1 /*- 2 * Copyright (c) 2010 Isilon Systems, Inc. 3 * Copyright (c) 2010 iX Systems, Inc. 4 * Copyright (c) 2010 Panasas, Inc. 5 * Copyright (c) 2013-2021 Mellanox Technologies, Ltd. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 #include "opt_global.h" 32 #include "opt_stack.h" 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/malloc.h> 37 #include <sys/kernel.h> 38 #include <sys/sysctl.h> 39 #include <sys/proc.h> 40 #include <sys/sglist.h> 41 #include <sys/sleepqueue.h> 42 #include <sys/refcount.h> 43 #include <sys/lock.h> 44 #include <sys/mutex.h> 45 #include <sys/bus.h> 46 #include <sys/eventhandler.h> 47 #include <sys/fcntl.h> 48 #include <sys/file.h> 49 #include <sys/filio.h> 50 #include <sys/rwlock.h> 51 #include <sys/mman.h> 52 #include <sys/stack.h> 53 #include <sys/sysent.h> 54 #include <sys/time.h> 55 #include <sys/user.h> 56 57 #include <vm/vm.h> 58 #include <vm/pmap.h> 59 #include <vm/vm_object.h> 60 #include <vm/vm_page.h> 61 #include <vm/vm_pager.h> 62 63 #include <machine/stdarg.h> 64 65 #if defined(__i386__) || defined(__amd64__) 66 #include <machine/cputypes.h> 67 #include <machine/md_var.h> 68 #endif 69 70 #include <linux/kobject.h> 71 #include <linux/cpu.h> 72 #include <linux/device.h> 73 #include <linux/slab.h> 74 #include <linux/module.h> 75 #include <linux/moduleparam.h> 76 #include <linux/cdev.h> 77 #include <linux/file.h> 78 #include <linux/sysfs.h> 79 #include <linux/mm.h> 80 #include <linux/io.h> 81 #include <linux/vmalloc.h> 82 #include <linux/netdevice.h> 83 #include <linux/timer.h> 84 #include <linux/interrupt.h> 85 #include <linux/uaccess.h> 86 #include <linux/utsname.h> 87 #include <linux/list.h> 88 #include <linux/kthread.h> 89 #include <linux/kernel.h> 90 #include <linux/compat.h> 91 #include <linux/io-mapping.h> 92 #include <linux/poll.h> 93 #include <linux/smp.h> 94 #include <linux/wait_bit.h> 95 #include <linux/rcupdate.h> 96 #include <linux/interval_tree.h> 97 #include <linux/interval_tree_generic.h> 98 99 #if defined(__i386__) || defined(__amd64__) 100 #include <asm/smp.h> 101 #include <asm/processor.h> 102 #endif 103 104 #include <xen/xen.h> 105 #ifdef XENHVM 106 #undef xen_pv_domain 107 #undef xen_initial_domain 108 /* xen/xen-os.h redefines __must_check */ 109 #undef __must_check 110 #include <xen/xen-os.h> 111 #endif 112 113 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 114 "LinuxKPI parameters"); 115 116 int linuxkpi_debug; 117 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, 118 &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); 119 120 int linuxkpi_rcu_debug; 121 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, rcu_debug, CTLFLAG_RWTUN, 122 &linuxkpi_rcu_debug, 0, "Set to enable RCU warning. Clear to disable."); 123 124 int linuxkpi_warn_dump_stack = 0; 125 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN, 126 &linuxkpi_warn_dump_stack, 0, 127 "Set to enable stack traces from WARN_ON(). Clear to disable."); 128 129 static struct timeval lkpi_net_lastlog; 130 static int lkpi_net_curpps; 131 static int lkpi_net_maxpps = 99; 132 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN, 133 &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second."); 134 135 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat"); 136 137 #include <linux/rbtree.h> 138 /* Undo Linux compat changes. */ 139 #undef RB_ROOT 140 #undef file 141 #undef cdev 142 #define RB_ROOT(head) (head)->rbh_root 143 144 static void linux_destroy_dev(struct linux_cdev *); 145 static void linux_cdev_deref(struct linux_cdev *ldev); 146 static struct vm_area_struct *linux_cdev_handle_find(void *handle); 147 148 cpumask_t cpu_online_mask; 149 static cpumask_t **static_single_cpu_mask; 150 static cpumask_t *static_single_cpu_mask_lcs; 151 struct kobject linux_class_root; 152 struct device linux_root_device; 153 struct class linux_class_misc; 154 struct list_head pci_drivers; 155 struct list_head pci_devices; 156 spinlock_t pci_lock; 157 struct uts_namespace init_uts_ns; 158 159 unsigned long linux_timer_hz_mask; 160 161 wait_queue_head_t linux_bit_waitq; 162 wait_queue_head_t linux_var_waitq; 163 164 int 165 panic_cmp(struct rb_node *one, struct rb_node *two) 166 { 167 panic("no cmp"); 168 } 169 170 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); 171 172 #define START(node) ((node)->start) 173 #define LAST(node) ((node)->last) 174 175 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START, 176 LAST,, lkpi_interval_tree) 177 178 static void 179 linux_device_release(struct device *dev) 180 { 181 pr_debug("linux_device_release: %s\n", dev_name(dev)); 182 kfree(dev); 183 } 184 185 static ssize_t 186 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) 187 { 188 struct class_attribute *dattr; 189 ssize_t error; 190 191 dattr = container_of(attr, struct class_attribute, attr); 192 error = -EIO; 193 if (dattr->show) 194 error = dattr->show(container_of(kobj, struct class, kobj), 195 dattr, buf); 196 return (error); 197 } 198 199 static ssize_t 200 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, 201 size_t count) 202 { 203 struct class_attribute *dattr; 204 ssize_t error; 205 206 dattr = container_of(attr, struct class_attribute, attr); 207 error = -EIO; 208 if (dattr->store) 209 error = dattr->store(container_of(kobj, struct class, kobj), 210 dattr, buf, count); 211 return (error); 212 } 213 214 static void 215 linux_class_release(struct kobject *kobj) 216 { 217 struct class *class; 218 219 class = container_of(kobj, struct class, kobj); 220 if (class->class_release) 221 class->class_release(class); 222 } 223 224 static const struct sysfs_ops linux_class_sysfs = { 225 .show = linux_class_show, 226 .store = linux_class_store, 227 }; 228 229 const struct kobj_type linux_class_ktype = { 230 .release = linux_class_release, 231 .sysfs_ops = &linux_class_sysfs 232 }; 233 234 static void 235 linux_dev_release(struct kobject *kobj) 236 { 237 struct device *dev; 238 239 dev = container_of(kobj, struct device, kobj); 240 /* This is the precedence defined by linux. */ 241 if (dev->release) 242 dev->release(dev); 243 else if (dev->class && dev->class->dev_release) 244 dev->class->dev_release(dev); 245 } 246 247 static ssize_t 248 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) 249 { 250 struct device_attribute *dattr; 251 ssize_t error; 252 253 dattr = container_of(attr, struct device_attribute, attr); 254 error = -EIO; 255 if (dattr->show) 256 error = dattr->show(container_of(kobj, struct device, kobj), 257 dattr, buf); 258 return (error); 259 } 260 261 static ssize_t 262 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, 263 size_t count) 264 { 265 struct device_attribute *dattr; 266 ssize_t error; 267 268 dattr = container_of(attr, struct device_attribute, attr); 269 error = -EIO; 270 if (dattr->store) 271 error = dattr->store(container_of(kobj, struct device, kobj), 272 dattr, buf, count); 273 return (error); 274 } 275 276 static const struct sysfs_ops linux_dev_sysfs = { 277 .show = linux_dev_show, 278 .store = linux_dev_store, 279 }; 280 281 const struct kobj_type linux_dev_ktype = { 282 .release = linux_dev_release, 283 .sysfs_ops = &linux_dev_sysfs 284 }; 285 286 struct device * 287 device_create(struct class *class, struct device *parent, dev_t devt, 288 void *drvdata, const char *fmt, ...) 289 { 290 struct device *dev; 291 va_list args; 292 293 dev = kzalloc(sizeof(*dev), M_WAITOK); 294 dev->parent = parent; 295 dev->class = class; 296 dev->devt = devt; 297 dev->driver_data = drvdata; 298 dev->release = linux_device_release; 299 va_start(args, fmt); 300 kobject_set_name_vargs(&dev->kobj, fmt, args); 301 va_end(args); 302 device_register(dev); 303 304 return (dev); 305 } 306 307 struct device * 308 device_create_groups_vargs(struct class *class, struct device *parent, 309 dev_t devt, void *drvdata, const struct attribute_group **groups, 310 const char *fmt, va_list args) 311 { 312 struct device *dev = NULL; 313 int retval = -ENODEV; 314 315 if (class == NULL || IS_ERR(class)) 316 goto error; 317 318 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 319 if (!dev) { 320 retval = -ENOMEM; 321 goto error; 322 } 323 324 dev->devt = devt; 325 dev->class = class; 326 dev->parent = parent; 327 dev->groups = groups; 328 dev->release = device_create_release; 329 /* device_initialize() needs the class and parent to be set */ 330 device_initialize(dev); 331 dev_set_drvdata(dev, drvdata); 332 333 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 334 if (retval) 335 goto error; 336 337 retval = device_add(dev); 338 if (retval) 339 goto error; 340 341 return dev; 342 343 error: 344 put_device(dev); 345 return ERR_PTR(retval); 346 } 347 348 struct class * 349 class_create(struct module *owner, const char *name) 350 { 351 struct class *class; 352 int error; 353 354 class = kzalloc(sizeof(*class), M_WAITOK); 355 class->owner = owner; 356 class->name = name; 357 class->class_release = linux_class_kfree; 358 error = class_register(class); 359 if (error) { 360 kfree(class); 361 return (NULL); 362 } 363 364 return (class); 365 } 366 367 static void 368 linux_kq_lock(void *arg) 369 { 370 spinlock_t *s = arg; 371 372 spin_lock(s); 373 } 374 static void 375 linux_kq_unlock(void *arg) 376 { 377 spinlock_t *s = arg; 378 379 spin_unlock(s); 380 } 381 382 static void 383 linux_kq_assert_lock(void *arg, int what) 384 { 385 #ifdef INVARIANTS 