drivers/staging/android/vsoc.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
vsoc_validate_inode
vsoc_validate_filep
shm_off_to_virtual_addr
shm_off_to_phys_addr
vsoc_region_from_inode
vsoc_region_from_filep
vsoc_device_region_size
do_create_fd_scoped_permission
do_destroy_fd_scoped_permission_node
do_destroy_fd_scoped_permission
do_vsoc_describe_region
handle_vsoc_cond_wait
do_vsoc_cond_wait
do_vsoc_cond_wake
vsoc_ioctl
vsoc_read
vsoc_lseek
vsoc_write
vsoc_interrupt
vsoc_probe_device
vsoc_remove_device
vsoc_cleanup_module
vsoc_init_module
vsoc_open
vsoc_release
vsoc_get_area
vsoc_mmap
   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * drivers/android/staging/vsoc.c
   4  *
   5  * Android Virtual System on a Chip (VSoC) driver
   6  *
   7  * Copyright (C) 2017 Google, Inc.
   8  *
   9  * Author: ghartman@google.com
  10  *
  11  * Based on drivers/char/kvm_ivshmem.c - driver for KVM Inter-VM shared memory
  12  *         Copyright 2009 Cam Macdonell <cam@cs.ualberta.ca>
  13  *
  14  * Based on cirrusfb.c and 8139cp.c:
  15  *   Copyright 1999-2001 Jeff Garzik
  16  *   Copyright 2001-2004 Jeff Garzik
  17  */
  18 
  19 #include <linux/dma-mapping.h>
  20 #include <linux/freezer.h>
  21 #include <linux/futex.h>
  22 #include <linux/init.h>
  23 #include <linux/kernel.h>
  24 #include <linux/module.h>
  25 #include <linux/mutex.h>
  26 #include <linux/pci.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/sched.h>
  29 #include <linux/syscalls.h>
  30 #include <linux/uaccess.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/cdev.h>
  33 #include <linux/file.h>
  34 #include "uapi/vsoc_shm.h"
  35 
  36 #define VSOC_DEV_NAME "vsoc"
  37 
  38 /*
  39  * Description of the ivshmem-doorbell PCI device used by QEmu. These
  40  * constants follow docs/specs/ivshmem-spec.txt, which can be found in
  41  * the QEmu repository. This was last reconciled with the version that
  42  * came out with 2.8
  43  */
  44 
  45 /*
  46  * These constants are determined KVM Inter-VM shared memory device
  47  * register offsets
  48  */
  49 enum {
  50         INTR_MASK = 0x00,       /* Interrupt Mask */
  51         INTR_STATUS = 0x04,     /* Interrupt Status */
  52         IV_POSITION = 0x08,     /* VM ID */
  53         DOORBELL = 0x0c,        /* Doorbell */
  54 };
  55 
  56 static const int REGISTER_BAR;  /* Equal to 0 */
  57 static const int MAX_REGISTER_BAR_LEN = 0x100;
  58 /*
  59  * The MSI-x BAR is not used directly.
  60  *
  61  * static const int MSI_X_BAR = 1;
  62  */
  63 static const int SHARED_MEMORY_BAR = 2;
  64 
  65 struct vsoc_region_data {
  66         char name[VSOC_DEVICE_NAME_SZ + 1];
  67         wait_queue_head_t interrupt_wait_queue;
  68         /* TODO(b/73664181): Use multiple futex wait queues */
  69         wait_queue_head_t futex_wait_queue;
  70         /* Flag indicating that an interrupt has been signalled by the host. */
  71         atomic_t *incoming_signalled;
  72         /* Flag indicating the guest has signalled the host. */
  73         atomic_t *outgoing_signalled;
  74         bool irq_requested;
  75         bool device_created;
  76 };
  77 
  78 struct vsoc_device {
  79         /* Kernel virtual address of REGISTER_BAR. */
  80         void __iomem *regs;
  81         /* Physical address of SHARED_MEMORY_BAR. */
  82         phys_addr_t shm_phys_start;
  83         /* Kernel virtual address of SHARED_MEMORY_BAR. */
  84         void __iomem *kernel_mapped_shm;
  85         /* Size of the entire shared memory window in bytes. */
  86         size_t shm_size;
  87         /*
  88          * Pointer to the virtual address of the shared memory layout structure.
  89          * This is probably identical to kernel_mapped_shm, but saving this
  90          * here saves a lot of annoying casts.
  91          */
  92         struct vsoc_shm_layout_descriptor *layout;
  93         /*
  94          * Points to a table of region descriptors in the kernel's virtual
  95          * address space. Calculated from
  96          * vsoc_shm_layout_descriptor.vsoc_region_desc_offset
  97          */
  98         struct vsoc_device_region *regions;
  99         /* Head of a list of permissions that have been granted. */
 100         struct list_head permissions;
 101         struct pci_dev *dev;
 102         /* Per-region (and therefore per-interrupt) information. */
 103         struct vsoc_region_data *regions_data;
 104         /*
 105          * Table of msi-x entries. This has to be separated from struct
 106          * vsoc_region_data because the kernel deals with them as an array.
 107          */
 108         struct msix_entry *msix_entries;
 109         /* Mutex that protectes the permission list */
 110         struct mutex mtx;
 111         /* Major number assigned by the kernel */
 112         int major;
 113         /* Character device assigned by the kernel */
 114         struct cdev cdev;
 115         /* Device class assigned by the kernel */
 116         struct class *class;
 117         /*
 118          * Flags that indicate what we've initialized. These are used to do an
 119          * orderly cleanup of the device.
 120          */
 121         bool enabled_device;
 122         bool requested_regions;
 123         bool cdev_added;
 124         bool class_added;
 125         bool msix_enabled;
 126 };
 127 
 128 static struct vsoc_device vsoc_dev;
 129 
 130 /*
 131  * TODO(ghartman): Add a /sys filesystem entry that summarizes the permissions.
