xref: /src/sys/dev/pci/pci.c (revision 7e7a1b61531a29b4a0a5cdac66b96f420e6c66e4)
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 "pcib_if.h"
85 #include "pci_if.h"
86 
87 #define	PCIR_IS_BIOS(cfg, reg)						\
88 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
89 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
90 
91 static device_probe_t	pci_probe;
92 
93 static bus_reset_post_t pci_reset_post;
94 static bus_reset_prepare_t pci_reset_prepare;
95 static bus_reset_child_t pci_reset_child;
96 static bus_hint_device_unit_t pci_hint_device_unit;
97 static bus_remap_intr_t pci_remap_intr_method;
98 
99 static pci_get_id_t	pci_get_id_method;
100 
101 static int		pci_has_quirk(uint32_t devid, int quirk);
102 static pci_addr_t	pci_mapbase(uint64_t mapreg);
103 static const char	*pci_maptype(uint64_t mapreg);
104 static int		pci_maprange(uint64_t mapreg);
105 static pci_addr_t	pci_rombase(uint64_t mapreg);
106 static int		pci_romsize(uint64_t testval);
107 static void		pci_fixancient(pcicfgregs *cfg);
108 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
109 
110 static int		pci_porten(device_t dev);
111 static int		pci_memen(device_t dev);
112 static void		pci_assign_interrupt(device_t bus, device_t dev,
113 			    int force_route);
114 static int		pci_add_map(device_t bus, device_t dev, int reg,
115 			    struct resource_list *rl, int force, int prefetch);
116 static void		pci_load_vendor_data(void);
117 static int		pci_describe_parse_line(char **ptr, int *vendor,
118 			    int *device, char **desc);
119 static char		*pci_describe_device(device_t dev);
120 static int		pci_modevent(module_t mod, int what, void *arg);
121 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
122 			    pcicfgregs *cfg);
123 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
124 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
125 			    int reg, uint32_t *data);
126 #if 0
127 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
128 			    int reg, uint32_t data);
129 #endif
130 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
131 static void		pci_mask_msix(device_t dev, u_int index);
132 static void		pci_unmask_msix(device_t dev, u_int index);
133 static int		pci_msi_blacklisted(void);
134 static int		pci_msix_blacklisted(void);
135 static void		pci_resume_msi(device_t dev);
136 static void		pci_resume_msix(device_t dev);
137 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
138     int b, int s, int f, uint16_t vid, uint16_t did);
139 
140 static device_method_t pci_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,		pci_probe),
143 	DEVMETHOD(device_attach,	pci_attach),
144 	DEVMETHOD(device_detach,	pci_detach),
145 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,	bus_generic_suspend),
147 	DEVMETHOD(device_resume,	pci_resume),
148 
149 	/* Bus interface */
150 	DEVMETHOD(bus_print_child,	pci_print_child),
151 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
152 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
153 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
154 	DEVMETHOD(bus_driver_added,	pci_driver_added),
155 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
156 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
157 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
158 	DEVMETHOD(bus_reset_post,	pci_reset_post),
159 	DEVMETHOD(bus_reset_child,	pci_reset_child),
160 
161 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
162 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
163 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
164 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
165 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
166 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
167 	DEVMETHOD(bus_adjust_resource,	pci_adjust_resource),
168 	DEVMETHOD(bus_release_resource,	pci_release_resource),
169 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
170 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
171 	DEVMETHOD(bus_map_resource,	pci_map_resource),
172 	DEVMETHOD(bus_unmap_resource,	pci_unmap_resource),
173 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
174 	DEVMETHOD(bus_child_detached,	pci_child_detached),
175 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
176 	DEVMETHOD(bus_child_location,	pci_child_location_method),
177 	DEVMETHOD(bus_get_device_path,	pci_get_device_path_method),
178 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
179 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
180 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
181 	DEVMETHOD(bus_resume_child,	pci_resume_child),
182 	DEVMETHOD(bus_rescan,		pci_rescan_method),
183 
184 	/* PCI interface */
185 	DEVMETHOD(pci_read_config,	pci_read_config_method),
186 	DEVMETHOD(pci_write_config,	pci_write_config_method),
187 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
188 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
189 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
190 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
191 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
192 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
193 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
194 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
195 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
196 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
197 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
198 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
199 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
200 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
201 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
202 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
203 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
204 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
205 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
206 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
207 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
208 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
209 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
210 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
211 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
212 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
213 	DEVMETHOD(pci_get_id,		pci_get_id_method),
214 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
215 	DEVMETHOD(pci_child_added,	pci_child_added_method),
216 #ifdef PCI_IOV
217 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
218 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
219 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
220 #endif
221 
222 	DEVMETHOD_END
223 };
224 
225 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
226 
227 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_modevent, NULL, BUS_PASS_BUS);
228 MODULE_VERSION(pci, 1);
229 
230 static char	*pci_vendordata;
231 static size_t	pci_vendordata_size;
232 
233 struct pci_quirk {
234 	uint32_t devid;	/* Vendor/device of the card */
235 	int	type;
236 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
237 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
238 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
239 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
240 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
241 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
243 #define	PCI_QUIRK_DISABLE_FLR	8 /* Function-Level Reset (FLR) not working. */
244 	int	arg1;
245 	int	arg2;
246 };
247 
248 static const struct pci_quirk pci_quirks[] = {
249 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
250 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
252 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
253 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
254 
255 	/*
256 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
257 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
258 	 */
259 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
260 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
261 
262 	/*
263 	 * MSI doesn't work on earlier Intel chipsets including
264 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
265 	 */
266 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
273 
274 	/*
275 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
276 	 * bridge.
277 	 */
278 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
279 
280 	/*
281 	 * Some virtualization environments emulate an older chipset
282 	 * but support MSI just fine.  QEMU uses the Intel 82440.
283 	 */
284 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
285 
286 	/*
287 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
288 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
289 	 * It prevents us from attaching hpet(4) when the bit is unset.
290 	 * Note this quirk only affects SB600 revision A13 and earlier.
291 	 * For SB600 A21 and later, firmware must set the bit to hide it.
292 	 * For SB700 and later, it is unused and hardcoded to zero.
293 	 */
294 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
295 
296 	/*
297 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
298 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
299 	 * of the command register is set.
300 	 */
301 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
306 
307 	/*
308 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
309 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
310 	 */
311 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
312 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
313 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
314 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
315 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
316 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
317 
318 	/*
319 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
320 	 * expected place.
321 	 */
322 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
323 
324 	/*
325 	 * With some MediaTek mt76 WiFi FLR does not work despite advertised.
326 	 */
327 	{ 0x061614c3, PCI_QUIRK_DISABLE_FLR,	0,	0 }, /* mt76 7922 */
328 
329 	/* end of table */
330 	{ 0 }
331 };
332 
333 /* map register information */
334 #define	PCI_MAPMEM	0x01	/* memory map */
335 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
336 #define	PCI_MAPPORT	0x04	/* port map */
337 
338 struct devlist pci_devq;
339 uint32_t pci_generation;
340 uint32_t pci_numdevs = 0;
341 static int pcie_chipset, pcix_chipset;
342 
343 /* sysctl vars */
344 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
345     "PCI bus tuning parameters");
346 
347 static int pci_enable_io_modes = 1;
348 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
349     &pci_enable_io_modes, 1,
350     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
351     " enable these bits correctly.  We'd like to do this all the time, but"
352     " there are some peripherals that this causes problems with.");
353 
354 static int pci_do_realloc_bars = 1;
355 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
356     &pci_do_realloc_bars, 0,
357     "Attempt to allocate a new range for any BARs whose original "
358     "firmware-assigned ranges fail to allocate during the initial device scan.");
359 
360 static int pci_do_power_nodriver = 0;
361 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
362     &pci_do_power_nodriver, 0,
363     "Place a function into D3 state when no driver attaches to it.  0 means"
364     " disable.  1 means conservatively place function into D3 state.  2 means"
365     " aggressively place function into D3 state.  3 means put absolutely"
366     " everything in D3 state.");
367 
368 int pci_do_power_resume = 1;
369 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
370     &pci_do_power_resume, 1,
371   "Transition from D3 -> D0 on resume.");
372 
373 int pci_do_power_suspend = 1;
374 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
375     &pci_do_power_suspend, 1,
376   "Transition from D0 -> D3 on suspend.");
377 
378 static int pci_do_msi = 1;
379 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
380     "Enable support for MSI interrupts");
381 
382 static int pci_do_msix = 1;
383 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
384     "Enable support for MSI-X interrupts");
385 
386 static int pci_msix_rewrite_table = 0;
387 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
388     &pci_msix_rewrite_table, 0,
389     "Rewrite entire MSI-X table when updating MSI-X entries");
390 
391 static int pci_honor_msi_blacklist = 1;
392 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
393     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
394 
395 #if defined(__i386__) || defined(__amd64__)
396 static int pci_usb_takeover = 1;
397 #else
398 static int pci_usb_takeover = 0;
399 #endif
400 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
401     &pci_usb_takeover, 1,
402     "Enable early takeover of USB controllers. Disable this if you depend on"
403     " BIOS emulation of USB devices, that is you use USB devices (like"
404     " keyboard or mouse) but do not load USB drivers");
405 
406 static int pci_clear_bars;
407 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
408     "Ignore firmware-assigned resources for BARs.");
409 
410 static int pci_clear_buses;
411 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
412     "Ignore firmware-assigned bus numbers.");
413 
414 static int pci_enable_ari = 1;
415 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
416     0, "Enable support for PCIe Alternative RID Interpretation");
417 
418 /*
419  * Some x86 firmware only enables PCIe hotplug if we claim to support aspm,
420  * however enabling it breaks some arm64 firmware as it powers off devices.
421  */
422 #if defined(__i386__) || defined(__amd64__)
423 int pci_enable_aspm = 1;
424 #else
425 int pci_enable_aspm = 0;
426 #endif
427 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
428     0, "Enable support for PCIe Active State Power Management");
429 
430 static int pci_clear_aer_on_attach = 0;
431 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
432     &pci_clear_aer_on_attach, 0,
433     "Clear port and device AER state on driver attach");
434 
435 static bool pci_enable_mps_tune = true;
436 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
437     &pci_enable_mps_tune, 1,
438     "Enable tuning of MPS(maximum payload size)." );
439 
440 static bool pci_intx_reroute = true;
441 SYSCTL_BOOL(_hw_pci, OID_AUTO, intx_reroute, CTLFLAG_RWTUN,
442     &pci_intx_reroute, 0, "Re-route INTx interrupts when scanning devices");
443 
444 static int
pci_has_quirk(uint32_t devid,int quirk)445 pci_has_quirk(uint32_t devid, int quirk)
446 {
447 	const struct pci_quirk *q;
448 
449 	for (q = &pci_quirks[0]; q->devid; q++) {
450 		if (q->devid == devid && q->type == quirk)
451 			return (1);
452 	}
453 	return (0);
454 }
455 
456 /* Find a device_t by bus/slot/function in domain 0 */
457 
458 device_t
pci_find_bsf(uint8_t bus,uint8_t slot,uint8_t func)459 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
460 {
461 
462 	return (pci_find_dbsf(0, bus, slot, func));
463 }
464 
465 /* Find a device_t by domain/bus/slot/function */
466 
467 device_t
pci_find_dbsf(uint32_t domain,uint8_t bus,uint8_t slot,uint8_t func)468 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
469 {
470 	struct pci_devinfo *dinfo = NULL;
471 
472 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
473 		if ((dinfo->cfg.domain == domain) &&
474 		    (dinfo->cfg.bus == bus) &&
475 		    (dinfo->cfg.slot == slot) &&
476 		    (dinfo->cfg.func == func)) {
477 			break;
478 		}
479 	}
480 
481 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
482 }
483 
484 /* Find a device_t by vendor/device ID */
485 
486 device_t
pci_find_device(uint16_t vendor,uint16_t device)487 pci_find_device(uint16_t vendor, uint16_t device)
488 {
489 	struct pci_devinfo *dinfo;
490 
491 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
492 		if ((dinfo->cfg.vendor == vendor) &&
493 		    (dinfo->cfg.device == device)) {
494 			return (dinfo->cfg.dev);
495 		}
496 	}
497 
498 	return (NULL);
499 }
500 
501 device_t
pci_find_class(uint8_t class,uint8_t subclass)502 pci_find_class(uint8_t class, uint8_t subclass)
503 {
504 	struct pci_devinfo *dinfo;
505 
506 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
507 		if (dinfo->cfg.baseclass == class &&
508 		    dinfo->cfg.subclass == subclass) {
509 			return (dinfo->cfg.dev);
510 		}
511 	}
512 
513 	return (NULL);
514 }
515 
516 device_t
pci_find_class_from(uint8_t class,uint8_t subclass,device_t from)517 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
518 {
519 	struct pci_devinfo *dinfo;
520 	bool found = false;
521 
522 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
523 		if (from != NULL && found == false) {
524 			if (from != dinfo->cfg.dev)
525 				continue;
526 			found = true;
527 			continue;
528 		}
529 		if (dinfo->cfg.baseclass == class &&
530 		    dinfo->cfg.subclass == subclass) {
531 			return (dinfo->cfg.dev);
532 		}
533 	}
534 
535 	return (NULL);
536 }
537 
538 device_t
pci_find_base_class_from(uint8_t class,device_t from)539 pci_find_base_class_from(uint8_t class, device_t from)
540 {
541 	struct pci_devinfo *dinfo;
542 	bool found = false;
543 
544 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
545 		if (from != NULL && found == false) {
546 			if (from != dinfo->cfg.dev)
547 				continue;
548 			found = true;
549 			continue;
550 		}
551 		if (dinfo->cfg.baseclass == class) {
552 			return (dinfo->cfg.dev);
553 		}
554 	}
555 
556 	return (NULL);
557 }
558 
559 static int
pci_printf(pcicfgregs * cfg,const char * fmt,...)560 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
561 {
562 	va_list ap;
563 	int retval;
564 
565 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
566 	    cfg->func);
567 	va_start(ap, fmt);
568 	retval += vprintf(fmt, ap);
569 	va_end(ap);
570 	return (retval);
571 }
572 
573 /* return base address of memory or port map */
574 
575 static pci_addr_t
pci_mapbase(uint64_t mapreg)576 pci_mapbase(uint64_t mapreg)
577 {
578 
579 	if (PCI_BAR_MEM(mapreg))
580 		return (mapreg & PCIM_BAR_MEM_BASE);
581 	else
582 		return (mapreg & PCIM_BAR_IO_BASE);
583 }
584 
585 /* return map type of memory or port map */
586 
587 static const char *
pci_maptype(uint64_t mapreg)588 pci_maptype(uint64_t mapreg)
589 {
590 
591 	if (PCI_BAR_IO(mapreg))
592 		return ("I/O Port");
593 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
594 		return ("Prefetchable Memory");
595 	return ("Memory");
596 }
597 
598 /* return log2 of map size decoded for memory or port map */
599 
600 int
pci_mapsize(uint64_t testval)601 pci_mapsize(uint64_t testval)
602 {
603 	int ln2size;
604 
605 	testval = pci_mapbase(testval);
606 	ln2size = 0;
607 	if (testval != 0) {
608 		while ((testval & 1) == 0)
609 		{
610 			ln2size++;
611 			testval >>= 1;
612 		}
613 	}
614 	return (ln2size);
615 }
616 
617 /* return base address of device ROM */
618 
619 static pci_addr_t
pci_rombase(uint64_t mapreg)620 pci_rombase(uint64_t mapreg)
621 {
622 
623 	return (mapreg & PCIM_BIOS_ADDR_MASK);
624 }
625 
626 /* return log2 of map size decided for device ROM */
627 
628 static int
pci_romsize(uint64_t testval)629 pci_romsize(uint64_t testval)
630 {
631 	int ln2size;
632 
633 	testval = pci_rombase(testval);
634 	ln2size = 0;
635 	if (testval != 0) {
636 		while ((testval & 1) == 0)
637 		{
638 			ln2size++;
639 			testval >>= 1;
640 		}
641 	}
642 	return (ln2size);
643 }
644 
645 /* return log2 of address range supported by map register */
646 
647 static int
pci_maprange(uint64_t mapreg)648 pci_maprange(uint64_t mapreg)
649 {
650 	int ln2range = 0;
651 
652 	if (PCI_BAR_IO(mapreg))
653 		ln2range = 32;
654 	else
655 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
656 		case PCIM_BAR_MEM_32:
657 			ln2range = 32;
658 			break;
659 		case PCIM_BAR_MEM_1MB:
660 			ln2range = 20;
661 			break;
662 		case PCIM_BAR_MEM_64:
663 			ln2range = 64;
664 			break;
665 		}
666 	return (ln2range);
667 }
668 
669 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
670 
671 static void
pci_fixancient(pcicfgregs * cfg)672 pci_fixancient(pcicfgregs *cfg)
673 {
674 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
675 		return;
676 
677 	/* PCI to PCI bridges use header type 1 */
678 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
679 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
680 }
681 
682 /* extract header type specific config data */
683 
684 static void
pci_hdrtypedata(device_t pcib,int b,int s,int f,pcicfgregs * cfg)685 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
686 {
687 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
688 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
689 	case PCIM_HDRTYPE_NORMAL:
690 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
691 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
692 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
693 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
694 		cfg->nummaps	    = PCI_MAXMAPS_0;
695 		break;
696 	case PCIM_HDRTYPE_BRIDGE:
697 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
698 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
699 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
700 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
701 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
702 		cfg->nummaps	    = PCI_MAXMAPS_1;
703 		break;
704 	case PCIM_HDRTYPE_CARDBUS:
705 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
706 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
707 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
708 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
709 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
710 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
711 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
712 		cfg->nummaps	    = PCI_MAXMAPS_2;
713 		break;
714 	}
715 #undef REG
716 }
717 
718 /* read configuration header into pcicfgregs structure */
719 struct pci_devinfo *
pci_read_device(device_t pcib,device_t bus,int d,int b,int s,int f)720 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
721 {
722 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
723 	uint16_t vid, did;
724 
725 	vid = REG(PCIR_VENDOR, 2);
726 	if (vid == PCIV_INVALID)
727 		return (NULL);
728 
729 	did = REG(PCIR_DEVICE, 2);
730 
731 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
732 }
733 
734 struct pci_devinfo *
pci_alloc_devinfo_method(device_t dev)735 pci_alloc_devinfo_method(device_t dev)
736 {
737 
738 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
739 	    M_WAITOK | M_ZERO));
740 }
741 
742 static struct pci_devinfo *
pci_fill_devinfo(device_t pcib,device_t bus,int d,int b,int s,int f,uint16_t vid,uint16_t did)743 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
744     uint16_t vid, uint16_t did)
745 {
746 	struct pci_devinfo *devlist_entry;
747 	pcicfgregs *cfg;
748 
749 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
750 
751 	cfg = &devlist_entry->cfg;
752 
753 	cfg->domain		= d;
754 	cfg->bus		= b;
755 	cfg->slot		= s;
756 	cfg->func		= f;
757 	cfg->vendor		= vid;
758 	cfg->device		= did;
759 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
760 	cfg->statreg		= REG(PCIR_STATUS, 2);
761 	cfg->baseclass		= REG(PCIR_CLASS, 1);
762 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
763 	cfg->progif		= REG(PCIR_PROGIF, 1);
764 	cfg->revid		= REG(PCIR_REVID, 1);
765 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
766 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
767 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
768 	cfg->intpin		= REG(PCIR_INTPIN, 1);
769 	cfg->intline		= REG(PCIR_INTLINE, 1);
770 
771 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
772 	cfg->hdrtype		&= ~PCIM_MFDEV;
773 	STAILQ_INIT(&cfg->maps);
774 
775 	cfg->iov		= NULL;
776 
777 	pci_fixancient(cfg);
778 	pci_hdrtypedata(pcib, b, s, f, cfg);
779 
780 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
781 		pci_read_cap(pcib, cfg);
782 
783 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
784 
785 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
786 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
787 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
788 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
789 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
790 
791 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
792 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
793 	devlist_entry->conf.pc_vendor = cfg->vendor;
794 	devlist_entry->conf.pc_device = cfg->device;
795 
796 	devlist_entry->conf.pc_class = cfg->baseclass;
797 	devlist_entry->conf.pc_subclass = cfg->subclass;
798 	devlist_entry->conf.pc_progif = cfg->progif;
799 	devlist_entry->conf.pc_revid = cfg->revid;
800 
801 	devlist_entry->conf.pc_secbus = cfg->bridge.br_secbus;
802 	devlist_entry->conf.pc_subbus = cfg->bridge.br_subbus;
803 
804 	pci_numdevs++;
805 	pci_generation++;
806 
807 	return (devlist_entry);
808 }
809 #undef REG
810 
811 static void
pci_ea_fill_info(device_t pcib,pcicfgregs * cfg)812 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
813 {
814 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
815     cfg->ea.ea_location + (n), w)
816 	int num_ent;
817 	int ptr;
818 	int a, b;
819 	uint32_t val;
820 	int ent_size;
821 	uint32_t dw[4];
822 	uint64_t base, max_offset;
823 	struct pci_ea_entry *eae;
824 
825 	if (cfg->ea.ea_location == 0)
826 		return;
827 
828 	STAILQ_INIT(&cfg->ea.ea_entries);
829 
830 	/* Determine the number of entries */
831 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
832 	num_ent &= PCIM_EA_NUM_ENT_MASK;
833 
834 	/* Find the first entry to care of */
835 	ptr = PCIR_EA_FIRST_ENT;
836 
837 	/* Skip DWORD 2 for type 1 functions */
838 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
839 		ptr += 4;
840 
841 	for (a = 0; a < num_ent; a++) {
842 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
843 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
844 
845 		/* Read a number of dwords in the entry */
846 		val = REG(ptr, 4);
847 		ptr += 4;
848 		ent_size = (val & PCIM_EA_ES);
849 
850 		for (b = 0; b < ent_size; b++) {
851 			dw[b] = REG(ptr, 4);
852 			ptr += 4;
853 		}
854 
855 		eae->eae_flags = val;
856 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
857 
858 		base = dw[0] & PCIM_EA_FIELD_MASK;
859 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
860 		b = 2;
861 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
862 			base |= (uint64_t)dw[b] << 32UL;
863 			b++;
864 		}
865 		if (((dw[1] & PCIM_EA_IS_64) != 0)
866 		    && (b < ent_size)) {
867 			max_offset |= (uint64_t)dw[b] << 32UL;
868 			b++;
869 		}
870 
871 		eae->eae_base = base;
872 		eae->eae_max_offset = max_offset;
873 
874 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
875 
876 		if (bootverbose) {
877 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
878 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
879 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
880 		}
881 	}
882 }
883 #undef REG
884 
885 static void
pci_read_cap(device_t pcib,pcicfgregs * cfg)886 pci_read_cap(device_t pcib, pcicfgregs *cfg)
887 {
888 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
889 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
890 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
891 	uint64_t addr;
892 #endif
893 	uint32_t val;
894 	int	ptr, nextptr, ptrptr;
895 
896 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
897 	case PCIM_HDRTYPE_NORMAL:
898 	case PCIM_HDRTYPE_BRIDGE:
899 		ptrptr = PCIR_CAP_PTR;
900 		break;
901 	case PCIM_HDRTYPE_CARDBUS:
902 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
903 		break;
904 	default:
905 		return;		/* no extended capabilities support */
906 	}
907 	nextptr = REG(ptrptr, 1);	/* sanity check? */
908 
909 	/*
910 	 * Read capability entries.
911 	 */
912 	while (nextptr != 0) {
913 		/* Sanity check */
914 		if (nextptr > 255) {
915 			printf("illegal PCI extended capability offset %d\n",
916 			    nextptr);
917 			return;
918 		}
919 		/* Find the next entry */
920 		ptr = nextptr;
921 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
922 
923 		/* Process this entry */
924 		switch (REG(ptr + PCICAP_ID, 1)) {
925 		case PCIY_PMG:		/* PCI power management */
926 			cfg->pp.pp_location = ptr;
927 			cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
928 			break;
929 		case PCIY_HT:		/* HyperTransport */
930 			/* Determine HT-specific capability type. */
931 			val = REG(ptr + PCIR_HT_COMMAND, 2);
932 
933 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
934 				cfg->ht.ht_slave = ptr;
935 
936 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
937 			switch (val & PCIM_HTCMD_CAP_MASK) {
938 			case PCIM_HTCAP_MSI_MAPPING:
939 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
940 					/* Sanity check the mapping window. */
941 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
942 					    4);
943 					addr <<= 32;
944 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
945 					    4);
946 					if (addr != MSI_INTEL_ADDR_BASE)
947 						device_printf(pcib,
948 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
949 						    cfg->domain, cfg->bus,
950 						    cfg->slot, cfg->func,
951 						    (long long)addr);
952 				} else
953 					addr = MSI_INTEL_ADDR_BASE;
954 
955 				cfg->ht.ht_msimap = ptr;
956 				cfg->ht.ht_msictrl = val;
957 				cfg->ht.ht_msiaddr = addr;
958 				break;
959 			}
960 #endif
961 			break;
962 		case PCIY_MSI:		/* PCI MSI */
963 			cfg->msi.msi_location = ptr;
964 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
965 			break;
966 		case PCIY_MSIX:		/* PCI MSI-X */
967 			cfg->msix.msix_location = ptr;
968 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
969 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
970 			cfg->msix.msix_table_bar = PCIR_BAR(val &
971 			    PCIM_MSIX_BIR_MASK);
972 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
973 			val = REG(ptr + PCIR_MSIX_PBA, 4);
974 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
975 			    PCIM_MSIX_BIR_MASK);
976 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
977 			break;
978 		case PCIY_VPD:		/* PCI Vital Product Data */
979 			cfg->vpd.vpd_reg = ptr;
980 			break;
981 		case PCIY_SUBVENDOR:
982 			/* Should always be true. */
983 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
984 			    PCIM_HDRTYPE_BRIDGE) {
985 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
986 				cfg->subvendor = val & 0xffff;
987 				cfg->subdevice = val >> 16;
988 			}
989 			break;
990 		case PCIY_PCIX:		/* PCI-X */
991 			/*
992 			 * Assume we have a PCI-X chipset if we have
993 			 * at least one PCI-PCI bridge with a PCI-X
994 			 * capability.  Note that some systems with
995 			 * PCI-express or HT chipsets might match on
996 			 * this check as well.
997 			 */
998 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
999 			    PCIM_HDRTYPE_BRIDGE)
1000 				pcix_chipset = 1;
1001 			cfg->pcix.pcix_location = ptr;
1002 			break;
1003 		case PCIY_EXPRESS:	/* PCI-express */
1004 			/*
1005 			 * Assume we have a PCI-express chipset if we have
1006 			 * at least one PCI-express device.
1007 			 */
1008 			pcie_chipset = 1;
1009 			cfg->pcie.pcie_location = ptr;
1010 			val = REG(ptr + PCIER_FLAGS, 2);
1011 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
1012 			break;
1013 		case PCIY_EA:		/* Enhanced Allocation */
1014 			cfg->ea.ea_location = ptr;
1015 			pci_ea_fill_info(pcib, cfg);
1016 			break;
1017 		default:
1018 			break;
1019 		}
1020 	}
1021 
1022 #if defined(__powerpc__)
1023 	/*
1024 	 * Enable the MSI mapping window for all HyperTransport
1025 	 * slaves.  PCI-PCI bridges have their windows enabled via
1026 	 * PCIB_MAP_MSI().
