xref: /freebsd/sys/dev/pci/pci.c (revision 346020138a0fd20085ebc285f090df38d7d18527)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_acpi.h"
33 #include "opt_iommu.h"
34 #include "opt_bus.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/linker.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 #include <sys/taskqueue.h>
51 #include <sys/tree.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 
57 #include <sys/bus.h>
58 #include <machine/bus.h>
59 #include <sys/rman.h>
60 #include <machine/resource.h>
61 #include <machine/stdarg.h>
62 
63 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
64 #include <machine/intr_machdep.h>
65 #endif
66 
67 #include <sys/pciio.h>
68 #include <dev/pci/pcireg.h>
69 #include <dev/pci/pcivar.h>
70 #include <dev/pci/pci_private.h>
71 
72 #ifdef PCI_IOV
73 #include <sys/nv.h>
74 #include <dev/pci/pci_iov_private.h>
75 #endif
76 
77 #include <dev/usb/controller/xhcireg.h>
78 #include <dev/usb/controller/ehcireg.h>
79 #include <dev/usb/controller/ohcireg.h>
80 #include <dev/usb/controller/uhcireg.h>
81 
82 #include <dev/iommu/iommu.h>
83 
84 #include "pcib_if.h"
85 #include "pci_if.h"
86 
87 #define	PCIR_IS_BIOS(cfg, reg)						\
88 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
89 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
90 
91 static 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 	int	arg1;
244 	int	arg2;
245 };
246 
247 static const struct pci_quirk pci_quirks[] = {
248 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
249 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
250 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
252 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 
254 	/*
255 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
256 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
257 	 */
258 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
259 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
260 
261 	/*
262 	 * MSI doesn't work on earlier Intel chipsets including
263 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
264 	 */
265 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 
273 	/*
274 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
275 	 * bridge.
276 	 */
277 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
278 
279 	/*
280 	 * Some virtualization environments emulate an older chipset
281 	 * but support MSI just fine.  QEMU uses the Intel 82440.
282 	 */
283 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
284 
285 	/*
286 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
287 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
288 	 * It prevents us from attaching hpet(4) when the bit is unset.
289 	 * Note this quirk only affects SB600 revision A13 and earlier.
290 	 * For SB600 A21 and later, firmware must set the bit to hide it.
291 	 * For SB700 and later, it is unused and hardcoded to zero.
292 	 */
293 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
294 
295 	/*
296 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
297 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
298 	 * of the command register is set.
299 	 */
300 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 
306 	/*
307 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
308 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
309 	 */
310 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
311 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
312 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
313 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
314 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
315 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
316 
317 	/*
318 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
319 	 * expected place.
320 	 */
321 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
322 	{ 0 }
323 };
324 
325 /* map register information */
326 #define	PCI_MAPMEM	0x01	/* memory map */
327 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
328 #define	PCI_MAPPORT	0x04	/* port map */
329 
330 struct devlist pci_devq;
331 uint32_t pci_generation;
332 uint32_t pci_numdevs = 0;
333 static int pcie_chipset, pcix_chipset;
334 
335 /* sysctl vars */
336 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
337     "PCI bus tuning parameters");
338 
339 static int pci_enable_io_modes = 1;
340 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
341     &pci_enable_io_modes, 1,
342     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
343     " enable these bits correctly.  We'd like to do this all the time, but"
344     " there are some peripherals that this causes problems with.");
345 
346 static int pci_do_realloc_bars = 1;
347 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
348     &pci_do_realloc_bars, 0,
349     "Attempt to allocate a new range for any BARs whose original "
350     "firmware-assigned ranges fail to allocate during the initial device scan.");
351 
352 static int pci_do_power_nodriver = 0;
353 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
354     &pci_do_power_nodriver, 0,
355     "Place a function into D3 state when no driver attaches to it.  0 means"
356     " disable.  1 means conservatively place function into D3 state.  2 means"
357     " aggressively place function into D3 state.  3 means put absolutely"
358     " everything in D3 state.");
359 
360 int pci_do_power_resume = 1;
361 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
362     &pci_do_power_resume, 1,
363   "Transition from D3 -> D0 on resume.");
364 
365 int pci_do_power_suspend = 1;
366 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
367     &pci_do_power_suspend, 1,
368   "Transition from D0 -> D3 on suspend.");
369 
370 static int pci_do_msi = 1;
371 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
372     "Enable support for MSI interrupts");
373 
374 static int pci_do_msix = 1;
375 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
376     "Enable support for MSI-X interrupts");
377 
378 static int pci_msix_rewrite_table = 0;
379 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
380     &pci_msix_rewrite_table, 0,
381     "Rewrite entire MSI-X table when updating MSI-X entries");
382 
383 static int pci_honor_msi_blacklist = 1;
384 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
385     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
386 
387 #if defined(__i386__) || defined(__amd64__)
388 static int pci_usb_takeover = 1;
389 #else
390 static int pci_usb_takeover = 0;
391 #endif
392 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
393     &pci_usb_takeover, 1,
394     "Enable early takeover of USB controllers. Disable this if you depend on"
395     " BIOS emulation of USB devices, that is you use USB devices (like"
396     " keyboard or mouse) but do not load USB drivers");
397 
398 static int pci_clear_bars;
399 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
400     "Ignore firmware-assigned resources for BARs.");
401 
402 static int pci_clear_buses;
403 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
404     "Ignore firmware-assigned bus numbers.");
405 
406 static int pci_enable_ari = 1;
407 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
408     0, "Enable support for PCIe Alternative RID Interpretation");
409 
410 /*
411  * Some x86 firmware only enables PCIe hotplug if we claim to support aspm,
412  * however enabling it breaks some arm64 firmware as it powers off devices.
413  */
414 #if defined(__i386__) || defined(__amd64__)
415 int pci_enable_aspm = 1;
416 #else
417 int pci_enable_aspm = 0;
418 #endif
419 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
420     0, "Enable support for PCIe Active State Power Management");
421 
422 static int pci_clear_aer_on_attach = 0;
423 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
424     &pci_clear_aer_on_attach, 0,
425     "Clear port and device AER state on driver attach");
426 
427 static bool pci_enable_mps_tune = true;
428 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
429     &pci_enable_mps_tune, 1,
430     "Enable tuning of MPS(maximum payload size)." );
431 
432 static int
pci_has_quirk(uint32_t devid,int quirk)433 pci_has_quirk(uint32_t devid, int quirk)
434 {
435 	const struct pci_quirk *q;
436 
437 	for (q = &pci_quirks[0]; q->devid; q++) {
438 		if (q->devid == devid && q->type == quirk)
439 			return (1);
440 	}
441 	return (0);
442 }
443 
444 /* Find a device_t by bus/slot/function in domain 0 */
445 
446 device_t
pci_find_bsf(uint8_t bus,uint8_t slot,uint8_t func)447 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
448 {
449 
450 	return (pci_find_dbsf(0, bus, slot, func));
451 }
452 
453 /* Find a device_t by domain/bus/slot/function */
454 
455 device_t
pci_find_dbsf(uint32_t domain,uint8_t bus,uint8_t slot,uint8_t func)456 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
457 {
458 	struct pci_devinfo *dinfo = NULL;
459 
460 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
461 		if ((dinfo->cfg.domain == domain) &&
462 		    (dinfo->cfg.bus == bus) &&
463 		    (dinfo->cfg.slot == slot) &&
464 		    (dinfo->cfg.func == func)) {
465 			break;
466 		}
467 	}
468 
469 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
470 }
471 
472 /* Find a device_t by vendor/device ID */
473 
474 device_t
pci_find_device(uint16_t vendor,uint16_t device)475 pci_find_device(uint16_t vendor, uint16_t device)
476 {
477 	struct pci_devinfo *dinfo;
478 
479 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
480 		if ((dinfo->cfg.vendor == vendor) &&
481 		    (dinfo->cfg.device == device)) {
482 			return (dinfo->cfg.dev);
483 		}
484 	}
485 
486 	return (NULL);
487 }
488 
489 device_t
pci_find_class(uint8_t class,uint8_t subclass)490 pci_find_class(uint8_t class, uint8_t subclass)
491 {
492 	struct pci_devinfo *dinfo;
493 
494 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
495 		if (dinfo->cfg.baseclass == class &&
496 		    dinfo->cfg.subclass == subclass) {
497 			return (dinfo->cfg.dev);
498 		}
499 	}
500 
501 	return (NULL);
502 }
503 
504 device_t
pci_find_class_from(uint8_t class,uint8_t subclass,device_t from)505 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
506 {
507 	struct pci_devinfo *dinfo;
508 	bool found = false;
509 
510 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
511 		if (from != NULL && found == false) {
512 			if (from != dinfo->cfg.dev)
513 				continue;
514 			found = true;
515 			continue;
516 		}
517 		if (dinfo->cfg.baseclass == class &&
518 		    dinfo->cfg.subclass == subclass) {
519 			return (dinfo->cfg.dev);
520 		}
521 	}
522 
523 	return (NULL);
524 }
525 
526 device_t
pci_find_base_class_from(uint8_t class,device_t from)527 pci_find_base_class_from(uint8_t class, device_t from)
528 {
529 	struct pci_devinfo *dinfo;
530 	bool found = false;
531 
532 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
533 		if (from != NULL && found == false) {
534 			if (from != dinfo->cfg.dev)
535 				continue;
536 			found = true;
537 			continue;
538 		}
539 		if (dinfo->cfg.baseclass == class) {
540 			return (dinfo->cfg.dev);
541 		}
542 	}
543 
544 	return (NULL);
545 }
546 
547 static int
pci_printf(pcicfgregs * cfg,const char * fmt,...)548 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
549 {
550 	va_list ap;
551 	int retval;
552 
553 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
554 	    cfg->func);
555 	va_start(ap, fmt);
556 	retval += vprintf(fmt, ap);
557 	va_end(ap);
558 	return (retval);
559 }
560 
561 /* return base address of memory or port map */
562 
563 static pci_addr_t
pci_mapbase(uint64_t mapreg)564 pci_mapbase(uint64_t mapreg)
565 {
566 
567 	if (PCI_BAR_MEM(mapreg))
568 		return (mapreg & PCIM_BAR_MEM_BASE);
569 	else
570 		return (mapreg & PCIM_BAR_IO_BASE);
571 }
572 
573 /* return map type of memory or port map */
574 
575 static const char *
pci_maptype(uint64_t mapreg)576 pci_maptype(uint64_t mapreg)
577 {
578 
579 	if (PCI_BAR_IO(mapreg))
580 		return ("I/O Port");
581 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
582 		return ("Prefetchable Memory");
583 	return ("Memory");
584 }
585 
586 /* return log2 of map size decoded for memory or port map */
587 
588 int
pci_mapsize(uint64_t testval)589 pci_mapsize(uint64_t testval)
590 {
591 	int ln2size;
592 
593 	testval = pci_mapbase(testval);
594 	ln2size = 0;
595 	if (testval != 0) {
596 		while ((testval & 1) == 0)
597 		{
598 			ln2size++;
599 			testval >>= 1;
600 		}
601 	}
602 	return (ln2size);
603 }
604 
605 /* return base address of device ROM */
606 
607 static pci_addr_t
pci_rombase(uint64_t mapreg)608 pci_rombase(uint64_t mapreg)
609 {
610 
611 	return (mapreg & PCIM_BIOS_ADDR_MASK);
612 }
613 
614 /* return log2 of map size decided for device ROM */
615 
616 static int
pci_romsize(uint64_t testval)617 pci_romsize(uint64_t testval)
618 {
619 	int ln2size;
620 
621 	testval = pci_rombase(testval);
622 	ln2size = 0;
623 	if (testval != 0) {
624 		while ((testval & 1) == 0)
625 		{
626 			ln2size++;
627 			testval >>= 1;
628 		}
629 	}
630 	return (ln2size);
631 }
632 
633 /* return log2 of address range supported by map register */
634 
635 static int
pci_maprange(uint64_t mapreg)636 pci_maprange(uint64_t mapreg)
637 {
638 	int ln2range = 0;
639 
640 	if (PCI_BAR_IO(mapreg))
641 		ln2range = 32;
642 	else
643 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
644 		case PCIM_BAR_MEM_32:
645 			ln2range = 32;
646 			break;
647 		case PCIM_BAR_MEM_1MB:
648 			ln2range = 20;
649 			break;
650 		case PCIM_BAR_MEM_64:
651 			ln2range = 64;
652 			break;
653 		}
654 	return (ln2range);
655 }
656 
657 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
658 
659 static void
pci_fixancient(pcicfgregs * cfg)660 pci_fixancient(pcicfgregs *cfg)
661 {
662 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
663 		return;
664 
665 	/* PCI to PCI bridges use header type 1 */
666 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
667 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
668 }
669 
670 /* extract header type specific config data */
671 
672 static void
pci_hdrtypedata(device_t pcib,int b,int s,int f,pcicfgregs * cfg)673 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
674 {
675 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
676 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
677 	case PCIM_HDRTYPE_NORMAL:
678 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
679 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
680 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
681 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
682 		cfg->nummaps	    = PCI_MAXMAPS_0;
683 		break;
684 	case PCIM_HDRTYPE_BRIDGE:
685 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
686 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
687 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
688 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
689 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
690 		cfg->nummaps	    = PCI_MAXMAPS_1;
691 		break;
692 	case PCIM_HDRTYPE_CARDBUS:
693 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
694 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
695 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
696 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
697 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
698 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
699 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
700 		cfg->nummaps	    = PCI_MAXMAPS_2;
701 		break;
702 	}
703 #undef REG
704 }
705 
706 /* read configuration header into pcicfgregs structure */
707 struct pci_devinfo *
pci_read_device(device_t pcib,device_t bus,int d,int b,int s,int f)708 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
709 {
710 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
711 	uint16_t vid, did;
712 
713 	vid = REG(PCIR_VENDOR, 2);
714 	if (vid == PCIV_INVALID)
715 		return (NULL);
716 
717 	did = REG(PCIR_DEVICE, 2);
718 
719 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
720 }
721 
722 struct pci_devinfo *
pci_alloc_devinfo_method(device_t dev)723 pci_alloc_devinfo_method(device_t dev)
724 {
725 
726 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
727 	    M_WAITOK | M_ZERO));
728 }
729 
730 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)731 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
732     uint16_t vid, uint16_t did)
733 {
734 	struct pci_devinfo *devlist_entry;
735 	pcicfgregs *cfg;
736 
737 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
738 
739 	cfg = &devlist_entry->cfg;
740 
741 	cfg->domain		= d;
742 	cfg->bus		= b;
743 	cfg->slot		= s;
744 	cfg->func		= f;
745 	cfg->vendor		= vid;
746 	cfg->device		= did;
747 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
748 	cfg->statreg		= REG(PCIR_STATUS, 2);
749 	cfg->baseclass		= REG(PCIR_CLASS, 1);
750 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
751 	cfg->progif		= REG(PCIR_PROGIF, 1);
752 	cfg->revid		= REG(PCIR_REVID, 1);
753 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
754 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
755 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
756 	cfg->intpin		= REG(PCIR_INTPIN, 1);
757 	cfg->intline		= REG(PCIR_INTLINE, 1);
758 
759 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
760 	cfg->hdrtype		&= ~PCIM_MFDEV;
761 	STAILQ_INIT(&cfg->maps);
762 
763 	cfg->iov		= NULL;
764 
765 	pci_fixancient(cfg);
766 	pci_hdrtypedata(pcib, b, s, f, cfg);
767 
768 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
769 		pci_read_cap(pcib, cfg);
770 
771 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
772 
773 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
774 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
775 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
776 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
777 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
778 
779 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
780 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
781 	devlist_entry->conf.pc_vendor = cfg->vendor;
782 	devlist_entry->conf.pc_device = cfg->device;
783 
784 	devlist_entry->conf.pc_class = cfg->baseclass;
785 	devlist_entry->conf.pc_subclass = cfg->subclass;
786 	devlist_entry->conf.pc_progif = cfg->progif;
787 	devlist_entry->conf.pc_revid = cfg->revid;
788 
789 	pci_numdevs++;
790 	pci_generation++;
791 
792 	return (devlist_entry);
793 }
794 #undef REG
795 
796 static void
pci_ea_fill_info(device_t pcib,pcicfgregs * cfg)797 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
798 {
799 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
800     cfg->ea.ea_location + (n), w)
801 	int num_ent;
802 	int ptr;
803 	int a, b;
804 	uint32_t val;
805 	int ent_size;
806 	uint32_t dw[4];
807 	uint64_t base, max_offset;
808 	struct pci_ea_entry *eae;
809 
810 	if (cfg->ea.ea_location == 0)
811 		return;
812 
813 	STAILQ_INIT(&cfg->ea.ea_entries);
814 
815 	/* Determine the number of entries */
816 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
817 	num_ent &= PCIM_EA_NUM_ENT_MASK;
818 
819 	/* Find the first entry to care of */
820 	ptr = PCIR_EA_FIRST_ENT;
821 
822 	/* Skip DWORD 2 for type 1 functions */
823 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
824 		ptr += 4;
825 
826 	for (a = 0; a < num_ent; a++) {
827 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
828 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
829 
830 		/* Read a number of dwords in the entry */
831 		val = REG(ptr, 4);
832 		ptr += 4;
833 		ent_size = (val & PCIM_EA_ES);
834 
835 		for (b = 0; b < ent_size; b++) {
836 			dw[b] = REG(ptr, 4);
837 			ptr += 4;
838 		}
839 
840 		eae->eae_flags = val;
841 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
842 
843 		base = dw[0] & PCIM_EA_FIELD_MASK;
844 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
845 		b = 2;
846 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
847 			base |= (uint64_t)dw[b] << 32UL;
848 			b++;
849 		}
850 		if (((dw[1] & PCIM_EA_IS_64) != 0)
851 		    && (b < ent_size)) {
852 			max_offset |= (uint64_t)dw[b] << 32UL;
853 			b++;
854 		}
855 
856 		eae->eae_base = base;
857 		eae->eae_max_offset = max_offset;
858 
859 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
860 
861 		if (bootverbose) {
862 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
863 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
864 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
865 		}
866 	}
867 }
868 #undef REG
869 
870 static void
pci_read_cap(device_t pcib,pcicfgregs * cfg)871 pci_read_cap(device_t pcib, pcicfgregs *cfg)
872 {
873 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
874 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
875 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
876 	uint64_t addr;
877 #endif
878 	uint32_t val;
879 	int	ptr, nextptr, ptrptr;
880 
881 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
882 	case PCIM_HDRTYPE_NORMAL:
883 	case PCIM_HDRTYPE_BRIDGE:
884 		ptrptr = PCIR_CAP_PTR;
885 		break;
886 	case PCIM_HDRTYPE_CARDBUS:
887 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
888 		break;
889 	default:
890 		return;		/* no extended capabilities support */
891 	}
892 	nextptr = REG(ptrptr, 1);	/* sanity check? */
893 
894 	/*
895 	 * Read capability entries.
896 	 */
897 	while (nextptr != 0) {
898 		/* Sanity check */
899 		if (nextptr > 255) {
900 			printf("illegal PCI extended capability offset %d\n",
901 			    nextptr);
902 			return;
903 		}
904 		/* Find the next entry */
905 		ptr = nextptr;
906 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
907 
908 		/* Process this entry */
909 		switch (REG(ptr + PCICAP_ID, 1)) {
910 		case PCIY_PMG:		/* PCI power management */
911 			if (cfg->pp.pp_cap == 0) {
912 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
913 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
914 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
915 				if ((nextptr - ptr) > PCIR_POWER_DATA)
916 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
917 			}
918 			break;
919 		case PCIY_HT:		/* HyperTransport */
920 			/* Determine HT-specific capability type. */
921 			val = REG(ptr + PCIR_HT_COMMAND, 2);
922 
923 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
924 				cfg->ht.ht_slave = ptr;
925 
926 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
927 			switch (val & PCIM_HTCMD_CAP_MASK) {
928 			case PCIM_HTCAP_MSI_MAPPING:
929 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
930 					/* Sanity check the mapping window. */
931 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
932 					    4);
933 					addr <<= 32;
934 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
935 					    4);
936 					if (addr != MSI_INTEL_ADDR_BASE)
937 						device_printf(pcib,
938 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
939 						    cfg->domain, cfg->bus,
940 						    cfg->slot, cfg->func,
941 						    (long long)addr);
942 				} else
943 					addr = MSI_INTEL_ADDR_BASE;
944 
945 				cfg->ht.ht_msimap = ptr;
946 				cfg->ht.ht_msictrl = val;
947 				cfg->ht.ht_msiaddr = addr;
948 				break;
949 			}
950 #endif
951 			break;
952 		case PCIY_MSI:		/* PCI MSI */
953 			cfg->msi.msi_location = ptr;
954 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
955 			break;
956 		case PCIY_MSIX:		/* PCI MSI-X */
957 			cfg->msix.msix_location = ptr;
958 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
959 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
960 			cfg->msix.msix_table_bar = PCIR_BAR(val &
961 			    PCIM_MSIX_BIR_MASK);
962 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
963 			val = REG(ptr + PCIR_MSIX_PBA, 4);
964 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
965 			    PCIM_MSIX_BIR_MASK);
966 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
967 			break;
968 		case PCIY_VPD:		/* PCI Vital Product Data */
969 			cfg->vpd.vpd_reg = ptr;
970 			break;
971 		case PCIY_SUBVENDOR:
972 			/* Should always be true. */
973 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
974 			    PCIM_HDRTYPE_BRIDGE) {
975 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
976 				cfg->subvendor = val & 0xffff;
977 				cfg->subdevice = val >> 16;
978 			}
979 			break;
980 		case PCIY_PCIX:		/* PCI-X */
981 			/*
982 			 * Assume we have a PCI-X chipset if we have
983 			 * at least one PCI-PCI bridge with a PCI-X
984 			 * capability.  Note that some systems with
985 			 * PCI-express or HT chipsets might match on
986 			 * this check as well.
987 			 */
988 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
989 			    PCIM_HDRTYPE_BRIDGE)
990 				pcix_chipset = 1;
991 			cfg->pcix.pcix_location = ptr;
992 			break;
993 		case PCIY_EXPRESS:	/* PCI-express */
994 			/*
995 			 * Assume we have a PCI-express chipset if we have
996 			 * at least one PCI-express device.
997 			 */
998 			pcie_chipset = 1;
999 			cfg->pcie.pcie_location = ptr;
1000 			val = REG(ptr + PCIER_FLAGS, 2);
1001 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
1002 			break;
1003 		case PCIY_EA:		/* Enhanced Allocation */
1004 			cfg->ea.ea_location = ptr;
1005 			pci_ea_fill_info(pcib, cfg);
1006 			break;
1007 		default:
1008 			break;
1009 		}
1010 	}
1011 
1012 #if defined(__powerpc__)
1013 	/*
1014 	 * Enable the MSI mapping window for all HyperTransport
1015 	 * slaves.  PCI-PCI bridges have their windows enabled via
1016 	 * PCIB_MAP_MSI().
