Linux Audio

Check our new training course

Loading...
v4.6
 
  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
 22static bool force_legacy = false;
 23
 24#if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
 25module_param(force_legacy, bool, 0444);
 26MODULE_PARM_DESC(force_legacy,
 27		 "Force legacy mode for transitional virtio 1 devices");
 28#endif
 29
 30/* wait for pending irq handlers */
 31void vp_synchronize_vectors(struct virtio_device *vdev)
 32{
 33	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 34	int i;
 35
 36	if (vp_dev->intx_enabled)
 37		synchronize_irq(vp_dev->pci_dev->irq);
 38
 39	for (i = 0; i < vp_dev->msix_vectors; ++i)
 40		synchronize_irq(vp_dev->msix_entries[i].vector);
 41}
 42
 43/* the notify function used when creating a virt queue */
 44bool vp_notify(struct virtqueue *vq)
 45{
 46	/* we write the queue's selector into the notification register to
 47	 * signal the other end */
 48	iowrite16(vq->index, (void __iomem *)vq->priv);
 49	return true;
 50}
 51
 52/* Handle a configuration change: Tell driver if it wants to know. */
 53static irqreturn_t vp_config_changed(int irq, void *opaque)
 54{
 55	struct virtio_pci_device *vp_dev = opaque;
 56
 57	virtio_config_changed(&vp_dev->vdev);
 58	return IRQ_HANDLED;
 59}
 60
 61/* Notify all virtqueues on an interrupt. */
 62static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 63{
 64	struct virtio_pci_device *vp_dev = opaque;
 65	struct virtio_pci_vq_info *info;
 66	irqreturn_t ret = IRQ_NONE;
 67	unsigned long flags;
 68
 69	spin_lock_irqsave(&vp_dev->lock, flags);
 70	list_for_each_entry(info, &vp_dev->virtqueues, node) {
 71		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 72			ret = IRQ_HANDLED;
 73	}
 74	spin_unlock_irqrestore(&vp_dev->lock, flags);
 75
 76	return ret;
 77}
 78
 79/* A small wrapper to also acknowledge the interrupt when it's handled.
 80 * I really need an EIO hook for the vring so I can ack the interrupt once we
 81 * know that we'll be handling the IRQ but before we invoke the callback since
 82 * the callback may notify the host which results in the host attempting to
 83 * raise an interrupt that we would then mask once we acknowledged the
 84 * interrupt. */
 85static irqreturn_t vp_interrupt(int irq, void *opaque)
 86{
 87	struct virtio_pci_device *vp_dev = opaque;
 88	u8 isr;
 89
 90	/* reading the ISR has the effect of also clearing it so it's very
 91	 * important to save off the value. */
 92	isr = ioread8(vp_dev->isr);
 93
 94	/* It's definitely not us if the ISR was not high */
 95	if (!isr)
 96		return IRQ_NONE;
 97
 98	/* Configuration change?  Tell driver if it wants to know. */
 99	if (isr & VIRTIO_PCI_ISR_CONFIG)
100		vp_config_changed(irq, opaque);
101
102	return vp_vring_interrupt(irq, opaque);
103}
104
105static void vp_free_vectors(struct virtio_device *vdev)
106{
107	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
108	int i;
109
110	if (vp_dev->intx_enabled) {
111		free_irq(vp_dev->pci_dev->irq, vp_dev);
112		vp_dev->intx_enabled = 0;
113	}
114
115	for (i = 0; i < vp_dev->msix_used_vectors; ++i)
116		free_irq(vp_dev->msix_entries[i].vector, vp_dev);
117
118	for (i = 0; i < vp_dev->msix_vectors; i++)
119		if (vp_dev->msix_affinity_masks[i])
120			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
121
122	if (vp_dev->msix_enabled) {
123		/* Disable the vector used for configuration */
124		vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
125
126		pci_disable_msix(vp_dev->pci_dev);
127		vp_dev->msix_enabled = 0;
128	}
129
130	vp_dev->msix_vectors = 0;
131	vp_dev->msix_used_vectors = 0;
132	kfree(vp_dev->msix_names);
133	vp_dev->msix_names = NULL;
134	kfree(vp_dev->msix_entries);
135	vp_dev->msix_entries = NULL;
136	kfree(vp_dev->msix_affinity_masks);
137	vp_dev->msix_affinity_masks = NULL;
138}
139
140static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
141				   bool per_vq_vectors)
142{
143	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
144	const char *name = dev_name(&vp_dev->vdev.dev);
 
