| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* VFIO PCI interrupt handling |
| 4 |
* |
| 5 |
* Copyright (C) 2012 Red Hat, Inc. All rights reserved. |
| 6 |
* Author: Alex Williamson <[email protected]> |
| 7 |
* |
| 8 |
* Derived from original vfio: |
| 9 |
* Copyright 2010 Cisco Systems, Inc. All rights reserved. |
| 10 |
* Author: Tom Lyon, [email protected] |
| 11 |
*/ |
| 12 |
|
| 13 |
#include <linux/device.h> |
| 14 |
#include <linux/interrupt.h> |
| 15 |
#include <linux/eventfd.h> |
| 16 |
#include <linux/msi.h> |
| 17 |
#include <linux/pci.h> |
| 18 |
#include <linux/file.h> |
| 19 |
#include <linux/vfio.h> |
| 20 |
#include <linux/wait.h> |
| 21 |
#include <linux/slab.h> |
| 22 |
|
| 23 |
#include "vfio_pci_priv.h" |
| 24 |
|
| 25 |
struct vfio_pci_irq_ctx { |
| 26 |
struct vfio_pci_core_device *vdev; |
| 27 |
struct eventfd_ctx *trigger; |
| 28 |
struct virqfd *unmask; |
| 29 |
struct virqfd *mask; |
| 30 |
char *name; |
| 31 |
bool masked; |
| 32 |
struct irq_bypass_producer producer; |
| 33 |
}; |
| 34 |
|
| 35 |
static bool irq_is(struct vfio_pci_core_device *vdev, int type) |
| 36 |
{ |
| 37 |
return vdev->irq_type == type; |
| 38 |
} |
| 39 |
|
| 40 |
static bool is_intx(struct vfio_pci_core_device *vdev) |
| 41 |
{ |
| 42 |
return vdev->irq_type == VFIO_PCI_INTX_IRQ_INDEX; |
| 43 |
} |
| 44 |
|
| 45 |
static bool is_irq_none(struct vfio_pci_core_device *vdev) |
| 46 |
{ |
| 47 |
return !(vdev->irq_type == VFIO_PCI_INTX_IRQ_INDEX || |
| 48 |
vdev->irq_type == VFIO_PCI_MSI_IRQ_INDEX || |
| 49 |
vdev->irq_type == VFIO_PCI_MSIX_IRQ_INDEX); |
| 50 |
} |
| 51 |
|
| 52 |
static |
| 53 |
struct vfio_pci_irq_ctx *vfio_irq_ctx_get(struct vfio_pci_core_device *vdev, |
| 54 |
unsigned long index) |
| 55 |
{ |
| 56 |
return xa_load(&vdev->ctx, index); |
| 57 |
} |
| 58 |
|
| 59 |
static void vfio_irq_ctx_free(struct vfio_pci_core_device *vdev, |
| 60 |
struct vfio_pci_irq_ctx *ctx, unsigned long index) |
| 61 |
{ |
| 62 |
xa_erase(&vdev->ctx, index); |
| 63 |
kfree(ctx); |
| 64 |
} |
| 65 |
|
| 66 |
static struct vfio_pci_irq_ctx * |
| 67 |
vfio_irq_ctx_alloc(struct vfio_pci_core_device *vdev, unsigned long index) |
| 68 |
{ |
| 69 |
struct vfio_pci_irq_ctx *ctx; |
| 70 |
int ret; |
| 71 |
|
| 72 |
ctx = kzalloc_obj(*ctx, GFP_KERNEL_ACCOUNT); |
| 73 |
if (!ctx) |
| 74 |
return NULL; |
| 75 |
|
| 76 |
ret = xa_insert(&vdev->ctx, index, ctx, GFP_KERNEL_ACCOUNT); |
| 77 |
if (ret) { |
| 78 |
kfree(ctx); |
| 79 |
return NULL; |
| 80 |
} |
| 81 |
|
| 82 |
return ctx; |
| 83 |
} |
| 84 |
|
| 85 |
/* |
| 86 |
* INTx |
| 87 |
*/ |
| 88 |
static void vfio_send_intx_eventfd(void *opaque, void *data) |
| 89 |
{ |
| 90 |
struct vfio_pci_core_device *vdev = opaque; |
| 91 |
|
| 92 |
if (likely(is_intx(vdev) && !vdev->virq_disabled)) { |
| 93 |
struct vfio_pci_irq_ctx *ctx = data; |
| 94 |
struct eventfd_ctx *trigger = READ_ONCE(ctx->trigger); |
| 95 |
|
| 96 |
if (likely(trigger)) |
| 97 |
eventfd_signal(trigger); |
| 98 |
} |
| 99 |
} |
| 100 |
|
| 101 |
/* Returns true if the INTx vfio_pci_irq_ctx.masked value is changed. */ |
| 102 |
static bool __vfio_pci_intx_mask(struct vfio_pci_core_device *vdev) |
| 103 |
{ |
| 104 |
struct pci_dev *pdev = vdev->pdev; |
| 105 |
struct vfio_pci_irq_ctx *ctx; |
| 106 |
unsigned long flags; |
| 107 |
bool masked_changed = false; |
| 108 |
|
| 109 |
lockdep_assert_held(&vdev->igate); |
| 110 |
|
| 111 |
spin_lock_irqsave(&vdev->irqlock, flags); |
| 112 |
|
| 113 |
/* |
| 114 |
* Masking can come from interrupt, ioctl, or config space |
| 115 |
* via INTx disable. The latter means this can get called |