1017 	 */
1018 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1019 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1020 		device_printf(pcib,
1021 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1022 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1023 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1024 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1025 		     2);
1026 	}
1027 #endif
1028 /* REG and WREG use carry through to next functions */
1029 }
1030 
1031 /*
1032  * PCI Vital Product Data
1033  */
1034 
1035 #define	PCI_VPD_TIMEOUT		1000000
1036 
1037 static int
pci_read_vpd_reg(device_t pcib,pcicfgregs * cfg,int reg,uint32_t * data)1038 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1039 {
1040 	int count = PCI_VPD_TIMEOUT;
1041 
1042 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1043 
1044 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1045 
1046 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1047 		if (--count < 0)
1048 			return (ENXIO);
1049 		DELAY(1);	/* limit looping */
1050 	}
1051 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1052 
1053 	return (0);
1054 }
1055 
1056 #if 0
1057 static int
1058 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1059 {
1060 	int count = PCI_VPD_TIMEOUT;
1061 
1062 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1063 
1064 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1065 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1066 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1067 		if (--count < 0)
1068 			return (ENXIO);
1069 		DELAY(1);	/* limit looping */
1070 	}
1071 
1072 	return (0);
1073 }
1074 #endif
1075 
1076 #undef PCI_VPD_TIMEOUT
1077 
1078 struct vpd_readstate {
1079 	device_t	pcib;
1080 	pcicfgregs	*cfg;
1081 	uint32_t	val;
1082 	int		bytesinval;
1083 	int		off;
1084 	uint8_t		cksum;
1085 };
1086 
1087 /* return 0 and one byte in *data if no read error, -1 else */
1088 static int
vpd_nextbyte(struct vpd_readstate * vrs,uint8_t * data)1089 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1090 {
1091 	uint32_t reg;
1092 	uint8_t byte;
1093 
1094 	if (vrs->bytesinval == 0) {
1095 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1096 			return (-1);
1097 		vrs->val = le32toh(reg);
1098 		vrs->off += 4;
1099 		byte = vrs->val & 0xff;
1100 		vrs->bytesinval = 3;
1101 	} else {
1102 		vrs->val = vrs->val >> 8;
1103 		byte = vrs->val & 0xff;
1104 		vrs->bytesinval--;
1105 	}
1106 
1107 	vrs->cksum += byte;
1108 	*data = byte;
1109 	return (0);
1110 }
1111 
1112 /* return 0 on match, -1 and "unget" byte on no match */
1113 static int
vpd_expectbyte(struct vpd_readstate * vrs,uint8_t expected)1114 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1115 {
1116 	uint8_t data;
1117 
1118 	if (vpd_nextbyte(vrs, &data) != 0)
1119 		return (-1);
1120 
1121 	if (data == expected)
1122 		return (0);
1123 
1124 	vrs->cksum -= data;
1125 	vrs->val = (vrs->val << 8) + data;
1126 	vrs->bytesinval++;
1127 	return (-1);
1128 }
1129 
1130 /* return size if tag matches, -1 on no match, -2 on read error */
1131 static int
vpd_read_tag_size(struct vpd_readstate * vrs,uint8_t vpd_tag)1132 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1133 {
1134 	uint8_t byte1, byte2;
1135 
1136 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1137 		return (-1);
1138 
1139 	if ((vpd_tag & 0x80) == 0)
1140 		return (vpd_tag & 0x07);
1141 
1142 	if (vpd_nextbyte(vrs, &byte1) != 0)
1143 		return (-2);
1144 	if (vpd_nextbyte(vrs, &byte2) != 0)
1145 		return (-2);
1146 
1147 	return ((byte2 << 8) + byte1);
1148 }
1149 
1150 /* (re)allocate buffer in multiples of 8 elements */
1151 static void*
alloc_buffer(void * buffer,size_t element_size,int needed)1152 alloc_buffer(void* buffer, size_t element_size, int needed)
1153 {
1154 	int alloc, new_alloc;
1155 
1156 	alloc = roundup2(needed, 8);
1157 	new_alloc = roundup2(needed + 1, 8);
1158 	if (alloc != new_alloc) {
1159 		buffer = reallocf(buffer,
1160 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1161 	}
1162 
1163 	return (buffer);
1164 }
1165 
1166 /* read VPD keyword and return element size, return -1 on read error */
1167 static int
vpd_read_elem_head(struct vpd_readstate * vrs,char keyword[2])1168 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1169 {
1170 	uint8_t data;
1171 
1172 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1173 		return (-1);
1174 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1175 		return (-1);
1176 	if (vpd_nextbyte(vrs, &data) != 0)
1177 		return (-1);
1178 
1179 	return (data);
1180 }
1181 
1182 /* read VPD data element of given size into allocated buffer */
1183 static char *
vpd_read_value(struct vpd_readstate * vrs,int size)1184 vpd_read_value(struct vpd_readstate *vrs, int size)
1185 {
1186 	int i;
1187 	char char1;
1188 	char *value;
1189 
1190 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1191 	for (i = 0; i < size; i++) {
1192 		if (vpd_nextbyte(vrs, &char1) != 0) {
1193 			free(value, M_DEVBUF);
1194 			return (NULL);
1195 		}
1196 		value[i] = char1;
1197 	}
1198 	value[size] = '\0';
1199 
1200 	return (value);
1201 }
1202 
1203 /* read VPD into *keyword and *value, return length of data element */
1204 static int
vpd_read_elem_data(struct vpd_readstate * vrs,char keyword[2],char ** value,int maxlen)1205 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1206 {
1207 	int len;
1208 
1209 	len = vpd_read_elem_head(vrs, keyword);
1210 	if (len < 0 || len > maxlen)
1211 		return (-1);
1212 	*value = vpd_read_value(vrs, len);
1213 
1214 	return (len);
1215 }
1216 
1217 /* subtract all data following first byte from checksum of RV element */
1218 static void
vpd_fixup_cksum(struct vpd_readstate * vrs,char * rvstring,int len)1219 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1220 {
1221 	int i;
1222 	uint8_t fixup;
1223 
1224 	fixup = 0;
1225 	for (i = 1; i < len; i++)
1226 		fixup += rvstring[i];
1227 	vrs->cksum -= fixup;
1228 }
1229 
1230 /* fetch one read-only element and return size of heading + data */
1231 static int
next_vpd_ro_elem(struct vpd_readstate * vrs,int maxsize)1232 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1233 {
1234 	struct pcicfg_vpd *vpd;
1235 	pcicfgregs *cfg;
1236 	struct vpd_readonly *vpd_ros;
1237 	int len;
1238 
1239 	cfg = vrs->cfg;
1240 	vpd = &cfg->vpd;
1241 
1242 	if (maxsize < 3)
1243 		return (-1);
1244 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1245 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1246 	maxsize -= 3;
1247 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1248 	if (vpd_ros->value == NULL)
1249 		return (-1);
1250 	vpd_ros->len = len;
1251 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1252 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1253 		if (vrs->cksum != 0) {
1254 			pci_printf(cfg,
1255 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1256 			return (-1);
1257 		}
1258 	}
1259 	vpd->vpd_rocnt++;
1260 
1261 	return (len + 3);
1262 }
1263 
1264 /* fetch one writable element and return size of heading + data */
1265 static int
next_vpd_rw_elem(struct vpd_readstate * vrs,int maxsize)1266 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1267 {
1268 	struct pcicfg_vpd *vpd;
1269 	pcicfgregs *cfg;
1270 	struct vpd_write *vpd_w;
1271 	int len;
1272 
1273 	cfg = vrs->cfg;
1274 	vpd = &cfg->vpd;
1275 
1276 	if (maxsize < 3)
1277 		return (-1);
1278 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1279 	if (vpd->vpd_w == NULL) {
1280 		pci_printf(cfg, "out of memory");
1281 		return (-1);
1282 	}
1283 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1284 	maxsize -= 3;
1285 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1286 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1287 	if (vpd_w->value == NULL)
1288 		return (-1);
1289 	vpd_w->len = len;
1290 	vpd->vpd_wcnt++;
1291 
1292 	return (len + 3);
1293 }
1294 
1295 /* free all memory allocated for VPD data */
1296 static void
vpd_free(struct pcicfg_vpd * vpd)1297 vpd_free(struct pcicfg_vpd *vpd)
1298 {
1299 	int i;
1300 
1301 	free(vpd->vpd_ident, M_DEVBUF);
1302 	for (i = 0; i < vpd->vpd_rocnt; i++)
1303 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1304 	free(vpd->vpd_ros, M_DEVBUF);
1305 	vpd->vpd_rocnt = 0;
1306 	for (i = 0; i < vpd->vpd_wcnt; i++)
1307 		free(vpd->vpd_w[i].value, M_DEVBUF);
1308 	free(vpd->vpd_w, M_DEVBUF);
1309 	vpd->vpd_wcnt = 0;
1310 }
1311 
1312 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1313 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1314 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1315 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1316 
1317 static int
pci_parse_vpd(device_t pcib,pcicfgregs * cfg)1318 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1319 {
1320 	struct vpd_readstate vrs;
1321 	int cksumvalid;
1322 	int size, elem_size;
1323 
1324 	/* init vpd reader */
1325 	vrs.bytesinval = 0;
1326 	vrs.off = 0;
1327 	vrs.pcib = pcib;
1328 	vrs.cfg = cfg;
1329 	vrs.cksum = 0;
1330 
1331 	/* read VPD ident element - mandatory */
1332 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1333 	if (size <= 0) {
1334 		pci_printf(cfg, "no VPD ident found\n");
1335 		return (0);
1336 	}
1337 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1338 	if (cfg->vpd.vpd_ident == NULL) {
1339 		pci_printf(cfg, "error accessing VPD ident data\n");
1340 		return (0);
1341 	}
1342 
1343 	/* read VPD RO elements - mandatory */
1344 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1345 	if (size <= 0) {
1346 		pci_printf(cfg, "no read-only VPD data found\n");
1347 		return (0);
1348 	}
1349 	while (size > 0) {
1350 		elem_size = next_vpd_ro_elem(&vrs, size);
1351 		if (elem_size < 0) {
1352 			pci_printf(cfg, "error accessing read-only VPD data\n");
1353 			return (-1);
1354 		}
1355 		size -= elem_size;
1356 	}
1357 	cksumvalid = (vrs.cksum == 0);
1358 	if (!cksumvalid)
1359 		return (-1);
1360 
1361 	/* read VPD RW elements - optional */
1362 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1363 	if (size == -2)
1364 		return (-1);
1365 	while (size > 0) {
1366 		elem_size = next_vpd_rw_elem(&vrs, size);
1367 		if (elem_size < 0) {
1368 			pci_printf(cfg, "error accessing writeable VPD data\n");
1369 			return (-1);
1370 		}
1371 		size -= elem_size;
1372 	}
1373 
1374 	/* read empty END tag - mandatory */
1375 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1376 	if (size != 0) {
1377 		pci_printf(cfg, "No valid VPD end tag found\n");
1378 	}
1379 	return (0);
1380 }
1381 
1382 static void
pci_read_vpd(device_t pcib,pcicfgregs * cfg)1383 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1384 {
1385 	int status;
1386 
1387 	status = pci_parse_vpd(pcib, cfg);
1388 	if (status < 0)
1389 		vpd_free(&cfg->vpd);
1390 	cfg->vpd.vpd_cached = 1;
1391 #undef REG
1392 #undef WREG
1393 }
1394 
1395 int
pci_get_vpd_ident_method(device_t dev,device_t child,const char ** identptr)1396 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1397 {
1398 	struct pci_devinfo *dinfo = device_get_ivars(child);
1399 	pcicfgregs *cfg = &dinfo->cfg;
1400 
1401 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1402 		pci_read_vpd(device_get_parent(dev), cfg);
1403 
1404 	*identptr = cfg->vpd.vpd_ident;
1405 
1406 	if (*identptr == NULL)
1407 		return (ENXIO);
1408 
1409 	return (0);
1410 }
1411 
1412 int
pci_get_vpd_readonly_method(device_t dev,device_t child,const char * kw,const char ** vptr)1413 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1414 	const char **vptr)
1415 {
1416 	struct pci_devinfo *dinfo = device_get_ivars(child);
1417 	pcicfgregs *cfg = &dinfo->cfg;
1418 	int i;
1419 
1420 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1421 		pci_read_vpd(device_get_parent(dev), cfg);
1422 
1423 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1424 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1425 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1426 			*vptr = cfg->vpd.vpd_ros[i].value;
1427 			return (0);
1428 		}
1429 
1430 	*vptr = NULL;
1431 	return (ENXIO);
1432 }
1433 
1434 struct pcicfg_vpd *
pci_fetch_vpd_list(device_t dev)1435 pci_fetch_vpd_list(device_t dev)
1436 {
1437 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1438 	pcicfgregs *cfg = &dinfo->cfg;
1439 
1440 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1441 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1442 	return (&cfg->vpd);
1443 }
1444 
1445 /*
1446  * Find the requested HyperTransport capability and return the offset
1447  * in configuration space via the pointer provided.  The function
1448  * returns 0 on success and an error code otherwise.
1449  */
1450 int
pci_find_htcap_method(device_t dev,device_t child,int capability,int * capreg)1451 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1452 {
1453 	int ptr, error;
1454 	uint16_t val;
1455 
1456 	error = pci_find_cap(child, PCIY_HT, &ptr);
1457 	if (error)
1458 		return (error);
1459 
1460 	/*
1461 	 * Traverse the capabilities list checking each HT capability
1462 	 * to see if it matches the requested HT capability.
1463 	 */
1464 	for (;;) {
1465 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1466 		if (capability == PCIM_HTCAP_SLAVE ||
1467 		    capability == PCIM_HTCAP_HOST)
1468 			val &= 0xe000;
1469 		else
1470 			val &= PCIM_HTCMD_CAP_MASK;
1471 		if (val == capability) {
1472 			if (capreg != NULL)
1473 				*capreg = ptr;
1474 			return (0);
1475 		}
1476 
1477 		/* Skip to the next HT capability. */
1478 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1479 			break;
1480 	}
1481 
1482 	return (ENOENT);
1483 }
1484 
1485 /*
1486  * Find the next requested HyperTransport capability after start and return
1487  * the offset in configuration space via the pointer provided.  The function
1488  * returns 0 on success and an error code otherwise.
1489  */
1490 int
pci_find_next_htcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1491 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1492     int start, int *capreg)
1493 {
1494 	int ptr;
1495 	uint16_t val;
1496 
1497 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1498 	    ("start capability is not HyperTransport capability"));
1499 	ptr = start;
1500 
1501 	/*
1502 	 * Traverse the capabilities list checking each HT capability
1503 	 * to see if it matches the requested HT capability.
1504 	 */
1505 	for (;;) {
1506 		/* Skip to the next HT capability. */
1507 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1508 			break;
1509 
1510 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1511 		if (capability == PCIM_HTCAP_SLAVE ||
1512 		    capability == PCIM_HTCAP_HOST)
1513 			val &= 0xe000;
1514 		else
1515 			val &= PCIM_HTCMD_CAP_MASK;
1516 		if (val == capability) {
1517 			if (capreg != NULL)
1518 				*capreg = ptr;
1519 			return (0);
1520 		}
1521 	}
1522 
1523 	return (ENOENT);
1524 }
1525 
1526 /*
1527  * Find the requested capability and return the offset in
1528  * configuration space via the pointer provided.  The function returns
1529  * 0 on success and an error code otherwise.
1530  */
1531 int
pci_find_cap_method(device_t dev,device_t child,int capability,int * capreg)1532 pci_find_cap_method(device_t dev, device_t child, int capability,
1533     int *capreg)
1534 {
1535 	struct pci_devinfo *dinfo = device_get_ivars(child);
1536 	pcicfgregs *cfg = &dinfo->cfg;
1537 	uint32_t status;
1538 	uint8_t ptr;
1539 	int cnt;
1540 
1541 	/*
1542 	 * Check the CAP_LIST bit of the PCI status register first.
1543 	 */
1544 	status = pci_read_config(child, PCIR_STATUS, 2);
1545 	if (!(status & PCIM_STATUS_CAPPRESENT))
1546 		return (ENXIO);
1547 
1548 	/*
1549 	 * Determine the start pointer of the capabilities list.
1550 	 */
1551 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1552 	case PCIM_HDRTYPE_NORMAL:
1553 	case PCIM_HDRTYPE_BRIDGE:
1554 		ptr = PCIR_CAP_PTR;
1555 		break;
1556 	case PCIM_HDRTYPE_CARDBUS:
1557 		ptr = PCIR_CAP_PTR_2;
1558 		break;
1559 	default:
1560 		/* XXX: panic? */
1561 		return (ENXIO);		/* no extended capabilities support */
1562 	}
1563 	ptr = pci_read_config(child, ptr, 1);
1564 
1565 	/*
1566 	 * Traverse the capabilities list.  Limit by total theoretical
1567 	 * maximum number of caps: capability needs at least id and
1568 	 * next registers, and any type X header cannot contain caps.
1569 	 */
1570 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1571 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1572 			if (capreg != NULL)
1573 				*capreg = ptr;
1574 			return (0);
1575 		}
1576 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1577 	}
1578 
1579 	return (ENOENT);
1580 }
1581 
1582 /*
1583  * Find the next requested capability after start and return the offset in
1584  * configuration space via the pointer provided.  The function returns
1585  * 0 on success and an error code otherwise.
1586  */
1587 int
pci_find_next_cap_method(device_t dev,device_t child,int capability,int start,int * capreg)1588 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1589     int start, int *capreg)
1590 {
1591 	uint8_t ptr;
1592 
1593 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1594 	    ("start capability is not expected capability"));
1595 
1596 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1597 	while (ptr != 0) {
1598 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1599 			if (capreg != NULL)
1600 				*capreg = ptr;
1601 			return (0);
1602 		}
1603 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1604 	}
1605 
1606 	return (ENOENT);
1607 }
1608 
1609 /*
1610  * Find the requested extended capability and return the offset in
1611  * configuration space via the pointer provided.  The function returns
1612  * 0 on success and an error code otherwise.
1613  */
1614 int
pci_find_extcap_method(device_t dev,device_t child,int capability,int * capreg)1615 pci_find_extcap_method(device_t dev, device_t child, int capability,
1616     int *capreg)
1617 {
1618 	struct pci_devinfo *dinfo = device_get_ivars(child);
1619 	pcicfgregs *cfg = &dinfo->cfg;
1620 	uint32_t ecap;
1621 	uint16_t ptr;
1622 
1623 	/* Only supported for PCI-express devices. */
1624 	if (cfg->pcie.pcie_location == 0)
1625 		return (ENXIO);
1626 
1627 	ptr = PCIR_EXTCAP;
1628 	ecap = pci_read_config(child, ptr, 4);
1629 	if (ecap == 0xffffffff || ecap == 0)
1630 		return (ENOENT);
1631 	for (;;) {
1632 		if (PCI_EXTCAP_ID(ecap) == capability) {
1633 			if (capreg != NULL)
1634 				*capreg = ptr;
1635 			return (0);
1636 		}
1637 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1638 		if (ptr == 0)
1639 			break;
1640 		ecap = pci_read_config(child, ptr, 4);
1641 	}
1642 
1643 	return (ENOENT);
1644 }
1645 
1646 /*
1647  * Find the next requested extended capability after start and return the
1648  * offset in configuration space via the pointer provided.  The function
1649  * returns 0 on success and an error code otherwise.
1650  */
1651 int
pci_find_next_extcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1652 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1653     int start, int *capreg)
1654 {
1655 	struct pci_devinfo *dinfo = device_get_ivars(child);
1656 	pcicfgregs *cfg = &dinfo->cfg;
1657 	uint32_t ecap;
1658 	uint16_t ptr;
1659 
1660 	/* Only supported for PCI-express devices. */
1661 	if (cfg->pcie.pcie_location == 0)
1662 		return (ENXIO);
1663 
1664 	ecap = pci_read_config(child, start, 4);
1665 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1666 	    ("start extended capability is not expected capability"));
1667 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1668 	while (ptr != 0) {
1669 		ecap = pci_read_config(child, ptr, 4);
1670 		if (PCI_EXTCAP_ID(ecap) == capability) {
1671 			if (capreg != NULL)
1672 				*capreg = ptr;
1673 			return (0);
1674 		}
1675 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1676 	}
1677 
1678 	return (ENOENT);
1679 }
1680 
1681 /*
1682  * Support for MSI-X message interrupts.
1683  */
1684 static void
pci_write_msix_entry(device_t dev,u_int index,uint64_t address,uint32_t data)1685 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1686 {
1687 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1688 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1689 	uint32_t offset;
1690 
1691 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1692 	offset = msix->msix_table_offset + index * 16;
1693 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1694 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1695 	bus_write_4(msix->msix_table_res, offset + 8, data);
1696 }
1697 
1698 void
pci_enable_msix_method(device_t dev,device_t child,u_int index,uint64_t address,uint32_t data)1699 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1700     uint64_t address, uint32_t data)
1701 {
1702 
1703 	if (pci_msix_rewrite_table) {
1704 		struct pci_devinfo *dinfo = device_get_ivars(child);
1705 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1706 
1707 		/*
1708 		 * Some VM hosts require MSIX to be disabled in the
1709 		 * control register before updating the MSIX table
1710 		 * entries are allowed. It is not enough to only
1711 		 * disable MSIX while updating a single entry. MSIX
1712 		 * must be disabled while updating all entries in the
1713 		 * table.
1714 		 */
1715 		pci_write_config(child,
1716 		    msix->msix_location + PCIR_MSIX_CTRL,
1717 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1718 		pci_resume_msix(child);
1719 	} else
1720 		pci_write_msix_entry(child, index, address, data);
1721 
1722 	/* Enable MSI -> HT mapping. */
1723 	pci_ht_map_msi(child, address);
1724 }
1725 
1726 void
pci_mask_msix(device_t dev,u_int index)1727 pci_mask_msix(device_t dev, u_int index)
1728 {
1729 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1730 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1731 	uint32_t offset, val;
1732 
1733 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1734 	offset = msix->msix_table_offset + index * 16 + 12;
1735 	val = bus_read_4(msix->msix_table_res, offset);
1736 	val |= PCIM_MSIX_VCTRL_MASK;
1737 
1738 	/*
1739 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1740 	 * register, so always write the new value.
1741 	 */
1742 	bus_write_4(msix->msix_table_res, offset, val);
1743 }
1744 
1745 void
pci_unmask_msix(device_t dev,u_int index)1746 pci_unmask_msix(device_t dev, u_int index)
1747 {
1748 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1749 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1750 	uint32_t offset, val;
1751 
1752 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1753 	offset = msix->msix_table_offset + index * 16 + 12;
1754 	val = bus_read_4(msix->msix_table_res, offset);
1755 	val &= ~PCIM_MSIX_VCTRL_MASK;
1756 
1757 	/*
1758 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1759 	 * register, so always write the new value.
1760 	 */
1761 	bus_write_4(msix->msix_table_res, offset, val);
1762 }
1763 
1764 int
pci_pending_msix(device_t dev,u_int index)1765 pci_pending_msix(device_t dev, u_int index)
1766 {
1767 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1768 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1769 	uint32_t offset, bit;
1770 
1771 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1772 	offset = msix->msix_pba_offset + (index / 32) * 4;
1773 	bit = 1 << index % 32;
1774 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1775 }
1776 
1777 /*
1778  * Restore MSI-X registers and table during resume.  If MSI-X is
1779  * enabled then walk the virtual table to restore the actual MSI-X
1780  * table.
1781  */
1782 static void
pci_resume_msix(device_t dev)1783 pci_resume_msix(device_t dev)
1784 {
1785 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1786 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1787 	struct msix_table_entry *mte;
1788 	struct msix_vector *mv;
1789 	u_int i, msgnum;
1790 
1791 	if (msix->msix_alloc > 0) {
1792 		msgnum = PCI_MSIX_MSGNUM(msix->msix_ctrl);
1793 
1794 		/* First, mask all vectors. */
1795 		for (i = 0; i < msgnum; i++)
1796 			pci_mask_msix(dev, i);
1797 
1798 		/* Second, program any messages with at least one handler. */
1799 		for (i = 0; i < msix->msix_table_len; i++) {
1800 			mte = &msix->msix_table[i];
1801 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1802 				continue;
1803 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1804 			pci_write_msix_entry(dev, i, mv->mv_address,
1805 			    mv->mv_data);
1806 			pci_unmask_msix(dev, i);
1807 		}
1808 	}
1809 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1810 	    msix->msix_ctrl, 2);
1811 }
1812 
1813 /*
1814  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1815  * returned in *count.  After this function returns, each message will be
1816  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1817  */
1818 int
pci_alloc_msix_method(device_t dev,device_t child,int * count)1819 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1820 {
1821 	struct pci_devinfo *dinfo = device_get_ivars(child);
1822 	pcicfgregs *cfg = &dinfo->cfg;
1823 	struct resource_list_entry *rle;
1824 	u_int actual, i, max;
1825 	int error, irq;
1826 	uint16_t ctrl, msgnum;
1827 
1828 	/* Don't let count == 0 get us into trouble. */
1829 	if (*count < 1)
1830 		return (EINVAL);
1831 
1832 	/* If rid 0 is allocated, then fail. */
1833 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1834 	if (rle != NULL && rle->res != NULL)
1835 		return (ENXIO);
1836 
1837 	/* Already have allocated messages? */
1838 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1839 		return (ENXIO);
1840 
1841 	/* If MSI-X is blacklisted for this system, fail. */
1842 	if (pci_msix_blacklisted())
1843 		return (ENXIO);
1844 
1845 	/* MSI-X capability present? */
1846 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1847 		return (ENODEV);
1848 
1849 	/* Make sure the appropriate BARs are mapped. */
1850 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1851 	    cfg->msix.msix_table_bar);
1852 	if (rle == NULL || rle->res == NULL ||
1853 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1854 		return (ENXIO);
1855 	cfg->msix.msix_table_res = rle->res;
1856 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1857 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1858 		    cfg->msix.msix_pba_bar);
1859 		if (rle == NULL || rle->res == NULL ||
1860 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1861 			return (ENXIO);
1862 	}
1863 	cfg->msix.msix_pba_res = rle->res;
1864 
1865 	ctrl = pci_read_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1866 	    2);
1867 	msgnum = PCI_MSIX_MSGNUM(ctrl);
1868 	if (bootverbose)
1869 		device_printf(child,
1870 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1871 		    *count, msgnum);
1872 	max = min(*count, msgnum);
1873 	for (i = 0; i < max; i++) {
1874 		/* Allocate a message. */
1875 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1876 		if (error) {
1877 			if (i == 0)
1878 				return (error);
1879 			break;
1880 		}
1881 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1882 		    irq, 1);
1883 	}
1884 	actual = i;
1885 
1886 	if (bootverbose) {
1887 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1888 		if (actual == 1)
1889 			device_printf(child, "using IRQ %ju for MSI-X\n",
1890 			    rle->start);
1891 		else {
1892 			bool run;
1893 
1894 			/*
1895 			 * Be fancy and try to print contiguous runs of
1896 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1897 			 * 'run' is true if we are in a range.
1898 			 */
1899 			device_printf(child, "using IRQs %ju", rle->start);
1900 			irq = rle->start;
1901 			run = false;
1902 			for (i = 1; i < actual; i++) {
1903 				rle = resource_list_find(&dinfo->resources,
1904 				    SYS_RES_IRQ, i + 1);
1905 
1906 				/* Still in a run? */
1907 				if (rle->start == irq + 1) {
1908 					run = true;
1909 					irq++;
1910 					continue;
1911 				}
1912 
1913 				/* Finish previous range. */
1914 				if (run) {
1915 					printf("-%d", irq);
1916 					run = false;
1917 				}
1918 
1919 				/* Start new range. */
1920 				printf(",%ju", rle->start);
1921 				irq = rle->start;
1922 			}
1923 
1924 			/* Unfinished range? */
1925 			if (run)
1926 				printf("-%d", irq);
1927 			printf(" for MSI-X\n");
1928 		}
1929 	}
1930 
1931 	/* Mask all vectors. */
1932 	for (i = 0; i < msgnum; i++)
1933 		pci_mask_msix(child, i);
1934 
1935 	/* Allocate and initialize vector data and virtual table. */
1936 	cfg->msix.msix_vectors = mallocarray(actual, sizeof(struct msix_vector),
1937 	    M_DEVBUF, M_WAITOK | M_ZERO);
1938 	cfg->msix.msix_table = mallocarray(actual,
1939 	    sizeof(struct msix_table_entry), M_DEVBUF, M_WAITOK | M_ZERO);
1940 	for (i = 0; i < actual; i++) {
1941 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1942 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1943 		cfg->msix.msix_table[i].mte_vector = i + 1;
1944 	}
1945 
1946 	/* Update control register to enable MSI-X. */
1947 	ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1948 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1949 	    ctrl, 2);
1950 	cfg->msix.msix_ctrl = ctrl;
1951 
1952 	/* Update counts of alloc'd messages. */
1953 	cfg->msix.msix_alloc = actual;
1954 	cfg->msix.msix_table_len = actual;
1955 	*count = actual;
1956 	return (0);
1957 }
1958 
1959 /*
1960  * By default, pci_alloc_msix() will assign the allocated IRQ
1961  * resources consecutively to the first N messages in the MSI-X table.
