2 * Virtio PCI driver - common functionality for all device versions
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.
7 * Copyright IBM Corp. 2007
8 * Copyright Red Hat, Inc. 2014
11 * Anthony Liguori <aliguori@us.ibm.com>
12 * Rusty Russell <rusty@rustcorp.com.au>
13 * Michael S. Tsirkin <mst@redhat.com>
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.
20 #include "virtio_pci_common.h"
22 static bool force_legacy
= false;
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");
30 /* wait for pending irq handlers */
31 void vp_synchronize_vectors(struct virtio_device
*vdev
)
33 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
36 if (vp_dev
->intx_enabled
)
37 synchronize_irq(vp_dev
->pci_dev
->irq
);
39 for (i
= 0; i
< vp_dev
->msix_vectors
; ++i
)
40 synchronize_irq(pci_irq_vector(vp_dev
->pci_dev
, i
));
43 /* the notify function used when creating a virt queue */
44 bool vp_notify(struct virtqueue
*vq
)
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
);
52 /* Handle a configuration change: Tell driver if it wants to know. */
53 static irqreturn_t
vp_config_changed(int irq
, void *opaque
)
55 struct virtio_pci_device
*vp_dev
= opaque
;
57 virtio_config_changed(&vp_dev
->vdev
);
61 /* Notify all virtqueues on an interrupt. */
62 static irqreturn_t
vp_vring_interrupt(int irq
, void *opaque
)
64 struct virtio_pci_device
*vp_dev
= opaque
;
65 struct virtio_pci_vq_info
*info
;
66 irqreturn_t ret
= IRQ_NONE
;
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
)
74 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
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
85 static irqreturn_t
vp_interrupt(int irq
, void *opaque
)
87 struct virtio_pci_device
*vp_dev
= opaque
;
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
);
94 /* It's definitely not us if the ISR was not high */
98 /* Configuration change? Tell driver if it wants to know. */
99 if (isr
& VIRTIO_PCI_ISR_CONFIG
)
100 vp_config_changed(irq
, opaque
);
102 return vp_vring_interrupt(irq
, opaque
);
105 static int vp_request_msix_vectors(struct virtio_device
*vdev
, int nvectors
,
106 bool per_vq_vectors
, struct irq_affinity
*desc
)
108 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
109 const char *name
= dev_name(&vp_dev
->vdev
.dev
);
113 vp_dev
->msix_vectors
= nvectors
;
115 vp_dev
->msix_names
= kmalloc(nvectors
* sizeof *vp_dev
->msix_names
,
117 if (!vp_dev
->msix_names
)
119 vp_dev
->msix_affinity_masks
120 = kzalloc(nvectors
* sizeof *vp_dev
->msix_affinity_masks
,
122 if (!vp_dev
->msix_affinity_masks
)
124 for (i
= 0; i
< nvectors
; ++i
)
125 if (!alloc_cpumask_var(&vp_dev
->msix_affinity_masks
[i
],
129 err
= pci_alloc_irq_vectors_affinity(vp_dev
->pci_dev
, nvectors
,
130 nvectors
, PCI_IRQ_MSIX
|
131 (desc
? PCI_IRQ_AFFINITY
: 0),
135 vp_dev
->msix_enabled
= 1;
137 /* Set the vector used for configuration */
138 v
= vp_dev
->msix_used_vectors
;
139 snprintf(vp_dev
->msix_names
[v
], sizeof *vp_dev
->msix_names
,
141 err
= request_irq(pci_irq_vector(vp_dev
->pci_dev
, v
),
142 vp_config_changed
, 0, vp_dev
->msix_names
[v
],
146 ++vp_dev
->msix_used_vectors
;
148 v
= vp_dev
->config_vector(vp_dev
, v
);
149 /* Verify we had enough resources to assign the vector */
150 if (v
== VIRTIO_MSI_NO_VECTOR
) {
155 if (!per_vq_vectors
) {