| 116 |
* even when not using intx delivery. In this case, just |
| 117 |
* try to have the physical bit follow the virtual bit. |
| 118 |
*/ |
| 119 |
if (unlikely(!is_intx(vdev))) { |
| 120 |
if (vdev->pci_2_3) |
| 121 |
pci_intx(pdev, 0); |
| 122 |
goto out_unlock; |
| 123 |
} |
| 124 |
|
| 125 |
ctx = vfio_irq_ctx_get(vdev, 0); |
| 126 |
if (WARN_ON_ONCE(!ctx)) |
| 127 |
goto out_unlock; |
| 128 |
|
| 129 |
if (!ctx->masked) { |
| 130 |
/* |
| 131 |
* Can't use check_and_mask here because we always want to |
| 132 |
* mask, not just when something is pending. |
| 133 |
*/ |
| 134 |
if (vdev->pci_2_3) |
| 135 |
pci_intx(pdev, 0); |
| 136 |
else |
| 137 |
disable_irq_nosync(pdev->irq); |
| 138 |
|
| 139 |
ctx->masked = true; |
| 140 |
masked_changed = true; |
| 141 |
} |
| 142 |
|
| 143 |
out_unlock: |
| 144 |
spin_unlock_irqrestore(&vdev->irqlock, flags); |
| 145 |
return masked_changed; |
| 146 |
} |
| 147 |
|
| 148 |
bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev) |
| 149 |
{ |
| 150 |
bool mask_changed; |
| 151 |
|
| 152 |
mutex_lock(&vdev->igate); |
| 153 |
mask_changed = __vfio_pci_intx_mask(vdev); |
| 154 |
mutex_unlock(&vdev->igate); |
| 155 |
|
| 156 |
return mask_changed; |
| 157 |
} |
| 158 |
|
| 159 |
/* |
| 160 |
* If this is triggered by an eventfd, we can't call eventfd_signal |
| 161 |
* or else we'll deadlock on the eventfd wait queue. Return >0 when |
| 162 |
* a signal is necessary, which can then be handled via a work queue |
| 163 |
* or directly depending on the caller. |
| 164 |
*/ |
| 165 |
static int vfio_pci_intx_unmask_handler(void *opaque, void *data) |
| 166 |
{ |
| 167 |
struct vfio_pci_core_device *vdev = opaque; |
| 168 |
struct pci_dev *pdev = vdev->pdev; |
| 169 |
struct vfio_pci_irq_ctx *ctx = data; |
| 170 |
unsigned long flags; |
| 171 |
int ret = 0; |
| 172 |
|
| 173 |
spin_lock_irqsave(&vdev->irqlock, flags); |
| 174 |
|
| 175 |
/* |
| 176 |
* Unmasking comes from ioctl or config, so again, have the |
| 177 |
* physical bit follow the virtual even when not using INTx. |
| 178 |
*/ |
| 179 |
if (unlikely(!is_intx(vdev))) { |
| 180 |
if (vdev->pci_2_3) |
| 181 |
pci_intx(pdev, 1); |
| 182 |
goto out_unlock; |
| 183 |
} |
| 184 |
|
| 185 |
if (ctx->masked && !vdev->virq_disabled) { |
| 186 |
/* |
| 187 |
* A pending interrupt here would immediately trigger, |
| 188 |
* but we can avoid that overhead by just re-sending |
| 189 |
* the interrupt to the user. |
| 190 |
*/ |
| 191 |
if (vdev->pci_2_3) { |
| 192 |
if (!pci_check_and_unmask_intx(pdev)) |
| 193 |
ret = 1; |
| 194 |
} else |
| 195 |
enable_irq(pdev->irq); |
| 196 |
|
| 197 |
ctx->masked = (ret > 0); |
| 198 |
} |
| 199 |
|
| 200 |
out_unlock: |
| 201 |
spin_unlock_irqrestore(&vdev->irqlock, flags); |
| 202 |
|
| 203 |
return ret; |
| 204 |
} |
| 205 |
|
| 206 |
static void __vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev) |
| 207 |
{ |
| 208 |
struct vfio_pci_irq_ctx *ctx = vfio_irq_ctx_get(vdev, 0); |
| 209 |
|
| 210 |
lockdep_assert_held(&vdev->igate); |
| 211 |
|
| 212 |
if (vfio_pci_intx_unmask_handler(vdev, ctx) > 0) |
| 213 |
vfio_send_intx_eventfd(vdev, ctx); |
| 214 |
} |
| 215 |
|
| 216 |
void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev) |
| 217 |
{ |
| 218 |
mutex_lock(&vdev->igate); |
| 219 |
__vfio_pci_intx_unmask(vdev); |
| 220 |
mutex_unlock(&vdev->igate); |
| 221 |
} |
| 222 |
|
| 223 |
static irqreturn_t vfio_intx_handler(int irq, void *dev_id) |
| 224 |
{ |
| 225 |
struct vfio_pci_irq_ctx *ctx = dev_id; |
| 226 |
struct vfio_pci_core_device *vdev = ctx->vdev; |
