xref: /linux/arch/x86/pci/xen.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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 
27 static 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
59 static 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 	if (set_pirq)
68 		pirq = gsi;
69 
70 	map_irq.domid = DOMID_SELF;
71 	map_irq.type = MAP_PIRQ_TYPE_GSI;
72 	map_irq.index = gsi;
73 	map_irq.pirq = pirq;
74 
75 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
76 	if (rc) {
77 		printk(KERN_WARNING "xen map irq failed %d\n", rc);
78 		return -1;
79 	}
80 
81 	if (triggering == ACPI_EDGE_SENSITIVE) {
82 		shareable = 0;
83 		name = "ioapic-edge";
84 	} else {
85 		shareable = 1;
86 		name = "ioapic-level";
87 	}
88 
89 	if (gsi_override >= 0)
90 		gsi = gsi_override;
91 
92 	irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
93 	if (irq < 0)
94 		goto out;
95 
96 	printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
97 out:
98 	return irq;
99 }
100 
101 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
102 				     int trigger, int polarity)
103 {
104 	if (!xen_hvm_domain())
105 		return -1;
106 
107 	return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
108 				 false /* no mapping of GSI to PIRQ */);
109 }
110 
111 #ifdef CONFIG_XEN_DOM0
112 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
113 {
114 	int rc, irq;
115 	struct physdev_setup_gsi setup_gsi;
116 
117 	if (!xen_pv_domain())
118 		return -1;
119 
120 	printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
121 			gsi, triggering, polarity);
122 
123 	irq = xen_register_pirq(gsi, gsi_override, triggering, true);
124 
125 	setup_gsi.gsi = gsi;
126 	setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
127 	setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
128 
129 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
130 	if (rc == -EEXIST)
131 		printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
132 	else if (rc) {
133 		printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
134 				gsi, rc);
135 	}
136 
137 	return irq;
138 }
139 
140 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
141 				 int trigger, int polarity)
142 {
143 	return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
144 }
145 #endif
146 #endif
147 
148 #if defined(CONFIG_PCI_MSI)
149 #include <linux/msi.h>
150 #include <asm/msidef.h>
151 
152 struct xen_pci_frontend_ops *xen_pci_frontend;
153 EXPORT_SYMBOL_GPL(xen_pci_frontend);
154 
155 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
156 {
157 	int irq, ret, i;
158 	struct msi_desc *msidesc;
159 	int *v;
160 
161 	v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
162 	if (!v)
163 		return -ENOMEM;
164 
165 	if (type == PCI_CAP_ID_MSIX)
166 		ret = xen_pci_frontend_enable_msix(dev, v, nvec);
167 	else
168 		ret = xen_pci_frontend_enable_msi(dev, v);
169 	if (ret)
170 		goto error;
171 	i = 0;
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) ?
175 					       "pcifront-msi-x" :
176 					       "pcifront-msi",
177 						DOMID_SELF);
178 		if (irq < 0)
179 			goto free;
180 		i++;
181 	}
182 	kfree(v);
183 	return 0;
184 
185 error:
186 	dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
187 free:
188 	kfree(v);
189 	return ret;
190 }
191 
192 #define XEN_PIRQ_MSI_DATA  (MSI_DATA_TRIGGER_EDGE | \
193 		MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
194 
195 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
196 		struct msi_msg *msg)
197 {
198 	/* We set vector == 0 to tell the hypervisor we don't care about it,
199 	 * but we want a pirq setup instead.
200 	 * We use the dest_id field to pass the pirq that we want. */
201 	msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
202 	msg->address_lo =
203 		MSI_ADDR_BASE_LO |
204 		MSI_ADDR_DEST_MODE_PHYSICAL |
205 		MSI_ADDR_REDIRECTION_CPU |
206 		MSI_ADDR_DEST_ID(pirq);
207 
208 	msg->data = XEN_PIRQ_MSI_DATA;
209 }
210 
211 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
212 {
213 	int irq, pirq;
214 	struct msi_desc *msidesc;
215 	struct msi_msg msg;
216 
217 	list_for_each_entry(msidesc, &dev->msi_list, list) {
218 		__read_msi_msg(msidesc, &msg);
219 		pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
220 			((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
221 		if (msg.data != XEN_PIRQ_MSI_DATA ||
222 		    xen_irq_from_pirq(pirq) < 0) {
223 			pirq = xen_allocate_pirq_msi(dev, msidesc);
224 			if (pirq < 0)
225 				goto error;
226 			xen_msi_compose_msg(dev, pirq, &msg);
227 			__write_msi_msg(msidesc, &msg);
228 			dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
229 		} else {
230 			dev_dbg(&dev->dev,
231 				"xen: msi already bound to pirq=%d\n", pirq);
232 		}
233 		irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq, 0,
234 					       (type == PCI_CAP_ID_MSIX) ?