386 spinlock_t *s = arg; 387 388 if (what == LA_LOCKED) 389 mtx_assert(s, MA_OWNED); 390 else 391 mtx_assert(s, MA_NOTOWNED); 392 #endif 393 } 394 395 static void 396 linux_file_kqfilter_poll(struct linux_file *, int); 397 398 struct linux_file * 399 linux_file_alloc(void) 400 { 401 struct linux_file *filp; 402 403 filp = kzalloc(sizeof(*filp), GFP_KERNEL); 404 405 /* set initial refcount */ 406 filp->f_count = 1; 407 408 /* setup fields needed by kqueue support */ 409 spin_lock_init(&filp->f_kqlock); 410 knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, 411 linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock); 412 413 return (filp); 414 } 415 416 void 417 linux_file_free(struct linux_file *filp) 418 { 419 if (filp->_file == NULL) { 420 if (filp->f_op != NULL && filp->f_op->release != NULL) 421 filp->f_op->release(filp->f_vnode, filp); 422 if (filp->f_shmem != NULL) 423 vm_object_deallocate(filp->f_shmem); 424 kfree_rcu(filp, rcu); 425 } else { 426 /* 427 * The close method of the character device or file 428 * will free the linux_file structure: 429 */ 430 _fdrop(filp->_file, curthread); 431 } 432 } 433 434 struct linux_cdev * 435 cdev_alloc(void) 436 { 437 struct linux_cdev *cdev; 438 439 cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK); 440 kobject_init(&cdev->kobj, &linux_cdev_ktype); 441 cdev->refs = 1; 442 return (cdev); 443 } 444 445 static int 446 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, 447 vm_page_t *mres) 448 { 449 struct vm_area_struct *vmap; 450 451 vmap = linux_cdev_handle_find(vm_obj->handle); 452 453 MPASS(vmap != NULL); 454 MPASS(vmap->vm_private_data == vm_obj->handle); 455 456 if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { 457 vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; 458 vm_page_t page; 459 460 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 461 /* 462 * If the passed in result page is a fake 463 * page, update it with the new physical 464 * address. 465 */ 466 page = *mres; 467 vm_page_updatefake(page, paddr, vm_obj->memattr); 468 } else { 469 /* 470 * Replace the passed in "mres" page with our 471 * own fake page and free up the all of the 472 * original pages. 473 */ 474 VM_OBJECT_WUNLOCK(vm_obj); 475 page = vm_page_getfake(paddr, vm_obj->memattr); 476 VM_OBJECT_WLOCK(vm_obj); 477 478 vm_page_replace(page, vm_obj, (*mres)->pindex, *mres); 479 *mres = page; 480 } 481 vm_page_valid(page); 482 return (VM_PAGER_OK); 483 } 484 return (VM_PAGER_FAIL); 485 } 486 487 static int 488 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, 489 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 490 { 491 struct vm_area_struct *vmap; 492 int err; 493 494 /* get VM area structure */ 495 vmap = linux_cdev_handle_find(vm_obj->handle); 496 MPASS(vmap != NULL); 497 MPASS(vmap->vm_private_data == vm_obj->handle); 498 499 VM_OBJECT_WUNLOCK(vm_obj); 500 501 linux_set_current(curthread); 502 503 down_write(&vmap->vm_mm->mmap_sem); 504 if (unlikely(vmap->vm_ops == NULL)) { 505 err = VM_FAULT_SIGBUS; 506 } else { 507 struct vm_fault vmf; 508 509 /* fill out VM fault structure */ 510 vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); 511 vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; 512 vmf.pgoff = 0; 513 vmf.page = NULL; 514 vmf.vma = vmap; 515 516 vmap->vm_pfn_count = 0; 517 vmap->vm_pfn_pcount = &vmap->vm_pfn_count; 518 vmap->vm_obj = vm_obj; 519 520 err = vmap->vm_ops->fault(&vmf); 521 522 while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { 523 kern_yield(PRI_USER); 524 err = vmap->vm_ops->fault(&vmf); 525 } 526 } 527 528 /* translate return code */ 529 switch (err) { 530 case VM_FAULT_OOM: 531 err = VM_PAGER_AGAIN; 532 break; 533 case VM_FAULT_SIGBUS: 534 err = VM_PAGER_BAD; 535 break; 536 case VM_FAULT_NOPAGE: 537 /* 538 * By contract the fault handler will return having 539 * busied all the pages itself. If pidx is already 540 * found in the object, it will simply xbusy the first 541 * page and return with vm_pfn_count set to 1. 542 */ 543 *first = vmap->vm_pfn_first; 544 *last = *first + vmap->vm_pfn_count - 1; 545 err = VM_PAGER_OK; 546 break; 547 default: 548 err = VM_PAGER_ERROR; 549 break; 550 } 551 up_write(&vmap->vm_mm->mmap_sem); 552 VM_OBJECT_WLOCK(vm_obj); 553 return (err); 554 } 555 556 static struct rwlock linux_vma_lock; 557 static TAILQ_HEAD(, vm_area_struct) linux_vma_head = 558 TAILQ_HEAD_INITIALIZER(linux_vma_head); 559 560 static void 561 linux_cdev_handle_free(struct vm_area_struct *vmap) 562 { 563 /* Drop reference on vm_file */ 564 if (vmap->vm_file != NULL) 565 fput(vmap->vm_file); 566 567 /* Drop reference on mm_struct */ 568 mmput(vmap->vm_mm); 569 570 kfree(vmap); 571 } 572 573 static void 574 linux_cdev_handle_remove(struct vm_area_struct *vmap) 575 { 576 rw_wlock(&linux_vma_lock); 577 TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); 578 rw_wunlock(&linux_vma_lock); 579 } 580 581 static struct vm_area_struct * 582 linux_cdev_handle_find(void *handle) 583 { 584 struct vm_area_struct *vmap; 585 586 rw_rlock(&linux_vma_lock); 587 TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { 588 if (vmap->vm_private_data == handle) 589 break; 590 } 591 rw_runlock(&linux_vma_lock); 592 return (vmap); 593 } 594 595 static int 596 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 597 vm_ooffset_t foff, struct ucred *cred, u_short *color) 598 { 599 600 MPASS(linux_cdev_handle_find(handle) != NULL); 601 *color = 0; 602 return (0); 603 } 604 605 static void 606 linux_cdev_pager_dtor(void *handle) 607 { 608 const struct vm_operations_struct *vm_ops; 609 struct vm_area_struct *vmap; 610 611 vmap = linux_cdev_handle_find(handle); 612 MPASS(vmap != NULL); 613 614 /* 615 * Remove handle before calling close operation to prevent 616 * other threads from reusing the handle pointer. 617 */ 618 linux_cdev_handle_remove(vmap); 619 620 down_write(&vmap->vm_mm->mmap_sem); 621 vm_ops = vmap->vm_ops; 622 if (likely(vm_ops != NULL)) 623 vm_ops->close(vmap); 624 up_write(&vmap->vm_mm->mmap_sem); 625 626 linux_cdev_handle_free(vmap); 627 } 628 629 static struct cdev_pager_ops linux_cdev_pager_ops[2] = { 630 { 631 /* OBJT_MGTDEVICE */ 632 .cdev_pg_populate = linux_cdev_pager_populate, 633 .cdev_pg_ctor = linux_cdev_pager_ctor, 634 .cdev_pg_dtor = linux_cdev_pager_dtor 635 }, 636 { 637 /* OBJT_DEVICE */ 638 .cdev_pg_fault = linux_cdev_pager_fault, 639 .cdev_pg_ctor = linux_cdev_pager_ctor, 640 .cdev_pg_dtor = linux_cdev_pager_dtor 641 }, 642 }; 643 644 int 645 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 646 unsigned long size) 647 { 648 vm_object_t obj; 649 vm_page_t m; 650 651 obj = vma->vm_obj; 652 if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) 653 return (-ENOTSUP); 654 VM_OBJECT_RLOCK(obj); 655 for (m = vm_page_find_least(obj, OFF_TO_IDX(address)); 656 m != NULL && m->pindex < OFF_TO_IDX(address + size); 657 m = TAILQ_NEXT(m, listq)) 658 pmap_remove_all(m); 659 VM_OBJECT_RUNLOCK(obj); 660 return (0); 661 } 662 663 void 664 vma_set_file(struct vm_area_struct *vma, struct linux_file *file) 665 { 666 struct linux_file *tmp; 667 668 /* Changing an anonymous vma with this is illegal */ 669 get_file(file); 670 tmp = vma->vm_file; 671 vma->vm_file = file; 672 fput(tmp); 673 } 674 675 static struct file_operations dummy_ldev_ops = { 676 /* XXXKIB */ 677 }; 678 679 static struct linux_cdev dummy_ldev = { 680 .ops = &dummy_ldev_ops, 681 }; 682 683 #define LDEV_SI_DTR 0x0001 684 #define LDEV_SI_REF 0x0002 685 686 static void 687 linux_get_fop(struct linux_file *filp, const struct file_operations **fop, 688 struct linux_cdev **dev) 689 { 690 struct linux_cdev *ldev; 691 u_int siref; 692 693 ldev = filp->f_cdev; 694 *fop = filp->f_op; 695 if (ldev != NULL) { 696 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 697 refcount_acquire(&ldev->refs); 698 } else { 699 for (siref = ldev->siref;;) { 700 if ((siref & LDEV_SI_DTR) != 0) { 701 ldev = &dummy_ldev; 702 *fop = ldev->ops; 703 siref = ldev->siref; 704 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 705 } else if (atomic_fcmpset_int(&ldev->siref, 706 &siref, siref + LDEV_SI_REF)) { 707 break; 708 } 709 } 710 } 711 } 712 *dev = ldev; 713 } 714 715 static void 716 linux_drop_fop(struct linux_cdev *ldev) 717 { 718 719 if (ldev == NULL) 720 return; 721 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 722 linux_cdev_deref(ldev); 723 } else { 724 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 725 MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); 726 atomic_subtract_int(&ldev->siref, LDEV_SI_REF); 727 } 728 } 729 730 #define OPW(fp,td,code) ({ \ 731 struct file *__fpop; \ 732 __typeof(code) __retval; \ 733 \ 734 __fpop = (td)->td_fpop; \ 735 (td)->td_fpop = (fp); \ 736 __retval = (code); \ 737 (td)->td_fpop = __fpop; \ 738 __retval; \ 739 }) 740 741 static int 742 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, 743 struct file *file) 744 { 745 struct linux_cdev *ldev; 746 struct linux_file *filp; 747 const struct file_operations *fop; 748 int error; 749 750 ldev = dev->si_drv1; 751 752 filp = linux_file_alloc(); 753 filp->f_dentry = &filp->f_dentry_store; 754 filp->f_op = ldev->ops; 755 filp->f_mode = file->f_flag; 756 filp->f_flags = file->f_flag; 757 filp->f_vnode = file->f_vnode; 758 filp->_file = file; 759 refcount_acquire(&ldev->refs); 760 filp->f_cdev = ldev; 761 762 linux_set_current(td); 