xref: /freebsd/sys/dev/pci/pci.c (revision 9b37d84c87e69dabc69d818aa4d2fea718bd8b74)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_acpi.h"
33 #include "opt_iommu.h"
34 #include "opt_bus.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/linker.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 #include <sys/taskqueue.h>
51 #include <sys/tree.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 
57 #include <sys/bus.h>
58 #include <machine/bus.h>
59 #include <sys/rman.h>
60 #include <machine/resource.h>
61 #include <machine/stdarg.h>
62 
63 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
64 #include <machine/intr_machdep.h>
65 #endif
66 
67 #include <sys/pciio.h>
68 #include <dev/pci/pcireg.h>
69 #include <dev/pci/pcivar.h>
70 #include <dev/pci/pci_private.h>
71 
72 #ifdef PCI_IOV
73 #include <sys/nv.h>
74 #include <dev/pci/pci_iov_private.h>
75 #endif
76 
77 #include <dev/usb/controller/xhcireg.h>
78 #include <dev/usb/controller/ehcireg.h>
79 #include <dev/usb/controller/ohcireg.h>
80 #include <dev/usb/controller/uhcireg.h>
81 
82 #include <dev/iommu/iommu.h>
83 
84 #include "pcib_if.h"
85 #include "pci_if.h"
86 
87 #define	PCIR_IS_BIOS(cfg, reg)						\
88 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
89 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
90 
91 static device_probe_t	pci_probe;
92 
93 static bus_reset_post_t pci_reset_post;
94 static bus_reset_prepare_t pci_reset_prepare;
95 static bus_reset_child_t pci_reset_child;
96 static bus_hint_device_unit_t pci_hint_device_unit;
97 static bus_remap_intr_t pci_remap_intr_method;
98 
99 static pci_get_id_t	pci_get_id_method;
100 
101 static int		pci_has_quirk(uint32_t devid, int quirk);
102 static pci_addr_t	pci_mapbase(uint64_t mapreg);
103 static const char	*pci_maptype(uint64_t mapreg);
104 static int		pci_maprange(uint64_t mapreg);
105 static pci_addr_t	pci_rombase(uint64_t mapreg);
106 static int		pci_romsize(uint64_t testval);
107 static void		pci_fixancient(pcicfgregs *cfg);
108 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
109 
110 static int		pci_porten(device_t dev);
111 static int		pci_memen(device_t dev);
112 static void		pci_assign_interrupt(device_t bus, device_t dev,
113 			    int force_route);
114 static int		pci_add_map(device_t bus, device_t dev, int reg,
115 			    struct resource_list *rl, int force, int prefetch);
116 static void		pci_load_vendor_data(void);
117 static int		pci_describe_parse_line(char **ptr, int *vendor,
118 			    int *device, char **desc);
119 static char		*pci_describe_device(device_t dev);
120 static int		pci_modevent(module_t mod, int what, void *arg);
121 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
122 			    pcicfgregs *cfg);
123 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
124 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
125 			    int reg, uint32_t *data);
126 #if 0
127 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
128 			    int reg, uint32_t data);
129 #endif
130 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
131 static void		pci_mask_msix(device_t dev, u_int index);
132 static void		pci_unmask_msix(device_t dev, u_int index);
133 static int		pci_msi_blacklisted(void);
134 static int		pci_msix_blacklisted(void);
135 static void		pci_resume_msi(device_t dev);
136 static void		pci_resume_msix(device_t dev);
137 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
138     int b, int s, int f, uint16_t vid, uint16_t did);
139 
140 static device_method_t pci_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,		pci_probe),
143 	DEVMETHOD(device_attach,	pci_attach),
144 	DEVMETHOD(device_detach,	pci_detach),
145 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,	bus_generic_suspend),
147 	DEVMETHOD(device_resume,	pci_resume),
148 
149 	/* Bus interface */
150 	DEVMETHOD(bus_print_child,	pci_print_child),
151 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
152 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
153 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
154 	DEVMETHOD(bus_driver_added,	pci_driver_added),
155 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
156 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
157 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
158 	DEVMETHOD(bus_reset_post,	pci_reset_post),
159 	DEVMETHOD(bus_reset_child,	pci_reset_child),
160 
161 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
162 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
163 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
164 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
165 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
166 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
167 	DEVMETHOD(bus_adjust_resource,	pci_adjust_resource),
168 	DEVMETHOD(bus_release_resource,	pci_release_resource),
169 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
170 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
171 	DEVMETHOD(bus_map_resource,	pci_map_resource),
172 	DEVMETHOD(bus_unmap_resource,	pci_unmap_resource),
173 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
174 	DEVMETHOD(bus_child_detached,	pci_child_detached),
175 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
176 	DEVMETHOD(bus_child_location,	pci_child_location_method),
177 	DEVMETHOD(bus_get_device_path,	pci_get_device_path_method),
178 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
179 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
180 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
181 	DEVMETHOD(bus_resume_child,	pci_resume_child),
182 	DEVMETHOD(bus_rescan,		pci_rescan_method),
183 
184 	/* PCI interface */
185 	DEVMETHOD(pci_read_config,	pci_read_config_method),
186 	DEVMETHOD(pci_write_config,	pci_write_config_method),
187 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
188 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
189 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
190 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
191 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
192 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
193 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
194 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
195 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
196 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
197 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
198 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
199 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
200 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
201 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
202 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
203 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
204 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
205 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
206 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
207 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
208 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
209 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
210 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
211 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
212 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
213 	DEVMETHOD(pci_get_id,		pci_get_id_method),
214 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
215 	DEVMETHOD(pci_child_added,	pci_child_added_method),
216 #ifdef PCI_IOV
217 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
218 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
219 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
220 #endif
221 
222 	DEVMETHOD_END
223 };
224 
225 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
226 
227 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_modevent, NULL, BUS_PASS_BUS);
228 MODULE_VERSION(pci, 1);
229 
230 static char	*pci_vendordata;
231 static size_t	pci_vendordata_size;
232 
233 struct pci_quirk {
234 	uint32_t devid;	/* Vendor/device of the card */
235 	int	type;
236 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
237 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
238 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
239 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
240 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
241 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
243 	int	arg1;
244 	int	arg2;
245 };
246 
247 static const struct pci_quirk pci_quirks[] = {
248 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
249 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
250 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
252 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 
254 	/*
255 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
256 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
257 	 */
258 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
259 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
260 
261 	/*
262 	 * MSI doesn't work on earlier Intel chipsets including
263 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
264 	 */
265 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 
273 	/*
274 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
275 	 * bridge.
276 	 */
277 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
278 
279 	/*
280 	 * Some virtualization environments emulate an older chipset
281 	 * but support MSI just fine.  QEMU uses the Intel 82440.
282 	 */
283 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
284 
285 	/*
286 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
287 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
288 	 * It prevents us from attaching hpet(4) when the bit is unset.
289 	 * Note this quirk only affects SB600 revision A13 and earlier.
290 	 * For SB600 A21 and later, firmware must set the bit to hide it.
291 	 * For SB700 and later, it is unused and hardcoded to zero.
292 	 */
293 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
294 
295 	/*
296 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
297 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
298 	 * of the command register is set.
299 	 */
300 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 
306 	/*
307 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
308 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
309 	 */
310 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
311 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
312 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
313 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
314 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
315 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
316 
317 	/*
318 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
319 	 * expected place.
320 	 */
321 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
322 	{ 0 }
323 };
324 
325 /* map register information */
326 #define	PCI_MAPMEM	0x01	/* memory map */
327 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
328 #define	PCI_MAPPORT	0x04	/* port map */
329 
330 struct devlist pci_devq;
331 uint32_t pci_generation;
332 uint32_t pci_numdevs = 0;
333 static int pcie_chipset, pcix_chipset;
334 
335 /* sysctl vars */
336 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
337     "PCI bus tuning parameters");
338 
339 static int pci_enable_io_modes = 1;
340 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
341     &pci_enable_io_modes, 1,
342     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
343     " enable these bits correctly.  We'd like to do this all the time, but"
344     " there are some peripherals that this causes problems with.");
345 
346 static int pci_do_realloc_bars = 1;
347 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
348     &pci_do_realloc_bars, 0,
349     "Attempt to allocate a new range for any BARs whose original "
350     "firmware-assigned ranges fail to allocate during the initial device scan.");
351 
352 static int pci_do_power_nodriver = 0;
353 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
354     &pci_do_power_nodriver, 0,
355     "Place a function into D3 state when no driver attaches to it.  0 means"
356     " disable.  1 means conservatively place function into D3 state.  2 means"
357     " aggressively place function into D3 state.  3 means put absolutely"
358     " everything in D3 state.");
359 
360 int pci_do_power_resume = 1;
361 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
362     &pci_do_power_resume, 1,
363   "Transition from D3 -> D0 on resume.");
364 
365 int pci_do_power_suspend = 1;
366 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
367     &pci_do_power_suspend, 1,
368   "Transition from D0 -> D3 on suspend.");
369 
370 static int pci_do_msi = 1;
371 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
372     "Enable support for MSI interrupts");
373 
374 static int pci_do_msix = 1;
375 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
376     "Enable support for MSI-X interrupts");
377 
378 static int pci_msix_rewrite_table = 0;
379 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
380     &pci_msix_rewrite_table, 0,
381     "Rewrite entire MSI-X table when updating MSI-X entries");
382 
383 static int pci_honor_msi_blacklist = 1;
384 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
385     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
386 
387 #if defined(__i386__) || defined(__amd64__)
388 static int pci_usb_takeover = 1;
389 #else
390 static int pci_usb_takeover = 0;
391 #endif
392 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
393     &pci_usb_takeover, 1,
394     "Enable early takeover of USB controllers. Disable this if you depend on"
395     " BIOS emulation of USB devices, that is you use USB devices (like"
396     " keyboard or mouse) but do not load USB drivers");
397 
398 static int pci_clear_bars;
399 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
400     "Ignore firmware-assigned resources for BARs.");
401 
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 
406 static int pci_enable_ari = 1;
407 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
408     0, "Enable support for PCIe Alternative RID Interpretation");
409 
410 /*
411  * Some x86 firmware only enables PCIe hotplug if we claim to support aspm,
412  * however enabling it breaks some arm64 firmware as it powers off devices.
413  */
414 #if defined(__i386__) || defined(__amd64__)
415 int pci_enable_aspm = 1;
416 #else
417 int pci_enable_aspm = 0;
418 #endif
419 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
420     0, "Enable support for PCIe Active State Power Management");
421 
422 static int pci_clear_aer_on_attach = 0;
423 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
424     &pci_clear_aer_on_attach, 0,
425     "Clear port and device AER state on driver attach");
426 
427 static bool pci_enable_mps_tune = true;
428 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
429     &pci_enable_mps_tune, 1,
430     "Enable tuning of MPS(maximum payload size)." );
431 
432 static bool pci_intx_reroute = true;
433 SYSCTL_BOOL(_hw_pci, OID_AUTO, intx_reroute, CTLFLAG_RWTUN,
434     &pci_intx_reroute, 0, "Re-route INTx interrupts when scanning devices");
435 
436 static int
437 pci_has_quirk(uint32_t devid, int quirk)
438 {
439 	const struct pci_quirk *q;
440 
441 	for (q = &pci_quirks[0]; q->devid; q++) {
442 		if (q->devid == devid && q->type == quirk)
443 			return (1);
444 	}
445 	return (0);
446 }
447 
448 /* Find a device_t by bus/slot/function in domain 0 */
449 
450 device_t
451 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
452 {
453 
454 	return (pci_find_dbsf(0, bus, slot, func));
455 }
456 
457 /* Find a device_t by domain/bus/slot/function */
458 
459 device_t
460 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
461 {
462 	struct pci_devinfo *dinfo = NULL;
463 
464 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
465 		if ((dinfo->cfg.domain == domain) &&
466 		    (dinfo->cfg.bus == bus) &&
467 		    (dinfo->cfg.slot == slot) &&
468 		    (dinfo->cfg.func == func)) {
469 			break;
470 		}
471 	}
472 
473 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
474 }
475 
476 /* Find a device_t by vendor/device ID */
477 
478 device_t
479 pci_find_device(uint16_t vendor, uint16_t device)
480 {
481 	struct pci_devinfo *dinfo;
482 
483 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
484 		if ((dinfo->cfg.vendor == vendor) &&
485 		    (dinfo->cfg.device == device)) {
486 			return (dinfo->cfg.dev);
487 		}
488 	}
489 
490 	return (NULL);
491 }
492 
493 device_t
494 pci_find_class(uint8_t class, uint8_t subclass)
495 {
496 	struct pci_devinfo *dinfo;
497 
498 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
499 		if (dinfo->cfg.baseclass == class &&
500 		    dinfo->cfg.subclass == subclass) {
501 			return (dinfo->cfg.dev);
502 		}
503 	}
504 
505 	return (NULL);
506 }
507 
508 device_t
509 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
510 {
511 	struct pci_devinfo *dinfo;
512 	bool found = false;
513 
514 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
515 		if (from != NULL && found == false) {
516 			if (from != dinfo->cfg.dev)
517 				continue;
518 			found = true;
519 			continue;
520 		}
521 		if (dinfo->cfg.baseclass == class &&
522 		    dinfo->cfg.subclass == subclass) {
523 			return (dinfo->cfg.dev);
524 		}
525 	}
526 
527 	return (NULL);
528 }
529 
530 device_t
531 pci_find_base_class_from(uint8_t class, device_t from)
532 {
533 	struct pci_devinfo *dinfo;
534 	bool found = false;
535 
536 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
537 		if (from != NULL && found == false) {
538 			if (from != dinfo->cfg.dev)
539 				continue;
540 			found = true;
541 			continue;
542 		}
543 		if (dinfo->cfg.baseclass == class) {
544 			return (dinfo->cfg.dev);
545 		}
546 	}
547 
548 	return (NULL);
549 }
550 
551 static int
552 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
553 {
554 	va_list ap;
555 	int retval;
556 
557 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
558 	    cfg->func);
559 	va_start(ap, fmt);
560 	retval += vprintf(fmt, ap);
561 	va_end(ap);
562 	return (retval);
563 }
564 
565 /* return base address of memory or port map */
566 
567 static pci_addr_t
568 pci_mapbase(uint64_t mapreg)
569 {
570 
571 	if (PCI_BAR_MEM(mapreg))
572 		return (mapreg & PCIM_BAR_MEM_BASE);
573 	else
574 		return (mapreg & PCIM_BAR_IO_BASE);
575 }
576 
577 /* return map type of memory or port map */
578 
579 static const char *
580 pci_maptype(uint64_t mapreg)
581 {
582 
583 	if (PCI_BAR_IO(mapreg))
584 		return ("I/O Port");
585 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
586 		return ("Prefetchable Memory");
587 	return ("Memory");
588 }
589 
590 /* return log2 of map size decoded for memory or port map */
591 
592 int
593 pci_mapsize(uint64_t testval)
594 {
595 	int ln2size;
596 
597 	testval = pci_mapbase(testval);
598 	ln2size = 0;
599 	if (testval != 0) {
600 		while ((testval & 1) == 0)
601 		{
602 			ln2size++;
603 			testval >>= 1;
604 		}
605 	}
606 	return (ln2size);
607 }
608 
609 /* return base address of device ROM */
610 
611 static pci_addr_t
612 pci_rombase(uint64_t mapreg)
613 {
614 
615 	return (mapreg & PCIM_BIOS_ADDR_MASK);
616 }
617 
618 /* return log2 of map size decided for device ROM */
619 
620 static int
621 pci_romsize(uint64_t testval)
622 {
623 	int ln2size;
624 
625 	testval = pci_rombase(testval);
626 	ln2size = 0;
627 	if (testval != 0) {
628 		while ((testval & 1) == 0)
629 		{
630 			ln2size++;
631 			testval >>= 1;
632 		}
633 	}
634 	return (ln2size);
635 }
636 
637 /* return log2 of address range supported by map register */
638 
639 static int
640 pci_maprange(uint64_t mapreg)
641 {
642 	int ln2range = 0;
643 
644 	if (PCI_BAR_IO(mapreg))
645 		ln2range = 32;
646 	else
647 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
648 		case PCIM_BAR_MEM_32:
649 			ln2range = 32;
650 			break;
651 		case PCIM_BAR_MEM_1MB:
652 			ln2range = 20;
653 			break;
654 		case PCIM_BAR_MEM_64:
655 			ln2range = 64;
656 			break;
657 		}
658 	return (ln2range);
659 }
660 
661 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
662 
663 static void
664 pci_fixancient(pcicfgregs *cfg)
665 {
666 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
667 		return;
668 
669 	/* PCI to PCI bridges use header type 1 */
670 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
671 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
672 }
673 
674 /* extract header type specific config data */
675 
676 static void
677 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
678 {
679 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
680 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
681 	case PCIM_HDRTYPE_NORMAL:
682 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
683 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
684 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
685 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
686 		cfg->nummaps	    = PCI_MAXMAPS_0;
687 		break;
688 	case PCIM_HDRTYPE_BRIDGE:
689 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
690 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
691 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
692 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
693 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
694 		cfg->nummaps	    = PCI_MAXMAPS_1;
695 		break;
696 	case PCIM_HDRTYPE_CARDBUS:
697 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
698 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
699 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
700 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
701 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
702 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
703 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
704 		cfg->nummaps	    = PCI_MAXMAPS_2;
705 		break;
706 	}
707 #undef REG
708 }
709 
710 /* read configuration header into pcicfgregs structure */
711 struct pci_devinfo *
712 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
713 {
714 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
715 	uint16_t vid, did;
716 
717 	vid = REG(PCIR_VENDOR, 2);
718 	if (vid == PCIV_INVALID)
719 		return (NULL);
720 
721 	did = REG(PCIR_DEVICE, 2);
722 
723 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
724 }
725 
726 struct pci_devinfo *
727 pci_alloc_devinfo_method(device_t dev)
728 {
729 
730 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
731 	    M_WAITOK | M_ZERO));
732 }
733 
734 static struct pci_devinfo *
735 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
736     uint16_t vid, uint16_t did)
737 {
738 	struct pci_devinfo *devlist_entry;
739 	pcicfgregs *cfg;
740 
741 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
742 
743 	cfg = &devlist_entry->cfg;
744 
745 	cfg->domain		= d;
746 	cfg->bus		= b;
747 	cfg->slot		= s;
748 	cfg->func		= f;
749 	cfg->vendor		= vid;
750 	cfg->device		= did;
751 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
752 	cfg->statreg		= REG(PCIR_STATUS, 2);
753 	cfg->baseclass		= REG(PCIR_CLASS, 1);
754 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
755 	cfg->progif		= REG(PCIR_PROGIF, 1);
756 	cfg->revid		= REG(PCIR_REVID, 1);
757 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
758 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
759 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
760 	cfg->intpin		= REG(PCIR_INTPIN, 1);
761 	cfg->intline		= REG(PCIR_INTLINE, 1);
762 
763 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
764 	cfg->hdrtype		&= ~PCIM_MFDEV;
765 	STAILQ_INIT(&cfg->maps);
766 
767 	cfg->iov		= NULL;
768 
769 	pci_fixancient(cfg);
770 	pci_hdrtypedata(pcib, b, s, f, cfg);
771 
772 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
773 		pci_read_cap(pcib, cfg);
774 
775 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
776 
777 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
778 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
779 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
780 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
781 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
782 
783 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
784 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
785 	devlist_entry->conf.pc_vendor = cfg->vendor;
786 	devlist_entry->conf.pc_device = cfg->device;
787 
788 	devlist_entry->conf.pc_class = cfg->baseclass;
789 	devlist_entry->conf.pc_subclass = cfg->subclass;
790 	devlist_entry->conf.pc_progif = cfg->progif;
791 	devlist_entry->conf.pc_revid = cfg->revid;
792 
793 	pci_numdevs++;
794 	pci_generation++;
795 
796 	return (devlist_entry);
797 }
798 #undef REG
799 
800 static void
801 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
802 {
803 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
804     cfg->ea.ea_location + (n), w)
805 	int num_ent;
806 	int ptr;
807 	int a, b;
808 	uint32_t val;
809 	int ent_size;
810 	uint32_t dw[4];
811 	uint64_t base, max_offset;
812 	struct pci_ea_entry *eae;
813 
814 	if (cfg->ea.ea_location == 0)
815 		return;
816 
817 	STAILQ_INIT(&cfg->ea.ea_entries);
818 
819 	/* Determine the number of entries */
820 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
821 	num_ent &= PCIM_EA_NUM_ENT_MASK;
822 
823 	/* Find the first entry to care of */
824 	ptr = PCIR_EA_FIRST_ENT;
825 
826 	/* Skip DWORD 2 for type 1 functions */
827 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
828 		ptr += 4;
829 
830 	for (a = 0; a < num_ent; a++) {
831 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
832 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
833 
834 		/* Read a number of dwords in the entry */
835 		val = REG(ptr, 4);
836 		ptr += 4;
837 		ent_size = (val & PCIM_EA_ES);
838 
839 		for (b = 0; b < ent_size; b++) {
840 			dw[b] = REG(ptr, 4);
841 			ptr += 4;
842 		}
843 
844 		eae->eae_flags = val;
845 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
846 
847 		base = dw[0] & PCIM_EA_FIELD_MASK;
848 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
849 		b = 2;
850 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
851 			base |= (uint64_t)dw[b] << 32UL;
852 			b++;
853 		}
854 		if (((dw[1] & PCIM_EA_IS_64) != 0)
855 		    && (b < ent_size)) {
856 			max_offset |= (uint64_t)dw[b] << 32UL;
857 			b++;
858 		}
859 
860 		eae->eae_base = base;
861 		eae->eae_max_offset = max_offset;
862 
863 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
864 
865 		if (bootverbose) {
866 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
867 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
868 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
869 		}
870 	}
871 }
872 #undef REG
873 
874 static void
875 pci_read_cap(device_t pcib, pcicfgregs *cfg)
876 {
877 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
878 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
879 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
880 	uint64_t addr;
881 #endif
882 	uint32_t val;
883 	int	ptr, nextptr, ptrptr;
884 
885 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
886 	case PCIM_HDRTYPE_NORMAL:
887 	case PCIM_HDRTYPE_BRIDGE:
888 		ptrptr = PCIR_CAP_PTR;
889 		break;
890 	case PCIM_HDRTYPE_CARDBUS:
891 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
892 		break;
893 	default:
894 		return;		/* no extended capabilities support */
895 	}
896 	nextptr = REG(ptrptr, 1);	/* sanity check? */
897 
898 	/*
899 	 * Read capability entries.
900 	 */
901 	while (nextptr != 0) {
902 		/* Sanity check */
903 		if (nextptr > 255) {
904 			printf("illegal PCI extended capability offset %d\n",
905 			    nextptr);
906 			return;
907 		}
908 		/* Find the next entry */
909 		ptr = nextptr;
910 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
911 
912 		/* Process this entry */
913 		switch (REG(ptr + PCICAP_ID, 1)) {
914 		case PCIY_PMG:		/* PCI power management */
915 			cfg->pp.pp_location = ptr;
916 			cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
917 			break;
918 		case PCIY_HT:		/* HyperTransport */
919 			/* Determine HT-specific capability type. */
920 			val = REG(ptr + PCIR_HT_COMMAND, 2);
921 
922 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
923 				cfg->ht.ht_slave = ptr;
924 
925 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
926 			switch (val & PCIM_HTCMD_CAP_MASK) {
927 			case PCIM_HTCAP_MSI_MAPPING:
928 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
929 					/* Sanity check the mapping window. */
930 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
931 					    4);
932 					addr <<= 32;
933 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
934 					    4);
935 					if (addr != MSI_INTEL_ADDR_BASE)
936 						device_printf(pcib,
937 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
938 						    cfg->domain, cfg->bus,
939 						    cfg->slot, cfg->func,
940 						    (long long)addr);
941 				} else
942 					addr = MSI_INTEL_ADDR_BASE;
943 
944 				cfg->ht.ht_msimap = ptr;
945 				cfg->ht.ht_msictrl = val;
946 				cfg->ht.ht_msiaddr = addr;
947 				break;
948 			}
949 #endif
950 			break;
951 		case PCIY_MSI:		/* PCI MSI */
952 			cfg->msi.msi_location = ptr;
953 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
954 			break;
955 		case PCIY_MSIX:		/* PCI MSI-X */
956 			cfg->msix.msix_location = ptr;
957 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
958 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
959 			cfg->msix.msix_table_bar = PCIR_BAR(val &
960 			    PCIM_MSIX_BIR_MASK);
961 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
962 			val = REG(ptr + PCIR_MSIX_PBA, 4);
963 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
964 			    PCIM_MSIX_BIR_MASK);
965 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
966 			break;
967 		case PCIY_VPD:		/* PCI Vital Product Data */
968 			cfg->vpd.vpd_reg = ptr;
969 			break;
970 		case PCIY_SUBVENDOR:
971 			/* Should always be true. */
972 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
973 			    PCIM_HDRTYPE_BRIDGE) {
974 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
975 				cfg->subvendor = val & 0xffff;
976 				cfg->subdevice = val >> 16;
977 			}
978 			break;
979 		case PCIY_PCIX:		/* PCI-X */
980 			/*
981 			 * Assume we have a PCI-X chipset if we have
982 			 * at least one PCI-PCI bridge with a PCI-X
983 			 * capability.  Note that some systems with
984 			 * PCI-express or HT chipsets might match on
985 			 * this check as well.