145	unsigned i, v;
146	int err = -ENOMEM;
147
148	vp_dev->msix_vectors = nvectors;
149
150	vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
151				       GFP_KERNEL);
152	if (!vp_dev->msix_entries)
153		goto error;
154	vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
155				     GFP_KERNEL);
156	if (!vp_dev->msix_names)
157		goto error;
158	vp_dev->msix_affinity_masks
159		= kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
160			  GFP_KERNEL);
161	if (!vp_dev->msix_affinity_masks)
162		goto error;
163	for (i = 0; i < nvectors; ++i)
164		if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
165					GFP_KERNEL))
166			goto error;
167
168	for (i = 0; i < nvectors; ++i)
169		vp_dev->msix_entries[i].entry = i;
 
 
170
171	err = pci_enable_msix_exact(vp_dev->pci_dev,
172				    vp_dev->msix_entries, nvectors);
173	if (err)
174		goto error;
175	vp_dev->msix_enabled = 1;
176
177	/* Set the vector used for configuration */
178	v = vp_dev->msix_used_vectors;
179	snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
180		 "%s-config", name);
181	err = request_irq(vp_dev->msix_entries[v].vector,
182			  vp_config_changed, 0, vp_dev->msix_names[v],
183			  vp_dev);
184	if (err)
185		goto error;
186	++vp_dev->msix_used_vectors;
187
188	v = vp_dev->config_vector(vp_dev, v);
189	/* Verify we had enough resources to assign the vector */
190	if (v == VIRTIO_MSI_NO_VECTOR) {
191		err = -EBUSY;
192		goto error;
193	}
194
195	if (!per_vq_vectors) {
196		/* Shared vector for all VQs */
197		v = vp_dev->msix_used_vectors;
198		snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
199			 "%s-virtqueues", name);
200		err = request_irq(vp_dev->msix_entries[v].vector,
201				  vp_vring_interrupt, 0, vp_dev->msix_names[v],
202				  vp_dev);
203		if (err)
204			goto error;
205		++vp_dev->msix_used_vectors;
206	}
207	return 0;
208error:
209	vp_free_vectors(vdev);
210	return err;
211}
212
213static int vp_request_intx(struct virtio_device *vdev)
214{
215	int err;
216	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
217
218	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
219			  IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
220	if (!err)
221		vp_dev->intx_enabled = 1;
222	return err;
223}
224
225static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index,
226				     void (*callback)(struct virtqueue *vq),
227				     const char *name,
 
228				     u16 msix_vec)
229{
230	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
231	struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
232	struct virtqueue *vq;
233	unsigned long flags;
234
235	/* fill out our structure that represents an active queue */
236	if (!info)
237		return ERR_PTR(-ENOMEM);
238
239	vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, msix_vec);
 
240	if (IS_ERR(vq))
241		goto out_info;
242
243	info->vq = vq;
244	if (callback) {
245		spin_lock_irqsave(&vp_dev->lock, flags);
246		list_add(&info->node, &vp_dev->virtqueues);
247		spin_unlock_irqrestore(&vp_dev->lock, flags);
248	} else {
249		INIT_LIST_HEAD(&info->node);
250	}
251
252	vp_dev->vqs[index] = info;
253	return vq;
254
255out_info:
256	kfree(info);
257	return vq;
258}
259
260static void vp_del_vq(struct virtqueue *vq)
261{
262	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
263	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
264	unsigned long flags;
265
266	spin_lock_irqsave(&vp_dev->lock, flags);
267	list_del(&info->node);
268	spin_unlock_irqrestore(&vp_dev->lock, flags);
269
270	vp_dev->del_vq(info);
271	kfree(info);
272}
273
274/* the config->del_vqs() implementation */
275void vp_del_vqs(struct virtio_device *vdev)
276{
277	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
278	struct virtqueue *vq, *n;
279	struct virtio_pci_vq_info *info;
280
281	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
282		info = vp_dev->vqs[vq->index];
283		if (vp_dev->per_vq_vectors &&
284			info->msix_vector != VIRTIO_MSI_NO_VECTOR)
285			free_irq(vp_dev->msix_entries[info->msix_vector].vector,
286				 vq);
 
 
 
 
 
287		vp_del_vq(vq);
288	}
289	vp_dev->per_vq_vectors = false;
290
291	vp_free_vectors(vdev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
292	kfree(vp_dev->vqs);
293	vp_dev->vqs = NULL;
294}
295
296static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
297			      struct virtqueue *vqs[],
298			      vq_callback_t *callbacks[],
299			      const char * const names[],
300			      bool use_msix,
301			      bool per_vq_vectors)
302{
303	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
304	u16 msix_vec;
305	int i, err, nvectors, allocated_vectors;
306
307	vp_dev->vqs = kmalloc(nvqs * sizeof *vp_dev->vqs, GFP_KERNEL);
308	if (!vp_dev->vqs)
309		return -ENOMEM;
310
311	if (!use_msix) {
312		/* Old style: one normal interrupt for change and all vqs. */
313		err = vp_request_intx(vdev);
314		if (err)
315			goto error_find;
 