763 linux_get_fop(filp, &fop, &ldev); 764 765 if (fop->open != NULL) { 766 error = -fop->open(file->f_vnode, filp); 767 if (error != 0) { 768 linux_drop_fop(ldev); 769 linux_cdev_deref(filp->f_cdev); 770 kfree(filp); 771 return (error); 772 } 773 } 774 775 /* hold on to the vnode - used for fstat() */ 776 vhold(filp->f_vnode); 777 778 /* release the file from devfs */ 779 finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); 780 linux_drop_fop(ldev); 781 return (ENXIO); 782 } 783 784 #define LINUX_IOCTL_MIN_PTR 0x10000UL 785 #define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) 786 787 static inline int 788 linux_remap_address(void **uaddr, size_t len) 789 { 790 uintptr_t uaddr_val = (uintptr_t)(*uaddr); 791 792 if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && 793 uaddr_val < LINUX_IOCTL_MAX_PTR)) { 794 struct task_struct *pts = current; 795 if (pts == NULL) { 796 *uaddr = NULL; 797 return (1); 798 } 799 800 /* compute data offset */ 801 uaddr_val -= LINUX_IOCTL_MIN_PTR; 802 803 /* check that length is within bounds */ 804 if ((len > IOCPARM_MAX) || 805 (uaddr_val + len) > pts->bsd_ioctl_len) { 806 *uaddr = NULL; 807 return (1); 808 } 809 810 /* re-add kernel buffer address */ 811 uaddr_val += (uintptr_t)pts->bsd_ioctl_data; 812 813 /* update address location */ 814 *uaddr = (void *)uaddr_val; 815 return (1); 816 } 817 return (0); 818 } 819 820 int 821 linux_copyin(const void *uaddr, void *kaddr, size_t len) 822 { 823 if (linux_remap_address(__DECONST(void **, &uaddr), len)) { 824 if (uaddr == NULL) 825 return (-EFAULT); 826 memcpy(kaddr, uaddr, len); 827 return (0); 828 } 829 return (-copyin(uaddr, kaddr, len)); 830 } 831 832 int 833 linux_copyout(const void *kaddr, void *uaddr, size_t len) 834 { 835 if (linux_remap_address(&uaddr, len)) { 836 if (uaddr == NULL) 837 return (-EFAULT); 838 memcpy(uaddr, kaddr, len); 839 return (0); 840 } 841 return (-copyout(kaddr, uaddr, len)); 842 } 843 844 size_t 845 linux_clear_user(void *_uaddr, size_t _len) 846 { 847 uint8_t *uaddr = _uaddr; 848 size_t len = _len; 849 850 /* make sure uaddr is aligned before going into the fast loop */ 851 while (((uintptr_t)uaddr & 7) != 0 && len > 7) { 852 if (subyte(uaddr, 0)) 853 return (_len); 854 uaddr++; 855 len--; 856 } 857 858 /* zero 8 bytes at a time */ 859 while (len > 7) { 860 #ifdef __LP64__ 861 if (suword64(uaddr, 0)) 862 return (_len); 863 #else 864 if (suword32(uaddr, 0)) 865 return (_len); 866 if (suword32(uaddr + 4, 0)) 867 return (_len); 868 #endif 869 uaddr += 8; 870 len -= 8; 871 } 872 873 /* zero fill end, if any */ 874 while (len > 0) { 875 if (subyte(uaddr, 0)) 876 return (_len); 877 uaddr++; 878 len--; 879 } 880 return (0); 881 } 882 883 int 884 linux_access_ok(const void *uaddr, size_t len) 885 { 886 uintptr_t saddr; 887 uintptr_t eaddr; 888 889 /* get start and end address */ 890 saddr = (uintptr_t)uaddr; 891 eaddr = (uintptr_t)uaddr + len; 892 893 /* verify addresses are valid for userspace */ 894 return ((saddr == eaddr) || 895 (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); 896 } 897 898 /* 899 * This function should return either EINTR or ERESTART depending on 900 * the signal type sent to this thread: 901 */ 902 static int 903 linux_get_error(struct task_struct *task, int error) 904 { 905 /* check for signal type interrupt code */ 906 if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { 907 error = -linux_schedule_get_interrupt_value(task); 908 if (error == 0) 909 error = EINTR; 910 } 911 return (error); 912 } 913 914 static int 915 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, 916 const struct file_operations *fop, u_long cmd, caddr_t data, 917 struct thread *td) 918 { 919 struct task_struct *task = current; 920 unsigned size; 921 int error; 922 923 size = IOCPARM_LEN(cmd); 924 /* refer to logic in sys_ioctl() */ 925 if (size > 0) { 926 /* 927 * Setup hint for linux_copyin() and linux_copyout(). 928 * 929 * Background: Linux code expects a user-space address 930 * while FreeBSD supplies a kernel-space address. 931 */ 932 task->bsd_ioctl_data = data; 933 task->bsd_ioctl_len = size; 934 data = (void *)LINUX_IOCTL_MIN_PTR; 935 } else { 936 /* fetch user-space pointer */ 937 data = *(void **)data; 938 } 939 #ifdef COMPAT_FREEBSD32 940 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 941 /* try the compat IOCTL handler first */ 942 if (fop->compat_ioctl != NULL) { 943 error = -OPW(fp, td, fop->compat_ioctl(filp, 944 cmd, (u_long)data)); 945 } else { 946 error = ENOTTY; 947 } 948 949 /* fallback to the regular IOCTL handler, if any */ 950 if (error == ENOTTY && fop->unlocked_ioctl != NULL) { 951 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 952 cmd, (u_long)data)); 953 } 954 } else 955 #endif 956 { 957 if (fop->unlocked_ioctl != NULL) { 958 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 959 cmd, (u_long)data)); 960 } else { 961 error = ENOTTY; 962 } 963 } 964 if (size > 0) { 965 task->bsd_ioctl_data = NULL; 966 task->bsd_ioctl_len = 0; 967 } 968 969 if (error == EWOULDBLOCK) { 970 /* update kqfilter status, if any */ 971 linux_file_kqfilter_poll(filp, 972 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 973 } else { 974 error = linux_get_error(task, error); 975 } 976 return (error); 977 } 978 979 #define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) 980 981 /* 982 * This function atomically updates the poll wakeup state and returns 983 * the previous state at the time of update. 984 */ 985 static uint8_t 986 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) 987 { 988 int c, old; 989 990 c = v->counter; 991 992 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) 993 c = old; 994 995 return (c); 996 } 997 998 static int 999 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) 1000 { 1001 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1002 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1003 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1004 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, 1005 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ 1006 }; 1007 struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); 1008 1009 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1010 case LINUX_FWQ_STATE_QUEUED: 1011 linux_poll_wakeup(filp); 1012 return (1); 1013 default: 1014 return (0); 1015 } 1016 } 1017 1018 void 1019 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) 1020 { 1021 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1022 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, 1023 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1024 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ 1025 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, 1026 }; 1027 1028 /* check if we are called inside the select system call */ 1029 if (p == LINUX_POLL_TABLE_NORMAL) 1030 selrecord(curthread, &filp->f_selinfo); 1031 1032 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1033 case LINUX_FWQ_STATE_INIT: 1034 /* NOTE: file handles can only belong to one wait-queue */ 1035 filp->f_wait_queue.wqh = wqh; 1036 filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; 1037 add_wait_queue(wqh, &filp->f_wait_queue.wq); 1038 atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); 1039 break; 1040 default: 1041 break; 1042 } 1043 } 1044 1045 static void 1046 linux_poll_wait_dequeue(struct linux_file *filp) 1047 { 1048 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1049 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1050 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, 1051 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, 1052 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, 1053 }; 1054 1055 seldrain(&filp->f_selinfo); 1056 1057 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1058 case LINUX_FWQ_STATE_NOT_READY: 1059 case LINUX_FWQ_STATE_QUEUED: 1060 case LINUX_FWQ_STATE_READY: 1061 remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); 1062 break; 1063 default: 1064 break; 1065 } 1066 } 1067 1068 void 1069 linux_poll_wakeup(struct linux_file *filp) 1070 { 1071 /* this function should be NULL-safe */ 1072 if (filp == NULL) 1073 return; 1074 1075 selwakeup(&filp->f_selinfo); 1076 1077 spin_lock(&filp->f_kqlock); 1078 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | 1079 LINUX_KQ_FLAG_NEED_WRITE; 1080 1081 /* make sure the "knote" gets woken up */ 1082 KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); 1083 spin_unlock(&filp->f_kqlock); 1084 } 1085 1086 static void 1087 linux_file_kqfilter_detach(struct knote *kn) 1088 { 1089 struct linux_file *filp = kn->kn_hook; 1090 1091 spin_lock(&filp->f_kqlock); 1092 knlist_remove(&filp->f_selinfo.si_note, kn, 1); 1093 spin_unlock(&filp->f_kqlock); 1094 } 1095 1096 static int 1097 linux_file_kqfilter_read_event(struct knote *kn, long hint) 1098 { 1099 struct linux_file *filp = kn->kn_hook; 1100 1101 mtx_assert(&filp->f_kqlock, MA_OWNED); 1102 1103 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); 1104 } 1105 1106 static int 1107 linux_file_kqfilter_write_event(struct knote *kn, long hint) 1108 { 1109 struct linux_file *filp = kn->kn_hook; 1110 1111 mtx_assert(&filp->f_kqlock, MA_OWNED); 1112 1113 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); 1114 } 1115 1116 static struct filterops linux_dev_kqfiltops_read = { 1117 .f_isfd = 1, 1118 .f_detach = linux_file_kqfilter_detach, 