156 /* Shared vector for all VQs */
157 v
= vp_dev
->msix_used_vectors
;
158 snprintf(vp_dev
->msix_names
[v
], sizeof *vp_dev
->msix_names
,
159 "%s-virtqueues", name
);
160 err
= request_irq(pci_irq_vector(vp_dev
->pci_dev
, v
),
161 vp_vring_interrupt
, 0, vp_dev
->msix_names
[v
],
165 ++vp_dev
->msix_used_vectors
;
172 static struct virtqueue
*vp_setup_vq(struct virtio_device
*vdev
, unsigned index
,
173 void (*callback
)(struct virtqueue
*vq
),
178 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
179 struct virtio_pci_vq_info
*info
= kmalloc(sizeof *info
, GFP_KERNEL
);
180 struct virtqueue
*vq
;
183 /* fill out our structure that represents an active queue */
185 return ERR_PTR(-ENOMEM
);
187 vq
= vp_dev
->setup_vq(vp_dev
, info
, index
, callback
, name
, ctx
,
194 spin_lock_irqsave(&vp_dev
->lock
, flags
);
195 list_add(&info
->node
, &vp_dev
->virtqueues
);
196 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
198 INIT_LIST_HEAD(&info
->node
);
201 vp_dev
->vqs
[index
] = info
;
209 static void vp_del_vq(struct virtqueue
*vq
)
211 struct virtio_pci_device
*vp_dev
= to_vp_device(vq
->vdev
);
212 struct virtio_pci_vq_info
*info
= vp_dev
->vqs
[vq
->index
];
215 spin_lock_irqsave(&vp_dev
->lock
, flags
);
216 list_del(&info
->node
);
217 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
219 vp_dev
->del_vq(info
);
223 /* the config->del_vqs() implementation */
224 void vp_del_vqs(struct virtio_device
*vdev
)
226 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
227 struct virtqueue
*vq
, *n
;
230 list_for_each_entry_safe(vq
, n
, &vdev
->vqs
, list
) {
231 if (vp_dev
->per_vq_vectors
) {
232 int v
= vp_dev
->vqs
[vq
->index
]->msix_vector
;
234 if (v
!= VIRTIO_MSI_NO_VECTOR
) {
235 int irq
= pci_irq_vector(vp_dev
->pci_dev
, v
);
237 irq_set_affinity_hint(irq
, NULL
);
243 vp_dev
->per_vq_vectors
= false;
245 if (vp_dev
->intx_enabled
) {
246 free_irq(vp_dev
->pci_dev
->irq
, vp_dev
);
247 vp_dev
->intx_enabled
= 0;
250 for (i
= 0; i
< vp_dev
->msix_used_vectors
; ++i
)
251 free_irq(pci_irq_vector(vp_dev
->pci_dev
, i
), vp_dev
);
253 for (i
= 0; i
< vp_dev
->msix_vectors
; i
++)
254 if (vp_dev
->msix_affinity_masks
[i
])
255 free_cpumask_var(vp_dev
->msix_affinity_masks
[i
]);
257 if (vp_dev
->msix_enabled
) {
258 /* Disable the vector used for configuration */
259 vp_dev
->config_vector(vp_dev
, VIRTIO_MSI_NO_VECTOR
);
261 pci_free_irq_vectors(vp_dev
->pci_dev
);
262 vp_dev
->msix_enabled
= 0;
265 vp_dev
->msix_vectors
= 0;
266 vp_dev
->msix_used_vectors
= 0;
267 kfree(vp_dev
->msix_names
);
268 vp_dev
->msix_names
= NULL
;
269 kfree(vp_dev
->msix_affinity_masks
);
270 vp_dev
->msix_affinity_masks
= NULL
;
275 static int vp_find_vqs_msix(struct virtio_device
*vdev
, unsigned nvqs
,
276 struct virtqueue
*vqs
[], vq_callback_t
*callbacks
[],
277 const char * const names
[], bool per_vq_vectors
,
279 struct irq_affinity
*desc
)
281 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
283 int i
, err
, nvectors
, allocated_vectors
;
285 vp_dev
->vqs
= kcalloc(nvqs
, sizeof(*vp_dev
->vqs
), GFP_KERNEL
);
289 if (per_vq_vectors
) {
290 /* Best option: one for change interrupt, one per vq. */
292 for (i
= 0; i
< nvqs
; ++i
)
296 /* Second best: one for change, shared for all vqs. */
300 err
= vp_request_msix_vectors(vdev
