xref: /freebsd/sys/dev/pci/pci.c (revision d0ff5773cefaf3fa41b1be3e44ca35bd9d5f68ee)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_acpi.h"
33 #include "opt_iommu.h"
34 #include "opt_bus.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/linker.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/stdarg.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_extern.h>
57 
58 #include <sys/bus.h>
59 #include <machine/bus.h>
60 #include <sys/rman.h>
61 #include <machine/resource.h>
62 
63 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
64 #include <machine/intr_machdep.h>
65 #endif
66 
67 #include <sys/pciio.h>
68 #include <dev/pci/pcireg.h>
69 #include <dev/pci/pcivar.h>
70 #include <dev/pci/pci_private.h>
71 
72 #ifdef PCI_IOV
73 #include <sys/nv.h>
74 #include <dev/pci/pci_iov_private.h>
75 #endif
76 
77 #include <dev/usb/controller/xhcireg.h>
78 #include <dev/usb/controller/ehcireg.h>
79 #include <dev/usb/controller/ohcireg.h>
80 #include <dev/usb/controller/uhcireg.h>
81 
82 #include <dev/iommu/iommu.h>
83 
84 #include <contrib/dev/acpica/include/acpi.h>
85 #include <dev/acpica/acpivar.h>
86 
87 #include "pcib_if.h"
88 #include "pci_if.h"
89 
90 #define	PCIR_IS_BIOS(cfg, reg)						\
91 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
92 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
93 
94 static device_probe_t	pci_probe;
95 
96 static bus_reset_post_t pci_reset_post;
97 static bus_reset_prepare_t pci_reset_prepare;
98 static bus_reset_child_t pci_reset_child;
99 static bus_hint_device_unit_t pci_hint_device_unit;
100 static bus_remap_intr_t pci_remap_intr_method;
101 
102 static pci_get_id_t	pci_get_id_method;
103 
104 static int		pci_has_quirk(uint32_t devid, int quirk);
105 static pci_addr_t	pci_mapbase(uint64_t mapreg);
106 static const char	*pci_maptype(uint64_t mapreg);
107 static int		pci_maprange(uint64_t mapreg);
108 static pci_addr_t	pci_rombase(uint64_t mapreg);
109 static int		pci_romsize(uint64_t testval);
110 static void		pci_fixancient(pcicfgregs *cfg);
111 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
112 
113 static int		pci_porten(device_t dev);
114 static int		pci_memen(device_t dev);
115 static void		pci_assign_interrupt(device_t bus, device_t dev,
116 			    int force_route);
117 static int		pci_add_map(device_t bus, device_t dev, int reg,
118 			    struct resource_list *rl, int force, int prefetch);
119 static void		pci_load_vendor_data(void);
120 static int		pci_describe_parse_line(char **ptr, int *vendor,
121 			    int *device, char **desc);
122 static char		*pci_describe_device(device_t dev);
123 static int		pci_modevent(module_t mod, int what, void *arg);
124 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
125 			    pcicfgregs *cfg);
126 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
127 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
128 			    int reg, uint32_t *data);
129 #if 0
130 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
131 			    int reg, uint32_t data);
132 #endif
133 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
134 static void		pci_mask_msix(device_t dev, u_int index);
135 static void		pci_unmask_msix(device_t dev, u_int index);
136 static int		pci_msi_blacklisted(void);
137 static int		pci_msix_blacklisted(void);
138 static void		pci_resume_msi(device_t dev);
139 static void		pci_resume_msix(device_t dev);
140 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
141     int b, int s, int f, uint16_t vid, uint16_t did);
142 
143 static device_method_t pci_methods[] = {
144 	/* Device interface */
145 	DEVMETHOD(device_probe,		pci_probe),
146 	DEVMETHOD(device_attach,	pci_attach),
147 	DEVMETHOD(device_detach,	pci_detach),
148 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
149 	DEVMETHOD(device_suspend,	bus_generic_suspend),
150 	DEVMETHOD(device_resume,	pci_resume),
151 
152 	/* Bus interface */
153 	DEVMETHOD(bus_print_child,	pci_print_child),
154 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
155 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
156 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
157 	DEVMETHOD(bus_driver_added,	pci_driver_added),
158 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
159 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
160 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
161 	DEVMETHOD(bus_reset_post,	pci_reset_post),
162 	DEVMETHOD(bus_reset_child,	pci_reset_child),
163 
164 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
165 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
166 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
167 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
168 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
169 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
170 	DEVMETHOD(bus_adjust_resource,	pci_adjust_resource),
171 	DEVMETHOD(bus_release_resource,	pci_release_resource),
172 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
173 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
174 	DEVMETHOD(bus_map_resource,	pci_map_resource),
175 	DEVMETHOD(bus_unmap_resource,	pci_unmap_resource),
176 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
177 	DEVMETHOD(bus_child_detached,	pci_child_detached),
178 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
179 	DEVMETHOD(bus_child_location,	pci_child_location_method),
180 	DEVMETHOD(bus_get_device_path,	pci_get_device_path_method),
181 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
182 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
183 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
184 	DEVMETHOD(bus_resume_child,	pci_resume_child),
185 	DEVMETHOD(bus_rescan,		pci_rescan_method),
186 
187 	/* PCI interface */
188 	DEVMETHOD(pci_read_config,	pci_read_config_method),
189 	DEVMETHOD(pci_write_config,	pci_write_config_method),
190 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
191 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
192 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
193 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
194 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
195 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
196 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
197 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
198 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
199 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
200 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
201 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
202 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
203 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
204 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
205 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
206 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
207 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
208 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
209 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
210 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
211 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
212 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
213 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
214 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
215 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
216 	DEVMETHOD(pci_get_id,		pci_get_id_method),
217 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
218 	DEVMETHOD(pci_child_added,	pci_child_added_method),
219 #ifdef PCI_IOV
220 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
221 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
222 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
223 #endif
224 
225 	DEVMETHOD_END
226 };
227 
228 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
229 
230 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_modevent, NULL, BUS_PASS_BUS);
231 MODULE_VERSION(pci, 1);
232 
233 static char	*pci_vendordata;
234 static size_t	pci_vendordata_size;
235 
236 struct pci_quirk {
237 	uint32_t devid;	/* Vendor/device of the card */
238 	int	type;
239 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
240 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
241 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
242 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
243 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
244 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
245 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
246 	int	arg1;
247 	int	arg2;
248 };
249 
250 static const struct pci_quirk pci_quirks[] = {
251 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
252 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
254 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
255 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
256 
257 	/*
258 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
259 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
260 	 */
261 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
262 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
263 
264 	/*
265 	 * MSI doesn't work on earlier Intel chipsets including
266 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
267 	 */
268 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
273 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
274 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
275 
276 	/*
277 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
278 	 * bridge.
279 	 */
280 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
281 
282 	/*
283 	 * Some virtualization environments emulate an older chipset
284 	 * but support MSI just fine.  QEMU uses the Intel 82440.
285 	 */
286 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
287 
288 	/*
289 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
290 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
291 	 * It prevents us from attaching hpet(4) when the bit is unset.
292 	 * Note this quirk only affects SB600 revision A13 and earlier.
293 	 * For SB600 A21 and later, firmware must set the bit to hide it.
294 	 * For SB700 and later, it is unused and hardcoded to zero.
295 	 */
296 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
297 
298 	/*
299 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
300 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
301 	 * of the command register is set.
302 	 */
303 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
306 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
307 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
308 
309 	/*
310 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
311 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
312 	 */
313 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
314 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
315 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
316 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
317 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
318 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
319 
320 	/*
321 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
322 	 * expected place.
323 	 */
324 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
325 	{ 0 }
326 };
327 
328 /* map register information */
329 #define	PCI_MAPMEM	0x01	/* memory map */
330 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
331 #define	PCI_MAPPORT	0x04	/* port map */
332 
333 struct devlist pci_devq;
334 uint32_t pci_generation;
335 uint32_t pci_numdevs = 0;
336 static int pcie_chipset, pcix_chipset;
337 
338 /* sysctl vars */
339 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
340     "PCI bus tuning parameters");
341 
342 static int pci_enable_io_modes = 1;
343 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
344     &pci_enable_io_modes, 1,
345     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
346     " enable these bits correctly.  We'd like to do this all the time, but"
347     " there are some peripherals that this causes problems with.");
348 
349 static int pci_do_realloc_bars = 1;
350 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
351     &pci_do_realloc_bars, 0,
352     "Attempt to allocate a new range for any BARs whose original "
353     "firmware-assigned ranges fail to allocate during the initial device scan.");
354 
355 static int pci_do_power_nodriver = 0;
356 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
357     &pci_do_power_nodriver, 0,
358     "Place a function into D3 state when no driver attaches to it.  0 means"
359     " disable.  1 means conservatively place function into D3 state.  2 means"
360     " aggressively place function into D3 state.  3 means put absolutely"
361     " everything in D3 state.");
362 
363 int pci_do_power_resume = 1;
364 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
365     &pci_do_power_resume, 1,
366   "Transition from D3 -> D0 on resume.");
367 
368 int pci_do_power_suspend = 1;
369 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
370     &pci_do_power_suspend, 1,
371   "Transition from D0 -> D3 on suspend.");
372 
373 static int pci_do_msi = 1;
374 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
375     "Enable support for MSI interrupts");
376 
377 static int pci_do_msix = 1;
378 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
379     "Enable support for MSI-X interrupts");
380 
381 static int pci_msix_rewrite_table = 0;
382 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
383     &pci_msix_rewrite_table, 0,
384     "Rewrite entire MSI-X table when updating MSI-X entries");
385 
386 static int pci_honor_msi_blacklist = 1;
387 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
388     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
389 
390 #if defined(__i386__) || defined(__amd64__)
391 static int pci_usb_takeover = 1;
392 #else
393 static int pci_usb_takeover = 0;
394 #endif
395 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
396     &pci_usb_takeover, 1,
397     "Enable early takeover of USB controllers. Disable this if you depend on"
398     " BIOS emulation of USB devices, that is you use USB devices (like"
399     " keyboard or mouse) but do not load USB drivers");
400 
401 static int pci_clear_bars;
402 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
403     "Ignore firmware-assigned resources for BARs.");
404 
405 static int pci_clear_buses;
406 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
407     "Ignore firmware-assigned bus numbers.");
408 
409 static int pci_enable_ari = 1;
410 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
411     0, "Enable support for PCIe Alternative RID Interpretation");
412 
413 /*
414  * Some x86 firmware only enables PCIe hotplug if we claim to support aspm,
415  * however enabling it breaks some arm64 firmware as it powers off devices.
416  */
417 #if defined(__i386__) || defined(__amd64__)
418 int pci_enable_aspm = 1;
419 #else
420 int pci_enable_aspm = 0;
421 #endif
422 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
423     0, "Enable support for PCIe Active State Power Management");
424 
425 static int pci_clear_aer_on_attach = 0;
426 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
427     &pci_clear_aer_on_attach, 0,
428     "Clear port and device AER state on driver attach");
429 
430 static bool pci_enable_mps_tune = true;
431 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
432     &pci_enable_mps_tune, 1,
433     "Enable tuning of MPS(maximum payload size)." );
434 
435 static bool pci_intx_reroute = true;
436 SYSCTL_BOOL(_hw_pci, OID_AUTO, intx_reroute, CTLFLAG_RWTUN,
437     &pci_intx_reroute, 0, "Re-route INTx interrupts when scanning devices");
438 
439 static int
440 pci_has_quirk(uint32_t devid, int quirk)
441 {
442 	const struct pci_quirk *q;
443 
444 	for (q = &pci_quirks[0]; q->devid; q++) {
445 		if (q->devid == devid && q->type == quirk)
446 			return (1);
447 	}
448 	return (0);
449 }
450 
451 /* Find a device_t by bus/slot/function in domain 0 */
452 
453 device_t
454 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
455 {
456 
457 	return (pci_find_dbsf(0, bus, slot, func));
458 }
459 
460 /* Find a device_t by domain/bus/slot/function */
461 
462 device_t
463 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
464 {
465 	struct pci_devinfo *dinfo = NULL;
466 
467 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
468 		if ((dinfo->cfg.domain == domain) &&
469 		    (dinfo->cfg.bus == bus) &&
470 		    (dinfo->cfg.slot == slot) &&
471 		    (dinfo->cfg.func == func)) {
472 			break;
473 		}
474 	}
475 
476 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
477 }
478 
479 /* Find a device_t by vendor/device ID */
480 
481 device_t
482 pci_find_device(uint16_t vendor, uint16_t device)
483 {
484 	struct pci_devinfo *dinfo;
485 
486 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
487 		if ((dinfo->cfg.vendor == vendor) &&
488 		    (dinfo->cfg.device == device)) {
489 			return (dinfo->cfg.dev);
490 		}
491 	}
492 
493 	return (NULL);
494 }
495 
496 device_t
497 pci_find_class(uint8_t class, uint8_t subclass)
498 {
499 	struct pci_devinfo *dinfo;
500 
501 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
502 		if (dinfo->cfg.baseclass == class &&
503 		    dinfo->cfg.subclass == subclass) {
504 			return (dinfo->cfg.dev);
505 		}
506 	}
507 
508 	return (NULL);
509 }
510 
511 device_t
512 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
513 {
514 	struct pci_devinfo *dinfo;
515 	bool found = false;
516 
517 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
518 		if (from != NULL && found == false) {
519 			if (from != dinfo->cfg.dev)
520 				continue;
521 			found = true;
522 			continue;
523 		}
524 		if (dinfo->cfg.baseclass == class &&
525 		    dinfo->cfg.subclass == subclass) {
526 			return (dinfo->cfg.dev);
527 		}
528 	}
529 
530 	return (NULL);
531 }
532 
533 device_t
534 pci_find_base_class_from(uint8_t class, device_t from)
535 {
536 	struct pci_devinfo *dinfo;
537 	bool found = false;
538 
539 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
540 		if (from != NULL && found == false) {
541 			if (from != dinfo->cfg.dev)
542 				continue;
543 			found = true;
544 			continue;
545 		}
546 		if (dinfo->cfg.baseclass == class) {
547 			return (dinfo->cfg.dev);
548 		}
549 	}
550 
551 	return (NULL);
552 }
553 
554 static int
555 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
556 {
557 	va_list ap;
558 	int retval;
559 
560 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
561 	    cfg->func);
562 	va_start(ap, fmt);
563 	retval += vprintf(fmt, ap);
564 	va_end(ap);
565 	return (retval);
566 }
567 
568 /* return base address of memory or port map */
569 
570 static pci_addr_t
571 pci_mapbase(uint64_t mapreg)
572 {
573 
574 	if (PCI_BAR_MEM(mapreg))
575 		return (mapreg & PCIM_BAR_MEM_BASE);
576 	else
577 		return (mapreg & PCIM_BAR_IO_BASE);
578 }
579 
580 /* return map type of memory or port map */
581 
582 static const char *
583 pci_maptype(uint64_t mapreg)
584 {
585 
586 	if (PCI_BAR_IO(mapreg))
587 		return ("I/O Port");
588 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
589 		return ("Prefetchable Memory");
590 	return ("Memory");
591 }
592 
593 /* return log2 of map size decoded for memory or port map */
594 
595 int
596 pci_mapsize(uint64_t testval)
597 {
598 	int ln2size;
599 
600 	testval = pci_mapbase(testval);
601 	ln2size = 0;
602 	if (testval != 0) {
603 		while ((testval & 1) == 0)
604 		{
605 			ln2size++;
606 			testval >>= 1;
607 		}
608 	}
609 	return (ln2size);
610 }
611 
612 /* return base address of device ROM */
613 
614 static pci_addr_t
615 pci_rombase(uint64_t mapreg)
616 {
617 
618 	return (mapreg & PCIM_BIOS_ADDR_MASK);
619 }
620 
621 /* return log2 of map size decided for device ROM */
622 
623 static int
624 pci_romsize(uint64_t testval)
625 {
626 	int ln2size;
627 
628 	testval = pci_rombase(testval);
629 	ln2size = 0;
630 	if (testval != 0) {
631 		while ((testval & 1) == 0)
632 		{
633 			ln2size++;
634 			testval >>= 1;
635 		}
636 	}
637 	return (ln2size);
638 }
639 
640 /* return log2 of address range supported by map register */
641 
642 static int
643 pci_maprange(uint64_t mapreg)
644 {
645 	int ln2range = 0;
646 
647 	if (PCI_BAR_IO(mapreg))
648 		ln2range = 32;
649 	else
650 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
651 		case PCIM_BAR_MEM_32:
652 			ln2range = 32;
653 			break;
654 		case PCIM_BAR_MEM_1MB:
655 			ln2range = 20;
656 			break;
657 		case PCIM_BAR_MEM_64:
658 			ln2range = 64;
659 			break;
660 		}
661 	return (ln2range);
662 }
663 
664 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
665 
666 static void
667 pci_fixancient(pcicfgregs *cfg)
668 {
669 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
670 		return;
671 
672 	/* PCI to PCI bridges use header type 1 */
673 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
674 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
675 }
676 
677 /* extract header type specific config data */
678 
679 static void
680 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
681 {
682 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
683 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
684 	case PCIM_HDRTYPE_NORMAL:
685 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
686 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
687 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
688 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
689 		cfg->nummaps	    = PCI_MAXMAPS_0;
690 		break;
691 	case PCIM_HDRTYPE_BRIDGE:
692 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
693 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
694 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
695 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
696 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
697 		cfg->nummaps	    = PCI_MAXMAPS_1;
698 		break;
699 	case PCIM_HDRTYPE_CARDBUS:
700 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
701 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
702 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
703 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
704 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
705 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
706 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
707 		cfg->nummaps	    = PCI_MAXMAPS_2;
708 		break;
709 	}
710 #undef REG
711 }
712 
713 /* read configuration header into pcicfgregs structure */
714 struct pci_devinfo *
715 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
716 {
717 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
718 	uint16_t vid, did;
719 
720 	vid = REG(PCIR_VENDOR, 2);
721 	if (vid == PCIV_INVALID)
722 		return (NULL);
723 
724 	did = REG(PCIR_DEVICE, 2);
725 
726 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
727 }
728 
729 struct pci_devinfo *
730 pci_alloc_devinfo_method(device_t dev)
731 {
732 
733 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
734 	    M_WAITOK | M_ZERO));
735 }
736 
737 static struct pci_devinfo *
738 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
739     uint16_t vid, uint16_t did)
740 {
741 	struct pci_devinfo *devlist_entry;
742 	pcicfgregs *cfg;
743 
744 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
745 
746 	cfg = &devlist_entry->cfg;
747 
748 	cfg->domain		= d;
749 	cfg->bus		= b;
750 	cfg->slot		= s;
751 	cfg->func		= f;
752 	cfg->vendor		= vid;
753 	cfg->device		= did;
754 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
755 	cfg->statreg		= REG(PCIR_STATUS, 2);
756 	cfg->baseclass		= REG(PCIR_CLASS, 1);
757 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
758 	cfg->progif		= REG(PCIR_PROGIF, 1);
759 	cfg->revid		= REG(PCIR_REVID, 1);
760 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
761 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
762 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
763 	cfg->intpin		= REG(PCIR_INTPIN, 1);
764 	cfg->intline		= REG(PCIR_INTLINE, 1);
765 
766 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
767 	cfg->hdrtype		&= ~PCIM_MFDEV;
768 	STAILQ_INIT(&cfg->maps);
769 
770 	cfg->iov		= NULL;
771 
772 	pci_fixancient(cfg);
773 	pci_hdrtypedata(pcib, b, s, f, cfg);
774 
775 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
776 		pci_read_cap(pcib, cfg);
777 
778 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
779 
780 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
781 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
782 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
783 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
784 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
785 
786 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
787 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
788 	devlist_entry->conf.pc_vendor = cfg->vendor;
789 	devlist_entry->conf.pc_device = cfg->device;
790 
791 	devlist_entry->conf.pc_class = cfg->baseclass;
792 	devlist_entry->conf.pc_subclass = cfg->subclass;
793 	devlist_entry->conf.pc_progif = cfg->progif;
794 	devlist_entry->conf.pc_revid = cfg->revid;
795 
796 	pci_numdevs++;
797 	pci_generation++;
798 
799 	return (devlist_entry);
800 }
801 #undef REG
802 
803 static void
804 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
805 {
806 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
807     cfg->ea.ea_location + (n), w)
808 	int num_ent;
809 	int ptr;
810 	int a, b;
811 	uint32_t val;
812 	int ent_size;
813 	uint32_t dw[4];
814 	uint64_t base, max_offset;
815 	struct pci_ea_entry *eae;
816 
817 	if (cfg->ea.ea_location == 0)
818 		return;
819 
820 	STAILQ_INIT(&cfg->ea.ea_entries);
821 
822 	/* Determine the number of entries */
823 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
824 	num_ent &= PCIM_EA_NUM_ENT_MASK;
825 
826 	/* Find the first entry to care of */
827 	ptr = PCIR_EA_FIRST_ENT;
828 
829 	/* Skip DWORD 2 for type 1 functions */
830 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
831 		ptr += 4;
832 
833 	for (a = 0; a < num_ent; a++) {
834 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
835 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
836 
837 		/* Read a number of dwords in the entry */
838 		val = REG(ptr, 4);
839 		ptr += 4;
840 		ent_size = (val & PCIM_EA_ES);
841 
842 		for (b = 0; b < ent_size; b++) {
843 			dw[b] = REG(ptr, 4);
844 			ptr += 4;
845 		}
846 
847 		eae->eae_flags = val;
848 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
849 
850 		base = dw[0] & PCIM_EA_FIELD_MASK;
851 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
852 		b = 2;
853 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
854 			base |= (uint64_t)dw[b] << 32UL;
855 			b++;
856 		}
857 		if (((dw[1] & PCIM_EA_IS_64) != 0)
858 		    && (b < ent_size)) {
859 			max_offset |= (uint64_t)dw[b] << 32UL;
860 			b++;
861 		}
862 
863 		eae->eae_base = base;
864 		eae->eae_max_offset = max_offset;
865 
866 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
867 
868 		if (bootverbose) {
869 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
870 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
871 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
872 		}
873 	}
874 }
875 #undef REG
876 
877 static void
878 pci_read_cap(device_t pcib, pcicfgregs *cfg)
879 {
880 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
881 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
882 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
883 	uint64_t addr;
884 #endif
885 	uint32_t val;
886 	int	ptr, nextptr, ptrptr;
887 
888 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
889 	case PCIM_HDRTYPE_NORMAL:
890 	case PCIM_HDRTYPE_BRIDGE:
891 		ptrptr = PCIR_CAP_PTR;
892 		break;
893 	case PCIM_HDRTYPE_CARDBUS:
894 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
895 		break;
896 	default:
897 		return;		/* no extended capabilities support */
898 	}
899 	nextptr = REG(ptrptr, 1);	/* sanity check? */
900 
901 	/*
902 	 * Read capability entries.
903 	 */
904 	while (nextptr != 0) {
905 		/* Sanity check */
906 		if (nextptr > 255) {
907 			printf("illegal PCI extended capability offset %d\n",
908 			    nextptr);
909 			return;
910 		}
911 		/* Find the next entry */
912 		ptr = nextptr;
913 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
914 
915 		/* Process this entry */
916 		switch (REG(ptr + PCICAP_ID, 1)) {
917 		case PCIY_PMG:		/* PCI power management */
918 			cfg->pp.pp_location = ptr;
919 			cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
920 			break;
921 		case PCIY_HT:		/* HyperTransport */
922 			/* Determine HT-specific capability type. */
923 			val = REG(ptr + PCIR_HT_COMMAND, 2);
924 
925 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
926 				cfg->ht.ht_slave = ptr;
927 
928 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
929 			switch (val & PCIM_HTCMD_CAP_MASK) {
930 			case PCIM_HTCAP_MSI_MAPPING:
931 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
932 					/* Sanity check the mapping window. */
933 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
934 					    4);
935 					addr <<= 32;
936 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
937 					    4);
938 					if (addr != MSI_INTEL_ADDR_BASE)
939 						device_printf(pcib,
940 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
941 						    cfg->domain, cfg->bus,
942 						    cfg->slot, cfg->func,
943 						    (long long)addr);
944 				} else
945 					addr = MSI_INTEL_ADDR_BASE;
946 
947 				cfg->ht.ht_msimap = ptr;
948 				cfg->ht.ht_msictrl = val;
949 				cfg->ht.ht_msiaddr = addr;
950 				break;
951 			}
952 #endif
953 			break;
954 		case PCIY_MSI:		/* PCI MSI */
955 			cfg->msi.msi_location = ptr;
956 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
957 			break;
958 		case PCIY_MSIX:		/* PCI MSI-X */
959 			cfg->msix.msix_location = ptr;
960 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
961 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
962 			cfg->msix.msix_table_bar = PCIR_BAR(val &
963 			    PCIM_MSIX_BIR_MASK);
964 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
965 			val = REG(ptr + PCIR_MSIX_PBA, 4);
966 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
967 			    PCIM_MSIX_BIR_MASK);
968 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
969 			break;
970 		case PCIY_VPD:		/* PCI Vital Product Data */
971 			cfg->vpd.vpd_reg = ptr;
972 			break;
973 		case PCIY_SUBVENDOR:
974 			/* Should always be true. */
975 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
976 			    PCIM_HDRTYPE_BRIDGE) {
977 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
978 				cfg->subvendor = val & 0xffff;
979 				cfg->subdevice = val >> 16;
980 			}
981 			break;
982 		case PCIY_PCIX:		/* PCI-X */
983 			/*
984 			 * Assume we have a PCI-X chipset if we have
985 			 * at least one PCI-PCI bridge with a PCI-X
986 			 * capability.  Note that some systems with
987 			 * PCI-express or HT chipsets might match on
988 			 * this check as well.
