Loading...
1/*
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.
7 *
8 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
9 * Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
10 * Stefano Stabellini <stefano.stabellini@eu.citrix.com>
11 */
12#include <linux/export.h>
13#include <linux/init.h>
14#include <linux/pci.h>
15#include <linux/acpi.h>
16
17#include <linux/io.h>
18#include <asm/io_apic.h>
19#include <asm/pci_x86.h>
20
21#include <asm/xen/hypervisor.h>
22
23#include <xen/features.h>
24#include <xen/events.h>
25#include <asm/xen/pci.h>
26#include <asm/xen/cpuid.h>
27#include <asm/apic.h>
28#include <asm/i8259.h>
29
30static int xen_pcifront_enable_irq(struct pci_dev *dev)
31{
32 int rc;
33 int share = 1;
34 int pirq;
35 u8 gsi;
36
37 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
38 if (rc < 0) {
39 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
40 rc);
41 return rc;
42 }
43 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
44 pirq = gsi;
45
46 if (gsi < nr_legacy_irqs())
47 share = 0;
48
49 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
50 if (rc < 0) {
51 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
52 gsi, pirq, rc);
53 return rc;
54 }
55
56 dev->irq = rc;
57 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
58 return 0;
59}
60
61#ifdef CONFIG_ACPI
62static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
63 bool set_pirq)
64{
65 int rc, pirq = -1, irq = -1;
66 struct physdev_map_pirq map_irq;
67 int shareable = 0;
68 char *name;
69
70 irq = xen_irq_from_gsi(gsi);
71 if (irq > 0)
72 return irq;
73
74 if (set_pirq)
75 pirq = gsi;
76
77 map_irq.domid = DOMID_SELF;
78 map_irq.type = MAP_PIRQ_TYPE_GSI;
79 map_irq.index = gsi;
80 map_irq.pirq = pirq;
81
82 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
83 if (rc) {
84 printk(KERN_WARNING "xen map irq failed %d\n", rc);
85 return -1;
86 }
87
88 if (triggering == ACPI_EDGE_SENSITIVE) {
89 shareable = 0;
90 name = "ioapic-edge";
91 } else {
92 shareable = 1;
93 name = "ioapic-level";
94 }
95
96 if (gsi_override >= 0)
97 gsi = gsi_override;
98
99 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
100 if (irq < 0)
101 goto out;
102
103 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
104out:
105 return irq;
106}
107
108static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
109 int trigger, int polarity)
110{
111 if (!xen_hvm_domain())
112 return -1;
113
114 return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
115 false /* no mapping of GSI to PIRQ */);
116}
117
118#ifdef CONFIG_XEN_DOM0
119static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
120{
121 int rc, irq;
122 struct physdev_setup_gsi setup_gsi;
123
124 if (!xen_pv_domain())
125 return -1;
126
127 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
128 gsi, triggering, polarity);
129
130 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
131
132 setup_gsi.gsi = gsi;
133 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
134 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
135
136 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
137 if (rc == -EEXIST)
138 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
139 else if (rc) {
140 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
141 gsi, rc);
142 }
143
144 return irq;
145}
146
147static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
148 int trigger, int polarity)
149{
150 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
151}
152#endif
153#endif
154
155#if defined(CONFIG_PCI_MSI)
156#include <linux/msi.h>
157#include <asm/msidef.h>
158
159struct xen_pci_frontend_ops *xen_pci_frontend;
160EXPORT_SYMBOL_GPL(xen_pci_frontend);
161
162static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
163{
164 int irq, ret, i;
165 struct msi_desc *msidesc;
166 int *v;
167
168 if (type == PCI_CAP_ID_MSI && nvec > 1)
169 return 1;
170
171 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
172 if (!v)
173 return -ENOMEM;
174
175 if (type == PCI_CAP_ID_MSIX)
176 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
177 else
178 ret = xen_pci_frontend_enable_msi(dev, v);
179 if (ret)
180 goto error;
181 i = 0;
182 for_each_pci_msi_entry(msidesc, dev) {
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) ?
