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>
39 #include <sys/endian.h>
40 #include <sys/eventhandler.h>
41 #include <sys/fcntl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/linker.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/queue.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
53 #include <vm/vm_extern.h>
56 #include <machine/bus.h>
58 #include <machine/resource.h>
59 #include <machine/stdarg.h>
61 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
62 #include <machine/intr_machdep.h>
65 #include <sys/pciio.h>
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pci_private.h>
72 #include <dev/pci/pci_iov_private.h>
75 #include <dev/usb/controller/xhcireg.h>
76 #include <dev/usb/controller/ehcireg.h>
77 #include <dev/usb/controller/ohcireg.h>
78 #include <dev/usb/controller/uhcireg.h>
83 #define PCIR_IS_BIOS(cfg, reg) \
84 (((cfg)->hdrtype == PCIM_HDRTYPE_NORMAL && reg == PCIR_BIOS) || \
85 ((cfg)->hdrtype == PCIM_HDRTYPE_BRIDGE && reg == PCIR_BIOS_1))
87 static int pci_has_quirk(uint32_t devid, int quirk);
88 static pci_addr_t pci_mapbase(uint64_t mapreg);
89 static const char *pci_maptype(uint64_t mapreg);
90 static int pci_maprange(uint64_t mapreg);
91 static pci_addr_t pci_rombase(uint64_t mapreg);
92 static int pci_romsize(uint64_t testval);
93 static void pci_fixancient(pcicfgregs *cfg);
94 static int pci_printf(pcicfgregs *cfg, const char *fmt, ...);
96 static int pci_porten(device_t dev);
97 static int pci_memen(device_t dev);
98 static void pci_assign_interrupt(device_t bus, device_t dev,
100 static int pci_add_map(device_t bus, device_t dev, int reg,
101 struct resource_list *rl, int force, int prefetch);
102 static int pci_probe(device_t dev);
103 static int pci_attach(device_t dev);
104 static int pci_detach(device_t dev);
105 static void pci_load_vendor_data(void);
106 static int pci_describe_parse_line(char **ptr, int *vendor,
107 int *device, char **desc);
108 static char *pci_describe_device(device_t dev);
109 static int pci_modevent(module_t mod, int what, void *arg);
110 static void pci_hdrtypedata(device_t pcib, int b, int s, int f,
112 static void pci_read_cap(device_t pcib, pcicfgregs *cfg);
113 static int pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg,
114 int reg, uint32_t *data);
116 static int pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg,
117 int reg, uint32_t data);
119 static void pci_read_vpd(device_t pcib, pcicfgregs *cfg);
120 static void pci_mask_msix(device_t dev, u_int index);
121 static void pci_unmask_msix(device_t dev, u_int index);
122 static int pci_msi_blacklisted(void);
123 static int pci_msix_blacklisted(void);
124 static void pci_resume_msi(device_t dev);
125 static void pci_resume_msix(device_t dev);
126 static int pci_remap_intr_method(device_t bus, device_t dev,
128 static void pci_hint_device_unit(device_t acdev, device_t child,
129 const char *name, int *unitp);
130 static int pci_reset_post(device_t dev, device_t child);
131 static int pci_reset_prepare(device_t dev, device_t child);
132 static int pci_reset_child(device_t dev, device_t child,
135 static int pci_get_id_method(device_t dev, device_t child,
136 enum pci_id_type type, uintptr_t *rid);
138 static struct pci_devinfo * pci_fill_devinfo(device_t pcib, device_t bus, int d,
139 int b, int s, int f, uint16_t vid, uint16_t did);
141 static device_method_t pci_methods[] = {
142 /* Device interface */
143 DEVMETHOD(device_probe, pci_probe),
144 DEVMETHOD(device_attach, pci_attach),
145 DEVMETHOD(device_detach, pci_detach),
146 DEVMETHOD(device_shutdown, bus_generic_shutdown),
147 DEVMETHOD(device_suspend, bus_generic_suspend),
148 DEVMETHOD(device_resume, pci_resume),
151 DEVMETHOD(bus_print_child, pci_print_child),
152 DEVMETHOD(bus_probe_nomatch, pci_probe_nomatch),
153 DEVMETHOD(bus_read_ivar, pci_read_ivar),
154 DEVMETHOD(bus_write_ivar, pci_write_ivar),
155 DEVMETHOD(bus_driver_added, pci_driver_added),
156 DEVMETHOD(bus_setup_intr, pci_setup_intr),
157 DEVMETHOD(bus_teardown_intr, pci_teardown_intr),
158 DEVMETHOD(bus_reset_prepare, pci_reset_prepare),
159 DEVMETHOD(bus_reset_post, pci_reset_post),
160 DEVMETHOD(bus_reset_child, pci_reset_child),
162 DEVMETHOD(bus_get_dma_tag, pci_get_dma_tag),
163 DEVMETHOD(bus_get_resource_list,pci_get_resource_list),
164 DEVMETHOD(bus_set_resource, bus_generic_rl_set_resource),
165 DEVMETHOD(bus_get_resource, bus_generic_rl_get_resource),
166 DEVMETHOD(bus_delete_resource, pci_delete_resource),
167 DEVMETHOD(bus_alloc_resource, pci_alloc_resource),
168 DEVMETHOD(bus_adjust_resource, bus_generic_adjust_resource),
169 DEVMETHOD(bus_release_resource, pci_release_resource),
170 DEVMETHOD(bus_activate_resource, pci_activate_resource),
171 DEVMETHOD(bus_deactivate_resource, pci_deactivate_resource),
172 DEVMETHOD(bus_child_deleted, pci_child_deleted),
173 DEVMETHOD(bus_child_detached, pci_child_detached),
174 DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method),
175 DEVMETHOD(bus_child_location_str, pci_child_location_str_method),
176 DEVMETHOD(bus_hint_device_unit, pci_hint_device_unit),
177 DEVMETHOD(bus_remap_intr, pci_remap_intr_method),
178 DEVMETHOD(bus_suspend_child, pci_suspend_child),
179 DEVMETHOD(bus_resume_child, pci_resume_child),
180 DEVMETHOD(bus_rescan, pci_rescan_method),
183 DEVMETHOD(pci_read_config, pci_read_config_method),
184 DEVMETHOD(pci_write_config, pci_write_config_method),
185 DEVMETHOD(pci_enable_busmaster, pci_enable_busmaster_method),
186 DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method),
187 DEVMETHOD(pci_enable_io, pci_enable_io_method),
188 DEVMETHOD(pci_disable_io, pci_disable_io_method),
189 DEVMETHOD(pci_get_vpd_ident, pci_get_vpd_ident_method),
190 DEVMETHOD(pci_get_vpd_readonly, pci_get_vpd_readonly_method),
191 DEVMETHOD(pci_get_powerstate, pci_get_powerstate_method),
192 DEVMETHOD(pci_set_powerstate, pci_set_powerstate_method),
193 DEVMETHOD(pci_assign_interrupt, pci_assign_interrupt_method),
194 DEVMETHOD(pci_find_cap, pci_find_cap_method),
195 DEVMETHOD(pci_find_next_cap, pci_find_next_cap_method),
196 DEVMETHOD(pci_find_extcap, pci_find_extcap_method),
197 DEVMETHOD(pci_find_next_extcap, pci_find_next_extcap_method),
198 DEVMETHOD(pci_find_htcap, pci_find_htcap_method),
199 DEVMETHOD(pci_find_next_htcap, pci_find_next_htcap_method),
200 DEVMETHOD(pci_alloc_msi, pci_alloc_msi_method),
201 DEVMETHOD(pci_alloc_msix, pci_alloc_msix_method),
202 DEVMETHOD(pci_enable_msi, pci_enable_msi_method),
203 DEVMETHOD(pci_enable_msix, pci_enable_msix_method),
204 DEVMETHOD(pci_disable_msi, pci_disable_msi_method),
205 DEVMETHOD(pci_remap_msix, pci_remap_msix_method),
206 DEVMETHOD(pci_release_msi, pci_release_msi_method),
207 DEVMETHOD(pci_msi_count, pci_msi_count_method),
208 DEVMETHOD(pci_msix_count, pci_msix_count_method),
209 DEVMETHOD(pci_msix_pba_bar, pci_msix_pba_bar_method),
210 DEVMETHOD(pci_msix_table_bar, pci_msix_table_bar_method),
211 DEVMETHOD(pci_get_id, pci_get_id_method),
212 DEVMETHOD(pci_alloc_devinfo, pci_alloc_devinfo_method),
213 DEVMETHOD(pci_child_added, pci_child_added_method),
215 DEVMETHOD(pci_iov_attach, pci_iov_attach_method),
216 DEVMETHOD(pci_iov_detach, pci_iov_detach_method),
217 DEVMETHOD(pci_create_iov_child, pci_create_iov_child_method),
223 DEFINE_CLASS_0(pci, pci_driver, pci_methods, sizeof(struct pci_softc));
225 static devclass_t pci_devclass;
226 EARLY_DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, NULL,
228 MODULE_VERSION(pci, 1);
230 static char *pci_vendordata;
231 static size_t pci_vendordata_size;
234 uint32_t devid; /* Vendor/device of the card */
236 #define PCI_QUIRK_MAP_REG 1 /* PCI map register in weird place */
237 #define PCI_QUIRK_DISABLE_MSI 2 /* Neither MSI nor MSI-X work */
238 #define PCI_QUIRK_ENABLE_MSI_VM 3 /* Older chipset in VM where MSI works */
239 #define PCI_QUIRK_UNMAP_REG 4 /* Ignore PCI map register */
240 #define PCI_QUIRK_DISABLE_MSIX 5 /* MSI-X doesn't work */
241 #define PCI_QUIRK_MSI_INTX_BUG 6 /* PCIM_CMD_INTxDIS disables MSI */
242 #define PCI_QUIRK_REALLOC_BAR 7 /* Can't allocate memory at the default address */
247 static const struct pci_quirk pci_quirks[] = {
248 /* The Intel 82371AB and 82443MX have a map register at offset 0x90. */
249 { 0x71138086, PCI_QUIRK_MAP_REG, 0x90, 0 },
250 { 0x719b8086, PCI_QUIRK_MAP_REG, 0x90, 0 },
251 /* As does the Serverworks OSB4 (the SMBus mapping register) */
252 { 0x02001166, PCI_QUIRK_MAP_REG, 0x90, 0 },
255 * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge
256 * or the CMIC-SL (AKA ServerWorks GC_LE).
258 { 0x00141166, PCI_QUIRK_DISABLE_MSI, 0, 0 },
259 { 0x00171166, PCI_QUIRK_DISABLE_MSI, 0, 0 },
262 * MSI doesn't work on earlier Intel chipsets including
263 * E7500, E7501, E7505, 845, 865, 875/E7210, and 855.
265 { 0x25408086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
266 { 0x254c8086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
267 { 0x25508086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
268 { 0x25608086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
269 { 0x25708086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
270 { 0x25788086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
271 { 0x35808086, PCI_QUIRK_DISABLE_MSI, 0, 0 },
274 * MSI doesn't work with devices behind the AMD 8131 HT-PCIX
277 { 0x74501022, PCI_QUIRK_DISABLE_MSI, 0, 0 },
280 * MSI-X allocation doesn't work properly for devices passed through
281 * by VMware up to at least ESXi 5.1.
283 { 0x079015ad, PCI_QUIRK_DISABLE_MSIX, 0, 0 }, /* PCI/PCI-X */
284 { 0x07a015ad, PCI_QUIRK_DISABLE_MSIX, 0, 0 }, /* PCIe */
287 * Some virtualization environments emulate an older chipset
288 * but support MSI just fine. QEMU uses the Intel 82440.
290 { 0x12378086, PCI_QUIRK_ENABLE_MSI_VM, 0, 0 },
293 * HPET MMIO base address may appear in Bar1 for AMD SB600 SMBus
294 * controller depending on SoftPciRst register (PM_IO 0x55 [7]).
295 * It prevents us from attaching hpet(4) when the bit is unset.
296 * Note this quirk only affects SB600 revision A13 and earlier.
297 * For SB600 A21 and later, firmware must set the bit to hide it.
298 * For SB700 and later, it is unused and hardcoded to zero.
300 { 0x43851002, PCI_QUIRK_UNMAP_REG, 0x14, 0 },
303 * Atheros AR8161/AR8162/E2200/E2400/E2500 Ethernet controllers have
304 * a bug that MSI interrupt does not assert if PCIM_CMD_INTxDIS bit
305 * of the command register is set.
307 { 0x10911969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
308 { 0xE0911969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
309 { 0xE0A11969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
310 { 0xE0B11969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
311 { 0x10901969, PCI_QUIRK_MSI_INTX_BUG, 0, 0 },
314 * Broadcom BCM5714(S)/BCM5715(S)/BCM5780(S) Ethernet MACs don't
315 * issue MSI interrupts with PCIM_CMD_INTxDIS set either.
317 { 0x166814e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5714 */
318 { 0x166914e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5714S */
319 { 0x166a14e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5780 */
320 { 0x166b14e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5780S */
321 { 0x167814e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5715 */
322 { 0x167914e4, PCI_QUIRK_MSI_INTX_BUG, 0, 0 }, /* BCM5715S */
325 * HPE Gen 10 VGA has a memory range that can't be allocated in the
328 { 0x98741002, PCI_QUIRK_REALLOC_BAR, 0, 0 },
333 /* map register information */
334 #define PCI_MAPMEM 0x01 /* memory map */
335 #define PCI_MAPMEMP 0x02 /* prefetchable memory map */
336 #define PCI_MAPPORT 0x04 /* port map */
338 struct devlist pci_devq;
339 uint32_t pci_generation;
340 uint32_t pci_numdevs = 0;
341 static int pcie_chipset, pcix_chipset;
344 SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters");
346 static int pci_enable_io_modes = 1;
347 SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RWTUN,
348 &pci_enable_io_modes, 1,
349 "Enable I/O and memory bits in the config register. Some BIOSes do not"
350 " enable these bits correctly. We'd like to do this all the time, but"
351 " there are some peripherals that this causes problems with.");
353 static int pci_do_realloc_bars = 1;
354 SYSCTL_INT(_hw_pci, OID_AUTO, realloc_bars, CTLFLAG_RWTUN,
355 &pci_do_realloc_bars, 0,
356 "Attempt to allocate a new range for any BARs whose original "
357 "firmware-assigned ranges fail to allocate during the initial device scan.");
359 static int pci_do_power_nodriver = 0;
360 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RWTUN,
361 &pci_do_power_nodriver, 0,
362 "Place a function into D3 state when no driver attaches to it. 0 means"
363 " disable. 1 means conservatively place devices into D3 state. 2 means"
364 " aggressively place devices into D3 state. 3 means put absolutely"
365 " everything in D3 state.");
367 int pci_do_power_resume = 1;
368 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RWTUN,
369 &pci_do_power_resume, 1,
370 "Transition from D3 -> D0 on resume.");
372 int pci_do_power_suspend = 1;
373 SYSCTL_INT(_hw_pci, OID_AUTO, do_power_suspend, CTLFLAG_RWTUN,
374 &pci_do_power_suspend, 1,
375 "Transition from D0 -> D3 on suspend.");
377 static int pci_do_msi = 1;
378 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RWTUN, &pci_do_msi, 1,
379 "Enable support for MSI interrupts");
381 static int pci_do_msix = 1;
382 SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RWTUN, &pci_do_msix, 1,
383 "Enable support for MSI-X interrupts");
385 static int pci_msix_rewrite_table = 0;
386 SYSCTL_INT(_hw_pci, OID_AUTO, msix_rewrite_table, CTLFLAG_RWTUN,
387 &pci_msix_rewrite_table, 0,
388 "Rewrite entire MSI-X table when updating MSI-X entries");
390 static int pci_honor_msi_blacklist = 1;
391 SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RDTUN,
392 &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI/MSI-X");
394 #if defined(__i386__) || defined(__amd64__)
395 static int pci_usb_takeover = 1;
397 static int pci_usb_takeover = 0;
399 SYSCTL_INT(_hw_pci, OID_AUTO, usb_early_takeover, CTLFLAG_RDTUN,
400 &pci_usb_takeover, 1,
401 "Enable early takeover of USB controllers. Disable this if you depend on"
402 " BIOS emulation of USB devices, that is you use USB devices (like"
403 " keyboard or mouse) but do not load USB drivers");
405 static int pci_clear_bars;
406 SYSCTL_INT(_hw_pci, OID_AUTO, clear_bars, CTLFLAG_RDTUN, &pci_clear_bars, 0,
407 "Ignore firmware-assigned resources for BARs.");
409 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
410 static int pci_clear_buses;
411 SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
412 "Ignore firmware-assigned bus numbers.");
415 static int pci_enable_ari = 1;
416 SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
417 0, "Enable support for PCIe Alternative RID Interpretation");
419 static int pci_clear_aer_on_attach = 0;
420 SYSCTL_INT(_hw_pci, OID_AUTO, clear_aer_on_attach, CTLFLAG_RWTUN,
421 &pci_clear_aer_on_attach, 0,
422 "Clear port and device AER state on driver attach");
425 pci_has_quirk(uint32_t devid, int quirk)
427 const struct pci_quirk *q;
429 for (q = &pci_quirks[0]; q->devid; q++) {
430 if (q->devid == devid && q->type == quirk)
436 /* Find a device_t by bus/slot/function in domain 0 */
439 pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func)
442 return (pci_find_dbsf(0, bus, slot, func));
445 /* Find a device_t by domain/bus/slot/function */
448 pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func)
450 struct pci_devinfo *dinfo;
452 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
453 if ((dinfo->cfg.domain == domain) &&
454 (dinfo->cfg.bus == bus) &&
455 (dinfo->cfg.slot == slot) &&
456 (dinfo->cfg.func == func)) {
457 return (dinfo->cfg.dev);
464 /* Find a device_t by vendor/device ID */
467 pci_find_device(uint16_t vendor, uint16_t device)
469 struct pci_devinfo *dinfo;
471 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
472 if ((dinfo->cfg.vendor == vendor) &&
473 (dinfo->cfg.device == device)) {
474 return (dinfo->cfg.dev);
482 pci_find_class(uint8_t class, uint8_t subclass)
484 struct pci_devinfo *dinfo;
486 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
487 if (dinfo->cfg.baseclass == class &&
488 dinfo->cfg.subclass == subclass) {
489 return (dinfo->cfg.dev);
497 pci_printf(pcicfgregs *cfg, const char *fmt, ...)
502 retval = printf("pci%d:%d:%d:%d: ", cfg->domain, cfg->bus, cfg->slot,
505 retval += vprintf(fmt, ap);
510 /* return base address of memory or port map */
513 pci_mapbase(uint64_t mapreg)
516 if (PCI_BAR_MEM(mapreg))
517 return (mapreg & PCIM_BAR_MEM_BASE);
519 return (mapreg & PCIM_BAR_IO_BASE);
522 /* return map type of memory or port map */
525 pci_maptype(uint64_t mapreg)
528 if (PCI_BAR_IO(mapreg))
530 if (mapreg & PCIM_BAR_MEM_PREFETCH)
531 return ("Prefetchable Memory");
535 /* return log2 of map size decoded for memory or port map */
538 pci_mapsize(uint64_t testval)
542 testval = pci_mapbase(testval);
545 while ((testval & 1) == 0)
554 /* return base address of device ROM */
557 pci_rombase(uint64_t mapreg)
560 return (mapreg & PCIM_BIOS_ADDR_MASK);
563 /* return log2 of map size decided for device ROM */
566 pci_romsize(uint64_t testval)
570 testval = pci_rombase(testval);
573 while ((testval & 1) == 0)
582 /* return log2 of address range supported by map register */
585 pci_maprange(uint64_t mapreg)
589 if (PCI_BAR_IO(mapreg))
592 switch (mapreg & PCIM_BAR_MEM_TYPE) {
593 case PCIM_BAR_MEM_32:
596 case PCIM_BAR_MEM_1MB:
599 case PCIM_BAR_MEM_64:
606 /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */
609 pci_fixancient(pcicfgregs *cfg)
611 if ((cfg->hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
614 /* PCI to PCI bridges use header type 1 */
615 if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI)
616 cfg->hdrtype = PCIM_HDRTYPE_BRIDGE;
619 /* extract header type specific config data */
622 pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg)
624 #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w)
625 switch (cfg->hdrtype & PCIM_HDRTYPE) {
626 case PCIM_HDRTYPE_NORMAL:
627 cfg->subvendor = REG(PCIR_SUBVEND_0, 2);
628 cfg->subdevice = REG(PCIR_SUBDEV_0, 2);
629 cfg->mingnt = REG(PCIR_MINGNT, 1);
630 cfg->maxlat = REG(PCIR_MAXLAT, 1);
631 cfg->nummaps = PCI_MAXMAPS_0;
633 case PCIM_HDRTYPE_BRIDGE:
634 cfg->bridge.br_seclat = REG(PCIR_SECLAT_1, 1);
635 cfg->bridge.br_subbus = REG(PCIR_SUBBUS_1, 1);
636 cfg->bridge.br_secbus = REG(PCIR_SECBUS_1, 1);
637 cfg->bridge.br_pribus = REG(PCIR_PRIBUS_1, 1);
638 cfg->bridge.br_control = REG(PCIR_BRIDGECTL_1, 2);
639 cfg->nummaps = PCI_MAXMAPS_1;
641 case PCIM_HDRTYPE_CARDBUS:
642 cfg->bridge.br_seclat = REG(PCIR_SECLAT_2, 1);
643 cfg->bridge.br_subbus = REG(PCIR_SUBBUS_2, 1);
644 cfg->bridge.br_secbus = REG(PCIR_SECBUS_2, 1);
645 cfg->bridge.br_pribus = REG(PCIR_PRIBUS_2, 1);
646 cfg->bridge.br_control = REG(PCIR_BRIDGECTL_2, 2);
647 cfg->subvendor = REG(PCIR_SUBVEND_2, 2);
648 cfg->subdevice = REG(PCIR_SUBDEV_2, 2);
649 cfg->nummaps = PCI_MAXMAPS_2;
655 /* read configuration header into pcicfgregs structure */
657 pci_read_device(device_t pcib, device_t bus, int d, int b, int s, int f)
659 #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w)
662 vid = REG(PCIR_VENDOR, 2);
663 did = REG(PCIR_DEVICE, 2);
665 return (pci_fill_devinfo(pcib, bus, d, b, s, f, vid, did));
671 pci_alloc_devinfo_method(device_t dev)
674 return (malloc(sizeof(struct pci_devinfo), M_DEVBUF,
678 static struct pci_devinfo *
679 pci_fill_devinfo(device_t pcib, device_t bus, int d, int b, int s, int f,
680 uint16_t vid, uint16_t did)
682 struct pci_devinfo *devlist_entry;
685 devlist_entry = PCI_ALLOC_DEVINFO(bus);
687 cfg = &devlist_entry->cfg;
695 cfg->cmdreg = REG(PCIR_COMMAND, 2);
696 cfg->statreg = REG(PCIR_STATUS, 2);
697 cfg->baseclass = REG(PCIR_CLASS, 1);
698 cfg->subclass = REG(PCIR_SUBCLASS, 1);
699 cfg->progif = REG(PCIR_PROGIF, 1);
700 cfg->revid = REG(PCIR_REVID, 1);
701 cfg->hdrtype = REG(PCIR_HDRTYPE, 1);
702 cfg->cachelnsz = REG(PCIR_CACHELNSZ, 1);
703 cfg->lattimer = REG(PCIR_LATTIMER, 1);
704 cfg->intpin = REG(PCIR_INTPIN, 1);
705 cfg->intline = REG(PCIR_INTLINE, 1);
707 cfg->mfdev = (cfg->hdrtype & PCIM_MFDEV) != 0;
708 cfg->hdrtype &= ~PCIM_MFDEV;
709 STAILQ_INIT(&cfg->maps);
714 pci_hdrtypedata(pcib, b, s, f, cfg);
716 if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT)
717 pci_read_cap(pcib, cfg);
719 STAILQ_INSERT_TAIL(&pci_devq, devlist_entry, pci_links);
721 devlist_entry->conf.pc_sel.pc_domain = cfg->domain;
722 devlist_entry->conf.pc_sel.pc_bus = cfg->bus;
723 devlist_entry->conf.pc_sel.pc_dev = cfg->slot;
724 devlist_entry->conf.pc_sel.pc_func = cfg->func;
725 devlist_entry->conf.pc_hdr = cfg->hdrtype;
727 devlist_entry->conf.pc_subvendor = cfg->subvendor;
728 devlist_entry->conf.pc_subdevice = cfg->subdevice;
729 devlist_entry->conf.pc_vendor = cfg->vendor;
730 devlist_entry->conf.pc_device = cfg->device;
732 devlist_entry->conf.pc_class = cfg->baseclass;
733 devlist_entry->conf.pc_subclass = cfg->subclass;
734 devlist_entry->conf.pc_progif = cfg->progif;
735 devlist_entry->conf.pc_revid = cfg->revid;
740 return (devlist_entry);
745 pci_ea_fill_info(device_t pcib, pcicfgregs *cfg)
747 #define REG(n, w) PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, \
748 cfg->ea.ea_location + (n), w)
755 uint64_t base, max_offset;
756 struct pci_ea_entry *eae;
758 if (cfg->ea.ea_location == 0)
761 STAILQ_INIT(&cfg->ea.ea_entries);
763 /* Determine the number of entries */
764 num_ent = REG(PCIR_EA_NUM_ENT, 2);
765 num_ent &= PCIM_EA_NUM_ENT_MASK;
767 /* Find the first entry to care of */
768 ptr = PCIR_EA_FIRST_ENT;
770 /* Skip DWORD 2 for type 1 functions */
771 if ((cfg->hdrtype & PCIM_HDRTYPE) == PCIM_HDRTYPE_BRIDGE)
774 for (a = 0; a < num_ent; a++) {
776 eae = malloc(sizeof(*eae), M_DEVBUF, M_WAITOK | M_ZERO);
777 eae->eae_cfg_offset = cfg->ea.ea_location + ptr;
779 /* Read a number of dwords in the entry */
782 ent_size = (val & PCIM_EA_ES);
784 for (b = 0; b < ent_size; b++) {
789 eae->eae_flags = val;
790 eae->eae_bei = (PCIM_EA_BEI & val) >> PCIM_EA_BEI_OFFSET;
792 base = dw[0] & PCIM_EA_FIELD_MASK;
793 max_offset = dw[1] | ~PCIM_EA_FIELD_MASK;
795 if (((dw[0] & PCIM_EA_IS_64) != 0) && (b < ent_size)) {
796 base |= (uint64_t)dw[b] << 32UL;
799 if (((dw[1] & PCIM_EA_IS_64) != 0)
801 max_offset |= (uint64_t)dw[b] << 32UL;
805 eae->eae_base = base;
806 eae->eae_max_offset = max_offset;
808 STAILQ_INSERT_TAIL(&cfg->ea.ea_entries, eae, eae_link);
811 printf("PCI(EA) dev %04x:%04x, bei %d, flags #%x, base #%jx, max_offset #%jx\n",
812 cfg->vendor, cfg->device, eae->eae_bei, eae->eae_flags,
813 (uintmax_t)eae->eae_base, (uintmax_t)eae->eae_max_offset);
820 pci_read_cap(device_t pcib, pcicfgregs *cfg)
822 #define REG(n, w) PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w)
823 #define WREG(n, v, w) PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w)
824 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
828 int ptr, nextptr, ptrptr;
830 switch (cfg->hdrtype & PCIM_HDRTYPE) {
831 case PCIM_HDRTYPE_NORMAL:
832 case PCIM_HDRTYPE_BRIDGE:
833 ptrptr = PCIR_CAP_PTR;
835 case PCIM_HDRTYPE_CARDBUS:
836 ptrptr = PCIR_CAP_PTR_2; /* cardbus capabilities ptr */
839 return; /* no extended capabilities support */
841 nextptr = REG(ptrptr, 1); /* sanity check? */
844 * Read capability entries.
