| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* Copyright (C) 2012 Red Hat, Inc. All rights reserved. |
| 4 |
* Author: Alex Williamson <[email protected]> |
| 5 |
* |
| 6 |
* Derived from original vfio: |
| 7 |
* Copyright 2010 Cisco Systems, Inc. All rights reserved. |
| 8 |
* Author: Tom Lyon, [email protected] |
| 9 |
*/ |
| 10 |
|
| 11 |
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 12 |
|
| 13 |
#include <linux/aperture.h> |
| 14 |
#include <linux/device.h> |
| 15 |
#include <linux/eventfd.h> |
| 16 |
#include <linux/file.h> |
| 17 |
#include <linux/interrupt.h> |
| 18 |
#include <linux/iommu.h> |
| 19 |
#include <linux/module.h> |
| 20 |
#include <linux/mutex.h> |
| 21 |
#include <linux/notifier.h> |
| 22 |
#include <linux/pci.h> |
| 23 |
#include <linux/pm_runtime.h> |
| 24 |
#include <linux/slab.h> |
| 25 |
#include <linux/types.h> |
| 26 |
#include <linux/uaccess.h> |
| 27 |
#include <linux/vgaarb.h> |
| 28 |
#include <linux/nospec.h> |
| 29 |
#include <linux/sched/mm.h> |
| 30 |
#include <linux/iommufd.h> |
| 31 |
#include <linux/pci-p2pdma.h> |
| 32 |
#if IS_ENABLED(CONFIG_EEH) |
| 33 |
#include <asm/eeh.h> |
| 34 |
#endif |
| 35 |
|
| 36 |
#include "vfio_pci_priv.h" |
| 37 |
|
| 38 |
#define DRIVER_AUTHOR "Alex Williamson <[email protected]>" |
| 39 |
#define DRIVER_DESC "core driver for VFIO based PCI devices" |
| 40 |
|
| 41 |
static bool nointxmask; |
| 42 |
static bool disable_vga; |
| 43 |
static bool disable_idle_d3; |
| 44 |
|
| 45 |
static void vfio_pci_eventfd_rcu_free(struct rcu_head *rcu) |
| 46 |
{ |
| 47 |
struct vfio_pci_eventfd *eventfd = |
| 48 |
container_of(rcu, struct vfio_pci_eventfd, rcu); |
| 49 |
|
| 50 |
eventfd_ctx_put(eventfd->ctx); |
| 51 |
kfree(eventfd); |
| 52 |
} |
| 53 |
|
| 54 |
int vfio_pci_eventfd_replace_locked(struct vfio_pci_core_device *vdev, |
| 55 |
struct vfio_pci_eventfd __rcu **peventfd, |
| 56 |
struct eventfd_ctx *ctx) |
| 57 |
{ |
| 58 |
struct vfio_pci_eventfd *new = NULL; |
| 59 |
struct vfio_pci_eventfd *old; |
| 60 |
|
| 61 |
lockdep_assert_held(&vdev->igate); |
| 62 |
|
| 63 |
if (ctx) { |
| 64 |
new = kzalloc_obj(*new, GFP_KERNEL_ACCOUNT); |
| 65 |
if (!new) |
| 66 |
return -ENOMEM; |
| 67 |
|
| 68 |
new->ctx = ctx; |
| 69 |
} |
| 70 |
|
| 71 |
old = rcu_replace_pointer(*peventfd, new, |
| 72 |
lockdep_is_held(&vdev->igate)); |
| 73 |
if (old) |
| 74 |
call_rcu(&old->rcu, vfio_pci_eventfd_rcu_free); |
| 75 |
|
| 76 |
return 0; |
| 77 |
} |
| 78 |
|
| 79 |
/* List of PF's that vfio_pci_core_sriov_configure() has been called on */ |
| 80 |
static DEFINE_MUTEX(vfio_pci_sriov_pfs_mutex); |
| 81 |
static LIST_HEAD(vfio_pci_sriov_pfs); |
| 82 |
|
| 83 |
struct vfio_pci_dummy_resource { |
| 84 |
struct resource resource; |
| 85 |
int index; |
| 86 |
struct list_head res_next; |
| 87 |
}; |
| 88 |
|
| 89 |
struct vfio_pci_vf_token { |
| 90 |
struct mutex lock; |
| 91 |
uuid_t uuid; |
| 92 |
int users; |
| 93 |
}; |
| 94 |
|
| 95 |
static inline bool vfio_vga_disabled(void) |
| 96 |
{ |
| 97 |
#ifdef CONFIG_VFIO_PCI_VGA |
| 98 |
return disable_vga; |
| 99 |
#else |
| 100 |
return true; |
| 101 |
#endif |
| 102 |
} |
| 103 |
|
| 104 |
/* |
| 105 |
* Our VGA arbiter participation is limited since we don't know anything |
| 106 |
* about the device itself. However, if the device is the only VGA device |
| 107 |
* downstream of a bridge and VFIO VGA support is disabled, then we can |
| 108 |
* safely return legacy VGA IO and memory as not decoded since the user |
| 109 |
* has no way to get to it and routing can be disabled externally at the |
| 110 |
* bridge. |
| 111 |
*/ |
| 112 |
static unsigned int vfio_pci_set_decode(struct pci_dev *pdev, bool single_vga) |
| 113 |
{ |
| 114 |
struct pci_dev *tmp = NULL; |
| 115 |
unsigned char max_busnr; |
| 116 |
unsigned int decodes; |
| 117 |
|
| 118 |
if (single_vga || !vfio_vga_disabled() || pci_is_root_bus(pdev->bus)) |
| 119 |
return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM | |
| 120 |
VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM; |
| 121 |
|
| 122 |
max_busnr = pci_bus_max_busnr(pdev->bus); |
| 123 |
decodes = VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; |
| 124 |
|
| 125 |
while ((tmp = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, tmp)) != NULL) { |
| 126 |
if (tmp == pdev || |
| 127 |
pci_domain_nr(tmp->bus) != pci_domain_nr(pdev->bus) || |
| 128 |
pci_is_root_bus(tmp->bus)) |
| 129 |
continue; |
| 130 |
|
| 131 |
if (tmp->bus->number >= pdev->bus->number && |
| 132 |
tmp->bus->number <= max_busnr) { |
| 133 |
pci_dev_put(tmp); |
| 134 |
decodes |= VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM; |
| 135 |
break; |
| 136 |
} |
| 137 |
} |
| 138 |
|
| 139 |
return decodes; |
| 140 |
} |
| 141 |
|
| 142 |
static void vfio_pci_probe_mmaps(struct vfio_pci_core_device *vdev) |
| 143 |
{ |
| 144 |
struct resource *res; |
| 145 |
int i; |
| 146 |
struct vfio_pci_dummy_resource *dummy_res; |
| 147 |
|
| 148 |
for (i = 0; i < PCI_STD_NUM_BARS; i++) { |
| 149 |
int bar = i + PCI_STD_RESOURCES; |
| 150 |
|
| 151 |
res = &vdev->pdev->resource[bar]; |
| 152 |
|
| 153 |
if (vdev->pdev->non_mappable_bars) |
| 154 |
goto no_mmap; |
| 155 |
|
| 156 |
if (!(res->flags & IORESOURCE_MEM)) |
| 157 |
goto no_mmap; |
| 158 |
|
| 159 |
/* |
| 160 |
* The PCI core shouldn't set up a resource with a |
| 161 |
* type but zero size. But there may be bugs that |
| 162 |
* cause us to do that. |
| 163 |
*/ |
| 164 |
if (!resource_size(res)) |
| 165 |
goto no_mmap; |
| 166 |
|
| 167 |
if (resource_size(res) >= PAGE_SIZE) { |
| 168 |
vdev->bar_mmap_supported[bar] = true; |
| 169 |
continue; |
| 170 |
} |
| 171 |
|
| 172 |
if (!(res->start & ~PAGE_MASK)) { |
| 173 |
/* |
| 174 |
* Add a dummy resource to reserve the remainder |
| 175 |
* of the exclusive page in case that hot-add |
| 176 |
* device's bar is assigned into it. |
| 177 |
*/ |
| 178 |
dummy_res = kzalloc_obj(*dummy_res, GFP_KERNEL_ACCOUNT); |
| 179 |
if (dummy_res == NULL) |
| 180 |
goto no_mmap; |
| 181 |
|
| 182 |
dummy_res->resource.name = "vfio sub-page reserved"; |
| 183 |
dummy_res->resource.start = res->end + 1; |
| 184 |
dummy_res->resource.end = res->start + PAGE_SIZE - 1; |
| 185 |
dummy_res->resource.flags = res->flags; |
| 186 |
if (request_resource(res->parent, |
| 187 |
&dummy_res->resource)) { |
| 188 |
kfree(dummy_res); |
| 189 |
goto no_mmap; |
| 190 |
} |
| 191 |
dummy_res->index = bar; |
| 192 |
list_add(&dummy_res->res_next, |
| 193 |
&vdev->dummy_resources_list); |
| 194 |
vdev->bar_mmap_supported[bar] = true; |
| 195 |
continue; |
| 196 |
} |
| 197 |
/* |
| 198 |
* Here we don't handle the case when the BAR is not page |
| 199 |
* aligned because we can't expect the BAR will be |
| 200 |
* assigned into the same location in a page in guest |
| 201 |
* when we passthrough the BAR. And it's hard to access |
| 202 |
* this BAR in userspace because we have no way to get |
| 203 |
* the BAR's location in a page. |
| 204 |
*/ |
| 205 |
no_mmap: |
| 206 |
vdev->bar_mmap_supported[bar] = false; |
| 207 |
} |
| 208 |
} |
| 209 |
|
| 210 |
struct vfio_pci_group_info; |
| 211 |
static void vfio_pci_dev_set_try_reset(struct vfio_device_set *dev_set); |
| 212 |
static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set, |
| 213 |
struct vfio_pci_group_info *groups, |
| 214 |
struct iommufd_ctx *iommufd_ctx); |
| 215 |
|
| 216 |
/* |
| 217 |
* INTx masking requires the ability to disable INTx signaling via PCI_COMMAND |
| 218 |
* _and_ the ability detect when the device is asserting INTx via PCI_STATUS. |
| 219 |
* If a device implements the former but not the latter we would typically |
| 220 |
* expect broken_intx_masking be set and require an exclusive interrupt. |
| 221 |
* However since we do have control of the device's ability to assert INTx, |
| 222 |
* we can instead pretend that the device does not implement INTx, virtualizing |
| 223 |
* the pin register to report zero and maintaining DisINTx set on the host. |
| 224 |
*/ |
| 225 |
static bool vfio_pci_nointx(struct pci_dev *pdev) |
| 226 |
{ |
| 227 |
switch (pdev->vendor) { |
| 228 |
case PCI_VENDOR_ID_INTEL: |
| 229 |
switch (pdev->device) { |
| 230 |
/* All i40e (XL710/X710/XXV710) 10/20/25/40GbE NICs */ |
| 231 |
case 0x1572: |
| 232 |
case 0x1574: |
| 233 |
case 0x1580 ... 0x1581: |
| 234 |
case 0x1583 ... 0x158b: |
| 235 |
case 0x37d0 ... 0x37d2: |
| 236 |
/* X550 */ |
| 237 |
case 0x1563: |
| 238 |
return true; |
| 239 |
default: |
| 240 |
return false; |
| 241 |
} |
| 242 |
} |
| 243 |
|
| 244 |
return false; |
| 245 |
} |
| 246 |
|
| 247 |
static void vfio_pci_probe_power_state(struct vfio_pci_core_device *vdev) |
| 248 |
{ |
| 249 |
struct pci_dev *pdev = vdev->pdev; |
| 250 |
u16 pmcsr; |
| 251 |
|
| 252 |
if (!pdev->pm_cap) |
| 253 |
return; |
| 254 |
|
| 255 |
pci_read_config_word(pdev, pdev->pm_cap + PCI_PM_CTRL, &pmcsr); |
| 256 |
|
| 257 |
vdev->needs_pm_restore = !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET); |
| 258 |
} |
| 259 |
|
| 260 |
/* |
| 261 |
* pci_set_power_state() wrapper handling devices which perform a soft reset on |
| 262 |
* D3->D0 transition. Save state prior to D0/1/2->D3, stash it on the vdev, |
| 263 |
* restore when returned to D0. Saved separately from pci_saved_state for use |
| 264 |
* by PM capability emulation and separately from pci_dev internal saved state |
| 265 |
* to avoid it being overwritten and consumed around other resets. |
| 266 |
*/ |
| 267 |
int vfio_pci_set_power_state(struct vfio_pci_core_device *vdev, pci_power_t state) |
| 268 |
{ |
| 269 |
struct pci_dev *pdev = vdev->pdev; |
| 270 |
bool needs_restore = false, needs_save = false; |
| 271 |
int ret; |
| 272 |
|
| 273 |
/* Prevent changing power state for PFs with VFs enabled */ |
| 274 |
if (pci_num_vf(pdev) && state > PCI_D0) |
| 275 |
return -EBUSY; |
| 276 |
|
| 277 |
if (vdev->needs_pm_restore) { |
| 278 |
if (pdev->current_state < PCI_D3hot && state >= PCI_D3hot) { |
| 279 |
pci_save_state(pdev); |
| 280 |
needs_save = true; |
| 281 |
} |
| 282 |
|
| 283 |
if (pdev->current_state >= PCI_D3hot && state <= PCI_D0) |
| 284 |
needs_restore = true; |
| 285 |
} |
| 286 |
|
| 287 |
ret = pci_set_power_state(pdev, state); |
| 288 |
|
| 289 |
if (!ret) { |
| 290 |
/* D3 might be unsupported via quirk, skip unless in D3 */ |
| 291 |
if (needs_save && pdev->current_state >= PCI_D3hot) { |
| 292 |
/* |
| 293 |
* The current PCI state will be saved locally in |
| 294 |
* 'pm_save' during the D3hot transition. When the |
| 295 |
* device state is changed to D0 again with the current |
| 296 |
* function, then pci_store_saved_state() will restore |
| 297 |
* the state and will free the memory pointed by |
| 298 |
* 'pm_save'. There are few cases where the PCI power |
| 299 |
* state can be changed to D0 without the involvement |
| 300 |
* of the driver. For these cases, free the earlier |
| 301 |
* allocated memory first before overwriting 'pm_save' |
| 302 |
* to prevent the memory leak. |
| 303 |
*/ |
| 304 |
kfree(vdev->pm_save); |
| 305 |
vdev->pm_save = pci_store_saved_state(pdev); |
| 306 |
} else if (needs_restore) { |
| 307 |
pci_load_and_free_saved_state(pdev, &vdev->pm_save); |
| 308 |
pci_restore_state(pdev); |
| 309 |
} |
| 310 |
} |
| 311 |
|
| 312 |
return ret; |
| 313 |
} |
| 314 |
|
| 315 |
static int vfio_pci_runtime_pm_entry(struct vfio_pci_core_device *vdev, |
| 316 |
struct eventfd_ctx *efdctx) |
| 317 |
{ |
| 318 |
/* |
| 319 |
* The vdev power related flags are protected with 'memory_lock' |
| 320 |
* semaphore. |
| 321 |
*/ |
| 322 |
vfio_pci_zap_and_down_write_memory_lock(vdev); |
| 323 |
vfio_pci_dma_buf_move(vdev, true); |
| 324 |
|
| 325 |
if (vdev->pm_runtime_engaged) { |
| 326 |
up_write(&vdev->memory_lock); |
| 327 |
return -EINVAL; |
| 328 |
} |
| 329 |