| 227 |
unsigned long flags; |
| 228 |
int ret = IRQ_NONE; |
| 229 |
|
| 230 |
spin_lock_irqsave(&vdev->irqlock, flags); |
| 231 |
|
| 232 |
if (!vdev->pci_2_3) { |
| 233 |
disable_irq_nosync(vdev->pdev->irq); |
| 234 |
ctx->masked = true; |
| 235 |
ret = IRQ_HANDLED; |
| 236 |
} else if (!ctx->masked && /* may be shared */ |
| 237 |
pci_check_and_mask_intx(vdev->pdev)) { |
| 238 |
ctx->masked = true; |
| 239 |
ret = IRQ_HANDLED; |
| 240 |
} |
| 241 |
|
| 242 |
spin_unlock_irqrestore(&vdev->irqlock, flags); |
| 243 |
|
| 244 |
if (ret == IRQ_HANDLED) |
| 245 |
vfio_send_intx_eventfd(vdev, ctx); |
| 246 |
|
| 247 |
return ret; |
| 248 |
} |
| 249 |
|
| 250 |
static int vfio_intx_enable(struct vfio_pci_core_device *vdev, |
| 251 |
struct eventfd_ctx *trigger) |
| 252 |
{ |
| 253 |
struct pci_dev *pdev = vdev->pdev; |
| 254 |
struct vfio_pci_irq_ctx *ctx; |
| 255 |
unsigned long irqflags; |
| 256 |
char *name; |
| 257 |
int ret; |
| 258 |
|
| 259 |
if (!is_irq_none(vdev)) |
| 260 |
return -EINVAL; |
| 261 |
|
| 262 |
if (!pdev->irq || pdev->irq == IRQ_NOTCONNECTED) |
| 263 |
return -ENODEV; |
| 264 |
|
| 265 |
name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-intx(%s)", pci_name(pdev)); |
| 266 |
if (!name) |
| 267 |
return -ENOMEM; |
| 268 |
|
| 269 |
ctx = vfio_irq_ctx_alloc(vdev, 0); |
| 270 |
if (!ctx) { |
| 271 |
kfree(name); |
| 272 |
return -ENOMEM; |
| 273 |
} |
| 274 |
|
| 275 |
ctx->name = name; |
| 276 |
ctx->trigger = trigger; |
| 277 |
ctx->vdev = vdev; |
| 278 |
|
| 279 |
/* |
| 280 |
* Fill the initial masked state based on virq_disabled. After |
| 281 |
* enable, changing the DisINTx bit in vconfig directly changes INTx |
| 282 |
* masking. igate prevents races during setup, once running masked |
| 283 |
* is protected via irqlock. |
| 284 |
* |
| 285 |
* Devices supporting DisINTx also reflect the current mask state in |
| 286 |
* the physical DisINTx bit, which is not affected during IRQ setup. |
| 287 |
* |
| 288 |
* Devices without DisINTx support require an exclusive interrupt. |
| 289 |
* IRQ masking is performed at the IRQ chip. Again, igate protects |
| 290 |
* against races during setup and IRQ handlers and irqfds are not |
| 291 |
* yet active, therefore masked is stable and can be used to |
| 292 |
* conditionally auto-enable the IRQ. |
| 293 |
* |
| 294 |
* irq_type must be stable while the IRQ handler is registered, |
| 295 |
* therefore it must be set before request_irq(). |
| 296 |
*/ |
| 297 |
ctx->masked = vdev->virq_disabled; |
| 298 |
if (vdev->pci_2_3) { |
| 299 |
pci_intx(pdev, !ctx->masked); |
| 300 |
irqflags = IRQF_SHARED; |
| 301 |
} else { |
| 302 |
irqflags = ctx->masked ? IRQF_NO_AUTOEN : 0; |
| 303 |
} |
| 304 |
|
| 305 |
vdev->irq_type = VFIO_PCI_INTX_IRQ_INDEX; |
| 306 |
|
| 307 |
if (!vdev->pci_2_3) |
| 308 |
irq_set_status_flags(pdev->irq, IRQ_DISABLE_UNLAZY); |
| 309 |
|
| 310 |
ret = request_irq(pdev->irq, vfio_intx_handler, |
| 311 |
irqflags, ctx->name, ctx); |
| 312 |
if (ret) { |
| 313 |
if (!vdev->pci_2_3) |
| 314 |
irq_clear_status_flags(pdev->irq, IRQ_DISABLE_UNLAZY); |
| 315 |
vdev->irq_type = VFIO_PCI_NUM_IRQS; |
| 316 |
kfree(name); |
| 317 |
vfio_irq_ctx_free(vdev, ctx, 0); |
| 318 |
return ret; |
| 319 |
} |
| 320 |
|
| 321 |
return 0; |
| 322 |
} |
| 323 |
|
| 324 |
static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev, |
| 325 |
struct eventfd_ctx *trigger) |
| 326 |
{ |
| 327 |
struct pci_dev *pdev = vdev->pdev; |
| 328 |
struct vfio_pci_irq_ctx *ctx; |
| 329 |
struct eventfd_ctx *old; |
| 330 |
|
| 331 |
ctx = vfio_irq_ctx_get(vdev, 0); |
| 332 |
