xref: /freebsd/sys/dev/pci/pci.c (revision 6882fc4cc43653dd52f6d49e15db9dada423455a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_acpi.h"
33 #include "opt_iommu.h"
34 #include "opt_bus.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/linker.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/stdarg.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_extern.h>
57 
58 #include <sys/bus.h>
59 #include <machine/bus.h>
60 #include <sys/rman.h>
61 #include <machine/resource.h>
62 
63 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
64 #include <machine/intr_machdep.h>
65 #endif
66 
67 #include <sys/pciio.h>
68 #include <dev/pci/pcireg.h>
69 #include <dev/pci/pcivar.h>
70 #include <dev/pci/pci_private.h>
71 
72 #ifdef PCI_IOV
73 #include <sys/nv.h>
74 #include <dev/pci/pci_iov_private.h>
75 #endif
76 
77 #include <dev/usb/controller/xhcireg.h>
78 #include <dev/usb/controller/ehcireg.h>
79 #include <dev/usb/controller/ohcireg.h>
80 #include <dev/usb/controller/uhcireg.h>
81 
82 #include <dev/iommu/iommu.h>
83 
84 #include "pcib_if.h"
85 #include "pci_if.h"
86 
87 #define	PCIR_IS_BIOS(cfg, reg)						\
88 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
89 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
90 
91 static device_probe_t	pci_probe;
92 
93 static bus_reset_post_t pci_reset_post;
94 static bus_reset_prepare_t pci_reset_prepare;
95 static bus_reset_child_t pci_reset_child;
96 static bus_hint_device_unit_t pci_hint_device_unit;
97 static bus_remap_intr_t pci_remap_intr_method;
98 
99 static pci_get_id_t	pci_get_id_method;
100 
101 static int		pci_has_quirk(uint32_t devid, int quirk);
102 static pci_addr_t	pci_mapbase(uint64_t mapreg);
103 static const char	*pci_maptype(uint64_t mapreg);
104 static int		pci_maprange(uint64_t mapreg);
105 static pci_addr_t	pci_rombase(uint64_t mapreg);
106 static int		pci_romsize(uint64_t testval);
107 static void		pci_fixancient(pcicfgregs *cfg);
108 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
109 
110 static int		pci_porten(device_t dev);
111 static int		pci_memen(device_t dev);
112 static void		pci_assign_interrupt(device_t bus, device_t dev,
113 			    int force_route);
114 static int		pci_add_map(device_t bus, device_t dev, int reg,
115 			    struct resource_list *rl, int force, int prefetch);
116 static void		pci_load_vendor_data(void);
117 static int		pci_describe_parse_line(char **ptr, int *vendor,
118 			    int *device, char **desc);
119 static char		*pci_describe_device(device_t dev);
120 static int		pci_modevent(module_t mod, int what, void *arg);
121 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
122 			    pcicfgregs *cfg);
123 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
124 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
125 			    int reg, uint32_t *data);
126 #if 0
127 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
128 			    int reg, uint32_t data);
129 #endif
130 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
131 static void		pci_mask_msix(device_t dev, u_int index);
132 static void		pci_unmask_msix(device_t dev, u_int index);
133 static int		pci_msi_blacklisted(void);
134 static int		pci_msix_blacklisted(void);
135 static void		pci_resume_msi(device_t dev);
136 static void		pci_resume_msix(device_t dev);
137 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
138     int b, int s, int f, uint16_t vid, uint16_t did);
139 
140 static device_method_t pci_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,		pci_probe),
143 	DEVMETHOD(device_attach,	pci_attach),
144 	DEVMETHOD(device_detach,	pci_detach),
145 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,	bus_generic_suspend),
147 	DEVMETHOD(device_resume,	pci_resume),
148 
149 	/* Bus interface */
150 	DEVMETHOD(bus_print_child,	pci_print_child),
151 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
152 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
153 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
154 	DEVMETHOD(bus_driver_added,	pci_driver_added),
155 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
156 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
157 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
158 	DEVMETHOD(bus_reset_post,	pci_reset_post),
159 	DEVMETHOD(bus_reset_child,	pci_reset_child),
160 
161 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
162 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
163 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
164 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
165 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
166 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
167 	DEVMETHOD(bus_adjust_resource,	pci_adjust_resource),
168 	DEVMETHOD(bus_release_resource,	pci_release_resource),
169 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
170 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
171 	DEVMETHOD(bus_map_resource,	pci_map_resource),
172 	DEVMETHOD(bus_unmap_resource,	pci_unmap_resource),
173 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
174 	DEVMETHOD(bus_child_detached,	pci_child_detached),
175 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
176 	DEVMETHOD(bus_child_location,	pci_child_location_method),
177 	DEVMETHOD(bus_get_device_path,	pci_get_device_path_method),
178 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
179 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
180 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
181 	DEVMETHOD(bus_resume_child,	pci_resume_child),
182 	DEVMETHOD(bus_rescan,		pci_rescan_method),
183 
184 	/* PCI interface */
185 	DEVMETHOD(pci_read_config,	pci_read_config_method),
186 	DEVMETHOD(pci_write_config,	pci_write_config_method),
187 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
188 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
189 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
190 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
191 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
192 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
193 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
194 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
195 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
196 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
197 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
198 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
199 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
200 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
201 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
202 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
203 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
204 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
205 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
206 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
207 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
208 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
209 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
210 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
211 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
212 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
213 	DEVMETHOD(pci_get_id,		pci_get_id_method),
214 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
215 	DEVMETHOD(pci_child_added,	pci_child_added_method),
216 #ifdef PCI_IOV
217 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
218 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
219 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
220 #endif
221 
222 	DEVMETHOD_END
223 };
224 
225 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
226 
227 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_modevent, NULL, BUS_PASS_BUS);
228 MODULE_VERSION(pci, 1);
229 
230 static char	*pci_vendordata;
231 static size_t	pci_vendordata_size;
232 
233 struct pci_quirk {
234 	uint32_t devid;	/* Vendor/device of the card */
235 	int	type;
236 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
237 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
238 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
239 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
240 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
241 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
243 	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 bool pci_intx_reroute = true;
433 SYSCTL_BOOL(_hw_pci, OID_AUTO, intx_reroute, CTLFLAG_RWTUN,
434     &pci_intx_reroute, 0, "Re-route INTx interrupts when scanning devices");
435 
436 static int
pci_has_quirk(uint32_t devid,int quirk)437 pci_has_quirk(uint32_t devid, int quirk)
438 {
439 	const struct pci_quirk *q;
440 
441 	for (q = &pci_quirks[0]; q->devid; q++) {
442 		if (q->devid == devid && q->type == quirk)
443 			return (1);
444 	}
445 	return (0);
446 }
447 
448 /* Find a device_t by bus/slot/function in domain 0 */
449 
450 device_t
pci_find_bsf(uint8_t bus,uint8_t slot,uint8_t func)451 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
452 {
453 
454 	return (pci_find_dbsf(0, bus, slot, func));
455 }
456 
457 /* Find a device_t by domain/bus/slot/function */
458 
459 device_t
pci_find_dbsf(uint32_t domain,uint8_t bus,uint8_t slot,uint8_t func)460 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
461 {
462 	struct pci_devinfo *dinfo = NULL;
463 
464 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
465 		if ((dinfo->cfg.domain == domain) &&
466 		    (dinfo->cfg.bus == bus) &&
467 		    (dinfo->cfg.slot == slot) &&
468 		    (dinfo->cfg.func == func)) {
469 			break;
470 		}
471 	}
472 
473 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
474 }
475 
476 /* Find a device_t by vendor/device ID */
477 
478 device_t
pci_find_device(uint16_t vendor,uint16_t device)479 pci_find_device(uint16_t vendor, uint16_t device)
480 {
481 	struct pci_devinfo *dinfo;
482 
483 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
484 		if ((dinfo->cfg.vendor == vendor) &&
485 		    (dinfo->cfg.device == device)) {
486 			return (dinfo->cfg.dev);
487 		}
488 	}
489 
490 	return (NULL);
491 }
492 
493 device_t
pci_find_class(uint8_t class,uint8_t subclass)494 pci_find_class(uint8_t class, uint8_t subclass)
495 {
496 	struct pci_devinfo *dinfo;
497 
498 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
499 		if (dinfo->cfg.baseclass == class &&
500 		    dinfo->cfg.subclass == subclass) {
501 			return (dinfo->cfg.dev);
502 		}
503 	}
504 
505 	return (NULL);
506 }
507 
508 device_t
pci_find_class_from(uint8_t class,uint8_t subclass,device_t from)509 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
510 {
511 	struct pci_devinfo *dinfo;
512 	bool found = false;
513 
514 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
515 		if (from != NULL && found == false) {
516 			if (from != dinfo->cfg.dev)
517 				continue;
518 			found = true;
519 			continue;
520 		}
521 		if (dinfo->cfg.baseclass == class &&
522 		    dinfo->cfg.subclass == subclass) {
523 			return (dinfo->cfg.dev);
524 		}
525 	}
526 
527 	return (NULL);
528 }
529 
530 device_t
pci_find_base_class_from(uint8_t class,device_t from)531 pci_find_base_class_from(uint8_t class, device_t from)
532 {
533 	struct pci_devinfo *dinfo;
534 	bool found = false;
535 
536 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
537 		if (from != NULL && found == false) {
538 			if (from != dinfo->cfg.dev)
539 				continue;
540 			found = true;
541 			continue;
542 		}
543 		if (dinfo->cfg.baseclass == class) {
544 			return (dinfo->cfg.dev);
545 		}
546 	}
547 
548 	return (NULL);
549 }
550 
551 static int
pci_printf(pcicfgregs * cfg,const char * fmt,...)552 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
553 {
554 	va_list ap;
555 	int retval;
556 
557 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
558 	    cfg->func);
559 	va_start(ap, fmt);
560 	retval += vprintf(fmt, ap);
561 	va_end(ap);
562 	return (retval);
563 }
564 
565 /* return base address of memory or port map */
566 
567 static pci_addr_t
pci_mapbase(uint64_t mapreg)568 pci_mapbase(uint64_t mapreg)
569 {
570 
571 	if (PCI_BAR_MEM(mapreg))
572 		return (mapreg & PCIM_BAR_MEM_BASE);
573 	else
574 		return (mapreg & PCIM_BAR_IO_BASE);
575 }
576 
577 /* return map type of memory or port map */
578 
579 static const char *
pci_maptype(uint64_t mapreg)580 pci_maptype(uint64_t mapreg)
581 {
582 
583 	if (PCI_BAR_IO(mapreg))
584 		return ("I/O Port");
585 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
586 		return ("Prefetchable Memory");
587 	return ("Memory");
588 }
589 
590 /* return log2 of map size decoded for memory or port map */
591 
592 int
pci_mapsize(uint64_t testval)593 pci_mapsize(uint64_t testval)
594 {
595 	int ln2size;
596 
597 	testval = pci_mapbase(testval);
598 	ln2size = 0;
599 	if (testval != 0) {
600 		while ((testval & 1) == 0)
601 		{
602 			ln2size++;
603 			testval >>= 1;
604 		}
605 	}
606 	return (ln2size);
607 }
608 
609 /* return base address of device ROM */
610 
611 static pci_addr_t
pci_rombase(uint64_t mapreg)612 pci_rombase(uint64_t mapreg)
613 {
614 
615 	return (mapreg & PCIM_BIOS_ADDR_MASK);
616 }
617 
618 /* return log2 of map size decided for device ROM */
619 
620 static int
pci_romsize(uint64_t testval)621 pci_romsize(uint64_t testval)
622 {
623 	int ln2size;
624 
625 	testval = pci_rombase(testval);
626 	ln2size = 0;
627 	if (testval != 0) {
628 		while ((testval & 1) == 0)
629 		{
630 			ln2size++;
631 			testval >>= 1;
632 		}
633 	}
634 	return (ln2size);
635 }
636 
637 /* return log2 of address range supported by map register */
638 
639 static int
pci_maprange(uint64_t mapreg)640 pci_maprange(uint64_t mapreg)
641 {
642 	int ln2range = 0;
643 
644 	if (PCI_BAR_IO(mapreg))
645 		ln2range = 32;
646 	else
647 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
648 		case PCIM_BAR_MEM_32:
649 			ln2range = 32;
650 			break;
651 		case PCIM_BAR_MEM_1MB:
652 			ln2range = 20;
653 			break;
654 		case PCIM_BAR_MEM_64:
655 			ln2range = 64;
656 			break;
657 		}
658 	return (ln2range);
659 }
660 
661 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
662 
663 static void
pci_fixancient(pcicfgregs * cfg)664 pci_fixancient(pcicfgregs *cfg)
665 {
666 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
667 		return;
668 
669 	/* PCI to PCI bridges use header type 1 */
670 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
671 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
672 }
673 
674 /* extract header type specific config data */
675 
676 static void
pci_hdrtypedata(device_t pcib,int b,int s,int f,pcicfgregs * cfg)677 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
678 {
679 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
680 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
681 	case PCIM_HDRTYPE_NORMAL:
682 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
683 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
684 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
685 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
686 		cfg->nummaps	    = PCI_MAXMAPS_0;
687 		break;
688 	case PCIM_HDRTYPE_BRIDGE:
689 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
690 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
691 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
692 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
693 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
694 		cfg->nummaps	    = PCI_MAXMAPS_1;
695 		break;
696 	case PCIM_HDRTYPE_CARDBUS:
697 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
698 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
699 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
700 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
701 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
702 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
703 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
704 		cfg->nummaps	    = PCI_MAXMAPS_2;
705 		break;
706 	}
707 #undef REG
708 }
709 
710 /* read configuration header into pcicfgregs structure */
711 struct pci_devinfo *
pci_read_device(device_t pcib,device_t bus,int d,int b,int s,int f)712 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
713 {
714 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
715 	uint16_t vid, did;
716 
717 	vid = REG(PCIR_VENDOR, 2);
718 	if (vid == PCIV_INVALID)
719 		return (NULL);
720 
721 	did = REG(PCIR_DEVICE, 2);
722 
723 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
724 }
725 
726 struct pci_devinfo *
pci_alloc_devinfo_method(device_t dev)727 pci_alloc_devinfo_method(device_t dev)
728 {
729 
730 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
731 	    M_WAITOK | M_ZERO));
732 }
733 
734 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)735 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
736     uint16_t vid, uint16_t did)
737 {
738 	struct pci_devinfo *devlist_entry;
739 	pcicfgregs *cfg;
740 
741 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
742 
743 	cfg = &devlist_entry->cfg;
744 
745 	cfg->domain		= d;
746 	cfg->bus		= b;
747 	cfg->slot		= s;
748 	cfg->func		= f;
749 	cfg->vendor		= vid;
750 	cfg->device		= did;
751 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
752 	cfg->statreg		= REG(PCIR_STATUS, 2);
753 	cfg->baseclass		= REG(PCIR_CLASS, 1);
754 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
755 	cfg->progif		= REG(PCIR_PROGIF, 1);
756 	cfg->revid		= REG(PCIR_REVID, 1);
757 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
758 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
759 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
760 	cfg->intpin		= REG(PCIR_INTPIN, 1);
761 	cfg->intline		= REG(PCIR_INTLINE, 1);
762 
763 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
764 	cfg->hdrtype		&= ~PCIM_MFDEV;
765 	STAILQ_INIT(&cfg->maps);
766 
767 	cfg->iov		= NULL;
768 
769 	pci_fixancient(cfg);
770 	pci_hdrtypedata(pcib, b, s, f, cfg);
771 
772 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
773 		pci_read_cap(pcib, cfg);
774 
775 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
776 
777 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
778 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
779 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
780 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
781 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
782 
783 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
784 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
785 	devlist_entry->conf.pc_vendor = cfg->vendor;
786 	devlist_entry->conf.pc_device = cfg->device;
787 
788 	devlist_entry->conf.pc_class = cfg->baseclass;
789 	devlist_entry->conf.pc_subclass = cfg->subclass;
790 	devlist_entry->conf.pc_progif = cfg->progif;
791 	devlist_entry->conf.pc_revid = cfg->revid;
792 
793 	pci_numdevs++;
794 	pci_generation++;
795 
796 	return (devlist_entry);
797 }
798 #undef REG
799 
800 static void
pci_ea_fill_info(device_t pcib,pcicfgregs * cfg)801 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
802 {
803 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
804     cfg->ea.ea_location + (n), w)
805 	int num_ent;
806 	int ptr;
807 	int a, b;
808 	uint32_t val;
809 	int ent_size;
810 	uint32_t dw[4];
811 	uint64_t base, max_offset;
812 	struct pci_ea_entry *eae;
813 
814 	if (cfg->ea.ea_location == 0)
815 		return;
816 
817 	STAILQ_INIT(&cfg->ea.ea_entries);
818 
819 	/* Determine the number of entries */
820 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
821 	num_ent &= PCIM_EA_NUM_ENT_MASK;
822 
823 	/* Find the first entry to care of */
824 	ptr = PCIR_EA_FIRST_ENT;
825 
826 	/* Skip DWORD 2 for type 1 functions */
827 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
828 		ptr += 4;
829 
830 	for (a = 0; a < num_ent; a++) {
831 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
832 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
833 
834 		/* Read a number of dwords in the entry */
835 		val = REG(ptr, 4);
836 		ptr += 4;
837 		ent_size = (val & PCIM_EA_ES);
838 
839 		for (b = 0; b < ent_size; b++) {
840 			dw[b] = REG(ptr, 4);
841 			ptr += 4;
842 		}
843 
844 		eae->eae_flags = val;
845 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
846 
847 		base = dw[0] & PCIM_EA_FIELD_MASK;
848 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
849 		b = 2;
850 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
851 			base |= (uint64_t)dw[b] << 32UL;
852 			b++;
853 		}
854 		if (((dw[1] & PCIM_EA_IS_64) != 0)
855 		    && (b < ent_size)) {
856 			max_offset |= (uint64_t)dw[b] << 32UL;
857 			b++;
858 		}
859 
860 		eae->eae_base = base;
861 		eae->eae_max_offset = max_offset;
862 
863 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
864 
865 		if (bootverbose) {
866 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
867 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
868 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
869 		}
870 	}
871 }
872 #undef REG
873 
874 static void
pci_read_cap(device_t pcib,pcicfgregs * cfg)875 pci_read_cap(device_t pcib, pcicfgregs *cfg)
876 {
877 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
878 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
879 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
880 	uint64_t addr;
881 #endif
882 	uint32_t val;
883 	int	ptr, nextptr, ptrptr;
884 
885 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
886 	case PCIM_HDRTYPE_NORMAL:
887 	case PCIM_HDRTYPE_BRIDGE:
888 		ptrptr = PCIR_CAP_PTR;
889 		break;
890 	case PCIM_HDRTYPE_CARDBUS:
891 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
892 		break;
893 	default:
894 		return;		/* no extended capabilities support */
895 	}
896 	nextptr = REG(ptrptr, 1);	/* sanity check? */
897 
898 	/*
899 	 * Read capability entries.
900 	 */
901 	while (nextptr != 0) {
902 		/* Sanity check */
903 		if (nextptr > 255) {
904 			printf("illegal PCI extended capability offset %d\n",
905 			    nextptr);
906 			return;
907 		}
908 		/* Find the next entry */
909 		ptr = nextptr;
910 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
911 
912 		/* Process this entry */
913 		switch (REG(ptr + PCICAP_ID, 1)) {
914 		case PCIY_PMG:		/* PCI power management */
915 			cfg->pp.pp_location = ptr;
916 			cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
917 			break;
918 		case PCIY_HT:		/* HyperTransport */
919 			/* Determine HT-specific capability type. */
920 			val = REG(ptr + PCIR_HT_COMMAND, 2);
921 
922 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
923 				cfg->ht.ht_slave = ptr;
924 
925 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
926 			switch (val & PCIM_HTCMD_CAP_MASK) {
927 			case PCIM_HTCAP_MSI_MAPPING:
928 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
929 					/* Sanity check the mapping window. */
930 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
931 					    4);
932 					addr <<= 32;
933 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
934 					    4);
935 					if (addr != MSI_INTEL_ADDR_BASE)
936 						device_printf(pcib,
937 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
938 						    cfg->domain, cfg->bus,
939 						    cfg->slot, cfg->func,
940 						    (long long)addr);
941 				} else
942 					addr = MSI_INTEL_ADDR_BASE;
943 
944 				cfg->ht.ht_msimap = ptr;
945 				cfg->ht.ht_msictrl = val;
946 				cfg->ht.ht_msiaddr = addr;
947 				break;
948 			}
949 #endif
950 			break;
951 		case PCIY_MSI:		/* PCI MSI */
952 			cfg->msi.msi_location = ptr;
953 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
954 			break;
955 		case PCIY_MSIX:		/* PCI MSI-X */
956 			cfg->msix.msix_location = ptr;
957 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
958 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
959 			cfg->msix.msix_table_bar = PCIR_BAR(val &
960 			    PCIM_MSIX_BIR_MASK);
961 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
962 			val = REG(ptr + PCIR_MSIX_PBA, 4);
963 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
964 			    PCIM_MSIX_BIR_MASK);
965 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
966 			break;
967 		case PCIY_VPD:		/* PCI Vital Product Data */
968 			cfg->vpd.vpd_reg = ptr;
969 			break;
970 		case PCIY_SUBVENDOR:
971 			/* Should always be true. */
972 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
973 			    PCIM_HDRTYPE_BRIDGE) {
974 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
975 				cfg->subvendor = val & 0xffff;
976 				cfg->subdevice = val >> 16;
977 			}
978 			break;
979 		case PCIY_PCIX:		/* PCI-X */
980 			/*
981 			 * Assume we have a PCI-X chipset if we have
982 			 * at least one PCI-PCI bridge with a PCI-X
983 			 * capability.  Note that some systems with
984 			 * PCI-express or HT chipsets might match on
985 			 * this check as well.
986 			 */
987 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
988 			    PCIM_HDRTYPE_BRIDGE)
989 				pcix_chipset = 1;
990 			cfg->pcix.pcix_location = ptr;
991 			break;
992 		case PCIY_EXPRESS:	/* PCI-express */
993 			/*
994 			 * Assume we have a PCI-express chipset if we have
995 			 * at least one PCI-express device.
996 			 */
997 			pcie_chipset = 1;
998 			cfg->pcie.pcie_location = ptr;
999 			val = REG(ptr + PCIER_FLAGS, 2);
1000 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
1001 			break;
1002 		case PCIY_EA:		/* Enhanced Allocation */
1003 			cfg->ea.ea_location = ptr;
1004 			pci_ea_fill_info(pcib, cfg);
1005 			break;
1006 		default:
1007 			break;
1008 		}
1009 	}
1010 
1011 #if defined(__powerpc__)
1012 	/*
1013 	 * Enable the MSI mapping window for all HyperTransport
1014 	 * slaves.  PCI-PCI bridges have their windows enabled via
1015 	 * PCIB_MAP_MSI().
