xref: /freebsd/sys/dev/pci/pci.c (revision 369d70ec9282cccf131ab0400b05dad13ba8116d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_iommu.h"
36 #include "opt_bus.h"
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/endian.h>
41 #include <sys/eventhandler.h>
42 #include <sys/fcntl.h>
43 #include <sys/kernel.h>
44 #include <sys/limits.h>
45 #include <sys/linker.h>
46 #include <sys/malloc.h>
47 #include <sys/module.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_extern.h>
57 
58 #include <sys/bus.h>
59 #include <machine/bus.h>
60 #include <sys/rman.h>
61 #include <machine/resource.h>
62 #include <machine/stdarg.h>
63 
64 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
65 #include <machine/intr_machdep.h>
66 #endif
67 
68 #include <sys/pciio.h>
69 #include <dev/pci/pcireg.h>
70 #include <dev/pci/pcivar.h>
71 #include <dev/pci/pci_private.h>
72 
73 #ifdef PCI_IOV
74 #include <sys/nv.h>
75 #include <dev/pci/pci_iov_private.h>
76 #endif
77 
78 #include <dev/usb/controller/xhcireg.h>
79 #include <dev/usb/controller/ehcireg.h>
80 #include <dev/usb/controller/ohcireg.h>
81 #include <dev/usb/controller/uhcireg.h>
82 
83 #include <dev/iommu/iommu.h>
84 
85 #include "pcib_if.h"
86 #include "pci_if.h"
87 
88 #define	PCIR_IS_BIOS(cfg, reg)						\
89 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
90 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
91 
92 static int		pci_has_quirk(uint32_t devid, int quirk);
93 static pci_addr_t	pci_mapbase(uint64_t mapreg);
94 static const char	*pci_maptype(uint64_t mapreg);
95 static int		pci_maprange(uint64_t mapreg);
96 static pci_addr_t	pci_rombase(uint64_t mapreg);
97 static int		pci_romsize(uint64_t testval);
98 static void		pci_fixancient(pcicfgregs *cfg);
99 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
100 
101 static int		pci_porten(device_t dev);
102 static int		pci_memen(device_t dev);
103 static void		pci_assign_interrupt(device_t bus, device_t dev,
104 			    int force_route);
105 static int		pci_add_map(device_t bus, device_t dev, int reg,
106 			    struct resource_list *rl, int force, int prefetch);
107 static int		pci_probe(device_t dev);
108 static void		pci_load_vendor_data(void);
109 static int		pci_describe_parse_line(char **ptr, int *vendor,
110 			    int *device, char **desc);
111 static char		*pci_describe_device(device_t dev);
112 static int		pci_modevent(module_t mod, int what, void *arg);
113 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
114 			    pcicfgregs *cfg);
115 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
116 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
117 			    int reg, uint32_t *data);
118 #if 0
119 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
120 			    int reg, uint32_t data);
121 #endif
122 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
123 static void		pci_mask_msix(device_t dev, u_int index);
124 static void		pci_unmask_msix(device_t dev, u_int index);
125 static int		pci_msi_blacklisted(void);
126 static int		pci_msix_blacklisted(void);
127 static void		pci_resume_msi(device_t dev);
128 static void		pci_resume_msix(device_t dev);
129 static int		pci_remap_intr_method(device_t bus, device_t dev,
130 			    u_int irq);
131 static void		pci_hint_device_unit(device_t acdev, device_t child,
132 			    const char *name, int *unitp);
133 static int		pci_reset_post(device_t dev, device_t child);
134 static int		pci_reset_prepare(device_t dev, device_t child);
135 static int		pci_reset_child(device_t dev, device_t child,
136 			    int flags);
137 
138 static int		pci_get_id_method(device_t dev, device_t child,
139 			    enum pci_id_type type, uintptr_t *rid);
140 
141 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
142     int b, int s, int f, uint16_t vid, uint16_t did);
143 
144 static device_method_t pci_methods[] = {
145 	/* Device interface */
146 	DEVMETHOD(device_probe,		pci_probe),
147 	DEVMETHOD(device_attach,	pci_attach),
148 	DEVMETHOD(device_detach,	pci_detach),
149 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
150 	DEVMETHOD(device_suspend,	bus_generic_suspend),
151 	DEVMETHOD(device_resume,	pci_resume),
152 
153 	/* Bus interface */
154 	DEVMETHOD(bus_print_child,	pci_print_child),
155 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
156 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
157 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
158 	DEVMETHOD(bus_driver_added,	pci_driver_added),
159 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
160 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
161 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
162 	DEVMETHOD(bus_reset_post,	pci_reset_post),
163 	DEVMETHOD(bus_reset_child,	pci_reset_child),
164 
165 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
166 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
167 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
168 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
169 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
170 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
171 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
172 	DEVMETHOD(bus_release_resource,	pci_release_resource),
173 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
174 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
175 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
176 	DEVMETHOD(bus_child_detached,	pci_child_detached),
177 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
178 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
179 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
180 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
181 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
182 	DEVMETHOD(bus_resume_child,	pci_resume_child),
183 	DEVMETHOD(bus_rescan,		pci_rescan_method),
184 
185 	/* PCI interface */
186 	DEVMETHOD(pci_read_config,	pci_read_config_method),
187 	DEVMETHOD(pci_write_config,	pci_write_config_method),
188 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
189 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
190 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
191 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
192 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
193 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
194 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
195 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
196 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
197 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
198 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
199 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
200 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
201 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
202 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
203 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
204 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
205 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
206 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
207 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
208 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
209 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
210 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
211 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
212 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
213 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
214 	DEVMETHOD(pci_get_id,		pci_get_id_method),
215 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
216 	DEVMETHOD(pci_child_added,	pci_child_added_method),
217 #ifdef PCI_IOV
218 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
219 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
220 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
221 #endif
222 
223 	DEVMETHOD_END
224 };
225 
226 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
227 
228 static devclass_t pci_devclass;
229 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
230     BUS_PASS_BUS);
231 MODULE_VERSION(pci, 1);
232 
233 static char	*pci_vendordata;
234 static size_t	pci_vendordata_size;
235 
236 struct pci_quirk {
237 	uint32_t devid;	/* Vendor/device of the card */
238 	int	type;
239 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
240 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
241 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
242 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
243 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
244 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
245 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
246 	int	arg1;
247 	int	arg2;
248 };
249 
250 static const struct pci_quirk pci_quirks[] = {
251 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
252 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
254 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
255 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
256 
257 	/*
258 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
259 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
260 	 */
261 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
262 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
263 
264 	/*
265 	 * MSI doesn't work on earlier Intel chipsets including
266 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
267 	 */
268 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
273 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
274 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
275 
276 	/*
277 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
278 	 * bridge.
279 	 */
280 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
281 
282 	/*
283 	 * Some virtualization environments emulate an older chipset
284 	 * but support MSI just fine.  QEMU uses the Intel 82440.
285 	 */
286 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
287 
288 	/*
289 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
290 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
291 	 * It prevents us from attaching hpet(4) when the bit is unset.
292 	 * Note this quirk only affects SB600 revision A13 and earlier.
293 	 * For SB600 A21 and later, firmware must set the bit to hide it.
294 	 * For SB700 and later, it is unused and hardcoded to zero.
295 	 */
296 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
297 
298 	/*
299 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
300 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
301 	 * of the command register is set.
302 	 */
303 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
306 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
307 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 
309 	/*
310 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
311 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
312 	 */
313 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
314 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
315 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
316 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
317 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
318 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
319 
320 	/*
321 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
322 	 * expected place.
323 	 */
324 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
325 	{ 0 }
326 };
327 
328 /* map register information */
329 #define	PCI_MAPMEM	0x01	/* memory map */
330 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
331 #define	PCI_MAPPORT	0x04	/* port map */
332 
333 struct devlist pci_devq;
334 uint32_t pci_generation;
335 uint32_t pci_numdevs = 0;
336 static int pcie_chipset, pcix_chipset;
337 
338 /* sysctl vars */
339 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
340     "PCI bus tuning parameters");
341 
342 static int pci_enable_io_modes = 1;
343 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
344     &pci_enable_io_modes, 1,
345     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
346     " enable these bits correctly.  We'd like to do this all the time, but"
347     " there are some peripherals that this causes problems with.");
348 
349 static int pci_do_realloc_bars = 1;
350 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
351     &pci_do_realloc_bars, 0,
352     "Attempt to allocate a new range for any BARs whose original "
353     "firmware-assigned ranges fail to allocate during the initial device scan.");
354 
355 static int pci_do_power_nodriver = 0;
356 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
357     &pci_do_power_nodriver, 0,
358     "Place a function into D3 state when no driver attaches to it.  0 means"
359     " disable.  1 means conservatively place devices into D3 state.  2 means"
360     " aggressively place devices into D3 state.  3 means put absolutely"
361     " everything in D3 state.");
362 
363 int pci_do_power_resume = 1;
364 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
365     &pci_do_power_resume, 1,
366   "Transition from D3 -> D0 on resume.");
367 
368 int pci_do_power_suspend = 1;
369 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
370     &pci_do_power_suspend, 1,
371   "Transition from D0 -> D3 on suspend.");
372 
373 static int pci_do_msi = 1;
374 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
375     "Enable support for MSI interrupts");
376 
377 static int pci_do_msix = 1;
378 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
379     "Enable support for MSI-X interrupts");
380 
381 static int pci_msix_rewrite_table = 0;
382 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
383     &pci_msix_rewrite_table, 0,
384     "Rewrite entire MSI-X table when updating MSI-X entries");
385 
386 static int pci_honor_msi_blacklist = 1;
387 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
388     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
389 
390 #if defined(__i386__) || defined(__amd64__)
391 static int pci_usb_takeover = 1;
392 #else
393 static int pci_usb_takeover = 0;
394 #endif
395 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
396     &pci_usb_takeover, 1,
397     "Enable early takeover of USB controllers. Disable this if you depend on"
398     " BIOS emulation of USB devices, that is you use USB devices (like"
399     " keyboard or mouse) but do not load USB drivers");
400 
401 static int pci_clear_bars;
402 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
403     "Ignore firmware-assigned resources for BARs.");
404 
405 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
406 static int pci_clear_buses;
407 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
408     "Ignore firmware-assigned bus numbers.");
409 #endif
410 
411 static int pci_enable_ari = 1;
412 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
413     0, "Enable support for PCIe Alternative RID Interpretation");
414 
415 int pci_enable_aspm = 1;
416 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
417     0, "Enable support for PCIe Active State Power Management");
418 
419 static int pci_clear_aer_on_attach = 0;
420 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
421     &pci_clear_aer_on_attach, 0,
422     "Clear port and device AER state on driver attach");
423 
424 static int
425 pci_has_quirk(uint32_t devid, int quirk)
426 {
427 	const struct pci_quirk *q;
428 
429 	for (q = &pci_quirks[0]; q->devid; q++) {
430 		if (q->devid == devid && q->type == quirk)
431 			return (1);
432 	}
433 	return (0);
434 }
435 
436 /* Find a device_t by bus/slot/function in domain 0 */
437 
438 device_t
439 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
440 {
441 
442 	return (pci_find_dbsf(0, bus, slot, func));
443 }
444 
445 /* Find a device_t by domain/bus/slot/function */
446 
447 device_t
448 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
449 {
450 	struct pci_devinfo *dinfo = NULL;
451 
452 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
453 		if ((dinfo->cfg.domain == domain) &&
454 		    (dinfo->cfg.bus == bus) &&
455 		    (dinfo->cfg.slot == slot) &&
456 		    (dinfo->cfg.func == func)) {
457 			break;
458 		}
459 	}
460 
461 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
462 }
463 
464 /* Find a device_t by vendor/device ID */
465 
466 device_t
467 pci_find_device(uint16_t vendor, uint16_t device)
468 {
469 	struct pci_devinfo *dinfo;
470 
471 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
472 		if ((dinfo->cfg.vendor == vendor) &&
473 		    (dinfo->cfg.device == device)) {
474 			return (dinfo->cfg.dev);
475 		}
476 	}
477 
478 	return (NULL);
479 }
480 
481 device_t
482 pci_find_class(uint8_t class, uint8_t subclass)
483 {
484 	struct pci_devinfo *dinfo;
485 
486 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
487 		if (dinfo->cfg.baseclass == class &&
488 		    dinfo->cfg.subclass == subclass) {
489 			return (dinfo->cfg.dev);
490 		}
491 	}
492 
493 	return (NULL);
494 }
495 
496 static int
497 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
498 {
499 	va_list ap;
500 	int retval;
501 
502 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
503 	    cfg->func);
504 	va_start(ap, fmt);
505 	retval += vprintf(fmt, ap);
506 	va_end(ap);
507 	return (retval);
508 }
509 
510 /* return base address of memory or port map */
511 
512 static pci_addr_t
513 pci_mapbase(uint64_t mapreg)
514 {
515 
516 	if (PCI_BAR_MEM(mapreg))
517 		return (mapreg & PCIM_BAR_MEM_BASE);
518 	else
519 		return (mapreg & PCIM_BAR_IO_BASE);
520 }
521 
522 /* return map type of memory or port map */
523 
524 static const char *
525 pci_maptype(uint64_t mapreg)
526 {
527 
528 	if (PCI_BAR_IO(mapreg))
529 		return ("I/O Port");
530 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
531 		return ("Prefetchable Memory");
532 	return ("Memory");
533 }
534 
535 /* return log2 of map size decoded for memory or port map */
536 
537 int
538 pci_mapsize(uint64_t testval)
539 {
540 	int ln2size;
541 
542 	testval = pci_mapbase(testval);
543 	ln2size = 0;
544 	if (testval != 0) {
545 		while ((testval & 1) == 0)
546 		{
547 			ln2size++;
548 			testval >>= 1;
549 		}
550 	}
551 	return (ln2size);
552 }
553 
554 /* return base address of device ROM */
555 
556 static pci_addr_t
557 pci_rombase(uint64_t mapreg)
558 {
559 
560 	return (mapreg & PCIM_BIOS_ADDR_MASK);
561 }
562 
563 /* return log2 of map size decided for device ROM */
564 
565 static int
566 pci_romsize(uint64_t testval)
567 {
568 	int ln2size;
569 
570 	testval = pci_rombase(testval);
571 	ln2size = 0;
572 	if (testval != 0) {
573 		while ((testval & 1) == 0)
574 		{
575 			ln2size++;
576 			testval >>= 1;
577 		}
578 	}
579 	return (ln2size);
580 }
581 
582 /* return log2 of address range supported by map register */
583 
584 static int
585 pci_maprange(uint64_t mapreg)
586 {
587 	int ln2range = 0;
588 
589 	if (PCI_BAR_IO(mapreg))
590 		ln2range = 32;
591 	else
592 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
593 		case PCIM_BAR_MEM_32:
594 			ln2range = 32;
595 			break;
596 		case PCIM_BAR_MEM_1MB:
597 			ln2range = 20;
598 			break;
599 		case PCIM_BAR_MEM_64:
600 			ln2range = 64;
601 			break;
602 		}
603 	return (ln2range);
604 }
605 
606 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
607 
608 static void
609 pci_fixancient(pcicfgregs *cfg)
610 {
611 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
612 		return;
613 
614 	/* PCI to PCI bridges use header type 1 */
615 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
616 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
617 }
618 
619 /* extract header type specific config data */
620 
621 static void
622 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
623 {
624 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
625 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
626 	case PCIM_HDRTYPE_NORMAL:
627 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
628 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
629 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
630 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
631 		cfg->nummaps	    = PCI_MAXMAPS_0;
632 		break;
633 	case PCIM_HDRTYPE_BRIDGE:
634 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
635 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
636 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
637 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
638 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
639 		cfg->nummaps	    = PCI_MAXMAPS_1;
640 		break;
641 	case PCIM_HDRTYPE_CARDBUS:
642 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
643 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
644 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
645 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
646 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
647 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
648 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
649 		cfg->nummaps	    = PCI_MAXMAPS_2;
650 		break;
651 	}
652 #undef REG
653 }
654 
655 /* read configuration header into pcicfgregs structure */
656 struct pci_devinfo *
657 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
658 {
659 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
660 	uint16_t vid, did;
661 
662 	vid = REG(PCIR_VENDOR, 2);
663 	did = REG(PCIR_DEVICE, 2);
664 	if (vid != 0xffff)
665 		return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
666 
667 	return (NULL);
668 }
669 
670 struct pci_devinfo *
671 pci_alloc_devinfo_method(device_t dev)
672 {
673 
674 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
675 	    M_WAITOK | M_ZERO));
676 }
677 
678 static struct pci_devinfo *
679 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
680     uint16_t vid, uint16_t did)
681 {
682 	struct pci_devinfo *devlist_entry;
683 	pcicfgregs *cfg;
684 
685 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
686 
687 	cfg = &devlist_entry->cfg;
688 
689 	cfg->domain		= d;
690 	cfg->bus		= b;
691 	cfg->slot		= s;
692 	cfg->func		= f;
693 	cfg->vendor		= vid;
694 	cfg->device		= did;
695 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
696 	cfg->statreg		= REG(PCIR_STATUS, 2);
697 	cfg->baseclass		= REG(PCIR_CLASS, 1);
698 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
699 	cfg->progif		= REG(PCIR_PROGIF, 1);
700 	cfg->revid		= REG(PCIR_REVID, 1);
701 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
702 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
703 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
704 	cfg->intpin		= REG(PCIR_INTPIN, 1);
705 	cfg->intline		= REG(PCIR_INTLINE, 1);
706 
707 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
708 	cfg->hdrtype		&= ~PCIM_MFDEV;
709 	STAILQ_INIT(&cfg->maps);
710 
711 	cfg->iov		= NULL;
712 
713 	pci_fixancient(cfg);
714 	pci_hdrtypedata(pcib, b, s, f, cfg);
715 
716 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
717 		pci_read_cap(pcib, cfg);
718 
719 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
720 
721 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
722 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
723 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
724 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
725 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
726 
727 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
728 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
729 	devlist_entry->conf.pc_vendor = cfg->vendor;
730 	devlist_entry->conf.pc_device = cfg->device;
731 
732 	devlist_entry->conf.pc_class = cfg->baseclass;
733 	devlist_entry->conf.pc_subclass = cfg->subclass;
734 	devlist_entry->conf.pc_progif = cfg->progif;
735 	devlist_entry->conf.pc_revid = cfg->revid;
736 
737 	pci_numdevs++;
738 	pci_generation++;
739 
740 	return (devlist_entry);
741 }
742 #undef REG
743 
744 static void
745 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
746 {
747 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
748     cfg->ea.ea_location + (n), w)
749 	int num_ent;
750 	int ptr;
751 	int a, b;
752 	uint32_t val;
753 	int ent_size;
754 	uint32_t dw[4];
755 	uint64_t base, max_offset;
756 	struct pci_ea_entry *eae;
757 
758 	if (cfg->ea.ea_location == 0)
759 		return;
760 
761 	STAILQ_INIT(&cfg->ea.ea_entries);
762 
763 	/* Determine the number of entries */
764 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
765 	num_ent &= PCIM_EA_NUM_ENT_MASK;
766 
767 	/* Find the first entry to care of */
768 	ptr = PCIR_EA_FIRST_ENT;
769 
770 	/* Skip DWORD 2 for type 1 functions */
771 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
772 		ptr += 4;
773 
774 	for (a = 0; a < num_ent; a++) {
775 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
776 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
777 
778 		/* Read a number of dwords in the entry */
779 		val = REG(ptr, 4);
780 		ptr += 4;
781 		ent_size = (val & PCIM_EA_ES);
782 
783 		for (b = 0; b < ent_size; b++) {
784 			dw[b] = REG(ptr, 4);
785 			ptr += 4;
786 		}
787 
788 		eae->eae_flags = val;
789 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
790 
791 		base = dw[0] & PCIM_EA_FIELD_MASK;
792 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
793 		b = 2;
794 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
795 			base |= (uint64_t)dw[b] << 32UL;
796 			b++;
797 		}
798 		if (((dw[1] & PCIM_EA_IS_64) != 0)
799 		    && (b < ent_size)) {
800 			max_offset |= (uint64_t)dw[b] << 32UL;
801 			b++;
802 		}
803 
804 		eae->eae_base = base;
805 		eae->eae_max_offset = max_offset;
806 
807 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
808 
809 		if (bootverbose) {
810 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
811 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
812 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
813 		}
814 	}
815 }
816 #undef REG
817 
818 static void
819 pci_read_cap(device_t pcib, pcicfgregs *cfg)
820 {
821 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
822 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
823 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
824 	uint64_t addr;
825 #endif
826 	uint32_t val;
827 	int	ptr, nextptr, ptrptr;
828 
829 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
830 	case PCIM_HDRTYPE_NORMAL:
831 	case PCIM_HDRTYPE_BRIDGE:
832 		ptrptr = PCIR_CAP_PTR;
833 		break;
834 	case PCIM_HDRTYPE_CARDBUS:
835 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
836 		break;
837 	default:
838 		return;		/* no extended capabilities support */
839 	}
840 	nextptr = REG(ptrptr, 1);	/* sanity check? */
841 
842 	/*
843 	 * Read capability entries.
844 	 */
845 	while (nextptr != 0) {
846 		/* Sanity check */
847 		if (nextptr > 255) {
848 			printf("illegal PCI extended capability offset %d\n",
849 			    nextptr);
850 			return;
851 		}
852 		/* Find the next entry */
853 		ptr = nextptr;
854 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
855 
856 		/* Process this entry */
857 		switch (REG(ptr + PCICAP_ID, 1)) {
858 		case PCIY_PMG:		/* PCI power management */
859 			if (cfg->pp.pp_cap == 0) {
860 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
861 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
862 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
863 				if ((nextptr - ptr) > PCIR_POWER_DATA)
864 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
865 			}
866 			break;
867 		case PCIY_HT:		/* HyperTransport */
868 			/* Determine HT-specific capability type. */
869 			val = REG(ptr + PCIR_HT_COMMAND, 2);
870 
871 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
872 				cfg->ht.ht_slave = ptr;
873 
874 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
875 			switch (val & PCIM_HTCMD_CAP_MASK) {
876 			case PCIM_HTCAP_MSI_MAPPING:
877 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
878 					/* Sanity check the mapping window. */
879 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
880 					    4);
881 					addr <<= 32;
882 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
883 					    4);
884 					if (addr != MSI_INTEL_ADDR_BASE)
885 						device_printf(pcib,
886 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
887 						    cfg->domain, cfg->bus,
888 						    cfg->slot, cfg->func,
889 						    (long long)addr);
890 				} else
891 					addr = MSI_INTEL_ADDR_BASE;
892 
893 				cfg->ht.ht_msimap = ptr;
894 				cfg->ht.ht_msictrl = val;
895 				cfg->ht.ht_msiaddr = addr;
896 				break;
897 			}
898 #endif
899 			break;
900 		case PCIY_MSI:		/* PCI MSI */
901 			cfg->msi.msi_location = ptr;
902 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
903 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
904 						     PCIM_MSICTRL_MMC_MASK)>>1);
905 			break;
906 		case PCIY_MSIX:		/* PCI MSI-X */
907 			cfg->msix.msix_location = ptr;
908 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
909 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
910 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
911 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
912 			cfg->msix.msix_table_bar = PCIR_BAR(val &
913 			    PCIM_MSIX_BIR_MASK);
914 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
915 			val = REG(ptr + PCIR_MSIX_PBA, 4);
916 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
917 			    PCIM_MSIX_BIR_MASK);
918 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
919 			break;
920 		case PCIY_VPD:		/* PCI Vital Product Data */
921 			cfg->vpd.vpd_reg = ptr;
922 			break;
923 		case PCIY_SUBVENDOR:
924 			/* Should always be true. */
925 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
926 			    PCIM_HDRTYPE_BRIDGE) {
927 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
928 				cfg->subvendor = val & 0xffff;
929 				cfg->subdevice = val >> 16;
930 			}
931 			break;
932 		case PCIY_PCIX:		/* PCI-X */
933 			/*
934 			 * Assume we have a PCI-X chipset if we have
935 			 * at least one PCI-PCI bridge with a PCI-X
936 			 * capability.  Note that some systems with
937 			 * PCI-express or HT chipsets might match on
938 			 * this check as well.
