xref: /freebsd/sys/dev/pci/pci.c (revision 6419bb52)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_bus.h"
36 
37 #include <sys/param.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 
55 #include <sys/bus.h>
56 #include <machine/bus.h>
57 #include <sys/rman.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
60 
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
63 #endif
64 
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
69 
70 #ifdef PCI_IOV
71 #include <sys/nv.h>
72 #include <dev/pci/pci_iov_private.h>
73 #endif
74 
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
79 
80 #include "pcib_if.h"
81 #include "pci_if.h"
82 
83 #define	PCIR_IS_BIOS(cfg, reg)						\
84 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
85 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 
87 static int		pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t	pci_mapbase(uint64_t mapreg);
89 static const char	*pci_maptype(uint64_t mapreg);
90 static int		pci_maprange(uint64_t mapreg);
91 static pci_addr_t	pci_rombase(uint64_t mapreg);
92 static int		pci_romsize(uint64_t testval);
93 static void		pci_fixancient(pcicfgregs *cfg);
94 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 
96 static int		pci_porten(device_t dev);
97 static int		pci_memen(device_t dev);
98 static void		pci_assign_interrupt(device_t bus, device_t dev,
99 			    int force_route);
100 static int		pci_add_map(device_t bus, device_t dev, int reg,
101 			    struct resource_list *rl, int force, int prefetch);
102 static int		pci_probe(device_t dev);
103 static void		pci_load_vendor_data(void);
104 static int		pci_describe_parse_line(char **ptr, int *vendor,
105 			    int *device, char **desc);
106 static char		*pci_describe_device(device_t dev);
107 static int		pci_modevent(module_t mod, int what, void *arg);
108 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
109 			    pcicfgregs *cfg);
110 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
111 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
112 			    int reg, uint32_t *data);
113 #if 0
114 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
115 			    int reg, uint32_t data);
116 #endif
117 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
118 static void		pci_mask_msix(device_t dev, u_int index);
119 static void		pci_unmask_msix(device_t dev, u_int index);
120 static int		pci_msi_blacklisted(void);
121 static int		pci_msix_blacklisted(void);
122 static void		pci_resume_msi(device_t dev);
123 static void		pci_resume_msix(device_t dev);
124 static int		pci_remap_intr_method(device_t bus, device_t dev,
125 			    u_int irq);
126 static void		pci_hint_device_unit(device_t acdev, device_t child,
127 			    const char *name, int *unitp);
128 static int		pci_reset_post(device_t dev, device_t child);
129 static int		pci_reset_prepare(device_t dev, device_t child);
130 static int		pci_reset_child(device_t dev, device_t child,
131 			    int flags);
132 
133 static int		pci_get_id_method(device_t dev, device_t child,
134 			    enum pci_id_type type, uintptr_t *rid);
135 
136 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
137     int b, int s, int f, uint16_t vid, uint16_t did);
138 
139 static device_method_t pci_methods[] = {
140 	/* Device interface */
141 	DEVMETHOD(device_probe,		pci_probe),
142 	DEVMETHOD(device_attach,	pci_attach),
143 	DEVMETHOD(device_detach,	pci_detach),
144 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
145 	DEVMETHOD(device_suspend,	bus_generic_suspend),
146 	DEVMETHOD(device_resume,	pci_resume),
147 
148 	/* Bus interface */
149 	DEVMETHOD(bus_print_child,	pci_print_child),
150 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
151 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
152 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
153 	DEVMETHOD(bus_driver_added,	pci_driver_added),
154 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
155 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
156 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
157 	DEVMETHOD(bus_reset_post,	pci_reset_post),
158 	DEVMETHOD(bus_reset_child,	pci_reset_child),
159 
160 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
161 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
162 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
163 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
164 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
165 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
166 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
167 	DEVMETHOD(bus_release_resource,	pci_release_resource),
168 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
169 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
170 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
171 	DEVMETHOD(bus_child_detached,	pci_child_detached),
172 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
173 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
174 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
175 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
176 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
177 	DEVMETHOD(bus_resume_child,	pci_resume_child),
178 	DEVMETHOD(bus_rescan,		pci_rescan_method),
179 
180 	/* PCI interface */
181 	DEVMETHOD(pci_read_config,	pci_read_config_method),
182 	DEVMETHOD(pci_write_config,	pci_write_config_method),
183 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
184 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
185 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
186 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
187 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
188 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
189 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
190 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
191 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
192 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
193 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
194 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
195 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
196 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
197 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
198 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
199 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
200 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
201 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
202 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
203 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
204 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
205 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
206 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
207 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
208 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
209 	DEVMETHOD(pci_get_id,		pci_get_id_method),
210 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
211 	DEVMETHOD(pci_child_added,	pci_child_added_method),
212 #ifdef PCI_IOV
213 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
214 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
215 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
216 #endif
217 
218 	DEVMETHOD_END
219 };
220 
221 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
222 
223 static devclass_t pci_devclass;
224 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
225     BUS_PASS_BUS);
226 MODULE_VERSION(pci, 1);
227 
228 static char	*pci_vendordata;
229 static size_t	pci_vendordata_size;
230 
231 struct pci_quirk {
232 	uint32_t devid;	/* Vendor/device of the card */
233 	int	type;
234 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
235 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
236 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
237 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
238 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
239 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
240 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
241 	int	arg1;
242 	int	arg2;
243 };
244 
245 static const struct pci_quirk pci_quirks[] = {
246 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
247 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
248 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
249 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
250 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 
252 	/*
253 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
254 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
255 	 */
256 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
257 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
258 
259 	/*
260 	 * MSI doesn't work on earlier Intel chipsets including
261 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
262 	 */
263 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
264 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
265 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 
271 	/*
272 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
273 	 * bridge.
274 	 */
275 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
276 
277 	/*
278 	 * Some virtualization environments emulate an older chipset
279 	 * but support MSI just fine.  QEMU uses the Intel 82440.
280 	 */
281 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
282 
283 	/*
284 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
285 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
286 	 * It prevents us from attaching hpet(4) when the bit is unset.
287 	 * Note this quirk only affects SB600 revision A13 and earlier.
288 	 * For SB600 A21 and later, firmware must set the bit to hide it.
289 	 * For SB700 and later, it is unused and hardcoded to zero.
290 	 */
291 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
292 
293 	/*
294 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
295 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
296 	 * of the command register is set.
297 	 */
298 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
299 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
300 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 
304 	/*
305 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
306 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
307 	 */
308 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
309 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
310 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
311 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
312 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
313 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
314 
315 	/*
316 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
317 	 * expected place.
318 	 */
319 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
320 
321 	{ 0 }
322 };
323 
324 /* map register information */
325 #define	PCI_MAPMEM	0x01	/* memory map */
326 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
327 #define	PCI_MAPPORT	0x04	/* port map */
328 
329 struct devlist pci_devq;
330 uint32_t pci_generation;
331 uint32_t pci_numdevs = 0;
332 static int pcie_chipset, pcix_chipset;
333 
334 /* sysctl vars */
335 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
336     "PCI bus tuning parameters");
337 
338 static int pci_enable_io_modes = 1;
339 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
340     &pci_enable_io_modes, 1,
341     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
342     " enable these bits correctly.  We'd like to do this all the time, but"
343     " there are some peripherals that this causes problems with.");
344 
345 static int pci_do_realloc_bars = 1;
346 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
347     &pci_do_realloc_bars, 0,
348     "Attempt to allocate a new range for any BARs whose original "
349     "firmware-assigned ranges fail to allocate during the initial device scan.");
350 
351 static int pci_do_power_nodriver = 0;
352 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
353     &pci_do_power_nodriver, 0,
354     "Place a function into D3 state when no driver attaches to it.  0 means"
355     " disable.  1 means conservatively place devices into D3 state.  2 means"
356     " aggressively place devices into D3 state.  3 means put absolutely"
357     " everything in D3 state.");
358 
359 int pci_do_power_resume = 1;
360 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
361     &pci_do_power_resume, 1,
362   "Transition from D3 -> D0 on resume.");
363 
364 int pci_do_power_suspend = 1;
365 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
366     &pci_do_power_suspend, 1,
367   "Transition from D0 -> D3 on suspend.");
368 
369 static int pci_do_msi = 1;
370 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
371     "Enable support for MSI interrupts");
372 
373 static int pci_do_msix = 1;
374 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
375     "Enable support for MSI-X interrupts");
376 
377 static int pci_msix_rewrite_table = 0;
378 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
379     &pci_msix_rewrite_table, 0,
380     "Rewrite entire MSI-X table when updating MSI-X entries");
381 
382 static int pci_honor_msi_blacklist = 1;
383 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
384     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
385 
386 #if defined(__i386__) || defined(__amd64__)
387 static int pci_usb_takeover = 1;
388 #else
389 static int pci_usb_takeover = 0;
390 #endif
391 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
392     &pci_usb_takeover, 1,
393     "Enable early takeover of USB controllers. Disable this if you depend on"
394     " BIOS emulation of USB devices, that is you use USB devices (like"
395     " keyboard or mouse) but do not load USB drivers");
396 
397 static int pci_clear_bars;
398 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
399     "Ignore firmware-assigned resources for BARs.");
400 
401 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
402 static int pci_clear_buses;
403 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
404     "Ignore firmware-assigned bus numbers.");
405 #endif
406 
407 static int pci_enable_ari = 1;
408 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
409     0, "Enable support for PCIe Alternative RID Interpretation");
410 
411 static int pci_clear_aer_on_attach = 0;
412 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
413     &pci_clear_aer_on_attach, 0,
414     "Clear port and device AER state on driver attach");
415 
416 static int
417 pci_has_quirk(uint32_t devid, int quirk)
418 {
419 	const struct pci_quirk *q;
420 
421 	for (q = &pci_quirks[0]; q->devid; q++) {
422 		if (q->devid == devid && q->type == quirk)
423 			return (1);
424 	}
425 	return (0);
426 }
427 
428 /* Find a device_t by bus/slot/function in domain 0 */
429 
430 device_t
431 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
432 {
433 
434 	return (pci_find_dbsf(0, bus, slot, func));
435 }
436 
437 /* Find a device_t by domain/bus/slot/function */
438 
439 device_t
440 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
441 {
442 	struct pci_devinfo *dinfo = NULL;
443 
444 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
445 		if ((dinfo->cfg.domain == domain) &&
446 		    (dinfo->cfg.bus == bus) &&
447 		    (dinfo->cfg.slot == slot) &&
448 		    (dinfo->cfg.func == func)) {
449 			break;
450 		}
451 	}
452 
453 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
454 }
455 
456 /* Find a device_t by vendor/device ID */
457 
458 device_t
459 pci_find_device(uint16_t vendor, uint16_t device)
460 {
461 	struct pci_devinfo *dinfo;
462 
463 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
464 		if ((dinfo->cfg.vendor == vendor) &&
465 		    (dinfo->cfg.device == device)) {
466 			return (dinfo->cfg.dev);
467 		}
468 	}
469 
470 	return (NULL);
471 }
472 
473 device_t
474 pci_find_class(uint8_t class, uint8_t subclass)
475 {
476 	struct pci_devinfo *dinfo;
477 
478 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
479 		if (dinfo->cfg.baseclass == class &&
480 		    dinfo->cfg.subclass == subclass) {
481 			return (dinfo->cfg.dev);
482 		}
483 	}
484 
485 	return (NULL);
486 }
487 
488 static int
489 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
490 {
491 	va_list ap;
492 	int retval;
493 
494 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
495 	    cfg->func);
496 	va_start(ap, fmt);
497 	retval += vprintf(fmt, ap);
498 	va_end(ap);
499 	return (retval);
500 }
501 
502 /* return base address of memory or port map */
503 
504 static pci_addr_t
505 pci_mapbase(uint64_t mapreg)
506 {
507 
508 	if (PCI_BAR_MEM(mapreg))
509 		return (mapreg & PCIM_BAR_MEM_BASE);
510 	else
511 		return (mapreg & PCIM_BAR_IO_BASE);
512 }
513 
514 /* return map type of memory or port map */
515 
516 static const char *
517 pci_maptype(uint64_t mapreg)
518 {
519 
520 	if (PCI_BAR_IO(mapreg))
521 		return ("I/O Port");
522 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
523 		return ("Prefetchable Memory");
524 	return ("Memory");
525 }
526 
527 /* return log2 of map size decoded for memory or port map */
528 
529 int
530 pci_mapsize(uint64_t testval)
531 {
532 	int ln2size;
533 
534 	testval = pci_mapbase(testval);
535 	ln2size = 0;
536 	if (testval != 0) {
537 		while ((testval & 1) == 0)
538 		{
539 			ln2size++;
540 			testval >>= 1;
541 		}
542 	}
543 	return (ln2size);
544 }
545 
546 /* return base address of device ROM */
547 
548 static pci_addr_t
549 pci_rombase(uint64_t mapreg)
550 {
551 
552 	return (mapreg & PCIM_BIOS_ADDR_MASK);
553 }
554 
555 /* return log2 of map size decided for device ROM */
556 
557 static int
558 pci_romsize(uint64_t testval)
559 {
560 	int ln2size;
561 
562 	testval = pci_rombase(testval);
563 	ln2size = 0;
564 	if (testval != 0) {
565 		while ((testval & 1) == 0)
566 		{
567 			ln2size++;
568 			testval >>= 1;
569 		}
570 	}
571 	return (ln2size);
572 }
573 
574 /* return log2 of address range supported by map register */
575 
576 static int
577 pci_maprange(uint64_t mapreg)
578 {
579 	int ln2range = 0;
580 
581 	if (PCI_BAR_IO(mapreg))
582 		ln2range = 32;
583 	else
584 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
585 		case PCIM_BAR_MEM_32:
586 			ln2range = 32;
587 			break;
588 		case PCIM_BAR_MEM_1MB:
589 			ln2range = 20;
590 			break;
591 		case PCIM_BAR_MEM_64:
592 			ln2range = 64;
593 			break;
594 		}
595 	return (ln2range);
596 }
597 
598 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
599 
600 static void
601 pci_fixancient(pcicfgregs *cfg)
602 {
603 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
604 		return;
605 
606 	/* PCI to PCI bridges use header type 1 */
607 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
608 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
609 }
610 
611 /* extract header type specific config data */
612 
613 static void
614 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
615 {
616 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
617 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
618 	case PCIM_HDRTYPE_NORMAL:
619 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
620 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
621 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
622 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
623 		cfg->nummaps	    = PCI_MAXMAPS_0;
624 		break;
625 	case PCIM_HDRTYPE_BRIDGE:
626 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
627 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
628 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
629 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
630 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
631 		cfg->nummaps	    = PCI_MAXMAPS_1;
632 		break;
633 	case PCIM_HDRTYPE_CARDBUS:
634 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
635 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
636 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
637 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
638 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
639 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
640 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
641 		cfg->nummaps	    = PCI_MAXMAPS_2;
642 		break;
643 	}
644 #undef REG
645 }
646 
647 /* read configuration header into pcicfgregs structure */
648 struct pci_devinfo *
649 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
650 {
651 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
652 	uint16_t vid, did;
653 
654 	vid = REG(PCIR_VENDOR, 2);
655 	did = REG(PCIR_DEVICE, 2);
656 	if (vid != 0xffff)
657 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
658 
659 	return (NULL);
660 }
661 
662 struct pci_devinfo *
663 pci_alloc_devinfo_method(device_t dev)
664 {
665 
666 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
667 	    M_WAITOK | M_ZERO));
668 }
669 
670 static struct pci_devinfo *
671 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
672     uint16_t vid, uint16_t did)
673 {
674 	struct pci_devinfo *devlist_entry;
675 	pcicfgregs *cfg;
676 
677 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
678 
679 	cfg = &devlist_entry->cfg;
680 
681 	cfg->domain		= d;
682 	cfg->bus		= b;
683 	cfg->slot		= s;
684 	cfg->func		= f;
685 	cfg->vendor		= vid;
686 	cfg->device		= did;
687 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
688 	cfg->statreg		= REG(PCIR_STATUS, 2);
689 	cfg->baseclass		= REG(PCIR_CLASS, 1);
690 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
691 	cfg->progif		= REG(PCIR_PROGIF, 1);
692 	cfg->revid		= REG(PCIR_REVID, 1);
693 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
694 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
695 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
696 	cfg->intpin		= REG(PCIR_INTPIN, 1);
697 	cfg->intline		= REG(PCIR_INTLINE, 1);
698 
699 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
700 	cfg->hdrtype		&= ~PCIM_MFDEV;
701 	STAILQ_INIT(&cfg->maps);
702 
703 	cfg->iov		= NULL;
704 
705 	pci_fixancient(cfg);
706 	pci_hdrtypedata(pcib, b, s, f, cfg);
707 
708 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
709 		pci_read_cap(pcib, cfg);
710 
711 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
712 
713 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
714 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
715 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
716 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
717 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
718 
719 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
720 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
721 	devlist_entry->conf.pc_vendor = cfg->vendor;
722 	devlist_entry->conf.pc_device = cfg->device;
723 
724 	devlist_entry->conf.pc_class = cfg->baseclass;
725 	devlist_entry->conf.pc_subclass = cfg->subclass;
726 	devlist_entry->conf.pc_progif = cfg->progif;
727 	devlist_entry->conf.pc_revid = cfg->revid;
728 
729 	pci_numdevs++;
730 	pci_generation++;
731 
732 	return (devlist_entry);
733 }
734 #undef REG
735 
736 static void
737 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
738 {
739 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
740     cfg->ea.ea_location + (n), w)
741 	int num_ent;
742 	int ptr;
743 	int a, b;
744 	uint32_t val;
745 	int ent_size;
746 	uint32_t dw[4];
747 	uint64_t base, max_offset;
748 	struct pci_ea_entry *eae;
749 
750 	if (cfg->ea.ea_location == 0)
751 		return;
752 
753 	STAILQ_INIT(&cfg->ea.ea_entries);
754 
755 	/* Determine the number of entries */
756 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
757 	num_ent &= PCIM_EA_NUM_ENT_MASK;
758 
759 	/* Find the first entry to care of */
760 	ptr = PCIR_EA_FIRST_ENT;
761 
762 	/* Skip DWORD 2 for type 1 functions */
763 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
764 		ptr += 4;
765 
766 	for (a = 0; a < num_ent; a++) {
767 
768 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
769 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
770 
771 		/* Read a number of dwords in the entry */
772 		val = REG(ptr, 4);
773 		ptr += 4;
774 		ent_size = (val & PCIM_EA_ES);
775 
776 		for (b = 0; b < ent_size; b++) {
777 			dw[b] = REG(ptr, 4);
778 			ptr += 4;
779 		}
780 
781 		eae->eae_flags = val;
782 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
783 
784 		base = dw[0] & PCIM_EA_FIELD_MASK;
785 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
786 		b = 2;
787 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
788 			base |= (uint64_t)dw[b] << 32UL;
789 			b++;
790 		}
791 		if (((dw[1] & PCIM_EA_IS_64) != 0)
792 		    && (b < ent_size)) {
793 			max_offset |= (uint64_t)dw[b] << 32UL;
794 			b++;
795 		}
796 
797 		eae->eae_base = base;
798 		eae->eae_max_offset = max_offset;
799 
800 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
801 
802 		if (bootverbose) {
803 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
804 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
805 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
806 		}
807 	}
808 }
809 #undef REG
810 
811 static void
812 pci_read_cap(device_t pcib, pcicfgregs *cfg)
813 {
814 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
815 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
816 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
817 	uint64_t addr;
818 #endif
819 	uint32_t val;
820 	int	ptr, nextptr, ptrptr;
821 
822 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
823 	case PCIM_HDRTYPE_NORMAL:
824 	case PCIM_HDRTYPE_BRIDGE:
825 		ptrptr = PCIR_CAP_PTR;
826 		break;
827 	case PCIM_HDRTYPE_CARDBUS:
828 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
829 		break;
830 	default:
831 		return;		/* no extended capabilities support */
832 	}
833 	nextptr = REG(ptrptr, 1);	/* sanity check? */
834 
835 	/*
836 	 * Read capability entries.