1962  * However, device drivers may want to use different layouts if they
1963  * either receive fewer messages than they asked for, or they wish to
1964  * populate the MSI-X table sparsely.  This method allows the driver
1965  * to specify what layout it wants.  It must be called after a
1966  * successful pci_alloc_msix() but before any of the associated
1967  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1968  *
1969  * The 'vectors' array contains 'count' message vectors.  The array
1970  * maps directly to the MSI-X table in that index 0 in the array
1971  * specifies the vector for the first message in the MSI-X table, etc.
1972  * The vector value in each array index can either be 0 to indicate
1973  * that no vector should be assigned to a message slot, or it can be a
1974  * number from 1 to N (where N is the count returned from a
1975  * succcessful call to pci_alloc_msix()) to indicate which message
1976  * vector (IRQ) to be used for the corresponding message.
1977  *
1978  * On successful return, each message with a non-zero vector will have
1979  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1980  * 1.  Additionally, if any of the IRQs allocated via the previous
1981  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1982  * will be freed back to the system automatically.
1983  *
1984  * For example, suppose a driver has a MSI-X table with 6 messages and
1985  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1986  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1987  * C.  After the call to pci_alloc_msix(), the device will be setup to
1988  * have an MSI-X table of ABC--- (where - means no vector assigned).
1989  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1990  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1991  * be freed back to the system.  This device will also have valid
1992  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1993  *
1994  * In any case, the SYS_RES_IRQ rid X will always map to the message
1995  * at MSI-X table index X - 1 and will only be valid if a vector is
1996  * assigned to that table entry.
1997  */
1998 int
pci_remap_msix_method(device_t dev,device_t child,int count,const u_int * vectors)1999 pci_remap_msix_method(device_t dev, device_t child, int count,
2000     const u_int *vectors)
2001 {
2002 	struct pci_devinfo *dinfo = device_get_ivars(child);
2003 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2004 	struct resource_list_entry *rle;
2005 	u_int i, irq, j;
2006 	bool *used;
2007 
2008 	/*
2009 	 * Have to have at least one message in the table but the
2010 	 * table can't be bigger than the actual MSI-X table in the
2011 	 * device.
2012 	 */
2013 	if (count < 1 || count > PCI_MSIX_MSGNUM(msix->msix_ctrl))
2014 		return (EINVAL);
2015 
2016 	/* Sanity check the vectors. */
2017 	for (i = 0; i < count; i++)
2018 		if (vectors[i] > msix->msix_alloc)
2019 			return (EINVAL);
2020 
2021 	/*
2022 	 * Make sure there aren't any holes in the vectors to be used.
2023 	 * It's a big pain to support it, and it doesn't really make
2024 	 * sense anyway.  Also, at least one vector must be used.
2025 	 */
2026 	used = mallocarray(msix->msix_alloc, sizeof(*used), M_DEVBUF, M_WAITOK |
2027 	    M_ZERO);
2028 	for (i = 0; i < count; i++)
2029 		if (vectors[i] != 0)
2030 			used[vectors[i] - 1] = true;
2031 	for (i = 0; i < msix->msix_alloc - 1; i++)
2032 		if (!used[i] && used[i + 1]) {
2033 			free(used, M_DEVBUF);
2034 			return (EINVAL);
2035 		}
2036 	if (!used[0]) {
2037 		free(used, M_DEVBUF);
2038 		return (EINVAL);
2039 	}
2040 
2041 	/* Make sure none of the resources are allocated. */
2042 	for (i = 0; i < msix->msix_table_len; i++) {
2043 		if (msix->msix_table[i].mte_vector == 0)
2044 			continue;
2045 		if (msix->msix_table[i].mte_handlers > 0) {
2046 			free(used, M_DEVBUF);
2047 			return (EBUSY);
2048 		}
2049 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2050 		KASSERT(rle != NULL, ("missing resource"));
2051 		if (rle->res != NULL) {
2052 			free(used, M_DEVBUF);
2053 			return (EBUSY);
2054 		}
2055 	}
2056 
2057 	/* Free the existing resource list entries. */
2058 	for (i = 0; i < msix->msix_table_len; i++) {
2059 		if (msix->msix_table[i].mte_vector == 0)
2060 			continue;
2061 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2062 	}
2063 
2064 	/*
2065 	 * Build the new virtual table keeping track of which vectors are
2066 	 * used.
2067 	 */
2068 	free(msix->msix_table, M_DEVBUF);
2069 	msix->msix_table = mallocarray(count, sizeof(struct msix_table_entry),
2070 	    M_DEVBUF, M_WAITOK | M_ZERO);
2071 	for (i = 0; i < count; i++)
2072 		msix->msix_table[i].mte_vector = vectors[i];
2073 	msix->msix_table_len = count;
2074 
2075 	/* Free any unused IRQs and resize the vectors array if necessary. */
2076 	j = msix->msix_alloc - 1;
2077 	if (!used[j]) {
2078 		struct msix_vector *vec;
2079 
2080 		while (!used[j]) {
2081 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2082 			    msix->msix_vectors[j].mv_irq);
2083 			j--;
2084 		}
2085 		vec = mallocarray(j + 1, sizeof(struct msix_vector), M_DEVBUF,
2086 		    M_WAITOK);
2087 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2088 		    (j + 1));
2089 		free(msix->msix_vectors, M_DEVBUF);
2090 		msix->msix_vectors = vec;
2091 		msix->msix_alloc = j + 1;
2092 	}
2093 	free(used, M_DEVBUF);
2094 
2095 	/* Map the IRQs onto the rids. */
2096 	for (i = 0; i < count; i++) {
2097 		if (vectors[i] == 0)
2098 			continue;
2099 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2100 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2101 		    irq, 1);
2102 	}
2103 
2104 	if (bootverbose) {
2105 		device_printf(child, "Remapped MSI-X IRQs as: ");
2106 		for (i = 0; i < count; i++) {
2107 			if (i != 0)
2108 				printf(", ");
2109 			if (vectors[i] == 0)
2110 				printf("---");
2111 			else
2112 				printf("%d",
2113 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2114 		}
2115 		printf("\n");
2116 	}
2117 
2118 	return (0);
2119 }
2120 
2121 static int
pci_release_msix(device_t dev,device_t child)2122 pci_release_msix(device_t dev, device_t child)
2123 {
2124 	struct pci_devinfo *dinfo = device_get_ivars(child);
2125 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2126 	struct resource_list_entry *rle;
2127 	u_int i;
2128 
2129 	/* Do we have any messages to release? */
2130 	if (msix->msix_alloc == 0)
2131 		return (ENODEV);
2132 
2133 	/* Make sure none of the resources are allocated. */
2134 	for (i = 0; i < msix->msix_table_len; i++) {
2135 		if (msix->msix_table[i].mte_vector == 0)
2136 			continue;
2137 		if (msix->msix_table[i].mte_handlers > 0)
2138 			return (EBUSY);
2139 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2140 		KASSERT(rle != NULL, ("missing resource"));
2141 		if (rle->res != NULL)
2142 			return (EBUSY);
2143 	}
2144 
2145 	/* Update control register to disable MSI-X. */
2146 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2147 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2148 	    msix->msix_ctrl, 2);
2149 
2150 	/* Free the resource list entries. */
2151 	for (i = 0; i < msix->msix_table_len; i++) {
2152 		if (msix->msix_table[i].mte_vector == 0)
2153 			continue;
2154 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2155 	}
2156 	free(msix->msix_table, M_DEVBUF);
2157 	msix->msix_table_len = 0;
2158 
2159 	/* Release the IRQs. */
2160 	for (i = 0; i < msix->msix_alloc; i++)
2161 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2162 		    msix->msix_vectors[i].mv_irq);
2163 	free(msix->msix_vectors, M_DEVBUF);
2164 	msix->msix_alloc = 0;
2165 	return (0);
2166 }
2167 
2168 /*
2169  * Return the max supported MSI-X messages this device supports.
2170  * Basically, assuming the MD code can alloc messages, this function
2171  * should return the maximum value that pci_alloc_msix() can return.
2172  * Thus, it is subject to the tunables, etc.
2173  */
2174 int
pci_msix_count_method(device_t dev,device_t child)2175 pci_msix_count_method(device_t dev, device_t child)
2176 {
2177 	struct pci_devinfo *dinfo = device_get_ivars(child);
2178 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2179 	uint16_t ctrl;
2180 
2181 	if (pci_do_msix && msix->msix_location != 0) {
2182 		ctrl = pci_read_config(child, msix->msix_location +
2183 		    PCIR_MSI_CTRL, 2);
2184 		return (PCI_MSIX_MSGNUM(ctrl));
2185 	}
2186 	return (0);
2187 }
2188 
2189 int
pci_msix_pba_bar_method(device_t dev,device_t child)2190 pci_msix_pba_bar_method(device_t dev, device_t child)
2191 {
2192 	struct pci_devinfo *dinfo = device_get_ivars(child);
2193 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2194 
2195 	if (pci_do_msix && msix->msix_location != 0)
2196 		return (msix->msix_pba_bar);
2197 	return (-1);
2198 }
2199 
2200 int
pci_msix_table_bar_method(device_t dev,device_t child)2201 pci_msix_table_bar_method(device_t dev, device_t child)
2202 {
2203 	struct pci_devinfo *dinfo = device_get_ivars(child);
2204 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2205 
2206 	if (pci_do_msix && msix->msix_location != 0)
2207 		return (msix->msix_table_bar);
2208 	return (-1);
2209 }
2210 
2211 /*
2212  * HyperTransport MSI mapping control
2213  */
2214 void
pci_ht_map_msi(device_t dev,uint64_t addr)2215 pci_ht_map_msi(device_t dev, uint64_t addr)
2216 {
2217 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2218 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2219 
2220 	if (!ht->ht_msimap)
2221 		return;
2222 
2223 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2224 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2225 		/* Enable MSI -> HT mapping. */
2226 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2227 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2228 		    ht->ht_msictrl, 2);
2229 	}
2230 
2231 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2232 		/* Disable MSI -> HT mapping. */
2233 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2234 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2235 		    ht->ht_msictrl, 2);
2236 	}
2237 }
2238 
2239 int
pci_get_relaxed_ordering_enabled(device_t dev)2240 pci_get_relaxed_ordering_enabled(device_t dev)
2241 {
2242 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2243 	int cap;
2244 	uint16_t val;
2245 
2246 	cap = dinfo->cfg.pcie.pcie_location;
2247 	if (cap == 0)
2248 		return (0);
2249 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2250 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2251 	return (val != 0);
2252 }
2253 
2254 int
pci_get_max_payload(device_t dev)2255 pci_get_max_payload(device_t dev)
2256 {
2257 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2258 	int cap;
2259 	uint16_t val;
2260 
2261 	cap = dinfo->cfg.pcie.pcie_location;
2262 	if (cap == 0)
2263 		return (0);
2264 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2265 	val &= PCIEM_CTL_MAX_PAYLOAD;
2266 	val >>= 5;
2267 	return (1 << (val + 7));
2268 }
2269 
2270 int
pci_get_max_read_req(device_t dev)2271 pci_get_max_read_req(device_t dev)
2272 {
2273 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2274 	int cap;
2275 	uint16_t val;
2276 
2277 	cap = dinfo->cfg.pcie.pcie_location;
2278 	if (cap == 0)
2279 		return (0);
2280 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2281 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2282 	val >>= 12;
2283 	return (1 << (val + 7));
2284 }
2285 
2286 int
pci_set_max_read_req(device_t dev,int size)2287 pci_set_max_read_req(device_t dev, int size)
2288 {
2289 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2290 	int cap;
2291 	uint16_t val;
2292 
2293 	cap = dinfo->cfg.pcie.pcie_location;
2294 	if (cap == 0)
2295 		return (0);
2296 	if (size < 128)
2297 		size = 128;
2298 	if (size > 4096)
2299 		size = 4096;
2300 	size = (1 << (fls(size) - 1));
2301 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2302 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2303 	val |= (fls(size) - 8) << 12;
2304 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2305 	return (size);
2306 }
2307 
2308 uint32_t
pcie_read_config(device_t dev,int reg,int width)2309 pcie_read_config(device_t dev, int reg, int width)
2310 {
2311 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2312 	int cap;
2313 
2314 	cap = dinfo->cfg.pcie.pcie_location;
2315 	if (cap == 0) {
2316 		if (width == 2)
2317 			return (0xffff);
2318 		return (0xffffffff);
2319 	}
2320 
2321 	return (pci_read_config(dev, cap + reg, width));
2322 }
2323 
2324 void
pcie_write_config(device_t dev,int reg,uint32_t value,int width)2325 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2326 {
2327 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2328 	int cap;
2329 
2330 	cap = dinfo->cfg.pcie.pcie_location;
2331 	if (cap == 0)
2332 		return;
2333 	pci_write_config(dev, cap + reg, value, width);
2334 }
2335 
2336 /*
2337  * Adjusts a PCI-e capability register by clearing the bits in mask
2338  * and setting the bits in (value & mask).  Bits not set in mask are
2339  * not adjusted.
2340  *
2341  * Returns the old value on success or all ones on failure.
2342  */
2343 uint32_t
pcie_adjust_config(device_t dev,int reg,uint32_t mask,uint32_t value,int width)2344 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2345     int width)
2346 {
2347 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2348 	uint32_t old, new;
2349 	int cap;
2350 
2351 	cap = dinfo->cfg.pcie.pcie_location;
2352 	if (cap == 0) {
2353 		if (width == 2)
2354 			return (0xffff);
2355 		return (0xffffffff);
2356 	}
2357 
2358 	old = pci_read_config(dev, cap + reg, width);
2359 	new = old & ~mask;
2360 	new |= (value & mask);
2361 	pci_write_config(dev, cap + reg, new, width);
2362 	return (old);
2363 }
2364 
2365 /*
2366  * Support for MSI message signalled interrupts.
2367  */
2368 void
pci_enable_msi_method(device_t dev,device_t child,uint64_t address,uint16_t data)2369 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2370     uint16_t data)
2371 {
2372 	struct pci_devinfo *dinfo = device_get_ivars(child);
2373 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2374 
2375 	/* Write data and address values. */
2376 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2377 	    address & 0xffffffff, 4);
2378 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2379 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2380 		    address >> 32, 4);
2381 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2382 		    data, 2);
2383 	} else
2384 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2385 		    2);
2386 
2387 	/* Enable MSI in the control register. */
2388 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2389 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2390 	    msi->msi_ctrl, 2);
2391 
2392 	/* Enable MSI -> HT mapping. */
2393 	pci_ht_map_msi(child, address);
2394 }
2395 
2396 void
pci_disable_msi_method(device_t dev,device_t child)2397 pci_disable_msi_method(device_t dev, device_t child)
2398 {
2399 	struct pci_devinfo *dinfo = device_get_ivars(child);
2400 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2401 
2402 	/* Disable MSI -> HT mapping. */
2403 	pci_ht_map_msi(child, 0);
2404 
2405 	/* Disable MSI in the control register. */
2406 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2407 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2408 	    msi->msi_ctrl, 2);
2409 }
2410 
2411 /*
2412  * Restore MSI registers during resume.  If MSI is enabled then
2413  * restore the data and address registers in addition to the control
2414  * register.
2415  */
2416 static void
pci_resume_msi(device_t dev)2417 pci_resume_msi(device_t dev)
2418 {
2419 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2420 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2421 	uint64_t address;
2422 	uint16_t data;
2423 
2424 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2425 		address = msi->msi_addr;
2426 		data = msi->msi_data;
2427 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2428 		    address & 0xffffffff, 4);
2429 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2430 			pci_write_config(dev, msi->msi_location +
2431 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2432 			pci_write_config(dev, msi->msi_location +
2433 			    PCIR_MSI_DATA_64BIT, data, 2);
2434 		} else
2435 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2436 			    data, 2);
2437 	}
2438 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2439 	    2);
2440 }
2441 
2442 static int
pci_remap_intr_method(device_t bus,device_t dev,u_int irq)2443 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2444 {
2445 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2446 	pcicfgregs *cfg = &dinfo->cfg;
2447 	struct resource_list_entry *rle;
2448 	struct msix_table_entry *mte;
2449 	struct msix_vector *mv;
2450 	uint64_t addr;
2451 	uint32_t data;
2452 	u_int i, j;
2453 	int error;
2454 
2455 	/*
2456 	 * Handle MSI first.  We try to find this IRQ among our list
2457 	 * of MSI IRQs.  If we find it, we request updated address and
2458 	 * data registers and apply the results.
2459 	 */
2460 	if (cfg->msi.msi_alloc > 0) {
2461 		/* If we don't have any active handlers, nothing to do. */
2462 		if (cfg->msi.msi_handlers == 0)
2463 			return (0);
2464 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2465 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2466 			    i + 1);
2467 			if (rle->start == irq) {
2468 				error = PCIB_MAP_MSI(device_get_parent(bus),
2469 				    dev, irq, &addr, &data);
2470 				if (error)
2471 					return (error);
2472 				pci_disable_msi(dev);
2473 				dinfo->cfg.msi.msi_addr = addr;
2474 				dinfo->cfg.msi.msi_data = data;
2475 				pci_enable_msi(dev, addr, data);
2476 				return (0);
2477 			}
2478 		}
2479 		return (ENOENT);
2480 	}
2481 
2482 	/*
2483 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2484 	 * we request the updated mapping info.  If that works, we go
2485 	 * through all the slots that use this IRQ and update them.
2486 	 */
2487 	if (cfg->msix.msix_alloc > 0) {
2488 		bool found = false;
2489 
2490 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2491 			mv = &cfg->msix.msix_vectors[i];
2492 			if (mv->mv_irq == irq) {
2493 				error = PCIB_MAP_MSI(device_get_parent(bus),
2494 				    dev, irq, &addr, &data);
2495 				if (error)
2496 					return (error);
2497 				mv->mv_address = addr;
2498 				mv->mv_data = data;
2499 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2500 					mte = &cfg->msix.msix_table[j];
2501 					if (mte->mte_vector != i + 1)
2502 						continue;
2503 					if (mte->mte_handlers == 0)
2504 						continue;
2505 					pci_mask_msix(dev, j);
2506 					pci_enable_msix(dev, j, addr, data);
2507 					pci_unmask_msix(dev, j);
2508 				}
2509 				found = true;
2510 			}
2511 		}
2512 		return (found ? 0 : ENOENT);
2513 	}
2514 
2515 	return (ENOENT);
2516 }
2517 
2518 /*
2519  * Returns true if the specified device is blacklisted because MSI
2520  * doesn't work.
2521  */
2522 int
pci_msi_device_blacklisted(device_t dev)2523 pci_msi_device_blacklisted(device_t dev)
2524 {
2525 
2526 	if (!pci_honor_msi_blacklist)
2527 		return (0);
2528 
2529 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2530 }
2531 
2532 /*
2533  * Determine if MSI is blacklisted globally on this system.  Currently,
2534  * we just check for blacklisted chipsets as represented by the
2535  * host-PCI bridge at device 0:0:0.  In the future, it may become
2536  * necessary to check other system attributes, such as the kenv values
2537  * that give the motherboard manufacturer and model number.
2538  */
2539 static int
pci_msi_blacklisted(void)2540 pci_msi_blacklisted(void)
2541 {
2542 	device_t dev;
2543 
2544 	if (!pci_honor_msi_blacklist)
2545 		return (0);
2546 
2547 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2548 	if (!(pcie_chipset || pcix_chipset)) {
2549 		if (vm_guest != VM_GUEST_NO) {
2550 			/*
2551 			 * Whitelist older chipsets in virtual
2552 			 * machines known to support MSI.
2553 			 */
2554 			dev = pci_find_bsf(0, 0, 0);
2555 			if (dev != NULL)
2556 				return (!pci_has_quirk(pci_get_devid(dev),
2557 					PCI_QUIRK_ENABLE_MSI_VM));
2558 		}
2559 		return (1);
2560 	}
2561 
2562 	dev = pci_find_bsf(0, 0, 0);
2563 	if (dev != NULL)
2564 		return (pci_msi_device_blacklisted(dev));
2565 	return (0);
2566 }
2567 
2568 /*
2569  * Returns true if the specified device is blacklisted because MSI-X
2570  * doesn't work.  Note that this assumes that if MSI doesn't work,
2571  * MSI-X doesn't either.
2572  */
2573 int
pci_msix_device_blacklisted(device_t dev)2574 pci_msix_device_blacklisted(device_t dev)
2575 {
2576 
2577 	if (!pci_honor_msi_blacklist)
2578 		return (0);
2579 
2580 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2581 		return (1);
2582 
2583 	return (pci_msi_device_blacklisted(dev));
2584 }
2585 
2586 /*
2587  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2588  * is blacklisted, assume that MSI-X is as well.  Check for additional
2589  * chipsets where MSI works but MSI-X does not.
2590  */
2591 static int
pci_msix_blacklisted(void)2592 pci_msix_blacklisted(void)
2593 {
2594 	device_t dev;
2595 
2596 	if (!pci_honor_msi_blacklist)
2597 		return (0);
2598 
2599 	dev = pci_find_bsf(0, 0, 0);
2600 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2601 	    PCI_QUIRK_DISABLE_MSIX))
2602 		return (1);
2603 
2604 	return (pci_msi_blacklisted());
2605 }
2606 
2607 /*
2608  * Attempt to allocate *count MSI messages.  The actual number allocated is
2609  * returned in *count.  After this function returns, each message will be
2610  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2611  */
2612 int
pci_alloc_msi_method(device_t dev,device_t child,int * count)2613 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2614 {
2615 	struct pci_devinfo *dinfo = device_get_ivars(child);
2616 	pcicfgregs *cfg = &dinfo->cfg;
2617 	struct resource_list_entry *rle;
2618 	u_int actual, i;
2619 	int error, irqs[32];
2620 	uint16_t ctrl, msgnum;
2621 
2622 	/* Don't let count == 0 get us into trouble. */
2623 	if (*count < 1)
2624 		return (EINVAL);
2625 
2626 	/* If rid 0 is allocated, then fail. */
2627 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2628 	if (rle != NULL && rle->res != NULL)
2629 		return (ENXIO);
2630 
2631 	/* Already have allocated messages? */
2632 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2633 		return (ENXIO);
2634 
2635 	/* If MSI is blacklisted for this system, fail. */
2636 	if (pci_msi_blacklisted())
2637 		return (ENXIO);
2638 
2639 	/* MSI capability present? */
2640 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2641 		return (ENODEV);
2642 
2643 	ctrl = pci_read_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, 2);
2644 	msgnum = PCI_MSI_MSGNUM(ctrl);
2645 	if (bootverbose)
2646 		device_printf(child,
2647 		    "attempting to allocate %d MSI vectors (%u supported)\n",
2648 		    *count, msgnum);
2649 
2650 	/* Don't ask for more than the device supports. */
2651 	actual = min(*count, msgnum);
2652 
2653 	/* Don't ask for more than 32 messages. */
2654 	actual = min(actual, 32);
2655 
2656 	/* MSI requires power of 2 number of messages. */
2657 	if (!powerof2(actual))
2658 		return (EINVAL);
2659 
2660 	for (;;) {
2661 		/* Try to allocate N messages. */
2662 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2663 		    actual, irqs);
2664 		if (error == 0)
2665 			break;
2666 		if (actual == 1)
2667 			return (error);
2668 
2669 		/* Try N / 2. */
2670 		actual >>= 1;
2671 	}
2672 
2673 	/*
2674 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2675 	 * resources in the irqs[] array, so add new resources
2676 	 * starting at rid 1.
2677 	 */
2678 	for (i = 0; i < actual; i++)
2679 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2680 		    irqs[i], irqs[i], 1);
2681 
2682 	if (bootverbose) {
2683 		if (actual == 1)
2684 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2685 		else {
2686 			bool run;
2687 
2688 			/*
2689 			 * Be fancy and try to print contiguous runs
2690 			 * of IRQ values as ranges.  'run' is true if
2691 			 * we are in a range.
2692 			 */
2693 			device_printf(child, "using IRQs %d", irqs[0]);
2694 			run = false;
2695 			for (i = 1; i < actual; i++) {
2696 				/* Still in a run? */
2697 				if (irqs[i] == irqs[i - 1] + 1) {
2698 					run = true;
2699 					continue;
2700 				}
2701 
2702 				/* Finish previous range. */
2703 				if (run) {
2704 					printf("-%d", irqs[i - 1]);
2705 					run = false;
2706 				}
2707 
2708 				/* Start new range. */
2709 				printf(",%d", irqs[i]);
2710 			}
2711 
2712 			/* Unfinished range? */
2713 			if (run)
2714 				printf("-%d", irqs[actual - 1]);
2715 			printf(" for MSI\n");
2716 		}
2717 	}
2718 
2719 	/* Update control register with actual count. */
2720 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2721 	ctrl |= (ffs(actual) - 1) << 4;
2722 	cfg->msi.msi_ctrl = ctrl;
2723 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2724 
2725 	/* Update counts of alloc'd messages. */
2726 	cfg->msi.msi_alloc = actual;
2727 	cfg->msi.msi_handlers = 0;
2728 	*count = actual;
2729 	return (0);
2730 }
2731 
2732 /* Release the MSI messages associated with this device. */
2733 int
pci_release_msi_method(device_t dev,device_t child)2734 pci_release_msi_method(device_t dev, device_t child)
2735 {
2736 	struct pci_devinfo *dinfo = device_get_ivars(child);
2737 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2738 	struct resource_list_entry *rle;
2739 	u_int i, irqs[32];
2740 	int error;
2741 
2742 	/* Try MSI-X first. */
2743 	error = pci_release_msix(dev, child);
2744 	if (error != ENODEV)
2745 		return (error);
2746 
2747 	/* Do we have any messages to release? */
2748 	if (msi->msi_alloc == 0)
2749 		return (ENODEV);
2750 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2751 
2752 	/* Make sure none of the resources are allocated. */
2753 	if (msi->msi_handlers > 0)
2754 		return (EBUSY);
2755 	for (i = 0; i < msi->msi_alloc; i++) {
2756 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2757 		KASSERT(rle != NULL, ("missing MSI resource"));
2758 		if (rle->res != NULL)
2759 			return (EBUSY);
2760 		irqs[i] = rle->start;
2761 	}
2762 
2763 	/* Update control register with 0 count. */
2764 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2765 	    ("%s: MSI still enabled", __func__));
2766 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2767 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2768 	    msi->msi_ctrl, 2);
2769 
2770 	/* Release the messages. */
2771 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2772 	for (i = 0; i < msi->msi_alloc; i++)
2773 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2774 
2775 	/* Update alloc count. */
2776 	msi->msi_alloc = 0;
2777 	msi->msi_addr = 0;
2778 	msi->msi_data = 0;
2779 	return (0);
2780 }
2781 
2782 /*
2783  * Return the max supported MSI messages this device supports.