1027 	 */
1028 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1029 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1030 		device_printf(pcib,
1031 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1032 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1033 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1034 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1035 		     2);
1036 	}
1037 #endif
1038 /* REG and WREG use carry through to next functions */
1039 }
1040 
1041 /*
1042  * PCI Vital Product Data
1043  */
1044 
1045 #define	PCI_VPD_TIMEOUT		1000000
1046 
1047 static int
pci_read_vpd_reg(device_t pcib,pcicfgregs * cfg,int reg,uint32_t * data)1048 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1049 {
1050 	int count = PCI_VPD_TIMEOUT;
1051 
1052 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1053 
1054 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1055 
1056 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1057 		if (--count < 0)
1058 			return (ENXIO);
1059 		DELAY(1);	/* limit looping */
1060 	}
1061 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1062 
1063 	return (0);
1064 }
1065 
1066 #if 0
1067 static int
1068 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1069 {
1070 	int count = PCI_VPD_TIMEOUT;
1071 
1072 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1073 
1074 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1075 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1076 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1077 		if (--count < 0)
1078 			return (ENXIO);
1079 		DELAY(1);	/* limit looping */
1080 	}
1081 
1082 	return (0);
1083 }
1084 #endif
1085 
1086 #undef PCI_VPD_TIMEOUT
1087 
1088 struct vpd_readstate {
1089 	device_t	pcib;
1090 	pcicfgregs	*cfg;
1091 	uint32_t	val;
1092 	int		bytesinval;
1093 	int		off;
1094 	uint8_t		cksum;
1095 };
1096 
1097 /* return 0 and one byte in *data if no read error, -1 else */
1098 static int
vpd_nextbyte(struct vpd_readstate * vrs,uint8_t * data)1099 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1100 {
1101 	uint32_t reg;
1102 	uint8_t byte;
1103 
1104 	if (vrs->bytesinval == 0) {
1105 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1106 			return (-1);
1107 		vrs->val = le32toh(reg);
1108 		vrs->off += 4;
1109 		byte = vrs->val & 0xff;
1110 		vrs->bytesinval = 3;
1111 	} else {
1112 		vrs->val = vrs->val >> 8;
1113 		byte = vrs->val & 0xff;
1114 		vrs->bytesinval--;
1115 	}
1116 
1117 	vrs->cksum += byte;
1118 	*data = byte;
1119 	return (0);
1120 }
1121 
1122 /* return 0 on match, -1 and "unget" byte on no match */
1123 static int
vpd_expectbyte(struct vpd_readstate * vrs,uint8_t expected)1124 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1125 {
1126 	uint8_t data;
1127 
1128 	if (vpd_nextbyte(vrs, &data) != 0)
1129 		return (-1);
1130 
1131 	if (data == expected)
1132 		return (0);
1133 
1134 	vrs->cksum -= data;
1135 	vrs->val = (vrs->val << 8) + data;
1136 	vrs->bytesinval++;
1137 	return (-1);
1138 }
1139 
1140 /* return size if tag matches, -1 on no match, -2 on read error */
1141 static int
vpd_read_tag_size(struct vpd_readstate * vrs,uint8_t vpd_tag)1142 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1143 {
1144 	uint8_t byte1, byte2;
1145 
1146 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1147 		return (-1);
1148 
1149 	if ((vpd_tag & 0x80) == 0)
1150 		return (vpd_tag & 0x07);
1151 
1152 	if (vpd_nextbyte(vrs, &byte1) != 0)
1153 		return (-2);
1154 	if (vpd_nextbyte(vrs, &byte2) != 0)
1155 		return (-2);
1156 
1157 	return ((byte2 << 8) + byte1);
1158 }
1159 
1160 /* (re)allocate buffer in multiples of 8 elements */
1161 static void*
alloc_buffer(void * buffer,size_t element_size,int needed)1162 alloc_buffer(void* buffer, size_t element_size, int needed)
1163 {
1164 	int alloc, new_alloc;
1165 
1166 	alloc = roundup2(needed, 8);
1167 	new_alloc = roundup2(needed + 1, 8);
1168 	if (alloc != new_alloc) {
1169 		buffer = reallocf(buffer,
1170 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1171 	}
1172 
1173 	return (buffer);
1174 }
1175 
1176 /* read VPD keyword and return element size, return -1 on read error */
1177 static int
vpd_read_elem_head(struct vpd_readstate * vrs,char keyword[2])1178 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1179 {
1180 	uint8_t data;
1181 
1182 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1183 		return (-1);
1184 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1185 		return (-1);
1186 	if (vpd_nextbyte(vrs, &data) != 0)
1187 		return (-1);
1188 
1189 	return (data);
1190 }
1191 
1192 /* read VPD data element of given size into allocated buffer */
1193 static char *
vpd_read_value(struct vpd_readstate * vrs,int size)1194 vpd_read_value(struct vpd_readstate *vrs, int size)
1195 {
1196 	int i;
1197 	char char1;
1198 	char *value;
1199 
1200 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1201 	for (i = 0; i < size; i++) {
1202 		if (vpd_nextbyte(vrs, &char1) != 0) {
1203 			free(value, M_DEVBUF);
1204 			return (NULL);
1205 		}
1206 		value[i] = char1;
1207 	}
1208 	value[size] = '\0';
1209 
1210 	return (value);
1211 }
1212 
1213 /* read VPD into *keyword and *value, return length of data element */
1214 static int
vpd_read_elem_data(struct vpd_readstate * vrs,char keyword[2],char ** value,int maxlen)1215 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1216 {
1217 	int len;
1218 
1219 	len = vpd_read_elem_head(vrs, keyword);
1220 	if (len < 0 || len > maxlen)
1221 		return (-1);
1222 	*value = vpd_read_value(vrs, len);
1223 
1224 	return (len);
1225 }
1226 
1227 /* subtract all data following first byte from checksum of RV element */
1228 static void
vpd_fixup_cksum(struct vpd_readstate * vrs,char * rvstring,int len)1229 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1230 {
1231 	int i;
1232 	uint8_t fixup;
1233 
1234 	fixup = 0;
1235 	for (i = 1; i < len; i++)
1236 		fixup += rvstring[i];
1237 	vrs->cksum -= fixup;
1238 }
1239 
1240 /* fetch one read-only element and return size of heading + data */
1241 static int
next_vpd_ro_elem(struct vpd_readstate * vrs,int maxsize)1242 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1243 {
1244 	struct pcicfg_vpd *vpd;
1245 	pcicfgregs *cfg;
1246 	struct vpd_readonly *vpd_ros;
1247 	int len;
1248 
1249 	cfg = vrs->cfg;
1250 	vpd = &cfg->vpd;
1251 
1252 	if (maxsize < 3)
1253 		return (-1);
1254 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1255 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1256 	maxsize -= 3;
1257 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1258 	if (vpd_ros->value == NULL)
1259 		return (-1);
1260 	vpd_ros->len = len;
1261 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1262 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1263 		if (vrs->cksum != 0) {
1264 			pci_printf(cfg,
1265 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1266 			return (-1);
1267 		}
1268 	}
1269 	vpd->vpd_rocnt++;
1270 
1271 	return (len + 3);
1272 }
1273 
1274 /* fetch one writable element and return size of heading + data */
1275 static int
next_vpd_rw_elem(struct vpd_readstate * vrs,int maxsize)1276 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1277 {
1278 	struct pcicfg_vpd *vpd;
1279 	pcicfgregs *cfg;
1280 	struct vpd_write *vpd_w;
1281 	int len;
1282 
1283 	cfg = vrs->cfg;
1284 	vpd = &cfg->vpd;
1285 
1286 	if (maxsize < 3)
1287 		return (-1);
1288 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1289 	if (vpd->vpd_w == NULL) {
1290 		pci_printf(cfg, "out of memory");
1291 		return (-1);
1292 	}
1293 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1294 	maxsize -= 3;
1295 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1296 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1297 	if (vpd_w->value == NULL)
1298 		return (-1);
1299 	vpd_w->len = len;
1300 	vpd->vpd_wcnt++;
1301 
1302 	return (len + 3);
1303 }
1304 
1305 /* free all memory allocated for VPD data */
1306 static void
vpd_free(struct pcicfg_vpd * vpd)1307 vpd_free(struct pcicfg_vpd *vpd)
1308 {
1309 	int i;
1310 
1311 	free(vpd->vpd_ident, M_DEVBUF);
1312 	for (i = 0; i < vpd->vpd_rocnt; i++)
1313 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1314 	free(vpd->vpd_ros, M_DEVBUF);
1315 	vpd->vpd_rocnt = 0;
1316 	for (i = 0; i < vpd->vpd_wcnt; i++)
1317 		free(vpd->vpd_w[i].value, M_DEVBUF);
1318 	free(vpd->vpd_w, M_DEVBUF);
1319 	vpd->vpd_wcnt = 0;
1320 }
1321 
1322 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1323 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1324 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1325 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1326 
1327 static int
pci_parse_vpd(device_t pcib,pcicfgregs * cfg)1328 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1329 {
1330 	struct vpd_readstate vrs;
1331 	int cksumvalid;
1332 	int size, elem_size;
1333 
1334 	/* init vpd reader */
1335 	vrs.bytesinval = 0;
1336 	vrs.off = 0;
1337 	vrs.pcib = pcib;
1338 	vrs.cfg = cfg;
1339 	vrs.cksum = 0;
1340 
1341 	/* read VPD ident element - mandatory */
1342 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1343 	if (size <= 0) {
1344 		pci_printf(cfg, "no VPD ident found\n");
1345 		return (0);
1346 	}
1347 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1348 	if (cfg->vpd.vpd_ident == NULL) {
1349 		pci_printf(cfg, "error accessing VPD ident data\n");
1350 		return (0);
1351 	}
1352 
1353 	/* read VPD RO elements - mandatory */
1354 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1355 	if (size <= 0) {
1356 		pci_printf(cfg, "no read-only VPD data found\n");
1357 		return (0);
1358 	}
1359 	while (size > 0) {
1360 		elem_size = next_vpd_ro_elem(&vrs, size);
1361 		if (elem_size < 0) {
1362 			pci_printf(cfg, "error accessing read-only VPD data\n");
1363 			return (-1);
1364 		}
1365 		size -= elem_size;
1366 	}
1367 	cksumvalid = (vrs.cksum == 0);
1368 	if (!cksumvalid)
1369 		return (-1);
1370 
1371 	/* read VPD RW elements - optional */
1372 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1373 	if (size == -2)
1374 		return (-1);
1375 	while (size > 0) {
1376 		elem_size = next_vpd_rw_elem(&vrs, size);
1377 		if (elem_size < 0) {
1378 			pci_printf(cfg, "error accessing writeable VPD data\n");
1379 			return (-1);
1380 		}
1381 		size -= elem_size;
1382 	}
1383 
1384 	/* read empty END tag - mandatory */
1385 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1386 	if (size != 0) {
1387 		pci_printf(cfg, "No valid VPD end tag found\n");
1388 	}
1389 	return (0);
1390 }
1391 
1392 static void
pci_read_vpd(device_t pcib,pcicfgregs * cfg)1393 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1394 {
1395 	int status;
1396 
1397 	status = pci_parse_vpd(pcib, cfg);
1398 	if (status < 0)
1399 		vpd_free(&cfg->vpd);
1400 	cfg->vpd.vpd_cached = 1;
1401 #undef REG
1402 #undef WREG
1403 }
1404 
1405 int
pci_get_vpd_ident_method(device_t dev,device_t child,const char ** identptr)1406 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1407 {
1408 	struct pci_devinfo *dinfo = device_get_ivars(child);
1409 	pcicfgregs *cfg = &dinfo->cfg;
1410 
1411 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1412 		pci_read_vpd(device_get_parent(dev), cfg);
1413 
1414 	*identptr = cfg->vpd.vpd_ident;
1415 
1416 	if (*identptr == NULL)
1417 		return (ENXIO);
1418 
1419 	return (0);
1420 }
1421 
1422 int
pci_get_vpd_readonly_method(device_t dev,device_t child,const char * kw,const char ** vptr)1423 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1424 	const char **vptr)
1425 {
1426 	struct pci_devinfo *dinfo = device_get_ivars(child);
1427 	pcicfgregs *cfg = &dinfo->cfg;
1428 	int i;
1429 
1430 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1431 		pci_read_vpd(device_get_parent(dev), cfg);
1432 
1433 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1434 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1435 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1436 			*vptr = cfg->vpd.vpd_ros[i].value;
1437 			return (0);
1438 		}
1439 
1440 	*vptr = NULL;
1441 	return (ENXIO);
1442 }
1443 
1444 struct pcicfg_vpd *
pci_fetch_vpd_list(device_t dev)1445 pci_fetch_vpd_list(device_t dev)
1446 {
1447 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1448 	pcicfgregs *cfg = &dinfo->cfg;
1449 
1450 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1451 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1452 	return (&cfg->vpd);
1453 }
1454 
1455 /*
1456  * Find the requested HyperTransport capability and return the offset
1457  * in configuration space via the pointer provided.  The function
1458  * returns 0 on success and an error code otherwise.
1459  */
1460 int
pci_find_htcap_method(device_t dev,device_t child,int capability,int * capreg)1461 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1462 {
1463 	int ptr, error;
1464 	uint16_t val;
1465 
1466 	error = pci_find_cap(child, PCIY_HT, &ptr);
1467 	if (error)
1468 		return (error);
1469 
1470 	/*
1471 	 * Traverse the capabilities list checking each HT capability
1472 	 * to see if it matches the requested HT capability.
1473 	 */
1474 	for (;;) {
1475 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1476 		if (capability == PCIM_HTCAP_SLAVE ||
1477 		    capability == PCIM_HTCAP_HOST)
1478 			val &= 0xe000;
1479 		else
1480 			val &= PCIM_HTCMD_CAP_MASK;
1481 		if (val == capability) {
1482 			if (capreg != NULL)
1483 				*capreg = ptr;
1484 			return (0);
1485 		}
1486 
1487 		/* Skip to the next HT capability. */
1488 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1489 			break;
1490 	}
1491 
1492 	return (ENOENT);
1493 }
1494 
1495 /*
1496  * Find the next requested HyperTransport capability after start and return
1497  * the offset in configuration space via the pointer provided.  The function
1498  * returns 0 on success and an error code otherwise.
1499  */
1500 int
pci_find_next_htcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1501 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1502     int start, int *capreg)
1503 {
1504 	int ptr;
1505 	uint16_t val;
1506 
1507 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1508 	    ("start capability is not HyperTransport capability"));
1509 	ptr = start;
1510 
1511 	/*
1512 	 * Traverse the capabilities list checking each HT capability
1513 	 * to see if it matches the requested HT capability.
1514 	 */
1515 	for (;;) {
1516 		/* Skip to the next HT capability. */
1517 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1518 			break;
1519 
1520 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1521 		if (capability == PCIM_HTCAP_SLAVE ||
1522 		    capability == PCIM_HTCAP_HOST)
1523 			val &= 0xe000;
1524 		else
1525 			val &= PCIM_HTCMD_CAP_MASK;
1526 		if (val == capability) {
1527 			if (capreg != NULL)
1528 				*capreg = ptr;
1529 			return (0);
1530 		}
1531 	}
1532 
1533 	return (ENOENT);
1534 }
1535 
1536 /*
1537  * Find the requested capability and return the offset in
1538  * configuration space via the pointer provided.  The function returns
1539  * 0 on success and an error code otherwise.
1540  */
1541 int
pci_find_cap_method(device_t dev,device_t child,int capability,int * capreg)1542 pci_find_cap_method(device_t dev, device_t child, int capability,
1543     int *capreg)
1544 {
1545 	struct pci_devinfo *dinfo = device_get_ivars(child);
1546 	pcicfgregs *cfg = &dinfo->cfg;
1547 	uint32_t status;
1548 	uint8_t ptr;
1549 	int cnt;
1550 
1551 	/*
1552 	 * Check the CAP_LIST bit of the PCI status register first.
1553 	 */
1554 	status = pci_read_config(child, PCIR_STATUS, 2);
1555 	if (!(status & PCIM_STATUS_CAPPRESENT))
1556 		return (ENXIO);
1557 
1558 	/*
1559 	 * Determine the start pointer of the capabilities list.
1560 	 */
1561 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1562 	case PCIM_HDRTYPE_NORMAL:
1563 	case PCIM_HDRTYPE_BRIDGE:
1564 		ptr = PCIR_CAP_PTR;
1565 		break;
1566 	case PCIM_HDRTYPE_CARDBUS:
1567 		ptr = PCIR_CAP_PTR_2;
1568 		break;
1569 	default:
1570 		/* XXX: panic? */
1571 		return (ENXIO);		/* no extended capabilities support */
1572 	}
1573 	ptr = pci_read_config(child, ptr, 1);
1574 
1575 	/*
1576 	 * Traverse the capabilities list.  Limit by total theoretical
1577 	 * maximum number of caps: capability needs at least id and
1578 	 * next registers, and any type X header cannot contain caps.
1579 	 */
1580 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1581 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1582 			if (capreg != NULL)
1583 				*capreg = ptr;
1584 			return (0);
1585 		}
1586 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1587 	}
1588 
1589 	return (ENOENT);
1590 }
1591 
1592 /*
1593  * Find the next requested capability after start and return the offset in
1594  * configuration space via the pointer provided.  The function returns
1595  * 0 on success and an error code otherwise.
1596  */
1597 int
pci_find_next_cap_method(device_t dev,device_t child,int capability,int start,int * capreg)1598 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1599     int start, int *capreg)
1600 {
1601 	uint8_t ptr;
1602 
1603 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1604 	    ("start capability is not expected capability"));
1605 
1606 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1607 	while (ptr != 0) {
1608 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1609 			if (capreg != NULL)
1610 				*capreg = ptr;
1611 			return (0);
1612 		}
1613 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1614 	}
1615 
1616 	return (ENOENT);
1617 }
1618 
1619 /*
1620  * Find the requested extended capability and return the offset in
1621  * configuration space via the pointer provided.  The function returns
1622  * 0 on success and an error code otherwise.
1623  */
1624 int
pci_find_extcap_method(device_t dev,device_t child,int capability,int * capreg)1625 pci_find_extcap_method(device_t dev, device_t child, int capability,
1626     int *capreg)
1627 {
1628 	struct pci_devinfo *dinfo = device_get_ivars(child);
1629 	pcicfgregs *cfg = &dinfo->cfg;
1630 	uint32_t ecap;
1631 	uint16_t ptr;
1632 
1633 	/* Only supported for PCI-express devices. */
1634 	if (cfg->pcie.pcie_location == 0)
1635 		return (ENXIO);
1636 
1637 	ptr = PCIR_EXTCAP;
1638 	ecap = pci_read_config(child, ptr, 4);
1639 	if (ecap == 0xffffffff || ecap == 0)
1640 		return (ENOENT);
1641 	for (;;) {
1642 		if (PCI_EXTCAP_ID(ecap) == capability) {
1643 			if (capreg != NULL)
1644 				*capreg = ptr;
1645 			return (0);
1646 		}
1647 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1648 		if (ptr == 0)
1649 			break;
1650 		ecap = pci_read_config(child, ptr, 4);
1651 	}
1652 
1653 	return (ENOENT);
1654 }
1655 
1656 /*
1657  * Find the next requested extended capability after start and return the
1658  * offset in configuration space via the pointer provided.  The function
1659  * returns 0 on success and an error code otherwise.
1660  */
1661 int
pci_find_next_extcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1662 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1663     int start, int *capreg)
1664 {
1665 	struct pci_devinfo *dinfo = device_get_ivars(child);
1666 	pcicfgregs *cfg = &dinfo->cfg;
1667 	uint32_t ecap;
1668 	uint16_t ptr;
1669 
1670 	/* Only supported for PCI-express devices. */
1671 	if (cfg->pcie.pcie_location == 0)
1672 		return (ENXIO);
1673 
1674 	ecap = pci_read_config(child, start, 4);
1675 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1676 	    ("start extended capability is not expected capability"));
1677 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1678 	while (ptr != 0) {
1679 		ecap = pci_read_config(child, ptr, 4);
1680 		if (PCI_EXTCAP_ID(ecap) == capability) {
1681 			if (capreg != NULL)
1682 				*capreg = ptr;
1683 			return (0);
1684 		}
1685 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1686 	}
1687 
1688 	return (ENOENT);
1689 }
1690 
1691 /*
1692  * Support for MSI-X message interrupts.
1693  */
1694 static void
pci_write_msix_entry(device_t dev,u_int index,uint64_t address,uint32_t data)1695 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1696 {
1697 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1698 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1699 	uint32_t offset;
1700 
1701 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1702 	offset = msix->msix_table_offset + index * 16;
1703 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1704 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1705 	bus_write_4(msix->msix_table_res, offset + 8, data);
1706 }
1707 
1708 void
pci_enable_msix_method(device_t dev,device_t child,u_int index,uint64_t address,uint32_t data)1709 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1710     uint64_t address, uint32_t data)
1711 {
1712 
1713 	if (pci_msix_rewrite_table) {
1714 		struct pci_devinfo *dinfo = device_get_ivars(child);
1715 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1716 
1717 		/*
1718 		 * Some VM hosts require MSIX to be disabled in the
1719 		 * control register before updating the MSIX table
1720 		 * entries are allowed. It is not enough to only
1721 		 * disable MSIX while updating a single entry. MSIX
1722 		 * must be disabled while updating all entries in the
1723 		 * table.
1724 		 */
1725 		pci_write_config(child,
1726 		    msix->msix_location + PCIR_MSIX_CTRL,
1727 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1728 		pci_resume_msix(child);
1729 	} else
1730 		pci_write_msix_entry(child, index, address, data);
1731 
1732 	/* Enable MSI -> HT mapping. */
1733 	pci_ht_map_msi(child, address);
1734 }
1735 
1736 void
pci_mask_msix(device_t dev,u_int index)1737 pci_mask_msix(device_t dev, u_int index)
1738 {
1739 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1740 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1741 	uint32_t offset, val;
1742 
1743 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1744 	offset = msix->msix_table_offset + index * 16 + 12;
1745 	val = bus_read_4(msix->msix_table_res, offset);
1746 	val |= PCIM_MSIX_VCTRL_MASK;
1747 
1748 	/*
1749 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1750 	 * register, so always write the new value.
1751 	 */
1752 	bus_write_4(msix->msix_table_res, offset, val);
1753 }
1754 
1755 void
pci_unmask_msix(device_t dev,u_int index)1756 pci_unmask_msix(device_t dev, u_int index)
1757 {
1758 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1759 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1760 	uint32_t offset, val;
1761 
1762 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1763 	offset = msix->msix_table_offset + index * 16 + 12;
1764 	val = bus_read_4(msix->msix_table_res, offset);
1765 	val &= ~PCIM_MSIX_VCTRL_MASK;
1766 
1767 	/*
1768 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1769 	 * register, so always write the new value.
1770 	 */
1771 	bus_write_4(msix->msix_table_res, offset, val);
1772 }
1773 
1774 int
pci_pending_msix(device_t dev,u_int index)1775 pci_pending_msix(device_t dev, u_int index)
1776 {
1777 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1778 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1779 	uint32_t offset, bit;
1780 
1781 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1782 	offset = msix->msix_pba_offset + (index / 32) * 4;
1783 	bit = 1 << index % 32;
1784 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1785 }
1786 
1787 /*
1788  * Restore MSI-X registers and table during resume.  If MSI-X is
1789  * enabled then walk the virtual table to restore the actual MSI-X
1790  * table.
1791  */
1792 static void
pci_resume_msix(device_t dev)1793 pci_resume_msix(device_t dev)
1794 {
1795 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1796 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1797 	struct msix_table_entry *mte;
1798 	struct msix_vector *mv;
1799 	u_int i, msgnum;
1800 
1801 	if (msix->msix_alloc > 0) {
1802 		msgnum = PCI_MSIX_MSGNUM(msix->msix_ctrl);
1803 
1804 		/* First, mask all vectors. */
1805 		for (i = 0; i < msgnum; i++)
1806 			pci_mask_msix(dev, i);
1807 
1808 		/* Second, program any messages with at least one handler. */
1809 		for (i = 0; i < msix->msix_table_len; i++) {
1810 			mte = &msix->msix_table[i];
1811 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1812 				continue;
1813 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1814 			pci_write_msix_entry(dev, i, mv->mv_address,
1815 			    mv->mv_data);
1816 			pci_unmask_msix(dev, i);
1817 		}
1818 	}
1819 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1820 	    msix->msix_ctrl, 2);
1821 }
1822 
1823 /*
1824  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1825  * returned in *count.  After this function returns, each message will be
1826  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1827  */
1828 int
pci_alloc_msix_method(device_t dev,device_t child,int * count)1829 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1830 {
1831 	struct pci_devinfo *dinfo = device_get_ivars(child);
1832 	pcicfgregs *cfg = &dinfo->cfg;
1833 	struct resource_list_entry *rle;
1834 	u_int actual, i, max;
1835 	int error, irq;
1836 	uint16_t ctrl, msgnum;
1837 
1838 	/* Don't let count == 0 get us into trouble. */
1839 	if (*count < 1)
1840 		return (EINVAL);
1841 
1842 	/* If rid 0 is allocated, then fail. */
1843 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1844 	if (rle != NULL && rle->res != NULL)
1845 		return (ENXIO);
1846 
1847 	/* Already have allocated messages? */
1848 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1849 		return (ENXIO);
1850 
1851 	/* If MSI-X is blacklisted for this system, fail. */
1852 	if (pci_msix_blacklisted())
1853 		return (ENXIO);
1854 
1855 	/* MSI-X capability present? */
1856 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1857 		return (ENODEV);
1858 
1859 	/* Make sure the appropriate BARs are mapped. */
1860 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1861 	    cfg->msix.msix_table_bar);
1862 	if (rle == NULL || rle->res == NULL ||
1863 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1864 		return (ENXIO);
1865 	cfg->msix.msix_table_res = rle->res;
1866 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1867 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1868 		    cfg->msix.msix_pba_bar);
1869 		if (rle == NULL || rle->res == NULL ||
1870 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1871 			return (ENXIO);
1872 	}
1873 	cfg->msix.msix_pba_res = rle->res;
1874 
1875 	ctrl = pci_read_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1876 	    2);
1877 	msgnum = PCI_MSIX_MSGNUM(ctrl);
1878 	if (bootverbose)
1879 		device_printf(child,
1880 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1881 		    *count, msgnum);
1882 	max = min(*count, msgnum);
1883 	for (i = 0; i < max; i++) {
1884 		/* Allocate a message. */
1885 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1886 		if (error) {
1887 			if (i == 0)
1888 				return (error);
1889 			break;
1890 		}
1891 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1892 		    irq, 1);
1893 	}
1894 	actual = i;
1895 
1896 	if (bootverbose) {
1897 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1898 		if (actual == 1)
1899 			device_printf(child, "using IRQ %ju for MSI-X\n",
1900 			    rle->start);
1901 		else {
1902 			bool run;
1903 
1904 			/*
1905 			 * Be fancy and try to print contiguous runs of
1906 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1907 			 * 'run' is true if we are in a range.