1016 	 */
1017 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1018 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1019 		device_printf(pcib,
1020 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1021 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1022 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1023 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1024 		     2);
1025 	}
1026 #endif
1027 /* REG and WREG use carry through to next functions */
1028 }
1029 
1030 /*
1031  * PCI Vital Product Data
1032  */
1033 
1034 #define	PCI_VPD_TIMEOUT		1000000
1035 
1036 static int
pci_read_vpd_reg(device_t pcib,pcicfgregs * cfg,int reg,uint32_t * data)1037 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1038 {
1039 	int count = PCI_VPD_TIMEOUT;
1040 
1041 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1042 
1043 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1044 
1045 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1046 		if (--count < 0)
1047 			return (ENXIO);
1048 		DELAY(1);	/* limit looping */
1049 	}
1050 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1051 
1052 	return (0);
1053 }
1054 
1055 #if 0
1056 static int
1057 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1058 {
1059 	int count = PCI_VPD_TIMEOUT;
1060 
1061 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1062 
1063 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1064 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1065 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1066 		if (--count < 0)
1067 			return (ENXIO);
1068 		DELAY(1);	/* limit looping */
1069 	}
1070 
1071 	return (0);
1072 }
1073 #endif
1074 
1075 #undef PCI_VPD_TIMEOUT
1076 
1077 struct vpd_readstate {
1078 	device_t	pcib;
1079 	pcicfgregs	*cfg;
1080 	uint32_t	val;
1081 	int		bytesinval;
1082 	int		off;
1083 	uint8_t		cksum;
1084 };
1085 
1086 /* return 0 and one byte in *data if no read error, -1 else */
1087 static int
vpd_nextbyte(struct vpd_readstate * vrs,uint8_t * data)1088 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1089 {
1090 	uint32_t reg;
1091 	uint8_t byte;
1092 
1093 	if (vrs->bytesinval == 0) {
1094 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1095 			return (-1);
1096 		vrs->val = le32toh(reg);
1097 		vrs->off += 4;
1098 		byte = vrs->val & 0xff;
1099 		vrs->bytesinval = 3;
1100 	} else {
1101 		vrs->val = vrs->val >> 8;
1102 		byte = vrs->val & 0xff;
1103 		vrs->bytesinval--;
1104 	}
1105 
1106 	vrs->cksum += byte;
1107 	*data = byte;
1108 	return (0);
1109 }
1110 
1111 /* return 0 on match, -1 and "unget" byte on no match */
1112 static int
vpd_expectbyte(struct vpd_readstate * vrs,uint8_t expected)1113 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1114 {
1115 	uint8_t data;
1116 
1117 	if (vpd_nextbyte(vrs, &data) != 0)
1118 		return (-1);
1119 
1120 	if (data == expected)
1121 		return (0);
1122 
1123 	vrs->cksum -= data;
1124 	vrs->val = (vrs->val << 8) + data;
1125 	vrs->bytesinval++;
1126 	return (-1);
1127 }
1128 
1129 /* return size if tag matches, -1 on no match, -2 on read error */
1130 static int
vpd_read_tag_size(struct vpd_readstate * vrs,uint8_t vpd_tag)1131 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1132 {
1133 	uint8_t byte1, byte2;
1134 
1135 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1136 		return (-1);
1137 
1138 	if ((vpd_tag & 0x80) == 0)
1139 		return (vpd_tag & 0x07);
1140 
1141 	if (vpd_nextbyte(vrs, &byte1) != 0)
1142 		return (-2);
1143 	if (vpd_nextbyte(vrs, &byte2) != 0)
1144 		return (-2);
1145 
1146 	return ((byte2 << 8) + byte1);
1147 }
1148 
1149 /* (re)allocate buffer in multiples of 8 elements */
1150 static void*
alloc_buffer(void * buffer,size_t element_size,int needed)1151 alloc_buffer(void* buffer, size_t element_size, int needed)
1152 {
1153 	int alloc, new_alloc;
1154 
1155 	alloc = roundup2(needed, 8);
1156 	new_alloc = roundup2(needed + 1, 8);
1157 	if (alloc != new_alloc) {
1158 		buffer = reallocf(buffer,
1159 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1160 	}
1161 
1162 	return (buffer);
1163 }
1164 
1165 /* read VPD keyword and return element size, return -1 on read error */
1166 static int
vpd_read_elem_head(struct vpd_readstate * vrs,char keyword[2])1167 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1168 {
1169 	uint8_t data;
1170 
1171 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1172 		return (-1);
1173 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1174 		return (-1);
1175 	if (vpd_nextbyte(vrs, &data) != 0)
1176 		return (-1);
1177 
1178 	return (data);
1179 }
1180 
1181 /* read VPD data element of given size into allocated buffer */
1182 static char *
vpd_read_value(struct vpd_readstate * vrs,int size)1183 vpd_read_value(struct vpd_readstate *vrs, int size)
1184 {
1185 	int i;
1186 	char char1;
1187 	char *value;
1188 
1189 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1190 	for (i = 0; i < size; i++) {
1191 		if (vpd_nextbyte(vrs, &char1) != 0) {
1192 			free(value, M_DEVBUF);
1193 			return (NULL);
1194 		}
1195 		value[i] = char1;
1196 	}
1197 	value[size] = '\0';
1198 
1199 	return (value);
1200 }
1201 
1202 /* read VPD into *keyword and *value, return length of data element */
1203 static int
vpd_read_elem_data(struct vpd_readstate * vrs,char keyword[2],char ** value,int maxlen)1204 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1205 {
1206 	int len;
1207 
1208 	len = vpd_read_elem_head(vrs, keyword);
1209 	if (len < 0 || len > maxlen)
1210 		return (-1);
1211 	*value = vpd_read_value(vrs, len);
1212 
1213 	return (len);
1214 }
1215 
1216 /* subtract all data following first byte from checksum of RV element */
1217 static void
vpd_fixup_cksum(struct vpd_readstate * vrs,char * rvstring,int len)1218 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1219 {
1220 	int i;
1221 	uint8_t fixup;
1222 
1223 	fixup = 0;
1224 	for (i = 1; i < len; i++)
1225 		fixup += rvstring[i];
1226 	vrs->cksum -= fixup;
1227 }
1228 
1229 /* fetch one read-only element and return size of heading + data */
1230 static int
next_vpd_ro_elem(struct vpd_readstate * vrs,int maxsize)1231 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1232 {
1233 	struct pcicfg_vpd *vpd;
1234 	pcicfgregs *cfg;
1235 	struct vpd_readonly *vpd_ros;
1236 	int len;
1237 
1238 	cfg = vrs->cfg;
1239 	vpd = &cfg->vpd;
1240 
1241 	if (maxsize < 3)
1242 		return (-1);
1243 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1244 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1245 	maxsize -= 3;
1246 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1247 	if (vpd_ros->value == NULL)
1248 		return (-1);
1249 	vpd_ros->len = len;
1250 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1251 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1252 		if (vrs->cksum != 0) {
1253 			pci_printf(cfg,
1254 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1255 			return (-1);
1256 		}
1257 	}
1258 	vpd->vpd_rocnt++;
1259 
1260 	return (len + 3);
1261 }
1262 
1263 /* fetch one writable element and return size of heading + data */
1264 static int
next_vpd_rw_elem(struct vpd_readstate * vrs,int maxsize)1265 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1266 {
1267 	struct pcicfg_vpd *vpd;
1268 	pcicfgregs *cfg;
1269 	struct vpd_write *vpd_w;
1270 	int len;
1271 
1272 	cfg = vrs->cfg;
1273 	vpd = &cfg->vpd;
1274 
1275 	if (maxsize < 3)
1276 		return (-1);
1277 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1278 	if (vpd->vpd_w == NULL) {
1279 		pci_printf(cfg, "out of memory");
1280 		return (-1);
1281 	}
1282 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1283 	maxsize -= 3;
1284 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1285 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1286 	if (vpd_w->value == NULL)
1287 		return (-1);
1288 	vpd_w->len = len;
1289 	vpd->vpd_wcnt++;
1290 
1291 	return (len + 3);
1292 }
1293 
1294 /* free all memory allocated for VPD data */
1295 static void
vpd_free(struct pcicfg_vpd * vpd)1296 vpd_free(struct pcicfg_vpd *vpd)
1297 {
1298 	int i;
1299 
1300 	free(vpd->vpd_ident, M_DEVBUF);
1301 	for (i = 0; i < vpd->vpd_rocnt; i++)
1302 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1303 	free(vpd->vpd_ros, M_DEVBUF);
1304 	vpd->vpd_rocnt = 0;
1305 	for (i = 0; i < vpd->vpd_wcnt; i++)
1306 		free(vpd->vpd_w[i].value, M_DEVBUF);
1307 	free(vpd->vpd_w, M_DEVBUF);
1308 	vpd->vpd_wcnt = 0;
1309 }
1310 
1311 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1312 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1313 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1314 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1315 
1316 static int
pci_parse_vpd(device_t pcib,pcicfgregs * cfg)1317 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1318 {
1319 	struct vpd_readstate vrs;
1320 	int cksumvalid;
1321 	int size, elem_size;
1322 
1323 	/* init vpd reader */
1324 	vrs.bytesinval = 0;
1325 	vrs.off = 0;
1326 	vrs.pcib = pcib;
1327 	vrs.cfg = cfg;
1328 	vrs.cksum = 0;
1329 
1330 	/* read VPD ident element - mandatory */
1331 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1332 	if (size <= 0) {
1333 		pci_printf(cfg, "no VPD ident found\n");
1334 		return (0);
1335 	}
1336 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1337 	if (cfg->vpd.vpd_ident == NULL) {
1338 		pci_printf(cfg, "error accessing VPD ident data\n");
1339 		return (0);
1340 	}
1341 
1342 	/* read VPD RO elements - mandatory */
1343 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1344 	if (size <= 0) {
1345 		pci_printf(cfg, "no read-only VPD data found\n");
1346 		return (0);
1347 	}
1348 	while (size > 0) {
1349 		elem_size = next_vpd_ro_elem(&vrs, size);
1350 		if (elem_size < 0) {
1351 			pci_printf(cfg, "error accessing read-only VPD data\n");
1352 			return (-1);
1353 		}
1354 		size -= elem_size;
1355 	}
1356 	cksumvalid = (vrs.cksum == 0);
1357 	if (!cksumvalid)
1358 		return (-1);
1359 
1360 	/* read VPD RW elements - optional */
1361 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1362 	if (size == -2)
1363 		return (-1);
1364 	while (size > 0) {
1365 		elem_size = next_vpd_rw_elem(&vrs, size);
1366 		if (elem_size < 0) {
1367 			pci_printf(cfg, "error accessing writeable VPD data\n");
1368 			return (-1);
1369 		}
1370 		size -= elem_size;
1371 	}
1372 
1373 	/* read empty END tag - mandatory */
1374 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1375 	if (size != 0) {
1376 		pci_printf(cfg, "No valid VPD end tag found\n");
1377 	}
1378 	return (0);
1379 }
1380 
1381 static void
pci_read_vpd(device_t pcib,pcicfgregs * cfg)1382 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1383 {
1384 	int status;
1385 
1386 	status = pci_parse_vpd(pcib, cfg);
1387 	if (status < 0)
1388 		vpd_free(&cfg->vpd);
1389 	cfg->vpd.vpd_cached = 1;
1390 #undef REG
1391 #undef WREG
1392 }
1393 
1394 int
pci_get_vpd_ident_method(device_t dev,device_t child,const char ** identptr)1395 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1396 {
1397 	struct pci_devinfo *dinfo = device_get_ivars(child);
1398 	pcicfgregs *cfg = &dinfo->cfg;
1399 
1400 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1401 		pci_read_vpd(device_get_parent(dev), cfg);
1402 
1403 	*identptr = cfg->vpd.vpd_ident;
1404 
1405 	if (*identptr == NULL)
1406 		return (ENXIO);
1407 
1408 	return (0);
1409 }
1410 
1411 int
pci_get_vpd_readonly_method(device_t dev,device_t child,const char * kw,const char ** vptr)1412 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1413 	const char **vptr)
1414 {
1415 	struct pci_devinfo *dinfo = device_get_ivars(child);
1416 	pcicfgregs *cfg = &dinfo->cfg;
1417 	int i;
1418 
1419 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1420 		pci_read_vpd(device_get_parent(dev), cfg);
1421 
1422 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1423 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1424 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1425 			*vptr = cfg->vpd.vpd_ros[i].value;
1426 			return (0);
1427 		}
1428 
1429 	*vptr = NULL;
1430 	return (ENXIO);
1431 }
1432 
1433 struct pcicfg_vpd *
pci_fetch_vpd_list(device_t dev)1434 pci_fetch_vpd_list(device_t dev)
1435 {
1436 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1437 	pcicfgregs *cfg = &dinfo->cfg;
1438 
1439 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1440 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1441 	return (&cfg->vpd);
1442 }
1443 
1444 /*
1445  * Find the requested HyperTransport capability and return the offset
1446  * in configuration space via the pointer provided.  The function
1447  * returns 0 on success and an error code otherwise.
1448  */
1449 int
pci_find_htcap_method(device_t dev,device_t child,int capability,int * capreg)1450 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1451 {
1452 	int ptr, error;
1453 	uint16_t val;
1454 
1455 	error = pci_find_cap(child, PCIY_HT, &ptr);
1456 	if (error)
1457 		return (error);
1458 
1459 	/*
1460 	 * Traverse the capabilities list checking each HT capability
1461 	 * to see if it matches the requested HT capability.
1462 	 */
1463 	for (;;) {
1464 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1465 		if (capability == PCIM_HTCAP_SLAVE ||
1466 		    capability == PCIM_HTCAP_HOST)
1467 			val &= 0xe000;
1468 		else
1469 			val &= PCIM_HTCMD_CAP_MASK;
1470 		if (val == capability) {
1471 			if (capreg != NULL)
1472 				*capreg = ptr;
1473 			return (0);
1474 		}
1475 
1476 		/* Skip to the next HT capability. */
1477 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1478 			break;
1479 	}
1480 
1481 	return (ENOENT);
1482 }
1483 
1484 /*
1485  * Find the next requested HyperTransport capability after start and return
1486  * the offset in configuration space via the pointer provided.  The function
1487  * returns 0 on success and an error code otherwise.
1488  */
1489 int
pci_find_next_htcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1490 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1491     int start, int *capreg)
1492 {
1493 	int ptr;
1494 	uint16_t val;
1495 
1496 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1497 	    ("start capability is not HyperTransport capability"));
1498 	ptr = start;
1499 
1500 	/*
1501 	 * Traverse the capabilities list checking each HT capability
1502 	 * to see if it matches the requested HT capability.
1503 	 */
1504 	for (;;) {
1505 		/* Skip to the next HT capability. */
1506 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1507 			break;
1508 
1509 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1510 		if (capability == PCIM_HTCAP_SLAVE ||
1511 		    capability == PCIM_HTCAP_HOST)
1512 			val &= 0xe000;
1513 		else
1514 			val &= PCIM_HTCMD_CAP_MASK;
1515 		if (val == capability) {
1516 			if (capreg != NULL)
1517 				*capreg = ptr;
1518 			return (0);
1519 		}
1520 	}
1521 
1522 	return (ENOENT);
1523 }
1524 
1525 /*
1526  * Find the requested capability and return the offset in
1527  * configuration space via the pointer provided.  The function returns
1528  * 0 on success and an error code otherwise.
1529  */
1530 int
pci_find_cap_method(device_t dev,device_t child,int capability,int * capreg)1531 pci_find_cap_method(device_t dev, device_t child, int capability,
1532     int *capreg)
1533 {
1534 	struct pci_devinfo *dinfo = device_get_ivars(child);
1535 	pcicfgregs *cfg = &dinfo->cfg;
1536 	uint32_t status;
1537 	uint8_t ptr;
1538 	int cnt;
1539 
1540 	/*
1541 	 * Check the CAP_LIST bit of the PCI status register first.
1542 	 */
1543 	status = pci_read_config(child, PCIR_STATUS, 2);
1544 	if (!(status & PCIM_STATUS_CAPPRESENT))
1545 		return (ENXIO);
1546 
1547 	/*
1548 	 * Determine the start pointer of the capabilities list.
1549 	 */
1550 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1551 	case PCIM_HDRTYPE_NORMAL:
1552 	case PCIM_HDRTYPE_BRIDGE:
1553 		ptr = PCIR_CAP_PTR;
1554 		break;
1555 	case PCIM_HDRTYPE_CARDBUS:
1556 		ptr = PCIR_CAP_PTR_2;
1557 		break;
1558 	default:
1559 		/* XXX: panic? */
1560 		return (ENXIO);		/* no extended capabilities support */
1561 	}
1562 	ptr = pci_read_config(child, ptr, 1);
1563 
1564 	/*
1565 	 * Traverse the capabilities list.  Limit by total theoretical
1566 	 * maximum number of caps: capability needs at least id and
1567 	 * next registers, and any type X header cannot contain caps.
1568 	 */
1569 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1570 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1571 			if (capreg != NULL)
1572 				*capreg = ptr;
1573 			return (0);
1574 		}
1575 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1576 	}
1577 
1578 	return (ENOENT);
1579 }
1580 
1581 /*
1582  * Find the next requested capability after start and return the offset in
1583  * configuration space via the pointer provided.  The function returns
1584  * 0 on success and an error code otherwise.
1585  */
1586 int
pci_find_next_cap_method(device_t dev,device_t child,int capability,int start,int * capreg)1587 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1588     int start, int *capreg)
1589 {
1590 	uint8_t ptr;
1591 
1592 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1593 	    ("start capability is not expected capability"));
1594 
1595 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1596 	while (ptr != 0) {
1597 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1598 			if (capreg != NULL)
1599 				*capreg = ptr;
1600 			return (0);
1601 		}
1602 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1603 	}
1604 
1605 	return (ENOENT);
1606 }
1607 
1608 /*
1609  * Find the requested extended capability and return the offset in
1610  * configuration space via the pointer provided.  The function returns
1611  * 0 on success and an error code otherwise.
1612  */
1613 int
pci_find_extcap_method(device_t dev,device_t child,int capability,int * capreg)1614 pci_find_extcap_method(device_t dev, device_t child, int capability,
1615     int *capreg)
1616 {
1617 	struct pci_devinfo *dinfo = device_get_ivars(child);
1618 	pcicfgregs *cfg = &dinfo->cfg;
1619 	uint32_t ecap;
1620 	uint16_t ptr;
1621 
1622 	/* Only supported for PCI-express devices. */
1623 	if (cfg->pcie.pcie_location == 0)
1624 		return (ENXIO);
1625 
1626 	ptr = PCIR_EXTCAP;
1627 	ecap = pci_read_config(child, ptr, 4);
1628 	if (ecap == 0xffffffff || ecap == 0)
1629 		return (ENOENT);
1630 	for (;;) {
1631 		if (PCI_EXTCAP_ID(ecap) == capability) {
1632 			if (capreg != NULL)
1633 				*capreg = ptr;
1634 			return (0);
1635 		}
1636 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1637 		if (ptr == 0)
1638 			break;
1639 		ecap = pci_read_config(child, ptr, 4);
1640 	}
1641 
1642 	return (ENOENT);
1643 }
1644 
1645 /*
1646  * Find the next requested extended capability after start and return the
1647  * offset in configuration space via the pointer provided.  The function
1648  * returns 0 on success and an error code otherwise.
1649  */
1650 int
pci_find_next_extcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1651 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1652     int start, int *capreg)
1653 {
1654 	struct pci_devinfo *dinfo = device_get_ivars(child);
1655 	pcicfgregs *cfg = &dinfo->cfg;
1656 	uint32_t ecap;
1657 	uint16_t ptr;
1658 
1659 	/* Only supported for PCI-express devices. */
1660 	if (cfg->pcie.pcie_location == 0)
1661 		return (ENXIO);
1662 
1663 	ecap = pci_read_config(child, start, 4);
1664 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1665 	    ("start extended capability is not expected capability"));
1666 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1667 	while (ptr != 0) {
1668 		ecap = pci_read_config(child, ptr, 4);
1669 		if (PCI_EXTCAP_ID(ecap) == capability) {
1670 			if (capreg != NULL)
1671 				*capreg = ptr;
1672 			return (0);
1673 		}
1674 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1675 	}
1676 
1677 	return (ENOENT);
1678 }
1679 
1680 /*
1681  * Support for MSI-X message interrupts.
1682  */
1683 static void
pci_write_msix_entry(device_t dev,u_int index,uint64_t address,uint32_t data)1684 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1685 {
1686 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1687 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1688 	uint32_t offset;
1689 
1690 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1691 	offset = msix->msix_table_offset + index * 16;
1692 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1693 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1694 	bus_write_4(msix->msix_table_res, offset + 8, data);
1695 }
1696 
1697 void
pci_enable_msix_method(device_t dev,device_t child,u_int index,uint64_t address,uint32_t data)1698 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1699     uint64_t address, uint32_t data)
1700 {
1701 
1702 	if (pci_msix_rewrite_table) {
1703 		struct pci_devinfo *dinfo = device_get_ivars(child);
1704 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1705 
1706 		/*
1707 		 * Some VM hosts require MSIX to be disabled in the
1708 		 * control register before updating the MSIX table
1709 		 * entries are allowed. It is not enough to only
1710 		 * disable MSIX while updating a single entry. MSIX
1711 		 * must be disabled while updating all entries in the
1712 		 * table.
1713 		 */
1714 		pci_write_config(child,
1715 		    msix->msix_location + PCIR_MSIX_CTRL,
1716 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1717 		pci_resume_msix(child);
1718 	} else
1719 		pci_write_msix_entry(child, index, address, data);
1720 
1721 	/* Enable MSI -> HT mapping. */
1722 	pci_ht_map_msi(child, address);
1723 }
1724 
1725 void
pci_mask_msix(device_t dev,u_int index)1726 pci_mask_msix(device_t dev, u_int index)
1727 {
1728 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1729 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1730 	uint32_t offset, val;
1731 
1732 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1733 	offset = msix->msix_table_offset + index * 16 + 12;
1734 	val = bus_read_4(msix->msix_table_res, offset);
1735 	val |= PCIM_MSIX_VCTRL_MASK;
1736 
1737 	/*
1738 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1739 	 * register, so always write the new value.
1740 	 */
1741 	bus_write_4(msix->msix_table_res, offset, val);
1742 }
1743 
1744 void
pci_unmask_msix(device_t dev,u_int index)1745 pci_unmask_msix(device_t dev, u_int index)
1746 {
1747 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1748 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1749 	uint32_t offset, val;
1750 
1751 	KASSERT(PCI_MSIX_MSGNUM(msix->msix_ctrl) > index, ("bogus index"));
1752 	offset = msix->msix_table_offset + index * 16 + 12;
1753 	val = bus_read_4(msix->msix_table_res, offset);
1754 	val &= ~PCIM_MSIX_VCTRL_MASK;
1755 
1756 	/*
1757 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1758 	 * register, so always write the new value.
1759 	 */
1760 	bus_write_4(msix->msix_table_res, offset, val);
1761 }
1762 
1763 int
pci_pending_msix(device_t dev,u_int index)1764 pci_pending_msix(device_t dev, u_int index)
1765 {
1766 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1767 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1768 	uint32_t offset, bit;
1769 
1770 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1771 	offset = msix->msix_pba_offset + (index / 32) * 4;
1772 	bit = 1 << index % 32;
1773 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1774 }
1775 
1776 /*
1777  * Restore MSI-X registers and table during resume.  If MSI-X is
1778  * enabled then walk the virtual table to restore the actual MSI-X
1779  * table.
1780  */
1781 static void
pci_resume_msix(device_t dev)1782 pci_resume_msix(device_t dev)
1783 {
1784 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1785 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1786 	struct msix_table_entry *mte;
1787 	struct msix_vector *mv;
1788 	u_int i, msgnum;
1789 
1790 	if (msix->msix_alloc > 0) {
1791 		msgnum = PCI_MSIX_MSGNUM(msix->msix_ctrl);
1792 
1793 		/* First, mask all vectors. */
1794 		for (i = 0; i < msgnum; i++)
1795 			pci_mask_msix(dev, i);
1796 
1797 		/* Second, program any messages with at least one handler. */
1798 		for (i = 0; i < msix->msix_table_len; i++) {
1799 			mte = &msix->msix_table[i];
1800 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1801 				continue;
1802 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1803 			pci_write_msix_entry(dev, i, mv->mv_address,
1804 			    mv->mv_data);
1805 			pci_unmask_msix(dev, i);
1806 		}
1807 	}
1808 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1809 	    msix->msix_ctrl, 2);
1810 }
1811 
1812 /*
1813  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1814  * returned in *count.  After this function returns, each message will be
1815  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1816  */
1817 int
pci_alloc_msix_method(device_t dev,device_t child,int * count)1818 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1819 {
1820 	struct pci_devinfo *dinfo = device_get_ivars(child);
1821 	pcicfgregs *cfg = &dinfo->cfg;
1822 	struct resource_list_entry *rle;
1823 	u_int actual, i, max;
1824 	int error, irq;
1825 	uint16_t ctrl, msgnum;
1826 
1827 	/* Don't let count == 0 get us into trouble. */
1828 	if (*count < 1)
1829 		return (EINVAL);
1830 
1831 	/* If rid 0 is allocated, then fail. */
1832 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1833 	if (rle != NULL && rle->res != NULL)
1834 		return (ENXIO);
1835 
1836 	/* Already have allocated messages? */
1837 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1838 		return (ENXIO);
1839 
1840 	/* If MSI-X is blacklisted for this system, fail. */
1841 	if (pci_msix_blacklisted())
1842 		return (ENXIO);
1843 
1844 	/* MSI-X capability present? */
1845 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1846 		return (ENODEV);
1847 
1848 	/* Make sure the appropriate BARs are mapped. */
1849 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1850 	    cfg->msix.msix_table_bar);
1851 	if (rle == NULL || rle->res == NULL ||
1852 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1853 		return (ENXIO);
1854 	cfg->msix.msix_table_res = rle->res;
1855 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1856 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1857 		    cfg->msix.msix_pba_bar);
1858 		if (rle == NULL || rle->res == NULL ||
1859 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1860 			return (ENXIO);
1861 	}
1862 	cfg->msix.msix_pba_res = rle->res;
1863 
1864 	ctrl = pci_read_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1865 	    2);
1866 	msgnum = PCI_MSIX_MSGNUM(ctrl);
1867 	if (bootverbose)
1868 		device_printf(child,
1869 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1870 		    *count, msgnum);
1871 	max = min(*count, msgnum);
1872 	for (i = 0; i < max; i++) {
1873 		/* Allocate a message. */
1874 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1875 		if (error) {
1876 			if (i == 0)
1877 				return (error);
1878 			break;
1879 		}
1880 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1881 		    irq, 1);
1882 	}
1883 	actual = i;
1884 
1885 	if (bootverbose) {
1886 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1887 		if (actual == 1)
1888 			device_printf(child, "using IRQ %ju for MSI-X\n",
1889 			    rle->start);
1890 		else {
1891 			bool run;
1892 
1893 			/*
1894 			 * Be fancy and try to print contiguous runs of
1895 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1896 			 * 'run' is true if we are in a range.
1897 			 */
1898 			device_printf(child, "using IRQs %ju", rle->start);
1899 			irq = rle->start;
1900 			run = false;
1901 			for (i = 1; i < actual; i++) {
1902 				rle = resource_list_find(&dinfo->resources,
1903 				    SYS_RES_IRQ, i + 1);
1904 
1905 				/* Still in a run? */
1906 				if (rle->start == irq + 1) {
1907 					run = true;
1908 					irq++;
1909 					continue;
1910 				}
1911 
1912 				/* Finish previous range. */
1913 				if (run) {
1914 					printf("-%d", irq);
1915 					run = false;
1916 				}
1917 
1918 				/* Start new range. */
1919 				printf(",%ju", rle->start);
1920 				irq = rle->start;
1921 			}
1922 
1923 			/* Unfinished range? */
1924 			if (run)
1925 				printf("-%d", irq);
1926 			printf(" for MSI-X\n");
1927 		}
1928 	}
1929 
1930 	/*
1931 	 * Mask all vectors. Note that the message index assertion in
1932 	 * pci_mask_msix requires msix_ctrl to be set.
1933 	 */
1934 	cfg->msix.msix_ctrl = ctrl;
1935 	for (i = 0; i < msgnum; i++)
1936 		pci_mask_msix(child, i);
1937 
1938 	/* Allocate and initialize vector data and virtual table. */
1939 	cfg->msix.msix_vectors = mallocarray(actual, sizeof(struct msix_vector),
1940 	    M_DEVBUF, M_WAITOK | M_ZERO);
1941 	cfg->msix.msix_table = mallocarray(actual,
1942 	    sizeof(struct msix_table_entry), M_DEVBUF, M_WAITOK | M_ZERO);
1943 	for (i = 0; i < actual; i++) {
1944 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1945 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1946 		cfg->msix.msix_table[i].mte_vector = i + 1;
1947 	}
1948 
1949 	/* Update control register to enable MSI-X. */
1950 	ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1951 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1952 	    ctrl, 2);
1953 	cfg->msix.msix_ctrl = ctrl;
1954 
1955 	/* Update counts of alloc'd messages. */
1956 	cfg->msix.msix_alloc = actual;
1957 	cfg->msix.msix_table_len = actual;
1958 	*count = actual;
1959 	return (0);
1960 }
1961 
1962 /*
1963  * By default, pci_alloc_msix() will assign the allocated IRQ
1964  * resources consecutively to the first N messages in the MSI-X table.
1965  * However, device drivers may want to use different layouts if they
1966  * either receive fewer messages than they asked for, or they wish to
1967  * populate the MSI-X table sparsely.  This method allows the driver
1968  * to specify what layout it wants.  It must be called after a
1969  * successful pci_alloc_msix() but before any of the associated
1970  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1971  *
1972  * The 'vectors' array contains 'count' message vectors.  The array
1973  * maps directly to the MSI-X table in that index 0 in the array
1974  * specifies the vector for the first message in the MSI-X table, etc.