846 while (nextptr != 0) {
849 printf("illegal PCI extended capability offset %d\n",
853 /* Find the next entry */
855 nextptr = REG(ptr + PCICAP_NEXTPTR, 1);
857 /* Process this entry */
858 switch (REG(ptr + PCICAP_ID, 1)) {
859 case PCIY_PMG: /* PCI power management */
860 if (cfg->pp.pp_cap == 0) {
861 cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2);
862 cfg->pp.pp_status = ptr + PCIR_POWER_STATUS;
863 cfg->pp.pp_bse = ptr + PCIR_POWER_BSE;
864 if ((nextptr - ptr) > PCIR_POWER_DATA)
865 cfg->pp.pp_data = ptr + PCIR_POWER_DATA;
868 case PCIY_HT: /* HyperTransport */
869 /* Determine HT-specific capability type. */
870 val = REG(ptr + PCIR_HT_COMMAND, 2);
872 if ((val & 0xe000) == PCIM_HTCAP_SLAVE)
873 cfg->ht.ht_slave = ptr;
875 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
876 switch (val & PCIM_HTCMD_CAP_MASK) {
877 case PCIM_HTCAP_MSI_MAPPING:
878 if (!(val & PCIM_HTCMD_MSI_FIXED)) {
879 /* Sanity check the mapping window. */
880 addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI,
883 addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO,
885 if (addr != MSI_INTEL_ADDR_BASE)
887 "HT device at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n",
888 cfg->domain, cfg->bus,
889 cfg->slot, cfg->func,
892 addr = MSI_INTEL_ADDR_BASE;
894 cfg->ht.ht_msimap = ptr;
895 cfg->ht.ht_msictrl = val;
896 cfg->ht.ht_msiaddr = addr;
901 case PCIY_MSI: /* PCI MSI */
902 cfg->msi.msi_location = ptr;
903 cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2);
904 cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl &
905 PCIM_MSICTRL_MMC_MASK)>>1);
907 case PCIY_MSIX: /* PCI MSI-X */
908 cfg->msix.msix_location = ptr;
909 cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2);
910 cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl &
911 PCIM_MSIXCTRL_TABLE_SIZE) + 1;
912 val = REG(ptr + PCIR_MSIX_TABLE, 4);
913 cfg->msix.msix_table_bar = PCIR_BAR(val &
915 cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK;
916 val = REG(ptr + PCIR_MSIX_PBA, 4);
917 cfg->msix.msix_pba_bar = PCIR_BAR(val &
919 cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK;
921 case PCIY_VPD: /* PCI Vital Product Data */
922 cfg->vpd.vpd_reg = ptr;
925 /* Should always be true. */
926 if ((cfg->hdrtype & PCIM_HDRTYPE) ==
927 PCIM_HDRTYPE_BRIDGE) {
928 val = REG(ptr + PCIR_SUBVENDCAP_ID, 4);
929 cfg->subvendor = val & 0xffff;
930 cfg->subdevice = val >> 16;
933 case PCIY_PCIX: /* PCI-X */
935 * Assume we have a PCI-X chipset if we have
936 * at least one PCI-PCI bridge with a PCI-X
937 * capability. Note that some systems with
938 * PCI-express or HT chipsets might match on
939 * this check as well.
941 if ((cfg->hdrtype & PCIM_HDRTYPE) ==
944 cfg->pcix.pcix_location = ptr;
946 case PCIY_EXPRESS: /* PCI-express */
948 * Assume we have a PCI-express chipset if we have
949 * at least one PCI-express device.
952 cfg->pcie.pcie_location = ptr;
953 val = REG(ptr + PCIER_FLAGS, 2);
954 cfg->pcie.pcie_type = val & PCIEM_FLAGS_TYPE;
956 case PCIY_EA: /* Enhanced Allocation */
957 cfg->ea.ea_location = ptr;
958 pci_ea_fill_info(pcib, cfg);
965 #if defined(__powerpc__)
967 * Enable the MSI mapping window for all HyperTransport
968 * slaves. PCI-PCI bridges have their windows enabled via
971 if (cfg->ht.ht_slave != 0 && cfg->ht.ht_msimap != 0 &&
972 !(cfg->ht.ht_msictrl & PCIM_HTCMD_MSI_ENABLE)) {
974 "Enabling MSI window for HyperTransport slave at pci%d:%d:%d:%d\n",
975 cfg->domain, cfg->bus, cfg->slot, cfg->func);
976 cfg->ht.ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
977 WREG(cfg->ht.ht_msimap + PCIR_HT_COMMAND, cfg->ht.ht_msictrl,
981 /* REG and WREG use carry through to next functions */
985 * PCI Vital Product Data
988 #define PCI_VPD_TIMEOUT 1000000
991 pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data)
993 int count = PCI_VPD_TIMEOUT;
995 KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
997 WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2);
999 while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) {
1002 DELAY(1); /* limit looping */
1004 *data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4));
1011 pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data)
1013 int count = PCI_VPD_TIMEOUT;
1015 KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned"));
1017 WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4);
1018 WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2);
1019 while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) {
1022 DELAY(1); /* limit looping */
1029 #undef PCI_VPD_TIMEOUT
1031 struct vpd_readstate {
1041 vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data)
1046 if (vrs->bytesinval == 0) {
1047 if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, ®))
1049 vrs->val = le32toh(reg);
1051 byte = vrs->val & 0xff;
1052 vrs->bytesinval = 3;
1054 vrs->val = vrs->val >> 8;
1055 byte = vrs->val & 0xff;
1065 pci_read_vpd(device_t pcib, pcicfgregs *cfg)
1067 struct vpd_readstate vrs;
1072 int alloc, off; /* alloc/off for RO/W arrays */
1078 /* init vpd reader */
1086 name = remain = i = 0; /* shut up stupid gcc */
1087 alloc = off = 0; /* shut up stupid gcc */
1088 dflen = 0; /* shut up stupid gcc */
1090 while (state >= 0) {
1091 if (vpd_nextbyte(&vrs, &byte)) {
1096 printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \
1097 "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val,
1098 vrs.off, vrs.bytesinval, byte, state, remain, name, i);
1101 case 0: /* item name */
1103 if (vpd_nextbyte(&vrs, &byte2)) {
1108 if (vpd_nextbyte(&vrs, &byte2)) {
1112 remain |= byte2 << 8;
1113 if (remain > (0x7f*4 - vrs.off)) {
1116 "invalid VPD data, remain %#x\n",
1121 remain = byte & 0x7;
1122 name = (byte >> 3) & 0xf;
1125 case 0x2: /* String */
1126 cfg->vpd.vpd_ident = malloc(remain + 1,
1127 M_DEVBUF, M_WAITOK);
1134 case 0x10: /* VPD-R */
1137 cfg->vpd.vpd_ros = malloc(alloc *
1138 sizeof(*cfg->vpd.vpd_ros), M_DEVBUF,
1142 case 0x11: /* VPD-W */
1145 cfg->vpd.vpd_w = malloc(alloc *
1146 sizeof(*cfg->vpd.vpd_w), M_DEVBUF,
1150 default: /* Invalid data, abort */
1156 case 1: /* Identifier String */
1157 cfg->vpd.vpd_ident[i++] = byte;
1160 cfg->vpd.vpd_ident[i] = '\0';
1165 case 2: /* VPD-R Keyword Header */
1167 cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1168 (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros),
1169 M_DEVBUF, M_WAITOK | M_ZERO);
1171 cfg->vpd.vpd_ros[off].keyword[0] = byte;
1172 if (vpd_nextbyte(&vrs, &byte2)) {
1176 cfg->vpd.vpd_ros[off].keyword[1] = byte2;
1177 if (vpd_nextbyte(&vrs, &byte2)) {
1181 cfg->vpd.vpd_ros[off].len = dflen = byte2;
1183 strncmp(cfg->vpd.vpd_ros[off].keyword, "RV",
1186 * if this happens, we can't trust the rest
1189 pci_printf(cfg, "bad keyword length: %d\n",
1194 } else if (dflen == 0) {
1195 cfg->vpd.vpd_ros[off].value = malloc(1 *
1196 sizeof(*cfg->vpd.vpd_ros[off].value),
1197 M_DEVBUF, M_WAITOK);
1198 cfg->vpd.vpd_ros[off].value[0] = '\x00';
1200 cfg->vpd.vpd_ros[off].value = malloc(
1202 sizeof(*cfg->vpd.vpd_ros[off].value),
1203 M_DEVBUF, M_WAITOK);
1206 /* keep in sync w/ state 3's transistions */
1207 if (dflen == 0 && remain == 0)
1209 else if (dflen == 0)
1215 case 3: /* VPD-R Keyword Value */
1216 cfg->vpd.vpd_ros[off].value[i++] = byte;
1217 if (strncmp(cfg->vpd.vpd_ros[off].keyword,
1218 "RV", 2) == 0 && cksumvalid == -1) {
1224 "bad VPD cksum, remain %hhu\n",
1233 /* keep in sync w/ state 2's transistions */
1235 cfg->vpd.vpd_ros[off++].value[i++] = '\0';
1236 if (dflen == 0 && remain == 0) {
1237 cfg->vpd.vpd_rocnt = off;
1238 cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros,
1239 off * sizeof(*cfg->vpd.vpd_ros),
1240 M_DEVBUF, M_WAITOK | M_ZERO);
1242 } else if (dflen == 0)
1252 case 5: /* VPD-W Keyword Header */
1254 cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1255 (alloc *= 2) * sizeof(*cfg->vpd.vpd_w),
1256 M_DEVBUF, M_WAITOK | M_ZERO);
1258 cfg->vpd.vpd_w[off].keyword[0] = byte;
1259 if (vpd_nextbyte(&vrs, &byte2)) {
1263 cfg->vpd.vpd_w[off].keyword[1] = byte2;
1264 if (vpd_nextbyte(&vrs, &byte2)) {
1268 cfg->vpd.vpd_w[off].len = dflen = byte2;
1269 cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval;
1270 cfg->vpd.vpd_w[off].value = malloc((dflen + 1) *
1271 sizeof(*cfg->vpd.vpd_w[off].value),
1272 M_DEVBUF, M_WAITOK);
1275 /* keep in sync w/ state 6's transistions */
1276 if (dflen == 0 && remain == 0)
1278 else if (dflen == 0)
1284 case 6: /* VPD-W Keyword Value */
1285 cfg->vpd.vpd_w[off].value[i++] = byte;
1288 /* keep in sync w/ state 5's transistions */
1290 cfg->vpd.vpd_w[off++].value[i++] = '\0';
1291 if (dflen == 0 && remain == 0) {
1292 cfg->vpd.vpd_wcnt = off;
1293 cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w,
1294 off * sizeof(*cfg->vpd.vpd_w),
1295 M_DEVBUF, M_WAITOK | M_ZERO);
1297 } else if (dflen == 0)
1302 pci_printf(cfg, "invalid state: %d\n", state);
1308 if (cksumvalid == 0 || state < -1) {
1309 /* read-only data bad, clean up */
1310 if (cfg->vpd.vpd_ros != NULL) {
1311 for (off = 0; cfg->vpd.vpd_ros[off].value; off++)
1312 free(cfg->vpd.vpd_ros[off].value, M_DEVBUF);
1313 free(cfg->vpd.vpd_ros, M_DEVBUF);
1314 cfg->vpd.vpd_ros = NULL;
1318 /* I/O error, clean up */
1319 pci_printf(cfg, "failed to read VPD data.\n");
1320 if (cfg->vpd.vpd_ident != NULL) {
1321 free(cfg->vpd.vpd_ident, M_DEVBUF);
1322 cfg->vpd.vpd_ident = NULL;
1324 if (cfg->vpd.vpd_w != NULL) {
1325 for (off = 0; cfg->vpd.vpd_w[off].value; off++)
1326 free(cfg->vpd.vpd_w[off].value, M_DEVBUF);
1327 free(cfg->vpd.vpd_w, M_DEVBUF);
1328 cfg->vpd.vpd_w = NULL;
1331 cfg->vpd.vpd_cached = 1;
1337 pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr)
1339 struct pci_devinfo *dinfo = device_get_ivars(child);
1340 pcicfgregs *cfg = &dinfo->cfg;
1342 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1343 pci_read_vpd(device_get_parent(dev), cfg);
1345 *identptr = cfg->vpd.vpd_ident;
1347 if (*identptr == NULL)
1354 pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw,
1357 struct pci_devinfo *dinfo = device_get_ivars(child);
1358 pcicfgregs *cfg = &dinfo->cfg;
1361 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1362 pci_read_vpd(device_get_parent(dev), cfg);
1364 for (i = 0; i < cfg->vpd.vpd_rocnt; i++)
1365 if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword,
1366 sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) {
1367 *vptr = cfg->vpd.vpd_ros[i].value;
1376 pci_fetch_vpd_list(device_t dev)
1378 struct pci_devinfo *dinfo = device_get_ivars(dev);
1379 pcicfgregs *cfg = &dinfo->cfg;
1381 if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0)
1382 pci_read_vpd(device_get_parent(device_get_parent(dev)), cfg);
1387 * Find the requested HyperTransport capability and return the offset
1388 * in configuration space via the pointer provided. The function
1389 * returns 0 on success and an error code otherwise.
1392 pci_find_htcap_method(device_t dev, device_t child, int capability, int *capreg)
1397 error = pci_find_cap(child, PCIY_HT, &ptr);
1402 * Traverse the capabilities list checking each HT capability
1403 * to see if it matches the requested HT capability.
1406 val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1407 if (capability == PCIM_HTCAP_SLAVE ||
1408 capability == PCIM_HTCAP_HOST)
1411 val &= PCIM_HTCMD_CAP_MASK;
1412 if (val == capability) {
1418 /* Skip to the next HT capability. */
1419 if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1427 * Find the next requested HyperTransport capability after start and return
1428 * the offset in configuration space via the pointer provided. The function
1429 * returns 0 on success and an error code otherwise.
1432 pci_find_next_htcap_method(device_t dev, device_t child, int capability,
1433 int start, int *capreg)
1438 KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == PCIY_HT,
1439 ("start capability is not HyperTransport capability"));
1443 * Traverse the capabilities list checking each HT capability
1444 * to see if it matches the requested HT capability.
1447 /* Skip to the next HT capability. */
1448 if (pci_find_next_cap(child, PCIY_HT, ptr, &ptr) != 0)
1451 val = pci_read_config(child, ptr + PCIR_HT_COMMAND, 2);
1452 if (capability == PCIM_HTCAP_SLAVE ||
1453 capability == PCIM_HTCAP_HOST)
1456 val &= PCIM_HTCMD_CAP_MASK;
1457 if (val == capability) {
1468 * Find the requested capability and return the offset in
1469 * configuration space via the pointer provided. The function returns
1470 * 0 on success and an error code otherwise.
1473 pci_find_cap_method(device_t dev, device_t child, int capability,
1476 struct pci_devinfo *dinfo = device_get_ivars(child);
1477 pcicfgregs *cfg = &dinfo->cfg;
1482 * Check the CAP_LIST bit of the PCI status register first.
1484 status = pci_read_config(child, PCIR_STATUS, 2);
1485 if (!(status & PCIM_STATUS_CAPPRESENT))
1489 * Determine the start pointer of the capabilities list.
1491 switch (cfg->hdrtype & PCIM_HDRTYPE) {
1492 case PCIM_HDRTYPE_NORMAL:
1493 case PCIM_HDRTYPE_BRIDGE:
1496 case PCIM_HDRTYPE_CARDBUS:
1497 ptr = PCIR_CAP_PTR_2;
1501 return (ENXIO); /* no extended capabilities support */
1503 ptr = pci_read_config(child, ptr, 1);
1506 * Traverse the capabilities list.
1509 if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1514 ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1521 * Find the next requested capability after start and return the offset in
1522 * configuration space via the pointer provided. The function returns
1523 * 0 on success and an error code otherwise.
1526 pci_find_next_cap_method(device_t dev, device_t child, int capability,
1527 int start, int *capreg)
1531 KASSERT(pci_read_config(child, start + PCICAP_ID, 1) == capability,
1532 ("start capability is not expected capability"));
1534 ptr = pci_read_config(child, start + PCICAP_NEXTPTR, 1);
1536 if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) {
1541 ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1);
1548 * Find the requested extended capability and return the offset in
1549 * configuration space via the pointer provided. The function returns
1550 * 0 on success and an error code otherwise.
1553 pci_find_extcap_method(device_t dev, device_t child, int capability,
1556 struct pci_devinfo *dinfo = device_get_ivars(child);
1557 pcicfgregs *cfg = &dinfo->cfg;
1561 /* Only supported for PCI-express devices. */
1562 if (cfg->pcie.pcie_location == 0)
1566 ecap = pci_read_config(child, ptr, 4);
1567 if (ecap == 0xffffffff || ecap == 0)
1570 if (PCI_EXTCAP_ID(ecap) == capability) {
1575 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1578 ecap = pci_read_config(child, ptr, 4);
1585 * Find the next requested extended capability after start and return the
1586 * offset in configuration space via the pointer provided. The function
1587 * returns 0 on success and an error code otherwise.
1590 pci_find_next_extcap_method(device_t dev, device_t child, int capability,
1591 int start, int *capreg)
1593 struct pci_devinfo *dinfo = device_get_ivars(child);
1594 pcicfgregs *cfg = &dinfo->cfg;
1598 /* Only supported for PCI-express devices. */
1599 if (cfg->pcie.pcie_location == 0)
1602 ecap = pci_read_config(child, start, 4);
1603 KASSERT(PCI_EXTCAP_ID(ecap) == capability,
1604 ("start extended capability is not expected capability"));
1605 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1607 ecap = pci_read_config(child, ptr, 4);
1608 if (PCI_EXTCAP_ID(ecap) == capability) {
1613 ptr = PCI_EXTCAP_NEXTPTR(ecap);
1620 * Support for MSI-X message interrupts.
1623 pci_write_msix_entry(device_t dev, u_int index, uint64_t address, uint32_t data)
1625 struct pci_devinfo *dinfo = device_get_ivars(dev);
1626 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1629 KASSERT(msix->msix_table_len > index, ("bogus index"));
1630 offset = msix->msix_table_offset + index * 16;
1631 bus_write_4(msix->msix_table_res, offset, address & 0xffffffff);
1632 bus_write_4(msix->msix_table_res, offset + 4, address >> 32);
1633 bus_write_4(msix->msix_table_res, offset + 8, data);
1637 pci_enable_msix_method(device_t dev, device_t child, u_int index,
1638 uint64_t address, uint32_t data)
1641 if (pci_msix_rewrite_table) {
1642 struct pci_devinfo *dinfo = device_get_ivars(child);
1643 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1646 * Some VM hosts require MSIX to be disabled in the
1647 * control register before updating the MSIX table
1648 * entries are allowed. It is not enough to only
1649 * disable MSIX while updating a single entry. MSIX
1650 * must be disabled while updating all entries in the
1653 pci_write_config(child,
1654 msix->msix_location + PCIR_MSIX_CTRL,
1655 msix->msix_ctrl & ~PCIM_MSIXCTRL_MSIX_ENABLE, 2);
1656 pci_resume_msix(child);
1658 pci_write_msix_entry(child, index, address, data);
1660 /* Enable MSI -> HT mapping. */
1661 pci_ht_map_msi(child, address);
1665 pci_mask_msix(device_t dev, u_int index)
1667 struct pci_devinfo *dinfo = device_get_ivars(dev);
1668 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1669 uint32_t offset, val;
1671 KASSERT(msix->msix_msgnum > index, ("bogus index"));
1672 offset = msix->msix_table_offset + index * 16 + 12;
1673 val = bus_read_4(msix->msix_table_res, offset);
1674 val |= PCIM_MSIX_VCTRL_MASK;
1677 * Some devices (e.g. Samsung PM961) do not support reads of this
1678 * register, so always write the new value.