 132  */
 133 
 134 struct fd_scoped_permission_node {
 135         struct fd_scoped_permission permission;
 136         struct list_head list;
 137 };
 138 
 139 struct vsoc_private_data {
 140         struct fd_scoped_permission_node *fd_scoped_permission_node;
 141 };
 142 
 143 static long vsoc_ioctl(struct file *, unsigned int, unsigned long);
 144 static int vsoc_mmap(struct file *, struct vm_area_struct *);
 145 static int vsoc_open(struct inode *, struct file *);
 146 static int vsoc_release(struct inode *, struct file *);
 147 static ssize_t vsoc_read(struct file *, char __user *, size_t, loff_t *);
 148 static ssize_t vsoc_write(struct file *, const char __user *, size_t, loff_t *);
 149 static loff_t vsoc_lseek(struct file *filp, loff_t offset, int origin);
 150 static int
 151 do_create_fd_scoped_permission(struct vsoc_device_region *region_p,
 152                                struct fd_scoped_permission_node *np,
 153                                struct fd_scoped_permission_arg __user *arg);
 154 static void
 155 do_destroy_fd_scoped_permission(struct vsoc_device_region *owner_region_p,
 156                                 struct fd_scoped_permission *perm);
 157 static long do_vsoc_describe_region(struct file *,
 158                                     struct vsoc_device_region __user *);
 159 static ssize_t vsoc_get_area(struct file *filp, __u32 *perm_off);
 160 
 161 /**
 162  * Validate arguments on entry points to the driver.
 163  */
 164 inline int vsoc_validate_inode(struct inode *inode)
 165 {
 166         if (iminor(inode) >= vsoc_dev.layout->region_count) {
 167                 dev_err(&vsoc_dev.dev->dev,
 168                         "describe_region: invalid region %d\n", iminor(inode));
 169                 return -ENODEV;
 170         }
 171         return 0;
 172 }
 173 
 174 inline int vsoc_validate_filep(struct file *filp)
 175 {
 176         int ret = vsoc_validate_inode(file_inode(filp));
 177 
 178         if (ret)
 179                 return ret;
 180         if (!filp->private_data) {
 181                 dev_err(&vsoc_dev.dev->dev,
 182                         "No private data on fd, region %d\n",
 183                         iminor(file_inode(filp)));
 184                 return -EBADFD;
 185         }
 186         return 0;
 187 }
 188 
 189 /* Converts from shared memory offset to virtual address */
 190 static inline void *shm_off_to_virtual_addr(__u32 offset)
 191 {
 192         return (void __force *)vsoc_dev.kernel_mapped_shm + offset;
 193 }
 194 
 195 /* Converts from shared memory offset to physical address */
 196 static inline phys_addr_t shm_off_to_phys_addr(__u32 offset)
 197 {
 198         return vsoc_dev.shm_phys_start + offset;
 199 }
 200 
 201 /**
 202  * Convenience functions to obtain the region from the inode or file.
 203  * Dangerous to call before validating the inode/file.
 204  */
 205 static
 206 inline struct vsoc_device_region *vsoc_region_from_inode(struct inode *inode)
 207 {
 208         return &vsoc_dev.regions[iminor(inode)];
 209 }
 210 
 211 static
 212 inline struct vsoc_device_region *vsoc_region_from_filep(struct file *inode)
 213 {
 214         return vsoc_region_from_inode(file_inode(inode));
 215 }
 216 
 217 static inline uint32_t vsoc_device_region_size(struct vsoc_device_region *r)
 218 {
 219         return r->region_end_offset - r->region_begin_offset;
 220 }
 221 
 222 static const struct file_operations vsoc_ops = {
 223         .owner = THIS_MODULE,
 224         .open = vsoc_open,
 225         .mmap = vsoc_mmap,
 226         .read = vsoc_read,
 227         .unlocked_ioctl = vsoc_ioctl,
 228         .compat_ioctl = vsoc_ioctl,
 229         .write = vsoc_write,
 230         .llseek = vsoc_lseek,
 231         .release = vsoc_release,
 232 };
 233 
 234 static struct pci_device_id vsoc_id_table[] = {
 235         {0x1af4, 0x1110, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
 236         {0},
 237 };
 238 
 239 MODULE_DEVICE_TABLE(pci, vsoc_id_table);
 240 
 241 static void vsoc_remove_device(struct pci_dev *pdev);
 242 static int vsoc_probe_device(struct pci_dev *pdev,
 243                              const struct pci_device_id *ent);
 244 
 245 static struct pci_driver vsoc_pci_driver = {
 246         .name = "vsoc",
 247         .id_table = vsoc_id_table,
 248         .probe = vsoc_probe_device,
 249         .remove = vsoc_remove_device,
 250 };
 251 
 252 static int
 253 do_create_fd_scoped_permission(struct vsoc_device_region *region_p,
 254                                struct fd_scoped_permission_node *np,
 255                                struct fd_scoped_permission_arg __user *arg)
 256 {
 257         struct file *managed_filp;
 258         s32 managed_fd;
 259         atomic_t *owner_ptr = NULL;
 260         struct vsoc_device_region *managed_region_p;
 261 
 262         if (copy_from_user(&np->permission,
 263                            &arg->perm, sizeof(np->permission)) ||
 264             copy_from_user(&managed_fd,
 265                            &arg->managed_region_fd, sizeof(managed_fd))) {
 266                 return -EFAULT;
 267         }
 268         managed_filp = fdget(managed_fd).file;
 269         /* Check that it's a valid fd, */
 270         if (!managed_filp || vsoc_validate_filep(managed_filp))
 271                 return -EPERM;
 272         /* EEXIST if the given fd already has a permission. */
 273         if (((struct vsoc_private_data *)managed_filp->private_data)->
 274             fd_scoped_permission_node)
 275                 return -EEXIST;
 276         managed_region_p = vsoc_region_from_filep(managed_filp);
 277         /* Check that the provided region is managed by this one */
 278         if (&vsoc_dev.regions[managed_region_p->managed_by] != region_p)
 279                 return -EPERM;
 280         /* The area must be well formed and have non-zero size */
 281         if (np->permission.begin_offset >= np->permission.end_offset)
 282                 return -EINVAL;
 283         /* The area must fit in the memory window */
 284         if (np->permission.end_offset >
 285             vsoc_device_region_size(managed_region_p))
 286                 return -ERANGE;
 287         /* The area must be in the region data section */
 288         if (np->permission.begin_offset <
 289             managed_region_p->offset_of_region_data)
 290                 return -ERANGE;
 291         /* The area must be page aligned */
 292         if (!PAGE_ALIGNED(np->permission.begin_offset) ||
 293             !PAGE_ALIGNED(np->permission.end_offset))
 294                 return -EINVAL;
 295         /* Owner offset must be naturally aligned in the window */
 296         if (np->permission.owner_offset &
 297             (sizeof(np->permission.owner_offset) - 1))
 298                 return -EINVAL;
 299         /* The owner flag must reside in the owner memory */
 300         if (np->permission.owner_offset + sizeof(np->permission.owner_offset) >
 301             vsoc_device_region_size(region_p))
 302                 return -ERANGE;
 303         /* The owner flag must reside in the data section */
 304         if (np->permission.owner_offset < region_p->offset_of_region_data)
 305                 return -EINVAL;
 306         /* The owner value must change to claim the memory */
 307         if (np->permission.owned_value == VSOC_REGION_FREE)
 308                 return -EINVAL;
 309         owner_ptr =
 310             (atomic_t *)shm_off_to_virtual_addr(region_p->region_begin_offset +
 311                                                 np->permission.owner_offset);
 312         /* We've already verified that this is in the shared memory window, so
 313          * it should be safe to write to this address.