1975  * The vector value in each array index can either be 0 to indicate
1976  * that no vector should be assigned to a message slot, or it can be a
1977  * number from 1 to N (where N is the count returned from a
1978  * succcessful call to pci_alloc_msix()) to indicate which message
1979  * vector (IRQ) to be used for the corresponding message.
1980  *
1981  * On successful return, each message with a non-zero vector will have
1982  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1983  * 1.  Additionally, if any of the IRQs allocated via the previous
1984  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1985  * will be freed back to the system automatically.
1986  *
1987  * For example, suppose a driver has a MSI-X table with 6 messages and
1988  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1989  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1990  * C.  After the call to pci_alloc_msix(), the device will be setup to
1991  * have an MSI-X table of ABC--- (where - means no vector assigned).
1992  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1993  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1994  * be freed back to the system.  This device will also have valid
1995  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1996  *
1997  * In any case, the SYS_RES_IRQ rid X will always map to the message
1998  * at MSI-X table index X - 1 and will only be valid if a vector is
1999  * assigned to that table entry.
2000  */
2001 int
pci_remap_msix_method(device_t dev,device_t child,int count,const u_int * vectors)2002 pci_remap_msix_method(device_t dev, device_t child, int count,
2003     const u_int *vectors)
2004 {
2005 	struct pci_devinfo *dinfo = device_get_ivars(child);
2006 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2007 	struct resource_list_entry *rle;
2008 	u_int i, irq, j;
2009 	bool *used;
2010 
2011 	/*
2012 	 * Have to have at least one message in the table but the
2013 	 * table can't be bigger than the actual MSI-X table in the
2014 	 * device.
2015 	 */
2016 	if (count < 1 || count > PCI_MSIX_MSGNUM(msix->msix_ctrl))
2017 		return (EINVAL);
2018 
2019 	/* Sanity check the vectors. */
2020 	for (i = 0; i < count; i++)
2021 		if (vectors[i] > msix->msix_alloc)
2022 			return (EINVAL);
2023 
2024 	/*
2025 	 * Make sure there aren't any holes in the vectors to be used.
2026 	 * It's a big pain to support it, and it doesn't really make
2027 	 * sense anyway.  Also, at least one vector must be used.
2028 	 */
2029 	used = mallocarray(msix->msix_alloc, sizeof(*used), M_DEVBUF, M_WAITOK |
2030 	    M_ZERO);
2031 	for (i = 0; i < count; i++)
2032 		if (vectors[i] != 0)
2033 			used[vectors[i] - 1] = true;
2034 	for (i = 0; i < msix->msix_alloc - 1; i++)
2035 		if (!used[i] && used[i + 1]) {
2036 			free(used, M_DEVBUF);
2037 			return (EINVAL);
2038 		}
2039 	if (!used[0]) {
2040 		free(used, M_DEVBUF);
2041 		return (EINVAL);
2042 	}
2043 
2044 	/* Make sure none of the resources are allocated. */
2045 	for (i = 0; i < msix->msix_table_len; i++) {
2046 		if (msix->msix_table[i].mte_vector == 0)
2047 			continue;
2048 		if (msix->msix_table[i].mte_handlers > 0) {
2049 			free(used, M_DEVBUF);
2050 			return (EBUSY);
2051 		}
2052 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2053 		KASSERT(rle != NULL, ("missing resource"));
2054 		if (rle->res != NULL) {
2055 			free(used, M_DEVBUF);
2056 			return (EBUSY);
2057 		}
2058 	}
2059 
2060 	/* Free the existing resource list entries. */
2061 	for (i = 0; i < msix->msix_table_len; i++) {
2062 		if (msix->msix_table[i].mte_vector == 0)
2063 			continue;
2064 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2065 	}
2066 
2067 	/*
2068 	 * Build the new virtual table keeping track of which vectors are
2069 	 * used.
2070 	 */
2071 	free(msix->msix_table, M_DEVBUF);
2072 	msix->msix_table = mallocarray(count, sizeof(struct msix_table_entry),
2073 	    M_DEVBUF, M_WAITOK | M_ZERO);
2074 	for (i = 0; i < count; i++)
2075 		msix->msix_table[i].mte_vector = vectors[i];
2076 	msix->msix_table_len = count;
2077 
2078 	/* Free any unused IRQs and resize the vectors array if necessary. */
2079 	j = msix->msix_alloc - 1;
2080 	if (!used[j]) {
2081 		struct msix_vector *vec;
2082 
2083 		while (!used[j]) {
2084 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2085 			    msix->msix_vectors[j].mv_irq);
2086 			j--;
2087 		}
2088 		vec = mallocarray(j + 1, sizeof(struct msix_vector), M_DEVBUF,
2089 		    M_WAITOK);
2090 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2091 		    (j + 1));
2092 		free(msix->msix_vectors, M_DEVBUF);
2093 		msix->msix_vectors = vec;
2094 		msix->msix_alloc = j + 1;
2095 	}
2096 	free(used, M_DEVBUF);
2097 
2098 	/* Map the IRQs onto the rids. */
2099 	for (i = 0; i < count; i++) {
2100 		if (vectors[i] == 0)
2101 			continue;
2102 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2103 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2104 		    irq, 1);
2105 	}
2106 
2107 	if (bootverbose) {
2108 		device_printf(child, "Remapped MSI-X IRQs as: ");
2109 		for (i = 0; i < count; i++) {
2110 			if (i != 0)
2111 				printf(", ");
2112 			if (vectors[i] == 0)
2113 				printf("---");
2114 			else
2115 				printf("%d",
2116 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2117 		}
2118 		printf("\n");
2119 	}
2120 
2121 	return (0);
2122 }
2123 
2124 static int
pci_release_msix(device_t dev,device_t child)2125 pci_release_msix(device_t dev, device_t child)
2126 {
2127 	struct pci_devinfo *dinfo = device_get_ivars(child);
2128 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2129 	struct resource_list_entry *rle;
2130 	u_int i;
2131 
2132 	/* Do we have any messages to release? */
2133 	if (msix->msix_alloc == 0)
2134 		return (ENODEV);
2135 
2136 	/* Make sure none of the resources are allocated. */
2137 	for (i = 0; i < msix->msix_table_len; i++) {
2138 		if (msix->msix_table[i].mte_vector == 0)
2139 			continue;
2140 		if (msix->msix_table[i].mte_handlers > 0)
2141 			return (EBUSY);
2142 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2143 		KASSERT(rle != NULL, ("missing resource"));
2144 		if (rle->res != NULL)
2145 			return (EBUSY);
2146 	}
2147 
2148 	/* Update control register to disable MSI-X. */
2149 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2150 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2151 	    msix->msix_ctrl, 2);
2152 
2153 	/* Free the resource list entries. */
2154 	for (i = 0; i < msix->msix_table_len; i++) {
2155 		if (msix->msix_table[i].mte_vector == 0)
2156 			continue;
2157 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2158 	}
2159 	free(msix->msix_table, M_DEVBUF);
2160 	msix->msix_table_len = 0;
2161 
2162 	/* Release the IRQs. */
2163 	for (i = 0; i < msix->msix_alloc; i++)
2164 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2165 		    msix->msix_vectors[i].mv_irq);
2166 	free(msix->msix_vectors, M_DEVBUF);
2167 	msix->msix_alloc = 0;
2168 	return (0);
2169 }
2170 
2171 /*
2172  * Return the max supported MSI-X messages this device supports.
2173  * Basically, assuming the MD code can alloc messages, this function
2174  * should return the maximum value that pci_alloc_msix() can return.
2175  * Thus, it is subject to the tunables, etc.
2176  */
2177 int
pci_msix_count_method(device_t dev,device_t child)2178 pci_msix_count_method(device_t dev, device_t child)
2179 {
2180 	struct pci_devinfo *dinfo = device_get_ivars(child);
2181 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2182 	uint16_t ctrl;
2183 
2184 	if (pci_do_msix && msix->msix_location != 0) {
2185 		ctrl = pci_read_config(child, msix->msix_location +
2186 		    PCIR_MSI_CTRL, 2);
2187 		return (PCI_MSIX_MSGNUM(ctrl));
2188 	}
2189 	return (0);
2190 }
2191 
2192 int
pci_msix_pba_bar_method(device_t dev,device_t child)2193 pci_msix_pba_bar_method(device_t dev, device_t child)
2194 {
2195 	struct pci_devinfo *dinfo = device_get_ivars(child);
2196 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2197 
2198 	if (pci_do_msix && msix->msix_location != 0)
2199 		return (msix->msix_pba_bar);
2200 	return (-1);
2201 }
2202 
2203 int
pci_msix_table_bar_method(device_t dev,device_t child)2204 pci_msix_table_bar_method(device_t dev, device_t child)
2205 {
2206 	struct pci_devinfo *dinfo = device_get_ivars(child);
2207 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2208 
2209 	if (pci_do_msix && msix->msix_location != 0)
2210 		return (msix->msix_table_bar);
2211 	return (-1);
2212 }
2213 
2214 /*
2215  * HyperTransport MSI mapping control
2216  */
2217 void
pci_ht_map_msi(device_t dev,uint64_t addr)2218 pci_ht_map_msi(device_t dev, uint64_t addr)
2219 {
2220 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2221 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2222 
2223 	if (!ht->ht_msimap)
2224 		return;
2225 
2226 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2227 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2228 		/* Enable MSI -> HT mapping. */
2229 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2230 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2231 		    ht->ht_msictrl, 2);
2232 	}
2233 
2234 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2235 		/* Disable MSI -> HT mapping. */
2236 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2237 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2238 		    ht->ht_msictrl, 2);
2239 	}
2240 }
2241 
2242 int
pci_get_relaxed_ordering_enabled(device_t dev)2243 pci_get_relaxed_ordering_enabled(device_t dev)
2244 {
2245 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2246 	int cap;
2247 	uint16_t val;
2248 
2249 	cap = dinfo->cfg.pcie.pcie_location;
2250 	if (cap == 0)
2251 		return (0);
2252 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2253 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2254 	return (val != 0);
2255 }
2256 
2257 int
pci_get_max_payload(device_t dev)2258 pci_get_max_payload(device_t dev)
2259 {
2260 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2261 	int cap;
2262 	uint16_t val;
2263 
2264 	cap = dinfo->cfg.pcie.pcie_location;
2265 	if (cap == 0)
2266 		return (0);
2267 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2268 	val &= PCIEM_CTL_MAX_PAYLOAD;
2269 	val >>= 5;
2270 	return (1 << (val + 7));
2271 }
2272 
2273 int
pci_get_max_read_req(device_t dev)2274 pci_get_max_read_req(device_t dev)
2275 {
2276 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2277 	int cap;
2278 	uint16_t val;
2279 
2280 	cap = dinfo->cfg.pcie.pcie_location;
2281 	if (cap == 0)
2282 		return (0);
2283 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2284 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2285 	val >>= 12;
2286 	return (1 << (val + 7));
2287 }
2288 
2289 int
pci_set_max_read_req(device_t dev,int size)2290 pci_set_max_read_req(device_t dev, int size)
2291 {
2292 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2293 	int cap;
2294 	uint16_t val;
2295 
2296 	cap = dinfo->cfg.pcie.pcie_location;
2297 	if (cap == 0)
2298 		return (0);
2299 	if (size < 128)
2300 		size = 128;
2301 	if (size > 4096)
2302 		size = 4096;
2303 	size = (1 << (fls(size) - 1));
2304 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2305 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2306 	val |= (fls(size) - 8) << 12;
2307 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2308 	return (size);
2309 }
2310 
2311 uint32_t
pcie_read_config(device_t dev,int reg,int width)2312 pcie_read_config(device_t dev, int reg, int width)
2313 {
2314 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2315 	int cap;
2316 
2317 	cap = dinfo->cfg.pcie.pcie_location;
2318 	if (cap == 0) {
2319 		if (width == 2)
2320 			return (0xffff);
2321 		return (0xffffffff);
2322 	}
2323 
2324 	return (pci_read_config(dev, cap + reg, width));
2325 }
2326 
2327 void
pcie_write_config(device_t dev,int reg,uint32_t value,int width)2328 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2329 {
2330 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2331 	int cap;
2332 
2333 	cap = dinfo->cfg.pcie.pcie_location;
2334 	if (cap == 0)
2335 		return;
2336 	pci_write_config(dev, cap + reg, value, width);
2337 }
2338 
2339 /*
2340  * Adjusts a PCI-e capability register by clearing the bits in mask
2341  * and setting the bits in (value & mask).  Bits not set in mask are
2342  * not adjusted.
2343  *
2344  * Returns the old value on success or all ones on failure.
2345  */
2346 uint32_t
pcie_adjust_config(device_t dev,int reg,uint32_t mask,uint32_t value,int width)2347 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2348     int width)
2349 {
2350 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2351 	uint32_t old, new;
2352 	int cap;
2353 
2354 	cap = dinfo->cfg.pcie.pcie_location;
2355 	if (cap == 0) {
2356 		if (width == 2)
2357 			return (0xffff);
2358 		return (0xffffffff);
2359 	}
2360 
2361 	old = pci_read_config(dev, cap + reg, width);
2362 	new = old & ~mask;
2363 	new |= (value & mask);
2364 	pci_write_config(dev, cap + reg, new, width);
2365 	return (old);
2366 }
2367 
2368 /*
2369  * Support for MSI message signalled interrupts.
2370  */
2371 void
pci_enable_msi_method(device_t dev,device_t child,uint64_t address,uint16_t data)2372 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2373     uint16_t data)
2374 {
2375 	struct pci_devinfo *dinfo = device_get_ivars(child);
2376 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2377 
2378 	/* Write data and address values. */
2379 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2380 	    address & 0xffffffff, 4);
2381 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2382 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2383 		    address >> 32, 4);
2384 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2385 		    data, 2);
2386 	} else
2387 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2388 		    2);
2389 
2390 	/* Enable MSI in the control register. */
2391 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2392 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2393 	    msi->msi_ctrl, 2);
2394 
2395 	/* Enable MSI -> HT mapping. */
2396 	pci_ht_map_msi(child, address);
2397 }
2398 
2399 void
pci_disable_msi_method(device_t dev,device_t child)2400 pci_disable_msi_method(device_t dev, device_t child)
2401 {
2402 	struct pci_devinfo *dinfo = device_get_ivars(child);
2403 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2404 
2405 	/* Disable MSI -> HT mapping. */
2406 	pci_ht_map_msi(child, 0);
2407 
2408 	/* Disable MSI in the control register. */
2409 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2410 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2411 	    msi->msi_ctrl, 2);
2412 }
2413 
2414 /*
2415  * Restore MSI registers during resume.  If MSI is enabled then
2416  * restore the data and address registers in addition to the control
2417  * register.
2418  */
2419 static void
pci_resume_msi(device_t dev)2420 pci_resume_msi(device_t dev)
2421 {
2422 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2423 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2424 	uint64_t address;
2425 	uint16_t data;
2426 
2427 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2428 		address = msi->msi_addr;
2429 		data = msi->msi_data;
2430 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2431 		    address & 0xffffffff, 4);
2432 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2433 			pci_write_config(dev, msi->msi_location +
2434 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2435 			pci_write_config(dev, msi->msi_location +
2436 			    PCIR_MSI_DATA_64BIT, data, 2);
2437 		} else
2438 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2439 			    data, 2);
2440 	}
2441 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2442 	    2);
2443 }
2444 
2445 static int
pci_remap_intr_method(device_t bus,device_t dev,u_int irq)2446 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2447 {
2448 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2449 	pcicfgregs *cfg = &dinfo->cfg;
2450 	struct resource_list_entry *rle;
2451 	struct msix_table_entry *mte;
2452 	struct msix_vector *mv;
2453 	uint64_t addr;
2454 	uint32_t data;
2455 	u_int i, j;
2456 	int error;
2457 
2458 	/*
2459 	 * Handle MSI first.  We try to find this IRQ among our list
2460 	 * of MSI IRQs.  If we find it, we request updated address and
2461 	 * data registers and apply the results.
2462 	 */
2463 	if (cfg->msi.msi_alloc > 0) {
2464 		/* If we don't have any active handlers, nothing to do. */
2465 		if (cfg->msi.msi_handlers == 0)
2466 			return (0);
2467 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2468 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2469 			    i + 1);
2470 			if (rle->start == irq) {
2471 				error = PCIB_MAP_MSI(device_get_parent(bus),
2472 				    dev, irq, &addr, &data);
2473 				if (error)
2474 					return (error);
2475 				pci_disable_msi(dev);
2476 				dinfo->cfg.msi.msi_addr = addr;
2477 				dinfo->cfg.msi.msi_data = data;
2478 				pci_enable_msi(dev, addr, data);
2479 				return (0);
2480 			}
2481 		}
2482 		return (ENOENT);
2483 	}
2484 
2485 	/*
2486 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2487 	 * we request the updated mapping info.  If that works, we go
2488 	 * through all the slots that use this IRQ and update them.
2489 	 */
2490 	if (cfg->msix.msix_alloc > 0) {
2491 		bool found = false;
2492 
2493 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2494 			mv = &cfg->msix.msix_vectors[i];
2495 			if (mv->mv_irq == irq) {
2496 				error = PCIB_MAP_MSI(device_get_parent(bus),
2497 				    dev, irq, &addr, &data);
2498 				if (error)
2499 					return (error);
2500 				mv->mv_address = addr;
2501 				mv->mv_data = data;
2502 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2503 					mte = &cfg->msix.msix_table[j];
2504 					if (mte->mte_vector != i + 1)
2505 						continue;
2506 					if (mte->mte_handlers == 0)
2507 						continue;
2508 					pci_mask_msix(dev, j);
2509 					pci_enable_msix(dev, j, addr, data);
2510 					pci_unmask_msix(dev, j);
2511 				}
2512 				found = true;
2513 			}
2514 		}
2515 		return (found ? 0 : ENOENT);
2516 	}
2517 
2518 	return (ENOENT);
2519 }
2520 
2521 /*
2522  * Returns true if the specified device is blacklisted because MSI
2523  * doesn't work.
2524  */
2525 int
pci_msi_device_blacklisted(device_t dev)2526 pci_msi_device_blacklisted(device_t dev)
2527 {
2528 
2529 	if (!pci_honor_msi_blacklist)
2530 		return (0);
2531 
2532 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2533 }
2534 
2535 /*
2536  * Determine if MSI is blacklisted globally on this system.  Currently,
2537  * we just check for blacklisted chipsets as represented by the
2538  * host-PCI bridge at device 0:0:0.  In the future, it may become
2539  * necessary to check other system attributes, such as the kenv values
2540  * that give the motherboard manufacturer and model number.
2541  */
2542 static int
pci_msi_blacklisted(void)2543 pci_msi_blacklisted(void)
2544 {
2545 	device_t dev;
2546 
2547 	if (!pci_honor_msi_blacklist)
2548 		return (0);
2549 
2550 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2551 	if (!(pcie_chipset || pcix_chipset)) {
2552 		if (vm_guest != VM_GUEST_NO) {
2553 			/*
2554 			 * Whitelist older chipsets in virtual
2555 			 * machines known to support MSI.
2556 			 */
2557 			dev = pci_find_bsf(0, 0, 0);
2558 			if (dev != NULL)
2559 				return (!pci_has_quirk(pci_get_devid(dev),
2560 					PCI_QUIRK_ENABLE_MSI_VM));
2561 		}
2562 		return (1);
2563 	}
2564 
2565 	dev = pci_find_bsf(0, 0, 0);
2566 	if (dev != NULL)
2567 		return (pci_msi_device_blacklisted(dev));
2568 	return (0);
2569 }
2570 
2571 /*
2572  * Returns true if the specified device is blacklisted because MSI-X
2573  * doesn't work.  Note that this assumes that if MSI doesn't work,
2574  * MSI-X doesn't either.
2575  */
2576 int
pci_msix_device_blacklisted(device_t dev)2577 pci_msix_device_blacklisted(device_t dev)
2578 {
2579 
2580 	if (!pci_honor_msi_blacklist)
2581 		return (0);
2582 
2583 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2584 		return (1);
2585 
2586 	return (pci_msi_device_blacklisted(dev));
2587 }
2588 
2589 /*
2590  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2591  * is blacklisted, assume that MSI-X is as well.  Check for additional
2592  * chipsets where MSI works but MSI-X does not.
2593  */
2594 static int
pci_msix_blacklisted(void)2595 pci_msix_blacklisted(void)
2596 {
2597 	device_t dev;
2598 
2599 	if (!pci_honor_msi_blacklist)
2600 		return (0);
2601 
2602 	dev = pci_find_bsf(0, 0, 0);
2603 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2604 	    PCI_QUIRK_DISABLE_MSIX))
2605 		return (1);
2606 
2607 	return (pci_msi_blacklisted());
2608 }
2609 
2610 /*
2611  * Attempt to allocate *count MSI messages.  The actual number allocated is
2612  * returned in *count.  After this function returns, each message will be
2613  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2614  */
2615 int
pci_alloc_msi_method(device_t dev,device_t child,int * count)2616 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2617 {
2618 	struct pci_devinfo *dinfo = device_get_ivars(child);
2619 	pcicfgregs *cfg = &dinfo->cfg;
2620 	struct resource_list_entry *rle;
2621 	u_int actual, i;
2622 	int error, irqs[32];
2623 	uint16_t ctrl, msgnum;
2624 
2625 	/* Don't let count == 0 get us into trouble. */
2626 	if (*count < 1)
2627 		return (EINVAL);
2628 
2629 	/* If rid 0 is allocated, then fail. */
2630 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2631 	if (rle != NULL && rle->res != NULL)
2632 		return (ENXIO);
2633 
2634 	/* Already have allocated messages? */
2635 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2636 		return (ENXIO);
2637 
2638 	/* If MSI is blacklisted for this system, fail. */
2639 	if (pci_msi_blacklisted())
2640 		return (ENXIO);
2641 
2642 	/* MSI capability present? */
2643 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2644 		return (ENODEV);
2645 
2646 	ctrl = pci_read_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, 2);
2647 	msgnum = PCI_MSI_MSGNUM(ctrl);
2648 	if (bootverbose)
2649 		device_printf(child,
2650 		    "attempting to allocate %d MSI vectors (%u supported)\n",
2651 		    *count, msgnum);
2652 
2653 	/* Don't ask for more than the device supports. */
2654 	actual = min(*count, msgnum);
2655 
2656 	/* Don't ask for more than 32 messages. */
2657 	actual = min(actual, 32);
2658 
2659 	/* MSI requires power of 2 number of messages. */
2660 	if (!powerof2(actual))
2661 		return (EINVAL);
2662 
2663 	for (;;) {
2664 		/* Try to allocate N messages. */
2665 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2666 		    actual, irqs);
2667 		if (error == 0)
2668 			break;
2669 		if (actual == 1)
2670 			return (error);
2671 
2672 		/* Try N / 2. */
2673 		actual >>= 1;
2674 	}
2675 
2676 	/*
2677 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2678 	 * resources in the irqs[] array, so add new resources
2679 	 * starting at rid 1.
2680 	 */
2681 	for (i = 0; i < actual; i++)
2682 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2683 		    irqs[i], irqs[i], 1);
2684 
2685 	if (bootverbose) {
2686 		if (actual == 1)
2687 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2688 		else {
2689 			bool run;
2690 
2691 			/*
2692 			 * Be fancy and try to print contiguous runs
2693 			 * of IRQ values as ranges.  'run' is true if
2694 			 * we are in a range.
2695 			 */
2696 			device_printf(child, "using IRQs %d", irqs[0]);
2697 			run = false;
2698 			for (i = 1; i < actual; i++) {
2699 				/* Still in a run? */
2700 				if (irqs[i] == irqs[i - 1] + 1) {
2701 					run = true;
2702 					continue;
2703 				}
2704 
2705 				/* Finish previous range. */
2706 				if (run) {
2707 					printf("-%d", irqs[i - 1]);
2708 					run = false;
2709 				}
2710 
2711 				/* Start new range. */
2712 				printf(",%d", irqs[i]);
2713 			}
2714 
2715 			/* Unfinished range? */
2716 			if (run)
2717 				printf("-%d", irqs[actual - 1]);
2718 			printf(" for MSI\n");
2719 		}
2720 	}
2721 
2722 	/* Update control register with actual count. */
2723 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2724 	ctrl |= (ffs(actual) - 1) << 4;
2725 	cfg->msi.msi_ctrl = ctrl;
2726 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2727 
2728 	/* Update counts of alloc'd messages. */
2729 	cfg->msi.msi_alloc = actual;
2730 	cfg->msi.msi_handlers = 0;
2731 	*count = actual;
2732 	return (0);
2733 }
2734 
2735 /* Release the MSI messages associated with this device. */
2736 int
pci_release_msi_method(device_t dev,device_t child)2737 pci_release_msi_method(device_t dev, device_t child)
2738 {
2739 	struct pci_devinfo *dinfo = device_get_ivars(child);
2740 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2741 	struct resource_list_entry *rle;
2742 	u_int i, irqs[32];
2743 	int error;
2744 
2745 	/* Try MSI-X first. */
2746 	error = pci_release_msix(dev, child);
2747 	if (error != ENODEV)
2748 		return (error);
2749 
2750 	/* Do we have any messages to release? */
2751 	if (msi->msi_alloc == 0)
2752 		return (ENODEV);
2753 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2754 
2755 	/* Make sure none of the resources are allocated. */
2756 	if (msi->msi_handlers > 0)
2757 		return (EBUSY);
2758 	for (i = 0; i < msi->msi_alloc; i++) {
2759 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2760 		KASSERT(rle != NULL, ("missing MSI resource"));
2761 		if (rle->res != NULL)
2762 			return (EBUSY);
2763 		irqs[i] = rle->start;
2764 	}
2765 
2766 	/* Update control register with 0 count. */
2767 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2768 	    ("%s: MSI still enabled", __func__));
2769 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2770 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2771 	    msi->msi_ctrl, 2);
2772 
2773 	/* Release the messages. */
2774 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2775 	for (i = 0; i < msi->msi_alloc; i++)
2776 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2777 
2778 	/* Update alloc count. */
2779 	msi->msi_alloc = 0;
2780 	msi->msi_addr = 0;
2781 	msi->msi_data = 0;
2782 	return (0);
2783 }
2784 
2785 /*
2786  * Return the max supported MSI messages this device supports.
