xref: /openbmc/linux/drivers/virtio/virtio_pci_common.c (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
1 /*
2  * Virtio PCI driver - common functionality for all device versions
3  *
4  * This module allows virtio devices to be used over a virtual PCI device.
5  * This can be used with QEMU based VMMs like KVM or Xen.
6  *
7  * Copyright IBM Corp. 2007
8  * Copyright Red Hat, Inc. 2014
9  *
10  * Authors:
11  *  Anthony Liguori  <aliguori@us.ibm.com>
12  *  Rusty Russell <rusty@rustcorp.com.au>
13  *  Michael S. Tsirkin <mst@redhat.com>
14  *
15  * This work is licensed under the terms of the GNU GPL, version 2 or later.
16  * See the COPYING file in the top-level directory.
17  *
18  */
19 
20 #include "virtio_pci_common.h"
21 
22 static bool force_legacy = false;
23 
24 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
25 module_param(force_legacy, bool, 0444);
26 MODULE_PARM_DESC(force_legacy,
27 		 "Force legacy mode for transitional virtio 1 devices");
28 #endif
29 
30 /* wait for pending irq handlers */
31 void vp_synchronize_vectors(struct virtio_device *vdev)
32 {
33 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
34 	int i;
35 
36 	synchronize_irq(pci_irq_vector(vp_dev->pci_dev, 0));
37 	for (i = 1; i < vp_dev->msix_vectors; i++)
38 		synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
39 }
40 
41 /* the notify function used when creating a virt queue */
42 bool vp_notify(struct virtqueue *vq)
43 {
44 	/* we write the queue's selector into the notification register to
45 	 * signal the other end */
46 	iowrite16(vq->index, (void __iomem *)vq->priv);
47 	return true;
48 }
49 
50 /* Handle a configuration change: Tell driver if it wants to know. */
51 static irqreturn_t vp_config_changed(int irq, void *opaque)
52 {
53 	struct virtio_pci_device *vp_dev = opaque;
54 
55 	virtio_config_changed(&vp_dev->vdev);
56 	return IRQ_HANDLED;
57 }
58 
59 /* Notify all virtqueues on an interrupt. */
60 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
61 {
62 	struct virtio_pci_device *vp_dev = opaque;
63 	irqreturn_t ret = IRQ_NONE;
64 	struct virtqueue *vq;
65 
66 	list_for_each_entry(vq, &vp_dev->vdev.vqs, list) {
67 		if (vq->callback && vring_interrupt(irq, vq) == IRQ_HANDLED)
68 			ret = IRQ_HANDLED;
69 	}
70 
71 	return ret;
72 }
73 
74 /* A small wrapper to also acknowledge the interrupt when it's handled.
75  * I really need an EIO hook for the vring so I can ack the interrupt once we
76  * know that we'll be handling the IRQ but before we invoke the callback since
77  * the callback may notify the host which results in the host attempting to
78  * raise an interrupt that we would then mask once we acknowledged the
79  * interrupt. */
80 static irqreturn_t vp_interrupt(int irq, void *opaque)
81 {
82 	struct virtio_pci_device *vp_dev = opaque;
83 	u8 isr;
84 
85 	/* reading the ISR has the effect of also clearing it so it's very
86 	 * important to save off the value. */
87 	isr = ioread8(vp_dev->isr);
88 
89 	/* It's definitely not us if the ISR was not high */
90 	if (!isr)
91 		return IRQ_NONE;
92 
93 	/* Configuration change?  Tell driver if it wants to know. */
94 	if (isr & VIRTIO_PCI_ISR_CONFIG)
95 		vp_config_changed(irq, opaque);
96 
97 	return vp_vring_interrupt(irq, opaque);
98 }
99 
100 static void vp_remove_vqs(struct virtio_device *vdev)
101 {
102 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
103 	struct virtqueue *vq, *n;
104 
105 	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
106 		if (vp_dev->msix_vector_map) {
107 			int v = vp_dev->msix_vector_map[vq->index];
108 
109 			if (v != VIRTIO_MSI_NO_VECTOR)
110 				free_irq(pci_irq_vector(vp_dev->pci_dev, v),
111 					vq);
112 		}
113 		vp_dev->del_vq(vq);
114 	}
115 }
116 
117 /* the config->del_vqs() implementation */
118 void vp_del_vqs(struct virtio_device *vdev)
119 {
120 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
121 	int i;
122 
123 	if (WARN_ON_ONCE(list_empty_careful(&vdev->vqs)))
124 		return;
125 
126 	vp_remove_vqs(vdev);
127 
128 	if (vp_dev->pci_dev->msix_enabled) {