1119 .f_event = linux_file_kqfilter_read_event, 1120 }; 1121 1122 static struct filterops linux_dev_kqfiltops_write = { 1123 .f_isfd = 1, 1124 .f_detach = linux_file_kqfilter_detach, 1125 .f_event = linux_file_kqfilter_write_event, 1126 }; 1127 1128 static void 1129 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) 1130 { 1131 struct thread *td; 1132 const struct file_operations *fop; 1133 struct linux_cdev *ldev; 1134 int temp; 1135 1136 if ((filp->f_kqflags & kqflags) == 0) 1137 return; 1138 1139 td = curthread; 1140 1141 linux_get_fop(filp, &fop, &ldev); 1142 /* get the latest polling state */ 1143 temp = OPW(filp->_file, td, fop->poll(filp, NULL)); 1144 linux_drop_fop(ldev); 1145 1146 spin_lock(&filp->f_kqlock); 1147 /* clear kqflags */ 1148 filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | 1149 LINUX_KQ_FLAG_NEED_WRITE); 1150 /* update kqflags */ 1151 if ((temp & (POLLIN | POLLOUT)) != 0) { 1152 if ((temp & POLLIN) != 0) 1153 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; 1154 if ((temp & POLLOUT) != 0) 1155 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; 1156 1157 /* make sure the "knote" gets woken up */ 1158 KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); 1159 } 1160 spin_unlock(&filp->f_kqlock); 1161 } 1162 1163 static int 1164 linux_file_kqfilter(struct file *file, struct knote *kn) 1165 { 1166 struct linux_file *filp; 1167 struct thread *td; 1168 int error; 1169 1170 td = curthread; 1171 filp = (struct linux_file *)file->f_data; 1172 filp->f_flags = file->f_flag; 1173 if (filp->f_op->poll == NULL) 1174 return (EINVAL); 1175 1176 spin_lock(&filp->f_kqlock); 1177 switch (kn->kn_filter) { 1178 case EVFILT_READ: 1179 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; 1180 kn->kn_fop = &linux_dev_kqfiltops_read; 1181 kn->kn_hook = filp; 1182 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1183 error = 0; 1184 break; 1185 case EVFILT_WRITE: 1186 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; 1187 kn->kn_fop = &linux_dev_kqfiltops_write; 1188 kn->kn_hook = filp; 1189 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1190 error = 0; 1191 break; 1192 default: 1193 error = EINVAL; 1194 break; 1195 } 1196 spin_unlock(&filp->f_kqlock); 1197 1198 if (error == 0) { 1199 linux_set_current(td); 1200 1201 /* update kqfilter status, if any */ 1202 linux_file_kqfilter_poll(filp, 1203 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 1204 } 1205 return (error); 1206 } 1207 1208 static int 1209 linux_file_mmap_single(struct file *fp, const struct file_operations *fop, 1210 vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, 1211 int nprot, bool is_shared, struct thread *td) 1212 { 1213 struct task_struct *task; 1214 struct vm_area_struct *vmap; 1215 struct mm_struct *mm; 1216 struct linux_file *filp; 1217 vm_memattr_t attr; 1218 int error; 1219 1220 filp = (struct linux_file *)fp->f_data; 1221 filp->f_flags = fp->f_flag; 1222 1223 if (fop->mmap == NULL) 1224 return (EOPNOTSUPP); 1225 1226 linux_set_current(td); 1227 1228 /* 1229 * The same VM object might be shared by multiple processes 1230 * and the mm_struct is usually freed when a process exits. 1231 * 1232 * The atomic reference below makes sure the mm_struct is 1233 * available as long as the vmap is in the linux_vma_head. 1234 */ 1235 task = current; 1236 mm = task->mm; 1237 if (atomic_inc_not_zero(&mm->mm_users) == 0) 1238 return (EINVAL); 1239 1240 vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); 1241 vmap->vm_start = 0; 1242 vmap->vm_end = size; 1243 vmap->vm_pgoff = *offset / PAGE_SIZE; 1244 vmap->vm_pfn = 0; 1245 vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); 1246 if (is_shared) 1247 vmap->vm_flags |= VM_SHARED; 1248 vmap->vm_ops = NULL; 1249 vmap->vm_file = get_file(filp); 1250 vmap->vm_mm = mm; 1251 1252 if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { 1253 error = linux_get_error(task, EINTR); 1254 } else { 1255 error = -OPW(fp, td, fop->mmap(filp, vmap)); 1256 error = linux_get_error(task, error); 1257 up_write(&vmap->vm_mm->mmap_sem); 1258 } 1259 1260 if (error != 0) { 1261 linux_cdev_handle_free(vmap); 1262 return (error); 1263 } 1264 1265 attr = pgprot2cachemode(vmap->vm_page_prot); 1266 1267 if (vmap->vm_ops != NULL) { 1268 struct vm_area_struct *ptr; 1269 void *vm_private_data; 1270 bool vm_no_fault; 1271 1272 if (vmap->vm_ops->open == NULL || 1273 vmap->vm_ops->close == NULL || 1274 vmap->vm_private_data == NULL) { 1275 /* free allocated VM area struct */ 1276 linux_cdev_handle_free(vmap); 1277 return (EINVAL); 1278 } 1279 1280 vm_private_data = vmap->vm_private_data; 1281 1282 rw_wlock(&linux_vma_lock); 1283 TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { 1284 if (ptr->vm_private_data == vm_private_data) 1285 break; 1286 } 1287 /* check if there is an existing VM area struct */ 1288 if (ptr != NULL) { 1289 /* check if the VM area structure is invalid */ 1290 if (ptr->vm_ops == NULL || 1291 ptr->vm_ops->open == NULL || 1292 ptr->vm_ops->close == NULL) { 1293 error = ESTALE; 1294 vm_no_fault = 1; 1295 } else { 1296 error = EEXIST; 1297 vm_no_fault = (ptr->vm_ops->fault == NULL); 1298 } 1299 } else { 1300 /* insert VM area structure into list */ 1301 TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); 1302 error = 0; 1303 vm_no_fault = (vmap->vm_ops->fault == NULL); 1304 } 1305 rw_wunlock(&linux_vma_lock); 1306 1307 if (error != 0) { 1308 /* free allocated VM area struct */ 1309 linux_cdev_handle_free(vmap); 1310 /* check for stale VM area struct */ 1311 if (error != EEXIST) 1312 return (error); 1313 } 1314 1315 /* check if there is no fault handler */ 1316 if (vm_no_fault) { 1317 *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, 1318 &linux_cdev_pager_ops[1], size, nprot, *offset, 1319 td->td_ucred); 1320 } else { 1321 *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, 1322 &linux_cdev_pager_ops[0], size, nprot, *offset, 1323 td->td_ucred); 1324 } 1325 1326 /* check if allocating the VM object failed */ 1327 if (*object == NULL) { 1328 if (error == 0) { 1329 /* remove VM area struct from list */ 1330 linux_cdev_handle_remove(vmap); 1331 /* free allocated VM area struct */ 1332 linux_cdev_handle_free(vmap); 1333 } 1334 return (EINVAL); 1335 } 1336 } else { 1337 struct sglist *sg; 1338 1339 sg = sglist_alloc(1, M_WAITOK); 1340 sglist_append_phys(sg, 1341 (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); 1342 1343 *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, 1344 nprot, 0, td->td_ucred); 1345 1346 linux_cdev_handle_free(vmap); 1347 1348 if (*object == NULL) { 1349 sglist_free(sg); 1350 return (EINVAL); 1351 } 1352 } 1353 1354 if (attr != VM_MEMATTR_DEFAULT) { 1355 VM_OBJECT_WLOCK(*object); 1356 vm_object_set_memattr(*object, attr); 1357 VM_OBJECT_WUNLOCK(*object); 1358 } 1359 *offset = 0; 1360 return (0); 1361 } 1362 1363 struct cdevsw linuxcdevsw = { 1364 .d_version = D_VERSION, 1365 .d_fdopen = linux_dev_fdopen, 1366 .d_name = "lkpidev", 1367 }; 1368 1369 static int 1370 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, 1371 int flags, struct thread *td) 1372 { 1373 struct linux_file *filp; 1374 const struct file_operations *fop; 1375 struct linux_cdev *ldev; 1376 ssize_t bytes; 1377 int error; 1378 1379 error = 0; 1380 filp = (struct linux_file *)file->f_data; 1381 filp->f_flags = file->f_flag; 1382 /* XXX no support for I/O vectors currently */ 1383 if (uio->uio_iovcnt != 1) 1384 return (EOPNOTSUPP); 1385 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1386 return (EINVAL); 1387 linux_set_current(td); 1388 linux_get_fop(filp, &fop, &ldev); 1389 if (fop->read != NULL) { 1390 bytes = OPW(file, td, fop->read(filp, 1391 uio->uio_iov->iov_base, 1392 uio->uio_iov->iov_len, &uio->uio_offset)); 1393 if (bytes >= 0) { 1394 uio->uio_iov->iov_base = 1395 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1396 uio->uio_iov->iov_len -= bytes; 1397 uio->uio_resid -= bytes; 1398 } else { 1399 error = linux_get_error(current, -bytes); 1400 } 1401 } else 1402 error = ENXIO; 1403 1404 /* update kqfilter status, if any */ 1405 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); 1406 linux_drop_fop(ldev); 1407 1408 return (error); 1409 } 1410 1411 static int 1412 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, 1413 int flags, struct thread *td) 1414 { 1415 struct linux_file *filp; 1416 const struct file_operations *fop; 1417 struct linux_cdev *ldev; 1418 ssize_t bytes; 1419 int error; 1420 1421 filp = (struct linux_file *)file->f_data; 1422 filp->f_flags = file->f_flag; 1423 /* XXX no support for I/O vectors currently */ 1424 if (uio->uio_iovcnt != 1) 1425 return (EOPNOTSUPP); 1426 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1427 return (EINVAL); 1428 linux_set_current(td); 1429 linux_get_fop(filp, &fop, &ldev); 1430 if (fop->write != NULL) { 1431 bytes = OPW(file, td, fop->write(filp, 1432 uio->uio_iov->iov_base, 1433 uio->uio_iov->iov_len, &uio->uio_offset)); 1434 if (bytes >= 0) { 1435 uio->uio_iov->iov_base = 1436 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1437 uio->uio_iov->iov_len -= bytes; 1438 uio->uio_resid -= bytes; 1439 error = 0; 1440 } else { 1441 error = linux_get_error(current, -bytes); 1442 } 1443 } else 1444 error = ENXIO; 1445 1446 /* update kqfilter status, if any */ 1447 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); 1448 1449 