989 			 */
990 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
991 			    PCIM_HDRTYPE_BRIDGE)
992 				pcix_chipset = 1;
993 			cfg->pcix.pcix_location = ptr;
994 			break;
995 		case PCIY_EXPRESS:	/* PCI-express */
996 			/*
997 			 * Assume we have a PCI-express chipset if we have
998 			 * at least one PCI-express device.
999 			 */
1000 			pcie_chipset = 1;
1001 			cfg->pcie.pcie_location = ptr;
1002 			val = REG(ptr + PCIER_FLAGS, 2);
1003 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
1004 			break;
1005 		case PCIY_EA:		/* Enhanced Allocation */
1006 			cfg->ea.ea_location = ptr;
1007 			pci_ea_fill_info(pcib, cfg);
1008 			break;
1009 		default:
1010 			break;
1011 		}
1012 	}
1013 
1014 #if defined(__powerpc__)
1015 	/*
1016 	 * Enable the MSI mapping window for all HyperTransport
1017 	 * slaves.  PCI-PCI bridges have their windows enabled via
1018 	 * PCIB_MAP_MSI().
1019 	 */
1020 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1021 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1022 		device_printf(pcib,
1023 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1024 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1025 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1026 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1027 		     2);
1028 	}
1029 #endif
1030 /* REG and WREG use carry through to next functions */
1031 }
1032 
1033 /*
1034  * PCI Vital Product Data
1035  */
1036 
1037 #define	PCI_VPD_TIMEOUT		1000000
1038 
1039 static int
1040 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1041 {
1042 	int count = PCI_VPD_TIMEOUT;
1043 
1044 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1045 
1046 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1047 
1048 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1049 		if (--count < 0)
1050 			return (ENXIO);
1051 		DELAY(1);	/* limit looping */
1052 	}
1053 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1054 
1055 	return (0);
1056 }
1057 
1058 #if 0
1059 static int
1060 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1061 {
1062 	int count = PCI_VPD_TIMEOUT;
1063 
1064 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1065 
1066 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1067 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1068 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1069 		if (--count < 0)
1070 			return (ENXIO);
1071 		DELAY(1);	/* limit looping */
1072 	}
1073 
1074 	return (0);
1075 }
1076 #endif
1077 
1078 #undef PCI_VPD_TIMEOUT
1079 
1080 struct vpd_readstate {
1081 	device_t	pcib;
1082 	pcicfgregs	*cfg;
1083 	uint32_t	val;
1084 	int		bytesinval;
1085 	int		off;
1086 	uint8_t		cksum;
1087 };
1088 
1089 /* return 0 and one byte in *data if no read error, -1 else */
1090 static int
1091 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1092 {
1093 	uint32_t reg;
1094 	uint8_t byte;
1095 
1096 	if (vrs->bytesinval == 0) {
1097 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1098 			return (-1);
1099 		vrs->val = le32toh(reg);
1100 		vrs->off += 4;
1101 		byte = vrs->val & 0xff;
1102 		vrs->bytesinval = 3;
1103 	} else {
1104 		vrs->val = vrs->val >> 8;
1105 		byte = vrs->val & 0xff;
1106 		vrs->bytesinval--;
1107 	}
1108 
1109 	vrs->cksum += byte;
1110 	*data = byte;
1111 	return (0);
1112 }
1113 
1114 /* return 0 on match, -1 and "unget" byte on no match */
1115 static int
1116 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1117 {
1118 	uint8_t data;
1119 
1120 	if (vpd_nextbyte(vrs, &data) != 0)
1121 		return (-1);
1122 
1123 	if (data == expected)
1124 		return (0);
1125 
1126 	vrs->cksum -= data;
1127 	vrs->val = (vrs->val << 8) + data;
1128 	vrs->bytesinval++;
1129 	return (-1);
1130 }
1131 
1132 /* return size if tag matches, -1 on no match, -2 on read error */
1133 static int
1134 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1135 {
1136 	uint8_t byte1, byte2;
1137 
1138 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1139 		return (-1);
1140 
1141 	if ((vpd_tag & 0x80) == 0)
1142 		return (vpd_tag & 0x07);
1143 
1144 	if (vpd_nextbyte(vrs, &byte1) != 0)
1145 		return (-2);
1146 	if (vpd_nextbyte(vrs, &byte2) != 0)
1147 		return (-2);
1148 
1149 	return ((byte2 << 8) + byte1);
1150 }
1151 
1152 /* (re)allocate buffer in multiples of 8 elements */
1153 static void*
1154 alloc_buffer(void* buffer, size_t element_size, int needed)
1155 {
1156 	int alloc, new_alloc;
1157 
1158 	alloc = roundup2(needed, 8);
1159 	new_alloc = roundup2(needed + 1, 8);
1160 	if (alloc != new_alloc) {
1161 		buffer = reallocf(buffer,
1162 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1163 	}
1164 
1165 	return (buffer);
1166 }
1167 
1168 /* read VPD keyword and return element size, return -1 on read error */
1169 static int
1170 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1171 {
1172 	uint8_t data;
1173 
1174 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1175 		return (-1);
1176 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1177 		return (-1);
1178 	if (vpd_nextbyte(vrs, &data) != 0)
1179 		return (-1);
1180 
1181 	return (data);
1182 }
1183 
1184 /* read VPD data element of given size into allocated buffer */
1185 static char *
1186 vpd_read_value(struct vpd_readstate *vrs, int size)
1187 {
1188 	int i;
1189 	char char1;
1190 	char *value;
1191 
1192 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1193 	for (i = 0; i < size; i++) {
1194 		if (vpd_nextbyte(vrs, &char1) != 0) {
1195 			free(value, M_DEVBUF);
1196 			return (NULL);
1197 		}
1198 		value[i] = char1;
1199 	}
1200 	value[size] = '\0';
1201 
1202 	return (value);
1203 }
1204 
1205 /* read VPD into *keyword and *value, return length of data element */
1206 static int
1207 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1208 {
1209 	int len;
1210 
1211 	len = vpd_read_elem_head(vrs, keyword);
1212 	if (len < 0 || len > maxlen)
1213 		return (-1);
1214 	*value = vpd_read_value(vrs, len);
1215 
1216 	return (len);
1217 }
1218 
1219 /* subtract all data following first byte from checksum of RV element */
1220 static void
1221 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1222 {
1223 	int i;
1224 	uint8_t fixup;
1225 
1226 	fixup = 0;
1227 	for (i = 1; i < len; i++)
1228 		fixup += rvstring[i];
1229 	vrs->cksum -= fixup;
1230 }
1231 
1232 /* fetch one read-only element and return size of heading + data */
1233 static int
1234 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1235 {
1236 	struct pcicfg_vpd *vpd;
1237 	pcicfgregs *cfg;
1238 	struct vpd_readonly *vpd_ros;
1239 	int len;
1240 
1241 	cfg = vrs->cfg;
1242 	vpd = &cfg->vpd;
1243 
1244 	if (maxsize < 3)
1245 		return (-1);
1246 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1247 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1248 	maxsize -= 3;
1249 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1250 	if (vpd_ros->value == NULL)
1251 		return (-1);
1252 	vpd_ros->len = len;
1253 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1254 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1255 		if (vrs->cksum != 0) {
1256 			pci_printf(cfg,
1257 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1258 			return (-1);
1259 		}
1260 	}
1261 	vpd->vpd_rocnt++;
1262 
1263 	return (len + 3);
1264 }
1265 
1266 /* fetch one writable element and return size of heading + data */
1267 static int
1268 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1269 {
1270 	struct pcicfg_vpd *vpd;
1271 	pcicfgregs *cfg;
1272 	struct vpd_write *vpd_w;
1273 	int len;
1274 
1275 	cfg = vrs->cfg;
1276 	vpd = &cfg->vpd;
1277 
1278 	if (maxsize < 3)
1279 		return (-1);
1280 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1281 	if (vpd->vpd_w == NULL) {
1282 		pci_printf(cfg, "out of memory");
1283 		return (-1);
1284 	}
1285 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1286 	maxsize -= 3;
1287 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1288 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1289 	if (vpd_w->value == NULL)
1290 		return (-1);
1291 	vpd_w->len = len;
1292 	vpd->vpd_wcnt++;
1293 
1294 	return (len + 3);
1295 }
1296 
1297 /* free all memory allocated for VPD data */
1298 static void
1299 vpd_free(struct pcicfg_vpd *vpd)
1300 {
1301 	int i;
1302 
1303 	free(vpd->vpd_ident, M_DEVBUF);
1304 	for (i = 0; i < vpd->vpd_rocnt; i++)
1305 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1306 	free(vpd->vpd_ros, M_DEVBUF);
1307 	vpd->vpd_rocnt = 0;
1308 	for (i = 0; i < vpd->vpd_wcnt; i++)
1309 		free(vpd->vpd_w[i].value, M_DEVBUF);
1310 	free(vpd->vpd_w, M_DEVBUF);
1311 	vpd->vpd_wcnt = 0;
1312 }
1313 
1314 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1315 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1316 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1317 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1318 
1319 static int
1320 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1321 {
1322 	struct vpd_readstate vrs;
1323 	int cksumvalid;
1324 	int size, elem_size;
1325 
1326 	/* init vpd reader */
1327 	vrs.bytesinval = 0;
1328 	vrs.off = 0;
1329 	vrs.pcib = pcib;
1330 	vrs.cfg = cfg;
1331 	vrs.cksum = 0;
1332 
1333 	/* read VPD ident element - mandatory */
1334 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1335 	if (size <= 0) {
1336 		pci_printf(cfg, "no VPD ident found\n");
1337 		return (0);
1338 	}
1339 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1340 	if (cfg->vpd.vpd_ident == NULL) {
1341 		pci_printf(cfg, "error accessing VPD ident data\n");
1342 		return (0);
1343 	}
1344 
1345 	/* read VPD RO elements - mandatory */
1346 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1347 	if (size <= 0) {
1348 		pci_printf(cfg, "no read-only VPD data found\n");
1349 		return (0);
1350 	}
1351 	while (size > 0) {
1352 		elem_size = next_vpd_ro_elem(&vrs, size);
1353 		if (elem_size < 0) {
1354 			pci_printf(cfg, "error accessing read-only VPD data\n");
1355 			return (-1);
1356 		}
1357 		size -= elem_size;
1358 	}
1359 	cksumvalid = (vrs.cksum == 0);
1360 	if (!cksumvalid)
1361 		return (-1);
1362 
1363 	/* read VPD RW elements - optional */
1364 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1365 	if (size == -2)
1366 		return (-1);
1367 	while (size > 0) {
1368 		elem_size = next_vpd_rw_elem(&vrs, size);
1369 		if (elem_size < 0) {
1370 			pci_printf(cfg, "error accessing writeable VPD data\n");
1371 			return (-1);
1372 		}
1373 		size -= elem_size;
1374 	}
1375 
1376 	/* read empty END tag - mandatory */
1377 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1378 	if (size != 0) {
1379 		pci_printf(cfg, "No valid VPD end tag found\n");
1380 	}
1381 	return (0);
1382 }
1383 
1384 static void
1385 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1386 {
1387 	int status;
1388 
1389 	status = pci_parse_vpd(pcib, cfg);
1390 	if (status < 0)
1391 		vpd_free(&cfg->vpd);
1392 	cfg->vpd.vpd_cached = 1;
1393 #undef REG
1394 #undef WREG
1395 }
1396 
1397 int
1398 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1399 {
1400 	struct pci_devinfo *dinfo = device_get_ivars(child);
1401 	pcicfgregs *cfg = &dinfo->cfg;
1402 
1403 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1404 		pci_read_vpd(device_get_parent(dev), cfg);
1405 
1406 	*identptr = cfg->vpd.vpd_ident;
1407 
1408 	if (*identptr == NULL)
1409 		return (ENXIO);
1410 
1411 	return (0);
1412 }
1413 
1414 int
1415 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1416 	const char **vptr)
1417 {
1418 	struct pci_devinfo *dinfo = device_get_ivars(child);
1419 	pcicfgregs *cfg = &dinfo->cfg;
1420 	int i;
1421 
1422 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1423 		pci_read_vpd(device_get_parent(dev), cfg);
1424 
1425 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1426 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1427 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1428 			*vptr = cfg->vpd.vpd_ros[i].value;
1429 			return (0);
1430 		}
1431 
1432 	*vptr = NULL;
1433 	return (ENXIO);
1434 }
1435 
1436 struct pcicfg_vpd *
1437 pci_fetch_vpd_list(device_t dev)
1438 {
1439 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1440 	pcicfgregs *cfg = &dinfo->cfg;
1441 
1442 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1443 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1444 	return (&cfg->vpd);
1445 }
1446 
1447 /*
1448  * Find the requested HyperTransport capability and return the offset
1449  * in configuration space via the pointer provided.  The function
1450  * returns 0 on success and an error code otherwise.
1451  */
1452 int
1453 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1454 {
1455 	int ptr, error;
1456 	uint16_t val;
1457 
1458 	error = pci_find_cap(child, PCIY_HT, &ptr);
1459 	if (error)
1460 		return (error);
1461 
1462 	/*
1463 	 * Traverse the capabilities list checking each HT capability
1464 	 * to see if it matches the requested HT capability.
1465 	 */
1466 	for (;;) {
1467 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1468 		if (capability == PCIM_HTCAP_SLAVE ||
1469 		    capability == PCIM_HTCAP_HOST)
1470 			val &= 0xe000;
1471 		else
1472 			val &= PCIM_HTCMD_CAP_MASK;
1473 		if (val == capability) {
1474 			if (capreg != NULL)
1475 				*capreg = ptr;
1476 			return (0);
1477 		}
1478 
1479 		/* Skip to the next HT capability. */
1480 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1481 			break;
1482 	}
1483 
1484 	return (ENOENT);
1485 }
1486 
1487 /*
1488  * Find the next requested HyperTransport capability after start and return
1489  * the offset in configuration space via the pointer provided.  The function
1490  * returns 0 on success and an error code otherwise.
1491  */
1492 int
1493 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1494     int start, int *capreg)
1495 {
1496 	int ptr;
1497 	uint16_t val;
1498 
1499 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1500 	    ("start capability is not HyperTransport capability"));
1501 	ptr = start;
1502 
1503 	/*
1504 	 * Traverse the capabilities list checking each HT capability
1505 	 * to see if it matches the requested HT capability.
1506 	 */
1507 	for (;;) {
1508 		/* Skip to the next HT capability. */
1509 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1510 			break;
1511 
1512 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1513 		if (capability == PCIM_HTCAP_SLAVE ||
1514 		    capability == PCIM_HTCAP_HOST)
1515 			val &= 0xe000;
1516 		else
1517 			val &= PCIM_HTCMD_CAP_MASK;
1518 		if (val == capability) {
1519 			if (capreg != NULL)
1520 				*capreg = ptr;
1521 			return (0);
1522 		}
1523 	}
1524 
1525 	return (ENOENT);
1526 }
1527 
1528 /*
1529  * Find the requested capability and return the offset in
1530  * configuration space via the pointer provided.  The function returns
1531  * 0 on success and an error code otherwise.
1532  */
1533 int
1534 pci_find_cap_method(device_t dev, device_t child, int capability,
1535     int *capreg)
1536 {
1537 	struct pci_devinfo *dinfo = device_get_ivars(child);
1538 	pcicfgregs *cfg = &dinfo->cfg;
1539 	uint32_t status;
1540 	uint8_t ptr;
1541 	int cnt;
1542 
1543 	/*
1544 	 * Check the CAP_LIST bit of the PCI status register first.
1545 	 */
1546 	status = pci_read_config(child, PCIR_STATUS, 2);
1547 	if (!(status & PCIM_STATUS_CAPPRESENT))
1548 		return (ENXIO);
1549 
1550 	/*
1551 	 * Determine the start pointer of the capabilities list.
1552 	 */
1553 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1554 	case PCIM_HDRTYPE_NORMAL:
1555 	case PCIM_HDRTYPE_BRIDGE:
1556 		ptr = PCIR_CAP_PTR;
1557 		break;
1558 	case PCIM_HDRTYPE_CARDBUS:
1559 		ptr = PCIR_CAP_PTR_2;
1560 		break;
1561 	default:
1562 		/* XXX: panic? */
1563 		return (ENXIO);		/* no extended capabilities support */
1564 	}
1565 	ptr = pci_read_config(child, ptr, 1);
1566 
1567 	/*
1568 	 * Traverse the capabilities list.  Limit by total theoretical
1569 	 * maximum number of caps: capability needs at least id and
1570 	 * next registers, and any type X header cannot contain caps.
1571 	 */
1572 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1573 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1574 			if (capreg != NULL)
1575 				*capreg = ptr;
1576 			return (0);
1577 		}
1578 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1579 	}
1580 
1581 	return (ENOENT);
1582 }
1583 
1584 /*
1585  * Find the next requested capability after start and return the offset in
1586  * configuration space via the pointer provided.  The function returns
1587  * 0 on success and an error code otherwise.
1588  */
1589 int
1590 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1591     int start, int *capreg)
1592 {
1593 	uint8_t ptr;
1594 
1595 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1596 	    ("start capability is not expected capability"));
1597 
1598 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1599 	while (ptr != 0) {
1600 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1601 			if (capreg != NULL)
1602 				*capreg = ptr;
1603 			return (0);
1604 		}
1605 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1606 	}
1607 
1608 	return (ENOENT);
1609 }
1610 
1611 /*
1612  * Find the requested extended capability and return the offset in
1613  * configuration space via the pointer provided.  The function returns
1614  * 0 on success and an error code otherwise.
1615  */
1616 int
1617 pci_find_extcap_method(device_t dev, device_t child, int capability,
1618     int *capreg)
1619 {
1620 	struct pci_devinfo *dinfo = device_get_ivars(child);
1621 	pcicfgregs *cfg = &dinfo->cfg;
1622 	uint32_t ecap;
1623 	uint16_t ptr;
1624 
1625 	/* Only supported for PCI-express devices. */
1626 	if (cfg->pcie.pcie_location == 0)
1627 		return (ENXIO);
1628 
1629 	ptr = PCIR_EXTCAP;
1630 	ecap = pci_read_config(child, ptr, 4);
1631 	if (ecap == 0xffffffff || ecap == 0)
1632 		return (ENOENT);
1633 	for (;;) {
1634 		if (PCI_EXTCAP_ID(ecap) == capability) {
1635 			if (capreg != NULL)
1636 				*capreg = ptr;
1637 			return (0);
1638 		}
1639 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1640 		if (ptr == 0)
1641 			break;
1642 		ecap = pci_read_config(child, ptr, 4);
1643 	}
1644 
1645 	return (ENOENT);
1646 }
1647 
1648 /*
1649  * Find the next requested extended capability after start and return the
1650  * offset in configuration space via the pointer provided.  The function
1651  * returns 0 on success and an error code otherwise.
1652  */
1653 int
1654 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1655     int start, int *capreg)
1656 {
1657 	struct pci_devinfo *dinfo = device_get_ivars(child);
1658 	pcicfgregs *cfg = &dinfo->cfg;
1659 	uint32_t ecap;
1660 	uint16_t ptr;
1661 
1662 	/* Only supported for PCI-express devices. */
1663 	if (cfg->pcie.pcie_location == 0)
1664 		return (ENXIO);
1665 
1666 	ecap = pci_read_config(child, start, 4);
1667 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1668 	    ("start extended capability is not expected capability"));
1669 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1670 	while (ptr != 0) {
1671 		ecap = pci_read_config(child, ptr, 4);
1672 		if (PCI_EXTCAP_ID(ecap) == capability) {
1673 			if (capreg != NULL)
1674 				*capreg = ptr;
1675 			return (0);
1676 		}
1677 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1678 	}
1679 
1680 	return (ENOENT);
1681 }
1682 
1683 /*
1684  * Support for MSI-X message interrupts.
1685  */
1686 static void
1687 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1688 {
1689 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1690 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1691 	uint32_t offset;
1692 
1693 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1694 	offset = msix->msix_table_offset + index * 16;
1695 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1696 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1697 	bus_write_4(msix->msix_table_res, offset + 8, data);
1698 }
1699 
1700 void
1701 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1702     uint64_t address, uint32_t data)
1703 {
1704 
1705 	if (pci_msix_rewrite_table) {
1706 		struct pci_devinfo *dinfo = device_get_ivars(child);
1707 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1708 
1709 		/*
1710 		 * Some VM hosts require MSIX to be disabled in the
1711 		 * control register before updating the MSIX table
1712 		 * entries are allowed. It is not enough to only
1713 		 * disable MSIX while updating a single entry. MSIX
1714 		 * must be disabled while updating all entries in the
1715 		 * table.
1716 		 */
1717 		pci_write_config(child,
1718 		    msix->msix_location + PCIR_MSIX_CTRL,
1719 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1720 		pci_resume_msix(child);
1721 	} else
1722 		pci_write_msix_entry(child, index, address, data);
1723 
1724 	/* Enable MSI -> HT mapping. */
1725 	pci_ht_map_msi(child, address);
1726 }
1727 
1728 void
1729 pci_mask_msix(device_t dev, u_int index)
1730 {
1731 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1732 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1733 	uint32_t offset, val;
1734 
1735 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1736 	offset = msix->msix_table_offset + index * 16 + 12;
1737 	val = bus_read_4(msix->msix_table_res, offset);
1738 	val |= PCIM_MSIX_VCTRL_MASK;
1739 
1740 	/*
1741 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1742 	 * register, so always write the new value.
1743 	 */
1744 	bus_write_4(msix->msix_table_res, offset, val);
1745 }
1746 
1747 void
1748 pci_unmask_msix(device_t dev, u_int index)
1749 {
1750 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1751 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1752 	uint32_t offset, val;
1753 
1754 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1755 	offset = msix->msix_table_offset + index * 16 + 12;
1756 	val = bus_read_4(msix->msix_table_res, offset);
1757 	val &= ~PCIM_MSIX_VCTRL_MASK;
1758 
1759 	/*
1760 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1761 	 * register, so always write the new value.
1762 	 */
1763 	bus_write_4(msix->msix_table_res, offset, val);
1764 }
1765 
1766 int
1767 pci_pending_msix(device_t dev, u_int index)
1768 {
1769 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1770 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1771 	uint32_t offset, bit;
1772 
1773 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1774 	offset = msix->msix_pba_offset + (index / 32) * 4;
1775 	bit = 1 << index % 32;
1776 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1777 }
1778 
1779 /*
1780  * Restore MSI-X registers and table during resume.  If MSI-X is
1781  * enabled then walk the virtual table to restore the actual MSI-X
1782  * table.
1783  */
1784 static void
1785 pci_resume_msix(device_t dev)
1786 {
1787 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1788 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1789 	struct msix_table_entry *mte;
1790 	struct msix_vector *mv;
1791 	u_int i, msgnum;
1792 
1793 	if (msix->msix_alloc > 0) {
1794 		msgnum = PCI_MSIX_MSGNUM(msix->msix_ctrl);
1795 
1796 		/* First, mask all vectors. */
1797 		for (i = 0; i < msgnum; i++)
1798 			pci_mask_msix(dev, i);
1799 
1800 		/* Second, program any messages with at least one handler. */
1801 		for (i = 0; i < msix->msix_table_len; i++) {
1802 			mte = &msix->msix_table[i];
1803 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1804 				continue;
1805 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1806 			pci_write_msix_entry(dev, i, mv->mv_address,
1807 			    mv->mv_data);
1808 			pci_unmask_msix(dev, i);
1809 		}
1810 	}
1811 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1812 	    msix->msix_ctrl, 2);
1813 }
1814 
1815 /*
1816  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1817  * returned in *count.  After this function returns, each message will be
1818  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1819  */
1820 int
1821 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1822 {
1823 	struct pci_devinfo *dinfo = device_get_ivars(child);
1824 	pcicfgregs *cfg = &dinfo->cfg;
1825 	struct resource_list_entry *rle;
1826 	u_int actual, i, max;
1827 	int error, irq;
1828 	uint16_t ctrl, msgnum;
1829 
1830 	/* Don't let count == 0 get us into trouble. */
1831 	if (*count < 1)
1832 		return (EINVAL);
1833 
1834 	/* If rid 0 is allocated, then fail. */
1835 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1836 	if (rle != NULL && rle->res != NULL)
1837 		return (ENXIO);
1838 
1839 	/* Already have allocated messages? */
1840 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1841 		return (ENXIO);
1842 
1843 	/* If MSI-X is blacklisted for this system, fail. */
1844 	if (pci_msix_blacklisted())
1845 		return (ENXIO);
1846 
1847 	/* MSI-X capability present? */
1848 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1849 		return (ENODEV);
1850 
1851 	/* Make sure the appropriate BARs are mapped. */
1852 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1853 	    cfg->msix.msix_table_bar);
1854 	if (rle == NULL || rle->res == NULL ||
1855 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1856 		return (ENXIO);
1857 	cfg->msix.msix_table_res = rle->res;
1858 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1859 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1860 		    cfg->msix.msix_pba_bar);
1861 		if (rle == NULL || rle->res == NULL ||
1862 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1863 			return (ENXIO);
1864 	}
1865 	cfg->msix.msix_pba_res = rle->res;
1866 
1867 	ctrl = pci_read_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1868 	    2);
1869 	msgnum = PCI_MSIX_MSGNUM(ctrl);
1870 	if (bootverbose)
1871 		device_printf(child,
1872 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1873 		    *count, msgnum);
1874 	max = min(*count, msgnum);
1875 	for (i = 0; i < max; i++) {
1876 		/* Allocate a message. */
1877 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1878 		if (error) {
1879 			if (i == 0)
1880 				return (error);
1881 			break;
1882 		}
1883 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1884 		    irq, 1);
1885 	}
1886 	actual = i;
1887 
1888 	if (bootverbose) {
1889 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1890 		if (actual == 1)
1891 			device_printf(child, "using IRQ %ju for MSI-X\n",
1892 			    rle->start);
1893 		else {
1894 			bool run;
1895 
1896 			/*
1897 			 * Be fancy and try to print contiguous runs of
1898 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1899 			 * 'run' is true if we are in a range.