|
| 330 |
vdev->pm_runtime_engaged = true; |
| 331 |
vdev->pm_wake_eventfd_ctx = efdctx; |
| 332 |
pm_runtime_put_noidle(&vdev->pdev->dev); |
| 333 |
up_write(&vdev->memory_lock); |
| 334 |
|
| 335 |
return 0; |
| 336 |
} |
| 337 |
|
| 338 |
static int vfio_pci_core_pm_entry(struct vfio_pci_core_device *vdev, u32 flags, |
| 339 |
void __user *arg, size_t argsz) |
| 340 |
{ |
| 341 |
int ret; |
| 342 |
|
| 343 |
ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_SET, 0); |
| 344 |
if (ret != 1) |
| 345 |
return ret; |
| 346 |
|
| 347 |
/* |
| 348 |
* Inside vfio_pci_runtime_pm_entry(), only the runtime PM usage count |
| 349 |
* will be decremented. The pm_runtime_put() will be invoked again |
| 350 |
* while returning from the ioctl and then the device can go into |
| 351 |
* runtime suspended state. |
| 352 |
*/ |
| 353 |
return vfio_pci_runtime_pm_entry(vdev, NULL); |
| 354 |
} |
| 355 |
|
| 356 |
static int vfio_pci_core_pm_entry_with_wakeup( |
| 357 |
struct vfio_pci_core_device *vdev, u32 flags, |
| 358 |
struct vfio_device_low_power_entry_with_wakeup __user *arg, |
| 359 |
size_t argsz) |
| 360 |
{ |
| 361 |
struct vfio_device_low_power_entry_with_wakeup entry; |
| 362 |
struct eventfd_ctx *efdctx; |
| 363 |
int ret; |
| 364 |
|
| 365 |
ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_SET, |
| 366 |
sizeof(entry)); |
| 367 |
if (ret != 1) |
| 368 |
return ret; |
| 369 |
|
| 370 |
if (copy_from_user(&entry, arg, sizeof(entry))) |
| 371 |
return -EFAULT; |
| 372 |
|
| 373 |
if (entry.wakeup_eventfd < 0) |
| 374 |
return -EINVAL; |
| 375 |
|
| 376 |
efdctx = eventfd_ctx_fdget(entry.wakeup_eventfd); |
| 377 |
if (IS_ERR(efdctx)) |
| 378 |
return PTR_ERR(efdctx); |
| 379 |
|
| 380 |
ret = vfio_pci_runtime_pm_entry(vdev, efdctx); |
| 381 |
if (ret) |
| 382 |
eventfd_ctx_put(efdctx); |
| 383 |
|
| 384 |
return ret; |
| 385 |
} |
| 386 |
|
| 387 |
static void __vfio_pci_runtime_pm_exit(struct vfio_pci_core_device *vdev) |
| 388 |
{ |
| 389 |
if (vdev->pm_runtime_engaged) { |
| 390 |
vdev->pm_runtime_engaged = false; |
| 391 |
pm_runtime_get_noresume(&vdev->pdev->dev); |
| 392 |
|
| 393 |
if (vdev->pm_wake_eventfd_ctx) { |
| 394 |
eventfd_ctx_put(vdev->pm_wake_eventfd_ctx); |
| 395 |
vdev->pm_wake_eventfd_ctx = NULL; |
| 396 |
} |
| 397 |
} |
| 398 |
} |
| 399 |
|
| 400 |
static void vfio_pci_runtime_pm_exit(struct vfio_pci_core_device *vdev) |
| 401 |
{ |
| 402 |
/* |
| 403 |
* The vdev power related flags are protected with 'memory_lock' |
| 404 |
* semaphore. |
| 405 |
*/ |
| 406 |
down_write(&vdev->memory_lock); |
| 407 |
__vfio_pci_runtime_pm_exit(vdev); |
| 408 |
if (__vfio_pci_memory_enabled(vdev)) |
| 409 |
vfio_pci_dma_buf_move(vdev, false); |
| 410 |
up_write(&vdev->memory_lock); |
| 411 |
} |
| 412 |
|
| 413 |
static int vfio_pci_core_pm_exit(struct vfio_pci_core_device *vdev, u32 flags, |
| 414 |
void __user *arg, size_t argsz) |
| 415 |
{ |
| 416 |
int ret; |
| 417 |
|
| 418 |
ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_SET, 0); |
| 419 |
if (ret != 1) |
| 420 |
return ret; |
| 421 |
|
| 422 |
/* |
| 423 |
* The device is always in the active state here due to pm wrappers |
| 424 |
* around ioctls. If the device had entered a low power state and |
| 425 |
* pm_wake_eventfd_ctx is valid, vfio_pci_core_runtime_resume() has |
| 426 |
* already signaled the eventfd and exited low power mode itself. |
| 427 |
* pm_runtime_engaged protects the redundant call here. |
| 428 |
*/ |
| 429 |
vfio_pci_runtime_pm_exit(vdev); |
| 430 |
return 0; |
| 431 |
} |
| 432 |
|
| 433 |
#ifdef CONFIG_PM |
| 434 |
static int vfio_pci_core_runtime_suspend(struct device *dev) |
| 435 |
{ |
| 436 |
struct vfio_pci_core_device *vdev = dev_get_drvdata(dev); |
| 437 |
|
| 438 |
down_write(&vdev->memory_lock); |
| 439 |
/* |
| 440 |
* The user can move the device into D3hot state before invoking |
| 441 |
* power management IOCTL. Move the device into D0 state here and then |
| 442 |
* the pci-driver core runtime PM suspend function will move the device |
| 443 |
* into the low power state. Also, for the devices which have |
| 444 |
* NoSoftRst-, it will help in restoring the original state |
| 445 |
* (saved locally in 'vdev->pm_save'). |
| 446 |
*/ |
| 447 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 448 |
up_write(&vdev->memory_lock); |
| 449 |
|
| 450 |
/* |
| 451 |
* If INTx is enabled, then mask INTx before going into the runtime |
| 452 |
* suspended state and unmask the same in the runtime resume. |
| 453 |
* If INTx has already been masked by the user, then |
| 454 |
* vfio_pci_intx_mask() will return false and in that case, INTx |
| 455 |
* should not be unmasked in the runtime resume. |
| 456 |
*/ |
| 457 |
vdev->pm_intx_masked = ((vdev->irq_type == VFIO_PCI_INTX_IRQ_INDEX) && |
| 458 |
vfio_pci_intx_mask(vdev)); |
| 459 |
|
| 460 |
return 0; |
| 461 |
} |
| 462 |
|
| 463 |
static int vfio_pci_core_runtime_resume(struct device *dev) |
| 464 |
{ |
| 465 |
struct vfio_pci_core_device *vdev = dev_get_drvdata(dev); |
| 466 |
|
| 467 |
/* |
| 468 |
* Resume with a pm_wake_eventfd_ctx signals the eventfd and exit |
| 469 |
* low power mode. |
| 470 |
*/ |
| 471 |
down_write(&vdev->memory_lock); |
| 472 |
if (vdev->pm_wake_eventfd_ctx) { |
| 473 |
eventfd_signal(vdev->pm_wake_eventfd_ctx); |
| 474 |
__vfio_pci_runtime_pm_exit(vdev); |
| 475 |
} |
| 476 |
up_write(&vdev->memory_lock); |
| 477 |
|
| 478 |
if (vdev->pm_intx_masked) |
| 479 |
vfio_pci_intx_unmask(vdev); |
| 480 |
|
| 481 |
return 0; |
| 482 |
} |
| 483 |
#endif /* CONFIG_PM */ |
| 484 |
|
| 485 |
/* |
| 486 |
* The pci-driver core runtime PM routines always save the device state |
| 487 |
* before going into suspended state. If the device is going into low power |
| 488 |
* state with only with runtime PM ops, then no explicit handling is needed |
| 489 |
* for the devices which have NoSoftRst-. |
| 490 |
*/ |
| 491 |
static const struct dev_pm_ops vfio_pci_core_pm_ops = { |
| 492 |
SET_RUNTIME_PM_OPS(vfio_pci_core_runtime_suspend, |
| 493 |
vfio_pci_core_runtime_resume, |
| 494 |
NULL) |
| 495 |
}; |
| 496 |
|
| 497 |
int vfio_pci_core_enable(struct vfio_pci_core_device *vdev) |
| 498 |
{ |
| 499 |
struct pci_dev *pdev = vdev->pdev; |
| 500 |
int ret; |
| 501 |
u16 cmd; |
| 502 |
u8 msix_pos; |
| 503 |
|
| 504 |
if (!disable_idle_d3) { |
| 505 |
ret = pm_runtime_resume_and_get(&pdev->dev); |
| 506 |
if (ret < 0) |
| 507 |
return ret; |
| 508 |
} |
| 509 |
|
| 510 |
/* Don't allow our initial saved state to include busmaster */ |
| 511 |
pci_clear_master(pdev); |
| 512 |
|
| 513 |
ret = pci_enable_device(pdev); |
| 514 |
if (ret) |
| 515 |
goto out_power; |
| 516 |
|
| 517 |
/* If reset fails because of the device lock, fail this path entirely */ |
| 518 |
ret = pci_try_reset_function(pdev); |
| 519 |
if (ret == -EAGAIN) |
| 520 |
goto out_disable_device; |
| 521 |
|
| 522 |
vdev->reset_works = !ret; |
| 523 |
pci_save_state(pdev); |
| 524 |
vdev->pci_saved_state = pci_store_saved_state(pdev); |
| 525 |
if (!vdev->pci_saved_state) |
| 526 |
pci_dbg(pdev, "%s: Couldn't store saved state\n", __func__); |
| 527 |
|
| 528 |
if (likely(!nointxmask)) { |
| 529 |
if (vfio_pci_nointx(pdev)) { |
| 530 |
pci_info(pdev, "Masking broken INTx support\n"); |
| 531 |
vdev->nointx = true; |
| 532 |
pci_intx(pdev, 0); |
| 533 |
} else |
| 534 |
vdev->pci_2_3 = pci_intx_mask_supported(pdev); |
| 535 |
} |
| 536 |
|
| 537 |
pci_read_config_word(pdev, PCI_COMMAND, &cmd); |
| 538 |
if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) { |
| 539 |
cmd &= ~PCI_COMMAND_INTX_DISABLE; |
| 540 |
pci_write_config_word(pdev, PCI_COMMAND, cmd); |
| 541 |
} |
| 542 |
|
| 543 |
ret = vfio_pci_zdev_open_device(vdev); |
| 544 |
if (ret) |
| 545 |
goto out_free_state; |
| 546 |
|
| 547 |
ret = vfio_config_init(vdev); |
| 548 |
if (ret) |
| 549 |
goto out_free_zdev; |
| 550 |
|
| 551 |
msix_pos = pdev->msix_cap; |
| 552 |
if (msix_pos) { |
| 553 |
u16 flags; |
| 554 |
u32 table; |
| 555 |
|
| 556 |
pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags); |
| 557 |
pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table); |
| 558 |
|
| 559 |
vdev->msix_bar = table & PCI_MSIX_TABLE_BIR; |
| 560 |
vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET; |
| 561 |
vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16; |
| 562 |
vdev->has_dyn_msix = pci_msix_can_alloc_dyn(pdev); |
| 563 |
} else { |
| 564 |
vdev->msix_bar = 0xFF; |
| 565 |
vdev->has_dyn_msix = false; |
| 566 |
} |
| 567 |
|
| 568 |
if (!vfio_vga_disabled() && vfio_pci_is_vga(pdev)) |
| 569 |
vdev->has_vga = true; |
| 570 |
|
| 571 |
|
| 572 |
return 0; |
| 573 |
|
| 574 |
out_free_zdev: |
| 575 |
vfio_pci_zdev_close_device(vdev); |
| 576 |
out_free_state: |
| 577 |
kfree(vdev->pci_saved_state); |
| 578 |
vdev->pci_saved_state = NULL; |
| 579 |
out_disable_device: |
| 580 |
pci_disable_device(pdev); |
| 581 |
out_power: |
| 582 |
if (!disable_idle_d3) |
| 583 |
pm_runtime_put(&pdev->dev); |
| 584 |
return ret; |
| 585 |
} |
| 586 |
EXPORT_SYMBOL_GPL(vfio_pci_core_enable); |
| 587 |
|
| 588 |
void vfio_pci_core_disable(struct vfio_pci_core_device *vdev) |
| 589 |
{ |
| 590 |
struct pci_dev *bridge; |
| 591 |
struct pci_dev *pdev = vdev->pdev; |
| 592 |
struct vfio_pci_dummy_resource *dummy_res, *tmp; |
| 593 |
struct vfio_pci_ioeventfd *ioeventfd, *ioeventfd_tmp; |
| 594 |
int i, bar; |
| 595 |
|
| 596 |
/* For needs_reset */ |
| 597 |
lockdep_assert_held(&vdev->vdev.dev_set->lock); |
| 598 |
|
| 599 |
/* |
| 600 |
* This function can be invoked while the power state is non-D0. |
| 601 |
* This non-D0 power state can be with or without runtime PM. |
| 602 |
* vfio_pci_runtime_pm_exit() will internally increment the usage |
| 603 |
* count corresponding to pm_runtime_put() called during low power |
| 604 |
* feature entry and then pm_runtime_resume() will wake up the device, |
| 605 |
* if the device has already gone into the suspended state. Otherwise, |
| 606 |
* the vfio_pci_set_power_state() will change the device power state |
| 607 |
* to D0. |
| 608 |
*/ |
| 609 |
vfio_pci_runtime_pm_exit(vdev); |
| 610 |
pm_runtime_resume(&pdev->dev); |
| 611 |
|
| 612 |
/* |
| 613 |
* This function calls __pci_reset_function_locked() which internally |
| 614 |
* can use pci_pm_reset() for the function reset. pci_pm_reset() will |
| 615 |
* fail if the power state is non-D0. Also, for the devices which |
| 616 |
* have NoSoftRst-, the reset function can cause the PCI config space |
| 617 |
* reset without restoring the original state (saved locally in |
| 618 |
* 'vdev->pm_save'). |
| 619 |
*/ |
| 620 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 621 |
|
| 622 |
/* Stop the device from further DMA */ |
| 623 |
pci_clear_master(pdev); |
| 624 |
|
| 625 |
vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE | |
| 626 |
VFIO_IRQ_SET_ACTION_TRIGGER, |
| 627 |
vdev->irq_type, 0, 0, NULL); |
| 628 |
|
| 629 |
/* Device closed, don't need mutex here */ |
| 630 |
list_for_each_entry_safe(ioeventfd, ioeventfd_tmp, |
| 631 |
&vdev->ioeventfds_list, next) { |
| 632 |
vfio_virqfd_disable(&ioeventfd->virqfd); |
| 633 |
list_del(&ioeventfd->next); |
| 634 |
kfree(ioeventfd); |
| 635 |
} |
| 636 |
vdev->ioeventfds_nr = 0; |
| 637 |
|
| 638 |
vdev->virq_disabled = false; |
| 639 |
|
| 640 |
for (i = 0; i < vdev->num_regions; i++) |
| 641 |
vdev->region[i].ops->release(vdev, &vdev->region[i]); |
| 642 |
|
| 643 |
vdev->num_regions = 0; |
| 644 |
kfree(vdev->region); |
| 645 |
vdev->region = NULL; /* don't krealloc a freed pointer */ |
| 646 |
|
| 647 |
vfio_config_free(vdev); |
| 648 |
|
| 649 |
for (i = 0; i < PCI_STD_NUM_BARS; i++) { |
| 650 |
bar = i + PCI_STD_RESOURCES; |
| 651 |
if (!vdev->barmap[bar]) |
| 652 |
continue; |
| 653 |
pci_iounmap(pdev, vdev->barmap[bar]); |
