xref: /freebsd/sys/dev/pci/pci.c (revision c6989859ae9388eeb46a24fe88f9b8d07101c710)
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 	{ 0 }
321 };
322 
323 /* map register information */
324 #define	PCI_MAPMEM	0x01	/* memory map */
325 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
326 #define	PCI_MAPPORT	0x04	/* port map */
327 
328 struct devlist pci_devq;
329 uint32_t pci_generation;
330 uint32_t pci_numdevs = 0;
331 static int pcie_chipset, pcix_chipset;
332 
333 /* sysctl vars */
334 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
335     "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 int pci_enable_aspm = 1;
411 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
412     0, "Enable support for PCIe Active State Power Management");
413 
414 static int pci_clear_aer_on_attach = 0;
415 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
416     &pci_clear_aer_on_attach, 0,
417     "Clear port and device AER state on driver attach");
418 
419 static int
420 pci_has_quirk(uint32_t devid, int quirk)
421 {
422 	const struct pci_quirk *q;
423 
424 	for (q = &pci_quirks[0]; q->devid; q++) {
425 		if (q->devid == devid && q->type == quirk)
426 			return (1);
427 	}
428 	return (0);
429 }
430 
431 /* Find a device_t by bus/slot/function in domain 0 */
432 
433 device_t
434 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
435 {
436 
437 	return (pci_find_dbsf(0, bus, slot, func));
438 }
439 
440 /* Find a device_t by domain/bus/slot/function */
441 
442 device_t
443 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
444 {
445 	struct pci_devinfo *dinfo = NULL;
446 
447 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
448 		if ((dinfo->cfg.domain == domain) &&
449 		    (dinfo->cfg.bus == bus) &&
450 		    (dinfo->cfg.slot == slot) &&
451 		    (dinfo->cfg.func == func)) {
452 			break;
453 		}
454 	}
455 
456 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
457 }
458 
459 /* Find a device_t by vendor/device ID */
460 
461 device_t
462 pci_find_device(uint16_t vendor, uint16_t device)
463 {
464 	struct pci_devinfo *dinfo;
465 
466 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
467 		if ((dinfo->cfg.vendor == vendor) &&
468 		    (dinfo->cfg.device == device)) {
469 			return (dinfo->cfg.dev);
470 		}
471 	}
472 
473 	return (NULL);
474 }
475 
476 device_t
477 pci_find_class(uint8_t class, uint8_t subclass)
478 {
479 	struct pci_devinfo *dinfo;
480 
481 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
482 		if (dinfo->cfg.baseclass == class &&
483 		    dinfo->cfg.subclass == subclass) {
484 			return (dinfo->cfg.dev);
485 		}
486 	}
487 
488 	return (NULL);
489 }
490 
491 static int
492 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
493 {
494 	va_list ap;
495 	int retval;
496 
497 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
498 	    cfg->func);
499 	va_start(ap, fmt);
500 	retval += vprintf(fmt, ap);
501 	va_end(ap);
502 	return (retval);
503 }
504 
505 /* return base address of memory or port map */
506 
507 static pci_addr_t
508 pci_mapbase(uint64_t mapreg)
509 {
510 
511 	if (PCI_BAR_MEM(mapreg))
512 		return (mapreg & PCIM_BAR_MEM_BASE);
513 	else
514 		return (mapreg & PCIM_BAR_IO_BASE);
515 }
516 
517 /* return map type of memory or port map */
518 
519 static const char *
520 pci_maptype(uint64_t mapreg)
521 {
522 
523 	if (PCI_BAR_IO(mapreg))
524 		return ("I/O Port");
525 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
526 		return ("Prefetchable Memory");
527 	return ("Memory");
528 }
529 
530 /* return log2 of map size decoded for memory or port map */
531 
532 int
533 pci_mapsize(uint64_t testval)
534 {
535 	int ln2size;
536 
537 	testval = pci_mapbase(testval);
538 	ln2size = 0;
539 	if (testval != 0) {
540 		while ((testval & 1) == 0)
541 		{
542 			ln2size++;
543 			testval >>= 1;
544 		}
545 	}
546 	return (ln2size);
547 }
548 
549 /* return base address of device ROM */
550 
551 static pci_addr_t
552 pci_rombase(uint64_t mapreg)
553 {
554 
555 	return (mapreg & PCIM_BIOS_ADDR_MASK);
556 }
557 
558 /* return log2 of map size decided for device ROM */
559 
560 static int
561 pci_romsize(uint64_t testval)
562 {
563 	int ln2size;
564 
565 	testval = pci_rombase(testval);
566 	ln2size = 0;
567 	if (testval != 0) {
568 		while ((testval & 1) == 0)
569 		{
570 			ln2size++;
571 			testval >>= 1;
572 		}
573 	}
574 	return (ln2size);
575 }
576 
577 /* return log2 of address range supported by map register */
578 
579 static int
580 pci_maprange(uint64_t mapreg)
581 {
582 	int ln2range = 0;
583 
584 	if (PCI_BAR_IO(mapreg))
585 		ln2range = 32;
586 	else
587 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
588 		case PCIM_BAR_MEM_32:
589 			ln2range = 32;
590 			break;
591 		case PCIM_BAR_MEM_1MB:
592 			ln2range = 20;
593 			break;
594 		case PCIM_BAR_MEM_64:
595 			ln2range = 64;
596 			break;
597 		}
598 	return (ln2range);
599 }
600 
601 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
602 
603 static void
604 pci_fixancient(pcicfgregs *cfg)
605 {
606 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
607 		return;
608 
609 	/* PCI to PCI bridges use header type 1 */
610 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
611 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
612 }
613 
614 /* extract header type specific config data */
615 
616 static void
617 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
618 {
619 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
620 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
621 	case PCIM_HDRTYPE_NORMAL:
622 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
623 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
624 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
625 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
626 		cfg->nummaps	    = PCI_MAXMAPS_0;
627 		break;
628 	case PCIM_HDRTYPE_BRIDGE:
629 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
630 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
631 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
632 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
633 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
634 		cfg->nummaps	    = PCI_MAXMAPS_1;
635 		break;
636 	case PCIM_HDRTYPE_CARDBUS:
637 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
638 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
639 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
640 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
641 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
642 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
643 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
644 		cfg->nummaps	    = PCI_MAXMAPS_2;
645 		break;
646 	}
647 #undef REG
648 }
649 
650 /* read configuration header into pcicfgregs structure */
651 struct pci_devinfo *
652 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
653 {
654 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
655 	uint16_t vid, did;
656 
657 	vid = REG(PCIR_VENDOR, 2);
658 	did = REG(PCIR_DEVICE, 2);
659 	if (vid != 0xffff)
660 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
661 
662 	return (NULL);
663 }
664 
665 struct pci_devinfo *
666 pci_alloc_devinfo_method(device_t dev)
667 {
668 
669 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
670 	    M_WAITOK | M_ZERO));
671 }
672 
673 static struct pci_devinfo *
674 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
675     uint16_t vid, uint16_t did)
676 {
677 	struct pci_devinfo *devlist_entry;
678 	pcicfgregs *cfg;
679 
680 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
681 
682 	cfg = &devlist_entry->cfg;
683 
684 	cfg->domain		= d;
685 	cfg->bus		= b;
686 	cfg->slot		= s;
687 	cfg->func		= f;
688 	cfg->vendor		= vid;
689 	cfg->device		= did;
690 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
691 	cfg->statreg		= REG(PCIR_STATUS, 2);
692 	cfg->baseclass		= REG(PCIR_CLASS, 1);
693 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
694 	cfg->progif		= REG(PCIR_PROGIF, 1);
695 	cfg->revid		= REG(PCIR_REVID, 1);
696 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
697 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
698 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
699 	cfg->intpin		= REG(PCIR_INTPIN, 1);
700 	cfg->intline		= REG(PCIR_INTLINE, 1);
701 
702 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
703 	cfg->hdrtype		&= ~PCIM_MFDEV;
704 	STAILQ_INIT(&cfg->maps);
705 
706 	cfg->iov		= NULL;
707 
708 	pci_fixancient(cfg);
709 	pci_hdrtypedata(pcib, b, s, f, cfg);
710 
711 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
712 		pci_read_cap(pcib, cfg);
713 
714 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
715 
716 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
717 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
718 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
719 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
720 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
721 
722 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
723 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
724 	devlist_entry->conf.pc_vendor = cfg->vendor;
725 	devlist_entry->conf.pc_device = cfg->device;
726 
727 	devlist_entry->conf.pc_class = cfg->baseclass;
728 	devlist_entry->conf.pc_subclass = cfg->subclass;
729 	devlist_entry->conf.pc_progif = cfg->progif;
730 	devlist_entry->conf.pc_revid = cfg->revid;
731 
732 	pci_numdevs++;
733 	pci_generation++;
734 
735 	return (devlist_entry);
736 }
737 #undef REG
738 
739 static void
740 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
741 {
742 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
743     cfg->ea.ea_location + (n), w)
744 	int num_ent;
745 	int ptr;
746 	int a, b;
747 	uint32_t val;
748 	int ent_size;
749 	uint32_t dw[4];
750 	uint64_t base, max_offset;
751 	struct pci_ea_entry *eae;
752 
753 	if (cfg->ea.ea_location == 0)
754 		return;
755 
756 	STAILQ_INIT(&cfg->ea.ea_entries);
757 
758 	/* Determine the number of entries */
759 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
760 	num_ent &= PCIM_EA_NUM_ENT_MASK;
761 
762 	/* Find the first entry to care of */
763 	ptr = PCIR_EA_FIRST_ENT;
764 
765 	/* Skip DWORD 2 for type 1 functions */
766 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
767 		ptr += 4;
768 
769 	for (a = 0; a < num_ent; a++) {
770 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
771 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
772 
773 		/* Read a number of dwords in the entry */
774 		val = REG(ptr, 4);
775 		ptr += 4;
776 		ent_size = (val & PCIM_EA_ES);
777 
778 		for (b = 0; b < ent_size; b++) {
779 			dw[b] = REG(ptr, 4);
780 			ptr += 4;
781 		}
782 
783 		eae->eae_flags = val;
784 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
785 
786 		base = dw[0] & PCIM_EA_FIELD_MASK;
787 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
788 		b = 2;
789 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
790 			base |= (uint64_t)dw[b] << 32UL;
791 			b++;
792 		}
793 		if (((dw[1] & PCIM_EA_IS_64) != 0)
794 		    && (b < ent_size)) {
795 			max_offset |= (uint64_t)dw[b] << 32UL;
796 			b++;
797 		}
798 
799 		eae->eae_base = base;
800 		eae->eae_max_offset = max_offset;
801 
802 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
803 
804 		if (bootverbose) {
805 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
806 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
807 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
808 		}
809 	}
810 }
811 #undef REG
812 
813 static void
814 pci_read_cap(device_t pcib, pcicfgregs *cfg)
815 {
816 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
817 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
818 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
819 	uint64_t addr;
820 #endif
821 	uint32_t val;
822 	int	ptr, nextptr, ptrptr;
823 
824 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
825 	case PCIM_HDRTYPE_NORMAL:
826 	case PCIM_HDRTYPE_BRIDGE:
827 		ptrptr = PCIR_CAP_PTR;
828 		break;
829 	case PCIM_HDRTYPE_CARDBUS:
830 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
831 		break;
832 	default:
833 		return;		/* no extended capabilities support */
834 	}
835 	nextptr = REG(ptrptr, 1);	/* sanity check? */
836 
837 	/*
838 	 * Read capability entries.
839 	 */
840 	while (nextptr != 0) {
841 		/* Sanity check */
842 		if (nextptr > 255) {
843 			printf("illegal PCI extended capability offset %d\n",
844 			    nextptr);
845 			return;
846 		}
847 		/* Find the next entry */
848 		ptr = nextptr;
849 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
850 
851 		/* Process this entry */
852 		switch (REG(ptr + PCICAP_ID, 1)) {
853 		case PCIY_PMG:		/* PCI power management */
854 			if (cfg->pp.pp_cap == 0) {
855 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
856 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
857 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
858 				if ((nextptr - ptr) > PCIR_POWER_DATA)
859 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
860 			}
861 			break;
862 		case PCIY_HT:		/* HyperTransport */
863 			/* Determine HT-specific capability type. */
864 			val = REG(ptr + PCIR_HT_COMMAND, 2);
865 
866 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
867 				cfg->ht.ht_slave = ptr;
868 
869 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
870 			switch (val & PCIM_HTCMD_CAP_MASK) {
871 			case PCIM_HTCAP_MSI_MAPPING:
872 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
873 					/* Sanity check the mapping window. */
874 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
875 					    4);
876 					addr <<= 32;
877 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
878 					    4);
879 					if (addr != MSI_INTEL_ADDR_BASE)
880 						device_printf(pcib,
881 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
882 						    cfg->domain, cfg->bus,
883 						    cfg->slot, cfg->func,
884 						    (long long)addr);
885 				} else
886 					addr = MSI_INTEL_ADDR_BASE;
887 
888 				cfg->ht.ht_msimap = ptr;
889 				cfg->ht.ht_msictrl = val;
890 				cfg->ht.ht_msiaddr = addr;
891 				break;
892 			}
893 #endif
894 			break;
895 		case PCIY_MSI:		/* PCI MSI */
896 			cfg->msi.msi_location = ptr;
897 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
898 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
899 						     PCIM_MSICTRL_MMC_MASK)>>1);
900 			break;
901 		case PCIY_MSIX:		/* PCI MSI-X */
902 			cfg->msix.msix_location = ptr;
903 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
904 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
905 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
906 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
907 			cfg->msix.msix_table_bar = PCIR_BAR(val &
908 			    PCIM_MSIX_BIR_MASK);
909 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
910 			val = REG(ptr + PCIR_MSIX_PBA, 4);
911 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
912 			    PCIM_MSIX_BIR_MASK);
913 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
914 			break;
915 		case PCIY_VPD:		/* PCI Vital Product Data */
916 			cfg->vpd.vpd_reg = ptr;
917 			break;
918 		case PCIY_SUBVENDOR:
919 			/* Should always be true. */
920 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
921 			    PCIM_HDRTYPE_BRIDGE) {
922 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
923 				cfg->subvendor = val & 0xffff;
924 				cfg->subdevice = val >> 16;
925 			}
926 			break;
927 		case PCIY_PCIX:		/* PCI-X */
928 			/*
929 			 * Assume we have a PCI-X chipset if we have
930 			 * at least one PCI-PCI bridge with a PCI-X
931 			 * capability.  Note that some systems with
932 			 * PCI-express or HT chipsets might match on
933 			 * this check as well.
934 			 */
935 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
936 			    PCIM_HDRTYPE_BRIDGE)
937 				pcix_chipset = 1;
938 			cfg->pcix.pcix_location = ptr;
939 			break;
940 		case PCIY_EXPRESS:	/* PCI-express */
941 			/*
942 			 * Assume we have a PCI-express chipset if we have
943 			 * at least one PCI-express device.
944 			 */
945 			pcie_chipset = 1;
946 			cfg->pcie.pcie_location = ptr;
947 			val = REG(ptr + PCIER_FLAGS, 2);
948 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
949 			break;
950 		case PCIY_EA:		/* Enhanced Allocation */
951 			cfg->ea.ea_location = ptr;
952 			pci_ea_fill_info(pcib, cfg);
953 			break;
954 		default:
955 			break;
956 		}
957 	}
958 
959 #if defined(__powerpc__)
960 	/*
961 	 * Enable the MSI mapping window for all HyperTransport
962 	 * slaves.  PCI-PCI bridges have their windows enabled via
963 	 * PCIB_MAP_MSI().