if (WARN_ON_ONCE(!ctx)) |
| 333 |
return -EINVAL; |
| 334 |
|
| 335 |
old = ctx->trigger; |
| 336 |
|
| 337 |
WRITE_ONCE(ctx->trigger, trigger); |
| 338 |
|
| 339 |
/* Releasing an old ctx requires synchronizing in-flight users */ |
| 340 |
if (old) { |
| 341 |
synchronize_irq(pdev->irq); |
| 342 |
vfio_virqfd_flush_thread(&ctx->unmask); |
| 343 |
eventfd_ctx_put(old); |
| 344 |
} |
| 345 |
|
| 346 |
return 0; |
| 347 |
} |
| 348 |
|
| 349 |
static void vfio_intx_disable(struct vfio_pci_core_device *vdev) |
| 350 |
{ |
| 351 |
struct pci_dev *pdev = vdev->pdev; |
| 352 |
struct vfio_pci_irq_ctx *ctx; |
| 353 |
|
| 354 |
ctx = vfio_irq_ctx_get(vdev, 0); |
| 355 |
WARN_ON_ONCE(!ctx); |
| 356 |
if (ctx) { |
| 357 |
vfio_virqfd_disable(&ctx->unmask); |
| 358 |
vfio_virqfd_disable(&ctx->mask); |
| 359 |
free_irq(pdev->irq, ctx); |
| 360 |
if (!vdev->pci_2_3) |
| 361 |
irq_clear_status_flags(pdev->irq, IRQ_DISABLE_UNLAZY); |
| 362 |
if (ctx->trigger) |
| 363 |
eventfd_ctx_put(ctx->trigger); |
| 364 |
kfree(ctx->name); |
| 365 |
vfio_irq_ctx_free(vdev, ctx, 0); |
| 366 |
} |
| 367 |
vdev->irq_type = VFIO_PCI_NUM_IRQS; |
| 368 |
} |
| 369 |
|
| 370 |
/* |
| 371 |
* MSI/MSI-X |
| 372 |
*/ |
| 373 |
static irqreturn_t vfio_msihandler(int irq, void *arg) |
| 374 |
{ |
| 375 |
struct eventfd_ctx *trigger = arg; |
| 376 |
|
| 377 |
eventfd_signal(trigger); |
| 378 |
return IRQ_HANDLED; |
| 379 |
} |
| 380 |
|
| 381 |
static int vfio_msi_enable(struct vfio_pci_core_device *vdev, int nvec, bool msix) |
| 382 |
{ |
| 383 |
struct pci_dev *pdev = vdev->pdev; |
| 384 |
unsigned int flag = msix ? PCI_IRQ_MSIX : PCI_IRQ_MSI; |
| 385 |
int ret; |
| 386 |
u16 cmd; |
| 387 |
|
| 388 |
if (!is_irq_none(vdev)) |
| 389 |
return -EINVAL; |
| 390 |
|
| 391 |
/* return the number of supported vectors if we can't get all: */ |
| 392 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 393 |
ret = pci_alloc_irq_vectors(pdev, 1, nvec, flag); |
| 394 |
if (ret < nvec) { |
| 395 |
if (ret > 0) |
| 396 |
pci_free_irq_vectors(pdev); |
| 397 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 398 |
return ret; |
| 399 |
} |
| 400 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 401 |
|
| 402 |
vdev->irq_type = msix ? VFIO_PCI_MSIX_IRQ_INDEX : |
| 403 |
VFIO_PCI_MSI_IRQ_INDEX; |
| 404 |
|
| 405 |
if (!msix) { |
| 406 |
/* |
| 407 |
* Compute the virtual hardware field for max msi vectors - |
| 408 |
* it is the log base 2 of the number of vectors. |
| 409 |
*/ |
| 410 |
vdev->msi_qmax = fls(nvec * 2 - 1) - 1; |
| 411 |
} |
| 412 |
|
| 413 |
return 0; |
| 414 |
} |
| 415 |
|
| 416 |
/* |
| 417 |
* vfio_msi_alloc_irq() returns the Linux IRQ number of an MSI or MSI-X device |
| 418 |
* interrupt vector. If a Linux IRQ number is not available then a new |
| 419 |
* interrupt is allocated if dynamic MSI-X is supported. |
| 420 |
* |
| 421 |
* Where is vfio_msi_free_irq()? Allocated interrupts are maintained, |
| 422 |
* essentially forming a cache that subsequent allocations can draw from. |
| 423 |
* Interrupts are freed using pci_free_irq_vectors() when MSI/MSI-X is |
| 424 |
* disabled. |
| 425 |
*/ |
| 426 |
static int vfio_msi_alloc_irq(struct vfio_pci_core_device *vdev, |
| 427 |
unsigned int vector, bool msix) |
| 428 |
{ |
| 429 |
struct pci_dev *pdev = vdev->pdev; |
| 430 |
struct msi_map map; |
| 431 |
int irq; |
| 432 |
u16 cmd; |
| 433 |
|
| 434 |
irq = pci_irq_vector(pdev, vector); |
| 435 |
if (WARN_ON_ONCE(irq == 0)) |
| 436 |
return -EINVAL; |
| 437 |
if (irq > 0 || !msix || !vdev->has_dyn_msix) |
| 438 |
return irq; |
| 439 |
|