1680 bus_write_4(msix->msix_table_res, offset, val);
1684 pci_unmask_msix(device_t dev, u_int index)
1686 struct pci_devinfo *dinfo = device_get_ivars(dev);
1687 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1688 uint32_t offset, val;
1690 KASSERT(msix->msix_table_len > index, ("bogus index"));
1691 offset = msix->msix_table_offset + index * 16 + 12;
1692 val = bus_read_4(msix->msix_table_res, offset);
1693 val &= ~PCIM_MSIX_VCTRL_MASK;
1696 * Some devices (e.g. Samsung PM961) do not support reads of this
1697 * register, so always write the new value.
1699 bus_write_4(msix->msix_table_res, offset, val);
1703 pci_pending_msix(device_t dev, u_int index)
1705 struct pci_devinfo *dinfo = device_get_ivars(dev);
1706 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1707 uint32_t offset, bit;
1709 KASSERT(msix->msix_table_len > index, ("bogus index"));
1710 offset = msix->msix_pba_offset + (index / 32) * 4;
1711 bit = 1 << index % 32;
1712 return (bus_read_4(msix->msix_pba_res, offset) & bit);
1716 * Restore MSI-X registers and table during resume. If MSI-X is
1717 * enabled then walk the virtual table to restore the actual MSI-X
1721 pci_resume_msix(device_t dev)
1723 struct pci_devinfo *dinfo = device_get_ivars(dev);
1724 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1725 struct msix_table_entry *mte;
1726 struct msix_vector *mv;
1729 if (msix->msix_alloc > 0) {
1730 /* First, mask all vectors. */
1731 for (i = 0; i < msix->msix_msgnum; i++)
1732 pci_mask_msix(dev, i);
1734 /* Second, program any messages with at least one handler. */
1735 for (i = 0; i < msix->msix_table_len; i++) {
1736 mte = &msix->msix_table[i];
1737 if (mte->mte_vector == 0 || mte->mte_handlers == 0)
1739 mv = &msix->msix_vectors[mte->mte_vector - 1];
1740 pci_write_msix_entry(dev, i, mv->mv_address,
1742 pci_unmask_msix(dev, i);
1745 pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL,
1746 msix->msix_ctrl, 2);
1750 * Attempt to allocate *count MSI-X messages. The actual number allocated is
1751 * returned in *count. After this function returns, each message will be
1752 * available to the driver as SYS_RES_IRQ resources starting at rid 1.
1755 pci_alloc_msix_method(device_t dev, device_t child, int *count)
1757 struct pci_devinfo *dinfo = device_get_ivars(child);
1758 pcicfgregs *cfg = &dinfo->cfg;
1759 struct resource_list_entry *rle;
1760 int actual, error, i, irq, max;
1762 /* Don't let count == 0 get us into trouble. */
1766 /* If rid 0 is allocated, then fail. */
1767 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
1768 if (rle != NULL && rle->res != NULL)
1771 /* Already have allocated messages? */
1772 if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
1775 /* If MSI-X is blacklisted for this system, fail. */
1776 if (pci_msix_blacklisted())
1779 /* MSI-X capability present? */
1780 if (cfg->msix.msix_location == 0 || !pci_do_msix)
1783 /* Make sure the appropriate BARs are mapped. */
1784 rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1785 cfg->msix.msix_table_bar);
1786 if (rle == NULL || rle->res == NULL ||
1787 !(rman_get_flags(rle->res) & RF_ACTIVE))
1789 cfg->msix.msix_table_res = rle->res;
1790 if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) {
1791 rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY,
1792 cfg->msix.msix_pba_bar);
1793 if (rle == NULL || rle->res == NULL ||
1794 !(rman_get_flags(rle->res) & RF_ACTIVE))
1797 cfg->msix.msix_pba_res = rle->res;
1800 device_printf(child,
1801 "attempting to allocate %d MSI-X vectors (%d supported)\n",
1802 *count, cfg->msix.msix_msgnum);
1803 max = min(*count, cfg->msix.msix_msgnum);
1804 for (i = 0; i < max; i++) {
1805 /* Allocate a message. */
1806 error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq);
1812 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
1818 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1);
1820 device_printf(child, "using IRQ %ju for MSI-X\n",
1826 * Be fancy and try to print contiguous runs of
1827 * IRQ values as ranges. 'irq' is the previous IRQ.
1828 * 'run' is true if we are in a range.
1830 device_printf(child, "using IRQs %ju", rle->start);
1833 for (i = 1; i < actual; i++) {
1834 rle = resource_list_find(&dinfo->resources,
1835 SYS_RES_IRQ, i + 1);
1837 /* Still in a run? */
1838 if (rle->start == irq + 1) {
1844 /* Finish previous range. */
1850 /* Start new range. */
1851 printf(",%ju", rle->start);
1855 /* Unfinished range? */
1858 printf(" for MSI-X\n");
1862 /* Mask all vectors. */
1863 for (i = 0; i < cfg->msix.msix_msgnum; i++)
1864 pci_mask_msix(child, i);
1866 /* Allocate and initialize vector data and virtual table. */
1867 cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual,
1868 M_DEVBUF, M_WAITOK | M_ZERO);
1869 cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual,
1870 M_DEVBUF, M_WAITOK | M_ZERO);
1871 for (i = 0; i < actual; i++) {
1872 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1873 cfg->msix.msix_vectors[i].mv_irq = rle->start;
1874 cfg->msix.msix_table[i].mte_vector = i + 1;
1877 /* Update control register to enable MSI-X. */
1878 cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
1879 pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL,
1880 cfg->msix.msix_ctrl, 2);
1882 /* Update counts of alloc'd messages. */
1883 cfg->msix.msix_alloc = actual;
1884 cfg->msix.msix_table_len = actual;
1890 * By default, pci_alloc_msix() will assign the allocated IRQ
1891 * resources consecutively to the first N messages in the MSI-X table.
1892 * However, device drivers may want to use different layouts if they
1893 * either receive fewer messages than they asked for, or they wish to
1894 * populate the MSI-X table sparsely. This method allows the driver
1895 * to specify what layout it wants. It must be called after a
1896 * successful pci_alloc_msix() but before any of the associated
1897 * SYS_RES_IRQ resources are allocated via bus_alloc_resource().
1899 * The 'vectors' array contains 'count' message vectors. The array
1900 * maps directly to the MSI-X table in that index 0 in the array
1901 * specifies the vector for the first message in the MSI-X table, etc.
1902 * The vector value in each array index can either be 0 to indicate
1903 * that no vector should be assigned to a message slot, or it can be a
1904 * number from 1 to N (where N is the count returned from a
1905 * succcessful call to pci_alloc_msix()) to indicate which message
1906 * vector (IRQ) to be used for the corresponding message.
1908 * On successful return, each message with a non-zero vector will have
1909 * an associated SYS_RES_IRQ whose rid is equal to the array index +
1910 * 1. Additionally, if any of the IRQs allocated via the previous
1911 * call to pci_alloc_msix() are not used in the mapping, those IRQs
1912 * will be freed back to the system automatically.
1914 * For example, suppose a driver has a MSI-X table with 6 messages and
1915 * asks for 6 messages, but pci_alloc_msix() only returns a count of
1916 * 3. Call the three vectors allocated by pci_alloc_msix() A, B, and
1917 * C. After the call to pci_alloc_msix(), the device will be setup to
1918 * have an MSI-X table of ABC--- (where - means no vector assigned).
1919 * If the driver then passes a vector array of { 1, 0, 1, 2, 0, 2 },
1920 * then the MSI-X table will look like A-AB-B, and the 'C' vector will
1921 * be freed back to the system. This device will also have valid
1922 * SYS_RES_IRQ rids of 1, 3, 4, and 6.
1924 * In any case, the SYS_RES_IRQ rid X will always map to the message
1925 * at MSI-X table index X - 1 and will only be valid if a vector is
1926 * assigned to that table entry.
1929 pci_remap_msix_method(device_t dev, device_t child, int count,
1930 const u_int *vectors)
1932 struct pci_devinfo *dinfo = device_get_ivars(child);
1933 struct pcicfg_msix *msix = &dinfo->cfg.msix;
1934 struct resource_list_entry *rle;
1935 int i, irq, j, *used;
1938 * Have to have at least one message in the table but the
1939 * table can't be bigger than the actual MSI-X table in the
1942 if (count == 0 || count > msix->msix_msgnum)
1945 /* Sanity check the vectors. */
1946 for (i = 0; i < count; i++)
1947 if (vectors[i] > msix->msix_alloc)
1951 * Make sure there aren't any holes in the vectors to be used.
1952 * It's a big pain to support it, and it doesn't really make
1953 * sense anyway. Also, at least one vector must be used.
1955 used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK |
1957 for (i = 0; i < count; i++)
1958 if (vectors[i] != 0)
1959 used[vectors[i] - 1] = 1;
1960 for (i = 0; i < msix->msix_alloc - 1; i++)
1961 if (used[i] == 0 && used[i + 1] == 1) {
1962 free(used, M_DEVBUF);
1966 free(used, M_DEVBUF);
1970 /* Make sure none of the resources are allocated. */
1971 for (i = 0; i < msix->msix_table_len; i++) {
1972 if (msix->msix_table[i].mte_vector == 0)
1974 if (msix->msix_table[i].mte_handlers > 0) {
1975 free(used, M_DEVBUF);
1978 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
1979 KASSERT(rle != NULL, ("missing resource"));
1980 if (rle->res != NULL) {
1981 free(used, M_DEVBUF);
1986 /* Free the existing resource list entries. */
1987 for (i = 0; i < msix->msix_table_len; i++) {
1988 if (msix->msix_table[i].mte_vector == 0)
1990 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
1994 * Build the new virtual table keeping track of which vectors are
1997 free(msix->msix_table, M_DEVBUF);
1998 msix->msix_table = malloc(sizeof(struct msix_table_entry) * count,
1999 M_DEVBUF, M_WAITOK | M_ZERO);
2000 for (i = 0; i < count; i++)
2001 msix->msix_table[i].mte_vector = vectors[i];
2002 msix->msix_table_len = count;
2004 /* Free any unused IRQs and resize the vectors array if necessary. */
2005 j = msix->msix_alloc - 1;
2007 struct msix_vector *vec;
2009 while (used[j] == 0) {
2010 PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2011 msix->msix_vectors[j].mv_irq);
2014 vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF,
2016 bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) *
2018 free(msix->msix_vectors, M_DEVBUF);
2019 msix->msix_vectors = vec;
2020 msix->msix_alloc = j + 1;
2022 free(used, M_DEVBUF);
2024 /* Map the IRQs onto the rids. */
2025 for (i = 0; i < count; i++) {
2026 if (vectors[i] == 0)
2028 irq = msix->msix_vectors[vectors[i] - 1].mv_irq;
2029 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq,
2034 device_printf(child, "Remapped MSI-X IRQs as: ");
2035 for (i = 0; i < count; i++) {
2038 if (vectors[i] == 0)
2042 msix->msix_vectors[vectors[i] - 1].mv_irq);
2051 pci_release_msix(device_t dev, device_t child)
2053 struct pci_devinfo *dinfo = device_get_ivars(child);
2054 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2055 struct resource_list_entry *rle;
2058 /* Do we have any messages to release? */
2059 if (msix->msix_alloc == 0)
2062 /* Make sure none of the resources are allocated. */
2063 for (i = 0; i < msix->msix_table_len; i++) {
2064 if (msix->msix_table[i].mte_vector == 0)
2066 if (msix->msix_table[i].mte_handlers > 0)
2068 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2069 KASSERT(rle != NULL, ("missing resource"));
2070 if (rle->res != NULL)
2074 /* Update control register to disable MSI-X. */
2075 msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE;
2076 pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL,
2077 msix->msix_ctrl, 2);
2079 /* Free the resource list entries. */
2080 for (i = 0; i < msix->msix_table_len; i++) {
2081 if (msix->msix_table[i].mte_vector == 0)
2083 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2085 free(msix->msix_table, M_DEVBUF);
2086 msix->msix_table_len = 0;
2088 /* Release the IRQs. */
2089 for (i = 0; i < msix->msix_alloc; i++)
2090 PCIB_RELEASE_MSIX(device_get_parent(dev), child,
2091 msix->msix_vectors[i].mv_irq);
2092 free(msix->msix_vectors, M_DEVBUF);
2093 msix->msix_alloc = 0;
2098 * Return the max supported MSI-X messages this device supports.
2099 * Basically, assuming the MD code can alloc messages, this function
2100 * should return the maximum value that pci_alloc_msix() can return.
2101 * Thus, it is subject to the tunables, etc.
2104 pci_msix_count_method(device_t dev, device_t child)
2106 struct pci_devinfo *dinfo = device_get_ivars(child);
2107 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2109 if (pci_do_msix && msix->msix_location != 0)
2110 return (msix->msix_msgnum);
2115 pci_msix_pba_bar_method(device_t dev, device_t child)
2117 struct pci_devinfo *dinfo = device_get_ivars(child);
2118 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2120 if (pci_do_msix && msix->msix_location != 0)
2121 return (msix->msix_pba_bar);
2126 pci_msix_table_bar_method(device_t dev, device_t child)
2128 struct pci_devinfo *dinfo = device_get_ivars(child);
2129 struct pcicfg_msix *msix = &dinfo->cfg.msix;
2131 if (pci_do_msix && msix->msix_location != 0)
2132 return (msix->msix_table_bar);
2137 * HyperTransport MSI mapping control
2140 pci_ht_map_msi(device_t dev, uint64_t addr)
2142 struct pci_devinfo *dinfo = device_get_ivars(dev);
2143 struct pcicfg_ht *ht = &dinfo->cfg.ht;
2148 if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) &&
2149 ht->ht_msiaddr >> 20 == addr >> 20) {
2150 /* Enable MSI -> HT mapping. */
2151 ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE;
2152 pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2156 if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) {
2157 /* Disable MSI -> HT mapping. */
2158 ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE;
2159 pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND,
2165 pci_get_max_payload(device_t dev)
2167 struct pci_devinfo *dinfo = device_get_ivars(dev);
2171 cap = dinfo->cfg.pcie.pcie_location;
2174 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2175 val &= PCIEM_CTL_MAX_PAYLOAD;
2177 return (1 << (val + 7));
2181 pci_get_max_read_req(device_t dev)
2183 struct pci_devinfo *dinfo = device_get_ivars(dev);
2187 cap = dinfo->cfg.pcie.pcie_location;
2190 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2191 val &= PCIEM_CTL_MAX_READ_REQUEST;
2193 return (1 << (val + 7));
2197 pci_set_max_read_req(device_t dev, int size)
2199 struct pci_devinfo *dinfo = device_get_ivars(dev);
2203 cap = dinfo->cfg.pcie.pcie_location;
2210 size = (1 << (fls(size) - 1));
2211 val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
2212 val &= ~PCIEM_CTL_MAX_READ_REQUEST;
2213 val |= (fls(size) - 8) << 12;
2214 pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2);
2219 pcie_read_config(device_t dev, int reg, int width)
2221 struct pci_devinfo *dinfo = device_get_ivars(dev);
2224 cap = dinfo->cfg.pcie.pcie_location;
2228 return (0xffffffff);
2231 return (pci_read_config(dev, cap + reg, width));
2235 pcie_write_config(device_t dev, int reg, uint32_t value, int width)
2237 struct pci_devinfo *dinfo = device_get_ivars(dev);
2240 cap = dinfo->cfg.pcie.pcie_location;
2243 pci_write_config(dev, cap + reg, value, width);
2247 * Adjusts a PCI-e capability register by clearing the bits in mask
2248 * and setting the bits in (value & mask). Bits not set in mask are
2251 * Returns the old value on success or all ones on failure.
2254 pcie_adjust_config(device_t dev, int reg, uint32_t mask, uint32_t value,
2257 struct pci_devinfo *dinfo = device_get_ivars(dev);
2261 cap = dinfo->cfg.pcie.pcie_location;
2265 return (0xffffffff);
2268 old = pci_read_config(dev, cap + reg, width);
2270 new |= (value & mask);
2271 pci_write_config(dev, cap + reg, new, width);
2276 * Support for MSI message signalled interrupts.
2279 pci_enable_msi_method(device_t dev, device_t child, uint64_t address,
2282 struct pci_devinfo *dinfo = device_get_ivars(child);
2283 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2285 /* Write data and address values. */
2286 pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR,
2287 address & 0xffffffff, 4);
2288 if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2289 pci_write_config(child, msi->msi_location + PCIR_MSI_ADDR_HIGH,
2291 pci_write_config(child, msi->msi_location + PCIR_MSI_DATA_64BIT,
2294 pci_write_config(child, msi->msi_location + PCIR_MSI_DATA, data,
2297 /* Enable MSI in the control register. */
2298 msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE;
2299 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2302 /* Enable MSI -> HT mapping. */
2303 pci_ht_map_msi(child, address);
2307 pci_disable_msi_method(device_t dev, device_t child)
2309 struct pci_devinfo *dinfo = device_get_ivars(child);
2310 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2312 /* Disable MSI -> HT mapping. */
2313 pci_ht_map_msi(child, 0);
2315 /* Disable MSI in the control register. */
2316 msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE;
2317 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2322 * Restore MSI registers during resume. If MSI is enabled then
2323 * restore the data and address registers in addition to the control
2327 pci_resume_msi(device_t dev)
2329 struct pci_devinfo *dinfo = device_get_ivars(dev);
2330 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2334 if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) {
2335 address = msi->msi_addr;
2336 data = msi->msi_data;
2337 pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR,
2338 address & 0xffffffff, 4);
2339 if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) {
2340 pci_write_config(dev, msi->msi_location +
2341 PCIR_MSI_ADDR_HIGH, address >> 32, 4);
2342 pci_write_config(dev, msi->msi_location +
2343 PCIR_MSI_DATA_64BIT, data, 2);
2345 pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA,
2348 pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl,
2353 pci_remap_intr_method(device_t bus, device_t dev, u_int irq)
2355 struct pci_devinfo *dinfo = device_get_ivars(dev);
2356 pcicfgregs *cfg = &dinfo->cfg;
2357 struct resource_list_entry *rle;
2358 struct msix_table_entry *mte;
2359 struct msix_vector *mv;
2365 * Handle MSI first. We try to find this IRQ among our list
2366 * of MSI IRQs. If we find it, we request updated address and
2367 * data registers and apply the results.
2369 if (cfg->msi.msi_alloc > 0) {
2371 /* If we don't have any active handlers, nothing to do. */
2372 if (cfg->msi.msi_handlers == 0)
2374 for (i = 0; i < cfg->msi.msi_alloc; i++) {
2375 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ,
2377 if (rle->start == irq) {
2378 error = PCIB_MAP_MSI(device_get_parent(bus),
2379 dev, irq, &addr, &data);
2382 pci_disable_msi(dev);
2383 dinfo->cfg.msi.msi_addr = addr;
2384 dinfo->cfg.msi.msi_data = data;
2385 pci_enable_msi(dev, addr, data);
2393 * For MSI-X, we check to see if we have this IRQ. If we do,
2394 * we request the updated mapping info. If that works, we go
2395 * through all the slots that use this IRQ and update them.
2397 if (cfg->msix.msix_alloc > 0) {
2398 for (i = 0; i < cfg->msix.msix_alloc; i++) {
2399 mv = &cfg->msix.msix_vectors[i];
2400 if (mv->mv_irq == irq) {
2401 error = PCIB_MAP_MSI(device_get_parent(bus),
2402 dev, irq, &addr, &data);
2405 mv->mv_address = addr;
2407 for (j = 0; j < cfg->msix.msix_table_len; j++) {
2408 mte = &cfg->msix.msix_table[j];
2409 if (mte->mte_vector != i + 1)
2411 if (mte->mte_handlers == 0)
2413 pci_mask_msix(dev, j);
2414 pci_enable_msix(dev, j, addr, data);
2415 pci_unmask_msix(dev, j);
2426 * Returns true if the specified device is blacklisted because MSI
2430 pci_msi_device_blacklisted(device_t dev)
2433 if (!pci_honor_msi_blacklist)
2436 return (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSI));
2440 * Determine if MSI is blacklisted globally on this system. Currently,
2441 * we just check for blacklisted chipsets as represented by the
2442 * host-PCI bridge at device 0:0:0. In the future, it may become
2443 * necessary to check other system attributes, such as the kenv values
2444 * that give the motherboard manufacturer and model number.
2447 pci_msi_blacklisted(void)
2451 if (!pci_honor_msi_blacklist)
2454 /* Blacklist all non-PCI-express and non-PCI-X chipsets. */
2455 if (!(pcie_chipset || pcix_chipset)) {
2456 if (vm_guest != VM_GUEST_NO) {
2458 * Whitelist older chipsets in virtual
2459 * machines known to support MSI.
2461 dev = pci_find_bsf(0, 0, 0);
2463 return (!pci_has_quirk(pci_get_devid(dev),
2464 PCI_QUIRK_ENABLE_MSI_VM));
2469 dev = pci_find_bsf(0, 0, 0);
2471 return (pci_msi_device_blacklisted(dev));
2476 * Returns true if the specified device is blacklisted because MSI-X
2477 * doesn't work. Note that this assumes that if MSI doesn't work,
2478 * MSI-X doesn't either.
2481 pci_msix_device_blacklisted(device_t dev)
2484 if (!pci_honor_msi_blacklist)
2487 if (pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_DISABLE_MSIX))
2490 return (pci_msi_device_blacklisted(dev));
2494 * Determine if MSI-X is blacklisted globally on this system. If MSI
2495 * is blacklisted, assume that MSI-X is as well. Check for additional
2496 * chipsets where MSI works but MSI-X does not.
2499 pci_msix_blacklisted(void)
2503 if (!pci_honor_msi_blacklist)
2506 dev = pci_find_bsf(0, 0, 0);
2507 if (dev != NULL && pci_has_quirk(pci_get_devid(dev),
2508 PCI_QUIRK_DISABLE_MSIX))
2511 return (pci_msi_blacklisted());
2515 * Attempt to allocate *count MSI messages. The actual number allocated is
2516 * returned in *count. After this function returns, each message will be
2517 * available to the driver as SYS_RES_IRQ resources starting at a rid 1.
2520 pci_alloc_msi_method(device_t dev, device_t child, int *count)
2522 struct pci_devinfo *dinfo = device_get_ivars(child);
2523 pcicfgregs *cfg = &dinfo->cfg;
2524 struct resource_list_entry *rle;
2525 int actual, error, i, irqs[32];
2528 /* Don't let count == 0 get us into trouble. */
2532 /* If rid 0 is allocated, then fail. */
2533 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
2534 if (rle != NULL && rle->res != NULL)
2537 /* Already have allocated messages? */
2538 if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0)
2541 /* If MSI is blacklisted for this system, fail. */
2542 if (pci_msi_blacklisted())
2545 /* MSI capability present? */
2546 if (cfg->msi.msi_location == 0 || !pci_do_msi)
2550 device_printf(child,
2551 "attempting to allocate %d MSI vectors (%d supported)\n",
2552 *count, cfg->msi.msi_msgnum);
2554 /* Don't ask for more than the device supports. */
2555 actual = min(*count, cfg->msi.msi_msgnum);
2557 /* Don't ask for more than 32 messages. */
2558 actual = min(actual, 32);
2560 /* MSI requires power of 2 number of messages. */
2561 if (!powerof2(actual))
2565 /* Try to allocate N messages. */
2566 error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual,
2578 * We now have N actual messages mapped onto SYS_RES_IRQ
2579 * resources in the irqs[] array, so add new resources
2580 * starting at rid 1.