 314          */
 315         if (atomic_cmpxchg(owner_ptr,
 316                            VSOC_REGION_FREE,
 317                            np->permission.owned_value) != VSOC_REGION_FREE) {
 318                 return -EBUSY;
 319         }
 320         ((struct vsoc_private_data *)managed_filp->private_data)->
 321             fd_scoped_permission_node = np;
 322         /* The file offset needs to be adjusted if the calling
 323          * process did any read/write operations on the fd
 324          * before creating the permission.
 325          */
 326         if (managed_filp->f_pos) {
 327                 if (managed_filp->f_pos > np->permission.end_offset) {
 328                         /* If the offset is beyond the permission end, set it
 329                          * to the end.
 330                          */
 331                         managed_filp->f_pos = np->permission.end_offset;
 332                 } else {
 333                         /* If the offset is within the permission interval
 334                          * keep it there otherwise reset it to zero.
 335                          */
 336                         if (managed_filp->f_pos < np->permission.begin_offset) {
 337                                 managed_filp->f_pos = 0;
 338                         } else {
 339                                 managed_filp->f_pos -=
 340                                     np->permission.begin_offset;
 341                         }
 342                 }
 343         }
 344         return 0;
 345 }
 346 
 347 static void
 348 do_destroy_fd_scoped_permission_node(struct vsoc_device_region *owner_region_p,
 349                                      struct fd_scoped_permission_node *node)
 350 {
 351         if (node) {
 352                 do_destroy_fd_scoped_permission(owner_region_p,
 353                                                 &node->permission);
 354                 mutex_lock(&vsoc_dev.mtx);
 355                 list_del(&node->list);
 356                 mutex_unlock(&vsoc_dev.mtx);
 357                 kfree(node);
 358         }
 359 }
 360 
 361 static void
 362 do_destroy_fd_scoped_permission(struct vsoc_device_region *owner_region_p,
 363                                 struct fd_scoped_permission *perm)
 364 {
 365         atomic_t *owner_ptr = NULL;
 366         int prev = 0;
 367 
 368         if (!perm)
 369                 return;
 370         owner_ptr = (atomic_t *)shm_off_to_virtual_addr
 371                 (owner_region_p->region_begin_offset + perm->owner_offset);
 372         prev = atomic_xchg(owner_ptr, VSOC_REGION_FREE);
 373         if (prev != perm->owned_value)
 374                 dev_err(&vsoc_dev.dev->dev,
 375                         "%x-%x: owner (%s) %x: expected to be %x was %x",
 376                         perm->begin_offset, perm->end_offset,
 377                         owner_region_p->device_name, perm->owner_offset,
 378                         perm->owned_value, prev);
 379 }
 380 
 381 static long do_vsoc_describe_region(struct file *filp,
 382                                     struct vsoc_device_region __user *dest)
 383 {
 384         struct vsoc_device_region *region_p;
 385         int retval = vsoc_validate_filep(filp);
 386 
 387         if (retval)
 388                 return retval;
 389         region_p = vsoc_region_from_filep(filp);
 390         if (copy_to_user(dest, region_p, sizeof(*region_p)))
 391                 return -EFAULT;
 392         return 0;
 393 }
 394 
 395 /**
 396  * Implements the inner logic of cond_wait. Copies to and from userspace are
 397  * done in the helper function below.
 398  */
 399 static int handle_vsoc_cond_wait(struct file *filp, struct vsoc_cond_wait *arg)
 400 {
 401         DEFINE_WAIT(wait);
 402         u32 region_number = iminor(file_inode(filp));
 403         struct vsoc_region_data *data = vsoc_dev.regions_data + region_number;
 404         struct hrtimer_sleeper timeout, *to = NULL;
 405         int ret = 0;
 406         struct vsoc_device_region *region_p = vsoc_region_from_filep(filp);
 407         atomic_t *address = NULL;
 408         ktime_t wake_time;
 409 
 410         /* Ensure that the offset is aligned */
 411         if (arg->offset & (sizeof(uint32_t) - 1))
 412                 return -EADDRNOTAVAIL;
 413         /* Ensure that the offset is within shared memory */
 414         if (((uint64_t)arg->offset) + region_p->region_begin_offset +
 415             sizeof(uint32_t) > region_p->region_end_offset)
 416                 return -E2BIG;
 417         address = shm_off_to_virtual_addr(region_p->region_begin_offset +
 418                                           arg->offset);
 419 
 420         /* Ensure that the type of wait is valid */
 421         switch (arg->wait_type) {
 422         case VSOC_WAIT_IF_EQUAL:
 423                 break;
 424         case VSOC_WAIT_IF_EQUAL_TIMEOUT:
 425                 to = &timeout;
 426                 break;
 427         default:
 428                 return -EINVAL;
 429         }
 430 
 431         if (to) {
 432                 /* Copy the user-supplied timesec into the kernel structure.
 433                  * We do things this way to flatten differences between 32 bit
 434                  * and 64 bit timespecs.
