request owner change\n");
3599 
3600 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3601 
3602 		/* wait a maximum of 5 second */
3603 
3604 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3605 			DELAY(1000);
3606 			bios_sem = bus_read_1(res, eecp +
3607 			    XHCI_XECP_BIOS_SEM);
3608 		}
3609 
3610 		if (bios_sem != 0) {
3611 			if (bootverbose)
3612 				printf("xhci early: "
3613 				    "SMM does not respond\n");
3614 		}
3615 
3616 		/* Disable interrupts */
3617 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3618 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3619 		bus_read_4(res, offs + XHCI_USBSTS);
3620 	}
3621 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3622 }
3623 
3624 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3625 static void
3626 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3627     struct resource_list *rl)
3628 {
3629 	struct resource *res;
3630 	char *cp;
3631 	rman_res_t start, end, count;
3632 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3633 
3634 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3635 	case PCIM_HDRTYPE_BRIDGE:
3636 		sec_reg = PCIR_SECBUS_1;
3637 		sub_reg = PCIR_SUBBUS_1;
3638 		break;
3639 	case PCIM_HDRTYPE_CARDBUS:
3640 		sec_reg = PCIR_SECBUS_2;
3641 		sub_reg = PCIR_SUBBUS_2;
3642 		break;
3643 	default:
3644 		return;
3645 	}
3646 
3647 	/*
3648 	 * If the existing bus range is valid, attempt to reserve it
3649 	 * from our parent.  If this fails for any reason, clear the
3650 	 * secbus and subbus registers.
3651 	 *
3652 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3653 	 * This would at least preserve the existing sec_bus if it is
3654 	 * valid.
3655 	 */
3656 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3657 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3658 
3659 	/* Quirk handling. */
3660 	switch (pci_get_devid(dev)) {
3661 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3662 		sup_bus = pci_read_config(dev, 0x41, 1);
3663 		if (sup_bus != 0xff) {
3664 			sec_bus = sup_bus + 1;
3665 			sub_bus = sup_bus + 1;
3666 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3667 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3668 		}
3669 		break;
3670 
3671 	case 0x00dd10de:
3672 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3673 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3674 			break;
3675 		if (strncmp(cp, "Compal", 6) != 0) {
3676 			freeenv(cp);
3677 			break;
3678 		}
3679 		freeenv(cp);
3680 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3681 			break;
3682 		if (strncmp(cp, "08A0", 4) != 0) {
3683 			freeenv(cp);
3684 			break;
3685 		}
3686 		freeenv(cp);
3687 		if (sub_bus < 0xa) {
3688 			sub_bus = 0xa;
3689 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3690 		}
3691 		break;
3692 	}
3693 
3694 	if (bootverbose)
3695 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3696 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3697 		start = sec_bus;
3698 		end = sub_bus;
3699 		count = end - start + 1;
3700 
3701 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3702 
3703 		/*
3704 		 * If requested, clear secondary bus registers in
3705 		 * bridge devices to force a complete renumbering
3706 		 * rather than reserving the existing range.  However,
3707 		 * preserve the existing size.
3708 		 */
3709 		if (pci_clear_buses)
3710 			goto clear;
3711 
3712 		rid = 0;
3713 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3714 		    start, end, count, 0);
3715 		if (res != NULL)
3716 			return;
3717 
3718 		if (bootverbose)
3719 			device_printf(bus,
3720 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3721 			    pci_get_domain(dev), pci_get_bus(dev),
3722 			    pci_get_slot(dev), pci_get_function(dev));
3723 	}
3724 
3725 clear:
3726 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3727 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3728 }
3729 
3730 static struct resource *
3731 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3732     rman_res_t end, rman_res_t count, u_int flags)
3733 {
3734 	struct pci_devinfo *dinfo;
3735 	pcicfgregs *cfg;
3736 	struct resource_list *rl;
3737 	struct resource *res;
3738 	int sec_reg, sub_reg;
3739 
3740 	dinfo = device_get_ivars(child);
3741 	cfg = &dinfo->cfg;
3742 	rl = &dinfo->resources;
3743 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3744 	case PCIM_HDRTYPE_BRIDGE:
3745 		sec_reg = PCIR_SECBUS_1;
3746 		sub_reg = PCIR_SUBBUS_1;
3747 		break;
3748 	case PCIM_HDRTYPE_CARDBUS:
3749 		sec_reg = PCIR_SECBUS_2;
3750 		sub_reg = PCIR_SUBBUS_2;
3751 		break;
3752 	default:
3753 		return (NULL);
3754 	}
3755 
3756 	if (*rid != 0)
3757 		return (NULL);
3758 
3759 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3760 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3761 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3762 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3763 		    start, end, count, flags & ~RF_ACTIVE);
3764 		if (res == NULL) {
3765 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3766 			device_printf(child, "allocating %ju bus%s failed\n",
3767 			    count, count == 1 ? "" : "es");
3768 			return (NULL);
3769 		}
3770 		if (bootverbose)
3771 			device_printf(child,
3772 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3773 			    count == 1 ? "" : "es", rman_get_start(res));
3774 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3775 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3776 	}
3777 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3778 	    end, count, flags));
3779 }
3780 #endif
3781 
3782 static int
3783 pci_ea_bei_to_rid(device_t dev, int bei)
3784 {
3785 #ifdef PCI_IOV
3786 	struct pci_devinfo *dinfo;
3787 	int iov_pos;
3788 	struct pcicfg_iov *iov;
3789 
3790 	dinfo = device_get_ivars(dev);
3791 	iov = dinfo->cfg.iov;
3792 	if (iov != NULL)
3793 		iov_pos = iov->iov_pos;
3794 	else
3795 		iov_pos = 0;
3796 #endif
3797 
3798 	/* Check if matches BAR */
3799 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3800 	    (bei <= PCIM_EA_BEI_BAR_5))
3801 		return (PCIR_BAR(bei));
3802 
3803 	/* Check ROM */
3804 	if (bei == PCIM_EA_BEI_ROM)
3805 		return (PCIR_BIOS);
3806 
3807 #ifdef PCI_IOV
3808 	/* Check if matches VF_BAR */
3809 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3810 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3811 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3812 		    iov_pos);
3813 #endif
3814 
3815 	return (-1);
3816 }
3817 
3818 int
3819 pci_ea_is_enabled(device_t dev, int rid)
3820 {
3821 	struct pci_ea_entry *ea;
3822 	struct pci_devinfo *dinfo;
3823 
3824 	dinfo = device_get_ivars(dev);
3825 
3826 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3827 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3828 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3829 	}
3830 
3831 	return (0);
3832 }
3833 
3834 void
3835 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3836 {
3837 	struct pci_ea_entry *ea;
3838 	struct pci_devinfo *dinfo;
3839 	pci_addr_t start, end, count;
3840 	struct resource_list *rl;
3841 	int type, flags, rid;
3842 	struct resource *res;
3843 	uint32_t tmp;
3844 #ifdef PCI_IOV
3845 	struct pcicfg_iov *iov;
3846 #endif
3847 
3848 	dinfo = device_get_ivars(dev);
3849 	rl = &dinfo->resources;
3850 	flags = 0;
3851 
3852 #ifdef PCI_IOV
3853 	iov = dinfo->cfg.iov;
3854 #endif
3855 
3856 	if (dinfo->cfg.ea.ea_location == 0)
3857 		return;
3858 
3859 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3860 
3861 		/*
3862 		 * TODO: Ignore EA-BAR if is not enabled.
3863 		 *   Currently the EA implementation supports
3864 		 *   only situation, where EA structure contains
3865 		 *   predefined entries. In case they are not enabled
3866 		 *   leave them unallocated and proceed with
3867 		 *   a legacy-BAR mechanism.
3868 		 */
3869 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3870 			continue;
3871 
3872 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3873 		case PCIM_EA_P_MEM_PREFETCH:
3874 		case PCIM_EA_P_VF_MEM_PREFETCH:
3875 			flags = RF_PREFETCHABLE;
3876 			/* FALLTHROUGH */
3877 		case PCIM_EA_P_VF_MEM:
3878 		case PCIM_EA_P_MEM:
3879 			type = SYS_RES_MEMORY;
3880 			break;
3881 		case PCIM_EA_P_IO:
3882 			type = SYS_RES_IOPORT;
3883 			break;
3884 		default:
3885 			continue;
3886 		}
3887 
3888 		if (alloc_iov != 0) {
3889 #ifdef PCI_IOV
3890 			/* Allocating IOV, confirm BEI matches */
3891 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3892 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3893 				continue;
3894 #else
3895 			continue;
3896 #endif
3897 		} else {
3898 			/* Allocating BAR, confirm BEI matches */
3899 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3900 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3901 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3902 				continue;
3903 		}
3904 
3905 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3906 		if (rid < 0)
3907 			continue;
3908 
3909 		/* Skip resources already allocated by EA */
3910 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3911 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3912 			continue;
3913 
3914 		start = ea->eae_base;
3915 		count = ea->eae_max_offset + 1;
3916 #ifdef PCI_IOV
3917 		if (iov != NULL)
3918 			count = count * iov->iov_num_vfs;
3919 #endif
3920 		end = start + count - 1;
3921 		if (count == 0)
3922 			continue;
3923 
3924 		resource_list_add(rl, type, rid, start, end, count);
3925 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3926 		    flags);
3927 		if (res == NULL) {
3928 			resource_list_delete(rl, type, rid);
3929 
3930 			/*
3931 			 * Failed to allocate using EA, disable entry.
3932 			 * Another attempt to allocation will be performed
3933 			 * further, but this time using legacy BAR registers
3934 			 */
3935 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3936 			tmp &= ~PCIM_EA_ENABLE;
3937 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3938 
3939 			/*
3940 			 * Disabling entry might fail in case it is hardwired.
3941 			 * Read flags again to match current status.
3942 			 */
3943 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3944 
3945 			continue;
3946 		}
3947 
3948 		/* As per specification, fill BAR with zeros */
3949 		pci_write_config(dev, rid, 0, 4);
3950 	}
3951 }
3952 
3953 void
3954 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3955 {
3956 	struct pci_devinfo *dinfo;
3957 	pcicfgregs *cfg;
3958 	struct resource_list *rl;
3959 	const struct pci_quirk *q;
3960 	uint32_t devid;
3961 	int i;
3962 
3963 	dinfo = device_get_ivars(dev);
3964 	cfg = &dinfo->cfg;
3965 	rl = &dinfo->resources;
3966 	devid = (cfg->device << 16) | cfg->vendor;
3967 
3968 	/* Allocate resources using Enhanced Allocation */
3969 	pci_add_resources_ea(bus, dev, 0);
3970 
3971 	/* ATA devices needs special map treatment */
3972 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3973 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3974 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3975 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3976 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3977 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3978 	else
3979 		for (i = 0; i < cfg->nummaps;) {
3980 			/* Skip resources already managed by EA */
3981 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3982 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3983 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3984 				i++;
3985 				continue;
3986 			}
3987 
3988 			/*
3989 			 * Skip quirked resources.