986 			 */
987 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
988 			    PCIM_HDRTYPE_BRIDGE)
989 				pcix_chipset = 1;
990 			cfg->pcix.pcix_location = ptr;
991 			break;
992 		case PCIY_EXPRESS:	/* PCI-express */
993 			/*
994 			 * Assume we have a PCI-express chipset if we have
995 			 * at least one PCI-express device.
996 			 */
997 			pcie_chipset = 1;
998 			cfg->pcie.pcie_location = ptr;
999 			val = REG(ptr + PCIER_FLAGS, 2);
1000 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
1001 			break;
1002 		case PCIY_EA:		/* Enhanced Allocation */
1003 			cfg->ea.ea_location = ptr;
1004 			pci_ea_fill_info(pcib, cfg);
1005 			break;
1006 		default:
1007 			break;
1008 		}
1009 	}
1010 
1011 #if defined(__powerpc__)
1012 	/*
1013 	 * Enable the MSI mapping window for all HyperTransport
1014 	 * slaves.  PCI-PCI bridges have their windows enabled via
1015 	 * PCIB_MAP_MSI().
1016 	 */
1017 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1018 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1019 		device_printf(pcib,
1020 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1021 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1022 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1023 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1024 		     2);
1025 	}
1026 #endif
1027 /* REG and WREG use carry through to next functions */
1028 }
1029 
1030 /*
1031  * PCI Vital Product Data
1032  */
1033 
1034 #define	PCI_VPD_TIMEOUT		1000000
1035 
1036 static int
1037 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1038 {
1039 	int count = PCI_VPD_TIMEOUT;
1040 
1041 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1042 
1043 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1044 
1045 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1046 		if (--count < 0)
1047 			return (ENXIO);
1048 		DELAY(1);	/* limit looping */
1049 	}
1050 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1051 
1052 	return (0);
1053 }
1054 
1055 #if 0
1056 static int
1057 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1058 {
1059 	int count = PCI_VPD_TIMEOUT;
1060 
1061 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1062 
1063 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1064 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1065 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1066 		if (--count < 0)
1067 			return (ENXIO);
1068 		DELAY(1);	/* limit looping */
1069 	}
1070 
1071 	return (0);
1072 }
1073 #endif
1074 
1075 #undef PCI_VPD_TIMEOUT
1076 
1077 struct vpd_readstate {
1078 	device_t	pcib;
1079 	pcicfgregs	*cfg;
1080 	uint32_t	val;
1081 	int		bytesinval;
1082 	int		off;
1083 	uint8_t		cksum;
1084 };
1085 
1086 /* return 0 and one byte in *data if no read error, -1 else */
1087 static int
1088 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1089 {
1090 	uint32_t reg;
1091 	uint8_t byte;
1092 
1093 	if (vrs->bytesinval == 0) {
1094 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1095 			return (-1);
1096 		vrs->val = le32toh(reg);
1097 		vrs->off += 4;
1098 		byte = vrs->val & 0xff;
1099 		vrs->bytesinval = 3;
1100 	} else {
1101 		vrs->val = vrs->val >> 8;
1102 		byte = vrs->val & 0xff;
1103 		vrs->bytesinval--;
1104 	}
1105 
1106 	vrs->cksum += byte;
1107 	*data = byte;
1108 	return (0);
1109 }
1110 
1111 /* return 0 on match, -1 and "unget" byte on no match */
1112 static int
1113 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1114 {
1115 	uint8_t data;
1116 
1117 	if (vpd_nextbyte(vrs, &data) != 0)
1118 		return (-1);
1119 
1120 	if (data == expected)
1121 		return (0);
1122 
1123 	vrs->cksum -= data;
1124 	vrs->val = (vrs->val << 8) + data;
1125 	vrs->bytesinval++;
1126 	return (-1);
1127 }
1128 
1129 /* return size if tag matches, -1 on no match, -2 on read error */
1130 static int
1131 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1132 {
1133 	uint8_t byte1, byte2;
1134 
1135 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1136 		return (-1);
1137 
1138 	if ((vpd_tag & 0x80) == 0)
1139 		return (vpd_tag & 0x07);
1140 
1141 	if (vpd_nextbyte(vrs, &byte1) != 0)
1142 		return (-2);
1143 	if (vpd_nextbyte(vrs, &byte2) != 0)
1144 		return (-2);
1145 
1146 	return ((byte2 << 8) + byte1);
1147 }
1148 
1149 /* (re)allocate buffer in multiples of 8 elements */
1150 static void*
1151 alloc_buffer(void* buffer, size_t element_size, int needed)
1152 {
1153 	int alloc, new_alloc;
1154 
1155 	alloc = roundup2(needed, 8);
1156 	new_alloc = roundup2(needed + 1, 8);
1157 	if (alloc != new_alloc) {
1158 		buffer = reallocf(buffer,
1159 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1160 	}
1161 
1162 	return (buffer);
1163 }
1164 
1165 /* read VPD keyword and return element size, return -1 on read error */
1166 static int
1167 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1168 {
1169 	uint8_t data;
1170 
1171 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1172 		return (-1);
1173 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1174 		return (-1);
1175 	if (vpd_nextbyte(vrs, &data) != 0)
1176 		return (-1);
1177 
1178 	return (data);
1179 }
1180 
1181 /* read VPD data element of given size into allocated buffer */
1182 static char *
1183 vpd_read_value(struct vpd_readstate *vrs, int size)
1184 {
1185 	int i;
1186 	char char1;
1187 	char *value;
1188 
1189 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1190 	for (i = 0; i < size; i++) {
1191 		if (vpd_nextbyte(vrs, &char1) != 0) {
1192 			free(value, M_DEVBUF);
1193 			return (NULL);
1194 		}
1195 		value[i] = char1;
1196 	}
1197 	value[size] = '\0';
1198 
1199 	return (value);
1200 }
1201 
1202 /* read VPD into *keyword and *value, return length of data element */
1203 static int
1204 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1205 {
1206 	int len;
1207 
1208 	len = vpd_read_elem_head(vrs, keyword);
1209 	if (len < 0 || len > maxlen)
1210 		return (-1);
1211 	*value = vpd_read_value(vrs, len);
1212 
1213 	return (len);
1214 }
1215 
1216 /* subtract all data following first byte from checksum of RV element */
1217 static void
1218 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1219 {
1220 	int i;
1221 	uint8_t fixup;
1222 
1223 	fixup = 0;
1224 	for (i = 1; i < len; i++)
1225 		fixup += rvstring[i];
1226 	vrs->cksum -= fixup;
1227 }
1228 
1229 /* fetch one read-only element and return size of heading + data */
1230 static int
1231 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1232 {
1233 	struct pcicfg_vpd *vpd;
1234 	pcicfgregs *cfg;
1235 	struct vpd_readonly *vpd_ros;
1236 	int len;
1237 
1238 	cfg = vrs->cfg;
1239 	vpd = &cfg->vpd;
1240 
1241 	if (maxsize < 3)
1242 		return (-1);
1243 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1244 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1245 	maxsize -= 3;
1246 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1247 	if (vpd_ros->value == NULL)
1248 		return (-1);
1249 	vpd_ros->len = len;
1250 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1251 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1252 		if (vrs->cksum != 0) {
1253 			pci_printf(cfg,
1254 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1255 			return (-1);
1256 		}
1257 	}
1258 	vpd->vpd_rocnt++;
1259 
1260 	return (len + 3);
1261 }
1262 
1263 /* fetch one writable element and return size of heading + data */
1264 static int
1265 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1266 {
1267 	struct pcicfg_vpd *vpd;
1268 	pcicfgregs *cfg;
1269 	struct vpd_write *vpd_w;
1270 	int len;
1271 
1272 	cfg = vrs->cfg;
1273 	vpd = &cfg->vpd;
1274 
1275 	if (maxsize < 3)
1276 		return (-1);
1277 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1278 	if (vpd->vpd_w == NULL) {
1279 		pci_printf(cfg, "out of memory");
1280 		return (-1);
1281 	}
1282 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1283 	maxsize -= 3;
1284 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1285 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1286 	if (vpd_w->value == NULL)
1287 		return (-1);
1288 	vpd_w->len = len;
1289 	vpd->vpd_wcnt++;
1290 
1291 	return (len + 3);
1292 }
1293 
1294 /* free all memory allocated for VPD data */
1295 static void
1296 vpd_free(struct pcicfg_vpd *vpd)
1297 {
1298 	int i;
1299 
1300 	free(vpd->vpd_ident, M_DEVBUF);
1301 	for (i = 0; i < vpd->vpd_rocnt; i++)
1302 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1303 	free(vpd->vpd_ros, M_DEVBUF);
1304 	vpd->vpd_rocnt = 0;
1305 	for (i = 0; i < vpd->vpd_wcnt; i++)
1306 		free(vpd->vpd_w[i].value, M_DEVBUF);
1307 	free(vpd->vpd_w, M_DEVBUF);
1308 	vpd->vpd_wcnt = 0;
1309 }
1310 
1311 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1312 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1313 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1314 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1315 
1316 static int
1317 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1318 {
1319 	struct vpd_readstate vrs;
1320 	int cksumvalid;
1321 	int size, elem_size;
1322 
1323 	/* init vpd reader */
1324 	vrs.bytesinval = 0;
1325 	vrs.off = 0;
1326 	vrs.pcib = pcib;
1327 	vrs.cfg = cfg;
1328 	vrs.cksum = 0;
1329 
1330 	/* read VPD ident element - mandatory */
1331 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1332 	if (size <= 0) {
1333 		pci_printf(cfg, "no VPD ident found\n");
1334 		return (0);
1335 	}
1336 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1337 	if (cfg->vpd.vpd_ident == NULL) {
1338 		pci_printf(cfg, "error accessing VPD ident data\n");
1339 		return (0);
1340 	}
1341 
1342 	/* read VPD RO elements - mandatory */
1343 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1344 	if (size <= 0) {
1345 		pci_printf(cfg, "no read-only VPD data found\n");
1346 		return (0);
1347 	}
1348 	while (size > 0) {
1349 		elem_size = next_vpd_ro_elem(&vrs, size);
1350 		if (elem_size < 0) {
1351 			pci_printf(cfg, "error accessing read-only VPD data\n");
1352 			return (-1);
1353 		}
1354 		size -= elem_size;
1355 	}
1356 	cksumvalid = (vrs.cksum == 0);
1357 	if (!cksumvalid)
1358 		return (-1);
1359 
1360 	/* read VPD RW elements - optional */
1361 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1362 	if (size == -2)
1363 		return (-1);
1364 	while (size > 0) {
1365 		elem_size = next_vpd_rw_elem(&vrs, size);
1366 		if (elem_size < 0) {
1367 			pci_printf(cfg, "error accessing writeable VPD data\n");
1368 			return (-1);
1369 		}
1370 		size -= elem_size;
1371 	}
1372 
1373 	/* read empty END tag - mandatory */
1374 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1375 	if (size != 0) {
1376 		pci_printf(cfg, "No valid VPD end tag found\n");
1377 	}
1378 	return (0);
1379 }
1380 
1381 static void
1382 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1383 {
1384 	int status;
1385 
1386 	status = pci_parse_vpd(pcib, cfg);
1387 	if (status < 0)
1388 		vpd_free(&cfg->vpd);
1389 	cfg->vpd.vpd_cached = 1;
1390 #undef REG
1391 #undef WREG
1392 }
1393 
1394 int
1395 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1396 {
1397 	struct pci_devinfo *dinfo = device_get_ivars(child);
1398 	pcicfgregs *cfg = &dinfo->cfg;
1399 
1400 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1401 		pci_read_vpd(device_get_parent(dev), cfg);
1402 
1403 	*identptr = cfg->vpd.vpd_ident;
1404 
1405 	if (*identptr == NULL)
1406 		return (ENXIO);
1407 
1408 	return (0);
1409 }
1410 
1411 int
1412 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1413 	const char **vptr)
1414 {
1415 	struct pci_devinfo *dinfo = device_get_ivars(child);
1416 	pcicfgregs *cfg = &dinfo->cfg;
1417 	int i;
1418 
1419 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1420 		pci_read_vpd(device_get_parent(dev), cfg);
1421 
1422 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1423 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1424 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1425 			*vptr = cfg->vpd.vpd_ros[i].value;
1426 			return (0);
1427 		}
1428 
1429 	*vptr = NULL;
1430 	return (ENXIO);
1431 }
1432 
1433 struct pcicfg_vpd *
1434 pci_fetch_vpd_list(device_t dev)
1435 {
1436 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1437 	pcicfgregs *cfg = &dinfo->cfg;
1438 
1439 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1440 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1441 	return (&cfg->vpd);
1442 }
1443 
1444 /*
1445  * Find the requested HyperTransport capability and return the offset
1446  * in configuration space via the pointer provided.  The function
1447  * returns 0 on success and an error code otherwise.
1448  */
1449 int
1450 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1451 {
1452 	int ptr, error;
1453 	uint16_t val;
1454 
1455 	error = pci_find_cap(child, PCIY_HT, &ptr);
1456 	if (error)
1457 		return (error);
1458 
1459 	/*
1460 	 * Traverse the capabilities list checking each HT capability
1461 	 * to see if it matches the requested HT capability.
1462 	 */
1463 	for (;;) {
1464 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1465 		if (capability == PCIM_HTCAP_SLAVE ||
1466 		    capability == PCIM_HTCAP_HOST)
1467 			val &= 0xe000;
1468 		else
1469 			val &= PCIM_HTCMD_CAP_MASK;
1470 		if (val == capability) {
1471 			if (capreg != NULL)
1472 				*capreg = ptr;
1473 			return (0);
1474 		}
1475 
1476 		/* Skip to the next HT capability. */
1477 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1478 			break;
1479 	}
1480 
1481 	return (ENOENT);
1482 }
1483 
1484 /*
1485  * Find the next requested HyperTransport capability after start and return
1486  * the offset in configuration space via the pointer provided.  The function
1487  * returns 0 on success and an error code otherwise.
1488  */
1489 int
1490 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1491     int start, int *capreg)
1492 {
1493 	int ptr;
1494 	uint16_t val;
1495 
1496 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1497 	    ("start capability is not HyperTransport capability"));
1498 	ptr = start;
1499 
1500 	/*
1501 	 * Traverse the capabilities list checking each HT capability
1502 	 * to see if it matches the requested HT capability.
1503 	 */
1504 	for (;;) {
1505 		/* Skip to the next HT capability. */
1506 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1507 			break;
1508 
1509 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1510 		if (capability == PCIM_HTCAP_SLAVE ||
1511 		    capability == PCIM_HTCAP_HOST)
1512 			val &= 0xe000;
1513 		else
1514 			val &= PCIM_HTCMD_CAP_MASK;
1515 		if (val == capability) {
1516 			if (capreg != NULL)
1517 				*capreg = ptr;
1518 			return (0);
1519 		}
1520 	}
1521 
1522 	return (ENOENT);
1523 }
1524 
1525 /*
1526  * Find the requested capability and return the offset in
1527  * configuration space via the pointer provided.  The function returns
1528  * 0 on success and an error code otherwise.
1529  */
1530 int
1531 pci_find_cap_method(device_t dev, device_t child, int capability,
1532     int *capreg)
1533 {
1534 	struct pci_devinfo *dinfo = device_get_ivars(child);
1535 	pcicfgregs *cfg = &dinfo->cfg;
1536 	uint32_t status;
1537 	uint8_t ptr;
1538 	int cnt;
1539 
1540 	/*
1541 	 * Check the CAP_LIST bit of the PCI status register first.
1542 	 */
1543 	status = pci_read_config(child, PCIR_STATUS, 2);
1544 	if (!(status & PCIM_STATUS_CAPPRESENT))
1545 		return (ENXIO);
1546 
1547 	/*
1548 	 * Determine the start pointer of the capabilities list.
1549 	 */
1550 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1551 	case PCIM_HDRTYPE_NORMAL:
1552 	case PCIM_HDRTYPE_BRIDGE:
1553 		ptr = PCIR_CAP_PTR;
1554 		break;
1555 	case PCIM_HDRTYPE_CARDBUS:
1556 		ptr = PCIR_CAP_PTR_2;
1557 		break;
1558 	default:
1559 		/* XXX: panic? */
1560 		return (ENXIO);		/* no extended capabilities support */
1561 	}
1562 	ptr = pci_read_config(child, ptr, 1);
1563 
1564 	/*
1565 	 * Traverse the capabilities list.  Limit by total theoretical
1566 	 * maximum number of caps: capability needs at least id and
1567 	 * next registers, and any type X header cannot contain caps.
1568 	 */
1569 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1570 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1571 			if (capreg != NULL)
1572 				*capreg = ptr;
1573 			return (0);
1574 		}
1575 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1576 	}
1577 
1578 	return (ENOENT);
1579 }
1580 
1581 /*
1582  * Find the next requested capability after start and return the offset in
1583  * configuration space via the pointer provided.  The function returns
1584  * 0 on success and an error code otherwise.
1585  */
1586 int
1587 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1588     int start, int *capreg)
1589 {
1590 	uint8_t ptr;
1591 
1592 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1593 	    ("start capability is not expected capability"));
1594 
1595 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1596 	while (ptr != 0) {
1597 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1598 			if (capreg != NULL)
1599 				*capreg = ptr;
1600 			return (0);
1601 		}
1602 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1603 	}
1604 
1605 	return (ENOENT);
1606 }
1607 
1608 /*
1609  * Find the requested extended capability and return the offset in
1610  * configuration space via the pointer provided.  The function returns
1611  * 0 on success and an error code otherwise.
1612  */
1613 int
1614 pci_find_extcap_method(device_t dev, device_t child, int capability,
1615     int *capreg)
1616 {
1617 	struct pci_devinfo *dinfo = device_get_ivars(child);
1618 	pcicfgregs *cfg = &dinfo->cfg;
1619 	uint32_t ecap;
1620 	uint16_t ptr;
1621 
1622 	/* Only supported for PCI-express devices. */
1623 	if (cfg->pcie.pcie_location == 0)
1624 		return (ENXIO);
1625 
1626 	ptr = PCIR_EXTCAP;
1627 	ecap = pci_read_config(child, ptr, 4);
1628 	if (ecap == 0xffffffff || ecap == 0)
1629 		return (ENOENT);
1630 	for (;;) {
1631 		if (PCI_EXTCAP_ID(ecap) == capability) {
1632 			if (capreg != NULL)
1633 				*capreg = ptr;
1634 			return (0);
1635 		}
1636 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1637 		if (ptr == 0)
1638 			break;
1639 		ecap = pci_read_config(child, ptr, 4);
1640 	}
1641 
1642 	return (ENOENT);
1643 }
1644 
1645 /*
1646  * Find the next requested extended capability after start and return the
1647  * offset in configuration space via the pointer provided.  The function
1648  * returns 0 on success and an error code otherwise.
1649  */
1650 int
1651 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1652     int start, int *capreg)
1653 {
1654 	struct pci_devinfo *dinfo = device_get_ivars(child);
1655 	pcicfgregs *cfg = &dinfo->cfg;
1656 	uint32_t ecap;
1657 	uint16_t ptr;
1658 
1659 	/* Only supported for PCI-express devices. */
1660 	if (cfg->pcie.pcie_location == 0)
1661 		return (ENXIO);
1662 
1663 	ecap = pci_read_config(child, start, 4);
1664 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1665 	    ("start extended capability is not expected capability"));
1666 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1667 	while (ptr != 0) {
1668 		ecap = pci_read_config(child, ptr, 4);
1669 		if (PCI_EXTCAP_ID(ecap) == capability) {
1670 			if (capreg != NULL)
1671 				*capreg = ptr;
1672 			return (0);
1673 		}
1674 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1675 	}
1676 
1677 	return (ENOENT);
1678 }
1679 
1680 /*
1681  * Support for MSI-X message interrupts.
1682  */
1683 static void
1684 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1685 {
1686 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1687 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1688 	uint32_t offset;
1689 
1690 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1691 	offset = msix->msix_table_offset + index * 16;
1692 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1693 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1694 	bus_write_4(msix->msix_table_res, offset + 8, data);
1695 }
1696 
1697 void
1698 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1699     uint64_t address, uint32_t data)
1700 {
1701 
1702 	if (pci_msix_rewrite_table) {
1703 		struct pci_devinfo *dinfo = device_get_ivars(child);
1704 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1705 
1706 		/*
1707 		 * Some VM hosts require MSIX to be disabled in the
1708 		 * control register before updating the MSIX table
1709 		 * entries are allowed. It is not enough to only
1710 		 * disable MSIX while updating a single entry. MSIX
1711 		 * must be disabled while updating all entries in the
1712 		 * table.
1713 		 */
1714 		pci_write_config(child,
1715 		    msix->msix_location + PCIR_MSIX_CTRL,
1716 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1717 		pci_resume_msix(child);
1718 	} else
1719 		pci_write_msix_entry(child, index, address, data);
1720 
1721 	/* Enable MSI -> HT mapping. */
1722 	pci_ht_map_msi(child, address);
1723 }
1724 
1725 void
1726 pci_mask_msix(device_t dev, u_int index)
1727 {
1728 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1729 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1730 	uint32_t offset, val;
1731 
1732 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1733 	offset = msix->msix_table_offset + index * 16 + 12;
1734 	val = bus_read_4(msix->msix_table_res, offset);
1735 	val |= PCIM_MSIX_VCTRL_MASK;
1736 
1737 	/*
1738 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1739 	 * register, so always write the new value.
1740 	 */
1741 	bus_write_4(msix->msix_table_res, offset, val);
1742 }
1743 
1744 void
1745 pci_unmask_msix(device_t dev, u_int index)
1746 {
1747 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1748 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1749 	uint32_t offset, val;
1750 
1751 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1752 	offset = msix->msix_table_offset + index * 16 + 12;
1753 	val = bus_read_4(msix->msix_table_res, offset);
1754 	val &= ~PCIM_MSIX_VCTRL_MASK;
1755 
1756 	/*
1757 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1758 	 * register, so always write the new value.
1759 	 */
1760 	bus_write_4(msix->msix_table_res, offset, val);
1761 }
1762 
1763 int
1764 pci_pending_msix(device_t dev, u_int index)
1765 {
1766 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1767 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1768 	uint32_t offset, bit;
1769 
1770 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1771 	offset = msix->msix_pba_offset + (index / 32) * 4;
1772 	bit = 1 << index % 32;
1773 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1774 }
1775 
1776 /*
1777  * Restore MSI-X registers and table during resume.  If MSI-X is
1778  * enabled then walk the virtual table to restore the actual MSI-X
1779  * table.
1780  */
1781 static void
1782 pci_resume_msix(device_t dev)
1783 {
1784 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1785 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1786 	struct msix_table_entry *mte;
1787 	struct msix_vector *mv;
1788 	u_int i, msgnum;
1789 
1790 	if (msix->msix_alloc > 0) {
1791 		msgnum = PCI_MSIX_MSGNUM(msix->msix_ctrl);
1792 
1793 		/* First, mask all vectors. */
1794 		for (i = 0; i < msgnum; i++)
1795 			pci_mask_msix(dev, i);
1796 
1797 		/* Second, program any messages with at least one handler. */
1798 		for (i = 0; i < msix->msix_table_len; i++) {
1799 			mte = &msix->msix_table[i];
1800 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1801 				continue;
1802 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1803 			pci_write_msix_entry(dev, i, mv->mv_address,
1804 			    mv->mv_data);
1805 			pci_unmask_msix(dev, i);
1806 		}
1807 	}
1808 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1809 	    msix->msix_ctrl, 2);
1810 }
1811 
1812 /*
1813  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1814  * returned in *count.  After this function returns, each message will be
1815  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1816  */
1817 int
1818 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1819 {
1820 	struct pci_devinfo *dinfo = device_get_ivars(child);
1821 	pcicfgregs *cfg = &dinfo->cfg;
1822 	struct resource_list_entry *rle;
1823 	u_int actual, i, max;
1824 	int error, irq;
1825 	uint16_t ctrl, msgnum;
1826 
1827 	/* Don't let count == 0 get us into trouble. */
1828 	if (*count < 1)
1829 		return (EINVAL);
1830 
1831 	/* If rid 0 is allocated, then fail. */
1832 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1833 	if (rle != NULL && rle->res != NULL)
1834 		return (ENXIO);
1835 
1836 	/* Already have allocated messages? */
1837 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1838 		return (ENXIO);
1839 
1840 	/* If MSI-X is blacklisted for this system, fail. */
1841 	if (pci_msix_blacklisted())
1842 		return (ENXIO);
1843 
1844 	/* MSI-X capability present? */
1845 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1846 		return (ENODEV);
1847 
1848 	/* Make sure the appropriate BARs are mapped. */
1849 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1850 	    cfg->msix.msix_table_bar);
1851 	if (rle == NULL || rle->res == NULL ||
1852 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1853 		return (ENXIO);
1854 	cfg->msix.msix_table_res = rle->res;
1855 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1856 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1857 		    cfg->msix.msix_pba_bar);
1858 		if (rle == NULL || rle->res == NULL ||
1859 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1860 			return (ENXIO);
1861 	}
1862 	cfg->msix.msix_pba_res = rle->res;
1863 
1864 	ctrl = pci_read_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1865 	    2);
1866 	msgnum = PCI_MSIX_MSGNUM(ctrl);
1867 	if (bootverbose)
1868 		device_printf(child,
1869 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1870 		    *count, msgnum);
1871 	max = min(*count, msgnum);
1872 	for (i = 0; i < max; i++) {
1873 		/* Allocate a message. */
1874 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1875 		if (error) {
1876 			if (i == 0)
1877 				return (error);
1878 			break;
1879 		}
1880 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1881 		    irq, 1);
1882 	}
1883 	actual = i;
1884 
1885 	if (bootverbose) {
1886 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1887 		if (actual == 1)
1888 			device_printf(child, "using IRQ %ju for MSI-X\n",
1889 			    rle->start);
1890 		else {
1891 			bool run;
1892 
1893 			/*
1894 			 * Be fancy and try to print contiguous runs of
1895 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1896 			 * 'run' is true if we are in a range.
