xref: /freebsd/sys/dev/pci/pci.c (revision 5884fab46153dee52bda660bcabca95c3cc97f44)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5  * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6  * Copyright (c) 2000, BSDi
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice unmodified, this list of conditions, and the following
14  *    disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_acpi.h"
33 #include "opt_iommu.h"
34 #include "opt_bus.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/endian.h>
39 #include <sys/eventhandler.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/linker.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 #include <sys/taskqueue.h>
51 #include <sys/tree.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_extern.h>
56 
57 #include <sys/bus.h>
58 #include <machine/bus.h>
59 #include <sys/rman.h>
60 #include <machine/resource.h>
61 #include <machine/stdarg.h>
62 
63 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
64 #include <machine/intr_machdep.h>
65 #endif
66 
67 #include <sys/pciio.h>
68 #include <dev/pci/pcireg.h>
69 #include <dev/pci/pcivar.h>
70 #include <dev/pci/pci_private.h>
71 
72 #ifdef PCI_IOV
73 #include <sys/nv.h>
74 #include <dev/pci/pci_iov_private.h>
75 #endif
76 
77 #include <dev/usb/controller/xhcireg.h>
78 #include <dev/usb/controller/ehcireg.h>
79 #include <dev/usb/controller/ohcireg.h>
80 #include <dev/usb/controller/uhcireg.h>
81 
82 #include <dev/iommu/iommu.h>
83 
84 #include "pcib_if.h"
85 #include "pci_if.h"
86 
87 #define	PCIR_IS_BIOS(cfg, reg)						\
88 	(((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) ||	\
89 	 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
90 
91 static device_probe_t	pci_probe;
92 
93 static bus_reset_post_t pci_reset_post;
94 static bus_reset_prepare_t pci_reset_prepare;
95 static bus_reset_child_t pci_reset_child;
96 static bus_hint_device_unit_t pci_hint_device_unit;
97 static bus_remap_intr_t pci_remap_intr_method;
98 
99 static pci_get_id_t	pci_get_id_method;
100 
101 static int		pci_has_quirk(uint32_t devid, int quirk);
102 static pci_addr_t	pci_mapbase(uint64_t mapreg);
103 static const char	*pci_maptype(uint64_t mapreg);
104 static int		pci_maprange(uint64_t mapreg);
105 static pci_addr_t	pci_rombase(uint64_t mapreg);
106 static int		pci_romsize(uint64_t testval);
107 static void		pci_fixancient(pcicfgregs *cfg);
108 static int		pci_printf(pcicfgregs *cfg, const char *fmt, ...);
109 
110 static int		pci_porten(device_t dev);
111 static int		pci_memen(device_t dev);
112 static void		pci_assign_interrupt(device_t bus, device_t dev,
113 			    int force_route);
114 static int		pci_add_map(device_t bus, device_t dev, int reg,
115 			    struct resource_list *rl, int force, int prefetch);
116 static void		pci_load_vendor_data(void);
117 static int		pci_describe_parse_line(char **ptr, int *vendor,
118 			    int *device, char **desc);
119 static char		*pci_describe_device(device_t dev);
120 static int		pci_modevent(module_t mod, int what, void *arg);
121 static void		pci_hdrtypedata(device_t pcib, int b, int s, int f,
122 			    pcicfgregs *cfg);
123 static void		pci_read_cap(device_t pcib, pcicfgregs *cfg);
124 static int		pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
125 			    int reg, uint32_t *data);
126 #if 0
127 static int		pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
128 			    int reg, uint32_t data);
129 #endif
130 static void		pci_read_vpd(device_t pcib, pcicfgregs *cfg);
131 static void		pci_mask_msix(device_t dev, u_int index);
132 static void		pci_unmask_msix(device_t dev, u_int index);
133 static int		pci_msi_blacklisted(void);
134 static int		pci_msix_blacklisted(void);
135 static void		pci_resume_msi(device_t dev);
136 static void		pci_resume_msix(device_t dev);
137 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
138     int b, int s, int f, uint16_t vid, uint16_t did);
139 
140 static device_method_t pci_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,		pci_probe),
143 	DEVMETHOD(device_attach,	pci_attach),
144 	DEVMETHOD(device_detach,	pci_detach),
145 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,	bus_generic_suspend),
147 	DEVMETHOD(device_resume,	pci_resume),
148 
149 	/* Bus interface */
150 	DEVMETHOD(bus_print_child,	pci_print_child),
151 	DEVMETHOD(bus_probe_nomatch,	pci_probe_nomatch),
152 	DEVMETHOD(bus_read_ivar,	pci_read_ivar),
153 	DEVMETHOD(bus_write_ivar,	pci_write_ivar),
154 	DEVMETHOD(bus_driver_added,	pci_driver_added),
155 	DEVMETHOD(bus_setup_intr,	pci_setup_intr),
156 	DEVMETHOD(bus_teardown_intr,	pci_teardown_intr),
157 	DEVMETHOD(bus_reset_prepare,	pci_reset_prepare),
158 	DEVMETHOD(bus_reset_post,	pci_reset_post),
159 	DEVMETHOD(bus_reset_child,	pci_reset_child),
160 
161 	DEVMETHOD(bus_get_dma_tag,	pci_get_dma_tag),
162 	DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
163 	DEVMETHOD(bus_set_resource,	bus_generic_rl_set_resource),
164 	DEVMETHOD(bus_get_resource,	bus_generic_rl_get_resource),
165 	DEVMETHOD(bus_delete_resource,	pci_delete_resource),
166 	DEVMETHOD(bus_alloc_resource,	pci_alloc_resource),
167 	DEVMETHOD(bus_adjust_resource,	pci_adjust_resource),
168 	DEVMETHOD(bus_release_resource,	pci_release_resource),
169 	DEVMETHOD(bus_activate_resource, pci_activate_resource),
170 	DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
171 	DEVMETHOD(bus_map_resource,	pci_map_resource),
172 	DEVMETHOD(bus_unmap_resource,	pci_unmap_resource),
173 	DEVMETHOD(bus_child_deleted,	pci_child_deleted),
174 	DEVMETHOD(bus_child_detached,	pci_child_detached),
175 	DEVMETHOD(bus_child_pnpinfo,	pci_child_pnpinfo_method),
176 	DEVMETHOD(bus_child_location,	pci_child_location_method),
177 	DEVMETHOD(bus_get_device_path,	pci_get_device_path_method),
178 	DEVMETHOD(bus_hint_device_unit,	pci_hint_device_unit),
179 	DEVMETHOD(bus_remap_intr,	pci_remap_intr_method),
180 	DEVMETHOD(bus_suspend_child,	pci_suspend_child),
181 	DEVMETHOD(bus_resume_child,	pci_resume_child),
182 	DEVMETHOD(bus_rescan,		pci_rescan_method),
183 
184 	/* PCI interface */
185 	DEVMETHOD(pci_read_config,	pci_read_config_method),
186 	DEVMETHOD(pci_write_config,	pci_write_config_method),
187 	DEVMETHOD(pci_enable_busmaster,	pci_enable_busmaster_method),
188 	DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
189 	DEVMETHOD(pci_enable_io,	pci_enable_io_method),
190 	DEVMETHOD(pci_disable_io,	pci_disable_io_method),
191 	DEVMETHOD(pci_get_vpd_ident,	pci_get_vpd_ident_method),
192 	DEVMETHOD(pci_get_vpd_readonly,	pci_get_vpd_readonly_method),
193 	DEVMETHOD(pci_get_powerstate,	pci_get_powerstate_method),
194 	DEVMETHOD(pci_set_powerstate,	pci_set_powerstate_method),
195 	DEVMETHOD(pci_assign_interrupt,	pci_assign_interrupt_method),
196 	DEVMETHOD(pci_find_cap,		pci_find_cap_method),
197 	DEVMETHOD(pci_find_next_cap,	pci_find_next_cap_method),
198 	DEVMETHOD(pci_find_extcap,	pci_find_extcap_method),
199 	DEVMETHOD(pci_find_next_extcap,	pci_find_next_extcap_method),
200 	DEVMETHOD(pci_find_htcap,	pci_find_htcap_method),
201 	DEVMETHOD(pci_find_next_htcap,	pci_find_next_htcap_method),
202 	DEVMETHOD(pci_alloc_msi,	pci_alloc_msi_method),
203 	DEVMETHOD(pci_alloc_msix,	pci_alloc_msix_method),
204 	DEVMETHOD(pci_enable_msi,	pci_enable_msi_method),
205 	DEVMETHOD(pci_enable_msix,	pci_enable_msix_method),
206 	DEVMETHOD(pci_disable_msi,	pci_disable_msi_method),
207 	DEVMETHOD(pci_remap_msix,	pci_remap_msix_method),
208 	DEVMETHOD(pci_release_msi,	pci_release_msi_method),
209 	DEVMETHOD(pci_msi_count,	pci_msi_count_method),
210 	DEVMETHOD(pci_msix_count,	pci_msix_count_method),
211 	DEVMETHOD(pci_msix_pba_bar,	pci_msix_pba_bar_method),
212 	DEVMETHOD(pci_msix_table_bar,	pci_msix_table_bar_method),
213 	DEVMETHOD(pci_get_id,		pci_get_id_method),
214 	DEVMETHOD(pci_alloc_devinfo,	pci_alloc_devinfo_method),
215 	DEVMETHOD(pci_child_added,	pci_child_added_method),
216 #ifdef PCI_IOV
217 	DEVMETHOD(pci_iov_attach,	pci_iov_attach_method),
218 	DEVMETHOD(pci_iov_detach,	pci_iov_detach_method),
219 	DEVMETHOD(pci_create_iov_child,	pci_create_iov_child_method),
220 #endif
221 
222 	DEVMETHOD_END
223 };
224 
225 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
226 
227 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_modevent, NULL, BUS_PASS_BUS);
228 MODULE_VERSION(pci, 1);
229 
230 static char	*pci_vendordata;
231 static size_t	pci_vendordata_size;
232 
233 struct pci_quirk {
234 	uint32_t devid;	/* Vendor/device of the card */
235 	int	type;
236 #define	PCI_QUIRK_MAP_REG	1 /* PCI map register in weird place */
237 #define	PCI_QUIRK_DISABLE_MSI	2 /* Neither MSI nor MSI-X work */
238 #define	PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI works */
239 #define	PCI_QUIRK_UNMAP_REG	4 /* Ignore PCI map register */
240 #define	PCI_QUIRK_DISABLE_MSIX	5 /* MSI-X doesn't work */
241 #define	PCI_QUIRK_MSI_INTX_BUG	6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define	PCI_QUIRK_REALLOC_BAR	7 /* Can't allocate memory at the default address */
243 	int	arg1;
244 	int	arg2;
245 };
246 
247 static const struct pci_quirk pci_quirks[] = {
248 	/* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
249 	{ 0x71138086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
250 	{ 0x719b8086, PCI_QUIRK_MAP_REG,	0x90,	 0 },
251 	/* As does the Serverworks OSB4 (the SMBus mapping register) */
252 	{ 0x02001166, PCI_QUIRK_MAP_REG,	0x90,	 0 },
253 
254 	/*
255 	 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
256 	 * or the CMIC-SL (AKA ServerWorks GC_LE).
257 	 */
258 	{ 0x00141166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
259 	{ 0x00171166, PCI_QUIRK_DISABLE_MSI,	0,	0 },
260 
261 	/*
262 	 * MSI doesn't work on earlier Intel chipsets including
263 	 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
264 	 */
265 	{ 0x25408086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
266 	{ 0x254c8086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
267 	{ 0x25508086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
268 	{ 0x25608086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
269 	{ 0x25708086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
270 	{ 0x25788086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
271 	{ 0x35808086, PCI_QUIRK_DISABLE_MSI,	0,	0 },
272 
273 	/*
274 	 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
275 	 * bridge.
276 	 */
277 	{ 0x74501022, PCI_QUIRK_DISABLE_MSI,	0,	0 },
278 
279 	/*
280 	 * Some virtualization environments emulate an older chipset
281 	 * but support MSI just fine.  QEMU uses the Intel 82440.
282 	 */
283 	{ 0x12378086, PCI_QUIRK_ENABLE_MSI_VM,	0,	0 },
284 
285 	/*
286 	 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
287 	 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
288 	 * It prevents us from attaching hpet(4) when the bit is unset.
289 	 * Note this quirk only affects SB600 revision A13 and earlier.
290 	 * For SB600 A21 and later, firmware must set the bit to hide it.
291 	 * For SB700 and later, it is unused and hardcoded to zero.
292 	 */
293 	{ 0x43851002, PCI_QUIRK_UNMAP_REG,	0x14,	0 },
294 
295 	/*
296 	 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
297 	 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
298 	 * of the command register is set.
299 	 */
300 	{ 0x10911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
301 	{ 0xE0911969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
302 	{ 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
303 	{ 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
304 	{ 0x10901969, PCI_QUIRK_MSI_INTX_BUG,	0,	0 },
305 
306 	/*
307 	 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
308 	 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
309 	 */
310 	{ 0x166814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714 */
311 	{ 0x166914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5714S */
312 	{ 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780 */
313 	{ 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5780S */
314 	{ 0x167814e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715 */
315 	{ 0x167914e4, PCI_QUIRK_MSI_INTX_BUG,	0,	0 }, /* BCM5715S */
316 
317 	/*
318 	 * HPE Gen 10 VGA has a memory range that can't be allocated in the
319 	 * expected place.
320 	 */
321 	{ 0x98741002, PCI_QUIRK_REALLOC_BAR,	0, 	0 },
322 	{ 0 }
323 };
324 
325 /* map register information */
326 #define	PCI_MAPMEM	0x01	/* memory map */
327 #define	PCI_MAPMEMP	0x02	/* prefetchable memory map */
328 #define	PCI_MAPPORT	0x04	/* port map */
329 
330 struct devlist pci_devq;
331 uint32_t pci_generation;
332 uint32_t pci_numdevs = 0;
333 static int pcie_chipset, pcix_chipset;
334 
335 /* sysctl vars */
336 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
337     "PCI bus tuning parameters");
338 
339 static int pci_enable_io_modes = 1;
340 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
341     &pci_enable_io_modes, 1,
342     "Enable I/O and memory bits in the config register.  Some BIOSes do not"
343     " enable these bits correctly.  We'd like to do this all the time, but"
344     " there are some peripherals that this causes problems with.");
345 
346 static int pci_do_realloc_bars = 1;
347 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
348     &pci_do_realloc_bars, 0,
349     "Attempt to allocate a new range for any BARs whose original "
350     "firmware-assigned ranges fail to allocate during the initial device scan.");
351 
352 static int pci_do_power_nodriver = 0;
353 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
354     &pci_do_power_nodriver, 0,
355     "Place a function into D3 state when no driver attaches to it.  0 means"
356     " disable.  1 means conservatively place devices into D3 state.  2 means"
357     " aggressively place devices into D3 state.  3 means put absolutely"
358     " everything in D3 state.");
359 
360 int pci_do_power_resume = 1;
361 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
362     &pci_do_power_resume, 1,
363   "Transition from D3 -> D0 on resume.");
364 
365 int pci_do_power_suspend = 1;
366 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
367     &pci_do_power_suspend, 1,
368   "Transition from D0 -> D3 on suspend.");
369 
370 static int pci_do_msi = 1;
371 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
372     "Enable support for MSI interrupts");
373 
374 static int pci_do_msix = 1;
375 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
376     "Enable support for MSI-X interrupts");
377 
378 static int pci_msix_rewrite_table = 0;
379 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
380     &pci_msix_rewrite_table, 0,
381     "Rewrite entire MSI-X table when updating MSI-X entries");
382 
383 static int pci_honor_msi_blacklist = 1;
384 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
385     &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
386 
387 #if defined(__i386__) || defined(__amd64__)
388 static int pci_usb_takeover = 1;
389 #else
390 static int pci_usb_takeover = 0;
391 #endif
392 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
393     &pci_usb_takeover, 1,
394     "Enable early takeover of USB controllers. Disable this if you depend on"
395     " BIOS emulation of USB devices, that is you use USB devices (like"
396     " keyboard or mouse) but do not load USB drivers");
397 
398 static int pci_clear_bars;
399 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
400     "Ignore firmware-assigned resources for BARs.");
401 
402 static int pci_clear_buses;
403 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
404     "Ignore firmware-assigned bus numbers.");
405 
406 static int pci_enable_ari = 1;
407 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
408     0, "Enable support for PCIe Alternative RID Interpretation");
409 
410 /*
411  * Some x86 firmware only enables PCIe hotplug if we claim to support aspm,
412  * however enabling it breaks some arm64 firmware as it powers off devices.
413  */
414 #if defined(__i386__) || defined(__amd64__)
415 int pci_enable_aspm = 1;
416 #else
417 int pci_enable_aspm = 0;
418 #endif
419 SYSCTL_INT(_hw_pci, OID_AUTO, enable_aspm, CTLFLAG_RDTUN, &pci_enable_aspm,
420     0, "Enable support for PCIe Active State Power Management");
421 
422 static int pci_clear_aer_on_attach = 0;
423 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
424     &pci_clear_aer_on_attach, 0,
425     "Clear port and device AER state on driver attach");
426 
427 static bool pci_enable_mps_tune = true;
428 SYSCTL_BOOL(_hw_pci, OID_AUTO, enable_mps_tune, CTLFLAG_RWTUN,
429     &pci_enable_mps_tune, 1,
430     "Enable tuning of MPS(maximum payload size)." );
431 
432 static int
pci_has_quirk(uint32_t devid,int quirk)433 pci_has_quirk(uint32_t devid, int quirk)
434 {
435 	const struct pci_quirk *q;
436 
437 	for (q = &pci_quirks[0]; q->devid; q++) {
438 		if (q->devid == devid && q->type == quirk)
439 			return (1);
440 	}
441 	return (0);
442 }
443 
444 /* Find a device_t by bus/slot/function in domain 0 */
445 
446 device_t
pci_find_bsf(uint8_t bus,uint8_t slot,uint8_t func)447 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
448 {
449 
450 	return (pci_find_dbsf(0, bus, slot, func));
451 }
452 
453 /* Find a device_t by domain/bus/slot/function */
454 
455 device_t
pci_find_dbsf(uint32_t domain,uint8_t bus,uint8_t slot,uint8_t func)456 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
457 {
458 	struct pci_devinfo *dinfo = NULL;
459 
460 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
461 		if ((dinfo->cfg.domain == domain) &&
462 		    (dinfo->cfg.bus == bus) &&
463 		    (dinfo->cfg.slot == slot) &&
464 		    (dinfo->cfg.func == func)) {
465 			break;
466 		}
467 	}
468 
469 	return (dinfo != NULL ? dinfo->cfg.dev : NULL);
470 }
471 
472 /* Find a device_t by vendor/device ID */
473 
474 device_t
pci_find_device(uint16_t vendor,uint16_t device)475 pci_find_device(uint16_t vendor, uint16_t device)
476 {
477 	struct pci_devinfo *dinfo;
478 
479 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
480 		if ((dinfo->cfg.vendor == vendor) &&
481 		    (dinfo->cfg.device == device)) {
482 			return (dinfo->cfg.dev);
483 		}
484 	}
485 
486 	return (NULL);
487 }
488 
489 device_t
pci_find_class(uint8_t class,uint8_t subclass)490 pci_find_class(uint8_t class, uint8_t subclass)
491 {
492 	struct pci_devinfo *dinfo;
493 
494 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
495 		if (dinfo->cfg.baseclass == class &&
496 		    dinfo->cfg.subclass == subclass) {
497 			return (dinfo->cfg.dev);
498 		}
499 	}
500 
501 	return (NULL);
502 }
503 
504 device_t
pci_find_class_from(uint8_t class,uint8_t subclass,device_t from)505 pci_find_class_from(uint8_t class, uint8_t subclass, device_t from)
506 {
507 	struct pci_devinfo *dinfo;
508 	bool found = false;
509 
510 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
511 		if (from != NULL && found == false) {
512 			if (from != dinfo->cfg.dev)
513 				continue;
514 			found = true;
515 			continue;
516 		}
517 		if (dinfo->cfg.baseclass == class &&
518 		    dinfo->cfg.subclass == subclass) {
519 			return (dinfo->cfg.dev);
520 		}
521 	}
522 
523 	return (NULL);
524 }
525 
526 static int
pci_printf(pcicfgregs * cfg,const char * fmt,...)527 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
528 {
529 	va_list ap;
530 	int retval;
531 
532 	retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
533 	    cfg->func);
534 	va_start(ap, fmt);
535 	retval += vprintf(fmt, ap);
536 	va_end(ap);
537 	return (retval);
538 }
539 
540 /* return base address of memory or port map */
541 
542 static pci_addr_t
pci_mapbase(uint64_t mapreg)543 pci_mapbase(uint64_t mapreg)
544 {
545 
546 	if (PCI_BAR_MEM(mapreg))
547 		return (mapreg & PCIM_BAR_MEM_BASE);
548 	else
549 		return (mapreg & PCIM_BAR_IO_BASE);
550 }
551 
552 /* return map type of memory or port map */
553 
554 static const char *
pci_maptype(uint64_t mapreg)555 pci_maptype(uint64_t mapreg)
556 {
557 
558 	if (PCI_BAR_IO(mapreg))
559 		return ("I/O Port");
560 	if (mapreg & PCIM_BAR_MEM_PREFETCH)
561 		return ("Prefetchable Memory");
562 	return ("Memory");
563 }
564 
565 /* return log2 of map size decoded for memory or port map */
566 
567 int
pci_mapsize(uint64_t testval)568 pci_mapsize(uint64_t testval)
569 {
570 	int ln2size;
571 
572 	testval = pci_mapbase(testval);
573 	ln2size = 0;
574 	if (testval != 0) {
575 		while ((testval & 1) == 0)
576 		{
577 			ln2size++;
578 			testval >>= 1;
579 		}
580 	}
581 	return (ln2size);
582 }
583 
584 /* return base address of device ROM */
585 
586 static pci_addr_t
pci_rombase(uint64_t mapreg)587 pci_rombase(uint64_t mapreg)
588 {
589 
590 	return (mapreg & PCIM_BIOS_ADDR_MASK);
591 }
592 
593 /* return log2 of map size decided for device ROM */
594 
595 static int
pci_romsize(uint64_t testval)596 pci_romsize(uint64_t testval)
597 {
598 	int ln2size;
599 
600 	testval = pci_rombase(testval);
601 	ln2size = 0;
602 	if (testval != 0) {
603 		while ((testval & 1) == 0)
604 		{
605 			ln2size++;
606 			testval >>= 1;
607 		}
608 	}
609 	return (ln2size);
610 }
611 
612 /* return log2 of address range supported by map register */
613 
614 static int
pci_maprange(uint64_t mapreg)615 pci_maprange(uint64_t mapreg)
616 {
617 	int ln2range = 0;
618 
619 	if (PCI_BAR_IO(mapreg))
620 		ln2range = 32;
621 	else
622 		switch (mapreg & PCIM_BAR_MEM_TYPE) {
623 		case PCIM_BAR_MEM_32:
624 			ln2range = 32;
625 			break;
626 		case PCIM_BAR_MEM_1MB:
627 			ln2range = 20;
628 			break;
629 		case PCIM_BAR_MEM_64:
630 			ln2range = 64;
631 			break;
632 		}
633 	return (ln2range);
634 }
635 
636 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
637 
638 static void
pci_fixancient(pcicfgregs * cfg)639 pci_fixancient(pcicfgregs *cfg)
640 {
641 	if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
642 		return;
643 
644 	/* PCI to PCI bridges use header type 1 */
645 	if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
646 		cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
647 }
648 
649 /* extract header type specific config data */
650 
651 static void
pci_hdrtypedata(device_t pcib,int b,int s,int f,pcicfgregs * cfg)652 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
653 {
654 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
655 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
656 	case PCIM_HDRTYPE_NORMAL:
657 		cfg->subvendor      = REG(PCIR_SUBVEND_0, 2);
658 		cfg->subdevice      = REG(PCIR_SUBDEV_0, 2);
659 		cfg->mingnt         = REG(PCIR_MINGNT, 1);
660 		cfg->maxlat         = REG(PCIR_MAXLAT, 1);
661 		cfg->nummaps	    = PCI_MAXMAPS_0;
662 		break;
663 	case PCIM_HDRTYPE_BRIDGE:
664 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
665 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
666 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
667 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
668 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
669 		cfg->nummaps	    = PCI_MAXMAPS_1;
670 		break;
671 	case PCIM_HDRTYPE_CARDBUS:
672 		cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
673 		cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
674 		cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
675 		cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
676 		cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
677 		cfg->subvendor      = REG(PCIR_SUBVEND_2, 2);
678 		cfg->subdevice      = REG(PCIR_SUBDEV_2, 2);
679 		cfg->nummaps	    = PCI_MAXMAPS_2;
680 		break;
681 	}
682 #undef REG
683 }
684 
685 /* read configuration header into pcicfgregs structure */
686 struct pci_devinfo *
pci_read_device(device_t pcib,device_t bus,int d,int b,int s,int f)687 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
688 {
689 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, b, s, f, n, w)
690 	uint16_t vid, did;
691 
692 	vid = REG(PCIR_VENDOR, 2);
693 	if (vid == PCIV_INVALID)
694 		return (NULL);
695 
696 	did = REG(PCIR_DEVICE, 2);
697 
698 	return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
699 }
700 
701 struct pci_devinfo *
pci_alloc_devinfo_method(device_t dev)702 pci_alloc_devinfo_method(device_t dev)
703 {
704 
705 	return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
706 	    M_WAITOK | M_ZERO));
707 }
708 
709 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)710 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
711     uint16_t vid, uint16_t did)
712 {
713 	struct pci_devinfo *devlist_entry;
714 	pcicfgregs *cfg;
715 
716 	devlist_entry = PCI_ALLOC_DEVINFO(bus);
717 
718 	cfg = &devlist_entry->cfg;
719 
720 	cfg->domain		= d;
721 	cfg->bus		= b;
722 	cfg->slot		= s;
723 	cfg->func		= f;
724 	cfg->vendor		= vid;
725 	cfg->device		= did;
726 	cfg->cmdreg		= REG(PCIR_COMMAND, 2);
727 	cfg->statreg		= REG(PCIR_STATUS, 2);
728 	cfg->baseclass		= REG(PCIR_CLASS, 1);
729 	cfg->subclass		= REG(PCIR_SUBCLASS, 1);
730 	cfg->progif		= REG(PCIR_PROGIF, 1);
731 	cfg->revid		= REG(PCIR_REVID, 1);
732 	cfg->hdrtype		= REG(PCIR_HDRTYPE, 1);
733 	cfg->cachelnsz		= REG(PCIR_CACHELNSZ, 1);
734 	cfg->lattimer		= REG(PCIR_LATTIMER, 1);
735 	cfg->intpin		= REG(PCIR_INTPIN, 1);
736 	cfg->intline		= REG(PCIR_INTLINE, 1);
737 
738 	cfg->mfdev		= (cfg->hdrtype & PCIM_MFDEV) != 0;
739 	cfg->hdrtype		&= ~PCIM_MFDEV;
740 	STAILQ_INIT(&cfg->maps);
741 
742 	cfg->iov		= NULL;
743 
744 	pci_fixancient(cfg);
745 	pci_hdrtypedata(pcib, b, s, f, cfg);
746 
747 	if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
748 		pci_read_cap(pcib, cfg);
749 
750 	STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
751 
752 	devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
753 	devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
754 	devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
755 	devlist_entry->conf.pc_sel.pc_func = cfg->func;
756 	devlist_entry->conf.pc_hdr = cfg->hdrtype;
757 
758 	devlist_entry->conf.pc_subvendor = cfg->subvendor;
759 	devlist_entry->conf.pc_subdevice = cfg->subdevice;
760 	devlist_entry->conf.pc_vendor = cfg->vendor;
761 	devlist_entry->conf.pc_device = cfg->device;
762 
763 	devlist_entry->conf.pc_class = cfg->baseclass;
764 	devlist_entry->conf.pc_subclass = cfg->subclass;
765 	devlist_entry->conf.pc_progif = cfg->progif;
766 	devlist_entry->conf.pc_revid = cfg->revid;
767 
768 	pci_numdevs++;
769 	pci_generation++;
770 
771 	return (devlist_entry);
772 }
773 #undef REG
774 
775 static void
pci_ea_fill_info(device_t pcib,pcicfgregs * cfg)776 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
777 {
778 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
779     cfg->ea.ea_location + (n), w)
780 	int num_ent;
781 	int ptr;
782 	int a, b;
783 	uint32_t val;
784 	int ent_size;
785 	uint32_t dw[4];
786 	uint64_t base, max_offset;
787 	struct pci_ea_entry *eae;
788 
789 	if (cfg->ea.ea_location == 0)
790 		return;
791 
792 	STAILQ_INIT(&cfg->ea.ea_entries);
793 
794 	/* Determine the number of entries */
795 	num_ent = REG(PCIR_EA_NUM_ENT, 2);
796 	num_ent &= PCIM_EA_NUM_ENT_MASK;
797 
798 	/* Find the first entry to care of */
799 	ptr = PCIR_EA_FIRST_ENT;
800 
801 	/* Skip DWORD 2 for type 1 functions */
802 	if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
803 		ptr += 4;
804 
805 	for (a = 0; a < num_ent; a++) {
806 		eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
807 		eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
808 
809 		/* Read a number of dwords in the entry */
810 		val = REG(ptr, 4);
811 		ptr += 4;
812 		ent_size = (val & PCIM_EA_ES);
813 
814 		for (b = 0; b < ent_size; b++) {
815 			dw[b] = REG(ptr, 4);
816 			ptr += 4;
817 		}
818 
819 		eae->eae_flags = val;
820 		eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
821 
822 		base = dw[0] & PCIM_EA_FIELD_MASK;
823 		max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
824 		b = 2;
825 		if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
826 			base |= (uint64_t)dw[b] << 32UL;
827 			b++;
828 		}
829 		if (((dw[1] & PCIM_EA_IS_64) != 0)
830 		    && (b < ent_size)) {
831 			max_offset |= (uint64_t)dw[b] << 32UL;
832 			b++;
833 		}
834 
835 		eae->eae_base = base;
836 		eae->eae_max_offset = max_offset;
837 
838 		STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
839 
840 		if (bootverbose) {
841 			printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
842 			    cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
843 			    (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
844 		}
845 	}
846 }
847 #undef REG
848 
849 static void
pci_read_cap(device_t pcib,pcicfgregs * cfg)850 pci_read_cap(device_t pcib, pcicfgregs *cfg)
851 {
852 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
853 #define	WREG(n, v, w)	PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
854 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
855 	uint64_t addr;
856 #endif
857 	uint32_t val;
858 	int	ptr, nextptr, ptrptr;
859 
860 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
861 	case PCIM_HDRTYPE_NORMAL:
862 	case PCIM_HDRTYPE_BRIDGE:
863 		ptrptr = PCIR_CAP_PTR;
864 		break;
865 	case PCIM_HDRTYPE_CARDBUS:
866 		ptrptr = PCIR_CAP_PTR_2;	/* cardbus capabilities ptr */
867 		break;
868 	default:
869 		return;		/* no extended capabilities support */
870 	}
871 	nextptr = REG(ptrptr, 1);	/* sanity check? */
872 
873 	/*
874 	 * Read capability entries.
