xref: /freebsd/sys/dev/pci/pci.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
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 int		pci_has_quirk(uint32_t devid, int quirk);
92 static pci_addr_t	pci_mapbase(uint64_t mapreg);
93 static const char	*pci_maptype(uint64_t mapreg);
94 static int		pci_maprange(uint64_t mapreg);
95 static pci_addr_t	pci_rombase(uint64_t mapreg);
96 static int		pci_romsize(uint64_t testval);
97 static void		pci_fixancient(pcicfgregs *cfg);
98 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
99 
100 static int		pci_porten(device_t dev);
101 static int		pci_memen(device_t dev);
102 static void		pci_assign_interrupt(device_t bus, device_t dev,
103 			    int force_route);
104 static int		pci_add_map(device_t bus, device_t dev, int reg,
105 			    struct resource_list *rl, int force, int prefetch);
106 static int		pci_probe(device_t dev);
107 static void		pci_load_vendor_data(void);
108 static int		pci_describe_parse_line(char **ptr, int *vendor,
109 			    int *device, char **desc);
110 static char		*pci_describe_device(device_t dev);
111 static int		pci_modevent(module_t mod, int what, void *arg);
112 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
113 			    pcicfgregs *cfg);
114 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
115 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
116 			    int reg, uint32_t *data);
117 #if 0
118 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
119 			    int reg, uint32_t data);
120 #endif
121 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
122 static void		pci_mask_msix(device_t dev, u_int index);
123 static void		pci_unmask_msix(device_t dev, u_int index);
124 static int		pci_msi_blacklisted(void);
125 static int		pci_msix_blacklisted(void);
126 static void		pci_resume_msi(device_t dev);
127 static void		pci_resume_msix(device_t dev);
128 static int		pci_remap_intr_method(device_t bus, device_t dev,
129 			    u_int irq);
130 static void		pci_hint_device_unit(device_t acdev, device_t child,
131 			    const char *name, int *unitp);
132 static int		pci_reset_post(device_t dev, device_t child);
133 static int		pci_reset_prepare(device_t dev, device_t child);
134 static int		pci_reset_child(device_t dev, device_t child,
135 			    int flags);
136 
137 static int		pci_get_id_method(device_t dev, device_t child,
138 			    enum pci_id_type type, uintptr_t *rid);
139 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
140     int b, int s, int f, uint16_t vid, uint16_t did);
141 
142 static device_method_t pci_methods[] = {
143 	/* Device interface */
144 	DEVMETHOD(device_probe,		pci_probe),
145 	DEVMETHOD(device_attach,	pci_attach),
146 	DEVMETHOD(device_detach,	pci_detach),
147 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
148 	DEVMETHOD(device_suspend,	bus_generic_suspend),
149 	DEVMETHOD(device_resume,	pci_resume),
150 
151 	/* Bus interface */
152 	DEVMETHOD(bus_print_child,	pci_print_child),
153 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
154 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
155 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
156 	DEVMETHOD(bus_driver_added,	pci_driver_added),
157 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
158 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
159 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
160 	DEVMETHOD(bus_reset_post,	pci_reset_post),
161 	DEVMETHOD(bus_reset_child,	pci_reset_child),
162 
163 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
164 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
165 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
166 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
167 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
168 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
169 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
170 	DEVMETHOD(bus_release_resource,	pci_release_resource),
171 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
172 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_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 devices into D3 state.  2 means"
357     " aggressively place devices 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 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
403 static int pci_clear_buses;
404 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
405     "Ignore firmware-assigned bus numbers.");
406 #endif
407 
408 static int pci_enable_ari = 1;
409 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
410     0, "Enable support for PCIe Alternative RID Interpretation");
411 
412 int pci_enable_aspm = 1;
413 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
414     0, "Enable support for PCIe Active State Power Management");
415 
416 static int pci_clear_aer_on_attach = 0;
417 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
418     &pci_clear_aer_on_attach, 0,
419     "Clear port and device AER state on driver attach");
420 
421 static bool pci_enable_mps_tune = true;
422 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
423     &pci_enable_mps_tune, 1,
424     "Enable tuning of MPS(maximum payload size)." );
425 
426 static int
427 pci_has_quirk(uint32_t devid, int quirk)
428 {
429 	const struct pci_quirk *q;
430 
431 	for (q = &pci_quirks[0]; q->devid; q++) {
432 		if (q->devid == devid && q->type == quirk)
433 			return (1);
434 	}
435 	return (0);
436 }
437 
438 /* Find a device_t by bus/slot/function in domain 0 */
439 
440 device_t
441 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
442 {
443 
444 	return (pci_find_dbsf(0, bus, slot, func));
445 }
446 
447 /* Find a device_t by domain/bus/slot/function */
448 
449 device_t
450 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
451 {
452 	struct pci_devinfo *dinfo = NULL;
453 
454 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
455 		if ((dinfo->cfg.domain == domain) &&
456 		    (dinfo->cfg.bus == bus) &&
457 		    (dinfo->cfg.slot == slot) &&
458 		    (dinfo->cfg.func == func)) {
459 			break;
460 		}
461 	}
462 
463 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
464 }
465 
466 /* Find a device_t by vendor/device ID */
467 
468 device_t
469 pci_find_device(uint16_t vendor, uint16_t device)
470 {
471 	struct pci_devinfo *dinfo;
472 
473 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
474 		if ((dinfo->cfg.vendor == vendor) &&
475 		    (dinfo->cfg.device == device)) {
476 			return (dinfo->cfg.dev);
477 		}
478 	}
479 
480 	return (NULL);
481 }
482 
483 device_t
484 pci_find_class(uint8_t class, uint8_t subclass)
485 {
486 	struct pci_devinfo *dinfo;
487 
488 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
489 		if (dinfo->cfg.baseclass == class &&
490 		    dinfo->cfg.subclass == subclass) {
491 			return (dinfo->cfg.dev);
492 		}
493 	}
494 
495 	return (NULL);
496 }
497 
498 device_t
499 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
500 {
501 	struct pci_devinfo *dinfo;
502 	bool found = false;
503 
504 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
505 		if (from != NULL && found == false) {
506 			if (from != dinfo->cfg.dev)
507 				continue;
508 			found = true;
509 			continue;
510 		}
511 		if (dinfo->cfg.baseclass == class &&
512 		    dinfo->cfg.subclass == subclass) {
513 			return (dinfo->cfg.dev);
514 		}
515 	}
516 
517 	return (NULL);
518 }
519 
520 static int
521 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
522 {
523 	va_list ap;
524 	int retval;
525 
526 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
527 	    cfg->func);
528 	va_start(ap, fmt);
529 	retval += vprintf(fmt, ap);
530 	va_end(ap);
531 	return (retval);
532 }
533 
534 /* return base address of memory or port map */
535 
536 static pci_addr_t
537 pci_mapbase(uint64_t mapreg)
538 {
539 
540 	if (PCI_BAR_MEM(mapreg))
541 		return (mapreg & PCIM_BAR_MEM_BASE);
542 	else
543 		return (mapreg & PCIM_BAR_IO_BASE);
544 }
545 
546 /* return map type of memory or port map */
547 
548 static const char *
549 pci_maptype(uint64_t mapreg)
550 {
551 
552 	if (PCI_BAR_IO(mapreg))
553 		return ("I/O Port");
554 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
555 		return ("Prefetchable Memory");
556 	return ("Memory");
557 }
558 
559 /* return log2 of map size decoded for memory or port map */
560 
561 int
562 pci_mapsize(uint64_t testval)
563 {
564 	int ln2size;
565 
566 	testval = pci_mapbase(testval);
567 	ln2size = 0;
568 	if (testval != 0) {
569 		while ((testval & 1) == 0)
570 		{
571 			ln2size++;
572 			testval >>= 1;
573 		}
574 	}
575 	return (ln2size);
576 }
577 
578 /* return base address of device ROM */
579 
580 static pci_addr_t
581 pci_rombase(uint64_t mapreg)
582 {
583 
584 	return (mapreg & PCIM_BIOS_ADDR_MASK);
585 }
586 
587 /* return log2 of map size decided for device ROM */
588 
589 static int
590 pci_romsize(uint64_t testval)
591 {
592 	int ln2size;
593 
594 	testval = pci_rombase(testval);
595 	ln2size = 0;
596 	if (testval != 0) {
597 		while ((testval & 1) == 0)
598 		{
599 			ln2size++;
600 			testval >>= 1;
601 		}
602 	}
603 	return (ln2size);
604 }
605 
606 /* return log2 of address range supported by map register */
607 
608 static int
609 pci_maprange(uint64_t mapreg)
610 {
611 	int ln2range = 0;
612 
613 	if (PCI_BAR_IO(mapreg))
614 		ln2range = 32;
615 	else
616 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
617 		case PCIM_BAR_MEM_32:
618 			ln2range = 32;
619 			break;
620 		case PCIM_BAR_MEM_1MB:
621 			ln2range = 20;
622 			break;
623 		case PCIM_BAR_MEM_64:
624 			ln2range = 64;
625 			break;
626 		}
627 	return (ln2range);
628 }
629 
630 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
631 
632 static void
633 pci_fixancient(pcicfgregs *cfg)
634 {
635 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
636 		return;
637 
638 	/* PCI to PCI bridges use header type 1 */
639 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
640 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
641 }
642 
643 /* extract header type specific config data */
644 
645 static void
646 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
647 {
648 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
649 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
650 	case PCIM_HDRTYPE_NORMAL:
651 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
652 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
653 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
654 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
655 		cfg->nummaps	    = PCI_MAXMAPS_0;
656 		break;
657 	case PCIM_HDRTYPE_BRIDGE:
658 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
659 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
660 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
661 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
662 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
663 		cfg->nummaps	    = PCI_MAXMAPS_1;
664 		break;
665 	case PCIM_HDRTYPE_CARDBUS:
666 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
667 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
668 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
669 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
670 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
671 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
672 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
673 		cfg->nummaps	    = PCI_MAXMAPS_2;
674 		break;
675 	}
676 #undef REG
677 }
678 
679 /* read configuration header into pcicfgregs structure */
680 struct pci_devinfo *
681 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
682 {
683 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
684 	uint16_t vid, did;
685 
686 	vid = REG(PCIR_VENDOR, 2);
687 	if (vid == PCIV_INVALID)
688 		return (NULL);
689 
690 	did = REG(PCIR_DEVICE, 2);
691 
692 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
693 }
694 
695 struct pci_devinfo *
696 pci_alloc_devinfo_method(device_t dev)
697 {
698 
699 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
700 	    M_WAITOK | M_ZERO));
701 }
702 
703 static struct pci_devinfo *
704 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
705     uint16_t vid, uint16_t did)
706 {
707 	struct pci_devinfo *devlist_entry;
708 	pcicfgregs *cfg;
709 
710 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
711 
712 	cfg = &devlist_entry->cfg;
713 
714 	cfg->domain		= d;
715 	cfg->bus		= b;
716 	cfg->slot		= s;
717 	cfg->func		= f;
718 	cfg->vendor		= vid;
719 	cfg->device		= did;
720 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
721 	cfg->statreg		= REG(PCIR_STATUS, 2);
722 	cfg->baseclass		= REG(PCIR_CLASS, 1);
723 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
724 	cfg->progif		= REG(PCIR_PROGIF, 1);
725 	cfg->revid		= REG(PCIR_REVID, 1);
726 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
727 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
728 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
729 	cfg->intpin		= REG(PCIR_INTPIN, 1);
730 	cfg->intline		= REG(PCIR_INTLINE, 1);
731 
732 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
733 	cfg->hdrtype		&= ~PCIM_MFDEV;
734 	STAILQ_INIT(&cfg->maps);
735 
736 	cfg->iov		= NULL;
737 
738 	pci_fixancient(cfg);
739 	pci_hdrtypedata(pcib, b, s, f, cfg);
740 
741 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
742 		pci_read_cap(pcib, cfg);
743 
744 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
745 
746 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
747 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
748 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
749 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
750 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
751 
752 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
753 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
754 	devlist_entry->conf.pc_vendor = cfg->vendor;
755 	devlist_entry->conf.pc_device = cfg->device;
756 
757 	devlist_entry->conf.pc_class = cfg->baseclass;
758 	devlist_entry->conf.pc_subclass = cfg->subclass;
759 	devlist_entry->conf.pc_progif = cfg->progif;
760 	devlist_entry->conf.pc_revid = cfg->revid;
761 
762 	pci_numdevs++;
763 	pci_generation++;
764 
765 	return (devlist_entry);
766 }
767 #undef REG
768 
769 static void
770 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
771 {
772 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
773     cfg->ea.ea_location + (n), w)
774 	int num_ent;
775 	int ptr;
776 	int a, b;
777 	uint32_t val;
778 	int ent_size;
779 	uint32_t dw[4];
780 	uint64_t base, max_offset;
781 	struct pci_ea_entry *eae;
782 
783 	if (cfg->ea.ea_location == 0)
784 		return;
785 
786 	STAILQ_INIT(&cfg->ea.ea_entries);
787 
788 	/* Determine the number of entries */
789 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
790 	num_ent &= PCIM_EA_NUM_ENT_MASK;
791 
792 	/* Find the first entry to care of */
793 	ptr = PCIR_EA_FIRST_ENT;
794 
795 	/* Skip DWORD 2 for type 1 functions */
796 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
797 		ptr += 4;
798 
799 	for (a = 0; a < num_ent; a++) {
800 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
801 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
802 
803 		/* Read a number of dwords in the entry */
804 		val = REG(ptr, 4);
805 		ptr += 4;
806 		ent_size = (val & PCIM_EA_ES);
807 
808 		for (b = 0; b < ent_size; b++) {
809 			dw[b] = REG(ptr, 4);
810 			ptr += 4;
811 		}
812 
813 		eae->eae_flags = val;
814 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
815 
816 		base = dw[0] & PCIM_EA_FIELD_MASK;
817 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
818 		b = 2;
819 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
820 			base |= (uint64_t)dw[b] << 32UL;
821 			b++;
822 		}
823 		if (((dw[1] & PCIM_EA_IS_64) != 0)
824 		    && (b < ent_size)) {
825 			max_offset |= (uint64_t)dw[b] << 32UL;
826 			b++;
827 		}
828 
829 		eae->eae_base = base;
830 		eae->eae_max_offset = max_offset;
831 
832 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
833 
834 		if (bootverbose) {
835 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
836 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
837 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
838 		}
839 	}
840 }
841 #undef REG
842 
843 static void
844 pci_read_cap(device_t pcib, pcicfgregs *cfg)
845 {
846 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
847 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
848 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
849 	uint64_t addr;
850 #endif
851 	uint32_t val;
852 	int	ptr, nextptr, ptrptr;
853 
854 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
855 	case PCIM_HDRTYPE_NORMAL:
856 	case PCIM_HDRTYPE_BRIDGE:
857 		ptrptr = PCIR_CAP_PTR;
858 		break;
859 	case PCIM_HDRTYPE_CARDBUS:
860 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
861 		break;
862 	default:
863 		return;		/* no extended capabilities support */
864 	}
865 	nextptr = REG(ptrptr, 1);	/* sanity check? */
866 
867 	/*
868 	 * Read capability entries.
869 	 */
870 	while (nextptr != 0) {
871 		/* Sanity check */
872 		if (nextptr > 255) {
873 			printf("illegal PCI extended capability offset %d\n",
874 			    nextptr);
875 			return;
876 		}
877 		/* Find the next entry */
878 		ptr = nextptr;
879 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
880 
881 		/* Process this entry */
882 		switch (REG(ptr + PCICAP_ID, 1)) {
883 		case PCIY_PMG:		/* PCI power management */
884 			if (cfg->pp.pp_cap == 0) {
885 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
886 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
887 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
888 				if ((nextptr - ptr) > PCIR_POWER_DATA)
889 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
890 			}
891 			break;
892 		case PCIY_HT:		/* HyperTransport */
893 			/* Determine HT-specific capability type. */
894 			val = REG(ptr + PCIR_HT_COMMAND, 2);
895 
896 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
897 				cfg->ht.ht_slave = ptr;
898 
899 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
900 			switch (val & PCIM_HTCMD_CAP_MASK) {
901 			case PCIM_HTCAP_MSI_MAPPING:
902 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
903 					/* Sanity check the mapping window. */
904 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
905 					    4);
906 					addr <<= 32;
907 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
908 					    4);
909 					if (addr != MSI_INTEL_ADDR_BASE)
910 						device_printf(pcib,
911 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
912 						    cfg->domain, cfg->bus,
913 						    cfg->slot, cfg->func,
914 						    (long long)addr);
915 				} else
916 					addr = MSI_INTEL_ADDR_BASE;
917 
918 				cfg->ht.ht_msimap = ptr;
919 				cfg->ht.ht_msictrl = val;
920 				cfg->ht.ht_msiaddr = addr;
921 				break;
922 			}
923 #endif
924 			break;
925 		case PCIY_MSI:		/* PCI MSI */
926 			cfg->msi.msi_location = ptr;
927 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
928 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
929 						     PCIM_MSICTRL_MMC_MASK)>>1);
930 			break;
931 		case PCIY_MSIX:		/* PCI MSI-X */
932 			cfg->msix.msix_location = ptr;
933 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
934 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
935 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
936 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
937 			cfg->msix.msix_table_bar = PCIR_BAR(val &
938 			    PCIM_MSIX_BIR_MASK);
939 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
940 			val = REG(ptr + PCIR_MSIX_PBA, 4);
941 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
942 			    PCIM_MSIX_BIR_MASK);
943 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
944 			break;
945 		case PCIY_VPD:		/* PCI Vital Product Data */
946 			cfg->vpd.vpd_reg = ptr;
947 			break;
948 		case PCIY_SUBVENDOR:
949 			/* Should always be true. */
950 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
951 			    PCIM_HDRTYPE_BRIDGE) {
952 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
953 				cfg->subvendor = val & 0xffff;
954 				cfg->subdevice = val >> 16;
955 			}
956 			break;
957 		case PCIY_PCIX:		/* PCI-X */
958 			/*
959 			 * Assume we have a PCI-X chipset if we have
960 			 * at least one PCI-PCI bridge with a PCI-X
961 			 * capability.  Note that some systems with
962 			 * PCI-express or HT chipsets might match on
963 			 * this check as well.
964 			 */
965 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
966 			    PCIM_HDRTYPE_BRIDGE)
967 				pcix_chipset = 1;
968 			cfg->pcix.pcix_location = ptr;
969 			break;
970 		case PCIY_EXPRESS:	/* PCI-express */
971 			/*
972 			 * Assume we have a PCI-express chipset if we have
973 			 * at least one PCI-express device.
974 			 */
975 			pcie_chipset = 1;
976 			cfg->pcie.pcie_location = ptr;
977 			val = REG(ptr + PCIER_FLAGS, 2);
978 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
979 			break;
980 		case PCIY_EA:		/* Enhanced Allocation */
981 			cfg->ea.ea_location = ptr;
982 			pci_ea_fill_info(pcib, cfg);
983 			break;
984 		default:
985 			break;
986 		}
987 	}
988 
989 #if defined(__powerpc__)
990 	/*
991 	 * Enable the MSI mapping window for all HyperTransport
992 	 * slaves.  PCI-PCI bridges have their windows enabled via
993 	 * PCIB_MAP_MSI().