1900 			 */
1901 			device_printf(child, "using IRQs %ju", rle->start);
1902 			irq = rle->start;
1903 			run = false;
1904 			for (i = 1; i < actual; i++) {
1905 				rle = resource_list_find(&dinfo->resources,
1906 				    SYS_RES_IRQ, i + 1);
1907 
1908 				/* Still in a run? */
1909 				if (rle->start == irq + 1) {
1910 					run = true;
1911 					irq++;
1912 					continue;
1913 				}
1914 
1915 				/* Finish previous range. */
1916 				if (run) {
1917 					printf("-%d", irq);
1918 					run = false;
1919 				}
1920 
1921 				/* Start new range. */
1922 				printf(",%ju", rle->start);
1923 				irq = rle->start;
1924 			}
1925 
1926 			/* Unfinished range? */
1927 			if (run)
1928 				printf("-%d", irq);
1929 			printf(" for MSI-X\n");
1930 		}
1931 	}
1932 
1933 	/*
1934 	 * Mask all vectors. Note that the message index assertion in
1935 	 * pci_mask_msix requires msix_ctrl to be set.
1936 	 */
1937 	cfg->msix.msix_ctrl = ctrl;
1938 	for (i = 0; i < msgnum; i++)
1939 		pci_mask_msix(child, i);
1940 
1941 	/* Allocate and initialize vector data and virtual table. */
1942 	cfg->msix.msix_vectors = mallocarray(actual, sizeof(struct msix_vector),
1943 	    M_DEVBUF, M_WAITOK | M_ZERO);
1944 	cfg->msix.msix_table = mallocarray(actual,
1945 	    sizeof(struct msix_table_entry), M_DEVBUF, M_WAITOK | M_ZERO);
1946 	for (i = 0; i < actual; i++) {
1947 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1948 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1949 		cfg->msix.msix_table[i].mte_vector = i + 1;
1950 	}
1951 
1952 	/* Update control register to enable MSI-X. */
1953 	ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1954 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1955 	    ctrl, 2);
1956 	cfg->msix.msix_ctrl = ctrl;
1957 
1958 	/* Update counts of alloc'd messages. */
1959 	cfg->msix.msix_alloc = actual;
1960 	cfg->msix.msix_table_len = actual;
1961 	*count = actual;
1962 	return (0);
1963 }
1964 
1965 /*
1966  * By default, pci_alloc_msix() will assign the allocated IRQ
1967  * resources consecutively to the first N messages in the MSI-X table.
1968  * However, device drivers may want to use different layouts if they
1969  * either receive fewer messages than they asked for, or they wish to
1970  * populate the MSI-X table sparsely.  This method allows the driver
1971  * to specify what layout it wants.  It must be called after a
1972  * successful pci_alloc_msix() but before any of the associated
1973  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1974  *
1975  * The 'vectors' array contains 'count' message vectors.  The array
1976  * maps directly to the MSI-X table in that index 0 in the array
1977  * specifies the vector for the first message in the MSI-X table, etc.
1978  * The vector value in each array index can either be 0 to indicate
1979  * that no vector should be assigned to a message slot, or it can be a
1980  * number from 1 to N (where N is the count returned from a
1981  * succcessful call to pci_alloc_msix()) to indicate which message
1982  * vector (IRQ) to be used for the corresponding message.
1983  *
1984  * On successful return, each message with a non-zero vector will have
1985  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1986  * 1.  Additionally, if any of the IRQs allocated via the previous
1987  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1988  * will be freed back to the system automatically.
1989  *
1990  * For example, suppose a driver has a MSI-X table with 6 messages and
1991  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1992  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1993  * C.  After the call to pci_alloc_msix(), the device will be setup to
1994  * have an MSI-X table of ABC--- (where - means no vector assigned).
1995  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1996  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1997  * be freed back to the system.  This device will also have valid
1998  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1999  *
2000  * In any case, the SYS_RES_IRQ rid X will always map to the message
2001  * at MSI-X table index X - 1 and will only be valid if a vector is
2002  * assigned to that table entry.
2003  */
2004 int
2005 pci_remap_msix_method(device_t dev, device_t child, int count,
2006     const u_int *vectors)
2007 {
2008 	struct pci_devinfo *dinfo = device_get_ivars(child);
2009 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2010 	struct resource_list_entry *rle;
2011 	u_int i, irq, j;
2012 	bool *used;
2013 
2014 	/*
2015 	 * Have to have at least one message in the table but the
2016 	 * table can't be bigger than the actual MSI-X table in the
2017 	 * device.
2018 	 */
2019 	if (count < 1 || count > PCI_MSIX_MSGNUM(msix->msix_ctrl))
2020 		return (EINVAL);
2021 
2022 	/* Sanity check the vectors. */
2023 	for (i = 0; i < count; i++)
2024 		if (vectors[i] > msix->msix_alloc)
2025 			return (EINVAL);
2026 
2027 	/*
2028 	 * Make sure there aren't any holes in the vectors to be used.
2029 	 * It's a big pain to support it, and it doesn't really make
2030 	 * sense anyway.  Also, at least one vector must be used.
2031 	 */
2032 	used = mallocarray(msix->msix_alloc, sizeof(*used), M_DEVBUF, M_WAITOK |
2033 	    M_ZERO);
2034 	for (i = 0; i < count; i++)
2035 		if (vectors[i] != 0)
2036 			used[vectors[i] - 1] = true;
2037 	for (i = 0; i < msix->msix_alloc - 1; i++)
2038 		if (!used[i] && used[i + 1]) {
2039 			free(used, M_DEVBUF);
2040 			return (EINVAL);
2041 		}
2042 	if (!used[0]) {
2043 		free(used, M_DEVBUF);
2044 		return (EINVAL);
2045 	}
2046 
2047 	/* Make sure none of the resources are allocated. */
2048 	for (i = 0; i < msix->msix_table_len; i++) {
2049 		if (msix->msix_table[i].mte_vector == 0)
2050 			continue;
2051 		if (msix->msix_table[i].mte_handlers > 0) {
2052 			free(used, M_DEVBUF);
2053 			return (EBUSY);
2054 		}
2055 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2056 		KASSERT(rle != NULL, ("missing resource"));
2057 		if (rle->res != NULL) {
2058 			free(used, M_DEVBUF);
2059 			return (EBUSY);
2060 		}
2061 	}
2062 
2063 	/* Free the existing resource list entries. */
2064 	for (i = 0; i < msix->msix_table_len; i++) {
2065 		if (msix->msix_table[i].mte_vector == 0)
2066 			continue;
2067 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2068 	}
2069 
2070 	/*
2071 	 * Build the new virtual table keeping track of which vectors are
2072 	 * used.
2073 	 */
2074 	free(msix->msix_table, M_DEVBUF);
2075 	msix->msix_table = mallocarray(count, sizeof(struct msix_table_entry),
2076 	    M_DEVBUF, M_WAITOK | M_ZERO);
2077 	for (i = 0; i < count; i++)
2078 		msix->msix_table[i].mte_vector = vectors[i];
2079 	msix->msix_table_len = count;
2080 
2081 	/* Free any unused IRQs and resize the vectors array if necessary. */
2082 	j = msix->msix_alloc - 1;
2083 	if (!used[j]) {
2084 		struct msix_vector *vec;
2085 
2086 		while (!used[j]) {
2087 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2088 			    msix->msix_vectors[j].mv_irq);
2089 			j--;
2090 		}
2091 		vec = mallocarray(j + 1, sizeof(struct msix_vector), M_DEVBUF,
2092 		    M_WAITOK);
2093 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2094 		    (j + 1));
2095 		free(msix->msix_vectors, M_DEVBUF);
2096 		msix->msix_vectors = vec;
2097 		msix->msix_alloc = j + 1;
2098 	}
2099 	free(used, M_DEVBUF);
2100 
2101 	/* Map the IRQs onto the rids. */
2102 	for (i = 0; i < count; i++) {
2103 		if (vectors[i] == 0)
2104 			continue;
2105 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2106 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2107 		    irq, 1);
2108 	}
2109 
2110 	if (bootverbose) {
2111 		device_printf(child, "Remapped MSI-X IRQs as: ");
2112 		for (i = 0; i < count; i++) {
2113 			if (i != 0)
2114 				printf(", ");
2115 			if (vectors[i] == 0)
2116 				printf("---");
2117 			else
2118 				printf("%d",
2119 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2120 		}
2121 		printf("\n");
2122 	}
2123 
2124 	return (0);
2125 }
2126 
2127 static int
2128 pci_release_msix(device_t dev, device_t child)
2129 {
2130 	struct pci_devinfo *dinfo = device_get_ivars(child);
2131 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2132 	struct resource_list_entry *rle;
2133 	u_int i;
2134 
2135 	/* Do we have any messages to release? */
2136 	if (msix->msix_alloc == 0)
2137 		return (ENODEV);
2138 
2139 	/* Make sure none of the resources are allocated. */
2140 	for (i = 0; i < msix->msix_table_len; i++) {
2141 		if (msix->msix_table[i].mte_vector == 0)
2142 			continue;
2143 		if (msix->msix_table[i].mte_handlers > 0)
2144 			return (EBUSY);
2145 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2146 		KASSERT(rle != NULL, ("missing resource"));
2147 		if (rle->res != NULL)
2148 			return (EBUSY);
2149 	}
2150 
2151 	/* Update control register to disable MSI-X. */
2152 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2153 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2154 	    msix->msix_ctrl, 2);
2155 
2156 	/* Free the resource list entries. */
2157 	for (i = 0; i < msix->msix_table_len; i++) {
2158 		if (msix->msix_table[i].mte_vector == 0)
2159 			continue;
2160 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2161 	}
2162 	free(msix->msix_table, M_DEVBUF);
2163 	msix->msix_table_len = 0;
2164 
2165 	/* Release the IRQs. */
2166 	for (i = 0; i < msix->msix_alloc; i++)
2167 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2168 		    msix->msix_vectors[i].mv_irq);
2169 	free(msix->msix_vectors, M_DEVBUF);
2170 	msix->msix_alloc = 0;
2171 	return (0);
2172 }
2173 
2174 /*
2175  * Return the max supported MSI-X messages this device supports.
2176  * Basically, assuming the MD code can alloc messages, this function
2177  * should return the maximum value that pci_alloc_msix() can return.
2178  * Thus, it is subject to the tunables, etc.
2179  */
2180 int
2181 pci_msix_count_method(device_t dev, device_t child)
2182 {
2183 	struct pci_devinfo *dinfo = device_get_ivars(child);
2184 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2185 	uint16_t ctrl;
2186 
2187 	if (pci_do_msix && msix->msix_location != 0) {
2188 		ctrl = pci_read_config(child, msix->msix_location +
2189 		    PCIR_MSI_CTRL, 2);
2190 		return (PCI_MSIX_MSGNUM(ctrl));
2191 	}
2192 	return (0);
2193 }
2194 
2195 int
2196 pci_msix_pba_bar_method(device_t dev, device_t child)
2197 {
2198 	struct pci_devinfo *dinfo = device_get_ivars(child);
2199 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2200 
2201 	if (pci_do_msix && msix->msix_location != 0)
2202 		return (msix->msix_pba_bar);
2203 	return (-1);
2204 }
2205 
2206 int
2207 pci_msix_table_bar_method(device_t dev, device_t child)
2208 {
2209 	struct pci_devinfo *dinfo = device_get_ivars(child);
2210 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2211 
2212 	if (pci_do_msix && msix->msix_location != 0)
2213 		return (msix->msix_table_bar);
2214 	return (-1);
2215 }
2216 
2217 /*
2218  * HyperTransport MSI mapping control
2219  */
2220 void
2221 pci_ht_map_msi(device_t dev, uint64_t addr)
2222 {
2223 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2224 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2225 
2226 	if (!ht->ht_msimap)
2227 		return;
2228 
2229 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2230 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2231 		/* Enable MSI -> HT mapping. */
2232 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2233 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2234 		    ht->ht_msictrl, 2);
2235 	}
2236 
2237 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2238 		/* Disable MSI -> HT mapping. */
2239 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2240 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2241 		    ht->ht_msictrl, 2);
2242 	}
2243 }
2244 
2245 int
2246 pci_get_relaxed_ordering_enabled(device_t dev)
2247 {
2248 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2249 	int cap;
2250 	uint16_t val;
2251 
2252 	cap = dinfo->cfg.pcie.pcie_location;
2253 	if (cap == 0)
2254 		return (0);
2255 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2256 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2257 	return (val != 0);
2258 }
2259 
2260 int
2261 pci_get_max_payload(device_t dev)
2262 {
2263 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2264 	int cap;
2265 	uint16_t val;
2266 
2267 	cap = dinfo->cfg.pcie.pcie_location;
2268 	if (cap == 0)
2269 		return (0);
2270 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2271 	val &= PCIEM_CTL_MAX_PAYLOAD;
2272 	val >>= 5;
2273 	return (1 << (val + 7));
2274 }
2275 
2276 int
2277 pci_get_max_read_req(device_t dev)
2278 {
2279 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2280 	int cap;
2281 	uint16_t val;
2282 
2283 	cap = dinfo->cfg.pcie.pcie_location;
2284 	if (cap == 0)
2285 		return (0);
2286 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2287 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2288 	val >>= 12;
2289 	return (1 << (val + 7));
2290 }
2291 
2292 int
2293 pci_set_max_read_req(device_t dev, int size)
2294 {
2295 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2296 	int cap;
2297 	uint16_t val;
2298 
2299 	cap = dinfo->cfg.pcie.pcie_location;
2300 	if (cap == 0)
2301 		return (0);
2302 	if (size < 128)
2303 		size = 128;
2304 	if (size > 4096)
2305 		size = 4096;
2306 	size = (1 << (fls(size) - 1));
2307 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2308 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2309 	val |= (fls(size) - 8) << 12;
2310 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2311 	return (size);
2312 }
2313 
2314 uint32_t
2315 pcie_read_config(device_t dev, int reg, int width)
2316 {
2317 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2318 	int cap;
2319 
2320 	cap = dinfo->cfg.pcie.pcie_location;
2321 	if (cap == 0) {
2322 		if (width == 2)
2323 			return (0xffff);
2324 		return (0xffffffff);
2325 	}
2326 
2327 	return (pci_read_config(dev, cap + reg, width));
2328 }
2329 
2330 void
2331 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2332 {
2333 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2334 	int cap;
2335 
2336 	cap = dinfo->cfg.pcie.pcie_location;
2337 	if (cap == 0)
2338 		return;
2339 	pci_write_config(dev, cap + reg, value, width);
2340 }
2341 
2342 /*
2343  * Adjusts a PCI-e capability register by clearing the bits in mask
2344  * and setting the bits in (value & mask).  Bits not set in mask are
2345  * not adjusted.
2346  *
2347  * Returns the old value on success or all ones on failure.
2348  */
2349 uint32_t
2350 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2351     int width)
2352 {
2353 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2354 	uint32_t old, new;
2355 	int cap;
2356 
2357 	cap = dinfo->cfg.pcie.pcie_location;
2358 	if (cap == 0) {
2359 		if (width == 2)
2360 			return (0xffff);
2361 		return (0xffffffff);
2362 	}
2363 
2364 	old = pci_read_config(dev, cap + reg, width);
2365 	new = old & ~mask;
2366 	new |= (value & mask);
2367 	pci_write_config(dev, cap + reg, new, width);
2368 	return (old);
2369 }
2370 
2371 /*
2372  * Support for MSI message signalled interrupts.
2373  */
2374 void
2375 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2376     uint16_t data)
2377 {
2378 	struct pci_devinfo *dinfo = device_get_ivars(child);
2379 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2380 
2381 	/* Write data and address values. */
2382 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2383 	    address & 0xffffffff, 4);
2384 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2385 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2386 		    address >> 32, 4);
2387 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2388 		    data, 2);
2389 	} else
2390 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2391 		    2);
2392 
2393 	/* Enable MSI in the control register. */
2394 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2395 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2396 	    msi->msi_ctrl, 2);
2397 
2398 	/* Enable MSI -> HT mapping. */
2399 	pci_ht_map_msi(child, address);
2400 }
2401 
2402 void
2403 pci_disable_msi_method(device_t dev, device_t child)
2404 {
2405 	struct pci_devinfo *dinfo = device_get_ivars(child);
2406 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2407 
2408 	/* Disable MSI -> HT mapping. */
2409 	pci_ht_map_msi(child, 0);
2410 
2411 	/* Disable MSI in the control register. */
2412 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2413 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2414 	    msi->msi_ctrl, 2);
2415 }
2416 
2417 /*
2418  * Restore MSI registers during resume.  If MSI is enabled then
2419  * restore the data and address registers in addition to the control
2420  * register.
2421  */
2422 static void
2423 pci_resume_msi(device_t dev)
2424 {
2425 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2426 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2427 	uint64_t address;
2428 	uint16_t data;
2429 
2430 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2431 		address = msi->msi_addr;
2432 		data = msi->msi_data;
2433 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2434 		    address & 0xffffffff, 4);
2435 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2436 			pci_write_config(dev, msi->msi_location +
2437 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2438 			pci_write_config(dev, msi->msi_location +
2439 			    PCIR_MSI_DATA_64BIT, data, 2);
2440 		} else
2441 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2442 			    data, 2);
2443 	}
2444 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2445 	    2);
2446 }
2447 
2448 static int
2449 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2450 {
2451 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2452 	pcicfgregs *cfg = &dinfo->cfg;
2453 	struct resource_list_entry *rle;
2454 	struct msix_table_entry *mte;
2455 	struct msix_vector *mv;
2456 	uint64_t addr;
2457 	uint32_t data;
2458 	u_int i, j;
2459 	int error;
2460 
2461 	/*
2462 	 * Handle MSI first.  We try to find this IRQ among our list
2463 	 * of MSI IRQs.  If we find it, we request updated address and
2464 	 * data registers and apply the results.
2465 	 */
2466 	if (cfg->msi.msi_alloc > 0) {
2467 		/* If we don't have any active handlers, nothing to do. */
2468 		if (cfg->msi.msi_handlers == 0)
2469 			return (0);
2470 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2471 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2472 			    i + 1);
2473 			if (rle->start == irq) {
2474 				error = PCIB_MAP_MSI(device_get_parent(bus),
2475 				    dev, irq, &addr, &data);
2476 				if (error)
2477 					return (error);
2478 				pci_disable_msi(dev);
2479 				dinfo->cfg.msi.msi_addr = addr;
2480 				dinfo->cfg.msi.msi_data = data;
2481 				pci_enable_msi(dev, addr, data);
2482 				return (0);
2483 			}
2484 		}
2485 		return (ENOENT);
2486 	}
2487 
2488 	/*
2489 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2490 	 * we request the updated mapping info.  If that works, we go
2491 	 * through all the slots that use this IRQ and update them.
2492 	 */
2493 	if (cfg->msix.msix_alloc > 0) {
2494 		bool found = false;
2495 
2496 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2497 			mv = &cfg->msix.msix_vectors[i];
2498 			if (mv->mv_irq == irq) {
2499 				error = PCIB_MAP_MSI(device_get_parent(bus),
2500 				    dev, irq, &addr, &data);
2501 				if (error)
2502 					return (error);
2503 				mv->mv_address = addr;
2504 				mv->mv_data = data;
2505 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2506 					mte = &cfg->msix.msix_table[j];
2507 					if (mte->mte_vector != i + 1)
2508 						continue;
2509 					if (mte->mte_handlers == 0)
2510 						continue;
2511 					pci_mask_msix(dev, j);
2512 					pci_enable_msix(dev, j, addr, data);
2513 					pci_unmask_msix(dev, j);
2514 				}
2515 				found = true;
2516 			}
2517 		}
2518 		return (found ? 0 : ENOENT);
2519 	}
2520 
2521 	return (ENOENT);
2522 }
2523 
2524 /*
2525  * Returns true if the specified device is blacklisted because MSI
2526  * doesn't work.
2527  */
2528 int
2529 pci_msi_device_blacklisted(device_t dev)
2530 {
2531 
2532 	if (!pci_honor_msi_blacklist)
2533 		return (0);
2534 
2535 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2536 }
2537 
2538 /*
2539  * Determine if MSI is blacklisted globally on this system.  Currently,
2540  * we just check for blacklisted chipsets as represented by the
2541  * host-PCI bridge at device 0:0:0.  In the future, it may become
2542  * necessary to check other system attributes, such as the kenv values
2543  * that give the motherboard manufacturer and model number.
2544  */
2545 static int
2546 pci_msi_blacklisted(void)
2547 {
2548 	device_t dev;
2549 
2550 	if (!pci_honor_msi_blacklist)
2551 		return (0);
2552 
2553 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2554 	if (!(pcie_chipset || pcix_chipset)) {
2555 		if (vm_guest != VM_GUEST_NO) {
2556 			/*
2557 			 * Whitelist older chipsets in virtual
2558 			 * machines known to support MSI.
2559 			 */
2560 			dev = pci_find_bsf(0, 0, 0);
2561 			if (dev != NULL)
2562 				return (!pci_has_quirk(pci_get_devid(dev),
2563 					PCI_QUIRK_ENABLE_MSI_VM));
2564 		}
2565 		return (1);
2566 	}
2567 
2568 	dev = pci_find_bsf(0, 0, 0);
2569 	if (dev != NULL)
2570 		return (pci_msi_device_blacklisted(dev));
2571 	return (0);
2572 }
2573 
2574 /*
2575  * Returns true if the specified device is blacklisted because MSI-X
2576  * doesn't work.  Note that this assumes that if MSI doesn't work,
2577  * MSI-X doesn't either.
2578  */
2579 int
2580 pci_msix_device_blacklisted(device_t dev)
2581 {
2582 
2583 	if (!pci_honor_msi_blacklist)
2584 		return (0);
2585 
2586 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2587 		return (1);
2588 
2589 	return (pci_msi_device_blacklisted(dev));
2590 }
2591 
2592 /*
2593  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2594  * is blacklisted, assume that MSI-X is as well.  Check for additional
2595  * chipsets where MSI works but MSI-X does not.
2596  */
2597 static int
2598 pci_msix_blacklisted(void)
2599 {
2600 	device_t dev;
2601 
2602 	if (!pci_honor_msi_blacklist)
2603 		return (0);
2604 
2605 	dev = pci_find_bsf(0, 0, 0);
2606 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2607 	    PCI_QUIRK_DISABLE_MSIX))
2608 		return (1);
2609 
2610 	return (pci_msi_blacklisted());
2611 }
2612 
2613 /*
2614  * Attempt to allocate *count MSI messages.  The actual number allocated is
2615  * returned in *count.  After this function returns, each message will be
2616  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2617  */
2618 int
2619 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2620 {
2621 	struct pci_devinfo *dinfo = device_get_ivars(child);
2622 	pcicfgregs *cfg = &dinfo->cfg;
2623 	struct resource_list_entry *rle;
2624 	u_int actual, i;
2625 	int error, irqs[32];
2626 	uint16_t ctrl, msgnum;
2627 
2628 	/* Don't let count == 0 get us into trouble. */
2629 	if (*count < 1)
2630 		return (EINVAL);
2631 
2632 	/* If rid 0 is allocated, then fail. */
2633 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2634 	if (rle != NULL && rle->res != NULL)
2635 		return (ENXIO);
2636 
2637 	/* Already have allocated messages? */
2638 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2639 		return (ENXIO);
2640 
2641 	/* If MSI is blacklisted for this system, fail. */
2642 	if (pci_msi_blacklisted())
2643 		return (ENXIO);
2644 
2645 	/* MSI capability present? */
2646 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2647 		return (ENODEV);
2648 
2649 	ctrl = pci_read_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, 2);
2650 	msgnum = PCI_MSI_MSGNUM(ctrl);
2651 	if (bootverbose)
2652 		device_printf(child,
2653 		    "attempting to allocate %d MSI vectors (%u supported)\n",
2654 		    *count, msgnum);
2655 
2656 	/* Don't ask for more than the device supports. */
2657 	actual = min(*count, msgnum);
2658 
2659 	/* Don't ask for more than 32 messages. */
2660 	actual = min(actual, 32);
2661 
2662 	/* MSI requires power of 2 number of messages. */
2663 	if (!powerof2(actual))
2664 		return (EINVAL);
2665 
2666 	for (;;) {
2667 		/* Try to allocate N messages. */
2668 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2669 		    actual, irqs);
2670 		if (error == 0)
2671 			break;
2672 		if (actual == 1)
2673 			return (error);
2674 
2675 		/* Try N / 2. */
2676 		actual >>= 1;
2677 	}
2678 
2679 	/*
2680 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2681 	 * resources in the irqs[] array, so add new resources
2682 	 * starting at rid 1.
2683 	 */
2684 	for (i = 0; i < actual; i++)
2685 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2686 		    irqs[i], irqs[i], 1);
2687 
2688 	if (bootverbose) {
2689 		if (actual == 1)
2690 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2691 		else {
2692 			bool run;
2693 
2694 			/*
2695 			 * Be fancy and try to print contiguous runs
2696 			 * of IRQ values as ranges.  'run' is true if
2697 			 * we are in a range.