| 440 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 441 |
map = pci_msix_alloc_irq_at(pdev, vector, NULL); |
| 442 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 443 |
|
| 444 |
return map.index < 0 ? map.index : map.virq; |
| 445 |
} |
| 446 |
|
| 447 |
static int vfio_msi_set_vector_signal(struct vfio_pci_core_device *vdev, |
| 448 |
unsigned int vector, int fd, bool msix) |
| 449 |
{ |
| 450 |
struct pci_dev *pdev = vdev->pdev; |
| 451 |
struct vfio_pci_irq_ctx *ctx; |
| 452 |
struct eventfd_ctx *trigger; |
| 453 |
int irq = -EINVAL, ret; |
| 454 |
u16 cmd; |
| 455 |
|
| 456 |
ctx = vfio_irq_ctx_get(vdev, vector); |
| 457 |
|
| 458 |
if (ctx) { |
| 459 |
irq_bypass_unregister_producer(&ctx->producer); |
| 460 |
irq = pci_irq_vector(pdev, vector); |
| 461 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 462 |
free_irq(irq, ctx->trigger); |
| 463 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 464 |
/* Interrupt stays allocated, will be freed at MSI-X disable. */ |
| 465 |
kfree(ctx->name); |
| 466 |
eventfd_ctx_put(ctx->trigger); |
| 467 |
vfio_irq_ctx_free(vdev, ctx, vector); |
| 468 |
} |
| 469 |
|
| 470 |
if (fd < 0) |
| 471 |
return 0; |
| 472 |
|
| 473 |
if (irq == -EINVAL) { |
| 474 |
/* Interrupt stays allocated, will be freed at MSI-X disable. */ |
| 475 |
irq = vfio_msi_alloc_irq(vdev, vector, msix); |
| 476 |
if (irq < 0) |
| 477 |
return irq; |
| 478 |
} |
| 479 |
|
| 480 |
ctx = vfio_irq_ctx_alloc(vdev, vector); |
| 481 |
if (!ctx) |
| 482 |
return -ENOMEM; |
| 483 |
|
| 484 |
ctx->name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-msi%s[%d](%s)", |
| 485 |
msix ? "x" : "", vector, pci_name(pdev)); |
| 486 |
if (!ctx->name) { |
| 487 |
ret = -ENOMEM; |
| 488 |
goto out_free_ctx; |
| 489 |
} |
| 490 |
|
| 491 |
trigger = eventfd_ctx_fdget(fd); |
| 492 |
if (IS_ERR(trigger)) { |
| 493 |
ret = PTR_ERR(trigger); |
| 494 |
goto out_free_name; |
| 495 |
} |
| 496 |
|
| 497 |
/* |
| 498 |
* If the vector was previously allocated, refresh the on-device |
| 499 |
* message data before enabling in case it had been cleared or |
| 500 |
* corrupted (e.g. due to backdoor resets) since writing. |
| 501 |
*/ |
| 502 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 503 |
if (msix) { |
| 504 |
struct msi_msg msg; |
| 505 |
|
| 506 |
get_cached_msi_msg(irq, &msg); |
| 507 |
pci_write_msi_msg(irq, &msg); |
| 508 |
} |
| 509 |
|
| 510 |
ret = request_irq(irq, vfio_msihandler, 0, ctx->name, trigger); |
| 511 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 512 |
if (ret) |
| 513 |
goto out_put_eventfd_ctx; |
| 514 |
|
| 515 |
ret = irq_bypass_register_producer(&ctx->producer, trigger, irq); |
| 516 |
if (unlikely(ret)) { |
| 517 |
dev_info(&pdev->dev, |
| 518 |
"irq bypass producer (eventfd %p) registration fails: %d\n", |
| 519 |
trigger, ret); |
| 520 |
} |
| 521 |
ctx->trigger = trigger; |
| 522 |
|
| 523 |
return 0; |
| 524 |
|
| 525 |
out_put_eventfd_ctx: |
| 526 |
eventfd_ctx_put(trigger); |
| 527 |
out_free_name: |
| 528 |
kfree(ctx->name); |
| 529 |
out_free_ctx: |
| 530 |
vfio_irq_ctx_free(vdev, ctx, vector); |
| 531 |
return ret; |
| 532 |
} |
| 533 |
|
| 534 |
static int vfio_msi_set_block(struct vfio_pci_core_device *vdev, unsigned start, |
| 535 |
unsigned count, int32_t *fds, bool msix) |
| 536 |
{ |
| 537 |
unsigned int i, j; |
| 538 |
int ret = 0; |
| 539 |
|
| 540 |
for (i = 0, j = start; i < count && !ret; i++, j++) { |
| 541 |
int fd = fds ? fds[i] : -1; |
| 542 |
ret = vfio_msi_set_vector_signal(vdev, j, fd, msix); |
| 543 |
} |
| 544 |
|
| 545 |
if (ret) { |
| 546 |