| 654 |
pci_release_selected_regions(pdev, 1 << bar); |
| 655 |
vdev->barmap[bar] = NULL; |
| 656 |
} |
| 657 |
|
| 658 |
list_for_each_entry_safe(dummy_res, tmp, |
| 659 |
&vdev->dummy_resources_list, res_next) { |
| 660 |
list_del(&dummy_res->res_next); |
| 661 |
release_resource(&dummy_res->resource); |
| 662 |
kfree(dummy_res); |
| 663 |
} |
| 664 |
|
| 665 |
vdev->needs_reset = true; |
| 666 |
|
| 667 |
vfio_pci_zdev_close_device(vdev); |
| 668 |
|
| 669 |
/* |
| 670 |
* If we have saved state, restore it. If we can reset the device, |
| 671 |
* even better. Resetting with current state seems better than |
| 672 |
* nothing, but saving and restoring current state without reset |
| 673 |
* is just busy work. |
| 674 |
*/ |
| 675 |
if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) { |
| 676 |
pci_info(pdev, "%s: Couldn't reload saved state\n", __func__); |
| 677 |
|
| 678 |
if (!vdev->reset_works) |
| 679 |
goto out; |
| 680 |
|
| 681 |
pci_save_state(pdev); |
| 682 |
} |
| 683 |
|
| 684 |
/* |
| 685 |
* Disable INTx and MSI, presumably to avoid spurious interrupts |
| 686 |
* during reset. Stolen from pci_reset_function() |
| 687 |
*/ |
| 688 |
pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE); |
| 689 |
|
| 690 |
/* |
| 691 |
* Try to get the locks ourselves to prevent a deadlock. The |
| 692 |
* success of this is dependent on being able to lock the device, |
| 693 |
* which is not always possible. |
| 694 |
* We can not use the "try" reset interface here, which will |
| 695 |
* overwrite the previously restored configuration information. |
| 696 |
*/ |
| 697 |
if (vdev->reset_works) { |
| 698 |
bridge = pci_upstream_bridge(pdev); |
| 699 |
if (bridge && !pci_dev_trylock(bridge)) |
| 700 |
goto out_restore_state; |
| 701 |
if (pci_dev_trylock(pdev)) { |
| 702 |
if (!__pci_reset_function_locked(pdev)) |
| 703 |
vdev->needs_reset = false; |
| 704 |
pci_dev_unlock(pdev); |
| 705 |
} |
| 706 |
if (bridge) |
| 707 |
pci_dev_unlock(bridge); |
| 708 |
} |
| 709 |
|
| 710 |
out_restore_state: |
| 711 |
pci_restore_state(pdev); |
| 712 |
out: |
| 713 |
pci_disable_device(pdev); |
| 714 |
|
| 715 |
vfio_pci_dev_set_try_reset(vdev->vdev.dev_set); |
| 716 |
|
| 717 |
/* Put the pm-runtime usage counter acquired during enable */ |
| 718 |
if (!disable_idle_d3) |
| 719 |
pm_runtime_put(&pdev->dev); |
| 720 |
} |
| 721 |
EXPORT_SYMBOL_GPL(vfio_pci_core_disable); |
| 722 |
|
| 723 |
void vfio_pci_core_close_device(struct vfio_device *core_vdev) |
| 724 |
{ |
| 725 |
struct vfio_pci_core_device *vdev = |
| 726 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 727 |
|
| 728 |
if (vdev->sriov_pf_core_dev) { |
| 729 |
mutex_lock(&vdev->sriov_pf_core_dev->vf_token->lock); |
| 730 |
WARN_ON(!vdev->sriov_pf_core_dev->vf_token->users); |
| 731 |
vdev->sriov_pf_core_dev->vf_token->users--; |
| 732 |
mutex_unlock(&vdev->sriov_pf_core_dev->vf_token->lock); |
| 733 |
} |
| 734 |
#if IS_ENABLED(CONFIG_EEH) |
| 735 |
eeh_dev_release(vdev->pdev); |
| 736 |
#endif |
| 737 |
vfio_pci_core_disable(vdev); |
| 738 |
|
| 739 |
vfio_pci_dma_buf_cleanup(vdev); |
| 740 |
|
| 741 |
mutex_lock(&vdev->igate); |
| 742 |
vfio_pci_eventfd_replace_locked(vdev, &vdev->err_trigger, NULL); |
| 743 |
vfio_pci_eventfd_replace_locked(vdev, &vdev->req_trigger, NULL); |
| 744 |
mutex_unlock(&vdev->igate); |
| 745 |
} |
| 746 |
EXPORT_SYMBOL_GPL(vfio_pci_core_close_device); |
| 747 |
|
| 748 |
void vfio_pci_core_finish_enable(struct vfio_pci_core_device *vdev) |
| 749 |
{ |
| 750 |
vfio_pci_probe_mmaps(vdev); |
| 751 |
#if IS_ENABLED(CONFIG_EEH) |
| 752 |
eeh_dev_open(vdev->pdev); |
| 753 |
#endif |
| 754 |
|
| 755 |
if (vdev->sriov_pf_core_dev) { |
| 756 |
mutex_lock(&vdev->sriov_pf_core_dev->vf_token->lock); |
| 757 |
vdev->sriov_pf_core_dev->vf_token->users++; |
| 758 |
mutex_unlock(&vdev->sriov_pf_core_dev->vf_token->lock); |
| 759 |
} |
| 760 |
} |
| 761 |
EXPORT_SYMBOL_GPL(vfio_pci_core_finish_enable); |
| 762 |
|
| 763 |
static int vfio_pci_get_irq_count(struct vfio_pci_core_device *vdev, int irq_type) |
| 764 |
{ |
| 765 |
if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) { |
| 766 |
return vdev->vconfig[PCI_INTERRUPT_PIN] ? 1 : 0; |
| 767 |
} else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) { |
| 768 |
u8 pos; |
| 769 |
u16 flags; |
| 770 |
|
| 771 |
pos = vdev->pdev->msi_cap; |
| 772 |
if (pos) { |
| 773 |
pci_read_config_word(vdev->pdev, |
| 774 |
pos + PCI_MSI_FLAGS, &flags); |
| 775 |
return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1); |
| 776 |
} |
| 777 |
} else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) { |
| 778 |
u8 pos; |
| 779 |
u16 flags; |
| 780 |
|
| 781 |
pos = vdev->pdev->msix_cap; |
| 782 |
if (pos) { |
| 783 |
pci_read_config_word(vdev->pdev, |
| 784 |
pos + PCI_MSIX_FLAGS, &flags); |
| 785 |
|
| 786 |
return (flags & PCI_MSIX_FLAGS_QSIZE) + 1; |
| 787 |
} |
| 788 |
} else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) { |
| 789 |
if (pci_is_pcie(vdev->pdev)) |
| 790 |
return 1; |
| 791 |
} else if (irq_type == VFIO_PCI_REQ_IRQ_INDEX) { |
| 792 |
return 1; |
| 793 |
} |
| 794 |
|
| 795 |
return 0; |
| 796 |
} |
| 797 |
|
| 798 |
static int vfio_pci_count_devs(struct pci_dev *pdev, void *data) |
| 799 |
{ |
| 800 |
(*(int *)data)++; |
| 801 |
return 0; |
| 802 |
} |
| 803 |
|
| 804 |
struct vfio_pci_fill_info { |
| 805 |
struct vfio_device *vdev; |
| 806 |
struct vfio_pci_dependent_device *devices; |
| 807 |
int nr_devices; |
| 808 |
u32 count; |
| 809 |
u32 flags; |
| 810 |
}; |
| 811 |
|
| 812 |
static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data) |
| 813 |
{ |
| 814 |
struct vfio_pci_dependent_device *info; |
| 815 |
struct vfio_pci_fill_info *fill = data; |
| 816 |
|
| 817 |
/* The topology changed since we counted devices */ |
| 818 |
if (fill->count >= fill->nr_devices) |
| 819 |
return -EAGAIN; |
| 820 |
|
| 821 |
info = &fill->devices[fill->count++]; |
| 822 |
info->segment = pci_domain_nr(pdev->bus); |
| 823 |
info->bus = pdev->bus->number; |
| 824 |
info->devfn = pdev->devfn; |
| 825 |
|
| 826 |
if (fill->flags & VFIO_PCI_HOT_RESET_FLAG_DEV_ID) { |
| 827 |
struct iommufd_ctx *iommufd = vfio_iommufd_device_ictx(fill->vdev); |
| 828 |
struct vfio_device_set *dev_set = fill->vdev->dev_set; |
| 829 |
struct vfio_device *vdev; |
| 830 |
|
| 831 |
/* |
| 832 |
* hot-reset requires all affected devices be represented in |
| 833 |
* the dev_set. |
| 834 |
*/ |
| 835 |
vdev = vfio_find_device_in_devset(dev_set, &pdev->dev); |
| 836 |
if (!vdev) { |
| 837 |
info->devid = VFIO_PCI_DEVID_NOT_OWNED; |
| 838 |
} else { |
| 839 |
int id = vfio_iommufd_get_dev_id(vdev, iommufd); |
| 840 |
|
| 841 |
if (id > 0) |
| 842 |
info->devid = id; |
| 843 |
else if (id == -ENOENT) |
| 844 |
info->devid = VFIO_PCI_DEVID_OWNED; |
| 845 |
else |
| 846 |
info->devid = VFIO_PCI_DEVID_NOT_OWNED; |
| 847 |
} |
| 848 |
/* If devid is VFIO_PCI_DEVID_NOT_OWNED, clear owned flag. */ |
| 849 |
if (info->devid == VFIO_PCI_DEVID_NOT_OWNED) |
| 850 |
fill->flags &= ~VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED; |
| 851 |
} else { |
| 852 |
struct iommu_group *iommu_group; |
| 853 |
|
| 854 |
iommu_group = iommu_group_get(&pdev->dev); |
| 855 |
if (!iommu_group) |
| 856 |
return -EPERM; /* Cannot reset non-isolated devices */ |
| 857 |
|
| 858 |
info->group_id = iommu_group_id(iommu_group); |
| 859 |
iommu_group_put(iommu_group); |
| 860 |
} |
| 861 |
|
| 862 |
return 0; |
| 863 |
} |
| 864 |
|
| 865 |
struct vfio_pci_group_info { |
| 866 |
int count; |
| 867 |
struct file **files; |
| 868 |
}; |
| 869 |
|
| 870 |
static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot) |
| 871 |
{ |
| 872 |
for (; pdev; pdev = pdev->bus->self) |
| 873 |
if (pdev->bus == slot->bus) |
| 874 |
return (pdev->slot == slot); |
| 875 |
return false; |
| 876 |
} |
| 877 |
|
| 878 |
struct vfio_pci_walk_info { |
| 879 |
int (*fn)(struct pci_dev *pdev, void *data); |
| 880 |
void *data; |
| 881 |
struct pci_dev *pdev; |
| 882 |
bool slot; |
| 883 |
int ret; |
| 884 |
}; |
| 885 |
|
| 886 |
static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data) |
| 887 |
{ |
| 888 |
struct vfio_pci_walk_info *walk = data; |
| 889 |
|
| 890 |
if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot)) |
| 891 |
walk->ret = walk->fn(pdev, walk->data); |
| 892 |
|
| 893 |
return walk->ret; |
| 894 |
} |
| 895 |
|
| 896 |
static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev, |
| 897 |
int (*fn)(struct pci_dev *, |
| 898 |
void *data), void *data, |
| 899 |
bool slot) |
| 900 |
{ |
| 901 |
struct vfio_pci_walk_info walk = { |
| 902 |
.fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0, |
| 903 |
}; |
| 904 |
|
| 905 |
pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk); |
| 906 |
|
| 907 |
return walk.ret; |
| 908 |
} |
| 909 |
|
| 910 |
static int msix_mmappable_cap(struct vfio_pci_core_device *vdev, |
| 911 |
struct vfio_info_cap *caps) |
| 912 |
{ |
| 913 |
struct vfio_info_cap_header header = { |
| 914 |
.id = VFIO_REGION_INFO_CAP_MSIX_MAPPABLE, |
| 915 |
.version = 1 |
| 916 |
}; |
| 917 |
|
| 918 |
return vfio_info_add_capability(caps, &header, sizeof(header)); |
| 919 |
} |
| 920 |
|
| 921 |
int vfio_pci_core_register_dev_region(struct vfio_pci_core_device *vdev, |
| 922 |
unsigned int type, unsigned int subtype, |
| 923 |
const struct vfio_pci_regops *ops, |
| 924 |
size_t size, u32 flags, void *data) |
| 925 |
{ |
| 926 |
struct vfio_pci_region *region; |
| 927 |
|
| 928 |
region = krealloc(vdev->region, |
| 929 |
(vdev->num_regions + 1) * sizeof(*region), |
| 930 |
GFP_KERNEL_ACCOUNT); |
| 931 |
if (!region) |
| 932 |
return -ENOMEM; |
| 933 |
|
| 934 |
vdev->region = region; |
| 935 |
vdev->region[vdev->num_regions].type = type; |
| 936 |
vdev->region[vdev->num_regions].subtype = subtype; |
| 937 |
vdev->region[vdev->num_regions].ops = ops; |
| 938 |
vdev->region[vdev->num_regions].size = size; |
| 939 |
vdev->region[vdev->num_regions].flags = flags; |
| 940 |
vdev->region[vdev->num_regions].data = data; |
| 941 |
|
| 942 |
vdev->num_regions++; |
| 943 |
|
| 944 |
return 0; |
| 945 |
} |
| 946 |
EXPORT_SYMBOL_GPL(vfio_pci_core_register_dev_region); |
| 947 |
|
| 948 |
static int vfio_pci_info_atomic_cap(struct vfio_pci_core_device *vdev, |
| 949 |
struct vfio_info_cap *caps) |
| 950 |
{ |
| 951 |
struct vfio_device_info_cap_pci_atomic_comp cap = { |
| 952 |
.header.id = VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP, |
| 953 |
.header.version = 1 |
| 954 |
}; |
| 955 |
struct pci_dev *pdev = pci_physfn(vdev->pdev); |
| 956 |
u32 devcap2; |
| 957 |
|
| 958 |
pcie_capability_read_dword(pdev, PCI_EXP_DEVCAP2, &devcap2); |
| 959 |
|
| 960 |
if ((devcap2 & PCI_EXP_DEVCAP2_ATOMIC_COMP32) && |
| 961 |
!pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32)) |
| 962 |
cap.flags |= VFIO_PCI_ATOMIC_COMP32; |
| 963 |
|
| 964 |
if ((devcap2 & PCI_EXP_DEVCAP2_ATOMIC_COMP64) && |
| 965 |
!pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64)) |
| 966 |
cap.flags |= VFIO_PCI_ATOMIC_COMP64; |
| 967 |
|
| 968 |
if ((devcap2 & PCI_EXP_DEVCAP2_ATOMIC_COMP128) && |
| 969 |
!pci_enable_atomic_ops_to_root(pdev, |
| 970 |
PCI_EXP_DEVCAP2_ATOMIC_COMP128)) |
| 971 |
cap.flags |= VFIO_PCI_ATOMIC_COMP128; |
| 972 |
|
| 973 |
if (!cap.flags) |
| 974 |
return -ENODEV; |
| 975 |
|
| 976 |
return vfio_info_add_capability(caps, &cap.header, sizeof(cap)); |
| 977 |
} |
| 978 |
|
| 979 |
static int vfio_pci_ioctl_get_info(struct vfio_pci_core_device *vdev, |
| 980 |
struct vfio_device_info __user *arg) |
| 981 |
{ |
| 982 |
unsigned long minsz = offsetofend(struct vfio_device_info, num_irqs); |
| 983 |
struct vfio_device_info info = {}; |
| 984 |
struct vfio_info_cap caps = { .buf = NULL, .size = 0 }; |
| 985 |
int ret; |
| 986 |
|
| 987 |
if (copy_from_user(&info, arg, minsz)) |