316	} else {
317		if (per_vq_vectors) {
318			/* Best option: one for change interrupt, one per vq. */
319			nvectors = 1;
320			for (i = 0; i < nvqs; ++i)
321				if (callbacks[i])
322					++nvectors;
323		} else {
324			/* Second best: one for change, shared for all vqs. */
325			nvectors = 2;
326		}
327
328		err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
329		if (err)
330			goto error_find;
331	}
332
 
 
 
 
 
333	vp_dev->per_vq_vectors = per_vq_vectors;
334	allocated_vectors = vp_dev->msix_used_vectors;
335	for (i = 0; i < nvqs; ++i) {
336		if (!names[i]) {
337			vqs[i] = NULL;
338			continue;
339		} else if (!callbacks[i] || !vp_dev->msix_enabled)
 
 
340			msix_vec = VIRTIO_MSI_NO_VECTOR;
341		else if (vp_dev->per_vq_vectors)
342			msix_vec = allocated_vectors++;
343		else
344			msix_vec = VP_MSIX_VQ_VECTOR;
345		vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 
 
346		if (IS_ERR(vqs[i])) {
347			err = PTR_ERR(vqs[i]);
348			goto error_find;
349		}
350
351		if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
352			continue;
353
354		/* allocate per-vq irq if available and necessary */
355		snprintf(vp_dev->msix_names[msix_vec],
356			 sizeof *vp_dev->msix_names,
357			 "%s-%s",
358			 dev_name(&vp_dev->vdev.dev), names[i]);
359		err = request_irq(vp_dev->msix_entries[msix_vec].vector,
360				  vring_interrupt, 0,
361				  vp_dev->msix_names[msix_vec],
362				  vqs[i]);
363		if (err) {
364			vp_del_vq(vqs[i]);
365			goto error_find;
366		}
367	}
368	return 0;
369
370error_find:
371	vp_del_vqs(vdev);
372	return err;
373}
374
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
375/* the config->find_vqs() implementation */
376int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
377		struct virtqueue *vqs[],
378		vq_callback_t *callbacks[],
379		const char * const names[])
380{
381	int err;
382
383	/* Try MSI-X with one vector per queue. */
384	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
385	if (!err)
386		return 0;
387	/* Fallback: MSI-X with one vector for config, one shared for queues. */
388	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
389				 true, false);
390	if (!err)
391		return 0;
392	/* Finally fall back to regular interrupts. */
393	return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
394				  false, false);
395}
396
397const char *vp_bus_name(struct virtio_device *vdev)
398{
399	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
400
401	return pci_name(vp_dev->pci_dev);
402}
403
404/* Setup the affinity for a virtqueue:
405 * - force the affinity for per vq vector
406 * - OR over all affinities for shared MSI
407 * - ignore the affinity request if we're using INTX
408 */
409int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
410{
411	struct virtio_device *vdev = vq->vdev;
412	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
413	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
414	struct cpumask *mask;
415	unsigned int irq;
416
417	if (!vq->callback)
418		return -EINVAL;
419
420	if (vp_dev->msix_enabled) {
421		mask = vp_dev->msix_affinity_masks[info->msix_vector];
422		irq = vp_dev->msix_entries[info->msix_vector].vector;
423		if (cpu == -1)
424			irq_set_affinity_hint(irq, NULL);
425		else {
426			cpumask_clear(mask);
427			cpumask_set_cpu(cpu, mask);
428			irq_set_affinity_hint(irq, mask);
429		}
430	}
431	return 0;
432}
433
 
 
 
 
 
 
 
 
 
 
 