3990 			 */
3991 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3992 				if (q->devid == devid &&
3993 				    q->type == PCI_QUIRK_UNMAP_REG &&
3994 				    q->arg1 == PCIR_BAR(i))
3995 					break;
3996 			if (q->devid != 0) {
3997 				i++;
3998 				continue;
3999 			}
4000 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4001 			    prefetchmask & (1 << i));
4002 		}
4003 
4004 	/*
4005 	 * Add additional, quirked resources.
4006 	 */
4007 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4008 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4009 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4010 
4011 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4012 #ifdef __PCI_REROUTE_INTERRUPT
4013 		/*
4014 		 * Try to re-route interrupts. Sometimes the BIOS or
4015 		 * firmware may leave bogus values in these registers.
4016 		 * If the re-route fails, then just stick with what we
4017 		 * have.
4018 		 */
4019 		pci_assign_interrupt(bus, dev, 1);
4020 #else
4021 		pci_assign_interrupt(bus, dev, 0);
4022 #endif
4023 	}
4024 
4025 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4026 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4027 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4028 			xhci_early_takeover(dev);
4029 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4030 			ehci_early_takeover(dev);
4031 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4032 			ohci_early_takeover(dev);
4033 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4034 			uhci_early_takeover(dev);
4035 	}
4036 
4037 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4038 	/*
4039 	 * Reserve resources for secondary bus ranges behind bridge
4040 	 * devices.
4041 	 */
4042 	pci_reserve_secbus(bus, dev, cfg, rl);
4043 #endif
4044 }
4045 
4046 static struct pci_devinfo *
4047 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4048     int slot, int func)
4049 {
4050 	struct pci_devinfo *dinfo;
4051 
4052 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4053 	if (dinfo != NULL)
4054 		pci_add_child(dev, dinfo);
4055 
4056 	return (dinfo);
4057 }
4058 
4059 void
4060 pci_add_children(device_t dev, int domain, int busno)
4061 {
4062 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4063 	device_t pcib = device_get_parent(dev);
4064 	struct pci_devinfo *dinfo;
4065 	int maxslots;
4066 	int s, f, pcifunchigh;
4067 	uint8_t hdrtype;
4068 	int first_func;
4069 
4070 	/*
4071 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4072 	 * enable ARI.  We must enable ARI before detecting the rest of the
4073 	 * functions on this bus as ARI changes the set of slots and functions
4074 	 * that are legal on this bus.
4075 	 */
4076 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4077 	if (dinfo != NULL && pci_enable_ari)
4078 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4079 
4080 	/*
4081 	 * Start looking for new devices on slot 0 at function 1 because we
4082 	 * just identified the device at slot 0, function 0.
4083 	 */
4084 	first_func = 1;
4085 
4086 	maxslots = PCIB_MAXSLOTS(pcib);
4087 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4088 		pcifunchigh = 0;
4089 		f = 0;
4090 		DELAY(1);
4091 		hdrtype = REG(PCIR_HDRTYPE, 1);
4092 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4093 			continue;
4094 		if (hdrtype & PCIM_MFDEV)
4095 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4096 		for (f = first_func; f <= pcifunchigh; f++)
4097 			pci_identify_function(pcib, dev, domain, busno, s, f);
4098 	}
4099 #undef REG
4100 }
4101 
4102 int
4103 pci_rescan_method(device_t dev)
4104 {
4105 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4106 	device_t pcib = device_get_parent(dev);
4107 	device_t child, *devlist, *unchanged;
4108 	int devcount, error, i, j, maxslots, oldcount;
4109 	int busno, domain, s, f, pcifunchigh;
4110 	uint8_t hdrtype;
4111 
4112 	/* No need to check for ARI on a rescan. */
4113 	error = device_get_children(dev, &devlist, &devcount);
4114 	if (error)
4115 		return (error);
4116 	if (devcount != 0) {
4117 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4118 		    M_NOWAIT | M_ZERO);
4119 		if (unchanged == NULL) {
4120 			free(devlist, M_TEMP);
4121 			return (ENOMEM);
4122 		}
4123 	} else
4124 		unchanged = NULL;
4125 
4126 	domain = pcib_get_domain(dev);
4127 	busno = pcib_get_bus(dev);
4128 	maxslots = PCIB_MAXSLOTS(pcib);
4129 	for (s = 0; s <= maxslots; s++) {
4130 		/* If function 0 is not present, skip to the next slot. */
4131 		f = 0;
4132 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4133 			continue;
4134 		pcifunchigh = 0;
4135 		hdrtype = REG(PCIR_HDRTYPE, 1);
4136 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4137 			continue;
4138 		if (hdrtype & PCIM_MFDEV)
4139 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4140 		for (f = 0; f <= pcifunchigh; f++) {
4141 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4142 				continue;
4143 
4144 			/*
4145 			 * Found a valid function.  Check if a
4146 			 * device_t for this device already exists.
4147 			 */
4148 			for (i = 0; i < devcount; i++) {
4149 				child = devlist[i];
4150 				if (child == NULL)
4151 					continue;
4152 				if (pci_get_slot(child) == s &&
4153 				    pci_get_function(child) == f) {
4154 					unchanged[i] = child;
4155 					goto next_func;
4156 				}
4157 			}
4158 
4159 			pci_identify_function(pcib, dev, domain, busno, s, f);
4160 		next_func:;
4161 		}
4162 	}
4163 
4164 	/* Remove devices that are no longer present. */
4165 	for (i = 0; i < devcount; i++) {
4166 		if (unchanged[i] != NULL)
4167 			continue;
4168 		device_delete_child(dev, devlist[i]);
4169 	}
4170 
4171 	free(devlist, M_TEMP);
4172 	oldcount = devcount;
4173 
4174 	/* Try to attach the devices just added. */
4175 	error = device_get_children(dev, &devlist, &devcount);
4176 	if (error) {
4177 		free(unchanged, M_TEMP);
4178 		return (error);
4179 	}
4180 
4181 	for (i = 0; i < devcount; i++) {
4182 		for (j = 0; j < oldcount; j++) {
4183 			if (devlist[i] == unchanged[j])
4184 				goto next_device;
4185 		}
4186 
4187 		device_probe_and_attach(devlist[i]);
4188 	next_device:;
4189 	}
4190 
4191 	free(unchanged, M_TEMP);
4192 	free(devlist, M_TEMP);
4193 	return (0);
4194 #undef REG
4195 }
4196 
4197 #ifdef PCI_IOV
4198 device_t
4199 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4200     uint16_t did)
4201 {
4202 	struct pci_devinfo *vf_dinfo;
4203 	device_t pcib;
4204 	int busno, slot, func;
4205 
4206 	pcib = device_get_parent(bus);
4207 
4208 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4209 
4210 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4211 	    slot, func, vid, did);
4212 
4213 	vf_dinfo->cfg.flags |= PCICFG_VF;
4214 	pci_add_child(bus, vf_dinfo);
4215 
4216 	return (vf_dinfo->cfg.dev);
4217 }
4218 
4219 device_t
4220 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4221     uint16_t vid, uint16_t did)
4222 {
4223 
4224 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4225 }
4226 #endif
4227 
4228 static void
4229 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4230 {
4231 	int aer;
4232 	uint32_t r;
4233 	uint16_t r2;
4234 
4235 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4236 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4237 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4238 		    PCIER_ROOT_CTL, 2);
4239 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4240 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4241 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4242 		    PCIER_ROOT_CTL, r2, 2);
4243 	}
4244 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4245 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4246 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4247 		if (r != 0 && bootverbose) {
4248 			pci_printf(&dinfo->cfg,
4249 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4250 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4251 			    4));
4252 		}
4253 
4254 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4255 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4256 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4257 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4258 		    PCIM_AER_UC_POISONED_TLP |
4259 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4260 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4261 		    PCIM_AER_UC_COMPLETER_ABORT |
4262 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4263 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4264 		    PCIM_AER_UC_MALFORMED_TLP |
4265 		    PCIM_AER_UC_ECRC_ERROR |
4266 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4267 		    PCIM_AER_UC_ACS_VIOLATION |
4268 		    PCIM_AER_UC_INTERNAL_ERROR |
4269 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4270 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4271 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4272 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4273 
4274 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4275 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4276 		if (r != 0 && bootverbose) {
4277 			pci_printf(&dinfo->cfg,
4278 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4279 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4280 			    4));
4281 		}
4282 
4283 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4284 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4285 		    PCIM_AER_COR_BAD_TLP |
4286 		    PCIM_AER_COR_BAD_DLLP |
4287 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4288 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4289 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4290 		    PCIM_AER_COR_INTERNAL_ERROR |
4291 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4292 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4293 
4294 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4295 		    PCIER_DEVICE_CTL, 2);
4296 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4297 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4298 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4299 		    PCIER_DEVICE_CTL, r, 2);
4300 	}
4301 }
4302 
4303 void
4304 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4305 {
4306 	device_t dev;
4307 
4308 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4309 	device_set_ivars(dev, dinfo);
4310 	resource_list_init(&dinfo->resources);
4311 	pci_cfg_save(dev, dinfo, 0);
4312 	pci_cfg_restore(dev, dinfo);
4313 	pci_print_verbose(dinfo);
4314 	pci_add_resources(bus, dev, 0, 0);
4315 	pci_child_added(dinfo->cfg.dev);
4316 
4317 	if (pci_clear_aer_on_attach)
4318 		pci_add_child_clear_aer(dev, dinfo);
4319 
4320 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4321 }
4322 
4323 void
4324 pci_child_added_method(device_t dev, device_t child)
4325 {
4326 
4327 }
4328 
4329 static int
4330 pci_probe(device_t dev)
4331 {
4332 
4333 	device_set_desc(dev, "PCI bus");
4334 
4335 	/* Allow other subclasses to override this driver. */
4336 	return (BUS_PROBE_GENERIC);
4337 }
4338 
4339 int
4340 pci_attach_common(device_t dev)
4341 {
4342 	struct pci_softc *sc;
4343 	int busno, domain;
4344 #ifdef PCI_RES_BUS
4345 	int rid;
4346 #endif
4347 
4348 	sc = device_get_softc(dev);
4349 	domain = pcib_get_domain(dev);
4350 	busno = pcib_get_bus(dev);
4351 #ifdef PCI_RES_BUS
4352 	rid = 0;
4353 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4354 	    1, 0);
4355 	if (sc->sc_bus == NULL) {
4356 		device_printf(dev, "failed to allocate bus number\n");
4357 		return (ENXIO);
4358 	}
4359 #endif
4360 	if (bootverbose)
4361 		device_printf(dev, "domain=%d, physical bus=%d\n",
4362 		    domain, busno);
4363 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4364 	return (0);
4365 }
4366 
4367 int
4368 pci_attach(device_t dev)
4369 {
4370 	int busno, domain, error;
4371 
4372 	error = pci_attach_common(dev);
4373 	if (error)
4374 		return (error);
4375 
4376 	/*
4377 	 * Since there can be multiple independently numbered PCI
4378 	 * buses on systems with multiple PCI domains, we can't use
4379 	 * the unit number to decide which bus we are probing. We ask
4380 	 * the parent pcib what our domain and bus numbers are.
