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/stdarg.h> 54 #include <sys/sysent.h> 55 #include <sys/time.h> 56 #include <sys/user.h> 57 58 #include <vm/vm.h> 59 #include <vm/pmap.h> 60 #include <vm/vm_object.h> 61 #include <vm/vm_page.h> 62 #include <vm/vm_pager.h> 63 #include <vm/vm_radix.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/fs.h> 79 #include <linux/sysfs.h> 80 #include <linux/mm.h> 81 #include <linux/io.h> 82 #include <linux/vmalloc.h> 83 #include <linux/netdevice.h> 84 #include <linux/timer.h> 85 #include <linux/interrupt.h> 86 #include <linux/uaccess.h> 87 #include <linux/utsname.h> 88 #include <linux/list.h> 89 #include <linux/kthread.h> 90 #include <linux/kernel.h> 91 #include <linux/compat.h> 92 #include <linux/io-mapping.h> 93 #include <linux/poll.h> 94 #include <linux/smp.h> 95 #include <linux/wait_bit.h> 96 #include <linux/rcupdate.h> 97 #include <linux/interval_tree.h> 98 #include <linux/interval_tree_generic.h> 99 #include <linux/printk.h> 100 #include <linux/seq_file.h> 101 102 #if defined(__i386__) || defined(__amd64__) 103 #include <asm/smp.h> 104 #include <asm/processor.h> 105 #endif 106 107 #include <xen/xen.h> 108 #ifdef XENHVM 109 #undef xen_pv_domain 110 #undef xen_initial_domain 111 /* xen/xen-os.h redefines __must_check */ 112 #undef __must_check 113 #include <xen/xen-os.h> 114 #endif 115 116 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 117 "LinuxKPI parameters"); 118 119 int linuxkpi_debug; 120 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, 121 &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); 122 123 int linuxkpi_rcu_debug; 124 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, rcu_debug, CTLFLAG_RWTUN, 125 &linuxkpi_rcu_debug, 0, "Set to enable RCU warning. Clear to disable."); 126 127 int linuxkpi_warn_dump_stack = 0; 128 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN, 129 &linuxkpi_warn_dump_stack, 0, 130 "Set to enable stack traces from WARN_ON(). Clear to disable."); 131 132 static struct timeval lkpi_net_lastlog; 133 static int lkpi_net_curpps; 134 static int lkpi_net_maxpps = 99; 135 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN, 136 &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second."); 137 138 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat"); 139 140 #include <linux/rbtree.h> 141 /* Undo Linux compat changes. */ 142 #undef RB_ROOT 143 #undef file 144 #undef cdev 145 #define RB_ROOT(head) (head)->rbh_root 146 147 static void linux_destroy_dev(struct linux_cdev *); 148 static void linux_cdev_deref(struct linux_cdev *ldev); 149 static struct vm_area_struct *linux_cdev_handle_find(void *handle); 150 151 cpumask_t cpu_online_mask; 152 static cpumask_t **static_single_cpu_mask; 153 static cpumask_t *static_single_cpu_mask_lcs; 154 struct kobject linux_class_root; 155 struct device linux_root_device; 156 struct class linux_class_misc; 157 struct list_head pci_drivers; 158 struct list_head pci_devices; 159 spinlock_t pci_lock; 160 struct uts_namespace init_uts_ns; 161 162 unsigned long linux_timer_hz_mask; 163 164 wait_queue_head_t linux_bit_waitq; 165 wait_queue_head_t linux_var_waitq; 166 167 int 168 panic_cmp(struct rb_node *one, struct rb_node *two) 169 { 170 panic("no cmp"); 171 } 172 173 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); 174 175 #define START(node) ((node)->start) 176 #define LAST(node) ((node)->last) 177 178 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START, 179 LAST,, lkpi_interval_tree) 180 181 static void 182 linux_device_release(struct device *dev) 183 { 184 pr_debug("linux_device_release: %s\n", dev_name(dev)); 185 kfree(dev); 186 } 187 188 static ssize_t 189 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) 190 { 191 struct class_attribute *dattr; 192 ssize_t error; 193 194 dattr = container_of(attr, struct class_attribute, attr); 195 error = -EIO; 196 if (dattr->show) 197 error = dattr->show(container_of(kobj, struct class, kobj), 198 dattr, buf); 199 return (error); 200 } 201 202 static ssize_t 203 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, 204 size_t count) 205 { 206 struct class_attribute *dattr; 207 ssize_t error; 208 209 dattr = container_of(attr, struct class_attribute, attr); 210 error = -EIO; 211 if (dattr->store) 212 error = dattr->store(container_of(kobj, struct class, kobj), 213 dattr, buf, count); 214 return (error); 215 } 216 217 static void 218 linux_class_release(struct kobject *kobj) 219 { 220 struct class *class; 221 222 class = container_of(kobj, struct class, kobj); 223 if (class->class_release) 224 class->class_release(class); 225 } 226 227 static const struct sysfs_ops linux_class_sysfs = { 228 .show = linux_class_show, 229 .store = linux_class_store, 230 }; 231 232 const struct kobj_type linux_class_ktype = { 233 .release = linux_class_release, 234 .sysfs_ops = &linux_class_sysfs 235 }; 236 237 static void 238 linux_dev_release(struct kobject *kobj) 239 { 240 struct device *dev; 241 242 dev = container_of(kobj, struct device, kobj); 243 /* This is the precedence defined by linux. */ 244 if (dev->release) 245 dev->release(dev); 246 else if (dev->class && dev->class->dev_release) 247 dev->class->dev_release(dev); 248 } 249 250 static ssize_t 251 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) 252 { 253 struct device_attribute *dattr; 254 ssize_t error; 255 256 dattr = container_of(attr, struct device_attribute, attr); 257 error = -EIO; 258 if (dattr->show) 259 error = dattr->show(container_of(kobj, struct device, kobj), 260 dattr, buf); 261 return (error); 262 } 263 264 static ssize_t 265 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, 266 size_t count) 267 { 268 struct device_attribute *dattr; 269 ssize_t error; 270 271 dattr = container_of(attr, struct device_attribute, attr); 272 error = -EIO; 273 if (dattr->store) 274 error = dattr->store(container_of(kobj, struct device, kobj), 275 dattr, buf, count); 276 return (error); 277 } 278 279 static const struct sysfs_ops linux_dev_sysfs = { 280 .show = linux_dev_show, 281 .store = linux_dev_store, 282 }; 283 284 const struct kobj_type linux_dev_ktype = { 285 .release = linux_dev_release, 286 .sysfs_ops = &linux_dev_sysfs 287 }; 288 289 struct device * 290 device_create(struct class *class, struct device *parent, dev_t devt, 291 void *drvdata, const char *fmt, ...) 292 { 293 struct device *dev; 294 va_list args; 295 296 dev = kzalloc(sizeof(*dev), M_WAITOK); 297 dev->parent = parent; 298 dev->class = class; 299 dev->devt = devt; 300 dev->driver_data = drvdata; 301 dev->release = linux_device_release; 302 va_start(args, fmt); 303 kobject_set_name_vargs(&dev->kobj, fmt, args); 304 va_end(args); 305 device_register(dev); 306 307 return (dev); 308 } 309 310 struct device * 311 device_create_groups_vargs(struct class *class, struct device *parent, 312 dev_t devt, void *drvdata, const struct attribute_group **groups, 313 const char *fmt, va_list args) 314 { 315 struct device *dev = NULL; 316 int retval = -ENODEV; 317 318 if (class == NULL || IS_ERR(class)) 319 goto error; 320 321 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 322 if (!dev) { 323 retval = -ENOMEM; 324 goto error; 325 } 326 327 dev->devt = devt; 328 dev->class = class; 329 dev->parent = parent; 330 dev->groups = groups; 331 dev->release = device_create_release; 332 /* device_initialize() needs the class and parent to be set */ 333 device_initialize(dev); 334 dev_set_drvdata(dev, drvdata); 335 336 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 337 if (retval) 338 goto error; 339 340 retval = device_add(dev); 341 if (retval) 342 goto error; 343 344 return dev; 345 346 error: 347 put_device(dev); 348 return ERR_PTR(retval); 349 } 350 351 struct class * 352 lkpi_class_create(const char *name) 353 { 354 struct class *class; 355 int error; 356 357 class = kzalloc(sizeof(*class), M_WAITOK); 358 class->name = name; 359 class->class_release = linux_class_kfree; 360 error = class_register(class); 361 if (error) { 362 kfree(class); 363 return (NULL); 364 } 365 366 return (class); 367 } 368 369 static void 370 linux_kq_lock(void *arg) 371 { 372 spinlock_t *s = arg; 373 374 spin_lock(s); 375 } 376 static void 377 linux_kq_unlock(void *arg) 378 { 379 spinlock_t *s = arg; 380 381 spin_unlock(s); 382 } 383 384 static void 385 linux_kq_assert_lock(void *arg, int what) 386 { 387 #ifdef INVARIANTS 388 spinlock_t *s = arg; 389 390 if (what == LA_LOCKED) 391 mtx_assert(s, MA_OWNED); 392 else 393 mtx_assert(s, MA_NOTOWNED); 394 #endif 395 } 396 397 static void 398 linux_file_kqfilter_poll(struct linux_file *, int); 399 400 struct linux_file * 401 linux_file_alloc(void) 402 { 403 struct linux_file *filp; 404 405 filp = kzalloc(sizeof(*filp), GFP_KERNEL); 406 407 /* set initial refcount */ 408 filp->f_count = 1; 409 410 /* setup fields needed by kqueue support */ 411 spin_lock_init(&filp->f_kqlock); 412 knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, 413 linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock); 414 415 return (filp); 416 } 417 418 void 419 linux_file_free(struct linux_file *filp) 420 { 421 if (filp->_file == NULL) { 422 if (filp->f_op != NULL && filp->f_op->release != NULL) 423 filp->f_op->release(filp->f_vnode, filp); 424 if (filp->f_shmem != NULL) 425 vm_object_deallocate(filp->f_shmem); 426 kfree_rcu(filp, rcu); 427 } else { 428 /* 429 * The close method of the character device or file 430 * will free the linux_file structure: 431 */ 432 _fdrop(filp->_file, curthread); 433 } 434 } 435 436 struct linux_cdev * 437 cdev_alloc(void) 438 { 439 struct linux_cdev *cdev; 440 441 cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK); 442 kobject_init(&cdev->kobj, &linux_cdev_ktype); 443 cdev->refs = 1; 444 return (cdev); 445 } 446 447 static int 448 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, 449 vm_page_t *mres) 450 { 451 struct vm_area_struct *vmap; 452 453 vmap = linux_cdev_handle_find(vm_obj->handle); 454 455 MPASS(vmap != NULL); 456 MPASS(vmap->vm_private_data == vm_obj->handle); 457 458 if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { 459 vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; 460 vm_page_t page; 461 462 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 463 /* 464 * If the passed in result page is a fake 465 * page, update it with the new physical 466 * address. 