994 	 */
995 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
996 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
997 		device_printf(pcib,
998 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
999 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1000 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1001 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1002 		     2);
1003 	}
1004 #endif
1005 /* REG and WREG use carry through to next functions */
1006 }
1007 
1008 /*
1009  * PCI Vital Product Data
1010  */
1011 
1012 #define	PCI_VPD_TIMEOUT		1000000
1013 
1014 static int
1015 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1016 {
1017 	int count = PCI_VPD_TIMEOUT;
1018 
1019 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1020 
1021 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1022 
1023 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1024 		if (--count < 0)
1025 			return (ENXIO);
1026 		DELAY(1);	/* limit looping */
1027 	}
1028 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1029 
1030 	return (0);
1031 }
1032 
1033 #if 0
1034 static int
1035 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1036 {
1037 	int count = PCI_VPD_TIMEOUT;
1038 
1039 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1040 
1041 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1042 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1043 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1044 		if (--count < 0)
1045 			return (ENXIO);
1046 		DELAY(1);	/* limit looping */
1047 	}
1048 
1049 	return (0);
1050 }
1051 #endif
1052 
1053 #undef PCI_VPD_TIMEOUT
1054 
1055 struct vpd_readstate {
1056 	device_t	pcib;
1057 	pcicfgregs	*cfg;
1058 	uint32_t	val;
1059 	int		bytesinval;
1060 	int		off;
1061 	uint8_t		cksum;
1062 };
1063 
1064 /* return 0 and one byte in *data if no read error, -1 else */
1065 static int
1066 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1067 {
1068 	uint32_t reg;
1069 	uint8_t byte;
1070 
1071 	if (vrs->bytesinval == 0) {
1072 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1073 			return (-1);
1074 		vrs->val = le32toh(reg);
1075 		vrs->off += 4;
1076 		byte = vrs->val & 0xff;
1077 		vrs->bytesinval = 3;
1078 	} else {
1079 		vrs->val = vrs->val >> 8;
1080 		byte = vrs->val & 0xff;
1081 		vrs->bytesinval--;
1082 	}
1083 
1084 	vrs->cksum += byte;
1085 	*data = byte;
1086 	return (0);
1087 }
1088 
1089 /* return 0 on match, -1 and "unget" byte on no match */
1090 static int
1091 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1092 {
1093 	uint8_t data;
1094 
1095 	if (vpd_nextbyte(vrs, &data) != 0)
1096 		return (-1);
1097 
1098 	if (data == expected)
1099 		return (0);
1100 
1101 	vrs->cksum -= data;
1102 	vrs->val = (vrs->val << 8) + data;
1103 	vrs->bytesinval++;
1104 	return (-1);
1105 }
1106 
1107 /* return size if tag matches, -1 on no match, -2 on read error */
1108 static int
1109 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1110 {
1111 	uint8_t byte1, byte2;
1112 
1113 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1114 		return (-1);
1115 
1116 	if ((vpd_tag & 0x80) == 0)
1117 		return (vpd_tag & 0x07);
1118 
1119 	if (vpd_nextbyte(vrs, &byte1) != 0)
1120 		return (-2);
1121 	if (vpd_nextbyte(vrs, &byte2) != 0)
1122 		return (-2);
1123 
1124 	return ((byte2 << 8) + byte1);
1125 }
1126 
1127 /* (re)allocate buffer in multiples of 8 elements */
1128 static void*
1129 alloc_buffer(void* buffer, size_t element_size, int needed)
1130 {
1131 	int alloc, new_alloc;
1132 
1133 	alloc = roundup2(needed, 8);
1134 	new_alloc = roundup2(needed + 1, 8);
1135 	if (alloc != new_alloc) {
1136 		buffer = reallocf(buffer,
1137 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1138 	}
1139 
1140 	return (buffer);
1141 }
1142 
1143 /* read VPD keyword and return element size, return -1 on read error */
1144 static int
1145 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1146 {
1147 	uint8_t data;
1148 
1149 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1150 		return (-1);
1151 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1152 		return (-1);
1153 	if (vpd_nextbyte(vrs, &data) != 0)
1154 		return (-1);
1155 
1156 	return (data);
1157 }
1158 
1159 /* read VPD data element of given size into allocated buffer */
1160 static char *
1161 vpd_read_value(struct vpd_readstate *vrs, int size)
1162 {
1163 	int i;
1164 	char char1;
1165 	char *value;
1166 
1167 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1168 	for (i = 0; i < size; i++) {
1169 		if (vpd_nextbyte(vrs, &char1) != 0) {
1170 			free(value, M_DEVBUF);
1171 			return (NULL);
1172 		}
1173 		value[i] = char1;
1174 	}
1175 	value[size] = '\0';
1176 
1177 	return (value);
1178 }
1179 
1180 /* read VPD into *keyword and *value, return length of data element */
1181 static int
1182 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1183 {
1184 	int len;
1185 
1186 	len = vpd_read_elem_head(vrs, keyword);
1187 	if (len > maxlen)
1188 		return (-1);
1189 	*value = vpd_read_value(vrs, len);
1190 
1191 	return (len);
1192 }
1193 
1194 /* subtract all data following first byte from checksum of RV element */
1195 static void
1196 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1197 {
1198 	int i;
1199 	uint8_t fixup;
1200 
1201 	fixup = 0;
1202 	for (i = 1; i < len; i++)
1203 		fixup += rvstring[i];
1204 	vrs->cksum -= fixup;
1205 }
1206 
1207 /* fetch one read-only element and return size of heading + data */
1208 static size_t
1209 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1210 {
1211 	struct pcicfg_vpd *vpd;
1212 	pcicfgregs *cfg;
1213 	struct vpd_readonly *vpd_ros;
1214 	int len;
1215 
1216 	cfg = vrs->cfg;
1217 	vpd = &cfg->vpd;
1218 
1219 	if (maxsize < 3)
1220 		return (-1);
1221 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1222 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1223 	maxsize -= 3;
1224 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1225 	if (vpd_ros->value == NULL)
1226 		return (-1);
1227 	vpd_ros->len = len;
1228 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1229 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1230 		if (vrs->cksum != 0) {
1231 			pci_printf(cfg,
1232 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1233 			return (-1);
1234 		}
1235 	}
1236 	vpd->vpd_rocnt++;
1237 
1238 	return (len + 3);
1239 }
1240 
1241 /* fetch one writable element and return size of heading + data */
1242 static size_t
1243 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1244 {
1245 	struct pcicfg_vpd *vpd;
1246 	pcicfgregs *cfg;
1247 	struct vpd_write *vpd_w;
1248 	int len;
1249 
1250 	cfg = vrs->cfg;
1251 	vpd = &cfg->vpd;
1252 
1253 	if (maxsize < 3)
1254 		return (-1);
1255 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1256 	if (vpd->vpd_w == NULL) {
1257 		pci_printf(cfg, "out of memory");
1258 		return (-1);
1259 	}
1260 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1261 	maxsize -= 3;
1262 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1263 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1264 	if (vpd_w->value == NULL)
1265 		return (-1);
1266 	vpd_w->len = len;
1267 	vpd->vpd_wcnt++;
1268 
1269 	return (len + 3);
1270 }
1271 
1272 /* free all memory allocated for VPD data */
1273 static void
1274 vpd_free(struct pcicfg_vpd *vpd)
1275 {
1276 	int i;
1277 
1278 	free(vpd->vpd_ident, M_DEVBUF);
1279 	for (i = 0; i < vpd->vpd_rocnt; i++)
1280 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1281 	free(vpd->vpd_ros, M_DEVBUF);
1282 	vpd->vpd_rocnt = 0;
1283 	for (i = 0; i < vpd->vpd_wcnt; i++)
1284 		free(vpd->vpd_w[i].value, M_DEVBUF);
1285 	free(vpd->vpd_w, M_DEVBUF);
1286 	vpd->vpd_wcnt = 0;
1287 }
1288 
1289 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1290 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1291 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1292 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1293 
1294 static int
1295 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1296 {
1297 	struct vpd_readstate vrs;
1298 	int cksumvalid;
1299 	int size, elem_size;
1300 
1301 	/* init vpd reader */
1302 	vrs.bytesinval = 0;
1303 	vrs.off = 0;
1304 	vrs.pcib = pcib;
1305 	vrs.cfg = cfg;
1306 	vrs.cksum = 0;
1307 
1308 	/* read VPD ident element - mandatory */
1309 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1310 	if (size <= 0) {
1311 		pci_printf(cfg, "no VPD ident found\n");
1312 		return (0);
1313 	}
1314 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1315 	if (cfg->vpd.vpd_ident == NULL) {
1316 		pci_printf(cfg, "error accessing VPD ident data\n");
1317 		return (0);
1318 	}
1319 
1320 	/* read VPD RO elements - mandatory */
1321 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1322 	if (size <= 0) {
1323 		pci_printf(cfg, "no read-only VPD data found\n");
1324 		return (0);
1325 	}
1326 	while (size > 0) {
1327 		elem_size = next_vpd_ro_elem(&vrs, size);
1328 		if (elem_size < 0) {
1329 			pci_printf(cfg, "error accessing read-only VPD data\n");
1330 			return (-1);
1331 		}
1332 		size -= elem_size;
1333 	}
1334 	cksumvalid = (vrs.cksum == 0);
1335 	if (!cksumvalid)
1336 		return (-1);
1337 
1338 	/* read VPD RW elements - optional */
1339 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1340 	if (size == -2)
1341 		return (-1);
1342 	while (size > 0) {
1343 		elem_size = next_vpd_rw_elem(&vrs, size);
1344 		if (elem_size < 0) {
1345 			pci_printf(cfg, "error accessing writeable VPD data\n");
1346 			return (-1);
1347 		}
1348 		size -= elem_size;
1349 	}
1350 
1351 	/* read empty END tag - mandatory */
1352 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1353 	if (size != 0) {
1354 		pci_printf(cfg, "No valid VPD end tag found\n");
1355 	}
1356 	return (0);
1357 }
1358 
1359 static void
1360 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1361 {
1362 	int status;
1363 
1364 	status = pci_parse_vpd(pcib, cfg);
1365 	if (status < 0)
1366 		vpd_free(&cfg->vpd);
1367 	cfg->vpd.vpd_cached = 1;
1368 #undef REG
1369 #undef WREG
1370 }
1371 
1372 int
1373 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1374 {
1375 	struct pci_devinfo *dinfo = device_get_ivars(child);
1376 	pcicfgregs *cfg = &dinfo->cfg;
1377 
1378 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1379 		pci_read_vpd(device_get_parent(dev), cfg);
1380 
1381 	*identptr = cfg->vpd.vpd_ident;
1382 
1383 	if (*identptr == NULL)
1384 		return (ENXIO);
1385 
1386 	return (0);
1387 }
1388 
1389 int
1390 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1391 	const char **vptr)
1392 {
1393 	struct pci_devinfo *dinfo = device_get_ivars(child);
1394 	pcicfgregs *cfg = &dinfo->cfg;
1395 	int i;
1396 
1397 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1398 		pci_read_vpd(device_get_parent(dev), cfg);
1399 
1400 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1401 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1402 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1403 			*vptr = cfg->vpd.vpd_ros[i].value;
1404 			return (0);
1405 		}
1406 
1407 	*vptr = NULL;
1408 	return (ENXIO);
1409 }
1410 
1411 struct pcicfg_vpd *
1412 pci_fetch_vpd_list(device_t dev)
1413 {
1414 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1415 	pcicfgregs *cfg = &dinfo->cfg;
1416 
1417 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1418 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1419 	return (&cfg->vpd);
1420 }
1421 
1422 /*
1423  * Find the requested HyperTransport capability and return the offset
1424  * in configuration space via the pointer provided.  The function
1425  * returns 0 on success and an error code otherwise.
1426  */
1427 int
1428 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1429 {
1430 	int ptr, error;
1431 	uint16_t val;
1432 
1433 	error = pci_find_cap(child, PCIY_HT, &ptr);
1434 	if (error)
1435 		return (error);
1436 
1437 	/*
1438 	 * Traverse the capabilities list checking each HT capability
1439 	 * to see if it matches the requested HT capability.
1440 	 */
1441 	for (;;) {
1442 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1443 		if (capability == PCIM_HTCAP_SLAVE ||
1444 		    capability == PCIM_HTCAP_HOST)
1445 			val &= 0xe000;
1446 		else
1447 			val &= PCIM_HTCMD_CAP_MASK;
1448 		if (val == capability) {
1449 			if (capreg != NULL)
1450 				*capreg = ptr;
1451 			return (0);
1452 		}
1453 
1454 		/* Skip to the next HT capability. */
1455 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1456 			break;
1457 	}
1458 
1459 	return (ENOENT);
1460 }
1461 
1462 /*
1463  * Find the next requested HyperTransport capability after start and return
1464  * the offset in configuration space via the pointer provided.  The function
1465  * returns 0 on success and an error code otherwise.
1466  */
1467 int
1468 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1469     int start, int *capreg)
1470 {
1471 	int ptr;
1472 	uint16_t val;
1473 
1474 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1475 	    ("start capability is not HyperTransport capability"));
1476 	ptr = start;
1477 
1478 	/*
1479 	 * Traverse the capabilities list checking each HT capability
1480 	 * to see if it matches the requested HT capability.
1481 	 */
1482 	for (;;) {
1483 		/* Skip to the next HT capability. */
1484 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1485 			break;
1486 
1487 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1488 		if (capability == PCIM_HTCAP_SLAVE ||
1489 		    capability == PCIM_HTCAP_HOST)
1490 			val &= 0xe000;
1491 		else
1492 			val &= PCIM_HTCMD_CAP_MASK;
1493 		if (val == capability) {
1494 			if (capreg != NULL)
1495 				*capreg = ptr;
1496 			return (0);
1497 		}
1498 	}
1499 
1500 	return (ENOENT);
1501 }
1502 
1503 /*
1504  * Find the requested capability and return the offset in
1505  * configuration space via the pointer provided.  The function returns
1506  * 0 on success and an error code otherwise.
1507  */
1508 int
1509 pci_find_cap_method(device_t dev, device_t child, int capability,
1510     int *capreg)
1511 {
1512 	struct pci_devinfo *dinfo = device_get_ivars(child);
1513 	pcicfgregs *cfg = &dinfo->cfg;
1514 	uint32_t status;
1515 	uint8_t ptr;
1516 
1517 	/*
1518 	 * Check the CAP_LIST bit of the PCI status register first.
1519 	 */
1520 	status = pci_read_config(child, PCIR_STATUS, 2);
1521 	if (!(status & PCIM_STATUS_CAPPRESENT))
1522 		return (ENXIO);
1523 
1524 	/*
1525 	 * Determine the start pointer of the capabilities list.
1526 	 */
1527 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1528 	case PCIM_HDRTYPE_NORMAL:
1529 	case PCIM_HDRTYPE_BRIDGE:
1530 		ptr = PCIR_CAP_PTR;
1531 		break;
1532 	case PCIM_HDRTYPE_CARDBUS:
1533 		ptr = PCIR_CAP_PTR_2;
1534 		break;
1535 	default:
1536 		/* XXX: panic? */
1537 		return (ENXIO);		/* no extended capabilities support */
1538 	}
1539 	ptr = pci_read_config(child, ptr, 1);
1540 
1541 	/*
1542 	 * Traverse the capabilities list.
1543 	 */
1544 	while (ptr != 0) {
1545 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1546 			if (capreg != NULL)
1547 				*capreg = ptr;
1548 			return (0);
1549 		}
1550 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1551 	}
1552 
1553 	return (ENOENT);
1554 }
1555 
1556 /*
1557  * Find the next requested capability after start and return the offset in
1558  * configuration space via the pointer provided.  The function returns
1559  * 0 on success and an error code otherwise.
1560  */
1561 int
1562 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1563     int start, int *capreg)
1564 {
1565 	uint8_t ptr;
1566 
1567 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1568 	    ("start capability is not expected capability"));
1569 
1570 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1571 	while (ptr != 0) {
1572 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1573 			if (capreg != NULL)
1574 				*capreg = ptr;
1575 			return (0);
1576 		}
1577 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1578 	}
1579 
1580 	return (ENOENT);
1581 }
1582 
1583 /*
1584  * Find the requested extended capability and return the offset in
1585  * configuration space via the pointer provided.  The function returns
1586  * 0 on success and an error code otherwise.
1587  */
1588 int
1589 pci_find_extcap_method(device_t dev, device_t child, int capability,
1590     int *capreg)
1591 {
1592 	struct pci_devinfo *dinfo = device_get_ivars(child);
1593 	pcicfgregs *cfg = &dinfo->cfg;
1594 	uint32_t ecap;
1595 	uint16_t ptr;
1596 
1597 	/* Only supported for PCI-express devices. */
1598 	if (cfg->pcie.pcie_location == 0)
1599 		return (ENXIO);
1600 
1601 	ptr = PCIR_EXTCAP;
1602 	ecap = pci_read_config(child, ptr, 4);
1603 	if (ecap == 0xffffffff || ecap == 0)
1604 		return (ENOENT);
1605 	for (;;) {
1606 		if (PCI_EXTCAP_ID(ecap) == capability) {
1607 			if (capreg != NULL)
1608 				*capreg = ptr;
1609 			return (0);
1610 		}
1611 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1612 		if (ptr == 0)
1613 			break;
1614 		ecap = pci_read_config(child, ptr, 4);
1615 	}
1616 
1617 	return (ENOENT);
1618 }
1619 
1620 /*
1621  * Find the next requested extended capability after start and return the
1622  * offset in configuration space via the pointer provided.  The function
1623  * returns 0 on success and an error code otherwise.
1624  */
1625 int
1626 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1627     int start, int *capreg)
1628 {
1629 	struct pci_devinfo *dinfo = device_get_ivars(child);
1630 	pcicfgregs *cfg = &dinfo->cfg;
1631 	uint32_t ecap;
1632 	uint16_t ptr;
1633 
1634 	/* Only supported for PCI-express devices. */
1635 	if (cfg->pcie.pcie_location == 0)
1636 		return (ENXIO);
1637 
1638 	ecap = pci_read_config(child, start, 4);
1639 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1640 	    ("start extended capability is not expected capability"));
1641 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1642 	while (ptr != 0) {
1643 		ecap = pci_read_config(child, ptr, 4);
1644 		if (PCI_EXTCAP_ID(ecap) == capability) {
1645 			if (capreg != NULL)
1646 				*capreg = ptr;
1647 			return (0);
1648 		}
1649 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1650 	}
1651 
1652 	return (ENOENT);
1653 }
1654 
1655 /*
1656  * Support for MSI-X message interrupts.
1657  */
1658 static void
1659 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1660 {
1661 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1662 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1663 	uint32_t offset;
1664 
1665 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1666 	offset = msix->msix_table_offset + index * 16;
1667 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1668 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1669 	bus_write_4(msix->msix_table_res, offset + 8, data);
1670 }
1671 
1672 void
1673 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1674     uint64_t address, uint32_t data)
1675 {
1676 
1677 	if (pci_msix_rewrite_table) {
1678 		struct pci_devinfo *dinfo = device_get_ivars(child);
1679 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1680 
1681 		/*
1682 		 * Some VM hosts require MSIX to be disabled in the
1683 		 * control register before updating the MSIX table
1684 		 * entries are allowed. It is not enough to only
1685 		 * disable MSIX while updating a single entry. MSIX
1686 		 * must be disabled while updating all entries in the
1687 		 * table.
1688 		 */
1689 		pci_write_config(child,
1690 		    msix->msix_location + PCIR_MSIX_CTRL,
1691 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1692 		pci_resume_msix(child);
1693 	} else
1694 		pci_write_msix_entry(child, index, address, data);
1695 
1696 	/* Enable MSI -> HT mapping. */
1697 	pci_ht_map_msi(child, address);
1698 }
1699 
1700 void
1701 pci_mask_msix(device_t dev, u_int index)
1702 {
1703 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1704 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1705 	uint32_t offset, val;
1706 
1707 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1708 	offset = msix->msix_table_offset + index * 16 + 12;
1709 	val = bus_read_4(msix->msix_table_res, offset);
1710 	val |= PCIM_MSIX_VCTRL_MASK;
1711 
1712 	/*
1713 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1714 	 * register, so always write the new value.
1715 	 */
1716 	bus_write_4(msix->msix_table_res, offset, val);
1717 }
1718 
1719 void
1720 pci_unmask_msix(device_t dev, u_int index)
1721 {
1722 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1723 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1724 	uint32_t offset, val;
1725 
1726 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1727 	offset = msix->msix_table_offset + index * 16 + 12;
1728 	val = bus_read_4(msix->msix_table_res, offset);
1729 	val &= ~PCIM_MSIX_VCTRL_MASK;
1730 
1731 	/*
1732 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1733 	 * register, so always write the new value.
1734 	 */
1735 	bus_write_4(msix->msix_table_res, offset, val);
1736 }
1737 
1738 int
1739 pci_pending_msix(device_t dev, u_int index)
1740 {
1741 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1742 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1743 	uint32_t offset, bit;
1744 
1745 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1746 	offset = msix->msix_pba_offset + (index / 32) * 4;
1747 	bit = 1 << index % 32;
1748 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1749 }
1750 
1751 /*
1752  * Restore MSI-X registers and table during resume.  If MSI-X is
1753  * enabled then walk the virtual table to restore the actual MSI-X
1754  * table.
1755  */
1756 static void
1757 pci_resume_msix(device_t dev)
1758 {
1759 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1760 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1761 	struct msix_table_entry *mte;
1762 	struct msix_vector *mv;
1763 	int i;
1764 
1765 	if (msix->msix_alloc > 0) {
1766 		/* First, mask all vectors. */
1767 		for (i = 0; i < msix->msix_msgnum; i++)
1768 			pci_mask_msix(dev, i);
1769 
1770 		/* Second, program any messages with at least one handler. */
1771 		for (i = 0; i < msix->msix_table_len; i++) {
1772 			mte = &msix->msix_table[i];
1773 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1774 				continue;
1775 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1776 			pci_write_msix_entry(dev, i, mv->mv_address,
1777 			    mv->mv_data);
1778 			pci_unmask_msix(dev, i);
1779 		}
1780 	}
1781 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1782 	    msix->msix_ctrl, 2);
1783 }
1784 
1785 /*
1786  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1787  * returned in *count.  After this function returns, each message will be
1788  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1789  */
1790 int
1791 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1792 {
1793 	struct pci_devinfo *dinfo = device_get_ivars(child);
1794 	pcicfgregs *cfg = &dinfo->cfg;
1795 	struct resource_list_entry *rle;
1796 	int actual, error, i, irq, max;
1797 
1798 	/* Don't let count == 0 get us into trouble. */
1799 	if (*count == 0)
1800 		return (EINVAL);
1801 
1802 	/* If rid 0 is allocated, then fail. */
1803 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1804 	if (rle != NULL && rle->res != NULL)
1805 		return (ENXIO);
1806 
1807 	/* Already have allocated messages? */
1808 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1809 		return (ENXIO);
1810 
1811 	/* If MSI-X is blacklisted for this system, fail. */
1812 	if (pci_msix_blacklisted())
1813 		return (ENXIO);
1814 
1815 	/* MSI-X capability present? */
1816 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1817 		return (ENODEV);
1818 
1819 	/* Make sure the appropriate BARs are mapped. */
1820 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1821 	    cfg->msix.msix_table_bar);
1822 	if (rle == NULL || rle->res == NULL ||
1823 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1824 		return (ENXIO);
1825 	cfg->msix.msix_table_res = rle->res;
1826 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1827 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1828 		    cfg->msix.msix_pba_bar);
1829 		if (rle == NULL || rle->res == NULL ||
1830 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1831 			return (ENXIO);
1832 	}
1833 	cfg->msix.msix_pba_res = rle->res;
1834 
1835 	if (bootverbose)
1836 		device_printf(child,
1837 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1838 		    *count, cfg->msix.msix_msgnum);
1839 	max = min(*count, cfg->msix.msix_msgnum);
1840 	for (i = 0; i < max; i++) {
1841 		/* Allocate a message. */
1842 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1843 		if (error) {
1844 			if (i == 0)
1845 				return (error);
1846 			break;
1847 		}
1848 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1849 		    irq, 1);
1850 	}
1851 	actual = i;
1852 
1853 	if (bootverbose) {
1854 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1855 		if (actual == 1)
1856 			device_printf(child, "using IRQ %ju for MSI-X\n",
1857 			    rle->start);
1858 		else {
1859 			int run;
1860 
1861 			/*
1862 			 * Be fancy and try to print contiguous runs of
1863 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1864 			 * 'run' is true if we are in a range.
1865 			 */
1866 			device_printf(child, "using IRQs %ju", rle->start);
1867 			irq = rle->start;
1868 			run = 0;
1869 			for (i = 1; i < actual; i++) {
1870 				rle = resource_list_find(&dinfo->resources,
1871 				    SYS_RES_IRQ, i + 1);
1872 
1873 				/* Still in a run? */
1874 				if (rle->start == irq + 1) {
1875 					run = 1;
1876 					irq++;
1877 					continue;
1878 				}
1879 
1880 				/* Finish previous range. */
1881 				if (run) {
1882 					printf("-%d", irq);
1883 					run = 0;
1884 				}
1885 
1886 				/* Start new range. */
1887 				printf(",%ju", rle->start);
1888 				irq = rle->start;
1889 			}
1890 
1891 			/* Unfinished range? */
1892 			if (run)
1893 				printf("-%d", irq);
1894 			printf(" for MSI-X\n");
1895 		}
1896 	}
1897 
1898 	/* Mask all vectors. */
1899 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1900 		pci_mask_msix(child, i);
1901 
1902 	/* Allocate and initialize vector data and virtual table. */
1903 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1904 	    M_DEVBUF, M_WAITOK | M_ZERO);
1905 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1906 	    M_DEVBUF, M_WAITOK | M_ZERO);
1907 	for (i = 0; i < actual; i++) {
1908 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1909 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1910 		cfg->msix.msix_table[i].mte_vector = i + 1;
1911 	}
1912 
1913 	/* Update control register to enable MSI-X. */
1914 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1915 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1916 	    cfg->msix.msix_ctrl, 2);
1917 
1918 	/* Update counts of alloc'd messages. */
1919 	cfg->msix.msix_alloc = actual;
1920 	cfg->msix.msix_table_len = actual;
1921 	*count = actual;
1922 	return (0);
1923 }
1924 
1925 /*
1926  * By default, pci_alloc_msix() will assign the allocated IRQ
1927  * resources consecutively to the first N messages in the MSI-X table.
1928  * However, device drivers may want to use different layouts if they
1929  * either receive fewer messages than they asked for, or they wish to
1930  * populate the MSI-X table sparsely.  This method allows the driver
1931  * to specify what layout it wants.  It must be called after a
1932  * successful pci_alloc_msix() but before any of the associated
1933  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1934  *
1935  * The 'vectors' array contains 'count' message vectors.  The array
1936  * maps directly to the MSI-X table in that index 0 in the array
1937  * specifies the vector for the first message in the MSI-X table, etc.
1938  * The vector value in each array index can either be 0 to indicate
1939  * that no vector should be assigned to a message slot, or it can be a
1940  * number from 1 to N (where N is the count returned from a
1941  * succcessful call to pci_alloc_msix()) to indicate which message
1942  * vector (IRQ) to be used for the corresponding message.
1943  *
1944  * On successful return, each message with a non-zero vector will have
1945  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1946  * 1.  Additionally, if any of the IRQs allocated via the previous
1947  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1948  * will be freed back to the system automatically.
1949  *
1950  * For example, suppose a driver has a MSI-X table with 6 messages and
1951  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1952  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1953  * C.  After the call to pci_alloc_msix(), the device will be setup to
1954  * have an MSI-X table of ABC--- (where - means no vector assigned).
1955  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1956  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1957  * be freed back to the system.  This device will also have valid
1958  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1959  *
1960  * In any case, the SYS_RES_IRQ rid X will always map to the message
1961  * at MSI-X table index X - 1 and will only be valid if a vector is
1962  * assigned to that table entry.
