2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1997, Stefan Esser <se@freebsd.org>
5 * Copyright (c) 2000, Michael Smith <msmith@freebsd.org>
6 * Copyright (c) 2000, BSDi
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice unmodified, this list of conditions, and the following
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.
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.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/limits.h>
42 #include <sys/linker.h>
43 #include <sys/fcntl.h>
45 #include <sys/kernel.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/endian.h>
52 #include <vm/vm_extern.h>
55 #include <machine/bus.h>
57 #include <machine/resource.h>
58 #include <machine/stdarg.h>
60 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
61 #include <machine/intr_machdep.h>
64 #include <sys/pciio.h>
65 #include <dev/pci/pcireg.h>
66 #include <dev/pci/pcivar.h>
67 #include <dev/pci/pci_private.h>
71 #include <dev/pci/pci_iov_private.h>
74 #include <dev/usb/controller/xhcireg.h>
75 #include <dev/usb/controller/ehcireg.h>
76 #include <dev/usb/controller/ohcireg.h>
77 #include <dev/usb/controller/uhcireg.h>
82 #define PCIR_IS_BIOS(cfg, reg) \
83 (((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) || \
84 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
86 static int pci_has_quirk(uint32_t devid, int quirk);
87 static pci_addr_t pci_mapbase(uint64_t mapreg);
88 static const char *pci_maptype(uint64_t mapreg);
89 static int pci_maprange(uint64_t mapreg);
90 static pci_addr_t pci_rombase(uint64_t mapreg);
91 static int pci_romsize(uint64_t testval);
92 static void pci_fixancient(pcicfgregs *cfg);
93 static int pci_printf(pcicfgregs *cfg, const char *fmt, ...);
95 static int pci_porten(device_t dev);
96 static int pci_memen(device_t dev);
97 static void pci_assign_interrupt(device_t bus, device_t dev,
99 static int pci_add_map(device_t bus, device_t dev, int reg,
100 struct resource_list *rl, int force, int prefetch);
101 static int pci_probe(device_t dev);
102 static int pci_attach(device_t dev);
103 static int pci_detach(device_t dev);
104 static void pci_load_vendor_data(void);
105 static int pci_describe_parse_line(char **ptr, int *vendor,
106 int *device, char **desc);
107 static char *pci_describe_device(device_t dev);
108 static int pci_modevent(module_t mod, int what, void *arg);
109 static void pci_hdrtypedata(device_t pcib, int b, int s, int f,
111 static void pci_read_cap(device_t pcib, pcicfgregs *cfg);
112 static int pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
113 int reg, uint32_t *data);
115 static int pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
116 int reg, uint32_t data);
118 static void pci_read_vpd(device_t pcib, pcicfgregs *cfg);
119 static void pci_mask_msix(device_t dev, u_int index);
120 static void pci_unmask_msix(device_t dev, u_int index);
121 static int pci_msi_blacklisted(void);
122 static int pci_msix_blacklisted(void);
123 static void pci_resume_msi(device_t dev);
124 static void pci_resume_msix(device_t dev);
125 static int pci_remap_intr_method(device_t bus, device_t dev,
127 static void pci_hint_device_unit(device_t acdev, device_t child,
128 const char *name, int *unitp);
129 static int pci_reset_post(device_t dev, device_t child);
130 static int pci_reset_prepare(device_t dev, device_t child);
131 static int pci_reset_child(device_t dev, device_t child,
134 static int pci_get_id_method(device_t dev, device_t child,
135 enum pci_id_type type, uintptr_t *rid);
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);
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),
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),
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, bus_generic_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_child_deleted, pci_child_deleted),
172 DEVMETHOD(bus_child_detached, pci_child_detached),
173 DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
174 DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
175 DEVMETHOD(bus_hint_device_unit, pci_hint_device_unit),
176 DEVMETHOD(bus_remap_intr, pci_remap_intr_method),
177 DEVMETHOD(bus_suspend_child, pci_suspend_child),
178 DEVMETHOD(bus_resume_child, pci_resume_child),
179 DEVMETHOD(bus_rescan, pci_rescan_method),
182 DEVMETHOD(pci_read_config, pci_read_config_method),
183 DEVMETHOD(pci_write_config, pci_write_config_method),
184 DEVMETHOD(pci_enable_busmaster, pci_enable_busmaster_method),
185 DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
186 DEVMETHOD(pci_enable_io, pci_enable_io_method),
187 DEVMETHOD(pci_disable_io, pci_disable_io_method),
188 DEVMETHOD(pci_get_vpd_ident, pci_get_vpd_ident_method),
189 DEVMETHOD(pci_get_vpd_readonly, pci_get_vpd_readonly_method),
190 DEVMETHOD(pci_get_powerstate, pci_get_powerstate_method),
191 DEVMETHOD(pci_set_powerstate, pci_set_powerstate_method),
192 DEVMETHOD(pci_assign_interrupt, pci_assign_interrupt_method),
193 DEVMETHOD(pci_find_cap, pci_find_cap_method),
194 DEVMETHOD(pci_find_next_cap, pci_find_next_cap_method),
195 DEVMETHOD(pci_find_extcap, pci_find_extcap_method),
196 DEVMETHOD(pci_find_next_extcap, pci_find_next_extcap_method),
197 DEVMETHOD(pci_find_htcap, pci_find_htcap_method),
198 DEVMETHOD(pci_find_next_htcap, pci_find_next_htcap_method),
199 DEVMETHOD(pci_alloc_msi, pci_alloc_msi_method),
200 DEVMETHOD(pci_alloc_msix, pci_alloc_msix_method),
201 DEVMETHOD(pci_enable_msi, pci_enable_msi_method),
202 DEVMETHOD(pci_enable_msix, pci_enable_msix_method),
203 DEVMETHOD(pci_disable_msi, pci_disable_msi_method),
204 DEVMETHOD(pci_remap_msix, pci_remap_msix_method),
205 DEVMETHOD(pci_release_msi, pci_release_msi_method),
206 DEVMETHOD(pci_msi_count, pci_msi_count_method),
207 DEVMETHOD(pci_msix_count, pci_msix_count_method),
208 DEVMETHOD(pci_msix_pba_bar, pci_msix_pba_bar_method),
209 DEVMETHOD(pci_msix_table_bar, pci_msix_table_bar_method),
210 DEVMETHOD(pci_get_id, pci_get_id_method),
211 DEVMETHOD(pci_alloc_devinfo, pci_alloc_devinfo_method),
212 DEVMETHOD(pci_child_added, pci_child_added_method),
214 DEVMETHOD(pci_iov_attach, pci_iov_attach_method),
215 DEVMETHOD(pci_iov_detach, pci_iov_detach_method),
216 DEVMETHOD(pci_create_iov_child, pci_create_iov_child_method),
222 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
224 static devclass_t pci_devclass;
225 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
227 MODULE_VERSION(pci, 1);
229 static char *pci_vendordata;
230 static size_t pci_vendordata_size;
233 uint32_t devid; /* Vendor/device of the card */
235 #define PCI_QUIRK_MAP_REG 1 /* PCI map register in weird place */
236 #define PCI_QUIRK_DISABLE_MSI 2 /* Neither MSI nor MSI-X work */
237 #define PCI_QUIRK_ENABLE_MSI_VM 3 /* Older chipset in VM where MSI works */
238 #define PCI_QUIRK_UNMAP_REG 4 /* Ignore PCI map register */
239 #define PCI_QUIRK_DISABLE_MSIX 5 /* MSI-X doesn't work */
240 #define PCI_QUIRK_MSI_INTX_BUG 6 /* PCIM_CMD_INTxDIS disables MSI */
241 #define PCI_QUIRK_REALLOC_BAR 7 /* Can't allocate memory at the default address */
246 static const struct pci_quirk pci_quirks[] = {
247 /* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
248 { 0x71138086, PCI_QUIRK_MAP_REG, 0x90, 0 },
249 { 0x719b8086, PCI_QUIRK_MAP_REG, 0x90, 0 },
250 /* As does the Serverworks OSB4 (the SMBus mapping register) */
251 { 0x02001166, PCI_QUIRK_MAP_REG, 0x90, 0 },
254 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
255 * or the CMIC-SL (AKA ServerWorks GC_LE).
257 { 0x00141166, PCI_QUIRK_DISABLE_MSI, 0, 0 },
258 { 0x00171166, PCI_QUIRK_DISABLE_MSI, 0, 0 },
261 * MSI doesn't work on earlier Intel chipsets including
262 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
264 { 0x25408086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
265 { 0x254c8086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
266 { 0x25508086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
267 { 0x25608086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
268 { 0x25708086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
269 { 0x25788086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
270 { 0x35808086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
273 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
276 { 0x74501022, PCI_QUIRK_DISABLE_MSI, 0, 0 },
279 * MSI-X allocation doesn't work properly for devices passed through
280 * by VMware up to at least ESXi 5.1.
282 { 0x079015ad, PCI_QUIRK_DISABLE_MSIX, 0, 0 }, /* PCI/PCI-X */
283 { 0x07a015ad, PCI_QUIRK_DISABLE_MSIX, 0, 0 }, /* PCIe */
286 * Some virtualization environments emulate an older chipset
287 * but support MSI just fine. QEMU uses the Intel 82440.
289 { 0x12378086, PCI_QUIRK_ENABLE_MSI_VM, 0, 0 },
292 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
293 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
294 * It prevents us from attaching hpet(4) when the bit is unset.
295 * Note this quirk only affects SB600 revision A13 and earlier.
296 * For SB600 A21 and later, firmware must set the bit to hide it.
297 * For SB700 and later, it is unused and hardcoded to zero.
299 { 0x43851002, PCI_QUIRK_UNMAP_REG, 0x14, 0 },
302 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
303 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
304 * of the command register is set.
306 { 0x10911969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
307 { 0xE0911969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
308 { 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
309 { 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
310 { 0x10901969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
313 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
314 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
316 { 0x166814e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5714 */
317 { 0x166914e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5714S */
318 { 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5780 */
319 { 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5780S */
320 { 0x167814e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5715 */
321 { 0x167914e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5715S */
324 * HPE Gen 10 VGA has a memory range that can't be allocated in the
327 { 0x98741002, PCI_QUIRK_REALLOC_BAR, 0, 0 },
332 /* map register information */
333 #define PCI_MAPMEM 0x01 /* memory map */
334 #define PCI_MAPMEMP 0x02 /* prefetchable memory map */
335 #define PCI_MAPPORT 0x04 /* port map */
337 struct devlist pci_devq;
338 uint32_t pci_generation;
339 uint32_t pci_numdevs = 0;
340 static int pcie_chipset, pcix_chipset;
343 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
345 static int pci_enable_io_modes = 1;
346 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
347 &pci_enable_io_modes, 1,
348 "Enable I/O and memory bits in the config register. Some BIOSes do not"
349 " enable these bits correctly. We'd like to do this all the time, but"
350 " there are some peripherals that this causes problems with.");
352 static int pci_do_realloc_bars = 1;
353 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
354 &pci_do_realloc_bars, 0,
355 "Attempt to allocate a new range for any BARs whose original "
356 "firmware-assigned ranges fail to allocate during the initial device scan.");
358 static int pci_do_power_nodriver = 0;
359 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
360 &pci_do_power_nodriver, 0,
361 "Place a function into D3 state when no driver attaches to it. 0 means"
362 " disable. 1 means conservatively place devices into D3 state. 2 means"
363 " aggressively place devices into D3 state. 3 means put absolutely"
364 " everything in D3 state.");
366 int pci_do_power_resume = 1;
367 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
368 &pci_do_power_resume, 1,
369 "Transition from D3 -> D0 on resume.");
371 int pci_do_power_suspend = 1;
372 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
373 &pci_do_power_suspend, 1,
374 "Transition from D0 -> D3 on suspend.");
376 static int pci_do_msi = 1;
377 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
378 "Enable support for MSI interrupts");
380 static int pci_do_msix = 1;
381 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
382 "Enable support for MSI-X interrupts");
384 static int pci_msix_rewrite_table = 0;
385 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
386 &pci_msix_rewrite_table, 0,
387 "Rewrite entire MSI-X table when updating MSI-X entries");
389 static int pci_honor_msi_blacklist = 1;
390 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
391 &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
393 #if defined(__i386__) || defined(__amd64__)
394 static int pci_usb_takeover = 1;
396 static int pci_usb_takeover = 0;
398 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
399 &pci_usb_takeover, 1,
400 "Enable early takeover of USB controllers. Disable this if you depend on"
401 " BIOS emulation of USB devices, that is you use USB devices (like"
402 " keyboard or mouse) but do not load USB drivers");
404 static int pci_clear_bars;
405 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
406 "Ignore firmware-assigned resources for BARs.");
408 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
409 static int pci_clear_buses;
410 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
411 "Ignore firmware-assigned bus numbers.");
414 static int pci_enable_ari = 1;
415 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
416 0, "Enable support for PCIe Alternative RID Interpretation");
418 static int pci_clear_aer_on_attach = 0;
419 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
420 &pci_clear_aer_on_attach, 0,
421 "Clear port and device AER state on driver attach");
424 pci_has_quirk(uint32_t devid, int quirk)
426 const struct pci_quirk *q;
428 for (q = &pci_quirks[0]; q->devid; q++) {
429 if (q->devid == devid && q->type == quirk)
435 /* Find a device_t by bus/slot/function in domain 0 */
438 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
441 return (pci_find_dbsf(0, bus, slot, func));
444 /* Find a device_t by domain/bus/slot/function */
447 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
449 struct pci_devinfo *dinfo;
451 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
452 if ((dinfo->cfg.domain == domain) &&
453 (dinfo->cfg.bus == bus) &&
454 (dinfo->cfg.slot == slot) &&
455 (dinfo->cfg.func == func)) {
456 return (dinfo->cfg.dev);
463 /* Find a device_t by vendor/device ID */
466 pci_find_device(uint16_t vendor, uint16_t device)
468 struct pci_devinfo *dinfo;
470 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
471 if ((dinfo->cfg.vendor == vendor) &&
472 (dinfo->cfg.device == device)) {
473 return (dinfo->cfg.dev);
481 pci_find_class(uint8_t class, uint8_t subclass)
483 struct pci_devinfo *dinfo;
485 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
486 if (dinfo->cfg.baseclass == class &&
487 dinfo->cfg.subclass == subclass) {
488 return (dinfo->cfg.dev);
496 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
501 retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
504 retval += vprintf(fmt, ap);
509 /* return base address of memory or port map */
512 pci_mapbase(uint64_t mapreg)
515 if (PCI_BAR_MEM(mapreg))
516 return (mapreg & PCIM_BAR_MEM_BASE);
518 return (mapreg & PCIM_BAR_IO_BASE);
521 /* return map type of memory or port map */
524 pci_maptype(uint64_t mapreg)
527 if (PCI_BAR_IO(mapreg))
529 if (mapreg & PCIM_BAR_MEM_PREFETCH)
530 return ("Prefetchable Memory");
534 /* return log2 of map size decoded for memory or port map */
537 pci_mapsize(uint64_t testval)
541 testval = pci_mapbase(testval);
544 while ((testval & 1) == 0)
553 /* return base address of device ROM */
556 pci_rombase(uint64_t mapreg)
559 return (mapreg & PCIM_BIOS_ADDR_MASK);
562 /* return log2 of map size decided for device ROM */
565 pci_romsize(uint64_t testval)
569 testval = pci_rombase(testval);
572 while ((testval & 1) == 0)
581 /* return log2 of address range supported by map register */
584 pci_maprange(uint64_t mapreg)
588 if (PCI_BAR_IO(mapreg))
591 switch (mapreg & PCIM_BAR_MEM_TYPE) {
592 case PCIM_BAR_MEM_32:
595 case PCIM_BAR_MEM_1MB:
598 case PCIM_BAR_MEM_64:
605 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
608 pci_fixancient(pcicfgregs *cfg)
610 if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
613 /* PCI to PCI bridges use header type 1 */
614 if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
615 cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
618 /* extract header type specific config data */
621 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
623 #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w)
624 switch (cfg->hdrtype & PCIM_HDRTYPE) {
625 case PCIM_HDRTYPE_NORMAL:
626 cfg->subvendor = REG(PCIR_SUBVEND_0, 2);
627 cfg->subdevice = REG(PCIR_SUBDEV_0, 2);
628 cfg->mingnt = REG(PCIR_MINGNT, 1);
629 cfg->maxlat = REG(PCIR_MAXLAT, 1);
630 cfg->nummaps = PCI_MAXMAPS_0;
632 case PCIM_HDRTYPE_BRIDGE:
633 cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
634 cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
635 cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
636 cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
637 cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
638 cfg->nummaps = PCI_MAXMAPS_1;
640 case PCIM_HDRTYPE_CARDBUS:
641 cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
642 cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
643 cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
644 cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
645 cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
646 cfg->subvendor = REG(PCIR_SUBVEND_2, 2);
647 cfg->subdevice = REG(PCIR_SUBDEV_2, 2);
648 cfg->nummaps = PCI_MAXMAPS_2;
654 /* read configuration header into pcicfgregs structure */
656 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
658 #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w)
661 vid = REG(PCIR_VENDOR, 2);
662 did = REG(PCIR_DEVICE, 2);
664 return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
670 pci_alloc_devinfo_method(device_t dev)
673 return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
677 static struct pci_devinfo *
678 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
679 uint16_t vid, uint16_t did)
681 struct pci_devinfo *devlist_entry;
684 devlist_entry = PCI_ALLOC_DEVINFO(bus);
686 cfg = &devlist_entry->cfg;
694 cfg->cmdreg = REG(PCIR_COMMAND, 2);
695 cfg->statreg = REG(PCIR_STATUS, 2);
696 cfg->baseclass = REG(PCIR_CLASS, 1);
697 cfg->subclass = REG(PCIR_SUBCLASS, 1);
698 cfg->progif = REG(PCIR_PROGIF, 1);
699 cfg->revid = REG(PCIR_REVID, 1);
700 cfg->hdrtype = REG(PCIR_HDRTYPE, 1);
701 cfg->cachelnsz = REG(PCIR_CACHELNSZ, 1);
702 cfg->lattimer = REG(PCIR_LATTIMER, 1);
703 cfg->intpin = REG(PCIR_INTPIN, 1);
704 cfg->intline = REG(PCIR_INTLINE, 1);
706 cfg->mfdev = (cfg->hdrtype & PCIM_MFDEV) != 0;
707 cfg->hdrtype &= ~PCIM_MFDEV;
708 STAILQ_INIT(&cfg->maps);
713 pci_hdrtypedata(pcib, b, s, f, cfg);
715 if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
716 pci_read_cap(pcib, cfg);
718 STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
720 devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
721 devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
722 devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
723 devlist_entry->conf.pc_sel.pc_func = cfg->func;
724 devlist_entry->conf.pc_hdr = cfg->hdrtype;
726 devlist_entry->conf.pc_subvendor = cfg->subvendor;
727 devlist_entry->conf.pc_subdevice = cfg->subdevice;
728 devlist_entry->conf.pc_vendor = cfg->vendor;
729 devlist_entry->conf.pc_device = cfg->device;
731 devlist_entry->conf.pc_class = cfg->baseclass;
732 devlist_entry->conf.pc_subclass = cfg->subclass;
733 devlist_entry->conf.pc_progif = cfg->progif;
734 devlist_entry->conf.pc_revid = cfg->revid;
739 return (devlist_entry);
744 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
746 #define REG(n, w) PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
747 cfg->ea.ea_location + (n), w)
754 uint64_t base, max_offset;
755 struct pci_ea_entry *eae;
757 if (cfg->ea.ea_location == 0)
760 STAILQ_INIT(&cfg->ea.ea_entries);
762 /* Determine the number of entries */
763 num_ent = REG(PCIR_EA_NUM_ENT, 2);
764 num_ent &= PCIM_EA_NUM_ENT_MASK;
766 /* Find the first entry to care of */
767 ptr = PCIR_EA_FIRST_ENT;
769 /* Skip DWORD 2 for type 1 functions */
770 if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
773 for (a = 0; a < num_ent; a++) {
775 eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
776 eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
778 /* Read a number of dwords in the entry */
781 ent_size = (val & PCIM_EA_ES);
783 for (b = 0; b < ent_size; b++) {
788 eae->eae_flags = val;
789 eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
791 base = dw[0] & PCIM_EA_FIELD_MASK;
792 max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
794 if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
795 base |= (uint64_t)dw[b] << 32UL;
798 if (((dw[1] & PCIM_EA_IS_64) != 0)
800 max_offset |= (uint64_t)dw[b] << 32UL;
804 eae->eae_base = base;
805 eae->eae_max_offset = max_offset;
807 STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
810 printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
811 cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
812 (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
819 pci_read_cap(device_t pcib, pcicfgregs *cfg)
821 #define REG(n, w) PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
822 #define WREG(n, v, w) PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
823 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
827 int ptr, nextptr, ptrptr;
829 switch (cfg->hdrtype & PCIM_HDRTYPE) {
830 case PCIM_HDRTYPE_NORMAL:
831 case PCIM_HDRTYPE_BRIDGE:
832 ptrptr = PCIR_CAP_PTR;
834 case PCIM_HDRTYPE_CARDBUS:
835 ptrptr = PCIR_CAP_PTR_2; /* cardbus capabilities ptr */
838 return; /* no extended capabilities support */
840 nextptr = REG(ptrptr, 1); /* sanity check? */
843 * Read capability entries.
