xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
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_warn_dump_stack = 0;
121 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN,
122     &linuxkpi_warn_dump_stack, 0,
123     "Set to enable stack traces from WARN_ON(). Clear to disable.");
124 
125 static struct timeval lkpi_net_lastlog;
126 static int lkpi_net_curpps;
127 static int lkpi_net_maxpps = 99;
128 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
129     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
130 
131 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat");
132 
133 #include <linux/rbtree.h>
134 /* Undo Linux compat changes. */
135 #undef RB_ROOT
136 #undef file
137 #undef cdev
138 #define	RB_ROOT(head)	(head)->rbh_root
139 
140 static void linux_destroy_dev(struct linux_cdev *);
141 static void linux_cdev_deref(struct linux_cdev *ldev);
142 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
143 
144 cpumask_t cpu_online_mask;
145 static cpumask_t **static_single_cpu_mask;
146 static cpumask_t *static_single_cpu_mask_lcs;
147 struct kobject linux_class_root;
148 struct device linux_root_device;
149 struct class linux_class_misc;
150 struct list_head pci_drivers;
151 struct list_head pci_devices;
152 spinlock_t pci_lock;
153 struct uts_namespace init_uts_ns;
154 
155 unsigned long linux_timer_hz_mask;
156 
157 wait_queue_head_t linux_bit_waitq;
158 wait_queue_head_t linux_var_waitq;
159 
160 int
161 panic_cmp(struct rb_node *one, struct rb_node *two)
162 {
163 	panic("no cmp");
164 }
165 
166 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
167 
168 #define	START(node)	((node)->start)
169 #define	LAST(node)	((node)->last)
170 
171 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
172     LAST,, lkpi_interval_tree)
173 
174 static void
175 linux_device_release(struct device *dev)
176 {
177 	pr_debug("linux_device_release: %s\n", dev_name(dev));
178 	kfree(dev);
179 }
180 
181 static ssize_t
182 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
183 {
184 	struct class_attribute *dattr;
185 	ssize_t error;
186 
187 	dattr = container_of(attr, struct class_attribute, attr);
188 	error = -EIO;
189 	if (dattr->show)
190 		error = dattr->show(container_of(kobj, struct class, kobj),
191 		    dattr, buf);
192 	return (error);
193 }
194 
195 static ssize_t
196 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
197     size_t count)
198 {
199 	struct class_attribute *dattr;
200 	ssize_t error;
201 
202 	dattr = container_of(attr, struct class_attribute, attr);
203 	error = -EIO;
204 	if (dattr->store)
205 		error = dattr->store(container_of(kobj, struct class, kobj),
206 		    dattr, buf, count);
207 	return (error);
208 }
209 
210 static void
211 linux_class_release(struct kobject *kobj)
212 {
213 	struct class *class;
214 
215 	class = container_of(kobj, struct class, kobj);
216 	if (class->class_release)
217 		class->class_release(class);
218 }
219 
220 static const struct sysfs_ops linux_class_sysfs = {
221 	.show  = linux_class_show,
222 	.store = linux_class_store,
223 };
224 
225 const struct kobj_type linux_class_ktype = {
226 	.release = linux_class_release,
227 	.sysfs_ops = &linux_class_sysfs
228 };
229 
230 static void
231 linux_dev_release(struct kobject *kobj)
232 {
233 	struct device *dev;
234 
235 	dev = container_of(kobj, struct device, kobj);
236 	/* This is the precedence defined by linux. */
237 	if (dev->release)
238 		dev->release(dev);
239 	else if (dev->class && dev->class->dev_release)
240 		dev->class->dev_release(dev);
241 }
242 
243 static ssize_t
244 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
245 {
246 	struct device_attribute *dattr;
247 	ssize_t error;
248 
249 	dattr = container_of(attr, struct device_attribute, attr);
250 	error = -EIO;
251 	if (dattr->show)
252 		error = dattr->show(container_of(kobj, struct device, kobj),
253 		    dattr, buf);
254 	return (error);
255 }
256 
257 static ssize_t
258 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
259     size_t count)
260 {
261 	struct device_attribute *dattr;
262 	ssize_t error;
263 
264 	dattr = container_of(attr, struct device_attribute, attr);
265 	error = -EIO;
266 	if (dattr->store)
267 		error = dattr->store(container_of(kobj, struct device, kobj),
268 		    dattr, buf, count);
269 	return (error);
270 }
271 
272 static const struct sysfs_ops linux_dev_sysfs = {
273 	.show  = linux_dev_show,
274 	.store = linux_dev_store,
275 };
276 
277 const struct kobj_type linux_dev_ktype = {
278 	.release = linux_dev_release,
279 	.sysfs_ops = &linux_dev_sysfs
280 };
281 
282 struct device *
283 device_create(struct class *class, struct device *parent, dev_t devt,
284     void *drvdata, const char *fmt, ...)
285 {
286 	struct device *dev;
287 	va_list args;
288 
289 	dev = kzalloc(sizeof(*dev), M_WAITOK);
290 	dev->parent = parent;
291 	dev->class = class;
292 	dev->devt = devt;
293 	dev->driver_data = drvdata;
294 	dev->release = linux_device_release;
295 	va_start(args, fmt);
296 	kobject_set_name_vargs(&dev->kobj, fmt, args);
297 	va_end(args);
298 	device_register(dev);
299 
300 	return (dev);
301 }
302 
303 struct device *
304 device_create_groups_vargs(struct class *class, struct device *parent,
305     dev_t devt, void *drvdata, const struct attribute_group **groups,
306     const char *fmt, va_list args)
307 {
308 	struct device *dev = NULL;
309 	int retval = -ENODEV;
310 
311 	if (class == NULL || IS_ERR(class))
312 		goto error;
313 
314 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
315 	if (!dev) {
316 		retval = -ENOMEM;
317 		goto error;
318 	}
319 
320 	dev->devt = devt;
321 	dev->class = class;
322 	dev->parent = parent;
323 	dev->groups = groups;
324 	dev->release = device_create_release;
325 	/* device_initialize() needs the class and parent to be set */
326 	device_initialize(dev);
327 	dev_set_drvdata(dev, drvdata);
328 
329 	retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
330 	if (retval)
331 		goto error;
332 
333 	retval = device_add(dev);
334 	if (retval)
335 		goto error;
336 
337 	return dev;
338 
339 error:
340 	put_device(dev);
341 	return ERR_PTR(retval);
342 }
343 
344 struct class *
345 class_create(struct module *owner, const char *name)
346 {
347 	struct class *class;
348 	int error;
349 
350 	class = kzalloc(sizeof(*class), M_WAITOK);
351 	class->owner = owner;
352 	class->name = name;
353 	class->class_release = linux_class_kfree;
354 	error = class_register(class);
355 	if (error) {
356 		kfree(class);
357 		return (NULL);
358 	}
359 
360 	return (class);
361 }
362 
363 static void
364 linux_kq_lock(void *arg)
365 {
366 	spinlock_t *s = arg;
367 
368 	spin_lock(s);
369 }
370 static void
371 linux_kq_unlock(void *arg)
372 {
373 	spinlock_t *s = arg;
374 
375 	spin_unlock(s);
376 }
377 
378 static void
379 linux_kq_assert_lock(void *arg, int what)
380 {
381 #ifdef INVARIANTS
382 	spinlock_t *s = arg;
383 
384 	if (what == LA_LOCKED)
385 		mtx_assert(s, MA_OWNED);
386 	else
387 		mtx_assert(s, MA_NOTOWNED);
388 #endif
389 }
390 
391 static void
392 linux_file_kqfilter_poll(struct linux_file *, int);
393 
394 struct linux_file *
395 linux_file_alloc(void)
396 {
397 	struct linux_file *filp;
398 
399 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
400 
401 	/* set initial refcount */
402 	filp->f_count = 1;
403 
404 	/* setup fields needed by kqueue support */
405 	spin_lock_init(&filp->f_kqlock);
406 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
407 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
408 
409 	return (filp);
410 }
411 
412 void
413 linux_file_free(struct linux_file *filp)
414 {
415 	if (filp->_file == NULL) {
416 		if (filp->f_op != NULL && filp->f_op->release != NULL)
417 			filp->f_op->release(filp->f_vnode, filp);
418 		if (filp->f_shmem != NULL)
419 			vm_object_deallocate(filp->f_shmem);
420 		kfree_rcu(filp, rcu);
421 	} else {
422 		/*
423 		 * The close method of the character device or file
424 		 * will free the linux_file structure:
425 		 */
426 		_fdrop(filp->_file, curthread);
427 	}
428 }
429 
430 struct linux_cdev *
431 cdev_alloc(void)
432 {
433 	struct linux_cdev *cdev;
434 
435 	cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
436 	kobject_init(&cdev->kobj, &linux_cdev_ktype);
437 	cdev->refs = 1;
438 	return (cdev);
439 }
440 
441 static int
442 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
443     vm_page_t *mres)
444 {
445 	struct vm_area_struct *vmap;
446 
447 	vmap = linux_cdev_handle_find(vm_obj->handle);
448 
449 	MPASS(vmap != NULL);
450 	MPASS(vmap->vm_private_data == vm_obj->handle);
451 
452 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
453 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
454 		vm_page_t page;
455 
456 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
457 			/*
458 			 * If the passed in result page is a fake
459 			 * page, update it with the new physical
460 			 * address.
461 			 */
462 			page = *mres;
463 			vm_page_updatefake(page, paddr, vm_obj->memattr);
464 		} else {
465 			/*
466 			 * Replace the passed in "mres" page with our
467 			 * own fake page and free up the all of the
468 			 * original pages.
