2 * Xen PCI - handle PCI (INTx) and MSI infrastructure calls for PV, HVM and
3 * initial domain support. We also handle the DSDT _PRT callbacks for GSI's
4 * used in HVM and initial domain mode (PV does not parse ACPI, so it has no
5 * concept of GSIs). Under PV we hook under the pnbbios API for IRQs and
6 * 0xcf8 PCI configuration read/write.
8 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
9 * Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
10 * Stefano Stabellini <stefano.stabellini@eu.citrix.com>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/pci.h>
15 #include <linux/acpi.h>
18 #include <asm/io_apic.h>
19 #include <asm/pci_x86.h>
21 #include <asm/xen/hypervisor.h>
23 #include <xen/features.h>
24 #include <xen/events.h>
25 #include <asm/xen/pci.h>
26 #include <asm/xen/cpuid.h>
28 #include <asm/i8259.h>
30 static int xen_pcifront_enable_irq(struct pci_dev *dev)
37 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
39 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
43 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
46 if (gsi < nr_legacy_irqs())
49 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
51 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
57 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
62 static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
65 int rc, pirq = -1, irq = -1;
66 struct physdev_map_pirq map_irq;
70 irq = xen_irq_from_gsi(gsi);
77 map_irq.domid = DOMID_SELF;
78 map_irq.type = MAP_PIRQ_TYPE_GSI;
82 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
84 printk(KERN_WARNING "xen map irq failed %d\n", rc);
88 if (triggering == ACPI_EDGE_SENSITIVE) {
93 name = "ioapic-level";
96 if (gsi_override >= 0)
99 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
103 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
108 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
109 int trigger, int polarity)
111 if (!xen_hvm_domain())
114 return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
115 false /* no mapping of GSI to PIRQ */);
118 #ifdef CONFIG_XEN_DOM0
119 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
122 struct physdev_setup_gsi setup_gsi;
124 if (!xen_pv_domain())
127 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
128 gsi, triggering, polarity);
130 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
133 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
134 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
136 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
138 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
140 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
147 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
148 int trigger, int polarity)
150 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
155 #if defined(CONFIG_PCI_MSI)
156 #include <linux/msi.h>
157 #include <asm/msidef.h>
159 struct xen_pci_frontend_ops *xen_pci_frontend;
160 EXPORT_SYMBOL_GPL(xen_pci_frontend);
162 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
165 struct msi_desc *msidesc;
168 if (type == PCI_CAP_ID_MSI && nvec > 1)
171 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
175 if (type == PCI_CAP_ID_MSIX)
176 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
178 ret = xen_pci_frontend_enable_msi(dev, v);
182 list_for_each_entry(msidesc, &dev->msi_list, list) {
183 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i],
184 (type == PCI_CAP_ID_MSI) ? nvec : 1,
185 (type == PCI_CAP_ID_MSIX) ?
199 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
205 #define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
206 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
208 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
211 /* We set vector == 0 to tell the hypervisor we don't care about it,
212 * but we want a pirq setup instead.
213 * We use the dest_id field to pass the pirq that we want. */
214 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
217 MSI_ADDR_DEST_MODE_PHYSICAL |
218 MSI_ADDR_REDIRECTION_CPU |
219 MSI_ADDR_DEST_ID(pirq);
221 msg->data = XEN_PIRQ_MSI_DATA;
224 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
227 struct msi_desc *msidesc;
230 if (type == PCI_CAP_ID_MSI && nvec > 1)
233 list_for_each_entry(msidesc, &dev->msi_list, list) {
234 __pci_read_msi_msg(msidesc, &msg);
235 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
236 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
237 if (msg.data != XEN_PIRQ_MSI_DATA ||
238 xen_irq_from_pirq(pirq) < 0) {
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);
249 "xen: msi already bound to pirq=%d\n", pirq);
251 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
252 (type == PCI_CAP_ID_MSI) ? nvec : 1,
253 (type == PCI_CAP_ID_MSIX) ?