linux_drop_fop(ldev); 1450 1451 return (error); 1452 } 1453 1454 static int 1455 linux_file_poll(struct file *file, int events, struct ucred *active_cred, 1456 struct thread *td) 1457 { 1458 struct linux_file *filp; 1459 const struct file_operations *fop; 1460 struct linux_cdev *ldev; 1461 int revents; 1462 1463 filp = (struct linux_file *)file->f_data; 1464 filp->f_flags = file->f_flag; 1465 linux_set_current(td); 1466 linux_get_fop(filp, &fop, &ldev); 1467 if (fop->poll != NULL) { 1468 revents = OPW(file, td, fop->poll(filp, 1469 LINUX_POLL_TABLE_NORMAL)) & events; 1470 } else { 1471 revents = 0; 1472 } 1473 linux_drop_fop(ldev); 1474 return (revents); 1475 } 1476 1477 static int 1478 linux_file_close(struct file *file, struct thread *td) 1479 { 1480 struct linux_file *filp; 1481 int (*release)(struct inode *, struct linux_file *); 1482 const struct file_operations *fop; 1483 struct linux_cdev *ldev; 1484 int error; 1485 1486 filp = (struct linux_file *)file->f_data; 1487 1488 KASSERT(file_count(filp) == 0, 1489 ("File refcount(%d) is not zero", file_count(filp))); 1490 1491 if (td == NULL) 1492 td = curthread; 1493 1494 error = 0; 1495 filp->f_flags = file->f_flag; 1496 linux_set_current(td); 1497 linux_poll_wait_dequeue(filp); 1498 linux_get_fop(filp, &fop, &ldev); 1499 /* 1500 * Always use the real release function, if any, to avoid 1501 * leaking device resources: 1502 */ 1503 release = filp->f_op->release; 1504 if (release != NULL) 1505 error = -OPW(file, td, release(filp->f_vnode, filp)); 1506 funsetown(&filp->f_sigio); 1507 if (filp->f_vnode != NULL) 1508 vdrop(filp->f_vnode); 1509 linux_drop_fop(ldev); 1510 ldev = filp->f_cdev; 1511 if (ldev != NULL) 1512 linux_cdev_deref(ldev); 1513 linux_synchronize_rcu(RCU_TYPE_REGULAR); 1514 kfree(filp); 1515 1516 return (error); 1517 } 1518 1519 static int 1520 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, 1521 struct thread *td) 1522 { 1523 struct linux_file *filp; 1524 const struct file_operations *fop; 1525 struct linux_cdev *ldev; 1526 struct fiodgname_arg *fgn; 1527 const char *p; 1528 int error, i; 1529 1530 error = 0; 1531 filp = (struct linux_file *)fp->f_data; 1532 filp->f_flags = fp->f_flag; 1533 linux_get_fop(filp, &fop, &ldev); 1534 1535 linux_set_current(td); 1536 switch (cmd) { 1537 case FIONBIO: 1538 break; 1539 case FIOASYNC: 1540 if (fop->fasync == NULL) 1541 break; 1542 error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); 1543 break; 1544 case FIOSETOWN: 1545 error = fsetown(*(int *)data, &filp->f_sigio); 1546 if (error == 0) { 1547 if (fop->fasync == NULL) 1548 break; 1549 error = -OPW(fp, td, fop->fasync(0, filp, 1550 fp->f_flag & FASYNC)); 1551 } 1552 break; 1553 case FIOGETOWN: 1554 *(int *)data = fgetown(&filp->f_sigio); 1555 break; 1556 case FIODGNAME: 1557 #ifdef COMPAT_FREEBSD32 1558 case FIODGNAME_32: 1559 #endif 1560 if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { 1561 error = ENXIO; 1562 break; 1563 } 1564 fgn = data; 1565 p = devtoname(filp->f_cdev->cdev); 1566 i = strlen(p) + 1; 1567 if (i > fgn->len) { 1568 error = EINVAL; 1569 break; 1570 } 1571 error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i); 1572 break; 1573 default: 1574 error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); 1575 break; 1576 } 1577 linux_drop_fop(ldev); 1578 return (error); 1579 } 1580 1581 static int 1582 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1583 vm_prot_t maxprot, int flags, struct file *fp, 1584 vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) 1585 { 1586 /* 1587 * Character devices do not provide private mappings 1588 * of any kind: 1589 */ 1590 if ((maxprot & VM_PROT_WRITE) == 0 && 1591 (prot & VM_PROT_WRITE) != 0) 1592 return (EACCES); 1593 if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0) 1594 return (EINVAL); 1595 1596 return (linux_file_mmap_single(fp, fop, foff, objsize, objp, 1597 (int)prot, (flags & MAP_SHARED) ? true : false, td)); 1598 } 1599 1600 static int 1601 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1602 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1603 struct thread *td) 1604 { 1605 struct linux_file *filp; 1606 const struct file_operations *fop; 1607 struct linux_cdev *ldev; 1608 struct mount *mp; 1609 struct vnode *vp; 1610 vm_object_t object; 1611 vm_prot_t maxprot; 1612 int error; 1613 1614 filp = (struct linux_file *)fp->f_data; 1615 1616 vp = filp->f_vnode; 1617 if (vp == NULL) 1618 return (EOPNOTSUPP); 1619 1620 /* 1621 * Ensure that file and memory protections are 1622 * compatible. 1623 */ 1624 mp = vp->v_mount; 1625 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1626 maxprot = VM_PROT_NONE; 1627 if ((prot & VM_PROT_EXECUTE) != 0) 1628 return (EACCES); 1629 } else 1630 maxprot = VM_PROT_EXECUTE; 1631 if ((fp->f_flag & FREAD) != 0) 1632 maxprot |= VM_PROT_READ; 1633 else if ((prot & VM_PROT_READ) != 0) 1634 return (EACCES); 1635 1636 /* 1637 * If we are sharing potential changes via MAP_SHARED and we 1638 * are trying to get write permission although we opened it 1639 * without asking for it, bail out. 1640 * 1641 * Note that most character devices always share mappings. 1642 * 1643 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE 1644 * requests rather than doing it here. 1645 */ 1646 if ((flags & MAP_SHARED) != 0) { 1647 if ((fp->f_flag & FWRITE) != 0) 1648 maxprot |= VM_PROT_WRITE; 1649 else if ((prot & VM_PROT_WRITE) != 0) 1650 return (EACCES); 1651 } 1652 maxprot &= cap_maxprot; 1653 1654 linux_get_fop(filp, &fop, &ldev); 1655 error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp, 1656 &foff, fop, &object); 1657 if (error != 0) 1658 goto out; 1659 1660 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1661 foff, FALSE, td); 1662 if (error != 0) 1663 vm_object_deallocate(object); 1664 out: 1665 linux_drop_fop(ldev); 1666 return (error); 1667 } 1668 1669 static int 1670 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 1671 { 1672 struct linux_file *filp; 1673 struct vnode *vp; 1674 int error; 1675 1676 filp = (struct linux_file *)fp->f_data; 1677 if (filp->f_vnode == NULL) 1678 return (EOPNOTSUPP); 1679 1680 vp = filp->f_vnode; 1681 1682 vn_lock(vp, LK_SHARED | LK_RETRY); 1683 error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED); 1684 VOP_UNLOCK(vp); 1685 1686 return (error); 1687 } 1688 1689 static int 1690 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1691 struct filedesc *fdp) 1692 { 1693 struct linux_file *filp; 1694 struct vnode *vp; 1695 int error; 1696 1697 filp = fp->f_data; 1698 vp = filp->f_vnode; 1699 if (vp == NULL) { 1700 error = 0; 1701 kif->kf_type = KF_TYPE_DEV; 1702 } else { 1703 vref(vp); 1704 FILEDESC_SUNLOCK(fdp); 1705 error = vn_fill_kinfo_vnode(vp, kif); 1706 vrele(vp); 1707 kif->kf_type = KF_TYPE_VNODE; 1708 FILEDESC_SLOCK(fdp); 1709 } 1710 return (error); 1711 } 1712 1713 unsigned int 1714 linux_iminor(struct inode *inode) 1715 { 1716 struct linux_cdev *ldev; 1717 1718 if (inode == NULL || inode->v_rdev == NULL || 1719 inode->v_rdev->si_devsw != &linuxcdevsw) 1720 return (-1U); 1721 ldev = inode->v_rdev->si_drv1; 1722 if (ldev == NULL) 1723 return (-1U); 1724 1725 return (minor(ldev->dev)); 1726 } 1727 1728 static int 1729 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td) 1730 { 1731 struct linux_file *filp1, *filp2; 1732 1733 if (fp2->f_type != DTYPE_DEV) 1734 return (3); 1735 1736 filp1 = fp1->f_data; 1737 filp2 = fp2->f_data; 1738 return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev)); 1739 } 1740 1741 struct fileops linuxfileops = { 1742 .fo_read = linux_file_read, 1743 .fo_write = linux_file_write, 1744 .fo_truncate = invfo_truncate, 1745 .fo_kqfilter = linux_file_kqfilter, 1746 .fo_stat = linux_file_stat, 1747 .fo_fill_kinfo = linux_file_fill_kinfo, 1748 .fo_poll = linux_file_poll, 1749 .fo_close = linux_file_close, 1750 .fo_ioctl = linux_file_ioctl, 1751 .fo_mmap = linux_file_mmap, 1752 .fo_chmod = invfo_chmod, 1753 .fo_chown = invfo_chown, 1754 .fo_sendfile = invfo_sendfile, 1755 .fo_cmp = linux_file_kcmp, 1756 .fo_flags = DFLAG_PASSABLE, 1757 }; 1758 1759 /* 1760 * Hash of vmmap addresses. This is infrequently accessed and does not 1761 * need to be particularly large. This is done because we must store the 1762 * caller's idea of the map size to properly unmap. 