837 	 */
838 	while (nextptr != 0) {
839 		/* Sanity check */
840 		if (nextptr > 255) {
841 			printf("illegal PCI extended capability offset %d\n",
842 			    nextptr);
843 			return;
844 		}
845 		/* Find the next entry */
846 		ptr = nextptr;
847 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
848 
849 		/* Process this entry */
850 		switch (REG(ptr + PCICAP_ID, 1)) {
851 		case PCIY_PMG:		/* PCI power management */
852 			if (cfg->pp.pp_cap == 0) {
853 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
854 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
855 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
856 				if ((nextptr - ptr) > PCIR_POWER_DATA)
857 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
858 			}
859 			break;
860 		case PCIY_HT:		/* HyperTransport */
861 			/* Determine HT-specific capability type. */
862 			val = REG(ptr + PCIR_HT_COMMAND, 2);
863 
864 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
865 				cfg->ht.ht_slave = ptr;
866 
867 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
868 			switch (val & PCIM_HTCMD_CAP_MASK) {
869 			case PCIM_HTCAP_MSI_MAPPING:
870 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
871 					/* Sanity check the mapping window. */
872 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
873 					    4);
874 					addr <<= 32;
875 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
876 					    4);
877 					if (addr != MSI_INTEL_ADDR_BASE)
878 						device_printf(pcib,
879 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
880 						    cfg->domain, cfg->bus,
881 						    cfg->slot, cfg->func,
882 						    (long long)addr);
883 				} else
884 					addr = MSI_INTEL_ADDR_BASE;
885 
886 				cfg->ht.ht_msimap = ptr;
887 				cfg->ht.ht_msictrl = val;
888 				cfg->ht.ht_msiaddr = addr;
889 				break;
890 			}
891 #endif
892 			break;
893 		case PCIY_MSI:		/* PCI MSI */
894 			cfg->msi.msi_location = ptr;
895 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
896 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
897 						     PCIM_MSICTRL_MMC_MASK)>>1);
898 			break;
899 		case PCIY_MSIX:		/* PCI MSI-X */
900 			cfg->msix.msix_location = ptr;
901 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
902 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
903 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
904 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
905 			cfg->msix.msix_table_bar = PCIR_BAR(val &
906 			    PCIM_MSIX_BIR_MASK);
907 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
908 			val = REG(ptr + PCIR_MSIX_PBA, 4);
909 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
910 			    PCIM_MSIX_BIR_MASK);
911 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
912 			break;
913 		case PCIY_VPD:		/* PCI Vital Product Data */
914 			cfg->vpd.vpd_reg = ptr;
915 			break;
916 		case PCIY_SUBVENDOR:
917 			/* Should always be true. */
918 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
919 			    PCIM_HDRTYPE_BRIDGE) {
920 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
921 				cfg->subvendor = val & 0xffff;
922 				cfg->subdevice = val >> 16;
923 			}
924 			break;
925 		case PCIY_PCIX:		/* PCI-X */
926 			/*
927 			 * Assume we have a PCI-X chipset if we have
928 			 * at least one PCI-PCI bridge with a PCI-X
929 			 * capability.  Note that some systems with
930 			 * PCI-express or HT chipsets might match on
931 			 * this check as well.
932 			 */
933 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
934 			    PCIM_HDRTYPE_BRIDGE)
935 				pcix_chipset = 1;
936 			cfg->pcix.pcix_location = ptr;
937 			break;
938 		case PCIY_EXPRESS:	/* PCI-express */
939 			/*
940 			 * Assume we have a PCI-express chipset if we have
941 			 * at least one PCI-express device.
942 			 */
943 			pcie_chipset = 1;
944 			cfg->pcie.pcie_location = ptr;
945 			val = REG(ptr + PCIER_FLAGS, 2);
946 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
947 			break;
948 		case PCIY_EA:		/* Enhanced Allocation */
949 			cfg->ea.ea_location = ptr;
950 			pci_ea_fill_info(pcib, cfg);
951 			break;
952 		default:
953 			break;
954 		}
955 	}
956 
957 #if defined(__powerpc__)
958 	/*
959 	 * Enable the MSI mapping window for all HyperTransport
960 	 * slaves.  PCI-PCI bridges have their windows enabled via
961 	 * PCIB_MAP_MSI().
962 	 */
963 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
964 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
965 		device_printf(pcib,
966 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
967 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
968 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
969 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
970 		     2);
971 	}
972 #endif
973 /* REG and WREG use carry through to next functions */
974 }
975 
976 /*
977  * PCI Vital Product Data
978  */
979 
980 #define	PCI_VPD_TIMEOUT		1000000
981 
982 static int
983 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
984 {
985 	int count = PCI_VPD_TIMEOUT;
986 
987 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
988 
989 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
990 
991 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
992 		if (--count < 0)
993 			return (ENXIO);
994 		DELAY(1);	/* limit looping */
995 	}
996 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
997 
998 	return (0);
999 }
1000 
1001 #if 0
1002 static int
1003 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1004 {
1005 	int count = PCI_VPD_TIMEOUT;
1006 
1007 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1008 
1009 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1010 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1011 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1012 		if (--count < 0)
1013 			return (ENXIO);
1014 		DELAY(1);	/* limit looping */
1015 	}
1016 
1017 	return (0);
1018 }
1019 #endif
1020 
1021 #undef PCI_VPD_TIMEOUT
1022 
1023 struct vpd_readstate {
1024 	device_t	pcib;
1025 	pcicfgregs	*cfg;
1026 	uint32_t	val;
1027 	int		bytesinval;
1028 	int		off;
1029 	uint8_t		cksum;
1030 };
1031 
1032 static int
1033 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1034 {
1035 	uint32_t reg;
1036 	uint8_t byte;
1037 
1038 	if (vrs->bytesinval == 0) {
1039 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1040 			return (ENXIO);
1041 		vrs->val = le32toh(reg);
1042 		vrs->off += 4;
1043 		byte = vrs->val & 0xff;
1044 		vrs->bytesinval = 3;
1045 	} else {
1046 		vrs->val = vrs->val >> 8;
1047 		byte = vrs->val & 0xff;
1048 		vrs->bytesinval--;
1049 	}
1050 
1051 	vrs->cksum += byte;
1052 	*data = byte;
1053 	return (0);
1054 }
1055 
1056 static void
1057 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1058 {
1059 	struct vpd_readstate vrs;
1060 	int state;
1061 	int name;
1062 	int remain;
1063 	int i;
1064 	int alloc, off;		/* alloc/off for RO/W arrays */
1065 	int cksumvalid;
1066 	int dflen;
1067 	uint8_t byte;
1068 	uint8_t byte2;
1069 
1070 	/* init vpd reader */
1071 	vrs.bytesinval = 0;
1072 	vrs.off = 0;
1073 	vrs.pcib = pcib;
1074 	vrs.cfg = cfg;
1075 	vrs.cksum = 0;
1076 
1077 	state = 0;
1078 	name = remain = i = 0;	/* shut up stupid gcc */
1079 	alloc = off = 0;	/* shut up stupid gcc */
1080 	dflen = 0;		/* shut up stupid gcc */
1081 	cksumvalid = -1;
1082 	while (state >= 0) {
1083 		if (vpd_nextbyte(&vrs, &byte)) {
1084 			state = -2;
1085 			break;
1086 		}
1087 #if 0
1088 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1089 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1090 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1091 #endif
1092 		switch (state) {
1093 		case 0:		/* item name */
1094 			if (byte & 0x80) {
1095 				if (vpd_nextbyte(&vrs, &byte2)) {
1096 					state = -2;
1097 					break;
1098 				}
1099 				remain = byte2;
1100 				if (vpd_nextbyte(&vrs, &byte2)) {
1101 					state = -2;
1102 					break;
1103 				}
1104 				remain |= byte2 << 8;
1105 				name = byte & 0x7f;
1106 			} else {
1107 				remain = byte & 0x7;
1108 				name = (byte >> 3) & 0xf;
1109 			}
1110 			if (vrs.off + remain - vrs.bytesinval > 0x8000) {
1111 				pci_printf(cfg,
1112 				    "VPD data overflow, remain %#x\n", remain);
1113 				state = -1;
1114 				break;
1115 			}
1116 			switch (name) {
1117 			case 0x2:	/* String */
1118 				cfg->vpd.vpd_ident = malloc(remain + 1,
1119 				    M_DEVBUF, M_WAITOK);
1120 				i = 0;
1121 				state = 1;
1122 				break;
1123 			case 0xf:	/* End */
1124 				state = -1;
1125 				break;
1126 			case 0x10:	/* VPD-R */
1127 				alloc = 8;
1128 				off = 0;
1129 				cfg->vpd.vpd_ros = malloc(alloc *
1130 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1131 				    M_WAITOK | M_ZERO);
1132 				state = 2;
1133 				break;
1134 			case 0x11:	/* VPD-W */
1135 				alloc = 8;
1136 				off = 0;
1137 				cfg->vpd.vpd_w = malloc(alloc *
1138 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1139 				    M_WAITOK | M_ZERO);
1140 				state = 5;
1141 				break;
1142 			default:	/* Invalid data, abort */
1143 				state = -1;
1144 				break;
1145 			}
1146 			break;
1147 
1148 		case 1:	/* Identifier String */
1149 			cfg->vpd.vpd_ident[i++] = byte;
1150 			remain--;
1151 			if (remain == 0)  {
1152 				cfg->vpd.vpd_ident[i] = '\0';
1153 				state = 0;
1154 			}
1155 			break;
1156 
1157 		case 2:	/* VPD-R Keyword Header */
1158 			if (off == alloc) {
1159 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1160 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1161 				    M_DEVBUF, M_WAITOK | M_ZERO);
1162 			}
1163 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1164 			if (vpd_nextbyte(&vrs, &byte2)) {
1165 				state = -2;
1166 				break;
1167 			}
1168 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1169 			if (vpd_nextbyte(&vrs, &byte2)) {
1170 				state = -2;
1171 				break;
1172 			}
1173 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1174 			if (dflen == 0 &&
1175 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1176 			    2) == 0) {
1177 				/*
1178 				 * if this happens, we can't trust the rest
1179 				 * of the VPD.
1180 				 */
1181 				pci_printf(cfg, "bad keyword length: %d\n",
1182 				    dflen);
1183 				cksumvalid = 0;
1184 				state = -1;
1185 				break;
1186 			} else if (dflen == 0) {
1187 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1188 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1189 				    M_DEVBUF, M_WAITOK);
1190 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1191 			} else
1192 				cfg->vpd.vpd_ros[off].value = malloc(
1193 				    (dflen + 1) *
1194 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1195 				    M_DEVBUF, M_WAITOK);
1196 			remain -= 3;
1197 			i = 0;
1198 			/* keep in sync w/ state 3's transistions */
1199 			if (dflen == 0 && remain == 0)
1200 				state = 0;
1201 			else if (dflen == 0)
1202 				state = 2;
1203 			else
1204 				state = 3;
1205 			break;
1206 
1207 		case 3:	/* VPD-R Keyword Value */
1208 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1209 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1210 			    "RV", 2) == 0 && cksumvalid == -1) {
1211 				if (vrs.cksum == 0)
1212 					cksumvalid = 1;
1213 				else {
1214 					if (bootverbose)
1215 						pci_printf(cfg,
1216 					    "bad VPD cksum, remain %hhu\n",
1217 						    vrs.cksum);
1218 					cksumvalid = 0;
1219 					state = -1;
1220 					break;
1221 				}
1222 			}
1223 			dflen--;
1224 			remain--;
1225 			/* keep in sync w/ state 2's transistions */
1226 			if (dflen == 0)
1227 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1228 			if (dflen == 0 && remain == 0) {
1229 				cfg->vpd.vpd_rocnt = off;
1230 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1231 				    off * sizeof(*cfg->vpd.vpd_ros),
1232 				    M_DEVBUF, M_WAITOK | M_ZERO);
1233 				state = 0;
1234 			} else if (dflen == 0)
1235 				state = 2;
1236 			break;
1237 
1238 		case 4:
1239 			remain--;
1240 			if (remain == 0)
1241 				state = 0;
1242 			break;
1243 
1244 		case 5:	/* VPD-W Keyword Header */
1245 			if (off == alloc) {
1246 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1247 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1248 				    M_DEVBUF, M_WAITOK | M_ZERO);
1249 			}
1250 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1251 			if (vpd_nextbyte(&vrs, &byte2)) {
1252 				state = -2;
1253 				break;
1254 			}
1255 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1256 			if (vpd_nextbyte(&vrs, &byte2)) {
1257 				state = -2;
1258 				break;
1259 			}
1260 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1261 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1262 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1263 			    sizeof(*cfg->vpd.vpd_w[off].value),
1264 			    M_DEVBUF, M_WAITOK);
1265 			remain -= 3;
1266 			i = 0;
1267 			/* keep in sync w/ state 6's transistions */
1268 			if (dflen == 0 && remain == 0)
1269 				state = 0;
1270 			else if (dflen == 0)
1271 				state = 5;
1272 			else
1273 				state = 6;
1274 			break;
1275 
1276 		case 6:	/* VPD-W Keyword Value */
1277 			cfg->vpd.vpd_w[off].value[i++] = byte;
1278 			dflen--;
1279 			remain--;
1280 			/* keep in sync w/ state 5's transistions */
1281 			if (dflen == 0)
1282 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1283 			if (dflen == 0 && remain == 0) {
1284 				cfg->vpd.vpd_wcnt = off;
1285 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1286 				    off * sizeof(*cfg->vpd.vpd_w),
1287 				    M_DEVBUF, M_WAITOK | M_ZERO);
1288 				state = 0;
1289 			} else if (dflen == 0)
1290 				state = 5;
1291 			break;
1292 
1293 		default:
1294 			pci_printf(cfg, "invalid state: %d\n", state);
1295 			state = -1;
1296 			break;
1297 		}
1298 	}
1299 
1300 	if (cksumvalid == 0 || state < -1) {
1301 		/* read-only data bad, clean up */
1302 		if (cfg->vpd.vpd_ros != NULL) {
1303 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1304 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1305 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1306 			cfg->vpd.vpd_ros = NULL;
1307 		}
1308 	}
1309 	if (state < -1) {
1310 		/* I/O error, clean up */
1311 		pci_printf(cfg, "failed to read VPD data.\n");
1312 		if (cfg->vpd.vpd_ident != NULL) {
1313 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1314 			cfg->vpd.vpd_ident = NULL;
1315 		}
1316 		if (cfg->vpd.vpd_w != NULL) {
1317 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1318 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1319 			free(cfg->vpd.vpd_w, M_DEVBUF);
1320 			cfg->vpd.vpd_w = NULL;
1321 		}
1322 	}
1323 	cfg->vpd.vpd_cached = 1;
1324 #undef REG
1325 #undef WREG
1326 }
1327 
1328 int
1329 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1330 {
1331 	struct pci_devinfo *dinfo = device_get_ivars(child);
1332 	pcicfgregs *cfg = &dinfo->cfg;
1333 
1334 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1335 		pci_read_vpd(device_get_parent(dev), cfg);
1336 
1337 	*identptr = cfg->vpd.vpd_ident;
1338 
1339 	if (*identptr == NULL)
1340 		return (ENXIO);
1341 
1342 	return (0);
1343 }
1344 
1345 int
1346 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1347 	const char **vptr)
1348 {
1349 	struct pci_devinfo *dinfo = device_get_ivars(child);
1350 	pcicfgregs *cfg = &dinfo->cfg;
1351 	int i;
1352 
1353 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1354 		pci_read_vpd(device_get_parent(dev), cfg);
1355 
1356 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1357 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1358 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1359 			*vptr = cfg->vpd.vpd_ros[i].value;
1360 			return (0);
1361 		}
1362 
1363 	*vptr = NULL;
1364 	return (ENXIO);
1365 }
1366 
1367 struct pcicfg_vpd *
1368 pci_fetch_vpd_list(device_t dev)
1369 {
1370 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1371 	pcicfgregs *cfg = &dinfo->cfg;
1372 
1373 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1374 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1375 	return (&cfg->vpd);
1376 }
1377 
1378 /*
1379  * Find the requested HyperTransport capability and return the offset
1380  * in configuration space via the pointer provided.  The function
1381  * returns 0 on success and an error code otherwise.
1382  */
1383 int
1384 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1385 {
1386 	int ptr, error;
1387 	uint16_t val;
1388 
1389 	error = pci_find_cap(child, PCIY_HT, &ptr);
1390 	if (error)
1391 		return (error);
1392 
1393 	/*
1394 	 * Traverse the capabilities list checking each HT capability
1395 	 * to see if it matches the requested HT capability.
1396 	 */
1397 	for (;;) {
1398 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1399 		if (capability == PCIM_HTCAP_SLAVE ||
1400 		    capability == PCIM_HTCAP_HOST)
1401 			val &= 0xe000;
1402 		else
1403 			val &= PCIM_HTCMD_CAP_MASK;
1404 		if (val == capability) {
1405 			if (capreg != NULL)
1406 				*capreg = ptr;
1407 			return (0);
1408 		}
1409 
1410 		/* Skip to the next HT capability. */
1411 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1412 			break;
1413 	}
1414 
1415 	return (ENOENT);
1416 }
1417 
1418 /*
1419  * Find the next requested HyperTransport capability after start and return
1420  * the offset in configuration space via the pointer provided.  The function
1421  * returns 0 on success and an error code otherwise.
1422  */
1423 int
1424 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1425     int start, int *capreg)
1426 {
1427 	int ptr;
1428 	uint16_t val;
1429 
1430 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1431 	    ("start capability is not HyperTransport capability"));
1432 	ptr = start;
1433 
1434 	/*
1435 	 * Traverse the capabilities list checking each HT capability
1436 	 * to see if it matches the requested HT capability.
1437 	 */
1438 	for (;;) {
1439 		/* Skip to the next HT capability. */
1440 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1441 			break;
1442 
1443 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1444 		if (capability == PCIM_HTCAP_SLAVE ||
1445 		    capability == PCIM_HTCAP_HOST)
1446 			val &= 0xe000;
1447 		else
1448 			val &= PCIM_HTCMD_CAP_MASK;
1449 		if (val == capability) {
1450 			if (capreg != NULL)
1451 				*capreg = ptr;
1452 			return (0);
1453 		}
1454 	}
1455 
1456 	return (ENOENT);
1457 }
1458 
1459 /*
1460  * Find the requested capability and return the offset in
1461  * configuration space via the pointer provided.  The function returns
1462  * 0 on success and an error code otherwise.
