xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 68e709cb296e23b3c827e705d6328cd2261a1d71)
1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2017 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 <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/malloc.h>
36 #include <sys/kernel.h>
37 #include <sys/sysctl.h>
38 #include <sys/proc.h>
39 #include <sys/sglist.h>
40 #include <sys/sleepqueue.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/bus.h>
44 #include <sys/fcntl.h>
45 #include <sys/file.h>
46 #include <sys/filio.h>
47 #include <sys/rwlock.h>
48 #include <sys/mman.h>
49 
50 #include <vm/vm.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 
56 #include <machine/stdarg.h>
57 
58 #if defined(__i386__) || defined(__amd64__)
59 #include <machine/md_var.h>
60 #endif
61 
62 #include <linux/kobject.h>
63 #include <linux/device.h>
64 #include <linux/slab.h>
65 #include <linux/module.h>
66 #include <linux/moduleparam.h>
67 #include <linux/cdev.h>
68 #include <linux/file.h>
69 #include <linux/sysfs.h>
70 #include <linux/mm.h>
71 #include <linux/io.h>
72 #include <linux/vmalloc.h>
73 #include <linux/netdevice.h>
74 #include <linux/timer.h>
75 #include <linux/interrupt.h>
76 #include <linux/uaccess.h>
77 #include <linux/list.h>
78 #include <linux/kthread.h>
79 #include <linux/kernel.h>
80 #include <linux/compat.h>
81 #include <linux/poll.h>
82 #include <linux/smp.h>
83 
84 #if defined(__i386__) || defined(__amd64__)
85 #include <asm/smp.h>
86 #endif
87 
88 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters");
89 
90 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
91 
92 #include <linux/rbtree.h>
93 /* Undo Linux compat changes. */
94 #undef RB_ROOT
95 #undef file
96 #undef cdev
97 #define	RB_ROOT(head)	(head)->rbh_root
98 
99 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
100 
101 struct kobject linux_class_root;
102 struct device linux_root_device;
103 struct class linux_class_misc;
104 struct list_head pci_drivers;
105 struct list_head pci_devices;
106 spinlock_t pci_lock;
107 
108 unsigned long linux_timer_hz_mask;
109 
110 int
111 panic_cmp(struct rb_node *one, struct rb_node *two)
112 {
113 	panic("no cmp");
114 }
115 
116 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
117 
118 int
119 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
120 {
121 	va_list tmp_va;
122 	int len;
123 	char *old;
124 	char *name;
125 	char dummy;
126 
127 	old = kobj->name;
128 
129 	if (old && fmt == NULL)
130 		return (0);
131 
132 	/* compute length of string */
133 	va_copy(tmp_va, args);
134 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
135 	va_end(tmp_va);
136 
137 	/* account for zero termination */
138 	len++;
139 
140 	/* check for error */
141 	if (len < 1)
142 		return (-EINVAL);
143 
144 	/* allocate memory for string */
145 	name = kzalloc(len, GFP_KERNEL);
146 	if (name == NULL)
147 		return (-ENOMEM);
148 	vsnprintf(name, len, fmt, args);
149 	kobj->name = name;
150 
151 	/* free old string */
152 	kfree(old);
153 
154 	/* filter new string */
155 	for (; *name != '\0'; name++)
156 		if (*name == '/')
157 			*name = '!';
158 	return (0);
159 }
160 
161 int
162 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
163 {
164 	va_list args;
165 	int error;
166 
167 	va_start(args, fmt);
168 	error = kobject_set_name_vargs(kobj, fmt, args);
169 	va_end(args);
170 
171 	return (error);
172 }
173 
174 static int
175 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
176 {
177 	const struct kobj_type *t;
178 	int error;
179 
180 	kobj->parent = parent;
181 	error = sysfs_create_dir(kobj);
182 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
183 		struct attribute **attr;
184 		t = kobj->ktype;
185 
186 		for (attr = t->default_attrs; *attr != NULL; attr++) {
187 			error = sysfs_create_file(kobj, *attr);
188 			if (error)
189 				break;
190 		}
191 		if (error)
192 			sysfs_remove_dir(kobj);
193 
194 	}
