xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision c7046f76c2c027b00c0e6ba57cfd28f1a78f5e23)
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 __FBSDID("$FreeBSD$");
32 
33 #include "opt_stack.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/kernel.h>
39 #include <sys/sysctl.h>
40 #include <sys/proc.h>
41 #include <sys/sglist.h>
42 #include <sys/sleepqueue.h>
43 #include <sys/refcount.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/bus.h>
47 #include <sys/eventhandler.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/filio.h>
51 #include <sys/rwlock.h>
52 #include <sys/mman.h>
53 #include <sys/stack.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 
64 #include <machine/stdarg.h>
65 
66 #if defined(__i386__) || defined(__amd64__)
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/list.h>
87 #include <linux/kthread.h>
88 #include <linux/kernel.h>
89 #include <linux/compat.h>
90 #include <linux/io-mapping.h>
91 #include <linux/poll.h>
92 #include <linux/smp.h>
93 #include <linux/wait_bit.h>
94 #include <linux/rcupdate.h>
95 #include <linux/interval_tree.h>
96 #include <linux/interval_tree_generic.h>
97 
98 #if defined(__i386__) || defined(__amd64__)
99 #include <asm/smp.h>
100 #include <asm/processor.h>
101 #endif
102 
103 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
104     "LinuxKPI parameters");
105 
106 int linuxkpi_debug;
107 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
108     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
109 
110 int linuxkpi_warn_dump_stack = 0;
111 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN,
112     &linuxkpi_warn_dump_stack, 0,
113     "Set to enable stack traces from WARN_ON(). Clear to disable.");
114 
115 static struct timeval lkpi_net_lastlog;
116 static int lkpi_net_curpps;
117 static int lkpi_net_maxpps = 99;
118 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
119     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
120 
121 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat");
122 
123 #include <linux/rbtree.h>
124 /* Undo Linux compat changes. */
125 #undef RB_ROOT
126 #undef file
127 #undef cdev
128 #define	RB_ROOT(head)	(head)->rbh_root
129 
130 static void linux_destroy_dev(struct linux_cdev *);
131 static void linux_cdev_deref(struct linux_cdev *ldev);
132 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
133 
134 cpumask_t cpu_online_mask;
135 struct kobject linux_class_root;
136 struct device linux_root_device;
137 struct class linux_class_misc;
138 struct list_head pci_drivers;
139 struct list_head pci_devices;
140 spinlock_t pci_lock;
141 
142 unsigned long linux_timer_hz_mask;
143 
144 wait_queue_head_t linux_bit_waitq;
145 wait_queue_head_t linux_var_waitq;
146 
147 int
148 panic_cmp(struct rb_node *one, struct rb_node *two)
149 {
150 	panic("no cmp");
151 }
152 
153 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
154 
155 #define	START(node)	((node)->start)
156 #define	LAST(node)	((node)->last)
157 
158 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
159     LAST,, lkpi_interval_tree)
160 
161 struct kobject *
162 kobject_create(void)
163 {
164 	struct kobject *kobj;
165 
166 	kobj = kzalloc(sizeof(*kobj), GFP_KERNEL);
167 	if (kobj == NULL)
168 		return (NULL);
169 	kobject_init(kobj, &linux_kfree_type);
170 
171 	return (kobj);
172 }
173 
174 
175 int
176 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
177 {
178 	va_list tmp_va;
179 	int len;
180 	char *old;
181 	char *name;
182 	char dummy;
183 
184 	old = kobj->name;
185 
186 	if (old && fmt == NULL)
187 		return (0);
188 
189 	/* compute length of string */
190 	va_copy(tmp_va, args);
191 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
192 	va_end(tmp_va);
193 
194 	/* account for zero termination */
195 	len++;
196 
197 	/* check for error */
198 	if (len < 1)
199 		return (-EINVAL);
200 
201 	/* allocate memory for string */
202 	name = kzalloc(len, GFP_KERNEL);
203 	if (name == NULL)
204 		return (-ENOMEM);
205 	vsnprintf(name, len, fmt, args);
206 	kobj->name = name;
207 
208 	/* free old string */
209 	kfree(old);
210 
211 	/* filter new string */
212 	for (; *name != '\0'; name++)
213 		if (*name == '/')
214 			*name = '!';
215 	return (0);
216 }
217 
218 int
219 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
220 {
221 	va_list args;
222 	int error;
223 
224 	va_start(args, fmt);
225 	error = kobject_set_name_vargs(kobj, fmt, args);
226 	va_end(args);
227 
228 	return (error);
229 }
230 
231 static int
232 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
233 {
234 	const struct kobj_type *t;
235 	int error;
236 
237 	kobj->parent = parent;
238 	error = sysfs_create_dir(kobj);
239 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
240 		struct attribute **attr;
241 		t = kobj->ktype;
242 
243 		for (attr = t->default_attrs; *attr != NULL; attr++) {
244 			error = sysfs_create_file(kobj, *attr);
245 			if (error)
246 				break;
247 		}
248 		if (error)
249 			sysfs_remove_dir(kobj);
250 	}
251 	return (error);
252 }
253 
254 int
255 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
256 {
257 	va_list args;
258 	int error;
259 
260 	va_start(args, fmt);
261 	error = kobject_set_name_vargs(kobj, fmt, args);
262 	va_end(args);
263 	if (error)
264 		return (error);
265 
266 	return kobject_add_complete(kobj, parent);
267 }
268 
269 void
270 linux_kobject_release(struct kref *kref)
271 {
272 	struct kobject *kobj;
273 	char *name;
274 
275 	kobj = container_of(kref, struct kobject, kref);
276 	sysfs_remove_dir(kobj);
277 	name = kobj->name;
278 	if (kobj->ktype && kobj->ktype->release)
279 		kobj->ktype->release(kobj);
280 	kfree(name);
281 }
282 
283 static void
284 linux_kobject_kfree(struct kobject *kobj)
285 {
286 	kfree(kobj);
287 }
288 
289 static void
290 linux_kobject_kfree_name(struct kobject *kobj)
291 {
292 	if (kobj) {
293 		kfree(kobj->name);
294 	}
295 }
296 
297 const struct kobj_type linux_kfree_type = {
298 	.release = linux_kobject_kfree
299 };
300 
301 static ssize_t
302 lkpi_kobj_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
303 {
304 	struct kobj_attribute *ka =
305 	    container_of(attr, struct kobj_attribute, attr);
306 
307 	if (ka->show == NULL)
308 		return (-EIO);
309 
310 	return (ka->show(kobj, ka, buf));
311 }
312 
313 static ssize_t
314 lkpi_kobj_attr_store(struct kobject *kobj, struct attribute *attr,
315     const char *buf, size_t count)
316 {
317 	struct kobj_attribute *ka =
318 	    container_of(attr, struct kobj_attribute, attr);
319 
320 	if (ka->store == NULL)
321 		return (-EIO);
322 
323 	return (ka->store(kobj, ka, buf, count));
324 }
325 
326 const struct sysfs_ops kobj_sysfs_ops = {
327 	.show	= lkpi_kobj_attr_show,
328 	.store	= lkpi_kobj_attr_store,
329 };
330 
331 static void
332 linux_device_release(struct device *dev)
333 {
334 	pr_debug("linux_device_release: %s\n", dev_name(dev));
335 	kfree(dev);
336 }
337 
338 static ssize_t
339 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
340 {
341 	struct class_attribute *dattr;
342 	ssize_t error;
343 
344 	dattr = container_of(attr, struct class_attribute, attr);
345 	error = -EIO;
346 	if (dattr->show)
347 		error = dattr->show(container_of(kobj, struct class, kobj),
348 		    dattr, buf);
349 	return (error);
350 }
351 
352 static ssize_t
353 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
354     size_t count)
355 {
356 	struct class_attribute *dattr;
357 	ssize_t error;
358 
359 	dattr = container_of(attr, struct class_attribute, attr);
360 	error = -EIO;
361 	if (dattr->store)
362 		error = dattr->store(container_of(kobj, struct class, kobj),
363 		    dattr, buf, count);
364 	return (error);
365 }
366 
367 static void
368 linux_class_release(struct kobject *kobj)
369 {
370 	struct class *class;
371 
372 	class = container_of(kobj, struct class, kobj);
373 	if (class->class_release)
374 		class->class_release(class);
375 }
376 
377 static const struct sysfs_ops linux_class_sysfs = {
378 	.show  = linux_class_show,
379 	.store = linux_class_store,
380 };
381 
382 const struct kobj_type linux_class_ktype = {
383 	.release = linux_class_release,
384 	.sysfs_ops = &linux_class_sysfs
385 };
386 
387 static void
388 linux_dev_release(struct kobject *kobj)
389 {
390 	struct device *dev;
391 
392 	dev = container_of(kobj, struct device, kobj);
393 	/* This is the precedence defined by linux. */
394 	if (dev->release)
395 		dev->release(dev);
396 	else if (dev->class && dev->class->dev_release)
397 		dev->class->dev_release(dev);
398 }
399 
400 static ssize_t
401 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
402 {
403 	struct device_attribute *dattr;
404 	ssize_t error;
405 
406 	dattr = container_of(attr, struct device_attribute, attr);
407 	error = -EIO;
408 	if (dattr->show)
409 		error = dattr->show(container_of(kobj, struct device, kobj),
410 		    dattr, buf);
411 	return (error);
412 }
413 
414 static ssize_t
415 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
416     size_t count)
417 {
418 	struct device_attribute *dattr;
419 	ssize_t error;
420 
421 	dattr = container_of(attr, struct device_attribute, attr);
422 	error = -EIO;
423 	if (dattr->store)
424 		error = dattr->store(container_of(kobj, struct device, kobj),
425 		    dattr, buf, count);
426 	return (error);
427 }
428 
429 static const struct sysfs_ops linux_dev_sysfs = {
430 	.show  = linux_dev_show,
431 	.store = linux_dev_store,
432 };
433 
434 const struct kobj_type linux_dev_ktype = {
435 	.release = linux_dev_release,
436 	.sysfs_ops = &linux_dev_sysfs
437 };
438 
439 struct device *
440 device_create(struct class *class, struct device *parent, dev_t devt,
441     void *drvdata, const char *fmt, ...)