186 "pcifront-msi-x" :
187 "pcifront-msi",
188 DOMID_SELF);
189 if (irq < 0) {
190 ret = irq;
191 goto free;
192 }
193 i++;
194 }
195 kfree(v);
196 return 0;
197
198error:
199 if (ret == -ENOSYS)
200 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
201 else if (ret)
202 dev_err(&dev->dev, "Xen PCI frontend error: %d!\n", ret);
203free:
204 kfree(v);
205 return ret;
206}
207
208#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
209 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
210
211static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
212 struct msi_msg *msg)
213{
214 /* We set vector == 0 to tell the hypervisor we don't care about it,
215 * but we want a pirq setup instead.
216 * We use the dest_id field to pass the pirq that we want. */
217 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
218 msg->address_lo =
219 MSI_ADDR_BASE_LO |
220 MSI_ADDR_DEST_MODE_PHYSICAL |
221 MSI_ADDR_REDIRECTION_CPU |
222 MSI_ADDR_DEST_ID(pirq);
223
224 msg->data = XEN_PIRQ_MSI_DATA;
225}
226
227static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
228{
229 int irq, pirq;
230 struct msi_desc *msidesc;
231 struct msi_msg msg;
232
233 if (type == PCI_CAP_ID_MSI && nvec > 1)
234 return 1;
235
236 for_each_pci_msi_entry(msidesc, dev) {
237 pirq = xen_allocate_pirq_msi(dev, msidesc);
238 if (pirq < 0) {
239 irq = -ENODEV;
240 goto error;
241 }
242 xen_msi_compose_msg(dev, pirq, &msg);
243 __pci_write_msi_msg(msidesc, &msg);
244 dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
245 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
246 (type == PCI_CAP_ID_MSI) ? nvec : 1,
247 (type == PCI_CAP_ID_MSIX) ?
248 "msi-x" : "msi",
249 DOMID_SELF);
250 if (irq < 0)
251 goto error;
252 dev_dbg(&dev->dev,
253 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
254 }
255 return 0;
256
257error:
258 dev_err(&dev->dev, "Failed to create MSI%s! ret=%d!\n",
259 type == PCI_CAP_ID_MSI ? "" : "-X", irq);
260 return irq;
261}
262
263#ifdef CONFIG_XEN_DOM0
264static bool __read_mostly pci_seg_supported = true;
265
266static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
267{
268 int ret = 0;
269 struct msi_desc *msidesc;
270
271 for_each_pci_msi_entry(msidesc, dev) {
272 struct physdev_map_pirq map_irq;
273 domid_t domid;
274
275 domid = ret = xen_find_device_domain_owner(dev);
276 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
277 * hence check ret value for < 0. */
278 if (ret < 0)
279 domid = DOMID_SELF;
280
281 memset(&map_irq, 0, sizeof(map_irq));
282 map_irq.domid = domid;
283 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
284 map_irq.index = -1;
285 map_irq.pirq = -1;
286 map_irq.bus = dev->bus->number |
287 (pci_domain_nr(dev->bus) << 16);
288 map_irq.devfn = dev->devfn;
289
290 if (type == PCI_CAP_ID_MSI && nvec > 1) {
291 map_irq.type = MAP_PIRQ_TYPE_MULTI_MSI;
292 map_irq.entry_nr = nvec;
293 } else if (type == PCI_CAP_ID_MSIX) {
294 int pos;
295 unsigned long flags;
296 u32 table_offset, bir;
297
298 pos = dev->msix_cap;
299 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
300 &table_offset);
301 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
302 flags = pci_resource_flags(dev, bir);
303 if (!flags || (flags & IORESOURCE_UNSET))
304 return -EINVAL;
305
306 map_irq.table_base = pci_resource_start(dev, bir);
307 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
308 }
309
310 ret = -EINVAL;
311 if (pci_seg_supported)
312 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
313 &map_irq);
314 if (type == PCI_CAP_ID_MSI && nvec > 1 && ret) {
315 /*
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
319 * single MSI.
