1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Virtio PCI driver - modern (virtio 1.0) device support
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 <linux/delay.h>
18 #define VIRTIO_PCI_NO_LEGACY
19 #define VIRTIO_RING_NO_LEGACY
20 #include "virtio_pci_common.h"
22 static u64 vp_get_features(struct virtio_device *vdev)
24 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
26 return vp_modern_get_features(&vp_dev->mdev);
29 static void vp_transport_features(struct virtio_device *vdev, u64 features)
31 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
32 struct pci_dev *pci_dev = vp_dev->pci_dev;
34 if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) &&
35 pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV))
36 __virtio_set_bit(vdev, VIRTIO_F_SR_IOV);
39 /* virtio config->finalize_features() implementation */
40 static int vp_finalize_features(struct virtio_device *vdev)
42 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
43 u64 features = vdev->features;
45 /* Give virtio_ring a chance to accept features. */
46 vring_transport_features(vdev);
48 /* Give virtio_pci a chance to accept features. */
49 vp_transport_features(vdev, features);
51 if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
52 dev_err(&vdev->dev, "virtio: device uses modern interface "
53 "but does not have VIRTIO_F_VERSION_1\n");
57 vp_modern_set_features(&vp_dev->mdev, vdev->features);
62 /* virtio config->get() implementation */
63 static void vp_get(struct virtio_device *vdev, unsigned offset,
64 void *buf, unsigned len)
66 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
67 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
68 void __iomem *device = mdev->device;
73 BUG_ON(offset + len > mdev->device_len);
77 b = ioread8(device + offset);
78 memcpy(buf, &b, sizeof b);
81 w = cpu_to_le16(ioread16(device + offset));
82 memcpy(buf, &w, sizeof w);
85 l = cpu_to_le32(ioread32(device + offset));
86 memcpy(buf, &l, sizeof l);
89 l = cpu_to_le32(ioread32(device + offset));
90 memcpy(buf, &l, sizeof l);
91 l = cpu_to_le32(ioread32(device + offset + sizeof l));
92 memcpy(buf + sizeof l, &l, sizeof l);
99 /* the config->set() implementation. it's symmetric to the config->get()
101 static void vp_set(struct virtio_device *vdev, unsigned offset,
102 const void *buf, unsigned len)
104 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
105 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
106 void __iomem *device = mdev->device;
111 BUG_ON(offset + len > mdev->device_len);
115 memcpy(&b, buf, sizeof b);
116 iowrite8(b, device + offset);
119 memcpy(&w, buf, sizeof w);
120 iowrite16(le16_to_cpu(w), device + offset);
123 memcpy(&l, buf, sizeof l);
124 iowrite32(le32_to_cpu(l), device + offset);
127 memcpy(&l, buf, sizeof l);
128 iowrite32(le32_to_cpu(l), device + offset);
129 memcpy(&l, buf + sizeof l, sizeof l);
130 iowrite32(le32_to_cpu(l), device + offset + sizeof l);
137 static u32 vp_generation(struct virtio_device *vdev)
139 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
141 return vp_modern_generation(&vp_dev->mdev);
144 /* config->{get,set}_status() implementations */
145 static u8 vp_get_status(struct virtio_device *vdev)
147 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
149 return vp_modern_get_status(&vp_dev->mdev);
152 static void vp_set_status(struct virtio_device *vdev, u8 status)
154 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
156 /* We should never be setting status to 0. */
158 vp_modern_set_status(&vp_dev->mdev, status);
161 static void vp_reset(struct virtio_device *vdev)
163 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
164 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
166 /* 0 status means a reset. */
167 vp_modern_set_status(mdev, 0);
168 /* After writing 0 to device_status, the driver MUST wait for a read of
169 * device_status to return 0 before reinitializing the device.
170 * This will flush out the status write, and flush in device writes,
171 * including MSI-X interrupts, if any.
173 while (vp_modern_get_status(mdev))
175 /* Flush pending VQ/configuration callbacks. */
176 vp_synchronize_vectors(vdev);
179 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
181 return vp_modern_config_vector(&vp_dev->mdev, vector);
184 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
185 struct virtio_pci_vq_info *info,
187 void (*callback)(struct virtqueue *vq),
193 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
194 struct virtqueue *vq;
198 if (index >= vp_modern_get_num_queues(mdev))
199 return ERR_PTR(-ENOENT);
201 /* Check if queue is either not available or already active. */
202 num = vp_modern_get_queue_size(mdev, index);
203 if (!num || vp_modern_get_queue_enable(mdev, index))
204 return ERR_PTR(-ENOENT);
206 if (num & (num - 1)) {
207 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
208 return ERR_PTR(-EINVAL);
211 /* get offset of notification word for this vq */
212 off = vp_modern_get_queue_notify_off(mdev, index);
214 info->msix_vector = msix_vec;
216 /* create the vring */
217 vq = vring_create_virtqueue(index, num,
218 SMP_CACHE_BYTES, &vp_dev->vdev,
220 vp_notify, callback, name);
222 return ERR_PTR(-ENOMEM);
224 /* activate the queue */
225 vp_modern_set_queue_size(mdev, index, virtqueue_get_vring_size(vq));
226 vp_modern_queue_address(mdev, index, virtqueue_get_desc_addr(vq),
227 virtqueue_get_avail_addr(vq),
228 virtqueue_get_used_addr(vq));
230 if (mdev->notify_base) {
231 /* offset should not wrap */
232 if ((u64)off * mdev->notify_offset_multiplier + 2
233 > mdev->notify_len) {
234 dev_warn(&mdev->pci_dev->dev,
235 "bad notification offset %u (x %u) "
236 "for queue %u > %zd",
237 off, mdev->notify_offset_multiplier,
238 index, mdev->notify_len);
242 vq->priv = (void __force *)mdev->notify_base +
243 off * mdev->notify_offset_multiplier;
245 vq->priv = (void __force *)vp_modern_map_capability(mdev,
246 mdev->notify_map_cap, 2, 2,
247 off * mdev->notify_offset_multiplier, 2,
256 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
257 msix_vec = vp_modern_queue_vector(mdev, index, msix_vec);
258 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
260 goto err_assign_vector;
267 if (!mdev->notify_base)
268 pci_iounmap(mdev->pci_dev, (void __iomem __force *)vq->priv);
270 vring_del_virtqueue(vq);
274 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
275 struct virtqueue *vqs[],
276 vq_callback_t *callbacks[],
277 const char * const names[], const bool *ctx,
278 struct irq_affinity *desc)
280 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
281 struct virtqueue *vq;
282 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
287 /* Select and activate all queues. Has to be done last: once we do
288 * this, there's no way to go back except reset.