442 {
443 	struct device *dev;
444 	va_list args;
445 
446 	dev = kzalloc(sizeof(*dev), M_WAITOK);
447 	dev->parent = parent;
448 	dev->class = class;
449 	dev->devt = devt;
450 	dev->driver_data = drvdata;
451 	dev->release = linux_device_release;
452 	va_start(args, fmt);
453 	kobject_set_name_vargs(&dev->kobj, fmt, args);
454 	va_end(args);
455 	device_register(dev);
456 
457 	return (dev);
458 }
459 
460 struct device *
461 device_create_groups_vargs(struct class *class, struct device *parent,
462     dev_t devt, void *drvdata, const struct attribute_group **groups,
463     const char *fmt, va_list args)
464 {
465 	struct device *dev = NULL;
466 	int retval = -ENODEV;
467 
468 	if (class == NULL || IS_ERR(class))
469 		goto error;
470 
471 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
472 	if (!dev) {
473 		retval = -ENOMEM;
474 		goto error;
475 	}
476 
477 	dev->devt = devt;
478 	dev->class = class;
479 	dev->parent = parent;
480 	dev->groups = groups;
481 	dev->release = device_create_release;
482 	/* device_initialize() needs the class and parent to be set */
483 	device_initialize(dev);
484 	dev_set_drvdata(dev, drvdata);
485 
486 	retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
487 	if (retval)
488 		goto error;
489 
490 	retval = device_add(dev);
491 	if (retval)
492 		goto error;
493 
494 	return dev;
495 
496 error:
497 	put_device(dev);
498 	return ERR_PTR(retval);
499 }
500 
501 struct class *
502 class_create(struct module *owner, const char *name)
503 {
504 	struct class *class;
505 	int error;
506 
507 	class = kzalloc(sizeof(*class), M_WAITOK);
508 	class->owner = owner;
509 	class->name = name;
510 	class->class_release = linux_class_kfree;
511 	error = class_register(class);
512 	if (error) {
513 		kfree(class);
514 		return (NULL);
515 	}
516 
517 	return (class);
518 }
519 
520 int
521 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
522     struct kobject *parent, const char *fmt, ...)
523 {
524 	va_list args;
525 	int error;
526 
527 	kobject_init(kobj, ktype);
528 	kobj->ktype = ktype;
529 	kobj->parent = parent;
530 	kobj->name = NULL;
531 
532 	va_start(args, fmt);
533 	error = kobject_set_name_vargs(kobj, fmt, args);
534 	va_end(args);
535 	if (error)
536 		return (error);
537 	return kobject_add_complete(kobj, parent);
538 }
539 
540 static void
541 linux_kq_lock(void *arg)
542 {
543 	spinlock_t *s = arg;
544 
545 	spin_lock(s);
546 }
547 static void
548 linux_kq_unlock(void *arg)
549 {
550 	spinlock_t *s = arg;
551 
552 	spin_unlock(s);
553 }
554 
555 static void
556 linux_kq_assert_lock(void *arg, int what)
557 {
558 #ifdef INVARIANTS
559 	spinlock_t *s = arg;
560 
561 	if (what == LA_LOCKED)
562 		mtx_assert(&s->m, MA_OWNED);
563 	else
564 		mtx_assert(&s->m, MA_NOTOWNED);
565 #endif
566 }
567 
568 static void
569 linux_file_kqfilter_poll(struct linux_file *, int);
570 
571 struct linux_file *
572 linux_file_alloc(void)
573 {
574 	struct linux_file *filp;
575 
576 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
577 
578 	/* set initial refcount */
579 	filp->f_count = 1;
580 
581 	/* setup fields needed by kqueue support */
582 	spin_lock_init(&filp->f_kqlock);
583 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
584 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
585 
586 	return (filp);
587 }
588 
589 void
590 linux_file_free(struct linux_file *filp)
591 {
592 	if (filp->_file == NULL) {
593 		if (filp->f_op != NULL && filp->f_op->release != NULL)
594 			filp->f_op->release(filp->f_vnode, filp);
595 		if (filp->f_shmem != NULL)
596 			vm_object_deallocate(filp->f_shmem);
597 		kfree_rcu(filp, rcu);
598 	} else {
599 		/*
600 		 * The close method of the character device or file
601 		 * will free the linux_file structure:
602 		 */
603 		_fdrop(filp->_file, curthread);
604 	}
605 }
606 
607 struct linux_cdev *
608 cdev_alloc(void)
609 {
610 	struct linux_cdev *cdev;
611 
612 	cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
613 	kobject_init(&cdev->kobj, &linux_cdev_ktype);
614 	cdev->refs = 1;
615 	return (cdev);
616 }
617 
618 static int
619 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
620     vm_page_t *mres)
621 {
622 	struct vm_area_struct *vmap;
623 
624 	vmap = linux_cdev_handle_find(vm_obj->handle);
625 
626 	MPASS(vmap != NULL);
627 	MPASS(vmap->vm_private_data == vm_obj->handle);
628 
629 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
630 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
631 		vm_page_t page;
632 
633 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
634 			/*
635 			 * If the passed in result page is a fake
636 			 * page, update it with the new physical
637 			 * address.
638 			 */
639 			page = *mres;
640 			vm_page_updatefake(page, paddr, vm_obj->memattr);
641 		} else {
642 			/*
643 			 * Replace the passed in "mres" page with our
644 			 * own fake page and free up the all of the
645 			 * original pages.
646 			 */
647 			VM_OBJECT_WUNLOCK(vm_obj);
648 			page = vm_page_getfake(paddr, vm_obj->memattr);
649 			VM_OBJECT_WLOCK(vm_obj);
650 
651 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
652 			*mres = page;
653 		}
654 		vm_page_valid(page);
655 		return (VM_PAGER_OK);
656 	}
657 	return (VM_PAGER_FAIL);
658 }
659 
660 static int
661 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
662     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
663 {
664 	struct vm_area_struct *vmap;
665 	int err;
666 
667 	/* get VM area structure */
668 	vmap = linux_cdev_handle_find(vm_obj->handle);
669 	MPASS(vmap != NULL);
670 	MPASS(vmap->vm_private_data == vm_obj->handle);
671 
672 	VM_OBJECT_WUNLOCK(vm_obj);
673 
674 	linux_set_current(curthread);
675 
676 	down_write(&vmap->vm_mm->mmap_sem);
677 	if (unlikely(vmap->vm_ops == NULL)) {
678 		err = VM_FAULT_SIGBUS;
679 	} else {
680 		struct vm_fault vmf;
681 
682 		/* fill out VM fault structure */
683 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
684 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
685 		vmf.pgoff = 0;
686 		vmf.page = NULL;
687 		vmf.vma = vmap;
688 
689 		vmap->vm_pfn_count = 0;
690 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
691 		vmap->vm_obj = vm_obj;
692 
693 		err = vmap->vm_ops->fault(&vmf);
694 
695 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
696 			kern_yield(PRI_USER);
697 			err = vmap->vm_ops->fault(&vmf);
698 		}
699 	}
700 
701 	/* translate return code */
702 	switch (err) {
703 	case VM_FAULT_OOM:
704 		err = VM_PAGER_AGAIN;
705 		break;
706 	case VM_FAULT_SIGBUS:
707 		err = VM_PAGER_BAD;
708 		break;
709 	case VM_FAULT_NOPAGE:
710 		/*
711 		 * By contract the fault handler will return having
712 		 * busied all the pages itself. If pidx is already
713 		 * found in the object, it will simply xbusy the first
714 		 * page and return with vm_pfn_count set to 1.
715 		 */
716 		*first = vmap->vm_pfn_first;
717 		*last = *first + vmap->vm_pfn_count - 1;
718 		err = VM_PAGER_OK;
719 		break;
720 	default:
721 		err = VM_PAGER_ERROR;
722 		break;
723 	}
724 	up_write(&vmap->vm_mm->mmap_sem);
725 	VM_OBJECT_WLOCK(vm_obj);
726 	return (err);
727 }
728 
729 static struct rwlock linux_vma_lock;
730 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
731     TAILQ_HEAD_INITIALIZER(linux_vma_head);
732 
733 static void
734 linux_cdev_handle_free(struct vm_area_struct *vmap)
735 {
736 	/* Drop reference on vm_file */
737 	if (vmap->vm_file != NULL)
738 		fput(vmap->vm_file);
739 
740 	/* Drop reference on mm_struct */
741 	mmput(vmap->vm_mm);
742 
743 	kfree(vmap);
744 }
745 
746 static void
747 linux_cdev_handle_remove(struct vm_area_struct *vmap)
748 {
749 	rw_wlock(&linux_vma_lock);
750 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
751 	rw_wunlock(&linux_vma_lock);
752 }
753 
754 static struct vm_area_struct *
755 linux_cdev_handle_find(void *handle)
756 {
757 	struct vm_area_struct *vmap;
758 
759 	rw_rlock(&linux_vma_lock);
760 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
761 		if (vmap->vm_private_data == handle)
762 			break;
763 	}
764 	rw_runlock(&linux_vma_lock);
765 	return (vmap);
766 }
767 
768 static int
769 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
770 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
771 {
772 
773 	MPASS(linux_cdev_handle_find(handle) != NULL);
774 	*color = 0;
775 	return (0);
776 }
777 
778 static void
779 linux_cdev_pager_dtor(void *handle)
780 {
781 	const struct vm_operations_struct *vm_ops;
782 	struct vm_area_struct *vmap;
783 
784 	vmap = linux_cdev_handle_find(handle);
785 	MPASS(vmap != NULL);
786 
787 	/*
788 	 * Remove handle before calling close operation to prevent
789 	 * other threads from reusing the handle pointer.
