1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Virtio PCI driver - common functionality for all device versions
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.
8 * Copyright IBM Corp. 2007
9 * Copyright Red Hat, Inc. 2014
12 * Anthony Liguori <aliguori@us.ibm.com>
13 * Rusty Russell <rusty@rustcorp.com.au>
14 * Michael S. Tsirkin <mst@redhat.com>
17 #include "virtio_pci_common.h"
19 static bool force_legacy
= false;
21 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
22 module_param(force_legacy
, bool, 0444);
23 MODULE_PARM_DESC(force_legacy
,
24 "Force legacy mode for transitional virtio 1 devices");
27 /* wait for pending irq handlers */
28 void vp_synchronize_vectors(struct virtio_device
*vdev
)
30 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
33 if (vp_dev
->intx_enabled
)
34 synchronize_irq(vp_dev
->pci_dev
->irq
);
36 for (i
= 0; i
< vp_dev
->msix_vectors
; ++i
)
37 synchronize_irq(pci_irq_vector(vp_dev
->pci_dev
, i
));
40 /* the notify function used when creating a virt queue */
41 bool vp_notify(struct virtqueue
*vq
)
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
);
49 /* Handle a configuration change: Tell driver if it wants to know. */
50 static irqreturn_t
vp_config_changed(int irq
, void *opaque
)
52 struct virtio_pci_device
*vp_dev
= opaque
;
54 virtio_config_changed(&vp_dev
->vdev
);
58 /* Notify all virtqueues on an interrupt. */
59 static irqreturn_t
vp_vring_interrupt(int irq
, void *opaque
)
61 struct virtio_pci_device
*vp_dev
= opaque
;
62 struct virtio_pci_vq_info
*info
;
63 irqreturn_t ret
= IRQ_NONE
;
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
)
71 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
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
82 static irqreturn_t
vp_interrupt(int irq
, void *opaque
)
84 struct virtio_pci_device
*vp_dev
= opaque
;
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
);
91 /* It's definitely not us if the ISR was not high */
95 /* Configuration change? Tell driver if it wants to know. */
96 if (isr
& VIRTIO_PCI_ISR_CONFIG
)
97 vp_config_changed(irq
, opaque
);
99 return vp_vring_interrupt(irq
, opaque
);
102 static int vp_request_msix_vectors(struct virtio_device
*vdev
, int nvectors
,
103 bool per_vq_vectors
, struct irq_affinity
*desc
)
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
;
111 vp_dev
->msix_vectors
= nvectors
;
113 vp_dev
->msix_names
= kmalloc_array(nvectors
,
114 sizeof(*vp_dev
->msix_names
),
116 if (!vp_dev
->msix_names
)
118 vp_dev
->msix_affinity_masks
119 = kcalloc(nvectors
, sizeof(*vp_dev
->msix_affinity_masks
),
121 if (!vp_dev
->msix_affinity_masks
)
123 for (i
= 0; i
< nvectors
; ++i
)
124 if (!alloc_cpumask_var(&vp_dev
->msix_affinity_masks
[i
],
129 flags
|= PCI_IRQ_AFFINITY
;
130 desc
->pre_vectors
++; /* virtio config vector */
133 err
= pci_alloc_irq_vectors_affinity(vp_dev
->pci_dev
, nvectors
,
134 nvectors
, flags
, desc
);
137 vp_dev
->msix_enabled
= 1;
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
,
143 err
= request_irq(pci_irq_vector(vp_dev
->pci_dev
, v
),
144 vp_config_changed
, 0, vp_dev
->msix_names
[v
],
148 ++vp_dev
->msix_used_vectors
;
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
) {
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
],
167 ++vp_dev
->msix_used_vectors
;
174 static struct virtqueue
*vp_setup_vq(struct virtio_device
*vdev
, unsigned index
,
175 void (*callback
)(struct virtqueue
*vq
),
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
;
185 /* fill out our structure that represents an active queue */
187 return ERR_PTR(-ENOMEM
);
189 vq
= vp_dev
->setup_vq(vp_dev
, info
, index
, callback
, name
, ctx
,
196 spin_lock_irqsave(&vp_dev
->lock
, flags
);
197 list_add(&info
->node
, &vp_dev
->virtqueues
);
198 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