469 			 */
470 			VM_OBJECT_WUNLOCK(vm_obj);
471 			page = vm_page_getfake(paddr, vm_obj->memattr);
472 			VM_OBJECT_WLOCK(vm_obj);
473 
474 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
475 			*mres = page;
476 		}
477 		vm_page_valid(page);
478 		return (VM_PAGER_OK);
479 	}
480 	return (VM_PAGER_FAIL);
481 }
482 
483 static int
484 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
485     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
486 {
487 	struct vm_area_struct *vmap;
488 	int err;
489 
490 	/* get VM area structure */
491 	vmap = linux_cdev_handle_find(vm_obj->handle);
492 	MPASS(vmap != NULL);
493 	MPASS(vmap->vm_private_data == vm_obj->handle);
494 
495 	VM_OBJECT_WUNLOCK(vm_obj);
496 
497 	linux_set_current(curthread);
498 
499 	down_write(&vmap->vm_mm->mmap_sem);
500 	if (unlikely(vmap->vm_ops == NULL)) {
501 		err = VM_FAULT_SIGBUS;
502 	} else {
503 		struct vm_fault vmf;
504 
505 		/* fill out VM fault structure */
506 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
507 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
508 		vmf.pgoff = 0;
509 		vmf.page = NULL;
510 		vmf.vma = vmap;
511 
512 		vmap->vm_pfn_count = 0;
513 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
514 		vmap->vm_obj = vm_obj;
515 
516 		err = vmap->vm_ops->fault(&vmf);
517 
518 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
519 			kern_yield(PRI_USER);
520 			err = vmap->vm_ops->fault(&vmf);
521 		}
522 	}
523 
524 	/* translate return code */
525 	switch (err) {
526 	case VM_FAULT_OOM:
527 		err = VM_PAGER_AGAIN;
528 		break;
529 	case VM_FAULT_SIGBUS:
530 		err = VM_PAGER_BAD;
531 		break;
532 	case VM_FAULT_NOPAGE:
533 		/*
534 		 * By contract the fault handler will return having
535 		 * busied all the pages itself. If pidx is already
536 		 * found in the object, it will simply xbusy the first
537 		 * page and return with vm_pfn_count set to 1.
538 		 */
539 		*first = vmap->vm_pfn_first;
540 		*last = *first + vmap->vm_pfn_count - 1;
541 		err = VM_PAGER_OK;
542 		break;
543 	default:
544 		err = VM_PAGER_ERROR;
545 		break;
546 	}
547 	up_write(&vmap->vm_mm->mmap_sem);
548 	VM_OBJECT_WLOCK(vm_obj);
549 	return (err);
550 }
551 
552 static struct rwlock linux_vma_lock;
553 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
554     TAILQ_HEAD_INITIALIZER(linux_vma_head);
555 
556 static void
557 linux_cdev_handle_free(struct vm_area_struct *vmap)
558 {
559 	/* Drop reference on vm_file */
560 	if (vmap->vm_file != NULL)
561 		fput(vmap->vm_file);
562 
563 	/* Drop reference on mm_struct */
564 	mmput(vmap->vm_mm);
565 
566 	kfree(vmap);
567 }
568 
569 static void
570 linux_cdev_handle_remove(struct vm_area_struct *vmap)
571 {
572 	rw_wlock(&linux_vma_lock);
573 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
574 	rw_wunlock(&linux_vma_lock);
575 }
576 
577 static struct vm_area_struct *
578 linux_cdev_handle_find(void *handle)
579 {
580 	struct vm_area_struct *vmap;
581 
582 	rw_rlock(&linux_vma_lock);
583 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
584 		if (vmap->vm_private_data == handle)
585 			break;
586 	}
587 	rw_runlock(&linux_vma_lock);
588 	return (vmap);
589 }
590 
591 static int
592 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
593 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
594 {
595 
596 	MPASS(linux_cdev_handle_find(handle) != NULL);
597 	*color = 0;
598 	return (0);
599 }
600 
601 static void
602 linux_cdev_pager_dtor(void *handle)
603 {
604 	const struct vm_operations_struct *vm_ops;
605 	struct vm_area_struct *vmap;
606 
607 	vmap = linux_cdev_handle_find(handle);
608 	MPASS(vmap != NULL);
609 
610 	/*
611 	 * Remove handle before calling close operation to prevent
612 	 * other threads from reusing the handle pointer.
613 	 */
614 	linux_cdev_handle_remove(vmap);
615 
616 	down_write(&vmap->vm_mm->mmap_sem);
617 	vm_ops = vmap->vm_ops;
618 	if (likely(vm_ops != NULL))
619 		vm_ops->close(vmap);
620 	up_write(&vmap->vm_mm->mmap_sem);
621 
622 	linux_cdev_handle_free(vmap);
623 }
624 
625 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
626   {
627 	/* OBJT_MGTDEVICE */
628 	.cdev_pg_populate	= linux_cdev_pager_populate,
629 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
630 	.cdev_pg_dtor	= linux_cdev_pager_dtor
631   },
632   {
633 	/* OBJT_DEVICE */
634 	.cdev_pg_fault	= linux_cdev_pager_fault,
635 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
636 	.cdev_pg_dtor	= linux_cdev_pager_dtor
637   },
638 };
639 
640 int
641 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
642     unsigned long size)
643 {
644 	vm_object_t obj;
645 	vm_page_t m;
646 
647 	obj = vma->vm_obj;
648 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
649 		return (-ENOTSUP);
650 	VM_OBJECT_RLOCK(obj);
651 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
652 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
653 	    m = TAILQ_NEXT(m, listq))
654 		pmap_remove_all(m);
655 	VM_OBJECT_RUNLOCK(obj);
656 	return (0);
657 }
658 
659 void
660 vma_set_file(struct vm_area_struct *vma, struct linux_file *file)
661 {
662 	struct linux_file *tmp;
663 
664 	/* Changing an anonymous vma with this is illegal */
665 	get_file(file);
666 	tmp = vma->vm_file;
667 	vma->vm_file = file;
668 	fput(tmp);
669 }
670 
671 static struct file_operations dummy_ldev_ops = {
672 	/* XXXKIB */
673 };
674 
675 static struct linux_cdev dummy_ldev = {
676 	.ops = &dummy_ldev_ops,
677 };
678 
679 #define	LDEV_SI_DTR	0x0001
680 #define	LDEV_SI_REF	0x0002
681 
682 static void
683 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
684     struct linux_cdev **dev)
685 {
686 	struct linux_cdev *ldev;
687 	u_int siref;
688 
689 	ldev = filp->f_cdev;
690 	*fop = filp->f_op;
691 	if (ldev != NULL) {
692 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
693 			refcount_acquire(&ldev->refs);
694 		} else {
695 			for (siref = ldev->siref;;) {
696 				if ((siref & LDEV_SI_DTR) != 0) {
697 					ldev = &dummy_ldev;
698 					*fop = ldev->ops;
699 					siref = ldev->siref;
700 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
701 				} else if (atomic_fcmpset_int(&ldev->siref,
702 				    &siref, siref + LDEV_SI_REF)) {
703 					break;
704 				}
705 			}
706 		}
707 	}
708 	*dev = ldev;
709 }
710 
711 static void
712 linux_drop_fop(struct linux_cdev *ldev)
713 {
714 
715 	if (ldev == NULL)
716 		return;
717 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
718 		linux_cdev_deref(ldev);
719 	} else {
720 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
721 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
722 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
723 	}
724 }
725 
726 #define	OPW(fp,td,code) ({			\
727 	struct file *__fpop;			\
728 	__typeof(code) __retval;		\
729 						\
730 	__fpop = (td)->td_fpop;			\
731 	(td)->td_fpop = (fp);			\
732 	__retval = (code);			\
733 	(td)->td_fpop = __fpop;			\
734 	__retval;				\
735 })
736 
737 static int
738 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
739     struct file *file)
740 {
741 	struct linux_cdev *ldev;
742 	struct linux_file *filp;
743 	const struct file_operations *fop;
744 	int error;
745 
746 	ldev = dev->si_drv1;
747 
748 	filp = linux_file_alloc();
749 	filp->f_dentry = &filp->f_dentry_store;
750 	filp->f_op = ldev->ops;
751 	filp->f_mode = file->f_flag;
752 	filp->f_flags = file->f_flag;
753 	filp->f_vnode = file->f_vnode;
754 	filp->_file = file;
755 	refcount_acquire(&ldev->refs);
756 	filp->f_cdev = ldev;
757 
758 	linux_set_current(td);
759 	linux_get_fop(filp, &fop, &ldev);
760 
761 	if (fop->open != NULL) {
762 		error = -fop->open(file->f_vnode, filp);
763 		if (error != 0) {
764 			linux_drop_fop(ldev);
765 			linux_cdev_deref(filp->f_cdev);
766 			kfree(filp);
767 			return (error);
768 		}
769 	}
770 
771 	/* hold on to the vnode - used for fstat() */
772 	vhold(filp->f_vnode);
773 
774 	/* release the file from devfs */
775 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
776 	linux_drop_fop(ldev);
777 	return (ENXIO);
778 }
779 
780 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
781 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
782 
783 static inline int
784 linux_remap_address(void **uaddr, size_t len)
785 {
786 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
787 
788 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
789 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
790 		struct task_struct *pts = current;
791 		if (pts == NULL) {
792 			*uaddr = NULL;
793 			return (1);
794 		}
795 
796 		/* compute data offset */
797 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
798 
799 		/* check that length is within bounds */
800 		if ((len > IOCPARM_MAX) ||
801 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
802 			*uaddr = NULL;
803 			return (1);
804 		}
805 
806 		/* re-add kernel buffer address */
807 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
808 
809 		/* update address location */
810 		*uaddr = (void *)uaddr_val;
811 		return (1);
812 	}
813 	return (0);
814 }
815 
816 int
817 linux_copyin(const void *uaddr, void *kaddr, size_t len)
818 {
819 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
820 		if (uaddr == NULL)
821 			return (-EFAULT);
822 		memcpy(kaddr, uaddr, len);
823 		return (0);
824 	}
825 	return (-copyin(uaddr, kaddr, len));
826 }
827 
828 int
829 linux_copyout(const void *kaddr, void *uaddr, size_t len)
830 {
831 	if (linux_remap_address(&uaddr, len)) {
832 		if (uaddr == NULL)
833 			return (-EFAULT);
834 		memcpy(uaddr, kaddr, len);
835 		return (0);
836 	}
837 	return (-copyout(kaddr, uaddr, len));
838 }
839 
840 size_t
841 linux_clear_user(void *_uaddr, size_t _len)
842 {
843 	uint8_t *uaddr = _uaddr;
844 	size_t len = _len;
845 
846 	/* make sure uaddr is aligned before going into the fast loop */
847 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
848 		if (subyte(uaddr, 0))
849 			return (_len);
850 		uaddr++;
851 		len--;
852 	}
853 
854 	/* zero 8 bytes at a time */
855 	while (len > 7) {
856 #ifdef __LP64__
857 		if (suword64(uaddr, 0))
858 			return (_len);
859 #else
860 		if (suword32(uaddr, 0))
861 			return (_len);
862 		if (suword32(uaddr + 4, 0))
863 			return (_len);
864 #endif
865 		uaddr += 8;
866 		len -= 8;
867 	}
868 
869 	/* zero fill end, if any */
870 	while (len > 0) {
871 		if (subyte(uaddr, 0))
872 			return (_len);
873 		uaddr++;
874 		len--;
875 	}
876 	return (0);
877 }
878 
879 int
880 linux_access_ok(const void *uaddr, size_t len)
881 {
882 	uintptr_t saddr;
883 	uintptr_t eaddr;
884 
885 	/* get start and end address */
886 	saddr = (uintptr_t)uaddr;
887 	eaddr = (uintptr_t)uaddr + len;
888 
889 	/* verify addresses are valid for userspace */
890 	return ((saddr == eaddr) ||
891 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
892 }
893 
894 /*
895  * This function should return either EINTR or ERESTART depending on
896  * the signal type sent to this thread:
897  */
898 static int
899 linux_get_error(struct task_struct *task, int error)
900 {
901 	/* check for signal type interrupt code */
902 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
903 		error = -linux_schedule_get_interrupt_value(task);
904 		if (error == 0)
905 			error = EINTR;
906 	}
907 	return (error);
908 }
909 
910 static int
911 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
912     const struct file_operations *fop, u_long cmd, caddr_t data,
913     struct thread *td)
914 {
915 	struct task_struct *task = current;
916 	unsigned size;
917 	int error;
918 
919 	size = IOCPARM_LEN(cmd);
920 	/* refer to logic in sys_ioctl() */
921 	if (size > 0) {
922 		/*
923 		 * Setup hint for linux_copyin() and linux_copyout().