845 while (nextptr != 0) {
848 printf("illegal PCI extended capability offset %d\n",
852 /* Find the next entry */
854 nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
856 /* Process this entry */
857 switch (REG(ptr + PCICAP_ID, 1)) {
858 case PCIY_PMG: /* PCI power management */
859 if (cfg->pp.pp_cap == 0) {
860 cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
861 cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
862 cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
863 if ((nextptr - ptr) > PCIR_POWER_DATA)
864 cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
867 case PCIY_HT: /* HyperTransport */
868 /* Determine HT-specific capability type. */
869 val = REG(ptr + PCIR_HT_COMMAND, 2);
871 if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
872 cfg->ht.ht_slave = ptr;
874 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
875 switch (val & PCIM_HTCMD_CAP_MASK) {
876 case PCIM_HTCAP_MSI_MAPPING:
877 if (!(val & PCIM_HTCMD_MSI_FIXED)) {
878 /* Sanity check the mapping window. */
879 addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
882 addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
884 if (addr != MSI_INTEL_ADDR_BASE)
886 "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
887 cfg->domain, cfg->bus,
888 cfg->slot, cfg->func,
891 addr = MSI_INTEL_ADDR_BASE;
893 cfg->ht.ht_msimap = ptr;
894 cfg->ht.ht_msictrl = val;
895 cfg->ht.ht_msiaddr = addr;
900 case PCIY_MSI: /* PCI MSI */
901 cfg->msi.msi_location = ptr;
902 cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
903 cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
904 PCIM_MSICTRL_MMC_MASK)>>1);
906 case PCIY_MSIX: /* PCI MSI-X */
907 cfg->msix.msix_location = ptr;
908 cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
909 cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
910 PCIM_MSIXCTRL_TABLE_SIZE) + 1;
911 val = REG(ptr + PCIR_MSIX_TABLE, 4);
912 cfg->msix.msix_table_bar = PCIR_BAR(val &
914 cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
915 val = REG(ptr + PCIR_MSIX_PBA, 4);
916 cfg->msix.msix_pba_bar = PCIR_BAR(val &
918 cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
920 case PCIY_VPD: /* PCI Vital Product Data */
921 cfg->vpd.vpd_reg = ptr;
924 /* Should always be true. */
925 if ((cfg->hdrtype & PCIM_HDRTYPE) ==
926 PCIM_HDRTYPE_BRIDGE) {
927 val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
928 cfg->subvendor = val & 0xffff;
929 cfg->subdevice = val >> 16;
932 case PCIY_PCIX: /* PCI-X */
934 * Assume we have a PCI-X chipset if we have
935 * at least one PCI-PCI bridge with a PCI-X
936 * capability. Note that some systems with
937 * PCI-express or HT chipsets might match on
938 * this check as well.
940 if ((cfg->hdrtype & PCIM_HDRTYPE) ==
943 cfg->pcix.pcix_location = ptr;
945 case PCIY_EXPRESS: /* PCI-express */
947 * Assume we have a PCI-express chipset if we have
948 * at least one PCI-express device.
951 cfg->pcie.pcie_location = ptr;
952 val = REG(ptr + PCIER_FLAGS, 2);
953 cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
955 case PCIY_EA: /* Enhanced Allocation */
956 cfg->ea.ea_location = ptr;
957 pci_ea_fill_info(pcib, cfg);
964 #if defined(__powerpc__)
966 * Enable the MSI mapping window for all HyperTransport
967 * slaves. PCI-PCI bridges have their windows enabled via
970 if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
971 !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
973 "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
974 cfg->domain, cfg->bus, cfg->slot, cfg->func);
975 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
976 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
980 /* REG and WREG use carry through to next functions */
984 * PCI Vital Product Data
987 #define PCI_VPD_TIMEOUT 1000000
990 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
992 int count = PCI_VPD_TIMEOUT;
994 KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
996 WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
998 while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1001 DELAY(1); /* limit looping */
1003 *data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1010 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1012 int count = PCI_VPD_TIMEOUT;
1014 KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1016 WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1017 WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1018 while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1021 DELAY(1); /* limit looping */
1028 #undef PCI_VPD_TIMEOUT
1030 struct vpd_readstate {
1040 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1045 if (vrs->bytesinval == 0) {
1046 if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, ®))
1048 vrs->val = le32toh(reg);
1050 byte = vrs->val & 0xff;
1051 vrs->bytesinval = 3;
1053 vrs->val = vrs->val >> 8;
1054 byte = vrs->val & 0xff;
1064 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1066 struct vpd_readstate vrs;
1071 int alloc, off; /* alloc/off for RO/W arrays */
1077 /* init vpd reader */
1085 name = remain = i = 0; /* shut up stupid gcc */
1086 alloc = off = 0; /* shut up stupid gcc */
1087 dflen = 0; /* shut up stupid gcc */
1089 while (state >= 0) {
1090 if (vpd_nextbyte(&vrs, &byte)) {
1095 printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1096 "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1097 vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1100 case 0: /* item name */
1102 if (vpd_nextbyte(&vrs, &byte2)) {
1107 if (vpd_nextbyte(&vrs, &byte2)) {
1111 remain |= byte2 << 8;
1112 if (remain > (0x7f*4 - vrs.off)) {
1115 "invalid VPD data, remain %#x\n",
1120 remain = byte & 0x7;
1121 name = (byte >> 3) & 0xf;
1124 case 0x2: /* String */
1125 cfg->vpd.vpd_ident = malloc(remain + 1,
1126 M_DEVBUF, M_WAITOK);
1133 case 0x10: /* VPD-R */
1136 cfg->vpd.vpd_ros = malloc(alloc *
1137 sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1141 case 0x11: /* VPD-W */
1144 cfg->vpd.vpd_w = malloc(alloc *
1145 sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1149 default: /* Invalid data, abort */
1155 case 1: /* Identifier String */
1156 cfg->vpd.vpd_ident[i++] = byte;
1159 cfg->vpd.vpd_ident[i] = '\0';
1164 case 2: /* VPD-R Keyword Header */
1166 cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1167 (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1168 M_DEVBUF, M_WAITOK | M_ZERO);
1170 cfg->vpd.vpd_ros[off].keyword[0] = byte;
1171 if (vpd_nextbyte(&vrs, &byte2)) {
1175 cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1176 if (vpd_nextbyte(&vrs, &byte2)) {
1180 cfg->vpd.vpd_ros[off].len = dflen = byte2;
1182 strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1185 * if this happens, we can't trust the rest
1188 pci_printf(cfg, "bad keyword length: %d\n",
1193 } else if (dflen == 0) {
1194 cfg->vpd.vpd_ros[off].value = malloc(1 *
1195 sizeof(*cfg->vpd.vpd_ros[off].value),
1196 M_DEVBUF, M_WAITOK);
1197 cfg->vpd.vpd_ros[off].value[0] = '\x00';
1199 cfg->vpd.vpd_ros[off].value = malloc(
1201 sizeof(*cfg->vpd.vpd_ros[off].value),
1202 M_DEVBUF, M_WAITOK);
1205 /* keep in sync w/ state 3's transistions */
1206 if (dflen == 0 && remain == 0)
1208 else if (dflen == 0)
1214 case 3: /* VPD-R Keyword Value */
1215 cfg->vpd.vpd_ros[off].value[i++] = byte;
1216 if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1217 "RV", 2) == 0 && cksumvalid == -1) {
1223 "bad VPD cksum, remain %hhu\n",
1232 /* keep in sync w/ state 2's transistions */
1234 cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1235 if (dflen == 0 && remain == 0) {
1236 cfg->vpd.vpd_rocnt = off;
1237 cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1238 off * sizeof(*cfg->vpd.vpd_ros),
1239 M_DEVBUF, M_WAITOK | M_ZERO);
1241 } else if (dflen == 0)
1251 case 5: /* VPD-W Keyword Header */
1253 cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1254 (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1255 M_DEVBUF, M_WAITOK | M_ZERO);
1257 cfg->vpd.vpd_w[off].keyword[0] = byte;
1258 if (vpd_nextbyte(&vrs, &byte2)) {
1262 cfg->vpd.vpd_w[off].keyword[1] = byte2;
1263 if (vpd_nextbyte(&vrs, &byte2)) {
1267 cfg->vpd.vpd_w[off].len = dflen = byte2;
1268 cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1269 cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1270 sizeof(*cfg->vpd.vpd_w[off].value),
1271 M_DEVBUF, M_WAITOK);
1274 /* keep in sync w/ state 6's transistions */
1275 if (dflen == 0 && remain == 0)
1277 else if (dflen == 0)
1283 case 6: /* VPD-W Keyword Value */
1284 cfg->vpd.vpd_w[off].value[i++] = byte;
1287 /* keep in sync w/ state 5's transistions */
1289 cfg->vpd.vpd_w[off++].value[i++] = '\0';
1290 if (dflen == 0 && remain == 0) {
1291 cfg->vpd.vpd_wcnt = off;
1292 cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1293 off * sizeof(*cfg->vpd.vpd_w),
1294 M_DEVBUF, M_WAITOK | M_ZERO);
1296 } else if (dflen == 0)
1301 pci_printf(cfg, "invalid state: %d\n", state);
1307 if (cksumvalid == 0 || state < -1) {
1308 /* read-only data bad, clean up */
1309 if (cfg->vpd.vpd_ros != NULL) {
1310 for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1311 free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1312 free(cfg->vpd.vpd_ros, M_DEVBUF);
1313 cfg->vpd.vpd_ros = NULL;
1317 /* I/O error, clean up */
1318 pci_printf(cfg, "failed to read VPD data.\n");
1319 if (cfg->vpd.vpd_ident != NULL) {
1320 free(cfg->vpd.vpd_ident, M_DEVBUF);
1321 cfg->vpd.vpd_ident = NULL;
1323 if (cfg->vpd.vpd_w != NULL) {
1324 for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1325 free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1326 free(cfg->vpd.vpd_w, M_DEVBUF);
1327 cfg->vpd.vpd_w = NULL;
1330 cfg->vpd.vpd_cached = 1;
1336 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1338 struct pci_devinfo *dinfo = device_get_ivars(child);
1339 pcicfgregs *cfg = &dinfo->cfg;
1341 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1342 pci_read_vpd(device_get_parent(dev), cfg);
1344 *identptr = cfg->vpd.vpd_ident;
1346 if (*identptr == NULL)
1353 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1356 struct pci_devinfo *dinfo = device_get_ivars(child);
1357 pcicfgregs *cfg = &dinfo->cfg;
1360 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1361 pci_read_vpd(device_get_parent(dev), cfg);
1363 for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1364 if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1365 sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1366 *vptr = cfg->vpd.vpd_ros[i].value;
1375 pci_fetch_vpd_list(device_t dev)
1377 struct pci_devinfo *dinfo = device_get_ivars(dev);
1378 pcicfgregs *cfg = &dinfo->cfg;
1380 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1381 pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1386 * Find the requested HyperTransport capability and return the offset
1387 * in configuration space via the pointer provided. The function
1388 * returns 0 on success and an error code otherwise.
1391 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1396 error = pci_find_cap(child, PCIY_HT, &ptr);
1401 * Traverse the capabilities list checking each HT capability
1402 * to see if it matches the requested HT capability.
1405 val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1406 if (capability == PCIM_HTCAP_SLAVE ||
1407 capability == PCIM_HTCAP_HOST)
1410 val &= PCIM_HTCMD_CAP_MASK;
1411 if (val == capability) {
1417 /* Skip to the next HT capability. */
1418 if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1426 * Find the next requested HyperTransport capability after start and return
1427 * the offset in configuration space via the pointer provided. The function
1428 * returns 0 on success and an error code otherwise.
1431 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1432 int start, int *capreg)
1437 KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1438 ("start capability is not HyperTransport capability"));
1442 * Traverse the capabilities list checking each HT capability
1443 * to see if it matches the requested HT capability.
1446 /* Skip to the next HT capability. */
1447 if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1450 val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1451 if (capability == PCIM_HTCAP_SLAVE ||
1452 capability == PCIM_HTCAP_HOST)
1455 val &= PCIM_HTCMD_CAP_MASK;
1456 if (val == capability) {
1467 * Find the requested capability and return the offset in
1468 * configuration space via the pointer provided. The function returns
1469 * 0 on success and an error code otherwise.
1472 pci_find_cap_method(device_t dev, device_t child, int capability,
1475 struct pci_devinfo *dinfo = device_get_ivars(child);
1476 pcicfgregs *cfg = &dinfo->cfg;
1481 * Check the CAP_LIST bit of the PCI status register first.
1483 status = pci_read_config(child, PCIR_STATUS, 2);
1484 if (!(status & PCIM_STATUS_CAPPRESENT))
1488 * Determine the start pointer of the capabilities list.
1490 switch (cfg->hdrtype & PCIM_HDRTYPE) {
1491 case PCIM_HDRTYPE_NORMAL:
1492 case PCIM_HDRTYPE_BRIDGE:
1495 case PCIM_HDRTYPE_CARDBUS:
1496 ptr = PCIR_CAP_PTR_2;
1500 return (ENXIO); /* no extended capabilities support */
1502 ptr = pci_read_config(child, ptr, 1);
1505 * Traverse the capabilities list.
1508 if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1513 ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1520 * Find the next requested capability after start and return the offset in
1521 * configuration space via the pointer provided. The function returns
1522 * 0 on success and an error code otherwise.
1525 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1526 int start, int *capreg)
1530 KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1531 ("start capability is not expected capability"));
1533 ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1535 if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1540 ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1547 * Find the requested extended capability and return the offset in
1548 * configuration space via the pointer provided. The function returns
1549 * 0 on success and an error code otherwise.
1552 pci_find_extcap_method(device_t dev, device_t child, int capability,
1555 struct pci_devinfo *dinfo = device_get_ivars(child);
1556 pcicfgregs *cfg = &dinfo->cfg;
1560 /* Only supported for PCI-express devices. */
1561 if (cfg->pcie.pcie_location == 0)
1565 ecap = pci_read_config(child, ptr, 4);
1566 if (ecap == 0xffffffff || ecap == 0)
1569 if (PCI_EXTCAP_ID(ecap) == capability) {
1574 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1577 ecap = pci_read_config(child, ptr, 4);
1584 * Find the next requested extended capability after start and return the
1585 * offset in configuration space via the pointer provided. The function
1586 * returns 0 on success and an error code otherwise.
1589 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1590 int start, int *capreg)
1592 struct pci_devinfo *dinfo = device_get_ivars(child);
1593 pcicfgregs *cfg = &dinfo->cfg;
1597 /* Only supported for PCI-express devices. */
1598 if (cfg->pcie.pcie_location == 0)
1601 ecap = pci_read_config(child, start, 4);
1602 KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1603 ("start extended capability is not expected capability"));
1604 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1606 ecap = pci_read_config(child, ptr, 4);
1607 if (PCI_EXTCAP_ID(ecap) == capability) {
1612 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1619 * Support for MSI-X message interrupts.