964 	 */
965 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
966 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
967 		device_printf(pcib,
968 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
969 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
970 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
971 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
972 		     2);
973 	}
974 #endif
975 /* REG and WREG use carry through to next functions */
976 }
977 
978 /*
979  * PCI Vital Product Data
980  */
981 
982 #define	PCI_VPD_TIMEOUT		1000000
983 
984 static int
985 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
986 {
987 	int count = PCI_VPD_TIMEOUT;
988 
989 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
990 
991 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
992 
993 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
994 		if (--count < 0)
995 			return (ENXIO);
996 		DELAY(1);	/* limit looping */
997 	}
998 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
999 
1000 	return (0);
1001 }
1002 
1003 #if 0
1004 static int
1005 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1006 {
1007 	int count = PCI_VPD_TIMEOUT;
1008 
1009 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1010 
1011 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1012 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1013 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1014 		if (--count < 0)
1015 			return (ENXIO);
1016 		DELAY(1);	/* limit looping */
1017 	}
1018 
1019 	return (0);
1020 }
1021 #endif
1022 
1023 #undef PCI_VPD_TIMEOUT
1024 
1025 struct vpd_readstate {
1026 	device_t	pcib;
1027 	pcicfgregs	*cfg;
1028 	uint32_t	val;
1029 	int		bytesinval;
1030 	int		off;
1031 	uint8_t		cksum;
1032 };
1033 
1034 static int
1035 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1036 {
1037 	uint32_t reg;
1038 	uint8_t byte;
1039 
1040 	if (vrs->bytesinval == 0) {
1041 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1042 			return (ENXIO);
1043 		vrs->val = le32toh(reg);
1044 		vrs->off += 4;
1045 		byte = vrs->val & 0xff;
1046 		vrs->bytesinval = 3;
1047 	} else {
1048 		vrs->val = vrs->val >> 8;
1049 		byte = vrs->val & 0xff;
1050 		vrs->bytesinval--;
1051 	}
1052 
1053 	vrs->cksum += byte;
1054 	*data = byte;
1055 	return (0);
1056 }
1057 
1058 static void
1059 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1060 {
1061 	struct vpd_readstate vrs;
1062 	int state;
1063 	int name;
1064 	int remain;
1065 	int i;
1066 	int alloc, off;		/* alloc/off for RO/W arrays */
1067 	int cksumvalid;
1068 	int dflen;
1069 	uint8_t byte;
1070 	uint8_t byte2;
1071 
1072 	/* init vpd reader */
1073 	vrs.bytesinval = 0;
1074 	vrs.off = 0;
1075 	vrs.pcib = pcib;
1076 	vrs.cfg = cfg;
1077 	vrs.cksum = 0;
1078 
1079 	state = 0;
1080 	name = remain = i = 0;	/* shut up stupid gcc */
1081 	alloc = off = 0;	/* shut up stupid gcc */
1082 	dflen = 0;		/* shut up stupid gcc */
1083 	cksumvalid = -1;
1084 	while (state >= 0) {
1085 		if (vpd_nextbyte(&vrs, &byte)) {
1086 			state = -2;
1087 			break;
1088 		}
1089 #if 0
1090 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1091 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1092 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1093 #endif
1094 		switch (state) {
1095 		case 0:		/* item name */
1096 			if (byte & 0x80) {
1097 				if (vpd_nextbyte(&vrs, &byte2)) {
1098 					state = -2;
1099 					break;
1100 				}
1101 				remain = byte2;
1102 				if (vpd_nextbyte(&vrs, &byte2)) {
1103 					state = -2;
1104 					break;
1105 				}
1106 				remain |= byte2 << 8;
1107 				name = byte & 0x7f;
1108 			} else {
1109 				remain = byte & 0x7;
1110 				name = (byte >> 3) & 0xf;
1111 			}
1112 			if (vrs.off + remain - vrs.bytesinval > 0x8000) {
1113 				pci_printf(cfg,
1114 				    "VPD data overflow, remain %#x\n", remain);
1115 				state = -1;
1116 				break;
1117 			}
1118 			switch (name) {
1119 			case 0x2:	/* String */
1120 				cfg->vpd.vpd_ident = malloc(remain + 1,
1121 				    M_DEVBUF, M_WAITOK);
1122 				i = 0;
1123 				state = 1;
1124 				break;
1125 			case 0xf:	/* End */
1126 				state = -1;
1127 				break;
1128 			case 0x10:	/* VPD-R */
1129 				alloc = 8;
1130 				off = 0;
1131 				cfg->vpd.vpd_ros = malloc(alloc *
1132 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1133 				    M_WAITOK | M_ZERO);
1134 				state = 2;
1135 				break;
1136 			case 0x11:	/* VPD-W */
1137 				alloc = 8;
1138 				off = 0;
1139 				cfg->vpd.vpd_w = malloc(alloc *
1140 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1141 				    M_WAITOK | M_ZERO);
1142 				state = 5;
1143 				break;
1144 			default:	/* Invalid data, abort */
1145 				state = -1;
1146 				break;
1147 			}
1148 			break;
1149 
1150 		case 1:	/* Identifier String */
1151 			cfg->vpd.vpd_ident[i++] = byte;
1152 			remain--;
1153 			if (remain == 0)  {
1154 				cfg->vpd.vpd_ident[i] = '\0';
1155 				state = 0;
1156 			}
1157 			break;
1158 
1159 		case 2:	/* VPD-R Keyword Header */
1160 			if (off == alloc) {
1161 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1162 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1163 				    M_DEVBUF, M_WAITOK | M_ZERO);
1164 			}
1165 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1166 			if (vpd_nextbyte(&vrs, &byte2)) {
1167 				state = -2;
1168 				break;
1169 			}
1170 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1171 			if (vpd_nextbyte(&vrs, &byte2)) {
1172 				state = -2;
1173 				break;
1174 			}
1175 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1176 			if (dflen == 0 &&
1177 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1178 			    2) == 0) {
1179 				/*
1180 				 * if this happens, we can't trust the rest
1181 				 * of the VPD.
1182 				 */
1183 				pci_printf(cfg, "bad keyword length: %d\n",
1184 				    dflen);
1185 				cksumvalid = 0;
1186 				state = -1;
1187 				break;
1188 			} else if (dflen == 0) {
1189 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1190 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1191 				    M_DEVBUF, M_WAITOK);
1192 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1193 			} else
1194 				cfg->vpd.vpd_ros[off].value = malloc(
1195 				    (dflen + 1) *
1196 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1197 				    M_DEVBUF, M_WAITOK);
1198 			remain -= 3;
1199 			i = 0;
1200 			/* keep in sync w/ state 3's transistions */
1201 			if (dflen == 0 && remain == 0)
1202 				state = 0;
1203 			else if (dflen == 0)
1204 				state = 2;
1205 			else
1206 				state = 3;
1207 			break;
1208 
1209 		case 3:	/* VPD-R Keyword Value */
1210 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1211 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1212 			    "RV", 2) == 0 && cksumvalid == -1) {
1213 				if (vrs.cksum == 0)
1214 					cksumvalid = 1;
1215 				else {
1216 					if (bootverbose)
1217 						pci_printf(cfg,
1218 					    "bad VPD cksum, remain %hhu\n",
1219 						    vrs.cksum);
1220 					cksumvalid = 0;
1221 					state = -1;
1222 					break;
1223 				}
1224 			}
1225 			dflen--;
1226 			remain--;
1227 			/* keep in sync w/ state 2's transistions */
1228 			if (dflen == 0)
1229 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1230 			if (dflen == 0 && remain == 0) {
1231 				cfg->vpd.vpd_rocnt = off;
1232 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1233 				    off * sizeof(*cfg->vpd.vpd_ros),
1234 				    M_DEVBUF, M_WAITOK | M_ZERO);
1235 				state = 0;
1236 			} else if (dflen == 0)
1237 				state = 2;
1238 			break;
1239 
1240 		case 4:
1241 			remain--;
1242 			if (remain == 0)
1243 				state = 0;
1244 			break;
1245 
1246 		case 5:	/* VPD-W Keyword Header */
1247 			if (off == alloc) {
1248 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1249 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1250 				    M_DEVBUF, M_WAITOK | M_ZERO);
1251 			}
1252 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1253 			if (vpd_nextbyte(&vrs, &byte2)) {
1254 				state = -2;
1255 				break;
1256 			}
1257 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1258 			if (vpd_nextbyte(&vrs, &byte2)) {
1259 				state = -2;
1260 				break;
1261 			}
1262 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1263 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1264 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1265 			    sizeof(*cfg->vpd.vpd_w[off].value),
1266 			    M_DEVBUF, M_WAITOK);
1267 			remain -= 3;
1268 			i = 0;
1269 			/* keep in sync w/ state 6's transistions */
1270 			if (dflen == 0 && remain == 0)
1271 				state = 0;
1272 			else if (dflen == 0)
1273 				state = 5;
1274 			else
1275 				state = 6;
1276 			break;
1277 
1278 		case 6:	/* VPD-W Keyword Value */
1279 			cfg->vpd.vpd_w[off].value[i++] = byte;
1280 			dflen--;
1281 			remain--;
1282 			/* keep in sync w/ state 5's transistions */
1283 			if (dflen == 0)
1284 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1285 			if (dflen == 0 && remain == 0) {
1286 				cfg->vpd.vpd_wcnt = off;
1287 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1288 				    off * sizeof(*cfg->vpd.vpd_w),
1289 				    M_DEVBUF, M_WAITOK | M_ZERO);
1290 				state = 0;
1291 			} else if (dflen == 0)
1292 				state = 5;
1293 			break;
1294 
1295 		default:
1296 			pci_printf(cfg, "invalid state: %d\n", state);
1297 			state = -1;
1298 			break;
1299 		}
1300 	}
1301 
1302 	if (cksumvalid == 0 || state < -1) {
1303 		/* read-only data bad, clean up */
1304 		if (cfg->vpd.vpd_ros != NULL) {
1305 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1306 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1307 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1308 			cfg->vpd.vpd_ros = NULL;
1309 		}
1310 	}
1311 	if (state < -1) {
1312 		/* I/O error, clean up */
1313 		pci_printf(cfg, "failed to read VPD data.\n");
1314 		if (cfg->vpd.vpd_ident != NULL) {
1315 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1316 			cfg->vpd.vpd_ident = NULL;
1317 		}
1318 		if (cfg->vpd.vpd_w != NULL) {
1319 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1320 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1321 			free(cfg->vpd.vpd_w, M_DEVBUF);
1322 			cfg->vpd.vpd_w = NULL;
1323 		}
1324 	}
1325 	cfg->vpd.vpd_cached = 1;
1326 #undef REG
1327 #undef WREG
1328 }
1329 
1330 int
1331 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1332 {
1333 	struct pci_devinfo *dinfo = device_get_ivars(child);
1334 	pcicfgregs *cfg = &dinfo->cfg;
1335 
1336 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1337 		pci_read_vpd(device_get_parent(dev), cfg);
1338 
1339 	*identptr = cfg->vpd.vpd_ident;
1340 
1341 	if (*identptr == NULL)
1342 		return (ENXIO);
1343 
1344 	return (0);
1345 }
1346 
1347 int
1348 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1349 	const char **vptr)
1350 {
1351 	struct pci_devinfo *dinfo = device_get_ivars(child);
1352 	pcicfgregs *cfg = &dinfo->cfg;
1353 	int i;
1354 
1355 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1356 		pci_read_vpd(device_get_parent(dev), cfg);
1357 
1358 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1359 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1360 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1361 			*vptr = cfg->vpd.vpd_ros[i].value;
1362 			return (0);
1363 		}
1364 
1365 	*vptr = NULL;
1366 	return (ENXIO);
1367 }
1368 
1369 struct pcicfg_vpd *
1370 pci_fetch_vpd_list(device_t dev)
1371 {
1372 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1373 	pcicfgregs *cfg = &dinfo->cfg;
1374 
1375 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1376 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1377 	return (&cfg->vpd);
1378 }
1379 
1380 /*
1381  * Find the requested HyperTransport capability and return the offset
1382  * in configuration space via the pointer provided.  The function
1383  * returns 0 on success and an error code otherwise.
1384  */
1385 int
1386 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1387 {
1388 	int ptr, error;
1389 	uint16_t val;
1390 
1391 	error = pci_find_cap(child, PCIY_HT, &ptr);
1392 	if (error)
1393 		return (error);
1394 
1395 	/*
1396 	 * Traverse the capabilities list checking each HT capability
1397 	 * to see if it matches the requested HT capability.
1398 	 */
1399 	for (;;) {
1400 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1401 		if (capability == PCIM_HTCAP_SLAVE ||
1402 		    capability == PCIM_HTCAP_HOST)
1403 			val &= 0xe000;
1404 		else
1405 			val &= PCIM_HTCMD_CAP_MASK;
1406 		if (val == capability) {
1407 			if (capreg != NULL)
1408 				*capreg = ptr;
1409 			return (0);
1410 		}
1411 
1412 		/* Skip to the next HT capability. */
1413 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1414 			break;
1415 	}
1416 
1417 	return (ENOENT);
1418 }
1419 
1420 /*
1421  * Find the next requested HyperTransport capability after start and return
1422  * the offset in configuration space via the pointer provided.  The function
1423  * returns 0 on success and an error code otherwise.
1424  */
1425 int
1426 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1427     int start, int *capreg)
1428 {
1429 	int ptr;
1430 	uint16_t val;
1431 
1432 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1433 	    ("start capability is not HyperTransport capability"));
1434 	ptr = start;
1435 
1436 	/*
1437 	 * Traverse the capabilities list checking each HT capability
1438 	 * to see if it matches the requested HT capability.
1439 	 */
1440 	for (;;) {
1441 		/* Skip to the next HT capability. */
1442 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1443 			break;
1444 
1445 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1446 		if (capability == PCIM_HTCAP_SLAVE ||
1447 		    capability == PCIM_HTCAP_HOST)
1448 			val &= 0xe000;
1449 		else
1450 			val &= PCIM_HTCMD_CAP_MASK;
1451 		if (val == capability) {
1452 			if (capreg != NULL)
1453 				*capreg = ptr;
1454 			return (0);
1455 		}
1456 	}
1457 
1458 	return (ENOENT);
1459 }
1460 
1461 /*
1462  * Find the requested capability and return the offset in
1463  * configuration space via the pointer provided.  The function returns
1464  * 0 on success and an error code otherwise.
1465  */
1466 int
1467 pci_find_cap_method(device_t dev, device_t child, int capability,
1468     int *capreg)
1469 {
1470 	struct pci_devinfo *dinfo = device_get_ivars(child);
1471 	pcicfgregs *cfg = &dinfo->cfg;
1472 	uint32_t status;
1473 	uint8_t ptr;
1474 
1475 	/*
1476 	 * Check the CAP_LIST bit of the PCI status register first.
1477 	 */
1478 	status = pci_read_config(child, PCIR_STATUS, 2);
1479 	if (!(status & PCIM_STATUS_CAPPRESENT))
1480 		return (ENXIO);
1481 
1482 	/*
1483 	 * Determine the start pointer of the capabilities list.
1484 	 */
1485 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1486 	case PCIM_HDRTYPE_NORMAL:
1487 	case PCIM_HDRTYPE_BRIDGE:
1488 		ptr = PCIR_CAP_PTR;
1489 		break;
1490 	case PCIM_HDRTYPE_CARDBUS:
1491 		ptr = PCIR_CAP_PTR_2;
1492 		break;
1493 	default:
1494 		/* XXX: panic? */
1495 		return (ENXIO);		/* no extended capabilities support */
1496 	}
1497 	ptr = pci_read_config(child, ptr, 1);
1498 
1499 	/*
1500 	 * Traverse the capabilities list.
1501 	 */
1502 	while (ptr != 0) {
1503 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1504 			if (capreg != NULL)
1505 				*capreg = ptr;
1506 			return (0);
1507 		}
1508 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1509 	}
1510 
1511 	return (ENOENT);
1512 }
1513 
1514 /*
1515  * Find the next requested capability after start and return the offset in
1516  * configuration space via the pointer provided.  The function returns
1517  * 0 on success and an error code otherwise.
1518  */
1519 int
1520 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1521     int start, int *capreg)
1522 {
1523 	uint8_t ptr;
1524 
1525 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1526 	    ("start capability is not expected capability"));
1527 
1528 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1529 	while (ptr != 0) {
1530 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1531 			if (capreg != NULL)
1532 				*capreg = ptr;
1533 			return (0);
1534 		}
1535 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1536 	}
1537 
1538 	return (ENOENT);
1539 }
1540 
1541 /*
1542  * Find the requested extended capability and return the offset in
1543  * configuration space via the pointer provided.  The function returns
1544  * 0 on success and an error code otherwise.
1545  */
1546 int
1547 pci_find_extcap_method(device_t dev, device_t child, int capability,
1548     int *capreg)
1549 {
1550 	struct pci_devinfo *dinfo = device_get_ivars(child);
1551 	pcicfgregs *cfg = &dinfo->cfg;
1552 	uint32_t ecap;
1553 	uint16_t ptr;
1554 
1555 	/* Only supported for PCI-express devices. */
1556 	if (cfg->pcie.pcie_location == 0)
1557 		return (ENXIO);
1558 
1559 	ptr = PCIR_EXTCAP;
1560 	ecap = pci_read_config(child, ptr, 4);
1561 	if (ecap == 0xffffffff || ecap == 0)
1562 		return (ENOENT);
1563 	for (;;) {
1564 		if (PCI_EXTCAP_ID(ecap) == capability) {
1565 			if (capreg != NULL)
1566 				*capreg = ptr;
1567 			return (0);
1568 		}
1569 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1570 		if (ptr == 0)
1571 			break;
1572 		ecap = pci_read_config(child, ptr, 4);
1573 	}
1574 
1575 	return (ENOENT);
1576 }
1577 
1578 /*
1579  * Find the next requested extended capability after start and return the
1580  * offset in configuration space via the pointer provided.  The function
1581  * returns 0 on success and an error code otherwise.
1582  */
1583 int
1584 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1585     int start, int *capreg)
1586 {
1587 	struct pci_devinfo *dinfo = device_get_ivars(child);
1588 	pcicfgregs *cfg = &dinfo->cfg;
1589 	uint32_t ecap;
1590 	uint16_t ptr;
1591 
1592 	/* Only supported for PCI-express devices. */
1593 	if (cfg->pcie.pcie_location == 0)
1594 		return (ENXIO);
1595 
1596 	ecap = pci_read_config(child, start, 4);
1597 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1598 	    ("start extended capability is not expected capability"));
1599 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1600 	while (ptr != 0) {
1601 		ecap = pci_read_config(child, ptr, 4);
1602 		if (PCI_EXTCAP_ID(ecap) == capability) {
1603 			if (capreg != NULL)
1604 				*capreg = ptr;
1605 			return (0);
1606 		}
1607 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1608 	}
1609 
1610 	return (ENOENT);
1611 }
1612 
1613 /*
1614  * Support for MSI-X message interrupts.
1615  */
1616 static void
1617 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1618 {
1619 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1620 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1621 	uint32_t offset;
1622 
1623 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1624 	offset = msix->msix_table_offset + index * 16;
1625 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1626 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1627 	bus_write_4(msix->msix_table_res, offset + 8, data);
1628 }
1629 
1630 void
1631 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1632     uint64_t address, uint32_t data)
1633 {
1634 
1635 	if (pci_msix_rewrite_table) {
1636 		struct pci_devinfo *dinfo = device_get_ivars(child);
1637 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1638 
1639 		/*
1640 		 * Some VM hosts require MSIX to be disabled in the
1641 		 * control register before updating the MSIX table
1642 		 * entries are allowed. It is not enough to only
1643 		 * disable MSIX while updating a single entry. MSIX
1644 		 * must be disabled while updating all entries in the
1645 		 * table.
1646 		 */
1647 		pci_write_config(child,
1648 		    msix->msix_location + PCIR_MSIX_CTRL,
1649 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1650 		pci_resume_msix(child);
1651 	} else
1652 		pci_write_msix_entry(child, index, address, data);
1653 
1654 	/* Enable MSI -> HT mapping. */
1655 	pci_ht_map_msi(child, address);
1656 }
1657 
1658 void
1659 pci_mask_msix(device_t dev, u_int index)
1660 {
1661 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1662 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1663 	uint32_t offset, val;
1664 
1665 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1666 	offset = msix->msix_table_offset + index * 16 + 12;
1667 	val = bus_read_4(msix->msix_table_res, offset);
1668 	val |= PCIM_MSIX_VCTRL_MASK;
1669 
1670 	/*
1671 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1672 	 * register, so always write the new value.