| 988 |
return -EFAULT; |
| 989 |
|
| 990 |
if (info.argsz < minsz) |
| 991 |
return -EINVAL; |
| 992 |
|
| 993 |
minsz = min_t(size_t, info.argsz, sizeof(info)); |
| 994 |
|
| 995 |
info.flags = VFIO_DEVICE_FLAGS_PCI; |
| 996 |
|
| 997 |
if (vdev->reset_works) |
| 998 |
info.flags |= VFIO_DEVICE_FLAGS_RESET; |
| 999 |
|
| 1000 |
info.num_regions = VFIO_PCI_NUM_REGIONS + vdev->num_regions; |
| 1001 |
info.num_irqs = VFIO_PCI_NUM_IRQS; |
| 1002 |
|
| 1003 |
ret = vfio_pci_info_zdev_add_caps(vdev, &caps); |
| 1004 |
if (ret && ret != -ENODEV) { |
| 1005 |
pci_warn(vdev->pdev, |
| 1006 |
"Failed to setup zPCI info capabilities\n"); |
| 1007 |
return ret; |
| 1008 |
} |
| 1009 |
|
| 1010 |
ret = vfio_pci_info_atomic_cap(vdev, &caps); |
| 1011 |
if (ret && ret != -ENODEV) { |
| 1012 |
pci_warn(vdev->pdev, |
| 1013 |
"Failed to setup AtomicOps info capability\n"); |
| 1014 |
return ret; |
| 1015 |
} |
| 1016 |
|
| 1017 |
if (caps.size) { |
| 1018 |
info.flags |= VFIO_DEVICE_FLAGS_CAPS; |
| 1019 |
if (info.argsz < sizeof(info) + caps.size) { |
| 1020 |
info.argsz = sizeof(info) + caps.size; |
| 1021 |
} else { |
| 1022 |
vfio_info_cap_shift(&caps, sizeof(info)); |
| 1023 |
if (copy_to_user(arg + 1, caps.buf, caps.size)) { |
| 1024 |
kfree(caps.buf); |
| 1025 |
return -EFAULT; |
| 1026 |
} |
| 1027 |
info.cap_offset = sizeof(*arg); |
| 1028 |
} |
| 1029 |
|
| 1030 |
kfree(caps.buf); |
| 1031 |
} |
| 1032 |
|
| 1033 |
return copy_to_user(arg, &info, minsz) ? -EFAULT : 0; |
| 1034 |
} |
| 1035 |
|
| 1036 |
int vfio_pci_ioctl_get_region_info(struct vfio_device *core_vdev, |
| 1037 |
struct vfio_region_info *info, |
| 1038 |
struct vfio_info_cap *caps) |
| 1039 |
{ |
| 1040 |
struct vfio_pci_core_device *vdev = |
| 1041 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1042 |
struct pci_dev *pdev = vdev->pdev; |
| 1043 |
int i, ret; |
| 1044 |
|
| 1045 |
switch (info->index) { |
| 1046 |
case VFIO_PCI_CONFIG_REGION_INDEX: |
| 1047 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 1048 |
info->size = pdev->cfg_size; |
| 1049 |
info->flags = VFIO_REGION_INFO_FLAG_READ | |
| 1050 |
VFIO_REGION_INFO_FLAG_WRITE; |
| 1051 |
break; |
| 1052 |
case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: |
| 1053 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 1054 |
info->size = pci_resource_len(pdev, info->index); |
| 1055 |
if (!info->size) { |
| 1056 |
info->flags = 0; |
| 1057 |
break; |
| 1058 |
} |
| 1059 |
|
| 1060 |
info->flags = VFIO_REGION_INFO_FLAG_READ | |
| 1061 |
VFIO_REGION_INFO_FLAG_WRITE; |
| 1062 |
if (vdev->bar_mmap_supported[info->index]) { |
| 1063 |
info->flags |= VFIO_REGION_INFO_FLAG_MMAP; |
| 1064 |
if (info->index == vdev->msix_bar) { |
| 1065 |
ret = msix_mmappable_cap(vdev, caps); |
| 1066 |
if (ret) |
| 1067 |
return ret; |
| 1068 |
} |
| 1069 |
} |
| 1070 |
|
| 1071 |
break; |
| 1072 |
case VFIO_PCI_ROM_REGION_INDEX: { |
| 1073 |
void __iomem *io; |
| 1074 |
size_t size; |
| 1075 |
u16 cmd; |
| 1076 |
|
| 1077 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 1078 |
info->flags = 0; |
| 1079 |
info->size = 0; |
| 1080 |
|
| 1081 |
if (pci_resource_start(pdev, PCI_ROM_RESOURCE)) { |
| 1082 |
/* |
| 1083 |
* Check ROM content is valid. Need to enable memory |
| 1084 |
* decode for ROM access in pci_map_rom(). |
| 1085 |
*/ |
| 1086 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 1087 |
io = pci_map_rom(pdev, &size); |
| 1088 |
if (io) { |
| 1089 |
info->flags = VFIO_REGION_INFO_FLAG_READ; |
| 1090 |
/* Report the BAR size, not the ROM size. */ |
| 1091 |
info->size = pci_resource_len(pdev, |
| 1092 |
PCI_ROM_RESOURCE); |
| 1093 |
pci_unmap_rom(pdev, io); |
| 1094 |
} |
| 1095 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 1096 |
} else if (pdev->rom && pdev->romlen) { |
| 1097 |
info->flags = VFIO_REGION_INFO_FLAG_READ; |
| 1098 |
/* Report BAR size as power of two. */ |
| 1099 |
info->size = roundup_pow_of_two(pdev->romlen); |
| 1100 |
} |
| 1101 |
|
| 1102 |
break; |
| 1103 |
} |
| 1104 |
case VFIO_PCI_VGA_REGION_INDEX: |
| 1105 |
if (!vdev->has_vga) |
| 1106 |
return -EINVAL; |
| 1107 |
|
| 1108 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 1109 |
info->size = 0xc0000; |
| 1110 |
info->flags = VFIO_REGION_INFO_FLAG_READ | |
| 1111 |
VFIO_REGION_INFO_FLAG_WRITE; |
| 1112 |
|
| 1113 |
break; |
| 1114 |
default: { |
| 1115 |
struct vfio_region_info_cap_type cap_type = { |
| 1116 |
.header.id = VFIO_REGION_INFO_CAP_TYPE, |
| 1117 |
.header.version = 1 |
| 1118 |
}; |
| 1119 |
|
| 1120 |
if (info->index >= VFIO_PCI_NUM_REGIONS + vdev->num_regions) |
| 1121 |
return -EINVAL; |
| 1122 |
info->index = array_index_nospec( |
| 1123 |
info->index, VFIO_PCI_NUM_REGIONS + vdev->num_regions); |
| 1124 |
|
| 1125 |
i = info->index - VFIO_PCI_NUM_REGIONS; |
| 1126 |
|
| 1127 |
info->offset = VFIO_PCI_INDEX_TO_OFFSET(info->index); |
| 1128 |
info->size = vdev->region[i].size; |
| 1129 |
info->flags = vdev->region[i].flags; |
| 1130 |
|
| 1131 |
cap_type.type = vdev->region[i].type; |
| 1132 |
cap_type.subtype = vdev->region[i].subtype; |
| 1133 |
|
| 1134 |
ret = vfio_info_add_capability(caps, &cap_type.header, |
| 1135 |
sizeof(cap_type)); |
| 1136 |
if (ret) |
| 1137 |
return ret; |
| 1138 |
|
| 1139 |
if (vdev->region[i].ops->add_capability) { |
| 1140 |
ret = vdev->region[i].ops->add_capability( |
| 1141 |
vdev, &vdev->region[i], caps); |
| 1142 |
if (ret) |
| 1143 |
return ret; |
| 1144 |
} |
| 1145 |
} |
| 1146 |
} |
| 1147 |
return 0; |
| 1148 |
} |
| 1149 |
EXPORT_SYMBOL_GPL(vfio_pci_ioctl_get_region_info); |
| 1150 |
|
| 1151 |
static int vfio_pci_ioctl_get_irq_info(struct vfio_pci_core_device *vdev, |
| 1152 |
struct vfio_irq_info __user *arg) |
| 1153 |
{ |
| 1154 |
unsigned long minsz = offsetofend(struct vfio_irq_info, count); |
| 1155 |
struct vfio_irq_info info; |
| 1156 |
|
| 1157 |
if (copy_from_user(&info, arg, minsz)) |
| 1158 |
return -EFAULT; |
| 1159 |
|
| 1160 |
if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS) |
| 1161 |
return -EINVAL; |
| 1162 |
|
| 1163 |
switch (info.index) { |
| 1164 |
case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX: |
| 1165 |
case VFIO_PCI_REQ_IRQ_INDEX: |
| 1166 |
break; |
| 1167 |
case VFIO_PCI_ERR_IRQ_INDEX: |
| 1168 |
if (pci_is_pcie(vdev->pdev)) |
| 1169 |
break; |
| 1170 |
fallthrough; |
| 1171 |
default: |
| 1172 |
return -EINVAL; |
| 1173 |
} |
| 1174 |
|
| 1175 |
info.flags = VFIO_IRQ_INFO_EVENTFD; |
| 1176 |
|
| 1177 |
info.count = vfio_pci_get_irq_count(vdev, info.index); |
| 1178 |
|
| 1179 |
if (info.index == VFIO_PCI_INTX_IRQ_INDEX) |
| 1180 |
info.flags |= |
| 1181 |
(VFIO_IRQ_INFO_MASKABLE | VFIO_IRQ_INFO_AUTOMASKED); |
| 1182 |
else if (info.index != VFIO_PCI_MSIX_IRQ_INDEX || !vdev->has_dyn_msix) |
| 1183 |
info.flags |= VFIO_IRQ_INFO_NORESIZE; |
| 1184 |
|
| 1185 |
return copy_to_user(arg, &info, minsz) ? -EFAULT : 0; |
| 1186 |
} |
| 1187 |
|
| 1188 |
static int vfio_pci_ioctl_set_irqs(struct vfio_pci_core_device *vdev, |
| 1189 |
struct vfio_irq_set __user *arg) |
| 1190 |
{ |
| 1191 |
unsigned long minsz = offsetofend(struct vfio_irq_set, count); |
| 1192 |
struct vfio_irq_set hdr; |
| 1193 |
u8 *data = NULL; |
| 1194 |
int max, ret = 0; |
| 1195 |
size_t data_size = 0; |
| 1196 |
|
| 1197 |
if (copy_from_user(&hdr, arg, minsz)) |
| 1198 |
return -EFAULT; |
| 1199 |
|
| 1200 |
max = vfio_pci_get_irq_count(vdev, hdr.index); |
| 1201 |
|
| 1202 |
ret = vfio_set_irqs_validate_and_prepare(&hdr, max, VFIO_PCI_NUM_IRQS, |
| 1203 |
&data_size); |
| 1204 |
if (ret) |
| 1205 |
return ret; |
| 1206 |
|
| 1207 |
if (data_size) { |
| 1208 |
data = memdup_user(&arg->data, data_size); |
| 1209 |
if (IS_ERR(data)) |
| 1210 |
return PTR_ERR(data); |
| 1211 |
} |
| 1212 |
|
| 1213 |
mutex_lock(&vdev->igate); |
| 1214 |
|
| 1215 |
ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index, hdr.start, |
| 1216 |
hdr.count, data); |
| 1217 |
|
| 1218 |
mutex_unlock(&vdev->igate); |
| 1219 |
kfree(data); |
| 1220 |
|
| 1221 |
return ret; |
| 1222 |
} |
| 1223 |
|
| 1224 |
static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev, |
| 1225 |
void __user *arg) |
| 1226 |
{ |
| 1227 |
int ret; |
| 1228 |
|
| 1229 |
if (!vdev->reset_works) |
| 1230 |
return -EINVAL; |
| 1231 |
|
| 1232 |
vfio_pci_zap_and_down_write_memory_lock(vdev); |
| 1233 |
|
| 1234 |
/* |
| 1235 |
* This function can be invoked while the power state is non-D0. If |
| 1236 |
* pci_try_reset_function() has been called while the power state is |
| 1237 |
* non-D0, then pci_try_reset_function() will internally set the power |
| 1238 |
* state to D0 without vfio driver involvement. For the devices which |
| 1239 |
* have NoSoftRst-, the reset function can cause the PCI config space |
| 1240 |
* reset without restoring the original state (saved locally in |
| 1241 |
* 'vdev->pm_save'). |
| 1242 |
*/ |
| 1243 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 1244 |
|
| 1245 |
vfio_pci_dma_buf_move(vdev, true); |
| 1246 |
ret = pci_try_reset_function(vdev->pdev); |
| 1247 |
if (__vfio_pci_memory_enabled(vdev)) |
| 1248 |
vfio_pci_dma_buf_move(vdev, false); |
| 1249 |
up_write(&vdev->memory_lock); |
| 1250 |
|
| 1251 |
return ret; |
| 1252 |
} |
| 1253 |
|
| 1254 |
static int vfio_pci_ioctl_get_pci_hot_reset_info( |
| 1255 |
struct vfio_pci_core_device *vdev, |
| 1256 |
struct vfio_pci_hot_reset_info __user *arg) |
| 1257 |
{ |
| 1258 |
unsigned long minsz = |
| 1259 |
offsetofend(struct vfio_pci_hot_reset_info, count); |
| 1260 |
struct vfio_pci_dependent_device *devices = NULL; |
| 1261 |
struct vfio_pci_hot_reset_info hdr; |
| 1262 |
struct vfio_pci_fill_info fill = {}; |
| 1263 |
bool slot = false; |
| 1264 |
int ret, count = 0; |
| 1265 |
|
| 1266 |
if (copy_from_user(&hdr, arg, minsz)) |
| 1267 |
return -EFAULT; |
| 1268 |
|
| 1269 |
if (hdr.argsz < minsz) |
| 1270 |
return -EINVAL; |
| 1271 |
|
| 1272 |
hdr.flags = 0; |
| 1273 |
|
| 1274 |
/* Can we do a slot or bus reset or neither? */ |
| 1275 |
if (!pci_probe_reset_slot(vdev->pdev->slot)) |
| 1276 |
slot = true; |
| 1277 |
else if (pci_probe_reset_bus(vdev->pdev->bus)) |
| 1278 |
return -ENODEV; |
| 1279 |
|
| 1280 |
ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs, |
| 1281 |
&count, slot); |
| 1282 |
if (ret) |
| 1283 |
return ret; |
| 1284 |
|
| 1285 |
if (WARN_ON(!count)) /* Should always be at least one */ |
| 1286 |
return -ERANGE; |
| 1287 |
|
| 1288 |
if (count > (hdr.argsz - sizeof(hdr)) / sizeof(*devices)) { |
| 1289 |
hdr.count = count; |
| 1290 |
ret = -ENOSPC; |
| 1291 |
goto header; |
| 1292 |
} |
| 1293 |
|
| 1294 |
devices = kzalloc_objs(*devices, count); |
| 1295 |
if (!devices) |
| 1296 |
return -ENOMEM; |
| 1297 |
|
| 1298 |
fill.devices = devices; |
| 1299 |
fill.nr_devices = count; |
| 1300 |
fill.vdev = &vdev->vdev; |
| 1301 |
|
| 1302 |
if (vfio_device_cdev_opened(&vdev->vdev)) |
| 1303 |
fill.flags |= VFIO_PCI_HOT_RESET_FLAG_DEV_ID | |
| 1304 |
VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED; |
| 1305 |
|
| 1306 |
mutex_lock(&vdev->vdev.dev_set->lock); |
| 1307 |
ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_fill_devs, |
| 1308 |
&fill, slot); |
| 1309 |
mutex_unlock(&vdev->vdev.dev_set->lock); |
| 1310 |
if (ret) |
| 1311 |
goto out; |
| 1312 |
|
| 1313 |
if (copy_to_user(arg->devices, devices, |
| 1314 |
sizeof(*devices) * fill.count)) { |
| 1315 |
ret = -EFAULT; |
| 1316 |
goto out; |
| 1317 |
} |
| 1318 |
|
| 1319 |
hdr.count = fill.count; |
| 1320 |
hdr.flags = fill.flags; |
| 1321 |
|
| 1322 |
header: |
| 1323 |
if (copy_to_user(arg, &hdr, minsz)) |
| 1324 |
ret = -EFAULT; |
| 1325 |
out: |
| 1326 |
kfree(devices); |
| 1327 |
return ret; |
| 1328 |
} |
| 1329 |
|
| 1330 |
static int |
| 1331 |
vfio_pci_ioctl_pci_hot_reset_groups(struct vfio_pci_core_device *vdev, |
| 1332 |