 
434#ifdef CONFIG_PM_SLEEP
435static int virtio_pci_freeze(struct device *dev)
436{
437	struct pci_dev *pci_dev = to_pci_dev(dev);
438	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
439	int ret;
440
441	ret = virtio_device_freeze(&vp_dev->vdev);
442
443	if (!ret)
444		pci_disable_device(pci_dev);
445	return ret;
446}
447
448static int virtio_pci_restore(struct device *dev)
449{
450	struct pci_dev *pci_dev = to_pci_dev(dev);
451	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
452	int ret;
453
454	ret = pci_enable_device(pci_dev);
455	if (ret)
456		return ret;
457
458	pci_set_master(pci_dev);
459	return virtio_device_restore(&vp_dev->vdev);
460}
461
462static const struct dev_pm_ops virtio_pci_pm_ops = {
463	SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
464};
465#endif
466
467
468/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
469static const struct pci_device_id virtio_pci_id_table[] = {
470	{ PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
471	{ 0 }
472};
473
474MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
475
476static void virtio_pci_release_dev(struct device *_d)
477{
478	struct virtio_device *vdev = dev_to_virtio(_d);
479	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
480
481	/* As struct device is a kobject, it's not safe to
482	 * free the memory (including the reference counter itself)
483	 * until it's release callback. */
484	kfree(vp_dev);
485}
486
487static int virtio_pci_probe(struct pci_dev *pci_dev,
488			    const struct pci_device_id *id)
489{
490	struct virtio_pci_device *vp_dev;
491	int rc;
492
493	/* allocate our structure and fill it out */
494	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
495	if (!vp_dev)
496		return -ENOMEM;
497
498	pci_set_drvdata(pci_dev, vp_dev);
499	vp_dev->vdev.dev.parent = &pci_dev->dev;
500	vp_dev->vdev.dev.release = virtio_pci_release_dev;
501	vp_dev->pci_dev = pci_dev;
502	INIT_LIST_HEAD(&vp_dev->virtqueues);
503	spin_lock_init(&vp_dev->lock);
504
505	/* enable the device */
506	rc = pci_enable_device(pci_dev);
507	if (rc)
508		goto err_enable_device;
509
510	if (force_legacy) {
511		rc = virtio_pci_legacy_probe(vp_dev);
512		/* Also try modern mode if we can't map BAR0 (no IO space). */
513		if (rc == -ENODEV || rc == -ENOMEM)
514			rc = virtio_pci_modern_probe(vp_dev);
515		if (rc)
516			goto err_probe;
517	} else {
518		rc = virtio_pci_modern_probe(vp_dev);
519		if (rc == -ENODEV)
520			rc = virtio_pci_legacy_probe(vp_dev);
521		if (rc)
522			goto err_probe;
523	}
524
525	pci_set_master(pci_dev);
526
527	rc = register_virtio_device(&vp_dev->vdev);
 
528	if (rc)
529		goto err_register;
530
531	return 0;
532
533err_register:
534	if (vp_dev->ioaddr)
535	     virtio_pci_legacy_remove(vp_dev);
536	else
537	     virtio_pci_modern_remove(vp_dev);
538err_probe:
539	pci_disable_device(pci_dev);
540err_enable_device:
541	kfree(vp_dev);
 
 
 
542	return rc;
543}
544
545static void virtio_pci_remove(struct pci_dev *pci_dev)
546{
547	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
548	struct device *dev = get_device(&vp_dev->vdev.dev);
549
 
 
550	unregister_virtio_device(&vp_dev->vdev);
551
552	if (vp_dev->ioaddr)
553		virtio_pci_legacy_remove(vp_dev);
554	else
555		virtio_pci_modern_remove(vp_dev);
556
557	pci_disable_device(pci_dev);
558	put_device(dev);
559}
560
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
561static struct pci_driver virtio_pci_driver = {
562	.name		= "virtio-pci",
563	.id_table	= virtio_pci_id_table,
564	.probe		= virtio_pci_probe,
565	.remove		= virtio_pci_remove,
566#ifdef CONFIG_PM_SLEEP
567	.driver.pm	= &virtio_pci_pm_ops,
568#endif
 
569};
570
571module_pci_driver(virtio_pci_driver);
572
573MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
574MODULE_DESCRIPTION("virtio-pci");
575MODULE_LICENSE("GPL");
576MODULE_VERSION("1");
v5.9
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * Virtio PCI driver - common functionality for all device versions
  4 *
  5 * This module allows virtio devices to be used over a virtual PCI device.
  6 * This can be used with QEMU based VMMs like KVM or Xen.
  7 *
  8 * Copyright IBM Corp. 2007
  9 * Copyright Red Hat, Inc. 2014
 10 *
 11 * Authors:
 12 *  Anthony Liguori  <aliguori@us.ibm.com>
 13 *  Rusty Russell <rusty@rustcorp.com.au>
 14 *  Michael S. Tsirkin <mst@redhat.com>
 
 
 