1673 	 */
1674 	bus_write_4(msix->msix_table_res, offset, val);
1675 }
1676 
1677 void
1678 pci_unmask_msix(device_t dev, u_int index)
1679 {
1680 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1681 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1682 	uint32_t offset, val;
1683 
1684 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1685 	offset = msix->msix_table_offset + index * 16 + 12;
1686 	val = bus_read_4(msix->msix_table_res, offset);
1687 	val &= ~PCIM_MSIX_VCTRL_MASK;
1688 
1689 	/*
1690 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1691 	 * register, so always write the new value.
1692 	 */
1693 	bus_write_4(msix->msix_table_res, offset, val);
1694 }
1695 
1696 int
1697 pci_pending_msix(device_t dev, u_int index)
1698 {
1699 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1700 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1701 	uint32_t offset, bit;
1702 
1703 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1704 	offset = msix->msix_pba_offset + (index / 32) * 4;
1705 	bit = 1 << index % 32;
1706 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1707 }
1708 
1709 /*
1710  * Restore MSI-X registers and table during resume.  If MSI-X is
1711  * enabled then walk the virtual table to restore the actual MSI-X
1712  * table.
1713  */
1714 static void
1715 pci_resume_msix(device_t dev)
1716 {
1717 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1718 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1719 	struct msix_table_entry *mte;
1720 	struct msix_vector *mv;
1721 	int i;
1722 
1723 	if (msix->msix_alloc > 0) {
1724 		/* First, mask all vectors. */
1725 		for (i = 0; i < msix->msix_msgnum; i++)
1726 			pci_mask_msix(dev, i);
1727 
1728 		/* Second, program any messages with at least one handler. */
1729 		for (i = 0; i < msix->msix_table_len; i++) {
1730 			mte = &msix->msix_table[i];
1731 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1732 				continue;
1733 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1734 			pci_write_msix_entry(dev, i, mv->mv_address,
1735 			    mv->mv_data);
1736 			pci_unmask_msix(dev, i);
1737 		}
1738 	}
1739 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1740 	    msix->msix_ctrl, 2);
1741 }
1742 
1743 /*
1744  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1745  * returned in *count.  After this function returns, each message will be
1746  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1747  */
1748 int
1749 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1750 {
1751 	struct pci_devinfo *dinfo = device_get_ivars(child);
1752 	pcicfgregs *cfg = &dinfo->cfg;
1753 	struct resource_list_entry *rle;
1754 	int actual, error, i, irq, max;
1755 
1756 	/* Don't let count == 0 get us into trouble. */
1757 	if (*count == 0)
1758 		return (EINVAL);
1759 
1760 	/* If rid 0 is allocated, then fail. */
1761 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1762 	if (rle != NULL && rle->res != NULL)
1763 		return (ENXIO);
1764 
1765 	/* Already have allocated messages? */
1766 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1767 		return (ENXIO);
1768 
1769 	/* If MSI-X is blacklisted for this system, fail. */
1770 	if (pci_msix_blacklisted())
1771 		return (ENXIO);
1772 
1773 	/* MSI-X capability present? */
1774 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1775 		return (ENODEV);
1776 
1777 	/* Make sure the appropriate BARs are mapped. */
1778 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1779 	    cfg->msix.msix_table_bar);
1780 	if (rle == NULL || rle->res == NULL ||
1781 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1782 		return (ENXIO);
1783 	cfg->msix.msix_table_res = rle->res;
1784 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1785 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1786 		    cfg->msix.msix_pba_bar);
1787 		if (rle == NULL || rle->res == NULL ||
1788 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1789 			return (ENXIO);
1790 	}
1791 	cfg->msix.msix_pba_res = rle->res;
1792 
1793 	if (bootverbose)
1794 		device_printf(child,
1795 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1796 		    *count, cfg->msix.msix_msgnum);
1797 	max = min(*count, cfg->msix.msix_msgnum);
1798 	for (i = 0; i < max; i++) {
1799 		/* Allocate a message. */
1800 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1801 		if (error) {
1802 			if (i == 0)
1803 				return (error);
1804 			break;
1805 		}
1806 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1807 		    irq, 1);
1808 	}
1809 	actual = i;
1810 
1811 	if (bootverbose) {
1812 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1813 		if (actual == 1)
1814 			device_printf(child, "using IRQ %ju for MSI-X\n",
1815 			    rle->start);
1816 		else {
1817 			int run;
1818 
1819 			/*
1820 			 * Be fancy and try to print contiguous runs of
1821 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1822 			 * 'run' is true if we are in a range.
1823 			 */
1824 			device_printf(child, "using IRQs %ju", rle->start);
1825 			irq = rle->start;
1826 			run = 0;
1827 			for (i = 1; i < actual; i++) {
1828 				rle = resource_list_find(&dinfo->resources,
1829 				    SYS_RES_IRQ, i + 1);
1830 
1831 				/* Still in a run? */
1832 				if (rle->start == irq + 1) {
1833 					run = 1;
1834 					irq++;
1835 					continue;
1836 				}
1837 
1838 				/* Finish previous range. */
1839 				if (run) {
1840 					printf("-%d", irq);
1841 					run = 0;
1842 				}
1843 
1844 				/* Start new range. */
1845 				printf(",%ju", rle->start);
1846 				irq = rle->start;
1847 			}
1848 
1849 			/* Unfinished range? */
1850 			if (run)
1851 				printf("-%d", irq);
1852 			printf(" for MSI-X\n");
1853 		}
1854 	}
1855 
1856 	/* Mask all vectors. */
1857 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1858 		pci_mask_msix(child, i);
1859 
1860 	/* Allocate and initialize vector data and virtual table. */
1861 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1862 	    M_DEVBUF, M_WAITOK | M_ZERO);
1863 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1864 	    M_DEVBUF, M_WAITOK | M_ZERO);
1865 	for (i = 0; i < actual; i++) {
1866 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1867 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1868 		cfg->msix.msix_table[i].mte_vector = i + 1;
1869 	}
1870 
1871 	/* Update control register to enable MSI-X. */
1872 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1873 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1874 	    cfg->msix.msix_ctrl, 2);
1875 
1876 	/* Update counts of alloc'd messages. */
1877 	cfg->msix.msix_alloc = actual;
1878 	cfg->msix.msix_table_len = actual;
1879 	*count = actual;
1880 	return (0);
1881 }
1882 
1883 /*
1884  * By default, pci_alloc_msix() will assign the allocated IRQ
1885  * resources consecutively to the first N messages in the MSI-X table.
1886  * However, device drivers may want to use different layouts if they
1887  * either receive fewer messages than they asked for, or they wish to
1888  * populate the MSI-X table sparsely.  This method allows the driver
1889  * to specify what layout it wants.  It must be called after a
1890  * successful pci_alloc_msix() but before any of the associated
1891  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1892  *
1893  * The 'vectors' array contains 'count' message vectors.  The array
1894  * maps directly to the MSI-X table in that index 0 in the array
1895  * specifies the vector for the first message in the MSI-X table, etc.
1896  * The vector value in each array index can either be 0 to indicate
1897  * that no vector should be assigned to a message slot, or it can be a
1898  * number from 1 to N (where N is the count returned from a
1899  * succcessful call to pci_alloc_msix()) to indicate which message
1900  * vector (IRQ) to be used for the corresponding message.
1901  *
1902  * On successful return, each message with a non-zero vector will have
1903  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1904  * 1.  Additionally, if any of the IRQs allocated via the previous
1905  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1906  * will be freed back to the system automatically.
1907  *
1908  * For example, suppose a driver has a MSI-X table with 6 messages and
1909  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1910  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1911  * C.  After the call to pci_alloc_msix(), the device will be setup to
1912  * have an MSI-X table of ABC--- (where - means no vector assigned).
1913  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1914  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1915  * be freed back to the system.  This device will also have valid
1916  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1917  *
1918  * In any case, the SYS_RES_IRQ rid X will always map to the message
1919  * at MSI-X table index X - 1 and will only be valid if a vector is
1920  * assigned to that table entry.
1921  */
1922 int
1923 pci_remap_msix_method(device_t dev, device_t child, int count,
1924     const u_int *vectors)
1925 {
1926 	struct pci_devinfo *dinfo = device_get_ivars(child);
1927 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1928 	struct resource_list_entry *rle;
1929 	int i, irq, j, *used;
1930 
1931 	/*
1932 	 * Have to have at least one message in the table but the
1933 	 * table can't be bigger than the actual MSI-X table in the
1934 	 * device.
1935 	 */
1936 	if (count == 0 || count > msix->msix_msgnum)
1937 		return (EINVAL);
1938 
1939 	/* Sanity check the vectors. */
1940 	for (i = 0; i < count; i++)
1941 		if (vectors[i] > msix->msix_alloc)
1942 			return (EINVAL);
1943 
1944 	/*
1945 	 * Make sure there aren't any holes in the vectors to be used.
1946 	 * It's a big pain to support it, and it doesn't really make
1947 	 * sense anyway.  Also, at least one vector must be used.
1948 	 */
1949 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1950 	    M_ZERO);
1951 	for (i = 0; i < count; i++)
1952 		if (vectors[i] != 0)
1953 			used[vectors[i] - 1] = 1;
1954 	for (i = 0; i < msix->msix_alloc - 1; i++)
1955 		if (used[i] == 0 && used[i + 1] == 1) {
1956 			free(used, M_DEVBUF);
1957 			return (EINVAL);
1958 		}
1959 	if (used[0] != 1) {
1960 		free(used, M_DEVBUF);
1961 		return (EINVAL);
1962 	}
1963 
1964 	/* Make sure none of the resources are allocated. */
1965 	for (i = 0; i < msix->msix_table_len; i++) {
1966 		if (msix->msix_table[i].mte_vector == 0)
1967 			continue;
1968 		if (msix->msix_table[i].mte_handlers > 0) {
1969 			free(used, M_DEVBUF);
1970 			return (EBUSY);
1971 		}
1972 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1973 		KASSERT(rle != NULL, ("missing resource"));
1974 		if (rle->res != NULL) {
1975 			free(used, M_DEVBUF);
1976 			return (EBUSY);
1977 		}
1978 	}
1979 
1980 	/* Free the existing resource list entries. */
1981 	for (i = 0; i < msix->msix_table_len; i++) {
1982 		if (msix->msix_table[i].mte_vector == 0)
1983 			continue;
1984 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1985 	}
1986 
1987 	/*
1988 	 * Build the new virtual table keeping track of which vectors are
1989 	 * used.
1990 	 */
1991 	free(msix->msix_table, M_DEVBUF);
1992 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1993 	    M_DEVBUF, M_WAITOK | M_ZERO);
1994 	for (i = 0; i < count; i++)
1995 		msix->msix_table[i].mte_vector = vectors[i];
1996 	msix->msix_table_len = count;
1997 
1998 	/* Free any unused IRQs and resize the vectors array if necessary. */
1999 	j = msix->msix_alloc - 1;
2000 	if (used[j] == 0) {
2001 		struct msix_vector *vec;
2002 
2003 		while (used[j] == 0) {
2004 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2005 			    msix->msix_vectors[j].mv_irq);
2006 			j--;
2007 		}
2008 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2009 		    M_WAITOK);
2010 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2011 		    (j + 1));
2012 		free(msix->msix_vectors, M_DEVBUF);
2013 		msix->msix_vectors = vec;
2014 		msix->msix_alloc = j + 1;
2015 	}
2016 	free(used, M_DEVBUF);
2017 
2018 	/* Map the IRQs onto the rids. */
2019 	for (i = 0; i < count; i++) {
2020 		if (vectors[i] == 0)
2021 			continue;
2022 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2023 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2024 		    irq, 1);
2025 	}
2026 
2027 	if (bootverbose) {
2028 		device_printf(child, "Remapped MSI-X IRQs as: ");
2029 		for (i = 0; i < count; i++) {
2030 			if (i != 0)
2031 				printf(", ");
2032 			if (vectors[i] == 0)
2033 				printf("---");
2034 			else
2035 				printf("%d",
2036 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2037 		}
2038 		printf("\n");
2039 	}
2040 
2041 	return (0);
2042 }
2043 
2044 static int
2045 pci_release_msix(device_t dev, device_t child)
2046 {
2047 	struct pci_devinfo *dinfo = device_get_ivars(child);
2048 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2049 	struct resource_list_entry *rle;
2050 	int i;
2051 
2052 	/* Do we have any messages to release? */
2053 	if (msix->msix_alloc == 0)
2054 		return (ENODEV);
2055 
2056 	/* Make sure none of the resources are allocated. */
2057 	for (i = 0; i < msix->msix_table_len; i++) {
2058 		if (msix->msix_table[i].mte_vector == 0)
2059 			continue;
2060 		if (msix->msix_table[i].mte_handlers > 0)
2061 			return (EBUSY);
2062 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2063 		KASSERT(rle != NULL, ("missing resource"));
2064 		if (rle->res != NULL)
2065 			return (EBUSY);
2066 	}
2067 
2068 	/* Update control register to disable MSI-X. */
2069 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2070 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2071 	    msix->msix_ctrl, 2);
2072 
2073 	/* Free the resource list entries. */
2074 	for (i = 0; i < msix->msix_table_len; i++) {
2075 		if (msix->msix_table[i].mte_vector == 0)
2076 			continue;
2077 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2078 	}
2079 	free(msix->msix_table, M_DEVBUF);
2080 	msix->msix_table_len = 0;
2081 
2082 	/* Release the IRQs. */
2083 	for (i = 0; i < msix->msix_alloc; i++)
2084 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2085 		    msix->msix_vectors[i].mv_irq);
2086 	free(msix->msix_vectors, M_DEVBUF);
2087 	msix->msix_alloc = 0;
2088 	return (0);
2089 }
2090 
2091 /*
2092  * Return the max supported MSI-X messages this device supports.
2093  * Basically, assuming the MD code can alloc messages, this function
2094  * should return the maximum value that pci_alloc_msix() can return.
2095  * Thus, it is subject to the tunables, etc.
2096  */
2097 int
2098 pci_msix_count_method(device_t dev, device_t child)
2099 {
2100 	struct pci_devinfo *dinfo = device_get_ivars(child);
2101 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2102 
2103 	if (pci_do_msix && msix->msix_location != 0)
2104 		return (msix->msix_msgnum);
2105 	return (0);
2106 }
2107 
2108 int
2109 pci_msix_pba_bar_method(device_t dev, device_t child)
2110 {
2111 	struct pci_devinfo *dinfo = device_get_ivars(child);
2112 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2113 
2114 	if (pci_do_msix && msix->msix_location != 0)
2115 		return (msix->msix_pba_bar);
2116 	return (-1);
2117 }
2118 
2119 int
2120 pci_msix_table_bar_method(device_t dev, device_t child)
2121 {
2122 	struct pci_devinfo *dinfo = device_get_ivars(child);
2123 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2124 
2125 	if (pci_do_msix && msix->msix_location != 0)
2126 		return (msix->msix_table_bar);
2127 	return (-1);
2128 }
2129 
2130 /*
2131  * HyperTransport MSI mapping control
2132  */
2133 void
2134 pci_ht_map_msi(device_t dev, uint64_t addr)
2135 {
2136 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2137 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2138 
2139 	if (!ht->ht_msimap)
2140 		return;
2141 
2142 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2143 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2144 		/* Enable MSI -> HT mapping. */
2145 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2146 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2147 		    ht->ht_msictrl, 2);
2148 	}
2149 
2150 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2151 		/* Disable MSI -> HT mapping. */
2152 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2153 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2154 		    ht->ht_msictrl, 2);
2155 	}
2156 }
2157 
2158 int
2159 pci_get_max_payload(device_t dev)
2160 {
2161 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2162 	int cap;
2163 	uint16_t val;
2164 
2165 	cap = dinfo->cfg.pcie.pcie_location;
2166 	if (cap == 0)
2167 		return (0);
2168 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2169 	val &= PCIEM_CTL_MAX_PAYLOAD;
2170 	val >>= 5;
2171 	return (1 << (val + 7));
2172 }
2173 
2174 int
2175 pci_get_max_read_req(device_t dev)
2176 {
2177 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2178 	int cap;
2179 	uint16_t val;
2180 
2181 	cap = dinfo->cfg.pcie.pcie_location;
2182 	if (cap == 0)
2183 		return (0);
2184 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2185 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2186 	val >>= 12;
2187 	return (1 << (val + 7));
2188 }
2189 
2190 int
2191 pci_set_max_read_req(device_t dev, int size)
2192 {
2193 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2194 	int cap;
2195 	uint16_t val;
2196 
2197 	cap = dinfo->cfg.pcie.pcie_location;
2198 	if (cap == 0)
2199 		return (0);
2200 	if (size < 128)
2201 		size = 128;
2202 	if (size > 4096)
2203 		size = 4096;
2204 	size = (1 << (fls(size) - 1));
2205 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2206 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2207 	val |= (fls(size) - 8) << 12;
2208 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2209 	return (size);
2210 }
2211 
2212 uint32_t
2213 pcie_read_config(device_t dev, int reg, int width)
2214 {
2215 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2216 	int cap;
2217 
2218 	cap = dinfo->cfg.pcie.pcie_location;
2219 	if (cap == 0) {
2220 		if (width == 2)
2221 			return (0xffff);
2222 		return (0xffffffff);
2223 	}
2224 
2225 	return (pci_read_config(dev, cap + reg, width));
2226 }
2227 
2228 void
2229 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2230 {
2231 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2232 	int cap;
2233 
2234 	cap = dinfo->cfg.pcie.pcie_location;
2235 	if (cap == 0)
2236 		return;
2237 	pci_write_config(dev, cap + reg, value, width);
2238 }
2239 
2240 /*
2241  * Adjusts a PCI-e capability register by clearing the bits in mask
2242  * and setting the bits in (value & mask).  Bits not set in mask are
2243  * not adjusted.