467 */ 468 page = *mres; 469 vm_page_updatefake(page, paddr, vm_obj->memattr); 470 } else { 471 /* 472 * Replace the passed in "mres" page with our 473 * own fake page and free up the all of the 474 * original pages. 475 */ 476 VM_OBJECT_WUNLOCK(vm_obj); 477 page = vm_page_getfake(paddr, vm_obj->memattr); 478 VM_OBJECT_WLOCK(vm_obj); 479 480 vm_page_replace(page, vm_obj, (*mres)->pindex, *mres); 481 *mres = page; 482 } 483 vm_page_valid(page); 484 return (VM_PAGER_OK); 485 } 486 return (VM_PAGER_FAIL); 487 } 488 489 static int 490 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, 491 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 492 { 493 struct vm_area_struct *vmap; 494 int err; 495 496 /* get VM area structure */ 497 vmap = linux_cdev_handle_find(vm_obj->handle); 498 MPASS(vmap != NULL); 499 MPASS(vmap->vm_private_data == vm_obj->handle); 500 501 VM_OBJECT_WUNLOCK(vm_obj); 502 503 linux_set_current(curthread); 504 505 down_write(&vmap->vm_mm->mmap_sem); 506 if (unlikely(vmap->vm_ops == NULL)) { 507 err = VM_FAULT_SIGBUS; 508 } else { 509 struct vm_fault vmf; 510 511 /* fill out VM fault structure */ 512 vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); 513 vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; 514 vmf.pgoff = 0; 515 vmf.page = NULL; 516 vmf.vma = vmap; 517 518 vmap->vm_pfn_count = 0; 519 vmap->vm_pfn_pcount = &vmap->vm_pfn_count; 520 vmap->vm_obj = vm_obj; 521 522 err = vmap->vm_ops->fault(&vmf); 523 524 while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { 525 kern_yield(PRI_USER); 526 err = vmap->vm_ops->fault(&vmf); 527 } 528 } 529 530 /* translate return code */ 531 switch (err) { 532 case VM_FAULT_OOM: 533 err = VM_PAGER_AGAIN; 534 break; 535 case VM_FAULT_SIGBUS: 536 err = VM_PAGER_BAD; 537 break; 538 case VM_FAULT_NOPAGE: 539 /* 540 * By contract the fault handler will return having 541 * busied all the pages itself. If pidx is already 542 * found in the object, it will simply xbusy the first 543 * page and return with vm_pfn_count set to 1. 544 */ 545 *first = vmap->vm_pfn_first; 546 *last = *first + vmap->vm_pfn_count - 1; 547 err = VM_PAGER_OK; 548 break; 549 default: 550 err = VM_PAGER_ERROR; 551 break; 552 } 553 up_write(&vmap->vm_mm->mmap_sem); 554 VM_OBJECT_WLOCK(vm_obj); 555 return (err); 556 } 557 558 static struct rwlock linux_vma_lock; 559 static TAILQ_HEAD(, vm_area_struct) linux_vma_head = 560 TAILQ_HEAD_INITIALIZER(linux_vma_head); 561 562 static void 563 linux_cdev_handle_free(struct vm_area_struct *vmap) 564 { 565 /* Drop reference on vm_file */ 566 if (vmap->vm_file != NULL) 567 fput(vmap->vm_file); 568 569 /* Drop reference on mm_struct */ 570 mmput(vmap->vm_mm); 571 572 kfree(vmap); 573 } 574 575 static void 576 linux_cdev_handle_remove(struct vm_area_struct *vmap) 577 { 578 rw_wlock(&linux_vma_lock); 579 TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); 580 rw_wunlock(&linux_vma_lock); 581 } 582 583 static struct vm_area_struct * 584 linux_cdev_handle_find(void *handle) 585 { 586 struct vm_area_struct *vmap; 587 588 rw_rlock(&linux_vma_lock); 589 TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { 590 if (vmap->vm_private_data == handle) 591 break; 592 } 593 rw_runlock(&linux_vma_lock); 594 return (vmap); 595 } 596 597 static int 598 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 599 vm_ooffset_t foff, struct ucred *cred, u_short *color) 600 { 601 602 MPASS(linux_cdev_handle_find(handle) != NULL); 603 *color = 0; 604 return (0); 605 } 606 607 static void 608 linux_cdev_pager_dtor(void *handle) 609 { 610 const struct vm_operations_struct *vm_ops; 611 struct vm_area_struct *vmap; 612 613 vmap = linux_cdev_handle_find(handle); 614 MPASS(vmap != NULL); 615 616 /* 617 * Remove handle before calling close operation to prevent 618 * other threads from reusing the handle pointer. 619 */ 620 linux_cdev_handle_remove(vmap); 621 622 down_write(&vmap->vm_mm->mmap_sem); 623 vm_ops = vmap->vm_ops; 624 if (likely(vm_ops != NULL)) 625 vm_ops->close(vmap); 626 up_write(&vmap->vm_mm->mmap_sem); 627 628 linux_cdev_handle_free(vmap); 629 } 630 631 static struct cdev_pager_ops linux_cdev_pager_ops[2] = { 632 { 633 /* OBJT_MGTDEVICE */ 634 .cdev_pg_populate = linux_cdev_pager_populate, 635 .cdev_pg_ctor = linux_cdev_pager_ctor, 636 .cdev_pg_dtor = linux_cdev_pager_dtor 637 }, 638 { 639 /* OBJT_DEVICE */ 640 .cdev_pg_fault = linux_cdev_pager_fault, 641 .cdev_pg_ctor = linux_cdev_pager_ctor, 642 .cdev_pg_dtor = linux_cdev_pager_dtor 643 }, 644 }; 645 646 int 647 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 648 unsigned long size) 649 { 650 struct pctrie_iter pages; 651 vm_object_t obj; 652 vm_page_t m; 653 654 obj = vma->vm_obj; 655 if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) 656 return (-ENOTSUP); 657 VM_OBJECT_RLOCK(obj); 658 vm_page_iter_limit_init(&pages, obj, OFF_TO_IDX(address + size)); 659 VM_RADIX_FOREACH_FROM(m, &pages, OFF_TO_IDX(address)) 660 pmap_remove_all(m); 661 VM_OBJECT_RUNLOCK(obj); 662 return (0); 663 } 664 665 void 666 vma_set_file(struct vm_area_struct *vma, struct linux_file *file) 667 { 668 struct linux_file *tmp; 669 670 /* Changing an anonymous vma with this is illegal */ 671 get_file(file); 672 tmp = vma->vm_file; 673 vma->vm_file = file; 674 fput(tmp); 675 } 676 677 static struct file_operations dummy_ldev_ops = { 678 /* XXXKIB */ 679 }; 680 681 static struct linux_cdev dummy_ldev = { 682 .ops = &dummy_ldev_ops, 683 }; 684 685 #define LDEV_SI_DTR 0x0001 686 #define LDEV_SI_REF 0x0002 687 688 static void 689 linux_get_fop(struct linux_file *filp, const struct file_operations **fop, 690 struct linux_cdev **dev) 691 { 692 struct linux_cdev *ldev; 693 u_int siref; 694 695 ldev = filp->f_cdev; 696 *fop = filp->f_op; 697 if (ldev != NULL) { 698 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 699 refcount_acquire(&ldev->refs); 700 } else { 701 for (siref = ldev->siref;;) { 702 if ((siref & LDEV_SI_DTR) != 0) { 703 ldev = &dummy_ldev; 704 *fop = ldev->ops; 705 siref = ldev->siref; 706 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 707 } else if (atomic_fcmpset_int(&ldev->siref, 708 &siref, siref + LDEV_SI_REF)) { 709 break; 710 } 711 } 712 } 713 } 714 *dev = ldev; 715 } 716 717 static void 718 linux_drop_fop(struct linux_cdev *ldev) 719 { 720 721 if (ldev == NULL) 722 return; 723 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 724 linux_cdev_deref(ldev); 725 } else { 726 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 727 MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); 728 atomic_subtract_int(&ldev->siref, LDEV_SI_REF); 729 } 730 } 731 732 #define OPW(fp,td,code) ({ \ 733 struct file *__fpop; \ 734 __typeof(code) __retval; \ 735 \ 736 __fpop = (td)->td_fpop; \ 737 (td)->td_fpop = (fp); \ 738 __retval = (code); \ 739 (td)->td_fpop = __fpop; \ 740 __retval; \ 741 }) 742 743 static int 744 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, 745 struct file *file) 746 { 747 struct linux_cdev *ldev; 748 struct linux_file *filp; 749 const struct file_operations *fop; 750 int error; 751 752 ldev = dev->si_drv1; 753 754 filp = linux_file_alloc(); 755 filp->f_dentry = &filp->f_dentry_store; 756 filp->f_op = ldev->ops; 757 filp->f_mode = file->f_flag; 758 filp->f_flags = file->f_flag; 759 filp->f_vnode = file->f_vnode; 760 filp->_file = file; 761 refcount_acquire(&ldev->refs); 762 filp->f_cdev = ldev; 763 764 linux_set_current(td); 765 linux_get_fop(filp, &fop, &ldev); 766 767 if (fop->open != NULL) { 768 error = -fop->open(file->f_vnode, filp); 769 if (error != 0) { 770 linux_drop_fop(ldev); 771 linux_cdev_deref(filp->f_cdev); 772 kfree(filp); 773 return (error); 774 } 775 } 776 777 /* hold on to the vnode - used for fstat() */ 778 vref(filp->f_vnode); 779 780 /* release the file from devfs */ 781 finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); 782 linux_drop_fop(ldev); 783 return (ENXIO); 784 } 785 786 #define LINUX_IOCTL_MIN_PTR 0x10000UL 787 #define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) 788 789 static inline int 790 linux_remap_address(void **uaddr, size_t len) 791 { 792 uintptr_t uaddr_val = (uintptr_t)(*uaddr); 793 794 if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && 795 uaddr_val < LINUX_IOCTL_MAX_PTR)) { 796 struct task_struct *pts = current; 797 if (pts == NULL) { 798 *uaddr = NULL; 799 return (1); 800 } 801 802 /* compute data offset */ 803 uaddr_val -= LINUX_IOCTL_MIN_PTR; 804 805 /* check that length is within bounds */ 806 if ((len > IOCPARM_MAX) || 807 (uaddr_val + len) > pts->bsd_ioctl_len) { 808 *uaddr = NULL; 809 return (1); 810 } 811 812 /* re-add kernel buffer address */ 813 uaddr_val += (uintptr_t)pts->bsd_ioctl_data; 814 815 /* update address location */ 816 *uaddr = (void *)uaddr_val; 817 return (1); 818 } 819 return (0); 820 } 821 822 int 823 linux_copyin(const void *uaddr, void *kaddr, size_t len) 824 { 825 if (linux_remap_address(__DECONST(void **, &uaddr), len)) { 826 if (uaddr == NULL) 827 return (-EFAULT); 828 memcpy(kaddr, uaddr, len); 