2787  * Basically, assuming the MD code can alloc messages, this function
2788  * should return the maximum value that pci_alloc_msi() can return.
2789  * Thus, it is subject to the tunables, etc.
2790  */
2791 int
pci_msi_count_method(device_t dev,device_t child)2792 pci_msi_count_method(device_t dev, device_t child)
2793 {
2794 	struct pci_devinfo *dinfo = device_get_ivars(child);
2795 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2796 	uint16_t ctrl;
2797 
2798 	if (pci_do_msi && msi->msi_location != 0) {
2799 		ctrl = pci_read_config(child, msi->msi_location + PCIR_MSI_CTRL,
2800 		    2);
2801 		return (PCI_MSI_MSGNUM(ctrl));
2802 	}
2803 	return (0);
2804 }
2805 
2806 /* free pcicfgregs structure and all depending data structures */
2807 
2808 int
pci_freecfg(struct pci_devinfo * dinfo)2809 pci_freecfg(struct pci_devinfo *dinfo)
2810 {
2811 	struct devlist *devlist_head;
2812 	struct pci_map *pm, *next;
2813 
2814 	devlist_head = &pci_devq;
2815 
2816 	if (dinfo->cfg.vpd.vpd_reg)
2817 		vpd_free(&dinfo->cfg.vpd);
2818 
2819 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2820 		free(pm, M_DEVBUF);
2821 	}
2822 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2823 	free(dinfo, M_DEVBUF);
2824 
2825 	/* increment the generation count */
2826 	pci_generation++;
2827 
2828 	/* we're losing one device */
2829 	pci_numdevs--;
2830 	return (0);
2831 }
2832 
2833 /*
2834  * PCI power manangement
2835  */
2836 int
pci_set_powerstate_method(device_t dev,device_t child,int state)2837 pci_set_powerstate_method(device_t dev, device_t child, int state)
2838 {
2839 	struct pci_devinfo *dinfo = device_get_ivars(child);
2840 	pcicfgregs *cfg = &dinfo->cfg;
2841 	uint16_t status;
2842 	int oldstate, highest, delay;
2843 
2844 	if (cfg->pp.pp_location == 0)
2845 		return (EOPNOTSUPP);
2846 
2847 	/*
2848 	 * Optimize a no state change request away.  While it would be OK to
2849 	 * write to the hardware in theory, some devices have shown odd
2850 	 * behavior when going from D3 -> D3.
2851 	 */
2852 	oldstate = pci_get_powerstate(child);
2853 	if (oldstate == state)
2854 		return (0);
2855 
2856 	/*
2857 	 * The PCI power management specification states that after a state
2858 	 * transition between PCI power states, system software must
2859 	 * guarantee a minimal delay before the function accesses the device.
2860 	 * Compute the worst case delay that we need to guarantee before we
2861 	 * access the device.  Many devices will be responsive much more
2862 	 * quickly than this delay, but there are some that don't respond
2863 	 * instantly to state changes.  Transitions to/from D3 state require
2864 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2865 	 * is done below with DELAY rather than a sleeper function because
2866 	 * this function can be called from contexts where we cannot sleep.
2867 	 */
2868 	highest = (oldstate > state) ? oldstate : state;
2869 	if (highest == PCI_POWERSTATE_D3)
2870 	    delay = 10000;
2871 	else if (highest == PCI_POWERSTATE_D2)
2872 	    delay = 200;
2873 	else
2874 	    delay = 0;
2875 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2876 	    PCIR_POWER_STATUS, 2) & ~PCIM_PSTAT_DMASK;
2877 	switch (state) {
2878 	case PCI_POWERSTATE_D0:
2879 		status |= PCIM_PSTAT_D0;
2880 		break;
2881 	case PCI_POWERSTATE_D1:
2882 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2883 			return (EOPNOTSUPP);
2884 		status |= PCIM_PSTAT_D1;
2885 		break;
2886 	case PCI_POWERSTATE_D2:
2887 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2888 			return (EOPNOTSUPP);
2889 		status |= PCIM_PSTAT_D2;
2890 		break;
2891 	case PCI_POWERSTATE_D3:
2892 		status |= PCIM_PSTAT_D3;
2893 		break;
2894 	default:
2895 		return (EINVAL);
2896 	}
2897 
2898 	if (bootverbose)
2899 		pci_printf(cfg, "Transition from %s to %s\n",
2900 		    pci_powerstate_to_str(oldstate),
2901 		    pci_powerstate_to_str(state));
2902 
2903 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_location + PCIR_POWER_STATUS,
2904 	    status, 2);
2905 	if (delay)
2906 		DELAY(delay);
2907 	return (0);
2908 }
2909 
2910 int
pci_get_powerstate_method(device_t dev,device_t child)2911 pci_get_powerstate_method(device_t dev, device_t child)
2912 {
2913 	struct pci_devinfo *dinfo = device_get_ivars(child);
2914 	pcicfgregs *cfg = &dinfo->cfg;
2915 	uint16_t status;
2916 	int result;
2917 
2918 	if (cfg->pp.pp_location != 0) {
2919 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_location +
2920 		    PCIR_POWER_STATUS, 2);
2921 		switch (status & PCIM_PSTAT_DMASK) {
2922 		case PCIM_PSTAT_D0:
2923 			result = PCI_POWERSTATE_D0;
2924 			break;
2925 		case PCIM_PSTAT_D1:
2926 			result = PCI_POWERSTATE_D1;
2927 			break;
2928 		case PCIM_PSTAT_D2:
2929 			result = PCI_POWERSTATE_D2;
2930 			break;
2931 		case PCIM_PSTAT_D3:
2932 			result = PCI_POWERSTATE_D3;
2933 			break;
2934 		default:
2935 			result = PCI_POWERSTATE_UNKNOWN;
2936 			break;
2937 		}
2938 	} else {
2939 		/* No support, device is always at D0 */
2940 		result = PCI_POWERSTATE_D0;
2941 	}
2942 	return (result);
2943 }
2944 
2945 /* Clear any active PME# and disable PME# generation. */
2946 void
pci_clear_pme(device_t dev)2947 pci_clear_pme(device_t dev)
2948 {
2949 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2950 	pcicfgregs *cfg = &dinfo->cfg;
2951 	uint16_t status;
2952 
2953 	if (cfg->pp.pp_location != 0) {
2954 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2955 		    PCIR_POWER_STATUS, 2);
2956 		status &= ~PCIM_PSTAT_PMEENABLE;
2957 		status |= PCIM_PSTAT_PME;
2958 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2959 		    PCIR_POWER_STATUS, status, 2);
2960 	}
2961 }
2962 
2963 /* Clear any active PME# and enable PME# generation. */
2964 void
pci_enable_pme(device_t dev)2965 pci_enable_pme(device_t dev)
2966 {
2967 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2968 	pcicfgregs *cfg = &dinfo->cfg;
2969 	uint16_t status;
2970 
2971 	if (cfg->pp.pp_location != 0) {
2972 		status = pci_read_config(dev, dinfo->cfg.pp.pp_location +
2973 		    PCIR_POWER_STATUS, 2);
2974 		status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2975 		pci_write_config(dev, dinfo->cfg.pp.pp_location +
2976 		    PCIR_POWER_STATUS, status, 2);
2977 	}
2978 }
2979 
2980 bool
pci_has_pm(device_t dev)2981 pci_has_pm(device_t dev)
2982 {
2983 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2984 	pcicfgregs *cfg = &dinfo->cfg;
2985 
2986 	return (cfg->pp.pp_location != 0);
2987 }
2988 
2989 /*
2990  * Some convenience functions for PCI device drivers.
2991  */
2992 
2993 static __inline void
pci_set_command_bit(device_t dev,device_t child,uint16_t bit)2994 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2995 {
2996 	uint16_t	command;
2997 
2998 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2999 	command |= bit;
3000 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3001 }
3002 
3003 static __inline void
pci_clear_command_bit(device_t dev,device_t child,uint16_t bit)3004 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
3005 {
3006 	uint16_t	command;
3007 
3008 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
3009 	command &= ~bit;
3010 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
3011 }
3012 
3013 int
pci_enable_busmaster_method(device_t dev,device_t child)3014 pci_enable_busmaster_method(device_t dev, device_t child)
3015 {
3016 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3017 	return (0);
3018 }
3019 
3020 int
pci_disable_busmaster_method(device_t dev,device_t child)3021 pci_disable_busmaster_method(device_t dev, device_t child)
3022 {
3023 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
3024 	return (0);
3025 }
3026 
3027 int
pci_enable_io_method(device_t dev,device_t child,int space)3028 pci_enable_io_method(device_t dev, device_t child, int space)
3029 {
3030 	uint16_t bit;
3031 
3032 	switch(space) {
3033 	case SYS_RES_IOPORT:
3034 		bit = PCIM_CMD_PORTEN;
3035 		break;
3036 	case SYS_RES_MEMORY:
3037 		bit = PCIM_CMD_MEMEN;
3038 		break;
3039 	default:
3040 		return (EINVAL);
3041 	}
3042 	pci_set_command_bit(dev, child, bit);
3043 	return (0);
3044 }
3045 
3046 int
pci_disable_io_method(device_t dev,device_t child,int space)3047 pci_disable_io_method(device_t dev, device_t child, int space)
3048 {
3049 	uint16_t bit;
3050 
3051 	switch(space) {
3052 	case SYS_RES_IOPORT:
3053 		bit = PCIM_CMD_PORTEN;
3054 		break;
3055 	case SYS_RES_MEMORY:
3056 		bit = PCIM_CMD_MEMEN;
3057 		break;
3058 	default:
3059 		return (EINVAL);
3060 	}
3061 	pci_clear_command_bit(dev, child, bit);
3062 	return (0);
3063 }
3064 
3065 /*
3066  * New style pci driver.  Parent device is either a pci-host-bridge or a
3067  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
3068  */
3069 
3070 void
pci_print_verbose(struct pci_devinfo * dinfo)3071 pci_print_verbose(struct pci_devinfo *dinfo)
3072 {
3073 
3074 	if (bootverbose) {
3075 		pcicfgregs *cfg = &dinfo->cfg;
3076 
3077 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
3078 		    cfg->vendor, cfg->device, cfg->revid);
3079 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
3080 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
3081 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
3082 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
3083 		    cfg->mfdev);
3084 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
3085 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
3086 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
3087 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3088 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3089 		if (cfg->intpin > 0)
3090 			printf("\tintpin=%c, irq=%d\n",
3091 			    cfg->intpin +'a' -1, cfg->intline);
3092 		if (cfg->pp.pp_location) {
3093 			uint16_t status;
3094 
3095 			status = pci_read_config(cfg->dev, cfg->pp.pp_location +
3096 			    PCIR_POWER_STATUS, 2);
3097 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3098 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3099 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3100 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3101 			    status & PCIM_PSTAT_DMASK);
3102 		}
3103 		if (cfg->msi.msi_location) {
3104 			uint16_t ctrl, msgnum;
3105 
3106 			ctrl = cfg->msi.msi_ctrl;
3107 			msgnum = PCI_MSI_MSGNUM(ctrl);
3108 			printf("\tMSI supports %d message%s%s%s\n",
3109 			    msgnum, (msgnum == 1) ? "" : "s",
3110 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3111 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3112 		}
3113 		if (cfg->msix.msix_location) {
3114 			uint16_t msgnum;
3115 
3116 			msgnum = PCI_MSIX_MSGNUM(cfg->msix.msix_ctrl);
3117 			printf("\tMSI-X supports %d message%s ",
3118 			    msgnum, (msgnum == 1) ? "" : "s");
3119 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3120 				printf("in map 0x%x\n",
3121 				    cfg->msix.msix_table_bar);
3122 			else
3123 				printf("in maps 0x%x and 0x%x\n",
3124 				    cfg->msix.msix_table_bar,
3125 				    cfg->msix.msix_pba_bar);
3126 		}
3127 	}
3128 }
3129 
3130 static int
pci_porten(device_t dev)3131 pci_porten(device_t dev)
3132 {
3133 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3134 }
3135 
3136 static int
pci_memen(device_t dev)3137 pci_memen(device_t dev)
3138 {
3139 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3140 }
3141 
3142 void
pci_read_bar(device_t dev,int reg,pci_addr_t * mapp,pci_addr_t * testvalp,int * bar64)3143 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3144     int *bar64)
3145 {
3146 	struct pci_devinfo *dinfo;
3147 	pci_addr_t map, testval;
3148 	int ln2range;
3149 	uint16_t cmd;
3150 
3151 	/*
3152 	 * The device ROM BAR is special.  It is always a 32-bit
3153 	 * memory BAR.  Bit 0 is special and should not be set when
3154 	 * sizing the BAR.
3155 	 */
3156 	dinfo = device_get_ivars(dev);
3157 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3158 		map = pci_read_config(dev, reg, 4);
3159 		pci_write_config(dev, reg, 0xfffffffe, 4);
3160 		testval = pci_read_config(dev, reg, 4);
3161 		pci_write_config(dev, reg, map, 4);
3162 		*mapp = map;
3163 		*testvalp = testval;
3164 		if (bar64 != NULL)
3165 			*bar64 = 0;
3166 		return;
3167 	}
3168 
3169 	map = pci_read_config(dev, reg, 4);
3170 	ln2range = pci_maprange(map);
3171 	if (ln2range == 64)
3172 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3173 
3174 	/*
3175 	 * Disable decoding via the command register before
3176 	 * determining the BAR's length since we will be placing it in
3177 	 * a weird state.
3178 	 */
3179 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3180 	pci_write_config(dev, PCIR_COMMAND,
3181 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3182 
3183 	/*
3184 	 * Determine the BAR's length by writing all 1's.  The bottom
3185 	 * log_2(size) bits of the BAR will stick as 0 when we read
3186 	 * the value back.
3187 	 *
3188 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3189 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3190 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3191 	 *
3192 	 * Writes to both registers must be performed before attempting to
3193 	 * read back the size value.
3194 	 */
3195 	testval = 0;
3196 	pci_write_config(dev, reg, 0xffffffff, 4);
3197 	if (ln2range == 64) {
3198 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3199 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3200 	}
3201 	testval |= pci_read_config(dev, reg, 4);
3202 
3203 	/*
3204 	 * Restore the original value of the BAR.  We may have reprogrammed
3205 	 * the BAR of the low-level console device and when booting verbose,
3206 	 * we need the console device addressable.
3207 	 */
3208 	pci_write_config(dev, reg, map, 4);
3209 	if (ln2range == 64)
3210 		pci_write_config(dev, reg + 4, map >> 32, 4);
3211 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3212 
3213 	*mapp = map;
3214 	*testvalp = testval;
3215 	if (bar64 != NULL)
3216 		*bar64 = (ln2range == 64);
3217 }
3218 
3219 static void
pci_write_bar(device_t dev,struct pci_map * pm,pci_addr_t base)3220 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3221 {
3222 	struct pci_devinfo *dinfo;
3223 	int ln2range;
3224 
3225 	/* The device ROM BAR is always a 32-bit memory BAR. */
3226 	dinfo = device_get_ivars(dev);
3227 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3228 		ln2range = 32;
3229 	else
3230 		ln2range = pci_maprange(pm->pm_value);
3231 	pci_write_config(dev, pm->pm_reg, base, 4);
3232 	if (ln2range == 64)
3233 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3234 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3235 	if (ln2range == 64)
3236 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3237 		    pm->pm_reg + 4, 4) << 32;
3238 }
3239 
3240 struct pci_map *
pci_find_bar(device_t dev,int reg)3241 pci_find_bar(device_t dev, int reg)
3242 {
3243 	struct pci_devinfo *dinfo;
3244 	struct pci_map *pm;
3245 
3246 	dinfo = device_get_ivars(dev);
3247 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3248 		if (pm->pm_reg == reg)
3249 			return (pm);
3250 	}
3251 	return (NULL);
3252 }
3253 
3254 struct pci_map *
pci_first_bar(device_t dev)3255 pci_first_bar(device_t dev)
3256 {
3257 	struct pci_devinfo *dinfo;
3258 
3259 	dinfo = device_get_ivars(dev);
3260 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3261 }
3262 
3263 struct pci_map *
pci_next_bar(struct pci_map * pm)3264 pci_next_bar(struct pci_map *pm)
3265 {
3266 	return (STAILQ_NEXT(pm, pm_link));
3267 }
3268 
3269 int
pci_bar_enabled(device_t dev,struct pci_map * pm)3270 pci_bar_enabled(device_t dev, struct pci_map *pm)
3271 {
3272 	struct pci_devinfo *dinfo;
3273 	uint16_t cmd;
3274 
3275 	dinfo = device_get_ivars(dev);
3276 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3277 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3278 		return (0);
3279 #ifdef PCI_IOV
3280 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3281 		struct pcicfg_iov *iov;
3282 
3283 		iov = dinfo->cfg.iov;
3284 		cmd = pci_read_config(iov->iov_pf,
3285 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3286 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3287 	}
3288 #endif
3289 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3290 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3291 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3292 	else
3293 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3294 }
3295 
3296 struct pci_map *
pci_add_bar(device_t dev,int reg,pci_addr_t value,pci_addr_t size)3297 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3298 {
3299 	struct pci_devinfo *dinfo;
3300 	struct pci_map *pm, *prev;
3301 
3302 	dinfo = device_get_ivars(dev);
3303 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3304 	pm->pm_reg = reg;
3305 	pm->pm_value = value;
3306 	pm->pm_size = size;
3307 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3308 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3309 		    reg));
3310 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3311 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3312 			break;
3313 	}
3314 	if (prev != NULL)
3315 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3316 	else
3317 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3318 	return (pm);
3319 }
3320 
3321 static void
pci_restore_bars(device_t dev)3322 pci_restore_bars(device_t dev)
3323 {
3324 	struct pci_devinfo *dinfo;
3325 	struct pci_map *pm;
3326 	int ln2range;
3327 
3328 	dinfo = device_get_ivars(dev);
3329 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3330 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3331 			ln2range = 32;
3332 		else
3333 			ln2range = pci_maprange(pm->pm_value);
3334 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3335 		if (ln2range == 64)
3336 			pci_write_config(dev, pm->pm_reg + 4,
3337 			    pm->pm_value >> 32, 4);
3338 	}
3339 }
3340 
3341 /*
3342  * Add a resource based on a pci map register. Return 1 if the map
3343  * register is a 32bit map register or 2 if it is a 64bit register.
3344  */
3345 static int
pci_add_map(device_t bus,device_t dev,int reg,struct resource_list * rl,int force,int prefetch)3346 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3347     int force, int prefetch)
3348 {
3349 	struct pci_map *pm;
3350 	pci_addr_t base, map, testval;
3351 	pci_addr_t start, end, count;
3352 	int barlen, basezero, flags, maprange, mapsize, type;
3353 	uint16_t cmd;
3354 	struct resource *res;
3355 
3356 	/*
3357 	 * The BAR may already exist if the device is a CardBus card
3358 	 * whose CIS is stored in this BAR.
3359 	 */
3360 	pm = pci_find_bar(dev, reg);
3361 	if (pm != NULL) {
3362 		maprange = pci_maprange(pm->pm_value);
3363 		barlen = maprange == 64 ? 2 : 1;
3364 		return (barlen);
3365 	}
3366 
3367 	pci_read_bar(dev, reg, &map, &testval, NULL);
3368 	if (PCI_BAR_MEM(map)) {
3369 		type = SYS_RES_MEMORY;
3370 		if (map & PCIM_BAR_MEM_PREFETCH)
3371 			prefetch = 1;
3372 	} else
3373 		type = SYS_RES_IOPORT;
3374 	mapsize = pci_mapsize(testval);
3375 	base = pci_mapbase(map);
3376 #ifdef __PCI_BAR_ZERO_VALID
3377 	basezero = 0;
3378 #else
3379 	basezero = base == 0;
3380 #endif
3381 	maprange = pci_maprange(map);
3382 	barlen = maprange == 64 ? 2 : 1;
3383 
3384 	/*
3385 	 * For I/O registers, if bottom bit is set, and the next bit up
3386 	 * isn't clear, we know we have a BAR that doesn't conform to the
3387 	 * spec, so ignore it.  Also, sanity check the size of the data
3388 	 * areas to the type of memory involved.  Memory must be at least
3389 	 * 16 bytes in size, while I/O ranges must be at least 4.
3390 	 */
3391 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3392 		return (barlen);
3393 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3394 	    (type == SYS_RES_IOPORT && mapsize < 2))
3395 		return (barlen);
3396 
3397 	/* Save a record of this BAR. */
3398 	pm = pci_add_bar(dev, reg, map, mapsize);
3399 	if (bootverbose) {
3400 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3401 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3402 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3403 			printf(", port disabled\n");
3404 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3405 			printf(", memory disabled\n");
3406 		else
3407 			printf(", enabled\n");
3408 	}
3409 
3410 	/*
3411 	 * If base is 0, then we have problems if this architecture does
3412 	 * not allow that.  It is best to ignore such entries for the
3413 	 * moment.  These will be allocated later if the driver specifically
3414 	 * requests them.  However, some removable buses look better when
3415 	 * all resources are allocated, so allow '0' to be overridden.
3416 	 *
3417 	 * Similarly treat maps whose values is the same as the test value
3418 	 * read back.  These maps have had all f's written to them by the
3419 	 * BIOS in an attempt to disable the resources.
3420 	 */
3421 	if (!force && (basezero || map == testval))
3422 		return (barlen);
3423 	if ((u_long)base != base) {
3424 		device_printf(bus,
3425 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3426 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3427 		    pci_get_function(dev), reg);
3428 		return (barlen);
3429 	}
3430 
3431 	/*
3432 	 * This code theoretically does the right thing, but has
3433 	 * undesirable side effects in some cases where peripherals
3434 	 * respond oddly to having these bits enabled.  Let the user
3435 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3436 	 * default).
3437 	 */
3438 	if (pci_enable_io_modes) {
3439 		/* Turn on resources that have been left off by a lazy BIOS */
3440 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3441 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3442 			cmd |= PCIM_CMD_PORTEN;
3443 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3444 		}
3445 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3446 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3447 			cmd |= PCIM_CMD_MEMEN;
3448 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3449 		}
3450 	} else {
3451 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3452 			return (barlen);
3453 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3454 			return (barlen);
3455 	}
3456 
3457 	count = (pci_addr_t)1 << mapsize;
3458 	flags = RF_ALIGNMENT_LOG2(mapsize);
3459 	if (prefetch)
3460 		flags |= RF_PREFETCHABLE;
3461 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3462 		start = 0;	/* Let the parent decide. */
3463 		end = ~0;
3464 	} else {
3465 		start = base;
3466 		end = base + count - 1;
3467 	}
3468 	resource_list_add(rl, type, reg, start, end, count);
3469 
3470 	/*
3471 	 * Try to allocate the resource for this BAR from our parent
3472 	 * so that this resource range is already reserved.  The
3473 	 * driver for this device will later inherit this resource in
3474 	 * pci_alloc_resource().
3475 	 */
3476 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3477 	    flags);
3478 	if ((pci_do_realloc_bars
3479 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3480 	    && res == NULL && (start != 0 || end != ~0)) {
3481 		/*
3482 		 * If the allocation fails, try to allocate a resource for
3483 		 * this BAR using any available range.  The firmware felt
3484 		 * it was important enough to assign a resource, so don't
3485 		 * disable decoding if we can help it.
3486 		 */
3487 		resource_list_delete(rl, type, reg);
3488 		resource_list_add(rl, type, reg, 0, ~0, count);
3489 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3490 		    count, flags);
3491 	}
3492 	if (res == NULL) {
3493 		/*
3494 		 * If the allocation fails, delete the resource list entry
3495 		 * and disable decoding for this device.
3496 		 *
3497 		 * If the driver requests this resource in the future,
3498 		 * pci_reserve_map() will try to allocate a fresh
3499 		 * resource range.
3500 		 */
3501 		resource_list_delete(rl, type, reg);
3502 		pci_disable_io(dev, type);
3503 		if (bootverbose)
3504 			device_printf(bus,
3505 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3506 			    pci_get_domain(dev), pci_get_bus(dev),
3507 			    pci_get_slot(dev), pci_get_function(dev), reg);
3508 	} else {
3509 		start = rman_get_start(res);
3510 		pci_write_bar(dev, pm, start);
3511 	}
3512 	return (barlen);
3513 }
3514 
3515 /*
3516  * For ATA devices we need to decide early what addressing mode to use.
3517  * Legacy demands that the primary and secondary ATA ports sits on the
3518  * same addresses that old ISA hardware did. This dictates that we use
3519  * those addresses and ignore the BAR's if we cannot set PCI native
3520  * addressing mode.
3521  */
3522 static void
pci_ata_maps(device_t bus,device_t dev,struct resource_list * rl,int force,uint32_t prefetchmask)3523 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3524     uint32_t prefetchmask)
3525 {
3526 	int rid, type, progif;
3527 #if 0
3528 	/* if this device supports PCI native addressing use it */
3529 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3530 	if ((progif & 0x8a) == 0x8a) {
3531 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3532 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3533 			printf("Trying ATA native PCI addressing mode\n");
3534 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3535 		}
3536 	}
3537 #endif
3538 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3539 	type = SYS_RES_IOPORT;
3540 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3541 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3542 		    prefetchmask & (1 << 0));
3543 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3544 		    prefetchmask & (1 << 1));
3545 	} else {
3546 		rid = PCIR_BAR(0);
3547 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3548 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3549 		    0x1f7, 8, 0);
3550 		rid = PCIR_BAR(1);
3551 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3552 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3553 		    0x3f6, 1, 0);
3554 	}
3555 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3556 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3557 		    prefetchmask & (1 << 2));
3558 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3559 		    prefetchmask & (1 << 3));
3560 	} else {
3561 		rid = PCIR_BAR(2);
3562 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3563 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3564 		    0x177, 8, 0);
3565 		rid = PCIR_BAR(3);
3566 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3567 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3568 		    0x376, 1, 0);
3569 	}
3570 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3571 	    prefetchmask & (1 << 4));
3572 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3573 	    prefetchmask & (1 << 5));
3574 }
3575 
3576 static void
pci_assign_interrupt(device_t bus,device_t dev,int force_route)3577 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3578 {
3579 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3580 	pcicfgregs *cfg = &dinfo->cfg;
3581 	char tunable_name[64];
3582 	int irq;
3583 
3584 	/* Has to have an intpin to have an interrupt. */
3585 	if (cfg->intpin == 0)
3586 		return;
3587 
3588 	/* Let the user override the IRQ with a tunable. */
3589 	irq = PCI_INVALID_IRQ;
3590 	snprintf(tunable_name, sizeof(tunable_name),
3591 	    "hw.pci%d.%d.%d.INT%c.irq",
3592 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3593 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3594 		irq = PCI_INVALID_IRQ;
3595 
3596 	/*
3597 	 * If we didn't get an IRQ via the tunable, then we either use the
3598 	 * IRQ value in the intline register or we ask the bus to route an
3599 	 * interrupt for us.  If force_route is true, then we only use the
3600 	 * value in the intline register if the bus was unable to assign an
3601 	 * IRQ.