1463  */
1464 int
1465 pci_find_cap_method(device_t dev, device_t child, int capability,
1466     int *capreg)
1467 {
1468 	struct pci_devinfo *dinfo = device_get_ivars(child);
1469 	pcicfgregs *cfg = &dinfo->cfg;
1470 	uint32_t status;
1471 	uint8_t ptr;
1472 
1473 	/*
1474 	 * Check the CAP_LIST bit of the PCI status register first.
1475 	 */
1476 	status = pci_read_config(child, PCIR_STATUS, 2);
1477 	if (!(status & PCIM_STATUS_CAPPRESENT))
1478 		return (ENXIO);
1479 
1480 	/*
1481 	 * Determine the start pointer of the capabilities list.
1482 	 */
1483 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1484 	case PCIM_HDRTYPE_NORMAL:
1485 	case PCIM_HDRTYPE_BRIDGE:
1486 		ptr = PCIR_CAP_PTR;
1487 		break;
1488 	case PCIM_HDRTYPE_CARDBUS:
1489 		ptr = PCIR_CAP_PTR_2;
1490 		break;
1491 	default:
1492 		/* XXX: panic? */
1493 		return (ENXIO);		/* no extended capabilities support */
1494 	}
1495 	ptr = pci_read_config(child, ptr, 1);
1496 
1497 	/*
1498 	 * Traverse the capabilities list.
1499 	 */
1500 	while (ptr != 0) {
1501 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1502 			if (capreg != NULL)
1503 				*capreg = ptr;
1504 			return (0);
1505 		}
1506 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1507 	}
1508 
1509 	return (ENOENT);
1510 }
1511 
1512 /*
1513  * Find the next requested capability after start and return the offset in
1514  * configuration space via the pointer provided.  The function returns
1515  * 0 on success and an error code otherwise.
1516  */
1517 int
1518 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1519     int start, int *capreg)
1520 {
1521 	uint8_t ptr;
1522 
1523 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1524 	    ("start capability is not expected capability"));
1525 
1526 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1527 	while (ptr != 0) {
1528 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1529 			if (capreg != NULL)
1530 				*capreg = ptr;
1531 			return (0);
1532 		}
1533 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1534 	}
1535 
1536 	return (ENOENT);
1537 }
1538 
1539 /*
1540  * Find the requested extended capability and return the offset in
1541  * configuration space via the pointer provided.  The function returns
1542  * 0 on success and an error code otherwise.
1543  */
1544 int
1545 pci_find_extcap_method(device_t dev, device_t child, int capability,
1546     int *capreg)
1547 {
1548 	struct pci_devinfo *dinfo = device_get_ivars(child);
1549 	pcicfgregs *cfg = &dinfo->cfg;
1550 	uint32_t ecap;
1551 	uint16_t ptr;
1552 
1553 	/* Only supported for PCI-express devices. */
1554 	if (cfg->pcie.pcie_location == 0)
1555 		return (ENXIO);
1556 
1557 	ptr = PCIR_EXTCAP;
1558 	ecap = pci_read_config(child, ptr, 4);
1559 	if (ecap == 0xffffffff || ecap == 0)
1560 		return (ENOENT);
1561 	for (;;) {
1562 		if (PCI_EXTCAP_ID(ecap) == capability) {
1563 			if (capreg != NULL)
1564 				*capreg = ptr;
1565 			return (0);
1566 		}
1567 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1568 		if (ptr == 0)
1569 			break;
1570 		ecap = pci_read_config(child, ptr, 4);
1571 	}
1572 
1573 	return (ENOENT);
1574 }
1575 
1576 /*
1577  * Find the next requested extended capability after start and return the
1578  * offset in configuration space via the pointer provided.  The function
1579  * returns 0 on success and an error code otherwise.
1580  */
1581 int
1582 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1583     int start, int *capreg)
1584 {
1585 	struct pci_devinfo *dinfo = device_get_ivars(child);
1586 	pcicfgregs *cfg = &dinfo->cfg;
1587 	uint32_t ecap;
1588 	uint16_t ptr;
1589 
1590 	/* Only supported for PCI-express devices. */
1591 	if (cfg->pcie.pcie_location == 0)
1592 		return (ENXIO);
1593 
1594 	ecap = pci_read_config(child, start, 4);
1595 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1596 	    ("start extended capability is not expected capability"));
1597 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1598 	while (ptr != 0) {
1599 		ecap = pci_read_config(child, ptr, 4);
1600 		if (PCI_EXTCAP_ID(ecap) == capability) {
1601 			if (capreg != NULL)
1602 				*capreg = ptr;
1603 			return (0);
1604 		}
1605 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1606 	}
1607 
1608 	return (ENOENT);
1609 }
1610 
1611 /*
1612  * Support for MSI-X message interrupts.
1613  */
1614 static void
1615 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1616 {
1617 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1618 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1619 	uint32_t offset;
1620 
1621 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1622 	offset = msix->msix_table_offset + index * 16;
1623 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1624 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1625 	bus_write_4(msix->msix_table_res, offset + 8, data);
1626 }
1627 
1628 void
1629 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1630     uint64_t address, uint32_t data)
1631 {
1632 
1633 	if (pci_msix_rewrite_table) {
1634 		struct pci_devinfo *dinfo = device_get_ivars(child);
1635 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1636 
1637 		/*
1638 		 * Some VM hosts require MSIX to be disabled in the
1639 		 * control register before updating the MSIX table
1640 		 * entries are allowed. It is not enough to only
1641 		 * disable MSIX while updating a single entry. MSIX
1642 		 * must be disabled while updating all entries in the
1643 		 * table.
1644 		 */
1645 		pci_write_config(child,
1646 		    msix->msix_location + PCIR_MSIX_CTRL,
1647 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1648 		pci_resume_msix(child);
1649 	} else
1650 		pci_write_msix_entry(child, index, address, data);
1651 
1652 	/* Enable MSI -> HT mapping. */
1653 	pci_ht_map_msi(child, address);
1654 }
1655 
1656 void
1657 pci_mask_msix(device_t dev, u_int index)
1658 {
1659 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1660 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1661 	uint32_t offset, val;
1662 
1663 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1664 	offset = msix->msix_table_offset + index * 16 + 12;
1665 	val = bus_read_4(msix->msix_table_res, offset);
1666 	val |= PCIM_MSIX_VCTRL_MASK;
1667 
1668 	/*
1669 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1670 	 * register, so always write the new value.
1671 	 */
1672 	bus_write_4(msix->msix_table_res, offset, val);
1673 }
1674 
1675 void
1676 pci_unmask_msix(device_t dev, u_int index)
1677 {
1678 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1679 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1680 	uint32_t offset, val;
1681 
1682 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1683 	offset = msix->msix_table_offset + index * 16 + 12;
1684 	val = bus_read_4(msix->msix_table_res, offset);
1685 	val &= ~PCIM_MSIX_VCTRL_MASK;
1686 
1687 	/*
1688 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1689 	 * register, so always write the new value.
1690 	 */
1691 	bus_write_4(msix->msix_table_res, offset, val);
1692 }
1693 
1694 int
1695 pci_pending_msix(device_t dev, u_int index)
1696 {
1697 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1698 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1699 	uint32_t offset, bit;
1700 
1701 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1702 	offset = msix->msix_pba_offset + (index / 32) * 4;
1703 	bit = 1 << index % 32;
1704 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1705 }
1706 
1707 /*
1708  * Restore MSI-X registers and table during resume.  If MSI-X is
1709  * enabled then walk the virtual table to restore the actual MSI-X
1710  * table.
1711  */
1712 static void
1713 pci_resume_msix(device_t dev)
1714 {
1715 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1716 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1717 	struct msix_table_entry *mte;
1718 	struct msix_vector *mv;
1719 	int i;
1720 
1721 	if (msix->msix_alloc > 0) {
1722 		/* First, mask all vectors. */
1723 		for (i = 0; i < msix->msix_msgnum; i++)
1724 			pci_mask_msix(dev, i);
1725 
1726 		/* Second, program any messages with at least one handler. */
1727 		for (i = 0; i < msix->msix_table_len; i++) {
1728 			mte = &msix->msix_table[i];
1729 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1730 				continue;
1731 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1732 			pci_write_msix_entry(dev, i, mv->mv_address,
1733 			    mv->mv_data);
1734 			pci_unmask_msix(dev, i);
1735 		}
1736 	}
1737 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1738 	    msix->msix_ctrl, 2);
1739 }
1740 
1741 /*
1742  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1743  * returned in *count.  After this function returns, each message will be
1744  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1745  */
1746 int
1747 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1748 {
1749 	struct pci_devinfo *dinfo = device_get_ivars(child);
1750 	pcicfgregs *cfg = &dinfo->cfg;
1751 	struct resource_list_entry *rle;
1752 	int actual, error, i, irq, max;
1753 
1754 	/* Don't let count == 0 get us into trouble. */
1755 	if (*count == 0)
1756 		return (EINVAL);
1757 
1758 	/* If rid 0 is allocated, then fail. */
1759 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1760 	if (rle != NULL && rle->res != NULL)
1761 		return (ENXIO);
1762 
1763 	/* Already have allocated messages? */
1764 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1765 		return (ENXIO);
1766 
1767 	/* If MSI-X is blacklisted for this system, fail. */
1768 	if (pci_msix_blacklisted())
1769 		return (ENXIO);
1770 
1771 	/* MSI-X capability present? */
1772 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1773 		return (ENODEV);
1774 
1775 	/* Make sure the appropriate BARs are mapped. */
1776 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1777 	    cfg->msix.msix_table_bar);
1778 	if (rle == NULL || rle->res == NULL ||
1779 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1780 		return (ENXIO);
1781 	cfg->msix.msix_table_res = rle->res;
1782 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1783 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1784 		    cfg->msix.msix_pba_bar);
1785 		if (rle == NULL || rle->res == NULL ||
1786 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1787 			return (ENXIO);
1788 	}
1789 	cfg->msix.msix_pba_res = rle->res;
1790 
1791 	if (bootverbose)
1792 		device_printf(child,
1793 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1794 		    *count, cfg->msix.msix_msgnum);
1795 	max = min(*count, cfg->msix.msix_msgnum);
1796 	for (i = 0; i < max; i++) {
1797 		/* Allocate a message. */
1798 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1799 		if (error) {
1800 			if (i == 0)
1801 				return (error);
1802 			break;
1803 		}
1804 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1805 		    irq, 1);
1806 	}
1807 	actual = i;
1808 
1809 	if (bootverbose) {
1810 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1811 		if (actual == 1)
1812 			device_printf(child, "using IRQ %ju for MSI-X\n",
1813 			    rle->start);
1814 		else {
1815 			int run;
1816 
1817 			/*
1818 			 * Be fancy and try to print contiguous runs of
1819 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1820 			 * 'run' is true if we are in a range.
1821 			 */
1822 			device_printf(child, "using IRQs %ju", rle->start);
1823 			irq = rle->start;
1824 			run = 0;
1825 			for (i = 1; i < actual; i++) {
1826 				rle = resource_list_find(&dinfo->resources,
1827 				    SYS_RES_IRQ, i + 1);
1828 
1829 				/* Still in a run? */
1830 				if (rle->start == irq + 1) {
1831 					run = 1;
1832 					irq++;
1833 					continue;
1834 				}
1835 
1836 				/* Finish previous range. */
1837 				if (run) {
1838 					printf("-%d", irq);
1839 					run = 0;
1840 				}
1841 
1842 				/* Start new range. */
1843 				printf(",%ju", rle->start);
1844 				irq = rle->start;
1845 			}
1846 
1847 			/* Unfinished range? */
1848 			if (run)
1849 				printf("-%d", irq);
1850 			printf(" for MSI-X\n");
1851 		}
1852 	}
1853 
1854 	/* Mask all vectors. */
1855 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1856 		pci_mask_msix(child, i);
1857 
1858 	/* Allocate and initialize vector data and virtual table. */
1859 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1860 	    M_DEVBUF, M_WAITOK | M_ZERO);
1861 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1862 	    M_DEVBUF, M_WAITOK | M_ZERO);
1863 	for (i = 0; i < actual; i++) {
1864 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1865 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1866 		cfg->msix.msix_table[i].mte_vector = i + 1;
1867 	}
1868 
1869 	/* Update control register to enable MSI-X. */
1870 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1871 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1872 	    cfg->msix.msix_ctrl, 2);
1873 
1874 	/* Update counts of alloc'd messages. */
1875 	cfg->msix.msix_alloc = actual;
1876 	cfg->msix.msix_table_len = actual;
1877 	*count = actual;
1878 	return (0);
1879 }
1880 
1881 /*
1882  * By default, pci_alloc_msix() will assign the allocated IRQ
1883  * resources consecutively to the first N messages in the MSI-X table.
1884  * However, device drivers may want to use different layouts if they
1885  * either receive fewer messages than they asked for, or they wish to
1886  * populate the MSI-X table sparsely.  This method allows the driver
1887  * to specify what layout it wants.  It must be called after a
1888  * successful pci_alloc_msix() but before any of the associated
1889  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1890  *
1891  * The 'vectors' array contains 'count' message vectors.  The array
1892  * maps directly to the MSI-X table in that index 0 in the array
1893  * specifies the vector for the first message in the MSI-X table, etc.
1894  * The vector value in each array index can either be 0 to indicate
1895  * that no vector should be assigned to a message slot, or it can be a
1896  * number from 1 to N (where N is the count returned from a
1897  * succcessful call to pci_alloc_msix()) to indicate which message
1898  * vector (IRQ) to be used for the corresponding message.
1899  *
1900  * On successful return, each message with a non-zero vector will have
1901  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1902  * 1.  Additionally, if any of the IRQs allocated via the previous
1903  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1904  * will be freed back to the system automatically.
1905  *
1906  * For example, suppose a driver has a MSI-X table with 6 messages and
1907  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1908  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1909  * C.  After the call to pci_alloc_msix(), the device will be setup to
1910  * have an MSI-X table of ABC--- (where - means no vector assigned).
1911  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1912  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1913  * be freed back to the system.  This device will also have valid
1914  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1915  *
1916  * In any case, the SYS_RES_IRQ rid X will always map to the message
1917  * at MSI-X table index X - 1 and will only be valid if a vector is
1918  * assigned to that table entry.
1919  */
1920 int
1921 pci_remap_msix_method(device_t dev, device_t child, int count,
1922     const u_int *vectors)
1923 {
1924 	struct pci_devinfo *dinfo = device_get_ivars(child);
1925 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1926 	struct resource_list_entry *rle;
1927 	int i, irq, j, *used;
1928 
1929 	/*
1930 	 * Have to have at least one message in the table but the
1931 	 * table can't be bigger than the actual MSI-X table in the
1932 	 * device.
1933 	 */
1934 	if (count == 0 || count > msix->msix_msgnum)
1935 		return (EINVAL);
1936 
1937 	/* Sanity check the vectors. */
1938 	for (i = 0; i < count; i++)
1939 		if (vectors[i] > msix->msix_alloc)
1940 			return (EINVAL);
1941 
1942 	/*
1943 	 * Make sure there aren't any holes in the vectors to be used.
1944 	 * It's a big pain to support it, and it doesn't really make
1945 	 * sense anyway.  Also, at least one vector must be used.
1946 	 */
1947 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1948 	    M_ZERO);
1949 	for (i = 0; i < count; i++)
1950 		if (vectors[i] != 0)
1951 			used[vectors[i] - 1] = 1;
1952 	for (i = 0; i < msix->msix_alloc - 1; i++)
1953 		if (used[i] == 0 && used[i + 1] == 1) {
1954 			free(used, M_DEVBUF);
1955 			return (EINVAL);
1956 		}
1957 	if (used[0] != 1) {
1958 		free(used, M_DEVBUF);
1959 		return (EINVAL);
1960 	}
1961 
1962 	/* Make sure none of the resources are allocated. */
1963 	for (i = 0; i < msix->msix_table_len; i++) {
1964 		if (msix->msix_table[i].mte_vector == 0)
1965 			continue;
1966 		if (msix->msix_table[i].mte_handlers > 0) {
1967 			free(used, M_DEVBUF);
1968 			return (EBUSY);
1969 		}
1970 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1971 		KASSERT(rle != NULL, ("missing resource"));
1972 		if (rle->res != NULL) {
1973 			free(used, M_DEVBUF);
1974 			return (EBUSY);
1975 		}
1976 	}
1977 
1978 	/* Free the existing resource list entries. */
1979 	for (i = 0; i < msix->msix_table_len; i++) {
1980 		if (msix->msix_table[i].mte_vector == 0)
1981 			continue;
1982 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1983 	}
1984 
1985 	/*
1986 	 * Build the new virtual table keeping track of which vectors are
1987 	 * used.
1988 	 */
1989 	free(msix->msix_table, M_DEVBUF);
1990 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1991 	    M_DEVBUF, M_WAITOK | M_ZERO);
1992 	for (i = 0; i < count; i++)
1993 		msix->msix_table[i].mte_vector = vectors[i];
1994 	msix->msix_table_len = count;
1995 
1996 	/* Free any unused IRQs and resize the vectors array if necessary. */
1997 	j = msix->msix_alloc - 1;
1998 	if (used[j] == 0) {
1999 		struct msix_vector *vec;
2000 
2001 		while (used[j] == 0) {
2002 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2003 			    msix->msix_vectors[j].mv_irq);
2004 			j--;
2005 		}
2006 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2007 		    M_WAITOK);
2008 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2009 		    (j + 1));
2010 		free(msix->msix_vectors, M_DEVBUF);
2011 		msix->msix_vectors = vec;
2012 		msix->msix_alloc = j + 1;
2013 	}
2014 	free(used, M_DEVBUF);
2015 
2016 	/* Map the IRQs onto the rids. */
2017 	for (i = 0; i < count; i++) {
2018 		if (vectors[i] == 0)
2019 			continue;
2020 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2021 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2022 		    irq, 1);
2023 	}
2024 
2025 	if (bootverbose) {
2026 		device_printf(child, "Remapped MSI-X IRQs as: ");
2027 		for (i = 0; i < count; i++) {
2028 			if (i != 0)
2029 				printf(", ");
2030 			if (vectors[i] == 0)
2031 				printf("---");
2032 			else
2033 				printf("%d",
2034 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2035 		}
2036 		printf("\n");
2037 	}
2038 
2039 	return (0);
2040 }
2041 
2042 static int
2043 pci_release_msix(device_t dev, device_t child)
2044 {
2045 	struct pci_devinfo *dinfo = device_get_ivars(child);
2046 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2047 	struct resource_list_entry *rle;
2048 	int i;
2049 
2050 	/* Do we have any messages to release? */
2051 	if (msix->msix_alloc == 0)
2052 		return (ENODEV);
2053 
2054 	/* Make sure none of the resources are allocated. */
2055 	for (i = 0; i < msix->msix_table_len; i++) {
2056 		if (msix->msix_table[i].mte_vector == 0)
2057 			continue;
2058 		if (msix->msix_table[i].mte_handlers > 0)
2059 			return (EBUSY);
2060 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2061 		KASSERT(rle != NULL, ("missing resource"));
2062 		if (rle->res != NULL)
2063 			return (EBUSY);
2064 	}
2065 
2066 	/* Update control register to disable MSI-X. */
2067 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2068 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2069 	    msix->msix_ctrl, 2);
2070 
2071 	/* Free the resource list entries. */
2072 	for (i = 0; i < msix->msix_table_len; i++) {
2073 		if (msix->msix_table[i].mte_vector == 0)
2074 			continue;
2075 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2076 	}
2077 	free(msix->msix_table, M_DEVBUF);
2078 	msix->msix_table_len = 0;
2079 
2080 	/* Release the IRQs. */
2081 	for (i = 0; i < msix->msix_alloc; i++)
2082 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2083 		    msix->msix_vectors[i].mv_irq);
2084 	free(msix->msix_vectors, M_DEVBUF);
2085 	msix->msix_alloc = 0;
2086 	return (0);
2087 }
2088 
2089 /*
2090  * Return the max supported MSI-X messages this device supports.