1908 			 */
1909 			device_printf(child, "using IRQs %ju", rle->start);
1910 			irq = rle->start;
1911 			run = false;
1912 			for (i = 1; i < actual; i++) {
1913 				rle = resource_list_find(&dinfo->resources,
1914 				    SYS_RES_IRQ, i + 1);
1915 
1916 				/* Still in a run? */
1917 				if (rle->start == irq + 1) {
1918 					run = true;
1919 					irq++;
1920 					continue;
1921 				}
1922 
1923 				/* Finish previous range. */
1924 				if (run) {
1925 					printf("-%d", irq);
1926 					run = false;
1927 				}
1928 
1929 				/* Start new range. */
1930 				printf(",%ju", rle->start);
1931 				irq = rle->start;
1932 			}
1933 
1934 			/* Unfinished range? */
1935 			if (run)
1936 				printf("-%d", irq);
1937 			printf(" for MSI-X\n");
1938 		}
1939 	}
1940 
1941 	/*
1942 	 * Mask all vectors. Note that the message index assertion in
1943 	 * pci_mask_msix requires msix_ctrl to be set.
1944 	 */
1945 	cfg->msix.msix_ctrl = ctrl;
1946 	for (i = 0; i < msgnum; i++)
1947 		pci_mask_msix(child, i);
1948 
1949 	/* Allocate and initialize vector data and virtual table. */
1950 	cfg->msix.msix_vectors = mallocarray(actual, sizeof(struct msix_vector),
1951 	    M_DEVBUF, M_WAITOK | M_ZERO);
1952 	cfg->msix.msix_table = mallocarray(actual,
1953 	    sizeof(struct msix_table_entry), M_DEVBUF, M_WAITOK | M_ZERO);
1954 	for (i = 0; i < actual; i++) {
1955 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1956 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1957 		cfg->msix.msix_table[i].mte_vector = i + 1;
1958 	}
1959 
1960 	/* Update control register to enable MSI-X. */
1961 	ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1962 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1963 	    ctrl, 2);
1964 	cfg->msix.msix_ctrl = ctrl;
1965 
1966 	/* Update counts of alloc'd messages. */
1967 	cfg->msix.msix_alloc = actual;
1968 	cfg->msix.msix_table_len = actual;
1969 	*count = actual;
1970 	return (0);
1971 }
1972 
1973 /*
1974  * By default, pci_alloc_msix() will assign the allocated IRQ
1975  * resources consecutively to the first N messages in the MSI-X table.
1976  * However, device drivers may want to use different layouts if they
1977  * either receive fewer messages than they asked for, or they wish to
1978  * populate the MSI-X table sparsely.  This method allows the driver
1979  * to specify what layout it wants.  It must be called after a
1980  * successful pci_alloc_msix() but before any of the associated
1981  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1982  *
1983  * The 'vectors' array contains 'count' message vectors.  The array
1984  * maps directly to the MSI-X table in that index 0 in the array
1985  * specifies the vector for the first message in the MSI-X table, etc.
1986  * The vector value in each array index can either be 0 to indicate
1987  * that no vector should be assigned to a message slot, or it can be a
1988  * number from 1 to N (where N is the count returned from a
1989  * succcessful call to pci_alloc_msix()) to indicate which message
1990  * vector (IRQ) to be used for the corresponding message.
1991  *
1992  * On successful return, each message with a non-zero vector will have
1993  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1994  * 1.  Additionally, if any of the IRQs allocated via the previous
1995  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1996  * will be freed back to the system automatically.
1997  *
1998  * For example, suppose a driver has a MSI-X table with 6 messages and
1999  * asks for 6 messages, but pci_alloc_msix() only returns a count of
2000  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
2001  * C.  After the call to pci_alloc_msix(), the device will be setup to
2002  * have an MSI-X table of ABC--- (where - means no vector assigned).
2003  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
2004  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
2005  * be freed back to the system.  This device will also have valid
2006  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
2007  *
2008  * In any case, the SYS_RES_IRQ rid X will always map to the message
2009  * at MSI-X table index X - 1 and will only be valid if a vector is
2010  * assigned to that table entry.
2011  */
2012 int
pci_remap_msix_method(device_t dev,device_t child,int count,const u_int * vectors)2013 pci_remap_msix_method(device_t dev, device_t child, int count,
2014     const u_int *vectors)
2015 {
2016 	struct pci_devinfo *dinfo = device_get_ivars(child);
2017 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2018 	struct resource_list_entry *rle;
2019 	u_int i, irq, j;
2020 	bool *used;
2021 
2022 	/*
2023 	 * Have to have at least one message in the table but the
2024 	 * table can't be bigger than the actual MSI-X table in the
2025 	 * device.
2026 	 */
2027 	if (count < 1 || count > PCI_MSIX_MSGNUM(msix->msix_ctrl))
2028 		return (EINVAL);
2029 
2030 	/* Sanity check the vectors. */
2031 	for (i = 0; i < count; i++)
2032 		if (vectors[i] > msix->msix_alloc)
2033 			return (EINVAL);
2034 
2035 	/*
2036 	 * Make sure there aren't any holes in the vectors to be used.
2037 	 * It's a big pain to support it, and it doesn't really make
2038 	 * sense anyway.  Also, at least one vector must be used.
2039 	 */
2040 	used = mallocarray(msix->msix_alloc, sizeof(*used), M_DEVBUF, M_WAITOK |
2041 	    M_ZERO);
2042 	for (i = 0; i < count; i++)
2043 		if (vectors[i] != 0)
2044 			used[vectors[i] - 1] = true;
2045 	for (i = 0; i < msix->msix_alloc - 1; i++)
2046 		if (!used[i] && used[i + 1]) {
2047 			free(used, M_DEVBUF);
2048 			return (EINVAL);
2049 		}
2050 	if (!used[0]) {
2051 		free(used, M_DEVBUF);
2052 		return (EINVAL);
2053 	}
2054 
2055 	/* Make sure none of the resources are allocated. */
2056 	for (i = 0; i < msix->msix_table_len; i++) {
2057 		if (msix->msix_table[i].mte_vector == 0)
2058 			continue;
2059 		if (msix->msix_table[i].mte_handlers > 0) {
2060 			free(used, M_DEVBUF);
2061 			return (EBUSY);
2062 		}
2063 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2064 		KASSERT(rle != NULL, ("missing resource"));
2065 		if (rle->res != NULL) {
2066 			free(used, M_DEVBUF);
2067 			return (EBUSY);
2068 		}
2069 	}
2070 
2071 	/* Free the existing resource list entries. */
2072 	for (i = 0; i < msix->msix_table_len; i++) {
2073 		if (msix->msix_table[i].mte_vector == 0)
2074 			continue;
2075 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2076 	}
2077 
2078 	/*
2079 	 * Build the new virtual table keeping track of which vectors are
2080 	 * used.
2081 	 */
2082 	free(msix->msix_table, M_DEVBUF);
2083 	msix->msix_table = mallocarray(count, sizeof(struct msix_table_entry),
2084 	    M_DEVBUF, M_WAITOK | M_ZERO);
2085 	for (i = 0; i < count; i++)
2086 		msix->msix_table[i].mte_vector = vectors[i];
2087 	msix->msix_table_len = count;
2088 
2089 	/* Free any unused IRQs and resize the vectors array if necessary. */
2090 	j = msix->msix_alloc - 1;
2091 	if (!used[j]) {
2092 		struct msix_vector *vec;
2093 
2094 		while (!used[j]) {
2095 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2096 			    msix->msix_vectors[j].mv_irq);
2097 			j--;
2098 		}
2099 		vec = mallocarray(j + 1, sizeof(struct msix_vector), M_DEVBUF,
2100 		    M_WAITOK);
2101 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2102 		    (j + 1));
2103 		free(msix->msix_vectors, M_DEVBUF);
2104 		msix->msix_vectors = vec;
2105 		msix->msix_alloc = j + 1;
2106 	}
2107 	free(used, M_DEVBUF);
2108 
2109 	/* Map the IRQs onto the rids. */
2110 	for (i = 0; i < count; i++) {
2111 		if (vectors[i] == 0)
2112 			continue;
2113 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2114 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2115 		    irq, 1);
2116 	}
2117 
2118 	if (bootverbose) {
2119 		device_printf(child, "Remapped MSI-X IRQs as: ");
2120 		for (i = 0; i < count; i++) {
2121 			if (i != 0)
2122 				printf(", ");
2123 			if (vectors[i] == 0)
2124 				printf("---");
2125 			else
2126 				printf("%d",
2127 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2128 		}
2129 		printf("\n");
2130 	}
2131 
2132 	return (0);
2133 }
2134 
2135 static int
pci_release_msix(device_t dev,device_t child)2136 pci_release_msix(device_t dev, device_t child)
2137 {
2138 	struct pci_devinfo *dinfo = device_get_ivars(child);
2139 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2140 	struct resource_list_entry *rle;
2141 	u_int i;
2142 
2143 	/* Do we have any messages to release? */
2144 	if (msix->msix_alloc == 0)
2145 		return (ENODEV);
2146 
2147 	/* Make sure none of the resources are allocated. */
2148 	for (i = 0; i < msix->msix_table_len; i++) {
2149 		if (msix->msix_table[i].mte_vector == 0)
2150 			continue;
2151 		if (msix->msix_table[i].mte_handlers > 0)
2152 			return (EBUSY);
2153 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2154 		KASSERT(rle != NULL, ("missing resource"));
2155 		if (rle->res != NULL)
2156 			return (EBUSY);
2157 	}
2158 
2159 	/* Update control register to disable MSI-X. */
2160 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2161 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2162 	    msix->msix_ctrl, 2);
2163 
2164 	/* Free the resource list entries. */
2165 	for (i = 0; i < msix->msix_table_len; i++) {
2166 		if (msix->msix_table[i].mte_vector == 0)
2167 			continue;
2168 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2169 	}
2170 	free(msix->msix_table, M_DEVBUF);
2171 	msix->msix_table_len = 0;
2172 
2173 	/* Release the IRQs. */
2174 	for (i = 0; i < msix->msix_alloc; i++)
2175 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2176 		    msix->msix_vectors[i].mv_irq);
2177 	free(msix->msix_vectors, M_DEVBUF);
2178 	msix->msix_alloc = 0;
2179 	return (0);
2180 }
2181 
2182 /*
2183  * Return the max supported MSI-X messages this device supports.
2184  * Basically, assuming the MD code can alloc messages, this function
2185  * should return the maximum value that pci_alloc_msix() can return.
2186  * Thus, it is subject to the tunables, etc.
2187  */
2188 int
pci_msix_count_method(device_t dev,device_t child)2189 pci_msix_count_method(device_t dev, device_t child)
2190 {
2191 	struct pci_devinfo *dinfo = device_get_ivars(child);
2192 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2193 	uint16_t ctrl;
2194 
2195 	if (pci_do_msix && msix->msix_location != 0) {
2196 		ctrl = pci_read_config(child, msix->msix_location +
2197 		    PCIR_MSI_CTRL, 2);
2198 		return (PCI_MSIX_MSGNUM(ctrl));
2199 	}
2200 	return (0);
2201 }
2202 
2203 int
pci_msix_pba_bar_method(device_t dev,device_t child)2204 pci_msix_pba_bar_method(device_t dev, device_t child)
2205 {
2206 	struct pci_devinfo *dinfo = device_get_ivars(child);
2207 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2208 
2209 	if (pci_do_msix && msix->msix_location != 0)
2210 		return (msix->msix_pba_bar);
2211 	return (-1);
2212 }
2213 
2214 int
pci_msix_table_bar_method(device_t dev,device_t child)2215 pci_msix_table_bar_method(device_t dev, device_t child)
2216 {
2217 	struct pci_devinfo *dinfo = device_get_ivars(child);
2218 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2219 
2220 	if (pci_do_msix && msix->msix_location != 0)
2221 		return (msix->msix_table_bar);
2222 	return (-1);
2223 }
2224 
2225 /*
2226  * HyperTransport MSI mapping control
2227  */
2228 void
pci_ht_map_msi(device_t dev,uint64_t addr)2229 pci_ht_map_msi(device_t dev, uint64_t addr)
2230 {
2231 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2232 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2233 
2234 	if (!ht->ht_msimap)
2235 		return;
2236 
2237 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2238 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2239 		/* Enable MSI -> HT mapping. */
2240 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2241 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2242 		    ht->ht_msictrl, 2);
2243 	}
2244 
2245 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2246 		/* Disable MSI -> HT mapping. */
2247 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2248 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2249 		    ht->ht_msictrl, 2);
2250 	}
2251 }
2252 
2253 int
pci_get_relaxed_ordering_enabled(device_t dev)2254 pci_get_relaxed_ordering_enabled(device_t dev)
2255 {
2256 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2257 	int cap;
2258 	uint16_t val;
2259 
2260 	cap = dinfo->cfg.pcie.pcie_location;
2261 	if (cap == 0)
2262 		return (0);
2263 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2264 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2265 	return (val != 0);
2266 }
2267 
2268 int
pci_get_max_payload(device_t dev)2269 pci_get_max_payload(device_t dev)
2270 {
2271 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2272 	int cap;
2273 	uint16_t val;
2274 
2275 	cap = dinfo->cfg.pcie.pcie_location;
2276 	if (cap == 0)
2277 		return (0);
2278 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2279 	val &= PCIEM_CTL_MAX_PAYLOAD;
2280 	val >>= 5;
2281 	return (1 << (val + 7));
2282 }
2283 
2284 int
pci_get_max_read_req(device_t dev)2285 pci_get_max_read_req(device_t dev)
2286 {
2287 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2288 	int cap;
2289 	uint16_t val;
2290 
2291 	cap = dinfo->cfg.pcie.pcie_location;
2292 	if (cap == 0)
2293 		return (0);
2294 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2295 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2296 	val >>= 12;
2297 	return (1 << (val + 7));
2298 }
2299 
2300 int
pci_set_max_read_req(device_t dev,int size)2301 pci_set_max_read_req(device_t dev, int size)
2302 {
2303 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2304 	int cap;
2305 	uint16_t val;
2306 
2307 	cap = dinfo->cfg.pcie.pcie_location;
2308 	if (cap == 0)
2309 		return (0);
2310 	if (size < 128)
2311 		size = 128;
2312 	if (size > 4096)
2313 		size = 4096;
2314 	size = (1 << (fls(size) - 1));
2315 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2316 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2317 	val |= (fls(size) - 8) << 12;
2318 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2319 	return (size);
2320 }
2321 
2322 uint32_t
pcie_read_config(device_t dev,int reg,int width)2323 pcie_read_config(device_t dev, int reg, int width)
2324 {
2325 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2326 	int cap;
2327 
2328 	cap = dinfo->cfg.pcie.pcie_location;
2329 	if (cap == 0) {
2330 		if (width == 2)
2331 			return (0xffff);
2332 		return (0xffffffff);
2333 	}
2334 
2335 	return (pci_read_config(dev, cap + reg, width));
2336 }
2337 
2338 void
pcie_write_config(device_t dev,int reg,uint32_t value,int width)2339 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2340 {
2341 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2342 	int cap;
2343 
2344 	cap = dinfo->cfg.pcie.pcie_location;
2345 	if (cap == 0)
2346 		return;
2347 	pci_write_config(dev, cap + reg, value, width);
2348 }
2349 
2350 /*
2351  * Adjusts a PCI-e capability register by clearing the bits in mask
2352  * and setting the bits in (value & mask).  Bits not set in mask are
2353  * not adjusted.
2354  *
2355  * Returns the old value on success or all ones on failure.
2356  */
2357 uint32_t
pcie_adjust_config(device_t dev,int reg,uint32_t mask,uint32_t value,int width)2358 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2359     int width)
2360 {
2361 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2362 	uint32_t old, new;
2363 	int cap;
2364 
2365 	cap = dinfo->cfg.pcie.pcie_location;
2366 	if (cap == 0) {
2367 		if (width == 2)
2368 			return (0xffff);
2369 		return (0xffffffff);
2370 	}
2371 
2372 	old = pci_read_config(dev, cap + reg, width);
2373 	new = old & ~mask;
2374 	new |= (value & mask);
2375 	pci_write_config(dev, cap + reg, new, width);
2376 	return (old);
2377 }
2378 
2379 /*
2380  * Support for MSI message signalled interrupts.
2381  */
2382 void
pci_enable_msi_method(device_t dev,device_t child,uint64_t address,uint16_t data)2383 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2384     uint16_t data)
2385 {
2386 	struct pci_devinfo *dinfo = device_get_ivars(child);
2387 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2388 
2389 	/* Write data and address values. */
2390 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2391 	    address & 0xffffffff, 4);
2392 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2393 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2394 		    address >> 32, 4);
2395 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2396 		    data, 2);
2397 	} else
2398 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2399 		    2);
2400 
2401 	/* Enable MSI in the control register. */
2402 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2403 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2404 	    msi->msi_ctrl, 2);
2405 
2406 	/* Enable MSI -> HT mapping. */
2407 	pci_ht_map_msi(child, address);
2408 }
2409 
2410 void
pci_disable_msi_method(device_t dev,device_t child)2411 pci_disable_msi_method(device_t dev, device_t child)
2412 {
2413 	struct pci_devinfo *dinfo = device_get_ivars(child);
2414 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2415 
2416 	/* Disable MSI -> HT mapping. */
2417 	pci_ht_map_msi(child, 0);
2418 
2419 	/* Disable MSI in the control register. */
2420 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2421 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2422 	    msi->msi_ctrl, 2);
2423 }
2424 
2425 /*
2426  * Restore MSI registers during resume.  If MSI is enabled then
2427  * restore the data and address registers in addition to the control
2428  * register.
2429  */
2430 static void
pci_resume_msi(device_t dev)2431 pci_resume_msi(device_t dev)
2432 {
2433 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2434 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2435 	uint64_t address;
2436 	uint16_t data;
2437 
2438 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2439 		address = msi->msi_addr;
2440 		data = msi->msi_data;
2441 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2442 		    address & 0xffffffff, 4);
2443 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2444 			pci_write_config(dev, msi->msi_location +
2445 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2446 			pci_write_config(dev, msi->msi_location +
2447 			    PCIR_MSI_DATA_64BIT, data, 2);
2448 		} else
2449 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2450 			    data, 2);
2451 	}
2452 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2453 	    2);
2454 }
2455 
2456 static int
pci_remap_intr_method(device_t bus,device_t dev,u_int irq)2457 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2458 {
2459 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2460 	pcicfgregs *cfg = &dinfo->cfg;
2461 	struct resource_list_entry *rle;
2462 	struct msix_table_entry *mte;
2463 	struct msix_vector *mv;
2464 	uint64_t addr;
2465 	uint32_t data;
2466 	u_int i, j;
2467 	int error;
2468 
2469 	/*
2470 	 * Handle MSI first.  We try to find this IRQ among our list
2471 	 * of MSI IRQs.  If we find it, we request updated address and
2472 	 * data registers and apply the results.
2473 	 */
2474 	if (cfg->msi.msi_alloc > 0) {
2475 		/* If we don't have any active handlers, nothing to do. */
2476 		if (cfg->msi.msi_handlers == 0)
2477 			return (0);
2478 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2479 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2480 			    i + 1);
2481 			if (rle->start == irq) {
2482 				error = PCIB_MAP_MSI(device_get_parent(bus),
2483 				    dev, irq, &addr, &data);
2484 				if (error)
2485 					return (error);
2486 				pci_disable_msi(dev);
2487 				dinfo->cfg.msi.msi_addr = addr;
2488 				dinfo->cfg.msi.msi_data = data;
2489 				pci_enable_msi(dev, addr, data);
2490 				return (0);
2491 			}
2492 		}
2493 		return (ENOENT);
2494 	}
2495 
2496 	/*
2497 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2498 	 * we request the updated mapping info.  If that works, we go
2499 	 * through all the slots that use this IRQ and update them.
2500 	 */
2501 	if (cfg->msix.msix_alloc > 0) {
2502 		bool found = false;
2503 
2504 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2505 			mv = &cfg->msix.msix_vectors[i];
2506 			if (mv->mv_irq == irq) {
2507 				error = PCIB_MAP_MSI(device_get_parent(bus),
2508 				    dev, irq, &addr, &data);
2509 				if (error)
2510 					return (error);
2511 				mv->mv_address = addr;
2512 				mv->mv_data = data;
2513 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2514 					mte = &cfg->msix.msix_table[j];
2515 					if (mte->mte_vector != i + 1)
2516 						continue;
2517 					if (mte->mte_handlers == 0)
2518 						continue;
2519 					pci_mask_msix(dev, j);
2520 					pci_enable_msix(dev, j, addr, data);
2521 					pci_unmask_msix(dev, j);
2522 				}
2523 				found = true;
2524 			}
2525 		}
2526 		return (found ? 0 : ENOENT);
2527 	}
2528 
2529 	return (ENOENT);
2530 }
2531 
2532 /*
2533  * Returns true if the specified device is blacklisted because MSI
2534  * doesn't work.
2535  */
2536 int
pci_msi_device_blacklisted(device_t dev)2537 pci_msi_device_blacklisted(device_t dev)
2538 {
2539 
2540 	if (!pci_honor_msi_blacklist)
2541 		return (0);
2542 
2543 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2544 }
2545 
2546 /*
2547  * Determine if MSI is blacklisted globally on this system.  Currently,
2548  * we just check for blacklisted chipsets as represented by the
2549  * host-PCI bridge at device 0:0:0.  In the future, it may become
2550  * necessary to check other system attributes, such as the kenv values
2551  * that give the motherboard manufacturer and model number.
2552  */
2553 static int
pci_msi_blacklisted(void)2554 pci_msi_blacklisted(void)
2555 {
2556 	device_t dev;
2557 
2558 	if (!pci_honor_msi_blacklist)
2559 		return (0);
2560 
2561 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2562 	if (!(pcie_chipset || pcix_chipset)) {
2563 		if (vm_guest != VM_GUEST_NO) {
2564 			/*
2565 			 * Whitelist older chipsets in virtual
2566 			 * machines known to support MSI.
2567 			 */
2568 			dev = pci_find_bsf(0, 0, 0);
2569 			if (dev != NULL)
2570 				return (!pci_has_quirk(pci_get_devid(dev),
2571 					PCI_QUIRK_ENABLE_MSI_VM));
2572 		}
2573 		return (1);
2574 	}
2575 
2576 	dev = pci_find_bsf(0, 0, 0);
2577 	if (dev != NULL)
2578 		return (pci_msi_device_blacklisted(dev));
2579 	return (0);
2580 }
2581 
2582 /*
2583  * Returns true if the specified device is blacklisted because MSI-X
2584  * doesn't work.  Note that this assumes that if MSI doesn't work,
2585  * MSI-X doesn't either.
2586  */
2587 int
pci_msix_device_blacklisted(device_t dev)2588 pci_msix_device_blacklisted(device_t dev)
2589 {
2590 
2591 	if (!pci_honor_msi_blacklist)
2592 		return (0);
2593 
2594 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2595 		return (1);
2596 
2597 	return (pci_msi_device_blacklisted(dev));
2598 }
2599 
2600 /*
2601  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2602  * is blacklisted, assume that MSI-X is as well.  Check for additional
2603  * chipsets where MSI works but MSI-X does not.
2604  */
2605 static int
pci_msix_blacklisted(void)2606 pci_msix_blacklisted(void)
2607 {
2608 	device_t dev;
2609 
2610 	if (!pci_honor_msi_blacklist)
2611 		return (0);
2612 
2613 	dev = pci_find_bsf(0, 0, 0);
2614 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2615 	    PCI_QUIRK_DISABLE_MSIX))
2616 		return (1);
2617 
2618 	return (pci_msi_blacklisted());
2619 }
2620 
2621 /*
2622  * Attempt to allocate *count MSI messages.  The actual number allocated is
2623  * returned in *count.  After this function returns, each message will be
2624  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2625  */
2626 int
pci_alloc_msi_method(device_t dev,device_t child,int * count)2627 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2628 {
2629 	struct pci_devinfo *dinfo = device_get_ivars(child);
2630 	pcicfgregs *cfg = &dinfo->cfg;
2631 	struct resource_list_entry *rle;
2632 	u_int actual, i;
2633 	int error, irqs[32];
2634 	uint16_t ctrl, msgnum;
2635 
2636 	/* Don't let count == 0 get us into trouble. */
2637 	if (*count < 1)
2638 		return (EINVAL);
2639 
2640 	/* If rid 0 is allocated, then fail. */
2641 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2642 	if (rle != NULL && rle->res != NULL)
2643 		return (ENXIO);
2644 
2645 	/* Already have allocated messages? */
2646 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2647 		return (ENXIO);
2648 
2649 	/* If MSI is blacklisted for this system, fail. */
2650 	if (pci_msi_blacklisted())
2651 		return (ENXIO);
2652 
2653 	/* MSI capability present? */
2654 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2655 		return (ENODEV);
2656 
2657 	ctrl = pci_read_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, 2);
2658 	msgnum = PCI_MSI_MSGNUM(ctrl);
2659 	if (bootverbose)
2660 		device_printf(child,
2661 		    "attempting to allocate %d MSI vectors (%u supported)\n",
2662 		    *count, msgnum);
2663 
2664 	/* Don't ask for more than the device supports. */
2665 	actual = min(*count, msgnum);
2666 
2667 	/* Don't ask for more than 32 messages. */
2668 	actual = min(actual, 32);
2669 
2670 	/* MSI requires power of 2 number of messages. */
2671 	if (!powerof2(actual))
2672 		return (EINVAL);
2673 
2674 	for (;;) {
2675 		/* Try to allocate N messages. */
2676 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2677 		    actual, irqs);
2678 		if (error == 0)
2679 			break;
2680 		if (actual == 1)
2681 			return (error);
2682 
2683 		/* Try N / 2. */
2684 		actual >>= 1;
2685 	}
2686 
2687 	/*
2688 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2689 	 * resources in the irqs[] array, so add new resources
2690 	 * starting at rid 1.
2691 	 */
2692 	for (i = 0; i < actual; i++)
2693 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2694 		    irqs[i], irqs[i], 1);
2695 
2696 	if (bootverbose) {
2697 		if (actual == 1)
2698 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2699 		else {
2700 			bool run;
2701 
2702 			/*
2703 			 * Be fancy and try to print contiguous runs
2704 			 * of IRQ values as ranges.  'run' is true if
2705 			 * we are in a range.