2784  * Basically, assuming the MD code can alloc messages, this function
2785  * should return the maximum value that pci_alloc_msi() can return.
2786  * Thus, it is subject to the tunables, etc.
2787  */
2788 int
pci_msi_count_method(device_t dev,device_t child)2789 pci_msi_count_method(device_t dev, device_t child)
2790 {
2791 	struct pci_devinfo *dinfo = device_get_ivars(child);
2792 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2793 	uint16_t ctrl;
2794 
2795 	if (pci_do_msi && msi->msi_location != 0) {
2796 		ctrl = pci_read_config(child, msi->msi_location + PCIR_MSI_CTRL,
2797 		    2);
2798 		return (PCI_MSI_MSGNUM(ctrl));
2799 	}
2800 	return (0);
2801 }
2802 
2803 /* free pcicfgregs structure and all depending data structures */
2804 
2805 int
pci_freecfg(struct pci_devinfo * dinfo)2806 pci_freecfg(struct pci_devinfo *dinfo)
2807 {
2808 	struct devlist *devlist_head;
2809 	struct pci_map *pm, *next;
2810 
2811 	devlist_head = &pci_devq;
2812 
2813 	if (dinfo->cfg.vpd.vpd_reg)
2814 		vpd_free(&dinfo->cfg.vpd);
2815 
2816 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2817 		free(pm, M_DEVBUF);
2818 	}
2819 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2820 	free(dinfo, M_DEVBUF);
2821 
2822 	/* increment the generation count */
2823 	pci_generation++;
2824 
2825 	/* we're losing one device */
2826 	pci_numdevs--;
2827 	return (0);
2828 }
2829 
2830 /*
2831  * PCI power manangement
2832  */
2833 int
pci_set_powerstate_method(device_t dev,device_t child,int state)2834 pci_set_powerstate_method(device_t dev, device_t child, int state)
2835 {
2836 	struct pci_devinfo *dinfo = device_get_ivars(child);
2837 	pcicfgregs *cfg = &dinfo->cfg;
2838 	uint16_t status;
2839 	int oldstate, highest, delay;
2840 
2841 	if (cfg->pp.pp_cap == 0)
2842 		return (EOPNOTSUPP);
2843 
2844 	/*
2845 	 * Optimize a no state change request away.  While it would be OK to
2846 	 * write to the hardware in theory, some devices have shown odd
2847 	 * behavior when going from D3 -> D3.
2848 	 */
2849 	oldstate = pci_get_powerstate(child);
2850 	if (oldstate == state)
2851 		return (0);
2852 
2853 	/*
2854 	 * The PCI power management specification states that after a state
2855 	 * transition between PCI power states, system software must
2856 	 * guarantee a minimal delay before the function accesses the device.
2857 	 * Compute the worst case delay that we need to guarantee before we
2858 	 * access the device.  Many devices will be responsive much more
2859 	 * quickly than this delay, but there are some that don't respond
2860 	 * instantly to state changes.  Transitions to/from D3 state require
2861 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2862 	 * is done below with DELAY rather than a sleeper function because
2863 	 * this function can be called from contexts where we cannot sleep.
2864 	 */
2865 	highest = (oldstate > state) ? oldstate : state;
2866 	if (highest == PCI_POWERSTATE_D3)
2867 	    delay = 10000;
2868 	else if (highest == PCI_POWERSTATE_D2)
2869 	    delay = 200;
2870 	else
2871 	    delay = 0;
2872 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2873 	    & ~PCIM_PSTAT_DMASK;
2874 	switch (state) {
2875 	case PCI_POWERSTATE_D0:
2876 		status |= PCIM_PSTAT_D0;
2877 		break;
2878 	case PCI_POWERSTATE_D1:
2879 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2880 			return (EOPNOTSUPP);
2881 		status |= PCIM_PSTAT_D1;
2882 		break;
2883 	case PCI_POWERSTATE_D2:
2884 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2885 			return (EOPNOTSUPP);
2886 		status |= PCIM_PSTAT_D2;
2887 		break;
2888 	case PCI_POWERSTATE_D3:
2889 		status |= PCIM_PSTAT_D3;
2890 		break;
2891 	default:
2892 		return (EINVAL);
2893 	}
2894 
2895 	if (bootverbose)
2896 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2897 		    state);
2898 
2899 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2900 	if (delay)
2901 		DELAY(delay);
2902 	return (0);
2903 }
2904 
2905 int
pci_get_powerstate_method(device_t dev,device_t child)2906 pci_get_powerstate_method(device_t dev, device_t child)
2907 {
2908 	struct pci_devinfo *dinfo = device_get_ivars(child);
2909 	pcicfgregs *cfg = &dinfo->cfg;
2910 	uint16_t status;
2911 	int result;
2912 
2913 	if (cfg->pp.pp_cap != 0) {
2914 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2915 		switch (status & PCIM_PSTAT_DMASK) {
2916 		case PCIM_PSTAT_D0:
2917 			result = PCI_POWERSTATE_D0;
2918 			break;
2919 		case PCIM_PSTAT_D1:
2920 			result = PCI_POWERSTATE_D1;
2921 			break;
2922 		case PCIM_PSTAT_D2:
2923 			result = PCI_POWERSTATE_D2;
2924 			break;
2925 		case PCIM_PSTAT_D3:
2926 			result = PCI_POWERSTATE_D3;
2927 			break;
2928 		default:
2929 			result = PCI_POWERSTATE_UNKNOWN;
2930 			break;
2931 		}
2932 	} else {
2933 		/* No support, device is always at D0 */
2934 		result = PCI_POWERSTATE_D0;
2935 	}
2936 	return (result);
2937 }
2938 
2939 /*
2940  * Some convenience functions for PCI device drivers.
2941  */
2942 
2943 static __inline void
pci_set_command_bit(device_t dev,device_t child,uint16_t bit)2944 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2945 {
2946 	uint16_t	command;
2947 
2948 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2949 	command |= bit;
2950 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2951 }
2952 
2953 static __inline void
pci_clear_command_bit(device_t dev,device_t child,uint16_t bit)2954 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2955 {
2956 	uint16_t	command;
2957 
2958 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2959 	command &= ~bit;
2960 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2961 }
2962 
2963 int
pci_enable_busmaster_method(device_t dev,device_t child)2964 pci_enable_busmaster_method(device_t dev, device_t child)
2965 {
2966 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2967 	return (0);
2968 }
2969 
2970 int
pci_disable_busmaster_method(device_t dev,device_t child)2971 pci_disable_busmaster_method(device_t dev, device_t child)
2972 {
2973 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2974 	return (0);
2975 }
2976 
2977 int
pci_enable_io_method(device_t dev,device_t child,int space)2978 pci_enable_io_method(device_t dev, device_t child, int space)
2979 {
2980 	uint16_t bit;
2981 
2982 	switch(space) {
2983 	case SYS_RES_IOPORT:
2984 		bit = PCIM_CMD_PORTEN;
2985 		break;
2986 	case SYS_RES_MEMORY:
2987 		bit = PCIM_CMD_MEMEN;
2988 		break;
2989 	default:
2990 		return (EINVAL);
2991 	}
2992 	pci_set_command_bit(dev, child, bit);
2993 	return (0);
2994 }
2995 
2996 int
pci_disable_io_method(device_t dev,device_t child,int space)2997 pci_disable_io_method(device_t dev, device_t child, int space)
2998 {
2999 	uint16_t bit;
3000 
3001 	switch(space) {
3002 	case SYS_RES_IOPORT:
3003 		bit = PCIM_CMD_PORTEN;
3004 		break;
3005 	case SYS_RES_MEMORY:
3006 		bit = PCIM_CMD_MEMEN;
3007 		break;
3008 	default:
3009 		return (EINVAL);
3010 	}
3011 	pci_clear_command_bit(dev, child, bit);
3012 	return (0);
3013 }
3014 
3015 /*
3016  * New style pci driver.  Parent device is either a pci-host-bridge or a
3017  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
3018  */
3019 
3020 void
pci_print_verbose(struct pci_devinfo * dinfo)3021 pci_print_verbose(struct pci_devinfo *dinfo)
3022 {
3023 
3024 	if (bootverbose) {
3025 		pcicfgregs *cfg = &dinfo->cfg;
3026 
3027 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
3028 		    cfg->vendor, cfg->device, cfg->revid);
3029 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
3030 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
3031 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
3032 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
3033 		    cfg->mfdev);
3034 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
3035 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
3036 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
3037 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3038 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3039 		if (cfg->intpin > 0)
3040 			printf("\tintpin=%c, irq=%d\n",
3041 			    cfg->intpin +'a' -1, cfg->intline);
3042 		if (cfg->pp.pp_cap) {
3043 			uint16_t status;
3044 
3045 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
3046 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3047 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3048 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3049 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3050 			    status & PCIM_PSTAT_DMASK);
3051 		}
3052 		if (cfg->msi.msi_location) {
3053 			uint16_t ctrl, msgnum;
3054 
3055 			ctrl = cfg->msi.msi_ctrl;
3056 			msgnum = PCI_MSI_MSGNUM(ctrl);
3057 			printf("\tMSI supports %d message%s%s%s\n",
3058 			    msgnum, (msgnum == 1) ? "" : "s",
3059 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3060 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3061 		}
3062 		if (cfg->msix.msix_location) {
3063 			uint16_t msgnum;
3064 
3065 			msgnum = PCI_MSIX_MSGNUM(cfg->msix.msix_ctrl);
3066 			printf("\tMSI-X supports %d message%s ",
3067 			    msgnum, (msgnum == 1) ? "" : "s");
3068 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3069 				printf("in map 0x%x\n",
3070 				    cfg->msix.msix_table_bar);
3071 			else
3072 				printf("in maps 0x%x and 0x%x\n",
3073 				    cfg->msix.msix_table_bar,
3074 				    cfg->msix.msix_pba_bar);
3075 		}
3076 	}
3077 }
3078 
3079 static int
pci_porten(device_t dev)3080 pci_porten(device_t dev)
3081 {
3082 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3083 }
3084 
3085 static int
pci_memen(device_t dev)3086 pci_memen(device_t dev)
3087 {
3088 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3089 }
3090 
3091 void
pci_read_bar(device_t dev,int reg,pci_addr_t * mapp,pci_addr_t * testvalp,int * bar64)3092 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3093     int *bar64)
3094 {
3095 	struct pci_devinfo *dinfo;
3096 	pci_addr_t map, testval;
3097 	int ln2range;
3098 	uint16_t cmd;
3099 
3100 	/*
3101 	 * The device ROM BAR is special.  It is always a 32-bit
3102 	 * memory BAR.  Bit 0 is special and should not be set when
3103 	 * sizing the BAR.
3104 	 */
3105 	dinfo = device_get_ivars(dev);
3106 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3107 		map = pci_read_config(dev, reg, 4);
3108 		pci_write_config(dev, reg, 0xfffffffe, 4);
3109 		testval = pci_read_config(dev, reg, 4);
3110 		pci_write_config(dev, reg, map, 4);
3111 		*mapp = map;
3112 		*testvalp = testval;
3113 		if (bar64 != NULL)
3114 			*bar64 = 0;
3115 		return;
3116 	}
3117 
3118 	map = pci_read_config(dev, reg, 4);
3119 	ln2range = pci_maprange(map);
3120 	if (ln2range == 64)
3121 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3122 
3123 	/*
3124 	 * Disable decoding via the command register before
3125 	 * determining the BAR's length since we will be placing it in
3126 	 * a weird state.
3127 	 */
3128 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3129 	pci_write_config(dev, PCIR_COMMAND,
3130 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3131 
3132 	/*
3133 	 * Determine the BAR's length by writing all 1's.  The bottom
3134 	 * log_2(size) bits of the BAR will stick as 0 when we read
3135 	 * the value back.
3136 	 *
3137 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3138 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3139 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3140 	 *
3141 	 * Writes to both registers must be performed before attempting to
3142 	 * read back the size value.
3143 	 */
3144 	testval = 0;
3145 	pci_write_config(dev, reg, 0xffffffff, 4);
3146 	if (ln2range == 64) {
3147 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3148 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3149 	}
3150 	testval |= pci_read_config(dev, reg, 4);
3151 
3152 	/*
3153 	 * Restore the original value of the BAR.  We may have reprogrammed
3154 	 * the BAR of the low-level console device and when booting verbose,
3155 	 * we need the console device addressable.
3156 	 */
3157 	pci_write_config(dev, reg, map, 4);
3158 	if (ln2range == 64)
3159 		pci_write_config(dev, reg + 4, map >> 32, 4);
3160 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3161 
3162 	*mapp = map;
3163 	*testvalp = testval;
3164 	if (bar64 != NULL)
3165 		*bar64 = (ln2range == 64);
3166 }
3167 
3168 static void
pci_write_bar(device_t dev,struct pci_map * pm,pci_addr_t base)3169 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3170 {
3171 	struct pci_devinfo *dinfo;
3172 	int ln2range;
3173 
3174 	/* The device ROM BAR is always a 32-bit memory BAR. */
3175 	dinfo = device_get_ivars(dev);
3176 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3177 		ln2range = 32;
3178 	else
3179 		ln2range = pci_maprange(pm->pm_value);
3180 	pci_write_config(dev, pm->pm_reg, base, 4);
3181 	if (ln2range == 64)
3182 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3183 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3184 	if (ln2range == 64)
3185 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3186 		    pm->pm_reg + 4, 4) << 32;
3187 }
3188 
3189 struct pci_map *
pci_find_bar(device_t dev,int reg)3190 pci_find_bar(device_t dev, int reg)
3191 {
3192 	struct pci_devinfo *dinfo;
3193 	struct pci_map *pm;
3194 
3195 	dinfo = device_get_ivars(dev);
3196 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3197 		if (pm->pm_reg == reg)
3198 			return (pm);
3199 	}
3200 	return (NULL);
3201 }
3202 
3203 struct pci_map *
pci_first_bar(device_t dev)3204 pci_first_bar(device_t dev)
3205 {
3206 	struct pci_devinfo *dinfo;
3207 
3208 	dinfo = device_get_ivars(dev);
3209 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3210 }
3211 
3212 struct pci_map *
pci_next_bar(struct pci_map * pm)3213 pci_next_bar(struct pci_map *pm)
3214 {
3215 	return (STAILQ_NEXT(pm, pm_link));
3216 }
3217 
3218 int
pci_bar_enabled(device_t dev,struct pci_map * pm)3219 pci_bar_enabled(device_t dev, struct pci_map *pm)
3220 {
3221 	struct pci_devinfo *dinfo;
3222 	uint16_t cmd;
3223 
3224 	dinfo = device_get_ivars(dev);
3225 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3226 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3227 		return (0);
3228 #ifdef PCI_IOV
3229 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3230 		struct pcicfg_iov *iov;
3231 
3232 		iov = dinfo->cfg.iov;
3233 		cmd = pci_read_config(iov->iov_pf,
3234 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3235 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3236 	}
3237 #endif
3238 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3239 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3240 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3241 	else
3242 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3243 }
3244 
3245 struct pci_map *
pci_add_bar(device_t dev,int reg,pci_addr_t value,pci_addr_t size)3246 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3247 {
3248 	struct pci_devinfo *dinfo;
3249 	struct pci_map *pm, *prev;
3250 
3251 	dinfo = device_get_ivars(dev);
3252 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3253 	pm->pm_reg = reg;
3254 	pm->pm_value = value;
3255 	pm->pm_size = size;
3256 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3257 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3258 		    reg));
3259 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3260 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3261 			break;
3262 	}
3263 	if (prev != NULL)
3264 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3265 	else
3266 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3267 	return (pm);
3268 }
3269 
3270 static void
pci_restore_bars(device_t dev)3271 pci_restore_bars(device_t dev)
3272 {
3273 	struct pci_devinfo *dinfo;
3274 	struct pci_map *pm;
3275 	int ln2range;
3276 
3277 	dinfo = device_get_ivars(dev);
3278 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3279 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3280 			ln2range = 32;
3281 		else
3282 			ln2range = pci_maprange(pm->pm_value);
3283 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3284 		if (ln2range == 64)
3285 			pci_write_config(dev, pm->pm_reg + 4,
3286 			    pm->pm_value >> 32, 4);
3287 	}
3288 }
3289 
3290 /*
3291  * Add a resource based on a pci map register. Return 1 if the map
3292  * register is a 32bit map register or 2 if it is a 64bit register.
3293  */
3294 static int
pci_add_map(device_t bus,device_t dev,int reg,struct resource_list * rl,int force,int prefetch)3295 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3296     int force, int prefetch)
3297 {
3298 	struct pci_map *pm;
3299 	pci_addr_t base, map, testval;
3300 	pci_addr_t start, end, count;
3301 	int barlen, basezero, flags, maprange, mapsize, type;
3302 	uint16_t cmd;
3303 	struct resource *res;
3304 
3305 	/*
3306 	 * The BAR may already exist if the device is a CardBus card
3307 	 * whose CIS is stored in this BAR.
3308 	 */
3309 	pm = pci_find_bar(dev, reg);
3310 	if (pm != NULL) {
3311 		maprange = pci_maprange(pm->pm_value);
3312 		barlen = maprange == 64 ? 2 : 1;
3313 		return (barlen);
3314 	}
3315 
3316 	pci_read_bar(dev, reg, &map, &testval, NULL);
3317 	if (PCI_BAR_MEM(map)) {
3318 		type = SYS_RES_MEMORY;
3319 		if (map & PCIM_BAR_MEM_PREFETCH)
3320 			prefetch = 1;
3321 	} else
3322 		type = SYS_RES_IOPORT;
3323 	mapsize = pci_mapsize(testval);
3324 	base = pci_mapbase(map);
3325 #ifdef __PCI_BAR_ZERO_VALID
3326 	basezero = 0;
3327 #else
3328 	basezero = base == 0;
3329 #endif
3330 	maprange = pci_maprange(map);
3331 	barlen = maprange == 64 ? 2 : 1;
3332 
3333 	/*
3334 	 * For I/O registers, if bottom bit is set, and the next bit up
3335 	 * isn't clear, we know we have a BAR that doesn't conform to the
3336 	 * spec, so ignore it.  Also, sanity check the size of the data
3337 	 * areas to the type of memory involved.  Memory must be at least
3338 	 * 16 bytes in size, while I/O ranges must be at least 4.
3339 	 */
3340 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3341 		return (barlen);
3342 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3343 	    (type == SYS_RES_IOPORT && mapsize < 2))
3344 		return (barlen);
3345 
3346 	/* Save a record of this BAR. */
3347 	pm = pci_add_bar(dev, reg, map, mapsize);
3348 	if (bootverbose) {
3349 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3350 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3351 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3352 			printf(", port disabled\n");
3353 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3354 			printf(", memory disabled\n");
3355 		else
3356 			printf(", enabled\n");
3357 	}
3358 
3359 	/*
3360 	 * If base is 0, then we have problems if this architecture does
3361 	 * not allow that.  It is best to ignore such entries for the
3362 	 * moment.  These will be allocated later if the driver specifically
3363 	 * requests them.  However, some removable buses look better when
3364 	 * all resources are allocated, so allow '0' to be overridden.
3365 	 *
3366 	 * Similarly treat maps whose values is the same as the test value
3367 	 * read back.  These maps have had all f's written to them by the
3368 	 * BIOS in an attempt to disable the resources.
3369 	 */
3370 	if (!force && (basezero || map == testval))
3371 		return (barlen);
3372 	if ((u_long)base != base) {
3373 		device_printf(bus,
3374 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3375 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3376 		    pci_get_function(dev), reg);
3377 		return (barlen);
3378 	}
3379 
3380 	/*
3381 	 * This code theoretically does the right thing, but has
3382 	 * undesirable side effects in some cases where peripherals
3383 	 * respond oddly to having these bits enabled.  Let the user
3384 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3385 	 * default).
3386 	 */
3387 	if (pci_enable_io_modes) {
3388 		/* Turn on resources that have been left off by a lazy BIOS */
3389 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3390 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3391 			cmd |= PCIM_CMD_PORTEN;
3392 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3393 		}
3394 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3395 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3396 			cmd |= PCIM_CMD_MEMEN;
3397 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3398 		}
3399 	} else {
3400 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3401 			return (barlen);
3402 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3403 			return (barlen);
3404 	}
3405 
3406 	count = (pci_addr_t)1 << mapsize;
3407 	flags = RF_ALIGNMENT_LOG2(mapsize);
3408 	if (prefetch)
3409 		flags |= RF_PREFETCHABLE;
3410 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3411 		start = 0;	/* Let the parent decide. */
3412 		end = ~0;
3413 	} else {
3414 		start = base;
3415 		end = base + count - 1;
3416 	}
3417 	resource_list_add(rl, type, reg, start, end, count);
3418 
3419 	/*
3420 	 * Try to allocate the resource for this BAR from our parent
3421 	 * so that this resource range is already reserved.  The
3422 	 * driver for this device will later inherit this resource in
3423 	 * pci_alloc_resource().
3424 	 */
3425 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3426 	    flags);
3427 	if ((pci_do_realloc_bars
3428 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3429 	    && res == NULL && (start != 0 || end != ~0)) {
3430 		/*
3431 		 * If the allocation fails, try to allocate a resource for
3432 		 * this BAR using any available range.  The firmware felt
3433 		 * it was important enough to assign a resource, so don't
3434 		 * disable decoding if we can help it.
3435 		 */
3436 		resource_list_delete(rl, type, reg);
3437 		resource_list_add(rl, type, reg, 0, ~0, count);
3438 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3439 		    count, flags);
3440 	}
3441 	if (res == NULL) {
3442 		/*
3443 		 * If the allocation fails, delete the resource list entry
3444 		 * and disable decoding for this device.
3445 		 *
3446 		 * If the driver requests this resource in the future,
3447 		 * pci_reserve_map() will try to allocate a fresh
3448 		 * resource range.
3449 		 */
3450 		resource_list_delete(rl, type, reg);
3451 		pci_disable_io(dev, type);
3452 		if (bootverbose)
3453 			device_printf(bus,
3454 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3455 			    pci_get_domain(dev), pci_get_bus(dev),
3456 			    pci_get_slot(dev), pci_get_function(dev), reg);
3457 	} else {
3458 		start = rman_get_start(res);
3459 		pci_write_bar(dev, pm, start);
3460 	}
3461 	return (barlen);
3462 }
3463 
3464 /*
3465  * For ATA devices we need to decide early what addressing mode to use.
3466  * Legacy demands that the primary and secondary ATA ports sits on the
3467  * same addresses that old ISA hardware did. This dictates that we use
3468  * those addresses and ignore the BAR's if we cannot set PCI native
3469  * addressing mode.
3470  */
3471 static void
pci_ata_maps(device_t bus,device_t dev,struct resource_list * rl,int force,uint32_t prefetchmask)3472 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3473     uint32_t prefetchmask)
3474 {
3475 	int rid, type, progif;
3476 #if 0
3477 	/* if this device supports PCI native addressing use it */
3478 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3479 	if ((progif & 0x8a) == 0x8a) {
3480 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3481 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3482 			printf("Trying ATA native PCI addressing mode\n");
3483 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3484 		}
3485 	}
3486 #endif
3487 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3488 	type = SYS_RES_IOPORT;
3489 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3490 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3491 		    prefetchmask & (1 << 0));
3492 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3493 		    prefetchmask & (1 << 1));
3494 	} else {
3495 		rid = PCIR_BAR(0);
3496 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3497 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3498 		    0x1f7, 8, 0);
3499 		rid = PCIR_BAR(1);
3500 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3501 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3502 		    0x3f6, 1, 0);
3503 	}
3504 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3505 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3506 		    prefetchmask & (1 << 2));
3507 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3508 		    prefetchmask & (1 << 3));
3509 	} else {
3510 		rid = PCIR_BAR(2);
3511 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3512 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3513 		    0x177, 8, 0);
3514 		rid = PCIR_BAR(3);
3515 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3516 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3517 		    0x376, 1, 0);
3518 	}
3519 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3520 	    prefetchmask & (1 << 4));
3521 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3522 	    prefetchmask & (1 << 5));
3523 }
3524 
3525 static void
pci_assign_interrupt(device_t bus,device_t dev,int force_route)3526 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3527 {
3528 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3529 	pcicfgregs *cfg = &dinfo->cfg;
3530 	char tunable_name[64];
3531 	int irq;
3532 
3533 	/* Has to have an intpin to have an interrupt. */
3534 	if (cfg->intpin == 0)
3535 		return;
3536 
3537 	/* Let the user override the IRQ with a tunable. */
3538 	irq = PCI_INVALID_IRQ;
3539 	snprintf(tunable_name, sizeof(tunable_name),
3540 	    "hw.pci%d.%d.%d.INT%c.irq",
3541 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3542 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3543 		irq = PCI_INVALID_IRQ;
3544 
3545 	/*
3546 	 * If we didn't get an IRQ via the tunable, then we either use the
3547 	 * IRQ value in the intline register or we ask the bus to route an
3548 	 * interrupt for us.  If force_route is true, then we only use the
3549 	 * value in the intline register if the bus was unable to assign an
3550 	 * IRQ.