2091  * Basically, assuming the MD code can alloc messages, this function
2092  * should return the maximum value that pci_alloc_msix() can return.
2093  * Thus, it is subject to the tunables, etc.
2094  */
2095 int
2096 pci_msix_count_method(device_t dev, device_t child)
2097 {
2098 	struct pci_devinfo *dinfo = device_get_ivars(child);
2099 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2100 
2101 	if (pci_do_msix && msix->msix_location != 0)
2102 		return (msix->msix_msgnum);
2103 	return (0);
2104 }
2105 
2106 int
2107 pci_msix_pba_bar_method(device_t dev, device_t child)
2108 {
2109 	struct pci_devinfo *dinfo = device_get_ivars(child);
2110 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2111 
2112 	if (pci_do_msix && msix->msix_location != 0)
2113 		return (msix->msix_pba_bar);
2114 	return (-1);
2115 }
2116 
2117 int
2118 pci_msix_table_bar_method(device_t dev, device_t child)
2119 {
2120 	struct pci_devinfo *dinfo = device_get_ivars(child);
2121 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2122 
2123 	if (pci_do_msix && msix->msix_location != 0)
2124 		return (msix->msix_table_bar);
2125 	return (-1);
2126 }
2127 
2128 /*
2129  * HyperTransport MSI mapping control
2130  */
2131 void
2132 pci_ht_map_msi(device_t dev, uint64_t addr)
2133 {
2134 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2135 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2136 
2137 	if (!ht->ht_msimap)
2138 		return;
2139 
2140 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2141 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2142 		/* Enable MSI -> HT mapping. */
2143 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2144 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2145 		    ht->ht_msictrl, 2);
2146 	}
2147 
2148 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2149 		/* Disable MSI -> HT mapping. */
2150 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2151 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2152 		    ht->ht_msictrl, 2);
2153 	}
2154 }
2155 
2156 int
2157 pci_get_max_payload(device_t dev)
2158 {
2159 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2160 	int cap;
2161 	uint16_t val;
2162 
2163 	cap = dinfo->cfg.pcie.pcie_location;
2164 	if (cap == 0)
2165 		return (0);
2166 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2167 	val &= PCIEM_CTL_MAX_PAYLOAD;
2168 	val >>= 5;
2169 	return (1 << (val + 7));
2170 }
2171 
2172 int
2173 pci_get_max_read_req(device_t dev)
2174 {
2175 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2176 	int cap;
2177 	uint16_t val;
2178 
2179 	cap = dinfo->cfg.pcie.pcie_location;
2180 	if (cap == 0)
2181 		return (0);
2182 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2183 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2184 	val >>= 12;
2185 	return (1 << (val + 7));
2186 }
2187 
2188 int
2189 pci_set_max_read_req(device_t dev, int size)
2190 {
2191 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2192 	int cap;
2193 	uint16_t val;
2194 
2195 	cap = dinfo->cfg.pcie.pcie_location;
2196 	if (cap == 0)
2197 		return (0);
2198 	if (size < 128)
2199 		size = 128;
2200 	if (size > 4096)
2201 		size = 4096;
2202 	size = (1 << (fls(size) - 1));
2203 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2204 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2205 	val |= (fls(size) - 8) << 12;
2206 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2207 	return (size);
2208 }
2209 
2210 uint32_t
2211 pcie_read_config(device_t dev, int reg, int width)
2212 {
2213 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2214 	int cap;
2215 
2216 	cap = dinfo->cfg.pcie.pcie_location;
2217 	if (cap == 0) {
2218 		if (width == 2)
2219 			return (0xffff);
2220 		return (0xffffffff);
2221 	}
2222 
2223 	return (pci_read_config(dev, cap + reg, width));
2224 }
2225 
2226 void
2227 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2228 {
2229 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2230 	int cap;
2231 
2232 	cap = dinfo->cfg.pcie.pcie_location;
2233 	if (cap == 0)
2234 		return;
2235 	pci_write_config(dev, cap + reg, value, width);
2236 }
2237 
2238 /*
2239  * Adjusts a PCI-e capability register by clearing the bits in mask
2240  * and setting the bits in (value & mask).  Bits not set in mask are
2241  * not adjusted.
2242  *
2243  * Returns the old value on success or all ones on failure.
2244  */
2245 uint32_t
2246 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2247     int width)
2248 {
2249 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2250 	uint32_t old, new;
2251 	int cap;
2252 
2253 	cap = dinfo->cfg.pcie.pcie_location;
2254 	if (cap == 0) {
2255 		if (width == 2)
2256 			return (0xffff);
2257 		return (0xffffffff);
2258 	}
2259 
2260 	old = pci_read_config(dev, cap + reg, width);
2261 	new = old & ~mask;
2262 	new |= (value & mask);
2263 	pci_write_config(dev, cap + reg, new, width);
2264 	return (old);
2265 }
2266 
2267 /*
2268  * Support for MSI message signalled interrupts.
2269  */
2270 void
2271 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2272     uint16_t data)
2273 {
2274 	struct pci_devinfo *dinfo = device_get_ivars(child);
2275 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2276 
2277 	/* Write data and address values. */
2278 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2279 	    address & 0xffffffff, 4);
2280 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2281 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2282 		    address >> 32, 4);
2283 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2284 		    data, 2);
2285 	} else
2286 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2287 		    2);
2288 
2289 	/* Enable MSI in the control register. */
2290 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2291 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2292 	    msi->msi_ctrl, 2);
2293 
2294 	/* Enable MSI -> HT mapping. */
2295 	pci_ht_map_msi(child, address);
2296 }
2297 
2298 void
2299 pci_disable_msi_method(device_t dev, device_t child)
2300 {
2301 	struct pci_devinfo *dinfo = device_get_ivars(child);
2302 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2303 
2304 	/* Disable MSI -> HT mapping. */
2305 	pci_ht_map_msi(child, 0);
2306 
2307 	/* Disable MSI in the control register. */
2308 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2309 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2310 	    msi->msi_ctrl, 2);
2311 }
2312 
2313 /*
2314  * Restore MSI registers during resume.  If MSI is enabled then
2315  * restore the data and address registers in addition to the control
2316  * register.
2317  */
2318 static void
2319 pci_resume_msi(device_t dev)
2320 {
2321 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2322 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2323 	uint64_t address;
2324 	uint16_t data;
2325 
2326 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2327 		address = msi->msi_addr;
2328 		data = msi->msi_data;
2329 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2330 		    address & 0xffffffff, 4);
2331 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2332 			pci_write_config(dev, msi->msi_location +
2333 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2334 			pci_write_config(dev, msi->msi_location +
2335 			    PCIR_MSI_DATA_64BIT, data, 2);
2336 		} else
2337 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2338 			    data, 2);
2339 	}
2340 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2341 	    2);
2342 }
2343 
2344 static int
2345 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2346 {
2347 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2348 	pcicfgregs *cfg = &dinfo->cfg;
2349 	struct resource_list_entry *rle;
2350 	struct msix_table_entry *mte;
2351 	struct msix_vector *mv;
2352 	uint64_t addr;
2353 	uint32_t data;
2354 	int error, i, j;
2355 
2356 	/*
2357 	 * Handle MSI first.  We try to find this IRQ among our list
2358 	 * of MSI IRQs.  If we find it, we request updated address and
2359 	 * data registers and apply the results.
2360 	 */
2361 	if (cfg->msi.msi_alloc > 0) {
2362 
2363 		/* If we don't have any active handlers, nothing to do. */
2364 		if (cfg->msi.msi_handlers == 0)
2365 			return (0);
2366 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2367 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2368 			    i + 1);
2369 			if (rle->start == irq) {
2370 				error = PCIB_MAP_MSI(device_get_parent(bus),
2371 				    dev, irq, &addr, &data);
2372 				if (error)
2373 					return (error);
2374 				pci_disable_msi(dev);
2375 				dinfo->cfg.msi.msi_addr = addr;
2376 				dinfo->cfg.msi.msi_data = data;
2377 				pci_enable_msi(dev, addr, data);
2378 				return (0);
2379 			}
2380 		}
2381 		return (ENOENT);
2382 	}
2383 
2384 	/*
2385 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2386 	 * we request the updated mapping info.  If that works, we go
2387 	 * through all the slots that use this IRQ and update them.
2388 	 */
2389 	if (cfg->msix.msix_alloc > 0) {
2390 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2391 			mv = &cfg->msix.msix_vectors[i];
2392 			if (mv->mv_irq == irq) {
2393 				error = PCIB_MAP_MSI(device_get_parent(bus),
2394 				    dev, irq, &addr, &data);
2395 				if (error)
2396 					return (error);
2397 				mv->mv_address = addr;
2398 				mv->mv_data = data;
2399 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2400 					mte = &cfg->msix.msix_table[j];
2401 					if (mte->mte_vector != i + 1)
2402 						continue;
2403 					if (mte->mte_handlers == 0)
2404 						continue;
2405 					pci_mask_msix(dev, j);
2406 					pci_enable_msix(dev, j, addr, data);
2407 					pci_unmask_msix(dev, j);
2408 				}
2409 			}
2410 		}
2411 		return (ENOENT);
2412 	}
2413 
2414 	return (ENOENT);
2415 }
2416 
2417 /*
2418  * Returns true if the specified device is blacklisted because MSI
2419  * doesn't work.
2420  */
2421 int
2422 pci_msi_device_blacklisted(device_t dev)
2423 {
2424 
2425 	if (!pci_honor_msi_blacklist)
2426 		return (0);
2427 
2428 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2429 }
2430 
2431 /*
2432  * Determine if MSI is blacklisted globally on this system.  Currently,
2433  * we just check for blacklisted chipsets as represented by the
2434  * host-PCI bridge at device 0:0:0.  In the future, it may become
2435  * necessary to check other system attributes, such as the kenv values
2436  * that give the motherboard manufacturer and model number.
2437  */
2438 static int
2439 pci_msi_blacklisted(void)
2440 {
2441 	device_t dev;
2442 
2443 	if (!pci_honor_msi_blacklist)
2444 		return (0);
2445 
2446 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2447 	if (!(pcie_chipset || pcix_chipset)) {
2448 		if (vm_guest != VM_GUEST_NO) {
2449 			/*
2450 			 * Whitelist older chipsets in virtual
2451 			 * machines known to support MSI.
2452 			 */
2453 			dev = pci_find_bsf(0, 0, 0);
2454 			if (dev != NULL)
2455 				return (!pci_has_quirk(pci_get_devid(dev),
2456 					PCI_QUIRK_ENABLE_MSI_VM));
2457 		}
2458 		return (1);
2459 	}
2460 
2461 	dev = pci_find_bsf(0, 0, 0);
2462 	if (dev != NULL)
2463 		return (pci_msi_device_blacklisted(dev));
2464 	return (0);
2465 }
2466 
2467 /*
2468  * Returns true if the specified device is blacklisted because MSI-X
2469  * doesn't work.  Note that this assumes that if MSI doesn't work,
2470  * MSI-X doesn't either.
2471  */
2472 int
2473 pci_msix_device_blacklisted(device_t dev)
2474 {
2475 
2476 	if (!pci_honor_msi_blacklist)
2477 		return (0);
2478 
2479 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2480 		return (1);
2481 
2482 	return (pci_msi_device_blacklisted(dev));
2483 }
2484 
2485 /*
2486  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2487  * is blacklisted, assume that MSI-X is as well.  Check for additional
2488  * chipsets where MSI works but MSI-X does not.
2489  */
2490 static int
2491 pci_msix_blacklisted(void)
2492 {
2493 	device_t dev;
2494 
2495 	if (!pci_honor_msi_blacklist)
2496 		return (0);
2497 
2498 	dev = pci_find_bsf(0, 0, 0);
2499 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2500 	    PCI_QUIRK_DISABLE_MSIX))
2501 		return (1);
2502 
2503 	return (pci_msi_blacklisted());
2504 }
2505 
2506 /*
2507  * Attempt to allocate *count MSI messages.  The actual number allocated is
2508  * returned in *count.  After this function returns, each message will be
2509  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2510  */
2511 int
2512 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2513 {
2514 	struct pci_devinfo *dinfo = device_get_ivars(child);
2515 	pcicfgregs *cfg = &dinfo->cfg;
2516 	struct resource_list_entry *rle;
2517 	int actual, error, i, irqs[32];
2518 	uint16_t ctrl;
2519 
2520 	/* Don't let count == 0 get us into trouble. */
2521 	if (*count == 0)
2522 		return (EINVAL);
2523 
2524 	/* If rid 0 is allocated, then fail. */
2525 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2526 	if (rle != NULL && rle->res != NULL)
2527 		return (ENXIO);
2528 
2529 	/* Already have allocated messages? */
2530 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2531 		return (ENXIO);
2532 
2533 	/* If MSI is blacklisted for this system, fail. */
2534 	if (pci_msi_blacklisted())
2535 		return (ENXIO);
2536 
2537 	/* MSI capability present? */
2538 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2539 		return (ENODEV);
2540 
2541 	if (bootverbose)
2542 		device_printf(child,
2543 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2544 		    *count, cfg->msi.msi_msgnum);
2545 
2546 	/* Don't ask for more than the device supports. */
2547 	actual = min(*count, cfg->msi.msi_msgnum);
2548 
2549 	/* Don't ask for more than 32 messages. */
2550 	actual = min(actual, 32);
2551 
2552 	/* MSI requires power of 2 number of messages. */
2553 	if (!powerof2(actual))
2554 		return (EINVAL);
2555 
2556 	for (;;) {
2557 		/* Try to allocate N messages. */
2558 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2559 		    actual, irqs);
2560 		if (error == 0)
2561 			break;
2562 		if (actual == 1)
2563 			return (error);
2564 
2565 		/* Try N / 2. */
2566 		actual >>= 1;
2567 	}
2568 
2569 	/*
2570 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2571 	 * resources in the irqs[] array, so add new resources
2572 	 * starting at rid 1.
2573 	 */
2574 	for (i = 0; i < actual; i++)
2575 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2576 		    irqs[i], irqs[i], 1);
2577 
2578 	if (bootverbose) {
2579 		if (actual == 1)
2580 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2581 		else {
2582 			int run;
2583 
2584 			/*
2585 			 * Be fancy and try to print contiguous runs
2586 			 * of IRQ values as ranges.  'run' is true if
2587 			 * we are in a range.
2588 			 */
2589 			device_printf(child, "using IRQs %d", irqs[0]);
2590 			run = 0;
2591 			for (i = 1; i < actual; i++) {
2592 
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 		/*
3863 		 * TODO: Ignore EA-BAR if is not enabled.
3864 		 *   Currently the EA implementation supports
3865 		 *   only situation, where EA structure contains
3866 		 *   predefined entries. In case they are not enabled
3867 		 *   leave them unallocated and proceed with
3868 		 *   a legacy-BAR mechanism.
3869 		 */
3870 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3871 			continue;
3872 
3873 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3874 		case PCIM_EA_P_MEM_PREFETCH:
3875 		case PCIM_EA_P_VF_MEM_PREFETCH:
3876 			flags = RF_PREFETCHABLE;
3877 			/* FALLTHROUGH */
3878 		case PCIM_EA_P_VF_MEM:
3879 		case PCIM_EA_P_MEM:
3880 			type = SYS_RES_MEMORY;
3881 			break;
3882 		case PCIM_EA_P_IO:
3883 			type = SYS_RES_IOPORT;
3884 			break;
3885 		default:
3886 			continue;
3887 		}
3888 
3889 		if (alloc_iov != 0) {
3890 #ifdef PCI_IOV
3891 			/* Allocating IOV, confirm BEI matches */
3892 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3893 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3894 				continue;
3895 #else
3896 			continue;
3897 #endif
3898 		} else {
3899 			/* Allocating BAR, confirm BEI matches */
3900 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3901 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3902 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3903 				continue;
3904 		}
3905 
3906 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3907 		if (rid < 0)
3908 			continue;
3909 
3910 		/* Skip resources already allocated by EA */
3911 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3912 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3913 			continue;
3914 
3915 		start = ea->eae_base;
3916 		count = ea->eae_max_offset + 1;
3917 #ifdef PCI_IOV
3918 		if (iov != NULL)
3919 			count = count * iov->iov_num_vfs;
3920 #endif
3921 		end = start + count - 1;
3922 		if (count == 0)
3923 			continue;
3924 
3925 		resource_list_add(rl, type, rid, start, end, count);
3926 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3927 		    flags);
3928 		if (res == NULL) {
3929 			resource_list_delete(rl, type, rid);
3930 
3931 			/*
3932 			 * Failed to allocate using EA, disable entry.
3933 			 * Another attempt to allocation will be performed
3934 			 * further, but this time using legacy BAR registers
3935 			 */
3936 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3937 			tmp &= ~PCIM_EA_ENABLE;
3938 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3939 
3940 			/*
3941 			 * Disabling entry might fail in case it is hardwired.
3942 			 * Read flags again to match current status.
3943 			 */
3944 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3945 
3946 			continue;
3947 		}
3948 
3949 		/* As per specification, fill BAR with zeros */
3950 		pci_write_config(dev, rid, 0, 4);
3951 	}
3952 }
3953 
3954 void
3955 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3956 {
3957 	struct pci_devinfo *dinfo;
3958 	pcicfgregs *cfg;
3959 	struct resource_list *rl;
3960 	const struct pci_quirk *q;
3961 	uint32_t devid;
3962 	int i;
3963 
3964 	dinfo = device_get_ivars(dev);
3965 	cfg = &dinfo->cfg;
3966 	rl = &dinfo->resources;
3967 	devid = (cfg->device << 16) | cfg->vendor;
3968 
3969 	/* Allocate resources using Enhanced Allocation */
3970 	pci_add_resources_ea(bus, dev, 0);
3971 
3972 	/* ATA devices needs special map treatment */
3973 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3974 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3975 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3976 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3977 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3978 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3979 	else
3980 		for (i = 0; i < cfg->nummaps;) {
3981 			/* Skip resources already managed by EA */
3982 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3983 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3984 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3985 				i++;
3986 				continue;
3987 			}
3988 
3989 			/*
3990 			 * Skip quirked resources.