, nvectors
, per_vq_vectors
,
301 per_vq_vectors
? desc
: NULL
);
305 vp_dev
->per_vq_vectors
= per_vq_vectors
;
306 allocated_vectors
= vp_dev
->msix_used_vectors
;
307 for (i
= 0; i
< nvqs
; ++i
) {
314 msix_vec
= VIRTIO_MSI_NO_VECTOR
;
315 else if (vp_dev
->per_vq_vectors
)
316 msix_vec
= allocated_vectors
++;
318 msix_vec
= VP_MSIX_VQ_VECTOR
;
319 vqs
[i
] = vp_setup_vq(vdev
, i
, callbacks
[i
], names
[i
],
320 ctx
? ctx
[i
] : false,
322 if (IS_ERR(vqs
[i
])) {
323 err
= PTR_ERR(vqs
[i
]);
327 if (!vp_dev
->per_vq_vectors
|| msix_vec
== VIRTIO_MSI_NO_VECTOR
)
330 /* allocate per-vq irq if available and necessary */
331 snprintf(vp_dev
->msix_names
[msix_vec
],
332 sizeof *vp_dev
->msix_names
,
334 dev_name(&vp_dev
->vdev
.dev
), names
[i
]);
335 err
= request_irq(pci_irq_vector(vp_dev
->pci_dev
, msix_vec
),
337 vp_dev
->msix_names
[msix_vec
],
349 static int vp_find_vqs_intx(struct virtio_device
*vdev
, unsigned nvqs
,
350 struct virtqueue
*vqs
[], vq_callback_t
*callbacks
[],
351 const char * const names
[], const bool *ctx
)
353 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
356 vp_dev
->vqs
= kcalloc(nvqs
, sizeof(*vp_dev
->vqs
), GFP_KERNEL
);
360 err
= request_irq(vp_dev
->pci_dev
->irq
, vp_interrupt
, IRQF_SHARED
,
361 dev_name(&vdev
->dev
), vp_dev
);
365 vp_dev
->intx_enabled
= 1;
366 vp_dev
->per_vq_vectors
= false;
367 for (i
= 0; i
< nvqs
; ++i
) {
372 vqs
[i
] = vp_setup_vq(vdev
, i
, callbacks
[i
], names
[i
],
373 ctx
? ctx
[i
] : false,
374 VIRTIO_MSI_NO_VECTOR
);
375 if (IS_ERR(vqs
[i
])) {
376 err
= PTR_ERR(vqs
[i
]);
387 /* the config->find_vqs() implementation */
388 int vp_find_vqs(struct virtio_device
*vdev
, unsigned nvqs
,
389 struct virtqueue
*vqs
[], vq_callback_t
*callbacks
[],
390 const char * const names
[], const bool *ctx
,
391 struct irq_affinity
*desc
)
395 /* Try MSI-X with one vector per queue. */
396 err
= vp_find_vqs_msix(vdev
, nvqs
, vqs
, callbacks
, names
, true, ctx
, desc
);
399 /* Fallback: MSI-X with one vector for config, one shared for queues. */
400 err
= vp_find_vqs_msix(vdev
, nvqs
, vqs
, callbacks
, names
, false, ctx
, desc
);
403 /* Finally fall back to regular interrupts. */
404 return vp_find_vqs_intx(vdev
, nvqs
, vqs
, callbacks
, names
, ctx
);
407 const char *vp_bus_name(struct virtio_device
*vdev
)
409 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
411 return pci_name(vp_dev
->pci_dev
);
414 /* Setup the affinity for a virtqueue:
415 * - force the affinity for per vq vector
416 * - OR over all affinities for shared MSI
417 * - ignore the affinity request if we're using INTX
419 int vp_set_vq_affinity(struct virtqueue
*vq
, int cpu
)
421 struct virtio_device
*vdev
= vq
->vdev
;
422 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
423 struct virtio_pci_vq_info
*info
= vp_dev
->vqs
[vq
->index
];
424 struct cpumask
*mask
;
430 if (vp_dev
->msix_enabled
) {
431 mask
= vp_dev
->msix_affinity_masks
[info
->msix_vector
];
432 irq
= pci_irq_vector(vp_dev
->pci_dev
, info
->msix_vector
);
434 irq_set_affinity_hint(irq
, NULL
);
437 cpumask_set_cpu(cpu
, mask
);
438 irq_set_affinity_hint(irq
, mask
);
444 const struct cpumask
*vp_get_vq_affinity(struct virtio_device
*vdev
, int index
)
446 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
448 if (!vp_dev
