1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/acpi.h>
4 #include <linux/init.h>
7 #include <linux/slab.h>
8 #include <linux/pci-acpi.h>
10 #include <asm/pci_x86.h>
11 #include <asm/e820/api.h>
13 struct pci_root_info {
14 struct acpi_pci_root_info common;
15 struct pci_sysdata sd;
16 #ifdef CONFIG_PCI_MMCONFIG
23 static bool pci_use_e820 = true;
24 static bool pci_use_crs = true;
25 static bool pci_ignore_seg;
27 static int __init set_use_crs(const struct dmi_system_id *id)
33 static int __init set_nouse_crs(const struct dmi_system_id *id)
39 static int __init set_ignore_seg(const struct dmi_system_id *id)
41 printk(KERN_INFO "PCI: %s detected: ignoring ACPI _SEG\n", id->ident);
42 pci_ignore_seg = true;
46 static int __init set_no_e820(const struct dmi_system_id *id)
48 printk(KERN_INFO "PCI: %s detected: not clipping E820 regions from _CRS\n",
54 static const struct dmi_system_id pci_crs_quirks[] __initconst = {
55 /* http://bugzilla.kernel.org/show_bug.cgi?id=14183 */
57 .callback = set_use_crs,
58 .ident = "IBM System x3800",
60 DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
61 DMI_MATCH(DMI_PRODUCT_NAME, "x3800"),
64 /* https://bugzilla.kernel.org/show_bug.cgi?id=16007 */
65 /* 2006 AMD HT/VIA system with two host bridges */
67 .callback = set_use_crs,
68 .ident = "ASRock ALiveSATA2-GLAN",
70 DMI_MATCH(DMI_PRODUCT_NAME, "ALiveSATA2-GLAN"),
73 /* https://bugzilla.kernel.org/show_bug.cgi?id=30552 */
74 /* 2006 AMD HT/VIA system with two host bridges */
76 .callback = set_use_crs,
77 .ident = "ASUS M2V-MX SE",
79 DMI_MATCH(DMI_BOARD_VENDOR, "ASUSTeK Computer INC."),
80 DMI_MATCH(DMI_BOARD_NAME, "M2V-MX SE"),
81 DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
84 /* https://bugzilla.kernel.org/show_bug.cgi?id=42619 */
86 .callback = set_use_crs,
87 .ident = "MSI MS-7253",
89 DMI_MATCH(DMI_BOARD_VENDOR, "MICRO-STAR INTERNATIONAL CO., LTD"),
90 DMI_MATCH(DMI_BOARD_NAME, "MS-7253"),
91 DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies, LTD"),
94 /* https://bugs.launchpad.net/ubuntu/+source/alsa-driver/+bug/931368 */
95 /* https://bugs.launchpad.net/ubuntu/+source/alsa-driver/+bug/1033299 */
97 .callback = set_use_crs,
98 .ident = "Foxconn K8M890-8237A",
100 DMI_MATCH(DMI_BOARD_VENDOR, "Foxconn"),
101 DMI_MATCH(DMI_BOARD_NAME, "K8M890-8237A"),
102 DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies, LTD"),
106 /* Now for the blacklist.. */
108 /* https://bugzilla.redhat.com/show_bug.cgi?id=769657 */
110 .callback = set_nouse_crs,
111 .ident = "Dell Studio 1557",
113 DMI_MATCH(DMI_BOARD_VENDOR, "Dell Inc."),
114 DMI_MATCH(DMI_PRODUCT_NAME, "Studio 1557"),
115 DMI_MATCH(DMI_BIOS_VERSION, "A09"),
118 /* https://bugzilla.redhat.com/show_bug.cgi?id=769657 */
120 .callback = set_nouse_crs,
121 .ident = "Thinkpad SL510",
123 DMI_MATCH(DMI_BOARD_VENDOR, "LENOVO"),
124 DMI_MATCH(DMI_BOARD_NAME, "2847DFG"),
125 DMI_MATCH(DMI_BIOS_VERSION, "6JET85WW (1.43 )"),
128 /* https://bugzilla.kernel.org/show_bug.cgi?id=42606 */
130 .callback = set_nouse_crs,
131 .ident = "Supermicro X8DTH",
133 DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
134 DMI_MATCH(DMI_PRODUCT_NAME, "X8DTH-i/6/iF/6F"),
135 DMI_MATCH(DMI_BIOS_VERSION, "2.0a"),
139 /* https://bugzilla.kernel.org/show_bug.cgi?id=15362 */
141 .callback = set_ignore_seg,
142 .ident = "HP xw9300",
144 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
145 DMI_MATCH(DMI_PRODUCT_NAME, "HP xw9300 Workstation"),
150 * Many Lenovo models with "IIL" in their DMI_PRODUCT_VERSION have
151 * an E820 reserved region that covers the entire 32-bit host
152 * bridge memory window from _CRS. Using the E820 region to clip
153 * _CRS means no space is available for hot-added or uninitialized
154 * PCI devices. This typically breaks I2C controllers for touchpads
155 * and hot-added Thunderbolt devices. See the commit log for
156 * models known to require this quirk and related bug reports.