1897 			 */
1898 			device_printf(child, "using IRQs %ju", rle->start);
1899 			irq = rle->start;
1900 			run = false;
1901 			for (i = 1; i < actual; i++) {
1902 				rle = resource_list_find(&dinfo->resources,
1903 				    SYS_RES_IRQ, i + 1);
1904 
1905 				/* Still in a run? */
1906 				if (rle->start == irq + 1) {
1907 					run = true;
1908 					irq++;
1909 					continue;
1910 				}
1911 
1912 				/* Finish previous range. */
1913 				if (run) {
1914 					printf("-%d", irq);
1915 					run = false;
1916 				}
1917 
1918 				/* Start new range. */
1919 				printf(",%ju", rle->start);
1920 				irq = rle->start;
1921 			}
1922 
1923 			/* Unfinished range? */
1924 			if (run)
1925 				printf("-%d", irq);
1926 			printf(" for MSI-X\n");
1927 		}
1928 	}
1929 
1930 	/* Mask all vectors. */
1931 	for (i = 0; i < msgnum; i++)
1932 		pci_mask_msix(child, i);
1933 
1934 	/* Allocate and initialize vector data and virtual table. */
1935 	cfg->msix.msix_vectors = mallocarray(actual, sizeof(struct msix_vector),
1936 	    M_DEVBUF, M_WAITOK | M_ZERO);
1937 	cfg->msix.msix_table = mallocarray(actual,
1938 	    sizeof(struct msix_table_entry), M_DEVBUF, M_WAITOK | M_ZERO);
1939 	for (i = 0; i < actual; i++) {
1940 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1941 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1942 		cfg->msix.msix_table[i].mte_vector = i + 1;
1943 	}
1944 
1945 	/* Update control register to enable MSI-X. */
1946 	ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1947 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1948 	    ctrl, 2);
1949 	cfg->msix.msix_ctrl = ctrl;
1950 
1951 	/* Update counts of alloc'd messages. */
1952 	cfg->msix.msix_alloc = actual;
1953 	cfg->msix.msix_table_len = actual;
1954 	*count = actual;
1955 	return (0);
1956 }
1957 
1958 /*
1959  * By default, pci_alloc_msix() will assign the allocated IRQ
1960  * resources consecutively to the first N messages in the MSI-X table.
1961  * However, device drivers may want to use different layouts if they
1962  * either receive fewer messages than they asked for, or they wish to
1963  * populate the MSI-X table sparsely.  This method allows the driver
1964  * to specify what layout it wants.  It must be called after a
1965  * successful pci_alloc_msix() but before any of the associated
1966  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1967  *
1968  * The 'vectors' array contains 'count' message vectors.  The array
1969  * maps directly to the MSI-X table in that index 0 in the array
1970  * specifies the vector for the first message in the MSI-X table, etc.
1971  * The vector value in each array index can either be 0 to indicate
1972  * that no vector should be assigned to a message slot, or it can be a
1973  * number from 1 to N (where N is the count returned from a
1974  * succcessful call to pci_alloc_msix()) to indicate which message
1975  * vector (IRQ) to be used for the corresponding message.
1976  *
1977  * On successful return, each message with a non-zero vector will have
1978  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1979  * 1.  Additionally, if any of the IRQs allocated via the previous
1980  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1981  * will be freed back to the system automatically.
1982  *
1983  * For example, suppose a driver has a MSI-X table with 6 messages and
1984  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1985  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1986  * C.  After the call to pci_alloc_msix(), the device will be setup to
1987  * have an MSI-X table of ABC--- (where - means no vector assigned).
1988  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1989  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1990  * be freed back to the system.  This device will also have valid
1991  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1992  *
1993  * In any case, the SYS_RES_IRQ rid X will always map to the message
1994  * at MSI-X table index X - 1 and will only be valid if a vector is
1995  * assigned to that table entry.
1996  */
1997 int
1998 pci_remap_msix_method(device_t dev, device_t child, int count,
1999     const u_int *vectors)
2000 {
2001 	struct pci_devinfo *dinfo = device_get_ivars(child);
2002 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2003 	struct resource_list_entry *rle;
2004 	u_int i, irq, j;
2005 	bool *used;
2006 
2007 	/*
2008 	 * Have to have at least one message in the table but the
2009 	 * table can't be bigger than the actual MSI-X table in the
2010 	 * device.
2011 	 */
2012 	if (count < 1 || count > PCI_MSIX_MSGNUM(msix->msix_ctrl))
2013 		return (EINVAL);
2014 
2015 	/* Sanity check the vectors. */
2016 	for (i = 0; i < count; i++)
2017 		if (vectors[i] > msix->msix_alloc)
2018 			return (EINVAL);
2019 
2020 	/*
2021 	 * Make sure there aren't any holes in the vectors to be used.
2022 	 * It's a big pain to support it, and it doesn't really make
2023 	 * sense anyway.  Also, at least one vector must be used.
2024 	 */
2025 	used = mallocarray(msix->msix_alloc, sizeof(*used), M_DEVBUF, M_WAITOK |
2026 	    M_ZERO);
2027 	for (i = 0; i < count; i++)
2028 		if (vectors[i] != 0)
2029 			used[vectors[i] - 1] = true;
2030 	for (i = 0; i < msix->msix_alloc - 1; i++)
2031 		if (!used[i] && used[i + 1]) {
2032 			free(used, M_DEVBUF);
2033 			return (EINVAL);
2034 		}
2035 	if (!used[0]) {
2036 		free(used, M_DEVBUF);
2037 		return (EINVAL);
2038 	}
2039 
2040 	/* Make sure none of the resources are allocated. */
2041 	for (i = 0; i < msix->msix_table_len; i++) {
2042 		if (msix->msix_table[i].mte_vector == 0)
2043 			continue;
2044 		if (msix->msix_table[i].mte_handlers > 0) {
2045 			free(used, M_DEVBUF);
2046 			return (EBUSY);
2047 		}
2048 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2049 		KASSERT(rle != NULL, ("missing resource"));
2050 		if (rle->res != NULL) {
2051 			free(used, M_DEVBUF);
2052 			return (EBUSY);
2053 		}
2054 	}
2055 
2056 	/* Free the existing resource list entries. */
2057 	for (i = 0; i < msix->msix_table_len; i++) {
2058 		if (msix->msix_table[i].mte_vector == 0)
2059 			continue;
2060 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2061 	}
2062 
2063 	/*
2064 	 * Build the new virtual table keeping track of which vectors are
2065 	 * used.
2066 	 */
2067 	free(msix->msix_table, M_DEVBUF);
2068 	msix->msix_table = mallocarray(count, sizeof(struct msix_table_entry),
2069 	    M_DEVBUF, M_WAITOK | M_ZERO);
2070 	for (i = 0; i < count; i++)
2071 		msix->msix_table[i].mte_vector = vectors[i];
2072 	msix->msix_table_len = count;
2073 
2074 	/* Free any unused IRQs and resize the vectors array if necessary. */
2075 	j = msix->msix_alloc - 1;
2076 	if (!used[j]) {
2077 		struct msix_vector *vec;
2078 
2079 		while (!used[j]) {
2080 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2081 			    msix->msix_vectors[j].mv_irq);
2082 			j--;
2083 		}
2084 		vec = mallocarray(j + 1, sizeof(struct msix_vector), M_DEVBUF,
2085 		    M_WAITOK);
2086 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2087 		    (j + 1));
2088 		free(msix->msix_vectors, M_DEVBUF);
2089 		msix->msix_vectors = vec;
2090 		msix->msix_alloc = j + 1;
2091 	}
2092 	free(used, M_DEVBUF);
2093 
2094 	/* Map the IRQs onto the rids. */
2095 	for (i = 0; i < count; i++) {
2096 		if (vectors[i] == 0)
2097 			continue;
2098 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2099 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2100 		    irq, 1);
2101 	}
2102 
2103 	if (bootverbose) {
2104 		device_printf(child, "Remapped MSI-X IRQs as: ");
2105 		for (i = 0; i < count; i++) {
2106 			if (i != 0)
2107 				printf(", ");
2108 			if (vectors[i] == 0)
2109 				printf("---");
2110 			else
2111 				printf("%d",
2112 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2113 		}
2114 		printf("\n");
2115 	}
2116 
2117 	return (0);
2118 }
2119 
2120 static int
2121 pci_release_msix(device_t dev, device_t child)
2122 {
2123 	struct pci_devinfo *dinfo = device_get_ivars(child);
2124 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2125 	struct resource_list_entry *rle;
2126 	u_int i;
2127 
2128 	/* Do we have any messages to release? */
2129 	if (msix->msix_alloc == 0)
2130 		return (ENODEV);
2131 
2132 	/* Make sure none of the resources are allocated. */
2133 	for (i = 0; i < msix->msix_table_len; i++) {
2134 		if (msix->msix_table[i].mte_vector == 0)
2135 			continue;
2136 		if (msix->msix_table[i].mte_handlers > 0)
2137 			return (EBUSY);
2138 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2139 		KASSERT(rle != NULL, ("missing resource"));
2140 		if (rle->res != NULL)
2141 			return (EBUSY);
2142 	}
2143 
2144 	/* Update control register to disable MSI-X. */
2145 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2146 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2147 	    msix->msix_ctrl, 2);
2148 
2149 	/* Free the resource list entries. */
2150 	for (i = 0; i < msix->msix_table_len; i++) {
2151 		if (msix->msix_table[i].mte_vector == 0)
2152 			continue;
2153 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2154 	}
2155 	free(msix->msix_table, M_DEVBUF);
2156 	msix->msix_table_len = 0;
2157 
2158 	/* Release the IRQs. */
2159 	for (i = 0; i < msix->msix_alloc; i++)
2160 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2161 		    msix->msix_vectors[i].mv_irq);
2162 	free(msix->msix_vectors, M_DEVBUF);
2163 	msix->msix_alloc = 0;
2164 	return (0);
2165 }
2166 
2167 /*
2168  * Return the max supported MSI-X messages this device supports.
2169  * Basically, assuming the MD code can alloc messages, this function
2170  * should return the maximum value that pci_alloc_msix() can return.
2171  * Thus, it is subject to the tunables, etc.
2172  */
2173 int
2174 pci_msix_count_method(device_t dev, device_t child)
2175 {
2176 	struct pci_devinfo *dinfo = device_get_ivars(child);
2177 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2178 	uint16_t ctrl;
2179 
2180 	if (pci_do_msix && msix->msix_location != 0) {
2181 		ctrl = pci_read_config(child, msix->msix_location +
2182 		    PCIR_MSI_CTRL, 2);
2183 		return (PCI_MSIX_MSGNUM(ctrl));
2184 	}
2185 	return (0);
2186 }
2187 
2188 int
2189 pci_msix_pba_bar_method(device_t dev, device_t child)
2190 {
2191 	struct pci_devinfo *dinfo = device_get_ivars(child);
2192 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2193 
2194 	if (pci_do_msix && msix->msix_location != 0)
2195 		return (msix->msix_pba_bar);
2196 	return (-1);
2197 }
2198 
2199 int
2200 pci_msix_table_bar_method(device_t dev, device_t child)
2201 {
2202 	struct pci_devinfo *dinfo = device_get_ivars(child);
2203 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2204 
2205 	if (pci_do_msix && msix->msix_location != 0)
2206 		return (msix->msix_table_bar);
2207 	return (-1);
2208 }
2209 
2210 /*
2211  * HyperTransport MSI mapping control
2212  */
2213 void
2214 pci_ht_map_msi(device_t dev, uint64_t addr)
2215 {
2216 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2217 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2218 
2219 	if (!ht->ht_msimap)
2220 		return;
2221 
2222 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2223 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2224 		/* Enable MSI -> HT mapping. */
2225 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2226 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2227 		    ht->ht_msictrl, 2);
2228 	}
2229 
2230 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2231 		/* Disable MSI -> HT mapping. */
2232 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2233 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2234 		    ht->ht_msictrl, 2);
2235 	}
2236 }
2237 
2238 int
2239 pci_get_relaxed_ordering_enabled(device_t dev)
2240 {
2241 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2242 	int cap;
2243 	uint16_t val;
2244 
2245 	cap = dinfo->cfg.pcie.pcie_location;
2246 	if (cap == 0)
2247 		return (0);
2248 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2249 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2250 	return (val != 0);
2251 }
2252 
2253 int
2254 pci_get_max_payload(device_t dev)
2255 {
2256 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2257 	int cap;
2258 	uint16_t val;
2259 
2260 	cap = dinfo->cfg.pcie.pcie_location;
2261 	if (cap == 0)
2262 		return (0);
2263 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2264 	val &= PCIEM_CTL_MAX_PAYLOAD;
2265 	val >>= 5;
2266 	return (1 << (val + 7));
2267 }
2268 
2269 int
2270 pci_get_max_read_req(device_t dev)
2271 {
2272 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2273 	int cap;
2274 	uint16_t val;
2275 
2276 	cap = dinfo->cfg.pcie.pcie_location;
2277 	if (cap == 0)
2278 		return (0);
2279 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2280 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2281 	val >>= 12;
2282 	return (1 << (val + 7));
2283 }
2284 
2285 int
2286 pci_set_max_read_req(device_t dev, int size)
2287 {
2288 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2289 	int cap;
2290 	uint16_t val;
2291 
2292 	cap = dinfo->cfg.pcie.pcie_location;
2293 	if (cap == 0)
2294 		return (0);
2295 	if (size < 128)
2296 		size = 128;
2297 	if (size > 4096)
2298 		size = 4096;
2299 	size = (1 << (fls(size) - 1));
2300 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2301 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2302 	val |= (fls(size) - 8) << 12;
2303 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2304 	return (size);
2305 }
2306 
2307 uint32_t
2308 pcie_read_config(device_t dev, int reg, int width)
2309 {
2310 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2311 	int cap;
2312 
2313 	cap = dinfo->cfg.pcie.pcie_location;
2314 	if (cap == 0) {
2315 		if (width == 2)
2316 			return (0xffff);
2317 		return (0xffffffff);
2318 	}
2319 
2320 	return (pci_read_config(dev, cap + reg, width));
2321 }
2322 
2323 void
2324 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2325 {
2326 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2327 	int cap;
2328 
2329 	cap = dinfo->cfg.pcie.pcie_location;
2330 	if (cap == 0)
2331 		return;
2332 	pci_write_config(dev, cap + reg, value, width);
2333 }
2334 
2335 /*
2336  * Adjusts a PCI-e capability register by clearing the bits in mask
2337  * and setting the bits in (value & mask).  Bits not set in mask are
2338  * not adjusted.
2339  *
2340  * Returns the old value on success or all ones on failure.
2341  */
2342 uint32_t
2343 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2344     int width)
2345 {
2346 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2347 	uint32_t old, new;
2348 	int cap;
2349 
2350 	cap = dinfo->cfg.pcie.pcie_location;
2351 	if (cap == 0) {
2352 		if (width == 2)
2353 			return (0xffff);
2354 		return (0xffffffff);
2355 	}
2356 
2357 	old = pci_read_config(dev, cap + reg, width);
2358 	new = old & ~mask;
2359 	new |= (value & mask);
2360 	pci_write_config(dev, cap + reg, new, width);
2361 	return (old);
2362 }
2363 
2364 /*
2365  * Support for MSI message signalled interrupts.
2366  */
2367 void
2368 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2369     uint16_t data)
2370 {
2371 	struct pci_devinfo *dinfo = device_get_ivars(child);
2372 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2373 
2374 	/* Write data and address values. */
2375 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2376 	    address & 0xffffffff, 4);
2377 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2378 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2379 		    address >> 32, 4);
2380 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2381 		    data, 2);
2382 	} else
2383 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2384 		    2);
2385 
2386 	/* Enable MSI in the control register. */
2387 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2388 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2389 	    msi->msi_ctrl, 2);
2390 
2391 	/* Enable MSI -> HT mapping. */
2392 	pci_ht_map_msi(child, address);
2393 }
2394 
2395 void
2396 pci_disable_msi_method(device_t dev, device_t child)
2397 {
2398 	struct pci_devinfo *dinfo = device_get_ivars(child);
2399 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2400 
2401 	/* Disable MSI -> HT mapping. */
2402 	pci_ht_map_msi(child, 0);
2403 
2404 	/* Disable MSI in the control register. */
2405 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2406 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2407 	    msi->msi_ctrl, 2);
2408 }
2409 
2410 /*
2411  * Restore MSI registers during resume.  If MSI is enabled then
2412  * restore the data and address registers in addition to the control
2413  * register.
2414  */
2415 static void
2416 pci_resume_msi(device_t dev)
2417 {
2418 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2419 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2420 	uint64_t address;
2421 	uint16_t data;
2422 
2423 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2424 		address = msi->msi_addr;
2425 		data = msi->msi_data;
2426 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2427 		    address & 0xffffffff, 4);
2428 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2429 			pci_write_config(dev, msi->msi_location +
2430 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2431 			pci_write_config(dev, msi->msi_location +
2432 			    PCIR_MSI_DATA_64BIT, data, 2);
2433 		} else
2434 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2435 			    data, 2);
2436 	}
2437 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2438 	    2);
2439 }
2440 
2441 static int
2442 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2443 {
2444 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2445 	pcicfgregs *cfg = &dinfo->cfg;
2446 	struct resource_list_entry *rle;
2447 	struct msix_table_entry *mte;
2448 	struct msix_vector *mv;
2449 	uint64_t addr;
2450 	uint32_t data;
2451 	u_int i, j;
2452 	int error;
2453 
2454 	/*
2455 	 * Handle MSI first.  We try to find this IRQ among our list
2456 	 * of MSI IRQs.  If we find it, we request updated address and
2457 	 * data registers and apply the results.
2458 	 */
2459 	if (cfg->msi.msi_alloc > 0) {
2460 		/* If we don't have any active handlers, nothing to do. */
2461 		if (cfg->msi.msi_handlers == 0)
2462 			return (0);
2463 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2464 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2465 			    i + 1);
2466 			if (rle->start == irq) {
2467 				error = PCIB_MAP_MSI(device_get_parent(bus),
2468 				    dev, irq, &addr, &data);
2469 				if (error)
2470 					return (error);
2471 				pci_disable_msi(dev);
2472 				dinfo->cfg.msi.msi_addr = addr;
2473 				dinfo->cfg.msi.msi_data = data;
2474 				pci_enable_msi(dev, addr, data);
2475 				return (0);
2476 			}
2477 		}
2478 		return (ENOENT);
2479 	}
2480 
2481 	/*
2482 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2483 	 * we request the updated mapping info.  If that works, we go
2484 	 * through all the slots that use this IRQ and update them.
2485 	 */
2486 	if (cfg->msix.msix_alloc > 0) {
2487 		bool found = false;
2488 
2489 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2490 			mv = &cfg->msix.msix_vectors[i];
2491 			if (mv->mv_irq == irq) {
2492 				error = PCIB_MAP_MSI(device_get_parent(bus),
2493 				    dev, irq, &addr, &data);
2494 				if (error)
2495 					return (error);
2496 				mv->mv_address = addr;
2497 				mv->mv_data = data;
2498 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2499 					mte = &cfg->msix.msix_table[j];
2500 					if (mte->mte_vector != i + 1)
2501 						continue;
2502 					if (mte->mte_handlers == 0)
2503 						continue;
2504 					pci_mask_msix(dev, j);
2505 					pci_enable_msix(dev, j, addr, data);
2506 					pci_unmask_msix(dev, j);
2507 				}
2508 				found = true;
2509 			}
2510 		}
2511 		return (found ? 0 : ENOENT);
2512 	}
2513 
2514 	return (ENOENT);
2515 }
2516 
2517 /*
2518  * Returns true if the specified device is blacklisted because MSI
2519  * doesn't work.
2520  */
2521 int
2522 pci_msi_device_blacklisted(device_t dev)
2523 {
2524 
2525 	if (!pci_honor_msi_blacklist)
2526 		return (0);
2527 
2528 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2529 }
2530 
2531 /*
2532  * Determine if MSI is blacklisted globally on this system.  Currently,
2533  * we just check for blacklisted chipsets as represented by the
2534  * host-PCI bridge at device 0:0:0.  In the future, it may become
2535  * necessary to check other system attributes, such as the kenv values
2536  * that give the motherboard manufacturer and model number.
2537  */
2538 static int
2539 pci_msi_blacklisted(void)
2540 {
2541 	device_t dev;
2542 
2543 	if (!pci_honor_msi_blacklist)
2544 		return (0);
2545 
2546 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2547 	if (!(pcie_chipset || pcix_chipset)) {
2548 		if (vm_guest != VM_GUEST_NO) {
2549 			/*
2550 			 * Whitelist older chipsets in virtual
2551 			 * machines known to support MSI.
2552 			 */
2553 			dev = pci_find_bsf(0, 0, 0);
2554 			if (dev != NULL)
2555 				return (!pci_has_quirk(pci_get_devid(dev),
2556 					PCI_QUIRK_ENABLE_MSI_VM));
2557 		}
2558 		return (1);
2559 	}
2560 
2561 	dev = pci_find_bsf(0, 0, 0);
2562 	if (dev != NULL)
2563 		return (pci_msi_device_blacklisted(dev));
2564 	return (0);
2565 }
2566 
2567 /*
2568  * Returns true if the specified device is blacklisted because MSI-X
2569  * doesn't work.  Note that this assumes that if MSI doesn't work,
2570  * MSI-X doesn't either.
2571  */
2572 int
2573 pci_msix_device_blacklisted(device_t dev)
2574 {
2575 
2576 	if (!pci_honor_msi_blacklist)
2577 		return (0);
2578 
2579 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2580 		return (1);
2581 
2582 	return (pci_msi_device_blacklisted(dev));
2583 }
2584 
2585 /*
2586  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2587  * is blacklisted, assume that MSI-X is as well.  Check for additional
2588  * chipsets where MSI works but MSI-X does not.
2589  */
2590 static int
2591 pci_msix_blacklisted(void)
2592 {
2593 	device_t dev;
2594 
2595 	if (!pci_honor_msi_blacklist)
2596 		return (0);
2597 
2598 	dev = pci_find_bsf(0, 0, 0);
2599 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2600 	    PCI_QUIRK_DISABLE_MSIX))
2601 		return (1);
2602 
2603 	return (pci_msi_blacklisted());
2604 }
2605 
2606 /*
2607  * Attempt to allocate *count MSI messages.  The actual number allocated is
2608  * returned in *count.  After this function returns, each message will be
2609  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2610  */
2611 int
2612 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2613 {
2614 	struct pci_devinfo *dinfo = device_get_ivars(child);
2615 	pcicfgregs *cfg = &dinfo->cfg;
2616 	struct resource_list_entry *rle;
2617 	u_int actual, i;
2618 	int error, irqs[32];
2619 	uint16_t ctrl, msgnum;
2620 
2621 	/* Don't let count == 0 get us into trouble. */
2622 	if (*count < 1)
2623 		return (EINVAL);
2624 
2625 	/* If rid 0 is allocated, then fail. */
2626 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2627 	if (rle != NULL && rle->res != NULL)
2628 		return (ENXIO);
2629 
2630 	/* Already have allocated messages? */
2631 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2632 		return (ENXIO);
2633 
2634 	/* If MSI is blacklisted for this system, fail. */
2635 	if (pci_msi_blacklisted())
2636 		return (ENXIO);
2637 
2638 	/* MSI capability present? */
2639 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2640 		return (ENODEV);
2641 
2642 	ctrl = pci_read_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, 2);
2643 	msgnum = PCI_MSI_MSGNUM(ctrl);
2644 	if (bootverbose)
2645 		device_printf(child,
2646 		    "attempting to allocate %d MSI vectors (%u supported)\n",
2647 		    *count, msgnum);
2648 
2649 	/* Don't ask for more than the device supports. */
2650 	actual = min(*count, msgnum);
2651 
2652 	/* Don't ask for more than 32 messages. */
2653 	actual = min(actual, 32);
2654 
2655 	/* MSI requires power of 2 number of messages. */
2656 	if (!powerof2(actual))
2657 		return (EINVAL);
2658 
2659 	for (;;) {
2660 		/* Try to allocate N messages. */
2661 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2662 		    actual, irqs);
2663 		if (error == 0)
2664 			break;
2665 		if (actual == 1)
2666 			return (error);
2667 
2668 		/* Try N / 2. */
2669 		actual >>= 1;
2670 	}
2671 
2672 	/*
2673 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2674 	 * resources in the irqs[] array, so add new resources
2675 	 * starting at rid 1.
2676 	 */
2677 	for (i = 0; i < actual; i++)
2678 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2679 		    irqs[i], irqs[i], 1);
2680 
2681 	if (bootverbose) {
2682 		if (actual == 1)
2683 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2684 		else {
2685 			bool run;
2686 
2687 			/*
2688 			 * Be fancy and try to print contiguous runs
2689 			 * of IRQ values as ranges.  'run' is true if
2690 			 * we are in a range.