129 		for (i = 0; i < vp_dev->msix_vectors; i++)
130 			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
131 
132 		/* Disable the vector used for configuration */
133 		vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
134 
135 		kfree(vp_dev->msix_affinity_masks);
136 		kfree(vp_dev->msix_names);
137 		kfree(vp_dev->msix_vector_map);
138 	}
139 
140 	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
141 	pci_free_irq_vectors(vp_dev->pci_dev);
142 }
143 
144 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
145 		struct virtqueue *vqs[], vq_callback_t *callbacks[],
146 		const char * const names[], struct irq_affinity *desc)
147 {
148 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
149 	const char *name = dev_name(&vp_dev->vdev.dev);
150 	int i, j, err = -ENOMEM, allocated_vectors, nvectors;
151 	unsigned flags = PCI_IRQ_MSIX;
152 	bool shared = false;
153 	u16 msix_vec;
154 
155 	if (desc) {
156 		flags |= PCI_IRQ_AFFINITY;
157 		desc->pre_vectors++; /* virtio config vector */
158 	}
159 
160 	nvectors = 1;
161 	for (i = 0; i < nvqs; i++)
162 		if (callbacks[i])
163 			nvectors++;
164 
165 	/* Try one vector per queue first. */
166 	err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
167 			nvectors, flags, desc);
168 	if (err < 0) {
169 		/* Fallback to one vector for config, one shared for queues. */
170 		shared = true;
171 		err = pci_alloc_irq_vectors(vp_dev->pci_dev, 2, 2,
172 				PCI_IRQ_MSIX);
173 		if (err < 0)
174 			return err;
175 	}
176 	if (err < 0)
177 		return err;
178 
179 	vp_dev->msix_vectors = nvectors;
180 	vp_dev->msix_names = kmalloc_array(nvectors,
181 			sizeof(*vp_dev->msix_names), GFP_KERNEL);
182 	if (!vp_dev->msix_names)
183 		goto out_free_irq_vectors;
184 
185 	vp_dev->msix_affinity_masks = kcalloc(nvectors,
186 			sizeof(*vp_dev->msix_affinity_masks), GFP_KERNEL);
187 	if (!vp_dev->msix_affinity_masks)
188 		goto out_free_msix_names;
189 
190 	for (i = 0; i < nvectors; ++i) {
191 		if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
192 				GFP_KERNEL))
193 			goto out_free_msix_affinity_masks;
194 	}
195 
196 	/* Set the vector used for configuration */
197 	snprintf(vp_dev->msix_names[0], sizeof(*vp_dev->msix_names),
198 		 "%s-config", name);
199 	err = request_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_config_changed,
200 			0, vp_dev->msix_names[0], vp_dev);
201 	if (err)
202 		goto out_free_msix_affinity_masks;
203 
204 	/* Verify we had enough resources to assign the vector */
205 	if (vp_dev->config_vector(vp_dev, 0) == VIRTIO_MSI_NO_VECTOR) {
206 		err = -EBUSY;
207 		goto out_free_config_irq;
208 	}
209 
210 	vp_dev->msix_vector_map = kmalloc_array(nvqs,
211 			sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
212 	if (!vp_dev->msix_vector_map)
213 		goto out_disable_config_irq;
214 
215 	allocated_vectors = j = 1; /* vector 0 is the config interrupt */
216 	for (i = 0; i < nvqs; ++i) {
217 		if (!names[i]) {
218 			vqs[i] = NULL;
219 			continue;
220 		}
221 
222 		if (callbacks[i])
223 			msix_vec = allocated_vectors;
224 		else
225 			msix_vec = VIRTIO_MSI_NO_VECTOR;
226 
227 		vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
228 				msix_vec);
229 		if (IS_ERR(vqs[i])) {
230 			err = PTR_ERR(vqs[i]);
231 			goto out_remove_vqs;
232 		}
233 
234 		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
235 			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
236 			continue;
237 		}
238 
239 		snprintf(vp_dev->msix_names[j],
240 			 sizeof(*vp_dev->msix_names), "%s-%s",
241 			 dev_name(&vp_dev->vdev.dev), names[i]);
242 		err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
243 				  vring_interrupt, IRQF_SHARED,
244 				  vp_dev->msix_names[j], vqs[i]);
245 		if (err) {
246 			/* don't free this irq on error */
247 			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
248 			goto out_remove_vqs;
249 		}
250 		vp_dev->msix_vector_map[i] = msix_vec;
251 		j++;
252 
253 		/*
254 		 * Use a different vector for each queue if they are available,
255 		 * else share the same vector for all VQs.