4381 	 */
4382 	domain = pcib_get_domain(dev);
4383 	busno = pcib_get_bus(dev);
4384 	pci_add_children(dev, domain, busno);
4385 	return (bus_generic_attach(dev));
4386 }
4387 
4388 int
4389 pci_detach(device_t dev)
4390 {
4391 #ifdef PCI_RES_BUS
4392 	struct pci_softc *sc;
4393 #endif
4394 	int error;
4395 
4396 	error = bus_generic_detach(dev);
4397 	if (error)
4398 		return (error);
4399 #ifdef PCI_RES_BUS
4400 	sc = device_get_softc(dev);
4401 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4402 	if (error)
4403 		return (error);
4404 #endif
4405 	return (device_delete_children(dev));
4406 }
4407 
4408 static void
4409 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4410 {
4411 	int line, unit;
4412 	const char *at;
4413 	char me1[24], me2[32];
4414 	uint8_t b, s, f;
4415 	uint32_t d;
4416 
4417 	d = pci_get_domain(child);
4418 	b = pci_get_bus(child);
4419 	s = pci_get_slot(child);
4420 	f = pci_get_function(child);
4421 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4422 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4423 	line = 0;
4424 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4425 		resource_string_value(name, unit, "at", &at);
4426 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4427 			continue; /* No match, try next candidate */
4428 		*unitp = unit;
4429 		return;
4430 	}
4431 }
4432 
4433 static void
4434 pci_set_power_child(device_t dev, device_t child, int state)
4435 {
4436 	device_t pcib;
4437 	int dstate;
4438 
4439 	/*
4440 	 * Set the device to the given state.  If the firmware suggests
4441 	 * a different power state, use it instead.  If power management
4442 	 * is not present, the firmware is responsible for managing
4443 	 * device power.  Skip children who aren't attached since they
4444 	 * are handled separately.
4445 	 */
4446 	pcib = device_get_parent(dev);
4447 	dstate = state;
4448 	if (device_is_attached(child) &&
4449 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4450 		pci_set_powerstate(child, dstate);
4451 }
4452 
4453 int
4454 pci_suspend_child(device_t dev, device_t child)
4455 {
4456 	struct pci_devinfo *dinfo;
4457 	struct resource_list_entry *rle;
4458 	int error;
4459 
4460 	dinfo = device_get_ivars(child);
4461 
4462 	/*
4463 	 * Save the PCI configuration space for the child and set the
4464 	 * device in the appropriate power state for this sleep state.
4465 	 */
4466 	pci_cfg_save(child, dinfo, 0);
4467 
4468 	/* Suspend devices before potentially powering them down. */
4469 	error = bus_generic_suspend_child(dev, child);
4470 
4471 	if (error)
4472 		return (error);
4473 
4474 	if (pci_do_power_suspend) {
4475 		/*
4476 		 * Make sure this device's interrupt handler is not invoked
4477 		 * in the case the device uses a shared interrupt that can
4478 		 * be raised by some other device.
4479 		 * This is applicable only to regular (legacy) PCI interrupts
4480 		 * as MSI/MSI-X interrupts are never shared.
4481 		 */
4482 		rle = resource_list_find(&dinfo->resources,
4483 		    SYS_RES_IRQ, 0);
4484 		if (rle != NULL && rle->res != NULL)
4485 			(void)bus_suspend_intr(child, rle->res);
4486 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4487 	}
4488 
4489 	return (0);
4490 }
4491 
4492 int
4493 pci_resume_child(device_t dev, device_t child)
4494 {
4495 	struct pci_devinfo *dinfo;
4496 	struct resource_list_entry *rle;
4497 
4498 	if (pci_do_power_resume)
4499 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4500 
4501 	dinfo = device_get_ivars(child);
4502 	pci_cfg_restore(child, dinfo);
4503 	if (!device_is_attached(child))
4504 		pci_cfg_save(child, dinfo, 1);
4505 
4506 	bus_generic_resume_child(dev, child);
4507 
4508 	/*
4509 	 * Allow interrupts only after fully resuming the driver and hardware.
4510 	 */
4511 	if (pci_do_power_suspend) {
4512 		/* See pci_suspend_child for details. */
4513 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4514 		if (rle != NULL && rle->res != NULL)
4515 			(void)bus_resume_intr(child, rle->res);
4516 	}
4517 
4518 	return (0);
4519 }
4520 
4521 int
4522 pci_resume(device_t dev)
4523 {
4524 	device_t child, *devlist;
4525 	int error, i, numdevs;
4526 
4527 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4528 		return (error);
4529 
4530 	/*
4531 	 * Resume critical devices first, then everything else later.
4532 	 */
4533 	for (i = 0; i < numdevs; i++) {
4534 		child = devlist[i];
4535 		switch (pci_get_class(child)) {
4536 		case PCIC_DISPLAY:
4537 		case PCIC_MEMORY:
4538 		case PCIC_BRIDGE:
4539 		case PCIC_BASEPERIPH:
4540 			BUS_RESUME_CHILD(dev, child);
4541 			break;
4542 		}
4543 	}
4544 	for (i = 0; i < numdevs; i++) {
4545 		child = devlist[i];
4546 		switch (pci_get_class(child)) {
4547 		case PCIC_DISPLAY:
4548 		case PCIC_MEMORY:
4549 		case PCIC_BRIDGE:
4550 		case PCIC_BASEPERIPH:
4551 			break;
4552 		default:
4553 			BUS_RESUME_CHILD(dev, child);
4554 		}
4555 	}
4556 	free(devlist, M_TEMP);
4557 	return (0);
4558 }
4559 
4560 static void
4561 pci_load_vendor_data(void)
4562 {
4563 	caddr_t data;
4564 	void *ptr;
4565 	size_t sz;
4566 
4567 	data = preload_search_by_type("pci_vendor_data");
4568 	if (data != NULL) {
4569 		ptr = preload_fetch_addr(data);
4570 		sz = preload_fetch_size(data);
4571 		if (ptr != NULL && sz != 0) {
4572 			pci_vendordata = ptr;
4573 			pci_vendordata_size = sz;
4574 			/* terminate the database */
4575 			pci_vendordata[pci_vendordata_size] = '\n';
4576 		}
4577 	}
4578 }
4579 
4580 void
4581 pci_driver_added(device_t dev, driver_t *driver)
4582 {
4583 	int numdevs;
4584 	device_t *devlist;
4585 	device_t child;
4586 	struct pci_devinfo *dinfo;
4587 	int i;
4588 
4589 	if (bootverbose)
4590 		device_printf(dev, "driver added\n");
4591 	DEVICE_IDENTIFY(driver, dev);
4592 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4593 		return;
4594 	for (i = 0; i < numdevs; i++) {
4595 		child = devlist[i];
4596 		if (device_get_state(child) != DS_NOTPRESENT)
4597 			continue;
4598 		dinfo = device_get_ivars(child);
4599 		pci_print_verbose(dinfo);
4600 		if (bootverbose)
4601 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4602 		pci_cfg_restore(child, dinfo);
4603 		if (device_probe_and_attach(child) != 0)
4604 			pci_child_detached(dev, child);
4605 	}
4606 	free(devlist, M_TEMP);
4607 }
4608 
4609 int
4610 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4611     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4612 {
4613 	struct pci_devinfo *dinfo;
4614 	struct msix_table_entry *mte;
4615 	struct msix_vector *mv;
4616 	uint64_t addr;
4617 	uint32_t data;
4618 	void *cookie;
4619 	int error, rid;
4620 
4621 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4622 	    arg, &cookie);
4623 	if (error)
4624 		return (error);
4625 
4626 	/* If this is not a direct child, just bail out. */
4627 	if (device_get_parent(child) != dev) {
4628 		*cookiep = cookie;
4629 		return(0);
4630 	}
4631 
4632 	rid = rman_get_rid(irq);
4633 	if (rid == 0) {
4634 		/* Make sure that INTx is enabled */
4635 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4636 	} else {
4637 		/*
4638 		 * Check to see if the interrupt is MSI or MSI-X.
4639 		 * Ask our parent to map the MSI and give
4640 		 * us the address and data register values.
4641 		 * If we fail for some reason, teardown the
4642 		 * interrupt handler.
4643 		 */
4644 		dinfo = device_get_ivars(child);
4645 		if (dinfo->cfg.msi.msi_alloc > 0) {
4646 			if (dinfo->cfg.msi.msi_addr == 0) {
4647 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4648 			    ("MSI has handlers, but vectors not mapped"));
4649 				error = PCIB_MAP_MSI(device_get_parent(dev),
4650 				    child, rman_get_start(irq), &addr, &data);
4651 				if (error)
4652 					goto bad;
4653 				dinfo->cfg.msi.msi_addr = addr;
4654 				dinfo->cfg.msi.msi_data = data;
4655 			}
4656 			if (dinfo->cfg.msi.msi_handlers == 0)
4657 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4658 				    dinfo->cfg.msi.msi_data);
4659 			dinfo->cfg.msi.msi_handlers++;
4660 		} else {
4661 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4662 			    ("No MSI or MSI-X interrupts allocated"));
4663 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4664 			    ("MSI-X index too high"));
4665 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4666 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4667 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4668 			KASSERT(mv->mv_irq == rman_get_start(irq),
4669 			    ("IRQ mismatch"));
4670 			if (mv->mv_address == 0) {
4671 				KASSERT(mte->mte_handlers == 0,
4672 		    ("MSI-X table entry has handlers, but vector not mapped"));
4673 				error = PCIB_MAP_MSI(device_get_parent(dev),
4674 				    child, rman_get_start(irq), &addr, &data);
4675 				if (error)
4676 					goto bad;
4677 				mv->mv_address = addr;
4678 				mv->mv_data = data;
4679 			}
4680 
4681 			/*
4682 			 * The MSIX table entry must be made valid by
4683 			 * incrementing the mte_handlers before
4684 			 * calling pci_enable_msix() and
4685 			 * pci_resume_msix(). Else the MSIX rewrite
4686 			 * table quirk will not work as expected.
4687 			 */
4688 			mte->mte_handlers++;
4689 			if (mte->mte_handlers == 1) {
4690 				pci_enable_msix(child, rid - 1, mv->mv_address,
4691 				    mv->mv_data);
4692 				pci_unmask_msix(child, rid - 1);
4693 			}
4694 		}
4695 
4696 		/*
4697 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4698 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4699 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4700 		 */
4701 		if (!pci_has_quirk(pci_get_devid(child),
4702 		    PCI_QUIRK_MSI_INTX_BUG))
4703 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4704 		else
4705 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4706 	bad:
4707 		if (error) {
4708 			(void)bus_generic_teardown_intr(dev, child, irq,
4709 			    cookie);
4710 			return (error);
4711 		}
4712 	}
4713 	*cookiep = cookie;
4714 	return (0);
4715 }
4716 
4717 int
4718 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4719     void *cookie)
4720 {
4721 	struct msix_table_entry *mte;
4722 	struct resource_list_entry *rle;
4723 	struct pci_devinfo *dinfo;
4724 	int error, rid;
4725 
4726 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4727 		return (EINVAL);
4728 
4729 	/* If this isn't a direct child, just bail out */
4730 	if (device_get_parent(child) != dev)
4731 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4732 
4733 	rid = rman_get_rid(irq);
4734 	if (rid == 0) {
4735 		/* Mask INTx */
4736 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4737 	} else {
4738 		/*
4739 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4740 		 * decrement the appropriate handlers count and mask the
4741 		 * MSI-X message, or disable MSI messages if the count
4742 		 * drops to 0.