2244  *
2245  * Returns the old value on success or all ones on failure.
2246  */
2247 uint32_t
2248 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2249     int width)
2250 {
2251 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2252 	uint32_t old, new;
2253 	int cap;
2254 
2255 	cap = dinfo->cfg.pcie.pcie_location;
2256 	if (cap == 0) {
2257 		if (width == 2)
2258 			return (0xffff);
2259 		return (0xffffffff);
2260 	}
2261 
2262 	old = pci_read_config(dev, cap + reg, width);
2263 	new = old & ~mask;
2264 	new |= (value & mask);
2265 	pci_write_config(dev, cap + reg, new, width);
2266 	return (old);
2267 }
2268 
2269 /*
2270  * Support for MSI message signalled interrupts.
2271  */
2272 void
2273 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2274     uint16_t data)
2275 {
2276 	struct pci_devinfo *dinfo = device_get_ivars(child);
2277 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2278 
2279 	/* Write data and address values. */
2280 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2281 	    address & 0xffffffff, 4);
2282 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2283 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2284 		    address >> 32, 4);
2285 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2286 		    data, 2);
2287 	} else
2288 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2289 		    2);
2290 
2291 	/* Enable MSI in the control register. */
2292 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2293 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2294 	    msi->msi_ctrl, 2);
2295 
2296 	/* Enable MSI -> HT mapping. */
2297 	pci_ht_map_msi(child, address);
2298 }
2299 
2300 void
2301 pci_disable_msi_method(device_t dev, device_t child)
2302 {
2303 	struct pci_devinfo *dinfo = device_get_ivars(child);
2304 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2305 
2306 	/* Disable MSI -> HT mapping. */
2307 	pci_ht_map_msi(child, 0);
2308 
2309 	/* Disable MSI in the control register. */
2310 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2311 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2312 	    msi->msi_ctrl, 2);
2313 }
2314 
2315 /*
2316  * Restore MSI registers during resume.  If MSI is enabled then
2317  * restore the data and address registers in addition to the control
2318  * register.
2319  */
2320 static void
2321 pci_resume_msi(device_t dev)
2322 {
2323 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2324 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2325 	uint64_t address;
2326 	uint16_t data;
2327 
2328 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2329 		address = msi->msi_addr;
2330 		data = msi->msi_data;
2331 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2332 		    address & 0xffffffff, 4);
2333 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2334 			pci_write_config(dev, msi->msi_location +
2335 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2336 			pci_write_config(dev, msi->msi_location +
2337 			    PCIR_MSI_DATA_64BIT, data, 2);
2338 		} else
2339 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2340 			    data, 2);
2341 	}
2342 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2343 	    2);
2344 }
2345 
2346 static int
2347 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2348 {
2349 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2350 	pcicfgregs *cfg = &dinfo->cfg;
2351 	struct resource_list_entry *rle;
2352 	struct msix_table_entry *mte;
2353 	struct msix_vector *mv;
2354 	uint64_t addr;
2355 	uint32_t data;
2356 	int error, i, j;
2357 
2358 	/*
2359 	 * Handle MSI first.  We try to find this IRQ among our list
2360 	 * of MSI IRQs.  If we find it, we request updated address and
2361 	 * data registers and apply the results.
2362 	 */
2363 	if (cfg->msi.msi_alloc > 0) {
2364 		/* If we don't have any active handlers, nothing to do. */
2365 		if (cfg->msi.msi_handlers == 0)
2366 			return (0);
2367 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2368 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2369 			    i + 1);
2370 			if (rle->start == irq) {
2371 				error = PCIB_MAP_MSI(device_get_parent(bus),
2372 				    dev, irq, &addr, &data);
2373 				if (error)
2374 					return (error);
2375 				pci_disable_msi(dev);
2376 				dinfo->cfg.msi.msi_addr = addr;
2377 				dinfo->cfg.msi.msi_data = data;
2378 				pci_enable_msi(dev, addr, data);
2379 				return (0);
2380 			}
2381 		}
2382 		return (ENOENT);
2383 	}
2384 
2385 	/*
2386 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2387 	 * we request the updated mapping info.  If that works, we go
2388 	 * through all the slots that use this IRQ and update them.
2389 	 */
2390 	if (cfg->msix.msix_alloc > 0) {
2391 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2392 			mv = &cfg->msix.msix_vectors[i];
2393 			if (mv->mv_irq == irq) {
2394 				error = PCIB_MAP_MSI(device_get_parent(bus),
2395 				    dev, irq, &addr, &data);
2396 				if (error)
2397 					return (error);
2398 				mv->mv_address = addr;
2399 				mv->mv_data = data;
2400 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2401 					mte = &cfg->msix.msix_table[j];
2402 					if (mte->mte_vector != i + 1)
2403 						continue;
2404 					if (mte->mte_handlers == 0)
2405 						continue;
2406 					pci_mask_msix(dev, j);
2407 					pci_enable_msix(dev, j, addr, data);
2408 					pci_unmask_msix(dev, j);
2409 				}
2410 			}
2411 		}
2412 		return (ENOENT);
2413 	}
2414 
2415 	return (ENOENT);
2416 }
2417 
2418 /*
2419  * Returns true if the specified device is blacklisted because MSI
2420  * doesn't work.
2421  */
2422 int
2423 pci_msi_device_blacklisted(device_t dev)
2424 {
2425 
2426 	if (!pci_honor_msi_blacklist)
2427 		return (0);
2428 
2429 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2430 }
2431 
2432 /*
2433  * Determine if MSI is blacklisted globally on this system.  Currently,
2434  * we just check for blacklisted chipsets as represented by the
2435  * host-PCI bridge at device 0:0:0.  In the future, it may become
2436  * necessary to check other system attributes, such as the kenv values
2437  * that give the motherboard manufacturer and model number.
2438  */
2439 static int
2440 pci_msi_blacklisted(void)
2441 {
2442 	device_t dev;
2443 
2444 	if (!pci_honor_msi_blacklist)
2445 		return (0);
2446 
2447 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2448 	if (!(pcie_chipset || pcix_chipset)) {
2449 		if (vm_guest != VM_GUEST_NO) {
2450 			/*
2451 			 * Whitelist older chipsets in virtual
2452 			 * machines known to support MSI.
2453 			 */
2454 			dev = pci_find_bsf(0, 0, 0);
2455 			if (dev != NULL)
2456 				return (!pci_has_quirk(pci_get_devid(dev),
2457 					PCI_QUIRK_ENABLE_MSI_VM));
2458 		}
2459 		return (1);
2460 	}
2461 
2462 	dev = pci_find_bsf(0, 0, 0);
2463 	if (dev != NULL)
2464 		return (pci_msi_device_blacklisted(dev));
2465 	return (0);
2466 }
2467 
2468 /*
2469  * Returns true if the specified device is blacklisted because MSI-X
2470  * doesn't work.  Note that this assumes that if MSI doesn't work,
2471  * MSI-X doesn't either.
2472  */
2473 int
2474 pci_msix_device_blacklisted(device_t dev)
2475 {
2476 
2477 	if (!pci_honor_msi_blacklist)
2478 		return (0);
2479 
2480 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2481 		return (1);
2482 
2483 	return (pci_msi_device_blacklisted(dev));
2484 }
2485 
2486 /*
2487  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2488  * is blacklisted, assume that MSI-X is as well.  Check for additional
2489  * chipsets where MSI works but MSI-X does not.
2490  */
2491 static int
2492 pci_msix_blacklisted(void)
2493 {
2494 	device_t dev;
2495 
2496 	if (!pci_honor_msi_blacklist)
2497 		return (0);
2498 
2499 	dev = pci_find_bsf(0, 0, 0);
2500 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2501 	    PCI_QUIRK_DISABLE_MSIX))
2502 		return (1);
2503 
2504 	return (pci_msi_blacklisted());
2505 }
2506 
2507 /*
2508  * Attempt to allocate *count MSI messages.  The actual number allocated is
2509  * returned in *count.  After this function returns, each message will be
2510  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2511  */
2512 int
2513 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2514 {
2515 	struct pci_devinfo *dinfo = device_get_ivars(child);
2516 	pcicfgregs *cfg = &dinfo->cfg;
2517 	struct resource_list_entry *rle;
2518 	int actual, error, i, irqs[32];
2519 	uint16_t ctrl;
2520 
2521 	/* Don't let count == 0 get us into trouble. */
2522 	if (*count == 0)
2523 		return (EINVAL);
2524 
2525 	/* If rid 0 is allocated, then fail. */
2526 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2527 	if (rle != NULL && rle->res != NULL)
2528 		return (ENXIO);
2529 
2530 	/* Already have allocated messages? */
2531 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2532 		return (ENXIO);
2533 
2534 	/* If MSI is blacklisted for this system, fail. */
2535 	if (pci_msi_blacklisted())
2536 		return (ENXIO);
2537 
2538 	/* MSI capability present? */
2539 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2540 		return (ENODEV);
2541 
2542 	if (bootverbose)
2543 		device_printf(child,
2544 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2545 		    *count, cfg->msi.msi_msgnum);
2546 
2547 	/* Don't ask for more than the device supports. */
2548 	actual = min(*count, cfg->msi.msi_msgnum);
2549 
2550 	/* Don't ask for more than 32 messages. */
2551 	actual = min(actual, 32);
2552 
2553 	/* MSI requires power of 2 number of messages. */
2554 	if (!powerof2(actual))
2555 		return (EINVAL);
2556 
2557 	for (;;) {
2558 		/* Try to allocate N messages. */
2559 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2560 		    actual, irqs);
2561 		if (error == 0)
2562 			break;
2563 		if (actual == 1)
2564 			return (error);
2565 
2566 		/* Try N / 2. */
2567 		actual >>= 1;
2568 	}
2569 
2570 	/*
2571 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2572 	 * resources in the irqs[] array, so add new resources
2573 	 * starting at rid 1.
2574 	 */
2575 	for (i = 0; i < actual; i++)
2576 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2577 		    irqs[i], irqs[i], 1);
2578 
2579 	if (bootverbose) {
2580 		if (actual == 1)
2581 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2582 		else {
2583 			int run;
2584 
2585 			/*
2586 			 * Be fancy and try to print contiguous runs
2587 			 * of IRQ values as ranges.  'run' is true if
2588 			 * we are in a range.
2589 			 */
2590 			device_printf(child, "using IRQs %d", irqs[0]);
2591 			run = 0;
2592 			for (i = 1; i < actual; i++) {
2593 				/* Still in a run? */
2594 				if (irqs[i] == irqs[i - 1] + 1) {
2595 					run = 1;
2596 					continue;
2597 				}
2598 
2599 				/* Finish previous range. */
2600 				if (run) {
2601 					printf("-%d", irqs[i - 1]);
2602 					run = 0;
2603 				}
2604 
2605 				/* Start new range. */
2606 				printf(",%d", irqs[i]);
2607 			}
2608 
2609 			/* Unfinished range? */
2610 			if (run)
2611 				printf("-%d", irqs[actual - 1]);
2612 			printf(" for MSI\n");
2613 		}
2614 	}
2615 
2616 	/* Update control register with actual count. */
2617 	ctrl = cfg->msi.msi_ctrl;
2618 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2619 	ctrl |= (ffs(actual) - 1) << 4;
2620 	cfg->msi.msi_ctrl = ctrl;
2621 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2622 
2623 	/* Update counts of alloc'd messages. */
2624 	cfg->msi.msi_alloc = actual;
2625 	cfg->msi.msi_handlers = 0;
2626 	*count = actual;
2627 	return (0);
2628 }
2629 
2630 /* Release the MSI messages associated with this device. */
2631 int
2632 pci_release_msi_method(device_t dev, device_t child)
2633 {
2634 	struct pci_devinfo *dinfo = device_get_ivars(child);
2635 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2636 	struct resource_list_entry *rle;
2637 	int error, i, irqs[32];
2638 
2639 	/* Try MSI-X first. */
2640 	error = pci_release_msix(dev, child);
2641 	if (error != ENODEV)
2642 		return (error);
2643 
2644 	/* Do we have any messages to release? */
2645 	if (msi->msi_alloc == 0)
2646 		return (ENODEV);
2647 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2648 
2649 	/* Make sure none of the resources are allocated. */
2650 	if (msi->msi_handlers > 0)
2651 		return (EBUSY);
2652 	for (i = 0; i < msi->msi_alloc; i++) {
2653 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2654 		KASSERT(rle != NULL, ("missing MSI resource"));
2655 		if (rle->res != NULL)
2656 			return (EBUSY);
2657 		irqs[i] = rle->start;
2658 	}
2659 
2660 	/* Update control register with 0 count. */
2661 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2662 	    ("%s: MSI still enabled", __func__));
2663 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2664 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2665 	    msi->msi_ctrl, 2);
2666 
2667 	/* Release the messages. */
2668 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2669 	for (i = 0; i < msi->msi_alloc; i++)
2670 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2671 
2672 	/* Update alloc count. */
2673 	msi->msi_alloc = 0;
2674 	msi->msi_addr = 0;
2675 	msi->msi_data = 0;
2676 	return (0);
2677 }
2678 
2679 /*
2680  * Return the max supported MSI messages this device supports.
2681  * Basically, assuming the MD code can alloc messages, this function
2682  * should return the maximum value that pci_alloc_msi() can return.
2683  * Thus, it is subject to the tunables, etc.
2684  */
2685 int
2686 pci_msi_count_method(device_t dev, device_t child)
2687 {
2688 	struct pci_devinfo *dinfo = device_get_ivars(child);
2689 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2690 
2691 	if (pci_do_msi && msi->msi_location != 0)
2692 		return (msi->msi_msgnum);
2693 	return (0);
2694 }
2695 
2696 /* free pcicfgregs structure and all depending data structures */
2697 
2698 int
2699 pci_freecfg(struct pci_devinfo *dinfo)
2700 {
2701 	struct devlist *devlist_head;
2702 	struct pci_map *pm, *next;
2703 	int i;
2704 
2705 	devlist_head = &pci_devq;
2706 
2707 	if (dinfo->cfg.vpd.vpd_reg) {
2708 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2709 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2710 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2711 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2712 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2713 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2714 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2715 	}
2716 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2717 		free(pm, M_DEVBUF);
2718 	}
2719 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2720 	free(dinfo, M_DEVBUF);
2721 
2722 	/* increment the generation count */
2723 	pci_generation++;
2724 
2725 	/* we're losing one device */
2726 	pci_numdevs--;
2727 	return (0);
2728 }
2729 
2730 /*
2731  * PCI power manangement
2732  */
2733 int
2734 pci_set_powerstate_method(device_t dev, device_t child, int state)
2735 {
2736 	struct pci_devinfo *dinfo = device_get_ivars(child);
2737 	pcicfgregs *cfg = &dinfo->cfg;
2738 	uint16_t status;
2739 	int oldstate, highest, delay;
2740 
2741 	if (cfg->pp.pp_cap == 0)
2742 		return (EOPNOTSUPP);
2743 
2744 	/*
2745 	 * Optimize a no state change request away.  While it would be OK to
2746 	 * write to the hardware in theory, some devices have shown odd
2747 	 * behavior when going from D3 -> D3.
2748 	 */
2749 	oldstate = pci_get_powerstate(child);
2750 	if (oldstate == state)
2751 		return (0);
2752 
2753 	/*
2754 	 * The PCI power management specification states that after a state
2755 	 * transition between PCI power states, system software must
2756 	 * guarantee a minimal delay before the function accesses the device.
2757 	 * Compute the worst case delay that we need to guarantee before we
2758 	 * access the device.  Many devices will be responsive much more
2759 	 * quickly than this delay, but there are some that don't respond
2760 	 * instantly to state changes.  Transitions to/from D3 state require
2761 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2762 	 * is done below with DELAY rather than a sleeper function because
2763 	 * this function can be called from contexts where we cannot sleep.
2764 	 */
2765 	highest = (oldstate > state) ? oldstate : state;
2766 	if (highest == PCI_POWERSTATE_D3)
2767 	    delay = 10000;
2768 	else if (highest == PCI_POWERSTATE_D2)
2769 	    delay = 200;
2770 	else
2771 	    delay = 0;
2772 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2773 	    & ~PCIM_PSTAT_DMASK;
2774 	switch (state) {
2775 	case PCI_POWERSTATE_D0:
2776 		status |= PCIM_PSTAT_D0;
2777 		break;
2778 	case PCI_POWERSTATE_D1:
2779 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2780 			return (EOPNOTSUPP);
2781 		status |= PCIM_PSTAT_D1;
2782 		break;
2783 	case PCI_POWERSTATE_D2:
2784 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2785 			return (EOPNOTSUPP);
2786 		status |= PCIM_PSTAT_D2;
2787 		break;
2788 	case PCI_POWERSTATE_D3:
2789 		status |= PCIM_PSTAT_D3;
2790 		break;
2791 	default:
2792 		return (EINVAL);
2793 	}
2794 
2795 	if (bootverbose)
2796 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2797 		    state);
2798 
2799 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2800 	if (delay)
2801 		DELAY(delay);
2802 	return (0);
2803 }
2804 
2805 int
2806 pci_get_powerstate_method(device_t dev, device_t child)
2807 {
2808 	struct pci_devinfo *dinfo = device_get_ivars(child);
2809 	pcicfgregs *cfg = &dinfo->cfg;
2810 	uint16_t status;
2811 	int result;
2812 
2813 	if (cfg->pp.pp_cap != 0) {
2814 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2815 		switch (status & PCIM_PSTAT_DMASK) {
2816 		case PCIM_PSTAT_D0:
2817 			result = PCI_POWERSTATE_D0;
2818 			break;
2819 		case PCIM_PSTAT_D1:
2820 			result = PCI_POWERSTATE_D1;
2821 			break;
2822 		case PCIM_PSTAT_D2:
2823 			result = PCI_POWERSTATE_D2;
2824 			break;
2825 		case PCIM_PSTAT_D3:
2826 			result = PCI_POWERSTATE_D3;
2827 			break;
2828 		default:
2829 			result = PCI_POWERSTATE_UNKNOWN;
2830 			break;
2831 		}
2832 	} else {
2833 		/* No support, device is always at D0 */
2834 		result = PCI_POWERSTATE_D0;
2835 	}
2836 	return (result);
2837 }
2838 
2839 /*
2840  * Some convenience functions for PCI device drivers.
