xref: /freebsd/sys/dev/pci/pci.c (revision e7c5d9d3f2a6b76b7528aeaf2e9e284345ba62b0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_bus.h"
36 
37 #include <sys/param.h>
38 #include <sys/conf.h>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_extern.h>
54 
55 #include <sys/bus.h>
56 #include <machine/bus.h>
57 #include <sys/rman.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
60 
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
63 #endif
64 
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
69 
70 #ifdef PCI_IOV
71 #include <sys/nv.h>
72 #include <dev/pci/pci_iov_private.h>
73 #endif
74 
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
79 
80 #include "pcib_if.h"
81 #include "pci_if.h"
82 
83 #define	PCIR_IS_BIOS(cfg, reg)						\
84 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
85 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 
87 static int		pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t	pci_mapbase(uint64_t mapreg);
89 static const char	*pci_maptype(uint64_t mapreg);
90 static int		pci_maprange(uint64_t mapreg);
91 static pci_addr_t	pci_rombase(uint64_t mapreg);
92 static int		pci_romsize(uint64_t testval);
93 static void		pci_fixancient(pcicfgregs *cfg);
94 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 
96 static int		pci_porten(device_t dev);
97 static int		pci_memen(device_t dev);
98 static void		pci_assign_interrupt(device_t bus, device_t dev,
99 			    int force_route);
100 static int		pci_add_map(device_t bus, device_t dev, int reg,
101 			    struct resource_list *rl, int force, int prefetch);
102 static int		pci_probe(device_t dev);
103 static int		pci_attach(device_t dev);
104 static int		pci_detach(device_t dev);
105 static void		pci_load_vendor_data(void);
106 static int		pci_describe_parse_line(char **ptr, int *vendor,
107 			    int *device, char **desc);
108 static char		*pci_describe_device(device_t dev);
109 static int		pci_modevent(module_t mod, int what, void *arg);
110 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
111 			    pcicfgregs *cfg);
112 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
113 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
114 			    int reg, uint32_t *data);
115 #if 0
116 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
117 			    int reg, uint32_t data);
118 #endif
119 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
120 static void		pci_mask_msix(device_t dev, u_int index);
121 static void		pci_unmask_msix(device_t dev, u_int index);
122 static int		pci_msi_blacklisted(void);
123 static int		pci_msix_blacklisted(void);
124 static void		pci_resume_msi(device_t dev);
125 static void		pci_resume_msix(device_t dev);
126 static int		pci_remap_intr_method(device_t bus, device_t dev,
127 			    u_int irq);
128 static void		pci_hint_device_unit(device_t acdev, device_t child,
129 			    const char *name, int *unitp);
130 static int		pci_reset_post(device_t dev, device_t child);
131 static int		pci_reset_prepare(device_t dev, device_t child);
132 static int		pci_reset_child(device_t dev, device_t child,
133 			    int flags);
134 
135 static int		pci_get_id_method(device_t dev, device_t child,
136 			    enum pci_id_type type, uintptr_t *rid);
137 
138 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
139     int b, int s, int f, uint16_t vid, uint16_t did);
140 
141 static device_method_t pci_methods[] = {
142 	/* Device interface */
143 	DEVMETHOD(device_probe,		pci_probe),
144 	DEVMETHOD(device_attach,	pci_attach),
145 	DEVMETHOD(device_detach,	pci_detach),
146 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
147 	DEVMETHOD(device_suspend,	bus_generic_suspend),
148 	DEVMETHOD(device_resume,	pci_resume),
149 
150 	/* Bus interface */
151 	DEVMETHOD(bus_print_child,	pci_print_child),
152 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
153 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
154 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
155 	DEVMETHOD(bus_driver_added,	pci_driver_added),
156 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
157 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
158 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
159 	DEVMETHOD(bus_reset_post,	pci_reset_post),
160 	DEVMETHOD(bus_reset_child,	pci_reset_child),
161 
162 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
163 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
164 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
165 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
166 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
167 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
168 	DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
169 	DEVMETHOD(bus_release_resource,	pci_release_resource),
170 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
171 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
172 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
173 	DEVMETHOD(bus_child_detached,	pci_child_detached),
174 	DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
175 	DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
176 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
177 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
178 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
179 	DEVMETHOD(bus_resume_child,	pci_resume_child),
180 	DEVMETHOD(bus_rescan,		pci_rescan_method),
181 
182 	/* PCI interface */
183 	DEVMETHOD(pci_read_config,	pci_read_config_method),
184 	DEVMETHOD(pci_write_config,	pci_write_config_method),
185 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
186 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
187 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
188 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
189 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
190 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
191 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
192 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
193 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
194 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
195 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
196 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
197 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
198 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
199 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
200 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
201 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
202 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
203 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
204 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
205 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
206 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
207 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
208 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
209 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
210 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
211 	DEVMETHOD(pci_get_id,		pci_get_id_method),
212 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
213 	DEVMETHOD(pci_child_added,	pci_child_added_method),
214 #ifdef PCI_IOV
215 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
216 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
217 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
218 #endif
219 
220 	DEVMETHOD_END
221 };
222 
223 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
224 
225 static devclass_t pci_devclass;
226 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
227     BUS_PASS_BUS);
228 MODULE_VERSION(pci, 1);
229 
230 static char	*pci_vendordata;
231 static size_t	pci_vendordata_size;
232 
233 struct pci_quirk {
234 	uint32_t devid;	/* Vendor/device of the card */
235 	int	type;
236 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
237 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
238 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
239 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
240 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
241 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
243 	int	arg1;
244 	int	arg2;
245 };
246 
247 static const struct pci_quirk pci_quirks[] = {
248 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
249 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
250 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
252 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 
254 	/*
255 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
256 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
257 	 */
258 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
259 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
260 
261 	/*
262 	 * MSI doesn't work on earlier Intel chipsets including
263 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
264 	 */
265 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 
273 	/*
274 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
275 	 * bridge.
276 	 */
277 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
278 
279 	/*
280 	 * MSI-X allocation doesn't work properly for devices passed through
281 	 * by VMware up to at least ESXi 5.1.
282 	 */
283 	{ 0x079015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCI/PCI-X */
284 	{ 0x07a015ad, PCI_QUIRK_DISABLE_MSIX,	0,	0 }, /* PCIe */
285 
286 	/*
287 	 * Some virtualization environments emulate an older chipset
288 	 * but support MSI just fine.  QEMU uses the Intel 82440.
289 	 */
290 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
291 
292 	/*
293 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
294 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
295 	 * It prevents us from attaching hpet(4) when the bit is unset.
296 	 * Note this quirk only affects SB600 revision A13 and earlier.
297 	 * For SB600 A21 and later, firmware must set the bit to hide it.
298 	 * For SB700 and later, it is unused and hardcoded to zero.
299 	 */
300 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
301 
302 	/*
303 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
304 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
305 	 * of the command register is set.
306 	 */
307 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
309 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
310 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
311 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
312 
313 	/*
314 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
315 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
316 	 */
317 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
318 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
319 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
320 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
321 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
322 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
323 
324 	/*
325 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
326 	 * expected place.
327 	 */
328 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
329 
330 	{ 0 }
331 };
332 
333 /* map register information */
334 #define	PCI_MAPMEM	0x01	/* memory map */
335 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
336 #define	PCI_MAPPORT	0x04	/* port map */
337 
338 struct devlist pci_devq;
339 uint32_t pci_generation;
340 uint32_t pci_numdevs = 0;
341 static int pcie_chipset, pcix_chipset;
342 
343 /* sysctl vars */
344 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
345 
346 static int pci_enable_io_modes = 1;
347 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
348     &pci_enable_io_modes, 1,
349     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
350     " enable these bits correctly.  We'd like to do this all the time, but"
351     " there are some peripherals that this causes problems with.");
352 
353 static int pci_do_realloc_bars = 1;
354 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
355     &pci_do_realloc_bars, 0,
356     "Attempt to allocate a new range for any BARs whose original "
357     "firmware-assigned ranges fail to allocate during the initial device scan.");
358 
359 static int pci_do_power_nodriver = 0;
360 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
361     &pci_do_power_nodriver, 0,
362     "Place a function into D3 state when no driver attaches to it.  0 means"
363     " disable.  1 means conservatively place devices into D3 state.  2 means"
364     " aggressively place devices into D3 state.  3 means put absolutely"
365     " everything in D3 state.");
366 
367 int pci_do_power_resume = 1;
368 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
369     &pci_do_power_resume, 1,
370   "Transition from D3 -> D0 on resume.");
371 
372 int pci_do_power_suspend = 1;
373 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
374     &pci_do_power_suspend, 1,
375   "Transition from D0 -> D3 on suspend.");
376 
377 static int pci_do_msi = 1;
378 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
379     "Enable support for MSI interrupts");
380 
381 static int pci_do_msix = 1;
382 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
383     "Enable support for MSI-X interrupts");
384 
385 static int pci_msix_rewrite_table = 0;
386 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
387     &pci_msix_rewrite_table, 0,
388     "Rewrite entire MSI-X table when updating MSI-X entries");
389 
390 static int pci_honor_msi_blacklist = 1;
391 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
392     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
393 
394 #if defined(__i386__) || defined(__amd64__)
395 static int pci_usb_takeover = 1;
396 #else
397 static int pci_usb_takeover = 0;
398 #endif
399 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
400     &pci_usb_takeover, 1,
401     "Enable early takeover of USB controllers. Disable this if you depend on"
402     " BIOS emulation of USB devices, that is you use USB devices (like"
403     " keyboard or mouse) but do not load USB drivers");
404 
405 static int pci_clear_bars;
406 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
407     "Ignore firmware-assigned resources for BARs.");
408 
409 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
410 static int pci_clear_buses;
411 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
412     "Ignore firmware-assigned bus numbers.");
413 #endif
414 
415 static int pci_enable_ari = 1;
416 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
417     0, "Enable support for PCIe Alternative RID Interpretation");
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;
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 			return (dinfo->cfg.dev);
458 		}
459 	}
460 
461 	return (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 
776 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
777 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
778 
779 		/* Read a number of dwords in the entry */
780 		val = REG(ptr, 4);
781 		ptr += 4;
782 		ent_size = (val & PCIM_EA_ES);
783 
784 		for (b = 0; b < ent_size; b++) {
785 			dw[b] = REG(ptr, 4);
786 			ptr += 4;
787 		}
788 
789 		eae->eae_flags = val;
790 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
791 
792 		base = dw[0] & PCIM_EA_FIELD_MASK;
793 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
794 		b = 2;
795 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
796 			base |= (uint64_t)dw[b] << 32UL;
797 			b++;
798 		}
799 		if (((dw[1] & PCIM_EA_IS_64) != 0)
800 		    && (b < ent_size)) {
801 			max_offset |= (uint64_t)dw[b] << 32UL;
802 			b++;
803 		}
804 
805 		eae->eae_base = base;
806 		eae->eae_max_offset = max_offset;
807 
808 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
809 
810 		if (bootverbose) {
811 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
812 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
813 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
814 		}
815 	}
816 }
817 #undef REG
818 
819 static void
820 pci_read_cap(device_t pcib, pcicfgregs *cfg)
821 {
822 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
823 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
824 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
825 	uint64_t addr;
826 #endif
827 	uint32_t val;
828 	int	ptr, nextptr, ptrptr;
829 
830 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
831 	case PCIM_HDRTYPE_NORMAL:
832 	case PCIM_HDRTYPE_BRIDGE:
833 		ptrptr = PCIR_CAP_PTR;
834 		break;
835 	case PCIM_HDRTYPE_CARDBUS:
836 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
837 		break;
838 	default:
839 		return;		/* no extended capabilities support */
840 	}
841 	nextptr = REG(ptrptr, 1);	/* sanity check? */
842 
843 	/*
844 	 * Read capability entries.
845 	 */
846 	while (nextptr != 0) {
847 		/* Sanity check */
848 		if (nextptr > 255) {
849 			printf("illegal PCI extended capability offset %d\n",
850 			    nextptr);
851 			return;
852 		}
853 		/* Find the next entry */
854 		ptr = nextptr;
855 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
856 
857 		/* Process this entry */
858 		switch (REG(ptr + PCICAP_ID, 1)) {
859 		case PCIY_PMG:		/* PCI power management */
860 			if (cfg->pp.pp_cap == 0) {
861 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
862 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
863 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
864 				if ((nextptr - ptr) > PCIR_POWER_DATA)
865 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
866 			}
867 			break;
868 		case PCIY_HT:		/* HyperTransport */
869 			/* Determine HT-specific capability type. */
870 			val = REG(ptr + PCIR_HT_COMMAND, 2);
871 
872 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
873 				cfg->ht.ht_slave = ptr;
874 
875 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
876 			switch (val & PCIM_HTCMD_CAP_MASK) {
877 			case PCIM_HTCAP_MSI_MAPPING:
878 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
879 					/* Sanity check the mapping window. */
880 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
881 					    4);
882 					addr <<= 32;
883 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
884 					    4);
885 					if (addr != MSI_INTEL_ADDR_BASE)
886 						device_printf(pcib,
887 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
888 						    cfg->domain, cfg->bus,
889 						    cfg->slot, cfg->func,
890 						    (long long)addr);
891 				} else
892 					addr = MSI_INTEL_ADDR_BASE;
893 
894 				cfg->ht.ht_msimap = ptr;
895 				cfg->ht.ht_msictrl = val;
896 				cfg->ht.ht_msiaddr = addr;
897 				break;
898 			}
899 #endif
900 			break;
901 		case PCIY_MSI:		/* PCI MSI */
902 			cfg->msi.msi_location = ptr;
903 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
904 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
905 						     PCIM_MSICTRL_MMC_MASK)>>1);
906 			break;
907 		case PCIY_MSIX:		/* PCI MSI-X */
908 			cfg->msix.msix_location = ptr;
909 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
910 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
911 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
912 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
913 			cfg->msix.msix_table_bar = PCIR_BAR(val &
914 			    PCIM_MSIX_BIR_MASK);
915 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
916 			val = REG(ptr + PCIR_MSIX_PBA, 4);
917 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
918 			    PCIM_MSIX_BIR_MASK);
919 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
920 			break;
921 		case PCIY_VPD:		/* PCI Vital Product Data */
922 			cfg->vpd.vpd_reg = ptr;
923 			break;
924 		case PCIY_SUBVENDOR:
925 			/* Should always be true. */
926 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
927 			    PCIM_HDRTYPE_BRIDGE) {
928 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
929 				cfg->subvendor = val & 0xffff;
930 				cfg->subdevice = val >> 16;
931 			}
932 			break;
933 		case PCIY_PCIX:		/* PCI-X */
934 			/*
935 			 * Assume we have a PCI-X chipset if we have
936 			 * at least one PCI-PCI bridge with a PCI-X
937 			 * capability.  Note that some systems with
938 			 * PCI-express or HT chipsets might match on
939 			 * this check as well.
940 			 */
941 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
942 			    PCIM_HDRTYPE_BRIDGE)
943 				pcix_chipset = 1;
944 			cfg->pcix.pcix_location = ptr;
945 			break;
946 		case PCIY_EXPRESS:	/* PCI-express */
947 			/*
948 			 * Assume we have a PCI-express chipset if we have
949 			 * at least one PCI-express device.
950 			 */
951 			pcie_chipset = 1;
952 			cfg->pcie.pcie_location = ptr;
953 			val = REG(ptr + PCIER_FLAGS, 2);
954 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
955 			break;
956 		case PCIY_EA:		/* Enhanced Allocation */
957 			cfg->ea.ea_location = ptr;
958 			pci_ea_fill_info(pcib, cfg);
959 			break;
960 		default:
961 			break;
962 		}
963 	}
964 
965 #if defined(__powerpc__)
966 	/*
967 	 * Enable the MSI mapping window for all HyperTransport
968 	 * slaves.  PCI-PCI bridges have their windows enabled via
969 	 * PCIB_MAP_MSI().
970 	 */
971 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
972 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
973 		device_printf(pcib,
974 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
975 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
976 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
977 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
978 		     2);
979 	}
980 #endif
981 /* REG and WREG use carry through to next functions */
982 }
983 
984 /*
985  * PCI Vital Product Data
986  */
987 
988 #define	PCI_VPD_TIMEOUT		1000000
989 
990 static int
991 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
992 {
993 	int count = PCI_VPD_TIMEOUT;
994 
995 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
996 
997 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
998 
999 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1000 		if (--count < 0)
1001 			return (ENXIO);
1002 		DELAY(1);	/* limit looping */
1003 	}
1004 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1005 
1006 	return (0);
1007 }
1008 
1009 #if 0
1010 static int
1011 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1012 {
1013 	int count = PCI_VPD_TIMEOUT;
1014 
1015 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1016 
1017 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1018 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1019 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1020 		if (--count < 0)
1021 			return (ENXIO);
1022 		DELAY(1);	/* limit looping */
1023 	}
1024 
1025 	return (0);
1026 }
1027 #endif
1028 
1029 #undef PCI_VPD_TIMEOUT
1030 
1031 struct vpd_readstate {
1032 	device_t	pcib;
1033 	pcicfgregs	*cfg;
1034 	uint32_t	val;
1035 	int		bytesinval;
1036 	int		off;
1037 	uint8_t		cksum;
1038 };
1039 
1040 static int
1041 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1042 {
1043 	uint32_t reg;
1044 	uint8_t byte;
1045 
1046 	if (vrs->bytesinval == 0) {
1047 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1048 			return (ENXIO);
1049 		vrs->val = le32toh(reg);
1050 		vrs->off += 4;
1051 		byte = vrs->val & 0xff;
1052 		vrs->bytesinval = 3;
1053 	} else {
1054 		vrs->val = vrs->val >> 8;
1055 		byte = vrs->val & 0xff;
1056 		vrs->bytesinval--;
1057 	}
1058 
1059 	vrs->cksum += byte;
1060 	*data = byte;
1061 	return (0);
1062 }
1063 
1064 static void
1065 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1066 {
1067 	struct vpd_readstate vrs;
1068 	int state;
1069 	int name;
1070 	int remain;
1071 	int i;
1072 	int alloc, off;		/* alloc/off for RO/W arrays */
1073 	int cksumvalid;
1074 	int dflen;
1075 	uint8_t byte;
1076 	uint8_t byte2;
1077 
1078 	/* init vpd reader */
1079 	vrs.bytesinval = 0;
1080 	vrs.off = 0;
1081 	vrs.pcib = pcib;
1082 	vrs.cfg = cfg;
1083 	vrs.cksum = 0;
1084 
1085 	state = 0;
1086 	name = remain = i = 0;	/* shut up stupid gcc */
1087 	alloc = off = 0;	/* shut up stupid gcc */
1088 	dflen = 0;		/* shut up stupid gcc */
1089 	cksumvalid = -1;
1090 	while (state >= 0) {
1091 		if (vpd_nextbyte(&vrs, &byte)) {
1092 			state = -2;
1093 			break;
1094 		}
1095 #if 0
1096 		printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1097 		    "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1098 		    vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1099 #endif
1100 		switch (state) {
1101 		case 0:		/* item name */
1102 			if (byte & 0x80) {
1103 				if (vpd_nextbyte(&vrs, &byte2)) {
1104 					state = -2;
1105 					break;
1106 				}
1107 				remain = byte2;
1108 				if (vpd_nextbyte(&vrs, &byte2)) {
1109 					state = -2;
1110 					break;
1111 				}
1112 				remain |= byte2 << 8;
1113 				if (remain > (0x7f*4 - vrs.off)) {
1114 					state = -1;
1115 					pci_printf(cfg,
1116 					    "invalid VPD data, remain %#x\n",
1117 					    remain);
1118 				}
1119 				name = byte & 0x7f;
1120 			} else {
1121 				remain = byte & 0x7;
1122 				name = (byte >> 3) & 0xf;
1123 			}
1124 			switch (name) {
1125 			case 0x2:	/* String */
1126 				cfg->vpd.vpd_ident = malloc(remain + 1,
1127 				    M_DEVBUF, M_WAITOK);
1128 				i = 0;
1129 				state = 1;
1130 				break;
1131 			case 0xf:	/* End */
1132 				state = -1;
1133 				break;
1134 			case 0x10:	/* VPD-R */
1135 				alloc = 8;
1136 				off = 0;
1137 				cfg->vpd.vpd_ros = malloc(alloc *
1138 				    sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1139 				    M_WAITOK | M_ZERO);
1140 				state = 2;
1141 				break;
1142 			case 0x11:	/* VPD-W */
1143 				alloc = 8;
1144 				off = 0;
1145 				cfg->vpd.vpd_w = malloc(alloc *
1146 				    sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1147 				    M_WAITOK | M_ZERO);
1148 				state = 5;
1149 				break;
1150 			default:	/* Invalid data, abort */
1151 				state = -1;
1152 				break;
1153 			}
1154 			break;
1155 
1156 		case 1:	/* Identifier String */
1157 			cfg->vpd.vpd_ident[i++] = byte;
1158 			remain--;
1159 			if (remain == 0)  {
1160 				cfg->vpd.vpd_ident[i] = '\0';
1161 				state = 0;
1162 			}
1163 			break;
1164 
1165 		case 2:	/* VPD-R Keyword Header */
1166 			if (off == alloc) {
1167 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1168 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1169 				    M_DEVBUF, M_WAITOK | M_ZERO);
1170 			}
1171 			cfg->vpd.vpd_ros[off].keyword[0] = byte;
1172 			if (vpd_nextbyte(&vrs, &byte2)) {
1173 				state = -2;
1174 				break;
1175 			}
1176 			cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1177 			if (vpd_nextbyte(&vrs, &byte2)) {
1178 				state = -2;
1179 				break;
1180 			}
1181 			cfg->vpd.vpd_ros[off].len = dflen = byte2;
1182 			if (dflen == 0 &&
1183 			    strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1184 			    2) == 0) {
1185 				/*
1186 				 * if this happens, we can't trust the rest
1187 				 * of the VPD.