for (i = start; i < j; i++) |
| 547 |
vfio_msi_set_vector_signal(vdev, i, -1, msix); |
| 548 |
} |
| 549 |
|
| 550 |
return ret; |
| 551 |
} |
| 552 |
|
| 553 |
static void vfio_msi_disable(struct vfio_pci_core_device *vdev, bool msix) |
| 554 |
{ |
| 555 |
struct pci_dev *pdev = vdev->pdev; |
| 556 |
struct vfio_pci_irq_ctx *ctx; |
| 557 |
unsigned long i; |
| 558 |
u16 cmd; |
| 559 |
|
| 560 |
xa_for_each(&vdev->ctx, i, ctx) { |
| 561 |
vfio_virqfd_disable(&ctx->unmask); |
| 562 |
vfio_virqfd_disable(&ctx->mask); |
| 563 |
vfio_msi_set_vector_signal(vdev, i, -1, msix); |
| 564 |
} |
| 565 |
|
| 566 |
cmd = vfio_pci_memory_lock_and_enable(vdev); |
| 567 |
pci_free_irq_vectors(pdev); |
| 568 |
vfio_pci_memory_unlock_and_restore(vdev, cmd); |
| 569 |
|
| 570 |
/* |
| 571 |
* Both disable paths above use pci_intx_for_msi() to clear DisINTx |
| 572 |
* via their shutdown paths. Restore for NoINTx devices. |
| 573 |
*/ |
| 574 |
if (vdev->nointx) |
| 575 |
pci_intx(pdev, 0); |
| 576 |
|
| 577 |
vdev->irq_type = VFIO_PCI_NUM_IRQS; |
| 578 |
} |
| 579 |
|
| 580 |
/* |
| 581 |
* IOCTL support |
| 582 |
*/ |
| 583 |
static int vfio_pci_set_intx_unmask(struct vfio_pci_core_device *vdev, |
| 584 |
unsigned index, unsigned start, |
| 585 |
unsigned count, uint32_t flags, void *data) |
| 586 |
{ |
| 587 |
if (!is_intx(vdev) || start != 0 || count != 1) |
| 588 |
return -EINVAL; |
| 589 |
|
| 590 |
if (flags & VFIO_IRQ_SET_DATA_NONE) { |
| 591 |
__vfio_pci_intx_unmask(vdev); |
| 592 |
} else if (flags & VFIO_IRQ_SET_DATA_BOOL) { |
| 593 |
uint8_t unmask = *(uint8_t *)data; |
| 594 |
if (unmask) |
| 595 |
__vfio_pci_intx_unmask(vdev); |
| 596 |
} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) { |
| 597 |
struct vfio_pci_irq_ctx *ctx = vfio_irq_ctx_get(vdev, 0); |
| 598 |
int32_t fd = *(int32_t *)data; |
| 599 |
|
| 600 |
if (WARN_ON_ONCE(!ctx)) |
| 601 |
return -EINVAL; |
| 602 |
if (fd >= 0) |
| 603 |
return vfio_virqfd_enable((void *) vdev, |
| 604 |
vfio_pci_intx_unmask_handler, |
| 605 |
vfio_send_intx_eventfd, ctx, |
| 606 |
&ctx->unmask, fd); |
| 607 |
|
| 608 |
vfio_virqfd_disable(&ctx->unmask); |
| 609 |
} |
| 610 |
|
| 611 |
return 0; |
| 612 |
} |
| 613 |
|
| 614 |
static int vfio_pci_set_intx_mask(struct vfio_pci_core_device *vdev, |
| 615 |
unsigned index, unsigned start, |
| 616 |
unsigned count, uint32_t flags, void *data) |
| 617 |
{ |
| 618 |
if (!is_intx(vdev) || start != 0 || count != 1) |
| 619 |
return -EINVAL; |
| 620 |
|
| 621 |
if (flags & VFIO_IRQ_SET_DATA_NONE) { |
| 622 |
__vfio_pci_intx_mask(vdev); |
| 623 |
} else if (flags & VFIO_IRQ_SET_DATA_BOOL) { |
| 624 |
uint8_t mask = *(uint8_t *)data; |
| 625 |
if (mask) |
| 626 |
__vfio_pci_intx_mask(vdev); |
| 627 |
} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) { |
| 628 |
return -ENOTTY; /* XXX implement me */ |
| 629 |
} |
| 630 |
|
| 631 |
return 0; |
| 632 |
} |
| 633 |
|
| 634 |
static int vfio_pci_set_intx_trigger(struct vfio_pci_core_device *vdev, |
| 635 |
unsigned index, unsigned start, |
| 636 |
unsigned count, uint32_t flags, void *data) |
| 637 |
{ |
| 638 |
if (is_intx(vdev) && !count && (flags & VFIO_IRQ_SET_DATA_NONE)) { |
| 639 |
vfio_intx_disable(vdev); |
| 640 |
return 0; |
| 641 |
} |
| 642 |
|
| 643 |
if (!(is_intx(vdev) || is_irq_none(vdev)) || start != 0 || count != 1) |
| 644 |
return -EINVAL; |
| 645 |
|
| 646 |
if (flags & VFIO_IRQ_SET_DATA_EVENTFD) { |
| 647 |
struct eventfd_ctx *trigger = NULL; |
| 648 |
int32_t fd = *(int32_t *)data; |
| 649 |
int ret; |
| 650 |
|
| 651 |
if (fd >= 0) { |
| 652 |
trigger = eventfd_ctx_fdget(fd); |