939 			 */
940 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
941 			    PCIM_HDRTYPE_BRIDGE)
942 				pcix_chipset = 1;
943 			cfg->pcix.pcix_location = ptr;
944 			break;
945 		case PCIY_EXPRESS:	/* PCI-express */
946 			/*
947 			 * Assume we have a PCI-express chipset if we have
948 			 * at least one PCI-express device.
949 			 */
950 			pcie_chipset = 1;
951 			cfg->pcie.pcie_location = ptr;
952 			val = REG(ptr + PCIER_FLAGS, 2);
953 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
954 			break;
955 		case PCIY_EA:		/* Enhanced Allocation */
956 			cfg->ea.ea_location = ptr;
957 			pci_ea_fill_info(pcib, cfg);
958 			break;
959 		default:
960 			break;
961 		}
962 	}
963 
964 #if defined(__powerpc__)
965 	/*
966 	 * Enable the MSI mapping window for all HyperTransport
967 	 * slaves.  PCI-PCI bridges have their windows enabled via
968 	 * PCIB_MAP_MSI().
969 	 */
970 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
971 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
972 		device_printf(pcib,
973 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
974 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
975 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
976 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
977 		     2);
978 	}
979 #endif
980 /* REG and WREG use carry through to next functions */
981 }
982 
983 /*
984  * PCI Vital Product Data
985  */
986 
987 #define	PCI_VPD_TIMEOUT		1000000
988 
989 static int
990 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
991 {
992 	int count = PCI_VPD_TIMEOUT;
993 
994 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
995 
996 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
997 
998 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
999 		if (--count < 0)
1000 			return (ENXIO);
1001 		DELAY(1);	/* limit looping */
1002 	}
1003 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1004 
1005 	return (0);
1006 }
1007 
1008 #if 0
1009 static int
1010 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1011 {
1012 	int count = PCI_VPD_TIMEOUT;
1013 
1014 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1015 
1016 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1017 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1018 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1019 		if (--count < 0)
1020 			return (ENXIO);
1021 		DELAY(1);	/* limit looping */
1022 	}
1023 
1024 	return (0);
1025 }
1026 #endif
1027 
1028 #undef PCI_VPD_TIMEOUT
1029 
1030 struct vpd_readstate {
1031 	device_t	pcib;
1032 	pcicfgregs	*cfg;
1033 	uint32_t	val;
1034 	int		bytesinval;
1035 	int		off;
1036 	uint8_t		cksum;
1037 };
1038 
1039 static int
1040 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1041 {
1042 	uint32_t reg;
1043 	uint8_t byte;
1044 
1045 	if (vrs->bytesinval == 0) {
1046 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1047 			return (ENXIO);
1048 		vrs->val = le32toh(reg);
1049 		vrs->off += 4;
1050 		byte = vrs->val & 0xff;
1051 		vrs->bytesinval = 3;
1052 	} else {
1053 		vrs->val = vrs->val >> 8;
1054 		byte = vrs->val & 0xff;
1055 		vrs->bytesinval--;
1056 	}
1057 
1058 	vrs->cksum += byte;
1059 	*data = byte;
1060 	return (0);
1061 }
1062 
1063 static void
1064 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1065 {
1066 	struct vpd_readstate vrs;
1067 	int state;
1068 	int name;
1069 	int remain;
1070 	int i;
1071 	int alloc, off;		/* alloc/off for RO/W arrays */
1072 	int cksumvalid;
1073 	int dflen;
1074 	uint8_t byte;
1075 	uint8_t byte2;
1076 
1077 	/* init vpd reader */
1078 	vrs.bytesinval = 0;
1079 	vrs.off = 0;
1080 	vrs.pcib = pcib;
1081 	vrs.cfg = cfg;
1082 	vrs.cksum = 0;
1083 
1084 	state = 0;
1085 	name = remain = i = 0;	/* shut up stupid gcc */
1086 	alloc = off = 0;	/* shut up stupid gcc */
1087 	dflen = 0;		/* shut up stupid gcc */
1088 	cksumvalid = -1;
1089 	while (state >= 0) {
1090 		if (vpd_nextbyte(&vrs, &byte)) {
1091 			state = -2;
1092 			break;
1093 		}
1094 #if 0
1095 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1096 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1097 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1098 #endif
1099 		switch (state) {
1100 		case 0:		/* item name */
1101 			if (byte & 0x80) {
1102 				if (vpd_nextbyte(&vrs, &byte2)) {
1103 					state = -2;
1104 					break;
1105 				}
1106 				remain = byte2;
1107 				if (vpd_nextbyte(&vrs, &byte2)) {
1108 					state = -2;
1109 					break;
1110 				}
1111 				remain |= byte2 << 8;
1112 				name = byte & 0x7f;
1113 			} else {
1114 				remain = byte & 0x7;
1115 				name = (byte >> 3) & 0xf;
1116 			}
1117 			if (vrs.off + remain - vrs.bytesinval > 0x8000) {
1118 				pci_printf(cfg,
1119 				    "VPD data overflow, remain %#x\n", remain);
1120 				state = -1;
1121 				break;
1122 			}
1123 			switch (name) {
1124 			case 0x2:	/* String */
1125 				cfg->vpd.vpd_ident = malloc(remain + 1,
1126 				    M_DEVBUF, M_WAITOK);
1127 				i = 0;
1128 				state = 1;
1129 				break;
1130 			case 0xf:	/* End */
1131 				state = -1;
1132 				break;
1133 			case 0x10:	/* VPD-R */
1134 				alloc = 8;
1135 				off = 0;
1136 				cfg->vpd.vpd_ros = malloc(alloc *
1137 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1138 				    M_WAITOK | M_ZERO);
1139 				state = 2;
1140 				break;
1141 			case 0x11:	/* VPD-W */
1142 				alloc = 8;
1143 				off = 0;
1144 				cfg->vpd.vpd_w = malloc(alloc *
1145 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1146 				    M_WAITOK | M_ZERO);
1147 				state = 5;
1148 				break;
1149 			default:	/* Invalid data, abort */
1150 				state = -1;
1151 				break;
1152 			}
1153 			break;
1154 
1155 		case 1:	/* Identifier String */
1156 			cfg->vpd.vpd_ident[i++] = byte;
1157 			remain--;
1158 			if (remain == 0)  {
1159 				cfg->vpd.vpd_ident[i] = '\0';
1160 				state = 0;
1161 			}
1162 			break;
1163 
1164 		case 2:	/* VPD-R Keyword Header */
1165 			if (off == alloc) {
1166 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1167 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1168 				    M_DEVBUF, M_WAITOK | M_ZERO);
1169 			}
1170 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1171 			if (vpd_nextbyte(&vrs, &byte2)) {
1172 				state = -2;
1173 				break;
1174 			}
1175 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1176 			if (vpd_nextbyte(&vrs, &byte2)) {
1177 				state = -2;
1178 				break;
1179 			}
1180 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1181 			if (dflen == 0 &&
1182 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1183 			    2) == 0) {
1184 				/*
1185 				 * if this happens, we can't trust the rest
1186 				 * of the VPD.
1187 				 */
1188 				pci_printf(cfg, "bad keyword length: %d\n",
1189 				    dflen);
1190 				cksumvalid = 0;
1191 				state = -1;
1192 				break;
1193 			} else if (dflen == 0) {
1194 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1195 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1196 				    M_DEVBUF, M_WAITOK);
1197 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1198 			} else
1199 				cfg->vpd.vpd_ros[off].value = malloc(
1200 				    (dflen + 1) *
1201 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1202 				    M_DEVBUF, M_WAITOK);
1203 			remain -= 3;
1204 			i = 0;
1205 			/* keep in sync w/ state 3's transistions */
1206 			if (dflen == 0 && remain == 0)
1207 				state = 0;
1208 			else if (dflen == 0)
1209 				state = 2;
1210 			else
1211 				state = 3;
1212 			break;
1213 
1214 		case 3:	/* VPD-R Keyword Value */
1215 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1216 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1217 			    "RV", 2) == 0 && cksumvalid == -1) {
1218 				if (vrs.cksum == 0)
1219 					cksumvalid = 1;
1220 				else {
1221 					if (bootverbose)
1222 						pci_printf(cfg,
1223 					    "bad VPD cksum, remain %hhu\n",
1224 						    vrs.cksum);
1225 					cksumvalid = 0;
1226 					state = -1;
1227 					break;
1228 				}
1229 			}
1230 			dflen--;
1231 			remain--;
1232 			/* keep in sync w/ state 2's transistions */
1233 			if (dflen == 0)
1234 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1235 			if (dflen == 0 && remain == 0) {
1236 				cfg->vpd.vpd_rocnt = off;
1237 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1238 				    off * sizeof(*cfg->vpd.vpd_ros),
1239 				    M_DEVBUF, M_WAITOK | M_ZERO);
1240 				state = 0;
1241 			} else if (dflen == 0)
1242 				state = 2;
1243 			break;
1244 
1245 		case 4:
1246 			remain--;
1247 			if (remain == 0)
1248 				state = 0;
1249 			break;
1250 
1251 		case 5:	/* VPD-W Keyword Header */
1252 			if (off == alloc) {
1253 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1254 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1255 				    M_DEVBUF, M_WAITOK | M_ZERO);
1256 			}
1257 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1258 			if (vpd_nextbyte(&vrs, &byte2)) {
1259 				state = -2;
1260 				break;
1261 			}
1262 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1263 			if (vpd_nextbyte(&vrs, &byte2)) {
1264 				state = -2;
1265 				break;
1266 			}
1267 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1268 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1269 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1270 			    sizeof(*cfg->vpd.vpd_w[off].value),
1271 			    M_DEVBUF, M_WAITOK);
1272 			remain -= 3;
1273 			i = 0;
1274 			/* keep in sync w/ state 6's transistions */
1275 			if (dflen == 0 && remain == 0)
1276 				state = 0;
1277 			else if (dflen == 0)
1278 				state = 5;
1279 			else
1280 				state = 6;
1281 			break;
1282 
1283 		case 6:	/* VPD-W Keyword Value */
1284 			cfg->vpd.vpd_w[off].value[i++] = byte;
1285 			dflen--;
1286 			remain--;
1287 			/* keep in sync w/ state 5's transistions */
1288 			if (dflen == 0)
1289 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1290 			if (dflen == 0 && remain == 0) {
1291 				cfg->vpd.vpd_wcnt = off;
1292 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1293 				    off * sizeof(*cfg->vpd.vpd_w),
1294 				    M_DEVBUF, M_WAITOK | M_ZERO);
1295 				state = 0;
1296 			} else if (dflen == 0)
1297 				state = 5;
1298 			break;
1299 
1300 		default:
1301 			pci_printf(cfg, "invalid state: %d\n", state);
1302 			state = -1;
1303 			break;
1304 		}
1305 	}
1306 
1307 	if (cksumvalid == 0 || state < -1) {
1308 		/* read-only data bad, clean up */
1309 		if (cfg->vpd.vpd_ros != NULL) {
1310 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1311 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1312 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1313 			cfg->vpd.vpd_ros = NULL;
1314 		}
1315 	}
1316 	if (state < -1) {
1317 		/* I/O error, clean up */
1318 		pci_printf(cfg, "failed to read VPD data.\n");
1319 		if (cfg->vpd.vpd_ident != NULL) {
1320 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1321 			cfg->vpd.vpd_ident = NULL;
1322 		}
1323 		if (cfg->vpd.vpd_w != NULL) {
1324 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1325 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1326 			free(cfg->vpd.vpd_w, M_DEVBUF);
1327 			cfg->vpd.vpd_w = NULL;
1328 		}
1329 	}
1330 	cfg->vpd.vpd_cached = 1;
1331 #undef REG
1332 #undef WREG
1333 }
1334 
1335 int
1336 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1337 {
1338 	struct pci_devinfo *dinfo = device_get_ivars(child);
1339 	pcicfgregs *cfg = &dinfo->cfg;
1340 
1341 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1342 		pci_read_vpd(device_get_parent(dev), cfg);
1343 
1344 	*identptr = cfg->vpd.vpd_ident;
1345 
1346 	if (*identptr == NULL)
1347 		return (ENXIO);
1348 
1349 	return (0);
1350 }
1351 
1352 int
1353 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1354 	const char **vptr)
1355 {
1356 	struct pci_devinfo *dinfo = device_get_ivars(child);
1357 	pcicfgregs *cfg = &dinfo->cfg;
1358 	int i;
1359 
1360 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1361 		pci_read_vpd(device_get_parent(dev), cfg);
1362 
1363 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1364 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1365 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1366 			*vptr = cfg->vpd.vpd_ros[i].value;
1367 			return (0);
1368 		}
1369 
1370 	*vptr = NULL;
1371 	return (ENXIO);
1372 }
1373 
1374 struct pcicfg_vpd *
1375 pci_fetch_vpd_list(device_t dev)
1376 {
1377 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1378 	pcicfgregs *cfg = &dinfo->cfg;
1379 
1380 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1381 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1382 	return (&cfg->vpd);
1383 }
1384 
1385 /*
1386  * Find the requested HyperTransport capability and return the offset
1387  * in configuration space via the pointer provided.  The function
1388  * returns 0 on success and an error code otherwise.
1389  */
1390 int
1391 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1392 {
1393 	int ptr, error;
1394 	uint16_t val;
1395 
1396 	error = pci_find_cap(child, PCIY_HT, &ptr);
1397 	if (error)
1398 		return (error);
1399 
1400 	/*
1401 	 * Traverse the capabilities list checking each HT capability
1402 	 * to see if it matches the requested HT capability.
1403 	 */
1404 	for (;;) {
1405 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1406 		if (capability == PCIM_HTCAP_SLAVE ||
1407 		    capability == PCIM_HTCAP_HOST)
1408 			val &= 0xe000;
1409 		else
1410 			val &= PCIM_HTCMD_CAP_MASK;
1411 		if (val == capability) {
1412 			if (capreg != NULL)
1413 				*capreg = ptr;
1414 			return (0);
1415 		}
1416 
1417 		/* Skip to the next HT capability. */
1418 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1419 			break;
1420 	}
1421 
1422 	return (ENOENT);
1423 }
1424 
1425 /*
1426  * Find the next requested HyperTransport capability after start and return
1427  * the offset in configuration space via the pointer provided.  The function
1428  * returns 0 on success and an error code otherwise.
1429  */
1430 int
1431 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1432     int start, int *capreg)
1433 {
1434 	int ptr;
1435 	uint16_t val;
1436 
1437 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1438 	    ("start capability is not HyperTransport capability"));
1439 	ptr = start;
1440 
1441 	/*
1442 	 * Traverse the capabilities list checking each HT capability
1443 	 * to see if it matches the requested HT capability.
1444 	 */
1445 	for (;;) {
1446 		/* Skip to the next HT capability. */
1447 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1448 			break;
1449 
1450 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1451 		if (capability == PCIM_HTCAP_SLAVE ||
1452 		    capability == PCIM_HTCAP_HOST)
1453 			val &= 0xe000;
1454 		else
1455 			val &= PCIM_HTCMD_CAP_MASK;
1456 		if (val == capability) {
1457 			if (capreg != NULL)
1458 				*capreg = ptr;
1459 			return (0);
1460 		}
1461 	}
1462 
1463 	return (ENOENT);
1464 }
1465 
1466 /*
1467  * Find the requested capability and return the offset in
1468  * configuration space via the pointer provided.  The function returns
1469  * 0 on success and an error code otherwise.
1470  */
1471 int
1472 pci_find_cap_method(device_t dev, device_t child, int capability,
1473     int *capreg)
1474 {
1475 	struct pci_devinfo *dinfo = device_get_ivars(child);
1476 	pcicfgregs *cfg = &dinfo->cfg;
1477 	uint32_t status;
1478 	uint8_t ptr;
1479 
1480 	/*
1481 	 * Check the CAP_LIST bit of the PCI status register first.
1482 	 */
1483 	status = pci_read_config(child, PCIR_STATUS, 2);
1484 	if (!(status & PCIM_STATUS_CAPPRESENT))
1485 		return (ENXIO);
1486 
1487 	/*
1488 	 * Determine the start pointer of the capabilities list.
1489 	 */
1490 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1491 	case PCIM_HDRTYPE_NORMAL:
1492 	case PCIM_HDRTYPE_BRIDGE:
1493 		ptr = PCIR_CAP_PTR;
1494 		break;
1495 	case PCIM_HDRTYPE_CARDBUS:
1496 		ptr = PCIR_CAP_PTR_2;
1497 		break;
1498 	default:
1499 		/* XXX: panic? */
1500 		return (ENXIO);		/* no extended capabilities support */
1501 	}
1502 	ptr = pci_read_config(child, ptr, 1);
1503 
1504 	/*
1505 	 * Traverse the capabilities list.
1506 	 */
1507 	while (ptr != 0) {
1508 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1509 			if (capreg != NULL)
1510 				*capreg = ptr;
1511 			return (0);
1512 		}
1513 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1514 	}
1515 
1516 	return (ENOENT);
1517 }
1518 
1519 /*
1520  * Find the next requested capability after start and return the offset in
1521  * configuration space via the pointer provided.  The function returns
1522  * 0 on success and an error code otherwise.
1523  */
1524 int
1525 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1526     int start, int *capreg)
1527 {
1528 	uint8_t ptr;
1529 
1530 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1531 	    ("start capability is not expected capability"));
1532 
1533 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1534 	while (ptr != 0) {
1535 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1536 			if (capreg != NULL)
1537 				*capreg = ptr;
1538 			return (0);
1539 		}
1540 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1541 	}
1542 
1543 	return (ENOENT);
1544 }
1545 
1546 /*
1547  * Find the requested extended capability and return the offset in
1548  * configuration space via the pointer provided.  The function returns
1549  * 0 on success and an error code otherwise.
1550  */
1551 int
1552 pci_find_extcap_method(device_t dev, device_t child, int capability,
1553     int *capreg)
1554 {
1555 	struct pci_devinfo *dinfo = device_get_ivars(child);
1556 	pcicfgregs *cfg = &dinfo->cfg;
1557 	uint32_t ecap;
1558 	uint16_t ptr;
1559 
1560 	/* Only supported for PCI-express devices. */
1561 	if (cfg->pcie.pcie_location == 0)
1562 		return (ENXIO);
1563 
1564 	ptr = PCIR_EXTCAP;
1565 	ecap = pci_read_config(child, ptr, 4);
1566 	if (ecap == 0xffffffff || ecap == 0)
1567 		return (ENOENT);
1568 	for (;;) {
1569 		if (PCI_EXTCAP_ID(ecap) == capability) {
1570 			if (capreg != NULL)
1571 				*capreg = ptr;
1572 			return (0);
1573 		}
1574 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1575 		if (ptr == 0)
1576 			break;
1577 		ecap = pci_read_config(child, ptr, 4);
1578 	}
1579 
1580 	return (ENOENT);
1581 }
1582 
1583 /*
1584  * Find the next requested extended capability after start and return the
1585  * offset in configuration space via the pointer provided.  The function
1586  * returns 0 on success and an error code otherwise.
1587  */
1588 int
1589 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1590     int start, 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 	ecap = pci_read_config(child, start, 4);
1602 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1603 	    ("start extended capability is not expected capability"));
1604 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1605 	while (ptr != 0) {
1606 		ecap = pci_read_config(child, ptr, 4);
1607 		if (PCI_EXTCAP_ID(ecap) == capability) {
1608 			if (capreg != NULL)
1609 				*capreg = ptr;
1610 			return (0);
1611 		}
1612 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1613 	}
1614 
1615 	return (ENOENT);
1616 }
1617 
1618 /*
1619  * Support for MSI-X message interrupts.
1620  */
1621 static void
1622 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1623 {
1624 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1625 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1626 	uint32_t offset;
1627 
1628 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1629 	offset = msix->msix_table_offset + index * 16;
1630 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1631 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1632 	bus_write_4(msix->msix_table_res, offset + 8, data);
1633 }
1634 
1635 void
1636 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1637     uint64_t address, uint32_t data)
1638 {
1639 
1640 	if (pci_msix_rewrite_table) {
1641 		struct pci_devinfo *dinfo = device_get_ivars(child);
1642 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1643 
1644 		/*
1645 		 * Some VM hosts require MSIX to be disabled in the
1646 		 * control register before updating the MSIX table
1647 		 * entries are allowed. It is not enough to only
1648 		 * disable MSIX while updating a single entry. MSIX
1649 		 * must be disabled while updating all entries in the
1650 		 * table.
1651 		 */
1652 		pci_write_config(child,
1653 		    msix->msix_location + PCIR_MSIX_CTRL,
1654 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1655 		pci_resume_msix(child);
1656 	} else
1657 		pci_write_msix_entry(child, index, address, data);
1658 
1659 	/* Enable MSI -> HT mapping. */
1660 	pci_ht_map_msi(child, address);
1661 }
1662 
1663 void
1664 pci_mask_msix(device_t dev, u_int index)
1665 {
1666 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1667 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1668 	uint32_t offset, val;
1669 
1670 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1671 	offset = msix->msix_table_offset + index * 16 + 12;
1672 	val = bus_read_4(msix->msix_table_res, offset);
1673 	val |= PCIM_MSIX_VCTRL_MASK;
1674 
1675 	/*
1676 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1677 	 * register, so always write the new value.
1678 	 */
1679 	bus_write_4(msix->msix_table_res, offset, val);
1680 }
1681 
1682 void
1683 pci_unmask_msix(device_t dev, u_int index)
1684 {
1685 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1686 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1687 	uint32_t offset, val;
1688 
1689 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1690 	offset = msix->msix_table_offset + index * 16 + 12;
1691 	val = bus_read_4(msix->msix_table_res, offset);
1692 	val &= ~PCIM_MSIX_VCTRL_MASK;
1693 
1694 	/*
1695 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1696 	 * register, so always write the new value.
1697 	 */
1698 	bus_write_4(msix->msix_table_res, offset, val);
1699 }
1700 
1701 int
1702 pci_pending_msix(device_t dev, u_int index)
1703 {
1704 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1705 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1706 	uint32_t offset, bit;
1707 
1708 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1709 	offset = msix->msix_pba_offset + (index / 32) * 4;
1710 	bit = 1 << index % 32;
1711 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1712 }
1713 
1714 /*
1715  * Restore MSI-X registers and table during resume.  If MSI-X is
1716  * enabled then walk the virtual table to restore the actual MSI-X
1717  * table.