256 		 */
257 		if (!shared)
258 			allocated_vectors++;
259 	}
260 
261 	return 0;
262 
263 out_remove_vqs:
264 	vp_remove_vqs(vdev);
265 	kfree(vp_dev->msix_vector_map);
266 out_disable_config_irq:
267 	vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
268 out_free_config_irq:
269 	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
270 out_free_msix_affinity_masks:
271 	for (i = 0; i < nvectors; i++) {
272 		if (vp_dev->msix_affinity_masks[i])
273 			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
274 	}
275 	kfree(vp_dev->msix_affinity_masks);
276 out_free_msix_names:
277 	kfree(vp_dev->msix_names);
278 out_free_irq_vectors:
279 	pci_free_irq_vectors(vp_dev->pci_dev);
280 	return err;
281 }
282 
283 static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
284 		struct virtqueue *vqs[], vq_callback_t *callbacks[],
285 		const char * const names[])
286 {
287 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
288 	int i, err;
289 
290 	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
291 			dev_name(&vdev->dev), vp_dev);
292 	if (err)
293 		return err;
294 
295 	for (i = 0; i < nvqs; ++i) {
296 		if (!names[i]) {
297 			vqs[i] = NULL;
298 			continue;
299 		}
300 		vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
301 				VIRTIO_MSI_NO_VECTOR);
302 		if (IS_ERR(vqs[i])) {
303 			err = PTR_ERR(vqs[i]);
304 			goto out_remove_vqs;
305 		}
306 	}
307 
308 	return 0;
309 
310 out_remove_vqs:
311 	vp_remove_vqs(vdev);
312 	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
313 	return err;
314 }
315 
316 /* the config->find_vqs() implementation */
317 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
318 		struct virtqueue *vqs[], vq_callback_t *callbacks[],
319 		const char * const names[], struct irq_affinity *desc)
320 {
321 	int err;
322 
323 	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, desc);
324 	if (!err)
325 		return 0;
326 	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
327 }
328 
329 const char *vp_bus_name(struct virtio_device *vdev)
330 {
331 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
332 
333 	return pci_name(vp_dev->pci_dev);
334 }
335 
336 /* Setup the affinity for a virtqueue:
337  * - force the affinity for per vq vector
338  * - OR over all affinities for shared MSI
339  * - ignore the affinity request if we're using INTX
340  */
341 int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
342 {
343 	struct virtio_device *vdev = vq->vdev;
344 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
345 
346 	if (!vq->callback)
347 		return -EINVAL;
348 
349 	if (vp_dev->pci_dev->msix_enabled) {
350 		int vec = vp_dev->msix_vector_map[vq->index];
351 		struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
352 		unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
353 
354 		if (cpu == -1)
355 			irq_set_affinity_hint(irq, NULL);
356 		else {
357 			cpumask_clear(mask);
358 			cpumask_set_cpu(cpu, mask);
359 			irq_set_affinity_hint(irq, mask);
360 		}
361 	}
362 	return 0;
363 }
364 
365 const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
366 {
367 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
368 	unsigned int *map = vp_dev->msix_vector_map;
369 
370 	if (!map || map[index] == VIRTIO_MSI_NO_VECTOR)
371 		return NULL;
372 
373 	return pci_irq_get_affinity(vp_dev->pci_dev, map[index]);
374 }
375 
376 #ifdef CONFIG_PM_SLEEP
377 static int virtio_pci_freeze(struct device *dev)
378 {
379 	struct pci_dev *pci_dev = to_pci_dev(dev);
380 	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
381 	int ret;
382 
383 	ret = virtio_device_freeze(&vp_dev->vdev);