 435                  */
 436                 if (arg->wake_time_nsec >= NSEC_PER_SEC)
 437                         return -EINVAL;
 438                 wake_time = ktime_set(arg->wake_time_sec, arg->wake_time_nsec);
 439 
 440                 hrtimer_init_sleeper_on_stack(to, CLOCK_MONOTONIC,
 441                                               HRTIMER_MODE_ABS);
 442                 hrtimer_set_expires_range_ns(&to->timer, wake_time,
 443                                              current->timer_slack_ns);
 444         }
 445 
 446         while (1) {
 447                 prepare_to_wait(&data->futex_wait_queue, &wait,
 448                                 TASK_INTERRUPTIBLE);
 449                 /*
 450                  * Check the sentinel value after prepare_to_wait. If the value
 451                  * changes after this check the writer will call signal,
 452                  * changing the task state from INTERRUPTIBLE to RUNNING. That
 453                  * will ensure that schedule() will eventually schedule this
 454                  * task.
 455                  */
 456                 if (atomic_read(address) != arg->value) {
 457                         ret = 0;
 458                         break;
 459                 }
 460                 if (to) {
 461                         hrtimer_sleeper_start_expires(to, HRTIMER_MODE_ABS);
 462                         if (likely(to->task))
 463                                 freezable_schedule();
 464                         hrtimer_cancel(&to->timer);
 465                         if (!to->task) {
 466                                 ret = -ETIMEDOUT;
 467                                 break;
 468                         }
 469                 } else {
 470                         freezable_schedule();
 471                 }
 472                 /* Count the number of times that we woke up. This is useful
 473                  * for unit testing.
 474                  */
 475                 ++arg->wakes;
 476                 if (signal_pending(current)) {
 477                         ret = -EINTR;
 478                         break;
 479                 }
 480         }
 481         finish_wait(&data->futex_wait_queue, &wait);
 482         if (to)
 483                 destroy_hrtimer_on_stack(&to->timer);
 484         return ret;
 485 }
 486 
 487 /**
 488  * Handles the details of copying from/to userspace to ensure that the copies
 489  * happen on all of the return paths of cond_wait.
 490  */
 491 static int do_vsoc_cond_wait(struct file *filp,
 492                              struct vsoc_cond_wait __user *untrusted_in)
 493 {
 494         struct vsoc_cond_wait arg;
 495         int rval = 0;
 496 
 497         if (copy_from_user(&arg, untrusted_in, sizeof(arg)))
 498                 return -EFAULT;
 499         /* wakes is an out parameter. Initialize it to something sensible. */
 500         arg.wakes = 0;
 501         rval = handle_vsoc_cond_wait(filp, &arg);
 502         if (copy_to_user(untrusted_in, &arg, sizeof(arg)))
 503                 return -EFAULT;
 504         return rval;
 505 }
 506 
 507 static int do_vsoc_cond_wake(struct file *filp, uint32_t offset)
 508 {
 509         struct vsoc_device_region *region_p = vsoc_region_from_filep(filp);
 510         u32 region_number = iminor(file_inode(filp));
 511         struct vsoc_region_data *data = vsoc_dev.regions_data + region_number;
 512         /* Ensure that the offset is aligned */
 513         if (offset & (sizeof(uint32_t) - 1))
 514                 return -EADDRNOTAVAIL;
 515         /* Ensure that the offset is within shared memory */
 516         if (((uint64_t)offset) + region_p->region_begin_offset +
 517             sizeof(uint32_t) > region_p->region_end_offset)
 518                 return -E2BIG;
 519         /*
 520          * TODO(b/73664181): Use multiple futex wait queues.
 521          * We need to wake every sleeper when the condition changes. Typically
 522          * only a single thread will be waiting on the condition, but there
 523          * are exceptions. The worst case is about 10 threads.
 524          */
 525         wake_up_interruptible_all(&data->futex_wait_queue);
 526         return 0;
 527 }
 528 
 529 static long vsoc_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 530 {
 531         int rv = 0;
 532         struct vsoc_device_region *region_p;
 533         u32 reg_num;
 534         struct vsoc_region_data *reg_data;
 535         int retval = vsoc_validate_filep(filp);
 536 
 537         if (retval)
 538                 return retval;
 539         region_p = vsoc_region_from_filep(filp);
 540         reg_num = iminor(file_inode(filp));
 541         reg_data = vsoc_dev.regions_data + reg_num;
 542         switch (cmd) {
 543         case VSOC_CREATE_FD_SCOPED_PERMISSION:
 544                 {
 545                         struct fd_scoped_permission_node *node = NULL;
 546 
 547                         node = kzalloc(sizeof(*node), GFP_KERNEL);
 548                         /* We can't allocate memory for the permission */
 549                         if (!node)
 550                                 return -ENOMEM;
 551                         INIT_LIST_HEAD(&node->list);
 552                         rv = do_create_fd_scoped_permission
 553                                 (region_p,
 554                                  node,
 555                                  (struct fd_scoped_permission_arg __user *)arg);
 556                         if (!rv) {
 557                                 mutex_lock(&vsoc_dev.mtx);
 558                                 list_add(&node->list, &vsoc_dev.permissions);
 559                                 mutex_unlock(&vsoc_dev.mtx);
 560                         } else {
 561                                 kfree(node);
 562                                 return rv;
 563                         }