1763 */ 1764 struct vmmap { 1765 LIST_ENTRY(vmmap) vm_next; 1766 void *vm_addr; 1767 unsigned long vm_size; 1768 }; 1769 1770 struct vmmaphd { 1771 struct vmmap *lh_first; 1772 }; 1773 #define VMMAP_HASH_SIZE 64 1774 #define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) 1775 #define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK 1776 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; 1777 static struct mtx vmmaplock; 1778 1779 static void 1780 vmmap_add(void *addr, unsigned long size) 1781 { 1782 struct vmmap *vmmap; 1783 1784 vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); 1785 mtx_lock(&vmmaplock); 1786 vmmap->vm_size = size; 1787 vmmap->vm_addr = addr; 1788 LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); 1789 mtx_unlock(&vmmaplock); 1790 } 1791 1792 static struct vmmap * 1793 vmmap_remove(void *addr) 1794 { 1795 struct vmmap *vmmap; 1796 1797 mtx_lock(&vmmaplock); 1798 LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) 1799 if (vmmap->vm_addr == addr) 1800 break; 1801 if (vmmap) 1802 LIST_REMOVE(vmmap, vm_next); 1803 mtx_unlock(&vmmaplock); 1804 1805 return (vmmap); 1806 } 1807 1808 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) 1809 void * 1810 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) 1811 { 1812 void *addr; 1813 1814 addr = pmap_mapdev_attr(phys_addr, size, attr); 1815 if (addr == NULL) 1816 return (NULL); 1817 vmmap_add(addr, size); 1818 1819 return (addr); 1820 } 1821 #endif 1822 1823 void 1824 iounmap(void *addr) 1825 { 1826 struct vmmap *vmmap; 1827 1828 vmmap = vmmap_remove(addr); 1829 if (vmmap == NULL) 1830 return; 1831 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) 1832 pmap_unmapdev(addr, vmmap->vm_size); 1833 #endif 1834 kfree(vmmap); 1835 } 1836 1837 void * 1838 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) 1839 { 1840 vm_offset_t off; 1841 size_t size; 1842 1843 size = count * PAGE_SIZE; 1844 off = kva_alloc(size); 1845 if (off == 0) 1846 return (NULL); 1847 vmmap_add((void *)off, size); 1848 pmap_qenter(off, pages, count); 1849 1850 return ((void *)off); 1851 } 1852 1853 void 1854 vunmap(void *addr) 1855 { 1856 struct vmmap *vmmap; 1857 1858 vmmap = vmmap_remove(addr); 1859 if (vmmap == NULL) 1860 return; 1861 pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); 1862 kva_free((vm_offset_t)addr, vmmap->vm_size); 1863 kfree(vmmap); 1864 } 1865 1866 static char * 1867 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap) 1868 { 1869 unsigned int len; 1870 char *p; 1871 va_list aq; 1872 1873 va_copy(aq, ap); 1874 len = vsnprintf(NULL, 0, fmt, aq); 1875 va_end(aq); 1876 1877 if (dev != NULL) 1878 p = devm_kmalloc(dev, len + 1, gfp); 1879 else 1880 p = kmalloc(len + 1, gfp); 1881 if (p != NULL) 1882 vsnprintf(p, len + 1, fmt, ap); 1883 1884 return (p); 1885 } 1886 1887 char * 1888 kvasprintf(gfp_t gfp, const char *fmt, va_list ap) 1889 { 1890 1891 return (devm_kvasprintf(NULL, gfp, fmt, ap)); 1892 } 1893 1894 char * 1895 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...) 1896 { 1897 va_list ap; 1898 char *p; 1899 1900 va_start(ap, fmt); 1901 p = devm_kvasprintf(dev, gfp, fmt, ap); 1902 va_end(ap); 1903 1904 return (p); 1905 } 1906 1907 char * 1908 kasprintf(gfp_t gfp, const char *fmt, ...) 1909 { 1910 va_list ap; 1911 char *p; 1912 1913 va_start(ap, fmt); 1914 p = kvasprintf(gfp, fmt, ap); 1915 va_end(ap); 1916 1917 return (p); 1918 } 1919 1920 static void 1921 linux_timer_callback_wrapper(void *context) 1922 { 1923 struct timer_list *timer; 1924 1925 timer = context; 1926 1927 /* the timer is about to be shutdown permanently */ 1928 if (timer->function == NULL) 1929 return; 1930 1931 if (linux_set_current_flags(curthread, M_NOWAIT)) { 1932 /* try again later */ 1933 callout_reset(&timer->callout, 1, 1934 &linux_timer_callback_wrapper, timer); 1935 return; 1936 } 1937 1938 timer->function(timer->data); 1939 } 1940 1941 int 1942 mod_timer(struct timer_list *timer, int expires) 1943 { 1944 int ret; 1945 1946 timer->expires = expires; 1947 ret = callout_reset(&timer->callout, 1948 linux_timer_jiffies_until(expires), 1949 &linux_timer_callback_wrapper, timer); 1950 1951 MPASS(ret == 0 || ret == 1); 1952 1953 return (ret == 1); 1954 } 1955 1956 void 1957 add_timer(struct timer_list *timer) 1958 { 1959 1960 callout_reset(&timer->callout, 1961 linux_timer_jiffies_until(timer->expires), 1962 &linux_timer_callback_wrapper, timer); 1963 } 1964 1965 void 1966 add_timer_on(struct timer_list *timer, int cpu) 1967 { 1968 1969 callout_reset_on(&timer->callout, 1970 linux_timer_jiffies_until(timer->expires), 1971 &linux_timer_callback_wrapper, timer, cpu); 1972 } 1973 1974 int 1975 del_timer(struct timer_list *timer) 1976 { 1977 1978 if (callout_stop(&(timer)->callout) == -1) 1979 return (0); 1980 return (1); 1981 } 1982 1983 int 1984 del_timer_sync(struct timer_list *timer) 1985 { 1986 1987 if (callout_drain(&(timer)->callout) == -1) 1988 return (0); 1989 return (1); 1990 } 1991 1992 int 1993 timer_delete_sync(struct timer_list *timer) 1994 { 1995 1996 return (del_timer_sync(timer)); 1997 } 1998 1999 int 2000 timer_shutdown_sync(struct timer_list *timer) 2001 { 2002 2003 timer->function = NULL; 2004 return (del_timer_sync(timer)); 2005 } 2006 2007 /* greatest common divisor, Euclid equation */ 2008 static uint64_t 2009 lkpi_gcd_64(uint64_t a, uint64_t b) 2010 { 2011 uint64_t an; 2012 uint64_t bn; 2013 2014 while (b != 0) { 2015 an = b; 2016 bn = a % b; 2017 a = an; 2018 b = bn; 2019 } 2020 return (a); 2021 } 2022 2023 uint64_t lkpi_nsec2hz_rem; 2024 uint64_t lkpi_nsec2hz_div = 1000000000ULL; 2025 uint64_t lkpi_nsec2hz_max; 2026 2027 uint64_t lkpi_usec2hz_rem; 2028 uint64_t lkpi_usec2hz_div = 1000000ULL; 2029 uint64_t lkpi_usec2hz_max; 2030 2031 uint64_t lkpi_msec2hz_rem; 2032 uint64_t lkpi_msec2hz_div = 1000ULL; 2033 uint64_t lkpi_msec2hz_max; 2034 2035 static void 2036 linux_timer_init(void *arg) 2037 { 2038 uint64_t gcd; 2039 2040 /* 2041 * Compute an internal HZ value which can divide 2**32 to 2042 * avoid timer rounding problems when the tick value wraps 2043 * around 2**32: 2044 */ 2045 linux_timer_hz_mask = 1; 2046 while (linux_timer_hz_mask < (unsigned long)hz) 2047 linux_timer_hz_mask *= 2; 2048 linux_timer_hz_mask--; 2049 2050 /* compute some internal constants */ 2051 2052 lkpi_nsec2hz_rem = hz; 2053 lkpi_usec2hz_rem = hz; 2054 lkpi_msec2hz_rem = hz; 2055 2056 gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); 2057 lkpi_nsec2hz_rem /= gcd; 2058 lkpi_nsec2hz_div /= gcd; 2059 lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; 2060 2061 gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); 2062 lkpi_usec2hz_rem /= gcd; 2063 lkpi_usec2hz_div /= gcd; 2064 lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; 2065 2066 gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); 2067 lkpi_msec2hz_rem /= gcd; 2068 lkpi_msec2hz_div /= gcd; 2069 lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; 2070 } 2071 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); 2072 2073 void 2074 linux_complete_common(struct completion *c, int all) 2075 { 2076 int wakeup_swapper; 2077 2078 sleepq_lock(c); 2079 if (all) { 2080 c->done = UINT_MAX; 2081 wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); 2082 } else { 2083 if (c->done != UINT_MAX) 2084 c->done++; 2085 wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); 2086 } 2087 sleepq_release(c); 2088 if (wakeup_swapper) 2089 kick_proc0(); 2090 } 2091 2092 /* 2093 * Indefinite wait for done != 0 with or without signals. 2094 */ 2095 int 2096 linux_wait_for_common(struct completion *c, int flags) 2097 { 2098 struct task_struct *task; 2099 int error; 2100 2101 if (SCHEDULER_STOPPED()) 2102 return (0); 2103 2104 task = current; 2105 2106 if (flags != 0) 2107 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2108 else 2109 flags = SLEEPQ_SLEEP; 2110 error = 0; 2111 for (;;) { 2112 sleepq_lock(c); 2113 if (c->done) 2114 break; 2115 sleepq_add(c, NULL, "completion", flags, 0); 2116 if (flags & SLEEPQ_INTERRUPTIBLE) { 2117 DROP_GIANT(); 2118 error = -sleepq_wait_sig(c, 0); 2119 PICKUP_GIANT(); 2120 if (error != 0) { 2121 linux_schedule_save_interrupt_value(task, error); 2122 error = -ERESTARTSYS; 2123 goto intr; 2124 } 2125 } else { 2126 DROP_GIANT(); 2127 sleepq_wait(c, 0); 2128 PICKUP_GIANT(); 2129 } 2130 } 2131 if (c->done != UINT_MAX) 2132 c->done--; 2133 sleepq_release(c); 2134 2135 intr: 2136 return (error); 2137 } 2138 2139 /* 2140 * Time limited wait for done != 0 with or without signals. 2141 */ 2142 int 2143 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags) 2144 { 2145 struct task_struct *task; 2146 int end = jiffies + timeout; 2147 int error; 2148 2149 if (SCHEDULER_STOPPED()) 2150 return (0); 2151 2152 task = current; 2153 2154 if (flags != 0) 2155 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2156 else 2157 flags = SLEEPQ_SLEEP; 2158 2159 for (;;) { 2160 sleepq_lock(c); 2161 if (c->done) 2162 break; 2163 sleepq_add(c, NULL, "completion", flags, 0); 2164 sleepq_set_timeout(c, linux_timer_jiffies_until(end)); 2165 2166 DROP_GIANT(); 2167 if (flags & SLEEPQ_INTERRUPTIBLE) 2168 error = -sleepq_timedwait_sig(c, 0); 2169 else 2170 error = -sleepq_timedwait(c, 0); 2171 PICKUP_GIANT(); 2172 2173 if (error != 0) { 2174 /* check for timeout */ 2175 if (error == -EWOULDBLOCK) { 2176 error = 0; /* timeout */ 2177 } else { 2178 /* signal happened */ 2179 linux_schedule_save_interrupt_value(task, error); 2180 error = -ERESTARTSYS; 2181 } 2182 goto done; 2183 } 2184 } 2185 if (c->done != UINT_MAX) 2186 c->done--; 2187 sleepq_release(c); 2188 2189 /* return how many jiffies are left */ 2190 error = linux_timer_jiffies_until(end); 2191 done: 2192 return (error); 2193 } 2194 2195 int 2196 linux_try_wait_for_completion(struct completion *c) 2197 { 2198 int isdone; 2199 2200 sleepq_lock(c); 2201 isdone = (c->done != 0); 2202 if (c->done != 0 && c->done != UINT_MAX) 2203 c->done--; 2204 sleepq_release(c); 2205 return (isdone); 2206 } 2207 2208 int 2209 linux_completion_done(struct completion *c) 2210 { 2211 int isdone; 2212 2213 sleepq_lock(c); 2214 isdone = (c->done != 0); 2215 sleepq_release(c); 2216 return (isdone); 2217 } 2218 2219 static void 2220 linux_cdev_deref(struct linux_cdev *ldev) 2221 { 2222 if (refcount_release(&ldev->refs) && 2223 ldev->kobj.ktype == &linux_cdev_ktype) 2224 kfree(ldev); 2225 } 2226 2227 static void 2228 linux_cdev_release(struct kobject *kobj) 2229 { 2230 struct