3991 			 */
3992 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3993 				if (q->devid == devid &&
3994 				    q->type == PCI_QUIRK_UNMAP_REG &&
3995 				    q->arg1 == PCIR_BAR(i))
3996 					break;
3997 			if (q->devid != 0) {
3998 				i++;
3999 				continue;
4000 			}
4001 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4002 			    prefetchmask & (1 << i));
4003 		}
4004 
4005 	/*
4006 	 * Add additional, quirked resources.
4007 	 */
4008 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4009 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4010 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4011 
4012 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4013 #ifdef __PCI_REROUTE_INTERRUPT
4014 		/*
4015 		 * Try to re-route interrupts. Sometimes the BIOS or
4016 		 * firmware may leave bogus values in these registers.
4017 		 * If the re-route fails, then just stick with what we
4018 		 * have.
4019 		 */
4020 		pci_assign_interrupt(bus, dev, 1);
4021 #else
4022 		pci_assign_interrupt(bus, dev, 0);
4023 #endif
4024 	}
4025 
4026 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4027 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4028 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4029 			xhci_early_takeover(dev);
4030 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4031 			ehci_early_takeover(dev);
4032 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4033 			ohci_early_takeover(dev);
4034 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4035 			uhci_early_takeover(dev);
4036 	}
4037 
4038 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4039 	/*
4040 	 * Reserve resources for secondary bus ranges behind bridge
4041 	 * devices.
4042 	 */
4043 	pci_reserve_secbus(bus, dev, cfg, rl);
4044 #endif
4045 }
4046 
4047 static struct pci_devinfo *
4048 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4049     int slot, int func)
4050 {
4051 	struct pci_devinfo *dinfo;
4052 
4053 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4054 	if (dinfo != NULL)
4055 		pci_add_child(dev, dinfo);
4056 
4057 	return (dinfo);
4058 }
4059 
4060 void
4061 pci_add_children(device_t dev, int domain, int busno)
4062 {
4063 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4064 	device_t pcib = device_get_parent(dev);
4065 	struct pci_devinfo *dinfo;
4066 	int maxslots;
4067 	int s, f, pcifunchigh;
4068 	uint8_t hdrtype;
4069 	int first_func;
4070 
4071 	/*
4072 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4073 	 * enable ARI.  We must enable ARI before detecting the rest of the
4074 	 * functions on this bus as ARI changes the set of slots and functions
4075 	 * that are legal on this bus.
4076 	 */
4077 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4078 	if (dinfo != NULL && pci_enable_ari)
4079 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4080 
4081 	/*
4082 	 * Start looking for new devices on slot 0 at function 1 because we
4083 	 * just identified the device at slot 0, function 0.
4084 	 */
4085 	first_func = 1;
4086 
4087 	maxslots = PCIB_MAXSLOTS(pcib);
4088 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4089 		pcifunchigh = 0;
4090 		f = 0;
4091 		DELAY(1);
4092 		hdrtype = REG(PCIR_HDRTYPE, 1);
4093 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4094 			continue;
4095 		if (hdrtype & PCIM_MFDEV)
4096 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4097 		for (f = first_func; f <= pcifunchigh; f++)
4098 			pci_identify_function(pcib, dev, domain, busno, s, f);
4099 	}
4100 #undef REG
4101 }
4102 
4103 int
4104 pci_rescan_method(device_t dev)
4105 {
4106 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4107 	device_t pcib = device_get_parent(dev);
4108 	device_t child, *devlist, *unchanged;
4109 	int devcount, error, i, j, maxslots, oldcount;
4110 	int busno, domain, s, f, pcifunchigh;
4111 	uint8_t hdrtype;
4112 
4113 	/* No need to check for ARI on a rescan. */
4114 	error = device_get_children(dev, &devlist, &devcount);
4115 	if (error)
4116 		return (error);
4117 	if (devcount != 0) {
4118 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4119 		    M_NOWAIT | M_ZERO);
4120 		if (unchanged == NULL) {
4121 			free(devlist, M_TEMP);
4122 			return (ENOMEM);
4123 		}
4124 	} else
4125 		unchanged = NULL;
4126 
4127 	domain = pcib_get_domain(dev);
4128 	busno = pcib_get_bus(dev);
4129 	maxslots = PCIB_MAXSLOTS(pcib);
4130 	for (s = 0; s <= maxslots; s++) {
4131 		/* If function 0 is not present, skip to the next slot. */
4132 		f = 0;
4133 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4134 			continue;
4135 		pcifunchigh = 0;
4136 		hdrtype = REG(PCIR_HDRTYPE, 1);
4137 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4138 			continue;
4139 		if (hdrtype & PCIM_MFDEV)
4140 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4141 		for (f = 0; f <= pcifunchigh; f++) {
4142 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4143 				continue;
4144 
4145 			/*
4146 			 * Found a valid function.  Check if a
4147 			 * device_t for this device already exists.
4148 			 */
4149 			for (i = 0; i < devcount; i++) {
4150 				child = devlist[i];
4151 				if (child == NULL)
4152 					continue;
4153 				if (pci_get_slot(child) == s &&
4154 				    pci_get_function(child) == f) {
4155 					unchanged[i] = child;
4156 					goto next_func;
4157 				}
4158 			}
4159 
4160 			pci_identify_function(pcib, dev, domain, busno, s, f);
4161 		next_func:;
4162 		}
4163 	}
4164 
4165 	/* Remove devices that are no longer present. */
4166 	for (i = 0; i < devcount; i++) {
4167 		if (unchanged[i] != NULL)
4168 			continue;
4169 		device_delete_child(dev, devlist[i]);
4170 	}
4171 
4172 	free(devlist, M_TEMP);
4173 	oldcount = devcount;
4174 
4175 	/* Try to attach the devices just added. */
4176 	error = device_get_children(dev, &devlist, &devcount);
4177 	if (error) {
4178 		free(unchanged, M_TEMP);
4179 		return (error);
4180 	}
4181 
4182 	for (i = 0; i < devcount; i++) {
4183 		for (j = 0; j < oldcount; j++) {
4184 			if (devlist[i] == unchanged[j])
4185 				goto next_device;
4186 		}
4187 
4188 		device_probe_and_attach(devlist[i]);
4189 	next_device:;
4190 	}
4191 
4192 	free(unchanged, M_TEMP);
4193 	free(devlist, M_TEMP);
4194 	return (0);
4195 #undef REG
4196 }
4197 
4198 #ifdef PCI_IOV
4199 device_t
4200 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4201     uint16_t did)
4202 {
4203 	struct pci_devinfo *vf_dinfo;
4204 	device_t pcib;
4205 	int busno, slot, func;
4206 
4207 	pcib = device_get_parent(bus);
4208 
4209 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4210 
4211 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4212 	    slot, func, vid, did);
4213 
4214 	vf_dinfo->cfg.flags |= PCICFG_VF;
4215 	pci_add_child(bus, vf_dinfo);
4216 
4217 	return (vf_dinfo->cfg.dev);
4218 }
4219 
4220 device_t
4221 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4222     uint16_t vid, uint16_t did)
4223 {
4224 
4225 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4226 }
4227 #endif
4228 
4229 static void
4230 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4231 {
4232 	int aer;
4233 	uint32_t r;
4234 	uint16_t r2;
4235 
4236 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4237 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4238 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4239 		    PCIER_ROOT_CTL, 2);
4240 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4241 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4242 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4243 		    PCIER_ROOT_CTL, r2, 2);
4244 	}
4245 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4246 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4247 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4248 		if (r != 0 && bootverbose) {
4249 			pci_printf(&dinfo->cfg,
4250 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4251 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4252 			    4));
4253 		}
4254 
4255 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4256 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4257 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4258 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4259 		    PCIM_AER_UC_POISONED_TLP |
4260 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4261 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4262 		    PCIM_AER_UC_COMPLETER_ABORT |
4263 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4264 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4265 		    PCIM_AER_UC_MALFORMED_TLP |
4266 		    PCIM_AER_UC_ECRC_ERROR |
4267 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4268 		    PCIM_AER_UC_ACS_VIOLATION |
4269 		    PCIM_AER_UC_INTERNAL_ERROR |
4270 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4271 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4272 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4273 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4274 
4275 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4276 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4277 		if (r != 0 && bootverbose) {
4278 			pci_printf(&dinfo->cfg,
4279 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4280 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4281 			    4));
4282 		}
4283 
4284 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4285 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4286 		    PCIM_AER_COR_BAD_TLP |
4287 		    PCIM_AER_COR_BAD_DLLP |
4288 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4289 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4290 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4291 		    PCIM_AER_COR_INTERNAL_ERROR |
4292 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4293 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4294 
4295 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4296 		    PCIER_DEVICE_CTL, 2);
4297 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4298 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4299 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4300 		    PCIER_DEVICE_CTL, r, 2);
4301 	}
4302 }
4303 
4304 void
4305 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4306 {
4307 	device_t dev;
4308 
4309 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4310 	device_set_ivars(dev, dinfo);
4311 	resource_list_init(&dinfo->resources);
4312 	pci_cfg_save(dev, dinfo, 0);
4313 	pci_cfg_restore(dev, dinfo);
4314 	pci_print_verbose(dinfo);
4315 	pci_add_resources(bus, dev, 0, 0);
4316 	pci_child_added(dinfo->cfg.dev);
4317 
4318 	if (pci_clear_aer_on_attach)
4319 		pci_add_child_clear_aer(dev, dinfo);
4320 
4321 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4322 }
4323 
4324 void
4325 pci_child_added_method(device_t dev, device_t child)
4326 {
4327 
4328 }
4329 
4330 static int
4331 pci_probe(device_t dev)
4332 {
4333 
4334 	device_set_desc(dev, "PCI bus");
4335 
4336 	/* Allow other subclasses to override this driver. */
4337 	return (BUS_PROBE_GENERIC);
4338 }
4339 
4340 int
4341 pci_attach_common(device_t dev)
4342 {
4343 	struct pci_softc *sc;
4344 	int busno, domain;
4345 #ifdef PCI_RES_BUS
4346 	int rid;
4347 #endif
4348 
4349 	sc = device_get_softc(dev);
4350 	domain = pcib_get_domain(dev);
4351 	busno = pcib_get_bus(dev);
4352 #ifdef PCI_RES_BUS
4353 	rid = 0;
4354 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4355 	    1, 0);
4356 	if (sc->sc_bus == NULL) {
4357 		device_printf(dev, "failed to allocate bus number\n");
4358 		return (ENXIO);
4359 	}
4360 #endif
4361 	if (bootverbose)
4362 		device_printf(dev, "domain=%d, physical bus=%d\n",
4363 		    domain, busno);
4364 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4365 	return (0);
4366 }
4367 
4368 int
4369 pci_attach(device_t dev)
4370 {
4371 	int busno, domain, error;
4372 
4373 	error = pci_attach_common(dev);
4374 	if (error)
4375 		return (error);
4376 
4377 	/*
4378 	 * Since there can be multiple independently numbered PCI
4379 	 * buses on systems with multiple PCI domains, we can't use
4380 	 * the unit number to decide which bus we are probing. We ask
4381 	 * the parent pcib what our domain and bus numbers are.
4382 	 */
4383 	domain = pcib_get_domain(dev);
4384 	busno = pcib_get_bus(dev);
4385 	pci_add_children(dev, domain, busno);
4386 	return (bus_generic_attach(dev));
4387 }
4388 
4389 int
4390 pci_detach(device_t dev)
4391 {
4392 #ifdef PCI_RES_BUS
4393 	struct pci_softc *sc;
4394 #endif
4395 	int error;
4396 
4397 	error = bus_generic_detach(dev);
4398 	if (error)
4399 		return (error);
4400 #ifdef PCI_RES_BUS
4401 	sc = device_get_softc(dev);
4402 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4403 	if (error)
4404 		return (error);
4405 #endif
4406 	return (device_delete_children(dev));
4407 }
4408 
4409 static void
4410 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4411 {
4412 	int line, unit;
4413 	const char *at;
4414 	char me1[24], me2[32];
4415 	uint8_t b, s, f;
4416 	uint32_t d;
4417 
4418 	d = pci_get_domain(child);
4419 	b = pci_get_bus(child);
4420 	s = pci_get_slot(child);
4421 	f = pci_get_function(child);
4422 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4423 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4424 	line = 0;
4425 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4426 		resource_string_value(name, unit, "at", &at);
4427 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4428 			continue; /* No match, try next candidate */
4429 		*unitp = unit;
4430 		return;
4431 	}
4432 }
4433 
4434 static void
4435 pci_set_power_child(device_t dev, device_t child, int state)
4436 {
4437 	device_t pcib;
4438 	int dstate;
4439 
4440 	/*
4441 	 * Set the device to the given state.  If the firmware suggests
4442 	 * a different power state, use it instead.  If power management
4443 	 * is not present, the firmware is responsible for managing
4444 	 * device power.  Skip children who aren't attached since they
4445 	 * are handled separately.
4446 	 */
4447 	pcib = device_get_parent(dev);
4448 	dstate = state;
4449 	if (device_is_attached(child) &&
4450 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4451 		pci_set_powerstate(child, dstate);
4452 }
4453 
4454 int
4455 pci_suspend_child(device_t dev, device_t child)
4456 {
4457 	struct pci_devinfo *dinfo;
4458 	struct resource_list_entry *rle;
4459 	int error;
4460 
4461 	dinfo = device_get_ivars(child);
4462 
4463 	/*
4464 	 * Save the PCI configuration space for the child and set the
4465 	 * device in the appropriate power state for this sleep state.
4466 	 */
4467 	pci_cfg_save(child, dinfo, 0);
4468 
4469 	/* Suspend devices before potentially powering them down. */
4470 	error = bus_generic_suspend_child(dev, child);
4471 
4472 	if (error)
4473 		return (error);
4474 
4475 	if (pci_do_power_suspend) {
4476 		/*
4477 		 * Make sure this device's interrupt handler is not invoked
4478 		 * in the case the device uses a shared interrupt that can
4479 		 * be raised by some other device.
4480 		 * This is applicable only to regular (legacy) PCI interrupts
4481 		 * as MSI/MSI-X interrupts are never shared.
4482 		 */
4483 		rle = resource_list_find(&dinfo->resources,
4484 		    SYS_RES_IRQ, 0);
4485 		if (rle != NULL && rle->res != NULL)
4486 			(void)bus_suspend_intr(child, rle->res);
4487 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4488 	}
4489 
4490 	return (0);
4491 }
4492 
4493 int
4494 pci_resume_child(device_t dev, device_t child)
4495 {
4496 	struct pci_devinfo *dinfo;
4497 	struct resource_list_entry *rle;
4498 
4499 	if (pci_do_power_resume)
4500 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4501 
4502 	dinfo = device_get_ivars(child);
4503 	pci_cfg_restore(child, dinfo);
4504 	if (!device_is_attached(child))
4505 		pci_cfg_save(child, dinfo, 1);
4506 
4507 	bus_generic_resume_child(dev, child);
4508 
4509 	/*
4510 	 * Allow interrupts only after fully resuming the driver and hardware.
4511 	 */
4512 	if (pci_do_power_suspend) {
4513 		/* See pci_suspend_child for details. */
4514 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4515 		if (rle != NULL && rle->res != NULL)
4516 			(void)bus_resume_intr(child, rle->res);
4517 	}
4518 
4519 	return (0);
4520 }
4521 
4522 int
4523 pci_resume(device_t dev)
4524 {
4525 	device_t child, *devlist;
4526 	int error, i, numdevs;
4527 
4528 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4529 		return (error);
4530 
4531 	/*
4532 	 * Resume critical devices first, then everything else later.
4533 	 */
4534 	for (i = 0; i < numdevs; i++) {
4535 		child = devlist[i];
4536 		switch (pci_get_class(child)) {
4537 		case PCIC_DISPLAY:
4538 		case PCIC_MEMORY:
4539 		case PCIC_BRIDGE:
4540 		case PCIC_BASEPERIPH:
4541 			BUS_RESUME_CHILD(dev, child);
4542 			break;
4543 		}
4544 	}
4545 	for (i = 0; i < numdevs; i++) {
4546 		child = devlist[i];
4547 		switch (pci_get_class(child)) {
4548 		case PCIC_DISPLAY:
4549 		case PCIC_MEMORY:
4550 		case PCIC_BRIDGE:
4551 		case PCIC_BASEPERIPH:
4552 			break;
4553 		default:
4554 			BUS_RESUME_CHILD(dev, child);
4555 		}
4556 	}
4557 	free(devlist, M_TEMP);
4558 	return (0);
4559 }
4560 
4561 static void
4562 pci_load_vendor_data(void)
4563 {
4564 	caddr_t data;
4565 	void *ptr;
4566 	size_t sz;
4567 
4568 	data = preload_search_by_type("pci_vendor_data");
4569 	if (data != NULL) {
4570 		ptr = preload_fetch_addr(data);
4571 		sz = preload_fetch_size(data);
4572 		if (ptr != NULL && sz != 0) {
4573 			pci_vendordata = ptr;
4574 			pci_vendordata_size = sz;
4575 			/* terminate the database */
4576 			pci_vendordata[pci_vendordata_size] = '\n';
4577 		}
4578 	}
4579 }
4580 
4581 void
4582 pci_driver_added(device_t dev, driver_t *driver)
4583 {
4584 	int numdevs;
4585 	device_t *devlist;
4586 	device_t child;
4587 	struct pci_devinfo *dinfo;
4588 	int i;
4589 
4590 	if (bootverbose)
4591 		device_printf(dev, "driver added\n");
4592 	DEVICE_IDENTIFY(driver, dev);
4593 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4594 		return;
4595 	for (i = 0; i < numdevs; i++) {
4596 		child = devlist[i];
4597 		if (device_get_state(child) != DS_NOTPRESENT)
4598 			continue;
4599 		dinfo = device_get_ivars(child);
4600 		pci_print_verbose(dinfo);
4601 		if (bootverbose)
4602 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4603 		pci_cfg_restore(child, dinfo);
4604 		if (device_probe_and_attach(child) != 0)
4605 			pci_child_detached(dev, child);
4606 	}
4607 	free(devlist, M_TEMP);
4608 }
4609 
4610 int
4611 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4612     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4613 {
4614 	struct pci_devinfo *dinfo;
4615 	struct msix_table_entry *mte;
4616 	struct msix_vector *mv;
4617 	uint64_t addr;
4618 	uint32_t data;
4619 	void *cookie;
4620 	int error, rid;
4621 
4622 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4623 	    arg, &cookie);
4624 	if (error)
4625 		return (error);
4626 
4627 	/* If this is not a direct child, just bail out. */
4628 	if (device_get_parent(child) != dev) {
4629 		*cookiep = cookie;
4630 		return(0);
4631 	}
4632 
4633 	rid = rman_get_rid(irq);
4634 	if (rid == 0) {
4635 		/* Make sure that INTx is enabled */
4636 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4637 	} else {
4638 		/*
4639 		 * Check to see if the interrupt is MSI or MSI-X.
4640 		 * Ask our parent to map the MSI and give
4641 		 * us the address and data register values.
4642 		 * If we fail for some reason, teardown the
4643 		 * interrupt handler.