 564                 }
 565                 break;
 566 
 567         case VSOC_GET_FD_SCOPED_PERMISSION:
 568                 {
 569                         struct fd_scoped_permission_node *node =
 570                             ((struct vsoc_private_data *)filp->private_data)->
 571                             fd_scoped_permission_node;
 572                         if (!node)
 573                                 return -ENOENT;
 574                         if (copy_to_user
 575                             ((struct fd_scoped_permission __user *)arg,
 576                              &node->permission, sizeof(node->permission)))
 577                                 return -EFAULT;
 578                 }
 579                 break;
 580 
 581         case VSOC_MAYBE_SEND_INTERRUPT_TO_HOST:
 582                 if (!atomic_xchg(reg_data->outgoing_signalled, 1)) {
 583                         writel(reg_num, vsoc_dev.regs + DOORBELL);
 584                         return 0;
 585                 } else {
 586                         return -EBUSY;
 587                 }
 588                 break;
 589 
 590         case VSOC_SEND_INTERRUPT_TO_HOST:
 591                 writel(reg_num, vsoc_dev.regs + DOORBELL);
 592                 return 0;
 593         case VSOC_WAIT_FOR_INCOMING_INTERRUPT:
 594                 wait_event_interruptible
 595                         (reg_data->interrupt_wait_queue,
 596                          (atomic_read(reg_data->incoming_signalled) != 0));
 597                 break;
 598 
 599         case VSOC_DESCRIBE_REGION:
 600                 return do_vsoc_describe_region
 601                         (filp,
 602                          (struct vsoc_device_region __user *)arg);
 603 
 604         case VSOC_SELF_INTERRUPT:
 605                 atomic_set(reg_data->incoming_signalled, 1);
 606                 wake_up_interruptible(&reg_data->interrupt_wait_queue);
 607                 break;
 608 
 609         case VSOC_COND_WAIT:
 610                 return do_vsoc_cond_wait(filp,
 611                                          (struct vsoc_cond_wait __user *)arg);
 612         case VSOC_COND_WAKE:
 613                 return do_vsoc_cond_wake(filp, arg);
 614 
 615         default:
 616                 return -EINVAL;
 617         }
 618         return 0;
 619 }
 620 
 621 static ssize_t vsoc_read(struct file *filp, char __user *buffer, size_t len,
 622                          loff_t *poffset)
 623 {
 624         __u32 area_off;
 625         const void *area_p;
 626         ssize_t area_len;
 627         int retval = vsoc_validate_filep(filp);
 628 
 629         if (retval)
 630                 return retval;
 631         area_len = vsoc_get_area(filp, &area_off);
 632         area_p = shm_off_to_virtual_addr(area_off);
 633         area_p += *poffset;
 634         area_len -= *poffset;
 635         if (area_len <= 0)
 636                 return 0;
 637         if (area_len < len)
 638                 len = area_len;
 639         if (copy_to_user(buffer, area_p, len))
 640                 return -EFAULT;
 641         *poffset += len;
 642         return len;
 643 }
 644 
 645 static loff_t vsoc_lseek(struct file *filp, loff_t offset, int origin)
 646 {
 647         ssize_t area_len = 0;
 648         int retval = vsoc_validate_filep(filp);
 649 
 650         if (retval)
 651                 return retval;
 652         area_len = vsoc_get_area(filp, NULL);
 653         switch (origin) {
 654         case SEEK_SET:
 655                 break;
 656 
 657         case SEEK_CUR:
 658                 if (offset > 0 && offset + filp->f_pos < 0)
 659                         return -EOVERFLOW;
 660                 offset += filp->f_pos;
 661                 break;
 662 
 663         case SEEK_END:
 664                 if (offset > 0 && offset + area_len < 0)
 665                         return -EOVERFLOW;
 666                 offset += area_len;
 667                 break;
 668 
 669         case SEEK_DATA:
 670                 if (offset >= area_len)
 671                         return -EINVAL;
 672                 if (offset < 0)
 673                         offset = 0;
 674                 break;
 675 
 676         case SEEK_HOLE:
 677                 /* Next hole is always the end of the region, unless offset is
 678                  * beyond that
 679                  */
 680                 if (offset < area_len)
 681                         offset = area_len;
 682                 break;
 683 
 684         default:
 685                 return -EINVAL;
 686         }
 687 
 688         if (offset < 0 || offset > area_len)
 689                 return -EINVAL;
 690         filp->f_pos = offset;
 691 
 692         return offset;
 693 }
 694 
 695 static ssize_t vsoc_write(struct file *filp, const char __user *buffer,
 696                           size_t len, loff_t *poffset)
 697 {
 698         __u32 area_off;
 699         void *area_p;
 700         ssize_t area_len;
 701         int retval = vsoc_validate_filep(filp);
 702 
 703         if (retval)
 704                 return retval;
 705         area_len = vsoc_get_area(filp, &area_off);
 706         area_p = shm_off_to_virtual_addr(area_off);
 707         area_p += *poffset;
 708         area_len -= *poffset;
 709         if (area_len <= 0)
 710                 return 0;
 711         if (area_len < len)
 712                 len = area_len;
 713         if (copy_from_user(area_p, buffer, len))
 714                 return -EFAULT;
 715         *poffset += len;
 716         return len;
 717 }
 718 
 719 static irqreturn_t vsoc_interrupt(int irq, void *region_data_v)
 720 {
 721         struct vsoc_region_data *region_data =
 722             (struct vsoc_region_data *)region_data_v;
 723         int reg_num = region_data - vsoc_dev.regions_data;
 724 
 725         if (unlikely(!region_data))
 726                 return IRQ_NONE;