875 	 */
876 	while (nextptr != 0) {
877 		/* Sanity check */
878 		if (nextptr > 255) {
879 			printf("illegal PCI extended capability offset %d\n",
880 			    nextptr);
881 			return;
882 		}
883 		/* Find the next entry */
884 		ptr = nextptr;
885 		nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
886 
887 		/* Process this entry */
888 		switch (REG(ptr + PCICAP_ID, 1)) {
889 		case PCIY_PMG:		/* PCI power management */
890 			if (cfg->pp.pp_cap == 0) {
891 				cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
892 				cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
893 				cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
894 				if ((nextptr - ptr) > PCIR_POWER_DATA)
895 					cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
896 			}
897 			break;
898 		case PCIY_HT:		/* HyperTransport */
899 			/* Determine HT-specific capability type. */
900 			val = REG(ptr + PCIR_HT_COMMAND, 2);
901 
902 			if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
903 				cfg->ht.ht_slave = ptr;
904 
905 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
906 			switch (val & PCIM_HTCMD_CAP_MASK) {
907 			case PCIM_HTCAP_MSI_MAPPING:
908 				if (!(val & PCIM_HTCMD_MSI_FIXED)) {
909 					/* Sanity check the mapping window. */
910 					addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
911 					    4);
912 					addr <<= 32;
913 					addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
914 					    4);
915 					if (addr != MSI_INTEL_ADDR_BASE)
916 						device_printf(pcib,
917 	    "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
918 						    cfg->domain, cfg->bus,
919 						    cfg->slot, cfg->func,
920 						    (long long)addr);
921 				} else
922 					addr = MSI_INTEL_ADDR_BASE;
923 
924 				cfg->ht.ht_msimap = ptr;
925 				cfg->ht.ht_msictrl = val;
926 				cfg->ht.ht_msiaddr = addr;
927 				break;
928 			}
929 #endif
930 			break;
931 		case PCIY_MSI:		/* PCI MSI */
932 			cfg->msi.msi_location = ptr;
933 			cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
934 			cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
935 						     PCIM_MSICTRL_MMC_MASK)>>1);
936 			break;
937 		case PCIY_MSIX:		/* PCI MSI-X */
938 			cfg->msix.msix_location = ptr;
939 			cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
940 			cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
941 			    PCIM_MSIXCTRL_TABLE_SIZE) + 1;
942 			val = REG(ptr + PCIR_MSIX_TABLE, 4);
943 			cfg->msix.msix_table_bar = PCIR_BAR(val &
944 			    PCIM_MSIX_BIR_MASK);
945 			cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
946 			val = REG(ptr + PCIR_MSIX_PBA, 4);
947 			cfg->msix.msix_pba_bar = PCIR_BAR(val &
948 			    PCIM_MSIX_BIR_MASK);
949 			cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
950 			break;
951 		case PCIY_VPD:		/* PCI Vital Product Data */
952 			cfg->vpd.vpd_reg = ptr;
953 			break;
954 		case PCIY_SUBVENDOR:
955 			/* Should always be true. */
956 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
957 			    PCIM_HDRTYPE_BRIDGE) {
958 				val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
959 				cfg->subvendor = val & 0xffff;
960 				cfg->subdevice = val >> 16;
961 			}
962 			break;
963 		case PCIY_PCIX:		/* PCI-X */
964 			/*
965 			 * Assume we have a PCI-X chipset if we have
966 			 * at least one PCI-PCI bridge with a PCI-X
967 			 * capability.  Note that some systems with
968 			 * PCI-express or HT chipsets might match on
969 			 * this check as well.
970 			 */
971 			if ((cfg->hdrtype & PCIM_HDRTYPE) ==
972 			    PCIM_HDRTYPE_BRIDGE)
973 				pcix_chipset = 1;
974 			cfg->pcix.pcix_location = ptr;
975 			break;
976 		case PCIY_EXPRESS:	/* PCI-express */
977 			/*
978 			 * Assume we have a PCI-express chipset if we have
979 			 * at least one PCI-express device.
980 			 */
981 			pcie_chipset = 1;
982 			cfg->pcie.pcie_location = ptr;
983 			val = REG(ptr + PCIER_FLAGS, 2);
984 			cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
985 			break;
986 		case PCIY_EA:		/* Enhanced Allocation */
987 			cfg->ea.ea_location = ptr;
988 			pci_ea_fill_info(pcib, cfg);
989 			break;
990 		default:
991 			break;
992 		}
993 	}
994 
995 #if defined(__powerpc__)
996 	/*
997 	 * Enable the MSI mapping window for all HyperTransport
998 	 * slaves.  PCI-PCI bridges have their windows enabled via
999 	 * PCIB_MAP_MSI().
1000 	 */
1001 	if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
1002 	    !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
1003 		device_printf(pcib,
1004 	    "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
1005 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
1006 		 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
1007 		 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
1008 		     2);
1009 	}
1010 #endif
1011 /* REG and WREG use carry through to next functions */
1012 }
1013 
1014 /*
1015  * PCI Vital Product Data
1016  */
1017 
1018 #define	PCI_VPD_TIMEOUT		1000000
1019 
1020 static int
pci_read_vpd_reg(device_t pcib,pcicfgregs * cfg,int reg,uint32_t * data)1021 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
1022 {
1023 	int count = PCI_VPD_TIMEOUT;
1024 
1025 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1026 
1027 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
1028 
1029 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1030 		if (--count < 0)
1031 			return (ENXIO);
1032 		DELAY(1);	/* limit looping */
1033 	}
1034 	*data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1035 
1036 	return (0);
1037 }
1038 
1039 #if 0
1040 static int
1041 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1042 {
1043 	int count = PCI_VPD_TIMEOUT;
1044 
1045 	KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1046 
1047 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1048 	WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1049 	while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1050 		if (--count < 0)
1051 			return (ENXIO);
1052 		DELAY(1);	/* limit looping */
1053 	}
1054 
1055 	return (0);
1056 }
1057 #endif
1058 
1059 #undef PCI_VPD_TIMEOUT
1060 
1061 struct vpd_readstate {
1062 	device_t	pcib;
1063 	pcicfgregs	*cfg;
1064 	uint32_t	val;
1065 	int		bytesinval;
1066 	int		off;
1067 	uint8_t		cksum;
1068 };
1069 
1070 /* return 0 and one byte in *data if no read error, -1 else */
1071 static int
vpd_nextbyte(struct vpd_readstate * vrs,uint8_t * data)1072 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1073 {
1074 	uint32_t reg;
1075 	uint8_t byte;
1076 
1077 	if (vrs->bytesinval == 0) {
1078 		if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, &reg))
1079 			return (-1);
1080 		vrs->val = le32toh(reg);
1081 		vrs->off += 4;
1082 		byte = vrs->val & 0xff;
1083 		vrs->bytesinval = 3;
1084 	} else {
1085 		vrs->val = vrs->val >> 8;
1086 		byte = vrs->val & 0xff;
1087 		vrs->bytesinval--;
1088 	}
1089 
1090 	vrs->cksum += byte;
1091 	*data = byte;
1092 	return (0);
1093 }
1094 
1095 /* return 0 on match, -1 and "unget" byte on no match */
1096 static int
vpd_expectbyte(struct vpd_readstate * vrs,uint8_t expected)1097 vpd_expectbyte(struct vpd_readstate *vrs, uint8_t expected)
1098 {
1099 	uint8_t data;
1100 
1101 	if (vpd_nextbyte(vrs, &data) != 0)
1102 		return (-1);
1103 
1104 	if (data == expected)
1105 		return (0);
1106 
1107 	vrs->cksum -= data;
1108 	vrs->val = (vrs->val << 8) + data;
1109 	vrs->bytesinval++;
1110 	return (-1);
1111 }
1112 
1113 /* return size if tag matches, -1 on no match, -2 on read error */
1114 static int
vpd_read_tag_size(struct vpd_readstate * vrs,uint8_t vpd_tag)1115 vpd_read_tag_size(struct vpd_readstate *vrs, uint8_t vpd_tag)
1116 {
1117 	uint8_t byte1, byte2;
1118 
1119 	if (vpd_expectbyte(vrs, vpd_tag) != 0)
1120 		return (-1);
1121 
1122 	if ((vpd_tag & 0x80) == 0)
1123 		return (vpd_tag & 0x07);
1124 
1125 	if (vpd_nextbyte(vrs, &byte1) != 0)
1126 		return (-2);
1127 	if (vpd_nextbyte(vrs, &byte2) != 0)
1128 		return (-2);
1129 
1130 	return ((byte2 << 8) + byte1);
1131 }
1132 
1133 /* (re)allocate buffer in multiples of 8 elements */
1134 static void*
alloc_buffer(void * buffer,size_t element_size,int needed)1135 alloc_buffer(void* buffer, size_t element_size, int needed)
1136 {
1137 	int alloc, new_alloc;
1138 
1139 	alloc = roundup2(needed, 8);
1140 	new_alloc = roundup2(needed + 1, 8);
1141 	if (alloc != new_alloc) {
1142 		buffer = reallocf(buffer,
1143 		    new_alloc * element_size, M_DEVBUF, M_WAITOK | M_ZERO);
1144 	}
1145 
1146 	return (buffer);
1147 }
1148 
1149 /* read VPD keyword and return element size, return -1 on read error */
1150 static int
vpd_read_elem_head(struct vpd_readstate * vrs,char keyword[2])1151 vpd_read_elem_head(struct vpd_readstate *vrs, char keyword[2])
1152 {
1153 	uint8_t data;
1154 
1155 	if (vpd_nextbyte(vrs, &keyword[0]) != 0)
1156 		return (-1);
1157 	if (vpd_nextbyte(vrs, &keyword[1]) != 0)
1158 		return (-1);
1159 	if (vpd_nextbyte(vrs, &data) != 0)
1160 		return (-1);
1161 
1162 	return (data);
1163 }
1164 
1165 /* read VPD data element of given size into allocated buffer */
1166 static char *
vpd_read_value(struct vpd_readstate * vrs,int size)1167 vpd_read_value(struct vpd_readstate *vrs, int size)
1168 {
1169 	int i;
1170 	char char1;
1171 	char *value;
1172 
1173 	value = malloc(size + 1, M_DEVBUF, M_WAITOK);
1174 	for (i = 0; i < size; i++) {
1175 		if (vpd_nextbyte(vrs, &char1) != 0) {
1176 			free(value, M_DEVBUF);
1177 			return (NULL);
1178 		}
1179 		value[i] = char1;
1180 	}
1181 	value[size] = '\0';
1182 
1183 	return (value);
1184 }
1185 
1186 /* read VPD into *keyword and *value, return length of data element */
1187 static int
vpd_read_elem_data(struct vpd_readstate * vrs,char keyword[2],char ** value,int maxlen)1188 vpd_read_elem_data(struct vpd_readstate *vrs, char keyword[2], char **value, int maxlen)
1189 {
1190 	int len;
1191 
1192 	len = vpd_read_elem_head(vrs, keyword);
1193 	if (len < 0 || len > maxlen)
1194 		return (-1);
1195 	*value = vpd_read_value(vrs, len);
1196 
1197 	return (len);
1198 }
1199 
1200 /* subtract all data following first byte from checksum of RV element */
1201 static void
vpd_fixup_cksum(struct vpd_readstate * vrs,char * rvstring,int len)1202 vpd_fixup_cksum(struct vpd_readstate *vrs, char *rvstring, int len)
1203 {
1204 	int i;
1205 	uint8_t fixup;
1206 
1207 	fixup = 0;
1208 	for (i = 1; i < len; i++)
1209 		fixup += rvstring[i];
1210 	vrs->cksum -= fixup;
1211 }
1212 
1213 /* fetch one read-only element and return size of heading + data */
1214 static int
next_vpd_ro_elem(struct vpd_readstate * vrs,int maxsize)1215 next_vpd_ro_elem(struct vpd_readstate *vrs, int maxsize)
1216 {
1217 	struct pcicfg_vpd *vpd;
1218 	pcicfgregs *cfg;
1219 	struct vpd_readonly *vpd_ros;
1220 	int len;
1221 
1222 	cfg = vrs->cfg;
1223 	vpd = &cfg->vpd;
1224 
1225 	if (maxsize < 3)
1226 		return (-1);
1227 	vpd->vpd_ros = alloc_buffer(vpd->vpd_ros, sizeof(*vpd->vpd_ros), vpd->vpd_rocnt);
1228 	vpd_ros = &vpd->vpd_ros[vpd->vpd_rocnt];
1229 	maxsize -= 3;
1230 	len = vpd_read_elem_data(vrs, vpd_ros->keyword, &vpd_ros->value, maxsize);
1231 	if (vpd_ros->value == NULL)
1232 		return (-1);
1233 	vpd_ros->len = len;
1234 	if (vpd_ros->keyword[0] == 'R' && vpd_ros->keyword[1] == 'V') {
1235 		vpd_fixup_cksum(vrs, vpd_ros->value, len);
1236 		if (vrs->cksum != 0) {
1237 			pci_printf(cfg,
1238 			    "invalid VPD checksum %#hhx\n", vrs->cksum);
1239 			return (-1);
1240 		}
1241 	}
1242 	vpd->vpd_rocnt++;
1243 
1244 	return (len + 3);
1245 }
1246 
1247 /* fetch one writable element and return size of heading + data */
1248 static int
next_vpd_rw_elem(struct vpd_readstate * vrs,int maxsize)1249 next_vpd_rw_elem(struct vpd_readstate *vrs, int maxsize)
1250 {
1251 	struct pcicfg_vpd *vpd;
1252 	pcicfgregs *cfg;
1253 	struct vpd_write *vpd_w;
1254 	int len;
1255 
1256 	cfg = vrs->cfg;
1257 	vpd = &cfg->vpd;
1258 
1259 	if (maxsize < 3)
1260 		return (-1);
1261 	vpd->vpd_w = alloc_buffer(vpd->vpd_w, sizeof(*vpd->vpd_w), vpd->vpd_wcnt);
1262 	if (vpd->vpd_w == NULL) {
1263 		pci_printf(cfg, "out of memory");
1264 		return (-1);
1265 	}
1266 	vpd_w = &vpd->vpd_w[vpd->vpd_wcnt];
1267 	maxsize -= 3;
1268 	vpd_w->start = vrs->off + 3 - vrs->bytesinval;
1269 	len = vpd_read_elem_data(vrs, vpd_w->keyword, &vpd_w->value, maxsize);
1270 	if (vpd_w->value == NULL)
1271 		return (-1);
1272 	vpd_w->len = len;
1273 	vpd->vpd_wcnt++;
1274 
1275 	return (len + 3);
1276 }
1277 
1278 /* free all memory allocated for VPD data */
1279 static void
vpd_free(struct pcicfg_vpd * vpd)1280 vpd_free(struct pcicfg_vpd *vpd)
1281 {
1282 	int i;
1283 
1284 	free(vpd->vpd_ident, M_DEVBUF);
1285 	for (i = 0; i < vpd->vpd_rocnt; i++)
1286 		free(vpd->vpd_ros[i].value, M_DEVBUF);
1287 	free(vpd->vpd_ros, M_DEVBUF);
1288 	vpd->vpd_rocnt = 0;
1289 	for (i = 0; i < vpd->vpd_wcnt; i++)
1290 		free(vpd->vpd_w[i].value, M_DEVBUF);
1291 	free(vpd->vpd_w, M_DEVBUF);
1292 	vpd->vpd_wcnt = 0;
1293 }
1294 
1295 #define VPD_TAG_END	((0x0f << 3) | 0)	/* small tag, len == 0 */
1296 #define VPD_TAG_IDENT	(0x02 | 0x80)		/* large tag */
1297 #define VPD_TAG_RO	(0x10 | 0x80)		/* large tag */
1298 #define VPD_TAG_RW	(0x11 | 0x80)		/* large tag */
1299 
1300 static int
pci_parse_vpd(device_t pcib,pcicfgregs * cfg)1301 pci_parse_vpd(device_t pcib, pcicfgregs *cfg)
1302 {
1303 	struct vpd_readstate vrs;
1304 	int cksumvalid;
1305 	int size, elem_size;
1306 
1307 	/* init vpd reader */
1308 	vrs.bytesinval = 0;
1309 	vrs.off = 0;
1310 	vrs.pcib = pcib;
1311 	vrs.cfg = cfg;
1312 	vrs.cksum = 0;
1313 
1314 	/* read VPD ident element - mandatory */
1315 	size = vpd_read_tag_size(&vrs, VPD_TAG_IDENT);
1316 	if (size <= 0) {
1317 		pci_printf(cfg, "no VPD ident found\n");
1318 		return (0);
1319 	}
1320 	cfg->vpd.vpd_ident = vpd_read_value(&vrs, size);
1321 	if (cfg->vpd.vpd_ident == NULL) {
1322 		pci_printf(cfg, "error accessing VPD ident data\n");
1323 		return (0);
1324 	}
1325 
1326 	/* read VPD RO elements - mandatory */
1327 	size = vpd_read_tag_size(&vrs, VPD_TAG_RO);
1328 	if (size <= 0) {
1329 		pci_printf(cfg, "no read-only VPD data found\n");
1330 		return (0);
1331 	}
1332 	while (size > 0) {
1333 		elem_size = next_vpd_ro_elem(&vrs, size);
1334 		if (elem_size < 0) {
1335 			pci_printf(cfg, "error accessing read-only VPD data\n");
1336 			return (-1);
1337 		}
1338 		size -= elem_size;
1339 	}
1340 	cksumvalid = (vrs.cksum == 0);
1341 	if (!cksumvalid)
1342 		return (-1);
1343 
1344 	/* read VPD RW elements - optional */
1345 	size = vpd_read_tag_size(&vrs, VPD_TAG_RW);
1346 	if (size == -2)
1347 		return (-1);
1348 	while (size > 0) {
1349 		elem_size = next_vpd_rw_elem(&vrs, size);
1350 		if (elem_size < 0) {
1351 			pci_printf(cfg, "error accessing writeable VPD data\n");
1352 			return (-1);
1353 		}
1354 		size -= elem_size;
1355 	}
1356 
1357 	/* read empty END tag - mandatory */
1358 	size = vpd_read_tag_size(&vrs, VPD_TAG_END);
1359 	if (size != 0) {
1360 		pci_printf(cfg, "No valid VPD end tag found\n");
1361 	}
1362 	return (0);
1363 }
1364 
1365 static void
pci_read_vpd(device_t pcib,pcicfgregs * cfg)1366 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1367 {
1368 	int status;
1369 
1370 	status = pci_parse_vpd(pcib, cfg);
1371 	if (status < 0)
1372 		vpd_free(&cfg->vpd);
1373 	cfg->vpd.vpd_cached = 1;
1374 #undef REG
1375 #undef WREG
1376 }
1377 
1378 int
pci_get_vpd_ident_method(device_t dev,device_t child,const char ** identptr)1379 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1380 {
1381 	struct pci_devinfo *dinfo = device_get_ivars(child);
1382 	pcicfgregs *cfg = &dinfo->cfg;
1383 
1384 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1385 		pci_read_vpd(device_get_parent(dev), cfg);
1386 
1387 	*identptr = cfg->vpd.vpd_ident;
1388 
1389 	if (*identptr == NULL)
1390 		return (ENXIO);
1391 
1392 	return (0);
1393 }
1394 
1395 int
pci_get_vpd_readonly_method(device_t dev,device_t child,const char * kw,const char ** vptr)1396 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1397 	const char **vptr)
1398 {
1399 	struct pci_devinfo *dinfo = device_get_ivars(child);
1400 	pcicfgregs *cfg = &dinfo->cfg;
1401 	int i;
1402 
1403 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1404 		pci_read_vpd(device_get_parent(dev), cfg);
1405 
1406 	for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1407 		if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1408 		    sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1409 			*vptr = cfg->vpd.vpd_ros[i].value;
1410 			return (0);
1411 		}
1412 
1413 	*vptr = NULL;
1414 	return (ENXIO);
1415 }
1416 
1417 struct pcicfg_vpd *
pci_fetch_vpd_list(device_t dev)1418 pci_fetch_vpd_list(device_t dev)
1419 {
1420 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1421 	pcicfgregs *cfg = &dinfo->cfg;
1422 
1423 	if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1424 		pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1425 	return (&cfg->vpd);
1426 }
1427 
1428 /*
1429  * Find the requested HyperTransport capability and return the offset
1430  * in configuration space via the pointer provided.  The function
1431  * returns 0 on success and an error code otherwise.
1432  */
1433 int
pci_find_htcap_method(device_t dev,device_t child,int capability,int * capreg)1434 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1435 {
1436 	int ptr, error;
1437 	uint16_t val;
1438 
1439 	error = pci_find_cap(child, PCIY_HT, &ptr);
1440 	if (error)
1441 		return (error);
1442 
1443 	/*
1444 	 * Traverse the capabilities list checking each HT capability
1445 	 * to see if it matches the requested HT capability.
1446 	 */
1447 	for (;;) {
1448 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1449 		if (capability == PCIM_HTCAP_SLAVE ||
1450 		    capability == PCIM_HTCAP_HOST)
1451 			val &= 0xe000;
1452 		else
1453 			val &= PCIM_HTCMD_CAP_MASK;
1454 		if (val == capability) {
1455 			if (capreg != NULL)
1456 				*capreg = ptr;
1457 			return (0);
1458 		}
1459 
1460 		/* Skip to the next HT capability. */
1461 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1462 			break;
1463 	}
1464 
1465 	return (ENOENT);
1466 }
1467 
1468 /*
1469  * Find the next requested HyperTransport capability after start and return
1470  * the offset in configuration space via the pointer provided.  The function
1471  * returns 0 on success and an error code otherwise.
1472  */
1473 int
pci_find_next_htcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1474 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1475     int start, int *capreg)
1476 {
1477 	int ptr;
1478 	uint16_t val;
1479 
1480 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1481 	    ("start capability is not HyperTransport capability"));
1482 	ptr = start;
1483 
1484 	/*
1485 	 * Traverse the capabilities list checking each HT capability
1486 	 * to see if it matches the requested HT capability.
1487 	 */
1488 	for (;;) {
1489 		/* Skip to the next HT capability. */
1490 		if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1491 			break;
1492 
1493 		val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1494 		if (capability == PCIM_HTCAP_SLAVE ||
1495 		    capability == PCIM_HTCAP_HOST)
1496 			val &= 0xe000;
1497 		else
1498 			val &= PCIM_HTCMD_CAP_MASK;
1499 		if (val == capability) {
1500 			if (capreg != NULL)
1501 				*capreg = ptr;
1502 			return (0);
1503 		}
1504 	}
1505 
1506 	return (ENOENT);
1507 }
1508 
1509 /*
1510  * Find the requested capability and return the offset in
1511  * configuration space via the pointer provided.  The function returns
1512  * 0 on success and an error code otherwise.
1513  */
1514 int
pci_find_cap_method(device_t dev,device_t child,int capability,int * capreg)1515 pci_find_cap_method(device_t dev, device_t child, int capability,
1516     int *capreg)
1517 {
1518 	struct pci_devinfo *dinfo = device_get_ivars(child);
1519 	pcicfgregs *cfg = &dinfo->cfg;
1520 	uint32_t status;
1521 	uint8_t ptr;
1522 	int cnt;
1523 
1524 	/*
1525 	 * Check the CAP_LIST bit of the PCI status register first.
1526 	 */
1527 	status = pci_read_config(child, PCIR_STATUS, 2);
1528 	if (!(status & PCIM_STATUS_CAPPRESENT))
1529 		return (ENXIO);
1530 
1531 	/*
1532 	 * Determine the start pointer of the capabilities list.
1533 	 */
1534 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
1535 	case PCIM_HDRTYPE_NORMAL:
1536 	case PCIM_HDRTYPE_BRIDGE:
1537 		ptr = PCIR_CAP_PTR;
1538 		break;
1539 	case PCIM_HDRTYPE_CARDBUS:
1540 		ptr = PCIR_CAP_PTR_2;
1541 		break;
1542 	default:
1543 		/* XXX: panic? */
1544 		return (ENXIO);		/* no extended capabilities support */
1545 	}
1546 	ptr = pci_read_config(child, ptr, 1);
1547 
1548 	/*
1549 	 * Traverse the capabilities list.  Limit by total theoretical
1550 	 * maximum number of caps: capability needs at least id and
1551 	 * next registers, and any type X header cannot contain caps.
1552 	 */
1553 	for (cnt = 0; ptr != 0 && cnt < (PCIE_REGMAX - 0x40) / 2; cnt++) {
1554 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1555 			if (capreg != NULL)
1556 				*capreg = ptr;
1557 			return (0);
1558 		}
1559 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1560 	}
1561 
1562 	return (ENOENT);
1563 }
1564 
1565 /*
1566  * Find the next requested capability after start and return the offset in
1567  * configuration space via the pointer provided.  The function returns
1568  * 0 on success and an error code otherwise.
1569  */
1570 int
pci_find_next_cap_method(device_t dev,device_t child,int capability,int start,int * capreg)1571 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1572     int start, int *capreg)
1573 {
1574 	uint8_t ptr;
1575 
1576 	KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1577 	    ("start capability is not expected capability"));
1578 
1579 	ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1580 	while (ptr != 0) {
1581 		if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1582 			if (capreg != NULL)
1583 				*capreg = ptr;
1584 			return (0);
1585 		}
1586 		ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1587 	}
1588 
1589 	return (ENOENT);
1590 }
1591 
1592 /*
1593  * Find the requested extended capability and return the offset in
1594  * configuration space via the pointer provided.  The function returns
1595  * 0 on success and an error code otherwise.
1596  */
1597 int
pci_find_extcap_method(device_t dev,device_t child,int capability,int * capreg)1598 pci_find_extcap_method(device_t dev, device_t child, int capability,
1599     int *capreg)
1600 {
1601 	struct pci_devinfo *dinfo = device_get_ivars(child);
1602 	pcicfgregs *cfg = &dinfo->cfg;
1603 	uint32_t ecap;
1604 	uint16_t ptr;
1605 
1606 	/* Only supported for PCI-express devices. */
1607 	if (cfg->pcie.pcie_location == 0)
1608 		return (ENXIO);
1609 
1610 	ptr = PCIR_EXTCAP;
1611 	ecap = pci_read_config(child, ptr, 4);
1612 	if (ecap == 0xffffffff || ecap == 0)
1613 		return (ENOENT);
1614 	for (;;) {
1615 		if (PCI_EXTCAP_ID(ecap) == capability) {
1616 			if (capreg != NULL)
1617 				*capreg = ptr;
1618 			return (0);
1619 		}
1620 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1621 		if (ptr == 0)
1622 			break;
1623 		ecap = pci_read_config(child, ptr, 4);
1624 	}
1625 
1626 	return (ENOENT);
1627 }
1628 
1629 /*
1630  * Find the next requested extended capability after start and return the
1631  * offset in configuration space via the pointer provided.  The function
1632  * returns 0 on success and an error code otherwise.
1633  */
1634 int
pci_find_next_extcap_method(device_t dev,device_t child,int capability,int start,int * capreg)1635 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1636     int start, int *capreg)
1637 {
1638 	struct pci_devinfo *dinfo = device_get_ivars(child);
1639 	pcicfgregs *cfg = &dinfo->cfg;
1640 	uint32_t ecap;
1641 	uint16_t ptr;
1642 
1643 	/* Only supported for PCI-express devices. */
1644 	if (cfg->pcie.pcie_location == 0)
1645 		return (ENXIO);
1646 
1647 	ecap = pci_read_config(child, start, 4);
1648 	KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1649 	    ("start extended capability is not expected capability"));
1650 	ptr = PCI_EXTCAP_NEXTPTR(ecap);
1651 	while (ptr != 0) {
1652 		ecap = pci_read_config(child, ptr, 4);
1653 		if (PCI_EXTCAP_ID(ecap) == capability) {
1654 			if (capreg != NULL)
1655 				*capreg = ptr;
1656 			return (0);
1657 		}
1658 		ptr = PCI_EXTCAP_NEXTPTR(ecap);
1659 	}
1660 
1661 	return (ENOENT);
1662 }
1663 
1664 /*
1665  * Support for MSI-X message interrupts.
1666  */
1667 static void
pci_write_msix_entry(device_t dev,u_int index,uint64_t address,uint32_t data)1668 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1669 {
1670 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1671 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1672 	uint32_t offset;
1673 
1674 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1675 	offset = msix->msix_table_offset + index * 16;
1676 	bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1677 	bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1678 	bus_write_4(msix->msix_table_res, offset + 8, data);
1679 }
1680 
1681 void
pci_enable_msix_method(device_t dev,device_t child,u_int index,uint64_t address,uint32_t data)1682 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1683     uint64_t address, uint32_t data)
1684 {
1685 
1686 	if (pci_msix_rewrite_table) {
1687 		struct pci_devinfo *dinfo = device_get_ivars(child);
1688 		struct pcicfg_msix *msix = &dinfo->cfg.msix;
1689 
1690 		/*
1691 		 * Some VM hosts require MSIX to be disabled in the
1692 		 * control register before updating the MSIX table
1693 		 * entries are allowed. It is not enough to only
1694 		 * disable MSIX while updating a single entry. MSIX
1695 		 * must be disabled while updating all entries in the
1696 		 * table.
1697 		 */
1698 		pci_write_config(child,
1699 		    msix->msix_location + PCIR_MSIX_CTRL,
1700 		    msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1701 		pci_resume_msix(child);
1702 	} else
1703 		pci_write_msix_entry(child, index, address, data);
1704 
1705 	/* Enable MSI -> HT mapping. */
1706 	pci_ht_map_msi(child, address);
1707 }
1708 
1709 void
pci_mask_msix(device_t dev,u_int index)1710 pci_mask_msix(device_t dev, u_int index)
1711 {
1712 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1713 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1714 	uint32_t offset, val;
1715 
1716 	KASSERT(msix->msix_msgnum > index, ("bogus index"));
1717 	offset = msix->msix_table_offset + index * 16 + 12;
1718 	val = bus_read_4(msix->msix_table_res, offset);
1719 	val |= PCIM_MSIX_VCTRL_MASK;
1720 
1721 	/*
1722 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1723 	 * register, so always write the new value.
