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/i8259.h>
28 static int xen_pcifront_enable_irq(struct pci_dev *dev)
35 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
37 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
41 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
44 if (gsi < nr_legacy_irqs())
47 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
49 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
55 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
60 static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
63 int rc, pirq = -1, irq = -1;
64 struct physdev_map_pirq map_irq;
68 irq = xen_irq_from_gsi(gsi);
75 map_irq.domid = DOMID_SELF;
76 map_irq.type = MAP_PIRQ_TYPE_GSI;
80 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
82 printk(KERN_WARNING "xen map irq failed %d\n", rc);
86 if (triggering == ACPI_EDGE_SENSITIVE) {
91 name = "ioapic-level";
94 if (gsi_override >= 0)
97 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
101 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
106 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
107 int trigger, int polarity)
109 if (!xen_hvm_domain())
112 return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
113 false /* no mapping of GSI to PIRQ */);
116 #ifdef CONFIG_XEN_DOM0
117 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
120 struct physdev_setup_gsi setup_gsi;
122 if (!xen_pv_domain())
125 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
126 gsi, triggering, polarity);
128 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
131 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
132 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
134 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
136 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
138 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
145 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
146 int trigger, int polarity)
148 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
153 #if defined(CONFIG_PCI_MSI)
154 #include <linux/msi.h>
155 #include <asm/msidef.h>
157 struct xen_pci_frontend_ops *xen_pci_frontend;
158 EXPORT_SYMBOL_GPL(xen_pci_frontend);
160 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
163 struct msi_desc *msidesc;
166 if (type == PCI_CAP_ID_MSI && nvec > 1)
169 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
173 if (type == PCI_CAP_ID_MSIX)
174 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
176 ret = xen_pci_frontend_enable_msi(dev, v);
180 list_for_each_entry(msidesc, &dev->msi_list, list) {
181 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i],
182 (type == PCI_CAP_ID_MSI) ? nvec : 1,
183 (type == PCI_CAP_ID_MSIX) ?
197 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
203 #define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
204 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
206 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
209 /* We set vector == 0 to tell the hypervisor we don't care about it,
210 * but we want a pirq setup instead.
211 * We use the dest_id field to pass the pirq that we want. */
212 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
215 MSI_ADDR_DEST_MODE_PHYSICAL |
216 MSI_ADDR_REDIRECTION_CPU |
217 MSI_ADDR_DEST_ID(pirq);
219 msg->data = XEN_PIRQ_MSI_DATA;
222 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
225 struct msi_desc *msidesc;
228 if (type == PCI_CAP_ID_MSI && nvec > 1)
231 list_for_each_entry(msidesc, &dev->msi_list, list) {
232 __read_msi_msg(msidesc, &msg);
233 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
234 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
235 if (msg.data != XEN_PIRQ_MSI_DATA ||
236 xen_irq_from_pirq(pirq) < 0) {
237 pirq = xen_allocate_pirq_msi(dev, msidesc);
242 xen_msi_compose_msg(dev, pirq, &msg);
243 __write_msi_msg(msidesc, &msg);
244 dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
247 "xen: msi already bound to pirq=%d\n", pirq);
249 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
250 (type == PCI_CAP_ID_MSI) ? nvec : 1,
251 (type == PCI_CAP_ID_MSIX) ?
257 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
263 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
267 #ifdef CONFIG_XEN_DOM0
268 static bool __read_mostly pci_seg_supported = true;
270 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
273 struct msi_desc *msidesc;
275 list_for_each_entry(msidesc, &dev->msi_list, list) {
276 struct physdev_map_pirq map_irq;
279 domid = ret = xen_find_device_domain_owner(dev);
280 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
281 * hence check ret value for < 0. */
285 memset(&map_irq, 0, sizeof(map_irq));
286 map_irq.domid = domid;
287 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
290 map_irq.bus = dev->bus->number |
291 (pci_domain_nr(dev->bus) << 16);
292 map_irq.devfn = dev->devfn;
294 if (type == PCI_CAP_ID_MSI && nvec > 1) {
295 map_irq.type = MAP_PIRQ_TYPE_MULTI_MSI;
296 map_irq.entry_nr = nvec;
297 } else if (type == PCI_CAP_ID_MSIX) {
299 u32 table_offset, bir;
302 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
304 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
306 map_irq.table_base = pci_resource_start(dev, bir);
307 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
311 if (pci_seg_supported)
312 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
314 if (type == PCI_CAP_ID_MSI && nvec > 1 && ret) {
316 * If MAP_PIRQ_TYPE_MULTI_MSI is not available
317 * there's nothing else we can do in this case.