 727 
 728         if (unlikely(reg_num < 0 ||
 729                      reg_num >= vsoc_dev.layout->region_count)) {
 730                 dev_err(&vsoc_dev.dev->dev,
 731                         "invalid irq @%p reg_num=0x%04x\n",
 732                         region_data, reg_num);
 733                 return IRQ_NONE;
 734         }
 735         if (unlikely(vsoc_dev.regions_data + reg_num != region_data)) {
 736                 dev_err(&vsoc_dev.dev->dev,
 737                         "irq not aligned @%p reg_num=0x%04x\n",
 738                         region_data, reg_num);
 739                 return IRQ_NONE;
 740         }
 741         wake_up_interruptible(&region_data->interrupt_wait_queue);
 742         return IRQ_HANDLED;
 743 }
 744 
 745 static int vsoc_probe_device(struct pci_dev *pdev,
 746                              const struct pci_device_id *ent)
 747 {
 748         int result;
 749         int i;
 750         resource_size_t reg_size;
 751         dev_t devt;
 752 
 753         vsoc_dev.dev = pdev;
 754         result = pci_enable_device(pdev);
 755         if (result) {
 756                 dev_err(&pdev->dev,
 757                         "pci_enable_device failed %s: error %d\n",
 758                         pci_name(pdev), result);
 759                 return result;
 760         }
 761         vsoc_dev.enabled_device = true;
 762         result = pci_request_regions(pdev, "vsoc");
 763         if (result < 0) {
 764                 dev_err(&pdev->dev, "pci_request_regions failed\n");
 765                 vsoc_remove_device(pdev);
 766                 return -EBUSY;
 767         }
 768         vsoc_dev.requested_regions = true;
 769         /* Set up the control registers in BAR 0 */
 770         reg_size = pci_resource_len(pdev, REGISTER_BAR);
 771         if (reg_size > MAX_REGISTER_BAR_LEN)
 772                 vsoc_dev.regs =
 773                     pci_iomap(pdev, REGISTER_BAR, MAX_REGISTER_BAR_LEN);
 774         else
 775                 vsoc_dev.regs = pci_iomap(pdev, REGISTER_BAR, reg_size);
 776 
 777         if (!vsoc_dev.regs) {
 778                 dev_err(&pdev->dev,
 779                         "cannot map registers of size %zu\n",
 780                        (size_t)reg_size);
 781                 vsoc_remove_device(pdev);
 782                 return -EBUSY;
 783         }
 784 
 785         /* Map the shared memory in BAR 2 */
 786         vsoc_dev.shm_phys_start = pci_resource_start(pdev, SHARED_MEMORY_BAR);
 787         vsoc_dev.shm_size = pci_resource_len(pdev, SHARED_MEMORY_BAR);
 788 
 789         dev_info(&pdev->dev, "shared memory @ DMA %pa size=0x%zx\n",
 790                  &vsoc_dev.shm_phys_start, vsoc_dev.shm_size);
 791         vsoc_dev.kernel_mapped_shm = pci_iomap_wc(pdev, SHARED_MEMORY_BAR, 0);
 792         if (!vsoc_dev.kernel_mapped_shm) {
 793                 dev_err(&vsoc_dev.dev->dev, "cannot iomap region\n");
 794                 vsoc_remove_device(pdev);
 795                 return -EBUSY;
 796         }
 797 
 798         vsoc_dev.layout = (struct vsoc_shm_layout_descriptor __force *)
 799                                 vsoc_dev.kernel_mapped_shm;
 800         dev_info(&pdev->dev, "major_version: %d\n",
 801                  vsoc_dev.layout->major_version);
 802         dev_info(&pdev->dev, "minor_version: %d\n",
 803                  vsoc_dev.layout->minor_version);
 804         dev_info(&pdev->dev, "size: 0x%x\n", vsoc_dev.layout->size);
 805         dev_info(&pdev->dev, "regions: %d\n", vsoc_dev.layout->region_count);
 806         if (vsoc_dev.layout->major_version !=
 807             CURRENT_VSOC_LAYOUT_MAJOR_VERSION) {
 808                 dev_err(&vsoc_dev.dev->dev,
 809                         "driver supports only major_version %d\n",
 810                         CURRENT_VSOC_LAYOUT_MAJOR_VERSION);
 811                 vsoc_remove_device(pdev);
 812                 return -EBUSY;
 813         }
 814         result = alloc_chrdev_region(&devt, 0, vsoc_dev.layout->region_count,
 815                                      VSOC_DEV_NAME);
 816         if (result) {
 817                 dev_err(&vsoc_dev.dev->dev, "alloc_chrdev_region failed\n");
 818                 vsoc_remove_device(pdev);
 819                 return -EBUSY;
 820         }
 821         vsoc_dev.major = MAJOR(devt);
 822         cdev_init(&vsoc_dev.cdev, &vsoc_ops);
 823         vsoc_dev.cdev.owner = THIS_MODULE;
 824         result = cdev_add(&vsoc_dev.cdev, devt, vsoc_dev.layout->region_count);
 825         if (result) {
 826                 dev_err(&vsoc_dev.dev->dev, "cdev_add error\n");
 827                 vsoc_remove_device(pdev);
 828                 return -EBUSY;
 829         }
 830         vsoc_dev.cdev_added = true;
 831         vsoc_dev.class = class_create(THIS_MODULE, VSOC_DEV_NAME);
 832         if (IS_ERR(vsoc_dev.class)) {
 833                 dev_err(&vsoc_dev.dev->dev, "class_create failed\n");
 834                 vsoc_remove_device(pdev);
 835                 return PTR_ERR(vsoc_dev.class);
 836         }
 837         vsoc_dev.class_added = true;
 838         vsoc_dev.regions = (struct vsoc_device_region __force *)
 839                 ((void *)vsoc_dev.layout +
 840                  vsoc_dev.layout->vsoc_region_desc_offset);
 841         vsoc_dev.msix_entries =
 842                 kcalloc(vsoc_dev.layout->region_count,
 843                         sizeof(vsoc_dev.msix_entries[0]), GFP_KERNEL);
 844         if (!vsoc_dev.msix_entries) {
 845                 dev_err(&vsoc_dev.dev->dev,
 846                         "unable to allocate msix_entries\n");
 847                 vsoc_remove_device(pdev);
 848                 return -ENOSPC;
 849         }
 850         vsoc_dev.regions_data =
 851                 kcalloc(vsoc_dev.layout->region_count,
 852                         sizeof(vsoc_dev.regions_data[0]), GFP_KERNEL);