4743 		 */
4744 		dinfo = device_get_ivars(child);
4745 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4746 		if (rle->res != irq)
4747 			return (EINVAL);
4748 		if (dinfo->cfg.msi.msi_alloc > 0) {
4749 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4750 			    ("MSI-X index too high"));
4751 			if (dinfo->cfg.msi.msi_handlers == 0)
4752 				return (EINVAL);
4753 			dinfo->cfg.msi.msi_handlers--;
4754 			if (dinfo->cfg.msi.msi_handlers == 0)
4755 				pci_disable_msi(child);
4756 		} else {
4757 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4758 			    ("No MSI or MSI-X interrupts allocated"));
4759 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4760 			    ("MSI-X index too high"));
4761 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4762 			if (mte->mte_handlers == 0)
4763 				return (EINVAL);
4764 			mte->mte_handlers--;
4765 			if (mte->mte_handlers == 0)
4766 				pci_mask_msix(child, rid - 1);
4767 		}
4768 	}
4769 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4770 	if (rid > 0)
4771 		KASSERT(error == 0,
4772 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4773 	return (error);
4774 }
4775 
4776 int
4777 pci_print_child(device_t dev, device_t child)
4778 {
4779 	struct pci_devinfo *dinfo;
4780 	struct resource_list *rl;
4781 	int retval = 0;
4782 
4783 	dinfo = device_get_ivars(child);
4784 	rl = &dinfo->resources;
4785 
4786 	retval += bus_print_child_header(dev, child);
4787 
4788 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4789 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4790 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4791 	if (device_get_flags(dev))
4792 		retval += printf(" flags %#x", device_get_flags(dev));
4793 
4794 	retval += printf(" at device %d.%d", pci_get_slot(child),
4795 	    pci_get_function(child));
4796 
4797 	retval += bus_print_child_domain(dev, child);
4798 	retval += bus_print_child_footer(dev, child);
4799 
4800 	return (retval);
4801 }
4802 
4803 static const struct
4804 {
4805 	int		class;
4806 	int		subclass;
4807 	int		report; /* 0 = bootverbose, 1 = always */
4808 	const char	*desc;
4809 } pci_nomatch_tab[] = {
4810 	{PCIC_OLD,		-1,			1, "old"},
4811 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4812 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4813 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4814 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4815 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4816 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4817 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4818 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4819 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4820 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4821 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4822 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4823 	{PCIC_NETWORK,		-1,			1, "network"},
4824 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4825 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4826 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4827 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4828 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4829 	{PCIC_DISPLAY,		-1,			1, "display"},
4830 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4831 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4832 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4833 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4834 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4835 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4836 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4837 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4838 	{PCIC_MEMORY,		-1,			1, "memory"},
4839 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4840 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4841 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4842 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4843 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4844 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4845 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4846 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4847 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4848 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4849 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4850 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4851 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4852 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4853 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4854 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4855 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4856 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4857 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4858 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4859 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4860 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4861 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4862 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4863 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4864 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4865 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4866 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4867 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4868 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4869 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4870 	{PCIC_DOCKING,		-1,			1, "docking station"},
4871 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4872 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4873 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4874 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4875 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4876 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4877 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4878 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4879 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4880 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4881 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4882 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4883 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4884 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4885 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4886 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4887 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4888 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4889 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4890 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4891 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4892 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4893 	{PCIC_DASP,		-1,			0, "dasp"},
4894 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4895 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4896 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4897 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4898 	{0, 0, 0,		NULL}
4899 };
4900 
4901 void
4902 pci_probe_nomatch(device_t dev, device_t child)
4903 {
4904 	int i, report;
4905 	const char *cp, *scp;
4906 	char *device;
4907 
4908 	/*
4909 	 * Look for a listing for this device in a loaded device database.
4910 	 */
4911 	report = 1;
4912 	if ((device = pci_describe_device(child)) != NULL) {
4913 		device_printf(dev, "<%s>", device);
4914 		free(device, M_DEVBUF);
4915 	} else {
4916 		/*
4917 		 * Scan the class/subclass descriptions for a general
4918 		 * description.
4919 		 */
4920 		cp = "unknown";
4921 		scp = NULL;
4922 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4923 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4924 				if (pci_nomatch_tab[i].subclass == -1) {
4925 					cp = pci_nomatch_tab[i].desc;
4926 					report = pci_nomatch_tab[i].report;
4927 				} else if (pci_nomatch_tab[i].subclass ==
4928 				    pci_get_subclass(child)) {
4929 					scp = pci_nomatch_tab[i].desc;
4930 					report = pci_nomatch_tab[i].report;
4931 				}
4932 			}
4933 		}
4934 		if (report || bootverbose) {
4935 			device_printf(dev, "<%s%s%s>",
4936 			    cp ? cp : "",
4937 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4938 			    scp ? scp : "");
4939 		}
4940 	}
4941 	if (report || bootverbose) {
4942 		printf(" at device %d.%d (no driver attached)\n",
4943 		    pci_get_slot(child), pci_get_function(child));
4944 	}
4945 	pci_cfg_save(child, device_get_ivars(child), 1);
4946 }
4947 
4948 void
4949 pci_child_detached(device_t dev, device_t child)
4950 {
4951 	struct pci_devinfo *dinfo;
4952 	struct resource_list *rl;
4953 
4954 	dinfo = device_get_ivars(child);
4955 	rl = &dinfo->resources;
4956 
4957 	/*
4958 	 * Have to deallocate IRQs before releasing any MSI messages and
4959 	 * have to release MSI messages before deallocating any memory
4960 	 * BARs.
4961 	 */
4962 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4963 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4964 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4965 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4966 		(void)pci_release_msi(child);
4967 	}
4968 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4969 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4970 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4971 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4972 #ifdef PCI_RES_BUS
4973 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4974 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4975 #endif
4976 
4977 	pci_cfg_save(child, dinfo, 1);
4978 }
4979 
4980 /*
4981  * Parse the PCI device database, if loaded, and return a pointer to a
4982  * description of the device.
4983  *
4984  * The database is flat text formatted as follows:
4985  *
4986  * Any line not in a valid format is ignored.
4987  * Lines are terminated with newline '\n' characters.
4988  *
4989  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4990  * the vendor name.
4991  *
4992  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
4993  * - devices cannot be listed without a corresponding VENDOR line.
4994  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
4995  * another TAB, then the device name.
4996  */
4997 
4998 /*
4999  * Assuming (ptr) points to the beginning of a line in the database,
5000  * return the vendor or device and description of the next entry.
5001  * The value of (vendor) or (device) inappropriate for the entry type
5002  * is set to -1.  Returns nonzero at the end of the database.
5003  *
5004  * Note that this is slightly unrobust in the face of corrupt data;
5005  * we attempt to safeguard against this by spamming the end of the
5006  * database with a newline when we initialise.
5007  */
5008 static int
5009 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5010 {
5011 	char	*cp = *ptr;
5012 	int	left;
5013 
5014 	*device = -1;
5015 	*vendor = -1;
5016 	**desc = '\0';
5017 	for (;;) {
5018 		left = pci_vendordata_size - (cp - pci_vendordata);
5019 		if (left <= 0) {
5020 			*ptr = cp;
5021 			return(1);
5022 		}
5023 
5024 		/* vendor entry? */
5025 		if (*cp != '\t' &&
5026 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5027 			break;
5028 		/* device entry? */
5029 		if (*cp == '\t' &&
5030 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5031 			break;
5032 
5033 		/* skip to next line */
5034 		while (*cp != '\n' && left > 0) {
5035 			cp++;
5036 			left--;
5037 		}
5038 		if (*cp == '\n') {
5039 			cp++;
5040 			left--;
5041 		}
5042 	}
5043 	/* skip to next line */
5044 	while (*cp != '\n' && left > 0) {
5045 		cp++;
5046 		left--;
5047 	}
5048 	if (*cp == '\n' && left > 0)
5049 		cp++;
5050 	*ptr = cp;
5051 	return(0);
5052 }
5053 
5054 static char *
5055 pci_describe_device(device_t dev)
5056 {
5057 	int	vendor, device;
5058 	char	*desc, *vp, *dp, *line;
5059 
5060 	desc = vp = dp = NULL;
5061 
5062 	/*
5063 	 * If we have no vendor data, we can't do anything.
5064 	 */
5065 	if (pci_vendordata == NULL)
5066 		goto out;
5067 
5068 	/*
5069 	 * Scan the vendor data looking for this device
5070 	 */
5071 	line = pci_vendordata;
5072 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5073 		goto out;
5074 	for (;;) {
5075 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5076 			goto out;
5077 		if (vendor == pci_get_vendor(dev))
5078 			break;
5079 	}
5080 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5081 		goto out;
5082 	for (;;) {
5083 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5084 			*dp = 0;
5085 			break;
5086 		}
5087 		if (vendor != -1) {
5088 			*dp = 0;
5089 			break;
5090 		}
5091 		if (device == pci_get_device(dev))
5092 			break;
5093 	}
5094 	if (dp[0] == '\0')
5095 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5096 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5097 	    NULL)
5098 		sprintf(desc, "%s, %s", vp, dp);
5099 out:
5100 	if (vp != NULL)
5101 		free(vp, M_DEVBUF);
5102 	if (dp != NULL)
5103 		free(dp, M_DEVBUF);
5104 	return(desc);
5105 }
5106 
5107 int
5108 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5109 {
5110 	struct pci_devinfo *dinfo;
5111 	pcicfgregs *cfg;
5112 
5113 	dinfo = device_get_ivars(child);
5114 	cfg = &dinfo->cfg;
5115 
5116 	switch (which) {
5117 	case PCI_IVAR_ETHADDR:
5118 		/*
5119 		 * The generic accessor doesn't deal with failure, so
5120 		 * we set the return value, then return an error.