195 	return (error);
196 }
197 
198 int
199 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
200 {
201 	va_list args;
202 	int error;
203 
204 	va_start(args, fmt);
205 	error = kobject_set_name_vargs(kobj, fmt, args);
206 	va_end(args);
207 	if (error)
208 		return (error);
209 
210 	return kobject_add_complete(kobj, parent);
211 }
212 
213 void
214 linux_kobject_release(struct kref *kref)
215 {
216 	struct kobject *kobj;
217 	char *name;
218 
219 	kobj = container_of(kref, struct kobject, kref);
220 	sysfs_remove_dir(kobj);
221 	name = kobj->name;
222 	if (kobj->ktype && kobj->ktype->release)
223 		kobj->ktype->release(kobj);
224 	kfree(name);
225 }
226 
227 static void
228 linux_kobject_kfree(struct kobject *kobj)
229 {
230 	kfree(kobj);
231 }
232 
233 static void
234 linux_kobject_kfree_name(struct kobject *kobj)
235 {
236 	if (kobj) {
237 		kfree(kobj->name);
238 	}
239 }
240 
241 const struct kobj_type linux_kfree_type = {
242 	.release = linux_kobject_kfree
243 };
244 
245 static void
246 linux_device_release(struct device *dev)
247 {
248 	pr_debug("linux_device_release: %s\n", dev_name(dev));
249 	kfree(dev);
250 }
251 
252 static ssize_t
253 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
254 {
255 	struct class_attribute *dattr;
256 	ssize_t error;
257 
258 	dattr = container_of(attr, struct class_attribute, attr);
259 	error = -EIO;
260 	if (dattr->show)
261 		error = dattr->show(container_of(kobj, struct class, kobj),
262 		    dattr, buf);
263 	return (error);
264 }
265 
266 static ssize_t
267 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
268     size_t count)
269 {
270 	struct class_attribute *dattr;
271 	ssize_t error;
272 
273 	dattr = container_of(attr, struct class_attribute, attr);
274 	error = -EIO;
275 	if (dattr->store)
276 		error = dattr->store(container_of(kobj, struct class, kobj),
277 		    dattr, buf, count);
278 	return (error);
279 }
280 
281 static void
282 linux_class_release(struct kobject *kobj)
283 {
284 	struct class *class;
285 
286 	class = container_of(kobj, struct class, kobj);
287 	if (class->class_release)
288 		class->class_release(class);
289 }
290 
291 static const struct sysfs_ops linux_class_sysfs = {
292 	.show  = linux_class_show,
293 	.store = linux_class_store,
294 };
295 
296 const struct kobj_type linux_class_ktype = {
297 	.release = linux_class_release,
298 	.sysfs_ops = &linux_class_sysfs
299 };
300 
301 static void
302 linux_dev_release(struct kobject *kobj)
303 {
304 	struct device *dev;
305 
306 	dev = container_of(kobj, struct device, kobj);
307 	/* This is the precedence defined by linux. */
308 	if (dev->release)
309 		dev->release(dev);
310 	else if (dev->class && dev->class->dev_release)
311 		dev->class->dev_release(dev);
312 }
313 
314 static ssize_t
315 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
316 {
317 	struct device_attribute *dattr;
318 	ssize_t error;
319 
320 	dattr = container_of(attr, struct device_attribute, attr);
321 	error = -EIO;
322 	if (dattr->show)
323 		error = dattr->show(container_of(kobj, struct device, kobj),
324 		    dattr, buf);
325 	return (error);
326 }
327 
328 static ssize_t
329 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
330     size_t count)
331 {
332 	struct device_attribute *dattr;
333 	ssize_t error;
334 
335 	dattr = container_of(attr, struct device_attribute, attr);
336 	error = -EIO;
337 	if (dattr->store)
338 		error = dattr->store(container_of(kobj, struct device, kobj),
339 		    dattr, buf, count);
340 	return (error);
341 }
342 
343 static const struct sysfs_ops linux_dev_sysfs = {
344 	.show  = linux_dev_show,
345 	.store = linux_dev_store,
346 };
347 
348 const struct kobj_type linux_dev_ktype = {
349 	.release = linux_dev_release,
350 	.sysfs_ops = &linux_dev_sysfs
351 };
352 
353 struct device *
354 device_create(struct class *class, struct device *parent, dev_t devt,
355     void *drvdata, const char *fmt, ...)