2706 			 */
2707 			device_printf(child, "using IRQs %d", irqs[0]);
2708 			run = false;
2709 			for (i = 1; i < actual; i++) {
2710 				/* Still in a run? */
2711 				if (irqs[i] == irqs[i - 1] + 1) {
2712 					run = true;
2713 					continue;
2714 				}
2715 
2716 				/* Finish previous range. */
2717 				if (run) {
2718 					printf("-%d", irqs[i - 1]);
2719 					run = false;
2720 				}
2721 
2722 				/* Start new range. */
2723 				printf(",%d", irqs[i]);
2724 			}
2725 
2726 			/* Unfinished range? */
2727 			if (run)
2728 				printf("-%d", irqs[actual - 1]);
2729 			printf(" for MSI\n");
2730 		}
2731 	}
2732 
2733 	/* Update control register with actual count. */
2734 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2735 	ctrl |= (ffs(actual) - 1) << 4;
2736 	cfg->msi.msi_ctrl = ctrl;
2737 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2738 
2739 	/* Update counts of alloc'd messages. */
2740 	cfg->msi.msi_alloc = actual;
2741 	cfg->msi.msi_handlers = 0;
2742 	*count = actual;
2743 	return (0);
2744 }
2745 
2746 /* Release the MSI messages associated with this device. */
2747 int
pci_release_msi_method(device_t dev,device_t child)2748 pci_release_msi_method(device_t dev, device_t child)
2749 {
2750 	struct pci_devinfo *dinfo = device_get_ivars(child);
2751 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2752 	struct resource_list_entry *rle;
2753 	u_int i, irqs[32];
2754 	int error;
2755 
2756 	/* Try MSI-X first. */
2757 	error = pci_release_msix(dev, child);
2758 	if (error != ENODEV)
2759 		return (error);
2760 
2761 	/* Do we have any messages to release? */
2762 	if (msi->msi_alloc == 0)
2763 		return (ENODEV);
2764 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2765 
2766 	/* Make sure none of the resources are allocated. */
2767 	if (msi->msi_handlers > 0)
2768 		return (EBUSY);
2769 	for (i = 0; i < msi->msi_alloc; i++) {
2770 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2771 		KASSERT(rle != NULL, ("missing MSI resource"));
2772 		if (rle->res != NULL)
2773 			return (EBUSY);
2774 		irqs[i] = rle->start;
2775 	}
2776 
2777 	/* Update control register with 0 count. */
2778 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2779 	    ("%s: MSI still enabled", __func__));
2780 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2781 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2782 	    msi->msi_ctrl, 2);
2783 
2784 	/* Release the messages. */
2785 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2786 	for (i = 0; i < msi->msi_alloc; i++)
2787 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2788 
2789 	/* Update alloc count. */
2790 	msi->msi_alloc = 0;
2791 	msi->msi_addr = 0;
2792 	msi->msi_data = 0;
2793 	return (0);
2794 }
2795 
2796 /*
2797  * Return the max supported MSI messages this device supports.
2798  * Basically, assuming the MD code can alloc messages, this function
2799  * should return the maximum value that pci_alloc_msi() can return.
2800  * Thus, it is subject to the tunables, etc.
2801  */
2802 int
pci_msi_count_method(device_t dev,device_t child)2803 pci_msi_count_method(device_t dev, device_t child)
2804 {
2805 	struct pci_devinfo *dinfo = device_get_ivars(child);
2806 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2807 	uint16_t ctrl;
2808 
2809 	if (pci_do_msi && msi->msi_location != 0) {
2810 		ctrl = pci_read_config(child, msi->msi_location + PCIR_MSI_CTRL,
2811 		    2);
2812 		return (PCI_MSI_MSGNUM(ctrl));
2813 	}
2814 	return (0);
2815 }
2816 
2817 /* free pcicfgregs structure and all depending data structures */
2818 
2819 int
pci_freecfg(struct pci_devinfo * dinfo)2820 pci_freecfg(struct pci_devinfo *dinfo)
2821 {
2822 	struct devlist *devlist_head;
2823 	struct pci_map *pm, *next;
2824 
2825 	devlist_head = &pci_devq;
2826 
2827 	if (dinfo->cfg.vpd.vpd_reg)
2828 		vpd_free(&dinfo->cfg.vpd);
2829 
2830 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2831 		free(pm, M_DEVBUF);
2832 	}
2833 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2834 	free(dinfo, M_DEVBUF);
2835 
2836 	/* increment the generation count */
2837 	pci_generation++;
2838 
2839 	/* we're losing one device */
2840 	pci_numdevs--;
2841 	return (0);
2842 }
2843 
2844 /*
2845  * PCI power manangement
2846  */
2847 int
pci_set_powerstate_method(device_t dev,device_t child,int state)2848 pci_set_powerstate_method(device_t dev, device_t child, int state)
2849 {
2850 	struct pci_devinfo *dinfo = device_get_ivars(child);
2851 	pcicfgregs *cfg = &dinfo->cfg;
2852 	uint16_t status;
2853 	int oldstate, highest, delay;
2854 
2855 	if (cfg->pp.pp_location == 0)
2856 		return (EOPNOTSUPP);
2857 
2858 	/*
2859 	 * Optimize a no state change request away.  While it would be OK to
2860 	 * write to the hardware in theory, some devices have shown odd
2861 	 * behavior when going from D3 -> D3.
2862 	 */
2863 	oldstate = pci_get_powerstate(child);
2864 	if (oldstate == state)
2865 		return (0);
2866 
2867 	/*
2868 	 * The PCI power management specification states that after a state
2869 	 * transition between PCI power states, system software must
2870 	 * guarantee a minimal delay before the function accesses the device.
2871 	 * Compute the worst case delay that we need to guarantee before we
2872 	 * access the device.  Many devices will be responsive much more
2873 	 * quickly than this delay, but there are some that don't respond
2874 	 * instantly to state changes.  Transitions to/from D3 state require
2875 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2876 	 * is done below with DELAY rather than a sleeper function because
2877 	 * this function can be called from contexts where we cannot sleep.
2878 	 */
2879 	highest = (oldstate > state) ? oldstate : state;
2880 	if (highest == PCI_POWERSTATE_D3)
2881 	    delay = 10000;
2882 	else if (highest == PCI_POWERSTATE_D2)
2883 	    delay = 200;
2884 	else
2885 	    delay = 0;
2886 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2887 	    PCIR_POWER_STATUS, 2) & ~PCIM_PSTAT_DMASK;
2888 	switch (state) {
2889 	case PCI_POWERSTATE_D0:
2890 		status |= PCIM_PSTAT_D0;
2891 		break;
2892 	case PCI_POWERSTATE_D1:
2893 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2894 			return (EOPNOTSUPP);
2895 		status |= PCIM_PSTAT_D1;
2896 		break;
2897 	case PCI_POWERSTATE_D2:
2898 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2899 			return (EOPNOTSUPP);
2900 		status |= PCIM_PSTAT_D2;
2901 		break;
2902 	case PCI_POWERSTATE_D3:
2903 		status |= PCIM_PSTAT_D3;
2904 		break;
2905 	default:
2906 		return (EINVAL);
2907 	}
2908 
2909 	if (bootverbose)
2910 		pci_printf(cfg, "Transition from %s to %s\n",
2911 		    pci_powerstate_to_str(oldstate),
2912 		    pci_powerstate_to_str(state));
2913 
2914 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_location + PCIR_POWER_STATUS,
2915 	    status, 2);
2916 	if (delay)
2917 		DELAY(delay);
2918 	return (0);
2919 }
2920 
2921 int
pci_get_powerstate_method(device_t dev,device_t child)2922 pci_get_powerstate_method(device_t dev, device_t child)
2923 {
2924 	struct pci_devinfo *dinfo = device_get_ivars(child);
2925 	pcicfgregs *cfg = &dinfo->cfg;
2926 	uint16_t status;
2927 	int result;
2928 
2929 	if (cfg->pp.pp_location != 0) {
2930 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2931 		    PCIR_POWER_STATUS, 2);
2932 		switch (status & PCIM_PSTAT_DMASK) {
2933 		case PCIM_PSTAT_D0:
2934 			result = PCI_POWERSTATE_D0;
2935 			break;
2936 		case PCIM_PSTAT_D1:
2937 			result = PCI_POWERSTATE_D1;
2938 			break;
2939 		case PCIM_PSTAT_D2:
2940 			result = PCI_POWERSTATE_D2;
2941 			break;
2942 		case PCIM_PSTAT_D3:
2943 			result = PCI_POWERSTATE_D3;
2944 			break;
2945 		default:
2946 			result = PCI_POWERSTATE_UNKNOWN;
2947 			break;
2948 		}
2949 	} else {
2950 		/* No support, device is always at D0 */
2951 		result = PCI_POWERSTATE_D0;
2952 	}
2953 	return (result);
2954 }
2955 
2956 /* Clear any active PME# and disable PME# generation. */
2957 void
pci_clear_pme(device_t dev)2958 pci_clear_pme(device_t dev)
2959 {
2960 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2961 	pcicfgregs *cfg = &dinfo->cfg;
2962 	uint16_t status;
2963 
2964 	if (cfg->pp.pp_location != 0) {
2965 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2966 		    PCIR_POWER_STATUS, 2);
2967 		status &= ~PCIM_PSTAT_PMEENABLE;
2968 		status |= PCIM_PSTAT_PME;
2969 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2970 		    PCIR_POWER_STATUS, status, 2);
2971 	}
2972 }
2973 
2974 /* Clear any active PME# and enable PME# generation. */
2975 void
pci_enable_pme(device_t dev)2976 pci_enable_pme(device_t dev)
2977 {
2978 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2979 	pcicfgregs *cfg = &dinfo->cfg;
2980 	uint16_t status;
2981 
2982 	if (cfg->pp.pp_location != 0) {
2983 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2984 		    PCIR_POWER_STATUS, 2);
2985 		status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2986 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2987 		    PCIR_POWER_STATUS, status, 2);
2988 	}
2989 }
2990 
2991 bool
pci_has_pm(device_t dev)2992 pci_has_pm(device_t dev)
2993 {
2994 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2995 	pcicfgregs *cfg = &dinfo->cfg;
2996 
2997 	return (cfg->pp.pp_location != 0);
2998 }
2999 
3000 /*
3001  * Some convenience functions for PCI device drivers.
3002  */
3003 
3004 static __inline void
pci_set_command_bit(device_t dev,device_t child,uint16_t bit)3005 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
3006 {
3007 	uint16_t	command;
3008 
3009 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3010 	command |= bit;
3011 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3012 }
3013 
3014 static __inline void
pci_clear_command_bit(device_t dev,device_t child,uint16_t bit)3015 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
3016 {
3017 	uint16_t	command;
3018 
3019 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3020 	command &= ~bit;
3021 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3022 }
3023 
3024 int
pci_enable_busmaster_method(device_t dev,device_t child)3025 pci_enable_busmaster_method(device_t dev, device_t child)
3026 {
3027 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3028 	return (0);
3029 }
3030 
3031 int
pci_disable_busmaster_method(device_t dev,device_t child)3032 pci_disable_busmaster_method(device_t dev, device_t child)
3033 {
3034 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3035 	return (0);
3036 }
3037 
3038 int
pci_enable_io_method(device_t dev,device_t child,int space)3039 pci_enable_io_method(device_t dev, device_t child, int space)
3040 {
3041 	uint16_t bit;
3042 
3043 	switch(space) {
3044 	case SYS_RES_IOPORT:
3045 		bit = PCIM_CMD_PORTEN;
3046 		break;
3047 	case SYS_RES_MEMORY:
3048 		bit = PCIM_CMD_MEMEN;
3049 		break;
3050 	default:
3051 		return (EINVAL);
3052 	}
3053 	pci_set_command_bit(dev, child, bit);
3054 	return (0);
3055 }
3056 
3057 int
pci_disable_io_method(device_t dev,device_t child,int space)3058 pci_disable_io_method(device_t dev, device_t child, int space)
3059 {
3060 	uint16_t bit;
3061 
3062 	switch(space) {
3063 	case SYS_RES_IOPORT:
3064 		bit = PCIM_CMD_PORTEN;
3065 		break;
3066 	case SYS_RES_MEMORY:
3067 		bit = PCIM_CMD_MEMEN;
3068 		break;
3069 	default:
3070 		return (EINVAL);
3071 	}
3072 	pci_clear_command_bit(dev, child, bit);
3073 	return (0);
3074 }
3075 
3076 /*
3077  * New style pci driver.  Parent device is either a pci-host-bridge or a
3078  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
3079  */
3080 
3081 void
pci_print_verbose(struct pci_devinfo * dinfo)3082 pci_print_verbose(struct pci_devinfo *dinfo)
3083 {
3084 
3085 	if (bootverbose) {
3086 		pcicfgregs *cfg = &dinfo->cfg;
3087 
3088 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
3089 		    cfg->vendor, cfg->device, cfg->revid);
3090 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
3091 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
3092 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
3093 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
3094 		    cfg->mfdev);
3095 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
3096 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
3097 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
3098 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3099 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3100 		if (cfg->intpin > 0)
3101 			printf("\tintpin=%c, irq=%d\n",
3102 			    cfg->intpin +'a' -1, cfg->intline);
3103 		if (cfg->pp.pp_location) {
3104 			uint16_t status;
3105 
3106 			status = pci_read_config(cfg->dev, cfg->pp.pp_location +
3107 			    PCIR_POWER_STATUS, 2);
3108 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3109 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3110 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3111 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3112 			    status & PCIM_PSTAT_DMASK);
3113 		}
3114 		if (cfg->msi.msi_location) {
3115 			uint16_t ctrl, msgnum;
3116 
3117 			ctrl = cfg->msi.msi_ctrl;
3118 			msgnum = PCI_MSI_MSGNUM(ctrl);
3119 			printf("\tMSI supports %d message%s%s%s\n",
3120 			    msgnum, (msgnum == 1) ? "" : "s",
3121 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3122 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3123 		}
3124 		if (cfg->msix.msix_location) {
3125 			uint16_t msgnum;
3126 
3127 			msgnum = PCI_MSIX_MSGNUM(cfg->msix.msix_ctrl);
3128 			printf("\tMSI-X supports %d message%s ",
3129 			    msgnum, (msgnum == 1) ? "" : "s");
3130 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3131 				printf("in map 0x%x\n",
3132 				    cfg->msix.msix_table_bar);
3133 			else
3134 				printf("in maps 0x%x and 0x%x\n",
3135 				    cfg->msix.msix_table_bar,
3136 				    cfg->msix.msix_pba_bar);
3137 		}
3138 	}
3139 }
3140 
3141 static int
pci_porten(device_t dev)3142 pci_porten(device_t dev)
3143 {
3144 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3145 }
3146 
3147 static int
pci_memen(device_t dev)3148 pci_memen(device_t dev)
3149 {
3150 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3151 }
3152 
3153 void
pci_read_bar(device_t dev,int reg,pci_addr_t * mapp,pci_addr_t * testvalp,int * bar64)3154 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3155     int *bar64)
3156 {
3157 	struct pci_devinfo *dinfo;
3158 	pci_addr_t map, testval;
3159 	int ln2range;
3160 	uint16_t cmd;
3161 
3162 	/*
3163 	 * The device ROM BAR is special.  It is always a 32-bit
3164 	 * memory BAR.  Bit 0 is special and should not be set when
3165 	 * sizing the BAR.
3166 	 */
3167 	dinfo = device_get_ivars(dev);
3168 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3169 		map = pci_read_config(dev, reg, 4);
3170 		pci_write_config(dev, reg, 0xfffffffe, 4);
3171 		testval = pci_read_config(dev, reg, 4);
3172 		pci_write_config(dev, reg, map, 4);
3173 		*mapp = map;
3174 		*testvalp = testval;
3175 		if (bar64 != NULL)
3176 			*bar64 = 0;
3177 		return;
3178 	}
3179 
3180 	map = pci_read_config(dev, reg, 4);
3181 	ln2range = pci_maprange(map);
3182 	if (ln2range == 64)
3183 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3184 
3185 	/*
3186 	 * Disable decoding via the command register before
3187 	 * determining the BAR's length since we will be placing it in
3188 	 * a weird state.
3189 	 */
3190 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3191 	pci_write_config(dev, PCIR_COMMAND,
3192 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3193 
3194 	/*
3195 	 * Determine the BAR's length by writing all 1's.  The bottom
3196 	 * log_2(size) bits of the BAR will stick as 0 when we read
3197 	 * the value back.
3198 	 *
3199 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3200 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3201 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3202 	 *
3203 	 * Writes to both registers must be performed before attempting to
3204 	 * read back the size value.
3205 	 */
3206 	testval = 0;
3207 	pci_write_config(dev, reg, 0xffffffff, 4);
3208 	if (ln2range == 64) {
3209 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3210 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3211 	}
3212 	testval |= pci_read_config(dev, reg, 4);
3213 
3214 	/*
3215 	 * Restore the original value of the BAR.  We may have reprogrammed
3216 	 * the BAR of the low-level console device and when booting verbose,
3217 	 * we need the console device addressable.
3218 	 */
3219 	pci_write_config(dev, reg, map, 4);
3220 	if (ln2range == 64)
3221 		pci_write_config(dev, reg + 4, map >> 32, 4);
3222 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3223 
3224 	*mapp = map;
3225 	*testvalp = testval;
3226 	if (bar64 != NULL)
3227 		*bar64 = (ln2range == 64);
3228 }
3229 
3230 static void
pci_write_bar(device_t dev,struct pci_map * pm,pci_addr_t base)3231 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3232 {
3233 	struct pci_devinfo *dinfo;
3234 	int ln2range;
3235 
3236 	/* The device ROM BAR is always a 32-bit memory BAR. */
3237 	dinfo = device_get_ivars(dev);
3238 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3239 		ln2range = 32;
3240 	else
3241 		ln2range = pci_maprange(pm->pm_value);
3242 	pci_write_config(dev, pm->pm_reg, base, 4);
3243 	if (ln2range == 64)
3244 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3245 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3246 	if (ln2range == 64)
3247 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3248 		    pm->pm_reg + 4, 4) << 32;
3249 }
3250 
3251 struct pci_map *
pci_find_bar(device_t dev,int reg)3252 pci_find_bar(device_t dev, int reg)
3253 {
3254 	struct pci_devinfo *dinfo;
3255 	struct pci_map *pm;
3256 
3257 	dinfo = device_get_ivars(dev);
3258 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3259 		if (pm->pm_reg == reg)
3260 			return (pm);
3261 	}
3262 	return (NULL);
3263 }
3264 
3265 struct pci_map *
pci_first_bar(device_t dev)3266 pci_first_bar(device_t dev)
3267 {
3268 	struct pci_devinfo *dinfo;
3269 
3270 	dinfo = device_get_ivars(dev);
3271 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3272 }
3273 
3274 struct pci_map *
pci_next_bar(struct pci_map * pm)3275 pci_next_bar(struct pci_map *pm)
3276 {
3277 	return (STAILQ_NEXT(pm, pm_link));
3278 }
3279 
3280 int
pci_bar_enabled(device_t dev,struct pci_map * pm)3281 pci_bar_enabled(device_t dev, struct pci_map *pm)
3282 {
3283 	struct pci_devinfo *dinfo;
3284 	uint16_t cmd;
3285 
3286 	dinfo = device_get_ivars(dev);
3287 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3288 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3289 		return (0);
3290 #ifdef PCI_IOV
3291 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3292 		struct pcicfg_iov *iov;
3293 
3294 		iov = dinfo->cfg.iov;
3295 		cmd = pci_read_config(iov->iov_pf,
3296 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3297 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3298 	}
3299 #endif
3300 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3301 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3302 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3303 	else
3304 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3305 }
3306 
3307 struct pci_map *
pci_add_bar(device_t dev,int reg,pci_addr_t value,pci_addr_t size)3308 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3309 {
3310 	struct pci_devinfo *dinfo;
3311 	struct pci_map *pm, *prev;
3312 
3313 	dinfo = device_get_ivars(dev);
3314 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3315 	pm->pm_reg = reg;
3316 	pm->pm_value = value;
3317 	pm->pm_size = size;
3318 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3319 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3320 		    reg));
3321 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3322 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3323 			break;
3324 	}
3325 	if (prev != NULL)
3326 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3327 	else
3328 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3329 	return (pm);
3330 }
3331 
3332 static void
pci_restore_bars(device_t dev)3333 pci_restore_bars(device_t dev)
3334 {
3335 	struct pci_devinfo *dinfo;
3336 	struct pci_map *pm;
3337 	int ln2range;
3338 
3339 	dinfo = device_get_ivars(dev);
3340 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3341 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3342 			ln2range = 32;
3343 		else
3344 			ln2range = pci_maprange(pm->pm_value);
3345 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3346 		if (ln2range == 64)
3347 			pci_write_config(dev, pm->pm_reg + 4,
3348 			    pm->pm_value >> 32, 4);
3349 	}
3350 }
3351 
3352 /*
3353  * Add a resource based on a pci map register. Return 1 if the map
3354  * register is a 32bit map register or 2 if it is a 64bit register.
3355  */
3356 static int
pci_add_map(device_t bus,device_t dev,int reg,struct resource_list * rl,int force,int prefetch)3357 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3358     int force, int prefetch)
3359 {
3360 	struct pci_map *pm;
3361 	pci_addr_t base, map, testval;
3362 	pci_addr_t start, end, count;
3363 	int barlen, basezero, flags, maprange, mapsize, type;
3364 	uint16_t cmd;
3365 	struct resource *res;
3366 
3367 	/*
3368 	 * The BAR may already exist if the device is a CardBus card
3369 	 * whose CIS is stored in this BAR.
3370 	 */
3371 	pm = pci_find_bar(dev, reg);
3372 	if (pm != NULL) {
3373 		maprange = pci_maprange(pm->pm_value);
3374 		barlen = maprange == 64 ? 2 : 1;
3375 		return (barlen);
3376 	}
3377 
3378 	pci_read_bar(dev, reg, &map, &testval, NULL);
3379 	if (PCI_BAR_MEM(map)) {
3380 		type = SYS_RES_MEMORY;
3381 		if (map & PCIM_BAR_MEM_PREFETCH)
3382 			prefetch = 1;
3383 	} else
3384 		type = SYS_RES_IOPORT;
3385 	mapsize = pci_mapsize(testval);
3386 	base = pci_mapbase(map);
3387 #ifdef __PCI_BAR_ZERO_VALID
3388 	basezero = 0;
3389 #else
3390 	basezero = base == 0;
3391 #endif
3392 	maprange = pci_maprange(map);
3393 	barlen = maprange == 64 ? 2 : 1;
3394 
3395 	/*
3396 	 * For I/O registers, if bottom bit is set, and the next bit up
3397 	 * isn't clear, we know we have a BAR that doesn't conform to the
3398 	 * spec, so ignore it.  Also, sanity check the size of the data
3399 	 * areas to the type of memory involved.  Memory must be at least
3400 	 * 16 bytes in size, while I/O ranges must be at least 4.
3401 	 */
3402 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3403 		return (barlen);
3404 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3405 	    (type == SYS_RES_IOPORT && mapsize < 2))
3406 		return (barlen);
3407 
3408 	/* Save a record of this BAR. */
3409 	pm = pci_add_bar(dev, reg, map, mapsize);
3410 	if (bootverbose) {
3411 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3412 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3413 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3414 			printf(", port disabled\n");
3415 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3416 			printf(", memory disabled\n");
3417 		else
3418 			printf(", enabled\n");
3419 	}
3420 
3421 	/*
3422 	 * If base is 0, then we have problems if this architecture does
3423 	 * not allow that.  It is best to ignore such entries for the
3424 	 * moment.  These will be allocated later if the driver specifically
3425 	 * requests them.  However, some removable buses look better when
3426 	 * all resources are allocated, so allow '0' to be overridden.
3427 	 *
3428 	 * Similarly treat maps whose values is the same as the test value
3429 	 * read back.  These maps have had all f's written to them by the
3430 	 * BIOS in an attempt to disable the resources.
3431 	 */
3432 	if (!force && (basezero || map == testval))
3433 		return (barlen);
3434 	if ((u_long)base != base) {
3435 		device_printf(bus,
3436 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3437 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3438 		    pci_get_function(dev), reg);
3439 		return (barlen);
3440 	}
3441 
3442 	/*
3443 	 * This code theoretically does the right thing, but has
3444 	 * undesirable side effects in some cases where peripherals
3445 	 * respond oddly to having these bits enabled.  Let the user
3446 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3447 	 * default).
3448 	 */
3449 	if (pci_enable_io_modes) {
3450 		/* Turn on resources that have been left off by a lazy BIOS */
3451 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3452 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3453 			cmd |= PCIM_CMD_PORTEN;
3454 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3455 		}
3456 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3457 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3458 			cmd |= PCIM_CMD_MEMEN;
3459 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3460 		}
3461 	} else {
3462 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3463 			return (barlen);
3464 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3465 			return (barlen);
3466 	}
3467 
3468 	count = (pci_addr_t)1 << mapsize;
3469 	flags = RF_ALIGNMENT_LOG2(mapsize);
3470 	if (prefetch)
3471 		flags |= RF_PREFETCHABLE;
3472 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3473 		start = 0;	/* Let the parent decide. */
3474 		end = ~0;
3475 	} else {
3476 		start = base;
3477 		end = base + count - 1;
3478 	}
3479 	resource_list_add(rl, type, reg, start, end, count);
3480 
3481 	/*
3482 	 * Try to allocate the resource for this BAR from our parent
3483 	 * so that this resource range is already reserved.  The
3484 	 * driver for this device will later inherit this resource in
3485 	 * pci_alloc_resource().
3486 	 */
3487 	res = resource_list_reserve(rl, bus, dev, type, reg, start, end, count,
3488 	    flags);
3489 	if ((pci_do_realloc_bars
3490 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3491 	    && res == NULL && (start != 0 || end != ~0)) {
3492 		/*
3493 		 * If the allocation fails, try to allocate a resource for
3494 		 * this BAR using any available range.  The firmware felt
3495 		 * it was important enough to assign a resource, so don't
3496 		 * disable decoding if we can help it.
3497 		 */
3498 		resource_list_delete(rl, type, reg);
3499 		resource_list_add(rl, type, reg, 0, ~0, count);
3500 		res = resource_list_reserve(rl, bus, dev, type, reg, 0, ~0,
3501 		    count, flags);
3502 	}
3503 	if (res == NULL) {
3504 		/*
3505 		 * If the allocation fails, delete the resource list entry
3506 		 * and disable decoding for this device.
3507 		 *
3508 		 * If the driver requests this resource in the future,
3509 		 * pci_reserve_map() will try to allocate a fresh
3510 		 * resource range.
3511 		 */
3512 		resource_list_delete(rl, type, reg);
3513 		pci_disable_io(dev, type);
3514 		if (bootverbose)
3515 			device_printf(bus,
3516 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3517 			    pci_get_domain(dev), pci_get_bus(dev),
3518 			    pci_get_slot(dev), pci_get_function(dev), reg);
3519 	} else {
3520 		start = rman_get_start(res);
3521 		pci_write_bar(dev, pm, start);
3522 	}
3523 	return (barlen);
3524 }
3525 
3526 /*
3527  * For ATA devices we need to decide early what addressing mode to use.
3528  * Legacy demands that the primary and secondary ATA ports sits on the
3529  * same addresses that old ISA hardware did. This dictates that we use
3530  * those addresses and ignore the BAR's if we cannot set PCI native
3531  * addressing mode.