1188 				 */
1189 				pci_printf(cfg, "bad keyword length: %d\n",
1190 				    dflen);
1191 				cksumvalid = 0;
1192 				state = -1;
1193 				break;
1194 			} else if (dflen == 0) {
1195 				cfg->vpd.vpd_ros[off].value = malloc(1 *
1196 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1197 				    M_DEVBUF, M_WAITOK);
1198 				cfg->vpd.vpd_ros[off].value[0] = '\x00';
1199 			} else
1200 				cfg->vpd.vpd_ros[off].value = malloc(
1201 				    (dflen + 1) *
1202 				    sizeof(*cfg->vpd.vpd_ros[off].value),
1203 				    M_DEVBUF, M_WAITOK);
1204 			remain -= 3;
1205 			i = 0;
1206 			/* keep in sync w/ state 3's transistions */
1207 			if (dflen == 0 && remain == 0)
1208 				state = 0;
1209 			else if (dflen == 0)
1210 				state = 2;
1211 			else
1212 				state = 3;
1213 			break;
1214 
1215 		case 3:	/* VPD-R Keyword Value */
1216 			cfg->vpd.vpd_ros[off].value[i++] = byte;
1217 			if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1218 			    "RV", 2) == 0 && cksumvalid == -1) {
1219 				if (vrs.cksum == 0)
1220 					cksumvalid = 1;
1221 				else {
1222 					if (bootverbose)
1223 						pci_printf(cfg,
1224 					    "bad VPD cksum, remain %hhu\n",
1225 						    vrs.cksum);
1226 					cksumvalid = 0;
1227 					state = -1;
1228 					break;
1229 				}
1230 			}
1231 			dflen--;
1232 			remain--;
1233 			/* keep in sync w/ state 2's transistions */
1234 			if (dflen == 0)
1235 				cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1236 			if (dflen == 0 && remain == 0) {
1237 				cfg->vpd.vpd_rocnt = off;
1238 				cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1239 				    off * sizeof(*cfg->vpd.vpd_ros),
1240 				    M_DEVBUF, M_WAITOK | M_ZERO);
1241 				state = 0;
1242 			} else if (dflen == 0)
1243 				state = 2;
1244 			break;
1245 
1246 		case 4:
1247 			remain--;
1248 			if (remain == 0)
1249 				state = 0;
1250 			break;
1251 
1252 		case 5:	/* VPD-W Keyword Header */
1253 			if (off == alloc) {
1254 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1255 				    (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1256 				    M_DEVBUF, M_WAITOK | M_ZERO);
1257 			}
1258 			cfg->vpd.vpd_w[off].keyword[0] = byte;
1259 			if (vpd_nextbyte(&vrs, &byte2)) {
1260 				state = -2;
1261 				break;
1262 			}
1263 			cfg->vpd.vpd_w[off].keyword[1] = byte2;
1264 			if (vpd_nextbyte(&vrs, &byte2)) {
1265 				state = -2;
1266 				break;
1267 			}
1268 			cfg->vpd.vpd_w[off].len = dflen = byte2;
1269 			cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1270 			cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1271 			    sizeof(*cfg->vpd.vpd_w[off].value),
1272 			    M_DEVBUF, M_WAITOK);
1273 			remain -= 3;
1274 			i = 0;
1275 			/* keep in sync w/ state 6's transistions */
1276 			if (dflen == 0 && remain == 0)
1277 				state = 0;
1278 			else if (dflen == 0)
1279 				state = 5;
1280 			else
1281 				state = 6;
1282 			break;
1283 
1284 		case 6:	/* VPD-W Keyword Value */
1285 			cfg->vpd.vpd_w[off].value[i++] = byte;
1286 			dflen--;
1287 			remain--;
1288 			/* keep in sync w/ state 5's transistions */
1289 			if (dflen == 0)
1290 				cfg->vpd.vpd_w[off++].value[i++] = '\0';
1291 			if (dflen == 0 && remain == 0) {
1292 				cfg->vpd.vpd_wcnt = off;
1293 				cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1294 				    off * sizeof(*cfg->vpd.vpd_w),
1295 				    M_DEVBUF, M_WAITOK | M_ZERO);
1296 				state = 0;
1297 			} else if (dflen == 0)
1298 				state = 5;
1299 			break;
1300 
1301 		default:
1302 			pci_printf(cfg, "invalid state: %d\n", state);
1303 			state = -1;
1304 			break;
1305 		}
1306 	}
1307 
1308 	if (cksumvalid == 0 || state < -1) {
1309 		/* read-only data bad, clean up */
1310 		if (cfg->vpd.vpd_ros != NULL) {
1311 			for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1312 				free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1313 			free(cfg->vpd.vpd_ros, M_DEVBUF);
1314 			cfg->vpd.vpd_ros = NULL;
1315 		}
1316 	}
1317 	if (state < -1) {
1318 		/* I/O error, clean up */
1319 		pci_printf(cfg, "failed to read VPD data.\n");
1320 		if (cfg->vpd.vpd_ident != NULL) {
1321 			free(cfg->vpd.vpd_ident, M_DEVBUF);
1322 			cfg->vpd.vpd_ident = NULL;
1323 		}
1324 		if (cfg->vpd.vpd_w != NULL) {
1325 			for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1326 				free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1327 			free(cfg->vpd.vpd_w, M_DEVBUF);
1328 			cfg->vpd.vpd_w = NULL;
1329 		}
1330 	}
1331 	cfg->vpd.vpd_cached = 1;
1332 #undef REG
1333 #undef WREG
1334 }
1335 
1336 int
1337 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1338 {
1339 	struct pci_devinfo *dinfo = device_get_ivars(child);
1340 	pcicfgregs *cfg = &dinfo->cfg;
1341 
1342 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1343 		pci_read_vpd(device_get_parent(dev), cfg);
1344 
1345 	*identptr = cfg->vpd.vpd_ident;
1346 
1347 	if (*identptr == NULL)
1348 		return (ENXIO);
1349 
1350 	return (0);
1351 }
1352 
1353 int
1354 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1355 	const char **vptr)
1356 {
1357 	struct pci_devinfo *dinfo = device_get_ivars(child);
1358 	pcicfgregs *cfg = &dinfo->cfg;
1359 	int i;
1360 
1361 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1362 		pci_read_vpd(device_get_parent(dev), cfg);
1363 
1364 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1365 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1366 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1367 			*vptr = cfg->vpd.vpd_ros[i].value;
1368 			return (0);
1369 		}
1370 
1371 	*vptr = NULL;
1372 	return (ENXIO);
1373 }
1374 
1375 struct pcicfg_vpd *
1376 pci_fetch_vpd_list(device_t dev)
1377 {
1378 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1379 	pcicfgregs *cfg = &dinfo->cfg;
1380 
1381 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1382 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1383 	return (&cfg->vpd);
1384 }
1385 
1386 /*
1387  * Find the requested HyperTransport capability and return the offset
1388  * in configuration space via the pointer provided.  The function
1389  * returns 0 on success and an error code otherwise.
1390  */
1391 int
1392 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1393 {
1394 	int ptr, error;
1395 	uint16_t val;
1396 
1397 	error = pci_find_cap(child, PCIY_HT, &ptr);
1398 	if (error)
1399 		return (error);
1400 
1401 	/*
1402 	 * Traverse the capabilities list checking each HT capability
1403 	 * to see if it matches the requested HT capability.
1404 	 */
1405 	for (;;) {
1406 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1407 		if (capability == PCIM_HTCAP_SLAVE ||
1408 		    capability == PCIM_HTCAP_HOST)
1409 			val &= 0xe000;
1410 		else
1411 			val &= PCIM_HTCMD_CAP_MASK;
1412 		if (val == capability) {
1413 			if (capreg != NULL)
1414 				*capreg = ptr;
1415 			return (0);
1416 		}
1417 
1418 		/* Skip to the next HT capability. */
1419 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1420 			break;
1421 	}
1422 
1423 	return (ENOENT);
1424 }
1425 
1426 /*
1427  * Find the next requested HyperTransport capability after start and return
1428  * the offset in configuration space via the pointer provided.  The function
1429  * returns 0 on success and an error code otherwise.
1430  */
1431 int
1432 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1433     int start, int *capreg)
1434 {
1435 	int ptr;
1436 	uint16_t val;
1437 
1438 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1439 	    ("start capability is not HyperTransport capability"));
1440 	ptr = start;
1441 
1442 	/*
1443 	 * Traverse the capabilities list checking each HT capability
1444 	 * to see if it matches the requested HT capability.
1445 	 */
1446 	for (;;) {
1447 		/* Skip to the next HT capability. */
1448 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1449 			break;
1450 
1451 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1452 		if (capability == PCIM_HTCAP_SLAVE ||
1453 		    capability == PCIM_HTCAP_HOST)
1454 			val &= 0xe000;
1455 		else
1456 			val &= PCIM_HTCMD_CAP_MASK;
1457 		if (val == capability) {
1458 			if (capreg != NULL)
1459 				*capreg = ptr;
1460 			return (0);
1461 		}
1462 	}
1463 
1464 	return (ENOENT);
1465 }
1466 
1467 /*
1468  * Find the requested capability and return the offset in
1469  * configuration space via the pointer provided.  The function returns
1470  * 0 on success and an error code otherwise.
1471  */
1472 int
1473 pci_find_cap_method(device_t dev, device_t child, int capability,
1474     int *capreg)
1475 {
1476 	struct pci_devinfo *dinfo = device_get_ivars(child);
1477 	pcicfgregs *cfg = &dinfo->cfg;
1478 	uint32_t status;
1479 	uint8_t ptr;
1480 
1481 	/*
1482 	 * Check the CAP_LIST bit of the PCI status register first.
1483 	 */
1484 	status = pci_read_config(child, PCIR_STATUS, 2);
1485 	if (!(status & PCIM_STATUS_CAPPRESENT))
1486 		return (ENXIO);
1487 
1488 	/*
1489 	 * Determine the start pointer of the capabilities list.
1490 	 */
1491 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1492 	case PCIM_HDRTYPE_NORMAL:
1493 	case PCIM_HDRTYPE_BRIDGE:
1494 		ptr = PCIR_CAP_PTR;
1495 		break;
1496 	case PCIM_HDRTYPE_CARDBUS:
1497 		ptr = PCIR_CAP_PTR_2;
1498 		break;
1499 	default:
1500 		/* XXX: panic? */
1501 		return (ENXIO);		/* no extended capabilities support */
1502 	}
1503 	ptr = pci_read_config(child, ptr, 1);
1504 
1505 	/*
1506 	 * Traverse the capabilities list.
1507 	 */
1508 	while (ptr != 0) {
1509 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1510 			if (capreg != NULL)
1511 				*capreg = ptr;
1512 			return (0);
1513 		}
1514 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1515 	}
1516 
1517 	return (ENOENT);
1518 }
1519 
1520 /*
1521  * Find the next requested capability after start and return the offset in
1522  * configuration space via the pointer provided.  The function returns
1523  * 0 on success and an error code otherwise.
1524  */
1525 int
1526 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1527     int start, int *capreg)
1528 {
1529 	uint8_t ptr;
1530 
1531 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1532 	    ("start capability is not expected capability"));
1533 
1534 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1535 	while (ptr != 0) {
1536 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1537 			if (capreg != NULL)
1538 				*capreg = ptr;
1539 			return (0);
1540 		}
1541 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1542 	}
1543 
1544 	return (ENOENT);
1545 }
1546 
1547 /*
1548  * Find the requested extended capability and return the offset in
1549  * configuration space via the pointer provided.  The function returns
1550  * 0 on success and an error code otherwise.
1551  */
1552 int
1553 pci_find_extcap_method(device_t dev, device_t child, int capability,
1554     int *capreg)
1555 {
1556 	struct pci_devinfo *dinfo = device_get_ivars(child);
1557 	pcicfgregs *cfg = &dinfo->cfg;
1558 	uint32_t ecap;
1559 	uint16_t ptr;
1560 
1561 	/* Only supported for PCI-express devices. */
1562 	if (cfg->pcie.pcie_location == 0)
1563 		return (ENXIO);
1564 
1565 	ptr = PCIR_EXTCAP;
1566 	ecap = pci_read_config(child, ptr, 4);
1567 	if (ecap == 0xffffffff || ecap == 0)
1568 		return (ENOENT);
1569 	for (;;) {
1570 		if (PCI_EXTCAP_ID(ecap) == capability) {
1571 			if (capreg != NULL)
1572 				*capreg = ptr;
1573 			return (0);
1574 		}
1575 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1576 		if (ptr == 0)
1577 			break;
1578 		ecap = pci_read_config(child, ptr, 4);
1579 	}
1580 
1581 	return (ENOENT);
1582 }
1583 
1584 /*
1585  * Find the next requested extended capability after start and return the
1586  * offset in configuration space via the pointer provided.  The function
1587  * returns 0 on success and an error code otherwise.
1588  */
1589 int
1590 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1591     int start, int *capreg)
1592 {
1593 	struct pci_devinfo *dinfo = device_get_ivars(child);
1594 	pcicfgregs *cfg = &dinfo->cfg;
1595 	uint32_t ecap;
1596 	uint16_t ptr;
1597 
1598 	/* Only supported for PCI-express devices. */
1599 	if (cfg->pcie.pcie_location == 0)
1600 		return (ENXIO);
1601 
1602 	ecap = pci_read_config(child, start, 4);
1603 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1604 	    ("start extended capability is not expected capability"));
1605 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1606 	while (ptr != 0) {
1607 		ecap = pci_read_config(child, ptr, 4);
1608 		if (PCI_EXTCAP_ID(ecap) == capability) {
1609 			if (capreg != NULL)
1610 				*capreg = ptr;
1611 			return (0);
1612 		}
1613 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1614 	}
1615 
1616 	return (ENOENT);
1617 }
1618 
1619 /*
1620  * Support for MSI-X message interrupts.
1621  */
1622 static void
1623 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1624 {
1625 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1626 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1627 	uint32_t offset;
1628 
1629 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1630 	offset = msix->msix_table_offset + index * 16;
1631 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1632 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1633 	bus_write_4(msix->msix_table_res, offset + 8, data);
1634 }
1635 
1636 void
1637 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1638     uint64_t address, uint32_t data)
1639 {
1640 
1641 	if (pci_msix_rewrite_table) {
1642 		struct pci_devinfo *dinfo = device_get_ivars(child);
1643 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1644 
1645 		/*
1646 		 * Some VM hosts require MSIX to be disabled in the
1647 		 * control register before updating the MSIX table
1648 		 * entries are allowed. It is not enough to only
1649 		 * disable MSIX while updating a single entry. MSIX
1650 		 * must be disabled while updating all entries in the
1651 		 * table.
1652 		 */
1653 		pci_write_config(child,
1654 		    msix->msix_location + PCIR_MSIX_CTRL,
1655 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1656 		pci_resume_msix(child);
1657 	} else
1658 		pci_write_msix_entry(child, index, address, data);
1659 
1660 	/* Enable MSI -> HT mapping. */
1661 	pci_ht_map_msi(child, address);
1662 }
1663 
1664 void
1665 pci_mask_msix(device_t dev, u_int index)
1666 {
1667 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1668 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1669 	uint32_t offset, val;
1670 
1671 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1672 	offset = msix->msix_table_offset + index * 16 + 12;
1673 	val = bus_read_4(msix->msix_table_res, offset);
1674 	val |= PCIM_MSIX_VCTRL_MASK;
1675 
1676 	/*
1677 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1678 	 * register, so always write the new value.
1679 	 */
1680 	bus_write_4(msix->msix_table_res, offset, val);
1681 }
1682 
1683 void
1684 pci_unmask_msix(device_t dev, u_int index)
1685 {
1686 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1687 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1688 	uint32_t offset, val;
1689 
1690 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1691 	offset = msix->msix_table_offset + index * 16 + 12;
1692 	val = bus_read_4(msix->msix_table_res, offset);
1693 	val &= ~PCIM_MSIX_VCTRL_MASK;
1694 
1695 	/*
1696 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1697 	 * register, so always write the new value.
1698 	 */
1699 	bus_write_4(msix->msix_table_res, offset, val);
1700 }
1701 
1702 int
1703 pci_pending_msix(device_t dev, u_int index)
1704 {
1705 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1706 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1707 	uint32_t offset, bit;
1708 
1709 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1710 	offset = msix->msix_pba_offset + (index / 32) * 4;
1711 	bit = 1 << index % 32;
1712 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1713 }
1714 
1715 /*
1716  * Restore MSI-X registers and table during resume.  If MSI-X is
1717  * enabled then walk the virtual table to restore the actual MSI-X
1718  * table.
1719  */
1720 static void
1721 pci_resume_msix(device_t dev)
1722 {
1723 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1724 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1725 	struct msix_table_entry *mte;
1726 	struct msix_vector *mv;
1727 	int i;
1728 
1729 	if (msix->msix_alloc > 0) {
1730 		/* First, mask all vectors. */
1731 		for (i = 0; i < msix->msix_msgnum; i++)
1732 			pci_mask_msix(dev, i);
1733 
1734 		/* Second, program any messages with at least one handler. */
1735 		for (i = 0; i < msix->msix_table_len; i++) {
1736 			mte = &msix->msix_table[i];
1737 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1738 				continue;
1739 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1740 			pci_write_msix_entry(dev, i, mv->mv_address,
1741 			    mv->mv_data);
1742 			pci_unmask_msix(dev, i);
1743 		}
1744 	}
1745 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1746 	    msix->msix_ctrl, 2);
1747 }
1748 
1749 /*
1750  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1751  * returned in *count.  After this function returns, each message will be
1752  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1753  */
1754 int
1755 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1756 {
1757 	struct pci_devinfo *dinfo = device_get_ivars(child);
1758 	pcicfgregs *cfg = &dinfo->cfg;
1759 	struct resource_list_entry *rle;
1760 	int actual, error, i, irq, max;
1761 
1762 	/* Don't let count == 0 get us into trouble. */
1763 	if (*count == 0)
1764 		return (EINVAL);
1765 
1766 	/* If rid 0 is allocated, then fail. */
1767 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1768 	if (rle != NULL && rle->res != NULL)
1769 		return (ENXIO);
1770 
1771 	/* Already have allocated messages? */
1772 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1773 		return (ENXIO);
1774 
1775 	/* If MSI-X is blacklisted for this system, fail. */
1776 	if (pci_msix_blacklisted())
1777 		return (ENXIO);
1778 
1779 	/* MSI-X capability present? */
1780 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1781 		return (ENODEV);
1782 
1783 	/* Make sure the appropriate BARs are mapped. */
1784 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1785 	    cfg->msix.msix_table_bar);
1786 	if (rle == NULL || rle->res == NULL ||
1787 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1788 		return (ENXIO);
1789 	cfg->msix.msix_table_res = rle->res;
1790 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1791 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1792 		    cfg->msix.msix_pba_bar);
1793 		if (rle == NULL || rle->res == NULL ||
1794 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1795 			return (ENXIO);
1796 	}
1797 	cfg->msix.msix_pba_res = rle->res;
1798 
1799 	if (bootverbose)
1800 		device_printf(child,
1801 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1802 		    *count, cfg->msix.msix_msgnum);
1803 	max = min(*count, cfg->msix.msix_msgnum);
1804 	for (i = 0; i < max; i++) {
1805 		/* Allocate a message. */
1806 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1807 		if (error) {
1808 			if (i == 0)
1809 				return (error);
1810 			break;
1811 		}
1812 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1813 		    irq, 1);
1814 	}
1815 	actual = i;
1816 
1817 	if (bootverbose) {
1818 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1819 		if (actual == 1)
1820 			device_printf(child, "using IRQ %ju for MSI-X\n",
1821 			    rle->start);
1822 		else {
1823 			int run;
1824 
1825 			/*
1826 			 * Be fancy and try to print contiguous runs of
1827 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1828 			 * 'run' is true if we are in a range.
1829 			 */
1830 			device_printf(child, "using IRQs %ju", rle->start);
1831 			irq = rle->start;
1832 			run = 0;
1833 			for (i = 1; i < actual; i++) {
1834 				rle = resource_list_find(&dinfo->resources,
1835 				    SYS_RES_IRQ, i + 1);
1836 
1837 				/* Still in a run? */
1838 				if (rle->start == irq + 1) {
1839 					run = 1;
1840 					irq++;
1841 					continue;
1842 				}
1843 
1844 				/* Finish previous range. */
1845 				if (run) {
1846 					printf("-%d", irq);
1847 					run = 0;
1848 				}
1849 
1850 				/* Start new range. */
1851 				printf(",%ju", rle->start);
1852 				irq = rle->start;
1853 			}
1854 
1855 			/* Unfinished range? */
1856 			if (run)
1857 				printf("-%d", irq);
1858 			printf(" for MSI-X\n");
1859 		}
1860 	}
1861 
1862 	/* Mask all vectors. */
1863 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1864 		pci_mask_msix(child, i);
1865 
1866 	/* Allocate and initialize vector data and virtual table. */
1867 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1868 	    M_DEVBUF, M_WAITOK | M_ZERO);
1869 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1870 	    M_DEVBUF, M_WAITOK | M_ZERO);
1871 	for (i = 0; i < actual; i++) {
1872 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1873 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1874 		cfg->msix.msix_table[i].mte_vector = i + 1;
1875 	}
1876 
1877 	/* Update control register to enable MSI-X. */
1878 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1879 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1880 	    cfg->msix.msix_ctrl, 2);
1881 
1882 	/* Update counts of alloc'd messages. */
1883 	cfg->msix.msix_alloc = actual;
1884 	cfg->msix.msix_table_len = actual;
1885 	*count = actual;
1886 	return (0);
1887 }
1888 
1889 /*
1890  * By default, pci_alloc_msix() will assign the allocated IRQ
1891  * resources consecutively to the first N messages in the MSI-X table.
1892  * However, device drivers may want to use different layouts if they
1893  * either receive fewer messages than they asked for, or they wish to
1894  * populate the MSI-X table sparsely.  This method allows the driver
1895  * to specify what layout it wants.  It must be called after a
1896  * successful pci_alloc_msix() but before any of the associated
1897  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1898  *
1899  * The 'vectors' array contains 'count' message vectors.  The array
1900  * maps directly to the MSI-X table in that index 0 in the array
1901  * specifies the vector for the first message in the MSI-X table, etc.
1902  * The vector value in each array index can either be 0 to indicate
1903  * that no vector should be assigned to a message slot, or it can be a
1904  * number from 1 to N (where N is the count returned from a
1905  * succcessful call to pci_alloc_msix()) to indicate which message
1906  * vector (IRQ) to be used for the corresponding message.
1907  *
1908  * On successful return, each message with a non-zero vector will have
1909  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1910  * 1.  Additionally, if any of the IRQs allocated via the previous
1911  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1912  * will be freed back to the system automatically.
1913  *
1914  * For example, suppose a driver has a MSI-X table with 6 messages and
1915  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1916  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1917  * C.  After the call to pci_alloc_msix(), the device will be setup to
1918  * have an MSI-X table of ABC--- (where - means no vector assigned).
1919  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1920  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1921  * be freed back to the system.  This device will also have valid
1922  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1923  *
1924  * In any case, the SYS_RES_IRQ rid X will always map to the message
1925  * at MSI-X table index X - 1 and will only be valid if a vector is
1926  * assigned to that table entry.
1927  */
1928 int
1929 pci_remap_msix_method(device_t dev, device_t child, int count,
1930     const u_int *vectors)
1931 {
1932 	struct pci_devinfo *dinfo = device_get_ivars(child);
1933 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1934 	struct resource_list_entry *rle;
1935 	int i, irq, j, *used;
1936 
1937 	/*
1938 	 * Have to have at least one message in the table but the
1939 	 * table can't be bigger than the actual MSI-X table in the
1940 	 * device.
1941 	 */
1942 	if (count == 0 || count > msix->msix_msgnum)
1943 		return (EINVAL);
1944 
1945 	/* Sanity check the vectors. */
1946 	for (i = 0; i < count; i++)
1947 		if (vectors[i] > msix->msix_alloc)
1948 			return (EINVAL);
1949 
1950 	/*
1951 	 * Make sure there aren't any holes in the vectors to be used.