3602 	 */
3603 	if (!PCI_INTERRUPT_VALID(irq)) {
3604 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3605 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3606 		if (!PCI_INTERRUPT_VALID(irq))
3607 			irq = cfg->intline;
3608 	}
3609 
3610 	/* If after all that we don't have an IRQ, just bail. */
3611 	if (!PCI_INTERRUPT_VALID(irq))
3612 		return;
3613 
3614 	/* Update the config register if it changed. */
3615 	if (irq != cfg->intline) {
3616 		cfg->intline = irq;
3617 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3618 	}
3619 
3620 	/* Add this IRQ as rid 0 interrupt resource. */
3621 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3622 }
3623 
3624 /* Perform early OHCI takeover from SMM. */
3625 static void
ohci_early_takeover(device_t self)3626 ohci_early_takeover(device_t self)
3627 {
3628 	struct resource *res;
3629 	uint32_t ctl;
3630 	int rid;
3631 	int i;
3632 
3633 	rid = PCIR_BAR(0);
3634 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3635 	if (res == NULL)
3636 		return;
3637 
3638 	ctl = bus_read_4(res, OHCI_CONTROL);
3639 	if (ctl & OHCI_IR) {
3640 		if (bootverbose)
3641 			printf("ohci early: "
3642 			    "SMM active, request owner change\n");
3643 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3644 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3645 			DELAY(1000);
3646 			ctl = bus_read_4(res, OHCI_CONTROL);
3647 		}
3648 		if (ctl & OHCI_IR) {
3649 			if (bootverbose)
3650 				printf("ohci early: "
3651 				    "SMM does not respond, resetting\n");
3652 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3653 		}
3654 		/* Disable interrupts */
3655 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3656 	}
3657 
3658 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3659 }
3660 
3661 /* Perform early UHCI takeover from SMM. */
3662 static void
uhci_early_takeover(device_t self)3663 uhci_early_takeover(device_t self)
3664 {
3665 	struct resource *res;
3666 	int rid;
3667 
3668 	/*
3669 	 * Set the PIRQD enable bit and switch off all the others. We don't
3670 	 * want legacy support to interfere with us XXX Does this also mean
3671 	 * that the BIOS won't touch the keyboard anymore if it is connected
3672 	 * to the ports of the root hub?
3673 	 */
3674 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3675 
3676 	/* Disable interrupts */
3677 	rid = PCI_UHCI_BASE_REG;
3678 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3679 	if (res != NULL) {
3680 		bus_write_2(res, UHCI_INTR, 0);
3681 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3682 	}
3683 }
3684 
3685 /* Perform early EHCI takeover from SMM. */
3686 static void
ehci_early_takeover(device_t self)3687 ehci_early_takeover(device_t self)
3688 {
3689 	struct resource *res;
3690 	uint32_t cparams;
3691 	uint32_t eec;
3692 	uint8_t eecp;
3693 	uint8_t bios_sem;
3694 	uint8_t offs;
3695 	int rid;
3696 	int i;
3697 
3698 	rid = PCIR_BAR(0);
3699 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3700 	if (res == NULL)
3701 		return;
3702 
3703 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3704 
3705 	/* Synchronise with the BIOS if it owns the controller. */
3706 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3707 	    eecp = EHCI_EECP_NEXT(eec)) {
3708 		eec = pci_read_config(self, eecp, 4);
3709 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3710 			continue;
3711 		}
3712 		bios_sem = pci_read_config(self, eecp +
3713 		    EHCI_LEGSUP_BIOS_SEM, 1);
3714 		if (bios_sem == 0) {
3715 			continue;
3716 		}
3717 		if (bootverbose)
3718 			printf("ehci early: "
3719 			    "SMM active, request owner change\n");
3720 
3721 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3722 
3723 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3724 			DELAY(1000);
3725 			bios_sem = pci_read_config(self, eecp +
3726 			    EHCI_LEGSUP_BIOS_SEM, 1);
3727 		}
3728 
3729 		if (bios_sem != 0) {
3730 			if (bootverbose)
3731 				printf("ehci early: "
3732 				    "SMM does not respond\n");
3733 		}
3734 		/* Disable interrupts */
3735 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3736 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3737 	}
3738 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3739 }
3740 
3741 /* Perform early XHCI takeover from SMM. */
3742 static void
xhci_early_takeover(device_t self)3743 xhci_early_takeover(device_t self)
3744 {
3745 	struct resource *res;
3746 	uint32_t cparams;
3747 	uint32_t eec;
3748 	uint8_t eecp;
3749 	uint8_t bios_sem;
3750 	uint8_t offs;
3751 	int rid;
3752 	int i;
3753 
3754 	rid = PCIR_BAR(0);
3755 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3756 	if (res == NULL)
3757 		return;
3758 
3759 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3760 
3761 	eec = -1;
3762 
3763 	/* Synchronise with the BIOS if it owns the controller. */
3764 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3765 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3766 		eec = bus_read_4(res, eecp);
3767 
3768 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3769 			continue;
3770 
3771 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3772 		if (bios_sem == 0)
3773 			continue;
3774 
3775 		if (bootverbose)
3776 			printf("xhci early: "
3777 			    "SMM active, request owner change\n");
3778 
3779 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3780 
3781 		/* wait a maximum of 5 second */
3782 
3783 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3784 			DELAY(1000);
3785 			bios_sem = bus_read_1(res, eecp +
3786 			    XHCI_XECP_BIOS_SEM);
3787 		}
3788 
3789 		if (bios_sem != 0) {
3790 			if (bootverbose)
3791 				printf("xhci early: "
3792 				    "SMM does not respond\n");
3793 		}
3794 
3795 		/* Disable interrupts */
3796 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3797 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3798 		bus_read_4(res, offs + XHCI_USBSTS);
3799 	}
3800 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3801 }
3802 
3803 static void
pci_reserve_secbus(device_t bus,device_t dev,pcicfgregs * cfg,struct resource_list * rl)3804 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3805     struct resource_list *rl)
3806 {
3807 	struct resource *res;
3808 	char *cp;
3809 	rman_res_t start, end, count;
3810 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3811 
3812 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3813 	case PCIM_HDRTYPE_BRIDGE:
3814 		sec_reg = PCIR_SECBUS_1;
3815 		sub_reg = PCIR_SUBBUS_1;
3816 		break;
3817 	case PCIM_HDRTYPE_CARDBUS:
3818 		sec_reg = PCIR_SECBUS_2;
3819 		sub_reg = PCIR_SUBBUS_2;
3820 		break;
3821 	default:
3822 		return;
3823 	}
3824 
3825 	/*
3826 	 * If the existing bus range is valid, attempt to reserve it
3827 	 * from our parent.  If this fails for any reason, clear the
3828 	 * secbus and subbus registers.
3829 	 *
3830 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3831 	 * This would at least preserve the existing sec_bus if it is
3832 	 * valid.
3833 	 */
3834 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3835 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3836 
3837 	/* Quirk handling. */
3838 	switch (pci_get_devid(dev)) {
3839 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3840 		sup_bus = pci_read_config(dev, 0x41, 1);
3841 		if (sup_bus != 0xff) {
3842 			sec_bus = sup_bus + 1;
3843 			sub_bus = sup_bus + 1;
3844 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3845 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3846 		}
3847 		break;
3848 
3849 	case 0x00dd10de:
3850 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3851 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3852 			break;
3853 		if (strncmp(cp, "Compal", 6) != 0) {
3854 			freeenv(cp);
3855 			break;
3856 		}
3857 		freeenv(cp);
3858 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3859 			break;
3860 		if (strncmp(cp, "08A0", 4) != 0) {
3861 			freeenv(cp);
3862 			break;
3863 		}
3864 		freeenv(cp);
3865 		if (sub_bus < 0xa) {
3866 			sub_bus = 0xa;
3867 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3868 		}
3869 		break;
3870 	}
3871 
3872 	if (bootverbose)
3873 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3874 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3875 		start = sec_bus;
3876 		end = sub_bus;
3877 		count = end - start + 1;
3878 
3879 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3880 
3881 		/*
3882 		 * If requested, clear secondary bus registers in
3883 		 * bridge devices to force a complete renumbering
3884 		 * rather than reserving the existing range.  However,
3885 		 * preserve the existing size.
3886 		 */
3887 		if (pci_clear_buses)
3888 			goto clear;
3889 
3890 		rid = 0;
3891 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3892 		    start, end, count, 0);
3893 		if (res != NULL)
3894 			return;
3895 
3896 		if (bootverbose)
3897 			device_printf(bus,
3898 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3899 			    pci_get_domain(dev), pci_get_bus(dev),
3900 			    pci_get_slot(dev), pci_get_function(dev));
3901 	}
3902 
3903 clear:
3904 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3905 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3906 }
3907 
3908 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)3909 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3910     rman_res_t end, rman_res_t count, u_int flags)
3911 {
3912 	struct pci_devinfo *dinfo;
3913 	pcicfgregs *cfg;
3914 	struct resource_list *rl;
3915 	struct resource *res;
3916 	int sec_reg, sub_reg;
3917 
3918 	dinfo = device_get_ivars(child);
3919 	cfg = &dinfo->cfg;
3920 	rl = &dinfo->resources;
3921 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3922 	case PCIM_HDRTYPE_BRIDGE:
3923 		sec_reg = PCIR_SECBUS_1;
3924 		sub_reg = PCIR_SUBBUS_1;
3925 		break;
3926 	case PCIM_HDRTYPE_CARDBUS:
3927 		sec_reg = PCIR_SECBUS_2;
3928 		sub_reg = PCIR_SUBBUS_2;
3929 		break;
3930 	default:
3931 		return (NULL);
3932 	}
3933 
3934 	if (*rid != 0)
3935 		return (NULL);
3936 
3937 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3938 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3939 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3940 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3941 		    start, end, count, flags & ~RF_ACTIVE);
3942 		if (res == NULL) {
3943 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3944 			device_printf(child, "allocating %ju bus%s failed\n",
3945 			    count, count == 1 ? "" : "es");
3946 			return (NULL);
3947 		}
3948 		if (bootverbose)
3949 			device_printf(child,
3950 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3951 			    count == 1 ? "" : "es", rman_get_start(res));
3952 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3953 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3954 	}
3955 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3956 	    end, count, flags));
3957 }
3958 
3959 static int
pci_ea_bei_to_rid(device_t dev,int bei)3960 pci_ea_bei_to_rid(device_t dev, int bei)
3961 {
3962 #ifdef PCI_IOV
3963 	struct pci_devinfo *dinfo;
3964 	int iov_pos;
3965 	struct pcicfg_iov *iov;
3966 
3967 	dinfo = device_get_ivars(dev);
3968 	iov = dinfo->cfg.iov;
3969 	if (iov != NULL)
3970 		iov_pos = iov->iov_pos;
3971 	else
3972 		iov_pos = 0;
3973 #endif
3974 
3975 	/* Check if matches BAR */
3976 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3977 	    (bei <= PCIM_EA_BEI_BAR_5))
3978 		return (PCIR_BAR(bei));
3979 
3980 	/* Check ROM */
3981 	if (bei == PCIM_EA_BEI_ROM)
3982 		return (PCIR_BIOS);
3983 
3984 #ifdef PCI_IOV
3985 	/* Check if matches VF_BAR */
3986 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3987 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3988 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3989 		    iov_pos);
3990 #endif
3991 
3992 	return (-1);
3993 }
3994 
3995 int
pci_ea_is_enabled(device_t dev,int rid)3996 pci_ea_is_enabled(device_t dev, int rid)
3997 {
3998 	struct pci_ea_entry *ea;
3999 	struct pci_devinfo *dinfo;
4000 
4001 	dinfo = device_get_ivars(dev);
4002 
4003 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4004 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
4005 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
4006 	}
4007 
4008 	return (0);
4009 }
4010 
4011 void
pci_add_resources_ea(device_t bus,device_t dev,int alloc_iov)4012 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
4013 {
4014 	struct pci_ea_entry *ea;
4015 	struct pci_devinfo *dinfo;
4016 	pci_addr_t start, end, count;
4017 	struct resource_list *rl;
4018 	int type, flags, rid;
4019 	struct resource *res;
4020 	uint32_t tmp;
4021 #ifdef PCI_IOV
4022 	struct pcicfg_iov *iov;
4023 #endif
4024 
4025 	dinfo = device_get_ivars(dev);
4026 	rl = &dinfo->resources;
4027 	flags = 0;
4028 
4029 #ifdef PCI_IOV
4030 	iov = dinfo->cfg.iov;
4031 #endif
4032 
4033 	if (dinfo->cfg.ea.ea_location == 0)
4034 		return;
4035 
4036 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
4037 		/*
4038 		 * TODO: Ignore EA-BAR if is not enabled.
4039 		 *   Currently the EA implementation supports
4040 		 *   only situation, where EA structure contains
4041 		 *   predefined entries. In case they are not enabled
4042 		 *   leave them unallocated and proceed with
4043 		 *   a legacy-BAR mechanism.
4044 		 */
4045 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
4046 			continue;
4047 
4048 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
4049 		case PCIM_EA_P_MEM_PREFETCH:
4050 		case PCIM_EA_P_VF_MEM_PREFETCH:
4051 			flags = RF_PREFETCHABLE;
4052 			/* FALLTHROUGH */
4053 		case PCIM_EA_P_VF_MEM:
4054 		case PCIM_EA_P_MEM:
4055 			type = SYS_RES_MEMORY;
4056 			break;
4057 		case PCIM_EA_P_IO:
4058 			type = SYS_RES_IOPORT;
4059 			break;
4060 		default:
4061 			continue;
4062 		}
4063 
4064 		if (alloc_iov != 0) {
4065 #ifdef PCI_IOV
4066 			/* Allocating IOV, confirm BEI matches */
4067 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
4068 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
4069 				continue;
4070 #else
4071 			continue;
4072 #endif
4073 		} else {
4074 			/* Allocating BAR, confirm BEI matches */
4075 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
4076 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
4077 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
4078 				continue;
4079 		}
4080 
4081 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
4082 		if (rid < 0)
4083 			continue;
4084 
4085 		/* Skip resources already allocated by EA */
4086 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
4087 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
4088 			continue;
4089 
4090 		start = ea->eae_base;
4091 		count = ea->eae_max_offset + 1;
4092 #ifdef PCI_IOV
4093 		if (iov != NULL)
4094 			count = count * iov->iov_num_vfs;
4095 #endif
4096 		end = start + count - 1;
4097 		if (count == 0)
4098 			continue;
4099 
4100 		resource_list_add(rl, type, rid, start, end, count);
4101 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4102 		    flags);
4103 		if (res == NULL) {
4104 			resource_list_delete(rl, type, rid);
4105 
4106 			/*
4107 			 * Failed to allocate using EA, disable entry.
4108 			 * Another attempt to allocation will be performed
4109 			 * further, but this time using legacy BAR registers
4110 			 */
4111 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4112 			tmp &= ~PCIM_EA_ENABLE;
4113 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4114 
4115 			/*
4116 			 * Disabling entry might fail in case it is hardwired.
4117 			 * Read flags again to match current status.
4118 			 */
4119 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4120 
4121 			continue;
4122 		}
4123 
4124 		/* As per specification, fill BAR with zeros */
4125 		pci_write_config(dev, rid, 0, 4);
4126 	}
4127 }
4128 
4129 void
pci_add_resources(device_t bus,device_t dev,int force,uint32_t prefetchmask)4130 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4131 {
4132 	struct pci_devinfo *dinfo;
4133 	pcicfgregs *cfg;
4134 	struct resource_list *rl;
4135 	const struct pci_quirk *q;
4136 	uint32_t devid;
4137 	int i;
4138 
4139 	dinfo = device_get_ivars(dev);
4140 	cfg = &dinfo->cfg;
4141 	rl = &dinfo->resources;
4142 	devid = (cfg->device << 16) | cfg->vendor;
4143 
4144 	/* Allocate resources using Enhanced Allocation */
4145 	pci_add_resources_ea(bus, dev, 0);
4146 
4147 	/* ATA devices needs special map treatment */
4148 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4149 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4150 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4151 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4152 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4153 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4154 	else
4155 		for (i = 0; i < cfg->nummaps;) {
4156 			/* Skip resources already managed by EA */
4157 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4158 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4159 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4160 				i++;
4161 				continue;
4162 			}
4163 
4164 			/*
4165 			 * Skip quirked resources.
4166 			 */
4167 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4168 				if (q->devid == devid &&
4169 				    q->type == PCI_QUIRK_UNMAP_REG &&
4170 				    q->arg1 == PCIR_BAR(i))
4171 					break;
4172 			if (q->devid != 0) {
4173 				i++;
4174 				continue;
4175 			}
4176 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4177 			    prefetchmask & (1 << i));
4178 		}
4179 
4180 	/*
4181 	 * Add additional, quirked resources.
4182 	 */
4183 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4184 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4185 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4186 
4187 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline) &&
4188 	    pci_intx_reroute) {
4189 		/*
4190 		 * Try to re-route interrupts. Sometimes the BIOS or
4191 		 * firmware may leave bogus values in these registers.
4192 		 * If the re-route fails, then just stick with what we
4193 		 * have.
4194 		 */
4195 		pci_assign_interrupt(bus, dev, 1);
4196 	}
4197 
4198 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4199 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4200 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4201 			xhci_early_takeover(dev);
4202 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4203 			ehci_early_takeover(dev);
4204 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4205 			ohci_early_takeover(dev);
4206 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4207 			uhci_early_takeover(dev);
4208 	}
4209 
4210 	/*
4211 	 * Reserve resources for secondary bus ranges behind bridge
4212 	 * devices.
4213 	 */
4214 	pci_reserve_secbus(bus, dev, cfg, rl);
4215 }
4216 
4217 static struct pci_devinfo *
pci_identify_function(device_t pcib,device_t dev,int domain,int busno,int slot,int func)4218 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4219     int slot, int func)
4220 {
4221 	struct pci_devinfo *dinfo;
4222 
4223 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4224 	if (dinfo != NULL)
4225 		pci_add_child(dev, dinfo);
4226 
4227 	return (dinfo);
4228 }
4229 
4230 void
pci_add_children(device_t dev,int domain,int busno)4231 pci_add_children(device_t dev, int domain, int busno)
4232 {
4233 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4234 	device_t pcib = device_get_parent(dev);
4235 	struct pci_devinfo *dinfo;
4236 	int maxslots;
4237 	int s, f, pcifunchigh;
4238 	uint8_t hdrtype;
4239 	int first_func;
4240 
4241 	/*
4242 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4243 	 * enable ARI.  We must enable ARI before detecting the rest of the
4244 	 * functions on this bus as ARI changes the set of slots and functions
4245 	 * that are legal on this bus.
4246 	 */
4247 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4248 	if (dinfo != NULL && pci_enable_ari)
4249 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4250 
4251 	/*
4252 	 * Start looking for new devices on slot 0 at function 1 because we
4253 	 * just identified the device at slot 0, function 0.
4254 	 */
4255 	first_func = 1;
4256 
4257 	maxslots = PCIB_MAXSLOTS(pcib);
4258 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4259 		pcifunchigh = 0;
4260 		f = 0;
4261 		DELAY(1);
4262 
4263 		/* If function 0 is not present, skip to the next slot. */
4264 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4265 			continue;
4266 		hdrtype = REG(PCIR_HDRTYPE, 1);
4267 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4268 			continue;
4269 		if (hdrtype & PCIM_MFDEV)
4270 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4271 		for (f = first_func; f <= pcifunchigh; f++)
4272 			pci_identify_function(pcib, dev, domain, busno, s, f);
4273 	}
4274 #undef REG
4275 }
4276 
4277 int
pci_rescan_method(device_t dev)4278 pci_rescan_method(device_t dev)
4279 {
4280 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4281 	device_t pcib = device_get_parent(dev);
4282 	device_t child, *devlist, *unchanged;
4283 	int devcount, error, i, j, maxslots, oldcount;
4284 	int busno, domain, s, f, pcifunchigh;
4285 	uint8_t hdrtype;
4286 
4287 	/* No need to check for ARI on a rescan. */
4288 	error = device_get_children(dev, &devlist, &devcount);
4289 	if (error)
4290 		return (error);
4291 	if (devcount != 0) {
4292 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4293 		    M_NOWAIT | M_ZERO);
4294 		if (unchanged == NULL) {
4295 			free(devlist, M_TEMP);
4296 			return (ENOMEM);
4297 		}
4298 	} else
4299 		unchanged = NULL;
4300 
4301 	domain = pcib_get_domain(dev);
4302 	busno = pcib_get_bus(dev);
4303 	maxslots = PCIB_MAXSLOTS(pcib);
4304 	for (s = 0; s <= maxslots; s++) {
4305 		/* If function 0 is not present, skip to the next slot. */
4306 		f = 0;
4307 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4308 			continue;
4309 		pcifunchigh = 0;
4310 		hdrtype = REG(PCIR_HDRTYPE, 1);
4311 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4312 			continue;
4313 		if (hdrtype & PCIM_MFDEV)
4314 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4315 		for (f = 0; f <= pcifunchigh; f++) {
4316 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4317 				continue;
4318 
4319 			/*
4320 			 * Found a valid function.  Check if a
4321 			 * device_t for this device already exists.
4322 			 */
4323 			for (i = 0; i < devcount; i++) {
4324 				child = devlist[i];
4325 				if (child == NULL)
4326 					continue;
4327 				if (pci_get_slot(child) == s &&
4328 				    pci_get_function(child) == f) {
4329 					unchanged[i] = child;
4330 					goto next_func;
4331 				}
4332 			}
4333 
4334 			pci_identify_function(pcib, dev, domain, busno, s, f);
4335 		next_func:;
4336 		}
4337 	}
4338 
4339 	/* Remove devices that are no longer present. */
4340 	for (i = 0; i < devcount; i++) {
4341 		if (unchanged[i] != NULL)
4342 			continue;
4343 		device_delete_child(dev, devlist[i]);
4344 	}
4345 
4346 	free(devlist, M_TEMP);
4347 	oldcount = devcount;
4348 
4349 	/* Try to attach the devices just added. */
4350 	error = device_get_children(dev, &devlist, &devcount);
4351 	if (error) {
4352 		free(unchanged, M_TEMP);
4353 		return (error);
4354 	}
4355 
4356 	for (i = 0; i < devcount; i++) {
4357 		for (j = 0; j < oldcount; j++) {
4358 			if (devlist[i] == unchanged[j])
4359 				goto next_device;
4360 		}
4361 
4362 		device_probe_and_attach(devlist[i]);
4363 	next_device:;
4364 	}
4365 
4366 	free(unchanged, M_TEMP);
4367 	free(devlist, M_TEMP);
4368 	return (0);
4369 #undef REG
4370 }
4371 
4372 #ifdef PCI_IOV
4373 device_t
pci_add_iov_child(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4374 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4375     uint16_t did)
4376 {
4377 	struct pci_devinfo *vf_dinfo;
4378 	device_t pcib;
4379 	int busno, slot, func;
4380 
4381 	pcib = device_get_parent(bus);
4382 
4383 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4384 
4385 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4386 	    slot, func, vid, did);
4387 
4388 	vf_dinfo->cfg.flags |= PCICFG_VF;
4389 	pci_add_child(bus, vf_dinfo);
4390 
4391 	return (vf_dinfo->cfg.dev);
4392 }
4393 
4394 device_t
pci_create_iov_child_method(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4395 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4396     uint16_t vid, uint16_t did)
4397 {
4398 
4399 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4400 }
4401 #endif
4402 
4403 /*
4404  * For PCIe device set Max_Payload_Size to match PCIe root's.
4405  */
4406 static void
pcie_setup_mps(device_t dev)4407 pcie_setup_mps(device_t dev)
4408 {
4409 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4410 	device_t root;
4411 	uint16_t rmps, mmps, mps;
4412 
4413 	if (dinfo->cfg.pcie.pcie_location == 0)
4414 		return;
4415 	root = pci_find_pcie_root_port(dev);
4416 	if (root == NULL)
4417 		return;
4418 	/* Check whether the MPS is already configured. */
4419 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4420 	    PCIEM_CTL_MAX_PAYLOAD;
4421 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4422 	    PCIEM_CTL_MAX_PAYLOAD;
4423 	if (mps == rmps)
4424 		return;
4425 	/* Check whether the device is capable of the root's MPS. */
4426 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4427 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4428 	if (rmps > mmps) {
4429 		/*
4430 		 * The device is unable to handle root's MPS.  Limit root.
4431 		 * XXX: We should traverse through all the tree, applying
4432 		 * it to all the devices.