290 list_for_each_entry(vq, &vdev->vqs, list)
291 vp_modern_set_queue_enable(&vp_dev->mdev, vq->index, true);
296 static void del_vq(struct virtio_pci_vq_info *info)
298 struct virtqueue *vq = info->vq;
299 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
300 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
302 if (vp_dev->msix_enabled)
303 vp_modern_queue_vector(mdev, vq->index,
304 VIRTIO_MSI_NO_VECTOR);
306 if (!mdev->notify_base)
307 pci_iounmap(mdev->pci_dev, (void __force __iomem *)vq->priv);
309 vring_del_virtqueue(vq);
312 static int virtio_pci_find_shm_cap(struct pci_dev *dev, u8 required_id,
313 u8 *bar, u64 *offset, u64 *len)
317 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR); pos > 0;
318 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
319 u8 type, cap_len, id;
321 u64 res_offset, res_length;
323 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
325 if (type != VIRTIO_PCI_CAP_SHARED_MEMORY_CFG)
328 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
330 if (cap_len != sizeof(struct virtio_pci_cap64)) {
331 dev_err(&dev->dev, "%s: shm cap with bad size offset:"
332 " %d size: %d\n", __func__, pos, cap_len);
336 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
338 if (id != required_id)
341 /* Type, and ID match, looks good */
342 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
345 /* Read the lower 32bit of length and offset */
346 pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
349 pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
353 /* and now the top half */
354 pci_read_config_dword(dev,
355 pos + offsetof(struct virtio_pci_cap64,
357 res_offset |= ((u64)tmp32) << 32;
358 pci_read_config_dword(dev,
359 pos + offsetof(struct virtio_pci_cap64,
361 res_length |= ((u64)tmp32) << 32;
363 *offset = res_offset;
371 static bool vp_get_shm_region(struct virtio_device *vdev,
372 struct virtio_shm_region *region, u8 id)
374 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
375 struct pci_dev *pci_dev = vp_dev->pci_dev;
378 phys_addr_t phys_addr;
381 if (!virtio_pci_find_shm_cap(pci_dev, id, &bar, &offset, &len))
384 phys_addr = pci_resource_start(pci_dev, bar);
385 bar_len = pci_resource_len(pci_dev, bar);
387 if ((offset + len) < offset) {
388 dev_err(&pci_dev->dev, "%s: cap offset+len overflow detected\n",
393 if (offset + len > bar_len) {
394 dev_err(&pci_dev->dev, "%s: bar shorter than cap offset+len\n",
400 region->addr = (u64) phys_addr + offset;
405 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
408 .generation = vp_generation,
409 .get_status = vp_get_status,
410 .set_status = vp_set_status,
412 .find_vqs = vp_modern_find_vqs,
413 .del_vqs = vp_del_vqs,
414 .get_features = vp_get_features,
415 .finalize_features = vp_finalize_features,
416 .bus_name = vp_bus_name,
417 .set_vq_affinity = vp_set_vq_affinity,
418 .get_vq_affinity = vp_get_vq_affinity,
419 .get_shm_region = vp_get_shm_region,
422 static const struct virtio_config_ops virtio_pci_config_ops = {
425 .generation = vp_generation,
426 .get_status = vp_get_status,
427 .set_status = vp_set_status,
429 .find_vqs = vp_modern_find_vqs,
430 .del_vqs = vp_del_vqs,
431 .get_features = vp_get_features,
432 .finalize_features = vp_finalize_features,
433 .bus_name = vp_bus_name,
434 .set_vq_affinity = vp_set_vq_affinity,
435 .get_vq_affinity = vp_get_vq_affinity,
436 .get_shm_region = vp_get_shm_region,
439 /* the PCI probing function */
440 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
442 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
443 struct pci_dev *pci_dev = vp_dev->pci_dev;
446 mdev->pci_dev = pci_dev;
448 err = vp_modern_probe(mdev);
453 vp_dev->vdev.config = &virtio_pci_config_ops;
455 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
457 vp_dev->config_vector = vp_config_vector;
458 vp_dev->setup_vq = setup_vq;
459 vp_dev->del_vq = del_vq;
460 vp_dev->isr = mdev->isr;
461 vp_dev->vdev.id = mdev->id;
466 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
468 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
470 vp_modern_remove(mdev);