1718  */
1719 static void
1720 pci_resume_msix(device_t dev)
1721 {
1722 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1723 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1724 	struct msix_table_entry *mte;
1725 	struct msix_vector *mv;
1726 	int i;
1727 
1728 	if (msix->msix_alloc > 0) {
1729 		/* First, mask all vectors. */
1730 		for (i = 0; i < msix->msix_msgnum; i++)
1731 			pci_mask_msix(dev, i);
1732 
1733 		/* Second, program any messages with at least one handler. */
1734 		for (i = 0; i < msix->msix_table_len; i++) {
1735 			mte = &msix->msix_table[i];
1736 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1737 				continue;
1738 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1739 			pci_write_msix_entry(dev, i, mv->mv_address,
1740 			    mv->mv_data);
1741 			pci_unmask_msix(dev, i);
1742 		}
1743 	}
1744 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1745 	    msix->msix_ctrl, 2);
1746 }
1747 
1748 /*
1749  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1750  * returned in *count.  After this function returns, each message will be
1751  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1752  */
1753 int
1754 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1755 {
1756 	struct pci_devinfo *dinfo = device_get_ivars(child);
1757 	pcicfgregs *cfg = &dinfo->cfg;
1758 	struct resource_list_entry *rle;
1759 	int actual, error, i, irq, max;
1760 
1761 	/* Don't let count == 0 get us into trouble. */
1762 	if (*count == 0)
1763 		return (EINVAL);
1764 
1765 	/* If rid 0 is allocated, then fail. */
1766 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1767 	if (rle != NULL && rle->res != NULL)
1768 		return (ENXIO);
1769 
1770 	/* Already have allocated messages? */
1771 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1772 		return (ENXIO);
1773 
1774 	/* If MSI-X is blacklisted for this system, fail. */
1775 	if (pci_msix_blacklisted())
1776 		return (ENXIO);
1777 
1778 	/* MSI-X capability present? */
1779 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1780 		return (ENODEV);
1781 
1782 	/* Make sure the appropriate BARs are mapped. */
1783 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1784 	    cfg->msix.msix_table_bar);
1785 	if (rle == NULL || rle->res == NULL ||
1786 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1787 		return (ENXIO);
1788 	cfg->msix.msix_table_res = rle->res;
1789 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1790 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1791 		    cfg->msix.msix_pba_bar);
1792 		if (rle == NULL || rle->res == NULL ||
1793 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1794 			return (ENXIO);
1795 	}
1796 	cfg->msix.msix_pba_res = rle->res;
1797 
1798 	if (bootverbose)
1799 		device_printf(child,
1800 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1801 		    *count, cfg->msix.msix_msgnum);
1802 	max = min(*count, cfg->msix.msix_msgnum);
1803 	for (i = 0; i < max; i++) {
1804 		/* Allocate a message. */
1805 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1806 		if (error) {
1807 			if (i == 0)
1808 				return (error);
1809 			break;
1810 		}
1811 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1812 		    irq, 1);
1813 	}
1814 	actual = i;
1815 
1816 	if (bootverbose) {
1817 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1818 		if (actual == 1)
1819 			device_printf(child, "using IRQ %ju for MSI-X\n",
1820 			    rle->start);
1821 		else {
1822 			int run;
1823 
1824 			/*
1825 			 * Be fancy and try to print contiguous runs of
1826 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1827 			 * 'run' is true if we are in a range.
1828 			 */
1829 			device_printf(child, "using IRQs %ju", rle->start);
1830 			irq = rle->start;
1831 			run = 0;
1832 			for (i = 1; i < actual; i++) {
1833 				rle = resource_list_find(&dinfo->resources,
1834 				    SYS_RES_IRQ, i + 1);
1835 
1836 				/* Still in a run? */
1837 				if (rle->start == irq + 1) {
1838 					run = 1;
1839 					irq++;
1840 					continue;
1841 				}
1842 
1843 				/* Finish previous range. */
1844 				if (run) {
1845 					printf("-%d", irq);
1846 					run = 0;
1847 				}
1848 
1849 				/* Start new range. */
1850 				printf(",%ju", rle->start);
1851 				irq = rle->start;
1852 			}
1853 
1854 			/* Unfinished range? */
1855 			if (run)
1856 				printf("-%d", irq);
1857 			printf(" for MSI-X\n");
1858 		}
1859 	}
1860 
1861 	/* Mask all vectors. */
1862 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1863 		pci_mask_msix(child, i);
1864 
1865 	/* Allocate and initialize vector data and virtual table. */
1866 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1867 	    M_DEVBUF, M_WAITOK | M_ZERO);
1868 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1869 	    M_DEVBUF, M_WAITOK | M_ZERO);
1870 	for (i = 0; i < actual; i++) {
1871 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1872 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1873 		cfg->msix.msix_table[i].mte_vector = i + 1;
1874 	}
1875 
1876 	/* Update control register to enable MSI-X. */
1877 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1878 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1879 	    cfg->msix.msix_ctrl, 2);
1880 
1881 	/* Update counts of alloc'd messages. */
1882 	cfg->msix.msix_alloc = actual;
1883 	cfg->msix.msix_table_len = actual;
1884 	*count = actual;
1885 	return (0);
1886 }
1887 
1888 /*
1889  * By default, pci_alloc_msix() will assign the allocated IRQ
1890  * resources consecutively to the first N messages in the MSI-X table.
1891  * However, device drivers may want to use different layouts if they
1892  * either receive fewer messages than they asked for, or they wish to
1893  * populate the MSI-X table sparsely.  This method allows the driver
1894  * to specify what layout it wants.  It must be called after a
1895  * successful pci_alloc_msix() but before any of the associated
1896  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1897  *
1898  * The 'vectors' array contains 'count' message vectors.  The array
1899  * maps directly to the MSI-X table in that index 0 in the array
1900  * specifies the vector for the first message in the MSI-X table, etc.
1901  * The vector value in each array index can either be 0 to indicate
1902  * that no vector should be assigned to a message slot, or it can be a
1903  * number from 1 to N (where N is the count returned from a
1904  * succcessful call to pci_alloc_msix()) to indicate which message
1905  * vector (IRQ) to be used for the corresponding message.
1906  *
1907  * On successful return, each message with a non-zero vector will have
1908  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1909  * 1.  Additionally, if any of the IRQs allocated via the previous
1910  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1911  * will be freed back to the system automatically.
1912  *
1913  * For example, suppose a driver has a MSI-X table with 6 messages and
1914  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1915  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1916  * C.  After the call to pci_alloc_msix(), the device will be setup to
1917  * have an MSI-X table of ABC--- (where - means no vector assigned).
1918  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1919  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1920  * be freed back to the system.  This device will also have valid
1921  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1922  *
1923  * In any case, the SYS_RES_IRQ rid X will always map to the message
1924  * at MSI-X table index X - 1 and will only be valid if a vector is
1925  * assigned to that table entry.
1926  */
1927 int
1928 pci_remap_msix_method(device_t dev, device_t child, int count,
1929     const u_int *vectors)
1930 {
1931 	struct pci_devinfo *dinfo = device_get_ivars(child);
1932 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1933 	struct resource_list_entry *rle;
1934 	int i, irq, j, *used;
1935 
1936 	/*
1937 	 * Have to have at least one message in the table but the
1938 	 * table can't be bigger than the actual MSI-X table in the
1939 	 * device.
1940 	 */
1941 	if (count == 0 || count > msix->msix_msgnum)
1942 		return (EINVAL);
1943 
1944 	/* Sanity check the vectors. */
1945 	for (i = 0; i < count; i++)
1946 		if (vectors[i] > msix->msix_alloc)
1947 			return (EINVAL);
1948 
1949 	/*
1950 	 * Make sure there aren't any holes in the vectors to be used.
1951 	 * It's a big pain to support it, and it doesn't really make
1952 	 * sense anyway.  Also, at least one vector must be used.
1953 	 */
1954 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1955 	    M_ZERO);
1956 	for (i = 0; i < count; i++)
1957 		if (vectors[i] != 0)
1958 			used[vectors[i] - 1] = 1;
1959 	for (i = 0; i < msix->msix_alloc - 1; i++)
1960 		if (used[i] == 0 && used[i + 1] == 1) {
1961 			free(used, M_DEVBUF);
1962 			return (EINVAL);
1963 		}
1964 	if (used[0] != 1) {
1965 		free(used, M_DEVBUF);
1966 		return (EINVAL);
1967 	}
1968 
1969 	/* Make sure none of the resources are allocated. */
1970 	for (i = 0; i < msix->msix_table_len; i++) {
1971 		if (msix->msix_table[i].mte_vector == 0)
1972 			continue;
1973 		if (msix->msix_table[i].mte_handlers > 0) {
1974 			free(used, M_DEVBUF);
1975 			return (EBUSY);
1976 		}
1977 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1978 		KASSERT(rle != NULL, ("missing resource"));
1979 		if (rle->res != NULL) {
1980 			free(used, M_DEVBUF);
1981 			return (EBUSY);
1982 		}
1983 	}
1984 
1985 	/* Free the existing resource list entries. */
1986 	for (i = 0; i < msix->msix_table_len; i++) {
1987 		if (msix->msix_table[i].mte_vector == 0)
1988 			continue;
1989 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1990 	}
1991 
1992 	/*
1993 	 * Build the new virtual table keeping track of which vectors are
1994 	 * used.
1995 	 */
1996 	free(msix->msix_table, M_DEVBUF);
1997 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1998 	    M_DEVBUF, M_WAITOK | M_ZERO);
1999 	for (i = 0; i < count; i++)
2000 		msix->msix_table[i].mte_vector = vectors[i];
2001 	msix->msix_table_len = count;
2002 
2003 	/* Free any unused IRQs and resize the vectors array if necessary. */
2004 	j = msix->msix_alloc - 1;
2005 	if (used[j] == 0) {
2006 		struct msix_vector *vec;
2007 
2008 		while (used[j] == 0) {
2009 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2010 			    msix->msix_vectors[j].mv_irq);
2011 			j--;
2012 		}
2013 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2014 		    M_WAITOK);
2015 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2016 		    (j + 1));
2017 		free(msix->msix_vectors, M_DEVBUF);
2018 		msix->msix_vectors = vec;
2019 		msix->msix_alloc = j + 1;
2020 	}
2021 	free(used, M_DEVBUF);
2022 
2023 	/* Map the IRQs onto the rids. */
2024 	for (i = 0; i < count; i++) {
2025 		if (vectors[i] == 0)
2026 			continue;
2027 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2028 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2029 		    irq, 1);
2030 	}
2031 
2032 	if (bootverbose) {
2033 		device_printf(child, "Remapped MSI-X IRQs as: ");
2034 		for (i = 0; i < count; i++) {
2035 			if (i != 0)
2036 				printf(", ");
2037 			if (vectors[i] == 0)
2038 				printf("---");
2039 			else
2040 				printf("%d",
2041 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2042 		}
2043 		printf("\n");
2044 	}
2045 
2046 	return (0);
2047 }
2048 
2049 static int
2050 pci_release_msix(device_t dev, device_t child)
2051 {
2052 	struct pci_devinfo *dinfo = device_get_ivars(child);
2053 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2054 	struct resource_list_entry *rle;
2055 	int i;
2056 
2057 	/* Do we have any messages to release? */
2058 	if (msix->msix_alloc == 0)
2059 		return (ENODEV);
2060 
2061 	/* Make sure none of the resources are allocated. */
2062 	for (i = 0; i < msix->msix_table_len; i++) {
2063 		if (msix->msix_table[i].mte_vector == 0)
2064 			continue;
2065 		if (msix->msix_table[i].mte_handlers > 0)
2066 			return (EBUSY);
2067 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2068 		KASSERT(rle != NULL, ("missing resource"));
2069 		if (rle->res != NULL)
2070 			return (EBUSY);
2071 	}
2072 
2073 	/* Update control register to disable MSI-X. */
2074 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2075 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2076 	    msix->msix_ctrl, 2);
2077 
2078 	/* Free the resource list entries. */
2079 	for (i = 0; i < msix->msix_table_len; i++) {
2080 		if (msix->msix_table[i].mte_vector == 0)
2081 			continue;
2082 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2083 	}
2084 	free(msix->msix_table, M_DEVBUF);
2085 	msix->msix_table_len = 0;
2086 
2087 	/* Release the IRQs. */
2088 	for (i = 0; i < msix->msix_alloc; i++)
2089 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2090 		    msix->msix_vectors[i].mv_irq);
2091 	free(msix->msix_vectors, M_DEVBUF);
2092 	msix->msix_alloc = 0;
2093 	return (0);
2094 }
2095 
2096 /*
2097  * Return the max supported MSI-X messages this device supports.
2098  * Basically, assuming the MD code can alloc messages, this function
2099  * should return the maximum value that pci_alloc_msix() can return.
2100  * Thus, it is subject to the tunables, etc.
2101  */
2102 int
2103 pci_msix_count_method(device_t dev, device_t child)
2104 {
2105 	struct pci_devinfo *dinfo = device_get_ivars(child);
2106 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2107 
2108 	if (pci_do_msix && msix->msix_location != 0)
2109 		return (msix->msix_msgnum);
2110 	return (0);
2111 }
2112 
2113 int
2114 pci_msix_pba_bar_method(device_t dev, device_t child)
2115 {
2116 	struct pci_devinfo *dinfo = device_get_ivars(child);
2117 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2118 
2119 	if (pci_do_msix && msix->msix_location != 0)
2120 		return (msix->msix_pba_bar);
2121 	return (-1);
2122 }
2123 
2124 int
2125 pci_msix_table_bar_method(device_t dev, device_t child)
2126 {
2127 	struct pci_devinfo *dinfo = device_get_ivars(child);
2128 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2129 
2130 	if (pci_do_msix && msix->msix_location != 0)
2131 		return (msix->msix_table_bar);
2132 	return (-1);
2133 }
2134 
2135 /*
2136  * HyperTransport MSI mapping control
2137  */
2138 void
2139 pci_ht_map_msi(device_t dev, uint64_t addr)
2140 {
2141 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2142 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2143 
2144 	if (!ht->ht_msimap)
2145 		return;
2146 
2147 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2148 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2149 		/* Enable MSI -> HT mapping. */
2150 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2151 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2152 		    ht->ht_msictrl, 2);
2153 	}
2154 
2155 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2156 		/* Disable MSI -> HT mapping. */
2157 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2158 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2159 		    ht->ht_msictrl, 2);
2160 	}
2161 }
2162 
2163 int
2164 pci_get_max_payload(device_t dev)
2165 {
2166 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2167 	int cap;
2168 	uint16_t val;
2169 
2170 	cap = dinfo->cfg.pcie.pcie_location;
2171 	if (cap == 0)
2172 		return (0);
2173 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2174 	val &= PCIEM_CTL_MAX_PAYLOAD;
2175 	val >>= 5;
2176 	return (1 << (val + 7));
2177 }
2178 
2179 int
2180 pci_get_max_read_req(device_t dev)
2181 {
2182 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2183 	int cap;
2184 	uint16_t val;
2185 
2186 	cap = dinfo->cfg.pcie.pcie_location;
2187 	if (cap == 0)
2188 		return (0);
2189 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2190 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2191 	val >>= 12;
2192 	return (1 << (val + 7));
2193 }
2194 
2195 int
2196 pci_set_max_read_req(device_t dev, int size)
2197 {
2198 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2199 	int cap;
2200 	uint16_t val;
2201 
2202 	cap = dinfo->cfg.pcie.pcie_location;
2203 	if (cap == 0)
2204 		return (0);
2205 	if (size < 128)
2206 		size = 128;
2207 	if (size > 4096)
2208 		size = 4096;
2209 	size = (1 << (fls(size) - 1));
2210 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2211 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2212 	val |= (fls(size) - 8) << 12;
2213 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2214 	return (size);
2215 }
2216 
2217 uint32_t
2218 pcie_read_config(device_t dev, int reg, int width)
2219 {
2220 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2221 	int cap;
2222 
2223 	cap = dinfo->cfg.pcie.pcie_location;
2224 	if (cap == 0) {
2225 		if (width == 2)
2226 			return (0xffff);
2227 		return (0xffffffff);
2228 	}
2229 
2230 	return (pci_read_config(dev, cap + reg, width));
2231 }
2232 
2233 void
2234 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2235 {
2236 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2237 	int cap;
2238 
2239 	cap = dinfo->cfg.pcie.pcie_location;
2240 	if (cap == 0)
2241 		return;
2242 	pci_write_config(dev, cap + reg, value, width);
2243 }
2244 
2245 /*
2246  * Adjusts a PCI-e capability register by clearing the bits in mask
2247  * and setting the bits in (value & mask).  Bits not set in mask are
2248  * not adjusted.
2249  *
2250  * Returns the old value on success or all ones on failure.
2251  */
2252 uint32_t
2253 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2254     int width)
2255 {
2256 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2257 	uint32_t old, new;
2258 	int cap;
2259 
2260 	cap = dinfo->cfg.pcie.pcie_location;
2261 	if (cap == 0) {
2262 		if (width == 2)
2263 			return (0xffff);
2264 		return (0xffffffff);
2265 	}
2266 
2267 	old = pci_read_config(dev, cap + reg, width);
2268 	new = old & ~mask;
2269 	new |= (value & mask);
2270 	pci_write_config(dev, cap + reg, new, width);
2271 	return (old);
2272 }
2273 
2274 /*
2275  * Support for MSI message signalled interrupts.
2276  */
2277 void
2278 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2279     uint16_t data)
2280 {
2281 	struct pci_devinfo *dinfo = device_get_ivars(child);
2282 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2283 
2284 	/* Write data and address values. */
2285 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2286 	    address & 0xffffffff, 4);
2287 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2288 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2289 		    address >> 32, 4);
2290 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2291 		    data, 2);
2292 	} else
2293 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2294 		    2);
2295 
2296 	/* Enable MSI in the control register. */
2297 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2298 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2299 	    msi->msi_ctrl, 2);
2300 
2301 	/* Enable MSI -> HT mapping. */
2302 	pci_ht_map_msi(child, address);
2303 }
2304 
2305 void
2306 pci_disable_msi_method(device_t dev, device_t child)
2307 {
2308 	struct pci_devinfo *dinfo = device_get_ivars(child);
2309 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2310 
2311 	/* Disable MSI -> HT mapping. */
2312 	pci_ht_map_msi(child, 0);
2313 
2314 	/* Disable MSI in the control register. */
2315 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2316 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2317 	    msi->msi_ctrl, 2);
2318 }
2319 
2320 /*
2321  * Restore MSI registers during resume.  If MSI is enabled then
2322  * restore the data and address registers in addition to the control
2323  * register.
2324  */
2325 static void
2326 pci_resume_msi(device_t dev)
2327 {
2328 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2329 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2330 	uint64_t address;
2331 	uint16_t data;
2332 
2333 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2334 		address = msi->msi_addr;
2335 		data = msi->msi_data;
2336 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2337 		    address & 0xffffffff, 4);
2338 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2339 			pci_write_config(dev, msi->msi_location +
2340 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2341 			pci_write_config(dev, msi->msi_location +
2342 			    PCIR_MSI_DATA_64BIT, data, 2);
2343 		} else
2344 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2345 			    data, 2);
2346 	}
2347 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2348 	    2);
2349 }
2350 
2351 static int
2352 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2353 {
2354 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2355 	pcicfgregs *cfg = &dinfo->cfg;
2356 	struct resource_list_entry *rle;
2357 	struct msix_table_entry *mte;
2358 	struct msix_vector *mv;
2359 	uint64_t addr;
2360 	uint32_t data;
2361 	int error, i, j;
2362 
2363 	/*
2364 	 * Handle MSI first.  We try to find this IRQ among our list
2365 	 * of MSI IRQs.  If we find it, we request updated address and
2366 	 * data registers and apply the results.
2367 	 */
2368 	if (cfg->msi.msi_alloc > 0) {
2369 		/* If we don't have any active handlers, nothing to do. */
2370 		if (cfg->msi.msi_handlers == 0)
2371 			return (0);
2372 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2373 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2374 			    i + 1);
2375 			if (rle->start == irq) {
2376 				error = PCIB_MAP_MSI(device_get_parent(bus),
2377 				    dev, irq, &addr, &data);
2378 				if (error)
2379 					return (error);
2380 				pci_disable_msi(dev);
2381 				dinfo->cfg.msi.msi_addr = addr;
2382 				dinfo->cfg.msi.msi_data = data;
2383 				pci_enable_msi(dev, addr, data);
2384 				return (0);
2385 			}
2386 		}
2387 		return (ENOENT);
2388 	}
2389 
2390 	/*
2391 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2392 	 * we request the updated mapping info.  If that works, we go
2393 	 * through all the slots that use this IRQ and update them.
2394 	 */
2395 	if (cfg->msix.msix_alloc > 0) {
2396 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2397 			mv = &cfg->msix.msix_vectors[i];
2398 			if (mv->mv_irq == irq) {
2399 				error = PCIB_MAP_MSI(device_get_parent(bus),
2400 				    dev, irq, &addr, &data);
2401 				if (error)
2402 					return (error);
2403 				mv->mv_address = addr;
2404 				mv->mv_data = data;
2405 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2406 					mte = &cfg->msix.msix_table[j];
2407 					if (mte->mte_vector != i + 1)
2408 						continue;
2409 					if (mte->mte_handlers == 0)
2410 						continue;
2411 					pci_mask_msix(dev, j);
2412 					pci_enable_msix(dev, j, addr, data);
2413 					pci_unmask_msix(dev, j);
2414 				}
2415 			}
2416 		}
2417 		return (ENOENT);
2418 	}
2419 
2420 	return (ENOENT);
2421 }
2422 
2423 /*
2424  * Returns true if the specified device is blacklisted because MSI
2425  * doesn't work.
2426  */
2427 int
2428 pci_msi_device_blacklisted(device_t dev)
2429 {
2430 
2431 	if (!pci_honor_msi_blacklist)
2432 		return (0);
2433 
2434 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2435 }
2436 
2437 /*
2438  * Determine if MSI is blacklisted globally on this system.  Currently,
2439  * we just check for blacklisted chipsets as represented by the
2440  * host-PCI bridge at device 0:0:0.  In the future, it may become
2441  * necessary to check other system attributes, such as the kenv values
2442  * that give the motherboard manufacturer and model number.
2443  */
2444 static int
2445 pci_msi_blacklisted(void)
2446 {
2447 	device_t dev;
2448 
2449 	if (!pci_honor_msi_blacklist)
2450 		return (0);
2451 
2452 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2453 	if (!(pcie_chipset || pcix_chipset)) {
2454 		if (vm_guest != VM_GUEST_NO) {
2455 			/*
2456 			 * Whitelist older chipsets in virtual
2457 			 * machines known to support MSI.
2458 			 */
2459 			dev = pci_find_bsf(0, 0, 0);
2460 			if (dev != NULL)
2461 				return (!pci_has_quirk(pci_get_devid(dev),
2462 					PCI_QUIRK_ENABLE_MSI_VM));
2463 		}
2464 		return (1);
2465 	}
2466 
2467 	dev = pci_find_bsf(0, 0, 0);
2468 	if (dev != NULL)
2469 		return (pci_msi_device_blacklisted(dev));
2470 	return (0);
2471 }
2472 
2473 /*
2474  * Returns true if the specified device is blacklisted because MSI-X
2475  * doesn't work.  Note that this assumes that if MSI doesn't work,
2476  * MSI-X doesn't either.
2477  */
2478 int
2479 pci_msix_device_blacklisted(device_t dev)
2480 {
2481 
2482 	if (!pci_honor_msi_blacklist)
2483 		return (0);
2484 
2485 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2486 		return (1);
2487 
2488 	return (pci_msi_device_blacklisted(dev));
2489 }
2490 
2491 /*
2492  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2493  * is blacklisted, assume that MSI-X is as well.  Check for additional
2494  * chipsets where MSI works but MSI-X does not.
2495  */
2496 static int
2497 pci_msix_blacklisted(void)
2498 {
2499 	device_t dev;
2500 
2501 	if (!pci_honor_msi_blacklist)
2502 		return (0);
2503 
2504 	dev = pci_find_bsf(0, 0, 0);
2505 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2506 	    PCI_QUIRK_DISABLE_MSIX))
2507 		return (1);
2508 
2509 	return (pci_msi_blacklisted());
2510 }
2511 
2512 /*
2513  * Attempt to allocate *count MSI messages.  The actual number allocated is
2514  * returned in *count.  After this function returns, each message will be
2515  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2516  */
2517 int
2518 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2519 {
2520 	struct pci_devinfo *dinfo = device_get_ivars(child);
2521 	pcicfgregs *cfg = &dinfo->cfg;
2522 	struct resource_list_entry *rle;
2523 	int actual, error, i, irqs[32];
2524 	uint16_t ctrl;
2525 
2526 	/* Don't let count == 0 get us into trouble. */
2527 	if (*count == 0)
2528 		return (EINVAL);
2529 
2530 	/* If rid 0 is allocated, then fail. */
2531 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2532 	if (rle != NULL && rle->res != NULL)
2533 		return (ENXIO);
2534 
2535 	/* Already have allocated messages? */
2536 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2537 		return (ENXIO);
2538 
2539 	/* If MSI is blacklisted for this system, fail. */
2540 	if (pci_msi_blacklisted())
2541 		return (ENXIO);
2542 
2543 	/* MSI capability present? */
2544 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2545 		return (ENODEV);
2546 
2547 	if (bootverbose)
2548 		device_printf(child,
2549 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2550 		    *count, cfg->msi.msi_msgnum);
2551 
2552 	/* Don't ask for more than the device supports. */
2553 	actual = min(*count, cfg->msi.msi_msgnum);
2554 
2555 	/* Don't ask for more than 32 messages. */
2556 	actual = min(actual, 32);
2557 
2558 	/* MSI requires power of 2 number of messages. */
2559 	if (!powerof2(actual))
2560 		return (EINVAL);
2561 
2562 	for (;;) {
2563 		/* Try to allocate N messages. */
2564 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2565 		    actual, irqs);
2566 		if (error == 0)
2567 			break;
2568 		if (actual == 1)
2569 			return (error);
2570 
2571 		/* Try N / 2. */
2572 		actual >>= 1;
2573 	}
2574 
2575 	/*
2576 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2577 	 * resources in the irqs[] array, so add new resources
2578 	 * starting at rid 1.