1622 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1624 struct pci_devinfo *dinfo = device_get_ivars(dev);
1625 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1628 KASSERT(msix->msix_table_len > index, ("bogus index"));
1629 offset = msix->msix_table_offset + index * 16;
1630 bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1631 bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1632 bus_write_4(msix->msix_table_res, offset + 8, data);
1636 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1637 uint64_t address, uint32_t data)
1640 if (pci_msix_rewrite_table) {
1641 struct pci_devinfo *dinfo = device_get_ivars(child);
1642 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1645 * Some VM hosts require MSIX to be disabled in the
1646 * control register before updating the MSIX table
1647 * entries are allowed. It is not enough to only
1648 * disable MSIX while updating a single entry. MSIX
1649 * must be disabled while updating all entries in the
1652 pci_write_config(child,
1653 msix->msix_location + PCIR_MSIX_CTRL,
1654 msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1655 pci_resume_msix(child);
1657 pci_write_msix_entry(child, index, address, data);
1659 /* Enable MSI -> HT mapping. */
1660 pci_ht_map_msi(child, address);
1664 pci_mask_msix(device_t dev, u_int index)
1666 struct pci_devinfo *dinfo = device_get_ivars(dev);
1667 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1668 uint32_t offset, val;
1670 KASSERT(msix->msix_msgnum > index, ("bogus index"));
1671 offset = msix->msix_table_offset + index * 16 + 12;
1672 val = bus_read_4(msix->msix_table_res, offset);
1673 if (!(val & PCIM_MSIX_VCTRL_MASK)) {
1674 val |= PCIM_MSIX_VCTRL_MASK;
1675 bus_write_4(msix->msix_table_res, offset, val);
1680 pci_unmask_msix(device_t dev, u_int index)
1682 struct pci_devinfo *dinfo = device_get_ivars(dev);
1683 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1684 uint32_t offset, val;
1686 KASSERT(msix->msix_table_len > index, ("bogus index"));
1687 offset = msix->msix_table_offset + index * 16 + 12;
1688 val = bus_read_4(msix->msix_table_res, offset);
1689 if (val & PCIM_MSIX_VCTRL_MASK) {
1690 val &= ~PCIM_MSIX_VCTRL_MASK;
1691 bus_write_4(msix->msix_table_res, offset, val);
1696 pci_pending_msix(device_t dev, u_int index)
1698 struct pci_devinfo *dinfo = device_get_ivars(dev);
1699 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1700 uint32_t offset, bit;
1702 KASSERT(msix->msix_table_len > index, ("bogus index"));
1703 offset = msix->msix_pba_offset + (index / 32) * 4;
1704 bit = 1 << index % 32;
1705 return (bus_read_4(msix->msix_pba_res, offset) & bit);
1709 * Restore MSI-X registers and table during resume. If MSI-X is
1710 * enabled then walk the virtual table to restore the actual MSI-X
1714 pci_resume_msix(device_t dev)
1716 struct pci_devinfo *dinfo = device_get_ivars(dev);
1717 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1718 struct msix_table_entry *mte;
1719 struct msix_vector *mv;
1722 if (msix->msix_alloc > 0) {
1723 /* First, mask all vectors. */
1724 for (i = 0; i < msix->msix_msgnum; i++)
1725 pci_mask_msix(dev, i);
1727 /* Second, program any messages with at least one handler. */
1728 for (i = 0; i < msix->msix_table_len; i++) {
1729 mte = &msix->msix_table[i];
1730 if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1732 mv = &msix->msix_vectors[mte->mte_vector - 1];
1733 pci_write_msix_entry(dev, i, mv->mv_address,
1735 pci_unmask_msix(dev, i);
1738 pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1739 msix->msix_ctrl, 2);
1743 * Attempt to allocate *count MSI-X messages. The actual number allocated is
1744 * returned in *count. After this function returns, each message will be
1745 * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1748 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1750 struct pci_devinfo *dinfo = device_get_ivars(child);
1751 pcicfgregs *cfg = &dinfo->cfg;
1752 struct resource_list_entry *rle;
1753 int actual, error, i, irq, max;
1755 /* Don't let count == 0 get us into trouble. */
1759 /* If rid 0 is allocated, then fail. */
1760 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1761 if (rle != NULL && rle->res != NULL)
1764 /* Already have allocated messages? */
1765 if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1768 /* If MSI-X is blacklisted for this system, fail. */
1769 if (pci_msix_blacklisted())
1772 /* MSI-X capability present? */
1773 if (cfg->msix.msix_location == 0 || !pci_do_msix)
1776 /* Make sure the appropriate BARs are mapped. */
1777 rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1778 cfg->msix.msix_table_bar);
1779 if (rle == NULL || rle->res == NULL ||
1780 !(rman_get_flags(rle->res) & RF_ACTIVE))
1782 cfg->msix.msix_table_res = rle->res;
1783 if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1784 rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1785 cfg->msix.msix_pba_bar);
1786 if (rle == NULL || rle->res == NULL ||
1787 !(rman_get_flags(rle->res) & RF_ACTIVE))
1790 cfg->msix.msix_pba_res = rle->res;
1793 device_printf(child,
1794 "attempting to allocate %d MSI-X vectors (%d supported)\n",
1795 *count, cfg->msix.msix_msgnum);
1796 max = min(*count, cfg->msix.msix_msgnum);
1797 for (i = 0; i < max; i++) {
1798 /* Allocate a message. */
1799 error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1805 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1811 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1813 device_printf(child, "using IRQ %ju for MSI-X\n",
1819 * Be fancy and try to print contiguous runs of
1820 * IRQ values as ranges. 'irq' is the previous IRQ.
1821 * 'run' is true if we are in a range.
1823 device_printf(child, "using IRQs %ju", rle->start);
1826 for (i = 1; i < actual; i++) {
1827 rle = resource_list_find(&dinfo->resources,
1828 SYS_RES_IRQ, i + 1);
1830 /* Still in a run? */
1831 if (rle->start == irq + 1) {
1837 /* Finish previous range. */
1843 /* Start new range. */
1844 printf(",%ju", rle->start);
1848 /* Unfinished range? */
1851 printf(" for MSI-X\n");
1855 /* Mask all vectors. */
1856 for (i = 0; i < cfg->msix.msix_msgnum; i++)
1857 pci_mask_msix(child, i);
1859 /* Allocate and initialize vector data and virtual table. */
1860 cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1861 M_DEVBUF, M_WAITOK | M_ZERO);
1862 cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1863 M_DEVBUF, M_WAITOK | M_ZERO);
1864 for (i = 0; i < actual; i++) {
1865 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1866 cfg->msix.msix_vectors[i].mv_irq = rle->start;
1867 cfg->msix.msix_table[i].mte_vector = i + 1;
1870 /* Update control register to enable MSI-X. */
1871 cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1872 pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1873 cfg->msix.msix_ctrl, 2);
1875 /* Update counts of alloc'd messages. */
1876 cfg->msix.msix_alloc = actual;
1877 cfg->msix.msix_table_len = actual;
1883 * By default, pci_alloc_msix() will assign the allocated IRQ
1884 * resources consecutively to the first N messages in the MSI-X table.
1885 * However, device drivers may want to use different layouts if they
1886 * either receive fewer messages than they asked for, or they wish to
1887 * populate the MSI-X table sparsely. This method allows the driver
1888 * to specify what layout it wants. It must be called after a
1889 * successful pci_alloc_msix() but before any of the associated
1890 * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1892 * The 'vectors' array contains 'count' message vectors. The array
1893 * maps directly to the MSI-X table in that index 0 in the array
1894 * specifies the vector for the first message in the MSI-X table, etc.
1895 * The vector value in each array index can either be 0 to indicate
1896 * that no vector should be assigned to a message slot, or it can be a
1897 * number from 1 to N (where N is the count returned from a
1898 * succcessful call to pci_alloc_msix()) to indicate which message
1899 * vector (IRQ) to be used for the corresponding message.
1901 * On successful return, each message with a non-zero vector will have
1902 * an associated SYS_RES_IRQ whose rid is equal to the array index +
1903 * 1. Additionally, if any of the IRQs allocated via the previous
1904 * call to pci_alloc_msix() are not used in the mapping, those IRQs
1905 * will be freed back to the system automatically.
1907 * For example, suppose a driver has a MSI-X table with 6 messages and
1908 * asks for 6 messages, but pci_alloc_msix() only returns a count of
1909 * 3. Call the three vectors allocated by pci_alloc_msix() A, B, and
1910 * C. After the call to pci_alloc_msix(), the device will be setup to
1911 * have an MSI-X table of ABC--- (where - means no vector assigned).
1912 * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1913 * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1914 * be freed back to the system. This device will also have valid
1915 * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1917 * In any case, the SYS_RES_IRQ rid X will always map to the message
1918 * at MSI-X table index X - 1 and will only be valid if a vector is
1919 * assigned to that table entry.
1922 pci_remap_msix_method(device_t dev, device_t child, int count,
1923 const u_int *vectors)
1925 struct pci_devinfo *dinfo = device_get_ivars(child);
1926 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1927 struct resource_list_entry *rle;
1928 int i, irq, j, *used;
1931 * Have to have at least one message in the table but the
1932 * table can't be bigger than the actual MSI-X table in the
1935 if (count == 0 || count > msix->msix_msgnum)
1938 /* Sanity check the vectors. */
1939 for (i = 0; i < count; i++)
1940 if (vectors[i] > msix->msix_alloc)
1944 * Make sure there aren't any holes in the vectors to be used.
1945 * It's a big pain to support it, and it doesn't really make
1946 * sense anyway. Also, at least one vector must be used.
1948 used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1950 for (i = 0; i < count; i++)
1951 if (vectors[i] != 0)
1952 used[vectors[i] - 1] = 1;
1953 for (i = 0; i < msix->msix_alloc - 1; i++)
1954 if (used[i] == 0 && used[i + 1] == 1) {
1955 free(used, M_DEVBUF);
1959 free(used, M_DEVBUF);
1963 /* Make sure none of the resources are allocated. */
1964 for (i = 0; i < msix->msix_table_len; i++) {
1965 if (msix->msix_table[i].mte_vector == 0)
1967 if (msix->msix_table[i].mte_handlers > 0) {
1968 free(used, M_DEVBUF);
1971 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1972 KASSERT(rle != NULL, ("missing resource"));
1973 if (rle->res != NULL) {
1974 free(used, M_DEVBUF);
1979 /* Free the existing resource list entries. */
1980 for (i = 0; i < msix->msix_table_len; i++) {
1981 if (msix->msix_table[i].mte_vector == 0)
1983 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1987 * Build the new virtual table keeping track of which vectors are
1990 free(msix->msix_table, M_DEVBUF);
1991 msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1992 M_DEVBUF, M_WAITOK | M_ZERO);
1993 for (i = 0; i < count; i++)
1994 msix->msix_table[i].mte_vector = vectors[i];
1995 msix->msix_table_len = count;
1997 /* Free any unused IRQs and resize the vectors array if necessary. */
1998 j = msix->msix_alloc - 1;
2000 struct msix_vector *vec;
2002 while (used[j] == 0) {
2003 PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2004 msix->msix_vectors[j].mv_irq);
2007 vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2009 bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2011 free(msix->msix_vectors, M_DEVBUF);
2012 msix->msix_vectors = vec;
2013 msix->msix_alloc = j + 1;
2015 free(used, M_DEVBUF);
2017 /* Map the IRQs onto the rids. */
2018 for (i = 0; i < count; i++) {
2019 if (vectors[i] == 0)
2021 irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2022 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2027 device_printf(child, "Remapped MSI-X IRQs as: ");
2028 for (i = 0; i < count; i++) {
2031 if (vectors[i] == 0)
2035 msix->msix_vectors[vectors[i] - 1].mv_irq);
2044 pci_release_msix(device_t dev, device_t child)
2046 struct pci_devinfo *dinfo = device_get_ivars(child);
2047 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2048 struct resource_list_entry *rle;
2051 /* Do we have any messages to release? */
2052 if (msix->msix_alloc == 0)
2055 /* Make sure none of the resources are allocated. */
2056 for (i = 0; i < msix->msix_table_len; i++) {
2057 if (msix->msix_table[i].mte_vector == 0)
2059 if (msix->msix_table[i].mte_handlers > 0)
2061 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2062 KASSERT(rle != NULL, ("missing resource"));
2063 if (rle->res != NULL)
2067 /* Update control register to disable MSI-X. */
2068 msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2069 pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2070 msix->msix_ctrl, 2);
2072 /* Free the resource list entries. */
2073 for (i = 0; i < msix->msix_table_len; i++) {
2074 if (msix->msix_table[i].mte_vector == 0)
2076 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2078 free(msix->msix_table, M_DEVBUF);
2079 msix->msix_table_len = 0;
2081 /* Release the IRQs. */
2082 for (i = 0; i < msix->msix_alloc; i++)
2083 PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2084 msix->msix_vectors[i].mv_irq);
2085 free(msix->msix_vectors, M_DEVBUF);
2086 msix->msix_alloc = 0;
2091 * Return the max supported MSI-X messages this device supports.
2092 * Basically, assuming the MD code can alloc messages, this function
2093 * should return the maximum value that pci_alloc_msix() can return.
2094 * Thus, it is subject to the tunables, etc.
2097 pci_msix_count_method(device_t dev, device_t child)
2099 struct pci_devinfo *dinfo = device_get_ivars(child);
2100 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2102 if (pci_do_msix && msix->msix_location != 0)
2103 return (msix->msix_msgnum);
2108 pci_msix_pba_bar_method(device_t dev, device_t child)
2110 struct pci_devinfo *dinfo = device_get_ivars(child);
2111 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2113 if (pci_do_msix && msix->msix_location != 0)
2114 return (msix->msix_pba_bar);
2119 pci_msix_table_bar_method(device_t dev, device_t child)
2121 struct pci_devinfo *dinfo = device_get_ivars(child);
2122 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2124 if (pci_do_msix && msix->msix_location != 0)
2125 return (msix->msix_table_bar);
2130 * HyperTransport MSI mapping control
2133 pci_ht_map_msi(device_t dev, uint64_t addr)
2135 struct pci_devinfo *dinfo = device_get_ivars(dev);
2136 struct pcicfg_ht *ht = &dinfo->cfg.ht;
2141 if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2142 ht->ht_msiaddr >> 20 == addr >> 20) {
2143 /* Enable MSI -> HT mapping. */
2144 ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2145 pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2149 if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2150 /* Disable MSI -> HT mapping. */
2151 ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2152 pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2158 pci_get_max_payload(device_t dev)
2160 struct pci_devinfo *dinfo = device_get_ivars(dev);
2164 cap = dinfo->cfg.pcie.pcie_location;
2167 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2168 val &= PCIEM_CTL_MAX_PAYLOAD;
2170 return (1 << (val + 7));
2174 pci_get_max_read_req(device_t dev)
2176 struct pci_devinfo *dinfo = device_get_ivars(dev);
2180 cap = dinfo->cfg.pcie.pcie_location;
2183 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2184 val &= PCIEM_CTL_MAX_READ_REQUEST;
2186 return (1 << (val + 7));
2190 pci_set_max_read_req(device_t dev, int size)
2192 struct pci_devinfo *dinfo = device_get_ivars(dev);
2196 cap = dinfo->cfg.pcie.pcie_location;
2203 size = (1 << (fls(size) - 1));
2204 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2205 val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2206 val |= (fls(size) - 8) << 12;
2207 pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2212 pcie_read_config(device_t dev, int reg, int width)
2214 struct pci_devinfo *dinfo = device_get_ivars(dev);
2217 cap = dinfo->cfg.pcie.pcie_location;
2221 return (0xffffffff);
2224 return (pci_read_config(dev, cap + reg, width));
2228 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2230 struct pci_devinfo *dinfo = device_get_ivars(dev);
2233 cap = dinfo->cfg.pcie.pcie_location;
2236 pci_write_config(dev, cap + reg, value, width);
2240 * Adjusts a PCI-e capability register by clearing the bits in mask
2241 * and setting the bits in (value & mask). Bits not set in mask are
2244 * Returns the old value on success or all ones on failure.
2247 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2250 struct pci_devinfo *dinfo = device_get_ivars(dev);
2254 cap = dinfo->cfg.pcie.pcie_location;
2258 return (0xffffffff);
2261 old = pci_read_config(dev, cap + reg, width);
2263 new |= (value & mask);
2264 pci_write_config(dev, cap + reg, new, width);
2269 * Support for MSI message signalled interrupts.
2272 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2275 struct pci_devinfo *dinfo = device_get_ivars(child);
2276 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2278 /* Write data and address values. */
2279 pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2280 address & 0xffffffff, 4);
2281 if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2282 pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2284 pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2287 pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2290 /* Enable MSI in the control register. */
2291 msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2292 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2295 /* Enable MSI -> HT mapping. */
2296 pci_ht_map_msi(child, address);
2300 pci_disable_msi_method(device_t dev, device_t child)
2302 struct pci_devinfo *dinfo = device_get_ivars(child);
2303 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2305 /* Disable MSI -> HT mapping. */
2306 pci_ht_map_msi(child, 0);
2308 /* Disable MSI in the control register. */
2309 msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2310 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2315 * Restore MSI registers during resume. If MSI is enabled then
2316 * restore the data and address registers in addition to the control
2320 pci_resume_msi(device_t dev)
2322 struct pci_devinfo *dinfo = device_get_ivars(dev);
2323 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2327 if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2328 address = msi->msi_addr;
2329 data = msi->msi_data;
2330 pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2331 address & 0xffffffff, 4);
2332 if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2333 pci_write_config(dev, msi->msi_location +
2334 PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2335 pci_write_config(dev, msi->msi_location +
2336 PCIR_MSI_DATA_64BIT, data, 2);
2338 pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2341 pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2346 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2348 struct pci_devinfo *dinfo = device_get_ivars(dev);
2349 pcicfgregs *cfg = &dinfo->cfg;
2350 struct resource_list_entry *rle;
2351 struct msix_table_entry *mte;
2352 struct msix_vector *mv;
2358 * Handle MSI first. We try to find this IRQ among our list
2359 * of MSI IRQs. If we find it, we request updated address and
2360 * data registers and apply the results.
2362 if (cfg->msi.msi_alloc > 0) {
2364 /* If we don't have any active handlers, nothing to do. */
2365 if (cfg->msi.msi_handlers == 0)
2367 for (i = 0; i < cfg->msi.msi_alloc; i++) {
2368 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2370 if (rle->start == irq) {
2371 error = PCIB_MAP_MSI(device_get_parent(bus),
2372 dev, irq, &addr, &data);
2375 pci_disable_msi(dev);
2376 dinfo->cfg.msi.msi_addr = addr;
2377 dinfo->cfg.msi.msi_data = data;
2378 pci_enable_msi(dev, addr, data);
2386 * For MSI-X, we check to see if we have this IRQ. If we do,
2387 * we request the updated mapping info. If that works, we go
2388 * through all the slots that use this IRQ and update them.
2390 if (cfg->msix.msix_alloc > 0) {
2391 for (i = 0; i < cfg->msix.msix_alloc; i++) {
2392 mv = &cfg->msix.msix_vectors[i];
2393 if (mv->mv_irq == irq) {
2394 error = PCIB_MAP_MSI(device_get_parent(bus),
2395 dev, irq, &addr, &data);
2398 mv->mv_address = addr;
2400 for (j = 0; j < cfg->msix.msix_table_len; j++) {
2401 mte = &cfg->msix.msix_table[j];
2402 if (mte->mte_vector != i + 1)
2404 if (mte->mte_handlers == 0)
2406 pci_mask_msix(dev, j);
2407 pci_enable_msix(dev, j, addr, data);
2408 pci_unmask_msix(dev, j);
2419 * Returns true if the specified device is blacklisted because MSI
2423 pci_msi_device_blacklisted(device_t dev)
2426 if (!pci_honor_msi_blacklist)
2429 return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2433 * Determine if MSI is blacklisted globally on this system. Currently,
2434 * we just check for blacklisted chipsets as represented by the
2435 * host-PCI bridge at device 0:0:0. In the future, it may become
2436 * necessary to check other system attributes, such as the kenv values
2437 * that give the motherboard manufacturer and model number.
2440 pci_msi_blacklisted(void)
2444 if (!pci_honor_msi_blacklist)
2447 /* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2448 if (!(pcie_chipset || pcix_chipset)) {
2449 if (vm_guest != VM_GUEST_NO) {
2451 * Whitelist older chipsets in virtual
2452 * machines known to support MSI.
2454 dev = pci_find_bsf(0, 0, 0);
2456 return (!pci_has_quirk(pci_get_devid(dev),
2457 PCI_QUIRK_ENABLE_MSI_VM));
2462 dev = pci_find_bsf(0, 0, 0);
2464 return (pci_msi_device_blacklisted(dev));
2469 * Returns true if the specified device is blacklisted because MSI-X
2470 * doesn't work. Note that this assumes that if MSI doesn't work,
2471 * MSI-X doesn't either.
