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>
27 static int xen_pcifront_enable_irq(struct pci_dev *dev)
34 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
36 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
40 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
43 if (gsi < NR_IRQS_LEGACY)
46 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
48 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
54 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
59 static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
62 int rc, pirq = -1, irq = -1;
63 struct physdev_map_pirq map_irq;
70 map_irq.domid = DOMID_SELF;
71 map_irq.type = MAP_PIRQ_TYPE_GSI;
75 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
77 printk(KERN_WARNING "xen map irq failed %d\n", rc);
81 if (triggering == ACPI_EDGE_SENSITIVE) {
86 name = "ioapic-level";
89 if (gsi_override >= 0)
92 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
96 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
101 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
102 int trigger, int polarity)
104 if (!xen_hvm_domain())
107 return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
108 false /* no mapping of GSI to PIRQ */);
111 #ifdef CONFIG_XEN_DOM0
112 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
115 struct physdev_setup_gsi setup_gsi;
117 if (!xen_pv_domain())
120 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
121 gsi, triggering, polarity);
123 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
126 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
127 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
129 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
131 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
133 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
140 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
141 int trigger, int polarity)
143 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
148 #if defined(CONFIG_PCI_MSI)
149 #include <linux/msi.h>
150 #include <asm/msidef.h>
152 struct xen_pci_frontend_ops *xen_pci_frontend;
153 EXPORT_SYMBOL_GPL(xen_pci_frontend);
155 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
158 struct msi_desc *msidesc;
161 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
165 if (type == PCI_CAP_ID_MSIX)
166 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
168 ret = xen_pci_frontend_enable_msi(dev, v);
172 list_for_each_entry(msidesc, &dev->msi_list, list) {
173 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i], 0,
174 (type == PCI_CAP_ID_MSIX) ?
188 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
194 #define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
195 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
197 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
200 /* We set vector == 0 to tell the hypervisor we don't care about it,
201 * but we want a pirq setup instead.
202 * We use the dest_id field to pass the pirq that we want. */
203 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
206 MSI_ADDR_DEST_MODE_PHYSICAL |
207 MSI_ADDR_REDIRECTION_CPU |
208 MSI_ADDR_DEST_ID(pirq);
210 msg->data = XEN_PIRQ_MSI_DATA;
213 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
216 struct msi_desc *msidesc;
219 list_for_each_entry(msidesc, &dev->msi_list, list) {
220 __read_msi_msg(msidesc, &msg);
221 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
222 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
223 if (msg.data != XEN_PIRQ_MSI_DATA ||
224 xen_irq_from_pirq(pirq) < 0) {
225 pirq = xen_allocate_pirq_msi(dev, msidesc);
230 xen_msi_compose_msg(dev, pirq, &msg);
231 __write_msi_msg(msidesc, &msg);
232 dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
235 "xen: msi already bound to pirq=%d\n", pirq);
237 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq, 0,
238 (type == PCI_CAP_ID_MSIX) ?
244 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
250 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
254 #ifdef CONFIG_XEN_DOM0
255 static bool __read_mostly pci_seg_supported = true;
257 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
260 struct msi_desc *msidesc;
262 list_for_each_entry(msidesc, &dev->msi_list, list) {
263 struct physdev_map_pirq map_irq;
266 domid = ret = xen_find_device_domain_owner(dev);
267 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
268 * hence check ret value for < 0. */
272 memset(&map_irq, 0, sizeof(map_irq));
273 map_irq.domid = domid;
274 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
277 map_irq.bus = dev->bus->number |
278 (pci_domain_nr(dev->bus) << 16);
279 map_irq.devfn = dev->devfn;
281 if (type == PCI_CAP_ID_MSIX) {
283 u32 table_offset, bir;
285 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
287 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
289 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
291 map_irq.table_base = pci_resource_start(dev, bir);
292 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
296 if (pci_seg_supported)
297 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
299 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
300 map_irq.type = MAP_PIRQ_TYPE_MSI;
303 map_irq.bus = dev->bus->number;
304 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
307 pci_seg_supported = false;
310 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
315 ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
316 map_irq.pirq, map_irq.index,
317 (type == PCI_CAP_ID_MSIX) ?