829 return (0); 830 } 831 return (-copyin(uaddr, kaddr, len)); 832 } 833 834 int 835 linux_copyout(const void *kaddr, void *uaddr, size_t len) 836 { 837 if (linux_remap_address(&uaddr, len)) { 838 if (uaddr == NULL) 839 return (-EFAULT); 840 memcpy(uaddr, kaddr, len); 841 return (0); 842 } 843 return (-copyout(kaddr, uaddr, len)); 844 } 845 846 size_t 847 linux_clear_user(void *_uaddr, size_t _len) 848 { 849 uint8_t *uaddr = _uaddr; 850 size_t len = _len; 851 852 /* make sure uaddr is aligned before going into the fast loop */ 853 while (((uintptr_t)uaddr & 7) != 0 && len > 7) { 854 if (subyte(uaddr, 0)) 855 return (_len); 856 uaddr++; 857 len--; 858 } 859 860 /* zero 8 bytes at a time */ 861 while (len > 7) { 862 #ifdef __LP64__ 863 if (suword64(uaddr, 0)) 864 return (_len); 865 #else 866 if (suword32(uaddr, 0)) 867 return (_len); 868 if (suword32(uaddr + 4, 0)) 869 return (_len); 870 #endif 871 uaddr += 8; 872 len -= 8; 873 } 874 875 /* zero fill end, if any */ 876 while (len > 0) { 877 if (subyte(uaddr, 0)) 878 return (_len); 879 uaddr++; 880 len--; 881 } 882 return (0); 883 } 884 885 int 886 linux_access_ok(const void *uaddr, size_t len) 887 { 888 uintptr_t saddr; 889 uintptr_t eaddr; 890 891 /* get start and end address */ 892 saddr = (uintptr_t)uaddr; 893 eaddr = (uintptr_t)uaddr + len; 894 895 /* verify addresses are valid for userspace */ 896 return ((saddr == eaddr) || 897 (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); 898 } 899 900 /* 901 * This function should return either EINTR or ERESTART depending on 902 * the signal type sent to this thread: 903 */ 904 static int 905 linux_get_error(struct task_struct *task, int error) 906 { 907 /* check for signal type interrupt code */ 908 if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { 909 error = -linux_schedule_get_interrupt_value(task); 910 if (error == 0) 911 error = EINTR; 912 } 913 return (error); 914 } 915 916 static int 917 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, 918 const struct file_operations *fop, u_long cmd, caddr_t data, 919 struct thread *td) 920 { 921 struct task_struct *task = current; 922 unsigned size; 923 int error; 924 925 size = IOCPARM_LEN(cmd); 926 /* refer to logic in sys_ioctl() */ 927 if (size > 0) { 928 /* 929 * Setup hint for linux_copyin() and linux_copyout(). 930 * 931 * Background: Linux code expects a user-space address 932 * while FreeBSD supplies a kernel-space address. 933 */ 934 task->bsd_ioctl_data = data; 935 task->bsd_ioctl_len = size; 936 data = (void *)LINUX_IOCTL_MIN_PTR; 937 } else { 938 /* fetch user-space pointer */ 939 data = *(void **)data; 940 } 941 #ifdef COMPAT_FREEBSD32 942 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 943 /* try the compat IOCTL handler first */ 944 if (fop->compat_ioctl != NULL) { 945 error = -OPW(fp, td, fop->compat_ioctl(filp, 946 cmd, (u_long)data)); 947 } else { 948 error = ENOTTY; 949 } 950 951 /* fallback to the regular IOCTL handler, if any */ 952 if (error == ENOTTY && fop->unlocked_ioctl != NULL) { 953 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 954 cmd, (u_long)data)); 955 } 956 } else 957 #endif 958 { 959 if (fop->unlocked_ioctl != NULL) { 960 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 961 cmd, (u_long)data)); 962 } else { 963 error = ENOTTY; 964 } 965 } 966 if (size > 0) { 967 task->bsd_ioctl_data = NULL; 968 task->bsd_ioctl_len = 0; 969 } 970 971 if (error == EWOULDBLOCK) { 972 /* update kqfilter status, if any */ 973 linux_file_kqfilter_poll(filp, 974 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 975 } else { 976 error = linux_get_error(task, error); 977 } 978 return (error); 979 } 980 981 #define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) 982 983 /* 984 * This function atomically updates the poll wakeup state and returns 985 * the previous state at the time of update. 986 */ 987 static uint8_t 988 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) 989 { 990 int c, old; 991 992 c = v->counter; 993 994 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) 995 c = old; 996 997 return (c); 998 } 999 1000 static int 1001 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) 1002 { 1003 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1004 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1005 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1006 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, 1007 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ 1008 }; 1009 struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); 1010 1011 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1012 case LINUX_FWQ_STATE_QUEUED: 1013 linux_poll_wakeup(filp); 1014 return (1); 1015 default: 1016 return (0); 1017 } 1018 } 1019 1020 void 1021 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) 1022 { 1023 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1024 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, 1025 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1026 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ 1027 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, 1028 }; 1029 1030 /* check if we are called inside the select system call */ 1031 if (p == LINUX_POLL_TABLE_NORMAL) 1032 selrecord(curthread, &filp->f_selinfo); 1033 1034 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1035 case LINUX_FWQ_STATE_INIT: 1036 /* NOTE: file handles can only belong to one wait-queue */ 1037 filp->f_wait_queue.wqh = wqh; 1038 filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; 1039 add_wait_queue(wqh, &filp->f_wait_queue.wq); 1040 atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); 1041 break; 1042 default: 1043 break; 1044 } 1045 } 1046 1047 static void 1048 linux_poll_wait_dequeue(struct linux_file *filp) 1049 { 1050 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1051 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1052 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, 1053 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, 1054 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, 1055 }; 1056 1057 seldrain(&filp->f_selinfo); 1058 1059 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1060 case LINUX_FWQ_STATE_NOT_READY: 1061 case LINUX_FWQ_STATE_QUEUED: 1062 case LINUX_FWQ_STATE_READY: 1063 remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); 1064 break; 1065 default: 1066 break; 1067 } 1068 } 1069 1070 void 1071 linux_poll_wakeup(struct linux_file *filp) 1072 { 1073 /* this function should be NULL-safe */ 1074 if (filp == NULL) 1075 return; 1076 1077 selwakeup(&filp->f_selinfo); 1078 1079 spin_lock(&filp->f_kqlock); 1080 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | 1081 LINUX_KQ_FLAG_NEED_WRITE; 1082 1083 /* make sure the "knote" gets woken up */ 1084 KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); 1085 spin_unlock(&filp->f_kqlock); 1086 } 1087 1088 static struct linux_file * 1089 __get_file_rcu(struct linux_file **f) 1090 { 1091 struct linux_file *file1, *file2; 1092 1093 file1 = READ_ONCE(*f); 1094 if (file1 == NULL) 1095 return (NULL); 1096 1097 if (!refcount_acquire_if_not_zero( 1098 file1->_file == NULL ? &file1->f_count : &file1->_file->f_count)) 1099 return (ERR_PTR(-EAGAIN)); 1100 1101 file2 = READ_ONCE(*f); 1102 if (file2 == file1) 1103 return (file2); 1104 1105 fput(file1); 1106 return (ERR_PTR(-EAGAIN)); 1107 } 1108 1109 struct linux_file * 1110 linux_get_file_rcu(struct linux_file **f) 1111 { 1112 struct linux_file *file1; 1113 1114 for (;;) { 1115 file1 = __get_file_rcu(f); 1116 if (file1 == NULL) 1117 return (NULL); 1118 1119 if (IS_ERR(file1)) 1120 continue; 1121 1122 return (file1); 1123 } 1124 } 1125 1126 struct linux_file * 1127 get_file_active(struct linux_file **f) 1128 { 1129 struct linux_file *file1; 1130 1131 rcu_read_lock(); 1132 file1 = __get_file_rcu(f); 1133 rcu_read_unlock(); 1134 if (IS_ERR(file1)) 1135 file1 = NULL; 1136 1137 return (file1); 1138 } 1139 1140 static void 1141 linux_file_kqfilter_detach(struct knote *kn) 1142 { 1143 struct linux_file *filp = kn->kn_hook; 1144 1145 spin_lock(&filp->f_kqlock); 1146 knlist_remove(&filp->f_selinfo.si_note, kn, 1); 1147 spin_unlock(&filp->f_kqlock); 1148 } 1149 1150 static int 1151 linux_file_kqfilter_read_event(struct knote *kn, long hint) 1152 { 1153 struct linux_file *filp = kn->kn_hook; 1154 1155 mtx_assert(&filp->f_kqlock, MA_OWNED); 1156 1157 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); 1158 } 1159 1160 static int 1161 linux_file_kqfilter_write_event(struct knote *kn, long hint) 1162 { 1163 struct linux_file *filp = kn->kn_hook; 1164 1165 mtx_assert(&filp->f_kqlock, MA_OWNED); 1166 1167 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); 1168 } 1169 1170 static const struct filterops linux_dev_kqfiltops_read = { 1171 .f_isfd = 1, 1172 .f_detach = linux_file_kqfilter_detach, 1173 .f_event = linux_file_kqfilter_read_event, 1174 }; 1175 1176 static const struct filterops linux_dev_kqfiltops_write = { 1177 .f_isfd = 1, 1178 .f_detach = linux_file_kqfilter_detach, 1179 .f_event = linux_file_kqfilter_write_event, 1180 }; 1181 1182 static void 1183 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) 1184 { 1185 struct thread *td; 1186 const struct file_operations *fop; 1187 struct linux_cdev *ldev; 1188 int temp; 1189 1190 if ((filp->f_kqflags & kqflags) == 0) 1191 return; 1192 1193 td = curthread; 1194 1195 linux_get_fop(filp, &fop, &ldev); 1196 /* get the latest polling state */ 1197 temp = OPW(filp->_file, td, fop->poll(filp, NULL)); 1198 linux_drop_fop(ldev); 1199 1200 spin_lock(&filp->f_kqlock); 1201 /* clear kqflags */ 1202 filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | 1203 LINUX_KQ_FLAG_NEED_WRITE); 1204 /* update kqflags */ 1205 if ((temp & (POLLIN | POLLOUT)) != 0) { 1206 if ((temp & POLLIN) != 0) 1207 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; 1208 if ((temp & POLLOUT) != 0) 1209 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; 1210 1211 /* make sure the "knote" gets woken up */ 1212 KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); 1213 } 1214 spin_unlock(&filp->f_kqlock); 1215 } 1216 1217 static int 1218 linux_file_kqfilter(struct file *file, struct knote *kn) 1219 { 1220 struct linux_file *filp; 1221 struct thread *td; 1222 int error; 1223 1224 td = curthread; 1225 filp = (struct linux_file *)file->f_data; 1226 filp->f_flags = file->f_flag; 1227 if (filp->f_op->poll == NULL) 1228 return (EINVAL); 1229 1230 spin_lock(&filp->f_kqlock); 1231 switch (kn->kn_filter) { 1232 case EVFILT_READ: 1233 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; 1234 kn->kn_fop = &linux_dev_kqfiltops_read; 1235 kn->kn_hook = filp; 1236 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1237 error = 0; 1238 break; 1239 case EVFILT_WRITE: 1240 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; 1241 kn->kn_fop = &linux_dev_kqfiltops_write; 1242 kn->kn_hook = filp; 1243 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1244 error = 0; 1245 break; 1246 default: 1247 error = EINVAL; 1248 break; 1249 } 1250 spin_unlock(&filp->f_kqlock); 1251 1252 if (error == 0) { 1253 linux_set_current(td); 1254 1255 /* update kqfilter status, if any */ 1256 linux_file_kqfilter_poll(filp, 1257 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 1258 } 1259 return (error); 1260 } 1261 1262 static int 1263 linux_file_mmap_single(struct file *fp, const struct file_operations *fop, 1264 vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, 1265 int nprot, bool is_shared, struct thread *td) 1266 { 1267 struct task_struct *task; 1268 struct vm_area_struct *vmap; 1269 struct mm_struct *mm; 1270 struct linux_file *filp; 1271 vm_memattr_t attr; 1272 int error; 1273 1274 filp = (struct linux_file *)fp->f_data; 1275 filp->f_flags = fp->f_flag; 1276 1277 if (fop->mmap == NULL) 1278 return (EOPNOTSUPP); 1279 1280 linux_set_current(td); 1281 1282 /* 1283 * The same VM object might be shared by multiple processes 1284 * and the mm_struct is usually freed when a process exits. 1285 * 1286 * The atomic reference below makes sure the mm_struct is 1287 * available as long as the vmap is in the linux_vma_head. 1288 */ 1289 task = current; 1290 mm = task->mm; 1291 if (atomic_inc_not_zero(&mm->mm_users) == 0) 1292 return (EINVAL); 1293 1294 vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); 1295 vmap->vm_start = 0; 1296 vmap->vm_end = size; 1297 vmap->vm_pgoff = *offset / PAGE_SIZE; 1298 vmap->vm_pfn = 0; 1299 vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); 1300 if (is_shared) 1301 vmap->vm_flags |= VM_SHARED; 1302 vmap->vm_ops = NULL; 1303 vmap->vm_file = get_file(filp); 1304 vmap->vm_mm = mm; 1305 1306 if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { 1307 error = linux_get_error(task, EINTR); 1308 } else { 1309 error = -OPW(fp, td, fop->mmap(filp, vmap)); 1310 error = linux_get_error(task, error); 1311 up_write(&vmap->vm_mm->mmap_sem); 1312 } 1313 1314 if (error != 0) { 1315 linux_cdev_handle_free(vmap); 1316 return (error); 1317 } 1318 1319 attr = pgprot2cachemode(vmap->vm_page_prot); 1320 1321 if (vmap->vm_ops != NULL) { 1322 struct vm_area_struct *ptr; 1323 void *vm_private_data; 1324 bool vm_no_fault; 1325 1326 if (vmap->vm_ops->open == NULL || 1327 vmap->vm_ops->close == NULL || 1328 vmap->vm_private_data == NULL) { 1329 /* free allocated VM area struct */ 1330 linux_cdev_handle_free(vmap); 1331 return (EINVAL); 1332 } 1333 1334 vm_private_data = vmap->vm_private_data; 1335 1336 rw_wlock(&linux_vma_lock); 1337 TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { 1338 if (ptr->vm_private_data == vm_private_data) 1339 break; 1340 } 1341 /* check if there is an existing VM area struct */ 1342 if (ptr != NULL) { 1343 /* check if the VM area structure is invalid */ 1344 if (ptr->vm_ops == NULL || 1345 ptr->vm_ops->open == NULL || 1346 ptr->vm_ops->close == NULL) { 1347 error = ESTALE; 1348 vm_no_fault = 1; 1349 } else { 1350 error = EEXIST; 1351 vm_no_fault = (ptr->vm_ops->fault == NULL); 1352 } 1353 } else { 1354 /* insert VM area structure into list */ 1355 TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); 1356 error = 0; 1357 vm_no_fault = (vmap->vm_ops->fault == NULL); 1358 } 1359 rw_wunlock(&linux_vma_lock); 1360 1361 if (error != 0) { 1362 /* free allocated VM area struct */ 1363 linux_cdev_handle_free(vmap); 1364 /* check for stale VM area struct */ 1365 if (error != EEXIST) 1366 return (error); 1367 } 1368 1369 /* check if there is no fault handler */ 1370 if (vm_no_fault) { 1371 *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, 1372 &linux_cdev_pager_ops[1], size, nprot, *offset, 1373 td->td_ucred); 1374 } else { 1375 *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, 1376 &linux_cdev_pager_ops[0], size, nprot, *offset, 1377 td->td_ucred); 1378 } 1379 1380 /* check if allocating the VM object failed */ 1381 if (*object == NULL) { 1382 if (error == 0) { 1383 /* remove VM area struct from list */ 1384 linux_cdev_handle_remove(vmap); 1385 /* free allocated VM area struct */ 1386 linux_cdev_handle_free(vmap); 1387 } 1388 return (EINVAL); 1389 } 1390 } else { 1391 struct sglist *sg; 1392 1393 sg = sglist_alloc(1, M_WAITOK); 1394 sglist_append_phys(sg, 1395 (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); 1396 1397 *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, 1398 nprot, 0, td->td_ucred); 1399 1400 linux_cdev_handle_free(vmap); 1401 1402 if (*object == NULL) { 1403 sglist_free(sg); 1404 return (EINVAL); 1405 } 1406 } 1407 1408 if (attr != VM_MEMATTR_DEFAULT) { 1409 VM_OBJECT_WLOCK(*object); 1410 vm_object_set_memattr(*object, attr); 1411 VM_OBJECT_WUNLOCK(*object); 1412 } 1413 *offset = 0; 1414 return (0); 1415 } 1416 1417 struct cdevsw linuxcdevsw = { 1418 .d_version = D_VERSION, 1419 .d_fdopen = linux_dev_fdopen, 1420 .d_name = "lkpidev", 1421 }; 1422 1423 static int 1424 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, 1425 int flags, struct thread *td) 1426 { 1427 struct linux_file *filp; 1428 const struct file_operations *fop; 1429 struct linux_cdev *ldev; 1430 ssize_t bytes; 1431 int error; 1432 1433 error = 0; 1434 filp = (struct linux_file *)file->f_data; 1435 filp->f_flags = file->f_flag; 1436 /* XXX no support for I/O vectors currently */ 1437 if (uio->uio_iovcnt != 1) 1438 return (EOPNOTSUPP); 1439 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1440 return (EINVAL); 1441 linux_set_current(td); 1442 linux_get_fop(filp, &fop, &ldev); 1443 if (fop->read != NULL) { 1444 bytes = OPW(file, td, fop->read(filp, 1445 uio->uio_iov->iov_base, 1446 uio->uio_iov->iov_len, &uio->uio_offset)); 1447 if (bytes >= 0) { 1448 uio->uio_iov->iov_base = 1449 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1450 uio->uio_iov->iov_len -= bytes; 1451 uio->uio_resid -= bytes; 1452 } else { 1453 error = linux_get_error(current, -bytes); 1454 } 1455 } else 1456 error = ENXIO; 1457 1458 /* update kqfilter status, if any */ 1459 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); 1460 linux_drop_fop(ldev); 1461 1462 return (error); 1463 } 1464 1465 static int 1466 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, 1467 int flags, struct thread *td) 1468 { 1469 struct linux_file *filp; 1470 const struct file_operations *fop; 1471 struct linux_cdev *ldev; 1472 ssize_t bytes; 1473 int error; 1474 1475 filp = (struct linux_file *)file->f_data; 1476 filp->f_flags = file->f_flag; 1477 /* XXX no support for I/O vectors currently */ 1478 if (uio->uio_iovcnt != 1) 1479 return (EOPNOTSUPP); 1480 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1481 return (EINVAL); 1482 linux_set_current(td); 1483 linux_get_fop(filp, &fop, &ldev); 1484 if (fop->write != NULL) { 1485 bytes = OPW(file, td, fop->write(filp, 1486 uio->uio_iov->iov_base, 1487 uio->uio_iov->iov_len, &uio->uio_offset)); 1488 if (bytes >= 0) { 1489 uio->uio_iov->iov_base = 1490 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1491 uio->uio_iov->iov_len -= bytes; 1492 uio->uio_resid -= bytes; 1493 error = 0; 1494 } else { 1495 error = linux_get_error(current, -bytes); 1496 } 1497 } else 1498 error = ENXIO; 1499 1500 /* update kqfilter status, if any */ 1501 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); 1502 1503 linux_drop_fop(ldev); 1504 1505 return (error); 1506 } 1507 1508 static int 1509 linux_file_poll(struct file *file, int events, struct ucred *active_cred, 1510 struct thread *td) 1511 { 1512 struct linux_file *filp; 1513 const struct file_operations *fop; 1514 struct linux_cdev *ldev; 1515 int revents; 1516 1517 filp = (struct linux_file *)file->f_data; 1518 filp->f_flags = file->f_flag; 1519 linux_set_current(td); 1520 linux_get_fop(filp, &fop, &ldev); 1521 if (fop->poll != NULL) { 1522 revents = OPW(file, td, fop->poll(filp, 1523 LINUX_POLL_TABLE_NORMAL)) & events; 1524 } else { 1525 revents = 0; 1526 } 1527 linux_drop_fop(ldev); 1528 return (revents); 1529 } 1530 1531 static int 1532 linux_file_close(struct file *file, struct thread *td) 1533 { 1534 struct linux_file *filp; 1535 int (*release)(struct inode *, struct linux_file *); 1536 const struct file_operations *fop; 1537 struct linux_cdev *ldev; 1538 int error; 1539 1540 filp = (struct linux_file *)file->f_data; 1541 1542 KASSERT(file_count(filp) == 0, 1543 ("File refcount(%d) is not zero", file_count(filp))); 1544 1545 if (td == NULL) 1546 td = curthread; 1547 