235 					       "msi-x" : "msi",
236 					       DOMID_SELF);
237 		if (irq < 0)
238 			goto error;
239 		dev_dbg(&dev->dev,
240 			"xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
241 	}
242 	return 0;
243 
244 error:
245 	dev_err(&dev->dev,
246 		"Xen PCI frontend has not registered MSI/MSI-X support!\n");
247 	return -ENODEV;
248 }
249 
250 #ifdef CONFIG_XEN_DOM0
251 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
252 {
253 	int ret = 0;
254 	struct msi_desc *msidesc;
255 
256 	list_for_each_entry(msidesc, &dev->msi_list, list) {
257 		struct physdev_map_pirq map_irq;
258 		domid_t domid;
259 
260 		domid = ret = xen_find_device_domain_owner(dev);
261 		/* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
262 		 * hence check ret value for < 0. */
263 		if (ret < 0)
264 			domid = DOMID_SELF;
265 
266 		memset(&map_irq, 0, sizeof(map_irq));
267 		map_irq.domid = domid;
268 		map_irq.type = MAP_PIRQ_TYPE_MSI;
269 		map_irq.index = -1;
270 		map_irq.pirq = -1;
271 		map_irq.bus = dev->bus->number;
272 		map_irq.devfn = dev->devfn;
273 
274 		if (type == PCI_CAP_ID_MSIX) {
275 			int pos;
276 			u32 table_offset, bir;
277 
278 			pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
279 
280 			pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
281 					      &table_offset);
282 			bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
283 
284 			map_irq.table_base = pci_resource_start(dev, bir);
285 			map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
286 		}
287 
288 		ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
289 		if (ret) {
290 			dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
291 				 ret, domid);
292 			goto out;
293 		}
294 
295 		ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
296 					       map_irq.pirq, map_irq.index,
297 					       (type == PCI_CAP_ID_MSIX) ?
298 					       "msi-x" : "msi",
299 						domid);
300 		if (ret < 0)
301 			goto out;
302 	}
303 	ret = 0;
304 out:
305 	return ret;
306 }
307 #endif
308 
309 static void xen_teardown_msi_irqs(struct pci_dev *dev)
310 {
311 	struct msi_desc *msidesc;
312 
313 	msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
314 	if (msidesc->msi_attrib.is_msix)
315 		xen_pci_frontend_disable_msix(dev);
316 	else
317 		xen_pci_frontend_disable_msi(dev);
318 
319 	/* Free the IRQ's and the msidesc using the generic code. */
320 	default_teardown_msi_irqs(dev);
321 }
322 
323 static void xen_teardown_msi_irq(unsigned int irq)
324 {
325 	xen_destroy_irq(irq);
326 }
327 
328 #endif
329 
330 int __init pci_xen_init(void)
331 {
332 	if (!xen_pv_domain() || xen_initial_domain())
333 		return -ENODEV;
334 
335 	printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
336 
337 	pcibios_set_cache_line_size();
338 
339 	pcibios_enable_irq = xen_pcifront_enable_irq;
340 	pcibios_disable_irq = NULL;
341 
342 #ifdef CONFIG_ACPI
343 	/* Keep ACPI out of the picture */
344 	acpi_noirq = 1;
345 #endif
346 
347 #ifdef CONFIG_PCI_MSI
348 	x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
349 	x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
350 	x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
351 #endif
352 	return 0;
353 }
354 
355 int __init pci_xen_hvm_init(void)
356 {
357 	if (!xen_feature(XENFEAT_hvm_pirqs))
358 		return 0;
359 
360 #ifdef CONFIG_ACPI
361 	/*
362 	 * We don't want to change the actual ACPI delivery model,
363 	 * just how GSIs get registered.