u32 array_count, bool slot, |
| 1333 |
struct vfio_pci_hot_reset __user *arg) |
| 1334 |
{ |
| 1335 |
int32_t *group_fds; |
| 1336 |
struct file **files; |
| 1337 |
struct vfio_pci_group_info info; |
| 1338 |
int file_idx, count = 0, ret = 0; |
| 1339 |
|
| 1340 |
/* |
| 1341 |
* We can't let userspace give us an arbitrarily large buffer to copy, |
| 1342 |
* so verify how many we think there could be. Note groups can have |
| 1343 |
* multiple devices so one group per device is the max. |
| 1344 |
*/ |
| 1345 |
ret = vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs, |
| 1346 |
&count, slot); |
| 1347 |
if (ret) |
| 1348 |
return ret; |
| 1349 |
|
| 1350 |
if (array_count > count) |
| 1351 |
return -EINVAL; |
| 1352 |
|
| 1353 |
group_fds = kzalloc_objs(*group_fds, array_count); |
| 1354 |
files = kzalloc_objs(*files, array_count); |
| 1355 |
if (!group_fds || !files) { |
| 1356 |
kfree(group_fds); |
| 1357 |
kfree(files); |
| 1358 |
return -ENOMEM; |
| 1359 |
} |
| 1360 |
|
| 1361 |
if (copy_from_user(group_fds, arg->group_fds, |
| 1362 |
array_count * sizeof(*group_fds))) { |
| 1363 |
kfree(group_fds); |
| 1364 |
kfree(files); |
| 1365 |
return -EFAULT; |
| 1366 |
} |
| 1367 |
|
| 1368 |
/* |
| 1369 |
* Get the group file for each fd to ensure the group is held across |
| 1370 |
* the reset |
| 1371 |
*/ |
| 1372 |
for (file_idx = 0; file_idx < array_count; file_idx++) { |
| 1373 |
struct file *file = fget(group_fds[file_idx]); |
| 1374 |
|
| 1375 |
if (!file) { |
| 1376 |
ret = -EBADF; |
| 1377 |
break; |
| 1378 |
} |
| 1379 |
|
| 1380 |
/* Ensure the FD is a vfio group FD.*/ |
| 1381 |
if (!vfio_file_is_group(file)) { |
| 1382 |
fput(file); |
| 1383 |
ret = -EINVAL; |
| 1384 |
break; |
| 1385 |
} |
| 1386 |
|
| 1387 |
files[file_idx] = file; |
| 1388 |
} |
| 1389 |
|
| 1390 |
kfree(group_fds); |
| 1391 |
|
| 1392 |
/* release reference to groups on error */ |
| 1393 |
if (ret) |
| 1394 |
goto hot_reset_release; |
| 1395 |
|
| 1396 |
info.count = array_count; |
| 1397 |
info.files = files; |
| 1398 |
|
| 1399 |
ret = vfio_pci_dev_set_hot_reset(vdev->vdev.dev_set, &info, NULL); |
| 1400 |
|
| 1401 |
hot_reset_release: |
| 1402 |
for (file_idx--; file_idx >= 0; file_idx--) |
| 1403 |
fput(files[file_idx]); |
| 1404 |
|
| 1405 |
kfree(files); |
| 1406 |
return ret; |
| 1407 |
} |
| 1408 |
|
| 1409 |
static int vfio_pci_ioctl_pci_hot_reset(struct vfio_pci_core_device *vdev, |
| 1410 |
struct vfio_pci_hot_reset __user *arg) |
| 1411 |
{ |
| 1412 |
unsigned long minsz = offsetofend(struct vfio_pci_hot_reset, count); |
| 1413 |
struct vfio_pci_hot_reset hdr; |
| 1414 |
bool slot = false; |
| 1415 |
|
| 1416 |
if (copy_from_user(&hdr, arg, minsz)) |
| 1417 |
return -EFAULT; |
| 1418 |
|
| 1419 |
if (hdr.argsz < minsz || hdr.flags) |
| 1420 |
return -EINVAL; |
| 1421 |
|
| 1422 |
/* zero-length array is only for cdev opened devices */ |
| 1423 |
if (!!hdr.count == vfio_device_cdev_opened(&vdev->vdev)) |
| 1424 |
return -EINVAL; |
| 1425 |
|
| 1426 |
/* Can we do a slot or bus reset or neither? */ |
| 1427 |
if (!pci_probe_reset_slot(vdev->pdev->slot)) |
| 1428 |
slot = true; |
| 1429 |
else if (pci_probe_reset_bus(vdev->pdev->bus)) |
| 1430 |
return -ENODEV; |
| 1431 |
|
| 1432 |
if (hdr.count) |
| 1433 |
return vfio_pci_ioctl_pci_hot_reset_groups(vdev, hdr.count, slot, arg); |
| 1434 |
|
| 1435 |
return vfio_pci_dev_set_hot_reset(vdev->vdev.dev_set, NULL, |
| 1436 |
vfio_iommufd_device_ictx(&vdev->vdev)); |
| 1437 |
} |
| 1438 |
|
| 1439 |
static int vfio_pci_ioctl_ioeventfd(struct vfio_pci_core_device *vdev, |
| 1440 |
struct vfio_device_ioeventfd __user *arg) |
| 1441 |
{ |
| 1442 |
unsigned long minsz = offsetofend(struct vfio_device_ioeventfd, fd); |
| 1443 |
struct vfio_device_ioeventfd ioeventfd; |
| 1444 |
int count; |
| 1445 |
|
| 1446 |
if (copy_from_user(&ioeventfd, arg, minsz)) |
| 1447 |
return -EFAULT; |
| 1448 |
|
| 1449 |
if (ioeventfd.argsz < minsz) |
| 1450 |
return -EINVAL; |
| 1451 |
|
| 1452 |
if (ioeventfd.flags & ~VFIO_DEVICE_IOEVENTFD_SIZE_MASK) |
| 1453 |
return -EINVAL; |
| 1454 |
|
| 1455 |
count = ioeventfd.flags & VFIO_DEVICE_IOEVENTFD_SIZE_MASK; |
| 1456 |
|
| 1457 |
if (hweight8(count) != 1 || ioeventfd.fd < -1) |
| 1458 |
return -EINVAL; |
| 1459 |
|
| 1460 |
return vfio_pci_ioeventfd(vdev, ioeventfd.offset, ioeventfd.data, count, |
| 1461 |
ioeventfd.fd); |
| 1462 |
} |
| 1463 |
|
| 1464 |
long vfio_pci_core_ioctl(struct vfio_device *core_vdev, unsigned int cmd, |
| 1465 |
unsigned long arg) |
| 1466 |
{ |
| 1467 |
struct vfio_pci_core_device *vdev = |
| 1468 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1469 |
void __user *uarg = (void __user *)arg; |
| 1470 |
|
| 1471 |
switch (cmd) { |
| 1472 |
case VFIO_DEVICE_GET_INFO: |
| 1473 |
return vfio_pci_ioctl_get_info(vdev, uarg); |
| 1474 |
case VFIO_DEVICE_GET_IRQ_INFO: |
| 1475 |
return vfio_pci_ioctl_get_irq_info(vdev, uarg); |
| 1476 |
case VFIO_DEVICE_GET_PCI_HOT_RESET_INFO: |
| 1477 |
return vfio_pci_ioctl_get_pci_hot_reset_info(vdev, uarg); |
| 1478 |
case VFIO_DEVICE_IOEVENTFD: |
| 1479 |
return vfio_pci_ioctl_ioeventfd(vdev, uarg); |
| 1480 |
case VFIO_DEVICE_PCI_HOT_RESET: |
| 1481 |
return vfio_pci_ioctl_pci_hot_reset(vdev, uarg); |
| 1482 |
case VFIO_DEVICE_RESET: |
| 1483 |
return vfio_pci_ioctl_reset(vdev, uarg); |
| 1484 |
case VFIO_DEVICE_SET_IRQS: |
| 1485 |
return vfio_pci_ioctl_set_irqs(vdev, uarg); |
| 1486 |
default: |
| 1487 |
return -ENOTTY; |
| 1488 |
} |
| 1489 |
} |
| 1490 |
EXPORT_SYMBOL_GPL(vfio_pci_core_ioctl); |
| 1491 |
|
| 1492 |
static int vfio_pci_core_feature_token(struct vfio_pci_core_device *vdev, |
| 1493 |
u32 flags, uuid_t __user *arg, |
| 1494 |
size_t argsz) |
| 1495 |
{ |
| 1496 |
uuid_t uuid; |
| 1497 |
int ret; |
| 1498 |
|
| 1499 |
if (!vdev->vf_token) |
| 1500 |
return -ENOTTY; |
| 1501 |
/* |
| 1502 |
* We do not support GET of the VF Token UUID as this could |
| 1503 |
* expose the token of the previous device user. |
| 1504 |
*/ |
| 1505 |
ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_SET, |
| 1506 |
sizeof(uuid)); |
| 1507 |
if (ret != 1) |
| 1508 |
return ret; |
| 1509 |
|
| 1510 |
if (copy_from_user(&uuid, arg, sizeof(uuid))) |
| 1511 |
return -EFAULT; |
| 1512 |
|
| 1513 |
mutex_lock(&vdev->vf_token->lock); |
| 1514 |
uuid_copy(&vdev->vf_token->uuid, &uuid); |
| 1515 |
mutex_unlock(&vdev->vf_token->lock); |
| 1516 |
return 0; |
| 1517 |
} |
| 1518 |
|
| 1519 |
int vfio_pci_core_ioctl_feature(struct vfio_device *device, u32 flags, |
| 1520 |
void __user *arg, size_t argsz) |
| 1521 |
{ |
| 1522 |
struct vfio_pci_core_device *vdev = |
| 1523 |
container_of(device, struct vfio_pci_core_device, vdev); |
| 1524 |
|
| 1525 |
switch (flags & VFIO_DEVICE_FEATURE_MASK) { |
| 1526 |
case VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY: |
| 1527 |
return vfio_pci_core_pm_entry(vdev, flags, arg, argsz); |
| 1528 |
case VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP: |
| 1529 |
return vfio_pci_core_pm_entry_with_wakeup(vdev, flags, |
| 1530 |
arg, argsz); |
| 1531 |
case VFIO_DEVICE_FEATURE_LOW_POWER_EXIT: |
| 1532 |
return vfio_pci_core_pm_exit(vdev, flags, arg, argsz); |
| 1533 |
case VFIO_DEVICE_FEATURE_PCI_VF_TOKEN: |
| 1534 |
return vfio_pci_core_feature_token(vdev, flags, arg, argsz); |
| 1535 |
case VFIO_DEVICE_FEATURE_DMA_BUF: |
| 1536 |
return vfio_pci_core_feature_dma_buf(vdev, flags, arg, argsz); |
| 1537 |
default: |
| 1538 |
return -ENOTTY; |
| 1539 |
} |
| 1540 |
} |
| 1541 |
EXPORT_SYMBOL_GPL(vfio_pci_core_ioctl_feature); |
| 1542 |
|
| 1543 |
static ssize_t vfio_pci_rw(struct vfio_pci_core_device *vdev, char __user *buf, |
| 1544 |
size_t count, loff_t *ppos, bool iswrite) |
| 1545 |
{ |
| 1546 |
unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); |
| 1547 |
int ret; |
| 1548 |
|
| 1549 |
if (index >= VFIO_PCI_NUM_REGIONS + vdev->num_regions) |
| 1550 |
return -EINVAL; |
| 1551 |
|
| 1552 |
ret = pm_runtime_resume_and_get(&vdev->pdev->dev); |
| 1553 |
if (ret) { |
| 1554 |
pci_info_ratelimited(vdev->pdev, "runtime resume failed %d\n", |
| 1555 |
ret); |
| 1556 |
return -EIO; |
| 1557 |
} |
| 1558 |
|
| 1559 |
switch (index) { |
| 1560 |
case VFIO_PCI_CONFIG_REGION_INDEX: |
| 1561 |
ret = vfio_pci_config_rw(vdev, buf, count, ppos, iswrite); |
| 1562 |
break; |
| 1563 |
|
| 1564 |
case VFIO_PCI_ROM_REGION_INDEX: |
| 1565 |
if (iswrite) |
| 1566 |
ret = -EINVAL; |
| 1567 |
else |
| 1568 |
ret = vfio_pci_bar_rw(vdev, buf, count, ppos, false); |
| 1569 |
break; |
| 1570 |
|
| 1571 |
case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: |
| 1572 |
ret = vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite); |
| 1573 |
break; |
| 1574 |
|
| 1575 |
case VFIO_PCI_VGA_REGION_INDEX: |
| 1576 |
ret = vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite); |
| 1577 |
break; |
| 1578 |
|
| 1579 |
default: |
| 1580 |
index -= VFIO_PCI_NUM_REGIONS; |
| 1581 |
ret = vdev->region[index].ops->rw(vdev, buf, |
| 1582 |
count, ppos, iswrite); |
| 1583 |
break; |
| 1584 |
} |
| 1585 |
|
| 1586 |
pm_runtime_put(&vdev->pdev->dev); |
| 1587 |
return ret; |
| 1588 |
} |
| 1589 |
|
| 1590 |
ssize_t vfio_pci_core_read(struct vfio_device *core_vdev, char __user *buf, |
| 1591 |
size_t count, loff_t *ppos) |
| 1592 |
{ |
| 1593 |
struct vfio_pci_core_device *vdev = |
| 1594 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1595 |
|
| 1596 |
if (!count) |
| 1597 |
return 0; |
| 1598 |
|
| 1599 |
return vfio_pci_rw(vdev, buf, count, ppos, false); |
| 1600 |
} |
| 1601 |
EXPORT_SYMBOL_GPL(vfio_pci_core_read); |
| 1602 |
|
| 1603 |
ssize_t vfio_pci_core_write(struct vfio_device *core_vdev, const char __user *buf, |
| 1604 |
size_t count, loff_t *ppos) |
| 1605 |
{ |
| 1606 |
struct vfio_pci_core_device *vdev = |
| 1607 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1608 |
|
| 1609 |
if (!count) |
| 1610 |
return 0; |
| 1611 |
|
| 1612 |
return vfio_pci_rw(vdev, (char __user *)buf, count, ppos, true); |
| 1613 |
} |
| 1614 |
EXPORT_SYMBOL_GPL(vfio_pci_core_write); |
| 1615 |
|
| 1616 |
static void vfio_pci_zap_bars(struct vfio_pci_core_device *vdev) |
| 1617 |
{ |
| 1618 |
struct vfio_device *core_vdev = &vdev->vdev; |
| 1619 |
loff_t start = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_BAR0_REGION_INDEX); |
| 1620 |
loff_t end = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_ROM_REGION_INDEX); |
| 1621 |
loff_t len = end - start; |
| 1622 |
|
| 1623 |
unmap_mapping_range(core_vdev->inode->i_mapping, start, len, true); |
| 1624 |
} |
| 1625 |
|
| 1626 |
void vfio_pci_zap_and_down_write_memory_lock(struct vfio_pci_core_device *vdev) |
| 1627 |
{ |
| 1628 |
down_write(&vdev->memory_lock); |
| 1629 |
vfio_pci_zap_bars(vdev); |
| 1630 |
} |
| 1631 |
|
| 1632 |
u16 vfio_pci_memory_lock_and_enable(struct vfio_pci_core_device *vdev) |
| 1633 |
{ |
| 1634 |
u16 cmd; |
| 1635 |
|
| 1636 |
down_write(&vdev->memory_lock); |
| 1637 |
pci_read_config_word(vdev->pdev, PCI_COMMAND, &cmd); |
| 1638 |
if (!(cmd & PCI_COMMAND_MEMORY)) |
| 1639 |
pci_write_config_word(vdev->pdev, PCI_COMMAND, |
| 1640 |
cmd | PCI_COMMAND_MEMORY); |
| 1641 |
|
| 1642 |
return cmd; |
| 1643 |
} |
| 1644 |
|
| 1645 |
void vfio_pci_memory_unlock_and_restore(struct vfio_pci_core_device *vdev, u16 cmd) |
| 1646 |
{ |
| 1647 |
pci_write_config_word(vdev->pdev, PCI_COMMAND, cmd); |
| 1648 |
up_write(&vdev->memory_lock); |
| 1649 |
} |
| 1650 |
|
| 1651 |
static unsigned long vma_to_pfn(struct vm_area_struct *vma) |
| 1652 |
{ |
| 1653 |
struct vfio_pci_core_device *vdev = vma->vm_private_data; |
| 1654 |
int index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); |
| 1655 |
u64 pgoff; |
| 1656 |
|
| 1657 |
pgoff = vma->vm_pgoff & |
| 1658 |
((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); |
| 1659 |
|
| 1660 |