 
 15 */
 16
 17#include "virtio_pci_common.h"
 18
 19static bool force_legacy = false;
 20
 21#if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
 22module_param(force_legacy, bool, 0444);
 23MODULE_PARM_DESC(force_legacy,
 24		 "Force legacy mode for transitional virtio 1 devices");
 25#endif
 26
 27/* wait for pending irq handlers */
 28void vp_synchronize_vectors(struct virtio_device *vdev)
 29{
 30	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 31	int i;
 32
 33	if (vp_dev->intx_enabled)
 34		synchronize_irq(vp_dev->pci_dev->irq);
 35
 36	for (i = 0; i < vp_dev->msix_vectors; ++i)
 37		synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
 38}
 39
 40/* the notify function used when creating a virt queue */
 41bool vp_notify(struct virtqueue *vq)
 42{
 43	/* we write the queue's selector into the notification register to
 44	 * signal the other end */
 45	iowrite16(vq->index, (void __iomem *)vq->priv);
 46	return true;
 47}
 48
 49/* Handle a configuration change: Tell driver if it wants to know. */
 50static irqreturn_t vp_config_changed(int irq, void *opaque)
 51{
 52	struct virtio_pci_device *vp_dev = opaque;
 53
 54	virtio_config_changed(&vp_dev->vdev);
 55	return IRQ_HANDLED;
 56}
 57
 58/* Notify all virtqueues on an interrupt. */
 59static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 60{
 61	struct virtio_pci_device *vp_dev = opaque;
 62	struct virtio_pci_vq_info *info;
 63	irqreturn_t ret = IRQ_NONE;
 64	unsigned long flags;
 65
 66	spin_lock_irqsave(&vp_dev->lock, flags);
 67	list_for_each_entry(info, &vp_dev->virtqueues, node) {
 68		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 69			ret = IRQ_HANDLED;
 70	}
 71	spin_unlock_irqrestore(&vp_dev->lock, flags);
 72
 73	return ret;
 74}
 75
 76/* A small wrapper to also acknowledge the interrupt when it's handled.
 77 * I really need an EIO hook for the vring so I can ack the interrupt once we
 78 * know that we'll be handling the IRQ but before we invoke the callback since
 79 * the callback may notify the host which results in the host attempting to
 80 * raise an interrupt that we would then mask once we acknowledged the
 81 * interrupt. */
 82static irqreturn_t vp_interrupt(int irq, void *opaque)
 83{
 84	struct virtio_pci_device *vp_dev = opaque;
 85	u8 isr;
 86
 87	/* reading the ISR has the effect of also clearing it so it's very
 88	 * important to save off the value. */
 89	isr = ioread8(vp_dev->isr);
 90
 91	/* It's definitely not us if the ISR was not high */
 92	if (!isr)
 93		return IRQ_NONE;
 94
 95	/* Configuration change?  Tell driver if it wants to know. */
 96	if (isr & VIRTIO_PCI_ISR_CONFIG)
 97		vp_config_changed(irq, opaque);
 98
 99	return vp_vring_interrupt(irq, opaque);
100}
101
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
102static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
103				   bool per_vq_vectors, struct irq_affinity *desc)
104{
105	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
106	const char *name = dev_name(&vp_dev->vdev.dev);
107	unsigned flags = PCI_IRQ_MSIX;
108	unsigned i, v;
109	int err = -ENOMEM;
110
111	vp_dev->msix_vectors = nvectors;
112
113	vp_dev->msix_names = kmalloc_array(nvectors,
114					   sizeof(*vp_dev->msix_names),
115					   GFP_KERNEL);
 
 
 
116	if (!vp_dev->msix_names)
117		goto error;
118	vp_dev->msix_affinity_masks
119		= kcalloc(nvectors, sizeof(*vp_dev->msix_affinity_masks),
120			  GFP_KERNEL);
121	if (!vp_dev->msix_affinity_masks)
122		goto error;
123	for (i = 0; i < nvectors; ++i)
124		if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
125					GFP_KERNEL))
126			goto error;
127
128	if (desc) {
129		flags |= PCI_IRQ_AFFINITY;
130		desc->pre_vectors++; /* virtio config vector */
131	}
132
133	err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
134					     nvectors, flags, desc);
135	if (err < 0)
136		goto error;
137	vp_dev->msix_enabled = 1;
138
139	/* Set the vector used for configuration */
140	v = vp_dev->msix_used_vectors;
141	snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
142		 "%s-config", name);
143	err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
144			  vp_config_changed, 0, vp_dev->msix_names[v],
145			  vp_dev);
146	if (err)
147		goto error;
148	++vp_dev->msix_used_vectors;
149
150	v = vp_dev->config_vector(vp_dev, v);
151	/* Verify we had enough resources to assign the vector */
152	if (v == VIRTIO_MSI_NO_VECTOR) {
153		err = -EBUSY;
154		goto error;
155	}
156
157	if (!per_vq_vectors) {
158		/* Shared vector for all VQs */
159		v = vp_dev->msix_used_vectors;
160		snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
161			 "%s-virtqueues", name);
162		err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
163				  vp_vring_interrupt, 0, vp_dev->msix_names[v],
164				  vp_dev);
165		if (err)
166			goto error;
167		++vp_dev->msix_used_vectors;
168	}
169	return 0;
170error:
 