364 	 */
365 	__acpi_register_gsi = acpi_register_gsi_xen_hvm;
366 #endif
367 
368 #ifdef CONFIG_PCI_MSI
369 	x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
370 	x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
371 #endif
372 	return 0;
373 }
374 
375 #ifdef CONFIG_XEN_DOM0
376 static __init void xen_setup_acpi_sci(void)
377 {
378 	int rc;
379 	int trigger, polarity;
380 	int gsi = acpi_sci_override_gsi;
381 	int irq = -1;
382 	int gsi_override = -1;
383 
384 	if (!gsi)
385 		return;
386 
387 	rc = acpi_get_override_irq(gsi, &trigger, &polarity);
388 	if (rc) {
389 		printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
390 				" sci, rc=%d\n", rc);
391 		return;
392 	}
393 	trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
394 	polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
395 
396 	printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
397 			"polarity=%d\n", gsi, trigger, polarity);
398 
399 	/* Before we bind the GSI to a Linux IRQ, check whether
400 	 * we need to override it with bus_irq (IRQ) value. Usually for
401 	 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
402 	 *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
403 	 * but there are oddballs where the IRQ != GSI:
404 	 *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
405 	 * which ends up being: gsi_to_irq[9] == 20
406 	 * (which is what acpi_gsi_to_irq ends up calling when starting the
407 	 * the ACPI interpreter and keels over since IRQ 9 has not been
408 	 * setup as we had setup IRQ 20 for it).
409 	 */
410 	if (acpi_gsi_to_irq(gsi, &irq) == 0) {
411 		/* Use the provided value if it's valid. */
412 		if (irq >= 0)
413 			gsi_override = irq;
414 	}
415 
416 	gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
417 	printk(KERN_INFO "xen: acpi sci %d\n", gsi);
418 
419 	return;
420 }
421 
422 int __init pci_xen_initial_domain(void)
423 {
424 	int irq;
425 
426 #ifdef CONFIG_PCI_MSI
427 	x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
428 	x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
429 #endif
430 	xen_setup_acpi_sci();
431 	__acpi_register_gsi = acpi_register_gsi_xen;
432 	/* Pre-allocate legacy irqs */
433 	for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
434 		int trigger, polarity;
435 
436 		if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
437 			continue;
438 
439 		xen_register_pirq(irq, -1 /* no GSI override */,
440 			trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
441 			true /* Map GSI to PIRQ */);
442 	}
443 	if (0 == nr_ioapics) {
444 		for (irq = 0; irq < NR_IRQS_LEGACY; irq++)
445 			xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
446 	}
447 	return 0;
448 }
449 
450 struct xen_device_domain_owner {
451 	domid_t domain;
452 	struct pci_dev *dev;
453 	struct list_head list;
454 };
455 
456 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
457 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
458 
459 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
460 {
461 	struct xen_device_domain_owner *owner;
462 
463 	list_for_each_entry(owner, &dev_domain_list, list) {
464 		if (owner->dev == dev)
465 			return owner;
466 	}
467 	return NULL;
468 }
469 
470 int xen_find_device_domain_owner(struct pci_dev *dev)
471 {
472 	struct xen_device_domain_owner *owner;
473 	int domain = -ENODEV;
474 
475 	spin_lock(&dev_domain_list_spinlock);
476 	owner = find_device(dev);
477 	if (owner)
478 		domain = owner->domain;
479 	spin_unlock(&dev_domain_list_spinlock);
480 	return domain;
481 }
482 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
483 
484 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
485 {
486 	struct xen_device_domain_owner *owner;
487 
488 	owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
489 	if (!owner)
490 		return -ENODEV;
491 
492 	spin_lock(&dev_domain_list_spinlock);
493 	if (find_device(dev)) {
494 		spin_unlock(&dev_domain_list_spinlock);
495 		kfree(owner);
496 		return -EEXIST;
497 	}
498 	owner->domain = domain;
499 	owner->dev = dev;
500 	list_add_tail(&owner->list, &dev_domain_list);
501 	spin_unlock(&dev_domain_list_spinlock);
502 	return 0;
503 }
504 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
505 
506 int xen_unregister_device_domain_owner(struct pci_dev *dev)
507 {
508 	struct xen_device_domain_owner *owner;
509 
510 	spin_lock(&dev_domain_list_spinlock);
511 	owner = find_device(dev);
512 	if (!owner) {
513 		spin_unlock(&dev_domain_list_spinlock);
514 		return -ENODEV;
515 	}
516 	list_del(&owner->list);
517 	spin_unlock(&dev_domain_list_spinlock);
518 	kfree(owner);
519 	return 0;
520 }
521 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
522 #endif
523