| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* Copyright (c) 2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved |
| 4 |
*/ |
| 5 |
|
| 6 |
#include <linux/device.h> |
| 7 |
#include <linux/module.h> |
| 8 |
#include <linux/mutex.h> |
| 9 |
#include <linux/pci.h> |
| 10 |
#include <linux/pm_runtime.h> |
| 11 |
#include <linux/types.h> |
| 12 |
#include <linux/uaccess.h> |
| 13 |
#include <linux/vfio.h> |
| 14 |
#include <linux/vfio_pci_core.h> |
| 15 |
#include <linux/virtio_pci.h> |
| 16 |
#include <linux/virtio_net.h> |
| 17 |
#include <linux/virtio_pci_admin.h> |
| 18 |
|
| 19 |
#include "common.h" |
| 20 |
|
| 21 |
static int |
| 22 |
virtiovf_issue_legacy_rw_cmd(struct virtiovf_pci_core_device *virtvdev, |
| 23 |
loff_t pos, char __user *buf, |
| 24 |
size_t count, bool read) |
| 25 |
{ |
| 26 |
bool msix_enabled = |
| 27 |
(virtvdev->core_device.irq_type == VFIO_PCI_MSIX_IRQ_INDEX); |
| 28 |
struct pci_dev *pdev = virtvdev->core_device.pdev; |
| 29 |
u8 *bar0_buf = virtvdev->bar0_virtual_buf; |
| 30 |
bool common; |
| 31 |
u8 offset; |
| 32 |
int ret; |
| 33 |
|
| 34 |
common = pos < VIRTIO_PCI_CONFIG_OFF(msix_enabled); |
| 35 |
/* offset within the relevant configuration area */ |
| 36 |
offset = common ? pos : pos - VIRTIO_PCI_CONFIG_OFF(msix_enabled); |
| 37 |
mutex_lock(&virtvdev->bar_mutex); |
| 38 |
if (read) { |
| 39 |
if (common) |
| 40 |
ret = virtio_pci_admin_legacy_common_io_read(pdev, offset, |
| 41 |
count, bar0_buf + pos); |
| 42 |
else |
| 43 |
ret = virtio_pci_admin_legacy_device_io_read(pdev, offset, |
| 44 |
count, bar0_buf + pos); |
| 45 |
if (ret) |
| 46 |
goto out; |
| 47 |
if (copy_to_user(buf, bar0_buf + pos, count)) |
| 48 |
ret = -EFAULT; |
| 49 |
} else { |
| 50 |
if (copy_from_user(bar0_buf + pos, buf, count)) { |
| 51 |
ret = -EFAULT; |
| 52 |
goto out; |
| 53 |
} |
| 54 |
|
| 55 |
if (common) |
| 56 |
ret = virtio_pci_admin_legacy_common_io_write(pdev, offset, |
| 57 |
count, bar0_buf + pos); |
| 58 |
else |
| 59 |
ret = virtio_pci_admin_legacy_device_io_write(pdev, offset, |
| 60 |
count, bar0_buf + pos); |
| 61 |
} |
| 62 |
out: |
| 63 |
mutex_unlock(&virtvdev->bar_mutex); |
| 64 |
return ret; |
| 65 |
} |
| 66 |
|
| 67 |
static int |
| 68 |
virtiovf_pci_bar0_rw(struct virtiovf_pci_core_device *virtvdev, |
| 69 |
loff_t pos, char __user *buf, |
| 70 |
size_t count, bool read) |
| 71 |
{ |
| 72 |
struct vfio_pci_core_device *core_device = &virtvdev->core_device; |
| 73 |
struct pci_dev *pdev = core_device->pdev; |
| 74 |
u16 queue_notify; |
| 75 |
int ret; |
| 76 |
|
| 77 |
if (!(le16_to_cpu(virtvdev->pci_cmd) & PCI_COMMAND_IO)) |
| 78 |
return -EIO; |
| 79 |
|
| 80 |
if (pos + count > virtvdev->bar0_virtual_buf_size) |
| 81 |
return -EINVAL; |
| 82 |
|
| 83 |
ret = pm_runtime_resume_and_get(&pdev->dev); |
| 84 |
if (ret) { |
| 85 |