320 */
321 ret = 1;
322 goto out;
323 }
324 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
325 map_irq.type = MAP_PIRQ_TYPE_MSI;
326 map_irq.index = -1;
327 map_irq.pirq = -1;
328 map_irq.bus = dev->bus->number;
329 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
330 &map_irq);
331 if (ret != -EINVAL)
332 pci_seg_supported = false;
333 }
334 if (ret) {
335 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
336 ret, domid);
337 goto out;
338 }
339
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",
343 domid);
344 if (ret < 0)
345 goto out;
346 }
347 ret = 0;
348out:
349 return ret;
350}
351
352static void xen_initdom_restore_msi_irqs(struct pci_dev *dev)
353{
354 int ret = 0;
355
356 if (pci_seg_supported) {
357 struct physdev_pci_device restore_ext;
358
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,
363 &restore_ext);
364 if (ret == -ENOSYS)
365 pci_seg_supported = false;
366 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
367 }
368 if (!pci_seg_supported) {
369 struct physdev_restore_msi restore;
370
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);
375 }
376}
377#endif
378
379static void xen_teardown_msi_irqs(struct pci_dev *dev)
380{
381 struct msi_desc *msidesc;
382
383 msidesc = first_pci_msi_entry(dev);
384 if (msidesc->msi_attrib.is_msix)
385 xen_pci_frontend_disable_msix(dev);
386 else
387 xen_pci_frontend_disable_msi(dev);
388
389 /* Free the IRQ's and the msidesc using the generic code. */
390 default_teardown_msi_irqs(dev);
391}
392
393static void xen_teardown_msi_irq(unsigned int irq)
394{
395 xen_destroy_irq(irq);
396}
397
398#endif
399
400int __init pci_xen_init(void)
401{
402 if (!xen_pv_domain() || xen_initial_domain())
403 return -ENODEV;
404
405 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
406
407 pcibios_set_cache_line_size();
408
409 pcibios_enable_irq = xen_pcifront_enable_irq;
410 pcibios_disable_irq = NULL;
411
412#ifdef CONFIG_ACPI
413 /* Keep ACPI out of the picture */
414 acpi_noirq = 1;
415#endif
416
417#ifdef CONFIG_PCI_MSI
418 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
419 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
420 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
421 pci_msi_ignore_mask = 1;
422#endif
423 return 0;
424}
425
426#ifdef CONFIG_PCI_MSI
427void __init xen_msi_init(void)
428{
429 if (!disable_apic) {
430 /*
431 * If hardware supports (x2)APIC virtualization (as indicated
432 * by hypervisor's leaf 4) then we don't need to use pirqs/
433 * event channels for MSI handling and instead use regular
434 * APIC processing
435 */
436 uint32_t eax = cpuid_eax(xen_cpuid_base() + 4);
437
438 if (((eax & XEN_HVM_CPUID_X2APIC_VIRT) && x2apic_mode) ||
439 ((eax & XEN_HVM_CPUID_APIC_ACCESS_VIRT) && boot_cpu_has(X86_FEATURE_APIC)))
440 return;
441 }
442
443 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
444 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
445}
446#endif
447
448int __init pci_xen_hvm_init(void)
449{
450 if (!xen_feature(XENFEAT_hvm_pirqs))
451 return 0;
452
453#ifdef CONFIG_ACPI
454 /*
455 * We don't want to change the actual ACPI delivery model,
456 * just how GSIs get registered.
457 */
458 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
459 __acpi_unregister_gsi = NULL;
460#endif
461
462#ifdef CONFIG_PCI_MSI
463 /*
464 * We need to wait until after x2apic is initialized
465 * before we can set MSI IRQ ops.
466 */
467 x86_platform.apic_post_init = xen_msi_init;
468#endif
469 return 0;
470}
471
472#ifdef CONFIG_XEN_DOM0
473int __init pci_xen_initial_domain(void)
474{
475 int irq;
476
477#ifdef CONFIG_PCI_MSI
478 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
479 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
480 x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
481 pci_msi_ignore_mask = 1;
482#endif
483 __acpi_register_gsi = acpi_register_gsi_xen;
484 __acpi_unregister_gsi = NULL;
485 /*
486 * Pre-allocate the legacy IRQs. Use NR_LEGACY_IRQS here
487 * because we don't have a PIC and thus nr_legacy_irqs() is zero.