2582 for (i = 0; i < actual; i++)
2583 resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1,
2584 irqs[i], irqs[i], 1);
2588 device_printf(child, "using IRQ %d for MSI\n", irqs[0]);
2593 * Be fancy and try to print contiguous runs
2594 * of IRQ values as ranges. 'run' is true if
2595 * we are in a range.
2597 device_printf(child, "using IRQs %d", irqs[0]);
2599 for (i = 1; i < actual; i++) {
2601 /* Still in a run? */
2602 if (irqs[i] == irqs[i - 1] + 1) {
2607 /* Finish previous range. */
2609 printf("-%d", irqs[i - 1]);
2613 /* Start new range. */
2614 printf(",%d", irqs[i]);
2617 /* Unfinished range? */
2619 printf("-%d", irqs[actual - 1]);
2620 printf(" for MSI\n");
2624 /* Update control register with actual count. */
2625 ctrl = cfg->msi.msi_ctrl;
2626 ctrl &= ~PCIM_MSICTRL_MME_MASK;
2627 ctrl |= (ffs(actual) - 1) << 4;
2628 cfg->msi.msi_ctrl = ctrl;
2629 pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2);
2631 /* Update counts of alloc'd messages. */
2632 cfg->msi.msi_alloc = actual;
2633 cfg->msi.msi_handlers = 0;
2638 /* Release the MSI messages associated with this device. */
2640 pci_release_msi_method(device_t dev, device_t child)
2642 struct pci_devinfo *dinfo = device_get_ivars(child);
2643 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2644 struct resource_list_entry *rle;
2645 int error, i, irqs[32];
2647 /* Try MSI-X first. */
2648 error = pci_release_msix(dev, child);
2649 if (error != ENODEV)
2652 /* Do we have any messages to release? */
2653 if (msi->msi_alloc == 0)
2655 KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages"));
2657 /* Make sure none of the resources are allocated. */
2658 if (msi->msi_handlers > 0)
2660 for (i = 0; i < msi->msi_alloc; i++) {
2661 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1);
2662 KASSERT(rle != NULL, ("missing MSI resource"));
2663 if (rle->res != NULL)
2665 irqs[i] = rle->start;
2668 /* Update control register with 0 count. */
2669 KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE),
2670 ("%s: MSI still enabled", __func__));
2671 msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK;
2672 pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL,
2675 /* Release the messages. */
2676 PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs);
2677 for (i = 0; i < msi->msi_alloc; i++)
2678 resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1);
2680 /* Update alloc count. */
2688 * Return the max supported MSI messages this device supports.
2689 * Basically, assuming the MD code can alloc messages, this function
2690 * should return the maximum value that pci_alloc_msi() can return.
2691 * Thus, it is subject to the tunables, etc.
2694 pci_msi_count_method(device_t dev, device_t child)
2696 struct pci_devinfo *dinfo = device_get_ivars(child);
2697 struct pcicfg_msi *msi = &dinfo->cfg.msi;
2699 if (pci_do_msi && msi->msi_location != 0)
2700 return (msi->msi_msgnum);
2704 /* free pcicfgregs structure and all depending data structures */
2707 pci_freecfg(struct pci_devinfo *dinfo)
2709 struct devlist *devlist_head;
2710 struct pci_map *pm, *next;
2713 devlist_head = &pci_devq;
2715 if (dinfo->cfg.vpd.vpd_reg) {
2716 free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF);
2717 for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++)
2718 free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF);
2719 free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF);
2720 for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++)
2721 free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF);
2722 free(dinfo->cfg.vpd.vpd_w, M_DEVBUF);
2724 STAILQ_FOREACH_SAFE(pm, &dinfo->cfg.maps, pm_link, next) {
2727 STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links);
2728 free(dinfo, M_DEVBUF);
2730 /* increment the generation count */
2733 /* we're losing one device */
2739 * PCI power manangement
2742 pci_set_powerstate_method(device_t dev, device_t child, int state)
2744 struct pci_devinfo *dinfo = device_get_ivars(child);
2745 pcicfgregs *cfg = &dinfo->cfg;
2747 int oldstate, highest, delay;
2749 if (cfg->pp.pp_cap == 0)
2750 return (EOPNOTSUPP);
2753 * Optimize a no state change request away. While it would be OK to
2754 * write to the hardware in theory, some devices have shown odd
2755 * behavior when going from D3 -> D3.
2757 oldstate = pci_get_powerstate(child);
2758 if (oldstate == state)
2762 * The PCI power management specification states that after a state
2763 * transition between PCI power states, system software must
2764 * guarantee a minimal delay before the function accesses the device.
2765 * Compute the worst case delay that we need to guarantee before we
2766 * access the device. Many devices will be responsive much more
2767 * quickly than this delay, but there are some that don't respond
2768 * instantly to state changes. Transitions to/from D3 state require
2769 * 10ms, while D2 requires 200us, and D0/1 require none. The delay
2770 * is done below with DELAY rather than a sleeper function because
2771 * this function can be called from contexts where we cannot sleep.
2773 highest = (oldstate > state) ? oldstate : state;
2774 if (highest == PCI_POWERSTATE_D3)
2776 else if (highest == PCI_POWERSTATE_D2)
2780 status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2)
2781 & ~PCIM_PSTAT_DMASK;
2783 case PCI_POWERSTATE_D0:
2784 status |= PCIM_PSTAT_D0;
2786 case PCI_POWERSTATE_D1:
2787 if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0)
2788 return (EOPNOTSUPP);
2789 status |= PCIM_PSTAT_D1;
2791 case PCI_POWERSTATE_D2:
2792 if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0)
2793 return (EOPNOTSUPP);
2794 status |= PCIM_PSTAT_D2;
2796 case PCI_POWERSTATE_D3:
2797 status |= PCIM_PSTAT_D3;
2804 pci_printf(cfg, "Transition from D%d to D%d\n", oldstate,
2807 PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2);
2814 pci_get_powerstate_method(device_t dev, device_t child)
2816 struct pci_devinfo *dinfo = device_get_ivars(child);
2817 pcicfgregs *cfg = &dinfo->cfg;
2821 if (cfg->pp.pp_cap != 0) {
2822 status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2);
2823 switch (status & PCIM_PSTAT_DMASK) {
2825 result = PCI_POWERSTATE_D0;
2828 result = PCI_POWERSTATE_D1;
2831 result = PCI_POWERSTATE_D2;
2834 result = PCI_POWERSTATE_D3;
2837 result = PCI_POWERSTATE_UNKNOWN;
2841 /* No support, device is always at D0 */
2842 result = PCI_POWERSTATE_D0;
2848 * Some convenience functions for PCI device drivers.
2851 static __inline void
2852 pci_set_command_bit(device_t dev, device_t child, uint16_t bit)
2856 command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2858 PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2861 static __inline void
2862 pci_clear_command_bit(device_t dev, device_t child, uint16_t bit)
2866 command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2);
2868 PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2);
2872 pci_enable_busmaster_method(device_t dev, device_t child)
2874 pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2879 pci_disable_busmaster_method(device_t dev, device_t child)
2881 pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN);
2886 pci_enable_io_method(device_t dev, device_t child, int space)
2891 case SYS_RES_IOPORT:
2892 bit = PCIM_CMD_PORTEN;
2894 case SYS_RES_MEMORY:
2895 bit = PCIM_CMD_MEMEN;
2900 pci_set_command_bit(dev, child, bit);
2905 pci_disable_io_method(device_t dev, device_t child, int space)
2910 case SYS_RES_IOPORT:
2911 bit = PCIM_CMD_PORTEN;
2913 case SYS_RES_MEMORY:
2914 bit = PCIM_CMD_MEMEN;
2919 pci_clear_command_bit(dev, child, bit);
2924 * New style pci driver. Parent device is either a pci-host-bridge or a
2925 * pci-pci-bridge. Both kinds are represented by instances of pcib.
2929 pci_print_verbose(struct pci_devinfo *dinfo)
2933 pcicfgregs *cfg = &dinfo->cfg;
2935 printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n",
2936 cfg->vendor, cfg->device, cfg->revid);
2937 printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n",
2938 cfg->domain, cfg->bus, cfg->slot, cfg->func);
2939 printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n",
2940 cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype,
2942 printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n",
2943 cfg->cmdreg, cfg->statreg, cfg->cachelnsz);
2944 printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n",
2945 cfg->lattimer, cfg->lattimer * 30, cfg->mingnt,
2946 cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250);
2947 if (cfg->intpin > 0)
2948 printf("\tintpin=%c, irq=%d\n",
2949 cfg->intpin +'a' -1, cfg->intline);
2950 if (cfg->pp.pp_cap) {
2953 status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2);
2954 printf("\tpowerspec %d supports D0%s%s D3 current D%d\n",
2955 cfg->pp.pp_cap & PCIM_PCAP_SPEC,
2956 cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "",
2957 cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "",
2958 status & PCIM_PSTAT_DMASK);
2960 if (cfg->msi.msi_location) {
2963 ctrl = cfg->msi.msi_ctrl;
2964 printf("\tMSI supports %d message%s%s%s\n",
2965 cfg->msi.msi_msgnum,
2966 (cfg->msi.msi_msgnum == 1) ? "" : "s",
2967 (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "",
2968 (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":"");
2970 if (cfg->msix.msix_location) {
2971 printf("\tMSI-X supports %d message%s ",
2972 cfg->msix.msix_msgnum,
2973 (cfg->msix.msix_msgnum == 1) ? "" : "s");
2974 if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar)
2975 printf("in map 0x%x\n",
2976 cfg->msix.msix_table_bar);
2978 printf("in maps 0x%x and 0x%x\n",
2979 cfg->msix.msix_table_bar,
2980 cfg->msix.msix_pba_bar);
2986 pci_porten(device_t dev)
2988 return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0;
2992 pci_memen(device_t dev)
2994 return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0;
2998 pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp,
3001 struct pci_devinfo *dinfo;
3002 pci_addr_t map, testval;
3007 * The device ROM BAR is special. It is always a 32-bit
3008 * memory BAR. Bit 0 is special and should not be set when
3011 dinfo = device_get_ivars(dev);
3012 if (PCIR_IS_BIOS(&dinfo->cfg, reg)) {
3013 map = pci_read_config(dev, reg, 4);
3014 pci_write_config(dev, reg, 0xfffffffe, 4);
3015 testval = pci_read_config(dev, reg, 4);
3016 pci_write_config(dev, reg, map, 4);
3018 *testvalp = testval;
3024 map = pci_read_config(dev, reg, 4);
3025 ln2range = pci_maprange(map);
3027 map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3030 * Disable decoding via the command register before
3031 * determining the BAR's length since we will be placing it in
3034 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3035 pci_write_config(dev, PCIR_COMMAND,
3036 cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
3039 * Determine the BAR's length by writing all 1's. The bottom
3040 * log_2(size) bits of the BAR will stick as 0 when we read
3043 * NB: according to the PCI Local Bus Specification, rev. 3.0:
3044 * "Software writes 0FFFFFFFFh to both registers, reads them back,
3045 * and combines the result into a 64-bit value." (section 6.2.5.1)
3047 * Writes to both registers must be performed before attempting to
3048 * read back the size value.
3051 pci_write_config(dev, reg, 0xffffffff, 4);
3052 if (ln2range == 64) {
3053 pci_write_config(dev, reg + 4, 0xffffffff, 4);
3054 testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32;
3056 testval |= pci_read_config(dev, reg, 4);
3059 * Restore the original value of the BAR. We may have reprogrammed
3060 * the BAR of the low-level console device and when booting verbose,
3061 * we need the console device addressable.
3063 pci_write_config(dev, reg, map, 4);
3065 pci_write_config(dev, reg + 4, map >> 32, 4);
3066 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3069 *testvalp = testval;
3071 *bar64 = (ln2range == 64);
3075 pci_write_bar(device_t dev, struct pci_map *pm, pci_addr_t base)
3077 struct pci_devinfo *dinfo;
3080 /* The device ROM BAR is always a 32-bit memory BAR. */
3081 dinfo = device_get_ivars(dev);
3082 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3085 ln2range = pci_maprange(pm->pm_value);
3086 pci_write_config(dev, pm->pm_reg, base, 4);
3088 pci_write_config(dev, pm->pm_reg + 4, base >> 32, 4);
3089 pm->pm_value = pci_read_config(dev, pm->pm_reg, 4);
3091 pm->pm_value |= (pci_addr_t)pci_read_config(dev,
3092 pm->pm_reg + 4, 4) << 32;
3096 pci_find_bar(device_t dev, int reg)
3098 struct pci_devinfo *dinfo;
3101 dinfo = device_get_ivars(dev);
3102 STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3103 if (pm->pm_reg == reg)
3110 pci_bar_enabled(device_t dev, struct pci_map *pm)
3112 struct pci_devinfo *dinfo;
3115 dinfo = device_get_ivars(dev);
3116 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) &&
3117 !(pm->pm_value & PCIM_BIOS_ENABLE))
3119 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3120 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg) || PCI_BAR_MEM(pm->pm_value))
3121 return ((cmd & PCIM_CMD_MEMEN) != 0);
3123 return ((cmd & PCIM_CMD_PORTEN) != 0);
3127 pci_add_bar(device_t dev, int reg, pci_addr_t value, pci_addr_t size)
3129 struct pci_devinfo *dinfo;
3130 struct pci_map *pm, *prev;
3132 dinfo = device_get_ivars(dev);
3133 pm = malloc(sizeof(*pm), M_DEVBUF, M_WAITOK | M_ZERO);
3135 pm->pm_value = value;
3137 STAILQ_FOREACH(prev, &dinfo->cfg.maps, pm_link) {
3138 KASSERT(prev->pm_reg != pm->pm_reg, ("duplicate map %02x",
3140 if (STAILQ_NEXT(prev, pm_link) == NULL ||
3141 STAILQ_NEXT(prev, pm_link)->pm_reg > pm->pm_reg)
3145 STAILQ_INSERT_AFTER(&dinfo->cfg.maps, prev, pm, pm_link);
3147 STAILQ_INSERT_TAIL(&dinfo->cfg.maps, pm, pm_link);
3152 pci_restore_bars(device_t dev)
3154 struct pci_devinfo *dinfo;
3158 dinfo = device_get_ivars(dev);
3159 STAILQ_FOREACH(pm, &dinfo->cfg.maps, pm_link) {
3160 if (PCIR_IS_BIOS(&dinfo->cfg, pm->pm_reg))
3163 ln2range = pci_maprange(pm->pm_value);
3164 pci_write_config(dev, pm->pm_reg, pm->pm_value, 4);
3166 pci_write_config(dev, pm->pm_reg + 4,
3167 pm->pm_value >> 32, 4);
3172 * Add a resource based on a pci map register. Return 1 if the map
3173 * register is a 32bit map register or 2 if it is a 64bit register.
3176 pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl,
3177 int force, int prefetch)
3180 pci_addr_t base, map, testval;
3181 pci_addr_t start, end, count;
3182 int barlen, basezero, flags, maprange, mapsize, type;
3184 struct resource *res;
3187 * The BAR may already exist if the device is a CardBus card
3188 * whose CIS is stored in this BAR.
3190 pm = pci_find_bar(dev, reg);
3192 maprange = pci_maprange(pm->pm_value);
3193 barlen = maprange == 64 ? 2 : 1;
3197 pci_read_bar(dev, reg, &map, &testval, NULL);
3198 if (PCI_BAR_MEM(map)) {
3199 type = SYS_RES_MEMORY;
3200 if (map & PCIM_BAR_MEM_PREFETCH)
3203 type = SYS_RES_IOPORT;
3204 mapsize = pci_mapsize(testval);
3205 base = pci_mapbase(map);
3206 #ifdef __PCI_BAR_ZERO_VALID
3209 basezero = base == 0;
3211 maprange = pci_maprange(map);
3212 barlen = maprange == 64 ? 2 : 1;
3215 * For I/O registers, if bottom bit is set, and the next bit up
3216 * isn't clear, we know we have a BAR that doesn't conform to the
3217 * spec, so ignore it. Also, sanity check the size of the data
3218 * areas to the type of memory involved. Memory must be at least
3219 * 16 bytes in size, while I/O ranges must be at least 4.
3221 if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0)
3223 if ((type == SYS_RES_MEMORY && mapsize < 4) ||
3224 (type == SYS_RES_IOPORT && mapsize < 2))
3227 /* Save a record of this BAR. */
3228 pm = pci_add_bar(dev, reg, map, mapsize);
3230 printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d",
3231 reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize);
3232 if (type == SYS_RES_IOPORT && !pci_porten(dev))
3233 printf(", port disabled\n");
3234 else if (type == SYS_RES_MEMORY && !pci_memen(dev))
3235 printf(", memory disabled\n");
3237 printf(", enabled\n");
3241 * If base is 0, then we have problems if this architecture does
3242 * not allow that. It is best to ignore such entries for the
3243 * moment. These will be allocated later if the driver specifically
3244 * requests them. However, some removable buses look better when
3245 * all resources are allocated, so allow '0' to be overriden.
3247 * Similarly treat maps whose values is the same as the test value
3248 * read back. These maps have had all f's written to them by the
3249 * BIOS in an attempt to disable the resources.
3251 if (!force && (basezero || map == testval))
3253 if ((u_long)base != base) {
3255 "pci%d:%d:%d:%d bar %#x too many address bits",
3256 pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev),
3257 pci_get_function(dev), reg);
3262 * This code theoretically does the right thing, but has
3263 * undesirable side effects in some cases where peripherals
3264 * respond oddly to having these bits enabled. Let the user
3265 * be able to turn them off (since pci_enable_io_modes is 1 by
3268 if (pci_enable_io_modes) {
3269 /* Turn on resources that have been left off by a lazy BIOS */
3270 if (type == SYS_RES_IOPORT && !pci_porten(dev)) {
3271 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3272 cmd |= PCIM_CMD_PORTEN;
3273 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3275 if (type == SYS_RES_MEMORY && !pci_memen(dev)) {
3276 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
3277 cmd |= PCIM_CMD_MEMEN;
3278 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
3281 if (type == SYS_RES_IOPORT && !pci_porten(dev))
3283 if (type == SYS_RES_MEMORY && !pci_memen(dev))
3287 count = (pci_addr_t)1 << mapsize;
3288 flags = RF_ALIGNMENT_LOG2(mapsize);
3290 flags |= RF_PREFETCHABLE;
3291 if (basezero || base == pci_mapbase(testval) || pci_clear_bars) {
3292 start = 0; /* Let the parent decide. */
3296 end = base + count - 1;
3298 resource_list_add(rl, type, reg, start, end, count);
3301 * Try to allocate the resource for this BAR from our parent
3302 * so that this resource range is already reserved. The
3303 * driver for this device will later inherit this resource in
3304 * pci_alloc_resource().
3306 res = resource_list_reserve(rl, bus, dev, type, ®, start, end, count,
3308 if ((pci_do_realloc_bars
3309 || pci_has_quirk(pci_get_devid(dev), PCI_QUIRK_REALLOC_BAR))
3310 && res == NULL && (start != 0 || end != ~0)) {
3312 * If the allocation fails, try to allocate a resource for
3313 * this BAR using any available range. The firmware felt
3314 * it was important enough to assign a resource, so don't
3315 * disable decoding if we can help it.
3317 resource_list_delete(rl, type, reg);
3318 resource_list_add(rl, type, reg, 0, ~0, count);
3319 res = resource_list_reserve(rl, bus, dev, type, ®, 0, ~0,
3324 * If the allocation fails, delete the resource list entry
3325 * and disable decoding for this device.
3327 * If the driver requests this resource in the future,
3328 * pci_reserve_map() will try to allocate a fresh
3331 resource_list_delete(rl, type, reg);
3332 pci_disable_io(dev, type);
3335 "pci%d:%d:%d:%d bar %#x failed to allocate\n",
3336 pci_get_domain(dev), pci_get_bus(dev),
3337 pci_get_slot(dev), pci_get_function(dev), reg);
3339 start = rman_get_start(res);
3340 pci_write_bar(dev, pm, start);
3346 * For ATA devices we need to decide early what addressing mode to use.
3347 * Legacy demands that the primary and secondary ATA ports sits on the
3348 * same addresses that old ISA hardware did. This dictates that we use
3349 * those addresses and ignore the BAR's if we cannot set PCI native
3353 pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force,
3354 uint32_t prefetchmask)
3356 int rid, type, progif;
3358 /* if this device supports PCI native addressing use it */
3359 progif = pci_read_config(dev, PCIR_PROGIF, 1);
3360 if ((progif & 0x8a) == 0x8a) {
3361 if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) &&
3362 pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) {
3363 printf("Trying ATA native PCI addressing mode\n");
3364 pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1);
3368 progif = pci_read_config(dev, PCIR_PROGIF, 1);
3369 type = SYS_RES_IOPORT;
3370 if (progif & PCIP_STORAGE_IDE_MODEPRIM) {
3371 pci_add_map(bus, dev, PCIR_BAR(0), rl, force,
3372 prefetchmask & (1 << 0));
3373 pci_add_map(bus, dev, PCIR_BAR(1), rl, force,
3374 prefetchmask & (1 << 1));
3377 resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8);
3378 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x1f0,
3381 resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1);
3382 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x3f6,
3385 if (progif & PCIP_STORAGE_IDE_MODESEC) {
3386 pci_add_map(bus, dev, PCIR_BAR(2), rl, force,
3387 prefetchmask & (1 << 2));
3388 pci_add_map(bus, dev, PCIR_BAR(3), rl, force,
3389 prefetchmask & (1 << 3));
3392 resource_list_add(rl, type, rid, 0x170, 0x177, 8);
3393 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x170,
3396 resource_list_add(rl, type, rid, 0x376, 0x376, 1);
3397 (void)resource_list_reserve(rl, bus, dev, type, &rid, 0x376,
3400 pci_add_map(bus, dev, PCIR_BAR(4), rl, force,
3401 prefetchmask & (1 << 4));
3402 pci_add_map(bus, dev, PCIR_BAR(5), rl, force,
3403 prefetchmask & (1 << 5));
3407 pci_assign_interrupt(device_t bus, device_t dev, int force_route)
3409 struct pci_devinfo *dinfo = device_get_ivars(dev);
3410 pcicfgregs *cfg = &dinfo->cfg;
3411 char tunable_name[64];
3414 /* Has to have an intpin to have an interrupt. */
3415 if (cfg->intpin == 0)
3418 /* Let the user override the IRQ with a tunable. */
3419 irq = PCI_INVALID_IRQ;
3420 snprintf(tunable_name, sizeof(tunable_name),
3421 "hw.pci%d.%d.%d.INT%c.irq",
3422 cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1);
3423 if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0))
3424 irq = PCI_INVALID_IRQ;
3427 * If we didn't get an IRQ via the tunable, then we either use the
3428 * IRQ value in the intline register or we ask the bus to route an
3429 * interrupt for us. If force_route is true, then we only use the
3430 * value in the intline register if the bus was unable to assign an
3433 if (!PCI_INTERRUPT_VALID(irq)) {
3434 if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route)
3435 irq = PCI_ASSIGN_INTERRUPT(bus, dev);
3436 if (!PCI_INTERRUPT_VALID(irq))
3440 /* If after all that we don't have an IRQ, just bail. */
3441 if (!PCI_INTERRUPT_VALID(irq))
3444 /* Update the config register if it changed. */
3445 if (irq != cfg->intline) {
3447 pci_write_config(dev, PCIR_INTLINE, irq, 1);
3450 /* Add this IRQ as rid 0 interrupt resource. */
3451 resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1);
3454 /* Perform early OHCI takeover from SMM. */
3456 ohci_early_takeover(device_t self)
3458 struct resource *res;
3464 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3468 ctl = bus_read_4(res, OHCI_CONTROL);
3469 if (ctl & OHCI_IR) {
3471 printf("ohci early: "
3472 "SMM active, request owner change\n");
3473 bus_write_4(res, OHCI_COMMAND_STATUS, OHCI_OCR);
3474 for (i = 0; (i < 100) && (ctl & OHCI_IR); i++) {
3476 ctl = bus_read_4(res, OHCI_CONTROL);
3478 if (ctl & OHCI_IR) {
3480 printf("ohci early: "
3481 "SMM does not respond, resetting\n");
3482 bus_write_4(res, OHCI_CONTROL, OHCI_HCFS_RESET);
3484 /* Disable interrupts */
3485 bus_write_4(res, OHCI_INTERRUPT_DISABLE, OHCI_ALL_INTRS);
3488 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3491 /* Perform early UHCI takeover from SMM. */
3493 uhci_early_takeover(device_t self)
3495 struct resource *res;
3499 * Set the PIRQD enable bit and switch off all the others. We don't
3500 * want legacy support to interfere with us XXX Does this also mean
3501 * that the BIOS won't touch the keyboard anymore if it is connected
3502 * to the ports of the root hub?