3532  */
3533 static void
pci_ata_maps(device_t bus,device_t dev,struct resource_list * rl,int force,uint32_t prefetchmask)3534 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3535     uint32_t prefetchmask)
3536 {
3537 	int rid, type, progif;
3538 #if 0
3539 	/* if this device supports PCI native addressing use it */
3540 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3541 	if ((progif & 0x8a) == 0x8a) {
3542 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3543 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3544 			printf("Trying ATA native PCI addressing mode\n");
3545 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3546 		}
3547 	}
3548 #endif
3549 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3550 	type = SYS_RES_IOPORT;
3551 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3552 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3553 		    prefetchmask & (1 << 0));
3554 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3555 		    prefetchmask & (1 << 1));
3556 	} else {
3557 		rid = PCIR_BAR(0);
3558 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3559 		(void)resource_list_reserve(rl, bus, dev, type, rid, 0x1f0,
3560 		    0x1f7, 8, 0);
3561 		rid = PCIR_BAR(1);
3562 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3563 		(void)resource_list_reserve(rl, bus, dev, type, rid, 0x3f6,
3564 		    0x3f6, 1, 0);
3565 	}
3566 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3567 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3568 		    prefetchmask & (1 << 2));
3569 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3570 		    prefetchmask & (1 << 3));
3571 	} else {
3572 		rid = PCIR_BAR(2);
3573 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3574 		(void)resource_list_reserve(rl, bus, dev, type, rid, 0x170,
3575 		    0x177, 8, 0);
3576 		rid = PCIR_BAR(3);
3577 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3578 		(void)resource_list_reserve(rl, bus, dev, type, rid, 0x376,
3579 		    0x376, 1, 0);
3580 	}
3581 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3582 	    prefetchmask & (1 << 4));
3583 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3584 	    prefetchmask & (1 << 5));
3585 }
3586 
3587 static void
pci_assign_interrupt(device_t bus,device_t dev,int force_route)3588 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3589 {
3590 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3591 	pcicfgregs *cfg = &dinfo->cfg;
3592 	char tunable_name[64];
3593 	int irq;
3594 
3595 	/* Has to have an intpin to have an interrupt. */
3596 	if (cfg->intpin == 0)
3597 		return;
3598 
3599 	/* Let the user override the IRQ with a tunable. */
3600 	irq = PCI_INVALID_IRQ;
3601 	snprintf(tunable_name, sizeof(tunable_name),
3602 	    "hw.pci%d.%d.%d.INT%c.irq",
3603 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3604 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3605 		irq = PCI_INVALID_IRQ;
3606 
3607 	/*
3608 	 * If we didn't get an IRQ via the tunable, then we either use the
3609 	 * IRQ value in the intline register or we ask the bus to route an
3610 	 * interrupt for us.  If force_route is true, then we only use the
3611 	 * value in the intline register if the bus was unable to assign an
3612 	 * IRQ.
3613 	 */
3614 	if (!PCI_INTERRUPT_VALID(irq)) {
3615 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3616 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3617 		if (!PCI_INTERRUPT_VALID(irq))
3618 			irq = cfg->intline;
3619 	}
3620 
3621 	/* If after all that we don't have an IRQ, just bail. */
3622 	if (!PCI_INTERRUPT_VALID(irq))
3623 		return;
3624 
3625 	/* Update the config register if it changed. */
3626 	if (irq != cfg->intline) {
3627 		cfg->intline = irq;
3628 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3629 	}
3630 
3631 	/* Add this IRQ as rid 0 interrupt resource. */
3632 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3633 }
3634 
3635 /* Perform early OHCI takeover from SMM. */
3636 static void
ohci_early_takeover(device_t self)3637 ohci_early_takeover(device_t self)
3638 {
3639 	struct resource *res;
3640 	uint32_t ctl;
3641 	int rid;
3642 	int i;
3643 
3644 	rid = PCIR_BAR(0);
3645 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3646 	if (res == NULL)
3647 		return;
3648 
3649 	ctl = bus_read_4(res, OHCI_CONTROL);
3650 	if (ctl & OHCI_IR) {
3651 		if (bootverbose)
3652 			printf("ohci early: "
3653 			    "SMM active, request owner change\n");
3654 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3655 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3656 			DELAY(1000);
3657 			ctl = bus_read_4(res, OHCI_CONTROL);
3658 		}
3659 		if (ctl & OHCI_IR) {
3660 			if (bootverbose)
3661 				printf("ohci early: "
3662 				    "SMM does not respond, resetting\n");
3663 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3664 		}
3665 		/* Disable interrupts */
3666 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3667 	}
3668 
3669 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3670 }
3671 
3672 /* Perform early UHCI takeover from SMM. */
3673 static void
uhci_early_takeover(device_t self)3674 uhci_early_takeover(device_t self)
3675 {
3676 	struct resource *res;
3677 	int rid;
3678 
3679 	/*
3680 	 * Set the PIRQD enable bit and switch off all the others. We don't
3681 	 * want legacy support to interfere with us XXX Does this also mean
3682 	 * that the BIOS won't touch the keyboard anymore if it is connected
3683 	 * to the ports of the root hub?
3684 	 */
3685 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3686 
3687 	/* Disable interrupts */
3688 	rid = PCI_UHCI_BASE_REG;
3689 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3690 	if (res != NULL) {
3691 		bus_write_2(res, UHCI_INTR, 0);
3692 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3693 	}
3694 }
3695 
3696 /* Perform early EHCI takeover from SMM. */
3697 static void
ehci_early_takeover(device_t self)3698 ehci_early_takeover(device_t self)
3699 {
3700 	struct resource *res;
3701 	uint32_t cparams;
3702 	uint32_t eec;
3703 	uint8_t eecp;
3704 	uint8_t bios_sem;
3705 	uint8_t offs;
3706 	int rid;
3707 	int i;
3708 
3709 	rid = PCIR_BAR(0);
3710 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3711 	if (res == NULL)
3712 		return;
3713 
3714 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3715 
3716 	/* Synchronise with the BIOS if it owns the controller. */
3717 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3718 	    eecp = EHCI_EECP_NEXT(eec)) {
3719 		eec = pci_read_config(self, eecp, 4);
3720 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3721 			continue;
3722 		}
3723 		bios_sem = pci_read_config(self, eecp +
3724 		    EHCI_LEGSUP_BIOS_SEM, 1);
3725 		if (bios_sem == 0) {
3726 			continue;
3727 		}
3728 		if (bootverbose)
3729 			printf("ehci early: "
3730 			    "SMM active, request owner change\n");
3731 
3732 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3733 
3734 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3735 			DELAY(1000);
3736 			bios_sem = pci_read_config(self, eecp +
3737 			    EHCI_LEGSUP_BIOS_SEM, 1);
3738 		}
3739 
3740 		if (bios_sem != 0) {
3741 			if (bootverbose)
3742 				printf("ehci early: "
3743 				    "SMM does not respond\n");
3744 		}
3745 		/* Disable interrupts */
3746 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3747 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3748 	}
3749 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3750 }
3751 
3752 /* Perform early XHCI takeover from SMM. */
3753 static void
xhci_early_takeover(device_t self)3754 xhci_early_takeover(device_t self)
3755 {
3756 	struct resource *res;
3757 	uint32_t cparams;
3758 	uint32_t eec;
3759 	uint8_t eecp;
3760 	uint8_t bios_sem;
3761 	uint8_t offs;
3762 	int rid;
3763 	int i;
3764 
3765 	rid = PCIR_BAR(0);
3766 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3767 	if (res == NULL)
3768 		return;
3769 
3770 	cparams = bus_read_4(res, XHCI_HCCPARAMS1);
3771 
3772 	eec = -1;
3773 
3774 	/* Synchronise with the BIOS if it owns the controller. */
3775 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3776 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3777 		eec = bus_read_4(res, eecp);
3778 
3779 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3780 			continue;
3781 
3782 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3783 		if (bios_sem == 0)
3784 			continue;
3785 
3786 		if (bootverbose)
3787 			printf("xhci early: "
3788 			    "SMM active, request owner change\n");
3789 
3790 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3791 
3792 		/* wait a maximum of 5 second */
3793 
3794 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3795 			DELAY(1000);
3796 			bios_sem = bus_read_1(res, eecp +
3797 			    XHCI_XECP_BIOS_SEM);
3798 		}
3799 
3800 		if (bios_sem != 0) {
3801 			if (bootverbose)
3802 				printf("xhci early: "
3803 				    "SMM does not respond\n");
3804 		}
3805 
3806 		/* Disable interrupts */
3807 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3808 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3809 		bus_read_4(res, offs + XHCI_USBSTS);
3810 	}
3811 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3812 }
3813 
3814 static void
pci_reserve_secbus(device_t bus,device_t dev,pcicfgregs * cfg,struct resource_list * rl)3815 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3816     struct resource_list *rl)
3817 {
3818 	struct resource *res;
3819 	char *cp;
3820 	rman_res_t start, end, count;
3821 	int sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3822 
3823 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3824 	case PCIM_HDRTYPE_BRIDGE:
3825 		sec_reg = PCIR_SECBUS_1;
3826 		sub_reg = PCIR_SUBBUS_1;
3827 		break;
3828 	case PCIM_HDRTYPE_CARDBUS:
3829 		sec_reg = PCIR_SECBUS_2;
3830 		sub_reg = PCIR_SUBBUS_2;
3831 		break;
3832 	default:
3833 		return;
3834 	}
3835 
3836 	/*
3837 	 * If the existing bus range is valid, attempt to reserve it
3838 	 * from our parent.  If this fails for any reason, clear the
3839 	 * secbus and subbus registers.
3840 	 *
3841 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3842 	 * This would at least preserve the existing sec_bus if it is
3843 	 * valid.
3844 	 */
3845 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3846 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3847 
3848 	/* Quirk handling. */
3849 	switch (pci_get_devid(dev)) {
3850 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3851 		sup_bus = pci_read_config(dev, 0x41, 1);
3852 		if (sup_bus != 0xff) {
3853 			sec_bus = sup_bus + 1;
3854 			sub_bus = sup_bus + 1;
3855 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3856 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3857 		}
3858 		break;
3859 
3860 	case 0x00dd10de:
3861 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3862 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3863 			break;
3864 		if (strncmp(cp, "Compal", 6) != 0) {
3865 			freeenv(cp);
3866 			break;
3867 		}
3868 		freeenv(cp);
3869 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3870 			break;
3871 		if (strncmp(cp, "08A0", 4) != 0) {
3872 			freeenv(cp);
3873 			break;
3874 		}
3875 		freeenv(cp);
3876 		if (sub_bus < 0xa) {
3877 			sub_bus = 0xa;
3878 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3879 		}
3880 		break;
3881 	}
3882 
3883 	if (bootverbose)
3884 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3885 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3886 		start = sec_bus;
3887 		end = sub_bus;
3888 		count = end - start + 1;
3889 
3890 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3891 
3892 		/*
3893 		 * If requested, clear secondary bus registers in
3894 		 * bridge devices to force a complete renumbering
3895 		 * rather than reserving the existing range.  However,
3896 		 * preserve the existing size.
3897 		 */
3898 		if (pci_clear_buses)
3899 			goto clear;
3900 
3901 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, 0,
3902 		    start, end, count, 0);
3903 		if (res != NULL)
3904 			return;
3905 
3906 		if (bootverbose)
3907 			device_printf(bus,
3908 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3909 			    pci_get_domain(dev), pci_get_bus(dev),
3910 			    pci_get_slot(dev), pci_get_function(dev));
3911 	}
3912 
3913 clear:
3914 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3915 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3916 }
3917 
3918 static struct resource *
pci_alloc_secbus(device_t dev,device_t child,int rid,rman_res_t start,rman_res_t end,rman_res_t count,u_int flags)3919 pci_alloc_secbus(device_t dev, device_t child, int rid, rman_res_t start,
3920     rman_res_t end, rman_res_t count, u_int flags)
3921 {
3922 	struct pci_devinfo *dinfo;
3923 	pcicfgregs *cfg;
3924 	struct resource_list *rl;
3925 	struct resource *res;
3926 	int sec_reg, sub_reg;
3927 
3928 	dinfo = device_get_ivars(child);
3929 	cfg = &dinfo->cfg;
3930 	rl = &dinfo->resources;
3931 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3932 	case PCIM_HDRTYPE_BRIDGE:
3933 		sec_reg = PCIR_SECBUS_1;
3934 		sub_reg = PCIR_SUBBUS_1;
3935 		break;
3936 	case PCIM_HDRTYPE_CARDBUS:
3937 		sec_reg = PCIR_SECBUS_2;
3938 		sub_reg = PCIR_SUBBUS_2;
3939 		break;
3940 	default:
3941 		return (NULL);
3942 	}
3943 
3944 	if (rid != 0)
3945 		return (NULL);
3946 
3947 	if (resource_list_find(rl, PCI_RES_BUS, rid) == NULL)
3948 		resource_list_add(rl, PCI_RES_BUS, rid, start, end, count);
3949 	if (!resource_list_reserved(rl, PCI_RES_BUS, rid)) {
3950 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3951 		    start, end, count, flags & ~RF_ACTIVE);
3952 		if (res == NULL) {
3953 			resource_list_delete(rl, PCI_RES_BUS, rid);
3954 			device_printf(child, "allocating %ju bus%s failed\n",
3955 			    count, count == 1 ? "" : "es");
3956 			return (NULL);
3957 		}
3958 		if (bootverbose)
3959 			device_printf(child,
3960 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3961 			    count == 1 ? "" : "es", rman_get_start(res));
3962 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3963 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3964 	}
3965 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3966 	    end, count, flags));
3967 }
3968 
3969 static int
pci_ea_bei_to_rid(device_t dev,int bei)3970 pci_ea_bei_to_rid(device_t dev, int bei)
3971 {
3972 #ifdef PCI_IOV
3973 	struct pci_devinfo *dinfo;
3974 	int iov_pos;
3975 	struct pcicfg_iov *iov;
3976 
3977 	dinfo = device_get_ivars(dev);
3978 	iov = dinfo->cfg.iov;
3979 	if (iov != NULL)
3980 		iov_pos = iov->iov_pos;
3981 	else
3982 		iov_pos = 0;
3983 #endif
3984 
3985 	/* Check if matches BAR */
3986 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3987 	    (bei <= PCIM_EA_BEI_BAR_5))
3988 		return (PCIR_BAR(bei));
3989 
3990 	/* Check ROM */
3991 	if (bei == PCIM_EA_BEI_ROM)
3992 		return (PCIR_BIOS);
3993 
3994 #ifdef PCI_IOV
3995 	/* Check if matches VF_BAR */
3996 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3997 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3998 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3999 		    iov_pos);
4000 #endif
4001 
4002 	return (-1);
4003 }
4004 
4005 int
pci_ea_is_enabled(device_t dev,int rid)4006 pci_ea_is_enabled(device_t dev, int rid)
4007 {
4008 	struct pci_ea_entry *ea;
4009 	struct pci_devinfo *dinfo;
4010 
4011 	dinfo = device_get_ivars(dev);
4012 
4013 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4014 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
4015 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
4016 	}
4017 
4018 	return (0);
4019 }
4020 
4021 void
pci_add_resources_ea(device_t bus,device_t dev,int alloc_iov)4022 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
4023 {
4024 	struct pci_ea_entry *ea;
4025 	struct pci_devinfo *dinfo;
4026 	pci_addr_t start, end, count;
4027 	struct resource_list *rl;
4028 	int type, flags, rid;
4029 	struct resource *res;
4030 	uint32_t tmp;
4031 #ifdef PCI_IOV
4032 	struct pcicfg_iov *iov;
4033 #endif
4034 
4035 	dinfo = device_get_ivars(dev);
4036 	rl = &dinfo->resources;
4037 	flags = 0;
4038 
4039 #ifdef PCI_IOV
4040 	iov = dinfo->cfg.iov;
4041 #endif
4042 
4043 	if (dinfo->cfg.ea.ea_location == 0)
4044 		return;
4045 
4046 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4047 		/*
4048 		 * TODO: Ignore EA-BAR if is not enabled.
4049 		 *   Currently the EA implementation supports
4050 		 *   only situation, where EA structure contains
4051 		 *   predefined entries. In case they are not enabled
4052 		 *   leave them unallocated and proceed with
4053 		 *   a legacy-BAR mechanism.
4054 		 */
4055 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
4056 			continue;
4057 
4058 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
4059 		case PCIM_EA_P_MEM_PREFETCH:
4060 		case PCIM_EA_P_VF_MEM_PREFETCH:
4061 			flags = RF_PREFETCHABLE;
4062 			/* FALLTHROUGH */
4063 		case PCIM_EA_P_VF_MEM:
4064 		case PCIM_EA_P_MEM:
4065 			type = SYS_RES_MEMORY;
4066 			break;
4067 		case PCIM_EA_P_IO:
4068 			type = SYS_RES_IOPORT;
4069 			break;
4070 		default:
4071 			continue;
4072 		}
4073 
4074 		if (alloc_iov != 0) {
4075 #ifdef PCI_IOV
4076 			/* Allocating IOV, confirm BEI matches */
4077 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
4078 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
4079 				continue;
4080 #else
4081 			continue;
4082 #endif
4083 		} else {
4084 			/* Allocating BAR, confirm BEI matches */
4085 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
4086 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
4087 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
4088 				continue;
4089 		}
4090 
4091 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
4092 		if (rid < 0)
4093 			continue;
4094 
4095 		/* Skip resources already allocated by EA */
4096 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
4097 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
4098 			continue;
4099 
4100 		start = ea->eae_base;
4101 		count = ea->eae_max_offset + 1;
4102 #ifdef PCI_IOV
4103 		if (iov != NULL)
4104 			count = count * iov->iov_num_vfs;
4105 #endif
4106 		end = start + count - 1;
4107 		if (count == 0)
4108 			continue;
4109 
4110 		resource_list_add(rl, type, rid, start, end, count);
4111 		res = resource_list_reserve(rl, bus, dev, type, rid, start, end, count,
4112 		    flags);
4113 		if (res == NULL) {
4114 			resource_list_delete(rl, type, rid);
4115 
4116 			/*
4117 			 * Failed to allocate using EA, disable entry.
4118 			 * Another attempt to allocation will be performed
4119 			 * further, but this time using legacy BAR registers
4120 			 */
4121 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4122 			tmp &= ~PCIM_EA_ENABLE;
4123 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4124 
4125 			/*
4126 			 * Disabling entry might fail in case it is hardwired.
4127 			 * Read flags again to match current status.
4128 			 */
4129 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4130 
4131 			continue;
4132 		}
4133 
4134 		/* As per specification, fill BAR with zeros */
4135 		pci_write_config(dev, rid, 0, 4);
4136 	}
4137 }
4138 
4139 void
pci_add_resources(device_t bus,device_t dev,int force,uint32_t prefetchmask)4140 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4141 {
4142 	struct pci_devinfo *dinfo;
4143 	pcicfgregs *cfg;
4144 	struct resource_list *rl;
4145 	const struct pci_quirk *q;
4146 	uint32_t devid;
4147 	int i;
4148 
4149 	dinfo = device_get_ivars(dev);
4150 	cfg = &dinfo->cfg;
4151 	rl = &dinfo->resources;
4152 	devid = (cfg->device << 16) | cfg->vendor;
4153 
4154 	/* Allocate resources using Enhanced Allocation */
4155 	pci_add_resources_ea(bus, dev, 0);
4156 
4157 	/* ATA devices needs special map treatment */
4158 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4159 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4160 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4161 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4162 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4163 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4164 	else
4165 		for (i = 0; i < cfg->nummaps;) {
4166 			/* Skip resources already managed by EA */
4167 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4168 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4169 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4170 				i++;
4171 				continue;
4172 			}
4173 
4174 			/*
4175 			 * Skip quirked resources.
4176 			 */
4177 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4178 				if (q->devid == devid &&
4179 				    q->type == PCI_QUIRK_UNMAP_REG &&
4180 				    q->arg1 == PCIR_BAR(i))
4181 					break;
4182 			if (q->devid != 0) {
4183 				i++;
4184 				continue;
4185 			}
4186 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4187 			    prefetchmask & (1 << i));
4188 		}
4189 
4190 	/*
4191 	 * Add additional, quirked resources.
4192 	 */
4193 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4194 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4195 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4196 
4197 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline) &&
4198 	    pci_intx_reroute) {
4199 		/*
4200 		 * Try to re-route interrupts. Sometimes the BIOS or
4201 		 * firmware may leave bogus values in these registers.
4202 		 * If the re-route fails, then just stick with what we
4203 		 * have.
4204 		 */
4205 		pci_assign_interrupt(bus, dev, 1);
4206 	}
4207 
4208 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4209 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4210 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4211 			xhci_early_takeover(dev);
4212 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4213 			ehci_early_takeover(dev);
4214 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4215 			ohci_early_takeover(dev);
4216 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4217 			uhci_early_takeover(dev);
4218 	}
4219 
4220 	/*
4221 	 * Reserve resources for secondary bus ranges behind bridge
4222 	 * devices.
4223 	 */
4224 	pci_reserve_secbus(bus, dev, cfg, rl);
4225 }
4226 
4227 static struct pci_devinfo *
pci_identify_function(device_t pcib,device_t dev,int domain,int busno,int slot,int func)4228 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4229     int slot, int func)
4230 {
4231 	struct pci_devinfo *dinfo;
4232 
4233 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4234 	if (dinfo != NULL)
4235 		pci_add_child(dev, dinfo);
4236 
4237 	return (dinfo);
4238 }
4239 
4240 void
pci_add_children(device_t dev,int domain,int busno)4241 pci_add_children(device_t dev, int domain, int busno)
4242 {
4243 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4244 	device_t pcib = device_get_parent(dev);
4245 	struct pci_devinfo *dinfo;
4246 	int maxslots;
4247 	int s, f, pcifunchigh;
4248 	uint8_t hdrtype;
4249 	int first_func;
4250 
4251 	/*
4252 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4253 	 * enable ARI.  We must enable ARI before detecting the rest of the
4254 	 * functions on this bus as ARI changes the set of slots and functions
4255 	 * that are legal on this bus.
4256 	 */
4257 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4258 	if (dinfo != NULL && pci_enable_ari)
4259 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4260 
4261 	/*
4262 	 * Start looking for new devices on slot 0 at function 1 because we
4263 	 * just identified the device at slot 0, function 0.
4264 	 */
4265 	first_func = 1;
4266 
4267 	maxslots = PCIB_MAXSLOTS(pcib);
4268 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4269 		pcifunchigh = 0;
4270 		f = 0;
4271 		DELAY(1);
4272 
4273 		/* If function 0 is not present, skip to the next slot. */
4274 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4275 			continue;
4276 		hdrtype = REG(PCIR_HDRTYPE, 1);
4277 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4278 			continue;
4279 		if (hdrtype & PCIM_MFDEV)
4280 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4281 		for (f = first_func; f <= pcifunchigh; f++)
4282 			pci_identify_function(pcib, dev, domain, busno, s, f);
4283 	}
4284 #undef REG
4285 }
4286 
4287 int
pci_rescan_method(device_t dev)4288 pci_rescan_method(device_t dev)
4289 {
4290 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4291 	device_t pcib = device_get_parent(dev);
4292 	device_t child, *devlist, *unchanged;
4293 	int devcount, error, i, j, maxslots, oldcount;
4294 	int busno, domain, s, f, pcifunchigh;
4295 	uint8_t hdrtype;
4296 
4297 	/* No need to check for ARI on a rescan. */
4298 	error = device_get_children(dev, &devlist, &devcount);
4299 	if (error)
4300 		return (error);
4301 	if (devcount != 0) {
4302 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4303 		    M_NOWAIT | M_ZERO);
4304 		if (unchanged == NULL) {
4305 			free(devlist, M_TEMP);
4306 			return (ENOMEM);
4307 		}
4308 	} else
4309 		unchanged = NULL;
4310 
4311 	domain = pcib_get_domain(dev);
4312 	busno = pcib_get_bus(dev);
4313 	maxslots = PCIB_MAXSLOTS(pcib);
4314 	for (s = 0; s <= maxslots; s++) {
4315 		/* If function 0 is not present, skip to the next slot. */
4316 		f = 0;
4317 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4318 			continue;
4319 		pcifunchigh = 0;
4320 		hdrtype = REG(PCIR_HDRTYPE, 1);
4321 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4322 			continue;
4323 		if (hdrtype & PCIM_MFDEV)
4324 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4325 		for (f = 0; f <= pcifunchigh; f++) {
4326 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4327 				continue;
4328 
4329 			/*
4330 			 * Found a valid function.  Check if a
4331 			 * device_t for this device already exists.
4332 			 */
4333 			for (i = 0; i < devcount; i++) {
4334 				child = devlist[i];
4335 				if (child == NULL)
4336 					continue;
4337 				if (pci_get_slot(child) == s &&
4338 				    pci_get_function(child) == f) {
4339 					unchanged[i] = child;
4340 					goto next_func;
4341 				}
4342 			}
4343 
4344 			pci_identify_function(pcib, dev, domain, busno, s, f);
4345 		next_func:;
4346 		}
4347 	}
4348 
4349 	/* Remove devices that are no longer present. */
4350 	for (i = 0; i < devcount; i++) {
4351 		if (unchanged[i] != NULL)
4352 			continue;
4353 		device_delete_child(dev, devlist[i]);
4354 	}
4355 
4356 	free(devlist, M_TEMP);
4357 	oldcount = devcount;
4358 
4359 	/* Try to attach the devices just added. */
4360 	error = device_get_children(dev, &devlist, &devcount);
4361 	if (error) {
4362 		free(unchanged, M_TEMP);
4363 		return (error);
4364 	}
4365 
4366 	for (i = 0; i < devcount; i++) {
4367 		for (j = 0; j < oldcount; j++) {
4368 			if (devlist[i] == unchanged[j])
4369 				goto next_device;
4370 		}
4371 
4372 		device_probe_and_attach(devlist[i]);
4373 	next_device:;
4374 	}
4375 
4376 	free(unchanged, M_TEMP);
4377 	free(devlist, M_TEMP);
4378 	return (0);
4379 #undef REG
4380 }
4381 
4382 #ifdef PCI_IOV
4383 device_t
pci_add_iov_child(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4384 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4385     uint16_t did)
4386 {
4387 	struct pci_devinfo *vf_dinfo;
4388 	device_t pcib;
4389 	int busno, slot, func;
4390 
4391 	pcib = device_get_parent(bus);
4392 
4393 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4394 
4395 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4396 	    slot, func, vid, did);
4397 
4398 	vf_dinfo->cfg.flags |= PCICFG_VF;
4399 	pci_add_child(bus, vf_dinfo);
4400 
4401 	return (vf_dinfo->cfg.dev);
4402 }
4403 
4404 device_t
pci_create_iov_child_method(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4405 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4406     uint16_t vid, uint16_t did)
4407 {
4408 
4409 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4410 }
4411 #endif
4412 
4413 /*
4414  * For PCIe device set Max_Payload_Size to match PCIe root's.
4415  */
4416 static void
pcie_setup_mps(device_t dev)4417 pcie_setup_mps(device_t dev)
4418 {
4419 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4420 	device_t root;
4421 	uint16_t rmps, mmps, mps;
4422 
4423 	if (dinfo->cfg.pcie.pcie_location == 0)
4424 		return;
4425 	root = pci_find_pcie_root_port(dev);
4426 	if (root == NULL)
4427 		return;
4428 	/* Check whether the MPS is already configured. */
4429 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4430 	    PCIEM_CTL_MAX_PAYLOAD;
4431 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4432 	    PCIEM_CTL_MAX_PAYLOAD;
4433 	if (mps == rmps)
4434 		return;
4435 	/* Check whether the device is capable of the root's MPS. */
4436 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4437 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4438 	if (rmps > mmps) {
4439 		/*
4440 		 * The device is unable to handle root's MPS.  Limit root.