return (pci_resource_start(vdev->pdev, index) >> PAGE_SHIFT) + pgoff; |
| 1661 |
} |
| 1662 |
|
| 1663 |
vm_fault_t vfio_pci_vmf_insert_pfn(struct vfio_pci_core_device *vdev, |
| 1664 |
struct vm_fault *vmf, |
| 1665 |
unsigned long pfn, |
| 1666 |
unsigned int order) |
| 1667 |
{ |
| 1668 |
lockdep_assert_held_read(&vdev->memory_lock); |
| 1669 |
|
| 1670 |
if (vdev->pm_runtime_engaged || !__vfio_pci_memory_enabled(vdev)) |
| 1671 |
return VM_FAULT_SIGBUS; |
| 1672 |
|
| 1673 |
switch (order) { |
| 1674 |
case 0: |
| 1675 |
return vmf_insert_pfn(vmf->vma, vmf->address, pfn); |
| 1676 |
#ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP |
| 1677 |
case PMD_ORDER: |
| 1678 |
return vmf_insert_pfn_pmd(vmf, pfn, false); |
| 1679 |
#endif |
| 1680 |
#ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP |
| 1681 |
case PUD_ORDER: |
| 1682 |
return vmf_insert_pfn_pud(vmf, pfn, false); |
| 1683 |
break; |
| 1684 |
#endif |
| 1685 |
default: |
| 1686 |
return VM_FAULT_FALLBACK; |
| 1687 |
} |
| 1688 |
} |
| 1689 |
EXPORT_SYMBOL_GPL(vfio_pci_vmf_insert_pfn); |
| 1690 |
|
| 1691 |
static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf, |
| 1692 |
unsigned int order) |
| 1693 |
{ |
| 1694 |
struct vm_area_struct *vma = vmf->vma; |
| 1695 |
struct vfio_pci_core_device *vdev = vma->vm_private_data; |
| 1696 |
unsigned long addr = vmf->address & ~((PAGE_SIZE << order) - 1); |
| 1697 |
unsigned long pgoff = (addr - vma->vm_start) >> PAGE_SHIFT; |
| 1698 |
unsigned long pfn = vma_to_pfn(vma) + pgoff; |
| 1699 |
vm_fault_t ret = VM_FAULT_FALLBACK; |
| 1700 |
|
| 1701 |
if (is_aligned_for_order(vma, addr, pfn, order)) { |
| 1702 |
scoped_guard(rwsem_read, &vdev->memory_lock) |
| 1703 |
ret = vfio_pci_vmf_insert_pfn(vdev, vmf, pfn, order); |
| 1704 |
} |
| 1705 |
|
| 1706 |
dev_dbg_ratelimited(&vdev->pdev->dev, |
| 1707 |
"%s(,order = %d) BAR %ld page offset 0x%lx: 0x%x\n", |
| 1708 |
__func__, order, |
| 1709 |
vma->vm_pgoff >> |
| 1710 |
(VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT), |
| 1711 |
pgoff, (unsigned int)ret); |
| 1712 |
|
| 1713 |
return ret; |
| 1714 |
} |
| 1715 |
|
| 1716 |
static vm_fault_t vfio_pci_mmap_page_fault(struct vm_fault *vmf) |
| 1717 |
{ |
| 1718 |
return vfio_pci_mmap_huge_fault(vmf, 0); |
| 1719 |
} |
| 1720 |
|
| 1721 |
static const struct vm_operations_struct vfio_pci_mmap_ops = { |
| 1722 |
.fault = vfio_pci_mmap_page_fault, |
| 1723 |
#ifdef CONFIG_ARCH_SUPPORTS_HUGE_PFNMAP |
| 1724 |
.huge_fault = vfio_pci_mmap_huge_fault, |
| 1725 |
#endif |
| 1726 |
}; |
| 1727 |
|
| 1728 |
int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma) |
| 1729 |
{ |
| 1730 |
struct vfio_pci_core_device *vdev = |
| 1731 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1732 |
struct pci_dev *pdev = vdev->pdev; |
| 1733 |
unsigned int index; |
| 1734 |
u64 phys_len, req_len, pgoff, req_start; |
| 1735 |
int ret; |
| 1736 |
|
| 1737 |
index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); |
| 1738 |
|
| 1739 |
if (index >= VFIO_PCI_NUM_REGIONS + vdev->num_regions) |
| 1740 |
return -EINVAL; |
| 1741 |
if (vma->vm_end < vma->vm_start) |
| 1742 |
return -EINVAL; |
| 1743 |
if ((vma->vm_flags & VM_SHARED) == 0) |
| 1744 |
return -EINVAL; |
| 1745 |
if (index >= VFIO_PCI_NUM_REGIONS) { |
| 1746 |
int regnum = index - VFIO_PCI_NUM_REGIONS; |
| 1747 |
struct vfio_pci_region *region = vdev->region + regnum; |
| 1748 |
|
| 1749 |
if (region->ops && region->ops->mmap && |
| 1750 |
(region->flags & VFIO_REGION_INFO_FLAG_MMAP)) |
| 1751 |
return region->ops->mmap(vdev, region, vma); |
| 1752 |
return -EINVAL; |
| 1753 |
} |
| 1754 |
if (index >= VFIO_PCI_ROM_REGION_INDEX) |
| 1755 |
return -EINVAL; |
| 1756 |
if (!vdev->bar_mmap_supported[index]) |
| 1757 |
return -EINVAL; |
| 1758 |
|
| 1759 |
phys_len = PAGE_ALIGN(pci_resource_len(pdev, index)); |
| 1760 |
req_len = vma->vm_end - vma->vm_start; |
| 1761 |
pgoff = vma->vm_pgoff & |
| 1762 |
((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); |
| 1763 |
req_start = pgoff << PAGE_SHIFT; |
| 1764 |
|
| 1765 |
if (req_start + req_len > phys_len) |
| 1766 |
return -EINVAL; |
| 1767 |
|
| 1768 |
/* |
| 1769 |
* Even though we don't make use of the barmap for the mmap, |
| 1770 |
* we need to request the region and the barmap tracks that. |
| 1771 |
*/ |
| 1772 |
ret = vfio_pci_core_setup_barmap(vdev, index); |
| 1773 |
if (ret) |
| 1774 |
return ret; |
| 1775 |
|
| 1776 |
vma->vm_private_data = vdev; |
| 1777 |
vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); |
| 1778 |
vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot); |
| 1779 |
|
| 1780 |
/* |
| 1781 |
* Set vm_flags now, they should not be changed in the fault handler. |
| 1782 |
* We want the same flags and page protection (decrypted above) as |
| 1783 |
* io_remap_pfn_range() would set. |
| 1784 |
* |
| 1785 |
* VM_ALLOW_ANY_UNCACHED: The VMA flag is implemented for ARM64, |
| 1786 |
* allowing KVM stage 2 device mapping attributes to use Normal-NC |
| 1787 |
* rather than DEVICE_nGnRE, which allows guest mappings |
| 1788 |
* supporting write-combining attributes (WC). ARM does not |
| 1789 |
* architecturally guarantee this is safe, and indeed some MMIO |
| 1790 |
* regions like the GICv2 VCPU interface can trigger uncontained |
| 1791 |
* faults if Normal-NC is used. |
| 1792 |
* |
| 1793 |
* To safely use VFIO in KVM the platform must guarantee full |
| 1794 |
* safety in the guest where no action taken against a MMIO |
| 1795 |
* mapping can trigger an uncontained failure. The assumption is |
| 1796 |
* that most VFIO PCI platforms support this for both mapping types, |
| 1797 |
* at least in common flows, based on some expectations of how |
| 1798 |
* PCI IP is integrated. Hence VM_ALLOW_ANY_UNCACHED is set in |
| 1799 |
* the VMA flags. |
| 1800 |
*/ |
| 1801 |
vm_flags_set(vma, VM_ALLOW_ANY_UNCACHED | VM_IO | VM_PFNMAP | |
| 1802 |
VM_DONTEXPAND | VM_DONTDUMP); |
| 1803 |
vma->vm_ops = &vfio_pci_mmap_ops; |
| 1804 |
|
| 1805 |
return 0; |
| 1806 |
} |
| 1807 |
EXPORT_SYMBOL_GPL(vfio_pci_core_mmap); |
| 1808 |
|
| 1809 |
void vfio_pci_core_request(struct vfio_device *core_vdev, unsigned int count) |
| 1810 |
{ |
| 1811 |
struct vfio_pci_core_device *vdev = |
| 1812 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1813 |
struct pci_dev *pdev = vdev->pdev; |
| 1814 |
struct vfio_pci_eventfd *eventfd; |
| 1815 |
|
| 1816 |
rcu_read_lock(); |
| 1817 |
eventfd = rcu_dereference(vdev->req_trigger); |
| 1818 |
if (eventfd) { |
| 1819 |
if (!(count % 10)) |
| 1820 |
pci_notice_ratelimited(pdev, |
| 1821 |
"Relaying device request to user (#%u)\n", |
| 1822 |
count); |
| 1823 |
eventfd_signal(eventfd->ctx); |
| 1824 |
} else if (count == 0) { |
| 1825 |
pci_warn(pdev, |
| 1826 |
"No device request channel registered, blocked until released by user\n"); |
| 1827 |
} |
| 1828 |
rcu_read_unlock(); |
| 1829 |
} |
| 1830 |
EXPORT_SYMBOL_GPL(vfio_pci_core_request); |
| 1831 |
|
| 1832 |
int vfio_pci_core_match_token_uuid(struct vfio_device *core_vdev, |
| 1833 |
const uuid_t *uuid) |
| 1834 |
|
| 1835 |
{ |
| 1836 |
struct vfio_pci_core_device *vdev = |
| 1837 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1838 |
|
| 1839 |
/* |
| 1840 |
* There's always some degree of trust or collaboration between SR-IOV |
| 1841 |
* PF and VFs, even if just that the PF hosts the SR-IOV capability and |
| 1842 |
* can disrupt VFs with a reset, but often the PF has more explicit |
| 1843 |
* access to deny service to the VF or access data passed through the |
| 1844 |
* VF. We therefore require an opt-in via a shared VF token (UUID) to |
| 1845 |
* represent this trust. This both prevents that a VF driver might |
| 1846 |
* assume the PF driver is a trusted, in-kernel driver, and also that |
| 1847 |
* a PF driver might be replaced with a rogue driver, unknown to in-use |
| 1848 |
* VF drivers. |
| 1849 |
* |
| 1850 |
* Therefore when presented with a VF, if the PF is a vfio device and |
| 1851 |
* it is bound to the vfio-pci driver, the user needs to provide a VF |
| 1852 |
* token to access the device, in the form of appending a vf_token to |
| 1853 |
* the device name, for example: |
| 1854 |
* |
| 1855 |
* "0000:04:10.0 vf_token=bd8d9d2b-5a5f-4f5a-a211-f591514ba1f3" |
| 1856 |
* |
| 1857 |
* When presented with a PF which has VFs in use, the user must also |
| 1858 |
* provide the current VF token to prove collaboration with existing |
| 1859 |
* VF users. If VFs are not in use, the VF token provided for the PF |
| 1860 |
* device will act to set the VF token. |
| 1861 |
* |
| 1862 |
* If the VF token is provided but unused, an error is generated. |
| 1863 |
*/ |
| 1864 |
if (vdev->pdev->is_virtfn) { |
| 1865 |
struct vfio_pci_core_device *pf_vdev = vdev->sriov_pf_core_dev; |
| 1866 |
bool match; |
| 1867 |
|
| 1868 |
if (!pf_vdev) { |
| 1869 |
if (!uuid) |
| 1870 |
return 0; /* PF is not vfio-pci, no VF token */ |
| 1871 |
|
| 1872 |
pci_info_ratelimited(vdev->pdev, |
| 1873 |
"VF token incorrectly provided, PF not bound to vfio-pci\n"); |
| 1874 |
return -EINVAL; |
| 1875 |
} |
| 1876 |
|
| 1877 |
if (!uuid) { |
| 1878 |
pci_info_ratelimited(vdev->pdev, |
| 1879 |
"VF token required to access device\n"); |
| 1880 |
return -EACCES; |
| 1881 |
} |
| 1882 |
|
| 1883 |
mutex_lock(&pf_vdev->vf_token->lock); |
| 1884 |
match = uuid_equal(uuid, &pf_vdev->vf_token->uuid); |
| 1885 |
mutex_unlock(&pf_vdev->vf_token->lock); |
| 1886 |
|
| 1887 |
if (!match) { |
| 1888 |
pci_info_ratelimited(vdev->pdev, |
| 1889 |
"Incorrect VF token provided for device\n"); |
| 1890 |
return -EACCES; |
| 1891 |
} |
| 1892 |
} else if (vdev->vf_token) { |
| 1893 |
mutex_lock(&vdev->vf_token->lock); |
| 1894 |
if (vdev->vf_token->users) { |
| 1895 |
if (!uuid) { |
| 1896 |
mutex_unlock(&vdev->vf_token->lock); |
| 1897 |
pci_info_ratelimited(vdev->pdev, |
| 1898 |
"VF token required to access device\n"); |
| 1899 |
return -EACCES; |
| 1900 |
} |
| 1901 |
|
| 1902 |
if (!uuid_equal(uuid, &vdev->vf_token->uuid)) { |
| 1903 |
mutex_unlock(&vdev->vf_token->lock); |
| 1904 |
pci_info_ratelimited(vdev->pdev, |
| 1905 |
"Incorrect VF token provided for device\n"); |
| 1906 |
return -EACCES; |
| 1907 |
} |
| 1908 |
} else if (uuid) { |
| 1909 |
uuid_copy(&vdev->vf_token->uuid, uuid); |
| 1910 |
} |
| 1911 |
|
| 1912 |
mutex_unlock(&vdev->vf_token->lock); |
| 1913 |
} else if (uuid) { |
| 1914 |
pci_info_ratelimited(vdev->pdev, |
| 1915 |
"VF token incorrectly provided, not a PF or VF\n"); |
| 1916 |
return -EINVAL; |
| 1917 |
} |
| 1918 |
|
| 1919 |
return 0; |
| 1920 |
} |
| 1921 |
EXPORT_SYMBOL_GPL(vfio_pci_core_match_token_uuid); |
| 1922 |
|
| 1923 |
#define VF_TOKEN_ARG "vf_token=" |
| 1924 |
|
| 1925 |
int vfio_pci_core_match(struct vfio_device *core_vdev, char *buf) |
| 1926 |
{ |
| 1927 |
struct vfio_pci_core_device *vdev = |
| 1928 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 1929 |
bool vf_token = false; |
| 1930 |
uuid_t uuid; |
| 1931 |
int ret; |
| 1932 |
|
| 1933 |
if (strncmp(pci_name(vdev->pdev), buf, strlen(pci_name(vdev->pdev)))) |
| 1934 |
return 0; /* No match */ |
| 1935 |
|
| 1936 |
if (strlen(buf) > strlen(pci_name(vdev->pdev))) { |
| 1937 |
buf += strlen(pci_name(vdev->pdev)); |
| 1938 |
|
| 1939 |
if (*buf != ' ') |
| 1940 |
return 0; /* No match: non-whitespace after name */ |
| 1941 |
|
| 1942 |
while (*buf) { |
| 1943 |
if (*buf == ' ') { |
| 1944 |
buf++; |
| 1945 |
continue; |
| 1946 |
} |
| 1947 |
|
| 1948 |
if (!vf_token && !strncmp(buf, VF_TOKEN_ARG, |
| 1949 |
strlen(VF_TOKEN_ARG))) { |
| 1950 |
buf += strlen(VF_TOKEN_ARG); |
| 1951 |
|
| 1952 |
if (strlen(buf) < UUID_STRING_LEN) |
| 1953 |
return -EINVAL; |
| 1954 |
|
| 1955 |
ret = uuid_parse(buf, &uuid); |
| 1956 |
if (ret) |
| 1957 |
return ret; |
| 1958 |
|
| 1959 |
vf_token = true; |
| 1960 |