4644 		 */
4645 		dinfo = device_get_ivars(child);
4646 		if (dinfo->cfg.msi.msi_alloc > 0) {
4647 			if (dinfo->cfg.msi.msi_addr == 0) {
4648 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4649 			    ("MSI has handlers, but vectors not mapped"));
4650 				error = PCIB_MAP_MSI(device_get_parent(dev),
4651 				    child, rman_get_start(irq), &addr, &data);
4652 				if (error)
4653 					goto bad;
4654 				dinfo->cfg.msi.msi_addr = addr;
4655 				dinfo->cfg.msi.msi_data = data;
4656 			}
4657 			if (dinfo->cfg.msi.msi_handlers == 0)
4658 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4659 				    dinfo->cfg.msi.msi_data);
4660 			dinfo->cfg.msi.msi_handlers++;
4661 		} else {
4662 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4663 			    ("No MSI or MSI-X interrupts allocated"));
4664 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4665 			    ("MSI-X index too high"));
4666 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4667 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4668 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4669 			KASSERT(mv->mv_irq == rman_get_start(irq),
4670 			    ("IRQ mismatch"));
4671 			if (mv->mv_address == 0) {
4672 				KASSERT(mte->mte_handlers == 0,
4673 		    ("MSI-X table entry has handlers, but vector not mapped"));
4674 				error = PCIB_MAP_MSI(device_get_parent(dev),
4675 				    child, rman_get_start(irq), &addr, &data);
4676 				if (error)
4677 					goto bad;
4678 				mv->mv_address = addr;
4679 				mv->mv_data = data;
4680 			}
4681 
4682 			/*
4683 			 * The MSIX table entry must be made valid by
4684 			 * incrementing the mte_handlers before
4685 			 * calling pci_enable_msix() and
4686 			 * pci_resume_msix(). Else the MSIX rewrite
4687 			 * table quirk will not work as expected.
4688 			 */
4689 			mte->mte_handlers++;
4690 			if (mte->mte_handlers == 1) {
4691 				pci_enable_msix(child, rid - 1, mv->mv_address,
4692 				    mv->mv_data);
4693 				pci_unmask_msix(child, rid - 1);
4694 			}
4695 		}
4696 
4697 		/*
4698 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4699 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4700 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4701 		 */
4702 		if (!pci_has_quirk(pci_get_devid(child),
4703 		    PCI_QUIRK_MSI_INTX_BUG))
4704 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4705 		else
4706 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4707 	bad:
4708 		if (error) {
4709 			(void)bus_generic_teardown_intr(dev, child, irq,
4710 			    cookie);
4711 			return (error);
4712 		}
4713 	}
4714 	*cookiep = cookie;
4715 	return (0);
4716 }
4717 
4718 int
4719 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4720     void *cookie)
4721 {
4722 	struct msix_table_entry *mte;
4723 	struct resource_list_entry *rle;
4724 	struct pci_devinfo *dinfo;
4725 	int error, rid;
4726 
4727 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4728 		return (EINVAL);
4729 
4730 	/* If this isn't a direct child, just bail out */
4731 	if (device_get_parent(child) != dev)
4732 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4733 
4734 	rid = rman_get_rid(irq);
4735 	if (rid == 0) {
4736 		/* Mask INTx */
4737 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4738 	} else {
4739 		/*
4740 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4741 		 * decrement the appropriate handlers count and mask the
4742 		 * MSI-X message, or disable MSI messages if the count
4743 		 * drops to 0.
4744 		 */
4745 		dinfo = device_get_ivars(child);
4746 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4747 		if (rle->res != irq)
4748 			return (EINVAL);
4749 		if (dinfo->cfg.msi.msi_alloc > 0) {
4750 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4751 			    ("MSI-X index too high"));
4752 			if (dinfo->cfg.msi.msi_handlers == 0)
4753 				return (EINVAL);
4754 			dinfo->cfg.msi.msi_handlers--;
4755 			if (dinfo->cfg.msi.msi_handlers == 0)
4756 				pci_disable_msi(child);
4757 		} else {
4758 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4759 			    ("No MSI or MSI-X interrupts allocated"));
4760 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4761 			    ("MSI-X index too high"));
4762 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4763 			if (mte->mte_handlers == 0)
4764 				return (EINVAL);
4765 			mte->mte_handlers--;
4766 			if (mte->mte_handlers == 0)
4767 				pci_mask_msix(child, rid - 1);
4768 		}
4769 	}
4770 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4771 	if (rid > 0)
4772 		KASSERT(error == 0,
4773 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4774 	return (error);
4775 }
4776 
4777 int
4778 pci_print_child(device_t dev, device_t child)
4779 {
4780 	struct pci_devinfo *dinfo;
4781 	struct resource_list *rl;
4782 	int retval = 0;
4783 
4784 	dinfo = device_get_ivars(child);
4785 	rl = &dinfo->resources;
4786 
4787 	retval += bus_print_child_header(dev, child);
4788 
4789 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4790 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4791 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4792 	if (device_get_flags(dev))
4793 		retval += printf(" flags %#x", device_get_flags(dev));
4794 
4795 	retval += printf(" at device %d.%d", pci_get_slot(child),
4796 	    pci_get_function(child));
4797 
4798 	retval += bus_print_child_domain(dev, child);
4799 	retval += bus_print_child_footer(dev, child);
4800 
4801 	return (retval);
4802 }
4803 
4804 static const struct
4805 {
4806 	int		class;
4807 	int		subclass;
4808 	int		report; /* 0 = bootverbose, 1 = always */
4809 	const char	*desc;
4810 } pci_nomatch_tab[] = {
4811 	{PCIC_OLD,		-1,			1, "old"},
4812 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4813 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4814 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4815 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4816 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4817 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4818 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4819 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4820 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4821 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4822 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4823 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4824 	{PCIC_NETWORK,		-1,			1, "network"},
4825 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4826 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4827 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4828 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4829 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4830 	{PCIC_DISPLAY,		-1,			1, "display"},
4831 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4832 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4833 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4834 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4835 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4836 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4837 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4838 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4839 	{PCIC_MEMORY,		-1,			1, "memory"},
4840 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4841 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4842 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4843 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4844 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4845 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4846 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4847 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4848 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4849 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4850 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4851 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4852 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4853 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4854 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4855 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4856 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4857 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4858 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4859 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4860 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4861 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4862 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4863 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4864 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4865 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4866 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4867 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4868 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4869 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4870 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4871 	{PCIC_DOCKING,		-1,			1, "docking station"},
4872 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4873 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4874 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4875 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4876 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4877 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4878 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4879 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4880 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4881 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4882 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4883 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4884 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4885 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4886 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4887 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4888 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4889 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4890 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4891 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4892 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4893 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4894 	{PCIC_DASP,		-1,			0, "dasp"},
4895 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4896 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4897 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4898 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4899 	{0, 0, 0,		NULL}
4900 };
4901 
4902 void
4903 pci_probe_nomatch(device_t dev, device_t child)
4904 {
4905 	int i, report;
4906 	const char *cp, *scp;
4907 	char *device;
4908 
4909 	/*
4910 	 * Look for a listing for this device in a loaded device database.
4911 	 */
4912 	report = 1;
4913 	if ((device = pci_describe_device(child)) != NULL) {
4914 		device_printf(dev, "<%s>", device);
4915 		free(device, M_DEVBUF);
4916 	} else {
4917 		/*
4918 		 * Scan the class/subclass descriptions for a general
4919 		 * description.
4920 		 */
4921 		cp = "unknown";
4922 		scp = NULL;
4923 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4924 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4925 				if (pci_nomatch_tab[i].subclass == -1) {
4926 					cp = pci_nomatch_tab[i].desc;
4927 					report = pci_nomatch_tab[i].report;
4928 				} else if (pci_nomatch_tab[i].subclass ==
4929 				    pci_get_subclass(child)) {
4930 					scp = pci_nomatch_tab[i].desc;
4931 					report = pci_nomatch_tab[i].report;
4932 				}
4933 			}
4934 		}
4935 		if (report || bootverbose) {
4936 			device_printf(dev, "<%s%s%s>",
4937 			    cp ? cp : "",
4938 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4939 			    scp ? scp : "");
4940 		}
4941 	}
4942 	if (report || bootverbose) {
4943 		printf(" at device %d.%d (no driver attached)\n",
4944 		    pci_get_slot(child), pci_get_function(child));
4945 	}
4946 	pci_cfg_save(child, device_get_ivars(child), 1);
4947 }
4948 
4949 void
4950 pci_child_detached(device_t dev, device_t child)
4951 {
4952 	struct pci_devinfo *dinfo;
4953 	struct resource_list *rl;
4954 
4955 	dinfo = device_get_ivars(child);
4956 	rl = &dinfo->resources;
4957 
4958 	/*
4959 	 * Have to deallocate IRQs before releasing any MSI messages and
4960 	 * have to release MSI messages before deallocating any memory
4961 	 * BARs.
4962 	 */
4963 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4964 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4965 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4966 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4967 		(void)pci_release_msi(child);
4968 	}
4969 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4970 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4971 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4972 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4973 #ifdef PCI_RES_BUS
4974 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4975 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4976 #endif
4977 
4978 	pci_cfg_save(child, dinfo, 1);
4979 }
4980 
4981 /*
4982  * Parse the PCI device database, if loaded, and return a pointer to a
4983  * description of the device.
4984  *
4985  * The database is flat text formatted as follows:
4986  *
4987  * Any line not in a valid format is ignored.
4988  * Lines are terminated with newline '\n' characters.
4989  *
4990  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4991  * the vendor name.
4992  *
4993  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
4994  * - devices cannot be listed without a corresponding VENDOR line.
4995  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
4996  * another TAB, then the device name.
4997  */
4998 
4999 /*
5000  * Assuming (ptr) points to the beginning of a line in the database,
5001  * return the vendor or device and description of the next entry.
5002  * The value of (vendor) or (device) inappropriate for the entry type
5003  * is set to -1.  Returns nonzero at the end of the database.
5004  *
5005  * Note that this is slightly unrobust in the face of corrupt data;
5006  * we attempt to safeguard against this by spamming the end of the
5007  * database with a newline when we initialise.
5008  */
5009 static int
5010 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5011 {
5012 	char	*cp = *ptr;
5013 	int	left;
5014 
5015 	*device = -1;
5016 	*vendor = -1;
5017 	**desc = '\0';
5018 	for (;;) {
5019 		left = pci_vendordata_size - (cp - pci_vendordata);
5020 		if (left <= 0) {
5021 			*ptr = cp;
5022 			return(1);
5023 		}
5024 
5025 		/* vendor entry? */
5026 		if (*cp != '\t' &&
5027 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5028 			break;
5029 		/* device entry? */
5030 		if (*cp == '\t' &&
5031 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5032 			break;
5033 
5034 		/* skip to next line */
5035 		while (*cp != '\n' && left > 0) {
5036 			cp++;
5037 			left--;
5038 		}
5039 		if (*cp == '\n') {
5040 			cp++;
5041 			left--;
5042 		}
5043 	}
5044 	/* skip to next line */
5045 	while (*cp != '\n' && left > 0) {
5046 		cp++;
5047 		left--;
5048 	}
5049 	if (*cp == '\n' && left > 0)
5050 		cp++;
5051 	*ptr = cp;
5052 	return(0);
5053 }
5054 
5055 static char *
5056 pci_describe_device(device_t dev)
5057 {
5058 	int	vendor, device;
5059 	char	*desc, *vp, *dp, *line;
5060 
5061 	desc = vp = dp = NULL;
5062 
5063 	/*
5064 	 * If we have no vendor data, we can't do anything.
5065 	 */
5066 	if (pci_vendordata == NULL)
5067 		goto out;
5068 
5069 	/*
5070 	 * Scan the vendor data looking for this device
5071 	 */
5072 	line = pci_vendordata;
5073 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5074 		goto out;
5075 	for (;;) {
5076 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5077 			goto out;
5078 		if (vendor == pci_get_vendor(dev))
5079 			break;
5080 	}
5081 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5082 		goto out;
5083 	for (;;) {
5084 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5085 			*dp = 0;
5086 			break;
5087 		}
5088 		if (vendor != -1) {
5089 			*dp = 0;
5090 			break;
5091 		}
5092 		if (device == pci_get_device(dev))
5093 			break;
5094 	}
5095 	if (dp[0] == '\0')
5096 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5097 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5098 	    NULL)
5099 		sprintf(desc, "%s, %s", vp, dp);
5100 out:
5101 	if (vp != NULL)
5102 		free(vp, M_DEVBUF);
5103 	if (dp != NULL)
5104 		free(dp, M_DEVBUF);
5105 	return(desc);
5106 }
5107 
5108 int
5109 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5110 {
5111 	struct pci_devinfo *dinfo;
5112 	pcicfgregs *cfg;
5113 
5114 	dinfo = device_get_ivars(child);
5115 	cfg = &dinfo->cfg;
5116 
5117 	switch (which) {
5118 	case PCI_IVAR_ETHADDR:
5119 		/*
5120 		 * The generic accessor doesn't deal with failure, so
5121 		 * we set the return value, then return an error.
5122 		 */
5123 		*((uint8_t **) result) = NULL;
5124 		return (EINVAL);
5125 	case PCI_IVAR_SUBVENDOR:
5126 		*result = cfg->subvendor;
5127 		break;
5128 	case PCI_IVAR_SUBDEVICE:
5129 		*result = cfg->subdevice;
5130 		break;
5131 	case PCI_IVAR_VENDOR:
5132 		*result = cfg->vendor;
5133 		break;
5134 	case PCI_IVAR_DEVICE:
5135 		*result = cfg->device;
5136 		break;
5137 	case PCI_IVAR_DEVID:
5138 		*result = (cfg->device << 16) | cfg->vendor;
5139 		break;
5140 	case PCI_IVAR_CLASS:
5141 		*result = cfg->baseclass;
5142 		break;
5143 	case PCI_IVAR_SUBCLASS:
5144 		*result = cfg->subclass;
5145 		break;
5146 	case PCI_IVAR_PROGIF:
5147 		*result = cfg->progif;
5148 		break;
5149 	case PCI_IVAR_REVID:
5150 		*result = cfg->revid;
5151 		break;
5152 	case PCI_IVAR_INTPIN:
5153 		*result = cfg->intpin;
5154 		break;
5155 	case PCI_IVAR_IRQ:
5156 		*result = cfg->intline;
5157 		break;
5158 	case PCI_IVAR_DOMAIN:
5159 		*result = cfg->domain;
5160 		break;
5161 	case PCI_IVAR_BUS:
5162 		*result = cfg->bus;
5163 		break;
5164 	case PCI_IVAR_SLOT:
5165 		*result = cfg->slot;
5166 		break;
5167 	case PCI_IVAR_FUNCTION:
5168 		*result = cfg->func;
5169 		break;
5170 	case PCI_IVAR_CMDREG:
5171 		*result = cfg->cmdreg;
5172 		break;
5173 	case PCI_IVAR_CACHELNSZ:
5174 		*result = cfg->cachelnsz;
5175 		break;
5176 	case PCI_IVAR_MINGNT:
5177 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5178 			*result = -1;
5179 			return (EINVAL);
5180 		}
5181 		*result = cfg->mingnt;
5182 		break;
5183 	case PCI_IVAR_MAXLAT:
5184 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5185 			*result = -1;
5186 			return (EINVAL);
5187 		}
5188 		*result = cfg->maxlat;
5189 		break;
5190 	case PCI_IVAR_LATTIMER:
5191 		*result = cfg->lattimer;
5192 		break;
5193 	default:
5194 		return (ENOENT);
5195 	}
5196 	return (0);
5197 }
5198 
5199 int
5200 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5201 {
5202 	struct pci_devinfo *dinfo;
5203 
5204 	dinfo = device_get_ivars(child);
5205 
5206 	switch (which) {
5207 	case PCI_IVAR_INTPIN:
5208 		dinfo->cfg.intpin = value;
5209 		return (0);
5210 	case PCI_IVAR_ETHADDR:
5211 	case PCI_IVAR_SUBVENDOR:
5212 	case PCI_IVAR_SUBDEVICE:
5213 	case PCI_IVAR_VENDOR:
5214 	case PCI_IVAR_DEVICE:
5215 	case PCI_IVAR_DEVID:
5216 	case PCI_IVAR_CLASS:
5217 	case PCI_IVAR_SUBCLASS:
5218 	case PCI_IVAR_PROGIF:
5219 	case PCI_IVAR_REVID:
5220 	case PCI_IVAR_IRQ:
5221 	case PCI_IVAR_DOMAIN:
5222 	case PCI_IVAR_BUS:
5223 	case PCI_IVAR_SLOT:
5224 	case PCI_IVAR_FUNCTION:
5225 		return (EINVAL);	/* disallow for now */
5226 
5227 	default:
5228 		return (ENOENT);
5229 	}
5230 }
5231 
5232 #include "opt_ddb.h"
5233 #ifdef DDB
5234 #include <ddb/ddb.h>
5235 #include <sys/cons.h>
5236 
5237 /*
5238  * List resources based on pci map registers, used for within ddb
5239  */
5240 
5241 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5242 {
5243 	struct pci_devinfo *dinfo;
5244 	struct devlist *devlist_head;
5245 	struct pci_conf *p;
5246 	const char *name;
5247 	int i, error, none_count;
5248 
5249 	none_count = 0;
5250 	/* get the head of the device queue */
5251 	devlist_head = &pci_devq;
5252 
5253 	/*
5254 	 * Go through the list of devices and print out devices
5255 	 */
5256 	for (error = 0, i = 0,
5257 	     dinfo = STAILQ_FIRST(devlist_head);
5258 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5259 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5260 
5261 		/* Populate pd_name and pd_unit */
5262 		name = NULL;
5263 		if (dinfo->cfg.dev)
5264 			name = device_get_name(dinfo->cfg.dev);
5265 
5266 		p = &dinfo->conf;
5267 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5268 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5269 			(name && *name) ? name : "none",
5270 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5271 			none_count++,
5272 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5273 			p->pc_sel.pc_func, (p->pc_class << 16) |
5274 			(p->pc_subclass << 8) | p->pc_progif,
5275 			(p->pc_subdevice << 16) | p->pc_subvendor,
5276 			(p->pc_device << 16) | p->pc_vendor,
5277 			p->pc_revid, p->pc_hdr);
5278 	}
5279 }
5280 #endif /* DDB */
5281 
5282 static struct resource *
5283 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5284     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5285     u_int flags)
5286 {
5287 	struct pci_devinfo *dinfo = device_get_ivars(child);
5288 	struct resource_list *rl = &dinfo->resources;
5289 	struct resource *res;
5290 	struct pci_map *pm;
5291 	uint16_t cmd;
5292 	pci_addr_t map, testval;
5293 	int mapsize;
5294 
5295 	res = NULL;
5296 
5297 	/* If rid is managed by EA, ignore it */
5298 	if (pci_ea_is_enabled(child, *rid))
5299 		goto out;
5300 
5301 	pm = pci_find_bar(child, *rid);
5302 	if (pm != NULL) {
5303 		/* This is a BAR that we failed to allocate earlier. */
5304 		mapsize = pm->pm_size;
5305 		map = pm->pm_value;
5306 	} else {
5307 		/*
5308 		 * Weed out the bogons, and figure out how large the
5309 		 * BAR/map is.  BARs that read back 0 here are bogus
5310 		 * and unimplemented.  Note: atapci in legacy mode are
5311 		 * special and handled elsewhere in the code.  If you
5312 		 * have a atapci device in legacy mode and it fails
5313 		 * here, that other code is broken.