2841  */
2842 
2843 static __inline void
2844 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2845 {
2846 	uint16_t	command;
2847 
2848 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2849 	command |= bit;
2850 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2851 }
2852 
2853 static __inline void
2854 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2855 {
2856 	uint16_t	command;
2857 
2858 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2859 	command &= ~bit;
2860 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2861 }
2862 
2863 int
2864 pci_enable_busmaster_method(device_t dev, device_t child)
2865 {
2866 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2867 	return (0);
2868 }
2869 
2870 int
2871 pci_disable_busmaster_method(device_t dev, device_t child)
2872 {
2873 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2874 	return (0);
2875 }
2876 
2877 int
2878 pci_enable_io_method(device_t dev, device_t child, int space)
2879 {
2880 	uint16_t bit;
2881 
2882 	switch(space) {
2883 	case SYS_RES_IOPORT:
2884 		bit = PCIM_CMD_PORTEN;
2885 		break;
2886 	case SYS_RES_MEMORY:
2887 		bit = PCIM_CMD_MEMEN;
2888 		break;
2889 	default:
2890 		return (EINVAL);
2891 	}
2892 	pci_set_command_bit(dev, child, bit);
2893 	return (0);
2894 }
2895 
2896 int
2897 pci_disable_io_method(device_t dev, device_t child, int space)
2898 {
2899 	uint16_t bit;
2900 
2901 	switch(space) {
2902 	case SYS_RES_IOPORT:
2903 		bit = PCIM_CMD_PORTEN;
2904 		break;
2905 	case SYS_RES_MEMORY:
2906 		bit = PCIM_CMD_MEMEN;
2907 		break;
2908 	default:
2909 		return (EINVAL);
2910 	}
2911 	pci_clear_command_bit(dev, child, bit);
2912 	return (0);
2913 }
2914 
2915 /*
2916  * New style pci driver.  Parent device is either a pci-host-bridge or a
2917  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2918  */
2919 
2920 void
2921 pci_print_verbose(struct pci_devinfo *dinfo)
2922 {
2923 
2924 	if (bootverbose) {
2925 		pcicfgregs *cfg = &dinfo->cfg;
2926 
2927 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2928 		    cfg->vendor, cfg->device, cfg->revid);
2929 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2930 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2931 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2932 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2933 		    cfg->mfdev);
2934 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2935 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2936 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2937 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2938 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2939 		if (cfg->intpin > 0)
2940 			printf("\tintpin=%c, irq=%d\n",
2941 			    cfg->intpin +'a' -1, cfg->intline);
2942 		if (cfg->pp.pp_cap) {
2943 			uint16_t status;
2944 
2945 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2946 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2947 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2948 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2949 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2950 			    status & PCIM_PSTAT_DMASK);
2951 		}
2952 		if (cfg->msi.msi_location) {
2953 			int ctrl;
2954 
2955 			ctrl = cfg->msi.msi_ctrl;
2956 			printf("\tMSI supports %d message%s%s%s\n",
2957 			    cfg->msi.msi_msgnum,
2958 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2959 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2960 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2961 		}
2962 		if (cfg->msix.msix_location) {
2963 			printf("\tMSI-X supports %d message%s ",
2964 			    cfg->msix.msix_msgnum,
2965 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2966 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2967 				printf("in map 0x%x\n",
2968 				    cfg->msix.msix_table_bar);
2969 			else
2970 				printf("in maps 0x%x and 0x%x\n",
2971 				    cfg->msix.msix_table_bar,
2972 				    cfg->msix.msix_pba_bar);
2973 		}
2974 	}
2975 }
2976 
2977 static int
2978 pci_porten(device_t dev)
2979 {
2980 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2981 }
2982 
2983 static int
2984 pci_memen(device_t dev)
2985 {
2986 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2987 }
2988 
2989 void
2990 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2991     int *bar64)
2992 {
2993 	struct pci_devinfo *dinfo;
2994 	pci_addr_t map, testval;
2995 	int ln2range;
2996 	uint16_t cmd;
2997 
2998 	/*
2999 	 * The device ROM BAR is special.  It is always a 32-bit
3000 	 * memory BAR.  Bit 0 is special and should not be set when
3001 	 * sizing the BAR.
3002 	 */
3003 	dinfo = device_get_ivars(dev);
3004 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3005 		map = pci_read_config(dev, reg, 4);
3006 		pci_write_config(dev, reg, 0xfffffffe, 4);
3007 		testval = pci_read_config(dev, reg, 4);
3008 		pci_write_config(dev, reg, map, 4);
3009 		*mapp = map;
3010 		*testvalp = testval;
3011 		if (bar64 != NULL)
3012 			*bar64 = 0;
3013 		return;
3014 	}
3015 
3016 	map = pci_read_config(dev, reg, 4);
3017 	ln2range = pci_maprange(map);
3018 	if (ln2range == 64)
3019 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3020 
3021 	/*
3022 	 * Disable decoding via the command register before
3023 	 * determining the BAR's length since we will be placing it in
3024 	 * a weird state.
3025 	 */
3026 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3027 	pci_write_config(dev, PCIR_COMMAND,
3028 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3029 
3030 	/*
3031 	 * Determine the BAR's length by writing all 1's.  The bottom
3032 	 * log_2(size) bits of the BAR will stick as 0 when we read
3033 	 * the value back.
3034 	 *
3035 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3036 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3037 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3038 	 *
3039 	 * Writes to both registers must be performed before attempting to
3040 	 * read back the size value.
3041 	 */
3042 	testval = 0;
3043 	pci_write_config(dev, reg, 0xffffffff, 4);
3044 	if (ln2range == 64) {
3045 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3046 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3047 	}
3048 	testval |= pci_read_config(dev, reg, 4);
3049 
3050 	/*
3051 	 * Restore the original value of the BAR.  We may have reprogrammed
3052 	 * the BAR of the low-level console device and when booting verbose,
3053 	 * we need the console device addressable.
3054 	 */
3055 	pci_write_config(dev, reg, map, 4);
3056 	if (ln2range == 64)
3057 		pci_write_config(dev, reg + 4, map >> 32, 4);
3058 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3059 
3060 	*mapp = map;
3061 	*testvalp = testval;
3062 	if (bar64 != NULL)
3063 		*bar64 = (ln2range == 64);
3064 }
3065 
3066 static void
3067 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3068 {
3069 	struct pci_devinfo *dinfo;
3070 	int ln2range;
3071 
3072 	/* The device ROM BAR is always a 32-bit memory BAR. */
3073 	dinfo = device_get_ivars(dev);
3074 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3075 		ln2range = 32;
3076 	else
3077 		ln2range = pci_maprange(pm->pm_value);
3078 	pci_write_config(dev, pm->pm_reg, base, 4);
3079 	if (ln2range == 64)
3080 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3081 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3082 	if (ln2range == 64)
3083 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3084 		    pm->pm_reg + 4, 4) << 32;
3085 }
3086 
3087 struct pci_map *
3088 pci_find_bar(device_t dev, int reg)
3089 {
3090 	struct pci_devinfo *dinfo;
3091 	struct pci_map *pm;
3092 
3093 	dinfo = device_get_ivars(dev);
3094 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3095 		if (pm->pm_reg == reg)
3096 			return (pm);
3097 	}
3098 	return (NULL);
3099 }
3100 
3101 int
3102 pci_bar_enabled(device_t dev, struct pci_map *pm)
3103 {
3104 	struct pci_devinfo *dinfo;
3105 	uint16_t cmd;
3106 
3107 	dinfo = device_get_ivars(dev);
3108 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3109 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3110 		return (0);
3111 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3112 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3113 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3114 	else
3115 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3116 }
3117 
3118 struct pci_map *
3119 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3120 {
3121 	struct pci_devinfo *dinfo;
3122 	struct pci_map *pm, *prev;
3123 
3124 	dinfo = device_get_ivars(dev);
3125 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3126 	pm->pm_reg = reg;
3127 	pm->pm_value = value;
3128 	pm->pm_size = size;
3129 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3130 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3131 		    reg));
3132 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3133 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3134 			break;
3135 	}
3136 	if (prev != NULL)
3137 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3138 	else
3139 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3140 	return (pm);
3141 }
3142 
3143 static void
3144 pci_restore_bars(device_t dev)
3145 {
3146 	struct pci_devinfo *dinfo;
3147 	struct pci_map *pm;
3148 	int ln2range;
3149 
3150 	dinfo = device_get_ivars(dev);
3151 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3152 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3153 			ln2range = 32;
3154 		else
3155 			ln2range = pci_maprange(pm->pm_value);
3156 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3157 		if (ln2range == 64)
3158 			pci_write_config(dev, pm->pm_reg + 4,
3159 			    pm->pm_value >> 32, 4);
3160 	}
3161 }
3162 
3163 /*
3164  * Add a resource based on a pci map register. Return 1 if the map
3165  * register is a 32bit map register or 2 if it is a 64bit register.
3166  */
3167 static int
3168 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3169     int force, int prefetch)
3170 {
3171 	struct pci_map *pm;
3172 	pci_addr_t base, map, testval;
3173 	pci_addr_t start, end, count;
3174 	int barlen, basezero, flags, maprange, mapsize, type;
3175 	uint16_t cmd;
3176 	struct resource *res;
3177 
3178 	/*
3179 	 * The BAR may already exist if the device is a CardBus card
3180 	 * whose CIS is stored in this BAR.
3181 	 */
3182 	pm = pci_find_bar(dev, reg);
3183 	if (pm != NULL) {
3184 		maprange = pci_maprange(pm->pm_value);
3185 		barlen = maprange == 64 ? 2 : 1;
3186 		return (barlen);
3187 	}
3188 
3189 	pci_read_bar(dev, reg, &map, &testval, NULL);
3190 	if (PCI_BAR_MEM(map)) {
3191 		type = SYS_RES_MEMORY;
3192 		if (map & PCIM_BAR_MEM_PREFETCH)
3193 			prefetch = 1;
3194 	} else
3195 		type = SYS_RES_IOPORT;
3196 	mapsize = pci_mapsize(testval);
3197 	base = pci_mapbase(map);
3198 #ifdef __PCI_BAR_ZERO_VALID
3199 	basezero = 0;
3200 #else
3201 	basezero = base == 0;
3202 #endif
3203 	maprange = pci_maprange(map);
3204 	barlen = maprange == 64 ? 2 : 1;
3205 
3206 	/*
3207 	 * For I/O registers, if bottom bit is set, and the next bit up
3208 	 * isn't clear, we know we have a BAR that doesn't conform to the
3209 	 * spec, so ignore it.  Also, sanity check the size of the data
3210 	 * areas to the type of memory involved.  Memory must be at least
3211 	 * 16 bytes in size, while I/O ranges must be at least 4.
3212 	 */
3213 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3214 		return (barlen);
3215 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3216 	    (type == SYS_RES_IOPORT && mapsize < 2))
3217 		return (barlen);
3218 
3219 	/* Save a record of this BAR. */
3220 	pm = pci_add_bar(dev, reg, map, mapsize);
3221 	if (bootverbose) {
3222 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3223 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3224 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3225 			printf(", port disabled\n");
3226 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3227 			printf(", memory disabled\n");
3228 		else
3229 			printf(", enabled\n");
3230 	}
3231 
3232 	/*
3233 	 * If base is 0, then we have problems if this architecture does
3234 	 * not allow that.  It is best to ignore such entries for the
3235 	 * moment.  These will be allocated later if the driver specifically
3236 	 * requests them.  However, some removable buses look better when
3237 	 * all resources are allocated, so allow '0' to be overriden.
3238 	 *
3239 	 * Similarly treat maps whose values is the same as the test value
3240 	 * read back.  These maps have had all f's written to them by the
3241 	 * BIOS in an attempt to disable the resources.
3242 	 */
3243 	if (!force && (basezero || map == testval))
3244 		return (barlen);
3245 	if ((u_long)base != base) {
3246 		device_printf(bus,
3247 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3248 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3249 		    pci_get_function(dev), reg);
3250 		return (barlen);
3251 	}
3252 
3253 	/*
3254 	 * This code theoretically does the right thing, but has
3255 	 * undesirable side effects in some cases where peripherals
3256 	 * respond oddly to having these bits enabled.  Let the user
3257 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3258 	 * default).
3259 	 */
3260 	if (pci_enable_io_modes) {
3261 		/* Turn on resources that have been left off by a lazy BIOS */
3262 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3263 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3264 			cmd |= PCIM_CMD_PORTEN;
3265 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3266 		}
3267 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3268 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3269 			cmd |= PCIM_CMD_MEMEN;
3270 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3271 		}
3272 	} else {
3273 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3274 			return (barlen);
3275 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3276 			return (barlen);
3277 	}
3278 
3279 	count = (pci_addr_t)1 << mapsize;
3280 	flags = RF_ALIGNMENT_LOG2(mapsize);
3281 	if (prefetch)
3282 		flags |= RF_PREFETCHABLE;
3283 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3284 		start = 0;	/* Let the parent decide. */
3285 		end = ~0;
3286 	} else {
3287 		start = base;
3288 		end = base + count - 1;
3289 	}
3290 	resource_list_add(rl, type, reg, start, end, count);
3291 
3292 	/*
3293 	 * Try to allocate the resource for this BAR from our parent
3294 	 * so that this resource range is already reserved.  The
3295 	 * driver for this device will later inherit this resource in
3296 	 * pci_alloc_resource().
3297 	 */
3298 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3299 	    flags);
3300 	if ((pci_do_realloc_bars
3301 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3302 	    && res == NULL && (start != 0 || end != ~0)) {
3303 		/*
3304 		 * If the allocation fails, try to allocate a resource for
3305 		 * this BAR using any available range.  The firmware felt
3306 		 * it was important enough to assign a resource, so don't
3307 		 * disable decoding if we can help it.
3308 		 */
3309 		resource_list_delete(rl, type, reg);
3310 		resource_list_add(rl, type, reg, 0, ~0, count);
3311 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3312 		    count, flags);
3313 	}
3314 	if (res == NULL) {
3315 		/*
3316 		 * If the allocation fails, delete the resource list entry
3317 		 * and disable decoding for this device.
3318 		 *
3319 		 * If the driver requests this resource in the future,
3320 		 * pci_reserve_map() will try to allocate a fresh
3321 		 * resource range.
3322 		 */
3323 		resource_list_delete(rl, type, reg);
3324 		pci_disable_io(dev, type);
3325 		if (bootverbose)
3326 			device_printf(bus,
3327 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3328 			    pci_get_domain(dev), pci_get_bus(dev),
3329 			    pci_get_slot(dev), pci_get_function(dev), reg);
3330 	} else {
3331 		start = rman_get_start(res);
3332 		pci_write_bar(dev, pm, start);
3333 	}
3334 	return (barlen);
3335 }
3336 
3337 /*
3338  * For ATA devices we need to decide early what addressing mode to use.
3339  * Legacy demands that the primary and secondary ATA ports sits on the
3340  * same addresses that old ISA hardware did. This dictates that we use
3341  * those addresses and ignore the BAR's if we cannot set PCI native
3342  * addressing mode.
3343  */
3344 static void
3345 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3346     uint32_t prefetchmask)
3347 {
3348 	int rid, type, progif;
3349 #if 0
3350 	/* if this device supports PCI native addressing use it */
3351 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3352 	if ((progif & 0x8a) == 0x8a) {
3353 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3354 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3355 			printf("Trying ATA native PCI addressing mode\n");
3356 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3357 		}
3358 	}
3359 #endif
3360 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3361 	type = SYS_RES_IOPORT;
3362 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3363 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3364 		    prefetchmask & (1 << 0));
3365 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3366 		    prefetchmask & (1 << 1));
3367 	} else {
3368 		rid = PCIR_BAR(0);
3369 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3370 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3371 		    0x1f7, 8, 0);
3372 		rid = PCIR_BAR(1);
3373 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3374 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3375 		    0x3f6, 1, 0);
3376 	}
3377 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3378 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3379 		    prefetchmask & (1 << 2));
3380 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3381 		    prefetchmask & (1 << 3));
3382 	} else {
3383 		rid = PCIR_BAR(2);
3384 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3385 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3386 		    0x177, 8, 0);
3387 		rid = PCIR_BAR(3);
3388 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3389 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3390 		    0x376, 1, 0);
3391 	}
3392 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3393 	    prefetchmask & (1 << 4));
3394 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3395 	    prefetchmask & (1 << 5));
3396 }
3397 
3398 static void
3399 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3400 {
3401 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3402 	pcicfgregs *cfg = &dinfo->cfg;
3403 	char tunable_name[64];
3404 	int irq;
3405 
3406 	/* Has to have an intpin to have an interrupt. */
3407 	if (cfg->intpin == 0)
3408 		return;
3409 
3410 	/* Let the user override the IRQ with a tunable. */
3411 	irq = PCI_INVALID_IRQ;
3412 	snprintf(tunable_name, sizeof(tunable_name),
3413 	    "hw.pci%d.%d.%d.INT%c.irq",
3414 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3415 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3416 		irq = PCI_INVALID_IRQ;
3417 
3418 	/*
3419 	 * If we didn't get an IRQ via the tunable, then we either use the
3420 	 * IRQ value in the intline register or we ask the bus to route an
3421 	 * interrupt for us.  If force_route is true, then we only use the
3422 	 * value in the intline register if the bus was unable to assign an
3423 	 * IRQ.