pci_info_ratelimited(pdev, "runtime resume failed %d\n", ret); |
| 86 |
return -EIO; |
| 87 |
} |
| 88 |
|
| 89 |
switch (pos) { |
| 90 |
case VIRTIO_PCI_QUEUE_NOTIFY: |
| 91 |
if (count != sizeof(queue_notify)) { |
| 92 |
ret = -EINVAL; |
| 93 |
goto end; |
| 94 |
} |
| 95 |
if (read) { |
| 96 |
ret = vfio_pci_core_ioread16(core_device, true, &queue_notify, |
| 97 |
virtvdev->notify_addr); |
| 98 |
if (ret) |
| 99 |
goto end; |
| 100 |
if (copy_to_user(buf, &queue_notify, |
| 101 |
sizeof(queue_notify))) { |
| 102 |
ret = -EFAULT; |
| 103 |
goto end; |
| 104 |
} |
| 105 |
} else { |
| 106 |
if (copy_from_user(&queue_notify, buf, count)) { |
| 107 |
ret = -EFAULT; |
| 108 |
goto end; |
| 109 |
} |
| 110 |
ret = vfio_pci_core_iowrite16(core_device, true, queue_notify, |
| 111 |
virtvdev->notify_addr); |
| 112 |
} |
| 113 |
break; |
| 114 |
default: |
| 115 |
ret = virtiovf_issue_legacy_rw_cmd(virtvdev, pos, buf, count, |
| 116 |
read); |
| 117 |
} |
| 118 |
|
| 119 |
end: |
| 120 |
pm_runtime_put(&pdev->dev); |
| 121 |
return ret ? ret : count; |
| 122 |
} |
| 123 |
|
| 124 |
static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev, |
| 125 |
char __user *buf, size_t count, |
| 126 |
loff_t *ppos) |
| 127 |
{ |
| 128 |
struct virtiovf_pci_core_device *virtvdev = container_of( |
| 129 |
core_vdev, struct virtiovf_pci_core_device, core_device.vdev); |
| 130 |
loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; |
| 131 |
size_t register_offset; |
| 132 |
loff_t copy_offset; |
| 133 |
size_t copy_count; |
| 134 |
__le32 val32; |
| 135 |
__le16 val16; |
| 136 |
u8 val8; |
| 137 |
int ret; |
| 138 |
|
| 139 |
ret = vfio_pci_core_read(core_vdev, buf, count, ppos); |
| 140 |
if (ret < 0) |
| 141 |
return ret; |
| 142 |
|
| 143 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_DEVICE_ID, |
| 144 |
sizeof(val16), ©_offset, |
| 145 |
©_count, ®ister_offset)) { |
| 146 |
val16 = cpu_to_le16(VIRTIO_TRANS_ID_NET); |
| 147 |
if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, copy_count)) |
| 148 |
return -EFAULT; |
| 149 |
} |
| 150 |
|
| 151 |
if ((le16_to_cpu(virtvdev->pci_cmd) & PCI_COMMAND_IO) && |
| 152 |
vfio_pci_core_range_intersect_range(pos, count, PCI_COMMAND, |
| 153 |
sizeof(val16), ©_offset, |
| 154 |
©_count, ®ister_offset)) { |
| 155 |
if (copy_from_user((void *)&val16 + register_offset, buf + copy_offset, |
| 156 |
copy_count)) |
| 157 |
return -EFAULT; |
| 158 |
val16 |= cpu_to_le16(PCI_COMMAND_IO); |
| 159 |
if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, |
| 160 |
copy_count)) |
| 161 |
return -EFAULT; |
| 162 |
} |
| 163 |
|
| 164 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_REVISION_ID, |
| 165 |
sizeof(val8), ©_offset, |
| 166 |
©_count, ®ister_offset)) { |
| 167 |
/* Transional needs to have revision 0 */ |
| 168 |
val8 = 0; |
| 169 |
if (copy_to_user(buf + copy_offset, &val8, copy_count)) |
| 170 |