2691 			 */
2692 			device_printf(child, "using IRQs %d", irqs[0]);
2693 			run = false;
2694 			for (i = 1; i < actual; i++) {
2695 				/* Still in a run? */
2696 				if (irqs[i] == irqs[i - 1] + 1) {
2697 					run = true;
2698 					continue;
2699 				}
2700 
2701 				/* Finish previous range. */
2702 				if (run) {
2703 					printf("-%d", irqs[i - 1]);
2704 					run = false;
2705 				}
2706 
2707 				/* Start new range. */
2708 				printf(",%d", irqs[i]);
2709 			}
2710 
2711 			/* Unfinished range? */
2712 			if (run)
2713 				printf("-%d", irqs[actual - 1]);
2714 			printf(" for MSI\n");
2715 		}
2716 	}
2717 
2718 	/* Update control register with actual count. */
2719 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2720 	ctrl |= (ffs(actual) - 1) << 4;
2721 	cfg->msi.msi_ctrl = ctrl;
2722 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2723 
2724 	/* Update counts of alloc'd messages. */
2725 	cfg->msi.msi_alloc = actual;
2726 	cfg->msi.msi_handlers = 0;
2727 	*count = actual;
2728 	return (0);
2729 }
2730 
2731 /* Release the MSI messages associated with this device. */
2732 int
2733 pci_release_msi_method(device_t dev, device_t child)
2734 {
2735 	struct pci_devinfo *dinfo = device_get_ivars(child);
2736 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2737 	struct resource_list_entry *rle;
2738 	u_int i, irqs[32];
2739 	int error;
2740 
2741 	/* Try MSI-X first. */
2742 	error = pci_release_msix(dev, child);
2743 	if (error != ENODEV)
2744 		return (error);
2745 
2746 	/* Do we have any messages to release? */
2747 	if (msi->msi_alloc == 0)
2748 		return (ENODEV);
2749 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2750 
2751 	/* Make sure none of the resources are allocated. */
2752 	if (msi->msi_handlers > 0)
2753 		return (EBUSY);
2754 	for (i = 0; i < msi->msi_alloc; i++) {
2755 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2756 		KASSERT(rle != NULL, ("missing MSI resource"));
2757 		if (rle->res != NULL)
2758 			return (EBUSY);
2759 		irqs[i] = rle->start;
2760 	}
2761 
2762 	/* Update control register with 0 count. */
2763 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2764 	    ("%s: MSI still enabled", __func__));
2765 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2766 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2767 	    msi->msi_ctrl, 2);
2768 
2769 	/* Release the messages. */
2770 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2771 	for (i = 0; i < msi->msi_alloc; i++)
2772 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2773 
2774 	/* Update alloc count. */
2775 	msi->msi_alloc = 0;
2776 	msi->msi_addr = 0;
2777 	msi->msi_data = 0;
2778 	return (0);
2779 }
2780 
2781 /*
2782  * Return the max supported MSI messages this device supports.
2783  * Basically, assuming the MD code can alloc messages, this function
2784  * should return the maximum value that pci_alloc_msi() can return.
2785  * Thus, it is subject to the tunables, etc.
2786  */
2787 int
2788 pci_msi_count_method(device_t dev, device_t child)
2789 {
2790 	struct pci_devinfo *dinfo = device_get_ivars(child);
2791 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2792 	uint16_t ctrl;
2793 
2794 	if (pci_do_msi && msi->msi_location != 0) {
2795 		ctrl = pci_read_config(child, msi->msi_location + PCIR_MSI_CTRL,
2796 		    2);
2797 		return (PCI_MSI_MSGNUM(ctrl));
2798 	}
2799 	return (0);
2800 }
2801 
2802 /* free pcicfgregs structure and all depending data structures */
2803 
2804 int
2805 pci_freecfg(struct pci_devinfo *dinfo)
2806 {
2807 	struct devlist *devlist_head;
2808 	struct pci_map *pm, *next;
2809 
2810 	devlist_head = &pci_devq;
2811 
2812 	if (dinfo->cfg.vpd.vpd_reg)
2813 		vpd_free(&dinfo->cfg.vpd);
2814 
2815 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2816 		free(pm, M_DEVBUF);
2817 	}
2818 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2819 	free(dinfo, M_DEVBUF);
2820 
2821 	/* increment the generation count */
2822 	pci_generation++;
2823 
2824 	/* we're losing one device */
2825 	pci_numdevs--;
2826 	return (0);
2827 }
2828 
2829 /*
2830  * PCI power manangement
2831  */
2832 int
2833 pci_set_powerstate_method(device_t dev, device_t child, int state)
2834 {
2835 	struct pci_devinfo *dinfo = device_get_ivars(child);
2836 	pcicfgregs *cfg = &dinfo->cfg;
2837 	uint16_t status;
2838 	int oldstate, highest, delay;
2839 
2840 	if (cfg->pp.pp_location == 0)
2841 		return (EOPNOTSUPP);
2842 
2843 	/*
2844 	 * Optimize a no state change request away.  While it would be OK to
2845 	 * write to the hardware in theory, some devices have shown odd
2846 	 * behavior when going from D3 -> D3.
2847 	 */
2848 	oldstate = pci_get_powerstate(child);
2849 	if (oldstate == state)
2850 		return (0);
2851 
2852 	/*
2853 	 * The PCI power management specification states that after a state
2854 	 * transition between PCI power states, system software must
2855 	 * guarantee a minimal delay before the function accesses the device.
2856 	 * Compute the worst case delay that we need to guarantee before we
2857 	 * access the device.  Many devices will be responsive much more
2858 	 * quickly than this delay, but there are some that don't respond
2859 	 * instantly to state changes.  Transitions to/from D3 state require
2860 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2861 	 * is done below with DELAY rather than a sleeper function because
2862 	 * this function can be called from contexts where we cannot sleep.
2863 	 */
2864 	highest = (oldstate > state) ? oldstate : state;
2865 	if (highest == PCI_POWERSTATE_D3)
2866 	    delay = 10000;
2867 	else if (highest == PCI_POWERSTATE_D2)
2868 	    delay = 200;
2869 	else
2870 	    delay = 0;
2871 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2872 	    PCIR_POWER_STATUS, 2) & ~PCIM_PSTAT_DMASK;
2873 	switch (state) {
2874 	case PCI_POWERSTATE_D0:
2875 		status |= PCIM_PSTAT_D0;
2876 		break;
2877 	case PCI_POWERSTATE_D1:
2878 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2879 			return (EOPNOTSUPP);
2880 		status |= PCIM_PSTAT_D1;
2881 		break;
2882 	case PCI_POWERSTATE_D2:
2883 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2884 			return (EOPNOTSUPP);
2885 		status |= PCIM_PSTAT_D2;
2886 		break;
2887 	case PCI_POWERSTATE_D3:
2888 		status |= PCIM_PSTAT_D3;
2889 		break;
2890 	default:
2891 		return (EINVAL);
2892 	}
2893 
2894 	if (bootverbose)
2895 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2896 		    state);
2897 
2898 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_location + PCIR_POWER_STATUS,
2899 	    status, 2);
2900 	if (delay)
2901 		DELAY(delay);
2902 	return (0);
2903 }
2904 
2905 int
2906 pci_get_powerstate_method(device_t dev, device_t child)
2907 {
2908 	struct pci_devinfo *dinfo = device_get_ivars(child);
2909 	pcicfgregs *cfg = &dinfo->cfg;
2910 	uint16_t status;
2911 	int result;
2912 
2913 	if (cfg->pp.pp_location != 0) {
2914 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2915 		    PCIR_POWER_STATUS, 2);
2916 		switch (status & PCIM_PSTAT_DMASK) {
2917 		case PCIM_PSTAT_D0:
2918 			result = PCI_POWERSTATE_D0;
2919 			break;
2920 		case PCIM_PSTAT_D1:
2921 			result = PCI_POWERSTATE_D1;
2922 			break;
2923 		case PCIM_PSTAT_D2:
2924 			result = PCI_POWERSTATE_D2;
2925 			break;
2926 		case PCIM_PSTAT_D3:
2927 			result = PCI_POWERSTATE_D3;
2928 			break;
2929 		default:
2930 			result = PCI_POWERSTATE_UNKNOWN;
2931 			break;
2932 		}
2933 	} else {
2934 		/* No support, device is always at D0 */
2935 		result = PCI_POWERSTATE_D0;
2936 	}
2937 	return (result);
2938 }
2939 
2940 /* Clear any active PME# and disable PME# generation. */
2941 void
2942 pci_clear_pme(device_t dev)
2943 {
2944 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2945 	pcicfgregs *cfg = &dinfo->cfg;
2946 	uint16_t status;
2947 
2948 	if (cfg->pp.pp_location != 0) {
2949 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2950 		    PCIR_POWER_STATUS, 2);
2951 		status &= ~PCIM_PSTAT_PMEENABLE;
2952 		status |= PCIM_PSTAT_PME;
2953 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2954 		    PCIR_POWER_STATUS, status, 2);
2955 	}
2956 }
2957 
2958 /* Clear any active PME# and enable PME# generation. */
2959 void
2960 pci_enable_pme(device_t dev)
2961 {
2962 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2963 	pcicfgregs *cfg = &dinfo->cfg;
2964 	uint16_t status;
2965 
2966 	if (cfg->pp.pp_location != 0) {
2967 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2968 		    PCIR_POWER_STATUS, 2);
2969 		status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2970 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2971 		    PCIR_POWER_STATUS, status, 2);
2972 	}
2973 }
2974 
2975 bool
2976 pci_has_pm(device_t dev)
2977 {
2978 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2979 	pcicfgregs *cfg = &dinfo->cfg;
2980 
2981 	return (cfg->pp.pp_location != 0);
2982 }
2983 
2984 /*
2985  * Some convenience functions for PCI device drivers.
2986  */
2987 
2988 static __inline void
2989 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2990 {
2991 	uint16_t	command;
2992 
2993 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2994 	command |= bit;
2995 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2996 }
2997 
2998 static __inline void
2999 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
3000 {
3001 	uint16_t	command;
3002 
3003 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3004 	command &= ~bit;
3005 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3006 }
3007 
3008 int
3009 pci_enable_busmaster_method(device_t dev, device_t child)
3010 {
3011 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3012 	return (0);
3013 }
3014 
3015 int
3016 pci_disable_busmaster_method(device_t dev, device_t child)
3017 {
3018 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3019 	return (0);
3020 }
3021 
3022 int
3023 pci_enable_io_method(device_t dev, device_t child, int space)
3024 {
3025 	uint16_t bit;
3026 
3027 	switch(space) {
3028 	case SYS_RES_IOPORT:
3029 		bit = PCIM_CMD_PORTEN;
3030 		break;
3031 	case SYS_RES_MEMORY:
3032 		bit = PCIM_CMD_MEMEN;
3033 		break;
3034 	default:
3035 		return (EINVAL);
3036 	}
3037 	pci_set_command_bit(dev, child, bit);
3038 	return (0);
3039 }
3040 
3041 int
3042 pci_disable_io_method(device_t dev, device_t child, int space)
3043 {
3044 	uint16_t bit;
3045 
3046 	switch(space) {
3047 	case SYS_RES_IOPORT:
3048 		bit = PCIM_CMD_PORTEN;
3049 		break;
3050 	case SYS_RES_MEMORY:
3051 		bit = PCIM_CMD_MEMEN;
3052 		break;
3053 	default:
3054 		return (EINVAL);
3055 	}
3056 	pci_clear_command_bit(dev, child, bit);
3057 	return (0);
3058 }
3059 
3060 /*
3061  * New style pci driver.  Parent device is either a pci-host-bridge or a
3062  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
3063  */
3064 
3065 void
3066 pci_print_verbose(struct pci_devinfo *dinfo)
3067 {
3068 
3069 	if (bootverbose) {
3070 		pcicfgregs *cfg = &dinfo->cfg;
3071 
3072 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
3073 		    cfg->vendor, cfg->device, cfg->revid);
3074 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
3075 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
3076 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
3077 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
3078 		    cfg->mfdev);
3079 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
3080 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
3081 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
3082 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3083 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3084 		if (cfg->intpin > 0)
3085 			printf("\tintpin=%c, irq=%d\n",
3086 			    cfg->intpin +'a' -1, cfg->intline);
3087 		if (cfg->pp.pp_location) {
3088 			uint16_t status;
3089 
3090 			status = pci_read_config(cfg->dev, cfg->pp.pp_location +
3091 			    PCIR_POWER_STATUS, 2);
3092 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3093 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3094 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3095 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3096 			    status & PCIM_PSTAT_DMASK);
3097 		}
3098 		if (cfg->msi.msi_location) {
3099 			uint16_t ctrl, msgnum;
3100 
3101 			ctrl = cfg->msi.msi_ctrl;
3102 			msgnum = PCI_MSI_MSGNUM(ctrl);
3103 			printf("\tMSI supports %d message%s%s%s\n",
3104 			    msgnum, (msgnum == 1) ? "" : "s",
3105 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3106 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3107 		}
3108 		if (cfg->msix.msix_location) {
3109 			uint16_t msgnum;
3110 
3111 			msgnum = PCI_MSIX_MSGNUM(cfg->msix.msix_ctrl);
3112 			printf("\tMSI-X supports %d message%s ",
3113 			    msgnum, (msgnum == 1) ? "" : "s");
3114 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3115 				printf("in map 0x%x\n",
3116 				    cfg->msix.msix_table_bar);
3117 			else
3118 				printf("in maps 0x%x and 0x%x\n",
3119 				    cfg->msix.msix_table_bar,
3120 				    cfg->msix.msix_pba_bar);
3121 		}
3122 	}
3123 }
3124 
3125 static int
3126 pci_porten(device_t dev)
3127 {
3128 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3129 }
3130 
3131 static int
3132 pci_memen(device_t dev)
3133 {
3134 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3135 }
3136 
3137 void
3138 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3139     int *bar64)
3140 {
3141 	struct pci_devinfo *dinfo;
3142 	pci_addr_t map, testval;
3143 	int ln2range;
3144 	uint16_t cmd;
3145 
3146 	/*
3147 	 * The device ROM BAR is special.  It is always a 32-bit
3148 	 * memory BAR.  Bit 0 is special and should not be set when
3149 	 * sizing the BAR.
3150 	 */
3151 	dinfo = device_get_ivars(dev);
3152 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3153 		map = pci_read_config(dev, reg, 4);
3154 		pci_write_config(dev, reg, 0xfffffffe, 4);
3155 		testval = pci_read_config(dev, reg, 4);
3156 		pci_write_config(dev, reg, map, 4);
3157 		*mapp = map;
3158 		*testvalp = testval;
3159 		if (bar64 != NULL)
3160 			*bar64 = 0;
3161 		return;
3162 	}
3163 
3164 	map = pci_read_config(dev, reg, 4);
3165 	ln2range = pci_maprange(map);
3166 	if (ln2range == 64)
3167 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3168 
3169 	/*
3170 	 * Disable decoding via the command register before
3171 	 * determining the BAR's length since we will be placing it in
3172 	 * a weird state.
3173 	 */
3174 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3175 	pci_write_config(dev, PCIR_COMMAND,
3176 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3177 
3178 	/*
3179 	 * Determine the BAR's length by writing all 1's.  The bottom
3180 	 * log_2(size) bits of the BAR will stick as 0 when we read
3181 	 * the value back.
3182 	 *
3183 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3184 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3185 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3186 	 *
3187 	 * Writes to both registers must be performed before attempting to
3188 	 * read back the size value.
3189 	 */
3190 	testval = 0;
3191 	pci_write_config(dev, reg, 0xffffffff, 4);
3192 	if (ln2range == 64) {
3193 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3194 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3195 	}
3196 	testval |= pci_read_config(dev, reg, 4);
3197 
3198 	/*
3199 	 * Restore the original value of the BAR.  We may have reprogrammed
3200 	 * the BAR of the low-level console device and when booting verbose,
3201 	 * we need the console device addressable.
3202 	 */
3203 	pci_write_config(dev, reg, map, 4);
3204 	if (ln2range == 64)
3205 		pci_write_config(dev, reg + 4, map >> 32, 4);
3206 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3207 
3208 	*mapp = map;
3209 	*testvalp = testval;
3210 	if (bar64 != NULL)
3211 		*bar64 = (ln2range == 64);
3212 }
3213 
3214 static void
3215 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3216 {
3217 	struct pci_devinfo *dinfo;
3218 	int ln2range;
3219 
3220 	/* The device ROM BAR is always a 32-bit memory BAR. */
3221 	dinfo = device_get_ivars(dev);
3222 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3223 		ln2range = 32;
3224 	else
3225 		ln2range = pci_maprange(pm->pm_value);
3226 	pci_write_config(dev, pm->pm_reg, base, 4);
3227 	if (ln2range == 64)
3228 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3229 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3230 	if (ln2range == 64)
3231 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3232 		    pm->pm_reg + 4, 4) << 32;
3233 }
3234 
3235 struct pci_map *
3236 pci_find_bar(device_t dev, int reg)
3237 {
3238 	struct pci_devinfo *dinfo;
3239 	struct pci_map *pm;
3240 
3241 	dinfo = device_get_ivars(dev);
3242 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3243 		if (pm->pm_reg == reg)
3244 			return (pm);
3245 	}
3246 	return (NULL);
3247 }
3248 
3249 struct pci_map *
3250 pci_first_bar(device_t dev)
3251 {
3252 	struct pci_devinfo *dinfo;
3253 
3254 	dinfo = device_get_ivars(dev);
3255 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3256 }
3257 
3258 struct pci_map *
3259 pci_next_bar(struct pci_map *pm)
3260 {
3261 	return (STAILQ_NEXT(pm, pm_link));
3262 }
3263 
3264 int
3265 pci_bar_enabled(device_t dev, struct pci_map *pm)
3266 {
3267 	struct pci_devinfo *dinfo;
3268 	uint16_t cmd;
3269 
3270 	dinfo = device_get_ivars(dev);
3271 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3272 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3273 		return (0);
3274 #ifdef PCI_IOV
3275 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3276 		struct pcicfg_iov *iov;
3277 
3278 		iov = dinfo->cfg.iov;
3279 		cmd = pci_read_config(iov->iov_pf,
3280 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3281 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3282 	}
3283 #endif
3284 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3285 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3286 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3287 	else
3288 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3289 }
3290 
3291 struct pci_map *
3292 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3293 {
3294 	struct pci_devinfo *dinfo;
3295 	struct pci_map *pm, *prev;
3296 
3297 	dinfo = device_get_ivars(dev);
3298 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3299 	pm->pm_reg = reg;
3300 	pm->pm_value = value;
3301 	pm->pm_size = size;
3302 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3303 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3304 		    reg));
3305 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3306 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3307 			break;
3308 	}
3309 	if (prev != NULL)
3310 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3311 	else
3312 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3313 	return (pm);
3314 }
3315 
3316 static void
3317 pci_restore_bars(device_t dev)
3318 {
3319 	struct pci_devinfo *dinfo;
3320 	struct pci_map *pm;
3321 	int ln2range;
3322 
3323 	dinfo = device_get_ivars(dev);
3324 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3325 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3326 			ln2range = 32;
3327 		else
3328 			ln2range = pci_maprange(pm->pm_value);
3329 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3330 		if (ln2range == 64)
3331 			pci_write_config(dev, pm->pm_reg + 4,
3332 			    pm->pm_value >> 32, 4);
3333 	}
3334 }
3335 
3336 /*
3337  * Add a resource based on a pci map register. Return 1 if the map
3338  * register is a 32bit map register or 2 if it is a 64bit register.
3339  */
3340 static int
3341 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3342     int force, int prefetch)
3343 {
3344 	struct pci_map *pm;
3345 	pci_addr_t base, map, testval;
3346 	pci_addr_t start, end, count;
3347 	int barlen, basezero, flags, maprange, mapsize, type;
3348 	uint16_t cmd;
3349 	struct resource *res;
3350 
3351 	/*
3352 	 * The BAR may already exist if the device is a CardBus card
3353 	 * whose CIS is stored in this BAR.
3354 	 */
3355 	pm = pci_find_bar(dev, reg);
3356 	if (pm != NULL) {
3357 		maprange = pci_maprange(pm->pm_value);
3358 		barlen = maprange == 64 ? 2 : 1;
3359 		return (barlen);
3360 	}
3361 
3362 	pci_read_bar(dev, reg, &map, &testval, NULL);
3363 	if (PCI_BAR_MEM(map)) {
3364 		type = SYS_RES_MEMORY;
3365 		if (map & PCIM_BAR_MEM_PREFETCH)
3366 			prefetch = 1;
3367 	} else
3368 		type = SYS_RES_IOPORT;
3369 	mapsize = pci_mapsize(testval);
3370 	base = pci_mapbase(map);
3371 #ifdef __PCI_BAR_ZERO_VALID
3372 	basezero = 0;
3373 #else
3374 	basezero = base == 0;
3375 #endif
3376 	maprange = pci_maprange(map);
3377 	barlen = maprange == 64 ? 2 : 1;
3378 
3379 	/*
3380 	 * For I/O registers, if bottom bit is set, and the next bit up
3381 	 * isn't clear, we know we have a BAR that doesn't conform to the
3382 	 * spec, so ignore it.  Also, sanity check the size of the data
3383 	 * areas to the type of memory involved.  Memory must be at least
3384 	 * 16 bytes in size, while I/O ranges must be at least 4.
3385 	 */
3386 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3387 		return (barlen);
3388 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3389 	    (type == SYS_RES_IOPORT && mapsize < 2))
3390 		return (barlen);
3391 
3392 	/* Save a record of this BAR. */
3393 	pm = pci_add_bar(dev, reg, map, mapsize);
3394 	if (bootverbose) {
3395 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3396 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3397 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3398 			printf(", port disabled\n");
3399 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3400 			printf(", memory disabled\n");
3401 		else
3402 			printf(", enabled\n");
3403 	}
3404 
3405 	/*
3406 	 * If base is 0, then we have problems if this architecture does
3407 	 * not allow that.  It is best to ignore such entries for the
3408 	 * moment.  These will be allocated later if the driver specifically
3409 	 * requests them.  However, some removable buses look better when
3410 	 * all resources are allocated, so allow '0' to be overridden.
3411 	 *
3412 	 * Similarly treat maps whose values is the same as the test value
3413 	 * read back.  These maps have had all f's written to them by the
3414 	 * BIOS in an attempt to disable the resources.
3415 	 */
3416 	if (!force && (basezero || map == testval))
3417 		return (barlen);
3418 	if ((u_long)base != base) {
3419 		device_printf(bus,
3420 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3421 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3422 		    pci_get_function(dev), reg);
3423 		return (barlen);
3424 	}
3425 
3426 	/*
3427 	 * This code theoretically does the right thing, but has
3428 	 * undesirable side effects in some cases where peripherals
3429 	 * respond oddly to having these bits enabled.  Let the user
3430 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3431 	 * default).
3432 	 */
3433 	if (pci_enable_io_modes) {
3434 		/* Turn on resources that have been left off by a lazy BIOS */
3435 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3436 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3437 			cmd |= PCIM_CMD_PORTEN;
3438 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3439 		}
3440 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3441 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3442 			cmd |= PCIM_CMD_MEMEN;
3443 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3444 		}
3445 	} else {
3446 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3447 			return (barlen);
3448 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3449 			return (barlen);
3450 	}
3451 
3452 	count = (pci_addr_t)1 << mapsize;
3453 	flags = RF_ALIGNMENT_LOG2(mapsize);
3454 	if (prefetch)
3455 		flags |= RF_PREFETCHABLE;
3456 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3457 		start = 0;	/* Let the parent decide. */
3458 		end = ~0;
3459 	} else {
3460 		start = base;
3461 		end = base + count - 1;
3462 	}
3463 	resource_list_add(rl, type, reg, start, end, count);
3464 
3465 	/*
3466 	 * Try to allocate the resource for this BAR from our parent
3467 	 * so that this resource range is already reserved.  The
3468 	 * driver for this device will later inherit this resource in
3469 	 * pci_alloc_resource().
3470 	 */
3471 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3472 	    flags);
3473 	if ((pci_do_realloc_bars
3474 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3475 	    && res == NULL && (start != 0 || end != ~0)) {
3476 		/*
3477 		 * If the allocation fails, try to allocate a resource for
3478 		 * this BAR using any available range.  The firmware felt
3479 		 * it was important enough to assign a resource, so don't
3480 		 * disable decoding if we can help it.
3481 		 */
3482 		resource_list_delete(rl, type, reg);
3483 		resource_list_add(rl, type, reg, 0, ~0, count);
3484 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3485 		    count, flags);
3486 	}
3487 	if (res == NULL) {
3488 		/*
3489 		 * If the allocation fails, delete the resource list entry
3490 		 * and disable decoding for this device.
3491 		 *
3492 		 * If the driver requests this resource in the future,
3493 		 * pci_reserve_map() will try to allocate a fresh
3494 		 * resource range.
3495 		 */
3496 		resource_list_delete(rl, type, reg);
3497 		pci_disable_io(dev, type);
3498 		if (bootverbose)
3499 			device_printf(bus,
3500 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3501 			    pci_get_domain(dev), pci_get_bus(dev),
3502 			    pci_get_slot(dev), pci_get_function(dev), reg);
3503 	} else {
3504 		start = rman_get_start(res);
3505 		pci_write_bar(dev, pm, start);
3506 	}
3507 	return (barlen);
3508 }
3509 
3510 /*
3511  * For ATA devices we need to decide early what addressing mode to use.
3512  * Legacy demands that the primary and secondary ATA ports sits on the
3513  * same addresses that old ISA hardware did. This dictates that we use
3514  * those addresses and ignore the BAR's if we cannot set PCI native
3515  * addressing mode.