356 {
357 	struct device *dev;
358 	va_list args;
359 
360 	dev = kzalloc(sizeof(*dev), M_WAITOK);
361 	dev->parent = parent;
362 	dev->class = class;
363 	dev->devt = devt;
364 	dev->driver_data = drvdata;
365 	dev->release = linux_device_release;
366 	va_start(args, fmt);
367 	kobject_set_name_vargs(&dev->kobj, fmt, args);
368 	va_end(args);
369 	device_register(dev);
370 
371 	return (dev);
372 }
373 
374 int
375 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
376     struct kobject *parent, const char *fmt, ...)
377 {
378 	va_list args;
379 	int error;
380 
381 	kobject_init(kobj, ktype);
382 	kobj->ktype = ktype;
383 	kobj->parent = parent;
384 	kobj->name = NULL;
385 
386 	va_start(args, fmt);
387 	error = kobject_set_name_vargs(kobj, fmt, args);
388 	va_end(args);
389 	if (error)
390 		return (error);
391 	return kobject_add_complete(kobj, parent);
392 }
393 
394 static void
395 linux_kq_lock(void *arg)
396 {
397 	spinlock_t *s = arg;
398 
399 	spin_lock(s);
400 }
401 static void
402 linux_kq_unlock(void *arg)
403 {
404 	spinlock_t *s = arg;
405 
406 	spin_unlock(s);
407 }
408 
409 static void
410 linux_kq_lock_owned(void *arg)
411 {
412 #ifdef INVARIANTS
413 	spinlock_t *s = arg;
414 
415 	mtx_assert(&s->m, MA_OWNED);
416 #endif
417 }
418 
419 static void
420 linux_kq_lock_unowned(void *arg)
421 {
422 #ifdef INVARIANTS
423 	spinlock_t *s = arg;
424 
425 	mtx_assert(&s->m, MA_NOTOWNED);
426 #endif
427 }
428 
429 static void
430 linux_file_kqfilter_poll(struct linux_file *, int);
431 
432 struct linux_file *
433 linux_file_alloc(void)
434 {
435 	struct linux_file *filp;
436 
437 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
438 
439 	/* set initial refcount */
440 	filp->f_count = 1;
441 
442 	/* setup fields needed by kqueue support */
443 	spin_lock_init(&filp->f_kqlock);
444 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
445 	    linux_kq_lock, linux_kq_unlock,
446 	    linux_kq_lock_owned, linux_kq_lock_unowned);
447 
448 	return (filp);
449 }
450 
451 void
452 linux_file_free(struct linux_file *filp)
453 {
454 	if (filp->_file == NULL) {
455 		if (filp->f_shmem != NULL)
456 			vm_object_deallocate(filp->f_shmem);
457 		kfree(filp);
458 	} else {
459 		/*
460 		 * The close method of the character device or file
461 		 * will free the linux_file structure:
462 		 */
463 		_fdrop(filp->_file, curthread);
464 	}
465 }
466 
467 static int
468 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
469     vm_page_t *mres)
470 {
471 	struct vm_area_struct *vmap;
472 
473 	vmap = linux_cdev_handle_find(vm_obj->handle);
474 
475 	MPASS(vmap != NULL);
476 	MPASS(vmap->vm_private_data == vm_obj->handle);
477 
478 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
479 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
480 		vm_page_t page;
481 
482 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
483 			/*
484 			 * If the passed in result page is a fake
485 			 * page, update it with the new physical
486 			 * address.
487 			 */
488 			page = *mres;
489 			vm_page_updatefake(page, paddr, vm_obj->memattr);
490 		} else {
491 			/*
492 			 * Replace the passed in "mres" page with our
493 			 * own fake page and free up the all of the
494 			 * original pages.