2579 	 */
2580 	for (i = 0; i < actual; i++)
2581 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2582 		    irqs[i], irqs[i], 1);
2583 
2584 	if (bootverbose) {
2585 		if (actual == 1)
2586 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2587 		else {
2588 			int run;
2589 
2590 			/*
2591 			 * Be fancy and try to print contiguous runs
2592 			 * of IRQ values as ranges.  'run' is true if
2593 			 * we are in a range.
2594 			 */
2595 			device_printf(child, "using IRQs %d", irqs[0]);
2596 			run = 0;
2597 			for (i = 1; i < actual; i++) {
2598 				/* Still in a run? */
2599 				if (irqs[i] == irqs[i - 1] + 1) {
2600 					run = 1;
2601 					continue;
2602 				}
2603 
2604 				/* Finish previous range. */
2605 				if (run) {
2606 					printf("-%d", irqs[i - 1]);
2607 					run = 0;
2608 				}
2609 
2610 				/* Start new range. */
2611 				printf(",%d", irqs[i]);
2612 			}
2613 
2614 			/* Unfinished range? */
2615 			if (run)
2616 				printf("-%d", irqs[actual - 1]);
2617 			printf(" for MSI\n");
2618 		}
2619 	}
2620 
2621 	/* Update control register with actual count. */
2622 	ctrl = cfg->msi.msi_ctrl;
2623 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2624 	ctrl |= (ffs(actual) - 1) << 4;
2625 	cfg->msi.msi_ctrl = ctrl;
2626 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2627 
2628 	/* Update counts of alloc'd messages. */
2629 	cfg->msi.msi_alloc = actual;
2630 	cfg->msi.msi_handlers = 0;
2631 	*count = actual;
2632 	return (0);
2633 }
2634 
2635 /* Release the MSI messages associated with this device. */
2636 int
2637 pci_release_msi_method(device_t dev, device_t child)
2638 {
2639 	struct pci_devinfo *dinfo = device_get_ivars(child);
2640 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2641 	struct resource_list_entry *rle;
2642 	int error, i, irqs[32];
2643 
2644 	/* Try MSI-X first. */
2645 	error = pci_release_msix(dev, child);
2646 	if (error != ENODEV)
2647 		return (error);
2648 
2649 	/* Do we have any messages to release? */
2650 	if (msi->msi_alloc == 0)
2651 		return (ENODEV);
2652 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2653 
2654 	/* Make sure none of the resources are allocated. */
2655 	if (msi->msi_handlers > 0)
2656 		return (EBUSY);
2657 	for (i = 0; i < msi->msi_alloc; i++) {
2658 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2659 		KASSERT(rle != NULL, ("missing MSI resource"));
2660 		if (rle->res != NULL)
2661 			return (EBUSY);
2662 		irqs[i] = rle->start;
2663 	}
2664 
2665 	/* Update control register with 0 count. */
2666 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2667 	    ("%s: MSI still enabled", __func__));
2668 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2669 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2670 	    msi->msi_ctrl, 2);
2671 
2672 	/* Release the messages. */
2673 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2674 	for (i = 0; i < msi->msi_alloc; i++)
2675 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2676 
2677 	/* Update alloc count. */
2678 	msi->msi_alloc = 0;
2679 	msi->msi_addr = 0;
2680 	msi->msi_data = 0;
2681 	return (0);
2682 }
2683 
2684 /*
2685  * Return the max supported MSI messages this device supports.
2686  * Basically, assuming the MD code can alloc messages, this function
2687  * should return the maximum value that pci_alloc_msi() can return.
2688  * Thus, it is subject to the tunables, etc.
2689  */
2690 int
2691 pci_msi_count_method(device_t dev, device_t child)
2692 {
2693 	struct pci_devinfo *dinfo = device_get_ivars(child);
2694 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2695 
2696 	if (pci_do_msi && msi->msi_location != 0)
2697 		return (msi->msi_msgnum);
2698 	return (0);
2699 }
2700 
2701 /* free pcicfgregs structure and all depending data structures */
2702 
2703 int
2704 pci_freecfg(struct pci_devinfo *dinfo)
2705 {
2706 	struct devlist *devlist_head;
2707 	struct pci_map *pm, *next;
2708 	int i;
2709 
2710 	devlist_head = &pci_devq;
2711 
2712 	if (dinfo->cfg.vpd.vpd_reg) {
2713 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2714 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2715 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2716 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2717 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2718 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2719 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2720 	}
2721 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2722 		free(pm, M_DEVBUF);
2723 	}
2724 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2725 	free(dinfo, M_DEVBUF);
2726 
2727 	/* increment the generation count */
2728 	pci_generation++;
2729 
2730 	/* we're losing one device */
2731 	pci_numdevs--;
2732 	return (0);
2733 }
2734 
2735 /*
2736  * PCI power manangement
2737  */
2738 int
2739 pci_set_powerstate_method(device_t dev, device_t child, int state)
2740 {
2741 	struct pci_devinfo *dinfo = device_get_ivars(child);
2742 	pcicfgregs *cfg = &dinfo->cfg;
2743 	uint16_t status;
2744 	int oldstate, highest, delay;
2745 
2746 	if (cfg->pp.pp_cap == 0)
2747 		return (EOPNOTSUPP);
2748 
2749 	/*
2750 	 * Optimize a no state change request away.  While it would be OK to
2751 	 * write to the hardware in theory, some devices have shown odd
2752 	 * behavior when going from D3 -> D3.
2753 	 */
2754 	oldstate = pci_get_powerstate(child);
2755 	if (oldstate == state)
2756 		return (0);
2757 
2758 	/*
2759 	 * The PCI power management specification states that after a state
2760 	 * transition between PCI power states, system software must
2761 	 * guarantee a minimal delay before the function accesses the device.
2762 	 * Compute the worst case delay that we need to guarantee before we
2763 	 * access the device.  Many devices will be responsive much more
2764 	 * quickly than this delay, but there are some that don't respond
2765 	 * instantly to state changes.  Transitions to/from D3 state require
2766 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2767 	 * is done below with DELAY rather than a sleeper function because
2768 	 * this function can be called from contexts where we cannot sleep.
2769 	 */
2770 	highest = (oldstate > state) ? oldstate : state;
2771 	if (highest == PCI_POWERSTATE_D3)
2772 	    delay = 10000;
2773 	else if (highest == PCI_POWERSTATE_D2)
2774 	    delay = 200;
2775 	else
2776 	    delay = 0;
2777 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2778 	    & ~PCIM_PSTAT_DMASK;
2779 	switch (state) {
2780 	case PCI_POWERSTATE_D0:
2781 		status |= PCIM_PSTAT_D0;
2782 		break;
2783 	case PCI_POWERSTATE_D1:
2784 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2785 			return (EOPNOTSUPP);
2786 		status |= PCIM_PSTAT_D1;
2787 		break;
2788 	case PCI_POWERSTATE_D2:
2789 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2790 			return (EOPNOTSUPP);
2791 		status |= PCIM_PSTAT_D2;
2792 		break;
2793 	case PCI_POWERSTATE_D3:
2794 		status |= PCIM_PSTAT_D3;
2795 		break;
2796 	default:
2797 		return (EINVAL);
2798 	}
2799 
2800 	if (bootverbose)
2801 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2802 		    state);
2803 
2804 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2805 	if (delay)
2806 		DELAY(delay);
2807 	return (0);
2808 }
2809 
2810 int
2811 pci_get_powerstate_method(device_t dev, device_t child)
2812 {
2813 	struct pci_devinfo *dinfo = device_get_ivars(child);
2814 	pcicfgregs *cfg = &dinfo->cfg;
2815 	uint16_t status;
2816 	int result;
2817 
2818 	if (cfg->pp.pp_cap != 0) {
2819 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2820 		switch (status & PCIM_PSTAT_DMASK) {
2821 		case PCIM_PSTAT_D0:
2822 			result = PCI_POWERSTATE_D0;
2823 			break;
2824 		case PCIM_PSTAT_D1:
2825 			result = PCI_POWERSTATE_D1;
2826 			break;
2827 		case PCIM_PSTAT_D2:
2828 			result = PCI_POWERSTATE_D2;
2829 			break;
2830 		case PCIM_PSTAT_D3:
2831 			result = PCI_POWERSTATE_D3;
2832 			break;
2833 		default:
2834 			result = PCI_POWERSTATE_UNKNOWN;
2835 			break;
2836 		}
2837 	} else {
2838 		/* No support, device is always at D0 */
2839 		result = PCI_POWERSTATE_D0;
2840 	}
2841 	return (result);
2842 }
2843 
2844 /*
2845  * Some convenience functions for PCI device drivers.
2846  */
2847 
2848 static __inline void
2849 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2850 {
2851 	uint16_t	command;
2852 
2853 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2854 	command |= bit;
2855 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2856 }
2857 
2858 static __inline void
2859 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2860 {
2861 	uint16_t	command;
2862 
2863 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2864 	command &= ~bit;
2865 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2866 }
2867 
2868 int
2869 pci_enable_busmaster_method(device_t dev, device_t child)
2870 {
2871 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2872 	return (0);
2873 }
2874 
2875 int
2876 pci_disable_busmaster_method(device_t dev, device_t child)
2877 {
2878 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2879 	return (0);
2880 }
2881 
2882 int
2883 pci_enable_io_method(device_t dev, device_t child, int space)
2884 {
2885 	uint16_t bit;
2886 
2887 	switch(space) {
2888 	case SYS_RES_IOPORT:
2889 		bit = PCIM_CMD_PORTEN;
2890 		break;
2891 	case SYS_RES_MEMORY:
2892 		bit = PCIM_CMD_MEMEN;
2893 		break;
2894 	default:
2895 		return (EINVAL);
2896 	}
2897 	pci_set_command_bit(dev, child, bit);
2898 	return (0);
2899 }
2900 
2901 int
2902 pci_disable_io_method(device_t dev, device_t child, int space)
2903 {
2904 	uint16_t bit;
2905 
2906 	switch(space) {
2907 	case SYS_RES_IOPORT:
2908 		bit = PCIM_CMD_PORTEN;
2909 		break;
2910 	case SYS_RES_MEMORY:
2911 		bit = PCIM_CMD_MEMEN;
2912 		break;
2913 	default:
2914 		return (EINVAL);
2915 	}
2916 	pci_clear_command_bit(dev, child, bit);
2917 	return (0);
2918 }
2919 
2920 /*
2921  * New style pci driver.  Parent device is either a pci-host-bridge or a
2922  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2923  */
2924 
2925 void
2926 pci_print_verbose(struct pci_devinfo *dinfo)
2927 {
2928 
2929 	if (bootverbose) {
2930 		pcicfgregs *cfg = &dinfo->cfg;
2931 
2932 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2933 		    cfg->vendor, cfg->device, cfg->revid);
2934 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2935 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2936 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2937 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2938 		    cfg->mfdev);
2939 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2940 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2941 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2942 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2943 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2944 		if (cfg->intpin > 0)
2945 			printf("\tintpin=%c, irq=%d\n",
2946 			    cfg->intpin +'a' -1, cfg->intline);
2947 		if (cfg->pp.pp_cap) {
2948 			uint16_t status;
2949 
2950 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2951 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2952 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2953 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2954 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2955 			    status & PCIM_PSTAT_DMASK);
2956 		}
2957 		if (cfg->msi.msi_location) {
2958 			int ctrl;
2959 
2960 			ctrl = cfg->msi.msi_ctrl;
2961 			printf("\tMSI supports %d message%s%s%s\n",
2962 			    cfg->msi.msi_msgnum,
2963 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2964 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2965 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2966 		}
2967 		if (cfg->msix.msix_location) {
2968 			printf("\tMSI-X supports %d message%s ",
2969 			    cfg->msix.msix_msgnum,
2970 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2971 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2972 				printf("in map 0x%x\n",
2973 				    cfg->msix.msix_table_bar);
2974 			else
2975 				printf("in maps 0x%x and 0x%x\n",
2976 				    cfg->msix.msix_table_bar,
2977 				    cfg->msix.msix_pba_bar);
2978 		}
2979 	}
2980 }
2981 
2982 static int
2983 pci_porten(device_t dev)
2984 {
2985 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2986 }
2987 
2988 static int
2989 pci_memen(device_t dev)
2990 {
2991 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2992 }
2993 
2994 void
2995 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2996     int *bar64)
2997 {
2998 	struct pci_devinfo *dinfo;
2999 	pci_addr_t map, testval;
3000 	int ln2range;
3001 	uint16_t cmd;
3002 
3003 	/*
3004 	 * The device ROM BAR is special.  It is always a 32-bit
3005 	 * memory BAR.  Bit 0 is special and should not be set when
3006 	 * sizing the BAR.
3007 	 */
3008 	dinfo = device_get_ivars(dev);
3009 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3010 		map = pci_read_config(dev, reg, 4);
3011 		pci_write_config(dev, reg, 0xfffffffe, 4);
3012 		testval = pci_read_config(dev, reg, 4);
3013 		pci_write_config(dev, reg, map, 4);
3014 		*mapp = map;
3015 		*testvalp = testval;
3016 		if (bar64 != NULL)
3017 			*bar64 = 0;
3018 		return;
3019 	}
3020 
3021 	map = pci_read_config(dev, reg, 4);
3022 	ln2range = pci_maprange(map);
3023 	if (ln2range == 64)
3024 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3025 
3026 	/*
3027 	 * Disable decoding via the command register before
3028 	 * determining the BAR's length since we will be placing it in
3029 	 * a weird state.
3030 	 */
3031 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3032 	pci_write_config(dev, PCIR_COMMAND,
3033 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3034 
3035 	/*
3036 	 * Determine the BAR's length by writing all 1's.  The bottom
3037 	 * log_2(size) bits of the BAR will stick as 0 when we read
3038 	 * the value back.
3039 	 *
3040 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3041 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3042 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3043 	 *
3044 	 * Writes to both registers must be performed before attempting to
3045 	 * read back the size value.
3046 	 */
3047 	testval = 0;
3048 	pci_write_config(dev, reg, 0xffffffff, 4);
3049 	if (ln2range == 64) {
3050 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3051 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3052 	}
3053 	testval |= pci_read_config(dev, reg, 4);
3054 
3055 	/*
3056 	 * Restore the original value of the BAR.  We may have reprogrammed
3057 	 * the BAR of the low-level console device and when booting verbose,
3058 	 * we need the console device addressable.
3059 	 */
3060 	pci_write_config(dev, reg, map, 4);
3061 	if (ln2range == 64)
3062 		pci_write_config(dev, reg + 4, map >> 32, 4);
3063 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3064 
3065 	*mapp = map;
3066 	*testvalp = testval;
3067 	if (bar64 != NULL)
3068 		*bar64 = (ln2range == 64);
3069 }
3070 
3071 static void
3072 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3073 {
3074 	struct pci_devinfo *dinfo;
3075 	int ln2range;
3076 
3077 	/* The device ROM BAR is always a 32-bit memory BAR. */
3078 	dinfo = device_get_ivars(dev);
3079 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3080 		ln2range = 32;
3081 	else
3082 		ln2range = pci_maprange(pm->pm_value);
3083 	pci_write_config(dev, pm->pm_reg, base, 4);
3084 	if (ln2range == 64)
3085 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3086 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3087 	if (ln2range == 64)
3088 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3089 		    pm->pm_reg + 4, 4) << 32;
3090 }
3091 
3092 struct pci_map *
3093 pci_find_bar(device_t dev, int reg)
3094 {
3095 	struct pci_devinfo *dinfo;
3096 	struct pci_map *pm;
3097 
3098 	dinfo = device_get_ivars(dev);
3099 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3100 		if (pm->pm_reg == reg)
3101 			return (pm);
3102 	}
3103 	return (NULL);
3104 }
3105 
3106 int
3107 pci_bar_enabled(device_t dev, struct pci_map *pm)
3108 {
3109 	struct pci_devinfo *dinfo;
3110 	uint16_t cmd;
3111 
3112 	dinfo = device_get_ivars(dev);
3113 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3114 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3115 		return (0);
3116 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3117 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3118 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3119 	else
3120 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3121 }
3122 
3123 struct pci_map *
3124 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3125 {
3126 	struct pci_devinfo *dinfo;
3127 	struct pci_map *pm, *prev;
3128 
3129 	dinfo = device_get_ivars(dev);
3130 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3131 	pm->pm_reg = reg;
3132 	pm->pm_value = value;
3133 	pm->pm_size = size;
3134 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3135 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3136 		    reg));
3137 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3138 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3139 			break;
3140 	}
3141 	if (prev != NULL)
3142 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3143 	else
3144 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3145 	return (pm);
3146 }
3147 
3148 static void
3149 pci_restore_bars(device_t dev)
3150 {
3151 	struct pci_devinfo *dinfo;
3152 	struct pci_map *pm;
3153 	int ln2range;
3154 
3155 	dinfo = device_get_ivars(dev);
3156 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3157 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3158 			ln2range = 32;
3159 		else
3160 			ln2range = pci_maprange(pm->pm_value);
3161 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3162 		if (ln2range == 64)
3163 			pci_write_config(dev, pm->pm_reg + 4,
3164 			    pm->pm_value >> 32, 4);
3165 	}
3166 }
3167 
3168 /*
3169  * Add a resource based on a pci map register. Return 1 if the map
3170  * register is a 32bit map register or 2 if it is a 64bit register.
3171  */
3172 static int
3173 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3174     int force, int prefetch)
3175 {
3176 	struct pci_map *pm;
3177 	pci_addr_t base, map, testval;
3178 	pci_addr_t start, end, count;
3179 	int barlen, basezero, flags, maprange, mapsize, type;
3180 	uint16_t cmd;
3181 	struct resource *res;
3182 
3183 	/*
3184 	 * The BAR may already exist if the device is a CardBus card
3185 	 * whose CIS is stored in this BAR.
3186 	 */
3187 	pm = pci_find_bar(dev, reg);
3188 	if (pm != NULL) {
3189 		maprange = pci_maprange(pm->pm_value);
3190 		barlen = maprange == 64 ? 2 : 1;
3191 		return (barlen);
3192 	}
3193 
3194 	pci_read_bar(dev, reg, &map, &testval, NULL);
3195 	if (PCI_BAR_MEM(map)) {
3196 		type = SYS_RES_MEMORY;
3197 		if (map & PCIM_BAR_MEM_PREFETCH)
3198 			prefetch = 1;
3199 	} else
3200 		type = SYS_RES_IOPORT;
3201 	mapsize = pci_mapsize(testval);
3202 	base = pci_mapbase(map);
3203 #ifdef __PCI_BAR_ZERO_VALID
3204 	basezero = 0;
3205 #else
3206 	basezero = base == 0;
3207 #endif
3208 	maprange = pci_maprange(map);
3209 	barlen = maprange == 64 ? 2 : 1;
3210 
3211 	/*
3212 	 * For I/O registers, if bottom bit is set, and the next bit up
3213 	 * isn't clear, we know we have a BAR that doesn't conform to the
3214 	 * spec, so ignore it.  Also, sanity check the size of the data
3215 	 * areas to the type of memory involved.  Memory must be at least
3216 	 * 16 bytes in size, while I/O ranges must be at least 4.
3217 	 */
3218 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3219 		return (barlen);
3220 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3221 	    (type == SYS_RES_IOPORT && mapsize < 2))
3222 		return (barlen);
3223 
3224 	/* Save a record of this BAR. */
3225 	pm = pci_add_bar(dev, reg, map, mapsize);
3226 	if (bootverbose) {
3227 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3228 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3229 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3230 			printf(", port disabled\n");
3231 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3232 			printf(", memory disabled\n");
3233 		else
3234 			printf(", enabled\n");
3235 	}
3236 
3237 	/*
3238 	 * If base is 0, then we have problems if this architecture does
3239 	 * not allow that.  It is best to ignore such entries for the
3240 	 * moment.  These will be allocated later if the driver specifically
3241 	 * requests them.  However, some removable buses look better when
3242 	 * all resources are allocated, so allow '0' to be overriden.
3243 	 *
3244 	 * Similarly treat maps whose values is the same as the test value
3245 	 * read back.  These maps have had all f's written to them by the
3246 	 * BIOS in an attempt to disable the resources.
3247 	 */
3248 	if (!force && (basezero || map == testval))
3249 		return (barlen);
3250 	if ((u_long)base != base) {
3251 		device_printf(bus,
3252 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3253 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3254 		    pci_get_function(dev), reg);
3255 		return (barlen);
3256 	}
3257 
3258 	/*
3259 	 * This code theoretically does the right thing, but has
3260 	 * undesirable side effects in some cases where peripherals
3261 	 * respond oddly to having these bits enabled.  Let the user
3262 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3263 	 * default).
3264 	 */
3265 	if (pci_enable_io_modes) {
3266 		/* Turn on resources that have been left off by a lazy BIOS */
3267 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3268 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3269 			cmd |= PCIM_CMD_PORTEN;
3270 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3271 		}
3272 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3273 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3274 			cmd |= PCIM_CMD_MEMEN;
3275 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3276 		}
3277 	} else {
3278 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3279 			return (barlen);
3280 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3281 			return (barlen);
3282 	}
3283 
3284 	count = (pci_addr_t)1 << mapsize;
3285 	flags = RF_ALIGNMENT_LOG2(mapsize);
3286 	if (prefetch)
3287 		flags |= RF_PREFETCHABLE;
3288 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3289 		start = 0;	/* Let the parent decide. */
3290 		end = ~0;
3291 	} else {
3292 		start = base;
3293 		end = base + count - 1;
3294 	}
3295 	resource_list_add(rl, type, reg, start, end, count);
3296 
3297 	/*
3298 	 * Try to allocate the resource for this BAR from our parent
3299 	 * so that this resource range is already reserved.  The
3300 	 * driver for this device will later inherit this resource in
3301 	 * pci_alloc_resource().
3302 	 */
3303 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3304 	    flags);
3305 	if ((pci_do_realloc_bars
3306 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3307 	    && res == NULL && (start != 0 || end != ~0)) {
3308 		/*
3309 		 * If the allocation fails, try to allocate a resource for
3310 		 * this BAR using any available range.  The firmware felt
3311 		 * it was important enough to assign a resource, so don't
3312 		 * disable decoding if we can help it.
3313 		 */
3314 		resource_list_delete(rl, type, reg);
3315 		resource_list_add(rl, type, reg, 0, ~0, count);
3316 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3317 		    count, flags);
3318 	}
3319 	if (res == NULL) {
3320 		/*
3321 		 * If the allocation fails, delete the resource list entry
3322 		 * and disable decoding for this device.
3323 		 *
3324 		 * If the driver requests this resource in the future,
3325 		 * pci_reserve_map() will try to allocate a fresh
3326 		 * resource range.
3327 		 */
3328 		resource_list_delete(rl, type, reg);
3329 		pci_disable_io(dev, type);
3330 		if (bootverbose)
3331 			device_printf(bus,
3332 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3333 			    pci_get_domain(dev), pci_get_bus(dev),
3334 			    pci_get_slot(dev), pci_get_function(dev), reg);
3335 	} else {
3336 		start = rman_get_start(res);
3337 		pci_write_bar(dev, pm, start);
3338 	}
3339 	return (barlen);
3340 }
3341 
3342 /*
3343  * For ATA devices we need to decide early what addressing mode to use.