 
 
 
 
 
 
 
 
 
 
 
 
171	return err;
172}
173
174static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index,
175				     void (*callback)(struct virtqueue *vq),
176				     const char *name,
177				     bool ctx,
178				     u16 msix_vec)
179{
180	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
181	struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
182	struct virtqueue *vq;
183	unsigned long flags;
184
185	/* fill out our structure that represents an active queue */
186	if (!info)
187		return ERR_PTR(-ENOMEM);
188
189	vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, ctx,
190			      msix_vec);
191	if (IS_ERR(vq))
192		goto out_info;
193
194	info->vq = vq;
195	if (callback) {
196		spin_lock_irqsave(&vp_dev->lock, flags);
197		list_add(&info->node, &vp_dev->virtqueues);
198		spin_unlock_irqrestore(&vp_dev->lock, flags);
199	} else {
200		INIT_LIST_HEAD(&info->node);
201	}
202
203	vp_dev->vqs[index] = info;
204	return vq;
205
206out_info:
207	kfree(info);
208	return vq;
209}
210
211static void vp_del_vq(struct virtqueue *vq)
212{
213	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
214	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
215	unsigned long flags;
216
217	spin_lock_irqsave(&vp_dev->lock, flags);
218	list_del(&info->node);
219	spin_unlock_irqrestore(&vp_dev->lock, flags);
220
221	vp_dev->del_vq(info);
222	kfree(info);
223}
224
225/* the config->del_vqs() implementation */
226void vp_del_vqs(struct virtio_device *vdev)
227{
228	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
229	struct virtqueue *vq, *n;
230	int i;
231
232	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
233		if (vp_dev->per_vq_vectors) {
234			int v = vp_dev->vqs[vq->index]->msix_vector;
235
236			if (v != VIRTIO_MSI_NO_VECTOR) {
237				int irq = pci_irq_vector(vp_dev->pci_dev, v);
238
239				irq_set_affinity_hint(irq, NULL);
240				free_irq(irq, vq);
241			}
242		}
243		vp_del_vq(vq);
244	}
245	vp_dev->per_vq_vectors = false;
246
247	if (vp_dev->intx_enabled) {
248		free_irq(vp_dev->pci_dev->irq, vp_dev);
249		vp_dev->intx_enabled = 0;
250	}
251
252	for (i = 0; i < vp_dev->msix_used_vectors; ++i)
253		free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
254
255	if (vp_dev->msix_affinity_masks) {
256		for (i = 0; i < vp_dev->msix_vectors; i++)
257			if (vp_dev->msix_affinity_masks[i])
258				free_cpumask_var(vp_dev->msix_affinity_masks[i]);
259	}
260
261	if (vp_dev->msix_enabled) {
262		/* Disable the vector used for configuration */
263		vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
264
265		pci_free_irq_vectors(vp_dev->pci_dev);
266		vp_dev->msix_enabled = 0;
267	}
268
269	vp_dev->msix_vectors = 0;
270	vp_dev->msix_used_vectors = 0;
271	kfree(vp_dev->msix_names);
272	vp_dev->msix_names = NULL;
273	kfree(vp_dev->msix_affinity_masks);
274	vp_dev->msix_affinity_masks = NULL;
275	kfree(vp_dev->vqs);
276	vp_dev->vqs = NULL;
277}
278
279static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
280		struct virtqueue *vqs[], vq_callback_t *callbacks[],
281		const char * const names[], bool per_vq_vectors,
282		const bool *ctx,
283		struct irq_affinity *desc)
 
284{
285	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
286	u16 msix_vec;
287	int i, err, nvectors, allocated_vectors, queue_idx = 0;
288
289	vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
290	if (!vp_dev->vqs)
291		return -ENOMEM;
292
293	if (per_vq_vectors) {
294		/* Best option: one for change interrupt, one per vq. */
295		nvectors = 1;
296		for (i = 0; i < nvqs; ++i)
297			if (names[i] && callbacks[i])
298				++nvectors;
299	} else {
300		/* Second best: one for change, shared for all vqs. */
301		nvectors = 2;
 
 
 
 
 
 
 
 
 
 
 