2474 pci_msix_device_blacklisted(device_t dev)
2477 if (!pci_honor_msi_blacklist)
2480 if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2483 return (pci_msi_device_blacklisted(dev));
2487 * Determine if MSI-X is blacklisted globally on this system. If MSI
2488 * is blacklisted, assume that MSI-X is as well. Check for additional
2489 * chipsets where MSI works but MSI-X does not.
2492 pci_msix_blacklisted(void)
2496 if (!pci_honor_msi_blacklist)
2499 dev = pci_find_bsf(0, 0, 0);
2500 if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2501 PCI_QUIRK_DISABLE_MSIX))
2504 return (pci_msi_blacklisted());
2508 * Attempt to allocate *count MSI messages. The actual number allocated is
2509 * returned in *count. After this function returns, each message will be
2510 * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2513 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2515 struct pci_devinfo *dinfo = device_get_ivars(child);
2516 pcicfgregs *cfg = &dinfo->cfg;
2517 struct resource_list_entry *rle;
2518 int actual, error, i, irqs[32];
2521 /* Don't let count == 0 get us into trouble. */
2525 /* If rid 0 is allocated, then fail. */
2526 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2527 if (rle != NULL && rle->res != NULL)
2530 /* Already have allocated messages? */
2531 if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2534 /* If MSI is blacklisted for this system, fail. */
2535 if (pci_msi_blacklisted())
2538 /* MSI capability present? */
2539 if (cfg->msi.msi_location == 0 || !pci_do_msi)
2543 device_printf(child,
2544 "attempting to allocate %d MSI vectors (%d supported)\n",
2545 *count, cfg->msi.msi_msgnum);
2547 /* Don't ask for more than the device supports. */
2548 actual = min(*count, cfg->msi.msi_msgnum);
2550 /* Don't ask for more than 32 messages. */
2551 actual = min(actual, 32);
2553 /* MSI requires power of 2 number of messages. */
2554 if (!powerof2(actual))
2558 /* Try to allocate N messages. */
2559 error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2571 * We now have N actual messages mapped onto SYS_RES_IRQ
2572 * resources in the irqs[] array, so add new resources
2573 * starting at rid 1.
2575 for (i = 0; i < actual; i++)
2576 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2577 irqs[i], irqs[i], 1);
2581 device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2586 * Be fancy and try to print contiguous runs
2587 * of IRQ values as ranges. 'run' is true if
2588 * we are in a range.
2590 device_printf(child, "using IRQs %d", irqs[0]);
2592 for (i = 1; i < actual; i++) {
2594 /* Still in a run? */
2595 if (irqs[i] == irqs[i - 1] + 1) {
2600 /* Finish previous range. */
2602 printf("-%d", irqs[i - 1]);
2606 /* Start new range. */
2607 printf(",%d", irqs[i]);
2610 /* Unfinished range? */
2612 printf("-%d", irqs[actual - 1]);
2613 printf(" for MSI\n");
2617 /* Update control register with actual count. */
2618 ctrl = cfg->msi.msi_ctrl;
2619 ctrl &= ~PCIM_MSICTRL_MME_MASK;
2620 ctrl |= (ffs(actual) - 1) << 4;
2621 cfg->msi.msi_ctrl = ctrl;
2622 pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2624 /* Update counts of alloc'd messages. */
2625 cfg->msi.msi_alloc = actual;
2626 cfg->msi.msi_handlers = 0;
2631 /* Release the MSI messages associated with this device. */
2633 pci_release_msi_method(device_t dev, device_t child)
2635 struct pci_devinfo *dinfo = device_get_ivars(child);
2636 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2637 struct resource_list_entry *rle;
2638 int error, i, irqs[32];
2640 /* Try MSI-X first. */
2641 error = pci_release_msix(dev, child);
2642 if (error != ENODEV)
2645 /* Do we have any messages to release? */
2646 if (msi->msi_alloc == 0)
2648 KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2650 /* Make sure none of the resources are allocated. */
2651 if (msi->msi_handlers > 0)
2653 for (i = 0; i < msi->msi_alloc; i++) {
2654 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2655 KASSERT(rle != NULL, ("missing MSI resource"));
2656 if (rle->res != NULL)
2658 irqs[i] = rle->start;
2661 /* Update control register with 0 count. */
2662 KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2663 ("%s: MSI still enabled", __func__));
2664 msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2665 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2668 /* Release the messages. */
2669 PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2670 for (i = 0; i < msi->msi_alloc; i++)
2671 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2673 /* Update alloc count. */
2681 * Return the max supported MSI messages this device supports.
2682 * Basically, assuming the MD code can alloc messages, this function
2683 * should return the maximum value that pci_alloc_msi() can return.
2684 * Thus, it is subject to the tunables, etc.
2687 pci_msi_count_method(device_t dev, device_t child)
2689 struct pci_devinfo *dinfo = device_get_ivars(child);
2690 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2692 if (pci_do_msi && msi->msi_location != 0)
2693 return (msi->msi_msgnum);
2697 /* free pcicfgregs structure and all depending data structures */
2700 pci_freecfg(struct pci_devinfo *dinfo)
2702 struct devlist *devlist_head;
2703 struct pci_map *pm, *next;
2706 devlist_head = &pci_devq;
2708 if (dinfo->cfg.vpd.vpd_reg) {
2709 free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2710 for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2711 free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2712 free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2713 for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2714 free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2715 free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2717 STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2720 STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2721 free(dinfo, M_DEVBUF);
2723 /* increment the generation count */
2726 /* we're losing one device */
2732 * PCI power manangement
2735 pci_set_powerstate_method(device_t dev, device_t child, int state)
2737 struct pci_devinfo *dinfo = device_get_ivars(child);
2738 pcicfgregs *cfg = &dinfo->cfg;
2740 int oldstate, highest, delay;
2742 if (cfg->pp.pp_cap == 0)
2743 return (EOPNOTSUPP);
2746 * Optimize a no state change request away. While it would be OK to
2747 * write to the hardware in theory, some devices have shown odd
2748 * behavior when going from D3 -> D3.
2750 oldstate = pci_get_powerstate(child);
2751 if (oldstate == state)
2755 * The PCI power management specification states that after a state
2756 * transition between PCI power states, system software must
2757 * guarantee a minimal delay before the function accesses the device.
2758 * Compute the worst case delay that we need to guarantee before we
2759 * access the device. Many devices will be responsive much more
2760 * quickly than this delay, but there are some that don't respond
2761 * instantly to state changes. Transitions to/from D3 state require
2762 * 10ms, while D2 requires 200us, and D0/1 require none. The delay
2763 * is done below with DELAY rather than a sleeper function because
2764 * this function can be called from contexts where we cannot sleep.
2766 highest = (oldstate > state) ? oldstate : state;
2767 if (highest == PCI_POWERSTATE_D3)
2769 else if (highest == PCI_POWERSTATE_D2)
2773 status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2774 & ~PCIM_PSTAT_DMASK;
2776 case PCI_POWERSTATE_D0:
2777 status |= PCIM_PSTAT_D0;
2779 case PCI_POWERSTATE_D1:
2780 if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2781 return (EOPNOTSUPP);
2782 status |= PCIM_PSTAT_D1;
2784 case PCI_POWERSTATE_D2:
2785 if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2786 return (EOPNOTSUPP);
2787 status |= PCIM_PSTAT_D2;
2789 case PCI_POWERSTATE_D3:
2790 status |= PCIM_PSTAT_D3;
2797 pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2800 PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2807 pci_get_powerstate_method(device_t dev, device_t child)
2809 struct pci_devinfo *dinfo = device_get_ivars(child);
2810 pcicfgregs *cfg = &dinfo->cfg;
2814 if (cfg->pp.pp_cap != 0) {
2815 status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2816 switch (status & PCIM_PSTAT_DMASK) {
2818 result = PCI_POWERSTATE_D0;
2821 result = PCI_POWERSTATE_D1;
2824 result = PCI_POWERSTATE_D2;
2827 result = PCI_POWERSTATE_D3;
2830 result = PCI_POWERSTATE_UNKNOWN;
2834 /* No support, device is always at D0 */
2835 result = PCI_POWERSTATE_D0;
2841 * Some convenience functions for PCI device drivers.
2844 static __inline void
2845 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2849 command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2851 PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2854 static __inline void
2855 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2859 command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2861 PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2865 pci_enable_busmaster_method(device_t dev, device_t child)
2867 pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2872 pci_disable_busmaster_method(device_t dev, device_t child)
2874 pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2879 pci_enable_io_method(device_t dev, device_t child, int space)
2884 case SYS_RES_IOPORT:
2885 bit = PCIM_CMD_PORTEN;
2887 case SYS_RES_MEMORY:
2888 bit = PCIM_CMD_MEMEN;
2893 pci_set_command_bit(dev, child, bit);
2898 pci_disable_io_method(device_t dev, device_t child, int space)
2903 case SYS_RES_IOPORT:
2904 bit = PCIM_CMD_PORTEN;
2906 case SYS_RES_MEMORY:
2907 bit = PCIM_CMD_MEMEN;
2912 pci_clear_command_bit(dev, child, bit);
2917 * New style pci driver. Parent device is either a pci-host-bridge or a
2918 * pci-pci-bridge. Both kinds are represented by instances of pcib.
2922 pci_print_verbose(struct pci_devinfo *dinfo)
2926 pcicfgregs *cfg = &dinfo->cfg;
2928 printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2929 cfg->vendor, cfg->device, cfg->revid);
2930 printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2931 cfg->domain, cfg->bus, cfg->slot, cfg->func);
2932 printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2933 cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2935 printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2936 cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2937 printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2938 cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2939 cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2940 if (cfg->intpin > 0)
2941 printf("\tintpin=%c, irq=%d\n",
2942 cfg->intpin +'a' -1, cfg->intline);
2943 if (cfg->pp.pp_cap) {
2946 status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2947 printf("\tpowerspec %d supports D0%s%s D3 current D%d\n",
2948 cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2949 cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2950 cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2951 status & PCIM_PSTAT_DMASK);
2953 if (cfg->msi.msi_location) {
2956 ctrl = cfg->msi.msi_ctrl;
2957 printf("\tMSI supports %d message%s%s%s\n",
2958 cfg->msi.msi_msgnum,
2959 (cfg->msi.msi_msgnum == 1) ? "" : "s",
2960 (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2961 (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2963 if (cfg->msix.msix_location) {
2964 printf("\tMSI-X supports %d message%s ",
2965 cfg->msix.msix_msgnum,
2966 (cfg->msix.msix_msgnum == 1) ? "" : "s");
2967 if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2968 printf("in map 0x%x\n",
2969 cfg->msix.msix_table_bar);
2971 printf("in maps 0x%x and 0x%x\n",
2972 cfg->msix.msix_table_bar,
2973 cfg->msix.msix_pba_bar);
2979 pci_porten(device_t dev)
2981 return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2985 pci_memen(device_t dev)
2987 return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2991 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
2994 struct pci_devinfo *dinfo;
2995 pci_addr_t map, testval;
3000 * The device ROM BAR is special. It is always a 32-bit
3001 * memory BAR. Bit 0 is special and should not be set when
3004 dinfo = device_get_ivars(dev);
3005 if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3006 map = pci_read_config(dev, reg, 4);
3007 pci_write_config(dev, reg, 0xfffffffe, 4);
3008 testval = pci_read_config(dev, reg, 4);
3009 pci_write_config(dev, reg, map, 4);
3011 *testvalp = testval;
3017 map = pci_read_config(dev, reg, 4);
3018 ln2range = pci_maprange(map);
3020 map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3023 * Disable decoding via the command register before
3024 * determining the BAR's length since we will be placing it in
3027 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3028 pci_write_config(dev, PCIR_COMMAND,
3029 cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3032 * Determine the BAR's length by writing all 1's. The bottom
3033 * log_2(size) bits of the BAR will stick as 0 when we read
3036 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3037 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3038 * and combines the result into a 64-bit value." (section 6.2.5.1)
3040 * Writes to both registers must be performed before attempting to
3041 * read back the size value.
3044 pci_write_config(dev, reg, 0xffffffff, 4);
3045 if (ln2range == 64) {
3046 pci_write_config(dev, reg + 4, 0xffffffff, 4);
3047 testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3049 testval |= pci_read_config(dev, reg, 4);
3052 * Restore the original value of the BAR. We may have reprogrammed
3053 * the BAR of the low-level console device and when booting verbose,
3054 * we need the console device addressable.
3056 pci_write_config(dev, reg, map, 4);
3058 pci_write_config(dev, reg + 4, map >> 32, 4);
3059 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3062 *testvalp = testval;
3064 *bar64 = (ln2range == 64);
3068 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3070 struct pci_devinfo *dinfo;
3073 /* The device ROM BAR is always a 32-bit memory BAR. */
3074 dinfo = device_get_ivars(dev);
3075 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3078 ln2range = pci_maprange(pm->pm_value);
3079 pci_write_config(dev, pm->pm_reg, base, 4);
3081 pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3082 pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3084 pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3085 pm->pm_reg + 4, 4) << 32;
3089 pci_find_bar(device_t dev, int reg)
3091 struct pci_devinfo *dinfo;
3094 dinfo = device_get_ivars(dev);
3095 STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3096 if (pm->pm_reg == reg)
3103 pci_bar_enabled(device_t dev, struct pci_map *pm)
3105 struct pci_devinfo *dinfo;
3108 dinfo = device_get_ivars(dev);
3109 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3110 !(pm->pm_value & PCIM_BIOS_ENABLE))
3112 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3113 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3114 return ((cmd & PCIM_CMD_MEMEN) != 0);
3116 return ((cmd & PCIM_CMD_PORTEN) != 0);
3120 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3122 struct pci_devinfo *dinfo;
3123 struct pci_map *pm, *prev;
3125 dinfo = device_get_ivars(dev);
3126 pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3128 pm->pm_value = value;
3130 STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3131 KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3133 if (STAILQ_NEXT(prev, pm_link) == NULL ||
3134 STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3138 STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3140 STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3145 pci_restore_bars(device_t dev)
3147 struct pci_devinfo *dinfo;
3151 dinfo = device_get_ivars(dev);
3152 STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3153 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3156 ln2range = pci_maprange(pm->pm_value);
3157 pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3159 pci_write_config(dev, pm->pm_reg + 4,
3160 pm->pm_value >> 32, 4);
3165 * Add a resource based on a pci map register. Return 1 if the map
3166 * register is a 32bit map register or 2 if it is a 64bit register.
3169 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3170 int force, int prefetch)
3173 pci_addr_t base, map, testval;
3174 pci_addr_t start, end, count;
3175 int barlen, basezero, flags, maprange, mapsize, type;
3177 struct resource *res;
3180 * The BAR may already exist if the device is a CardBus card
3181 * whose CIS is stored in this BAR.
3183 pm = pci_find_bar(dev, reg);
3185 maprange = pci_maprange(pm->pm_value);
3186 barlen = maprange == 64 ? 2 : 1;
3190 pci_read_bar(dev, reg, &map, &testval, NULL);
3191 if (PCI_BAR_MEM(map)) {
3192 type = SYS_RES_MEMORY;
3193 if (map & PCIM_BAR_MEM_PREFETCH)
3196 type = SYS_RES_IOPORT;
3197 mapsize = pci_mapsize(testval);
3198 base = pci_mapbase(map);
3199 #ifdef __PCI_BAR_ZERO_VALID
3202 basezero = base == 0;
3204 maprange = pci_maprange(map);
3205 barlen = maprange == 64 ? 2 : 1;
3208 * For I/O registers, if bottom bit is set, and the next bit up
3209 * isn't clear, we know we have a BAR that doesn't conform to the
3210 * spec, so ignore it. Also, sanity check the size of the data
3211 * areas to the type of memory involved. Memory must be at least
3212 * 16 bytes in size, while I/O ranges must be at least 4.
3214 if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3216 if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3217 (type == SYS_RES_IOPORT && mapsize < 2))
3220 /* Save a record of this BAR. */
3221 pm = pci_add_bar(dev, reg, map, mapsize);
3223 printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3224 reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3225 if (type == SYS_RES_IOPORT && !pci_porten(dev))
3226 printf(", port disabled\n");
3227 else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3228 printf(", memory disabled\n");
3230 printf(", enabled\n");
3234 * If base is 0, then we have problems if this architecture does
3235 * not allow that. It is best to ignore such entries for the
3236 * moment. These will be allocated later if the driver specifically
3237 * requests them. However, some removable buses look better when
3238 * all resources are allocated, so allow '0' to be overriden.
3240 * Similarly treat maps whose values is the same as the test value
3241 * read back. These maps have had all f's written to them by the
3242 * BIOS in an attempt to disable the resources.
3244 if (!force && (basezero || map == testval))
3246 if ((u_long)base != base) {
3248 "pci%d:%d:%d:%d bar %#x too many address bits",
3249 pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3250 pci_get_function(dev), reg);
3255 * This code theoretically does the right thing, but has
3256 * undesirable side effects in some cases where peripherals
3257 * respond oddly to having these bits enabled. Let the user
3258 * be able to turn them off (since pci_enable_io_modes is 1 by
3261 if (pci_enable_io_modes) {
3262 /* Turn on resources that have been left off by a lazy BIOS */
3263 if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3264 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3265 cmd |= PCIM_CMD_PORTEN;
3266 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3268 if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3269 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3270 cmd |= PCIM_CMD_MEMEN;
3271 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3274 if (type == SYS_RES_IOPORT && !pci_porten(dev))
3276 if (type == SYS_RES_MEMORY && !pci_memen(dev))
3280 count = (pci_addr_t)1 << mapsize;
3281 flags = RF_ALIGNMENT_LOG2(mapsize);
3283 flags |= RF_PREFETCHABLE;
3284 if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3285 start = 0; /* Let the parent decide. */
3289 end = base + count - 1;
3291 resource_list_add(rl, type, reg, start, end, count);
3294 * Try to allocate the resource for this BAR from our parent
3295 * so that this resource range is already reserved. The
3296 * driver for this device will later inherit this resource in
3297 * pci_alloc_resource().
3299 res = resource_list_reserve(rl, bus, dev, type, ®, start, end, count,
3301 if ((pci_do_realloc_bars
3302 || pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3303 && res == NULL && (start != 0 || end != ~0)) {
3305 * If the allocation fails, try to allocate a resource for
3306 * this BAR using any available range. The firmware felt
3307 * it was important enough to assign a resource, so don't
3308 * disable decoding if we can help it.
3310 resource_list_delete(rl, type, reg);
3311 resource_list_add(rl, type, reg, 0, ~0, count);
3312 res = resource_list_reserve(rl, bus, dev, type, ®, 0, ~0,
3317 * If the allocation fails, delete the resource list entry
3318 * and disable decoding for this device.