488 */
489 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
490 int trigger, polarity;
491
492 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
493 continue;
494
495 xen_register_pirq(irq, -1 /* no GSI override */,
496 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
497 true /* Map GSI to PIRQ */);
498 }
499 if (0 == nr_ioapics) {
500 for (irq = 0; irq < nr_legacy_irqs(); irq++)
501 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
502 }
503 return 0;
504}
505
506struct xen_device_domain_owner {
507 domid_t domain;
508 struct pci_dev *dev;
509 struct list_head list;
510};
511
512static DEFINE_SPINLOCK(dev_domain_list_spinlock);
513static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
514
515static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
516{
517 struct xen_device_domain_owner *owner;
518
519 list_for_each_entry(owner, &dev_domain_list, list) {
520 if (owner->dev == dev)
521 return owner;
522 }
523 return NULL;
524}
525
526int xen_find_device_domain_owner(struct pci_dev *dev)
527{
528 struct xen_device_domain_owner *owner;
529 int domain = -ENODEV;
530
531 spin_lock(&dev_domain_list_spinlock);
532 owner = find_device(dev);
533 if (owner)
534 domain = owner->domain;
535 spin_unlock(&dev_domain_list_spinlock);
536 return domain;
537}
538EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
539
540int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
541{
542 struct xen_device_domain_owner *owner;
543
544 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
545 if (!owner)
546 return -ENODEV;
547
548 spin_lock(&dev_domain_list_spinlock);
549 if (find_device(dev)) {
550 spin_unlock(&dev_domain_list_spinlock);
551 kfree(owner);
552 return -EEXIST;
553 }
554 owner->domain = domain;
555 owner->dev = dev;
556 list_add_tail(&owner->list, &dev_domain_list);
557 spin_unlock(&dev_domain_list_spinlock);
558 return 0;
559}
560EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
561
562int xen_unregister_device_domain_owner(struct pci_dev *dev)
563{
564 struct xen_device_domain_owner *owner;
565
566 spin_lock(&dev_domain_list_spinlock);
567 owner = find_device(dev);
568 if (!owner) {
569 spin_unlock(&dev_domain_list_spinlock);
570 return -ENODEV;
571 }
572 list_del(&owner->list);
573 spin_unlock(&dev_domain_list_spinlock);
574 kfree(owner);
575 return 0;
576}
577EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
578#endif
1/*
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.
7 *
8 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
9 * Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
10 * Stefano Stabellini <stefano.stabellini@eu.citrix.com>
11 */
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/pci.h>
15#include <linux/acpi.h>
16
17#include <linux/io.h>
18#include <asm/io_apic.h>
19#include <asm/pci_x86.h>
20
21#include <asm/xen/hypervisor.h>
22
23#include <xen/features.h>
24#include <xen/events.h>
25#include <asm/xen/pci.h>
26
27static int xen_pcifront_enable_irq(struct pci_dev *dev)
28{
29 int rc;
30 int share = 1;
31 int pirq;
32 u8 gsi;
33
34 rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
35 if (rc < 0) {
36 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
37 rc);
38 return rc;
39 }
40 /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
41 pirq = gsi;
42
43 if (gsi < NR_IRQS_LEGACY)
44 share = 0;
45
46 rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
47 if (rc < 0) {
48 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
49 gsi, pirq, rc);
50 return rc;
51 }
52
53 dev->irq = rc;
54 dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
55 return 0;
56}
57
58#ifdef CONFIG_ACPI
59static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
60 bool set_pirq)
61{
62 int rc, pirq = -1, irq = -1;
63 struct physdev_map_pirq map_irq;
64 int shareable = 0;
65 char *name;
66
67 irq = xen_irq_from_gsi(gsi);
68 if (irq > 0)
69 return irq;
70
71 if (set_pirq)
72 pirq = gsi;
73
74 map_irq.domid = DOMID_SELF;
75 map_irq.type = MAP_PIRQ_TYPE_GSI;
76 map_irq.index = gsi;
77 map_irq.pirq = pirq;
78
79 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
80 if (rc) {