495 			 */
496 			VM_OBJECT_WUNLOCK(vm_obj);
497 			page = vm_page_getfake(paddr, vm_obj->memattr);
498 			VM_OBJECT_WLOCK(vm_obj);
499 
500 			vm_page_replace_checked(page, vm_obj,
501 			    (*mres)->pindex, *mres);
502 
503 			vm_page_lock(*mres);
504 			vm_page_free(*mres);
505 			vm_page_unlock(*mres);
506 			*mres = page;
507 		}
508 		page->valid = VM_PAGE_BITS_ALL;
509 		return (VM_PAGER_OK);
510 	}
511 	return (VM_PAGER_FAIL);
512 }
513 
514 static int
515 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
516     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
517 {
518 	struct vm_area_struct *vmap;
519 	int err;
520 
521 	linux_set_current(curthread);
522 
523 	/* get VM area structure */
524 	vmap = linux_cdev_handle_find(vm_obj->handle);
525 	MPASS(vmap != NULL);
526 	MPASS(vmap->vm_private_data == vm_obj->handle);
527 
528 	VM_OBJECT_WUNLOCK(vm_obj);
529 
530 	down_write(&vmap->vm_mm->mmap_sem);
531 	if (unlikely(vmap->vm_ops == NULL)) {
532 		err = VM_FAULT_SIGBUS;
533 	} else {
534 		struct vm_fault vmf;
535 
536 		/* fill out VM fault structure */
537 		vmf.virtual_address = (void *)((uintptr_t)pidx << PAGE_SHIFT);
538 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
539 		vmf.pgoff = 0;
540 		vmf.page = NULL;
541 
542 		vmap->vm_pfn_count = 0;
543 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
544 		vmap->vm_obj = vm_obj;
545 
546 		err = vmap->vm_ops->fault(vmap, &vmf);
547 
548 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
549 			kern_yield(PRI_USER);
550 			err = vmap->vm_ops->fault(vmap, &vmf);
551 		}
552 	}
553 
554 	/* translate return code */
555 	switch (err) {
556 	case VM_FAULT_OOM:
557 		err = VM_PAGER_AGAIN;
558 		break;
559 	case VM_FAULT_SIGBUS:
560 		err = VM_PAGER_BAD;
561 		break;
562 	case VM_FAULT_NOPAGE:
563 		/*
564 		 * By contract the fault handler will return having
565 		 * busied all the pages itself. If pidx is already
566 		 * found in the object, it will simply xbusy the first
567 		 * page and return with vm_pfn_count set to 1.
568 		 */
569 		*first = vmap->vm_pfn_first;
570 		*last = *first + vmap->vm_pfn_count - 1;
571 		err = VM_PAGER_OK;
572 		break;
573 	default:
574 		err = VM_PAGER_ERROR;
575 		break;
576 	}
577 	up_write(&vmap->vm_mm->mmap_sem);
578 	VM_OBJECT_WLOCK(vm_obj);
579 	return (err);
580 }
581 
582 static struct rwlock linux_vma_lock;
583 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
584     TAILQ_HEAD_INITIALIZER(linux_vma_head);
585 
586 static void
587 linux_cdev_handle_free(struct vm_area_struct *vmap)
588 {
589 	/* Drop reference on vm_file */
590 	if (vmap->vm_file != NULL)
591 		fput(vmap->vm_file);
592 
593 	/* Drop reference on mm_struct */
594 	mmput(vmap->vm_mm);
595 
596 	kfree(vmap);
597 }
598 
599 static void
600 linux_cdev_handle_remove(struct vm_area_struct *vmap)
601 {
602 	rw_wlock(&linux_vma_lock);
603 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
604 	rw_wunlock(&linux_vma_lock);
605 }
606 
607 static struct vm_area_struct *
608 linux_cdev_handle_find(void *handle)
609 {
610 	struct vm_area_struct *vmap;
611 
612 	rw_rlock(&linux_vma_lock);
613 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
614 		if (vmap->vm_private_data == handle)
615 			break;
616 	}
617 	rw_runlock(&linux_vma_lock);
618 	return (vmap);
619 }
620 
621 static int
622 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
623 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
624 {
625 
626 	MPASS(linux_cdev_handle_find(handle) != NULL);
627 	*color = 0;
628 	return (0);
629 }
630 
631 static void
632 linux_cdev_pager_dtor(void *handle)
633 {
634 	const struct vm_operations_struct *vm_ops;
635 	struct vm_area_struct *vmap;
636 
637 	vmap = linux_cdev_handle_find(handle);
638 	MPASS(vmap != NULL);
639 
640 	/*
641 	 * Remove handle before calling close operation to prevent
642 	 * other threads from reusing the handle pointer.
