1 // SPDX-License-Identifier: GPL-2.0-only
3 * Xen PCI - handle PCI (INTx) and MSI infrastructure calls for PV, HVM and
4 * initial domain support. We also handle the DSDT _PRT callbacks for GSI's
5 * used in HVM and initial domain mode (PV does not parse ACPI, so it has no
6 * concept of GSIs). Under PV we hook under the pnbbios API for IRQs and
7 * 0xcf8 PCI configuration read/write.
13 #include <linux/export.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 #include <linux/acpi.h>
19 #include <asm/io_apic.h>
20 #include <asm/pci_x86.h>
22 #include <asm/xen/hypervisor.h>
24 #include <xen/features.h>
25 #include <xen/events.h>
27 #include <asm/xen/pci.h>
28 #include <asm/xen/cpuid.h>
31 #include <asm/i8259.h>
33 static int xen_pcifront_enable_irq(struct pci_dev *dev)
40 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
42 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
46 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
49 if (gsi < nr_legacy_irqs())
52 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
54 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
60 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
65 static int xen_register_pirq(u32 gsi, int triggering, bool set_pirq)
67 int rc, pirq = -1, irq;
68 struct physdev_map_pirq map_irq;
72 irq = xen_irq_from_gsi(gsi);
79 map_irq.domid = DOMID_SELF;
80 map_irq.type = MAP_PIRQ_TYPE_GSI;
84 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
86 printk(KERN_WARNING "xen map irq failed %d\n", rc);
90 if (triggering == ACPI_EDGE_SENSITIVE) {
95 name = "ioapic-level";
98 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
102 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
107 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
108 int trigger, int polarity)
110 if (!xen_hvm_domain())
113 return xen_register_pirq(gsi, trigger,
114 false /* no mapping of GSI to PIRQ */);
117 #ifdef CONFIG_XEN_PV_DOM0
118 static int xen_register_gsi(u32 gsi, int triggering, int polarity)
121 struct physdev_setup_gsi setup_gsi;
123 if (!xen_pv_domain())
126 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
127 gsi, triggering, polarity);
129 irq = xen_register_pirq(gsi, triggering, true);
132 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
133 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
135 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
137 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
139 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
146 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
147 int trigger, int polarity)
149 return xen_register_gsi(gsi, trigger, polarity);
154 #if defined(CONFIG_PCI_MSI)
155 #include <linux/msi.h>
157 struct xen_pci_frontend_ops *xen_pci_frontend;
158 EXPORT_SYMBOL_GPL(xen_pci_frontend);
161 int (*setup_msi_irqs)(struct pci_dev *dev, int nvec, int type);
162 void (*teardown_msi_irqs)(struct pci_dev *dev);
165 static struct xen_msi_ops xen_msi_ops __ro_after_init;
167 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
170 struct msi_desc *msidesc;
173 if (type == PCI_CAP_ID_MSI && nvec > 1)
176 v = kcalloc(max(1, nvec), sizeof(int), GFP_KERNEL);
180 if (type == PCI_CAP_ID_MSIX)
181 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
183 ret = xen_pci_frontend_enable_msi(dev, v);
187 msi_for_each_desc(msidesc, &dev->dev, MSI_DESC_NOTASSOCIATED) {
188 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i],
189 (type == PCI_CAP_ID_MSI) ? nvec : 1,
190 (type == PCI_CAP_ID_MSIX) ?
201 return msi_device_populate_sysfs(&dev->dev);
205 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
207 dev_err(&dev->dev, "Xen PCI frontend error: %d!\n", ret);
213 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
217 * We set vector == 0 to tell the hypervisor we don't care about
218 * it, but we want a pirq setup instead. We use the dest_id fields
219 * to pass the pirq that we want.
221 memset(msg, 0, sizeof(*msg));
222 msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
223 msg->arch_addr_hi.destid_8_31 = pirq >> 8;
224 msg->arch_addr_lo.destid_0_7 = pirq & 0xFF;
225 msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
226 msg->arch_data.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
229 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
232 struct msi_desc *msidesc;
235 if (type == PCI_CAP_ID_MSI && nvec > 1)
238 msi_for_each_desc(msidesc, &dev->dev, MSI_DESC_NOTASSOCIATED) {
239 pirq = xen_allocate_pirq_msi(dev, msidesc);
244 xen_msi_compose_msg(dev, pirq, &msg);
245 __pci_write_msi_msg(msidesc, &msg);
246 dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
247 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
248 (type == PCI_CAP_ID_MSI) ? nvec : 1,
249 (type == PCI_CAP_ID_MSIX) ?