1724 	 */
1725 	bus_write_4(msix->msix_table_res, offset, val);
1726 }
1727 
1728 void
pci_unmask_msix(device_t dev,u_int index)1729 pci_unmask_msix(device_t dev, u_int index)
1730 {
1731 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1732 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1733 	uint32_t offset, val;
1734 
1735 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1736 	offset = msix->msix_table_offset + index * 16 + 12;
1737 	val = bus_read_4(msix->msix_table_res, offset);
1738 	val &= ~PCIM_MSIX_VCTRL_MASK;
1739 
1740 	/*
1741 	 * Some devices (e.g. Samsung PM961) do not support reads of this
1742 	 * register, so always write the new value.
1743 	 */
1744 	bus_write_4(msix->msix_table_res, offset, val);
1745 }
1746 
1747 int
pci_pending_msix(device_t dev,u_int index)1748 pci_pending_msix(device_t dev, u_int index)
1749 {
1750 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1751 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1752 	uint32_t offset, bit;
1753 
1754 	KASSERT(msix->msix_table_len > index, ("bogus index"));
1755 	offset = msix->msix_pba_offset + (index / 32) * 4;
1756 	bit = 1 << index % 32;
1757 	return (bus_read_4(msix->msix_pba_res, offset) & bit);
1758 }
1759 
1760 /*
1761  * Restore MSI-X registers and table during resume.  If MSI-X is
1762  * enabled then walk the virtual table to restore the actual MSI-X
1763  * table.
1764  */
1765 static void
pci_resume_msix(device_t dev)1766 pci_resume_msix(device_t dev)
1767 {
1768 	struct pci_devinfo *dinfo = device_get_ivars(dev);
1769 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1770 	struct msix_table_entry *mte;
1771 	struct msix_vector *mv;
1772 	int i;
1773 
1774 	if (msix->msix_alloc > 0) {
1775 		/* First, mask all vectors. */
1776 		for (i = 0; i < msix->msix_msgnum; i++)
1777 			pci_mask_msix(dev, i);
1778 
1779 		/* Second, program any messages with at least one handler. */
1780 		for (i = 0; i < msix->msix_table_len; i++) {
1781 			mte = &msix->msix_table[i];
1782 			if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1783 				continue;
1784 			mv = &msix->msix_vectors[mte->mte_vector - 1];
1785 			pci_write_msix_entry(dev, i, mv->mv_address,
1786 			    mv->mv_data);
1787 			pci_unmask_msix(dev, i);
1788 		}
1789 	}
1790 	pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1791 	    msix->msix_ctrl, 2);
1792 }
1793 
1794 /*
1795  * Attempt to allocate *count MSI-X messages.  The actual number allocated is
1796  * returned in *count.  After this function returns, each message will be
1797  * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1798  */
1799 int
pci_alloc_msix_method(device_t dev,device_t child,int * count)1800 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1801 {
1802 	struct pci_devinfo *dinfo = device_get_ivars(child);
1803 	pcicfgregs *cfg = &dinfo->cfg;
1804 	struct resource_list_entry *rle;
1805 	int actual, error, i, irq, max;
1806 
1807 	/* Don't let count == 0 get us into trouble. */
1808 	if (*count == 0)
1809 		return (EINVAL);
1810 
1811 	/* If rid 0 is allocated, then fail. */
1812 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1813 	if (rle != NULL && rle->res != NULL)
1814 		return (ENXIO);
1815 
1816 	/* Already have allocated messages? */
1817 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1818 		return (ENXIO);
1819 
1820 	/* If MSI-X is blacklisted for this system, fail. */
1821 	if (pci_msix_blacklisted())
1822 		return (ENXIO);
1823 
1824 	/* MSI-X capability present? */
1825 	if (cfg->msix.msix_location == 0 || !pci_do_msix)
1826 		return (ENODEV);
1827 
1828 	/* Make sure the appropriate BARs are mapped. */
1829 	rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1830 	    cfg->msix.msix_table_bar);
1831 	if (rle == NULL || rle->res == NULL ||
1832 	    !(rman_get_flags(rle->res) & RF_ACTIVE))
1833 		return (ENXIO);
1834 	cfg->msix.msix_table_res = rle->res;
1835 	if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1836 		rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1837 		    cfg->msix.msix_pba_bar);
1838 		if (rle == NULL || rle->res == NULL ||
1839 		    !(rman_get_flags(rle->res) & RF_ACTIVE))
1840 			return (ENXIO);
1841 	}
1842 	cfg->msix.msix_pba_res = rle->res;
1843 
1844 	if (bootverbose)
1845 		device_printf(child,
1846 		    "attempting to allocate %d MSI-X vectors (%d supported)\n",
1847 		    *count, cfg->msix.msix_msgnum);
1848 	max = min(*count, cfg->msix.msix_msgnum);
1849 	for (i = 0; i < max; i++) {
1850 		/* Allocate a message. */
1851 		error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1852 		if (error) {
1853 			if (i == 0)
1854 				return (error);
1855 			break;
1856 		}
1857 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1858 		    irq, 1);
1859 	}
1860 	actual = i;
1861 
1862 	if (bootverbose) {
1863 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1864 		if (actual == 1)
1865 			device_printf(child, "using IRQ %ju for MSI-X\n",
1866 			    rle->start);
1867 		else {
1868 			int run;
1869 
1870 			/*
1871 			 * Be fancy and try to print contiguous runs of
1872 			 * IRQ values as ranges.  'irq' is the previous IRQ.
1873 			 * 'run' is true if we are in a range.
1874 			 */
1875 			device_printf(child, "using IRQs %ju", rle->start);
1876 			irq = rle->start;
1877 			run = 0;
1878 			for (i = 1; i < actual; i++) {
1879 				rle = resource_list_find(&dinfo->resources,
1880 				    SYS_RES_IRQ, i + 1);
1881 
1882 				/* Still in a run? */
1883 				if (rle->start == irq + 1) {
1884 					run = 1;
1885 					irq++;
1886 					continue;
1887 				}
1888 
1889 				/* Finish previous range. */
1890 				if (run) {
1891 					printf("-%d", irq);
1892 					run = 0;
1893 				}
1894 
1895 				/* Start new range. */
1896 				printf(",%ju", rle->start);
1897 				irq = rle->start;
1898 			}
1899 
1900 			/* Unfinished range? */
1901 			if (run)
1902 				printf("-%d", irq);
1903 			printf(" for MSI-X\n");
1904 		}
1905 	}
1906 
1907 	/* Mask all vectors. */
1908 	for (i = 0; i < cfg->msix.msix_msgnum; i++)
1909 		pci_mask_msix(child, i);
1910 
1911 	/* Allocate and initialize vector data and virtual table. */
1912 	cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1913 	    M_DEVBUF, M_WAITOK | M_ZERO);
1914 	cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1915 	    M_DEVBUF, M_WAITOK | M_ZERO);
1916 	for (i = 0; i < actual; i++) {
1917 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1918 		cfg->msix.msix_vectors[i].mv_irq = rle->start;
1919 		cfg->msix.msix_table[i].mte_vector = i + 1;
1920 	}
1921 
1922 	/* Update control register to enable MSI-X. */
1923 	cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1924 	pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1925 	    cfg->msix.msix_ctrl, 2);
1926 
1927 	/* Update counts of alloc'd messages. */
1928 	cfg->msix.msix_alloc = actual;
1929 	cfg->msix.msix_table_len = actual;
1930 	*count = actual;
1931 	return (0);
1932 }
1933 
1934 /*
1935  * By default, pci_alloc_msix() will assign the allocated IRQ
1936  * resources consecutively to the first N messages in the MSI-X table.
1937  * However, device drivers may want to use different layouts if they
1938  * either receive fewer messages than they asked for, or they wish to
1939  * populate the MSI-X table sparsely.  This method allows the driver
1940  * to specify what layout it wants.  It must be called after a
1941  * successful pci_alloc_msix() but before any of the associated
1942  * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1943  *
1944  * The 'vectors' array contains 'count' message vectors.  The array
1945  * maps directly to the MSI-X table in that index 0 in the array
1946  * specifies the vector for the first message in the MSI-X table, etc.
1947  * The vector value in each array index can either be 0 to indicate
1948  * that no vector should be assigned to a message slot, or it can be a
1949  * number from 1 to N (where N is the count returned from a
1950  * succcessful call to pci_alloc_msix()) to indicate which message
1951  * vector (IRQ) to be used for the corresponding message.
1952  *
1953  * On successful return, each message with a non-zero vector will have
1954  * an associated SYS_RES_IRQ whose rid is equal to the array index +
1955  * 1.  Additionally, if any of the IRQs allocated via the previous
1956  * call to pci_alloc_msix() are not used in the mapping, those IRQs
1957  * will be freed back to the system automatically.
1958  *
1959  * For example, suppose a driver has a MSI-X table with 6 messages and
1960  * asks for 6 messages, but pci_alloc_msix() only returns a count of
1961  * 3.  Call the three vectors allocated by pci_alloc_msix() A, B, and
1962  * C.  After the call to pci_alloc_msix(), the device will be setup to
1963  * have an MSI-X table of ABC--- (where - means no vector assigned).
1964  * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1965  * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1966  * be freed back to the system.  This device will also have valid
1967  * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1968  *
1969  * In any case, the SYS_RES_IRQ rid X will always map to the message
1970  * at MSI-X table index X - 1 and will only be valid if a vector is
1971  * assigned to that table entry.
1972  */
1973 int
pci_remap_msix_method(device_t dev,device_t child,int count,const u_int * vectors)1974 pci_remap_msix_method(device_t dev, device_t child, int count,
1975     const u_int *vectors)
1976 {
1977 	struct pci_devinfo *dinfo = device_get_ivars(child);
1978 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
1979 	struct resource_list_entry *rle;
1980 	int i, irq, j, *used;
1981 
1982 	/*
1983 	 * Have to have at least one message in the table but the
1984 	 * table can't be bigger than the actual MSI-X table in the
1985 	 * device.
1986 	 */
1987 	if (count == 0 || count > msix->msix_msgnum)
1988 		return (EINVAL);
1989 
1990 	/* Sanity check the vectors. */
1991 	for (i = 0; i < count; i++)
1992 		if (vectors[i] > msix->msix_alloc)
1993 			return (EINVAL);
1994 
1995 	/*
1996 	 * Make sure there aren't any holes in the vectors to be used.
1997 	 * It's a big pain to support it, and it doesn't really make
1998 	 * sense anyway.  Also, at least one vector must be used.
1999 	 */
2000 	used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
2001 	    M_ZERO);
2002 	for (i = 0; i < count; i++)
2003 		if (vectors[i] != 0)
2004 			used[vectors[i] - 1] = 1;
2005 	for (i = 0; i < msix->msix_alloc - 1; i++)
2006 		if (used[i] == 0 && used[i + 1] == 1) {
2007 			free(used, M_DEVBUF);
2008 			return (EINVAL);
2009 		}
2010 	if (used[0] != 1) {
2011 		free(used, M_DEVBUF);
2012 		return (EINVAL);
2013 	}
2014 
2015 	/* Make sure none of the resources are allocated. */
2016 	for (i = 0; i < msix->msix_table_len; i++) {
2017 		if (msix->msix_table[i].mte_vector == 0)
2018 			continue;
2019 		if (msix->msix_table[i].mte_handlers > 0) {
2020 			free(used, M_DEVBUF);
2021 			return (EBUSY);
2022 		}
2023 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2024 		KASSERT(rle != NULL, ("missing resource"));
2025 		if (rle->res != NULL) {
2026 			free(used, M_DEVBUF);
2027 			return (EBUSY);
2028 		}
2029 	}
2030 
2031 	/* Free the existing resource list entries. */
2032 	for (i = 0; i < msix->msix_table_len; i++) {
2033 		if (msix->msix_table[i].mte_vector == 0)
2034 			continue;
2035 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2036 	}
2037 
2038 	/*
2039 	 * Build the new virtual table keeping track of which vectors are
2040 	 * used.
2041 	 */
2042 	free(msix->msix_table, M_DEVBUF);
2043 	msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
2044 	    M_DEVBUF, M_WAITOK | M_ZERO);
2045 	for (i = 0; i < count; i++)
2046 		msix->msix_table[i].mte_vector = vectors[i];
2047 	msix->msix_table_len = count;
2048 
2049 	/* Free any unused IRQs and resize the vectors array if necessary. */
2050 	j = msix->msix_alloc - 1;
2051 	if (used[j] == 0) {
2052 		struct msix_vector *vec;
2053 
2054 		while (used[j] == 0) {
2055 			PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2056 			    msix->msix_vectors[j].mv_irq);
2057 			j--;
2058 		}
2059 		vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2060 		    M_WAITOK);
2061 		bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2062 		    (j + 1));
2063 		free(msix->msix_vectors, M_DEVBUF);
2064 		msix->msix_vectors = vec;
2065 		msix->msix_alloc = j + 1;
2066 	}
2067 	free(used, M_DEVBUF);
2068 
2069 	/* Map the IRQs onto the rids. */
2070 	for (i = 0; i < count; i++) {
2071 		if (vectors[i] == 0)
2072 			continue;
2073 		irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2074 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2075 		    irq, 1);
2076 	}
2077 
2078 	if (bootverbose) {
2079 		device_printf(child, "Remapped MSI-X IRQs as: ");
2080 		for (i = 0; i < count; i++) {
2081 			if (i != 0)
2082 				printf(", ");
2083 			if (vectors[i] == 0)
2084 				printf("---");
2085 			else
2086 				printf("%d",
2087 				    msix->msix_vectors[vectors[i] - 1].mv_irq);
2088 		}
2089 		printf("\n");
2090 	}
2091 
2092 	return (0);
2093 }
2094 
2095 static int
pci_release_msix(device_t dev,device_t child)2096 pci_release_msix(device_t dev, device_t child)
2097 {
2098 	struct pci_devinfo *dinfo = device_get_ivars(child);
2099 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2100 	struct resource_list_entry *rle;
2101 	int i;
2102 
2103 	/* Do we have any messages to release? */
2104 	if (msix->msix_alloc == 0)
2105 		return (ENODEV);
2106 
2107 	/* Make sure none of the resources are allocated. */
2108 	for (i = 0; i < msix->msix_table_len; i++) {
2109 		if (msix->msix_table[i].mte_vector == 0)
2110 			continue;
2111 		if (msix->msix_table[i].mte_handlers > 0)
2112 			return (EBUSY);
2113 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2114 		KASSERT(rle != NULL, ("missing resource"));
2115 		if (rle->res != NULL)
2116 			return (EBUSY);
2117 	}
2118 
2119 	/* Update control register to disable MSI-X. */
2120 	msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2121 	pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2122 	    msix->msix_ctrl, 2);
2123 
2124 	/* Free the resource list entries. */
2125 	for (i = 0; i < msix->msix_table_len; i++) {
2126 		if (msix->msix_table[i].mte_vector == 0)
2127 			continue;
2128 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2129 	}
2130 	free(msix->msix_table, M_DEVBUF);
2131 	msix->msix_table_len = 0;
2132 
2133 	/* Release the IRQs. */
2134 	for (i = 0; i < msix->msix_alloc; i++)
2135 		PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2136 		    msix->msix_vectors[i].mv_irq);
2137 	free(msix->msix_vectors, M_DEVBUF);
2138 	msix->msix_alloc = 0;
2139 	return (0);
2140 }
2141 
2142 /*
2143  * Return the max supported MSI-X messages this device supports.
2144  * Basically, assuming the MD code can alloc messages, this function
2145  * should return the maximum value that pci_alloc_msix() can return.
2146  * Thus, it is subject to the tunables, etc.
2147  */
2148 int
pci_msix_count_method(device_t dev,device_t child)2149 pci_msix_count_method(device_t dev, device_t child)
2150 {
2151 	struct pci_devinfo *dinfo = device_get_ivars(child);
2152 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2153 
2154 	if (pci_do_msix && msix->msix_location != 0)
2155 		return (msix->msix_msgnum);
2156 	return (0);
2157 }
2158 
2159 int
pci_msix_pba_bar_method(device_t dev,device_t child)2160 pci_msix_pba_bar_method(device_t dev, device_t child)
2161 {
2162 	struct pci_devinfo *dinfo = device_get_ivars(child);
2163 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2164 
2165 	if (pci_do_msix && msix->msix_location != 0)
2166 		return (msix->msix_pba_bar);
2167 	return (-1);
2168 }
2169 
2170 int
pci_msix_table_bar_method(device_t dev,device_t child)2171 pci_msix_table_bar_method(device_t dev, device_t child)
2172 {
2173 	struct pci_devinfo *dinfo = device_get_ivars(child);
2174 	struct pcicfg_msix *msix = &dinfo->cfg.msix;
2175 
2176 	if (pci_do_msix && msix->msix_location != 0)
2177 		return (msix->msix_table_bar);
2178 	return (-1);
2179 }
2180 
2181 /*
2182  * HyperTransport MSI mapping control
2183  */
2184 void
pci_ht_map_msi(device_t dev,uint64_t addr)2185 pci_ht_map_msi(device_t dev, uint64_t addr)
2186 {
2187 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2188 	struct pcicfg_ht *ht = &dinfo->cfg.ht;
2189 
2190 	if (!ht->ht_msimap)
2191 		return;
2192 
2193 	if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2194 	    ht->ht_msiaddr >> 20 == addr >> 20) {
2195 		/* Enable MSI -> HT mapping. */
2196 		ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2197 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2198 		    ht->ht_msictrl, 2);
2199 	}
2200 
2201 	if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2202 		/* Disable MSI -> HT mapping. */
2203 		ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2204 		pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2205 		    ht->ht_msictrl, 2);
2206 	}
2207 }
2208 
2209 int
pci_get_relaxed_ordering_enabled(device_t dev)2210 pci_get_relaxed_ordering_enabled(device_t dev)
2211 {
2212 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2213 	int cap;
2214 	uint16_t val;
2215 
2216 	cap = dinfo->cfg.pcie.pcie_location;
2217 	if (cap == 0)
2218 		return (0);
2219 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2220 	val &= PCIEM_CTL_RELAXED_ORD_ENABLE;
2221 	return (val != 0);
2222 }
2223 
2224 int
pci_get_max_payload(device_t dev)2225 pci_get_max_payload(device_t dev)
2226 {
2227 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2228 	int cap;
2229 	uint16_t val;
2230 
2231 	cap = dinfo->cfg.pcie.pcie_location;
2232 	if (cap == 0)
2233 		return (0);
2234 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2235 	val &= PCIEM_CTL_MAX_PAYLOAD;
2236 	val >>= 5;
2237 	return (1 << (val + 7));
2238 }
2239 
2240 int
pci_get_max_read_req(device_t dev)2241 pci_get_max_read_req(device_t dev)
2242 {
2243 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2244 	int cap;
2245 	uint16_t val;
2246 
2247 	cap = dinfo->cfg.pcie.pcie_location;
2248 	if (cap == 0)
2249 		return (0);
2250 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2251 	val &= PCIEM_CTL_MAX_READ_REQUEST;
2252 	val >>= 12;
2253 	return (1 << (val + 7));
2254 }
2255 
2256 int
pci_set_max_read_req(device_t dev,int size)2257 pci_set_max_read_req(device_t dev, int size)
2258 {
2259 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2260 	int cap;
2261 	uint16_t val;
2262 
2263 	cap = dinfo->cfg.pcie.pcie_location;
2264 	if (cap == 0)
2265 		return (0);
2266 	if (size < 128)
2267 		size = 128;
2268 	if (size > 4096)
2269 		size = 4096;
2270 	size = (1 << (fls(size) - 1));
2271 	val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2272 	val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2273 	val |= (fls(size) - 8) << 12;
2274 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2275 	return (size);
2276 }
2277 
2278 uint32_t
pcie_read_config(device_t dev,int reg,int width)2279 pcie_read_config(device_t dev, int reg, int width)
2280 {
2281 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2282 	int cap;
2283 
2284 	cap = dinfo->cfg.pcie.pcie_location;
2285 	if (cap == 0) {
2286 		if (width == 2)
2287 			return (0xffff);
2288 		return (0xffffffff);
2289 	}
2290 
2291 	return (pci_read_config(dev, cap + reg, width));
2292 }
2293 
2294 void
pcie_write_config(device_t dev,int reg,uint32_t value,int width)2295 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2296 {
2297 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2298 	int cap;
2299 
2300 	cap = dinfo->cfg.pcie.pcie_location;
2301 	if (cap == 0)
2302 		return;
2303 	pci_write_config(dev, cap + reg, value, width);
2304 }
2305 
2306 /*
2307  * Adjusts a PCI-e capability register by clearing the bits in mask
2308  * and setting the bits in (value & mask).  Bits not set in mask are
2309  * not adjusted.
2310  *
2311  * Returns the old value on success or all ones on failure.
2312  */
2313 uint32_t
pcie_adjust_config(device_t dev,int reg,uint32_t mask,uint32_t value,int width)2314 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2315     int width)
2316 {
2317 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2318 	uint32_t old, new;
2319 	int cap;
2320 
2321 	cap = dinfo->cfg.pcie.pcie_location;
2322 	if (cap == 0) {
2323 		if (width == 2)
2324 			return (0xffff);
2325 		return (0xffffffff);
2326 	}
2327 
2328 	old = pci_read_config(dev, cap + reg, width);
2329 	new = old & ~mask;
2330 	new |= (value & mask);
2331 	pci_write_config(dev, cap + reg, new, width);
2332 	return (old);
2333 }
2334 
2335 /*
2336  * Support for MSI message signalled interrupts.
2337  */
2338 void
pci_enable_msi_method(device_t dev,device_t child,uint64_t address,uint16_t data)2339 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2340     uint16_t data)
2341 {
2342 	struct pci_devinfo *dinfo = device_get_ivars(child);
2343 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2344 
2345 	/* Write data and address values. */
2346 	pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2347 	    address & 0xffffffff, 4);
2348 	if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2349 		pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2350 		    address >> 32, 4);
2351 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2352 		    data, 2);
2353 	} else
2354 		pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2355 		    2);
2356 
2357 	/* Enable MSI in the control register. */
2358 	msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2359 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2360 	    msi->msi_ctrl, 2);
2361 
2362 	/* Enable MSI -> HT mapping. */
2363 	pci_ht_map_msi(child, address);
2364 }
2365 
2366 void
pci_disable_msi_method(device_t dev,device_t child)2367 pci_disable_msi_method(device_t dev, device_t child)
2368 {
2369 	struct pci_devinfo *dinfo = device_get_ivars(child);
2370 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2371 
2372 	/* Disable MSI -> HT mapping. */
2373 	pci_ht_map_msi(child, 0);
2374 
2375 	/* Disable MSI in the control register. */
2376 	msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2377 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2378 	    msi->msi_ctrl, 2);
2379 }
2380 
2381 /*
2382  * Restore MSI registers during resume.  If MSI is enabled then
2383  * restore the data and address registers in addition to the control
2384  * register.
2385  */
2386 static void
pci_resume_msi(device_t dev)2387 pci_resume_msi(device_t dev)
2388 {
2389 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2390 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2391 	uint64_t address;
2392 	uint16_t data;
2393 
2394 	if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2395 		address = msi->msi_addr;
2396 		data = msi->msi_data;
2397 		pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2398 		    address & 0xffffffff, 4);
2399 		if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2400 			pci_write_config(dev, msi->msi_location +
2401 			    PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2402 			pci_write_config(dev, msi->msi_location +
2403 			    PCIR_MSI_DATA_64BIT, data, 2);
2404 		} else
2405 			pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2406 			    data, 2);
2407 	}
2408 	pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2409 	    2);
2410 }
2411 
2412 static int
pci_remap_intr_method(device_t bus,device_t dev,u_int irq)2413 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2414 {
2415 	struct pci_devinfo *dinfo = device_get_ivars(dev);
2416 	pcicfgregs *cfg = &dinfo->cfg;
2417 	struct resource_list_entry *rle;
2418 	struct msix_table_entry *mte;
2419 	struct msix_vector *mv;
2420 	uint64_t addr;
2421 	uint32_t data;
2422 	int error, i, j;
2423 
2424 	/*
2425 	 * Handle MSI first.  We try to find this IRQ among our list
2426 	 * of MSI IRQs.  If we find it, we request updated address and
2427 	 * data registers and apply the results.
2428 	 */
2429 	if (cfg->msi.msi_alloc > 0) {
2430 		/* If we don't have any active handlers, nothing to do. */
2431 		if (cfg->msi.msi_handlers == 0)
2432 			return (0);
2433 		for (i = 0; i < cfg->msi.msi_alloc; i++) {
2434 			rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2435 			    i + 1);
2436 			if (rle->start == irq) {
2437 				error = PCIB_MAP_MSI(device_get_parent(bus),
2438 				    dev, irq, &addr, &data);
2439 				if (error)
2440 					return (error);
2441 				pci_disable_msi(dev);
2442 				dinfo->cfg.msi.msi_addr = addr;
2443 				dinfo->cfg.msi.msi_data = data;
2444 				pci_enable_msi(dev, addr, data);
2445 				return (0);
2446 			}
2447 		}
2448 		return (ENOENT);
2449 	}
2450 
2451 	/*
2452 	 * For MSI-X, we check to see if we have this IRQ.  If we do,
2453 	 * we request the updated mapping info.  If that works, we go
2454 	 * through all the slots that use this IRQ and update them.
2455 	 */
2456 	if (cfg->msix.msix_alloc > 0) {
2457 		bool found = false;
2458 
2459 		for (i = 0; i < cfg->msix.msix_alloc; i++) {
2460 			mv = &cfg->msix.msix_vectors[i];
2461 			if (mv->mv_irq == irq) {
2462 				error = PCIB_MAP_MSI(device_get_parent(bus),
2463 				    dev, irq, &addr, &data);
2464 				if (error)
2465 					return (error);
2466 				mv->mv_address = addr;
2467 				mv->mv_data = data;
2468 				for (j = 0; j < cfg->msix.msix_table_len; j++) {
2469 					mte = &cfg->msix.msix_table[j];
2470 					if (mte->mte_vector != i + 1)
2471 						continue;
2472 					if (mte->mte_handlers == 0)
2473 						continue;
2474 					pci_mask_msix(dev, j);
2475 					pci_enable_msix(dev, j, addr, data);
2476 					pci_unmask_msix(dev, j);
2477 				}
2478 				found = true;
2479 			}
2480 		}
2481 		return (found ? 0 : ENOENT);
2482 	}
2483 
2484 	return (ENOENT);
2485 }
2486 
2487 /*
2488  * Returns true if the specified device is blacklisted because MSI
2489  * doesn't work.
2490  */
2491 int
pci_msi_device_blacklisted(device_t dev)2492 pci_msi_device_blacklisted(device_t dev)
2493 {
2494 
2495 	if (!pci_honor_msi_blacklist)
2496 		return (0);
2497 
2498 	return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2499 }
2500 
2501 /*
2502  * Determine if MSI is blacklisted globally on this system.  Currently,
2503  * we just check for blacklisted chipsets as represented by the
2504  * host-PCI bridge at device 0:0:0.  In the future, it may become
2505  * necessary to check other system attributes, such as the kenv values
2506  * that give the motherboard manufacturer and model number.
2507  */
2508 static int
pci_msi_blacklisted(void)2509 pci_msi_blacklisted(void)
2510 {
2511 	device_t dev;
2512 
2513 	if (!pci_honor_msi_blacklist)
2514 		return (0);
2515 
2516 	/* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2517 	if (!(pcie_chipset || pcix_chipset)) {
2518 		if (vm_guest != VM_GUEST_NO) {
2519 			/*
2520 			 * Whitelist older chipsets in virtual
2521 			 * machines known to support MSI.
2522 			 */
2523 			dev = pci_find_bsf(0, 0, 0);
2524 			if (dev != NULL)
2525 				return (!pci_has_quirk(pci_get_devid(dev),
2526 					PCI_QUIRK_ENABLE_MSI_VM));
2527 		}
2528 		return (1);
2529 	}
2530 
2531 	dev = pci_find_bsf(0, 0, 0);
2532 	if (dev != NULL)
2533 		return (pci_msi_device_blacklisted(dev));
2534 	return (0);
2535 }
2536 
2537 /*
2538  * Returns true if the specified device is blacklisted because MSI-X
2539  * doesn't work.  Note that this assumes that if MSI doesn't work,
2540  * MSI-X doesn't either.
2541  */
2542 int
pci_msix_device_blacklisted(device_t dev)2543 pci_msix_device_blacklisted(device_t dev)
2544 {
2545 
2546 	if (!pci_honor_msi_blacklist)
2547 		return (0);
2548 
2549 	if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2550 		return (1);
2551 
2552 	return (pci_msi_device_blacklisted(dev));
2553 }
2554 
2555 /*
2556  * Determine if MSI-X is blacklisted globally on this system.  If MSI
2557  * is blacklisted, assume that MSI-X is as well.  Check for additional
2558  * chipsets where MSI works but MSI-X does not.
2559  */
2560 static int
pci_msix_blacklisted(void)2561 pci_msix_blacklisted(void)
2562 {
2563 	device_t dev;
2564 
2565 	if (!pci_honor_msi_blacklist)
2566 		return (0);
2567 
2568 	dev = pci_find_bsf(0, 0, 0);
2569 	if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2570 	    PCI_QUIRK_DISABLE_MSIX))
2571 		return (1);
2572 
2573 	return (pci_msi_blacklisted());
2574 }
2575 
2576 /*
2577  * Attempt to allocate *count MSI messages.  The actual number allocated is
2578  * returned in *count.  After this function returns, each message will be
2579  * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2580  */
2581 int
pci_alloc_msi_method(device_t dev,device_t child,int * count)2582 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2583 {
2584 	struct pci_devinfo *dinfo = device_get_ivars(child);
2585 	pcicfgregs *cfg = &dinfo->cfg;
2586 	struct resource_list_entry *rle;
2587 	int actual, error, i, irqs[32];
2588 	uint16_t ctrl;
2589 
2590 	/* Don't let count == 0 get us into trouble. */
2591 	if (*count == 0)
2592 		return (EINVAL);
2593 
2594 	/* If rid 0 is allocated, then fail. */
2595 	rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2596 	if (rle != NULL && rle->res != NULL)
2597 		return (ENXIO);
2598 
2599 	/* Already have allocated messages? */
2600 	if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2601 		return (ENXIO);
2602 
2603 	/* If MSI is blacklisted for this system, fail. */
2604 	if (pci_msi_blacklisted())
2605 		return (ENXIO);
2606 
2607 	/* MSI capability present? */
2608 	if (cfg->msi.msi_location == 0 || !pci_do_msi)
2609 		return (ENODEV);
2610 
2611 	if (bootverbose)
2612 		device_printf(child,
2613 		    "attempting to allocate %d MSI vectors (%d supported)\n",
2614 		    *count, cfg->msi.msi_msgnum);
2615 
2616 	/* Don't ask for more than the device supports. */
2617 	actual = min(*count, cfg->msi.msi_msgnum);
2618 
2619 	/* Don't ask for more than 32 messages. */
2620 	actual = min(actual, 32);
2621 
2622 	/* MSI requires power of 2 number of messages. */
2623 	if (!powerof2(actual))
2624 		return (EINVAL);
2625 
2626 	for (;;) {
2627 		/* Try to allocate N messages. */
2628 		error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2629 		    actual, irqs);
2630 		if (error == 0)
2631 			break;
2632 		if (actual == 1)
2633 			return (error);
2634 
2635 		/* Try N / 2. */
2636 		actual >>= 1;
2637 	}
2638 
2639 	/*
2640 	 * We now have N actual messages mapped onto SYS_RES_IRQ
2641 	 * resources in the irqs[] array, so add new resources
2642 	 * starting at rid 1.