buf += UUID_STRING_LEN; |
| 1961 |
} else { |
| 1962 |
/* Unknown/duplicate option */ |
| 1963 |
return -EINVAL; |
| 1964 |
} |
| 1965 |
} |
| 1966 |
} |
| 1967 |
|
| 1968 |
ret = core_vdev->ops->match_token_uuid(core_vdev, |
| 1969 |
vf_token ? &uuid : NULL); |
| 1970 |
if (ret) |
| 1971 |
return ret; |
| 1972 |
|
| 1973 |
return 1; /* Match */ |
| 1974 |
} |
| 1975 |
EXPORT_SYMBOL_GPL(vfio_pci_core_match); |
| 1976 |
|
| 1977 |
static int vfio_pci_bus_notifier(struct notifier_block *nb, |
| 1978 |
unsigned long action, void *data) |
| 1979 |
{ |
| 1980 |
struct vfio_pci_core_device *vdev = container_of(nb, |
| 1981 |
struct vfio_pci_core_device, nb); |
| 1982 |
struct device *dev = data; |
| 1983 |
struct pci_dev *pdev = to_pci_dev(dev); |
| 1984 |
struct pci_dev *physfn = pci_physfn(pdev); |
| 1985 |
|
| 1986 |
if (action == BUS_NOTIFY_ADD_DEVICE && |
| 1987 |
pdev->is_virtfn && physfn == vdev->pdev) { |
| 1988 |
pci_info(vdev->pdev, "Captured SR-IOV VF %s driver_override\n", |
| 1989 |
pci_name(pdev)); |
| 1990 |
pdev->driver_override = kasprintf(GFP_KERNEL, "%s", |
| 1991 |
vdev->vdev.ops->name); |
| 1992 |
WARN_ON(!pdev->driver_override); |
| 1993 |
} else if (action == BUS_NOTIFY_BOUND_DRIVER && |
| 1994 |
pdev->is_virtfn && physfn == vdev->pdev) { |
| 1995 |
struct pci_driver *drv = pci_dev_driver(pdev); |
| 1996 |
|
| 1997 |
if (drv && drv != pci_dev_driver(vdev->pdev)) |
| 1998 |
pci_warn(vdev->pdev, |
| 1999 |
"VF %s bound to driver %s while PF bound to driver %s\n", |
| 2000 |
pci_name(pdev), drv->name, |
| 2001 |
pci_dev_driver(vdev->pdev)->name); |
| 2002 |
} |
| 2003 |
|
| 2004 |
return 0; |
| 2005 |
} |
| 2006 |
|
| 2007 |
static int vfio_pci_vf_init(struct vfio_pci_core_device *vdev) |
| 2008 |
{ |
| 2009 |
struct pci_dev *pdev = vdev->pdev; |
| 2010 |
struct vfio_pci_core_device *cur; |
| 2011 |
struct pci_dev *physfn; |
| 2012 |
int ret; |
| 2013 |
|
| 2014 |
if (pdev->is_virtfn) { |
| 2015 |
/* |
| 2016 |
* If this VF was created by our vfio_pci_core_sriov_configure() |
| 2017 |
* then we can find the PF vfio_pci_core_device now, and due to |
| 2018 |
* the locking in pci_disable_sriov() it cannot change until |
| 2019 |
* this VF device driver is removed. |
| 2020 |
*/ |
| 2021 |
physfn = pci_physfn(vdev->pdev); |
| 2022 |
mutex_lock(&vfio_pci_sriov_pfs_mutex); |
| 2023 |
list_for_each_entry(cur, &vfio_pci_sriov_pfs, sriov_pfs_item) { |
| 2024 |
if (cur->pdev == physfn) { |
| 2025 |
vdev->sriov_pf_core_dev = cur; |
| 2026 |
break; |
| 2027 |
} |
| 2028 |
} |
| 2029 |
mutex_unlock(&vfio_pci_sriov_pfs_mutex); |
| 2030 |
return 0; |
| 2031 |
} |
| 2032 |
|
| 2033 |
/* Not a SRIOV PF */ |
| 2034 |
if (!pdev->is_physfn) |
| 2035 |
return 0; |
| 2036 |
|
| 2037 |
vdev->vf_token = kzalloc_obj(*vdev->vf_token); |
| 2038 |
if (!vdev->vf_token) |
| 2039 |
return -ENOMEM; |
| 2040 |
|
| 2041 |
mutex_init(&vdev->vf_token->lock); |
| 2042 |
uuid_gen(&vdev->vf_token->uuid); |
| 2043 |
|
| 2044 |
vdev->nb.notifier_call = vfio_pci_bus_notifier; |
| 2045 |
ret = bus_register_notifier(&pci_bus_type, &vdev->nb); |
| 2046 |
if (ret) { |
| 2047 |
kfree(vdev->vf_token); |
| 2048 |
return ret; |
| 2049 |
} |
| 2050 |
return 0; |
| 2051 |
} |
| 2052 |
|
| 2053 |
static void vfio_pci_vf_uninit(struct vfio_pci_core_device *vdev) |
| 2054 |
{ |
| 2055 |
if (!vdev->vf_token) |
| 2056 |
return; |
| 2057 |
|
| 2058 |
bus_unregister_notifier(&pci_bus_type, &vdev->nb); |
| 2059 |
WARN_ON(vdev->vf_token->users); |
| 2060 |
mutex_destroy(&vdev->vf_token->lock); |
| 2061 |
kfree(vdev->vf_token); |
| 2062 |
} |
| 2063 |
|
| 2064 |
static int vfio_pci_vga_init(struct vfio_pci_core_device *vdev) |
| 2065 |
{ |
| 2066 |
struct pci_dev *pdev = vdev->pdev; |
| 2067 |
int ret; |
| 2068 |
|
| 2069 |
if (!vfio_pci_is_vga(pdev)) |
| 2070 |
return 0; |
| 2071 |
|
| 2072 |
ret = aperture_remove_conflicting_pci_devices(pdev, vdev->vdev.ops->name); |
| 2073 |
if (ret) |
| 2074 |
return ret; |
| 2075 |
|
| 2076 |
ret = vga_client_register(pdev, vfio_pci_set_decode); |
| 2077 |
if (ret) |
| 2078 |
return ret; |
| 2079 |
vga_set_legacy_decoding(pdev, vfio_pci_set_decode(pdev, false)); |
| 2080 |
return 0; |
| 2081 |
} |
| 2082 |
|
| 2083 |
static void vfio_pci_vga_uninit(struct vfio_pci_core_device *vdev) |
| 2084 |
{ |
| 2085 |
struct pci_dev *pdev = vdev->pdev; |
| 2086 |
|
| 2087 |
if (!vfio_pci_is_vga(pdev)) |
| 2088 |
return; |
| 2089 |
vga_client_unregister(pdev); |
| 2090 |
vga_set_legacy_decoding(pdev, VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM | |
| 2091 |
VGA_RSRC_LEGACY_IO | |
| 2092 |
VGA_RSRC_LEGACY_MEM); |
| 2093 |
} |
| 2094 |
|
| 2095 |
int vfio_pci_core_init_dev(struct vfio_device *core_vdev) |
| 2096 |
{ |
| 2097 |
struct vfio_pci_core_device *vdev = |
| 2098 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 2099 |
int ret; |
| 2100 |
|
| 2101 |
vdev->pdev = to_pci_dev(core_vdev->dev); |
| 2102 |
vdev->irq_type = VFIO_PCI_NUM_IRQS; |
| 2103 |
mutex_init(&vdev->igate); |
| 2104 |
spin_lock_init(&vdev->irqlock); |
| 2105 |
mutex_init(&vdev->ioeventfds_lock); |
| 2106 |
INIT_LIST_HEAD(&vdev->dummy_resources_list); |
| 2107 |
INIT_LIST_HEAD(&vdev->ioeventfds_list); |
| 2108 |
INIT_LIST_HEAD(&vdev->sriov_pfs_item); |
| 2109 |
ret = pcim_p2pdma_init(vdev->pdev); |
| 2110 |
if (ret && ret != -EOPNOTSUPP) |
| 2111 |
return ret; |
| 2112 |
INIT_LIST_HEAD(&vdev->dmabufs); |
| 2113 |
init_rwsem(&vdev->memory_lock); |
| 2114 |
xa_init(&vdev->ctx); |
| 2115 |
|
| 2116 |
return 0; |
| 2117 |
} |
| 2118 |
EXPORT_SYMBOL_GPL(vfio_pci_core_init_dev); |
| 2119 |
|
| 2120 |
void vfio_pci_core_release_dev(struct vfio_device *core_vdev) |
| 2121 |
{ |
| 2122 |
struct vfio_pci_core_device *vdev = |
| 2123 |
container_of(core_vdev, struct vfio_pci_core_device, vdev); |
| 2124 |
|
| 2125 |
mutex_destroy(&vdev->igate); |
| 2126 |
mutex_destroy(&vdev->ioeventfds_lock); |
| 2127 |
kfree(vdev->region); |
| 2128 |
kfree(vdev->pm_save); |
| 2129 |
} |
| 2130 |
EXPORT_SYMBOL_GPL(vfio_pci_core_release_dev); |
| 2131 |
|
| 2132 |
int vfio_pci_core_register_device(struct vfio_pci_core_device *vdev) |
| 2133 |
{ |
| 2134 |
struct pci_dev *pdev = vdev->pdev; |
| 2135 |
struct device *dev = &pdev->dev; |
| 2136 |
int ret; |
| 2137 |
|
| 2138 |
/* Drivers must set the vfio_pci_core_device to their drvdata */ |
| 2139 |
if (WARN_ON(vdev != dev_get_drvdata(dev))) |
| 2140 |
return -EINVAL; |
| 2141 |
|
| 2142 |
if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL) |
| 2143 |
return -EINVAL; |
| 2144 |
|
| 2145 |
if (vdev->vdev.mig_ops) { |
| 2146 |
if (!(vdev->vdev.mig_ops->migration_get_state && |
| 2147 |
vdev->vdev.mig_ops->migration_set_state && |
| 2148 |
vdev->vdev.mig_ops->migration_get_data_size) || |
| 2149 |
!(vdev->vdev.migration_flags & VFIO_MIGRATION_STOP_COPY)) |
| 2150 |
return -EINVAL; |
| 2151 |
} |
| 2152 |
|
| 2153 |
if (vdev->vdev.log_ops && !(vdev->vdev.log_ops->log_start && |
| 2154 |
vdev->vdev.log_ops->log_stop && |
| 2155 |
vdev->vdev.log_ops->log_read_and_clear)) |
| 2156 |
return -EINVAL; |
| 2157 |
|
| 2158 |
/* |
| 2159 |
* Prevent binding to PFs with VFs enabled, the VFs might be in use |
| 2160 |
* by the host or other users. We cannot capture the VFs if they |
| 2161 |
* already exist, nor can we track VF users. Disabling SR-IOV here |
| 2162 |
* would initiate removing the VFs, which would unbind the driver, |
| 2163 |
* which is prone to blocking if that VF is also in use by vfio-pci. |
| 2164 |
* Just reject these PFs and let the user sort it out. |
| 2165 |
*/ |
| 2166 |
if (pci_num_vf(pdev)) { |
| 2167 |
pci_warn(pdev, "Cannot bind to PF with SR-IOV enabled\n"); |
| 2168 |
return -EBUSY; |
| 2169 |
} |
| 2170 |
|
| 2171 |
if (pci_is_root_bus(pdev->bus) || pdev->is_virtfn) { |
| 2172 |
ret = vfio_assign_device_set(&vdev->vdev, vdev); |
| 2173 |
} else if (!pci_probe_reset_slot(pdev->slot)) { |
| 2174 |
ret = vfio_assign_device_set(&vdev->vdev, pdev->slot); |
| 2175 |
} else { |
| 2176 |
/* |
| 2177 |
* If there is no slot reset support for this device, the whole |
| 2178 |
* bus needs to be grouped together to support bus-wide resets. |
| 2179 |
*/ |
| 2180 |
ret = vfio_assign_device_set(&vdev->vdev, pdev->bus); |
| 2181 |
} |
| 2182 |
|
| 2183 |
if (ret) |
| 2184 |
return ret; |
| 2185 |
ret = vfio_pci_vf_init(vdev); |
| 2186 |
if (ret) |
| 2187 |
return ret; |
| 2188 |
ret = vfio_pci_vga_init(vdev); |
| 2189 |
if (ret) |
| 2190 |
goto out_vf; |
| 2191 |
|
| 2192 |
vfio_pci_probe_power_state(vdev); |
| 2193 |
|
| 2194 |
/* |
| 2195 |
* pci-core sets the device power state to an unknown value at |
| 2196 |
* bootup and after being removed from a driver. The only |
| 2197 |
* transition it allows from this unknown state is to D0, which |
| 2198 |
* typically happens when a driver calls pci_enable_device(). |
| 2199 |
* We're not ready to enable the device yet, but we do want to |
| 2200 |
* be able to get to D3. Therefore first do a D0 transition |
| 2201 |
* before enabling runtime PM. |
| 2202 |
*/ |
| 2203 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 2204 |
|
| 2205 |
dev->driver->pm = &vfio_pci_core_pm_ops; |
| 2206 |
pm_runtime_allow(dev); |
| 2207 |
if (!disable_idle_d3) |
| 2208 |
pm_runtime_put(dev); |
| 2209 |
|
| 2210 |
ret = vfio_register_group_dev(&vdev->vdev); |
| 2211 |
if (ret) |
| 2212 |
goto out_power; |
| 2213 |
return 0; |
| 2214 |
|
| 2215 |
out_power: |
| 2216 |
if (!disable_idle_d3) |
| 2217 |
pm_runtime_get_noresume(dev); |
| 2218 |
|
| 2219 |
pm_runtime_forbid(dev); |
| 2220 |
out_vf: |
| 2221 |
vfio_pci_vf_uninit(vdev); |
| 2222 |
return ret; |
| 2223 |
} |
| 2224 |
EXPORT_SYMBOL_GPL(vfio_pci_core_register_device); |
| 2225 |
|
| 2226 |
void vfio_pci_core_unregister_device(struct vfio_pci_core_device *vdev) |
| 2227 |
{ |
| 2228 |
vfio_pci_core_sriov_configure(vdev, 0); |
| 2229 |
|
| 2230 |
vfio_unregister_group_dev(&vdev->vdev); |
| 2231 |
|
| 2232 |
vfio_pci_vf_uninit(vdev); |
| 2233 |
vfio_pci_vga_uninit(vdev); |
| 2234 |
|
| 2235 |
if (!disable_idle_d3) |
| 2236 |
pm_runtime_get_noresume(&vdev->pdev->dev); |
| 2237 |
|
| 2238 |
pm_runtime_forbid(&vdev->pdev->dev); |
| 2239 |
} |
| 2240 |
EXPORT_SYMBOL_GPL(vfio_pci_core_unregister_device); |
| 2241 |
|
| 2242 |
pci_ers_result_t vfio_pci_core_aer_err_detected(struct pci_dev *pdev, |
| 2243 |
pci_channel_state_t state) |
| 2244 |
{ |
| 2245 |
struct vfio_pci_core_device *vdev = dev_get_drvdata(&pdev->dev); |
| 2246 |
struct vfio_pci_eventfd *eventfd; |
| 2247 |
|
| 2248 |
rcu_read_lock(); |
| 2249 |
eventfd = rcu_dereference(vdev->err_trigger); |
| 2250 |
if (eventfd) |
| 2251 |
eventfd_signal(eventfd->ctx); |
| 2252 |
rcu_read_unlock(); |
| 2253 |
|
| 2254 |
return PCI_ERS_RESULT_CAN_RECOVER; |
| 2255 |
} |
| 2256 |
EXPORT_SYMBOL_GPL(vfio_pci_core_aer_err_detected); |
| 2257 |
|
| 2258 |
int vfio_pci_core_sriov_configure(struct vfio_pci_core_device *vdev, |
| 2259 |
int nr_virtfn) |
| 2260 |
{ |
| 2261 |
struct pci_dev *pdev = vdev->pdev; |
| 2262 |
int ret = 0; |
| 2263 |
|
| 2264 |
device_lock_assert(&pdev->dev); |
| 2265 |
|
| 2266 |
if (nr_virtfn) { |
| 2267 |
mutex_lock(&vfio_pci_sriov_pfs_mutex); |
| 2268 |
/* |
| 2269 |
* The thread that adds the vdev to the list is the only thread |
| 2270 |
* that gets to call pci_enable_sriov() and we will only allow |
| 2271 |
* it to be called once without going through |
| 2272 |
* pci_disable_sriov() |
| 2273 |
*/ |
| 2274 |
if (!list_empty(&vdev->sriov_pfs_item)) { |
| 2275 |
ret = -EINVAL; |
| 2276 |
goto out_unlock; |
| 2277 |
} |
| 2278 |
list_add_tail(&vdev->sriov_pfs_item, &vfio_pci_sriov_pfs); |
| 2279 |
mutex_unlock(&vfio_pci_sriov_pfs_mutex); |
| 2280 |
|
| 2281 |
/* |
| 2282 |
* The PF power state should always be higher than the VF power |
| 2283 |
* state. The PF can be in low power state either with runtime |
| 2284 |