3344  * Legacy demands that the primary and secondary ATA ports sits on the
3345  * same addresses that old ISA hardware did. This dictates that we use
3346  * those addresses and ignore the BAR's if we cannot set PCI native
3347  * addressing mode.
3348  */
3349 static void
3350 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3351     uint32_t prefetchmask)
3352 {
3353 	int rid, type, progif;
3354 #if 0
3355 	/* if this device supports PCI native addressing use it */
3356 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3357 	if ((progif & 0x8a) == 0x8a) {
3358 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3359 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3360 			printf("Trying ATA native PCI addressing mode\n");
3361 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3362 		}
3363 	}
3364 #endif
3365 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3366 	type = SYS_RES_IOPORT;
3367 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3368 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3369 		    prefetchmask & (1 << 0));
3370 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3371 		    prefetchmask & (1 << 1));
3372 	} else {
3373 		rid = PCIR_BAR(0);
3374 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3375 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3376 		    0x1f7, 8, 0);
3377 		rid = PCIR_BAR(1);
3378 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3379 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3380 		    0x3f6, 1, 0);
3381 	}
3382 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3383 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3384 		    prefetchmask & (1 << 2));
3385 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3386 		    prefetchmask & (1 << 3));
3387 	} else {
3388 		rid = PCIR_BAR(2);
3389 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3390 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3391 		    0x177, 8, 0);
3392 		rid = PCIR_BAR(3);
3393 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3394 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3395 		    0x376, 1, 0);
3396 	}
3397 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3398 	    prefetchmask & (1 << 4));
3399 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3400 	    prefetchmask & (1 << 5));
3401 }
3402 
3403 static void
3404 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3405 {
3406 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3407 	pcicfgregs *cfg = &dinfo->cfg;
3408 	char tunable_name[64];
3409 	int irq;
3410 
3411 	/* Has to have an intpin to have an interrupt. */
3412 	if (cfg->intpin == 0)
3413 		return;
3414 
3415 	/* Let the user override the IRQ with a tunable. */
3416 	irq = PCI_INVALID_IRQ;
3417 	snprintf(tunable_name, sizeof(tunable_name),
3418 	    "hw.pci%d.%d.%d.INT%c.irq",
3419 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3420 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3421 		irq = PCI_INVALID_IRQ;
3422 
3423 	/*
3424 	 * If we didn't get an IRQ via the tunable, then we either use the
3425 	 * IRQ value in the intline register or we ask the bus to route an
3426 	 * interrupt for us.  If force_route is true, then we only use the
3427 	 * value in the intline register if the bus was unable to assign an
3428 	 * IRQ.
3429 	 */
3430 	if (!PCI_INTERRUPT_VALID(irq)) {
3431 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3432 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3433 		if (!PCI_INTERRUPT_VALID(irq))
3434 			irq = cfg->intline;
3435 	}
3436 
3437 	/* If after all that we don't have an IRQ, just bail. */
3438 	if (!PCI_INTERRUPT_VALID(irq))
3439 		return;
3440 
3441 	/* Update the config register if it changed. */
3442 	if (irq != cfg->intline) {
3443 		cfg->intline = irq;
3444 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3445 	}
3446 
3447 	/* Add this IRQ as rid 0 interrupt resource. */
3448 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3449 }
3450 
3451 /* Perform early OHCI takeover from SMM. */
3452 static void
3453 ohci_early_takeover(device_t self)
3454 {
3455 	struct resource *res;
3456 	uint32_t ctl;
3457 	int rid;
3458 	int i;
3459 
3460 	rid = PCIR_BAR(0);
3461 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3462 	if (res == NULL)
3463 		return;
3464 
3465 	ctl = bus_read_4(res, OHCI_CONTROL);
3466 	if (ctl & OHCI_IR) {
3467 		if (bootverbose)
3468 			printf("ohci early: "
3469 			    "SMM active, request owner change\n");
3470 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3471 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3472 			DELAY(1000);
3473 			ctl = bus_read_4(res, OHCI_CONTROL);
3474 		}
3475 		if (ctl & OHCI_IR) {
3476 			if (bootverbose)
3477 				printf("ohci early: "
3478 				    "SMM does not respond, resetting\n");
3479 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3480 		}
3481 		/* Disable interrupts */
3482 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3483 	}
3484 
3485 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3486 }
3487 
3488 /* Perform early UHCI takeover from SMM. */
3489 static void
3490 uhci_early_takeover(device_t self)
3491 {
3492 	struct resource *res;
3493 	int rid;
3494 
3495 	/*
3496 	 * Set the PIRQD enable bit and switch off all the others. We don't
3497 	 * want legacy support to interfere with us XXX Does this also mean
3498 	 * that the BIOS won't touch the keyboard anymore if it is connected
3499 	 * to the ports of the root hub?
3500 	 */
3501 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3502 
3503 	/* Disable interrupts */
3504 	rid = PCI_UHCI_BASE_REG;
3505 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3506 	if (res != NULL) {
3507 		bus_write_2(res, UHCI_INTR, 0);
3508 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3509 	}
3510 }
3511 
3512 /* Perform early EHCI takeover from SMM. */
3513 static void
3514 ehci_early_takeover(device_t self)
3515 {
3516 	struct resource *res;
3517 	uint32_t cparams;
3518 	uint32_t eec;
3519 	uint8_t eecp;
3520 	uint8_t bios_sem;
3521 	uint8_t offs;
3522 	int rid;
3523 	int i;
3524 
3525 	rid = PCIR_BAR(0);
3526 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3527 	if (res == NULL)
3528 		return;
3529 
3530 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3531 
3532 	/* Synchronise with the BIOS if it owns the controller. */
3533 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3534 	    eecp = EHCI_EECP_NEXT(eec)) {
3535 		eec = pci_read_config(self, eecp, 4);
3536 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3537 			continue;
3538 		}
3539 		bios_sem = pci_read_config(self, eecp +
3540 		    EHCI_LEGSUP_BIOS_SEM, 1);
3541 		if (bios_sem == 0) {
3542 			continue;
3543 		}
3544 		if (bootverbose)
3545 			printf("ehci early: "
3546 			    "SMM active, request owner change\n");
3547 
3548 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3549 
3550 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3551 			DELAY(1000);
3552 			bios_sem = pci_read_config(self, eecp +
3553 			    EHCI_LEGSUP_BIOS_SEM, 1);
3554 		}
3555 
3556 		if (bios_sem != 0) {
3557 			if (bootverbose)
3558 				printf("ehci early: "
3559 				    "SMM does not respond\n");
3560 		}
3561 		/* Disable interrupts */
3562 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3563 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3564 	}
3565 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3566 }
3567 
3568 /* Perform early XHCI takeover from SMM. */
3569 static void
3570 xhci_early_takeover(device_t self)
3571 {
3572 	struct resource *res;
3573 	uint32_t cparams;
3574 	uint32_t eec;
3575 	uint8_t eecp;
3576 	uint8_t bios_sem;
3577 	uint8_t offs;
3578 	int rid;
3579 	int i;
3580 
3581 	rid = PCIR_BAR(0);
3582 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3583 	if (res == NULL)
3584 		return;
3585 
3586 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3587 
3588 	eec = -1;
3589 
3590 	/* Synchronise with the BIOS if it owns the controller. */
3591 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3592 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3593 		eec = bus_read_4(res, eecp);
3594 
3595 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3596 			continue;
3597 
3598 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3599 		if (bios_sem == 0)
3600 			continue;
3601 
3602 		if (bootverbose)
3603 			printf("xhci early: "
3604 			    "SMM active, request owner change\n");
3605 
3606 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3607 
3608 		/* wait a maximum of 5 second */
3609 
3610 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3611 			DELAY(1000);
3612 			bios_sem = bus_read_1(res, eecp +
3613 			    XHCI_XECP_BIOS_SEM);
3614 		}
3615 
3616 		if (bios_sem != 0) {
3617 			if (bootverbose)
3618 				printf("xhci early: "
3619 				    "SMM does not respond\n");
3620 		}
3621 
3622 		/* Disable interrupts */
3623 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3624 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3625 		bus_read_4(res, offs + XHCI_USBSTS);
3626 	}
3627 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3628 }
3629 
3630 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3631 static void
3632 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3633     struct resource_list *rl)
3634 {
3635 	struct resource *res;
3636 	char *cp;
3637 	rman_res_t start, end, count;
3638 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3639 
3640 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3641 	case PCIM_HDRTYPE_BRIDGE:
3642 		sec_reg = PCIR_SECBUS_1;
3643 		sub_reg = PCIR_SUBBUS_1;
3644 		break;
3645 	case PCIM_HDRTYPE_CARDBUS:
3646 		sec_reg = PCIR_SECBUS_2;
3647 		sub_reg = PCIR_SUBBUS_2;
3648 		break;
3649 	default:
3650 		return;
3651 	}
3652 
3653 	/*
3654 	 * If the existing bus range is valid, attempt to reserve it
3655 	 * from our parent.  If this fails for any reason, clear the
3656 	 * secbus and subbus registers.
3657 	 *
3658 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3659 	 * This would at least preserve the existing sec_bus if it is
3660 	 * valid.
3661 	 */
3662 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3663 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3664 
3665 	/* Quirk handling. */
3666 	switch (pci_get_devid(dev)) {
3667 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3668 		sup_bus = pci_read_config(dev, 0x41, 1);
3669 		if (sup_bus != 0xff) {
3670 			sec_bus = sup_bus + 1;
3671 			sub_bus = sup_bus + 1;
3672 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3673 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3674 		}
3675 		break;
3676 
3677 	case 0x00dd10de:
3678 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3679 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3680 			break;
3681 		if (strncmp(cp, "Compal", 6) != 0) {
3682 			freeenv(cp);
3683 			break;
3684 		}
3685 		freeenv(cp);
3686 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3687 			break;
3688 		if (strncmp(cp, "08A0", 4) != 0) {
3689 			freeenv(cp);
3690 			break;
3691 		}
3692 		freeenv(cp);
3693 		if (sub_bus < 0xa) {
3694 			sub_bus = 0xa;
3695 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3696 		}
3697 		break;
3698 	}
3699 
3700 	if (bootverbose)
3701 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3702 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3703 		start = sec_bus;
3704 		end = sub_bus;
3705 		count = end - start + 1;
3706 
3707 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3708 
3709 		/*
3710 		 * If requested, clear secondary bus registers in
3711 		 * bridge devices to force a complete renumbering
3712 		 * rather than reserving the existing range.  However,
3713 		 * preserve the existing size.
3714 		 */
3715 		if (pci_clear_buses)
3716 			goto clear;
3717 
3718 		rid = 0;
3719 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3720 		    start, end, count, 0);
3721 		if (res != NULL)
3722 			return;
3723 
3724 		if (bootverbose)
3725 			device_printf(bus,
3726 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3727 			    pci_get_domain(dev), pci_get_bus(dev),
3728 			    pci_get_slot(dev), pci_get_function(dev));
3729 	}
3730 
3731 clear:
3732 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3733 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3734 }
3735 
3736 static struct resource *
3737 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3738     rman_res_t end, rman_res_t count, u_int flags)
3739 {
3740 	struct pci_devinfo *dinfo;
3741 	pcicfgregs *cfg;
3742 	struct resource_list *rl;
3743 	struct resource *res;
3744 	int sec_reg, sub_reg;
3745 
3746 	dinfo = device_get_ivars(child);
3747 	cfg = &dinfo->cfg;
3748 	rl = &dinfo->resources;
3749 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3750 	case PCIM_HDRTYPE_BRIDGE:
3751 		sec_reg = PCIR_SECBUS_1;
3752 		sub_reg = PCIR_SUBBUS_1;
3753 		break;
3754 	case PCIM_HDRTYPE_CARDBUS:
3755 		sec_reg = PCIR_SECBUS_2;
3756 		sub_reg = PCIR_SUBBUS_2;
3757 		break;
3758 	default:
3759 		return (NULL);
3760 	}
3761 
3762 	if (*rid != 0)
3763 		return (NULL);
3764 
3765 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3766 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3767 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3768 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3769 		    start, end, count, flags & ~RF_ACTIVE);
3770 		if (res == NULL) {
3771 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3772 			device_printf(child, "allocating %ju bus%s failed\n",
3773 			    count, count == 1 ? "" : "es");
3774 			return (NULL);
3775 		}
3776 		if (bootverbose)
3777 			device_printf(child,
3778 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3779 			    count == 1 ? "" : "es", rman_get_start(res));
3780 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3781 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3782 	}
3783 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3784 	    end, count, flags));
3785 }
3786 #endif
3787 
3788 static int
3789 pci_ea_bei_to_rid(device_t dev, int bei)
3790 {
3791 #ifdef PCI_IOV
3792 	struct pci_devinfo *dinfo;
3793 	int iov_pos;
3794 	struct pcicfg_iov *iov;
3795 
3796 	dinfo = device_get_ivars(dev);
3797 	iov = dinfo->cfg.iov;
3798 	if (iov != NULL)
3799 		iov_pos = iov->iov_pos;
3800 	else
3801 		iov_pos = 0;
3802 #endif
3803 
3804 	/* Check if matches BAR */
3805 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3806 	    (bei <= PCIM_EA_BEI_BAR_5))
3807 		return (PCIR_BAR(bei));
3808 
3809 	/* Check ROM */
3810 	if (bei == PCIM_EA_BEI_ROM)
3811 		return (PCIR_BIOS);
3812 
3813 #ifdef PCI_IOV
3814 	/* Check if matches VF_BAR */
3815 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3816 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3817 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3818 		    iov_pos);
3819 #endif
3820 
3821 	return (-1);
3822 }
3823 
3824 int
3825 pci_ea_is_enabled(device_t dev, int rid)
3826 {
3827 	struct pci_ea_entry *ea;
3828 	struct pci_devinfo *dinfo;
3829 
3830 	dinfo = device_get_ivars(dev);
3831 
3832 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3833 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3834 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3835 	}
3836 
3837 	return (0);
3838 }
3839 
3840 void
3841 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3842 {
3843 	struct pci_ea_entry *ea;
3844 	struct pci_devinfo *dinfo;
3845 	pci_addr_t start, end, count;
3846 	struct resource_list *rl;
3847 	int type, flags, rid;
3848 	struct resource *res;
3849 	uint32_t tmp;
3850 #ifdef PCI_IOV
3851 	struct pcicfg_iov *iov;
3852 #endif
3853 
3854 	dinfo = device_get_ivars(dev);
3855 	rl = &dinfo->resources;
3856 	flags = 0;
3857 
3858 #ifdef PCI_IOV
3859 	iov = dinfo->cfg.iov;
3860 #endif
3861 
3862 	if (dinfo->cfg.ea.ea_location == 0)
3863 		return;
3864 
3865 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3866 		/*
3867 		 * TODO: Ignore EA-BAR if is not enabled.
3868 		 *   Currently the EA implementation supports
3869 		 *   only situation, where EA structure contains
3870 		 *   predefined entries. In case they are not enabled
3871 		 *   leave them unallocated and proceed with
3872 		 *   a legacy-BAR mechanism.
3873 		 */
3874 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3875 			continue;
3876 
3877 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3878 		case PCIM_EA_P_MEM_PREFETCH:
3879 		case PCIM_EA_P_VF_MEM_PREFETCH:
3880 			flags = RF_PREFETCHABLE;
3881 			/* FALLTHROUGH */
3882 		case PCIM_EA_P_VF_MEM:
3883 		case PCIM_EA_P_MEM:
3884 			type = SYS_RES_MEMORY;
3885 			break;
3886 		case PCIM_EA_P_IO:
3887 			type = SYS_RES_IOPORT;
3888 			break;
3889 		default:
3890 			continue;
3891 		}
3892 
3893 		if (alloc_iov != 0) {
3894 #ifdef PCI_IOV
3895 			/* Allocating IOV, confirm BEI matches */
3896 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3897 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3898 				continue;
3899 #else
3900 			continue;
3901 #endif
3902 		} else {
3903 			/* Allocating BAR, confirm BEI matches */
3904 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3905 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3906 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3907 				continue;
3908 		}
3909 
3910 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3911 		if (rid < 0)
3912 			continue;
3913 
3914 		/* Skip resources already allocated by EA */
3915 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3916 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3917 			continue;
3918 
3919 		start = ea->eae_base;
3920 		count = ea->eae_max_offset + 1;
3921 #ifdef PCI_IOV
3922 		if (iov != NULL)
3923 			count = count * iov->iov_num_vfs;
3924 #endif
3925 		end = start + count - 1;
3926 		if (count == 0)
3927 			continue;
3928 
3929 		resource_list_add(rl, type, rid, start, end, count);
3930 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3931 		    flags);
3932 		if (res == NULL) {
3933 			resource_list_delete(rl, type, rid);
3934 
3935 			/*
3936 			 * Failed to allocate using EA, disable entry.
3937 			 * Another attempt to allocation will be performed
3938 			 * further, but this time using legacy BAR registers
3939 			 */
3940 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3941 			tmp &= ~PCIM_EA_ENABLE;
3942 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3943 
3944 			/*
3945 			 * Disabling entry might fail in case it is hardwired.
3946 			 * Read flags again to match current status.
3947 			 */
3948 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3949 
3950 			continue;
3951 		}
3952 
3953 		/* As per specification, fill BAR with zeros */
3954 		pci_write_config(dev, rid, 0, 4);
3955 	}
3956 }
3957 
3958 void
3959 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3960 {
3961 	struct pci_devinfo *dinfo;
3962 	pcicfgregs *cfg;
3963 	struct resource_list *rl;
3964 	const struct pci_quirk *q;
3965 	uint32_t devid;
3966 	int i;
3967 
3968 	dinfo = device_get_ivars(dev);
3969 	cfg = &dinfo->cfg;
3970 	rl = &dinfo->resources;
3971 	devid = (cfg->device << 16) | cfg->vendor;
3972 
3973 	/* Allocate resources using Enhanced Allocation */
3974 	pci_add_resources_ea(bus, dev, 0);
3975 
3976 	/* ATA devices needs special map treatment */
3977 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3978 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3979 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3980 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3981 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3982 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3983 	else
3984 		for (i = 0; i < cfg->nummaps;) {
3985 			/* Skip resources already managed by EA */
3986 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3987 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3988 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3989 				i++;
3990 				continue;
3991 			}
3992 
3993 			/*
3994 			 * Skip quirked resources.
3995 			 */
3996 			for (q = &pci_quirks[0]; q->devid != 0; q++)
3997 				if (q->devid == devid &&
3998 				    q->type == PCI_QUIRK_UNMAP_REG &&
3999 				    q->arg1 == PCIR_BAR(i))
4000 					break;
4001 			if (q->devid != 0) {
4002 				i++;
4003 				continue;
4004 			}
4005 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4006 			    prefetchmask & (1 << i));
4007 		}
4008 
4009 	/*
4010 	 * Add additional, quirked resources.
4011 	 */
4012 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4013 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4014 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4015 
4016 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4017 #ifdef __PCI_REROUTE_INTERRUPT
4018 		/*
4019 		 * Try to re-route interrupts. Sometimes the BIOS or
4020 		 * firmware may leave bogus values in these registers.
4021 		 * If the re-route fails, then just stick with what we
4022 		 * have.
4023 		 */
4024 		pci_assign_interrupt(bus, dev, 1);
4025 #else
4026 		pci_assign_interrupt(bus, dev, 0);
4027 #endif
4028 	}
4029 
4030 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4031 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4032 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4033 			xhci_early_takeover(dev);
4034 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4035 			ehci_early_takeover(dev);
4036 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4037 			ohci_early_takeover(dev);
4038 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4039 			uhci_early_takeover(dev);
4040 	}
4041 
4042 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4043 	/*
4044 	 * Reserve resources for secondary bus ranges behind bridge
4045 	 * devices.
4046 	 */
4047 	pci_reserve_secbus(bus, dev, cfg, rl);
4048 #endif
4049 }
4050 
4051 static struct pci_devinfo *
4052 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4053     int slot, int func)
4054 {
4055 	struct pci_devinfo *dinfo;
4056 
4057 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4058 	if (dinfo != NULL)
4059 		pci_add_child(dev, dinfo);
4060 
4061 	return (dinfo);
4062 }
4063 
4064 void
4065 pci_add_children(device_t dev, int domain, int busno)
4066 {
4067 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4068 	device_t pcib = device_get_parent(dev);
4069 	struct pci_devinfo *dinfo;
4070 	int maxslots;
4071 	int s, f, pcifunchigh;
4072 	uint8_t hdrtype;
4073 	int first_func;
4074 
4075 	/*
4076 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4077 	 * enable ARI.  We must enable ARI before detecting the rest of the
4078 	 * functions on this bus as ARI changes the set of slots and functions
4079 	 * that are legal on this bus.
4080 	 */
4081 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4082 	if (dinfo != NULL && pci_enable_ari)
4083 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4084 
4085 	/*
4086 	 * Start looking for new devices on slot 0 at function 1 because we
4087 	 * just identified the device at slot 0, function 0.
4088 	 */
4089 	first_func = 1;
4090 
4091 	maxslots = PCIB_MAXSLOTS(pcib);
4092 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4093 		pcifunchigh = 0;
4094 		f = 0;
4095 		DELAY(1);
4096 		hdrtype = REG(PCIR_HDRTYPE, 1);
4097 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4098 			continue;
4099 		if (hdrtype & PCIM_MFDEV)
4100 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4101 		for (f = first_func; f <= pcifunchigh; f++)
4102 			pci_identify_function(pcib, dev, domain, busno, s, f);
4103 	}
4104 #undef REG
4105 }
4106 
4107 int
4108 pci_rescan_method(device_t dev)
4109 {
4110 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4111 	device_t pcib = device_get_parent(dev);
4112 	device_t child, *devlist, *unchanged;
4113 	int devcount, error, i, j, maxslots, oldcount;
4114 	int busno, domain, s, f, pcifunchigh;
4115 	uint8_t hdrtype;
4116 
4117 	/* No need to check for ARI on a rescan. */
4118 	error = device_get_children(dev, &devlist, &devcount);
4119 	if (error)
4120 		return (error);
4121 	if (devcount != 0) {
4122 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4123 		    M_NOWAIT | M_ZERO);
4124 		if (unchanged == NULL) {
4125 			free(devlist, M_TEMP);
4126 			return (ENOMEM);
4127 		}
4128 	} else
4129 		unchanged = NULL;
4130 
4131 	domain = pcib_get_domain(dev);
4132 	busno = pcib_get_bus(dev);
4133 	maxslots = PCIB_MAXSLOTS(pcib);
4134 	for (s = 0; s <= maxslots; s++) {
4135 		/* If function 0 is not present, skip to the next slot. */
4136 		f = 0;
4137 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4138 			continue;
4139 		pcifunchigh = 0;
4140 		hdrtype = REG(PCIR_HDRTYPE, 1);
4141 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4142 			continue;
4143 		if (hdrtype & PCIM_MFDEV)
4144 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4145 		for (f = 0; f <= pcifunchigh; f++) {
4146 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4147 				continue;
4148 
4149 			/*
4150 			 * Found a valid function.  Check if a
4151 			 * device_t for this device already exists.