4433 		 */
4434 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4435 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4436 	} else {
4437 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4438 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4439 	}
4440 }
4441 
4442 static void
pci_add_child_clear_aer(device_t dev,struct pci_devinfo * dinfo)4443 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4444 {
4445 	int aer;
4446 	uint32_t r;
4447 	uint16_t r2;
4448 
4449 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4450 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4451 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4452 		    PCIER_ROOT_CTL, 2);
4453 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4454 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4455 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4456 		    PCIER_ROOT_CTL, r2, 2);
4457 	}
4458 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4459 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4460 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4461 		if (r != 0 && bootverbose) {
4462 			pci_printf(&dinfo->cfg,
4463 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4464 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4465 			    4));
4466 		}
4467 
4468 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4469 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4470 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4471 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4472 		    PCIM_AER_UC_POISONED_TLP |
4473 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4474 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4475 		    PCIM_AER_UC_COMPLETER_ABORT |
4476 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4477 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4478 		    PCIM_AER_UC_MALFORMED_TLP |
4479 		    PCIM_AER_UC_ECRC_ERROR |
4480 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4481 		    PCIM_AER_UC_ACS_VIOLATION |
4482 		    PCIM_AER_UC_INTERNAL_ERROR |
4483 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4484 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4485 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4486 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4487 
4488 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4489 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4490 		if (r != 0 && bootverbose) {
4491 			pci_printf(&dinfo->cfg,
4492 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4493 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4494 			    4));
4495 		}
4496 
4497 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4498 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4499 		    PCIM_AER_COR_BAD_TLP |
4500 		    PCIM_AER_COR_BAD_DLLP |
4501 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4502 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4503 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4504 		    PCIM_AER_COR_INTERNAL_ERROR |
4505 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4506 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4507 
4508 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4509 		    PCIER_DEVICE_CTL, 2);
4510 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4511 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4512 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4513 		    PCIER_DEVICE_CTL, r, 2);
4514 	}
4515 }
4516 
4517 void
pci_add_child(device_t bus,struct pci_devinfo * dinfo)4518 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4519 {
4520 	device_t dev;
4521 
4522 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4523 	device_set_ivars(dev, dinfo);
4524 	resource_list_init(&dinfo->resources);
4525 	pci_cfg_save(dev, dinfo, 0);
4526 	pci_cfg_restore(dev, dinfo);
4527 	pci_clear_pme(dev);
4528 	pci_print_verbose(dinfo);
4529 	pci_add_resources(bus, dev, 0, 0);
4530 	if (pci_enable_mps_tune)
4531 		pcie_setup_mps(dev);
4532 	pci_child_added(dinfo->cfg.dev);
4533 
4534 	if (pci_clear_aer_on_attach)
4535 		pci_add_child_clear_aer(dev, dinfo);
4536 
4537 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4538 }
4539 
4540 void
pci_child_added_method(device_t dev,device_t child)4541 pci_child_added_method(device_t dev, device_t child)
4542 {
4543 
4544 }
4545 
4546 static int
pci_probe(device_t dev)4547 pci_probe(device_t dev)
4548 {
4549 
4550 	device_set_desc(dev, "PCI bus");
4551 
4552 	/* Allow other subclasses to override this driver. */
4553 	return (BUS_PROBE_GENERIC);
4554 }
4555 
4556 int
pci_attach_common(device_t dev)4557 pci_attach_common(device_t dev)
4558 {
4559 	struct pci_softc *sc;
4560 	int busno, domain;
4561 	int rid;
4562 
4563 	sc = device_get_softc(dev);
4564 	domain = pcib_get_domain(dev);
4565 	busno = pcib_get_bus(dev);
4566 	rid = 0;
4567 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4568 	    1, 0);
4569 	if (sc->sc_bus == NULL) {
4570 		device_printf(dev, "failed to allocate bus number\n");
4571 		return (ENXIO);
4572 	}
4573 	if (bootverbose)
4574 		device_printf(dev, "domain=%d, physical bus=%d\n",
4575 		    domain, busno);
4576 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4577 	return (0);
4578 }
4579 
4580 int
pci_attach(device_t dev)4581 pci_attach(device_t dev)
4582 {
4583 	int busno, domain, error;
4584 
4585 	error = pci_attach_common(dev);
4586 	if (error)
4587 		return (error);
4588 
4589 	/*
4590 	 * Since there can be multiple independently numbered PCI
4591 	 * buses on systems with multiple PCI domains, we can't use
4592 	 * the unit number to decide which bus we are probing. We ask
4593 	 * the parent pcib what our domain and bus numbers are.
4594 	 */
4595 	domain = pcib_get_domain(dev);
4596 	busno = pcib_get_bus(dev);
4597 	pci_add_children(dev, domain, busno);
4598 	bus_attach_children(dev);
4599 	return (0);
4600 }
4601 
4602 int
pci_detach(device_t dev)4603 pci_detach(device_t dev)
4604 {
4605 	struct pci_softc *sc;
4606 	int error;
4607 
4608 	error = bus_generic_detach(dev);
4609 	if (error)
4610 		return (error);
4611 	sc = device_get_softc(dev);
4612 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4613 	return (error);
4614 }
4615 
4616 static void
pci_hint_device_unit(device_t dev,device_t child,const char * name,int * unitp)4617 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4618 {
4619 	int line, unit;
4620 	const char *at;
4621 	char me1[24], me2[32];
4622 	uint8_t b, s, f;
4623 	uint32_t d;
4624 	device_location_cache_t *cache;
4625 
4626 	d = pci_get_domain(child);
4627 	b = pci_get_bus(child);
4628 	s = pci_get_slot(child);
4629 	f = pci_get_function(child);
4630 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4631 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4632 	line = 0;
4633 	cache = dev_wired_cache_init();
4634 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4635 		resource_string_value(name, unit, "at", &at);
4636 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4637 			*unitp = unit;
4638 			break;
4639 		}
4640 		if (dev_wired_cache_match(cache, child, at)) {
4641 			*unitp = unit;
4642 			break;
4643 		}
4644 	}
4645 	dev_wired_cache_fini(cache);
4646 }
4647 
4648 static void
pci_set_power_child(device_t dev,device_t child,int state)4649 pci_set_power_child(device_t dev, device_t child, int state)
4650 {
4651 	device_t pcib;
4652 	int dstate;
4653 
4654 	/*
4655 	 * Set the device to the given state.  If the firmware suggests
4656 	 * a different power state, use it instead.  If power management
4657 	 * is not present, the firmware is responsible for managing
4658 	 * device power.  Skip children who aren't attached since they
4659 	 * are handled separately.
4660 	 */
4661 	pcib = device_get_parent(dev);
4662 	dstate = state;
4663 	if (device_is_attached(child) &&
4664 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4665 		pci_set_powerstate(child, dstate);
4666 }
4667 
4668 int
pci_suspend_child(device_t dev,device_t child)4669 pci_suspend_child(device_t dev, device_t child)
4670 {
4671 	struct pci_devinfo *dinfo;
4672 	struct resource_list_entry *rle;
4673 	int error;
4674 
4675 	dinfo = device_get_ivars(child);
4676 
4677 	/*
4678 	 * Save the PCI configuration space for the child and set the
4679 	 * device in the appropriate power state for this sleep state.
4680 	 */
4681 	pci_cfg_save(child, dinfo, 0);
4682 
4683 	/* Suspend devices before potentially powering them down. */
4684 	error = bus_generic_suspend_child(dev, child);
4685 
4686 	if (error)
4687 		return (error);
4688 
4689 	if (pci_do_power_suspend) {
4690 		/*
4691 		 * Make sure this device's interrupt handler is not invoked
4692 		 * in the case the device uses a shared interrupt that can
4693 		 * be raised by some other device.
4694 		 * This is applicable only to regular (legacy) PCI interrupts
4695 		 * as MSI/MSI-X interrupts are never shared.
4696 		 */
4697 		rle = resource_list_find(&dinfo->resources,
4698 		    SYS_RES_IRQ, 0);
4699 		if (rle != NULL && rle->res != NULL)
4700 			(void)bus_suspend_intr(child, rle->res);
4701 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4702 	}
4703 
4704 	return (0);
4705 }
4706 
4707 int
pci_resume_child(device_t dev,device_t child)4708 pci_resume_child(device_t dev, device_t child)
4709 {
4710 	struct pci_devinfo *dinfo;
4711 	struct resource_list_entry *rle;
4712 
4713 	if (pci_do_power_resume)
4714 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4715 
4716 	dinfo = device_get_ivars(child);
4717 	pci_cfg_restore(child, dinfo);
4718 	pci_clear_pme(child);
4719 	if (!device_is_attached(child))
4720 		pci_cfg_save(child, dinfo, 1);
4721 
4722 	bus_generic_resume_child(dev, child);
4723 
4724 	/*
4725 	 * Allow interrupts only after fully resuming the driver and hardware.
4726 	 */
4727 	if (pci_do_power_suspend) {
4728 		/* See pci_suspend_child for details. */
4729 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4730 		if (rle != NULL && rle->res != NULL)
4731 			(void)bus_resume_intr(child, rle->res);
4732 	}
4733 
4734 	return (0);
4735 }
4736 
4737 int
pci_resume(device_t dev)4738 pci_resume(device_t dev)
4739 {
4740 	device_t child, *devlist;
4741 	int error, i, numdevs;
4742 
4743 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4744 		return (error);
4745 
4746 	/*
4747 	 * Resume critical devices first, then everything else later.
4748 	 */
4749 	for (i = 0; i < numdevs; i++) {
4750 		child = devlist[i];
4751 		switch (pci_get_class(child)) {
4752 		case PCIC_DISPLAY:
4753 		case PCIC_MEMORY:
4754 		case PCIC_BRIDGE:
4755 		case PCIC_BASEPERIPH:
4756 			BUS_RESUME_CHILD(dev, child);
4757 			break;
4758 		}
4759 	}
4760 	for (i = 0; i < numdevs; i++) {
4761 		child = devlist[i];
4762 		switch (pci_get_class(child)) {
4763 		case PCIC_DISPLAY:
4764 		case PCIC_MEMORY:
4765 		case PCIC_BRIDGE:
4766 		case PCIC_BASEPERIPH:
4767 			break;
4768 		default:
4769 			BUS_RESUME_CHILD(dev, child);
4770 		}
4771 	}
4772 	free(devlist, M_TEMP);
4773 	return (0);
4774 }
4775 
4776 static void
pci_load_vendor_data(void)4777 pci_load_vendor_data(void)
4778 {
4779 	caddr_t data;
4780 	void *ptr;
4781 	size_t sz;
4782 
4783 	data = preload_search_by_type("pci_vendor_data");
4784 	if (data != NULL) {
4785 		ptr = preload_fetch_addr(data);
4786 		sz = preload_fetch_size(data);
4787 		if (ptr != NULL && sz != 0) {
4788 			pci_vendordata = ptr;
4789 			pci_vendordata_size = sz;
4790 			/* terminate the database */
4791 			pci_vendordata[pci_vendordata_size] = '\n';
4792 		}
4793 	}
4794 }
4795 
4796 void
pci_driver_added(device_t dev,driver_t * driver)4797 pci_driver_added(device_t dev, driver_t *driver)
4798 {
4799 	int numdevs;
4800 	device_t *devlist;
4801 	device_t child;
4802 	struct pci_devinfo *dinfo;
4803 	int i;
4804 
4805 	if (bootverbose)
4806 		device_printf(dev, "driver added\n");
4807 	DEVICE_IDENTIFY(driver, dev);
4808 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4809 		return;
4810 	for (i = 0; i < numdevs; i++) {
4811 		child = devlist[i];
4812 		if (device_get_state(child) != DS_NOTPRESENT)
4813 			continue;
4814 		dinfo = device_get_ivars(child);
4815 		pci_print_verbose(dinfo);
4816 		if (bootverbose)
4817 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4818 		pci_cfg_restore(child, dinfo);
4819 		if (device_probe_and_attach(child) != 0)
4820 			pci_child_detached(dev, child);
4821 	}
4822 	free(devlist, M_TEMP);
4823 }
4824 
4825 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)4826 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4827     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4828 {
4829 	struct pci_devinfo *dinfo;
4830 	struct msix_table_entry *mte;
4831 	struct msix_vector *mv;
4832 	uint64_t addr;
4833 	uint32_t data;
4834 	void *cookie;
4835 	int error, rid;
4836 
4837 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4838 	    arg, &cookie);
4839 	if (error)
4840 		return (error);
4841 
4842 	/* If this is not a direct child, just bail out. */
4843 	if (device_get_parent(child) != dev) {
4844 		*cookiep = cookie;
4845 		return(0);
4846 	}
4847 
4848 	rid = rman_get_rid(irq);
4849 	if (rid == 0) {
4850 		/* Make sure that INTx is enabled */
4851 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4852 	} else {
4853 		/*
4854 		 * Check to see if the interrupt is MSI or MSI-X.
4855 		 * Ask our parent to map the MSI and give
4856 		 * us the address and data register values.
4857 		 * If we fail for some reason, teardown the
4858 		 * interrupt handler.
4859 		 */
4860 		dinfo = device_get_ivars(child);
4861 		if (dinfo->cfg.msi.msi_alloc > 0) {
4862 			if (dinfo->cfg.msi.msi_addr == 0) {
4863 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4864 			    ("MSI has handlers, but vectors not mapped"));
4865 				error = PCIB_MAP_MSI(device_get_parent(dev),
4866 				    child, rman_get_start(irq), &addr, &data);
4867 				if (error)
4868 					goto bad;
4869 				dinfo->cfg.msi.msi_addr = addr;
4870 				dinfo->cfg.msi.msi_data = data;
4871 			}
4872 			if (dinfo->cfg.msi.msi_handlers == 0)
4873 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4874 				    dinfo->cfg.msi.msi_data);
4875 			dinfo->cfg.msi.msi_handlers++;
4876 		} else {
4877 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4878 			    ("No MSI or MSI-X interrupts allocated"));
4879 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4880 			    ("MSI-X index too high"));
4881 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4882 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4883 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4884 			KASSERT(mv->mv_irq == rman_get_start(irq),
4885 			    ("IRQ mismatch"));
4886 			if (mv->mv_address == 0) {
4887 				KASSERT(mte->mte_handlers == 0,
4888 		    ("MSI-X table entry has handlers, but vector not mapped"));
4889 				error = PCIB_MAP_MSI(device_get_parent(dev),
4890 				    child, rman_get_start(irq), &addr, &data);
4891 				if (error)
4892 					goto bad;
4893 				mv->mv_address = addr;
4894 				mv->mv_data = data;
4895 			}
4896 
4897 			/*
4898 			 * The MSIX table entry must be made valid by
4899 			 * incrementing the mte_handlers before
4900 			 * calling pci_enable_msix() and
4901 			 * pci_resume_msix(). Else the MSIX rewrite
4902 			 * table quirk will not work as expected.
4903 			 */
4904 			mte->mte_handlers++;
4905 			if (mte->mte_handlers == 1) {
4906 				pci_enable_msix(child, rid - 1, mv->mv_address,
4907 				    mv->mv_data);
4908 				pci_unmask_msix(child, rid - 1);
4909 			}
4910 		}
4911 
4912 		/*
4913 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4914 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4915 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4916 		 */
4917 		if (!pci_has_quirk(pci_get_devid(child),
4918 		    PCI_QUIRK_MSI_INTX_BUG))
4919 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4920 		else
4921 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4922 	bad:
4923 		if (error) {
4924 			(void)bus_generic_teardown_intr(dev, child, irq,
4925 			    cookie);
4926 			return (error);
4927 		}
4928 	}
4929 	*cookiep = cookie;
4930 	return (0);
4931 }
4932 
4933 int
pci_teardown_intr(device_t dev,device_t child,struct resource * irq,void * cookie)4934 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4935     void *cookie)
4936 {
4937 	struct msix_table_entry *mte;
4938 	struct resource_list_entry *rle;
4939 	struct pci_devinfo *dinfo;
4940 	int error, rid;
4941 
4942 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4943 		return (EINVAL);
4944 
4945 	/* If this isn't a direct child, just bail out */
4946 	if (device_get_parent(child) != dev)
4947 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4948 
4949 	rid = rman_get_rid(irq);
4950 	if (rid == 0) {
4951 		/* Mask INTx */
4952 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4953 	} else {
4954 		/*
4955 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4956 		 * decrement the appropriate handlers count and mask the
4957 		 * MSI-X message, or disable MSI messages if the count
4958 		 * drops to 0.
4959 		 */
4960 		dinfo = device_get_ivars(child);
4961 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4962 		if (rle->res != irq)
4963 			return (EINVAL);
4964 		if (dinfo->cfg.msi.msi_alloc > 0) {
4965 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4966 			    ("MSI-X index too high"));
4967 			if (dinfo->cfg.msi.msi_handlers == 0)
4968 				return (EINVAL);
4969 			dinfo->cfg.msi.msi_handlers--;
4970 			if (dinfo->cfg.msi.msi_handlers == 0)
4971 				pci_disable_msi(child);
4972 		} else {
4973 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4974 			    ("No MSI or MSI-X interrupts allocated"));
4975 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4976 			    ("MSI-X index too high"));
4977 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4978 			if (mte->mte_handlers == 0)
4979 				return (EINVAL);
4980 			mte->mte_handlers--;
4981 			if (mte->mte_handlers == 0)
4982 				pci_mask_msix(child, rid - 1);
4983 		}
4984 	}
4985 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4986 	if (rid > 0)
4987 		KASSERT(error == 0,
4988 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4989 	return (error);
4990 }
4991 
4992 int
pci_print_child(device_t dev,device_t child)4993 pci_print_child(device_t dev, device_t child)
4994 {
4995 	struct pci_devinfo *dinfo;
4996 	struct resource_list *rl;
4997 	int retval = 0;
4998 
4999 	dinfo = device_get_ivars(child);
5000 	rl = &dinfo->resources;
5001 
5002 	retval += bus_print_child_header(dev, child);
5003 
5004 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
5005 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
5006 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
5007 	if (device_get_flags(dev))
5008 		retval += printf(" flags %#x", device_get_flags(dev));
5009 
5010 	retval += printf(" at device %d.%d", pci_get_slot(child),
5011 	    pci_get_function(child));
5012 
5013 	retval += bus_print_child_domain(dev, child);
5014 	retval += bus_print_child_footer(dev, child);
5015 
5016 	return (retval);
5017 }
5018 
5019 static const struct
5020 {
5021 	int		class;
5022 	int		subclass;
5023 	int		report; /* 0 = bootverbose, 1 = always */
5024 	const char	*desc;
5025 } pci_nomatch_tab[] = {
5026 	{PCIC_OLD,		-1,			1, "old"},
5027 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
5028 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
5029 	{PCIC_STORAGE,		-1,			1, "mass storage"},
5030 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
5031 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
5032 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
5033 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
5034 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
5035 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
5036 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
5037 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
5038 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
5039 	{PCIC_NETWORK,		-1,			1, "network"},
5040 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
5041 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
5042 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
5043 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
5044 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
5045 	{PCIC_DISPLAY,		-1,			1, "display"},
5046 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
5047 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
5048 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
5049 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
5050 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
5051 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
5052 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
5053 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
5054 	{PCIC_MEMORY,		-1,			1, "memory"},
5055 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
5056 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
5057 	{PCIC_BRIDGE,		-1,			1, "bridge"},
5058 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
5059 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
5060 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
5061 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
5062 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
5063 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
5064 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
5065 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
5066 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
5067 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
5068 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
5069 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
5070 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
5071 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
5072 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
5073 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
5074 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
5075 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
5076 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
5077 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
5078 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
5079 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
5080 	{PCIC_INPUTDEV,		-1,			1, "input device"},
5081 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
5082 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
5083 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
5084 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
5085 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
5086 	{PCIC_DOCKING,		-1,			1, "docking station"},
5087 	{PCIC_PROCESSOR,	-1,			1, "processor"},
5088 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
5089 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
5090 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
5091 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
5092 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
5093 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5094 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5095 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5096 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5097 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5098 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5099 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5100 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5101 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5102 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5103 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5104 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5105 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5106 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5107 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5108 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5109 	{PCIC_DASP,		-1,			0, "dasp"},
5110 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5111 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5112 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5113 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5114 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5115 	{0, 0, 0,		NULL}
5116 };
5117 
5118 void
pci_probe_nomatch(device_t dev,device_t child)5119 pci_probe_nomatch(device_t dev, device_t child)
5120 {
5121 	int i, report;
5122 	const char *cp, *scp;
5123 	char *device;
5124 
5125 	/*
5126 	 * Look for a listing for this device in a loaded device database.
5127 	 */
5128 	report = 1;
5129 	if ((device = pci_describe_device(child)) != NULL) {
5130 		device_printf(dev, "<%s>", device);
5131 		free(device, M_DEVBUF);
5132 	} else {
5133 		/*
5134 		 * Scan the class/subclass descriptions for a general
5135 		 * description.
5136 		 */
5137 		cp = "unknown";
5138 		scp = NULL;
5139 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5140 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5141 				if (pci_nomatch_tab[i].subclass == -1) {
5142 					cp = pci_nomatch_tab[i].desc;
5143 					report = pci_nomatch_tab[i].report;
5144 				} else if (pci_nomatch_tab[i].subclass ==
5145 				    pci_get_subclass(child)) {
5146 					scp = pci_nomatch_tab[i].desc;
5147 					report = pci_nomatch_tab[i].report;
5148 				}
5149 			}
5150 		}
5151 		if (report || bootverbose) {
5152 			device_printf(dev, "<%s%s%s>",
5153 			    cp ? cp : "",
5154 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5155 			    scp ? scp : "");
5156 		}
5157 	}
5158 	if (report || bootverbose) {
5159 		printf(" at device %d.%d (no driver attached)\n",
5160 		    pci_get_slot(child), pci_get_function(child));
5161 	}
5162 	pci_cfg_save(child, device_get_ivars(child), 1);
5163 }
5164 
5165 void
pci_child_detached(device_t dev,device_t child)5166 pci_child_detached(device_t dev, device_t child)
5167 {
5168 	struct pci_devinfo *dinfo;
5169 	struct resource_list *rl;
5170 
5171 	dinfo = device_get_ivars(child);
5172 	rl = &dinfo->resources;
5173 
5174 	/*
5175 	 * Have to deallocate IRQs before releasing any MSI messages and
5176 	 * have to release MSI messages before deallocating any memory
5177 	 * BARs.
5178 	 */
5179 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5180 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5181 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5182 		if (dinfo->cfg.msi.msi_alloc != 0)
5183 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5184 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5185 		else
5186 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5187 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5188 		(void)pci_release_msi(child);
5189 	}
5190 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5191 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5192 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5193 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5194 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5195 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5196 
5197 	pci_cfg_save(child, dinfo, 1);
5198 }
5199 
5200 /*
5201  * Parse the PCI device database, if loaded, and return a pointer to a
5202  * description of the device.
5203  *
5204  * The database is flat text formatted as follows:
5205  *
5206  * Any line not in a valid format is ignored.
5207  * Lines are terminated with newline '\n' characters.
5208  *
5209  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5210  * the vendor name.
5211  *
5212  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5213  * - devices cannot be listed without a corresponding VENDOR line.
5214  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5215  * another TAB, then the device name.
5216  */
5217 
5218 /*
5219  * Assuming (ptr) points to the beginning of a line in the database,
5220  * return the vendor or device and description of the next entry.
5221  * The value of (vendor) or (device) inappropriate for the entry type
5222  * is set to -1.  Returns nonzero at the end of the database.
5223  *
5224  * Note that this is slightly unrobust in the face of corrupt data;
5225  * we attempt to safeguard against this by spamming the end of the
5226  * database with a newline when we initialise.
5227  */
5228 static int
pci_describe_parse_line(char ** ptr,int * vendor,int * device,char ** desc)5229 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5230 {
5231 	char	*cp = *ptr;
5232 	int	left;
5233 
5234 	*device = -1;
5235 	*vendor = -1;
5236 	**desc = '\0';
5237 	for (;;) {
5238 		left = pci_vendordata_size - (cp - pci_vendordata);
5239 		if (left <= 0) {
5240 			*ptr = cp;
5241 			return(1);
5242 		}
5243 
5244 		/* vendor entry? */
5245 		if (*cp != '\t' &&
5246 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5247 			break;
5248 		/* device entry? */
5249 		if (*cp == '\t' &&
5250 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5251 			break;
5252 
5253 		/* skip to next line */
5254 		while (*cp != '\n' && left > 0) {
5255 			cp++;
5256 			left--;
5257 		}
5258 		if (*cp == '\n') {
5259 			cp++;
5260 			left--;
5261 		}
5262 	}
5263 	/* skip to next line */
5264 	while (*cp != '\n' && left > 0) {
5265 		cp++;
5266 		left--;
5267 	}
5268 	if (*cp == '\n' && left > 0)
5269 		cp++;
5270 	*ptr = cp;
5271 	return(0);
5272 }
5273 
5274 static char *
pci_describe_device(device_t dev)5275 pci_describe_device(device_t dev)
5276 {
5277 	int	vendor, device;
5278 	char	*desc, *vp, *dp, *line;
5279 
5280 	desc = vp = dp = NULL;
5281 
5282 	/*
5283 	 * If we have no vendor data, we can't do anything.
5284 	 */
5285 	if (pci_vendordata == NULL)
5286 		goto out;
5287 
5288 	/*
5289 	 * Scan the vendor data looking for this device
5290 	 */
5291 	line = pci_vendordata;
5292 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5293 		goto out;
5294 	for (;;) {
5295 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5296 			goto out;
5297 		if (vendor == pci_get_vendor(dev))
5298 			break;
5299 	}
5300 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5301 		goto out;
5302 	for (;;) {
5303 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5304 			*dp = 0;
5305 			break;
5306 		}
5307 		if (vendor != -1) {
5308 			*dp = 0;
5309 			break;
5310 		}
5311 		if (device == pci_get_device(dev))
5312 			break;
5313 	}
5314 	if (dp[0] == '\0')
5315 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5316 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5317 	    NULL)
5318 		sprintf(desc, "%s, %s", vp, dp);
5319 out:
5320 	if (vp != NULL)
5321 		free(vp, M_DEVBUF);
5322 	if (dp != NULL)
5323 		free(dp, M_DEVBUF);
5324 	return(desc);
5325 }
5326 
5327 int
pci_read_ivar(device_t dev,device_t child,int which,uintptr_t * result)5328 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5329 {
5330 	struct pci_devinfo *dinfo;
5331 	pcicfgregs *cfg;
5332 
5333 	dinfo = device_get_ivars(child);
5334 	cfg = &dinfo->cfg;
5335 
5336 	switch (which) {
5337 	case PCI_IVAR_ETHADDR:
5338 		/*
5339 		 * The generic accessor doesn't deal with failure, so
5340 		 * we set the return value, then return an error.