1952 	 * It's a big pain to support it, and it doesn't really make
1953 	 * sense anyway.  Also, at least one vector must be used.
1954 	 */
1955 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1956 	    M_ZERO);
1957 	for (i = 0; i < count; i++)
1958 		if (vectors[i] != 0)
1959 			used[vectors[i] - 1] = 1;
1960 	for (i = 0; i < msix->msix_alloc - 1; i++)
1961 		if (used[i] == 0 && used[i + 1] == 1) {
1962 			free(used, M_DEVBUF);
1963 			return (EINVAL);
1964 		}
1965 	if (used[0] != 1) {
1966 		free(used, M_DEVBUF);
1967 		return (EINVAL);
1968 	}
1969 
1970 	/* Make sure none of the resources are allocated. */
1971 	for (i = 0; i < msix->msix_table_len; i++) {
1972 		if (msix->msix_table[i].mte_vector == 0)
1973 			continue;
1974 		if (msix->msix_table[i].mte_handlers > 0) {
1975 			free(used, M_DEVBUF);
1976 			return (EBUSY);
1977 		}
1978 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1979 		KASSERT(rle != NULL, ("missing resource"));
1980 		if (rle->res != NULL) {
1981 			free(used, M_DEVBUF);
1982 			return (EBUSY);
1983 		}
1984 	}
1985 
1986 	/* Free the existing resource list entries. */
1987 	for (i = 0; i < msix->msix_table_len; i++) {
1988 		if (msix->msix_table[i].mte_vector == 0)
1989 			continue;
1990 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1991 	}
1992 
1993 	/*
1994 	 * Build the new virtual table keeping track of which vectors are
1995 	 * used.
1996 	 */
1997 	free(msix->msix_table, M_DEVBUF);
1998 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1999 	    M_DEVBUF, M_WAITOK | M_ZERO);
2000 	for (i = 0; i < count; i++)
2001 		msix->msix_table[i].mte_vector = vectors[i];
2002 	msix->msix_table_len = count;
2003 
2004 	/* Free any unused IRQs and resize the vectors array if necessary. */
2005 	j = msix->msix_alloc - 1;
2006 	if (used[j] == 0) {
2007 		struct msix_vector *vec;
2008 
2009 		while (used[j] == 0) {
2010 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2011 			    msix->msix_vectors[j].mv_irq);
2012 			j--;
2013 		}
2014 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2015 		    M_WAITOK);
2016 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2017 		    (j + 1));
2018 		free(msix->msix_vectors, M_DEVBUF);
2019 		msix->msix_vectors = vec;
2020 		msix->msix_alloc = j + 1;
2021 	}
2022 	free(used, M_DEVBUF);
2023 
2024 	/* Map the IRQs onto the rids. */
2025 	for (i = 0; i < count; i++) {
2026 		if (vectors[i] == 0)
2027 			continue;
2028 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2029 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2030 		    irq, 1);
2031 	}
2032 
2033 	if (bootverbose) {
2034 		device_printf(child, "Remapped MSI-X IRQs as: ");
2035 		for (i = 0; i < count; i++) {
2036 			if (i != 0)
2037 				printf(", ");
2038 			if (vectors[i] == 0)
2039 				printf("---");
2040 			else
2041 				printf("%d",
2042 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2043 		}
2044 		printf("\n");
2045 	}
2046 
2047 	return (0);
2048 }
2049 
2050 static int
2051 pci_release_msix(device_t dev, device_t child)
2052 {
2053 	struct pci_devinfo *dinfo = device_get_ivars(child);
2054 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2055 	struct resource_list_entry *rle;
2056 	int i;
2057 
2058 	/* Do we have any messages to release? */
2059 	if (msix->msix_alloc == 0)
2060 		return (ENODEV);
2061 
2062 	/* Make sure none of the resources are allocated. */
2063 	for (i = 0; i < msix->msix_table_len; i++) {
2064 		if (msix->msix_table[i].mte_vector == 0)
2065 			continue;
2066 		if (msix->msix_table[i].mte_handlers > 0)
2067 			return (EBUSY);
2068 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2069 		KASSERT(rle != NULL, ("missing resource"));
2070 		if (rle->res != NULL)
2071 			return (EBUSY);
2072 	}
2073 
2074 	/* Update control register to disable MSI-X. */
2075 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2076 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2077 	    msix->msix_ctrl, 2);
2078 
2079 	/* Free the resource list entries. */
2080 	for (i = 0; i < msix->msix_table_len; i++) {
2081 		if (msix->msix_table[i].mte_vector == 0)
2082 			continue;
2083 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2084 	}
2085 	free(msix->msix_table, M_DEVBUF);
2086 	msix->msix_table_len = 0;
2087 
2088 	/* Release the IRQs. */
2089 	for (i = 0; i < msix->msix_alloc; i++)
2090 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2091 		    msix->msix_vectors[i].mv_irq);
2092 	free(msix->msix_vectors, M_DEVBUF);
2093 	msix->msix_alloc = 0;
2094 	return (0);
2095 }
2096 
2097 /*
2098  * Return the max supported MSI-X messages this device supports.
2099  * Basically, assuming the MD code can alloc messages, this function
2100  * should return the maximum value that pci_alloc_msix() can return.
2101  * Thus, it is subject to the tunables, etc.
2102  */
2103 int
2104 pci_msix_count_method(device_t dev, device_t child)
2105 {
2106 	struct pci_devinfo *dinfo = device_get_ivars(child);
2107 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2108 
2109 	if (pci_do_msix && msix->msix_location != 0)
2110 		return (msix->msix_msgnum);
2111 	return (0);
2112 }
2113 
2114 int
2115 pci_msix_pba_bar_method(device_t dev, device_t child)
2116 {
2117 	struct pci_devinfo *dinfo = device_get_ivars(child);
2118 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2119 
2120 	if (pci_do_msix && msix->msix_location != 0)
2121 		return (msix->msix_pba_bar);
2122 	return (-1);
2123 }
2124 
2125 int
2126 pci_msix_table_bar_method(device_t dev, device_t child)
2127 {
2128 	struct pci_devinfo *dinfo = device_get_ivars(child);
2129 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2130 
2131 	if (pci_do_msix && msix->msix_location != 0)
2132 		return (msix->msix_table_bar);
2133 	return (-1);
2134 }
2135 
2136 /*
2137  * HyperTransport MSI mapping control
2138  */
2139 void
2140 pci_ht_map_msi(device_t dev, uint64_t addr)
2141 {
2142 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2143 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2144 
2145 	if (!ht->ht_msimap)
2146 		return;
2147 
2148 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2149 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2150 		/* Enable MSI -> HT mapping. */
2151 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2152 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2153 		    ht->ht_msictrl, 2);
2154 	}
2155 
2156 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2157 		/* Disable MSI -> HT mapping. */
2158 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2159 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2160 		    ht->ht_msictrl, 2);
2161 	}
2162 }
2163 
2164 int
2165 pci_get_max_payload(device_t dev)
2166 {
2167 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2168 	int cap;
2169 	uint16_t val;
2170 
2171 	cap = dinfo->cfg.pcie.pcie_location;
2172 	if (cap == 0)
2173 		return (0);
2174 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2175 	val &= PCIEM_CTL_MAX_PAYLOAD;
2176 	val >>= 5;
2177 	return (1 << (val + 7));
2178 }
2179 
2180 int
2181 pci_get_max_read_req(device_t dev)
2182 {
2183 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2184 	int cap;
2185 	uint16_t val;
2186 
2187 	cap = dinfo->cfg.pcie.pcie_location;
2188 	if (cap == 0)
2189 		return (0);
2190 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2191 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2192 	val >>= 12;
2193 	return (1 << (val + 7));
2194 }
2195 
2196 int
2197 pci_set_max_read_req(device_t dev, int size)
2198 {
2199 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2200 	int cap;
2201 	uint16_t val;
2202 
2203 	cap = dinfo->cfg.pcie.pcie_location;
2204 	if (cap == 0)
2205 		return (0);
2206 	if (size < 128)
2207 		size = 128;
2208 	if (size > 4096)
2209 		size = 4096;
2210 	size = (1 << (fls(size) - 1));
2211 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2212 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2213 	val |= (fls(size) - 8) << 12;
2214 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2215 	return (size);
2216 }
2217 
2218 uint32_t
2219 pcie_read_config(device_t dev, int reg, int width)
2220 {
2221 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2222 	int cap;
2223 
2224 	cap = dinfo->cfg.pcie.pcie_location;
2225 	if (cap == 0) {
2226 		if (width == 2)
2227 			return (0xffff);
2228 		return (0xffffffff);
2229 	}
2230 
2231 	return (pci_read_config(dev, cap + reg, width));
2232 }
2233 
2234 void
2235 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2236 {
2237 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2238 	int cap;
2239 
2240 	cap = dinfo->cfg.pcie.pcie_location;
2241 	if (cap == 0)
2242 		return;
2243 	pci_write_config(dev, cap + reg, value, width);
2244 }
2245 
2246 /*
2247  * Adjusts a PCI-e capability register by clearing the bits in mask
2248  * and setting the bits in (value & mask).  Bits not set in mask are
2249  * not adjusted.
2250  *
2251  * Returns the old value on success or all ones on failure.
2252  */
2253 uint32_t
2254 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2255     int width)
2256 {
2257 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2258 	uint32_t old, new;
2259 	int cap;
2260 
2261 	cap = dinfo->cfg.pcie.pcie_location;
2262 	if (cap == 0) {
2263 		if (width == 2)
2264 			return (0xffff);
2265 		return (0xffffffff);
2266 	}
2267 
2268 	old = pci_read_config(dev, cap + reg, width);
2269 	new = old & ~mask;
2270 	new |= (value & mask);
2271 	pci_write_config(dev, cap + reg, new, width);
2272 	return (old);
2273 }
2274 
2275 /*
2276  * Support for MSI message signalled interrupts.
2277  */
2278 void
2279 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2280     uint16_t data)
2281 {
2282 	struct pci_devinfo *dinfo = device_get_ivars(child);
2283 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2284 
2285 	/* Write data and address values. */
2286 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2287 	    address & 0xffffffff, 4);
2288 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2289 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2290 		    address >> 32, 4);
2291 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2292 		    data, 2);
2293 	} else
2294 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2295 		    2);
2296 
2297 	/* Enable MSI in the control register. */
2298 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2299 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2300 	    msi->msi_ctrl, 2);
2301 
2302 	/* Enable MSI -> HT mapping. */
2303 	pci_ht_map_msi(child, address);
2304 }
2305 
2306 void
2307 pci_disable_msi_method(device_t dev, device_t child)
2308 {
2309 	struct pci_devinfo *dinfo = device_get_ivars(child);
2310 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2311 
2312 	/* Disable MSI -> HT mapping. */
2313 	pci_ht_map_msi(child, 0);
2314 
2315 	/* Disable MSI in the control register. */
2316 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2317 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2318 	    msi->msi_ctrl, 2);
2319 }
2320 
2321 /*
2322  * Restore MSI registers during resume.  If MSI is enabled then
2323  * restore the data and address registers in addition to the control
2324  * register.
2325  */
2326 static void
2327 pci_resume_msi(device_t dev)
2328 {
2329 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2330 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2331 	uint64_t address;
2332 	uint16_t data;
2333 
2334 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2335 		address = msi->msi_addr;
2336 		data = msi->msi_data;
2337 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2338 		    address & 0xffffffff, 4);
2339 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2340 			pci_write_config(dev, msi->msi_location +
2341 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2342 			pci_write_config(dev, msi->msi_location +
2343 			    PCIR_MSI_DATA_64BIT, data, 2);
2344 		} else
2345 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2346 			    data, 2);
2347 	}
2348 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2349 	    2);
2350 }
2351 
2352 static int
2353 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2354 {
2355 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2356 	pcicfgregs *cfg = &dinfo->cfg;
2357 	struct resource_list_entry *rle;
2358 	struct msix_table_entry *mte;
2359 	struct msix_vector *mv;
2360 	uint64_t addr;
2361 	uint32_t data;
2362 	int error, i, j;
2363 
2364 	/*
2365 	 * Handle MSI first.  We try to find this IRQ among our list
2366 	 * of MSI IRQs.  If we find it, we request updated address and
2367 	 * data registers and apply the results.
2368 	 */
2369 	if (cfg->msi.msi_alloc > 0) {
2370 
2371 		/* If we don't have any active handlers, nothing to do. */
2372 		if (cfg->msi.msi_handlers == 0)
2373 			return (0);
2374 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2375 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2376 			    i + 1);
2377 			if (rle->start == irq) {
2378 				error = PCIB_MAP_MSI(device_get_parent(bus),
2379 				    dev, irq, &addr, &data);
2380 				if (error)
2381 					return (error);
2382 				pci_disable_msi(dev);
2383 				dinfo->cfg.msi.msi_addr = addr;
2384 				dinfo->cfg.msi.msi_data = data;
2385 				pci_enable_msi(dev, addr, data);
2386 				return (0);
2387 			}
2388 		}
2389 		return (ENOENT);
2390 	}
2391 
2392 	/*
2393 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2394 	 * we request the updated mapping info.  If that works, we go
2395 	 * through all the slots that use this IRQ and update them.
2396 	 */
2397 	if (cfg->msix.msix_alloc > 0) {
2398 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2399 			mv = &cfg->msix.msix_vectors[i];
2400 			if (mv->mv_irq == irq) {
2401 				error = PCIB_MAP_MSI(device_get_parent(bus),
2402 				    dev, irq, &addr, &data);
2403 				if (error)
2404 					return (error);
2405 				mv->mv_address = addr;
2406 				mv->mv_data = data;
2407 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2408 					mte = &cfg->msix.msix_table[j];
2409 					if (mte->mte_vector != i + 1)
2410 						continue;
2411 					if (mte->mte_handlers == 0)
2412 						continue;
2413 					pci_mask_msix(dev, j);
2414 					pci_enable_msix(dev, j, addr, data);
2415 					pci_unmask_msix(dev, j);
2416 				}
2417 			}
2418 		}
2419 		return (ENOENT);
2420 	}
2421 
2422 	return (ENOENT);
2423 }
2424 
2425 /*
2426  * Returns true if the specified device is blacklisted because MSI
2427  * doesn't work.
2428  */
2429 int
2430 pci_msi_device_blacklisted(device_t dev)
2431 {
2432 
2433 	if (!pci_honor_msi_blacklist)
2434 		return (0);
2435 
2436 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2437 }
2438 
2439 /*
2440  * Determine if MSI is blacklisted globally on this system.  Currently,
2441  * we just check for blacklisted chipsets as represented by the
2442  * host-PCI bridge at device 0:0:0.  In the future, it may become
2443  * necessary to check other system attributes, such as the kenv values
2444  * that give the motherboard manufacturer and model number.
2445  */
2446 static int
2447 pci_msi_blacklisted(void)
2448 {
2449 	device_t dev;
2450 
2451 	if (!pci_honor_msi_blacklist)
2452 		return (0);
2453 
2454 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2455 	if (!(pcie_chipset || pcix_chipset)) {
2456 		if (vm_guest != VM_GUEST_NO) {
2457 			/*
2458 			 * Whitelist older chipsets in virtual
2459 			 * machines known to support MSI.
2460 			 */
2461 			dev = pci_find_bsf(0, 0, 0);
2462 			if (dev != NULL)
2463 				return (!pci_has_quirk(pci_get_devid(dev),
2464 					PCI_QUIRK_ENABLE_MSI_VM));
2465 		}
2466 		return (1);
2467 	}
2468 
2469 	dev = pci_find_bsf(0, 0, 0);
2470 	if (dev != NULL)
2471 		return (pci_msi_device_blacklisted(dev));
2472 	return (0);
2473 }
2474 
2475 /*
2476  * Returns true if the specified device is blacklisted because MSI-X
2477  * doesn't work.  Note that this assumes that if MSI doesn't work,
2478  * MSI-X doesn't either.
2479  */
2480 int
2481 pci_msix_device_blacklisted(device_t dev)
2482 {
2483 
2484 	if (!pci_honor_msi_blacklist)
2485 		return (0);
2486 
2487 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2488 		return (1);
2489 
2490 	return (pci_msi_device_blacklisted(dev));
2491 }
2492 
2493 /*
2494  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2495  * is blacklisted, assume that MSI-X is as well.  Check for additional
2496  * chipsets where MSI works but MSI-X does not.
2497  */
2498 static int
2499 pci_msix_blacklisted(void)
2500 {
2501 	device_t dev;
2502 
2503 	if (!pci_honor_msi_blacklist)
2504 		return (0);
2505 
2506 	dev = pci_find_bsf(0, 0, 0);
2507 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2508 	    PCI_QUIRK_DISABLE_MSIX))
2509 		return (1);
2510 
2511 	return (pci_msi_blacklisted());
2512 }
2513 
2514 /*
2515  * Attempt to allocate *count MSI messages.  The actual number allocated is
2516  * returned in *count.  After this function returns, each message will be
2517  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2518  */
2519 int
2520 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2521 {
2522 	struct pci_devinfo *dinfo = device_get_ivars(child);
2523 	pcicfgregs *cfg = &dinfo->cfg;
2524 	struct resource_list_entry *rle;
2525 	int actual, error, i, irqs[32];
2526 	uint16_t ctrl;
2527 
2528 	/* Don't let count == 0 get us into trouble. */
2529 	if (*count == 0)
2530 		return (EINVAL);
2531 
2532 	/* If rid 0 is allocated, then fail. */
2533 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2534 	if (rle != NULL && rle->res != NULL)
2535 		return (ENXIO);
2536 
2537 	/* Already have allocated messages? */
2538 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2539 		return (ENXIO);
2540 
2541 	/* If MSI is blacklisted for this system, fail. */
2542 	if (pci_msi_blacklisted())
2543 		return (ENXIO);
2544 
2545 	/* MSI capability present? */
2546 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2547 		return (ENODEV);
2548 
2549 	if (bootverbose)
2550 		device_printf(child,
2551 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2552 		    *count, cfg->msi.msi_msgnum);
2553 
2554 	/* Don't ask for more than the device supports. */
2555 	actual = min(*count, cfg->msi.msi_msgnum);
2556 
2557 	/* Don't ask for more than 32 messages. */
2558 	actual = min(actual, 32);
2559 
2560 	/* MSI requires power of 2 number of messages. */
2561 	if (!powerof2(actual))
2562 		return (EINVAL);
2563 
2564 	for (;;) {
2565 		/* Try to allocate N messages. */
2566 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2567 		    actual, irqs);
2568 		if (error == 0)
2569 			break;
2570 		if (actual == 1)
2571 			return (error);
2572 
2573 		/* Try N / 2. */
2574 		actual >>= 1;
2575 	}
2576 
2577 	/*
2578 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2579 	 * resources in the irqs[] array, so add new resources
2580 	 * starting at rid 1.
2581 	 */
2582 	for (i = 0; i < actual; i++)
2583 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2584 		    irqs[i], irqs[i], 1);
2585 
2586 	if (bootverbose) {
2587 		if (actual == 1)
2588 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2589 		else {
2590 			int run;
2591 
2592 			/*
2593 			 * Be fancy and try to print contiguous runs
2594 			 * of IRQ values as ranges.  'run' is true if
2595 			 * we are in a range.
2596 			 */
2597 			device_printf(child, "using IRQs %d", irqs[0]);
2598 			run = 0;
2599 			for (i = 1; i < actual; i++) {
2600 
2601 				/* Still in a run? */
2602 				if (irqs[i] == irqs[i - 1] + 1) {
2603 					run = 1;
2604 					continue;
2605 				}
2606 
2607 				/* Finish previous range. */
2608 				if (run) {
2609 					printf("-%d", irqs[i - 1]);
2610 					run = 0;
2611 				}
2612 
2613 				/* Start new range. */
2614 				printf(",%d", irqs[i]);
2615 			}
2616 
2617 			/* Unfinished range? */
2618 			if (run)
2619 				printf("-%d", irqs[actual - 1]);
2620 			printf(" for MSI\n");
2621 		}
2622 	}
2623 
2624 	/* Update control register with actual count. */
2625 	ctrl = cfg->msi.msi_ctrl;
2626 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2627 	ctrl |= (ffs(actual) - 1) << 4;
2628 	cfg->msi.msi_ctrl = ctrl;
2629 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2630 
2631 	/* Update counts of alloc'd messages. */
2632 	cfg->msi.msi_alloc = actual;
2633 	cfg->msi.msi_handlers = 0;
2634 	*count = actual;
2635 	return (0);
2636 }
2637 
2638 /* Release the MSI messages associated with this device. */
2639 int
2640 pci_release_msi_method(device_t dev, device_t child)
2641 {
2642 	struct pci_devinfo *dinfo = device_get_ivars(child);
2643 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2644 	struct resource_list_entry *rle;
2645 	int error, i, irqs[32];
2646 
2647 	/* Try MSI-X first. */
2648 	error = pci_release_msix(dev, child);
2649 	if (error != ENODEV)
2650 		return (error);
2651 
2652 	/* Do we have any messages to release? */
2653 	if (msi->msi_alloc == 0)
2654 		return (ENODEV);
2655 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2656 
2657 	/* Make sure none of the resources are allocated. */
2658 	if (msi->msi_handlers > 0)
2659 		return (EBUSY);
2660 	for (i = 0; i < msi->msi_alloc; i++) {
2661 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2662 		KASSERT(rle != NULL, ("missing MSI resource"));
2663 		if (rle->res != NULL)
2664 			return (EBUSY);
2665 		irqs[i] = rle->start;
2666 	}
2667 
2668 	/* Update control register with 0 count. */
2669 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2670 	    ("%s: MSI still enabled", __func__));
2671 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2672 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2673 	    msi->msi_ctrl, 2);
2674 
2675 	/* Release the messages. */
2676 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2677 	for (i = 0; i < msi->msi_alloc; i++)
2678 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2679 
2680 	/* Update alloc count. */
2681 	msi->msi_alloc = 0;
2682 	msi->msi_addr = 0;
2683 	msi->msi_data = 0;
2684 	return (0);
2685 }
2686 
2687 /*
2688  * Return the max supported MSI messages this device supports.
2689  * Basically, assuming the MD code can alloc messages, this function
2690  * should return the maximum value that pci_alloc_msi() can return.
2691  * Thus, it is subject to the tunables, etc.