329 static void xen_teardown_msi_irqs(struct pci_dev *dev)
331 struct msi_desc *msidesc;
333 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
334 if (msidesc->msi_attrib.is_msix)
335 xen_pci_frontend_disable_msix(dev);
337 xen_pci_frontend_disable_msi(dev);
339 /* Free the IRQ's and the msidesc using the generic code. */
340 default_teardown_msi_irqs(dev);
343 static void xen_teardown_msi_irq(unsigned int irq)
345 xen_destroy_irq(irq);
350 int __init pci_xen_init(void)
352 if (!xen_pv_domain() || xen_initial_domain())
355 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
357 pcibios_set_cache_line_size();
359 pcibios_enable_irq = xen_pcifront_enable_irq;
360 pcibios_disable_irq = NULL;
363 /* Keep ACPI out of the picture */
367 #ifdef CONFIG_PCI_MSI
368 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
369 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
370 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
375 int __init pci_xen_hvm_init(void)
377 if (!xen_feature(XENFEAT_hvm_pirqs))
382 * We don't want to change the actual ACPI delivery model,
383 * just how GSIs get registered.
385 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
388 #ifdef CONFIG_PCI_MSI
389 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
390 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
395 #ifdef CONFIG_XEN_DOM0
396 static __init void xen_setup_acpi_sci(void)
399 int trigger, polarity;
400 int gsi = acpi_sci_override_gsi;
402 int gsi_override = -1;
407 rc = acpi_get_override_irq(gsi, &trigger, &polarity);
409 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
410 " sci, rc=%d\n", rc);
413 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
414 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
416 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
417 "polarity=%d\n", gsi, trigger, polarity);
419 /* Before we bind the GSI to a Linux IRQ, check whether
420 * we need to override it with bus_irq (IRQ) value. Usually for
421 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
422 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
423 * but there are oddballs where the IRQ != GSI:
424 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
425 * which ends up being: gsi_to_irq[9] == 20
426 * (which is what acpi_gsi_to_irq ends up calling when starting the
427 * the ACPI interpreter and keels over since IRQ 9 has not been
428 * setup as we had setup IRQ 20 for it).
430 if (acpi_gsi_to_irq(gsi, &irq) == 0) {
431 /* Use the provided value if it's valid. */
436 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
437 printk(KERN_INFO "xen: acpi sci %d\n", gsi);
442 int __init pci_xen_initial_domain(void)
446 #ifdef CONFIG_PCI_MSI
447 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
448 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
450 xen_setup_acpi_sci();
451 __acpi_register_gsi = acpi_register_gsi_xen;
452 /* Pre-allocate legacy irqs */
453 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
454 int trigger, polarity;
456 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
459 xen_register_pirq(irq, -1 /* no GSI override */,
460 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
461 true /* Map GSI to PIRQ */);
463 if (0 == nr_ioapics) {
464 for (irq = 0; irq < NR_IRQS_LEGACY; irq++)
465 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
470 struct xen_device_domain_owner {
473 struct list_head list;
476 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
477 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
479 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
481 struct xen_device_domain_owner *owner;
483 list_for_each_entry(owner, &dev_domain_list, list) {
484 if (owner->dev == dev)
490 int xen_find_device_domain_owner(struct pci_dev *dev)
492 struct xen_device_domain_owner *owner;
493 int domain = -ENODEV;
495 spin_lock(&dev_domain_list_spinlock);
496 owner = find_device(dev);
498 domain = owner->domain;
499 spin_unlock(&dev_domain_list_spinlock);
502 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
504 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
506 struct xen_device_domain_owner *owner;
508 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
512 spin_lock(&dev_domain_list_spinlock);
513 if (find_device(dev)) {
514 spin_unlock(&dev_domain_list_spinlock);
518 owner->domain = domain;
520 list_add_tail(&owner->list, &dev_domain_list);
521 spin_unlock(&dev_domain_list_spinlock);
524 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
526 int xen_unregister_device_domain_owner(struct pci_dev *dev)
528 struct xen_device_domain_owner *owner;
530 spin_lock(&dev_domain_list_spinlock);
531 owner = find_device(dev);
533 spin_unlock(&dev_domain_list_spinlock);
536 list_del(&owner->list);
537 spin_unlock(&dev_domain_list_spinlock);
541 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);