 
302	}
303
304	err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors,
305				      per_vq_vectors ? desc : NULL);
306	if (err)
307		goto error_find;
308
309	vp_dev->per_vq_vectors = per_vq_vectors;
310	allocated_vectors = vp_dev->msix_used_vectors;
311	for (i = 0; i < nvqs; ++i) {
312		if (!names[i]) {
313			vqs[i] = NULL;
314			continue;
315		}
316
317		if (!callbacks[i])
318			msix_vec = VIRTIO_MSI_NO_VECTOR;
319		else if (vp_dev->per_vq_vectors)
320			msix_vec = allocated_vectors++;
321		else
322			msix_vec = VP_MSIX_VQ_VECTOR;
323		vqs[i] = vp_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
324				     ctx ? ctx[i] : false,
325				     msix_vec);
326		if (IS_ERR(vqs[i])) {
327			err = PTR_ERR(vqs[i]);
328			goto error_find;
329		}
330
331		if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
332			continue;
333
334		/* allocate per-vq irq if available and necessary */
335		snprintf(vp_dev->msix_names[msix_vec],
336			 sizeof *vp_dev->msix_names,
337			 "%s-%s",
338			 dev_name(&vp_dev->vdev.dev), names[i]);
339		err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
340				  vring_interrupt, 0,
341				  vp_dev->msix_names[msix_vec],
342				  vqs[i]);
343		if (err)
 
344			goto error_find;
 
345	}
346	return 0;
347
348error_find:
349	vp_del_vqs(vdev);
350	return err;
351}
352
353static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
354		struct virtqueue *vqs[], vq_callback_t *callbacks[],
355		const char * const names[], const bool *ctx)
356{
357	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
358	int i, err, queue_idx = 0;
359
360	vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
361	if (!vp_dev->vqs)
362		return -ENOMEM;
363
364	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
365			dev_name(&vdev->dev), vp_dev);
366	if (err)
367		goto out_del_vqs;
368
369	vp_dev->intx_enabled = 1;
370	vp_dev->per_vq_vectors = false;
371	for (i = 0; i < nvqs; ++i) {
372		if (!names[i]) {
373			vqs[i] = NULL;
374			continue;
375		}
376		vqs[i] = vp_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
377				     ctx ? ctx[i] : false,
378				     VIRTIO_MSI_NO_VECTOR);
379		if (IS_ERR(vqs[i])) {
380			err = PTR_ERR(vqs[i]);
381			goto out_del_vqs;
382		}
383	}
384
385	return 0;
386out_del_vqs:
387	vp_del_vqs(vdev);
388	return err;
389}
390
391/* the config->find_vqs() implementation */
392int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
393		struct virtqueue *vqs[], vq_callback_t *callbacks[],
394		const char * const names[], const bool *ctx,
395		struct irq_affinity *desc)
396{
397	int err;
398
399	/* Try MSI-X with one vector per queue. */
400	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true, ctx, desc);
401	if (!err)
402		return 0;
403	/* Fallback: MSI-X with one vector for config, one shared for queues. */
404	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
 
405	if (!err)
406		return 0;
407	/* Finally fall back to regular interrupts. */
408	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
 
409}
410
411const char *vp_bus_name(struct virtio_device *vdev)
412{
413	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
414
415	return pci_name(vp_dev->pci_dev);
416}
417
418/* Setup the affinity for a virtqueue:
419 * - force the affinity for per vq vector
420 * - OR over all affinities for shared MSI
421 * - ignore the affinity request if we're using INTX
422 */
423int vp_set_vq_affinity(struct virtqueue *vq, const struct cpumask *cpu_mask)
424{
425	struct virtio_device *vdev = vq->vdev;
426	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
427	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
428	struct cpumask *mask;
429	unsigned int irq;
430
431	if (!vq->callback)
432		return -EINVAL;
433
434	if (vp_dev->msix_enabled) {
435		mask = vp_dev->msix_affinity_masks[info->msix_vector];
436		irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
437		if (!cpu_mask)
438			irq_set_affinity_hint(irq, NULL);
439		else {
440			cpumask_copy(mask, cpu_mask);
 