3424 	 */
3425 	if (!PCI_INTERRUPT_VALID(irq)) {
3426 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3427 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3428 		if (!PCI_INTERRUPT_VALID(irq))
3429 			irq = cfg->intline;
3430 	}
3431 
3432 	/* If after all that we don't have an IRQ, just bail. */
3433 	if (!PCI_INTERRUPT_VALID(irq))
3434 		return;
3435 
3436 	/* Update the config register if it changed. */
3437 	if (irq != cfg->intline) {
3438 		cfg->intline = irq;
3439 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3440 	}
3441 
3442 	/* Add this IRQ as rid 0 interrupt resource. */
3443 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3444 }
3445 
3446 /* Perform early OHCI takeover from SMM. */
3447 static void
3448 ohci_early_takeover(device_t self)
3449 {
3450 	struct resource *res;
3451 	uint32_t ctl;
3452 	int rid;
3453 	int i;
3454 
3455 	rid = PCIR_BAR(0);
3456 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3457 	if (res == NULL)
3458 		return;
3459 
3460 	ctl = bus_read_4(res, OHCI_CONTROL);
3461 	if (ctl & OHCI_IR) {
3462 		if (bootverbose)
3463 			printf("ohci early: "
3464 			    "SMM active, request owner change\n");
3465 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3466 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3467 			DELAY(1000);
3468 			ctl = bus_read_4(res, OHCI_CONTROL);
3469 		}
3470 		if (ctl & OHCI_IR) {
3471 			if (bootverbose)
3472 				printf("ohci early: "
3473 				    "SMM does not respond, resetting\n");
3474 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3475 		}
3476 		/* Disable interrupts */
3477 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3478 	}
3479 
3480 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3481 }
3482 
3483 /* Perform early UHCI takeover from SMM. */
3484 static void
3485 uhci_early_takeover(device_t self)
3486 {
3487 	struct resource *res;
3488 	int rid;
3489 
3490 	/*
3491 	 * Set the PIRQD enable bit and switch off all the others. We don't
3492 	 * want legacy support to interfere with us XXX Does this also mean
3493 	 * that the BIOS won't touch the keyboard anymore if it is connected
3494 	 * to the ports of the root hub?
3495 	 */
3496 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3497 
3498 	/* Disable interrupts */
3499 	rid = PCI_UHCI_BASE_REG;
3500 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3501 	if (res != NULL) {
3502 		bus_write_2(res, UHCI_INTR, 0);
3503 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3504 	}
3505 }
3506 
3507 /* Perform early EHCI takeover from SMM. */
3508 static void
3509 ehci_early_takeover(device_t self)
3510 {
3511 	struct resource *res;
3512 	uint32_t cparams;
3513 	uint32_t eec;
3514 	uint8_t eecp;
3515 	uint8_t bios_sem;
3516 	uint8_t offs;
3517 	int rid;
3518 	int i;
3519 
3520 	rid = PCIR_BAR(0);
3521 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3522 	if (res == NULL)
3523 		return;
3524 
3525 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3526 
3527 	/* Synchronise with the BIOS if it owns the controller. */
3528 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3529 	    eecp = EHCI_EECP_NEXT(eec)) {
3530 		eec = pci_read_config(self, eecp, 4);
3531 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3532 			continue;
3533 		}
3534 		bios_sem = pci_read_config(self, eecp +
3535 		    EHCI_LEGSUP_BIOS_SEM, 1);
3536 		if (bios_sem == 0) {
3537 			continue;
3538 		}
3539 		if (bootverbose)
3540 			printf("ehci early: "
3541 			    "SMM active, request owner change\n");
3542 
3543 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3544 
3545 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3546 			DELAY(1000);
3547 			bios_sem = pci_read_config(self, eecp +
3548 			    EHCI_LEGSUP_BIOS_SEM, 1);
3549 		}
3550 
3551 		if (bios_sem != 0) {
3552 			if (bootverbose)
3553 				printf("ehci early: "
3554 				    "SMM does not respond\n");
3555 		}
3556 		/* Disable interrupts */
3557 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3558 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3559 	}
3560 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3561 }
3562 
3563 /* Perform early XHCI takeover from SMM. */
3564 static void
3565 xhci_early_takeover(device_t self)
3566 {
3567 	struct resource *res;
3568 	uint32_t cparams;
3569 	uint32_t eec;
3570 	uint8_t eecp;
3571 	uint8_t bios_sem;
3572 	uint8_t offs;
3573 	int rid;
3574 	int i;
3575 
3576 	rid = PCIR_BAR(0);
3577 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3578 	if (res == NULL)
3579 		return;
3580 
3581 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3582 
3583 	eec = -1;
3584 
3585 	/* Synchronise with the BIOS if it owns the controller. */
3586 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3587 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3588 		eec = bus_read_4(res, eecp);
3589 
3590 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3591 			continue;
3592 
3593 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3594 		if (bios_sem == 0)
3595 			continue;
3596 
3597 		if (bootverbose)
3598 			printf("xhci early: "
3599 			    "SMM active, request owner change\n");
3600 
3601 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3602 
3603 		/* wait a maximum of 5 second */
3604 
3605 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3606 			DELAY(1000);
3607 			bios_sem = bus_read_1(res, eecp +
3608 			    XHCI_XECP_BIOS_SEM);
3609 		}
3610 
3611 		if (bios_sem != 0) {
3612 			if (bootverbose)
3613 				printf("xhci early: "
3614 				    "SMM does not respond\n");
3615 		}
3616 
3617 		/* Disable interrupts */
3618 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3619 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3620 		bus_read_4(res, offs + XHCI_USBSTS);
3621 	}
3622 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3623 }
3624 
3625 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3626 static void
3627 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3628     struct resource_list *rl)
3629 {
3630 	struct resource *res;
3631 	char *cp;
3632 	rman_res_t start, end, count;
3633 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3634 
3635 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3636 	case PCIM_HDRTYPE_BRIDGE:
3637 		sec_reg = PCIR_SECBUS_1;
3638 		sub_reg = PCIR_SUBBUS_1;
3639 		break;
3640 	case PCIM_HDRTYPE_CARDBUS:
3641 		sec_reg = PCIR_SECBUS_2;
3642 		sub_reg = PCIR_SUBBUS_2;
3643 		break;
3644 	default:
3645 		return;
3646 	}
3647 
3648 	/*
3649 	 * If the existing bus range is valid, attempt to reserve it
3650 	 * from our parent.  If this fails for any reason, clear the
3651 	 * secbus and subbus registers.
3652 	 *
3653 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3654 	 * This would at least preserve the existing sec_bus if it is
3655 	 * valid.
3656 	 */
3657 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3658 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3659 
3660 	/* Quirk handling. */
3661 	switch (pci_get_devid(dev)) {
3662 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3663 		sup_bus = pci_read_config(dev, 0x41, 1);
3664 		if (sup_bus != 0xff) {
3665 			sec_bus = sup_bus + 1;
3666 			sub_bus = sup_bus + 1;
3667 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3668 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3669 		}
3670 		break;
3671 
3672 	case 0x00dd10de:
3673 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3674 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3675 			break;
3676 		if (strncmp(cp, "Compal", 6) != 0) {
3677 			freeenv(cp);
3678 			break;
3679 		}
3680 		freeenv(cp);
3681 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3682 			break;
3683 		if (strncmp(cp, "08A0", 4) != 0) {
3684 			freeenv(cp);
3685 			break;
3686 		}
3687 		freeenv(cp);
3688 		if (sub_bus < 0xa) {
3689 			sub_bus = 0xa;
3690 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3691 		}
3692 		break;
3693 	}
3694 
3695 	if (bootverbose)
3696 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3697 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3698 		start = sec_bus;
3699 		end = sub_bus;
3700 		count = end - start + 1;
3701 
3702 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3703 
3704 		/*
3705 		 * If requested, clear secondary bus registers in
3706 		 * bridge devices to force a complete renumbering
3707 		 * rather than reserving the existing range.  However,
3708 		 * preserve the existing size.
3709 		 */
3710 		if (pci_clear_buses)
3711 			goto clear;
3712 
3713 		rid = 0;
3714 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3715 		    start, end, count, 0);
3716 		if (res != NULL)
3717 			return;
3718 
3719 		if (bootverbose)
3720 			device_printf(bus,
3721 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3722 			    pci_get_domain(dev), pci_get_bus(dev),
3723 			    pci_get_slot(dev), pci_get_function(dev));
3724 	}
3725 
3726 clear:
3727 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3728 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3729 }
3730 
3731 static struct resource *
3732 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3733     rman_res_t end, rman_res_t count, u_int flags)
3734 {
3735 	struct pci_devinfo *dinfo;
3736 	pcicfgregs *cfg;
3737 	struct resource_list *rl;
3738 	struct resource *res;
3739 	int sec_reg, sub_reg;
3740 
3741 	dinfo = device_get_ivars(child);
3742 	cfg = &dinfo->cfg;
3743 	rl = &dinfo->resources;
3744 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3745 	case PCIM_HDRTYPE_BRIDGE:
3746 		sec_reg = PCIR_SECBUS_1;
3747 		sub_reg = PCIR_SUBBUS_1;
3748 		break;
3749 	case PCIM_HDRTYPE_CARDBUS:
3750 		sec_reg = PCIR_SECBUS_2;
3751 		sub_reg = PCIR_SUBBUS_2;
3752 		break;
3753 	default:
3754 		return (NULL);
3755 	}
3756 
3757 	if (*rid != 0)
3758 		return (NULL);
3759 
3760 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3761 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3762 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3763 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3764 		    start, end, count, flags & ~RF_ACTIVE);
3765 		if (res == NULL) {
3766 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3767 			device_printf(child, "allocating %ju bus%s failed\n",
3768 			    count, count == 1 ? "" : "es");
3769 			return (NULL);
3770 		}
3771 		if (bootverbose)
3772 			device_printf(child,
3773 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3774 			    count == 1 ? "" : "es", rman_get_start(res));
3775 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3776 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3777 	}
3778 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3779 	    end, count, flags));
3780 }
3781 #endif
3782 
3783 static int
3784 pci_ea_bei_to_rid(device_t dev, int bei)
3785 {
3786 #ifdef PCI_IOV
3787 	struct pci_devinfo *dinfo;
3788 	int iov_pos;
3789 	struct pcicfg_iov *iov;
3790 
3791 	dinfo = device_get_ivars(dev);
3792 	iov = dinfo->cfg.iov;
3793 	if (iov != NULL)
3794 		iov_pos = iov->iov_pos;
3795 	else
3796 		iov_pos = 0;
3797 #endif
3798 
3799 	/* Check if matches BAR */
3800 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3801 	    (bei <= PCIM_EA_BEI_BAR_5))
3802 		return (PCIR_BAR(bei));
3803 
3804 	/* Check ROM */
3805 	if (bei == PCIM_EA_BEI_ROM)
3806 		return (PCIR_BIOS);
3807 
3808 #ifdef PCI_IOV
3809 	/* Check if matches VF_BAR */
3810 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3811 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3812 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3813 		    iov_pos);
3814 #endif
3815 
3816 	return (-1);
3817 }
3818 
3819 int
3820 pci_ea_is_enabled(device_t dev, int rid)
3821 {
3822 	struct pci_ea_entry *ea;
3823 	struct pci_devinfo *dinfo;
3824 
3825 	dinfo = device_get_ivars(dev);
3826 
3827 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3828 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3829 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3830 	}
3831 
3832 	return (0);
3833 }
3834 
3835 void
3836 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3837 {
3838 	struct pci_ea_entry *ea;
3839 	struct pci_devinfo *dinfo;
3840 	pci_addr_t start, end, count;
3841 	struct resource_list *rl;
3842 	int type, flags, rid;
3843 	struct resource *res;
3844 	uint32_t tmp;
3845 #ifdef PCI_IOV
3846 	struct pcicfg_iov *iov;
3847 #endif
3848 
3849 	dinfo = device_get_ivars(dev);
3850 	rl = &dinfo->resources;
3851 	flags = 0;
3852 
3853 #ifdef PCI_IOV
3854 	iov = dinfo->cfg.iov;
3855 #endif
3856 
3857 	if (dinfo->cfg.ea.ea_location == 0)
3858 		return;
3859 
3860 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
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 		/* Populate pd_name and pd_unit */
5260 		name = NULL;
5261 		if (dinfo->cfg.dev)
5262 			name = device_get_name(dinfo->cfg.dev);
5263 
5264 		p = &dinfo->conf;
5265 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5266 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5267 			(name && *name) ? name : "none",
5268 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5269 			none_count++,
5270 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5271 			p->pc_sel.pc_func, (p->pc_class << 16) |
5272 			(p->pc_subclass << 8) | p->pc_progif,
5273 			(p->pc_subdevice << 16) | p->pc_subvendor,
5274 			(p->pc_device << 16) | p->pc_vendor,
5275 			p->pc_revid, p->pc_hdr);
5276 	}
5277 }
5278 #endif /* DDB */
5279 
5280 static struct resource *
5281 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5282     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5283     u_int flags)
5284 {
5285 	struct pci_devinfo *dinfo = device_get_ivars(child);
5286 	struct resource_list *rl = &dinfo->resources;
5287 	struct resource *res;
5288 	struct pci_map *pm;
5289 	uint16_t cmd;
5290 	pci_addr_t map, testval;
5291 	int mapsize;
5292 
5293 	res = NULL;
5294 
5295 	/* If rid is managed by EA, ignore it */
5296 	if (pci_ea_is_enabled(child, *rid))
5297 		goto out;
5298 
5299 	pm = pci_find_bar(child, *rid);
5300 	if (pm != NULL) {
5301 		/* This is a BAR that we failed to allocate earlier. */
5302 		mapsize = pm->pm_size;
5303 		map = pm->pm_value;
5304 	} else {
5305 		/*
5306 		 * Weed out the bogons, and figure out how large the
5307 		 * BAR/map is.  BARs that read back 0 here are bogus
5308 		 * and unimplemented.  Note: atapci in legacy mode are
5309 		 * special and handled elsewhere in the code.  If you
5310 		 * have a atapci device in legacy mode and it fails
5311 		 * here, that other code is broken.
5312 		 */
5313 		pci_read_bar(child, *rid, &map, &testval, NULL);
5314 
5315 		/*
5316 		 * Determine the size of the BAR and ignore BARs with a size
5317 		 * of 0.  Device ROM BARs use a different mask value.
5318 		 */
5319 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5320 			mapsize = pci_romsize(testval);
5321 		else
5322 			mapsize = pci_mapsize(testval);
5323 		if (mapsize == 0)
5324 			goto out;
5325 		pm = pci_add_bar(child, *rid, map, mapsize);
5326 	}
5327 
5328 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5329 		if (type != SYS_RES_MEMORY) {
5330 			if (bootverbose)
5331 				device_printf(dev,
5332 				    "child %s requested type %d for rid %#x,"
5333 				    " but the BAR says it is an memio\n",
5334 				    device_get_nameunit(child), type, *rid);
5335 			goto out;
5336 		}
5337 	} else {
5338 		if (type != SYS_RES_IOPORT) {
5339 			if (bootverbose)
5340 				device_printf(dev,
5341 				    "child %s requested type %d for rid %#x,"
5342 				    " but the BAR says it is an ioport\n",
5343 				    device_get_nameunit(child), type, *rid);
5344 			goto out;
5345 		}
5346 	}
5347 
5348 	/*
5349 	 * For real BARs, we need to override the size that
5350 	 * the driver requests, because that's what the BAR
5351 	 * actually uses and we would otherwise have a
5352 	 * situation where we might allocate the excess to
5353 	 * another driver, which won't work.
5354 	 */
5355 	count = ((pci_addr_t)1 << mapsize) * num;
5356 	if (RF_ALIGNMENT(flags) < mapsize)
5357 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5358 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5359 		flags |= RF_PREFETCHABLE;
5360 
5361 	/*
5362 	 * Allocate enough resource, and then write back the
5363 	 * appropriate BAR for that resource.
5364 	 */
5365 	resource_list_add(rl, type, *rid, start, end, count);
5366 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5367 	    count, flags & ~RF_ACTIVE);
5368 	if (res == NULL) {
5369 		resource_list_delete(rl, type, *rid);
5370 		device_printf(child,
5371 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5372 		    count, *rid, type, start, end);
5373 		goto out;
5374 	}
5375 	if (bootverbose)
5376 		device_printf(child,
5377 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5378 		    count, *rid, type, rman_get_start(res));
5379 
5380 	/* Disable decoding via the CMD register before updating the BAR */
5381 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5382 	pci_write_config(child, PCIR_COMMAND,
5383 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5384 
5385 	map = rman_get_start(res);
5386 	pci_write_bar(child, pm, map);
5387 
5388 	/* Restore the original value of the CMD register */
5389 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5390 out:
5391 	return (res);
5392 }
5393 
5394 struct resource *
5395 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5396     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5397     u_int flags)
5398 {
5399 	struct pci_devinfo *dinfo;
5400 	struct resource_list *rl;
5401 	struct resource_list_entry *rle;
5402 	struct resource *res;
5403 	pcicfgregs *cfg;
5404 
5405 	/*
5406 	 * Perform lazy resource allocation
5407 	 */
5408 	dinfo = device_get_ivars(child);
5409 	rl = &dinfo->resources;
5410 	cfg = &dinfo->cfg;
5411 	switch (type) {
5412 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5413 	case PCI_RES_BUS:
5414 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5415 		    flags));
5416 #endif
5417 	case SYS_RES_IRQ:
5418 		/*
5419 		 * Can't alloc legacy interrupt once MSI messages have
5420 		 * been allocated.
5421 		 */
5422 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5423 		    cfg->msix.msix_alloc > 0))
5424 			return (NULL);
5425 
5426 		/*
5427 		 * If the child device doesn't have an interrupt
5428 		 * routed and is deserving of an interrupt, try to
5429 		 * assign it one.