3551 	 */
3552 	if (!PCI_INTERRUPT_VALID(irq)) {
3553 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3554 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3555 		if (!PCI_INTERRUPT_VALID(irq))
3556 			irq = cfg->intline;
3557 	}
3558 
3559 	/* If after all that we don't have an IRQ, just bail. */
3560 	if (!PCI_INTERRUPT_VALID(irq))
3561 		return;
3562 
3563 	/* Update the config register if it changed. */
3564 	if (irq != cfg->intline) {
3565 		cfg->intline = irq;
3566 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3567 	}
3568 
3569 	/* Add this IRQ as rid 0 interrupt resource. */
3570 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3571 }
3572 
3573 /* Perform early OHCI takeover from SMM. */
3574 static void
ohci_early_takeover(device_t self)3575 ohci_early_takeover(device_t self)
3576 {
3577 	struct resource *res;
3578 	uint32_t ctl;
3579 	int rid;
3580 	int i;
3581 
3582 	rid = PCIR_BAR(0);
3583 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3584 	if (res == NULL)
3585 		return;
3586 
3587 	ctl = bus_read_4(res, OHCI_CONTROL);
3588 	if (ctl & OHCI_IR) {
3589 		if (bootverbose)
3590 			printf("ohci early: "
3591 			    "SMM active, request owner change\n");
3592 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3593 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3594 			DELAY(1000);
3595 			ctl = bus_read_4(res, OHCI_CONTROL);
3596 		}
3597 		if (ctl & OHCI_IR) {
3598 			if (bootverbose)
3599 				printf("ohci early: "
3600 				    "SMM does not respond, resetting\n");
3601 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3602 		}
3603 		/* Disable interrupts */
3604 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3605 	}
3606 
3607 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3608 }
3609 
3610 /* Perform early UHCI takeover from SMM. */
3611 static void
uhci_early_takeover(device_t self)3612 uhci_early_takeover(device_t self)
3613 {
3614 	struct resource *res;
3615 	int rid;
3616 
3617 	/*
3618 	 * Set the PIRQD enable bit and switch off all the others. We don't
3619 	 * want legacy support to interfere with us XXX Does this also mean
3620 	 * that the BIOS won't touch the keyboard anymore if it is connected
3621 	 * to the ports of the root hub?
3622 	 */
3623 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3624 
3625 	/* Disable interrupts */
3626 	rid = PCI_UHCI_BASE_REG;
3627 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3628 	if (res != NULL) {
3629 		bus_write_2(res, UHCI_INTR, 0);
3630 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3631 	}
3632 }
3633 
3634 /* Perform early EHCI takeover from SMM. */
3635 static void
ehci_early_takeover(device_t self)3636 ehci_early_takeover(device_t self)
3637 {
3638 	struct resource *res;
3639 	uint32_t cparams;
3640 	uint32_t eec;
3641 	uint8_t eecp;
3642 	uint8_t bios_sem;
3643 	uint8_t offs;
3644 	int rid;
3645 	int i;
3646 
3647 	rid = PCIR_BAR(0);
3648 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3649 	if (res == NULL)
3650 		return;
3651 
3652 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3653 
3654 	/* Synchronise with the BIOS if it owns the controller. */
3655 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3656 	    eecp = EHCI_EECP_NEXT(eec)) {
3657 		eec = pci_read_config(self, eecp, 4);
3658 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3659 			continue;
3660 		}
3661 		bios_sem = pci_read_config(self, eecp +
3662 		    EHCI_LEGSUP_BIOS_SEM, 1);
3663 		if (bios_sem == 0) {
3664 			continue;
3665 		}
3666 		if (bootverbose)
3667 			printf("ehci early: "
3668 			    "SMM active, request owner change\n");
3669 
3670 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3671 
3672 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3673 			DELAY(1000);
3674 			bios_sem = pci_read_config(self, eecp +
3675 			    EHCI_LEGSUP_BIOS_SEM, 1);
3676 		}
3677 
3678 		if (bios_sem != 0) {
3679 			if (bootverbose)
3680 				printf("ehci early: "
3681 				    "SMM does not respond\n");
3682 		}
3683 		/* Disable interrupts */
3684 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3685 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3686 	}
3687 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3688 }
3689 
3690 /* Perform early XHCI takeover from SMM. */
3691 static void
xhci_early_takeover(device_t self)3692 xhci_early_takeover(device_t self)
3693 {
3694 	struct resource *res;
3695 	uint32_t cparams;
3696 	uint32_t eec;
3697 	uint8_t eecp;
3698 	uint8_t bios_sem;
3699 	uint8_t offs;
3700 	int rid;
3701 	int i;
3702 
3703 	rid = PCIR_BAR(0);
3704 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3705 	if (res == NULL)
3706 		return;
3707 
3708 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3709 
3710 	eec = -1;
3711 
3712 	/* Synchronise with the BIOS if it owns the controller. */
3713 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3714 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3715 		eec = bus_read_4(res, eecp);
3716 
3717 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3718 			continue;
3719 
3720 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3721 		if (bios_sem == 0)
3722 			continue;
3723 
3724 		if (bootverbose)
3725 			printf("xhci early: "
3726 			    "SMM active, request owner change\n");
3727 
3728 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3729 
3730 		/* wait a maximum of 5 second */
3731 
3732 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3733 			DELAY(1000);
3734 			bios_sem = bus_read_1(res, eecp +
3735 			    XHCI_XECP_BIOS_SEM);
3736 		}
3737 
3738 		if (bios_sem != 0) {
3739 			if (bootverbose)
3740 				printf("xhci early: "
3741 				    "SMM does not respond\n");
3742 		}
3743 
3744 		/* Disable interrupts */
3745 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3746 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3747 		bus_read_4(res, offs + XHCI_USBSTS);
3748 	}
3749 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3750 }
3751 
3752 static void
pci_reserve_secbus(device_t bus,device_t dev,pcicfgregs * cfg,struct resource_list * rl)3753 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3754     struct resource_list *rl)
3755 {
3756 	struct resource *res;
3757 	char *cp;
3758 	rman_res_t start, end, count;
3759 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3760 
3761 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3762 	case PCIM_HDRTYPE_BRIDGE:
3763 		sec_reg = PCIR_SECBUS_1;
3764 		sub_reg = PCIR_SUBBUS_1;
3765 		break;
3766 	case PCIM_HDRTYPE_CARDBUS:
3767 		sec_reg = PCIR_SECBUS_2;
3768 		sub_reg = PCIR_SUBBUS_2;
3769 		break;
3770 	default:
3771 		return;
3772 	}
3773 
3774 	/*
3775 	 * If the existing bus range is valid, attempt to reserve it
3776 	 * from our parent.  If this fails for any reason, clear the
3777 	 * secbus and subbus registers.
3778 	 *
3779 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3780 	 * This would at least preserve the existing sec_bus if it is
3781 	 * valid.
3782 	 */
3783 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3784 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3785 
3786 	/* Quirk handling. */
3787 	switch (pci_get_devid(dev)) {
3788 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3789 		sup_bus = pci_read_config(dev, 0x41, 1);
3790 		if (sup_bus != 0xff) {
3791 			sec_bus = sup_bus + 1;
3792 			sub_bus = sup_bus + 1;
3793 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3794 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3795 		}
3796 		break;
3797 
3798 	case 0x00dd10de:
3799 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3800 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3801 			break;
3802 		if (strncmp(cp, "Compal", 6) != 0) {
3803 			freeenv(cp);
3804 			break;
3805 		}
3806 		freeenv(cp);
3807 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3808 			break;
3809 		if (strncmp(cp, "08A0", 4) != 0) {
3810 			freeenv(cp);
3811 			break;
3812 		}
3813 		freeenv(cp);
3814 		if (sub_bus < 0xa) {
3815 			sub_bus = 0xa;
3816 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3817 		}
3818 		break;
3819 	}
3820 
3821 	if (bootverbose)
3822 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3823 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3824 		start = sec_bus;
3825 		end = sub_bus;
3826 		count = end - start + 1;
3827 
3828 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3829 
3830 		/*
3831 		 * If requested, clear secondary bus registers in
3832 		 * bridge devices to force a complete renumbering
3833 		 * rather than reserving the existing range.  However,
3834 		 * preserve the existing size.
3835 		 */
3836 		if (pci_clear_buses)
3837 			goto clear;
3838 
3839 		rid = 0;
3840 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3841 		    start, end, count, 0);
3842 		if (res != NULL)
3843 			return;
3844 
3845 		if (bootverbose)
3846 			device_printf(bus,
3847 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3848 			    pci_get_domain(dev), pci_get_bus(dev),
3849 			    pci_get_slot(dev), pci_get_function(dev));
3850 	}
3851 
3852 clear:
3853 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3854 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3855 }
3856 
3857 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)3858 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3859     rman_res_t end, rman_res_t count, u_int flags)
3860 {
3861 	struct pci_devinfo *dinfo;
3862 	pcicfgregs *cfg;
3863 	struct resource_list *rl;
3864 	struct resource *res;
3865 	int sec_reg, sub_reg;
3866 
3867 	dinfo = device_get_ivars(child);
3868 	cfg = &dinfo->cfg;
3869 	rl = &dinfo->resources;
3870 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3871 	case PCIM_HDRTYPE_BRIDGE:
3872 		sec_reg = PCIR_SECBUS_1;
3873 		sub_reg = PCIR_SUBBUS_1;
3874 		break;
3875 	case PCIM_HDRTYPE_CARDBUS:
3876 		sec_reg = PCIR_SECBUS_2;
3877 		sub_reg = PCIR_SUBBUS_2;
3878 		break;
3879 	default:
3880 		return (NULL);
3881 	}
3882 
3883 	if (*rid != 0)
3884 		return (NULL);
3885 
3886 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3887 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3888 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3889 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3890 		    start, end, count, flags & ~RF_ACTIVE);
3891 		if (res == NULL) {
3892 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3893 			device_printf(child, "allocating %ju bus%s failed\n",
3894 			    count, count == 1 ? "" : "es");
3895 			return (NULL);
3896 		}
3897 		if (bootverbose)
3898 			device_printf(child,
3899 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3900 			    count == 1 ? "" : "es", rman_get_start(res));
3901 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3902 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3903 	}
3904 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3905 	    end, count, flags));
3906 }
3907 
3908 static int
pci_ea_bei_to_rid(device_t dev,int bei)3909 pci_ea_bei_to_rid(device_t dev, int bei)
3910 {
3911 #ifdef PCI_IOV
3912 	struct pci_devinfo *dinfo;
3913 	int iov_pos;
3914 	struct pcicfg_iov *iov;
3915 
3916 	dinfo = device_get_ivars(dev);
3917 	iov = dinfo->cfg.iov;
3918 	if (iov != NULL)
3919 		iov_pos = iov->iov_pos;
3920 	else
3921 		iov_pos = 0;
3922 #endif
3923 
3924 	/* Check if matches BAR */
3925 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3926 	    (bei <= PCIM_EA_BEI_BAR_5))
3927 		return (PCIR_BAR(bei));
3928 
3929 	/* Check ROM */
3930 	if (bei == PCIM_EA_BEI_ROM)
3931 		return (PCIR_BIOS);
3932 
3933 #ifdef PCI_IOV
3934 	/* Check if matches VF_BAR */
3935 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3936 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3937 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3938 		    iov_pos);
3939 #endif
3940 
3941 	return (-1);
3942 }
3943 
3944 int
pci_ea_is_enabled(device_t dev,int rid)3945 pci_ea_is_enabled(device_t dev, int rid)
3946 {
3947 	struct pci_ea_entry *ea;
3948 	struct pci_devinfo *dinfo;
3949 
3950 	dinfo = device_get_ivars(dev);
3951 
3952 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3953 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3954 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3955 	}
3956 
3957 	return (0);
3958 }
3959 
3960 void
pci_add_resources_ea(device_t bus,device_t dev,int alloc_iov)3961 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3962 {
3963 	struct pci_ea_entry *ea;
3964 	struct pci_devinfo *dinfo;
3965 	pci_addr_t start, end, count;
3966 	struct resource_list *rl;
3967 	int type, flags, rid;
3968 	struct resource *res;
3969 	uint32_t tmp;
3970 #ifdef PCI_IOV
3971 	struct pcicfg_iov *iov;
3972 #endif
3973 
3974 	dinfo = device_get_ivars(dev);
3975 	rl = &dinfo->resources;
3976 	flags = 0;
3977 
3978 #ifdef PCI_IOV
3979 	iov = dinfo->cfg.iov;
3980 #endif
3981 
3982 	if (dinfo->cfg.ea.ea_location == 0)
3983 		return;
3984 
3985 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3986 		/*
3987 		 * TODO: Ignore EA-BAR if is not enabled.
3988 		 *   Currently the EA implementation supports
3989 		 *   only situation, where EA structure contains
3990 		 *   predefined entries. In case they are not enabled
3991 		 *   leave them unallocated and proceed with
3992 		 *   a legacy-BAR mechanism.
3993 		 */
3994 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3995 			continue;
3996 
3997 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3998 		case PCIM_EA_P_MEM_PREFETCH:
3999 		case PCIM_EA_P_VF_MEM_PREFETCH:
4000 			flags = RF_PREFETCHABLE;
4001 			/* FALLTHROUGH */
4002 		case PCIM_EA_P_VF_MEM:
4003 		case PCIM_EA_P_MEM:
4004 			type = SYS_RES_MEMORY;
4005 			break;
4006 		case PCIM_EA_P_IO:
4007 			type = SYS_RES_IOPORT;
4008 			break;
4009 		default:
4010 			continue;
4011 		}
4012 
4013 		if (alloc_iov != 0) {
4014 #ifdef PCI_IOV
4015 			/* Allocating IOV, confirm BEI matches */
4016 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
4017 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
4018 				continue;
4019 #else
4020 			continue;
4021 #endif
4022 		} else {
4023 			/* Allocating BAR, confirm BEI matches */
4024 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
4025 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
4026 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
4027 				continue;
4028 		}
4029 
4030 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
4031 		if (rid < 0)
4032 			continue;
4033 
4034 		/* Skip resources already allocated by EA */
4035 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
4036 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
4037 			continue;
4038 
4039 		start = ea->eae_base;
4040 		count = ea->eae_max_offset + 1;
4041 #ifdef PCI_IOV
4042 		if (iov != NULL)
4043 			count = count * iov->iov_num_vfs;
4044 #endif
4045 		end = start + count - 1;
4046 		if (count == 0)
4047 			continue;
4048 
4049 		resource_list_add(rl, type, rid, start, end, count);
4050 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4051 		    flags);
4052 		if (res == NULL) {
4053 			resource_list_delete(rl, type, rid);
4054 
4055 			/*
4056 			 * Failed to allocate using EA, disable entry.
4057 			 * Another attempt to allocation will be performed
4058 			 * further, but this time using legacy BAR registers
4059 			 */
4060 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4061 			tmp &= ~PCIM_EA_ENABLE;
4062 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4063 
4064 			/*
4065 			 * Disabling entry might fail in case it is hardwired.
4066 			 * Read flags again to match current status.
4067 			 */
4068 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4069 
4070 			continue;
4071 		}
4072 
4073 		/* As per specification, fill BAR with zeros */
4074 		pci_write_config(dev, rid, 0, 4);
4075 	}
4076 }
4077 
4078 void
pci_add_resources(device_t bus,device_t dev,int force,uint32_t prefetchmask)4079 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4080 {
4081 	struct pci_devinfo *dinfo;
4082 	pcicfgregs *cfg;
4083 	struct resource_list *rl;
4084 	const struct pci_quirk *q;
4085 	uint32_t devid;
4086 	int i;
4087 
4088 	dinfo = device_get_ivars(dev);
4089 	cfg = &dinfo->cfg;
4090 	rl = &dinfo->resources;
4091 	devid = (cfg->device << 16) | cfg->vendor;
4092 
4093 	/* Allocate resources using Enhanced Allocation */
4094 	pci_add_resources_ea(bus, dev, 0);
4095 
4096 	/* ATA devices needs special map treatment */
4097 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4098 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4099 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4100 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4101 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4102 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4103 	else
4104 		for (i = 0; i < cfg->nummaps;) {
4105 			/* Skip resources already managed by EA */
4106 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4107 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4108 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4109 				i++;
4110 				continue;
4111 			}
4112 
4113 			/*
4114 			 * Skip quirked resources.
4115 			 */
4116 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4117 				if (q->devid == devid &&
4118 				    q->type == PCI_QUIRK_UNMAP_REG &&
4119 				    q->arg1 == PCIR_BAR(i))
4120 					break;
4121 			if (q->devid != 0) {
4122 				i++;
4123 				continue;
4124 			}
4125 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4126 			    prefetchmask & (1 << i));
4127 		}
4128 
4129 	/*
4130 	 * Add additional, quirked resources.
4131 	 */
4132 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4133 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4134 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4135 
4136 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4137 		/*
4138 		 * Try to re-route interrupts. Sometimes the BIOS or
4139 		 * firmware may leave bogus values in these registers.
4140 		 * If the re-route fails, then just stick with what we
4141 		 * have.
4142 		 */
4143 		pci_assign_interrupt(bus, dev, 1);
4144 	}
4145 
4146 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4147 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4148 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4149 			xhci_early_takeover(dev);
4150 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4151 			ehci_early_takeover(dev);
4152 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4153 			ohci_early_takeover(dev);
4154 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4155 			uhci_early_takeover(dev);
4156 	}
4157 
4158 	/*
4159 	 * Reserve resources for secondary bus ranges behind bridge
4160 	 * devices.
4161 	 */
4162 	pci_reserve_secbus(bus, dev, cfg, rl);
4163 }
4164 
4165 static struct pci_devinfo *
pci_identify_function(device_t pcib,device_t dev,int domain,int busno,int slot,int func)4166 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4167     int slot, int func)
4168 {
4169 	struct pci_devinfo *dinfo;
4170 
4171 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4172 	if (dinfo != NULL)
4173 		pci_add_child(dev, dinfo);
4174 
4175 	return (dinfo);
4176 }
4177 
4178 void
pci_add_children(device_t dev,int domain,int busno)4179 pci_add_children(device_t dev, int domain, int busno)
4180 {
4181 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4182 	device_t pcib = device_get_parent(dev);
4183 	struct pci_devinfo *dinfo;
4184 	int maxslots;
4185 	int s, f, pcifunchigh;
4186 	uint8_t hdrtype;
4187 	int first_func;
4188 
4189 	/*
4190 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4191 	 * enable ARI.  We must enable ARI before detecting the rest of the
4192 	 * functions on this bus as ARI changes the set of slots and functions
4193 	 * that are legal on this bus.
4194 	 */
4195 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4196 	if (dinfo != NULL && pci_enable_ari)
4197 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4198 
4199 	/*
4200 	 * Start looking for new devices on slot 0 at function 1 because we
4201 	 * just identified the device at slot 0, function 0.
4202 	 */
4203 	first_func = 1;
4204 
4205 	maxslots = PCIB_MAXSLOTS(pcib);
4206 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4207 		pcifunchigh = 0;
4208 		f = 0;
4209 		DELAY(1);
4210 
4211 		/* If function 0 is not present, skip to the next slot. */
4212 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4213 			continue;
4214 		hdrtype = REG(PCIR_HDRTYPE, 1);
4215 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4216 			continue;
4217 		if (hdrtype & PCIM_MFDEV)
4218 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4219 		for (f = first_func; f <= pcifunchigh; f++)
4220 			pci_identify_function(pcib, dev, domain, busno, s, f);
4221 	}
4222 #undef REG
4223 }
4224 
4225 int
pci_rescan_method(device_t dev)4226 pci_rescan_method(device_t dev)
4227 {
4228 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4229 	device_t pcib = device_get_parent(dev);
4230 	device_t child, *devlist, *unchanged;
4231 	int devcount, error, i, j, maxslots, oldcount;
4232 	int busno, domain, s, f, pcifunchigh;
4233 	uint8_t hdrtype;
4234 
4235 	/* No need to check for ARI on a rescan. */
4236 	error = device_get_children(dev, &devlist, &devcount);
4237 	if (error)
4238 		return (error);
4239 	if (devcount != 0) {
4240 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4241 		    M_NOWAIT | M_ZERO);
4242 		if (unchanged == NULL) {
4243 			free(devlist, M_TEMP);
4244 			return (ENOMEM);
4245 		}
4246 	} else
4247 		unchanged = NULL;
4248 
4249 	domain = pcib_get_domain(dev);
4250 	busno = pcib_get_bus(dev);
4251 	maxslots = PCIB_MAXSLOTS(pcib);
4252 	for (s = 0; s <= maxslots; s++) {
4253 		/* If function 0 is not present, skip to the next slot. */
4254 		f = 0;
4255 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4256 			continue;
4257 		pcifunchigh = 0;
4258 		hdrtype = REG(PCIR_HDRTYPE, 1);
4259 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4260 			continue;
4261 		if (hdrtype & PCIM_MFDEV)
4262 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4263 		for (f = 0; f <= pcifunchigh; f++) {
4264 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4265 				continue;
4266 
4267 			/*
4268 			 * Found a valid function.  Check if a
4269 			 * device_t for this device already exists.
4270 			 */
4271 			for (i = 0; i < devcount; i++) {
4272 				child = devlist[i];
4273 				if (child == NULL)
4274 					continue;
4275 				if (pci_get_slot(child) == s &&
4276 				    pci_get_function(child) == f) {
4277 					unchanged[i] = child;
4278 					goto next_func;
4279 				}
4280 			}
4281 
4282 			pci_identify_function(pcib, dev, domain, busno, s, f);
4283 		next_func:;
4284 		}
4285 	}
4286 
4287 	/* Remove devices that are no longer present. */
4288 	for (i = 0; i < devcount; i++) {
4289 		if (unchanged[i] != NULL)
4290 			continue;
4291 		device_delete_child(dev, devlist[i]);
4292 	}
4293 
4294 	free(devlist, M_TEMP);
4295 	oldcount = devcount;
4296 
4297 	/* Try to attach the devices just added. */
4298 	error = device_get_children(dev, &devlist, &devcount);
4299 	if (error) {
4300 		free(unchanged, M_TEMP);
4301 		return (error);
4302 	}
4303 
4304 	for (i = 0; i < devcount; i++) {
4305 		for (j = 0; j < oldcount; j++) {
4306 			if (devlist[i] == unchanged[j])
4307 				goto next_device;
4308 		}
4309 
4310 		device_probe_and_attach(devlist[i]);
4311 	next_device:;
4312 	}
4313 
4314 	free(unchanged, M_TEMP);
4315 	free(devlist, M_TEMP);
4316 	return (0);
4317 #undef REG
4318 }
4319 
4320 #ifdef PCI_IOV
4321 device_t
pci_add_iov_child(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4322 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4323     uint16_t did)
4324 {
4325 	struct pci_devinfo *vf_dinfo;
4326 	device_t pcib;
4327 	int busno, slot, func;
4328 
4329 	pcib = device_get_parent(bus);
4330 
4331 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4332 
4333 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4334 	    slot, func, vid, did);
4335 
4336 	vf_dinfo->cfg.flags |= PCICFG_VF;
4337 	pci_add_child(bus, vf_dinfo);
4338 
4339 	return (vf_dinfo->cfg.dev);
4340 }
4341 
4342 device_t
pci_create_iov_child_method(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4343 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4344     uint16_t vid, uint16_t did)
4345 {
4346 
4347 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4348 }
4349 #endif
4350 
4351 /*
4352  * For PCIe device set Max_Payload_Size to match PCIe root's.
4353  */
4354 static void
pcie_setup_mps(device_t dev)4355 pcie_setup_mps(device_t dev)
4356 {
4357 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4358 	device_t root;
4359 	uint16_t rmps, mmps, mps;
4360 
4361 	if (dinfo->cfg.pcie.pcie_location == 0)
4362 		return;
4363 	root = pci_find_pcie_root_port(dev);
4364 	if (root == NULL)
4365 		return;
4366 	/* Check whether the MPS is already configured. */
4367 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4368 	    PCIEM_CTL_MAX_PAYLOAD;
4369 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4370 	    PCIEM_CTL_MAX_PAYLOAD;
4371 	if (mps == rmps)
4372 		return;
4373 	/* Check whether the device is capable of the root's MPS. */
4374 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4375 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4376 	if (rmps > mmps) {
4377 		/*
4378 		 * The device is unable to handle root's MPS.  Limit root.
4379 		 * XXX: We should traverse through all the tree, applying
4380 		 * it to all the devices.