1963  */
1964 int
1965 pci_remap_msix_method(device_t dev, device_t child, int count,
1966     const u_int *vectors)
1967 {
1968 	struct pci_devinfo *dinfo = device_get_ivars(child);
1969 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1970 	struct resource_list_entry *rle;
1971 	int i, irq, j, *used;
1972 
1973 	/*
1974 	 * Have to have at least one message in the table but the
1975 	 * table can't be bigger than the actual MSI-X table in the
1976 	 * device.
1977 	 */
1978 	if (count == 0 || count > msix->msix_msgnum)
1979 		return (EINVAL);
1980 
1981 	/* Sanity check the vectors. */
1982 	for (i = 0; i < count; i++)
1983 		if (vectors[i] > msix->msix_alloc)
1984 			return (EINVAL);
1985 
1986 	/*
1987 	 * Make sure there aren't any holes in the vectors to be used.
1988 	 * It's a big pain to support it, and it doesn't really make
1989 	 * sense anyway.  Also, at least one vector must be used.
1990 	 */
1991 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1992 	    M_ZERO);
1993 	for (i = 0; i < count; i++)
1994 		if (vectors[i] != 0)
1995 			used[vectors[i] - 1] = 1;
1996 	for (i = 0; i < msix->msix_alloc - 1; i++)
1997 		if (used[i] == 0 && used[i + 1] == 1) {
1998 			free(used, M_DEVBUF);
1999 			return (EINVAL);
2000 		}
2001 	if (used[0] != 1) {
2002 		free(used, M_DEVBUF);
2003 		return (EINVAL);
2004 	}
2005 
2006 	/* Make sure none of the resources are allocated. */
2007 	for (i = 0; i < msix->msix_table_len; i++) {
2008 		if (msix->msix_table[i].mte_vector == 0)
2009 			continue;
2010 		if (msix->msix_table[i].mte_handlers > 0) {
2011 			free(used, M_DEVBUF);
2012 			return (EBUSY);
2013 		}
2014 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2015 		KASSERT(rle != NULL, ("missing resource"));
2016 		if (rle->res != NULL) {
2017 			free(used, M_DEVBUF);
2018 			return (EBUSY);
2019 		}
2020 	}
2021 
2022 	/* Free the existing resource list entries. */
2023 	for (i = 0; i < msix->msix_table_len; i++) {
2024 		if (msix->msix_table[i].mte_vector == 0)
2025 			continue;
2026 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2027 	}
2028 
2029 	/*
2030 	 * Build the new virtual table keeping track of which vectors are
2031 	 * used.
2032 	 */
2033 	free(msix->msix_table, M_DEVBUF);
2034 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
2035 	    M_DEVBUF, M_WAITOK | M_ZERO);
2036 	for (i = 0; i < count; i++)
2037 		msix->msix_table[i].mte_vector = vectors[i];
2038 	msix->msix_table_len = count;
2039 
2040 	/* Free any unused IRQs and resize the vectors array if necessary. */
2041 	j = msix->msix_alloc - 1;
2042 	if (used[j] == 0) {
2043 		struct msix_vector *vec;
2044 
2045 		while (used[j] == 0) {
2046 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2047 			    msix->msix_vectors[j].mv_irq);
2048 			j--;
2049 		}
2050 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2051 		    M_WAITOK);
2052 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2053 		    (j + 1));
2054 		free(msix->msix_vectors, M_DEVBUF);
2055 		msix->msix_vectors = vec;
2056 		msix->msix_alloc = j + 1;
2057 	}
2058 	free(used, M_DEVBUF);
2059 
2060 	/* Map the IRQs onto the rids. */
2061 	for (i = 0; i < count; i++) {
2062 		if (vectors[i] == 0)
2063 			continue;
2064 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2065 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2066 		    irq, 1);
2067 	}
2068 
2069 	if (bootverbose) {
2070 		device_printf(child, "Remapped MSI-X IRQs as: ");
2071 		for (i = 0; i < count; i++) {
2072 			if (i != 0)
2073 				printf(", ");
2074 			if (vectors[i] == 0)
2075 				printf("---");
2076 			else
2077 				printf("%d",
2078 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2079 		}
2080 		printf("\n");
2081 	}
2082 
2083 	return (0);
2084 }
2085 
2086 static int
2087 pci_release_msix(device_t dev, device_t child)
2088 {
2089 	struct pci_devinfo *dinfo = device_get_ivars(child);
2090 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2091 	struct resource_list_entry *rle;
2092 	int i;
2093 
2094 	/* Do we have any messages to release? */
2095 	if (msix->msix_alloc == 0)
2096 		return (ENODEV);
2097 
2098 	/* Make sure none of the resources are allocated. */
2099 	for (i = 0; i < msix->msix_table_len; i++) {
2100 		if (msix->msix_table[i].mte_vector == 0)
2101 			continue;
2102 		if (msix->msix_table[i].mte_handlers > 0)
2103 			return (EBUSY);
2104 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2105 		KASSERT(rle != NULL, ("missing resource"));
2106 		if (rle->res != NULL)
2107 			return (EBUSY);
2108 	}
2109 
2110 	/* Update control register to disable MSI-X. */
2111 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2112 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2113 	    msix->msix_ctrl, 2);
2114 
2115 	/* Free the resource list entries. */
2116 	for (i = 0; i < msix->msix_table_len; i++) {
2117 		if (msix->msix_table[i].mte_vector == 0)
2118 			continue;
2119 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2120 	}
2121 	free(msix->msix_table, M_DEVBUF);
2122 	msix->msix_table_len = 0;
2123 
2124 	/* Release the IRQs. */
2125 	for (i = 0; i < msix->msix_alloc; i++)
2126 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2127 		    msix->msix_vectors[i].mv_irq);
2128 	free(msix->msix_vectors, M_DEVBUF);
2129 	msix->msix_alloc = 0;
2130 	return (0);
2131 }
2132 
2133 /*
2134  * Return the max supported MSI-X messages this device supports.
2135  * Basically, assuming the MD code can alloc messages, this function
2136  * should return the maximum value that pci_alloc_msix() can return.
2137  * Thus, it is subject to the tunables, etc.
2138  */
2139 int
2140 pci_msix_count_method(device_t dev, device_t child)
2141 {
2142 	struct pci_devinfo *dinfo = device_get_ivars(child);
2143 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2144 
2145 	if (pci_do_msix && msix->msix_location != 0)
2146 		return (msix->msix_msgnum);
2147 	return (0);
2148 }
2149 
2150 int
2151 pci_msix_pba_bar_method(device_t dev, device_t child)
2152 {
2153 	struct pci_devinfo *dinfo = device_get_ivars(child);
2154 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2155 
2156 	if (pci_do_msix && msix->msix_location != 0)
2157 		return (msix->msix_pba_bar);
2158 	return (-1);
2159 }
2160 
2161 int
2162 pci_msix_table_bar_method(device_t dev, device_t child)
2163 {
2164 	struct pci_devinfo *dinfo = device_get_ivars(child);
2165 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2166 
2167 	if (pci_do_msix && msix->msix_location != 0)
2168 		return (msix->msix_table_bar);
2169 	return (-1);
2170 }
2171 
2172 /*
2173  * HyperTransport MSI mapping control
2174  */
2175 void
2176 pci_ht_map_msi(device_t dev, uint64_t addr)
2177 {
2178 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2179 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2180 
2181 	if (!ht->ht_msimap)
2182 		return;
2183 
2184 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2185 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2186 		/* Enable MSI -> HT mapping. */
2187 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2188 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2189 		    ht->ht_msictrl, 2);
2190 	}
2191 
2192 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2193 		/* Disable MSI -> HT mapping. */
2194 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2195 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2196 		    ht->ht_msictrl, 2);
2197 	}
2198 }
2199 
2200 int
2201 pci_get_relaxed_ordering_enabled(device_t dev)
2202 {
2203 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2204 	int cap;
2205 	uint16_t val;
2206 
2207 	cap = dinfo->cfg.pcie.pcie_location;
2208 	if (cap == 0)
2209 		return (0);
2210 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2211 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2212 	return (val != 0);
2213 }
2214 
2215 int
2216 pci_get_max_payload(device_t dev)
2217 {
2218 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2219 	int cap;
2220 	uint16_t val;
2221 
2222 	cap = dinfo->cfg.pcie.pcie_location;
2223 	if (cap == 0)
2224 		return (0);
2225 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2226 	val &= PCIEM_CTL_MAX_PAYLOAD;
2227 	val >>= 5;
2228 	return (1 << (val + 7));
2229 }
2230 
2231 int
2232 pci_get_max_read_req(device_t dev)
2233 {
2234 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2235 	int cap;
2236 	uint16_t val;
2237 
2238 	cap = dinfo->cfg.pcie.pcie_location;
2239 	if (cap == 0)
2240 		return (0);
2241 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2242 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2243 	val >>= 12;
2244 	return (1 << (val + 7));
2245 }
2246 
2247 int
2248 pci_set_max_read_req(device_t dev, int size)
2249 {
2250 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2251 	int cap;
2252 	uint16_t val;
2253 
2254 	cap = dinfo->cfg.pcie.pcie_location;
2255 	if (cap == 0)
2256 		return (0);
2257 	if (size < 128)
2258 		size = 128;
2259 	if (size > 4096)
2260 		size = 4096;
2261 	size = (1 << (fls(size) - 1));
2262 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2263 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2264 	val |= (fls(size) - 8) << 12;
2265 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2266 	return (size);
2267 }
2268 
2269 uint32_t
2270 pcie_read_config(device_t dev, int reg, int width)
2271 {
2272 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2273 	int cap;
2274 
2275 	cap = dinfo->cfg.pcie.pcie_location;
2276 	if (cap == 0) {
2277 		if (width == 2)
2278 			return (0xffff);
2279 		return (0xffffffff);
2280 	}
2281 
2282 	return (pci_read_config(dev, cap + reg, width));
2283 }
2284 
2285 void
2286 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2287 {
2288 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2289 	int cap;
2290 
2291 	cap = dinfo->cfg.pcie.pcie_location;
2292 	if (cap == 0)
2293 		return;
2294 	pci_write_config(dev, cap + reg, value, width);
2295 }
2296 
2297 /*
2298  * Adjusts a PCI-e capability register by clearing the bits in mask
2299  * and setting the bits in (value & mask).  Bits not set in mask are
2300  * not adjusted.
2301  *
2302  * Returns the old value on success or all ones on failure.
2303  */
2304 uint32_t
2305 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2306     int width)
2307 {
2308 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2309 	uint32_t old, new;
2310 	int cap;
2311 
2312 	cap = dinfo->cfg.pcie.pcie_location;
2313 	if (cap == 0) {
2314 		if (width == 2)
2315 			return (0xffff);
2316 		return (0xffffffff);
2317 	}
2318 
2319 	old = pci_read_config(dev, cap + reg, width);
2320 	new = old & ~mask;
2321 	new |= (value & mask);
2322 	pci_write_config(dev, cap + reg, new, width);
2323 	return (old);
2324 }
2325 
2326 /*
2327  * Support for MSI message signalled interrupts.
2328  */
2329 void
2330 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2331     uint16_t data)
2332 {
2333 	struct pci_devinfo *dinfo = device_get_ivars(child);
2334 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2335 
2336 	/* Write data and address values. */
2337 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2338 	    address & 0xffffffff, 4);
2339 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2340 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2341 		    address >> 32, 4);
2342 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2343 		    data, 2);
2344 	} else
2345 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2346 		    2);
2347 
2348 	/* Enable MSI in the control register. */
2349 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2350 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2351 	    msi->msi_ctrl, 2);
2352 
2353 	/* Enable MSI -> HT mapping. */
2354 	pci_ht_map_msi(child, address);
2355 }
2356 
2357 void
2358 pci_disable_msi_method(device_t dev, device_t child)
2359 {
2360 	struct pci_devinfo *dinfo = device_get_ivars(child);
2361 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2362 
2363 	/* Disable MSI -> HT mapping. */
2364 	pci_ht_map_msi(child, 0);
2365 
2366 	/* Disable MSI in the control register. */
2367 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2368 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2369 	    msi->msi_ctrl, 2);
2370 }
2371 
2372 /*
2373  * Restore MSI registers during resume.  If MSI is enabled then
2374  * restore the data and address registers in addition to the control
2375  * register.
2376  */
2377 static void
2378 pci_resume_msi(device_t dev)
2379 {
2380 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2381 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2382 	uint64_t address;
2383 	uint16_t data;
2384 
2385 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2386 		address = msi->msi_addr;
2387 		data = msi->msi_data;
2388 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2389 		    address & 0xffffffff, 4);
2390 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2391 			pci_write_config(dev, msi->msi_location +
2392 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2393 			pci_write_config(dev, msi->msi_location +
2394 			    PCIR_MSI_DATA_64BIT, data, 2);
2395 		} else
2396 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2397 			    data, 2);
2398 	}
2399 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2400 	    2);
2401 }
2402 
2403 static int
2404 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2405 {
2406 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2407 	pcicfgregs *cfg = &dinfo->cfg;
2408 	struct resource_list_entry *rle;
2409 	struct msix_table_entry *mte;
2410 	struct msix_vector *mv;
2411 	uint64_t addr;
2412 	uint32_t data;
2413 	int error, i, j;
2414 
2415 	/*
2416 	 * Handle MSI first.  We try to find this IRQ among our list
2417 	 * of MSI IRQs.  If we find it, we request updated address and
2418 	 * data registers and apply the results.
2419 	 */
2420 	if (cfg->msi.msi_alloc > 0) {
2421 		/* If we don't have any active handlers, nothing to do. */
2422 		if (cfg->msi.msi_handlers == 0)
2423 			return (0);
2424 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2425 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2426 			    i + 1);
2427 			if (rle->start == irq) {
2428 				error = PCIB_MAP_MSI(device_get_parent(bus),
2429 				    dev, irq, &addr, &data);
2430 				if (error)
2431 					return (error);
2432 				pci_disable_msi(dev);
2433 				dinfo->cfg.msi.msi_addr = addr;
2434 				dinfo->cfg.msi.msi_data = data;
2435 				pci_enable_msi(dev, addr, data);
2436 				return (0);
2437 			}
2438 		}
2439 		return (ENOENT);
2440 	}
2441 
2442 	/*
2443 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2444 	 * we request the updated mapping info.  If that works, we go
2445 	 * through all the slots that use this IRQ and update them.
2446 	 */
2447 	if (cfg->msix.msix_alloc > 0) {
2448 		bool found = false;
2449 
2450 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2451 			mv = &cfg->msix.msix_vectors[i];
2452 			if (mv->mv_irq == irq) {
2453 				error = PCIB_MAP_MSI(device_get_parent(bus),
2454 				    dev, irq, &addr, &data);
2455 				if (error)
2456 					return (error);
2457 				mv->mv_address = addr;
2458 				mv->mv_data = data;
2459 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2460 					mte = &cfg->msix.msix_table[j];
2461 					if (mte->mte_vector != i + 1)
2462 						continue;
2463 					if (mte->mte_handlers == 0)
2464 						continue;
2465 					pci_mask_msix(dev, j);
2466 					pci_enable_msix(dev, j, addr, data);
2467 					pci_unmask_msix(dev, j);
2468 				}
2469 				found = true;
2470 			}
2471 		}
2472 		return (found ? 0 : ENOENT);
2473 	}
2474 
2475 	return (ENOENT);
2476 }
2477 
2478 /*
2479  * Returns true if the specified device is blacklisted because MSI
2480  * doesn't work.
2481  */
2482 int
2483 pci_msi_device_blacklisted(device_t dev)
2484 {
2485 
2486 	if (!pci_honor_msi_blacklist)
2487 		return (0);
2488 
2489 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2490 }
2491 
2492 /*
2493  * Determine if MSI is blacklisted globally on this system.  Currently,
2494  * we just check for blacklisted chipsets as represented by the
2495  * host-PCI bridge at device 0:0:0.  In the future, it may become
2496  * necessary to check other system attributes, such as the kenv values
2497  * that give the motherboard manufacturer and model number.
2498  */
2499 static int
2500 pci_msi_blacklisted(void)
2501 {
2502 	device_t dev;
2503 
2504 	if (!pci_honor_msi_blacklist)
2505 		return (0);
2506 
2507 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2508 	if (!(pcie_chipset || pcix_chipset)) {
2509 		if (vm_guest != VM_GUEST_NO) {
2510 			/*
2511 			 * Whitelist older chipsets in virtual
2512 			 * machines known to support MSI.
2513 			 */
2514 			dev = pci_find_bsf(0, 0, 0);
2515 			if (dev != NULL)
2516 				return (!pci_has_quirk(pci_get_devid(dev),
2517 					PCI_QUIRK_ENABLE_MSI_VM));
2518 		}
2519 		return (1);
2520 	}
2521 
2522 	dev = pci_find_bsf(0, 0, 0);
2523 	if (dev != NULL)
2524 		return (pci_msi_device_blacklisted(dev));
2525 	return (0);
2526 }
2527 
2528 /*
2529  * Returns true if the specified device is blacklisted because MSI-X
2530  * doesn't work.  Note that this assumes that if MSI doesn't work,
2531  * MSI-X doesn't either.
2532  */
2533 int
2534 pci_msix_device_blacklisted(device_t dev)
2535 {
2536 
2537 	if (!pci_honor_msi_blacklist)
2538 		return (0);
2539 
2540 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2541 		return (1);
2542 
2543 	return (pci_msi_device_blacklisted(dev));
2544 }
2545 
2546 /*
2547  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2548  * is blacklisted, assume that MSI-X is as well.  Check for additional
2549  * chipsets where MSI works but MSI-X does not.
2550  */
2551 static int
2552 pci_msix_blacklisted(void)
2553 {
2554 	device_t dev;
2555 
2556 	if (!pci_honor_msi_blacklist)
2557 		return (0);
2558 
2559 	dev = pci_find_bsf(0, 0, 0);
2560 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2561 	    PCI_QUIRK_DISABLE_MSIX))
2562 		return (1);
2563 
2564 	return (pci_msi_blacklisted());
2565 }
2566 
2567 /*
2568  * Attempt to allocate *count MSI messages.  The actual number allocated is
2569  * returned in *count.  After this function returns, each message will be
2570  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2571  */
2572 int
2573 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2574 {
2575 	struct pci_devinfo *dinfo = device_get_ivars(child);
2576 	pcicfgregs *cfg = &dinfo->cfg;
2577 	struct resource_list_entry *rle;
2578 	int actual, error, i, irqs[32];
2579 	uint16_t ctrl;
2580 
2581 	/* Don't let count == 0 get us into trouble. */
2582 	if (*count == 0)
2583 		return (EINVAL);
2584 
2585 	/* If rid 0 is allocated, then fail. */
2586 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2587 	if (rle != NULL && rle->res != NULL)
2588 		return (ENXIO);
2589 
2590 	/* Already have allocated messages? */
2591 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2592 		return (ENXIO);
2593 
2594 	/* If MSI is blacklisted for this system, fail. */
2595 	if (pci_msi_blacklisted())
2596 		return (ENXIO);
2597 
2598 	/* MSI capability present? */
2599 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2600 		return (ENODEV);
2601 
2602 	if (bootverbose)
2603 		device_printf(child,
2604 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2605 		    *count, cfg->msi.msi_msgnum);
2606 
2607 	/* Don't ask for more than the device supports. */
2608 	actual = min(*count, cfg->msi.msi_msgnum);
2609 
2610 	/* Don't ask for more than 32 messages. */
2611 	actual = min(actual, 32);
2612 
2613 	/* MSI requires power of 2 number of messages. */
2614 	if (!powerof2(actual))
2615 		return (EINVAL);
2616 
2617 	for (;;) {
2618 		/* Try to allocate N messages. */
2619 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2620 		    actual, irqs);
2621 		if (error == 0)
2622 			break;
2623 		if (actual == 1)
2624 			return (error);
2625 
2626 		/* Try N / 2. */
2627 		actual >>= 1;
2628 	}
2629 
2630 	/*
2631 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2632 	 * resources in the irqs[] array, so add new resources
2633 	 * starting at rid 1.
2634 	 */
2635 	for (i = 0; i < actual; i++)
2636 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2637 		    irqs[i], irqs[i], 1);
2638 
2639 	if (bootverbose) {
2640 		if (actual == 1)
2641 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2642 		else {
2643 			int run;
2644 
2645 			/*
2646 			 * Be fancy and try to print contiguous runs
2647 			 * of IRQ values as ranges.  'run' is true if
2648 			 * we are in a range.
2649 			 */
2650 			device_printf(child, "using IRQs %d", irqs[0]);
2651 			run = 0;
2652 			for (i = 1; i < actual; i++) {
2653 				/* Still in a run? */
2654 				if (irqs[i] == irqs[i - 1] + 1) {
2655 					run = 1;
2656 					continue;
2657 				}
2658 
2659 				/* Finish previous range. */
2660 				if (run) {
2661 					printf("-%d", irqs[i - 1]);
2662 					run = 0;
2663 				}
2664 
2665 				/* Start new range. */
2666 				printf(",%d", irqs[i]);
2667 			}
2668 
2669 			/* Unfinished range? */
2670 			if (run)
2671 				printf("-%d", irqs[actual - 1]);
2672 			printf(" for MSI\n");
2673 		}
2674 	}
2675 
2676 	/* Update control register with actual count. */
2677 	ctrl = cfg->msi.msi_ctrl;
2678 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2679 	ctrl |= (ffs(actual) - 1) << 4;
2680 	cfg->msi.msi_ctrl = ctrl;
2681 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2682 
2683 	/* Update counts of alloc'd messages. */
2684 	cfg->msi.msi_alloc = actual;
2685 	cfg->msi.msi_handlers = 0;
2686 	*count = actual;
2687 	return (0);
2688 }
2689 
2690 /* Release the MSI messages associated with this device. */
2691 int
2692 pci_release_msi_method(device_t dev, device_t child)
2693 {
2694 	struct pci_devinfo *dinfo = device_get_ivars(child);
2695 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2696 	struct resource_list_entry *rle;
2697 	int error, i, irqs[32];
2698 
2699 	/* Try MSI-X first. */
2700 	error = pci_release_msix(dev, child);
2701 	if (error != ENODEV)
2702 		return (error);
2703 
2704 	/* Do we have any messages to release? */
2705 	if (msi->msi_alloc == 0)
2706 		return (ENODEV);
2707 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2708 
2709 	/* Make sure none of the resources are allocated. */
2710 	if (msi->msi_handlers > 0)
2711 		return (EBUSY);
2712 	for (i = 0; i < msi->msi_alloc; i++) {
2713 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2714 		KASSERT(rle != NULL, ("missing MSI resource"));
2715 		if (rle->res != NULL)
2716 			return (EBUSY);
2717 		irqs[i] = rle->start;
2718 	}
2719 
2720 	/* Update control register with 0 count. */
2721 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2722 	    ("%s: MSI still enabled", __func__));
2723 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2724 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2725 	    msi->msi_ctrl, 2);
2726 
2727 	/* Release the messages. */
2728 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2729 	for (i = 0; i < msi->msi_alloc; i++)
2730 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2731 
2732 	/* Update alloc count. */
2733 	msi->msi_alloc = 0;
2734 	msi->msi_addr = 0;
2735 	msi->msi_data = 0;
2736 	return (0);
2737 }
2738 
2739 /*
2740  * Return the max supported MSI messages this device supports.
2741  * Basically, assuming the MD code can alloc messages, this function
2742  * should return the maximum value that pci_alloc_msi() can return.
2743  * Thus, it is subject to the tunables, etc.