200 INIT_LIST_HEAD(&info
->node
);
203 vp_dev
->vqs
[index
] = info
;
211 static void vp_del_vq(struct virtqueue
*vq
)
213 struct virtio_pci_device
*vp_dev
= to_vp_device(vq
->vdev
);
214 struct virtio_pci_vq_info
*info
= vp_dev
->vqs
[vq
->index
];
217 spin_lock_irqsave(&vp_dev
->lock
, flags
);
218 list_del(&info
->node
);
219 spin_unlock_irqrestore(&vp_dev
->lock
, flags
);
221 vp_dev
->del_vq(info
);
225 /* the config->del_vqs() implementation */
226 void vp_del_vqs(struct virtio_device
*vdev
)
228 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
229 struct virtqueue
*vq
, *n
;
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
;
236 if (v
!= VIRTIO_MSI_NO_VECTOR
) {
237 int irq
= pci_irq_vector(vp_dev
->pci_dev
, v
);
239 irq_set_affinity_hint(irq
, NULL
);
245 vp_dev
->per_vq_vectors
= false;
247 if (vp_dev
->intx_enabled
) {
248 free_irq(vp_dev
->pci_dev
->irq
, vp_dev
);
249 vp_dev
->intx_enabled
= 0;
252 for (i
= 0; i
< vp_dev
->msix_used_vectors
; ++i
)
253 free_irq(pci_irq_vector(vp_dev
->pci_dev
, i
), vp_dev
);
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
]);
261 if (vp_dev
->msix_enabled
) {
262 /* Disable the vector used for configuration */
263 vp_dev
->config_vector(vp_dev
, VIRTIO_MSI_NO_VECTOR
);
265 pci_free_irq_vectors(vp_dev
->pci_dev
);
266 vp_dev
->msix_enabled
= 0;
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
;
279 static 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
,
283 struct irq_affinity
*desc
)
285 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
287 int i
, err
, nvectors
, allocated_vectors
, queue_idx
= 0;
289 vp_dev
->vqs
= kcalloc(nvqs
, sizeof(*vp_dev
->vqs
), GFP_KERNEL
);
293 if (per_vq_vectors
) {
294 /* Best option: one for change interrupt, one per vq. */
296 for (i
= 0; i
< nvqs
; ++i
)
297 if (names
[i
] && callbacks
[i
])
300 /* Second best: one for change, shared for all vqs. */
304 err
= vp_request_msix_vectors(vdev
, nvectors
, per_vq_vectors
,
305 per_vq_vectors
? desc
: NULL
);
309 vp_dev
->per_vq_vectors
= per_vq_vectors
;
310 allocated_vectors
= vp_dev
->msix_used_vectors
;
311 for (i
= 0; i
< nvqs
; ++i
) {
318 msix_vec
= VIRTIO_MSI_NO_VECTOR
;
319 else if (vp_dev
->per_vq_vectors
)
320 msix_vec
= allocated_vectors
++;
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,
326 if (IS_ERR(vqs
[i
])) {
327 err
= PTR_ERR(vqs
[i
]);
331 if (!vp_dev
->per_vq_vectors
|| msix_vec
== VIRTIO_MSI_NO_VECTOR
)
334 /* allocate per-vq irq if available and necessary */
335 snprintf(vp_dev
->msix_names
[msix_vec
],
336 sizeof *vp_dev
->msix_names
,
338 dev_name(&vp_dev
->vdev
.dev
), names
[i
]);
339 err
= request_irq(pci_irq_vector(vp_dev
->pci_dev
, msix_vec
),
341 vp_dev
->msix_names
[msix_vec
],
353 static 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
)
357 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
358 int i
, err
, queue_idx
= 0;
360 vp_dev
->vqs
= kcalloc(nvqs
, sizeof(*vp_dev
->vqs
), GFP_KERNEL
);
364 err
= request_irq(vp_dev
->pci_dev
->irq
, vp_interrupt
, IRQF_SHARED
,
365 dev_name(&vdev
->dev
), vp_dev
);
369 vp_dev
->intx_enabled
= 1;
370 vp_dev
->per_vq_vectors
= false;
371 for (i
= 0; i
< nvqs
; ++i
) {
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
]);
391 /* the config->find_vqs() implementation */
392 int 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
)
399 /* Try MSI-X with one vector per queue. */
400 err
= vp_find_vqs_msix(vdev
, nvqs
, vqs
, callbacks
, names
, true, ctx
, desc
);
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
);
407 /* Finally fall back to regular interrupts. */
408 return vp_find_vqs_intx(vdev
, nvqs
, vqs
, callbacks
, names
, ctx
);
411 const char *vp_bus_name(struct virtio_device
*vdev
)
413 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