5314 		 */
5315 		pci_read_bar(child, *rid, &map, &testval, NULL);
5316 
5317 		/*
5318 		 * Determine the size of the BAR and ignore BARs with a size
5319 		 * of 0.  Device ROM BARs use a different mask value.
5320 		 */
5321 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5322 			mapsize = pci_romsize(testval);
5323 		else
5324 			mapsize = pci_mapsize(testval);
5325 		if (mapsize == 0)
5326 			goto out;
5327 		pm = pci_add_bar(child, *rid, map, mapsize);
5328 	}
5329 
5330 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5331 		if (type != SYS_RES_MEMORY) {
5332 			if (bootverbose)
5333 				device_printf(dev,
5334 				    "child %s requested type %d for rid %#x,"
5335 				    " but the BAR says it is an memio\n",
5336 				    device_get_nameunit(child), type, *rid);
5337 			goto out;
5338 		}
5339 	} else {
5340 		if (type != SYS_RES_IOPORT) {
5341 			if (bootverbose)
5342 				device_printf(dev,
5343 				    "child %s requested type %d for rid %#x,"
5344 				    " but the BAR says it is an ioport\n",
5345 				    device_get_nameunit(child), type, *rid);
5346 			goto out;
5347 		}
5348 	}
5349 
5350 	/*
5351 	 * For real BARs, we need to override the size that
5352 	 * the driver requests, because that's what the BAR
5353 	 * actually uses and we would otherwise have a
5354 	 * situation where we might allocate the excess to
5355 	 * another driver, which won't work.
5356 	 */
5357 	count = ((pci_addr_t)1 << mapsize) * num;
5358 	if (RF_ALIGNMENT(flags) < mapsize)
5359 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5360 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5361 		flags |= RF_PREFETCHABLE;
5362 
5363 	/*
5364 	 * Allocate enough resource, and then write back the
5365 	 * appropriate BAR for that resource.
5366 	 */
5367 	resource_list_add(rl, type, *rid, start, end, count);
5368 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5369 	    count, flags & ~RF_ACTIVE);
5370 	if (res == NULL) {
5371 		resource_list_delete(rl, type, *rid);
5372 		device_printf(child,
5373 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5374 		    count, *rid, type, start, end);
5375 		goto out;
5376 	}
5377 	if (bootverbose)
5378 		device_printf(child,
5379 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5380 		    count, *rid, type, rman_get_start(res));
5381 
5382 	/* Disable decoding via the CMD register before updating the BAR */
5383 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5384 	pci_write_config(child, PCIR_COMMAND,
5385 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5386 
5387 	map = rman_get_start(res);
5388 	pci_write_bar(child, pm, map);
5389 
5390 	/* Restore the original value of the CMD register */
5391 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5392 out:
5393 	return (res);
5394 }
5395 
5396 struct resource *
5397 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5398     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5399     u_int flags)
5400 {
5401 	struct pci_devinfo *dinfo;
5402 	struct resource_list *rl;
5403 	struct resource_list_entry *rle;
5404 	struct resource *res;
5405 	pcicfgregs *cfg;
5406 
5407 	/*
5408 	 * Perform lazy resource allocation
5409 	 */
5410 	dinfo = device_get_ivars(child);
5411 	rl = &dinfo->resources;
5412 	cfg = &dinfo->cfg;
5413 	switch (type) {
5414 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5415 	case PCI_RES_BUS:
5416 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5417 		    flags));
5418 #endif
5419 	case SYS_RES_IRQ:
5420 		/*
5421 		 * Can't alloc legacy interrupt once MSI messages have
5422 		 * been allocated.
5423 		 */
5424 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5425 		    cfg->msix.msix_alloc > 0))
5426 			return (NULL);
5427 
5428 		/*
5429 		 * If the child device doesn't have an interrupt
5430 		 * routed and is deserving of an interrupt, try to
5431 		 * assign it one.
5432 		 */
5433 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5434 		    (cfg->intpin != 0))
5435 			pci_assign_interrupt(dev, child, 0);
5436 		break;
5437 	case SYS_RES_IOPORT:
5438 	case SYS_RES_MEMORY:
5439 #ifdef NEW_PCIB
5440 		/*
5441 		 * PCI-PCI bridge I/O window resources are not BARs.
5442 		 * For those allocations just pass the request up the
5443 		 * tree.
5444 		 */
5445 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5446 			switch (*rid) {
5447 			case PCIR_IOBASEL_1:
5448 			case PCIR_MEMBASE_1:
5449 			case PCIR_PMBASEL_1:
5450 				/*
5451 				 * XXX: Should we bother creating a resource
5452 				 * list entry?
5453 				 */
5454 				return (bus_generic_alloc_resource(dev, child,
5455 				    type, rid, start, end, count, flags));
5456 			}
5457 		}
5458 #endif
5459 		/* Reserve resources for this BAR if needed. */
5460 		rle = resource_list_find(rl, type, *rid);
5461 		if (rle == NULL) {
5462 			res = pci_reserve_map(dev, child, type, rid, start, end,
5463 			    count, num, flags);
5464 			if (res == NULL)
5465 				return (NULL);
5466 		}
5467 	}
5468 	return (resource_list_alloc(rl, dev, child, type, rid,
5469 	    start, end, count, flags));
5470 }
5471 
5472 struct resource *
5473 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5474     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5475 {
5476 #ifdef PCI_IOV
5477 	struct pci_devinfo *dinfo;
5478 #endif
5479 
5480 	if (device_get_parent(child) != dev)
5481 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5482 		    type, rid, start, end, count, flags));
5483 
5484 #ifdef PCI_IOV
5485 	dinfo = device_get_ivars(child);
5486 	if (dinfo->cfg.flags & PCICFG_VF) {
5487 		switch (type) {
5488 		/* VFs can't have I/O BARs. */
5489 		case SYS_RES_IOPORT:
5490 			return (NULL);
5491 		case SYS_RES_MEMORY:
5492 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5493 			    start, end, count, flags));
5494 		}
5495 
5496 		/* Fall through for other types of resource allocations. */
5497 	}
5498 #endif
5499 
5500 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5501 	    count, 1, flags));
5502 }
5503 
5504 int
5505 pci_release_resource(device_t dev, device_t child, int type, int rid,
5506     struct resource *r)
5507 {
5508 	struct pci_devinfo *dinfo;
5509 	struct resource_list *rl;
5510 	pcicfgregs *cfg;
5511 
5512 	if (device_get_parent(child) != dev)
5513 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5514 		    type, rid, r));
5515 
5516 	dinfo = device_get_ivars(child);
5517 	cfg = &dinfo->cfg;
5518 
5519 #ifdef PCI_IOV
5520 	if (dinfo->cfg.flags & PCICFG_VF) {
5521 		switch (type) {
5522 		/* VFs can't have I/O BARs. */
5523 		case SYS_RES_IOPORT:
5524 			return (EDOOFUS);
5525 		case SYS_RES_MEMORY:
5526 			return (pci_vf_release_mem_resource(dev, child, rid,
5527 			    r));
5528 		}
5529 
5530 		/* Fall through for other types of resource allocations. */
5531 	}
5532 #endif
5533 
5534 #ifdef NEW_PCIB
5535 	/*
5536 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5537 	 * those allocations just pass the request up the tree.
5538 	 */
5539 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5540 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5541 		switch (rid) {
5542 		case PCIR_IOBASEL_1:
5543 		case PCIR_MEMBASE_1:
5544 		case PCIR_PMBASEL_1:
5545 			return (bus_generic_release_resource(dev, child, type,
5546 			    rid, r));
5547 		}
5548 	}
5549 #endif
5550 
5551 	rl = &dinfo->resources;
5552 	return (resource_list_release(rl, dev, child, type, rid, r));
5553 }
5554 
5555 int
5556 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5557     struct resource *r)
5558 {
5559 	struct pci_devinfo *dinfo;
5560 	int error;
5561 
5562 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5563 	if (error)
5564 		return (error);
5565 
5566 	/* Enable decoding in the command register when activating BARs. */
5567 	if (device_get_parent(child) == dev) {
5568 		/* Device ROMs need their decoding explicitly enabled. */
5569 		dinfo = device_get_ivars(child);
5570 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5571 			pci_write_bar(child, pci_find_bar(child, rid),
5572 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5573 		switch (type) {
5574 		case SYS_RES_IOPORT:
5575 		case SYS_RES_MEMORY:
5576 			error = PCI_ENABLE_IO(dev, child, type);
5577 			break;
5578 		}
5579 	}
5580 	return (error);
5581 }
5582 
5583 int
5584 pci_deactivate_resource(device_t dev, device_t child, int type,
5585     int rid, struct resource *r)
5586 {
5587 	struct pci_devinfo *dinfo;
5588 	int error;
5589 
5590 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5591 	if (error)
5592 		return (error);
5593 
5594 	/* Disable decoding for device ROMs. */
5595 	if (device_get_parent(child) == dev) {
5596 		dinfo = device_get_ivars(child);
5597 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5598 			pci_write_bar(child, pci_find_bar(child, rid),
5599 			    rman_get_start(r));
5600 	}
5601 	return (0);
5602 }
5603 
5604 void
5605 pci_child_deleted(device_t dev, device_t child)
5606 {
5607 	struct resource_list_entry *rle;
5608 	struct resource_list *rl;
5609 	struct pci_devinfo *dinfo;
5610 
5611 	dinfo = device_get_ivars(child);
5612 	rl = &dinfo->resources;
5613 
5614 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5615 
5616 	/* Turn off access to resources we're about to free */
5617 	if (bus_child_present(child) != 0) {
5618 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5619 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5620 
5621 		pci_disable_busmaster(child);
5622 	}
5623 
5624 	/* Free all allocated resources */
5625 	STAILQ_FOREACH(rle, rl, link) {
5626 		if (rle->res) {
5627 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5628 			    resource_list_busy(rl, rle->type, rle->rid)) {
5629 				pci_printf(&dinfo->cfg,
5630 				    "Resource still owned, oops. "
5631 				    "(type=%d, rid=%d, addr=%lx)\n",
5632 				    rle->type, rle->rid,
5633 				    rman_get_start(rle->res));
5634 				bus_release_resource(child, rle->type, rle->rid,
5635 				    rle->res);
5636 			}
5637 			resource_list_unreserve(rl, dev, child, rle->type,
5638 			    rle->rid);
5639 		}
5640 	}
5641 	resource_list_free(rl);
5642 
5643 	pci_freecfg(dinfo);
5644 }
5645 
5646 void
5647 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5648 {
5649 	struct pci_devinfo *dinfo;
5650 	struct resource_list *rl;
5651 	struct resource_list_entry *rle;
5652 
5653 	if (device_get_parent(child) != dev)
5654 		return;
5655 
5656 	dinfo = device_get_ivars(child);
5657 	rl = &dinfo->resources;
5658 	rle = resource_list_find(rl, type, rid);
5659 	if (rle == NULL)
5660 		return;
5661 
5662 	if (rle->res) {
5663 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5664 		    resource_list_busy(rl, type, rid)) {
5665 			device_printf(dev, "delete_resource: "
5666 			    "Resource still owned by child, oops. "
5667 			    "(type=%d, rid=%d, addr=%jx)\n",
5668 			    type, rid, rman_get_start(rle->res));
5669 			return;
5670 		}
5671 		resource_list_unreserve(rl, dev, child, type, rid);
5672 	}
5673 	resource_list_delete(rl, type, rid);
5674 }
5675 
5676 struct resource_list *
5677 pci_get_resource_list (device_t dev, device_t child)
5678 {
5679 	struct pci_devinfo *dinfo = device_get_ivars(child);
5680 
5681 	return (&dinfo->resources);
5682 }
5683 
5684 #ifdef ACPI_DMAR
5685 bus_dma_tag_t acpi_iommu_get_dma_tag(device_t dev, device_t child);
5686 bus_dma_tag_t
5687 pci_get_dma_tag(device_t bus, device_t dev)
5688 {
5689 	bus_dma_tag_t tag;
5690 	struct pci_softc *sc;
5691 
5692 	if (device_get_parent(dev) == bus) {
5693 		/* try iommu and return if it works */
5694 		tag = acpi_iommu_get_dma_tag(bus, dev);
5695 	} else
5696 		tag = NULL;
5697 	if (tag == NULL) {
5698 		sc = device_get_softc(bus);
5699 		tag = sc->sc_dma_tag;
5700 	}
5701 	return (tag);
5702 }
5703 #else
5704 bus_dma_tag_t
5705 pci_get_dma_tag(device_t bus, device_t dev)
5706 {
5707 	struct pci_softc *sc = device_get_softc(bus);
5708 
5709 	return (sc->sc_dma_tag);
5710 }
5711 #endif
5712 
5713 uint32_t
5714 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5715 {
5716 	struct pci_devinfo *dinfo = device_get_ivars(child);
5717 	pcicfgregs *cfg = &dinfo->cfg;
5718 
5719 #ifdef PCI_IOV
5720 	/*
5721 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5722 	 * emulate them here.
5723 	 */
5724 	if (cfg->flags & PCICFG_VF) {
5725 		if (reg == PCIR_VENDOR) {
5726 			switch (width) {
5727 			case 4:
5728 				return (cfg->device << 16 | cfg->vendor);
5729 			case 2:
5730 				return (cfg->vendor);
5731 			case 1:
5732 				return (cfg->vendor & 0xff);
5733 			default:
5734 				return (0xffffffff);
5735 			}
5736 		} else if (reg == PCIR_DEVICE) {
5737 			switch (width) {
5738 			/* Note that an unaligned 4-byte read is an error. */
5739 			case 2:
5740 				return (cfg->device);
5741 			case 1:
5742 				return (cfg->device & 0xff);
5743 			default:
5744 				return (0xffffffff);
5745 			}
5746 		}
5747 	}
5748 #endif
5749 
5750 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5751 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5752 }
5753 
5754 void
5755 pci_write_config_method(device_t dev, device_t child, int reg,
5756     uint32_t val, int width)
5757 {
5758 	struct pci_devinfo *dinfo = device_get_ivars(child);
5759 	pcicfgregs *cfg = &dinfo->cfg;
5760 
5761 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5762 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5763 }
5764 
5765 int
5766 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5767     size_t buflen)
5768 {
5769 
5770 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5771 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5772 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5773 	return (0);
5774 }
5775 
5776 int
5777 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5778     size_t buflen)
5779 {
5780 	struct pci_devinfo *dinfo;
5781 	pcicfgregs *cfg;
5782 
5783 	dinfo = device_get_ivars(child);
5784 	cfg = &dinfo->cfg;
5785 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5786 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5787 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5788 	    cfg->progif);
5789 	return (0);
5790 }
5791 
5792 int
5793 pci_assign_interrupt_method(device_t dev, device_t child)
5794 {
5795 	struct pci_devinfo *dinfo = device_get_ivars(child);
5796 	pcicfgregs *cfg = &dinfo->cfg;
5797 
5798 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5799 	    cfg->intpin));
5800 }
5801 
5802 static void
5803 pci_lookup(void *arg, const char *name, device_t *dev)
5804 {
5805 	long val;
5806 	char *end;
5807 	int domain, bus, slot, func;
5808 
5809 	if (*dev != NULL)
5810 		return;
5811 
5812 	/*
5813 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5814 	 * pciB:S:F.  In the latter case, the domain is assumed to
5815 	 * be zero.
5816 	 */
5817 	if (strncmp(name, "pci", 3) != 0)
5818 		return;
5819 	val = strtol(name + 3, &end, 10);
5820 	if (val < 0 || val > INT_MAX || *end != ':')
5821 		return;
5822 	domain = val;
5823 	val = strtol(end + 1, &end, 10);
5824 	if (val < 0 || val > INT_MAX || *end != ':')
5825 		return;
5826 	bus = val;
5827 	val = strtol(end + 1, &end, 10);
5828 	if (val < 0 || val > INT_MAX)
5829 		return;
5830 	slot = val;
5831 	if (*end == ':') {
5832 		val = strtol(end + 1, &end, 10);
5833 		if (val < 0 || val > INT_MAX || *end != '\0')
5834 			return;
5835 		func = val;
5836 	} else if (*end == '\0') {
5837 		func = slot;
5838 		slot = bus;
5839 		bus = domain;
5840 		domain = 0;
5841 	} else
5842 		return;
5843 
5844 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5845 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5846 		return;
5847 
5848 	*dev = pci_find_dbsf(domain, bus, slot, func);
5849 }
5850 
5851 static int
5852 pci_modevent(module_t mod, int what, void *arg)
5853 {
5854 	static struct cdev *pci_cdev;
5855 	static eventhandler_tag tag;
5856 
5857 	switch (what) {
5858 	case MOD_LOAD:
5859 		STAILQ_INIT(&pci_devq);
5860 		pci_generation = 0;
5861 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5862 		    "pci");
5863 		pci_load_vendor_data();
5864 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5865 		    1000);
5866 		break;
5867 
5868 	case MOD_UNLOAD:
5869 		if (tag != NULL)
5870 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5871 		destroy_dev(pci_cdev);
5872 		break;
5873 	}
5874 
5875 	return (0);
5876 }
5877 
5878 static void
5879 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5880 {
5881 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5882 	struct pcicfg_pcie *cfg;
5883 	int version, pos;
5884 
5885 	cfg = &dinfo->cfg.pcie;
5886 	pos = cfg->pcie_location;
5887 
5888 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5889 
5890 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5891 
5892 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5893 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5894 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5895 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5896 
5897 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5898 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5899 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5900 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5901 
5902 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5903 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5904 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5905 
5906 	if (version > 1) {
5907 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5908 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5909 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5910 	}
5911 #undef WREG
5912 }
5913 
5914 static void
5915 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5916 {
5917 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5918 	    dinfo->cfg.pcix.pcix_command,  2);
5919 }
5920 
5921 void
5922 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5923 {
5924 
5925 	/*
5926 	 * Restore the device to full power mode.  We must do this
5927 	 * before we restore the registers because moving from D3 to
5928 	 * D0 will cause the chip's BARs and some other registers to
5929 	 * be reset to some unknown power on reset values.  Cut down
5930 	 * the noise on boot by doing nothing if we are already in
5931 	 * state D0.