790 	 */
791 	linux_cdev_handle_remove(vmap);
792 
793 	down_write(&vmap->vm_mm->mmap_sem);
794 	vm_ops = vmap->vm_ops;
795 	if (likely(vm_ops != NULL))
796 		vm_ops->close(vmap);
797 	up_write(&vmap->vm_mm->mmap_sem);
798 
799 	linux_cdev_handle_free(vmap);
800 }
801 
802 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
803   {
804 	/* OBJT_MGTDEVICE */
805 	.cdev_pg_populate	= linux_cdev_pager_populate,
806 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
807 	.cdev_pg_dtor	= linux_cdev_pager_dtor
808   },
809   {
810 	/* OBJT_DEVICE */
811 	.cdev_pg_fault	= linux_cdev_pager_fault,
812 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
813 	.cdev_pg_dtor	= linux_cdev_pager_dtor
814   },
815 };
816 
817 int
818 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
819     unsigned long size)
820 {
821 	vm_object_t obj;
822 	vm_page_t m;
823 
824 	obj = vma->vm_obj;
825 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
826 		return (-ENOTSUP);
827 	VM_OBJECT_RLOCK(obj);
828 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
829 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
830 	    m = TAILQ_NEXT(m, listq))
831 		pmap_remove_all(m);
832 	VM_OBJECT_RUNLOCK(obj);
833 	return (0);
834 }
835 
836 static struct file_operations dummy_ldev_ops = {
837 	/* XXXKIB */
838 };
839 
840 static struct linux_cdev dummy_ldev = {
841 	.ops = &dummy_ldev_ops,
842 };
843 
844 #define	LDEV_SI_DTR	0x0001
845 #define	LDEV_SI_REF	0x0002
846 
847 static void
848 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
849     struct linux_cdev **dev)
850 {
851 	struct linux_cdev *ldev;
852 	u_int siref;
853 
854 	ldev = filp->f_cdev;
855 	*fop = filp->f_op;
856 	if (ldev != NULL) {
857 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
858 			refcount_acquire(&ldev->refs);
859 		} else {
860 			for (siref = ldev->siref;;) {
861 				if ((siref & LDEV_SI_DTR) != 0) {
862 					ldev = &dummy_ldev;
863 					*fop = ldev->ops;
864 					siref = ldev->siref;
865 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
866 				} else if (atomic_fcmpset_int(&ldev->siref,
867 				    &siref, siref + LDEV_SI_REF)) {
868 					break;
869 				}
870 			}
871 		}
872 	}
873 	*dev = ldev;
874 }
875 
876 static void
877 linux_drop_fop(struct linux_cdev *ldev)
878 {
879 
880 	if (ldev == NULL)
881 		return;
882 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
883 		linux_cdev_deref(ldev);
884 	} else {
885 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
886 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
887 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
888 	}
889 }
890 
891 #define	OPW(fp,td,code) ({			\
892 	struct file *__fpop;			\
893 	__typeof(code) __retval;		\
894 						\
895 	__fpop = (td)->td_fpop;			\
896 	(td)->td_fpop = (fp);			\
897 	__retval = (code);			\
898 	(td)->td_fpop = __fpop;			\
899 	__retval;				\
900 })
901 
902 static int
903 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
904     struct file *file)
905 {
906 	struct linux_cdev *ldev;
907 	struct linux_file *filp;
908 	const struct file_operations *fop;
909 	int error;
910 
911 	ldev = dev->si_drv1;
912 
913 	filp = linux_file_alloc();
914 	filp->f_dentry = &filp->f_dentry_store;
915 	filp->f_op = ldev->ops;
916 	filp->f_mode = file->f_flag;
917 	filp->f_flags = file->f_flag;
918 	filp->f_vnode = file->f_vnode;
919 	filp->_file = file;
920 	refcount_acquire(&ldev->refs);
921 	filp->f_cdev = ldev;
922 
923 	linux_set_current(td);
924 	linux_get_fop(filp, &fop, &ldev);
925 
926 	if (fop->open != NULL) {
927 		error = -fop->open(file->f_vnode, filp);
928 		if (error != 0) {
929 			linux_drop_fop(ldev);
930 			linux_cdev_deref(filp->f_cdev);
931 			kfree(filp);
932 			return (error);
933 		}
934 	}
935 
936 	/* hold on to the vnode - used for fstat() */
937 	vhold(filp->f_vnode);
938 
939 	/* release the file from devfs */
940 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
941 	linux_drop_fop(ldev);
942 	return (ENXIO);
943 }
944 
945 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
946 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
947 
948 static inline int
949 linux_remap_address(void **uaddr, size_t len)
950 {
951 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
952 
953 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
954 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
955 		struct task_struct *pts = current;
956 		if (pts == NULL) {
957 			*uaddr = NULL;
958 			return (1);
959 		}
960 
961 		/* compute data offset */
962 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
963 
964 		/* check that length is within bounds */
965 		if ((len > IOCPARM_MAX) ||
966 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
967 			*uaddr = NULL;
968 			return (1);
969 		}
970 
971 		/* re-add kernel buffer address */
972 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
973 
974 		/* update address location */
975 		*uaddr = (void *)uaddr_val;
976 		return (1);
977 	}
978 	return (0);
979 }
980 
981 int
982 linux_copyin(const void *uaddr, void *kaddr, size_t len)
983 {
984 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
985 		if (uaddr == NULL)
986 			return (-EFAULT);
987 		memcpy(kaddr, uaddr, len);
988 		return (0);
989 	}
990 	return (-copyin(uaddr, kaddr, len));
991 }
992 
993 int
994 linux_copyout(const void *kaddr, void *uaddr, size_t len)
995 {
996 	if (linux_remap_address(&uaddr, len)) {
997 		if (uaddr == NULL)
998 			return (-EFAULT);
999 		memcpy(uaddr, kaddr, len);
1000 		return (0);
1001 	}
1002 	return (-copyout(kaddr, uaddr, len));
1003 }
1004 
1005 size_t
1006 linux_clear_user(void *_uaddr, size_t _len)
1007 {
1008 	uint8_t *uaddr = _uaddr;
1009 	size_t len = _len;
1010 
1011 	/* make sure uaddr is aligned before going into the fast loop */
1012 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
1013 		if (subyte(uaddr, 0))
1014 			return (_len);
1015 		uaddr++;
1016 		len--;
1017 	}
1018 
1019 	/* zero 8 bytes at a time */
1020 	while (len > 7) {
1021 #ifdef __LP64__
1022 		if (suword64(uaddr, 0))
1023 			return (_len);
1024 #else
1025 		if (suword32(uaddr, 0))
1026 			return (_len);
1027 		if (suword32(uaddr + 4, 0))
1028 			return (_len);
1029 #endif
1030 		uaddr += 8;
1031 		len -= 8;
1032 	}
1033 
1034 	/* zero fill end, if any */
1035 	while (len > 0) {
1036 		if (subyte(uaddr, 0))
1037 			return (_len);
1038 		uaddr++;
1039 		len--;
1040 	}
1041 	return (0);
1042 }
1043 
1044 int
1045 linux_access_ok(const void *uaddr, size_t len)
1046 {
1047 	uintptr_t saddr;
1048 	uintptr_t eaddr;
1049 
1050 	/* get start and end address */
1051 	saddr = (uintptr_t)uaddr;
1052 	eaddr = (uintptr_t)uaddr + len;
1053 
1054 	/* verify addresses are valid for userspace */
1055 	return ((saddr == eaddr) ||
1056 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
1057 }
1058 
1059 /*
1060  * This function should return either EINTR or ERESTART depending on
1061  * the signal type sent to this thread:
1062  */
1063 static int
1064 linux_get_error(struct task_struct *task, int error)
1065 {
1066 	/* check for signal type interrupt code */
1067 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
1068 		error = -linux_schedule_get_interrupt_value(task);
1069 		if (error == 0)
1070 			error = EINTR;
1071 	}
1072 	return (error);
1073 }
1074 
1075 static int
1076 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
1077     const struct file_operations *fop, u_long cmd, caddr_t data,
1078     struct thread *td)
1079 {
1080 	struct task_struct *task = current;
1081 	unsigned size;
1082 	int error;
1083 
1084 	size = IOCPARM_LEN(cmd);
1085 	/* refer to logic in sys_ioctl() */
1086 	if (size > 0) {
1087 		/*
1088 		 * Setup hint for linux_copyin() and linux_copyout().
1089 		 *
1090 		 * Background: Linux code expects a user-space address
1091 		 * while FreeBSD supplies a kernel-space address.