4441 		 * XXX: We should traverse through all the tree, applying
4442 		 * it to all the devices.
4443 		 */
4444 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4445 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4446 	} else {
4447 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4448 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4449 	}
4450 }
4451 
4452 static void
pci_add_child_clear_aer(device_t dev,struct pci_devinfo * dinfo)4453 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4454 {
4455 	int aer;
4456 	uint32_t r;
4457 	uint16_t r2;
4458 
4459 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4460 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4461 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4462 		    PCIER_ROOT_CTL, 2);
4463 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4464 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4465 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4466 		    PCIER_ROOT_CTL, r2, 2);
4467 	}
4468 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4469 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4470 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4471 		if (r != 0 && bootverbose) {
4472 			pci_printf(&dinfo->cfg,
4473 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4474 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4475 			    4));
4476 		}
4477 
4478 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4479 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4480 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4481 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4482 		    PCIM_AER_UC_POISONED_TLP |
4483 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4484 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4485 		    PCIM_AER_UC_COMPLETER_ABORT |
4486 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4487 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4488 		    PCIM_AER_UC_MALFORMED_TLP |
4489 		    PCIM_AER_UC_ECRC_ERROR |
4490 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4491 		    PCIM_AER_UC_ACS_VIOLATION |
4492 		    PCIM_AER_UC_INTERNAL_ERROR |
4493 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4494 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4495 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4496 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4497 
4498 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4499 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4500 		if (r != 0 && bootverbose) {
4501 			pci_printf(&dinfo->cfg,
4502 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4503 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4504 			    4));
4505 		}
4506 
4507 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4508 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4509 		    PCIM_AER_COR_BAD_TLP |
4510 		    PCIM_AER_COR_BAD_DLLP |
4511 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4512 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4513 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4514 		    PCIM_AER_COR_INTERNAL_ERROR |
4515 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4516 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4517 
4518 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4519 		    PCIER_DEVICE_CTL, 2);
4520 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4521 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4522 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4523 		    PCIER_DEVICE_CTL, r, 2);
4524 	}
4525 }
4526 
4527 void
pci_add_child(device_t bus,struct pci_devinfo * dinfo)4528 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4529 {
4530 	device_t dev;
4531 
4532 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4533 	device_set_ivars(dev, dinfo);
4534 	resource_list_init(&dinfo->resources);
4535 	pci_cfg_save(dev, dinfo, 0);
4536 	pci_cfg_restore(dev, dinfo);
4537 	pci_clear_pme(dev);
4538 	pci_print_verbose(dinfo);
4539 	pci_add_resources(bus, dev, 0, 0);
4540 	if (pci_enable_mps_tune)
4541 		pcie_setup_mps(dev);
4542 	pci_child_added(dinfo->cfg.dev);
4543 
4544 	if (pci_clear_aer_on_attach)
4545 		pci_add_child_clear_aer(dev, dinfo);
4546 
4547 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4548 }
4549 
4550 void
pci_child_added_method(device_t dev,device_t child)4551 pci_child_added_method(device_t dev, device_t child)
4552 {
4553 
4554 }
4555 
4556 static int
pci_probe(device_t dev)4557 pci_probe(device_t dev)
4558 {
4559 
4560 	device_set_desc(dev, "PCI bus");
4561 
4562 	/* Allow other subclasses to override this driver. */
4563 	return (BUS_PROBE_GENERIC);
4564 }
4565 
4566 int
pci_attach_common(device_t dev)4567 pci_attach_common(device_t dev)
4568 {
4569 	struct pci_softc *sc;
4570 	int busno, domain;
4571 	int rid;
4572 
4573 	sc = device_get_softc(dev);
4574 	domain = pcib_get_domain(dev);
4575 	busno = pcib_get_bus(dev);
4576 	rid = 0;
4577 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4578 	    1, 0);
4579 	if (sc->sc_bus == NULL) {
4580 		device_printf(dev, "failed to allocate bus number\n");
4581 		return (ENXIO);
4582 	}
4583 	if (bootverbose)
4584 		device_printf(dev, "domain=%d, physical bus=%d\n",
4585 		    domain, busno);
4586 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4587 	return (0);
4588 }
4589 
4590 int
pci_attach(device_t dev)4591 pci_attach(device_t dev)
4592 {
4593 	int busno, domain, error;
4594 
4595 	error = pci_attach_common(dev);
4596 	if (error)
4597 		return (error);
4598 
4599 	/*
4600 	 * Since there can be multiple independently numbered PCI
4601 	 * buses on systems with multiple PCI domains, we can't use
4602 	 * the unit number to decide which bus we are probing. We ask
4603 	 * the parent pcib what our domain and bus numbers are.
4604 	 */
4605 	domain = pcib_get_domain(dev);
4606 	busno = pcib_get_bus(dev);
4607 	pci_add_children(dev, domain, busno);
4608 	bus_attach_children(dev);
4609 	return (0);
4610 }
4611 
4612 int
pci_detach(device_t dev)4613 pci_detach(device_t dev)
4614 {
4615 	struct pci_softc *sc;
4616 	int error;
4617 
4618 	error = bus_generic_detach(dev);
4619 	if (error)
4620 		return (error);
4621 	sc = device_get_softc(dev);
4622 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4623 	return (error);
4624 }
4625 
4626 static void
pci_hint_device_unit(device_t dev,device_t child,const char * name,int * unitp)4627 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4628 {
4629 	int line, unit;
4630 	const char *at;
4631 	char me1[24], me2[32];
4632 	uint8_t b, s, f;
4633 	uint32_t d;
4634 	device_location_cache_t *cache;
4635 
4636 	d = pci_get_domain(child);
4637 	b = pci_get_bus(child);
4638 	s = pci_get_slot(child);
4639 	f = pci_get_function(child);
4640 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4641 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4642 	line = 0;
4643 	cache = dev_wired_cache_init();
4644 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4645 		resource_string_value(name, unit, "at", &at);
4646 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4647 			*unitp = unit;
4648 			break;
4649 		}
4650 		if (dev_wired_cache_match(cache, child, at)) {
4651 			*unitp = unit;
4652 			break;
4653 		}
4654 	}
4655 	dev_wired_cache_fini(cache);
4656 }
4657 
4658 static void
pci_set_power_child(device_t dev,device_t child,int state)4659 pci_set_power_child(device_t dev, device_t child, int state)
4660 {
4661 	device_t pcib;
4662 	int dstate;
4663 
4664 	/*
4665 	 * Set the device to the given state.  If the firmware suggests
4666 	 * a different power state, use it instead.  If power management
4667 	 * is not present, the firmware is responsible for managing
4668 	 * device power.  Skip children who aren't attached since they
4669 	 * are handled separately.
4670 	 */
4671 	pcib = device_get_parent(dev);
4672 	dstate = state;
4673 	if (device_is_attached(child) &&
4674 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4675 		pci_set_powerstate(child, dstate);
4676 }
4677 
4678 int
pci_suspend_child(device_t dev,device_t child)4679 pci_suspend_child(device_t dev, device_t child)
4680 {
4681 	struct pci_devinfo *dinfo;
4682 	struct resource_list_entry *rle;
4683 	int error;
4684 
4685 	dinfo = device_get_ivars(child);
4686 
4687 	/*
4688 	 * Save the PCI configuration space for the child and set the
4689 	 * device in the appropriate power state for this sleep state.
4690 	 */
4691 	pci_cfg_save(child, dinfo, 0);
4692 
4693 	/* Suspend devices before potentially powering them down. */
4694 	error = bus_generic_suspend_child(dev, child);
4695 
4696 	if (error)
4697 		return (error);
4698 
4699 	if (pci_do_power_suspend) {
4700 		/*
4701 		 * Make sure this device's interrupt handler is not invoked
4702 		 * in the case the device uses a shared interrupt that can
4703 		 * be raised by some other device.
4704 		 * This is applicable only to regular (legacy) PCI interrupts
4705 		 * as MSI/MSI-X interrupts are never shared.
4706 		 */
4707 		rle = resource_list_find(&dinfo->resources,
4708 		    SYS_RES_IRQ, 0);
4709 		if (rle != NULL && rle->res != NULL)
4710 			(void)bus_suspend_intr(child, rle->res);
4711 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4712 	}
4713 
4714 	return (0);
4715 }
4716 
4717 int
pci_resume_child(device_t dev,device_t child)4718 pci_resume_child(device_t dev, device_t child)
4719 {
4720 	struct pci_devinfo *dinfo;
4721 	struct resource_list_entry *rle;
4722 
4723 	if (pci_do_power_resume)
4724 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4725 
4726 	dinfo = device_get_ivars(child);
4727 	pci_cfg_restore(child, dinfo);
4728 	pci_clear_pme(child);
4729 	if (!device_is_attached(child))
4730 		pci_cfg_save(child, dinfo, 1);
4731 
4732 	bus_generic_resume_child(dev, child);
4733 
4734 	/*
4735 	 * Allow interrupts only after fully resuming the driver and hardware.
4736 	 */
4737 	if (pci_do_power_suspend) {
4738 		/* See pci_suspend_child for details. */
4739 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4740 		if (rle != NULL && rle->res != NULL)
4741 			(void)bus_resume_intr(child, rle->res);
4742 	}
4743 
4744 	return (0);
4745 }
4746 
4747 int
pci_resume(device_t dev)4748 pci_resume(device_t dev)
4749 {
4750 	device_t child, *devlist;
4751 	int error, i, numdevs;
4752 
4753 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4754 		return (error);
4755 
4756 	/*
4757 	 * Resume critical devices first, then everything else later.
4758 	 */
4759 	for (i = 0; i < numdevs; i++) {
4760 		child = devlist[i];
4761 		switch (pci_get_class(child)) {
4762 		case PCIC_DISPLAY:
4763 		case PCIC_MEMORY:
4764 		case PCIC_BRIDGE:
4765 		case PCIC_BASEPERIPH:
4766 			BUS_RESUME_CHILD(dev, child);
4767 			break;
4768 		}
4769 	}
4770 	for (i = 0; i < numdevs; i++) {
4771 		child = devlist[i];
4772 		switch (pci_get_class(child)) {
4773 		case PCIC_DISPLAY:
4774 		case PCIC_MEMORY:
4775 		case PCIC_BRIDGE:
4776 		case PCIC_BASEPERIPH:
4777 			break;
4778 		default:
4779 			BUS_RESUME_CHILD(dev, child);
4780 		}
4781 	}
4782 	free(devlist, M_TEMP);
4783 	return (0);
4784 }
4785 
4786 static void
pci_load_vendor_data(void)4787 pci_load_vendor_data(void)
4788 {
4789 	caddr_t data;
4790 	void *ptr;
4791 	size_t sz;
4792 
4793 	data = preload_search_by_type("pci_vendor_data");
4794 	if (data != NULL) {
4795 		ptr = preload_fetch_addr(data);
4796 		sz = preload_fetch_size(data);
4797 		if (ptr != NULL && sz != 0) {
4798 			pci_vendordata = ptr;
4799 			pci_vendordata_size = sz;
4800 			/* terminate the database */
4801 			pci_vendordata[pci_vendordata_size] = '\n';
4802 		}
4803 	}
4804 }
4805 
4806 void
pci_driver_added(device_t dev,driver_t * driver)4807 pci_driver_added(device_t dev, driver_t *driver)
4808 {
4809 	int numdevs;
4810 	device_t *devlist;
4811 	device_t child;
4812 	struct pci_devinfo *dinfo;
4813 	int i;
4814 
4815 	if (bootverbose)
4816 		device_printf(dev, "driver added\n");
4817 	DEVICE_IDENTIFY(driver, dev);
4818 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4819 		return;
4820 	for (i = 0; i < numdevs; i++) {
4821 		child = devlist[i];
4822 		if (device_get_state(child) != DS_NOTPRESENT)
4823 			continue;
4824 		dinfo = device_get_ivars(child);
4825 		pci_print_verbose(dinfo);
4826 		if (bootverbose)
4827 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4828 		pci_cfg_restore(child, dinfo);
4829 		if (device_probe_and_attach(child) != 0)
4830 			pci_child_detached(dev, child);
4831 	}
4832 	free(devlist, M_TEMP);
4833 }
4834 
4835 int
pci_setup_intr(device_t dev,device_t child,struct resource * irq,int flags,driver_filter_t * filter,driver_intr_t * intr,void * arg,void ** cookiep)4836 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4837     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4838 {
4839 	struct pci_devinfo *dinfo;
4840 	struct msix_table_entry *mte;
4841 	struct msix_vector *mv;
4842 	uint64_t addr;
4843 	uint32_t data;
4844 	void *cookie;
4845 	int error, rid;
4846 
4847 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4848 	    arg, &cookie);
4849 	if (error)
4850 		return (error);
4851 
4852 	/* If this is not a direct child, just bail out. */
4853 	if (device_get_parent(child) != dev) {
4854 		*cookiep = cookie;
4855 		return(0);
4856 	}
4857 
4858 	rid = rman_get_rid(irq);
4859 	if (rid == 0) {
4860 		/* Make sure that INTx is enabled */
4861 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4862 	} else {
4863 		/*
4864 		 * Check to see if the interrupt is MSI or MSI-X.
4865 		 * Ask our parent to map the MSI and give
4866 		 * us the address and data register values.
4867 		 * If we fail for some reason, teardown the
4868 		 * interrupt handler.
4869 		 */
4870 		dinfo = device_get_ivars(child);
4871 		if (dinfo->cfg.msi.msi_alloc > 0) {
4872 			if (dinfo->cfg.msi.msi_addr == 0) {
4873 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4874 			    ("MSI has handlers, but vectors not mapped"));
4875 				error = PCIB_MAP_MSI(device_get_parent(dev),
4876 				    child, rman_get_start(irq), &addr, &data);
4877 				if (error)
4878 					goto bad;
4879 				dinfo->cfg.msi.msi_addr = addr;
4880 				dinfo->cfg.msi.msi_data = data;
4881 			}
4882 			if (dinfo->cfg.msi.msi_handlers == 0)
4883 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4884 				    dinfo->cfg.msi.msi_data);
4885 			dinfo->cfg.msi.msi_handlers++;
4886 		} else {
4887 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4888 			    ("No MSI or MSI-X interrupts allocated"));
4889 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4890 			    ("MSI-X index too high"));
4891 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4892 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4893 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4894 			KASSERT(mv->mv_irq == rman_get_start(irq),
4895 			    ("IRQ mismatch"));
4896 			if (mv->mv_address == 0) {
4897 				KASSERT(mte->mte_handlers == 0,
4898 		    ("MSI-X table entry has handlers, but vector not mapped"));
4899 				error = PCIB_MAP_MSI(device_get_parent(dev),
4900 				    child, rman_get_start(irq), &addr, &data);
4901 				if (error)
4902 					goto bad;
4903 				mv->mv_address = addr;
4904 				mv->mv_data = data;
4905 			}
4906 
4907 			/*
4908 			 * The MSIX table entry must be made valid by
4909 			 * incrementing the mte_handlers before
4910 			 * calling pci_enable_msix() and
4911 			 * pci_resume_msix(). Else the MSIX rewrite
4912 			 * table quirk will not work as expected.
4913 			 */
4914 			mte->mte_handlers++;
4915 			if (mte->mte_handlers == 1) {
4916 				pci_enable_msix(child, rid - 1, mv->mv_address,
4917 				    mv->mv_data);
4918 				pci_unmask_msix(child, rid - 1);
4919 			}
4920 		}
4921 
4922 		/*
4923 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4924 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4925 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4926 		 */
4927 		if (!pci_has_quirk(pci_get_devid(child),
4928 		    PCI_QUIRK_MSI_INTX_BUG))
4929 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4930 		else
4931 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4932 	bad:
4933 		if (error) {
4934 			(void)bus_generic_teardown_intr(dev, child, irq,
4935 			    cookie);
4936 			return (error);
4937 		}
4938 	}
4939 	*cookiep = cookie;
4940 	return (0);
4941 }
4942 
4943 int
pci_teardown_intr(device_t dev,device_t child,struct resource * irq,void * cookie)4944 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4945     void *cookie)
4946 {
4947 	struct msix_table_entry *mte;
4948 	struct resource_list_entry *rle;
4949 	struct pci_devinfo *dinfo;
4950 	int error, rid;
4951 
4952 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4953 		return (EINVAL);
4954 
4955 	/* If this isn't a direct child, just bail out */
4956 	if (device_get_parent(child) != dev)
4957 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4958 
4959 	rid = rman_get_rid(irq);
4960 	if (rid == 0) {
4961 		/* Mask INTx */
4962 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4963 	} else {
4964 		/*
4965 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4966 		 * decrement the appropriate handlers count and mask the
4967 		 * MSI-X message, or disable MSI messages if the count
4968 		 * drops to 0.
4969 		 */
4970 		dinfo = device_get_ivars(child);
4971 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4972 		if (rle->res != irq)
4973 			return (EINVAL);
4974 		if (dinfo->cfg.msi.msi_alloc > 0) {
4975 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4976 			    ("MSI-X index too high"));
4977 			if (dinfo->cfg.msi.msi_handlers == 0)
4978 				return (EINVAL);
4979 			dinfo->cfg.msi.msi_handlers--;
4980 			if (dinfo->cfg.msi.msi_handlers == 0)
4981 				pci_disable_msi(child);
4982 		} else {
4983 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4984 			    ("No MSI or MSI-X interrupts allocated"));
4985 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4986 			    ("MSI-X index too high"));
4987 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4988 			if (mte->mte_handlers == 0)
4989 				return (EINVAL);
4990 			mte->mte_handlers--;
4991 			if (mte->mte_handlers == 0)
4992 				pci_mask_msix(child, rid - 1);
4993 		}
4994 	}
4995 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4996 	if (rid > 0)
4997 		KASSERT(error == 0,
4998 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4999 	return (error);
5000 }
5001 
5002 int
pci_print_child(device_t dev,device_t child)5003 pci_print_child(device_t dev, device_t child)
5004 {
5005 	struct pci_devinfo *dinfo;
5006 	struct resource_list *rl;
5007 	int retval = 0;
5008 
5009 	dinfo = device_get_ivars(child);
5010 	rl = &dinfo->resources;
5011 
5012 	retval += bus_print_child_header(dev, child);
5013 
5014 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
5015 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
5016 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
5017 	if (device_get_flags(dev))
5018 		retval += printf(" flags %#x", device_get_flags(dev));
5019 
5020 	retval += printf(" at device %d.%d", pci_get_slot(child),
5021 	    pci_get_function(child));
5022 
5023 	retval += bus_print_child_domain(dev, child);
5024 	retval += bus_print_child_footer(dev, child);
5025 
5026 	return (retval);
5027 }
5028 
5029 static const struct
5030 {
5031 	int		class;
5032 	int		subclass;
5033 	int		report; /* 0 = bootverbose, 1 = always */
5034 	const char	*desc;
5035 } pci_nomatch_tab[] = {
5036 	{PCIC_OLD,		-1,			1, "old"},
5037 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
5038 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
5039 	{PCIC_STORAGE,		-1,			1, "mass storage"},
5040 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
5041 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
5042 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
5043 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
5044 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
5045 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
5046 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
5047 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
5048 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
5049 	{PCIC_NETWORK,		-1,			1, "network"},
5050 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
5051 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
5052 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
5053 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
5054 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
5055 	{PCIC_DISPLAY,		-1,			1, "display"},
5056 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
5057 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
5058 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
5059 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
5060 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
5061 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
5062 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
5063 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
5064 	{PCIC_MEMORY,		-1,			1, "memory"},
5065 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
5066 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
5067 	{PCIC_BRIDGE,		-1,			1, "bridge"},
5068 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
5069 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
5070 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
5071 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
5072 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
5073 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
5074 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
5075 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
5076 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
5077 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
5078 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
5079 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
5080 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
5081 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
5082 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
5083 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
5084 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
5085 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
5086 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
5087 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
5088 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
5089 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
5090 	{PCIC_INPUTDEV,		-1,			1, "input device"},
5091 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
5092 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
5093 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
5094 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
5095 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5096 	{PCIC_DOCKING,		-1,			1, "docking station"},
5097 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5098 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5099 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5100 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5101 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5102 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5103 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5104 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5105 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5106 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5107 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5108 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5109 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5110 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5111 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5112 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5113 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5114 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5115 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5116 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5117 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5118 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5119 	{PCIC_DASP,		-1,			0, "dasp"},
5120 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5121 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5122 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5123 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5124 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5125 	{0, 0, 0,		NULL}
5126 };
5127 
5128 void
pci_probe_nomatch(device_t dev,device_t child)5129 pci_probe_nomatch(device_t dev, device_t child)
5130 {
5131 	int i, report;
5132 	const char *cp, *scp;
5133 	char *device;
5134 
5135 	/*
5136 	 * Look for a listing for this device in a loaded device database.
5137 	 */
5138 	report = 1;
5139 	if ((device = pci_describe_device(child)) != NULL) {
5140 		device_printf(dev, "<%s>", device);
5141 		free(device, M_DEVBUF);
5142 	} else {
5143 		/*
5144 		 * Scan the class/subclass descriptions for a general
5145 		 * description.
5146 		 */
5147 		cp = "unknown";
5148 		scp = NULL;
5149 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5150 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5151 				if (pci_nomatch_tab[i].subclass == -1) {
5152 					cp = pci_nomatch_tab[i].desc;
5153 					report = pci_nomatch_tab[i].report;
5154 				} else if (pci_nomatch_tab[i].subclass ==
5155 				    pci_get_subclass(child)) {
5156 					scp = pci_nomatch_tab[i].desc;
5157 					report = pci_nomatch_tab[i].report;
5158 				}
5159 			}
5160 		}
5161 		if (report || bootverbose) {
5162 			device_printf(dev, "<%s%s%s>",
5163 			    cp ? cp : "",
5164 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5165 			    scp ? scp : "");
5166 		}
5167 	}
5168 	if (report || bootverbose) {
5169 		printf(" at device %d.%d (no driver attached)\n",
5170 		    pci_get_slot(child), pci_get_function(child));
5171 	}
5172 	pci_cfg_save(child, device_get_ivars(child), 1);
5173 }
5174 
5175 void
pci_child_detached(device_t dev,device_t child)5176 pci_child_detached(device_t dev, device_t child)
5177 {
5178 	struct pci_devinfo *dinfo;
5179 	struct resource_list *rl;
5180 
5181 	dinfo = device_get_ivars(child);
5182 	rl = &dinfo->resources;
5183 
5184 	/*
5185 	 * Have to deallocate IRQs before releasing any MSI messages and
5186 	 * have to release MSI messages before deallocating any memory
5187 	 * BARs.
5188 	 */
5189 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5190 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5191 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5192 		if (dinfo->cfg.msi.msi_alloc != 0)
5193 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5194 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5195 		else
5196 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5197 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5198 		(void)pci_release_msi(child);
5199 	}
5200 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5201 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5202 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5203 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5204 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5205 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5206 
5207 	pci_cfg_save(child, dinfo, 1);
5208 }
5209 
5210 /*
5211  * Parse the PCI device database, if loaded, and return a pointer to a
5212  * description of the device.
5213  *
5214  * The database is flat text formatted as follows:
5215  *
5216  * Any line not in a valid format is ignored.
5217  * Lines are terminated with newline '\n' characters.
5218  *
5219  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5220  * the vendor name.
5221  *
5222  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5223  * - devices cannot be listed without a corresponding VENDOR line.
5224  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5225  * another TAB, then the device name.
5226  */
5227 
5228 /*
5229  * Assuming (ptr) points to the beginning of a line in the database,
5230  * return the vendor or device and description of the next entry.
5231  * The value of (vendor) or (device) inappropriate for the entry type
5232  * is set to -1.  Returns nonzero at the end of the database.
5233  *
5234  * Note that this is slightly unrobust in the face of corrupt data;
5235  * we attempt to safeguard against this by spamming the end of the
5236  * database with a newline when we initialise.
5237  */
5238 static int
pci_describe_parse_line(char ** ptr,int * vendor,int * device,char ** desc)5239 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5240 {
5241 	char	*cp = *ptr;
5242 	int	left;
5243 
5244 	*device = -1;
5245 	*vendor = -1;
5246 	**desc = '\0';
5247 	for (;;) {
5248 		left = pci_vendordata_size - (cp - pci_vendordata);
5249 		if (left <= 0) {
5250 			*ptr = cp;
5251 			return(1);
5252 		}
5253 
5254 		/* vendor entry? */
5255 		if (*cp != '\t' &&
5256 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5257 			break;
5258 		/* device entry? */
5259 		if (*cp == '\t' &&
5260 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5261 			break;
5262 
5263 		/* skip to next line */
5264 		while (*cp != '\n' && left > 0) {
5265 			cp++;
5266 			left--;
5267 		}
5268 		if (*cp == '\n') {
5269 			cp++;
5270 			left--;
5271 		}
5272 	}
5273 	/* skip to next line */
5274 	while (*cp != '\n' && left > 0) {
5275 		cp++;
5276 		left--;
5277 	}
5278 	if (*cp == '\n' && left > 0)
5279 		cp++;
5280 	*ptr = cp;
5281 	return(0);
5282 }
5283 
5284 static char *
pci_describe_device(device_t dev)5285 pci_describe_device(device_t dev)
5286 {
5287 	int	vendor, device;
5288 	char	*desc, *vp, *dp, *line;
5289 
5290 	desc = vp = dp = NULL;
5291 
5292 	/*
5293 	 * If we have no vendor data, we can't do anything.
5294 	 */
5295 	if (pci_vendordata == NULL)
5296 		goto out;
5297 
5298 	/*
5299 	 * Scan the vendor data looking for this device
5300 	 */
5301 	line = pci_vendordata;
5302 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5303 		goto out;
5304 	for (;;) {
5305 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5306 			goto out;
5307 		if (vendor == pci_get_vendor(dev))
5308 			break;
5309 	}
5310 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5311 		goto out;
5312 	for (;;) {
5313 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5314 			*dp = 0;
5315 			break;
5316 		}
5317 		if (vendor != -1) {
5318 			*dp = 0;
5319 			break;
5320 		}
5321 		if (device == pci_get_device(dev))
5322 			break;
5323 	}
5324 	if (dp[0] == '\0')
5325 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5326 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5327 	    NULL)
5328 		sprintf(desc, "%s, %s", vp, dp);
5329 out:
5330 	if (vp != NULL)
5331 		free(vp, M_DEVBUF);
5332 	if (dp != NULL)
5333 		free(dp, M_DEVBUF);
5334 	return(desc);
5335 }
5336 
5337 int
pci_read_ivar(device_t dev,device_t child,int which,uintptr_t * result)5338 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5339 {
5340 	struct pci_devinfo *dinfo;
5341 	pcicfgregs *cfg;
5342 
5343 	dinfo = device_get_ivars(child);
5344 	cfg = &dinfo->cfg;
5345 
5346 	switch (which) {
5347 	case PCI_IVAR_ETHADDR:
5348 		/*
5349 		 * The generic accessor doesn't deal with failure, so
5350 		 * we set the return value, then return an error.