318 * Just set ret > 0 so driver can retry with
324 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
325 map_irq.type = MAP_PIRQ_TYPE_MSI;
328 map_irq.bus = dev->bus->number;
329 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
332 pci_seg_supported = false;
335 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
340 ret = xen_bind_pirq_msi_to_irq(dev, msidesc, map_irq.pirq,
341 (type == PCI_CAP_ID_MSI) ? nvec : 1,
342 (type == PCI_CAP_ID_MSIX) ? "msi-x" : "msi",
352 static void xen_initdom_restore_msi_irqs(struct pci_dev *dev)
356 if (pci_seg_supported) {
357 struct physdev_pci_device restore_ext;
359 restore_ext.seg = pci_domain_nr(dev->bus);
360 restore_ext.bus = dev->bus->number;
361 restore_ext.devfn = dev->devfn;
362 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
365 pci_seg_supported = false;
366 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
368 if (!pci_seg_supported) {
369 struct physdev_restore_msi restore;
371 restore.bus = dev->bus->number;
372 restore.devfn = dev->devfn;
373 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
374 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
379 static void xen_teardown_msi_irqs(struct pci_dev *dev)
381 struct msi_desc *msidesc;
383 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
384 if (msidesc->msi_attrib.is_msix)
385 xen_pci_frontend_disable_msix(dev);
387 xen_pci_frontend_disable_msi(dev);
389 /* Free the IRQ's and the msidesc using the generic code. */
390 default_teardown_msi_irqs(dev);
393 static void xen_teardown_msi_irq(unsigned int irq)
395 xen_destroy_irq(irq);
397 static u32 xen_nop_msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
401 static u32 xen_nop_msix_mask_irq(struct msi_desc *desc, u32 flag)
407 int __init pci_xen_init(void)
409 if (!xen_pv_domain() || xen_initial_domain())
412 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
414 pcibios_set_cache_line_size();
416 pcibios_enable_irq = xen_pcifront_enable_irq;
417 pcibios_disable_irq = NULL;
420 /* Keep ACPI out of the picture */
424 #ifdef CONFIG_PCI_MSI
425 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
426 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
427 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
428 x86_msi.msi_mask_irq = xen_nop_msi_mask_irq;
429 x86_msi.msix_mask_irq = xen_nop_msix_mask_irq;
434 int __init pci_xen_hvm_init(void)
436 if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
441 * We don't want to change the actual ACPI delivery model,
442 * just how GSIs get registered.
444 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
447 #ifdef CONFIG_PCI_MSI
448 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
449 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
454 #ifdef CONFIG_XEN_DOM0
455 static __init void xen_setup_acpi_sci(void)
458 int trigger, polarity;
459 int gsi = acpi_sci_override_gsi;
461 int gsi_override = -1;
466 rc = acpi_get_override_irq(gsi, &trigger, &polarity);
468 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
469 " sci, rc=%d\n", rc);
472 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
473 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
475 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
476 "polarity=%d\n", gsi, trigger, polarity);
478 /* Before we bind the GSI to a Linux IRQ, check whether
479 * we need to override it with bus_irq (IRQ) value. Usually for
480 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
481 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
482 * but there are oddballs where the IRQ != GSI:
483 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
484 * which ends up being: gsi_to_irq[9] == 20
485 * (which is what acpi_gsi_to_irq ends up calling when starting the
486 * the ACPI interpreter and keels over since IRQ 9 has not been
487 * setup as we had setup IRQ 20 for it).
489 if (acpi_gsi_to_irq(gsi, &irq) == 0) {
490 /* Use the provided value if it's valid. */
495 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
496 printk(KERN_INFO "xen: acpi sci %d\n", gsi);
501 int __init pci_xen_initial_domain(void)
505 #ifdef CONFIG_PCI_MSI
506 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
507 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
508 x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
509 x86_msi.msi_mask_irq = xen_nop_msi_mask_irq;
510 x86_msi.msix_mask_irq = xen_nop_msix_mask_irq;
512 xen_setup_acpi_sci();
513 __acpi_register_gsi = acpi_register_gsi_xen;
514 /* Pre-allocate legacy irqs */
515 for (irq = 0; irq < nr_legacy_irqs(); irq++) {
516 int trigger, polarity;
518 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
521 xen_register_pirq(irq, -1 /* no GSI override */,
522 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
523 true /* Map GSI to PIRQ */);
525 if (0 == nr_ioapics) {
526 for (irq = 0; irq < nr_legacy_irqs(); irq++)
527 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
532 struct xen_device_domain_owner {
535 struct list_head list;
538 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
539 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
541 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
543 struct xen_device_domain_owner *owner;
545 list_for_each_entry(owner, &dev_domain_list, list) {
546 if (owner->dev == dev)
552 int xen_find_device_domain_owner(struct pci_dev *dev)
554 struct xen_device_domain_owner *owner;
555 int domain = -ENODEV;
557 spin_lock(&dev_domain_list_spinlock);
558 owner = find_device(dev);
560 domain = owner->domain;
561 spin_unlock(&dev_domain_list_spinlock);
564 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
566 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
568 struct xen_device_domain_owner *owner;
570 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
574 spin_lock(&dev_domain_list_spinlock);
575 if (find_device(dev)) {
576 spin_unlock(&dev_domain_list_spinlock);
580 owner->domain = domain;
582 list_add_tail(&owner->list, &dev_domain_list);
583 spin_unlock(&dev_domain_list_spinlock);
586 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
588 int xen_unregister_device_domain_owner(struct pci_dev *dev)
590 struct xen_device_domain_owner *owner;
592 spin_lock(&dev_domain_list_spinlock);
593 owner = find_device(dev);
595 spin_unlock(&dev_domain_list_spinlock);
598 list_del(&owner->list);
599 spin_unlock(&dev_domain_list_spinlock);
603 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);