3504 pci_write_config(self, PCI_LEGSUP, PCI_LEGSUP_USBPIRQDEN, 2);
3506 /* Disable interrupts */
3507 rid = PCI_UHCI_BASE_REG;
3508 res = bus_alloc_resource_any(self, SYS_RES_IOPORT, &rid, RF_ACTIVE);
3510 bus_write_2(res, UHCI_INTR, 0);
3511 bus_release_resource(self, SYS_RES_IOPORT, rid, res);
3515 /* Perform early EHCI takeover from SMM. */
3517 ehci_early_takeover(device_t self)
3519 struct resource *res;
3529 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3533 cparams = bus_read_4(res, EHCI_HCCPARAMS);
3535 /* Synchronise with the BIOS if it owns the controller. */
3536 for (eecp = EHCI_HCC_EECP(cparams); eecp != 0;
3537 eecp = EHCI_EECP_NEXT(eec)) {
3538 eec = pci_read_config(self, eecp, 4);
3539 if (EHCI_EECP_ID(eec) != EHCI_EC_LEGSUP) {
3542 bios_sem = pci_read_config(self, eecp +
3543 EHCI_LEGSUP_BIOS_SEM, 1);
3544 if (bios_sem == 0) {
3548 printf("ehci early: "
3549 "SMM active, request owner change\n");
3551 pci_write_config(self, eecp + EHCI_LEGSUP_OS_SEM, 1, 1);
3553 for (i = 0; (i < 100) && (bios_sem != 0); i++) {
3555 bios_sem = pci_read_config(self, eecp +
3556 EHCI_LEGSUP_BIOS_SEM, 1);
3559 if (bios_sem != 0) {
3561 printf("ehci early: "
3562 "SMM does not respond\n");
3564 /* Disable interrupts */
3565 offs = EHCI_CAPLENGTH(bus_read_4(res, EHCI_CAPLEN_HCIVERSION));
3566 bus_write_4(res, offs + EHCI_USBINTR, 0);
3568 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3571 /* Perform early XHCI takeover from SMM. */
3573 xhci_early_takeover(device_t self)
3575 struct resource *res;
3585 res = bus_alloc_resource_any(self, SYS_RES_MEMORY, &rid, RF_ACTIVE);
3589 cparams = bus_read_4(res, XHCI_HCSPARAMS0);
3593 /* Synchronise with the BIOS if it owns the controller. */
3594 for (eecp = XHCI_HCS0_XECP(cparams) << 2; eecp != 0 && XHCI_XECP_NEXT(eec);
3595 eecp += XHCI_XECP_NEXT(eec) << 2) {
3596 eec = bus_read_4(res, eecp);
3598 if (XHCI_XECP_ID(eec) != XHCI_ID_USB_LEGACY)
3601 bios_sem = bus_read_1(res, eecp + XHCI_XECP_BIOS_SEM);
3606 printf("xhci early: "
3607 "SMM active, request owner change\n");
3609 bus_write_1(res, eecp + XHCI_XECP_OS_SEM, 1);
3611 /* wait a maximum of 5 second */
3613 for (i = 0; (i < 5000) && (bios_sem != 0); i++) {
3615 bios_sem = bus_read_1(res, eecp +
3616 XHCI_XECP_BIOS_SEM);
3619 if (bios_sem != 0) {
3621 printf("xhci early: "
3622 "SMM does not respond\n");
3625 /* Disable interrupts */
3626 offs = bus_read_1(res, XHCI_CAPLENGTH);
3627 bus_write_4(res, offs + XHCI_USBCMD, 0);
3628 bus_read_4(res, offs + XHCI_USBSTS);
3630 bus_release_resource(self, SYS_RES_MEMORY, rid, res);
3633 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
3635 pci_reserve_secbus(device_t bus, device_t dev, pcicfgregs *cfg,
3636 struct resource_list *rl)
3638 struct resource *res;
3640 rman_res_t start, end, count;
3641 int rid, sec_bus, sec_reg, sub_bus, sub_reg, sup_bus;
3643 switch (cfg->hdrtype & PCIM_HDRTYPE) {
3644 case PCIM_HDRTYPE_BRIDGE:
3645 sec_reg = PCIR_SECBUS_1;
3646 sub_reg = PCIR_SUBBUS_1;
3648 case PCIM_HDRTYPE_CARDBUS:
3649 sec_reg = PCIR_SECBUS_2;
3650 sub_reg = PCIR_SUBBUS_2;
3657 * If the existing bus range is valid, attempt to reserve it
3658 * from our parent. If this fails for any reason, clear the
3659 * secbus and subbus registers.
3661 * XXX: Should we reset sub_bus to sec_bus if it is < sec_bus?
3662 * This would at least preserve the existing sec_bus if it is
3665 sec_bus = PCI_READ_CONFIG(bus, dev, sec_reg, 1);
3666 sub_bus = PCI_READ_CONFIG(bus, dev, sub_reg, 1);
3668 /* Quirk handling. */
3669 switch (pci_get_devid(dev)) {
3670 case 0x12258086: /* Intel 82454KX/GX (Orion) */
3671 sup_bus = pci_read_config(dev, 0x41, 1);
3672 if (sup_bus != 0xff) {
3673 sec_bus = sup_bus + 1;
3674 sub_bus = sup_bus + 1;
3675 PCI_WRITE_CONFIG(bus, dev, sec_reg, sec_bus, 1);
3676 PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3681 /* Compaq R3000 BIOS sets wrong subordinate bus number. */
3682 if ((cp = kern_getenv("smbios.planar.maker")) == NULL)
3684 if (strncmp(cp, "Compal", 6) != 0) {
3689 if ((cp = kern_getenv("smbios.planar.product")) == NULL)
3691 if (strncmp(cp, "08A0", 4) != 0) {
3696 if (sub_bus < 0xa) {
3698 PCI_WRITE_CONFIG(bus, dev, sub_reg, sub_bus, 1);
3704 printf("\tsecbus=%d, subbus=%d\n", sec_bus, sub_bus);
3705 if (sec_bus > 0 && sub_bus >= sec_bus) {
3708 count = end - start + 1;
3710 resource_list_add(rl, PCI_RES_BUS, 0, 0, ~0, count);
3713 * If requested, clear secondary bus registers in
3714 * bridge devices to force a complete renumbering
3715 * rather than reserving the existing range. However,
3716 * preserve the existing size.
3718 if (pci_clear_buses)
3722 res = resource_list_reserve(rl, bus, dev, PCI_RES_BUS, &rid,
3723 start, end, count, 0);
3729 "pci%d:%d:%d:%d secbus failed to allocate\n",
3730 pci_get_domain(dev), pci_get_bus(dev),
3731 pci_get_slot(dev), pci_get_function(dev));
3735 PCI_WRITE_CONFIG(bus, dev, sec_reg, 0, 1);
3736 PCI_WRITE_CONFIG(bus, dev, sub_reg, 0, 1);
3739 static struct resource *
3740 pci_alloc_secbus(device_t dev, device_t child, int *rid, rman_res_t start,
3741 rman_res_t end, rman_res_t count, u_int flags)
3743 struct pci_devinfo *dinfo;
3745 struct resource_list *rl;
3746 struct resource *res;
3747 int sec_reg, sub_reg;
3749 dinfo = device_get_ivars(child);
3751 rl = &dinfo->resources;
3752 switch (cfg->hdrtype & PCIM_HDRTYPE) {
3753 case PCIM_HDRTYPE_BRIDGE:
3754 sec_reg = PCIR_SECBUS_1;
3755 sub_reg = PCIR_SUBBUS_1;
3757 case PCIM_HDRTYPE_CARDBUS:
3758 sec_reg = PCIR_SECBUS_2;
3759 sub_reg = PCIR_SUBBUS_2;
3768 if (resource_list_find(rl, PCI_RES_BUS, *rid) == NULL)
3769 resource_list_add(rl, PCI_RES_BUS, *rid, start, end, count);
3770 if (!resource_list_reserved(rl, PCI_RES_BUS, *rid)) {
3771 res = resource_list_reserve(rl, dev, child, PCI_RES_BUS, rid,
3772 start, end, count, flags & ~RF_ACTIVE);
3774 resource_list_delete(rl, PCI_RES_BUS, *rid);
3775 device_printf(child, "allocating %ju bus%s failed\n",
3776 count, count == 1 ? "" : "es");
3780 device_printf(child,
3781 "Lazy allocation of %ju bus%s at %ju\n", count,
3782 count == 1 ? "" : "es", rman_get_start(res));
3783 PCI_WRITE_CONFIG(dev, child, sec_reg, rman_get_start(res), 1);
3784 PCI_WRITE_CONFIG(dev, child, sub_reg, rman_get_end(res), 1);
3786 return (resource_list_alloc(rl, dev, child, PCI_RES_BUS, rid, start,
3787 end, count, flags));
3792 pci_ea_bei_to_rid(device_t dev, int bei)
3795 struct pci_devinfo *dinfo;
3797 struct pcicfg_iov *iov;
3799 dinfo = device_get_ivars(dev);
3800 iov = dinfo->cfg.iov;
3802 iov_pos = iov->iov_pos;
3807 /* Check if matches BAR */
3808 if ((bei >= PCIM_EA_BEI_BAR_0) &&
3809 (bei <= PCIM_EA_BEI_BAR_5))
3810 return (PCIR_BAR(bei));
3813 if (bei == PCIM_EA_BEI_ROM)
3817 /* Check if matches VF_BAR */
3818 if ((iov != NULL) && (bei >= PCIM_EA_BEI_VF_BAR_0) &&
3819 (bei <= PCIM_EA_BEI_VF_BAR_5))
3820 return (PCIR_SRIOV_BAR(bei - PCIM_EA_BEI_VF_BAR_0) +
3828 pci_ea_is_enabled(device_t dev, int rid)
3830 struct pci_ea_entry *ea;
3831 struct pci_devinfo *dinfo;
3833 dinfo = device_get_ivars(dev);
3835 STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3836 if (pci_ea_bei_to_rid(dev, ea->eae_bei) == rid)
3837 return ((ea->eae_flags & PCIM_EA_ENABLE) > 0);
3844 pci_add_resources_ea(device_t bus, device_t dev, int alloc_iov)
3846 struct pci_ea_entry *ea;
3847 struct pci_devinfo *dinfo;
3848 pci_addr_t start, end, count;
3849 struct resource_list *rl;
3850 int type, flags, rid;
3851 struct resource *res;
3854 struct pcicfg_iov *iov;
3857 dinfo = device_get_ivars(dev);
3858 rl = &dinfo->resources;
3862 iov = dinfo->cfg.iov;
3865 if (dinfo->cfg.ea.ea_location == 0)
3868 STAILQ_FOREACH(ea, &dinfo->cfg.ea.ea_entries, eae_link) {
3871 * TODO: Ignore EA-BAR if is not enabled.
3872 * Currently the EA implementation supports
3873 * only situation, where EA structure contains
3874 * predefined entries. In case they are not enabled
3875 * leave them unallocated and proceed with
3876 * a legacy-BAR mechanism.
3878 if ((ea->eae_flags & PCIM_EA_ENABLE) == 0)
3881 switch ((ea->eae_flags & PCIM_EA_PP) >> PCIM_EA_PP_OFFSET) {
3882 case PCIM_EA_P_MEM_PREFETCH:
3883 case PCIM_EA_P_VF_MEM_PREFETCH:
3884 flags = RF_PREFETCHABLE;
3886 case PCIM_EA_P_VF_MEM:
3888 type = SYS_RES_MEMORY;
3891 type = SYS_RES_IOPORT;
3897 if (alloc_iov != 0) {
3899 /* Allocating IOV, confirm BEI matches */
3900 if ((ea->eae_bei < PCIM_EA_BEI_VF_BAR_0) ||
3901 (ea->eae_bei > PCIM_EA_BEI_VF_BAR_5))
3907 /* Allocating BAR, confirm BEI matches */
3908 if (((ea->eae_bei < PCIM_EA_BEI_BAR_0) ||
3909 (ea->eae_bei > PCIM_EA_BEI_BAR_5)) &&
3910 (ea->eae_bei != PCIM_EA_BEI_ROM))
3914 rid = pci_ea_bei_to_rid(dev, ea->eae_bei);
3918 /* Skip resources already allocated by EA */
3919 if ((resource_list_find(rl, SYS_RES_MEMORY, rid) != NULL) ||
3920 (resource_list_find(rl, SYS_RES_IOPORT, rid) != NULL))
3923 start = ea->eae_base;
3924 count = ea->eae_max_offset + 1;
3927 count = count * iov->iov_num_vfs;
3929 end = start + count - 1;
3933 resource_list_add(rl, type, rid, start, end, count);
3934 res = resource_list_reserve(rl, bus, dev, type, &rid, start, end, count,
3937 resource_list_delete(rl, type, rid);
3940 * Failed to allocate using EA, disable entry.
3941 * Another attempt to allocation will be performed
3942 * further, but this time using legacy BAR registers
3944 tmp = pci_read_config(dev, ea->eae_cfg_offset, 4);
3945 tmp &= ~PCIM_EA_ENABLE;
3946 pci_write_config(dev, ea->eae_cfg_offset, tmp, 4);
3949 * Disabling entry might fail in case it is hardwired.
3950 * Read flags again to match current status.
3952 ea->eae_flags = pci_read_config(dev, ea->eae_cfg_offset, 4);
3957 /* As per specification, fill BAR with zeros */
3958 pci_write_config(dev, rid, 0, 4);
3963 pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
3965 struct pci_devinfo *dinfo;
3967 struct resource_list *rl;
3968 const struct pci_quirk *q;
3972 dinfo = device_get_ivars(dev);
3974 rl = &dinfo->resources;
3975 devid = (cfg->device << 16) | cfg->vendor;
3977 /* Allocate resources using Enhanced Allocation */
3978 pci_add_resources_ea(bus, dev, 0);
3980 /* ATA devices needs special map treatment */
3981 if ((pci_get_class(dev) == PCIC_STORAGE) &&
3982 (pci_get_subclass(dev) == PCIS_STORAGE_IDE) &&
3983 ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) ||
3984 (!pci_read_config(dev, PCIR_BAR(0), 4) &&
3985 !pci_read_config(dev, PCIR_BAR(2), 4))) )
3986 pci_ata_maps(bus, dev, rl, force, prefetchmask);
3988 for (i = 0; i < cfg->nummaps;) {
3989 /* Skip resources already managed by EA */
3990 if ((resource_list_find(rl, SYS_RES_MEMORY, PCIR_BAR(i)) != NULL) ||
3991 (resource_list_find(rl, SYS_RES_IOPORT, PCIR_BAR(i)) != NULL) ||
3992 pci_ea_is_enabled(dev, PCIR_BAR(i))) {
3998 * Skip quirked resources.
4000 for (q = &pci_quirks[0]; q->devid != 0; q++)
4001 if (q->devid == devid &&
4002 q->type == PCI_QUIRK_UNMAP_REG &&
4003 q->arg1 == PCIR_BAR(i))
4005 if (q->devid != 0) {
4009 i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force,
4010 prefetchmask & (1 << i));
4014 * Add additional, quirked resources.
4016 for (q = &pci_quirks[0]; q->devid != 0; q++)
4017 if (q->devid == devid && q->type == PCI_QUIRK_MAP_REG)
4018 pci_add_map(bus, dev, q->arg1, rl, force, 0);
4020 if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) {
4021 #ifdef __PCI_REROUTE_INTERRUPT
4023 * Try to re-route interrupts. Sometimes the BIOS or
4024 * firmware may leave bogus values in these registers.
4025 * If the re-route fails, then just stick with what we
4028 pci_assign_interrupt(bus, dev, 1);
4030 pci_assign_interrupt(bus, dev, 0);
4034 if (pci_usb_takeover && pci_get_class(dev) == PCIC_SERIALBUS &&
4035 pci_get_subclass(dev) == PCIS_SERIALBUS_USB) {
4036 if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_XHCI)
4037 xhci_early_takeover(dev);
4038 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_EHCI)
4039 ehci_early_takeover(dev);
4040 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_OHCI)
4041 ohci_early_takeover(dev);
4042 else if (pci_get_progif(dev) == PCIP_SERIALBUS_USB_UHCI)
4043 uhci_early_takeover(dev);
4046 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
4048 * Reserve resources for secondary bus ranges behind bridge
4051 pci_reserve_secbus(bus, dev, cfg, rl);
4055 static struct pci_devinfo *
4056 pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
4059 struct pci_devinfo *dinfo;
4061 dinfo = pci_read_device(pcib, dev, domain, busno, slot, func);
4063 pci_add_child(dev, dinfo);
4069 pci_add_children(device_t dev, int domain, int busno)
4071 #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4072 device_t pcib = device_get_parent(dev);
4073 struct pci_devinfo *dinfo;
4075 int s, f, pcifunchigh;
4080 * Try to detect a device at slot 0, function 0. If it exists, try to
4081 * enable ARI. We must enable ARI before detecting the rest of the
4082 * functions on this bus as ARI changes the set of slots and functions
4083 * that are legal on this bus.
4085 dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0);
4086 if (dinfo != NULL && pci_enable_ari)
4087 PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
4090 * Start looking for new devices on slot 0 at function 1 because we
4091 * just identified the device at slot 0, function 0.
4095 maxslots = PCIB_MAXSLOTS(pcib);
4096 for (s = 0; s <= maxslots; s++, first_func = 0) {
4100 hdrtype = REG(PCIR_HDRTYPE, 1);
4101 if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4103 if (hdrtype & PCIM_MFDEV)
4104 pcifunchigh = PCIB_MAXFUNCS(pcib);
4105 for (f = first_func; f <= pcifunchigh; f++)
4106 pci_identify_function(pcib, dev, domain, busno, s, f);
4112 pci_rescan_method(device_t dev)
4114 #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w)
4115 device_t pcib = device_get_parent(dev);
4116 device_t child, *devlist, *unchanged;
4117 int devcount, error, i, j, maxslots, oldcount;
4118 int busno, domain, s, f, pcifunchigh;
4121 /* No need to check for ARI on a rescan. */
4122 error = device_get_children(dev, &devlist, &devcount);
4125 if (devcount != 0) {
4126 unchanged = malloc(devcount * sizeof(device_t), M_TEMP,
4128 if (unchanged == NULL) {
4129 free(devlist, M_TEMP);
4135 domain = pcib_get_domain(dev);
4136 busno = pcib_get_bus(dev);
4137 maxslots = PCIB_MAXSLOTS(pcib);
4138 for (s = 0; s <= maxslots; s++) {
4139 /* If function 0 is not present, skip to the next slot. */
4141 if (REG(PCIR_VENDOR, 2) == 0xffff)
4144 hdrtype = REG(PCIR_HDRTYPE, 1);
4145 if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
4147 if (hdrtype & PCIM_MFDEV)
4148 pcifunchigh = PCIB_MAXFUNCS(pcib);
4149 for (f = 0; f <= pcifunchigh; f++) {
4150 if (REG(PCIR_VENDOR, 2) == 0xffff)
4154 * Found a valid function. Check if a
4155 * device_t for this device already exists.