2643 	 */
2644 	for (i = 0; i < actual; i++)
2645 		resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2646 		    irqs[i], irqs[i], 1);
2647 
2648 	if (bootverbose) {
2649 		if (actual == 1)
2650 			device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2651 		else {
2652 			int run;
2653 
2654 			/*
2655 			 * Be fancy and try to print contiguous runs
2656 			 * of IRQ values as ranges.  'run' is true if
2657 			 * we are in a range.
2658 			 */
2659 			device_printf(child, "using IRQs %d", irqs[0]);
2660 			run = 0;
2661 			for (i = 1; i < actual; i++) {
2662 				/* Still in a run? */
2663 				if (irqs[i] == irqs[i - 1] + 1) {
2664 					run = 1;
2665 					continue;
2666 				}
2667 
2668 				/* Finish previous range. */
2669 				if (run) {
2670 					printf("-%d", irqs[i - 1]);
2671 					run = 0;
2672 				}
2673 
2674 				/* Start new range. */
2675 				printf(",%d", irqs[i]);
2676 			}
2677 
2678 			/* Unfinished range? */
2679 			if (run)
2680 				printf("-%d", irqs[actual - 1]);
2681 			printf(" for MSI\n");
2682 		}
2683 	}
2684 
2685 	/* Update control register with actual count. */
2686 	ctrl = cfg->msi.msi_ctrl;
2687 	ctrl &= ~PCIM_MSICTRL_MME_MASK;
2688 	ctrl |= (ffs(actual) - 1) << 4;
2689 	cfg->msi.msi_ctrl = ctrl;
2690 	pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2691 
2692 	/* Update counts of alloc'd messages. */
2693 	cfg->msi.msi_alloc = actual;
2694 	cfg->msi.msi_handlers = 0;
2695 	*count = actual;
2696 	return (0);
2697 }
2698 
2699 /* Release the MSI messages associated with this device. */
2700 int
pci_release_msi_method(device_t dev,device_t child)2701 pci_release_msi_method(device_t dev, device_t child)
2702 {
2703 	struct pci_devinfo *dinfo = device_get_ivars(child);
2704 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2705 	struct resource_list_entry *rle;
2706 	int error, i, irqs[32];
2707 
2708 	/* Try MSI-X first. */
2709 	error = pci_release_msix(dev, child);
2710 	if (error != ENODEV)
2711 		return (error);
2712 
2713 	/* Do we have any messages to release? */
2714 	if (msi->msi_alloc == 0)
2715 		return (ENODEV);
2716 	KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2717 
2718 	/* Make sure none of the resources are allocated. */
2719 	if (msi->msi_handlers > 0)
2720 		return (EBUSY);
2721 	for (i = 0; i < msi->msi_alloc; i++) {
2722 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2723 		KASSERT(rle != NULL, ("missing MSI resource"));
2724 		if (rle->res != NULL)
2725 			return (EBUSY);
2726 		irqs[i] = rle->start;
2727 	}
2728 
2729 	/* Update control register with 0 count. */
2730 	KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2731 	    ("%s: MSI still enabled", __func__));
2732 	msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2733 	pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2734 	    msi->msi_ctrl, 2);
2735 
2736 	/* Release the messages. */
2737 	PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2738 	for (i = 0; i < msi->msi_alloc; i++)
2739 		resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2740 
2741 	/* Update alloc count. */
2742 	msi->msi_alloc = 0;
2743 	msi->msi_addr = 0;
2744 	msi->msi_data = 0;
2745 	return (0);
2746 }
2747 
2748 /*
2749  * Return the max supported MSI messages this device supports.
2750  * Basically, assuming the MD code can alloc messages, this function
2751  * should return the maximum value that pci_alloc_msi() can return.
2752  * Thus, it is subject to the tunables, etc.
2753  */
2754 int
pci_msi_count_method(device_t dev,device_t child)2755 pci_msi_count_method(device_t dev, device_t child)
2756 {
2757 	struct pci_devinfo *dinfo = device_get_ivars(child);
2758 	struct pcicfg_msi *msi = &dinfo->cfg.msi;
2759 
2760 	if (pci_do_msi && msi->msi_location != 0)
2761 		return (msi->msi_msgnum);
2762 	return (0);
2763 }
2764 
2765 /* free pcicfgregs structure and all depending data structures */
2766 
2767 int
pci_freecfg(struct pci_devinfo * dinfo)2768 pci_freecfg(struct pci_devinfo *dinfo)
2769 {
2770 	struct devlist *devlist_head;
2771 	struct pci_map *pm, *next;
2772 
2773 	devlist_head = &pci_devq;
2774 
2775 	if (dinfo->cfg.vpd.vpd_reg)
2776 		vpd_free(&dinfo->cfg.vpd);
2777 
2778 	STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2779 		free(pm, M_DEVBUF);
2780 	}
2781 	STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2782 	free(dinfo, M_DEVBUF);
2783 
2784 	/* increment the generation count */
2785 	pci_generation++;
2786 
2787 	/* we're losing one device */
2788 	pci_numdevs--;
2789 	return (0);
2790 }
2791 
2792 /*
2793  * PCI power manangement
2794  */
2795 int
pci_set_powerstate_method(device_t dev,device_t child,int state)2796 pci_set_powerstate_method(device_t dev, device_t child, int state)
2797 {
2798 	struct pci_devinfo *dinfo = device_get_ivars(child);
2799 	pcicfgregs *cfg = &dinfo->cfg;
2800 	uint16_t status;
2801 	int oldstate, highest, delay;
2802 
2803 	if (cfg->pp.pp_cap == 0)
2804 		return (EOPNOTSUPP);
2805 
2806 	/*
2807 	 * Optimize a no state change request away.  While it would be OK to
2808 	 * write to the hardware in theory, some devices have shown odd
2809 	 * behavior when going from D3 -> D3.
2810 	 */
2811 	oldstate = pci_get_powerstate(child);
2812 	if (oldstate == state)
2813 		return (0);
2814 
2815 	/*
2816 	 * The PCI power management specification states that after a state
2817 	 * transition between PCI power states, system software must
2818 	 * guarantee a minimal delay before the function accesses the device.
2819 	 * Compute the worst case delay that we need to guarantee before we
2820 	 * access the device.  Many devices will be responsive much more
2821 	 * quickly than this delay, but there are some that don't respond
2822 	 * instantly to state changes.  Transitions to/from D3 state require
2823 	 * 10ms, while D2 requires 200us, and D0/1 require none.  The delay
2824 	 * is done below with DELAY rather than a sleeper function because
2825 	 * this function can be called from contexts where we cannot sleep.
2826 	 */
2827 	highest = (oldstate > state) ? oldstate : state;
2828 	if (highest == PCI_POWERSTATE_D3)
2829 	    delay = 10000;
2830 	else if (highest == PCI_POWERSTATE_D2)
2831 	    delay = 200;
2832 	else
2833 	    delay = 0;
2834 	status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2835 	    & ~PCIM_PSTAT_DMASK;
2836 	switch (state) {
2837 	case PCI_POWERSTATE_D0:
2838 		status |= PCIM_PSTAT_D0;
2839 		break;
2840 	case PCI_POWERSTATE_D1:
2841 		if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2842 			return (EOPNOTSUPP);
2843 		status |= PCIM_PSTAT_D1;
2844 		break;
2845 	case PCI_POWERSTATE_D2:
2846 		if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2847 			return (EOPNOTSUPP);
2848 		status |= PCIM_PSTAT_D2;
2849 		break;
2850 	case PCI_POWERSTATE_D3:
2851 		status |= PCIM_PSTAT_D3;
2852 		break;
2853 	default:
2854 		return (EINVAL);
2855 	}
2856 
2857 	if (bootverbose)
2858 		pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2859 		    state);
2860 
2861 	PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2862 	if (delay)
2863 		DELAY(delay);
2864 	return (0);
2865 }
2866 
2867 int
pci_get_powerstate_method(device_t dev,device_t child)2868 pci_get_powerstate_method(device_t dev, device_t child)
2869 {
2870 	struct pci_devinfo *dinfo = device_get_ivars(child);
2871 	pcicfgregs *cfg = &dinfo->cfg;
2872 	uint16_t status;
2873 	int result;
2874 
2875 	if (cfg->pp.pp_cap != 0) {
2876 		status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2877 		switch (status & PCIM_PSTAT_DMASK) {
2878 		case PCIM_PSTAT_D0:
2879 			result = PCI_POWERSTATE_D0;
2880 			break;
2881 		case PCIM_PSTAT_D1:
2882 			result = PCI_POWERSTATE_D1;
2883 			break;
2884 		case PCIM_PSTAT_D2:
2885 			result = PCI_POWERSTATE_D2;
2886 			break;
2887 		case PCIM_PSTAT_D3:
2888 			result = PCI_POWERSTATE_D3;
2889 			break;
2890 		default:
2891 			result = PCI_POWERSTATE_UNKNOWN;
2892 			break;
2893 		}
2894 	} else {
2895 		/* No support, device is always at D0 */
2896 		result = PCI_POWERSTATE_D0;
2897 	}
2898 	return (result);
2899 }
2900 
2901 /*
2902  * Some convenience functions for PCI device drivers.
2903  */
2904 
2905 static __inline void
pci_set_command_bit(device_t dev,device_t child,uint16_t bit)2906 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2907 {
2908 	uint16_t	command;
2909 
2910 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2911 	command |= bit;
2912 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2913 }
2914 
2915 static __inline void
pci_clear_command_bit(device_t dev,device_t child,uint16_t bit)2916 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2917 {
2918 	uint16_t	command;
2919 
2920 	command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2921 	command &= ~bit;
2922 	PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2923 }
2924 
2925 int
pci_enable_busmaster_method(device_t dev,device_t child)2926 pci_enable_busmaster_method(device_t dev, device_t child)
2927 {
2928 	pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2929 	return (0);
2930 }
2931 
2932 int
pci_disable_busmaster_method(device_t dev,device_t child)2933 pci_disable_busmaster_method(device_t dev, device_t child)
2934 {
2935 	pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2936 	return (0);
2937 }
2938 
2939 int
pci_enable_io_method(device_t dev,device_t child,int space)2940 pci_enable_io_method(device_t dev, device_t child, int space)
2941 {
2942 	uint16_t bit;
2943 
2944 	switch(space) {
2945 	case SYS_RES_IOPORT:
2946 		bit = PCIM_CMD_PORTEN;
2947 		break;
2948 	case SYS_RES_MEMORY:
2949 		bit = PCIM_CMD_MEMEN;
2950 		break;
2951 	default:
2952 		return (EINVAL);
2953 	}
2954 	pci_set_command_bit(dev, child, bit);
2955 	return (0);
2956 }
2957 
2958 int
pci_disable_io_method(device_t dev,device_t child,int space)2959 pci_disable_io_method(device_t dev, device_t child, int space)
2960 {
2961 	uint16_t bit;
2962 
2963 	switch(space) {
2964 	case SYS_RES_IOPORT:
2965 		bit = PCIM_CMD_PORTEN;
2966 		break;
2967 	case SYS_RES_MEMORY:
2968 		bit = PCIM_CMD_MEMEN;
2969 		break;
2970 	default:
2971 		return (EINVAL);
2972 	}
2973 	pci_clear_command_bit(dev, child, bit);
2974 	return (0);
2975 }
2976 
2977 /*
2978  * New style pci driver.  Parent device is either a pci-host-bridge or a
2979  * pci-pci-bridge.  Both kinds are represented by instances of pcib.
2980  */
2981 
2982 void
pci_print_verbose(struct pci_devinfo * dinfo)2983 pci_print_verbose(struct pci_devinfo *dinfo)
2984 {
2985 
2986 	if (bootverbose) {
2987 		pcicfgregs *cfg = &dinfo->cfg;
2988 
2989 		printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2990 		    cfg->vendor, cfg->device, cfg->revid);
2991 		printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2992 		    cfg->domain, cfg->bus, cfg->slot, cfg->func);
2993 		printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2994 		    cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2995 		    cfg->mfdev);
2996 		printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2997 		    cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2998 		printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2999 		    cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
3000 		    cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
3001 		if (cfg->intpin > 0)
3002 			printf("\tintpin=%c, irq=%d\n",
3003 			    cfg->intpin +'a' -1, cfg->intline);
3004 		if (cfg->pp.pp_cap) {
3005 			uint16_t status;
3006 
3007 			status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
3008 			printf("\tpowerspec %d  supports D0%s%s D3  current D%d\n",
3009 			    cfg->pp.pp_cap & PCIM_PCAP_SPEC,
3010 			    cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
3011 			    cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
3012 			    status & PCIM_PSTAT_DMASK);
3013 		}
3014 		if (cfg->msi.msi_location) {
3015 			int ctrl;
3016 
3017 			ctrl = cfg->msi.msi_ctrl;
3018 			printf("\tMSI supports %d message%s%s%s\n",
3019 			    cfg->msi.msi_msgnum,
3020 			    (cfg->msi.msi_msgnum == 1) ? "" : "s",
3021 			    (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
3022 			    (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
3023 		}
3024 		if (cfg->msix.msix_location) {
3025 			printf("\tMSI-X supports %d message%s ",
3026 			    cfg->msix.msix_msgnum,
3027 			    (cfg->msix.msix_msgnum == 1) ? "" : "s");
3028 			if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
3029 				printf("in map 0x%x\n",
3030 				    cfg->msix.msix_table_bar);
3031 			else
3032 				printf("in maps 0x%x and 0x%x\n",
3033 				    cfg->msix.msix_table_bar,
3034 				    cfg->msix.msix_pba_bar);
3035 		}
3036 	}
3037 }
3038 
3039 static int
pci_porten(device_t dev)3040 pci_porten(device_t dev)
3041 {
3042 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
3043 }
3044 
3045 static int
pci_memen(device_t dev)3046 pci_memen(device_t dev)
3047 {
3048 	return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
3049 }
3050 
3051 void
pci_read_bar(device_t dev,int reg,pci_addr_t * mapp,pci_addr_t * testvalp,int * bar64)3052 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3053     int *bar64)
3054 {
3055 	struct pci_devinfo *dinfo;
3056 	pci_addr_t map, testval;
3057 	int ln2range;
3058 	uint16_t cmd;
3059 
3060 	/*
3061 	 * The device ROM BAR is special.  It is always a 32-bit
3062 	 * memory BAR.  Bit 0 is special and should not be set when
3063 	 * sizing the BAR.
3064 	 */
3065 	dinfo = device_get_ivars(dev);
3066 	if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3067 		map = pci_read_config(dev, reg, 4);
3068 		pci_write_config(dev, reg, 0xfffffffe, 4);
3069 		testval = pci_read_config(dev, reg, 4);
3070 		pci_write_config(dev, reg, map, 4);
3071 		*mapp = map;
3072 		*testvalp = testval;
3073 		if (bar64 != NULL)
3074 			*bar64 = 0;
3075 		return;
3076 	}
3077 
3078 	map = pci_read_config(dev, reg, 4);
3079 	ln2range = pci_maprange(map);
3080 	if (ln2range == 64)
3081 		map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3082 
3083 	/*
3084 	 * Disable decoding via the command register before
3085 	 * determining the BAR's length since we will be placing it in
3086 	 * a weird state.
3087 	 */
3088 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3089 	pci_write_config(dev, PCIR_COMMAND,
3090 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3091 
3092 	/*
3093 	 * Determine the BAR's length by writing all 1's.  The bottom
3094 	 * log_2(size) bits of the BAR will stick as 0 when we read
3095 	 * the value back.
3096 	 *
3097 	 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3098 	 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3099 	 * and combines the result into a 64-bit value." (section 6.2.5.1)
3100 	 *
3101 	 * Writes to both registers must be performed before attempting to
3102 	 * read back the size value.
3103 	 */
3104 	testval = 0;
3105 	pci_write_config(dev, reg, 0xffffffff, 4);
3106 	if (ln2range == 64) {
3107 		pci_write_config(dev, reg + 4, 0xffffffff, 4);
3108 		testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3109 	}
3110 	testval |= pci_read_config(dev, reg, 4);
3111 
3112 	/*
3113 	 * Restore the original value of the BAR.  We may have reprogrammed
3114 	 * the BAR of the low-level console device and when booting verbose,
3115 	 * we need the console device addressable.
3116 	 */
3117 	pci_write_config(dev, reg, map, 4);
3118 	if (ln2range == 64)
3119 		pci_write_config(dev, reg + 4, map >> 32, 4);
3120 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3121 
3122 	*mapp = map;
3123 	*testvalp = testval;
3124 	if (bar64 != NULL)
3125 		*bar64 = (ln2range == 64);
3126 }
3127 
3128 static void
pci_write_bar(device_t dev,struct pci_map * pm,pci_addr_t base)3129 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3130 {
3131 	struct pci_devinfo *dinfo;
3132 	int ln2range;
3133 
3134 	/* The device ROM BAR is always a 32-bit memory BAR. */
3135 	dinfo = device_get_ivars(dev);
3136 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3137 		ln2range = 32;
3138 	else
3139 		ln2range = pci_maprange(pm->pm_value);
3140 	pci_write_config(dev, pm->pm_reg, base, 4);
3141 	if (ln2range == 64)
3142 		pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3143 	pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3144 	if (ln2range == 64)
3145 		pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3146 		    pm->pm_reg + 4, 4) << 32;
3147 }
3148 
3149 struct pci_map *
pci_find_bar(device_t dev,int reg)3150 pci_find_bar(device_t dev, int reg)
3151 {
3152 	struct pci_devinfo *dinfo;
3153 	struct pci_map *pm;
3154 
3155 	dinfo = device_get_ivars(dev);
3156 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3157 		if (pm->pm_reg == reg)
3158 			return (pm);
3159 	}
3160 	return (NULL);
3161 }
3162 
3163 struct pci_map *
pci_first_bar(device_t dev)3164 pci_first_bar(device_t dev)
3165 {
3166 	struct pci_devinfo *dinfo;
3167 
3168 	dinfo = device_get_ivars(dev);
3169 	return (STAILQ_FIRST(&dinfo->cfg.maps));
3170 }
3171 
3172 struct pci_map *
pci_next_bar(struct pci_map * pm)3173 pci_next_bar(struct pci_map *pm)
3174 {
3175 	return (STAILQ_NEXT(pm, pm_link));
3176 }
3177 
3178 int
pci_bar_enabled(device_t dev,struct pci_map * pm)3179 pci_bar_enabled(device_t dev, struct pci_map *pm)
3180 {
3181 	struct pci_devinfo *dinfo;
3182 	uint16_t cmd;
3183 
3184 	dinfo = device_get_ivars(dev);
3185 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3186 	    !(pm->pm_value & PCIM_BIOS_ENABLE))
3187 		return (0);
3188 #ifdef PCI_IOV
3189 	if ((dinfo->cfg.flags & PCICFG_VF) != 0) {
3190 		struct pcicfg_iov *iov;
3191 
3192 		iov = dinfo->cfg.iov;
3193 		cmd = pci_read_config(iov->iov_pf,
3194 		    iov->iov_pos + PCIR_SRIOV_CTL, 2);
3195 		return ((cmd & PCIM_SRIOV_VF_MSE) != 0);
3196 	}
3197 #endif
3198 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3199 	if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3200 		return ((cmd & PCIM_CMD_MEMEN) != 0);
3201 	else
3202 		return ((cmd & PCIM_CMD_PORTEN) != 0);
3203 }
3204 
3205 struct pci_map *
pci_add_bar(device_t dev,int reg,pci_addr_t value,pci_addr_t size)3206 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3207 {
3208 	struct pci_devinfo *dinfo;
3209 	struct pci_map *pm, *prev;
3210 
3211 	dinfo = device_get_ivars(dev);
3212 	pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3213 	pm->pm_reg = reg;
3214 	pm->pm_value = value;
3215 	pm->pm_size = size;
3216 	STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3217 		KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3218 		    reg));
3219 		if (STAILQ_NEXT(prev, pm_link) == NULL ||
3220 		    STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3221 			break;
3222 	}
3223 	if (prev != NULL)
3224 		STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3225 	else
3226 		STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3227 	return (pm);
3228 }
3229 
3230 static void
pci_restore_bars(device_t dev)3231 pci_restore_bars(device_t dev)
3232 {
3233 	struct pci_devinfo *dinfo;
3234 	struct pci_map *pm;
3235 	int ln2range;
3236 
3237 	dinfo = device_get_ivars(dev);
3238 	STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3239 		if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3240 			ln2range = 32;
3241 		else
3242 			ln2range = pci_maprange(pm->pm_value);
3243 		pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3244 		if (ln2range == 64)
3245 			pci_write_config(dev, pm->pm_reg + 4,
3246 			    pm->pm_value >> 32, 4);
3247 	}
3248 }
3249 
3250 /*
3251  * Add a resource based on a pci map register. Return 1 if the map
3252  * register is a 32bit map register or 2 if it is a 64bit register.
3253  */
3254 static int
pci_add_map(device_t bus,device_t dev,int reg,struct resource_list * rl,int force,int prefetch)3255 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3256     int force, int prefetch)
3257 {
3258 	struct pci_map *pm;
3259 	pci_addr_t base, map, testval;
3260 	pci_addr_t start, end, count;
3261 	int barlen, basezero, flags, maprange, mapsize, type;
3262 	uint16_t cmd;
3263 	struct resource *res;
3264 
3265 	/*
3266 	 * The BAR may already exist if the device is a CardBus card
3267 	 * whose CIS is stored in this BAR.
3268 	 */
3269 	pm = pci_find_bar(dev, reg);
3270 	if (pm != NULL) {
3271 		maprange = pci_maprange(pm->pm_value);
3272 		barlen = maprange == 64 ? 2 : 1;
3273 		return (barlen);
3274 	}
3275 
3276 	pci_read_bar(dev, reg, &map, &testval, NULL);
3277 	if (PCI_BAR_MEM(map)) {
3278 		type = SYS_RES_MEMORY;
3279 		if (map & PCIM_BAR_MEM_PREFETCH)
3280 			prefetch = 1;
3281 	} else
3282 		type = SYS_RES_IOPORT;
3283 	mapsize = pci_mapsize(testval);
3284 	base = pci_mapbase(map);
3285 #ifdef __PCI_BAR_ZERO_VALID
3286 	basezero = 0;
3287 #else
3288 	basezero = base == 0;
3289 #endif
3290 	maprange = pci_maprange(map);
3291 	barlen = maprange == 64 ? 2 : 1;
3292 
3293 	/*
3294 	 * For I/O registers, if bottom bit is set, and the next bit up
3295 	 * isn't clear, we know we have a BAR that doesn't conform to the
3296 	 * spec, so ignore it.  Also, sanity check the size of the data
3297 	 * areas to the type of memory involved.  Memory must be at least
3298 	 * 16 bytes in size, while I/O ranges must be at least 4.
3299 	 */
3300 	if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3301 		return (barlen);
3302 	if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3303 	    (type == SYS_RES_IOPORT && mapsize < 2))
3304 		return (barlen);
3305 
3306 	/* Save a record of this BAR. */
3307 	pm = pci_add_bar(dev, reg, map, mapsize);
3308 	if (bootverbose) {
3309 		printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3310 		    reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3311 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3312 			printf(", port disabled\n");
3313 		else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3314 			printf(", memory disabled\n");
3315 		else
3316 			printf(", enabled\n");
3317 	}
3318 
3319 	/*
3320 	 * If base is 0, then we have problems if this architecture does
3321 	 * not allow that.  It is best to ignore such entries for the
3322 	 * moment.  These will be allocated later if the driver specifically
3323 	 * requests them.  However, some removable buses look better when
3324 	 * all resources are allocated, so allow '0' to be overridden.
3325 	 *
3326 	 * Similarly treat maps whose values is the same as the test value
3327 	 * read back.  These maps have had all f's written to them by the
3328 	 * BIOS in an attempt to disable the resources.
3329 	 */
3330 	if (!force && (basezero || map == testval))
3331 		return (barlen);
3332 	if ((u_long)base != base) {
3333 		device_printf(bus,
3334 		    "pci%d:%d:%d:%d bar %#x too many address bits",
3335 		    pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3336 		    pci_get_function(dev), reg);
3337 		return (barlen);
3338 	}
3339 
3340 	/*
3341 	 * This code theoretically does the right thing, but has
3342 	 * undesirable side effects in some cases where peripherals
3343 	 * respond oddly to having these bits enabled.  Let the user
3344 	 * be able to turn them off (since pci_enable_io_modes is 1 by
3345 	 * default).
3346 	 */
3347 	if (pci_enable_io_modes) {
3348 		/* Turn on resources that have been left off by a lazy BIOS */
3349 		if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3350 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3351 			cmd |= PCIM_CMD_PORTEN;
3352 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3353 		}
3354 		if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3355 			cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3356 			cmd |= PCIM_CMD_MEMEN;
3357 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3358 		}
3359 	} else {
3360 		if (type == SYS_RES_IOPORT && !pci_porten(dev))
3361 			return (barlen);
3362 		if (type == SYS_RES_MEMORY && !pci_memen(dev))
3363 			return (barlen);
3364 	}
3365 
3366 	count = (pci_addr_t)1 << mapsize;
3367 	flags = RF_ALIGNMENT_LOG2(mapsize);
3368 	if (prefetch)
3369 		flags |= RF_PREFETCHABLE;
3370 	if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3371 		start = 0;	/* Let the parent decide. */
3372 		end = ~0;
3373 	} else {
3374 		start = base;
3375 		end = base + count - 1;
3376 	}
3377 	resource_list_add(rl, type, reg, start, end, count);
3378 
3379 	/*
3380 	 * Try to allocate the resource for this BAR from our parent
3381 	 * so that this resource range is already reserved.  The
3382 	 * driver for this device will later inherit this resource in
3383 	 * pci_alloc_resource().
3384 	 */
3385 	res = resource_list_reserve(rl, bus, dev, type, &reg, start, end, count,
3386 	    flags);
3387 	if ((pci_do_realloc_bars
3388 		|| pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3389 	    && res == NULL && (start != 0 || end != ~0)) {
3390 		/*
3391 		 * If the allocation fails, try to allocate a resource for
3392 		 * this BAR using any available range.  The firmware felt
3393 		 * it was important enough to assign a resource, so don't
3394 		 * disable decoding if we can help it.
3395 		 */
3396 		resource_list_delete(rl, type, reg);
3397 		resource_list_add(rl, type, reg, 0, ~0, count);
3398 		res = resource_list_reserve(rl, bus, dev, type, &reg, 0, ~0,
3399 		    count, flags);
3400 	}
3401 	if (res == NULL) {
3402 		/*
3403 		 * If the allocation fails, delete the resource list entry
3404 		 * and disable decoding for this device.
3405 		 *
3406 		 * If the driver requests this resource in the future,
3407 		 * pci_reserve_map() will try to allocate a fresh
3408 		 * resource range.
3409 		 */
3410 		resource_list_delete(rl, type, reg);
3411 		pci_disable_io(dev, type);
3412 		if (bootverbose)
3413 			device_printf(bus,
3414 			    "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3415 			    pci_get_domain(dev), pci_get_bus(dev),
3416 			    pci_get_slot(dev), pci_get_function(dev), reg);
3417 	} else {
3418 		start = rman_get_start(res);
3419 		pci_write_bar(dev, pm, start);
3420 	}
3421 	return (barlen);
3422 }
3423 
3424 /*
3425  * For ATA devices we need to decide early what addressing mode to use.
3426  * Legacy demands that the primary and secondary ATA ports sits on the
3427  * same addresses that old ISA hardware did. This dictates that we use
3428  * those addresses and ignore the BAR's if we cannot set PCI native
3429  * addressing mode.
3430  */
3431 static void
pci_ata_maps(device_t bus,device_t dev,struct resource_list * rl,int force,uint32_t prefetchmask)3432 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3433     uint32_t prefetchmask)
3434 {
3435 	int rid, type, progif;
3436 #if 0
3437 	/* if this device supports PCI native addressing use it */
3438 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3439 	if ((progif & 0x8a) == 0x8a) {
3440 		if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3441 		    pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3442 			printf("Trying ATA native PCI addressing mode\n");
3443 			pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3444 		}
3445 	}
3446 #endif
3447 	progif = pci_read_config(dev, PCIR_PROGIF, 1);
3448 	type = SYS_RES_IOPORT;
3449 	if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3450 		pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3451 		    prefetchmask & (1 << 0));
3452 		pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3453 		    prefetchmask & (1 << 1));
3454 	} else {
3455 		rid = PCIR_BAR(0);
3456 		resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3457 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3458 		    0x1f7, 8, 0);
3459 		rid = PCIR_BAR(1);
3460 		resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3461 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3462 		    0x3f6, 1, 0);
3463 	}
3464 	if (progif & PCIP_STORAGE_IDE_MODESEC) {
3465 		pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3466 		    prefetchmask & (1 << 2));
3467 		pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3468 		    prefetchmask & (1 << 3));
3469 	} else {
3470 		rid = PCIR_BAR(2);
3471 		resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3472 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3473 		    0x177, 8, 0);
3474 		rid = PCIR_BAR(3);
3475 		resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3476 		(void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3477 		    0x376, 1, 0);
3478 	}
3479 	pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3480 	    prefetchmask & (1 << 4));
3481 	pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3482 	    prefetchmask & (1 << 5));
3483 }
3484 
3485 static void
pci_assign_interrupt(device_t bus,device_t dev,int force_route)3486 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3487 {
3488 	struct pci_devinfo *dinfo = device_get_ivars(dev);
3489 	pcicfgregs *cfg = &dinfo->cfg;
3490 	char tunable_name[64];
3491 	int irq;
3492 
3493 	/* Has to have an intpin to have an interrupt. */
3494 	if (cfg->intpin == 0)
3495 		return;
3496 
3497 	/* Let the user override the IRQ with a tunable. */
3498 	irq = PCI_INVALID_IRQ;
3499 	snprintf(tunable_name, sizeof(tunable_name),
3500 	    "hw.pci%d.%d.%d.INT%c.irq",
3501 	    cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3502 	if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3503 		irq = PCI_INVALID_IRQ;
3504 
3505 	/*
3506 	 * If we didn't get an IRQ via the tunable, then we either use the
3507 	 * IRQ value in the intline register or we ask the bus to route an
3508 	 * interrupt for us.  If force_route is true, then we only use the
3509 	 * value in the intline register if the bus was unable to assign an
3510 	 * IRQ.