255 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
257 return msi_device_populate_sysfs(&dev->dev);
260 dev_err(&dev->dev, "Failed to create MSI%s! ret=%d!\n",
261 type == PCI_CAP_ID_MSI ? "" : "-X", irq);
265 #ifdef CONFIG_XEN_PV_DOM0
266 static bool __read_mostly pci_seg_supported = true;
268 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
271 struct msi_desc *msidesc;
273 msi_for_each_desc(msidesc, &dev->dev, MSI_DESC_NOTASSOCIATED) {
274 struct physdev_map_pirq map_irq;
277 domid = ret = xen_find_device_domain_owner(dev);
278 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
279 * hence check ret value for < 0. */
283 memset(&map_irq, 0, sizeof(map_irq));
284 map_irq.domid = domid;
285 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
288 map_irq.bus = dev->bus->number |
289 (pci_domain_nr(dev->bus) << 16);
290 map_irq.devfn = dev->devfn;
292 if (type == PCI_CAP_ID_MSI && nvec > 1) {
293 map_irq.type = MAP_PIRQ_TYPE_MULTI_MSI;
294 map_irq.entry_nr = nvec;
295 } else if (type == PCI_CAP_ID_MSIX) {
298 u32 table_offset, bir;
301 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
303 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
304 flags = pci_resource_flags(dev, bir);
305 if (!flags || (flags & IORESOURCE_UNSET))
308 map_irq.table_base = pci_resource_start(dev, bir);
309 map_irq.entry_nr = msidesc->msi_index;
313 if (pci_seg_supported)
314 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
316 if (type == PCI_CAP_ID_MSI && nvec > 1 && ret) {
318 * If MAP_PIRQ_TYPE_MULTI_MSI is not available
319 * there's nothing else we can do in this case.
320 * Just set ret > 0 so driver can retry with
326 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
327 map_irq.type = MAP_PIRQ_TYPE_MSI;
330 map_irq.bus = dev->bus->number;
331 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
334 pci_seg_supported = false;
337 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
342 ret = xen_bind_pirq_msi_to_irq(dev, msidesc, map_irq.pirq,
343 (type == PCI_CAP_ID_MSI) ? nvec : 1,
344 (type == PCI_CAP_ID_MSIX) ? "msi-x" : "msi",
349 ret = msi_device_populate_sysfs(&dev->dev);
354 bool xen_initdom_restore_msi(struct pci_dev *dev)
358 if (!xen_initial_domain())
361 if (pci_seg_supported) {
362 struct physdev_pci_device restore_ext;
364 restore_ext.seg = pci_domain_nr(dev->bus);
365 restore_ext.bus = dev->bus->number;
366 restore_ext.devfn = dev->devfn;
367 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
370 pci_seg_supported = false;
371 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
373 if (!pci_seg_supported) {
374 struct physdev_restore_msi restore;
376 restore.bus = dev->bus->number;
377 restore.devfn = dev->devfn;
378 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
379 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
383 #else /* CONFIG_XEN_PV_DOM0 */
384 #define xen_initdom_setup_msi_irqs NULL
385 #endif /* !CONFIG_XEN_PV_DOM0 */
387 static void xen_teardown_msi_irqs(struct pci_dev *dev)
389 struct msi_desc *msidesc;
392 msi_for_each_desc(msidesc, &dev->dev, MSI_DESC_ASSOCIATED) {
393 for (i = 0; i < msidesc->nvec_used; i++)
394 xen_destroy_irq(msidesc->irq + i);
398 msi_device_destroy_sysfs(&dev->dev);
401 static void xen_pv_teardown_msi_irqs(struct pci_dev *dev)
403 if (dev->msix_enabled)
404 xen_pci_frontend_disable_msix(dev);
406 xen_pci_frontend_disable_msi(dev);
408 xen_teardown_msi_irqs(dev);
411 static int xen_msi_domain_alloc_irqs(struct irq_domain *domain,
412 struct device *dev, int nvec)
416 if (WARN_ON_ONCE(!dev_is_pci(dev)))
419 type = to_pci_dev(dev)->msix_enabled ? PCI_CAP_ID_MSIX : PCI_CAP_ID_MSI;
421 return xen_msi_ops.setup_msi_irqs(to_pci_dev(dev), nvec, type);
424 static void xen_msi_domain_free_irqs(struct irq_domain *domain,
427 if (WARN_ON_ONCE(!dev_is_pci(dev)))
430 xen_msi_ops.teardown_msi_irqs(to_pci_dev(dev));
433 static struct msi_domain_ops xen_pci_msi_domain_ops = {
434 .domain_alloc_irqs = xen_msi_domain_alloc_irqs,
435 .domain_free_irqs = xen_msi_domain_free_irqs,
438 static struct msi_domain_info xen_pci_msi_domain_info = {
439 .flags = MSI_FLAG_PCI_MSIX | MSI_FLAG_FREE_MSI_DESCS | MSI_FLAG_DEV_SYSFS,
440 .ops = &xen_pci_msi_domain_ops,
444 * This irq domain is a blatant violation of the irq domain design, but
445 * distangling XEN into real irq domains is not a job for mere mortals with
446 * limited XENology. But it's the least dangerous way for a mere mortal to
447 * get rid of the arch_*_msi_irqs() hackery in order to store the irq
448 * domain pointer in struct device. This irq domain wrappery allows to do
449 * that without breaking XEN terminally.