5341 		 */
5342 		*((uint8_t **) result) = NULL;
5343 		return (EINVAL);
5344 	case PCI_IVAR_SUBVENDOR:
5345 		*result = cfg->subvendor;
5346 		break;
5347 	case PCI_IVAR_SUBDEVICE:
5348 		*result = cfg->subdevice;
5349 		break;
5350 	case PCI_IVAR_VENDOR:
5351 		*result = cfg->vendor;
5352 		break;
5353 	case PCI_IVAR_DEVICE:
5354 		*result = cfg->device;
5355 		break;
5356 	case PCI_IVAR_DEVID:
5357 		*result = (cfg->device << 16) | cfg->vendor;
5358 		break;
5359 	case PCI_IVAR_CLASS:
5360 		*result = cfg->baseclass;
5361 		break;
5362 	case PCI_IVAR_SUBCLASS:
5363 		*result = cfg->subclass;
5364 		break;
5365 	case PCI_IVAR_PROGIF:
5366 		*result = cfg->progif;
5367 		break;
5368 	case PCI_IVAR_REVID:
5369 		*result = cfg->revid;
5370 		break;
5371 	case PCI_IVAR_INTPIN:
5372 		*result = cfg->intpin;
5373 		break;
5374 	case PCI_IVAR_IRQ:
5375 		*result = cfg->intline;
5376 		break;
5377 	case PCI_IVAR_DOMAIN:
5378 		*result = cfg->domain;
5379 		break;
5380 	case PCI_IVAR_BUS:
5381 		*result = cfg->bus;
5382 		break;
5383 	case PCI_IVAR_SLOT:
5384 		*result = cfg->slot;
5385 		break;
5386 	case PCI_IVAR_FUNCTION:
5387 		*result = cfg->func;
5388 		break;
5389 	case PCI_IVAR_CMDREG:
5390 		*result = cfg->cmdreg;
5391 		break;
5392 	case PCI_IVAR_CACHELNSZ:
5393 		*result = cfg->cachelnsz;
5394 		break;
5395 	case PCI_IVAR_MINGNT:
5396 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5397 			*result = -1;
5398 			return (EINVAL);
5399 		}
5400 		*result = cfg->mingnt;
5401 		break;
5402 	case PCI_IVAR_MAXLAT:
5403 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5404 			*result = -1;
5405 			return (EINVAL);
5406 		}
5407 		*result = cfg->maxlat;
5408 		break;
5409 	case PCI_IVAR_LATTIMER:
5410 		*result = cfg->lattimer;
5411 		break;
5412 	default:
5413 		return (ENOENT);
5414 	}
5415 	return (0);
5416 }
5417 
5418 int
pci_write_ivar(device_t dev,device_t child,int which,uintptr_t value)5419 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5420 {
5421 	struct pci_devinfo *dinfo;
5422 
5423 	dinfo = device_get_ivars(child);
5424 
5425 	switch (which) {
5426 	case PCI_IVAR_INTPIN:
5427 		dinfo->cfg.intpin = value;
5428 		return (0);
5429 	case PCI_IVAR_ETHADDR:
5430 	case PCI_IVAR_SUBVENDOR:
5431 	case PCI_IVAR_SUBDEVICE:
5432 	case PCI_IVAR_VENDOR:
5433 	case PCI_IVAR_DEVICE:
5434 	case PCI_IVAR_DEVID:
5435 	case PCI_IVAR_CLASS:
5436 	case PCI_IVAR_SUBCLASS:
5437 	case PCI_IVAR_PROGIF:
5438 	case PCI_IVAR_REVID:
5439 	case PCI_IVAR_IRQ:
5440 	case PCI_IVAR_DOMAIN:
5441 	case PCI_IVAR_BUS:
5442 	case PCI_IVAR_SLOT:
5443 	case PCI_IVAR_FUNCTION:
5444 		return (EINVAL);	/* disallow for now */
5445 
5446 	default:
5447 		return (ENOENT);
5448 	}
5449 }
5450 
5451 #include "opt_ddb.h"
5452 #ifdef DDB
5453 #include <ddb/ddb.h>
5454 #include <sys/cons.h>
5455 
5456 /*
5457  * List resources based on pci map registers, used for within ddb
5458  */
5459 
DB_SHOW_COMMAND_FLAGS(pciregs,db_pci_dump,DB_CMD_MEMSAFE)5460 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5461 {
5462 	struct pci_devinfo *dinfo;
5463 	struct devlist *devlist_head;
5464 	struct pci_conf *p;
5465 	const char *name;
5466 	int i, error, none_count;
5467 
5468 	none_count = 0;
5469 	/* get the head of the device queue */
5470 	devlist_head = &pci_devq;
5471 
5472 	/*
5473 	 * Go through the list of devices and print out devices
5474 	 */
5475 	for (error = 0, i = 0,
5476 	     dinfo = STAILQ_FIRST(devlist_head);
5477 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5478 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5479 		/* Populate pd_name and pd_unit */
5480 		name = NULL;
5481 		if (dinfo->cfg.dev)
5482 			name = device_get_name(dinfo->cfg.dev);
5483 
5484 		p = &dinfo->conf;
5485 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5486 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5487 			(name && *name) ? name : "none",
5488 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5489 			none_count++,
5490 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5491 			p->pc_sel.pc_func, (p->pc_class << 16) |
5492 			(p->pc_subclass << 8) | p->pc_progif,
5493 			(p->pc_subdevice << 16) | p->pc_subvendor,
5494 			(p->pc_device << 16) | p->pc_vendor,
5495 			p->pc_revid, p->pc_hdr);
5496 	}
5497 }
5498 #endif /* DDB */
5499 
5500 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)5501 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5502     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5503     u_int flags)
5504 {
5505 	struct pci_devinfo *dinfo = device_get_ivars(child);
5506 	struct resource_list *rl = &dinfo->resources;
5507 	struct resource *res;
5508 	struct pci_map *pm;
5509 	uint16_t cmd;
5510 	pci_addr_t map, testval;
5511 	int mapsize;
5512 
5513 	res = NULL;
5514 
5515 	/* If rid is managed by EA, ignore it */
5516 	if (pci_ea_is_enabled(child, *rid))
5517 		goto out;
5518 
5519 	pm = pci_find_bar(child, *rid);
5520 	if (pm != NULL) {
5521 		/* This is a BAR that we failed to allocate earlier. */
5522 		mapsize = pm->pm_size;
5523 		map = pm->pm_value;
5524 	} else {
5525 		/*
5526 		 * Weed out the bogons, and figure out how large the
5527 		 * BAR/map is.  BARs that read back 0 here are bogus
5528 		 * and unimplemented.  Note: atapci in legacy mode are
5529 		 * special and handled elsewhere in the code.  If you
5530 		 * have a atapci device in legacy mode and it fails
5531 		 * here, that other code is broken.
5532 		 */
5533 		pci_read_bar(child, *rid, &map, &testval, NULL);
5534 
5535 		/*
5536 		 * Determine the size of the BAR and ignore BARs with a size
5537 		 * of 0.  Device ROM BARs use a different mask value.
5538 		 */
5539 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5540 			mapsize = pci_romsize(testval);
5541 		else
5542 			mapsize = pci_mapsize(testval);
5543 		if (mapsize == 0)
5544 			goto out;
5545 		pm = pci_add_bar(child, *rid, map, mapsize);
5546 	}
5547 
5548 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5549 		if (type != SYS_RES_MEMORY) {
5550 			if (bootverbose)
5551 				device_printf(dev,
5552 				    "child %s requested type %d for rid %#x,"
5553 				    " but the BAR says it is an memio\n",
5554 				    device_get_nameunit(child), type, *rid);
5555 			goto out;
5556 		}
5557 	} else {
5558 		if (type != SYS_RES_IOPORT) {
5559 			if (bootverbose)
5560 				device_printf(dev,
5561 				    "child %s requested type %d for rid %#x,"
5562 				    " but the BAR says it is an ioport\n",
5563 				    device_get_nameunit(child), type, *rid);
5564 			goto out;
5565 		}
5566 	}
5567 
5568 	/*
5569 	 * For real BARs, we need to override the size that
5570 	 * the driver requests, because that's what the BAR
5571 	 * actually uses and we would otherwise have a
5572 	 * situation where we might allocate the excess to
5573 	 * another driver, which won't work.
5574 	 */
5575 	count = ((pci_addr_t)1 << mapsize) * num;
5576 	if (RF_ALIGNMENT(flags) < mapsize)
5577 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5578 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5579 		flags |= RF_PREFETCHABLE;
5580 
5581 	/*
5582 	 * Allocate enough resource, and then write back the
5583 	 * appropriate BAR for that resource.
5584 	 */
5585 	resource_list_add(rl, type, *rid, start, end, count);
5586 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5587 	    count, flags & ~RF_ACTIVE);
5588 	if (res == NULL) {
5589 		resource_list_delete(rl, type, *rid);
5590 		device_printf(child,
5591 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5592 		    count, *rid, type, start, end);
5593 		goto out;
5594 	}
5595 	if (bootverbose)
5596 		device_printf(child,
5597 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5598 		    count, *rid, type, rman_get_start(res));
5599 
5600 	/* Disable decoding via the CMD register before updating the BAR */
5601 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5602 	pci_write_config(child, PCIR_COMMAND,
5603 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5604 
5605 	map = rman_get_start(res);
5606 	pci_write_bar(child, pm, map);
5607 
5608 	/* Restore the original value of the CMD register */
5609 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5610 out:
5611 	return (res);
5612 }
5613 
5614 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)5615 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5616     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5617     u_int flags)
5618 {
5619 	struct pci_devinfo *dinfo;
5620 	struct resource_list *rl;
5621 	struct resource_list_entry *rle;
5622 	struct resource *res;
5623 	pcicfgregs *cfg;
5624 
5625 	/*
5626 	 * Perform lazy resource allocation
5627 	 */
5628 	dinfo = device_get_ivars(child);
5629 	rl = &dinfo->resources;
5630 	cfg = &dinfo->cfg;
5631 	switch (type) {
5632 	case PCI_RES_BUS:
5633 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5634 		    flags));
5635 	case SYS_RES_IRQ:
5636 		/*
5637 		 * Can't alloc legacy interrupt once MSI messages have
5638 		 * been allocated.
5639 		 */
5640 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5641 		    cfg->msix.msix_alloc > 0))
5642 			return (NULL);
5643 
5644 		/*
5645 		 * If the child device doesn't have an interrupt
5646 		 * routed and is deserving of an interrupt, try to
5647 		 * assign it one.
5648 		 */
5649 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5650 		    (cfg->intpin != 0))
5651 			pci_assign_interrupt(dev, child, 0);
5652 		break;
5653 	case SYS_RES_IOPORT:
5654 	case SYS_RES_MEMORY:
5655 		/*
5656 		 * PCI-PCI bridge I/O window resources are not BARs.
5657 		 * For those allocations just pass the request up the
5658 		 * tree.
5659 		 */
5660 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5661 			switch (*rid) {
5662 			case PCIR_IOBASEL_1:
5663 			case PCIR_MEMBASE_1:
5664 			case PCIR_PMBASEL_1:
5665 				/*
5666 				 * XXX: Should we bother creating a resource
5667 				 * list entry?
5668 				 */
5669 				return (bus_generic_alloc_resource(dev, child,
5670 				    type, rid, start, end, count, flags));
5671 			}
5672 		}
5673 		/* Reserve resources for this BAR if needed. */
5674 		rle = resource_list_find(rl, type, *rid);
5675 		if (rle == NULL) {
5676 			res = pci_reserve_map(dev, child, type, rid, start, end,
5677 			    count, num, flags);
5678 			if (res == NULL)
5679 				return (NULL);
5680 		}
5681 	}
5682 	return (resource_list_alloc(rl, dev, child, type, rid,
5683 	    start, end, count, flags));
5684 }
5685 
5686 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)5687 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5688     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5689 {
5690 #ifdef PCI_IOV
5691 	struct pci_devinfo *dinfo;
5692 #endif
5693 
5694 	if (device_get_parent(child) != dev)
5695 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5696 		    type, rid, start, end, count, flags));
5697 
5698 #ifdef PCI_IOV
5699 	dinfo = device_get_ivars(child);
5700 	if (dinfo->cfg.flags & PCICFG_VF) {
5701 		switch (type) {
5702 		/* VFs can't have I/O BARs. */
5703 		case SYS_RES_IOPORT:
5704 			return (NULL);
5705 		case SYS_RES_MEMORY:
5706 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5707 			    start, end, count, flags));
5708 		}
5709 
5710 		/* Fall through for other types of resource allocations. */
5711 	}
5712 #endif
5713 
5714 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5715 	    count, 1, flags));
5716 }
5717 
5718 int
pci_release_resource(device_t dev,device_t child,struct resource * r)5719 pci_release_resource(device_t dev, device_t child, struct resource *r)
5720 {
5721 	struct pci_devinfo *dinfo;
5722 	struct resource_list *rl;
5723 	pcicfgregs *cfg __unused;
5724 
5725 	if (device_get_parent(child) != dev)
5726 		return (bus_generic_release_resource(dev, child, r));
5727 
5728 	dinfo = device_get_ivars(child);
5729 	cfg = &dinfo->cfg;
5730 
5731 #ifdef PCI_IOV
5732 	if (cfg->flags & PCICFG_VF) {
5733 		switch (rman_get_type(r)) {
5734 		/* VFs can't have I/O BARs. */
5735 		case SYS_RES_IOPORT:
5736 			return (EDOOFUS);
5737 		case SYS_RES_MEMORY:
5738 			return (pci_vf_release_mem_resource(dev, child, r));
5739 		}
5740 
5741 		/* Fall through for other types of resource allocations. */
5742 	}
5743 #endif
5744 
5745 	/*
5746 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5747 	 * those allocations just pass the request up the tree.
5748 	 */
5749 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5750 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5751 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5752 		switch (rman_get_rid(r)) {
5753 		case PCIR_IOBASEL_1:
5754 		case PCIR_MEMBASE_1:
5755 		case PCIR_PMBASEL_1:
5756 			return (bus_generic_release_resource(dev, child, r));
5757 		}
5758 	}
5759 
5760 	rl = &dinfo->resources;
5761 	return (resource_list_release(rl, dev, child, r));
5762 }
5763 
5764 int
pci_activate_resource(device_t dev,device_t child,struct resource * r)5765 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5766 {
5767 	struct pci_devinfo *dinfo;
5768 	int error, rid, type;
5769 
5770 	if (device_get_parent(child) != dev)
5771 		return (bus_generic_activate_resource(dev, child, r));
5772 
5773 	dinfo = device_get_ivars(child);
5774 #ifdef PCI_IOV
5775 	if (dinfo->cfg.flags & PCICFG_VF) {
5776 		switch (rman_get_type(r)) {
5777 		/* VFs can't have I/O BARs. */
5778 		case SYS_RES_IOPORT:
5779 			error = EINVAL;
5780 			break;
5781 		case SYS_RES_MEMORY:
5782 			error = pci_vf_activate_mem_resource(dev, child, r);
5783 			break;
5784 		default:
5785 			error = bus_generic_activate_resource(dev, child, r);
5786 			break;
5787 		}
5788 	} else
5789 #endif
5790 		error = bus_generic_activate_resource(dev, child, r);
5791 	if (error)
5792 		return (error);
5793 
5794 	rid = rman_get_rid(r);
5795 	type = rman_get_type(r);
5796 
5797 	/* Device ROMs need their decoding explicitly enabled. */
5798 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5799 		pci_write_bar(child, pci_find_bar(child, rid),
5800 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5801 
5802 	/* Enable decoding in the command register when activating BARs. */
5803 	switch (type) {
5804 	case SYS_RES_IOPORT:
5805 	case SYS_RES_MEMORY:
5806 		error = PCI_ENABLE_IO(dev, child, type);
5807 		break;
5808 	}
5809 	return (error);
5810 }
5811 
5812 int
pci_deactivate_resource(device_t dev,device_t child,struct resource * r)5813 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5814 {
5815 	struct pci_devinfo *dinfo;
5816 	int error, rid, type;
5817 
5818 	if (device_get_parent(child) != dev)
5819 		return (bus_generic_deactivate_resource(dev, child, r));
5820 
5821 	dinfo = device_get_ivars(child);
5822 #ifdef PCI_IOV
5823 	if (dinfo->cfg.flags & PCICFG_VF) {
5824 		switch (rman_get_type(r)) {
5825 		/* VFs can't have I/O BARs. */
5826 		case SYS_RES_IOPORT:
5827 			error = EINVAL;
5828 			break;
5829 		case SYS_RES_MEMORY:
5830 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5831 			break;
5832 		default:
5833 			error = bus_generic_deactivate_resource(dev, child, r);
5834 			break;
5835 		}
5836 	} else
5837 #endif
5838 		error = bus_generic_deactivate_resource(dev, child, r);
5839 	if (error)
5840 		return (error);
5841 
5842 	/* Disable decoding for device ROMs. */
5843 	rid = rman_get_rid(r);
5844 	type = rman_get_type(r);
5845 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5846 		pci_write_bar(child, pci_find_bar(child, rid),
5847 		    rman_get_start(r));
5848 	return (0);
5849 }
5850 
5851 int
pci_adjust_resource(device_t dev,device_t child,struct resource * r,rman_res_t start,rman_res_t end)5852 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5853     rman_res_t start, rman_res_t end)
5854 {
5855 #ifdef PCI_IOV
5856 	struct pci_devinfo *dinfo;
5857 
5858 	if (device_get_parent(child) != dev)
5859 		return (bus_generic_adjust_resource(dev, child, r, start,
5860 		    end));
5861 
5862 	dinfo = device_get_ivars(child);
5863 	if (dinfo->cfg.flags & PCICFG_VF) {
5864 		switch (rman_get_type(r)) {
5865 		/* VFs can't have I/O BARs. */
5866 		case SYS_RES_IOPORT:
5867 			return (EINVAL);
5868 		case SYS_RES_MEMORY:
5869 			return (pci_vf_adjust_mem_resource(dev, child, r,
5870 			    start, end));
5871 		}
5872 
5873 		/* Fall through for other types of resource allocations. */
5874 	}
5875 #endif
5876 
5877 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5878 }
5879 
5880 int
pci_map_resource(device_t dev,device_t child,struct resource * r,struct resource_map_request * argsp,struct resource_map * map)5881 pci_map_resource(device_t dev, device_t child, struct resource *r,
5882     struct resource_map_request *argsp, struct resource_map *map)
5883 {
5884 #ifdef PCI_IOV
5885 	struct pci_devinfo *dinfo;
5886 
5887 	if (device_get_parent(child) != dev)
5888 		return (bus_generic_map_resource(dev, child, r, argsp,
5889 		    map));
5890 
5891 	dinfo = device_get_ivars(child);
5892 	if (dinfo->cfg.flags & PCICFG_VF) {
5893 		switch (rman_get_type(r)) {
5894 		/* VFs can't have I/O BARs. */
5895 		case SYS_RES_IOPORT:
5896 			return (EINVAL);
5897 		case SYS_RES_MEMORY:
5898 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5899 			    map));
5900 		}
5901 
5902 		/* Fall through for other types of resource allocations. */
5903 	}
5904 #endif
5905 
5906 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5907 }
5908 
5909 int
pci_unmap_resource(device_t dev,device_t child,struct resource * r,struct resource_map * map)5910 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5911     struct resource_map *map)
5912 {
5913 #ifdef PCI_IOV
5914 	struct pci_devinfo *dinfo;
5915 
5916 	if (device_get_parent(child) != dev)
5917 		return (bus_generic_unmap_resource(dev, child, r, map));
5918 
5919 	dinfo = device_get_ivars(child);
5920 	if (dinfo->cfg.flags & PCICFG_VF) {
5921 		switch (rman_get_type(r)) {
5922 		/* VFs can't have I/O BARs. */
5923 		case SYS_RES_IOPORT:
5924 			return (EINVAL);
5925 		case SYS_RES_MEMORY:
5926 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5927 		}
5928 
5929 		/* Fall through for other types of resource allocations. */
5930 	}
5931 #endif
5932 
5933 	return (bus_generic_unmap_resource(dev, child, r, map));
5934 }
5935 
5936 void
pci_child_deleted(device_t dev,device_t child)5937 pci_child_deleted(device_t dev, device_t child)
5938 {
5939 	struct resource_list_entry *rle;
5940 	struct resource_list *rl;
5941 	struct pci_devinfo *dinfo;
5942 
5943 	dinfo = device_get_ivars(child);
5944 	rl = &dinfo->resources;
5945 
5946 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5947 
5948 	/* Turn off access to resources we're about to free */
5949 	if (bus_child_present(child) != 0) {
5950 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5951 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5952 
5953 		pci_disable_busmaster(child);
5954 	}
5955 
5956 	/* Free all allocated resources */
5957 	STAILQ_FOREACH(rle, rl, link) {
5958 		if (rle->res) {
5959 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5960 			    resource_list_busy(rl, rle->type, rle->rid)) {
5961 				pci_printf(&dinfo->cfg,
5962 				    "Resource still owned, oops. "
5963 				    "(type=%d, rid=%d, addr=%lx)\n",
5964 				    rle->type, rle->rid,
5965 				    rman_get_start(rle->res));
5966 				bus_release_resource(child, rle->type, rle->rid,
5967 				    rle->res);
5968 			}
5969 			resource_list_unreserve(rl, dev, child, rle->type,
5970 			    rle->rid);
5971 		}
5972 	}
5973 	resource_list_free(rl);
5974 
5975 	pci_freecfg(dinfo);
5976 }
5977 
5978 void
pci_delete_resource(device_t dev,device_t child,int type,int rid)5979 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5980 {
5981 	struct pci_devinfo *dinfo;
5982 	struct resource_list *rl;
5983 	struct resource_list_entry *rle;
5984 
5985 	if (device_get_parent(child) != dev)
5986 		return;
5987 
5988 	dinfo = device_get_ivars(child);
5989 	rl = &dinfo->resources;
5990 	rle = resource_list_find(rl, type, rid);
5991 	if (rle == NULL)
5992 		return;
5993 
5994 	if (rle->res) {
5995 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5996 		    resource_list_busy(rl, type, rid)) {
5997 			device_printf(dev, "delete_resource: "
5998 			    "Resource still owned by child, oops. "
5999 			    "(type=%d, rid=%d, addr=%jx)\n",
6000 			    type, rid, rman_get_start(rle->res));
6001 			return;
6002 		}
6003 		resource_list_unreserve(rl, dev, child, type, rid);
6004 	}
6005 	resource_list_delete(rl, type, rid);
6006 }
6007 
6008 struct resource_list *
pci_get_resource_list(device_t dev,device_t child)6009 pci_get_resource_list (device_t dev, device_t child)
6010 {
6011 	struct pci_devinfo *dinfo = device_get_ivars(child);
6012 
6013 	return (&dinfo->resources);
6014 }
6015 
6016 #ifdef IOMMU
6017 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)6018 pci_get_dma_tag(device_t bus, device_t dev)
6019 {
6020 	bus_dma_tag_t tag;
6021 	struct pci_softc *sc;
6022 
6023 	if (device_get_parent(dev) == bus) {
6024 		/* try iommu and return if it works */
6025 		tag = iommu_get_dma_tag(bus, dev);
6026 	} else
6027 		tag = NULL;
6028 	if (tag == NULL) {
6029 		sc = device_get_softc(bus);
6030 		tag = sc->sc_dma_tag;
6031 	}
6032 	return (tag);
6033 }
6034 #else
6035 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)6036 pci_get_dma_tag(device_t bus, device_t dev)
6037 {
6038 	struct pci_softc *sc = device_get_softc(bus);
6039 
6040 	return (sc->sc_dma_tag);
6041 }
6042 #endif
6043 
6044 uint32_t
pci_read_config_method(device_t dev,device_t child,int reg,int width)6045 pci_read_config_method(device_t dev, device_t child, int reg, int width)
6046 {
6047 	struct pci_devinfo *dinfo = device_get_ivars(child);
6048 	pcicfgregs *cfg = &dinfo->cfg;
6049 
6050 #ifdef PCI_IOV
6051 	/*
6052 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
6053 	 * emulate them here.
6054 	 */
6055 	if (cfg->flags & PCICFG_VF) {
6056 		if (reg == PCIR_VENDOR) {
6057 			switch (width) {
6058 			case 4:
6059 				return (cfg->device << 16 | cfg->vendor);
6060 			case 2:
6061 				return (cfg->vendor);
6062 			case 1:
6063 				return (cfg->vendor & 0xff);
6064 			default:
6065 				return (0xffffffff);
6066 			}
6067 		} else if (reg == PCIR_DEVICE) {
6068 			switch (width) {
6069 			/* Note that an unaligned 4-byte read is an error. */
6070 			case 2:
6071 				return (cfg->device);
6072 			case 1:
6073 				return (cfg->device & 0xff);
6074 			default:
6075 				return (0xffffffff);
6076 			}
6077 		}
6078 	}
6079 #endif
6080 
6081 	return (PCIB_READ_CONFIG(device_get_parent(dev),
6082 	    cfg->bus, cfg->slot, cfg->func, reg, width));
6083 }
6084 
6085 void
pci_write_config_method(device_t dev,device_t child,int reg,uint32_t val,int width)6086 pci_write_config_method(device_t dev, device_t child, int reg,
6087     uint32_t val, int width)
6088 {
6089 	struct pci_devinfo *dinfo = device_get_ivars(child);
6090 	pcicfgregs *cfg = &dinfo->cfg;
6091 
6092 	PCIB_WRITE_CONFIG(device_get_parent(dev),
6093 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6094 }
6095 
6096 int
pci_child_location_method(device_t dev,device_t child,struct sbuf * sb)6097 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6098 {
6099 
6100 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6101 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6102 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6103 	return (0);
6104 }
6105 
6106 int
pci_child_pnpinfo_method(device_t dev,device_t child,struct sbuf * sb)6107 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6108 {
6109 	struct pci_devinfo *dinfo;
6110 	pcicfgregs *cfg;
6111 
6112 	dinfo = device_get_ivars(child);
6113 	cfg = &dinfo->cfg;
6114 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6115 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6116 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6117 	    cfg->progif);
6118 	return (0);
6119 }
6120 
6121 int
pci_get_device_path_method(device_t bus,device_t child,const char * locator,struct sbuf * sb)6122 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6123     struct sbuf *sb)
6124 {
6125 	device_t parent = device_get_parent(bus);
6126 	int rv;
6127 
6128 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6129 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6130 		if (rv == 0) {
6131 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6132 			    pci_get_function(child));
6133 		}
6134 		return (0);
6135 	}
6136 	return (bus_generic_get_device_path(bus, child, locator, sb));
6137 }
6138 
6139 int
pci_assign_interrupt_method(device_t dev,device_t child)6140 pci_assign_interrupt_method(device_t dev, device_t child)
6141 {
6142 	struct pci_devinfo *dinfo = device_get_ivars(child);
6143 	pcicfgregs *cfg = &dinfo->cfg;
6144 
6145 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6146 	    cfg->intpin));
6147 }
6148 
6149 static void
pci_lookup(void * arg,const char * name,device_t * dev)6150 pci_lookup(void *arg, const char *name, device_t *dev)
6151 {
6152 	long val;
6153 	char *end;
6154 	int domain, bus, slot, func;
6155 
6156 	if (*dev != NULL)
6157 		return;
6158 
6159 	/*
6160 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6161 	 * pciB:S:F.  In the latter case, the domain is assumed to
6162 	 * be zero.