2692  */
2693 int
2694 pci_msi_count_method(device_t dev, device_t child)
2695 {
2696 	struct pci_devinfo *dinfo = device_get_ivars(child);
2697 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2698 
2699 	if (pci_do_msi && msi->msi_location != 0)
2700 		return (msi->msi_msgnum);
2701 	return (0);
2702 }
2703 
2704 /* free pcicfgregs structure and all depending data structures */
2705 
2706 int
2707 pci_freecfg(struct pci_devinfo *dinfo)
2708 {
2709 	struct devlist *devlist_head;
2710 	struct pci_map *pm, *next;
2711 	int i;
2712 
2713 	devlist_head = &pci_devq;
2714 
2715 	if (dinfo->cfg.vpd.vpd_reg) {
2716 		free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2717 		for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2718 			free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2719 		free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2720 		for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2721 			free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2722 		free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2723 	}
2724 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2725 		free(pm, M_DEVBUF);
2726 	}
2727 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2728 	free(dinfo, M_DEVBUF);
2729 
2730 	/* increment the generation count */
2731 	pci_generation++;
2732 
2733 	/* we're losing one device */
2734 	pci_numdevs--;
2735 	return (0);
2736 }
2737 
2738 /*
2739  * PCI power manangement
2740  */
2741 int
2742 pci_set_powerstate_method(device_t dev, device_t child, int state)
2743 {
2744 	struct pci_devinfo *dinfo = device_get_ivars(child);
2745 	pcicfgregs *cfg = &dinfo->cfg;
2746 	uint16_t status;
2747 	int oldstate, highest, delay;
2748 
2749 	if (cfg->pp.pp_cap == 0)
2750 		return (EOPNOTSUPP);
2751 
2752 	/*
2753 	 * Optimize a no state change request away.  While it would be OK to
2754 	 * write to the hardware in theory, some devices have shown odd
2755 	 * behavior when going from D3 -> D3.
2756 	 */
2757 	oldstate = pci_get_powerstate(child);
2758 	if (oldstate == state)
2759 		return (0);
2760 
2761 	/*
2762 	 * The PCI power management specification states that after a state
2763 	 * transition between PCI power states, system software must
2764 	 * guarantee a minimal delay before the function accesses the device.
2765 	 * Compute the worst case delay that we need to guarantee before we
2766 	 * access the device.  Many devices will be responsive much more
2767 	 * quickly than this delay, but there are some that don't respond
2768 	 * instantly to state changes.  Transitions to/from D3 state require
2769 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2770 	 * is done below with DELAY rather than a sleeper function because
2771 	 * this function can be called from contexts where we cannot sleep.
2772 	 */
2773 	highest = (oldstate > state) ? oldstate : state;
2774 	if (highest == PCI_POWERSTATE_D3)
2775 	    delay = 10000;
2776 	else if (highest == PCI_POWERSTATE_D2)
2777 	    delay = 200;
2778 	else
2779 	    delay = 0;
2780 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2781 	    & ~PCIM_PSTAT_DMASK;
2782 	switch (state) {
2783 	case PCI_POWERSTATE_D0:
2784 		status |= PCIM_PSTAT_D0;
2785 		break;
2786 	case PCI_POWERSTATE_D1:
2787 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2788 			return (EOPNOTSUPP);
2789 		status |= PCIM_PSTAT_D1;
2790 		break;
2791 	case PCI_POWERSTATE_D2:
2792 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2793 			return (EOPNOTSUPP);
2794 		status |= PCIM_PSTAT_D2;
2795 		break;
2796 	case PCI_POWERSTATE_D3:
2797 		status |= PCIM_PSTAT_D3;
2798 		break;
2799 	default:
2800 		return (EINVAL);
2801 	}
2802 
2803 	if (bootverbose)
2804 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2805 		    state);
2806 
2807 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2808 	if (delay)
2809 		DELAY(delay);
2810 	return (0);
2811 }
2812 
2813 int
2814 pci_get_powerstate_method(device_t dev, device_t child)
2815 {
2816 	struct pci_devinfo *dinfo = device_get_ivars(child);
2817 	pcicfgregs *cfg = &dinfo->cfg;
2818 	uint16_t status;
2819 	int result;
2820 
2821 	if (cfg->pp.pp_cap != 0) {
2822 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2823 		switch (status & PCIM_PSTAT_DMASK) {
2824 		case PCIM_PSTAT_D0:
2825 			result = PCI_POWERSTATE_D0;
2826 			break;
2827 		case PCIM_PSTAT_D1:
2828 			result = PCI_POWERSTATE_D1;
2829 			break;
2830 		case PCIM_PSTAT_D2:
2831 			result = PCI_POWERSTATE_D2;
2832 			break;
2833 		case PCIM_PSTAT_D3:
2834 			result = PCI_POWERSTATE_D3;
2835 			break;
2836 		default:
2837 			result = PCI_POWERSTATE_UNKNOWN;
2838 			break;
2839 		}
2840 	} else {
2841 		/* No support, device is always at D0 */
2842 		result = PCI_POWERSTATE_D0;
2843 	}
2844 	return (result);
2845 }
2846 
2847 /*
2848  * Some convenience functions for PCI device drivers.
2849  */
2850 
2851 static __inline void
2852 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2853 {
2854 	uint16_t	command;
2855 
2856 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2857 	command |= bit;
2858 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2859 }
2860 
2861 static __inline void
2862 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2863 {
2864 	uint16_t	command;
2865 
2866 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2867 	command &= ~bit;
2868 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2869 }
2870 
2871 int
2872 pci_enable_busmaster_method(device_t dev, device_t child)
2873 {
2874 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2875 	return (0);
2876 }
2877 
2878 int
2879 pci_disable_busmaster_method(device_t dev, device_t child)
2880 {
2881 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2882 	return (0);
2883 }
2884 
2885 int
2886 pci_enable_io_method(device_t dev, device_t child, int space)
2887 {
2888 	uint16_t bit;
2889 
2890 	switch(space) {
2891 	case SYS_RES_IOPORT:
2892 		bit = PCIM_CMD_PORTEN;
2893 		break;
2894 	case SYS_RES_MEMORY:
2895 		bit = PCIM_CMD_MEMEN;
2896 		break;
2897 	default:
2898 		return (EINVAL);
2899 	}
2900 	pci_set_command_bit(dev, child, bit);
2901 	return (0);
2902 }
2903 
2904 int
2905 pci_disable_io_method(device_t dev, device_t child, int space)
2906 {
2907 	uint16_t bit;
2908 
2909 	switch(space) {
2910 	case SYS_RES_IOPORT:
2911 		bit = PCIM_CMD_PORTEN;
2912 		break;
2913 	case SYS_RES_MEMORY:
2914 		bit = PCIM_CMD_MEMEN;
2915 		break;
2916 	default:
2917 		return (EINVAL);
2918 	}
2919 	pci_clear_command_bit(dev, child, bit);
2920 	return (0);
2921 }
2922 
2923 /*
2924  * New style pci driver.  Parent device is either a pci-host-bridge or a
2925  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2926  */
2927 
2928 void
2929 pci_print_verbose(struct pci_devinfo *dinfo)
2930 {
2931 
2932 	if (bootverbose) {
2933 		pcicfgregs *cfg = &dinfo->cfg;
2934 
2935 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2936 		    cfg->vendor, cfg->device, cfg->revid);
2937 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2938 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2939 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2940 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2941 		    cfg->mfdev);
2942 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2943 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2944 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2945 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2946 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2947 		if (cfg->intpin > 0)
2948 			printf("\tintpin=%c, irq=%d\n",
2949 			    cfg->intpin +'a' -1, cfg->intline);
2950 		if (cfg->pp.pp_cap) {
2951 			uint16_t status;
2952 
2953 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2954 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
2955 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2956 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2957 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2958 			    status & PCIM_PSTAT_DMASK);
2959 		}
2960 		if (cfg->msi.msi_location) {
2961 			int ctrl;
2962 
2963 			ctrl = cfg->msi.msi_ctrl;
2964 			printf("\tMSI supports %d message%s%s%s\n",
2965 			    cfg->msi.msi_msgnum,
2966 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
2967 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2968 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2969 		}
2970 		if (cfg->msix.msix_location) {
2971 			printf("\tMSI-X supports %d message%s ",
2972 			    cfg->msix.msix_msgnum,
2973 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
2974 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2975 				printf("in map 0x%x\n",
2976 				    cfg->msix.msix_table_bar);
2977 			else
2978 				printf("in maps 0x%x and 0x%x\n",
2979 				    cfg->msix.msix_table_bar,
2980 				    cfg->msix.msix_pba_bar);
2981 		}
2982 	}
2983 }
2984 
2985 static int
2986 pci_porten(device_t dev)
2987 {
2988 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2989 }
2990 
2991 static int
2992 pci_memen(device_t dev)
2993 {
2994 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2995 }
2996 
2997 void
2998 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2999     int *bar64)
3000 {
3001 	struct pci_devinfo *dinfo;
3002 	pci_addr_t map, testval;
3003 	int ln2range;
3004 	uint16_t cmd;
3005 
3006 	/*
3007 	 * The device ROM BAR is special.  It is always a 32-bit
3008 	 * memory BAR.  Bit 0 is special and should not be set when
3009 	 * sizing the BAR.
3010 	 */
3011 	dinfo = device_get_ivars(dev);
3012 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3013 		map = pci_read_config(dev, reg, 4);
3014 		pci_write_config(dev, reg, 0xfffffffe, 4);
3015 		testval = pci_read_config(dev, reg, 4);
3016 		pci_write_config(dev, reg, map, 4);
3017 		*mapp = map;
3018 		*testvalp = testval;
3019 		if (bar64 != NULL)
3020 			*bar64 = 0;
3021 		return;
3022 	}
3023 
3024 	map = pci_read_config(dev, reg, 4);
3025 	ln2range = pci_maprange(map);
3026 	if (ln2range == 64)
3027 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3028 
3029 	/*
3030 	 * Disable decoding via the command register before
3031 	 * determining the BAR's length since we will be placing it in
3032 	 * a weird state.
3033 	 */
3034 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3035 	pci_write_config(dev, PCIR_COMMAND,
3036 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3037 
3038 	/*
3039 	 * Determine the BAR's length by writing all 1's.  The bottom
3040 	 * log_2(size) bits of the BAR will stick as 0 when we read
3041 	 * the value back.
3042 	 *
3043 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3044 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3045 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3046 	 *
3047 	 * Writes to both registers must be performed before attempting to
3048 	 * read back the size value.
3049 	 */
3050 	testval = 0;
3051 	pci_write_config(dev, reg, 0xffffffff, 4);
3052 	if (ln2range == 64) {
3053 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3054 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3055 	}
3056 	testval |= pci_read_config(dev, reg, 4);
3057 
3058 	/*
3059 	 * Restore the original value of the BAR.  We may have reprogrammed
3060 	 * the BAR of the low-level console device and when booting verbose,
3061 	 * we need the console device addressable.
3062 	 */
3063 	pci_write_config(dev, reg, map, 4);
3064 	if (ln2range == 64)
3065 		pci_write_config(dev, reg + 4, map >> 32, 4);
3066 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3067 
3068 	*mapp = map;
3069 	*testvalp = testval;
3070 	if (bar64 != NULL)
3071 		*bar64 = (ln2range == 64);
3072 }
3073 
3074 static void
3075 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3076 {
3077 	struct pci_devinfo *dinfo;
3078 	int ln2range;
3079 
3080 	/* The device ROM BAR is always a 32-bit memory BAR. */
3081 	dinfo = device_get_ivars(dev);
3082 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3083 		ln2range = 32;
3084 	else
3085 		ln2range = pci_maprange(pm->pm_value);
3086 	pci_write_config(dev, pm->pm_reg, base, 4);
3087 	if (ln2range == 64)
3088 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3089 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3090 	if (ln2range == 64)
3091 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3092 		    pm->pm_reg + 4, 4) << 32;
3093 }
3094 
3095 struct pci_map *
3096 pci_find_bar(device_t dev, int reg)
3097 {
3098 	struct pci_devinfo *dinfo;
3099 	struct pci_map *pm;
3100 
3101 	dinfo = device_get_ivars(dev);
3102 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3103 		if (pm->pm_reg == reg)
3104 			return (pm);
3105 	}
3106 	return (NULL);
3107 }
3108 
3109 int
3110 pci_bar_enabled(device_t dev, struct pci_map *pm)
3111 {
3112 	struct pci_devinfo *dinfo;
3113 	uint16_t cmd;
3114 
3115 	dinfo = device_get_ivars(dev);
3116 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3117 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3118 		return (0);
3119 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3120 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3121 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3122 	else
3123 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3124 }
3125 
3126 struct pci_map *
3127 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3128 {
3129 	struct pci_devinfo *dinfo;
3130 	struct pci_map *pm, *prev;
3131 
3132 	dinfo = device_get_ivars(dev);
3133 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3134 	pm->pm_reg = reg;
3135 	pm->pm_value = value;
3136 	pm->pm_size = size;
3137 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3138 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3139 		    reg));
3140 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3141 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3142 			break;
3143 	}
3144 	if (prev != NULL)
3145 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3146 	else
3147 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3148 	return (pm);
3149 }
3150 
3151 static void
3152 pci_restore_bars(device_t dev)
3153 {
3154 	struct pci_devinfo *dinfo;
3155 	struct pci_map *pm;
3156 	int ln2range;
3157 
3158 	dinfo = device_get_ivars(dev);
3159 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3160 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3161 			ln2range = 32;
3162 		else
3163 			ln2range = pci_maprange(pm->pm_value);
3164 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3165 		if (ln2range == 64)
3166 			pci_write_config(dev, pm->pm_reg + 4,
3167 			    pm->pm_value >> 32, 4);
3168 	}
3169 }
3170 
3171 /*
3172  * Add a resource based on a pci map register. Return 1 if the map
3173  * register is a 32bit map register or 2 if it is a 64bit register.
3174  */
3175 static int
3176 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3177     int force, int prefetch)
3178 {
3179 	struct pci_map *pm;
3180 	pci_addr_t base, map, testval;
3181 	pci_addr_t start, end, count;
3182 	int barlen, basezero, flags, maprange, mapsize, type;
3183 	uint16_t cmd;
3184 	struct resource *res;
3185 
3186 	/*
3187 	 * The BAR may already exist if the device is a CardBus card
3188 	 * whose CIS is stored in this BAR.
3189 	 */
3190 	pm = pci_find_bar(dev, reg);
3191 	if (pm != NULL) {
3192 		maprange = pci_maprange(pm->pm_value);
3193 		barlen = maprange == 64 ? 2 : 1;
3194 		return (barlen);
3195 	}
3196 
3197 	pci_read_bar(dev, reg, &map, &testval, NULL);
3198 	if (PCI_BAR_MEM(map)) {
3199 		type = SYS_RES_MEMORY;
3200 		if (map & PCIM_BAR_MEM_PREFETCH)
3201 			prefetch = 1;
3202 	} else
3203 		type = SYS_RES_IOPORT;
3204 	mapsize = pci_mapsize(testval);
3205 	base = pci_mapbase(map);
3206 #ifdef __PCI_BAR_ZERO_VALID
3207 	basezero = 0;
3208 #else
3209 	basezero = base == 0;
3210 #endif
3211 	maprange = pci_maprange(map);
3212 	barlen = maprange == 64 ? 2 : 1;
3213 
3214 	/*
3215 	 * For I/O registers, if bottom bit is set, and the next bit up
3216 	 * isn't clear, we know we have a BAR that doesn't conform to the
3217 	 * spec, so ignore it.  Also, sanity check the size of the data
3218 	 * areas to the type of memory involved.  Memory must be at least
3219 	 * 16 bytes in size, while I/O ranges must be at least 4.
3220 	 */
3221 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3222 		return (barlen);
3223 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3224 	    (type == SYS_RES_IOPORT && mapsize < 2))
3225 		return (barlen);
3226 
3227 	/* Save a record of this BAR. */
3228 	pm = pci_add_bar(dev, reg, map, mapsize);
3229 	if (bootverbose) {
3230 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3231 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3232 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3233 			printf(", port disabled\n");
3234 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3235 			printf(", memory disabled\n");
3236 		else
3237 			printf(", enabled\n");
3238 	}
3239 
3240 	/*
3241 	 * If base is 0, then we have problems if this architecture does
3242 	 * not allow that.  It is best to ignore such entries for the
3243 	 * moment.  These will be allocated later if the driver specifically
3244 	 * requests them.  However, some removable buses look better when
3245 	 * all resources are allocated, so allow '0' to be overriden.
3246 	 *
3247 	 * Similarly treat maps whose values is the same as the test value
3248 	 * read back.  These maps have had all f's written to them by the
3249 	 * BIOS in an attempt to disable the resources.
3250 	 */
3251 	if (!force && (basezero || map == testval))
3252 		return (barlen);
3253 	if ((u_long)base != base) {
3254 		device_printf(bus,
3255 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3256 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3257 		    pci_get_function(dev), reg);
3258 		return (barlen);
3259 	}
3260 
3261 	/*
3262 	 * This code theoretically does the right thing, but has
3263 	 * undesirable side effects in some cases where peripherals
3264 	 * respond oddly to having these bits enabled.  Let the user
3265 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3266 	 * default).
3267 	 */
3268 	if (pci_enable_io_modes) {
3269 		/* Turn on resources that have been left off by a lazy BIOS */
3270 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3271 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3272 			cmd |= PCIM_CMD_PORTEN;
3273 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3274 		}
3275 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3276 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3277 			cmd |= PCIM_CMD_MEMEN;
3278 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3279 		}
3280 	} else {
3281 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3282 			return (barlen);
3283 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3284 			return (barlen);
3285 	}
3286 
3287 	count = (pci_addr_t)1 << mapsize;
3288 	flags = RF_ALIGNMENT_LOG2(mapsize);
3289 	if (prefetch)
3290 		flags |= RF_PREFETCHABLE;
3291 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3292 		start = 0;	/* Let the parent decide. */
3293 		end = ~0;
3294 	} else {
3295 		start = base;
3296 		end = base + count - 1;
3297 	}
3298 	resource_list_add(rl, type, reg, start, end, count);
3299 
3300 	/*
3301 	 * Try to allocate the resource for this BAR from our parent
3302 	 * so that this resource range is already reserved.  The
3303 	 * driver for this device will later inherit this resource in
3304 	 * pci_alloc_resource().
3305 	 */
3306 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3307 	    flags);
3308 	if ((pci_do_realloc_bars
3309 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3310 	    && res == NULL && (start != 0 || end != ~0)) {
3311 		/*
3312 		 * If the allocation fails, try to allocate a resource for
3313 		 * this BAR using any available range.  The firmware felt
3314 		 * it was important enough to assign a resource, so don't
3315 		 * disable decoding if we can help it.
3316 		 */
3317 		resource_list_delete(rl, type, reg);
3318 		resource_list_add(rl, type, reg, 0, ~0, count);
3319 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3320 		    count, flags);
3321 	}
3322 	if (res == NULL) {
3323 		/*
3324 		 * If the allocation fails, delete the resource list entry
3325 		 * and disable decoding for this device.
3326 		 *
3327 		 * If the driver requests this resource in the future,
3328 		 * pci_reserve_map() will try to allocate a fresh
3329 		 * resource range.
3330 		 */
3331 		resource_list_delete(rl, type, reg);
3332 		pci_disable_io(dev, type);
3333 		if (bootverbose)
3334 			device_printf(bus,
3335 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3336 			    pci_get_domain(dev), pci_get_bus(dev),
3337 			    pci_get_slot(dev), pci_get_function(dev), reg);
3338 	} else {
3339 		start = rman_get_start(res);
3340 		pci_write_bar(dev, pm, start);
3341 	}
3342 	return (barlen);
3343 }
3344 
3345 /*
3346  * For ATA devices we need to decide early what addressing mode to use.
3347  * Legacy demands that the primary and secondary ATA ports sits on the
3348  * same addresses that old ISA hardware did. This dictates that we use
3349  * those addresses and ignore the BAR's if we cannot set PCI native
3350  * addressing mode.
3351  */
3352 static void
3353 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3354     uint32_t prefetchmask)
3355 {
3356 	int rid, type, progif;
3357 #if 0
3358 	/* if this device supports PCI native addressing use it */
3359 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3360 	if ((progif & 0x8a) == 0x8a) {
3361 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3362 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3363 			printf("Trying ATA native PCI addressing mode\n");
3364 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3365 		}
3366 	}
3367 #endif
3368 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3369 	type = SYS_RES_IOPORT;
3370 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3371 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3372 		    prefetchmask & (1 << 0));
3373 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3374 		    prefetchmask & (1 << 1));
3375 	} else {
3376 		rid = PCIR_BAR(0);
3377 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3378 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3379 		    0x1f7, 8, 0);
3380 		rid = PCIR_BAR(1);
3381 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3382 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3383 		    0x3f6, 1, 0);
3384 	}
3385 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3386 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3387 		    prefetchmask & (1 << 2));
3388 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3389 		    prefetchmask & (1 << 3));
3390 	} else {
3391 		rid = PCIR_BAR(2);
3392 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3393 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3394 		    0x177, 8, 0);
3395 		rid = PCIR_BAR(3);
3396 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3397 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3398 		    0x376, 1, 0);
3399 	}
3400 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3401 	    prefetchmask & (1 << 4));
3402 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3403 	    prefetchmask & (1 << 5));
3404 }
3405 
3406 static void
3407 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3408 {
3409 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3410 	pcicfgregs *cfg = &dinfo->cfg;
3411 	char tunable_name[64];
3412 	int irq;
3413 
3414 	/* Has to have an intpin to have an interrupt. */
3415 	if (cfg->intpin == 0)
3416 		return;
3417 
3418 	/* Let the user override the IRQ with a tunable. */
3419 	irq = PCI_INVALID_IRQ;
3420 	snprintf(tunable_name, sizeof(tunable_name),
3421 	    "hw.pci%d.%d.%d.INT%c.irq",
3422 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3423 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3424 		irq = PCI_INVALID_IRQ;
3425 
3426 	/*
3427 	 * If we didn't get an IRQ via the tunable, then we either use the
3428 	 * IRQ value in the intline register or we ask the bus to route an
3429 	 * interrupt for us.  If force_route is true, then we only use the
3430 	 * value in the intline register if the bus was unable to assign an
3431 	 * IRQ.
3432 	 */
3433 	if (!PCI_INTERRUPT_VALID(irq)) {
3434 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3435 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3436 		if (!PCI_INTERRUPT_VALID(irq))
3437 			irq = cfg->intline;
3438 	}
3439 
3440 	/* If after all that we don't have an IRQ, just bail. */
3441 	if (!PCI_INTERRUPT_VALID(irq))
3442 		return;
3443 
3444 	/* Update the config register if it changed. */
3445 	if (irq != cfg->intline) {
3446 		cfg->intline = irq;
3447 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3448 	}
3449 
3450 	/* Add this IRQ as rid 0 interrupt resource. */
3451 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3452 }
3453 
3454 /* Perform early OHCI takeover from SMM. */
3455 static void
3456 ohci_early_takeover(device_t self)
3457 {
3458 	struct resource *res;
3459 	uint32_t ctl;
3460 	int rid;
3461 	int i;
3462 
3463 	rid = PCIR_BAR(0);
3464 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3465 	if (res == NULL)
3466 		return;
3467 
3468 	ctl = bus_read_4(res, OHCI_CONTROL);
3469 	if (ctl & OHCI_IR) {
3470 		if (bootverbose)
3471 			printf("ohci early: "
3472 			    "SMM active, request owner change\n");
3473 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3474 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3475 			DELAY(1000);
3476 			ctl = bus_read_4(res, OHCI_CONTROL);
3477 		}
3478 		if (ctl & OHCI_IR) {
3479 			if (bootverbose)
3480 				printf("ohci early: "
3481 				    "SMM does not respond, resetting\n");
3482 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3483 		}
3484 		/* Disable interrupts */
3485 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3486 	}
3487 
3488 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3489 }
3490 
3491 /* Perform early UHCI takeover from SMM. */
3492 static void
3493 uhci_early_takeover(device_t self)
3494 {
3495 	struct resource *res;
3496 	int rid;
3497 
3498 	/*
3499 	 * Set the PIRQD enable bit and switch off all the others. We don't
3500 	 * want legacy support to interfere with us XXX Does this also mean
3501 	 * that the BIOS won't touch the keyboard anymore if it is connected
3502 	 * to the ports of the root hub?