924 		 *
925 		 * Background: Linux code expects a user-space address
926 		 * while FreeBSD supplies a kernel-space address.
927 		 */
928 		task->bsd_ioctl_data = data;
929 		task->bsd_ioctl_len = size;
930 		data = (void *)LINUX_IOCTL_MIN_PTR;
931 	} else {
932 		/* fetch user-space pointer */
933 		data = *(void **)data;
934 	}
935 #ifdef COMPAT_FREEBSD32
936 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
937 		/* try the compat IOCTL handler first */
938 		if (fop->compat_ioctl != NULL) {
939 			error = -OPW(fp, td, fop->compat_ioctl(filp,
940 			    cmd, (u_long)data));
941 		} else {
942 			error = ENOTTY;
943 		}
944 
945 		/* fallback to the regular IOCTL handler, if any */
946 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
947 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
948 			    cmd, (u_long)data));
949 		}
950 	} else
951 #endif
952 	{
953 		if (fop->unlocked_ioctl != NULL) {
954 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
955 			    cmd, (u_long)data));
956 		} else {
957 			error = ENOTTY;
958 		}
959 	}
960 	if (size > 0) {
961 		task->bsd_ioctl_data = NULL;
962 		task->bsd_ioctl_len = 0;
963 	}
964 
965 	if (error == EWOULDBLOCK) {
966 		/* update kqfilter status, if any */
967 		linux_file_kqfilter_poll(filp,
968 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
969 	} else {
970 		error = linux_get_error(task, error);
971 	}
972 	return (error);
973 }
974 
975 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
976 
977 /*
978  * This function atomically updates the poll wakeup state and returns
979  * the previous state at the time of update.
980  */
981 static uint8_t
982 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
983 {
984 	int c, old;
985 
986 	c = v->counter;
987 
988 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
989 		c = old;
990 
991 	return (c);
992 }
993 
994 static int
995 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
996 {
997 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
998 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
999 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1000 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1001 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1002 	};
1003 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1004 
1005 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1006 	case LINUX_FWQ_STATE_QUEUED:
1007 		linux_poll_wakeup(filp);
1008 		return (1);
1009 	default:
1010 		return (0);
1011 	}
1012 }
1013 
1014 void
1015 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1016 {
1017 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1018 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1019 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1020 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1021 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1022 	};
1023 
1024 	/* check if we are called inside the select system call */
1025 	if (p == LINUX_POLL_TABLE_NORMAL)
1026 		selrecord(curthread, &filp->f_selinfo);
1027 
1028 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1029 	case LINUX_FWQ_STATE_INIT:
1030 		/* NOTE: file handles can only belong to one wait-queue */
1031 		filp->f_wait_queue.wqh = wqh;
1032 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1033 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1034 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1035 		break;
1036 	default:
1037 		break;
1038 	}
1039 }
1040 
1041 static void
1042 linux_poll_wait_dequeue(struct linux_file *filp)
1043 {
1044 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1045 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1046 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1047 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1048 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1049 	};
1050 
1051 	seldrain(&filp->f_selinfo);
1052 
1053 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1054 	case LINUX_FWQ_STATE_NOT_READY:
1055 	case LINUX_FWQ_STATE_QUEUED:
1056 	case LINUX_FWQ_STATE_READY:
1057 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1058 		break;
1059 	default:
1060 		break;
1061 	}
1062 }
1063 
1064 void
1065 linux_poll_wakeup(struct linux_file *filp)
1066 {
1067 	/* this function should be NULL-safe */
1068 	if (filp == NULL)
1069 		return;
1070 
1071 	selwakeup(&filp->f_selinfo);
1072 
1073 	spin_lock(&filp->f_kqlock);
1074 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1075 	    LINUX_KQ_FLAG_NEED_WRITE;
1076 
1077 	/* make sure the "knote" gets woken up */
1078 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1079 	spin_unlock(&filp->f_kqlock);
1080 }
1081 
1082 static void
1083 linux_file_kqfilter_detach(struct knote *kn)
1084 {
1085 	struct linux_file *filp = kn->kn_hook;
1086 
1087 	spin_lock(&filp->f_kqlock);
1088 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1089 	spin_unlock(&filp->f_kqlock);
1090 }
1091 
1092 static int
1093 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1094 {
1095 	struct linux_file *filp = kn->kn_hook;
1096 
1097 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1098 
1099 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1100 }
1101 
1102 static int
1103 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1104 {
1105 	struct linux_file *filp = kn->kn_hook;
1106 
1107 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1108 
1109 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1110 }
1111 
1112 static struct filterops linux_dev_kqfiltops_read = {
1113 	.f_isfd = 1,
1114 	.f_detach = linux_file_kqfilter_detach,
1115 	.f_event = linux_file_kqfilter_read_event,
1116 };
1117 
1118 static struct filterops linux_dev_kqfiltops_write = {
1119 	.f_isfd = 1,
1120 	.f_detach = linux_file_kqfilter_detach,
1121 	.f_event = linux_file_kqfilter_write_event,
1122 };
1123 
1124 static void
1125 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1126 {
1127 	struct thread *td;
1128 	const struct file_operations *fop;
1129 	struct linux_cdev *ldev;
1130 	int temp;
1131 
1132 	if ((filp->f_kqflags & kqflags) == 0)
1133 		return;
1134 
1135 	td = curthread;
1136 
1137 	linux_get_fop(filp, &fop, &ldev);
1138 	/* get the latest polling state */
1139 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1140 	linux_drop_fop(ldev);
1141 
1142 	spin_lock(&filp->f_kqlock);
1143 	/* clear kqflags */
1144 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1145 	    LINUX_KQ_FLAG_NEED_WRITE);
1146 	/* update kqflags */
1147 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1148 		if ((temp & POLLIN) != 0)
1149 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1150 		if ((temp & POLLOUT) != 0)
1151 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1152 
1153 		/* make sure the "knote" gets woken up */
1154 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1155 	}
1156 	spin_unlock(&filp->f_kqlock);
1157 }
1158 
1159 static int
1160 linux_file_kqfilter(struct file *file, struct knote *kn)
1161 {
1162 	struct linux_file *filp;
1163 	struct thread *td;
1164 	int error;
1165 
1166 	td = curthread;
1167 	filp = (struct linux_file *)file->f_data;
1168 	filp->f_flags = file->f_flag;
1169 	if (filp->f_op->poll == NULL)
1170 		return (EINVAL);
1171 
1172 	spin_lock(&filp->f_kqlock);
1173 	switch (kn->kn_filter) {
1174 	case EVFILT_READ:
1175 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1176 		kn->kn_fop = &linux_dev_kqfiltops_read;
1177 		kn->kn_hook = filp;
1178 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1179 		error = 0;
1180 		break;
1181 	case EVFILT_WRITE:
1182 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1183 		kn->kn_fop = &linux_dev_kqfiltops_write;
1184 		kn->kn_hook = filp;
1185 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1186 		error = 0;
1187 		break;
1188 	default:
1189 		error = EINVAL;
1190 		break;
1191 	}
1192 	spin_unlock(&filp->f_kqlock);
1193 
1194 	if (error == 0) {
1195 		linux_set_current(td);
1196 
1197 		/* update kqfilter status, if any */
1198 		linux_file_kqfilter_poll(filp,
1199 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1200 	}
1201 	return (error);
1202 }
1203 
1204 static int
1205 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1206     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1207     int nprot, bool is_shared, struct thread *td)
1208 {
1209 	struct task_struct *task;
1210 	struct vm_area_struct *vmap;
1211 	struct mm_struct *mm;
1212 	struct linux_file *filp;
1213 	vm_memattr_t attr;
1214 	int error;
1215 
1216 	filp = (struct linux_file *)fp->f_data;
1217 	filp->f_flags = fp->f_flag;
1218 
1219 	if (fop->mmap == NULL)
1220 		return (EOPNOTSUPP);
1221 
1222 	linux_set_current(td);
1223 
1224 	/*
1225 	 * The same VM object might be shared by multiple processes
1226 	 * and the mm_struct is usually freed when a process exits.