643 	 */
644 	linux_cdev_handle_remove(vmap);
645 
646 	down_write(&vmap->vm_mm->mmap_sem);
647 	vm_ops = vmap->vm_ops;
648 	if (likely(vm_ops != NULL))
649 		vm_ops->close(vmap);
650 	up_write(&vmap->vm_mm->mmap_sem);
651 
652 	linux_cdev_handle_free(vmap);
653 }
654 
655 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
656   {
657 	/* OBJT_MGTDEVICE */
658 	.cdev_pg_populate	= linux_cdev_pager_populate,
659 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
660 	.cdev_pg_dtor	= linux_cdev_pager_dtor
661   },
662   {
663 	/* OBJT_DEVICE */
664 	.cdev_pg_fault	= linux_cdev_pager_fault,
665 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
666 	.cdev_pg_dtor	= linux_cdev_pager_dtor
667   },
668 };
669 
670 #define	OPW(fp,td,code) ({			\
671 	struct file *__fpop;			\
672 	int __retval;				\
673 						\
674 	__fpop = (td)->td_fpop;			\
675 	(td)->td_fpop = (fp);			\
676 	__retval = (code);			\
677 	(td)->td_fpop = __fpop;			\
678 	__retval;				\
679 })
680 
681 static int
682 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, struct file *file)
683 {
684 	struct linux_cdev *ldev;
685 	struct linux_file *filp;
686 	int error;
687 
688 	ldev = dev->si_drv1;
689 
690 	filp = linux_file_alloc();
691 	filp->f_dentry = &filp->f_dentry_store;
692 	filp->f_op = ldev->ops;
693 	filp->f_mode = file->f_flag;
694 	filp->f_flags = file->f_flag;
695 	filp->f_vnode = file->f_vnode;
696 	filp->_file = file;
697 
698 	linux_set_current(td);
699 
700 	if (filp->f_op->open) {
701 		error = -filp->f_op->open(file->f_vnode, filp);
702 		if (error) {
703 			kfree(filp);
704 			return (error);
705 		}
706 	}
707 
708 	/* hold on to the vnode - used for fstat() */
709 	vhold(filp->f_vnode);
710 
711 	/* release the file from devfs */
712 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
713 	return (ENXIO);
714 }
715 
716 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
717 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
718 
719 static inline int
720 linux_remap_address(void **uaddr, size_t len)
721 {
722 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
723 
724 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
725 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
726 		struct task_struct *pts = current;
727 		if (pts == NULL) {
728 			*uaddr = NULL;
729 			return (1);
730 		}
731 
732 		/* compute data offset */
733 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
734 
735 		/* check that length is within bounds */
736 		if ((len > IOCPARM_MAX) ||
737 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
738 			*uaddr = NULL;
739 			return (1);
740 		}
741 
742 		/* re-add kernel buffer address */
743 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
744 
745 		/* update address location */
746 		*uaddr = (void *)uaddr_val;
747 		return (1);
748 	}
749 	return (0);
750 }
751 
752 int
753 linux_copyin(const void *uaddr, void *kaddr, size_t len)
754 {
755 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
756 		if (uaddr == NULL)
757 			return (-EFAULT);
758 		memcpy(kaddr, uaddr, len);
759 		return (0);
760 	}
761 	return (-copyin(uaddr, kaddr, len));
762 }
763 
764 int
765 linux_copyout(const void *kaddr, void *uaddr, size_t len)
766 {
767 	if (linux_remap_address(&uaddr, len)) {
768 		if (uaddr == NULL)
769 			return (-EFAULT);
770 		memcpy(uaddr, kaddr, len);
771 		return (0);
772 	}
773 	return (-copyout(kaddr, uaddr, len));
774 }
775 
776 size_t
777 linux_clear_user(void *_uaddr, size_t _len)
778 {
779 	uint8_t *uaddr = _uaddr;
780 	size_t len = _len;
781 
782 	/* make sure uaddr is aligned before going into the fast loop */
783 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
784 		if (subyte(uaddr, 0))
785 			return (_len);
786 		uaddr++;
787 		len--;
788 	}
789 
790 	/* zero 8 bytes at a time */
791 	while (len > 7) {
792 #ifdef __LP64__
793 		if (suword64(uaddr, 0))
794 			return (_len);
795 #else
796 		if (suword32(uaddr, 0))
797 			return (_len);
798 		if (suword32(uaddr + 4, 0))
799 			return (_len);
800 #endif
801 		uaddr += 8;
802 		len -= 8;
803 	}
804 
805 	/* zero fill end, if any */
806 	while (len > 0) {
807 		if (subyte(uaddr, 0))
808 			return (_len);
809 		uaddr++;
810 		len--;
811 	}
812 	return (0);
813 }
814 
815 int
816 linux_access_ok(int rw, const void *uaddr, size_t len)
817 {
818 	uintptr_t saddr;
819 	uintptr_t eaddr;
820 
821 	/* get start and end address */
822 	saddr = (uintptr_t)uaddr;
823 	eaddr = (uintptr_t)uaddr + len;
824 
825 	/* verify addresses are valid for userspace */
826 	return ((saddr == eaddr) ||
827 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
828 }
829 
830 static int
831 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
832     u_long cmd, caddr_t data, struct thread *td)
833 {
834 	unsigned size;
835 	int error;
836 
837 	size = IOCPARM_LEN(cmd);
838 	/* refer to logic in sys_ioctl() */
839 	if (size > 0) {
840 		/*
841 		 * Setup hint for linux_copyin() and linux_copyout().
