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",
41 rc = xen_allocate_pirq_gsi(gsi);
43 dev_warn(&dev->dev, "Xen PCI: failed to allocate a PIRQ for GSI%d: %d\n",
49 if (gsi < NR_IRQS_LEGACY)
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 gsi_override, int triggering,
68 int rc, pirq = -1, irq = -1;
69 struct physdev_map_pirq map_irq;
74 pirq = xen_allocate_pirq_gsi(gsi);
78 map_irq.domid = DOMID_SELF;
79 map_irq.type = MAP_PIRQ_TYPE_GSI;
83 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
85 printk(KERN_WARNING "xen map irq failed %d\n", rc);
89 if (triggering == ACPI_EDGE_SENSITIVE) {
94 name = "ioapic-level";
97 if (gsi_override >= 0)
100 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
104 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
109 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
110 int trigger, int polarity)
112 if (!xen_hvm_domain())
115 return xen_register_pirq(gsi, -1 /* no GSI override */,
116 trigger, false /* no PIRQ allocation */);
119 #ifdef CONFIG_XEN_DOM0
120 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
123 struct physdev_setup_gsi setup_gsi;
125 if (!xen_pv_domain())
128 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
129 gsi, triggering, polarity);
131 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
134 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
135 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
137 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
139 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
141 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
148 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
149 int trigger, int polarity)
151 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
156 #if defined(CONFIG_PCI_MSI)
157 #include <linux/msi.h>
158 #include <asm/msidef.h>
160 struct xen_pci_frontend_ops *xen_pci_frontend;
161 EXPORT_SYMBOL_GPL(xen_pci_frontend);
163 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
166 struct msi_desc *msidesc;
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], 0,
182 (type == PCI_CAP_ID_MSIX) ?
194 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
200 #define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
201 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
203 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
206 /* We set vector == 0 to tell the hypervisor we don't care about it,
207 * but we want a pirq setup instead.
208 * We use the dest_id field to pass the pirq that we want. */
209 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
212 MSI_ADDR_DEST_MODE_PHYSICAL |
213 MSI_ADDR_REDIRECTION_CPU |
214 MSI_ADDR_DEST_ID(pirq);
216 msg->data = XEN_PIRQ_MSI_DATA;
219 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
222 struct msi_desc *msidesc;
225 list_for_each_entry(msidesc, &dev->msi_list, list) {
226 __read_msi_msg(msidesc, &msg);
227 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
228 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
229 if (msg.data != XEN_PIRQ_MSI_DATA ||
230 xen_irq_from_pirq(pirq) < 0) {
231 pirq = xen_allocate_pirq_msi(dev, msidesc);
234 xen_msi_compose_msg(dev, pirq, &msg);
235 __write_msi_msg(msidesc, &msg);
236 dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
239 "xen: msi already bound to pirq=%d\n", pirq);
241 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq, 0,
242 (type == PCI_CAP_ID_MSIX) ?
248 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
254 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
258 #ifdef CONFIG_XEN_DOM0
259 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
262 struct msi_desc *msidesc;
264 list_for_each_entry(msidesc, &dev->msi_list, list) {
265 struct physdev_map_pirq map_irq;
268 domid = ret = xen_find_device_domain_owner(dev);
269 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
270 * hence check ret value for < 0. */
274 memset(&map_irq, 0, sizeof(map_irq));
275 map_irq.domid = domid;
276 map_irq.type = MAP_PIRQ_TYPE_MSI;
279 map_irq.bus = dev->bus->number;
280 map_irq.devfn = dev->devfn;
282 if (type == PCI_CAP_ID_MSIX) {
284 u32 table_offset, bir;
286 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
288 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
290 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
292 map_irq.table_base = pci_resource_start(dev, bir);
293 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
296 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
298 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
303 ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
304 map_irq.pirq, map_irq.index,
305 (type == PCI_CAP_ID_MSIX) ?