return -EFAULT; |
| 171 |
} |
| 172 |
|
| 173 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_0, |
| 174 |
sizeof(val32), ©_offset, |
| 175 |
©_count, ®ister_offset)) { |
| 176 |
u32 bar_mask = ~(virtvdev->bar0_virtual_buf_size - 1); |
| 177 |
u32 pci_base_addr_0 = le32_to_cpu(virtvdev->pci_base_addr_0); |
| 178 |
|
| 179 |
val32 = cpu_to_le32((pci_base_addr_0 & bar_mask) | PCI_BASE_ADDRESS_SPACE_IO); |
| 180 |
if (copy_to_user(buf + copy_offset, (void *)&val32 + register_offset, copy_count)) |
| 181 |
return -EFAULT; |
| 182 |
} |
| 183 |
|
| 184 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_SUBSYSTEM_ID, |
| 185 |
sizeof(val16), ©_offset, |
| 186 |
©_count, ®ister_offset)) { |
| 187 |
/* |
| 188 |
* Transitional devices use the PCI subsystem device id as |
| 189 |
* virtio device id, same as legacy driver always did. |
| 190 |
*/ |
| 191 |
val16 = cpu_to_le16(VIRTIO_ID_NET); |
| 192 |
if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, |
| 193 |
copy_count)) |
| 194 |
return -EFAULT; |
| 195 |
} |
| 196 |
|
| 197 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_SUBSYSTEM_VENDOR_ID, |
| 198 |
sizeof(val16), ©_offset, |
| 199 |
©_count, ®ister_offset)) { |
| 200 |
val16 = cpu_to_le16(PCI_VENDOR_ID_REDHAT_QUMRANET); |
| 201 |
if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, |
| 202 |
copy_count)) |
| 203 |
return -EFAULT; |
| 204 |
} |
| 205 |
|
| 206 |
return count; |
| 207 |
} |
| 208 |
|
| 209 |
ssize_t virtiovf_pci_core_read(struct vfio_device *core_vdev, char __user *buf, |
| 210 |
size_t count, loff_t *ppos) |
| 211 |
{ |
| 212 |
struct virtiovf_pci_core_device *virtvdev = container_of( |
| 213 |
core_vdev, struct virtiovf_pci_core_device, core_device.vdev); |
| 214 |
unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); |
| 215 |
loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; |
| 216 |
|
| 217 |
if (!count) |
| 218 |
return 0; |
| 219 |
|
| 220 |
if (index == VFIO_PCI_CONFIG_REGION_INDEX) |
| 221 |
return virtiovf_pci_read_config(core_vdev, buf, count, ppos); |
| 222 |
|
| 223 |
if (index == VFIO_PCI_BAR0_REGION_INDEX) |
| 224 |
return virtiovf_pci_bar0_rw(virtvdev, pos, buf, count, true); |
| 225 |
|
| 226 |
return vfio_pci_core_read(core_vdev, buf, count, ppos); |
| 227 |
} |
| 228 |
|
| 229 |
static ssize_t virtiovf_pci_write_config(struct vfio_device *core_vdev, |
| 230 |
const char __user *buf, size_t count, |
| 231 |
loff_t *ppos) |
| 232 |
{ |
| 233 |
struct virtiovf_pci_core_device *virtvdev = container_of( |
| 234 |
core_vdev, struct virtiovf_pci_core_device, core_device.vdev); |
| 235 |
loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; |
| 236 |
size_t register_offset; |
| 237 |
loff_t copy_offset; |
| 238 |
size_t copy_count; |
| 239 |
|
| 240 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_COMMAND, |
| 241 |
sizeof(virtvdev->pci_cmd), |