441			irq_set_affinity_hint(irq, mask);
442		}
443	}
444	return 0;
445}
446
447const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
448{
449	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
450
451	if (!vp_dev->per_vq_vectors ||
452	    vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR)
453		return NULL;
454
455	return pci_irq_get_affinity(vp_dev->pci_dev,
456				    vp_dev->vqs[index]->msix_vector);
457}
458
459#ifdef CONFIG_PM_SLEEP
460static int virtio_pci_freeze(struct device *dev)
461{
462	struct pci_dev *pci_dev = to_pci_dev(dev);
463	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
464	int ret;
465
466	ret = virtio_device_freeze(&vp_dev->vdev);
467
468	if (!ret)
469		pci_disable_device(pci_dev);
470	return ret;
471}
472
473static int virtio_pci_restore(struct device *dev)
474{
475	struct pci_dev *pci_dev = to_pci_dev(dev);
476	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
477	int ret;
478
479	ret = pci_enable_device(pci_dev);
480	if (ret)
481		return ret;
482
483	pci_set_master(pci_dev);
484	return virtio_device_restore(&vp_dev->vdev);
485}
486
487static const struct dev_pm_ops virtio_pci_pm_ops = {
488	SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
489};
490#endif
491
492
493/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
494static const struct pci_device_id virtio_pci_id_table[] = {
495	{ PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
496	{ 0 }
497};
498
499MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
500
501static void virtio_pci_release_dev(struct device *_d)
502{
503	struct virtio_device *vdev = dev_to_virtio(_d);
504	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
505
506	/* As struct device is a kobject, it's not safe to
507	 * free the memory (including the reference counter itself)
508	 * until it's release callback. */
509	kfree(vp_dev);
510}
511
512static int virtio_pci_probe(struct pci_dev *pci_dev,
513			    const struct pci_device_id *id)
514{
515	struct virtio_pci_device *vp_dev, *reg_dev = NULL;
516	int rc;
517
518	/* allocate our structure and fill it out */
519	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
520	if (!vp_dev)
521		return -ENOMEM;
522
523	pci_set_drvdata(pci_dev, vp_dev);
524	vp_dev->vdev.dev.parent = &pci_dev->dev;
525	vp_dev->vdev.dev.release = virtio_pci_release_dev;
526	vp_dev->pci_dev = pci_dev;
527	INIT_LIST_HEAD(&vp_dev->virtqueues);
528	spin_lock_init(&vp_dev->lock);
529
530	/* enable the device */
531	rc = pci_enable_device(pci_dev);
532	if (rc)
533		goto err_enable_device;
534
535	if (force_legacy) {
536		rc = virtio_pci_legacy_probe(vp_dev);
537		/* Also try modern mode if we can't map BAR0 (no IO space). */
538		if (rc == -ENODEV || rc == -ENOMEM)
539			rc = virtio_pci_modern_probe(vp_dev);
540		if (rc)
541			goto err_probe;
542	} else {
543		rc = virtio_pci_modern_probe(vp_dev);
544		if (rc == -ENODEV)
545			rc = virtio_pci_legacy_probe(vp_dev);
546		if (rc)
547			goto err_probe;
548	}
549
550	pci_set_master(pci_dev);
551
552	rc = register_virtio_device(&vp_dev->vdev);
553	reg_dev = vp_dev;
554	if (rc)
555		goto err_register;
556
557	return 0;
558
559err_register:
560	if (vp_dev->ioaddr)
561	     virtio_pci_legacy_remove(vp_dev);
562	else
563	     virtio_pci_modern_remove(vp_dev);
564err_probe:
565	pci_disable_device(pci_dev);
566err_enable_device:
567	if (reg_dev)
568		put_device(&vp_dev->vdev.dev);
569	else
570		kfree(vp_dev);
571	return rc;
572}
573
574static void virtio_pci_remove(struct pci_dev *pci_dev)
575{
576	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
577	struct device *dev = get_device(&vp_dev->vdev.dev);
578
579	pci_disable_sriov(pci_dev);
580
581	unregister_virtio_device(&vp_dev->vdev);
582
583	if (vp_dev->ioaddr)
584		virtio_pci_legacy_remove(vp_dev);
585	else
586		virtio_pci_modern_remove(vp_dev);
587
588	pci_disable_device(pci_dev);
589	put_device(dev);
590}
591
592static int virtio_pci_sriov_configure(struct pci_dev *pci_dev, int num_vfs)
593{
594	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
595	struct virtio_device *vdev = &vp_dev->vdev;
596	int ret;
597
598	if (!(vdev->config->get_status(vdev) & VIRTIO_CONFIG_S_DRIVER_OK))
599		return -EBUSY;
600
601	if (!__virtio_test_bit(vdev, VIRTIO_F_SR_IOV))
602		return -EINVAL;
603
604	if (pci_vfs_assigned(pci_dev))
605		return -EPERM;
606
607	if (num_vfs == 0) {
608		pci_disable_sriov(pci_dev);
609		return 0;
610	}
611
612	ret = pci_enable_sriov(pci_dev, num_vfs);
613	if (ret < 0)
614		return ret;
615
616	return num_vfs;
617}
618
619static struct pci_driver virtio_pci_driver = {
620	.name		= "virtio-pci",
621	.id_table	= virtio_pci_id_table,
622	.probe		= virtio_pci_probe,
623	.remove		= virtio_pci_remove,
624#ifdef CONFIG_PM_SLEEP
625	.driver.pm	= &virtio_pci_pm_ops,
626#endif
627	.sriov_configure = virtio_pci_sriov_configure,
628};
629
630module_pci_driver(virtio_pci_driver);
631
632MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
633MODULE_DESCRIPTION("virtio-pci");
634MODULE_LICENSE("GPL");
635MODULE_VERSION("1");