2698 			 */
2699 			device_printf(child, "using IRQs %d", irqs[0]);
2700 			run = false;
2701 			for (i = 1; i < actual; i++) {
2702 				/* Still in a run? */
2703 				if (irqs[i] == irqs[i - 1] + 1) {
2704 					run = true;
2705 					continue;
2706 				}
2707 
2708 				/* Finish previous range. */
2709 				if (run) {
2710 					printf("-%d", irqs[i - 1]);
2711 					run = false;
2712 				}
2713 
2714 				/* Start new range. */
2715 				printf(",%d", irqs[i]);
2716 			}
2717 
2718 			/* Unfinished range? */
2719 			if (run)
2720 				printf("-%d", irqs[actual - 1]);
2721 			printf(" for MSI\n");
2722 		}
2723 	}
2724 
2725 	/* Update control register with actual count. */
2726 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2727 	ctrl |= (ffs(actual) - 1) << 4;
2728 	cfg->msi.msi_ctrl = ctrl;
2729 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2730 
2731 	/* Update counts of alloc'd messages. */
2732 	cfg->msi.msi_alloc = actual;
2733 	cfg->msi.msi_handlers = 0;
2734 	*count = actual;
2735 	return (0);
2736 }
2737 
2738 /* Release the MSI messages associated with this device. */
2739 int
2740 pci_release_msi_method(device_t dev, device_t child)
2741 {
2742 	struct pci_devinfo *dinfo = device_get_ivars(child);
2743 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2744 	struct resource_list_entry *rle;
2745 	u_int i, irqs[32];
2746 	int error;
2747 
2748 	/* Try MSI-X first. */
2749 	error = pci_release_msix(dev, child);
2750 	if (error != ENODEV)
2751 		return (error);
2752 
2753 	/* Do we have any messages to release? */
2754 	if (msi->msi_alloc == 0)
2755 		return (ENODEV);
2756 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2757 
2758 	/* Make sure none of the resources are allocated. */
2759 	if (msi->msi_handlers > 0)
2760 		return (EBUSY);
2761 	for (i = 0; i < msi->msi_alloc; i++) {
2762 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2763 		KASSERT(rle != NULL, ("missing MSI resource"));
2764 		if (rle->res != NULL)
2765 			return (EBUSY);
2766 		irqs[i] = rle->start;
2767 	}
2768 
2769 	/* Update control register with 0 count. */
2770 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2771 	    ("%s: MSI still enabled", __func__));
2772 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2773 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2774 	    msi->msi_ctrl, 2);
2775 
2776 	/* Release the messages. */
2777 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2778 	for (i = 0; i < msi->msi_alloc; i++)
2779 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2780 
2781 	/* Update alloc count. */
2782 	msi->msi_alloc = 0;
2783 	msi->msi_addr = 0;
2784 	msi->msi_data = 0;
2785 	return (0);
2786 }
2787 
2788 /*
2789  * Return the max supported MSI messages this device supports.
2790  * Basically, assuming the MD code can alloc messages, this function
2791  * should return the maximum value that pci_alloc_msi() can return.
2792  * Thus, it is subject to the tunables, etc.
2793  */
2794 int
2795 pci_msi_count_method(device_t dev, device_t child)
2796 {
2797 	struct pci_devinfo *dinfo = device_get_ivars(child);
2798 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2799 	uint16_t ctrl;
2800 
2801 	if (pci_do_msi && msi->msi_location != 0) {
2802 		ctrl = pci_read_config(child, msi->msi_location + PCIR_MSI_CTRL,
2803 		    2);
2804 		return (PCI_MSI_MSGNUM(ctrl));
2805 	}
2806 	return (0);
2807 }
2808 
2809 /* free pcicfgregs structure and all depending data structures */
2810 
2811 int
2812 pci_freecfg(struct pci_devinfo *dinfo)
2813 {
2814 	struct devlist *devlist_head;
2815 	struct pci_map *pm, *next;
2816 
2817 	devlist_head = &pci_devq;
2818 
2819 	if (dinfo->cfg.vpd.vpd_reg)
2820 		vpd_free(&dinfo->cfg.vpd);
2821 
2822 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2823 		free(pm, M_DEVBUF);
2824 	}
2825 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2826 	free(dinfo, M_DEVBUF);
2827 
2828 	/* increment the generation count */
2829 	pci_generation++;
2830 
2831 	/* we're losing one device */
2832 	pci_numdevs--;
2833 	return (0);
2834 }
2835 
2836 /*
2837  * PCI power manangement
2838  */
2839 int
2840 pci_set_powerstate_method(device_t dev, device_t child, int state)
2841 {
2842 	struct pci_devinfo *dinfo = device_get_ivars(child);
2843 	pcicfgregs *cfg = &dinfo->cfg;
2844 	uint16_t status;
2845 	int oldstate, highest, delay;
2846 
2847 	if (cfg->pp.pp_location == 0)
2848 		return (EOPNOTSUPP);
2849 
2850 	/*
2851 	 * Optimize a no state change request away.  While it would be OK to
2852 	 * write to the hardware in theory, some devices have shown odd
2853 	 * behavior when going from D3 -> D3.
2854 	 */
2855 	oldstate = pci_get_powerstate(child);
2856 	if (oldstate == state)
2857 		return (0);
2858 
2859 	/*
2860 	 * The PCI power management specification states that after a state
2861 	 * transition between PCI power states, system software must
2862 	 * guarantee a minimal delay before the function accesses the device.
2863 	 * Compute the worst case delay that we need to guarantee before we
2864 	 * access the device.  Many devices will be responsive much more
2865 	 * quickly than this delay, but there are some that don't respond
2866 	 * instantly to state changes.  Transitions to/from D3 state require
2867 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2868 	 * is done below with DELAY rather than a sleeper function because
2869 	 * this function can be called from contexts where we cannot sleep.
2870 	 */
2871 	highest = (oldstate > state) ? oldstate : state;
2872 	if (highest == PCI_POWERSTATE_D3)
2873 	    delay = 10000;
2874 	else if (highest == PCI_POWERSTATE_D2)
2875 	    delay = 200;
2876 	else
2877 	    delay = 0;
2878 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2879 	    PCIR_POWER_STATUS, 2) & ~PCIM_PSTAT_DMASK;
2880 	switch (state) {
2881 	case PCI_POWERSTATE_D0:
2882 		status |= PCIM_PSTAT_D0;
2883 		break;
2884 	case PCI_POWERSTATE_D1:
2885 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2886 			return (EOPNOTSUPP);
2887 		status |= PCIM_PSTAT_D1;
2888 		break;
2889 	case PCI_POWERSTATE_D2:
2890 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2891 			return (EOPNOTSUPP);
2892 		status |= PCIM_PSTAT_D2;
2893 		break;
2894 	case PCI_POWERSTATE_D3:
2895 		status |= PCIM_PSTAT_D3;
2896 		break;
2897 	default:
2898 		return (EINVAL);
2899 	}
2900 
2901 	if (bootverbose)
2902 		pci_printf(cfg, "Transition from %s to %s\n",
2903 		    acpi_d_state_to_str(oldstate), acpi_d_state_to_str(state));
2904 
2905 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_location + PCIR_POWER_STATUS,
2906 	    status, 2);
2907 	if (delay)
2908 		DELAY(delay);
2909 	return (0);
2910 }
2911 
2912 int
2913 pci_get_powerstate_method(device_t dev, device_t child)
2914 {
2915 	struct pci_devinfo *dinfo = device_get_ivars(child);
2916 	pcicfgregs *cfg = &dinfo->cfg;
2917 	uint16_t status;
2918 	int result;
2919 
2920 	if (cfg->pp.pp_location != 0) {
2921 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2922 		    PCIR_POWER_STATUS, 2);
2923 		switch (status & PCIM_PSTAT_DMASK) {
2924 		case PCIM_PSTAT_D0:
2925 			result = PCI_POWERSTATE_D0;
2926 			break;
2927 		case PCIM_PSTAT_D1:
2928 			result = PCI_POWERSTATE_D1;
2929 			break;
2930 		case PCIM_PSTAT_D2:
2931 			result = PCI_POWERSTATE_D2;
2932 			break;
2933 		case PCIM_PSTAT_D3:
2934 			result = PCI_POWERSTATE_D3;
2935 			break;
2936 		default:
2937 			result = PCI_POWERSTATE_UNKNOWN;
2938 			break;
2939 		}
2940 	} else {
2941 		/* No support, device is always at D0 */
2942 		result = PCI_POWERSTATE_D0;
2943 	}
2944 	return (result);
2945 }
2946 
2947 /* Clear any active PME# and disable PME# generation. */
2948 void
2949 pci_clear_pme(device_t dev)
2950 {
2951 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2952 	pcicfgregs *cfg = &dinfo->cfg;
2953 	uint16_t status;
2954 
2955 	if (cfg->pp.pp_location != 0) {
2956 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2957 		    PCIR_POWER_STATUS, 2);
2958 		status &= ~PCIM_PSTAT_PMEENABLE;
2959 		status |= PCIM_PSTAT_PME;
2960 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2961 		    PCIR_POWER_STATUS, status, 2);
2962 	}
2963 }
2964 
2965 /* Clear any active PME# and enable PME# generation. */
2966 void
2967 pci_enable_pme(device_t dev)
2968 {
2969 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2970 	pcicfgregs *cfg = &dinfo->cfg;
2971 	uint16_t status;
2972 
2973 	if (cfg->pp.pp_location != 0) {
2974 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2975 		    PCIR_POWER_STATUS, 2);
2976 		status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2977 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2978 		    PCIR_POWER_STATUS, status, 2);
2979 	}
2980 }
2981 
2982 bool
2983 pci_has_pm(device_t dev)
2984 {
2985 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2986 	pcicfgregs *cfg = &dinfo->cfg;
2987 
2988 	return (cfg->pp.pp_location != 0);
2989 }
2990 
2991 /*
2992  * Some convenience functions for PCI device drivers.
2993  */
2994 
2995 static __inline void
2996 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2997 {
2998 	uint16_t	command;
2999 
3000 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3001 	command |= bit;
3002 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3003 }
3004 
3005 static __inline void
3006 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
3007 {
3008 	uint16_t	command;
3009 
3010 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3011 	command &= ~bit;
3012 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3013 }
3014 
3015 int
3016 pci_enable_busmaster_method(device_t dev, device_t child)
3017 {
3018 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3019 	return (0);
3020 }
3021 
3022 int
3023 pci_disable_busmaster_method(device_t dev, device_t child)
3024 {
3025 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3026 	return (0);
3027 }
3028 
3029 int
3030 pci_enable_io_method(device_t dev, device_t child, int space)
3031 {
3032 	uint16_t bit;
3033 
3034 	switch(space) {
3035 	case SYS_RES_IOPORT:
3036 		bit = PCIM_CMD_PORTEN;
3037 		break;
3038 	case SYS_RES_MEMORY:
3039 		bit = PCIM_CMD_MEMEN;
3040 		break;
3041 	default:
3042 		return (EINVAL);
3043 	}
3044 	pci_set_command_bit(dev, child, bit);
3045 	return (0);
3046 }
3047 
3048 int
3049 pci_disable_io_method(device_t dev, device_t child, int space)
3050 {
3051 	uint16_t bit;
3052 
3053 	switch(space) {
3054 	case SYS_RES_IOPORT:
3055 		bit = PCIM_CMD_PORTEN;
3056 		break;
3057 	case SYS_RES_MEMORY:
3058 		bit = PCIM_CMD_MEMEN;
3059 		break;
3060 	default:
3061 		return (EINVAL);
3062 	}
3063 	pci_clear_command_bit(dev, child, bit);
3064 	return (0);
3065 }
3066 
3067 /*
3068  * New style pci driver.  Parent device is either a pci-host-bridge or a
3069  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
3070  */
3071 
3072 void
3073 pci_print_verbose(struct pci_devinfo *dinfo)
3074 {
3075 
3076 	if (bootverbose) {
3077 		pcicfgregs *cfg = &dinfo->cfg;
3078 
3079 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
3080 		    cfg->vendor, cfg->device, cfg->revid);
3081 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
3082 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
3083 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
3084 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
3085 		    cfg->mfdev);
3086 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
3087 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
3088 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
3089 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3090 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3091 		if (cfg->intpin > 0)
3092 			printf("\tintpin=%c, irq=%d\n",
3093 			    cfg->intpin +'a' -1, cfg->intline);
3094 		if (cfg->pp.pp_location) {
3095 			uint16_t status;
3096 
3097 			status = pci_read_config(cfg->dev, cfg->pp.pp_location +
3098 			    PCIR_POWER_STATUS, 2);
3099 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3100 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3101 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3102 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3103 			    status & PCIM_PSTAT_DMASK);
3104 		}
3105 		if (cfg->msi.msi_location) {
3106 			uint16_t ctrl, msgnum;
3107 
3108 			ctrl = cfg->msi.msi_ctrl;
3109 			msgnum = PCI_MSI_MSGNUM(ctrl);
3110 			printf("\tMSI supports %d message%s%s%s\n",
3111 			    msgnum, (msgnum == 1) ? "" : "s",
3112 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3113 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3114 		}
3115 		if (cfg->msix.msix_location) {
3116 			uint16_t msgnum;
3117 
3118 			msgnum = PCI_MSIX_MSGNUM(cfg->msix.msix_ctrl);
3119 			printf("\tMSI-X supports %d message%s ",
3120 			    msgnum, (msgnum == 1) ? "" : "s");
3121 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3122 				printf("in map 0x%x\n",
3123 				    cfg->msix.msix_table_bar);
3124 			else
3125 				printf("in maps 0x%x and 0x%x\n",
3126 				    cfg->msix.msix_table_bar,
3127 				    cfg->msix.msix_pba_bar);
3128 		}
3129 	}
3130 }
3131 
3132 static int
3133 pci_porten(device_t dev)
3134 {
3135 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3136 }
3137 
3138 static int
3139 pci_memen(device_t dev)
3140 {
3141 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3142 }
3143 
3144 void
3145 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3146     int *bar64)
3147 {
3148 	struct pci_devinfo *dinfo;
3149 	pci_addr_t map, testval;
3150 	int ln2range;
3151 	uint16_t cmd;
3152 
3153 	/*
3154 	 * The device ROM BAR is special.  It is always a 32-bit
3155 	 * memory BAR.  Bit 0 is special and should not be set when
3156 	 * sizing the BAR.
3157 	 */
3158 	dinfo = device_get_ivars(dev);
3159 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3160 		map = pci_read_config(dev, reg, 4);
3161 		pci_write_config(dev, reg, 0xfffffffe, 4);
3162 		testval = pci_read_config(dev, reg, 4);
3163 		pci_write_config(dev, reg, map, 4);
3164 		*mapp = map;
3165 		*testvalp = testval;
3166 		if (bar64 != NULL)
3167 			*bar64 = 0;
3168 		return;
3169 	}
3170 
3171 	map = pci_read_config(dev, reg, 4);
3172 	ln2range = pci_maprange(map);
3173 	if (ln2range == 64)
3174 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3175 
3176 	/*
3177 	 * Disable decoding via the command register before
3178 	 * determining the BAR's length since we will be placing it in
3179 	 * a weird state.
3180 	 */
3181 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3182 	pci_write_config(dev, PCIR_COMMAND,
3183 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3184 
3185 	/*
3186 	 * Determine the BAR's length by writing all 1's.  The bottom
3187 	 * log_2(size) bits of the BAR will stick as 0 when we read
3188 	 * the value back.
3189 	 *
3190 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3191 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3192 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3193 	 *
3194 	 * Writes to both registers must be performed before attempting to
3195 	 * read back the size value.
3196 	 */
3197 	testval = 0;
3198 	pci_write_config(dev, reg, 0xffffffff, 4);
3199 	if (ln2range == 64) {
3200 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3201 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3202 	}
3203 	testval |= pci_read_config(dev, reg, 4);
3204 
3205 	/*
3206 	 * Restore the original value of the BAR.  We may have reprogrammed
3207 	 * the BAR of the low-level console device and when booting verbose,
3208 	 * we need the console device addressable.
3209 	 */
3210 	pci_write_config(dev, reg, map, 4);
3211 	if (ln2range == 64)
3212 		pci_write_config(dev, reg + 4, map >> 32, 4);
3213 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3214 
3215 	*mapp = map;
3216 	*testvalp = testval;
3217 	if (bar64 != NULL)
3218 		*bar64 = (ln2range == 64);
3219 }
3220 
3221 static void
3222 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3223 {
3224 	struct pci_devinfo *dinfo;
3225 	int ln2range;
3226 
3227 	/* The device ROM BAR is always a 32-bit memory BAR. */
3228 	dinfo = device_get_ivars(dev);
3229 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3230 		ln2range = 32;
3231 	else
3232 		ln2range = pci_maprange(pm->pm_value);
3233 	pci_write_config(dev, pm->pm_reg, base, 4);
3234 	if (ln2range == 64)
3235 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3236 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3237 	if (ln2range == 64)
3238 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3239 		    pm->pm_reg + 4, 4) << 32;
3240 }
3241 
3242 struct pci_map *
3243 pci_find_bar(device_t dev, int reg)
3244 {
3245 	struct pci_devinfo *dinfo;
3246 	struct pci_map *pm;
3247 
3248 	dinfo = device_get_ivars(dev);
3249 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3250 		if (pm->pm_reg == reg)
3251 			return (pm);
3252 	}
3253 	return (NULL);
3254 }
3255 
3256 struct pci_map *
3257 pci_first_bar(device_t dev)
3258 {
3259 	struct pci_devinfo *dinfo;
3260 
3261 	dinfo = device_get_ivars(dev);
3262 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3263 }
3264 
3265 struct pci_map *
3266 pci_next_bar(struct pci_map *pm)
3267 {
3268 	return (STAILQ_NEXT(pm, pm_link));
3269 }
3270 
3271 int
3272 pci_bar_enabled(device_t dev, struct pci_map *pm)
3273 {
3274 	struct pci_devinfo *dinfo;
3275 	uint16_t cmd;
3276 
3277 	dinfo = device_get_ivars(dev);
3278 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3279 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3280 		return (0);
3281 #ifdef PCI_IOV
3282 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3283 		struct pcicfg_iov *iov;
3284 
3285 		iov = dinfo->cfg.iov;
3286 		cmd = pci_read_config(iov->iov_pf,
3287 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3288 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3289 	}
3290 #endif
3291 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3292 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3293 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3294 	else
3295 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3296 }
3297 
3298 struct pci_map *
3299 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3300 {
3301 	struct pci_devinfo *dinfo;
3302 	struct pci_map *pm, *prev;
3303 
3304 	dinfo = device_get_ivars(dev);
3305 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3306 	pm->pm_reg = reg;
3307 	pm->pm_value = value;
3308 	pm->pm_size = size;
3309 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3310 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3311 		    reg));
3312 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3313 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3314 			break;
3315 	}
3316 	if (prev != NULL)
3317 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3318 	else
3319 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3320 	return (pm);
3321 }
3322 
3323 static void
3324 pci_restore_bars(device_t dev)
3325 {
3326 	struct pci_devinfo *dinfo;
3327 	struct pci_map *pm;
3328 	int ln2range;
3329 
3330 	dinfo = device_get_ivars(dev);
3331 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3332 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3333 			ln2range = 32;
3334 		else
3335 			ln2range = pci_maprange(pm->pm_value);
3336 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3337 		if (ln2range == 64)
3338 			pci_write_config(dev, pm->pm_reg + 4,
3339 			    pm->pm_value >> 32, 4);
3340 	}
3341 }
3342 
3343 /*
3344  * Add a resource based on a pci map register. Return 1 if the map
3345  * register is a 32bit map register or 2 if it is a 64bit register.
3346  */
3347 static int
3348 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3349     int force, int prefetch)
3350 {
3351 	struct pci_map *pm;
3352 	pci_addr_t base, map, testval;
3353 	pci_addr_t start, end, count;
3354 	int barlen, basezero, flags, maprange, mapsize, type;
3355 	uint16_t cmd;
3356 	struct resource *res;
3357 
3358 	/*
3359 	 * The BAR may already exist if the device is a CardBus card
3360 	 * whose CIS is stored in this BAR.
3361 	 */
3362 	pm = pci_find_bar(dev, reg);
3363 	if (pm != NULL) {
3364 		maprange = pci_maprange(pm->pm_value);
3365 		barlen = maprange == 64 ? 2 : 1;
3366 		return (barlen);
3367 	}
3368 
3369 	pci_read_bar(dev, reg, &map, &testval, NULL);
3370 	if (PCI_BAR_MEM(map)) {
3371 		type = SYS_RES_MEMORY;
3372 		if (map & PCIM_BAR_MEM_PREFETCH)
3373 			prefetch = 1;
3374 	} else
3375 		type = SYS_RES_IOPORT;
3376 	mapsize = pci_mapsize(testval);
3377 	base = pci_mapbase(map);
3378 #ifdef __PCI_BAR_ZERO_VALID
3379 	basezero = 0;
3380 #else
3381 	basezero = base == 0;
3382 #endif
3383 	maprange = pci_maprange(map);
3384 	barlen = maprange == 64 ? 2 : 1;
3385 
3386 	/*
3387 	 * For I/O registers, if bottom bit is set, and the next bit up
3388 	 * isn't clear, we know we have a BAR that doesn't conform to the
3389 	 * spec, so ignore it.  Also, sanity check the size of the data
3390 	 * areas to the type of memory involved.  Memory must be at least
3391 	 * 16 bytes in size, while I/O ranges must be at least 4.
3392 	 */
3393 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3394 		return (barlen);
3395 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3396 	    (type == SYS_RES_IOPORT && mapsize < 2))
3397 		return (barlen);
3398 
3399 	/* Save a record of this BAR. */
3400 	pm = pci_add_bar(dev, reg, map, mapsize);
3401 	if (bootverbose) {
3402 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3403 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3404 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3405 			printf(", port disabled\n");
3406 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3407 			printf(", memory disabled\n");
3408 		else
3409 			printf(", enabled\n");
3410 	}
3411 
3412 	/*
3413 	 * If base is 0, then we have problems if this architecture does
3414 	 * not allow that.  It is best to ignore such entries for the
3415 	 * moment.  These will be allocated later if the driver specifically
3416 	 * requests them.  However, some removable buses look better when
3417 	 * all resources are allocated, so allow '0' to be overridden.
3418 	 *
3419 	 * Similarly treat maps whose values is the same as the test value
3420 	 * read back.  These maps have had all f's written to them by the
3421 	 * BIOS in an attempt to disable the resources.
3422 	 */
3423 	if (!force && (basezero || map == testval))
3424 		return (barlen);
3425 	if ((u_long)base != base) {
3426 		device_printf(bus,
3427 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3428 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3429 		    pci_get_function(dev), reg);
3430 		return (barlen);
3431 	}
3432 
3433 	/*
3434 	 * This code theoretically does the right thing, but has
3435 	 * undesirable side effects in some cases where peripherals
3436 	 * respond oddly to having these bits enabled.  Let the user
3437 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3438 	 * default).
3439 	 */
3440 	if (pci_enable_io_modes) {
3441 		/* Turn on resources that have been left off by a lazy BIOS */
3442 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3443 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3444 			cmd |= PCIM_CMD_PORTEN;
3445 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3446 		}
3447 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3448 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3449 			cmd |= PCIM_CMD_MEMEN;
3450 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3451 		}
3452 	} else {
3453 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3454 			return (barlen);
3455 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3456 			return (barlen);
3457 	}
3458 
3459 	count = (pci_addr_t)1 << mapsize;
3460 	flags = RF_ALIGNMENT_LOG2(mapsize);
3461 	if (prefetch)
3462 		flags |= RF_PREFETCHABLE;
3463 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3464 		start = 0;	/* Let the parent decide. */
3465 		end = ~0;
3466 	} else {
3467 		start = base;
3468 		end = base + count - 1;
3469 	}
3470 	resource_list_add(rl, type, reg, start, end, count);
3471 
3472 	/*
3473 	 * Try to allocate the resource for this BAR from our parent
3474 	 * so that this resource range is already reserved.  The
3475 	 * driver for this device will later inherit this resource in
3476 	 * pci_alloc_resource().
3477 	 */
3478 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3479 	    flags);
3480 	if ((pci_do_realloc_bars
3481 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3482 	    && res == NULL && (start != 0 || end != ~0)) {
3483 		/*
3484 		 * If the allocation fails, try to allocate a resource for
3485 		 * this BAR using any available range.  The firmware felt
3486 		 * it was important enough to assign a resource, so don't
3487 		 * disable decoding if we can help it.
3488 		 */
3489 		resource_list_delete(rl, type, reg);
3490 		resource_list_add(rl, type, reg, 0, ~0, count);
3491 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3492 		    count, flags);
3493 	}
3494 	if (res == NULL) {
3495 		/*
3496 		 * If the allocation fails, delete the resource list entry
3497 		 * and disable decoding for this device.
3498 		 *
3499 		 * If the driver requests this resource in the future,
3500 		 * pci_reserve_map() will try to allocate a fresh
3501 		 * resource range.
3502 		 */
3503 		resource_list_delete(rl, type, reg);
3504 		pci_disable_io(dev, type);
3505 		if (bootverbose)
3506 			device_printf(bus,
3507 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3508 			    pci_get_domain(dev), pci_get_bus(dev),
3509 			    pci_get_slot(dev), pci_get_function(dev), reg);
3510 	} else {
3511 		start = rman_get_start(res);
3512 		pci_write_bar(dev, pm, start);
3513 	}
3514 	return (barlen);
3515 }
3516 
3517 /*
3518  * For ATA devices we need to decide early what addressing mode to use.