| 242 |
©_offset, ©_count, |
| 243 |
®ister_offset)) { |
| 244 |
if (copy_from_user((void *)&virtvdev->pci_cmd + register_offset, |
| 245 |
buf + copy_offset, |
| 246 |
copy_count)) |
| 247 |
return -EFAULT; |
| 248 |
} |
| 249 |
|
| 250 |
if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_0, |
| 251 |
sizeof(virtvdev->pci_base_addr_0), |
| 252 |
©_offset, ©_count, |
| 253 |
®ister_offset)) { |
| 254 |
if (copy_from_user((void *)&virtvdev->pci_base_addr_0 + register_offset, |
| 255 |
buf + copy_offset, |
| 256 |
copy_count)) |
| 257 |
return -EFAULT; |
| 258 |
} |
| 259 |
|
| 260 |
return vfio_pci_core_write(core_vdev, buf, count, ppos); |
| 261 |
} |
| 262 |
|
| 263 |
ssize_t virtiovf_pci_core_write(struct vfio_device *core_vdev, const char __user *buf, |
| 264 |
size_t count, loff_t *ppos) |
| 265 |
{ |
| 266 |
struct virtiovf_pci_core_device *virtvdev = container_of( |
| 267 |
core_vdev, struct virtiovf_pci_core_device, core_device.vdev); |
| 268 |
unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); |
| 269 |
loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; |
| 270 |
|
| 271 |
if (!count) |
| 272 |
return 0; |
| 273 |
|
| 274 |
if (index == VFIO_PCI_CONFIG_REGION_INDEX) |
| 275 |
return virtiovf_pci_write_config(core_vdev, buf, count, ppos); |
| 276 |
|
| 277 |
if (index == VFIO_PCI_BAR0_REGION_INDEX) |
| 278 |
return virtiovf_pci_bar0_rw(virtvdev, pos, (char __user *)buf, count, false); |
| 279 |
|
| 280 |
return vfio_pci_core_write(core_vdev, buf, count, ppos); |
| 281 |
} |
| 282 |
|
| 283 |
int virtiovf_pci_ioctl_get_region_info(struct vfio_device *core_vdev, |
| 284 |
struct vfio_region_info *info, |
| 285 |
struct vfio_info_cap *caps) |
| 286 |
{ |
| 287 |
struct virtiovf_pci_core_device *virtvdev = container_of( |
| 288 |
core_vdev, struct virtiovf_pci_core_device, core_device.vdev); |
| 289 |
|
| 290 |
if (info->index != VFIO_PCI_BAR0_REGION_INDEX) |
| 291 |
return vfio_pci_ioctl_get_region_info(core_vdev, info, caps); |
| 292 |
|
| 293 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 294 |
info->size = virtvdev->bar0_virtual_buf_size; |
| 295 |
info->flags = VFIO_REGION_INFO_FLAG_READ | VFIO_REGION_INFO_FLAG_WRITE; |
| 296 |
return 0; |
| 297 |
} |
| 298 |
|
| 299 |
static int virtiovf_set_notify_addr(struct virtiovf_pci_core_device *virtvdev) |
| 300 |
{ |
| 301 |
struct vfio_pci_core_device *core_device = &virtvdev->core_device; |
| 302 |
int ret; |
| 303 |
|
| 304 |
/* |
| 305 |
* Setup the BAR where the 'notify' exists to be used by vfio as well |
| 306 |
* This will let us mmap it only once and use it when needed. |
| 307 |
*/ |
| 308 |
ret = vfio_pci_core_setup_barmap(core_device, |
| 309 |
virtvdev->notify_bar); |
| 310 |
if (ret) |
| 311 |
return ret; |
| 312 |
|
| 313 |
virtvdev->notify_addr = core_device->barmap[virtvdev->notify_bar] + |
| 314 |
virtvdev->notify_offset; |