384 
385 	if (!ret)
386 		pci_disable_device(pci_dev);
387 	return ret;
388 }
389 
390 static int virtio_pci_restore(struct device *dev)
391 {
392 	struct pci_dev *pci_dev = to_pci_dev(dev);
393 	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
394 	int ret;
395 
396 	ret = pci_enable_device(pci_dev);
397 	if (ret)
398 		return ret;
399 
400 	pci_set_master(pci_dev);
401 	return virtio_device_restore(&vp_dev->vdev);
402 }
403 
404 static const struct dev_pm_ops virtio_pci_pm_ops = {
405 	SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
406 };
407 #endif
408 
409 
410 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
411 static const struct pci_device_id virtio_pci_id_table[] = {
412 	{ PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
413 	{ 0 }
414 };
415 
416 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
417 
418 static void virtio_pci_release_dev(struct device *_d)
419 {
420 	struct virtio_device *vdev = dev_to_virtio(_d);
421 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
422 
423 	/* As struct device is a kobject, it's not safe to
424 	 * free the memory (including the reference counter itself)
425 	 * until it's release callback. */
426 	kfree(vp_dev);
427 }
428 
429 static int virtio_pci_probe(struct pci_dev *pci_dev,
430 			    const struct pci_device_id *id)
431 {
432 	struct virtio_pci_device *vp_dev;
433 	int rc;
434 
435 	/* allocate our structure and fill it out */
436 	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
437 	if (!vp_dev)
438 		return -ENOMEM;
439 
440 	pci_set_drvdata(pci_dev, vp_dev);
441 	vp_dev->vdev.dev.parent = &pci_dev->dev;
442 	vp_dev->vdev.dev.release = virtio_pci_release_dev;
443 	vp_dev->pci_dev = pci_dev;
444 
445 	/* enable the device */
446 	rc = pci_enable_device(pci_dev);
447 	if (rc)
448 		goto err_enable_device;
449 
450 	if (force_legacy) {
451 		rc = virtio_pci_legacy_probe(vp_dev);
452 		/* Also try modern mode if we can't map BAR0 (no IO space). */
453 		if (rc == -ENODEV || rc == -ENOMEM)
454 			rc = virtio_pci_modern_probe(vp_dev);
455 		if (rc)
456 			goto err_probe;
457 	} else {
458 		rc = virtio_pci_modern_probe(vp_dev);
459 		if (rc == -ENODEV)
460 			rc = virtio_pci_legacy_probe(vp_dev);
461 		if (rc)
462 			goto err_probe;
463 	}
464 
465 	pci_set_master(pci_dev);
466 
467 	rc = register_virtio_device(&vp_dev->vdev);
468 	if (rc)
469 		goto err_register;
470 
471 	return 0;
472 
473 err_register:
474 	if (vp_dev->ioaddr)
475 	     virtio_pci_legacy_remove(vp_dev);
476 	else
477 	     virtio_pci_modern_remove(vp_dev);
478 err_probe:
479 	pci_disable_device(pci_dev);
480 err_enable_device:
481 	kfree(vp_dev);
482 	return rc;
483 }
484 
485 static void virtio_pci_remove(struct pci_dev *pci_dev)
486 {
487 	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
488 	struct device *dev = get_device(&vp_dev->vdev.dev);
489 
490 	unregister_virtio_device(&vp_dev->vdev);
491 
492 	if (vp_dev->ioaddr)
493 		virtio_pci_legacy_remove(vp_dev);
494 	else
495 		virtio_pci_modern_remove(vp_dev);
496 
497 	pci_disable_device(pci_dev);
498 	put_device(dev);
499 }
500 
501 static struct pci_driver virtio_pci_driver = {
502 	.name		= "virtio-pci",
503 	.id_table	= virtio_pci_id_table,
504 	.probe		= virtio_pci_probe,
505 	.remove		= virtio_pci_remove,
506 #ifdef CONFIG_PM_SLEEP
507 	.driver.pm	= &virtio_pci_pm_ops,
508 #endif
509 };
510 
511 module_pci_driver(virtio_pci_driver);
512 
513 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
514 MODULE_DESCRIPTION("virtio-pci");
515 MODULE_LICENSE("GPL");
516 MODULE_VERSION("1");
517