415 return pci_name(vp_dev
->pci_dev
);
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
423 int vp_set_vq_affinity(struct virtqueue
*vq
, const struct cpumask
*cpu_mask
)
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
;
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
);
438 irq_set_affinity_hint(irq
, NULL
);
440 cpumask_copy(mask
, cpu_mask
);
441 irq_set_affinity_hint(irq
, mask
);
447 const struct cpumask
*vp_get_vq_affinity(struct virtio_device
*vdev
, int index
)
449 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
451 if (!vp_dev
->per_vq_vectors
||
452 vp_dev
->vqs
[index
]->msix_vector
== VIRTIO_MSI_NO_VECTOR
)
455 return pci_irq_get_affinity(vp_dev
->pci_dev
,
456 vp_dev
->vqs
[index
]->msix_vector
);
459 #ifdef CONFIG_PM_SLEEP
460 static int virtio_pci_freeze(struct device
*dev
)
462 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
463 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
466 ret
= virtio_device_freeze(&vp_dev
->vdev
);
469 pci_disable_device(pci_dev
);
473 static int virtio_pci_restore(struct device
*dev
)
475 struct pci_dev
*pci_dev
= to_pci_dev(dev
);
476 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
479 ret
= pci_enable_device(pci_dev
);
483 pci_set_master(pci_dev
);
484 return virtio_device_restore(&vp_dev
->vdev
);
487 static const struct dev_pm_ops virtio_pci_pm_ops
= {
488 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze
, virtio_pci_restore
)
493 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
494 static const struct pci_device_id virtio_pci_id_table
[] = {
495 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET
, PCI_ANY_ID
) },
499 MODULE_DEVICE_TABLE(pci
, virtio_pci_id_table
);
501 static void virtio_pci_release_dev(struct device
*_d
)
503 struct virtio_device
*vdev
= dev_to_virtio(_d
);
504 struct virtio_pci_device
*vp_dev
= to_vp_device(vdev
);
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. */
512 static int virtio_pci_probe(struct pci_dev
*pci_dev
,
513 const struct pci_device_id
*id
)
515 struct virtio_pci_device
*vp_dev
, *reg_dev
= NULL
;
518 /* allocate our structure and fill it out */
519 vp_dev
= kzalloc(sizeof(struct virtio_pci_device
), GFP_KERNEL
);
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
);
530 /* enable the device */
531 rc
= pci_enable_device(pci_dev
);
533 goto err_enable_device
;
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
);
543 rc
= virtio_pci_modern_probe(vp_dev
);
545 rc
= virtio_pci_legacy_probe(vp_dev
);
550 pci_set_master(pci_dev
);
552 rc
= register_virtio_device(&vp_dev
->vdev
);
561 virtio_pci_legacy_remove(vp_dev
);
563 virtio_pci_modern_remove(vp_dev
);
565 pci_disable_device(pci_dev
);
568 put_device(&vp_dev
->vdev
.dev
);
574 static void virtio_pci_remove(struct pci_dev
*pci_dev
)
576 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
577 struct device
*dev
= get_device(&vp_dev
->vdev
.dev
);
579 pci_disable_sriov(pci_dev
);
581 unregister_virtio_device(&vp_dev
->vdev
);
584 virtio_pci_legacy_remove(vp_dev
);
586 virtio_pci_modern_remove(vp_dev
);
588 pci_disable_device(pci_dev
);
592 static int virtio_pci_sriov_configure(struct pci_dev
*pci_dev
, int num_vfs
)
594 struct virtio_pci_device
*vp_dev
= pci_get_drvdata(pci_dev
);
595 struct virtio_device
*vdev
= &vp_dev
->vdev
;
598 if (!(vdev
->config
->get_status(vdev
) & VIRTIO_CONFIG_S_DRIVER_OK
))
601 if (!__virtio_test_bit(vdev
, VIRTIO_F_SR_IOV
))
604 if (pci_vfs_assigned(pci_dev
))
608 pci_disable_sriov(pci_dev
);
612 ret
= pci_enable_sriov(pci_dev
, num_vfs
);
619 static 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
,
627 .sriov_configure
= virtio_pci_sriov_configure
,
630 module_pci_driver(virtio_pci_driver
);
632 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
633 MODULE_DESCRIPTION("virtio-pci");
634 MODULE_LICENSE("GPL");