4381 		 */
4382 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4383 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4384 	} else {
4385 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4386 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4387 	}
4388 }
4389 
4390 static void
pci_add_child_clear_aer(device_t dev,struct pci_devinfo * dinfo)4391 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4392 {
4393 	int aer;
4394 	uint32_t r;
4395 	uint16_t r2;
4396 
4397 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4398 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4399 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4400 		    PCIER_ROOT_CTL, 2);
4401 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4402 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4403 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4404 		    PCIER_ROOT_CTL, r2, 2);
4405 	}
4406 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4407 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4408 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4409 		if (r != 0 && bootverbose) {
4410 			pci_printf(&dinfo->cfg,
4411 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4412 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4413 			    4));
4414 		}
4415 
4416 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4417 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4418 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4419 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4420 		    PCIM_AER_UC_POISONED_TLP |
4421 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4422 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4423 		    PCIM_AER_UC_COMPLETER_ABORT |
4424 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4425 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4426 		    PCIM_AER_UC_MALFORMED_TLP |
4427 		    PCIM_AER_UC_ECRC_ERROR |
4428 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4429 		    PCIM_AER_UC_ACS_VIOLATION |
4430 		    PCIM_AER_UC_INTERNAL_ERROR |
4431 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4432 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4433 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4434 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4435 
4436 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4437 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4438 		if (r != 0 && bootverbose) {
4439 			pci_printf(&dinfo->cfg,
4440 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4441 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4442 			    4));
4443 		}
4444 
4445 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4446 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4447 		    PCIM_AER_COR_BAD_TLP |
4448 		    PCIM_AER_COR_BAD_DLLP |
4449 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4450 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4451 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4452 		    PCIM_AER_COR_INTERNAL_ERROR |
4453 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4454 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4455 
4456 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4457 		    PCIER_DEVICE_CTL, 2);
4458 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4459 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4460 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4461 		    PCIER_DEVICE_CTL, r, 2);
4462 	}
4463 }
4464 
4465 void
pci_add_child(device_t bus,struct pci_devinfo * dinfo)4466 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4467 {
4468 	device_t dev;
4469 
4470 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4471 	device_set_ivars(dev, dinfo);
4472 	resource_list_init(&dinfo->resources);
4473 	pci_cfg_save(dev, dinfo, 0);
4474 	pci_cfg_restore(dev, dinfo);
4475 	pci_print_verbose(dinfo);
4476 	pci_add_resources(bus, dev, 0, 0);
4477 	if (pci_enable_mps_tune)
4478 		pcie_setup_mps(dev);
4479 	pci_child_added(dinfo->cfg.dev);
4480 
4481 	if (pci_clear_aer_on_attach)
4482 		pci_add_child_clear_aer(dev, dinfo);
4483 
4484 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4485 }
4486 
4487 void
pci_child_added_method(device_t dev,device_t child)4488 pci_child_added_method(device_t dev, device_t child)
4489 {
4490 
4491 }
4492 
4493 static int
pci_probe(device_t dev)4494 pci_probe(device_t dev)
4495 {
4496 
4497 	device_set_desc(dev, "PCI bus");
4498 
4499 	/* Allow other subclasses to override this driver. */
4500 	return (BUS_PROBE_GENERIC);
4501 }
4502 
4503 int
pci_attach_common(device_t dev)4504 pci_attach_common(device_t dev)
4505 {
4506 	struct pci_softc *sc;
4507 	int busno, domain;
4508 	int rid;
4509 
4510 	sc = device_get_softc(dev);
4511 	domain = pcib_get_domain(dev);
4512 	busno = pcib_get_bus(dev);
4513 	rid = 0;
4514 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4515 	    1, 0);
4516 	if (sc->sc_bus == NULL) {
4517 		device_printf(dev, "failed to allocate bus number\n");
4518 		return (ENXIO);
4519 	}
4520 	if (bootverbose)
4521 		device_printf(dev, "domain=%d, physical bus=%d\n",
4522 		    domain, busno);
4523 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4524 	return (0);
4525 }
4526 
4527 int
pci_attach(device_t dev)4528 pci_attach(device_t dev)
4529 {
4530 	int busno, domain, error;
4531 
4532 	error = pci_attach_common(dev);
4533 	if (error)
4534 		return (error);
4535 
4536 	/*
4537 	 * Since there can be multiple independently numbered PCI
4538 	 * buses on systems with multiple PCI domains, we can't use
4539 	 * the unit number to decide which bus we are probing. We ask
4540 	 * the parent pcib what our domain and bus numbers are.
4541 	 */
4542 	domain = pcib_get_domain(dev);
4543 	busno = pcib_get_bus(dev);
4544 	pci_add_children(dev, domain, busno);
4545 	bus_attach_children(dev);
4546 	return (0);
4547 }
4548 
4549 int
pci_detach(device_t dev)4550 pci_detach(device_t dev)
4551 {
4552 	struct pci_softc *sc;
4553 	int error;
4554 
4555 	error = bus_generic_detach(dev);
4556 	if (error)
4557 		return (error);
4558 	sc = device_get_softc(dev);
4559 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4560 	return (error);
4561 }
4562 
4563 static void
pci_hint_device_unit(device_t dev,device_t child,const char * name,int * unitp)4564 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4565 {
4566 	int line, unit;
4567 	const char *at;
4568 	char me1[24], me2[32];
4569 	uint8_t b, s, f;
4570 	uint32_t d;
4571 	device_location_cache_t *cache;
4572 
4573 	d = pci_get_domain(child);
4574 	b = pci_get_bus(child);
4575 	s = pci_get_slot(child);
4576 	f = pci_get_function(child);
4577 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4578 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4579 	line = 0;
4580 	cache = dev_wired_cache_init();
4581 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4582 		resource_string_value(name, unit, "at", &at);
4583 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4584 			*unitp = unit;
4585 			break;
4586 		}
4587 		if (dev_wired_cache_match(cache, child, at)) {
4588 			*unitp = unit;
4589 			break;
4590 		}
4591 	}
4592 	dev_wired_cache_fini(cache);
4593 }
4594 
4595 static void
pci_set_power_child(device_t dev,device_t child,int state)4596 pci_set_power_child(device_t dev, device_t child, int state)
4597 {
4598 	device_t pcib;
4599 	int dstate;
4600 
4601 	/*
4602 	 * Set the device to the given state.  If the firmware suggests
4603 	 * a different power state, use it instead.  If power management
4604 	 * is not present, the firmware is responsible for managing
4605 	 * device power.  Skip children who aren't attached since they
4606 	 * are handled separately.
4607 	 */
4608 	pcib = device_get_parent(dev);
4609 	dstate = state;
4610 	if (device_is_attached(child) &&
4611 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4612 		pci_set_powerstate(child, dstate);
4613 }
4614 
4615 int
pci_suspend_child(device_t dev,device_t child)4616 pci_suspend_child(device_t dev, device_t child)
4617 {
4618 	struct pci_devinfo *dinfo;
4619 	struct resource_list_entry *rle;
4620 	int error;
4621 
4622 	dinfo = device_get_ivars(child);
4623 
4624 	/*
4625 	 * Save the PCI configuration space for the child and set the
4626 	 * device in the appropriate power state for this sleep state.
4627 	 */
4628 	pci_cfg_save(child, dinfo, 0);
4629 
4630 	/* Suspend devices before potentially powering them down. */
4631 	error = bus_generic_suspend_child(dev, child);
4632 
4633 	if (error)
4634 		return (error);
4635 
4636 	if (pci_do_power_suspend) {
4637 		/*
4638 		 * Make sure this device's interrupt handler is not invoked
4639 		 * in the case the device uses a shared interrupt that can
4640 		 * be raised by some other device.
4641 		 * This is applicable only to regular (legacy) PCI interrupts
4642 		 * as MSI/MSI-X interrupts are never shared.
4643 		 */
4644 		rle = resource_list_find(&dinfo->resources,
4645 		    SYS_RES_IRQ, 0);
4646 		if (rle != NULL && rle->res != NULL)
4647 			(void)bus_suspend_intr(child, rle->res);
4648 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4649 	}
4650 
4651 	return (0);
4652 }
4653 
4654 int
pci_resume_child(device_t dev,device_t child)4655 pci_resume_child(device_t dev, device_t child)
4656 {
4657 	struct pci_devinfo *dinfo;
4658 	struct resource_list_entry *rle;
4659 
4660 	if (pci_do_power_resume)
4661 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4662 
4663 	dinfo = device_get_ivars(child);
4664 	pci_cfg_restore(child, dinfo);
4665 	if (!device_is_attached(child))
4666 		pci_cfg_save(child, dinfo, 1);
4667 
4668 	bus_generic_resume_child(dev, child);
4669 
4670 	/*
4671 	 * Allow interrupts only after fully resuming the driver and hardware.
4672 	 */
4673 	if (pci_do_power_suspend) {
4674 		/* See pci_suspend_child for details. */
4675 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4676 		if (rle != NULL && rle->res != NULL)
4677 			(void)bus_resume_intr(child, rle->res);
4678 	}
4679 
4680 	return (0);
4681 }
4682 
4683 int
pci_resume(device_t dev)4684 pci_resume(device_t dev)
4685 {
4686 	device_t child, *devlist;
4687 	int error, i, numdevs;
4688 
4689 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4690 		return (error);
4691 
4692 	/*
4693 	 * Resume critical devices first, then everything else later.
4694 	 */
4695 	for (i = 0; i < numdevs; i++) {
4696 		child = devlist[i];
4697 		switch (pci_get_class(child)) {
4698 		case PCIC_DISPLAY:
4699 		case PCIC_MEMORY:
4700 		case PCIC_BRIDGE:
4701 		case PCIC_BASEPERIPH:
4702 			BUS_RESUME_CHILD(dev, child);
4703 			break;
4704 		}
4705 	}
4706 	for (i = 0; i < numdevs; i++) {
4707 		child = devlist[i];
4708 		switch (pci_get_class(child)) {
4709 		case PCIC_DISPLAY:
4710 		case PCIC_MEMORY:
4711 		case PCIC_BRIDGE:
4712 		case PCIC_BASEPERIPH:
4713 			break;
4714 		default:
4715 			BUS_RESUME_CHILD(dev, child);
4716 		}
4717 	}
4718 	free(devlist, M_TEMP);
4719 	return (0);
4720 }
4721 
4722 static void
pci_load_vendor_data(void)4723 pci_load_vendor_data(void)
4724 {
4725 	caddr_t data;
4726 	void *ptr;
4727 	size_t sz;
4728 
4729 	data = preload_search_by_type("pci_vendor_data");
4730 	if (data != NULL) {
4731 		ptr = preload_fetch_addr(data);
4732 		sz = preload_fetch_size(data);
4733 		if (ptr != NULL && sz != 0) {
4734 			pci_vendordata = ptr;
4735 			pci_vendordata_size = sz;
4736 			/* terminate the database */
4737 			pci_vendordata[pci_vendordata_size] = '\n';
4738 		}
4739 	}
4740 }
4741 
4742 void
pci_driver_added(device_t dev,driver_t * driver)4743 pci_driver_added(device_t dev, driver_t *driver)
4744 {
4745 	int numdevs;
4746 	device_t *devlist;
4747 	device_t child;
4748 	struct pci_devinfo *dinfo;
4749 	int i;
4750 
4751 	if (bootverbose)
4752 		device_printf(dev, "driver added\n");
4753 	DEVICE_IDENTIFY(driver, dev);
4754 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4755 		return;
4756 	for (i = 0; i < numdevs; i++) {
4757 		child = devlist[i];
4758 		if (device_get_state(child) != DS_NOTPRESENT)
4759 			continue;
4760 		dinfo = device_get_ivars(child);
4761 		pci_print_verbose(dinfo);
4762 		if (bootverbose)
4763 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4764 		pci_cfg_restore(child, dinfo);
4765 		if (device_probe_and_attach(child) != 0)
4766 			pci_child_detached(dev, child);
4767 	}
4768 	free(devlist, M_TEMP);
4769 }
4770 
4771 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)4772 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4773     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4774 {
4775 	struct pci_devinfo *dinfo;
4776 	struct msix_table_entry *mte;
4777 	struct msix_vector *mv;
4778 	uint64_t addr;
4779 	uint32_t data;
4780 	void *cookie;
4781 	int error, rid;
4782 
4783 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4784 	    arg, &cookie);
4785 	if (error)
4786 		return (error);
4787 
4788 	/* If this is not a direct child, just bail out. */
4789 	if (device_get_parent(child) != dev) {
4790 		*cookiep = cookie;
4791 		return(0);
4792 	}
4793 
4794 	rid = rman_get_rid(irq);
4795 	if (rid == 0) {
4796 		/* Make sure that INTx is enabled */
4797 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4798 	} else {
4799 		/*
4800 		 * Check to see if the interrupt is MSI or MSI-X.
4801 		 * Ask our parent to map the MSI and give
4802 		 * us the address and data register values.
4803 		 * If we fail for some reason, teardown the
4804 		 * interrupt handler.
4805 		 */
4806 		dinfo = device_get_ivars(child);
4807 		if (dinfo->cfg.msi.msi_alloc > 0) {
4808 			if (dinfo->cfg.msi.msi_addr == 0) {
4809 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4810 			    ("MSI has handlers, but vectors not mapped"));
4811 				error = PCIB_MAP_MSI(device_get_parent(dev),
4812 				    child, rman_get_start(irq), &addr, &data);
4813 				if (error)
4814 					goto bad;
4815 				dinfo->cfg.msi.msi_addr = addr;
4816 				dinfo->cfg.msi.msi_data = data;
4817 			}
4818 			if (dinfo->cfg.msi.msi_handlers == 0)
4819 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4820 				    dinfo->cfg.msi.msi_data);
4821 			dinfo->cfg.msi.msi_handlers++;
4822 		} else {
4823 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4824 			    ("No MSI or MSI-X interrupts allocated"));
4825 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4826 			    ("MSI-X index too high"));
4827 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4828 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4829 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4830 			KASSERT(mv->mv_irq == rman_get_start(irq),
4831 			    ("IRQ mismatch"));
4832 			if (mv->mv_address == 0) {
4833 				KASSERT(mte->mte_handlers == 0,
4834 		    ("MSI-X table entry has handlers, but vector not mapped"));
4835 				error = PCIB_MAP_MSI(device_get_parent(dev),
4836 				    child, rman_get_start(irq), &addr, &data);
4837 				if (error)
4838 					goto bad;
4839 				mv->mv_address = addr;
4840 				mv->mv_data = data;
4841 			}
4842 
4843 			/*
4844 			 * The MSIX table entry must be made valid by
4845 			 * incrementing the mte_handlers before
4846 			 * calling pci_enable_msix() and
4847 			 * pci_resume_msix(). Else the MSIX rewrite
4848 			 * table quirk will not work as expected.
4849 			 */
4850 			mte->mte_handlers++;
4851 			if (mte->mte_handlers == 1) {
4852 				pci_enable_msix(child, rid - 1, mv->mv_address,
4853 				    mv->mv_data);
4854 				pci_unmask_msix(child, rid - 1);
4855 			}
4856 		}
4857 
4858 		/*
4859 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4860 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4861 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4862 		 */
4863 		if (!pci_has_quirk(pci_get_devid(child),
4864 		    PCI_QUIRK_MSI_INTX_BUG))
4865 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4866 		else
4867 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4868 	bad:
4869 		if (error) {
4870 			(void)bus_generic_teardown_intr(dev, child, irq,
4871 			    cookie);
4872 			return (error);
4873 		}
4874 	}
4875 	*cookiep = cookie;
4876 	return (0);
4877 }
4878 
4879 int
pci_teardown_intr(device_t dev,device_t child,struct resource * irq,void * cookie)4880 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4881     void *cookie)
4882 {
4883 	struct msix_table_entry *mte;
4884 	struct resource_list_entry *rle;
4885 	struct pci_devinfo *dinfo;
4886 	int error, rid;
4887 
4888 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4889 		return (EINVAL);
4890 
4891 	/* If this isn't a direct child, just bail out */
4892 	if (device_get_parent(child) != dev)
4893 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4894 
4895 	rid = rman_get_rid(irq);
4896 	if (rid == 0) {
4897 		/* Mask INTx */
4898 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4899 	} else {
4900 		/*
4901 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4902 		 * decrement the appropriate handlers count and mask the
4903 		 * MSI-X message, or disable MSI messages if the count
4904 		 * drops to 0.
4905 		 */
4906 		dinfo = device_get_ivars(child);
4907 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4908 		if (rle->res != irq)
4909 			return (EINVAL);
4910 		if (dinfo->cfg.msi.msi_alloc > 0) {
4911 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4912 			    ("MSI-X index too high"));
4913 			if (dinfo->cfg.msi.msi_handlers == 0)
4914 				return (EINVAL);
4915 			dinfo->cfg.msi.msi_handlers--;
4916 			if (dinfo->cfg.msi.msi_handlers == 0)
4917 				pci_disable_msi(child);
4918 		} else {
4919 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4920 			    ("No MSI or MSI-X interrupts allocated"));
4921 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4922 			    ("MSI-X index too high"));
4923 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4924 			if (mte->mte_handlers == 0)
4925 				return (EINVAL);
4926 			mte->mte_handlers--;
4927 			if (mte->mte_handlers == 0)
4928 				pci_mask_msix(child, rid - 1);
4929 		}
4930 	}
4931 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4932 	if (rid > 0)
4933 		KASSERT(error == 0,
4934 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4935 	return (error);
4936 }
4937 
4938 int
pci_print_child(device_t dev,device_t child)4939 pci_print_child(device_t dev, device_t child)
4940 {
4941 	struct pci_devinfo *dinfo;
4942 	struct resource_list *rl;
4943 	int retval = 0;
4944 
4945 	dinfo = device_get_ivars(child);
4946 	rl = &dinfo->resources;
4947 
4948 	retval += bus_print_child_header(dev, child);
4949 
4950 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4951 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4952 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4953 	if (device_get_flags(dev))
4954 		retval += printf(" flags %#x", device_get_flags(dev));
4955 
4956 	retval += printf(" at device %d.%d", pci_get_slot(child),
4957 	    pci_get_function(child));
4958 
4959 	retval += bus_print_child_domain(dev, child);
4960 	retval += bus_print_child_footer(dev, child);
4961 
4962 	return (retval);
4963 }
4964 
4965 static const struct
4966 {
4967 	int		class;
4968 	int		subclass;
4969 	int		report; /* 0 = bootverbose, 1 = always */
4970 	const char	*desc;
4971 } pci_nomatch_tab[] = {
4972 	{PCIC_OLD,		-1,			1, "old"},
4973 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4974 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4975 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4976 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4977 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4978 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4979 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4980 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4981 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4982 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4983 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4984 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4985 	{PCIC_NETWORK,		-1,			1, "network"},
4986 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4987 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4988 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4989 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4990 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4991 	{PCIC_DISPLAY,		-1,			1, "display"},
4992 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4993 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4994 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4995 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4996 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4997 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4998 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4999 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
5000 	{PCIC_MEMORY,		-1,			1, "memory"},
5001 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
5002 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
5003 	{PCIC_BRIDGE,		-1,			1, "bridge"},
5004 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
5005 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
5006 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
5007 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
5008 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
5009 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
5010 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
5011 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
5012 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
5013 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
5014 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
5015 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
5016 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
5017 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
5018 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
5019 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
5020 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
5021 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
5022 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
5023 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
5024 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
5025 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
5026 	{PCIC_INPUTDEV,		-1,			1, "input device"},
5027 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
5028 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
5029 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
5030 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
5031 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5032 	{PCIC_DOCKING,		-1,			1, "docking station"},
5033 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5034 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5035 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5036 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5037 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5038 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5039 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5040 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5041 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5042 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5043 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5044 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5045 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5046 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5047 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5048 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5049 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5050 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5051 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5052 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5053 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5054 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5055 	{PCIC_DASP,		-1,			0, "dasp"},
5056 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5057 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5058 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5059 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5060 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5061 	{0, 0, 0,		NULL}
5062 };
5063 
5064 void
pci_probe_nomatch(device_t dev,device_t child)5065 pci_probe_nomatch(device_t dev, device_t child)
5066 {
5067 	int i, report;
5068 	const char *cp, *scp;
5069 	char *device;
5070 
5071 	/*
5072 	 * Look for a listing for this device in a loaded device database.
5073 	 */
5074 	report = 1;
5075 	if ((device = pci_describe_device(child)) != NULL) {
5076 		device_printf(dev, "<%s>", device);
5077 		free(device, M_DEVBUF);
5078 	} else {
5079 		/*
5080 		 * Scan the class/subclass descriptions for a general
5081 		 * description.
5082 		 */
5083 		cp = "unknown";
5084 		scp = NULL;
5085 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5086 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5087 				if (pci_nomatch_tab[i].subclass == -1) {
5088 					cp = pci_nomatch_tab[i].desc;
5089 					report = pci_nomatch_tab[i].report;
5090 				} else if (pci_nomatch_tab[i].subclass ==
5091 				    pci_get_subclass(child)) {
5092 					scp = pci_nomatch_tab[i].desc;
5093 					report = pci_nomatch_tab[i].report;
5094 				}
5095 			}
5096 		}
5097 		if (report || bootverbose) {
5098 			device_printf(dev, "<%s%s%s>",
5099 			    cp ? cp : "",
5100 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5101 			    scp ? scp : "");
5102 		}
5103 	}
5104 	if (report || bootverbose) {
5105 		printf(" at device %d.%d (no driver attached)\n",
5106 		    pci_get_slot(child), pci_get_function(child));
5107 	}
5108 	pci_cfg_save(child, device_get_ivars(child), 1);
5109 }
5110 
5111 void
pci_child_detached(device_t dev,device_t child)5112 pci_child_detached(device_t dev, device_t child)
5113 {
5114 	struct pci_devinfo *dinfo;
5115 	struct resource_list *rl;
5116 
5117 	dinfo = device_get_ivars(child);
5118 	rl = &dinfo->resources;
5119 
5120 	/*
5121 	 * Have to deallocate IRQs before releasing any MSI messages and
5122 	 * have to release MSI messages before deallocating any memory
5123 	 * BARs.
5124 	 */
5125 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5126 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5127 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5128 		if (dinfo->cfg.msi.msi_alloc != 0)
5129 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5130 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5131 		else
5132 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5133 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5134 		(void)pci_release_msi(child);
5135 	}
5136 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5137 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5138 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5139 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5140 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5141 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5142 
5143 	pci_cfg_save(child, dinfo, 1);
5144 }
5145 
5146 /*
5147  * Parse the PCI device database, if loaded, and return a pointer to a
5148  * description of the device.
5149  *
5150  * The database is flat text formatted as follows:
5151  *
5152  * Any line not in a valid format is ignored.
5153  * Lines are terminated with newline '\n' characters.
5154  *
5155  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5156  * the vendor name.
5157  *
5158  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5159  * - devices cannot be listed without a corresponding VENDOR line.
5160  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5161  * another TAB, then the device name.
5162  */
5163 
5164 /*
5165  * Assuming (ptr) points to the beginning of a line in the database,
5166  * return the vendor or device and description of the next entry.
5167  * The value of (vendor) or (device) inappropriate for the entry type
5168  * is set to -1.  Returns nonzero at the end of the database.
5169  *
5170  * Note that this is slightly unrobust in the face of corrupt data;
5171  * we attempt to safeguard against this by spamming the end of the
5172  * database with a newline when we initialise.
5173  */
5174 static int
pci_describe_parse_line(char ** ptr,int * vendor,int * device,char ** desc)5175 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5176 {
5177 	char	*cp = *ptr;
5178 	int	left;
5179 
5180 	*device = -1;
5181 	*vendor = -1;
5182 	**desc = '\0';
5183 	for (;;) {
5184 		left = pci_vendordata_size - (cp - pci_vendordata);
5185 		if (left <= 0) {
5186 			*ptr = cp;
5187 			return(1);
5188 		}
5189 
5190 		/* vendor entry? */
5191 		if (*cp != '\t' &&
5192 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5193 			break;
5194 		/* device entry? */
5195 		if (*cp == '\t' &&
5196 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5197 			break;
5198 
5199 		/* skip to next line */
5200 		while (*cp != '\n' && left > 0) {
5201 			cp++;
5202 			left--;
5203 		}
5204 		if (*cp == '\n') {
5205 			cp++;
5206 			left--;
5207 		}
5208 	}
5209 	/* skip to next line */
5210 	while (*cp != '\n' && left > 0) {
5211 		cp++;
5212 		left--;
5213 	}
5214 	if (*cp == '\n' && left > 0)
5215 		cp++;
5216 	*ptr = cp;
5217 	return(0);
5218 }
5219 
5220 static char *
pci_describe_device(device_t dev)5221 pci_describe_device(device_t dev)
5222 {
5223 	int	vendor, device;
5224 	char	*desc, *vp, *dp, *line;
5225 
5226 	desc = vp = dp = NULL;
5227 
5228 	/*
5229 	 * If we have no vendor data, we can't do anything.
5230 	 */
5231 	if (pci_vendordata == NULL)
5232 		goto out;
5233 
5234 	/*
5235 	 * Scan the vendor data looking for this device
5236 	 */
5237 	line = pci_vendordata;
5238 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5239 		goto out;
5240 	for (;;) {
5241 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5242 			goto out;
5243 		if (vendor == pci_get_vendor(dev))
5244 			break;
5245 	}
5246 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5247 		goto out;
5248 	for (;;) {
5249 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5250 			*dp = 0;
5251 			break;
5252 		}
5253 		if (vendor != -1) {
5254 			*dp = 0;
5255 			break;
5256 		}
5257 		if (device == pci_get_device(dev))
5258 			break;
5259 	}
5260 	if (dp[0] == '\0')
5261 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5262 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5263 	    NULL)
5264 		sprintf(desc, "%s, %s", vp, dp);
5265 out:
5266 	if (vp != NULL)
5267 		free(vp, M_DEVBUF);
5268 	if (dp != NULL)
5269 		free(dp, M_DEVBUF);
5270 	return(desc);
5271 }
5272 
5273 int
pci_read_ivar(device_t dev,device_t child,int which,uintptr_t * result)5274 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5275 {
5276 	struct pci_devinfo *dinfo;
5277 	pcicfgregs *cfg;
5278 
5279 	dinfo = device_get_ivars(child);
5280 	cfg = &dinfo->cfg;
5281 
5282 	switch (which) {
5283 	case PCI_IVAR_ETHADDR:
5284 		/*
5285 		 * The generic accessor doesn't deal with failure, so
5286 		 * we set the return value, then return an error.