81 printk(KERN_WARNING "xen map irq failed %d\n", rc);
82 return -1;
83 }
84
85 if (triggering == ACPI_EDGE_SENSITIVE) {
86 shareable = 0;
87 name = "ioapic-edge";
88 } else {
89 shareable = 1;
90 name = "ioapic-level";
91 }
92
93 if (gsi_override >= 0)
94 gsi = gsi_override;
95
96 irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
97 if (irq < 0)
98 goto out;
99
100 printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
101out:
102 return irq;
103}
104
105static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
106 int trigger, int polarity)
107{
108 if (!xen_hvm_domain())
109 return -1;
110
111 return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
112 false /* no mapping of GSI to PIRQ */);
113}
114
115#ifdef CONFIG_XEN_DOM0
116static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
117{
118 int rc, irq;
119 struct physdev_setup_gsi setup_gsi;
120
121 if (!xen_pv_domain())
122 return -1;
123
124 printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
125 gsi, triggering, polarity);
126
127 irq = xen_register_pirq(gsi, gsi_override, triggering, true);
128
129 setup_gsi.gsi = gsi;
130 setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
131 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
132
133 rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
134 if (rc == -EEXIST)
135 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
136 else if (rc) {
137 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
138 gsi, rc);
139 }
140
141 return irq;
142}
143
144static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
145 int trigger, int polarity)
146{
147 return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
148}
149#endif
150#endif
151
152#if defined(CONFIG_PCI_MSI)
153#include <linux/msi.h>
154#include <asm/msidef.h>
155
156struct xen_pci_frontend_ops *xen_pci_frontend;
157EXPORT_SYMBOL_GPL(xen_pci_frontend);
158
159static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
160{
161 int irq, ret, i;
162 struct msi_desc *msidesc;
163 int *v;
164
165 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
166 if (!v)
167 return -ENOMEM;
168
169 if (type == PCI_CAP_ID_MSIX)
170 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
171 else
172 ret = xen_pci_frontend_enable_msi(dev, v);
173 if (ret)
174 goto error;
175 i = 0;
176 list_for_each_entry(msidesc, &dev->msi_list, list) {
177 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i], 0,
178 (type == PCI_CAP_ID_MSIX) ?
179 "pcifront-msi-x" :
180 "pcifront-msi",
181 DOMID_SELF);
182 if (irq < 0) {
183 ret = irq;
184 goto free;
185 }
186 i++;
187 }
188 kfree(v);
189 return 0;
190
191error:
192 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
193free:
194 kfree(v);
195 return ret;
196}
197
198#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \
199 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
200
201static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
202 struct msi_msg *msg)
203{
204 /* We set vector == 0 to tell the hypervisor we don't care about it,
205 * but we want a pirq setup instead.
206 * We use the dest_id field to pass the pirq that we want. */
207 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
208 msg->address_lo =
209 MSI_ADDR_BASE_LO |
210 MSI_ADDR_DEST_MODE_PHYSICAL |
211 MSI_ADDR_REDIRECTION_CPU |
212 MSI_ADDR_DEST_ID(pirq);
213
214 msg->data = XEN_PIRQ_MSI_DATA;
215}
216
217static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
218{
219 int irq, pirq;
220 struct msi_desc *msidesc;
221 struct msi_msg msg;
222
223 list_for_each_entry(msidesc, &dev->msi_list, list) {
224 __read_msi_msg(msidesc, &msg);
225 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
226 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
227 if (msg.data != XEN_PIRQ_MSI_DATA ||
228 xen_irq_from_pirq(pirq) < 0) {
229 pirq = xen_allocate_pirq_msi(dev, msidesc);
230 if (pirq < 0) {
231 irq = -ENODEV;
232 goto error;
233 }
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);
237 } else {
238 dev_dbg(&dev->dev,
239 "xen: msi already bound to pirq=%d\n", pirq);
240 }
241 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq, 0,
242 (type == PCI_CAP_ID_MSIX) ?