2744  */
2745 int
2746 pci_msi_count_method(device_t dev, device_t child)
2747 {
2748 	struct pci_devinfo *dinfo = device_get_ivars(child);
2749 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2750 
2751 	if (pci_do_msi && msi->msi_location != 0)
2752 		return (msi->msi_msgnum);
2753 	return (0);
2754 }
2755 
2756 /* free pcicfgregs structure and all depending data structures */
2757 
2758 int
2759 pci_freecfg(struct pci_devinfo *dinfo)
2760 {
2761 	struct devlist *devlist_head;
2762 	struct pci_map *pm, *next;
2763 
2764 	devlist_head = &pci_devq;
2765 
2766 	if (dinfo->cfg.vpd.vpd_reg)
2767 		vpd_free(&dinfo->cfg.vpd);
2768 
2769 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2770 		free(pm, M_DEVBUF);
2771 	}
2772 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2773 	free(dinfo, M_DEVBUF);
2774 
2775 	/* increment the generation count */
2776 	pci_generation++;
2777 
2778 	/* we're losing one device */
2779 	pci_numdevs--;
2780 	return (0);
2781 }
2782 
2783 /*
2784  * PCI power manangement
2785  */
2786 int
2787 pci_set_powerstate_method(device_t dev, device_t child, int state)
2788 {
2789 	struct pci_devinfo *dinfo = device_get_ivars(child);
2790 	pcicfgregs *cfg = &dinfo->cfg;
2791 	uint16_t status;
2792 	int oldstate, highest, delay;
2793 
2794 	if (cfg->pp.pp_cap == 0)
2795 		return (EOPNOTSUPP);
2796 
2797 	/*
2798 	 * Optimize a no state change request away.  While it would be OK to
2799 	 * write to the hardware in theory, some devices have shown odd
2800 	 * behavior when going from D3 -> D3.
2801 	 */
2802 	oldstate = pci_get_powerstate(child);
2803 	if (oldstate == state)
2804 		return (0);
2805 
2806 	/*
2807 	 * The PCI power management specification states that after a state
2808 	 * transition between PCI power states, system software must
2809 	 * guarantee a minimal delay before the function accesses the device.
2810 	 * Compute the worst case delay that we need to guarantee before we
2811 	 * access the device.  Many devices will be responsive much more
2812 	 * quickly than this delay, but there are some that don't respond
2813 	 * instantly to state changes.  Transitions to/from D3 state require
2814 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2815 	 * is done below with DELAY rather than a sleeper function because
2816 	 * this function can be called from contexts where we cannot sleep.
2817 	 */
2818 	highest = (oldstate > state) ? oldstate : state;
2819 	if (highest == PCI_POWERSTATE_D3)
2820 	    delay = 10000;
2821 	else if (highest == PCI_POWERSTATE_D2)
2822 	    delay = 200;
2823 	else
2824 	    delay = 0;
2825 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2826 	    & ~PCIM_PSTAT_DMASK;
2827 	switch (state) {
2828 	case PCI_POWERSTATE_D0:
2829 		status |= PCIM_PSTAT_D0;
2830 		break;
2831 	case PCI_POWERSTATE_D1:
2832 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2833 			return (EOPNOTSUPP);
2834 		status |= PCIM_PSTAT_D1;
2835 		break;
2836 	case PCI_POWERSTATE_D2:
2837 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2838 			return (EOPNOTSUPP);
2839 		status |= PCIM_PSTAT_D2;
2840 		break;
2841 	case PCI_POWERSTATE_D3:
2842 		status |= PCIM_PSTAT_D3;
2843 		break;
2844 	default:
2845 		return (EINVAL);
2846 	}
2847 
2848 	if (bootverbose)
2849 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2850 		    state);
2851 
2852 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2853 	if (delay)
2854 		DELAY(delay);
2855 	return (0);
2856 }
2857 
2858 int
2859 pci_get_powerstate_method(device_t dev, device_t child)
2860 {
2861 	struct pci_devinfo *dinfo = device_get_ivars(child);
2862 	pcicfgregs *cfg = &dinfo->cfg;
2863 	uint16_t status;
2864 	int result;
2865 
2866 	if (cfg->pp.pp_cap != 0) {
2867 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2868 		switch (status & PCIM_PSTAT_DMASK) {
2869 		case PCIM_PSTAT_D0:
2870 			result = PCI_POWERSTATE_D0;
2871 			break;
2872 		case PCIM_PSTAT_D1:
2873 			result = PCI_POWERSTATE_D1;
2874 			break;
2875 		case PCIM_PSTAT_D2:
2876 			result = PCI_POWERSTATE_D2;
2877 			break;
2878 		case PCIM_PSTAT_D3:
2879 			result = PCI_POWERSTATE_D3;
2880 			break;
2881 		default:
2882 			result = PCI_POWERSTATE_UNKNOWN;
2883 			break;
2884 		}
2885 	} else {
2886 		/* No support, device is always at D0 */
2887 		result = PCI_POWERSTATE_D0;
2888 	}
2889 	return (result);
2890 }
2891 
2892 /*
2893  * Some convenience functions for PCI device drivers.
2894  */
2895 
2896 static __inline void
2897 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2898 {
2899 	uint16_t	command;
2900 
2901 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2902 	command |= bit;
2903 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2904 }
2905 
2906 static __inline void
2907 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2908 {
2909 	uint16_t	command;
2910 
2911 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2912 	command &= ~bit;
2913 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2914 }
2915 
2916 int
2917 pci_enable_busmaster_method(device_t dev, device_t child)
2918 {
2919 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2920 	return (0);
2921 }
2922 
2923 int
2924 pci_disable_busmaster_method(device_t dev, device_t child)
2925 {
2926 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2927 	return (0);
2928 }
2929 
2930 int
2931 pci_enable_io_method(device_t dev, device_t child, int space)
2932 {
2933 	uint16_t bit;
2934 
2935 	switch(space) {
2936 	case SYS_RES_IOPORT:
2937 		bit = PCIM_CMD_PORTEN;
2938 		break;
2939 	case SYS_RES_MEMORY:
2940 		bit = PCIM_CMD_MEMEN;
2941 		break;
2942 	default:
2943 		return (EINVAL);
2944 	}
2945 	pci_set_command_bit(dev, child, bit);
2946 	return (0);
2947 }
2948 
2949 int
2950 pci_disable_io_method(device_t dev, device_t child, int space)
2951 {
2952 	uint16_t bit;
2953 
2954 	switch(space) {
2955 	case SYS_RES_IOPORT:
2956 		bit = PCIM_CMD_PORTEN;
2957 		break;
2958 	case SYS_RES_MEMORY:
2959 		bit = PCIM_CMD_MEMEN;
2960 		break;
2961 	default:
2962 		return (EINVAL);
2963 	}
2964 	pci_clear_command_bit(dev, child, bit);
2965 	return (0);
2966 }
2967 
2968 /*
2969  * New style pci driver.  Parent device is either a pci-host-bridge or a
2970  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2971  */
2972 
2973 void
2974 pci_print_verbose(struct pci_devinfo *dinfo)
2975 {
2976 
2977 	if (bootverbose) {
2978 		pcicfgregs *cfg = &dinfo->cfg;
2979 
2980 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2981 		    cfg->vendor, cfg->device, cfg->revid);
2982 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2983 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2984 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2985 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2986 		    cfg->mfdev);
2987 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2988 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2989 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2990 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2991 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2992 		if (cfg->intpin > 0)
2993 			printf("\tintpin=%c, irq=%d\n",
2994 			    cfg->intpin +'a' -1, cfg->intline);
2995 		if (cfg->pp.pp_cap) {
2996 			uint16_t status;
2997 
2998 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2999 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3000 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3001 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3002 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3003 			    status & PCIM_PSTAT_DMASK);
3004 		}
3005 		if (cfg->msi.msi_location) {
3006 			int ctrl;
3007 
3008 			ctrl = cfg->msi.msi_ctrl;
3009 			printf("\tMSI supports %d message%s%s%s\n",
3010 			    cfg->msi.msi_msgnum,
3011 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
3012 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3013 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3014 		}
3015 		if (cfg->msix.msix_location) {
3016 			printf("\tMSI-X supports %d message%s ",
3017 			    cfg->msix.msix_msgnum,
3018 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
3019 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3020 				printf("in map 0x%x\n",
3021 				    cfg->msix.msix_table_bar);
3022 			else
3023 				printf("in maps 0x%x and 0x%x\n",
3024 				    cfg->msix.msix_table_bar,
3025 				    cfg->msix.msix_pba_bar);
3026 		}
3027 	}
3028 }
3029 
3030 static int
3031 pci_porten(device_t dev)
3032 {
3033 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3034 }
3035 
3036 static int
3037 pci_memen(device_t dev)
3038 {
3039 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3040 }
3041 
3042 void
3043 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3044     int *bar64)
3045 {
3046 	struct pci_devinfo *dinfo;
3047 	pci_addr_t map, testval;
3048 	int ln2range;
3049 	uint16_t cmd;
3050 
3051 	/*
3052 	 * The device ROM BAR is special.  It is always a 32-bit
3053 	 * memory BAR.  Bit 0 is special and should not be set when
3054 	 * sizing the BAR.
3055 	 */
3056 	dinfo = device_get_ivars(dev);
3057 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3058 		map = pci_read_config(dev, reg, 4);
3059 		pci_write_config(dev, reg, 0xfffffffe, 4);
3060 		testval = pci_read_config(dev, reg, 4);
3061 		pci_write_config(dev, reg, map, 4);
3062 		*mapp = map;
3063 		*testvalp = testval;
3064 		if (bar64 != NULL)
3065 			*bar64 = 0;
3066 		return;
3067 	}
3068 
3069 	map = pci_read_config(dev, reg, 4);
3070 	ln2range = pci_maprange(map);
3071 	if (ln2range == 64)
3072 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3073 
3074 	/*
3075 	 * Disable decoding via the command register before
3076 	 * determining the BAR's length since we will be placing it in
3077 	 * a weird state.
3078 	 */
3079 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3080 	pci_write_config(dev, PCIR_COMMAND,
3081 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3082 
3083 	/*
3084 	 * Determine the BAR's length by writing all 1's.  The bottom
3085 	 * log_2(size) bits of the BAR will stick as 0 when we read
3086 	 * the value back.
3087 	 *
3088 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3089 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3090 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3091 	 *
3092 	 * Writes to both registers must be performed before attempting to
3093 	 * read back the size value.
3094 	 */
3095 	testval = 0;
3096 	pci_write_config(dev, reg, 0xffffffff, 4);
3097 	if (ln2range == 64) {
3098 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3099 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3100 	}
3101 	testval |= pci_read_config(dev, reg, 4);
3102 
3103 	/*
3104 	 * Restore the original value of the BAR.  We may have reprogrammed
3105 	 * the BAR of the low-level console device and when booting verbose,
3106 	 * we need the console device addressable.
3107 	 */
3108 	pci_write_config(dev, reg, map, 4);
3109 	if (ln2range == 64)
3110 		pci_write_config(dev, reg + 4, map >> 32, 4);
3111 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3112 
3113 	*mapp = map;
3114 	*testvalp = testval;
3115 	if (bar64 != NULL)
3116 		*bar64 = (ln2range == 64);
3117 }
3118 
3119 static void
3120 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3121 {
3122 	struct pci_devinfo *dinfo;
3123 	int ln2range;
3124 
3125 	/* The device ROM BAR is always a 32-bit memory BAR. */
3126 	dinfo = device_get_ivars(dev);
3127 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3128 		ln2range = 32;
3129 	else
3130 		ln2range = pci_maprange(pm->pm_value);
3131 	pci_write_config(dev, pm->pm_reg, base, 4);
3132 	if (ln2range == 64)
3133 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3134 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3135 	if (ln2range == 64)
3136 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3137 		    pm->pm_reg + 4, 4) << 32;
3138 }
3139 
3140 struct pci_map *
3141 pci_find_bar(device_t dev, int reg)
3142 {
3143 	struct pci_devinfo *dinfo;
3144 	struct pci_map *pm;
3145 
3146 	dinfo = device_get_ivars(dev);
3147 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3148 		if (pm->pm_reg == reg)
3149 			return (pm);
3150 	}
3151 	return (NULL);
3152 }
3153 
3154 struct pci_map *
3155 pci_first_bar(device_t dev)
3156 {
3157 	struct pci_devinfo *dinfo;
3158 
3159 	dinfo = device_get_ivars(dev);
3160 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3161 }
3162 
3163 struct pci_map *
3164 pci_next_bar(struct pci_map *pm)
3165 {
3166 	return (STAILQ_NEXT(pm, pm_link));
3167 }
3168 
3169 int
3170 pci_bar_enabled(device_t dev, struct pci_map *pm)
3171 {
3172 	struct pci_devinfo *dinfo;
3173 	uint16_t cmd;
3174 
3175 	dinfo = device_get_ivars(dev);
3176 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3177 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3178 		return (0);
3179 #ifdef PCI_IOV
3180 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3181 		struct pcicfg_iov *iov;
3182 
3183 		iov = dinfo->cfg.iov;
3184 		cmd = pci_read_config(iov->iov_pf,
3185 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3186 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3187 	}
3188 #endif
3189 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3190 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3191 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3192 	else
3193 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3194 }
3195 
3196 struct pci_map *
3197 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3198 {
3199 	struct pci_devinfo *dinfo;
3200 	struct pci_map *pm, *prev;
3201 
3202 	dinfo = device_get_ivars(dev);
3203 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3204 	pm->pm_reg = reg;
3205 	pm->pm_value = value;
3206 	pm->pm_size = size;
3207 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3208 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3209 		    reg));
3210 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3211 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3212 			break;
3213 	}
3214 	if (prev != NULL)
3215 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3216 	else
3217 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3218 	return (pm);
3219 }
3220 
3221 static void
3222 pci_restore_bars(device_t dev)
3223 {
3224 	struct pci_devinfo *dinfo;
3225 	struct pci_map *pm;
3226 	int ln2range;
3227 
3228 	dinfo = device_get_ivars(dev);
3229 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3230 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3231 			ln2range = 32;
3232 		else
3233 			ln2range = pci_maprange(pm->pm_value);
3234 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3235 		if (ln2range == 64)
3236 			pci_write_config(dev, pm->pm_reg + 4,
3237 			    pm->pm_value >> 32, 4);
3238 	}
3239 }
3240 
3241 /*
3242  * Add a resource based on a pci map register. Return 1 if the map
3243  * register is a 32bit map register or 2 if it is a 64bit register.
3244  */
3245 static int
3246 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3247     int force, int prefetch)
3248 {
3249 	struct pci_map *pm;
3250 	pci_addr_t base, map, testval;
3251 	pci_addr_t start, end, count;
3252 	int barlen, basezero, flags, maprange, mapsize, type;
3253 	uint16_t cmd;
3254 	struct resource *res;
3255 
3256 	/*
3257 	 * The BAR may already exist if the device is a CardBus card
3258 	 * whose CIS is stored in this BAR.
3259 	 */
3260 	pm = pci_find_bar(dev, reg);
3261 	if (pm != NULL) {
3262 		maprange = pci_maprange(pm->pm_value);
3263 		barlen = maprange == 64 ? 2 : 1;
3264 		return (barlen);
3265 	}
3266 
3267 	pci_read_bar(dev, reg, &map, &testval, NULL);
3268 	if (PCI_BAR_MEM(map)) {
3269 		type = SYS_RES_MEMORY;
3270 		if (map & PCIM_BAR_MEM_PREFETCH)
3271 			prefetch = 1;
3272 	} else
3273 		type = SYS_RES_IOPORT;
3274 	mapsize = pci_mapsize(testval);
3275 	base = pci_mapbase(map);
3276 #ifdef __PCI_BAR_ZERO_VALID
3277 	basezero = 0;
3278 #else
3279 	basezero = base == 0;
3280 #endif
3281 	maprange = pci_maprange(map);
3282 	barlen = maprange == 64 ? 2 : 1;
3283 
3284 	/*
3285 	 * For I/O registers, if bottom bit is set, and the next bit up
3286 	 * isn't clear, we know we have a BAR that doesn't conform to the
3287 	 * spec, so ignore it.  Also, sanity check the size of the data
3288 	 * areas to the type of memory involved.  Memory must be at least
3289 	 * 16 bytes in size, while I/O ranges must be at least 4.
3290 	 */
3291 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3292 		return (barlen);
3293 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3294 	    (type == SYS_RES_IOPORT && mapsize < 2))
3295 		return (barlen);
3296 
3297 	/* Save a record of this BAR. */
3298 	pm = pci_add_bar(dev, reg, map, mapsize);
3299 	if (bootverbose) {
3300 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3301 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3302 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3303 			printf(", port disabled\n");
3304 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3305 			printf(", memory disabled\n");
3306 		else
3307 			printf(", enabled\n");
3308 	}
3309 
3310 	/*
3311 	 * If base is 0, then we have problems if this architecture does
3312 	 * not allow that.  It is best to ignore such entries for the
3313 	 * moment.  These will be allocated later if the driver specifically
3314 	 * requests them.  However, some removable buses look better when
3315 	 * all resources are allocated, so allow '0' to be overridden.
3316 	 *
3317 	 * Similarly treat maps whose values is the same as the test value
3318 	 * read back.  These maps have had all f's written to them by the
3319 	 * BIOS in an attempt to disable the resources.
3320 	 */
3321 	if (!force && (basezero || map == testval))
3322 		return (barlen);
3323 	if ((u_long)base != base) {
3324 		device_printf(bus,
3325 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3326 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3327 		    pci_get_function(dev), reg);
3328 		return (barlen);
3329 	}
3330 
3331 	/*
3332 	 * This code theoretically does the right thing, but has
3333 	 * undesirable side effects in some cases where peripherals
3334 	 * respond oddly to having these bits enabled.  Let the user
3335 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3336 	 * default).
3337 	 */
3338 	if (pci_enable_io_modes) {
3339 		/* Turn on resources that have been left off by a lazy BIOS */
3340 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3341 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3342 			cmd |= PCIM_CMD_PORTEN;
3343 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3344 		}
3345 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3346 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3347 			cmd |= PCIM_CMD_MEMEN;
3348 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3349 		}
3350 	} else {
3351 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3352 			return (barlen);
3353 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3354 			return (barlen);
3355 	}
3356 
3357 	count = (pci_addr_t)1 << mapsize;
3358 	flags = RF_ALIGNMENT_LOG2(mapsize);
3359 	if (prefetch)
3360 		flags |= RF_PREFETCHABLE;
3361 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3362 		start = 0;	/* Let the parent decide. */
3363 		end = ~0;
3364 	} else {
3365 		start = base;
3366 		end = base + count - 1;
3367 	}
3368 	resource_list_add(rl, type, reg, start, end, count);
3369 
3370 	/*
3371 	 * Try to allocate the resource for this BAR from our parent
3372 	 * so that this resource range is already reserved.  The
3373 	 * driver for this device will later inherit this resource in
3374 	 * pci_alloc_resource().
3375 	 */
3376 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3377 	    flags);
3378 	if ((pci_do_realloc_bars
3379 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3380 	    && res == NULL && (start != 0 || end != ~0)) {
3381 		/*
3382 		 * If the allocation fails, try to allocate a resource for
3383 		 * this BAR using any available range.  The firmware felt
3384 		 * it was important enough to assign a resource, so don't
3385 		 * disable decoding if we can help it.
3386 		 */
3387 		resource_list_delete(rl, type, reg);
3388 		resource_list_add(rl, type, reg, 0, ~0, count);
3389 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3390 		    count, flags);
3391 	}
3392 	if (res == NULL) {
3393 		/*
3394 		 * If the allocation fails, delete the resource list entry
3395 		 * and disable decoding for this device.
3396 		 *
3397 		 * If the driver requests this resource in the future,
3398 		 * pci_reserve_map() will try to allocate a fresh
3399 		 * resource range.
3400 		 */
3401 		resource_list_delete(rl, type, reg);
3402 		pci_disable_io(dev, type);
3403 		if (bootverbose)
3404 			device_printf(bus,
3405 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3406 			    pci_get_domain(dev), pci_get_bus(dev),
3407 			    pci_get_slot(dev), pci_get_function(dev), reg);
3408 	} else {
3409 		start = rman_get_start(res);
3410 		pci_write_bar(dev, pm, start);
3411 	}
3412 	return (barlen);
3413 }
3414 
3415 /*
3416  * For ATA devices we need to decide early what addressing mode to use.
3417  * Legacy demands that the primary and secondary ATA ports sits on the
3418  * same addresses that old ISA hardware did. This dictates that we use
3419  * those addresses and ignore the BAR's if we cannot set PCI native
3420  * addressing mode.
3421  */
3422 static void
3423 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3424     uint32_t prefetchmask)
3425 {
3426 	int rid, type, progif;
3427 #if 0
3428 	/* if this device supports PCI native addressing use it */
3429 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3430 	if ((progif & 0x8a) == 0x8a) {
3431 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3432 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3433 			printf("Trying ATA native PCI addressing mode\n");
3434 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3435 		}
3436 	}
3437 #endif
3438 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3439 	type = SYS_RES_IOPORT;
3440 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3441 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3442 		    prefetchmask & (1 << 0));
3443 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3444 		    prefetchmask & (1 << 1));
3445 	} else {
3446 		rid = PCIR_BAR(0);
3447 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3448 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3449 		    0x1f7, 8, 0);
3450 		rid = PCIR_BAR(1);
3451 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3452 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3453 		    0x3f6, 1, 0);
3454 	}
3455 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3456 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3457 		    prefetchmask & (1 << 2));
3458 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3459 		    prefetchmask & (1 << 3));
3460 	} else {
3461 		rid = PCIR_BAR(2);
3462 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3463 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3464 		    0x177, 8, 0);
3465 		rid = PCIR_BAR(3);
3466 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3467 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3468 		    0x376, 1, 0);
3469 	}
3470 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3471 	    prefetchmask & (1 << 4));
3472 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3473 	    prefetchmask & (1 << 5));
3474 }
3475 
3476 static void
3477 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3478 {
3479 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3480 	pcicfgregs *cfg = &dinfo->cfg;
3481 	char tunable_name[64];
3482 	int irq;
3483 
3484 	/* Has to have an intpin to have an interrupt. */
3485 	if (cfg->intpin == 0)
3486 		return;
3487 
3488 	/* Let the user override the IRQ with a tunable. */
3489 	irq = PCI_INVALID_IRQ;
3490 	snprintf(tunable_name, sizeof(tunable_name),
3491 	    "hw.pci%d.%d.%d.INT%c.irq",
3492 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3493 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3494 		irq = PCI_INVALID_IRQ;
3495 
3496 	/*
3497 	 * If we didn't get an IRQ via the tunable, then we either use the
3498 	 * IRQ value in the intline register or we ask the bus to route an
3499 	 * interrupt for us.  If force_route is true, then we only use the
3500 	 * value in the intline register if the bus was unable to assign an
3501 	 * IRQ.