5351 		 */
5352 		*((uint8_t **) result) = NULL;
5353 		return (EINVAL);
5354 	case PCI_IVAR_SUBVENDOR:
5355 		*result = cfg->subvendor;
5356 		break;
5357 	case PCI_IVAR_SUBDEVICE:
5358 		*result = cfg->subdevice;
5359 		break;
5360 	case PCI_IVAR_VENDOR:
5361 		*result = cfg->vendor;
5362 		break;
5363 	case PCI_IVAR_DEVICE:
5364 		*result = cfg->device;
5365 		break;
5366 	case PCI_IVAR_DEVID:
5367 		*result = (cfg->device << 16) | cfg->vendor;
5368 		break;
5369 	case PCI_IVAR_CLASS:
5370 		*result = cfg->baseclass;
5371 		break;
5372 	case PCI_IVAR_SUBCLASS:
5373 		*result = cfg->subclass;
5374 		break;
5375 	case PCI_IVAR_PROGIF:
5376 		*result = cfg->progif;
5377 		break;
5378 	case PCI_IVAR_REVID:
5379 		*result = cfg->revid;
5380 		break;
5381 	case PCI_IVAR_INTPIN:
5382 		*result = cfg->intpin;
5383 		break;
5384 	case PCI_IVAR_IRQ:
5385 		*result = cfg->intline;
5386 		break;
5387 	case PCI_IVAR_DOMAIN:
5388 		*result = cfg->domain;
5389 		break;
5390 	case PCI_IVAR_BUS:
5391 		*result = cfg->bus;
5392 		break;
5393 	case PCI_IVAR_SLOT:
5394 		*result = cfg->slot;
5395 		break;
5396 	case PCI_IVAR_FUNCTION:
5397 		*result = cfg->func;
5398 		break;
5399 	case PCI_IVAR_CMDREG:
5400 		*result = cfg->cmdreg;
5401 		break;
5402 	case PCI_IVAR_CACHELNSZ:
5403 		*result = cfg->cachelnsz;
5404 		break;
5405 	case PCI_IVAR_MINGNT:
5406 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5407 			*result = -1;
5408 			return (EINVAL);
5409 		}
5410 		*result = cfg->mingnt;
5411 		break;
5412 	case PCI_IVAR_MAXLAT:
5413 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5414 			*result = -1;
5415 			return (EINVAL);
5416 		}
5417 		*result = cfg->maxlat;
5418 		break;
5419 	case PCI_IVAR_LATTIMER:
5420 		*result = cfg->lattimer;
5421 		break;
5422 	default:
5423 		return (ENOENT);
5424 	}
5425 	return (0);
5426 }
5427 
5428 int
pci_write_ivar(device_t dev,device_t child,int which,uintptr_t value)5429 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5430 {
5431 	struct pci_devinfo *dinfo;
5432 
5433 	dinfo = device_get_ivars(child);
5434 
5435 	switch (which) {
5436 	case PCI_IVAR_INTPIN:
5437 		dinfo->cfg.intpin = value;
5438 		return (0);
5439 	case PCI_IVAR_ETHADDR:
5440 	case PCI_IVAR_SUBVENDOR:
5441 	case PCI_IVAR_SUBDEVICE:
5442 	case PCI_IVAR_VENDOR:
5443 	case PCI_IVAR_DEVICE:
5444 	case PCI_IVAR_DEVID:
5445 	case PCI_IVAR_CLASS:
5446 	case PCI_IVAR_SUBCLASS:
5447 	case PCI_IVAR_PROGIF:
5448 	case PCI_IVAR_REVID:
5449 	case PCI_IVAR_IRQ:
5450 	case PCI_IVAR_DOMAIN:
5451 	case PCI_IVAR_BUS:
5452 	case PCI_IVAR_SLOT:
5453 	case PCI_IVAR_FUNCTION:
5454 		return (EINVAL);	/* disallow for now */
5455 
5456 	default:
5457 		return (ENOENT);
5458 	}
5459 }
5460 
5461 #include "opt_ddb.h"
5462 #ifdef DDB
5463 #include <ddb/ddb.h>
5464 #include <sys/cons.h>
5465 
5466 /*
5467  * List resources based on pci map registers, used for within ddb
5468  */
5469 
DB_SHOW_COMMAND_FLAGS(pciregs,db_pci_dump,DB_CMD_MEMSAFE)5470 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5471 {
5472 	struct pci_devinfo *dinfo;
5473 	struct devlist *devlist_head;
5474 	struct pci_conf *p;
5475 	const char *name;
5476 	int i, error, none_count;
5477 
5478 	none_count = 0;
5479 	/* get the head of the device queue */
5480 	devlist_head = &pci_devq;
5481 
5482 	/*
5483 	 * Go through the list of devices and print out devices
5484 	 */
5485 	for (error = 0, i = 0,
5486 	     dinfo = STAILQ_FIRST(devlist_head);
5487 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5488 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5489 		/* Populate pd_name and pd_unit */
5490 		name = NULL;
5491 		if (dinfo->cfg.dev)
5492 			name = device_get_name(dinfo->cfg.dev);
5493 
5494 		p = &dinfo->conf;
5495 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5496 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5497 			(name && *name) ? name : "none",
5498 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5499 			none_count++,
5500 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5501 			p->pc_sel.pc_func, (p->pc_class << 16) |
5502 			(p->pc_subclass << 8) | p->pc_progif,
5503 			(p->pc_subdevice << 16) | p->pc_subvendor,
5504 			(p->pc_device << 16) | p->pc_vendor,
5505 			p->pc_revid, p->pc_hdr);
5506 	}
5507 }
5508 #endif /* DDB */
5509 
5510 struct resource *
pci_reserve_map(device_t dev,device_t child,int type,int rid,rman_res_t start,rman_res_t end,rman_res_t count,u_int num,u_int flags)5511 pci_reserve_map(device_t dev, device_t child, int type, int rid,
5512     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5513     u_int flags)
5514 {
5515 	struct pci_devinfo *dinfo = device_get_ivars(child);
5516 	struct resource_list *rl = &dinfo->resources;
5517 	struct resource *res;
5518 	struct pci_map *pm;
5519 	uint16_t cmd;
5520 	pci_addr_t map, testval;
5521 	int mapsize;
5522 
5523 	res = NULL;
5524 
5525 	/* If rid is managed by EA, ignore it */
5526 	if (pci_ea_is_enabled(child, rid))
5527 		goto out;
5528 
5529 	pm = pci_find_bar(child, rid);
5530 	if (pm != NULL) {
5531 		/* This is a BAR that we failed to allocate earlier. */
5532 		mapsize = pm->pm_size;
5533 		map = pm->pm_value;
5534 	} else {
5535 		/*
5536 		 * Weed out the bogons, and figure out how large the
5537 		 * BAR/map is.  BARs that read back 0 here are bogus
5538 		 * and unimplemented.  Note: atapci in legacy mode are
5539 		 * special and handled elsewhere in the code.  If you
5540 		 * have a atapci device in legacy mode and it fails
5541 		 * here, that other code is broken.
5542 		 */
5543 		pci_read_bar(child, rid, &map, &testval, NULL);
5544 
5545 		/*
5546 		 * Determine the size of the BAR and ignore BARs with a size
5547 		 * of 0.  Device ROM BARs use a different mask value.
5548 		 */
5549 		if (PCIR_IS_BIOS(&dinfo->cfg, rid))
5550 			mapsize = pci_romsize(testval);
5551 		else
5552 			mapsize = pci_mapsize(testval);
5553 		if (mapsize == 0)
5554 			goto out;
5555 		pm = pci_add_bar(child, rid, map, mapsize);
5556 	}
5557 
5558 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, rid)) {
5559 		if (type != SYS_RES_MEMORY) {
5560 			if (bootverbose)
5561 				device_printf(dev,
5562 				    "child %s requested type %d for rid %#x,"
5563 				    " but the BAR says it is an memio\n",
5564 				    device_get_nameunit(child), type, rid);
5565 			goto out;
5566 		}
5567 	} else {
5568 		if (type != SYS_RES_IOPORT) {
5569 			if (bootverbose)
5570 				device_printf(dev,
5571 				    "child %s requested type %d for rid %#x,"
5572 				    " but the BAR says it is an ioport\n",
5573 				    device_get_nameunit(child), type, rid);
5574 			goto out;
5575 		}
5576 	}
5577 
5578 	/*
5579 	 * For real BARs, we need to override the size that
5580 	 * the driver requests, because that's what the BAR
5581 	 * actually uses and we would otherwise have a
5582 	 * situation where we might allocate the excess to
5583 	 * another driver, which won't work.
5584 	 */
5585 	count = ((pci_addr_t)1 << mapsize) * num;
5586 	if (RF_ALIGNMENT(flags) < mapsize)
5587 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5588 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5589 		flags |= RF_PREFETCHABLE;
5590 
5591 	/*
5592 	 * Allocate enough resource, and then write back the
5593 	 * appropriate BAR for that resource.
5594 	 */
5595 	resource_list_add(rl, type, rid, start, end, count);
5596 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5597 	    count, flags & ~RF_ACTIVE);
5598 	if (res == NULL) {
5599 		resource_list_delete(rl, type, rid);
5600 		device_printf(child,
5601 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5602 		    count, rid, type, start, end);
5603 		goto out;
5604 	}
5605 	if (bootverbose)
5606 		device_printf(child,
5607 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5608 		    count, rid, type, rman_get_start(res));
5609 
5610 	/* Disable decoding via the CMD register before updating the BAR */
5611 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5612 	pci_write_config(child, PCIR_COMMAND,
5613 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5614 
5615 	map = rman_get_start(res);
5616 	pci_write_bar(child, pm, map);
5617 
5618 	/* Restore the original value of the CMD register */
5619 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5620 out:
5621 	return (res);
5622 }
5623 
5624 struct resource *
pci_alloc_multi_resource(device_t dev,device_t child,int type,int rid,rman_res_t start,rman_res_t end,rman_res_t count,u_long num,u_int flags)5625 pci_alloc_multi_resource(device_t dev, device_t child, int type, int rid,
5626     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5627     u_int flags)
5628 {
5629 	struct pci_devinfo *dinfo;
5630 	struct resource_list *rl;
5631 	struct resource_list_entry *rle;
5632 	struct resource *res;
5633 	pcicfgregs *cfg;
5634 
5635 	/*
5636 	 * Perform lazy resource allocation
5637 	 */
5638 	dinfo = device_get_ivars(child);
5639 	rl = &dinfo->resources;
5640 	cfg = &dinfo->cfg;
5641 	switch (type) {
5642 	case PCI_RES_BUS:
5643 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5644 		    flags));
5645 	case SYS_RES_IRQ:
5646 		/*
5647 		 * Can't alloc legacy interrupt once MSI messages have
5648 		 * been allocated.
5649 		 */
5650 		if (rid == 0 && (cfg->msi.msi_alloc > 0 ||
5651 		    cfg->msix.msix_alloc > 0))
5652 			return (NULL);
5653 
5654 		/*
5655 		 * If the child device doesn't have an interrupt
5656 		 * routed and is deserving of an interrupt, try to
5657 		 * assign it one.
5658 		 */
5659 		if (rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5660 		    (cfg->intpin != 0))
5661 			pci_assign_interrupt(dev, child, 0);
5662 		break;
5663 	case SYS_RES_IOPORT:
5664 	case SYS_RES_MEMORY:
5665 		/*
5666 		 * PCI-PCI bridge I/O window resources are not BARs.
5667 		 * For those allocations just pass the request up the
5668 		 * tree.
5669 		 */
5670 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5671 			switch (rid) {
5672 			case PCIR_IOBASEL_1:
5673 			case PCIR_MEMBASE_1:
5674 			case PCIR_PMBASEL_1:
5675 				/*
5676 				 * XXX: Should we bother creating a resource
5677 				 * list entry?
5678 				 */
5679 				return (bus_generic_alloc_resource(dev, child,
5680 				    type, rid, start, end, count, flags));
5681 			}
5682 		}
5683 		/* Reserve resources for this BAR if needed. */
5684 		rle = resource_list_find(rl, type, rid);
5685 		if (rle == NULL) {
5686 			res = pci_reserve_map(dev, child, type, rid, start, end,
5687 			    count, num, flags);
5688 			if (res == NULL)
5689 				return (NULL);
5690 		}
5691 	}
5692 	return (resource_list_alloc(rl, dev, child, type, rid,
5693 	    start, end, count, flags));
5694 }
5695 
5696 struct resource *
pci_alloc_resource(device_t dev,device_t child,int type,int rid,rman_res_t start,rman_res_t end,rman_res_t count,u_int flags)5697 pci_alloc_resource(device_t dev, device_t child, int type, int rid,
5698     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5699 {
5700 #ifdef PCI_IOV
5701 	struct pci_devinfo *dinfo;
5702 #endif
5703 
5704 	if (device_get_parent(child) != dev)
5705 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5706 		    type, rid, start, end, count, flags));
5707 
5708 #ifdef PCI_IOV
5709 	dinfo = device_get_ivars(child);
5710 	if (dinfo->cfg.flags & PCICFG_VF) {
5711 		switch (type) {
5712 		/* VFs can't have I/O BARs. */
5713 		case SYS_RES_IOPORT:
5714 			return (NULL);
5715 		case SYS_RES_MEMORY:
5716 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5717 			    start, end, count, flags));
5718 		}
5719 
5720 		/* Fall through for other types of resource allocations. */
5721 	}
5722 #endif
5723 
5724 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5725 	    count, 1, flags));
5726 }
5727 
5728 int
pci_release_resource(device_t dev,device_t child,struct resource * r)5729 pci_release_resource(device_t dev, device_t child, struct resource *r)
5730 {
5731 	struct pci_devinfo *dinfo;
5732 	struct resource_list *rl;
5733 	pcicfgregs *cfg __unused;
5734 
5735 	if (device_get_parent(child) != dev)
5736 		return (bus_generic_release_resource(dev, child, r));
5737 
5738 	dinfo = device_get_ivars(child);
5739 	cfg = &dinfo->cfg;
5740 
5741 #ifdef PCI_IOV
5742 	if (cfg->flags & PCICFG_VF) {
5743 		switch (rman_get_type(r)) {
5744 		/* VFs can't have I/O BARs. */
5745 		case SYS_RES_IOPORT:
5746 			return (EDOOFUS);
5747 		case SYS_RES_MEMORY:
5748 			return (pci_vf_release_mem_resource(dev, child, r));
5749 		}
5750 
5751 		/* Fall through for other types of resource allocations. */
5752 	}
5753 #endif
5754 
5755 	/*
5756 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5757 	 * those allocations just pass the request up the tree.
5758 	 */
5759 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5760 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5761 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5762 		switch (rman_get_rid(r)) {
5763 		case PCIR_IOBASEL_1:
5764 		case PCIR_MEMBASE_1:
5765 		case PCIR_PMBASEL_1:
5766 			return (bus_generic_release_resource(dev, child, r));
5767 		}
5768 	}
5769 
5770 	rl = &dinfo->resources;
5771 	return (resource_list_release(rl, dev, child, r));
5772 }
5773 
5774 int
pci_activate_resource(device_t dev,device_t child,struct resource * r)5775 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5776 {
5777 	struct pci_devinfo *dinfo;
5778 	int error, rid, type;
5779 
5780 	if (device_get_parent(child) != dev)
5781 		return (bus_generic_activate_resource(dev, child, r));
5782 
5783 	dinfo = device_get_ivars(child);
5784 #ifdef PCI_IOV
5785 	if (dinfo->cfg.flags & PCICFG_VF) {
5786 		switch (rman_get_type(r)) {
5787 		/* VFs can't have I/O BARs. */
5788 		case SYS_RES_IOPORT:
5789 			error = EINVAL;
5790 			break;
5791 		case SYS_RES_MEMORY:
5792 			error = pci_vf_activate_mem_resource(dev, child, r);
5793 			break;
5794 		default:
5795 			error = bus_generic_activate_resource(dev, child, r);
5796 			break;
5797 		}
5798 	} else
5799 #endif
5800 		error = bus_generic_activate_resource(dev, child, r);
5801 	if (error)
5802 		return (error);
5803 
5804 	rid = rman_get_rid(r);
5805 	type = rman_get_type(r);
5806 
5807 	/* Device ROMs need their decoding explicitly enabled. */
5808 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5809 		pci_write_bar(child, pci_find_bar(child, rid),
5810 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5811 
5812 	/* Enable decoding in the command register when activating BARs. */
5813 	switch (type) {
5814 	case SYS_RES_IOPORT:
5815 	case SYS_RES_MEMORY:
5816 		error = PCI_ENABLE_IO(dev, child, type);
5817 		break;
5818 	}
5819 	return (error);
5820 }
5821 
5822 int
pci_deactivate_resource(device_t dev,device_t child,struct resource * r)5823 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5824 {
5825 	struct pci_devinfo *dinfo;
5826 	int error, rid, type;
5827 
5828 	if (device_get_parent(child) != dev)
5829 		return (bus_generic_deactivate_resource(dev, child, r));
5830 
5831 	dinfo = device_get_ivars(child);
5832 #ifdef PCI_IOV
5833 	if (dinfo->cfg.flags & PCICFG_VF) {
5834 		switch (rman_get_type(r)) {
5835 		/* VFs can't have I/O BARs. */
5836 		case SYS_RES_IOPORT:
5837 			error = EINVAL;
5838 			break;
5839 		case SYS_RES_MEMORY:
5840 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5841 			break;
5842 		default:
5843 			error = bus_generic_deactivate_resource(dev, child, r);
5844 			break;
5845 		}
5846 	} else
5847 #endif
5848 		error = bus_generic_deactivate_resource(dev, child, r);
5849 	if (error)
5850 		return (error);
5851 
5852 	/* Disable decoding for device ROMs. */
5853 	rid = rman_get_rid(r);
5854 	type = rman_get_type(r);
5855 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5856 		pci_write_bar(child, pci_find_bar(child, rid),
5857 		    rman_get_start(r));
5858 	return (0);
5859 }
5860 
5861 int
pci_adjust_resource(device_t dev,device_t child,struct resource * r,rman_res_t start,rman_res_t end)5862 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5863     rman_res_t start, rman_res_t end)
5864 {
5865 #ifdef PCI_IOV
5866 	struct pci_devinfo *dinfo;
5867 
5868 	if (device_get_parent(child) != dev)
5869 		return (bus_generic_adjust_resource(dev, child, r, start,
5870 		    end));
5871 
5872 	dinfo = device_get_ivars(child);
5873 	if (dinfo->cfg.flags & PCICFG_VF) {
5874 		switch (rman_get_type(r)) {
5875 		/* VFs can't have I/O BARs. */
5876 		case SYS_RES_IOPORT:
5877 			return (EINVAL);
5878 		case SYS_RES_MEMORY:
5879 			return (pci_vf_adjust_mem_resource(dev, child, r,
5880 			    start, end));
5881 		}
5882 
5883 		/* Fall through for other types of resource allocations. */
5884 	}
5885 #endif
5886 
5887 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5888 }
5889 
5890 int
pci_map_resource(device_t dev,device_t child,struct resource * r,struct resource_map_request * argsp,struct resource_map * map)5891 pci_map_resource(device_t dev, device_t child, struct resource *r,
5892     struct resource_map_request *argsp, struct resource_map *map)
5893 {
5894 #ifdef PCI_IOV
5895 	struct pci_devinfo *dinfo;
5896 
5897 	if (device_get_parent(child) != dev)
5898 		return (bus_generic_map_resource(dev, child, r, argsp,
5899 		    map));
5900 
5901 	dinfo = device_get_ivars(child);
5902 	if (dinfo->cfg.flags & PCICFG_VF) {
5903 		switch (rman_get_type(r)) {
5904 		/* VFs can't have I/O BARs. */
5905 		case SYS_RES_IOPORT:
5906 			return (EINVAL);
5907 		case SYS_RES_MEMORY:
5908 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5909 			    map));
5910 		}
5911 
5912 		/* Fall through for other types of resource allocations. */
5913 	}
5914 #endif
5915 
5916 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5917 }
5918 
5919 int
pci_unmap_resource(device_t dev,device_t child,struct resource * r,struct resource_map * map)5920 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5921     struct resource_map *map)
5922 {
5923 #ifdef PCI_IOV
5924 	struct pci_devinfo *dinfo;
5925 
5926 	if (device_get_parent(child) != dev)
5927 		return (bus_generic_unmap_resource(dev, child, r, map));
5928 
5929 	dinfo = device_get_ivars(child);
5930 	if (dinfo->cfg.flags & PCICFG_VF) {
5931 		switch (rman_get_type(r)) {
5932 		/* VFs can't have I/O BARs. */
5933 		case SYS_RES_IOPORT:
5934 			return (EINVAL);
5935 		case SYS_RES_MEMORY:
5936 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5937 		}
5938 
5939 		/* Fall through for other types of resource allocations. */
5940 	}
5941 #endif
5942 
5943 	return (bus_generic_unmap_resource(dev, child, r, map));
5944 }
5945 
5946 void
pci_child_deleted(device_t dev,device_t child)5947 pci_child_deleted(device_t dev, device_t child)
5948 {
5949 	struct resource_list_entry *rle;
5950 	struct resource_list *rl;
5951 	struct pci_devinfo *dinfo;
5952 
5953 	dinfo = device_get_ivars(child);
5954 	rl = &dinfo->resources;
5955 
5956 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5957 
5958 	/* Turn off access to resources we're about to free */
5959 	if (bus_child_present(child) != 0) {
5960 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5961 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5962 
5963 		pci_disable_busmaster(child);
5964 	}
5965 
5966 	/* Free all allocated resources */
5967 	STAILQ_FOREACH(rle, rl, link) {
5968 		if (rle->res) {
5969 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5970 			    resource_list_busy(rl, rle->type, rle->rid)) {
5971 				pci_printf(&dinfo->cfg,
5972 				    "Resource still owned, oops. "
5973 				    "(type=%d, rid=%d, addr=%lx)\n",
5974 				    rle->type, rle->rid,
5975 				    rman_get_start(rle->res));
5976 				bus_release_resource(child, rle->type, rle->rid,
5977 				    rle->res);
5978 			}
5979 			resource_list_unreserve(rl, dev, child, rle->type,
5980 			    rle->rid);
5981 		}
5982 	}
5983 	resource_list_free(rl);
5984 
5985 	pci_freecfg(dinfo);
5986 }
5987 
5988 void
pci_delete_resource(device_t dev,device_t child,int type,int rid)5989 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5990 {
5991 	struct pci_devinfo *dinfo;
5992 	struct resource_list *rl;
5993 	struct resource_list_entry *rle;
5994 
5995 	if (device_get_parent(child) != dev)
5996 		return;
5997 
5998 	dinfo = device_get_ivars(child);
5999 	rl = &dinfo->resources;
6000 	rle = resource_list_find(rl, type, rid);
6001 	if (rle == NULL)
6002 		return;
6003 
6004 	if (rle->res) {
6005 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
6006 		    resource_list_busy(rl, type, rid)) {
6007 			device_printf(dev, "delete_resource: "
6008 			    "Resource still owned by child, oops. "
6009 			    "(type=%d, rid=%d, addr=%jx)\n",
6010 			    type, rid, rman_get_start(rle->res));
6011 			return;
6012 		}
6013 		resource_list_unreserve(rl, dev, child, type, rid);
6014 	}
6015 	resource_list_delete(rl, type, rid);
6016 }
6017 
6018 struct resource_list *
pci_get_resource_list(device_t dev,device_t child)6019 pci_get_resource_list (device_t dev, device_t child)
6020 {
6021 	struct pci_devinfo *dinfo = device_get_ivars(child);
6022 
6023 	return (&dinfo->resources);
6024 }
6025 
6026 #ifdef IOMMU
6027 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)6028 pci_get_dma_tag(device_t bus, device_t dev)
6029 {
6030 	bus_dma_tag_t tag;
6031 	struct pci_softc *sc;
6032 
6033 	if (device_get_parent(dev) == bus) {
6034 		/* try iommu and return if it works */
6035 		tag = iommu_get_dma_tag(bus, dev);
6036 	} else
6037 		tag = NULL;
6038 	if (tag == NULL) {
6039 		sc = device_get_softc(bus);
6040 		tag = sc->sc_dma_tag;
6041 	}
6042 	return (tag);
6043 }
6044 #else
6045 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)6046 pci_get_dma_tag(device_t bus, device_t dev)
6047 {
6048 	struct pci_softc *sc = device_get_softc(bus);
6049 
6050 	return (sc->sc_dma_tag);
6051 }
6052 #endif
6053 
6054 uint32_t
pci_read_config_method(device_t dev,device_t child,int reg,int width)6055 pci_read_config_method(device_t dev, device_t child, int reg, int width)
6056 {
6057 	struct pci_devinfo *dinfo = device_get_ivars(child);
6058 	pcicfgregs *cfg = &dinfo->cfg;
6059 
6060 #ifdef PCI_IOV
6061 	/*
6062 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
6063 	 * emulate them here.
6064 	 */
6065 	if (cfg->flags & PCICFG_VF) {
6066 		if (reg == PCIR_VENDOR) {
6067 			switch (width) {
6068 			case 4:
6069 				return (cfg->device << 16 | cfg->vendor);
6070 			case 2:
6071 				return (cfg->vendor);
6072 			case 1:
6073 				return (cfg->vendor & 0xff);
6074 			default:
6075 				return (0xffffffff);
6076 			}
6077 		} else if (reg == PCIR_DEVICE) {
6078 			switch (width) {
6079 			/* Note that an unaligned 4-byte read is an error. */
6080 			case 2:
6081 				return (cfg->device);
6082 			case 1:
6083 				return (cfg->device & 0xff);
6084 			default:
6085 				return (0xffffffff);
6086 			}
6087 		}
6088 	}
6089 #endif
6090 
6091 	return (PCIB_READ_CONFIG(device_get_parent(dev),
6092 	    cfg->bus, cfg->slot, cfg->func, reg, width));
6093 }
6094 
6095 void
pci_write_config_method(device_t dev,device_t child,int reg,uint32_t val,int width)6096 pci_write_config_method(device_t dev, device_t child, int reg,
6097     uint32_t val, int width)
6098 {
6099 	struct pci_devinfo *dinfo = device_get_ivars(child);
6100 	pcicfgregs *cfg = &dinfo->cfg;
6101 
6102 	PCIB_WRITE_CONFIG(device_get_parent(dev),
6103 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6104 }
6105 
6106 int
pci_child_location_method(device_t dev,device_t child,struct sbuf * sb)6107 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6108 {
6109 
6110 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6111 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6112 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6113 	return (0);
6114 }
6115 
6116 int
pci_child_pnpinfo_method(device_t dev,device_t child,struct sbuf * sb)6117 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6118 {
6119 	struct pci_devinfo *dinfo;
6120 	pcicfgregs *cfg;
6121 
6122 	dinfo = device_get_ivars(child);
6123 	cfg = &dinfo->cfg;
6124 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6125 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6126 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6127 	    cfg->progif);
6128 	return (0);
6129 }
6130 
6131 int
pci_get_device_path_method(device_t bus,device_t child,const char * locator,struct sbuf * sb)6132 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6133     struct sbuf *sb)
6134 {
6135 	device_t parent = device_get_parent(bus);
6136 	int rv;
6137 
6138 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6139 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6140 		if (rv == 0) {
6141 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6142 			    pci_get_function(child));
6143 		}
6144 		return (0);
6145 	}
6146 	return (bus_generic_get_device_path(bus, child, locator, sb));
6147 }
6148 
6149 int
pci_assign_interrupt_method(device_t dev,device_t child)6150 pci_assign_interrupt_method(device_t dev, device_t child)
6151 {
6152 	struct pci_devinfo *dinfo = device_get_ivars(child);
6153 	pcicfgregs *cfg = &dinfo->cfg;
6154 
6155 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6156 	    cfg->intpin));
6157 }
6158 
6159 static void
pci_lookup(void * arg,const char * name,device_t * dev)6160 pci_lookup(void *arg, const char *name, device_t *dev)
6161 {
6162 	long val;
6163 	char *end;
6164 	int domain, bus, slot, func;
6165 
6166 	if (*dev != NULL)
6167 		return;
6168 
6169 	/*
6170 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6171 	 * pciB:S:F.  In the latter case, the domain is assumed to
6172 	 * be zero.