1227 	 *
1228 	 * The atomic reference below makes sure the mm_struct is
1229 	 * available as long as the vmap is in the linux_vma_head.
1230 	 */
1231 	task = current;
1232 	mm = task->mm;
1233 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1234 		return (EINVAL);
1235 
1236 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1237 	vmap->vm_start = 0;
1238 	vmap->vm_end = size;
1239 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1240 	vmap->vm_pfn = 0;
1241 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1242 	if (is_shared)
1243 		vmap->vm_flags |= VM_SHARED;
1244 	vmap->vm_ops = NULL;
1245 	vmap->vm_file = get_file(filp);
1246 	vmap->vm_mm = mm;
1247 
1248 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1249 		error = linux_get_error(task, EINTR);
1250 	} else {
1251 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1252 		error = linux_get_error(task, error);
1253 		up_write(&vmap->vm_mm->mmap_sem);
1254 	}
1255 
1256 	if (error != 0) {
1257 		linux_cdev_handle_free(vmap);
1258 		return (error);
1259 	}
1260 
1261 	attr = pgprot2cachemode(vmap->vm_page_prot);
1262 
1263 	if (vmap->vm_ops != NULL) {
1264 		struct vm_area_struct *ptr;
1265 		void *vm_private_data;
1266 		bool vm_no_fault;
1267 
1268 		if (vmap->vm_ops->open == NULL ||
1269 		    vmap->vm_ops->close == NULL ||
1270 		    vmap->vm_private_data == NULL) {
1271 			/* free allocated VM area struct */
1272 			linux_cdev_handle_free(vmap);
1273 			return (EINVAL);
1274 		}
1275 
1276 		vm_private_data = vmap->vm_private_data;
1277 
1278 		rw_wlock(&linux_vma_lock);
1279 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1280 			if (ptr->vm_private_data == vm_private_data)
1281 				break;
1282 		}
1283 		/* check if there is an existing VM area struct */
1284 		if (ptr != NULL) {
1285 			/* check if the VM area structure is invalid */
1286 			if (ptr->vm_ops == NULL ||
1287 			    ptr->vm_ops->open == NULL ||
1288 			    ptr->vm_ops->close == NULL) {
1289 				error = ESTALE;
1290 				vm_no_fault = 1;
1291 			} else {
1292 				error = EEXIST;
1293 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1294 			}
1295 		} else {
1296 			/* insert VM area structure into list */
1297 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1298 			error = 0;
1299 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1300 		}
1301 		rw_wunlock(&linux_vma_lock);
1302 
1303 		if (error != 0) {
1304 			/* free allocated VM area struct */
1305 			linux_cdev_handle_free(vmap);
1306 			/* check for stale VM area struct */
1307 			if (error != EEXIST)
1308 				return (error);
1309 		}
1310 
1311 		/* check if there is no fault handler */
1312 		if (vm_no_fault) {
1313 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1314 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1315 			    td->td_ucred);
1316 		} else {
1317 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1318 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1319 			    td->td_ucred);
1320 		}
1321 
1322 		/* check if allocating the VM object failed */
1323 		if (*object == NULL) {
1324 			if (error == 0) {
1325 				/* remove VM area struct from list */
1326 				linux_cdev_handle_remove(vmap);
1327 				/* free allocated VM area struct */
1328 				linux_cdev_handle_free(vmap);
1329 			}
1330 			return (EINVAL);
1331 		}
1332 	} else {
1333 		struct sglist *sg;
1334 
1335 		sg = sglist_alloc(1, M_WAITOK);
1336 		sglist_append_phys(sg,
1337 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1338 
1339 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1340 		    nprot, 0, td->td_ucred);
1341 
1342 		linux_cdev_handle_free(vmap);
1343 
1344 		if (*object == NULL) {
1345 			sglist_free(sg);
1346 			return (EINVAL);
1347 		}
1348 	}
1349 
1350 	if (attr != VM_MEMATTR_DEFAULT) {
1351 		VM_OBJECT_WLOCK(*object);
1352 		vm_object_set_memattr(*object, attr);
1353 		VM_OBJECT_WUNLOCK(*object);
1354 	}
1355 	*offset = 0;
1356 	return (0);
1357 }
1358 
1359 struct cdevsw linuxcdevsw = {
1360 	.d_version = D_VERSION,
1361 	.d_fdopen = linux_dev_fdopen,
1362 	.d_name = "lkpidev",
1363 };
1364 
1365 static int
1366 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1367     int flags, struct thread *td)
1368 {
1369 	struct linux_file *filp;
1370 	const struct file_operations *fop;
1371 	struct linux_cdev *ldev;
1372 	ssize_t bytes;
1373 	int error;
1374 
1375 	error = 0;
1376 	filp = (struct linux_file *)file->f_data;
1377 	filp->f_flags = file->f_flag;
1378 	/* XXX no support for I/O vectors currently */
1379 	if (uio->uio_iovcnt != 1)
1380 		return (EOPNOTSUPP);
1381 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1382 		return (EINVAL);
1383 	linux_set_current(td);
1384 	linux_get_fop(filp, &fop, &ldev);
1385 	if (fop->read != NULL) {
1386 		bytes = OPW(file, td, fop->read(filp,
1387 		    uio->uio_iov->iov_base,
1388 		    uio->uio_iov->iov_len, &uio->uio_offset));
1389 		if (bytes >= 0) {
1390 			uio->uio_iov->iov_base =
1391 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1392 			uio->uio_iov->iov_len -= bytes;
1393 			uio->uio_resid -= bytes;
1394 		} else {
1395 			error = linux_get_error(current, -bytes);
1396 		}
1397 	} else
1398 		error = ENXIO;
1399 
1400 	/* update kqfilter status, if any */
1401 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1402 	linux_drop_fop(ldev);
1403 
1404 	return (error);
1405 }
1406 
1407 static int
1408 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1409     int flags, struct thread *td)
1410 {
1411 	struct linux_file *filp;
1412 	const struct file_operations *fop;
1413 	struct linux_cdev *ldev;
1414 	ssize_t bytes;
1415 	int error;
1416 
1417 	filp = (struct linux_file *)file->f_data;
1418 	filp->f_flags = file->f_flag;
1419 	/* XXX no support for I/O vectors currently */
1420 	if (uio->uio_iovcnt != 1)
1421 		return (EOPNOTSUPP);
1422 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1423 		return (EINVAL);
1424 	linux_set_current(td);
1425 	linux_get_fop(filp, &fop, &ldev);
1426 	if (fop->write != NULL) {
1427 		bytes = OPW(file, td, fop->write(filp,
1428 		    uio->uio_iov->iov_base,
1429 		    uio->uio_iov->iov_len, &uio->uio_offset));
1430 		if (bytes >= 0) {
1431 			uio->uio_iov->iov_base =
1432 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1433 			uio->uio_iov->iov_len -= bytes;
1434 			uio->uio_resid -= bytes;
1435 			error = 0;
1436 		} else {
1437 			error = linux_get_error(current, -bytes);
1438 		}
1439 	} else
1440 		error = ENXIO;
1441 
1442 	/* update kqfilter status, if any */
1443 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1444 
1445 	linux_drop_fop(ldev);
1446 
1447 	return (error);
1448 }
1449 
1450 static int
1451 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1452     struct thread *td)
1453 {
1454 	struct linux_file *filp;
1455 	const struct file_operations *fop;
1456 	struct linux_cdev *ldev;
1457 	int revents;
1458 
1459 	filp = (struct linux_file *)file->f_data;
1460 	filp->f_flags = file->f_flag;
1461 	linux_set_current(td);
1462 	linux_get_fop(filp, &fop, &ldev);
1463 	if (fop->poll != NULL) {
1464 		revents = OPW(file, td, fop->poll(filp,
1465 		    LINUX_POLL_TABLE_NORMAL)) & events;
1466 	} else {
1467 		revents = 0;
1468 	}
1469 	linux_drop_fop(ldev);
1470 	return (revents);
1471 }
1472 
1473 static int
1474 linux_file_close(struct file *file, struct thread *td)
1475 {
1476 	struct linux_file *filp;
1477 	int (*release)(struct inode *, struct linux_file *);
1478 	const struct file_operations *fop;
1479 	struct linux_cdev *ldev;
1480 	int error;
1481 
1482 	filp = (struct linux_file *)file->f_data;
1483 