5121 		 */
5122 		*((uint8_t **) result) = NULL;
5123 		return (EINVAL);
5124 	case PCI_IVAR_SUBVENDOR:
5125 		*result = cfg->subvendor;
5126 		break;
5127 	case PCI_IVAR_SUBDEVICE:
5128 		*result = cfg->subdevice;
5129 		break;
5130 	case PCI_IVAR_VENDOR:
5131 		*result = cfg->vendor;
5132 		break;
5133 	case PCI_IVAR_DEVICE:
5134 		*result = cfg->device;
5135 		break;
5136 	case PCI_IVAR_DEVID:
5137 		*result = (cfg->device << 16) | cfg->vendor;
5138 		break;
5139 	case PCI_IVAR_CLASS:
5140 		*result = cfg->baseclass;
5141 		break;
5142 	case PCI_IVAR_SUBCLASS:
5143 		*result = cfg->subclass;
5144 		break;
5145 	case PCI_IVAR_PROGIF:
5146 		*result = cfg->progif;
5147 		break;
5148 	case PCI_IVAR_REVID:
5149 		*result = cfg->revid;
5150 		break;
5151 	case PCI_IVAR_INTPIN:
5152 		*result = cfg->intpin;
5153 		break;
5154 	case PCI_IVAR_IRQ:
5155 		*result = cfg->intline;
5156 		break;
5157 	case PCI_IVAR_DOMAIN:
5158 		*result = cfg->domain;
5159 		break;
5160 	case PCI_IVAR_BUS:
5161 		*result = cfg->bus;
5162 		break;
5163 	case PCI_IVAR_SLOT:
5164 		*result = cfg->slot;
5165 		break;
5166 	case PCI_IVAR_FUNCTION:
5167 		*result = cfg->func;
5168 		break;
5169 	case PCI_IVAR_CMDREG:
5170 		*result = cfg->cmdreg;
5171 		break;
5172 	case PCI_IVAR_CACHELNSZ:
5173 		*result = cfg->cachelnsz;
5174 		break;
5175 	case PCI_IVAR_MINGNT:
5176 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5177 			*result = -1;
5178 			return (EINVAL);
5179 		}
5180 		*result = cfg->mingnt;
5181 		break;
5182 	case PCI_IVAR_MAXLAT:
5183 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5184 			*result = -1;
5185 			return (EINVAL);
5186 		}
5187 		*result = cfg->maxlat;
5188 		break;
5189 	case PCI_IVAR_LATTIMER:
5190 		*result = cfg->lattimer;
5191 		break;
5192 	default:
5193 		return (ENOENT);
5194 	}
5195 	return (0);
5196 }
5197 
5198 int
5199 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5200 {
5201 	struct pci_devinfo *dinfo;
5202 
5203 	dinfo = device_get_ivars(child);
5204 
5205 	switch (which) {
5206 	case PCI_IVAR_INTPIN:
5207 		dinfo->cfg.intpin = value;
5208 		return (0);
5209 	case PCI_IVAR_ETHADDR:
5210 	case PCI_IVAR_SUBVENDOR:
5211 	case PCI_IVAR_SUBDEVICE:
5212 	case PCI_IVAR_VENDOR:
5213 	case PCI_IVAR_DEVICE:
5214 	case PCI_IVAR_DEVID:
5215 	case PCI_IVAR_CLASS:
5216 	case PCI_IVAR_SUBCLASS:
5217 	case PCI_IVAR_PROGIF:
5218 	case PCI_IVAR_REVID:
5219 	case PCI_IVAR_IRQ:
5220 	case PCI_IVAR_DOMAIN:
5221 	case PCI_IVAR_BUS:
5222 	case PCI_IVAR_SLOT:
5223 	case PCI_IVAR_FUNCTION:
5224 		return (EINVAL);	/* disallow for now */
5225 
5226 	default:
5227 		return (ENOENT);
5228 	}
5229 }
5230 
5231 #include "opt_ddb.h"
5232 #ifdef DDB
5233 #include <ddb/ddb.h>
5234 #include <sys/cons.h>
5235 
5236 /*
5237  * List resources based on pci map registers, used for within ddb
5238  */
5239 
5240 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5241 {
5242 	struct pci_devinfo *dinfo;
5243 	struct devlist *devlist_head;
5244 	struct pci_conf *p;
5245 	const char *name;
5246 	int i, error, none_count;
5247 
5248 	none_count = 0;
5249 	/* get the head of the device queue */
5250 	devlist_head = &pci_devq;
5251 
5252 	/*
5253 	 * Go through the list of devices and print out devices
5254 	 */
5255 	for (error = 0, i = 0,
5256 	     dinfo = STAILQ_FIRST(devlist_head);
5257 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5258 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5259 
5260 		/* Populate pd_name and pd_unit */
5261 		name = NULL;
5262 		if (dinfo->cfg.dev)
5263 			name = device_get_name(dinfo->cfg.dev);
5264 
5265 		p = &dinfo->conf;
5266 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5267 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5268 			(name && *name) ? name : "none",
5269 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5270 			none_count++,
5271 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5272 			p->pc_sel.pc_func, (p->pc_class << 16) |
5273 			(p->pc_subclass << 8) | p->pc_progif,
5274 			(p->pc_subdevice << 16) | p->pc_subvendor,
5275 			(p->pc_device << 16) | p->pc_vendor,
5276 			p->pc_revid, p->pc_hdr);
5277 	}
5278 }
5279 #endif /* DDB */
5280 
5281 static struct resource *
5282 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5283     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5284     u_int flags)
5285 {
5286 	struct pci_devinfo *dinfo = device_get_ivars(child);
5287 	struct resource_list *rl = &dinfo->resources;
5288 	struct resource *res;
5289 	struct pci_map *pm;
5290 	uint16_t cmd;
5291 	pci_addr_t map, testval;
5292 	int mapsize;
5293 
5294 	res = NULL;
5295 
5296 	/* If rid is managed by EA, ignore it */
5297 	if (pci_ea_is_enabled(child, *rid))
5298 		goto out;
5299 
5300 	pm = pci_find_bar(child, *rid);
5301 	if (pm != NULL) {
5302 		/* This is a BAR that we failed to allocate earlier. */
5303 		mapsize = pm->pm_size;
5304 		map = pm->pm_value;
5305 	} else {
5306 		/*
5307 		 * Weed out the bogons, and figure out how large the
5308 		 * BAR/map is.  BARs that read back 0 here are bogus
5309 		 * and unimplemented.  Note: atapci in legacy mode are
5310 		 * special and handled elsewhere in the code.  If you
5311 		 * have a atapci device in legacy mode and it fails
5312 		 * here, that other code is broken.
5313 		 */
5314 		pci_read_bar(child, *rid, &map, &testval, NULL);
5315 
5316 		/*
5317 		 * Determine the size of the BAR and ignore BARs with a size
5318 		 * of 0.  Device ROM BARs use a different mask value.
5319 		 */
5320 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5321 			mapsize = pci_romsize(testval);
5322 		else
5323 			mapsize = pci_mapsize(testval);
5324 		if (mapsize == 0)
5325 			goto out;
5326 		pm = pci_add_bar(child, *rid, map, mapsize);
5327 	}
5328 
5329 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5330 		if (type != SYS_RES_MEMORY) {
5331 			if (bootverbose)
5332 				device_printf(dev,
5333 				    "child %s requested type %d for rid %#x,"
5334 				    " but the BAR says it is an memio\n",
5335 				    device_get_nameunit(child), type, *rid);
5336 			goto out;
5337 		}
5338 	} else {
5339 		if (type != SYS_RES_IOPORT) {
5340 			if (bootverbose)
5341 				device_printf(dev,
5342 				    "child %s requested type %d for rid %#x,"
5343 				    " but the BAR says it is an ioport\n",
5344 				    device_get_nameunit(child), type, *rid);
5345 			goto out;
5346 		}
5347 	}
5348 
5349 	/*
5350 	 * For real BARs, we need to override the size that
5351 	 * the driver requests, because that's what the BAR
5352 	 * actually uses and we would otherwise have a
5353 	 * situation where we might allocate the excess to
5354 	 * another driver, which won't work.
5355 	 */
5356 	count = ((pci_addr_t)1 << mapsize) * num;
5357 	if (RF_ALIGNMENT(flags) < mapsize)
5358 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5359 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5360 		flags |= RF_PREFETCHABLE;
5361 
5362 	/*
5363 	 * Allocate enough resource, and then write back the
5364 	 * appropriate BAR for that resource.
5365 	 */
5366 	resource_list_add(rl, type, *rid, start, end, count);
5367 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5368 	    count, flags & ~RF_ACTIVE);
5369 	if (res == NULL) {
5370 		resource_list_delete(rl, type, *rid);
5371 		device_printf(child,
5372 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5373 		    count, *rid, type, start, end);
5374 		goto out;
5375 	}
5376 	if (bootverbose)
5377 		device_printf(child,
5378 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5379 		    count, *rid, type, rman_get_start(res));
5380 
5381 	/* Disable decoding via the CMD register before updating the BAR */
5382 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5383 	pci_write_config(child, PCIR_COMMAND,
5384 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5385 
5386 	map = rman_get_start(res);
5387 	pci_write_bar(child, pm, map);
5388 
5389 	/* Restore the original value of the CMD register */
5390 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5391 out:
5392 	return (res);
5393 }
5394 
5395 struct resource *
5396 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5397     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5398     u_int flags)
5399 {
5400 	struct pci_devinfo *dinfo;
5401 	struct resource_list *rl;
5402 	struct resource_list_entry *rle;
5403 	struct resource *res;
5404 	pcicfgregs *cfg;
5405 
5406 	/*
5407 	 * Perform lazy resource allocation
5408 	 */
5409 	dinfo = device_get_ivars(child);
5410 	rl = &dinfo->resources;
5411 	cfg = &dinfo->cfg;
5412 	switch (type) {
5413 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5414 	case PCI_RES_BUS:
5415 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5416 		    flags));
5417 #endif
5418 	case SYS_RES_IRQ:
5419 		/*
5420 		 * Can't alloc legacy interrupt once MSI messages have
5421 		 * been allocated.
5422 		 */
5423 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5424 		    cfg->msix.msix_alloc > 0))
5425 			return (NULL);
5426 
5427 		/*
5428 		 * If the child device doesn't have an interrupt
5429 		 * routed and is deserving of an interrupt, try to
5430 		 * assign it one.
5431 		 */
5432 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5433 		    (cfg->intpin != 0))
5434 			pci_assign_interrupt(dev, child, 0);
5435 		break;
5436 	case SYS_RES_IOPORT:
5437 	case SYS_RES_MEMORY:
5438 #ifdef NEW_PCIB
5439 		/*
5440 		 * PCI-PCI bridge I/O window resources are not BARs.
5441 		 * For those allocations just pass the request up the
5442 		 * tree.
5443 		 */
5444 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5445 			switch (*rid) {
5446 			case PCIR_IOBASEL_1:
5447 			case PCIR_MEMBASE_1:
5448 			case PCIR_PMBASEL_1:
5449 				/*
5450 				 * XXX: Should we bother creating a resource
5451 				 * list entry?