317 static void xen_teardown_msi_irqs(struct pci_dev *dev)
319 struct msi_desc *msidesc;
321 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
322 if (msidesc->msi_attrib.is_msix)
323 xen_pci_frontend_disable_msix(dev);
325 xen_pci_frontend_disable_msi(dev);
327 /* Free the IRQ's and the msidesc using the generic code. */
328 default_teardown_msi_irqs(dev);
331 static void xen_teardown_msi_irq(unsigned int irq)
333 xen_destroy_irq(irq);
338 int __init pci_xen_init(void)
340 if (!xen_pv_domain() || xen_initial_domain())
343 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
345 pcibios_set_cache_line_size();
347 pcibios_enable_irq = xen_pcifront_enable_irq;
348 pcibios_disable_irq = NULL;
351 /* Keep ACPI out of the picture */
355 #ifdef CONFIG_PCI_MSI
356 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
357 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
358 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
363 int __init pci_xen_hvm_init(void)
365 if (!xen_feature(XENFEAT_hvm_pirqs))
370 * We don't want to change the actual ACPI delivery model,
371 * just how GSIs get registered.
373 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
376 #ifdef CONFIG_PCI_MSI
377 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
378 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
383 #ifdef CONFIG_XEN_DOM0
385 static __init void xen_setup_acpi_sci(void)
388 int trigger, polarity;
389 int gsi = acpi_sci_override_gsi;
391 int gsi_override = -1;
396 rc = acpi_get_override_irq(gsi, &trigger, &polarity);
398 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
399 " sci, rc=%d\n", rc);
402 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
403 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
405 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
406 "polarity=%d\n", gsi, trigger, polarity);
408 /* Before we bind the GSI to a Linux IRQ, check whether
409 * we need to override it with bus_irq (IRQ) value. Usually for
410 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
411 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
412 * but there are oddballs where the IRQ != GSI:
413 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
414 * which ends up being: gsi_to_irq[9] == 20
415 * (which is what acpi_gsi_to_irq ends up calling when starting the
416 * the ACPI interpreter and keels over since IRQ 9 has not been
417 * setup as we had setup IRQ 20 for it).
419 /* Check whether the GSI != IRQ */
420 if (acpi_gsi_to_irq(gsi, &irq) == 0) {
421 if (irq >= 0 && irq != gsi)
422 /* Bugger, we MUST have that IRQ. */
426 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
427 printk(KERN_INFO "xen: acpi sci %d\n", gsi);
432 static int __init pci_xen_initial_domain(void)
434 #ifdef CONFIG_PCI_MSI
435 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
436 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
439 xen_setup_acpi_sci();
440 __acpi_register_gsi = acpi_register_gsi_xen;
445 void __init xen_setup_pirqs(void)
449 pci_xen_initial_domain();
451 if (0 == nr_ioapics) {
452 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
453 pirq = xen_allocate_pirq_gsi(irq);
455 "Could not allocate PIRQ for legacy interrupt\n"))
457 irq = xen_bind_pirq_gsi_to_irq(irq, pirq, 0, "xt-pic");
462 /* Pre-allocate legacy irqs */
463 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
464 int trigger, polarity;
466 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
469 xen_register_pirq(irq, -1 /* no GSI override */,
470 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
471 true /* allocate IRQ */);
476 struct xen_device_domain_owner {
479 struct list_head list;
482 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
483 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
485 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
487 struct xen_device_domain_owner *owner;
489 list_for_each_entry(owner, &dev_domain_list, list) {
490 if (owner->dev == dev)
496 int xen_find_device_domain_owner(struct pci_dev *dev)
498 struct xen_device_domain_owner *owner;
499 int domain = -ENODEV;
501 spin_lock(&dev_domain_list_spinlock);
502 owner = find_device(dev);
504 domain = owner->domain;
505 spin_unlock(&dev_domain_list_spinlock);
508 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
510 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
512 struct xen_device_domain_owner *owner;
514 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
518 spin_lock(&dev_domain_list_spinlock);
519 if (find_device(dev)) {
520 spin_unlock(&dev_domain_list_spinlock);
524 owner->domain = domain;
526 list_add_tail(&owner->list, &dev_domain_list);
527 spin_unlock(&dev_domain_list_spinlock);
530 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
532 int xen_unregister_device_domain_owner(struct pci_dev *dev)
534 struct xen_device_domain_owner *owner;
536 spin_lock(&dev_domain_list_spinlock);
537 owner = find_device(dev);
539 spin_unlock(&dev_domain_list_spinlock);
542 list_del(&owner->list);
543 spin_unlock(&dev_domain_list_spinlock);
547 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);