 853         if (!vsoc_dev.regions_data) {
 854                 dev_err(&vsoc_dev.dev->dev,
 855                         "unable to allocate regions' data\n");
 856                 vsoc_remove_device(pdev);
 857                 return -ENOSPC;
 858         }
 859         for (i = 0; i < vsoc_dev.layout->region_count; ++i)
 860                 vsoc_dev.msix_entries[i].entry = i;
 861 
 862         result = pci_enable_msix_exact(vsoc_dev.dev, vsoc_dev.msix_entries,
 863                                        vsoc_dev.layout->region_count);
 864         if (result) {
 865                 dev_info(&pdev->dev, "pci_enable_msix failed: %d\n", result);
 866                 vsoc_remove_device(pdev);
 867                 return -ENOSPC;
 868         }
 869         /* Check that all regions are well formed */
 870         for (i = 0; i < vsoc_dev.layout->region_count; ++i) {
 871                 const struct vsoc_device_region *region = vsoc_dev.regions + i;
 872 
 873                 if (!PAGE_ALIGNED(region->region_begin_offset) ||
 874                     !PAGE_ALIGNED(region->region_end_offset)) {
 875                         dev_err(&vsoc_dev.dev->dev,
 876                                 "region %d not aligned (%x:%x)", i,
 877                                 region->region_begin_offset,
 878                                 region->region_end_offset);
 879                         vsoc_remove_device(pdev);
 880                         return -EFAULT;
 881                 }
 882                 if (region->region_begin_offset >= region->region_end_offset ||
 883                     region->region_end_offset > vsoc_dev.shm_size) {
 884                         dev_err(&vsoc_dev.dev->dev,
 885                                 "region %d offsets are wrong: %x %x %zx",
 886                                 i, region->region_begin_offset,
 887                                 region->region_end_offset, vsoc_dev.shm_size);
 888                         vsoc_remove_device(pdev);
 889                         return -EFAULT;
 890                 }
 891                 if (region->managed_by >= vsoc_dev.layout->region_count) {
 892                         dev_err(&vsoc_dev.dev->dev,
 893                                 "region %d has invalid owner: %u",
 894                                 i, region->managed_by);
 895                         vsoc_remove_device(pdev);
 896                         return -EFAULT;
 897                 }
 898         }
 899         vsoc_dev.msix_enabled = true;
 900         for (i = 0; i < vsoc_dev.layout->region_count; ++i) {
 901                 const struct vsoc_device_region *region = vsoc_dev.regions + i;
 902                 size_t name_sz = sizeof(vsoc_dev.regions_data[i].name) - 1;
 903                 const struct vsoc_signal_table_layout *h_to_g_signal_table =
 904                         &region->host_to_guest_signal_table;
 905                 const struct vsoc_signal_table_layout *g_to_h_signal_table =
 906                         &region->guest_to_host_signal_table;
 907 
 908                 vsoc_dev.regions_data[i].name[name_sz] = '\0';
 909                 memcpy(vsoc_dev.regions_data[i].name, region->device_name,
 910                        name_sz);
 911                 dev_info(&pdev->dev, "region %d name=%s\n",
 912                          i, vsoc_dev.regions_data[i].name);
 913                 init_waitqueue_head
 914                         (&vsoc_dev.regions_data[i].interrupt_wait_queue);
 915                 init_waitqueue_head(&vsoc_dev.regions_data[i].futex_wait_queue);
 916                 vsoc_dev.regions_data[i].incoming_signalled =
 917                         shm_off_to_virtual_addr(region->region_begin_offset) +
 918                         h_to_g_signal_table->interrupt_signalled_offset;
 919                 vsoc_dev.regions_data[i].outgoing_signalled =
 920                         shm_off_to_virtual_addr(region->region_begin_offset) +
 921                         g_to_h_signal_table->interrupt_signalled_offset;
 922                 result = request_irq(vsoc_dev.msix_entries[i].vector,
 923                                      vsoc_interrupt, 0,
 924                                      vsoc_dev.regions_data[i].name,
 925                                      vsoc_dev.regions_data + i);
 926                 if (result) {
 927                         dev_info(&pdev->dev,
 928                                  "request_irq failed irq=%d vector=%d\n",
 929                                 i, vsoc_dev.msix_entries[i].vector);
 930                         vsoc_remove_device(pdev);
 931                         return -ENOSPC;
 932                 }
 933                 vsoc_dev.regions_data[i].irq_requested = true;
 934                 if (!device_create(vsoc_dev.class, NULL,
 935                                    MKDEV(vsoc_dev.major, i),
 936                                    NULL, vsoc_dev.regions_data[i].name)) {
 937                         dev_err(&vsoc_dev.dev->dev, "device_create failed\n");
 938                         vsoc_remove_device(pdev);
 939                         return -EBUSY;
 940                 }
 941                 vsoc_dev.regions_data[i].device_created = true;
 942         }
 943         return 0;
 944 }
 945 
 946 /*
 947  * This should undo all of the allocations in the probe function in reverse
 948  * order.
 949  *
 950  * Notes:
 951  *
 952  *   The device may have been partially initialized, so double check
 953  *   that the allocations happened.
 954  *
 955  *   This function may be called multiple times, so mark resources as freed
 956  *   as they are deallocated.
 957  */
 958 static void vsoc_remove_device(struct pci_dev *pdev)
 959 {
 960         int i;
 961         /*
 962          * pdev is the first thing to be set on probe and the last thing
 963          * to be cleared here. If it's NULL then there is no cleanup.