5452 				 */
5453 				return (bus_generic_alloc_resource(dev, child,
5454 				    type, rid, start, end, count, flags));
5455 			}
5456 		}
5457 #endif
5458 		/* Reserve resources for this BAR if needed. */
5459 		rle = resource_list_find(rl, type, *rid);
5460 		if (rle == NULL) {
5461 			res = pci_reserve_map(dev, child, type, rid, start, end,
5462 			    count, num, flags);
5463 			if (res == NULL)
5464 				return (NULL);
5465 		}
5466 	}
5467 	return (resource_list_alloc(rl, dev, child, type, rid,
5468 	    start, end, count, flags));
5469 }
5470 
5471 struct resource *
5472 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5473     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5474 {
5475 #ifdef PCI_IOV
5476 	struct pci_devinfo *dinfo;
5477 #endif
5478 
5479 	if (device_get_parent(child) != dev)
5480 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5481 		    type, rid, start, end, count, flags));
5482 
5483 #ifdef PCI_IOV
5484 	dinfo = device_get_ivars(child);
5485 	if (dinfo->cfg.flags & PCICFG_VF) {
5486 		switch (type) {
5487 		/* VFs can't have I/O BARs. */
5488 		case SYS_RES_IOPORT:
5489 			return (NULL);
5490 		case SYS_RES_MEMORY:
5491 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5492 			    start, end, count, flags));
5493 		}
5494 
5495 		/* Fall through for other types of resource allocations. */
5496 	}
5497 #endif
5498 
5499 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5500 	    count, 1, flags));
5501 }
5502 
5503 int
5504 pci_release_resource(device_t dev, device_t child, int type, int rid,
5505     struct resource *r)
5506 {
5507 	struct pci_devinfo *dinfo;
5508 	struct resource_list *rl;
5509 	pcicfgregs *cfg;
5510 
5511 	if (device_get_parent(child) != dev)
5512 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5513 		    type, rid, r));
5514 
5515 	dinfo = device_get_ivars(child);
5516 	cfg = &dinfo->cfg;
5517 
5518 #ifdef PCI_IOV
5519 	if (dinfo->cfg.flags & PCICFG_VF) {
5520 		switch (type) {
5521 		/* VFs can't have I/O BARs. */
5522 		case SYS_RES_IOPORT:
5523 			return (EDOOFUS);
5524 		case SYS_RES_MEMORY:
5525 			return (pci_vf_release_mem_resource(dev, child, rid,
5526 			    r));
5527 		}
5528 
5529 		/* Fall through for other types of resource allocations. */
5530 	}
5531 #endif
5532 
5533 #ifdef NEW_PCIB
5534 	/*
5535 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5536 	 * those allocations just pass the request up the tree.
5537 	 */
5538 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5539 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5540 		switch (rid) {
5541 		case PCIR_IOBASEL_1:
5542 		case PCIR_MEMBASE_1:
5543 		case PCIR_PMBASEL_1:
5544 			return (bus_generic_release_resource(dev, child, type,
5545 			    rid, r));
5546 		}
5547 	}
5548 #endif
5549 
5550 	rl = &dinfo->resources;
5551 	return (resource_list_release(rl, dev, child, type, rid, r));
5552 }
5553 
5554 int
5555 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5556     struct resource *r)
5557 {
5558 	struct pci_devinfo *dinfo;
5559 	int error;
5560 
5561 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5562 	if (error)
5563 		return (error);
5564 
5565 	/* Enable decoding in the command register when activating BARs. */
5566 	if (device_get_parent(child) == dev) {
5567 		/* Device ROMs need their decoding explicitly enabled. */
5568 		dinfo = device_get_ivars(child);
5569 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5570 			pci_write_bar(child, pci_find_bar(child, rid),
5571 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5572 		switch (type) {
5573 		case SYS_RES_IOPORT:
5574 		case SYS_RES_MEMORY:
5575 			error = PCI_ENABLE_IO(dev, child, type);
5576 			break;
5577 		}
5578 	}
5579 	return (error);
5580 }
5581 
5582 int
5583 pci_deactivate_resource(device_t dev, device_t child, int type,
5584     int rid, struct resource *r)
5585 {
5586 	struct pci_devinfo *dinfo;
5587 	int error;
5588 
5589 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5590 	if (error)
5591 		return (error);
5592 
5593 	/* Disable decoding for device ROMs. */
5594 	if (device_get_parent(child) == dev) {
5595 		dinfo = device_get_ivars(child);
5596 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5597 			pci_write_bar(child, pci_find_bar(child, rid),
5598 			    rman_get_start(r));
5599 	}
5600 	return (0);
5601 }
5602 
5603 void
5604 pci_child_deleted(device_t dev, device_t child)
5605 {
5606 	struct resource_list_entry *rle;
5607 	struct resource_list *rl;
5608 	struct pci_devinfo *dinfo;
5609 
5610 	dinfo = device_get_ivars(child);
5611 	rl = &dinfo->resources;
5612 
5613 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5614 
5615 	/* Turn off access to resources we're about to free */
5616 	if (bus_child_present(child) != 0) {
5617 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5618 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5619 
5620 		pci_disable_busmaster(child);
5621 	}
5622 
5623 	/* Free all allocated resources */
5624 	STAILQ_FOREACH(rle, rl, link) {
5625 		if (rle->res) {
5626 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5627 			    resource_list_busy(rl, rle->type, rle->rid)) {
5628 				pci_printf(&dinfo->cfg,
5629 				    "Resource still owned, oops. "
5630 				    "(type=%d, rid=%d, addr=%lx)\n",
5631 				    rle->type, rle->rid,
5632 				    rman_get_start(rle->res));
5633 				bus_release_resource(child, rle->type, rle->rid,
5634 				    rle->res);
5635 			}
5636 			resource_list_unreserve(rl, dev, child, rle->type,
5637 			    rle->rid);
5638 		}
5639 	}
5640 	resource_list_free(rl);
5641 
5642 	pci_freecfg(dinfo);
5643 }
5644 
5645 void
5646 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5647 {
5648 	struct pci_devinfo *dinfo;
5649 	struct resource_list *rl;
5650 	struct resource_list_entry *rle;
5651 
5652 	if (device_get_parent(child) != dev)
5653 		return;
5654 
5655 	dinfo = device_get_ivars(child);
5656 	rl = &dinfo->resources;
5657 	rle = resource_list_find(rl, type, rid);
5658 	if (rle == NULL)
5659 		return;
5660 
5661 	if (rle->res) {
5662 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5663 		    resource_list_busy(rl, type, rid)) {
5664 			device_printf(dev, "delete_resource: "
5665 			    "Resource still owned by child, oops. "
5666 			    "(type=%d, rid=%d, addr=%jx)\n",
5667 			    type, rid, rman_get_start(rle->res));
5668 			return;
5669 		}
5670 		resource_list_unreserve(rl, dev, child, type, rid);
5671 	}
5672 	resource_list_delete(rl, type, rid);
5673 }
5674 
5675 struct resource_list *
5676 pci_get_resource_list (device_t dev, device_t child)
5677 {
5678 	struct pci_devinfo *dinfo = device_get_ivars(child);
5679 
5680 	return (&dinfo->resources);
5681 }
5682 
5683 #ifdef ACPI_DMAR
5684 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5685 bus_dma_tag_t
5686 pci_get_dma_tag(device_t bus, device_t dev)
5687 {
5688 	bus_dma_tag_t tag;
5689 	struct pci_softc *sc;
5690 
5691 	if (device_get_parent(dev) == bus) {
5692 		/* try dmar and return if it works */
5693 		tag = dmar_get_dma_tag(bus, dev);
5694 	} else
5695 		tag = NULL;
5696 	if (tag == NULL) {
5697 		sc = device_get_softc(bus);
5698 		tag = sc->sc_dma_tag;
5699 	}
5700 	return (tag);
5701 }
5702 #else
5703 bus_dma_tag_t
5704 pci_get_dma_tag(device_t bus, device_t dev)
5705 {
5706 	struct pci_softc *sc = device_get_softc(bus);
5707 
5708 	return (sc->sc_dma_tag);
5709 }
5710 #endif
5711 
5712 uint32_t
5713 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5714 {
5715 	struct pci_devinfo *dinfo = device_get_ivars(child);
5716 	pcicfgregs *cfg = &dinfo->cfg;
5717 
5718 #ifdef PCI_IOV
5719 	/*
5720 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5721 	 * emulate them here.
5722 	 */
5723 	if (cfg->flags & PCICFG_VF) {
5724 		if (reg == PCIR_VENDOR) {
5725 			switch (width) {
5726 			case 4:
5727 				return (cfg->device << 16 | cfg->vendor);
5728 			case 2:
5729 				return (cfg->vendor);
5730 			case 1:
5731 				return (cfg->vendor & 0xff);
5732 			default:
5733 				return (0xffffffff);
5734 			}
5735 		} else if (reg == PCIR_DEVICE) {
5736 			switch (width) {
5737 			/* Note that an unaligned 4-byte read is an error. */
5738 			case 2:
5739 				return (cfg->device);
5740 			case 1:
5741 				return (cfg->device & 0xff);
5742 			default:
5743 				return (0xffffffff);
5744 			}
5745 		}
5746 	}
5747 #endif
5748 
5749 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5750 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5751 }
5752 
5753 void
5754 pci_write_config_method(device_t dev, device_t child, int reg,
5755     uint32_t val, int width)
5756 {
5757 	struct pci_devinfo *dinfo = device_get_ivars(child);
5758 	pcicfgregs *cfg = &dinfo->cfg;
5759 
5760 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5761 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5762 }
5763 
5764 int
5765 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5766     size_t buflen)
5767 {
5768 
5769 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5770 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5771 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5772 	return (0);
5773 }
5774 
5775 int
5776 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5777     size_t buflen)
5778 {
5779 	struct pci_devinfo *dinfo;
5780 	pcicfgregs *cfg;
5781 
5782 	dinfo = device_get_ivars(child);
5783 	cfg = &dinfo->cfg;
5784 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5785 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5786 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5787 	    cfg->progif);
5788 	return (0);
5789 }
5790 
5791 int
5792 pci_assign_interrupt_method(device_t dev, device_t child)
5793 {
5794 	struct pci_devinfo *dinfo = device_get_ivars(child);
5795 	pcicfgregs *cfg = &dinfo->cfg;
5796 
5797 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5798 	    cfg->intpin));
5799 }
5800 
5801 static void
5802 pci_lookup(void *arg, const char *name, device_t *dev)
5803 {
5804 	long val;
5805 	char *end;
5806 	int domain, bus, slot, func;
5807 
5808 	if (*dev != NULL)
5809 		return;
5810 
5811 	/*
5812 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5813 	 * pciB:S:F.  In the latter case, the domain is assumed to
5814 	 * be zero.