5932 	 */
5933 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5934 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5935 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5936 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5937 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5938 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5939 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5940 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5941 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5942 	case PCIM_HDRTYPE_NORMAL:
5943 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5944 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5945 		break;
5946 	case PCIM_HDRTYPE_BRIDGE:
5947 		pci_write_config(dev, PCIR_SECLAT_1,
5948 		    dinfo->cfg.bridge.br_seclat, 1);
5949 		pci_write_config(dev, PCIR_SUBBUS_1,
5950 		    dinfo->cfg.bridge.br_subbus, 1);
5951 		pci_write_config(dev, PCIR_SECBUS_1,
5952 		    dinfo->cfg.bridge.br_secbus, 1);
5953 		pci_write_config(dev, PCIR_PRIBUS_1,
5954 		    dinfo->cfg.bridge.br_pribus, 1);
5955 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5956 		    dinfo->cfg.bridge.br_control, 2);
5957 		break;
5958 	case PCIM_HDRTYPE_CARDBUS:
5959 		pci_write_config(dev, PCIR_SECLAT_2,
5960 		    dinfo->cfg.bridge.br_seclat, 1);
5961 		pci_write_config(dev, PCIR_SUBBUS_2,
5962 		    dinfo->cfg.bridge.br_subbus, 1);
5963 		pci_write_config(dev, PCIR_SECBUS_2,
5964 		    dinfo->cfg.bridge.br_secbus, 1);
5965 		pci_write_config(dev, PCIR_PRIBUS_2,
5966 		    dinfo->cfg.bridge.br_pribus, 1);
5967 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5968 		    dinfo->cfg.bridge.br_control, 2);
5969 		break;
5970 	}
5971 	pci_restore_bars(dev);
5972 
5973 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
5974 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5975 
5976 	/*
5977 	 * Restore extended capabilities for PCI-Express and PCI-X
5978 	 */
5979 	if (dinfo->cfg.pcie.pcie_location != 0)
5980 		pci_cfg_restore_pcie(dev, dinfo);
5981 	if (dinfo->cfg.pcix.pcix_location != 0)
5982 		pci_cfg_restore_pcix(dev, dinfo);
5983 
5984 	/* Restore MSI and MSI-X configurations if they are present. */
5985 	if (dinfo->cfg.msi.msi_location != 0)
5986 		pci_resume_msi(dev);
5987 	if (dinfo->cfg.msix.msix_location != 0)
5988 		pci_resume_msix(dev);
5989 
5990 #ifdef PCI_IOV
5991 	if (dinfo->cfg.iov != NULL)
5992 		pci_iov_cfg_restore(dev, dinfo);
5993 #endif
5994 }
5995 
5996 static void
5997 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
5998 {
5999 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6000 	struct pcicfg_pcie *cfg;
6001 	int version, pos;
6002 
6003 	cfg = &dinfo->cfg.pcie;
6004 	pos = cfg->pcie_location;
6005 
6006 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6007 
6008 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6009 
6010 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6011 
6012 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6013 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6014 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6015 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6016 
6017 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6018 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6019 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6020 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6021 
6022 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6023 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6024 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6025 
6026 	if (version > 1) {
6027 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6028 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6029 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6030 	}
6031 #undef RREG
6032 }
6033 
6034 static void
6035 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6036 {
6037 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6038 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6039 }
6040 
6041 void
6042 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6043 {
6044 	uint32_t cls;
6045 	int ps;
6046 
6047 	/*
6048 	 * Some drivers apparently write to these registers w/o updating our
6049 	 * cached copy.  No harm happens if we update the copy, so do so here
6050 	 * so we can restore them.  The COMMAND register is modified by the
6051 	 * bus w/o updating the cache.  This should represent the normally
6052 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6053 	 */
6054 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6055 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6056 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6057 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6058 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6059 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6060 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6061 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6062 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6063 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6064 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6065 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6066 	case PCIM_HDRTYPE_NORMAL:
6067 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6068 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6069 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6070 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6071 		break;
6072 	case PCIM_HDRTYPE_BRIDGE:
6073 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6074 		    PCIR_SECLAT_1, 1);
6075 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6076 		    PCIR_SUBBUS_1, 1);
6077 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6078 		    PCIR_SECBUS_1, 1);
6079 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6080 		    PCIR_PRIBUS_1, 1);
6081 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6082 		    PCIR_BRIDGECTL_1, 2);
6083 		break;
6084 	case PCIM_HDRTYPE_CARDBUS:
6085 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6086 		    PCIR_SECLAT_2, 1);
6087 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6088 		    PCIR_SUBBUS_2, 1);
6089 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6090 		    PCIR_SECBUS_2, 1);
6091 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6092 		    PCIR_PRIBUS_2, 1);
6093 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6094 		    PCIR_BRIDGECTL_2, 2);
6095 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6096 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6097 		break;
6098 	}
6099 
6100 	if (dinfo->cfg.pcie.pcie_location != 0)
6101 		pci_cfg_save_pcie(dev, dinfo);
6102 
6103 	if (dinfo->cfg.pcix.pcix_location != 0)
6104 		pci_cfg_save_pcix(dev, dinfo);
6105 
6106 #ifdef PCI_IOV
6107 	if (dinfo->cfg.iov != NULL)
6108 		pci_iov_cfg_save(dev, dinfo);
6109 #endif
6110 
6111 	/*
6112 	 * don't set the state for display devices, base peripherals and
6113 	 * memory devices since bad things happen when they are powered down.
6114 	 * We should (a) have drivers that can easily detach and (b) use
6115 	 * generic drivers for these devices so that some device actually
6116 	 * attaches.  We need to make sure that when we implement (a) we don't
6117 	 * power the device down on a reattach.
6118 	 */
6119 	cls = pci_get_class(dev);
6120 	if (!setstate)
6121 		return;
6122 	switch (pci_do_power_nodriver)
6123 	{
6124 		case 0:		/* NO powerdown at all */
6125 			return;
6126 		case 1:		/* Conservative about what to power down */
6127 			if (cls == PCIC_STORAGE)
6128 				return;
6129 			/*FALLTHROUGH*/
6130 		case 2:		/* Aggressive about what to power down */
6131 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6132 			    cls == PCIC_BASEPERIPH)
6133 				return;
6134 			/*FALLTHROUGH*/
6135 		case 3:		/* Power down everything */
6136 			break;
6137 	}
6138 	/*
6139 	 * PCI spec says we can only go into D3 state from D0 state.
6140 	 * Transition from D[12] into D0 before going to D3 state.
6141 	 */
6142 	ps = pci_get_powerstate(dev);
6143 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6144 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6145 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6146 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6147 }
6148 
6149 /* Wrapper APIs suitable for device driver use. */
6150 void
6151 pci_save_state(device_t dev)
6152 {
6153 	struct pci_devinfo *dinfo;
6154 
6155 	dinfo = device_get_ivars(dev);
6156 	pci_cfg_save(dev, dinfo, 0);
6157 }
6158 
6159 void
6160 pci_restore_state(device_t dev)
6161 {
6162 	struct pci_devinfo *dinfo;
6163 
6164 	dinfo = device_get_ivars(dev);
6165 	pci_cfg_restore(dev, dinfo);
6166 }
6167 
6168 static int
6169 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6170     uintptr_t *id)
6171 {
6172 
6173 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6174 }
6175 
6176 /* Find the upstream port of a given PCI device in a root complex. */
6177 device_t
6178 pci_find_pcie_root_port(device_t dev)
6179 {
6180 	struct pci_devinfo *dinfo;
6181 	devclass_t pci_class;
6182 	device_t pcib, bus;
6183 
6184 	pci_class = devclass_find("pci");
6185 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6186 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6187 
6188 	/*
6189 	 * Walk the bridge hierarchy until we find a PCI-e root
6190 	 * port or a non-PCI device.
6191 	 */
6192 	for (;;) {
6193 		bus = device_get_parent(dev);
6194 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6195 		    device_get_nameunit(dev)));
6196 
6197 		pcib = device_get_parent(bus);
6198 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6199 		    device_get_nameunit(bus)));
6200 
6201 		/*
6202 		 * pcib's parent must be a PCI bus for this to be a
6203 		 * PCI-PCI bridge.
6204 		 */
6205 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6206 			return (NULL);
6207 
6208 		dinfo = device_get_ivars(pcib);
6209 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6210 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6211 			return (pcib);
6212 
6213 		dev = pcib;
6214 	}
6215 }
6216 
6217 /*
6218  * Wait for pending transactions to complete on a PCI-express function.
6219  *
6220  * The maximum delay is specified in milliseconds in max_delay.  Note
6221  * that this function may sleep.
6222  *
6223  * Returns true if the function is idle and false if the timeout is
6224  * exceeded.  If dev is not a PCI-express function, this returns true.
6225  */
6226 bool
6227 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6228 {
6229 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6230 	uint16_t sta;
6231 	int cap;
6232 
6233 	cap = dinfo->cfg.pcie.pcie_location;
6234 	if (cap == 0)
6235 		return (true);
6236 
6237 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6238 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6239 		if (max_delay == 0)
6240 			return (false);
6241 
6242 		/* Poll once every 100 milliseconds up to the timeout. */
6243 		if (max_delay > 100) {
6244 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6245 			max_delay -= 100;
6246 		} else {
6247 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6248 			    C_HARDCLOCK);
6249 			max_delay = 0;
6250 		}
6251 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6252 	}
6253 
6254 	return (true);
6255 }
6256 
6257 /*
6258  * Determine the maximum Completion Timeout in microseconds.
6259  *
6260  * For non-PCI-express functions this returns 0.
6261  */
6262 int
6263 pcie_get_max_completion_timeout(device_t dev)
6264 {
6265 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6266 	int cap;
6267 
6268 	cap = dinfo->cfg.pcie.pcie_location;
6269 	if (cap == 0)
6270 		return (0);
6271 
6272 	/*
6273 	 * Functions using the 1.x spec use the default timeout range of
6274 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6275 	 * support programmable timeouts also use this range.
6276 	 */
6277 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6278 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6279 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6280 		return (50 * 1000);
6281 
6282 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6283 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6284 	case PCIEM_CTL2_COMP_TIMO_100US:
6285 		return (100);
6286 	case PCIEM_CTL2_COMP_TIMO_10MS:
6287 		return (10 * 1000);
6288 	case PCIEM_CTL2_COMP_TIMO_55MS:
6289 		return (55 * 1000);
6290 	case PCIEM_CTL2_COMP_TIMO_210MS:
6291 		return (210 * 1000);
6292 	case PCIEM_CTL2_COMP_TIMO_900MS:
6293 		return (900 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6295 		return (3500 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_13S:
6297 		return (13 * 1000 * 1000);
6298 	case PCIEM_CTL2_COMP_TIMO_64S:
6299 		return (64 * 1000 * 1000);
6300 	default:
6301 		return (50 * 1000);
6302 	}
6303 }
6304 
6305 /*
6306  * Perform a Function Level Reset (FLR) on a device.
6307  *
6308  * This function first waits for any pending transactions to complete
6309  * within the timeout specified by max_delay.  If transactions are
6310  * still pending, the function will return false without attempting a
6311  * reset.
6312  *
6313  * If dev is not a PCI-express function or does not support FLR, this
6314  * function returns false.
6315  *
6316  * Note that no registers are saved or restored.  The caller is
6317  * responsible for saving and restoring any registers including
6318  * PCI-standard registers via pci_save_state() and
6319  * pci_restore_state().
6320  */
6321 bool
6322 pcie_flr(device_t dev, u_int max_delay, bool force)
6323 {
6324 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6325 	uint16_t cmd, ctl;
6326 	int compl_delay;
6327 	int cap;
6328 
6329 	cap = dinfo->cfg.pcie.pcie_location;
6330 	if (cap == 0)
6331 		return (false);
6332 
6333 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6334 		return (false);
6335 
6336 	/*
6337 	 * Disable busmastering to prevent generation of new
6338 	 * transactions while waiting for the device to go idle.  If
6339 	 * the idle timeout fails, the command register is restored
6340 	 * which will re-enable busmastering.
6341 	 */
6342 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6343 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6344 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6345 		if (!force) {
6346 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6347 			return (false);
6348 		}
6349 		pci_printf(&dinfo->cfg,
6350 		    "Resetting with transactions pending after %d ms\n",
6351 		    max_delay);
6352 
6353 		/*
6354 		 * Extend the post-FLR delay to cover the maximum
6355 		 * Completion Timeout delay of anything in flight
6356 		 * during the FLR delay.  Enforce a minimum delay of
6357 		 * at least 10ms.
6358 		 */
6359 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6360 		if (compl_delay < 10)
6361 			compl_delay = 10;
6362 	} else
6363 		compl_delay = 0;
6364 
6365 	/* Initiate the reset. */
6366 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6367 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6368 	    PCIEM_CTL_INITIATE_FLR, 2);
6369 
6370 	/* Wait for 100ms. */
6371 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6372 
6373 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6374 	    PCIEM_STA_TRANSACTION_PND)
6375 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6376 	return (true);
6377 }
6378 
6379 /*
6380  * Attempt a power-management reset by cycling the device in/out of D3
6381  * state.  PCI spec says we can only go into D3 state from D0 state.
6382  * Transition from D[12] into D0 before going to D3 state.
6383  */
6384 int
6385 pci_power_reset(device_t dev)
6386 {
6387 	int ps;
6388 
6389 	ps = pci_get_powerstate(dev);
6390 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6391 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6392 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6393 	pci_set_powerstate(dev, ps);
6394 	return (0);
6395 }
6396 
6397 /*
6398  * Try link drop and retrain of the downstream port of upstream
6399  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6400  * cause Conventional Hot reset of the device in the slot.
6401  * Alternative, for PCIe, could be the secondary bus reset initiatied
6402  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6403  */
6404 int
6405 pcie_link_reset(device_t port, int pcie_location)
6406 {
6407 	uint16_t v;
6408 
6409 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6410 	v |= PCIEM_LINK_CTL_LINK_DIS;
6411 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6412 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6413 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6414 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6415 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6416 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6417 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6418 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6419 }
6420 
6421 static int
6422 pci_reset_post(device_t dev, device_t child)
6423 {
6424 
6425 	if (dev == device_get_parent(child))
6426 		pci_restore_state(child);
6427 	return (0);
6428 }
6429 
6430 static int
6431 pci_reset_prepare(device_t dev, device_t child)
6432 {
6433 
6434 	if (dev == device_get_parent(child))
6435 		pci_save_state(child);
6436 	return (0);
6437 }
6438 
6439 static int
6440 pci_reset_child(device_t dev, device_t child, int flags)
6441 {
6442 	int error;
6443 
6444 	if (dev == NULL || device_get_parent(child) != dev)
6445 		return (0);
6446 	if ((flags & DEVF_RESET_DETACH) != 0) {
6447 		error = device_get_state(child) == DS_ATTACHED ?
6448 		    device_detach(child) : 0;
6449 	} else {
6450 		error = BUS_SUSPEND_CHILD(dev, child);
6451 	}
6452 	if (error == 0) {
6453 		if (!pcie_flr(child, 1000, false)) {
6454 			error = BUS_RESET_PREPARE(dev, child);
6455 			if (error == 0)
6456 				pci_power_reset(child);
6457 			BUS_RESET_POST(dev, child);
6458 		}
6459 		if ((flags & DEVF_RESET_DETACH) != 0)
6460 			device_probe_and_attach(child);
6461 		else
6462 			BUS_RESUME_CHILD(dev, child);
6463 	}
6464 	return (error);
6465 }
6466 
6467 const struct pci_device_table *
6468 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6469 {
6470 	bool match;
6471 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6472 
6473 	vendor = pci_get_vendor(child);
6474 	device = pci_get_device(child);
6475 	subvendor = pci_get_subvendor(child);
6476 	subdevice = pci_get_subdevice(child);
6477 	class = pci_get_class(child);
6478 	subclass = pci_get_subclass(child);
6479 	revid = pci_get_revid(child);
6480 	while (nelt-- > 0) {
6481 		match = true;
6482 		if (id->match_flag_vendor)
6483 			match &= vendor == id->vendor;
6484 		if (id->match_flag_device)
6485 			match &= device == id->device;
6486 		if (id->match_flag_subvendor)
6487 			match &= subvendor == id->subvendor;
6488 		if (id->match_flag_subdevice)
6489 			match &= subdevice == id->subdevice;
6490 		if (id->match_flag_class)
6491 			match &= class == id->class_id;
6492 		if (id->match_flag_subclass)
6493 			match &= subclass == id->subclass;
6494 		if (id->match_flag_revid)
6495 			match &= revid == id->revid;
6496 		if (match)
6497 			return (id);
6498 		id++;
6499 	}
6500 	return (NULL);
6501 }
6502 
6503 static void
6504 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6505 {
6506 	const char *dev_name;
6507 	device_t dev;
6508 
6509 	dev = dinfo->cfg.dev;
6510 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6511 	    dinfo->cfg.slot, dinfo->cfg.func);
6512 	dev_name = device_get_name(dev);
6513 	if (dev_name != NULL)
6514 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6515 }
6516 
6517 void
6518 pci_print_faulted_dev(void)
6519 {
6520 	struct pci_devinfo *dinfo;
6521 	device_t dev;
6522 	int aer, i;
6523 	uint32_t r1, r2;
6524 	uint16_t status;
6525 
6526 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6527 		dev = dinfo->cfg.dev;
6528 		status = pci_read_config(dev, PCIR_STATUS, 2);
6529 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6530 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6531 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6532 		if (status != 0) {
6533 			pci_print_faulted_dev_name(dinfo);
6534 			printf(" error 0x%04x\n", status);
6535 		}
6536 		if (dinfo->cfg.pcie.pcie_location != 0) {
6537 			status = pci_read_config(dev,
6538 			    dinfo->cfg.pcie.pcie_location +
6539 			    PCIER_DEVICE_STA, 2);
6540 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6541 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6542 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6543 				pci_print_faulted_dev_name(dinfo);
6544 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6545 				    pci_read_config(dev,
6546 				    dinfo->cfg.pcie.pcie_location +
6547 				    PCIER_DEVICE_CTL, 2),
6548 				    status);
6549 			}
6550 		}
6551 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6552 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6553 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6554 			if (r1 != 0 || r2 != 0) {
6555 				pci_print_faulted_dev_name(dinfo);
6556 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6557 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6558 				    r1, pci_read_config(dev, aer +
6559 				    PCIR_AER_UC_MASK, 4),
6560 				    pci_read_config(dev, aer +
6561 				    PCIR_AER_UC_SEVERITY, 4),
6562 				    r2, pci_read_config(dev, aer +
6563 				    PCIR_AER_COR_MASK, 4),
6564 				    pci_read_config(dev, aer +
6565 				    PCIR_AER_CAP_CONTROL, 4));
6566 				for (i = 0; i < 4; i++) {
6567 					r1 = pci_read_config(dev, aer +
6568 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6569 					printf("    HL%d: 0x%08x\n", i, r1);
6570 				}
6571 			}
6572 		}
6573 	}
6574 }
6575 
6576 #ifdef DDB
6577 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6578 {
6579 
6580 	pci_print_faulted_dev();
6581 }
6582 
6583 static void
6584 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6585 {
6586 	device_t dev;
6587 	int aer;
6588 	uint32_t r;
6589 
6590 	dev = dinfo->cfg.dev;
6591 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6592 	    PCIER_DEVICE_STA, 2);
6593 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6594 	    PCIER_DEVICE_STA, r, 2);
6595 
6596 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6597 		return;
6598 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6599 	if (r != 0)
6600 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6601 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6602 	if (r != 0)
6603 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6604 }
6605 
6606 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6607 {
6608 	struct pci_devinfo *dinfo;
6609 	device_t dev;
6610 	uint16_t status, status1;
6611 
6612 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6613 		dev = dinfo->cfg.dev;
6614 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6615 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6616 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6617 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6618 		if (status1 != 0) {
6619 			status &= ~status1;
6620 			pci_write_config(dev, PCIR_STATUS, status, 2);
6621 		}
6622 		if (dinfo->cfg.pcie.pcie_location != 0)
6623 			db_clear_pcie_errors(dinfo);
6624 	}
6625 }
6626 #endif
6627