1484 	KASSERT(file_count(filp) == 0,
1485 	    ("File refcount(%d) is not zero", file_count(filp)));
1486 
1487 	if (td == NULL)
1488 		td = curthread;
1489 
1490 	error = 0;
1491 	filp->f_flags = file->f_flag;
1492 	linux_set_current(td);
1493 	linux_poll_wait_dequeue(filp);
1494 	linux_get_fop(filp, &fop, &ldev);
1495 	/*
1496 	 * Always use the real release function, if any, to avoid
1497 	 * leaking device resources:
1498 	 */
1499 	release = filp->f_op->release;
1500 	if (release != NULL)
1501 		error = -OPW(file, td, release(filp->f_vnode, filp));
1502 	funsetown(&filp->f_sigio);
1503 	if (filp->f_vnode != NULL)
1504 		vdrop(filp->f_vnode);
1505 	linux_drop_fop(ldev);
1506 	ldev = filp->f_cdev;
1507 	if (ldev != NULL)
1508 		linux_cdev_deref(ldev);
1509 	linux_synchronize_rcu(RCU_TYPE_REGULAR);
1510 	kfree(filp);
1511 
1512 	return (error);
1513 }
1514 
1515 static int
1516 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1517     struct thread *td)
1518 {
1519 	struct linux_file *filp;
1520 	const struct file_operations *fop;
1521 	struct linux_cdev *ldev;
1522 	struct fiodgname_arg *fgn;
1523 	const char *p;
1524 	int error, i;
1525 
1526 	error = 0;
1527 	filp = (struct linux_file *)fp->f_data;
1528 	filp->f_flags = fp->f_flag;
1529 	linux_get_fop(filp, &fop, &ldev);
1530 
1531 	linux_set_current(td);
1532 	switch (cmd) {
1533 	case FIONBIO:
1534 		break;
1535 	case FIOASYNC:
1536 		if (fop->fasync == NULL)
1537 			break;
1538 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1539 		break;
1540 	case FIOSETOWN:
1541 		error = fsetown(*(int *)data, &filp->f_sigio);
1542 		if (error == 0) {
1543 			if (fop->fasync == NULL)
1544 				break;
1545 			error = -OPW(fp, td, fop->fasync(0, filp,
1546 			    fp->f_flag & FASYNC));
1547 		}
1548 		break;
1549 	case FIOGETOWN:
1550 		*(int *)data = fgetown(&filp->f_sigio);
1551 		break;
1552 	case FIODGNAME:
1553 #ifdef	COMPAT_FREEBSD32
1554 	case FIODGNAME_32:
1555 #endif
1556 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1557 			error = ENXIO;
1558 			break;
1559 		}
1560 		fgn = data;
1561 		p = devtoname(filp->f_cdev->cdev);
1562 		i = strlen(p) + 1;
1563 		if (i > fgn->len) {
1564 			error = EINVAL;
1565 			break;
1566 		}
1567 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1568 		break;
1569 	default:
1570 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1571 		break;
1572 	}
1573 	linux_drop_fop(ldev);
1574 	return (error);
1575 }
1576 
1577 static int
1578 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1579     vm_prot_t maxprot, int flags, struct file *fp,
1580     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1581 {
1582 	/*
1583 	 * Character devices do not provide private mappings
1584 	 * of any kind:
1585 	 */
1586 	if ((maxprot & VM_PROT_WRITE) == 0 &&
1587 	    (prot & VM_PROT_WRITE) != 0)
1588 		return (EACCES);
1589 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1590 		return (EINVAL);
1591 
1592 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1593 	    (int)prot, (flags & MAP_SHARED) ? true : false, td));
1594 }
1595 
1596 static int
1597 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1598     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1599     struct thread *td)
1600 {
1601 	struct linux_file *filp;
1602 	const struct file_operations *fop;
1603 	struct linux_cdev *ldev;
1604 	struct mount *mp;
1605 	struct vnode *vp;
1606 	vm_object_t object;
1607 	vm_prot_t maxprot;
1608 	int error;
1609 
1610 	filp = (struct linux_file *)fp->f_data;
1611 
1612 	vp = filp->f_vnode;
1613 	if (vp == NULL)
1614 		return (EOPNOTSUPP);
1615 
1616 	/*
1617 	 * Ensure that file and memory protections are
1618 	 * compatible.
1619 	 */
1620 	mp = vp->v_mount;
1621 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1622 		maxprot = VM_PROT_NONE;
1623 		if ((prot & VM_PROT_EXECUTE) != 0)
1624 			return (EACCES);
1625 	} else
1626 		maxprot = VM_PROT_EXECUTE;
1627 	if ((fp->f_flag & FREAD) != 0)
1628 		maxprot |= VM_PROT_READ;
1629 	else if ((prot & VM_PROT_READ) != 0)
1630 		return (EACCES);
1631 
1632 	/*
1633 	 * If we are sharing potential changes via MAP_SHARED and we
1634 	 * are trying to get write permission although we opened it
1635 	 * without asking for it, bail out.
1636 	 *
1637 	 * Note that most character devices always share mappings.
1638 	 *
1639 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1640 	 * requests rather than doing it here.
1641 	 */
1642 	if ((flags & MAP_SHARED) != 0) {
1643 		if ((fp->f_flag & FWRITE) != 0)
1644 			maxprot |= VM_PROT_WRITE;
1645 		else if ((prot & VM_PROT_WRITE) != 0)
1646 			return (EACCES);
1647 	}
1648 	maxprot &= cap_maxprot;
1649 
1650 	linux_get_fop(filp, &fop, &ldev);
1651 	error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1652 	    &foff, fop, &object);
1653 	if (error != 0)
1654 		goto out;
1655 
1656 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1657 	    foff, FALSE, td);
1658 	if (error != 0)
1659 		vm_object_deallocate(object);
1660 out:
1661 	linux_drop_fop(ldev);
1662 	return (error);
1663 }
1664 
1665 static int
1666 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1667 {
1668 	struct linux_file *filp;
1669 	struct vnode *vp;
1670 	int error;
1671 
1672 	filp = (struct linux_file *)fp->f_data;
1673 	if (filp->f_vnode == NULL)
1674 		return (EOPNOTSUPP);
1675 
1676 	vp = filp->f_vnode;
1677 
1678 	vn_lock(vp, LK_SHARED | LK_RETRY);
1679 	error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1680 	VOP_UNLOCK(vp);
1681 
1682 	return (error);
1683 }
1684 
1685 static int
1686 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1687     struct filedesc *fdp)
1688 {
1689 	struct linux_file *filp;
1690 	struct vnode *vp;
1691 	int error;
1692 
1693 	filp = fp->f_data;
1694 	vp = filp->f_vnode;
1695 	if (vp == NULL) {
1696 		error = 0;
1697 		kif->kf_type = KF_TYPE_DEV;
1698 	} else {
1699 		vref(vp);
1700 		FILEDESC_SUNLOCK(fdp);
1701 		error = vn_fill_kinfo_vnode(vp, kif);
1702 		vrele(vp);
1703 		kif->kf_type = KF_TYPE_VNODE;
1704 		FILEDESC_SLOCK(fdp);
1705 	}
1706 	return (error);
1707 }
1708 
1709 unsigned int
1710 linux_iminor(struct inode *inode)
1711 {
1712 	struct linux_cdev *ldev;
1713 
1714 	if (inode == NULL || inode->v_rdev == NULL ||
1715 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1716 		return (-1U);
1717 	ldev = inode->v_rdev->si_drv1;
1718 	if (ldev == NULL)
1719 		return (-1U);
1720 
1721 	return (minor(ldev->dev));
1722 }
1723 
1724 static int
1725 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td)
1726 {
1727 	struct linux_file *filp1, *filp2;
1728 
1729 	if (fp2->f_type != DTYPE_DEV)
1730 		return (3);
1731 
1732 	filp1 = fp1->f_data;
1733 	filp2 = fp2->f_data;
1734 	return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev));
1735 }
1736 
1737 struct fileops linuxfileops = {
1738 	.fo_read = linux_file_read,
1739 	.fo_write = linux_file_write,
1740 	.fo_truncate = invfo_truncate,
1741 	.fo_kqfilter = linux_file_kqfilter,
1742 	.fo_stat = linux_file_stat,
1743 	.fo_fill_kinfo = linux_file_fill_kinfo,
1744 	.fo_poll = linux_file_poll,
1745 	.fo_close = linux_file_close,
1746 	.fo_ioctl = linux_file_ioctl,
1747 	.fo_mmap = linux_file_mmap,
1748 	.fo_chmod = invfo_chmod,
1749 	.fo_chown = invfo_chown,
1750 	.fo_sendfile = invfo_sendfile,
1751 	.fo_cmp = linux_file_kcmp,
1752 	.fo_flags = DFLAG_PASSABLE,
1753 };
1754 
1755 /*
1756  * Hash of vmmap addresses.  This is infrequently accessed and does not
1757  * need to be particularly large.  This is done because we must store the
1758  * caller's idea of the map size to properly unmap.