3503 	 */
3504 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3505 
3506 	/* Disable interrupts */
3507 	rid = PCI_UHCI_BASE_REG;
3508 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3509 	if (res != NULL) {
3510 		bus_write_2(res, UHCI_INTR, 0);
3511 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3512 	}
3513 }
3514 
3515 /* Perform early EHCI takeover from SMM. */
3516 static void
3517 ehci_early_takeover(device_t self)
3518 {
3519 	struct resource *res;
3520 	uint32_t cparams;
3521 	uint32_t eec;
3522 	uint8_t eecp;
3523 	uint8_t bios_sem;
3524 	uint8_t offs;
3525 	int rid;
3526 	int i;
3527 
3528 	rid = PCIR_BAR(0);
3529 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3530 	if (res == NULL)
3531 		return;
3532 
3533 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3534 
3535 	/* Synchronise with the BIOS if it owns the controller. */
3536 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3537 	    eecp = EHCI_EECP_NEXT(eec)) {
3538 		eec = pci_read_config(self, eecp, 4);
3539 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3540 			continue;
3541 		}
3542 		bios_sem = pci_read_config(self, eecp +
3543 		    EHCI_LEGSUP_BIOS_SEM, 1);
3544 		if (bios_sem == 0) {
3545 			continue;
3546 		}
3547 		if (bootverbose)
3548 			printf("ehci early: "
3549 			    "SMM active, request owner change\n");
3550 
3551 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3552 
3553 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3554 			DELAY(1000);
3555 			bios_sem = pci_read_config(self, eecp +
3556 			    EHCI_LEGSUP_BIOS_SEM, 1);
3557 		}
3558 
3559 		if (bios_sem != 0) {
3560 			if (bootverbose)
3561 				printf("ehci early: "
3562 				    "SMM does not respond\n");
3563 		}
3564 		/* Disable interrupts */
3565 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3566 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3567 	}
3568 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3569 }
3570 
3571 /* Perform early XHCI takeover from SMM. */
3572 static void
3573 xhci_early_takeover(device_t self)
3574 {
3575 	struct resource *res;
3576 	uint32_t cparams;
3577 	uint32_t eec;
3578 	uint8_t eecp;
3579 	uint8_t bios_sem;
3580 	uint8_t offs;
3581 	int rid;
3582 	int i;
3583 
3584 	rid = PCIR_BAR(0);
3585 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3586 	if (res == NULL)
3587 		return;
3588 
3589 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3590 
3591 	eec = -1;
3592 
3593 	/* Synchronise with the BIOS if it owns the controller. */
3594 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3595 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3596 		eec = bus_read_4(res, eecp);
3597 
3598 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3599 			continue;
3600 
3601 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3602 		if (bios_sem == 0)
3603 			continue;
3604 
3605 		if (bootverbose)
3606 			printf("xhci early: "
3607 			    "SMM active, request owner change\n");
3608 
3609 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3610 
3611 		/* wait a maximum of 5 second */
3612 
3613 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3614 			DELAY(1000);
3615 			bios_sem = bus_read_1(res, eecp +
3616 			    XHCI_XECP_BIOS_SEM);
3617 		}
3618 
3619 		if (bios_sem != 0) {
3620 			if (bootverbose)
3621 				printf("xhci early: "
3622 				    "SMM does not respond\n");
3623 		}
3624 
3625 		/* Disable interrupts */
3626 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3627 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3628 		bus_read_4(res, offs + XHCI_USBSTS);
3629 	}
3630 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3631 }
3632 
3633 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3634 static void
3635 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3636     struct resource_list *rl)
3637 {
3638 	struct resource *res;
3639 	char *cp;
3640 	rman_res_t start, end, count;
3641 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3642 
3643 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3644 	case PCIM_HDRTYPE_BRIDGE:
3645 		sec_reg = PCIR_SECBUS_1;
3646 		sub_reg = PCIR_SUBBUS_1;
3647 		break;
3648 	case PCIM_HDRTYPE_CARDBUS:
3649 		sec_reg = PCIR_SECBUS_2;
3650 		sub_reg = PCIR_SUBBUS_2;
3651 		break;
3652 	default:
3653 		return;
3654 	}
3655 
3656 	/*
3657 	 * If the existing bus range is valid, attempt to reserve it
3658 	 * from our parent.  If this fails for any reason, clear the
3659 	 * secbus and subbus registers.
3660 	 *
3661 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3662 	 * This would at least preserve the existing sec_bus if it is
3663 	 * valid.
3664 	 */
3665 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3666 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3667 
3668 	/* Quirk handling. */
3669 	switch (pci_get_devid(dev)) {
3670 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3671 		sup_bus = pci_read_config(dev, 0x41, 1);
3672 		if (sup_bus != 0xff) {
3673 			sec_bus = sup_bus + 1;
3674 			sub_bus = sup_bus + 1;
3675 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3676 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3677 		}
3678 		break;
3679 
3680 	case 0x00dd10de:
3681 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3682 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3683 			break;
3684 		if (strncmp(cp, "Compal", 6) != 0) {
3685 			freeenv(cp);
3686 			break;
3687 		}
3688 		freeenv(cp);
3689 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3690 			break;
3691 		if (strncmp(cp, "08A0", 4) != 0) {
3692 			freeenv(cp);
3693 			break;
3694 		}
3695 		freeenv(cp);
3696 		if (sub_bus < 0xa) {
3697 			sub_bus = 0xa;
3698 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3699 		}
3700 		break;
3701 	}
3702 
3703 	if (bootverbose)
3704 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3705 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3706 		start = sec_bus;
3707 		end = sub_bus;
3708 		count = end - start + 1;
3709 
3710 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3711 
3712 		/*
3713 		 * If requested, clear secondary bus registers in
3714 		 * bridge devices to force a complete renumbering
3715 		 * rather than reserving the existing range.  However,
3716 		 * preserve the existing size.
3717 		 */
3718 		if (pci_clear_buses)
3719 			goto clear;
3720 
3721 		rid = 0;
3722 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3723 		    start, end, count, 0);
3724 		if (res != NULL)
3725 			return;
3726 
3727 		if (bootverbose)
3728 			device_printf(bus,
3729 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3730 			    pci_get_domain(dev), pci_get_bus(dev),
3731 			    pci_get_slot(dev), pci_get_function(dev));
3732 	}
3733 
3734 clear:
3735 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3736 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3737 }
3738 
3739 static struct resource *
3740 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3741     rman_res_t end, rman_res_t count, u_int flags)
3742 {
3743 	struct pci_devinfo *dinfo;
3744 	pcicfgregs *cfg;
3745 	struct resource_list *rl;
3746 	struct resource *res;
3747 	int sec_reg, sub_reg;
3748 
3749 	dinfo = device_get_ivars(child);
3750 	cfg = &dinfo->cfg;
3751 	rl = &dinfo->resources;
3752 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3753 	case PCIM_HDRTYPE_BRIDGE:
3754 		sec_reg = PCIR_SECBUS_1;
3755 		sub_reg = PCIR_SUBBUS_1;
3756 		break;
3757 	case PCIM_HDRTYPE_CARDBUS:
3758 		sec_reg = PCIR_SECBUS_2;
3759 		sub_reg = PCIR_SUBBUS_2;
3760 		break;
3761 	default:
3762 		return (NULL);
3763 	}
3764 
3765 	if (*rid != 0)
3766 		return (NULL);
3767 
3768 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3769 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3770 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3771 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3772 		    start, end, count, flags & ~RF_ACTIVE);
3773 		if (res == NULL) {
3774 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3775 			device_printf(child, "allocating %ju bus%s failed\n",
3776 			    count, count == 1 ? "" : "es");
3777 			return (NULL);
3778 		}
3779 		if (bootverbose)
3780 			device_printf(child,
3781 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3782 			    count == 1 ? "" : "es", rman_get_start(res));
3783 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3784 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3785 	}
3786 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3787 	    end, count, flags));
3788 }
3789 #endif
3790 
3791 static int
3792 pci_ea_bei_to_rid(device_t dev, int bei)
3793 {
3794 #ifdef PCI_IOV
3795 	struct pci_devinfo *dinfo;
3796 	int iov_pos;
3797 	struct pcicfg_iov *iov;
3798 
3799 	dinfo = device_get_ivars(dev);
3800 	iov = dinfo->cfg.iov;
3801 	if (iov != NULL)
3802 		iov_pos = iov->iov_pos;
3803 	else
3804 		iov_pos = 0;
3805 #endif
3806 
3807 	/* Check if matches BAR */
3808 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3809 	    (bei <= PCIM_EA_BEI_BAR_5))
3810 		return (PCIR_BAR(bei));
3811 
3812 	/* Check ROM */
3813 	if (bei == PCIM_EA_BEI_ROM)
3814 		return (PCIR_BIOS);
3815 
3816 #ifdef PCI_IOV
3817 	/* Check if matches VF_BAR */
3818 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3819 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3820 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3821 		    iov_pos);
3822 #endif
3823 
3824 	return (-1);
3825 }
3826 
3827 int
3828 pci_ea_is_enabled(device_t dev, int rid)
3829 {
3830 	struct pci_ea_entry *ea;
3831 	struct pci_devinfo *dinfo;
3832 
3833 	dinfo = device_get_ivars(dev);
3834 
3835 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3836 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3837 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3838 	}
3839 
3840 	return (0);
3841 }
3842 
3843 void
3844 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3845 {
3846 	struct pci_ea_entry *ea;
3847 	struct pci_devinfo *dinfo;
3848 	pci_addr_t start, end, count;
3849 	struct resource_list *rl;
3850 	int type, flags, rid;
3851 	struct resource *res;
3852 	uint32_t tmp;
3853 #ifdef PCI_IOV
3854 	struct pcicfg_iov *iov;
3855 #endif
3856 
3857 	dinfo = device_get_ivars(dev);
3858 	rl = &dinfo->resources;
3859 	flags = 0;
3860 
3861 #ifdef PCI_IOV
3862 	iov = dinfo->cfg.iov;
3863 #endif
3864 
3865 	if (dinfo->cfg.ea.ea_location == 0)
3866 		return;
3867 
3868 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3869 
3870 		/*
3871 		 * TODO: Ignore EA-BAR if is not enabled.
3872 		 *   Currently the EA implementation supports
3873 		 *   only situation, where EA structure contains
3874 		 *   predefined entries. In case they are not enabled
3875 		 *   leave them unallocated and proceed with
3876 		 *   a legacy-BAR mechanism.
3877 		 */
3878 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3879 			continue;
3880 
3881 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3882 		case PCIM_EA_P_MEM_PREFETCH:
3883 		case PCIM_EA_P_VF_MEM_PREFETCH:
3884 			flags = RF_PREFETCHABLE;
3885 			/* FALLTHROUGH */
3886 		case PCIM_EA_P_VF_MEM:
3887 		case PCIM_EA_P_MEM:
3888 			type = SYS_RES_MEMORY;
3889 			break;
3890 		case PCIM_EA_P_IO:
3891 			type = SYS_RES_IOPORT;
3892 			break;
3893 		default:
3894 			continue;
3895 		}
3896 
3897 		if (alloc_iov != 0) {
3898 #ifdef PCI_IOV
3899 			/* Allocating IOV, confirm BEI matches */
3900 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3901 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3902 				continue;
3903 #else
3904 			continue;
3905 #endif
3906 		} else {
3907 			/* Allocating BAR, confirm BEI matches */
3908 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3909 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3910 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3911 				continue;
3912 		}
3913 
3914 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3915 		if (rid < 0)
3916 			continue;
3917 
3918 		/* Skip resources already allocated by EA */
3919 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3920 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3921 			continue;
3922 
3923 		start = ea->eae_base;
3924 		count = ea->eae_max_offset + 1;
3925 #ifdef PCI_IOV
3926 		if (iov != NULL)
3927 			count = count * iov->iov_num_vfs;
3928 #endif
3929 		end = start + count - 1;
3930 		if (count == 0)
3931 			continue;
3932 
3933 		resource_list_add(rl, type, rid, start, end, count);
3934 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3935 		    flags);
3936 		if (res == NULL) {
3937 			resource_list_delete(rl, type, rid);
3938 
3939 			/*
3940 			 * Failed to allocate using EA, disable entry.
3941 			 * Another attempt to allocation will be performed
3942 			 * further, but this time using legacy BAR registers
3943 			 */
3944 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3945 			tmp &= ~PCIM_EA_ENABLE;
3946 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3947 
3948 			/*
3949 			 * Disabling entry might fail in case it is hardwired.
3950 			 * Read flags again to match current status.
3951 			 */
3952 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3953 
3954 			continue;
3955 		}
3956 
3957 		/* As per specification, fill BAR with zeros */
3958 		pci_write_config(dev, rid, 0, 4);
3959 	}
3960 }
3961 
3962 void
3963 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3964 {
3965 	struct pci_devinfo *dinfo;
3966 	pcicfgregs *cfg;
3967 	struct resource_list *rl;
3968 	const struct pci_quirk *q;
3969 	uint32_t devid;
3970 	int i;
3971 
3972 	dinfo = device_get_ivars(dev);
3973 	cfg = &dinfo->cfg;
3974 	rl = &dinfo->resources;
3975 	devid = (cfg->device << 16) | cfg->vendor;
3976 
3977 	/* Allocate resources using Enhanced Allocation */
3978 	pci_add_resources_ea(bus, dev, 0);
3979 
3980 	/* ATA devices needs special map treatment */
3981 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
3982 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3983 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3984 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3985 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
3986 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
3987 	else
3988 		for (i = 0; i < cfg->nummaps;) {
3989 			/* Skip resources already managed by EA */
3990 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3991 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3992 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3993 				i++;
3994 				continue;
3995 			}
3996 
3997 			/*
3998 			 * Skip quirked resources.
3999 			 */
4000 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4001 				if (q->devid == devid &&
4002 				    q->type == PCI_QUIRK_UNMAP_REG &&
4003 				    q->arg1 == PCIR_BAR(i))
4004 					break;
4005 			if (q->devid != 0) {
4006 				i++;
4007 				continue;
4008 			}
4009 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4010 			    prefetchmask & (1 << i));
4011 		}
4012 
4013 	/*
4014 	 * Add additional, quirked resources.
4015 	 */
4016 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4017 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4018 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4019 
4020 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4021 #ifdef __PCI_REROUTE_INTERRUPT
4022 		/*
4023 		 * Try to re-route interrupts. Sometimes the BIOS or
4024 		 * firmware may leave bogus values in these registers.
4025 		 * If the re-route fails, then just stick with what we
4026 		 * have.
4027 		 */
4028 		pci_assign_interrupt(bus, dev, 1);
4029 #else
4030 		pci_assign_interrupt(bus, dev, 0);
4031 #endif
4032 	}
4033 
4034 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4035 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4036 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4037 			xhci_early_takeover(dev);
4038 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4039 			ehci_early_takeover(dev);
4040 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4041 			ohci_early_takeover(dev);
4042 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4043 			uhci_early_takeover(dev);
4044 	}
4045 
4046 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4047 	/*
4048 	 * Reserve resources for secondary bus ranges behind bridge
4049 	 * devices.
4050 	 */
4051 	pci_reserve_secbus(bus, dev, cfg, rl);
4052 #endif
4053 }
4054 
4055 static struct pci_devinfo *
4056 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4057     int slot, int func)
4058 {
4059 	struct pci_devinfo *dinfo;
4060 
4061 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4062 	if (dinfo != NULL)
4063 		pci_add_child(dev, dinfo);
4064 
4065 	return (dinfo);
4066 }
4067 
4068 void
4069 pci_add_children(device_t dev, int domain, int busno)
4070 {
4071 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4072 	device_t pcib = device_get_parent(dev);
4073 	struct pci_devinfo *dinfo;
4074 	int maxslots;
4075 	int s, f, pcifunchigh;
4076 	uint8_t hdrtype;
4077 	int first_func;
4078 
4079 	/*
4080 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4081 	 * enable ARI.  We must enable ARI before detecting the rest of the
4082 	 * functions on this bus as ARI changes the set of slots and functions
4083 	 * that are legal on this bus.
4084 	 */
4085 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4086 	if (dinfo != NULL && pci_enable_ari)
4087 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4088 
4089 	/*
4090 	 * Start looking for new devices on slot 0 at function 1 because we
4091 	 * just identified the device at slot 0, function 0.
4092 	 */
4093 	first_func = 1;
4094 
4095 	maxslots = PCIB_MAXSLOTS(pcib);
4096 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4097 		pcifunchigh = 0;
4098 		f = 0;
4099 		DELAY(1);
4100 		hdrtype = REG(PCIR_HDRTYPE, 1);
4101 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4102 			continue;
4103 		if (hdrtype & PCIM_MFDEV)
4104 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4105 		for (f = first_func; f <= pcifunchigh; f++)
4106 			pci_identify_function(pcib, dev, domain, busno, s, f);
4107 	}
4108 #undef REG
4109 }
4110 
4111 int
4112 pci_rescan_method(device_t dev)
4113 {
4114 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4115 	device_t pcib = device_get_parent(dev);
4116 	device_t child, *devlist, *unchanged;
4117 	int devcount, error, i, j, maxslots, oldcount;
4118 	int busno, domain, s, f, pcifunchigh;
4119 	uint8_t hdrtype;
4120 
4121 	/* No need to check for ARI on a rescan. */
4122 	error = device_get_children(dev, &devlist, &devcount);
4123 	if (error)
4124 		return (error);
4125 	if (devcount != 0) {
4126 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4127 		    M_NOWAIT | M_ZERO);
4128 		if (unchanged == NULL) {
4129 			free(devlist, M_TEMP);
4130 			return (ENOMEM);
4131 		}
4132 	} else
4133 		unchanged = NULL;
4134 
4135 	domain = pcib_get_domain(dev);
4136 	busno = pcib_get_bus(dev);
4137 	maxslots = PCIB_MAXSLOTS(pcib);
4138 	for (s = 0; s <= maxslots; s++) {
4139 		/* If function 0 is not present, skip to the next slot. */
4140 		f = 0;
4141 		if (REG(PCIR_VENDOR, 2) == 0xffff)
4142 			continue;
4143 		pcifunchigh = 0;
4144 		hdrtype = REG(PCIR_HDRTYPE, 1);
4145 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4146 			continue;
4147 		if (hdrtype & PCIM_MFDEV)
4148 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4149 		for (f = 0; f <= pcifunchigh; f++) {
4150 			if (REG(PCIR_VENDOR, 2) == 0xffff)
4151 				continue;
4152 
4153 			/*
4154 			 * Found a valid function.  Check if a
4155 			 * device_t for this device already exists.
4156 			 */
4157 			for (i = 0; i < devcount; i++) {
4158 				child = devlist[i];
4159 				if (child == NULL)
4160 					continue;
4161 				if (pci_get_slot(child) == s &&
4162 				    pci_get_function(child) == f) {
4163 					unchanged[i] = child;
4164 					goto next_func;
4165 				}
4166 			}
4167 
4168 			pci_identify_function(pcib, dev, domain, busno, s, f);
4169 		next_func:;
4170 		}
4171 	}
4172 
4173 	/* Remove devices that are no longer present. */
4174 	for (i = 0; i < devcount; i++) {
4175 		if (unchanged[i] != NULL)
4176 			continue;
4177 		device_delete_child(dev, devlist[i]);
4178 	}
4179 
4180 	free(devlist, M_TEMP);
4181 	oldcount = devcount;
4182 
4183 	/* Try to attach the devices just added. */
4184 	error = device_get_children(dev, &devlist, &devcount);
4185 	if (error) {
4186 		free(unchanged, M_TEMP);
4187 		return (error);
4188 	}
4189 
4190 	for (i = 0; i < devcount; i++) {
4191 		for (j = 0; j < oldcount; j++) {
4192 			if (devlist[i] == unchanged[j])
4193 				goto next_device;
4194 		}
4195 
4196 		device_probe_and_attach(devlist[i]);
4197 	next_device:;
4198 	}
4199 
4200 	free(unchanged, M_TEMP);
4201 	free(devlist, M_TEMP);
4202 	return (0);
4203 #undef REG
4204 }
4205 
4206 #ifdef PCI_IOV
4207 device_t
4208 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4209     uint16_t did)
4210 {
4211 	struct pci_devinfo *vf_dinfo;
4212 	device_t pcib;
4213 	int busno, slot, func;
4214 
4215 	pcib = device_get_parent(bus);
4216 
4217 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4218 
4219 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4220 	    slot, func, vid, did);
4221 
4222 	vf_dinfo->cfg.flags |= PCICFG_VF;
4223 	pci_add_child(bus, vf_dinfo);
4224 
4225 	return (vf_dinfo->cfg.dev);
4226 }
4227 
4228 device_t
4229 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4230     uint16_t vid, uint16_t did)
4231 {
4232 
4233 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4234 }
4235 #endif
4236 
4237 static void
4238 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4239 {
4240 	int aer;
4241 	uint32_t r;
4242 	uint16_t r2;
4243 
4244 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4245 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4246 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4247 		    PCIER_ROOT_CTL, 2);
4248 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4249 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4250 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4251 		    PCIER_ROOT_CTL, r2, 2);
4252 	}
4253 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4254 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4255 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4256 		if (r != 0 && bootverbose) {
4257 			pci_printf(&dinfo->cfg,
4258 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4259 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4260 			    4));
4261 		}
4262 
4263 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4264 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4265 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4266 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4267 		    PCIM_AER_UC_POISONED_TLP |
4268 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4269 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4270 		    PCIM_AER_UC_COMPLETER_ABORT |
4271 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4272 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4273 		    PCIM_AER_UC_MALFORMED_TLP |
4274 		    PCIM_AER_UC_ECRC_ERROR |
4275 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4276 		    PCIM_AER_UC_ACS_VIOLATION |
4277 		    PCIM_AER_UC_INTERNAL_ERROR |
4278 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4279 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4280 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4281 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4282 
4283 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4284 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4285 		if (r != 0 && bootverbose) {
4286 			pci_printf(&dinfo->cfg,
4287 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4288 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4289 			    4));
4290 		}
4291 
4292 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4293 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4294 		    PCIM_AER_COR_BAD_TLP |
4295 		    PCIM_AER_COR_BAD_DLLP |
4296 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4297 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4298 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4299 		    PCIM_AER_COR_INTERNAL_ERROR |
4300 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4301 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4302 
4303 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4304 		    PCIER_DEVICE_CTL, 2);
4305 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4306 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4307 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4308 		    PCIER_DEVICE_CTL, r, 2);
4309 	}
4310 }
4311 
4312 void
4313 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4314 {
4315 	device_t dev;
4316 
4317 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4318 	device_set_ivars(dev, dinfo);
4319 	resource_list_init(&dinfo->resources);
4320 	pci_cfg_save(dev, dinfo, 0);
4321 	pci_cfg_restore(dev, dinfo);
4322 	pci_print_verbose(dinfo);
4323 	pci_add_resources(bus, dev, 0, 0);
4324 	pci_child_added(dinfo->cfg.dev);
4325 
4326 	if (pci_clear_aer_on_attach)
4327 		pci_add_child_clear_aer(dev, dinfo);
4328 
4329 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4330 }
4331 
4332 void
4333 pci_child_added_method(device_t dev, device_t child)
4334 {
4335 
4336 }
4337 
4338 static int
4339 pci_probe(device_t dev)
4340 {
4341 
4342 	device_set_desc(dev, "PCI bus");
4343 
4344 	/* Allow other subclasses to override this driver. */
4345 	return (BUS_PROBE_GENERIC);
4346 }
4347 
4348 int
4349 pci_attach_common(device_t dev)
4350 {
4351 	struct pci_softc *sc;
4352 	int busno, domain;
4353 #ifdef PCI_RES_BUS
4354 	int rid;
4355 #endif
4356 
4357 	sc = device_get_softc(dev);
4358 	domain = pcib_get_domain(dev);
4359 	busno = pcib_get_bus(dev);
4360 #ifdef PCI_RES_BUS
4361 	rid = 0;
4362 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4363 	    1, 0);
4364 	if (sc->sc_bus == NULL) {
4365 		device_printf(dev, "failed to allocate bus number\n");
4366 		return (ENXIO);
4367 	}
4368 #endif
4369 	if (bootverbose)
4370 		device_printf(dev, "domain=%d, physical bus=%d\n",
4371 		    domain, busno);
4372 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4373 	return (0);
4374 }
4375 
4376 static int
4377 pci_attach(device_t dev)
4378 {
4379 	int busno, domain, error;
4380 
4381 	error = pci_attach_common(dev);
4382 	if (error)
4383 		return (error);
4384 
4385 	/*
4386 	 * Since there can be multiple independently numbered PCI
4387 	 * buses on systems with multiple PCI domains, we can't use
4388 	 * the unit number to decide which bus we are probing. We ask
4389 	 * the parent pcib what our domain and bus numbers are.