243 "msi-x" : "msi",
244 DOMID_SELF);
245 if (irq < 0)
246 goto error;
247 dev_dbg(&dev->dev,
248 "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
249 }
250 return 0;
251
252error:
253 dev_err(&dev->dev,
254 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
255 return irq;
256}
257
258#ifdef CONFIG_XEN_DOM0
259static bool __read_mostly pci_seg_supported = true;
260
261static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
262{
263 int ret = 0;
264 struct msi_desc *msidesc;
265
266 list_for_each_entry(msidesc, &dev->msi_list, list) {
267 struct physdev_map_pirq map_irq;
268 domid_t domid;
269
270 domid = ret = xen_find_device_domain_owner(dev);
271 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
272 * hence check ret value for < 0. */
273 if (ret < 0)
274 domid = DOMID_SELF;
275
276 memset(&map_irq, 0, sizeof(map_irq));
277 map_irq.domid = domid;
278 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
279 map_irq.index = -1;
280 map_irq.pirq = -1;
281 map_irq.bus = dev->bus->number |
282 (pci_domain_nr(dev->bus) << 16);
283 map_irq.devfn = dev->devfn;
284
285 if (type == PCI_CAP_ID_MSIX) {
286 int pos;
287 u32 table_offset, bir;
288
289 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
290
291 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
292 &table_offset);
293 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
294
295 map_irq.table_base = pci_resource_start(dev, bir);
296 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
297 }
298
299 ret = -EINVAL;
300 if (pci_seg_supported)
301 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
302 &map_irq);
303 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
304 map_irq.type = MAP_PIRQ_TYPE_MSI;
305 map_irq.index = -1;
306 map_irq.pirq = -1;
307 map_irq.bus = dev->bus->number;
308 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
309 &map_irq);
310 if (ret != -EINVAL)
311 pci_seg_supported = false;
312 }
313 if (ret) {
314 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
315 ret, domid);
316 goto out;
317 }
318
319 ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
320 map_irq.pirq, map_irq.index,
321 (type == PCI_CAP_ID_MSIX) ?
322 "msi-x" : "msi",
323 domid);
324 if (ret < 0)
325 goto out;
326 }
327 ret = 0;
328out:
329 return ret;
330}
331
332static void xen_initdom_restore_msi_irqs(struct pci_dev *dev, int irq)
333{
334 int ret = 0;
335
336 if (pci_seg_supported) {
337 struct physdev_pci_device restore_ext;
338
339 restore_ext.seg = pci_domain_nr(dev->bus);
340 restore_ext.bus = dev->bus->number;
341 restore_ext.devfn = dev->devfn;
342 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
343 &restore_ext);
344 if (ret == -ENOSYS)
345 pci_seg_supported = false;
346 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
347 }
348 if (!pci_seg_supported) {
349 struct physdev_restore_msi restore;
350
351 restore.bus = dev->bus->number;
352 restore.devfn = dev->devfn;
353 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
354 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
355 }
356}
357#endif
358
359static void xen_teardown_msi_irqs(struct pci_dev *dev)
360{
361 struct msi_desc *msidesc;
362
363 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
364 if (msidesc->msi_attrib.is_msix)
365 xen_pci_frontend_disable_msix(dev);
366 else
367 xen_pci_frontend_disable_msi(dev);
368
369 /* Free the IRQ's and the msidesc using the generic code. */
370 default_teardown_msi_irqs(dev);
371}
372
373static void xen_teardown_msi_irq(unsigned int irq)
374{
375 xen_destroy_irq(irq);
376}
377
378#endif
379
380int __init pci_xen_init(void)
381{
382 if (!xen_pv_domain() || xen_initial_domain())
383 return -ENODEV;
384
385 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
386
387 pcibios_set_cache_line_size();
388
389 pcibios_enable_irq = xen_pcifront_enable_irq;
390 pcibios_disable_irq = NULL;
391
392#ifdef CONFIG_ACPI
393 /* Keep ACPI out of the picture */
394 acpi_noirq = 1;
395#endif
396
397#ifdef CONFIG_PCI_MSI
398 x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
399 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
400 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
401#endif
402 return 0;
403}
404
405int __init pci_xen_hvm_init(void)
406{
407 if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
408 return 0;
409
410#ifdef CONFIG_ACPI
411 /*
412 * We don't want to change the actual ACPI delivery model,
413 * just how GSIs get registered.