1759  */
1760 struct vmmap {
1761 	LIST_ENTRY(vmmap)	vm_next;
1762 	void 			*vm_addr;
1763 	unsigned long		vm_size;
1764 };
1765 
1766 struct vmmaphd {
1767 	struct vmmap *lh_first;
1768 };
1769 #define	VMMAP_HASH_SIZE	64
1770 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1771 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1772 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1773 static struct mtx vmmaplock;
1774 
1775 static void
1776 vmmap_add(void *addr, unsigned long size)
1777 {
1778 	struct vmmap *vmmap;
1779 
1780 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1781 	mtx_lock(&vmmaplock);
1782 	vmmap->vm_size = size;
1783 	vmmap->vm_addr = addr;
1784 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1785 	mtx_unlock(&vmmaplock);
1786 }
1787 
1788 static struct vmmap *
1789 vmmap_remove(void *addr)
1790 {
1791 	struct vmmap *vmmap;
1792 
1793 	mtx_lock(&vmmaplock);
1794 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1795 		if (vmmap->vm_addr == addr)
1796 			break;
1797 	if (vmmap)
1798 		LIST_REMOVE(vmmap, vm_next);
1799 	mtx_unlock(&vmmaplock);
1800 
1801 	return (vmmap);
1802 }
1803 
1804 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1805 void *
1806 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1807 {
1808 	void *addr;
1809 
1810 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1811 	if (addr == NULL)
1812 		return (NULL);
1813 	vmmap_add(addr, size);
1814 
1815 	return (addr);
1816 }
1817 #endif
1818 
1819 void
1820 iounmap(void *addr)
1821 {
1822 	struct vmmap *vmmap;
1823 
1824 	vmmap = vmmap_remove(addr);
1825 	if (vmmap == NULL)
1826 		return;
1827 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1828 	pmap_unmapdev(addr, vmmap->vm_size);
1829 #endif
1830 	kfree(vmmap);
1831 }
1832 
1833 void *
1834 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1835 {
1836 	vm_offset_t off;
1837 	size_t size;
1838 
1839 	size = count * PAGE_SIZE;
1840 	off = kva_alloc(size);
1841 	if (off == 0)
1842 		return (NULL);
1843 	vmmap_add((void *)off, size);
1844 	pmap_qenter(off, pages, count);
1845 
1846 	return ((void *)off);
1847 }
1848 
1849 void
1850 vunmap(void *addr)
1851 {
1852 	struct vmmap *vmmap;
1853 
1854 	vmmap = vmmap_remove(addr);
1855 	if (vmmap == NULL)
1856 		return;
1857 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1858 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1859 	kfree(vmmap);
1860 }
1861 
1862 static char *
1863 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1864 {
1865 	unsigned int len;
1866 	char *p;
1867 	va_list aq;
1868 
1869 	va_copy(aq, ap);
1870 	len = vsnprintf(NULL, 0, fmt, aq);
1871 	va_end(aq);
1872 
1873 	if (dev != NULL)
1874 		p = devm_kmalloc(dev, len + 1, gfp);
1875 	else
1876 		p = kmalloc(len + 1, gfp);
1877 	if (p != NULL)
1878 		vsnprintf(p, len + 1, fmt, ap);
1879 
1880 	return (p);
1881 }
1882 
1883 char *
1884 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1885 {
1886 
1887 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1888 }
1889 
1890 char *
1891 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1892 {
1893 	va_list ap;
1894 	char *p;
1895 
1896 	va_start(ap, fmt);
1897 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1898 	va_end(ap);
1899 
1900 	return (p);
1901 }
1902 
1903 char *
1904 kasprintf(gfp_t gfp, const char *fmt, ...)
1905 {
1906 	va_list ap;
1907 	char *p;
1908 
1909 	va_start(ap, fmt);
1910 	p = kvasprintf(gfp, fmt, ap);
1911 	va_end(ap);
1912 
1913 	return (p);
1914 }
1915 
1916 static void
1917 linux_timer_callback_wrapper(void *context)
1918 {
1919 	struct timer_list *timer;
1920 
1921 	timer = context;
1922 
1923 	/* the timer is about to be shutdown permanently */
1924 	if (timer->function == NULL)
1925 		return;
1926 
1927 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
1928 		/* try again later */
1929 		callout_reset(&timer->callout, 1,
1930 		    &linux_timer_callback_wrapper, timer);
1931 		return;
1932 	}
1933 
1934 	timer->function(timer->data);
1935 }
1936 
1937 int
1938 mod_timer(struct timer_list *timer, int expires)
1939 {
1940 	int ret;
1941 
1942 	timer->expires = expires;
1943 	ret = callout_reset(&timer->callout,
1944 	    linux_timer_jiffies_until(expires),
1945 	    &linux_timer_callback_wrapper, timer);
1946 
1947 	MPASS(ret == 0 || ret == 1);
1948 
1949 	return (ret == 1);
1950 }
1951 
1952 void
1953 add_timer(struct timer_list *timer)
1954 {
1955 
1956 	callout_reset(&timer->callout,
1957 	    linux_timer_jiffies_until(timer->expires),
1958 	    &linux_timer_callback_wrapper, timer);
1959 }
1960 
1961 void
1962 add_timer_on(struct timer_list *timer, int cpu)
1963 {
1964 
1965 	callout_reset_on(&timer->callout,
1966 	    linux_timer_jiffies_until(timer->expires),
1967 	    &linux_timer_callback_wrapper, timer, cpu);
1968 }
1969 
1970 int
1971 del_timer(struct timer_list *timer)
1972 {
1973 
1974 	if (callout_stop(&(timer)->callout) == -1)
1975 		return (0);
1976 	return (1);
1977 }
1978 
1979 int
1980 del_timer_sync(struct timer_list *timer)
1981 {
1982 
1983 	if (callout_drain(&(timer)->callout) == -1)
1984 		return (0);
1985 	return (1);
1986 }
1987 
1988 int
1989 timer_delete_sync(struct timer_list *timer)
1990 {
1991 
1992 	return (del_timer_sync(timer));
1993 }
1994 
1995 int
1996 timer_shutdown_sync(struct timer_list *timer)
1997 {
1998 
1999 	timer->function = NULL;
2000 	return (del_timer_sync(timer));
2001 }
2002 
2003 /* greatest common divisor, Euclid equation */
2004 static uint64_t
2005 lkpi_gcd_64(uint64_t a, uint64_t b)
2006 {
2007 	uint64_t an;
2008 	uint64_t bn;
2009 
2010 	while (b != 0) {
2011 		an = b;
2012 		bn = a % b;
2013 		a = an;
2014 		b = bn;
2015 	}
2016 	return (a);
2017 }
2018 
2019 uint64_t lkpi_nsec2hz_rem;
2020 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2021 uint64_t lkpi_nsec2hz_max;
2022 
2023 uint64_t lkpi_usec2hz_rem;
2024 uint64_t lkpi_usec2hz_div = 1000000ULL;
2025 uint64_t lkpi_usec2hz_max;
2026 
2027 uint64_t lkpi_msec2hz_rem;
2028 uint64_t lkpi_msec2hz_div = 1000ULL;
2029 uint64_t lkpi_msec2hz_max;
2030 
2031 static void
2032 linux_timer_init(void *arg)
2033 {
2034 	uint64_t gcd;
2035 
2036 	/*
2037 	 * Compute an internal HZ value which can divide 2**32 to
2038 	 * avoid timer rounding problems when the tick value wraps
2039 	 * around 2**32:
2040 	 */
2041 	linux_timer_hz_mask = 1;
2042 	while (linux_timer_hz_mask < (unsigned long)hz)
2043 		linux_timer_hz_mask *= 2;
2044 	linux_timer_hz_mask--;
2045 
2046 	/* compute some internal constants */
2047 
2048 	lkpi_nsec2hz_rem = hz;
2049 	lkpi_usec2hz_rem = hz;
2050 	lkpi_msec2hz_rem = hz;
2051 
2052 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2053 	lkpi_nsec2hz_rem /= gcd;
2054 	lkpi_nsec2hz_div /= gcd;
2055 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2056 
2057 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2058 	lkpi_usec2hz_rem /= gcd;
2059 	lkpi_usec2hz_div /= gcd;
2060 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2061 
2062 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2063 	lkpi_msec2hz_rem /= gcd;
2064 	lkpi_msec2hz_div /= gcd;
2065 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2066 }
2067 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2068 
2069 void
2070 linux_complete_common(struct completion *c, int all)
2071 {
2072 	int wakeup_swapper;
2073 
2074 	sleepq_lock(c);
2075 	if (all) {
2076 		c->done = UINT_MAX;
2077 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2078 	} else {
2079 		if (c->done != UINT_MAX)
2080 			c->done++;
2081 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2082 	}
2083 	sleepq_release(c);
2084 	if (wakeup_swapper)
2085 		kick_proc0();
2086 }
2087 
2088 /*
2089  * Indefinite wait for done != 0 with or without signals.
2090  */
2091 int
2092 linux_wait_for_common(struct completion *c, int flags)
2093 {
2094 	struct task_struct *task;
2095 	int error;
2096 
2097 	if (SCHEDULER_STOPPED())
2098 		return (0);
2099 
2100 	task = current;
2101 
2102 	if (flags != 0)
2103 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2104 	else
2105 		flags = SLEEPQ_SLEEP;
2106 	error = 0;
2107 	for (;;) {
2108 		sleepq_lock(c);
2109 		if (c->done)
2110 			break;
2111 		sleepq_add(c, NULL, "completion", flags, 0);
2112 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2113 			DROP_GIANT();
2114 			error = -sleepq_wait_sig(c, 0);
2115 			PICKUP_GIANT();
2116 			if (error != 0) {
2117 				linux_schedule_save_interrupt_value(task, error);
2118 				error = -ERESTARTSYS;
2119 				goto intr;
2120 			}
2121 		} else {
2122 			DROP_GIANT();
2123 			sleepq_wait(c, 0);
2124 			PICKUP_GIANT();
2125 		}
2126 	}
2127 	if (c->done != UINT_MAX)
2128 		c->done--;
2129 	sleepq_release(c);
2130 
2131 intr:
2132 	return (error);
2133 }
2134 
2135 /*
2136  * Time limited wait for done != 0 with or without signals.