842 		 *
843 		 * Background: Linux code expects a user-space address
844 		 * while FreeBSD supplies a kernel-space address.
845 		 */
846 		current->bsd_ioctl_data = data;
847 		current->bsd_ioctl_len = size;
848 		data = (void *)LINUX_IOCTL_MIN_PTR;
849 	} else {
850 		/* fetch user-space pointer */
851 		data = *(void **)data;
852 	}
853 #if defined(__amd64__)
854 	if (td->td_proc->p_elf_machine == EM_386) {
855 		/* try the compat IOCTL handler first */
856 		if (filp->f_op->compat_ioctl != NULL)
857 			error = -OPW(fp, td, filp->f_op->compat_ioctl(filp, cmd, (u_long)data));
858 		else
859 			error = ENOTTY;
860 
861 		/* fallback to the regular IOCTL handler, if any */
862 		if (error == ENOTTY && filp->f_op->unlocked_ioctl != NULL)
863 			error = -OPW(fp, td, filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data));
864 	} else
865 #endif
866 	if (filp->f_op->unlocked_ioctl != NULL)
867 		error = -OPW(fp, td, filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data));
868 	else
869 		error = ENOTTY;
870 	if (size > 0) {
871 		current->bsd_ioctl_data = NULL;
872 		current->bsd_ioctl_len = 0;
873 	}
874 
875 	if (error == EWOULDBLOCK) {
876 		/* update kqfilter status, if any */
877 		linux_file_kqfilter_poll(filp,
878 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
879 	} else if (error == ERESTARTSYS)
880 		error = ERESTART;
881 	return (error);
882 }
883 
884 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
885 
886 /*
887  * This function atomically updates the poll wakeup state and returns
888  * the previous state at the time of update.
889  */
890 static uint8_t
891 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
892 {
893 	int c, old;
894 
895 	c = v->counter;
896 
897 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
898 		c = old;
899 
900 	return (c);
901 }
902 
903 
904 static int
905 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
906 {
907 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
908 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
909 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
910 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
911 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
912 	};
913 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
914 
915 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
916 	case LINUX_FWQ_STATE_QUEUED:
917 		linux_poll_wakeup(filp);
918 		return (1);
919 	default:
920 		return (0);
921 	}
922 }
923 
924 void
925 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
926 {
927 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
928 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
929 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
930 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
931 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
932 	};
933 
934 	/* check if we are called inside the select system call */
935 	if (p == LINUX_POLL_TABLE_NORMAL)
936 		selrecord(curthread, &filp->f_selinfo);
937 
938 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
939 	case LINUX_FWQ_STATE_INIT:
940 		/* NOTE: file handles can only belong to one wait-queue */
941 		filp->f_wait_queue.wqh = wqh;
942 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
943 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
944 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
945 		break;
946 	default:
947 		break;
948 	}
949 }
950 
951 static void
952 linux_poll_wait_dequeue(struct linux_file *filp)
953 {
954 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
955 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
956 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
957 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
958 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
959 	};
960 
961 	seldrain(&filp->f_selinfo);
962 
963 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
964 	case LINUX_FWQ_STATE_NOT_READY:
965 	case LINUX_FWQ_STATE_QUEUED:
966 	case LINUX_FWQ_STATE_READY:
967 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
968 		break;
969 	default:
970 		break;
971 	}
972 }
973 
974 void
975 linux_poll_wakeup(struct linux_file *filp)
976 {
977 	/* this function should be NULL-safe */
978 	if (filp == NULL)
979 		return;
980 