3511 	 */
3512 	if (!PCI_INTERRUPT_VALID(irq)) {
3513 		if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3514 			irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3515 		if (!PCI_INTERRUPT_VALID(irq))
3516 			irq = cfg->intline;
3517 	}
3518 
3519 	/* If after all that we don't have an IRQ, just bail. */
3520 	if (!PCI_INTERRUPT_VALID(irq))
3521 		return;
3522 
3523 	/* Update the config register if it changed. */
3524 	if (irq != cfg->intline) {
3525 		cfg->intline = irq;
3526 		pci_write_config(dev, PCIR_INTLINE, irq, 1);
3527 	}
3528 
3529 	/* Add this IRQ as rid 0 interrupt resource. */
3530 	resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3531 }
3532 
3533 /* Perform early OHCI takeover from SMM. */
3534 static void
ohci_early_takeover(device_t self)3535 ohci_early_takeover(device_t self)
3536 {
3537 	struct resource *res;
3538 	uint32_t ctl;
3539 	int rid;
3540 	int i;
3541 
3542 	rid = PCIR_BAR(0);
3543 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3544 	if (res == NULL)
3545 		return;
3546 
3547 	ctl = bus_read_4(res, OHCI_CONTROL);
3548 	if (ctl & OHCI_IR) {
3549 		if (bootverbose)
3550 			printf("ohci early: "
3551 			    "SMM active, request owner change\n");
3552 		bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3553 		for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3554 			DELAY(1000);
3555 			ctl = bus_read_4(res, OHCI_CONTROL);
3556 		}
3557 		if (ctl & OHCI_IR) {
3558 			if (bootverbose)
3559 				printf("ohci early: "
3560 				    "SMM does not respond, resetting\n");
3561 			bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3562 		}
3563 		/* Disable interrupts */
3564 		bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3565 	}
3566 
3567 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3568 }
3569 
3570 /* Perform early UHCI takeover from SMM. */
3571 static void
uhci_early_takeover(device_t self)3572 uhci_early_takeover(device_t self)
3573 {
3574 	struct resource *res;
3575 	int rid;
3576 
3577 	/*
3578 	 * Set the PIRQD enable bit and switch off all the others. We don't
3579 	 * want legacy support to interfere with us XXX Does this also mean
3580 	 * that the BIOS won't touch the keyboard anymore if it is connected
3581 	 * to the ports of the root hub?
3582 	 */
3583 	pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3584 
3585 	/* Disable interrupts */
3586 	rid = PCI_UHCI_BASE_REG;
3587 	res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3588 	if (res != NULL) {
3589 		bus_write_2(res, UHCI_INTR, 0);
3590 		bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3591 	}
3592 }
3593 
3594 /* Perform early EHCI takeover from SMM. */
3595 static void
ehci_early_takeover(device_t self)3596 ehci_early_takeover(device_t self)
3597 {
3598 	struct resource *res;
3599 	uint32_t cparams;
3600 	uint32_t eec;
3601 	uint8_t eecp;
3602 	uint8_t bios_sem;
3603 	uint8_t offs;
3604 	int rid;
3605 	int i;
3606 
3607 	rid = PCIR_BAR(0);
3608 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3609 	if (res == NULL)
3610 		return;
3611 
3612 	cparams = bus_read_4(res, EHCI_HCCPARAMS);
3613 
3614 	/* Synchronise with the BIOS if it owns the controller. */
3615 	for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3616 	    eecp = EHCI_EECP_NEXT(eec)) {
3617 		eec = pci_read_config(self, eecp, 4);
3618 		if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3619 			continue;
3620 		}
3621 		bios_sem = pci_read_config(self, eecp +
3622 		    EHCI_LEGSUP_BIOS_SEM, 1);
3623 		if (bios_sem == 0) {
3624 			continue;
3625 		}
3626 		if (bootverbose)
3627 			printf("ehci early: "
3628 			    "SMM active, request owner change\n");
3629 
3630 		pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3631 
3632 		for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3633 			DELAY(1000);
3634 			bios_sem = pci_read_config(self, eecp +
3635 			    EHCI_LEGSUP_BIOS_SEM, 1);
3636 		}
3637 
3638 		if (bios_sem != 0) {
3639 			if (bootverbose)
3640 				printf("ehci early: "
3641 				    "SMM does not respond\n");
3642 		}
3643 		/* Disable interrupts */
3644 		offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3645 		bus_write_4(res, offs + EHCI_USBINTR, 0);
3646 	}
3647 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3648 }
3649 
3650 /* Perform early XHCI takeover from SMM. */
3651 static void
xhci_early_takeover(device_t self)3652 xhci_early_takeover(device_t self)
3653 {
3654 	struct resource *res;
3655 	uint32_t cparams;
3656 	uint32_t eec;
3657 	uint8_t eecp;
3658 	uint8_t bios_sem;
3659 	uint8_t offs;
3660 	int rid;
3661 	int i;
3662 
3663 	rid = PCIR_BAR(0);
3664 	res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3665 	if (res == NULL)
3666 		return;
3667 
3668 	cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3669 
3670 	eec = -1;
3671 
3672 	/* Synchronise with the BIOS if it owns the controller. */
3673 	for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3674 	    eecp += XHCI_XECP_NEXT(eec) << 2) {
3675 		eec = bus_read_4(res, eecp);
3676 
3677 		if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3678 			continue;
3679 
3680 		bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3681 		if (bios_sem == 0)
3682 			continue;
3683 
3684 		if (bootverbose)
3685 			printf("xhci early: "
3686 			    "SMM active, request owner change\n");
3687 
3688 		bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3689 
3690 		/* wait a maximum of 5 second */
3691 
3692 		for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3693 			DELAY(1000);
3694 			bios_sem = bus_read_1(res, eecp +
3695 			    XHCI_XECP_BIOS_SEM);
3696 		}
3697 
3698 		if (bios_sem != 0) {
3699 			if (bootverbose)
3700 				printf("xhci early: "
3701 				    "SMM does not respond\n");
3702 		}
3703 
3704 		/* Disable interrupts */
3705 		offs = bus_read_1(res, XHCI_CAPLENGTH);
3706 		bus_write_4(res, offs + XHCI_USBCMD, 0);
3707 		bus_read_4(res, offs + XHCI_USBSTS);
3708 	}
3709 	bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3710 }
3711 
3712 static void
pci_reserve_secbus(device_t bus,device_t dev,pcicfgregs * cfg,struct resource_list * rl)3713 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3714     struct resource_list *rl)
3715 {
3716 	struct resource *res;
3717 	char *cp;
3718 	rman_res_t start, end, count;
3719 	int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3720 
3721 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3722 	case PCIM_HDRTYPE_BRIDGE:
3723 		sec_reg = PCIR_SECBUS_1;
3724 		sub_reg = PCIR_SUBBUS_1;
3725 		break;
3726 	case PCIM_HDRTYPE_CARDBUS:
3727 		sec_reg = PCIR_SECBUS_2;
3728 		sub_reg = PCIR_SUBBUS_2;
3729 		break;
3730 	default:
3731 		return;
3732 	}
3733 
3734 	/*
3735 	 * If the existing bus range is valid, attempt to reserve it
3736 	 * from our parent.  If this fails for any reason, clear the
3737 	 * secbus and subbus registers.
3738 	 *
3739 	 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3740 	 * This would at least preserve the existing sec_bus if it is
3741 	 * valid.
3742 	 */
3743 	sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3744 	sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3745 
3746 	/* Quirk handling. */
3747 	switch (pci_get_devid(dev)) {
3748 	case 0x12258086:		/* Intel 82454KX/GX (Orion) */
3749 		sup_bus = pci_read_config(dev, 0x41, 1);
3750 		if (sup_bus != 0xff) {
3751 			sec_bus = sup_bus + 1;
3752 			sub_bus = sup_bus + 1;
3753 			PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3754 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3755 		}
3756 		break;
3757 
3758 	case 0x00dd10de:
3759 		/* Compaq R3000 BIOS sets wrong subordinate bus number. */
3760 		if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3761 			break;
3762 		if (strncmp(cp, "Compal", 6) != 0) {
3763 			freeenv(cp);
3764 			break;
3765 		}
3766 		freeenv(cp);
3767 		if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3768 			break;
3769 		if (strncmp(cp, "08A0", 4) != 0) {
3770 			freeenv(cp);
3771 			break;
3772 		}
3773 		freeenv(cp);
3774 		if (sub_bus < 0xa) {
3775 			sub_bus = 0xa;
3776 			PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3777 		}
3778 		break;
3779 	}
3780 
3781 	if (bootverbose)
3782 		printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3783 	if (sec_bus > 0 && sub_bus >= sec_bus) {
3784 		start = sec_bus;
3785 		end = sub_bus;
3786 		count = end - start + 1;
3787 
3788 		resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3789 
3790 		/*
3791 		 * If requested, clear secondary bus registers in
3792 		 * bridge devices to force a complete renumbering
3793 		 * rather than reserving the existing range.  However,
3794 		 * preserve the existing size.
3795 		 */
3796 		if (pci_clear_buses)
3797 			goto clear;
3798 
3799 		rid = 0;
3800 		res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3801 		    start, end, count, 0);
3802 		if (res != NULL)
3803 			return;
3804 
3805 		if (bootverbose)
3806 			device_printf(bus,
3807 			    "pci%d:%d:%d:%d secbus failed to allocate\n",
3808 			    pci_get_domain(dev), pci_get_bus(dev),
3809 			    pci_get_slot(dev), pci_get_function(dev));
3810 	}
3811 
3812 clear:
3813 	PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3814 	PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3815 }
3816 
3817 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)3818 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3819     rman_res_t end, rman_res_t count, u_int flags)
3820 {
3821 	struct pci_devinfo *dinfo;
3822 	pcicfgregs *cfg;
3823 	struct resource_list *rl;
3824 	struct resource *res;
3825 	int sec_reg, sub_reg;
3826 
3827 	dinfo = device_get_ivars(child);
3828 	cfg = &dinfo->cfg;
3829 	rl = &dinfo->resources;
3830 	switch (cfg->hdrtype & PCIM_HDRTYPE) {
3831 	case PCIM_HDRTYPE_BRIDGE:
3832 		sec_reg = PCIR_SECBUS_1;
3833 		sub_reg = PCIR_SUBBUS_1;
3834 		break;
3835 	case PCIM_HDRTYPE_CARDBUS:
3836 		sec_reg = PCIR_SECBUS_2;
3837 		sub_reg = PCIR_SUBBUS_2;
3838 		break;
3839 	default:
3840 		return (NULL);
3841 	}
3842 
3843 	if (*rid != 0)
3844 		return (NULL);
3845 
3846 	if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3847 		resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3848 	if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3849 		res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3850 		    start, end, count, flags & ~RF_ACTIVE);
3851 		if (res == NULL) {
3852 			resource_list_delete(rl, PCI_RES_BUS, *rid);
3853 			device_printf(child, "allocating %ju bus%s failed\n",
3854 			    count, count == 1 ? "" : "es");
3855 			return (NULL);
3856 		}
3857 		if (bootverbose)
3858 			device_printf(child,
3859 			    "Lazy allocation of %ju bus%s at %ju\n", count,
3860 			    count == 1 ? "" : "es", rman_get_start(res));
3861 		PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3862 		PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3863 	}
3864 	return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3865 	    end, count, flags));
3866 }
3867 
3868 static int
pci_ea_bei_to_rid(device_t dev,int bei)3869 pci_ea_bei_to_rid(device_t dev, int bei)
3870 {
3871 #ifdef PCI_IOV
3872 	struct pci_devinfo *dinfo;
3873 	int iov_pos;
3874 	struct pcicfg_iov *iov;
3875 
3876 	dinfo = device_get_ivars(dev);
3877 	iov = dinfo->cfg.iov;
3878 	if (iov != NULL)
3879 		iov_pos = iov->iov_pos;
3880 	else
3881 		iov_pos = 0;
3882 #endif
3883 
3884 	/* Check if matches BAR */
3885 	if ((bei >= PCIM_EA_BEI_BAR_0) &&
3886 	    (bei <= PCIM_EA_BEI_BAR_5))
3887 		return (PCIR_BAR(bei));
3888 
3889 	/* Check ROM */
3890 	if (bei == PCIM_EA_BEI_ROM)
3891 		return (PCIR_BIOS);
3892 
3893 #ifdef PCI_IOV
3894 	/* Check if matches VF_BAR */
3895 	if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3896 	    (bei <= PCIM_EA_BEI_VF_BAR_5))
3897 		return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3898 		    iov_pos);
3899 #endif
3900 
3901 	return (-1);
3902 }
3903 
3904 int
pci_ea_is_enabled(device_t dev,int rid)3905 pci_ea_is_enabled(device_t dev, int rid)
3906 {
3907 	struct pci_ea_entry *ea;
3908 	struct pci_devinfo *dinfo;
3909 
3910 	dinfo = device_get_ivars(dev);
3911 
3912 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3913 		if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3914 			return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3915 	}
3916 
3917 	return (0);
3918 }
3919 
3920 void
pci_add_resources_ea(device_t bus,device_t dev,int alloc_iov)3921 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3922 {
3923 	struct pci_ea_entry *ea;
3924 	struct pci_devinfo *dinfo;
3925 	pci_addr_t start, end, count;
3926 	struct resource_list *rl;
3927 	int type, flags, rid;
3928 	struct resource *res;
3929 	uint32_t tmp;
3930 #ifdef PCI_IOV
3931 	struct pcicfg_iov *iov;
3932 #endif
3933 
3934 	dinfo = device_get_ivars(dev);
3935 	rl = &dinfo->resources;
3936 	flags = 0;
3937 
3938 #ifdef PCI_IOV
3939 	iov = dinfo->cfg.iov;
3940 #endif
3941 
3942 	if (dinfo->cfg.ea.ea_location == 0)
3943 		return;
3944 
3945 	STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3946 		/*
3947 		 * TODO: Ignore EA-BAR if is not enabled.
3948 		 *   Currently the EA implementation supports
3949 		 *   only situation, where EA structure contains
3950 		 *   predefined entries. In case they are not enabled
3951 		 *   leave them unallocated and proceed with
3952 		 *   a legacy-BAR mechanism.
3953 		 */
3954 		if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3955 			continue;
3956 
3957 		switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3958 		case PCIM_EA_P_MEM_PREFETCH:
3959 		case PCIM_EA_P_VF_MEM_PREFETCH:
3960 			flags = RF_PREFETCHABLE;
3961 			/* FALLTHROUGH */
3962 		case PCIM_EA_P_VF_MEM:
3963 		case PCIM_EA_P_MEM:
3964 			type = SYS_RES_MEMORY;
3965 			break;
3966 		case PCIM_EA_P_IO:
3967 			type = SYS_RES_IOPORT;
3968 			break;
3969 		default:
3970 			continue;
3971 		}
3972 
3973 		if (alloc_iov != 0) {
3974 #ifdef PCI_IOV
3975 			/* Allocating IOV, confirm BEI matches */
3976 			if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3977 			    (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3978 				continue;
3979 #else
3980 			continue;
3981 #endif
3982 		} else {
3983 			/* Allocating BAR, confirm BEI matches */
3984 			if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3985 			    (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3986 			    (ea->eae_bei != PCIM_EA_BEI_ROM))
3987 				continue;
3988 		}
3989 
3990 		rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3991 		if (rid < 0)
3992 			continue;
3993 
3994 		/* Skip resources already allocated by EA */
3995 		if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3996 		    (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3997 			continue;
3998 
3999 		start = ea->eae_base;
4000 		count = ea->eae_max_offset + 1;
4001 #ifdef PCI_IOV
4002 		if (iov != NULL)
4003 			count = count * iov->iov_num_vfs;
4004 #endif
4005 		end = start + count - 1;
4006 		if (count == 0)
4007 			continue;
4008 
4009 		resource_list_add(rl, type, rid, start, end, count);
4010 		res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
4011 		    flags);
4012 		if (res == NULL) {
4013 			resource_list_delete(rl, type, rid);
4014 
4015 			/*
4016 			 * Failed to allocate using EA, disable entry.
4017 			 * Another attempt to allocation will be performed
4018 			 * further, but this time using legacy BAR registers
4019 			 */
4020 			tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
4021 			tmp &= ~PCIM_EA_ENABLE;
4022 			pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
4023 
4024 			/*
4025 			 * Disabling entry might fail in case it is hardwired.
4026 			 * Read flags again to match current status.
4027 			 */
4028 			ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
4029 
4030 			continue;
4031 		}
4032 
4033 		/* As per specification, fill BAR with zeros */
4034 		pci_write_config(dev, rid, 0, 4);
4035 	}
4036 }
4037 
4038 void
pci_add_resources(device_t bus,device_t dev,int force,uint32_t prefetchmask)4039 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
4040 {
4041 	struct pci_devinfo *dinfo;
4042 	pcicfgregs *cfg;
4043 	struct resource_list *rl;
4044 	const struct pci_quirk *q;
4045 	uint32_t devid;
4046 	int i;
4047 
4048 	dinfo = device_get_ivars(dev);
4049 	cfg = &dinfo->cfg;
4050 	rl = &dinfo->resources;
4051 	devid = (cfg->device << 16) | cfg->vendor;
4052 
4053 	/* Allocate resources using Enhanced Allocation */
4054 	pci_add_resources_ea(bus, dev, 0);
4055 
4056 	/* ATA devices needs special map treatment */
4057 	if ((pci_get_class(dev) == PCIC_STORAGE) &&
4058 	    (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
4059 	    ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
4060 	     (!pci_read_config(dev, PCIR_BAR(0), 4) &&
4061 	      !pci_read_config(dev, PCIR_BAR(2), 4))) )
4062 		pci_ata_maps(bus, dev, rl, force, prefetchmask);
4063 	else
4064 		for (i = 0; i < cfg->nummaps;) {
4065 			/* Skip resources already managed by EA */
4066 			if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
4067 			    (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
4068 			    pci_ea_is_enabled(dev, PCIR_BAR(i))) {
4069 				i++;
4070 				continue;
4071 			}
4072 
4073 			/*
4074 			 * Skip quirked resources.
4075 			 */
4076 			for (q = &pci_quirks[0]; q->devid != 0; q++)
4077 				if (q->devid == devid &&
4078 				    q->type == PCI_QUIRK_UNMAP_REG &&
4079 				    q->arg1 == PCIR_BAR(i))
4080 					break;
4081 			if (q->devid != 0) {
4082 				i++;
4083 				continue;
4084 			}
4085 			i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4086 			    prefetchmask & (1 << i));
4087 		}
4088 
4089 	/*
4090 	 * Add additional, quirked resources.
4091 	 */
4092 	for (q = &pci_quirks[0]; q->devid != 0; q++)
4093 		if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4094 			pci_add_map(bus, dev, q->arg1, rl, force, 0);
4095 
4096 	if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4097 		/*
4098 		 * Try to re-route interrupts. Sometimes the BIOS or
4099 		 * firmware may leave bogus values in these registers.
4100 		 * If the re-route fails, then just stick with what we
4101 		 * have.
4102 		 */
4103 		pci_assign_interrupt(bus, dev, 1);
4104 	}
4105 
4106 	if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4107 	    pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4108 		if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4109 			xhci_early_takeover(dev);
4110 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4111 			ehci_early_takeover(dev);
4112 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4113 			ohci_early_takeover(dev);
4114 		else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4115 			uhci_early_takeover(dev);
4116 	}
4117 
4118 	/*
4119 	 * Reserve resources for secondary bus ranges behind bridge
4120 	 * devices.
4121 	 */
4122 	pci_reserve_secbus(bus, dev, cfg, rl);
4123 }
4124 
4125 static struct pci_devinfo *
pci_identify_function(device_t pcib,device_t dev,int domain,int busno,int slot,int func)4126 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4127     int slot, int func)
4128 {
4129 	struct pci_devinfo *dinfo;
4130 
4131 	dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4132 	if (dinfo != NULL)
4133 		pci_add_child(dev, dinfo);
4134 
4135 	return (dinfo);
4136 }
4137 
4138 void
pci_add_children(device_t dev,int domain,int busno)4139 pci_add_children(device_t dev, int domain, int busno)
4140 {
4141 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4142 	device_t pcib = device_get_parent(dev);
4143 	struct pci_devinfo *dinfo;
4144 	int maxslots;
4145 	int s, f, pcifunchigh;
4146 	uint8_t hdrtype;
4147 	int first_func;
4148 
4149 	/*
4150 	 * Try to detect a device at slot 0, function 0.  If it exists, try to
4151 	 * enable ARI.  We must enable ARI before detecting the rest of the
4152 	 * functions on this bus as ARI changes the set of slots and functions
4153 	 * that are legal on this bus.
4154 	 */
4155 	dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4156 	if (dinfo != NULL && pci_enable_ari)
4157 		PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4158 
4159 	/*
4160 	 * Start looking for new devices on slot 0 at function 1 because we
4161 	 * just identified the device at slot 0, function 0.
4162 	 */
4163 	first_func = 1;
4164 
4165 	maxslots = PCIB_MAXSLOTS(pcib);
4166 	for (s = 0; s <= maxslots; s++, first_func = 0) {
4167 		pcifunchigh = 0;
4168 		f = 0;
4169 		DELAY(1);
4170 
4171 		/* If function 0 is not present, skip to the next slot. */
4172 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4173 			continue;
4174 		hdrtype = REG(PCIR_HDRTYPE, 1);
4175 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4176 			continue;
4177 		if (hdrtype & PCIM_MFDEV)
4178 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4179 		for (f = first_func; f <= pcifunchigh; f++)
4180 			pci_identify_function(pcib, dev, domain, busno, s, f);
4181 	}
4182 #undef REG
4183 }
4184 
4185 int
pci_rescan_method(device_t dev)4186 pci_rescan_method(device_t dev)
4187 {
4188 #define	REG(n, w)	PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4189 	device_t pcib = device_get_parent(dev);
4190 	device_t child, *devlist, *unchanged;
4191 	int devcount, error, i, j, maxslots, oldcount;
4192 	int busno, domain, s, f, pcifunchigh;
4193 	uint8_t hdrtype;
4194 
4195 	/* No need to check for ARI on a rescan. */
4196 	error = device_get_children(dev, &devlist, &devcount);
4197 	if (error)
4198 		return (error);
4199 	if (devcount != 0) {
4200 		unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4201 		    M_NOWAIT | M_ZERO);
4202 		if (unchanged == NULL) {
4203 			free(devlist, M_TEMP);
4204 			return (ENOMEM);
4205 		}
4206 	} else
4207 		unchanged = NULL;
4208 
4209 	domain = pcib_get_domain(dev);
4210 	busno = pcib_get_bus(dev);
4211 	maxslots = PCIB_MAXSLOTS(pcib);
4212 	for (s = 0; s <= maxslots; s++) {
4213 		/* If function 0 is not present, skip to the next slot. */
4214 		f = 0;
4215 		if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4216 			continue;
4217 		pcifunchigh = 0;
4218 		hdrtype = REG(PCIR_HDRTYPE, 1);
4219 		if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4220 			continue;
4221 		if (hdrtype & PCIM_MFDEV)
4222 			pcifunchigh = PCIB_MAXFUNCS(pcib);
4223 		for (f = 0; f <= pcifunchigh; f++) {
4224 			if (REG(PCIR_VENDOR, 2) == PCIV_INVALID)
4225 				continue;
4226 
4227 			/*
4228 			 * Found a valid function.  Check if a
4229 			 * device_t for this device already exists.
4230 			 */
4231 			for (i = 0; i < devcount; i++) {
4232 				child = devlist[i];
4233 				if (child == NULL)
4234 					continue;
4235 				if (pci_get_slot(child) == s &&
4236 				    pci_get_function(child) == f) {
4237 					unchanged[i] = child;
4238 					goto next_func;
4239 				}
4240 			}
4241 
4242 			pci_identify_function(pcib, dev, domain, busno, s, f);
4243 		next_func:;
4244 		}
4245 	}
4246 
4247 	/* Remove devices that are no longer present. */
4248 	for (i = 0; i < devcount; i++) {
4249 		if (unchanged[i] != NULL)
4250 			continue;
4251 		device_delete_child(dev, devlist[i]);
4252 	}
4253 
4254 	free(devlist, M_TEMP);
4255 	oldcount = devcount;
4256 
4257 	/* Try to attach the devices just added. */
4258 	error = device_get_children(dev, &devlist, &devcount);
4259 	if (error) {
4260 		free(unchanged, M_TEMP);
4261 		return (error);
4262 	}
4263 
4264 	for (i = 0; i < devcount; i++) {
4265 		for (j = 0; j < oldcount; j++) {
4266 			if (devlist[i] == unchanged[j])
4267 				goto next_device;
4268 		}
4269 
4270 		device_probe_and_attach(devlist[i]);
4271 	next_device:;
4272 	}
4273 
4274 	free(unchanged, M_TEMP);
4275 	free(devlist, M_TEMP);
4276 	return (0);
4277 #undef REG
4278 }
4279 
4280 #ifdef PCI_IOV
4281 device_t
pci_add_iov_child(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4282 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4283     uint16_t did)
4284 {
4285 	struct pci_devinfo *vf_dinfo;
4286 	device_t pcib;
4287 	int busno, slot, func;
4288 
4289 	pcib = device_get_parent(bus);
4290 
4291 	PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4292 
4293 	vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4294 	    slot, func, vid, did);
4295 
4296 	vf_dinfo->cfg.flags |= PCICFG_VF;
4297 	pci_add_child(bus, vf_dinfo);
4298 
4299 	return (vf_dinfo->cfg.dev);
4300 }
4301 
4302 device_t
pci_create_iov_child_method(device_t bus,device_t pf,uint16_t rid,uint16_t vid,uint16_t did)4303 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4304     uint16_t vid, uint16_t did)
4305 {
4306 
4307 	return (pci_add_iov_child(bus, pf, rid, vid, did));
4308 }
4309 #endif
4310 
4311 /*
4312  * For PCIe device set Max_Payload_Size to match PCIe root's.
4313  */
4314 static void
pcie_setup_mps(device_t dev)4315 pcie_setup_mps(device_t dev)
4316 {
4317 	struct pci_devinfo *dinfo = device_get_ivars(dev);
4318 	device_t root;
4319 	uint16_t rmps, mmps, mps;
4320 
4321 	if (dinfo->cfg.pcie.pcie_location == 0)
4322 		return;
4323 	root = pci_find_pcie_root_port(dev);
4324 	if (root == NULL)
4325 		return;
4326 	/* Check whether the MPS is already configured. */
4327 	rmps = pcie_read_config(root, PCIER_DEVICE_CTL, 2) &
4328 	    PCIEM_CTL_MAX_PAYLOAD;
4329 	mps = pcie_read_config(dev, PCIER_DEVICE_CTL, 2) &
4330 	    PCIEM_CTL_MAX_PAYLOAD;
4331 	if (mps == rmps)
4332 		return;
4333 	/* Check whether the device is capable of the root's MPS. */
4334 	mmps = (pcie_read_config(dev, PCIER_DEVICE_CAP, 2) &
4335 	    PCIEM_CAP_MAX_PAYLOAD) << 5;
4336 	if (rmps > mmps) {
4337 		/*
4338 		 * The device is unable to handle root's MPS.  Limit root.
4339 		 * XXX: We should traverse through all the tree, applying
4340 		 * it to all the devices.