6163 	 */
6164 	if (strncmp(name, "pci", 3) != 0)
6165 		return;
6166 	val = strtol(name + 3, &end, 10);
6167 	if (val < 0 || val > INT_MAX || *end != ':')
6168 		return;
6169 	domain = val;
6170 	val = strtol(end + 1, &end, 10);
6171 	if (val < 0 || val > INT_MAX || *end != ':')
6172 		return;
6173 	bus = val;
6174 	val = strtol(end + 1, &end, 10);
6175 	if (val < 0 || val > INT_MAX)
6176 		return;
6177 	slot = val;
6178 	if (*end == ':') {
6179 		val = strtol(end + 1, &end, 10);
6180 		if (val < 0 || val > INT_MAX || *end != '\0')
6181 			return;
6182 		func = val;
6183 	} else if (*end == '\0') {
6184 		func = slot;
6185 		slot = bus;
6186 		bus = domain;
6187 		domain = 0;
6188 	} else
6189 		return;
6190 
6191 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6192 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6193 		return;
6194 
6195 	*dev = pci_find_dbsf(domain, bus, slot, func);
6196 }
6197 
6198 static int
pci_modevent(module_t mod,int what,void * arg)6199 pci_modevent(module_t mod, int what, void *arg)
6200 {
6201 	static struct cdev *pci_cdev;
6202 	static eventhandler_tag tag;
6203 
6204 	switch (what) {
6205 	case MOD_LOAD:
6206 		STAILQ_INIT(&pci_devq);
6207 		pci_generation = 0;
6208 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6209 		    "pci");
6210 		pci_load_vendor_data();
6211 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6212 		    1000);
6213 		break;
6214 
6215 	case MOD_UNLOAD:
6216 		if (tag != NULL)
6217 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6218 		destroy_dev(pci_cdev);
6219 		break;
6220 	}
6221 
6222 	return (0);
6223 }
6224 
6225 static void
pci_cfg_restore_pcie(device_t dev,struct pci_devinfo * dinfo)6226 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6227 {
6228 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6229 	struct pcicfg_pcie *cfg;
6230 	int version, pos;
6231 
6232 	cfg = &dinfo->cfg.pcie;
6233 	pos = cfg->pcie_location;
6234 
6235 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6236 
6237 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6238 
6239 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6240 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6241 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6242 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6243 
6244 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6245 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6246 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6247 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6248 
6249 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6250 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6251 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6252 
6253 	if (version > 1) {
6254 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6255 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6256 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6257 	}
6258 #undef WREG
6259 }
6260 
6261 static void
pci_cfg_restore_pcix(device_t dev,struct pci_devinfo * dinfo)6262 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6263 {
6264 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6265 	    dinfo->cfg.pcix.pcix_command,  2);
6266 }
6267 
6268 void
pci_cfg_restore(device_t dev,struct pci_devinfo * dinfo)6269 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6270 {
6271 
6272 	/*
6273 	 * Restore the device to full power mode.  We must do this
6274 	 * before we restore the registers because moving from D3 to
6275 	 * D0 will cause the chip's BARs and some other registers to
6276 	 * be reset to some unknown power on reset values.  Cut down
6277 	 * the noise on boot by doing nothing if we are already in
6278 	 * state D0.
6279 	 */
6280 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6281 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6282 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6283 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6284 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6285 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6286 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6287 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6288 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6289 	case PCIM_HDRTYPE_NORMAL:
6290 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6291 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6292 		break;
6293 	case PCIM_HDRTYPE_BRIDGE:
6294 		pci_write_config(dev, PCIR_SECLAT_1,
6295 		    dinfo->cfg.bridge.br_seclat, 1);
6296 		pci_write_config(dev, PCIR_SUBBUS_1,
6297 		    dinfo->cfg.bridge.br_subbus, 1);
6298 		pci_write_config(dev, PCIR_SECBUS_1,
6299 		    dinfo->cfg.bridge.br_secbus, 1);
6300 		pci_write_config(dev, PCIR_PRIBUS_1,
6301 		    dinfo->cfg.bridge.br_pribus, 1);
6302 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6303 		    dinfo->cfg.bridge.br_control, 2);
6304 		break;
6305 	case PCIM_HDRTYPE_CARDBUS:
6306 		pci_write_config(dev, PCIR_SECLAT_2,
6307 		    dinfo->cfg.bridge.br_seclat, 1);
6308 		pci_write_config(dev, PCIR_SUBBUS_2,
6309 		    dinfo->cfg.bridge.br_subbus, 1);
6310 		pci_write_config(dev, PCIR_SECBUS_2,
6311 		    dinfo->cfg.bridge.br_secbus, 1);
6312 		pci_write_config(dev, PCIR_PRIBUS_2,
6313 		    dinfo->cfg.bridge.br_pribus, 1);
6314 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6315 		    dinfo->cfg.bridge.br_control, 2);
6316 		break;
6317 	}
6318 	pci_restore_bars(dev);
6319 
6320 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6321 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6322 
6323 	/*
6324 	 * Restore extended capabilities for PCI-Express and PCI-X
6325 	 */
6326 	if (dinfo->cfg.pcie.pcie_location != 0)
6327 		pci_cfg_restore_pcie(dev, dinfo);
6328 	if (dinfo->cfg.pcix.pcix_location != 0)
6329 		pci_cfg_restore_pcix(dev, dinfo);
6330 
6331 	/* Restore MSI and MSI-X configurations if they are present. */
6332 	if (dinfo->cfg.msi.msi_location != 0)
6333 		pci_resume_msi(dev);
6334 	if (dinfo->cfg.msix.msix_location != 0)
6335 		pci_resume_msix(dev);
6336 
6337 #ifdef PCI_IOV
6338 	if (dinfo->cfg.iov != NULL)
6339 		pci_iov_cfg_restore(dev, dinfo);
6340 #endif
6341 }
6342 
6343 static void
pci_cfg_save_pcie(device_t dev,struct pci_devinfo * dinfo)6344 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6345 {
6346 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6347 	struct pcicfg_pcie *cfg;
6348 	int version, pos;
6349 
6350 	cfg = &dinfo->cfg.pcie;
6351 	pos = cfg->pcie_location;
6352 
6353 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6354 
6355 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6356 
6357 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6358 
6359 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6360 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6361 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6362 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6363 
6364 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6365 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6366 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6367 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6368 
6369 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6370 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6371 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6372 
6373 	if (version > 1) {
6374 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6375 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6376 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6377 	}
6378 #undef RREG
6379 }
6380 
6381 static void
pci_cfg_save_pcix(device_t dev,struct pci_devinfo * dinfo)6382 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6383 {
6384 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6385 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6386 }
6387 
6388 void
pci_cfg_save(device_t dev,struct pci_devinfo * dinfo,int setstate)6389 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6390 {
6391 	uint32_t cls;
6392 	int ps;
6393 
6394 	/*
6395 	 * Some drivers apparently write to these registers w/o updating our
6396 	 * cached copy.  No harm happens if we update the copy, so do so here
6397 	 * so we can restore them.  The COMMAND register is modified by the
6398 	 * bus w/o updating the cache.  This should represent the normally
6399 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6400 	 */
6401 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6402 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6403 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6404 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6405 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6406 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6407 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6408 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6409 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6410 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6411 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6412 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6413 	case PCIM_HDRTYPE_NORMAL:
6414 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6415 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6416 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6417 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6418 		break;
6419 	case PCIM_HDRTYPE_BRIDGE:
6420 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6421 		    PCIR_SECLAT_1, 1);
6422 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6423 		    PCIR_SUBBUS_1, 1);
6424 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6425 		    PCIR_SECBUS_1, 1);
6426 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6427 		    PCIR_PRIBUS_1, 1);
6428 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6429 		    PCIR_BRIDGECTL_1, 2);
6430 		break;
6431 	case PCIM_HDRTYPE_CARDBUS:
6432 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6433 		    PCIR_SECLAT_2, 1);
6434 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6435 		    PCIR_SUBBUS_2, 1);
6436 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6437 		    PCIR_SECBUS_2, 1);
6438 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6439 		    PCIR_PRIBUS_2, 1);
6440 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6441 		    PCIR_BRIDGECTL_2, 2);
6442 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6443 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6444 		break;
6445 	}
6446 
6447 	if (dinfo->cfg.pcie.pcie_location != 0)
6448 		pci_cfg_save_pcie(dev, dinfo);
6449 
6450 	if (dinfo->cfg.pcix.pcix_location != 0)
6451 		pci_cfg_save_pcix(dev, dinfo);
6452 
6453 #ifdef PCI_IOV
6454 	if (dinfo->cfg.iov != NULL)
6455 		pci_iov_cfg_save(dev, dinfo);
6456 #endif
6457 
6458 	/*
6459 	 * don't set the state for display devices, base peripherals and
6460 	 * memory devices since bad things happen when they are powered down.
6461 	 * We should (a) have drivers that can easily detach and (b) use
6462 	 * generic drivers for these devices so that some device actually
6463 	 * attaches.  We need to make sure that when we implement (a) we don't
6464 	 * power the device down on a reattach.
6465 	 */
6466 	cls = pci_get_class(dev);
6467 	if (!setstate)
6468 		return;
6469 	switch (pci_do_power_nodriver)
6470 	{
6471 		case 0:		/* NO powerdown at all */
6472 			return;
6473 		case 1:		/* Conservative about what to power down */
6474 			if (cls == PCIC_STORAGE)
6475 				return;
6476 			/*FALLTHROUGH*/
6477 		case 2:		/* Aggressive about what to power down */
6478 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6479 			    cls == PCIC_BASEPERIPH)
6480 				return;
6481 			/*FALLTHROUGH*/
6482 		case 3:		/* Power down everything */
6483 			break;
6484 	}
6485 	/*
6486 	 * PCI spec says we can only go into D3 state from D0 state.
6487 	 * Transition from D[12] into D0 before going to D3 state.
6488 	 */
6489 	ps = pci_get_powerstate(dev);
6490 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6491 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6492 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6493 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6494 }
6495 
6496 /* Wrapper APIs suitable for device driver use. */
6497 void
pci_save_state(device_t dev)6498 pci_save_state(device_t dev)
6499 {
6500 	struct pci_devinfo *dinfo;
6501 
6502 	dinfo = device_get_ivars(dev);
6503 	pci_cfg_save(dev, dinfo, 0);
6504 }
6505 
6506 void
pci_restore_state(device_t dev)6507 pci_restore_state(device_t dev)
6508 {
6509 	struct pci_devinfo *dinfo;
6510 
6511 	dinfo = device_get_ivars(dev);
6512 	pci_cfg_restore(dev, dinfo);
6513 }
6514 
6515 static int
pci_get_id_method(device_t dev,device_t child,enum pci_id_type type,uintptr_t * id)6516 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6517     uintptr_t *id)
6518 {
6519 
6520 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6521 }
6522 
6523 /* Find the upstream port of a given PCI device in a root complex. */
6524 device_t
pci_find_pcie_root_port(device_t dev)6525 pci_find_pcie_root_port(device_t dev)
6526 {
6527 	struct pci_devinfo *dinfo;
6528 	devclass_t pci_class;
6529 	device_t pcib, bus;
6530 
6531 	pci_class = devclass_find("pci");
6532 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6533 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6534 
6535 	/*
6536 	 * Walk the bridge hierarchy until we find a PCI-e root
6537 	 * port or a non-PCI device.
6538 	 */
6539 	for (;;) {
6540 		bus = device_get_parent(dev);
6541 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6542 		    device_get_nameunit(dev)));
6543 
6544 		pcib = device_get_parent(bus);
6545 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6546 		    device_get_nameunit(bus)));
6547 
6548 		/*
6549 		 * pcib's parent must be a PCI bus for this to be a
6550 		 * PCI-PCI bridge.
6551 		 */
6552 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6553 			return (NULL);
6554 
6555 		dinfo = device_get_ivars(pcib);
6556 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6557 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6558 			return (pcib);
6559 
6560 		dev = pcib;
6561 	}
6562 }
6563 
6564 /*
6565  * Wait for pending transactions to complete on a PCI-express function.
6566  *
6567  * The maximum delay is specified in milliseconds in max_delay.  Note
6568  * that this function may sleep.
6569  *
6570  * Returns true if the function is idle and false if the timeout is
6571  * exceeded.  If dev is not a PCI-express function, this returns true.
6572  */
6573 bool
pcie_wait_for_pending_transactions(device_t dev,u_int max_delay)6574 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6575 {
6576 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6577 	uint16_t sta;
6578 	int cap;
6579 
6580 	cap = dinfo->cfg.pcie.pcie_location;
6581 	if (cap == 0)
6582 		return (true);
6583 
6584 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6585 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6586 		if (max_delay == 0)
6587 			return (false);
6588 
6589 		/* Poll once every 100 milliseconds up to the timeout. */
6590 		if (max_delay > 100) {
6591 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6592 			max_delay -= 100;
6593 		} else {
6594 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6595 			    C_HARDCLOCK);
6596 			max_delay = 0;
6597 		}
6598 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6599 	}
6600 
6601 	return (true);
6602 }
6603 
6604 /*
6605  * Determine the maximum Completion Timeout in microseconds.
6606  *
6607  * For non-PCI-express functions this returns 0.
6608  */
6609 int
pcie_get_max_completion_timeout(device_t dev)6610 pcie_get_max_completion_timeout(device_t dev)
6611 {
6612 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6613 	int cap;
6614 
6615 	cap = dinfo->cfg.pcie.pcie_location;
6616 	if (cap == 0)
6617 		return (0);
6618 
6619 	/*
6620 	 * Functions using the 1.x spec use the default timeout range of
6621 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6622 	 * support programmable timeouts also use this range.
6623 	 */
6624 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6625 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6626 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6627 		return (50 * 1000);
6628 
6629 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6630 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6631 	case PCIEM_CTL2_COMP_TIMO_100US:
6632 		return (100);
6633 	case PCIEM_CTL2_COMP_TIMO_10MS:
6634 		return (10 * 1000);
6635 	case PCIEM_CTL2_COMP_TIMO_55MS:
6636 		return (55 * 1000);
6637 	case PCIEM_CTL2_COMP_TIMO_210MS:
6638 		return (210 * 1000);
6639 	case PCIEM_CTL2_COMP_TIMO_900MS:
6640 		return (900 * 1000);
6641 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6642 		return (3500 * 1000);
6643 	case PCIEM_CTL2_COMP_TIMO_13S:
6644 		return (13 * 1000 * 1000);
6645 	case PCIEM_CTL2_COMP_TIMO_64S:
6646 		return (64 * 1000 * 1000);
6647 	default:
6648 		return (50 * 1000);
6649 	}
6650 }
6651 
6652 void
pcie_apei_error(device_t dev,int sev,uint8_t * aerp)6653 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6654 {
6655 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6656 	const char *s;
6657 	int aer;
6658 	uint32_t r, r1;
6659 	uint16_t rs;
6660 
6661 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6662 		s = "Correctable";
6663 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6664 		s = "Uncorrectable (Non-Fatal)";
6665 	else
6666 		s = "Uncorrectable (Fatal)";
6667 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6668 	if (aerp) {
6669 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6670 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6671 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6672 		} else {
6673 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6674 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6675 		}
6676 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6677 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6678 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6679 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6680 			printf(" severity 0x%08x first %d\n",
6681 			    r, rs & 0x1f);
6682 		} else
6683 			printf("\n");
6684 	}
6685 
6686 	/* As kind of recovery just report and clear the error statuses. */
6687 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6688 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6689 		if (r != 0) {
6690 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6691 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6692 		}
6693 
6694 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6695 		if (r != 0) {
6696 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6697 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6698 		}
6699 	}
6700 	if (dinfo->cfg.pcie.pcie_location != 0) {
6701 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6702 		    PCIER_DEVICE_STA, 2);
6703 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6704 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6705 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6706 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6707 			    PCIER_DEVICE_STA, rs, 2);
6708 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6709 		}
6710 	}
6711 }
6712 
6713 /*
6714  * Perform a Function Level Reset (FLR) on a device.
6715  *
6716  * This function first waits for any pending transactions to complete
6717  * within the timeout specified by max_delay.  If transactions are
6718  * still pending, the function will return false without attempting a
6719  * reset.
6720  *
6721  * If dev is not a PCI-express function or does not support FLR, this
6722  * function returns false.
6723  *
6724  * Note that no registers are saved or restored.  The caller is
6725  * responsible for saving and restoring any registers including
6726  * PCI-standard registers via pci_save_state() and
6727  * pci_restore_state().
6728  */
6729 bool
pcie_flr(device_t dev,u_int max_delay,bool force)6730 pcie_flr(device_t dev, u_int max_delay, bool force)
6731 {
6732 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6733 	uint16_t cmd, ctl;
6734 	int compl_delay;
6735 	int cap;
6736 
6737 	cap = dinfo->cfg.pcie.pcie_location;
6738 	if (cap == 0)
6739 		return (false);
6740 
6741 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6742 		return (false);
6743 
6744 	/*
6745 	 * Disable busmastering to prevent generation of new
6746 	 * transactions while waiting for the device to go idle.  If
6747 	 * the idle timeout fails, the command register is restored
6748 	 * which will re-enable busmastering.
6749 	 */
6750 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6751 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6752 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6753 		if (!force) {
6754 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6755 			return (false);
6756 		}
6757 		pci_printf(&dinfo->cfg,
6758 		    "Resetting with transactions pending after %d ms\n",
6759 		    max_delay);
6760 
6761 		/*
6762 		 * Extend the post-FLR delay to cover the maximum
6763 		 * Completion Timeout delay of anything in flight
6764 		 * during the FLR delay.  Enforce a minimum delay of
6765 		 * at least 10ms.
6766 		 */
6767 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6768 		if (compl_delay < 10)
6769 			compl_delay = 10;
6770 	} else
6771 		compl_delay = 0;
6772 
6773 	/* Initiate the reset. */
6774 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6775 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6776 	    PCIEM_CTL_INITIATE_FLR, 2);
6777 
6778 	/* Wait for 100ms. */
6779 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6780 
6781 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6782 	    PCIEM_STA_TRANSACTION_PND)
6783 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6784 	return (true);
6785 }
6786 
6787 /*
6788  * Attempt a power-management reset by cycling the device in/out of D3
6789  * state.  PCI spec says we can only go into D3 state from D0 state.
6790  * Transition from D[12] into D0 before going to D3 state.
6791  */
6792 int
pci_power_reset(device_t dev)6793 pci_power_reset(device_t dev)
6794 {
6795 	int ps;
6796 
6797 	ps = pci_get_powerstate(dev);
6798 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6799 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6800 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6801 	pci_set_powerstate(dev, ps);
6802 	return (0);
6803 }
6804 
6805 /*
6806  * Try link drop and retrain of the downstream port of upstream
6807  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6808  * cause Conventional Hot reset of the device in the slot.
6809  * Alternative, for PCIe, could be the secondary bus reset initiatied
6810  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6811  */
6812 int
pcie_link_reset(device_t port,int pcie_location)6813 pcie_link_reset(device_t port, int pcie_location)
6814 {
6815 	uint16_t v;
6816 
6817 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6818 	v |= PCIEM_LINK_CTL_LINK_DIS;
6819 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6820 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6821 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6822 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6823 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6824 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6825 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6826 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6827 }
6828 
6829 static int
pci_reset_post(device_t dev,device_t child)6830 pci_reset_post(device_t dev, device_t child)
6831 {
6832 
6833 	if (dev == device_get_parent(child))
6834 		pci_restore_state(child);
6835 	return (0);
6836 }
6837 
6838 static int
pci_reset_prepare(device_t dev,device_t child)6839 pci_reset_prepare(device_t dev, device_t child)
6840 {
6841 
6842 	if (dev == device_get_parent(child))
6843 		pci_save_state(child);
6844 	return (0);
6845 }
6846 
6847 static int
pci_reset_child(device_t dev,device_t child,int flags)6848 pci_reset_child(device_t dev, device_t child, int flags)
6849 {
6850 	int error;
6851 
6852 	if (dev == NULL || device_get_parent(child) != dev)
6853 		return (0);
6854 	if ((flags & DEVF_RESET_DETACH) != 0) {
6855 		error = device_get_state(child) == DS_ATTACHED ?
6856 		    device_detach(child) : 0;
6857 	} else {
6858 		error = BUS_SUSPEND_CHILD(dev, child);
6859 	}
6860 	if (error == 0) {
6861 		if (!pcie_flr(child, 1000, false)) {
6862 			error = BUS_RESET_PREPARE(dev, child);
6863 			if (error == 0)
6864 				pci_power_reset(child);
6865 			BUS_RESET_POST(dev, child);
6866 		}
6867 		if ((flags & DEVF_RESET_DETACH) != 0)
6868 			device_probe_and_attach(child);
6869 		else
6870 			BUS_RESUME_CHILD(dev, child);
6871 	}
6872 	return (error);
6873 }
6874 
6875 const struct pci_device_table *
pci_match_device(device_t child,const struct pci_device_table * id,size_t nelt)6876 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6877 {
6878 	bool match;
6879 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6880 
6881 	vendor = pci_get_vendor(child);
6882 	device = pci_get_device(child);
6883 	subvendor = pci_get_subvendor(child);
6884 	subdevice = pci_get_subdevice(child);
6885 	class = pci_get_class(child);
6886 	subclass = pci_get_subclass(child);
6887 	revid = pci_get_revid(child);
6888 	while (nelt-- > 0) {
6889 		match = true;
6890 		if (id->match_flag_vendor)
6891 			match &= vendor == id->vendor;
6892 		if (id->match_flag_device)
6893 			match &= device == id->device;
6894 		if (id->match_flag_subvendor)
6895 			match &= subvendor == id->subvendor;
6896 		if (id->match_flag_subdevice)
6897 			match &= subdevice == id->subdevice;
6898 		if (id->match_flag_class)
6899 			match &= class == id->class_id;
6900 		if (id->match_flag_subclass)
6901 			match &= subclass == id->subclass;
6902 		if (id->match_flag_revid)
6903 			match &= revid == id->revid;
6904 		if (match)
6905 			return (id);
6906 		id++;
6907 	}
6908 	return (NULL);
6909 }
6910 
6911 static void
pci_print_faulted_dev_name(const struct pci_devinfo * dinfo)6912 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6913 {
6914 	const char *dev_name;
6915 	device_t dev;
6916 
6917 	dev = dinfo->cfg.dev;
6918 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6919 	    dinfo->cfg.slot, dinfo->cfg.func);
6920 	dev_name = device_get_name(dev);
6921 	if (dev_name != NULL)
6922 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6923 }
6924 
6925 void
pci_print_faulted_dev(void)6926 pci_print_faulted_dev(void)
6927 {
6928 	struct pci_devinfo *dinfo;
6929 	device_t dev;
6930 	int aer, i;
6931 	uint32_t r1, r2;
6932 	uint16_t status;
6933 
6934 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6935 		dev = dinfo->cfg.dev;
6936 		status = pci_read_config(dev, PCIR_STATUS, 2);
6937 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6938 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6939 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6940 		if (status != 0) {
6941 			pci_print_faulted_dev_name(dinfo);
6942 			printf(" error 0x%04x\n", status);
6943 		}
6944 		if (dinfo->cfg.pcie.pcie_location != 0) {
6945 			status = pci_read_config(dev,
6946 			    dinfo->cfg.pcie.pcie_location +
6947 			    PCIER_DEVICE_STA, 2);
6948 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6949 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6950 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6951 				pci_print_faulted_dev_name(dinfo);
6952 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6953 				    pci_read_config(dev,
6954 				    dinfo->cfg.pcie.pcie_location +
6955 				    PCIER_DEVICE_CTL, 2),
6956 				    status);
6957 			}
6958 		}
6959 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6960 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6961 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6962 			if (r1 != 0 || r2 != 0) {
6963 				pci_print_faulted_dev_name(dinfo);
6964 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6965 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6966 				    r1, pci_read_config(dev, aer +
6967 				    PCIR_AER_UC_MASK, 4),
6968 				    pci_read_config(dev, aer +
6969 				    PCIR_AER_UC_SEVERITY, 4),
6970 				    r2, pci_read_config(dev, aer +
6971 				    PCIR_AER_COR_MASK, 4),
6972 				    pci_read_config(dev, aer +
6973 				    PCIR_AER_CAP_CONTROL, 4));
6974 				for (i = 0; i < 4; i++) {
6975 					r1 = pci_read_config(dev, aer +
6976 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6977 					printf("    HL%d: 0x%08x\n", i, r1);
6978 				}
6979 			}
6980 		}
6981 	}
6982 }
6983 
6984 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(pcierr,pci_print_faulted_dev_db,DB_CMD_MEMSAFE)6985 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6986 {
6987 
6988 	pci_print_faulted_dev();
6989 }
6990 
6991 static void
db_clear_pcie_errors(const struct pci_devinfo * dinfo)6992 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6993 {
6994 	device_t dev;
6995 	int aer;
6996 	uint32_t r;
6997 
6998 	dev = dinfo->cfg.dev;
6999 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
7000 	    PCIER_DEVICE_STA, 2);
7001 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
7002 	    PCIER_DEVICE_STA, r, 2);
7003 
7004 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
7005 		return;
7006 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
7007 	if (r != 0)
7008 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
7009 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
7010 	if (r != 0)
7011 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
7012 }
7013 
DB_COMMAND_FLAGS(pci_clearerr,db_pci_clearerr,DB_CMD_MEMSAFE)7014 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
7015 {
7016 	struct pci_devinfo *dinfo;
7017 	device_t dev;
7018 	uint16_t status, status1;
7019 
7020 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
7021 		dev = dinfo->cfg.dev;
7022 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
7023 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
7024 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
7025 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
7026 		if (status1 != 0) {
7027 			status &= ~status1;
7028 			pci_write_config(dev, PCIR_STATUS, status, 2);
7029 		}
7030 		if (dinfo->cfg.pcie.pcie_location != 0)
7031 			db_clear_pcie_errors(dinfo);
7032 	}
7033 }
7034 #endif
7035