| 315 |
return 0; |
| 316 |
} |
| 317 |
|
| 318 |
int virtiovf_open_legacy_io(struct virtiovf_pci_core_device *virtvdev) |
| 319 |
{ |
| 320 |
if (!virtvdev->bar0_virtual_buf) |
| 321 |
return 0; |
| 322 |
|
| 323 |
/* |
| 324 |
* Upon close_device() the vfio_pci_core_disable() is called |
| 325 |
* and will close all the previous mmaps, so it seems that the |
| 326 |
* valid life cycle for the 'notify' addr is per open/close. |
| 327 |
*/ |
| 328 |
return virtiovf_set_notify_addr(virtvdev); |
| 329 |
} |
| 330 |
|
| 331 |
static int virtiovf_get_device_config_size(unsigned short device) |
| 332 |
{ |
| 333 |
/* Network card */ |
| 334 |
return offsetofend(struct virtio_net_config, status); |
| 335 |
} |
| 336 |
|
| 337 |
static int virtiovf_read_notify_info(struct virtiovf_pci_core_device *virtvdev) |
| 338 |
{ |
| 339 |
u64 offset; |
| 340 |
int ret; |
| 341 |
u8 bar; |
| 342 |
|
| 343 |
ret = virtio_pci_admin_legacy_io_notify_info(virtvdev->core_device.pdev, |
| 344 |
VIRTIO_ADMIN_CMD_NOTIFY_INFO_FLAGS_OWNER_MEM, |
| 345 |
&bar, &offset); |
| 346 |
if (ret) |
| 347 |
return ret; |
| 348 |
|
| 349 |
virtvdev->notify_bar = bar; |
| 350 |
virtvdev->notify_offset = offset; |
| 351 |
return 0; |
| 352 |
} |
| 353 |
|
| 354 |
static bool virtiovf_bar0_exists(struct pci_dev *pdev) |
| 355 |
{ |
| 356 |
struct resource *res = pdev->resource; |
| 357 |
|
| 358 |
return res->flags; |
| 359 |
} |
| 360 |
|
| 361 |
bool virtiovf_support_legacy_io(struct pci_dev *pdev) |
| 362 |
{ |
| 363 |
/* For now, the legacy IO functionality is supported only for virtio-net */ |
| 364 |
return pdev->device == 0x1041 && virtio_pci_admin_has_legacy_io(pdev) && |
| 365 |
!virtiovf_bar0_exists(pdev); |
| 366 |
} |
| 367 |
|
| 368 |
int virtiovf_init_legacy_io(struct virtiovf_pci_core_device *virtvdev) |
| 369 |
{ |
| 370 |
struct pci_dev *pdev = virtvdev->core_device.pdev; |
| 371 |
int ret; |
| 372 |
|
| 373 |
ret = virtiovf_read_notify_info(virtvdev); |
| 374 |
if (ret) |
| 375 |
return ret; |
| 376 |
|
| 377 |
virtvdev->bar0_virtual_buf_size = VIRTIO_PCI_CONFIG_OFF(true) + |
| 378 |
virtiovf_get_device_config_size(pdev->device); |
| 379 |
BUILD_BUG_ON(!is_power_of_2(virtvdev->bar0_virtual_buf_size)); |
| 380 |
virtvdev->bar0_virtual_buf = kzalloc(virtvdev->bar0_virtual_buf_size, |
| 381 |
GFP_KERNEL); |
| 382 |
if (!virtvdev->bar0_virtual_buf) |
| 383 |
return -ENOMEM; |
| 384 |
mutex_init(&virtvdev->bar_mutex); |
| 385 |
return 0; |
| 386 |
} |
| 387 |
|
| 388 |
void virtiovf_release_legacy_io(struct virtiovf_pci_core_device *virtvdev) |
| 389 |
{ |
| 390 |
kfree(virtvdev->bar0_virtual_buf); |
| 391 |
} |
| 392 |
|
| 393 |
void virtiovf_legacy_io_reset_done(struct pci_dev *pdev) |
| 394 |
{ |
| 395 |
struct virtiovf_pci_core_device *virtvdev = dev_get_drvdata(&pdev->dev); |
| 396 |
|
| 397 |
virtvdev->pci_cmd = 0; |
| 398 |
} |