3519  * Legacy demands that the primary and secondary ATA ports sits on the
3520  * same addresses that old ISA hardware did. This dictates that we use
3521  * those addresses and ignore the BAR's if we cannot set PCI native
3522  * addressing mode.
3523  */
3524 static void
3525 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3526     uint32_t prefetchmask)
3527 {
3528 	int rid, type, progif;
3529 #if 0
3530 	/* if this device supports PCI native addressing use it */
3531 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3532 	if ((progif & 0x8a) == 0x8a) {
3533 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3534 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3535 			printf("Trying ATA native PCI addressing mode\n");
3536 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3537 		}
3538 	}
3539 #endif
3540 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3541 	type = SYS_RES_IOPORT;
3542 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3543 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3544 		    prefetchmask & (1 << 0));
3545 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3546 		    prefetchmask & (1 << 1));
3547 	} else {
3548 		rid = PCIR_BAR(0);
3549 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3550 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3551 		    0x1f7, 8, 0);
3552 		rid = PCIR_BAR(1);
3553 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3554 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3555 		    0x3f6, 1, 0);
3556 	}
3557 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3558 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3559 		    prefetchmask & (1 << 2));
3560 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3561 		    prefetchmask & (1 << 3));
3562 	} else {
3563 		rid = PCIR_BAR(2);
3564 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3565 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3566 		    0x177, 8, 0);
3567 		rid = PCIR_BAR(3);
3568 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3569 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3570 		    0x376, 1, 0);
3571 	}
3572 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3573 	    prefetchmask & (1 << 4));
3574 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3575 	    prefetchmask & (1 << 5));
3576 }
3577 
3578 static void
3579 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3580 {
3581 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3582 	pcicfgregs *cfg = &dinfo->cfg;
3583 	char tunable_name[64];
3584 	int irq;
3585 
3586 	/* Has to have an intpin to have an interrupt. */
3587 	if (cfg->intpin == 0)
3588 		return;
3589 
3590 	/* Let the user override the IRQ with a tunable. */
3591 	irq = PCI_INVALID_IRQ;
3592 	snprintf(tunable_name, sizeof(tunable_name),
3593 	    "hw.pci%d.%d.%d.INT%c.irq",
3594 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3595 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3596 		irq = PCI_INVALID_IRQ;
3597 
3598 	/*
3599 	 * If we didn't get an IRQ via the tunable, then we either use the
3600 	 * IRQ value in the intline register or we ask the bus to route an
3601 	 * interrupt for us.  If force_route is true, then we only use the
3602 	 * value in the intline register if the bus was unable to assign an
3603 	 * IRQ.
3604 	 */
3605 	if (!PCI_INTERRUPT_VALID(irq)) {
3606 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3607 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3608 		if (!PCI_INTERRUPT_VALID(irq))
3609 			irq = cfg->intline;
3610 	}
3611 
3612 	/* If after all that we don't have an IRQ, just bail. */
3613 	if (!PCI_INTERRUPT_VALID(irq))
3614 		return;
3615 
3616 	/* Update the config register if it changed. */
3617 	if (irq != cfg->intline) {
3618 		cfg->intline = irq;
3619 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3620 	}
3621 
3622 	/* Add this IRQ as rid 0 interrupt resource. */
3623 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3624 }
3625 
3626 /* Perform early OHCI takeover from SMM. */
3627 static void
3628 ohci_early_takeover(device_t self)
3629 {
3630 	struct resource *res;
3631 	uint32_t ctl;
3632 	int rid;
3633 	int i;
3634 
3635 	rid = PCIR_BAR(0);
3636 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3637 	if (res == NULL)
3638 		return;
3639 
3640 	ctl = bus_read_4(res, OHCI_CONTROL);
3641 	if (ctl & OHCI_IR) {
3642 		if (bootverbose)
3643 			printf("ohci early: "
3644 			    "SMM active, request owner change\n");
3645 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3646 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3647 			DELAY(1000);
3648 			ctl = bus_read_4(res, OHCI_CONTROL);
3649 		}
3650 		if (ctl & OHCI_IR) {
3651 			if (bootverbose)
3652 				printf("ohci early: "
3653 				    "SMM does not respond, resetting\n");
3654 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3655 		}
3656 		/* Disable interrupts */
3657 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3658 	}
3659 
3660 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3661 }
3662 
3663 /* Perform early UHCI takeover from SMM. */
3664 static void
3665 uhci_early_takeover(device_t self)
3666 {
3667 	struct resource *res;
3668 	int rid;
3669 
3670 	/*
3671 	 * Set the PIRQD enable bit and switch off all the others. We don't
3672 	 * want legacy support to interfere with us XXX Does this also mean
3673 	 * that the BIOS won't touch the keyboard anymore if it is connected
3674 	 * to the ports of the root hub?
3675 	 */
3676 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3677 
3678 	/* Disable interrupts */
3679 	rid = PCI_UHCI_BASE_REG;
3680 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3681 	if (res != NULL) {
3682 		bus_write_2(res, UHCI_INTR, 0);
3683 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3684 	}
3685 }
3686 
3687 /* Perform early EHCI takeover from SMM. */
3688 static void
3689 ehci_early_takeover(device_t self)
3690 {
3691 	struct resource *res;
3692 	uint32_t cparams;
3693 	uint32_t eec;
3694 	uint8_t eecp;
3695 	uint8_t bios_sem;
3696 	uint8_t offs;
3697 	int rid;
3698 	int i;
3699 
3700 	rid = PCIR_BAR(0);
3701 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3702 	if (res == NULL)
3703 		return;
3704 
3705 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3706 
3707 	/* Synchronise with the BIOS if it owns the controller. */
3708 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3709 	    eecp = EHCI_EECP_NEXT(eec)) {
3710 		eec = pci_read_config(self, eecp, 4);
3711 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3712 			continue;
3713 		}
3714 		bios_sem = pci_read_config(self, eecp +
3715 		    EHCI_LEGSUP_BIOS_SEM, 1);
3716 		if (bios_sem == 0) {
3717 			continue;
3718 		}
3719 		if (bootverbose)
3720 			printf("ehci early: "
3721 			    "SMM active, request owner change\n");
3722 
3723 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3724 
3725 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3726 			DELAY(1000);
3727 			bios_sem = pci_read_config(self, eecp +
3728 			    EHCI_LEGSUP_BIOS_SEM, 1);
3729 		}
3730 
3731 		if (bios_sem != 0) {
3732 			if (bootverbose)
3733 				printf("ehci early: "
3734 				    "SMM does not respond\n");
3735 		}
3736 		/* Disable interrupts */
3737 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3738 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3739 	}
3740 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3741 }
3742 
3743 /* Perform early XHCI takeover from SMM. */
3744 static void
3745 xhci_early_takeover(device_t self)
3746 {
3747 	struct resource *res;
3748 	uint32_t cparams;
3749 	uint32_t eec;
3750 	uint8_t eecp;
3751 	uint8_t bios_sem;
3752 	uint8_t offs;
3753 	int rid;
3754 	int i;
3755 
3756 	rid = PCIR_BAR(0);
3757 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3758 	if (res == NULL)
3759 		return;
3760 
3761 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3762 
3763 	eec = -1;
3764 
3765 	/* Synchronise with the BIOS if it owns the controller. */
3766 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3767 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3768 		eec = bus_read_4(res, eecp);
3769 
3770 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3771 			continue;
3772 
3773 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3774 		if (bios_sem == 0)
3775 			continue;
3776 
3777 		if (bootverbose)
3778 			printf("xhci early: "
3779 			    "SMM active, request owner change\n");
3780 
3781 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3782 
3783 		/* wait a maximum of 5 second */
3784 
3785 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3786 			DELAY(1000);
3787 			bios_sem = bus_read_1(res, eecp +
3788 			    XHCI_XECP_BIOS_SEM);
3789 		}
3790 
3791 		if (bios_sem != 0) {
3792 			if (bootverbose)
3793 				printf("xhci early: "
3794 				    "SMM does not respond\n");
3795 		}
3796 
3797 		/* Disable interrupts */
3798 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3799 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3800 		bus_read_4(res, offs + XHCI_USBSTS);
3801 	}
3802 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3803 }
3804 
3805 static void
3806 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3807     struct resource_list *rl)
3808 {
3809 	struct resource *res;
3810 	char *cp;
3811 	rman_res_t start, end, count;
3812 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3813 
3814 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3815 	case PCIM_HDRTYPE_BRIDGE:
3816 		sec_reg = PCIR_SECBUS_1;
3817 		sub_reg = PCIR_SUBBUS_1;
3818 		break;
3819 	case PCIM_HDRTYPE_CARDBUS:
3820 		sec_reg = PCIR_SECBUS_2;
3821 		sub_reg = PCIR_SUBBUS_2;
3822 		break;
3823 	default:
3824 		return;
3825 	}
3826 
3827 	/*
3828 	 * If the existing bus range is valid, attempt to reserve it
3829 	 * from our parent.  If this fails for any reason, clear the
3830 	 * secbus and subbus registers.
3831 	 *
3832 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3833 	 * This would at least preserve the existing sec_bus if it is
3834 	 * valid.
3835 	 */
3836 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3837 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3838 
3839 	/* Quirk handling. */
3840 	switch (pci_get_devid(dev)) {
3841 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3842 		sup_bus = pci_read_config(dev, 0x41, 1);
3843 		if (sup_bus != 0xff) {
3844 			sec_bus = sup_bus + 1;
3845 			sub_bus = sup_bus + 1;
3846 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3847 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3848 		}
3849 		break;
3850 
3851 	case 0x00dd10de:
3852 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3853 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3854 			break;
3855 		if (strncmp(cp, "Compal", 6) != 0) {
3856 			freeenv(cp);
3857 			break;
3858 		}
3859 		freeenv(cp);
3860 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3861 			break;
3862 		if (strncmp(cp, "08A0", 4) != 0) {
3863 			freeenv(cp);
3864 			break;
3865 		}
3866 		freeenv(cp);
3867 		if (sub_bus < 0xa) {
3868 			sub_bus = 0xa;
3869 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3870 		}
3871 		break;
3872 	}
3873 
3874 	if (bootverbose)
3875 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3876 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3877 		start = sec_bus;
3878 		end = sub_bus;
3879 		count = end - start + 1;
3880 
3881 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3882 
3883 		/*
3884 		 * If requested, clear secondary bus registers in
3885 		 * bridge devices to force a complete renumbering
3886 		 * rather than reserving the existing range.  However,
3887 		 * preserve the existing size.
3888 		 */
3889 		if (pci_clear_buses)
3890 			goto clear;
3891 
3892 		rid = 0;
3893 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3894 		    start, end, count, 0);
3895 		if (res != NULL)
3896 			return;
3897 
3898 		if (bootverbose)
3899 			device_printf(bus,
3900 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3901 			    pci_get_domain(dev), pci_get_bus(dev),
3902 			    pci_get_slot(dev), pci_get_function(dev));
3903 	}
3904 
3905 clear:
3906 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3907 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3908 }
3909 
3910 static struct resource *
3911 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3912     rman_res_t end, rman_res_t count, u_int flags)
3913 {
3914 	struct pci_devinfo *dinfo;
3915 	pcicfgregs *cfg;
3916 	struct resource_list *rl;
3917 	struct resource *res;
3918 	int sec_reg, sub_reg;
3919 
3920 	dinfo = device_get_ivars(child);
3921 	cfg = &dinfo->cfg;
3922 	rl = &dinfo->resources;
3923 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3924 	case PCIM_HDRTYPE_BRIDGE:
3925 		sec_reg = PCIR_SECBUS_1;
3926 		sub_reg = PCIR_SUBBUS_1;
3927 		break;
3928 	case PCIM_HDRTYPE_CARDBUS:
3929 		sec_reg = PCIR_SECBUS_2;
3930 		sub_reg = PCIR_SUBBUS_2;
3931 		break;
3932 	default:
3933 		return (NULL);
3934 	}
3935 
3936 	if (*rid != 0)
3937 		return (NULL);
3938 
3939 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3940 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3941 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3942 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3943 		    start, end, count, flags & ~RF_ACTIVE);
3944 		if (res == NULL) {
3945 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3946 			device_printf(child, "allocating %ju bus%s failed\n",
3947 			    count, count == 1 ? "" : "es");
3948 			return (NULL);
3949 		}
3950 		if (bootverbose)
3951 			device_printf(child,
3952 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3953 			    count == 1 ? "" : "es", rman_get_start(res));
3954 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3955 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3956 	}
3957 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3958 	    end, count, flags));
3959 }
3960 
3961 static int
3962 pci_ea_bei_to_rid(device_t dev, int bei)
3963 {
3964 #ifdef PCI_IOV
3965 	struct pci_devinfo *dinfo;
3966 	int iov_pos;
3967 	struct pcicfg_iov *iov;
3968 
3969 	dinfo = device_get_ivars(dev);
3970 	iov = dinfo->cfg.iov;
3971 	if (iov != NULL)
3972 		iov_pos = iov->iov_pos;
3973 	else
3974 		iov_pos = 0;
3975 #endif
3976 
3977 	/* Check if matches BAR */
3978 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3979 	    (bei <= PCIM_EA_BEI_BAR_5))
3980 		return (PCIR_BAR(bei));
3981 
3982 	/* Check ROM */
3983 	if (bei == PCIM_EA_BEI_ROM)
3984 		return (PCIR_BIOS);
3985 
3986 #ifdef PCI_IOV
3987 	/* Check if matches VF_BAR */
3988 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3989 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3990 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3991 		    iov_pos);
3992 #endif
3993 
3994 	return (-1);
3995 }
3996 
3997 int
3998 pci_ea_is_enabled(device_t dev, int rid)
3999 {
4000 	struct pci_ea_entry *ea;
4001 	struct pci_devinfo *dinfo;
4002 
4003 	dinfo = device_get_ivars(dev);
4004 
4005 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4006 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
4007 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
4008 	}
4009 
4010 	return (0);
4011 }
4012 
4013 void
4014 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
4015 {
4016 	struct pci_ea_entry *ea;
4017 	struct pci_devinfo *dinfo;
4018 	pci_addr_t start, end, count;
4019 	struct resource_list *rl;
4020 	int type, flags, rid;
4021 	struct resource *res;
4022 	uint32_t tmp;
4023 #ifdef PCI_IOV
4024 	struct pcicfg_iov *iov;
4025 #endif
4026 
4027 	dinfo = device_get_ivars(dev);
4028 	rl = &dinfo->resources;
4029 	flags = 0;
4030 
4031 #ifdef PCI_IOV
4032 	iov = dinfo->cfg.iov;
4033 #endif
4034 
4035 	if (dinfo->cfg.ea.ea_location == 0)
4036 		return;
4037 
4038 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4039 		/*
4040 		 * TODO: Ignore EA-BAR if is not enabled.
4041 		 *   Currently the EA implementation supports
4042 		 *   only situation, where EA structure contains
4043 		 *   predefined entries. In case they are not enabled
4044 		 *   leave them unallocated and proceed with
4045 		 *   a legacy-BAR mechanism.
4046 		 */
4047 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
4048 			continue;
4049 
4050 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
4051 		case PCIM_EA_P_MEM_PREFETCH:
4052 		case PCIM_EA_P_VF_MEM_PREFETCH:
4053 			flags = RF_PREFETCHABLE;
4054 			/* FALLTHROUGH */
4055 		case PCIM_EA_P_VF_MEM:
4056 		case PCIM_EA_P_MEM:
4057 			type = SYS_RES_MEMORY;
4058 			break;
4059 		case PCIM_EA_P_IO:
4060 			type = SYS_RES_IOPORT;
4061 			break;
4062 		default:
4063 			continue;
4064 		}
4065 
4066 		if (alloc_iov != 0) {
4067 #ifdef PCI_IOV
4068 			/* Allocating IOV, confirm BEI matches */
4069 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
4070 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
4071 				continue;
4072 #else
4073 			continue;
4074 #endif
4075 		} else {
4076 			/* Allocating BAR, confirm BEI matches */
4077 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
4078 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
4079 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
4080 				continue;
4081 		}
4082 
4083 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
4084 		if (rid < 0)
4085 			continue;
4086 
4087 		/* Skip resources already allocated by EA */
4088 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
4089 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
4090 			continue;
4091 
4092 		start = ea->eae_base;
4093 		count = ea->eae_max_offset + 1;
4094 #ifdef PCI_IOV
4095 		if (iov != NULL)
4096 			count = count * iov->iov_num_vfs;
4097 #endif
4098 		end = start + count - 1;
4099 		if (count == 0)
4100 			continue;
4101 
4102 		resource_list_add(rl, type, rid, start, end, count);
4103 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4104 		    flags);
4105 		if (res == NULL) {
4106 			resource_list_delete(rl, type, rid);
4107 
4108 			/*
4109 			 * Failed to allocate using EA, disable entry.
4110 			 * Another attempt to allocation will be performed
4111 			 * further, but this time using legacy BAR registers
4112 			 */
4113 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4114 			tmp &= ~PCIM_EA_ENABLE;
4115 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4116 
4117 			/*
4118 			 * Disabling entry might fail in case it is hardwired.
4119 			 * Read flags again to match current status.
4120 			 */
4121 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4122 
4123 			continue;
4124 		}
4125 
4126 		/* As per specification, fill BAR with zeros */
4127 		pci_write_config(dev, rid, 0, 4);
4128 	}
4129 }
4130 
4131 void
4132 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4133 {
4134 	struct pci_devinfo *dinfo;
4135 	pcicfgregs *cfg;
4136 	struct resource_list *rl;
4137 	const struct pci_quirk *q;
4138 	uint32_t devid;
4139 	int i;
4140 
4141 	dinfo = device_get_ivars(dev);
4142 	cfg = &dinfo->cfg;
4143 	rl = &dinfo->resources;
4144 	devid = (cfg->device << 16) | cfg->vendor;
4145 
4146 	/* Allocate resources using Enhanced Allocation */
4147 	pci_add_resources_ea(bus, dev, 0);
4148 
4149 	/* ATA devices needs special map treatment */
4150 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4151 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4152 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4153 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4154 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4155 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4156 	else
4157 		for (i = 0; i < cfg->nummaps;) {
4158 			/* Skip resources already managed by EA */
4159 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4160 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4161 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4162 				i++;
4163 				continue;
4164 			}
4165 
4166 			/*
4167 			 * Skip quirked resources.
4168 			 */
4169 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4170 				if (q->devid == devid &&
4171 				    q->type == PCI_QUIRK_UNMAP_REG &&
4172 				    q->arg1 == PCIR_BAR(i))
4173 					break;
4174 			if (q->devid != 0) {
4175 				i++;
4176 				continue;
4177 			}
4178 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4179 			    prefetchmask & (1 << i));
4180 		}
4181 
4182 	/*
4183 	 * Add additional, quirked resources.
4184 	 */
4185 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4186 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4187 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4188 
4189 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline) &&
4190 	    pci_intx_reroute) {
4191 		/*
4192 		 * Try to re-route interrupts. Sometimes the BIOS or
4193 		 * firmware may leave bogus values in these registers.
4194 		 * If the re-route fails, then just stick with what we
4195 		 * have.
4196 		 */
4197 		pci_assign_interrupt(bus, dev, 1);
4198 	}
4199 
4200 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4201 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4202 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4203 			xhci_early_takeover(dev);
4204 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4205 			ehci_early_takeover(dev);
4206 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4207 			ohci_early_takeover(dev);
4208 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4209 			uhci_early_takeover(dev);
4210 	}
4211 
4212 	/*
4213 	 * Reserve resources for secondary bus ranges behind bridge
4214 	 * devices.
4215 	 */
4216 	pci_reserve_secbus(bus, dev, cfg, rl);
4217 }
4218 
4219 static struct pci_devinfo *
4220 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4221     int slot, int func)
4222 {
4223 	struct pci_devinfo *dinfo;
4224 
4225 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4226 	if (dinfo != NULL)
4227 		pci_add_child(dev, dinfo);
4228 
4229 	return (dinfo);
4230 }
4231 
4232 void
4233 pci_add_children(device_t dev, int domain, int busno)
4234 {
4235 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4236 	device_t pcib = device_get_parent(dev);
4237 	struct pci_devinfo *dinfo;
4238 	int maxslots;
4239 	int s, f, pcifunchigh;
4240 	uint8_t hdrtype;
4241 	int first_func;
4242 
4243 	/*
4244 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4245 	 * enable ARI.  We must enable ARI before detecting the rest of the
4246 	 * functions on this bus as ARI changes the set of slots and functions
4247 	 * that are legal on this bus.
4248 	 */
4249 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4250 	if (dinfo != NULL && pci_enable_ari)
4251 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4252 
4253 	/*
4254 	 * Start looking for new devices on slot 0 at function 1 because we
4255 	 * just identified the device at slot 0, function 0.
4256 	 */
4257 	first_func = 1;
4258 
4259 	maxslots = PCIB_MAXSLOTS(pcib);
4260 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4261 		pcifunchigh = 0;
4262 		f = 0;
4263 		DELAY(1);
4264 
4265 		/* If function 0 is not present, skip to the next slot. */
4266 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4267 			continue;
4268 		hdrtype = REG(PCIR_HDRTYPE, 1);
4269 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4270 			continue;
4271 		if (hdrtype & PCIM_MFDEV)
4272 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4273 		for (f = first_func; f <= pcifunchigh; f++)
4274 			pci_identify_function(pcib, dev, domain, busno, s, f);
4275 	}
4276 #undef REG
4277 }
4278 
4279 int
4280 pci_rescan_method(device_t dev)
4281 {
4282 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4283 	device_t pcib = device_get_parent(dev);
4284 	device_t child, *devlist, *unchanged;
4285 	int devcount, error, i, j, maxslots, oldcount;
4286 	int busno, domain, s, f, pcifunchigh;
4287 	uint8_t hdrtype;
4288 
4289 	/* No need to check for ARI on a rescan. */
4290 	error = device_get_children(dev, &devlist, &devcount);
4291 	if (error)
4292 		return (error);
4293 	if (devcount != 0) {
4294 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4295 		    M_NOWAIT | M_ZERO);
4296 		if (unchanged == NULL) {
4297 			free(devlist, M_TEMP);
4298 			return (ENOMEM);
4299 		}
4300 	} else
4301 		unchanged = NULL;
4302 
4303 	domain = pcib_get_domain(dev);
4304 	busno = pcib_get_bus(dev);
4305 	maxslots = PCIB_MAXSLOTS(pcib);
4306 	for (s = 0; s <= maxslots; s++) {
4307 		/* If function 0 is not present, skip to the next slot. */
4308 		f = 0;
4309 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4310 			continue;
4311 		pcifunchigh = 0;
4312 		hdrtype = REG(PCIR_HDRTYPE, 1);
4313 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4314 			continue;
4315 		if (hdrtype & PCIM_MFDEV)
4316 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4317 		for (f = 0; f <= pcifunchigh; f++) {
4318 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4319 				continue;
4320 
4321 			/*
4322 			 * Found a valid function.  Check if a
4323 			 * device_t for this device already exists.
4324 			 */
4325 			for (i = 0; i < devcount; i++) {
4326 				child = devlist[i];
4327 				if (child == NULL)
4328 					continue;
4329 				if (pci_get_slot(child) == s &&
4330 				    pci_get_function(child) == f) {
4331 					unchanged[i] = child;
4332 					goto next_func;
4333 				}
4334 			}
4335 
4336 			pci_identify_function(pcib, dev, domain, busno, s, f);
4337 		next_func:;
4338 		}
4339 	}
4340 
4341 	/* Remove devices that are no longer present. */
4342 	for (i = 0; i < devcount; i++) {
4343 		if (unchanged[i] != NULL)
4344 			continue;
4345 		device_delete_child(dev, devlist[i]);
4346 	}
4347 
4348 	free(devlist, M_TEMP);
4349 	oldcount = devcount;
4350 
4351 	/* Try to attach the devices just added. */
4352 	error = device_get_children(dev, &devlist, &devcount);
4353 	if (error) {
4354 		free(unchanged, M_TEMP);
4355 		return (error);
4356 	}
4357 
4358 	for (i = 0; i < devcount; i++) {
4359 		for (j = 0; j < oldcount; j++) {
4360 			if (devlist[i] == unchanged[j])
4361 				goto next_device;
4362 		}
4363 
4364 		device_probe_and_attach(devlist[i]);
4365 	next_device:;
4366 	}
4367 
4368 	free(unchanged, M_TEMP);
4369 	free(devlist, M_TEMP);
4370 	return (0);
4371 #undef REG
4372 }
4373 
4374 #ifdef PCI_IOV
4375 device_t
4376 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4377     uint16_t did)
4378 {
4379 	struct pci_devinfo *vf_dinfo;
4380 	device_t pcib;
4381 	int busno, slot, func;
4382 
4383 	pcib = device_get_parent(bus);
4384 
4385 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4386 
4387 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4388 	    slot, func, vid, did);
4389 
4390 	vf_dinfo->cfg.flags |= PCICFG_VF;
4391 	pci_add_child(bus, vf_dinfo);
4392 
4393 	return (vf_dinfo->cfg.dev);
4394 }
4395 
4396 device_t
4397 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4398     uint16_t vid, uint16_t did)
4399 {
4400 
4401 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4402 }
4403 #endif
4404 
4405 /*
4406  * For PCIe device set Max_Payload_Size to match PCIe root's.