3320 * If the driver requests this resource in the future,
3321 * pci_reserve_map() will try to allocate a fresh
3324 resource_list_delete(rl, type, reg);
3325 pci_disable_io(dev, type);
3328 "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3329 pci_get_domain(dev), pci_get_bus(dev),
3330 pci_get_slot(dev), pci_get_function(dev), reg);
3332 start = rman_get_start(res);
3333 pci_write_bar(dev, pm, start);
3339 * For ATA devices we need to decide early what addressing mode to use.
3340 * Legacy demands that the primary and secondary ATA ports sits on the
3341 * same addresses that old ISA hardware did. This dictates that we use
3342 * those addresses and ignore the BAR's if we cannot set PCI native
3346 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3347 uint32_t prefetchmask)
3349 int rid, type, progif;
3351 /* if this device supports PCI native addressing use it */
3352 progif = pci_read_config(dev, PCIR_PROGIF, 1);
3353 if ((progif & 0x8a) == 0x8a) {
3354 if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3355 pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3356 printf("Trying ATA native PCI addressing mode\n");
3357 pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3361 progif = pci_read_config(dev, PCIR_PROGIF, 1);
3362 type = SYS_RES_IOPORT;
3363 if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3364 pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3365 prefetchmask & (1 << 0));
3366 pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3367 prefetchmask & (1 << 1));
3370 resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3371 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3374 resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3375 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3378 if (progif & PCIP_STORAGE_IDE_MODESEC) {
3379 pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3380 prefetchmask & (1 << 2));
3381 pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3382 prefetchmask & (1 << 3));
3385 resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3386 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3389 resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3390 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3393 pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3394 prefetchmask & (1 << 4));
3395 pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3396 prefetchmask & (1 << 5));
3400 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3402 struct pci_devinfo *dinfo = device_get_ivars(dev);
3403 pcicfgregs *cfg = &dinfo->cfg;
3404 char tunable_name[64];
3407 /* Has to have an intpin to have an interrupt. */
3408 if (cfg->intpin == 0)
3411 /* Let the user override the IRQ with a tunable. */
3412 irq = PCI_INVALID_IRQ;
3413 snprintf(tunable_name, sizeof(tunable_name),
3414 "hw.pci%d.%d.%d.INT%c.irq",
3415 cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3416 if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3417 irq = PCI_INVALID_IRQ;
3420 * If we didn't get an IRQ via the tunable, then we either use the
3421 * IRQ value in the intline register or we ask the bus to route an
3422 * interrupt for us. If force_route is true, then we only use the
3423 * value in the intline register if the bus was unable to assign an
3426 if (!PCI_INTERRUPT_VALID(irq)) {
3427 if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3428 irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3429 if (!PCI_INTERRUPT_VALID(irq))
3433 /* If after all that we don't have an IRQ, just bail. */
3434 if (!PCI_INTERRUPT_VALID(irq))
3437 /* Update the config register if it changed. */
3438 if (irq != cfg->intline) {
3440 pci_write_config(dev, PCIR_INTLINE, irq, 1);
3443 /* Add this IRQ as rid 0 interrupt resource. */
3444 resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3447 /* Perform early OHCI takeover from SMM. */
3449 ohci_early_takeover(device_t self)
3451 struct resource *res;
3457 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3461 ctl = bus_read_4(res, OHCI_CONTROL);
3462 if (ctl & OHCI_IR) {
3464 printf("ohci early: "
3465 "SMM active, request owner change\n");
3466 bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3467 for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3469 ctl = bus_read_4(res, OHCI_CONTROL);
3471 if (ctl & OHCI_IR) {
3473 printf("ohci early: "
3474 "SMM does not respond, resetting\n");
3475 bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3477 /* Disable interrupts */
3478 bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3481 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3484 /* Perform early UHCI takeover from SMM. */
3486 uhci_early_takeover(device_t self)
3488 struct resource *res;
3492 * Set the PIRQD enable bit and switch off all the others. We don't
3493 * want legacy support to interfere with us XXX Does this also mean
3494 * that the BIOS won't touch the keyboard anymore if it is connected
3495 * to the ports of the root hub?
3497 pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3499 /* Disable interrupts */
3500 rid = PCI_UHCI_BASE_REG;
3501 res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3503 bus_write_2(res, UHCI_INTR, 0);
3504 bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3508 /* Perform early EHCI takeover from SMM. */
3510 ehci_early_takeover(device_t self)
3512 struct resource *res;
3522 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3526 cparams = bus_read_4(res, EHCI_HCCPARAMS);
3528 /* Synchronise with the BIOS if it owns the controller. */
3529 for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3530 eecp = EHCI_EECP_NEXT(eec)) {
3531 eec = pci_read_config(self, eecp, 4);
3532 if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3535 bios_sem = pci_read_config(self, eecp +
3536 EHCI_LEGSUP_BIOS_SEM, 1);
3537 if (bios_sem == 0) {
3541 printf("ehci early: "
3542 "SMM active, request owner change\n");
3544 pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3546 for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3548 bios_sem = pci_read_config(self, eecp +
3549 EHCI_LEGSUP_BIOS_SEM, 1);
3552 if (bios_sem != 0) {
3554 printf("ehci early: "
3555 "SMM does not respond\n");
3557 /* Disable interrupts */
3558 offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3559 bus_write_4(res, offs + EHCI_USBINTR, 0);
3561 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3564 /* Perform early XHCI takeover from SMM. */
3566 xhci_early_takeover(device_t self)
3568 struct resource *res;
3578 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3582 cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3586 /* Synchronise with the BIOS if it owns the controller. */
3587 for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3588 eecp += XHCI_XECP_NEXT(eec) << 2) {
3589 eec = bus_read_4(res, eecp);
3591 if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3594 bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3599 printf("xhci early: "
3600 "SMM active, request owner change\n");
3602 bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3604 /* wait a maximum of 5 second */
3606 for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3608 bios_sem = bus_read_1(res, eecp +
3609 XHCI_XECP_BIOS_SEM);
3612 if (bios_sem != 0) {
3614 printf("xhci early: "
3615 "SMM does not respond\n");
3618 /* Disable interrupts */
3619 offs = bus_read_1(res, XHCI_CAPLENGTH);
3620 bus_write_4(res, offs + XHCI_USBCMD, 0);
3621 bus_read_4(res, offs + XHCI_USBSTS);
3623 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3626 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3628 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3629 struct resource_list *rl)
3631 struct resource *res;
3633 rman_res_t start, end, count;
3634 int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3636 switch (cfg->hdrtype & PCIM_HDRTYPE) {
3637 case PCIM_HDRTYPE_BRIDGE:
3638 sec_reg = PCIR_SECBUS_1;
3639 sub_reg = PCIR_SUBBUS_1;
3641 case PCIM_HDRTYPE_CARDBUS:
3642 sec_reg = PCIR_SECBUS_2;
3643 sub_reg = PCIR_SUBBUS_2;
3650 * If the existing bus range is valid, attempt to reserve it
3651 * from our parent. If this fails for any reason, clear the
3652 * secbus and subbus registers.
3654 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3655 * This would at least preserve the existing sec_bus if it is
3658 sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3659 sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3661 /* Quirk handling. */
3662 switch (pci_get_devid(dev)) {
3663 case 0x12258086: /* Intel 82454KX/GX (Orion) */
3664 sup_bus = pci_read_config(dev, 0x41, 1);
3665 if (sup_bus != 0xff) {
3666 sec_bus = sup_bus + 1;
3667 sub_bus = sup_bus + 1;
3668 PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3669 PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3674 /* Compaq R3000 BIOS sets wrong subordinate bus number. */
3675 if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3677 if (strncmp(cp, "Compal", 6) != 0) {
3682 if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3684 if (strncmp(cp, "08A0", 4) != 0) {
3689 if (sub_bus < 0xa) {
3691 PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3697 printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3698 if (sec_bus > 0 && sub_bus >= sec_bus) {
3701 count = end - start + 1;
3703 resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3706 * If requested, clear secondary bus registers in
3707 * bridge devices to force a complete renumbering
3708 * rather than reserving the existing range. However,
3709 * preserve the existing size.
3711 if (pci_clear_buses)
3715 res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3716 start, end, count, 0);
3722 "pci%d:%d:%d:%d secbus failed to allocate\n",
3723 pci_get_domain(dev), pci_get_bus(dev),
3724 pci_get_slot(dev), pci_get_function(dev));
3728 PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3729 PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3732 static struct resource *
3733 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3734 rman_res_t end, rman_res_t count, u_int flags)
3736 struct pci_devinfo *dinfo;
3738 struct resource_list *rl;
3739 struct resource *res;
3740 int sec_reg, sub_reg;
3742 dinfo = device_get_ivars(child);
3744 rl = &dinfo->resources;
3745 switch (cfg->hdrtype & PCIM_HDRTYPE) {
3746 case PCIM_HDRTYPE_BRIDGE:
3747 sec_reg = PCIR_SECBUS_1;
3748 sub_reg = PCIR_SUBBUS_1;
3750 case PCIM_HDRTYPE_CARDBUS:
3751 sec_reg = PCIR_SECBUS_2;
3752 sub_reg = PCIR_SUBBUS_2;
3761 if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3762 resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3763 if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3764 res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3765 start, end, count, flags & ~RF_ACTIVE);
3767 resource_list_delete(rl, PCI_RES_BUS, *rid);
3768 device_printf(child, "allocating %ju bus%s failed\n",
3769 count, count == 1 ? "" : "es");
3773 device_printf(child,
3774 "Lazy allocation of %ju bus%s at %ju\n", count,
3775 count == 1 ? "" : "es", rman_get_start(res));
3776 PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3777 PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3779 return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3780 end, count, flags));
3785 pci_ea_bei_to_rid(device_t dev, int bei)
3788 struct pci_devinfo *dinfo;
3790 struct pcicfg_iov *iov;
3792 dinfo = device_get_ivars(dev);
3793 iov = dinfo->cfg.iov;
3795 iov_pos = iov->iov_pos;
3800 /* Check if matches BAR */
3801 if ((bei >= PCIM_EA_BEI_BAR_0) &&
3802 (bei <= PCIM_EA_BEI_BAR_5))
3803 return (PCIR_BAR(bei));
3806 if (bei == PCIM_EA_BEI_ROM)
3810 /* Check if matches VF_BAR */
3811 if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3812 (bei <= PCIM_EA_BEI_VF_BAR_5))
3813 return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3821 pci_ea_is_enabled(device_t dev, int rid)
3823 struct pci_ea_entry *ea;
3824 struct pci_devinfo *dinfo;
3826 dinfo = device_get_ivars(dev);
3828 STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3829 if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3830 return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3837 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3839 struct pci_ea_entry *ea;
3840 struct pci_devinfo *dinfo;
3841 pci_addr_t start, end, count;
3842 struct resource_list *rl;
3843 int type, flags, rid;
3844 struct resource *res;
3847 struct pcicfg_iov *iov;
3850 dinfo = device_get_ivars(dev);
3851 rl = &dinfo->resources;
3855 iov = dinfo->cfg.iov;
3858 if (dinfo->cfg.ea.ea_location == 0)
3861 STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3864 * TODO: Ignore EA-BAR if is not enabled.
3865 * Currently the EA implementation supports
3866 * only situation, where EA structure contains
3867 * predefined entries. In case they are not enabled
3868 * leave them unallocated and proceed with
3869 * a legacy-BAR mechanism.
3871 if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3874 switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3875 case PCIM_EA_P_MEM_PREFETCH:
3876 case PCIM_EA_P_VF_MEM_PREFETCH:
3877 flags = RF_PREFETCHABLE;
3879 case PCIM_EA_P_VF_MEM:
3881 type = SYS_RES_MEMORY;
3884 type = SYS_RES_IOPORT;
3890 if (alloc_iov != 0) {
3892 /* Allocating IOV, confirm BEI matches */
3893 if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3894 (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3900 /* Allocating BAR, confirm BEI matches */
3901 if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3902 (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3903 (ea->eae_bei != PCIM_EA_BEI_ROM))
3907 rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3911 /* Skip resources already allocated by EA */
3912 if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3913 (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3916 start = ea->eae_base;
3917 count = ea->eae_max_offset + 1;
3920 count = count * iov->iov_num_vfs;
3922 end = start + count - 1;
3926 resource_list_add(rl, type, rid, start, end, count);
3927 res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3930 resource_list_delete(rl, type, rid);
3933 * Failed to allocate using EA, disable entry.
3934 * Another attempt to allocation will be performed
3935 * further, but this time using legacy BAR registers
3937 tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3938 tmp &= ~PCIM_EA_ENABLE;
3939 pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3942 * Disabling entry might fail in case it is hardwired.
3943 * Read flags again to match current status.
3945 ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3950 /* As per specification, fill BAR with zeros */
3951 pci_write_config(dev, rid, 0, 4);
3956 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3958 struct pci_devinfo *dinfo;
3960 struct resource_list *rl;
3961 const struct pci_quirk *q;
3965 dinfo = device_get_ivars(dev);
3967 rl = &dinfo->resources;
3968 devid = (cfg->device << 16) | cfg->vendor;
3970 /* Allocate resources using Enhanced Allocation */
3971 pci_add_resources_ea(bus, dev, 0);
3973 /* ATA devices needs special map treatment */
3974 if ((pci_get_class(dev) == PCIC_STORAGE) &&
3975 (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3976 ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3977 (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3978 !pci_read_config(dev, PCIR_BAR(2), 4))) )
3979 pci_ata_maps(bus, dev, rl, force, prefetchmask);
3981 for (i = 0; i < cfg->nummaps;) {
3982 /* Skip resources already managed by EA */
3983 if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3984 (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3985 pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3991 * Skip quirked resources.
3993 for (q = &pci_quirks[0]; q->devid != 0; q++)
3994 if (q->devid == devid &&
3995 q->type == PCI_QUIRK_UNMAP_REG &&
3996 q->arg1 == PCIR_BAR(i))
3998 if (q->devid != 0) {
4002 i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4003 prefetchmask & (1 << i));
4007 * Add additional, quirked resources.
4009 for (q = &pci_quirks[0]; q->devid != 0; q++)
4010 if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4011 pci_add_map(bus, dev, q->arg1, rl, force, 0);
4013 if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4014 #ifdef __PCI_REROUTE_INTERRUPT
4016 * Try to re-route interrupts. Sometimes the BIOS or
4017 * firmware may leave bogus values in these registers.
4018 * If the re-route fails, then just stick with what we
4021 pci_assign_interrupt(bus, dev, 1);
4023 pci_assign_interrupt(bus, dev, 0);
4027 if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4028 pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4029 if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4030 xhci_early_takeover(dev);
4031 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4032 ehci_early_takeover(dev);
4033 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4034 ohci_early_takeover(dev);
4035 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4036 uhci_early_takeover(dev);
4039 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4041 * Reserve resources for secondary bus ranges behind bridge
4044 pci_reserve_secbus(bus, dev, cfg, rl);
4048 static struct pci_devinfo *
4049 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4052 struct pci_devinfo *dinfo;
4054 dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4056 pci_add_child(dev, dinfo);
4062 pci_add_children(device_t dev, int domain, int busno)
4064 #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4065 device_t pcib = device_get_parent(dev);
4066 struct pci_devinfo *dinfo;
4068 int s, f, pcifunchigh;
4073 * Try to detect a device at slot 0, function 0. If it exists, try to
4074 * enable ARI. We must enable ARI before detecting the rest of the
4075 * functions on this bus as ARI changes the set of slots and functions
4076 * that are legal on this bus.
4078 dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4079 if (dinfo != NULL && pci_enable_ari)
4080 PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4083 * Start looking for new devices on slot 0 at function 1 because we
4084 * just identified the device at slot 0, function 0.
4088 maxslots = PCIB_MAXSLOTS(pcib);
4089 for (s = 0; s <= maxslots; s++, first_func = 0) {
4093 hdrtype = REG(PCIR_HDRTYPE, 1);
4094 if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4096 if (hdrtype & PCIM_MFDEV)
4097 pcifunchigh = PCIB_MAXFUNCS(pcib);
4098 for (f = first_func; f <= pcifunchigh; f++)
4099 pci_identify_function(pcib, dev, domain, busno, s, f);
4105 pci_rescan_method(device_t dev)
4107 #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4108 device_t pcib = device_get_parent(dev);
4109 device_t child, *devlist, *unchanged;
4110 int devcount, error, i, j, maxslots, oldcount;
4111 int busno, domain, s, f, pcifunchigh;
4114 /* No need to check for ARI on a rescan. */
4115 error = device_get_children(dev, &devlist, &devcount);
4118 if (devcount != 0) {
4119 unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4121 if (unchanged == NULL) {
4122 free(devlist, M_TEMP);
4128 domain = pcib_get_domain(dev);
4129 busno = pcib_get_bus(dev);
4130 maxslots = PCIB_MAXSLOTS(pcib);
4131 for (s = 0; s <= maxslots; s++) {
4132 /* If function 0 is not present, skip to the next slot. */
4134 if (REG(PCIR_VENDOR, 2) == 0xffff)
4137 hdrtype = REG(PCIR_HDRTYPE, 1);
4138 if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4140 if (hdrtype & PCIM_MFDEV)
4141 pcifunchigh = PCIB_MAXFUNCS(pcib);
4142 for (f = 0; f <= pcifunchigh; f++) {
4143 if (REG(PCIR_VENDOR, 2) == 0xffff)
4147 * Found a valid function. Check if a
4148 * device_t for this device already exists.