2137  */
2138 int
2139 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2140 {
2141 	struct task_struct *task;
2142 	int end = jiffies + timeout;
2143 	int error;
2144 
2145 	if (SCHEDULER_STOPPED())
2146 		return (0);
2147 
2148 	task = current;
2149 
2150 	if (flags != 0)
2151 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2152 	else
2153 		flags = SLEEPQ_SLEEP;
2154 
2155 	for (;;) {
2156 		sleepq_lock(c);
2157 		if (c->done)
2158 			break;
2159 		sleepq_add(c, NULL, "completion", flags, 0);
2160 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2161 
2162 		DROP_GIANT();
2163 		if (flags & SLEEPQ_INTERRUPTIBLE)
2164 			error = -sleepq_timedwait_sig(c, 0);
2165 		else
2166 			error = -sleepq_timedwait(c, 0);
2167 		PICKUP_GIANT();
2168 
2169 		if (error != 0) {
2170 			/* check for timeout */
2171 			if (error == -EWOULDBLOCK) {
2172 				error = 0;	/* timeout */
2173 			} else {
2174 				/* signal happened */
2175 				linux_schedule_save_interrupt_value(task, error);
2176 				error = -ERESTARTSYS;
2177 			}
2178 			goto done;
2179 		}
2180 	}
2181 	if (c->done != UINT_MAX)
2182 		c->done--;
2183 	sleepq_release(c);
2184 
2185 	/* return how many jiffies are left */
2186 	error = linux_timer_jiffies_until(end);
2187 done:
2188 	return (error);
2189 }
2190 
2191 int
2192 linux_try_wait_for_completion(struct completion *c)
2193 {
2194 	int isdone;
2195 
2196 	sleepq_lock(c);
2197 	isdone = (c->done != 0);
2198 	if (c->done != 0 && c->done != UINT_MAX)
2199 		c->done--;
2200 	sleepq_release(c);
2201 	return (isdone);
2202 }
2203 
2204 int
2205 linux_completion_done(struct completion *c)
2206 {
2207 	int isdone;
2208 
2209 	sleepq_lock(c);
2210 	isdone = (c->done != 0);
2211 	sleepq_release(c);
2212 	return (isdone);
2213 }
2214 
2215 static void
2216 linux_cdev_deref(struct linux_cdev *ldev)
2217 {
2218 	if (refcount_release(&ldev->refs) &&
2219 	    ldev->kobj.ktype == &linux_cdev_ktype)
2220 		kfree(ldev);
2221 }
2222 
2223 static void
2224 linux_cdev_release(struct kobject *kobj)
2225 {
2226 	struct linux_cdev *cdev;
2227 	struct kobject *parent;
2228 
2229 	cdev = container_of(kobj, struct linux_cdev, kobj);
2230 	parent = kobj->parent;
2231 	linux_destroy_dev(cdev);
2232 	linux_cdev_deref(cdev);
2233 	kobject_put(parent);
2234 }
2235 
2236 static void
2237 linux_cdev_static_release(struct kobject *kobj)
2238 {
2239 	struct cdev *cdev;
2240 	struct linux_cdev *ldev;
2241 
2242 	ldev = container_of(kobj, struct linux_cdev, kobj);
2243 	cdev = ldev->cdev;
2244 	if (cdev != NULL) {
2245 		destroy_dev(cdev);
2246 		ldev->cdev = NULL;
2247 	}
2248 	kobject_put(kobj->parent);
2249 }
2250 
2251 int
2252 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2253 {
2254 	int ret;
2255 
2256 	if (dev->devt != 0) {
2257 		/* Set parent kernel object. */
2258 		ldev->kobj.parent = &dev->kobj;
2259 
2260 		/*
2261 		 * Unlike Linux we require the kobject of the
2262 		 * character device structure to have a valid name
2263 		 * before calling this function:
2264 		 */
2265 		if (ldev->kobj.name == NULL)
2266 			return (-EINVAL);
2267 
2268 		ret = cdev_add(ldev, dev->devt, 1);
2269 		if (ret)
2270 			return (ret);
2271 	}
2272 	ret = device_add(dev);
2273 	if (ret != 0 && dev->devt != 0)
2274 		cdev_del(ldev);
2275 	return (ret);
2276 }
2277 
2278 void
2279 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2280 {
2281 	device_del(dev);
2282 
2283 	if (dev->devt != 0)
2284 		cdev_del(ldev);
2285 }
2286 
2287 static void
2288 linux_destroy_dev(struct linux_cdev *ldev)
2289 {
2290 
2291 	if (ldev->cdev == NULL)
2292 		return;
2293 
2294 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2295 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2296 
2297 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2298 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2299 		pause("ldevdtr", hz / 4);
2300 
2301 	destroy_dev(ldev->cdev);
2302 	ldev->cdev = NULL;
2303 }
2304 
2305 const struct kobj_type linux_cdev_ktype = {
2306 	.release = linux_cdev_release,
2307 };
2308 
2309 const struct kobj_type linux_cdev_static_ktype = {
2310 	.release = linux_cdev_static_release,
2311 };
2312 
2313 static void
2314 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2315 {
2316 	struct notifier_block *nb;
2317 	struct netdev_notifier_info ni;
2318 
2319 	nb = arg;
2320 	ni.ifp = ifp;
2321 	ni.dev = (struct net_device *)ifp;
2322 	if (linkstate == LINK_STATE_UP)
2323 		nb->notifier_call(nb, NETDEV_UP, &ni);
2324 	else
2325 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2326 }
2327 
2328 static void
2329 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2330 {
2331 	struct notifier_block *nb;
2332 	struct netdev_notifier_info ni;
2333 
2334 	nb = arg;
2335 	ni.ifp = ifp;
2336 	ni.dev = (struct net_device *)ifp;
2337 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2338 }
2339 
2340 static void
2341 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2342 {
2343 	struct notifier_block *nb;
2344 	struct netdev_notifier_info ni;
2345 
2346 	nb = arg;
2347 	ni.ifp = ifp;
2348 	ni.dev = (struct net_device *)ifp;
2349 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2350 }
2351 
2352 static void
2353 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2354 {
2355 	struct notifier_block *nb;
2356 	struct netdev_notifier_info ni;
2357 
2358 	nb = arg;
2359 	ni.ifp = ifp;
2360 	ni.dev = (struct net_device *)ifp;
2361 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2362 }
2363 
2364 static void
2365 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2366 {
2367 	struct notifier_block *nb;
2368 	struct netdev_notifier_info ni;
2369 
2370 	nb = arg;
2371 	ni.ifp = ifp;
2372 	ni.dev = (struct net_device *)ifp;
2373 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2374 }
2375 
2376 int
2377 register_netdevice_notifier(struct notifier_block *nb)
2378 {
2379 
2380 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2381 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2382 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2383 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2384 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2385 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2386 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2387 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2388 
2389 	return (0);
2390 }
2391 
2392 int
2393 register_inetaddr_notifier(struct notifier_block *nb)
2394 {
2395 
2396 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2397 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2398 	return (0);
2399 }
2400 
2401 int
2402 unregister_netdevice_notifier(struct notifier_block *nb)
2403 {
2404 
2405 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2406 	    nb->tags[NETDEV_UP]);
2407 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2408 	    nb->tags[NETDEV_REGISTER]);
2409 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2410 	    nb->tags[NETDEV_UNREGISTER]);
2411 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2412 	    nb->tags[NETDEV_CHANGEADDR]);
2413 
2414 	return (0);
2415 }
2416 
2417 int
2418 unregister_inetaddr_notifier(struct notifier_block *nb)
2419 {
2420 
2421 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2422 	    nb->tags[NETDEV_CHANGEIFADDR]);
2423 
2424 	return (0);
2425 }
2426 
2427 struct list_sort_thunk {
2428 	int (*cmp)(void *, struct list_head *, struct list_head *);
2429 	void *priv;
2430 };
2431 
2432 static inline int
2433 linux_le_cmp(const void *d1, const void *d2, void *priv)
2434 {
2435 	struct list_head *le1, *le2;
2436 	struct list_sort_thunk *thunk;
2437 
2438 	thunk = priv;
2439 	le1 = *(__DECONST(struct list_head **, d1));
2440 	le2 = *(__DECONST(struct list_head **, d2));
2441 	return ((thunk->cmp)(thunk->priv, le1, le2));
2442 }
2443 
2444 void
2445 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2446     struct list_head *a, struct list_head *b))
2447 {
2448 	struct list_sort_thunk thunk;
2449 	struct list_head **ar, *le;
2450 	size_t count, i;
2451 
2452 	count = 0;
2453 	list_for_each(le, head)
2454 		count++;
2455 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2456 	i = 0;
2457 	list_for_each(le, head)
2458 		ar[i++] = le;
2459 	thunk.cmp = cmp;
2460 	thunk.priv = priv;
2461 	qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk);
2462 	INIT_LIST_HEAD(head);
2463 	for (i = 0; i < count; i++)
2464 		list_add_tail(ar[i], head);
2465 	free(ar, M_KMALLOC);
2466 }
2467 
2468 #if defined(__i386__) || defined(__amd64__)
2469 int
2470 linux_wbinvd_on_all_cpus(void)
2471 {
2472 
2473 	pmap_invalidate_cache();
2474 	return (0);
2475 }
2476 #endif
2477 
2478 int
2479 linux_on_each_cpu(void callback(void *), void *data)
2480 {
2481 
2482 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2483 	    smp_no_rendezvous_barrier, data);
2484 	return (0);
2485 }
2486 
2487 int
2488 linux_in_atomic(void)
2489 {
2490 
2491 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2492 }
2493 
2494 struct linux_cdev *
2495 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2496 {
2497 	dev_t dev = MKDEV(major, minor);
2498 	struct cdev *cdev;
2499 
2500 	dev_lock();
2501 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2502 		struct linux_cdev *ldev = cdev->si_drv1;
2503 		if (ldev->dev == dev &&
2504 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2505 			break;
2506 		}
2507 	}
2508 	dev_unlock();
2509 
2510 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2511 }
2512 
2513 int
2514 __register_chrdev(unsigned int major, unsigned int baseminor,
2515     unsigned int count, const char *name,
2516     const struct file_operations *fops)
2517 {
2518 	struct linux_cdev *cdev;
2519 	int ret = 0;
2520 	int i;
2521 
2522 	for (i = baseminor; i < baseminor + count; i++) {
2523 		cdev = cdev_alloc();
2524 		cdev->ops = fops;
2525 		kobject_set_name(&cdev->kobj, name);
2526 
2527 		ret = cdev_add(cdev, makedev(major, i), 1);
2528 		if (ret != 0)
2529 			break;
2530 	}
2531 	return (ret);
2532 }
2533 
2534 int