3516  */
3517 static void
3518 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3519     uint32_t prefetchmask)
3520 {
3521 	int rid, type, progif;
3522 #if 0
3523 	/* if this device supports PCI native addressing use it */
3524 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3525 	if ((progif & 0x8a) == 0x8a) {
3526 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3527 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3528 			printf("Trying ATA native PCI addressing mode\n");
3529 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3530 		}
3531 	}
3532 #endif
3533 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3534 	type = SYS_RES_IOPORT;
3535 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3536 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3537 		    prefetchmask & (1 << 0));
3538 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3539 		    prefetchmask & (1 << 1));
3540 	} else {
3541 		rid = PCIR_BAR(0);
3542 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3543 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3544 		    0x1f7, 8, 0);
3545 		rid = PCIR_BAR(1);
3546 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3547 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3548 		    0x3f6, 1, 0);
3549 	}
3550 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3551 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3552 		    prefetchmask & (1 << 2));
3553 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3554 		    prefetchmask & (1 << 3));
3555 	} else {
3556 		rid = PCIR_BAR(2);
3557 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3558 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3559 		    0x177, 8, 0);
3560 		rid = PCIR_BAR(3);
3561 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3562 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3563 		    0x376, 1, 0);
3564 	}
3565 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3566 	    prefetchmask & (1 << 4));
3567 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3568 	    prefetchmask & (1 << 5));
3569 }
3570 
3571 static void
3572 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3573 {
3574 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3575 	pcicfgregs *cfg = &dinfo->cfg;
3576 	char tunable_name[64];
3577 	int irq;
3578 
3579 	/* Has to have an intpin to have an interrupt. */
3580 	if (cfg->intpin == 0)
3581 		return;
3582 
3583 	/* Let the user override the IRQ with a tunable. */
3584 	irq = PCI_INVALID_IRQ;
3585 	snprintf(tunable_name, sizeof(tunable_name),
3586 	    "hw.pci%d.%d.%d.INT%c.irq",
3587 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3588 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3589 		irq = PCI_INVALID_IRQ;
3590 
3591 	/*
3592 	 * If we didn't get an IRQ via the tunable, then we either use the
3593 	 * IRQ value in the intline register or we ask the bus to route an
3594 	 * interrupt for us.  If force_route is true, then we only use the
3595 	 * value in the intline register if the bus was unable to assign an
3596 	 * IRQ.
3597 	 */
3598 	if (!PCI_INTERRUPT_VALID(irq)) {
3599 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3600 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3601 		if (!PCI_INTERRUPT_VALID(irq))
3602 			irq = cfg->intline;
3603 	}
3604 
3605 	/* If after all that we don't have an IRQ, just bail. */
3606 	if (!PCI_INTERRUPT_VALID(irq))
3607 		return;
3608 
3609 	/* Update the config register if it changed. */
3610 	if (irq != cfg->intline) {
3611 		cfg->intline = irq;
3612 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3613 	}
3614 
3615 	/* Add this IRQ as rid 0 interrupt resource. */
3616 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3617 }
3618 
3619 /* Perform early OHCI takeover from SMM. */
3620 static void
3621 ohci_early_takeover(device_t self)
3622 {
3623 	struct resource *res;
3624 	uint32_t ctl;
3625 	int rid;
3626 	int i;
3627 
3628 	rid = PCIR_BAR(0);
3629 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3630 	if (res == NULL)
3631 		return;
3632 
3633 	ctl = bus_read_4(res, OHCI_CONTROL);
3634 	if (ctl & OHCI_IR) {
3635 		if (bootverbose)
3636 			printf("ohci early: "
3637 			    "SMM active, request owner change\n");
3638 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3639 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3640 			DELAY(1000);
3641 			ctl = bus_read_4(res, OHCI_CONTROL);
3642 		}
3643 		if (ctl & OHCI_IR) {
3644 			if (bootverbose)
3645 				printf("ohci early: "
3646 				    "SMM does not respond, resetting\n");
3647 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3648 		}
3649 		/* Disable interrupts */
3650 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3651 	}
3652 
3653 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3654 }
3655 
3656 /* Perform early UHCI takeover from SMM. */
3657 static void
3658 uhci_early_takeover(device_t self)
3659 {
3660 	struct resource *res;
3661 	int rid;
3662 
3663 	/*
3664 	 * Set the PIRQD enable bit and switch off all the others. We don't
3665 	 * want legacy support to interfere with us XXX Does this also mean
3666 	 * that the BIOS won't touch the keyboard anymore if it is connected
3667 	 * to the ports of the root hub?
3668 	 */
3669 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3670 
3671 	/* Disable interrupts */
3672 	rid = PCI_UHCI_BASE_REG;
3673 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3674 	if (res != NULL) {
3675 		bus_write_2(res, UHCI_INTR, 0);
3676 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3677 	}
3678 }
3679 
3680 /* Perform early EHCI takeover from SMM. */
3681 static void
3682 ehci_early_takeover(device_t self)
3683 {
3684 	struct resource *res;
3685 	uint32_t cparams;
3686 	uint32_t eec;
3687 	uint8_t eecp;
3688 	uint8_t bios_sem;
3689 	uint8_t offs;
3690 	int rid;
3691 	int i;
3692 
3693 	rid = PCIR_BAR(0);
3694 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3695 	if (res == NULL)
3696 		return;
3697 
3698 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3699 
3700 	/* Synchronise with the BIOS if it owns the controller. */
3701 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3702 	    eecp = EHCI_EECP_NEXT(eec)) {
3703 		eec = pci_read_config(self, eecp, 4);
3704 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3705 			continue;
3706 		}
3707 		bios_sem = pci_read_config(self, eecp +
3708 		    EHCI_LEGSUP_BIOS_SEM, 1);
3709 		if (bios_sem == 0) {
3710 			continue;
3711 		}
3712 		if (bootverbose)
3713 			printf("ehci early: "
3714 			    "SMM active, request owner change\n");
3715 
3716 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3717 
3718 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3719 			DELAY(1000);
3720 			bios_sem = pci_read_config(self, eecp +
3721 			    EHCI_LEGSUP_BIOS_SEM, 1);
3722 		}
3723 
3724 		if (bios_sem != 0) {
3725 			if (bootverbose)
3726 				printf("ehci early: "
3727 				    "SMM does not respond\n");
3728 		}
3729 		/* Disable interrupts */
3730 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3731 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3732 	}
3733 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3734 }
3735 
3736 /* Perform early XHCI takeover from SMM. */
3737 static void
3738 xhci_early_takeover(device_t self)
3739 {
3740 	struct resource *res;
3741 	uint32_t cparams;
3742 	uint32_t eec;
3743 	uint8_t eecp;
3744 	uint8_t bios_sem;
3745 	uint8_t offs;
3746 	int rid;
3747 	int i;
3748 
3749 	rid = PCIR_BAR(0);
3750 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3751 	if (res == NULL)
3752 		return;
3753 
3754 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3755 
3756 	eec = -1;
3757 
3758 	/* Synchronise with the BIOS if it owns the controller. */
3759 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3760 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3761 		eec = bus_read_4(res, eecp);
3762 
3763 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3764 			continue;
3765 
3766 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3767 		if (bios_sem == 0)
3768 			continue;
3769 
3770 		if (bootverbose)
3771 			printf("xhci early: "
3772 			    "SMM active, request owner change\n");
3773 
3774 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3775 
3776 		/* wait a maximum of 5 second */
3777 
3778 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3779 			DELAY(1000);
3780 			bios_sem = bus_read_1(res, eecp +
3781 			    XHCI_XECP_BIOS_SEM);
3782 		}
3783 
3784 		if (bios_sem != 0) {
3785 			if (bootverbose)
3786 				printf("xhci early: "
3787 				    "SMM does not respond\n");
3788 		}
3789 
3790 		/* Disable interrupts */
3791 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3792 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3793 		bus_read_4(res, offs + XHCI_USBSTS);
3794 	}
3795 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3796 }
3797 
3798 static void
3799 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3800     struct resource_list *rl)
3801 {
3802 	struct resource *res;
3803 	char *cp;
3804 	rman_res_t start, end, count;
3805 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3806 
3807 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3808 	case PCIM_HDRTYPE_BRIDGE:
3809 		sec_reg = PCIR_SECBUS_1;
3810 		sub_reg = PCIR_SUBBUS_1;
3811 		break;
3812 	case PCIM_HDRTYPE_CARDBUS:
3813 		sec_reg = PCIR_SECBUS_2;
3814 		sub_reg = PCIR_SUBBUS_2;
3815 		break;
3816 	default:
3817 		return;
3818 	}
3819 
3820 	/*
3821 	 * If the existing bus range is valid, attempt to reserve it
3822 	 * from our parent.  If this fails for any reason, clear the
3823 	 * secbus and subbus registers.
3824 	 *
3825 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3826 	 * This would at least preserve the existing sec_bus if it is
3827 	 * valid.
3828 	 */
3829 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3830 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3831 
3832 	/* Quirk handling. */
3833 	switch (pci_get_devid(dev)) {
3834 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3835 		sup_bus = pci_read_config(dev, 0x41, 1);
3836 		if (sup_bus != 0xff) {
3837 			sec_bus = sup_bus + 1;
3838 			sub_bus = sup_bus + 1;
3839 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3840 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3841 		}
3842 		break;
3843 
3844 	case 0x00dd10de:
3845 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3846 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3847 			break;
3848 		if (strncmp(cp, "Compal", 6) != 0) {
3849 			freeenv(cp);
3850 			break;
3851 		}
3852 		freeenv(cp);
3853 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3854 			break;
3855 		if (strncmp(cp, "08A0", 4) != 0) {
3856 			freeenv(cp);
3857 			break;
3858 		}
3859 		freeenv(cp);
3860 		if (sub_bus < 0xa) {
3861 			sub_bus = 0xa;
3862 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3863 		}
3864 		break;
3865 	}
3866 
3867 	if (bootverbose)
3868 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3869 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3870 		start = sec_bus;
3871 		end = sub_bus;
3872 		count = end - start + 1;
3873 
3874 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3875 
3876 		/*
3877 		 * If requested, clear secondary bus registers in
3878 		 * bridge devices to force a complete renumbering
3879 		 * rather than reserving the existing range.  However,
3880 		 * preserve the existing size.
3881 		 */
3882 		if (pci_clear_buses)
3883 			goto clear;
3884 
3885 		rid = 0;
3886 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3887 		    start, end, count, 0);
3888 		if (res != NULL)
3889 			return;
3890 
3891 		if (bootverbose)
3892 			device_printf(bus,
3893 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3894 			    pci_get_domain(dev), pci_get_bus(dev),
3895 			    pci_get_slot(dev), pci_get_function(dev));
3896 	}
3897 
3898 clear:
3899 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3900 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3901 }
3902 
3903 static struct resource *
3904 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3905     rman_res_t end, rman_res_t count, u_int flags)
3906 {
3907 	struct pci_devinfo *dinfo;
3908 	pcicfgregs *cfg;
3909 	struct resource_list *rl;
3910 	struct resource *res;
3911 	int sec_reg, sub_reg;
3912 
3913 	dinfo = device_get_ivars(child);
3914 	cfg = &dinfo->cfg;
3915 	rl = &dinfo->resources;
3916 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3917 	case PCIM_HDRTYPE_BRIDGE:
3918 		sec_reg = PCIR_SECBUS_1;
3919 		sub_reg = PCIR_SUBBUS_1;
3920 		break;
3921 	case PCIM_HDRTYPE_CARDBUS:
3922 		sec_reg = PCIR_SECBUS_2;
3923 		sub_reg = PCIR_SUBBUS_2;
3924 		break;
3925 	default:
3926 		return (NULL);
3927 	}
3928 
3929 	if (*rid != 0)
3930 		return (NULL);
3931 
3932 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3933 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3934 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3935 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3936 		    start, end, count, flags & ~RF_ACTIVE);
3937 		if (res == NULL) {
3938 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3939 			device_printf(child, "allocating %ju bus%s failed\n",
3940 			    count, count == 1 ? "" : "es");
3941 			return (NULL);
3942 		}
3943 		if (bootverbose)
3944 			device_printf(child,
3945 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3946 			    count == 1 ? "" : "es", rman_get_start(res));
3947 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3948 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3949 	}
3950 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3951 	    end, count, flags));
3952 }
3953 
3954 static int
3955 pci_ea_bei_to_rid(device_t dev, int bei)
3956 {
3957 #ifdef PCI_IOV
3958 	struct pci_devinfo *dinfo;
3959 	int iov_pos;
3960 	struct pcicfg_iov *iov;
3961 
3962 	dinfo = device_get_ivars(dev);
3963 	iov = dinfo->cfg.iov;
3964 	if (iov != NULL)
3965 		iov_pos = iov->iov_pos;
3966 	else
3967 		iov_pos = 0;
3968 #endif
3969 
3970 	/* Check if matches BAR */
3971 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3972 	    (bei <= PCIM_EA_BEI_BAR_5))
3973 		return (PCIR_BAR(bei));
3974 
3975 	/* Check ROM */
3976 	if (bei == PCIM_EA_BEI_ROM)
3977 		return (PCIR_BIOS);
3978 
3979 #ifdef PCI_IOV
3980 	/* Check if matches VF_BAR */
3981 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3982 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3983 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3984 		    iov_pos);
3985 #endif
3986 
3987 	return (-1);
3988 }
3989 
3990 int
3991 pci_ea_is_enabled(device_t dev, int rid)
3992 {
3993 	struct pci_ea_entry *ea;
3994 	struct pci_devinfo *dinfo;
3995 
3996 	dinfo = device_get_ivars(dev);
3997 
3998 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3999 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
4000 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
4001 	}
4002 
4003 	return (0);
4004 }
4005 
4006 void
4007 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
4008 {
4009 	struct pci_ea_entry *ea;
4010 	struct pci_devinfo *dinfo;
4011 	pci_addr_t start, end, count;
4012 	struct resource_list *rl;
4013 	int type, flags, rid;
4014 	struct resource *res;
4015 	uint32_t tmp;
4016 #ifdef PCI_IOV
4017 	struct pcicfg_iov *iov;
4018 #endif
4019 
4020 	dinfo = device_get_ivars(dev);
4021 	rl = &dinfo->resources;
4022 	flags = 0;
4023 
4024 #ifdef PCI_IOV
4025 	iov = dinfo->cfg.iov;
4026 #endif
4027 
4028 	if (dinfo->cfg.ea.ea_location == 0)
4029 		return;
4030 
4031 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4032 		/*
4033 		 * TODO: Ignore EA-BAR if is not enabled.
4034 		 *   Currently the EA implementation supports
4035 		 *   only situation, where EA structure contains
4036 		 *   predefined entries. In case they are not enabled
4037 		 *   leave them unallocated and proceed with
4038 		 *   a legacy-BAR mechanism.
4039 		 */
4040 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
4041 			continue;
4042 
4043 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
4044 		case PCIM_EA_P_MEM_PREFETCH:
4045 		case PCIM_EA_P_VF_MEM_PREFETCH:
4046 			flags = RF_PREFETCHABLE;
4047 			/* FALLTHROUGH */
4048 		case PCIM_EA_P_VF_MEM:
4049 		case PCIM_EA_P_MEM:
4050 			type = SYS_RES_MEMORY;
4051 			break;
4052 		case PCIM_EA_P_IO:
4053 			type = SYS_RES_IOPORT;
4054 			break;
4055 		default:
4056 			continue;
4057 		}
4058 
4059 		if (alloc_iov != 0) {
4060 #ifdef PCI_IOV
4061 			/* Allocating IOV, confirm BEI matches */
4062 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
4063 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
4064 				continue;
4065 #else
4066 			continue;
4067 #endif
4068 		} else {
4069 			/* Allocating BAR, confirm BEI matches */
4070 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
4071 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
4072 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
4073 				continue;
4074 		}
4075 
4076 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
4077 		if (rid < 0)
4078 			continue;
4079 
4080 		/* Skip resources already allocated by EA */
4081 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
4082 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
4083 			continue;
4084 
4085 		start = ea->eae_base;
4086 		count = ea->eae_max_offset + 1;
4087 #ifdef PCI_IOV
4088 		if (iov != NULL)
4089 			count = count * iov->iov_num_vfs;
4090 #endif
4091 		end = start + count - 1;
4092 		if (count == 0)
4093 			continue;
4094 
4095 		resource_list_add(rl, type, rid, start, end, count);
4096 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4097 		    flags);
4098 		if (res == NULL) {
4099 			resource_list_delete(rl, type, rid);
4100 
4101 			/*
4102 			 * Failed to allocate using EA, disable entry.
4103 			 * Another attempt to allocation will be performed
4104 			 * further, but this time using legacy BAR registers
4105 			 */
4106 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4107 			tmp &= ~PCIM_EA_ENABLE;
4108 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4109 
4110 			/*
4111 			 * Disabling entry might fail in case it is hardwired.
4112 			 * Read flags again to match current status.
4113 			 */
4114 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4115 
4116 			continue;
4117 		}
4118 
4119 		/* As per specification, fill BAR with zeros */
4120 		pci_write_config(dev, rid, 0, 4);
4121 	}
4122 }
4123 
4124 void
4125 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4126 {
4127 	struct pci_devinfo *dinfo;
4128 	pcicfgregs *cfg;
4129 	struct resource_list *rl;
4130 	const struct pci_quirk *q;
4131 	uint32_t devid;
4132 	int i;
4133 
4134 	dinfo = device_get_ivars(dev);
4135 	cfg = &dinfo->cfg;
4136 	rl = &dinfo->resources;
4137 	devid = (cfg->device << 16) | cfg->vendor;
4138 
4139 	/* Allocate resources using Enhanced Allocation */
4140 	pci_add_resources_ea(bus, dev, 0);
4141 
4142 	/* ATA devices needs special map treatment */
4143 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4144 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4145 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4146 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4147 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4148 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4149 	else
4150 		for (i = 0; i < cfg->nummaps;) {
4151 			/* Skip resources already managed by EA */
4152 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4153 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4154 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4155 				i++;
4156 				continue;
4157 			}
4158 
4159 			/*
4160 			 * Skip quirked resources.
4161 			 */
4162 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4163 				if (q->devid == devid &&
4164 				    q->type == PCI_QUIRK_UNMAP_REG &&
4165 				    q->arg1 == PCIR_BAR(i))
4166 					break;
4167 			if (q->devid != 0) {
4168 				i++;
4169 				continue;
4170 			}
4171 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4172 			    prefetchmask & (1 << i));
4173 		}
4174 
4175 	/*
4176 	 * Add additional, quirked resources.
4177 	 */
4178 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4179 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4180 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4181 
4182 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline) &&
4183 	    pci_intx_reroute) {
4184 		/*
4185 		 * Try to re-route interrupts. Sometimes the BIOS or
4186 		 * firmware may leave bogus values in these registers.
4187 		 * If the re-route fails, then just stick with what we
4188 		 * have.
4189 		 */
4190 		pci_assign_interrupt(bus, dev, 1);
4191 	}
4192 
4193 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4194 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4195 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4196 			xhci_early_takeover(dev);
4197 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4198 			ehci_early_takeover(dev);
4199 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4200 			ohci_early_takeover(dev);
4201 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4202 			uhci_early_takeover(dev);
4203 	}
4204 
4205 	/*
4206 	 * Reserve resources for secondary bus ranges behind bridge
4207 	 * devices.
4208 	 */
4209 	pci_reserve_secbus(bus, dev, cfg, rl);
4210 }
4211 
4212 static struct pci_devinfo *
4213 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4214     int slot, int func)
4215 {
4216 	struct pci_devinfo *dinfo;
4217 
4218 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4219 	if (dinfo != NULL)
4220 		pci_add_child(dev, dinfo);
4221 
4222 	return (dinfo);
4223 }
4224 
4225 void
4226 pci_add_children(device_t dev, int domain, int busno)
4227 {
4228 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4229 	device_t pcib = device_get_parent(dev);
4230 	struct pci_devinfo *dinfo;
4231 	int maxslots;
4232 	int s, f, pcifunchigh;
4233 	uint8_t hdrtype;
4234 	int first_func;
4235 
4236 	/*
4237 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4238 	 * enable ARI.  We must enable ARI before detecting the rest of the
4239 	 * functions on this bus as ARI changes the set of slots and functions
4240 	 * that are legal on this bus.
4241 	 */
4242 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4243 	if (dinfo != NULL && pci_enable_ari)
4244 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4245 
4246 	/*
4247 	 * Start looking for new devices on slot 0 at function 1 because we
4248 	 * just identified the device at slot 0, function 0.
4249 	 */
4250 	first_func = 1;
4251 
4252 	maxslots = PCIB_MAXSLOTS(pcib);
4253 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4254 		pcifunchigh = 0;
4255 		f = 0;
4256 		DELAY(1);
4257 
4258 		/* If function 0 is not present, skip to the next slot. */
4259 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4260 			continue;
4261 		hdrtype = REG(PCIR_HDRTYPE, 1);
4262 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4263 			continue;
4264 		if (hdrtype & PCIM_MFDEV)
4265 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4266 		for (f = first_func; f <= pcifunchigh; f++)
4267 			pci_identify_function(pcib, dev, domain, busno, s, f);
4268 	}
4269 #undef REG
4270 }
4271 
4272 int
4273 pci_rescan_method(device_t dev)
4274 {
4275 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4276 	device_t pcib = device_get_parent(dev);
4277 	device_t child, *devlist, *unchanged;
4278 	int devcount, error, i, j, maxslots, oldcount;
4279 	int busno, domain, s, f, pcifunchigh;
4280 	uint8_t hdrtype;
4281 
4282 	/* No need to check for ARI on a rescan. */
4283 	error = device_get_children(dev, &devlist, &devcount);
4284 	if (error)
4285 		return (error);
4286 	if (devcount != 0) {
4287 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4288 		    M_NOWAIT | M_ZERO);
4289 		if (unchanged == NULL) {
4290 			free(devlist, M_TEMP);
4291 			return (ENOMEM);
4292 		}
4293 	} else
4294 		unchanged = NULL;
4295 
4296 	domain = pcib_get_domain(dev);
4297 	busno = pcib_get_bus(dev);
4298 	maxslots = PCIB_MAXSLOTS(pcib);
4299 	for (s = 0; s <= maxslots; s++) {
4300 		/* If function 0 is not present, skip to the next slot. */
4301 		f = 0;
4302 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4303 			continue;
4304 		pcifunchigh = 0;
4305 		hdrtype = REG(PCIR_HDRTYPE, 1);
4306 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4307 			continue;
4308 		if (hdrtype & PCIM_MFDEV)
4309 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4310 		for (f = 0; f <= pcifunchigh; f++) {
4311 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4312 				continue;
4313 
4314 			/*
4315 			 * Found a valid function.  Check if a
4316 			 * device_t for this device already exists.
4317 			 */
4318 			for (i = 0; i < devcount; i++) {
4319 				child = devlist[i];
4320 				if (child == NULL)
4321 					continue;
4322 				if (pci_get_slot(child) == s &&
4323 				    pci_get_function(child) == f) {
4324 					unchanged[i] = child;
4325 					goto next_func;
4326 				}
4327 			}
4328 
4329 			pci_identify_function(pcib, dev, domain, busno, s, f);
4330 		next_func:;
4331 		}
4332 	}
4333 
4334 	/* Remove devices that are no longer present. */
4335 	for (i = 0; i < devcount; i++) {
4336 		if (unchanged[i] != NULL)
4337 			continue;
4338 		device_delete_child(dev, devlist[i]);
4339 	}
4340 
4341 	free(devlist, M_TEMP);
4342 	oldcount = devcount;
4343 
4344 	/* Try to attach the devices just added. */
4345 	error = device_get_children(dev, &devlist, &devcount);
4346 	if (error) {
4347 		free(unchanged, M_TEMP);
4348 		return (error);
4349 	}
4350 
4351 	for (i = 0; i < devcount; i++) {
4352 		for (j = 0; j < oldcount; j++) {
4353 			if (devlist[i] == unchanged[j])
4354 				goto next_device;
4355 		}
4356 
4357 		device_probe_and_attach(devlist[i]);
4358 	next_device:;
4359 	}
4360 
4361 	free(unchanged, M_TEMP);
4362 	free(devlist, M_TEMP);
4363 	return (0);
4364 #undef REG
4365 }
4366 
4367 #ifdef PCI_IOV
4368 device_t
4369 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4370     uint16_t did)
4371 {
4372 	struct pci_devinfo *vf_dinfo;
4373 	device_t pcib;
4374 	int busno, slot, func;
4375 
4376 	pcib = device_get_parent(bus);
4377 
4378 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4379 
4380 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4381 	    slot, func, vid, did);
4382 
4383 	vf_dinfo->cfg.flags |= PCICFG_VF;
4384 	pci_add_child(bus, vf_dinfo);
4385 
4386 	return (vf_dinfo->cfg.dev);
4387 }
4388 
4389 device_t
4390 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4391     uint16_t vid, uint16_t did)
4392 {
4393 
4394 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4395 }
4396 #endif
4397 
4398 /*
4399  * For PCIe device set Max_Payload_Size to match PCIe root's.
4400  */
4401 static void
4402 pcie_setup_mps(device_t dev)
4403 {
4404 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4405 	device_t root;
4406 	uint16_t rmps, mmps, mps;
4407 
4408 	if (dinfo->cfg.pcie.pcie_location == 0)
4409 		return;
4410 	root = pci_find_pcie_root_port(dev);
4411 	if (root == NULL)
4412 		return;
4413 	/* Check whether the MPS is already configured. */
4414 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4415 	    PCIEM_CTL_MAX_PAYLOAD;
4416 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4417 	    PCIEM_CTL_MAX_PAYLOAD;
4418 	if (mps == rmps)
4419 		return;
4420 	/* Check whether the device is capable of the root's MPS. */
4421 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4422 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4423 	if (rmps > mmps) {
4424 		/*
4425 		 * The device is unable to handle root's MPS.  Limit root.