1092 		 */
1093 		task->bsd_ioctl_data = data;
1094 		task->bsd_ioctl_len = size;
1095 		data = (void *)LINUX_IOCTL_MIN_PTR;
1096 	} else {
1097 		/* fetch user-space pointer */
1098 		data = *(void **)data;
1099 	}
1100 #ifdef COMPAT_FREEBSD32
1101 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1102 		/* try the compat IOCTL handler first */
1103 		if (fop->compat_ioctl != NULL) {
1104 			error = -OPW(fp, td, fop->compat_ioctl(filp,
1105 			    cmd, (u_long)data));
1106 		} else {
1107 			error = ENOTTY;
1108 		}
1109 
1110 		/* fallback to the regular IOCTL handler, if any */
1111 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
1112 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
1113 			    cmd, (u_long)data));
1114 		}
1115 	} else
1116 #endif
1117 	{
1118 		if (fop->unlocked_ioctl != NULL) {
1119 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
1120 			    cmd, (u_long)data));
1121 		} else {
1122 			error = ENOTTY;
1123 		}
1124 	}
1125 	if (size > 0) {
1126 		task->bsd_ioctl_data = NULL;
1127 		task->bsd_ioctl_len = 0;
1128 	}
1129 
1130 	if (error == EWOULDBLOCK) {
1131 		/* update kqfilter status, if any */
1132 		linux_file_kqfilter_poll(filp,
1133 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1134 	} else {
1135 		error = linux_get_error(task, error);
1136 	}
1137 	return (error);
1138 }
1139 
1140 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
1141 
1142 /*
1143  * This function atomically updates the poll wakeup state and returns
1144  * the previous state at the time of update.
1145  */
1146 static uint8_t
1147 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1148 {
1149 	int c, old;
1150 
1151 	c = v->counter;
1152 
1153 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1154 		c = old;
1155 
1156 	return (c);
1157 }
1158 
1159 static int
1160 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1161 {
1162 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1163 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1164 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1165 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1166 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1167 	};
1168 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1169 
1170 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1171 	case LINUX_FWQ_STATE_QUEUED:
1172 		linux_poll_wakeup(filp);
1173 		return (1);
1174 	default:
1175 		return (0);
1176 	}
1177 }
1178 
1179 void
1180 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1181 {
1182 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1183 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1184 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1185 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1186 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1187 	};
1188 
1189 	/* check if we are called inside the select system call */
1190 	if (p == LINUX_POLL_TABLE_NORMAL)
1191 		selrecord(curthread, &filp->f_selinfo);
1192 
1193 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1194 	case LINUX_FWQ_STATE_INIT:
1195 		/* NOTE: file handles can only belong to one wait-queue */
1196 		filp->f_wait_queue.wqh = wqh;
1197 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1198 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1199 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1200 		break;
1201 	default:
1202 		break;
1203 	}
1204 }
1205 
1206 static void
1207 linux_poll_wait_dequeue(struct linux_file *filp)
1208 {
1209 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1210 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1211 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1212 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1213 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1214 	};
1215 
1216 	seldrain(&filp->f_selinfo);
1217 
1218 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1219 	case LINUX_FWQ_STATE_NOT_READY:
1220 	case LINUX_FWQ_STATE_QUEUED:
1221 	case LINUX_FWQ_STATE_READY:
1222 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1223 		break;
1224 	default:
1225 		break;
1226 	}
1227 }
1228 
1229 void
1230 linux_poll_wakeup(struct linux_file *filp)
1231 {
1232 	/* this function should be NULL-safe */
1233 	if (filp == NULL)
1234 		return;
1235 
1236 	selwakeup(&filp->f_selinfo);
1237 
1238 	spin_lock(&filp->f_kqlock);
1239 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1240 	    LINUX_KQ_FLAG_NEED_WRITE;
1241 
1242 	/* make sure the "knote" gets woken up */
1243 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1244 	spin_unlock(&filp->f_kqlock);
1245 }
1246 
1247 static void
1248 linux_file_kqfilter_detach(struct knote *kn)
1249 {
1250 	struct linux_file *filp = kn->kn_hook;
1251 
1252 	spin_lock(&filp->f_kqlock);
1253 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1254 	spin_unlock(&filp->f_kqlock);
1255 }
1256 
1257 static int
1258 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1259 {
1260 	struct linux_file *filp = kn->kn_hook;
1261 
1262 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1263 
1264 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1265 }
1266 
1267 static int
1268 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1269 {
1270 	struct linux_file *filp = kn->kn_hook;
1271 
1272 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1273 
1274 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1275 }
1276 
1277 static struct filterops linux_dev_kqfiltops_read = {
1278 	.f_isfd = 1,
1279 	.f_detach = linux_file_kqfilter_detach,
1280 	.f_event = linux_file_kqfilter_read_event,
1281 };
1282 
1283 static struct filterops linux_dev_kqfiltops_write = {
1284 	.f_isfd = 1,
1285 	.f_detach = linux_file_kqfilter_detach,
1286 	.f_event = linux_file_kqfilter_write_event,
1287 };
1288 
1289 static void
1290 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1291 {
1292 	struct thread *td;
1293 	const struct file_operations *fop;
1294 	struct linux_cdev *ldev;
1295 	int temp;
1296 
1297 	if ((filp->f_kqflags & kqflags) == 0)
1298 		return;
1299 
1300 	td = curthread;
1301 
1302 	linux_get_fop(filp, &fop, &ldev);
1303 	/* get the latest polling state */
1304 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1305 	linux_drop_fop(ldev);
1306 
1307 	spin_lock(&filp->f_kqlock);
1308 	/* clear kqflags */
1309 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1310 	    LINUX_KQ_FLAG_NEED_WRITE);
1311 	/* update kqflags */
1312 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1313 		if ((temp & POLLIN) != 0)
1314 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1315 		if ((temp & POLLOUT) != 0)
1316 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1317 
1318 		/* make sure the "knote" gets woken up */
1319 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1320 	}
1321 	spin_unlock(&filp->f_kqlock);
1322 }
1323 
1324 static int
1325 linux_file_kqfilter(struct file *file, struct knote *kn)
1326 {
1327 	struct linux_file *filp;
1328 	struct thread *td;
1329 	int error;
1330 
1331 	td = curthread;
1332 	filp = (struct linux_file *)file->f_data;
1333 	filp->f_flags = file->f_flag;
1334 	if (filp->f_op->poll == NULL)
1335 		return (EINVAL);
1336 
1337 	spin_lock(&filp->f_kqlock);
1338 	switch (kn->kn_filter) {
1339 	case EVFILT_READ:
1340 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1341 		kn->kn_fop = &linux_dev_kqfiltops_read;
1342 		kn->kn_hook = filp;
1343 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1344 		error = 0;
1345 		break;
1346 	case EVFILT_WRITE:
1347 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1348 		kn->kn_fop = &linux_dev_kqfiltops_write;
1349 		kn->kn_hook = filp;
1350 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1351 		error = 0;
1352 		break;
1353 	default:
1354 		error = EINVAL;
1355 		break;
1356 	}
1357 	spin_unlock(&filp->f_kqlock);
1358 
1359 	if (error == 0) {
1360 		linux_set_current(td);
1361 
1362 		/* update kqfilter status, if any */
1363 		linux_file_kqfilter_poll(filp,
1364 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1365 	}
1366 	return (error);
1367 }
1368 
1369 static int
1370 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1371     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1372     int nprot, bool is_shared, struct thread *td)
1373 {
1374 	struct task_struct *task;
1375 	struct vm_area_struct *vmap;
1376 	struct mm_struct *mm;
1377 	struct linux_file *filp;
1378 	vm_memattr_t attr;
1379 	int error;
1380 
1381 	filp = (struct linux_file *)fp->f_data;
1382 	filp->f_flags = fp->f_flag;
1383 
1384 	if (fop->mmap == NULL)
1385 		return (EOPNOTSUPP);
1386 
1387 	linux_set_current(td);
1388 
1389 	/*
1390 	 * The same VM object might be shared by multiple processes
1391 	 * and the mm_struct is usually freed when a process exits.
1392 	 *
1393 	 * The atomic reference below makes sure the mm_struct is
1394 	 * available as long as the vmap is in the linux_vma_head.