981 	selwakeup(&filp->f_selinfo);
982 
983 	spin_lock(&filp->f_kqlock);
984 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
985 	    LINUX_KQ_FLAG_NEED_WRITE;
986 
987 	/* make sure the "knote" gets woken up */
988 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
989 	spin_unlock(&filp->f_kqlock);
990 }
991 
992 static void
993 linux_file_kqfilter_detach(struct knote *kn)
994 {
995 	struct linux_file *filp = kn->kn_hook;
996 
997 	spin_lock(&filp->f_kqlock);
998 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
999 	spin_unlock(&filp->f_kqlock);
1000 }
1001 
1002 static int
1003 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1004 {
1005 	struct linux_file *filp = kn->kn_hook;
1006 
1007 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1008 
1009 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1010 }
1011 
1012 static int
1013 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1014 {
1015 	struct linux_file *filp = kn->kn_hook;
1016 
1017 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1018 
1019 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1020 }
1021 
1022 static struct filterops linux_dev_kqfiltops_read = {
1023 	.f_isfd = 1,
1024 	.f_detach = linux_file_kqfilter_detach,
1025 	.f_event = linux_file_kqfilter_read_event,
1026 };
1027 
1028 static struct filterops linux_dev_kqfiltops_write = {
1029 	.f_isfd = 1,
1030 	.f_detach = linux_file_kqfilter_detach,
1031 	.f_event = linux_file_kqfilter_write_event,
1032 };
1033 
1034 static void
1035 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1036 {
1037 	int temp;
1038 
1039 	if (filp->f_kqflags & kqflags) {
1040 		struct thread *td = curthread;
1041 
1042 		/* get the latest polling state */
1043 		temp = OPW(filp->_file, td, filp->f_op->poll(filp, NULL));
1044 
1045 		spin_lock(&filp->f_kqlock);
1046 		/* clear kqflags */
1047 		filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1048 		    LINUX_KQ_FLAG_NEED_WRITE);
1049 		/* update kqflags */
1050 		if (temp & (POLLIN | POLLOUT)) {
1051 			if (temp & POLLIN)
1052 				filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1053 			if (temp & POLLOUT)
1054 				filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1055 
1056 			/* make sure the "knote" gets woken up */
1057 			KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1058 		}
1059 		spin_unlock(&filp->f_kqlock);
1060 	}
1061 }
1062 
1063 static int
1064 linux_file_kqfilter(struct file *file, struct knote *kn)
1065 {
1066 	struct linux_file *filp;
1067 	struct thread *td;
1068 	int error;
1069 
1070 	td = curthread;
1071 	filp = (struct linux_file *)file->f_data;
1072 	filp->f_flags = file->f_flag;
1073 	if (filp->f_op->poll == NULL)
1074 		return (EINVAL);
1075 
1076 	spin_lock(&filp->f_kqlock);
1077 	switch (kn->kn_filter) {
1078 	case EVFILT_READ:
1079 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1080 		kn->kn_fop = &linux_dev_kqfiltops_read;
1081 		kn->kn_hook = filp;
1082 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1083 		error = 0;
1084 		break;
1085 	case EVFILT_WRITE:
1086 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1087 		kn->kn_fop = &linux_dev_kqfiltops_write;
1088 		kn->kn_hook = filp;
1089 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1090 		error = 0;
1091 		break;
1092 	default:
1093 		error = EINVAL;
1094 		break;
1095 	}
1096 	spin_unlock(&filp->f_kqlock);
1097 
1098 	if (error == 0) {
1099 		linux_set_current(td);
1100 
1101 		/* update kqfilter status, if any */
1102 		linux_file_kqfilter_poll(filp,
1103 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1104 	}
1105 	return (error);
1106 }
1107 
1108 static int
1109 linux_file_mmap_single(struct file *fp, vm_ooffset_t *offset,
1110     vm_size_t size, struct vm_object **object, int nprot,
1111     struct thread *td)
1112 {
1113 	struct vm_area_struct *vmap;
1114 	struct mm_struct *mm;
1115 	struct linux_file *filp;
1116 	vm_memattr_t attr;
1117 	int error;
1118 
1119 	filp = (struct linux_file *)fp->f_data;
1120 	filp->f_flags = fp->f_flag;
1121 
1122 	if (filp->f_op->mmap == NULL)
1123 		return (EOPNOTSUPP);
1124 
1125 	linux_set_current(td);
1126 
1127 	/*
1128 	 * The same VM object might be shared by multiple processes
1129 	 * and the mm_struct is usually freed when a process exits.