3502 	 */
3503 	if (!PCI_INTERRUPT_VALID(irq)) {
3504 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3505 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3506 		if (!PCI_INTERRUPT_VALID(irq))
3507 			irq = cfg->intline;
3508 	}
3509 
3510 	/* If after all that we don't have an IRQ, just bail. */
3511 	if (!PCI_INTERRUPT_VALID(irq))
3512 		return;
3513 
3514 	/* Update the config register if it changed. */
3515 	if (irq != cfg->intline) {
3516 		cfg->intline = irq;
3517 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3518 	}
3519 
3520 	/* Add this IRQ as rid 0 interrupt resource. */
3521 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3522 }
3523 
3524 /* Perform early OHCI takeover from SMM. */
3525 static void
3526 ohci_early_takeover(device_t self)
3527 {
3528 	struct resource *res;
3529 	uint32_t ctl;
3530 	int rid;
3531 	int i;
3532 
3533 	rid = PCIR_BAR(0);
3534 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3535 	if (res == NULL)
3536 		return;
3537 
3538 	ctl = bus_read_4(res, OHCI_CONTROL);
3539 	if (ctl & OHCI_IR) {
3540 		if (bootverbose)
3541 			printf("ohci early: "
3542 			    "SMM active, request owner change\n");
3543 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3544 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3545 			DELAY(1000);
3546 			ctl = bus_read_4(res, OHCI_CONTROL);
3547 		}
3548 		if (ctl & OHCI_IR) {
3549 			if (bootverbose)
3550 				printf("ohci early: "
3551 				    "SMM does not respond, resetting\n");
3552 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3553 		}
3554 		/* Disable interrupts */
3555 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3556 	}
3557 
3558 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3559 }
3560 
3561 /* Perform early UHCI takeover from SMM. */
3562 static void
3563 uhci_early_takeover(device_t self)
3564 {
3565 	struct resource *res;
3566 	int rid;
3567 
3568 	/*
3569 	 * Set the PIRQD enable bit and switch off all the others. We don't
3570 	 * want legacy support to interfere with us XXX Does this also mean
3571 	 * that the BIOS won't touch the keyboard anymore if it is connected
3572 	 * to the ports of the root hub?
3573 	 */
3574 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3575 
3576 	/* Disable interrupts */
3577 	rid = PCI_UHCI_BASE_REG;
3578 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3579 	if (res != NULL) {
3580 		bus_write_2(res, UHCI_INTR, 0);
3581 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3582 	}
3583 }
3584 
3585 /* Perform early EHCI takeover from SMM. */
3586 static void
3587 ehci_early_takeover(device_t self)
3588 {
3589 	struct resource *res;
3590 	uint32_t cparams;
3591 	uint32_t eec;
3592 	uint8_t eecp;
3593 	uint8_t bios_sem;
3594 	uint8_t offs;
3595 	int rid;
3596 	int i;
3597 
3598 	rid = PCIR_BAR(0);
3599 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3600 	if (res == NULL)
3601 		return;
3602 
3603 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3604 
3605 	/* Synchronise with the BIOS if it owns the controller. */
3606 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3607 	    eecp = EHCI_EECP_NEXT(eec)) {
3608 		eec = pci_read_config(self, eecp, 4);
3609 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3610 			continue;
3611 		}
3612 		bios_sem = pci_read_config(self, eecp +
3613 		    EHCI_LEGSUP_BIOS_SEM, 1);
3614 		if (bios_sem == 0) {
3615 			continue;
3616 		}
3617 		if (bootverbose)
3618 			printf("ehci early: "
3619 			    "SMM active, request owner change\n");
3620 
3621 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3622 
3623 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3624 			DELAY(1000);
3625 			bios_sem = pci_read_config(self, eecp +
3626 			    EHCI_LEGSUP_BIOS_SEM, 1);
3627 		}
3628 
3629 		if (bios_sem != 0) {
3630 			if (bootverbose)
3631 				printf("ehci early: "
3632 				    "SMM does not respond\n");
3633 		}
3634 		/* Disable interrupts */
3635 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3636 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3637 	}
3638 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3639 }
3640 
3641 /* Perform early XHCI takeover from SMM. */
3642 static void
3643 xhci_early_takeover(device_t self)
3644 {
3645 	struct resource *res;
3646 	uint32_t cparams;
3647 	uint32_t eec;
3648 	uint8_t eecp;
3649 	uint8_t bios_sem;
3650 	uint8_t offs;
3651 	int rid;
3652 	int i;
3653 
3654 	rid = PCIR_BAR(0);
3655 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3656 	if (res == NULL)
3657 		return;
3658 
3659 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3660 
3661 	eec = -1;
3662 
3663 	/* Synchronise with the BIOS if it owns the controller. */
3664 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3665 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3666 		eec = bus_read_4(res, eecp);
3667 
3668 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3669 			continue;
3670 
3671 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3672 		if (bios_sem == 0)
3673 			continue;
3674 
3675 		if (bootverbose)
3676 			printf("xhci early: "
3677 			    "SMM active, request owner change\n");
3678 
3679 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3680 
3681 		/* wait a maximum of 5 second */
3682 
3683 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3684 			DELAY(1000);
3685 			bios_sem = bus_read_1(res, eecp +
3686 			    XHCI_XECP_BIOS_SEM);
3687 		}
3688 
3689 		if (bios_sem != 0) {
3690 			if (bootverbose)
3691 				printf("xhci early: "
3692 				    "SMM does not respond\n");
3693 		}
3694 
3695 		/* Disable interrupts */
3696 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3697 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3698 		bus_read_4(res, offs + XHCI_USBSTS);
3699 	}
3700 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3701 }
3702 
3703 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3704 static void
3705 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3706     struct resource_list *rl)
3707 {
3708 	struct resource *res;
3709 	char *cp;
3710 	rman_res_t start, end, count;
3711 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3712 
3713 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3714 	case PCIM_HDRTYPE_BRIDGE:
3715 		sec_reg = PCIR_SECBUS_1;
3716 		sub_reg = PCIR_SUBBUS_1;
3717 		break;
3718 	case PCIM_HDRTYPE_CARDBUS:
3719 		sec_reg = PCIR_SECBUS_2;
3720 		sub_reg = PCIR_SUBBUS_2;
3721 		break;
3722 	default:
3723 		return;
3724 	}
3725 
3726 	/*
3727 	 * If the existing bus range is valid, attempt to reserve it
3728 	 * from our parent.  If this fails for any reason, clear the
3729 	 * secbus and subbus registers.
3730 	 *
3731 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3732 	 * This would at least preserve the existing sec_bus if it is
3733 	 * valid.
3734 	 */
3735 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3736 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3737 
3738 	/* Quirk handling. */
3739 	switch (pci_get_devid(dev)) {
3740 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3741 		sup_bus = pci_read_config(dev, 0x41, 1);
3742 		if (sup_bus != 0xff) {
3743 			sec_bus = sup_bus + 1;
3744 			sub_bus = sup_bus + 1;
3745 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3746 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3747 		}
3748 		break;
3749 
3750 	case 0x00dd10de:
3751 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3752 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3753 			break;
3754 		if (strncmp(cp, "Compal", 6) != 0) {
3755 			freeenv(cp);
3756 			break;
3757 		}
3758 		freeenv(cp);
3759 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3760 			break;
3761 		if (strncmp(cp, "08A0", 4) != 0) {
3762 			freeenv(cp);
3763 			break;
3764 		}
3765 		freeenv(cp);
3766 		if (sub_bus < 0xa) {
3767 			sub_bus = 0xa;
3768 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3769 		}
3770 		break;
3771 	}
3772 
3773 	if (bootverbose)
3774 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3775 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3776 		start = sec_bus;
3777 		end = sub_bus;
3778 		count = end - start + 1;
3779 
3780 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3781 
3782 		/*
3783 		 * If requested, clear secondary bus registers in
3784 		 * bridge devices to force a complete renumbering
3785 		 * rather than reserving the existing range.  However,
3786 		 * preserve the existing size.
3787 		 */
3788 		if (pci_clear_buses)
3789 			goto clear;
3790 
3791 		rid = 0;
3792 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3793 		    start, end, count, 0);
3794 		if (res != NULL)
3795 			return;
3796 
3797 		if (bootverbose)
3798 			device_printf(bus,
3799 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3800 			    pci_get_domain(dev), pci_get_bus(dev),
3801 			    pci_get_slot(dev), pci_get_function(dev));
3802 	}
3803 
3804 clear:
3805 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3806 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3807 }
3808 
3809 static struct resource *
3810 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3811     rman_res_t end, rman_res_t count, u_int flags)
3812 {
3813 	struct pci_devinfo *dinfo;
3814 	pcicfgregs *cfg;
3815 	struct resource_list *rl;
3816 	struct resource *res;
3817 	int sec_reg, sub_reg;
3818 
3819 	dinfo = device_get_ivars(child);
3820 	cfg = &dinfo->cfg;
3821 	rl = &dinfo->resources;
3822 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3823 	case PCIM_HDRTYPE_BRIDGE:
3824 		sec_reg = PCIR_SECBUS_1;
3825 		sub_reg = PCIR_SUBBUS_1;
3826 		break;
3827 	case PCIM_HDRTYPE_CARDBUS:
3828 		sec_reg = PCIR_SECBUS_2;
3829 		sub_reg = PCIR_SUBBUS_2;
3830 		break;
3831 	default:
3832 		return (NULL);
3833 	}
3834 
3835 	if (*rid != 0)
3836 		return (NULL);
3837 
3838 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3839 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3840 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3841 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3842 		    start, end, count, flags & ~RF_ACTIVE);
3843 		if (res == NULL) {
3844 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3845 			device_printf(child, "allocating %ju bus%s failed\n",
3846 			    count, count == 1 ? "" : "es");
3847 			return (NULL);
3848 		}
3849 		if (bootverbose)
3850 			device_printf(child,
3851 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3852 			    count == 1 ? "" : "es", rman_get_start(res));
3853 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3854 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3855 	}
3856 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3857 	    end, count, flags));
3858 }
3859 #endif
3860 
3861 static int
3862 pci_ea_bei_to_rid(device_t dev, int bei)
3863 {
3864 #ifdef PCI_IOV
3865 	struct pci_devinfo *dinfo;
3866 	int iov_pos;
3867 	struct pcicfg_iov *iov;
3868 
3869 	dinfo = device_get_ivars(dev);
3870 	iov = dinfo->cfg.iov;
3871 	if (iov != NULL)
3872 		iov_pos = iov->iov_pos;
3873 	else
3874 		iov_pos = 0;
3875 #endif
3876 
3877 	/* Check if matches BAR */
3878 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3879 	    (bei <= PCIM_EA_BEI_BAR_5))
3880 		return (PCIR_BAR(bei));
3881 
3882 	/* Check ROM */
3883 	if (bei == PCIM_EA_BEI_ROM)
3884 		return (PCIR_BIOS);
3885 
3886 #ifdef PCI_IOV
3887 	/* Check if matches VF_BAR */
3888 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3889 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3890 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3891 		    iov_pos);
3892 #endif
3893 
3894 	return (-1);
3895 }
3896 
3897 int
3898 pci_ea_is_enabled(device_t dev, int rid)
3899 {
3900 	struct pci_ea_entry *ea;
3901 	struct pci_devinfo *dinfo;
3902 
3903 	dinfo = device_get_ivars(dev);
3904 
3905 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3906 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3907 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3908 	}
3909 
3910 	return (0);
3911 }
3912 
3913 void
3914 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3915 {
3916 	struct pci_ea_entry *ea;
3917 	struct pci_devinfo *dinfo;
3918 	pci_addr_t start, end, count;
3919 	struct resource_list *rl;
3920 	int type, flags, rid;
3921 	struct resource *res;
3922 	uint32_t tmp;
3923 #ifdef PCI_IOV
3924 	struct pcicfg_iov *iov;
3925 #endif
3926 
3927 	dinfo = device_get_ivars(dev);
3928 	rl = &dinfo->resources;
3929 	flags = 0;
3930 
3931 #ifdef PCI_IOV
3932 	iov = dinfo->cfg.iov;
3933 #endif
3934 
3935 	if (dinfo->cfg.ea.ea_location == 0)
3936 		return;
3937 
3938 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3939 		/*
3940 		 * TODO: Ignore EA-BAR if is not enabled.
3941 		 *   Currently the EA implementation supports
3942 		 *   only situation, where EA structure contains
3943 		 *   predefined entries. In case they are not enabled
3944 		 *   leave them unallocated and proceed with
3945 		 *   a legacy-BAR mechanism.
3946 		 */
3947 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3948 			continue;
3949 
3950 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3951 		case PCIM_EA_P_MEM_PREFETCH:
3952 		case PCIM_EA_P_VF_MEM_PREFETCH:
3953 			flags = RF_PREFETCHABLE;
3954 			/* FALLTHROUGH */
3955 		case PCIM_EA_P_VF_MEM:
3956 		case PCIM_EA_P_MEM:
3957 			type = SYS_RES_MEMORY;
3958 			break;
3959 		case PCIM_EA_P_IO:
3960 			type = SYS_RES_IOPORT;
3961 			break;
3962 		default:
3963 			continue;
3964 		}
3965 
3966 		if (alloc_iov != 0) {
3967 #ifdef PCI_IOV
3968 			/* Allocating IOV, confirm BEI matches */
3969 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3970 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3971 				continue;
3972 #else
3973 			continue;
3974 #endif
3975 		} else {
3976 			/* Allocating BAR, confirm BEI matches */
3977 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3978 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3979 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3980 				continue;
3981 		}
3982 
3983 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3984 		if (rid < 0)
3985 			continue;
3986 
3987 		/* Skip resources already allocated by EA */
3988 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3989 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3990 			continue;
3991 
3992 		start = ea->eae_base;
3993 		count = ea->eae_max_offset + 1;
3994 #ifdef PCI_IOV
3995 		if (iov != NULL)
3996 			count = count * iov->iov_num_vfs;
3997 #endif
3998 		end = start + count - 1;
3999 		if (count == 0)
4000 			continue;
4001 
4002 		resource_list_add(rl, type, rid, start, end, count);
4003 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4004 		    flags);
4005 		if (res == NULL) {
4006 			resource_list_delete(rl, type, rid);
4007 
4008 			/*
4009 			 * Failed to allocate using EA, disable entry.
4010 			 * Another attempt to allocation will be performed
4011 			 * further, but this time using legacy BAR registers
4012 			 */
4013 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4014 			tmp &= ~PCIM_EA_ENABLE;
4015 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4016 
4017 			/*
4018 			 * Disabling entry might fail in case it is hardwired.
4019 			 * Read flags again to match current status.
4020 			 */
4021 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4022 
4023 			continue;
4024 		}
4025 
4026 		/* As per specification, fill BAR with zeros */
4027 		pci_write_config(dev, rid, 0, 4);
4028 	}
4029 }
4030 
4031 void
4032 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4033 {
4034 	struct pci_devinfo *dinfo;
4035 	pcicfgregs *cfg;
4036 	struct resource_list *rl;
4037 	const struct pci_quirk *q;
4038 	uint32_t devid;
4039 	int i;
4040 
4041 	dinfo = device_get_ivars(dev);
4042 	cfg = &dinfo->cfg;
4043 	rl = &dinfo->resources;
4044 	devid = (cfg->device << 16) | cfg->vendor;
4045 
4046 	/* Allocate resources using Enhanced Allocation */
4047 	pci_add_resources_ea(bus, dev, 0);
4048 
4049 	/* ATA devices needs special map treatment */
4050 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4051 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4052 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4053 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4054 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4055 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4056 	else
4057 		for (i = 0; i < cfg->nummaps;) {
4058 			/* Skip resources already managed by EA */
4059 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4060 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4061 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4062 				i++;
4063 				continue;
4064 			}
4065 
4066 			/*
4067 			 * Skip quirked resources.
4068 			 */
4069 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4070 				if (q->devid == devid &&
4071 				    q->type == PCI_QUIRK_UNMAP_REG &&
4072 				    q->arg1 == PCIR_BAR(i))
4073 					break;
4074 			if (q->devid != 0) {
4075 				i++;
4076 				continue;
4077 			}
4078 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4079 			    prefetchmask & (1 << i));
4080 		}
4081 
4082 	/*
4083 	 * Add additional, quirked resources.
4084 	 */
4085 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4086 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4087 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4088 
4089 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4090 #ifdef __PCI_REROUTE_INTERRUPT
4091 		/*
4092 		 * Try to re-route interrupts. Sometimes the BIOS or
4093 		 * firmware may leave bogus values in these registers.
4094 		 * If the re-route fails, then just stick with what we
4095 		 * have.
4096 		 */
4097 		pci_assign_interrupt(bus, dev, 1);
4098 #else
4099 		pci_assign_interrupt(bus, dev, 0);
4100 #endif
4101 	}
4102 
4103 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4104 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4105 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4106 			xhci_early_takeover(dev);
4107 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4108 			ehci_early_takeover(dev);
4109 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4110 			ohci_early_takeover(dev);
4111 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4112 			uhci_early_takeover(dev);
4113 	}
4114 
4115 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4116 	/*
4117 	 * Reserve resources for secondary bus ranges behind bridge
4118 	 * devices.
4119 	 */
4120 	pci_reserve_secbus(bus, dev, cfg, rl);
4121 #endif
4122 }
4123 
4124 static struct pci_devinfo *
4125 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4126     int slot, int func)
4127 {
4128 	struct pci_devinfo *dinfo;
4129 
4130 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4131 	if (dinfo != NULL)
4132 		pci_add_child(dev, dinfo);
4133 
4134 	return (dinfo);
4135 }
4136 
4137 void
4138 pci_add_children(device_t dev, int domain, int busno)
4139 {
4140 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4141 	device_t pcib = device_get_parent(dev);
4142 	struct pci_devinfo *dinfo;
4143 	int maxslots;
4144 	int s, f, pcifunchigh;
4145 	uint8_t hdrtype;
4146 	int first_func;
4147 
4148 	/*
4149 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4150 	 * enable ARI.  We must enable ARI before detecting the rest of the
4151 	 * functions on this bus as ARI changes the set of slots and functions
4152 	 * that are legal on this bus.
4153 	 */
4154 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4155 	if (dinfo != NULL && pci_enable_ari)
4156 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4157 
4158 	/*
4159 	 * Start looking for new devices on slot 0 at function 1 because we
4160 	 * just identified the device at slot 0, function 0.
4161 	 */
4162 	first_func = 1;
4163 
4164 	maxslots = PCIB_MAXSLOTS(pcib);
4165 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4166 		pcifunchigh = 0;
4167 		f = 0;
4168 		DELAY(1);
4169 
4170 		/* If function 0 is not present, skip to the next slot. */
4171 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4172 			continue;
4173 		hdrtype = REG(PCIR_HDRTYPE, 1);
4174 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4175 			continue;
4176 		if (hdrtype & PCIM_MFDEV)
4177 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4178 		for (f = first_func; f <= pcifunchigh; f++)
4179 			pci_identify_function(pcib, dev, domain, busno, s, f);
4180 	}
4181 #undef REG
4182 }
4183 
4184 int
4185 pci_rescan_method(device_t dev)
4186 {
4187 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4188 	device_t pcib = device_get_parent(dev);
4189 	device_t child, *devlist, *unchanged;
4190 	int devcount, error, i, j, maxslots, oldcount;
4191 	int busno, domain, s, f, pcifunchigh;
4192 	uint8_t hdrtype;
4193 
4194 	/* No need to check for ARI on a rescan. */
4195 	error = device_get_children(dev, &devlist, &devcount);
4196 	if (error)
4197 		return (error);
4198 	if (devcount != 0) {
4199 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4200 		    M_NOWAIT | M_ZERO);
4201 		if (unchanged == NULL) {
4202 			free(devlist, M_TEMP);
4203 			return (ENOMEM);
4204 		}
4205 	} else
4206 		unchanged = NULL;
4207 
4208 	domain = pcib_get_domain(dev);
4209 	busno = pcib_get_bus(dev);
4210 	maxslots = PCIB_MAXSLOTS(pcib);
4211 	for (s = 0; s <= maxslots; s++) {
4212 		/* If function 0 is not present, skip to the next slot. */
4213 		f = 0;
4214 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4215 			continue;
4216 		pcifunchigh = 0;
4217 		hdrtype = REG(PCIR_HDRTYPE, 1);
4218 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4219 			continue;
4220 		if (hdrtype & PCIM_MFDEV)
4221 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4222 		for (f = 0; f <= pcifunchigh; f++) {
4223 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4224 				continue;
4225 
4226 			/*
4227 			 * Found a valid function.  Check if a
4228 			 * device_t for this device already exists.
4229 			 */
4230 			for (i = 0; i < devcount; i++) {
4231 				child = devlist[i];
4232 				if (child == NULL)
4233 					continue;
4234 				if (pci_get_slot(child) == s &&
4235 				    pci_get_function(child) == f) {
4236 					unchanged[i] = child;
4237 					goto next_func;
4238 				}
4239 			}
4240 
4241 			pci_identify_function(pcib, dev, domain, busno, s, f);
4242 		next_func:;
4243 		}
4244 	}
4245 
4246 	/* Remove devices that are no longer present. */
4247 	for (i = 0; i < devcount; i++) {
4248 		if (unchanged[i] != NULL)
4249 			continue;
4250 		device_delete_child(dev, devlist[i]);
4251 	}
4252 
4253 	free(devlist, M_TEMP);
4254 	oldcount = devcount;
4255 
4256 	/* Try to attach the devices just added. */
4257 	error = device_get_children(dev, &devlist, &devcount);
4258 	if (error) {
4259 		free(unchanged, M_TEMP);
4260 		return (error);
4261 	}
4262 
4263 	for (i = 0; i < devcount; i++) {
4264 		for (j = 0; j < oldcount; j++) {
4265 			if (devlist[i] == unchanged[j])
4266 				goto next_device;
4267 		}
4268 
4269 		device_probe_and_attach(devlist[i]);
4270 	next_device:;
4271 	}
4272 
4273 	free(unchanged, M_TEMP);
4274 	free(devlist, M_TEMP);
4275 	return (0);
4276 #undef REG
4277 }
4278 
4279 #ifdef PCI_IOV
4280 device_t
4281 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4282     uint16_t did)
4283 {
4284 	struct pci_devinfo *vf_dinfo;
4285 	device_t pcib;
4286 	int busno, slot, func;
4287 
4288 	pcib = device_get_parent(bus);
4289 
4290 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4291 
4292 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4293 	    slot, func, vid, did);
4294 
4295 	vf_dinfo->cfg.flags |= PCICFG_VF;
4296 	pci_add_child(bus, vf_dinfo);
4297 
4298 	return (vf_dinfo->cfg.dev);
4299 }
4300 
4301 device_t
4302 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4303     uint16_t vid, uint16_t did)
4304 {
4305 
4306 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4307 }
4308 #endif
4309 
4310 /*
4311  * For PCIe device set Max_Payload_Size to match PCIe root's.
4312  */
4313 static void
4314 pcie_setup_mps(device_t dev)
4315 {
4316 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4317 	device_t root;
4318 	uint16_t rmps, mmps, mps;
4319 
4320 	if (dinfo->cfg.pcie.pcie_location == 0)
4321 		return;
4322 	root = pci_find_pcie_root_port(dev);
4323 	if (root == NULL)
4324 		return;
4325 	/* Check whether the MPS is already configured. */
4326 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4327 	    PCIEM_CTL_MAX_PAYLOAD;
4328 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4329 	    PCIEM_CTL_MAX_PAYLOAD;
4330 	if (mps == rmps)
4331 		return;
4332 	/* Check whether the device is capable of the root's MPS. */
4333 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4334 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4335 	if (rmps > mmps) {
4336 		/*
4337 		 * The device is unable to handle root's MPS.  Limit root.
4338 		 * XXX: We should traverse through all the tree, applying
4339 		 * it to all the devices.