4152 			 */
4153 			for (i = 0; i < devcount; i++) {
4154 				child = devlist[i];
4155 				if (child == NULL)
4156 					continue;
4157 				if (pci_get_slot(child) == s &&
4158 				    pci_get_function(child) == f) {
4159 					unchanged[i] = child;
4160 					goto next_func;
4161 				}
4162 			}
4163 
4164 			pci_identify_function(pcib, dev, domain, busno, s, f);
4165 		next_func:;
4166 		}
4167 	}
4168 
4169 	/* Remove devices that are no longer present. */
4170 	for (i = 0; i < devcount; i++) {
4171 		if (unchanged[i] != NULL)
4172 			continue;
4173 		device_delete_child(dev, devlist[i]);
4174 	}
4175 
4176 	free(devlist, M_TEMP);
4177 	oldcount = devcount;
4178 
4179 	/* Try to attach the devices just added. */
4180 	error = device_get_children(dev, &devlist, &devcount);
4181 	if (error) {
4182 		free(unchanged, M_TEMP);
4183 		return (error);
4184 	}
4185 
4186 	for (i = 0; i < devcount; i++) {
4187 		for (j = 0; j < oldcount; j++) {
4188 			if (devlist[i] == unchanged[j])
4189 				goto next_device;
4190 		}
4191 
4192 		device_probe_and_attach(devlist[i]);
4193 	next_device:;
4194 	}
4195 
4196 	free(unchanged, M_TEMP);
4197 	free(devlist, M_TEMP);
4198 	return (0);
4199 #undef REG
4200 }
4201 
4202 #ifdef PCI_IOV
4203 device_t
4204 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4205     uint16_t did)
4206 {
4207 	struct pci_devinfo *vf_dinfo;
4208 	device_t pcib;
4209 	int busno, slot, func;
4210 
4211 	pcib = device_get_parent(bus);
4212 
4213 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4214 
4215 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4216 	    slot, func, vid, did);
4217 
4218 	vf_dinfo->cfg.flags |= PCICFG_VF;
4219 	pci_add_child(bus, vf_dinfo);
4220 
4221 	return (vf_dinfo->cfg.dev);
4222 }
4223 
4224 device_t
4225 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4226     uint16_t vid, uint16_t did)
4227 {
4228 
4229 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4230 }
4231 #endif
4232 
4233 static void
4234 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4235 {
4236 	int aer;
4237 	uint32_t r;
4238 	uint16_t r2;
4239 
4240 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4241 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4242 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4243 		    PCIER_ROOT_CTL, 2);
4244 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4245 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4246 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4247 		    PCIER_ROOT_CTL, r2, 2);
4248 	}
4249 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4250 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4251 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4252 		if (r != 0 && bootverbose) {
4253 			pci_printf(&dinfo->cfg,
4254 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4255 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4256 			    4));
4257 		}
4258 
4259 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4260 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4261 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4262 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4263 		    PCIM_AER_UC_POISONED_TLP |
4264 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4265 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4266 		    PCIM_AER_UC_COMPLETER_ABORT |
4267 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4268 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4269 		    PCIM_AER_UC_MALFORMED_TLP |
4270 		    PCIM_AER_UC_ECRC_ERROR |
4271 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4272 		    PCIM_AER_UC_ACS_VIOLATION |
4273 		    PCIM_AER_UC_INTERNAL_ERROR |
4274 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4275 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4276 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4277 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4278 
4279 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4280 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4281 		if (r != 0 && bootverbose) {
4282 			pci_printf(&dinfo->cfg,
4283 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4284 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4285 			    4));
4286 		}
4287 
4288 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4289 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4290 		    PCIM_AER_COR_BAD_TLP |
4291 		    PCIM_AER_COR_BAD_DLLP |
4292 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4293 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4294 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4295 		    PCIM_AER_COR_INTERNAL_ERROR |
4296 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4297 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4298 
4299 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4300 		    PCIER_DEVICE_CTL, 2);
4301 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4302 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4303 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4304 		    PCIER_DEVICE_CTL, r, 2);
4305 	}
4306 }
4307 
4308 void
4309 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4310 {
4311 	device_t dev;
4312 
4313 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4314 	device_set_ivars(dev, dinfo);
4315 	resource_list_init(&dinfo->resources);
4316 	pci_cfg_save(dev, dinfo, 0);
4317 	pci_cfg_restore(dev, dinfo);
4318 	pci_print_verbose(dinfo);
4319 	pci_add_resources(bus, dev, 0, 0);
4320 	pci_child_added(dinfo->cfg.dev);
4321 
4322 	if (pci_clear_aer_on_attach)
4323 		pci_add_child_clear_aer(dev, dinfo);
4324 
4325 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4326 }
4327 
4328 void
4329 pci_child_added_method(device_t dev, device_t child)
4330 {
4331 
4332 }
4333 
4334 static int
4335 pci_probe(device_t dev)
4336 {
4337 
4338 	device_set_desc(dev, "PCI bus");
4339 
4340 	/* Allow other subclasses to override this driver. */
4341 	return (BUS_PROBE_GENERIC);
4342 }
4343 
4344 int
4345 pci_attach_common(device_t dev)
4346 {
4347 	struct pci_softc *sc;
4348 	int busno, domain;
4349 #ifdef PCI_RES_BUS
4350 	int rid;
4351 #endif
4352 
4353 	sc = device_get_softc(dev);
4354 	domain = pcib_get_domain(dev);
4355 	busno = pcib_get_bus(dev);
4356 #ifdef PCI_RES_BUS
4357 	rid = 0;
4358 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4359 	    1, 0);
4360 	if (sc->sc_bus == NULL) {
4361 		device_printf(dev, "failed to allocate bus number\n");
4362 		return (ENXIO);
4363 	}
4364 #endif
4365 	if (bootverbose)
4366 		device_printf(dev, "domain=%d, physical bus=%d\n",
4367 		    domain, busno);
4368 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4369 	return (0);
4370 }
4371 
4372 int
4373 pci_attach(device_t dev)
4374 {
4375 	int busno, domain, error;
4376 
4377 	error = pci_attach_common(dev);
4378 	if (error)
4379 		return (error);
4380 
4381 	/*
4382 	 * Since there can be multiple independently numbered PCI
4383 	 * buses on systems with multiple PCI domains, we can't use
4384 	 * the unit number to decide which bus we are probing. We ask
4385 	 * the parent pcib what our domain and bus numbers are.
4386 	 */
4387 	domain = pcib_get_domain(dev);
4388 	busno = pcib_get_bus(dev);
4389 	pci_add_children(dev, domain, busno);
4390 	return (bus_generic_attach(dev));
4391 }
4392 
4393 int
4394 pci_detach(device_t dev)
4395 {
4396 #ifdef PCI_RES_BUS
4397 	struct pci_softc *sc;
4398 #endif
4399 	int error;
4400 
4401 	error = bus_generic_detach(dev);
4402 	if (error)
4403 		return (error);
4404 #ifdef PCI_RES_BUS
4405 	sc = device_get_softc(dev);
4406 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4407 	if (error)
4408 		return (error);
4409 #endif
4410 	return (device_delete_children(dev));
4411 }
4412 
4413 static void
4414 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4415 {
4416 	int line, unit;
4417 	const char *at;
4418 	char me1[24], me2[32];
4419 	uint8_t b, s, f;
4420 	uint32_t d;
4421 
4422 	d = pci_get_domain(child);
4423 	b = pci_get_bus(child);
4424 	s = pci_get_slot(child);
4425 	f = pci_get_function(child);
4426 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4427 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4428 	line = 0;
4429 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4430 		resource_string_value(name, unit, "at", &at);
4431 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4432 			continue; /* No match, try next candidate */
4433 		*unitp = unit;
4434 		return;
4435 	}
4436 }
4437 
4438 static void
4439 pci_set_power_child(device_t dev, device_t child, int state)
4440 {
4441 	device_t pcib;
4442 	int dstate;
4443 
4444 	/*
4445 	 * Set the device to the given state.  If the firmware suggests
4446 	 * a different power state, use it instead.  If power management
4447 	 * is not present, the firmware is responsible for managing
4448 	 * device power.  Skip children who aren't attached since they
4449 	 * are handled separately.
4450 	 */
4451 	pcib = device_get_parent(dev);
4452 	dstate = state;
4453 	if (device_is_attached(child) &&
4454 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4455 		pci_set_powerstate(child, dstate);
4456 }
4457 
4458 int
4459 pci_suspend_child(device_t dev, device_t child)
4460 {
4461 	struct pci_devinfo *dinfo;
4462 	struct resource_list_entry *rle;
4463 	int error;
4464 
4465 	dinfo = device_get_ivars(child);
4466 
4467 	/*
4468 	 * Save the PCI configuration space for the child and set the
4469 	 * device in the appropriate power state for this sleep state.
4470 	 */
4471 	pci_cfg_save(child, dinfo, 0);
4472 
4473 	/* Suspend devices before potentially powering them down. */
4474 	error = bus_generic_suspend_child(dev, child);
4475 
4476 	if (error)
4477 		return (error);
4478 
4479 	if (pci_do_power_suspend) {
4480 		/*
4481 		 * Make sure this device's interrupt handler is not invoked
4482 		 * in the case the device uses a shared interrupt that can
4483 		 * be raised by some other device.
4484 		 * This is applicable only to regular (legacy) PCI interrupts
4485 		 * as MSI/MSI-X interrupts are never shared.
4486 		 */
4487 		rle = resource_list_find(&dinfo->resources,
4488 		    SYS_RES_IRQ, 0);
4489 		if (rle != NULL && rle->res != NULL)
4490 			(void)bus_suspend_intr(child, rle->res);
4491 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4492 	}
4493 
4494 	return (0);
4495 }
4496 
4497 int
4498 pci_resume_child(device_t dev, device_t child)
4499 {
4500 	struct pci_devinfo *dinfo;
4501 	struct resource_list_entry *rle;
4502 
4503 	if (pci_do_power_resume)
4504 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4505 
4506 	dinfo = device_get_ivars(child);
4507 	pci_cfg_restore(child, dinfo);
4508 	if (!device_is_attached(child))
4509 		pci_cfg_save(child, dinfo, 1);
4510 
4511 	bus_generic_resume_child(dev, child);
4512 
4513 	/*
4514 	 * Allow interrupts only after fully resuming the driver and hardware.
4515 	 */
4516 	if (pci_do_power_suspend) {
4517 		/* See pci_suspend_child for details. */
4518 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4519 		if (rle != NULL && rle->res != NULL)
4520 			(void)bus_resume_intr(child, rle->res);
4521 	}
4522 
4523 	return (0);
4524 }
4525 
4526 int
4527 pci_resume(device_t dev)
4528 {
4529 	device_t child, *devlist;
4530 	int error, i, numdevs;
4531 
4532 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4533 		return (error);
4534 
4535 	/*
4536 	 * Resume critical devices first, then everything else later.
4537 	 */
4538 	for (i = 0; i < numdevs; i++) {
4539 		child = devlist[i];
4540 		switch (pci_get_class(child)) {
4541 		case PCIC_DISPLAY:
4542 		case PCIC_MEMORY:
4543 		case PCIC_BRIDGE:
4544 		case PCIC_BASEPERIPH:
4545 			BUS_RESUME_CHILD(dev, child);
4546 			break;
4547 		}
4548 	}
4549 	for (i = 0; i < numdevs; i++) {
4550 		child = devlist[i];
4551 		switch (pci_get_class(child)) {
4552 		case PCIC_DISPLAY:
4553 		case PCIC_MEMORY:
4554 		case PCIC_BRIDGE:
4555 		case PCIC_BASEPERIPH:
4556 			break;
4557 		default:
4558 			BUS_RESUME_CHILD(dev, child);
4559 		}
4560 	}
4561 	free(devlist, M_TEMP);
4562 	return (0);
4563 }
4564 
4565 static void
4566 pci_load_vendor_data(void)
4567 {
4568 	caddr_t data;
4569 	void *ptr;
4570 	size_t sz;
4571 
4572 	data = preload_search_by_type("pci_vendor_data");
4573 	if (data != NULL) {
4574 		ptr = preload_fetch_addr(data);
4575 		sz = preload_fetch_size(data);
4576 		if (ptr != NULL && sz != 0) {
4577 			pci_vendordata = ptr;
4578 			pci_vendordata_size = sz;
4579 			/* terminate the database */
4580 			pci_vendordata[pci_vendordata_size] = '\n';
4581 		}
4582 	}
4583 }
4584 
4585 void
4586 pci_driver_added(device_t dev, driver_t *driver)
4587 {
4588 	int numdevs;
4589 	device_t *devlist;
4590 	device_t child;
4591 	struct pci_devinfo *dinfo;
4592 	int i;
4593 
4594 	if (bootverbose)
4595 		device_printf(dev, "driver added\n");
4596 	DEVICE_IDENTIFY(driver, dev);
4597 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4598 		return;
4599 	for (i = 0; i < numdevs; i++) {
4600 		child = devlist[i];
4601 		if (device_get_state(child) != DS_NOTPRESENT)
4602 			continue;
4603 		dinfo = device_get_ivars(child);
4604 		pci_print_verbose(dinfo);
4605 		if (bootverbose)
4606 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4607 		pci_cfg_restore(child, dinfo);
4608 		if (device_probe_and_attach(child) != 0)
4609 			pci_child_detached(dev, child);
4610 	}
4611 	free(devlist, M_TEMP);
4612 }
4613 
4614 int
4615 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4616     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4617 {
4618 	struct pci_devinfo *dinfo;
4619 	struct msix_table_entry *mte;
4620 	struct msix_vector *mv;
4621 	uint64_t addr;
4622 	uint32_t data;
4623 	void *cookie;
4624 	int error, rid;
4625 
4626 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4627 	    arg, &cookie);
4628 	if (error)
4629 		return (error);
4630 
4631 	/* If this is not a direct child, just bail out. */
4632 	if (device_get_parent(child) != dev) {
4633 		*cookiep = cookie;
4634 		return(0);
4635 	}
4636 
4637 	rid = rman_get_rid(irq);
4638 	if (rid == 0) {
4639 		/* Make sure that INTx is enabled */
4640 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4641 	} else {
4642 		/*
4643 		 * Check to see if the interrupt is MSI or MSI-X.
4644 		 * Ask our parent to map the MSI and give
4645 		 * us the address and data register values.
4646 		 * If we fail for some reason, teardown the
4647 		 * interrupt handler.
4648 		 */
4649 		dinfo = device_get_ivars(child);
4650 		if (dinfo->cfg.msi.msi_alloc > 0) {
4651 			if (dinfo->cfg.msi.msi_addr == 0) {
4652 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4653 			    ("MSI has handlers, but vectors not mapped"));
4654 				error = PCIB_MAP_MSI(device_get_parent(dev),
4655 				    child, rman_get_start(irq), &addr, &data);
4656 				if (error)
4657 					goto bad;
4658 				dinfo->cfg.msi.msi_addr = addr;
4659 				dinfo->cfg.msi.msi_data = data;
4660 			}
4661 			if (dinfo->cfg.msi.msi_handlers == 0)
4662 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4663 				    dinfo->cfg.msi.msi_data);
4664 			dinfo->cfg.msi.msi_handlers++;
4665 		} else {
4666 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4667 			    ("No MSI or MSI-X interrupts allocated"));
4668 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4669 			    ("MSI-X index too high"));
4670 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4671 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4672 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4673 			KASSERT(mv->mv_irq == rman_get_start(irq),
4674 			    ("IRQ mismatch"));
4675 			if (mv->mv_address == 0) {
4676 				KASSERT(mte->mte_handlers == 0,
4677 		    ("MSI-X table entry has handlers, but vector not mapped"));
4678 				error = PCIB_MAP_MSI(device_get_parent(dev),
4679 				    child, rman_get_start(irq), &addr, &data);
4680 				if (error)
4681 					goto bad;
4682 				mv->mv_address = addr;
4683 				mv->mv_data = data;
4684 			}
4685 
4686 			/*
4687 			 * The MSIX table entry must be made valid by
4688 			 * incrementing the mte_handlers before
4689 			 * calling pci_enable_msix() and
4690 			 * pci_resume_msix(). Else the MSIX rewrite
4691 			 * table quirk will not work as expected.
4692 			 */
4693 			mte->mte_handlers++;
4694 			if (mte->mte_handlers == 1) {
4695 				pci_enable_msix(child, rid - 1, mv->mv_address,
4696 				    mv->mv_data);
4697 				pci_unmask_msix(child, rid - 1);
4698 			}
4699 		}
4700 
4701 		/*
4702 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4703 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4704 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4705 		 */
4706 		if (!pci_has_quirk(pci_get_devid(child),
4707 		    PCI_QUIRK_MSI_INTX_BUG))
4708 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4709 		else
4710 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4711 	bad:
4712 		if (error) {
4713 			(void)bus_generic_teardown_intr(dev, child, irq,
4714 			    cookie);
4715 			return (error);
4716 		}
4717 	}
4718 	*cookiep = cookie;
4719 	return (0);
4720 }
4721 
4722 int
4723 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4724     void *cookie)
4725 {
4726 	struct msix_table_entry *mte;
4727 	struct resource_list_entry *rle;
4728 	struct pci_devinfo *dinfo;
4729 	int error, rid;
4730 
4731 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4732 		return (EINVAL);
4733 
4734 	/* If this isn't a direct child, just bail out */
4735 	if (device_get_parent(child) != dev)
4736 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4737 
4738 	rid = rman_get_rid(irq);
4739 	if (rid == 0) {
4740 		/* Mask INTx */
4741 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4742 	} else {
4743 		/*
4744 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4745 		 * decrement the appropriate handlers count and mask the
4746 		 * MSI-X message, or disable MSI messages if the count
4747 		 * drops to 0.
4748 		 */
4749 		dinfo = device_get_ivars(child);
4750 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4751 		if (rle->res != irq)
4752 			return (EINVAL);
4753 		if (dinfo->cfg.msi.msi_alloc > 0) {
4754 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4755 			    ("MSI-X index too high"));
4756 			if (dinfo->cfg.msi.msi_handlers == 0)
4757 				return (EINVAL);
4758 			dinfo->cfg.msi.msi_handlers--;
4759 			if (dinfo->cfg.msi.msi_handlers == 0)
4760 				pci_disable_msi(child);
4761 		} else {
4762 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4763 			    ("No MSI or MSI-X interrupts allocated"));
4764 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4765 			    ("MSI-X index too high"));
4766 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4767 			if (mte->mte_handlers == 0)
4768 				return (EINVAL);
4769 			mte->mte_handlers--;
4770 			if (mte->mte_handlers == 0)
4771 				pci_mask_msix(child, rid - 1);
4772 		}
4773 	}
4774 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4775 	if (rid > 0)
4776 		KASSERT(error == 0,
4777 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4778 	return (error);
4779 }
4780 
4781 int
4782 pci_print_child(device_t dev, device_t child)
4783 {
4784 	struct pci_devinfo *dinfo;
4785 	struct resource_list *rl;
4786 	int retval = 0;
4787 
4788 	dinfo = device_get_ivars(child);
4789 	rl = &dinfo->resources;
4790 
4791 	retval += bus_print_child_header(dev, child);
4792 
4793 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4794 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4795 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4796 	if (device_get_flags(dev))
4797 		retval += printf(" flags %#x", device_get_flags(dev));
4798 
4799 	retval += printf(" at device %d.%d", pci_get_slot(child),
4800 	    pci_get_function(child));
4801 
4802 	retval += bus_print_child_domain(dev, child);
4803 	retval += bus_print_child_footer(dev, child);
4804 
4805 	return (retval);
4806 }
4807 
4808 static const struct
4809 {
4810 	int		class;
4811 	int		subclass;
4812 	int		report; /* 0 = bootverbose, 1 = always */
4813 	const char	*desc;
4814 } pci_nomatch_tab[] = {
4815 	{PCIC_OLD,		-1,			1, "old"},
4816 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4817 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4818 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4819 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4820 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4821 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4822 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4823 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4824 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4825 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4826 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4827 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4828 	{PCIC_NETWORK,		-1,			1, "network"},
4829 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4830 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4831 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4832 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4833 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4834 	{PCIC_DISPLAY,		-1,			1, "display"},
4835 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4836 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4837 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4838 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4839 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4840 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4841 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4842 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4843 	{PCIC_MEMORY,		-1,			1, "memory"},
4844 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4845 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4846 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4847 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4848 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4849 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4850 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4851 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4852 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4853 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4854 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4855 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4856 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4857 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4858 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4859 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4860 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4861 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4862 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4863 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4864 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4865 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4866 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4867 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4868 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4869 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4870 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4871 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4872 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4873 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4874 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4875 	{PCIC_DOCKING,		-1,			1, "docking station"},
4876 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4877 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4878 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4879 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4880 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4881 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4882 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4883 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4884 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4885 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4886 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4887 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4888 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4889 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4890 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4891 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4892 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4893 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4894 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4895 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4896 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4897 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4898 	{PCIC_DASP,		-1,			0, "dasp"},
4899 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4900 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4901 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4902 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4903 	{0, 0, 0,		NULL}
4904 };
4905 
4906 void
4907 pci_probe_nomatch(device_t dev, device_t child)
4908 {
4909 	int i, report;
4910 	const char *cp, *scp;
4911 	char *device;
4912 
4913 	/*
4914 	 * Look for a listing for this device in a loaded device database.
4915 	 */
4916 	report = 1;
4917 	if ((device = pci_describe_device(child)) != NULL) {
4918 		device_printf(dev, "<%s>", device);
4919 		free(device, M_DEVBUF);
4920 	} else {
4921 		/*
4922 		 * Scan the class/subclass descriptions for a general
4923 		 * description.
4924 		 */
4925 		cp = "unknown";
4926 		scp = NULL;
4927 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4928 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4929 				if (pci_nomatch_tab[i].subclass == -1) {
4930 					cp = pci_nomatch_tab[i].desc;
4931 					report = pci_nomatch_tab[i].report;
4932 				} else if (pci_nomatch_tab[i].subclass ==
4933 				    pci_get_subclass(child)) {
4934 					scp = pci_nomatch_tab[i].desc;
4935 					report = pci_nomatch_tab[i].report;
4936 				}
4937 			}
4938 		}
4939 		if (report || bootverbose) {
4940 			device_printf(dev, "<%s%s%s>",
4941 			    cp ? cp : "",
4942 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4943 			    scp ? scp : "");
4944 		}
4945 	}
4946 	if (report || bootverbose) {
4947 		printf(" at device %d.%d (no driver attached)\n",
4948 		    pci_get_slot(child), pci_get_function(child));
4949 	}
4950 	pci_cfg_save(child, device_get_ivars(child), 1);
4951 }
4952 
4953 void
4954 pci_child_detached(device_t dev, device_t child)
4955 {
4956 	struct pci_devinfo *dinfo;
4957 	struct resource_list *rl;
4958 
4959 	dinfo = device_get_ivars(child);
4960 	rl = &dinfo->resources;
4961 
4962 	/*
4963 	 * Have to deallocate IRQs before releasing any MSI messages and
4964 	 * have to release MSI messages before deallocating any memory
4965 	 * BARs.
4966 	 */
4967 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4968 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4969 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4970 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4971 		(void)pci_release_msi(child);
4972 	}
4973 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4974 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4975 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4976 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4977 #ifdef PCI_RES_BUS
4978 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4979 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4980 #endif
4981 
4982 	pci_cfg_save(child, dinfo, 1);
4983 }
4984 
4985 /*
4986  * Parse the PCI device database, if loaded, and return a pointer to a
4987  * description of the device.
4988  *
4989  * The database is flat text formatted as follows:
4990  *
4991  * Any line not in a valid format is ignored.
4992  * Lines are terminated with newline '\n' characters.
4993  *
4994  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4995  * the vendor name.
4996  *
4997  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
4998  * - devices cannot be listed without a corresponding VENDOR line.
4999  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5000  * another TAB, then the device name.
5001  */
5002 
5003 /*
5004  * Assuming (ptr) points to the beginning of a line in the database,
5005  * return the vendor or device and description of the next entry.
5006  * The value of (vendor) or (device) inappropriate for the entry type
5007  * is set to -1.  Returns nonzero at the end of the database.
5008  *
5009  * Note that this is slightly unrobust in the face of corrupt data;
5010  * we attempt to safeguard against this by spamming the end of the
5011  * database with a newline when we initialise.