5287 		 */
5288 		*((uint8_t **) result) = NULL;
5289 		return (EINVAL);
5290 	case PCI_IVAR_SUBVENDOR:
5291 		*result = cfg->subvendor;
5292 		break;
5293 	case PCI_IVAR_SUBDEVICE:
5294 		*result = cfg->subdevice;
5295 		break;
5296 	case PCI_IVAR_VENDOR:
5297 		*result = cfg->vendor;
5298 		break;
5299 	case PCI_IVAR_DEVICE:
5300 		*result = cfg->device;
5301 		break;
5302 	case PCI_IVAR_DEVID:
5303 		*result = (cfg->device << 16) | cfg->vendor;
5304 		break;
5305 	case PCI_IVAR_CLASS:
5306 		*result = cfg->baseclass;
5307 		break;
5308 	case PCI_IVAR_SUBCLASS:
5309 		*result = cfg->subclass;
5310 		break;
5311 	case PCI_IVAR_PROGIF:
5312 		*result = cfg->progif;
5313 		break;
5314 	case PCI_IVAR_REVID:
5315 		*result = cfg->revid;
5316 		break;
5317 	case PCI_IVAR_INTPIN:
5318 		*result = cfg->intpin;
5319 		break;
5320 	case PCI_IVAR_IRQ:
5321 		*result = cfg->intline;
5322 		break;
5323 	case PCI_IVAR_DOMAIN:
5324 		*result = cfg->domain;
5325 		break;
5326 	case PCI_IVAR_BUS:
5327 		*result = cfg->bus;
5328 		break;
5329 	case PCI_IVAR_SLOT:
5330 		*result = cfg->slot;
5331 		break;
5332 	case PCI_IVAR_FUNCTION:
5333 		*result = cfg->func;
5334 		break;
5335 	case PCI_IVAR_CMDREG:
5336 		*result = cfg->cmdreg;
5337 		break;
5338 	case PCI_IVAR_CACHELNSZ:
5339 		*result = cfg->cachelnsz;
5340 		break;
5341 	case PCI_IVAR_MINGNT:
5342 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5343 			*result = -1;
5344 			return (EINVAL);
5345 		}
5346 		*result = cfg->mingnt;
5347 		break;
5348 	case PCI_IVAR_MAXLAT:
5349 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5350 			*result = -1;
5351 			return (EINVAL);
5352 		}
5353 		*result = cfg->maxlat;
5354 		break;
5355 	case PCI_IVAR_LATTIMER:
5356 		*result = cfg->lattimer;
5357 		break;
5358 	default:
5359 		return (ENOENT);
5360 	}
5361 	return (0);
5362 }
5363 
5364 int
pci_write_ivar(device_t dev,device_t child,int which,uintptr_t value)5365 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5366 {
5367 	struct pci_devinfo *dinfo;
5368 
5369 	dinfo = device_get_ivars(child);
5370 
5371 	switch (which) {
5372 	case PCI_IVAR_INTPIN:
5373 		dinfo->cfg.intpin = value;
5374 		return (0);
5375 	case PCI_IVAR_ETHADDR:
5376 	case PCI_IVAR_SUBVENDOR:
5377 	case PCI_IVAR_SUBDEVICE:
5378 	case PCI_IVAR_VENDOR:
5379 	case PCI_IVAR_DEVICE:
5380 	case PCI_IVAR_DEVID:
5381 	case PCI_IVAR_CLASS:
5382 	case PCI_IVAR_SUBCLASS:
5383 	case PCI_IVAR_PROGIF:
5384 	case PCI_IVAR_REVID:
5385 	case PCI_IVAR_IRQ:
5386 	case PCI_IVAR_DOMAIN:
5387 	case PCI_IVAR_BUS:
5388 	case PCI_IVAR_SLOT:
5389 	case PCI_IVAR_FUNCTION:
5390 		return (EINVAL);	/* disallow for now */
5391 
5392 	default:
5393 		return (ENOENT);
5394 	}
5395 }
5396 
5397 #include "opt_ddb.h"
5398 #ifdef DDB
5399 #include <ddb/ddb.h>
5400 #include <sys/cons.h>
5401 
5402 /*
5403  * List resources based on pci map registers, used for within ddb
5404  */
5405 
DB_SHOW_COMMAND_FLAGS(pciregs,db_pci_dump,DB_CMD_MEMSAFE)5406 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5407 {
5408 	struct pci_devinfo *dinfo;
5409 	struct devlist *devlist_head;
5410 	struct pci_conf *p;
5411 	const char *name;
5412 	int i, error, none_count;
5413 
5414 	none_count = 0;
5415 	/* get the head of the device queue */
5416 	devlist_head = &pci_devq;
5417 
5418 	/*
5419 	 * Go through the list of devices and print out devices
5420 	 */
5421 	for (error = 0, i = 0,
5422 	     dinfo = STAILQ_FIRST(devlist_head);
5423 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5424 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5425 		/* Populate pd_name and pd_unit */
5426 		name = NULL;
5427 		if (dinfo->cfg.dev)
5428 			name = device_get_name(dinfo->cfg.dev);
5429 
5430 		p = &dinfo->conf;
5431 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5432 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5433 			(name && *name) ? name : "none",
5434 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5435 			none_count++,
5436 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5437 			p->pc_sel.pc_func, (p->pc_class << 16) |
5438 			(p->pc_subclass << 8) | p->pc_progif,
5439 			(p->pc_subdevice << 16) | p->pc_subvendor,
5440 			(p->pc_device << 16) | p->pc_vendor,
5441 			p->pc_revid, p->pc_hdr);
5442 	}
5443 }
5444 #endif /* DDB */
5445 
5446 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)5447 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5448     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5449     u_int flags)
5450 {
5451 	struct pci_devinfo *dinfo = device_get_ivars(child);
5452 	struct resource_list *rl = &dinfo->resources;
5453 	struct resource *res;
5454 	struct pci_map *pm;
5455 	uint16_t cmd;
5456 	pci_addr_t map, testval;
5457 	int mapsize;
5458 
5459 	res = NULL;
5460 
5461 	/* If rid is managed by EA, ignore it */
5462 	if (pci_ea_is_enabled(child, *rid))
5463 		goto out;
5464 
5465 	pm = pci_find_bar(child, *rid);
5466 	if (pm != NULL) {
5467 		/* This is a BAR that we failed to allocate earlier. */
5468 		mapsize = pm->pm_size;
5469 		map = pm->pm_value;
5470 	} else {
5471 		/*
5472 		 * Weed out the bogons, and figure out how large the
5473 		 * BAR/map is.  BARs that read back 0 here are bogus
5474 		 * and unimplemented.  Note: atapci in legacy mode are
5475 		 * special and handled elsewhere in the code.  If you
5476 		 * have a atapci device in legacy mode and it fails
5477 		 * here, that other code is broken.
5478 		 */
5479 		pci_read_bar(child, *rid, &map, &testval, NULL);
5480 
5481 		/*
5482 		 * Determine the size of the BAR and ignore BARs with a size
5483 		 * of 0.  Device ROM BARs use a different mask value.
5484 		 */
5485 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5486 			mapsize = pci_romsize(testval);
5487 		else
5488 			mapsize = pci_mapsize(testval);
5489 		if (mapsize == 0)
5490 			goto out;
5491 		pm = pci_add_bar(child, *rid, map, mapsize);
5492 	}
5493 
5494 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5495 		if (type != SYS_RES_MEMORY) {
5496 			if (bootverbose)
5497 				device_printf(dev,
5498 				    "child %s requested type %d for rid %#x,"
5499 				    " but the BAR says it is an memio\n",
5500 				    device_get_nameunit(child), type, *rid);
5501 			goto out;
5502 		}
5503 	} else {
5504 		if (type != SYS_RES_IOPORT) {
5505 			if (bootverbose)
5506 				device_printf(dev,
5507 				    "child %s requested type %d for rid %#x,"
5508 				    " but the BAR says it is an ioport\n",
5509 				    device_get_nameunit(child), type, *rid);
5510 			goto out;
5511 		}
5512 	}
5513 
5514 	/*
5515 	 * For real BARs, we need to override the size that
5516 	 * the driver requests, because that's what the BAR
5517 	 * actually uses and we would otherwise have a
5518 	 * situation where we might allocate the excess to
5519 	 * another driver, which won't work.
5520 	 */
5521 	count = ((pci_addr_t)1 << mapsize) * num;
5522 	if (RF_ALIGNMENT(flags) < mapsize)
5523 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5524 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5525 		flags |= RF_PREFETCHABLE;
5526 
5527 	/*
5528 	 * Allocate enough resource, and then write back the
5529 	 * appropriate BAR for that resource.
5530 	 */
5531 	resource_list_add(rl, type, *rid, start, end, count);
5532 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5533 	    count, flags & ~RF_ACTIVE);
5534 	if (res == NULL) {
5535 		resource_list_delete(rl, type, *rid);
5536 		device_printf(child,
5537 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5538 		    count, *rid, type, start, end);
5539 		goto out;
5540 	}
5541 	if (bootverbose)
5542 		device_printf(child,
5543 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5544 		    count, *rid, type, rman_get_start(res));
5545 
5546 	/* Disable decoding via the CMD register before updating the BAR */
5547 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5548 	pci_write_config(child, PCIR_COMMAND,
5549 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5550 
5551 	map = rman_get_start(res);
5552 	pci_write_bar(child, pm, map);
5553 
5554 	/* Restore the original value of the CMD register */
5555 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5556 out:
5557 	return (res);
5558 }
5559 
5560 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)5561 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5562     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5563     u_int flags)
5564 {
5565 	struct pci_devinfo *dinfo;
5566 	struct resource_list *rl;
5567 	struct resource_list_entry *rle;
5568 	struct resource *res;
5569 	pcicfgregs *cfg;
5570 
5571 	/*
5572 	 * Perform lazy resource allocation
5573 	 */
5574 	dinfo = device_get_ivars(child);
5575 	rl = &dinfo->resources;
5576 	cfg = &dinfo->cfg;
5577 	switch (type) {
5578 	case PCI_RES_BUS:
5579 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5580 		    flags));
5581 	case SYS_RES_IRQ:
5582 		/*
5583 		 * Can't alloc legacy interrupt once MSI messages have
5584 		 * been allocated.
5585 		 */
5586 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5587 		    cfg->msix.msix_alloc > 0))
5588 			return (NULL);
5589 
5590 		/*
5591 		 * If the child device doesn't have an interrupt
5592 		 * routed and is deserving of an interrupt, try to
5593 		 * assign it one.
5594 		 */
5595 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5596 		    (cfg->intpin != 0))
5597 			pci_assign_interrupt(dev, child, 0);
5598 		break;
5599 	case SYS_RES_IOPORT:
5600 	case SYS_RES_MEMORY:
5601 		/*
5602 		 * PCI-PCI bridge I/O window resources are not BARs.
5603 		 * For those allocations just pass the request up the
5604 		 * tree.
5605 		 */
5606 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5607 			switch (*rid) {
5608 			case PCIR_IOBASEL_1:
5609 			case PCIR_MEMBASE_1:
5610 			case PCIR_PMBASEL_1:
5611 				/*
5612 				 * XXX: Should we bother creating a resource
5613 				 * list entry?
5614 				 */
5615 				return (bus_generic_alloc_resource(dev, child,
5616 				    type, rid, start, end, count, flags));
5617 			}
5618 		}
5619 		/* Reserve resources for this BAR if needed. */
5620 		rle = resource_list_find(rl, type, *rid);
5621 		if (rle == NULL) {
5622 			res = pci_reserve_map(dev, child, type, rid, start, end,
5623 			    count, num, flags);
5624 			if (res == NULL)
5625 				return (NULL);
5626 		}
5627 	}
5628 	return (resource_list_alloc(rl, dev, child, type, rid,
5629 	    start, end, count, flags));
5630 }
5631 
5632 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)5633 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5634     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5635 {
5636 #ifdef PCI_IOV
5637 	struct pci_devinfo *dinfo;
5638 #endif
5639 
5640 	if (device_get_parent(child) != dev)
5641 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5642 		    type, rid, start, end, count, flags));
5643 
5644 #ifdef PCI_IOV
5645 	dinfo = device_get_ivars(child);
5646 	if (dinfo->cfg.flags & PCICFG_VF) {
5647 		switch (type) {
5648 		/* VFs can't have I/O BARs. */
5649 		case SYS_RES_IOPORT:
5650 			return (NULL);
5651 		case SYS_RES_MEMORY:
5652 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5653 			    start, end, count, flags));
5654 		}
5655 
5656 		/* Fall through for other types of resource allocations. */
5657 	}
5658 #endif
5659 
5660 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5661 	    count, 1, flags));
5662 }
5663 
5664 int
pci_release_resource(device_t dev,device_t child,struct resource * r)5665 pci_release_resource(device_t dev, device_t child, struct resource *r)
5666 {
5667 	struct pci_devinfo *dinfo;
5668 	struct resource_list *rl;
5669 	pcicfgregs *cfg __unused;
5670 
5671 	if (device_get_parent(child) != dev)
5672 		return (bus_generic_release_resource(dev, child, r));
5673 
5674 	dinfo = device_get_ivars(child);
5675 	cfg = &dinfo->cfg;
5676 
5677 #ifdef PCI_IOV
5678 	if (cfg->flags & PCICFG_VF) {
5679 		switch (rman_get_type(r)) {
5680 		/* VFs can't have I/O BARs. */
5681 		case SYS_RES_IOPORT:
5682 			return (EDOOFUS);
5683 		case SYS_RES_MEMORY:
5684 			return (pci_vf_release_mem_resource(dev, child, r));
5685 		}
5686 
5687 		/* Fall through for other types of resource allocations. */
5688 	}
5689 #endif
5690 
5691 	/*
5692 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5693 	 * those allocations just pass the request up the tree.
5694 	 */
5695 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5696 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5697 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5698 		switch (rman_get_rid(r)) {
5699 		case PCIR_IOBASEL_1:
5700 		case PCIR_MEMBASE_1:
5701 		case PCIR_PMBASEL_1:
5702 			return (bus_generic_release_resource(dev, child, r));
5703 		}
5704 	}
5705 
5706 	rl = &dinfo->resources;
5707 	return (resource_list_release(rl, dev, child, r));
5708 }
5709 
5710 int
pci_activate_resource(device_t dev,device_t child,struct resource * r)5711 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5712 {
5713 	struct pci_devinfo *dinfo;
5714 	int error, rid, type;
5715 
5716 	if (device_get_parent(child) != dev)
5717 		return (bus_generic_activate_resource(dev, child, r));
5718 
5719 	dinfo = device_get_ivars(child);
5720 #ifdef PCI_IOV
5721 	if (dinfo->cfg.flags & PCICFG_VF) {
5722 		switch (rman_get_type(r)) {
5723 		/* VFs can't have I/O BARs. */
5724 		case SYS_RES_IOPORT:
5725 			error = EINVAL;
5726 			break;
5727 		case SYS_RES_MEMORY:
5728 			error = pci_vf_activate_mem_resource(dev, child, r);
5729 			break;
5730 		default:
5731 			error = bus_generic_activate_resource(dev, child, r);
5732 			break;
5733 		}
5734 	} else
5735 #endif
5736 		error = bus_generic_activate_resource(dev, child, r);
5737 	if (error)
5738 		return (error);
5739 
5740 	rid = rman_get_rid(r);
5741 	type = rman_get_type(r);
5742 
5743 	/* Device ROMs need their decoding explicitly enabled. */
5744 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5745 		pci_write_bar(child, pci_find_bar(child, rid),
5746 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5747 
5748 	/* Enable decoding in the command register when activating BARs. */
5749 	switch (type) {
5750 	case SYS_RES_IOPORT:
5751 	case SYS_RES_MEMORY:
5752 		error = PCI_ENABLE_IO(dev, child, type);
5753 		break;
5754 	}
5755 	return (error);
5756 }
5757 
5758 int
pci_deactivate_resource(device_t dev,device_t child,struct resource * r)5759 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5760 {
5761 	struct pci_devinfo *dinfo;
5762 	int error, rid, type;
5763 
5764 	if (device_get_parent(child) != dev)
5765 		return (bus_generic_deactivate_resource(dev, child, r));
5766 
5767 	dinfo = device_get_ivars(child);
5768 #ifdef PCI_IOV
5769 	if (dinfo->cfg.flags & PCICFG_VF) {
5770 		switch (rman_get_type(r)) {
5771 		/* VFs can't have I/O BARs. */
5772 		case SYS_RES_IOPORT:
5773 			error = EINVAL;
5774 			break;
5775 		case SYS_RES_MEMORY:
5776 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5777 			break;
5778 		default:
5779 			error = bus_generic_deactivate_resource(dev, child, r);
5780 			break;
5781 		}
5782 	} else
5783 #endif
5784 		error = bus_generic_deactivate_resource(dev, child, r);
5785 	if (error)
5786 		return (error);
5787 
5788 	/* Disable decoding for device ROMs. */
5789 	rid = rman_get_rid(r);
5790 	type = rman_get_type(r);
5791 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5792 		pci_write_bar(child, pci_find_bar(child, rid),
5793 		    rman_get_start(r));
5794 	return (0);
5795 }
5796 
5797 int
pci_adjust_resource(device_t dev,device_t child,struct resource * r,rman_res_t start,rman_res_t end)5798 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5799     rman_res_t start, rman_res_t end)
5800 {
5801 #ifdef PCI_IOV
5802 	struct pci_devinfo *dinfo;
5803 
5804 	if (device_get_parent(child) != dev)
5805 		return (bus_generic_adjust_resource(dev, child, r, start,
5806 		    end));
5807 
5808 	dinfo = device_get_ivars(child);
5809 	if (dinfo->cfg.flags & PCICFG_VF) {
5810 		switch (rman_get_type(r)) {
5811 		/* VFs can't have I/O BARs. */
5812 		case SYS_RES_IOPORT:
5813 			return (EINVAL);
5814 		case SYS_RES_MEMORY:
5815 			return (pci_vf_adjust_mem_resource(dev, child, r,
5816 			    start, end));
5817 		}
5818 
5819 		/* Fall through for other types of resource allocations. */
5820 	}
5821 #endif
5822 
5823 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5824 }
5825 
5826 int
pci_map_resource(device_t dev,device_t child,struct resource * r,struct resource_map_request * argsp,struct resource_map * map)5827 pci_map_resource(device_t dev, device_t child, struct resource *r,
5828     struct resource_map_request *argsp, struct resource_map *map)
5829 {
5830 #ifdef PCI_IOV
5831 	struct pci_devinfo *dinfo;
5832 
5833 	if (device_get_parent(child) != dev)
5834 		return (bus_generic_map_resource(dev, child, r, argsp,
5835 		    map));
5836 
5837 	dinfo = device_get_ivars(child);
5838 	if (dinfo->cfg.flags & PCICFG_VF) {
5839 		switch (rman_get_type(r)) {
5840 		/* VFs can't have I/O BARs. */
5841 		case SYS_RES_IOPORT:
5842 			return (EINVAL);
5843 		case SYS_RES_MEMORY:
5844 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5845 			    map));
5846 		}
5847 
5848 		/* Fall through for other types of resource allocations. */
5849 	}
5850 #endif
5851 
5852 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5853 }
5854 
5855 int
pci_unmap_resource(device_t dev,device_t child,struct resource * r,struct resource_map * map)5856 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5857     struct resource_map *map)
5858 {
5859 #ifdef PCI_IOV
5860 	struct pci_devinfo *dinfo;
5861 
5862 	if (device_get_parent(child) != dev)
5863 		return (bus_generic_unmap_resource(dev, child, r, map));
5864 
5865 	dinfo = device_get_ivars(child);
5866 	if (dinfo->cfg.flags & PCICFG_VF) {
5867 		switch (rman_get_type(r)) {
5868 		/* VFs can't have I/O BARs. */
5869 		case SYS_RES_IOPORT:
5870 			return (EINVAL);
5871 		case SYS_RES_MEMORY:
5872 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5873 		}
5874 
5875 		/* Fall through for other types of resource allocations. */
5876 	}
5877 #endif
5878 
5879 	return (bus_generic_unmap_resource(dev, child, r, map));
5880 }
5881 
5882 void
pci_child_deleted(device_t dev,device_t child)5883 pci_child_deleted(device_t dev, device_t child)
5884 {
5885 	struct resource_list_entry *rle;
5886 	struct resource_list *rl;
5887 	struct pci_devinfo *dinfo;
5888 
5889 	dinfo = device_get_ivars(child);
5890 	rl = &dinfo->resources;
5891 
5892 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5893 
5894 	/* Turn off access to resources we're about to free */
5895 	if (bus_child_present(child) != 0) {
5896 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5897 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5898 
5899 		pci_disable_busmaster(child);
5900 	}
5901 
5902 	/* Free all allocated resources */
5903 	STAILQ_FOREACH(rle, rl, link) {
5904 		if (rle->res) {
5905 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5906 			    resource_list_busy(rl, rle->type, rle->rid)) {
5907 				pci_printf(&dinfo->cfg,
5908 				    "Resource still owned, oops. "
5909 				    "(type=%d, rid=%d, addr=%lx)\n",
5910 				    rle->type, rle->rid,
5911 				    rman_get_start(rle->res));
5912 				bus_release_resource(child, rle->type, rle->rid,
5913 				    rle->res);
5914 			}
5915 			resource_list_unreserve(rl, dev, child, rle->type,
5916 			    rle->rid);
5917 		}
5918 	}
5919 	resource_list_free(rl);
5920 
5921 	pci_freecfg(dinfo);
5922 }
5923 
5924 void
pci_delete_resource(device_t dev,device_t child,int type,int rid)5925 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5926 {
5927 	struct pci_devinfo *dinfo;
5928 	struct resource_list *rl;
5929 	struct resource_list_entry *rle;
5930 
5931 	if (device_get_parent(child) != dev)
5932 		return;
5933 
5934 	dinfo = device_get_ivars(child);
5935 	rl = &dinfo->resources;
5936 	rle = resource_list_find(rl, type, rid);
5937 	if (rle == NULL)
5938 		return;
5939 
5940 	if (rle->res) {
5941 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5942 		    resource_list_busy(rl, type, rid)) {
5943 			device_printf(dev, "delete_resource: "
5944 			    "Resource still owned by child, oops. "
5945 			    "(type=%d, rid=%d, addr=%jx)\n",
5946 			    type, rid, rman_get_start(rle->res));
5947 			return;
5948 		}
5949 		resource_list_unreserve(rl, dev, child, type, rid);
5950 	}
5951 	resource_list_delete(rl, type, rid);
5952 }
5953 
5954 struct resource_list *
pci_get_resource_list(device_t dev,device_t child)5955 pci_get_resource_list (device_t dev, device_t child)
5956 {
5957 	struct pci_devinfo *dinfo = device_get_ivars(child);
5958 
5959 	return (&dinfo->resources);
5960 }
5961 
5962 #ifdef IOMMU
5963 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)5964 pci_get_dma_tag(device_t bus, device_t dev)
5965 {
5966 	bus_dma_tag_t tag;
5967 	struct pci_softc *sc;
5968 
5969 	if (device_get_parent(dev) == bus) {
5970 		/* try iommu and return if it works */
5971 		tag = iommu_get_dma_tag(bus, dev);
5972 	} else
5973 		tag = NULL;
5974 	if (tag == NULL) {
5975 		sc = device_get_softc(bus);
5976 		tag = sc->sc_dma_tag;
5977 	}
5978 	return (tag);
5979 }
5980 #else
5981 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)5982 pci_get_dma_tag(device_t bus, device_t dev)
5983 {
5984 	struct pci_softc *sc = device_get_softc(bus);
5985 
5986 	return (sc->sc_dma_tag);
5987 }
5988 #endif
5989 
5990 uint32_t
pci_read_config_method(device_t dev,device_t child,int reg,int width)5991 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5992 {
5993 	struct pci_devinfo *dinfo = device_get_ivars(child);
5994 	pcicfgregs *cfg = &dinfo->cfg;
5995 
5996 #ifdef PCI_IOV
5997 	/*
5998 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5999 	 * emulate them here.
6000 	 */
6001 	if (cfg->flags & PCICFG_VF) {
6002 		if (reg == PCIR_VENDOR) {
6003 			switch (width) {
6004 			case 4:
6005 				return (cfg->device << 16 | cfg->vendor);
6006 			case 2:
6007 				return (cfg->vendor);
6008 			case 1:
6009 				return (cfg->vendor & 0xff);
6010 			default:
6011 				return (0xffffffff);
6012 			}
6013 		} else if (reg == PCIR_DEVICE) {
6014 			switch (width) {
6015 			/* Note that an unaligned 4-byte read is an error. */
6016 			case 2:
6017 				return (cfg->device);
6018 			case 1:
6019 				return (cfg->device & 0xff);
6020 			default:
6021 				return (0xffffffff);
6022 			}
6023 		}
6024 	}
6025 #endif
6026 
6027 	return (PCIB_READ_CONFIG(device_get_parent(dev),
6028 	    cfg->bus, cfg->slot, cfg->func, reg, width));
6029 }
6030 
6031 void
pci_write_config_method(device_t dev,device_t child,int reg,uint32_t val,int width)6032 pci_write_config_method(device_t dev, device_t child, int reg,
6033     uint32_t val, int width)
6034 {
6035 	struct pci_devinfo *dinfo = device_get_ivars(child);
6036 	pcicfgregs *cfg = &dinfo->cfg;
6037 
6038 	PCIB_WRITE_CONFIG(device_get_parent(dev),
6039 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6040 }
6041 
6042 int
pci_child_location_method(device_t dev,device_t child,struct sbuf * sb)6043 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6044 {
6045 
6046 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6047 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6048 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6049 	return (0);
6050 }
6051 
6052 int
pci_child_pnpinfo_method(device_t dev,device_t child,struct sbuf * sb)6053 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6054 {
6055 	struct pci_devinfo *dinfo;
6056 	pcicfgregs *cfg;
6057 
6058 	dinfo = device_get_ivars(child);
6059 	cfg = &dinfo->cfg;
6060 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6061 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6062 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6063 	    cfg->progif);
6064 	return (0);
6065 }
6066 
6067 int
pci_get_device_path_method(device_t bus,device_t child,const char * locator,struct sbuf * sb)6068 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6069     struct sbuf *sb)
6070 {
6071 	device_t parent = device_get_parent(bus);
6072 	int rv;
6073 
6074 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6075 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6076 		if (rv == 0) {
6077 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6078 			    pci_get_function(child));
6079 		}
6080 		return (0);
6081 	}
6082 	return (bus_generic_get_device_path(bus, child, locator, sb));
6083 }
6084 
6085 int
pci_assign_interrupt_method(device_t dev,device_t child)6086 pci_assign_interrupt_method(device_t dev, device_t child)
6087 {
6088 	struct pci_devinfo *dinfo = device_get_ivars(child);
6089 	pcicfgregs *cfg = &dinfo->cfg;
6090 
6091 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6092 	    cfg->intpin));
6093 }
6094 
6095 static void
pci_lookup(void * arg,const char * name,device_t * dev)6096 pci_lookup(void *arg, const char *name, device_t *dev)
6097 {
6098 	long val;
6099 	char *end;
6100 	int domain, bus, slot, func;
6101 
6102 	if (*dev != NULL)
6103 		return;
6104 
6105 	/*
6106 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6107 	 * pciB:S:F.  In the latter case, the domain is assumed to
6108 	 * be zero.