4340 		 */
4341 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4342 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4343 	} else {
4344 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4345 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4346 	}
4347 }
4348 
4349 static void
4350 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4351 {
4352 	int aer;
4353 	uint32_t r;
4354 	uint16_t r2;
4355 
4356 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4357 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4358 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4359 		    PCIER_ROOT_CTL, 2);
4360 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4361 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4362 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4363 		    PCIER_ROOT_CTL, r2, 2);
4364 	}
4365 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4366 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4367 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4368 		if (r != 0 && bootverbose) {
4369 			pci_printf(&dinfo->cfg,
4370 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4371 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4372 			    4));
4373 		}
4374 
4375 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4376 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4377 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4378 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4379 		    PCIM_AER_UC_POISONED_TLP |
4380 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4381 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4382 		    PCIM_AER_UC_COMPLETER_ABORT |
4383 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4384 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4385 		    PCIM_AER_UC_MALFORMED_TLP |
4386 		    PCIM_AER_UC_ECRC_ERROR |
4387 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4388 		    PCIM_AER_UC_ACS_VIOLATION |
4389 		    PCIM_AER_UC_INTERNAL_ERROR |
4390 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4391 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4392 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4393 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4394 
4395 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4396 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4397 		if (r != 0 && bootverbose) {
4398 			pci_printf(&dinfo->cfg,
4399 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4400 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4401 			    4));
4402 		}
4403 
4404 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4405 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4406 		    PCIM_AER_COR_BAD_TLP |
4407 		    PCIM_AER_COR_BAD_DLLP |
4408 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4409 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4410 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4411 		    PCIM_AER_COR_INTERNAL_ERROR |
4412 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4413 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4414 
4415 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4416 		    PCIER_DEVICE_CTL, 2);
4417 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4418 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4419 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4420 		    PCIER_DEVICE_CTL, r, 2);
4421 	}
4422 }
4423 
4424 void
4425 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4426 {
4427 	device_t dev;
4428 
4429 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4430 	device_set_ivars(dev, dinfo);
4431 	resource_list_init(&dinfo->resources);
4432 	pci_cfg_save(dev, dinfo, 0);
4433 	pci_cfg_restore(dev, dinfo);
4434 	pci_print_verbose(dinfo);
4435 	pci_add_resources(bus, dev, 0, 0);
4436 	if (pci_enable_mps_tune)
4437 		pcie_setup_mps(dev);
4438 	pci_child_added(dinfo->cfg.dev);
4439 
4440 	if (pci_clear_aer_on_attach)
4441 		pci_add_child_clear_aer(dev, dinfo);
4442 
4443 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4444 }
4445 
4446 void
4447 pci_child_added_method(device_t dev, device_t child)
4448 {
4449 
4450 }
4451 
4452 static int
4453 pci_probe(device_t dev)
4454 {
4455 
4456 	device_set_desc(dev, "PCI bus");
4457 
4458 	/* Allow other subclasses to override this driver. */
4459 	return (BUS_PROBE_GENERIC);
4460 }
4461 
4462 int
4463 pci_attach_common(device_t dev)
4464 {
4465 	struct pci_softc *sc;
4466 	int busno, domain;
4467 #ifdef PCI_RES_BUS
4468 	int rid;
4469 #endif
4470 
4471 	sc = device_get_softc(dev);
4472 	domain = pcib_get_domain(dev);
4473 	busno = pcib_get_bus(dev);
4474 #ifdef PCI_RES_BUS
4475 	rid = 0;
4476 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4477 	    1, 0);
4478 	if (sc->sc_bus == NULL) {
4479 		device_printf(dev, "failed to allocate bus number\n");
4480 		return (ENXIO);
4481 	}
4482 #endif
4483 	if (bootverbose)
4484 		device_printf(dev, "domain=%d, physical bus=%d\n",
4485 		    domain, busno);
4486 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4487 	return (0);
4488 }
4489 
4490 int
4491 pci_attach(device_t dev)
4492 {
4493 	int busno, domain, error;
4494 
4495 	error = pci_attach_common(dev);
4496 	if (error)
4497 		return (error);
4498 
4499 	/*
4500 	 * Since there can be multiple independently numbered PCI
4501 	 * buses on systems with multiple PCI domains, we can't use
4502 	 * the unit number to decide which bus we are probing. We ask
4503 	 * the parent pcib what our domain and bus numbers are.
4504 	 */
4505 	domain = pcib_get_domain(dev);
4506 	busno = pcib_get_bus(dev);
4507 	pci_add_children(dev, domain, busno);
4508 	return (bus_generic_attach(dev));
4509 }
4510 
4511 int
4512 pci_detach(device_t dev)
4513 {
4514 #ifdef PCI_RES_BUS
4515 	struct pci_softc *sc;
4516 #endif
4517 	int error;
4518 
4519 	error = bus_generic_detach(dev);
4520 	if (error)
4521 		return (error);
4522 #ifdef PCI_RES_BUS
4523 	sc = device_get_softc(dev);
4524 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4525 	if (error)
4526 		return (error);
4527 #endif
4528 	return (device_delete_children(dev));
4529 }
4530 
4531 static void
4532 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4533 {
4534 	int line, unit;
4535 	const char *at;
4536 	char me1[24], me2[32];
4537 	uint8_t b, s, f;
4538 	uint32_t d;
4539 	device_location_cache_t *cache;
4540 
4541 	d = pci_get_domain(child);
4542 	b = pci_get_bus(child);
4543 	s = pci_get_slot(child);
4544 	f = pci_get_function(child);
4545 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4546 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4547 	line = 0;
4548 	cache = dev_wired_cache_init();
4549 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4550 		resource_string_value(name, unit, "at", &at);
4551 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4552 			*unitp = unit;
4553 			break;
4554 		}
4555 		if (dev_wired_cache_match(cache, child, at)) {
4556 			*unitp = unit;
4557 			break;
4558 		}
4559 	}
4560 	dev_wired_cache_fini(cache);
4561 }
4562 
4563 static void
4564 pci_set_power_child(device_t dev, device_t child, int state)
4565 {
4566 	device_t pcib;
4567 	int dstate;
4568 
4569 	/*
4570 	 * Set the device to the given state.  If the firmware suggests
4571 	 * a different power state, use it instead.  If power management
4572 	 * is not present, the firmware is responsible for managing
4573 	 * device power.  Skip children who aren't attached since they
4574 	 * are handled separately.
4575 	 */
4576 	pcib = device_get_parent(dev);
4577 	dstate = state;
4578 	if (device_is_attached(child) &&
4579 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4580 		pci_set_powerstate(child, dstate);
4581 }
4582 
4583 int
4584 pci_suspend_child(device_t dev, device_t child)
4585 {
4586 	struct pci_devinfo *dinfo;
4587 	struct resource_list_entry *rle;
4588 	int error;
4589 
4590 	dinfo = device_get_ivars(child);
4591 
4592 	/*
4593 	 * Save the PCI configuration space for the child and set the
4594 	 * device in the appropriate power state for this sleep state.
4595 	 */
4596 	pci_cfg_save(child, dinfo, 0);
4597 
4598 	/* Suspend devices before potentially powering them down. */
4599 	error = bus_generic_suspend_child(dev, child);
4600 
4601 	if (error)
4602 		return (error);
4603 
4604 	if (pci_do_power_suspend) {
4605 		/*
4606 		 * Make sure this device's interrupt handler is not invoked
4607 		 * in the case the device uses a shared interrupt that can
4608 		 * be raised by some other device.
4609 		 * This is applicable only to regular (legacy) PCI interrupts
4610 		 * as MSI/MSI-X interrupts are never shared.
4611 		 */
4612 		rle = resource_list_find(&dinfo->resources,
4613 		    SYS_RES_IRQ, 0);
4614 		if (rle != NULL && rle->res != NULL)
4615 			(void)bus_suspend_intr(child, rle->res);
4616 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4617 	}
4618 
4619 	return (0);
4620 }
4621 
4622 int
4623 pci_resume_child(device_t dev, device_t child)
4624 {
4625 	struct pci_devinfo *dinfo;
4626 	struct resource_list_entry *rle;
4627 
4628 	if (pci_do_power_resume)
4629 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4630 
4631 	dinfo = device_get_ivars(child);
4632 	pci_cfg_restore(child, dinfo);
4633 	if (!device_is_attached(child))
4634 		pci_cfg_save(child, dinfo, 1);
4635 
4636 	bus_generic_resume_child(dev, child);
4637 
4638 	/*
4639 	 * Allow interrupts only after fully resuming the driver and hardware.
4640 	 */
4641 	if (pci_do_power_suspend) {
4642 		/* See pci_suspend_child for details. */
4643 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4644 		if (rle != NULL && rle->res != NULL)
4645 			(void)bus_resume_intr(child, rle->res);
4646 	}
4647 
4648 	return (0);
4649 }
4650 
4651 int
4652 pci_resume(device_t dev)
4653 {
4654 	device_t child, *devlist;
4655 	int error, i, numdevs;
4656 
4657 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4658 		return (error);
4659 
4660 	/*
4661 	 * Resume critical devices first, then everything else later.
4662 	 */
4663 	for (i = 0; i < numdevs; i++) {
4664 		child = devlist[i];
4665 		switch (pci_get_class(child)) {
4666 		case PCIC_DISPLAY:
4667 		case PCIC_MEMORY:
4668 		case PCIC_BRIDGE:
4669 		case PCIC_BASEPERIPH:
4670 			BUS_RESUME_CHILD(dev, child);
4671 			break;
4672 		}
4673 	}
4674 	for (i = 0; i < numdevs; i++) {
4675 		child = devlist[i];
4676 		switch (pci_get_class(child)) {
4677 		case PCIC_DISPLAY:
4678 		case PCIC_MEMORY:
4679 		case PCIC_BRIDGE:
4680 		case PCIC_BASEPERIPH:
4681 			break;
4682 		default:
4683 			BUS_RESUME_CHILD(dev, child);
4684 		}
4685 	}
4686 	free(devlist, M_TEMP);
4687 	return (0);
4688 }
4689 
4690 static void
4691 pci_load_vendor_data(void)
4692 {
4693 	caddr_t data;
4694 	void *ptr;
4695 	size_t sz;
4696 
4697 	data = preload_search_by_type("pci_vendor_data");
4698 	if (data != NULL) {
4699 		ptr = preload_fetch_addr(data);
4700 		sz = preload_fetch_size(data);
4701 		if (ptr != NULL && sz != 0) {
4702 			pci_vendordata = ptr;
4703 			pci_vendordata_size = sz;
4704 			/* terminate the database */
4705 			pci_vendordata[pci_vendordata_size] = '\n';
4706 		}
4707 	}
4708 }
4709 
4710 void
4711 pci_driver_added(device_t dev, driver_t *driver)
4712 {
4713 	int numdevs;
4714 	device_t *devlist;
4715 	device_t child;
4716 	struct pci_devinfo *dinfo;
4717 	int i;
4718 
4719 	if (bootverbose)
4720 		device_printf(dev, "driver added\n");
4721 	DEVICE_IDENTIFY(driver, dev);
4722 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4723 		return;
4724 	for (i = 0; i < numdevs; i++) {
4725 		child = devlist[i];
4726 		if (device_get_state(child) != DS_NOTPRESENT)
4727 			continue;
4728 		dinfo = device_get_ivars(child);
4729 		pci_print_verbose(dinfo);
4730 		if (bootverbose)
4731 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4732 		pci_cfg_restore(child, dinfo);
4733 		if (device_probe_and_attach(child) != 0)
4734 			pci_child_detached(dev, child);
4735 	}
4736 	free(devlist, M_TEMP);
4737 }
4738 
4739 int
4740 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4741     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4742 {
4743 	struct pci_devinfo *dinfo;
4744 	struct msix_table_entry *mte;
4745 	struct msix_vector *mv;
4746 	uint64_t addr;
4747 	uint32_t data;
4748 	void *cookie;
4749 	int error, rid;
4750 
4751 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4752 	    arg, &cookie);
4753 	if (error)
4754 		return (error);
4755 
4756 	/* If this is not a direct child, just bail out. */
4757 	if (device_get_parent(child) != dev) {
4758 		*cookiep = cookie;
4759 		return(0);
4760 	}
4761 
4762 	rid = rman_get_rid(irq);
4763 	if (rid == 0) {
4764 		/* Make sure that INTx is enabled */
4765 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4766 	} else {
4767 		/*
4768 		 * Check to see if the interrupt is MSI or MSI-X.
4769 		 * Ask our parent to map the MSI and give
4770 		 * us the address and data register values.
4771 		 * If we fail for some reason, teardown the
4772 		 * interrupt handler.
4773 		 */
4774 		dinfo = device_get_ivars(child);
4775 		if (dinfo->cfg.msi.msi_alloc > 0) {
4776 			if (dinfo->cfg.msi.msi_addr == 0) {
4777 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4778 			    ("MSI has handlers, but vectors not mapped"));
4779 				error = PCIB_MAP_MSI(device_get_parent(dev),
4780 				    child, rman_get_start(irq), &addr, &data);
4781 				if (error)
4782 					goto bad;
4783 				dinfo->cfg.msi.msi_addr = addr;
4784 				dinfo->cfg.msi.msi_data = data;
4785 			}
4786 			if (dinfo->cfg.msi.msi_handlers == 0)
4787 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4788 				    dinfo->cfg.msi.msi_data);
4789 			dinfo->cfg.msi.msi_handlers++;
4790 		} else {
4791 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4792 			    ("No MSI or MSI-X interrupts allocated"));
4793 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4794 			    ("MSI-X index too high"));
4795 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4796 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4797 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4798 			KASSERT(mv->mv_irq == rman_get_start(irq),
4799 			    ("IRQ mismatch"));
4800 			if (mv->mv_address == 0) {
4801 				KASSERT(mte->mte_handlers == 0,
4802 		    ("MSI-X table entry has handlers, but vector not mapped"));
4803 				error = PCIB_MAP_MSI(device_get_parent(dev),
4804 				    child, rman_get_start(irq), &addr, &data);
4805 				if (error)
4806 					goto bad;
4807 				mv->mv_address = addr;
4808 				mv->mv_data = data;
4809 			}
4810 
4811 			/*
4812 			 * The MSIX table entry must be made valid by
4813 			 * incrementing the mte_handlers before
4814 			 * calling pci_enable_msix() and
4815 			 * pci_resume_msix(). Else the MSIX rewrite
4816 			 * table quirk will not work as expected.
4817 			 */
4818 			mte->mte_handlers++;
4819 			if (mte->mte_handlers == 1) {
4820 				pci_enable_msix(child, rid - 1, mv->mv_address,
4821 				    mv->mv_data);
4822 				pci_unmask_msix(child, rid - 1);
4823 			}
4824 		}
4825 
4826 		/*
4827 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4828 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4829 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4830 		 */
4831 		if (!pci_has_quirk(pci_get_devid(child),
4832 		    PCI_QUIRK_MSI_INTX_BUG))
4833 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4834 		else
4835 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4836 	bad:
4837 		if (error) {
4838 			(void)bus_generic_teardown_intr(dev, child, irq,
4839 			    cookie);
4840 			return (error);
4841 		}
4842 	}
4843 	*cookiep = cookie;
4844 	return (0);
4845 }
4846 
4847 int
4848 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4849     void *cookie)
4850 {
4851 	struct msix_table_entry *mte;
4852 	struct resource_list_entry *rle;
4853 	struct pci_devinfo *dinfo;
4854 	int error, rid;
4855 
4856 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4857 		return (EINVAL);
4858 
4859 	/* If this isn't a direct child, just bail out */
4860 	if (device_get_parent(child) != dev)
4861 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4862 
4863 	rid = rman_get_rid(irq);
4864 	if (rid == 0) {
4865 		/* Mask INTx */
4866 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4867 	} else {
4868 		/*
4869 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4870 		 * decrement the appropriate handlers count and mask the
4871 		 * MSI-X message, or disable MSI messages if the count
4872 		 * drops to 0.
4873 		 */
4874 		dinfo = device_get_ivars(child);
4875 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4876 		if (rle->res != irq)
4877 			return (EINVAL);
4878 		if (dinfo->cfg.msi.msi_alloc > 0) {
4879 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4880 			    ("MSI-X index too high"));
4881 			if (dinfo->cfg.msi.msi_handlers == 0)
4882 				return (EINVAL);
4883 			dinfo->cfg.msi.msi_handlers--;
4884 			if (dinfo->cfg.msi.msi_handlers == 0)
4885 				pci_disable_msi(child);
4886 		} else {
4887 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4888 			    ("No MSI or MSI-X interrupts allocated"));
4889 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4890 			    ("MSI-X index too high"));
4891 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4892 			if (mte->mte_handlers == 0)
4893 				return (EINVAL);
4894 			mte->mte_handlers--;
4895 			if (mte->mte_handlers == 0)
4896 				pci_mask_msix(child, rid - 1);
4897 		}
4898 	}
4899 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4900 	if (rid > 0)
4901 		KASSERT(error == 0,
4902 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4903 	return (error);
4904 }
4905 
4906 int
4907 pci_print_child(device_t dev, device_t child)
4908 {
4909 	struct pci_devinfo *dinfo;
4910 	struct resource_list *rl;
4911 	int retval = 0;
4912 
4913 	dinfo = device_get_ivars(child);
4914 	rl = &dinfo->resources;
4915 
4916 	retval += bus_print_child_header(dev, child);
4917 
4918 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4919 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4920 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4921 	if (device_get_flags(dev))
4922 		retval += printf(" flags %#x", device_get_flags(dev));
4923 
4924 	retval += printf(" at device %d.%d", pci_get_slot(child),
4925 	    pci_get_function(child));
4926 
4927 	retval += bus_print_child_domain(dev, child);
4928 	retval += bus_print_child_footer(dev, child);
4929 
4930 	return (retval);
4931 }
4932 
4933 static const struct
4934 {
4935 	int		class;
4936 	int		subclass;
4937 	int		report; /* 0 = bootverbose, 1 = always */
4938 	const char	*desc;
4939 } pci_nomatch_tab[] = {
4940 	{PCIC_OLD,		-1,			1, "old"},
4941 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4942 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4943 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4944 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4945 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4946 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4947 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4948 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4949 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4950 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4951 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4952 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4953 	{PCIC_NETWORK,		-1,			1, "network"},
4954 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4955 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4956 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4957 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4958 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4959 	{PCIC_DISPLAY,		-1,			1, "display"},
4960 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4961 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4962 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4963 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4964 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4965 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4966 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4967 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4968 	{PCIC_MEMORY,		-1,			1, "memory"},
4969 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4970 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4971 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4972 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4973 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4974 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4975 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4976 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4977 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4978 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4979 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4980 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4981 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4982 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4983 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4984 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4985 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4986 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4987 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4988 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4989 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4990 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4991 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4992 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4993 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4994 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4995 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4996 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4997 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4998 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4999 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5000 	{PCIC_DOCKING,		-1,			1, "docking station"},
5001 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5002 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5003 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5004 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5005 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5006 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5007 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5008 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5009 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5010 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5011 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5012 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5013 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5014 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5015 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5016 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5017 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5018 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5019 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5020 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5021 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5022 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5023 	{PCIC_DASP,		-1,			0, "dasp"},
5024 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5025 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5026 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5027 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5028 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5029 	{0, 0, 0,		NULL}
5030 };
5031 
5032 void
5033 pci_probe_nomatch(device_t dev, device_t child)
5034 {
5035 	int i, report;
5036 	const char *cp, *scp;
5037 	char *device;
5038 
5039 	/*
5040 	 * Look for a listing for this device in a loaded device database.
5041 	 */
5042 	report = 1;
5043 	if ((device = pci_describe_device(child)) != NULL) {
5044 		device_printf(dev, "<%s>", device);
5045 		free(device, M_DEVBUF);
5046 	} else {
5047 		/*
5048 		 * Scan the class/subclass descriptions for a general
5049 		 * description.
5050 		 */
5051 		cp = "unknown";
5052 		scp = NULL;
5053 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5054 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5055 				if (pci_nomatch_tab[i].subclass == -1) {
5056 					cp = pci_nomatch_tab[i].desc;
5057 					report = pci_nomatch_tab[i].report;
5058 				} else if (pci_nomatch_tab[i].subclass ==
5059 				    pci_get_subclass(child)) {
5060 					scp = pci_nomatch_tab[i].desc;
5061 					report = pci_nomatch_tab[i].report;
5062 				}
5063 			}
5064 		}
5065 		if (report || bootverbose) {
5066 			device_printf(dev, "<%s%s%s>",
5067 			    cp ? cp : "",
5068 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5069 			    scp ? scp : "");
5070 		}
5071 	}
5072 	if (report || bootverbose) {
5073 		printf(" at device %d.%d (no driver attached)\n",
5074 		    pci_get_slot(child), pci_get_function(child));
5075 	}
5076 	pci_cfg_save(child, device_get_ivars(child), 1);
5077 }
5078 
5079 void
5080 pci_child_detached(device_t dev, device_t child)
5081 {
5082 	struct pci_devinfo *dinfo;
5083 	struct resource_list *rl;
5084 
5085 	dinfo = device_get_ivars(child);
5086 	rl = &dinfo->resources;
5087 
5088 	/*
5089 	 * Have to deallocate IRQs before releasing any MSI messages and
5090 	 * have to release MSI messages before deallocating any memory
5091 	 * BARs.
5092 	 */
5093 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5094 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5095 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5096 		if (dinfo->cfg.msi.msi_alloc != 0)
5097 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5098 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5099 		else
5100 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5101 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5102 		(void)pci_release_msi(child);
5103 	}
5104 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5105 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5106 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5107 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5108 #ifdef PCI_RES_BUS
5109 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5110 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5111 #endif
5112 
5113 	pci_cfg_save(child, dinfo, 1);
5114 }
5115 
5116 /*
5117  * Parse the PCI device database, if loaded, and return a pointer to a
5118  * description of the device.
5119  *
5120  * The database is flat text formatted as follows:
5121  *
5122  * Any line not in a valid format is ignored.
5123  * Lines are terminated with newline '\n' characters.
5124  *
5125  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5126  * the vendor name.
5127  *
5128  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5129  * - devices cannot be listed without a corresponding VENDOR line.
5130  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5131  * another TAB, then the device name.
5132  */
5133 
5134 /*
5135  * Assuming (ptr) points to the beginning of a line in the database,
5136  * return the vendor or device and description of the next entry.
5137  * The value of (vendor) or (device) inappropriate for the entry type
5138  * is set to -1.  Returns nonzero at the end of the database.
5139  *
5140  * Note that this is slightly unrobust in the face of corrupt data;
5141  * we attempt to safeguard against this by spamming the end of the
5142  * database with a newline when we initialise.
5143  */
5144 static int
5145 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5146 {
5147 	char	*cp = *ptr;
5148 	int	left;
5149 
5150 	*device = -1;
5151 	*vendor = -1;
5152 	**desc = '\0';
5153 	for (;;) {
5154 		left = pci_vendordata_size - (cp - pci_vendordata);
5155 		if (left <= 0) {
5156 			*ptr = cp;
5157 			return(1);
5158 		}
5159 
5160 		/* vendor entry? */
5161 		if (*cp != '\t' &&
5162 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5163 			break;
5164 		/* device entry? */
5165 		if (*cp == '\t' &&
5166 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5167 			break;
5168 
5169 		/* skip to next line */
5170 		while (*cp != '\n' && left > 0) {
5171 			cp++;
5172 			left--;
5173 		}
5174 		if (*cp == '\n') {
5175 			cp++;
5176 			left--;
5177 		}
5178 	}
5179 	/* skip to next line */
5180 	while (*cp != '\n' && left > 0) {
5181 		cp++;
5182 		left--;
5183 	}
5184 	if (*cp == '\n' && left > 0)
5185 		cp++;
5186 	*ptr = cp;
5187 	return(0);
5188 }
5189 
5190 static char *
5191 pci_describe_device(device_t dev)
5192 {
5193 	int	vendor, device;
5194 	char	*desc, *vp, *dp, *line;
5195 
5196 	desc = vp = dp = NULL;
5197 
5198 	/*
5199 	 * If we have no vendor data, we can't do anything.