4157 for (i = 0; i < devcount; i++) {
4161 if (pci_get_slot(child) == s &&
4162 pci_get_function(child) == f) {
4163 unchanged[i] = child;
4168 pci_identify_function(pcib, dev, domain, busno, s, f);
4173 /* Remove devices that are no longer present. */
4174 for (i = 0; i < devcount; i++) {
4175 if (unchanged[i] != NULL)
4177 device_delete_child(dev, devlist[i]);
4180 free(devlist, M_TEMP);
4181 oldcount = devcount;
4183 /* Try to attach the devices just added. */
4184 error = device_get_children(dev, &devlist, &devcount);
4186 free(unchanged, M_TEMP);
4190 for (i = 0; i < devcount; i++) {
4191 for (j = 0; j < oldcount; j++) {
4192 if (devlist[i] == unchanged[j])
4196 device_probe_and_attach(devlist[i]);
4200 free(unchanged, M_TEMP);
4201 free(devlist, M_TEMP);
4208 pci_add_iov_child(device_t bus, device_t pf, uint16_t rid, uint16_t vid,
4211 struct pci_devinfo *vf_dinfo;
4213 int busno, slot, func;
4215 pcib = device_get_parent(bus);
4217 PCIB_DECODE_RID(pcib, rid, &busno, &slot, &func);
4219 vf_dinfo = pci_fill_devinfo(pcib, bus, pci_get_domain(pcib), busno,
4220 slot, func, vid, did);
4222 vf_dinfo->cfg.flags |= PCICFG_VF;
4223 pci_add_child(bus, vf_dinfo);
4225 return (vf_dinfo->cfg.dev);
4229 pci_create_iov_child_method(device_t bus, device_t pf, uint16_t rid,
4230 uint16_t vid, uint16_t did)
4233 return (pci_add_iov_child(bus, pf, rid, vid, did));
4238 pci_add_child_clear_aer(device_t dev, struct pci_devinfo *dinfo)
4244 if (dinfo->cfg.pcie.pcie_location != 0 &&
4245 dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT) {
4246 r2 = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4248 r2 &= ~(PCIEM_ROOT_CTL_SERR_CORR |
4249 PCIEM_ROOT_CTL_SERR_NONFATAL | PCIEM_ROOT_CTL_SERR_FATAL);
4250 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4251 PCIER_ROOT_CTL, r2, 2);
4253 if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
4254 r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
4255 pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
4256 if (r != 0 && bootverbose) {
4257 pci_printf(&dinfo->cfg,
4258 "clearing AER UC 0x%08x -> 0x%08x\n",
4259 r, pci_read_config(dev, aer + PCIR_AER_UC_STATUS,
4263 r = pci_read_config(dev, aer + PCIR_AER_UC_MASK, 4);
4264 r &= ~(PCIM_AER_UC_TRAINING_ERROR |
4265 PCIM_AER_UC_DL_PROTOCOL_ERROR |
4266 PCIM_AER_UC_SURPRISE_LINK_DOWN |
4267 PCIM_AER_UC_POISONED_TLP |
4268 PCIM_AER_UC_FC_PROTOCOL_ERROR |
4269 PCIM_AER_UC_COMPLETION_TIMEOUT |
4270 PCIM_AER_UC_COMPLETER_ABORT |
4271 PCIM_AER_UC_UNEXPECTED_COMPLETION |
4272 PCIM_AER_UC_RECEIVER_OVERFLOW |
4273 PCIM_AER_UC_MALFORMED_TLP |
4274 PCIM_AER_UC_ECRC_ERROR |
4275 PCIM_AER_UC_UNSUPPORTED_REQUEST |
4276 PCIM_AER_UC_ACS_VIOLATION |
4277 PCIM_AER_UC_INTERNAL_ERROR |
4278 PCIM_AER_UC_MC_BLOCKED_TLP |
4279 PCIM_AER_UC_ATOMIC_EGRESS_BLK |
4280 PCIM_AER_UC_TLP_PREFIX_BLOCKED);
4281 pci_write_config(dev, aer + PCIR_AER_UC_MASK, r, 4);
4283 r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
4284 pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
4285 if (r != 0 && bootverbose) {
4286 pci_printf(&dinfo->cfg,
4287 "clearing AER COR 0x%08x -> 0x%08x\n",
4288 r, pci_read_config(dev, aer + PCIR_AER_COR_STATUS,
4292 r = pci_read_config(dev, aer + PCIR_AER_COR_MASK, 4);
4293 r &= ~(PCIM_AER_COR_RECEIVER_ERROR |
4294 PCIM_AER_COR_BAD_TLP |
4295 PCIM_AER_COR_BAD_DLLP |
4296 PCIM_AER_COR_REPLAY_ROLLOVER |
4297 PCIM_AER_COR_REPLAY_TIMEOUT |
4298 PCIM_AER_COR_ADVISORY_NF_ERROR |
4299 PCIM_AER_COR_INTERNAL_ERROR |
4300 PCIM_AER_COR_HEADER_LOG_OVFLOW);
4301 pci_write_config(dev, aer + PCIR_AER_COR_MASK, r, 4);
4303 r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
4304 PCIER_DEVICE_CTL, 2);
4305 r |= PCIEM_CTL_COR_ENABLE | PCIEM_CTL_NFER_ENABLE |
4306 PCIEM_CTL_FER_ENABLE | PCIEM_CTL_URR_ENABLE;
4307 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
4308 PCIER_DEVICE_CTL, r, 2);
4313 pci_add_child(device_t bus, struct pci_devinfo *dinfo)
4317 dinfo->cfg.dev = dev = device_add_child(bus, NULL, -1);
4318 device_set_ivars(dev, dinfo);
4319 resource_list_init(&dinfo->resources);
4320 pci_cfg_save(dev, dinfo, 0);
4321 pci_cfg_restore(dev, dinfo);
4322 pci_print_verbose(dinfo);
4323 pci_add_resources(bus, dev, 0, 0);
4324 pci_child_added(dinfo->cfg.dev);
4326 if (pci_clear_aer_on_attach)
4327 pci_add_child_clear_aer(dev, dinfo);
4329 EVENTHANDLER_INVOKE(pci_add_device, dinfo->cfg.dev);
4333 pci_child_added_method(device_t dev, device_t child)
4339 pci_probe(device_t dev)
4342 device_set_desc(dev, "PCI bus");
4344 /* Allow other subclasses to override this driver. */
4345 return (BUS_PROBE_GENERIC);
4349 pci_attach_common(device_t dev)
4351 struct pci_softc *sc;
4357 sc = device_get_softc(dev);
4358 domain = pcib_get_domain(dev);
4359 busno = pcib_get_bus(dev);
4362 sc->sc_bus = bus_alloc_resource(dev, PCI_RES_BUS, &rid, busno, busno,
4364 if (sc->sc_bus == NULL) {
4365 device_printf(dev, "failed to allocate bus number\n");
4370 device_printf(dev, "domain=%d, physical bus=%d\n",
4372 sc->sc_dma_tag = bus_get_dma_tag(dev);
4377 pci_attach(device_t dev)
4379 int busno, domain, error;
4381 error = pci_attach_common(dev);
4386 * Since there can be multiple independently numbered PCI
4387 * buses on systems with multiple PCI domains, we can't use
4388 * the unit number to decide which bus we are probing. We ask
4389 * the parent pcib what our domain and bus numbers are.
4391 domain = pcib_get_domain(dev);
4392 busno = pcib_get_bus(dev);
4393 pci_add_children(dev, domain, busno);
4394 return (bus_generic_attach(dev));
4398 pci_detach(device_t dev)
4401 struct pci_softc *sc;
4405 error = bus_generic_detach(dev);
4409 sc = device_get_softc(dev);
4410 error = bus_release_resource(dev, PCI_RES_BUS, 0, sc->sc_bus);
4414 return (device_delete_children(dev));
4418 pci_hint_device_unit(device_t dev, device_t child, const char *name, int *unitp)
4422 char me1[24], me2[32];
4426 d = pci_get_domain(child);
4427 b = pci_get_bus(child);
4428 s = pci_get_slot(child);
4429 f = pci_get_function(child);
4430 snprintf(me1, sizeof(me1), "pci%u:%u:%u", b, s, f);
4431 snprintf(me2, sizeof(me2), "pci%u:%u:%u:%u", d, b, s, f);
4433 while (resource_find_dev(&line, name, &unit, "at", NULL) == 0) {
4434 resource_string_value(name, unit, "at", &at);
4435 if (strcmp(at, me1) != 0 && strcmp(at, me2) != 0)
4436 continue; /* No match, try next candidate */
4443 pci_set_power_child(device_t dev, device_t child, int state)
4449 * Set the device to the given state. If the firmware suggests
4450 * a different power state, use it instead. If power management
4451 * is not present, the firmware is responsible for managing
4452 * device power. Skip children who aren't attached since they
4453 * are handled separately.
4455 pcib = device_get_parent(dev);
4457 if (device_is_attached(child) &&
4458 PCIB_POWER_FOR_SLEEP(pcib, child, &dstate) == 0)
4459 pci_set_powerstate(child, dstate);
4463 pci_suspend_child(device_t dev, device_t child)
4465 struct pci_devinfo *dinfo;
4466 struct resource_list_entry *rle;
4469 dinfo = device_get_ivars(child);
4472 * Save the PCI configuration space for the child and set the
4473 * device in the appropriate power state for this sleep state.
4475 pci_cfg_save(child, dinfo, 0);
4477 /* Suspend devices before potentially powering them down. */
4478 error = bus_generic_suspend_child(dev, child);
4483 if (pci_do_power_suspend) {
4485 * Make sure this device's interrupt handler is not invoked
4486 * in the case the device uses a shared interrupt that can
4487 * be raised by some other device.
4488 * This is applicable only to regular (legacy) PCI interrupts
4489 * as MSI/MSI-X interrupts are never shared.
4491 rle = resource_list_find(&dinfo->resources,
4493 if (rle != NULL && rle->res != NULL)
4494 (void)bus_suspend_intr(child, rle->res);
4495 pci_set_power_child(dev, child, PCI_POWERSTATE_D3);
4502 pci_resume_child(device_t dev, device_t child)
4504 struct pci_devinfo *dinfo;
4505 struct resource_list_entry *rle;
4507 if (pci_do_power_resume)
4508 pci_set_power_child(dev, child, PCI_POWERSTATE_D0);
4510 dinfo = device_get_ivars(child);
4511 pci_cfg_restore(child, dinfo);
4512 if (!device_is_attached(child))
4513 pci_cfg_save(child, dinfo, 1);
4515 bus_generic_resume_child(dev, child);
4518 * Allow interrupts only after fully resuming the driver and hardware.
4520 if (pci_do_power_suspend) {
4521 /* See pci_suspend_child for details. */
4522 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0);
4523 if (rle != NULL && rle->res != NULL)
4524 (void)bus_resume_intr(child, rle->res);
4531 pci_resume(device_t dev)
4533 device_t child, *devlist;
4534 int error, i, numdevs;
4536 if ((error = device_get_children(dev, &devlist, &numdevs)) != 0)
4540 * Resume critical devices first, then everything else later.
4542 for (i = 0; i < numdevs; i++) {
4544 switch (pci_get_class(child)) {
4548 case PCIC_BASEPERIPH:
4549 BUS_RESUME_CHILD(dev, child);
4553 for (i = 0; i < numdevs; i++) {
4555 switch (pci_get_class(child)) {
4559 case PCIC_BASEPERIPH:
4562 BUS_RESUME_CHILD(dev, child);
4565 free(devlist, M_TEMP);
4570 pci_load_vendor_data(void)
4576 data = preload_search_by_type("pci_vendor_data");
4578 ptr = preload_fetch_addr(data);
4579 sz = preload_fetch_size(data);
4580 if (ptr != NULL && sz != 0) {
4581 pci_vendordata = ptr;
4582 pci_vendordata_size = sz;
4583 /* terminate the database */
4584 pci_vendordata[pci_vendordata_size] = '\n';
4590 pci_driver_added(device_t dev, driver_t *driver)
4595 struct pci_devinfo *dinfo;
4599 device_printf(dev, "driver added\n");
4600 DEVICE_IDENTIFY(driver, dev);
4601 if (device_get_children(dev, &devlist, &numdevs) != 0)
4603 for (i = 0; i < numdevs; i++) {
4605 if (device_get_state(child) != DS_NOTPRESENT)
4607 dinfo = device_get_ivars(child);
4608 pci_print_verbose(dinfo);
4610 pci_printf(&dinfo->cfg, "reprobing on driver added\n");
4611 pci_cfg_restore(child, dinfo);
4612 if (device_probe_and_attach(child) != 0)
4613 pci_child_detached(dev, child);
4615 free(devlist, M_TEMP);
4619 pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags,
4620 driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep)
4622 struct pci_devinfo *dinfo;
4623 struct msix_table_entry *mte;
4624 struct msix_vector *mv;
4630 error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr,
4635 /* If this is not a direct child, just bail out. */
4636 if (device_get_parent(child) != dev) {
4641 rid = rman_get_rid(irq);
4643 /* Make sure that INTx is enabled */
4644 pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4647 * Check to see if the interrupt is MSI or MSI-X.
4648 * Ask our parent to map the MSI and give
4649 * us the address and data register values.
4650 * If we fail for some reason, teardown the
4651 * interrupt handler.
4653 dinfo = device_get_ivars(child);
4654 if (dinfo->cfg.msi.msi_alloc > 0) {
4655 if (dinfo->cfg.msi.msi_addr == 0) {
4656 KASSERT(dinfo->cfg.msi.msi_handlers == 0,
4657 ("MSI has handlers, but vectors not mapped"));
4658 error = PCIB_MAP_MSI(device_get_parent(dev),
4659 child, rman_get_start(irq), &addr, &data);
4662 dinfo->cfg.msi.msi_addr = addr;
4663 dinfo->cfg.msi.msi_data = data;
4665 if (dinfo->cfg.msi.msi_handlers == 0)
4666 pci_enable_msi(child, dinfo->cfg.msi.msi_addr,
4667 dinfo->cfg.msi.msi_data);
4668 dinfo->cfg.msi.msi_handlers++;
4670 KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4671 ("No MSI or MSI-X interrupts allocated"));
4672 KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4673 ("MSI-X index too high"));
4674 mte = &dinfo->cfg.msix.msix_table[rid - 1];
4675 KASSERT(mte->mte_vector != 0, ("no message vector"));
4676 mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1];
4677 KASSERT(mv->mv_irq == rman_get_start(irq),
4679 if (mv->mv_address == 0) {
4680 KASSERT(mte->mte_handlers == 0,
4681 ("MSI-X table entry has handlers, but vector not mapped"));
4682 error = PCIB_MAP_MSI(device_get_parent(dev),
4683 child, rman_get_start(irq), &addr, &data);
4686 mv->mv_address = addr;
4691 * The MSIX table entry must be made valid by
4692 * incrementing the mte_handlers before
4693 * calling pci_enable_msix() and
4694 * pci_resume_msix(). Else the MSIX rewrite
4695 * table quirk will not work as expected.
4697 mte->mte_handlers++;
4698 if (mte->mte_handlers == 1) {
4699 pci_enable_msix(child, rid - 1, mv->mv_address,
4701 pci_unmask_msix(child, rid - 1);
4706 * Make sure that INTx is disabled if we are using MSI/MSI-X,
4707 * unless the device is affected by PCI_QUIRK_MSI_INTX_BUG,
4708 * in which case we "enable" INTx so MSI/MSI-X actually works.
4710 if (!pci_has_quirk(pci_get_devid(child),
4711 PCI_QUIRK_MSI_INTX_BUG))
4712 pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4714 pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS);
4717 (void)bus_generic_teardown_intr(dev, child, irq,
4727 pci_teardown_intr(device_t dev, device_t child, struct resource *irq,
4730 struct msix_table_entry *mte;
4731 struct resource_list_entry *rle;
4732 struct pci_devinfo *dinfo;
4735 if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE))
4738 /* If this isn't a direct child, just bail out */
4739 if (device_get_parent(child) != dev)
4740 return(bus_generic_teardown_intr(dev, child, irq, cookie));
4742 rid = rman_get_rid(irq);
4745 pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS);
4748 * Check to see if the interrupt is MSI or MSI-X. If so,
4749 * decrement the appropriate handlers count and mask the
4750 * MSI-X message, or disable MSI messages if the count
4753 dinfo = device_get_ivars(child);
4754 rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid);
4755 if (rle->res != irq)
4757 if (dinfo->cfg.msi.msi_alloc > 0) {
4758 KASSERT(rid <= dinfo->cfg.msi.msi_alloc,
4759 ("MSI-X index too high"));
4760 if (dinfo->cfg.msi.msi_handlers == 0)
4762 dinfo->cfg.msi.msi_handlers--;
4763 if (dinfo->cfg.msi.msi_handlers == 0)
4764 pci_disable_msi(child);
4766 KASSERT(dinfo->cfg.msix.msix_alloc > 0,
4767 ("No MSI or MSI-X interrupts allocated"));
4768 KASSERT(rid <= dinfo->cfg.msix.msix_table_len,
4769 ("MSI-X index too high"));
4770 mte = &dinfo->cfg.msix.msix_table[rid - 1];
4771 if (mte->mte_handlers == 0)
4773 mte->mte_handlers--;
4774 if (mte->mte_handlers == 0)
4775 pci_mask_msix(child, rid - 1);
4778 error = bus_generic_teardown_intr(dev, child, irq, cookie);
4781 ("%s: generic teardown failed for MSI/MSI-X", __func__));
4786 pci_print_child(device_t dev, device_t child)
4788 struct pci_devinfo *dinfo;
4789 struct resource_list *rl;
4792 dinfo = device_get_ivars(child);
4793 rl = &dinfo->resources;
4795 retval += bus_print_child_header(dev, child);
4797 retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#jx");
4798 retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#jx");
4799 retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%jd");
4800 if (device_get_flags(dev))
4801 retval += printf(" flags %#x", device_get_flags(dev));
4803 retval += printf(" at device %d.%d", pci_get_slot(child),
4804 pci_get_function(child));
4806 retval += bus_print_child_domain(dev, child);
4807 retval += bus_print_child_footer(dev, child);
4816 int report; /* 0 = bootverbose, 1 = always */
4818 } pci_nomatch_tab[] = {
4819 {PCIC_OLD, -1, 1, "old"},
4820 {PCIC_OLD, PCIS_OLD_NONVGA, 1, "non-VGA display device"},
4821 {PCIC_OLD, PCIS_OLD_VGA, 1, "VGA-compatible display device"},
4822 {PCIC_STORAGE, -1, 1, "mass storage"},
4823 {PCIC_STORAGE, PCIS_STORAGE_SCSI, 1, "SCSI"},
4824 {PCIC_STORAGE, PCIS_STORAGE_IDE, 1, "ATA"},
4825 {PCIC_STORAGE, PCIS_STORAGE_FLOPPY, 1, "floppy disk"},
4826 {PCIC_STORAGE, PCIS_STORAGE_IPI, 1, "IPI"},
4827 {PCIC_STORAGE, PCIS_STORAGE_RAID, 1, "RAID"},
4828 {PCIC_STORAGE, PCIS_STORAGE_ATA_ADMA, 1, "ATA (ADMA)"},
4829 {PCIC_STORAGE, PCIS_STORAGE_SATA, 1, "SATA"},
4830 {PCIC_STORAGE, PCIS_STORAGE_SAS, 1, "SAS"},
4831 {PCIC_STORAGE, PCIS_STORAGE_NVM, 1, "NVM"},
4832 {PCIC_NETWORK, -1, 1, "network"},
4833 {PCIC_NETWORK, PCIS_NETWORK_ETHERNET, 1, "ethernet"},
4834 {PCIC_NETWORK, PCIS_NETWORK_TOKENRING, 1, "token ring"},
4835 {PCIC_NETWORK, PCIS_NETWORK_FDDI, 1, "fddi"},
4836 {PCIC_NETWORK, PCIS_NETWORK_ATM, 1, "ATM"},
4837 {PCIC_NETWORK, PCIS_NETWORK_ISDN, 1, "ISDN"},
4838 {PCIC_DISPLAY, -1, 1, "display"},
4839 {PCIC_DISPLAY, PCIS_DISPLAY_VGA, 1, "VGA"},
4840 {PCIC_DISPLAY, PCIS_DISPLAY_XGA, 1, "XGA"},
4841 {PCIC_DISPLAY, PCIS_DISPLAY_3D, 1, "3D"},
4842 {PCIC_MULTIMEDIA, -1, 1, "multimedia"},
4843 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_VIDEO, 1, "video"},
4844 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_AUDIO, 1, "audio"},
4845 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_TELE, 1, "telephony"},
4846 {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_HDA, 1, "HDA"},
4847 {PCIC_MEMORY, -1, 1, "memory"},
4848 {PCIC_MEMORY, PCIS_MEMORY_RAM, 1, "RAM"},
4849 {PCIC_MEMORY, PCIS_MEMORY_FLASH, 1, "flash"},
4850 {PCIC_BRIDGE, -1, 1, "bridge"},
4851 {PCIC_BRIDGE, PCIS_BRIDGE_HOST, 1, "HOST-PCI"},
4852 {PCIC_BRIDGE, PCIS_BRIDGE_ISA, 1, "PCI-ISA"},
4853 {PCIC_BRIDGE, PCIS_BRIDGE_EISA, 1, "PCI-EISA"},
4854 {PCIC_BRIDGE, PCIS_BRIDGE_MCA, 1, "PCI-MCA"},
4855 {PCIC_BRIDGE, PCIS_BRIDGE_PCI, 1, "PCI-PCI"},
4856 {PCIC_BRIDGE, PCIS_BRIDGE_PCMCIA, 1, "PCI-PCMCIA"},
4857 {PCIC_BRIDGE, PCIS_BRIDGE_NUBUS, 1, "PCI-NuBus"},
4858 {PCIC_BRIDGE, PCIS_BRIDGE_CARDBUS, 1, "PCI-CardBus"},
4859 {PCIC_BRIDGE, PCIS_BRIDGE_RACEWAY, 1, "PCI-RACEway"},
4860 {PCIC_SIMPLECOMM, -1, 1, "simple comms"},
4861 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_UART, 1, "UART"}, /* could detect 16550 */
4862 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_PAR, 1, "parallel port"},
4863 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MULSER, 1, "multiport serial"},
4864 {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MODEM, 1, "generic modem"},
4865 {PCIC_BASEPERIPH, -1, 0, "base peripheral"},
4866 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PIC, 1, "interrupt controller"},
4867 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_DMA, 1, "DMA controller"},
4868 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_TIMER, 1, "timer"},
4869 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_RTC, 1, "realtime clock"},
4870 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PCIHOT, 1, "PCI hot-plug controller"},
4871 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_SDHC, 1, "SD host controller"},
4872 {PCIC_BASEPERIPH, PCIS_BASEPERIPH_IOMMU, 1, "IOMMU"},
4873 {PCIC_INPUTDEV, -1, 1, "input device"},
4874 {PCIC_INPUTDEV, PCIS_INPUTDEV_KEYBOARD, 1, "keyboard"},
4875 {PCIC_INPUTDEV, PCIS_INPUTDEV_DIGITIZER,1, "digitizer"},
4876 {PCIC_INPUTDEV, PCIS_INPUTDEV_MOUSE, 1, "mouse"},
4877 {PCIC_INPUTDEV, PCIS_INPUTDEV_SCANNER, 1, "scanner"},
4878 {PCIC_INPUTDEV, PCIS_INPUTDEV_GAMEPORT, 1, "gameport"},
4879 {PCIC_DOCKING, -1, 1, "docking station"},
4880 {PCIC_PROCESSOR, -1, 1, "processor"},
4881 {PCIC_SERIALBUS, -1, 1, "serial bus"},
4882 {PCIC_SERIALBUS, PCIS_SERIALBUS_FW, 1, "FireWire"},
4883 {PCIC_SERIALBUS, PCIS_SERIALBUS_ACCESS, 1, "AccessBus"},
4884 {PCIC_SERIALBUS, PCIS_SERIALBUS_SSA, 1, "SSA"},
4885 {PCIC_SERIALBUS, PCIS_SERIALBUS_USB, 1, "USB"},
4886 {PCIC_SERIALBUS, PCIS_SERIALBUS_FC, 1, "Fibre Channel"},
4887 {PCIC_SERIALBUS, PCIS_SERIALBUS_SMBUS, 0, "SMBus"},
4888 {PCIC_WIRELESS, -1, 1, "wireless controller"},
4889 {PCIC_WIRELESS, PCIS_WIRELESS_IRDA, 1, "iRDA"},
4890 {PCIC_WIRELESS, PCIS_WIRELESS_IR, 1, "IR"},
4891 {PCIC_WIRELESS, PCIS_WIRELESS_RF, 1, "RF"},
4892 {PCIC_INTELLIIO, -1, 1, "intelligent I/O controller"},
4893 {PCIC_INTELLIIO, PCIS_INTELLIIO_I2O, 1, "I2O"},
4894 {PCIC_SATCOM, -1, 1, "satellite communication"},
4895 {PCIC_SATCOM, PCIS_SATCOM_TV, 1, "sat TV"},
4896 {PCIC_SATCOM, PCIS_SATCOM_AUDIO, 1, "sat audio"},
4897 {PCIC_SATCOM, PCIS_SATCOM_VOICE, 1, "sat voice"},
4898 {PCIC_SATCOM, PCIS_SATCOM_DATA, 1, "sat data"},
4899 {PCIC_CRYPTO, -1, 1, "encrypt/decrypt"},
4900 {PCIC_CRYPTO, PCIS_CRYPTO_NETCOMP, 1, "network/computer crypto"},
4901 {PCIC_CRYPTO, PCIS_CRYPTO_ENTERTAIN, 1, "entertainment crypto"},
4902 {PCIC_DASP, -1, 0, "dasp"},
4903 {PCIC_DASP, PCIS_DASP_DPIO, 1, "DPIO module"},
4904 {PCIC_DASP, PCIS_DASP_PERFCNTRS, 1, "performance counters"},
4905 {PCIC_DASP, PCIS_DASP_COMM_SYNC, 1, "communication synchronizer"},
4906 {PCIC_DASP, PCIS_DASP_MGMT_CARD, 1, "signal processing management"},
4911 pci_probe_nomatch(device_t dev, device_t child)
4914 const char *cp, *scp;
4918 * Look for a listing for this device in a loaded device database.