->per_vq_vectors
||
449 vp_dev
->vqs
[index
]->msix_vector
== VIRTIO_MSI_NO_VECTOR
)
452 return pci_irq_get_affinity(vp_dev
->pci_dev
,
453 vp_dev
->vqs
[index
]->msix_vector
);
456 #ifdef CONFIG_PM_SLEEP
457 static int virtio_pci_freeze(struct device
*dev
)
459 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
460 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
463 ret
= virtio_device_freeze(&vp_dev
->vdev
);
466 pci_disable_device(pci_dev
);
470 static int virtio_pci_restore(struct device
*dev
)
472 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
473 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
476 ret
= pci_enable_device(pci_dev
);
480 pci_set_master(pci_dev
);
481 return virtio_device_restore(&vp_dev
->vdev
);
484 static const struct dev_pm_ops virtio_pci_pm_ops
= {
485 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze
, virtio_pci_restore
)
490 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
491 static const struct pci_device_id virtio_pci_id_table
[] = {
492 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET
, PCI_ANY_ID
) },
496 MODULE_DEVICE_TABLE(pci
, virtio_pci_id_table
);
498 static void virtio_pci_release_dev(struct device
*_d
)
500 struct virtio_device
*vdev
= dev_to_virtio(_d
);
501 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
503 /* As struct device is a kobject, it's not safe to
504 * free the memory (including the reference counter itself)
505 * until it's release callback. */
509 static int virtio_pci_probe(struct pci_dev
*pci_dev
,
510 const struct pci_device_id
*id
)
512 struct virtio_pci_device
*vp_dev
;
515 /* allocate our structure and fill it out */
516 vp_dev
= kzalloc(sizeof(struct virtio_pci_device
), GFP_KERNEL
);
520 pci_set_drvdata(pci_dev
, vp_dev
);
521 vp_dev
->vdev
.dev
.parent
= &pci_dev
->dev
;
522 vp_dev
->vdev
.dev
.release
= virtio_pci_release_dev
;
523 vp_dev
->pci_dev
= pci_dev
;
524 INIT_LIST_HEAD(&vp_dev
->virtqueues
);
525 spin_lock_init(&vp_dev
->lock
);
527 /* enable the device */
528 rc
= pci_enable_device(pci_dev
);
530 goto err_enable_device
;
533 rc
= virtio_pci_legacy_probe(vp_dev
);
534 /* Also try modern mode if we can't map BAR0 (no IO space). */
535 if (rc
== -ENODEV
|| rc
== -ENOMEM
)
536 rc
= virtio_pci_modern_probe(vp_dev
);
540 rc
= virtio_pci_modern_probe(vp_dev
);
542 rc
= virtio_pci_legacy_probe(vp_dev
);
547 pci_set_master(pci_dev
);
549 rc
= register_virtio_device(&vp_dev
->vdev
);
557 virtio_pci_legacy_remove(vp_dev
);
559 virtio_pci_modern_remove(vp_dev
);
561 pci_disable_device(pci_dev
);
567 static void virtio_pci_remove(struct pci_dev
*pci_dev
)
569 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
570 struct device
*dev
= get_device(&vp_dev
->vdev
.dev
);
572 unregister_virtio_device(&vp_dev
->vdev
);
575 virtio_pci_legacy_remove(vp_dev
);
577 virtio_pci_modern_remove(vp_dev
);
579 pci_disable_device(pci_dev
);
583 static struct pci_driver virtio_pci_driver
= {
584 .name
= "virtio-pci",
585 .id_table
= virtio_pci_id_table
,
586 .probe
= virtio_pci_probe
,
587 .remove
= virtio_pci_remove
,
588 #ifdef CONFIG_PM_SLEEP
589 .driver
.pm
= &virtio_pci_pm_ops
,
593 module_pci_driver(virtio_pci_driver
);
595 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
596 MODULE_DESCRIPTION("virtio-pci");
597 MODULE_LICENSE("GPL");