4426 		 * XXX: We should traverse through all the tree, applying
4427 		 * it to all the devices.
4428 		 */
4429 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4430 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4431 	} else {
4432 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4433 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4434 	}
4435 }
4436 
4437 static void
4438 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4439 {
4440 	int aer;
4441 	uint32_t r;
4442 	uint16_t r2;
4443 
4444 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4445 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4446 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4447 		    PCIER_ROOT_CTL, 2);
4448 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4449 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4450 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4451 		    PCIER_ROOT_CTL, r2, 2);
4452 	}
4453 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4454 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4455 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4456 		if (r != 0 && bootverbose) {
4457 			pci_printf(&dinfo->cfg,
4458 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4459 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4460 			    4));
4461 		}
4462 
4463 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4464 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4465 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4466 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4467 		    PCIM_AER_UC_POISONED_TLP |
4468 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4469 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4470 		    PCIM_AER_UC_COMPLETER_ABORT |
4471 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4472 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4473 		    PCIM_AER_UC_MALFORMED_TLP |
4474 		    PCIM_AER_UC_ECRC_ERROR |
4475 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4476 		    PCIM_AER_UC_ACS_VIOLATION |
4477 		    PCIM_AER_UC_INTERNAL_ERROR |
4478 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4479 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4480 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4481 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4482 
4483 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4484 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4485 		if (r != 0 && bootverbose) {
4486 			pci_printf(&dinfo->cfg,
4487 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4488 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4489 			    4));
4490 		}
4491 
4492 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4493 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4494 		    PCIM_AER_COR_BAD_TLP |
4495 		    PCIM_AER_COR_BAD_DLLP |
4496 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4497 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4498 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4499 		    PCIM_AER_COR_INTERNAL_ERROR |
4500 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4501 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4502 
4503 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4504 		    PCIER_DEVICE_CTL, 2);
4505 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4506 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4507 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4508 		    PCIER_DEVICE_CTL, r, 2);
4509 	}
4510 }
4511 
4512 void
4513 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4514 {
4515 	device_t dev;
4516 
4517 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4518 	device_set_ivars(dev, dinfo);
4519 	resource_list_init(&dinfo->resources);
4520 	pci_cfg_save(dev, dinfo, 0);
4521 	pci_cfg_restore(dev, dinfo);
4522 	pci_clear_pme(dev);
4523 	pci_print_verbose(dinfo);
4524 	pci_add_resources(bus, dev, 0, 0);
4525 	if (pci_enable_mps_tune)
4526 		pcie_setup_mps(dev);
4527 	pci_child_added(dinfo->cfg.dev);
4528 
4529 	if (pci_clear_aer_on_attach)
4530 		pci_add_child_clear_aer(dev, dinfo);
4531 
4532 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4533 }
4534 
4535 void
4536 pci_child_added_method(device_t dev, device_t child)
4537 {
4538 
4539 }
4540 
4541 static int
4542 pci_probe(device_t dev)
4543 {
4544 
4545 	device_set_desc(dev, "PCI bus");
4546 
4547 	/* Allow other subclasses to override this driver. */
4548 	return (BUS_PROBE_GENERIC);
4549 }
4550 
4551 int
4552 pci_attach_common(device_t dev)
4553 {
4554 	struct pci_softc *sc;
4555 	int busno, domain;
4556 	int rid;
4557 
4558 	sc = device_get_softc(dev);
4559 	domain = pcib_get_domain(dev);
4560 	busno = pcib_get_bus(dev);
4561 	rid = 0;
4562 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4563 	    1, 0);
4564 	if (sc->sc_bus == NULL) {
4565 		device_printf(dev, "failed to allocate bus number\n");
4566 		return (ENXIO);
4567 	}
4568 	if (bootverbose)
4569 		device_printf(dev, "domain=%d, physical bus=%d\n",
4570 		    domain, busno);
4571 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4572 	return (0);
4573 }
4574 
4575 int
4576 pci_attach(device_t dev)
4577 {
4578 	int busno, domain, error;
4579 
4580 	error = pci_attach_common(dev);
4581 	if (error)
4582 		return (error);
4583 
4584 	/*
4585 	 * Since there can be multiple independently numbered PCI
4586 	 * buses on systems with multiple PCI domains, we can't use
4587 	 * the unit number to decide which bus we are probing. We ask
4588 	 * the parent pcib what our domain and bus numbers are.
4589 	 */
4590 	domain = pcib_get_domain(dev);
4591 	busno = pcib_get_bus(dev);
4592 	pci_add_children(dev, domain, busno);
4593 	bus_attach_children(dev);
4594 	return (0);
4595 }
4596 
4597 int
4598 pci_detach(device_t dev)
4599 {
4600 	struct pci_softc *sc;
4601 	int error;
4602 
4603 	error = bus_generic_detach(dev);
4604 	if (error)
4605 		return (error);
4606 	sc = device_get_softc(dev);
4607 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4608 	return (error);
4609 }
4610 
4611 static void
4612 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4613 {
4614 	int line, unit;
4615 	const char *at;
4616 	char me1[24], me2[32];
4617 	uint8_t b, s, f;
4618 	uint32_t d;
4619 	device_location_cache_t *cache;
4620 
4621 	d = pci_get_domain(child);
4622 	b = pci_get_bus(child);
4623 	s = pci_get_slot(child);
4624 	f = pci_get_function(child);
4625 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4626 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4627 	line = 0;
4628 	cache = dev_wired_cache_init();
4629 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4630 		resource_string_value(name, unit, "at", &at);
4631 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4632 			*unitp = unit;
4633 			break;
4634 		}
4635 		if (dev_wired_cache_match(cache, child, at)) {
4636 			*unitp = unit;
4637 			break;
4638 		}
4639 	}
4640 	dev_wired_cache_fini(cache);
4641 }
4642 
4643 static void
4644 pci_set_power_child(device_t dev, device_t child, int state)
4645 {
4646 	device_t pcib;
4647 	int dstate;
4648 
4649 	/*
4650 	 * Set the device to the given state.  If the firmware suggests
4651 	 * a different power state, use it instead.  If power management
4652 	 * is not present, the firmware is responsible for managing
4653 	 * device power.  Skip children who aren't attached since they
4654 	 * are handled separately.
4655 	 */
4656 	pcib = device_get_parent(dev);
4657 	dstate = state;
4658 	if (device_is_attached(child) &&
4659 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4660 		pci_set_powerstate(child, dstate);
4661 }
4662 
4663 int
4664 pci_suspend_child(device_t dev, device_t child)
4665 {
4666 	struct pci_devinfo *dinfo;
4667 	struct resource_list_entry *rle;
4668 	int error;
4669 
4670 	dinfo = device_get_ivars(child);
4671 
4672 	/*
4673 	 * Save the PCI configuration space for the child and set the
4674 	 * device in the appropriate power state for this sleep state.
4675 	 */
4676 	pci_cfg_save(child, dinfo, 0);
4677 
4678 	/* Suspend devices before potentially powering them down. */
4679 	error = bus_generic_suspend_child(dev, child);
4680 
4681 	if (error)
4682 		return (error);
4683 
4684 	if (pci_do_power_suspend) {
4685 		/*
4686 		 * Make sure this device's interrupt handler is not invoked
4687 		 * in the case the device uses a shared interrupt that can
4688 		 * be raised by some other device.
4689 		 * This is applicable only to regular (legacy) PCI interrupts
4690 		 * as MSI/MSI-X interrupts are never shared.
4691 		 */
4692 		rle = resource_list_find(&dinfo->resources,
4693 		    SYS_RES_IRQ, 0);
4694 		if (rle != NULL && rle->res != NULL)
4695 			(void)bus_suspend_intr(child, rle->res);
4696 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4697 	}
4698 
4699 	return (0);
4700 }
4701 
4702 int
4703 pci_resume_child(device_t dev, device_t child)
4704 {
4705 	struct pci_devinfo *dinfo;
4706 	struct resource_list_entry *rle;
4707 
4708 	if (pci_do_power_resume)
4709 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4710 
4711 	dinfo = device_get_ivars(child);
4712 	pci_cfg_restore(child, dinfo);
4713 	pci_clear_pme(child);
4714 	if (!device_is_attached(child))
4715 		pci_cfg_save(child, dinfo, 1);
4716 
4717 	bus_generic_resume_child(dev, child);
4718 
4719 	/*
4720 	 * Allow interrupts only after fully resuming the driver and hardware.
4721 	 */
4722 	if (pci_do_power_suspend) {
4723 		/* See pci_suspend_child for details. */
4724 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4725 		if (rle != NULL && rle->res != NULL)
4726 			(void)bus_resume_intr(child, rle->res);
4727 	}
4728 
4729 	return (0);
4730 }
4731 
4732 int
4733 pci_resume(device_t dev)
4734 {
4735 	device_t child, *devlist;
4736 	int error, i, numdevs;
4737 
4738 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4739 		return (error);
4740 
4741 	/*
4742 	 * Resume critical devices first, then everything else later.
4743 	 */
4744 	for (i = 0; i < numdevs; i++) {
4745 		child = devlist[i];
4746 		switch (pci_get_class(child)) {
4747 		case PCIC_DISPLAY:
4748 		case PCIC_MEMORY:
4749 		case PCIC_BRIDGE:
4750 		case PCIC_BASEPERIPH:
4751 			BUS_RESUME_CHILD(dev, child);
4752 			break;
4753 		}
4754 	}
4755 	for (i = 0; i < numdevs; i++) {
4756 		child = devlist[i];
4757 		switch (pci_get_class(child)) {
4758 		case PCIC_DISPLAY:
4759 		case PCIC_MEMORY:
4760 		case PCIC_BRIDGE:
4761 		case PCIC_BASEPERIPH:
4762 			break;
4763 		default:
4764 			BUS_RESUME_CHILD(dev, child);
4765 		}
4766 	}
4767 	free(devlist, M_TEMP);
4768 	return (0);
4769 }
4770 
4771 static void
4772 pci_load_vendor_data(void)
4773 {
4774 	caddr_t data;
4775 	void *ptr;
4776 	size_t sz;
4777 
4778 	data = preload_search_by_type("pci_vendor_data");
4779 	if (data != NULL) {
4780 		ptr = preload_fetch_addr(data);
4781 		sz = preload_fetch_size(data);
4782 		if (ptr != NULL && sz != 0) {
4783 			pci_vendordata = ptr;
4784 			pci_vendordata_size = sz;
4785 			/* terminate the database */
4786 			pci_vendordata[pci_vendordata_size] = '\n';
4787 		}
4788 	}
4789 }
4790 
4791 void
4792 pci_driver_added(device_t dev, driver_t *driver)
4793 {
4794 	int numdevs;
4795 	device_t *devlist;
4796 	device_t child;
4797 	struct pci_devinfo *dinfo;
4798 	int i;
4799 
4800 	if (bootverbose)
4801 		device_printf(dev, "driver added\n");
4802 	DEVICE_IDENTIFY(driver, dev);
4803 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4804 		return;
4805 	for (i = 0; i < numdevs; i++) {
4806 		child = devlist[i];
4807 		if (device_get_state(child) != DS_NOTPRESENT)
4808 			continue;
4809 		dinfo = device_get_ivars(child);
4810 		pci_print_verbose(dinfo);
4811 		if (bootverbose)
4812 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4813 		pci_cfg_restore(child, dinfo);
4814 		if (device_probe_and_attach(child) != 0)
4815 			pci_child_detached(dev, child);
4816 	}
4817 	free(devlist, M_TEMP);
4818 }
4819 
4820 int
4821 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4822     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4823 {
4824 	struct pci_devinfo *dinfo;
4825 	struct msix_table_entry *mte;
4826 	struct msix_vector *mv;
4827 	uint64_t addr;
4828 	uint32_t data;
4829 	void *cookie;
4830 	int error, rid;
4831 
4832 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4833 	    arg, &cookie);
4834 	if (error)
4835 		return (error);
4836 
4837 	/* If this is not a direct child, just bail out. */
4838 	if (device_get_parent(child) != dev) {
4839 		*cookiep = cookie;
4840 		return(0);
4841 	}
4842 
4843 	rid = rman_get_rid(irq);
4844 	if (rid == 0) {
4845 		/* Make sure that INTx is enabled */
4846 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4847 	} else {
4848 		/*
4849 		 * Check to see if the interrupt is MSI or MSI-X.
4850 		 * Ask our parent to map the MSI and give
4851 		 * us the address and data register values.
4852 		 * If we fail for some reason, teardown the
4853 		 * interrupt handler.
4854 		 */
4855 		dinfo = device_get_ivars(child);
4856 		if (dinfo->cfg.msi.msi_alloc > 0) {
4857 			if (dinfo->cfg.msi.msi_addr == 0) {
4858 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4859 			    ("MSI has handlers, but vectors not mapped"));
4860 				error = PCIB_MAP_MSI(device_get_parent(dev),
4861 				    child, rman_get_start(irq), &addr, &data);
4862 				if (error)
4863 					goto bad;
4864 				dinfo->cfg.msi.msi_addr = addr;
4865 				dinfo->cfg.msi.msi_data = data;
4866 			}
4867 			if (dinfo->cfg.msi.msi_handlers == 0)
4868 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4869 				    dinfo->cfg.msi.msi_data);
4870 			dinfo->cfg.msi.msi_handlers++;
4871 		} else {
4872 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4873 			    ("No MSI or MSI-X interrupts allocated"));
4874 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4875 			    ("MSI-X index too high"));
4876 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4877 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4878 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4879 			KASSERT(mv->mv_irq == rman_get_start(irq),
4880 			    ("IRQ mismatch"));
4881 			if (mv->mv_address == 0) {
4882 				KASSERT(mte->mte_handlers == 0,
4883 		    ("MSI-X table entry has handlers, but vector not mapped"));
4884 				error = PCIB_MAP_MSI(device_get_parent(dev),
4885 				    child, rman_get_start(irq), &addr, &data);
4886 				if (error)
4887 					goto bad;
4888 				mv->mv_address = addr;
4889 				mv->mv_data = data;
4890 			}
4891 
4892 			/*
4893 			 * The MSIX table entry must be made valid by
4894 			 * incrementing the mte_handlers before
4895 			 * calling pci_enable_msix() and
4896 			 * pci_resume_msix(). Else the MSIX rewrite
4897 			 * table quirk will not work as expected.
4898 			 */
4899 			mte->mte_handlers++;
4900 			if (mte->mte_handlers == 1) {
4901 				pci_enable_msix(child, rid - 1, mv->mv_address,
4902 				    mv->mv_data);
4903 				pci_unmask_msix(child, rid - 1);
4904 			}
4905 		}
4906 
4907 		/*
4908 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4909 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4910 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4911 		 */
4912 		if (!pci_has_quirk(pci_get_devid(child),
4913 		    PCI_QUIRK_MSI_INTX_BUG))
4914 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4915 		else
4916 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4917 	bad:
4918 		if (error) {
4919 			(void)bus_generic_teardown_intr(dev, child, irq,
4920 			    cookie);
4921 			return (error);
4922 		}
4923 	}
4924 	*cookiep = cookie;
4925 	return (0);
4926 }
4927 
4928 int
4929 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4930     void *cookie)
4931 {
4932 	struct msix_table_entry *mte;
4933 	struct resource_list_entry *rle;
4934 	struct pci_devinfo *dinfo;
4935 	int error, rid;
4936 
4937 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4938 		return (EINVAL);
4939 
4940 	/* If this isn't a direct child, just bail out */
4941 	if (device_get_parent(child) != dev)
4942 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4943 
4944 	rid = rman_get_rid(irq);
4945 	if (rid == 0) {
4946 		/* Mask INTx */
4947 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4948 	} else {
4949 		/*
4950 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4951 		 * decrement the appropriate handlers count and mask the
4952 		 * MSI-X message, or disable MSI messages if the count
4953 		 * drops to 0.
4954 		 */
4955 		dinfo = device_get_ivars(child);
4956 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4957 		if (rle->res != irq)
4958 			return (EINVAL);
4959 		if (dinfo->cfg.msi.msi_alloc > 0) {
4960 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4961 			    ("MSI-X index too high"));
4962 			if (dinfo->cfg.msi.msi_handlers == 0)
4963 				return (EINVAL);
4964 			dinfo->cfg.msi.msi_handlers--;
4965 			if (dinfo->cfg.msi.msi_handlers == 0)
4966 				pci_disable_msi(child);
4967 		} else {
4968 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4969 			    ("No MSI or MSI-X interrupts allocated"));
4970 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4971 			    ("MSI-X index too high"));
4972 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4973 			if (mte->mte_handlers == 0)
4974 				return (EINVAL);
4975 			mte->mte_handlers--;
4976 			if (mte->mte_handlers == 0)
4977 				pci_mask_msix(child, rid - 1);
4978 		}
4979 	}
4980 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4981 	if (rid > 0)
4982 		KASSERT(error == 0,
4983 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4984 	return (error);
4985 }
4986 
4987 int
4988 pci_print_child(device_t dev, device_t child)
4989 {
4990 	struct pci_devinfo *dinfo;
4991 	struct resource_list *rl;
4992 	int retval = 0;
4993 
4994 	dinfo = device_get_ivars(child);
4995 	rl = &dinfo->resources;
4996 
4997 	retval += bus_print_child_header(dev, child);
4998 
4999 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
5000 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
5001 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
5002 	if (device_get_flags(dev))
5003 		retval += printf(" flags %#x", device_get_flags(dev));
5004 
5005 	retval += printf(" at device %d.%d", pci_get_slot(child),
5006 	    pci_get_function(child));
5007 
5008 	retval += bus_print_child_domain(dev, child);
5009 	retval += bus_print_child_footer(dev, child);
5010 
5011 	return (retval);
5012 }
5013 
5014 static const struct
5015 {
5016 	int		class;
5017 	int		subclass;
5018 	int		report; /* 0 = bootverbose, 1 = always */
5019 	const char	*desc;
5020 } pci_nomatch_tab[] = {
5021 	{PCIC_OLD,		-1,			1, "old"},
5022 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
5023 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
5024 	{PCIC_STORAGE,		-1,			1, "mass storage"},
5025 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
5026 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
5027 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
5028 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
5029 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
5030 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
5031 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
5032 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
5033 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
5034 	{PCIC_NETWORK,		-1,			1, "network"},
5035 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
5036 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
5037 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
5038 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
5039 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
5040 	{PCIC_DISPLAY,		-1,			1, "display"},
5041 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
5042 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
5043 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
5044 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
5045 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
5046 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
5047 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
5048 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
5049 	{PCIC_MEMORY,		-1,			1, "memory"},
5050 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
5051 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
5052 	{PCIC_BRIDGE,		-1,			1, "bridge"},
5053 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
5054 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
5055 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
5056 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
5057 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
5058 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
5059 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
5060 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
5061 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
5062 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
5063 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
5064 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
5065 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
5066 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
5067 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
5068 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
5069 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
5070 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
5071 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
5072 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
5073 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
5074 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
5075 	{PCIC_INPUTDEV,		-1,			1, "input device"},
5076 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
5077 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
5078 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
5079 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
5080 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5081 	{PCIC_DOCKING,		-1,			1, "docking station"},
5082 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5083 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5084 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5085 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5086 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5087 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5088 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5089 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5090 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5091 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5092 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5093 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5094 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5095 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5096 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5097 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5098 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5099 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5100 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5101 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5102 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5103 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5104 	{PCIC_DASP,		-1,			0, "dasp"},
5105 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5106 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5107 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5108 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5109 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5110 	{0, 0, 0,		NULL}
5111 };
5112 
5113 void
5114 pci_probe_nomatch(device_t dev, device_t child)
5115 {
5116 	int i, report;
5117 	const char *cp, *scp;
5118 	char *device;
5119 
5120 	/*
5121 	 * Look for a listing for this device in a loaded device database.
5122 	 */
5123 	report = 1;
5124 	if ((device = pci_describe_device(child)) != NULL) {
5125 		device_printf(dev, "<%s>", device);
5126 		free(device, M_DEVBUF);
5127 	} else {
5128 		/*
5129 		 * Scan the class/subclass descriptions for a general
5130 		 * description.
5131 		 */
5132 		cp = "unknown";
5133 		scp = NULL;
5134 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5135 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5136 				if (pci_nomatch_tab[i].subclass == -1) {
5137 					cp = pci_nomatch_tab[i].desc;
5138 					report = pci_nomatch_tab[i].report;
5139 				} else if (pci_nomatch_tab[i].subclass ==
5140 				    pci_get_subclass(child)) {
5141 					scp = pci_nomatch_tab[i].desc;
5142 					report = pci_nomatch_tab[i].report;
5143 				}
5144 			}
5145 		}
5146 		if (report || bootverbose) {
5147 			device_printf(dev, "<%s%s%s>",
5148 			    cp ? cp : "",
5149 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5150 			    scp ? scp : "");
5151 		}
5152 	}
5153 	if (report || bootverbose) {
5154 		printf(" at device %d.%d (no driver attached)\n",
5155 		    pci_get_slot(child), pci_get_function(child));
5156 	}
5157 	pci_cfg_save(child, device_get_ivars(child), 1);
5158 }
5159 
5160 void
5161 pci_child_detached(device_t dev, device_t child)
5162 {
5163 	struct pci_devinfo *dinfo;
5164 	struct resource_list *rl;
5165 
5166 	dinfo = device_get_ivars(child);
5167 	rl = &dinfo->resources;
5168 
5169 	/*
5170 	 * Have to deallocate IRQs before releasing any MSI messages and
5171 	 * have to release MSI messages before deallocating any memory
5172 	 * BARs.
5173 	 */
5174 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5175 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5176 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5177 		if (dinfo->cfg.msi.msi_alloc != 0)
5178 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5179 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5180 		else
5181 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5182 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5183 		(void)pci_release_msi(child);
5184 	}
5185 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5186 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5187 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5188 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5189 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5190 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5191 
5192 	pci_cfg_save(child, dinfo, 1);
5193 }
5194 
5195 /*
5196  * Parse the PCI device database, if loaded, and return a pointer to a
5197  * description of the device.
5198  *
5199  * The database is flat text formatted as follows:
5200  *
5201  * Any line not in a valid format is ignored.
5202  * Lines are terminated with newline '\n' characters.
5203  *
5204  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5205  * the vendor name.
5206  *
5207  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5208  * - devices cannot be listed without a corresponding VENDOR line.
5209  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5210  * another TAB, then the device name.
5211  */
5212 
5213 /*
5214  * Assuming (ptr) points to the beginning of a line in the database,
5215  * return the vendor or device and description of the next entry.
5216  * The value of (vendor) or (device) inappropriate for the entry type
5217  * is set to -1.  Returns nonzero at the end of the database.
5218  *
5219  * Note that this is slightly unrobust in the face of corrupt data;
5220  * we attempt to safeguard against this by spamming the end of the
5221  * database with a newline when we initialise.
5222  */
5223 static int
5224 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5225 {
5226 	char	*cp = *ptr;
5227 	int	left;
5228 
5229 	*device = -1;
5230 	*vendor = -1;
5231 	**desc = '\0';
5232 	for (;;) {
5233 		left = pci_vendordata_size - (cp - pci_vendordata);
5234 		if (left <= 0) {
5235 			*ptr = cp;
5236 			return(1);
5237 		}
5238 
5239 		/* vendor entry? */
5240 		if (*cp != '\t' &&
5241 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5242 			break;
5243 		/* device entry? */
5244 		if (*cp == '\t' &&
5245 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5246 			break;
5247 
5248 		/* skip to next line */
5249 		while (*cp != '\n' && left > 0) {
5250 			cp++;
5251 			left--;
5252 		}
5253 		if (*cp == '\n') {
5254 			cp++;
5255 			left--;
5256 		}
5257 	}
5258 	/* skip to next line */
5259 	while (*cp != '\n' && left > 0) {
5260 		cp++;
5261 		left--;
5262 	}
5263 	if (*cp == '\n' && left > 0)
5264 		cp++;
5265 	*ptr = cp;
5266 	return(0);
5267 }
5268 
5269 static char *
5270 pci_describe_device(device_t dev)
5271 {
5272 	int	vendor, device;
5273 	char	*desc, *vp, *dp, *line;
5274 
5275 	desc = vp = dp = NULL;
5276 
5277 	/*
5278 	 * If we have no vendor data, we can't do anything.
5279 	 */
5280 	if (pci_vendordata == NULL)
5281 		goto out;
5282 
5283 	/*
5284 	 * Scan the vendor data looking for this device
5285 	 */
5286 	line = pci_vendordata;
5287 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5288 		goto out;
5289 	for (;;) {
5290 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5291 			goto out;
5292 		if (vendor == pci_get_vendor(dev))
5293 			break;
5294 	}
5295 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5296 		goto out;
5297 	for (;;) {
5298 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5299 			*dp = 0;
5300 			break;
5301 		}
5302 		if (vendor != -1) {
5303 			*dp = 0;
5304 			break;
5305 		}
5306 		if (device == pci_get_device(dev))
5307 			break;
5308 	}
5309 	if (dp[0] == '\0')
5310 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5311 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5312 	    NULL)
5313 		sprintf(desc, "%s, %s", vp, dp);
5314 out:
5315 	if (vp != NULL)
5316 		free(vp, M_DEVBUF);
5317 	if (dp != NULL)
5318 		free(dp, M_DEVBUF);
5319 	return(desc);
5320 }
5321 
5322 int
5323 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5324 {
5325 	struct pci_devinfo *dinfo;
5326 	pcicfgregs *cfg;
5327 
5328 	dinfo = device_get_ivars(child);
5329 	cfg = &dinfo->cfg;
5330 
5331 	switch (which) {
5332 	case PCI_IVAR_ETHADDR:
5333 		/*
5334 		 * The generic accessor doesn't deal with failure, so
5335 		 * we set the return value, then return an error.