6109 	 */
6110 	if (strncmp(name, "pci", 3) != 0)
6111 		return;
6112 	val = strtol(name + 3, &end, 10);
6113 	if (val < 0 || val > INT_MAX || *end != ':')
6114 		return;
6115 	domain = val;
6116 	val = strtol(end + 1, &end, 10);
6117 	if (val < 0 || val > INT_MAX || *end != ':')
6118 		return;
6119 	bus = val;
6120 	val = strtol(end + 1, &end, 10);
6121 	if (val < 0 || val > INT_MAX)
6122 		return;
6123 	slot = val;
6124 	if (*end == ':') {
6125 		val = strtol(end + 1, &end, 10);
6126 		if (val < 0 || val > INT_MAX || *end != '\0')
6127 			return;
6128 		func = val;
6129 	} else if (*end == '\0') {
6130 		func = slot;
6131 		slot = bus;
6132 		bus = domain;
6133 		domain = 0;
6134 	} else
6135 		return;
6136 
6137 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6138 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6139 		return;
6140 
6141 	*dev = pci_find_dbsf(domain, bus, slot, func);
6142 }
6143 
6144 static int
pci_modevent(module_t mod,int what,void * arg)6145 pci_modevent(module_t mod, int what, void *arg)
6146 {
6147 	static struct cdev *pci_cdev;
6148 	static eventhandler_tag tag;
6149 
6150 	switch (what) {
6151 	case MOD_LOAD:
6152 		STAILQ_INIT(&pci_devq);
6153 		pci_generation = 0;
6154 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6155 		    "pci");
6156 		pci_load_vendor_data();
6157 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6158 		    1000);
6159 		break;
6160 
6161 	case MOD_UNLOAD:
6162 		if (tag != NULL)
6163 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6164 		destroy_dev(pci_cdev);
6165 		break;
6166 	}
6167 
6168 	return (0);
6169 }
6170 
6171 static void
pci_cfg_restore_pcie(device_t dev,struct pci_devinfo * dinfo)6172 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6173 {
6174 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6175 	struct pcicfg_pcie *cfg;
6176 	int version, pos;
6177 
6178 	cfg = &dinfo->cfg.pcie;
6179 	pos = cfg->pcie_location;
6180 
6181 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6182 
6183 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6184 
6185 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6186 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6187 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6188 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6189 
6190 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6191 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6192 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6193 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6194 
6195 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6196 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6197 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6198 
6199 	if (version > 1) {
6200 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6201 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6202 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6203 	}
6204 #undef WREG
6205 }
6206 
6207 static void
pci_cfg_restore_pcix(device_t dev,struct pci_devinfo * dinfo)6208 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6209 {
6210 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6211 	    dinfo->cfg.pcix.pcix_command,  2);
6212 }
6213 
6214 void
pci_cfg_restore(device_t dev,struct pci_devinfo * dinfo)6215 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6216 {
6217 
6218 	/*
6219 	 * Restore the device to full power mode.  We must do this
6220 	 * before we restore the registers because moving from D3 to
6221 	 * D0 will cause the chip's BARs and some other registers to
6222 	 * be reset to some unknown power on reset values.  Cut down
6223 	 * the noise on boot by doing nothing if we are already in
6224 	 * state D0.
6225 	 */
6226 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6227 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6228 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6229 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6230 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6231 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6232 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6233 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6234 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6235 	case PCIM_HDRTYPE_NORMAL:
6236 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6237 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6238 		break;
6239 	case PCIM_HDRTYPE_BRIDGE:
6240 		pci_write_config(dev, PCIR_SECLAT_1,
6241 		    dinfo->cfg.bridge.br_seclat, 1);
6242 		pci_write_config(dev, PCIR_SUBBUS_1,
6243 		    dinfo->cfg.bridge.br_subbus, 1);
6244 		pci_write_config(dev, PCIR_SECBUS_1,
6245 		    dinfo->cfg.bridge.br_secbus, 1);
6246 		pci_write_config(dev, PCIR_PRIBUS_1,
6247 		    dinfo->cfg.bridge.br_pribus, 1);
6248 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6249 		    dinfo->cfg.bridge.br_control, 2);
6250 		break;
6251 	case PCIM_HDRTYPE_CARDBUS:
6252 		pci_write_config(dev, PCIR_SECLAT_2,
6253 		    dinfo->cfg.bridge.br_seclat, 1);
6254 		pci_write_config(dev, PCIR_SUBBUS_2,
6255 		    dinfo->cfg.bridge.br_subbus, 1);
6256 		pci_write_config(dev, PCIR_SECBUS_2,
6257 		    dinfo->cfg.bridge.br_secbus, 1);
6258 		pci_write_config(dev, PCIR_PRIBUS_2,
6259 		    dinfo->cfg.bridge.br_pribus, 1);
6260 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6261 		    dinfo->cfg.bridge.br_control, 2);
6262 		break;
6263 	}
6264 	pci_restore_bars(dev);
6265 
6266 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6267 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6268 
6269 	/*
6270 	 * Restore extended capabilities for PCI-Express and PCI-X
6271 	 */
6272 	if (dinfo->cfg.pcie.pcie_location != 0)
6273 		pci_cfg_restore_pcie(dev, dinfo);
6274 	if (dinfo->cfg.pcix.pcix_location != 0)
6275 		pci_cfg_restore_pcix(dev, dinfo);
6276 
6277 	/* Restore MSI and MSI-X configurations if they are present. */
6278 	if (dinfo->cfg.msi.msi_location != 0)
6279 		pci_resume_msi(dev);
6280 	if (dinfo->cfg.msix.msix_location != 0)
6281 		pci_resume_msix(dev);
6282 
6283 #ifdef PCI_IOV
6284 	if (dinfo->cfg.iov != NULL)
6285 		pci_iov_cfg_restore(dev, dinfo);
6286 #endif
6287 }
6288 
6289 static void
pci_cfg_save_pcie(device_t dev,struct pci_devinfo * dinfo)6290 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6291 {
6292 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6293 	struct pcicfg_pcie *cfg;
6294 	int version, pos;
6295 
6296 	cfg = &dinfo->cfg.pcie;
6297 	pos = cfg->pcie_location;
6298 
6299 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6300 
6301 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6302 
6303 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6304 
6305 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6306 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6307 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6308 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6309 
6310 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6311 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6312 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6313 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6314 
6315 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6316 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6317 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6318 
6319 	if (version > 1) {
6320 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6321 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6322 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6323 	}
6324 #undef RREG
6325 }
6326 
6327 static void
pci_cfg_save_pcix(device_t dev,struct pci_devinfo * dinfo)6328 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6329 {
6330 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6331 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6332 }
6333 
6334 void
pci_cfg_save(device_t dev,struct pci_devinfo * dinfo,int setstate)6335 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6336 {
6337 	uint32_t cls;
6338 	int ps;
6339 
6340 	/*
6341 	 * Some drivers apparently write to these registers w/o updating our
6342 	 * cached copy.  No harm happens if we update the copy, so do so here
6343 	 * so we can restore them.  The COMMAND register is modified by the
6344 	 * bus w/o updating the cache.  This should represent the normally
6345 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6346 	 */
6347 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6348 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6349 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6350 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6351 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6352 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6353 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6354 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6355 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6356 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6357 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6358 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6359 	case PCIM_HDRTYPE_NORMAL:
6360 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6361 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6362 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6363 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6364 		break;
6365 	case PCIM_HDRTYPE_BRIDGE:
6366 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6367 		    PCIR_SECLAT_1, 1);
6368 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6369 		    PCIR_SUBBUS_1, 1);
6370 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6371 		    PCIR_SECBUS_1, 1);
6372 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6373 		    PCIR_PRIBUS_1, 1);
6374 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6375 		    PCIR_BRIDGECTL_1, 2);
6376 		break;
6377 	case PCIM_HDRTYPE_CARDBUS:
6378 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6379 		    PCIR_SECLAT_2, 1);
6380 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6381 		    PCIR_SUBBUS_2, 1);
6382 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6383 		    PCIR_SECBUS_2, 1);
6384 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6385 		    PCIR_PRIBUS_2, 1);
6386 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6387 		    PCIR_BRIDGECTL_2, 2);
6388 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6389 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6390 		break;
6391 	}
6392 
6393 	if (dinfo->cfg.pcie.pcie_location != 0)
6394 		pci_cfg_save_pcie(dev, dinfo);
6395 
6396 	if (dinfo->cfg.pcix.pcix_location != 0)
6397 		pci_cfg_save_pcix(dev, dinfo);
6398 
6399 #ifdef PCI_IOV
6400 	if (dinfo->cfg.iov != NULL)
6401 		pci_iov_cfg_save(dev, dinfo);
6402 #endif
6403 
6404 	/*
6405 	 * don't set the state for display devices, base peripherals and
6406 	 * memory devices since bad things happen when they are powered down.
6407 	 * We should (a) have drivers that can easily detach and (b) use
6408 	 * generic drivers for these devices so that some device actually
6409 	 * attaches.  We need to make sure that when we implement (a) we don't
6410 	 * power the device down on a reattach.
6411 	 */
6412 	cls = pci_get_class(dev);
6413 	if (!setstate)
6414 		return;
6415 	switch (pci_do_power_nodriver)
6416 	{
6417 		case 0:		/* NO powerdown at all */
6418 			return;
6419 		case 1:		/* Conservative about what to power down */
6420 			if (cls == PCIC_STORAGE)
6421 				return;
6422 			/*FALLTHROUGH*/
6423 		case 2:		/* Aggressive about what to power down */
6424 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6425 			    cls == PCIC_BASEPERIPH)
6426 				return;
6427 			/*FALLTHROUGH*/
6428 		case 3:		/* Power down everything */
6429 			break;
6430 	}
6431 	/*
6432 	 * PCI spec says we can only go into D3 state from D0 state.
6433 	 * Transition from D[12] into D0 before going to D3 state.
6434 	 */
6435 	ps = pci_get_powerstate(dev);
6436 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6437 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6438 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6439 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6440 }
6441 
6442 /* Wrapper APIs suitable for device driver use. */
6443 void
pci_save_state(device_t dev)6444 pci_save_state(device_t dev)
6445 {
6446 	struct pci_devinfo *dinfo;
6447 
6448 	dinfo = device_get_ivars(dev);
6449 	pci_cfg_save(dev, dinfo, 0);
6450 }
6451 
6452 void
pci_restore_state(device_t dev)6453 pci_restore_state(device_t dev)
6454 {
6455 	struct pci_devinfo *dinfo;
6456 
6457 	dinfo = device_get_ivars(dev);
6458 	pci_cfg_restore(dev, dinfo);
6459 }
6460 
6461 static int
pci_get_id_method(device_t dev,device_t child,enum pci_id_type type,uintptr_t * id)6462 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6463     uintptr_t *id)
6464 {
6465 
6466 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6467 }
6468 
6469 /* Find the upstream port of a given PCI device in a root complex. */
6470 device_t
pci_find_pcie_root_port(device_t dev)6471 pci_find_pcie_root_port(device_t dev)
6472 {
6473 	struct pci_devinfo *dinfo;
6474 	devclass_t pci_class;
6475 	device_t pcib, bus;
6476 
6477 	pci_class = devclass_find("pci");
6478 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6479 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6480 
6481 	/*
6482 	 * Walk the bridge hierarchy until we find a PCI-e root
6483 	 * port or a non-PCI device.
6484 	 */
6485 	for (;;) {
6486 		bus = device_get_parent(dev);
6487 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6488 		    device_get_nameunit(dev)));
6489 
6490 		pcib = device_get_parent(bus);
6491 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6492 		    device_get_nameunit(bus)));
6493 
6494 		/*
6495 		 * pcib's parent must be a PCI bus for this to be a
6496 		 * PCI-PCI bridge.
6497 		 */
6498 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6499 			return (NULL);
6500 
6501 		dinfo = device_get_ivars(pcib);
6502 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6503 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6504 			return (pcib);
6505 
6506 		dev = pcib;
6507 	}
6508 }
6509 
6510 /*
6511  * Wait for pending transactions to complete on a PCI-express function.
6512  *
6513  * The maximum delay is specified in milliseconds in max_delay.  Note
6514  * that this function may sleep.
6515  *
6516  * Returns true if the function is idle and false if the timeout is
6517  * exceeded.  If dev is not a PCI-express function, this returns true.
6518  */
6519 bool
pcie_wait_for_pending_transactions(device_t dev,u_int max_delay)6520 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6521 {
6522 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6523 	uint16_t sta;
6524 	int cap;
6525 
6526 	cap = dinfo->cfg.pcie.pcie_location;
6527 	if (cap == 0)
6528 		return (true);
6529 
6530 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6531 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6532 		if (max_delay == 0)
6533 			return (false);
6534 
6535 		/* Poll once every 100 milliseconds up to the timeout. */
6536 		if (max_delay > 100) {
6537 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6538 			max_delay -= 100;
6539 		} else {
6540 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6541 			    C_HARDCLOCK);
6542 			max_delay = 0;
6543 		}
6544 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6545 	}
6546 
6547 	return (true);
6548 }
6549 
6550 /*
6551  * Determine the maximum Completion Timeout in microseconds.
6552  *
6553  * For non-PCI-express functions this returns 0.
6554  */
6555 int
pcie_get_max_completion_timeout(device_t dev)6556 pcie_get_max_completion_timeout(device_t dev)
6557 {
6558 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6559 	int cap;
6560 
6561 	cap = dinfo->cfg.pcie.pcie_location;
6562 	if (cap == 0)
6563 		return (0);
6564 
6565 	/*
6566 	 * Functions using the 1.x spec use the default timeout range of
6567 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6568 	 * support programmable timeouts also use this range.
6569 	 */
6570 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6571 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6572 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6573 		return (50 * 1000);
6574 
6575 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6576 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6577 	case PCIEM_CTL2_COMP_TIMO_100US:
6578 		return (100);
6579 	case PCIEM_CTL2_COMP_TIMO_10MS:
6580 		return (10 * 1000);
6581 	case PCIEM_CTL2_COMP_TIMO_55MS:
6582 		return (55 * 1000);
6583 	case PCIEM_CTL2_COMP_TIMO_210MS:
6584 		return (210 * 1000);
6585 	case PCIEM_CTL2_COMP_TIMO_900MS:
6586 		return (900 * 1000);
6587 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6588 		return (3500 * 1000);
6589 	case PCIEM_CTL2_COMP_TIMO_13S:
6590 		return (13 * 1000 * 1000);
6591 	case PCIEM_CTL2_COMP_TIMO_64S:
6592 		return (64 * 1000 * 1000);
6593 	default:
6594 		return (50 * 1000);
6595 	}
6596 }
6597 
6598 void
pcie_apei_error(device_t dev,int sev,uint8_t * aerp)6599 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6600 {
6601 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6602 	const char *s;
6603 	int aer;
6604 	uint32_t r, r1;
6605 	uint16_t rs;
6606 
6607 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6608 		s = "Correctable";
6609 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6610 		s = "Uncorrectable (Non-Fatal)";
6611 	else
6612 		s = "Uncorrectable (Fatal)";
6613 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6614 	if (aerp) {
6615 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6616 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6617 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6618 		} else {
6619 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6620 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6621 		}
6622 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6623 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6624 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6625 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6626 			printf(" severity 0x%08x first %d\n",
6627 			    r, rs & 0x1f);
6628 		} else
6629 			printf("\n");
6630 	}
6631 
6632 	/* As kind of recovery just report and clear the error statuses. */
6633 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6634 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6635 		if (r != 0) {
6636 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6637 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6638 		}
6639 
6640 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6641 		if (r != 0) {
6642 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6643 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6644 		}
6645 	}
6646 	if (dinfo->cfg.pcie.pcie_location != 0) {
6647 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6648 		    PCIER_DEVICE_STA, 2);
6649 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6650 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6651 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6652 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6653 			    PCIER_DEVICE_STA, rs, 2);
6654 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6655 		}
6656 	}
6657 }
6658 
6659 /*
6660  * Perform a Function Level Reset (FLR) on a device.
6661  *
6662  * This function first waits for any pending transactions to complete
6663  * within the timeout specified by max_delay.  If transactions are
6664  * still pending, the function will return false without attempting a
6665  * reset.
6666  *
6667  * If dev is not a PCI-express function or does not support FLR, this
6668  * function returns false.
6669  *
6670  * Note that no registers are saved or restored.  The caller is
6671  * responsible for saving and restoring any registers including
6672  * PCI-standard registers via pci_save_state() and
6673  * pci_restore_state().
6674  */
6675 bool
pcie_flr(device_t dev,u_int max_delay,bool force)6676 pcie_flr(device_t dev, u_int max_delay, bool force)
6677 {
6678 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6679 	uint16_t cmd, ctl;
6680 	int compl_delay;
6681 	int cap;
6682 
6683 	cap = dinfo->cfg.pcie.pcie_location;
6684 	if (cap == 0)
6685 		return (false);
6686 
6687 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6688 		return (false);
6689 
6690 	/*
6691 	 * Disable busmastering to prevent generation of new
6692 	 * transactions while waiting for the device to go idle.  If
6693 	 * the idle timeout fails, the command register is restored
6694 	 * which will re-enable busmastering.
6695 	 */
6696 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6697 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6698 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6699 		if (!force) {
6700 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6701 			return (false);
6702 		}
6703 		pci_printf(&dinfo->cfg,
6704 		    "Resetting with transactions pending after %d ms\n",
6705 		    max_delay);
6706 
6707 		/*
6708 		 * Extend the post-FLR delay to cover the maximum
6709 		 * Completion Timeout delay of anything in flight
6710 		 * during the FLR delay.  Enforce a minimum delay of
6711 		 * at least 10ms.
6712 		 */
6713 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6714 		if (compl_delay < 10)
6715 			compl_delay = 10;
6716 	} else
6717 		compl_delay = 0;
6718 
6719 	/* Initiate the reset. */
6720 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6721 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6722 	    PCIEM_CTL_INITIATE_FLR, 2);
6723 
6724 	/* Wait for 100ms. */
6725 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6726 
6727 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6728 	    PCIEM_STA_TRANSACTION_PND)
6729 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6730 	return (true);
6731 }
6732 
6733 /*
6734  * Attempt a power-management reset by cycling the device in/out of D3
6735  * state.  PCI spec says we can only go into D3 state from D0 state.
6736  * Transition from D[12] into D0 before going to D3 state.
6737  */
6738 int
pci_power_reset(device_t dev)6739 pci_power_reset(device_t dev)
6740 {
6741 	int ps;
6742 
6743 	ps = pci_get_powerstate(dev);
6744 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6745 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6746 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6747 	pci_set_powerstate(dev, ps);
6748 	return (0);
6749 }
6750 
6751 /*
6752  * Try link drop and retrain of the downstream port of upstream
6753  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6754  * cause Conventional Hot reset of the device in the slot.
6755  * Alternative, for PCIe, could be the secondary bus reset initiatied
6756  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6757  */
6758 int
pcie_link_reset(device_t port,int pcie_location)6759 pcie_link_reset(device_t port, int pcie_location)
6760 {
6761 	uint16_t v;
6762 
6763 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6764 	v |= PCIEM_LINK_CTL_LINK_DIS;
6765 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6766 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6767 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6768 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6769 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6770 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6771 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6772 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6773 }
6774 
6775 static int
pci_reset_post(device_t dev,device_t child)6776 pci_reset_post(device_t dev, device_t child)
6777 {
6778 
6779 	if (dev == device_get_parent(child))
6780 		pci_restore_state(child);
6781 	return (0);
6782 }
6783 
6784 static int
pci_reset_prepare(device_t dev,device_t child)6785 pci_reset_prepare(device_t dev, device_t child)
6786 {
6787 
6788 	if (dev == device_get_parent(child))
6789 		pci_save_state(child);
6790 	return (0);
6791 }
6792 
6793 static int
pci_reset_child(device_t dev,device_t child,int flags)6794 pci_reset_child(device_t dev, device_t child, int flags)
6795 {
6796 	int error;
6797 
6798 	if (dev == NULL || device_get_parent(child) != dev)
6799 		return (0);
6800 	if ((flags & DEVF_RESET_DETACH) != 0) {
6801 		error = device_get_state(child) == DS_ATTACHED ?
6802 		    device_detach(child) : 0;
6803 	} else {
6804 		error = BUS_SUSPEND_CHILD(dev, child);
6805 	}
6806 	if (error == 0) {
6807 		if (!pcie_flr(child, 1000, false)) {
6808 			error = BUS_RESET_PREPARE(dev, child);
6809 			if (error == 0)
6810 				pci_power_reset(child);
6811 			BUS_RESET_POST(dev, child);
6812 		}
6813 		if ((flags & DEVF_RESET_DETACH) != 0)
6814 			device_probe_and_attach(child);
6815 		else
6816 			BUS_RESUME_CHILD(dev, child);
6817 	}
6818 	return (error);
6819 }
6820 
6821 const struct pci_device_table *
pci_match_device(device_t child,const struct pci_device_table * id,size_t nelt)6822 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6823 {
6824 	bool match;
6825 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6826 
6827 	vendor = pci_get_vendor(child);
6828 	device = pci_get_device(child);
6829 	subvendor = pci_get_subvendor(child);
6830 	subdevice = pci_get_subdevice(child);
6831 	class = pci_get_class(child);
6832 	subclass = pci_get_subclass(child);
6833 	revid = pci_get_revid(child);
6834 	while (nelt-- > 0) {
6835 		match = true;
6836 		if (id->match_flag_vendor)
6837 			match &= vendor == id->vendor;
6838 		if (id->match_flag_device)
6839 			match &= device == id->device;
6840 		if (id->match_flag_subvendor)
6841 			match &= subvendor == id->subvendor;
6842 		if (id->match_flag_subdevice)
6843 			match &= subdevice == id->subdevice;
6844 		if (id->match_flag_class)
6845 			match &= class == id->class_id;
6846 		if (id->match_flag_subclass)
6847 			match &= subclass == id->subclass;
6848 		if (id->match_flag_revid)
6849 			match &= revid == id->revid;
6850 		if (match)
6851 			return (id);
6852 		id++;
6853 	}
6854 	return (NULL);
6855 }
6856 
6857 static void
pci_print_faulted_dev_name(const struct pci_devinfo * dinfo)6858 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6859 {
6860 	const char *dev_name;
6861 	device_t dev;
6862 
6863 	dev = dinfo->cfg.dev;
6864 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6865 	    dinfo->cfg.slot, dinfo->cfg.func);
6866 	dev_name = device_get_name(dev);
6867 	if (dev_name != NULL)
6868 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6869 }
6870 
6871 void
pci_print_faulted_dev(void)6872 pci_print_faulted_dev(void)
6873 {
6874 	struct pci_devinfo *dinfo;
6875 	device_t dev;
6876 	int aer, i;
6877 	uint32_t r1, r2;
6878 	uint16_t status;
6879 
6880 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6881 		dev = dinfo->cfg.dev;
6882 		status = pci_read_config(dev, PCIR_STATUS, 2);
6883 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6884 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6885 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6886 		if (status != 0) {
6887 			pci_print_faulted_dev_name(dinfo);
6888 			printf(" error 0x%04x\n", status);
6889 		}
6890 		if (dinfo->cfg.pcie.pcie_location != 0) {
6891 			status = pci_read_config(dev,
6892 			    dinfo->cfg.pcie.pcie_location +
6893 			    PCIER_DEVICE_STA, 2);
6894 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6895 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6896 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6897 				pci_print_faulted_dev_name(dinfo);
6898 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6899 				    pci_read_config(dev,
6900 				    dinfo->cfg.pcie.pcie_location +
6901 				    PCIER_DEVICE_CTL, 2),
6902 				    status);
6903 			}
6904 		}
6905 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6906 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6907 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6908 			if (r1 != 0 || r2 != 0) {
6909 				pci_print_faulted_dev_name(dinfo);
6910 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6911 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6912 				    r1, pci_read_config(dev, aer +
6913 				    PCIR_AER_UC_MASK, 4),
6914 				    pci_read_config(dev, aer +
6915 				    PCIR_AER_UC_SEVERITY, 4),
6916 				    r2, pci_read_config(dev, aer +
6917 				    PCIR_AER_COR_MASK, 4),
6918 				    pci_read_config(dev, aer +
6919 				    PCIR_AER_CAP_CONTROL, 4));
6920 				for (i = 0; i < 4; i++) {
6921 					r1 = pci_read_config(dev, aer +
6922 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6923 					printf("    HL%d: 0x%08x\n", i, r1);
6924 				}
6925 			}
6926 		}
6927 	}
6928 }
6929 
6930 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(pcierr,pci_print_faulted_dev_db,DB_CMD_MEMSAFE)6931 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6932 {
6933 
6934 	pci_print_faulted_dev();
6935 }
6936 
6937 static void
db_clear_pcie_errors(const struct pci_devinfo * dinfo)6938 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6939 {
6940 	device_t dev;
6941 	int aer;
6942 	uint32_t r;
6943 
6944 	dev = dinfo->cfg.dev;
6945 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6946 	    PCIER_DEVICE_STA, 2);
6947 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6948 	    PCIER_DEVICE_STA, r, 2);
6949 
6950 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6951 		return;
6952 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6953 	if (r != 0)
6954 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6955 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6956 	if (r != 0)
6957 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6958 }
6959 
DB_COMMAND_FLAGS(pci_clearerr,db_pci_clearerr,DB_CMD_MEMSAFE)6960 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
6961 {
6962 	struct pci_devinfo *dinfo;
6963 	device_t dev;
6964 	uint16_t status, status1;
6965 
6966 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6967 		dev = dinfo->cfg.dev;
6968 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6969 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6970 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6971 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6972 		if (status1 != 0) {
6973 			status &= ~status1;
6974 			pci_write_config(dev, PCIR_STATUS, status, 2);
6975 		}
6976 		if (dinfo->cfg.pcie.pcie_location != 0)
6977 			db_clear_pcie_errors(dinfo);
6978 	}
6979 }
6980 #endif
6981