6173 	 */
6174 	if (strncmp(name, "pci", 3) != 0)
6175 		return;
6176 	val = strtol(name + 3, &end, 10);
6177 	if (val < 0 || val > INT_MAX || *end != ':')
6178 		return;
6179 	domain = val;
6180 	val = strtol(end + 1, &end, 10);
6181 	if (val < 0 || val > INT_MAX || *end != ':')
6182 		return;
6183 	bus = val;
6184 	val = strtol(end + 1, &end, 10);
6185 	if (val < 0 || val > INT_MAX)
6186 		return;
6187 	slot = val;
6188 	if (*end == ':') {
6189 		val = strtol(end + 1, &end, 10);
6190 		if (val < 0 || val > INT_MAX || *end != '\0')
6191 			return;
6192 		func = val;
6193 	} else if (*end == '\0') {
6194 		func = slot;
6195 		slot = bus;
6196 		bus = domain;
6197 		domain = 0;
6198 	} else
6199 		return;
6200 
6201 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6202 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6203 		return;
6204 
6205 	*dev = pci_find_dbsf(domain, bus, slot, func);
6206 }
6207 
6208 static int
pci_modevent(module_t mod,int what,void * arg)6209 pci_modevent(module_t mod, int what, void *arg)
6210 {
6211 	static struct cdev *pci_cdev;
6212 	static eventhandler_tag tag;
6213 
6214 	switch (what) {
6215 	case MOD_LOAD:
6216 		STAILQ_INIT(&pci_devq);
6217 		pci_generation = 0;
6218 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6219 		    "pci");
6220 		pci_load_vendor_data();
6221 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6222 		    1000);
6223 		break;
6224 
6225 	case MOD_UNLOAD:
6226 		if (tag != NULL)
6227 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6228 		destroy_dev(pci_cdev);
6229 		break;
6230 	}
6231 
6232 	return (0);
6233 }
6234 
6235 static void
pci_cfg_restore_pcie(device_t dev,struct pci_devinfo * dinfo)6236 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6237 {
6238 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6239 	struct pcicfg_pcie *cfg;
6240 	int version, pos;
6241 
6242 	cfg = &dinfo->cfg.pcie;
6243 	pos = cfg->pcie_location;
6244 
6245 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6246 
6247 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6248 
6249 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6250 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6251 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6252 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6253 
6254 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6255 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6256 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6257 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6258 
6259 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6260 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6261 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6262 
6263 	if (version > 1) {
6264 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6265 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6266 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6267 	}
6268 #undef WREG
6269 }
6270 
6271 static void
pci_cfg_restore_pcix(device_t dev,struct pci_devinfo * dinfo)6272 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6273 {
6274 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6275 	    dinfo->cfg.pcix.pcix_command,  2);
6276 }
6277 
6278 void
pci_cfg_restore(device_t dev,struct pci_devinfo * dinfo)6279 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6280 {
6281 
6282 	/*
6283 	 * Restore the device to full power mode.  We must do this
6284 	 * before we restore the registers because moving from D3 to
6285 	 * D0 will cause the chip's BARs and some other registers to
6286 	 * be reset to some unknown power on reset values.  Cut down
6287 	 * the noise on boot by doing nothing if we are already in
6288 	 * state D0.
6289 	 */
6290 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6291 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6292 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6293 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6294 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6295 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6296 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6297 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6298 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6299 	case PCIM_HDRTYPE_NORMAL:
6300 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6301 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6302 		break;
6303 	case PCIM_HDRTYPE_BRIDGE:
6304 		pci_write_config(dev, PCIR_SECLAT_1,
6305 		    dinfo->cfg.bridge.br_seclat, 1);
6306 		pci_write_config(dev, PCIR_SUBBUS_1,
6307 		    dinfo->cfg.bridge.br_subbus, 1);
6308 		pci_write_config(dev, PCIR_SECBUS_1,
6309 		    dinfo->cfg.bridge.br_secbus, 1);
6310 		pci_write_config(dev, PCIR_PRIBUS_1,
6311 		    dinfo->cfg.bridge.br_pribus, 1);
6312 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6313 		    dinfo->cfg.bridge.br_control, 2);
6314 		break;
6315 	case PCIM_HDRTYPE_CARDBUS:
6316 		pci_write_config(dev, PCIR_SECLAT_2,
6317 		    dinfo->cfg.bridge.br_seclat, 1);
6318 		pci_write_config(dev, PCIR_SUBBUS_2,
6319 		    dinfo->cfg.bridge.br_subbus, 1);
6320 		pci_write_config(dev, PCIR_SECBUS_2,
6321 		    dinfo->cfg.bridge.br_secbus, 1);
6322 		pci_write_config(dev, PCIR_PRIBUS_2,
6323 		    dinfo->cfg.bridge.br_pribus, 1);
6324 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6325 		    dinfo->cfg.bridge.br_control, 2);
6326 		break;
6327 	}
6328 	pci_restore_bars(dev);
6329 
6330 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6331 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6332 
6333 	/*
6334 	 * Restore extended capabilities for PCI-Express and PCI-X
6335 	 */
6336 	if (dinfo->cfg.pcie.pcie_location != 0)
6337 		pci_cfg_restore_pcie(dev, dinfo);
6338 	if (dinfo->cfg.pcix.pcix_location != 0)
6339 		pci_cfg_restore_pcix(dev, dinfo);
6340 
6341 	/* Restore MSI and MSI-X configurations if they are present. */
6342 	if (dinfo->cfg.msi.msi_location != 0)
6343 		pci_resume_msi(dev);
6344 	if (dinfo->cfg.msix.msix_location != 0)
6345 		pci_resume_msix(dev);
6346 
6347 #ifdef PCI_IOV
6348 	if (dinfo->cfg.iov != NULL)
6349 		pci_iov_cfg_restore(dev, dinfo);
6350 #endif
6351 }
6352 
6353 static void
pci_cfg_save_pcie(device_t dev,struct pci_devinfo * dinfo)6354 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6355 {
6356 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6357 	struct pcicfg_pcie *cfg;
6358 	int version, pos;
6359 
6360 	cfg = &dinfo->cfg.pcie;
6361 	pos = cfg->pcie_location;
6362 
6363 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6364 
6365 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6366 
6367 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6368 
6369 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6370 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6371 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6372 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6373 
6374 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6375 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6376 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6377 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6378 
6379 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6380 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6381 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6382 
6383 	if (version > 1) {
6384 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6385 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6386 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6387 	}
6388 #undef RREG
6389 }
6390 
6391 static void
pci_cfg_save_pcix(device_t dev,struct pci_devinfo * dinfo)6392 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6393 {
6394 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6395 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6396 }
6397 
6398 void
pci_cfg_save(device_t dev,struct pci_devinfo * dinfo,int setstate)6399 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6400 {
6401 	uint32_t cls;
6402 	int ps;
6403 
6404 	/*
6405 	 * Some drivers apparently write to these registers w/o updating our
6406 	 * cached copy.  No harm happens if we update the copy, so do so here
6407 	 * so we can restore them.  The COMMAND register is modified by the
6408 	 * bus w/o updating the cache.  This should represent the normally
6409 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6410 	 */
6411 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6412 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6413 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6414 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6415 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6416 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6417 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6418 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6419 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6420 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6421 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6422 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6423 	case PCIM_HDRTYPE_NORMAL:
6424 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6425 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6426 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6427 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6428 		break;
6429 	case PCIM_HDRTYPE_BRIDGE:
6430 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6431 		    PCIR_SECLAT_1, 1);
6432 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6433 		    PCIR_SUBBUS_1, 1);
6434 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6435 		    PCIR_SECBUS_1, 1);
6436 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6437 		    PCIR_PRIBUS_1, 1);
6438 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6439 		    PCIR_BRIDGECTL_1, 2);
6440 		break;
6441 	case PCIM_HDRTYPE_CARDBUS:
6442 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6443 		    PCIR_SECLAT_2, 1);
6444 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6445 		    PCIR_SUBBUS_2, 1);
6446 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6447 		    PCIR_SECBUS_2, 1);
6448 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6449 		    PCIR_PRIBUS_2, 1);
6450 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6451 		    PCIR_BRIDGECTL_2, 2);
6452 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6453 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6454 		break;
6455 	}
6456 
6457 	if (dinfo->cfg.pcie.pcie_location != 0)
6458 		pci_cfg_save_pcie(dev, dinfo);
6459 
6460 	if (dinfo->cfg.pcix.pcix_location != 0)
6461 		pci_cfg_save_pcix(dev, dinfo);
6462 
6463 #ifdef PCI_IOV
6464 	if (dinfo->cfg.iov != NULL)
6465 		pci_iov_cfg_save(dev, dinfo);
6466 #endif
6467 
6468 	/*
6469 	 * don't set the state for display devices, base peripherals and
6470 	 * memory devices since bad things happen when they are powered down.
6471 	 * We should (a) have drivers that can easily detach and (b) use
6472 	 * generic drivers for these devices so that some device actually
6473 	 * attaches.  We need to make sure that when we implement (a) we don't
6474 	 * power the device down on a reattach.
6475 	 */
6476 	cls = pci_get_class(dev);
6477 	if (!setstate)
6478 		return;
6479 	switch (pci_do_power_nodriver)
6480 	{
6481 		case 0:		/* NO powerdown at all */
6482 			return;
6483 		case 1:		/* Conservative about what to power down */
6484 			if (cls == PCIC_STORAGE)
6485 				return;
6486 			/*FALLTHROUGH*/
6487 		case 2:		/* Aggressive about what to power down */
6488 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6489 			    cls == PCIC_BASEPERIPH)
6490 				return;
6491 			/*FALLTHROUGH*/
6492 		case 3:		/* Power down everything */
6493 			break;
6494 	}
6495 	/*
6496 	 * PCI spec says we can only go into D3 state from D0 state.
6497 	 * Transition from D[12] into D0 before going to D3 state.
6498 	 */
6499 	ps = pci_get_powerstate(dev);
6500 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6501 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6502 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6503 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6504 }
6505 
6506 /* Wrapper APIs suitable for device driver use. */
6507 void
pci_save_state(device_t dev)6508 pci_save_state(device_t dev)
6509 {
6510 	struct pci_devinfo *dinfo;
6511 
6512 	dinfo = device_get_ivars(dev);
6513 	pci_cfg_save(dev, dinfo, 0);
6514 }
6515 
6516 void
pci_restore_state(device_t dev)6517 pci_restore_state(device_t dev)
6518 {
6519 	struct pci_devinfo *dinfo;
6520 
6521 	dinfo = device_get_ivars(dev);
6522 	pci_cfg_restore(dev, dinfo);
6523 }
6524 
6525 static int
pci_get_id_method(device_t dev,device_t child,enum pci_id_type type,uintptr_t * id)6526 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6527     uintptr_t *id)
6528 {
6529 
6530 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6531 }
6532 
6533 /* Find the upstream port of a given PCI device in a root complex. */
6534 device_t
pci_find_pcie_root_port(device_t dev)6535 pci_find_pcie_root_port(device_t dev)
6536 {
6537 	struct pci_devinfo *dinfo;
6538 	devclass_t pci_class;
6539 	device_t pcib, bus;
6540 
6541 	pci_class = devclass_find("pci");
6542 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6543 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6544 
6545 	/*
6546 	 * Walk the bridge hierarchy until we find a PCI-e root
6547 	 * port or a non-PCI device.
6548 	 */
6549 	for (;;) {
6550 		bus = device_get_parent(dev);
6551 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6552 		    device_get_nameunit(dev)));
6553 
6554 		pcib = device_get_parent(bus);
6555 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6556 		    device_get_nameunit(bus)));
6557 
6558 		/*
6559 		 * pcib's parent must be a PCI bus for this to be a
6560 		 * PCI-PCI bridge.
6561 		 */
6562 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6563 			return (NULL);
6564 
6565 		dinfo = device_get_ivars(pcib);
6566 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6567 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6568 			return (pcib);
6569 
6570 		dev = pcib;
6571 	}
6572 }
6573 
6574 /*
6575  * Wait for pending transactions to complete on a PCI-express function.
6576  *
6577  * The maximum delay is specified in milliseconds in max_delay.  Note
6578  * that this function may sleep.
6579  *
6580  * Returns true if the function is idle and false if the timeout is
6581  * exceeded.  If dev is not a PCI-express function, this returns true.
6582  */
6583 bool
pcie_wait_for_pending_transactions(device_t dev,u_int max_delay)6584 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6585 {
6586 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6587 	uint16_t sta;
6588 	int cap;
6589 
6590 	cap = dinfo->cfg.pcie.pcie_location;
6591 	if (cap == 0)
6592 		return (true);
6593 
6594 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6595 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6596 		if (max_delay == 0)
6597 			return (false);
6598 
6599 		/* Poll once every 100 milliseconds up to the timeout. */
6600 		if (max_delay > 100) {
6601 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6602 			max_delay -= 100;
6603 		} else {
6604 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6605 			    C_HARDCLOCK);
6606 			max_delay = 0;
6607 		}
6608 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6609 	}
6610 
6611 	return (true);
6612 }
6613 
6614 /*
6615  * Determine the maximum Completion Timeout in microseconds.
6616  *
6617  * For non-PCI-express functions this returns 0.
6618  */
6619 int
pcie_get_max_completion_timeout(device_t dev)6620 pcie_get_max_completion_timeout(device_t dev)
6621 {
6622 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6623 	int cap;
6624 
6625 	cap = dinfo->cfg.pcie.pcie_location;
6626 	if (cap == 0)
6627 		return (0);
6628 
6629 	/*
6630 	 * Functions using the 1.x spec use the default timeout range of
6631 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6632 	 * support programmable timeouts also use this range.
6633 	 */
6634 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6635 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6636 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6637 		return (50 * 1000);
6638 
6639 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6640 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6641 	case PCIEM_CTL2_COMP_TIMO_100US:
6642 		return (100);
6643 	case PCIEM_CTL2_COMP_TIMO_10MS:
6644 		return (10 * 1000);
6645 	case PCIEM_CTL2_COMP_TIMO_55MS:
6646 		return (55 * 1000);
6647 	case PCIEM_CTL2_COMP_TIMO_210MS:
6648 		return (210 * 1000);
6649 	case PCIEM_CTL2_COMP_TIMO_900MS:
6650 		return (900 * 1000);
6651 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6652 		return (3500 * 1000);
6653 	case PCIEM_CTL2_COMP_TIMO_13S:
6654 		return (13 * 1000 * 1000);
6655 	case PCIEM_CTL2_COMP_TIMO_64S:
6656 		return (64 * 1000 * 1000);
6657 	default:
6658 		return (50 * 1000);
6659 	}
6660 }
6661 
6662 void
pcie_apei_error(device_t dev,int sev,uint8_t * aerp)6663 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6664 {
6665 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6666 	const char *s;
6667 	int aer;
6668 	uint32_t r, r1;
6669 	uint16_t rs;
6670 
6671 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6672 		s = "Correctable";
6673 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6674 		s = "Uncorrectable (Non-Fatal)";
6675 	else
6676 		s = "Uncorrectable (Fatal)";
6677 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6678 	if (aerp) {
6679 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6680 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6681 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6682 		} else {
6683 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6684 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6685 		}
6686 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6687 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6688 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6689 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6690 			printf(" severity 0x%08x first %d\n",
6691 			    r, rs & 0x1f);
6692 		} else
6693 			printf("\n");
6694 	}
6695 
6696 	/* As kind of recovery just report and clear the error statuses. */
6697 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6698 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6699 		if (r != 0) {
6700 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6701 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6702 		}
6703 
6704 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6705 		if (r != 0) {
6706 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6707 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6708 		}
6709 	}
6710 	if (dinfo->cfg.pcie.pcie_location != 0) {
6711 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6712 		    PCIER_DEVICE_STA, 2);
6713 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6714 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6715 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6716 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6717 			    PCIER_DEVICE_STA, rs, 2);
6718 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6719 		}
6720 	}
6721 }
6722 
6723 /*
6724  * Perform a Function Level Reset (FLR) on a device.
6725  *
6726  * This function first waits for any pending transactions to complete
6727  * within the timeout specified by max_delay.  If transactions are
6728  * still pending, the function will return false without attempting a
6729  * reset.
6730  *
6731  * If dev is not a PCI-express function or does not support FLR, this
6732  * function returns false.
6733  *
6734  * Note that no registers are saved or restored.  The caller is
6735  * responsible for saving and restoring any registers including
6736  * PCI-standard registers via pci_save_state() and
6737  * pci_restore_state().
6738  */
6739 bool
pcie_flr(device_t dev,u_int max_delay,bool force)6740 pcie_flr(device_t dev, u_int max_delay, bool force)
6741 {
6742 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6743 	uint16_t cmd, ctl;
6744 	int compl_delay;
6745 	int cap;
6746 
6747 	cap = dinfo->cfg.pcie.pcie_location;
6748 	if (cap == 0)
6749 		return (false);
6750 
6751 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6752 		return (false);
6753 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_FLR))
6754 		return (false);
6755 
6756 	/*
6757 	 * Disable busmastering to prevent generation of new
6758 	 * transactions while waiting for the device to go idle.  If
6759 	 * the idle timeout fails, the command register is restored
6760 	 * which will re-enable busmastering.
6761 	 */
6762 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6763 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6764 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6765 		if (!force) {
6766 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6767 			return (false);
6768 		}
6769 		pci_printf(&dinfo->cfg,
6770 		    "Resetting with transactions pending after %d ms\n",
6771 		    max_delay);
6772 
6773 		/*
6774 		 * Extend the post-FLR delay to cover the maximum
6775 		 * Completion Timeout delay of anything in flight
6776 		 * during the FLR delay.  Enforce a minimum delay of
6777 		 * at least 10ms.
6778 		 */
6779 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6780 		if (compl_delay < 10)
6781 			compl_delay = 10;
6782 	} else
6783 		compl_delay = 0;
6784 
6785 	/* Initiate the reset. */
6786 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6787 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6788 	    PCIEM_CTL_INITIATE_FLR, 2);
6789 
6790 	/* Wait for 100ms. */
6791 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6792 
6793 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6794 	    PCIEM_STA_TRANSACTION_PND)
6795 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6796 	return (true);
6797 }
6798 
6799 /*
6800  * Attempt a power-management reset by cycling the device in/out of D3
6801  * state.  PCI spec says we can only go into D3 state from D0 state.
6802  * Transition from D[12] into D0 before going to D3 state.
6803  */
6804 int
pci_power_reset(device_t dev)6805 pci_power_reset(device_t dev)
6806 {
6807 	int ps;
6808 
6809 	ps = pci_get_powerstate(dev);
6810 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6811 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6812 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6813 	pci_set_powerstate(dev, ps);
6814 	return (0);
6815 }
6816 
6817 /*
6818  * Try link drop and retrain of the downstream port of upstream
6819  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6820  * cause Conventional Hot reset of the device in the slot.
6821  * Alternative, for PCIe, could be the secondary bus reset initiatied
6822  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6823  */
6824 int
pcie_link_reset(device_t port,int pcie_location)6825 pcie_link_reset(device_t port, int pcie_location)
6826 {
6827 	uint16_t v;
6828 
6829 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6830 	v |= PCIEM_LINK_CTL_LINK_DIS;
6831 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6832 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6833 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6834 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6835 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6836 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6837 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6838 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6839 }
6840 
6841 static int
pci_reset_post(device_t dev,device_t child)6842 pci_reset_post(device_t dev, device_t child)
6843 {
6844 
6845 	if (dev == device_get_parent(child))
6846 		pci_restore_state(child);
6847 	return (0);
6848 }
6849 
6850 static int
pci_reset_prepare(device_t dev,device_t child)6851 pci_reset_prepare(device_t dev, device_t child)
6852 {
6853 
6854 	if (dev == device_get_parent(child))
6855 		pci_save_state(child);
6856 	return (0);
6857 }
6858 
6859 static int
pci_reset_child(device_t dev,device_t child,int flags)6860 pci_reset_child(device_t dev, device_t child, int flags)
6861 {
6862 	int error;
6863 
6864 	if (dev == NULL || device_get_parent(child) != dev)
6865 		return (0);
6866 	if ((flags & DEVF_RESET_DETACH) != 0) {
6867 		error = device_get_state(child) == DS_ATTACHED ?
6868 		    device_detach(child) : 0;
6869 	} else {
6870 		error = BUS_SUSPEND_CHILD(dev, child);
6871 	}
6872 	if (error == 0) {
6873 		if (!pcie_flr(child, 1000, false)) {
6874 			error = BUS_RESET_PREPARE(dev, child);
6875 			if (error == 0)
6876 				pci_power_reset(child);
6877 			BUS_RESET_POST(dev, child);
6878 		}
6879 		if ((flags & DEVF_RESET_DETACH) != 0)
6880 			device_probe_and_attach(child);
6881 		else
6882 			BUS_RESUME_CHILD(dev, child);
6883 	}
6884 	return (error);
6885 }
6886 
6887 const struct pci_device_table *
pci_match_device(device_t child,const struct pci_device_table * id,size_t nelt)6888 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6889 {
6890 	bool match;
6891 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6892 
6893 	vendor = pci_get_vendor(child);
6894 	device = pci_get_device(child);
6895 	subvendor = pci_get_subvendor(child);
6896 	subdevice = pci_get_subdevice(child);
6897 	class = pci_get_class(child);
6898 	subclass = pci_get_subclass(child);
6899 	revid = pci_get_revid(child);
6900 	while (nelt-- > 0) {
6901 		match = true;
6902 		if (id->match_flag_vendor)
6903 			match &= vendor == id->vendor;
6904 		if (id->match_flag_device)
6905 			match &= device == id->device;
6906 		if (id->match_flag_subvendor)
6907 			match &= subvendor == id->subvendor;
6908 		if (id->match_flag_subdevice)
6909 			match &= subdevice == id->subdevice;
6910 		if (id->match_flag_class)
6911 			match &= class == id->class_id;
6912 		if (id->match_flag_subclass)
6913 			match &= subclass == id->subclass;
6914 		if (id->match_flag_revid)
6915 			match &= revid == id->revid;
6916 		if (match)
6917 			return (id);
6918 		id++;
6919 	}
6920 	return (NULL);
6921 }
6922 
6923 static void
pci_print_faulted_dev_name(const struct pci_devinfo * dinfo)6924 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6925 {
6926 	const char *dev_name;
6927 	device_t dev;
6928 
6929 	dev = dinfo->cfg.dev;
6930 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6931 	    dinfo->cfg.slot, dinfo->cfg.func);
6932 	dev_name = device_get_name(dev);
6933 	if (dev_name != NULL)
6934 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6935 }
6936 
6937 void
pci_print_faulted_dev(void)6938 pci_print_faulted_dev(void)
6939 {
6940 	struct pci_devinfo *dinfo;
6941 	device_t dev;
6942 	int aer, i;
6943 	uint32_t r1, r2;
6944 	uint16_t status;
6945 
6946 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6947 		dev = dinfo->cfg.dev;
6948 		status = pci_read_config(dev, PCIR_STATUS, 2);
6949 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6950 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6951 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6952 		if (status != 0) {
6953 			pci_print_faulted_dev_name(dinfo);
6954 			printf(" error 0x%04x\n", status);
6955 		}
6956 		if (dinfo->cfg.pcie.pcie_location != 0) {
6957 			status = pci_read_config(dev,
6958 			    dinfo->cfg.pcie.pcie_location +
6959 			    PCIER_DEVICE_STA, 2);
6960 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6961 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6962 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6963 				pci_print_faulted_dev_name(dinfo);
6964 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6965 				    pci_read_config(dev,
6966 				    dinfo->cfg.pcie.pcie_location +
6967 				    PCIER_DEVICE_CTL, 2),
6968 				    status);
6969 			}
6970 		}
6971 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6972 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6973 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6974 			if (r1 != 0 || r2 != 0) {
6975 				pci_print_faulted_dev_name(dinfo);
6976 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6977 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6978 				    r1, pci_read_config(dev, aer +
6979 				    PCIR_AER_UC_MASK, 4),
6980 				    pci_read_config(dev, aer +
6981 				    PCIR_AER_UC_SEVERITY, 4),
6982 				    r2, pci_read_config(dev, aer +
6983 				    PCIR_AER_COR_MASK, 4),
6984 				    pci_read_config(dev, aer +
6985 				    PCIR_AER_CAP_CONTROL, 4));
6986 				for (i = 0; i < 4; i++) {
6987 					r1 = pci_read_config(dev, aer +
6988 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6989 					printf("    HL%d: 0x%08x\n", i, r1);
6990 				}
6991 			}
6992 		}
6993 	}
6994 }
6995 
6996 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(pcierr,pci_print_faulted_dev_db,DB_CMD_MEMSAFE)6997 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6998 {
6999 
7000 	pci_print_faulted_dev();
7001 }
7002 
7003 static void
db_clear_pcie_errors(const struct pci_devinfo * dinfo)7004 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
7005 {
7006 	device_t dev;
7007 	int aer;
7008 	uint32_t r;
7009 
7010 	dev = dinfo->cfg.dev;
7011 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
7012 	    PCIER_DEVICE_STA, 2);
7013 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
7014 	    PCIER_DEVICE_STA, r, 2);
7015 
7016 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
7017 		return;
7018 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
7019 	if (r != 0)
7020 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
7021 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
7022 	if (r != 0)
7023 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
7024 }
7025 
DB_COMMAND_FLAGS(pci_clearerr,db_pci_clearerr,DB_CMD_MEMSAFE)7026 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
7027 {
7028 	struct pci_devinfo *dinfo;
7029 	device_t dev;
7030 	uint16_t status, status1;
7031 
7032 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
7033 		dev = dinfo->cfg.dev;
7034 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
7035 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
7036 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
7037 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
7038 		if (status1 != 0) {
7039 			status &= ~status1;
7040 			pci_write_config(dev, PCIR_STATUS, status, 2);
7041 		}
7042 		if (dinfo->cfg.pcie.pcie_location != 0)
7043 			db_clear_pcie_errors(dinfo);
7044 	}
7045 }
7046 #endif
7047