5336 		 */
5337 		*((uint8_t **) result) = NULL;
5338 		return (EINVAL);
5339 	case PCI_IVAR_SUBVENDOR:
5340 		*result = cfg->subvendor;
5341 		break;
5342 	case PCI_IVAR_SUBDEVICE:
5343 		*result = cfg->subdevice;
5344 		break;
5345 	case PCI_IVAR_VENDOR:
5346 		*result = cfg->vendor;
5347 		break;
5348 	case PCI_IVAR_DEVICE:
5349 		*result = cfg->device;
5350 		break;
5351 	case PCI_IVAR_DEVID:
5352 		*result = (cfg->device << 16) | cfg->vendor;
5353 		break;
5354 	case PCI_IVAR_CLASS:
5355 		*result = cfg->baseclass;
5356 		break;
5357 	case PCI_IVAR_SUBCLASS:
5358 		*result = cfg->subclass;
5359 		break;
5360 	case PCI_IVAR_PROGIF:
5361 		*result = cfg->progif;
5362 		break;
5363 	case PCI_IVAR_REVID:
5364 		*result = cfg->revid;
5365 		break;
5366 	case PCI_IVAR_INTPIN:
5367 		*result = cfg->intpin;
5368 		break;
5369 	case PCI_IVAR_IRQ:
5370 		*result = cfg->intline;
5371 		break;
5372 	case PCI_IVAR_DOMAIN:
5373 		*result = cfg->domain;
5374 		break;
5375 	case PCI_IVAR_BUS:
5376 		*result = cfg->bus;
5377 		break;
5378 	case PCI_IVAR_SLOT:
5379 		*result = cfg->slot;
5380 		break;
5381 	case PCI_IVAR_FUNCTION:
5382 		*result = cfg->func;
5383 		break;
5384 	case PCI_IVAR_CMDREG:
5385 		*result = cfg->cmdreg;
5386 		break;
5387 	case PCI_IVAR_CACHELNSZ:
5388 		*result = cfg->cachelnsz;
5389 		break;
5390 	case PCI_IVAR_MINGNT:
5391 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5392 			*result = -1;
5393 			return (EINVAL);
5394 		}
5395 		*result = cfg->mingnt;
5396 		break;
5397 	case PCI_IVAR_MAXLAT:
5398 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5399 			*result = -1;
5400 			return (EINVAL);
5401 		}
5402 		*result = cfg->maxlat;
5403 		break;
5404 	case PCI_IVAR_LATTIMER:
5405 		*result = cfg->lattimer;
5406 		break;
5407 	default:
5408 		return (ENOENT);
5409 	}
5410 	return (0);
5411 }
5412 
5413 int
5414 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5415 {
5416 	struct pci_devinfo *dinfo;
5417 
5418 	dinfo = device_get_ivars(child);
5419 
5420 	switch (which) {
5421 	case PCI_IVAR_INTPIN:
5422 		dinfo->cfg.intpin = value;
5423 		return (0);
5424 	case PCI_IVAR_ETHADDR:
5425 	case PCI_IVAR_SUBVENDOR:
5426 	case PCI_IVAR_SUBDEVICE:
5427 	case PCI_IVAR_VENDOR:
5428 	case PCI_IVAR_DEVICE:
5429 	case PCI_IVAR_DEVID:
5430 	case PCI_IVAR_CLASS:
5431 	case PCI_IVAR_SUBCLASS:
5432 	case PCI_IVAR_PROGIF:
5433 	case PCI_IVAR_REVID:
5434 	case PCI_IVAR_IRQ:
5435 	case PCI_IVAR_DOMAIN:
5436 	case PCI_IVAR_BUS:
5437 	case PCI_IVAR_SLOT:
5438 	case PCI_IVAR_FUNCTION:
5439 		return (EINVAL);	/* disallow for now */
5440 
5441 	default:
5442 		return (ENOENT);
5443 	}
5444 }
5445 
5446 #include "opt_ddb.h"
5447 #ifdef DDB
5448 #include <ddb/ddb.h>
5449 #include <sys/cons.h>
5450 
5451 /*
5452  * List resources based on pci map registers, used for within ddb
5453  */
5454 
5455 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5456 {
5457 	struct pci_devinfo *dinfo;
5458 	struct devlist *devlist_head;
5459 	struct pci_conf *p;
5460 	const char *name;
5461 	int i, error, none_count;
5462 
5463 	none_count = 0;
5464 	/* get the head of the device queue */
5465 	devlist_head = &pci_devq;
5466 
5467 	/*
5468 	 * Go through the list of devices and print out devices
5469 	 */
5470 	for (error = 0, i = 0,
5471 	     dinfo = STAILQ_FIRST(devlist_head);
5472 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5473 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5474 		/* Populate pd_name and pd_unit */
5475 		name = NULL;
5476 		if (dinfo->cfg.dev)
5477 			name = device_get_name(dinfo->cfg.dev);
5478 
5479 		p = &dinfo->conf;
5480 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5481 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5482 			(name && *name) ? name : "none",
5483 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5484 			none_count++,
5485 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5486 			p->pc_sel.pc_func, (p->pc_class << 16) |
5487 			(p->pc_subclass << 8) | p->pc_progif,
5488 			(p->pc_subdevice << 16) | p->pc_subvendor,
5489 			(p->pc_device << 16) | p->pc_vendor,
5490 			p->pc_revid, p->pc_hdr);
5491 	}
5492 }
5493 #endif /* DDB */
5494 
5495 struct resource *
5496 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5497     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5498     u_int flags)
5499 {
5500 	struct pci_devinfo *dinfo = device_get_ivars(child);
5501 	struct resource_list *rl = &dinfo->resources;
5502 	struct resource *res;
5503 	struct pci_map *pm;
5504 	uint16_t cmd;
5505 	pci_addr_t map, testval;
5506 	int mapsize;
5507 
5508 	res = NULL;
5509 
5510 	/* If rid is managed by EA, ignore it */
5511 	if (pci_ea_is_enabled(child, *rid))
5512 		goto out;
5513 
5514 	pm = pci_find_bar(child, *rid);
5515 	if (pm != NULL) {
5516 		/* This is a BAR that we failed to allocate earlier. */
5517 		mapsize = pm->pm_size;
5518 		map = pm->pm_value;
5519 	} else {
5520 		/*
5521 		 * Weed out the bogons, and figure out how large the
5522 		 * BAR/map is.  BARs that read back 0 here are bogus
5523 		 * and unimplemented.  Note: atapci in legacy mode are
5524 		 * special and handled elsewhere in the code.  If you
5525 		 * have a atapci device in legacy mode and it fails
5526 		 * here, that other code is broken.
5527 		 */
5528 		pci_read_bar(child, *rid, &map, &testval, NULL);
5529 
5530 		/*
5531 		 * Determine the size of the BAR and ignore BARs with a size
5532 		 * of 0.  Device ROM BARs use a different mask value.
5533 		 */
5534 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5535 			mapsize = pci_romsize(testval);
5536 		else
5537 			mapsize = pci_mapsize(testval);
5538 		if (mapsize == 0)
5539 			goto out;
5540 		pm = pci_add_bar(child, *rid, map, mapsize);
5541 	}
5542 
5543 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5544 		if (type != SYS_RES_MEMORY) {
5545 			if (bootverbose)
5546 				device_printf(dev,
5547 				    "child %s requested type %d for rid %#x,"
5548 				    " but the BAR says it is an memio\n",
5549 				    device_get_nameunit(child), type, *rid);
5550 			goto out;
5551 		}
5552 	} else {
5553 		if (type != SYS_RES_IOPORT) {
5554 			if (bootverbose)
5555 				device_printf(dev,
5556 				    "child %s requested type %d for rid %#x,"
5557 				    " but the BAR says it is an ioport\n",
5558 				    device_get_nameunit(child), type, *rid);
5559 			goto out;
5560 		}
5561 	}
5562 
5563 	/*
5564 	 * For real BARs, we need to override the size that
5565 	 * the driver requests, because that's what the BAR
5566 	 * actually uses and we would otherwise have a
5567 	 * situation where we might allocate the excess to
5568 	 * another driver, which won't work.
5569 	 */
5570 	count = ((pci_addr_t)1 << mapsize) * num;
5571 	if (RF_ALIGNMENT(flags) < mapsize)
5572 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5573 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5574 		flags |= RF_PREFETCHABLE;
5575 
5576 	/*
5577 	 * Allocate enough resource, and then write back the
5578 	 * appropriate BAR for that resource.
5579 	 */
5580 	resource_list_add(rl, type, *rid, start, end, count);
5581 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5582 	    count, flags & ~RF_ACTIVE);
5583 	if (res == NULL) {
5584 		resource_list_delete(rl, type, *rid);
5585 		device_printf(child,
5586 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5587 		    count, *rid, type, start, end);
5588 		goto out;
5589 	}
5590 	if (bootverbose)
5591 		device_printf(child,
5592 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5593 		    count, *rid, type, rman_get_start(res));
5594 
5595 	/* Disable decoding via the CMD register before updating the BAR */
5596 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5597 	pci_write_config(child, PCIR_COMMAND,
5598 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5599 
5600 	map = rman_get_start(res);
5601 	pci_write_bar(child, pm, map);
5602 
5603 	/* Restore the original value of the CMD register */
5604 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5605 out:
5606 	return (res);
5607 }
5608 
5609 struct resource *
5610 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5611     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5612     u_int flags)
5613 {
5614 	struct pci_devinfo *dinfo;
5615 	struct resource_list *rl;
5616 	struct resource_list_entry *rle;
5617 	struct resource *res;
5618 	pcicfgregs *cfg;
5619 
5620 	/*
5621 	 * Perform lazy resource allocation
5622 	 */
5623 	dinfo = device_get_ivars(child);
5624 	rl = &dinfo->resources;
5625 	cfg = &dinfo->cfg;
5626 	switch (type) {
5627 	case PCI_RES_BUS:
5628 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5629 		    flags));
5630 	case SYS_RES_IRQ:
5631 		/*
5632 		 * Can't alloc legacy interrupt once MSI messages have
5633 		 * been allocated.
5634 		 */
5635 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5636 		    cfg->msix.msix_alloc > 0))
5637 			return (NULL);
5638 
5639 		/*
5640 		 * If the child device doesn't have an interrupt
5641 		 * routed and is deserving of an interrupt, try to
5642 		 * assign it one.
5643 		 */
5644 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5645 		    (cfg->intpin != 0))
5646 			pci_assign_interrupt(dev, child, 0);
5647 		break;
5648 	case SYS_RES_IOPORT:
5649 	case SYS_RES_MEMORY:
5650 		/*
5651 		 * PCI-PCI bridge I/O window resources are not BARs.
5652 		 * For those allocations just pass the request up the
5653 		 * tree.
5654 		 */
5655 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5656 			switch (*rid) {
5657 			case PCIR_IOBASEL_1:
5658 			case PCIR_MEMBASE_1:
5659 			case PCIR_PMBASEL_1:
5660 				/*
5661 				 * XXX: Should we bother creating a resource
5662 				 * list entry?
5663 				 */
5664 				return (bus_generic_alloc_resource(dev, child,
5665 				    type, rid, start, end, count, flags));
5666 			}
5667 		}
5668 		/* Reserve resources for this BAR if needed. */
5669 		rle = resource_list_find(rl, type, *rid);
5670 		if (rle == NULL) {
5671 			res = pci_reserve_map(dev, child, type, rid, start, end,
5672 			    count, num, flags);
5673 			if (res == NULL)
5674 				return (NULL);
5675 		}
5676 	}
5677 	return (resource_list_alloc(rl, dev, child, type, rid,
5678 	    start, end, count, flags));
5679 }
5680 
5681 struct resource *
5682 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5683     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5684 {
5685 #ifdef PCI_IOV
5686 	struct pci_devinfo *dinfo;
5687 #endif
5688 
5689 	if (device_get_parent(child) != dev)
5690 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5691 		    type, rid, start, end, count, flags));
5692 
5693 #ifdef PCI_IOV
5694 	dinfo = device_get_ivars(child);
5695 	if (dinfo->cfg.flags & PCICFG_VF) {
5696 		switch (type) {
5697 		/* VFs can't have I/O BARs. */
5698 		case SYS_RES_IOPORT:
5699 			return (NULL);
5700 		case SYS_RES_MEMORY:
5701 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5702 			    start, end, count, flags));
5703 		}
5704 
5705 		/* Fall through for other types of resource allocations. */
5706 	}
5707 #endif
5708 
5709 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5710 	    count, 1, flags));
5711 }
5712 
5713 int
5714 pci_release_resource(device_t dev, device_t child, struct resource *r)
5715 {
5716 	struct pci_devinfo *dinfo;
5717 	struct resource_list *rl;
5718 	pcicfgregs *cfg __unused;
5719 
5720 	if (device_get_parent(child) != dev)
5721 		return (bus_generic_release_resource(dev, child, r));
5722 
5723 	dinfo = device_get_ivars(child);
5724 	cfg = &dinfo->cfg;
5725 
5726 #ifdef PCI_IOV
5727 	if (cfg->flags & PCICFG_VF) {
5728 		switch (rman_get_type(r)) {
5729 		/* VFs can't have I/O BARs. */
5730 		case SYS_RES_IOPORT:
5731 			return (EDOOFUS);
5732 		case SYS_RES_MEMORY:
5733 			return (pci_vf_release_mem_resource(dev, child, r));
5734 		}
5735 
5736 		/* Fall through for other types of resource allocations. */
5737 	}
5738 #endif
5739 
5740 	/*
5741 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5742 	 * those allocations just pass the request up the tree.
5743 	 */
5744 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5745 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5746 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5747 		switch (rman_get_rid(r)) {
5748 		case PCIR_IOBASEL_1:
5749 		case PCIR_MEMBASE_1:
5750 		case PCIR_PMBASEL_1:
5751 			return (bus_generic_release_resource(dev, child, r));
5752 		}
5753 	}
5754 
5755 	rl = &dinfo->resources;
5756 	return (resource_list_release(rl, dev, child, r));
5757 }
5758 
5759 int
5760 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5761 {
5762 	struct pci_devinfo *dinfo;
5763 	int error, rid, type;
5764 
5765 	if (device_get_parent(child) != dev)
5766 		return (bus_generic_activate_resource(dev, child, r));
5767 
5768 	dinfo = device_get_ivars(child);
5769 #ifdef PCI_IOV
5770 	if (dinfo->cfg.flags & PCICFG_VF) {
5771 		switch (rman_get_type(r)) {
5772 		/* VFs can't have I/O BARs. */
5773 		case SYS_RES_IOPORT:
5774 			error = EINVAL;
5775 			break;
5776 		case SYS_RES_MEMORY:
5777 			error = pci_vf_activate_mem_resource(dev, child, r);
5778 			break;
5779 		default:
5780 			error = bus_generic_activate_resource(dev, child, r);
5781 			break;
5782 		}
5783 	} else
5784 #endif
5785 		error = bus_generic_activate_resource(dev, child, r);
5786 	if (error)
5787 		return (error);
5788 
5789 	rid = rman_get_rid(r);
5790 	type = rman_get_type(r);
5791 
5792 	/* Device ROMs need their decoding explicitly enabled. */
5793 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5794 		pci_write_bar(child, pci_find_bar(child, rid),
5795 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5796 
5797 	/* Enable decoding in the command register when activating BARs. */
5798 	switch (type) {
5799 	case SYS_RES_IOPORT:
5800 	case SYS_RES_MEMORY:
5801 		error = PCI_ENABLE_IO(dev, child, type);
5802 		break;
5803 	}
5804 	return (error);
5805 }
5806 
5807 int
5808 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5809 {
5810 	struct pci_devinfo *dinfo;
5811 	int error, rid, type;
5812 
5813 	if (device_get_parent(child) != dev)
5814 		return (bus_generic_deactivate_resource(dev, child, r));
5815 
5816 	dinfo = device_get_ivars(child);
5817 #ifdef PCI_IOV
5818 	if (dinfo->cfg.flags & PCICFG_VF) {
5819 		switch (rman_get_type(r)) {
5820 		/* VFs can't have I/O BARs. */
5821 		case SYS_RES_IOPORT:
5822 			error = EINVAL;
5823 			break;
5824 		case SYS_RES_MEMORY:
5825 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5826 			break;
5827 		default:
5828 			error = bus_generic_deactivate_resource(dev, child, r);
5829 			break;
5830 		}
5831 	} else
5832 #endif
5833 		error = bus_generic_deactivate_resource(dev, child, r);
5834 	if (error)
5835 		return (error);
5836 
5837 	/* Disable decoding for device ROMs. */
5838 	rid = rman_get_rid(r);
5839 	type = rman_get_type(r);
5840 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5841 		pci_write_bar(child, pci_find_bar(child, rid),
5842 		    rman_get_start(r));
5843 	return (0);
5844 }
5845 
5846 int
5847 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5848     rman_res_t start, rman_res_t end)
5849 {
5850 #ifdef PCI_IOV
5851 	struct pci_devinfo *dinfo;
5852 
5853 	if (device_get_parent(child) != dev)
5854 		return (bus_generic_adjust_resource(dev, child, r, start,
5855 		    end));
5856 
5857 	dinfo = device_get_ivars(child);
5858 	if (dinfo->cfg.flags & PCICFG_VF) {
5859 		switch (rman_get_type(r)) {
5860 		/* VFs can't have I/O BARs. */
5861 		case SYS_RES_IOPORT:
5862 			return (EINVAL);
5863 		case SYS_RES_MEMORY:
5864 			return (pci_vf_adjust_mem_resource(dev, child, r,
5865 			    start, end));
5866 		}
5867 
5868 		/* Fall through for other types of resource allocations. */
5869 	}
5870 #endif
5871 
5872 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5873 }
5874 
5875 int
5876 pci_map_resource(device_t dev, device_t child, struct resource *r,
5877     struct resource_map_request *argsp, struct resource_map *map)
5878 {
5879 #ifdef PCI_IOV
5880 	struct pci_devinfo *dinfo;
5881 
5882 	if (device_get_parent(child) != dev)
5883 		return (bus_generic_map_resource(dev, child, r, argsp,
5884 		    map));
5885 
5886 	dinfo = device_get_ivars(child);
5887 	if (dinfo->cfg.flags & PCICFG_VF) {
5888 		switch (rman_get_type(r)) {
5889 		/* VFs can't have I/O BARs. */
5890 		case SYS_RES_IOPORT:
5891 			return (EINVAL);
5892 		case SYS_RES_MEMORY:
5893 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5894 			    map));
5895 		}
5896 
5897 		/* Fall through for other types of resource allocations. */
5898 	}
5899 #endif
5900 
5901 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5902 }
5903 
5904 int
5905 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5906     struct resource_map *map)
5907 {
5908 #ifdef PCI_IOV
5909 	struct pci_devinfo *dinfo;
5910 
5911 	if (device_get_parent(child) != dev)
5912 		return (bus_generic_unmap_resource(dev, child, r, map));
5913 
5914 	dinfo = device_get_ivars(child);
5915 	if (dinfo->cfg.flags & PCICFG_VF) {
5916 		switch (rman_get_type(r)) {
5917 		/* VFs can't have I/O BARs. */
5918 		case SYS_RES_IOPORT:
5919 			return (EINVAL);
5920 		case SYS_RES_MEMORY:
5921 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5922 		}
5923 
5924 		/* Fall through for other types of resource allocations. */
5925 	}
5926 #endif
5927 
5928 	return (bus_generic_unmap_resource(dev, child, r, map));
5929 }
5930 
5931 void
5932 pci_child_deleted(device_t dev, device_t child)
5933 {
5934 	struct resource_list_entry *rle;
5935 	struct resource_list *rl;
5936 	struct pci_devinfo *dinfo;
5937 
5938 	dinfo = device_get_ivars(child);
5939 	rl = &dinfo->resources;
5940 
5941 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5942 
5943 	/* Turn off access to resources we're about to free */
5944 	if (bus_child_present(child) != 0) {
5945 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5946 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5947 
5948 		pci_disable_busmaster(child);
5949 	}
5950 
5951 	/* Free all allocated resources */
5952 	STAILQ_FOREACH(rle, rl, link) {
5953 		if (rle->res) {
5954 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5955 			    resource_list_busy(rl, rle->type, rle->rid)) {
5956 				pci_printf(&dinfo->cfg,
5957 				    "Resource still owned, oops. "
5958 				    "(type=%d, rid=%d, addr=%lx)\n",
5959 				    rle->type, rle->rid,
5960 				    rman_get_start(rle->res));
5961 				bus_release_resource(child, rle->type, rle->rid,
5962 				    rle->res);
5963 			}
5964 			resource_list_unreserve(rl, dev, child, rle->type,
5965 			    rle->rid);
5966 		}
5967 	}
5968 	resource_list_free(rl);
5969 
5970 	pci_freecfg(dinfo);
5971 }
5972 
5973 void
5974 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5975 {
5976 	struct pci_devinfo *dinfo;
5977 	struct resource_list *rl;
5978 	struct resource_list_entry *rle;
5979 
5980 	if (device_get_parent(child) != dev)
5981 		return;
5982 
5983 	dinfo = device_get_ivars(child);
5984 	rl = &dinfo->resources;
5985 	rle = resource_list_find(rl, type, rid);
5986 	if (rle == NULL)
5987 		return;
5988 
5989 	if (rle->res) {
5990 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5991 		    resource_list_busy(rl, type, rid)) {
5992 			device_printf(dev, "delete_resource: "
5993 			    "Resource still owned by child, oops. "
5994 			    "(type=%d, rid=%d, addr=%jx)\n",
5995 			    type, rid, rman_get_start(rle->res));
5996 			return;
5997 		}
5998 		resource_list_unreserve(rl, dev, child, type, rid);
5999 	}
6000 	resource_list_delete(rl, type, rid);
6001 }
6002 
6003 struct resource_list *
6004 pci_get_resource_list (device_t dev, device_t child)
6005 {
6006 	struct pci_devinfo *dinfo = device_get_ivars(child);
6007 
6008 	return (&dinfo->resources);
6009 }
6010 
6011 #ifdef IOMMU
6012 bus_dma_tag_t
6013 pci_get_dma_tag(device_t bus, device_t dev)
6014 {
6015 	bus_dma_tag_t tag;
6016 	struct pci_softc *sc;
6017 
6018 	if (device_get_parent(dev) == bus) {
6019 		/* try iommu and return if it works */
6020 		tag = iommu_get_dma_tag(bus, dev);
6021 	} else
6022 		tag = NULL;
6023 	if (tag == NULL) {
6024 		sc = device_get_softc(bus);
6025 		tag = sc->sc_dma_tag;
6026 	}
6027 	return (tag);
6028 }
6029 #else
6030 bus_dma_tag_t
6031 pci_get_dma_tag(device_t bus, device_t dev)
6032 {
6033 	struct pci_softc *sc = device_get_softc(bus);
6034 
6035 	return (sc->sc_dma_tag);
6036 }
6037 #endif
6038 
6039 uint32_t
6040 pci_read_config_method(device_t dev, device_t child, int reg, int width)
6041 {
6042 	struct pci_devinfo *dinfo = device_get_ivars(child);
6043 	pcicfgregs *cfg = &dinfo->cfg;
6044 
6045 #ifdef PCI_IOV
6046 	/*
6047 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
6048 	 * emulate them here.
6049 	 */
6050 	if (cfg->flags & PCICFG_VF) {
6051 		if (reg == PCIR_VENDOR) {
6052 			switch (width) {
6053 			case 4:
6054 				return (cfg->device << 16 | cfg->vendor);
6055 			case 2:
6056 				return (cfg->vendor);
6057 			case 1:
6058 				return (cfg->vendor & 0xff);
6059 			default:
6060 				return (0xffffffff);
6061 			}
6062 		} else if (reg == PCIR_DEVICE) {
6063 			switch (width) {
6064 			/* Note that an unaligned 4-byte read is an error. */
6065 			case 2:
6066 				return (cfg->device);
6067 			case 1:
6068 				return (cfg->device & 0xff);
6069 			default:
6070 				return (0xffffffff);
6071 			}
6072 		}
6073 	}
6074 #endif
6075 
6076 	return (PCIB_READ_CONFIG(device_get_parent(dev),
6077 	    cfg->bus, cfg->slot, cfg->func, reg, width));
6078 }
6079 
6080 void
6081 pci_write_config_method(device_t dev, device_t child, int reg,
6082     uint32_t val, int width)
6083 {
6084 	struct pci_devinfo *dinfo = device_get_ivars(child);
6085 	pcicfgregs *cfg = &dinfo->cfg;
6086 
6087 	PCIB_WRITE_CONFIG(device_get_parent(dev),
6088 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6089 }
6090 
6091 int
6092 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6093 {
6094 
6095 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6096 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6097 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6098 	return (0);
6099 }
6100 
6101 int
6102 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6103 {
6104 	struct pci_devinfo *dinfo;
6105 	pcicfgregs *cfg;
6106 
6107 	dinfo = device_get_ivars(child);
6108 	cfg = &dinfo->cfg;
6109 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6110 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6111 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6112 	    cfg->progif);
6113 	return (0);
6114 }
6115 
6116 int
6117 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6118     struct sbuf *sb)
6119 {
6120 	device_t parent = device_get_parent(bus);
6121 	int rv;
6122 
6123 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6124 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6125 		if (rv == 0) {
6126 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6127 			    pci_get_function(child));
6128 		}
6129 		return (0);
6130 	}
6131 	return (bus_generic_get_device_path(bus, child, locator, sb));
6132 }
6133 
6134 int
6135 pci_assign_interrupt_method(device_t dev, device_t child)
6136 {
6137 	struct pci_devinfo *dinfo = device_get_ivars(child);
6138 	pcicfgregs *cfg = &dinfo->cfg;
6139 
6140 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6141 	    cfg->intpin));
6142 }
6143 
6144 static void
6145 pci_lookup(void *arg, const char *name, device_t *dev)
6146 {
6147 	long val;
6148 	char *end;
6149 	int domain, bus, slot, func;
6150 
6151 	if (*dev != NULL)
6152 		return;
6153 
6154 	/*
6155 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6156 	 * pciB:S:F.  In the latter case, the domain is assumed to
6157 	 * be zero.