5815 	 */
5816 	if (strncmp(name, "pci", 3) != 0)
5817 		return;
5818 	val = strtol(name + 3, &end, 10);
5819 	if (val < 0 || val > INT_MAX || *end != ':')
5820 		return;
5821 	domain = val;
5822 	val = strtol(end + 1, &end, 10);
5823 	if (val < 0 || val > INT_MAX || *end != ':')
5824 		return;
5825 	bus = val;
5826 	val = strtol(end + 1, &end, 10);
5827 	if (val < 0 || val > INT_MAX)
5828 		return;
5829 	slot = val;
5830 	if (*end == ':') {
5831 		val = strtol(end + 1, &end, 10);
5832 		if (val < 0 || val > INT_MAX || *end != '\0')
5833 			return;
5834 		func = val;
5835 	} else if (*end == '\0') {
5836 		func = slot;
5837 		slot = bus;
5838 		bus = domain;
5839 		domain = 0;
5840 	} else
5841 		return;
5842 
5843 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5844 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5845 		return;
5846 
5847 	*dev = pci_find_dbsf(domain, bus, slot, func);
5848 }
5849 
5850 static int
5851 pci_modevent(module_t mod, int what, void *arg)
5852 {
5853 	static struct cdev *pci_cdev;
5854 	static eventhandler_tag tag;
5855 
5856 	switch (what) {
5857 	case MOD_LOAD:
5858 		STAILQ_INIT(&pci_devq);
5859 		pci_generation = 0;
5860 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5861 		    "pci");
5862 		pci_load_vendor_data();
5863 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5864 		    1000);
5865 		break;
5866 
5867 	case MOD_UNLOAD:
5868 		if (tag != NULL)
5869 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5870 		destroy_dev(pci_cdev);
5871 		break;
5872 	}
5873 
5874 	return (0);
5875 }
5876 
5877 static void
5878 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5879 {
5880 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5881 	struct pcicfg_pcie *cfg;
5882 	int version, pos;
5883 
5884 	cfg = &dinfo->cfg.pcie;
5885 	pos = cfg->pcie_location;
5886 
5887 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5888 
5889 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5890 
5891 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5892 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5893 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5894 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5895 
5896 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5897 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5898 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5899 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5900 
5901 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5902 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5903 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5904 
5905 	if (version > 1) {
5906 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5907 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5908 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5909 	}
5910 #undef WREG
5911 }
5912 
5913 static void
5914 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5915 {
5916 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5917 	    dinfo->cfg.pcix.pcix_command,  2);
5918 }
5919 
5920 void
5921 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5922 {
5923 
5924 	/*
5925 	 * Restore the device to full power mode.  We must do this
5926 	 * before we restore the registers because moving from D3 to
5927 	 * D0 will cause the chip's BARs and some other registers to
5928 	 * be reset to some unknown power on reset values.  Cut down
5929 	 * the noise on boot by doing nothing if we are already in
5930 	 * state D0.
5931 	 */
5932 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5933 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5934 	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5935 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5936 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5937 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5938 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5939 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5940 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5941 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5942 	case PCIM_HDRTYPE_NORMAL:
5943 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5944 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5945 		break;
5946 	case PCIM_HDRTYPE_BRIDGE:
5947 		pci_write_config(dev, PCIR_SECLAT_1,
5948 		    dinfo->cfg.bridge.br_seclat, 1);
5949 		pci_write_config(dev, PCIR_SUBBUS_1,
5950 		    dinfo->cfg.bridge.br_subbus, 1);
5951 		pci_write_config(dev, PCIR_SECBUS_1,
5952 		    dinfo->cfg.bridge.br_secbus, 1);
5953 		pci_write_config(dev, PCIR_PRIBUS_1,
5954 		    dinfo->cfg.bridge.br_pribus, 1);
5955 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5956 		    dinfo->cfg.bridge.br_control, 2);
5957 		break;
5958 	case PCIM_HDRTYPE_CARDBUS:
5959 		pci_write_config(dev, PCIR_SECLAT_2,
5960 		    dinfo->cfg.bridge.br_seclat, 1);
5961 		pci_write_config(dev, PCIR_SUBBUS_2,
5962 		    dinfo->cfg.bridge.br_subbus, 1);
5963 		pci_write_config(dev, PCIR_SECBUS_2,
5964 		    dinfo->cfg.bridge.br_secbus, 1);
5965 		pci_write_config(dev, PCIR_PRIBUS_2,
5966 		    dinfo->cfg.bridge.br_pribus, 1);
5967 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5968 		    dinfo->cfg.bridge.br_control, 2);
5969 		break;
5970 	}
5971 	pci_restore_bars(dev);
5972 
5973 	/*
5974 	 * Restore extended capabilities for PCI-Express and PCI-X
5975 	 */
5976 	if (dinfo->cfg.pcie.pcie_location != 0)
5977 		pci_cfg_restore_pcie(dev, dinfo);
5978 	if (dinfo->cfg.pcix.pcix_location != 0)
5979 		pci_cfg_restore_pcix(dev, dinfo);
5980 
5981 	/* Restore MSI and MSI-X configurations if they are present. */
5982 	if (dinfo->cfg.msi.msi_location != 0)
5983 		pci_resume_msi(dev);
5984 	if (dinfo->cfg.msix.msix_location != 0)
5985 		pci_resume_msix(dev);
5986 
5987 #ifdef PCI_IOV
5988 	if (dinfo->cfg.iov != NULL)
5989 		pci_iov_cfg_restore(dev, dinfo);
5990 #endif
5991 }
5992 
5993 static void
5994 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
5995 {
5996 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
5997 	struct pcicfg_pcie *cfg;
5998 	int version, pos;
5999 
6000 	cfg = &dinfo->cfg.pcie;
6001 	pos = cfg->pcie_location;
6002 
6003 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6004 
6005 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6006 
6007 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6008 
6009 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6010 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6011 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6012 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6013 
6014 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6015 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6016 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6017 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6018 
6019 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6020 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6021 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6022 
6023 	if (version > 1) {
6024 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6025 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6026 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6027 	}
6028 #undef RREG
6029 }
6030 
6031 static void
6032 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6033 {
6034 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6035 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6036 }
6037 
6038 void
6039 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6040 {
6041 	uint32_t cls;
6042 	int ps;
6043 
6044 	/*
6045 	 * Some drivers apparently write to these registers w/o updating our
6046 	 * cached copy.  No harm happens if we update the copy, so do so here
6047 	 * so we can restore them.  The COMMAND register is modified by the
6048 	 * bus w/o updating the cache.  This should represent the normally
6049 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6050 	 */
6051 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6052 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6053 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6054 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6055 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6056 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6057 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6058 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6059 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6060 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6061 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6062 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6063 	case PCIM_HDRTYPE_NORMAL:
6064 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6065 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6066 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6067 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6068 		break;
6069 	case PCIM_HDRTYPE_BRIDGE:
6070 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6071 		    PCIR_SECLAT_1, 1);
6072 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6073 		    PCIR_SUBBUS_1, 1);
6074 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6075 		    PCIR_SECBUS_1, 1);
6076 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6077 		    PCIR_PRIBUS_1, 1);
6078 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6079 		    PCIR_BRIDGECTL_1, 2);
6080 		break;
6081 	case PCIM_HDRTYPE_CARDBUS:
6082 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6083 		    PCIR_SECLAT_2, 1);
6084 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6085 		    PCIR_SUBBUS_2, 1);
6086 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6087 		    PCIR_SECBUS_2, 1);
6088 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6089 		    PCIR_PRIBUS_2, 1);
6090 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6091 		    PCIR_BRIDGECTL_2, 2);
6092 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6093 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6094 		break;
6095 	}
6096 
6097 	if (dinfo->cfg.pcie.pcie_location != 0)
6098 		pci_cfg_save_pcie(dev, dinfo);
6099 
6100 	if (dinfo->cfg.pcix.pcix_location != 0)
6101 		pci_cfg_save_pcix(dev, dinfo);
6102 
6103 #ifdef PCI_IOV
6104 	if (dinfo->cfg.iov != NULL)
6105 		pci_iov_cfg_save(dev, dinfo);
6106 #endif
6107 
6108 	/*
6109 	 * don't set the state for display devices, base peripherals and
6110 	 * memory devices since bad things happen when they are powered down.
6111 	 * We should (a) have drivers that can easily detach and (b) use
6112 	 * generic drivers for these devices so that some device actually
6113 	 * attaches.  We need to make sure that when we implement (a) we don't
6114 	 * power the device down on a reattach.
6115 	 */
6116 	cls = pci_get_class(dev);
6117 	if (!setstate)
6118 		return;
6119 	switch (pci_do_power_nodriver)
6120 	{
6121 		case 0:		/* NO powerdown at all */
6122 			return;
6123 		case 1:		/* Conservative about what to power down */
6124 			if (cls == PCIC_STORAGE)
6125 				return;
6126 			/*FALLTHROUGH*/
6127 		case 2:		/* Aggressive about what to power down */
6128 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6129 			    cls == PCIC_BASEPERIPH)
6130 				return;
6131 			/*FALLTHROUGH*/
6132 		case 3:		/* Power down everything */
6133 			break;
6134 	}
6135 	/*
6136 	 * PCI spec says we can only go into D3 state from D0 state.
6137 	 * Transition from D[12] into D0 before going to D3 state.
6138 	 */
6139 	ps = pci_get_powerstate(dev);
6140 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6141 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6142 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6143 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6144 }
6145 
6146 /* Wrapper APIs suitable for device driver use. */
6147 void
6148 pci_save_state(device_t dev)
6149 {
6150 	struct pci_devinfo *dinfo;
6151 
6152 	dinfo = device_get_ivars(dev);
6153 	pci_cfg_save(dev, dinfo, 0);
6154 }
6155 
6156 void
6157 pci_restore_state(device_t dev)
6158 {
6159 	struct pci_devinfo *dinfo;
6160 
6161 	dinfo = device_get_ivars(dev);
6162 	pci_cfg_restore(dev, dinfo);
6163 }
6164 
6165 static int
6166 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6167     uintptr_t *id)
6168 {
6169 
6170 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6171 }
6172 
6173 /* Find the upstream port of a given PCI device in a root complex. */
6174 device_t
6175 pci_find_pcie_root_port(device_t dev)
6176 {
6177 	struct pci_devinfo *dinfo;
6178 	devclass_t pci_class;
6179 	device_t pcib, bus;
6180 
6181 	pci_class = devclass_find("pci");
6182 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6183 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6184 
6185 	/*
6186 	 * Walk the bridge hierarchy until we find a PCI-e root
6187 	 * port or a non-PCI device.
6188 	 */
6189 	for (;;) {
6190 		bus = device_get_parent(dev);
6191 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6192 		    device_get_nameunit(dev)));
6193 
6194 		pcib = device_get_parent(bus);
6195 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6196 		    device_get_nameunit(bus)));
6197 
6198 		/*
6199 		 * pcib's parent must be a PCI bus for this to be a
6200 		 * PCI-PCI bridge.
6201 		 */
6202 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6203 			return (NULL);
6204 
6205 		dinfo = device_get_ivars(pcib);
6206 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6207 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6208 			return (pcib);
6209 
6210 		dev = pcib;
6211 	}
6212 }
6213 
6214 /*
6215  * Wait for pending transactions to complete on a PCI-express function.
6216  *
6217  * The maximum delay is specified in milliseconds in max_delay.  Note
6218  * that this function may sleep.
6219  *
6220  * Returns true if the function is idle and false if the timeout is
6221  * exceeded.  If dev is not a PCI-express function, this returns true.
6222  */
6223 bool
6224 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6225 {
6226 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6227 	uint16_t sta;
6228 	int cap;
6229 
6230 	cap = dinfo->cfg.pcie.pcie_location;
6231 	if (cap == 0)
6232 		return (true);
6233 
6234 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6235 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6236 		if (max_delay == 0)
6237 			return (false);
6238 
6239 		/* Poll once every 100 milliseconds up to the timeout. */
6240 		if (max_delay > 100) {
6241 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6242 			max_delay -= 100;
6243 		} else {
6244 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6245 			    C_HARDCLOCK);
6246 			max_delay = 0;
6247 		}
6248 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6249 	}
6250 
6251 	return (true);
6252 }
6253 
6254 /*
6255  * Determine the maximum Completion Timeout in microseconds.