1395 	 */
1396 	task = current;
1397 	mm = task->mm;
1398 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1399 		return (EINVAL);
1400 
1401 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1402 	vmap->vm_start = 0;
1403 	vmap->vm_end = size;
1404 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1405 	vmap->vm_pfn = 0;
1406 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1407 	if (is_shared)
1408 		vmap->vm_flags |= VM_SHARED;
1409 	vmap->vm_ops = NULL;
1410 	vmap->vm_file = get_file(filp);
1411 	vmap->vm_mm = mm;
1412 
1413 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1414 		error = linux_get_error(task, EINTR);
1415 	} else {
1416 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1417 		error = linux_get_error(task, error);
1418 		up_write(&vmap->vm_mm->mmap_sem);
1419 	}
1420 
1421 	if (error != 0) {
1422 		linux_cdev_handle_free(vmap);
1423 		return (error);
1424 	}
1425 
1426 	attr = pgprot2cachemode(vmap->vm_page_prot);
1427 
1428 	if (vmap->vm_ops != NULL) {
1429 		struct vm_area_struct *ptr;
1430 		void *vm_private_data;
1431 		bool vm_no_fault;
1432 
1433 		if (vmap->vm_ops->open == NULL ||
1434 		    vmap->vm_ops->close == NULL ||
1435 		    vmap->vm_private_data == NULL) {
1436 			/* free allocated VM area struct */
1437 			linux_cdev_handle_free(vmap);
1438 			return (EINVAL);
1439 		}
1440 
1441 		vm_private_data = vmap->vm_private_data;
1442 
1443 		rw_wlock(&linux_vma_lock);
1444 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1445 			if (ptr->vm_private_data == vm_private_data)
1446 				break;
1447 		}
1448 		/* check if there is an existing VM area struct */
1449 		if (ptr != NULL) {
1450 			/* check if the VM area structure is invalid */
1451 			if (ptr->vm_ops == NULL ||
1452 			    ptr->vm_ops->open == NULL ||
1453 			    ptr->vm_ops->close == NULL) {
1454 				error = ESTALE;
1455 				vm_no_fault = 1;
1456 			} else {
1457 				error = EEXIST;
1458 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1459 			}
1460 		} else {
1461 			/* insert VM area structure into list */
1462 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1463 			error = 0;
1464 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1465 		}
1466 		rw_wunlock(&linux_vma_lock);
1467 
1468 		if (error != 0) {
1469 			/* free allocated VM area struct */
1470 			linux_cdev_handle_free(vmap);
1471 			/* check for stale VM area struct */
1472 			if (error != EEXIST)
1473 				return (error);
1474 		}
1475 
1476 		/* check if there is no fault handler */
1477 		if (vm_no_fault) {
1478 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1479 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1480 			    td->td_ucred);
1481 		} else {
1482 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1483 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1484 			    td->td_ucred);
1485 		}
1486 
1487 		/* check if allocating the VM object failed */
1488 		if (*object == NULL) {
1489 			if (error == 0) {
1490 				/* remove VM area struct from list */
1491 				linux_cdev_handle_remove(vmap);
1492 				/* free allocated VM area struct */
1493 				linux_cdev_handle_free(vmap);
1494 			}
1495 			return (EINVAL);
1496 		}
1497 	} else {
1498 		struct sglist *sg;
1499 
1500 		sg = sglist_alloc(1, M_WAITOK);
1501 		sglist_append_phys(sg,
1502 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1503 
1504 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1505 		    nprot, 0, td->td_ucred);
1506 
1507 		linux_cdev_handle_free(vmap);
1508 
1509 		if (*object == NULL) {
1510 			sglist_free(sg);
1511 			return (EINVAL);
1512 		}
1513 	}
1514 
1515 	if (attr != VM_MEMATTR_DEFAULT) {
1516 		VM_OBJECT_WLOCK(*object);
1517 		vm_object_set_memattr(*object, attr);
1518 		VM_OBJECT_WUNLOCK(*object);
1519 	}
1520 	*offset = 0;
1521 	return (0);
1522 }
1523 
1524 struct cdevsw linuxcdevsw = {
1525 	.d_version = D_VERSION,
1526 	.d_fdopen = linux_dev_fdopen,
1527 	.d_name = "lkpidev",
1528 };
1529 
1530 static int
1531 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1532     int flags, struct thread *td)
1533 {
1534 	struct linux_file *filp;
1535 	const struct file_operations *fop;
1536 	struct linux_cdev *ldev;
1537 	ssize_t bytes;
1538 	int error;
1539 
1540 	error = 0;
1541 	filp = (struct linux_file *)file->f_data;
1542 	filp->f_flags = file->f_flag;
1543 	/* XXX no support for I/O vectors currently */
1544 	if (uio->uio_iovcnt != 1)
1545 		return (EOPNOTSUPP);
1546 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1547 		return (EINVAL);
1548 	linux_set_current(td);
1549 	linux_get_fop(filp, &fop, &ldev);
1550 	if (fop->read != NULL) {
1551 		bytes = OPW(file, td, fop->read(filp,
1552 		    uio->uio_iov->iov_base,
1553 		    uio->uio_iov->iov_len, &uio->uio_offset));
1554 		if (bytes >= 0) {
1555 			uio->uio_iov->iov_base =
1556 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1557 			uio->uio_iov->iov_len -= bytes;
1558 			uio->uio_resid -= bytes;
1559 		} else {
1560 			error = linux_get_error(current, -bytes);
1561 		}
1562 	} else
1563 		error = ENXIO;
1564 
1565 	/* update kqfilter status, if any */
1566 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1567 	linux_drop_fop(ldev);
1568 
1569 	return (error);
1570 }
1571 
1572 static int
1573 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1574     int flags, struct thread *td)
1575 {
1576 	struct linux_file *filp;
1577 	const struct file_operations *fop;
1578 	struct linux_cdev *ldev;
1579 	ssize_t bytes;
1580 	int error;
1581 
1582 	filp = (struct linux_file *)file->f_data;
1583 	filp->f_flags = file->f_flag;
1584 	/* XXX no support for I/O vectors currently */
1585 	if (uio->uio_iovcnt != 1)
1586 		return (EOPNOTSUPP);
1587 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1588 		return (EINVAL);
1589 	linux_set_current(td);
1590 	linux_get_fop(filp, &fop, &ldev);
1591 	if (fop->write != NULL) {
1592 		bytes = OPW(file, td, fop->write(filp,
1593 		    uio->uio_iov->iov_base,
1594 		    uio->uio_iov->iov_len, &uio->uio_offset));
1595 		if (bytes >= 0) {
1596 			uio->uio_iov->iov_base =
1597 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1598 			uio->uio_iov->iov_len -= bytes;
1599 			uio->uio_resid -= bytes;
1600 			error = 0;
1601 		} else {
1602 			error = linux_get_error(current, -bytes);
1603 		}
1604 	} else
1605 		error = ENXIO;
1606 
1607 	/* update kqfilter status, if any */
1608 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1609 
1610 	linux_drop_fop(ldev);
1611 
1612 	return (error);
1613 }
1614 
1615 static int
1616 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1617     struct thread *td)
1618 {
1619 	struct linux_file *filp;
1620 	const struct file_operations *fop;
1621 	struct linux_cdev *ldev;
1622 	int revents;
1623 
1624 	filp = (struct linux_file *)file->f_data;
1625 	filp->f_flags = file->f_flag;
1626 	linux_set_current(td);
1627 	linux_get_fop(filp, &fop, &ldev);
1628 	if (fop->poll != NULL) {
1629 		revents = OPW(file, td, fop->poll(filp,
1630 		    LINUX_POLL_TABLE_NORMAL)) & events;
1631 	} else {
1632 		revents = 0;
1633 	}
1634 	linux_drop_fop(ldev);
1635 	return (revents);
1636 }
1637 
1638 static int
1639 linux_file_close(struct file *file, struct thread *td)
1640 {
1641 	struct linux_file *filp;
1642 	int (*release)(struct inode *, struct linux_file *);
1643 	const struct file_operations *fop;
1644 	struct linux_cdev *ldev;
1645 	int error;
1646 
1647 	filp = (struct linux_file *)file->f_data;
1648 
1649 	KASSERT(file_count(filp) == 0,
1650 	    ("File refcount(%d) is not zero", file_count(filp)));
1651 
1652 	if (td == NULL)
1653 		td = curthread;
1654 
1655 	error = 0;
1656 	filp->f_flags = file->f_flag;
1657 	linux_set_current(td);
1658 	linux_poll_wait_dequeue(filp);
1659 	linux_get_fop(filp, &fop, &ldev);
1660 	/*
1661 	 * Always use the real release function, if any, to avoid
1662 	 * leaking device resources:
1663 	 */
1664 	release = filp->f_op->release;
1665 	if (release != NULL)
1666 		error = -OPW(file, td, release(filp->f_vnode, filp));
1667 	funsetown(&filp->f_sigio);
1668 	if (filp->f_vnode != NULL)
1669 		vdrop(filp->f_vnode);
1670 	linux_drop_fop(ldev);
1671 	ldev = filp->f_cdev;
1672 	if (ldev != NULL)
1673 		linux_cdev_deref(ldev);
1674 	linux_synchronize_rcu(RCU_TYPE_REGULAR);
1675 	kfree(filp);
1676 
1677 	return (error);
1678 }
1679 
1680 static int
1681 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1682     struct thread *td)
1683 {
1684 	struct linux_file *filp;
1685 	const struct file_operations *fop;
1686 	struct linux_cdev *ldev;
1687 	struct fiodgname_arg *fgn;
1688 	const char *p;
1689 	int error, i;
1690 
1691 	error = 0;
1692 	filp = (struct linux_file *)fp->f_data;
1693 	filp->f_flags = fp->f_flag;
1694 	linux_get_fop(filp, &fop, &ldev);
1695 
1696 	linux_set_current(td);
1697 	switch (cmd) {
1698 	case FIONBIO:
1699 		break;
1700 	case FIOASYNC:
1701 		if (fop->fasync == NULL)
1702 			break;
1703 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1704 		break;
1705 	case FIOSETOWN:
1706 		error = fsetown(*(int *)data, &filp->f_sigio);
1707 		if (error == 0) {
1708 			if (fop->fasync == NULL)
1709 				break;
1710 			error = -OPW(fp, td, fop->fasync(0, filp,
1711 			    fp->f_flag & FASYNC));
1712 		}
1713 		break;
1714 	case FIOGETOWN:
1715 		*(int *)data = fgetown(&filp->f_sigio);
1716 		break;
1717 	case FIODGNAME:
1718 #ifdef	COMPAT_FREEBSD32
1719 	case FIODGNAME_32:
1720 #endif
1721 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1722 			error = ENXIO;
1723 			break;
1724 		}
1725 		fgn = data;
1726 		p = devtoname(filp->f_cdev->cdev);
1727 		i = strlen(p) + 1;
1728 		if (i > fgn->len) {
1729 			error = EINVAL;
1730 			break;
1731 		}
1732 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1733 		break;
1734 	default:
1735 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1736 		break;
1737 	}
1738 	linux_drop_fop(ldev);
1739 	return (error);
1740 }
1741 
1742 static int
1743 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1744     vm_prot_t maxprot, int flags, struct file *fp,
1745     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1746 {
1747 	/*
1748 	 * Character devices do not provide private mappings
1749 	 * of any kind:
1750 	 */
1751 	if ((maxprot & VM_PROT_WRITE) == 0 &&
1752 	    (prot & VM_PROT_WRITE) != 0)
1753 		return (EACCES);
1754 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1755 		return (EINVAL);
1756 
1757 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1758 	    (int)prot, (flags & MAP_SHARED) ? true : false, td));
1759 }
1760 
1761 static int
1762 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1763     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1764     struct thread *td)
1765 {
1766 	struct linux_file *filp;
1767 	const struct file_operations *fop;
1768 	struct linux_cdev *ldev;
1769 	struct mount *mp;
1770 	struct vnode *vp;
1771 	vm_object_t object;
1772 	vm_prot_t maxprot;
1773 	int error;
1774 
1775 	filp = (struct linux_file *)fp->f_data;
1776 
1777 	vp = filp->f_vnode;
1778 	if (vp == NULL)
1779 		return (EOPNOTSUPP);
1780 
1781 	/*
1782 	 * Ensure that file and memory protections are
1783 	 * compatible.