5200 	 */
5201 	if (pci_vendordata == NULL)
5202 		goto out;
5203 
5204 	/*
5205 	 * Scan the vendor data looking for this device
5206 	 */
5207 	line = pci_vendordata;
5208 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5209 		goto out;
5210 	for (;;) {
5211 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5212 			goto out;
5213 		if (vendor == pci_get_vendor(dev))
5214 			break;
5215 	}
5216 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5217 		goto out;
5218 	for (;;) {
5219 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5220 			*dp = 0;
5221 			break;
5222 		}
5223 		if (vendor != -1) {
5224 			*dp = 0;
5225 			break;
5226 		}
5227 		if (device == pci_get_device(dev))
5228 			break;
5229 	}
5230 	if (dp[0] == '\0')
5231 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5232 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5233 	    NULL)
5234 		sprintf(desc, "%s, %s", vp, dp);
5235 out:
5236 	if (vp != NULL)
5237 		free(vp, M_DEVBUF);
5238 	if (dp != NULL)
5239 		free(dp, M_DEVBUF);
5240 	return(desc);
5241 }
5242 
5243 int
5244 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5245 {
5246 	struct pci_devinfo *dinfo;
5247 	pcicfgregs *cfg;
5248 
5249 	dinfo = device_get_ivars(child);
5250 	cfg = &dinfo->cfg;
5251 
5252 	switch (which) {
5253 	case PCI_IVAR_ETHADDR:
5254 		/*
5255 		 * The generic accessor doesn't deal with failure, so
5256 		 * we set the return value, then return an error.
5257 		 */
5258 		*((uint8_t **) result) = NULL;
5259 		return (EINVAL);
5260 	case PCI_IVAR_SUBVENDOR:
5261 		*result = cfg->subvendor;
5262 		break;
5263 	case PCI_IVAR_SUBDEVICE:
5264 		*result = cfg->subdevice;
5265 		break;
5266 	case PCI_IVAR_VENDOR:
5267 		*result = cfg->vendor;
5268 		break;
5269 	case PCI_IVAR_DEVICE:
5270 		*result = cfg->device;
5271 		break;
5272 	case PCI_IVAR_DEVID:
5273 		*result = (cfg->device << 16) | cfg->vendor;
5274 		break;
5275 	case PCI_IVAR_CLASS:
5276 		*result = cfg->baseclass;
5277 		break;
5278 	case PCI_IVAR_SUBCLASS:
5279 		*result = cfg->subclass;
5280 		break;
5281 	case PCI_IVAR_PROGIF:
5282 		*result = cfg->progif;
5283 		break;
5284 	case PCI_IVAR_REVID:
5285 		*result = cfg->revid;
5286 		break;
5287 	case PCI_IVAR_INTPIN:
5288 		*result = cfg->intpin;
5289 		break;
5290 	case PCI_IVAR_IRQ:
5291 		*result = cfg->intline;
5292 		break;
5293 	case PCI_IVAR_DOMAIN:
5294 		*result = cfg->domain;
5295 		break;
5296 	case PCI_IVAR_BUS:
5297 		*result = cfg->bus;
5298 		break;
5299 	case PCI_IVAR_SLOT:
5300 		*result = cfg->slot;
5301 		break;
5302 	case PCI_IVAR_FUNCTION:
5303 		*result = cfg->func;
5304 		break;
5305 	case PCI_IVAR_CMDREG:
5306 		*result = cfg->cmdreg;
5307 		break;
5308 	case PCI_IVAR_CACHELNSZ:
5309 		*result = cfg->cachelnsz;
5310 		break;
5311 	case PCI_IVAR_MINGNT:
5312 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5313 			*result = -1;
5314 			return (EINVAL);
5315 		}
5316 		*result = cfg->mingnt;
5317 		break;
5318 	case PCI_IVAR_MAXLAT:
5319 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5320 			*result = -1;
5321 			return (EINVAL);
5322 		}
5323 		*result = cfg->maxlat;
5324 		break;
5325 	case PCI_IVAR_LATTIMER:
5326 		*result = cfg->lattimer;
5327 		break;
5328 	default:
5329 		return (ENOENT);
5330 	}
5331 	return (0);
5332 }
5333 
5334 int
5335 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5336 {
5337 	struct pci_devinfo *dinfo;
5338 
5339 	dinfo = device_get_ivars(child);
5340 
5341 	switch (which) {
5342 	case PCI_IVAR_INTPIN:
5343 		dinfo->cfg.intpin = value;
5344 		return (0);
5345 	case PCI_IVAR_ETHADDR:
5346 	case PCI_IVAR_SUBVENDOR:
5347 	case PCI_IVAR_SUBDEVICE:
5348 	case PCI_IVAR_VENDOR:
5349 	case PCI_IVAR_DEVICE:
5350 	case PCI_IVAR_DEVID:
5351 	case PCI_IVAR_CLASS:
5352 	case PCI_IVAR_SUBCLASS:
5353 	case PCI_IVAR_PROGIF:
5354 	case PCI_IVAR_REVID:
5355 	case PCI_IVAR_IRQ:
5356 	case PCI_IVAR_DOMAIN:
5357 	case PCI_IVAR_BUS:
5358 	case PCI_IVAR_SLOT:
5359 	case PCI_IVAR_FUNCTION:
5360 		return (EINVAL);	/* disallow for now */
5361 
5362 	default:
5363 		return (ENOENT);
5364 	}
5365 }
5366 
5367 #include "opt_ddb.h"
5368 #ifdef DDB
5369 #include <ddb/ddb.h>
5370 #include <sys/cons.h>
5371 
5372 /*
5373  * List resources based on pci map registers, used for within ddb
5374  */
5375 
5376 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5377 {
5378 	struct pci_devinfo *dinfo;
5379 	struct devlist *devlist_head;
5380 	struct pci_conf *p;
5381 	const char *name;
5382 	int i, error, none_count;
5383 
5384 	none_count = 0;
5385 	/* get the head of the device queue */
5386 	devlist_head = &pci_devq;
5387 
5388 	/*
5389 	 * Go through the list of devices and print out devices
5390 	 */
5391 	for (error = 0, i = 0,
5392 	     dinfo = STAILQ_FIRST(devlist_head);
5393 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5394 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5395 		/* Populate pd_name and pd_unit */
5396 		name = NULL;
5397 		if (dinfo->cfg.dev)
5398 			name = device_get_name(dinfo->cfg.dev);
5399 
5400 		p = &dinfo->conf;
5401 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5402 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5403 			(name && *name) ? name : "none",
5404 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5405 			none_count++,
5406 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5407 			p->pc_sel.pc_func, (p->pc_class << 16) |
5408 			(p->pc_subclass << 8) | p->pc_progif,
5409 			(p->pc_subdevice << 16) | p->pc_subvendor,
5410 			(p->pc_device << 16) | p->pc_vendor,
5411 			p->pc_revid, p->pc_hdr);
5412 	}
5413 }
5414 #endif /* DDB */
5415 
5416 struct resource *
5417 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5418     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5419     u_int flags)
5420 {
5421 	struct pci_devinfo *dinfo = device_get_ivars(child);
5422 	struct resource_list *rl = &dinfo->resources;
5423 	struct resource *res;
5424 	struct pci_map *pm;
5425 	uint16_t cmd;
5426 	pci_addr_t map, testval;
5427 	int mapsize;
5428 
5429 	res = NULL;
5430 
5431 	/* If rid is managed by EA, ignore it */
5432 	if (pci_ea_is_enabled(child, *rid))
5433 		goto out;
5434 
5435 	pm = pci_find_bar(child, *rid);
5436 	if (pm != NULL) {
5437 		/* This is a BAR that we failed to allocate earlier. */
5438 		mapsize = pm->pm_size;
5439 		map = pm->pm_value;
5440 	} else {
5441 		/*
5442 		 * Weed out the bogons, and figure out how large the
5443 		 * BAR/map is.  BARs that read back 0 here are bogus
5444 		 * and unimplemented.  Note: atapci in legacy mode are
5445 		 * special and handled elsewhere in the code.  If you
5446 		 * have a atapci device in legacy mode and it fails
5447 		 * here, that other code is broken.
5448 		 */
5449 		pci_read_bar(child, *rid, &map, &testval, NULL);
5450 
5451 		/*
5452 		 * Determine the size of the BAR and ignore BARs with a size
5453 		 * of 0.  Device ROM BARs use a different mask value.
5454 		 */
5455 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5456 			mapsize = pci_romsize(testval);
5457 		else
5458 			mapsize = pci_mapsize(testval);
5459 		if (mapsize == 0)
5460 			goto out;
5461 		pm = pci_add_bar(child, *rid, map, mapsize);
5462 	}
5463 
5464 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5465 		if (type != SYS_RES_MEMORY) {
5466 			if (bootverbose)
5467 				device_printf(dev,
5468 				    "child %s requested type %d for rid %#x,"
5469 				    " but the BAR says it is an memio\n",
5470 				    device_get_nameunit(child), type, *rid);
5471 			goto out;
5472 		}
5473 	} else {
5474 		if (type != SYS_RES_IOPORT) {
5475 			if (bootverbose)
5476 				device_printf(dev,
5477 				    "child %s requested type %d for rid %#x,"
5478 				    " but the BAR says it is an ioport\n",
5479 				    device_get_nameunit(child), type, *rid);
5480 			goto out;
5481 		}
5482 	}
5483 
5484 	/*
5485 	 * For real BARs, we need to override the size that
5486 	 * the driver requests, because that's what the BAR
5487 	 * actually uses and we would otherwise have a
5488 	 * situation where we might allocate the excess to
5489 	 * another driver, which won't work.
5490 	 */
5491 	count = ((pci_addr_t)1 << mapsize) * num;
5492 	if (RF_ALIGNMENT(flags) < mapsize)
5493 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5494 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5495 		flags |= RF_PREFETCHABLE;
5496 
5497 	/*
5498 	 * Allocate enough resource, and then write back the
5499 	 * appropriate BAR for that resource.
5500 	 */
5501 	resource_list_add(rl, type, *rid, start, end, count);
5502 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5503 	    count, flags & ~RF_ACTIVE);
5504 	if (res == NULL) {
5505 		resource_list_delete(rl, type, *rid);
5506 		device_printf(child,
5507 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5508 		    count, *rid, type, start, end);
5509 		goto out;
5510 	}
5511 	if (bootverbose)
5512 		device_printf(child,
5513 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5514 		    count, *rid, type, rman_get_start(res));
5515 
5516 	/* Disable decoding via the CMD register before updating the BAR */
5517 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5518 	pci_write_config(child, PCIR_COMMAND,
5519 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5520 
5521 	map = rman_get_start(res);
5522 	pci_write_bar(child, pm, map);
5523 
5524 	/* Restore the original value of the CMD register */
5525 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5526 out:
5527 	return (res);
5528 }
5529 
5530 struct resource *
5531 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5532     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5533     u_int flags)
5534 {
5535 	struct pci_devinfo *dinfo;
5536 	struct resource_list *rl;
5537 	struct resource_list_entry *rle;
5538 	struct resource *res;
5539 	pcicfgregs *cfg;
5540 
5541 	/*
5542 	 * Perform lazy resource allocation
5543 	 */
5544 	dinfo = device_get_ivars(child);
5545 	rl = &dinfo->resources;
5546 	cfg = &dinfo->cfg;
5547 	switch (type) {
5548 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5549 	case PCI_RES_BUS:
5550 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5551 		    flags));
5552 #endif
5553 	case SYS_RES_IRQ:
5554 		/*
5555 		 * Can't alloc legacy interrupt once MSI messages have
5556 		 * been allocated.
5557 		 */
5558 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5559 		    cfg->msix.msix_alloc > 0))
5560 			return (NULL);
5561 
5562 		/*
5563 		 * If the child device doesn't have an interrupt
5564 		 * routed and is deserving of an interrupt, try to
5565 		 * assign it one.
5566 		 */
5567 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5568 		    (cfg->intpin != 0))
5569 			pci_assign_interrupt(dev, child, 0);
5570 		break;
5571 	case SYS_RES_IOPORT:
5572 	case SYS_RES_MEMORY:
5573 #ifdef NEW_PCIB
5574 		/*
5575 		 * PCI-PCI bridge I/O window resources are not BARs.
5576 		 * For those allocations just pass the request up the
5577 		 * tree.
5578 		 */
5579 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5580 			switch (*rid) {
5581 			case PCIR_IOBASEL_1:
5582 			case PCIR_MEMBASE_1:
5583 			case PCIR_PMBASEL_1:
5584 				/*
5585 				 * XXX: Should we bother creating a resource
5586 				 * list entry?
5587 				 */
5588 				return (bus_generic_alloc_resource(dev, child,
5589 				    type, rid, start, end, count, flags));
5590 			}
5591 		}
5592 #endif
5593 		/* Reserve resources for this BAR if needed. */
5594 		rle = resource_list_find(rl, type, *rid);
5595 		if (rle == NULL) {
5596 			res = pci_reserve_map(dev, child, type, rid, start, end,
5597 			    count, num, flags);
5598 			if (res == NULL)
5599 				return (NULL);
5600 		}
5601 	}
5602 	return (resource_list_alloc(rl, dev, child, type, rid,
5603 	    start, end, count, flags));
5604 }
5605 
5606 struct resource *
5607 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5608     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5609 {
5610 #ifdef PCI_IOV
5611 	struct pci_devinfo *dinfo;
5612 #endif
5613 
5614 	if (device_get_parent(child) != dev)
5615 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5616 		    type, rid, start, end, count, flags));
5617 
5618 #ifdef PCI_IOV
5619 	dinfo = device_get_ivars(child);
5620 	if (dinfo->cfg.flags & PCICFG_VF) {
5621 		switch (type) {
5622 		/* VFs can't have I/O BARs. */
5623 		case SYS_RES_IOPORT:
5624 			return (NULL);
5625 		case SYS_RES_MEMORY:
5626 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5627 			    start, end, count, flags));
5628 		}
5629 
5630 		/* Fall through for other types of resource allocations. */
5631 	}
5632 #endif
5633 
5634 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5635 	    count, 1, flags));
5636 }
5637 
5638 int
5639 pci_release_resource(device_t dev, device_t child, int type, int rid,
5640     struct resource *r)
5641 {
5642 	struct pci_devinfo *dinfo;
5643 	struct resource_list *rl;
5644 	pcicfgregs *cfg __unused;
5645 
5646 	if (device_get_parent(child) != dev)
5647 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5648 		    type, rid, r));
5649 
5650 	dinfo = device_get_ivars(child);
5651 	cfg = &dinfo->cfg;
5652 
5653 #ifdef PCI_IOV
5654 	if (cfg->flags & PCICFG_VF) {
5655 		switch (type) {
5656 		/* VFs can't have I/O BARs. */
5657 		case SYS_RES_IOPORT:
5658 			return (EDOOFUS);
5659 		case SYS_RES_MEMORY:
5660 			return (pci_vf_release_mem_resource(dev, child, rid,
5661 			    r));
5662 		}
5663 
5664 		/* Fall through for other types of resource allocations. */
5665 	}
5666 #endif
5667 
5668 #ifdef NEW_PCIB
5669 	/*
5670 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5671 	 * those allocations just pass the request up the tree.
5672 	 */
5673 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5674 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5675 		switch (rid) {
5676 		case PCIR_IOBASEL_1:
5677 		case PCIR_MEMBASE_1:
5678 		case PCIR_PMBASEL_1:
5679 			return (bus_generic_release_resource(dev, child, type,
5680 			    rid, r));
5681 		}
5682 	}
5683 #endif
5684 
5685 	rl = &dinfo->resources;
5686 	return (resource_list_release(rl, dev, child, type, rid, r));
5687 }
5688 
5689 int
5690 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5691     struct resource *r)
5692 {
5693 	struct pci_devinfo *dinfo;
5694 	int error;
5695 
5696 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5697 	if (error)
5698 		return (error);
5699 
5700 	/* Enable decoding in the command register when activating BARs. */
5701 	if (device_get_parent(child) == dev) {
5702 		/* Device ROMs need their decoding explicitly enabled. */
5703 		dinfo = device_get_ivars(child);
5704 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5705 			pci_write_bar(child, pci_find_bar(child, rid),
5706 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5707 		switch (type) {
5708 		case SYS_RES_IOPORT:
5709 		case SYS_RES_MEMORY:
5710 			error = PCI_ENABLE_IO(dev, child, type);
5711 			break;
5712 		}
5713 	}
5714 	return (error);
5715 }
5716 
5717 int
5718 pci_deactivate_resource(device_t dev, device_t child, int type,
5719     int rid, struct resource *r)
5720 {
5721 	struct pci_devinfo *dinfo;
5722 	int error;
5723 
5724 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5725 	if (error)
5726 		return (error);
5727 
5728 	/* Disable decoding for device ROMs. */
5729 	if (device_get_parent(child) == dev) {
5730 		dinfo = device_get_ivars(child);
5731 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5732 			pci_write_bar(child, pci_find_bar(child, rid),
5733 			    rman_get_start(r));
5734 	}
5735 	return (0);
5736 }
5737 
5738 void
5739 pci_child_deleted(device_t dev, device_t child)
5740 {
5741 	struct resource_list_entry *rle;
5742 	struct resource_list *rl;
5743 	struct pci_devinfo *dinfo;
5744 
5745 	dinfo = device_get_ivars(child);
5746 	rl = &dinfo->resources;
5747 
5748 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5749 
5750 	/* Turn off access to resources we're about to free */
5751 	if (bus_child_present(child) != 0) {
5752 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5753 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5754 
5755 		pci_disable_busmaster(child);
5756 	}
5757 
5758 	/* Free all allocated resources */
5759 	STAILQ_FOREACH(rle, rl, link) {
5760 		if (rle->res) {
5761 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5762 			    resource_list_busy(rl, rle->type, rle->rid)) {
5763 				pci_printf(&dinfo->cfg,
5764 				    "Resource still owned, oops. "
5765 				    "(type=%d, rid=%d, addr=%lx)\n",
5766 				    rle->type, rle->rid,
5767 				    rman_get_start(rle->res));
5768 				bus_release_resource(child, rle->type, rle->rid,
5769 				    rle->res);
5770 			}
5771 			resource_list_unreserve(rl, dev, child, rle->type,
5772 			    rle->rid);
5773 		}
5774 	}
5775 	resource_list_free(rl);
5776 
5777 	pci_freecfg(dinfo);
5778 }
5779 
5780 void
5781 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5782 {
5783 	struct pci_devinfo *dinfo;
5784 	struct resource_list *rl;
5785 	struct resource_list_entry *rle;
5786 
5787 	if (device_get_parent(child) != dev)
5788 		return;
5789 
5790 	dinfo = device_get_ivars(child);
5791 	rl = &dinfo->resources;
5792 	rle = resource_list_find(rl, type, rid);
5793 	if (rle == NULL)
5794 		return;
5795 
5796 	if (rle->res) {
5797 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5798 		    resource_list_busy(rl, type, rid)) {
5799 			device_printf(dev, "delete_resource: "
5800 			    "Resource still owned by child, oops. "
5801 			    "(type=%d, rid=%d, addr=%jx)\n",
5802 			    type, rid, rman_get_start(rle->res));
5803 			return;
5804 		}
5805 		resource_list_unreserve(rl, dev, child, type, rid);
5806 	}
5807 	resource_list_delete(rl, type, rid);
5808 }
5809 
5810 struct resource_list *
5811 pci_get_resource_list (device_t dev, device_t child)
5812 {
5813 	struct pci_devinfo *dinfo = device_get_ivars(child);
5814 
5815 	return (&dinfo->resources);
5816 }
5817 
5818 #ifdef IOMMU
5819 bus_dma_tag_t
5820 pci_get_dma_tag(device_t bus, device_t dev)
5821 {
5822 	bus_dma_tag_t tag;
5823 	struct pci_softc *sc;
5824 
5825 	if (device_get_parent(dev) == bus) {
5826 		/* try iommu and return if it works */
5827 		tag = iommu_get_dma_tag(bus, dev);
5828 	} else
5829 		tag = NULL;
5830 	if (tag == NULL) {
5831 		sc = device_get_softc(bus);
5832 		tag = sc->sc_dma_tag;
5833 	}
5834 	return (tag);
5835 }
5836 #else
5837 bus_dma_tag_t
5838 pci_get_dma_tag(device_t bus, device_t dev)
5839 {
5840 	struct pci_softc *sc = device_get_softc(bus);
5841 
5842 	return (sc->sc_dma_tag);
5843 }
5844 #endif
5845 
5846 uint32_t
5847 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5848 {
5849 	struct pci_devinfo *dinfo = device_get_ivars(child);
5850 	pcicfgregs *cfg = &dinfo->cfg;
5851 
5852 #ifdef PCI_IOV
5853 	/*
5854 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5855 	 * emulate them here.
5856 	 */
5857 	if (cfg->flags & PCICFG_VF) {
5858 		if (reg == PCIR_VENDOR) {
5859 			switch (width) {
5860 			case 4:
5861 				return (cfg->device << 16 | cfg->vendor);
5862 			case 2:
5863 				return (cfg->vendor);
5864 			case 1:
5865 				return (cfg->vendor & 0xff);
5866 			default:
5867 				return (0xffffffff);
5868 			}
5869 		} else if (reg == PCIR_DEVICE) {
5870 			switch (width) {
5871 			/* Note that an unaligned 4-byte read is an error. */
5872 			case 2:
5873 				return (cfg->device);
5874 			case 1:
5875 				return (cfg->device & 0xff);
5876 			default:
5877 				return (0xffffffff);
5878 			}
5879 		}
5880 	}
5881 #endif
5882 
5883 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5884 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5885 }
5886 
5887 void
5888 pci_write_config_method(device_t dev, device_t child, int reg,
5889     uint32_t val, int width)
5890 {
5891 	struct pci_devinfo *dinfo = device_get_ivars(child);
5892 	pcicfgregs *cfg = &dinfo->cfg;
5893 
5894 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5895 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5896 }
5897 
5898 int
5899 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
5900 {
5901 
5902 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5903 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5904 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5905 	return (0);
5906 }
5907 
5908 int
5909 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
5910 {
5911 	struct pci_devinfo *dinfo;
5912 	pcicfgregs *cfg;
5913 
5914 	dinfo = device_get_ivars(child);
5915 	cfg = &dinfo->cfg;
5916 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5917 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5918 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5919 	    cfg->progif);
5920 	return (0);
5921 }
5922 
5923 int
5924 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
5925     struct sbuf *sb)
5926 {
5927 	device_t parent = device_get_parent(bus);
5928 	int rv;
5929 
5930 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
5931 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
5932 		if (rv == 0) {
5933 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
5934 			    pci_get_function(child));
5935 		}
5936 		return (0);
5937 	}
5938 	return (bus_generic_get_device_path(bus, child, locator, sb));
5939 }
5940 
5941 int
5942 pci_assign_interrupt_method(device_t dev, device_t child)
5943 {
5944 	struct pci_devinfo *dinfo = device_get_ivars(child);
5945 	pcicfgregs *cfg = &dinfo->cfg;
5946 
5947 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5948 	    cfg->intpin));
5949 }
5950 
5951 static void
5952 pci_lookup(void *arg, const char *name, device_t *dev)
5953 {
5954 	long val;
5955 	char *end;
5956 	int domain, bus, slot, func;
5957 
5958 	if (*dev != NULL)
5959 		return;
5960 
5961 	/*
5962 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5963 	 * pciB:S:F.  In the latter case, the domain is assumed to
5964 	 * be zero.