159 .callback = set_no_e820,
160 .ident = "Lenovo *IIL* product version",
162 DMI_MATCH(DMI_SYS_VENDOR, "LENOVO"),
163 DMI_MATCH(DMI_PRODUCT_VERSION, "IIL"),
168 * The Acer Spin 5 (SP513-54N) has the same E820 reservation covering
169 * the entire _CRS 32-bit window issue as the Lenovo *IIL* models.
170 * See https://bugs.launchpad.net/bugs/1884232
173 .callback = set_no_e820,
174 .ident = "Acer Spin 5 (SP513-54N)",
176 DMI_MATCH(DMI_SYS_VENDOR, "Acer"),
177 DMI_MATCH(DMI_PRODUCT_NAME, "Spin SP513-54N"),
182 * Clevo X170KM-G barebones have the same E820 reservation covering
183 * the entire _CRS 32-bit window issue as the Lenovo *IIL* models.
184 * See https://bugzilla.kernel.org/show_bug.cgi?id=214259
187 .callback = set_no_e820,
188 .ident = "Clevo X170KM-G Barebone",
190 DMI_MATCH(DMI_BOARD_NAME, "X170KM-G"),
196 void __init pci_acpi_crs_quirks(void)
198 int year = dmi_get_bios_year();
200 if (year >= 0 && year < 2008 && iomem_resource.end <= 0xffffffff)
204 * Some firmware includes unusable space (host bridge registers,
205 * hidden PCI device BARs, etc) in PCI host bridge _CRS. This is a
206 * firmware defect, and 4dc2287c1805 ("x86: avoid E820 regions when
207 * allocating address space") has clipped out the unusable space in
210 * But other firmware supplies E820 reserved regions that cover
211 * entire _CRS windows, so clipping throws away the entire window,
212 * leaving none for hot-added or uninitialized devices. These E820
213 * entries are probably *not* a firmware defect, so disable the
214 * clipping by default for post-2022 machines.
216 * We already have quirks to disable clipping for pre-2023
217 * machines, and we'll likely need quirks to *enable* clipping for
218 * post-2022 machines that incorrectly include unusable space in
222 pci_use_e820 = false;
224 dmi_check_system(pci_crs_quirks);
227 * If the user specifies "pci=use_crs" or "pci=nocrs" explicitly, that
228 * takes precedence over anything we figured out above.