1548 error = 0; 1549 filp->f_flags = file->f_flag; 1550 linux_set_current(td); 1551 linux_poll_wait_dequeue(filp); 1552 linux_get_fop(filp, &fop, &ldev); 1553 /* 1554 * Always use the real release function, if any, to avoid 1555 * leaking device resources: 1556 */ 1557 release = filp->f_op->release; 1558 if (release != NULL) 1559 error = -OPW(file, td, release(filp->f_vnode, filp)); 1560 funsetown(&filp->f_sigio); 1561 if (filp->f_vnode != NULL) 1562 vrele(filp->f_vnode); 1563 linux_drop_fop(ldev); 1564 ldev = filp->f_cdev; 1565 if (ldev != NULL) 1566 linux_cdev_deref(ldev); 1567 linux_synchronize_rcu(RCU_TYPE_REGULAR); 1568 kfree(filp); 1569 1570 return (error); 1571 } 1572 1573 static int 1574 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, 1575 struct thread *td) 1576 { 1577 struct linux_file *filp; 1578 const struct file_operations *fop; 1579 struct linux_cdev *ldev; 1580 struct fiodgname_arg *fgn; 1581 const char *p; 1582 int error, i; 1583 1584 error = 0; 1585 filp = (struct linux_file *)fp->f_data; 1586 filp->f_flags = fp->f_flag; 1587 linux_get_fop(filp, &fop, &ldev); 1588 1589 linux_set_current(td); 1590 switch (cmd) { 1591 case FIONBIO: 1592 break; 1593 case FIOASYNC: 1594 if (fop->fasync == NULL) 1595 break; 1596 error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); 1597 break; 1598 case FIOSETOWN: 1599 error = fsetown(*(int *)data, &filp->f_sigio); 1600 if (error == 0) { 1601 if (fop->fasync == NULL) 1602 break; 1603 error = -OPW(fp, td, fop->fasync(0, filp, 1604 fp->f_flag & FASYNC)); 1605 } 1606 break; 1607 case FIOGETOWN: 1608 *(int *)data = fgetown(&filp->f_sigio); 1609 break; 1610 case FIODGNAME: 1611 #ifdef COMPAT_FREEBSD32 1612 case FIODGNAME_32: 1613 #endif 1614 if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { 1615 error = ENXIO; 1616 break; 1617 } 1618 fgn = data; 1619 p = devtoname(filp->f_cdev->cdev); 1620 i = strlen(p) + 1; 1621 if (i > fgn->len) { 1622 error = EINVAL; 1623 break; 1624 } 1625 error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i); 1626 break; 1627 default: 1628 error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); 1629 break; 1630 } 1631 linux_drop_fop(ldev); 1632 return (error); 1633 } 1634 1635 static int 1636 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1637 vm_prot_t maxprot, int flags, struct file *fp, 1638 vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) 1639 { 1640 /* 1641 * Character devices do not provide private mappings 1642 * of any kind: 1643 */ 1644 if ((maxprot & VM_PROT_WRITE) == 0 && 1645 (prot & VM_PROT_WRITE) != 0) 1646 return (EACCES); 1647 if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0) 1648 return (EINVAL); 1649 1650 return (linux_file_mmap_single(fp, fop, foff, objsize, objp, 1651 (int)prot, (flags & MAP_SHARED) ? true : false, td)); 1652 } 1653 1654 static int 1655 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1656 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1657 struct thread *td) 1658 { 1659 struct linux_file *filp; 1660 const struct file_operations *fop; 1661 struct linux_cdev *ldev; 1662 struct mount *mp; 1663 struct vnode *vp; 1664 vm_object_t object; 1665 vm_prot_t maxprot; 1666 int error; 1667 1668 filp = (struct linux_file *)fp->f_data; 1669 1670 vp = filp->f_vnode; 1671 if (vp == NULL) 1672 return (EOPNOTSUPP); 1673 1674 /* 1675 * Ensure that file and memory protections are 1676 * compatible. 1677 */ 1678 mp = vp->v_mount; 1679 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1680 maxprot = VM_PROT_NONE; 1681 if ((prot & VM_PROT_EXECUTE) != 0) 1682 return (EACCES); 1683 } else 1684 maxprot = VM_PROT_EXECUTE; 1685 if ((fp->f_flag & FREAD) != 0) 1686 maxprot |= VM_PROT_READ; 1687 else if ((prot & VM_PROT_READ) != 0) 1688 return (EACCES); 1689 1690 /* 1691 * If we are sharing potential changes via MAP_SHARED and we 1692 * are trying to get write permission although we opened it 1693 * without asking for it, bail out. 1694 * 1695 * Note that most character devices always share mappings. 1696 * 1697 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE 1698 * requests rather than doing it here. 1699 */ 1700 if ((flags & MAP_SHARED) != 0) { 1701 if ((fp->f_flag & FWRITE) != 0) 1702 maxprot |= VM_PROT_WRITE; 1703 else if ((prot & VM_PROT_WRITE) != 0) 1704 return (EACCES); 1705 } 1706 maxprot &= cap_maxprot; 1707 1708 linux_get_fop(filp, &fop, &ldev); 1709 error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp, 1710 &foff, fop, &object); 1711 if (error != 0) 1712 goto out; 1713 1714 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1715 foff, FALSE, td); 1716 if (error != 0) 1717 vm_object_deallocate(object); 1718 out: 1719 linux_drop_fop(ldev); 1720 return (error); 1721 } 1722 1723 static int 1724 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 1725 { 1726 struct linux_file *filp; 1727 struct vnode *vp; 1728 int error; 1729 1730 filp = (struct linux_file *)fp->f_data; 1731 if (filp->f_vnode == NULL) 1732 return (EOPNOTSUPP); 1733 1734 vp = filp->f_vnode; 1735 1736 vn_lock(vp, LK_SHARED | LK_RETRY); 1737 error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED); 1738 VOP_UNLOCK(vp); 1739 1740 return (error); 1741 } 1742 1743 static int 1744 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1745 struct filedesc *fdp) 1746 { 1747 struct linux_file *filp; 1748 struct vnode *vp; 1749 int error; 1750 1751 filp = fp->f_data; 1752 vp = filp->f_vnode; 1753 if (vp == NULL) { 1754 error = 0; 1755 kif->kf_type = KF_TYPE_DEV; 1756 } else { 1757 vref(vp); 1758 FILEDESC_SUNLOCK(fdp); 1759 error = vn_fill_kinfo_vnode(vp, kif); 1760 vrele(vp); 1761 kif->kf_type = KF_TYPE_VNODE; 1762 FILEDESC_SLOCK(fdp); 1763 } 1764 return (error); 1765 } 1766 1767 unsigned int 1768 linux_iminor(struct inode *inode) 1769 { 1770 struct linux_cdev *ldev; 1771 1772 if (inode == NULL || inode->v_rdev == NULL || 1773 inode->v_rdev->si_devsw != &linuxcdevsw) 1774 return (-1U); 1775 ldev = inode->v_rdev->si_drv1; 1776 if (ldev == NULL) 1777 return (-1U); 1778 1779 return (minor(ldev->dev)); 1780 } 1781 1782 static int 1783 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td) 1784 { 1785 struct linux_file *filp1, *filp2; 1786 1787 if (fp2->f_type != DTYPE_DEV) 1788 return (3); 1789 1790 filp1 = fp1->f_data; 1791 filp2 = fp2->f_data; 1792 return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev)); 1793 } 1794 1795 const struct fileops linuxfileops = { 1796 .fo_read = linux_file_read, 1797 .fo_write = linux_file_write, 1798 .fo_truncate = invfo_truncate, 1799 .fo_kqfilter = linux_file_kqfilter, 1800 .fo_stat = linux_file_stat, 1801 .fo_fill_kinfo = linux_file_fill_kinfo, 1802 .fo_poll = linux_file_poll, 1803 .fo_close = linux_file_close, 1804 .fo_ioctl = linux_file_ioctl, 1805 .fo_mmap = linux_file_mmap, 1806 .fo_chmod = invfo_chmod, 1807 .fo_chown = invfo_chown, 1808 .fo_sendfile = invfo_sendfile, 1809 .fo_cmp = linux_file_kcmp, 1810 .fo_flags = DFLAG_PASSABLE, 1811 }; 1812 1813 /* 1814 * Hash of vmmap addresses. This is infrequently accessed and does not 1815 * need to be particularly large. This is done because we must store the 1816 * caller's idea of the map size to properly unmap. 1817 */ 1818 struct vmmap { 1819 LIST_ENTRY(vmmap) vm_next; 1820 void *vm_addr; 1821 unsigned long vm_size; 1822 }; 1823 1824 struct vmmaphd { 1825 struct vmmap *lh_first; 1826 }; 1827 #define VMMAP_HASH_SIZE 64 1828 #define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) 1829 #define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK 1830 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; 1831 static struct mtx vmmaplock; 1832 1833 static void 1834 vmmap_add(void *addr, unsigned long size) 1835 { 1836 struct vmmap *vmmap; 1837 1838 vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); 1839 mtx_lock(&vmmaplock); 1840 vmmap->vm_size = size; 1841 vmmap->vm_addr = addr; 1842 LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); 1843 mtx_unlock(&vmmaplock); 1844 } 1845 1846 static struct vmmap * 1847 vmmap_remove(void *addr) 1848 { 1849 struct vmmap *vmmap; 1850 1851 mtx_lock(&vmmaplock); 1852 LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) 1853 if (vmmap->vm_addr == addr) 1854 break; 1855 if (vmmap) 1856 LIST_REMOVE(vmmap, vm_next); 1857 mtx_unlock(&vmmaplock); 1858 1859 return (vmmap); 1860 } 1861 1862 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) 1863 void * 1864 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) 1865 { 1866 void *addr; 1867 1868 addr = pmap_mapdev_attr(phys_addr, size, attr); 1869 if (addr == NULL) 1870 return (NULL); 1871 vmmap_add(addr, size); 1872 1873 return (addr); 1874 } 1875 #endif 1876 1877 void 1878 iounmap(void *addr) 1879 { 1880 struct vmmap *vmmap; 1881 1882 vmmap = vmmap_remove(addr); 1883 if (vmmap == NULL) 1884 return; 1885 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv) 1886 pmap_unmapdev(addr, vmmap->vm_size); 1887 #endif 1888 kfree(vmmap); 1889 } 1890 1891 void * 1892 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) 1893 { 1894 vm_offset_t off; 1895 size_t size; 1896 1897 size = count * PAGE_SIZE; 1898 off = kva_alloc(size); 1899 if (off == 0) 1900 return (NULL); 1901 vmmap_add((void *)off, size); 1902 pmap_qenter(off, pages, count); 1903 1904 return ((void *)off); 1905 } 1906 1907 void 1908 vunmap(void *addr) 1909 { 1910 struct vmmap *vmmap; 1911 1912 vmmap = vmmap_remove(addr); 1913 if (vmmap == NULL) 1914 return; 1915 pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); 1916 kva_free((vm_offset_t)addr, vmmap->vm_size); 1917 kfree(vmmap); 1918 } 1919 1920 static char * 1921 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap) 1922 { 1923 unsigned int len; 1924 char *p; 1925 va_list aq; 1926 1927 va_copy(aq, ap); 1928 len = vsnprintf(NULL, 0, fmt, aq); 1929 va_end(aq); 1930 1931 if (dev != NULL) 1932 p = devm_kmalloc(dev, len + 1, gfp); 1933 else 1934 p = kmalloc(len + 1, gfp); 1935 if (p != NULL) 1936 vsnprintf(p, len + 1, fmt, ap); 1937 1938 return (p); 1939 } 1940 1941 char * 1942 kvasprintf(gfp_t gfp, const char *fmt, va_list ap) 1943 { 1944 1945 return (devm_kvasprintf(NULL, gfp, fmt, ap)); 1946 } 1947 1948 char * 1949 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...) 