1784 	 */
1785 	mp = vp->v_mount;
1786 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1787 		maxprot = VM_PROT_NONE;
1788 		if ((prot & VM_PROT_EXECUTE) != 0)
1789 			return (EACCES);
1790 	} else
1791 		maxprot = VM_PROT_EXECUTE;
1792 	if ((fp->f_flag & FREAD) != 0)
1793 		maxprot |= VM_PROT_READ;
1794 	else if ((prot & VM_PROT_READ) != 0)
1795 		return (EACCES);
1796 
1797 	/*
1798 	 * If we are sharing potential changes via MAP_SHARED and we
1799 	 * are trying to get write permission although we opened it
1800 	 * without asking for it, bail out.
1801 	 *
1802 	 * Note that most character devices always share mappings.
1803 	 *
1804 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1805 	 * requests rather than doing it here.
1806 	 */
1807 	if ((flags & MAP_SHARED) != 0) {
1808 		if ((fp->f_flag & FWRITE) != 0)
1809 			maxprot |= VM_PROT_WRITE;
1810 		else if ((prot & VM_PROT_WRITE) != 0)
1811 			return (EACCES);
1812 	}
1813 	maxprot &= cap_maxprot;
1814 
1815 	linux_get_fop(filp, &fop, &ldev);
1816 	error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1817 	    &foff, fop, &object);
1818 	if (error != 0)
1819 		goto out;
1820 
1821 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1822 	    foff, FALSE, td);
1823 	if (error != 0)
1824 		vm_object_deallocate(object);
1825 out:
1826 	linux_drop_fop(ldev);
1827 	return (error);
1828 }
1829 
1830 static int
1831 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1832 {
1833 	struct linux_file *filp;
1834 	struct vnode *vp;
1835 	int error;
1836 
1837 	filp = (struct linux_file *)fp->f_data;
1838 	if (filp->f_vnode == NULL)
1839 		return (EOPNOTSUPP);
1840 
1841 	vp = filp->f_vnode;
1842 
1843 	vn_lock(vp, LK_SHARED | LK_RETRY);
1844 	error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1845 	VOP_UNLOCK(vp);
1846 
1847 	return (error);
1848 }
1849 
1850 static int
1851 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1852     struct filedesc *fdp)
1853 {
1854 	struct linux_file *filp;
1855 	struct vnode *vp;
1856 	int error;
1857 
1858 	filp = fp->f_data;
1859 	vp = filp->f_vnode;
1860 	if (vp == NULL) {
1861 		error = 0;
1862 		kif->kf_type = KF_TYPE_DEV;
1863 	} else {
1864 		vref(vp);
1865 		FILEDESC_SUNLOCK(fdp);
1866 		error = vn_fill_kinfo_vnode(vp, kif);
1867 		vrele(vp);
1868 		kif->kf_type = KF_TYPE_VNODE;
1869 		FILEDESC_SLOCK(fdp);
1870 	}
1871 	return (error);
1872 }
1873 
1874 unsigned int
1875 linux_iminor(struct inode *inode)
1876 {
1877 	struct linux_cdev *ldev;
1878 
1879 	if (inode == NULL || inode->v_rdev == NULL ||
1880 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1881 		return (-1U);
1882 	ldev = inode->v_rdev->si_drv1;
1883 	if (ldev == NULL)
1884 		return (-1U);
1885 
1886 	return (minor(ldev->dev));
1887 }
1888 
1889 struct fileops linuxfileops = {
1890 	.fo_read = linux_file_read,
1891 	.fo_write = linux_file_write,
1892 	.fo_truncate = invfo_truncate,
1893 	.fo_kqfilter = linux_file_kqfilter,
1894 	.fo_stat = linux_file_stat,
1895 	.fo_fill_kinfo = linux_file_fill_kinfo,
1896 	.fo_poll = linux_file_poll,
1897 	.fo_close = linux_file_close,
1898 	.fo_ioctl = linux_file_ioctl,
1899 	.fo_mmap = linux_file_mmap,
1900 	.fo_chmod = invfo_chmod,
1901 	.fo_chown = invfo_chown,
1902 	.fo_sendfile = invfo_sendfile,
1903 	.fo_flags = DFLAG_PASSABLE,
1904 };
1905 
1906 /*
1907  * Hash of vmmap addresses.  This is infrequently accessed and does not
1908  * need to be particularly large.  This is done because we must store the
1909  * caller's idea of the map size to properly unmap.
1910  */
1911 struct vmmap {
1912 	LIST_ENTRY(vmmap)	vm_next;
1913 	void 			*vm_addr;
1914 	unsigned long		vm_size;
1915 };
1916 
1917 struct vmmaphd {
1918 	struct vmmap *lh_first;
1919 };
1920 #define	VMMAP_HASH_SIZE	64
1921 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1922 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1923 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1924 static struct mtx vmmaplock;
1925 
1926 static void
1927 vmmap_add(void *addr, unsigned long size)
1928 {
1929 	struct vmmap *vmmap;
1930 
1931 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1932 	mtx_lock(&vmmaplock);
1933 	vmmap->vm_size = size;
1934 	vmmap->vm_addr = addr;
1935 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1936 	mtx_unlock(&vmmaplock);
1937 }
1938 
1939 static struct vmmap *
1940 vmmap_remove(void *addr)
1941 {
1942 	struct vmmap *vmmap;
1943 
1944 	mtx_lock(&vmmaplock);
1945 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1946 		if (vmmap->vm_addr == addr)
1947 			break;
1948 	if (vmmap)
1949 		LIST_REMOVE(vmmap, vm_next);
1950 	mtx_unlock(&vmmaplock);
1951 
1952 	return (vmmap);
1953 }
1954 
1955 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1956 void *
1957 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1958 {
1959 	void *addr;
1960 
1961 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1962 	if (addr == NULL)
1963 		return (NULL);
1964 	vmmap_add(addr, size);
1965 
1966 	return (addr);
1967 }
1968 #endif
1969 
1970 void
1971 iounmap(void *addr)
1972 {
1973 	struct vmmap *vmmap;
1974 
1975 	vmmap = vmmap_remove(addr);
1976 	if (vmmap == NULL)
1977 		return;
1978 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1979 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1980 #endif
1981 	kfree(vmmap);
1982 }
1983 
1984 void *
1985 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1986 {
1987 	vm_offset_t off;
1988 	size_t size;
1989 
1990 	size = count * PAGE_SIZE;
1991 	off = kva_alloc(size);
1992 	if (off == 0)
1993 		return (NULL);
1994 	vmmap_add((void *)off, size);
1995 	pmap_qenter(off, pages, count);
1996 
1997 	return ((void *)off);
1998 }
1999 
2000 void
2001 vunmap(void *addr)
2002 {
2003 	struct vmmap *vmmap;
2004 
2005 	vmmap = vmmap_remove(addr);
2006 	if (vmmap == NULL)
2007 		return;
2008 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
2009 	kva_free((vm_offset_t)addr, vmmap->vm_size);
2010 	kfree(vmmap);
2011 }
2012 
2013 static char *
2014 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
2015 {
2016 	unsigned int len;
2017 	char *p;
2018 	va_list aq;
2019 
2020 	va_copy(aq, ap);
2021 	len = vsnprintf(NULL, 0, fmt, aq);
2022 	va_end(aq);
2023 
2024 	if (dev != NULL)
2025 		p = devm_kmalloc(dev, len + 1, gfp);
2026 	else
2027 		p = kmalloc(len + 1, gfp);
2028 	if (p != NULL)
2029 		vsnprintf(p, len + 1, fmt, ap);
2030 
2031 	return (p);
2032 }
2033 
2034 char *
2035 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
2036 {
2037 
2038 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
2039 }
2040 
2041 char *
2042 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
2043 {
2044 	va_list ap;
2045 	char *p;
2046 
2047 	va_start(ap, fmt);
2048 	p = devm_kvasprintf(dev, gfp, fmt, ap);
2049 	va_end(ap);
2050 
2051 	return (p);
2052 }
2053 
2054 char *
2055 kasprintf(gfp_t gfp, const char *fmt, ...)