5012  */
5013 static int
5014 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5015 {
5016 	char	*cp = *ptr;
5017 	int	left;
5018 
5019 	*device = -1;
5020 	*vendor = -1;
5021 	**desc = '\0';
5022 	for (;;) {
5023 		left = pci_vendordata_size - (cp - pci_vendordata);
5024 		if (left <= 0) {
5025 			*ptr = cp;
5026 			return(1);
5027 		}
5028 
5029 		/* vendor entry? */
5030 		if (*cp != '\t' &&
5031 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5032 			break;
5033 		/* device entry? */
5034 		if (*cp == '\t' &&
5035 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5036 			break;
5037 
5038 		/* skip to next line */
5039 		while (*cp != '\n' && left > 0) {
5040 			cp++;
5041 			left--;
5042 		}
5043 		if (*cp == '\n') {
5044 			cp++;
5045 			left--;
5046 		}
5047 	}
5048 	/* skip to next line */
5049 	while (*cp != '\n' && left > 0) {
5050 		cp++;
5051 		left--;
5052 	}
5053 	if (*cp == '\n' && left > 0)
5054 		cp++;
5055 	*ptr = cp;
5056 	return(0);
5057 }
5058 
5059 static char *
5060 pci_describe_device(device_t dev)
5061 {
5062 	int	vendor, device;
5063 	char	*desc, *vp, *dp, *line;
5064 
5065 	desc = vp = dp = NULL;
5066 
5067 	/*
5068 	 * If we have no vendor data, we can't do anything.
5069 	 */
5070 	if (pci_vendordata == NULL)
5071 		goto out;
5072 
5073 	/*
5074 	 * Scan the vendor data looking for this device
5075 	 */
5076 	line = pci_vendordata;
5077 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5078 		goto out;
5079 	for (;;) {
5080 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5081 			goto out;
5082 		if (vendor == pci_get_vendor(dev))
5083 			break;
5084 	}
5085 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5086 		goto out;
5087 	for (;;) {
5088 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5089 			*dp = 0;
5090 			break;
5091 		}
5092 		if (vendor != -1) {
5093 			*dp = 0;
5094 			break;
5095 		}
5096 		if (device == pci_get_device(dev))
5097 			break;
5098 	}
5099 	if (dp[0] == '\0')
5100 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5101 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5102 	    NULL)
5103 		sprintf(desc, "%s, %s", vp, dp);
5104 out:
5105 	if (vp != NULL)
5106 		free(vp, M_DEVBUF);
5107 	if (dp != NULL)
5108 		free(dp, M_DEVBUF);
5109 	return(desc);
5110 }
5111 
5112 int
5113 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5114 {
5115 	struct pci_devinfo *dinfo;
5116 	pcicfgregs *cfg;
5117 
5118 	dinfo = device_get_ivars(child);
5119 	cfg = &dinfo->cfg;
5120 
5121 	switch (which) {
5122 	case PCI_IVAR_ETHADDR:
5123 		/*
5124 		 * The generic accessor doesn't deal with failure, so
5125 		 * we set the return value, then return an error.
5126 		 */
5127 		*((uint8_t **) result) = NULL;
5128 		return (EINVAL);
5129 	case PCI_IVAR_SUBVENDOR:
5130 		*result = cfg->subvendor;
5131 		break;
5132 	case PCI_IVAR_SUBDEVICE:
5133 		*result = cfg->subdevice;
5134 		break;
5135 	case PCI_IVAR_VENDOR:
5136 		*result = cfg->vendor;
5137 		break;
5138 	case PCI_IVAR_DEVICE:
5139 		*result = cfg->device;
5140 		break;
5141 	case PCI_IVAR_DEVID:
5142 		*result = (cfg->device << 16) | cfg->vendor;
5143 		break;
5144 	case PCI_IVAR_CLASS:
5145 		*result = cfg->baseclass;
5146 		break;
5147 	case PCI_IVAR_SUBCLASS:
5148 		*result = cfg->subclass;
5149 		break;
5150 	case PCI_IVAR_PROGIF:
5151 		*result = cfg->progif;
5152 		break;
5153 	case PCI_IVAR_REVID:
5154 		*result = cfg->revid;
5155 		break;
5156 	case PCI_IVAR_INTPIN:
5157 		*result = cfg->intpin;
5158 		break;
5159 	case PCI_IVAR_IRQ:
5160 		*result = cfg->intline;
5161 		break;
5162 	case PCI_IVAR_DOMAIN:
5163 		*result = cfg->domain;
5164 		break;
5165 	case PCI_IVAR_BUS:
5166 		*result = cfg->bus;
5167 		break;
5168 	case PCI_IVAR_SLOT:
5169 		*result = cfg->slot;
5170 		break;
5171 	case PCI_IVAR_FUNCTION:
5172 		*result = cfg->func;
5173 		break;
5174 	case PCI_IVAR_CMDREG:
5175 		*result = cfg->cmdreg;
5176 		break;
5177 	case PCI_IVAR_CACHELNSZ:
5178 		*result = cfg->cachelnsz;
5179 		break;
5180 	case PCI_IVAR_MINGNT:
5181 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5182 			*result = -1;
5183 			return (EINVAL);
5184 		}
5185 		*result = cfg->mingnt;
5186 		break;
5187 	case PCI_IVAR_MAXLAT:
5188 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5189 			*result = -1;
5190 			return (EINVAL);
5191 		}
5192 		*result = cfg->maxlat;
5193 		break;
5194 	case PCI_IVAR_LATTIMER:
5195 		*result = cfg->lattimer;
5196 		break;
5197 	default:
5198 		return (ENOENT);
5199 	}
5200 	return (0);
5201 }
5202 
5203 int
5204 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5205 {
5206 	struct pci_devinfo *dinfo;
5207 
5208 	dinfo = device_get_ivars(child);
5209 
5210 	switch (which) {
5211 	case PCI_IVAR_INTPIN:
5212 		dinfo->cfg.intpin = value;
5213 		return (0);
5214 	case PCI_IVAR_ETHADDR:
5215 	case PCI_IVAR_SUBVENDOR:
5216 	case PCI_IVAR_SUBDEVICE:
5217 	case PCI_IVAR_VENDOR:
5218 	case PCI_IVAR_DEVICE:
5219 	case PCI_IVAR_DEVID:
5220 	case PCI_IVAR_CLASS:
5221 	case PCI_IVAR_SUBCLASS:
5222 	case PCI_IVAR_PROGIF:
5223 	case PCI_IVAR_REVID:
5224 	case PCI_IVAR_IRQ:
5225 	case PCI_IVAR_DOMAIN:
5226 	case PCI_IVAR_BUS:
5227 	case PCI_IVAR_SLOT:
5228 	case PCI_IVAR_FUNCTION:
5229 		return (EINVAL);	/* disallow for now */
5230 
5231 	default:
5232 		return (ENOENT);
5233 	}
5234 }
5235 
5236 #include "opt_ddb.h"
5237 #ifdef DDB
5238 #include <ddb/ddb.h>
5239 #include <sys/cons.h>
5240 
5241 /*
5242  * List resources based on pci map registers, used for within ddb
5243  */
5244 
5245 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5246 {
5247 	struct pci_devinfo *dinfo;
5248 	struct devlist *devlist_head;
5249 	struct pci_conf *p;
5250 	const char *name;
5251 	int i, error, none_count;
5252 
5253 	none_count = 0;
5254 	/* get the head of the device queue */
5255 	devlist_head = &pci_devq;
5256 
5257 	/*
5258 	 * Go through the list of devices and print out devices
5259 	 */
5260 	for (error = 0, i = 0,
5261 	     dinfo = STAILQ_FIRST(devlist_head);
5262 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5263 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5264 		/* Populate pd_name and pd_unit */
5265 		name = NULL;
5266 		if (dinfo->cfg.dev)
5267 			name = device_get_name(dinfo->cfg.dev);
5268 
5269 		p = &dinfo->conf;
5270 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5271 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5272 			(name && *name) ? name : "none",
5273 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5274 			none_count++,
5275 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5276 			p->pc_sel.pc_func, (p->pc_class << 16) |
5277 			(p->pc_subclass << 8) | p->pc_progif,
5278 			(p->pc_subdevice << 16) | p->pc_subvendor,
5279 			(p->pc_device << 16) | p->pc_vendor,
5280 			p->pc_revid, p->pc_hdr);
5281 	}
5282 }
5283 #endif /* DDB */
5284 
5285 static struct resource *
5286 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5287     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5288     u_int flags)
5289 {
5290 	struct pci_devinfo *dinfo = device_get_ivars(child);
5291 	struct resource_list *rl = &dinfo->resources;
5292 	struct resource *res;
5293 	struct pci_map *pm;
5294 	uint16_t cmd;
5295 	pci_addr_t map, testval;
5296 	int mapsize;
5297 
5298 	res = NULL;
5299 
5300 	/* If rid is managed by EA, ignore it */
5301 	if (pci_ea_is_enabled(child, *rid))
5302 		goto out;
5303 
5304 	pm = pci_find_bar(child, *rid);
5305 	if (pm != NULL) {
5306 		/* This is a BAR that we failed to allocate earlier. */
5307 		mapsize = pm->pm_size;
5308 		map = pm->pm_value;
5309 	} else {
5310 		/*
5311 		 * Weed out the bogons, and figure out how large the
5312 		 * BAR/map is.  BARs that read back 0 here are bogus
5313 		 * and unimplemented.  Note: atapci in legacy mode are
5314 		 * special and handled elsewhere in the code.  If you
5315 		 * have a atapci device in legacy mode and it fails
5316 		 * here, that other code is broken.
5317 		 */
5318 		pci_read_bar(child, *rid, &map, &testval, NULL);
5319 
5320 		/*
5321 		 * Determine the size of the BAR and ignore BARs with a size
5322 		 * of 0.  Device ROM BARs use a different mask value.
5323 		 */
5324 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5325 			mapsize = pci_romsize(testval);
5326 		else
5327 			mapsize = pci_mapsize(testval);
5328 		if (mapsize == 0)
5329 			goto out;
5330 		pm = pci_add_bar(child, *rid, map, mapsize);
5331 	}
5332 
5333 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5334 		if (type != SYS_RES_MEMORY) {
5335 			if (bootverbose)
5336 				device_printf(dev,
5337 				    "child %s requested type %d for rid %#x,"
5338 				    " but the BAR says it is an memio\n",
5339 				    device_get_nameunit(child), type, *rid);
5340 			goto out;
5341 		}
5342 	} else {
5343 		if (type != SYS_RES_IOPORT) {
5344 			if (bootverbose)
5345 				device_printf(dev,
5346 				    "child %s requested type %d for rid %#x,"
5347 				    " but the BAR says it is an ioport\n",
5348 				    device_get_nameunit(child), type, *rid);
5349 			goto out;
5350 		}
5351 	}
5352 
5353 	/*
5354 	 * For real BARs, we need to override the size that
5355 	 * the driver requests, because that's what the BAR
5356 	 * actually uses and we would otherwise have a
5357 	 * situation where we might allocate the excess to
5358 	 * another driver, which won't work.
5359 	 */
5360 	count = ((pci_addr_t)1 << mapsize) * num;
5361 	if (RF_ALIGNMENT(flags) < mapsize)
5362 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5363 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5364 		flags |= RF_PREFETCHABLE;
5365 
5366 	/*
5367 	 * Allocate enough resource, and then write back the
5368 	 * appropriate BAR for that resource.
5369 	 */
5370 	resource_list_add(rl, type, *rid, start, end, count);
5371 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5372 	    count, flags & ~RF_ACTIVE);
5373 	if (res == NULL) {
5374 		resource_list_delete(rl, type, *rid);
5375 		device_printf(child,
5376 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5377 		    count, *rid, type, start, end);
5378 		goto out;
5379 	}
5380 	if (bootverbose)
5381 		device_printf(child,
5382 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5383 		    count, *rid, type, rman_get_start(res));
5384 
5385 	/* Disable decoding via the CMD register before updating the BAR */
5386 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5387 	pci_write_config(child, PCIR_COMMAND,
5388 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5389 
5390 	map = rman_get_start(res);
5391 	pci_write_bar(child, pm, map);
5392 
5393 	/* Restore the original value of the CMD register */
5394 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5395 out:
5396 	return (res);
5397 }
5398 
5399 struct resource *
5400 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5401     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5402     u_int flags)
5403 {
5404 	struct pci_devinfo *dinfo;
5405 	struct resource_list *rl;
5406 	struct resource_list_entry *rle;
5407 	struct resource *res;
5408 	pcicfgregs *cfg;
5409 
5410 	/*
5411 	 * Perform lazy resource allocation
5412 	 */
5413 	dinfo = device_get_ivars(child);
5414 	rl = &dinfo->resources;
5415 	cfg = &dinfo->cfg;
5416 	switch (type) {
5417 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5418 	case PCI_RES_BUS:
5419 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5420 		    flags));
5421 #endif
5422 	case SYS_RES_IRQ:
5423 		/*
5424 		 * Can't alloc legacy interrupt once MSI messages have
5425 		 * been allocated.
5426 		 */
5427 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5428 		    cfg->msix.msix_alloc > 0))
5429 			return (NULL);
5430 
5431 		/*
5432 		 * If the child device doesn't have an interrupt
5433 		 * routed and is deserving of an interrupt, try to
5434 		 * assign it one.
5435 		 */
5436 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5437 		    (cfg->intpin != 0))
5438 			pci_assign_interrupt(dev, child, 0);
5439 		break;
5440 	case SYS_RES_IOPORT:
5441 	case SYS_RES_MEMORY:
5442 #ifdef NEW_PCIB
5443 		/*
5444 		 * PCI-PCI bridge I/O window resources are not BARs.
5445 		 * For those allocations just pass the request up the
5446 		 * tree.
5447 		 */
5448 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5449 			switch (*rid) {
5450 			case PCIR_IOBASEL_1:
5451 			case PCIR_MEMBASE_1:
5452 			case PCIR_PMBASEL_1:
5453 				/*
5454 				 * XXX: Should we bother creating a resource
5455 				 * list entry?
5456 				 */
5457 				return (bus_generic_alloc_resource(dev, child,
5458 				    type, rid, start, end, count, flags));
5459 			}
5460 		}
5461 #endif
5462 		/* Reserve resources for this BAR if needed. */
5463 		rle = resource_list_find(rl, type, *rid);
5464 		if (rle == NULL) {
5465 			res = pci_reserve_map(dev, child, type, rid, start, end,
5466 			    count, num, flags);
5467 			if (res == NULL)
5468 				return (NULL);
5469 		}
5470 	}
5471 	return (resource_list_alloc(rl, dev, child, type, rid,
5472 	    start, end, count, flags));
5473 }
5474 
5475 struct resource *
5476 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5477     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5478 {
5479 #ifdef PCI_IOV
5480 	struct pci_devinfo *dinfo;
5481 #endif
5482 
5483 	if (device_get_parent(child) != dev)
5484 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5485 		    type, rid, start, end, count, flags));
5486 
5487 #ifdef PCI_IOV
5488 	dinfo = device_get_ivars(child);
5489 	if (dinfo->cfg.flags & PCICFG_VF) {
5490 		switch (type) {
5491 		/* VFs can't have I/O BARs. */
5492 		case SYS_RES_IOPORT:
5493 			return (NULL);
5494 		case SYS_RES_MEMORY:
5495 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5496 			    start, end, count, flags));
5497 		}
5498 
5499 		/* Fall through for other types of resource allocations. */
5500 	}
5501 #endif
5502 
5503 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5504 	    count, 1, flags));
5505 }
5506 
5507 int
5508 pci_release_resource(device_t dev, device_t child, int type, int rid,
5509     struct resource *r)
5510 {
5511 	struct pci_devinfo *dinfo;
5512 	struct resource_list *rl;
5513 	pcicfgregs *cfg;
5514 
5515 	if (device_get_parent(child) != dev)
5516 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5517 		    type, rid, r));
5518 
5519 	dinfo = device_get_ivars(child);
5520 	cfg = &dinfo->cfg;
5521 
5522 #ifdef PCI_IOV
5523 	if (dinfo->cfg.flags & PCICFG_VF) {
5524 		switch (type) {
5525 		/* VFs can't have I/O BARs. */
5526 		case SYS_RES_IOPORT:
5527 			return (EDOOFUS);
5528 		case SYS_RES_MEMORY:
5529 			return (pci_vf_release_mem_resource(dev, child, rid,
5530 			    r));
5531 		}
5532 
5533 		/* Fall through for other types of resource allocations. */
5534 	}
5535 #endif
5536 
5537 #ifdef NEW_PCIB
5538 	/*
5539 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5540 	 * those allocations just pass the request up the tree.
5541 	 */
5542 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5543 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5544 		switch (rid) {
5545 		case PCIR_IOBASEL_1:
5546 		case PCIR_MEMBASE_1:
5547 		case PCIR_PMBASEL_1:
5548 			return (bus_generic_release_resource(dev, child, type,
5549 			    rid, r));
5550 		}
5551 	}
5552 #endif
5553 
5554 	rl = &dinfo->resources;
5555 	return (resource_list_release(rl, dev, child, type, rid, r));
5556 }
5557 
5558 int
5559 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5560     struct resource *r)
5561 {
5562 	struct pci_devinfo *dinfo;
5563 	int error;
5564 
5565 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5566 	if (error)
5567 		return (error);
5568 
5569 	/* Enable decoding in the command register when activating BARs. */
5570 	if (device_get_parent(child) == dev) {
5571 		/* Device ROMs need their decoding explicitly enabled. */
5572 		dinfo = device_get_ivars(child);
5573 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5574 			pci_write_bar(child, pci_find_bar(child, rid),
5575 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5576 		switch (type) {
5577 		case SYS_RES_IOPORT:
5578 		case SYS_RES_MEMORY:
5579 			error = PCI_ENABLE_IO(dev, child, type);
5580 			break;
5581 		}
5582 	}
5583 	return (error);
5584 }
5585 
5586 int
5587 pci_deactivate_resource(device_t dev, device_t child, int type,
5588     int rid, struct resource *r)
5589 {
5590 	struct pci_devinfo *dinfo;
5591 	int error;
5592 
5593 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5594 	if (error)
5595 		return (error);
5596 
5597 	/* Disable decoding for device ROMs. */
5598 	if (device_get_parent(child) == dev) {
5599 		dinfo = device_get_ivars(child);
5600 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5601 			pci_write_bar(child, pci_find_bar(child, rid),
5602 			    rman_get_start(r));
5603 	}
5604 	return (0);
5605 }
5606 
5607 void
5608 pci_child_deleted(device_t dev, device_t child)
5609 {
5610 	struct resource_list_entry *rle;
5611 	struct resource_list *rl;
5612 	struct pci_devinfo *dinfo;
5613 
5614 	dinfo = device_get_ivars(child);
5615 	rl = &dinfo->resources;
5616 
5617 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5618 
5619 	/* Turn off access to resources we're about to free */
5620 	if (bus_child_present(child) != 0) {
5621 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5622 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5623 
5624 		pci_disable_busmaster(child);
5625 	}
5626 
5627 	/* Free all allocated resources */
5628 	STAILQ_FOREACH(rle, rl, link) {
5629 		if (rle->res) {
5630 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5631 			    resource_list_busy(rl, rle->type, rle->rid)) {
5632 				pci_printf(&dinfo->cfg,
5633 				    "Resource still owned, oops. "
5634 				    "(type=%d, rid=%d, addr=%lx)\n",
5635 				    rle->type, rle->rid,
5636 				    rman_get_start(rle->res));
5637 				bus_release_resource(child, rle->type, rle->rid,
5638 				    rle->res);
5639 			}
5640 			resource_list_unreserve(rl, dev, child, rle->type,
5641 			    rle->rid);
5642 		}
5643 	}
5644 	resource_list_free(rl);
5645 
5646 	pci_freecfg(dinfo);
5647 }
5648 
5649 void
5650 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5651 {
5652 	struct pci_devinfo *dinfo;
5653 	struct resource_list *rl;
5654 	struct resource_list_entry *rle;
5655 
5656 	if (device_get_parent(child) != dev)
5657 		return;
5658 
5659 	dinfo = device_get_ivars(child);
5660 	rl = &dinfo->resources;
5661 	rle = resource_list_find(rl, type, rid);
5662 	if (rle == NULL)
5663 		return;
5664 
5665 	if (rle->res) {
5666 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5667 		    resource_list_busy(rl, type, rid)) {
5668 			device_printf(dev, "delete_resource: "
5669 			    "Resource still owned by child, oops. "
5670 			    "(type=%d, rid=%d, addr=%jx)\n",
5671 			    type, rid, rman_get_start(rle->res));
5672 			return;
5673 		}
5674 		resource_list_unreserve(rl, dev, child, type, rid);
5675 	}
5676 	resource_list_delete(rl, type, rid);
5677 }
5678 
5679 struct resource_list *
5680 pci_get_resource_list (device_t dev, device_t child)
5681 {
5682 	struct pci_devinfo *dinfo = device_get_ivars(child);
5683 
5684 	return (&dinfo->resources);
5685 }
5686 
5687 #ifdef IOMMU
5688 bus_dma_tag_t
5689 pci_get_dma_tag(device_t bus, device_t dev)
5690 {
5691 	bus_dma_tag_t tag;
5692 	struct pci_softc *sc;
5693 
5694 	if (device_get_parent(dev) == bus) {
5695 		/* try iommu and return if it works */
5696 		tag = iommu_get_dma_tag(bus, dev);
5697 	} else
5698 		tag = NULL;
5699 	if (tag == NULL) {
5700 		sc = device_get_softc(bus);
5701 		tag = sc->sc_dma_tag;
5702 	}
5703 	return (tag);
5704 }
5705 #else
5706 bus_dma_tag_t
5707 pci_get_dma_tag(device_t bus, device_t dev)
5708 {
5709 	struct pci_softc *sc = device_get_softc(bus);
5710 
5711 	return (sc->sc_dma_tag);
5712 }
5713 #endif
5714 
5715 uint32_t
5716 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5717 {
5718 	struct pci_devinfo *dinfo = device_get_ivars(child);
5719 	pcicfgregs *cfg = &dinfo->cfg;
5720 
5721 #ifdef PCI_IOV
5722 	/*
5723 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5724 	 * emulate them here.
5725 	 */
5726 	if (cfg->flags & PCICFG_VF) {
5727 		if (reg == PCIR_VENDOR) {
5728 			switch (width) {
5729 			case 4:
5730 				return (cfg->device << 16 | cfg->vendor);
5731 			case 2:
5732 				return (cfg->vendor);
5733 			case 1:
5734 				return (cfg->vendor & 0xff);
5735 			default:
5736 				return (0xffffffff);
5737 			}
5738 		} else if (reg == PCIR_DEVICE) {
5739 			switch (width) {
5740 			/* Note that an unaligned 4-byte read is an error. */
5741 			case 2:
5742 				return (cfg->device);
5743 			case 1:
5744 				return (cfg->device & 0xff);
5745 			default:
5746 				return (0xffffffff);
5747 			}
5748 		}
5749 	}
5750 #endif
5751 
5752 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5753 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5754 }
5755 
5756 void
5757 pci_write_config_method(device_t dev, device_t child, int reg,
5758     uint32_t val, int width)
5759 {
5760 	struct pci_devinfo *dinfo = device_get_ivars(child);
5761 	pcicfgregs *cfg = &dinfo->cfg;
5762 
5763 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5764 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5765 }
5766 
5767 int
5768 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5769     size_t buflen)
5770 {
5771 
5772 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5773 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5774 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5775 	return (0);
5776 }
5777 
5778 int
5779 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5780     size_t buflen)
5781 {
5782 	struct pci_devinfo *dinfo;
5783 	pcicfgregs *cfg;
5784 
5785 	dinfo = device_get_ivars(child);
5786 	cfg = &dinfo->cfg;
5787 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5788 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5789 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5790 	    cfg->progif);
5791 	return (0);
5792 }
5793 
5794 int
5795 pci_assign_interrupt_method(device_t dev, device_t child)
5796 {
5797 	struct pci_devinfo *dinfo = device_get_ivars(child);
5798 	pcicfgregs *cfg = &dinfo->cfg;
5799 
5800 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5801 	    cfg->intpin));
5802 }
5803 
5804 static void
5805 pci_lookup(void *arg, const char *name, device_t *dev)
5806 {
5807 	long val;
5808 	char *end;
5809 	int domain, bus, slot, func;
5810 
5811 	if (*dev != NULL)
5812 		return;
5813 
5814 	/*
5815 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5816 	 * pciB:S:F.  In the latter case, the domain is assumed to
5817 	 * be zero.