5430 		 */
5431 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5432 		    (cfg->intpin != 0))
5433 			pci_assign_interrupt(dev, child, 0);
5434 		break;
5435 	case SYS_RES_IOPORT:
5436 	case SYS_RES_MEMORY:
5437 #ifdef NEW_PCIB
5438 		/*
5439 		 * PCI-PCI bridge I/O window resources are not BARs.
5440 		 * For those allocations just pass the request up the
5441 		 * tree.
5442 		 */
5443 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5444 			switch (*rid) {
5445 			case PCIR_IOBASEL_1:
5446 			case PCIR_MEMBASE_1:
5447 			case PCIR_PMBASEL_1:
5448 				/*
5449 				 * XXX: Should we bother creating a resource
5450 				 * list entry?
5451 				 */
5452 				return (bus_generic_alloc_resource(dev, child,
5453 				    type, rid, start, end, count, flags));
5454 			}
5455 		}
5456 #endif
5457 		/* Reserve resources for this BAR if needed. */
5458 		rle = resource_list_find(rl, type, *rid);
5459 		if (rle == NULL) {
5460 			res = pci_reserve_map(dev, child, type, rid, start, end,
5461 			    count, num, flags);
5462 			if (res == NULL)
5463 				return (NULL);
5464 		}
5465 	}
5466 	return (resource_list_alloc(rl, dev, child, type, rid,
5467 	    start, end, count, flags));
5468 }
5469 
5470 struct resource *
5471 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5472     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5473 {
5474 #ifdef PCI_IOV
5475 	struct pci_devinfo *dinfo;
5476 #endif
5477 
5478 	if (device_get_parent(child) != dev)
5479 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5480 		    type, rid, start, end, count, flags));
5481 
5482 #ifdef PCI_IOV
5483 	dinfo = device_get_ivars(child);
5484 	if (dinfo->cfg.flags & PCICFG_VF) {
5485 		switch (type) {
5486 		/* VFs can't have I/O BARs. */
5487 		case SYS_RES_IOPORT:
5488 			return (NULL);
5489 		case SYS_RES_MEMORY:
5490 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5491 			    start, end, count, flags));
5492 		}
5493 
5494 		/* Fall through for other types of resource allocations. */
5495 	}
5496 #endif
5497 
5498 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5499 	    count, 1, flags));
5500 }
5501 
5502 int
5503 pci_release_resource(device_t dev, device_t child, int type, int rid,
5504     struct resource *r)
5505 {
5506 	struct pci_devinfo *dinfo;
5507 	struct resource_list *rl;
5508 	pcicfgregs *cfg;
5509 
5510 	if (device_get_parent(child) != dev)
5511 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5512 		    type, rid, r));
5513 
5514 	dinfo = device_get_ivars(child);
5515 	cfg = &dinfo->cfg;
5516 
5517 #ifdef PCI_IOV
5518 	if (dinfo->cfg.flags & PCICFG_VF) {
5519 		switch (type) {
5520 		/* VFs can't have I/O BARs. */
5521 		case SYS_RES_IOPORT:
5522 			return (EDOOFUS);
5523 		case SYS_RES_MEMORY:
5524 			return (pci_vf_release_mem_resource(dev, child, rid,
5525 			    r));
5526 		}
5527 
5528 		/* Fall through for other types of resource allocations. */
5529 	}
5530 #endif
5531 
5532 #ifdef NEW_PCIB
5533 	/*
5534 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5535 	 * those allocations just pass the request up the tree.
5536 	 */
5537 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5538 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5539 		switch (rid) {
5540 		case PCIR_IOBASEL_1:
5541 		case PCIR_MEMBASE_1:
5542 		case PCIR_PMBASEL_1:
5543 			return (bus_generic_release_resource(dev, child, type,
5544 			    rid, r));
5545 		}
5546 	}
5547 #endif
5548 
5549 	rl = &dinfo->resources;
5550 	return (resource_list_release(rl, dev, child, type, rid, r));
5551 }
5552 
5553 int
5554 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5555     struct resource *r)
5556 {
5557 	struct pci_devinfo *dinfo;
5558 	int error;
5559 
5560 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5561 	if (error)
5562 		return (error);
5563 
5564 	/* Enable decoding in the command register when activating BARs. */
5565 	if (device_get_parent(child) == dev) {
5566 		/* Device ROMs need their decoding explicitly enabled. */
5567 		dinfo = device_get_ivars(child);
5568 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5569 			pci_write_bar(child, pci_find_bar(child, rid),
5570 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5571 		switch (type) {
5572 		case SYS_RES_IOPORT:
5573 		case SYS_RES_MEMORY:
5574 			error = PCI_ENABLE_IO(dev, child, type);
5575 			break;
5576 		}
5577 	}
5578 	return (error);
5579 }
5580 
5581 int
5582 pci_deactivate_resource(device_t dev, device_t child, int type,
5583     int rid, struct resource *r)
5584 {
5585 	struct pci_devinfo *dinfo;
5586 	int error;
5587 
5588 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5589 	if (error)
5590 		return (error);
5591 
5592 	/* Disable decoding for device ROMs. */
5593 	if (device_get_parent(child) == dev) {
5594 		dinfo = device_get_ivars(child);
5595 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5596 			pci_write_bar(child, pci_find_bar(child, rid),
5597 			    rman_get_start(r));
5598 	}
5599 	return (0);
5600 }
5601 
5602 void
5603 pci_child_deleted(device_t dev, device_t child)
5604 {
5605 	struct resource_list_entry *rle;
5606 	struct resource_list *rl;
5607 	struct pci_devinfo *dinfo;
5608 
5609 	dinfo = device_get_ivars(child);
5610 	rl = &dinfo->resources;
5611 
5612 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5613 
5614 	/* Turn off access to resources we're about to free */
5615 	if (bus_child_present(child) != 0) {
5616 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5617 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5618 
5619 		pci_disable_busmaster(child);
5620 	}
5621 
5622 	/* Free all allocated resources */
5623 	STAILQ_FOREACH(rle, rl, link) {
5624 		if (rle->res) {
5625 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5626 			    resource_list_busy(rl, rle->type, rle->rid)) {
5627 				pci_printf(&dinfo->cfg,
5628 				    "Resource still owned, oops. "
5629 				    "(type=%d, rid=%d, addr=%lx)\n",
5630 				    rle->type, rle->rid,
5631 				    rman_get_start(rle->res));
5632 				bus_release_resource(child, rle->type, rle->rid,
5633 				    rle->res);
5634 			}
5635 			resource_list_unreserve(rl, dev, child, rle->type,
5636 			    rle->rid);
5637 		}
5638 	}
5639 	resource_list_free(rl);
5640 
5641 	pci_freecfg(dinfo);
5642 }
5643 
5644 void
5645 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5646 {
5647 	struct pci_devinfo *dinfo;
5648 	struct resource_list *rl;
5649 	struct resource_list_entry *rle;
5650 
5651 	if (device_get_parent(child) != dev)
5652 		return;
5653 
5654 	dinfo = device_get_ivars(child);
5655 	rl = &dinfo->resources;
5656 	rle = resource_list_find(rl, type, rid);
5657 	if (rle == NULL)
5658 		return;
5659 
5660 	if (rle->res) {
5661 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5662 		    resource_list_busy(rl, type, rid)) {
5663 			device_printf(dev, "delete_resource: "
5664 			    "Resource still owned by child, oops. "
5665 			    "(type=%d, rid=%d, addr=%jx)\n",
5666 			    type, rid, rman_get_start(rle->res));
5667 			return;
5668 		}
5669 		resource_list_unreserve(rl, dev, child, type, rid);
5670 	}
5671 	resource_list_delete(rl, type, rid);
5672 }
5673 
5674 struct resource_list *
5675 pci_get_resource_list (device_t dev, device_t child)
5676 {
5677 	struct pci_devinfo *dinfo = device_get_ivars(child);
5678 
5679 	return (&dinfo->resources);
5680 }
5681 
5682 #ifdef ACPI_DMAR
5683 bus_dma_tag_t acpi_iommu_get_dma_tag(device_t dev, device_t child);
5684 bus_dma_tag_t
5685 pci_get_dma_tag(device_t bus, device_t dev)
5686 {
5687 	bus_dma_tag_t tag;
5688 	struct pci_softc *sc;
5689 
5690 	if (device_get_parent(dev) == bus) {
5691 		/* try iommu and return if it works */
5692 		tag = acpi_iommu_get_dma_tag(bus, dev);
5693 	} else
5694 		tag = NULL;
5695 	if (tag == NULL) {
5696 		sc = device_get_softc(bus);
5697 		tag = sc->sc_dma_tag;
5698 	}
5699 	return (tag);
5700 }
5701 #else
5702 bus_dma_tag_t
5703 pci_get_dma_tag(device_t bus, device_t dev)
5704 {
5705 	struct pci_softc *sc = device_get_softc(bus);
5706 
5707 	return (sc->sc_dma_tag);
5708 }
5709 #endif
5710 
5711 uint32_t
5712 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5713 {
5714 	struct pci_devinfo *dinfo = device_get_ivars(child);
5715 	pcicfgregs *cfg = &dinfo->cfg;
5716 
5717 #ifdef PCI_IOV
5718 	/*
5719 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5720 	 * emulate them here.
5721 	 */
5722 	if (cfg->flags & PCICFG_VF) {
5723 		if (reg == PCIR_VENDOR) {
5724 			switch (width) {
5725 			case 4:
5726 				return (cfg->device << 16 | cfg->vendor);
5727 			case 2:
5728 				return (cfg->vendor);
5729 			case 1:
5730 				return (cfg->vendor & 0xff);
5731 			default:
5732 				return (0xffffffff);
5733 			}
5734 		} else if (reg == PCIR_DEVICE) {
5735 			switch (width) {
5736 			/* Note that an unaligned 4-byte read is an error. */
5737 			case 2:
5738 				return (cfg->device);
5739 			case 1:
5740 				return (cfg->device & 0xff);
5741 			default:
5742 				return (0xffffffff);
5743 			}
5744 		}
5745 	}
5746 #endif
5747 
5748 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5749 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5750 }
5751 
5752 void
5753 pci_write_config_method(device_t dev, device_t child, int reg,
5754     uint32_t val, int width)
5755 {
5756 	struct pci_devinfo *dinfo = device_get_ivars(child);
5757 	pcicfgregs *cfg = &dinfo->cfg;
5758 
5759 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5760 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5761 }
5762 
5763 int
5764 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5765     size_t buflen)
5766 {
5767 
5768 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5769 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5770 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5771 	return (0);
5772 }
5773 
5774 int
5775 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5776     size_t buflen)
5777 {
5778 	struct pci_devinfo *dinfo;
5779 	pcicfgregs *cfg;
5780 
5781 	dinfo = device_get_ivars(child);
5782 	cfg = &dinfo->cfg;
5783 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5784 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5785 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5786 	    cfg->progif);
5787 	return (0);
5788 }
5789 
5790 int
5791 pci_assign_interrupt_method(device_t dev, device_t child)
5792 {
5793 	struct pci_devinfo *dinfo = device_get_ivars(child);
5794 	pcicfgregs *cfg = &dinfo->cfg;
5795 
5796 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5797 	    cfg->intpin));
5798 }
5799 
5800 static void
5801 pci_lookup(void *arg, const char *name, device_t *dev)
5802 {
5803 	long val;
5804 	char *end;
5805 	int domain, bus, slot, func;
5806 
5807 	if (*dev != NULL)
5808 		return;
5809 
5810 	/*
5811 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5812 	 * pciB:S:F.  In the latter case, the domain is assumed to
5813 	 * be zero.
5814 	 */
5815 	if (strncmp(name, "pci", 3) != 0)
5816 		return;
5817 	val = strtol(name + 3, &end, 10);
5818 	if (val < 0 || val > INT_MAX || *end != ':')
5819 		return;
5820 	domain = val;
5821 	val = strtol(end + 1, &end, 10);
5822 	if (val < 0 || val > INT_MAX || *end != ':')
5823 		return;
5824 	bus = val;
5825 	val = strtol(end + 1, &end, 10);
5826 	if (val < 0 || val > INT_MAX)
5827 		return;
5828 	slot = val;
5829 	if (*end == ':') {
5830 		val = strtol(end + 1, &end, 10);
5831 		if (val < 0 || val > INT_MAX || *end != '\0')
5832 			return;
5833 		func = val;
5834 	} else if (*end == '\0') {
5835 		func = slot;
5836 		slot = bus;
5837 		bus = domain;
5838 		domain = 0;
5839 	} else
5840 		return;
5841 
5842 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5843 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5844 		return;
5845 
5846 	*dev = pci_find_dbsf(domain, bus, slot, func);
5847 }
5848 
5849 static int
5850 pci_modevent(module_t mod, int what, void *arg)
5851 {
5852 	static struct cdev *pci_cdev;
5853 	static eventhandler_tag tag;
5854 
5855 	switch (what) {
5856 	case MOD_LOAD:
5857 		STAILQ_INIT(&pci_devq);
5858 		pci_generation = 0;
5859 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5860 		    "pci");
5861 		pci_load_vendor_data();
5862 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5863 		    1000);
5864 		break;
5865 
5866 	case MOD_UNLOAD:
5867 		if (tag != NULL)
5868 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5869 		destroy_dev(pci_cdev);
5870 		break;
5871 	}
5872 
5873 	return (0);
5874 }
5875 
5876 static void
5877 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5878 {
5879 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5880 	struct pcicfg_pcie *cfg;
5881 	int version, pos;
5882 
5883 	cfg = &dinfo->cfg.pcie;
5884 	pos = cfg->pcie_location;
5885 
5886 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5887 
5888 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5889 
5890 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5891 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5892 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5893 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5894 
5895 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5896 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5897 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5898 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5899 
5900 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5901 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5902 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5903 
5904 	if (version > 1) {
5905 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5906 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5907 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5908 	}
5909 #undef WREG
5910 }
5911 
5912 static void
5913 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5914 {
5915 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5916 	    dinfo->cfg.pcix.pcix_command,  2);
5917 }
5918 
5919 void
5920 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5921 {
5922 
5923 	/*
5924 	 * Restore the device to full power mode.  We must do this
5925 	 * before we restore the registers because moving from D3 to
5926 	 * D0 will cause the chip's BARs and some other registers to
5927 	 * be reset to some unknown power on reset values.  Cut down
5928 	 * the noise on boot by doing nothing if we are already in
5929 	 * state D0.
5930 	 */
5931 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5932 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5933 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5934 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5935 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5936 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5937 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5938 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5939 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5940 	case PCIM_HDRTYPE_NORMAL:
5941 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5942 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5943 		break;
5944 	case PCIM_HDRTYPE_BRIDGE:
5945 		pci_write_config(dev, PCIR_SECLAT_1,
5946 		    dinfo->cfg.bridge.br_seclat, 1);
5947 		pci_write_config(dev, PCIR_SUBBUS_1,
5948 		    dinfo->cfg.bridge.br_subbus, 1);
5949 		pci_write_config(dev, PCIR_SECBUS_1,
5950 		    dinfo->cfg.bridge.br_secbus, 1);
5951 		pci_write_config(dev, PCIR_PRIBUS_1,
5952 		    dinfo->cfg.bridge.br_pribus, 1);
5953 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5954 		    dinfo->cfg.bridge.br_control, 2);
5955 		break;
5956 	case PCIM_HDRTYPE_CARDBUS:
5957 		pci_write_config(dev, PCIR_SECLAT_2,
5958 		    dinfo->cfg.bridge.br_seclat, 1);
5959 		pci_write_config(dev, PCIR_SUBBUS_2,
5960 		    dinfo->cfg.bridge.br_subbus, 1);
5961 		pci_write_config(dev, PCIR_SECBUS_2,
5962 		    dinfo->cfg.bridge.br_secbus, 1);
5963 		pci_write_config(dev, PCIR_PRIBUS_2,
5964 		    dinfo->cfg.bridge.br_pribus, 1);
5965 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5966 		    dinfo->cfg.bridge.br_control, 2);
5967 		break;
5968 	}
5969 	pci_restore_bars(dev);
5970 
5971 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
5972 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5973 
5974 	/*
5975 	 * Restore extended capabilities for PCI-Express and PCI-X
5976 	 */
5977 	if (dinfo->cfg.pcie.pcie_location != 0)
5978 		pci_cfg_restore_pcie(dev, dinfo);
5979 	if (dinfo->cfg.pcix.pcix_location != 0)
5980 		pci_cfg_restore_pcix(dev, dinfo);
5981 
5982 	/* Restore MSI and MSI-X configurations if they are present. */
5983 	if (dinfo->cfg.msi.msi_location != 0)
5984 		pci_resume_msi(dev);
5985 	if (dinfo->cfg.msix.msix_location != 0)
5986 		pci_resume_msix(dev);
5987 
5988 #ifdef PCI_IOV
5989 	if (dinfo->cfg.iov != NULL)
5990 		pci_iov_cfg_restore(dev, dinfo);
5991 #endif
5992 }
5993 
5994 static void
5995 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
5996 {
5997 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
5998 	struct pcicfg_pcie *cfg;
5999 	int version, pos;
6000 
6001 	cfg = &dinfo->cfg.pcie;
6002 	pos = cfg->pcie_location;
6003 
6004 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6005 
6006 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6007 
6008 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6009 
6010 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6011 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6012 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6013 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6014 
6015 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6016 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6017 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6018 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6019 
6020 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6021 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6022 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6023 
6024 	if (version > 1) {
6025 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6026 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6027 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6028 	}
6029 #undef RREG
6030 }
6031 
6032 static void
6033 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6034 {
6035 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6036 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6037 }
6038 
6039 void
6040 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6041 {
6042 	uint32_t cls;
6043 	int ps;
6044 
6045 	/*
6046 	 * Some drivers apparently write to these registers w/o updating our
6047 	 * cached copy.  No harm happens if we update the copy, so do so here
6048 	 * so we can restore them.  The COMMAND register is modified by the
6049 	 * bus w/o updating the cache.  This should represent the normally
6050 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6051 	 */
6052 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6053 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6054 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6055 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6056 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6057 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6058 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6059 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6060 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6061 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6062 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6063 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6064 	case PCIM_HDRTYPE_NORMAL:
6065 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6066 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6067 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6068 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6069 		break;
6070 	case PCIM_HDRTYPE_BRIDGE:
6071 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6072 		    PCIR_SECLAT_1, 1);
6073 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6074 		    PCIR_SUBBUS_1, 1);
6075 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6076 		    PCIR_SECBUS_1, 1);
6077 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6078 		    PCIR_PRIBUS_1, 1);
6079 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6080 		    PCIR_BRIDGECTL_1, 2);
6081 		break;
6082 	case PCIM_HDRTYPE_CARDBUS:
6083 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6084 		    PCIR_SECLAT_2, 1);
6085 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6086 		    PCIR_SUBBUS_2, 1);
6087 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6088 		    PCIR_SECBUS_2, 1);
6089 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6090 		    PCIR_PRIBUS_2, 1);
6091 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6092 		    PCIR_BRIDGECTL_2, 2);
6093 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6094 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6095 		break;
6096 	}
6097 
6098 	if (dinfo->cfg.pcie.pcie_location != 0)
6099 		pci_cfg_save_pcie(dev, dinfo);
6100 
6101 	if (dinfo->cfg.pcix.pcix_location != 0)
6102 		pci_cfg_save_pcix(dev, dinfo);
6103 
6104 #ifdef PCI_IOV
6105 	if (dinfo->cfg.iov != NULL)
6106 		pci_iov_cfg_save(dev, dinfo);
6107 #endif
6108 
6109 	/*
6110 	 * don't set the state for display devices, base peripherals and
6111 	 * memory devices since bad things happen when they are powered down.