2056 {
2057 	va_list ap;
2058 	char *p;
2059 
2060 	va_start(ap, fmt);
2061 	p = kvasprintf(gfp, fmt, ap);
2062 	va_end(ap);
2063 
2064 	return (p);
2065 }
2066 
2067 static void
2068 linux_timer_callback_wrapper(void *context)
2069 {
2070 	struct timer_list *timer;
2071 
2072 	timer = context;
2073 
2074 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
2075 		/* try again later */
2076 		callout_reset(&timer->callout, 1,
2077 		    &linux_timer_callback_wrapper, timer);
2078 		return;
2079 	}
2080 
2081 	timer->function(timer->data);
2082 }
2083 
2084 int
2085 mod_timer(struct timer_list *timer, int expires)
2086 {
2087 	int ret;
2088 
2089 	timer->expires = expires;
2090 	ret = callout_reset(&timer->callout,
2091 	    linux_timer_jiffies_until(expires),
2092 	    &linux_timer_callback_wrapper, timer);
2093 
2094 	MPASS(ret == 0 || ret == 1);
2095 
2096 	return (ret == 1);
2097 }
2098 
2099 void
2100 add_timer(struct timer_list *timer)
2101 {
2102 
2103 	callout_reset(&timer->callout,
2104 	    linux_timer_jiffies_until(timer->expires),
2105 	    &linux_timer_callback_wrapper, timer);
2106 }
2107 
2108 void
2109 add_timer_on(struct timer_list *timer, int cpu)
2110 {
2111 
2112 	callout_reset_on(&timer->callout,
2113 	    linux_timer_jiffies_until(timer->expires),
2114 	    &linux_timer_callback_wrapper, timer, cpu);
2115 }
2116 
2117 int
2118 del_timer(struct timer_list *timer)
2119 {
2120 
2121 	if (callout_stop(&(timer)->callout) == -1)
2122 		return (0);
2123 	return (1);
2124 }
2125 
2126 int
2127 del_timer_sync(struct timer_list *timer)
2128 {
2129 
2130 	if (callout_drain(&(timer)->callout) == -1)
2131 		return (0);
2132 	return (1);
2133 }
2134 
2135 /* greatest common divisor, Euclid equation */
2136 static uint64_t
2137 lkpi_gcd_64(uint64_t a, uint64_t b)
2138 {
2139 	uint64_t an;
2140 	uint64_t bn;
2141 
2142 	while (b != 0) {
2143 		an = b;
2144 		bn = a % b;
2145 		a = an;
2146 		b = bn;
2147 	}
2148 	return (a);
2149 }
2150 
2151 uint64_t lkpi_nsec2hz_rem;
2152 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2153 uint64_t lkpi_nsec2hz_max;
2154 
2155 uint64_t lkpi_usec2hz_rem;
2156 uint64_t lkpi_usec2hz_div = 1000000ULL;
2157 uint64_t lkpi_usec2hz_max;
2158 
2159 uint64_t lkpi_msec2hz_rem;
2160 uint64_t lkpi_msec2hz_div = 1000ULL;
2161 uint64_t lkpi_msec2hz_max;
2162 
2163 static void
2164 linux_timer_init(void *arg)
2165 {
2166 	uint64_t gcd;
2167 
2168 	/*
2169 	 * Compute an internal HZ value which can divide 2**32 to
2170 	 * avoid timer rounding problems when the tick value wraps
2171 	 * around 2**32:
2172 	 */
2173 	linux_timer_hz_mask = 1;
2174 	while (linux_timer_hz_mask < (unsigned long)hz)
2175 		linux_timer_hz_mask *= 2;
2176 	linux_timer_hz_mask--;
2177 
2178 	/* compute some internal constants */
2179 
2180 	lkpi_nsec2hz_rem = hz;
2181 	lkpi_usec2hz_rem = hz;
2182 	lkpi_msec2hz_rem = hz;
2183 
2184 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2185 	lkpi_nsec2hz_rem /= gcd;
2186 	lkpi_nsec2hz_div /= gcd;
2187 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2188 
2189 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2190 	lkpi_usec2hz_rem /= gcd;
2191 	lkpi_usec2hz_div /= gcd;
2192 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2193 
2194 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2195 	lkpi_msec2hz_rem /= gcd;
2196 	lkpi_msec2hz_div /= gcd;
2197 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2198 }
2199 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2200 
2201 void
2202 linux_complete_common(struct completion *c, int all)
2203 {
2204 	int wakeup_swapper;
2205 
2206 	sleepq_lock(c);
2207 	if (all) {
2208 		c->done = UINT_MAX;
2209 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2210 	} else {
2211 		if (c->done != UINT_MAX)
2212 			c->done++;
2213 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2214 	}
2215 	sleepq_release(c);
2216 	if (wakeup_swapper)
2217 		kick_proc0();
2218 }
2219 
2220 /*
2221  * Indefinite wait for done != 0 with or without signals.
2222  */
2223 int
2224 linux_wait_for_common(struct completion *c, int flags)
2225 {
2226 	struct task_struct *task;
2227 	int error;
2228 
2229 	if (SCHEDULER_STOPPED())
2230 		return (0);
2231 
2232 	task = current;
2233 
2234 	if (flags != 0)
2235 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2236 	else
2237 		flags = SLEEPQ_SLEEP;
2238 	error = 0;
2239 	for (;;) {
2240 		sleepq_lock(c);
2241 		if (c->done)
2242 			break;
2243 		sleepq_add(c, NULL, "completion", flags, 0);
2244 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2245 			DROP_GIANT();
2246 			error = -sleepq_wait_sig(c, 0);
2247 			PICKUP_GIANT();
2248 			if (error != 0) {
2249 				linux_schedule_save_interrupt_value(task, error);
2250 				error = -ERESTARTSYS;
2251 				goto intr;
2252 			}
2253 		} else {
2254 			DROP_GIANT();
2255 			sleepq_wait(c, 0);
2256 			PICKUP_GIANT();
2257 		}
2258 	}
2259 	if (c->done != UINT_MAX)
2260 		c->done--;
2261 	sleepq_release(c);
2262 
2263 intr:
2264 	return (error);
2265 }
2266 
2267 /*
2268  * Time limited wait for done != 0 with or without signals.
2269  */
2270 int
2271 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2272 {
2273 	struct task_struct *task;
2274 	int end = jiffies + timeout;
2275 	int error;
2276 
2277 	if (SCHEDULER_STOPPED())
2278 		return (0);
2279 
2280 	task = current;
2281 
2282 	if (flags != 0)
2283 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2284 	else
2285 		flags = SLEEPQ_SLEEP;
2286 
2287 	for (;;) {
2288 		sleepq_lock(c);
2289 		if (c->done)
2290 			break;
2291 		sleepq_add(c, NULL, "completion", flags, 0);
2292 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2293 
2294 		DROP_GIANT();
2295 		if (flags & SLEEPQ_INTERRUPTIBLE)
2296 			error = -sleepq_timedwait_sig(c, 0);
2297 		else
2298 			error = -sleepq_timedwait(c, 0);
2299 		PICKUP_GIANT();
2300 
2301 		if (error != 0) {
2302 			/* check for timeout */
2303 			if (error == -EWOULDBLOCK) {
2304 				error = 0;	/* timeout */
2305 			} else {
2306 				/* signal happened */
2307 				linux_schedule_save_interrupt_value(task, error);
2308 				error = -ERESTARTSYS;
2309 			}
2310 			goto done;
2311 		}
2312 	}
2313 	if (c->done != UINT_MAX)
2314 		c->done--;
2315 	sleepq_release(c);
2316 
2317 	/* return how many jiffies are left */
2318 	error = linux_timer_jiffies_until(end);
2319 done:
2320 	return (error);
2321 }
2322 
2323 int
2324 linux_try_wait_for_completion(struct completion *c)
2325 {
2326 	int isdone;
2327 
2328 	sleepq_lock(c);
2329 	isdone = (c->done != 0);
2330 	if (c->done != 0 && c->done != UINT_MAX)
2331 		c->done--;
2332 	sleepq_release(c);
2333 	return (isdone);
2334 }
2335 
2336 int
2337 linux_completion_done(struct completion *c)
2338 {
2339 	int isdone;
2340 
2341 	sleepq_lock(c);
2342 	isdone = (c->done != 0);
2343 	sleepq_release(c);
2344 	return (isdone);
2345 }
2346 
2347 static void
2348 linux_cdev_deref(struct linux_cdev *ldev)
2349 {
2350 	if (refcount_release(&ldev->refs) &&
2351 	    ldev->kobj.ktype == &linux_cdev_ktype)
2352 		kfree(ldev);
2353 }
2354 
2355 static void
2356 linux_cdev_release(struct kobject *kobj)
2357 {
2358 	struct linux_cdev *cdev;
2359 	struct kobject *parent;
2360 
2361 	cdev = container_of(kobj, struct linux_cdev, kobj);
2362 	parent = kobj->parent;
2363 	linux_destroy_dev(cdev);
2364 	linux_cdev_deref(cdev);
2365 	kobject_put(parent);
2366 }
2367 
2368 static void
2369 linux_cdev_static_release(struct kobject *kobj)
2370 {
2371 	struct cdev *cdev;
2372 	struct linux_cdev *ldev;
2373 
2374 	ldev = container_of(kobj, struct linux_cdev, kobj);
2375 	cdev = ldev->cdev;
2376 	if (cdev != NULL) {
2377 		destroy_dev(cdev);
2378 		ldev->cdev = NULL;
2379 	}
2380 	kobject_put(kobj->parent);
2381 }
2382 
2383 int
2384 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2385 {
2386 	int ret;
2387 
2388 	if (dev->devt != 0) {
2389 		/* Set parent kernel object. */
2390 		ldev->kobj.parent = &dev->kobj;
2391 
2392 		/*
2393 		 * Unlike Linux we require the kobject of the
2394 		 * character device structure to have a valid name
2395 		 * before calling this function:
2396 		 */
2397 		if (ldev->kobj.name == NULL)
2398 			return (-EINVAL);
2399 
2400 		ret = cdev_add(ldev, dev->devt, 1);
2401 		if (ret)
2402 			return (ret);
2403 	}
2404 	ret = device_add(dev);
2405 	if (ret != 0 && dev->devt != 0)
2406 		cdev_del(ldev);
2407 	return (ret);
2408 }
2409 
2410 void
2411 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2412 {
2413 	device_del(dev);
2414 
2415 	if (dev->devt != 0)
2416 		cdev_del(ldev);
2417 }
2418 
2419 static void
2420 linux_destroy_dev(struct linux_cdev *ldev)
2421 {
2422 
2423 	if (ldev->cdev == NULL)
2424 		return;
2425 
2426 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2427 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2428 
2429 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2430 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2431 		pause("ldevdtr", hz / 4);
2432 
2433 	destroy_dev(ldev->cdev);
2434 	ldev->cdev = NULL;
2435 }
2436 
2437 const struct kobj_type linux_cdev_ktype = {
2438 	.release = linux_cdev_release,
2439 };
2440 
2441 const struct kobj_type linux_cdev_static_ktype = {
2442 	.release = linux_cdev_static_release,
2443 };
2444 
2445 static void
2446 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2447 {
2448 	struct notifier_block *nb;
2449 	struct netdev_notifier_info ni;
2450 
2451 	nb = arg;
2452 	ni.ifp = ifp;
2453 	ni.dev = (struct net_device *)ifp;
2454 	if (linkstate == LINK_STATE_UP)
2455 		nb->notifier_call(nb, NETDEV_UP, &ni);
2456 	else