5818 	 */
5819 	if (strncmp(name, "pci", 3) != 0)
5820 		return;
5821 	val = strtol(name + 3, &end, 10);
5822 	if (val < 0 || val > INT_MAX || *end != ':')
5823 		return;
5824 	domain = val;
5825 	val = strtol(end + 1, &end, 10);
5826 	if (val < 0 || val > INT_MAX || *end != ':')
5827 		return;
5828 	bus = val;
5829 	val = strtol(end + 1, &end, 10);
5830 	if (val < 0 || val > INT_MAX)
5831 		return;
5832 	slot = val;
5833 	if (*end == ':') {
5834 		val = strtol(end + 1, &end, 10);
5835 		if (val < 0 || val > INT_MAX || *end != '\0')
5836 			return;
5837 		func = val;
5838 	} else if (*end == '\0') {
5839 		func = slot;
5840 		slot = bus;
5841 		bus = domain;
5842 		domain = 0;
5843 	} else
5844 		return;
5845 
5846 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5847 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5848 		return;
5849 
5850 	*dev = pci_find_dbsf(domain, bus, slot, func);
5851 }
5852 
5853 static int
5854 pci_modevent(module_t mod, int what, void *arg)
5855 {
5856 	static struct cdev *pci_cdev;
5857 	static eventhandler_tag tag;
5858 
5859 	switch (what) {
5860 	case MOD_LOAD:
5861 		STAILQ_INIT(&pci_devq);
5862 		pci_generation = 0;
5863 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5864 		    "pci");
5865 		pci_load_vendor_data();
5866 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5867 		    1000);
5868 		break;
5869 
5870 	case MOD_UNLOAD:
5871 		if (tag != NULL)
5872 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5873 		destroy_dev(pci_cdev);
5874 		break;
5875 	}
5876 
5877 	return (0);
5878 }
5879 
5880 static void
5881 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5882 {
5883 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5884 	struct pcicfg_pcie *cfg;
5885 	int version, pos;
5886 
5887 	cfg = &dinfo->cfg.pcie;
5888 	pos = cfg->pcie_location;
5889 
5890 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5891 
5892 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5893 
5894 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5895 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5896 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5897 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5898 
5899 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5900 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5901 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5902 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5903 
5904 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5905 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5906 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5907 
5908 	if (version > 1) {
5909 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5910 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5911 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5912 	}
5913 #undef WREG
5914 }
5915 
5916 static void
5917 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5918 {
5919 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5920 	    dinfo->cfg.pcix.pcix_command,  2);
5921 }
5922 
5923 void
5924 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5925 {
5926 
5927 	/*
5928 	 * Restore the device to full power mode.  We must do this
5929 	 * before we restore the registers because moving from D3 to
5930 	 * D0 will cause the chip's BARs and some other registers to
5931 	 * be reset to some unknown power on reset values.  Cut down
5932 	 * the noise on boot by doing nothing if we are already in
5933 	 * state D0.
5934 	 */
5935 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5936 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5937 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5938 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5939 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5940 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5941 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5942 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5943 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5944 	case PCIM_HDRTYPE_NORMAL:
5945 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5946 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5947 		break;
5948 	case PCIM_HDRTYPE_BRIDGE:
5949 		pci_write_config(dev, PCIR_SECLAT_1,
5950 		    dinfo->cfg.bridge.br_seclat, 1);
5951 		pci_write_config(dev, PCIR_SUBBUS_1,
5952 		    dinfo->cfg.bridge.br_subbus, 1);
5953 		pci_write_config(dev, PCIR_SECBUS_1,
5954 		    dinfo->cfg.bridge.br_secbus, 1);
5955 		pci_write_config(dev, PCIR_PRIBUS_1,
5956 		    dinfo->cfg.bridge.br_pribus, 1);
5957 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5958 		    dinfo->cfg.bridge.br_control, 2);
5959 		break;
5960 	case PCIM_HDRTYPE_CARDBUS:
5961 		pci_write_config(dev, PCIR_SECLAT_2,
5962 		    dinfo->cfg.bridge.br_seclat, 1);
5963 		pci_write_config(dev, PCIR_SUBBUS_2,
5964 		    dinfo->cfg.bridge.br_subbus, 1);
5965 		pci_write_config(dev, PCIR_SECBUS_2,
5966 		    dinfo->cfg.bridge.br_secbus, 1);
5967 		pci_write_config(dev, PCIR_PRIBUS_2,
5968 		    dinfo->cfg.bridge.br_pribus, 1);
5969 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5970 		    dinfo->cfg.bridge.br_control, 2);
5971 		break;
5972 	}
5973 	pci_restore_bars(dev);
5974 
5975 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
5976 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5977 
5978 	/*
5979 	 * Restore extended capabilities for PCI-Express and PCI-X
5980 	 */
5981 	if (dinfo->cfg.pcie.pcie_location != 0)
5982 		pci_cfg_restore_pcie(dev, dinfo);
5983 	if (dinfo->cfg.pcix.pcix_location != 0)
5984 		pci_cfg_restore_pcix(dev, dinfo);
5985 
5986 	/* Restore MSI and MSI-X configurations if they are present. */
5987 	if (dinfo->cfg.msi.msi_location != 0)
5988 		pci_resume_msi(dev);
5989 	if (dinfo->cfg.msix.msix_location != 0)
5990 		pci_resume_msix(dev);
5991 
5992 #ifdef PCI_IOV
5993 	if (dinfo->cfg.iov != NULL)
5994 		pci_iov_cfg_restore(dev, dinfo);
5995 #endif
5996 }
5997 
5998 static void
5999 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6000 {
6001 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6002 	struct pcicfg_pcie *cfg;
6003 	int version, pos;
6004 
6005 	cfg = &dinfo->cfg.pcie;
6006 	pos = cfg->pcie_location;
6007 
6008 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6009 
6010 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6011 
6012 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6013 
6014 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6015 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6016 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6017 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6018 
6019 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6020 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6021 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6022 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6023 
6024 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6025 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6026 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6027 
6028 	if (version > 1) {
6029 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6030 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6031 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6032 	}
6033 #undef RREG
6034 }
6035 
6036 static void
6037 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6038 {
6039 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6040 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6041 }
6042 
6043 void
6044 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6045 {
6046 	uint32_t cls;
6047 	int ps;
6048 
6049 	/*
6050 	 * Some drivers apparently write to these registers w/o updating our
6051 	 * cached copy.  No harm happens if we update the copy, so do so here
6052 	 * so we can restore them.  The COMMAND register is modified by the
6053 	 * bus w/o updating the cache.  This should represent the normally
6054 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6055 	 */
6056 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6057 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6058 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6059 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6060 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6061 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6062 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6063 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6064 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6065 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6066 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6067 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6068 	case PCIM_HDRTYPE_NORMAL:
6069 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6070 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6071 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6072 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6073 		break;
6074 	case PCIM_HDRTYPE_BRIDGE:
6075 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6076 		    PCIR_SECLAT_1, 1);
6077 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6078 		    PCIR_SUBBUS_1, 1);
6079 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6080 		    PCIR_SECBUS_1, 1);
6081 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6082 		    PCIR_PRIBUS_1, 1);
6083 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6084 		    PCIR_BRIDGECTL_1, 2);
6085 		break;
6086 	case PCIM_HDRTYPE_CARDBUS:
6087 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6088 		    PCIR_SECLAT_2, 1);
6089 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6090 		    PCIR_SUBBUS_2, 1);
6091 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6092 		    PCIR_SECBUS_2, 1);
6093 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6094 		    PCIR_PRIBUS_2, 1);
6095 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6096 		    PCIR_BRIDGECTL_2, 2);
6097 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6098 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6099 		break;
6100 	}
6101 
6102 	if (dinfo->cfg.pcie.pcie_location != 0)
6103 		pci_cfg_save_pcie(dev, dinfo);
6104 
6105 	if (dinfo->cfg.pcix.pcix_location != 0)
6106 		pci_cfg_save_pcix(dev, dinfo);
6107 
6108 #ifdef PCI_IOV
6109 	if (dinfo->cfg.iov != NULL)
6110 		pci_iov_cfg_save(dev, dinfo);
6111 #endif
6112 
6113 	/*
6114 	 * don't set the state for display devices, base peripherals and
6115 	 * memory devices since bad things happen when they are powered down.
6116 	 * We should (a) have drivers that can easily detach and (b) use
6117 	 * generic drivers for these devices so that some device actually
6118 	 * attaches.  We need to make sure that when we implement (a) we don't
6119 	 * power the device down on a reattach.
6120 	 */
6121 	cls = pci_get_class(dev);
6122 	if (!setstate)
6123 		return;
6124 	switch (pci_do_power_nodriver)
6125 	{
6126 		case 0:		/* NO powerdown at all */
6127 			return;
6128 		case 1:		/* Conservative about what to power down */
6129 			if (cls == PCIC_STORAGE)
6130 				return;
6131 			/*FALLTHROUGH*/
6132 		case 2:		/* Aggressive about what to power down */
6133 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6134 			    cls == PCIC_BASEPERIPH)
6135 				return;
6136 			/*FALLTHROUGH*/
6137 		case 3:		/* Power down everything */
6138 			break;
6139 	}
6140 	/*
6141 	 * PCI spec says we can only go into D3 state from D0 state.
6142 	 * Transition from D[12] into D0 before going to D3 state.
6143 	 */
6144 	ps = pci_get_powerstate(dev);
6145 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6146 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6147 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6148 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6149 }
6150 
6151 /* Wrapper APIs suitable for device driver use. */
6152 void
6153 pci_save_state(device_t dev)
6154 {
6155 	struct pci_devinfo *dinfo;
6156 
6157 	dinfo = device_get_ivars(dev);
6158 	pci_cfg_save(dev, dinfo, 0);
6159 }
6160 
6161 void
6162 pci_restore_state(device_t dev)
6163 {
6164 	struct pci_devinfo *dinfo;
6165 
6166 	dinfo = device_get_ivars(dev);
6167 	pci_cfg_restore(dev, dinfo);
6168 }
6169 
6170 static int
6171 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6172     uintptr_t *id)
6173 {
6174 
6175 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6176 }
6177 
6178 /* Find the upstream port of a given PCI device in a root complex. */
6179 device_t
6180 pci_find_pcie_root_port(device_t dev)
6181 {
6182 	struct pci_devinfo *dinfo;
6183 	devclass_t pci_class;
6184 	device_t pcib, bus;
6185 
6186 	pci_class = devclass_find("pci");
6187 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6188 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6189 
6190 	/*
6191 	 * Walk the bridge hierarchy until we find a PCI-e root
6192 	 * port or a non-PCI device.
6193 	 */
6194 	for (;;) {
6195 		bus = device_get_parent(dev);
6196 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6197 		    device_get_nameunit(dev)));
6198 
6199 		pcib = device_get_parent(bus);
6200 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6201 		    device_get_nameunit(bus)));
6202 
6203 		/*
6204 		 * pcib's parent must be a PCI bus for this to be a
6205 		 * PCI-PCI bridge.
6206 		 */
6207 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6208 			return (NULL);
6209 
6210 		dinfo = device_get_ivars(pcib);
6211 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6212 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6213 			return (pcib);
6214 
6215 		dev = pcib;
6216 	}
6217 }
6218 
6219 /*
6220  * Wait for pending transactions to complete on a PCI-express function.
6221  *
6222  * The maximum delay is specified in milliseconds in max_delay.  Note
6223  * that this function may sleep.
6224  *
6225  * Returns true if the function is idle and false if the timeout is
6226  * exceeded.  If dev is not a PCI-express function, this returns true.
6227  */
6228 bool
6229 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6230 {
6231 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6232 	uint16_t sta;
6233 	int cap;
6234 
6235 	cap = dinfo->cfg.pcie.pcie_location;
6236 	if (cap == 0)
6237 		return (true);
6238 
6239 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6240 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6241 		if (max_delay == 0)
6242 			return (false);
6243 
6244 		/* Poll once every 100 milliseconds up to the timeout. */
6245 		if (max_delay > 100) {
6246 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6247 			max_delay -= 100;
6248 		} else {
6249 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6250 			    C_HARDCLOCK);
6251 			max_delay = 0;
6252 		}
6253 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6254 	}
6255 
6256 	return (true);
6257 }
6258 
6259 /*
6260  * Determine the maximum Completion Timeout in microseconds.
6261  *
6262  * For non-PCI-express functions this returns 0.
6263  */
6264 int
6265 pcie_get_max_completion_timeout(device_t dev)
6266 {
6267 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6268 	int cap;
6269 
6270 	cap = dinfo->cfg.pcie.pcie_location;
6271 	if (cap == 0)
6272 		return (0);
6273 
6274 	/*
6275 	 * Functions using the 1.x spec use the default timeout range of
6276 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6277 	 * support programmable timeouts also use this range.
6278 	 */
6279 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6280 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6281 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6282 		return (50 * 1000);
6283 
6284 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6285 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6286 	case PCIEM_CTL2_COMP_TIMO_100US:
6287 		return (100);
6288 	case PCIEM_CTL2_COMP_TIMO_10MS:
6289 		return (10 * 1000);
6290 	case PCIEM_CTL2_COMP_TIMO_55MS:
6291 		return (55 * 1000);
6292 	case PCIEM_CTL2_COMP_TIMO_210MS:
6293 		return (210 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_900MS:
6295 		return (900 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6297 		return (3500 * 1000);
6298 	case PCIEM_CTL2_COMP_TIMO_13S:
6299 		return (13 * 1000 * 1000);
6300 	case PCIEM_CTL2_COMP_TIMO_64S:
6301 		return (64 * 1000 * 1000);
6302 	default:
6303 		return (50 * 1000);
6304 	}
6305 }
6306 
6307 void
6308 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6309 {
6310 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6311 	const char *s;
6312 	int aer;
6313 	uint32_t r, r1;
6314 	uint16_t rs;
6315 
6316 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6317 		s = "Correctable";
6318 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6319 		s = "Uncorrectable (Non-Fatal)";
6320 	else
6321 		s = "Uncorrectable (Fatal)";
6322 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6323 	if (aerp) {
6324 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6325 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6326 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6327 		} else {
6328 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6329 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6330 		}
6331 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6332 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6333 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6334 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6335 			printf(" severity 0x%08x first %d\n",
6336 			    r, rs & 0x1f);
6337 		} else
6338 			printf("\n");
6339 	}
6340 
6341 	/* As kind of recovery just report and clear the error statuses. */
6342 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6343 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6344 		if (r != 0) {
6345 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6346 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6347 		}
6348 
6349 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6350 		if (r != 0) {
6351 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6352 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6353 		}
6354 	}
6355 	if (dinfo->cfg.pcie.pcie_location != 0) {
6356 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6357 		    PCIER_DEVICE_STA, 2);
6358 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6359 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6360 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6361 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6362 			    PCIER_DEVICE_STA, rs, 2);
6363 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6364 		}
6365 	}
6366 }
6367 
6368 /*
6369  * Perform a Function Level Reset (FLR) on a device.
6370  *
6371  * This function first waits for any pending transactions to complete
6372  * within the timeout specified by max_delay.  If transactions are
6373  * still pending, the function will return false without attempting a
6374  * reset.
6375  *
6376  * If dev is not a PCI-express function or does not support FLR, this
6377  * function returns false.
6378  *
6379  * Note that no registers are saved or restored.  The caller is
6380  * responsible for saving and restoring any registers including
6381  * PCI-standard registers via pci_save_state() and
6382  * pci_restore_state().
6383  */
6384 bool
6385 pcie_flr(device_t dev, u_int max_delay, bool force)
6386 {
6387 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6388 	uint16_t cmd, ctl;
6389 	int compl_delay;
6390 	int cap;
6391 
6392 	cap = dinfo->cfg.pcie.pcie_location;
6393 	if (cap == 0)
6394 		return (false);
6395 
6396 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6397 		return (false);
6398 
6399 	/*
6400 	 * Disable busmastering to prevent generation of new
6401 	 * transactions while waiting for the device to go idle.  If
6402 	 * the idle timeout fails, the command register is restored
6403 	 * which will re-enable busmastering.
6404 	 */
6405 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6406 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6407 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6408 		if (!force) {
6409 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6410 			return (false);
6411 		}
6412 		pci_printf(&dinfo->cfg,
6413 		    "Resetting with transactions pending after %d ms\n",
6414 		    max_delay);
6415 
6416 		/*
6417 		 * Extend the post-FLR delay to cover the maximum
6418 		 * Completion Timeout delay of anything in flight
6419 		 * during the FLR delay.  Enforce a minimum delay of
6420 		 * at least 10ms.
6421 		 */
6422 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6423 		if (compl_delay < 10)
6424 			compl_delay = 10;
6425 	} else
6426 		compl_delay = 0;
6427 
6428 	/* Initiate the reset. */
6429 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6430 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6431 	    PCIEM_CTL_INITIATE_FLR, 2);
6432 
6433 	/* Wait for 100ms. */
6434 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6435 
6436 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6437 	    PCIEM_STA_TRANSACTION_PND)
6438 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6439 	return (true);
6440 }
6441 
6442 /*
6443  * Attempt a power-management reset by cycling the device in/out of D3
6444  * state.  PCI spec says we can only go into D3 state from D0 state.
6445  * Transition from D[12] into D0 before going to D3 state.
6446  */
6447 int
6448 pci_power_reset(device_t dev)
6449 {
6450 	int ps;
6451 
6452 	ps = pci_get_powerstate(dev);
6453 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6454 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6455 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6456 	pci_set_powerstate(dev, ps);
6457 	return (0);
6458 }
6459 
6460 /*
6461  * Try link drop and retrain of the downstream port of upstream
6462  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6463  * cause Conventional Hot reset of the device in the slot.
6464  * Alternative, for PCIe, could be the secondary bus reset initiatied
6465  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6466  */
6467 int
6468 pcie_link_reset(device_t port, int pcie_location)
6469 {
6470 	uint16_t v;
6471 
6472 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6473 	v |= PCIEM_LINK_CTL_LINK_DIS;
6474 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6475 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6476 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6477 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6478 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6479 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6480 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6481 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6482 }
6483 
6484 static int
6485 pci_reset_post(device_t dev, device_t child)
6486 {
6487 
6488 	if (dev == device_get_parent(child))
6489 		pci_restore_state(child);
6490 	return (0);
6491 }
6492 
6493 static int
6494 pci_reset_prepare(device_t dev, device_t child)
6495 {
6496 
6497 	if (dev == device_get_parent(child))
6498 		pci_save_state(child);
6499 	return (0);
6500 }
6501 
6502 static int
6503 pci_reset_child(device_t dev, device_t child, int flags)
6504 {
6505 	int error;
6506 
6507 	if (dev == NULL || device_get_parent(child) != dev)
6508 		return (0);
6509 	if ((flags & DEVF_RESET_DETACH) != 0) {
6510 		error = device_get_state(child) == DS_ATTACHED ?
6511 		    device_detach(child) : 0;
6512 	} else {
6513 		error = BUS_SUSPEND_CHILD(dev, child);
6514 	}
6515 	if (error == 0) {
6516 		if (!pcie_flr(child, 1000, false)) {
6517 			error = BUS_RESET_PREPARE(dev, child);
6518 			if (error == 0)
6519 				pci_power_reset(child);
6520 			BUS_RESET_POST(dev, child);
6521 		}
6522 		if ((flags & DEVF_RESET_DETACH) != 0)
6523 			device_probe_and_attach(child);
6524 		else
6525 			BUS_RESUME_CHILD(dev, child);
6526 	}
6527 	return (error);
6528 }
6529 
6530 const struct pci_device_table *
6531 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6532 {
6533 	bool match;
6534 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6535 
6536 	vendor = pci_get_vendor(child);
6537 	device = pci_get_device(child);
6538 	subvendor = pci_get_subvendor(child);
6539 	subdevice = pci_get_subdevice(child);
6540 	class = pci_get_class(child);
6541 	subclass = pci_get_subclass(child);
6542 	revid = pci_get_revid(child);
6543 	while (nelt-- > 0) {
6544 		match = true;
6545 		if (id->match_flag_vendor)
6546 			match &= vendor == id->vendor;
6547 		if (id->match_flag_device)
6548 			match &= device == id->device;
6549 		if (id->match_flag_subvendor)
6550 			match &= subvendor == id->subvendor;
6551 		if (id->match_flag_subdevice)
6552 			match &= subdevice == id->subdevice;
6553 		if (id->match_flag_class)
6554 			match &= class == id->class_id;
6555 		if (id->match_flag_subclass)
6556 			match &= subclass == id->subclass;
6557 		if (id->match_flag_revid)
6558 			match &= revid == id->revid;
6559 		if (match)
6560 			return (id);
6561 		id++;
6562 	}
6563 	return (NULL);
6564 }
6565 
6566 static void
6567 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6568 {
6569 	const char *dev_name;
6570 	device_t dev;
6571 
6572 	dev = dinfo->cfg.dev;
6573 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6574 	    dinfo->cfg.slot, dinfo->cfg.func);
6575 	dev_name = device_get_name(dev);
6576 	if (dev_name != NULL)
6577 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6578 }
6579 
6580 void
6581 pci_print_faulted_dev(void)
6582 {
6583 	struct pci_devinfo *dinfo;
6584 	device_t dev;
6585 	int aer, i;
6586 	uint32_t r1, r2;
6587 	uint16_t status;
6588 
6589 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6590 		dev = dinfo->cfg.dev;
6591 		status = pci_read_config(dev, PCIR_STATUS, 2);
6592 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6593 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6594 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6595 		if (status != 0) {
6596 			pci_print_faulted_dev_name(dinfo);
6597 			printf(" error 0x%04x\n", status);
6598 		}
6599 		if (dinfo->cfg.pcie.pcie_location != 0) {
6600 			status = pci_read_config(dev,
6601 			    dinfo->cfg.pcie.pcie_location +
6602 			    PCIER_DEVICE_STA, 2);
6603 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6604 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6605 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6606 				pci_print_faulted_dev_name(dinfo);
6607 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6608 				    pci_read_config(dev,
6609 				    dinfo->cfg.pcie.pcie_location +
6610 				    PCIER_DEVICE_CTL, 2),
6611 				    status);
6612 			}
6613 		}
6614 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6615 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6616 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6617 			if (r1 != 0 || r2 != 0) {
6618 				pci_print_faulted_dev_name(dinfo);
6619 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6620 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6621 				    r1, pci_read_config(dev, aer +
6622 				    PCIR_AER_UC_MASK, 4),
6623 				    pci_read_config(dev, aer +
6624 				    PCIR_AER_UC_SEVERITY, 4),
6625 				    r2, pci_read_config(dev, aer +
6626 				    PCIR_AER_COR_MASK, 4),
6627 				    pci_read_config(dev, aer +
6628 				    PCIR_AER_CAP_CONTROL, 4));
6629 				for (i = 0; i < 4; i++) {
6630 					r1 = pci_read_config(dev, aer +
6631 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6632 					printf("    HL%d: 0x%08x\n", i, r1);
6633 				}
6634 			}
6635 		}
6636 	}
6637 }
6638 
6639 #ifdef DDB
6640 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6641 {
6642 
6643 	pci_print_faulted_dev();
6644 }
6645 
6646 static void
6647 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6648 {
6649 	device_t dev;
6650 	int aer;
6651 	uint32_t r;
6652 
6653 	dev = dinfo->cfg.dev;
6654 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6655 	    PCIER_DEVICE_STA, 2);
6656 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6657 	    PCIER_DEVICE_STA, r, 2);
6658 
6659 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6660 		return;
6661 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6662 	if (r != 0)
6663 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6664 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6665 	if (r != 0)
6666 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6667 }
6668 
6669 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6670 {
6671 	struct pci_devinfo *dinfo;
6672 	device_t dev;
6673 	uint16_t status, status1;
6674 
6675 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6676 		dev = dinfo->cfg.dev;
6677 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6678 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6679 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6680 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6681 		if (status1 != 0) {
6682 			status &= ~status1;
6683 			pci_write_config(dev, PCIR_STATUS, status, 2);
6684 		}
6685 		if (dinfo->cfg.pcie.pcie_location != 0)
6686 			db_clear_pcie_errors(dinfo);
6687 	}
6688 }
6689 #endif
6690