4390 	 */
4391 	domain = pcib_get_domain(dev);
4392 	busno = pcib_get_bus(dev);
4393 	pci_add_children(dev, domain, busno);
4394 	return (bus_generic_attach(dev));
4395 }
4396 
4397 static int
4398 pci_detach(device_t dev)
4399 {
4400 #ifdef PCI_RES_BUS
4401 	struct pci_softc *sc;
4402 #endif
4403 	int error;
4404 
4405 	error = bus_generic_detach(dev);
4406 	if (error)
4407 		return (error);
4408 #ifdef PCI_RES_BUS
4409 	sc = device_get_softc(dev);
4410 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4411 	if (error)
4412 		return (error);
4413 #endif
4414 	return (device_delete_children(dev));
4415 }
4416 
4417 static void
4418 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4419 {
4420 	int line, unit;
4421 	const char *at;
4422 	char me1[24], me2[32];
4423 	uint8_t b, s, f;
4424 	uint32_t d;
4425 
4426 	d = pci_get_domain(child);
4427 	b = pci_get_bus(child);
4428 	s = pci_get_slot(child);
4429 	f = pci_get_function(child);
4430 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4431 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4432 	line = 0;
4433 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4434 		resource_string_value(name, unit, "at", &at);
4435 		if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4436 			continue; /* No match, try next candidate */
4437 		*unitp = unit;
4438 		return;
4439 	}
4440 }
4441 
4442 static void
4443 pci_set_power_child(device_t dev, device_t child, int state)
4444 {
4445 	device_t pcib;
4446 	int dstate;
4447 
4448 	/*
4449 	 * Set the device to the given state.  If the firmware suggests
4450 	 * a different power state, use it instead.  If power management
4451 	 * is not present, the firmware is responsible for managing
4452 	 * device power.  Skip children who aren't attached since they
4453 	 * are handled separately.
4454 	 */
4455 	pcib = device_get_parent(dev);
4456 	dstate = state;
4457 	if (device_is_attached(child) &&
4458 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4459 		pci_set_powerstate(child, dstate);
4460 }
4461 
4462 int
4463 pci_suspend_child(device_t dev, device_t child)
4464 {
4465 	struct pci_devinfo *dinfo;
4466 	struct resource_list_entry *rle;
4467 	int error;
4468 
4469 	dinfo = device_get_ivars(child);
4470 
4471 	/*
4472 	 * Save the PCI configuration space for the child and set the
4473 	 * device in the appropriate power state for this sleep state.
4474 	 */
4475 	pci_cfg_save(child, dinfo, 0);
4476 
4477 	/* Suspend devices before potentially powering them down. */
4478 	error = bus_generic_suspend_child(dev, child);
4479 
4480 	if (error)
4481 		return (error);
4482 
4483 	if (pci_do_power_suspend) {
4484 		/*
4485 		 * Make sure this device's interrupt handler is not invoked
4486 		 * in the case the device uses a shared interrupt that can
4487 		 * be raised by some other device.
4488 		 * This is applicable only to regular (legacy) PCI interrupts
4489 		 * as MSI/MSI-X interrupts are never shared.
4490 		 */
4491 		rle = resource_list_find(&dinfo->resources,
4492 		    SYS_RES_IRQ, 0);
4493 		if (rle != NULL && rle->res != NULL)
4494 			(void)bus_suspend_intr(child, rle->res);
4495 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4496 	}
4497 
4498 	return (0);
4499 }
4500 
4501 int
4502 pci_resume_child(device_t dev, device_t child)
4503 {
4504 	struct pci_devinfo *dinfo;
4505 	struct resource_list_entry *rle;
4506 
4507 	if (pci_do_power_resume)
4508 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4509 
4510 	dinfo = device_get_ivars(child);
4511 	pci_cfg_restore(child, dinfo);
4512 	if (!device_is_attached(child))
4513 		pci_cfg_save(child, dinfo, 1);
4514 
4515 	bus_generic_resume_child(dev, child);
4516 
4517 	/*
4518 	 * Allow interrupts only after fully resuming the driver and hardware.
4519 	 */
4520 	if (pci_do_power_suspend) {
4521 		/* See pci_suspend_child for details. */
4522 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4523 		if (rle != NULL && rle->res != NULL)
4524 			(void)bus_resume_intr(child, rle->res);
4525 	}
4526 
4527 	return (0);
4528 }
4529 
4530 int
4531 pci_resume(device_t dev)
4532 {
4533 	device_t child, *devlist;
4534 	int error, i, numdevs;
4535 
4536 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4537 		return (error);
4538 
4539 	/*
4540 	 * Resume critical devices first, then everything else later.
4541 	 */
4542 	for (i = 0; i < numdevs; i++) {
4543 		child = devlist[i];
4544 		switch (pci_get_class(child)) {
4545 		case PCIC_DISPLAY:
4546 		case PCIC_MEMORY:
4547 		case PCIC_BRIDGE:
4548 		case PCIC_BASEPERIPH:
4549 			BUS_RESUME_CHILD(dev, child);
4550 			break;
4551 		}
4552 	}
4553 	for (i = 0; i < numdevs; i++) {
4554 		child = devlist[i];
4555 		switch (pci_get_class(child)) {
4556 		case PCIC_DISPLAY:
4557 		case PCIC_MEMORY:
4558 		case PCIC_BRIDGE:
4559 		case PCIC_BASEPERIPH:
4560 			break;
4561 		default:
4562 			BUS_RESUME_CHILD(dev, child);
4563 		}
4564 	}
4565 	free(devlist, M_TEMP);
4566 	return (0);
4567 }
4568 
4569 static void
4570 pci_load_vendor_data(void)
4571 {
4572 	caddr_t data;
4573 	void *ptr;
4574 	size_t sz;
4575 
4576 	data = preload_search_by_type("pci_vendor_data");
4577 	if (data != NULL) {
4578 		ptr = preload_fetch_addr(data);
4579 		sz = preload_fetch_size(data);
4580 		if (ptr != NULL && sz != 0) {
4581 			pci_vendordata = ptr;
4582 			pci_vendordata_size = sz;
4583 			/* terminate the database */
4584 			pci_vendordata[pci_vendordata_size] = '\n';
4585 		}
4586 	}
4587 }
4588 
4589 void
4590 pci_driver_added(device_t dev, driver_t *driver)
4591 {
4592 	int numdevs;
4593 	device_t *devlist;
4594 	device_t child;
4595 	struct pci_devinfo *dinfo;
4596 	int i;
4597 
4598 	if (bootverbose)
4599 		device_printf(dev, "driver added\n");
4600 	DEVICE_IDENTIFY(driver, dev);
4601 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4602 		return;
4603 	for (i = 0; i < numdevs; i++) {
4604 		child = devlist[i];
4605 		if (device_get_state(child) != DS_NOTPRESENT)
4606 			continue;
4607 		dinfo = device_get_ivars(child);
4608 		pci_print_verbose(dinfo);
4609 		if (bootverbose)
4610 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4611 		pci_cfg_restore(child, dinfo);
4612 		if (device_probe_and_attach(child) != 0)
4613 			pci_child_detached(dev, child);
4614 	}
4615 	free(devlist, M_TEMP);
4616 }
4617 
4618 int
4619 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4620     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4621 {
4622 	struct pci_devinfo *dinfo;
4623 	struct msix_table_entry *mte;
4624 	struct msix_vector *mv;
4625 	uint64_t addr;
4626 	uint32_t data;
4627 	void *cookie;
4628 	int error, rid;
4629 
4630 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4631 	    arg, &cookie);
4632 	if (error)
4633 		return (error);
4634 
4635 	/* If this is not a direct child, just bail out. */
4636 	if (device_get_parent(child) != dev) {
4637 		*cookiep = cookie;
4638 		return(0);
4639 	}
4640 
4641 	rid = rman_get_rid(irq);
4642 	if (rid == 0) {
4643 		/* Make sure that INTx is enabled */
4644 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4645 	} else {
4646 		/*
4647 		 * Check to see if the interrupt is MSI or MSI-X.
4648 		 * Ask our parent to map the MSI and give
4649 		 * us the address and data register values.
4650 		 * If we fail for some reason, teardown the
4651 		 * interrupt handler.
4652 		 */
4653 		dinfo = device_get_ivars(child);
4654 		if (dinfo->cfg.msi.msi_alloc > 0) {
4655 			if (dinfo->cfg.msi.msi_addr == 0) {
4656 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4657 			    ("MSI has handlers, but vectors not mapped"));
4658 				error = PCIB_MAP_MSI(device_get_parent(dev),
4659 				    child, rman_get_start(irq), &addr, &data);
4660 				if (error)
4661 					goto bad;
4662 				dinfo->cfg.msi.msi_addr = addr;
4663 				dinfo->cfg.msi.msi_data = data;
4664 			}
4665 			if (dinfo->cfg.msi.msi_handlers == 0)
4666 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4667 				    dinfo->cfg.msi.msi_data);
4668 			dinfo->cfg.msi.msi_handlers++;
4669 		} else {
4670 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4671 			    ("No MSI or MSI-X interrupts allocated"));
4672 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4673 			    ("MSI-X index too high"));
4674 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4675 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4676 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4677 			KASSERT(mv->mv_irq == rman_get_start(irq),
4678 			    ("IRQ mismatch"));
4679 			if (mv->mv_address == 0) {
4680 				KASSERT(mte->mte_handlers == 0,
4681 		    ("MSI-X table entry has handlers, but vector not mapped"));
4682 				error = PCIB_MAP_MSI(device_get_parent(dev),
4683 				    child, rman_get_start(irq), &addr, &data);
4684 				if (error)
4685 					goto bad;
4686 				mv->mv_address = addr;
4687 				mv->mv_data = data;
4688 			}
4689 
4690 			/*
4691 			 * The MSIX table entry must be made valid by
4692 			 * incrementing the mte_handlers before
4693 			 * calling pci_enable_msix() and
4694 			 * pci_resume_msix(). Else the MSIX rewrite
4695 			 * table quirk will not work as expected.
4696 			 */
4697 			mte->mte_handlers++;
4698 			if (mte->mte_handlers == 1) {
4699 				pci_enable_msix(child, rid - 1, mv->mv_address,
4700 				    mv->mv_data);
4701 				pci_unmask_msix(child, rid - 1);
4702 			}
4703 		}
4704 
4705 		/*
4706 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4707 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4708 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4709 		 */
4710 		if (!pci_has_quirk(pci_get_devid(child),
4711 		    PCI_QUIRK_MSI_INTX_BUG))
4712 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4713 		else
4714 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4715 	bad:
4716 		if (error) {
4717 			(void)bus_generic_teardown_intr(dev, child, irq,
4718 			    cookie);
4719 			return (error);
4720 		}
4721 	}
4722 	*cookiep = cookie;
4723 	return (0);
4724 }
4725 
4726 int
4727 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4728     void *cookie)
4729 {
4730 	struct msix_table_entry *mte;
4731 	struct resource_list_entry *rle;
4732 	struct pci_devinfo *dinfo;
4733 	int error, rid;
4734 
4735 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4736 		return (EINVAL);
4737 
4738 	/* If this isn't a direct child, just bail out */
4739 	if (device_get_parent(child) != dev)
4740 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4741 
4742 	rid = rman_get_rid(irq);
4743 	if (rid == 0) {
4744 		/* Mask INTx */
4745 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4746 	} else {
4747 		/*
4748 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4749 		 * decrement the appropriate handlers count and mask the
4750 		 * MSI-X message, or disable MSI messages if the count
4751 		 * drops to 0.
4752 		 */
4753 		dinfo = device_get_ivars(child);
4754 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4755 		if (rle->res != irq)
4756 			return (EINVAL);
4757 		if (dinfo->cfg.msi.msi_alloc > 0) {
4758 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4759 			    ("MSI-X index too high"));
4760 			if (dinfo->cfg.msi.msi_handlers == 0)
4761 				return (EINVAL);
4762 			dinfo->cfg.msi.msi_handlers--;
4763 			if (dinfo->cfg.msi.msi_handlers == 0)
4764 				pci_disable_msi(child);
4765 		} else {
4766 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4767 			    ("No MSI or MSI-X interrupts allocated"));
4768 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4769 			    ("MSI-X index too high"));
4770 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4771 			if (mte->mte_handlers == 0)
4772 				return (EINVAL);
4773 			mte->mte_handlers--;
4774 			if (mte->mte_handlers == 0)
4775 				pci_mask_msix(child, rid - 1);
4776 		}
4777 	}
4778 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4779 	if (rid > 0)
4780 		KASSERT(error == 0,
4781 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4782 	return (error);
4783 }
4784 
4785 int
4786 pci_print_child(device_t dev, device_t child)
4787 {
4788 	struct pci_devinfo *dinfo;
4789 	struct resource_list *rl;
4790 	int retval = 0;
4791 
4792 	dinfo = device_get_ivars(child);
4793 	rl = &dinfo->resources;
4794 
4795 	retval += bus_print_child_header(dev, child);
4796 
4797 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4798 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4799 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4800 	if (device_get_flags(dev))
4801 		retval += printf(" flags %#x", device_get_flags(dev));
4802 
4803 	retval += printf(" at device %d.%d", pci_get_slot(child),
4804 	    pci_get_function(child));
4805 
4806 	retval += bus_print_child_domain(dev, child);
4807 	retval += bus_print_child_footer(dev, child);
4808 
4809 	return (retval);
4810 }
4811 
4812 static const struct
4813 {
4814 	int		class;
4815 	int		subclass;
4816 	int		report; /* 0 = bootverbose, 1 = always */
4817 	const char	*desc;
4818 } pci_nomatch_tab[] = {
4819 	{PCIC_OLD,		-1,			1, "old"},
4820 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4821 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4822 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4823 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4824 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4825 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4826 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4827 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4828 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4829 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4830 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4831 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4832 	{PCIC_NETWORK,		-1,			1, "network"},
4833 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4834 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4835 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4836 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4837 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4838 	{PCIC_DISPLAY,		-1,			1, "display"},
4839 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4840 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4841 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4842 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4843 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4844 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4845 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4846 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4847 	{PCIC_MEMORY,		-1,			1, "memory"},
4848 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4849 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4850 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4851 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4852 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4853 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4854 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4855 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4856 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4857 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4858 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4859 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4860 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4861 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4862 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4863 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4864 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4865 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4866 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4867 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4868 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4869 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4870 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4871 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4872 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4873 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4874 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4875 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4876 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4877 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4878 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4879 	{PCIC_DOCKING,		-1,			1, "docking station"},
4880 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4881 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4882 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4883 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4884 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4885 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4886 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
4887 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
4888 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
4889 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
4890 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
4891 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
4892 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
4893 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
4894 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
4895 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
4896 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
4897 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
4898 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
4899 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
4900 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
4901 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
4902 	{PCIC_DASP,		-1,			0, "dasp"},
4903 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
4904 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
4905 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
4906 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
4907 	{0, 0, 0,		NULL}
4908 };
4909 
4910 void
4911 pci_probe_nomatch(device_t dev, device_t child)
4912 {
4913 	int i, report;
4914 	const char *cp, *scp;
4915 	char *device;
4916 
4917 	/*
4918 	 * Look for a listing for this device in a loaded device database.
4919 	 */
4920 	report = 1;
4921 	if ((device = pci_describe_device(child)) != NULL) {
4922 		device_printf(dev, "<%s>", device);
4923 		free(device, M_DEVBUF);
4924 	} else {
4925 		/*
4926 		 * Scan the class/subclass descriptions for a general
4927 		 * description.
4928 		 */
4929 		cp = "unknown";
4930 		scp = NULL;
4931 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4932 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4933 				if (pci_nomatch_tab[i].subclass == -1) {
4934 					cp = pci_nomatch_tab[i].desc;
4935 					report = pci_nomatch_tab[i].report;
4936 				} else if (pci_nomatch_tab[i].subclass ==
4937 				    pci_get_subclass(child)) {
4938 					scp = pci_nomatch_tab[i].desc;
4939 					report = pci_nomatch_tab[i].report;
4940 				}
4941 			}
4942 		}
4943 		if (report || bootverbose) {
4944 			device_printf(dev, "<%s%s%s>",
4945 			    cp ? cp : "",
4946 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
4947 			    scp ? scp : "");
4948 		}
4949 	}
4950 	if (report || bootverbose) {
4951 		printf(" at device %d.%d (no driver attached)\n",
4952 		    pci_get_slot(child), pci_get_function(child));
4953 	}
4954 	pci_cfg_save(child, device_get_ivars(child), 1);
4955 }
4956 
4957 void
4958 pci_child_detached(device_t dev, device_t child)
4959 {
4960 	struct pci_devinfo *dinfo;
4961 	struct resource_list *rl;
4962 
4963 	dinfo = device_get_ivars(child);
4964 	rl = &dinfo->resources;
4965 
4966 	/*
4967 	 * Have to deallocate IRQs before releasing any MSI messages and
4968 	 * have to release MSI messages before deallocating any memory
4969 	 * BARs.
4970 	 */
4971 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4972 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4973 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4974 		pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4975 		(void)pci_release_msi(child);
4976 	}
4977 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4978 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4979 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4980 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4981 #ifdef PCI_RES_BUS
4982 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4983 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4984 #endif
4985 
4986 	pci_cfg_save(child, dinfo, 1);
4987 }
4988 
4989 /*
4990  * Parse the PCI device database, if loaded, and return a pointer to a
4991  * description of the device.
4992  *
4993  * The database is flat text formatted as follows:
4994  *
4995  * Any line not in a valid format is ignored.
4996  * Lines are terminated with newline '\n' characters.
4997  *
4998  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4999  * the vendor name.
5000  *
5001  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5002  * - devices cannot be listed without a corresponding VENDOR line.
5003  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5004  * another TAB, then the device name.
5005  */
5006 
5007 /*
5008  * Assuming (ptr) points to the beginning of a line in the database,
5009  * return the vendor or device and description of the next entry.
5010  * The value of (vendor) or (device) inappropriate for the entry type
5011  * is set to -1.  Returns nonzero at the end of the database.
5012  *
5013  * Note that this is slightly unrobust in the face of corrupt data;
5014  * we attempt to safeguard against this by spamming the end of the
5015  * database with a newline when we initialise.
5016  */
5017 static int
5018 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5019 {
5020 	char	*cp = *ptr;
5021 	int	left;
5022 
5023 	*device = -1;
5024 	*vendor = -1;
5025 	**desc = '\0';
5026 	for (;;) {
5027 		left = pci_vendordata_size - (cp - pci_vendordata);
5028 		if (left <= 0) {
5029 			*ptr = cp;
5030 			return(1);
5031 		}
5032 
5033 		/* vendor entry? */
5034 		if (*cp != '\t' &&
5035 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5036 			break;
5037 		/* device entry? */
5038 		if (*cp == '\t' &&
5039 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5040 			break;
5041 
5042 		/* skip to next line */
5043 		while (*cp != '\n' && left > 0) {
5044 			cp++;
5045 			left--;
5046 		}
5047 		if (*cp == '\n') {
5048 			cp++;
5049 			left--;
5050 		}
5051 	}
5052 	/* skip to next line */
5053 	while (*cp != '\n' && left > 0) {
5054 		cp++;
5055 		left--;
5056 	}
5057 	if (*cp == '\n' && left > 0)
5058 		cp++;
5059 	*ptr = cp;
5060 	return(0);
5061 }
5062 
5063 static char *
5064 pci_describe_device(device_t dev)
5065 {
5066 	int	vendor, device;
5067 	char	*desc, *vp, *dp, *line;
5068 
5069 	desc = vp = dp = NULL;
5070 
5071 	/*
5072 	 * If we have no vendor data, we can't do anything.