2457 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2458 }
2459 
2460 static void
2461 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2462 {
2463 	struct notifier_block *nb;
2464 	struct netdev_notifier_info ni;
2465 
2466 	nb = arg;
2467 	ni.ifp = ifp;
2468 	ni.dev = (struct net_device *)ifp;
2469 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2470 }
2471 
2472 static void
2473 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2474 {
2475 	struct notifier_block *nb;
2476 	struct netdev_notifier_info ni;
2477 
2478 	nb = arg;
2479 	ni.ifp = ifp;
2480 	ni.dev = (struct net_device *)ifp;
2481 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2482 }
2483 
2484 static void
2485 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2486 {
2487 	struct notifier_block *nb;
2488 	struct netdev_notifier_info ni;
2489 
2490 	nb = arg;
2491 	ni.ifp = ifp;
2492 	ni.dev = (struct net_device *)ifp;
2493 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2494 }
2495 
2496 static void
2497 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2498 {
2499 	struct notifier_block *nb;
2500 	struct netdev_notifier_info ni;
2501 
2502 	nb = arg;
2503 	ni.ifp = ifp;
2504 	ni.dev = (struct net_device *)ifp;
2505 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2506 }
2507 
2508 int
2509 register_netdevice_notifier(struct notifier_block *nb)
2510 {
2511 
2512 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2513 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2514 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2515 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2516 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2517 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2518 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2519 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2520 
2521 	return (0);
2522 }
2523 
2524 int
2525 register_inetaddr_notifier(struct notifier_block *nb)
2526 {
2527 
2528 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2529 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2530 	return (0);
2531 }
2532 
2533 int
2534 unregister_netdevice_notifier(struct notifier_block *nb)
2535 {
2536 
2537 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2538 	    nb->tags[NETDEV_UP]);
2539 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2540 	    nb->tags[NETDEV_REGISTER]);
2541 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2542 	    nb->tags[NETDEV_UNREGISTER]);
2543 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2544 	    nb->tags[NETDEV_CHANGEADDR]);
2545 
2546 	return (0);
2547 }
2548 
2549 int
2550 unregister_inetaddr_notifier(struct notifier_block *nb)
2551 {
2552 
2553 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2554 	    nb->tags[NETDEV_CHANGEIFADDR]);
2555 
2556 	return (0);
2557 }
2558 
2559 struct list_sort_thunk {
2560 	int (*cmp)(void *, struct list_head *, struct list_head *);
2561 	void *priv;
2562 };
2563 
2564 static inline int
2565 linux_le_cmp(void *priv, const void *d1, const void *d2)
2566 {
2567 	struct list_head *le1, *le2;
2568 	struct list_sort_thunk *thunk;
2569 
2570 	thunk = priv;
2571 	le1 = *(__DECONST(struct list_head **, d1));
2572 	le2 = *(__DECONST(struct list_head **, d2));
2573 	return ((thunk->cmp)(thunk->priv, le1, le2));
2574 }
2575 
2576 void
2577 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2578     struct list_head *a, struct list_head *b))
2579 {
2580 	struct list_sort_thunk thunk;
2581 	struct list_head **ar, *le;
2582 	size_t count, i;
2583 
2584 	count = 0;
2585 	list_for_each(le, head)
2586 		count++;
2587 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2588 	i = 0;
2589 	list_for_each(le, head)
2590 		ar[i++] = le;
2591 	thunk.cmp = cmp;
2592 	thunk.priv = priv;
2593 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2594 	INIT_LIST_HEAD(head);
2595 	for (i = 0; i < count; i++)
2596 		list_add_tail(ar[i], head);
2597 	free(ar, M_KMALLOC);
2598 }
2599 
2600 #if defined(__i386__) || defined(__amd64__)
2601 int
2602 linux_wbinvd_on_all_cpus(void)
2603 {
2604 
2605 	pmap_invalidate_cache();
2606 	return (0);
2607 }
2608 #endif
2609 
2610 int
2611 linux_on_each_cpu(void callback(void *), void *data)
2612 {
2613 
2614 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2615 	    smp_no_rendezvous_barrier, data);
2616 	return (0);
2617 }
2618 
2619 int
2620 linux_in_atomic(void)
2621 {
2622 
2623 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2624 }
2625 
2626 struct linux_cdev *
2627 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2628 {
2629 	dev_t dev = MKDEV(major, minor);
2630 	struct cdev *cdev;
2631 
2632 	dev_lock();
2633 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2634 		struct linux_cdev *ldev = cdev->si_drv1;
2635 		if (ldev->dev == dev &&
2636 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2637 			break;
2638 		}
2639 	}
2640 	dev_unlock();
2641 
2642 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2643 }
2644 
2645 int
2646 __register_chrdev(unsigned int major, unsigned int baseminor,
2647     unsigned int count, const char *name,
2648     const struct file_operations *fops)
2649 {
2650 	struct linux_cdev *cdev;
2651 	int ret = 0;
2652 	int i;
2653 
2654 	for (i = baseminor; i < baseminor + count; i++) {
2655 		cdev = cdev_alloc();
2656 		cdev->ops = fops;
2657 		kobject_set_name(&cdev->kobj, name);
2658 
2659 		ret = cdev_add(cdev, makedev(major, i), 1);
2660 		if (ret != 0)
2661 			break;
2662 	}
2663 	return (ret);
2664 }
2665 
2666 int
2667 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2668     unsigned int count, const char *name,
2669     const struct file_operations *fops, uid_t uid,
2670     gid_t gid, int mode)
2671 {
2672 	struct linux_cdev *cdev;
2673 	int ret = 0;
2674 	int i;
2675 
2676 	for (i = baseminor; i < baseminor + count; i++) {
2677 		cdev = cdev_alloc();
2678 		cdev->ops = fops;
2679 		kobject_set_name(&cdev->kobj, name);
2680 
2681 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2682 		if (ret != 0)
2683 			break;
2684 	}
2685 	return (ret);
2686 }
2687 
2688 void
2689 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2690     unsigned int count, const char *name)
2691 {
2692 	struct linux_cdev *cdevp;
2693 	int i;
2694 
2695 	for (i = baseminor; i < baseminor + count; i++) {
2696 		cdevp = linux_find_cdev(name, major, i);
2697 		if (cdevp != NULL)
2698 			cdev_del(cdevp);
2699 	}
2700 }
2701 
2702 void
2703 linux_dump_stack(void)
2704 {
2705 #ifdef STACK
2706 	struct stack st;
2707 
2708 	stack_save(&st);
2709 	stack_print(&st);
2710 #endif
2711 }
2712 
2713 int
2714 linuxkpi_net_ratelimit(void)
2715 {
2716 
2717 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2718 	   lkpi_net_maxpps));
2719 }
2720 
2721 struct io_mapping *
2722 io_mapping_create_wc(resource_size_t base, unsigned long size)
2723 {
2724 	struct io_mapping *mapping;
2725 
2726 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2727 	if (mapping == NULL)
2728 		return (NULL);
2729 	return (io_mapping_init_wc(mapping, base, size));
2730 }
2731 
2732 #if defined(__i386__) || defined(__amd64__)
2733 bool linux_cpu_has_clflush;
2734 struct cpuinfo_x86 boot_cpu_data;
2735 #endif
2736 
2737 static void
2738 linux_compat_init(void *arg)
2739 {
2740 	struct sysctl_oid *rootoid;
2741 	int i;
2742 
2743 #if defined(__i386__) || defined(__amd64__)
2744 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2745 	boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2746 	boot_cpu_data.x86 = ((cpu_id & 0xf0000) >> 12) | ((cpu_id & 0xf0) >> 4);
2747 #endif
2748 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2749 
2750 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2751 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2752 	kobject_init(&linux_class_root, &linux_class_ktype);
2753 	kobject_set_name(&linux_class_root, "class");
2754 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2755 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2756 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2757 	kobject_set_name(&linux_root_device.kobj, "device");
2758 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2759 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2760 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2761 	linux_root_device.bsddev = root_bus;
2762 	linux_class_misc.name = "misc";
2763 	class_register(&linux_class_misc);
2764 	INIT_LIST_HEAD(&pci_drivers);
2765 	INIT_LIST_HEAD(&pci_devices);
2766 	spin_lock_init(&pci_lock);
2767 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2768 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2769 		LIST_INIT(&vmmaphead[i]);
2770 	init_waitqueue_head(&linux_bit_waitq);
2771 	init_waitqueue_head(&linux_var_waitq);
2772 
2773 	CPU_COPY(&all_cpus, &cpu_online_mask);
2774 }
2775 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2776 
2777 static void
2778 linux_compat_uninit(void *arg)
2779 {
2780 	linux_kobject_kfree_name(&linux_class_root);
2781 	linux_kobject_kfree_name(&linux_root_device.kobj);
2782 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2783 
2784 	mtx_destroy(&vmmaplock);
2785 	spin_lock_destroy(&pci_lock);
2786 	rw_destroy(&linux_vma_lock);
2787 }
2788 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2789 
2790 /*
2791  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2792  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2793  * used. Assert these types have the same size, else some parts of the
2794  * LinuxKPI may not work like expected:
2795  */
2796 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2797