 964          */
 965         if (!pdev || !vsoc_dev.dev)
 966                 return;
 967         dev_info(&pdev->dev, "remove_device\n");
 968         if (vsoc_dev.regions_data) {
 969                 for (i = 0; i < vsoc_dev.layout->region_count; ++i) {
 970                         if (vsoc_dev.regions_data[i].device_created) {
 971                                 device_destroy(vsoc_dev.class,
 972                                                MKDEV(vsoc_dev.major, i));
 973                                 vsoc_dev.regions_data[i].device_created = false;
 974                         }
 975                         if (vsoc_dev.regions_data[i].irq_requested)
 976                                 free_irq(vsoc_dev.msix_entries[i].vector, NULL);
 977                         vsoc_dev.regions_data[i].irq_requested = false;
 978                 }
 979                 kfree(vsoc_dev.regions_data);
 980                 vsoc_dev.regions_data = NULL;
 981         }
 982         if (vsoc_dev.msix_enabled) {
 983                 pci_disable_msix(pdev);
 984                 vsoc_dev.msix_enabled = false;
 985         }
 986         kfree(vsoc_dev.msix_entries);
 987         vsoc_dev.msix_entries = NULL;
 988         vsoc_dev.regions = NULL;
 989         if (vsoc_dev.class_added) {
 990                 class_destroy(vsoc_dev.class);
 991                 vsoc_dev.class_added = false;
 992         }
 993         if (vsoc_dev.cdev_added) {
 994                 cdev_del(&vsoc_dev.cdev);
 995                 vsoc_dev.cdev_added = false;
 996         }
 997         if (vsoc_dev.major && vsoc_dev.layout) {
 998                 unregister_chrdev_region(MKDEV(vsoc_dev.major, 0),
 999                                          vsoc_dev.layout->region_count);
1000                 vsoc_dev.major = 0;
1001         }
1002         vsoc_dev.layout = NULL;
1003         if (vsoc_dev.kernel_mapped_shm) {
1004                 pci_iounmap(pdev, vsoc_dev.kernel_mapped_shm);
1005                 vsoc_dev.kernel_mapped_shm = NULL;
1006         }
1007         if (vsoc_dev.regs) {
1008                 pci_iounmap(pdev, vsoc_dev.regs);
1009                 vsoc_dev.regs = NULL;
1010         }
1011         if (vsoc_dev.requested_regions) {
1012                 pci_release_regions(pdev);
1013                 vsoc_dev.requested_regions = false;
1014         }
1015         if (vsoc_dev.enabled_device) {
1016                 pci_disable_device(pdev);
1017                 vsoc_dev.enabled_device = false;
1018         }
1019         /* Do this last: it indicates that the device is not initialized. */
1020         vsoc_dev.dev = NULL;
1021 }
1022 
1023 static void __exit vsoc_cleanup_module(void)
1024 {
1025         vsoc_remove_device(vsoc_dev.dev);
1026         pci_unregister_driver(&vsoc_pci_driver);
1027 }
1028 
1029 static int __init vsoc_init_module(void)
1030 {
1031         int err = -ENOMEM;
1032 
1033         INIT_LIST_HEAD(&vsoc_dev.permissions);
1034         mutex_init(&vsoc_dev.mtx);
1035 
1036         err = pci_register_driver(&vsoc_pci_driver);
1037         if (err < 0)
1038                 return err;
1039         return 0;
1040 }
1041 
1042 static int vsoc_open(struct inode *inode, struct file *filp)
1043 {
1044         /* Can't use vsoc_validate_filep because filp is still incomplete */
1045         int ret = vsoc_validate_inode(inode);
1046 
1047         if (ret)
1048                 return ret;
1049         filp->private_data =
1050                 kzalloc(sizeof(struct vsoc_private_data), GFP_KERNEL);
1051         if (!filp->private_data)
1052                 return -ENOMEM;
1053         return 0;
1054 }
1055 
1056 static int vsoc_release(struct inode *inode, struct file *filp)
1057 {
1058         struct vsoc_private_data *private_data = NULL;
1059         struct fd_scoped_permission_node *node = NULL;
1060         struct vsoc_device_region *owner_region_p = NULL;
1061         int retval = vsoc_validate_filep(filp);
1062 
1063         if (retval)
1064                 return retval;
1065         private_data = (struct vsoc_private_data *)filp->private_data;
1066         if (!private_data)
1067                 return 0;
1068 
1069         node = private_data->fd_scoped_permission_node;
1070         if (node) {
1071                 owner_region_p = vsoc_region_from_inode(inode);
1072                 if (owner_region_p->managed_by != VSOC_REGION_WHOLE) {
1073                         owner_region_p =
1074                             &vsoc_dev.regions[owner_region_p->managed_by];
1075                 }
1076                 do_destroy_fd_scoped_permission_node(owner_region_p, node);
1077                 private_data->fd_scoped_permission_node = NULL;
1078         }
1079         kfree(private_data);
1080         filp->private_data = NULL;
1081 
1082         return 0;
1083 }
1084 
1085 /*
1086  * Returns the device relative offset and length of the area specified by the
1087  * fd scoped permission. If there is no fd scoped permission set, a default
1088  * permission covering the entire region is assumed, unless the region is owned
1089  * by another one, in which case the default is a permission with zero size.
1090  */
1091 static ssize_t vsoc_get_area(struct file *filp, __u32 *area_offset)
1092 {
1093         __u32 off = 0;
1094         ssize_t length = 0;
1095         struct vsoc_device_region *region_p;
1096         struct fd_scoped_permission *perm;
1097 
1098         region_p = vsoc_region_from_filep(filp);
1099         off = region_p->region_begin_offset;
1100         perm = &((struct vsoc_private_data *)filp->private_data)->
1101                 fd_scoped_permission_node->permission;
1102         if (perm) {
1103                 off += perm->begin_offset;
1104                 length = perm->end_offset - perm->begin_offset;
1105         } else if (region_p->managed_by == VSOC_REGION_WHOLE) {
1106                 /* No permission set and the regions is not owned by another,
1107                  * default to full region access.
1108                  */
1109                 length = vsoc_device_region_size(region_p);
1110         } else {
1111                 /* return zero length, access is denied. */
1112                 length = 0;
1113         }
1114         if (area_offset)
1115                 *area_offset = off;
1116         return length;
1117 }
1118 
1119 static int vsoc_mmap(struct file *filp, struct vm_area_struct *vma)
1120 {
1121         unsigned long len = vma->vm_end - vma->vm_start;
1122         __u32 area_off;
1123         phys_addr_t mem_off;
1124         ssize_t area_len;
1125         int retval = vsoc_validate_filep(filp);
1126 
1127         if (retval)
1128                 return retval;
1129         area_len = vsoc_get_area(filp, &area_off);
1130         /* Add the requested offset */
1131         area_off += (vma->vm_pgoff << PAGE_SHIFT);
1132         area_len -= (vma->vm_pgoff << PAGE_SHIFT);
1133         if (area_len < len)
1134                 return -EINVAL;
1135         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1136         mem_off = shm_off_to_phys_addr(area_off);
1137         if (io_remap_pfn_range(vma, vma->vm_start, mem_off >> PAGE_SHIFT,
1138                                len, vma->vm_page_prot))
1139                 return -EAGAIN;
1140         return 0;
1141 }
1142 
1143 module_init(vsoc_init_module);
1144 module_exit(vsoc_cleanup_module);
1145 
1146 MODULE_LICENSE("GPL");
1147 MODULE_AUTHOR("Greg Hartman <ghartman@google.com>");
1148 MODULE_DESCRIPTION("VSoC interpretation of QEmu's ivshmem device");
1149 MODULE_VERSION("1.0");
/* [<][>][^][v][top][bottom][index][help] */
root/drivers/staging/android/vsoc.c

DEFINITIONS