5073 	 */
5074 	if (pci_vendordata == NULL)
5075 		goto out;
5076 
5077 	/*
5078 	 * Scan the vendor data looking for this device
5079 	 */
5080 	line = pci_vendordata;
5081 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5082 		goto out;
5083 	for (;;) {
5084 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5085 			goto out;
5086 		if (vendor == pci_get_vendor(dev))
5087 			break;
5088 	}
5089 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5090 		goto out;
5091 	for (;;) {
5092 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5093 			*dp = 0;
5094 			break;
5095 		}
5096 		if (vendor != -1) {
5097 			*dp = 0;
5098 			break;
5099 		}
5100 		if (device == pci_get_device(dev))
5101 			break;
5102 	}
5103 	if (dp[0] == '\0')
5104 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5105 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5106 	    NULL)
5107 		sprintf(desc, "%s, %s", vp, dp);
5108 out:
5109 	if (vp != NULL)
5110 		free(vp, M_DEVBUF);
5111 	if (dp != NULL)
5112 		free(dp, M_DEVBUF);
5113 	return(desc);
5114 }
5115 
5116 int
5117 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5118 {
5119 	struct pci_devinfo *dinfo;
5120 	pcicfgregs *cfg;
5121 
5122 	dinfo = device_get_ivars(child);
5123 	cfg = &dinfo->cfg;
5124 
5125 	switch (which) {
5126 	case PCI_IVAR_ETHADDR:
5127 		/*
5128 		 * The generic accessor doesn't deal with failure, so
5129 		 * we set the return value, then return an error.
5130 		 */
5131 		*((uint8_t **) result) = NULL;
5132 		return (EINVAL);
5133 	case PCI_IVAR_SUBVENDOR:
5134 		*result = cfg->subvendor;
5135 		break;
5136 	case PCI_IVAR_SUBDEVICE:
5137 		*result = cfg->subdevice;
5138 		break;
5139 	case PCI_IVAR_VENDOR:
5140 		*result = cfg->vendor;
5141 		break;
5142 	case PCI_IVAR_DEVICE:
5143 		*result = cfg->device;
5144 		break;
5145 	case PCI_IVAR_DEVID:
5146 		*result = (cfg->device << 16) | cfg->vendor;
5147 		break;
5148 	case PCI_IVAR_CLASS:
5149 		*result = cfg->baseclass;
5150 		break;
5151 	case PCI_IVAR_SUBCLASS:
5152 		*result = cfg->subclass;
5153 		break;
5154 	case PCI_IVAR_PROGIF:
5155 		*result = cfg->progif;
5156 		break;
5157 	case PCI_IVAR_REVID:
5158 		*result = cfg->revid;
5159 		break;
5160 	case PCI_IVAR_INTPIN:
5161 		*result = cfg->intpin;
5162 		break;
5163 	case PCI_IVAR_IRQ:
5164 		*result = cfg->intline;
5165 		break;
5166 	case PCI_IVAR_DOMAIN:
5167 		*result = cfg->domain;
5168 		break;
5169 	case PCI_IVAR_BUS:
5170 		*result = cfg->bus;
5171 		break;
5172 	case PCI_IVAR_SLOT:
5173 		*result = cfg->slot;
5174 		break;
5175 	case PCI_IVAR_FUNCTION:
5176 		*result = cfg->func;
5177 		break;
5178 	case PCI_IVAR_CMDREG:
5179 		*result = cfg->cmdreg;
5180 		break;
5181 	case PCI_IVAR_CACHELNSZ:
5182 		*result = cfg->cachelnsz;
5183 		break;
5184 	case PCI_IVAR_MINGNT:
5185 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5186 			*result = -1;
5187 			return (EINVAL);
5188 		}
5189 		*result = cfg->mingnt;
5190 		break;
5191 	case PCI_IVAR_MAXLAT:
5192 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5193 			*result = -1;
5194 			return (EINVAL);
5195 		}
5196 		*result = cfg->maxlat;
5197 		break;
5198 	case PCI_IVAR_LATTIMER:
5199 		*result = cfg->lattimer;
5200 		break;
5201 	default:
5202 		return (ENOENT);
5203 	}
5204 	return (0);
5205 }
5206 
5207 int
5208 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5209 {
5210 	struct pci_devinfo *dinfo;
5211 
5212 	dinfo = device_get_ivars(child);
5213 
5214 	switch (which) {
5215 	case PCI_IVAR_INTPIN:
5216 		dinfo->cfg.intpin = value;
5217 		return (0);
5218 	case PCI_IVAR_ETHADDR:
5219 	case PCI_IVAR_SUBVENDOR:
5220 	case PCI_IVAR_SUBDEVICE:
5221 	case PCI_IVAR_VENDOR:
5222 	case PCI_IVAR_DEVICE:
5223 	case PCI_IVAR_DEVID:
5224 	case PCI_IVAR_CLASS:
5225 	case PCI_IVAR_SUBCLASS:
5226 	case PCI_IVAR_PROGIF:
5227 	case PCI_IVAR_REVID:
5228 	case PCI_IVAR_IRQ:
5229 	case PCI_IVAR_DOMAIN:
5230 	case PCI_IVAR_BUS:
5231 	case PCI_IVAR_SLOT:
5232 	case PCI_IVAR_FUNCTION:
5233 		return (EINVAL);	/* disallow for now */
5234 
5235 	default:
5236 		return (ENOENT);
5237 	}
5238 }
5239 
5240 #include "opt_ddb.h"
5241 #ifdef DDB
5242 #include <ddb/ddb.h>
5243 #include <sys/cons.h>
5244 
5245 /*
5246  * List resources based on pci map registers, used for within ddb
5247  */
5248 
5249 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5250 {
5251 	struct pci_devinfo *dinfo;
5252 	struct devlist *devlist_head;
5253 	struct pci_conf *p;
5254 	const char *name;
5255 	int i, error, none_count;
5256 
5257 	none_count = 0;
5258 	/* get the head of the device queue */
5259 	devlist_head = &pci_devq;
5260 
5261 	/*
5262 	 * Go through the list of devices and print out devices
5263 	 */
5264 	for (error = 0, i = 0,
5265 	     dinfo = STAILQ_FIRST(devlist_head);
5266 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5267 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5268 
5269 		/* Populate pd_name and pd_unit */
5270 		name = NULL;
5271 		if (dinfo->cfg.dev)
5272 			name = device_get_name(dinfo->cfg.dev);
5273 
5274 		p = &dinfo->conf;
5275 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5276 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5277 			(name && *name) ? name : "none",
5278 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5279 			none_count++,
5280 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5281 			p->pc_sel.pc_func, (p->pc_class << 16) |
5282 			(p->pc_subclass << 8) | p->pc_progif,
5283 			(p->pc_subdevice << 16) | p->pc_subvendor,
5284 			(p->pc_device << 16) | p->pc_vendor,
5285 			p->pc_revid, p->pc_hdr);
5286 	}
5287 }
5288 #endif /* DDB */
5289 
5290 static struct resource *
5291 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5292     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5293     u_int flags)
5294 {
5295 	struct pci_devinfo *dinfo = device_get_ivars(child);
5296 	struct resource_list *rl = &dinfo->resources;
5297 	struct resource *res;
5298 	struct pci_map *pm;
5299 	uint16_t cmd;
5300 	pci_addr_t map, testval;
5301 	int mapsize;
5302 
5303 	res = NULL;
5304 
5305 	/* If rid is managed by EA, ignore it */
5306 	if (pci_ea_is_enabled(child, *rid))
5307 		goto out;
5308 
5309 	pm = pci_find_bar(child, *rid);
5310 	if (pm != NULL) {
5311 		/* This is a BAR that we failed to allocate earlier. */
5312 		mapsize = pm->pm_size;
5313 		map = pm->pm_value;
5314 	} else {
5315 		/*
5316 		 * Weed out the bogons, and figure out how large the
5317 		 * BAR/map is.  BARs that read back 0 here are bogus
5318 		 * and unimplemented.  Note: atapci in legacy mode are
5319 		 * special and handled elsewhere in the code.  If you
5320 		 * have a atapci device in legacy mode and it fails
5321 		 * here, that other code is broken.
5322 		 */
5323 		pci_read_bar(child, *rid, &map, &testval, NULL);
5324 
5325 		/*
5326 		 * Determine the size of the BAR and ignore BARs with a size
5327 		 * of 0.  Device ROM BARs use a different mask value.
5328 		 */
5329 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5330 			mapsize = pci_romsize(testval);
5331 		else
5332 			mapsize = pci_mapsize(testval);
5333 		if (mapsize == 0)
5334 			goto out;
5335 		pm = pci_add_bar(child, *rid, map, mapsize);
5336 	}
5337 
5338 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5339 		if (type != SYS_RES_MEMORY) {
5340 			if (bootverbose)
5341 				device_printf(dev,
5342 				    "child %s requested type %d for rid %#x,"
5343 				    " but the BAR says it is an memio\n",
5344 				    device_get_nameunit(child), type, *rid);
5345 			goto out;
5346 		}
5347 	} else {
5348 		if (type != SYS_RES_IOPORT) {
5349 			if (bootverbose)
5350 				device_printf(dev,
5351 				    "child %s requested type %d for rid %#x,"
5352 				    " but the BAR says it is an ioport\n",
5353 				    device_get_nameunit(child), type, *rid);
5354 			goto out;
5355 		}
5356 	}
5357 
5358 	/*
5359 	 * For real BARs, we need to override the size that
5360 	 * the driver requests, because that's what the BAR
5361 	 * actually uses and we would otherwise have a
5362 	 * situation where we might allocate the excess to
5363 	 * another driver, which won't work.
5364 	 */
5365 	count = ((pci_addr_t)1 << mapsize) * num;
5366 	if (RF_ALIGNMENT(flags) < mapsize)
5367 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5368 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5369 		flags |= RF_PREFETCHABLE;
5370 
5371 	/*
5372 	 * Allocate enough resource, and then write back the
5373 	 * appropriate BAR for that resource.
5374 	 */
5375 	resource_list_add(rl, type, *rid, start, end, count);
5376 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5377 	    count, flags & ~RF_ACTIVE);
5378 	if (res == NULL) {
5379 		resource_list_delete(rl, type, *rid);
5380 		device_printf(child,
5381 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5382 		    count, *rid, type, start, end);
5383 		goto out;
5384 	}
5385 	if (bootverbose)
5386 		device_printf(child,
5387 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5388 		    count, *rid, type, rman_get_start(res));
5389 
5390 	/* Disable decoding via the CMD register before updating the BAR */
5391 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5392 	pci_write_config(child, PCIR_COMMAND,
5393 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5394 
5395 	map = rman_get_start(res);
5396 	pci_write_bar(child, pm, map);
5397 
5398 	/* Restore the original value of the CMD register */
5399 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5400 out:
5401 	return (res);
5402 }
5403 
5404 struct resource *
5405 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5406     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5407     u_int flags)
5408 {
5409 	struct pci_devinfo *dinfo;
5410 	struct resource_list *rl;
5411 	struct resource_list_entry *rle;
5412 	struct resource *res;
5413 	pcicfgregs *cfg;
5414 
5415 	/*
5416 	 * Perform lazy resource allocation
5417 	 */
5418 	dinfo = device_get_ivars(child);
5419 	rl = &dinfo->resources;
5420 	cfg = &dinfo->cfg;
5421 	switch (type) {
5422 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5423 	case PCI_RES_BUS:
5424 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5425 		    flags));
5426 #endif
5427 	case SYS_RES_IRQ:
5428 		/*
5429 		 * Can't alloc legacy interrupt once MSI messages have
5430 		 * been allocated.
5431 		 */
5432 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5433 		    cfg->msix.msix_alloc > 0))
5434 			return (NULL);
5435 
5436 		/*
5437 		 * If the child device doesn't have an interrupt
5438 		 * routed and is deserving of an interrupt, try to
5439 		 * assign it one.
5440 		 */
5441 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5442 		    (cfg->intpin != 0))
5443 			pci_assign_interrupt(dev, child, 0);
5444 		break;
5445 	case SYS_RES_IOPORT:
5446 	case SYS_RES_MEMORY:
5447 #ifdef NEW_PCIB
5448 		/*
5449 		 * PCI-PCI bridge I/O window resources are not BARs.
5450 		 * For those allocations just pass the request up the
5451 		 * tree.
5452 		 */
5453 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5454 			switch (*rid) {
5455 			case PCIR_IOBASEL_1:
5456 			case PCIR_MEMBASE_1:
5457 			case PCIR_PMBASEL_1:
5458 				/*
5459 				 * XXX: Should we bother creating a resource
5460 				 * list entry?
5461 				 */
5462 				return (bus_generic_alloc_resource(dev, child,
5463 				    type, rid, start, end, count, flags));
5464 			}
5465 		}
5466 #endif
5467 		/* Reserve resources for this BAR if needed. */
5468 		rle = resource_list_find(rl, type, *rid);
5469 		if (rle == NULL) {
5470 			res = pci_reserve_map(dev, child, type, rid, start, end,
5471 			    count, num, flags);
5472 			if (res == NULL)
5473 				return (NULL);
5474 		}
5475 	}
5476 	return (resource_list_alloc(rl, dev, child, type, rid,
5477 	    start, end, count, flags));
5478 }
5479 
5480 struct resource *
5481 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5482     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5483 {
5484 #ifdef PCI_IOV
5485 	struct pci_devinfo *dinfo;
5486 #endif
5487 
5488 	if (device_get_parent(child) != dev)
5489 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5490 		    type, rid, start, end, count, flags));
5491 
5492 #ifdef PCI_IOV
5493 	dinfo = device_get_ivars(child);
5494 	if (dinfo->cfg.flags & PCICFG_VF) {
5495 		switch (type) {
5496 		/* VFs can't have I/O BARs. */
5497 		case SYS_RES_IOPORT:
5498 			return (NULL);
5499 		case SYS_RES_MEMORY:
5500 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5501 			    start, end, count, flags));
5502 		}
5503 
5504 		/* Fall through for other types of resource allocations. */
5505 	}
5506 #endif
5507 
5508 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5509 	    count, 1, flags));
5510 }
5511 
5512 int
5513 pci_release_resource(device_t dev, device_t child, int type, int rid,
5514     struct resource *r)
5515 {
5516 	struct pci_devinfo *dinfo;
5517 	struct resource_list *rl;
5518 	pcicfgregs *cfg;
5519 
5520 	if (device_get_parent(child) != dev)
5521 		return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5522 		    type, rid, r));
5523 
5524 	dinfo = device_get_ivars(child);
5525 	cfg = &dinfo->cfg;
5526 
5527 #ifdef PCI_IOV
5528 	if (dinfo->cfg.flags & PCICFG_VF) {
5529 		switch (type) {
5530 		/* VFs can't have I/O BARs. */
5531 		case SYS_RES_IOPORT:
5532 			return (EDOOFUS);
5533 		case SYS_RES_MEMORY:
5534 			return (pci_vf_release_mem_resource(dev, child, rid,
5535 			    r));
5536 		}
5537 
5538 		/* Fall through for other types of resource allocations. */
5539 	}
5540 #endif
5541 
5542 #ifdef NEW_PCIB
5543 	/*
5544 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5545 	 * those allocations just pass the request up the tree.
5546 	 */
5547 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5548 	    (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5549 		switch (rid) {
5550 		case PCIR_IOBASEL_1:
5551 		case PCIR_MEMBASE_1:
5552 		case PCIR_PMBASEL_1:
5553 			return (bus_generic_release_resource(dev, child, type,
5554 			    rid, r));
5555 		}
5556 	}
5557 #endif
5558 
5559 	rl = &dinfo->resources;
5560 	return (resource_list_release(rl, dev, child, type, rid, r));
5561 }
5562 
5563 int
5564 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5565     struct resource *r)
5566 {
5567 	struct pci_devinfo *dinfo;
5568 	int error;
5569 
5570 	error = bus_generic_activate_resource(dev, child, type, rid, r);
5571 	if (error)
5572 		return (error);
5573 
5574 	/* Enable decoding in the command register when activating BARs. */
5575 	if (device_get_parent(child) == dev) {
5576 		/* Device ROMs need their decoding explicitly enabled. */
5577 		dinfo = device_get_ivars(child);
5578 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5579 			pci_write_bar(child, pci_find_bar(child, rid),
5580 			    rman_get_start(r) | PCIM_BIOS_ENABLE);
5581 		switch (type) {
5582 		case SYS_RES_IOPORT:
5583 		case SYS_RES_MEMORY:
5584 			error = PCI_ENABLE_IO(dev, child, type);
5585 			break;
5586 		}
5587 	}
5588 	return (error);
5589 }
5590 
5591 int
5592 pci_deactivate_resource(device_t dev, device_t child, int type,
5593     int rid, struct resource *r)
5594 {
5595 	struct pci_devinfo *dinfo;
5596 	int error;
5597 
5598 	error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5599 	if (error)
5600 		return (error);
5601 
5602 	/* Disable decoding for device ROMs. */
5603 	if (device_get_parent(child) == dev) {
5604 		dinfo = device_get_ivars(child);
5605 		if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5606 			pci_write_bar(child, pci_find_bar(child, rid),
5607 			    rman_get_start(r));
5608 	}
5609 	return (0);
5610 }
5611 
5612 void
5613 pci_child_deleted(device_t dev, device_t child)
5614 {
5615 	struct resource_list_entry *rle;
5616 	struct resource_list *rl;
5617 	struct pci_devinfo *dinfo;
5618 
5619 	dinfo = device_get_ivars(child);
5620 	rl = &dinfo->resources;
5621 
5622 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5623 
5624 	/* Turn off access to resources we're about to free */
5625 	if (bus_child_present(child) != 0) {
5626 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5627 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5628 
5629 		pci_disable_busmaster(child);
5630 	}
5631 
5632 	/* Free all allocated resources */
5633 	STAILQ_FOREACH(rle, rl, link) {
5634 		if (rle->res) {
5635 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5636 			    resource_list_busy(rl, rle->type, rle->rid)) {
5637 				pci_printf(&dinfo->cfg,
5638 				    "Resource still owned, oops. "
5639 				    "(type=%d, rid=%d, addr=%lx)\n",
5640 				    rle->type, rle->rid,
5641 				    rman_get_start(rle->res));
5642 				bus_release_resource(child, rle->type, rle->rid,
5643 				    rle->res);
5644 			}
5645 			resource_list_unreserve(rl, dev, child, rle->type,
5646 			    rle->rid);
5647 		}
5648 	}
5649 	resource_list_free(rl);
5650 
5651 	pci_freecfg(dinfo);
5652 }
5653 
5654 void
5655 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5656 {
5657 	struct pci_devinfo *dinfo;
5658 	struct resource_list *rl;
5659 	struct resource_list_entry *rle;
5660 
5661 	if (device_get_parent(child) != dev)
5662 		return;
5663 
5664 	dinfo = device_get_ivars(child);
5665 	rl = &dinfo->resources;
5666 	rle = resource_list_find(rl, type, rid);
5667 	if (rle == NULL)
5668 		return;
5669 
5670 	if (rle->res) {
5671 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5672 		    resource_list_busy(rl, type, rid)) {
5673 			device_printf(dev, "delete_resource: "
5674 			    "Resource still owned by child, oops. "
5675 			    "(type=%d, rid=%d, addr=%jx)\n",
5676 			    type, rid, rman_get_start(rle->res));
5677 			return;
5678 		}
5679 		resource_list_unreserve(rl, dev, child, type, rid);
5680 	}
5681 	resource_list_delete(rl, type, rid);
5682 }
5683 
5684 struct resource_list *
5685 pci_get_resource_list (device_t dev, device_t child)
5686 {
5687 	struct pci_devinfo *dinfo = device_get_ivars(child);
5688 
5689 	return (&dinfo->resources);
5690 }
5691 
5692 #ifdef ACPI_DMAR
5693 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5694 bus_dma_tag_t
5695 pci_get_dma_tag(device_t bus, device_t dev)
5696 {
5697 	bus_dma_tag_t tag;
5698 	struct pci_softc *sc;
5699 
5700 	if (device_get_parent(dev) == bus) {
5701 		/* try dmar and return if it works */
5702 		tag = dmar_get_dma_tag(bus, dev);
5703 	} else
5704 		tag = NULL;
5705 	if (tag == NULL) {
5706 		sc = device_get_softc(bus);
5707 		tag = sc->sc_dma_tag;
5708 	}
5709 	return (tag);
5710 }
5711 #else
5712 bus_dma_tag_t
5713 pci_get_dma_tag(device_t bus, device_t dev)
5714 {
5715 	struct pci_softc *sc = device_get_softc(bus);
5716 
5717 	return (sc->sc_dma_tag);
5718 }
5719 #endif
5720 
5721 uint32_t
5722 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5723 {
5724 	struct pci_devinfo *dinfo = device_get_ivars(child);
5725 	pcicfgregs *cfg = &dinfo->cfg;
5726 
5727 #ifdef PCI_IOV
5728 	/*
5729 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5730 	 * emulate them here.
5731 	 */
5732 	if (cfg->flags & PCICFG_VF) {
5733 		if (reg == PCIR_VENDOR) {
5734 			switch (width) {
5735 			case 4:
5736 				return (cfg->device << 16 | cfg->vendor);
5737 			case 2:
5738 				return (cfg->vendor);
5739 			case 1:
5740 				return (cfg->vendor & 0xff);
5741 			default:
5742 				return (0xffffffff);
5743 			}
5744 		} else if (reg == PCIR_DEVICE) {
5745 			switch (width) {
5746 			/* Note that an unaligned 4-byte read is an error. */
5747 			case 2:
5748 				return (cfg->device);
5749 			case 1:
5750 				return (cfg->device & 0xff);
5751 			default:
5752 				return (0xffffffff);
5753 			}
5754 		}
5755 	}
5756 #endif
5757 
5758 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5759 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5760 }
5761 
5762 void
5763 pci_write_config_method(device_t dev, device_t child, int reg,
5764     uint32_t val, int width)
5765 {
5766 	struct pci_devinfo *dinfo = device_get_ivars(child);
5767 	pcicfgregs *cfg = &dinfo->cfg;
5768 
5769 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5770 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
5771 }
5772 
5773 int
5774 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5775     size_t buflen)
5776 {
5777 
5778 	snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5779 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5780 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5781 	return (0);
5782 }
5783 
5784 int
5785 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5786     size_t buflen)
5787 {
5788 	struct pci_devinfo *dinfo;
5789 	pcicfgregs *cfg;
5790 
5791 	dinfo = device_get_ivars(child);
5792 	cfg = &dinfo->cfg;
5793 	snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5794 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5795 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5796 	    cfg->progif);
5797 	return (0);
5798 }
5799 
5800 int
5801 pci_assign_interrupt_method(device_t dev, device_t child)
5802 {
5803 	struct pci_devinfo *dinfo = device_get_ivars(child);
5804 	pcicfgregs *cfg = &dinfo->cfg;
5805 
5806 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5807 	    cfg->intpin));
5808 }
5809 
5810 static void
5811 pci_lookup(void *arg, const char *name, device_t *dev)
5812 {
5813 	long val;
5814 	char *end;
5815 	int domain, bus, slot, func;
5816 
5817 	if (*dev != NULL)
5818 		return;
5819 
5820 	/*
5821 	 * Accept pciconf-style selectors of either pciD:B:S:F or
5822 	 * pciB:S:F.  In the latter case, the domain is assumed to
5823 	 * be zero.