* power management (when there is no user) or PCI_PM_CTRL |
| 2285 |
* register write by the user. If PF is in the low power state, |
| 2286 |
* then change the power state to D0 first before enabling |
| 2287 |
* SR-IOV. Also, this function can be called at any time, and |
| 2288 |
* userspace PCI_PM_CTRL write can race against this code path, |
| 2289 |
* so protect the same with 'memory_lock'. |
| 2290 |
*/ |
| 2291 |
ret = pm_runtime_resume_and_get(&pdev->dev); |
| 2292 |
if (ret) |
| 2293 |
goto out_del; |
| 2294 |
|
| 2295 |
down_write(&vdev->memory_lock); |
| 2296 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 2297 |
ret = pci_enable_sriov(pdev, nr_virtfn); |
| 2298 |
up_write(&vdev->memory_lock); |
| 2299 |
if (ret) { |
| 2300 |
pm_runtime_put(&pdev->dev); |
| 2301 |
goto out_del; |
| 2302 |
} |
| 2303 |
return nr_virtfn; |
| 2304 |
} |
| 2305 |
|
| 2306 |
if (pci_num_vf(pdev)) { |
| 2307 |
pci_disable_sriov(pdev); |
| 2308 |
pm_runtime_put(&pdev->dev); |
| 2309 |
} |
| 2310 |
|
| 2311 |
out_del: |
| 2312 |
mutex_lock(&vfio_pci_sriov_pfs_mutex); |
| 2313 |
list_del_init(&vdev->sriov_pfs_item); |
| 2314 |
out_unlock: |
| 2315 |
mutex_unlock(&vfio_pci_sriov_pfs_mutex); |
| 2316 |
return ret; |
| 2317 |
} |
| 2318 |
EXPORT_SYMBOL_GPL(vfio_pci_core_sriov_configure); |
| 2319 |
|
| 2320 |
const struct pci_error_handlers vfio_pci_core_err_handlers = { |
| 2321 |
.error_detected = vfio_pci_core_aer_err_detected, |
| 2322 |
}; |
| 2323 |
EXPORT_SYMBOL_GPL(vfio_pci_core_err_handlers); |
| 2324 |
|
| 2325 |
static bool vfio_dev_in_groups(struct vfio_device *vdev, |
| 2326 |
struct vfio_pci_group_info *groups) |
| 2327 |
{ |
| 2328 |
unsigned int i; |
| 2329 |
|
| 2330 |
if (!groups) |
| 2331 |
return false; |
| 2332 |
|
| 2333 |
for (i = 0; i < groups->count; i++) |
| 2334 |
if (vfio_file_has_dev(groups->files[i], vdev)) |
| 2335 |
return true; |
| 2336 |
return false; |
| 2337 |
} |
| 2338 |
|
| 2339 |
static int vfio_pci_is_device_in_set(struct pci_dev *pdev, void *data) |
| 2340 |
{ |
| 2341 |
struct vfio_device_set *dev_set = data; |
| 2342 |
|
| 2343 |
return vfio_find_device_in_devset(dev_set, &pdev->dev) ? 0 : -ENODEV; |
| 2344 |
} |
| 2345 |
|
| 2346 |
/* |
| 2347 |
* vfio-core considers a group to be viable and will create a vfio_device even |
| 2348 |
* if some devices are bound to drivers like pci-stub or pcieport. Here we |
| 2349 |
* require all PCI devices to be inside our dev_set since that ensures they stay |
| 2350 |
* put and that every driver controlling the device can co-ordinate with the |
| 2351 |
* device reset. |
| 2352 |
* |
| 2353 |
* Returns the pci_dev to pass to pci_reset_bus() if every PCI device to be |
| 2354 |
* reset is inside the dev_set, and pci_reset_bus() can succeed. NULL otherwise. |
| 2355 |
*/ |
| 2356 |
static struct pci_dev * |
| 2357 |
vfio_pci_dev_set_resettable(struct vfio_device_set *dev_set) |
| 2358 |
{ |
| 2359 |
struct pci_dev *pdev; |
| 2360 |
|
| 2361 |
lockdep_assert_held(&dev_set->lock); |
| 2362 |
|
| 2363 |
/* |
| 2364 |
* By definition all PCI devices in the dev_set share the same PCI |
| 2365 |
* reset, so any pci_dev will have the same outcomes for |
| 2366 |
* pci_probe_reset_*() and pci_reset_bus(). |
| 2367 |
*/ |
| 2368 |
pdev = list_first_entry(&dev_set->device_list, |
| 2369 |
struct vfio_pci_core_device, |
| 2370 |
vdev.dev_set_list)->pdev; |
| 2371 |
|
| 2372 |
/* pci_reset_bus() is supported */ |
| 2373 |
if (pci_probe_reset_slot(pdev->slot) && pci_probe_reset_bus(pdev->bus)) |
| 2374 |
return NULL; |
| 2375 |
|
| 2376 |
if (vfio_pci_for_each_slot_or_bus(pdev, vfio_pci_is_device_in_set, |
| 2377 |
dev_set, |
| 2378 |
!pci_probe_reset_slot(pdev->slot))) |
| 2379 |
return NULL; |
| 2380 |
return pdev; |
| 2381 |
} |
| 2382 |
|
| 2383 |
static int vfio_pci_dev_set_pm_runtime_get(struct vfio_device_set *dev_set) |
| 2384 |
{ |
| 2385 |
struct vfio_pci_core_device *cur; |
| 2386 |
int ret; |
| 2387 |
|
| 2388 |
list_for_each_entry(cur, &dev_set->device_list, vdev.dev_set_list) { |
| 2389 |
ret = pm_runtime_resume_and_get(&cur->pdev->dev); |
| 2390 |
if (ret) |
| 2391 |
goto unwind; |
| 2392 |
} |
| 2393 |
|
| 2394 |
return 0; |
| 2395 |
|
| 2396 |
unwind: |
| 2397 |
list_for_each_entry_continue_reverse(cur, &dev_set->device_list, |
| 2398 |
vdev.dev_set_list) |
| 2399 |
pm_runtime_put(&cur->pdev->dev); |
| 2400 |
|
| 2401 |
return ret; |
| 2402 |
} |
| 2403 |
|
| 2404 |
static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set, |
| 2405 |
struct vfio_pci_group_info *groups, |
| 2406 |
struct iommufd_ctx *iommufd_ctx) |
| 2407 |
{ |
| 2408 |
struct vfio_pci_core_device *vdev; |
| 2409 |
struct pci_dev *pdev; |
| 2410 |
int ret; |
| 2411 |
|
| 2412 |
mutex_lock(&dev_set->lock); |
| 2413 |
|
| 2414 |
pdev = vfio_pci_dev_set_resettable(dev_set); |
| 2415 |
if (!pdev) { |
| 2416 |
ret = -EINVAL; |
| 2417 |
goto err_unlock; |
| 2418 |
} |
| 2419 |
|
| 2420 |
/* |
| 2421 |
* Some of the devices in the dev_set can be in the runtime suspended |
| 2422 |
* state. Increment the usage count for all the devices in the dev_set |
| 2423 |
* before reset and decrement the same after reset. |
| 2424 |
*/ |
| 2425 |
ret = vfio_pci_dev_set_pm_runtime_get(dev_set); |
| 2426 |
if (ret) |
| 2427 |
goto err_unlock; |
| 2428 |
|
| 2429 |
list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list) { |
| 2430 |
bool owned; |
| 2431 |
|
| 2432 |
/* |
| 2433 |
* Test whether all the affected devices can be reset by the |
| 2434 |
* user. |
| 2435 |
* |
| 2436 |
* If called from a group opened device and the user provides |
| 2437 |
* a set of groups, all the devices in the dev_set should be |
| 2438 |
* contained by the set of groups provided by the user. |
| 2439 |
* |
| 2440 |
* If called from a cdev opened device and the user provides |
| 2441 |
* a zero-length array, all the devices in the dev_set must |
| 2442 |
* be bound to the same iommufd_ctx as the input iommufd_ctx. |
| 2443 |
* If there is any device that has not been bound to any |
| 2444 |
* iommufd_ctx yet, check if its iommu_group has any device |
| 2445 |
* bound to the input iommufd_ctx. Such devices can be |
| 2446 |
* considered owned by the input iommufd_ctx as the device |
| 2447 |
* cannot be owned by another iommufd_ctx when its iommu_group |
| 2448 |
* is owned. |
| 2449 |
* |
| 2450 |
* Otherwise, reset is not allowed. |
| 2451 |
*/ |
| 2452 |
if (iommufd_ctx) { |
| 2453 |
int devid = vfio_iommufd_get_dev_id(&vdev->vdev, |
| 2454 |
iommufd_ctx); |
| 2455 |
|
| 2456 |
owned = (devid > 0 || devid == -ENOENT); |
| 2457 |
} else { |
| 2458 |
owned = vfio_dev_in_groups(&vdev->vdev, groups); |
| 2459 |
} |
| 2460 |
|
| 2461 |
if (!owned) { |
| 2462 |
ret = -EINVAL; |
| 2463 |
break; |
| 2464 |
} |
| 2465 |
|
| 2466 |
/* |
| 2467 |
* Take the memory write lock for each device and zap BAR |
| 2468 |
* mappings to prevent the user accessing the device while in |
| 2469 |
* reset. Locking multiple devices is prone to deadlock, |
| 2470 |
* runaway and unwind if we hit contention. |
| 2471 |
*/ |
| 2472 |
if (!down_write_trylock(&vdev->memory_lock)) { |
| 2473 |
ret = -EBUSY; |
| 2474 |
break; |
| 2475 |
} |
| 2476 |
|
| 2477 |
vfio_pci_dma_buf_move(vdev, true); |
| 2478 |
vfio_pci_zap_bars(vdev); |
| 2479 |
} |
| 2480 |
|
| 2481 |
if (!list_entry_is_head(vdev, |
| 2482 |
&dev_set->device_list, vdev.dev_set_list)) { |
| 2483 |
vdev = list_prev_entry(vdev, vdev.dev_set_list); |
| 2484 |
goto err_undo; |
| 2485 |
} |
| 2486 |
|
| 2487 |
/* |
| 2488 |
* The pci_reset_bus() will reset all the devices in the bus. |
| 2489 |
* The power state can be non-D0 for some of the devices in the bus. |
| 2490 |
* For these devices, the pci_reset_bus() will internally set |
| 2491 |
* the power state to D0 without vfio driver involvement. |
| 2492 |
* For the devices which have NoSoftRst-, the reset function can |
| 2493 |
* cause the PCI config space reset without restoring the original |
| 2494 |
* state (saved locally in 'vdev->pm_save'). |
| 2495 |
*/ |
| 2496 |
list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list) |
| 2497 |
vfio_pci_set_power_state(vdev, PCI_D0); |
| 2498 |
|
| 2499 |
ret = pci_reset_bus(pdev); |
| 2500 |
|
| 2501 |
vdev = list_last_entry(&dev_set->device_list, |
| 2502 |
struct vfio_pci_core_device, vdev.dev_set_list); |
| 2503 |
|
| 2504 |
err_undo: |
| 2505 |
list_for_each_entry_from_reverse(vdev, &dev_set->device_list, |
| 2506 |
vdev.dev_set_list) { |
| 2507 |
if (vdev->vdev.open_count && __vfio_pci_memory_enabled(vdev)) |
| 2508 |
vfio_pci_dma_buf_move(vdev, false); |
| 2509 |
up_write(&vdev->memory_lock); |
| 2510 |
} |
| 2511 |
|
| 2512 |
list_for_each_entry(vdev, &dev_set->device_list, vdev.dev_set_list) |
| 2513 |
pm_runtime_put(&vdev->pdev->dev); |
| 2514 |
|
| 2515 |
err_unlock: |
| 2516 |
mutex_unlock(&dev_set->lock); |
| 2517 |
return ret; |
| 2518 |
} |
| 2519 |
|
| 2520 |
static bool vfio_pci_dev_set_needs_reset(struct vfio_device_set *dev_set) |
| 2521 |
{ |
| 2522 |
struct vfio_pci_core_device *cur; |
| 2523 |
bool needs_reset = false; |
| 2524 |
|
| 2525 |
/* No other VFIO device in the set can be open. */ |
| 2526 |
if (vfio_device_set_open_count(dev_set) > 1) |
| 2527 |
return false; |
| 2528 |
|
| 2529 |
list_for_each_entry(cur, &dev_set->device_list, vdev.dev_set_list) |
| 2530 |
needs_reset |= cur->needs_reset; |
| 2531 |
return needs_reset; |
| 2532 |
} |
| 2533 |
|
| 2534 |
/* |
| 2535 |
* If a bus or slot reset is available for the provided dev_set and: |
| 2536 |
* - All of the devices affected by that bus or slot reset are unused |
| 2537 |
* - At least one of the affected devices is marked dirty via |
| 2538 |
* needs_reset (such as by lack of FLR support) |
| 2539 |
* Then attempt to perform that bus or slot reset. |
| 2540 |
*/ |
| 2541 |
static void vfio_pci_dev_set_try_reset(struct vfio_device_set *dev_set) |
| 2542 |
{ |
| 2543 |
struct vfio_pci_core_device *cur; |
| 2544 |
struct pci_dev *pdev; |
| 2545 |
bool reset_done = false; |
| 2546 |
|
| 2547 |
if (!vfio_pci_dev_set_needs_reset(dev_set)) |
| 2548 |
return; |
| 2549 |
|
| 2550 |
pdev = vfio_pci_dev_set_resettable(dev_set); |
| 2551 |
if (!pdev) |
| 2552 |
return; |
| 2553 |
|
| 2554 |
/* |
| 2555 |
* Some of the devices in the bus can be in the runtime suspended |
| 2556 |
* state. Increment the usage count for all the devices in the dev_set |
| 2557 |
* before reset and decrement the same after reset. |
| 2558 |
*/ |
| 2559 |
if (!disable_idle_d3 && vfio_pci_dev_set_pm_runtime_get(dev_set)) |
| 2560 |
return; |
| 2561 |
|
| 2562 |
if (!pci_reset_bus(pdev)) |
| 2563 |
reset_done = true; |
| 2564 |
|
| 2565 |
list_for_each_entry(cur, &dev_set->device_list, vdev.dev_set_list) { |
| 2566 |
if (reset_done) |
| 2567 |
cur->needs_reset = false; |
| 2568 |
|
| 2569 |
if (!disable_idle_d3) |
| 2570 |
pm_runtime_put(&cur->pdev->dev); |
| 2571 |
} |
| 2572 |
} |
| 2573 |
|
| 2574 |
void vfio_pci_core_set_params(bool is_nointxmask, bool is_disable_vga, |
| 2575 |
bool is_disable_idle_d3) |
| 2576 |
{ |
| 2577 |
nointxmask = is_nointxmask; |
| 2578 |
disable_vga = is_disable_vga; |
| 2579 |
disable_idle_d3 = is_disable_idle_d3; |
| 2580 |
} |
| 2581 |
EXPORT_SYMBOL_GPL(vfio_pci_core_set_params); |
| 2582 |
|
| 2583 |
static void vfio_pci_core_cleanup(void) |
| 2584 |
{ |
| 2585 |
vfio_pci_uninit_perm_bits(); |
| 2586 |
} |
| 2587 |
|
| 2588 |
static int __init vfio_pci_core_init(void) |
| 2589 |
{ |
| 2590 |
/* Allocate shared config space permission data used by all devices */ |
| 2591 |
return vfio_pci_init_perm_bits(); |
| 2592 |
} |
| 2593 |
|
| 2594 |
module_init(vfio_pci_core_init); |
| 2595 |
module_exit(vfio_pci_core_cleanup); |
| 2596 |
|
| 2597 |
MODULE_LICENSE("GPL v2"); |
| 2598 |
MODULE_AUTHOR(DRIVER_AUTHOR); |
| 2599 |
MODULE_DESCRIPTION(DRIVER_DESC); |