| 653 |
if (IS_ERR(trigger)) |
| 654 |
return PTR_ERR(trigger); |
| 655 |
} |
| 656 |
|
| 657 |
if (is_intx(vdev)) |
| 658 |
ret = vfio_intx_set_signal(vdev, trigger); |
| 659 |
else |
| 660 |
ret = vfio_intx_enable(vdev, trigger); |
| 661 |
|
| 662 |
if (ret && trigger) |
| 663 |
eventfd_ctx_put(trigger); |
| 664 |
|
| 665 |
return ret; |
| 666 |
} |
| 667 |
|
| 668 |
if (!is_intx(vdev)) |
| 669 |
return -EINVAL; |
| 670 |
|
| 671 |
if (flags & VFIO_IRQ_SET_DATA_NONE) { |
| 672 |
vfio_send_intx_eventfd(vdev, vfio_irq_ctx_get(vdev, 0)); |
| 673 |
} else if (flags & VFIO_IRQ_SET_DATA_BOOL) { |
| 674 |
uint8_t trigger = *(uint8_t *)data; |
| 675 |
if (trigger) |
| 676 |
vfio_send_intx_eventfd(vdev, vfio_irq_ctx_get(vdev, 0)); |
| 677 |
} |
| 678 |
return 0; |
| 679 |
} |
| 680 |
|
| 681 |
static int vfio_pci_set_msi_trigger(struct vfio_pci_core_device *vdev, |
| 682 |
unsigned index, unsigned start, |
| 683 |
unsigned count, uint32_t flags, void *data) |
| 684 |
{ |
| 685 |
struct vfio_pci_irq_ctx *ctx; |
| 686 |
unsigned int i; |
| 687 |
bool msix = (index == VFIO_PCI_MSIX_IRQ_INDEX); |
| 688 |
|
| 689 |
if (irq_is(vdev, index) && !count && (flags & VFIO_IRQ_SET_DATA_NONE)) { |
| 690 |
vfio_msi_disable(vdev, msix); |
| 691 |
return 0; |
| 692 |
} |
| 693 |
|
| 694 |
if (!(irq_is(vdev, index) || is_irq_none(vdev))) |
| 695 |
return -EINVAL; |
| 696 |
|
| 697 |
if (flags & VFIO_IRQ_SET_DATA_EVENTFD) { |
| 698 |
int32_t *fds = data; |
| 699 |
int ret; |
| 700 |
|
| 701 |
if (vdev->irq_type == index) |
| 702 |
return vfio_msi_set_block(vdev, start, count, |
| 703 |
fds, msix); |
| 704 |
|
| 705 |
ret = vfio_msi_enable(vdev, start + count, msix); |
| 706 |
if (ret) |
| 707 |
return ret; |
| 708 |
|
| 709 |
ret = vfio_msi_set_block(vdev, start, count, fds, msix); |
| 710 |
if (ret) |
| 711 |
vfio_msi_disable(vdev, msix); |
| 712 |
|
| 713 |
return ret; |
| 714 |
} |
| 715 |
|
| 716 |
if (!irq_is(vdev, index)) |
| 717 |
return -EINVAL; |
| 718 |
|
| 719 |
for (i = start; i < start + count; i++) { |
| 720 |
ctx = vfio_irq_ctx_get(vdev, i); |
| 721 |
if (!ctx) |
| 722 |
continue; |
| 723 |
if (flags & VFIO_IRQ_SET_DATA_NONE) { |
| 724 |
eventfd_signal(ctx->trigger); |
| 725 |
} else if (flags & VFIO_IRQ_SET_DATA_BOOL) { |
| 726 |
uint8_t *bools = data; |
| 727 |
if (bools[i - start]) |
| 728 |
eventfd_signal(ctx->trigger); |
| 729 |
} |
| 730 |
} |
| 731 |
return 0; |
| 732 |
} |
| 733 |
|
| 734 |
static int vfio_pci_set_ctx_trigger_single(struct vfio_pci_core_device *vdev, |
| 735 |
struct vfio_pci_eventfd __rcu **peventfd, |
| 736 |
unsigned int count, uint32_t flags, |
| 737 |
void *data) |
| 738 |
{ |
| 739 |
/* DATA_NONE/DATA_BOOL enables loopback testing */ |
| 740 |
if (flags & VFIO_IRQ_SET_DATA_NONE) { |
| 741 |
struct vfio_pci_eventfd *eventfd; |
| 742 |
|
| 743 |
eventfd = rcu_dereference_protected(*peventfd, |
| 744 |
lockdep_is_held(&vdev->igate)); |
| 745 |
|
| 746 |
if (!eventfd) |
| 747 |
return -EINVAL; |
| 748 |
|
| 749 |
if (count) { |
| 750 |
eventfd_signal(eventfd->ctx); |
| 751 |
return 0; |
| 752 |
} |
| 753 |
|
| 754 |
return vfio_pci_eventfd_replace_locked(vdev, peventfd, NULL); |
| 755 |
} else if (flags & VFIO_IRQ_SET_DATA_BOOL) { |
| 756 |
uint8_t trigger; |
| 757 |
|
| 758 |
if (!count) |
| 759 |
return -EINVAL; |
| 760 |
|
| 761 |
trigger = *(uint8_t *)data; |
| 762 |
|
| 763 |
if (trigger) { |
| 764 |
struct vfio_pci_eventfd *eventfd = |
| 765 |
rcu_dereference_protected(*peventfd, |
| 766 |
lockdep_is_held(&vdev->igate)); |
| 767 |
|
| 768 |
if (eventfd) |
| 769 |
eventfd_signal(eventfd->ctx); |
| 770 |
} |