1950 { 1951 va_list ap; 1952 char *p; 1953 1954 va_start(ap, fmt); 1955 p = devm_kvasprintf(dev, gfp, fmt, ap); 1956 va_end(ap); 1957 1958 return (p); 1959 } 1960 1961 char * 1962 kasprintf(gfp_t gfp, const char *fmt, ...) 1963 { 1964 va_list ap; 1965 char *p; 1966 1967 va_start(ap, fmt); 1968 p = kvasprintf(gfp, fmt, ap); 1969 va_end(ap); 1970 1971 return (p); 1972 } 1973 1974 int 1975 __lkpi_hexdump_printf(void *arg1 __unused, const char *fmt, ...) 1976 { 1977 va_list ap; 1978 int result; 1979 1980 va_start(ap, fmt); 1981 result = vprintf(fmt, ap); 1982 va_end(ap); 1983 return (result); 1984 } 1985 1986 int 1987 __lkpi_hexdump_sbuf_printf(void *arg1, const char *fmt, ...) 1988 { 1989 va_list ap; 1990 int result; 1991 1992 va_start(ap, fmt); 1993 result = sbuf_vprintf(arg1, fmt, ap); 1994 va_end(ap); 1995 return (result); 1996 } 1997 1998 void 1999 lkpi_hex_dump(int(*_fpf)(void *, const char *, ...), void *arg1, 2000 const char *level, const char *prefix_str, 2001 const int prefix_type, const int rowsize, const int groupsize, 2002 const void *buf, size_t len, const bool ascii) 2003 { 2004 typedef const struct { long long value; } __packed *print_64p_t; 2005 typedef const struct { uint32_t value; } __packed *print_32p_t; 2006 typedef const struct { uint16_t value; } __packed *print_16p_t; 2007 const void *buf_old = buf; 2008 int row; 2009 2010 while (len > 0) { 2011 if (level != NULL) 2012 _fpf(arg1, "%s", level); 2013 if (prefix_str != NULL) 2014 _fpf(arg1, "%s ", prefix_str); 2015 2016 switch (prefix_type) { 2017 case DUMP_PREFIX_ADDRESS: 2018 _fpf(arg1, "[%p] ", buf); 2019 break; 2020 case DUMP_PREFIX_OFFSET: 2021 _fpf(arg1, "[%#tx] ", ((const char *)buf - 2022 (const char *)buf_old)); 2023 break; 2024 default: 2025 break; 2026 } 2027 for (row = 0; row != rowsize; row++) { 2028 if (groupsize == 8 && len > 7) { 2029 _fpf(arg1, "%016llx ", ((print_64p_t)buf)->value); 2030 buf = (const uint8_t *)buf + 8; 2031 len -= 8; 2032 } else if (groupsize == 4 && len > 3) { 2033 _fpf(arg1, "%08x ", ((print_32p_t)buf)->value); 2034 buf = (const uint8_t *)buf + 4; 2035 len -= 4; 2036 } else if (groupsize == 2 && len > 1) { 2037 _fpf(arg1, "%04x ", ((print_16p_t)buf)->value); 2038 buf = (const uint8_t *)buf + 2; 2039 len -= 2; 2040 } else if (len > 0) { 2041 _fpf(arg1, "%02x ", *(const uint8_t *)buf); 2042 buf = (const uint8_t *)buf + 1; 2043 len--; 2044 } else { 2045 break; 2046 } 2047 } 2048 _fpf(arg1, "\n"); 2049 } 2050 } 2051 2052 static void 2053 linux_timer_callback_wrapper(void *context) 2054 { 2055 struct timer_list *timer; 2056 2057 timer = context; 2058 2059 /* the timer is about to be shutdown permanently */ 2060 if (timer->function == NULL) 2061 return; 2062 2063 if (linux_set_current_flags(curthread, M_NOWAIT)) { 2064 /* try again later */ 2065 callout_reset(&timer->callout, 1, 2066 &linux_timer_callback_wrapper, timer); 2067 return; 2068 } 2069 2070 timer->function(timer->data); 2071 } 2072 2073 static int 2074 linux_timer_jiffies_until(unsigned long expires) 2075 { 2076 unsigned long delta = expires - jiffies; 2077 2078 /* 2079 * Guard against already expired values and make sure that the value can 2080 * be used as a tick count, rather than a jiffies count. 2081 */ 2082 if ((long)delta < 1) 2083 delta = 1; 2084 else if (delta > INT_MAX) 2085 delta = INT_MAX; 2086 return ((int)delta); 2087 } 2088 2089 int 2090 mod_timer(struct timer_list *timer, unsigned long expires) 2091 { 2092 int ret; 2093 2094 timer->expires = expires; 2095 ret = callout_reset(&timer->callout, 2096 linux_timer_jiffies_until(expires), 2097 &linux_timer_callback_wrapper, timer); 2098 2099 MPASS(ret == 0 || ret == 1); 2100 2101 return (ret == 1); 2102 } 2103 2104 void 2105 add_timer(struct timer_list *timer) 2106 { 2107 2108 callout_reset(&timer->callout, 2109 linux_timer_jiffies_until(timer->expires), 2110 &linux_timer_callback_wrapper, timer); 2111 } 2112 2113 void 2114 add_timer_on(struct timer_list *timer, int cpu) 2115 { 2116 2117 callout_reset_on(&timer->callout, 2118 linux_timer_jiffies_until(timer->expires), 2119 &linux_timer_callback_wrapper, timer, cpu); 2120 } 2121 2122 int 2123 timer_delete(struct timer_list *timer) 2124 { 2125 2126 if (callout_stop(&(timer)->callout) == -1) 2127 return (0); 2128 return (1); 2129 } 2130 2131 int 2132 timer_delete_sync(struct timer_list *timer) 2133 { 2134 2135 if (callout_drain(&(timer)->callout) == -1) 2136 return (0); 2137 return (1); 2138 } 2139 2140 int 2141 timer_shutdown_sync(struct timer_list *timer) 2142 { 2143 2144 timer->function = NULL; 2145 return (del_timer_sync(timer)); 2146 } 2147 2148 /* greatest common divisor, Euclid equation */ 2149 static uint64_t 2150 lkpi_gcd_64(uint64_t a, uint64_t b) 2151 { 2152 uint64_t an; 2153 uint64_t bn; 2154 2155 while (b != 0) { 2156 an = b; 2157 bn = a % b; 2158 a = an; 2159 b = bn; 2160 } 2161 return (a); 2162 } 2163 2164 uint64_t lkpi_nsec2hz_rem; 2165 uint64_t lkpi_nsec2hz_div = 1000000000ULL; 2166 uint64_t lkpi_nsec2hz_max; 2167 2168 uint64_t lkpi_usec2hz_rem; 2169 uint64_t lkpi_usec2hz_div = 1000000ULL; 2170 uint64_t lkpi_usec2hz_max; 2171 2172 uint64_t lkpi_msec2hz_rem; 2173 uint64_t lkpi_msec2hz_div = 1000ULL; 2174 uint64_t lkpi_msec2hz_max; 2175 2176 static void 2177 linux_timer_init(void *arg) 2178 { 2179 uint64_t gcd; 2180 2181 /* 2182 * Compute an internal HZ value which can divide 2**32 to 2183 * avoid timer rounding problems when the tick value wraps 2184 * around 2**32: 2185 */ 2186 linux_timer_hz_mask = 1; 2187 while (linux_timer_hz_mask < (unsigned long)hz) 2188 linux_timer_hz_mask *= 2; 2189 linux_timer_hz_mask--; 2190 2191 /* compute some internal constants */ 2192 2193 lkpi_nsec2hz_rem = hz; 2194 lkpi_usec2hz_rem = hz; 2195 lkpi_msec2hz_rem = hz; 2196 2197 gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); 2198 lkpi_nsec2hz_rem /= gcd; 2199 lkpi_nsec2hz_div /= gcd; 2200 lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; 2201 2202 gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); 2203 lkpi_usec2hz_rem /= gcd; 2204 lkpi_usec2hz_div /= gcd; 2205 lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; 2206 2207 gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); 2208 lkpi_msec2hz_rem /= gcd; 2209 lkpi_msec2hz_div /= gcd; 2210 lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; 2211 } 2212 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); 2213 2214 void 2215 linux_complete_common(struct completion *c, int all) 2216 { 2217 sleepq_lock(c); 2218 if (all) { 2219 c->done = UINT_MAX; 2220 sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); 2221 } else { 2222 if (c->done != UINT_MAX) 2223 c->done++; 2224 sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); 2225 } 2226 sleepq_release(c); 2227 } 2228 2229 /* 2230 * Indefinite wait for done != 0 with or without signals. 2231 */ 2232 int 2233 linux_wait_for_common(struct completion *c, int flags) 2234 { 2235 struct task_struct *task; 2236 int error; 2237 2238 if (SCHEDULER_STOPPED()) 2239 return (0); 2240 2241 task = current; 2242 2243 if (flags != 0) 2244 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2245 else 2246 flags = SLEEPQ_SLEEP; 2247 error = 0; 2248 for (;;) { 2249 sleepq_lock(c); 2250 if (c->done) 2251 break; 2252 sleepq_add(c, NULL, "completion", flags, 0); 2253 if (flags & SLEEPQ_INTERRUPTIBLE) { 2254 DROP_GIANT(); 2255 error = -sleepq_wait_sig(c, 0); 2256 PICKUP_GIANT(); 2257 if (error != 0) { 2258 linux_schedule_save_interrupt_value(task, error); 2259 error = -ERESTARTSYS; 2260 goto intr; 2261 } 2262 } else { 2263 DROP_GIANT(); 2264 sleepq_wait(c, 0); 2265 PICKUP_GIANT(); 2266 } 2267 } 2268 if (c->done != UINT_MAX) 2269 c->done--; 2270 sleepq_release(c); 2271 2272 intr: 2273 return (error); 2274 } 2275 2276 /* 2277 * Time limited wait for done != 0 with or without signals. 