4150 for (i = 0; i < devcount; i++) {
4154 if (pci_get_slot(child) == s &&
4155 pci_get_function(child) == f) {
4156 unchanged[i] = child;
4161 pci_identify_function(pcib, dev, domain, busno, s, f);
4166 /* Remove devices that are no longer present. */
4167 for (i = 0; i < devcount; i++) {
4168 if (unchanged[i] != NULL)
4170 device_delete_child(dev, devlist[i]);
4173 free(devlist, M_TEMP);
4174 oldcount = devcount;
4176 /* Try to attach the devices just added. */
4177 error = device_get_children(dev, &devlist, &devcount);
4179 free(unchanged, M_TEMP);
4183 for (i = 0; i < devcount; i++) {
4184 for (j = 0; j < oldcount; j++) {
4185 if (devlist[i] == unchanged[j])
4189 device_probe_and_attach(devlist[i]);
4193 free(unchanged, M_TEMP);
4194 free(devlist, M_TEMP);
4201 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4204 struct pci_devinfo *vf_dinfo;
4206 int busno, slot, func;
4208 pcib = device_get_parent(bus);
4210 PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4212 vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4213 slot, func, vid, did);
4215 vf_dinfo->cfg.flags |= PCICFG_VF;
4216 pci_add_child(bus, vf_dinfo);
4218 return (vf_dinfo->cfg.dev);
4222 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4223 uint16_t vid, uint16_t did)
4226 return (pci_add_iov_child(bus, pf, rid, vid, did));
4231 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4237 if (dinfo->cfg.pcie.pcie_location != 0 &&
4238 dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4239 r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4241 r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4242 PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4243 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4244 PCIER_ROOT_CTL, r2, 2);
4246 if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4247 r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4248 pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4249 if (r != 0 && bootverbose) {
4250 pci_printf(&dinfo->cfg,
4251 "clearing AER UC 0x%08x -> 0x%08x\n",
4252 r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4256 r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4257 r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4258 PCIM_AER_UC_DL_PROTOCOL_ERROR |
4259 PCIM_AER_UC_SURPRISE_LINK_DOWN |
4260 PCIM_AER_UC_POISONED_TLP |
4261 PCIM_AER_UC_FC_PROTOCOL_ERROR |
4262 PCIM_AER_UC_COMPLETION_TIMEOUT |
4263 PCIM_AER_UC_COMPLETER_ABORT |
4264 PCIM_AER_UC_UNEXPECTED_COMPLETION |
4265 PCIM_AER_UC_RECEIVER_OVERFLOW |
4266 PCIM_AER_UC_MALFORMED_TLP |
4267 PCIM_AER_UC_ECRC_ERROR |
4268 PCIM_AER_UC_UNSUPPORTED_REQUEST |
4269 PCIM_AER_UC_ACS_VIOLATION |
4270 PCIM_AER_UC_INTERNAL_ERROR |
4271 PCIM_AER_UC_MC_BLOCKED_TLP |
4272 PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4273 PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4274 pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4276 r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4277 pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4278 if (r != 0 && bootverbose) {
4279 pci_printf(&dinfo->cfg,
4280 "clearing AER COR 0x%08x -> 0x%08x\n",
4281 r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4285 r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4286 r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4287 PCIM_AER_COR_BAD_TLP |
4288 PCIM_AER_COR_BAD_DLLP |
4289 PCIM_AER_COR_REPLAY_ROLLOVER |
4290 PCIM_AER_COR_REPLAY_TIMEOUT |
4291 PCIM_AER_COR_ADVISORY_NF_ERROR |
4292 PCIM_AER_COR_INTERNAL_ERROR |
4293 PCIM_AER_COR_HEADER_LOG_OVFLOW);
4294 pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4296 r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4297 PCIER_DEVICE_CTL, 2);
4298 r |= PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4299 PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4300 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4301 PCIER_DEVICE_CTL, r, 2);
4306 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4310 dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4311 device_set_ivars(dev, dinfo);
4312 resource_list_init(&dinfo->resources);
4313 pci_cfg_save(dev, dinfo, 0);
4314 pci_cfg_restore(dev, dinfo);
4315 pci_print_verbose(dinfo);
4316 pci_add_resources(bus, dev, 0, 0);
4317 pci_child_added(dinfo->cfg.dev);
4319 if (pci_clear_aer_on_attach)
4320 pci_add_child_clear_aer(dev, dinfo);
4322 EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4326 pci_child_added_method(device_t dev, device_t child)
4332 pci_probe(device_t dev)
4335 device_set_desc(dev, "PCI bus");
4337 /* Allow other subclasses to override this driver. */
4338 return (BUS_PROBE_GENERIC);
4342 pci_attach_common(device_t dev)
4344 struct pci_softc *sc;
4350 sc = device_get_softc(dev);
4351 domain = pcib_get_domain(dev);
4352 busno = pcib_get_bus(dev);
4355 sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4357 if (sc->sc_bus == NULL) {
4358 device_printf(dev, "failed to allocate bus number\n");
4363 device_printf(dev, "domain=%d, physical bus=%d\n",
4365 sc->sc_dma_tag = bus_get_dma_tag(dev);
4370 pci_attach(device_t dev)
4372 int busno, domain, error;
4374 error = pci_attach_common(dev);
4379 * Since there can be multiple independently numbered PCI
4380 * buses on systems with multiple PCI domains, we can't use
4381 * the unit number to decide which bus we are probing. We ask
4382 * the parent pcib what our domain and bus numbers are.
4384 domain = pcib_get_domain(dev);
4385 busno = pcib_get_bus(dev);
4386 pci_add_children(dev, domain, busno);
4387 return (bus_generic_attach(dev));
4391 pci_detach(device_t dev)
4394 struct pci_softc *sc;
4398 error = bus_generic_detach(dev);
4402 sc = device_get_softc(dev);
4403 error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4407 return (device_delete_children(dev));
4411 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4415 char me1[24], me2[32];
4419 d = pci_get_domain(child);
4420 b = pci_get_bus(child);
4421 s = pci_get_slot(child);
4422 f = pci_get_function(child);
4423 snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4424 snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4426 while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4427 resource_string_value(name, unit, "at", &at);
4428 if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4429 continue; /* No match, try next candidate */
4436 pci_set_power_child(device_t dev, device_t child, int state)
4442 * Set the device to the given state. If the firmware suggests
4443 * a different power state, use it instead. If power management
4444 * is not present, the firmware is responsible for managing
4445 * device power. Skip children who aren't attached since they
4446 * are handled separately.
4448 pcib = device_get_parent(dev);
4450 if (device_is_attached(child) &&
4451 PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4452 pci_set_powerstate(child, dstate);
4456 pci_suspend_child(device_t dev, device_t child)
4458 struct pci_devinfo *dinfo;
4459 struct resource_list_entry *rle;
4462 dinfo = device_get_ivars(child);
4465 * Save the PCI configuration space for the child and set the
4466 * device in the appropriate power state for this sleep state.
4468 pci_cfg_save(child, dinfo, 0);
4470 /* Suspend devices before potentially powering them down. */
4471 error = bus_generic_suspend_child(dev, child);
4476 if (pci_do_power_suspend) {
4478 * Make sure this device's interrupt handler is not invoked
4479 * in the case the device uses a shared interrupt that can
4480 * be raised by some other device.
4481 * This is applicable only to regular (legacy) PCI interrupts
4482 * as MSI/MSI-X interrupts are never shared.
4484 rle = resource_list_find(&dinfo->resources,
4486 if (rle != NULL && rle->res != NULL)
4487 (void)bus_suspend_intr(child, rle->res);
4488 pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4495 pci_resume_child(device_t dev, device_t child)
4497 struct pci_devinfo *dinfo;
4498 struct resource_list_entry *rle;
4500 if (pci_do_power_resume)
4501 pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4503 dinfo = device_get_ivars(child);
4504 pci_cfg_restore(child, dinfo);
4505 if (!device_is_attached(child))
4506 pci_cfg_save(child, dinfo, 1);
4508 bus_generic_resume_child(dev, child);
4511 * Allow interrupts only after fully resuming the driver and hardware.
4513 if (pci_do_power_suspend) {
4514 /* See pci_suspend_child for details. */
4515 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4516 if (rle != NULL && rle->res != NULL)
4517 (void)bus_resume_intr(child, rle->res);
4524 pci_resume(device_t dev)
4526 device_t child, *devlist;
4527 int error, i, numdevs;
4529 if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4533 * Resume critical devices first, then everything else later.
4535 for (i = 0; i < numdevs; i++) {
4537 switch (pci_get_class(child)) {
4541 case PCIC_BASEPERIPH:
4542 BUS_RESUME_CHILD(dev, child);
4546 for (i = 0; i < numdevs; i++) {
4548 switch (pci_get_class(child)) {
4552 case PCIC_BASEPERIPH:
4555 BUS_RESUME_CHILD(dev, child);
4558 free(devlist, M_TEMP);
4563 pci_load_vendor_data(void)
4569 data = preload_search_by_type("pci_vendor_data");
4571 ptr = preload_fetch_addr(data);
4572 sz = preload_fetch_size(data);
4573 if (ptr != NULL && sz != 0) {
4574 pci_vendordata = ptr;
4575 pci_vendordata_size = sz;
4576 /* terminate the database */
4577 pci_vendordata[pci_vendordata_size] = '\n';
4583 pci_driver_added(device_t dev, driver_t *driver)
4588 struct pci_devinfo *dinfo;
4592 device_printf(dev, "driver added\n");
4593 DEVICE_IDENTIFY(driver, dev);
4594 if (device_get_children(dev, &devlist, &numdevs) != 0)
4596 for (i = 0; i < numdevs; i++) {
4598 if (device_get_state(child) != DS_NOTPRESENT)
4600 dinfo = device_get_ivars(child);
4601 pci_print_verbose(dinfo);
4603 pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4604 pci_cfg_restore(child, dinfo);
4605 if (device_probe_and_attach(child) != 0)
4606 pci_child_detached(dev, child);
4608 free(devlist, M_TEMP);
4612 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4613 driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4615 struct pci_devinfo *dinfo;
4616 struct msix_table_entry *mte;
4617 struct msix_vector *mv;
4623 error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4628 /* If this is not a direct child, just bail out. */
4629 if (device_get_parent(child) != dev) {
4634 rid = rman_get_rid(irq);
4636 /* Make sure that INTx is enabled */
4637 pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4640 * Check to see if the interrupt is MSI or MSI-X.
4641 * Ask our parent to map the MSI and give
4642 * us the address and data register values.
4643 * If we fail for some reason, teardown the
4644 * interrupt handler.
4646 dinfo = device_get_ivars(child);
4647 if (dinfo->cfg.msi.msi_alloc > 0) {
4648 if (dinfo->cfg.msi.msi_addr == 0) {
4649 KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4650 ("MSI has handlers, but vectors not mapped"));
4651 error = PCIB_MAP_MSI(device_get_parent(dev),
4652 child, rman_get_start(irq), &addr, &data);
4655 dinfo->cfg.msi.msi_addr = addr;
4656 dinfo->cfg.msi.msi_data = data;
4658 if (dinfo->cfg.msi.msi_handlers == 0)
4659 pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4660 dinfo->cfg.msi.msi_data);
4661 dinfo->cfg.msi.msi_handlers++;
4663 KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4664 ("No MSI or MSI-X interrupts allocated"));
4665 KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4666 ("MSI-X index too high"));
4667 mte = &dinfo->cfg.msix.msix_table[rid - 1];
4668 KASSERT(mte->mte_vector != 0, ("no message vector"));
4669 mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4670 KASSERT(mv->mv_irq == rman_get_start(irq),
4672 if (mv->mv_address == 0) {
4673 KASSERT(mte->mte_handlers == 0,
4674 ("MSI-X table entry has handlers, but vector not mapped"));
4675 error = PCIB_MAP_MSI(device_get_parent(dev),
4676 child, rman_get_start(irq), &addr, &data);
4679 mv->mv_address = addr;
4684 * The MSIX table entry must be made valid by
4685 * incrementing the mte_handlers before
4686 * calling pci_enable_msix() and
4687 * pci_resume_msix(). Else the MSIX rewrite
4688 * table quirk will not work as expected.
4690 mte->mte_handlers++;
4691 if (mte->mte_handlers == 1) {
4692 pci_enable_msix(child, rid - 1, mv->mv_address,
4694 pci_unmask_msix(child, rid - 1);
4699 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4700 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4701 * in which case we "enable" INTx so MSI/MSI-X actually works.
4703 if (!pci_has_quirk(pci_get_devid(child),
4704 PCI_QUIRK_MSI_INTX_BUG))
4705 pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4707 pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4710 (void)bus_generic_teardown_intr(dev, child, irq,
4720 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4723 struct msix_table_entry *mte;
4724 struct resource_list_entry *rle;
4725 struct pci_devinfo *dinfo;
4728 if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4731 /* If this isn't a direct child, just bail out */
4732 if (device_get_parent(child) != dev)
4733 return(bus_generic_teardown_intr(dev, child, irq, cookie));
4735 rid = rman_get_rid(irq);
4738 pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4741 * Check to see if the interrupt is MSI or MSI-X. If so,
4742 * decrement the appropriate handlers count and mask the
4743 * MSI-X message, or disable MSI messages if the count
4746 dinfo = device_get_ivars(child);
4747 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4748 if (rle->res != irq)
4750 if (dinfo->cfg.msi.msi_alloc > 0) {
4751 KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4752 ("MSI-X index too high"));
4753 if (dinfo->cfg.msi.msi_handlers == 0)
4755 dinfo->cfg.msi.msi_handlers--;
4756 if (dinfo->cfg.msi.msi_handlers == 0)
4757 pci_disable_msi(child);
4759 KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4760 ("No MSI or MSI-X interrupts allocated"));
4761 KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4762 ("MSI-X index too high"));
4763 mte = &dinfo->cfg.msix.msix_table[rid - 1];
4764 if (mte->mte_handlers == 0)
4766 mte->mte_handlers--;
4767 if (mte->mte_handlers == 0)
4768 pci_mask_msix(child, rid - 1);
4771 error = bus_generic_teardown_intr(dev, child, irq, cookie);
4774 ("%s: generic teardown failed for MSI/MSI-X", __func__));
4779 pci_print_child(device_t dev, device_t child)
4781 struct pci_devinfo *dinfo;
4782 struct resource_list *rl;
4785 dinfo = device_get_ivars(child);
4786 rl = &dinfo->resources;
4788 retval += bus_print_child_header(dev, child);
4790 retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4791 retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4792 retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4793 if (device_get_flags(dev))
4794 retval += printf(" flags %#x", device_get_flags(dev));
4796 retval += printf(" at device %d.%d", pci_get_slot(child),
4797 pci_get_function(child));
4799 retval += bus_print_child_domain(dev, child);
4800 retval += bus_print_child_footer(dev, child);
4809 int report; /* 0 = bootverbose, 1 = always */
4811 } pci_nomatch_tab[] = {
4812 {PCIC_OLD, -1, 1, "old"},
4813 {PCIC_OLD, PCIS_OLD_NONVGA, 1, "non-VGA display device"},
4814 {PCIC_OLD, PCIS_OLD_VGA, 1, "VGA-compatible display device"},
4815 {PCIC_STORAGE, -1, 1, "mass storage"},
4816 {PCIC_STORAGE, PCIS_STORAGE_SCSI, 1, "SCSI"},
4817 {PCIC_STORAGE, PCIS_STORAGE_IDE, 1, "ATA"},
4818 {PCIC_STORAGE, PCIS_STORAGE_FLOPPY, 1, "floppy disk"},
4819 {PCIC_STORAGE, PCIS_STORAGE_IPI, 1, "IPI"},
4820 {PCIC_STORAGE, PCIS_STORAGE_RAID, 1, "RAID"},
4821 {PCIC_STORAGE, PCIS_STORAGE_ATA_ADMA, 1, "ATA (ADMA)"},
4822 {PCIC_STORAGE, PCIS_STORAGE_SATA, 1, "SATA"},
4823 {PCIC_STORAGE, PCIS_STORAGE_SAS, 1, "SAS"},
4824 {PCIC_STORAGE, PCIS_STORAGE_NVM, 1, "NVM"},
4825 {PCIC_NETWORK, -1, 1, "network"},
4826 {PCIC_NETWORK, PCIS_NETWORK_ETHERNET, 1, "ethernet"},
4827 {PCIC_NETWORK, PCIS_NETWORK_TOKENRING, 1, "token ring"},
4828 {PCIC_NETWORK, PCIS_NETWORK_FDDI, 1, "fddi"},
4829 {PCIC_NETWORK, PCIS_NETWORK_ATM, 1, "ATM"},
4830 {PCIC_NETWORK, PCIS_NETWORK_ISDN, 1, "ISDN"},
4831 {PCIC_DISPLAY, -1, 1, "display"},
4832 {PCIC_DISPLAY, PCIS_DISPLAY_VGA, 1, "VGA"},
4833 {PCIC_DISPLAY, PCIS_DISPLAY_XGA, 1, "XGA"},
4834 {PCIC_DISPLAY, PCIS_DISPLAY_3D, 1, "3D"},
4835 {PCIC_MULTIMEDIA, -1, 1, "multimedia"},
4836 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_VIDEO, 1, "video"},
4837 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_AUDIO, 1, "audio"},
4838 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_TELE, 1, "telephony"},
4839 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_HDA, 1, "HDA"},
4840 {PCIC_MEMORY, -1, 1, "memory"},
4841 {PCIC_MEMORY, PCIS_MEMORY_RAM, 1, "RAM"},
4842 {PCIC_MEMORY, PCIS_MEMORY_FLASH, 1, "flash"},
4843 {PCIC_BRIDGE, -1, 1, "bridge"},
4844 {PCIC_BRIDGE, PCIS_BRIDGE_HOST, 1, "HOST-PCI"},
4845 {PCIC_BRIDGE, PCIS_BRIDGE_ISA, 1, "PCI-ISA"},
4846 {PCIC_BRIDGE, PCIS_BRIDGE_EISA, 1, "PCI-EISA"},
4847 {PCIC_BRIDGE, PCIS_BRIDGE_MCA, 1, "PCI-MCA"},
4848 {PCIC_BRIDGE, PCIS_BRIDGE_PCI, 1, "PCI-PCI"},
4849 {PCIC_BRIDGE, PCIS_BRIDGE_PCMCIA, 1, "PCI-PCMCIA"},
4850 {PCIC_BRIDGE, PCIS_BRIDGE_NUBUS, 1, "PCI-NuBus"},
4851 {PCIC_BRIDGE, PCIS_BRIDGE_CARDBUS, 1, "PCI-CardBus"},
4852 {PCIC_BRIDGE, PCIS_BRIDGE_RACEWAY, 1, "PCI-RACEway"},
4853 {PCIC_SIMPLECOMM, -1, 1, "simple comms"},
4854 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_UART, 1, "UART"}, /* could detect 16550 */
4855 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_PAR, 1, "parallel port"},
4856 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MULSER, 1, "multiport serial"},
4857 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MODEM, 1, "generic modem"},
4858 {PCIC_BASEPERIPH, -1, 0, "base peripheral"},
4859 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PIC, 1, "interrupt controller"},
4860 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_DMA, 1, "DMA controller"},
4861 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_TIMER, 1, "timer"},
4862 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_RTC, 1, "realtime clock"},
4863 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PCIHOT, 1, "PCI hot-plug controller"},
4864 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_SDHC, 1, "SD host controller"},
4865 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_IOMMU, 1, "IOMMU"},
4866 {PCIC_INPUTDEV, -1, 1, "input device"},
4867 {PCIC_INPUTDEV, PCIS_INPUTDEV_KEYBOARD, 1, "keyboard"},
4868 {PCIC_INPUTDEV, PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4869 {PCIC_INPUTDEV, PCIS_INPUTDEV_MOUSE, 1, "mouse"},
4870 {PCIC_INPUTDEV, PCIS_INPUTDEV_SCANNER, 1, "scanner"},
4871 {PCIC_INPUTDEV, PCIS_INPUTDEV_GAMEPORT, 1, "gameport"},
4872 {PCIC_DOCKING, -1, 1, "docking station"},
4873 {PCIC_PROCESSOR, -1, 1, "processor"},
4874 {PCIC_SERIALBUS, -1, 1, "serial bus"},
4875 {PCIC_SERIALBUS, PCIS_SERIALBUS_FW, 1, "FireWire"},
4876 {PCIC_SERIALBUS, PCIS_SERIALBUS_ACCESS, 1, "AccessBus"},
4877 {PCIC_SERIALBUS, PCIS_SERIALBUS_SSA, 1, "SSA"},
4878 {PCIC_SERIALBUS, PCIS_SERIALBUS_USB, 1, "USB"},
4879 {PCIC_SERIALBUS, PCIS_SERIALBUS_FC, 1, "Fibre Channel"},
4880 {PCIC_SERIALBUS, PCIS_SERIALBUS_SMBUS, 0, "SMBus"},
4881 {PCIC_WIRELESS, -1, 1, "wireless controller"},
4882 {PCIC_WIRELESS, PCIS_WIRELESS_IRDA, 1, "iRDA"},
4883 {PCIC_WIRELESS, PCIS_WIRELESS_IR, 1, "IR"},
4884 {PCIC_WIRELESS, PCIS_WIRELESS_RF, 1, "RF"},
4885 {PCIC_INTELLIIO, -1, 1, "intelligent I/O controller"},
4886 {PCIC_INTELLIIO, PCIS_INTELLIIO_I2O, 1, "I2O"},
4887 {PCIC_SATCOM, -1, 1, "satellite communication"},
4888 {PCIC_SATCOM, PCIS_SATCOM_TV, 1, "sat TV"},
4889 {PCIC_SATCOM, PCIS_SATCOM_AUDIO, 1, "sat audio"},
4890 {PCIC_SATCOM, PCIS_SATCOM_VOICE, 1, "sat voice"},
4891 {PCIC_SATCOM, PCIS_SATCOM_DATA, 1, "sat data"},
4892 {PCIC_CRYPTO, -1, 1, "encrypt/decrypt"},
4893 {PCIC_CRYPTO, PCIS_CRYPTO_NETCOMP, 1, "network/computer crypto"},
4894 {PCIC_CRYPTO, PCIS_CRYPTO_ENTERTAIN, 1, "entertainment crypto"},
4895 {PCIC_DASP, -1, 0, "dasp"},
4896 {PCIC_DASP, PCIS_DASP_DPIO, 1, "DPIO module"},
4897 {PCIC_DASP, PCIS_DASP_PERFCNTRS, 1, "performance counters"},
4898 {PCIC_DASP, PCIS_DASP_COMM_SYNC, 1, "communication synchronizer"},
4899 {PCIC_DASP, PCIS_DASP_MGMT_CARD, 1, "signal processing management"},
4904 pci_probe_nomatch(device_t dev, device_t child)
4907 const char *cp, *scp;
4911 * Look for a listing for this device in a loaded device database.