414 */
415 __acpi_register_gsi = acpi_register_gsi_xen_hvm;
416#endif
417
418#ifdef CONFIG_PCI_MSI
419 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
420 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
421#endif
422 return 0;
423}
424
425#ifdef CONFIG_XEN_DOM0
426static __init void xen_setup_acpi_sci(void)
427{
428 int rc;
429 int trigger, polarity;
430 int gsi = acpi_sci_override_gsi;
431 int irq = -1;
432 int gsi_override = -1;
433
434 if (!gsi)
435 return;
436
437 rc = acpi_get_override_irq(gsi, &trigger, &polarity);
438 if (rc) {
439 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
440 " sci, rc=%d\n", rc);
441 return;
442 }
443 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
444 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
445
446 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
447 "polarity=%d\n", gsi, trigger, polarity);
448
449 /* Before we bind the GSI to a Linux IRQ, check whether
450 * we need to override it with bus_irq (IRQ) value. Usually for
451 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
452 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
453 * but there are oddballs where the IRQ != GSI:
454 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
455 * which ends up being: gsi_to_irq[9] == 20
456 * (which is what acpi_gsi_to_irq ends up calling when starting the
457 * the ACPI interpreter and keels over since IRQ 9 has not been
458 * setup as we had setup IRQ 20 for it).
459 */
460 if (acpi_gsi_to_irq(gsi, &irq) == 0) {
461 /* Use the provided value if it's valid. */
462 if (irq >= 0)
463 gsi_override = irq;
464 }
465
466 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
467 printk(KERN_INFO "xen: acpi sci %d\n", gsi);
468
469 return;
470}
471
472int __init pci_xen_initial_domain(void)
473{
474 int irq;
475
476#ifdef CONFIG_PCI_MSI
477 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
478 x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
479 x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
480#endif
481 xen_setup_acpi_sci();
482 __acpi_register_gsi = acpi_register_gsi_xen;
483 /* Pre-allocate legacy irqs */
484 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
485 int trigger, polarity;
486
487 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
488 continue;
489
490 xen_register_pirq(irq, -1 /* no GSI override */,
491 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
492 true /* Map GSI to PIRQ */);
493 }
494 if (0 == nr_ioapics) {
495 for (irq = 0; irq < NR_IRQS_LEGACY; irq++)
496 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
497 }
498 return 0;
499}
500
501struct xen_device_domain_owner {
502 domid_t domain;
503 struct pci_dev *dev;
504 struct list_head list;
505};
506
507static DEFINE_SPINLOCK(dev_domain_list_spinlock);
508static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
509
510static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
511{
512 struct xen_device_domain_owner *owner;
513
514 list_for_each_entry(owner, &dev_domain_list, list) {
515 if (owner->dev == dev)
516 return owner;
517 }
518 return NULL;
519}
520
521int xen_find_device_domain_owner(struct pci_dev *dev)
522{
523 struct xen_device_domain_owner *owner;
524 int domain = -ENODEV;
525
526 spin_lock(&dev_domain_list_spinlock);
527 owner = find_device(dev);
528 if (owner)
529 domain = owner->domain;
530 spin_unlock(&dev_domain_list_spinlock);
531 return domain;
532}
533EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
534
535int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
536{
537 struct xen_device_domain_owner *owner;
538
539 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
540 if (!owner)
541 return -ENODEV;
542
543 spin_lock(&dev_domain_list_spinlock);
544 if (find_device(dev)) {
545 spin_unlock(&dev_domain_list_spinlock);
546 kfree(owner);
547 return -EEXIST;
548 }
549 owner->domain = domain;
550 owner->dev = dev;
551 list_add_tail(&owner->list, &dev_domain_list);
552 spin_unlock(&dev_domain_list_spinlock);
553 return 0;
554}
555EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
556
557int xen_unregister_device_domain_owner(struct pci_dev *dev)
558{
559 struct xen_device_domain_owner *owner;
560
561 spin_lock(&dev_domain_list_spinlock);
562 owner = find_device(dev);
563 if (!owner) {
564 spin_unlock(&dev_domain_list_spinlock);
565 return -ENODEV;
566 }
567 list_del(&owner->list);
568 spin_unlock(&dev_domain_list_spinlock);
569 kfree(owner);
570 return 0;
571}
572EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
573#endif