6112 	 * We should (a) have drivers that can easily detach and (b) use
6113 	 * generic drivers for these devices so that some device actually
6114 	 * attaches.  We need to make sure that when we implement (a) we don't
6115 	 * power the device down on a reattach.
6116 	 */
6117 	cls = pci_get_class(dev);
6118 	if (!setstate)
6119 		return;
6120 	switch (pci_do_power_nodriver)
6121 	{
6122 		case 0:		/* NO powerdown at all */
6123 			return;
6124 		case 1:		/* Conservative about what to power down */
6125 			if (cls == PCIC_STORAGE)
6126 				return;
6127 			/*FALLTHROUGH*/
6128 		case 2:		/* Aggressive about what to power down */
6129 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6130 			    cls == PCIC_BASEPERIPH)
6131 				return;
6132 			/*FALLTHROUGH*/
6133 		case 3:		/* Power down everything */
6134 			break;
6135 	}
6136 	/*
6137 	 * PCI spec says we can only go into D3 state from D0 state.
6138 	 * Transition from D[12] into D0 before going to D3 state.
6139 	 */
6140 	ps = pci_get_powerstate(dev);
6141 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6142 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6143 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6144 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6145 }
6146 
6147 /* Wrapper APIs suitable for device driver use. */
6148 void
6149 pci_save_state(device_t dev)
6150 {
6151 	struct pci_devinfo *dinfo;
6152 
6153 	dinfo = device_get_ivars(dev);
6154 	pci_cfg_save(dev, dinfo, 0);
6155 }
6156 
6157 void
6158 pci_restore_state(device_t dev)
6159 {
6160 	struct pci_devinfo *dinfo;
6161 
6162 	dinfo = device_get_ivars(dev);
6163 	pci_cfg_restore(dev, dinfo);
6164 }
6165 
6166 static int
6167 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6168     uintptr_t *id)
6169 {
6170 
6171 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6172 }
6173 
6174 /* Find the upstream port of a given PCI device in a root complex. */
6175 device_t
6176 pci_find_pcie_root_port(device_t dev)
6177 {
6178 	struct pci_devinfo *dinfo;
6179 	devclass_t pci_class;
6180 	device_t pcib, bus;
6181 
6182 	pci_class = devclass_find("pci");
6183 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6184 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6185 
6186 	/*
6187 	 * Walk the bridge hierarchy until we find a PCI-e root
6188 	 * port or a non-PCI device.
6189 	 */
6190 	for (;;) {
6191 		bus = device_get_parent(dev);
6192 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6193 		    device_get_nameunit(dev)));
6194 
6195 		pcib = device_get_parent(bus);
6196 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6197 		    device_get_nameunit(bus)));
6198 
6199 		/*
6200 		 * pcib's parent must be a PCI bus for this to be a
6201 		 * PCI-PCI bridge.
6202 		 */
6203 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6204 			return (NULL);
6205 
6206 		dinfo = device_get_ivars(pcib);
6207 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6208 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6209 			return (pcib);
6210 
6211 		dev = pcib;
6212 	}
6213 }
6214 
6215 /*
6216  * Wait for pending transactions to complete on a PCI-express function.
6217  *
6218  * The maximum delay is specified in milliseconds in max_delay.  Note
6219  * that this function may sleep.
6220  *
6221  * Returns true if the function is idle and false if the timeout is
6222  * exceeded.  If dev is not a PCI-express function, this returns true.
6223  */
6224 bool
6225 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6226 {
6227 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6228 	uint16_t sta;
6229 	int cap;
6230 
6231 	cap = dinfo->cfg.pcie.pcie_location;
6232 	if (cap == 0)
6233 		return (true);
6234 
6235 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6236 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6237 		if (max_delay == 0)
6238 			return (false);
6239 
6240 		/* Poll once every 100 milliseconds up to the timeout. */
6241 		if (max_delay > 100) {
6242 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6243 			max_delay -= 100;
6244 		} else {
6245 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6246 			    C_HARDCLOCK);
6247 			max_delay = 0;
6248 		}
6249 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6250 	}
6251 
6252 	return (true);
6253 }
6254 
6255 /*
6256  * Determine the maximum Completion Timeout in microseconds.
6257  *
6258  * For non-PCI-express functions this returns 0.
6259  */
6260 int
6261 pcie_get_max_completion_timeout(device_t dev)
6262 {
6263 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6264 	int cap;
6265 
6266 	cap = dinfo->cfg.pcie.pcie_location;
6267 	if (cap == 0)
6268 		return (0);
6269 
6270 	/*
6271 	 * Functions using the 1.x spec use the default timeout range of
6272 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6273 	 * support programmable timeouts also use this range.
6274 	 */
6275 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6276 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6277 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6278 		return (50 * 1000);
6279 
6280 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6281 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6282 	case PCIEM_CTL2_COMP_TIMO_100US:
6283 		return (100);
6284 	case PCIEM_CTL2_COMP_TIMO_10MS:
6285 		return (10 * 1000);
6286 	case PCIEM_CTL2_COMP_TIMO_55MS:
6287 		return (55 * 1000);
6288 	case PCIEM_CTL2_COMP_TIMO_210MS:
6289 		return (210 * 1000);
6290 	case PCIEM_CTL2_COMP_TIMO_900MS:
6291 		return (900 * 1000);
6292 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6293 		return (3500 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_13S:
6295 		return (13 * 1000 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_64S:
6297 		return (64 * 1000 * 1000);
6298 	default:
6299 		return (50 * 1000);
6300 	}
6301 }
6302 
6303 void
6304 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6305 {
6306 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6307 	const char *s;
6308 	int aer;
6309 	uint32_t r, r1;
6310 	uint16_t rs;
6311 
6312 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6313 		s = "Correctable";
6314 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6315 		s = "Uncorrectable (Non-Fatal)";
6316 	else
6317 		s = "Uncorrectable (Fatal)";
6318 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6319 	if (aerp) {
6320 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6321 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6322 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6323 		} else {
6324 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6325 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6326 		}
6327 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6328 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6329 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6330 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6331 			printf(" severity 0x%08x first %d\n",
6332 			    r, rs & 0x1f);
6333 		} else
6334 			printf("\n");
6335 	}
6336 
6337 	/* As kind of recovery just report and clear the error statuses. */
6338 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6339 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6340 		if (r != 0) {
6341 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6342 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6343 		}
6344 
6345 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6346 		if (r != 0) {
6347 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6348 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6349 		}
6350 	}
6351 	if (dinfo->cfg.pcie.pcie_location != 0) {
6352 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6353 		    PCIER_DEVICE_STA, 2);
6354 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6355 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6356 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6357 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6358 			    PCIER_DEVICE_STA, rs, 2);
6359 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6360 		}
6361 	}
6362 }
6363 
6364 /*
6365  * Perform a Function Level Reset (FLR) on a device.
6366  *
6367  * This function first waits for any pending transactions to complete
6368  * within the timeout specified by max_delay.  If transactions are
6369  * still pending, the function will return false without attempting a
6370  * reset.
6371  *
6372  * If dev is not a PCI-express function or does not support FLR, this
6373  * function returns false.
6374  *
6375  * Note that no registers are saved or restored.  The caller is
6376  * responsible for saving and restoring any registers including
6377  * PCI-standard registers via pci_save_state() and
6378  * pci_restore_state().
6379  */
6380 bool
6381 pcie_flr(device_t dev, u_int max_delay, bool force)
6382 {
6383 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6384 	uint16_t cmd, ctl;
6385 	int compl_delay;
6386 	int cap;
6387 
6388 	cap = dinfo->cfg.pcie.pcie_location;
6389 	if (cap == 0)
6390 		return (false);
6391 
6392 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6393 		return (false);
6394 
6395 	/*
6396 	 * Disable busmastering to prevent generation of new
6397 	 * transactions while waiting for the device to go idle.  If
6398 	 * the idle timeout fails, the command register is restored
6399 	 * which will re-enable busmastering.
6400 	 */
6401 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6402 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6403 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6404 		if (!force) {
6405 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6406 			return (false);
6407 		}
6408 		pci_printf(&dinfo->cfg,
6409 		    "Resetting with transactions pending after %d ms\n",
6410 		    max_delay);
6411 
6412 		/*
6413 		 * Extend the post-FLR delay to cover the maximum
6414 		 * Completion Timeout delay of anything in flight
6415 		 * during the FLR delay.  Enforce a minimum delay of
6416 		 * at least 10ms.
6417 		 */
6418 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6419 		if (compl_delay < 10)
6420 			compl_delay = 10;
6421 	} else
6422 		compl_delay = 0;
6423 
6424 	/* Initiate the reset. */
6425 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6426 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6427 	    PCIEM_CTL_INITIATE_FLR, 2);
6428 
6429 	/* Wait for 100ms. */
6430 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6431 
6432 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6433 	    PCIEM_STA_TRANSACTION_PND)
6434 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6435 	return (true);
6436 }
6437 
6438 /*
6439  * Attempt a power-management reset by cycling the device in/out of D3
6440  * state.  PCI spec says we can only go into D3 state from D0 state.
6441  * Transition from D[12] into D0 before going to D3 state.
6442  */
6443 int
6444 pci_power_reset(device_t dev)
6445 {
6446 	int ps;
6447 
6448 	ps = pci_get_powerstate(dev);
6449 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6450 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6451 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6452 	pci_set_powerstate(dev, ps);
6453 	return (0);
6454 }
6455 
6456 /*
6457  * Try link drop and retrain of the downstream port of upstream
6458  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6459  * cause Conventional Hot reset of the device in the slot.
6460  * Alternative, for PCIe, could be the secondary bus reset initiatied
6461  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6462  */
6463 int
6464 pcie_link_reset(device_t port, int pcie_location)
6465 {
6466 	uint16_t v;
6467 
6468 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6469 	v |= PCIEM_LINK_CTL_LINK_DIS;
6470 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6471 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6472 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6473 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6474 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6475 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6476 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6477 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6478 }
6479 
6480 static int
6481 pci_reset_post(device_t dev, device_t child)
6482 {
6483 
6484 	if (dev == device_get_parent(child))
6485 		pci_restore_state(child);
6486 	return (0);
6487 }
6488 
6489 static int
6490 pci_reset_prepare(device_t dev, device_t child)
6491 {
6492 
6493 	if (dev == device_get_parent(child))
6494 		pci_save_state(child);
6495 	return (0);
6496 }
6497 
6498 static int
6499 pci_reset_child(device_t dev, device_t child, int flags)
6500 {
6501 	int error;
6502 
6503 	if (dev == NULL || device_get_parent(child) != dev)
6504 		return (0);
6505 	if ((flags & DEVF_RESET_DETACH) != 0) {
6506 		error = device_get_state(child) == DS_ATTACHED ?
6507 		    device_detach(child) : 0;
6508 	} else {
6509 		error = BUS_SUSPEND_CHILD(dev, child);
6510 	}
6511 	if (error == 0) {
6512 		if (!pcie_flr(child, 1000, false)) {
6513 			error = BUS_RESET_PREPARE(dev, child);
6514 			if (error == 0)
6515 				pci_power_reset(child);
6516 			BUS_RESET_POST(dev, child);
6517 		}
6518 		if ((flags & DEVF_RESET_DETACH) != 0)
6519 			device_probe_and_attach(child);
6520 		else
6521 			BUS_RESUME_CHILD(dev, child);
6522 	}
6523 	return (error);
6524 }
6525 
6526 const struct pci_device_table *
6527 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6528 {
6529 	bool match;
6530 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6531 
6532 	vendor = pci_get_vendor(child);
6533 	device = pci_get_device(child);
6534 	subvendor = pci_get_subvendor(child);
6535 	subdevice = pci_get_subdevice(child);
6536 	class = pci_get_class(child);
6537 	subclass = pci_get_subclass(child);
6538 	revid = pci_get_revid(child);
6539 	while (nelt-- > 0) {
6540 		match = true;
6541 		if (id->match_flag_vendor)
6542 			match &= vendor == id->vendor;
6543 		if (id->match_flag_device)
6544 			match &= device == id->device;
6545 		if (id->match_flag_subvendor)
6546 			match &= subvendor == id->subvendor;
6547 		if (id->match_flag_subdevice)
6548 			match &= subdevice == id->subdevice;
6549 		if (id->match_flag_class)
6550 			match &= class == id->class_id;
6551 		if (id->match_flag_subclass)
6552 			match &= subclass == id->subclass;
6553 		if (id->match_flag_revid)
6554 			match &= revid == id->revid;
6555 		if (match)
6556 			return (id);
6557 		id++;
6558 	}
6559 	return (NULL);
6560 }
6561 
6562 static void
6563 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6564 {
6565 	const char *dev_name;
6566 	device_t dev;
6567 
6568 	dev = dinfo->cfg.dev;
6569 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6570 	    dinfo->cfg.slot, dinfo->cfg.func);
6571 	dev_name = device_get_name(dev);
6572 	if (dev_name != NULL)
6573 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6574 }
6575 
6576 void
6577 pci_print_faulted_dev(void)
6578 {
6579 	struct pci_devinfo *dinfo;
6580 	device_t dev;
6581 	int aer, i;
6582 	uint32_t r1, r2;
6583 	uint16_t status;
6584 
6585 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6586 		dev = dinfo->cfg.dev;
6587 		status = pci_read_config(dev, PCIR_STATUS, 2);
6588 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6589 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6590 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6591 		if (status != 0) {
6592 			pci_print_faulted_dev_name(dinfo);
6593 			printf(" error 0x%04x\n", status);
6594 		}
6595 		if (dinfo->cfg.pcie.pcie_location != 0) {
6596 			status = pci_read_config(dev,
6597 			    dinfo->cfg.pcie.pcie_location +
6598 			    PCIER_DEVICE_STA, 2);
6599 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6600 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6601 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6602 				pci_print_faulted_dev_name(dinfo);
6603 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6604 				    pci_read_config(dev,
6605 				    dinfo->cfg.pcie.pcie_location +
6606 				    PCIER_DEVICE_CTL, 2),
6607 				    status);
6608 			}
6609 		}
6610 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6611 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6612 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6613 			if (r1 != 0 || r2 != 0) {
6614 				pci_print_faulted_dev_name(dinfo);
6615 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6616 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6617 				    r1, pci_read_config(dev, aer +
6618 				    PCIR_AER_UC_MASK, 4),
6619 				    pci_read_config(dev, aer +
6620 				    PCIR_AER_UC_SEVERITY, 4),
6621 				    r2, pci_read_config(dev, aer +
6622 				    PCIR_AER_COR_MASK, 4),
6623 				    pci_read_config(dev, aer +
6624 				    PCIR_AER_CAP_CONTROL, 4));
6625 				for (i = 0; i < 4; i++) {
6626 					r1 = pci_read_config(dev, aer +
6627 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6628 					printf("    HL%d: 0x%08x\n", i, r1);
6629 				}
6630 			}
6631 		}
6632 	}
6633 }
6634 
6635 #ifdef DDB
6636 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6637 {
6638 
6639 	pci_print_faulted_dev();
6640 }
6641 
6642 static void
6643 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6644 {
6645 	device_t dev;
6646 	int aer;
6647 	uint32_t r;
6648 
6649 	dev = dinfo->cfg.dev;
6650 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6651 	    PCIER_DEVICE_STA, 2);
6652 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6653 	    PCIER_DEVICE_STA, r, 2);
6654 
6655 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6656 		return;
6657 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6658 	if (r != 0)
6659 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6660 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6661 	if (r != 0)
6662 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6663 }
6664 
6665 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6666 {
6667 	struct pci_devinfo *dinfo;
6668 	device_t dev;
6669 	uint16_t status, status1;
6670 
6671 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6672 		dev = dinfo->cfg.dev;
6673 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6674 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6675 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6676 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6677 		if (status1 != 0) {
6678 			status &= ~status1;
6679 			pci_write_config(dev, PCIR_STATUS, status, 2);
6680 		}
6681 		if (dinfo->cfg.pcie.pcie_location != 0)
6682 			db_clear_pcie_errors(dinfo);
6683 	}
6684 }
6685 #endif
6686