451 static __init struct irq_domain *xen_create_pci_msi_domain(void)
453 struct irq_domain *d = NULL;
454 struct fwnode_handle *fn;
456 fn = irq_domain_alloc_named_fwnode("XEN-MSI");
458 d = msi_create_irq_domain(fn, &xen_pci_msi_domain_info, NULL);
460 /* FIXME: No idea how to survive if this fails */
466 static __init void xen_setup_pci_msi(void)
468 if (xen_pv_domain()) {
469 if (xen_initial_domain())
470 xen_msi_ops.setup_msi_irqs = xen_initdom_setup_msi_irqs;
472 xen_msi_ops.setup_msi_irqs = xen_setup_msi_irqs;
473 xen_msi_ops.teardown_msi_irqs = xen_pv_teardown_msi_irqs;
474 } else if (xen_hvm_domain()) {
475 xen_msi_ops.setup_msi_irqs = xen_hvm_setup_msi_irqs;
476 xen_msi_ops.teardown_msi_irqs = xen_teardown_msi_irqs;
483 * Override the PCI/MSI irq domain init function. No point
484 * in allocating the native domain and never use it.
486 x86_init.irqs.create_pci_msi_domain = xen_create_pci_msi_domain;
488 * With XEN PIRQ/Eventchannels in use PCI/MSI[-X] masking is solely
489 * controlled by the hypervisor.
491 pci_msi_ignore_mask = 1;
494 #else /* CONFIG_PCI_MSI */
495 static inline void xen_setup_pci_msi(void) { }
496 #endif /* CONFIG_PCI_MSI */
498 int __init pci_xen_init(void)
500 if (!xen_pv_domain() || xen_initial_domain())
503 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
505 pcibios_set_cache_line_size();
507 pcibios_enable_irq = xen_pcifront_enable_irq;
508 pcibios_disable_irq = NULL;
510 /* Keep ACPI out of the picture */
517 #ifdef CONFIG_PCI_MSI
518 static void __init xen_hvm_msi_init(void)
520 if (!apic_is_disabled) {
522 * If hardware supports (x2)APIC virtualization (as indicated
523 * by hypervisor's leaf 4) then we don't need to use pirqs/
524 * event channels for MSI handling and instead use regular
527 uint32_t eax = cpuid_eax(xen_cpuid_base() + 4);
529 if (((eax & XEN_HVM_CPUID_X2APIC_VIRT) && x2apic_mode) ||
530 ((eax & XEN_HVM_CPUID_APIC_ACCESS_VIRT) && boot_cpu_has(X86_FEATURE_APIC)))
537 int __init pci_xen_hvm_init(void)
539 if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
544 * We don't want to change the actual ACPI delivery model,
545 * just how GSIs get registered.
547 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
548 __acpi_unregister_gsi = NULL;
551 #ifdef CONFIG_PCI_MSI
553 * We need to wait until after x2apic is initialized
554 * before we can set MSI IRQ ops.
556 x86_platform.apic_post_init = xen_hvm_msi_init;
561 #ifdef CONFIG_XEN_PV_DOM0
562 int __init pci_xen_initial_domain(void)
567 __acpi_register_gsi = acpi_register_gsi_xen;
568 __acpi_unregister_gsi = NULL;
570 * Pre-allocate the legacy IRQs. Use NR_LEGACY_IRQS here
571 * because we don't have a PIC and thus nr_legacy_irqs() is zero.
573 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
574 int trigger, polarity;
576 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
579 xen_register_pirq(irq,
580 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
581 true /* Map GSI to PIRQ */);
583 if (0 == nr_ioapics) {
584 for (irq = 0; irq < nr_legacy_irqs(); irq++)
585 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");