4341 		 */
4342 		pcie_adjust_config(root, PCIER_DEVICE_CTL,
4343 		    PCIEM_CTL_MAX_PAYLOAD, mmps, 2);
4344 	} else {
4345 		pcie_adjust_config(dev, PCIER_DEVICE_CTL,
4346 		    PCIEM_CTL_MAX_PAYLOAD, rmps, 2);
4347 	}
4348 }
4349 
4350 static void
pci_add_child_clear_aer(device_t dev,struct pci_devinfo * dinfo)4351 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4352 {
4353 	int aer;
4354 	uint32_t r;
4355 	uint16_t r2;
4356 
4357 	if (dinfo->cfg.pcie.pcie_location != 0 &&
4358 	    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4359 		r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4360 		    PCIER_ROOT_CTL, 2);
4361 		r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4362 		    PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4363 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4364 		    PCIER_ROOT_CTL, r2, 2);
4365 	}
4366 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4367 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4368 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4369 		if (r != 0 && bootverbose) {
4370 			pci_printf(&dinfo->cfg,
4371 			    "clearing AER UC 0x%08x -> 0x%08x\n",
4372 			    r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4373 			    4));
4374 		}
4375 
4376 		r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4377 		r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4378 		    PCIM_AER_UC_DL_PROTOCOL_ERROR |
4379 		    PCIM_AER_UC_SURPRISE_LINK_DOWN |
4380 		    PCIM_AER_UC_POISONED_TLP |
4381 		    PCIM_AER_UC_FC_PROTOCOL_ERROR |
4382 		    PCIM_AER_UC_COMPLETION_TIMEOUT |
4383 		    PCIM_AER_UC_COMPLETER_ABORT |
4384 		    PCIM_AER_UC_UNEXPECTED_COMPLETION |
4385 		    PCIM_AER_UC_RECEIVER_OVERFLOW |
4386 		    PCIM_AER_UC_MALFORMED_TLP |
4387 		    PCIM_AER_UC_ECRC_ERROR |
4388 		    PCIM_AER_UC_UNSUPPORTED_REQUEST |
4389 		    PCIM_AER_UC_ACS_VIOLATION |
4390 		    PCIM_AER_UC_INTERNAL_ERROR |
4391 		    PCIM_AER_UC_MC_BLOCKED_TLP |
4392 		    PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4393 		    PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4394 		pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4395 
4396 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4397 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4398 		if (r != 0 && bootverbose) {
4399 			pci_printf(&dinfo->cfg,
4400 			    "clearing AER COR 0x%08x -> 0x%08x\n",
4401 			    r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4402 			    4));
4403 		}
4404 
4405 		r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4406 		r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4407 		    PCIM_AER_COR_BAD_TLP |
4408 		    PCIM_AER_COR_BAD_DLLP |
4409 		    PCIM_AER_COR_REPLAY_ROLLOVER |
4410 		    PCIM_AER_COR_REPLAY_TIMEOUT |
4411 		    PCIM_AER_COR_ADVISORY_NF_ERROR |
4412 		    PCIM_AER_COR_INTERNAL_ERROR |
4413 		    PCIM_AER_COR_HEADER_LOG_OVFLOW);
4414 		pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4415 
4416 		r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4417 		    PCIER_DEVICE_CTL, 2);
4418 		r |=  PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4419 		    PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4420 		pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4421 		    PCIER_DEVICE_CTL, r, 2);
4422 	}
4423 }
4424 
4425 void
pci_add_child(device_t bus,struct pci_devinfo * dinfo)4426 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4427 {
4428 	device_t dev;
4429 
4430 	dinfo->cfg.dev = dev = device_add_child(bus, NULL, DEVICE_UNIT_ANY);
4431 	device_set_ivars(dev, dinfo);
4432 	resource_list_init(&dinfo->resources);
4433 	pci_cfg_save(dev, dinfo, 0);
4434 	pci_cfg_restore(dev, dinfo);
4435 	pci_print_verbose(dinfo);
4436 	pci_add_resources(bus, dev, 0, 0);
4437 	if (pci_enable_mps_tune)
4438 		pcie_setup_mps(dev);
4439 	pci_child_added(dinfo->cfg.dev);
4440 
4441 	if (pci_clear_aer_on_attach)
4442 		pci_add_child_clear_aer(dev, dinfo);
4443 
4444 	EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4445 }
4446 
4447 void
pci_child_added_method(device_t dev,device_t child)4448 pci_child_added_method(device_t dev, device_t child)
4449 {
4450 
4451 }
4452 
4453 static int
pci_probe(device_t dev)4454 pci_probe(device_t dev)
4455 {
4456 
4457 	device_set_desc(dev, "PCI bus");
4458 
4459 	/* Allow other subclasses to override this driver. */
4460 	return (BUS_PROBE_GENERIC);
4461 }
4462 
4463 int
pci_attach_common(device_t dev)4464 pci_attach_common(device_t dev)
4465 {
4466 	struct pci_softc *sc;
4467 	int busno, domain;
4468 	int rid;
4469 
4470 	sc = device_get_softc(dev);
4471 	domain = pcib_get_domain(dev);
4472 	busno = pcib_get_bus(dev);
4473 	rid = 0;
4474 	sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4475 	    1, 0);
4476 	if (sc->sc_bus == NULL) {
4477 		device_printf(dev, "failed to allocate bus number\n");
4478 		return (ENXIO);
4479 	}
4480 	if (bootverbose)
4481 		device_printf(dev, "domain=%d, physical bus=%d\n",
4482 		    domain, busno);
4483 	sc->sc_dma_tag = bus_get_dma_tag(dev);
4484 	return (0);
4485 }
4486 
4487 int
pci_attach(device_t dev)4488 pci_attach(device_t dev)
4489 {
4490 	int busno, domain, error;
4491 
4492 	error = pci_attach_common(dev);
4493 	if (error)
4494 		return (error);
4495 
4496 	/*
4497 	 * Since there can be multiple independently numbered PCI
4498 	 * buses on systems with multiple PCI domains, we can't use
4499 	 * the unit number to decide which bus we are probing. We ask
4500 	 * the parent pcib what our domain and bus numbers are.
4501 	 */
4502 	domain = pcib_get_domain(dev);
4503 	busno = pcib_get_bus(dev);
4504 	pci_add_children(dev, domain, busno);
4505 	bus_attach_children(dev);
4506 	return (0);
4507 }
4508 
4509 int
pci_detach(device_t dev)4510 pci_detach(device_t dev)
4511 {
4512 	struct pci_softc *sc;
4513 	int error;
4514 
4515 	error = bus_generic_detach(dev);
4516 	if (error)
4517 		return (error);
4518 	sc = device_get_softc(dev);
4519 	error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4520 	return (error);
4521 }
4522 
4523 static void
pci_hint_device_unit(device_t dev,device_t child,const char * name,int * unitp)4524 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4525 {
4526 	int line, unit;
4527 	const char *at;
4528 	char me1[24], me2[32];
4529 	uint8_t b, s, f;
4530 	uint32_t d;
4531 	device_location_cache_t *cache;
4532 
4533 	d = pci_get_domain(child);
4534 	b = pci_get_bus(child);
4535 	s = pci_get_slot(child);
4536 	f = pci_get_function(child);
4537 	snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4538 	snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4539 	line = 0;
4540 	cache = dev_wired_cache_init();
4541 	while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4542 		resource_string_value(name, unit, "at", &at);
4543 		if (strcmp(at, me1) == 0 || strcmp(at, me2) == 0) {
4544 			*unitp = unit;
4545 			break;
4546 		}
4547 		if (dev_wired_cache_match(cache, child, at)) {
4548 			*unitp = unit;
4549 			break;
4550 		}
4551 	}
4552 	dev_wired_cache_fini(cache);
4553 }
4554 
4555 static void
pci_set_power_child(device_t dev,device_t child,int state)4556 pci_set_power_child(device_t dev, device_t child, int state)
4557 {
4558 	device_t pcib;
4559 	int dstate;
4560 
4561 	/*
4562 	 * Set the device to the given state.  If the firmware suggests
4563 	 * a different power state, use it instead.  If power management
4564 	 * is not present, the firmware is responsible for managing
4565 	 * device power.  Skip children who aren't attached since they
4566 	 * are handled separately.
4567 	 */
4568 	pcib = device_get_parent(dev);
4569 	dstate = state;
4570 	if (device_is_attached(child) &&
4571 	    PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4572 		pci_set_powerstate(child, dstate);
4573 }
4574 
4575 int
pci_suspend_child(device_t dev,device_t child)4576 pci_suspend_child(device_t dev, device_t child)
4577 {
4578 	struct pci_devinfo *dinfo;
4579 	struct resource_list_entry *rle;
4580 	int error;
4581 
4582 	dinfo = device_get_ivars(child);
4583 
4584 	/*
4585 	 * Save the PCI configuration space for the child and set the
4586 	 * device in the appropriate power state for this sleep state.
4587 	 */
4588 	pci_cfg_save(child, dinfo, 0);
4589 
4590 	/* Suspend devices before potentially powering them down. */
4591 	error = bus_generic_suspend_child(dev, child);
4592 
4593 	if (error)
4594 		return (error);
4595 
4596 	if (pci_do_power_suspend) {
4597 		/*
4598 		 * Make sure this device's interrupt handler is not invoked
4599 		 * in the case the device uses a shared interrupt that can
4600 		 * be raised by some other device.
4601 		 * This is applicable only to regular (legacy) PCI interrupts
4602 		 * as MSI/MSI-X interrupts are never shared.
4603 		 */
4604 		rle = resource_list_find(&dinfo->resources,
4605 		    SYS_RES_IRQ, 0);
4606 		if (rle != NULL && rle->res != NULL)
4607 			(void)bus_suspend_intr(child, rle->res);
4608 		pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4609 	}
4610 
4611 	return (0);
4612 }
4613 
4614 int
pci_resume_child(device_t dev,device_t child)4615 pci_resume_child(device_t dev, device_t child)
4616 {
4617 	struct pci_devinfo *dinfo;
4618 	struct resource_list_entry *rle;
4619 
4620 	if (pci_do_power_resume)
4621 		pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4622 
4623 	dinfo = device_get_ivars(child);
4624 	pci_cfg_restore(child, dinfo);
4625 	if (!device_is_attached(child))
4626 		pci_cfg_save(child, dinfo, 1);
4627 
4628 	bus_generic_resume_child(dev, child);
4629 
4630 	/*
4631 	 * Allow interrupts only after fully resuming the driver and hardware.
4632 	 */
4633 	if (pci_do_power_suspend) {
4634 		/* See pci_suspend_child for details. */
4635 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4636 		if (rle != NULL && rle->res != NULL)
4637 			(void)bus_resume_intr(child, rle->res);
4638 	}
4639 
4640 	return (0);
4641 }
4642 
4643 int
pci_resume(device_t dev)4644 pci_resume(device_t dev)
4645 {
4646 	device_t child, *devlist;
4647 	int error, i, numdevs;
4648 
4649 	if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4650 		return (error);
4651 
4652 	/*
4653 	 * Resume critical devices first, then everything else later.
4654 	 */
4655 	for (i = 0; i < numdevs; i++) {
4656 		child = devlist[i];
4657 		switch (pci_get_class(child)) {
4658 		case PCIC_DISPLAY:
4659 		case PCIC_MEMORY:
4660 		case PCIC_BRIDGE:
4661 		case PCIC_BASEPERIPH:
4662 			BUS_RESUME_CHILD(dev, child);
4663 			break;
4664 		}
4665 	}
4666 	for (i = 0; i < numdevs; i++) {
4667 		child = devlist[i];
4668 		switch (pci_get_class(child)) {
4669 		case PCIC_DISPLAY:
4670 		case PCIC_MEMORY:
4671 		case PCIC_BRIDGE:
4672 		case PCIC_BASEPERIPH:
4673 			break;
4674 		default:
4675 			BUS_RESUME_CHILD(dev, child);
4676 		}
4677 	}
4678 	free(devlist, M_TEMP);
4679 	return (0);
4680 }
4681 
4682 static void
pci_load_vendor_data(void)4683 pci_load_vendor_data(void)
4684 {
4685 	caddr_t data;
4686 	void *ptr;
4687 	size_t sz;
4688 
4689 	data = preload_search_by_type("pci_vendor_data");
4690 	if (data != NULL) {
4691 		ptr = preload_fetch_addr(data);
4692 		sz = preload_fetch_size(data);
4693 		if (ptr != NULL && sz != 0) {
4694 			pci_vendordata = ptr;
4695 			pci_vendordata_size = sz;
4696 			/* terminate the database */
4697 			pci_vendordata[pci_vendordata_size] = '\n';
4698 		}
4699 	}
4700 }
4701 
4702 void
pci_driver_added(device_t dev,driver_t * driver)4703 pci_driver_added(device_t dev, driver_t *driver)
4704 {
4705 	int numdevs;
4706 	device_t *devlist;
4707 	device_t child;
4708 	struct pci_devinfo *dinfo;
4709 	int i;
4710 
4711 	if (bootverbose)
4712 		device_printf(dev, "driver added\n");
4713 	DEVICE_IDENTIFY(driver, dev);
4714 	if (device_get_children(dev, &devlist, &numdevs) != 0)
4715 		return;
4716 	for (i = 0; i < numdevs; i++) {
4717 		child = devlist[i];
4718 		if (device_get_state(child) != DS_NOTPRESENT)
4719 			continue;
4720 		dinfo = device_get_ivars(child);
4721 		pci_print_verbose(dinfo);
4722 		if (bootverbose)
4723 			pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4724 		pci_cfg_restore(child, dinfo);
4725 		if (device_probe_and_attach(child) != 0)
4726 			pci_child_detached(dev, child);
4727 	}
4728 	free(devlist, M_TEMP);
4729 }
4730 
4731 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)4732 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4733     driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4734 {
4735 	struct pci_devinfo *dinfo;
4736 	struct msix_table_entry *mte;
4737 	struct msix_vector *mv;
4738 	uint64_t addr;
4739 	uint32_t data;
4740 	void *cookie;
4741 	int error, rid;
4742 
4743 	error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4744 	    arg, &cookie);
4745 	if (error)
4746 		return (error);
4747 
4748 	/* If this is not a direct child, just bail out. */
4749 	if (device_get_parent(child) != dev) {
4750 		*cookiep = cookie;
4751 		return(0);
4752 	}
4753 
4754 	rid = rman_get_rid(irq);
4755 	if (rid == 0) {
4756 		/* Make sure that INTx is enabled */
4757 		pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4758 	} else {
4759 		/*
4760 		 * Check to see if the interrupt is MSI or MSI-X.
4761 		 * Ask our parent to map the MSI and give
4762 		 * us the address and data register values.
4763 		 * If we fail for some reason, teardown the
4764 		 * interrupt handler.
4765 		 */
4766 		dinfo = device_get_ivars(child);
4767 		if (dinfo->cfg.msi.msi_alloc > 0) {
4768 			if (dinfo->cfg.msi.msi_addr == 0) {
4769 				KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4770 			    ("MSI has handlers, but vectors not mapped"));
4771 				error = PCIB_MAP_MSI(device_get_parent(dev),
4772 				    child, rman_get_start(irq), &addr, &data);
4773 				if (error)
4774 					goto bad;
4775 				dinfo->cfg.msi.msi_addr = addr;
4776 				dinfo->cfg.msi.msi_data = data;
4777 			}
4778 			if (dinfo->cfg.msi.msi_handlers == 0)
4779 				pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4780 				    dinfo->cfg.msi.msi_data);
4781 			dinfo->cfg.msi.msi_handlers++;
4782 		} else {
4783 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4784 			    ("No MSI or MSI-X interrupts allocated"));
4785 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4786 			    ("MSI-X index too high"));
4787 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4788 			KASSERT(mte->mte_vector != 0, ("no message vector"));
4789 			mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4790 			KASSERT(mv->mv_irq == rman_get_start(irq),
4791 			    ("IRQ mismatch"));
4792 			if (mv->mv_address == 0) {
4793 				KASSERT(mte->mte_handlers == 0,
4794 		    ("MSI-X table entry has handlers, but vector not mapped"));
4795 				error = PCIB_MAP_MSI(device_get_parent(dev),
4796 				    child, rman_get_start(irq), &addr, &data);
4797 				if (error)
4798 					goto bad;
4799 				mv->mv_address = addr;
4800 				mv->mv_data = data;
4801 			}
4802 
4803 			/*
4804 			 * The MSIX table entry must be made valid by
4805 			 * incrementing the mte_handlers before
4806 			 * calling pci_enable_msix() and
4807 			 * pci_resume_msix(). Else the MSIX rewrite
4808 			 * table quirk will not work as expected.
4809 			 */
4810 			mte->mte_handlers++;
4811 			if (mte->mte_handlers == 1) {
4812 				pci_enable_msix(child, rid - 1, mv->mv_address,
4813 				    mv->mv_data);
4814 				pci_unmask_msix(child, rid - 1);
4815 			}
4816 		}
4817 
4818 		/*
4819 		 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4820 		 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4821 		 * in which case we "enable" INTx so MSI/MSI-X actually works.
4822 		 */
4823 		if (!pci_has_quirk(pci_get_devid(child),
4824 		    PCI_QUIRK_MSI_INTX_BUG))
4825 			pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4826 		else
4827 			pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4828 	bad:
4829 		if (error) {
4830 			(void)bus_generic_teardown_intr(dev, child, irq,
4831 			    cookie);
4832 			return (error);
4833 		}
4834 	}
4835 	*cookiep = cookie;
4836 	return (0);
4837 }
4838 
4839 int
pci_teardown_intr(device_t dev,device_t child,struct resource * irq,void * cookie)4840 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4841     void *cookie)
4842 {
4843 	struct msix_table_entry *mte;
4844 	struct resource_list_entry *rle;
4845 	struct pci_devinfo *dinfo;
4846 	int error, rid;
4847 
4848 	if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4849 		return (EINVAL);
4850 
4851 	/* If this isn't a direct child, just bail out */
4852 	if (device_get_parent(child) != dev)
4853 		return(bus_generic_teardown_intr(dev, child, irq, cookie));
4854 
4855 	rid = rman_get_rid(irq);
4856 	if (rid == 0) {
4857 		/* Mask INTx */
4858 		pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4859 	} else {
4860 		/*
4861 		 * Check to see if the interrupt is MSI or MSI-X.  If so,
4862 		 * decrement the appropriate handlers count and mask the
4863 		 * MSI-X message, or disable MSI messages if the count
4864 		 * drops to 0.
4865 		 */
4866 		dinfo = device_get_ivars(child);
4867 		rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4868 		if (rle->res != irq)
4869 			return (EINVAL);
4870 		if (dinfo->cfg.msi.msi_alloc > 0) {
4871 			KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4872 			    ("MSI-X index too high"));
4873 			if (dinfo->cfg.msi.msi_handlers == 0)
4874 				return (EINVAL);
4875 			dinfo->cfg.msi.msi_handlers--;
4876 			if (dinfo->cfg.msi.msi_handlers == 0)
4877 				pci_disable_msi(child);
4878 		} else {
4879 			KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4880 			    ("No MSI or MSI-X interrupts allocated"));
4881 			KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4882 			    ("MSI-X index too high"));
4883 			mte = &dinfo->cfg.msix.msix_table[rid - 1];
4884 			if (mte->mte_handlers == 0)
4885 				return (EINVAL);
4886 			mte->mte_handlers--;
4887 			if (mte->mte_handlers == 0)
4888 				pci_mask_msix(child, rid - 1);
4889 		}
4890 	}
4891 	error = bus_generic_teardown_intr(dev, child, irq, cookie);
4892 	if (rid > 0)
4893 		KASSERT(error == 0,
4894 		    ("%s: generic teardown failed for MSI/MSI-X", __func__));
4895 	return (error);
4896 }
4897 
4898 int
pci_print_child(device_t dev,device_t child)4899 pci_print_child(device_t dev, device_t child)
4900 {
4901 	struct pci_devinfo *dinfo;
4902 	struct resource_list *rl;
4903 	int retval = 0;
4904 
4905 	dinfo = device_get_ivars(child);
4906 	rl = &dinfo->resources;
4907 
4908 	retval += bus_print_child_header(dev, child);
4909 
4910 	retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4911 	retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4912 	retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4913 	if (device_get_flags(dev))
4914 		retval += printf(" flags %#x", device_get_flags(dev));
4915 
4916 	retval += printf(" at device %d.%d", pci_get_slot(child),
4917 	    pci_get_function(child));
4918 
4919 	retval += bus_print_child_domain(dev, child);
4920 	retval += bus_print_child_footer(dev, child);
4921 
4922 	return (retval);
4923 }
4924 
4925 static const struct
4926 {
4927 	int		class;
4928 	int		subclass;
4929 	int		report; /* 0 = bootverbose, 1 = always */
4930 	const char	*desc;
4931 } pci_nomatch_tab[] = {
4932 	{PCIC_OLD,		-1,			1, "old"},
4933 	{PCIC_OLD,		PCIS_OLD_NONVGA,	1, "non-VGA display device"},
4934 	{PCIC_OLD,		PCIS_OLD_VGA,		1, "VGA-compatible display device"},
4935 	{PCIC_STORAGE,		-1,			1, "mass storage"},
4936 	{PCIC_STORAGE,		PCIS_STORAGE_SCSI,	1, "SCSI"},
4937 	{PCIC_STORAGE,		PCIS_STORAGE_IDE,	1, "ATA"},
4938 	{PCIC_STORAGE,		PCIS_STORAGE_FLOPPY,	1, "floppy disk"},
4939 	{PCIC_STORAGE,		PCIS_STORAGE_IPI,	1, "IPI"},
4940 	{PCIC_STORAGE,		PCIS_STORAGE_RAID,	1, "RAID"},
4941 	{PCIC_STORAGE,		PCIS_STORAGE_ATA_ADMA,	1, "ATA (ADMA)"},
4942 	{PCIC_STORAGE,		PCIS_STORAGE_SATA,	1, "SATA"},
4943 	{PCIC_STORAGE,		PCIS_STORAGE_SAS,	1, "SAS"},
4944 	{PCIC_STORAGE,		PCIS_STORAGE_NVM,	1, "NVM"},
4945 	{PCIC_NETWORK,		-1,			1, "network"},
4946 	{PCIC_NETWORK,		PCIS_NETWORK_ETHERNET,	1, "ethernet"},
4947 	{PCIC_NETWORK,		PCIS_NETWORK_TOKENRING,	1, "token ring"},
4948 	{PCIC_NETWORK,		PCIS_NETWORK_FDDI,	1, "fddi"},
4949 	{PCIC_NETWORK,		PCIS_NETWORK_ATM,	1, "ATM"},
4950 	{PCIC_NETWORK,		PCIS_NETWORK_ISDN,	1, "ISDN"},
4951 	{PCIC_DISPLAY,		-1,			1, "display"},
4952 	{PCIC_DISPLAY,		PCIS_DISPLAY_VGA,	1, "VGA"},
4953 	{PCIC_DISPLAY,		PCIS_DISPLAY_XGA,	1, "XGA"},
4954 	{PCIC_DISPLAY,		PCIS_DISPLAY_3D,	1, "3D"},
4955 	{PCIC_MULTIMEDIA,	-1,			1, "multimedia"},
4956 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_VIDEO,	1, "video"},
4957 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_AUDIO,	1, "audio"},
4958 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_TELE,	1, "telephony"},
4959 	{PCIC_MULTIMEDIA,	PCIS_MULTIMEDIA_HDA,	1, "HDA"},
4960 	{PCIC_MEMORY,		-1,			1, "memory"},
4961 	{PCIC_MEMORY,		PCIS_MEMORY_RAM,	1, "RAM"},
4962 	{PCIC_MEMORY,		PCIS_MEMORY_FLASH,	1, "flash"},
4963 	{PCIC_BRIDGE,		-1,			1, "bridge"},
4964 	{PCIC_BRIDGE,		PCIS_BRIDGE_HOST,	1, "HOST-PCI"},
4965 	{PCIC_BRIDGE,		PCIS_BRIDGE_ISA,	1, "PCI-ISA"},
4966 	{PCIC_BRIDGE,		PCIS_BRIDGE_EISA,	1, "PCI-EISA"},
4967 	{PCIC_BRIDGE,		PCIS_BRIDGE_MCA,	1, "PCI-MCA"},
4968 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCI,	1, "PCI-PCI"},
4969 	{PCIC_BRIDGE,		PCIS_BRIDGE_PCMCIA,	1, "PCI-PCMCIA"},
4970 	{PCIC_BRIDGE,		PCIS_BRIDGE_NUBUS,	1, "PCI-NuBus"},
4971 	{PCIC_BRIDGE,		PCIS_BRIDGE_CARDBUS,	1, "PCI-CardBus"},
4972 	{PCIC_BRIDGE,		PCIS_BRIDGE_RACEWAY,	1, "PCI-RACEway"},
4973 	{PCIC_SIMPLECOMM,	-1,			1, "simple comms"},
4974 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_UART,	1, "UART"},	/* could detect 16550 */
4975 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_PAR,	1, "parallel port"},
4976 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MULSER,	1, "multiport serial"},
4977 	{PCIC_SIMPLECOMM,	PCIS_SIMPLECOMM_MODEM,	1, "generic modem"},
4978 	{PCIC_BASEPERIPH,	-1,			0, "base peripheral"},
4979 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PIC,	1, "interrupt controller"},
4980 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_DMA,	1, "DMA controller"},
4981 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_TIMER,	1, "timer"},
4982 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_RTC,	1, "realtime clock"},
4983 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_PCIHOT,	1, "PCI hot-plug controller"},
4984 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_SDHC,	1, "SD host controller"},
4985 	{PCIC_BASEPERIPH,	PCIS_BASEPERIPH_IOMMU,	1, "IOMMU"},
4986 	{PCIC_INPUTDEV,		-1,			1, "input device"},
4987 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_KEYBOARD,	1, "keyboard"},
4988 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4989 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_MOUSE,	1, "mouse"},
4990 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_SCANNER,	1, "scanner"},
4991 	{PCIC_INPUTDEV,		PCIS_INPUTDEV_GAMEPORT,	1, "gameport"},
4992 	{PCIC_DOCKING,		-1,			1, "docking station"},
4993 	{PCIC_PROCESSOR,	-1,			1, "processor"},
4994 	{PCIC_SERIALBUS,	-1,			1, "serial bus"},
4995 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FW,	1, "FireWire"},
4996 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_ACCESS,	1, "AccessBus"},
4997 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SSA,	1, "SSA"},
4998 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_USB,	1, "USB"},
4999 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_FC,	1, "Fibre Channel"},
5000 	{PCIC_SERIALBUS,	PCIS_SERIALBUS_SMBUS,	0, "SMBus"},
5001 	{PCIC_WIRELESS,		-1,			1, "wireless controller"},
5002 	{PCIC_WIRELESS,		PCIS_WIRELESS_IRDA,	1, "iRDA"},
5003 	{PCIC_WIRELESS,		PCIS_WIRELESS_IR,	1, "IR"},
5004 	{PCIC_WIRELESS,		PCIS_WIRELESS_RF,	1, "RF"},
5005 	{PCIC_INTELLIIO,	-1,			1, "intelligent I/O controller"},
5006 	{PCIC_INTELLIIO,	PCIS_INTELLIIO_I2O,	1, "I2O"},
5007 	{PCIC_SATCOM,		-1,			1, "satellite communication"},
5008 	{PCIC_SATCOM,		PCIS_SATCOM_TV,		1, "sat TV"},
5009 	{PCIC_SATCOM,		PCIS_SATCOM_AUDIO,	1, "sat audio"},
5010 	{PCIC_SATCOM,		PCIS_SATCOM_VOICE,	1, "sat voice"},
5011 	{PCIC_SATCOM,		PCIS_SATCOM_DATA,	1, "sat data"},
5012 	{PCIC_CRYPTO,		-1,			1, "encrypt/decrypt"},
5013 	{PCIC_CRYPTO,		PCIS_CRYPTO_NETCOMP,	1, "network/computer crypto"},
5014 	{PCIC_CRYPTO,		PCIS_CRYPTO_ENTERTAIN,	1, "entertainment crypto"},
5015 	{PCIC_DASP,		-1,			0, "dasp"},
5016 	{PCIC_DASP,		PCIS_DASP_DPIO,		1, "DPIO module"},
5017 	{PCIC_DASP,		PCIS_DASP_PERFCNTRS,	1, "performance counters"},
5018 	{PCIC_DASP,		PCIS_DASP_COMM_SYNC,	1, "communication synchronizer"},
5019 	{PCIC_DASP,		PCIS_DASP_MGMT_CARD,	1, "signal processing management"},
5020 	{PCIC_INSTRUMENT,	-1,			0, "non-essential instrumentation"},
5021 	{0, 0, 0,		NULL}
5022 };
5023 
5024 void
pci_probe_nomatch(device_t dev,device_t child)5025 pci_probe_nomatch(device_t dev, device_t child)
5026 {
5027 	int i, report;
5028 	const char *cp, *scp;
5029 	char *device;
5030 
5031 	/*
5032 	 * Look for a listing for this device in a loaded device database.
5033 	 */
5034 	report = 1;
5035 	if ((device = pci_describe_device(child)) != NULL) {
5036 		device_printf(dev, "<%s>", device);
5037 		free(device, M_DEVBUF);
5038 	} else {
5039 		/*
5040 		 * Scan the class/subclass descriptions for a general
5041 		 * description.
5042 		 */
5043 		cp = "unknown";
5044 		scp = NULL;
5045 		for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
5046 			if (pci_nomatch_tab[i].class == pci_get_class(child)) {
5047 				if (pci_nomatch_tab[i].subclass == -1) {
5048 					cp = pci_nomatch_tab[i].desc;
5049 					report = pci_nomatch_tab[i].report;
5050 				} else if (pci_nomatch_tab[i].subclass ==
5051 				    pci_get_subclass(child)) {
5052 					scp = pci_nomatch_tab[i].desc;
5053 					report = pci_nomatch_tab[i].report;
5054 				}
5055 			}
5056 		}
5057 		if (report || bootverbose) {
5058 			device_printf(dev, "<%s%s%s>",
5059 			    cp ? cp : "",
5060 			    ((cp != NULL) && (scp != NULL)) ? ", " : "",
5061 			    scp ? scp : "");
5062 		}
5063 	}
5064 	if (report || bootverbose) {
5065 		printf(" at device %d.%d (no driver attached)\n",
5066 		    pci_get_slot(child), pci_get_function(child));
5067 	}
5068 	pci_cfg_save(child, device_get_ivars(child), 1);
5069 }
5070 
5071 void
pci_child_detached(device_t dev,device_t child)5072 pci_child_detached(device_t dev, device_t child)
5073 {
5074 	struct pci_devinfo *dinfo;
5075 	struct resource_list *rl;
5076 
5077 	dinfo = device_get_ivars(child);
5078 	rl = &dinfo->resources;
5079 
5080 	/*
5081 	 * Have to deallocate IRQs before releasing any MSI messages and
5082 	 * have to release MSI messages before deallocating any memory
5083 	 * BARs.
5084 	 */
5085 	if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
5086 		pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
5087 	if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
5088 		if (dinfo->cfg.msi.msi_alloc != 0)
5089 			pci_printf(&dinfo->cfg, "Device leaked %d MSI "
5090 			    "vectors\n", dinfo->cfg.msi.msi_alloc);
5091 		else
5092 			pci_printf(&dinfo->cfg, "Device leaked %d MSI-X "
5093 			    "vectors\n", dinfo->cfg.msix.msix_alloc);
5094 		(void)pci_release_msi(child);
5095 	}
5096 	if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
5097 		pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
5098 	if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
5099 		pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
5100 	if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
5101 		pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
5102 
5103 	pci_cfg_save(child, dinfo, 1);
5104 }
5105 
5106 /*
5107  * Parse the PCI device database, if loaded, and return a pointer to a
5108  * description of the device.
5109  *
5110  * The database is flat text formatted as follows:
5111  *
5112  * Any line not in a valid format is ignored.
5113  * Lines are terminated with newline '\n' characters.
5114  *
5115  * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5116  * the vendor name.
5117  *
5118  * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5119  * - devices cannot be listed without a corresponding VENDOR line.
5120  * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5121  * another TAB, then the device name.
5122  */
5123 
5124 /*
5125  * Assuming (ptr) points to the beginning of a line in the database,
5126  * return the vendor or device and description of the next entry.
5127  * The value of (vendor) or (device) inappropriate for the entry type
5128  * is set to -1.  Returns nonzero at the end of the database.
5129  *
5130  * Note that this is slightly unrobust in the face of corrupt data;
5131  * we attempt to safeguard against this by spamming the end of the
5132  * database with a newline when we initialise.
5133  */
5134 static int
pci_describe_parse_line(char ** ptr,int * vendor,int * device,char ** desc)5135 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5136 {
5137 	char	*cp = *ptr;
5138 	int	left;
5139 
5140 	*device = -1;
5141 	*vendor = -1;
5142 	**desc = '\0';
5143 	for (;;) {
5144 		left = pci_vendordata_size - (cp - pci_vendordata);
5145 		if (left <= 0) {
5146 			*ptr = cp;
5147 			return(1);
5148 		}
5149 
5150 		/* vendor entry? */
5151 		if (*cp != '\t' &&
5152 		    sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5153 			break;
5154 		/* device entry? */
5155 		if (*cp == '\t' &&
5156 		    sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5157 			break;
5158 
5159 		/* skip to next line */
5160 		while (*cp != '\n' && left > 0) {
5161 			cp++;
5162 			left--;
5163 		}
5164 		if (*cp == '\n') {
5165 			cp++;
5166 			left--;
5167 		}
5168 	}
5169 	/* skip to next line */
5170 	while (*cp != '\n' && left > 0) {
5171 		cp++;
5172 		left--;
5173 	}
5174 	if (*cp == '\n' && left > 0)
5175 		cp++;
5176 	*ptr = cp;
5177 	return(0);
5178 }
5179 
5180 static char *
pci_describe_device(device_t dev)5181 pci_describe_device(device_t dev)
5182 {
5183 	int	vendor, device;
5184 	char	*desc, *vp, *dp, *line;
5185 
5186 	desc = vp = dp = NULL;
5187 
5188 	/*
5189 	 * If we have no vendor data, we can't do anything.