4407  */
4408 static void
4409 pcie_setup_mps(device_t dev)
4410 {
4411 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4412 	device_t root;
4413 	uint16_t rmps, mmps, mps;
4414 
4415 	if (dinfo->cfg.pcie.pcie_location == 0)
4416 		return;
4417 	root = pci_find_pcie_root_port(dev);
4418 	if (root == NULL)
4419 		return;
4420 	/* Check whether the MPS is already configured. */
4421 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4422 	    PCIEM_CTL_MAX_PAYLOAD;
4423 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4424 	    PCIEM_CTL_MAX_PAYLOAD;
4425 	if (mps == rmps)
4426 		return;
4427 	/* Check whether the device is capable of the root's MPS. */
4428 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4429 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4430 	if (rmps > mmps) {
4431 		/*
4432 		 * The device is unable to handle root's MPS.  Limit root.
4433 		 * XXX: We should traverse through all the tree, applying
4434 		 * it to all the devices.
4435 		 */
4436 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4437 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4438 	} else {
4439 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4440 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4441 	}
4442 }
4443 
4444 static void
4445 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4446 {
4447 	int aer;
4448 	uint32_t r;
4449 	uint16_t r2;
4450 
4451 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4452 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4453 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4454 		    PCIER_ROOT_CTL, 2);
4455 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4456 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4457 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4458 		    PCIER_ROOT_CTL, r2, 2);
4459 	}
4460 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4461 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4462 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4463 		if (r != 0 && bootverbose) {
4464 			pci_printf(&dinfo->cfg,
4465 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4466 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4467 			    4));
4468 		}
4469 
4470 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4471 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4472 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4473 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4474 		    PCIM_AER_UC_POISONED_TLP |
4475 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4476 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4477 		    PCIM_AER_UC_COMPLETER_ABORT |
4478 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4479 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4480 		    PCIM_AER_UC_MALFORMED_TLP |
4481 		    PCIM_AER_UC_ECRC_ERROR |
4482 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4483 		    PCIM_AER_UC_ACS_VIOLATION |
4484 		    PCIM_AER_UC_INTERNAL_ERROR |
4485 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4486 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4487 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4488 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4489 
4490 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4491 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4492 		if (r != 0 && bootverbose) {
4493 			pci_printf(&dinfo->cfg,
4494 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4495 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4496 			    4));
4497 		}
4498 
4499 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4500 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4501 		    PCIM_AER_COR_BAD_TLP |
4502 		    PCIM_AER_COR_BAD_DLLP |
4503 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4504 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4505 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4506 		    PCIM_AER_COR_INTERNAL_ERROR |
4507 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4508 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4509 
4510 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4511 		    PCIER_DEVICE_CTL, 2);
4512 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4513 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4514 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4515 		    PCIER_DEVICE_CTL, r, 2);
4516 	}
4517 }
4518 
4519 void
4520 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4521 {
4522 	device_t dev;
4523 
4524 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4525 	device_set_ivars(dev, dinfo);
4526 	resource_list_init(&dinfo->resources);
4527 	pci_cfg_save(dev, dinfo, 0);
4528 	pci_cfg_restore(dev, dinfo);
4529 	pci_clear_pme(dev);
4530 	pci_print_verbose(dinfo);
4531 	pci_add_resources(bus, dev, 0, 0);
4532 	if (pci_enable_mps_tune)
4533 		pcie_setup_mps(dev);
4534 	pci_child_added(dinfo->cfg.dev);
4535 
4536 	if (pci_clear_aer_on_attach)
4537 		pci_add_child_clear_aer(dev, dinfo);
4538 
4539 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4540 }
4541 
4542 void
4543 pci_child_added_method(device_t dev, device_t child)
4544 {
4545 
4546 }
4547 
4548 static int
4549 pci_probe(device_t dev)
4550 {
4551 
4552 	device_set_desc(dev, "PCI bus");
4553 
4554 	/* Allow other subclasses to override this driver. */
4555 	return (BUS_PROBE_GENERIC);
4556 }
4557 
4558 int
4559 pci_attach_common(device_t dev)
4560 {
4561 	struct pci_softc *sc;
4562 	int busno, domain;
4563 	int rid;
4564 
4565 	sc = device_get_softc(dev);
4566 	domain = pcib_get_domain(dev);
4567 	busno = pcib_get_bus(dev);
4568 	rid = 0;
4569 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4570 	    1, 0);
4571 	if (sc->sc_bus == NULL) {
4572 		device_printf(dev, "failed to allocate bus number\n");
4573 		return (ENXIO);
4574 	}
4575 	if (bootverbose)
4576 		device_printf(dev, "domain=%d, physical bus=%d\n",
4577 		    domain, busno);
4578 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4579 	return (0);
4580 }
4581 
4582 int
4583 pci_attach(device_t dev)
4584 {
4585 	int busno, domain, error;
4586 
4587 	error = pci_attach_common(dev);
4588 	if (error)
4589 		return (error);
4590 
4591 	/*
4592 	 * Since there can be multiple independently numbered PCI
4593 	 * buses on systems with multiple PCI domains, we can't use
4594 	 * the unit number to decide which bus we are probing. We ask
4595 	 * the parent pcib what our domain and bus numbers are.
4596 	 */
4597 	domain = pcib_get_domain(dev);
4598 	busno = pcib_get_bus(dev);
4599 	pci_add_children(dev, domain, busno);
4600 	bus_attach_children(dev);
4601 	return (0);
4602 }
4603 
4604 int
4605 pci_detach(device_t dev)
4606 {
4607 	struct pci_softc *sc;
4608 	int error;
4609 
4610 	error = bus_generic_detach(dev);
4611 	if (error)
4612 		return (error);
4613 	sc = device_get_softc(dev);
4614 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4615 	return (error);
4616 }
4617 
4618 static void
4619 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4620 {
4621 	int line, unit;
4622 	const char *at;
4623 	char me1[24], me2[32];
4624 	uint8_t b, s, f;
4625 	uint32_t d;
4626 	device_location_cache_t *cache;
4627 
4628 	d = pci_get_domain(child);
4629 	b = pci_get_bus(child);
4630 	s = pci_get_slot(child);
4631 	f = pci_get_function(child);
4632 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4633 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4634 	line = 0;
4635 	cache = dev_wired_cache_init();
4636 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4637 		resource_string_value(name, unit, "at", &at);
4638 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4639 			*unitp = unit;
4640 			break;
4641 		}
4642 		if (dev_wired_cache_match(cache, child, at)) {
4643 			*unitp = unit;
4644 			break;
4645 		}
4646 	}
4647 	dev_wired_cache_fini(cache);
4648 }
4649 
4650 static void
4651 pci_set_power_child(device_t dev, device_t child, int state)
4652 {
4653 	device_t pcib;
4654 	int dstate;
4655 
4656 	/*
4657 	 * Set the device to the given state.  If the firmware suggests
4658 	 * a different power state, use it instead.  If power management
4659 	 * is not present, the firmware is responsible for managing
4660 	 * device power.  Skip children who aren't attached since they
4661 	 * are handled separately.
4662 	 */
4663 	pcib = device_get_parent(dev);
4664 	dstate = state;
4665 	if (device_is_attached(child) &&
4666 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4667 		pci_set_powerstate(child, dstate);
4668 }
4669 
4670 int
4671 pci_suspend_child(device_t dev, device_t child)
4672 {
4673 	struct pci_devinfo *dinfo;
4674 	struct resource_list_entry *rle;
4675 	int error;
4676 
4677 	dinfo = device_get_ivars(child);
4678 
4679 	/*
4680 	 * Save the PCI configuration space for the child and set the
4681 	 * device in the appropriate power state for this sleep state.
4682 	 */
4683 	pci_cfg_save(child, dinfo, 0);
4684 
4685 	/* Suspend devices before potentially powering them down. */
4686 	error = bus_generic_suspend_child(dev, child);
4687 
4688 	if (error)
4689 		return (error);
4690 
4691 	if (pci_do_power_suspend) {
4692 		/*
4693 		 * Make sure this device's interrupt handler is not invoked
4694 		 * in the case the device uses a shared interrupt that can
4695 		 * be raised by some other device.
4696 		 * This is applicable only to regular (legacy) PCI interrupts
4697 		 * as MSI/MSI-X interrupts are never shared.
4698 		 */
4699 		rle = resource_list_find(&dinfo->resources,
4700 		    SYS_RES_IRQ, 0);
4701 		if (rle != NULL && rle->res != NULL)
4702 			(void)bus_suspend_intr(child, rle->res);
4703 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4704 	}
4705 
4706 	return (0);
4707 }
4708 
4709 int
4710 pci_resume_child(device_t dev, device_t child)
4711 {
4712 	struct pci_devinfo *dinfo;
4713 	struct resource_list_entry *rle;
4714 
4715 	if (pci_do_power_resume)
4716 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4717 
4718 	dinfo = device_get_ivars(child);
4719 	pci_cfg_restore(child, dinfo);
4720 	pci_clear_pme(child);
4721 	if (!device_is_attached(child))
4722 		pci_cfg_save(child, dinfo, 1);
4723 
4724 	bus_generic_resume_child(dev, child);
4725 
4726 	/*
4727 	 * Allow interrupts only after fully resuming the driver and hardware.
4728 	 */
4729 	if (pci_do_power_suspend) {
4730 		/* See pci_suspend_child for details. */
4731 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4732 		if (rle != NULL && rle->res != NULL)
4733 			(void)bus_resume_intr(child, rle->res);
4734 	}
4735 
4736 	return (0);
4737 }
4738 
4739 int
4740 pci_resume(device_t dev)
4741 {
4742 	device_t child, *devlist;
4743 	int error, i, numdevs;
4744 
4745 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4746 		return (error);
4747 
4748 	/*
4749 	 * Resume critical devices first, then everything else later.
4750 	 */
4751 	for (i = 0; i < numdevs; i++) {
4752 		child = devlist[i];
4753 		switch (pci_get_class(child)) {
4754 		case PCIC_DISPLAY:
4755 		case PCIC_MEMORY:
4756 		case PCIC_BRIDGE:
4757 		case PCIC_BASEPERIPH:
4758 			BUS_RESUME_CHILD(dev, child);
4759 			break;
4760 		}
4761 	}
4762 	for (i = 0; i < numdevs; i++) {
4763 		child = devlist[i];
4764 		switch (pci_get_class(child)) {
4765 		case PCIC_DISPLAY:
4766 		case PCIC_MEMORY:
4767 		case PCIC_BRIDGE:
4768 		case PCIC_BASEPERIPH:
4769 			break;
4770 		default:
4771 			BUS_RESUME_CHILD(dev, child);
4772 		}
4773 	}
4774 	free(devlist, M_TEMP);
4775 	return (0);
4776 }
4777 
4778 static void
4779 pci_load_vendor_data(void)
4780 {
4781 	caddr_t data;
4782 	void *ptr;
4783 	size_t sz;
4784 
4785 	data = preload_search_by_type("pci_vendor_data");
4786 	if (data != NULL) {
4787 		ptr = preload_fetch_addr(data);
4788 		sz = preload_fetch_size(data);
4789 		if (ptr != NULL && sz != 0) {
4790 			pci_vendordata = ptr;
4791 			pci_vendordata_size = sz;
4792 			/* terminate the database */
4793 			pci_vendordata[pci_vendordata_size] = '\n';
4794 		}
4795 	}
4796 }
4797 
4798 void
4799 pci_driver_added(device_t dev, driver_t *driver)
4800 {
4801 	int numdevs;
4802 	device_t *devlist;
4803 	device_t child;
4804 	struct pci_devinfo *dinfo;
4805 	int i;
4806 
4807 	if (bootverbose)
4808 		device_printf(dev, "driver added\n");
4809 	DEVICE_IDENTIFY(driver, dev);
4810 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4811 		return;
4812 	for (i = 0; i < numdevs; i++) {
4813 		child = devlist[i];
4814 		if (device_get_state(child) != DS_NOTPRESENT)
4815 			continue;
4816 		dinfo = device_get_ivars(child);
4817 		pci_print_verbose(dinfo);
4818 		if (bootverbose)
4819 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4820 		pci_cfg_restore(child, dinfo);
4821 		if (device_probe_and_attach(child) != 0)
4822 			pci_child_detached(dev, child);
4823 	}
4824 	free(devlist, M_TEMP);
4825 }
4826 
4827 int
4828 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4829     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4830 {
4831 	struct pci_devinfo *dinfo;
4832 	struct msix_table_entry *mte;
4833 	struct msix_vector *mv;
4834 	uint64_t addr;
4835 	uint32_t data;
4836 	void *cookie;
4837 	int error, rid;
4838 
4839 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4840 	    arg, &cookie);
4841 	if (error)
4842 		return (error);
4843 
4844 	/* If this is not a direct child, just bail out. */
4845 	if (device_get_parent(child) != dev) {
4846 		*cookiep = cookie;
4847 		return(0);
4848 	}
4849 
4850 	rid = rman_get_rid(irq);
4851 	if (rid == 0) {
4852 		/* Make sure that INTx is enabled */
4853 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4854 	} else {
4855 		/*
4856 		 * Check to see if the interrupt is MSI or MSI-X.
4857 		 * Ask our parent to map the MSI and give
4858 		 * us the address and data register values.
4859 		 * If we fail for some reason, teardown the
4860 		 * interrupt handler.
4861 		 */
4862 		dinfo = device_get_ivars(child);
4863 		if (dinfo->cfg.msi.msi_alloc > 0) {
4864 			if (dinfo->cfg.msi.msi_addr == 0) {
4865 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4866 			    ("MSI has handlers, but vectors not mapped"));
4867 				error = PCIB_MAP_MSI(device_get_parent(dev),
4868 				    child, rman_get_start(irq), &addr, &data);
4869 				if (error)
4870 					goto bad;
4871 				dinfo->cfg.msi.msi_addr = addr;
4872 				dinfo->cfg.msi.msi_data = data;
4873 			}
4874 			if (dinfo->cfg.msi.msi_handlers == 0)
4875 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4876 				    dinfo->cfg.msi.msi_data);
4877 			dinfo->cfg.msi.msi_handlers++;
4878 		} else {
4879 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4880 			    ("No MSI or MSI-X interrupts allocated"));
4881 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4882 			    ("MSI-X index too high"));
4883 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4884 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4885 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4886 			KASSERT(mv->mv_irq == rman_get_start(irq),
4887 			    ("IRQ mismatch"));
4888 			if (mv->mv_address == 0) {
4889 				KASSERT(mte->mte_handlers == 0,
4890 		    ("MSI-X table entry has handlers, but vector not mapped"));
4891 				error = PCIB_MAP_MSI(device_get_parent(dev),
4892 				    child, rman_get_start(irq), &addr, &data);
4893 				if (error)
4894 					goto bad;
4895 				mv->mv_address = addr;
4896 				mv->mv_data = data;
4897 			}
4898 
4899 			/*
4900 			 * The MSIX table entry must be made valid by
4901 			 * incrementing the mte_handlers before
4902 			 * calling pci_enable_msix() and
4903 			 * pci_resume_msix(). Else the MSIX rewrite
4904 			 * table quirk will not work as expected.
4905 			 */
4906 			mte->mte_handlers++;
4907 			if (mte->mte_handlers == 1) {
4908 				pci_enable_msix(child, rid - 1, mv->mv_address,
4909 				    mv->mv_data);
4910 				pci_unmask_msix(child, rid - 1);
4911 			}
4912 		}
4913 
4914 		/*
4915 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4916 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4917 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4918 		 */
4919 		if (!pci_has_quirk(pci_get_devid(child),
4920 		    PCI_QUIRK_MSI_INTX_BUG))
4921 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4922 		else
4923 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4924 	bad:
4925 		if (error) {
4926 			(void)bus_generic_teardown_intr(dev, child, irq,
4927 			    cookie);
4928 			return (error);
4929 		}
4930 	}
4931 	*cookiep = cookie;
4932 	return (0);
4933 }
4934 
4935 int
4936 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4937     void *cookie)
4938 {
4939 	struct msix_table_entry *mte;
4940 	struct resource_list_entry *rle;
4941 	struct pci_devinfo *dinfo;
4942 	int error, rid;
4943 
4944 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4945 		return (EINVAL);
4946 
4947 	/* If this isn't a direct child, just bail out */
4948 	if (device_get_parent(child) != dev)
4949 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4950 
4951 	rid = rman_get_rid(irq);
4952 	if (rid == 0) {
4953 		/* Mask INTx */
4954 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4955 	} else {
4956 		/*
4957 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4958 		 * decrement the appropriate handlers count and mask the
4959 		 * MSI-X message, or disable MSI messages if the count
4960 		 * drops to 0.
4961 		 */
4962 		dinfo = device_get_ivars(child);
4963 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4964 		if (rle->res != irq)
4965 			return (EINVAL);
4966 		if (dinfo->cfg.msi.msi_alloc > 0) {
4967 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4968 			    ("MSI-X index too high"));
4969 			if (dinfo->cfg.msi.msi_handlers == 0)
4970 				return (EINVAL);
4971 			dinfo->cfg.msi.msi_handlers--;
4972 			if (dinfo->cfg.msi.msi_handlers == 0)
4973 				pci_disable_msi(child);
4974 		} else {
4975 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4976 			    ("No MSI or MSI-X interrupts allocated"));
4977 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4978 			    ("MSI-X index too high"));
4979 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4980 			if (mte->mte_handlers == 0)
4981 				return (EINVAL);
4982 			mte->mte_handlers--;
4983 			if (mte->mte_handlers == 0)
4984 				pci_mask_msix(child, rid - 1);
4985 		}
4986 	}
4987 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4988 	if (rid > 0)
4989 		KASSERT(error == 0,
4990 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4991 	return (error);
4992 }
4993 
4994 int
4995 pci_print_child(device_t dev, device_t child)
4996 {
4997 	struct pci_devinfo *dinfo;
4998 	struct resource_list *rl;
4999 	int retval = 0;
5000 
5001 	dinfo = device_get_ivars(child);
5002 	rl = &dinfo->resources;
5003 
5004 	retval += bus_print_child_header(dev, child);
5005 
5006 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
5007 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
5008 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
5009 	if (device_get_flags(dev))
5010 		retval += printf(" flags %#x", device_get_flags(dev));
5011 
5012 	retval += printf(" at device %d.%d", pci_get_slot(child),
5013 	    pci_get_function(child));
5014 
5015 	retval += bus_print_child_domain(dev, child);
5016 	retval += bus_print_child_footer(dev, child);
5017 
5018 	return (retval);
5019 }
5020 
5021 static const struct
5022 {
5023 	int		class;
5024 	int		subclass;
5025 	int		report; /* 0 = bootverbose, 1 = always */
5026 	const char	*desc;
5027 } pci_nomatch_tab[] = {
5028 	{PCIC_OLD,		-1,			1, "old"},
5029 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
5030 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
5031 	{PCIC_STORAGE,		-1,			1, "mass storage"},
5032 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
5033 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
5034 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
5035 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
5036 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
5037 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
5038 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
5039 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
5040 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
5041 	{PCIC_NETWORK,		-1,			1, "network"},
5042 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
5043 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
5044 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
5045 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
5046 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
5047 	{PCIC_DISPLAY,		-1,			1, "display"},
5048 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
5049 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
5050 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
5051 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
5052 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
5053 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
5054 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
5055 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
5056 	{PCIC_MEMORY,		-1,			1, "memory"},
5057 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
5058 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
5059 	{PCIC_BRIDGE,		-1,			1, "bridge"},
5060 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
5061 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
5062 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
5063 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
5064 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
5065 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
5066 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
5067 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
5068 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
5069 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
5070 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
5071 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
5072 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
5073 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
5074 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
5075 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
5076 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
5077 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
5078 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
5079 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
5080 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
5081 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
5082 	{PCIC_INPUTDEV,		-1,			1, "input device"},
5083 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
5084 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
5085 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
5086 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
5087 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5088 	{PCIC_DOCKING,		-1,			1, "docking station"},
5089 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5090 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5091 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5092 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5093 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5094 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5095 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5096 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5097 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5098 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5099 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5100 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5101 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5102 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5103 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5104 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5105 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5106 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5107 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5108 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5109 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5110 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5111 	{PCIC_DASP,		-1,			0, "dasp"},
5112 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5113 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5114 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5115 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5116 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5117 	{0, 0, 0,		NULL}
5118 };
5119 
5120 void
5121 pci_probe_nomatch(device_t dev, device_t child)
5122 {
5123 	int i, report;
5124 	const char *cp, *scp;
5125 	char *device;
5126 
5127 	/*
5128 	 * Look for a listing for this device in a loaded device database.
5129 	 */
5130 	report = 1;
5131 	if ((device = pci_describe_device(child)) != NULL) {
5132 		device_printf(dev, "<%s>", device);
5133 		free(device, M_DEVBUF);
5134 	} else {
5135 		/*
5136 		 * Scan the class/subclass descriptions for a general
5137 		 * description.
5138 		 */
5139 		cp = "unknown";
5140 		scp = NULL;
5141 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5142 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5143 				if (pci_nomatch_tab[i].subclass == -1) {
5144 					cp = pci_nomatch_tab[i].desc;
5145 					report = pci_nomatch_tab[i].report;
5146 				} else if (pci_nomatch_tab[i].subclass ==
5147 				    pci_get_subclass(child)) {
5148 					scp = pci_nomatch_tab[i].desc;
5149 					report = pci_nomatch_tab[i].report;
5150 				}
5151 			}
5152 		}
5153 		if (report || bootverbose) {
5154 			device_printf(dev, "<%s%s%s>",
5155 			    cp ? cp : "",
5156 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5157 			    scp ? scp : "");
5158 		}
5159 	}
5160 	if (report || bootverbose) {
5161 		printf(" at device %d.%d (no driver attached)\n",
5162 		    pci_get_slot(child), pci_get_function(child));
5163 	}
5164 	pci_cfg_save(child, device_get_ivars(child), 1);
5165 }
5166 
5167 void
5168 pci_child_detached(device_t dev, device_t child)
5169 {
5170 	struct pci_devinfo *dinfo;
5171 	struct resource_list *rl;
5172 
5173 	dinfo = device_get_ivars(child);
5174 	rl = &dinfo->resources;
5175 
5176 	/*
5177 	 * Have to deallocate IRQs before releasing any MSI messages and
5178 	 * have to release MSI messages before deallocating any memory
5179 	 * BARs.
5180 	 */
5181 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5182 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5183 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5184 		if (dinfo->cfg.msi.msi_alloc != 0)
5185 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5186 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5187 		else
5188 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5189 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5190 		(void)pci_release_msi(child);
5191 	}
5192 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5193 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5194 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5195 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5196 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5197 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5198 
5199 	pci_cfg_save(child, dinfo, 1);
5200 }
5201 
5202 /*
5203  * Parse the PCI device database, if loaded, and return a pointer to a
5204  * description of the device.
5205  *
5206  * The database is flat text formatted as follows:
5207  *
5208  * Any line not in a valid format is ignored.
5209  * Lines are terminated with newline '\n' characters.
5210  *
5211  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5212  * the vendor name.
5213  *
5214  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5215  * - devices cannot be listed without a corresponding VENDOR line.
5216  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5217  * another TAB, then the device name.
5218  */
5219 
5220 /*
5221  * Assuming (ptr) points to the beginning of a line in the database,
5222  * return the vendor or device and description of the next entry.
5223  * The value of (vendor) or (device) inappropriate for the entry type
5224  * is set to -1.  Returns nonzero at the end of the database.
5225  *
5226  * Note that this is slightly unrobust in the face of corrupt data;
5227  * we attempt to safeguard against this by spamming the end of the
5228  * database with a newline when we initialise.
5229  */
5230 static int
5231 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5232 {
5233 	char	*cp = *ptr;
5234 	int	left;
5235 
5236 	*device = -1;
5237 	*vendor = -1;
5238 	**desc = '\0';
5239 	for (;;) {
5240 		left = pci_vendordata_size - (cp - pci_vendordata);
5241 		if (left <= 0) {
5242 			*ptr = cp;
5243 			return(1);
5244 		}
5245 
5246 		/* vendor entry? */
5247 		if (*cp != '\t' &&
5248 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5249 			break;
5250 		/* device entry? */
5251 		if (*cp == '\t' &&
5252 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5253 			break;
5254 
5255 		/* skip to next line */
5256 		while (*cp != '\n' && left > 0) {
5257 			cp++;
5258 			left--;
5259 		}
5260 		if (*cp == '\n') {
5261 			cp++;
5262 			left--;
5263 		}
5264 	}
5265 	/* skip to next line */
5266 	while (*cp != '\n' && left > 0) {
5267 		cp++;
5268 		left--;
5269 	}
5270 	if (*cp == '\n' && left > 0)
5271 		cp++;
5272 	*ptr = cp;
5273 	return(0);
5274 }
5275 
5276 static char *
5277 pci_describe_device(device_t dev)
5278 {
5279 	int	vendor, device;
5280 	char	*desc, *vp, *dp, *line;
5281 
5282 	desc = vp = dp = NULL;
5283 
5284 	/*
5285 	 * If we have no vendor data, we can't do anything.
5286 	 */
5287 	if (pci_vendordata == NULL)
5288 		goto out;
5289 
5290 	/*
5291 	 * Scan the vendor data looking for this device
5292 	 */
5293 	line = pci_vendordata;
5294 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5295 		goto out;
5296 	for (;;) {
5297 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5298 			goto out;
5299 		if (vendor == pci_get_vendor(dev))
5300 			break;
5301 	}
5302 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5303 		goto out;
5304 	for (;;) {
5305 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5306 			*dp = 0;
5307 			break;
5308 		}
5309 		if (vendor != -1) {
5310 			*dp = 0;
5311 			break;
5312 		}
5313 		if (device == pci_get_device(dev))
5314 			break;
5315 	}
5316 	if (dp[0] == '\0')
5317 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5318 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5319 	    NULL)
5320 		sprintf(desc, "%s, %s", vp, dp);
5321 out:
5322 	if (vp != NULL)
5323 		free(vp, M_DEVBUF);
5324 	if (dp != NULL)
5325 		free(dp, M_DEVBUF);
5326 	return(desc);
5327 }
5328 
5329 int
5330 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5331 {
5332 	struct pci_devinfo *dinfo;
5333 	pcicfgregs *cfg;
5334 
5335 	dinfo = device_get_ivars(child);
5336 	cfg = &dinfo->cfg;
5337 
5338 	switch (which) {
5339 	case PCI_IVAR_ETHADDR:
5340 		/*
5341 		 * The generic accessor doesn't deal with failure, so
5342 		 * we set the return value, then return an error.