4914 if ((device = pci_describe_device(child)) != NULL) {
4915 device_printf(dev, "<%s>", device);
4916 free(device, M_DEVBUF);
4919 * Scan the class/subclass descriptions for a general
4924 for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4925 if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4926 if (pci_nomatch_tab[i].subclass == -1) {
4927 cp = pci_nomatch_tab[i].desc;
4928 report = pci_nomatch_tab[i].report;
4929 } else if (pci_nomatch_tab[i].subclass ==
4930 pci_get_subclass(child)) {
4931 scp = pci_nomatch_tab[i].desc;
4932 report = pci_nomatch_tab[i].report;
4936 if (report || bootverbose) {
4937 device_printf(dev, "<%s%s%s>",
4939 ((cp != NULL) && (scp != NULL)) ? ", " : "",
4943 if (report || bootverbose) {
4944 printf(" at device %d.%d (no driver attached)\n",
4945 pci_get_slot(child), pci_get_function(child));
4947 pci_cfg_save(child, device_get_ivars(child), 1);
4951 pci_child_detached(device_t dev, device_t child)
4953 struct pci_devinfo *dinfo;
4954 struct resource_list *rl;
4956 dinfo = device_get_ivars(child);
4957 rl = &dinfo->resources;
4960 * Have to deallocate IRQs before releasing any MSI messages and
4961 * have to release MSI messages before deallocating any memory
4964 if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4965 pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4966 if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4967 pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4968 (void)pci_release_msi(child);
4970 if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4971 pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4972 if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4973 pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4975 if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4976 pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4979 pci_cfg_save(child, dinfo, 1);
4983 * Parse the PCI device database, if loaded, and return a pointer to a
4984 * description of the device.
4986 * The database is flat text formatted as follows:
4988 * Any line not in a valid format is ignored.
4989 * Lines are terminated with newline '\n' characters.
4991 * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
4994 * A DEVICE line is entered immediately below the corresponding VENDOR ID.
4995 * - devices cannot be listed without a corresponding VENDOR line.
4996 * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
4997 * another TAB, then the device name.
5001 * Assuming (ptr) points to the beginning of a line in the database,
5002 * return the vendor or device and description of the next entry.
5003 * The value of (vendor) or (device) inappropriate for the entry type
5004 * is set to -1. Returns nonzero at the end of the database.
5006 * Note that this is slightly unrobust in the face of corrupt data;
5007 * we attempt to safeguard against this by spamming the end of the
5008 * database with a newline when we initialise.
5011 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5020 left = pci_vendordata_size - (cp - pci_vendordata);
5028 sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5032 sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5035 /* skip to next line */
5036 while (*cp != '\n' && left > 0) {
5045 /* skip to next line */
5046 while (*cp != '\n' && left > 0) {
5050 if (*cp == '\n' && left > 0)
5057 pci_describe_device(device_t dev)
5060 char *desc, *vp, *dp, *line;
5062 desc = vp = dp = NULL;
5065 * If we have no vendor data, we can't do anything.
5067 if (pci_vendordata == NULL)
5071 * Scan the vendor data looking for this device
5073 line = pci_vendordata;
5074 if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5077 if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5079 if (vendor == pci_get_vendor(dev))
5082 if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5085 if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5093 if (device == pci_get_device(dev))
5097 snprintf(dp, 80, "0x%x", pci_get_device(dev));
5098 if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5100 sprintf(desc, "%s, %s", vp, dp);
5110 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5112 struct pci_devinfo *dinfo;
5115 dinfo = device_get_ivars(child);
5119 case PCI_IVAR_ETHADDR:
5121 * The generic accessor doesn't deal with failure, so
5122 * we set the return value, then return an error.
5124 *((uint8_t **) result) = NULL;
5126 case PCI_IVAR_SUBVENDOR:
5127 *result = cfg->subvendor;
5129 case PCI_IVAR_SUBDEVICE:
5130 *result = cfg->subdevice;
5132 case PCI_IVAR_VENDOR:
5133 *result = cfg->vendor;
5135 case PCI_IVAR_DEVICE:
5136 *result = cfg->device;
5138 case PCI_IVAR_DEVID:
5139 *result = (cfg->device << 16) | cfg->vendor;
5141 case PCI_IVAR_CLASS:
5142 *result = cfg->baseclass;
5144 case PCI_IVAR_SUBCLASS:
5145 *result = cfg->subclass;
5147 case PCI_IVAR_PROGIF:
5148 *result = cfg->progif;
5150 case PCI_IVAR_REVID:
5151 *result = cfg->revid;
5153 case PCI_IVAR_INTPIN:
5154 *result = cfg->intpin;
5157 *result = cfg->intline;
5159 case PCI_IVAR_DOMAIN:
5160 *result = cfg->domain;
5166 *result = cfg->slot;
5168 case PCI_IVAR_FUNCTION:
5169 *result = cfg->func;
5171 case PCI_IVAR_CMDREG:
5172 *result = cfg->cmdreg;
5174 case PCI_IVAR_CACHELNSZ:
5175 *result = cfg->cachelnsz;
5177 case PCI_IVAR_MINGNT:
5178 if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5182 *result = cfg->mingnt;
5184 case PCI_IVAR_MAXLAT:
5185 if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5189 *result = cfg->maxlat;
5191 case PCI_IVAR_LATTIMER:
5192 *result = cfg->lattimer;
5201 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5203 struct pci_devinfo *dinfo;
5205 dinfo = device_get_ivars(child);
5208 case PCI_IVAR_INTPIN:
5209 dinfo->cfg.intpin = value;
5211 case PCI_IVAR_ETHADDR:
5212 case PCI_IVAR_SUBVENDOR:
5213 case PCI_IVAR_SUBDEVICE:
5214 case PCI_IVAR_VENDOR:
5215 case PCI_IVAR_DEVICE:
5216 case PCI_IVAR_DEVID:
5217 case PCI_IVAR_CLASS:
5218 case PCI_IVAR_SUBCLASS:
5219 case PCI_IVAR_PROGIF:
5220 case PCI_IVAR_REVID:
5222 case PCI_IVAR_DOMAIN:
5225 case PCI_IVAR_FUNCTION:
5226 return (EINVAL); /* disallow for now */
5233 #include "opt_ddb.h"
5235 #include <ddb/ddb.h>
5236 #include <sys/cons.h>
5239 * List resources based on pci map registers, used for within ddb
5242 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5244 struct pci_devinfo *dinfo;
5245 struct devlist *devlist_head;
5248 int i, error, none_count;
5251 /* get the head of the device queue */
5252 devlist_head = &pci_devq;
5255 * Go through the list of devices and print out devices
5257 for (error = 0, i = 0,
5258 dinfo = STAILQ_FIRST(devlist_head);
5259 (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5260 dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5262 /* Populate pd_name and pd_unit */
5265 name = device_get_name(dinfo->cfg.dev);
5268 db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5269 "chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5270 (name && *name) ? name : "none",
5271 (name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5273 p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5274 p->pc_sel.pc_func, (p->pc_class << 16) |
5275 (p->pc_subclass << 8) | p->pc_progif,
5276 (p->pc_subdevice << 16) | p->pc_subvendor,
5277 (p->pc_device << 16) | p->pc_vendor,
5278 p->pc_revid, p->pc_hdr);
5283 static struct resource *
5284 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5285 rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5288 struct pci_devinfo *dinfo = device_get_ivars(child);
5289 struct resource_list *rl = &dinfo->resources;
5290 struct resource *res;
5293 pci_addr_t map, testval;
5298 /* If rid is managed by EA, ignore it */
5299 if (pci_ea_is_enabled(child, *rid))
5302 pm = pci_find_bar(child, *rid);
5304 /* This is a BAR that we failed to allocate earlier. */
5305 mapsize = pm->pm_size;
5309 * Weed out the bogons, and figure out how large the
5310 * BAR/map is. BARs that read back 0 here are bogus
5311 * and unimplemented. Note: atapci in legacy mode are
5312 * special and handled elsewhere in the code. If you
5313 * have a atapci device in legacy mode and it fails
5314 * here, that other code is broken.
5316 pci_read_bar(child, *rid, &map, &testval, NULL);
5319 * Determine the size of the BAR and ignore BARs with a size
5320 * of 0. Device ROM BARs use a different mask value.
5322 if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5323 mapsize = pci_romsize(testval);
5325 mapsize = pci_mapsize(testval);
5328 pm = pci_add_bar(child, *rid, map, mapsize);
5331 if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5332 if (type != SYS_RES_MEMORY) {
5335 "child %s requested type %d for rid %#x,"
5336 " but the BAR says it is an memio\n",
5337 device_get_nameunit(child), type, *rid);
5341 if (type != SYS_RES_IOPORT) {
5344 "child %s requested type %d for rid %#x,"
5345 " but the BAR says it is an ioport\n",
5346 device_get_nameunit(child), type, *rid);
5352 * For real BARs, we need to override the size that
5353 * the driver requests, because that's what the BAR
5354 * actually uses and we would otherwise have a
5355 * situation where we might allocate the excess to
5356 * another driver, which won't work.
5358 count = ((pci_addr_t)1 << mapsize) * num;
5359 if (RF_ALIGNMENT(flags) < mapsize)
5360 flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5361 if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5362 flags |= RF_PREFETCHABLE;
5365 * Allocate enough resource, and then write back the
5366 * appropriate BAR for that resource.
5368 resource_list_add(rl, type, *rid, start, end, count);
5369 res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5370 count, flags & ~RF_ACTIVE);
5372 resource_list_delete(rl, type, *rid);
5373 device_printf(child,
5374 "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5375 count, *rid, type, start, end);
5379 device_printf(child,
5380 "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5381 count, *rid, type, rman_get_start(res));
5383 /* Disable decoding via the CMD register before updating the BAR */
5384 cmd = pci_read_config(child, PCIR_COMMAND, 2);
5385 pci_write_config(child, PCIR_COMMAND,
5386 cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5388 map = rman_get_start(res);
5389 pci_write_bar(child, pm, map);
5391 /* Restore the original value of the CMD register */
5392 pci_write_config(child, PCIR_COMMAND, cmd, 2);
5398 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5399 rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5402 struct pci_devinfo *dinfo;
5403 struct resource_list *rl;
5404 struct resource_list_entry *rle;
5405 struct resource *res;
5409 * Perform lazy resource allocation
5411 dinfo = device_get_ivars(child);
5412 rl = &dinfo->resources;
5415 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5417 return (pci_alloc_secbus(dev, child, rid, start, end, count,
5422 * Can't alloc legacy interrupt once MSI messages have
5425 if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5426 cfg->msix.msix_alloc > 0))
5430 * If the child device doesn't have an interrupt
5431 * routed and is deserving of an interrupt, try to
5434 if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5436 pci_assign_interrupt(dev, child, 0);
5438 case SYS_RES_IOPORT:
5439 case SYS_RES_MEMORY:
5442 * PCI-PCI bridge I/O window resources are not BARs.
5443 * For those allocations just pass the request up the
5446 if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5448 case PCIR_IOBASEL_1:
5449 case PCIR_MEMBASE_1:
5450 case PCIR_PMBASEL_1:
5452 * XXX: Should we bother creating a resource
5455 return (bus_generic_alloc_resource(dev, child,
5456 type, rid, start, end, count, flags));
5460 /* Reserve resources for this BAR if needed. */
5461 rle = resource_list_find(rl, type, *rid);
5463 res = pci_reserve_map(dev, child, type, rid, start, end,
5469 return (resource_list_alloc(rl, dev, child, type, rid,
5470 start, end, count, flags));
5474 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5475 rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5478 struct pci_devinfo *dinfo;
5481 if (device_get_parent(child) != dev)
5482 return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5483 type, rid, start, end, count, flags));
5486 dinfo = device_get_ivars(child);
5487 if (dinfo->cfg.flags & PCICFG_VF) {
5489 /* VFs can't have I/O BARs. */
5490 case SYS_RES_IOPORT:
5492 case SYS_RES_MEMORY:
5493 return (pci_vf_alloc_mem_resource(dev, child, rid,
5494 start, end, count, flags));
5497 /* Fall through for other types of resource allocations. */
5501 return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5506 pci_release_resource(device_t dev, device_t child, int type, int rid,
5509 struct pci_devinfo *dinfo;
5510 struct resource_list *rl;
5513 if (device_get_parent(child) != dev)
5514 return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5517 dinfo = device_get_ivars(child);
5521 if (dinfo->cfg.flags & PCICFG_VF) {
5523 /* VFs can't have I/O BARs. */
5524 case SYS_RES_IOPORT:
5526 case SYS_RES_MEMORY:
5527 return (pci_vf_release_mem_resource(dev, child, rid,
5531 /* Fall through for other types of resource allocations. */
5537 * PCI-PCI bridge I/O window resources are not BARs. For
5538 * those allocations just pass the request up the tree.
5540 if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5541 (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5543 case PCIR_IOBASEL_1:
5544 case PCIR_MEMBASE_1:
5545 case PCIR_PMBASEL_1:
5546 return (bus_generic_release_resource(dev, child, type,
5552 rl = &dinfo->resources;
5553 return (resource_list_release(rl, dev, child, type, rid, r));
5557 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5560 struct pci_devinfo *dinfo;
5563 error = bus_generic_activate_resource(dev, child, type, rid, r);
5567 /* Enable decoding in the command register when activating BARs. */
5568 if (device_get_parent(child) == dev) {
5569 /* Device ROMs need their decoding explicitly enabled. */
5570 dinfo = device_get_ivars(child);
5571 if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5572 pci_write_bar(child, pci_find_bar(child, rid),
5573 rman_get_start(r) | PCIM_BIOS_ENABLE);
5575 case SYS_RES_IOPORT:
5576 case SYS_RES_MEMORY:
5577 error = PCI_ENABLE_IO(dev, child, type);
5585 pci_deactivate_resource(device_t dev, device_t child, int type,
5586 int rid, struct resource *r)
5588 struct pci_devinfo *dinfo;
5591 error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5595 /* Disable decoding for device ROMs. */
5596 if (device_get_parent(child) == dev) {
5597 dinfo = device_get_ivars(child);
5598 if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5599 pci_write_bar(child, pci_find_bar(child, rid),
5606 pci_child_deleted(device_t dev, device_t child)
5608 struct resource_list_entry *rle;
5609 struct resource_list *rl;
5610 struct pci_devinfo *dinfo;
5612 dinfo = device_get_ivars(child);
5613 rl = &dinfo->resources;
5615 EVENTHANDLER_INVOKE(pci_delete_device, child);
5617 /* Turn off access to resources we're about to free */
5618 if (bus_child_present(child) != 0) {
5619 pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5620 PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5622 pci_disable_busmaster(child);
5625 /* Free all allocated resources */
5626 STAILQ_FOREACH(rle, rl, link) {
5628 if (rman_get_flags(rle->res) & RF_ACTIVE ||
5629 resource_list_busy(rl, rle->type, rle->rid)) {
5630 pci_printf(&dinfo->cfg,
5631 "Resource still owned, oops. "
5632 "(type=%d, rid=%d, addr=%lx)\n",
5633 rle->type, rle->rid,
5634 rman_get_start(rle->res));
5635 bus_release_resource(child, rle->type, rle->rid,
5638 resource_list_unreserve(rl, dev, child, rle->type,
5642 resource_list_free(rl);
5648 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5650 struct pci_devinfo *dinfo;
5651 struct resource_list *rl;
5652 struct resource_list_entry *rle;
5654 if (device_get_parent(child) != dev)
5657 dinfo = device_get_ivars(child);
5658 rl = &dinfo->resources;
5659 rle = resource_list_find(rl, type, rid);
5664 if (rman_get_flags(rle->res) & RF_ACTIVE ||
5665 resource_list_busy(rl, type, rid)) {
5666 device_printf(dev, "delete_resource: "
5667 "Resource still owned by child, oops. "
5668 "(type=%d, rid=%d, addr=%jx)\n",
5669 type, rid, rman_get_start(rle->res));
5672 resource_list_unreserve(rl, dev, child, type, rid);
5674 resource_list_delete(rl, type, rid);
5677 struct resource_list *
5678 pci_get_resource_list (device_t dev, device_t child)
5680 struct pci_devinfo *dinfo = device_get_ivars(child);
5682 return (&dinfo->resources);
5686 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5688 pci_get_dma_tag(device_t bus, device_t dev)
5691 struct pci_softc *sc;
5693 if (device_get_parent(dev) == bus) {
5694 /* try dmar and return if it works */
5695 tag = dmar_get_dma_tag(bus, dev);
5699 sc = device_get_softc(bus);
5700 tag = sc->sc_dma_tag;
5706 pci_get_dma_tag(device_t bus, device_t dev)
5708 struct pci_softc *sc = device_get_softc(bus);
5710 return (sc->sc_dma_tag);
5715 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5717 struct pci_devinfo *dinfo = device_get_ivars(child);
5718 pcicfgregs *cfg = &dinfo->cfg;
5722 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5723 * emulate them here.