4921 if ((device = pci_describe_device(child)) != NULL) {
4922 device_printf(dev, "<%s>", device);
4923 free(device, M_DEVBUF);
4926 * Scan the class/subclass descriptions for a general
4931 for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) {
4932 if (pci_nomatch_tab[i].class == pci_get_class(child)) {
4933 if (pci_nomatch_tab[i].subclass == -1) {
4934 cp = pci_nomatch_tab[i].desc;
4935 report = pci_nomatch_tab[i].report;
4936 } else if (pci_nomatch_tab[i].subclass ==
4937 pci_get_subclass(child)) {
4938 scp = pci_nomatch_tab[i].desc;
4939 report = pci_nomatch_tab[i].report;
4943 if (report || bootverbose) {
4944 device_printf(dev, "<%s%s%s>",
4946 ((cp != NULL) && (scp != NULL)) ? ", " : "",
4950 if (report || bootverbose) {
4951 printf(" at device %d.%d (no driver attached)\n",
4952 pci_get_slot(child), pci_get_function(child));
4954 pci_cfg_save(child, device_get_ivars(child), 1);
4958 pci_child_detached(device_t dev, device_t child)
4960 struct pci_devinfo *dinfo;
4961 struct resource_list *rl;
4963 dinfo = device_get_ivars(child);
4964 rl = &dinfo->resources;
4967 * Have to deallocate IRQs before releasing any MSI messages and
4968 * have to release MSI messages before deallocating any memory
4971 if (resource_list_release_active(rl, dev, child, SYS_RES_IRQ) != 0)
4972 pci_printf(&dinfo->cfg, "Device leaked IRQ resources\n");
4973 if (dinfo->cfg.msi.msi_alloc != 0 || dinfo->cfg.msix.msix_alloc != 0) {
4974 pci_printf(&dinfo->cfg, "Device leaked MSI vectors\n");
4975 (void)pci_release_msi(child);
4977 if (resource_list_release_active(rl, dev, child, SYS_RES_MEMORY) != 0)
4978 pci_printf(&dinfo->cfg, "Device leaked memory resources\n");
4979 if (resource_list_release_active(rl, dev, child, SYS_RES_IOPORT) != 0)
4980 pci_printf(&dinfo->cfg, "Device leaked I/O resources\n");
4982 if (resource_list_release_active(rl, dev, child, PCI_RES_BUS) != 0)
4983 pci_printf(&dinfo->cfg, "Device leaked PCI bus numbers\n");
4986 pci_cfg_save(child, dinfo, 1);
4990 * Parse the PCI device database, if loaded, and return a pointer to a
4991 * description of the device.
4993 * The database is flat text formatted as follows:
4995 * Any line not in a valid format is ignored.
4996 * Lines are terminated with newline '\n' characters.
4998 * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then
5001 * A DEVICE line is entered immediately below the corresponding VENDOR ID.
5002 * - devices cannot be listed without a corresponding VENDOR line.
5003 * A DEVICE line consists of a TAB, the 4 digit (hex) device code,
5004 * another TAB, then the device name.
5008 * Assuming (ptr) points to the beginning of a line in the database,
5009 * return the vendor or device and description of the next entry.
5010 * The value of (vendor) or (device) inappropriate for the entry type
5011 * is set to -1. Returns nonzero at the end of the database.
5013 * Note that this is slightly unrobust in the face of corrupt data;
5014 * we attempt to safeguard against this by spamming the end of the
5015 * database with a newline when we initialise.
5018 pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc)
5027 left = pci_vendordata_size - (cp - pci_vendordata);
5035 sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2)
5039 sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2)
5042 /* skip to next line */
5043 while (*cp != '\n' && left > 0) {
5052 /* skip to next line */
5053 while (*cp != '\n' && left > 0) {
5057 if (*cp == '\n' && left > 0)
5064 pci_describe_device(device_t dev)
5067 char *desc, *vp, *dp, *line;
5069 desc = vp = dp = NULL;
5072 * If we have no vendor data, we can't do anything.
5074 if (pci_vendordata == NULL)
5078 * Scan the vendor data looking for this device
5080 line = pci_vendordata;
5081 if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5084 if (pci_describe_parse_line(&line, &vendor, &device, &vp))
5086 if (vendor == pci_get_vendor(dev))
5089 if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL)
5092 if (pci_describe_parse_line(&line, &vendor, &device, &dp)) {
5100 if (device == pci_get_device(dev))
5104 snprintf(dp, 80, "0x%x", pci_get_device(dev));
5105 if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) !=
5107 sprintf(desc, "%s, %s", vp, dp);
5117 pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
5119 struct pci_devinfo *dinfo;
5122 dinfo = device_get_ivars(child);
5126 case PCI_IVAR_ETHADDR:
5128 * The generic accessor doesn't deal with failure, so
5129 * we set the return value, then return an error.
5131 *((uint8_t **) result) = NULL;
5133 case PCI_IVAR_SUBVENDOR:
5134 *result = cfg->subvendor;
5136 case PCI_IVAR_SUBDEVICE:
5137 *result = cfg->subdevice;
5139 case PCI_IVAR_VENDOR:
5140 *result = cfg->vendor;
5142 case PCI_IVAR_DEVICE:
5143 *result = cfg->device;
5145 case PCI_IVAR_DEVID:
5146 *result = (cfg->device << 16) | cfg->vendor;
5148 case PCI_IVAR_CLASS:
5149 *result = cfg->baseclass;
5151 case PCI_IVAR_SUBCLASS:
5152 *result = cfg->subclass;
5154 case PCI_IVAR_PROGIF:
5155 *result = cfg->progif;
5157 case PCI_IVAR_REVID:
5158 *result = cfg->revid;
5160 case PCI_IVAR_INTPIN:
5161 *result = cfg->intpin;
5164 *result = cfg->intline;
5166 case PCI_IVAR_DOMAIN:
5167 *result = cfg->domain;
5173 *result = cfg->slot;
5175 case PCI_IVAR_FUNCTION:
5176 *result = cfg->func;
5178 case PCI_IVAR_CMDREG:
5179 *result = cfg->cmdreg;
5181 case PCI_IVAR_CACHELNSZ:
5182 *result = cfg->cachelnsz;
5184 case PCI_IVAR_MINGNT:
5185 if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5189 *result = cfg->mingnt;
5191 case PCI_IVAR_MAXLAT:
5192 if (cfg->hdrtype != PCIM_HDRTYPE_NORMAL) {
5196 *result = cfg->maxlat;
5198 case PCI_IVAR_LATTIMER:
5199 *result = cfg->lattimer;
5208 pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
5210 struct pci_devinfo *dinfo;
5212 dinfo = device_get_ivars(child);
5215 case PCI_IVAR_INTPIN:
5216 dinfo->cfg.intpin = value;
5218 case PCI_IVAR_ETHADDR:
5219 case PCI_IVAR_SUBVENDOR:
5220 case PCI_IVAR_SUBDEVICE:
5221 case PCI_IVAR_VENDOR:
5222 case PCI_IVAR_DEVICE:
5223 case PCI_IVAR_DEVID:
5224 case PCI_IVAR_CLASS:
5225 case PCI_IVAR_SUBCLASS:
5226 case PCI_IVAR_PROGIF:
5227 case PCI_IVAR_REVID:
5229 case PCI_IVAR_DOMAIN:
5232 case PCI_IVAR_FUNCTION:
5233 return (EINVAL); /* disallow for now */
5240 #include "opt_ddb.h"
5242 #include <ddb/ddb.h>
5243 #include <sys/cons.h>
5246 * List resources based on pci map registers, used for within ddb
5249 DB_SHOW_COMMAND(pciregs, db_pci_dump)
5251 struct pci_devinfo *dinfo;
5252 struct devlist *devlist_head;
5255 int i, error, none_count;
5258 /* get the head of the device queue */
5259 devlist_head = &pci_devq;
5262 * Go through the list of devices and print out devices
5264 for (error = 0, i = 0,
5265 dinfo = STAILQ_FIRST(devlist_head);
5266 (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit;
5267 dinfo = STAILQ_NEXT(dinfo, pci_links), i++) {
5269 /* Populate pd_name and pd_unit */
5272 name = device_get_name(dinfo->cfg.dev);
5275 db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x "
5276 "chip=0x%08x rev=0x%02x hdr=0x%02x\n",
5277 (name && *name) ? name : "none",
5278 (name && *name) ? (int)device_get_unit(dinfo->cfg.dev) :
5280 p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev,
5281 p->pc_sel.pc_func, (p->pc_class << 16) |
5282 (p->pc_subclass << 8) | p->pc_progif,
5283 (p->pc_subdevice << 16) | p->pc_subvendor,
5284 (p->pc_device << 16) | p->pc_vendor,
5285 p->pc_revid, p->pc_hdr);
5290 static struct resource *
5291 pci_reserve_map(device_t dev, device_t child, int type, int *rid,
5292 rman_res_t start, rman_res_t end, rman_res_t count, u_int num,
5295 struct pci_devinfo *dinfo = device_get_ivars(child);
5296 struct resource_list *rl = &dinfo->resources;
5297 struct resource *res;
5300 pci_addr_t map, testval;
5305 /* If rid is managed by EA, ignore it */
5306 if (pci_ea_is_enabled(child, *rid))
5309 pm = pci_find_bar(child, *rid);
5311 /* This is a BAR that we failed to allocate earlier. */
5312 mapsize = pm->pm_size;
5316 * Weed out the bogons, and figure out how large the
5317 * BAR/map is. BARs that read back 0 here are bogus
5318 * and unimplemented. Note: atapci in legacy mode are
5319 * special and handled elsewhere in the code. If you
5320 * have a atapci device in legacy mode and it fails
5321 * here, that other code is broken.
5323 pci_read_bar(child, *rid, &map, &testval, NULL);
5326 * Determine the size of the BAR and ignore BARs with a size
5327 * of 0. Device ROM BARs use a different mask value.
5329 if (PCIR_IS_BIOS(&dinfo->cfg, *rid))
5330 mapsize = pci_romsize(testval);
5332 mapsize = pci_mapsize(testval);
5335 pm = pci_add_bar(child, *rid, map, mapsize);
5338 if (PCI_BAR_MEM(map) || PCIR_IS_BIOS(&dinfo->cfg, *rid)) {
5339 if (type != SYS_RES_MEMORY) {
5342 "child %s requested type %d for rid %#x,"
5343 " but the BAR says it is an memio\n",
5344 device_get_nameunit(child), type, *rid);
5348 if (type != SYS_RES_IOPORT) {
5351 "child %s requested type %d for rid %#x,"
5352 " but the BAR says it is an ioport\n",
5353 device_get_nameunit(child), type, *rid);
5359 * For real BARs, we need to override the size that
5360 * the driver requests, because that's what the BAR
5361 * actually uses and we would otherwise have a
5362 * situation where we might allocate the excess to
5363 * another driver, which won't work.
5365 count = ((pci_addr_t)1 << mapsize) * num;
5366 if (RF_ALIGNMENT(flags) < mapsize)
5367 flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize);
5368 if (PCI_BAR_MEM(map) && (map & PCIM_BAR_MEM_PREFETCH))
5369 flags |= RF_PREFETCHABLE;
5372 * Allocate enough resource, and then write back the
5373 * appropriate BAR for that resource.
5375 resource_list_add(rl, type, *rid, start, end, count);
5376 res = resource_list_reserve(rl, dev, child, type, rid, start, end,
5377 count, flags & ~RF_ACTIVE);
5379 resource_list_delete(rl, type, *rid);
5380 device_printf(child,
5381 "%#jx bytes of rid %#x res %d failed (%#jx, %#jx).\n",
5382 count, *rid, type, start, end);
5386 device_printf(child,
5387 "Lazy allocation of %#jx bytes rid %#x type %d at %#jx\n",
5388 count, *rid, type, rman_get_start(res));
5390 /* Disable decoding via the CMD register before updating the BAR */
5391 cmd = pci_read_config(child, PCIR_COMMAND, 2);
5392 pci_write_config(child, PCIR_COMMAND,
5393 cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2);
5395 map = rman_get_start(res);
5396 pci_write_bar(child, pm, map);
5398 /* Restore the original value of the CMD register */
5399 pci_write_config(child, PCIR_COMMAND, cmd, 2);
5405 pci_alloc_multi_resource(device_t dev, device_t child, int type, int *rid,
5406 rman_res_t start, rman_res_t end, rman_res_t count, u_long num,
5409 struct pci_devinfo *dinfo;
5410 struct resource_list *rl;
5411 struct resource_list_entry *rle;
5412 struct resource *res;
5416 * Perform lazy resource allocation
5418 dinfo = device_get_ivars(child);
5419 rl = &dinfo->resources;
5422 #if defined(NEW_PCIB) && defined(PCI_RES_BUS)
5424 return (pci_alloc_secbus(dev, child, rid, start, end, count,
5429 * Can't alloc legacy interrupt once MSI messages have
5432 if (*rid == 0 && (cfg->msi.msi_alloc > 0 ||
5433 cfg->msix.msix_alloc > 0))
5437 * If the child device doesn't have an interrupt
5438 * routed and is deserving of an interrupt, try to
5441 if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) &&
5443 pci_assign_interrupt(dev, child, 0);
5445 case SYS_RES_IOPORT:
5446 case SYS_RES_MEMORY:
5449 * PCI-PCI bridge I/O window resources are not BARs.
5450 * For those allocations just pass the request up the
5453 if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE) {
5455 case PCIR_IOBASEL_1:
5456 case PCIR_MEMBASE_1:
5457 case PCIR_PMBASEL_1:
5459 * XXX: Should we bother creating a resource
5462 return (bus_generic_alloc_resource(dev, child,
5463 type, rid, start, end, count, flags));
5467 /* Reserve resources for this BAR if needed. */
5468 rle = resource_list_find(rl, type, *rid);
5470 res = pci_reserve_map(dev, child, type, rid, start, end,
5476 return (resource_list_alloc(rl, dev, child, type, rid,
5477 start, end, count, flags));
5481 pci_alloc_resource(device_t dev, device_t child, int type, int *rid,
5482 rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
5485 struct pci_devinfo *dinfo;
5488 if (device_get_parent(child) != dev)
5489 return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child,
5490 type, rid, start, end, count, flags));
5493 dinfo = device_get_ivars(child);
5494 if (dinfo->cfg.flags & PCICFG_VF) {
5496 /* VFs can't have I/O BARs. */
5497 case SYS_RES_IOPORT:
5499 case SYS_RES_MEMORY:
5500 return (pci_vf_alloc_mem_resource(dev, child, rid,
5501 start, end, count, flags));
5504 /* Fall through for other types of resource allocations. */
5508 return (pci_alloc_multi_resource(dev, child, type, rid, start, end,
5513 pci_release_resource(device_t dev, device_t child, int type, int rid,
5516 struct pci_devinfo *dinfo;
5517 struct resource_list *rl;
5520 if (device_get_parent(child) != dev)
5521 return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child,
5524 dinfo = device_get_ivars(child);
5528 if (dinfo->cfg.flags & PCICFG_VF) {
5530 /* VFs can't have I/O BARs. */
5531 case SYS_RES_IOPORT:
5533 case SYS_RES_MEMORY:
5534 return (pci_vf_release_mem_resource(dev, child, rid,
5538 /* Fall through for other types of resource allocations. */
5544 * PCI-PCI bridge I/O window resources are not BARs. For
5545 * those allocations just pass the request up the tree.
5547 if (cfg->hdrtype == PCIM_HDRTYPE_BRIDGE &&
5548 (type == SYS_RES_IOPORT || type == SYS_RES_MEMORY)) {
5550 case PCIR_IOBASEL_1:
5551 case PCIR_MEMBASE_1:
5552 case PCIR_PMBASEL_1:
5553 return (bus_generic_release_resource(dev, child, type,
5559 rl = &dinfo->resources;
5560 return (resource_list_release(rl, dev, child, type, rid, r));
5564 pci_activate_resource(device_t dev, device_t child, int type, int rid,
5567 struct pci_devinfo *dinfo;
5570 error = bus_generic_activate_resource(dev, child, type, rid, r);
5574 /* Enable decoding in the command register when activating BARs. */
5575 if (device_get_parent(child) == dev) {
5576 /* Device ROMs need their decoding explicitly enabled. */
5577 dinfo = device_get_ivars(child);
5578 if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5579 pci_write_bar(child, pci_find_bar(child, rid),
5580 rman_get_start(r) | PCIM_BIOS_ENABLE);
5582 case SYS_RES_IOPORT:
5583 case SYS_RES_MEMORY:
5584 error = PCI_ENABLE_IO(dev, child, type);
5592 pci_deactivate_resource(device_t dev, device_t child, int type,
5593 int rid, struct resource *r)
5595 struct pci_devinfo *dinfo;
5598 error = bus_generic_deactivate_resource(dev, child, type, rid, r);
5602 /* Disable decoding for device ROMs. */
5603 if (device_get_parent(child) == dev) {
5604 dinfo = device_get_ivars(child);
5605 if (type == SYS_RES_MEMORY && PCIR_IS_BIOS(&dinfo->cfg, rid))
5606 pci_write_bar(child, pci_find_bar(child, rid),
5613 pci_child_deleted(device_t dev, device_t child)
5615 struct resource_list_entry *rle;
5616 struct resource_list *rl;
5617 struct pci_devinfo *dinfo;
5619 dinfo = device_get_ivars(child);
5620 rl = &dinfo->resources;
5622 EVENTHANDLER_INVOKE(pci_delete_device, child);
5624 /* Turn off access to resources we're about to free */
5625 if (bus_child_present(child) != 0) {
5626 pci_write_config(child, PCIR_COMMAND, pci_read_config(child,
5627 PCIR_COMMAND, 2) & ~(PCIM_CMD_MEMEN | PCIM_CMD_PORTEN), 2);
5629 pci_disable_busmaster(child);
5632 /* Free all allocated resources */
5633 STAILQ_FOREACH(rle, rl, link) {
5635 if (rman_get_flags(rle->res) & RF_ACTIVE ||
5636 resource_list_busy(rl, rle->type, rle->rid)) {
5637 pci_printf(&dinfo->cfg,
5638 "Resource still owned, oops. "
5639 "(type=%d, rid=%d, addr=%lx)\n",
5640 rle->type, rle->rid,
5641 rman_get_start(rle->res));
5642 bus_release_resource(child, rle->type, rle->rid,
5645 resource_list_unreserve(rl, dev, child, rle->type,
5649 resource_list_free(rl);
5655 pci_delete_resource(device_t dev, device_t child, int type, int rid)
5657 struct pci_devinfo *dinfo;
5658 struct resource_list *rl;
5659 struct resource_list_entry *rle;
5661 if (device_get_parent(child) != dev)
5664 dinfo = device_get_ivars(child);
5665 rl = &dinfo->resources;
5666 rle = resource_list_find(rl, type, rid);
5671 if (rman_get_flags(rle->res) & RF_ACTIVE ||
5672 resource_list_busy(rl, type, rid)) {
5673 device_printf(dev, "delete_resource: "
5674 "Resource still owned by child, oops. "
5675 "(type=%d, rid=%d, addr=%jx)\n",
5676 type, rid, rman_get_start(rle->res));
5679 resource_list_unreserve(rl, dev, child, type, rid);
5681 resource_list_delete(rl, type, rid);
5684 struct resource_list *
5685 pci_get_resource_list (device_t dev, device_t child)
5687 struct pci_devinfo *dinfo = device_get_ivars(child);
5689 return (&dinfo->resources);
5693 bus_dma_tag_t dmar_get_dma_tag(device_t dev, device_t child);
5695 pci_get_dma_tag(device_t bus, device_t dev)
5698 struct pci_softc *sc;
5700 if (device_get_parent(dev) == bus) {
5701 /* try dmar and return if it works */
5702 tag = dmar_get_dma_tag(bus, dev);
5706 sc = device_get_softc(bus);
5707 tag = sc->sc_dma_tag;
5713 pci_get_dma_tag(device_t bus, device_t dev)
5715 struct pci_softc *sc = device_get_softc(bus);
5717 return (sc->sc_dma_tag);
5722 pci_read_config_method(device_t dev, device_t child, int reg, int width)
5724 struct pci_devinfo *dinfo = device_get_ivars(child);
5725 pcicfgregs *cfg = &dinfo->cfg;
5729 * SR-IOV VFs don't implement the VID or DID registers, so we have to
5730 * emulate them here.