5824 	 */
5825 	if (strncmp(name, "pci", 3) != 0)
5826 		return;
5827 	val = strtol(name + 3, &end, 10);
5828 	if (val < 0 || val > INT_MAX || *end != ':')
5829 		return;
5830 	domain = val;
5831 	val = strtol(end + 1, &end, 10);
5832 	if (val < 0 || val > INT_MAX || *end != ':')
5833 		return;
5834 	bus = val;
5835 	val = strtol(end + 1, &end, 10);
5836 	if (val < 0 || val > INT_MAX)
5837 		return;
5838 	slot = val;
5839 	if (*end == ':') {
5840 		val = strtol(end + 1, &end, 10);
5841 		if (val < 0 || val > INT_MAX || *end != '\0')
5842 			return;
5843 		func = val;
5844 	} else if (*end == '\0') {
5845 		func = slot;
5846 		slot = bus;
5847 		bus = domain;
5848 		domain = 0;
5849 	} else
5850 		return;
5851 
5852 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5853 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5854 		return;
5855 
5856 	*dev = pci_find_dbsf(domain, bus, slot, func);
5857 }
5858 
5859 static int
5860 pci_modevent(module_t mod, int what, void *arg)
5861 {
5862 	static struct cdev *pci_cdev;
5863 	static eventhandler_tag tag;
5864 
5865 	switch (what) {
5866 	case MOD_LOAD:
5867 		STAILQ_INIT(&pci_devq);
5868 		pci_generation = 0;
5869 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5870 		    "pci");
5871 		pci_load_vendor_data();
5872 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5873 		    1000);
5874 		break;
5875 
5876 	case MOD_UNLOAD:
5877 		if (tag != NULL)
5878 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5879 		destroy_dev(pci_cdev);
5880 		break;
5881 	}
5882 
5883 	return (0);
5884 }
5885 
5886 static void
5887 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5888 {
5889 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
5890 	struct pcicfg_pcie *cfg;
5891 	int version, pos;
5892 
5893 	cfg = &dinfo->cfg.pcie;
5894 	pos = cfg->pcie_location;
5895 
5896 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5897 
5898 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5899 
5900 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5901 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5902 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5903 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5904 
5905 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5906 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5907 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5908 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5909 
5910 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5911 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5912 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5913 
5914 	if (version > 1) {
5915 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5916 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5917 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5918 	}
5919 #undef WREG
5920 }
5921 
5922 static void
5923 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5924 {
5925 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5926 	    dinfo->cfg.pcix.pcix_command,  2);
5927 }
5928 
5929 void
5930 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5931 {
5932 
5933 	/*
5934 	 * Restore the device to full power mode.  We must do this
5935 	 * before we restore the registers because moving from D3 to
5936 	 * D0 will cause the chip's BARs and some other registers to
5937 	 * be reset to some unknown power on reset values.  Cut down
5938 	 * the noise on boot by doing nothing if we are already in
5939 	 * state D0.
5940 	 */
5941 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5942 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5943 	pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5944 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5945 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5946 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5947 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5948 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5949 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5950 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5951 	case PCIM_HDRTYPE_NORMAL:
5952 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5953 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5954 		break;
5955 	case PCIM_HDRTYPE_BRIDGE:
5956 		pci_write_config(dev, PCIR_SECLAT_1,
5957 		    dinfo->cfg.bridge.br_seclat, 1);
5958 		pci_write_config(dev, PCIR_SUBBUS_1,
5959 		    dinfo->cfg.bridge.br_subbus, 1);
5960 		pci_write_config(dev, PCIR_SECBUS_1,
5961 		    dinfo->cfg.bridge.br_secbus, 1);
5962 		pci_write_config(dev, PCIR_PRIBUS_1,
5963 		    dinfo->cfg.bridge.br_pribus, 1);
5964 		pci_write_config(dev, PCIR_BRIDGECTL_1,
5965 		    dinfo->cfg.bridge.br_control, 2);
5966 		break;
5967 	case PCIM_HDRTYPE_CARDBUS:
5968 		pci_write_config(dev, PCIR_SECLAT_2,
5969 		    dinfo->cfg.bridge.br_seclat, 1);
5970 		pci_write_config(dev, PCIR_SUBBUS_2,
5971 		    dinfo->cfg.bridge.br_subbus, 1);
5972 		pci_write_config(dev, PCIR_SECBUS_2,
5973 		    dinfo->cfg.bridge.br_secbus, 1);
5974 		pci_write_config(dev, PCIR_PRIBUS_2,
5975 		    dinfo->cfg.bridge.br_pribus, 1);
5976 		pci_write_config(dev, PCIR_BRIDGECTL_2,
5977 		    dinfo->cfg.bridge.br_control, 2);
5978 		break;
5979 	}
5980 	pci_restore_bars(dev);
5981 
5982 	/*
5983 	 * Restore extended capabilities for PCI-Express and PCI-X
5984 	 */
5985 	if (dinfo->cfg.pcie.pcie_location != 0)
5986 		pci_cfg_restore_pcie(dev, dinfo);
5987 	if (dinfo->cfg.pcix.pcix_location != 0)
5988 		pci_cfg_restore_pcix(dev, dinfo);
5989 
5990 	/* Restore MSI and MSI-X configurations if they are present. */
5991 	if (dinfo->cfg.msi.msi_location != 0)
5992 		pci_resume_msi(dev);
5993 	if (dinfo->cfg.msix.msix_location != 0)
5994 		pci_resume_msix(dev);
5995 
5996 #ifdef PCI_IOV
5997 	if (dinfo->cfg.iov != NULL)
5998 		pci_iov_cfg_restore(dev, dinfo);
5999 #endif
6000 }
6001 
6002 static void
6003 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6004 {
6005 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6006 	struct pcicfg_pcie *cfg;
6007 	int version, pos;
6008 
6009 	cfg = &dinfo->cfg.pcie;
6010 	pos = cfg->pcie_location;
6011 
6012 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6013 
6014 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6015 
6016 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6017 
6018 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6019 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6020 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6021 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6022 
6023 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6024 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6025 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6026 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6027 
6028 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6029 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6030 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6031 
6032 	if (version > 1) {
6033 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6034 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6035 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6036 	}
6037 #undef RREG
6038 }
6039 
6040 static void
6041 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6042 {
6043 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6044 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6045 }
6046 
6047 void
6048 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6049 {
6050 	uint32_t cls;
6051 	int ps;
6052 
6053 	/*
6054 	 * Some drivers apparently write to these registers w/o updating our
6055 	 * cached copy.  No harm happens if we update the copy, so do so here
6056 	 * so we can restore them.  The COMMAND register is modified by the
6057 	 * bus w/o updating the cache.  This should represent the normally
6058 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6059 	 */
6060 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6061 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6062 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6063 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6064 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6065 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6066 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6067 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6068 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6069 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6070 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6071 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6072 	case PCIM_HDRTYPE_NORMAL:
6073 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6074 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6075 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6076 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6077 		break;
6078 	case PCIM_HDRTYPE_BRIDGE:
6079 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6080 		    PCIR_SECLAT_1, 1);
6081 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6082 		    PCIR_SUBBUS_1, 1);
6083 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6084 		    PCIR_SECBUS_1, 1);
6085 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6086 		    PCIR_PRIBUS_1, 1);
6087 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6088 		    PCIR_BRIDGECTL_1, 2);
6089 		break;
6090 	case PCIM_HDRTYPE_CARDBUS:
6091 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6092 		    PCIR_SECLAT_2, 1);
6093 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6094 		    PCIR_SUBBUS_2, 1);
6095 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6096 		    PCIR_SECBUS_2, 1);
6097 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6098 		    PCIR_PRIBUS_2, 1);
6099 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6100 		    PCIR_BRIDGECTL_2, 2);
6101 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6102 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6103 		break;
6104 	}
6105 
6106 	if (dinfo->cfg.pcie.pcie_location != 0)
6107 		pci_cfg_save_pcie(dev, dinfo);
6108 
6109 	if (dinfo->cfg.pcix.pcix_location != 0)
6110 		pci_cfg_save_pcix(dev, dinfo);
6111 
6112 #ifdef PCI_IOV
6113 	if (dinfo->cfg.iov != NULL)
6114 		pci_iov_cfg_save(dev, dinfo);
6115 #endif
6116 
6117 	/*
6118 	 * don't set the state for display devices, base peripherals and
6119 	 * memory devices since bad things happen when they are powered down.
6120 	 * We should (a) have drivers that can easily detach and (b) use
6121 	 * generic drivers for these devices so that some device actually
6122 	 * attaches.  We need to make sure that when we implement (a) we don't
6123 	 * power the device down on a reattach.
6124 	 */
6125 	cls = pci_get_class(dev);
6126 	if (!setstate)
6127 		return;
6128 	switch (pci_do_power_nodriver)
6129 	{
6130 		case 0:		/* NO powerdown at all */
6131 			return;
6132 		case 1:		/* Conservative about what to power down */
6133 			if (cls == PCIC_STORAGE)
6134 				return;
6135 			/*FALLTHROUGH*/
6136 		case 2:		/* Aggressive about what to power down */
6137 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6138 			    cls == PCIC_BASEPERIPH)
6139 				return;
6140 			/*FALLTHROUGH*/
6141 		case 3:		/* Power down everything */
6142 			break;
6143 	}
6144 	/*
6145 	 * PCI spec says we can only go into D3 state from D0 state.
6146 	 * Transition from D[12] into D0 before going to D3 state.
6147 	 */
6148 	ps = pci_get_powerstate(dev);
6149 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6150 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6151 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6152 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6153 }
6154 
6155 /* Wrapper APIs suitable for device driver use. */
6156 void
6157 pci_save_state(device_t dev)
6158 {
6159 	struct pci_devinfo *dinfo;
6160 
6161 	dinfo = device_get_ivars(dev);
6162 	pci_cfg_save(dev, dinfo, 0);
6163 }
6164 
6165 void
6166 pci_restore_state(device_t dev)
6167 {
6168 	struct pci_devinfo *dinfo;
6169 
6170 	dinfo = device_get_ivars(dev);
6171 	pci_cfg_restore(dev, dinfo);
6172 }
6173 
6174 static int
6175 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6176     uintptr_t *id)
6177 {
6178 
6179 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6180 }
6181 
6182 /* Find the upstream port of a given PCI device in a root complex. */
6183 device_t
6184 pci_find_pcie_root_port(device_t dev)
6185 {
6186 	struct pci_devinfo *dinfo;
6187 	devclass_t pci_class;
6188 	device_t pcib, bus;
6189 
6190 	pci_class = devclass_find("pci");
6191 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6192 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6193 
6194 	/*
6195 	 * Walk the bridge hierarchy until we find a PCI-e root
6196 	 * port or a non-PCI device.
6197 	 */
6198 	for (;;) {
6199 		bus = device_get_parent(dev);
6200 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6201 		    device_get_nameunit(dev)));
6202 
6203 		pcib = device_get_parent(bus);
6204 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6205 		    device_get_nameunit(bus)));
6206 
6207 		/*
6208 		 * pcib's parent must be a PCI bus for this to be a
6209 		 * PCI-PCI bridge.
6210 		 */
6211 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6212 			return (NULL);
6213 
6214 		dinfo = device_get_ivars(pcib);
6215 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6216 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6217 			return (pcib);
6218 
6219 		dev = pcib;
6220 	}
6221 }
6222 
6223 /*
6224  * Wait for pending transactions to complete on a PCI-express function.
6225  *
6226  * The maximum delay is specified in milliseconds in max_delay.  Note
6227  * that this function may sleep.
6228  *
6229  * Returns true if the function is idle and false if the timeout is
6230  * exceeded.  If dev is not a PCI-express function, this returns true.
6231  */
6232 bool
6233 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6234 {
6235 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6236 	uint16_t sta;
6237 	int cap;
6238 
6239 	cap = dinfo->cfg.pcie.pcie_location;
6240 	if (cap == 0)
6241 		return (true);
6242 
6243 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6244 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6245 		if (max_delay == 0)
6246 			return (false);
6247 
6248 		/* Poll once every 100 milliseconds up to the timeout. */
6249 		if (max_delay > 100) {
6250 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6251 			max_delay -= 100;
6252 		} else {
6253 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6254 			    C_HARDCLOCK);
6255 			max_delay = 0;
6256 		}
6257 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6258 	}
6259 
6260 	return (true);
6261 }
6262 
6263 /*
6264  * Determine the maximum Completion Timeout in microseconds.
6265  *
6266  * For non-PCI-express functions this returns 0.
6267  */
6268 int
6269 pcie_get_max_completion_timeout(device_t dev)
6270 {
6271 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6272 	int cap;
6273 
6274 	cap = dinfo->cfg.pcie.pcie_location;
6275 	if (cap == 0)
6276 		return (0);
6277 
6278 	/*
6279 	 * Functions using the 1.x spec use the default timeout range of
6280 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6281 	 * support programmable timeouts also use this range.
6282 	 */
6283 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6284 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6285 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6286 		return (50 * 1000);
6287 
6288 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6289 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6290 	case PCIEM_CTL2_COMP_TIMO_100US:
6291 		return (100);
6292 	case PCIEM_CTL2_COMP_TIMO_10MS:
6293 		return (10 * 1000);
6294 	case PCIEM_CTL2_COMP_TIMO_55MS:
6295 		return (55 * 1000);
6296 	case PCIEM_CTL2_COMP_TIMO_210MS:
6297 		return (210 * 1000);
6298 	case PCIEM_CTL2_COMP_TIMO_900MS:
6299 		return (900 * 1000);
6300 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6301 		return (3500 * 1000);
6302 	case PCIEM_CTL2_COMP_TIMO_13S:
6303 		return (13 * 1000 * 1000);
6304 	case PCIEM_CTL2_COMP_TIMO_64S:
6305 		return (64 * 1000 * 1000);
6306 	default:
6307 		return (50 * 1000);
6308 	}
6309 }
6310 
6311 /*
6312  * Perform a Function Level Reset (FLR) on a device.
6313  *
6314  * This function first waits for any pending transactions to complete
6315  * within the timeout specified by max_delay.  If transactions are
6316  * still pending, the function will return false without attempting a
6317  * reset.
6318  *
6319  * If dev is not a PCI-express function or does not support FLR, this
6320  * function returns false.
6321  *
6322  * Note that no registers are saved or restored.  The caller is
6323  * responsible for saving and restoring any registers including
6324  * PCI-standard registers via pci_save_state() and
6325  * pci_restore_state().
6326  */
6327 bool
6328 pcie_flr(device_t dev, u_int max_delay, bool force)
6329 {
6330 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6331 	uint16_t cmd, ctl;
6332 	int compl_delay;
6333 	int cap;
6334 
6335 	cap = dinfo->cfg.pcie.pcie_location;
6336 	if (cap == 0)
6337 		return (false);
6338 
6339 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6340 		return (false);
6341 
6342 	/*
6343 	 * Disable busmastering to prevent generation of new
6344 	 * transactions while waiting for the device to go idle.  If
6345 	 * the idle timeout fails, the command register is restored
6346 	 * which will re-enable busmastering.
6347 	 */
6348 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6349 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6350 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6351 		if (!force) {
6352 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6353 			return (false);
6354 		}
6355 		pci_printf(&dinfo->cfg,
6356 		    "Resetting with transactions pending after %d ms\n",
6357 		    max_delay);
6358 
6359 		/*
6360 		 * Extend the post-FLR delay to cover the maximum
6361 		 * Completion Timeout delay of anything in flight
6362 		 * during the FLR delay.  Enforce a minimum delay of
6363 		 * at least 10ms.
6364 		 */
6365 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6366 		if (compl_delay < 10)
6367 			compl_delay = 10;
6368 	} else
6369 		compl_delay = 0;
6370 
6371 	/* Initiate the reset. */
6372 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6373 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6374 	    PCIEM_CTL_INITIATE_FLR, 2);
6375 
6376 	/* Wait for 100ms. */
6377 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6378 
6379 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6380 	    PCIEM_STA_TRANSACTION_PND)
6381 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6382 	return (true);
6383 }
6384 
6385 /*
6386  * Attempt a power-management reset by cycling the device in/out of D3
6387  * state.  PCI spec says we can only go into D3 state from D0 state.
6388  * Transition from D[12] into D0 before going to D3 state.
6389  */
6390 int
6391 pci_power_reset(device_t dev)
6392 {
6393 	int ps;
6394 
6395 	ps = pci_get_powerstate(dev);
6396 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6397 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6398 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6399 	pci_set_powerstate(dev, ps);
6400 	return (0);
6401 }
6402 
6403 /*
6404  * Try link drop and retrain of the downstream port of upstream
6405  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6406  * cause Conventional Hot reset of the device in the slot.
6407  * Alternative, for PCIe, could be the secondary bus reset initiatied
6408  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6409  */
6410 int
6411 pcie_link_reset(device_t port, int pcie_location)
6412 {
6413 	uint16_t v;
6414 
6415 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6416 	v |= PCIEM_LINK_CTL_LINK_DIS;
6417 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6418 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6419 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6420 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6421 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6422 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6423 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6424 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6425 }
6426 
6427 static int
6428 pci_reset_post(device_t dev, device_t child)
6429 {
6430 
6431 	if (dev == device_get_parent(child))
6432 		pci_restore_state(child);
6433 	return (0);
6434 }
6435 
6436 static int
6437 pci_reset_prepare(device_t dev, device_t child)
6438 {
6439 
6440 	if (dev == device_get_parent(child))
6441 		pci_save_state(child);
6442 	return (0);
6443 }
6444 
6445 static int
6446 pci_reset_child(device_t dev, device_t child, int flags)
6447 {
6448 	int error;
6449 
6450 	if (dev == NULL || device_get_parent(child) != dev)
6451 		return (0);
6452 	if ((flags & DEVF_RESET_DETACH) != 0) {
6453 		error = device_get_state(child) == DS_ATTACHED ?
6454 		    device_detach(child) : 0;
6455 	} else {
6456 		error = BUS_SUSPEND_CHILD(dev, child);
6457 	}
6458 	if (error == 0) {
6459 		if (!pcie_flr(child, 1000, false)) {
6460 			error = BUS_RESET_PREPARE(dev, child);
6461 			if (error == 0)
6462 				pci_power_reset(child);
6463 			BUS_RESET_POST(dev, child);
6464 		}
6465 		if ((flags & DEVF_RESET_DETACH) != 0)
6466 			device_probe_and_attach(child);
6467 		else
6468 			BUS_RESUME_CHILD(dev, child);
6469 	}
6470 	return (error);
6471 }
6472 
6473 const struct pci_device_table *
6474 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6475 {
6476 	bool match;
6477 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6478 
6479 	vendor = pci_get_vendor(child);
6480 	device = pci_get_device(child);
6481 	subvendor = pci_get_subvendor(child);
6482 	subdevice = pci_get_subdevice(child);
6483 	class = pci_get_class(child);
6484 	subclass = pci_get_subclass(child);
6485 	revid = pci_get_revid(child);
6486 	while (nelt-- > 0) {
6487 		match = true;
6488 		if (id->match_flag_vendor)
6489 			match &= vendor == id->vendor;
6490 		if (id->match_flag_device)
6491 			match &= device == id->device;
6492 		if (id->match_flag_subvendor)
6493 			match &= subvendor == id->subvendor;
6494 		if (id->match_flag_subdevice)
6495 			match &= subdevice == id->subdevice;
6496 		if (id->match_flag_class)
6497 			match &= class == id->class_id;
6498 		if (id->match_flag_subclass)
6499 			match &= subclass == id->subclass;
6500 		if (id->match_flag_revid)
6501 			match &= revid == id->revid;
6502 		if (match)
6503 			return (id);
6504 		id++;
6505 	}
6506 	return (NULL);
6507 }
6508 
6509 static void
6510 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6511 {
6512 	const char *dev_name;
6513 	device_t dev;
6514 
6515 	dev = dinfo->cfg.dev;
6516 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6517 	    dinfo->cfg.slot, dinfo->cfg.func);
6518 	dev_name = device_get_name(dev);
6519 	if (dev_name != NULL)
6520 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6521 }
6522 
6523 void
6524 pci_print_faulted_dev(void)
6525 {
6526 	struct pci_devinfo *dinfo;
6527 	device_t dev;
6528 	int aer, i;
6529 	uint32_t r1, r2;
6530 	uint16_t status;
6531 
6532 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6533 		dev = dinfo->cfg.dev;
6534 		status = pci_read_config(dev, PCIR_STATUS, 2);
6535 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6536 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6537 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6538 		if (status != 0) {
6539 			pci_print_faulted_dev_name(dinfo);
6540 			printf(" error 0x%04x\n", status);
6541 		}
6542 		if (dinfo->cfg.pcie.pcie_location != 0) {
6543 			status = pci_read_config(dev,
6544 			    dinfo->cfg.pcie.pcie_location +
6545 			    PCIER_DEVICE_STA, 2);
6546 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6547 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6548 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6549 				pci_print_faulted_dev_name(dinfo);
6550 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6551 				    pci_read_config(dev,
6552 				    dinfo->cfg.pcie.pcie_location +
6553 				    PCIER_DEVICE_CTL, 2),
6554 				    status);
6555 			}
6556 		}
6557 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6558 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6559 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6560 			if (r1 != 0 || r2 != 0) {
6561 				pci_print_faulted_dev_name(dinfo);
6562 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6563 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6564 				    r1, pci_read_config(dev, aer +
6565 				    PCIR_AER_UC_MASK, 4),
6566 				    pci_read_config(dev, aer +
6567 				    PCIR_AER_UC_SEVERITY, 4),
6568 				    r2, pci_read_config(dev, aer +
6569 				    PCIR_AER_COR_MASK, 4),
6570 				    pci_read_config(dev, aer +
6571 				    PCIR_AER_CAP_CONTROL, 4));
6572 				for (i = 0; i < 4; i++) {
6573 					r1 = pci_read_config(dev, aer +
6574 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6575 					printf("    HL%d: 0x%08x\n", i, r1);
6576 				}
6577 			}
6578 		}
6579 	}
6580 }
6581 
6582 #ifdef DDB
6583 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6584 {
6585 
6586 	pci_print_faulted_dev();
6587 }
6588 
6589 static void
6590 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6591 {
6592 	device_t dev;
6593 	int aer;
6594 	uint32_t r;
6595 
6596 	dev = dinfo->cfg.dev;
6597 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6598 	    PCIER_DEVICE_STA, 2);
6599 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6600 	    PCIER_DEVICE_STA, r, 2);
6601 
6602 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6603 		return;
6604 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6605 	if (r != 0)
6606 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6607 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6608 	if (r != 0)
6609 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6610 }
6611 
6612 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6613 {
6614 	struct pci_devinfo *dinfo;
6615 	device_t dev;
6616 	uint16_t status, status1;
6617 
6618 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6619 		dev = dinfo->cfg.dev;
6620 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6621 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6622 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6623 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6624 		if (status1 != 0) {
6625 			status &= ~status1;
6626 			pci_write_config(dev, PCIR_STATUS, status, 2);
6627 		}
6628 		if (dinfo->cfg.pcie.pcie_location != 0)
6629 			db_clear_pcie_errors(dinfo);
6630 	}
6631 }
6632 #endif
6633