linux_cdev *cdev; 2231 struct kobject *parent; 2232 2233 cdev = container_of(kobj, struct linux_cdev, kobj); 2234 parent = kobj->parent; 2235 linux_destroy_dev(cdev); 2236 linux_cdev_deref(cdev); 2237 kobject_put(parent); 2238 } 2239 2240 static void 2241 linux_cdev_static_release(struct kobject *kobj) 2242 { 2243 struct cdev *cdev; 2244 struct linux_cdev *ldev; 2245 2246 ldev = container_of(kobj, struct linux_cdev, kobj); 2247 cdev = ldev->cdev; 2248 if (cdev != NULL) { 2249 destroy_dev(cdev); 2250 ldev->cdev = NULL; 2251 } 2252 kobject_put(kobj->parent); 2253 } 2254 2255 int 2256 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev) 2257 { 2258 int ret; 2259 2260 if (dev->devt != 0) { 2261 /* Set parent kernel object. */ 2262 ldev->kobj.parent = &dev->kobj; 2263 2264 /* 2265 * Unlike Linux we require the kobject of the 2266 * character device structure to have a valid name 2267 * before calling this function: 2268 */ 2269 if (ldev->kobj.name == NULL) 2270 return (-EINVAL); 2271 2272 ret = cdev_add(ldev, dev->devt, 1); 2273 if (ret) 2274 return (ret); 2275 } 2276 ret = device_add(dev); 2277 if (ret != 0 && dev->devt != 0) 2278 cdev_del(ldev); 2279 return (ret); 2280 } 2281 2282 void 2283 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev) 2284 { 2285 device_del(dev); 2286 2287 if (dev->devt != 0) 2288 cdev_del(ldev); 2289 } 2290 2291 static void 2292 linux_destroy_dev(struct linux_cdev *ldev) 2293 { 2294 2295 if (ldev->cdev == NULL) 2296 return; 2297 2298 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 2299 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 2300 2301 atomic_set_int(&ldev->siref, LDEV_SI_DTR); 2302 while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) 2303 pause("ldevdtr", hz / 4); 2304 2305 destroy_dev(ldev->cdev); 2306 ldev->cdev = NULL; 2307 } 2308 2309 const struct kobj_type linux_cdev_ktype = { 2310 .release = linux_cdev_release, 2311 }; 2312 2313 const struct kobj_type linux_cdev_static_ktype = { 2314 .release = linux_cdev_static_release, 2315 }; 2316 2317 static void 2318 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) 2319 { 2320 struct notifier_block *nb; 2321 struct netdev_notifier_info ni; 2322 2323 nb = arg; 2324 ni.ifp = ifp; 2325 ni.dev = (struct net_device *)ifp; 2326 if (linkstate == LINK_STATE_UP) 2327 nb->notifier_call(nb, NETDEV_UP, &ni); 2328 else 2329 nb->notifier_call(nb, NETDEV_DOWN, &ni); 2330 } 2331 2332 static void 2333 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) 2334 { 2335 struct notifier_block *nb; 2336 struct netdev_notifier_info ni; 2337 2338 nb = arg; 2339 ni.ifp = ifp; 2340 ni.dev = (struct net_device *)ifp; 2341 nb->notifier_call(nb, NETDEV_REGISTER, &ni); 2342 } 2343 2344 static void 2345 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) 2346 { 2347 struct notifier_block *nb; 2348 struct netdev_notifier_info ni; 2349 2350 nb = arg; 2351 ni.ifp = ifp; 2352 ni.dev = (struct net_device *)ifp; 2353 nb->notifier_call(nb, NETDEV_UNREGISTER, &ni); 2354 } 2355 2356 static void 2357 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) 2358 { 2359 struct notifier_block *nb; 2360 struct netdev_notifier_info ni; 2361 2362 nb = arg; 2363 ni.ifp = ifp; 2364 ni.dev = (struct net_device *)ifp; 2365 nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni); 2366 } 2367 2368 static void 2369 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) 2370 { 2371 struct notifier_block *nb; 2372 struct netdev_notifier_info ni; 2373 2374 nb = arg; 2375 ni.ifp = ifp; 2376 ni.dev = (struct net_device *)ifp; 2377 nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni); 2378 } 2379 2380 int 2381 register_netdevice_notifier(struct notifier_block *nb) 2382 { 2383 2384 nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( 2385 ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); 2386 nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( 2387 ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); 2388 nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( 2389 ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); 2390 nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( 2391 iflladdr_event, linux_handle_iflladdr_event, nb, 0); 2392 2393 return (0); 2394 } 2395 2396 int 2397 register_inetaddr_notifier(struct notifier_block *nb) 2398 { 2399 2400 nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( 2401 ifaddr_event, linux_handle_ifaddr_event, nb, 0); 2402 return (0); 2403 } 2404 2405 int 2406 unregister_netdevice_notifier(struct notifier_block *nb) 2407 { 2408 2409 EVENTHANDLER_DEREGISTER(ifnet_link_event, 2410 nb->tags[NETDEV_UP]); 2411 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 2412 nb->tags[NETDEV_REGISTER]); 2413 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 2414 nb->tags[NETDEV_UNREGISTER]); 2415 EVENTHANDLER_DEREGISTER(iflladdr_event, 2416 nb->tags[NETDEV_CHANGEADDR]); 2417 2418 return (0); 2419 } 2420 2421 int 2422 unregister_inetaddr_notifier(struct notifier_block *nb) 2423 { 2424 2425 EVENTHANDLER_DEREGISTER(ifaddr_event, 2426 nb->tags[NETDEV_CHANGEIFADDR]); 2427 2428 return (0); 2429 } 2430 2431 struct list_sort_thunk { 2432 int (*cmp)(void *, struct list_head *, struct list_head *); 2433 void *priv; 2434 }; 2435 2436 static inline int 2437 linux_le_cmp(const void *d1, const void *d2, void *priv) 2438 { 2439 struct list_head *le1, *le2; 2440 struct list_sort_thunk *thunk; 2441 2442 thunk = priv; 2443 le1 = *(__DECONST(struct list_head **, d1)); 2444 le2 = *(__DECONST(struct list_head **, d2)); 2445 return ((thunk->cmp)(thunk->priv, le1, le2)); 2446 } 2447 2448 void 2449 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, 2450 struct list_head *a, struct list_head *b)) 2451 { 2452 struct list_sort_thunk thunk; 2453 struct list_head **ar, *le; 2454 size_t count, i; 2455 2456 count = 0; 2457 list_for_each(le, head) 2458 count++; 2459 ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); 2460 i = 0; 2461 list_for_each(le, head) 2462 ar[i++] = le; 2463 thunk.cmp = cmp; 2464 thunk.priv = priv; 2465 qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk); 2466 INIT_LIST_HEAD(head); 2467 for (i = 0; i < count; i++) 2468 list_add_tail(ar[i], head); 2469 free(ar, M_KMALLOC); 2470 } 2471 2472 #if defined(__i386__) || defined(__amd64__) 2473 int 2474 linux_wbinvd_on_all_cpus(void) 2475 { 2476 2477 pmap_invalidate_cache(); 2478 return (0); 2479 } 2480 #endif 2481 2482 int 2483 linux_on_each_cpu(void callback(void *), void *data) 2484 { 2485 2486 smp_rendezvous(smp_no_rendezvous_barrier, callback, 2487 smp_no_rendezvous_barrier, data); 2488 return (0); 2489 } 2490 2491 int 2492 linux_in_atomic(void) 2493 { 2494 2495 return ((curthread->td_pflags & TDP_NOFAULTING) != 0); 2496 } 2497 2498 struct linux_cdev * 2499 linux_find_cdev(const char *name, unsigned major, unsigned minor) 2500 { 2501 dev_t dev = MKDEV(major, minor); 2502 struct cdev *cdev; 2503 2504 dev_lock(); 2505 LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { 2506 struct linux_cdev *ldev = cdev->si_drv1; 2507 if (ldev->dev == dev && 2508 strcmp(kobject_name(&ldev->kobj), name) == 0) { 2509 break; 2510 } 2511 } 2512 dev_unlock(); 2513 2514 return (cdev != NULL ? cdev->si_drv1 : NULL); 2515 } 2516 2517 int 2518 __register_chrdev(unsigned int major, unsigned int baseminor, 2519 unsigned int count, const char *name, 2520 const struct file_operations *fops) 2521 { 2522 struct linux_cdev *cdev; 2523 int ret = 0; 2524 int i; 2525 2526 for (i = baseminor; i < baseminor + count; i++) { 2527 cdev = cdev_alloc(); 2528 cdev->ops = fops; 2529 kobject_set_name(&cdev->kobj, name); 2530 2531 ret = cdev_add(cdev, makedev(major, i), 1); 2532 if (ret != 0) 2533 break; 2534 } 2535 return (ret); 2536 } 2537 2538 int 2539 __register_chrdev_p(unsigned int major, unsigned int baseminor, 2540 unsigned int count, const char *name, 2541 const struct file_operations *fops, uid_t uid, 2542 gid_t gid, int mode) 2543 { 2544 struct linux_cdev *cdev; 2545 int ret = 0; 2546 int i; 2547 2548 for (i = baseminor; i < baseminor + count; i++) { 2549 cdev = cdev_alloc(); 2550 cdev->ops = fops; 2551 kobject_set_name(&cdev->kobj, name); 2552 2553 ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); 2554 if (ret != 0) 2555 break; 2556 } 2557 return (ret); 2558 } 2559 2560 void 2561 __unregister_chrdev(unsigned int major, unsigned int baseminor, 2562 unsigned int count, const char *name) 2563 { 2564 struct linux_cdev *cdevp; 2565 int i; 2566 2567 for (i = baseminor; i < baseminor + count; i++) { 2568 cdevp = linux_find_cdev(name, major, i); 2569 if (cdevp != NULL) 2570 cdev_del(cdevp); 2571 } 2572 } 2573 2574 void 2575 linux_dump_stack(void) 2576 { 2577 #ifdef STACK 2578 struct stack st; 2579 2580 stack_save(&st); 2581 stack_print(&st); 2582 #endif 2583 } 2584 2585 int 2586 linuxkpi_net_ratelimit(void) 2587 { 2588 2589 return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps, 2590 lkpi_net_maxpps)); 2591 } 2592 2593 struct io_mapping * 2594 io_mapping_create_wc(resource_size_t base, unsigned long size) 2595 { 2596 struct io_mapping *mapping; 2597 2598 mapping = kmalloc(sizeof(*mapping), GFP_KERNEL); 2599 if (mapping == NULL) 2600 return (NULL); 2601 return (io_mapping_init_wc(mapping, base, size)); 2602 } 2603 2604 /* We likely want a linuxkpi_device.c at some point. */ 2605 bool 2606 device_can_wakeup(struct device *dev) 2607 { 2608 2609 if (dev == NULL) 2610 return (false); 2611 /* 2612 * XXX-BZ iwlwifi queries it as part of enabling WoWLAN. 