5343 		 */
5344 		*((uint8_t **) result) = NULL;
5345 		return (EINVAL);
5346 	case PCI_IVAR_SUBVENDOR:
5347 		*result = cfg->subvendor;
5348 		break;
5349 	case PCI_IVAR_SUBDEVICE:
5350 		*result = cfg->subdevice;
5351 		break;
5352 	case PCI_IVAR_VENDOR:
5353 		*result = cfg->vendor;
5354 		break;
5355 	case PCI_IVAR_DEVICE:
5356 		*result = cfg->device;
5357 		break;
5358 	case PCI_IVAR_DEVID:
5359 		*result = (cfg->device << 16) | cfg->vendor;
5360 		break;
5361 	case PCI_IVAR_CLASS:
5362 		*result = cfg->baseclass;
5363 		break;
5364 	case PCI_IVAR_SUBCLASS:
5365 		*result = cfg->subclass;
5366 		break;
5367 	case PCI_IVAR_PROGIF:
5368 		*result = cfg->progif;
5369 		break;
5370 	case PCI_IVAR_REVID:
5371 		*result = cfg->revid;
5372 		break;
5373 	case PCI_IVAR_INTPIN:
5374 		*result = cfg->intpin;
5375 		break;
5376 	case PCI_IVAR_IRQ:
5377 		*result = cfg->intline;
5378 		break;
5379 	case PCI_IVAR_DOMAIN:
5380 		*result = cfg->domain;
5381 		break;
5382 	case PCI_IVAR_BUS:
5383 		*result = cfg->bus;
5384 		break;
5385 	case PCI_IVAR_SLOT:
5386 		*result = cfg->slot;
5387 		break;
5388 	case PCI_IVAR_FUNCTION:
5389 		*result = cfg->func;
5390 		break;
5391 	case PCI_IVAR_CMDREG:
5392 		*result = cfg->cmdreg;
5393 		break;
5394 	case PCI_IVAR_CACHELNSZ:
5395 		*result = cfg->cachelnsz;
5396 		break;
5397 	case PCI_IVAR_MINGNT:
5398 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5399 			*result = -1;
5400 			return (EINVAL);
5401 		}
5402 		*result = cfg->mingnt;
5403 		break;
5404 	case PCI_IVAR_MAXLAT:
5405 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5406 			*result = -1;
5407 			return (EINVAL);
5408 		}
5409 		*result = cfg->maxlat;
5410 		break;
5411 	case PCI_IVAR_LATTIMER:
5412 		*result = cfg->lattimer;
5413 		break;
5414 	default:
5415 		return (ENOENT);
5416 	}
5417 	return (0);
5418 }
5419 
5420 int
5421 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5422 {
5423 	struct pci_devinfo *dinfo;
5424 
5425 	dinfo = device_get_ivars(child);
5426 
5427 	switch (which) {
5428 	case PCI_IVAR_INTPIN:
5429 		dinfo->cfg.intpin = value;
5430 		return (0);
5431 	case PCI_IVAR_ETHADDR:
5432 	case PCI_IVAR_SUBVENDOR:
5433 	case PCI_IVAR_SUBDEVICE:
5434 	case PCI_IVAR_VENDOR:
5435 	case PCI_IVAR_DEVICE:
5436 	case PCI_IVAR_DEVID:
5437 	case PCI_IVAR_CLASS:
5438 	case PCI_IVAR_SUBCLASS:
5439 	case PCI_IVAR_PROGIF:
5440 	case PCI_IVAR_REVID:
5441 	case PCI_IVAR_IRQ:
5442 	case PCI_IVAR_DOMAIN:
5443 	case PCI_IVAR_BUS:
5444 	case PCI_IVAR_SLOT:
5445 	case PCI_IVAR_FUNCTION:
5446 		return (EINVAL);	/* disallow for now */
5447 
5448 	default:
5449 		return (ENOENT);
5450 	}
5451 }
5452 
5453 #include "opt_ddb.h"
5454 #ifdef DDB
5455 #include <ddb/ddb.h>
5456 #include <sys/cons.h>
5457 
5458 /*
5459  * List resources based on pci map registers, used for within ddb
5460  */
5461 
5462 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5463 {
5464 	struct pci_devinfo *dinfo;
5465 	struct devlist *devlist_head;
5466 	struct pci_conf *p;
5467 	const char *name;
5468 	int i, error, none_count;
5469 
5470 	none_count = 0;
5471 	/* get the head of the device queue */
5472 	devlist_head = &pci_devq;
5473 
5474 	/*
5475 	 * Go through the list of devices and print out devices
5476 	 */
5477 	for (error = 0, i = 0,
5478 	     dinfo = STAILQ_FIRST(devlist_head);
5479 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5480 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5481 		/* Populate pd_name and pd_unit */
5482 		name = NULL;
5483 		if (dinfo->cfg.dev)
5484 			name = device_get_name(dinfo->cfg.dev);
5485 
5486 		p = &dinfo->conf;
5487 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5488 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5489 			(name && *name) ? name : "none",
5490 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5491 			none_count++,
5492 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5493 			p->pc_sel.pc_func, (p->pc_class << 16) |
5494 			(p->pc_subclass << 8) | p->pc_progif,
5495 			(p->pc_subdevice << 16) | p->pc_subvendor,
5496 			(p->pc_device << 16) | p->pc_vendor,
5497 			p->pc_revid, p->pc_hdr);
5498 	}
5499 }
5500 #endif /* DDB */
5501 
5502 struct resource *
5503 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5504     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5505     u_int flags)
5506 {
5507 	struct pci_devinfo *dinfo = device_get_ivars(child);
5508 	struct resource_list *rl = &dinfo->resources;
5509 	struct resource *res;
5510 	struct pci_map *pm;
5511 	uint16_t cmd;
5512 	pci_addr_t map, testval;
5513 	int mapsize;
5514 
5515 	res = NULL;
5516 
5517 	/* If rid is managed by EA, ignore it */
5518 	if (pci_ea_is_enabled(child, *rid))
5519 		goto out;
5520 
5521 	pm = pci_find_bar(child, *rid);
5522 	if (pm != NULL) {
5523 		/* This is a BAR that we failed to allocate earlier. */
5524 		mapsize = pm->pm_size;
5525 		map = pm->pm_value;
5526 	} else {
5527 		/*
5528 		 * Weed out the bogons, and figure out how large the
5529 		 * BAR/map is.  BARs that read back 0 here are bogus
5530 		 * and unimplemented.  Note: atapci in legacy mode are
5531 		 * special and handled elsewhere in the code.  If you
5532 		 * have a atapci device in legacy mode and it fails
5533 		 * here, that other code is broken.
5534 		 */
5535 		pci_read_bar(child, *rid, &map, &testval, NULL);
5536 
5537 		/*
5538 		 * Determine the size of the BAR and ignore BARs with a size
5539 		 * of 0.  Device ROM BARs use a different mask value.
5540 		 */
5541 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5542 			mapsize = pci_romsize(testval);
5543 		else
5544 			mapsize = pci_mapsize(testval);
5545 		if (mapsize == 0)
5546 			goto out;
5547 		pm = pci_add_bar(child, *rid, map, mapsize);
5548 	}
5549 
5550 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5551 		if (type != SYS_RES_MEMORY) {
5552 			if (bootverbose)
5553 				device_printf(dev,
5554 				    "child %s requested type %d for rid %#x,"
5555 				    " but the BAR says it is an memio\n",
5556 				    device_get_nameunit(child), type, *rid);
5557 			goto out;
5558 		}
5559 	} else {
5560 		if (type != SYS_RES_IOPORT) {
5561 			if (bootverbose)
5562 				device_printf(dev,
5563 				    "child %s requested type %d for rid %#x,"
5564 				    " but the BAR says it is an ioport\n",
5565 				    device_get_nameunit(child), type, *rid);
5566 			goto out;
5567 		}
5568 	}
5569 
5570 	/*
5571 	 * For real BARs, we need to override the size that
5572 	 * the driver requests, because that's what the BAR
5573 	 * actually uses and we would otherwise have a
5574 	 * situation where we might allocate the excess to
5575 	 * another driver, which won't work.
5576 	 */
5577 	count = ((pci_addr_t)1 << mapsize) * num;
5578 	if (RF_ALIGNMENT(flags) < mapsize)
5579 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5580 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5581 		flags |= RF_PREFETCHABLE;
5582 
5583 	/*
5584 	 * Allocate enough resource, and then write back the
5585 	 * appropriate BAR for that resource.
5586 	 */
5587 	resource_list_add(rl, type, *rid, start, end, count);
5588 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5589 	    count, flags & ~RF_ACTIVE);
5590 	if (res == NULL) {
5591 		resource_list_delete(rl, type, *rid);
5592 		device_printf(child,
5593 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5594 		    count, *rid, type, start, end);
5595 		goto out;
5596 	}
5597 	if (bootverbose)
5598 		device_printf(child,
5599 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5600 		    count, *rid, type, rman_get_start(res));
5601 
5602 	/* Disable decoding via the CMD register before updating the BAR */
5603 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5604 	pci_write_config(child, PCIR_COMMAND,
5605 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5606 
5607 	map = rman_get_start(res);
5608 	pci_write_bar(child, pm, map);
5609 
5610 	/* Restore the original value of the CMD register */
5611 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5612 out:
5613 	return (res);
5614 }
5615 
5616 struct resource *
5617 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5618     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5619     u_int flags)
5620 {
5621 	struct pci_devinfo *dinfo;
5622 	struct resource_list *rl;
5623 	struct resource_list_entry *rle;
5624 	struct resource *res;
5625 	pcicfgregs *cfg;
5626 
5627 	/*
5628 	 * Perform lazy resource allocation
5629 	 */
5630 	dinfo = device_get_ivars(child);
5631 	rl = &dinfo->resources;
5632 	cfg = &dinfo->cfg;
5633 	switch (type) {
5634 	case PCI_RES_BUS:
5635 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5636 		    flags));
5637 	case SYS_RES_IRQ:
5638 		/*
5639 		 * Can't alloc legacy interrupt once MSI messages have
5640 		 * been allocated.
5641 		 */
5642 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5643 		    cfg->msix.msix_alloc > 0))
5644 			return (NULL);
5645 
5646 		/*
5647 		 * If the child device doesn't have an interrupt
5648 		 * routed and is deserving of an interrupt, try to
5649 		 * assign it one.
5650 		 */
5651 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5652 		    (cfg->intpin != 0))
5653 			pci_assign_interrupt(dev, child, 0);
5654 		break;
5655 	case SYS_RES_IOPORT:
5656 	case SYS_RES_MEMORY:
5657 		/*
5658 		 * PCI-PCI bridge I/O window resources are not BARs.
5659 		 * For those allocations just pass the request up the
5660 		 * tree.
5661 		 */
5662 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5663 			switch (*rid) {
5664 			case PCIR_IOBASEL_1:
5665 			case PCIR_MEMBASE_1:
5666 			case PCIR_PMBASEL_1:
5667 				/*
5668 				 * XXX: Should we bother creating a resource
5669 				 * list entry?
5670 				 */
5671 				return (bus_generic_alloc_resource(dev, child,
5672 				    type, rid, start, end, count, flags));
5673 			}
5674 		}
5675 		/* Reserve resources for this BAR if needed. */
5676 		rle = resource_list_find(rl, type, *rid);
5677 		if (rle == NULL) {
5678 			res = pci_reserve_map(dev, child, type, rid, start, end,
5679 			    count, num, flags);
5680 			if (res == NULL)
5681 				return (NULL);
5682 		}
5683 	}
5684 	return (resource_list_alloc(rl, dev, child, type, rid,
5685 	    start, end, count, flags));
5686 }
5687 
5688 struct resource *
5689 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5690     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5691 {
5692 #ifdef PCI_IOV
5693 	struct pci_devinfo *dinfo;
5694 #endif
5695 
5696 	if (device_get_parent(child) != dev)
5697 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5698 		    type, rid, start, end, count, flags));
5699 
5700 #ifdef PCI_IOV
5701 	dinfo = device_get_ivars(child);
5702 	if (dinfo->cfg.flags & PCICFG_VF) {
5703 		switch (type) {
5704 		/* VFs can't have I/O BARs. */
5705 		case SYS_RES_IOPORT:
5706 			return (NULL);
5707 		case SYS_RES_MEMORY:
5708 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5709 			    start, end, count, flags));
5710 		}
5711 
5712 		/* Fall through for other types of resource allocations. */
5713 	}
5714 #endif
5715 
5716 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5717 	    count, 1, flags));
5718 }
5719 
5720 int
5721 pci_release_resource(device_t dev, device_t child, struct resource *r)
5722 {
5723 	struct pci_devinfo *dinfo;
5724 	struct resource_list *rl;
5725 	pcicfgregs *cfg __unused;
5726 
5727 	if (device_get_parent(child) != dev)
5728 		return (bus_generic_release_resource(dev, child, r));
5729 
5730 	dinfo = device_get_ivars(child);
5731 	cfg = &dinfo->cfg;
5732 
5733 #ifdef PCI_IOV
5734 	if (cfg->flags & PCICFG_VF) {
5735 		switch (rman_get_type(r)) {
5736 		/* VFs can't have I/O BARs. */
5737 		case SYS_RES_IOPORT:
5738 			return (EDOOFUS);
5739 		case SYS_RES_MEMORY:
5740 			return (pci_vf_release_mem_resource(dev, child, r));
5741 		}
5742 
5743 		/* Fall through for other types of resource allocations. */
5744 	}
5745 #endif
5746 
5747 	/*
5748 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5749 	 * those allocations just pass the request up the tree.
5750 	 */
5751 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5752 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5753 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5754 		switch (rman_get_rid(r)) {
5755 		case PCIR_IOBASEL_1:
5756 		case PCIR_MEMBASE_1:
5757 		case PCIR_PMBASEL_1:
5758 			return (bus_generic_release_resource(dev, child, r));
5759 		}
5760 	}
5761 
5762 	rl = &dinfo->resources;
5763 	return (resource_list_release(rl, dev, child, r));
5764 }
5765 
5766 int
5767 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5768 {
5769 	struct pci_devinfo *dinfo;
5770 	int error, rid, type;
5771 
5772 	if (device_get_parent(child) != dev)
5773 		return (bus_generic_activate_resource(dev, child, r));
5774 
5775 	dinfo = device_get_ivars(child);
5776 #ifdef PCI_IOV
5777 	if (dinfo->cfg.flags & PCICFG_VF) {
5778 		switch (rman_get_type(r)) {
5779 		/* VFs can't have I/O BARs. */
5780 		case SYS_RES_IOPORT:
5781 			error = EINVAL;
5782 			break;
5783 		case SYS_RES_MEMORY:
5784 			error = pci_vf_activate_mem_resource(dev, child, r);
5785 			break;
5786 		default:
5787 			error = bus_generic_activate_resource(dev, child, r);
5788 			break;
5789 		}
5790 	} else
5791 #endif
5792 		error = bus_generic_activate_resource(dev, child, r);
5793 	if (error)
5794 		return (error);
5795 
5796 	rid = rman_get_rid(r);
5797 	type = rman_get_type(r);
5798 
5799 	/* Device ROMs need their decoding explicitly enabled. */
5800 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5801 		pci_write_bar(child, pci_find_bar(child, rid),
5802 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5803 
5804 	/* Enable decoding in the command register when activating BARs. */
5805 	switch (type) {
5806 	case SYS_RES_IOPORT:
5807 	case SYS_RES_MEMORY:
5808 		error = PCI_ENABLE_IO(dev, child, type);
5809 		break;
5810 	}
5811 	return (error);
5812 }
5813 
5814 int
5815 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5816 {
5817 	struct pci_devinfo *dinfo;
5818 	int error, rid, type;
5819 
5820 	if (device_get_parent(child) != dev)
5821 		return (bus_generic_deactivate_resource(dev, child, r));
5822 
5823 	dinfo = device_get_ivars(child);
5824 #ifdef PCI_IOV
5825 	if (dinfo->cfg.flags & PCICFG_VF) {
5826 		switch (rman_get_type(r)) {
5827 		/* VFs can't have I/O BARs. */
5828 		case SYS_RES_IOPORT:
5829 			error = EINVAL;
5830 			break;
5831 		case SYS_RES_MEMORY:
5832 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5833 			break;
5834 		default:
5835 			error = bus_generic_deactivate_resource(dev, child, r);
5836 			break;
5837 		}
5838 	} else
5839 #endif
5840 		error = bus_generic_deactivate_resource(dev, child, r);
5841 	if (error)
5842 		return (error);
5843 
5844 	/* Disable decoding for device ROMs. */
5845 	rid = rman_get_rid(r);
5846 	type = rman_get_type(r);
5847 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5848 		pci_write_bar(child, pci_find_bar(child, rid),
5849 		    rman_get_start(r));
5850 	return (0);
5851 }
5852 
5853 int
5854 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5855     rman_res_t start, rman_res_t end)
5856 {
5857 #ifdef PCI_IOV
5858 	struct pci_devinfo *dinfo;
5859 
5860 	if (device_get_parent(child) != dev)
5861 		return (bus_generic_adjust_resource(dev, child, r, start,
5862 		    end));
5863 
5864 	dinfo = device_get_ivars(child);
5865 	if (dinfo->cfg.flags & PCICFG_VF) {
5866 		switch (rman_get_type(r)) {
5867 		/* VFs can't have I/O BARs. */
5868 		case SYS_RES_IOPORT:
5869 			return (EINVAL);
5870 		case SYS_RES_MEMORY:
5871 			return (pci_vf_adjust_mem_resource(dev, child, r,
5872 			    start, end));
5873 		}
5874 
5875 		/* Fall through for other types of resource allocations. */
5876 	}
5877 #endif
5878 
5879 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5880 }
5881 
5882 int
5883 pci_map_resource(device_t dev, device_t child, struct resource *r,
5884     struct resource_map_request *argsp, struct resource_map *map)
5885 {
5886 #ifdef PCI_IOV
5887 	struct pci_devinfo *dinfo;
5888 
5889 	if (device_get_parent(child) != dev)
5890 		return (bus_generic_map_resource(dev, child, r, argsp,
5891 		    map));
5892 
5893 	dinfo = device_get_ivars(child);
5894 	if (dinfo->cfg.flags & PCICFG_VF) {
5895 		switch (rman_get_type(r)) {
5896 		/* VFs can't have I/O BARs. */
5897 		case SYS_RES_IOPORT:
5898 			return (EINVAL);
5899 		case SYS_RES_MEMORY:
5900 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5901 			    map));
5902 		}
5903 
5904 		/* Fall through for other types of resource allocations. */
5905 	}
5906 #endif
5907 
5908 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5909 }
5910 
5911 int
5912 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5913     struct resource_map *map)
5914 {
5915 #ifdef PCI_IOV
5916 	struct pci_devinfo *dinfo;
5917 
5918 	if (device_get_parent(child) != dev)
5919 		return (bus_generic_unmap_resource(dev, child, r, map));
5920 
5921 	dinfo = device_get_ivars(child);
5922 	if (dinfo->cfg.flags & PCICFG_VF) {
5923 		switch (rman_get_type(r)) {
5924 		/* VFs can't have I/O BARs. */
5925 		case SYS_RES_IOPORT:
5926 			return (EINVAL);
5927 		case SYS_RES_MEMORY:
5928 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5929 		}
5930 
5931 		/* Fall through for other types of resource allocations. */
5932 	}
5933 #endif
5934 
5935 	return (bus_generic_unmap_resource(dev, child, r, map));
5936 }
5937 
5938 void
5939 pci_child_deleted(device_t dev, device_t child)
5940 {
5941 	struct resource_list_entry *rle;
5942 	struct resource_list *rl;
5943 	struct pci_devinfo *dinfo;
5944 
5945 	dinfo = device_get_ivars(child);
5946 	rl = &dinfo->resources;
5947 
5948 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5949 
5950 	/* Turn off access to resources we're about to free */
5951 	if (bus_child_present(child) != 0) {
5952 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5953 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5954 
5955 		pci_disable_busmaster(child);
5956 	}
5957 
5958 	/* Free all allocated resources */
5959 	STAILQ_FOREACH(rle, rl, link) {
5960 		if (rle->res) {
5961 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5962 			    resource_list_busy(rl, rle->type, rle->rid)) {
5963 				pci_printf(&dinfo->cfg,
5964 				    "Resource still owned, oops. "
5965 				    "(type=%d, rid=%d, addr=%lx)\n",
5966 				    rle->type, rle->rid,
5967 				    rman_get_start(rle->res));
5968 				bus_release_resource(child, rle->type, rle->rid,
5969 				    rle->res);
5970 			}
5971 			resource_list_unreserve(rl, dev, child, rle->type,
5972 			    rle->rid);
5973 		}
5974 	}
5975 	resource_list_free(rl);
5976 
5977 	pci_freecfg(dinfo);
5978 }
5979 
5980 void
5981 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5982 {
5983 	struct pci_devinfo *dinfo;
5984 	struct resource_list *rl;
5985 	struct resource_list_entry *rle;
5986 
5987 	if (device_get_parent(child) != dev)
5988 		return;
5989 
5990 	dinfo = device_get_ivars(child);
5991 	rl = &dinfo->resources;
5992 	rle = resource_list_find(rl, type, rid);
5993 	if (rle == NULL)
5994 		return;
5995 
5996 	if (rle->res) {
5997 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5998 		    resource_list_busy(rl, type, rid)) {
5999 			device_printf(dev, "delete_resource: "
6000 			    "Resource still owned by child, oops. "
6001 			    "(type=%d, rid=%d, addr=%jx)\n",
6002 			    type, rid, rman_get_start(rle->res));
6003 			return;
6004 		}
6005 		resource_list_unreserve(rl, dev, child, type, rid);
6006 	}
6007 	resource_list_delete(rl, type, rid);
6008 }
6009 
6010 struct resource_list *
6011 pci_get_resource_list (device_t dev, device_t child)
6012 {
6013 	struct pci_devinfo *dinfo = device_get_ivars(child);
6014 
6015 	return (&dinfo->resources);
6016 }
6017 
6018 #ifdef IOMMU
6019 bus_dma_tag_t
6020 pci_get_dma_tag(device_t bus, device_t dev)
6021 {
6022 	bus_dma_tag_t tag;
6023 	struct pci_softc *sc;
6024 
6025 	if (device_get_parent(dev) == bus) {
6026 		/* try iommu and return if it works */
6027 		tag = iommu_get_dma_tag(bus, dev);
6028 	} else
6029 		tag = NULL;
6030 	if (tag == NULL) {
6031 		sc = device_get_softc(bus);
6032 		tag = sc->sc_dma_tag;
6033 	}
6034 	return (tag);
6035 }
6036 #else
6037 bus_dma_tag_t
6038 pci_get_dma_tag(device_t bus, device_t dev)
6039 {
6040 	struct pci_softc *sc = device_get_softc(bus);
6041 
6042 	return (sc->sc_dma_tag);
6043 }
6044 #endif
6045 
6046 uint32_t
6047 pci_read_config_method(device_t dev, device_t child, int reg, int width)
6048 {
6049 	struct pci_devinfo *dinfo = device_get_ivars(child);
6050 	pcicfgregs *cfg = &dinfo->cfg;
6051 
6052 #ifdef PCI_IOV
6053 	/*
6054 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
6055 	 * emulate them here.
6056 	 */
6057 	if (cfg->flags & PCICFG_VF) {
6058 		if (reg == PCIR_VENDOR) {
6059 			switch (width) {
6060 			case 4:
6061 				return (cfg->device << 16 | cfg->vendor);
6062 			case 2:
6063 				return (cfg->vendor);
6064 			case 1:
6065 				return (cfg->vendor & 0xff);
6066 			default:
6067 				return (0xffffffff);
6068 			}
6069 		} else if (reg == PCIR_DEVICE) {
6070 			switch (width) {
6071 			/* Note that an unaligned 4-byte read is an error. */
6072 			case 2:
6073 				return (cfg->device);
6074 			case 1:
6075 				return (cfg->device & 0xff);
6076 			default:
6077 				return (0xffffffff);
6078 			}
6079 		}
6080 	}
6081 #endif
6082 
6083 	return (PCIB_READ_CONFIG(device_get_parent(dev),
6084 	    cfg->bus, cfg->slot, cfg->func, reg, width));
6085 }
6086 
6087 void
6088 pci_write_config_method(device_t dev, device_t child, int reg,
6089     uint32_t val, int width)
6090 {
6091 	struct pci_devinfo *dinfo = device_get_ivars(child);
6092 	pcicfgregs *cfg = &dinfo->cfg;
6093 
6094 	PCIB_WRITE_CONFIG(device_get_parent(dev),
6095 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6096 }
6097 
6098 int
6099 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6100 {
6101 
6102 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6103 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6104 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6105 	return (0);
6106 }
6107 
6108 int
6109 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6110 {
6111 	struct pci_devinfo *dinfo;
6112 	pcicfgregs *cfg;
6113 
6114 	dinfo = device_get_ivars(child);
6115 	cfg = &dinfo->cfg;
6116 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6117 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6118 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6119 	    cfg->progif);
6120 	return (0);
6121 }
6122 
6123 int
6124 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6125     struct sbuf *sb)
6126 {
6127 	device_t parent = device_get_parent(bus);
6128 	int rv;
6129 
6130 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6131 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6132 		if (rv == 0) {
6133 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6134 			    pci_get_function(child));
6135 		}
6136 		return (0);
6137 	}
6138 	return (bus_generic_get_device_path(bus, child, locator, sb));
6139 }
6140 
6141 int
6142 pci_assign_interrupt_method(device_t dev, device_t child)
6143 {
6144 	struct pci_devinfo *dinfo = device_get_ivars(child);
6145 	pcicfgregs *cfg = &dinfo->cfg;
6146 
6147 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6148 	    cfg->intpin));
6149 }
6150 
6151 static void
6152 pci_lookup(void *arg, const char *name, device_t *dev)
6153 {
6154 	long val;
6155 	char *end;
6156 	int domain, bus, slot, func;
6157 
6158 	if (*dev != NULL)
6159 		return;
6160 
6161 	/*
6162 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6163 	 * pciB:S:F.  In the latter case, the domain is assumed to
6164 	 * be zero.