230 if (pci_probe & PCI_ROOT_NO_CRS)
232 else if (pci_probe & PCI_USE__CRS)
235 printk(KERN_INFO "PCI: %s host bridge windows from ACPI; "
236 "if necessary, use \"pci=%s\" and report a bug\n",
237 pci_use_crs ? "Using" : "Ignoring",
238 pci_use_crs ? "nocrs" : "use_crs");
240 /* "pci=use_e820"/"pci=no_e820" on the kernel cmdline takes precedence */
241 if (pci_probe & PCI_NO_E820)
242 pci_use_e820 = false;
243 else if (pci_probe & PCI_USE_E820)
246 printk(KERN_INFO "PCI: %s E820 reservations for host bridge windows\n",
247 pci_use_e820 ? "Using" : "Ignoring");
248 if (pci_probe & (PCI_NO_E820 | PCI_USE_E820))
249 printk(KERN_INFO "PCI: Please notify linux-pci@vger.kernel.org so future kernels can this automatically\n");
252 #ifdef CONFIG_PCI_MMCONFIG
253 static int check_segment(u16 seg, struct device *dev, char *estr)
257 "%s can't access PCI configuration "
258 "space under this host bridge.\n",
264 * Failure in adding MMCFG information is not fatal,
265 * just can't access extended configuration space of
266 * devices under this host bridge.
269 "%s can't access extended PCI configuration "
270 "space under this bridge.\n",
276 static int setup_mcfg_map(struct acpi_pci_root_info *ci)
279 struct pci_root_info *info;
280 struct acpi_pci_root *root = ci->root;
281 struct device *dev = &ci->bridge->dev;
283 info = container_of(ci, struct pci_root_info, common);
284 info->start_bus = (u8)root->secondary.start;
285 info->end_bus = (u8)root->secondary.end;
286 info->mcfg_added = false;
287 seg = info->sd.domain;
289 /* return success if MMCFG is not in use */
290 if (raw_pci_ext_ops && raw_pci_ext_ops != &pci_mmcfg)
293 if (!(pci_probe & PCI_PROBE_MMCONF))
294 return check_segment(seg, dev, "MMCONFIG is disabled,");
296 result = pci_mmconfig_insert(dev, seg, info->start_bus, info->end_bus,
299 /* enable MMCFG if it hasn't been enabled yet */
300 if (raw_pci_ext_ops == NULL)
301 raw_pci_ext_ops = &pci_mmcfg;
302 info->mcfg_added = true;
303 } else if (result != -EEXIST)
304 return check_segment(seg, dev,
305 "fail to add MMCONFIG information,");
310 static void teardown_mcfg_map(struct acpi_pci_root_info *ci)
312 struct pci_root_info *info;
314 info = container_of(ci, struct pci_root_info, common);
315 if (info->mcfg_added) {
316 pci_mmconfig_delete(info->sd.domain,
317 info->start_bus, info->end_bus);
318 info->mcfg_added = false;
322 static int setup_mcfg_map(struct acpi_pci_root_info *ci)
327 static void teardown_mcfg_map(struct acpi_pci_root_info *ci)
332 static int pci_acpi_root_get_node(struct acpi_pci_root *root)
334 int busnum = root->secondary.start;
335 struct acpi_device *device = root->device;
336 int node = acpi_get_node(device->handle);
338 if (node == NUMA_NO_NODE) {
339 node = x86_pci_root_bus_node(busnum);
340 if (node != 0 && node != NUMA_NO_NODE)
341 dev_info(&device->dev, FW_BUG "no _PXM; falling back to node %d from hardware (may be inconsistent with ACPI node numbers)\n",
344 if (node != NUMA_NO_NODE && !node_online(node))
350 static int pci_acpi_root_init_info(struct acpi_pci_root_info *ci)
352 return setup_mcfg_map(ci);
355 static void pci_acpi_root_release_info(struct acpi_pci_root_info *ci)
357 teardown_mcfg_map(ci);
358 kfree(container_of(ci, struct pci_root_info, common));
362 * An IO port or MMIO resource assigned to a PCI host bridge may be
363 * consumed by the host bridge itself or available to its child
364 * bus/devices. The ACPI specification defines a bit (Producer/Consumer)
365 * to tell whether the resource is consumed by the host bridge itself,
366 * but firmware hasn't used that bit consistently, so we can't rely on it.
368 * On x86 and IA64 platforms, all IO port and MMIO resources are assumed
369 * to be available to child bus/devices except one special case:
370 * IO port [0xCF8-0xCFF] is consumed by the host bridge itself
371 * to access PCI configuration space.