6256  *
6257  * For non-PCI-express functions this returns 0.
6258  */
6259 int
6260 pcie_get_max_completion_timeout(device_t dev)
6261 {
6262 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6263 	int cap;
6264 
6265 	cap = dinfo->cfg.pcie.pcie_location;
6266 	if (cap == 0)
6267 		return (0);
6268 
6269 	/*
6270 	 * Functions using the 1.x spec use the default timeout range of
6271 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6272 	 * support programmable timeouts also use this range.
6273 	 */
6274 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6275 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6276 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6277 		return (50 * 1000);
6278 
6279 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6280 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6281 	case PCIEM_CTL2_COMP_TIMO_100US:
6282 		return (100);
6283 	case PCIEM_CTL2_COMP_TIMO_10MS:
6284 		return (10 * 1000);
6285 	case PCIEM_CTL2_COMP_TIMO_55MS:
6286 		return (55 * 1000);
6287 	case PCIEM_CTL2_COMP_TIMO_210MS:
6288 		return (210 * 1000);
6289 	case PCIEM_CTL2_COMP_TIMO_900MS:
6290 		return (900 * 1000);
6291 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6292 		return (3500 * 1000);
6293 	case PCIEM_CTL2_COMP_TIMO_13S:
6294 		return (13 * 1000 * 1000);
6295 	case PCIEM_CTL2_COMP_TIMO_64S:
6296 		return (64 * 1000 * 1000);
6297 	default:
6298 		return (50 * 1000);
6299 	}
6300 }
6301 
6302 /*
6303  * Perform a Function Level Reset (FLR) on a device.
6304  *
6305  * This function first waits for any pending transactions to complete
6306  * within the timeout specified by max_delay.  If transactions are
6307  * still pending, the function will return false without attempting a
6308  * reset.
6309  *
6310  * If dev is not a PCI-express function or does not support FLR, this
6311  * function returns false.
6312  *
6313  * Note that no registers are saved or restored.  The caller is
6314  * responsible for saving and restoring any registers including
6315  * PCI-standard registers via pci_save_state() and
6316  * pci_restore_state().
6317  */
6318 bool
6319 pcie_flr(device_t dev, u_int max_delay, bool force)
6320 {
6321 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6322 	uint16_t cmd, ctl;
6323 	int compl_delay;
6324 	int cap;
6325 
6326 	cap = dinfo->cfg.pcie.pcie_location;
6327 	if (cap == 0)
6328 		return (false);
6329 
6330 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6331 		return (false);
6332 
6333 	/*
6334 	 * Disable busmastering to prevent generation of new
6335 	 * transactions while waiting for the device to go idle.  If
6336 	 * the idle timeout fails, the command register is restored
6337 	 * which will re-enable busmastering.
6338 	 */
6339 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6340 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6341 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6342 		if (!force) {
6343 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6344 			return (false);
6345 		}
6346 		pci_printf(&dinfo->cfg,
6347 		    "Resetting with transactions pending after %d ms\n",
6348 		    max_delay);
6349 
6350 		/*
6351 		 * Extend the post-FLR delay to cover the maximum
6352 		 * Completion Timeout delay of anything in flight
6353 		 * during the FLR delay.  Enforce a minimum delay of
6354 		 * at least 10ms.
6355 		 */
6356 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6357 		if (compl_delay < 10)
6358 			compl_delay = 10;
6359 	} else
6360 		compl_delay = 0;
6361 
6362 	/* Initiate the reset. */
6363 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6364 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6365 	    PCIEM_CTL_INITIATE_FLR, 2);
6366 
6367 	/* Wait for 100ms. */
6368 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6369 
6370 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6371 	    PCIEM_STA_TRANSACTION_PND)
6372 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6373 	return (true);
6374 }
6375 
6376 /*
6377  * Attempt a power-management reset by cycling the device in/out of D3
6378  * state.  PCI spec says we can only go into D3 state from D0 state.
6379  * Transition from D[12] into D0 before going to D3 state.
6380  */
6381 int
6382 pci_power_reset(device_t dev)
6383 {
6384 	int ps;
6385 
6386 	ps = pci_get_powerstate(dev);
6387 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6388 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6389 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6390 	pci_set_powerstate(dev, ps);
6391 	return (0);
6392 }
6393 
6394 /*
6395  * Try link drop and retrain of the downstream port of upstream
6396  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6397  * cause Conventional Hot reset of the device in the slot.
6398  * Alternative, for PCIe, could be the secondary bus reset initiatied
6399  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6400  */
6401 int
6402 pcie_link_reset(device_t port, int pcie_location)
6403 {
6404 	uint16_t v;
6405 
6406 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6407 	v |= PCIEM_LINK_CTL_LINK_DIS;
6408 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6409 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6410 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6411 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6412 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6413 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6414 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6415 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6416 }
6417 
6418 static int
6419 pci_reset_post(device_t dev, device_t child)
6420 {
6421 
6422 	if (dev == device_get_parent(child))
6423 		pci_restore_state(child);
6424 	return (0);
6425 }
6426 
6427 static int
6428 pci_reset_prepare(device_t dev, device_t child)
6429 {
6430 
6431 	if (dev == device_get_parent(child))
6432 		pci_save_state(child);
6433 	return (0);
6434 }
6435 
6436 static int
6437 pci_reset_child(device_t dev, device_t child, int flags)
6438 {
6439 	int error;
6440 
6441 	if (dev == NULL || device_get_parent(child) != dev)
6442 		return (0);
6443 	if ((flags & DEVF_RESET_DETACH) != 0) {
6444 		error = device_get_state(child) == DS_ATTACHED ?
6445 		    device_detach(child) : 0;
6446 	} else {
6447 		error = BUS_SUSPEND_CHILD(dev, child);
6448 	}
6449 	if (error == 0) {
6450 		if (!pcie_flr(child, 1000, false)) {
6451 			error = BUS_RESET_PREPARE(dev, child);
6452 			if (error == 0)
6453 				pci_power_reset(child);
6454 			BUS_RESET_POST(dev, child);
6455 		}
6456 		if ((flags & DEVF_RESET_DETACH) != 0)
6457 			device_probe_and_attach(child);
6458 		else
6459 			BUS_RESUME_CHILD(dev, child);
6460 	}
6461 	return (error);
6462 }
6463 
6464 const struct pci_device_table *
6465 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6466 {
6467 	bool match;
6468 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6469 
6470 	vendor = pci_get_vendor(child);
6471 	device = pci_get_device(child);
6472 	subvendor = pci_get_subvendor(child);
6473 	subdevice = pci_get_subdevice(child);
6474 	class = pci_get_class(child);
6475 	subclass = pci_get_subclass(child);
6476 	revid = pci_get_revid(child);
6477 	while (nelt-- > 0) {
6478 		match = true;
6479 		if (id->match_flag_vendor)
6480 			match &= vendor == id->vendor;
6481 		if (id->match_flag_device)
6482 			match &= device == id->device;
6483 		if (id->match_flag_subvendor)
6484 			match &= subvendor == id->subvendor;
6485 		if (id->match_flag_subdevice)
6486 			match &= subdevice == id->subdevice;
6487 		if (id->match_flag_class)
6488 			match &= class == id->class_id;
6489 		if (id->match_flag_subclass)
6490 			match &= subclass == id->subclass;
6491 		if (id->match_flag_revid)
6492 			match &= revid == id->revid;
6493 		if (match)
6494 			return (id);
6495 		id++;
6496 	}
6497 	return (NULL);
6498 }
6499 
6500 static void
6501 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6502 {
6503 	const char *dev_name;
6504 	device_t dev;
6505 
6506 	dev = dinfo->cfg.dev;
6507 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6508 	    dinfo->cfg.slot, dinfo->cfg.func);
6509 	dev_name = device_get_name(dev);
6510 	if (dev_name != NULL)
6511 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6512 }
6513 
6514 void
6515 pci_print_faulted_dev(void)
6516 {
6517 	struct pci_devinfo *dinfo;
6518 	device_t dev;
6519 	int aer, i;
6520 	uint32_t r1, r2;
6521 	uint16_t status;
6522 
6523 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6524 		dev = dinfo->cfg.dev;
6525 		status = pci_read_config(dev, PCIR_STATUS, 2);
6526 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6527 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6528 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6529 		if (status != 0) {
6530 			pci_print_faulted_dev_name(dinfo);
6531 			printf(" error 0x%04x\n", status);
6532 		}
6533 		if (dinfo->cfg.pcie.pcie_location != 0) {
6534 			status = pci_read_config(dev,
6535 			    dinfo->cfg.pcie.pcie_location +
6536 			    PCIER_DEVICE_STA, 2);
6537 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6538 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6539 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6540 				pci_print_faulted_dev_name(dinfo);
6541 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6542 				    pci_read_config(dev,
6543 				    dinfo->cfg.pcie.pcie_location +
6544 				    PCIER_DEVICE_CTL, 2),
6545 				    status);
6546 			}
6547 		}
6548 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6549 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6550 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6551 			if (r1 != 0 || r2 != 0) {
6552 				pci_print_faulted_dev_name(dinfo);
6553 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6554 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6555 				    r1, pci_read_config(dev, aer +
6556 				    PCIR_AER_UC_MASK, 4),
6557 				    pci_read_config(dev, aer +
6558 				    PCIR_AER_UC_SEVERITY, 4),
6559 				    r2, pci_read_config(dev, aer +
6560 				    PCIR_AER_COR_MASK, 4),
6561 				    pci_read_config(dev, aer +
6562 				    PCIR_AER_CAP_CONTROL, 4));
6563 				for (i = 0; i < 4; i++) {
6564 					r1 = pci_read_config(dev, aer +
6565 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6566 					printf("    HL%d: 0x%08x\n", i, r1);
6567 				}
6568 			}
6569 		}
6570 	}
6571 }
6572 
6573 #ifdef DDB
6574 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6575 {
6576 
6577 	pci_print_faulted_dev();
6578 }
6579 
6580 static void
6581 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6582 {
6583 	device_t dev;
6584 	int aer;
6585 	uint32_t r;
6586 
6587 	dev = dinfo->cfg.dev;
6588 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6589 	    PCIER_DEVICE_STA, 2);
6590 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6591 	    PCIER_DEVICE_STA, r, 2);
6592 
6593 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6594 		return;
6595 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6596 	if (r != 0)
6597 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6598 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6599 	if (r != 0)
6600 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6601 }
6602 
6603 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6604 {
6605 	struct pci_devinfo *dinfo;
6606 	device_t dev;
6607 	uint16_t status, status1;
6608 
6609 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6610 		dev = dinfo->cfg.dev;
6611 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6612 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6613 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6614 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6615 		if (status1 != 0) {
6616 			status &= ~status1;
6617 			pci_write_config(dev, PCIR_STATUS, status, 2);
6618 		}
6619 		if (dinfo->cfg.pcie.pcie_location != 0)
6620 			db_clear_pcie_errors(dinfo);
6621 	}
6622 }
6623 #endif
6624