2278 */ 2279 unsigned long 2280 linux_wait_for_timeout_common(struct completion *c, unsigned long timeout, 2281 int flags) 2282 { 2283 struct task_struct *task; 2284 unsigned long end = jiffies + timeout, error; 2285 2286 if (SCHEDULER_STOPPED()) 2287 return (0); 2288 2289 task = current; 2290 2291 if (flags != 0) 2292 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2293 else 2294 flags = SLEEPQ_SLEEP; 2295 2296 for (;;) { 2297 sleepq_lock(c); 2298 if (c->done) 2299 break; 2300 sleepq_add(c, NULL, "completion", flags, 0); 2301 sleepq_set_timeout(c, linux_timer_jiffies_until(end)); 2302 2303 DROP_GIANT(); 2304 if (flags & SLEEPQ_INTERRUPTIBLE) 2305 error = -sleepq_timedwait_sig(c, 0); 2306 else 2307 error = -sleepq_timedwait(c, 0); 2308 PICKUP_GIANT(); 2309 2310 if (error != 0) { 2311 /* check for timeout */ 2312 if (error == -EWOULDBLOCK) { 2313 error = 0; /* timeout */ 2314 } else { 2315 /* signal happened */ 2316 linux_schedule_save_interrupt_value(task, error); 2317 error = -ERESTARTSYS; 2318 } 2319 goto done; 2320 } 2321 } 2322 if (c->done != UINT_MAX) 2323 c->done--; 2324 sleepq_release(c); 2325 2326 /* return how many jiffies are left */ 2327 error = linux_timer_jiffies_until(end); 2328 done: 2329 return (error); 2330 } 2331 2332 int 2333 linux_try_wait_for_completion(struct completion *c) 2334 { 2335 int isdone; 2336 2337 sleepq_lock(c); 2338 isdone = (c->done != 0); 2339 if (c->done != 0 && c->done != UINT_MAX) 2340 c->done--; 2341 sleepq_release(c); 2342 return (isdone); 2343 } 2344 2345 int 2346 linux_completion_done(struct completion *c) 2347 { 2348 int isdone; 2349 2350 sleepq_lock(c); 2351 isdone = (c->done != 0); 2352 sleepq_release(c); 2353 return (isdone); 2354 } 2355 2356 static void 2357 linux_cdev_deref(struct linux_cdev *ldev) 2358 { 2359 if (refcount_release(&ldev->refs) && 2360 ldev->kobj.ktype == &linux_cdev_ktype) 2361 kfree(ldev); 2362 } 2363 2364 static void 2365 linux_cdev_release(struct kobject *kobj) 2366 { 2367 struct linux_cdev *cdev; 2368 struct kobject *parent; 2369 2370 cdev = container_of(kobj, struct linux_cdev, kobj); 2371 parent = kobj->parent; 2372 linux_destroy_dev(cdev); 2373 linux_cdev_deref(cdev); 2374 kobject_put(parent); 2375 } 2376 2377 static void 2378 linux_cdev_static_release(struct kobject *kobj) 2379 { 2380 struct cdev *cdev; 2381 struct linux_cdev *ldev; 2382 2383 ldev = container_of(kobj, struct linux_cdev, kobj); 2384 cdev = ldev->cdev; 2385 if (cdev != NULL) { 2386 destroy_dev(cdev); 2387 ldev->cdev = NULL; 2388 } 2389 kobject_put(kobj->parent); 2390 } 2391 2392 int 2393 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev) 2394 { 2395 int ret; 2396 2397 if (dev->devt != 0) { 2398 /* Set parent kernel object. */ 2399 ldev->kobj.parent = &dev->kobj; 2400 2401 /* 2402 * Unlike Linux we require the kobject of the 2403 * character device structure to have a valid name 2404 * before calling this function: 2405 */ 2406 if (ldev->kobj.name == NULL) 2407 return (-EINVAL); 2408 2409 ret = cdev_add(ldev, dev->devt, 1); 2410 if (ret) 2411 return (ret); 2412 } 2413 ret = device_add(dev); 2414 if (ret != 0 && dev->devt != 0) 2415 cdev_del(ldev); 2416 return (ret); 2417 } 2418 2419 void 2420 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev) 2421 { 2422 device_del(dev); 2423 2424 if (dev->devt != 0) 2425 cdev_del(ldev); 2426 } 2427 2428 static void 2429 linux_destroy_dev(struct linux_cdev *ldev) 2430 { 2431 2432 if (ldev->cdev == NULL) 2433 return; 2434 2435 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 2436 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 2437 2438 atomic_set_int(&ldev->siref, LDEV_SI_DTR); 2439 while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) 2440 pause("ldevdtr", hz / 4); 2441 2442 destroy_dev(ldev->cdev); 2443 ldev->cdev = NULL; 2444 } 2445 2446 const struct kobj_type linux_cdev_ktype = { 2447 .release = linux_cdev_release, 2448 }; 2449 2450 const struct kobj_type linux_cdev_static_ktype = { 2451 .release = linux_cdev_static_release, 2452 }; 2453 2454 static void 2455 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) 2456 { 2457 struct notifier_block *nb; 2458 struct netdev_notifier_info ni; 2459 2460 nb = arg; 2461 ni.ifp = ifp; 2462 ni.dev = (struct net_device *)ifp; 2463 if (linkstate == LINK_STATE_UP) 2464 nb->notifier_call(nb, NETDEV_UP, &ni); 2465 else 2466 nb->notifier_call(nb, NETDEV_DOWN, &ni); 2467 } 2468 2469 static void 2470 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) 2471 { 2472 struct notifier_block *nb; 2473 struct netdev_notifier_info ni; 2474 2475 nb = arg; 2476 ni.ifp = ifp; 2477 ni.dev = (struct net_device *)ifp; 2478 nb->notifier_call(nb, NETDEV_REGISTER, &ni); 2479 } 2480 2481 static void 2482 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) 2483 { 2484 struct notifier_block *nb; 2485 struct netdev_notifier_info ni; 2486 2487 nb = arg; 2488 ni.ifp = ifp; 2489 ni.dev = (struct net_device *)ifp; 2490 nb->notifier_call(nb, NETDEV_UNREGISTER, &ni); 2491 } 2492 2493 static void 2494 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) 2495 { 2496 struct notifier_block *nb; 2497 struct netdev_notifier_info ni; 2498 2499 nb = arg; 2500 ni.ifp = ifp; 2501 ni.dev = (struct net_device *)ifp; 2502 nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni); 2503 } 2504 2505 static void 2506 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) 2507 { 2508 struct notifier_block *nb; 2509 struct netdev_notifier_info ni; 2510 2511 nb = arg; 2512 ni.ifp = ifp; 2513 ni.dev = (struct net_device *)ifp; 2514 nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni); 2515 } 2516 2517 int 2518 register_netdevice_notifier(struct notifier_block *nb) 2519 { 2520 2521 nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( 2522 ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); 2523 nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( 2524 ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); 2525 nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( 2526 ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); 2527 nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( 2528 iflladdr_event, linux_handle_iflladdr_event, nb, 0); 2529 2530 return (0); 2531 } 2532 2533 int 2534 register_inetaddr_notifier(struct notifier_block *nb) 2535 { 2536 2537 nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( 2538 ifaddr_event, linux_handle_ifaddr_event, nb, 0); 2539 return (0); 2540 } 2541 2542 int 2543 unregister_netdevice_notifier(struct notifier_block *nb) 2544 { 2545 2546 EVENTHANDLER_DEREGISTER(ifnet_link_event, 2547 nb->tags[NETDEV_UP]); 2548 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 2549 nb->tags[NETDEV_REGISTER]); 2550 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 2551 nb->tags[NETDEV_UNREGISTER]); 2552 EVENTHANDLER_DEREGISTER(iflladdr_event, 2553 nb->tags[NETDEV_CHANGEADDR]); 2554 2555 return (0); 2556 } 2557 2558 int 2559 unregister_inetaddr_notifier(struct notifier_block *nb) 2560 { 2561 2562 EVENTHANDLER_DEREGISTER(ifaddr_event, 2563 nb->tags[NETDEV_CHANGEIFADDR]); 2564 2565 return (0); 2566 } 2567 2568 struct list_sort_thunk { 2569 int (*cmp)(void *, struct list_head *, struct list_head *); 2570 void *priv; 2571 }; 2572 2573 static inline int 2574 linux_le_cmp(const void *d1, const void *d2, void *priv) 2575 { 2576 struct list_head *le1, *le2; 2577 struct list_sort_thunk *thunk; 2578 2579 thunk = priv; 2580 le1 = *(__DECONST(struct list_head **, d1)); 2581 le2 = *(__DECONST(struct list_head **, d2)); 2582 return ((thunk->cmp)(thunk->priv, le1, le2)); 2583 } 2584 2585 void 2586 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, 2587 struct list_head *a, struct list_head *b)) 2588 { 2589 struct list_sort_thunk thunk; 2590 struct list_head **ar, *le; 2591 size_t count, i; 2592 2593 count = 0; 2594 list_for_each(le, head) 2595 count++; 2596 ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); 2597 i = 0; 2598 list_for_each(le, head) 2599 ar[i++] = le; 2600 thunk.cmp = cmp; 2601 thunk.priv = priv; 2602 qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk); 2603 INIT_LIST_HEAD(head); 2604 for (i = 0; i < count; i++) 2605 list_add_tail(ar[i], head); 2606 free(ar, M_KMALLOC); 2607 } 2608 2609 #if defined(__i386__) || defined(__amd64__) 2610 int 2611 linux_wbinvd_on_all_cpus(void) 2612 { 2613 2614 pmap_invalidate_cache(); 2615 return (0); 2616 } 2617 #endif 2618 2619 int 2620 linux_on_each_cpu(void callback(void *), void *data) 2621 { 2622 2623 smp_rendezvous(smp_no_rendezvous_barrier, callback, 2624 smp_no_rendezvous_barrier, data); 2625 return (0); 2626 } 2627 2628 int 2629 linux_in_atomic(void) 2630 { 2631 2632 return ((curthread->td_pflags & TDP_NOFAULTING) != 0); 2633 } 2634 2635 struct linux_cdev * 2636 linux_find_cdev(const char *name, unsigned major, unsigned minor) 2637 { 2638 dev_t dev = MKDEV(major, minor); 2639 struct cdev *cdev; 2640 2641 dev_lock(); 2642 LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { 2643 struct linux_cdev *ldev = cdev->si_drv1; 2644 if (ldev->dev == dev && 2645 strcmp(kobject_name(&ldev->kobj), name) == 0) { 2646 break; 2647 } 2648 } 2649 dev_unlock(); 2650 2651 return (cdev != NULL ? cdev->si_drv1 : NULL); 2652 } 2653 2654 int 2655 __register_chrdev(unsigned int major, unsigned int baseminor, 2656 unsigned int count, const char *name, 2657 const struct file_operations *fops) 2658 { 2659 struct linux_cdev *cdev; 2660 int ret = 0; 2661 int i; 2662 2663 for (i = baseminor; i < baseminor + count; i++) { 2664 cdev = cdev_alloc(); 2665 cdev->ops = fops; 2666 kobject_set_name(&cdev->kobj, name); 2667 2668 ret = cdev_add(cdev, makedev(major, i), 1); 2669 if (ret != 0) 2670 break; 2671 } 2672 return (ret); 2673 } 2674 2675 int 2676 __register_chrdev_p(unsigned int major, unsigned int baseminor, 2677 unsigned int count, const char *name, 2678 const struct file_operations *fops, uid_t uid, 2679 gid_t gid, int mode) 2680 { 2681 struct linux_cdev *cdev; 2682 int ret = 0; 2683 int i; 2684 2685 for (i = baseminor; i < baseminor + count; i++) { 2686 cdev = cdev_alloc(); 2687 cdev->ops = fops; 2688 kobject_set_name(&cdev->kobj, name); 2689 2690 ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); 2691 if (ret != 0) 2692 break; 2693 } 2694 return (ret); 2695 } 2696 2697 void 2698 __unregister_chrdev(unsigned int major, unsigned int baseminor, 2699 unsigned int count, const char *name) 2700 { 2701 struct linux_cdev *cdevp; 2702 int i; 2703 2704 for (i = baseminor; i < baseminor + count; i++) { 2705 cdevp = linux_find_cdev(name, major, i); 2706 if (cdevp != NULL) 2707 cdev_del(cdevp); 2708 } 2709 } 2710 2711 void 2712 linux_dump_stack(void) 2713 { 2714 #ifdef STACK 2715 struct stack st; 2716 2717 stack_save(&st); 2718 stack_print(&st); 2719 #endif 2720 } 2721 2722 int 2723 linuxkpi_net_ratelimit(void) 2724 { 2725 2726 return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps, 2727 lkpi_net_maxpps)); 2728 } 2729 2730 struct io_mapping * 2731 io_mapping_create_wc(resource_size_t base, unsigned long size) 2732 { 2733 struct io_mapping *mapping; 2734 2735 mapping = kmalloc(sizeof(*mapping), GFP_KERNEL); 2736 if (mapping == NULL) 2737 return (NULL); 2738 return (io_mapping_init_wc(mapping, base, size)); 2739 } 2740 2741 /* We likely want a linuxkpi_device.c at some point. */ 2742 bool 2743 device_can_wakeup(struct device *dev) 2744 { 2745 2746 if (dev == NULL) 2747 return (false); 2748 /* 2749 * XXX-BZ iwlwifi queries it as part of enabling WoWLAN. 