6165 	 */
6166 	if (strncmp(name, "pci", 3) != 0)
6167 		return;
6168 	val = strtol(name + 3, &end, 10);
6169 	if (val < 0 || val > INT_MAX || *end != ':')
6170 		return;
6171 	domain = val;
6172 	val = strtol(end + 1, &end, 10);
6173 	if (val < 0 || val > INT_MAX || *end != ':')
6174 		return;
6175 	bus = val;
6176 	val = strtol(end + 1, &end, 10);
6177 	if (val < 0 || val > INT_MAX)
6178 		return;
6179 	slot = val;
6180 	if (*end == ':') {
6181 		val = strtol(end + 1, &end, 10);
6182 		if (val < 0 || val > INT_MAX || *end != '\0')
6183 			return;
6184 		func = val;
6185 	} else if (*end == '\0') {
6186 		func = slot;
6187 		slot = bus;
6188 		bus = domain;
6189 		domain = 0;
6190 	} else
6191 		return;
6192 
6193 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6194 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6195 		return;
6196 
6197 	*dev = pci_find_dbsf(domain, bus, slot, func);
6198 }
6199 
6200 static int
6201 pci_modevent(module_t mod, int what, void *arg)
6202 {
6203 	static struct cdev *pci_cdev;
6204 	static eventhandler_tag tag;
6205 
6206 	switch (what) {
6207 	case MOD_LOAD:
6208 		STAILQ_INIT(&pci_devq);
6209 		pci_generation = 0;
6210 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6211 		    "pci");
6212 		pci_load_vendor_data();
6213 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6214 		    1000);
6215 		break;
6216 
6217 	case MOD_UNLOAD:
6218 		if (tag != NULL)
6219 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6220 		destroy_dev(pci_cdev);
6221 		break;
6222 	}
6223 
6224 	return (0);
6225 }
6226 
6227 static void
6228 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6229 {
6230 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6231 	struct pcicfg_pcie *cfg;
6232 	int version, pos;
6233 
6234 	cfg = &dinfo->cfg.pcie;
6235 	pos = cfg->pcie_location;
6236 
6237 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6238 
6239 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6240 
6241 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6242 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6243 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6244 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6245 
6246 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6247 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6248 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6249 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6250 
6251 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6252 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6253 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6254 
6255 	if (version > 1) {
6256 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6257 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6258 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6259 	}
6260 #undef WREG
6261 }
6262 
6263 static void
6264 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6265 {
6266 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6267 	    dinfo->cfg.pcix.pcix_command,  2);
6268 }
6269 
6270 void
6271 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6272 {
6273 
6274 	/*
6275 	 * Restore the device to full power mode.  We must do this
6276 	 * before we restore the registers because moving from D3 to
6277 	 * D0 will cause the chip's BARs and some other registers to
6278 	 * be reset to some unknown power on reset values.  Cut down
6279 	 * the noise on boot by doing nothing if we are already in
6280 	 * state D0.
6281 	 */
6282 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6283 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6284 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6285 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6286 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6287 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6288 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6289 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6290 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6291 	case PCIM_HDRTYPE_NORMAL:
6292 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6293 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6294 		break;
6295 	case PCIM_HDRTYPE_BRIDGE:
6296 		pci_write_config(dev, PCIR_SECLAT_1,
6297 		    dinfo->cfg.bridge.br_seclat, 1);
6298 		pci_write_config(dev, PCIR_SUBBUS_1,
6299 		    dinfo->cfg.bridge.br_subbus, 1);
6300 		pci_write_config(dev, PCIR_SECBUS_1,
6301 		    dinfo->cfg.bridge.br_secbus, 1);
6302 		pci_write_config(dev, PCIR_PRIBUS_1,
6303 		    dinfo->cfg.bridge.br_pribus, 1);
6304 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6305 		    dinfo->cfg.bridge.br_control, 2);
6306 		break;
6307 	case PCIM_HDRTYPE_CARDBUS:
6308 		pci_write_config(dev, PCIR_SECLAT_2,
6309 		    dinfo->cfg.bridge.br_seclat, 1);
6310 		pci_write_config(dev, PCIR_SUBBUS_2,
6311 		    dinfo->cfg.bridge.br_subbus, 1);
6312 		pci_write_config(dev, PCIR_SECBUS_2,
6313 		    dinfo->cfg.bridge.br_secbus, 1);
6314 		pci_write_config(dev, PCIR_PRIBUS_2,
6315 		    dinfo->cfg.bridge.br_pribus, 1);
6316 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6317 		    dinfo->cfg.bridge.br_control, 2);
6318 		break;
6319 	}
6320 	pci_restore_bars(dev);
6321 
6322 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6323 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6324 
6325 	/*
6326 	 * Restore extended capabilities for PCI-Express and PCI-X
6327 	 */
6328 	if (dinfo->cfg.pcie.pcie_location != 0)
6329 		pci_cfg_restore_pcie(dev, dinfo);
6330 	if (dinfo->cfg.pcix.pcix_location != 0)
6331 		pci_cfg_restore_pcix(dev, dinfo);
6332 
6333 	/* Restore MSI and MSI-X configurations if they are present. */
6334 	if (dinfo->cfg.msi.msi_location != 0)
6335 		pci_resume_msi(dev);
6336 	if (dinfo->cfg.msix.msix_location != 0)
6337 		pci_resume_msix(dev);
6338 
6339 #ifdef PCI_IOV
6340 	if (dinfo->cfg.iov != NULL)
6341 		pci_iov_cfg_restore(dev, dinfo);
6342 #endif
6343 }
6344 
6345 static void
6346 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6347 {
6348 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6349 	struct pcicfg_pcie *cfg;
6350 	int version, pos;
6351 
6352 	cfg = &dinfo->cfg.pcie;
6353 	pos = cfg->pcie_location;
6354 
6355 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6356 
6357 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6358 
6359 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6360 
6361 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6362 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6363 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6364 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6365 
6366 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6367 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6368 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6369 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6370 
6371 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6372 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6373 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6374 
6375 	if (version > 1) {
6376 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6377 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6378 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6379 	}
6380 #undef RREG
6381 }
6382 
6383 static void
6384 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6385 {
6386 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6387 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6388 }
6389 
6390 void
6391 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6392 {
6393 	uint32_t cls;
6394 	int ps;
6395 
6396 	/*
6397 	 * Some drivers apparently write to these registers w/o updating our
6398 	 * cached copy.  No harm happens if we update the copy, so do so here
6399 	 * so we can restore them.  The COMMAND register is modified by the
6400 	 * bus w/o updating the cache.  This should represent the normally
6401 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6402 	 */
6403 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6404 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6405 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6406 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6407 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6408 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6409 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6410 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6411 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6412 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6413 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6414 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6415 	case PCIM_HDRTYPE_NORMAL:
6416 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6417 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6418 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6419 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6420 		break;
6421 	case PCIM_HDRTYPE_BRIDGE:
6422 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6423 		    PCIR_SECLAT_1, 1);
6424 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6425 		    PCIR_SUBBUS_1, 1);
6426 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6427 		    PCIR_SECBUS_1, 1);
6428 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6429 		    PCIR_PRIBUS_1, 1);
6430 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6431 		    PCIR_BRIDGECTL_1, 2);
6432 		break;
6433 	case PCIM_HDRTYPE_CARDBUS:
6434 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6435 		    PCIR_SECLAT_2, 1);
6436 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6437 		    PCIR_SUBBUS_2, 1);
6438 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6439 		    PCIR_SECBUS_2, 1);
6440 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6441 		    PCIR_PRIBUS_2, 1);
6442 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6443 		    PCIR_BRIDGECTL_2, 2);
6444 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6445 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6446 		break;
6447 	}
6448 
6449 	if (dinfo->cfg.pcie.pcie_location != 0)
6450 		pci_cfg_save_pcie(dev, dinfo);
6451 
6452 	if (dinfo->cfg.pcix.pcix_location != 0)
6453 		pci_cfg_save_pcix(dev, dinfo);
6454 
6455 #ifdef PCI_IOV
6456 	if (dinfo->cfg.iov != NULL)
6457 		pci_iov_cfg_save(dev, dinfo);
6458 #endif
6459 
6460 	/*
6461 	 * don't set the state for display devices, base peripherals and
6462 	 * memory devices since bad things happen when they are powered down.
6463 	 * We should (a) have drivers that can easily detach and (b) use
6464 	 * generic drivers for these devices so that some device actually
6465 	 * attaches.  We need to make sure that when we implement (a) we don't
6466 	 * power the device down on a reattach.
6467 	 */
6468 	cls = pci_get_class(dev);
6469 	if (!setstate)
6470 		return;
6471 	switch (pci_do_power_nodriver)
6472 	{
6473 		case 0:		/* NO powerdown at all */
6474 			return;
6475 		case 1:		/* Conservative about what to power down */
6476 			if (cls == PCIC_STORAGE)
6477 				return;
6478 			/*FALLTHROUGH*/
6479 		case 2:		/* Aggressive about what to power down */
6480 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6481 			    cls == PCIC_BASEPERIPH)
6482 				return;
6483 			/*FALLTHROUGH*/
6484 		case 3:		/* Power down everything */
6485 			break;
6486 	}
6487 	/*
6488 	 * PCI spec says we can only go into D3 state from D0 state.
6489 	 * Transition from D[12] into D0 before going to D3 state.
6490 	 */
6491 	ps = pci_get_powerstate(dev);
6492 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6493 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6494 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6495 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6496 }
6497 
6498 /* Wrapper APIs suitable for device driver use. */
6499 void
6500 pci_save_state(device_t dev)
6501 {
6502 	struct pci_devinfo *dinfo;
6503 
6504 	dinfo = device_get_ivars(dev);
6505 	pci_cfg_save(dev, dinfo, 0);
6506 }
6507 
6508 void
6509 pci_restore_state(device_t dev)
6510 {
6511 	struct pci_devinfo *dinfo;
6512 
6513 	dinfo = device_get_ivars(dev);
6514 	pci_cfg_restore(dev, dinfo);
6515 }
6516 
6517 static int
6518 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6519     uintptr_t *id)
6520 {
6521 
6522 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6523 }
6524 
6525 /* Find the upstream port of a given PCI device in a root complex. */
6526 device_t
6527 pci_find_pcie_root_port(device_t dev)
6528 {
6529 	struct pci_devinfo *dinfo;
6530 	devclass_t pci_class;
6531 	device_t pcib, bus;
6532 
6533 	pci_class = devclass_find("pci");
6534 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6535 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6536 
6537 	/*
6538 	 * Walk the bridge hierarchy until we find a PCI-e root
6539 	 * port or a non-PCI device.
6540 	 */
6541 	for (;;) {
6542 		bus = device_get_parent(dev);
6543 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6544 		    device_get_nameunit(dev)));
6545 
6546 		pcib = device_get_parent(bus);
6547 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6548 		    device_get_nameunit(bus)));
6549 
6550 		/*
6551 		 * pcib's parent must be a PCI bus for this to be a
6552 		 * PCI-PCI bridge.
6553 		 */
6554 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6555 			return (NULL);
6556 
6557 		dinfo = device_get_ivars(pcib);
6558 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6559 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6560 			return (pcib);
6561 
6562 		dev = pcib;
6563 	}
6564 }
6565 
6566 /*
6567  * Wait for pending transactions to complete on a PCI-express function.
6568  *
6569  * The maximum delay is specified in milliseconds in max_delay.  Note
6570  * that this function may sleep.
6571  *
6572  * Returns true if the function is idle and false if the timeout is
6573  * exceeded.  If dev is not a PCI-express function, this returns true.
6574  */
6575 bool
6576 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6577 {
6578 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6579 	uint16_t sta;
6580 	int cap;
6581 
6582 	cap = dinfo->cfg.pcie.pcie_location;
6583 	if (cap == 0)
6584 		return (true);
6585 
6586 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6587 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6588 		if (max_delay == 0)
6589 			return (false);
6590 
6591 		/* Poll once every 100 milliseconds up to the timeout. */
6592 		if (max_delay > 100) {
6593 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6594 			max_delay -= 100;
6595 		} else {
6596 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6597 			    C_HARDCLOCK);
6598 			max_delay = 0;
6599 		}
6600 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6601 	}
6602 
6603 	return (true);
6604 }
6605 
6606 /*
6607  * Determine the maximum Completion Timeout in microseconds.
6608  *
6609  * For non-PCI-express functions this returns 0.
6610  */
6611 int
6612 pcie_get_max_completion_timeout(device_t dev)
6613 {
6614 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6615 	int cap;
6616 
6617 	cap = dinfo->cfg.pcie.pcie_location;
6618 	if (cap == 0)
6619 		return (0);
6620 
6621 	/*
6622 	 * Functions using the 1.x spec use the default timeout range of
6623 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6624 	 * support programmable timeouts also use this range.
6625 	 */
6626 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6627 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6628 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6629 		return (50 * 1000);
6630 
6631 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6632 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6633 	case PCIEM_CTL2_COMP_TIMO_100US:
6634 		return (100);
6635 	case PCIEM_CTL2_COMP_TIMO_10MS:
6636 		return (10 * 1000);
6637 	case PCIEM_CTL2_COMP_TIMO_55MS:
6638 		return (55 * 1000);
6639 	case PCIEM_CTL2_COMP_TIMO_210MS:
6640 		return (210 * 1000);
6641 	case PCIEM_CTL2_COMP_TIMO_900MS:
6642 		return (900 * 1000);
6643 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6644 		return (3500 * 1000);
6645 	case PCIEM_CTL2_COMP_TIMO_13S:
6646 		return (13 * 1000 * 1000);
6647 	case PCIEM_CTL2_COMP_TIMO_64S:
6648 		return (64 * 1000 * 1000);
6649 	default:
6650 		return (50 * 1000);
6651 	}
6652 }
6653 
6654 void
6655 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6656 {
6657 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6658 	const char *s;
6659 	int aer;
6660 	uint32_t r, r1;
6661 	uint16_t rs;
6662 
6663 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6664 		s = "Correctable";
6665 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6666 		s = "Uncorrectable (Non-Fatal)";
6667 	else
6668 		s = "Uncorrectable (Fatal)";
6669 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6670 	if (aerp) {
6671 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6672 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6673 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6674 		} else {
6675 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6676 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6677 		}
6678 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6679 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6680 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6681 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6682 			printf(" severity 0x%08x first %d\n",
6683 			    r, rs & 0x1f);
6684 		} else
6685 			printf("\n");
6686 	}
6687 
6688 	/* As kind of recovery just report and clear the error statuses. */
6689 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6690 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6691 		if (r != 0) {
6692 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6693 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6694 		}
6695 
6696 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6697 		if (r != 0) {
6698 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6699 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6700 		}
6701 	}
6702 	if (dinfo->cfg.pcie.pcie_location != 0) {
6703 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6704 		    PCIER_DEVICE_STA, 2);
6705 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6706 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6707 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6708 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6709 			    PCIER_DEVICE_STA, rs, 2);
6710 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6711 		}
6712 	}
6713 }
6714 
6715 /*
6716  * Perform a Function Level Reset (FLR) on a device.
6717  *
6718  * This function first waits for any pending transactions to complete
6719  * within the timeout specified by max_delay.  If transactions are
6720  * still pending, the function will return false without attempting a
6721  * reset.
6722  *
6723  * If dev is not a PCI-express function or does not support FLR, this
6724  * function returns false.
6725  *
6726  * Note that no registers are saved or restored.  The caller is
6727  * responsible for saving and restoring any registers including
6728  * PCI-standard registers via pci_save_state() and
6729  * pci_restore_state().
6730  */
6731 bool
6732 pcie_flr(device_t dev, u_int max_delay, bool force)
6733 {
6734 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6735 	uint16_t cmd, ctl;
6736 	int compl_delay;
6737 	int cap;
6738 
6739 	cap = dinfo->cfg.pcie.pcie_location;
6740 	if (cap == 0)
6741 		return (false);
6742 
6743 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6744 		return (false);
6745 
6746 	/*
6747 	 * Disable busmastering to prevent generation of new
6748 	 * transactions while waiting for the device to go idle.  If
6749 	 * the idle timeout fails, the command register is restored
6750 	 * which will re-enable busmastering.
6751 	 */
6752 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6753 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6754 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6755 		if (!force) {
6756 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6757 			return (false);
6758 		}
6759 		pci_printf(&dinfo->cfg,
6760 		    "Resetting with transactions pending after %d ms\n",
6761 		    max_delay);
6762 
6763 		/*
6764 		 * Extend the post-FLR delay to cover the maximum
6765 		 * Completion Timeout delay of anything in flight
6766 		 * during the FLR delay.  Enforce a minimum delay of
6767 		 * at least 10ms.
6768 		 */
6769 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6770 		if (compl_delay < 10)
6771 			compl_delay = 10;
6772 	} else
6773 		compl_delay = 0;
6774 
6775 	/* Initiate the reset. */
6776 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6777 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6778 	    PCIEM_CTL_INITIATE_FLR, 2);
6779 
6780 	/* Wait for 100ms. */
6781 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6782 
6783 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6784 	    PCIEM_STA_TRANSACTION_PND)
6785 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6786 	return (true);
6787 }
6788 
6789 /*
6790  * Attempt a power-management reset by cycling the device in/out of D3
6791  * state.  PCI spec says we can only go into D3 state from D0 state.
6792  * Transition from D[12] into D0 before going to D3 state.
6793  */
6794 int
6795 pci_power_reset(device_t dev)
6796 {
6797 	int ps;
6798 
6799 	ps = pci_get_powerstate(dev);
6800 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6801 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6802 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6803 	pci_set_powerstate(dev, ps);
6804 	return (0);
6805 }
6806 
6807 /*
6808  * Try link drop and retrain of the downstream port of upstream
6809  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6810  * cause Conventional Hot reset of the device in the slot.
6811  * Alternative, for PCIe, could be the secondary bus reset initiatied
6812  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6813  */
6814 int
6815 pcie_link_reset(device_t port, int pcie_location)
6816 {
6817 	uint16_t v;
6818 
6819 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6820 	v |= PCIEM_LINK_CTL_LINK_DIS;
6821 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6822 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6823 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6824 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6825 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6826 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6827 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6828 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6829 }
6830 
6831 static int
6832 pci_reset_post(device_t dev, device_t child)
6833 {
6834 
6835 	if (dev == device_get_parent(child))
6836 		pci_restore_state(child);
6837 	return (0);
6838 }
6839 
6840 static int
6841 pci_reset_prepare(device_t dev, device_t child)
6842 {
6843 
6844 	if (dev == device_get_parent(child))
6845 		pci_save_state(child);
6846 	return (0);
6847 }
6848 
6849 static int
6850 pci_reset_child(device_t dev, device_t child, int flags)
6851 {
6852 	int error;
6853 
6854 	if (dev == NULL || device_get_parent(child) != dev)
6855 		return (0);
6856 	if ((flags & DEVF_RESET_DETACH) != 0) {
6857 		error = device_get_state(child) == DS_ATTACHED ?
6858 		    device_detach(child) : 0;
6859 	} else {
6860 		error = BUS_SUSPEND_CHILD(dev, child);
6861 	}
6862 	if (error == 0) {
6863 		if (!pcie_flr(child, 1000, false)) {
6864 			error = BUS_RESET_PREPARE(dev, child);
6865 			if (error == 0)
6866 				pci_power_reset(child);
6867 			BUS_RESET_POST(dev, child);
6868 		}
6869 		if ((flags & DEVF_RESET_DETACH) != 0)
6870 			device_probe_and_attach(child);
6871 		else
6872 			BUS_RESUME_CHILD(dev, child);
6873 	}
6874 	return (error);
6875 }
6876 
6877 const struct pci_device_table *
6878 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6879 {
6880 	bool match;
6881 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6882 
6883 	vendor = pci_get_vendor(child);
6884 	device = pci_get_device(child);
6885 	subvendor = pci_get_subvendor(child);
6886 	subdevice = pci_get_subdevice(child);
6887 	class = pci_get_class(child);
6888 	subclass = pci_get_subclass(child);
6889 	revid = pci_get_revid(child);
6890 	while (nelt-- > 0) {
6891 		match = true;
6892 		if (id->match_flag_vendor)
6893 			match &= vendor == id->vendor;
6894 		if (id->match_flag_device)
6895 			match &= device == id->device;
6896 		if (id->match_flag_subvendor)
6897 			match &= subvendor == id->subvendor;
6898 		if (id->match_flag_subdevice)
6899 			match &= subdevice == id->subdevice;
6900 		if (id->match_flag_class)
6901 			match &= class == id->class_id;
6902 		if (id->match_flag_subclass)
6903 			match &= subclass == id->subclass;
6904 		if (id->match_flag_revid)
6905 			match &= revid == id->revid;
6906 		if (match)
6907 			return (id);
6908 		id++;
6909 	}
6910 	return (NULL);
6911 }
6912 
6913 static void
6914 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6915 {
6916 	const char *dev_name;
6917 	device_t dev;
6918 
6919 	dev = dinfo->cfg.dev;
6920 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6921 	    dinfo->cfg.slot, dinfo->cfg.func);
6922 	dev_name = device_get_name(dev);
6923 	if (dev_name != NULL)
6924 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6925 }
6926 
6927 void
6928 pci_print_faulted_dev(void)
6929 {
6930 	struct pci_devinfo *dinfo;
6931 	device_t dev;
6932 	int aer, i;
6933 	uint32_t r1, r2;
6934 	uint16_t status;
6935 
6936 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6937 		dev = dinfo->cfg.dev;
6938 		status = pci_read_config(dev, PCIR_STATUS, 2);
6939 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6940 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6941 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6942 		if (status != 0) {
6943 			pci_print_faulted_dev_name(dinfo);
6944 			printf(" error 0x%04x\n", status);
6945 		}
6946 		if (dinfo->cfg.pcie.pcie_location != 0) {
6947 			status = pci_read_config(dev,
6948 			    dinfo->cfg.pcie.pcie_location +
6949 			    PCIER_DEVICE_STA, 2);
6950 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6951 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6952 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6953 				pci_print_faulted_dev_name(dinfo);
6954 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6955 				    pci_read_config(dev,
6956 				    dinfo->cfg.pcie.pcie_location +
6957 				    PCIER_DEVICE_CTL, 2),
6958 				    status);
6959 			}
6960 		}
6961 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6962 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6963 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6964 			if (r1 != 0 || r2 != 0) {
6965 				pci_print_faulted_dev_name(dinfo);
6966 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6967 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6968 				    r1, pci_read_config(dev, aer +
6969 				    PCIR_AER_UC_MASK, 4),
6970 				    pci_read_config(dev, aer +
6971 				    PCIR_AER_UC_SEVERITY, 4),
6972 				    r2, pci_read_config(dev, aer +
6973 				    PCIR_AER_COR_MASK, 4),
6974 				    pci_read_config(dev, aer +
6975 				    PCIR_AER_CAP_CONTROL, 4));
6976 				for (i = 0; i < 4; i++) {
6977 					r1 = pci_read_config(dev, aer +
6978 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6979 					printf("    HL%d: 0x%08x\n", i, r1);
6980 				}
6981 			}
6982 		}
6983 	}
6984 }
6985 
6986 #ifdef DDB
6987 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6988 {
6989 
6990 	pci_print_faulted_dev();
6991 }
6992 
6993 static void
6994 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6995 {
6996 	device_t dev;
6997 	int aer;
6998 	uint32_t r;
6999 
7000 	dev = dinfo->cfg.dev;
7001 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
7002 	    PCIER_DEVICE_STA, 2);
7003 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
7004 	    PCIER_DEVICE_STA, r, 2);
7005 
7006 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
7007 		return;
7008 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
7009 	if (r != 0)
7010 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
7011 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
7012 	if (r != 0)
7013 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
7014 }
7015 
7016 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
7017 {
7018 	struct pci_devinfo *dinfo;
7019 	device_t dev;
7020 	uint16_t status, status1;
7021 
7022 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
7023 		dev = dinfo->cfg.dev;
7024 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
7025 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
7026 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
7027 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
7028 		if (status1 != 0) {
7029 			status &= ~status1;
7030 			pci_write_config(dev, PCIR_STATUS, status, 2);
7031 		}
7032 		if (dinfo->cfg.pcie.pcie_location != 0)
7033 			db_clear_pcie_errors(dinfo);
7034 	}
7035 }
7036 #endif
7037