2535 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2536     unsigned int count, const char *name,
2537     const struct file_operations *fops, uid_t uid,
2538     gid_t gid, int mode)
2539 {
2540 	struct linux_cdev *cdev;
2541 	int ret = 0;
2542 	int i;
2543 
2544 	for (i = baseminor; i < baseminor + count; i++) {
2545 		cdev = cdev_alloc();
2546 		cdev->ops = fops;
2547 		kobject_set_name(&cdev->kobj, name);
2548 
2549 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2550 		if (ret != 0)
2551 			break;
2552 	}
2553 	return (ret);
2554 }
2555 
2556 void
2557 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2558     unsigned int count, const char *name)
2559 {
2560 	struct linux_cdev *cdevp;
2561 	int i;
2562 
2563 	for (i = baseminor; i < baseminor + count; i++) {
2564 		cdevp = linux_find_cdev(name, major, i);
2565 		if (cdevp != NULL)
2566 			cdev_del(cdevp);
2567 	}
2568 }
2569 
2570 void
2571 linux_dump_stack(void)
2572 {
2573 #ifdef STACK
2574 	struct stack st;
2575 
2576 	stack_save(&st);
2577 	stack_print(&st);
2578 #endif
2579 }
2580 
2581 int
2582 linuxkpi_net_ratelimit(void)
2583 {
2584 
2585 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2586 	   lkpi_net_maxpps));
2587 }
2588 
2589 struct io_mapping *
2590 io_mapping_create_wc(resource_size_t base, unsigned long size)
2591 {
2592 	struct io_mapping *mapping;
2593 
2594 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2595 	if (mapping == NULL)
2596 		return (NULL);
2597 	return (io_mapping_init_wc(mapping, base, size));
2598 }
2599 
2600 #if defined(__i386__) || defined(__amd64__)
2601 bool linux_cpu_has_clflush;
2602 struct cpuinfo_x86 boot_cpu_data;
2603 struct cpuinfo_x86 *__cpu_data;
2604 #endif
2605 
2606 cpumask_t *
2607 lkpi_get_static_single_cpu_mask(int cpuid)
2608 {
2609 
2610 	KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n",
2611 	    __func__, cpuid));
2612 	KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n",
2613 	    __func__, cpuid));
2614 
2615 	return (static_single_cpu_mask[cpuid]);
2616 }
2617 
2618 bool
2619 lkpi_xen_initial_domain(void)
2620 {
2621 #ifdef XENHVM
2622 	return (xen_initial_domain());
2623 #else
2624 	return (false);
2625 #endif
2626 }
2627 
2628 bool
2629 lkpi_xen_pv_domain(void)
2630 {
2631 #ifdef XENHVM
2632 	return (xen_pv_domain());
2633 #else
2634 	return (false);
2635 #endif
2636 }
2637 
2638 static void
2639 linux_compat_init(void *arg)
2640 {
2641 	struct sysctl_oid *rootoid;
2642 	int i;
2643 
2644 #if defined(__i386__) || defined(__amd64__)
2645 	static const uint32_t x86_vendors[X86_VENDOR_NUM] = {
2646 		[X86_VENDOR_INTEL] = CPU_VENDOR_INTEL,
2647 		[X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX,
2648 		[X86_VENDOR_AMD] = CPU_VENDOR_AMD,
2649 		[X86_VENDOR_UMC] = CPU_VENDOR_UMC,
2650 		[X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR,
2651 		[X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA,
2652 		[X86_VENDOR_NSC] = CPU_VENDOR_NSC,
2653 		[X86_VENDOR_HYGON] = CPU_VENDOR_HYGON,
2654 	};
2655 	uint8_t x86_vendor = X86_VENDOR_UNKNOWN;
2656 
2657 	for (i = 0; i < X86_VENDOR_NUM; i++) {
2658 		if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) {
2659 			x86_vendor = i;
2660 			break;
2661 		}
2662 	}
2663 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2664 	boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2665 	boot_cpu_data.x86_max_cores = mp_ncpus;
2666 	boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id);
2667 	boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id);
2668 	boot_cpu_data.x86_vendor = x86_vendor;
2669 
2670 	__cpu_data = mallocarray(mp_maxid + 1,
2671 	    sizeof(*__cpu_data), M_KMALLOC, M_WAITOK | M_ZERO);
2672 	CPU_FOREACH(i) {
2673 		__cpu_data[i].x86_clflush_size = cpu_clflush_line_size;
2674 		__cpu_data[i].x86_max_cores = mp_ncpus;
2675 		__cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id);
2676 		__cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id);
2677 		__cpu_data[i].x86_vendor = x86_vendor;
2678 	}
2679 #endif
2680 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2681 
2682 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2683 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2684 	kobject_init(&linux_class_root, &linux_class_ktype);
2685 	kobject_set_name(&linux_class_root, "class");
2686 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2687 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2688 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2689 	kobject_set_name(&linux_root_device.kobj, "device");
2690 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2691 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2692 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2693 	linux_root_device.bsddev = root_bus;
2694 	linux_class_misc.name = "misc";
2695 	class_register(&linux_class_misc);
2696 	INIT_LIST_HEAD(&pci_drivers);
2697 	INIT_LIST_HEAD(&pci_devices);
2698 	spin_lock_init(&pci_lock);
2699 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2700 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2701 		LIST_INIT(&vmmaphead[i]);
2702 	init_waitqueue_head(&linux_bit_waitq);
2703 	init_waitqueue_head(&linux_var_waitq);
2704 
2705 	CPU_COPY(&all_cpus, &cpu_online_mask);
2706 	/*
2707 	 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system.
2708 	 * CPUs are indexed from 0..(mp_maxid).  The entry for cpuid 0 will only
2709 	 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on.
2710 	 * This is used by cpumask_of() (and possibly others in the future) for,
2711 	 * e.g., drivers to pass hints to irq_set_affinity_hint().
2712 	 */
2713 	static_single_cpu_mask = mallocarray(mp_maxid + 1,
2714 	    sizeof(static_single_cpu_mask), M_KMALLOC, M_WAITOK | M_ZERO);
2715 
2716 	/*
2717 	 * When the number of CPUs reach a threshold, we start to save memory
2718 	 * given the sets are static by overlapping those having their single
2719 	 * bit set at same position in a bitset word.  Asymptotically, this
2720 	 * regular scheme is in O(n²) whereas the overlapping one is in O(n)
2721 	 * only with n being the maximum number of CPUs, so the gain will become
2722 	 * huge quite quickly.  The threshold for 64-bit architectures is 128
2723 	 * CPUs.
2724 	 */
2725 	if (mp_ncpus < (2 * _BITSET_BITS)) {
2726 		cpumask_t *sscm_ptr;
2727 
2728 		/*
2729 		 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) *
2730 		 * (_BITSET_BITS / 8)' bytes (for comparison with the
2731 		 * overlapping scheme).
2732 		 */
2733 		static_single_cpu_mask_lcs = mallocarray(mp_ncpus,
2734 		    sizeof(*static_single_cpu_mask_lcs),
2735 		    M_KMALLOC, M_WAITOK | M_ZERO);
2736 
2737 		sscm_ptr = static_single_cpu_mask_lcs;
2738 		CPU_FOREACH(i) {
2739 			static_single_cpu_mask[i] = sscm_ptr++;
2740 			CPU_SET(i, static_single_cpu_mask[i]);
2741 		}
2742 	} else {
2743 		/* Pointer to a bitset word. */
2744 		__typeof(((cpuset_t *)NULL)->__bits[0]) *bwp;
2745 
2746 		/*
2747 		 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t'
2748 		 * really) with a single bit set that can be reused for all
2749 		 * single CPU masks by making them start at different offsets.
2750 		 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before
2751 		 * the word having its single bit set, and the same amount
2752 		 * after.
2753 		 */
2754 		static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS,
2755 		    (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8),
2756 		    M_KMALLOC, M_WAITOK | M_ZERO);
2757 
2758 		/*
2759 		 * We rely below on cpuset_t and the bitset generic
2760 		 * implementation assigning words in the '__bits' array in the
2761 		 * same order of bits (i.e., little-endian ordering, not to be
2762 		 * confused with machine endianness, which concerns bits in
2763 		 * words and other integers).  This is an imperfect test, but it
2764 		 * will detect a change to big-endian ordering.
2765 		 */
2766 		_Static_assert(
2767 		    __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1,
2768 		    "Assumes a bitset implementation that is little-endian "
2769 		    "on its words");
2770 
2771 		/* Initialize the single bit of each static span. */
2772 		bwp = (__typeof(bwp))static_single_cpu_mask_lcs +
2773 		    (__bitset_words(CPU_SETSIZE) - 1);
2774 		for (i = 0; i < _BITSET_BITS; i++) {
2775 			CPU_SET(i, (cpuset_t *)bwp);
2776 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1);
2777 		}
2778 
2779 		/*
2780 		 * Finally set all CPU masks to the proper word in their
2781 		 * relevant span.
2782 		 */
2783 		CPU_FOREACH(i) {
2784 			bwp = (__typeof(bwp))static_single_cpu_mask_lcs;
2785 			/* Find the non-zero word of the relevant span. */
2786 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) *
2787 			    (i % _BITSET_BITS) +
2788 			    __bitset_words(CPU_SETSIZE) - 1;
2789 			/* Shift to find the CPU mask start. */
2790 			bwp -= (i / _BITSET_BITS);
2791 			static_single_cpu_mask[i] = (cpuset_t *)bwp;
2792 		}
2793 	}
2794 
2795 	strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release));
2796 }
2797 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2798 
2799 static void
2800 linux_compat_uninit(void *arg)
2801 {
2802 	linux_kobject_kfree_name(&linux_class_root);
2803 	linux_kobject_kfree_name(&linux_root_device.kobj);
2804 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2805 
2806 	free(static_single_cpu_mask_lcs, M_KMALLOC);
2807 	free(static_single_cpu_mask, M_KMALLOC);
2808 #if defined(__i386__) || defined(__amd64__)
2809 	free(__cpu_data, M_KMALLOC);
2810 #endif
2811 
2812 	mtx_destroy(&vmmaplock);
2813 	spin_lock_destroy(&pci_lock);
2814 	rw_destroy(&linux_vma_lock);
2815 }
2816 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2817 
2818 /*
2819  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2820  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2821  * used. Assert these types have the same size, else some parts of the
2822  * LinuxKPI may not work like expected:
2823  */
2824 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2825