2750 * Normally this would be based on a bool in dev->power.XXX. 2751 * Check such as PCI PCIM_PCAP_*PME. We have no way to enable this yet. 2752 * We may get away by directly calling into bsddev for as long as 2753 * we can assume PCI only avoiding changing struct device breaking KBI. 2754 */ 2755 pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__); 2756 return (false); 2757 } 2758 2759 static void 2760 devm_device_group_remove(struct device *dev, void *p) 2761 { 2762 const struct attribute_group **dr = p; 2763 const struct attribute_group *group = *dr; 2764 2765 sysfs_remove_group(&dev->kobj, group); 2766 } 2767 2768 int 2769 lkpi_devm_device_add_group(struct device *dev, 2770 const struct attribute_group *group) 2771 { 2772 const struct attribute_group **dr; 2773 int ret; 2774 2775 dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL); 2776 if (dr == NULL) 2777 return (-ENOMEM); 2778 2779 ret = sysfs_create_group(&dev->kobj, group); 2780 if (ret == 0) { 2781 *dr = group; 2782 devres_add(dev, dr); 2783 } else 2784 devres_free(dr); 2785 2786 return (ret); 2787 } 2788 2789 #if defined(__i386__) || defined(__amd64__) 2790 bool linux_cpu_has_clflush; 2791 struct cpuinfo_x86 boot_cpu_data; 2792 struct cpuinfo_x86 *__cpu_data; 2793 #endif 2794 2795 cpumask_t * 2796 lkpi_get_static_single_cpu_mask(int cpuid) 2797 { 2798 2799 KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n", 2800 __func__, cpuid)); 2801 KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n", 2802 __func__, cpuid)); 2803 2804 return (static_single_cpu_mask[cpuid]); 2805 } 2806 2807 bool 2808 lkpi_xen_initial_domain(void) 2809 { 2810 #ifdef XENHVM 2811 return (xen_initial_domain()); 2812 #else 2813 return (false); 2814 #endif 2815 } 2816 2817 bool 2818 lkpi_xen_pv_domain(void) 2819 { 2820 #ifdef XENHVM 2821 return (xen_pv_domain()); 2822 #else 2823 return (false); 2824 #endif 2825 } 2826 2827 static void 2828 linux_compat_init(void *arg) 2829 { 2830 struct sysctl_oid *rootoid; 2831 int i; 2832 2833 #if defined(__i386__) || defined(__amd64__) 2834 static const uint32_t x86_vendors[X86_VENDOR_NUM] = { 2835 [X86_VENDOR_INTEL] = CPU_VENDOR_INTEL, 2836 [X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX, 2837 [X86_VENDOR_AMD] = CPU_VENDOR_AMD, 2838 [X86_VENDOR_UMC] = CPU_VENDOR_UMC, 2839 [X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR, 2840 [X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA, 2841 [X86_VENDOR_NSC] = CPU_VENDOR_NSC, 2842 [X86_VENDOR_HYGON] = CPU_VENDOR_HYGON, 2843 }; 2844 uint8_t x86_vendor = X86_VENDOR_UNKNOWN; 2845 2846 for (i = 0; i < X86_VENDOR_NUM; i++) { 2847 if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) { 2848 x86_vendor = i; 2849 break; 2850 } 2851 } 2852 linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); 2853 boot_cpu_data.x86_clflush_size = cpu_clflush_line_size; 2854 boot_cpu_data.x86_max_cores = mp_ncpus; 2855 boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id); 2856 boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id); 2857 boot_cpu_data.x86_vendor = x86_vendor; 2858 2859 __cpu_data = kmalloc_array(mp_maxid + 1, 2860 sizeof(*__cpu_data), M_WAITOK | M_ZERO); 2861 CPU_FOREACH(i) { 2862 __cpu_data[i].x86_clflush_size = cpu_clflush_line_size; 2863 __cpu_data[i].x86_max_cores = mp_ncpus; 2864 __cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id); 2865 __cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id); 2866 __cpu_data[i].x86_vendor = x86_vendor; 2867 } 2868 #endif 2869 rw_init(&linux_vma_lock, "lkpi-vma-lock"); 2870 2871 rootoid = SYSCTL_ADD_ROOT_NODE(NULL, 2872 OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); 2873 kobject_init(&linux_class_root, &linux_class_ktype); 2874 kobject_set_name(&linux_class_root, "class"); 2875 linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), 2876 OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); 2877 kobject_init(&linux_root_device.kobj, &linux_dev_ktype); 2878 kobject_set_name(&linux_root_device.kobj, "device"); 2879 linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, 2880 SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", 2881 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device"); 2882 linux_root_device.bsddev = root_bus; 2883 linux_class_misc.name = "misc"; 2884 class_register(&linux_class_misc); 2885 INIT_LIST_HEAD(&pci_drivers); 2886 INIT_LIST_HEAD(&pci_devices); 2887 spin_lock_init(&pci_lock); 2888 mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); 2889 for (i = 0; i < VMMAP_HASH_SIZE; i++) 2890 LIST_INIT(&vmmaphead[i]); 2891 init_waitqueue_head(&linux_bit_waitq); 2892 init_waitqueue_head(&linux_var_waitq); 2893 2894 CPU_COPY(&all_cpus, &cpu_online_mask); 2895 /* 2896 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system. 2897 * CPUs are indexed from 0..(mp_maxid). The entry for cpuid 0 will only 2898 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on. 2899 * This is used by cpumask_of() (and possibly others in the future) for, 2900 * e.g., drivers to pass hints to irq_set_affinity_hint(). 2901 */ 2902 static_single_cpu_mask = kmalloc_array(mp_maxid + 1, 2903 sizeof(static_single_cpu_mask), M_WAITOK | M_ZERO); 2904 2905 /* 2906 * When the number of CPUs reach a threshold, we start to save memory 2907 * given the sets are static by overlapping those having their single 2908 * bit set at same position in a bitset word. Asymptotically, this 2909 * regular scheme is in O(n²) whereas the overlapping one is in O(n) 2910 * only with n being the maximum number of CPUs, so the gain will become 2911 * huge quite quickly. The threshold for 64-bit architectures is 128 2912 * CPUs. 2913 */ 2914 if (mp_ncpus < (2 * _BITSET_BITS)) { 2915 cpumask_t *sscm_ptr; 2916 2917 /* 2918 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) * 2919 * (_BITSET_BITS / 8)' bytes (for comparison with the 2920 * overlapping scheme). 2921 */ 2922 static_single_cpu_mask_lcs = kmalloc_array(mp_ncpus, 2923 sizeof(*static_single_cpu_mask_lcs), 2924 M_WAITOK | M_ZERO); 2925 2926 sscm_ptr = static_single_cpu_mask_lcs; 2927 CPU_FOREACH(i) { 2928 static_single_cpu_mask[i] = sscm_ptr++; 2929 CPU_SET(i, static_single_cpu_mask[i]); 2930 } 2931 } else { 2932 /* Pointer to a bitset word. */ 2933 __typeof(((cpuset_t *)NULL)->__bits[0]) *bwp; 2934 2935 /* 2936 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t' 2937 * really) with a single bit set that can be reused for all 2938 * single CPU masks by making them start at different offsets. 2939 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before 2940 * the word having its single bit set, and the same amount 2941 * after. 2942 */ 2943 static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS, 2944 (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8), 2945 M_KMALLOC, M_WAITOK | M_ZERO); 2946 2947 /* 2948 * We rely below on cpuset_t and the bitset generic 2949 * implementation assigning words in the '__bits' array in the 2950 * same order of bits (i.e., little-endian ordering, not to be 2951 * confused with machine endianness, which concerns bits in 2952 * words and other integers). This is an imperfect test, but it 2953 * will detect a change to big-endian ordering. 2954 */ 2955 _Static_assert( 2956 __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1, 2957 "Assumes a bitset implementation that is little-endian " 2958 "on its words"); 2959 2960 /* Initialize the single bit of each static span. */ 2961 bwp = (__typeof(bwp))static_single_cpu_mask_lcs + 2962 (__bitset_words(CPU_SETSIZE) - 1); 2963 for (i = 0; i < _BITSET_BITS; i++) { 2964 CPU_SET(i, (cpuset_t *)bwp); 2965 bwp += (2 * __bitset_words(CPU_SETSIZE) - 1); 2966 } 2967 2968 /* 2969 * Finally set all CPU masks to the proper word in their 2970 * relevant span. 2971 */ 2972 CPU_FOREACH(i) { 2973 bwp = (__typeof(bwp))static_single_cpu_mask_lcs; 2974 /* Find the non-zero word of the relevant span. */ 2975 bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) * 2976 (i % _BITSET_BITS) + 2977 __bitset_words(CPU_SETSIZE) - 1; 2978 /* Shift to find the CPU mask start. */ 2979 bwp -= (i / _BITSET_BITS); 2980 static_single_cpu_mask[i] = (cpuset_t *)bwp; 2981 } 2982 } 2983 2984 strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release)); 2985 } 2986 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); 2987 2988 static void 2989 linux_compat_uninit(void *arg) 2990 { 2991 linux_kobject_kfree_name(&linux_class_root); 2992 linux_kobject_kfree_name(&linux_root_device.kobj); 2993 linux_kobject_kfree_name(&linux_class_misc.kobj); 2994 2995 free(static_single_cpu_mask_lcs, M_KMALLOC); 2996 free(static_single_cpu_mask, M_KMALLOC); 2997 #if defined(__i386__) || defined(__amd64__) 2998 free(__cpu_data, M_KMALLOC); 2999 #endif 3000 3001 mtx_destroy(&vmmaplock); 3002 spin_lock_destroy(&pci_lock); 3003 rw_destroy(&linux_vma_lock); 3004 } 3005 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); 3006 3007 /* 3008 * NOTE: Linux frequently uses "unsigned long" for pointer to integer 3009 * conversion and vice versa, where in FreeBSD "uintptr_t" would be 3010 * used. Assert these types have the same size, else some parts of the 3011 * LinuxKPI may not work like expected: 3012 */ 3013 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); 3014