2613 * Normally this would be based on a bool in dev->power.XXX. 2614 * Check such as PCI PCIM_PCAP_*PME. We have no way to enable this yet. 2615 * We may get away by directly calling into bsddev for as long as 2616 * we can assume PCI only avoiding changing struct device breaking KBI. 2617 */ 2618 pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__); 2619 return (false); 2620 } 2621 2622 static void 2623 devm_device_group_remove(struct device *dev, void *p) 2624 { 2625 const struct attribute_group **dr = p; 2626 const struct attribute_group *group = *dr; 2627 2628 sysfs_remove_group(&dev->kobj, group); 2629 } 2630 2631 int 2632 lkpi_devm_device_add_group(struct device *dev, 2633 const struct attribute_group *group) 2634 { 2635 const struct attribute_group **dr; 2636 int ret; 2637 2638 dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL); 2639 if (dr == NULL) 2640 return (-ENOMEM); 2641 2642 ret = sysfs_create_group(&dev->kobj, group); 2643 if (ret == 0) { 2644 *dr = group; 2645 devres_add(dev, dr); 2646 } else 2647 devres_free(dr); 2648 2649 return (ret); 2650 } 2651 2652 #if defined(__i386__) || defined(__amd64__) 2653 bool linux_cpu_has_clflush; 2654 struct cpuinfo_x86 boot_cpu_data; 2655 struct cpuinfo_x86 *__cpu_data; 2656 #endif 2657 2658 cpumask_t * 2659 lkpi_get_static_single_cpu_mask(int cpuid) 2660 { 2661 2662 KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n", 2663 __func__, cpuid)); 2664 KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n", 2665 __func__, cpuid)); 2666 2667 return (static_single_cpu_mask[cpuid]); 2668 } 2669 2670 bool 2671 lkpi_xen_initial_domain(void) 2672 { 2673 #ifdef XENHVM 2674 return (xen_initial_domain()); 2675 #else 2676 return (false); 2677 #endif 2678 } 2679 2680 bool 2681 lkpi_xen_pv_domain(void) 2682 { 2683 #ifdef XENHVM 2684 return (xen_pv_domain()); 2685 #else 2686 return (false); 2687 #endif 2688 } 2689 2690 static void 2691 linux_compat_init(void *arg) 2692 { 2693 struct sysctl_oid *rootoid; 2694 int i; 2695 2696 #if defined(__i386__) || defined(__amd64__) 2697 static const uint32_t x86_vendors[X86_VENDOR_NUM] = { 2698 [X86_VENDOR_INTEL] = CPU_VENDOR_INTEL, 2699 [X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX, 2700 [X86_VENDOR_AMD] = CPU_VENDOR_AMD, 2701 [X86_VENDOR_UMC] = CPU_VENDOR_UMC, 2702 [X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR, 2703 [X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA, 2704 [X86_VENDOR_NSC] = CPU_VENDOR_NSC, 2705 [X86_VENDOR_HYGON] = CPU_VENDOR_HYGON, 2706 }; 2707 uint8_t x86_vendor = X86_VENDOR_UNKNOWN; 2708 2709 for (i = 0; i < X86_VENDOR_NUM; i++) { 2710 if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) { 2711 x86_vendor = i; 2712 break; 2713 } 2714 } 2715 linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); 2716 boot_cpu_data.x86_clflush_size = cpu_clflush_line_size; 2717 boot_cpu_data.x86_max_cores = mp_ncpus; 2718 boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id); 2719 boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id); 2720 boot_cpu_data.x86_vendor = x86_vendor; 2721 2722 __cpu_data = mallocarray(mp_maxid + 1, 2723 sizeof(*__cpu_data), M_KMALLOC, M_WAITOK | M_ZERO); 2724 CPU_FOREACH(i) { 2725 __cpu_data[i].x86_clflush_size = cpu_clflush_line_size; 2726 __cpu_data[i].x86_max_cores = mp_ncpus; 2727 __cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id); 2728 __cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id); 2729 __cpu_data[i].x86_vendor = x86_vendor; 2730 } 2731 #endif 2732 rw_init(&linux_vma_lock, "lkpi-vma-lock"); 2733 2734 rootoid = SYSCTL_ADD_ROOT_NODE(NULL, 2735 OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); 2736 kobject_init(&linux_class_root, &linux_class_ktype); 2737 kobject_set_name(&linux_class_root, "class"); 2738 linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), 2739 OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); 2740 kobject_init(&linux_root_device.kobj, &linux_dev_ktype); 2741 kobject_set_name(&linux_root_device.kobj, "device"); 2742 linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, 2743 SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", 2744 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device"); 2745 linux_root_device.bsddev = root_bus; 2746 linux_class_misc.name = "misc"; 2747 class_register(&linux_class_misc); 2748 INIT_LIST_HEAD(&pci_drivers); 2749 INIT_LIST_HEAD(&pci_devices); 2750 spin_lock_init(&pci_lock); 2751 mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); 2752 for (i = 0; i < VMMAP_HASH_SIZE; i++) 2753 LIST_INIT(&vmmaphead[i]); 2754 init_waitqueue_head(&linux_bit_waitq); 2755 init_waitqueue_head(&linux_var_waitq); 2756 2757 CPU_COPY(&all_cpus, &cpu_online_mask); 2758 /* 2759 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system. 2760 * CPUs are indexed from 0..(mp_maxid). The entry for cpuid 0 will only 2761 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on. 2762 * This is used by cpumask_of() (and possibly others in the future) for, 2763 * e.g., drivers to pass hints to irq_set_affinity_hint(). 2764 */ 2765 static_single_cpu_mask = mallocarray(mp_maxid + 1, 2766 sizeof(static_single_cpu_mask), M_KMALLOC, M_WAITOK | M_ZERO); 2767 2768 /* 2769 * When the number of CPUs reach a threshold, we start to save memory 2770 * given the sets are static by overlapping those having their single 2771 * bit set at same position in a bitset word. Asymptotically, this 2772 * regular scheme is in O(n²) whereas the overlapping one is in O(n) 2773 * only with n being the maximum number of CPUs, so the gain will become 2774 * huge quite quickly. The threshold for 64-bit architectures is 128 2775 * CPUs. 2776 */ 2777 if (mp_ncpus < (2 * _BITSET_BITS)) { 2778 cpumask_t *sscm_ptr; 2779 2780 /* 2781 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) * 2782 * (_BITSET_BITS / 8)' bytes (for comparison with the 2783 * overlapping scheme). 2784 */ 2785 static_single_cpu_mask_lcs = mallocarray(mp_ncpus, 2786 sizeof(*static_single_cpu_mask_lcs), 2787 M_KMALLOC, M_WAITOK | M_ZERO); 2788 2789 sscm_ptr = static_single_cpu_mask_lcs; 2790 CPU_FOREACH(i) { 2791 static_single_cpu_mask[i] = sscm_ptr++; 2792 CPU_SET(i, static_single_cpu_mask[i]); 2793 } 2794 } else { 2795 /* Pointer to a bitset word. */ 2796 __typeof(((cpuset_t *)NULL)->__bits[0]) *bwp; 2797 2798 /* 2799 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t' 2800 * really) with a single bit set that can be reused for all 2801 * single CPU masks by making them start at different offsets. 2802 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before 2803 * the word having its single bit set, and the same amount 2804 * after. 2805 */ 2806 static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS, 2807 (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8), 2808 M_KMALLOC, M_WAITOK | M_ZERO); 2809 2810 /* 2811 * We rely below on cpuset_t and the bitset generic 2812 * implementation assigning words in the '__bits' array in the 2813 * same order of bits (i.e., little-endian ordering, not to be 2814 * confused with machine endianness, which concerns bits in 2815 * words and other integers). This is an imperfect test, but it 2816 * will detect a change to big-endian ordering. 2817 */ 2818 _Static_assert( 2819 __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1, 2820 "Assumes a bitset implementation that is little-endian " 2821 "on its words"); 2822 2823 /* Initialize the single bit of each static span. */ 2824 bwp = (__typeof(bwp))static_single_cpu_mask_lcs + 2825 (__bitset_words(CPU_SETSIZE) - 1); 2826 for (i = 0; i < _BITSET_BITS; i++) { 2827 CPU_SET(i, (cpuset_t *)bwp); 2828 bwp += (2 * __bitset_words(CPU_SETSIZE) - 1); 2829 } 2830 2831 /* 2832 * Finally set all CPU masks to the proper word in their 2833 * relevant span. 2834 */ 2835 CPU_FOREACH(i) { 2836 bwp = (__typeof(bwp))static_single_cpu_mask_lcs; 2837 /* Find the non-zero word of the relevant span. */ 2838 bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) * 2839 (i % _BITSET_BITS) + 2840 __bitset_words(CPU_SETSIZE) - 1; 2841 /* Shift to find the CPU mask start. */ 2842 bwp -= (i / _BITSET_BITS); 2843 static_single_cpu_mask[i] = (cpuset_t *)bwp; 2844 } 2845 } 2846 2847 strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release)); 2848 } 2849 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); 2850 2851 static void 2852 linux_compat_uninit(void *arg) 2853 { 2854 linux_kobject_kfree_name(&linux_class_root); 2855 linux_kobject_kfree_name(&linux_root_device.kobj); 2856 linux_kobject_kfree_name(&linux_class_misc.kobj); 2857 2858 free(static_single_cpu_mask_lcs, M_KMALLOC); 2859 free(static_single_cpu_mask, M_KMALLOC); 2860 #if defined(__i386__) || defined(__amd64__) 2861 free(__cpu_data, M_KMALLOC); 2862 #endif 2863 2864 mtx_destroy(&vmmaplock); 2865 spin_lock_destroy(&pci_lock); 2866 rw_destroy(&linux_vma_lock); 2867 } 2868 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); 2869 2870 /* 2871 * NOTE: Linux frequently uses "unsigned long" for pointer to integer 2872 * conversion and vice versa, where in FreeBSD "uintptr_t" would be 2873 * used. Assert these types have the same size, else some parts of the 2874 * LinuxKPI may not work like expected: 2875 */ 2876 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); 2877