5190 	 */
5191 	if (pci_vendordata == NULL)
5192 		goto out;
5193 
5194 	/*
5195 	 * Scan the vendor data looking for this device
5196 	 */
5197 	line = pci_vendordata;
5198 	if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5199 		goto out;
5200 	for (;;) {
5201 		if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5202 			goto out;
5203 		if (vendor == pci_get_vendor(dev))
5204 			break;
5205 	}
5206 	if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5207 		goto out;
5208 	for (;;) {
5209 		if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5210 			*dp = 0;
5211 			break;
5212 		}
5213 		if (vendor != -1) {
5214 			*dp = 0;
5215 			break;
5216 		}
5217 		if (device == pci_get_device(dev))
5218 			break;
5219 	}
5220 	if (dp[0] == '\0')
5221 		snprintf(dp, 80, "0x%x", pci_get_device(dev));
5222 	if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5223 	    NULL)
5224 		sprintf(desc, "%s, %s", vp, dp);
5225 out:
5226 	if (vp != NULL)
5227 		free(vp, M_DEVBUF);
5228 	if (dp != NULL)
5229 		free(dp, M_DEVBUF);
5230 	return(desc);
5231 }
5232 
5233 int
pci_read_ivar(device_t dev,device_t child,int which,uintptr_t * result)5234 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5235 {
5236 	struct pci_devinfo *dinfo;
5237 	pcicfgregs *cfg;
5238 
5239 	dinfo = device_get_ivars(child);
5240 	cfg = &dinfo->cfg;
5241 
5242 	switch (which) {
5243 	case PCI_IVAR_ETHADDR:
5244 		/*
5245 		 * The generic accessor doesn't deal with failure, so
5246 		 * we set the return value, then return an error.
5247 		 */
5248 		*((uint8_t **) result) = NULL;
5249 		return (EINVAL);
5250 	case PCI_IVAR_SUBVENDOR:
5251 		*result = cfg->subvendor;
5252 		break;
5253 	case PCI_IVAR_SUBDEVICE:
5254 		*result = cfg->subdevice;
5255 		break;
5256 	case PCI_IVAR_VENDOR:
5257 		*result = cfg->vendor;
5258 		break;
5259 	case PCI_IVAR_DEVICE:
5260 		*result = cfg->device;
5261 		break;
5262 	case PCI_IVAR_DEVID:
5263 		*result = (cfg->device << 16) | cfg->vendor;
5264 		break;
5265 	case PCI_IVAR_CLASS:
5266 		*result = cfg->baseclass;
5267 		break;
5268 	case PCI_IVAR_SUBCLASS:
5269 		*result = cfg->subclass;
5270 		break;
5271 	case PCI_IVAR_PROGIF:
5272 		*result = cfg->progif;
5273 		break;
5274 	case PCI_IVAR_REVID:
5275 		*result = cfg->revid;
5276 		break;
5277 	case PCI_IVAR_INTPIN:
5278 		*result = cfg->intpin;
5279 		break;
5280 	case PCI_IVAR_IRQ:
5281 		*result = cfg->intline;
5282 		break;
5283 	case PCI_IVAR_DOMAIN:
5284 		*result = cfg->domain;
5285 		break;
5286 	case PCI_IVAR_BUS:
5287 		*result = cfg->bus;
5288 		break;
5289 	case PCI_IVAR_SLOT:
5290 		*result = cfg->slot;
5291 		break;
5292 	case PCI_IVAR_FUNCTION:
5293 		*result = cfg->func;
5294 		break;
5295 	case PCI_IVAR_CMDREG:
5296 		*result = cfg->cmdreg;
5297 		break;
5298 	case PCI_IVAR_CACHELNSZ:
5299 		*result = cfg->cachelnsz;
5300 		break;
5301 	case PCI_IVAR_MINGNT:
5302 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5303 			*result = -1;
5304 			return (EINVAL);
5305 		}
5306 		*result = cfg->mingnt;
5307 		break;
5308 	case PCI_IVAR_MAXLAT:
5309 		if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5310 			*result = -1;
5311 			return (EINVAL);
5312 		}
5313 		*result = cfg->maxlat;
5314 		break;
5315 	case PCI_IVAR_LATTIMER:
5316 		*result = cfg->lattimer;
5317 		break;
5318 	default:
5319 		return (ENOENT);
5320 	}
5321 	return (0);
5322 }
5323 
5324 int
pci_write_ivar(device_t dev,device_t child,int which,uintptr_t value)5325 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5326 {
5327 	struct pci_devinfo *dinfo;
5328 
5329 	dinfo = device_get_ivars(child);
5330 
5331 	switch (which) {
5332 	case PCI_IVAR_INTPIN:
5333 		dinfo->cfg.intpin = value;
5334 		return (0);
5335 	case PCI_IVAR_ETHADDR:
5336 	case PCI_IVAR_SUBVENDOR:
5337 	case PCI_IVAR_SUBDEVICE:
5338 	case PCI_IVAR_VENDOR:
5339 	case PCI_IVAR_DEVICE:
5340 	case PCI_IVAR_DEVID:
5341 	case PCI_IVAR_CLASS:
5342 	case PCI_IVAR_SUBCLASS:
5343 	case PCI_IVAR_PROGIF:
5344 	case PCI_IVAR_REVID:
5345 	case PCI_IVAR_IRQ:
5346 	case PCI_IVAR_DOMAIN:
5347 	case PCI_IVAR_BUS:
5348 	case PCI_IVAR_SLOT:
5349 	case PCI_IVAR_FUNCTION:
5350 		return (EINVAL);	/* disallow for now */
5351 
5352 	default:
5353 		return (ENOENT);
5354 	}
5355 }
5356 
5357 #include "opt_ddb.h"
5358 #ifdef DDB
5359 #include <ddb/ddb.h>
5360 #include <sys/cons.h>
5361 
5362 /*
5363  * List resources based on pci map registers, used for within ddb
5364  */
5365 
DB_SHOW_COMMAND_FLAGS(pciregs,db_pci_dump,DB_CMD_MEMSAFE)5366 DB_SHOW_COMMAND_FLAGS(pciregs, db_pci_dump, DB_CMD_MEMSAFE)
5367 {
5368 	struct pci_devinfo *dinfo;
5369 	struct devlist *devlist_head;
5370 	struct pci_conf *p;
5371 	const char *name;
5372 	int i, error, none_count;
5373 
5374 	none_count = 0;
5375 	/* get the head of the device queue */
5376 	devlist_head = &pci_devq;
5377 
5378 	/*
5379 	 * Go through the list of devices and print out devices
5380 	 */
5381 	for (error = 0, i = 0,
5382 	     dinfo = STAILQ_FIRST(devlist_head);
5383 	     (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5384 	     dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5385 		/* Populate pd_name and pd_unit */
5386 		name = NULL;
5387 		if (dinfo->cfg.dev)
5388 			name = device_get_name(dinfo->cfg.dev);
5389 
5390 		p = &dinfo->conf;
5391 		db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5392 			"chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5393 			(name && *name) ? name : "none",
5394 			(name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5395 			none_count++,
5396 			p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5397 			p->pc_sel.pc_func, (p->pc_class << 16) |
5398 			(p->pc_subclass << 8) | p->pc_progif,
5399 			(p->pc_subdevice << 16) | p->pc_subvendor,
5400 			(p->pc_device << 16) | p->pc_vendor,
5401 			p->pc_revid, p->pc_hdr);
5402 	}
5403 }
5404 #endif /* DDB */
5405 
5406 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)5407 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5408     rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5409     u_int flags)
5410 {
5411 	struct pci_devinfo *dinfo = device_get_ivars(child);
5412 	struct resource_list *rl = &dinfo->resources;
5413 	struct resource *res;
5414 	struct pci_map *pm;
5415 	uint16_t cmd;
5416 	pci_addr_t map, testval;
5417 	int mapsize;
5418 
5419 	res = NULL;
5420 
5421 	/* If rid is managed by EA, ignore it */
5422 	if (pci_ea_is_enabled(child, *rid))
5423 		goto out;
5424 
5425 	pm = pci_find_bar(child, *rid);
5426 	if (pm != NULL) {
5427 		/* This is a BAR that we failed to allocate earlier. */
5428 		mapsize = pm->pm_size;
5429 		map = pm->pm_value;
5430 	} else {
5431 		/*
5432 		 * Weed out the bogons, and figure out how large the
5433 		 * BAR/map is.  BARs that read back 0 here are bogus
5434 		 * and unimplemented.  Note: atapci in legacy mode are
5435 		 * special and handled elsewhere in the code.  If you
5436 		 * have a atapci device in legacy mode and it fails
5437 		 * here, that other code is broken.
5438 		 */
5439 		pci_read_bar(child, *rid, &map, &testval, NULL);
5440 
5441 		/*
5442 		 * Determine the size of the BAR and ignore BARs with a size
5443 		 * of 0.  Device ROM BARs use a different mask value.
5444 		 */
5445 		if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5446 			mapsize = pci_romsize(testval);
5447 		else
5448 			mapsize = pci_mapsize(testval);
5449 		if (mapsize == 0)
5450 			goto out;
5451 		pm = pci_add_bar(child, *rid, map, mapsize);
5452 	}
5453 
5454 	if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5455 		if (type != SYS_RES_MEMORY) {
5456 			if (bootverbose)
5457 				device_printf(dev,
5458 				    "child %s requested type %d for rid %#x,"
5459 				    " but the BAR says it is an memio\n",
5460 				    device_get_nameunit(child), type, *rid);
5461 			goto out;
5462 		}
5463 	} else {
5464 		if (type != SYS_RES_IOPORT) {
5465 			if (bootverbose)
5466 				device_printf(dev,
5467 				    "child %s requested type %d for rid %#x,"
5468 				    " but the BAR says it is an ioport\n",
5469 				    device_get_nameunit(child), type, *rid);
5470 			goto out;
5471 		}
5472 	}
5473 
5474 	/*
5475 	 * For real BARs, we need to override the size that
5476 	 * the driver requests, because that's what the BAR
5477 	 * actually uses and we would otherwise have a
5478 	 * situation where we might allocate the excess to
5479 	 * another driver, which won't work.
5480 	 */
5481 	count = ((pci_addr_t)1 << mapsize) * num;
5482 	if (RF_ALIGNMENT(flags) < mapsize)
5483 		flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5484 	if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5485 		flags |= RF_PREFETCHABLE;
5486 
5487 	/*
5488 	 * Allocate enough resource, and then write back the
5489 	 * appropriate BAR for that resource.
5490 	 */
5491 	resource_list_add(rl, type, *rid, start, end, count);
5492 	res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5493 	    count, flags & ~RF_ACTIVE);
5494 	if (res == NULL) {
5495 		resource_list_delete(rl, type, *rid);
5496 		device_printf(child,
5497 		    "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5498 		    count, *rid, type, start, end);
5499 		goto out;
5500 	}
5501 	if (bootverbose)
5502 		device_printf(child,
5503 		    "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5504 		    count, *rid, type, rman_get_start(res));
5505 
5506 	/* Disable decoding via the CMD register before updating the BAR */
5507 	cmd = pci_read_config(child, PCIR_COMMAND, 2);
5508 	pci_write_config(child, PCIR_COMMAND,
5509 	    cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5510 
5511 	map = rman_get_start(res);
5512 	pci_write_bar(child, pm, map);
5513 
5514 	/* Restore the original value of the CMD register */
5515 	pci_write_config(child, PCIR_COMMAND, cmd, 2);
5516 out:
5517 	return (res);
5518 }
5519 
5520 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)5521 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5522     rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5523     u_int flags)
5524 {
5525 	struct pci_devinfo *dinfo;
5526 	struct resource_list *rl;
5527 	struct resource_list_entry *rle;
5528 	struct resource *res;
5529 	pcicfgregs *cfg;
5530 
5531 	/*
5532 	 * Perform lazy resource allocation
5533 	 */
5534 	dinfo = device_get_ivars(child);
5535 	rl = &dinfo->resources;
5536 	cfg = &dinfo->cfg;
5537 	switch (type) {
5538 	case PCI_RES_BUS:
5539 		return (pci_alloc_secbus(dev, child, rid, start, end, count,
5540 		    flags));
5541 	case SYS_RES_IRQ:
5542 		/*
5543 		 * Can't alloc legacy interrupt once MSI messages have
5544 		 * been allocated.
5545 		 */
5546 		if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5547 		    cfg->msix.msix_alloc > 0))
5548 			return (NULL);
5549 
5550 		/*
5551 		 * If the child device doesn't have an interrupt
5552 		 * routed and is deserving of an interrupt, try to
5553 		 * assign it one.
5554 		 */
5555 		if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5556 		    (cfg->intpin != 0))
5557 			pci_assign_interrupt(dev, child, 0);
5558 		break;
5559 	case SYS_RES_IOPORT:
5560 	case SYS_RES_MEMORY:
5561 		/*
5562 		 * PCI-PCI bridge I/O window resources are not BARs.
5563 		 * For those allocations just pass the request up the
5564 		 * tree.
5565 		 */
5566 		if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5567 			switch (*rid) {
5568 			case PCIR_IOBASEL_1:
5569 			case PCIR_MEMBASE_1:
5570 			case PCIR_PMBASEL_1:
5571 				/*
5572 				 * XXX: Should we bother creating a resource
5573 				 * list entry?
5574 				 */
5575 				return (bus_generic_alloc_resource(dev, child,
5576 				    type, rid, start, end, count, flags));
5577 			}
5578 		}
5579 		/* Reserve resources for this BAR if needed. */
5580 		rle = resource_list_find(rl, type, *rid);
5581 		if (rle == NULL) {
5582 			res = pci_reserve_map(dev, child, type, rid, start, end,
5583 			    count, num, flags);
5584 			if (res == NULL)
5585 				return (NULL);
5586 		}
5587 	}
5588 	return (resource_list_alloc(rl, dev, child, type, rid,
5589 	    start, end, count, flags));
5590 }
5591 
5592 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)5593 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5594     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5595 {
5596 #ifdef PCI_IOV
5597 	struct pci_devinfo *dinfo;
5598 #endif
5599 
5600 	if (device_get_parent(child) != dev)
5601 		return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5602 		    type, rid, start, end, count, flags));
5603 
5604 #ifdef PCI_IOV
5605 	dinfo = device_get_ivars(child);
5606 	if (dinfo->cfg.flags & PCICFG_VF) {
5607 		switch (type) {
5608 		/* VFs can't have I/O BARs. */
5609 		case SYS_RES_IOPORT:
5610 			return (NULL);
5611 		case SYS_RES_MEMORY:
5612 			return (pci_vf_alloc_mem_resource(dev, child, rid,
5613 			    start, end, count, flags));
5614 		}
5615 
5616 		/* Fall through for other types of resource allocations. */
5617 	}
5618 #endif
5619 
5620 	return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5621 	    count, 1, flags));
5622 }
5623 
5624 int
pci_release_resource(device_t dev,device_t child,struct resource * r)5625 pci_release_resource(device_t dev, device_t child, struct resource *r)
5626 {
5627 	struct pci_devinfo *dinfo;
5628 	struct resource_list *rl;
5629 	pcicfgregs *cfg __unused;
5630 
5631 	if (device_get_parent(child) != dev)
5632 		return (bus_generic_release_resource(dev, child, r));
5633 
5634 	dinfo = device_get_ivars(child);
5635 	cfg = &dinfo->cfg;
5636 
5637 #ifdef PCI_IOV
5638 	if (cfg->flags & PCICFG_VF) {
5639 		switch (rman_get_type(r)) {
5640 		/* VFs can't have I/O BARs. */
5641 		case SYS_RES_IOPORT:
5642 			return (EDOOFUS);
5643 		case SYS_RES_MEMORY:
5644 			return (pci_vf_release_mem_resource(dev, child, r));
5645 		}
5646 
5647 		/* Fall through for other types of resource allocations. */
5648 	}
5649 #endif
5650 
5651 	/*
5652 	 * PCI-PCI bridge I/O window resources are not BARs.  For
5653 	 * those allocations just pass the request up the tree.
5654 	 */
5655 	if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5656 	    (rman_get_type(r) == SYS_RES_IOPORT ||
5657 	    rman_get_type(r) == SYS_RES_MEMORY)) {
5658 		switch (rman_get_rid(r)) {
5659 		case PCIR_IOBASEL_1:
5660 		case PCIR_MEMBASE_1:
5661 		case PCIR_PMBASEL_1:
5662 			return (bus_generic_release_resource(dev, child, r));
5663 		}
5664 	}
5665 
5666 	rl = &dinfo->resources;
5667 	return (resource_list_release(rl, dev, child, r));
5668 }
5669 
5670 int
pci_activate_resource(device_t dev,device_t child,struct resource * r)5671 pci_activate_resource(device_t dev, device_t child, struct resource *r)
5672 {
5673 	struct pci_devinfo *dinfo;
5674 	int error, rid, type;
5675 
5676 	if (device_get_parent(child) != dev)
5677 		return (bus_generic_activate_resource(dev, child, r));
5678 
5679 	dinfo = device_get_ivars(child);
5680 #ifdef PCI_IOV
5681 	if (dinfo->cfg.flags & PCICFG_VF) {
5682 		switch (rman_get_type(r)) {
5683 		/* VFs can't have I/O BARs. */
5684 		case SYS_RES_IOPORT:
5685 			error = EINVAL;
5686 			break;
5687 		case SYS_RES_MEMORY:
5688 			error = pci_vf_activate_mem_resource(dev, child, r);
5689 			break;
5690 		default:
5691 			error = bus_generic_activate_resource(dev, child, r);
5692 			break;
5693 		}
5694 	} else
5695 #endif
5696 		error = bus_generic_activate_resource(dev, child, r);
5697 	if (error)
5698 		return (error);
5699 
5700 	rid = rman_get_rid(r);
5701 	type = rman_get_type(r);
5702 
5703 	/* Device ROMs need their decoding explicitly enabled. */
5704 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5705 		pci_write_bar(child, pci_find_bar(child, rid),
5706 		    rman_get_start(r) | PCIM_BIOS_ENABLE);
5707 
5708 	/* Enable decoding in the command register when activating BARs. */
5709 	switch (type) {
5710 	case SYS_RES_IOPORT:
5711 	case SYS_RES_MEMORY:
5712 		error = PCI_ENABLE_IO(dev, child, type);
5713 		break;
5714 	}
5715 	return (error);
5716 }
5717 
5718 int
pci_deactivate_resource(device_t dev,device_t child,struct resource * r)5719 pci_deactivate_resource(device_t dev, device_t child, struct resource *r)
5720 {
5721 	struct pci_devinfo *dinfo;
5722 	int error, rid, type;
5723 
5724 	if (device_get_parent(child) != dev)
5725 		return (bus_generic_deactivate_resource(dev, child, r));
5726 
5727 	dinfo = device_get_ivars(child);
5728 #ifdef PCI_IOV
5729 	if (dinfo->cfg.flags & PCICFG_VF) {
5730 		switch (rman_get_type(r)) {
5731 		/* VFs can't have I/O BARs. */
5732 		case SYS_RES_IOPORT:
5733 			error = EINVAL;
5734 			break;
5735 		case SYS_RES_MEMORY:
5736 			error = pci_vf_deactivate_mem_resource(dev, child, r);
5737 			break;
5738 		default:
5739 			error = bus_generic_deactivate_resource(dev, child, r);
5740 			break;
5741 		}
5742 	} else
5743 #endif
5744 		error = bus_generic_deactivate_resource(dev, child, r);
5745 	if (error)
5746 		return (error);
5747 
5748 	/* Disable decoding for device ROMs. */
5749 	rid = rman_get_rid(r);
5750 	type = rman_get_type(r);
5751 	if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5752 		pci_write_bar(child, pci_find_bar(child, rid),
5753 		    rman_get_start(r));
5754 	return (0);
5755 }
5756 
5757 int
pci_adjust_resource(device_t dev,device_t child,struct resource * r,rman_res_t start,rman_res_t end)5758 pci_adjust_resource(device_t dev, device_t child, struct resource *r,
5759     rman_res_t start, rman_res_t end)
5760 {
5761 #ifdef PCI_IOV
5762 	struct pci_devinfo *dinfo;
5763 
5764 	if (device_get_parent(child) != dev)
5765 		return (bus_generic_adjust_resource(dev, child, r, start,
5766 		    end));
5767 
5768 	dinfo = device_get_ivars(child);
5769 	if (dinfo->cfg.flags & PCICFG_VF) {
5770 		switch (rman_get_type(r)) {
5771 		/* VFs can't have I/O BARs. */
5772 		case SYS_RES_IOPORT:
5773 			return (EINVAL);
5774 		case SYS_RES_MEMORY:
5775 			return (pci_vf_adjust_mem_resource(dev, child, r,
5776 			    start, end));
5777 		}
5778 
5779 		/* Fall through for other types of resource allocations. */
5780 	}
5781 #endif
5782 
5783 	return (bus_generic_adjust_resource(dev, child, r, start, end));
5784 }
5785 
5786 int
pci_map_resource(device_t dev,device_t child,struct resource * r,struct resource_map_request * argsp,struct resource_map * map)5787 pci_map_resource(device_t dev, device_t child, struct resource *r,
5788     struct resource_map_request *argsp, struct resource_map *map)
5789 {
5790 #ifdef PCI_IOV
5791 	struct pci_devinfo *dinfo;
5792 
5793 	if (device_get_parent(child) != dev)
5794 		return (bus_generic_map_resource(dev, child, r, argsp,
5795 		    map));
5796 
5797 	dinfo = device_get_ivars(child);
5798 	if (dinfo->cfg.flags & PCICFG_VF) {
5799 		switch (rman_get_type(r)) {
5800 		/* VFs can't have I/O BARs. */
5801 		case SYS_RES_IOPORT:
5802 			return (EINVAL);
5803 		case SYS_RES_MEMORY:
5804 			return (pci_vf_map_mem_resource(dev, child, r, argsp,
5805 			    map));
5806 		}
5807 
5808 		/* Fall through for other types of resource allocations. */
5809 	}
5810 #endif
5811 
5812 	return (bus_generic_map_resource(dev, child, r, argsp, map));
5813 }
5814 
5815 int
pci_unmap_resource(device_t dev,device_t child,struct resource * r,struct resource_map * map)5816 pci_unmap_resource(device_t dev, device_t child, struct resource *r,
5817     struct resource_map *map)
5818 {
5819 #ifdef PCI_IOV
5820 	struct pci_devinfo *dinfo;
5821 
5822 	if (device_get_parent(child) != dev)
5823 		return (bus_generic_unmap_resource(dev, child, r, map));
5824 
5825 	dinfo = device_get_ivars(child);
5826 	if (dinfo->cfg.flags & PCICFG_VF) {
5827 		switch (rman_get_type(r)) {
5828 		/* VFs can't have I/O BARs. */
5829 		case SYS_RES_IOPORT:
5830 			return (EINVAL);
5831 		case SYS_RES_MEMORY:
5832 			return (pci_vf_unmap_mem_resource(dev, child, r, map));
5833 		}
5834 
5835 		/* Fall through for other types of resource allocations. */
5836 	}
5837 #endif
5838 
5839 	return (bus_generic_unmap_resource(dev, child, r, map));
5840 }
5841 
5842 void
pci_child_deleted(device_t dev,device_t child)5843 pci_child_deleted(device_t dev, device_t child)
5844 {
5845 	struct resource_list_entry *rle;
5846 	struct resource_list *rl;
5847 	struct pci_devinfo *dinfo;
5848 
5849 	dinfo = device_get_ivars(child);
5850 	rl = &dinfo->resources;
5851 
5852 	EVENTHANDLER_INVOKE(pci_delete_device, child);
5853 
5854 	/* Turn off access to resources we're about to free */
5855 	if (bus_child_present(child) != 0) {
5856 		pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5857 		    PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5858 
5859 		pci_disable_busmaster(child);
5860 	}
5861 
5862 	/* Free all allocated resources */
5863 	STAILQ_FOREACH(rle, rl, link) {
5864 		if (rle->res) {
5865 			if (rman_get_flags(rle->res) & RF_ACTIVE ||
5866 			    resource_list_busy(rl, rle->type, rle->rid)) {
5867 				pci_printf(&dinfo->cfg,
5868 				    "Resource still owned, oops. "
5869 				    "(type=%d, rid=%d, addr=%lx)\n",
5870 				    rle->type, rle->rid,
5871 				    rman_get_start(rle->res));
5872 				bus_release_resource(child, rle->type, rle->rid,
5873 				    rle->res);
5874 			}
5875 			resource_list_unreserve(rl, dev, child, rle->type,
5876 			    rle->rid);
5877 		}
5878 	}
5879 	resource_list_free(rl);
5880 
5881 	pci_freecfg(dinfo);
5882 }
5883 
5884 void
pci_delete_resource(device_t dev,device_t child,int type,int rid)5885 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5886 {
5887 	struct pci_devinfo *dinfo;
5888 	struct resource_list *rl;
5889 	struct resource_list_entry *rle;
5890 
5891 	if (device_get_parent(child) != dev)
5892 		return;
5893 
5894 	dinfo = device_get_ivars(child);
5895 	rl = &dinfo->resources;
5896 	rle = resource_list_find(rl, type, rid);
5897 	if (rle == NULL)
5898 		return;
5899 
5900 	if (rle->res) {
5901 		if (rman_get_flags(rle->res) & RF_ACTIVE ||
5902 		    resource_list_busy(rl, type, rid)) {
5903 			device_printf(dev, "delete_resource: "
5904 			    "Resource still owned by child, oops. "
5905 			    "(type=%d, rid=%d, addr=%jx)\n",
5906 			    type, rid, rman_get_start(rle->res));
5907 			return;
5908 		}
5909 		resource_list_unreserve(rl, dev, child, type, rid);
5910 	}
5911 	resource_list_delete(rl, type, rid);
5912 }
5913 
5914 struct resource_list *
pci_get_resource_list(device_t dev,device_t child)5915 pci_get_resource_list (device_t dev, device_t child)
5916 {
5917 	struct pci_devinfo *dinfo = device_get_ivars(child);
5918 
5919 	return (&dinfo->resources);
5920 }
5921 
5922 #ifdef IOMMU
5923 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)5924 pci_get_dma_tag(device_t bus, device_t dev)
5925 {
5926 	bus_dma_tag_t tag;
5927 	struct pci_softc *sc;
5928 
5929 	if (device_get_parent(dev) == bus) {
5930 		/* try iommu and return if it works */
5931 		tag = iommu_get_dma_tag(bus, dev);
5932 	} else
5933 		tag = NULL;
5934 	if (tag == NULL) {
5935 		sc = device_get_softc(bus);
5936 		tag = sc->sc_dma_tag;
5937 	}
5938 	return (tag);
5939 }
5940 #else
5941 bus_dma_tag_t
pci_get_dma_tag(device_t bus,device_t dev)5942 pci_get_dma_tag(device_t bus, device_t dev)
5943 {
5944 	struct pci_softc *sc = device_get_softc(bus);
5945 
5946 	return (sc->sc_dma_tag);
5947 }
5948 #endif
5949 
5950 uint32_t
pci_read_config_method(device_t dev,device_t child,int reg,int width)5951 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5952 {
5953 	struct pci_devinfo *dinfo = device_get_ivars(child);
5954 	pcicfgregs *cfg = &dinfo->cfg;
5955 
5956 #ifdef PCI_IOV
5957 	/*
5958 	 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5959 	 * emulate them here.
5960 	 */
5961 	if (cfg->flags & PCICFG_VF) {
5962 		if (reg == PCIR_VENDOR) {
5963 			switch (width) {
5964 			case 4:
5965 				return (cfg->device << 16 | cfg->vendor);
5966 			case 2:
5967 				return (cfg->vendor);
5968 			case 1:
5969 				return (cfg->vendor & 0xff);
5970 			default:
5971 				return (0xffffffff);
5972 			}
5973 		} else if (reg == PCIR_DEVICE) {
5974 			switch (width) {
5975 			/* Note that an unaligned 4-byte read is an error. */
5976 			case 2:
5977 				return (cfg->device);
5978 			case 1:
5979 				return (cfg->device & 0xff);
5980 			default:
5981 				return (0xffffffff);
5982 			}
5983 		}
5984 	}
5985 #endif
5986 
5987 	return (PCIB_READ_CONFIG(device_get_parent(dev),
5988 	    cfg->bus, cfg->slot, cfg->func, reg, width));
5989 }
5990 
5991 void
pci_write_config_method(device_t dev,device_t child,int reg,uint32_t val,int width)5992 pci_write_config_method(device_t dev, device_t child, int reg,
5993     uint32_t val, int width)
5994 {
5995 	struct pci_devinfo *dinfo = device_get_ivars(child);
5996 	pcicfgregs *cfg = &dinfo->cfg;
5997 
5998 	PCIB_WRITE_CONFIG(device_get_parent(dev),
5999 	    cfg->bus, cfg->slot, cfg->func, reg, val, width);
6000 }
6001 
6002 int
pci_child_location_method(device_t dev,device_t child,struct sbuf * sb)6003 pci_child_location_method(device_t dev, device_t child, struct sbuf *sb)
6004 {
6005 
6006 	sbuf_printf(sb, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
6007 	    pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
6008 	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
6009 	return (0);
6010 }
6011 
6012 int
pci_child_pnpinfo_method(device_t dev,device_t child,struct sbuf * sb)6013 pci_child_pnpinfo_method(device_t dev, device_t child, struct sbuf *sb)
6014 {
6015 	struct pci_devinfo *dinfo;
6016 	pcicfgregs *cfg;
6017 
6018 	dinfo = device_get_ivars(child);
6019 	cfg = &dinfo->cfg;
6020 	sbuf_printf(sb, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
6021 	    "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
6022 	    cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
6023 	    cfg->progif);
6024 	return (0);
6025 }
6026 
6027 int
pci_get_device_path_method(device_t bus,device_t child,const char * locator,struct sbuf * sb)6028 pci_get_device_path_method(device_t bus, device_t child, const char *locator,
6029     struct sbuf *sb)
6030 {
6031 	device_t parent = device_get_parent(bus);
6032 	int rv;
6033 
6034 	if (strcmp(locator, BUS_LOCATOR_UEFI) == 0) {
6035 		rv = bus_generic_get_device_path(parent, bus, locator, sb);
6036 		if (rv == 0) {
6037 			sbuf_printf(sb, "/Pci(0x%x,0x%x)", pci_get_slot(child),
6038 			    pci_get_function(child));
6039 		}
6040 		return (0);
6041 	}
6042 	return (bus_generic_get_device_path(bus, child, locator, sb));
6043 }
6044 
6045 int
pci_assign_interrupt_method(device_t dev,device_t child)6046 pci_assign_interrupt_method(device_t dev, device_t child)
6047 {
6048 	struct pci_devinfo *dinfo = device_get_ivars(child);
6049 	pcicfgregs *cfg = &dinfo->cfg;
6050 
6051 	return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
6052 	    cfg->intpin));
6053 }
6054 
6055 static void
pci_lookup(void * arg,const char * name,device_t * dev)6056 pci_lookup(void *arg, const char *name, device_t *dev)
6057 {
6058 	long val;
6059 	char *end;
6060 	int domain, bus, slot, func;
6061 
6062 	if (*dev != NULL)
6063 		return;
6064 
6065 	/*
6066 	 * Accept pciconf-style selectors of either pciD:B:S:F or
6067 	 * pciB:S:F.  In the latter case, the domain is assumed to
6068 	 * be zero.