259 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
265 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
269 #ifdef CONFIG_XEN_DOM0
270 static bool __read_mostly pci_seg_supported = true;
272 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
275 struct msi_desc *msidesc;
277 list_for_each_entry(msidesc, &dev->msi_list, list) {
278 struct physdev_map_pirq map_irq;
281 domid = ret = xen_find_device_domain_owner(dev);
282 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
283 * hence check ret value for < 0. */
287 memset(&map_irq, 0, sizeof(map_irq));
288 map_irq.domid = domid;
289 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
292 map_irq.bus = dev->bus->number |
293 (pci_domain_nr(dev->bus) << 16);
294 map_irq.devfn = dev->devfn;
296 if (type == PCI_CAP_ID_MSI && nvec > 1) {
297 map_irq.type = MAP_PIRQ_TYPE_MULTI_MSI;
298 map_irq.entry_nr = nvec;
299 } else if (type == PCI_CAP_ID_MSIX) {
301 u32 table_offset, bir;
304 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
306 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
308 map_irq.table_base = pci_resource_start(dev, bir);
309 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
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",
354 static void xen_initdom_restore_msi_irqs(struct pci_dev *dev)
358 if (pci_seg_supported) {
359 struct physdev_pci_device restore_ext;
361 restore_ext.seg = pci_domain_nr(dev->bus);
362 restore_ext.bus = dev->bus->number;
363 restore_ext.devfn = dev->devfn;
364 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
367 pci_seg_supported = false;
368 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
370 if (!pci_seg_supported) {
371 struct physdev_restore_msi restore;
373 restore.bus = dev->bus->number;
374 restore.devfn = dev->devfn;
375 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
376 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
381 static void xen_teardown_msi_irqs(struct pci_dev *dev)
383 struct msi_desc *msidesc;
385 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
386 if (msidesc->msi_attrib.is_msix)
387 xen_pci_frontend_disable_msix(dev);
389 xen_pci_frontend_disable_msi(dev);
391 /* Free the IRQ's and the msidesc using the generic code. */
392 default_teardown_msi_irqs(dev);
395 static void xen_teardown_msi_irq(unsigned int irq)
397 xen_destroy_irq(irq);
402 int __init pci_xen_init(void)
404 if (!xen_pv_domain() || xen_initial_domain())
407 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
409 pcibios_set_cache_line_size();
411 pcibios_enable_irq = xen_pcifront_enable_irq;
412 pcibios_disable_irq = NULL;
415 /* Keep ACPI out of the picture */
419 #ifdef CONFIG_PCI_MSI
420 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
421 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
422 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
423 pci_msi_ignore_mask = 1;
428 #ifdef CONFIG_PCI_MSI
429 void __init xen_msi_init(void)
433 * If hardware supports (x2)APIC virtualization (as indicated
434 * by hypervisor's leaf 4) then we don't need to use pirqs/
435 * event channels for MSI handling and instead use regular
438 uint32_t eax = cpuid_eax(xen_cpuid_base() + 4);
440 if (((eax & XEN_HVM_CPUID_X2APIC_VIRT) && x2apic_mode) ||
441 ((eax & XEN_HVM_CPUID_APIC_ACCESS_VIRT) && cpu_has_apic))
445 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
446 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
450 int __init pci_xen_hvm_init(void)
452 if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
457 * We don't want to change the actual ACPI delivery model,
458 * just how GSIs get registered.
460 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
463 #ifdef CONFIG_PCI_MSI
465 * We need to wait until after x2apic is initialized
466 * before we can set MSI IRQ ops.
468 x86_platform.apic_post_init = xen_msi_init;
473 #ifdef CONFIG_XEN_DOM0
474 static __init void xen_setup_acpi_sci(void)
477 int trigger, polarity;
478 int gsi = acpi_sci_override_gsi;
480 int gsi_override = -1;
485 rc = acpi_get_override_irq(gsi, &trigger, &polarity);
487 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
488 " sci, rc=%d\n", rc);
491 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
492 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
494 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
495 "polarity=%d\n", gsi, trigger, polarity);
497 /* Before we bind the GSI to a Linux IRQ, check whether
498 * we need to override it with bus_irq (IRQ) value. Usually for
499 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
500 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
501 * but there are oddballs where the IRQ != GSI:
502 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
503 * which ends up being: gsi_to_irq[9] == 20
504 * (which is what acpi_gsi_to_irq ends up calling when starting the
505 * the ACPI interpreter and keels over since IRQ 9 has not been
506 * setup as we had setup IRQ 20 for it).
508 if (acpi_gsi_to_irq(gsi, &irq) == 0) {
509 /* Use the provided value if it's valid. */
514 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
515 printk(KERN_INFO "xen: acpi sci %d\n", gsi);
520 int __init pci_xen_initial_domain(void)
524 #ifdef CONFIG_PCI_MSI
525 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
526 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
527 x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
528 pci_msi_ignore_mask = 1;
530 xen_setup_acpi_sci();
531 __acpi_register_gsi = acpi_register_gsi_xen;
532 /* Pre-allocate legacy irqs */
533 for (irq = 0; irq < nr_legacy_irqs(); irq++) {
534 int trigger, polarity;
536 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
539 xen_register_pirq(irq, -1 /* no GSI override */,
540 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
541 true /* Map GSI to PIRQ */);
543 if (0 == nr_ioapics) {
544 for (irq = 0; irq < nr_legacy_irqs(); irq++)
545 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
550 struct xen_device_domain_owner {
553 struct list_head list;
556 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
557 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
559 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
561 struct xen_device_domain_owner *owner;
563 list_for_each_entry(owner, &dev_domain_list, list) {
564 if (owner->dev == dev)
570 int xen_find_device_domain_owner(struct pci_dev *dev)
572 struct xen_device_domain_owner *owner;
573 int domain = -ENODEV;
575 spin_lock(&dev_domain_list_spinlock);
576 owner = find_device(dev);
578 domain = owner->domain;
579 spin_unlock(&dev_domain_list_spinlock);
582 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
584 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
586 struct xen_device_domain_owner *owner;
588 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
592 spin_lock(&dev_domain_list_spinlock);
593 if (find_device(dev)) {
594 spin_unlock(&dev_domain_list_spinlock);
598 owner->domain = domain;
600 list_add_tail(&owner->list, &dev_domain_list);
601 spin_unlock(&dev_domain_list_spinlock);
604 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
606 int xen_unregister_device_domain_owner(struct pci_dev *dev)
608 struct xen_device_domain_owner *owner;
610 spin_lock(&dev_domain_list_spinlock);
611 owner = find_device(dev);
613 spin_unlock(&dev_domain_list_spinlock);
616 list_del(&owner->list);
617 spin_unlock(&dev_domain_list_spinlock);
621 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);