6158 	 */
6159 	if (strncmp(name, "pci", 3) != 0)
6160 		return;
6161 	val = strtol(name + 3, &end, 10);
6162 	if (val < 0 || val > INT_MAX || *end != ':')
6163 		return;
6164 	domain = val;
6165 	val = strtol(end + 1, &end, 10);
6166 	if (val < 0 || val > INT_MAX || *end != ':')
6167 		return;
6168 	bus = val;
6169 	val = strtol(end + 1, &end, 10);
6170 	if (val < 0 || val > INT_MAX)
6171 		return;
6172 	slot = val;
6173 	if (*end == ':') {
6174 		val = strtol(end + 1, &end, 10);
6175 		if (val < 0 || val > INT_MAX || *end != '\0')
6176 			return;
6177 		func = val;
6178 	} else if (*end == '\0') {
6179 		func = slot;
6180 		slot = bus;
6181 		bus = domain;
6182 		domain = 0;
6183 	} else
6184 		return;
6185 
6186 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6187 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6188 		return;
6189 
6190 	*dev = pci_find_dbsf(domain, bus, slot, func);
6191 }
6192 
6193 static int
6194 pci_modevent(module_t mod, int what, void *arg)
6195 {
6196 	static struct cdev *pci_cdev;
6197 	static eventhandler_tag tag;
6198 
6199 	switch (what) {
6200 	case MOD_LOAD:
6201 		STAILQ_INIT(&pci_devq);
6202 		pci_generation = 0;
6203 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6204 		    "pci");
6205 		pci_load_vendor_data();
6206 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6207 		    1000);
6208 		break;
6209 
6210 	case MOD_UNLOAD:
6211 		if (tag != NULL)
6212 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6213 		destroy_dev(pci_cdev);
6214 		break;
6215 	}
6216 
6217 	return (0);
6218 }
6219 
6220 static void
6221 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6222 {
6223 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6224 	struct pcicfg_pcie *cfg;
6225 	int version, pos;
6226 
6227 	cfg = &dinfo->cfg.pcie;
6228 	pos = cfg->pcie_location;
6229 
6230 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6231 
6232 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6233 
6234 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6235 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6236 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6237 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6238 
6239 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6240 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6241 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6242 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6243 
6244 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6245 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6246 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6247 
6248 	if (version > 1) {
6249 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6250 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6251 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6252 	}
6253 #undef WREG
6254 }
6255 
6256 static void
6257 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6258 {
6259 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6260 	    dinfo->cfg.pcix.pcix_command,  2);
6261 }
6262 
6263 void
6264 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6265 {
6266 
6267 	/*
6268 	 * Restore the device to full power mode.  We must do this
6269 	 * before we restore the registers because moving from D3 to
6270 	 * D0 will cause the chip's BARs and some other registers to
6271 	 * be reset to some unknown power on reset values.  Cut down
6272 	 * the noise on boot by doing nothing if we are already in
6273 	 * state D0.
6274 	 */
6275 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6276 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6277 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6278 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6279 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6280 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6281 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6282 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6283 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6284 	case PCIM_HDRTYPE_NORMAL:
6285 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6286 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6287 		break;
6288 	case PCIM_HDRTYPE_BRIDGE:
6289 		pci_write_config(dev, PCIR_SECLAT_1,
6290 		    dinfo->cfg.bridge.br_seclat, 1);
6291 		pci_write_config(dev, PCIR_SUBBUS_1,
6292 		    dinfo->cfg.bridge.br_subbus, 1);
6293 		pci_write_config(dev, PCIR_SECBUS_1,
6294 		    dinfo->cfg.bridge.br_secbus, 1);
6295 		pci_write_config(dev, PCIR_PRIBUS_1,
6296 		    dinfo->cfg.bridge.br_pribus, 1);
6297 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6298 		    dinfo->cfg.bridge.br_control, 2);
6299 		break;
6300 	case PCIM_HDRTYPE_CARDBUS:
6301 		pci_write_config(dev, PCIR_SECLAT_2,
6302 		    dinfo->cfg.bridge.br_seclat, 1);
6303 		pci_write_config(dev, PCIR_SUBBUS_2,
6304 		    dinfo->cfg.bridge.br_subbus, 1);
6305 		pci_write_config(dev, PCIR_SECBUS_2,
6306 		    dinfo->cfg.bridge.br_secbus, 1);
6307 		pci_write_config(dev, PCIR_PRIBUS_2,
6308 		    dinfo->cfg.bridge.br_pribus, 1);
6309 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6310 		    dinfo->cfg.bridge.br_control, 2);
6311 		break;
6312 	}
6313 	pci_restore_bars(dev);
6314 
6315 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6316 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6317 
6318 	/*
6319 	 * Restore extended capabilities for PCI-Express and PCI-X
6320 	 */
6321 	if (dinfo->cfg.pcie.pcie_location != 0)
6322 		pci_cfg_restore_pcie(dev, dinfo);
6323 	if (dinfo->cfg.pcix.pcix_location != 0)
6324 		pci_cfg_restore_pcix(dev, dinfo);
6325 
6326 	/* Restore MSI and MSI-X configurations if they are present. */
6327 	if (dinfo->cfg.msi.msi_location != 0)
6328 		pci_resume_msi(dev);
6329 	if (dinfo->cfg.msix.msix_location != 0)
6330 		pci_resume_msix(dev);
6331 
6332 #ifdef PCI_IOV
6333 	if (dinfo->cfg.iov != NULL)
6334 		pci_iov_cfg_restore(dev, dinfo);
6335 #endif
6336 }
6337 
6338 static void
6339 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6340 {
6341 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6342 	struct pcicfg_pcie *cfg;
6343 	int version, pos;
6344 
6345 	cfg = &dinfo->cfg.pcie;
6346 	pos = cfg->pcie_location;
6347 
6348 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6349 
6350 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6351 
6352 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6353 
6354 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6355 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6356 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6357 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6358 
6359 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6360 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6361 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6362 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6363 
6364 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6365 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6366 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6367 
6368 	if (version > 1) {
6369 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6370 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6371 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6372 	}
6373 #undef RREG
6374 }
6375 
6376 static void
6377 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6378 {
6379 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6380 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6381 }
6382 
6383 void
6384 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6385 {
6386 	uint32_t cls;
6387 	int ps;
6388 
6389 	/*
6390 	 * Some drivers apparently write to these registers w/o updating our
6391 	 * cached copy.  No harm happens if we update the copy, so do so here
6392 	 * so we can restore them.  The COMMAND register is modified by the
6393 	 * bus w/o updating the cache.  This should represent the normally
6394 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6395 	 */
6396 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6397 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6398 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6399 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6400 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6401 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6402 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6403 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6404 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6405 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6406 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6407 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6408 	case PCIM_HDRTYPE_NORMAL:
6409 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6410 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6411 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6412 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6413 		break;
6414 	case PCIM_HDRTYPE_BRIDGE:
6415 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6416 		    PCIR_SECLAT_1, 1);
6417 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6418 		    PCIR_SUBBUS_1, 1);
6419 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6420 		    PCIR_SECBUS_1, 1);
6421 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6422 		    PCIR_PRIBUS_1, 1);
6423 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6424 		    PCIR_BRIDGECTL_1, 2);
6425 		break;
6426 	case PCIM_HDRTYPE_CARDBUS:
6427 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6428 		    PCIR_SECLAT_2, 1);
6429 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6430 		    PCIR_SUBBUS_2, 1);
6431 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6432 		    PCIR_SECBUS_2, 1);
6433 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6434 		    PCIR_PRIBUS_2, 1);
6435 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6436 		    PCIR_BRIDGECTL_2, 2);
6437 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6438 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6439 		break;
6440 	}
6441 
6442 	if (dinfo->cfg.pcie.pcie_location != 0)
6443 		pci_cfg_save_pcie(dev, dinfo);
6444 
6445 	if (dinfo->cfg.pcix.pcix_location != 0)
6446 		pci_cfg_save_pcix(dev, dinfo);
6447 
6448 #ifdef PCI_IOV
6449 	if (dinfo->cfg.iov != NULL)
6450 		pci_iov_cfg_save(dev, dinfo);
6451 #endif
6452 
6453 	/*
6454 	 * don't set the state for display devices, base peripherals and
6455 	 * memory devices since bad things happen when they are powered down.
6456 	 * We should (a) have drivers that can easily detach and (b) use
6457 	 * generic drivers for these devices so that some device actually
6458 	 * attaches.  We need to make sure that when we implement (a) we don't
6459 	 * power the device down on a reattach.
6460 	 */
6461 	cls = pci_get_class(dev);
6462 	if (!setstate)
6463 		return;
6464 	switch (pci_do_power_nodriver)
6465 	{
6466 		case 0:		/* NO powerdown at all */
6467 			return;
6468 		case 1:		/* Conservative about what to power down */
6469 			if (cls == PCIC_STORAGE)
6470 				return;
6471 			/*FALLTHROUGH*/
6472 		case 2:		/* Aggressive about what to power down */
6473 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6474 			    cls == PCIC_BASEPERIPH)
6475 				return;
6476 			/*FALLTHROUGH*/
6477 		case 3:		/* Power down everything */
6478 			break;
6479 	}
6480 	/*
6481 	 * PCI spec says we can only go into D3 state from D0 state.
6482 	 * Transition from D[12] into D0 before going to D3 state.
6483 	 */
6484 	ps = pci_get_powerstate(dev);
6485 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6486 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6487 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6488 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6489 }
6490 
6491 /* Wrapper APIs suitable for device driver use. */
6492 void
6493 pci_save_state(device_t dev)
6494 {
6495 	struct pci_devinfo *dinfo;
6496 
6497 	dinfo = device_get_ivars(dev);
6498 	pci_cfg_save(dev, dinfo, 0);
6499 }
6500 
6501 void
6502 pci_restore_state(device_t dev)
6503 {
6504 	struct pci_devinfo *dinfo;
6505 
6506 	dinfo = device_get_ivars(dev);
6507 	pci_cfg_restore(dev, dinfo);
6508 }
6509 
6510 static int
6511 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6512     uintptr_t *id)
6513 {
6514 
6515 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6516 }
6517 
6518 /* Find the upstream port of a given PCI device in a root complex. */
6519 device_t
6520 pci_find_pcie_root_port(device_t dev)
6521 {
6522 	struct pci_devinfo *dinfo;
6523 	devclass_t pci_class;
6524 	device_t pcib, bus;
6525 
6526 	pci_class = devclass_find("pci");
6527 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6528 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6529 
6530 	/*
6531 	 * Walk the bridge hierarchy until we find a PCI-e root
6532 	 * port or a non-PCI device.
6533 	 */
6534 	for (;;) {
6535 		bus = device_get_parent(dev);
6536 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6537 		    device_get_nameunit(dev)));
6538 
6539 		pcib = device_get_parent(bus);
6540 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6541 		    device_get_nameunit(bus)));
6542 
6543 		/*
6544 		 * pcib's parent must be a PCI bus for this to be a
6545 		 * PCI-PCI bridge.
6546 		 */
6547 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6548 			return (NULL);
6549 
6550 		dinfo = device_get_ivars(pcib);
6551 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6552 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6553 			return (pcib);
6554 
6555 		dev = pcib;
6556 	}
6557 }
6558 
6559 /*
6560  * Wait for pending transactions to complete on a PCI-express function.
6561  *
6562  * The maximum delay is specified in milliseconds in max_delay.  Note
6563  * that this function may sleep.
6564  *
6565  * Returns true if the function is idle and false if the timeout is
6566  * exceeded.  If dev is not a PCI-express function, this returns true.
6567  */
6568 bool
6569 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6570 {
6571 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6572 	uint16_t sta;
6573 	int cap;
6574 
6575 	cap = dinfo->cfg.pcie.pcie_location;
6576 	if (cap == 0)
6577 		return (true);
6578 
6579 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6580 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6581 		if (max_delay == 0)
6582 			return (false);
6583 
6584 		/* Poll once every 100 milliseconds up to the timeout. */
6585 		if (max_delay > 100) {
6586 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6587 			max_delay -= 100;
6588 		} else {
6589 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6590 			    C_HARDCLOCK);
6591 			max_delay = 0;
6592 		}
6593 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6594 	}
6595 
6596 	return (true);
6597 }
6598 
6599 /*
6600  * Determine the maximum Completion Timeout in microseconds.
6601  *
6602  * For non-PCI-express functions this returns 0.
6603  */
6604 int
6605 pcie_get_max_completion_timeout(device_t dev)
6606 {
6607 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6608 	int cap;
6609 
6610 	cap = dinfo->cfg.pcie.pcie_location;
6611 	if (cap == 0)
6612 		return (0);
6613 
6614 	/*
6615 	 * Functions using the 1.x spec use the default timeout range of
6616 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6617 	 * support programmable timeouts also use this range.
6618 	 */
6619 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6620 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6621 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6622 		return (50 * 1000);
6623 
6624 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6625 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6626 	case PCIEM_CTL2_COMP_TIMO_100US:
6627 		return (100);
6628 	case PCIEM_CTL2_COMP_TIMO_10MS:
6629 		return (10 * 1000);
6630 	case PCIEM_CTL2_COMP_TIMO_55MS:
6631 		return (55 * 1000);
6632 	case PCIEM_CTL2_COMP_TIMO_210MS:
6633 		return (210 * 1000);
6634 	case PCIEM_CTL2_COMP_TIMO_900MS:
6635 		return (900 * 1000);
6636 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6637 		return (3500 * 1000);
6638 	case PCIEM_CTL2_COMP_TIMO_13S:
6639 		return (13 * 1000 * 1000);
6640 	case PCIEM_CTL2_COMP_TIMO_64S:
6641 		return (64 * 1000 * 1000);
6642 	default:
6643 		return (50 * 1000);
6644 	}
6645 }
6646 
6647 void
6648 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6649 {
6650 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6651 	const char *s;
6652 	int aer;
6653 	uint32_t r, r1;
6654 	uint16_t rs;
6655 
6656 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6657 		s = "Correctable";
6658 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6659 		s = "Uncorrectable (Non-Fatal)";
6660 	else
6661 		s = "Uncorrectable (Fatal)";
6662 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6663 	if (aerp) {
6664 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6665 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6666 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6667 		} else {
6668 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6669 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6670 		}
6671 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6672 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6673 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6674 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6675 			printf(" severity 0x%08x first %d\n",
6676 			    r, rs & 0x1f);
6677 		} else
6678 			printf("\n");
6679 	}
6680 
6681 	/* As kind of recovery just report and clear the error statuses. */
6682 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6683 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6684 		if (r != 0) {
6685 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6686 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6687 		}
6688 
6689 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6690 		if (r != 0) {
6691 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6692 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6693 		}
6694 	}
6695 	if (dinfo->cfg.pcie.pcie_location != 0) {
6696 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6697 		    PCIER_DEVICE_STA, 2);
6698 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6699 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6700 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6701 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6702 			    PCIER_DEVICE_STA, rs, 2);
6703 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6704 		}
6705 	}
6706 }
6707 
6708 /*
6709  * Perform a Function Level Reset (FLR) on a device.
6710  *
6711  * This function first waits for any pending transactions to complete
6712  * within the timeout specified by max_delay.  If transactions are
6713  * still pending, the function will return false without attempting a
6714  * reset.
6715  *
6716  * If dev is not a PCI-express function or does not support FLR, this
6717  * function returns false.
6718  *
6719  * Note that no registers are saved or restored.  The caller is
6720  * responsible for saving and restoring any registers including
6721  * PCI-standard registers via pci_save_state() and
6722  * pci_restore_state().
6723  */
6724 bool
6725 pcie_flr(device_t dev, u_int max_delay, bool force)
6726 {
6727 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6728 	uint16_t cmd, ctl;
6729 	int compl_delay;
6730 	int cap;
6731 
6732 	cap = dinfo->cfg.pcie.pcie_location;
6733 	if (cap == 0)
6734 		return (false);
6735 
6736 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6737 		return (false);
6738 
6739 	/*
6740 	 * Disable busmastering to prevent generation of new
6741 	 * transactions while waiting for the device to go idle.  If
6742 	 * the idle timeout fails, the command register is restored
6743 	 * which will re-enable busmastering.
6744 	 */
6745 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6746 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6747 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6748 		if (!force) {
6749 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6750 			return (false);
6751 		}
6752 		pci_printf(&dinfo->cfg,
6753 		    "Resetting with transactions pending after %d ms\n",
6754 		    max_delay);
6755 
6756 		/*
6757 		 * Extend the post-FLR delay to cover the maximum
6758 		 * Completion Timeout delay of anything in flight
6759 		 * during the FLR delay.  Enforce a minimum delay of
6760 		 * at least 10ms.
6761 		 */
6762 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6763 		if (compl_delay < 10)
6764 			compl_delay = 10;
6765 	} else
6766 		compl_delay = 0;
6767 
6768 	/* Initiate the reset. */
6769 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6770 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6771 	    PCIEM_CTL_INITIATE_FLR, 2);
6772 
6773 	/* Wait for 100ms. */
6774 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6775 
6776 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6777 	    PCIEM_STA_TRANSACTION_PND)
6778 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6779 	return (true);
6780 }
6781 
6782 /*
6783  * Attempt a power-management reset by cycling the device in/out of D3
6784  * state.  PCI spec says we can only go into D3 state from D0 state.
6785  * Transition from D[12] into D0 before going to D3 state.
6786  */
6787 int
6788 pci_power_reset(device_t dev)
6789 {
6790 	int ps;
6791 
6792 	ps = pci_get_powerstate(dev);
6793 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6794 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6795 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6796 	pci_set_powerstate(dev, ps);
6797 	return (0);
6798 }
6799 
6800 /*
6801  * Try link drop and retrain of the downstream port of upstream
6802  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6803  * cause Conventional Hot reset of the device in the slot.
6804  * Alternative, for PCIe, could be the secondary bus reset initiatied
6805  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6806  */
6807 int
6808 pcie_link_reset(device_t port, int pcie_location)
6809 {
6810 	uint16_t v;
6811 
6812 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6813 	v |= PCIEM_LINK_CTL_LINK_DIS;
6814 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6815 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6816 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6817 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6818 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6819 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6820 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6821 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6822 }
6823 
6824 static int
6825 pci_reset_post(device_t dev, device_t child)
6826 {
6827 
6828 	if (dev == device_get_parent(child))
6829 		pci_restore_state(child);
6830 	return (0);
6831 }
6832 
6833 static int
6834 pci_reset_prepare(device_t dev, device_t child)
6835 {
6836 
6837 	if (dev == device_get_parent(child))
6838 		pci_save_state(child);
6839 	return (0);
6840 }
6841 
6842 static int
6843 pci_reset_child(device_t dev, device_t child, int flags)
6844 {
6845 	int error;
6846 
6847 	if (dev == NULL || device_get_parent(child) != dev)
6848 		return (0);
6849 	if ((flags & DEVF_RESET_DETACH) != 0) {
6850 		error = device_get_state(child) == DS_ATTACHED ?
6851 		    device_detach(child) : 0;
6852 	} else {
6853 		error = BUS_SUSPEND_CHILD(dev, child);
6854 	}
6855 	if (error == 0) {
6856 		if (!pcie_flr(child, 1000, false)) {
6857 			error = BUS_RESET_PREPARE(dev, child);
6858 			if (error == 0)
6859 				pci_power_reset(child);
6860 			BUS_RESET_POST(dev, child);
6861 		}
6862 		if ((flags & DEVF_RESET_DETACH) != 0)
6863 			device_probe_and_attach(child);
6864 		else
6865 			BUS_RESUME_CHILD(dev, child);
6866 	}
6867 	return (error);
6868 }
6869 
6870 const struct pci_device_table *
6871 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6872 {
6873 	bool match;
6874 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6875 
6876 	vendor = pci_get_vendor(child);
6877 	device = pci_get_device(child);
6878 	subvendor = pci_get_subvendor(child);
6879 	subdevice = pci_get_subdevice(child);
6880 	class = pci_get_class(child);
6881 	subclass = pci_get_subclass(child);
6882 	revid = pci_get_revid(child);
6883 	while (nelt-- > 0) {
6884 		match = true;
6885 		if (id->match_flag_vendor)
6886 			match &= vendor == id->vendor;
6887 		if (id->match_flag_device)
6888 			match &= device == id->device;
6889 		if (id->match_flag_subvendor)
6890 			match &= subvendor == id->subvendor;
6891 		if (id->match_flag_subdevice)
6892 			match &= subdevice == id->subdevice;
6893 		if (id->match_flag_class)
6894 			match &= class == id->class_id;
6895 		if (id->match_flag_subclass)
6896 			match &= subclass == id->subclass;
6897 		if (id->match_flag_revid)
6898 			match &= revid == id->revid;
6899 		if (match)
6900 			return (id);
6901 		id++;
6902 	}
6903 	return (NULL);
6904 }
6905 
6906 static void
6907 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6908 {
6909 	const char *dev_name;
6910 	device_t dev;
6911 
6912 	dev = dinfo->cfg.dev;
6913 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6914 	    dinfo->cfg.slot, dinfo->cfg.func);
6915 	dev_name = device_get_name(dev);
6916 	if (dev_name != NULL)
6917 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6918 }
6919 
6920 void
6921 pci_print_faulted_dev(void)
6922 {
6923 	struct pci_devinfo *dinfo;
6924 	device_t dev;
6925 	int aer, i;
6926 	uint32_t r1, r2;
6927 	uint16_t status;
6928 
6929 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6930 		dev = dinfo->cfg.dev;
6931 		status = pci_read_config(dev, PCIR_STATUS, 2);
6932 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6933 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6934 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6935 		if (status != 0) {
6936 			pci_print_faulted_dev_name(dinfo);
6937 			printf(" error 0x%04x\n", status);
6938 		}
6939 		if (dinfo->cfg.pcie.pcie_location != 0) {
6940 			status = pci_read_config(dev,
6941 			    dinfo->cfg.pcie.pcie_location +
6942 			    PCIER_DEVICE_STA, 2);
6943 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6944 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6945 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6946 				pci_print_faulted_dev_name(dinfo);
6947 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6948 				    pci_read_config(dev,
6949 				    dinfo->cfg.pcie.pcie_location +
6950 				    PCIER_DEVICE_CTL, 2),
6951 				    status);
6952 			}
6953 		}
6954 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6955 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6956 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6957 			if (r1 != 0 || r2 != 0) {
6958 				pci_print_faulted_dev_name(dinfo);
6959 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6960 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6961 				    r1, pci_read_config(dev, aer +
6962 				    PCIR_AER_UC_MASK, 4),
6963 				    pci_read_config(dev, aer +
6964 				    PCIR_AER_UC_SEVERITY, 4),
6965 				    r2, pci_read_config(dev, aer +
6966 				    PCIR_AER_COR_MASK, 4),
6967 				    pci_read_config(dev, aer +
6968 				    PCIR_AER_CAP_CONTROL, 4));
6969 				for (i = 0; i < 4; i++) {
6970 					r1 = pci_read_config(dev, aer +
6971 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6972 					printf("    HL%d: 0x%08x\n", i, r1);
6973 				}
6974 			}
6975 		}
6976 	}
6977 }
6978 
6979 #ifdef DDB
6980 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6981 {
6982 
6983 	pci_print_faulted_dev();
6984 }
6985 
6986 static void
6987 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6988 {
6989 	device_t dev;
6990 	int aer;
6991 	uint32_t r;
6992 
6993 	dev = dinfo->cfg.dev;
6994 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6995 	    PCIER_DEVICE_STA, 2);
6996 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6997 	    PCIER_DEVICE_STA, r, 2);
6998 
6999 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
7000 		return;
7001 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
7002 	if (r != 0)
7003 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
7004 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
7005 	if (r != 0)
7006 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
7007 }
7008 
7009 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
7010 {
7011 	struct pci_devinfo *dinfo;
7012 	device_t dev;
7013 	uint16_t status, status1;
7014 
7015 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
7016 		dev = dinfo->cfg.dev;
7017 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
7018 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
7019 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
7020 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
7021 		if (status1 != 0) {
7022 			status &= ~status1;
7023 			pci_write_config(dev, PCIR_STATUS, status, 2);
7024 		}
7025 		if (dinfo->cfg.pcie.pcie_location != 0)
7026 			db_clear_pcie_errors(dinfo);
7027 	}
7028 }
7029 #endif
7030