5732 if (cfg->flags & PCICFG_VF) {
5733 if (reg == PCIR_VENDOR) {
5736 return (cfg->device << 16 | cfg->vendor);
5738 return (cfg->vendor);
5740 return (cfg->vendor & 0xff);
5742 return (0xffffffff);
5744 } else if (reg == PCIR_DEVICE) {
5746 /* Note that an unaligned 4-byte read is an error. */
5748 return (cfg->device);
5750 return (cfg->device & 0xff);
5752 return (0xffffffff);
5758 return (PCIB_READ_CONFIG(device_get_parent(dev),
5759 cfg->bus, cfg->slot, cfg->func, reg, width));
5763 pci_write_config_method(device_t dev, device_t child, int reg,
5764 uint32_t val, int width)
5766 struct pci_devinfo *dinfo = device_get_ivars(child);
5767 pcicfgregs *cfg = &dinfo->cfg;
5769 PCIB_WRITE_CONFIG(device_get_parent(dev),
5770 cfg->bus, cfg->slot, cfg->func, reg, val, width);
5774 pci_child_location_str_method(device_t dev, device_t child, char *buf,
5778 snprintf(buf, buflen, "slot=%d function=%d dbsf=pci%d:%d:%d:%d",
5779 pci_get_slot(child), pci_get_function(child), pci_get_domain(child),
5780 pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
5785 pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf,
5788 struct pci_devinfo *dinfo;
5791 dinfo = device_get_ivars(child);
5793 snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x "
5794 "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device,
5795 cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass,
5801 pci_assign_interrupt_method(device_t dev, device_t child)
5803 struct pci_devinfo *dinfo = device_get_ivars(child);
5804 pcicfgregs *cfg = &dinfo->cfg;
5806 return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child,
5811 pci_lookup(void *arg, const char *name, device_t *dev)
5815 int domain, bus, slot, func;
5821 * Accept pciconf-style selectors of either pciD:B:S:F or
5822 * pciB:S:F. In the latter case, the domain is assumed to
5825 if (strncmp(name, "pci", 3) != 0)
5827 val = strtol(name + 3, &end, 10);
5828 if (val < 0 || val > INT_MAX || *end != ':')
5831 val = strtol(end + 1, &end, 10);
5832 if (val < 0 || val > INT_MAX || *end != ':')
5835 val = strtol(end + 1, &end, 10);
5836 if (val < 0 || val > INT_MAX)
5840 val = strtol(end + 1, &end, 10);
5841 if (val < 0 || val > INT_MAX || *end != '\0')
5844 } else if (*end == '\0') {
5852 if (domain > PCI_DOMAINMAX || bus > PCI_BUSMAX || slot > PCI_SLOTMAX ||
5853 func > PCIE_ARI_FUNCMAX || (slot != 0 && func > PCI_FUNCMAX))
5856 *dev = pci_find_dbsf(domain, bus, slot, func);
5860 pci_modevent(module_t mod, int what, void *arg)
5862 static struct cdev *pci_cdev;
5863 static eventhandler_tag tag;
5867 STAILQ_INIT(&pci_devq);
5869 pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644,
5871 pci_load_vendor_data();
5872 tag = EVENTHANDLER_REGISTER(dev_lookup, pci_lookup, NULL,
5878 EVENTHANDLER_DEREGISTER(dev_lookup, tag);
5879 destroy_dev(pci_cdev);
5887 pci_cfg_restore_pcie(device_t dev, struct pci_devinfo *dinfo)
5889 #define WREG(n, v) pci_write_config(dev, pos + (n), (v), 2)
5890 struct pcicfg_pcie *cfg;
5893 cfg = &dinfo->cfg.pcie;
5894 pos = cfg->pcie_location;
5896 version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
5898 WREG(PCIER_DEVICE_CTL, cfg->pcie_device_ctl);
5900 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5901 cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
5902 cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
5903 WREG(PCIER_LINK_CTL, cfg->pcie_link_ctl);
5905 if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5906 (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
5907 (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
5908 WREG(PCIER_SLOT_CTL, cfg->pcie_slot_ctl);
5910 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
5911 cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
5912 WREG(PCIER_ROOT_CTL, cfg->pcie_root_ctl);
5915 WREG(PCIER_DEVICE_CTL2, cfg->pcie_device_ctl2);
5916 WREG(PCIER_LINK_CTL2, cfg->pcie_link_ctl2);
5917 WREG(PCIER_SLOT_CTL2, cfg->pcie_slot_ctl2);
5923 pci_cfg_restore_pcix(device_t dev, struct pci_devinfo *dinfo)
5925 pci_write_config(dev, dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND,
5926 dinfo->cfg.pcix.pcix_command, 2);
5930 pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo)
5934 * Restore the device to full power mode. We must do this
5935 * before we restore the registers because moving from D3 to
5936 * D0 will cause the chip's BARs and some other registers to
5937 * be reset to some unknown power on reset values. Cut down
5938 * the noise on boot by doing nothing if we are already in
5941 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0)
5942 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
5943 pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2);
5944 pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1);
5945 pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1);
5946 pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1);
5947 pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1);
5948 pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1);
5949 pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1);
5950 switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
5951 case PCIM_HDRTYPE_NORMAL:
5952 pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1);
5953 pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1);
5955 case PCIM_HDRTYPE_BRIDGE:
5956 pci_write_config(dev, PCIR_SECLAT_1,
5957 dinfo->cfg.bridge.br_seclat, 1);
5958 pci_write_config(dev, PCIR_SUBBUS_1,
5959 dinfo->cfg.bridge.br_subbus, 1);
5960 pci_write_config(dev, PCIR_SECBUS_1,
5961 dinfo->cfg.bridge.br_secbus, 1);
5962 pci_write_config(dev, PCIR_PRIBUS_1,
5963 dinfo->cfg.bridge.br_pribus, 1);
5964 pci_write_config(dev, PCIR_BRIDGECTL_1,
5965 dinfo->cfg.bridge.br_control, 2);
5967 case PCIM_HDRTYPE_CARDBUS:
5968 pci_write_config(dev, PCIR_SECLAT_2,
5969 dinfo->cfg.bridge.br_seclat, 1);
5970 pci_write_config(dev, PCIR_SUBBUS_2,
5971 dinfo->cfg.bridge.br_subbus, 1);
5972 pci_write_config(dev, PCIR_SECBUS_2,
5973 dinfo->cfg.bridge.br_secbus, 1);
5974 pci_write_config(dev, PCIR_PRIBUS_2,
5975 dinfo->cfg.bridge.br_pribus, 1);
5976 pci_write_config(dev, PCIR_BRIDGECTL_2,
5977 dinfo->cfg.bridge.br_control, 2);
5980 pci_restore_bars(dev);
5983 * Restore extended capabilities for PCI-Express and PCI-X
5985 if (dinfo->cfg.pcie.pcie_location != 0)
5986 pci_cfg_restore_pcie(dev, dinfo);
5987 if (dinfo->cfg.pcix.pcix_location != 0)
5988 pci_cfg_restore_pcix(dev, dinfo);
5990 /* Restore MSI and MSI-X configurations if they are present. */
5991 if (dinfo->cfg.msi.msi_location != 0)
5992 pci_resume_msi(dev);
5993 if (dinfo->cfg.msix.msix_location != 0)
5994 pci_resume_msix(dev);
5997 if (dinfo->cfg.iov != NULL)
5998 pci_iov_cfg_restore(dev, dinfo);
6003 pci_cfg_save_pcie(device_t dev, struct pci_devinfo *dinfo)
6005 #define RREG(n) pci_read_config(dev, pos + (n), 2)
6006 struct pcicfg_pcie *cfg;
6009 cfg = &dinfo->cfg.pcie;
6010 pos = cfg->pcie_location;
6012 cfg->pcie_flags = RREG(PCIER_FLAGS);
6014 version = cfg->pcie_flags & PCIEM_FLAGS_VERSION;
6016 cfg->pcie_device_ctl = RREG(PCIER_DEVICE_CTL);
6018 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6019 cfg->pcie_type == PCIEM_TYPE_ENDPOINT ||
6020 cfg->pcie_type == PCIEM_TYPE_LEGACY_ENDPOINT)
6021 cfg->pcie_link_ctl = RREG(PCIER_LINK_CTL);
6023 if (version > 1 || (cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6024 (cfg->pcie_type == PCIEM_TYPE_DOWNSTREAM_PORT &&
6025 (cfg->pcie_flags & PCIEM_FLAGS_SLOT))))
6026 cfg->pcie_slot_ctl = RREG(PCIER_SLOT_CTL);
6028 if (version > 1 || cfg->pcie_type == PCIEM_TYPE_ROOT_PORT ||
6029 cfg->pcie_type == PCIEM_TYPE_ROOT_EC)
6030 cfg->pcie_root_ctl = RREG(PCIER_ROOT_CTL);
6033 cfg->pcie_device_ctl2 = RREG(PCIER_DEVICE_CTL2);
6034 cfg->pcie_link_ctl2 = RREG(PCIER_LINK_CTL2);
6035 cfg->pcie_slot_ctl2 = RREG(PCIER_SLOT_CTL2);
6041 pci_cfg_save_pcix(device_t dev, struct pci_devinfo *dinfo)
6043 dinfo->cfg.pcix.pcix_command = pci_read_config(dev,
6044 dinfo->cfg.pcix.pcix_location + PCIXR_COMMAND, 2);
6048 pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate)
6054 * Some drivers apparently write to these registers w/o updating our
6055 * cached copy. No harm happens if we update the copy, so do so here
6056 * so we can restore them. The COMMAND register is modified by the
6057 * bus w/o updating the cache. This should represent the normally
6058 * writable portion of the 'defined' part of type 0/1/2 headers.
6060 dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2);
6061 dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2);
6062 dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2);
6063 dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1);
6064 dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1);
6065 dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1);
6066 dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1);
6067 dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1);
6068 dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1);
6069 dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1);
6070 dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1);
6071 switch (dinfo->cfg.hdrtype & PCIM_HDRTYPE) {
6072 case PCIM_HDRTYPE_NORMAL:
6073 dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2);
6074 dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2);
6075 dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1);
6076 dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1);
6078 case PCIM_HDRTYPE_BRIDGE:
6079 dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6081 dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6083 dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6085 dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6087 dinfo->cfg.bridge.br_control = pci_read_config(dev,
6088 PCIR_BRIDGECTL_1, 2);
6090 case PCIM_HDRTYPE_CARDBUS:
6091 dinfo->cfg.bridge.br_seclat = pci_read_config(dev,
6093 dinfo->cfg.bridge.br_subbus = pci_read_config(dev,
6095 dinfo->cfg.bridge.br_secbus = pci_read_config(dev,
6097 dinfo->cfg.bridge.br_pribus = pci_read_config(dev,
6099 dinfo->cfg.bridge.br_control = pci_read_config(dev,
6100 PCIR_BRIDGECTL_2, 2);
6101 dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_2, 2);
6102 dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_2, 2);
6106 if (dinfo->cfg.pcie.pcie_location != 0)
6107 pci_cfg_save_pcie(dev, dinfo);
6109 if (dinfo->cfg.pcix.pcix_location != 0)
6110 pci_cfg_save_pcix(dev, dinfo);
6113 if (dinfo->cfg.iov != NULL)
6114 pci_iov_cfg_save(dev, dinfo);
6118 * don't set the state for display devices, base peripherals and
6119 * memory devices since bad things happen when they are powered down.
6120 * We should (a) have drivers that can easily detach and (b) use
6121 * generic drivers for these devices so that some device actually
6122 * attaches. We need to make sure that when we implement (a) we don't
6123 * power the device down on a reattach.
6125 cls = pci_get_class(dev);
6128 switch (pci_do_power_nodriver)
6130 case 0: /* NO powerdown at all */
6132 case 1: /* Conservative about what to power down */
6133 if (cls == PCIC_STORAGE)
6136 case 2: /* Aggressive about what to power down */
6137 if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY ||
6138 cls == PCIC_BASEPERIPH)
6141 case 3: /* Power down everything */
6145 * PCI spec says we can only go into D3 state from D0 state.
6146 * Transition from D[12] into D0 before going to D3 state.
6148 ps = pci_get_powerstate(dev);
6149 if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6150 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6151 if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3)
6152 pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6155 /* Wrapper APIs suitable for device driver use. */
6157 pci_save_state(device_t dev)
6159 struct pci_devinfo *dinfo;
6161 dinfo = device_get_ivars(dev);
6162 pci_cfg_save(dev, dinfo, 0);
6166 pci_restore_state(device_t dev)
6168 struct pci_devinfo *dinfo;
6170 dinfo = device_get_ivars(dev);
6171 pci_cfg_restore(dev, dinfo);
6175 pci_get_id_method(device_t dev, device_t child, enum pci_id_type type,
6179 return (PCIB_GET_ID(device_get_parent(dev), child, type, id));
6182 /* Find the upstream port of a given PCI device in a root complex. */
6184 pci_find_pcie_root_port(device_t dev)
6186 struct pci_devinfo *dinfo;
6187 devclass_t pci_class;
6190 pci_class = devclass_find("pci");
6191 KASSERT(device_get_devclass(device_get_parent(dev)) == pci_class,
6192 ("%s: non-pci device %s", __func__, device_get_nameunit(dev)));
6195 * Walk the bridge hierarchy until we find a PCI-e root
6196 * port or a non-PCI device.
6199 bus = device_get_parent(dev);
6200 KASSERT(bus != NULL, ("%s: null parent of %s", __func__,
6201 device_get_nameunit(dev)));
6203 pcib = device_get_parent(bus);
6204 KASSERT(pcib != NULL, ("%s: null bridge of %s", __func__,
6205 device_get_nameunit(bus)));
6208 * pcib's parent must be a PCI bus for this to be a
6211 if (device_get_devclass(device_get_parent(pcib)) != pci_class)
6214 dinfo = device_get_ivars(pcib);
6215 if (dinfo->cfg.pcie.pcie_location != 0 &&
6216 dinfo->cfg.pcie.pcie_type == PCIEM_TYPE_ROOT_PORT)
6224 * Wait for pending transactions to complete on a PCI-express function.
6226 * The maximum delay is specified in milliseconds in max_delay. Note
6227 * that this function may sleep.
6229 * Returns true if the function is idle and false if the timeout is
6230 * exceeded. If dev is not a PCI-express function, this returns true.
6233 pcie_wait_for_pending_transactions(device_t dev, u_int max_delay)
6235 struct pci_devinfo *dinfo = device_get_ivars(dev);
6239 cap = dinfo->cfg.pcie.pcie_location;
6243 sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6244 while (sta & PCIEM_STA_TRANSACTION_PND) {
6248 /* Poll once every 100 milliseconds up to the timeout. */
6249 if (max_delay > 100) {
6250 pause_sbt("pcietp", 100 * SBT_1MS, 0, C_HARDCLOCK);
6253 pause_sbt("pcietp", max_delay * SBT_1MS, 0,
6257 sta = pci_read_config(dev, cap + PCIER_DEVICE_STA, 2);
6264 * Determine the maximum Completion Timeout in microseconds.
6266 * For non-PCI-express functions this returns 0.
6269 pcie_get_max_completion_timeout(device_t dev)
6271 struct pci_devinfo *dinfo = device_get_ivars(dev);
6274 cap = dinfo->cfg.pcie.pcie_location;
6279 * Functions using the 1.x spec use the default timeout range of
6280 * 50 microseconds to 50 milliseconds. Functions that do not
6281 * support programmable timeouts also use this range.
6283 if ((dinfo->cfg.pcie.pcie_flags & PCIEM_FLAGS_VERSION) < 2 ||
6284 (pci_read_config(dev, cap + PCIER_DEVICE_CAP2, 4) &
6285 PCIEM_CAP2_COMP_TIMO_RANGES) == 0)
6288 switch (pci_read_config(dev, cap + PCIER_DEVICE_CTL2, 2) &
6289 PCIEM_CTL2_COMP_TIMO_VAL) {
6290 case PCIEM_CTL2_COMP_TIMO_100US:
6292 case PCIEM_CTL2_COMP_TIMO_10MS:
6294 case PCIEM_CTL2_COMP_TIMO_55MS:
6296 case PCIEM_CTL2_COMP_TIMO_210MS:
6297 return (210 * 1000);
6298 case PCIEM_CTL2_COMP_TIMO_900MS:
6299 return (900 * 1000);
6300 case PCIEM_CTL2_COMP_TIMO_3500MS:
6301 return (3500 * 1000);
6302 case PCIEM_CTL2_COMP_TIMO_13S:
6303 return (13 * 1000 * 1000);
6304 case PCIEM_CTL2_COMP_TIMO_64S:
6305 return (64 * 1000 * 1000);
6312 * Perform a Function Level Reset (FLR) on a device.
6314 * This function first waits for any pending transactions to complete
6315 * within the timeout specified by max_delay. If transactions are
6316 * still pending, the function will return false without attempting a
6319 * If dev is not a PCI-express function or does not support FLR, this
6320 * function returns false.
6322 * Note that no registers are saved or restored. The caller is
6323 * responsible for saving and restoring any registers including
6324 * PCI-standard registers via pci_save_state() and
6325 * pci_restore_state().
6328 pcie_flr(device_t dev, u_int max_delay, bool force)
6330 struct pci_devinfo *dinfo = device_get_ivars(dev);
6335 cap = dinfo->cfg.pcie.pcie_location;
6339 if (!(pci_read_config(dev, cap + PCIER_DEVICE_CAP, 4) & PCIEM_CAP_FLR))
6343 * Disable busmastering to prevent generation of new
6344 * transactions while waiting for the device to go idle. If
6345 * the idle timeout fails, the command register is restored
6346 * which will re-enable busmastering.
6348 cmd = pci_read_config(dev, PCIR_COMMAND, 2);
6349 pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCIM_CMD_BUSMASTEREN), 2);
6350 if (!pcie_wait_for_pending_transactions(dev, max_delay)) {
6352 pci_write_config(dev, PCIR_COMMAND, cmd, 2);
6355 pci_printf(&dinfo->cfg,
6356 "Resetting with transactions pending after %d ms\n",
6360 * Extend the post-FLR delay to cover the maximum
6361 * Completion Timeout delay of anything in flight
6362 * during the FLR delay. Enforce a minimum delay of
6365 compl_delay = pcie_get_max_completion_timeout(dev) / 1000;
6366 if (compl_delay < 10)
6371 /* Initiate the reset. */
6372 ctl = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2);
6373 pci_write_config(dev, cap + PCIER_DEVICE_CTL, ctl |
6374 PCIEM_CTL_INITIATE_FLR, 2);
6376 /* Wait for 100ms. */
6377 pause_sbt("pcieflr", (100 + compl_delay) * SBT_1MS, 0, C_HARDCLOCK);
6379 if (pci_read_config(dev, cap + PCIER_DEVICE_STA, 2) &
6380 PCIEM_STA_TRANSACTION_PND)
6381 pci_printf(&dinfo->cfg, "Transactions pending after FLR!\n");
6386 * Attempt a power-management reset by cycling the device in/out of D3
6387 * state. PCI spec says we can only go into D3 state from D0 state.
6388 * Transition from D[12] into D0 before going to D3 state.
6391 pci_power_reset(device_t dev)
6395 ps = pci_get_powerstate(dev);
6396 if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3)
6397 pci_set_powerstate(dev, PCI_POWERSTATE_D0);
6398 pci_set_powerstate(dev, PCI_POWERSTATE_D3);
6399 pci_set_powerstate(dev, ps);
6404 * Try link drop and retrain of the downstream port of upstream
6405 * switch, for PCIe. According to the PCIe 3.0 spec 6.6.1, this must
6406 * cause Conventional Hot reset of the device in the slot.
6407 * Alternative, for PCIe, could be the secondary bus reset initiatied
6408 * on the upstream switch PCIR_BRIDGECTL_1, bit 6.
6411 pcie_link_reset(device_t port, int pcie_location)
6415 v = pci_read_config(port, pcie_location + PCIER_LINK_CTL, 2);
6416 v |= PCIEM_LINK_CTL_LINK_DIS;
6417 pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6418 pause_sbt("pcier1", mstosbt(20), 0, 0);
6419 v &= ~PCIEM_LINK_CTL_LINK_DIS;
6420 v |= PCIEM_LINK_CTL_RETRAIN_LINK;
6421 pci_write_config(port, pcie_location + PCIER_LINK_CTL, v, 2);
6422 pause_sbt("pcier2", mstosbt(100), 0, 0); /* 100 ms */
6423 v = pci_read_config(port, pcie_location + PCIER_LINK_STA, 2);
6424 return ((v & PCIEM_LINK_STA_TRAINING) != 0 ? ETIMEDOUT : 0);
6428 pci_reset_post(device_t dev, device_t child)
6431 if (dev == device_get_parent(child))
6432 pci_restore_state(child);
6437 pci_reset_prepare(device_t dev, device_t child)
6440 if (dev == device_get_parent(child))
6441 pci_save_state(child);
6446 pci_reset_child(device_t dev, device_t child, int flags)
6450 if (dev == NULL || device_get_parent(child) != dev)
6452 if ((flags & DEVF_RESET_DETACH) != 0) {
6453 error = device_get_state(child) == DS_ATTACHED ?
6454 device_detach(child) : 0;
6456 error = BUS_SUSPEND_CHILD(dev, child);
6459 if (!pcie_flr(child, 1000, false)) {
6460 error = BUS_RESET_PREPARE(dev, child);
6462 pci_power_reset(child);
6463 BUS_RESET_POST(dev, child);
6465 if ((flags & DEVF_RESET_DETACH) != 0)
6466 device_probe_and_attach(child);
6468 BUS_RESUME_CHILD(dev, child);
6473 const struct pci_device_table *
6474 pci_match_device(device_t child, const struct pci_device_table *id, size_t nelt)
6477 uint16_t vendor, device, subvendor, subdevice, class, subclass, revid;
6479 vendor = pci_get_vendor(child);
6480 device = pci_get_device(child);
6481 subvendor = pci_get_subvendor(child);
6482 subdevice = pci_get_subdevice(child);
6483 class = pci_get_class(child);
6484 subclass = pci_get_subclass(child);
6485 revid = pci_get_revid(child);
6486 while (nelt-- > 0) {
6488 if (id->match_flag_vendor)
6489 match &= vendor == id->vendor;
6490 if (id->match_flag_device)
6491 match &= device == id->device;
6492 if (id->match_flag_subvendor)
6493 match &= subvendor == id->subvendor;
6494 if (id->match_flag_subdevice)
6495 match &= subdevice == id->subdevice;
6496 if (id->match_flag_class)
6497 match &= class == id->class_id;
6498 if (id->match_flag_subclass)
6499 match &= subclass == id->subclass;
6500 if (id->match_flag_revid)
6501 match &= revid == id->revid;
6510 pci_print_faulted_dev_name(const struct pci_devinfo *dinfo)
6512 const char *dev_name;
6515 dev = dinfo->cfg.dev;
6516 printf("pci%d:%d:%d:%d", dinfo->cfg.domain, dinfo->cfg.bus,
6517 dinfo->cfg.slot, dinfo->cfg.func);
6518 dev_name = device_get_name(dev);
6519 if (dev_name != NULL)
6520 printf(" (%s%d)", dev_name, device_get_unit(dev));
6524 pci_print_faulted_dev(void)
6526 struct pci_devinfo *dinfo;
6532 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6533 dev = dinfo->cfg.dev;
6534 status = pci_read_config(dev, PCIR_STATUS, 2);
6535 status &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6536 PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6537 PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6539 pci_print_faulted_dev_name(dinfo);
6540 printf(" error 0x%04x\n", status);
6542 if (dinfo->cfg.pcie.pcie_location != 0) {
6543 status = pci_read_config(dev,
6544 dinfo->cfg.pcie.pcie_location +
6545 PCIER_DEVICE_STA, 2);
6546 if ((status & (PCIEM_STA_CORRECTABLE_ERROR |
6547 PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
6548 PCIEM_STA_UNSUPPORTED_REQ)) != 0) {
6549 pci_print_faulted_dev_name(dinfo);
6550 printf(" PCIe DEVCTL 0x%04x DEVSTA 0x%04x\n",
6551 pci_read_config(dev,
6552 dinfo->cfg.pcie.pcie_location +
6553 PCIER_DEVICE_CTL, 2),
6557 if (pci_find_extcap(dev, PCIZ_AER, &aer) == 0) {
6558 r1 = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6559 r2 = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6560 if (r1 != 0 || r2 != 0) {
6561 pci_print_faulted_dev_name(dinfo);
6562 printf(" AER UC 0x%08x Mask 0x%08x Svr 0x%08x\n"
6563 " COR 0x%08x Mask 0x%08x Ctl 0x%08x\n",
6564 r1, pci_read_config(dev, aer +
6565 PCIR_AER_UC_MASK, 4),
6566 pci_read_config(dev, aer +
6567 PCIR_AER_UC_SEVERITY, 4),
6568 r2, pci_read_config(dev, aer +
6569 PCIR_AER_COR_MASK, 4),
6570 pci_read_config(dev, aer +
6571 PCIR_AER_CAP_CONTROL, 4));
6572 for (i = 0; i < 4; i++) {
6573 r1 = pci_read_config(dev, aer +
6574 PCIR_AER_HEADER_LOG + i * 4, 4);
6575 printf(" HL%d: 0x%08x\n", i, r1);
6583 DB_SHOW_COMMAND(pcierr, pci_print_faulted_dev_db)
6586 pci_print_faulted_dev();
6590 db_clear_pcie_errors(const struct pci_devinfo *dinfo)
6596 dev = dinfo->cfg.dev;
6597 r = pci_read_config(dev, dinfo->cfg.pcie.pcie_location +
6598 PCIER_DEVICE_STA, 2);
6599 pci_write_config(dev, dinfo->cfg.pcie.pcie_location +
6600 PCIER_DEVICE_STA, r, 2);
6602 if (pci_find_extcap(dev, PCIZ_AER, &aer) != 0)
6604 r = pci_read_config(dev, aer + PCIR_AER_UC_STATUS, 4);
6606 pci_write_config(dev, aer + PCIR_AER_UC_STATUS, r, 4);
6607 r = pci_read_config(dev, aer + PCIR_AER_COR_STATUS, 4);
6609 pci_write_config(dev, aer + PCIR_AER_COR_STATUS, r, 4);
6612 DB_COMMAND(pci_clearerr, db_pci_clearerr)
6614 struct pci_devinfo *dinfo;
6616 uint16_t status, status1;
6618 STAILQ_FOREACH(dinfo, &pci_devq, pci_links) {
6619 dev = dinfo->cfg.dev;
6620 status1 = status = pci_read_config(dev, PCIR_STATUS, 2);
6621 status1 &= PCIM_STATUS_MDPERR | PCIM_STATUS_STABORT |
6622 PCIM_STATUS_RTABORT | PCIM_STATUS_RMABORT |
6623 PCIM_STATUS_SERR | PCIM_STATUS_PERR;
6626 pci_write_config(dev, PCIR_STATUS, status, 2);
6628 if (dinfo->cfg.pcie.pcie_location != 0)
6629 db_clear_pcie_errors(dinfo);