| 771 |
|
| 772 |
return 0; |
| 773 |
} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) { |
| 774 |
int32_t fd; |
| 775 |
|
| 776 |
if (!count) |
| 777 |
return -EINVAL; |
| 778 |
|
| 779 |
fd = *(int32_t *)data; |
| 780 |
if (fd == -1) { |
| 781 |
return vfio_pci_eventfd_replace_locked(vdev, |
| 782 |
peventfd, NULL); |
| 783 |
} else if (fd >= 0) { |
| 784 |
struct eventfd_ctx *efdctx; |
| 785 |
int ret; |
| 786 |
|
| 787 |
efdctx = eventfd_ctx_fdget(fd); |
| 788 |
if (IS_ERR(efdctx)) |
| 789 |
return PTR_ERR(efdctx); |
| 790 |
|
| 791 |
ret = vfio_pci_eventfd_replace_locked(vdev, |
| 792 |
peventfd, efdctx); |
| 793 |
if (ret) |
| 794 |
eventfd_ctx_put(efdctx); |
| 795 |
|
| 796 |
return ret; |
| 797 |
} |
| 798 |
} |
| 799 |
|
| 800 |
return -EINVAL; |
| 801 |
} |
| 802 |
|
| 803 |
static int vfio_pci_set_err_trigger(struct vfio_pci_core_device *vdev, |
| 804 |
unsigned index, unsigned start, |
| 805 |
unsigned count, uint32_t flags, void *data) |
| 806 |
{ |
| 807 |
if (index != VFIO_PCI_ERR_IRQ_INDEX || start != 0 || count > 1) |
| 808 |
return -EINVAL; |
| 809 |
|
| 810 |
return vfio_pci_set_ctx_trigger_single(vdev, &vdev->err_trigger, |
| 811 |
count, flags, data); |
| 812 |
} |
| 813 |
|
| 814 |
static int vfio_pci_set_req_trigger(struct vfio_pci_core_device *vdev, |
| 815 |
unsigned index, unsigned start, |
| 816 |
unsigned count, uint32_t flags, void *data) |
| 817 |
{ |
| 818 |
if (index != VFIO_PCI_REQ_IRQ_INDEX || start != 0 || count > 1) |
| 819 |
return -EINVAL; |
| 820 |
|
| 821 |
return vfio_pci_set_ctx_trigger_single(vdev, &vdev->req_trigger, |
| 822 |
count, flags, data); |
| 823 |
} |
| 824 |
|
| 825 |
int vfio_pci_set_irqs_ioctl(struct vfio_pci_core_device *vdev, uint32_t flags, |
| 826 |
unsigned index, unsigned start, unsigned count, |
| 827 |
void *data) |
| 828 |
{ |
| 829 |
int (*func)(struct vfio_pci_core_device *vdev, unsigned index, |
| 830 |
unsigned start, unsigned count, uint32_t flags, |
| 831 |
void *data) = NULL; |
| 832 |
|
| 833 |
switch (index) { |
| 834 |
case VFIO_PCI_INTX_IRQ_INDEX: |
| 835 |
switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) { |
| 836 |
case VFIO_IRQ_SET_ACTION_MASK: |
| 837 |
func = vfio_pci_set_intx_mask; |
| 838 |
break; |
| 839 |
case VFIO_IRQ_SET_ACTION_UNMASK: |
| 840 |
func = vfio_pci_set_intx_unmask; |
| 841 |
break; |
| 842 |
case VFIO_IRQ_SET_ACTION_TRIGGER: |
| 843 |
func = vfio_pci_set_intx_trigger; |
| 844 |
break; |
| 845 |
} |
| 846 |
break; |
| 847 |
case VFIO_PCI_MSI_IRQ_INDEX: |
| 848 |
case VFIO_PCI_MSIX_IRQ_INDEX: |
| 849 |
switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) { |
| 850 |
case VFIO_IRQ_SET_ACTION_MASK: |
| 851 |
case VFIO_IRQ_SET_ACTION_UNMASK: |
| 852 |
/* XXX Need masking support exported */ |
| 853 |
break; |
| 854 |
case VFIO_IRQ_SET_ACTION_TRIGGER: |
| 855 |
func = vfio_pci_set_msi_trigger; |
| 856 |
break; |
| 857 |
} |
| 858 |
break; |
| 859 |
case VFIO_PCI_ERR_IRQ_INDEX: |
| 860 |
switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) { |
| 861 |
case VFIO_IRQ_SET_ACTION_TRIGGER: |
| 862 |
if (pci_is_pcie(vdev->pdev)) |
| 863 |
func = vfio_pci_set_err_trigger; |
| 864 |
break; |
| 865 |
} |
| 866 |
break; |
| 867 |
case VFIO_PCI_REQ_IRQ_INDEX: |
| 868 |
switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) { |
| 869 |
case VFIO_IRQ_SET_ACTION_TRIGGER: |
| 870 |
func = vfio_pci_set_req_trigger; |
| 871 |
break; |
| 872 |
} |
| 873 |
break; |
| 874 |
} |
| 875 |
|
| 876 |
if (!func) |
| 877 |
return -ENOTTY; |
| 878 |
|
| 879 |
return func(vdev, index, start, count, flags, data); |
| 880 |
} |