6069 	 */
6070 	if (strncmp(name, "pci", 3) != 0)
6071 		return;
6072 	val = strtol(name + 3, &end, 10);
6073 	if (val < 0 || val > INT_MAX || *end != ':')
6074 		return;
6075 	domain = val;
6076 	val = strtol(end + 1, &end, 10);
6077 	if (val < 0 || val > INT_MAX || *end != ':')
6078 		return;
6079 	bus = val;
6080 	val = strtol(end + 1, &end, 10);
6081 	if (val < 0 || val > INT_MAX)
6082 		return;
6083 	slot = val;
6084 	if (*end == ':') {
6085 		val = strtol(end + 1, &end, 10);
6086 		if (val < 0 || val > INT_MAX || *end != '\0')
6087 			return;
6088 		func = val;
6089 	} else if (*end == '\0') {
6090 		func = slot;
6091 		slot = bus;
6092 		bus = domain;
6093 		domain = 0;
6094 	} else
6095 		return;
6096 
6097 	if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
6098 	    func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
6099 		return;
6100 
6101 	*dev = pci_find_dbsf(domain, bus, slot, func);
6102 }
6103 
6104 static int
pci_modevent(module_t mod,int what,void * arg)6105 pci_modevent(module_t mod, int what, void *arg)
6106 {
6107 	static struct cdev *pci_cdev;
6108 	static eventhandler_tag tag;
6109 
6110 	switch (what) {
6111 	case MOD_LOAD:
6112 		STAILQ_INIT(&pci_devq);
6113 		pci_generation = 0;
6114 		pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
6115 		    "pci");
6116 		pci_load_vendor_data();
6117 		tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
6118 		    1000);
6119 		break;
6120 
6121 	case MOD_UNLOAD:
6122 		if (tag != NULL)
6123 			EVENTHANDLER_DEREGISTER(dev_lookup, tag);
6124 		destroy_dev(pci_cdev);
6125 		break;
6126 	}
6127 
6128 	return (0);
6129 }
6130 
6131 static void
pci_cfg_restore_pcie(device_t dev,struct pci_devinfo * dinfo)6132 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
6133 {
6134 #define	WREG(n, v)	pci_write_config(dev, pos + (n), (v), 2)
6135 	struct pcicfg_pcie *cfg;
6136 	int version, pos;
6137 
6138 	cfg = &dinfo->cfg.pcie;
6139 	pos = cfg->pcie_location;
6140 
6141 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6142 
6143 	WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
6144 
6145 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6146 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6147 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6148 		WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
6149 
6150 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6151 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6152 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6153 		WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
6154 
6155 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6156 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6157 		WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
6158 
6159 	if (version > 1) {
6160 		WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
6161 		WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
6162 		WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
6163 	}
6164 #undef WREG
6165 }
6166 
6167 static void
pci_cfg_restore_pcix(device_t dev,struct pci_devinfo * dinfo)6168 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
6169 {
6170 	pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
6171 	    dinfo->cfg.pcix.pcix_command,  2);
6172 }
6173 
6174 void
pci_cfg_restore(device_t dev,struct pci_devinfo * dinfo)6175 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
6176 {
6177 
6178 	/*
6179 	 * Restore the device to full power mode.  We must do this
6180 	 * before we restore the registers because moving from D3 to
6181 	 * D0 will cause the chip's BARs and some other registers to
6182 	 * be reset to some unknown power on reset values.  Cut down
6183 	 * the noise on boot by doing nothing if we are already in
6184 	 * state D0.
6185 	 */
6186 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
6187 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6188 	pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
6189 	pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
6190 	pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
6191 	pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
6192 	pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
6193 	pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
6194 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6195 	case PCIM_HDRTYPE_NORMAL:
6196 		pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
6197 		pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
6198 		break;
6199 	case PCIM_HDRTYPE_BRIDGE:
6200 		pci_write_config(dev, PCIR_SECLAT_1,
6201 		    dinfo->cfg.bridge.br_seclat, 1);
6202 		pci_write_config(dev, PCIR_SUBBUS_1,
6203 		    dinfo->cfg.bridge.br_subbus, 1);
6204 		pci_write_config(dev, PCIR_SECBUS_1,
6205 		    dinfo->cfg.bridge.br_secbus, 1);
6206 		pci_write_config(dev, PCIR_PRIBUS_1,
6207 		    dinfo->cfg.bridge.br_pribus, 1);
6208 		pci_write_config(dev, PCIR_BRIDGECTL_1,
6209 		    dinfo->cfg.bridge.br_control, 2);
6210 		break;
6211 	case PCIM_HDRTYPE_CARDBUS:
6212 		pci_write_config(dev, PCIR_SECLAT_2,
6213 		    dinfo->cfg.bridge.br_seclat, 1);
6214 		pci_write_config(dev, PCIR_SUBBUS_2,
6215 		    dinfo->cfg.bridge.br_subbus, 1);
6216 		pci_write_config(dev, PCIR_SECBUS_2,
6217 		    dinfo->cfg.bridge.br_secbus, 1);
6218 		pci_write_config(dev, PCIR_PRIBUS_2,
6219 		    dinfo->cfg.bridge.br_pribus, 1);
6220 		pci_write_config(dev, PCIR_BRIDGECTL_2,
6221 		    dinfo->cfg.bridge.br_control, 2);
6222 		break;
6223 	}
6224 	pci_restore_bars(dev);
6225 
6226 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_BRIDGE)
6227 		pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
6228 
6229 	/*
6230 	 * Restore extended capabilities for PCI-Express and PCI-X
6231 	 */
6232 	if (dinfo->cfg.pcie.pcie_location != 0)
6233 		pci_cfg_restore_pcie(dev, dinfo);
6234 	if (dinfo->cfg.pcix.pcix_location != 0)
6235 		pci_cfg_restore_pcix(dev, dinfo);
6236 
6237 	/* Restore MSI and MSI-X configurations if they are present. */
6238 	if (dinfo->cfg.msi.msi_location != 0)
6239 		pci_resume_msi(dev);
6240 	if (dinfo->cfg.msix.msix_location != 0)
6241 		pci_resume_msix(dev);
6242 
6243 #ifdef PCI_IOV
6244 	if (dinfo->cfg.iov != NULL)
6245 		pci_iov_cfg_restore(dev, dinfo);
6246 #endif
6247 }
6248 
6249 static void
pci_cfg_save_pcie(device_t dev,struct pci_devinfo * dinfo)6250 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6251 {
6252 #define	RREG(n)	pci_read_config(dev, pos + (n), 2)
6253 	struct pcicfg_pcie *cfg;
6254 	int version, pos;
6255 
6256 	cfg = &dinfo->cfg.pcie;
6257 	pos = cfg->pcie_location;
6258 
6259 	cfg->pcie_flags = RREG(PCIER_FLAGS);
6260 
6261 	version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6262 
6263 	cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6264 
6265 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6266 	    cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6267 	    cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6268 		cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6269 
6270 	if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6271 	    (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6272 	     (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6273 		cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6274 
6275 	if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6276 	    cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6277 		cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6278 
6279 	if (version > 1) {
6280 		cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6281 		cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6282 		cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6283 	}
6284 #undef RREG
6285 }
6286 
6287 static void
pci_cfg_save_pcix(device_t dev,struct pci_devinfo * dinfo)6288 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6289 {
6290 	dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6291 	    dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6292 }
6293 
6294 void
pci_cfg_save(device_t dev,struct pci_devinfo * dinfo,int setstate)6295 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6296 {
6297 	uint32_t cls;
6298 	int ps;
6299 
6300 	/*
6301 	 * Some drivers apparently write to these registers w/o updating our
6302 	 * cached copy.  No harm happens if we update the copy, so do so here
6303 	 * so we can restore them.  The COMMAND register is modified by the
6304 	 * bus w/o updating the cache.  This should represent the normally
6305 	 * writable portion of the 'defined' part of type 0/1/2 headers.
6306 	 */
6307 	dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6308 	dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6309 	dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6310 	dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6311 	dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6312 	dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6313 	dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6314 	dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6315 	dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6316 	dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6317 	dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6318 	switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6319 	case PCIM_HDRTYPE_NORMAL:
6320 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6321 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6322 		dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6323 		dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6324 		break;
6325 	case PCIM_HDRTYPE_BRIDGE:
6326 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6327 		    PCIR_SECLAT_1, 1);
6328 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6329 		    PCIR_SUBBUS_1, 1);
6330 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6331 		    PCIR_SECBUS_1, 1);
6332 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6333 		    PCIR_PRIBUS_1, 1);
6334 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6335 		    PCIR_BRIDGECTL_1, 2);
6336 		break;
6337 	case PCIM_HDRTYPE_CARDBUS:
6338 		dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6339 		    PCIR_SECLAT_2, 1);
6340 		dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6341 		    PCIR_SUBBUS_2, 1);
6342 		dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6343 		    PCIR_SECBUS_2, 1);
6344 		dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6345 		    PCIR_PRIBUS_2, 1);
6346 		dinfo->cfg.bridge.br_control = pci_read_config(dev,
6347 		    PCIR_BRIDGECTL_2, 2);
6348 		dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6349 		dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6350 		break;
6351 	}
6352 
6353 	if (dinfo->cfg.pcie.pcie_location != 0)
6354 		pci_cfg_save_pcie(dev, dinfo);
6355 
6356 	if (dinfo->cfg.pcix.pcix_location != 0)
6357 		pci_cfg_save_pcix(dev, dinfo);
6358 
6359 #ifdef PCI_IOV
6360 	if (dinfo->cfg.iov != NULL)
6361 		pci_iov_cfg_save(dev, dinfo);
6362 #endif
6363 
6364 	/*
6365 	 * don't set the state for display devices, base peripherals and
6366 	 * memory devices since bad things happen when they are powered down.
6367 	 * We should (a) have drivers that can easily detach and (b) use
6368 	 * generic drivers for these devices so that some device actually
6369 	 * attaches.  We need to make sure that when we implement (a) we don't
6370 	 * power the device down on a reattach.
6371 	 */
6372 	cls = pci_get_class(dev);
6373 	if (!setstate)
6374 		return;
6375 	switch (pci_do_power_nodriver)
6376 	{
6377 		case 0:		/* NO powerdown at all */
6378 			return;
6379 		case 1:		/* Conservative about what to power down */
6380 			if (cls == PCIC_STORAGE)
6381 				return;
6382 			/*FALLTHROUGH*/
6383 		case 2:		/* Aggressive about what to power down */
6384 			if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6385 			    cls == PCIC_BASEPERIPH)
6386 				return;
6387 			/*FALLTHROUGH*/
6388 		case 3:		/* Power down everything */
6389 			break;
6390 	}
6391 	/*
6392 	 * PCI spec says we can only go into D3 state from D0 state.
6393 	 * Transition from D[12] into D0 before going to D3 state.
6394 	 */
6395 	ps = pci_get_powerstate(dev);
6396 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6397 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6398 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6399 		pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6400 }
6401 
6402 /* Wrapper APIs suitable for device driver use. */
6403 void
pci_save_state(device_t dev)6404 pci_save_state(device_t dev)
6405 {
6406 	struct pci_devinfo *dinfo;
6407 
6408 	dinfo = device_get_ivars(dev);
6409 	pci_cfg_save(dev, dinfo, 0);
6410 }
6411 
6412 void
pci_restore_state(device_t dev)6413 pci_restore_state(device_t dev)
6414 {
6415 	struct pci_devinfo *dinfo;
6416 
6417 	dinfo = device_get_ivars(dev);
6418 	pci_cfg_restore(dev, dinfo);
6419 }
6420 
6421 static int
pci_get_id_method(device_t dev,device_t child,enum pci_id_type type,uintptr_t * id)6422 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6423     uintptr_t *id)
6424 {
6425 
6426 	return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6427 }
6428 
6429 /* Find the upstream port of a given PCI device in a root complex. */
6430 device_t
pci_find_pcie_root_port(device_t dev)6431 pci_find_pcie_root_port(device_t dev)
6432 {
6433 	struct pci_devinfo *dinfo;
6434 	devclass_t pci_class;
6435 	device_t pcib, bus;
6436 
6437 	pci_class = devclass_find("pci");
6438 	KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6439 	    ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6440 
6441 	/*
6442 	 * Walk the bridge hierarchy until we find a PCI-e root
6443 	 * port or a non-PCI device.
6444 	 */
6445 	for (;;) {
6446 		bus = device_get_parent(dev);
6447 		KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6448 		    device_get_nameunit(dev)));
6449 
6450 		pcib = device_get_parent(bus);
6451 		KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6452 		    device_get_nameunit(bus)));
6453 
6454 		/*
6455 		 * pcib's parent must be a PCI bus for this to be a
6456 		 * PCI-PCI bridge.
6457 		 */
6458 		if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6459 			return (NULL);
6460 
6461 		dinfo = device_get_ivars(pcib);
6462 		if (dinfo->cfg.pcie.pcie_location != 0 &&
6463 		    dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6464 			return (pcib);
6465 
6466 		dev = pcib;
6467 	}
6468 }
6469 
6470 /*
6471  * Wait for pending transactions to complete on a PCI-express function.
6472  *
6473  * The maximum delay is specified in milliseconds in max_delay.  Note
6474  * that this function may sleep.
6475  *
6476  * Returns true if the function is idle and false if the timeout is
6477  * exceeded.  If dev is not a PCI-express function, this returns true.
6478  */
6479 bool
pcie_wait_for_pending_transactions(device_t dev,u_int max_delay)6480 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6481 {
6482 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6483 	uint16_t sta;
6484 	int cap;
6485 
6486 	cap = dinfo->cfg.pcie.pcie_location;
6487 	if (cap == 0)
6488 		return (true);
6489 
6490 	sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6491 	while (sta & PCIEM_STA_TRANSACTION_PND) {
6492 		if (max_delay == 0)
6493 			return (false);
6494 
6495 		/* Poll once every 100 milliseconds up to the timeout. */
6496 		if (max_delay > 100) {
6497 			pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6498 			max_delay -= 100;
6499 		} else {
6500 			pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6501 			    C_HARDCLOCK);
6502 			max_delay = 0;
6503 		}
6504 		sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6505 	}
6506 
6507 	return (true);
6508 }
6509 
6510 /*
6511  * Determine the maximum Completion Timeout in microseconds.
6512  *
6513  * For non-PCI-express functions this returns 0.
6514  */
6515 int
pcie_get_max_completion_timeout(device_t dev)6516 pcie_get_max_completion_timeout(device_t dev)
6517 {
6518 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6519 	int cap;
6520 
6521 	cap = dinfo->cfg.pcie.pcie_location;
6522 	if (cap == 0)
6523 		return (0);
6524 
6525 	/*
6526 	 * Functions using the 1.x spec use the default timeout range of
6527 	 * 50 microseconds to 50 milliseconds.  Functions that do not
6528 	 * support programmable timeouts also use this range.
6529 	 */
6530 	if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6531 	    (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6532 	    PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6533 		return (50 * 1000);
6534 
6535 	switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6536 	    PCIEM_CTL2_COMP_TIMO_VAL) {
6537 	case PCIEM_CTL2_COMP_TIMO_100US:
6538 		return (100);
6539 	case PCIEM_CTL2_COMP_TIMO_10MS:
6540 		return (10 * 1000);
6541 	case PCIEM_CTL2_COMP_TIMO_55MS:
6542 		return (55 * 1000);
6543 	case PCIEM_CTL2_COMP_TIMO_210MS:
6544 		return (210 * 1000);
6545 	case PCIEM_CTL2_COMP_TIMO_900MS:
6546 		return (900 * 1000);
6547 	case PCIEM_CTL2_COMP_TIMO_3500MS:
6548 		return (3500 * 1000);
6549 	case PCIEM_CTL2_COMP_TIMO_13S:
6550 		return (13 * 1000 * 1000);
6551 	case PCIEM_CTL2_COMP_TIMO_64S:
6552 		return (64 * 1000 * 1000);
6553 	default:
6554 		return (50 * 1000);
6555 	}
6556 }
6557 
6558 void
pcie_apei_error(device_t dev,int sev,uint8_t * aerp)6559 pcie_apei_error(device_t dev, int sev, uint8_t *aerp)
6560 {
6561 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6562 	const char *s;
6563 	int aer;
6564 	uint32_t r, r1;
6565 	uint16_t rs;
6566 
6567 	if (sev == PCIEM_STA_CORRECTABLE_ERROR)
6568 		s = "Correctable";
6569 	else if (sev == PCIEM_STA_NON_FATAL_ERROR)
6570 		s = "Uncorrectable (Non-Fatal)";
6571 	else
6572 		s = "Uncorrectable (Fatal)";
6573 	device_printf(dev, "%s PCIe error reported by APEI\n", s);
6574 	if (aerp) {
6575 		if (sev == PCIEM_STA_CORRECTABLE_ERROR) {
6576 			r = le32dec(aerp + PCIR_AER_COR_STATUS);
6577 			r1 = le32dec(aerp + PCIR_AER_COR_MASK);
6578 		} else {
6579 			r = le32dec(aerp + PCIR_AER_UC_STATUS);
6580 			r1 = le32dec(aerp + PCIR_AER_UC_MASK);
6581 		}
6582 		device_printf(dev, "status 0x%08x mask 0x%08x", r, r1);
6583 		if (sev != PCIEM_STA_CORRECTABLE_ERROR) {
6584 			r = le32dec(aerp + PCIR_AER_UC_SEVERITY);
6585 			rs = le16dec(aerp + PCIR_AER_CAP_CONTROL);
6586 			printf(" severity 0x%08x first %d\n",
6587 			    r, rs & 0x1f);
6588 		} else
6589 			printf("\n");
6590 	}
6591 
6592 	/* As kind of recovery just report and clear the error statuses. */
6593 	if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6594 		r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6595 		if (r != 0) {
6596 			pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6597 			device_printf(dev, "Clearing UC AER errors 0x%08x\n", r);
6598 		}
6599 
6600 		r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6601 		if (r != 0) {
6602 			pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6603 			device_printf(dev, "Clearing COR AER errors 0x%08x\n", r);
6604 		}
6605 	}
6606 	if (dinfo->cfg.pcie.pcie_location != 0) {
6607 		rs = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6608 		    PCIER_DEVICE_STA, 2);
6609 		if ((rs & (PCIEM_STA_CORRECTABLE_ERROR |
6610 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6611 		    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6612 			pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6613 			    PCIER_DEVICE_STA, rs, 2);
6614 			device_printf(dev, "Clearing PCIe errors 0x%04x\n", rs);
6615 		}
6616 	}
6617 }
6618 
6619 /*
6620  * Perform a Function Level Reset (FLR) on a device.
6621  *
6622  * This function first waits for any pending transactions to complete
6623  * within the timeout specified by max_delay.  If transactions are
6624  * still pending, the function will return false without attempting a
6625  * reset.
6626  *
6627  * If dev is not a PCI-express function or does not support FLR, this
6628  * function returns false.
6629  *
6630  * Note that no registers are saved or restored.  The caller is
6631  * responsible for saving and restoring any registers including
6632  * PCI-standard registers via pci_save_state() and
6633  * pci_restore_state().
6634  */
6635 bool
pcie_flr(device_t dev,u_int max_delay,bool force)6636 pcie_flr(device_t dev, u_int max_delay, bool force)
6637 {
6638 	struct pci_devinfo *dinfo = device_get_ivars(dev);
6639 	uint16_t cmd, ctl;
6640 	int compl_delay;
6641 	int cap;
6642 
6643 	cap = dinfo->cfg.pcie.pcie_location;
6644 	if (cap == 0)
6645 		return (false);
6646 
6647 	if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6648 		return (false);
6649 
6650 	/*
6651 	 * Disable busmastering to prevent generation of new
6652 	 * transactions while waiting for the device to go idle.  If
6653 	 * the idle timeout fails, the command register is restored
6654 	 * which will re-enable busmastering.
6655 	 */
6656 	cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6657 	pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6658 	if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6659 		if (!force) {
6660 			pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6661 			return (false);
6662 		}
6663 		pci_printf(&dinfo->cfg,
6664 		    "Resetting with transactions pending after %d ms\n",
6665 		    max_delay);
6666 
6667 		/*
6668 		 * Extend the post-FLR delay to cover the maximum
6669 		 * Completion Timeout delay of anything in flight
6670 		 * during the FLR delay.  Enforce a minimum delay of
6671 		 * at least 10ms.
6672 		 */
6673 		compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6674 		if (compl_delay < 10)
6675 			compl_delay = 10;
6676 	} else
6677 		compl_delay = 0;
6678 
6679 	/* Initiate the reset. */
6680 	ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6681 	pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6682 	    PCIEM_CTL_INITIATE_FLR, 2);
6683 
6684 	/* Wait for 100ms. */
6685 	pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6686 
6687 	if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6688 	    PCIEM_STA_TRANSACTION_PND)
6689 		pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6690 	return (true);
6691 }
6692 
6693 /*
6694  * Attempt a power-management reset by cycling the device in/out of D3
6695  * state.  PCI spec says we can only go into D3 state from D0 state.
6696  * Transition from D[12] into D0 before going to D3 state.
6697  */
6698 int
pci_power_reset(device_t dev)6699 pci_power_reset(device_t dev)
6700 {
6701 	int ps;
6702 
6703 	ps = pci_get_powerstate(dev);
6704 	if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6705 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6706 	pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6707 	pci_set_powerstate(dev, ps);
6708 	return (0);
6709 }
6710 
6711 /*
6712  * Try link drop and retrain of the downstream port of upstream
6713  * switch, for PCIe.  According to the PCIe 3.0 spec 6.6.1, this must
6714  * cause Conventional Hot reset of the device in the slot.
6715  * Alternative, for PCIe, could be the secondary bus reset initiatied
6716  * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6717  */
6718 int
pcie_link_reset(device_t port,int pcie_location)6719 pcie_link_reset(device_t port, int pcie_location)
6720 {
6721 	uint16_t v;
6722 
6723 	v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6724 	v |= PCIEM_LINK_CTL_LINK_DIS;
6725 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6726 	pause_sbt("pcier1", mstosbt(20), 0, 0);
6727 	v &= ~PCIEM_LINK_CTL_LINK_DIS;
6728 	v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6729 	pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6730 	pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6731 	v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6732 	return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6733 }
6734 
6735 static int
pci_reset_post(device_t dev,device_t child)6736 pci_reset_post(device_t dev, device_t child)
6737 {
6738 
6739 	if (dev == device_get_parent(child))
6740 		pci_restore_state(child);
6741 	return (0);
6742 }
6743 
6744 static int
pci_reset_prepare(device_t dev,device_t child)6745 pci_reset_prepare(device_t dev, device_t child)
6746 {
6747 
6748 	if (dev == device_get_parent(child))
6749 		pci_save_state(child);
6750 	return (0);
6751 }
6752 
6753 static int
pci_reset_child(device_t dev,device_t child,int flags)6754 pci_reset_child(device_t dev, device_t child, int flags)
6755 {
6756 	int error;
6757 
6758 	if (dev == NULL || device_get_parent(child) != dev)
6759 		return (0);
6760 	if ((flags & DEVF_RESET_DETACH) != 0) {
6761 		error = device_get_state(child) == DS_ATTACHED ?
6762 		    device_detach(child) : 0;
6763 	} else {
6764 		error = BUS_SUSPEND_CHILD(dev, child);
6765 	}
6766 	if (error == 0) {
6767 		if (!pcie_flr(child, 1000, false)) {
6768 			error = BUS_RESET_PREPARE(dev, child);
6769 			if (error == 0)
6770 				pci_power_reset(child);
6771 			BUS_RESET_POST(dev, child);
6772 		}
6773 		if ((flags & DEVF_RESET_DETACH) != 0)
6774 			device_probe_and_attach(child);
6775 		else
6776 			BUS_RESUME_CHILD(dev, child);
6777 	}
6778 	return (error);
6779 }
6780 
6781 const struct pci_device_table *
pci_match_device(device_t child,const struct pci_device_table * id,size_t nelt)6782 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6783 {
6784 	bool match;
6785 	uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6786 
6787 	vendor = pci_get_vendor(child);
6788 	device = pci_get_device(child);
6789 	subvendor = pci_get_subvendor(child);
6790 	subdevice = pci_get_subdevice(child);
6791 	class = pci_get_class(child);
6792 	subclass = pci_get_subclass(child);
6793 	revid = pci_get_revid(child);
6794 	while (nelt-- > 0) {
6795 		match = true;
6796 		if (id->match_flag_vendor)
6797 			match &= vendor == id->vendor;
6798 		if (id->match_flag_device)
6799 			match &= device == id->device;
6800 		if (id->match_flag_subvendor)
6801 			match &= subvendor == id->subvendor;
6802 		if (id->match_flag_subdevice)
6803 			match &= subdevice == id->subdevice;
6804 		if (id->match_flag_class)
6805 			match &= class == id->class_id;
6806 		if (id->match_flag_subclass)
6807 			match &= subclass == id->subclass;
6808 		if (id->match_flag_revid)
6809 			match &= revid == id->revid;
6810 		if (match)
6811 			return (id);
6812 		id++;
6813 	}
6814 	return (NULL);
6815 }
6816 
6817 static void
pci_print_faulted_dev_name(const struct pci_devinfo * dinfo)6818 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6819 {
6820 	const char *dev_name;
6821 	device_t dev;
6822 
6823 	dev = dinfo->cfg.dev;
6824 	printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6825 	    dinfo->cfg.slot, dinfo->cfg.func);
6826 	dev_name = device_get_name(dev);
6827 	if (dev_name != NULL)
6828 		printf(" (%s%d)", dev_name, device_get_unit(dev));
6829 }
6830 
6831 void
pci_print_faulted_dev(void)6832 pci_print_faulted_dev(void)
6833 {
6834 	struct pci_devinfo *dinfo;
6835 	device_t dev;
6836 	int aer, i;
6837 	uint32_t r1, r2;
6838 	uint16_t status;
6839 
6840 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6841 		dev = dinfo->cfg.dev;
6842 		status = pci_read_config(dev, PCIR_STATUS, 2);
6843 		status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6844 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6845 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6846 		if (status != 0) {
6847 			pci_print_faulted_dev_name(dinfo);
6848 			printf(" error 0x%04x\n", status);
6849 		}
6850 		if (dinfo->cfg.pcie.pcie_location != 0) {
6851 			status = pci_read_config(dev,
6852 			    dinfo->cfg.pcie.pcie_location +
6853 			    PCIER_DEVICE_STA, 2);
6854 			if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6855 			    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6856 			    PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6857 				pci_print_faulted_dev_name(dinfo);
6858 				printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6859 				    pci_read_config(dev,
6860 				    dinfo->cfg.pcie.pcie_location +
6861 				    PCIER_DEVICE_CTL, 2),
6862 				    status);
6863 			}
6864 		}
6865 		if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6866 			r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6867 			r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6868 			if (r1 != 0 || r2 != 0) {
6869 				pci_print_faulted_dev_name(dinfo);
6870 				printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6871 				    "  COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6872 				    r1, pci_read_config(dev, aer +
6873 				    PCIR_AER_UC_MASK, 4),
6874 				    pci_read_config(dev, aer +
6875 				    PCIR_AER_UC_SEVERITY, 4),
6876 				    r2, pci_read_config(dev, aer +
6877 				    PCIR_AER_COR_MASK, 4),
6878 				    pci_read_config(dev, aer +
6879 				    PCIR_AER_CAP_CONTROL, 4));
6880 				for (i = 0; i < 4; i++) {
6881 					r1 = pci_read_config(dev, aer +
6882 					    PCIR_AER_HEADER_LOG + i * 4, 4);
6883 					printf("    HL%d: 0x%08x\n", i, r1);
6884 				}
6885 			}
6886 		}
6887 	}
6888 }
6889 
6890 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(pcierr,pci_print_faulted_dev_db,DB_CMD_MEMSAFE)6891 DB_SHOW_COMMAND_FLAGS(pcierr, pci_print_faulted_dev_db, DB_CMD_MEMSAFE)
6892 {
6893 
6894 	pci_print_faulted_dev();
6895 }
6896 
6897 static void
db_clear_pcie_errors(const struct pci_devinfo * dinfo)6898 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6899 {
6900 	device_t dev;
6901 	int aer;
6902 	uint32_t r;
6903 
6904 	dev = dinfo->cfg.dev;
6905 	r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6906 	    PCIER_DEVICE_STA, 2);
6907 	pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6908 	    PCIER_DEVICE_STA, r, 2);
6909 
6910 	if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6911 		return;
6912 	r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6913 	if (r != 0)
6914 		pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6915 	r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6916 	if (r != 0)
6917 		pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6918 }
6919 
DB_COMMAND_FLAGS(pci_clearerr,db_pci_clearerr,DB_CMD_MEMSAFE)6920 DB_COMMAND_FLAGS(pci_clearerr, db_pci_clearerr, DB_CMD_MEMSAFE)
6921 {
6922 	struct pci_devinfo *dinfo;
6923 	device_t dev;
6924 	uint16_t status, status1;
6925 
6926 	STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6927 		dev = dinfo->cfg.dev;
6928 		status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6929 		status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6930 		    PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6931 		    PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6932 		if (status1 != 0) {
6933 			status &= ~status1;
6934 			pci_write_config(dev, PCIR_STATUS, status, 2);
6935 		}
6936 		if (dinfo->cfg.pcie.pcie_location != 0)
6937 			db_clear_pcie_errors(dinfo);
6938 	}
6939 }
6940 #endif
6941