373 * So explicitly filter out PCI CFG IO ports[0xCF8-0xCFF].
375 static bool resource_is_pcicfg_ioport(struct resource *res)
377 return (res->flags & IORESOURCE_IO) &&
378 res->start == 0xCF8 && res->end == 0xCFF;
381 static int pci_acpi_root_prepare_resources(struct acpi_pci_root_info *ci)
383 struct acpi_device *device = ci->bridge;
384 int busnum = ci->root->secondary.start;
385 struct resource_entry *entry, *tmp;
388 status = acpi_pci_probe_root_resources(ci);
391 resource_list_for_each_entry(entry, &ci->resources)
392 remove_e820_regions(&device->dev, entry->res);
396 resource_list_for_each_entry_safe(entry, tmp, &ci->resources)
397 if (resource_is_pcicfg_ioport(entry->res))
398 resource_list_destroy_entry(entry);
402 resource_list_for_each_entry_safe(entry, tmp, &ci->resources) {
403 dev_printk(KERN_DEBUG, &device->dev,
404 "host bridge window %pR (ignored)\n", entry->res);
405 resource_list_destroy_entry(entry);
407 x86_pci_root_bus_resources(busnum, &ci->resources);
412 static struct acpi_pci_root_ops acpi_pci_root_ops = {
413 .pci_ops = &pci_root_ops,
414 .init_info = pci_acpi_root_init_info,
415 .release_info = pci_acpi_root_release_info,
416 .prepare_resources = pci_acpi_root_prepare_resources,
419 struct pci_bus *pci_acpi_scan_root(struct acpi_pci_root *root)
421 int domain = root->segment;
422 int busnum = root->secondary.start;
423 int node = pci_acpi_root_get_node(root);
427 root->segment = domain = 0;
429 if (domain && !pci_domains_supported) {
430 printk(KERN_WARNING "pci_bus %04x:%02x: "
431 "ignored (multiple domains not supported)\n",
436 bus = pci_find_bus(domain, busnum);
439 * If the desired bus has been scanned already, replace
442 struct pci_sysdata sd = {
445 .companion = root->device
448 memcpy(bus->sysdata, &sd, sizeof(sd));
450 struct pci_root_info *info;
452 info = kzalloc(sizeof(*info), GFP_KERNEL);
454 dev_err(&root->device->dev,
455 "pci_bus %04x:%02x: ignored (out of memory)\n",
458 info->sd.domain = domain;
459 info->sd.node = node;
460 info->sd.companion = root->device;
461 bus = acpi_pci_root_create(root, &acpi_pci_root_ops,
462 &info->common, &info->sd);
466 /* After the PCI-E bus has been walked and all devices discovered,
467 * configure any settings of the fabric that might be necessary.
470 struct pci_bus *child;
471 list_for_each_entry(child, &bus->children, node)
472 pcie_bus_configure_settings(child);
478 int pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
481 * We pass NULL as parent to pci_create_root_bus(), so if it is not NULL
482 * here, pci_create_root_bus() has been called by someone else and
483 * sysdata is likely to be different from what we expect. Let it go in
486 if (!bridge->dev.parent) {
487 struct pci_sysdata *sd = bridge->bus->sysdata;
488 ACPI_COMPANION_SET(&bridge->dev, sd->companion);
493 int __init pci_acpi_init(void)
495 struct pci_dev *dev = NULL;
500 printk(KERN_INFO "PCI: Using ACPI for IRQ routing\n");
501 acpi_irq_penalty_init();
502 pcibios_enable_irq = acpi_pci_irq_enable;
503 pcibios_disable_irq = acpi_pci_irq_disable;
504 x86_init.pci.init_irq = x86_init_noop;
508 * PCI IRQ routing is set up by pci_enable_device(), but we
509 * also do it here in case there are still broken drivers that
510 * don't use pci_enable_device().
512 printk(KERN_INFO "PCI: Routing PCI interrupts for all devices because \"pci=routeirq\" specified\n");
513 for_each_pci_dev(dev)
514 acpi_pci_irq_enable(dev);