5725 if (cfg->flags & PCICFG_VF) {
5726 if (reg == PCIR_VENDOR) {
5729 return (cfg->device << 16 | cfg->vendor);
5731 return (cfg->vendor);
5733 return (cfg->vendor & 0xff);
5735 return (0xffffffff);
5737 } else if (reg == PCIR_DEVICE) {
5739 /* Note that an unaligned 4-byte read is an error. */
5741 return (cfg->device);
5743 return (cfg->device & 0xff);
5745 return (0xffffffff);
5751 return (PCIB_READ_CONFIG(device_get_parent(dev),
5752 cfg->bus, cfg->slot, cfg->func, reg, width));
5756 pci_write_config_method(device_t dev, device_t child, int reg,
5757 uint32_t val, int width)
5759 struct pci_devinfo *dinfo = device_get_ivars(child);
5760 pcicfgregs *cfg = &dinfo->cfg;
5762 PCIB_WRITE_CONFIG(device_get_parent(dev),
5763 cfg->bus, cfg->slot, cfg->func, reg, val, width);
5767 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5771 snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5772 pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5773 pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5778 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5781 struct pci_devinfo *dinfo;
5784 dinfo = device_get_ivars(child);
5786 snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5787 "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5788 cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5794 pci_assign_interrupt_method(device_t dev, device_t child)
5796 struct pci_devinfo *dinfo = device_get_ivars(child);
5797 pcicfgregs *cfg = &dinfo->cfg;
5799 return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5804 pci_lookup(void *arg, const char *name, device_t *dev)
5808 int domain, bus, slot, func;
5814 * Accept pciconf-style selectors of either pciD:B:S:F or
5815 * pciB:S:F. In the latter case, the domain is assumed to
5818 if (strncmp(name, "pci", 3) != 0)
5820 val = strtol(name + 3, &end, 10);
5821 if (val < 0 || val > INT_MAX || *end != ':')
5824 val = strtol(end + 1, &end, 10);
5825 if (val < 0 || val > INT_MAX || *end != ':')
5828 val = strtol(end + 1, &end, 10);
5829 if (val < 0 || val > INT_MAX)
5833 val = strtol(end + 1, &end, 10);
5834 if (val < 0 || val > INT_MAX || *end != '\0')
5837 } else if (*end == '\0') {
5845 if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5846 func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5849 *dev = pci_find_dbsf(domain, bus, slot, func);
5853 pci_modevent(module_t mod, int what, void *arg)
5855 static struct cdev *pci_cdev;
5856 static eventhandler_tag tag;
5860 STAILQ_INIT(&pci_devq);
5862 pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5864 pci_load_vendor_data();
5865 tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5871 EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5872 destroy_dev(pci_cdev);
5880 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5882 #define WREG(n, v) pci_write_config(dev, pos + (n), (v), 2)
5883 struct pcicfg_pcie *cfg;
5886 cfg = &dinfo->cfg.pcie;
5887 pos = cfg->pcie_location;
5889 version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5891 WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5893 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5894 cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5895 cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5896 WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5898 if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5899 (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5900 (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5901 WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5903 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5904 cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5905 WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5908 WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5909 WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5910 WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5916 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5918 pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5919 dinfo->cfg.pcix.pcix_command, 2);
5923 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5927 * Restore the device to full power mode. We must do this
5928 * before we restore the registers because moving from D3 to
5929 * D0 will cause the chip's BARs and some other registers to
5930 * be reset to some unknown power on reset values. Cut down
5931 * the noise on boot by doing nothing if we are already in
5934 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5935 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5936 pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5937 pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5938 pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5939 pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5940 pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5941 pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5942 pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5943 switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5944 case PCIM_HDRTYPE_NORMAL:
5945 pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5946 pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5948 case PCIM_HDRTYPE_BRIDGE:
5949 pci_write_config(dev, PCIR_SECLAT_1,
5950 dinfo->cfg.bridge.br_seclat, 1);
5951 pci_write_config(dev, PCIR_SUBBUS_1,
5952 dinfo->cfg.bridge.br_subbus, 1);
5953 pci_write_config(dev, PCIR_SECBUS_1,
5954 dinfo->cfg.bridge.br_secbus, 1);
5955 pci_write_config(dev, PCIR_PRIBUS_1,
5956 dinfo->cfg.bridge.br_pribus, 1);
5957 pci_write_config(dev, PCIR_BRIDGECTL_1,
5958 dinfo->cfg.bridge.br_control, 2);
5960 case PCIM_HDRTYPE_CARDBUS:
5961 pci_write_config(dev, PCIR_SECLAT_2,
5962 dinfo->cfg.bridge.br_seclat, 1);
5963 pci_write_config(dev, PCIR_SUBBUS_2,
5964 dinfo->cfg.bridge.br_subbus, 1);
5965 pci_write_config(dev, PCIR_SECBUS_2,
5966 dinfo->cfg.bridge.br_secbus, 1);
5967 pci_write_config(dev, PCIR_PRIBUS_2,
5968 dinfo->cfg.bridge.br_pribus, 1);
5969 pci_write_config(dev, PCIR_BRIDGECTL_2,
5970 dinfo->cfg.bridge.br_control, 2);
5973 pci_restore_bars(dev);
5976 * Restore extended capabilities for PCI-Express and PCI-X
5978 if (dinfo->cfg.pcie.pcie_location != 0)
5979 pci_cfg_restore_pcie(dev, dinfo);
5980 if (dinfo->cfg.pcix.pcix_location != 0)
5981 pci_cfg_restore_pcix(dev, dinfo);
5983 /* Restore MSI and MSI-X configurations if they are present. */
5984 if (dinfo->cfg.msi.msi_location != 0)
5985 pci_resume_msi(dev);
5986 if (dinfo->cfg.msix.msix_location != 0)
5987 pci_resume_msix(dev);
5990 if (dinfo->cfg.iov != NULL)
5991 pci_iov_cfg_restore(dev, dinfo);
5996 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
5998 #define RREG(n) pci_read_config(dev, pos + (n), 2)
5999 struct pcicfg_pcie *cfg;
6002 cfg = &dinfo->cfg.pcie;
6003 pos = cfg->pcie_location;
6005 cfg->pcie_flags = RREG(PCIER_FLAGS);
6007 version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6009 cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6011 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6012 cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6013 cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6014 cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6016 if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6017 (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6018 (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6019 cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6021 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6022 cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6023 cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6026 cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6027 cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6028 cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6034 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6036 dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6037 dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6041 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6047 * Some drivers apparently write to these registers w/o updating our
6048 * cached copy. No harm happens if we update the copy, so do so here
6049 * so we can restore them. The COMMAND register is modified by the
6050 * bus w/o updating the cache. This should represent the normally
6051 * writable portion of the 'defined' part of type 0/1/2 headers.
6053 dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6054 dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6055 dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6056 dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6057 dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6058 dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6059 dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6060 dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6061 dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6062 dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6063 dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6064 switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6065 case PCIM_HDRTYPE_NORMAL:
6066 dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6067 dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6068 dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6069 dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6071 case PCIM_HDRTYPE_BRIDGE:
6072 dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6074 dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6076 dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6078 dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6080 dinfo->cfg.bridge.br_control = pci_read_config(dev,
6081 PCIR_BRIDGECTL_1, 2);
6083 case PCIM_HDRTYPE_CARDBUS:
6084 dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6086 dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6088 dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6090 dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6092 dinfo->cfg.bridge.br_control = pci_read_config(dev,
6093 PCIR_BRIDGECTL_2, 2);
6094 dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6095 dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6099 if (dinfo->cfg.pcie.pcie_location != 0)
6100 pci_cfg_save_pcie(dev, dinfo);
6102 if (dinfo->cfg.pcix.pcix_location != 0)
6103 pci_cfg_save_pcix(dev, dinfo);
6106 if (dinfo->cfg.iov != NULL)
6107 pci_iov_cfg_save(dev, dinfo);
6111 * don't set the state for display devices, base peripherals and
6112 * memory devices since bad things happen when they are powered down.
6113 * We should (a) have drivers that can easily detach and (b) use
6114 * generic drivers for these devices so that some device actually
6115 * attaches. We need to make sure that when we implement (a) we don't
6116 * power the device down on a reattach.
6118 cls = pci_get_class(dev);
6121 switch (pci_do_power_nodriver)
6123 case 0: /* NO powerdown at all */
6125 case 1: /* Conservative about what to power down */
6126 if (cls == PCIC_STORAGE)
6129 case 2: /* Aggressive about what to power down */
6130 if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6131 cls == PCIC_BASEPERIPH)
6134 case 3: /* Power down everything */
6138 * PCI spec says we can only go into D3 state from D0 state.
6139 * Transition from D[12] into D0 before going to D3 state.
6141 ps = pci_get_powerstate(dev);
6142 if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6143 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6144 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6145 pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6148 /* Wrapper APIs suitable for device driver use. */
6150 pci_save_state(device_t dev)
6152 struct pci_devinfo *dinfo;
6154 dinfo = device_get_ivars(dev);
6155 pci_cfg_save(dev, dinfo, 0);
6159 pci_restore_state(device_t dev)
6161 struct pci_devinfo *dinfo;
6163 dinfo = device_get_ivars(dev);
6164 pci_cfg_restore(dev, dinfo);
6168 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6172 return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6175 /* Find the upstream port of a given PCI device in a root complex. */
6177 pci_find_pcie_root_port(device_t dev)
6179 struct pci_devinfo *dinfo;
6180 devclass_t pci_class;
6183 pci_class = devclass_find("pci");
6184 KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6185 ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6188 * Walk the bridge hierarchy until we find a PCI-e root
6189 * port or a non-PCI device.
6192 bus = device_get_parent(dev);
6193 KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6194 device_get_nameunit(dev)));
6196 pcib = device_get_parent(bus);
6197 KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6198 device_get_nameunit(bus)));
6201 * pcib's parent must be a PCI bus for this to be a
6204 if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6207 dinfo = device_get_ivars(pcib);
6208 if (dinfo->cfg.pcie.pcie_location != 0 &&
6209 dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6217 * Wait for pending transactions to complete on a PCI-express function.
6219 * The maximum delay is specified in milliseconds in max_delay. Note
6220 * that this function may sleep.
6222 * Returns true if the function is idle and false if the timeout is
6223 * exceeded. If dev is not a PCI-express function, this returns true.
6226 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6228 struct pci_devinfo *dinfo = device_get_ivars(dev);
6232 cap = dinfo->cfg.pcie.pcie_location;
6236 sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6237 while (sta & PCIEM_STA_TRANSACTION_PND) {
6241 /* Poll once every 100 milliseconds up to the timeout. */
6242 if (max_delay > 100) {
6243 pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6246 pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6250 sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6257 * Determine the maximum Completion Timeout in microseconds.
6259 * For non-PCI-express functions this returns 0.
6262 pcie_get_max_completion_timeout(device_t dev)
6264 struct pci_devinfo *dinfo = device_get_ivars(dev);
6267 cap = dinfo->cfg.pcie.pcie_location;
6272 * Functions using the 1.x spec use the default timeout range of
6273 * 50 microseconds to 50 milliseconds. Functions that do not
6274 * support programmable timeouts also use this range.
6276 if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6277 (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6278 PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6281 switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6282 PCIEM_CTL2_COMP_TIMO_VAL) {
6283 case PCIEM_CTL2_COMP_TIMO_100US:
6285 case PCIEM_CTL2_COMP_TIMO_10MS:
6287 case PCIEM_CTL2_COMP_TIMO_55MS:
6289 case PCIEM_CTL2_COMP_TIMO_210MS:
6290 return (210 * 1000);
6291 case PCIEM_CTL2_COMP_TIMO_900MS:
6292 return (900 * 1000);
6293 case PCIEM_CTL2_COMP_TIMO_3500MS:
6294 return (3500 * 1000);
6295 case PCIEM_CTL2_COMP_TIMO_13S:
6296 return (13 * 1000 * 1000);
6297 case PCIEM_CTL2_COMP_TIMO_64S:
6298 return (64 * 1000 * 1000);
6305 * Perform a Function Level Reset (FLR) on a device.
6307 * This function first waits for any pending transactions to complete
6308 * within the timeout specified by max_delay. If transactions are
6309 * still pending, the function will return false without attempting a
6312 * If dev is not a PCI-express function or does not support FLR, this
6313 * function returns false.
6315 * Note that no registers are saved or restored. The caller is
6316 * responsible for saving and restoring any registers including
6317 * PCI-standard registers via pci_save_state() and
6318 * pci_restore_state().
6321 pcie_flr(device_t dev, u_int max_delay, bool force)
6323 struct pci_devinfo *dinfo = device_get_ivars(dev);
6328 cap = dinfo->cfg.pcie.pcie_location;
6332 if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6336 * Disable busmastering to prevent generation of new
6337 * transactions while waiting for the device to go idle. If
6338 * the idle timeout fails, the command register is restored
6339 * which will re-enable busmastering.
6341 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6342 pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6343 if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6345 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6348 pci_printf(&dinfo->cfg,
6349 "Resetting with transactions pending after %d ms\n",
6353 * Extend the post-FLR delay to cover the maximum
6354 * Completion Timeout delay of anything in flight
6355 * during the FLR delay. Enforce a minimum delay of
6358 compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6359 if (compl_delay < 10)
6364 /* Initiate the reset. */
6365 ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6366 pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6367 PCIEM_CTL_INITIATE_FLR, 2);
6369 /* Wait for 100ms. */
6370 pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6372 if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6373 PCIEM_STA_TRANSACTION_PND)
6374 pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6379 * Attempt a power-management reset by cycling the device in/out of D3
6380 * state. PCI spec says we can only go into D3 state from D0 state.
6381 * Transition from D[12] into D0 before going to D3 state.
6384 pci_power_reset(device_t dev)
6388 ps = pci_get_powerstate(dev);
6389 if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6390 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6391 pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6392 pci_set_powerstate(dev, ps);
6397 * Try link drop and retrain of the downstream port of upstream
6398 * switch, for PCIe. According to the PCIe 3.0 spec 6.6.1, this must
6399 * cause Conventional Hot reset of the device in the slot.
6400 * Alternative, for PCIe, could be the secondary bus reset initiatied
6401 * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6404 pcie_link_reset(device_t port, int pcie_location)
6408 v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6409 v |= PCIEM_LINK_CTL_LINK_DIS;
6410 pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6411 pause_sbt("pcier1", mstosbt(20), 0, 0);
6412 v &= ~PCIEM_LINK_CTL_LINK_DIS;
6413 v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6414 pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6415 pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6416 v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6417 return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6421 pci_reset_post(device_t dev, device_t child)
6424 if (dev == device_get_parent(child))
6425 pci_restore_state(child);
6430 pci_reset_prepare(device_t dev, device_t child)
6433 if (dev == device_get_parent(child))
6434 pci_save_state(child);
6439 pci_reset_child(device_t dev, device_t child, int flags)
6443 if (dev == NULL || device_get_parent(child) != dev)
6445 if ((flags & DEVF_RESET_DETACH) != 0) {
6446 error = device_get_state(child) == DS_ATTACHED ?
6447 device_detach(child) : 0;
6449 error = BUS_SUSPEND_CHILD(dev, child);
6452 if (!pcie_flr(child, 1000, false)) {
6453 error = BUS_RESET_PREPARE(dev, child);
6455 pci_power_reset(child);
6456 BUS_RESET_POST(dev, child);
6458 if ((flags & DEVF_RESET_DETACH) != 0)
6459 device_probe_and_attach(child);
6461 BUS_RESUME_CHILD(dev, child);
6466 const struct pci_device_table *
6467 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6470 uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6472 vendor = pci_get_vendor(child);
6473 device = pci_get_device(child);
6474 subvendor = pci_get_subvendor(child);
6475 subdevice = pci_get_subdevice(child);
6476 class = pci_get_class(child);
6477 subclass = pci_get_subclass(child);
6478 revid = pci_get_revid(child);
6479 while (nelt-- > 0) {
6481 if (id->match_flag_vendor)
6482 match &= vendor == id->vendor;
6483 if (id->match_flag_device)
6484 match &= device == id->device;
6485 if (id->match_flag_subvendor)
6486 match &= subvendor == id->subvendor;
6487 if (id->match_flag_subdevice)
6488 match &= subdevice == id->subdevice;
6489 if (id->match_flag_class)
6490 match &= class == id->class_id;
6491 if (id->match_flag_subclass)
6492 match &= subclass == id->subclass;
6493 if (id->match_flag_revid)
6494 match &= revid == id->revid;
6503 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6505 const char *dev_name;
6508 dev = dinfo->cfg.dev;
6509 printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6510 dinfo->cfg.slot, dinfo->cfg.func);
6511 dev_name = device_get_name(dev);
6512 if (dev_name != NULL)
6513 printf(" (%s%d)", dev_name, device_get_unit(dev));
6517 pci_print_faulted_dev(void)
6519 struct pci_devinfo *dinfo;
6525 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6526 dev = dinfo->cfg.dev;
6527 status = pci_read_config(dev, PCIR_STATUS, 2);
6528 status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6529 PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6530 PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6532 pci_print_faulted_dev_name(dinfo);
6533 printf(" error 0x%04x\n", status);
6535 if (dinfo->cfg.pcie.pcie_location != 0) {
6536 status = pci_read_config(dev,
6537 dinfo->cfg.pcie.pcie_location +
6538 PCIER_DEVICE_STA, 2);
6539 if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6540 PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6541 PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6542 pci_print_faulted_dev_name(dinfo);
6543 printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6544 pci_read_config(dev,
6545 dinfo->cfg.pcie.pcie_location +
6546 PCIER_DEVICE_CTL, 2),
6550 if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6551 r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6552 r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6553 if (r1 != 0 || r2 != 0) {
6554 pci_print_faulted_dev_name(dinfo);
6555 printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6556 " COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6557 r1, pci_read_config(dev, aer +
6558 PCIR_AER_UC_MASK, 4),
6559 pci_read_config(dev, aer +
6560 PCIR_AER_UC_SEVERITY, 4),
6561 r2, pci_read_config(dev, aer +
6562 PCIR_AER_COR_MASK, 4),
6563 pci_read_config(dev, aer +
6564 PCIR_AER_CAP_CONTROL, 4));
6565 for (i = 0; i < 4; i++) {
6566 r1 = pci_read_config(dev, aer +
6567 PCIR_AER_HEADER_LOG + i * 4, 4);
6568 printf(" HL%d: 0x%08x\n", i, r1);
6576 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6579 pci_print_faulted_dev();
6583 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6589 dev = dinfo->cfg.dev;
6590 r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6591 PCIER_DEVICE_STA, 2);
6592 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6593 PCIER_DEVICE_STA, r, 2);
6595 if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6597 r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6599 pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6600 r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6602 pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6605 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6607 struct pci_devinfo *dinfo;
6609 uint16_t status, status1;
6611 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6612 dev = dinfo->cfg.dev;
6613 status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6614 status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6615 PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6616 PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6619 pci_write_config(dev, PCIR_STATUS, status, 2);
6621 if (dinfo->cfg.pcie.pcie_location != 0)
6622 db_clear_pcie_errors(dinfo);