5965 	 */
5966 	if (strncmp(name, "pci", 3) != 0)
5967 		return;
5968 	val = strtol(name + 3, &end, 10);
5969 	if (val < 0 || val > INT_MAX || *end != ':')
5970 		return;
5971 	domain = val;
5972 	val = strtol(end + 1, &end, 10);
5973 	if (val < 0 || val > INT_MAX || *end != ':')
5974 		return;
5975 	bus = val;
5976 	val = strtol(end + 1, &end, 10);
5977 	if (val < 0 || val > INT_MAX)
5978 		return;
5979 	slot = val;
5980 	if (*end == ':') {
5981 		val = strtol(end + 1, &end, 10);
5982 		if (val < 0 || val > INT_MAX || *end != '\0')
5983 			return;
5984 		func = val;
5985 	} else if (*end == '\0') {
5986 		func = slot;
5987 		slot = bus;
5988 		bus = domain;
5989 		domain = 0;
5990 	} else
5991 		return;
5992 
5993 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5994 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5995 		return;
5996 
5997 	*dev = pci_find_dbsf(domain, bus, slot, func);
5998 }
5999 
6000 static int
6001 pci_modevent(module_t mod, int what, void *arg)
6002 {
6003 	static struct cdev *pci_cdev;
6004 	static eventhandler_tag tag;
6005 
6006 	switch (what) {
6007 	case MOD_LOAD:
6008 		STAILQ_INIT(&pci_devq);
6009 		pci_generation = 0;
6010 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6011 		    "pci");
6012 		pci_load_vendor_data();
6013 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6014 		    1000);
6015 		break;
6016 
6017 	case MOD_UNLOAD:
6018 		if (tag != NULL)
6019 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6020 		destroy_dev(pci_cdev);
6021 		break;
6022 	}
6023 
6024 	return (0);
6025 }
6026 
6027 static void
6028 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6029 {
6030 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6031 	struct pcicfg_pcie *cfg;
6032 	int version, pos;
6033 
6034 	cfg = &dinfo->cfg.pcie;
6035 	pos = cfg->pcie_location;
6036 
6037 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6038 
6039 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6040 
6041 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6042 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6043 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6044 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6045 
6046 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6047 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6048 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6049 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6050 
6051 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6052 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6053 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6054 
6055 	if (version > 1) {
6056 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6057 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6058 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6059 	}
6060 #undef WREG
6061 }
6062 
6063 static void
6064 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6065 {
6066 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6067 	    dinfo->cfg.pcix.pcix_command,  2);
6068 }
6069 
6070 void
6071 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6072 {
6073 
6074 	/*
6075 	 * Restore the device to full power mode.  We must do this
6076 	 * before we restore the registers because moving from D3 to
6077 	 * D0 will cause the chip's BARs and some other registers to
6078 	 * be reset to some unknown power on reset values.  Cut down
6079 	 * the noise on boot by doing nothing if we are already in
6080 	 * state D0.
6081 	 */
6082 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6083 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6084 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6085 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6086 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6087 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6088 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6089 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6090 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6091 	case PCIM_HDRTYPE_NORMAL:
6092 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6093 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6094 		break;
6095 	case PCIM_HDRTYPE_BRIDGE:
6096 		pci_write_config(dev, PCIR_SECLAT_1,
6097 		    dinfo->cfg.bridge.br_seclat, 1);
6098 		pci_write_config(dev, PCIR_SUBBUS_1,
6099 		    dinfo->cfg.bridge.br_subbus, 1);
6100 		pci_write_config(dev, PCIR_SECBUS_1,
6101 		    dinfo->cfg.bridge.br_secbus, 1);
6102 		pci_write_config(dev, PCIR_PRIBUS_1,
6103 		    dinfo->cfg.bridge.br_pribus, 1);
6104 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6105 		    dinfo->cfg.bridge.br_control, 2);
6106 		break;
6107 	case PCIM_HDRTYPE_CARDBUS:
6108 		pci_write_config(dev, PCIR_SECLAT_2,
6109 		    dinfo->cfg.bridge.br_seclat, 1);
6110 		pci_write_config(dev, PCIR_SUBBUS_2,
6111 		    dinfo->cfg.bridge.br_subbus, 1);
6112 		pci_write_config(dev, PCIR_SECBUS_2,
6113 		    dinfo->cfg.bridge.br_secbus, 1);
6114 		pci_write_config(dev, PCIR_PRIBUS_2,
6115 		    dinfo->cfg.bridge.br_pribus, 1);
6116 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6117 		    dinfo->cfg.bridge.br_control, 2);
6118 		break;
6119 	}
6120 	pci_restore_bars(dev);
6121 
6122 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6123 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6124 
6125 	/*
6126 	 * Restore extended capabilities for PCI-Express and PCI-X
6127 	 */
6128 	if (dinfo->cfg.pcie.pcie_location != 0)
6129 		pci_cfg_restore_pcie(dev, dinfo);
6130 	if (dinfo->cfg.pcix.pcix_location != 0)
6131 		pci_cfg_restore_pcix(dev, dinfo);
6132 
6133 	/* Restore MSI and MSI-X configurations if they are present. */
6134 	if (dinfo->cfg.msi.msi_location != 0)
6135 		pci_resume_msi(dev);
6136 	if (dinfo->cfg.msix.msix_location != 0)
6137 		pci_resume_msix(dev);
6138 
6139 #ifdef PCI_IOV
6140 	if (dinfo->cfg.iov != NULL)
6141 		pci_iov_cfg_restore(dev, dinfo);
6142 #endif
6143 }
6144 
6145 static void
6146 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6147 {
6148 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6149 	struct pcicfg_pcie *cfg;
6150 	int version, pos;
6151 
6152 	cfg = &dinfo->cfg.pcie;
6153 	pos = cfg->pcie_location;
6154 
6155 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6156 
6157 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6158 
6159 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6160 
6161 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6162 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6163 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6164 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6165 
6166 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6167 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6168 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6169 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6170 
6171 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6172 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6173 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6174 
6175 	if (version > 1) {
6176 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6177 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6178 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6179 	}
6180 #undef RREG
6181 }
6182 
6183 static void
6184 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6185 {
6186 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6187 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6188 }
6189 
6190 void
6191 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6192 {
6193 	uint32_t cls;
6194 	int ps;
6195 
6196 	/*
6197 	 * Some drivers apparently write to these registers w/o updating our
6198 	 * cached copy.  No harm happens if we update the copy, so do so here
6199 	 * so we can restore them.  The COMMAND register is modified by the
6200 	 * bus w/o updating the cache.  This should represent the normally
6201 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6202 	 */
6203 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6204 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6205 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6206 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6207 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6208 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6209 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6210 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6211 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6212 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6213 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6214 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6215 	case PCIM_HDRTYPE_NORMAL:
6216 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6217 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6218 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6219 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6220 		break;
6221 	case PCIM_HDRTYPE_BRIDGE:
6222 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6223 		    PCIR_SECLAT_1, 1);
6224 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6225 		    PCIR_SUBBUS_1, 1);
6226 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6227 		    PCIR_SECBUS_1, 1);
6228 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6229 		    PCIR_PRIBUS_1, 1);
6230 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6231 		    PCIR_BRIDGECTL_1, 2);
6232 		break;
6233 	case PCIM_HDRTYPE_CARDBUS:
6234 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6235 		    PCIR_SECLAT_2, 1);
6236 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6237 		    PCIR_SUBBUS_2, 1);
6238 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6239 		    PCIR_SECBUS_2, 1);
6240 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6241 		    PCIR_PRIBUS_2, 1);
6242 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6243 		    PCIR_BRIDGECTL_2, 2);
6244 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6245 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6246 		break;
6247 	}
6248 
6249 	if (dinfo->cfg.pcie.pcie_location != 0)
6250 		pci_cfg_save_pcie(dev, dinfo);
6251 
6252 	if (dinfo->cfg.pcix.pcix_location != 0)
6253 		pci_cfg_save_pcix(dev, dinfo);
6254 
6255 #ifdef PCI_IOV
6256 	if (dinfo->cfg.iov != NULL)
6257 		pci_iov_cfg_save(dev, dinfo);
6258 #endif
6259 
6260 	/*
6261 	 * don't set the state for display devices, base peripherals and
6262 	 * memory devices since bad things happen when they are powered down.
6263 	 * We should (a) have drivers that can easily detach and (b) use
6264 	 * generic drivers for these devices so that some device actually
6265 	 * attaches.  We need to make sure that when we implement (a) we don't
6266 	 * power the device down on a reattach.
6267 	 */
6268 	cls = pci_get_class(dev);
6269 	if (!setstate)
6270 		return;
6271 	switch (pci_do_power_nodriver)
6272 	{
6273 		case 0:		/* NO powerdown at all */
6274 			return;
6275 		case 1:		/* Conservative about what to power down */
6276 			if (cls == PCIC_STORAGE)
6277 				return;
6278 			/*FALLTHROUGH*/
6279 		case 2:		/* Aggressive about what to power down */
6280 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6281 			    cls == PCIC_BASEPERIPH)
6282 				return;
6283 			/*FALLTHROUGH*/
6284 		case 3:		/* Power down everything */
6285 			break;
6286 	}
6287 	/*
6288 	 * PCI spec says we can only go into D3 state from D0 state.
6289 	 * Transition from D[12] into D0 before going to D3 state.
6290 	 */
6291 	ps = pci_get_powerstate(dev);
6292 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6293 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6294 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6295 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6296 }
6297 
6298 /* Wrapper APIs suitable for device driver use. */
6299 void
6300 pci_save_state(device_t dev)
6301 {
6302 	struct pci_devinfo *dinfo;
6303 
6304 	dinfo = device_get_ivars(dev);
6305 	pci_cfg_save(dev, dinfo, 0);
6306 }
6307 
6308 void
6309 pci_restore_state(device_t dev)
6310 {
6311 	struct pci_devinfo *dinfo;
6312 
6313 	dinfo = device_get_ivars(dev);
6314 	pci_cfg_restore(dev, dinfo);
6315 }
6316 
6317 static int
6318 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6319     uintptr_t *id)
6320 {
6321 
6322 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6323 }
6324 
6325 /* Find the upstream port of a given PCI device in a root complex. */
6326 device_t
6327 pci_find_pcie_root_port(device_t dev)
6328 {
6329 	struct pci_devinfo *dinfo;
6330 	devclass_t pci_class;
6331 	device_t pcib, bus;
6332 
6333 	pci_class = devclass_find("pci");
6334 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6335 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6336 
6337 	/*
6338 	 * Walk the bridge hierarchy until we find a PCI-e root
6339 	 * port or a non-PCI device.
6340 	 */
6341 	for (;;) {
6342 		bus = device_get_parent(dev);
6343 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6344 		    device_get_nameunit(dev)));
6345 
6346 		pcib = device_get_parent(bus);
6347 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6348 		    device_get_nameunit(bus)));
6349 
6350 		/*
6351 		 * pcib's parent must be a PCI bus for this to be a
6352 		 * PCI-PCI bridge.
6353 		 */
6354 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6355 			return (NULL);
6356 
6357 		dinfo = device_get_ivars(pcib);
6358 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6359 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6360 			return (pcib);
6361 
6362 		dev = pcib;
6363 	}
6364 }
6365 
6366 /*
6367  * Wait for pending transactions to complete on a PCI-express function.
6368  *
6369  * The maximum delay is specified in milliseconds in max_delay.  Note
6370  * that this function may sleep.
6371  *
6372  * Returns true if the function is idle and false if the timeout is
6373  * exceeded.  If dev is not a PCI-express function, this returns true.
6374  */
6375 bool
6376 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6377 {
6378 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6379 	uint16_t sta;
6380 	int cap;
6381 
6382 	cap = dinfo->cfg.pcie.pcie_location;
6383 	if (cap == 0)
6384 		return (true);
6385 
6386 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6387 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6388 		if (max_delay == 0)
6389 			return (false);
6390 
6391 		/* Poll once every 100 milliseconds up to the timeout. */
6392 		if (max_delay > 100) {
6393 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6394 			max_delay -= 100;
6395 		} else {
6396 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6397 			    C_HARDCLOCK);
6398 			max_delay = 0;
6399 		}
6400 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6401 	}
6402 
6403 	return (true);
6404 }
6405 
6406 /*
6407  * Determine the maximum Completion Timeout in microseconds.
6408  *
6409  * For non-PCI-express functions this returns 0.
6410  */
6411 int
6412 pcie_get_max_completion_timeout(device_t dev)
6413 {
6414 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6415 	int cap;
6416 
6417 	cap = dinfo->cfg.pcie.pcie_location;
6418 	if (cap == 0)
6419 		return (0);
6420 
6421 	/*
6422 	 * Functions using the 1.x spec use the default timeout range of
6423 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6424 	 * support programmable timeouts also use this range.
6425 	 */
6426 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6427 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6428 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6429 		return (50 * 1000);
6430 
6431 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6432 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6433 	case PCIEM_CTL2_COMP_TIMO_100US:
6434 		return (100);
6435 	case PCIEM_CTL2_COMP_TIMO_10MS:
6436 		return (10 * 1000);
6437 	case PCIEM_CTL2_COMP_TIMO_55MS:
6438 		return (55 * 1000);
6439 	case PCIEM_CTL2_COMP_TIMO_210MS:
6440 		return (210 * 1000);
6441 	case PCIEM_CTL2_COMP_TIMO_900MS:
6442 		return (900 * 1000);
6443 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6444 		return (3500 * 1000);
6445 	case PCIEM_CTL2_COMP_TIMO_13S:
6446 		return (13 * 1000 * 1000);
6447 	case PCIEM_CTL2_COMP_TIMO_64S:
6448 		return (64 * 1000 * 1000);
6449 	default:
6450 		return (50 * 1000);
6451 	}
6452 }
6453 
6454 void
6455 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6456 {
6457 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6458 	const char *s;
6459 	int aer;
6460 	uint32_t r, r1;
6461 	uint16_t rs;
6462 
6463 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6464 		s = "Correctable";
6465 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6466 		s = "Uncorrectable (Non-Fatal)";
6467 	else
6468 		s = "Uncorrectable (Fatal)";
6469 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6470 	if (aerp) {
6471 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6472 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6473 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6474 		} else {
6475 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6476 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6477 		}
6478 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6479 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6480 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6481 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6482 			printf(" severity 0x%08x first %d\n",
6483 			    r, rs & 0x1f);
6484 		} else
6485 			printf("\n");
6486 	}
6487 
6488 	/* As kind of recovery just report and clear the error statuses. */
6489 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6490 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6491 		if (r != 0) {
6492 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6493 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6494 		}
6495 
6496 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6497 		if (r != 0) {
6498 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6499 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6500 		}
6501 	}
6502 	if (dinfo->cfg.pcie.pcie_location != 0) {
6503 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6504 		    PCIER_DEVICE_STA, 2);
6505 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6506 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6507 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6508 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6509 			    PCIER_DEVICE_STA, rs, 2);
6510 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6511 		}
6512 	}
6513 }
6514 
6515 /*
6516  * Perform a Function Level Reset (FLR) on a device.
6517  *
6518  * This function first waits for any pending transactions to complete
6519  * within the timeout specified by max_delay.  If transactions are
6520  * still pending, the function will return false without attempting a
6521  * reset.
6522  *
6523  * If dev is not a PCI-express function or does not support FLR, this
6524  * function returns false.
6525  *
6526  * Note that no registers are saved or restored.  The caller is
6527  * responsible for saving and restoring any registers including
6528  * PCI-standard registers via pci_save_state() and
6529  * pci_restore_state().
6530  */
6531 bool
6532 pcie_flr(device_t dev, u_int max_delay, bool force)
6533 {
6534 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6535 	uint16_t cmd, ctl;
6536 	int compl_delay;
6537 	int cap;
6538 
6539 	cap = dinfo->cfg.pcie.pcie_location;
6540 	if (cap == 0)
6541 		return (false);
6542 
6543 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6544 		return (false);
6545 
6546 	/*
6547 	 * Disable busmastering to prevent generation of new
6548 	 * transactions while waiting for the device to go idle.  If
6549 	 * the idle timeout fails, the command register is restored
6550 	 * which will re-enable busmastering.
6551 	 */
6552 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6553 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6554 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6555 		if (!force) {
6556 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6557 			return (false);
6558 		}
6559 		pci_printf(&dinfo->cfg,
6560 		    "Resetting with transactions pending after %d ms\n",
6561 		    max_delay);
6562 
6563 		/*
6564 		 * Extend the post-FLR delay to cover the maximum
6565 		 * Completion Timeout delay of anything in flight
6566 		 * during the FLR delay.  Enforce a minimum delay of
6567 		 * at least 10ms.
6568 		 */
6569 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6570 		if (compl_delay < 10)
6571 			compl_delay = 10;
6572 	} else
6573 		compl_delay = 0;
6574 
6575 	/* Initiate the reset. */
6576 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6577 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6578 	    PCIEM_CTL_INITIATE_FLR, 2);
6579 
6580 	/* Wait for 100ms. */
6581 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6582 
6583 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6584 	    PCIEM_STA_TRANSACTION_PND)
6585 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6586 	return (true);
6587 }
6588 
6589 /*
6590  * Attempt a power-management reset by cycling the device in/out of D3
6591  * state.  PCI spec says we can only go into D3 state from D0 state.
6592  * Transition from D[12] into D0 before going to D3 state.
6593  */
6594 int
6595 pci_power_reset(device_t dev)
6596 {
6597 	int ps;
6598 
6599 	ps = pci_get_powerstate(dev);
6600 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6601 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6602 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6603 	pci_set_powerstate(dev, ps);
6604 	return (0);
6605 }
6606 
6607 /*
6608  * Try link drop and retrain of the downstream port of upstream
6609  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6610  * cause Conventional Hot reset of the device in the slot.
6611  * Alternative, for PCIe, could be the secondary bus reset initiatied
6612  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6613  */
6614 int
6615 pcie_link_reset(device_t port, int pcie_location)
6616 {
6617 	uint16_t v;
6618 
6619 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6620 	v |= PCIEM_LINK_CTL_LINK_DIS;
6621 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6622 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6623 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6624 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6625 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6626 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6627 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6628 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6629 }
6630 
6631 static int
6632 pci_reset_post(device_t dev, device_t child)
6633 {
6634 
6635 	if (dev == device_get_parent(child))
6636 		pci_restore_state(child);
6637 	return (0);
6638 }
6639 
6640 static int
6641 pci_reset_prepare(device_t dev, device_t child)
6642 {
6643 
6644 	if (dev == device_get_parent(child))
6645 		pci_save_state(child);
6646 	return (0);
6647 }
6648 
6649 static int
6650 pci_reset_child(device_t dev, device_t child, int flags)
6651 {
6652 	int error;
6653 
6654 	if (dev == NULL || device_get_parent(child) != dev)
6655 		return (0);
6656 	if ((flags & DEVF_RESET_DETACH) != 0) {
6657 		error = device_get_state(child) == DS_ATTACHED ?
6658 		    device_detach(child) : 0;
6659 	} else {
6660 		error = BUS_SUSPEND_CHILD(dev, child);
6661 	}
6662 	if (error == 0) {
6663 		if (!pcie_flr(child, 1000, false)) {
6664 			error = BUS_RESET_PREPARE(dev, child);
6665 			if (error == 0)
6666 				pci_power_reset(child);
6667 			BUS_RESET_POST(dev, child);
6668 		}
6669 		if ((flags & DEVF_RESET_DETACH) != 0)
6670 			device_probe_and_attach(child);
6671 		else
6672 			BUS_RESUME_CHILD(dev, child);
6673 	}
6674 	return (error);
6675 }
6676 
6677 const struct pci_device_table *
6678 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6679 {
6680 	bool match;
6681 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6682 
6683 	vendor = pci_get_vendor(child);
6684 	device = pci_get_device(child);
6685 	subvendor = pci_get_subvendor(child);
6686 	subdevice = pci_get_subdevice(child);
6687 	class = pci_get_class(child);
6688 	subclass = pci_get_subclass(child);
6689 	revid = pci_get_revid(child);
6690 	while (nelt-- > 0) {
6691 		match = true;
6692 		if (id->match_flag_vendor)
6693 			match &= vendor == id->vendor;
6694 		if (id->match_flag_device)
6695 			match &= device == id->device;
6696 		if (id->match_flag_subvendor)
6697 			match &= subvendor == id->subvendor;
6698 		if (id->match_flag_subdevice)
6699 			match &= subdevice == id->subdevice;
6700 		if (id->match_flag_class)
6701 			match &= class == id->class_id;
6702 		if (id->match_flag_subclass)
6703 			match &= subclass == id->subclass;
6704 		if (id->match_flag_revid)
6705 			match &= revid == id->revid;
6706 		if (match)
6707 			return (id);
6708 		id++;
6709 	}
6710 	return (NULL);
6711 }
6712 
6713 static void
6714 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6715 {
6716 	const char *dev_name;
6717 	device_t dev;
6718 
6719 	dev = dinfo->cfg.dev;
6720 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6721 	    dinfo->cfg.slot, dinfo->cfg.func);
6722 	dev_name = device_get_name(dev);
6723 	if (dev_name != NULL)
6724 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6725 }
6726 
6727 void
6728 pci_print_faulted_dev(void)
6729 {
6730 	struct pci_devinfo *dinfo;
6731 	device_t dev;
6732 	int aer, i;
6733 	uint32_t r1, r2;
6734 	uint16_t status;
6735 
6736 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6737 		dev = dinfo->cfg.dev;
6738 		status = pci_read_config(dev, PCIR_STATUS, 2);
6739 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6740 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6741 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6742 		if (status != 0) {
6743 			pci_print_faulted_dev_name(dinfo);
6744 			printf(" error 0x%04x\n", status);
6745 		}
6746 		if (dinfo->cfg.pcie.pcie_location != 0) {
6747 			status = pci_read_config(dev,
6748 			    dinfo->cfg.pcie.pcie_location +
6749 			    PCIER_DEVICE_STA, 2);
6750 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6751 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6752 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6753 				pci_print_faulted_dev_name(dinfo);
6754 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6755 				    pci_read_config(dev,
6756 				    dinfo->cfg.pcie.pcie_location +
6757 				    PCIER_DEVICE_CTL, 2),
6758 				    status);
6759 			}
6760 		}
6761 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6762 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6763 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6764 			if (r1 != 0 || r2 != 0) {
6765 				pci_print_faulted_dev_name(dinfo);
6766 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6767 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6768 				    r1, pci_read_config(dev, aer +
6769 				    PCIR_AER_UC_MASK, 4),
6770 				    pci_read_config(dev, aer +
6771 				    PCIR_AER_UC_SEVERITY, 4),
6772 				    r2, pci_read_config(dev, aer +
6773 				    PCIR_AER_COR_MASK, 4),
6774 				    pci_read_config(dev, aer +
6775 				    PCIR_AER_CAP_CONTROL, 4));
6776 				for (i = 0; i < 4; i++) {
6777 					r1 = pci_read_config(dev, aer +
6778 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6779 					printf("    HL%d: 0x%08x\n", i, r1);
6780 				}
6781 			}
6782 		}
6783 	}
6784 }
6785 
6786 #ifdef DDB
6787 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6788 {
6789 
6790 	pci_print_faulted_dev();
6791 }
6792 
6793 static void
6794 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6795 {
6796 	device_t dev;
6797 	int aer;
6798 	uint32_t r;
6799 
6800 	dev = dinfo->cfg.dev;
6801 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6802 	    PCIER_DEVICE_STA, 2);
6803 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6804 	    PCIER_DEVICE_STA, r, 2);
6805 
6806 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6807 		return;
6808 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6809 	if (r != 0)
6810 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6811 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6812 	if (r != 0)
6813 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6814 }
6815 
6816 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
6817 {
6818 	struct pci_devinfo *dinfo;
6819 	device_t dev;
6820 	uint16_t status, status1;
6821 
6822 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6823 		dev = dinfo->cfg.dev;
6824 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6825 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6826 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6827 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6828 		if (status1 != 0) {
6829 			status &= ~status1;
6830 			pci_write_config(dev, PCIR_STATUS, status, 2);
6831 		}
6832 		if (dinfo->cfg.pcie.pcie_location != 0)
6833 			db_clear_pcie_errors(dinfo);
6834 	}
6835 }
6836 #endif
6837