1130 	 *
1131 	 * The atomic reference below makes sure the mm_struct is
1132 	 * available as long as the vmap is in the linux_vma_head.
1133 	 */
1134 	mm = current->mm;
1135 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1136 		return (EINVAL);
1137 
1138 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1139 	vmap->vm_start = 0;
1140 	vmap->vm_end = size;
1141 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1142 	vmap->vm_pfn = 0;
1143 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1144 	vmap->vm_ops = NULL;
1145 	vmap->vm_file = get_file(filp);
1146 	vmap->vm_mm = mm;
1147 
1148 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1149 		error = EINTR;
1150 	} else {
1151 		error = -OPW(fp, td, filp->f_op->mmap(filp, vmap));
1152 		up_write(&vmap->vm_mm->mmap_sem);
1153 	}
1154 
1155 	if (error != 0) {
1156 		linux_cdev_handle_free(vmap);
1157 		return (error);
1158 	}
1159 
1160 	attr = pgprot2cachemode(vmap->vm_page_prot);
1161 
1162 	if (vmap->vm_ops != NULL) {
1163 		struct vm_area_struct *ptr;
1164 		void *vm_private_data;
1165 		bool vm_no_fault;
1166 
1167 		if (vmap->vm_ops->open == NULL ||
1168 		    vmap->vm_ops->close == NULL ||
1169 		    vmap->vm_private_data == NULL) {
1170 			/* free allocated VM area struct */
1171 			linux_cdev_handle_free(vmap);
1172 			return (EINVAL);
1173 		}
1174 
1175 		vm_private_data = vmap->vm_private_data;
1176 
1177 		rw_wlock(&linux_vma_lock);
1178 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1179 			if (ptr->vm_private_data == vm_private_data)
1180 				break;
1181 		}
1182 		/* check if there is an existing VM area struct */
1183 		if (ptr != NULL) {
1184 			/* check if the VM area structure is invalid */
1185 			if (ptr->vm_ops == NULL ||
1186 			    ptr->vm_ops->open == NULL ||
1187 			    ptr->vm_ops->close == NULL) {
1188 				error = ESTALE;
1189 				vm_no_fault = 1;
1190 			} else {
1191 				error = EEXIST;
1192 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1193 			}
1194 		} else {
1195 			/* insert VM area structure into list */
1196 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1197 			error = 0;
1198 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1199 		}
1200 		rw_wunlock(&linux_vma_lock);
1201 
1202 		if (error != 0) {
1203 			/* free allocated VM area struct */
1204 			linux_cdev_handle_free(vmap);
1205 			/* check for stale VM area struct */
1206 			if (error != EEXIST)
1207 				return (error);
1208 		}
1209 
1210 		/* check if there is no fault handler */
1211 		if (vm_no_fault) {
1212 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1213 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1214 			    td->td_ucred);
1215 		} else {
1216 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1217 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1218 			    td->td_ucred);
1219 		}
1220 
1221 		/* check if allocating the VM object failed */
1222 		if (*object == NULL) {
1223 			if (error == 0) {
1224 				/* remove VM area struct from list */
1225 				linux_cdev_handle_remove(vmap);
1226 				/* free allocated VM area struct */
1227 				linux_cdev_handle_free(vmap);
1228 			}
1229 			return (EINVAL);
1230 		}
1231 	} else {
1232 		struct sglist *sg;
1233 
1234 		sg = sglist_alloc(1, M_WAITOK);
1235 		sglist_append_phys(sg,
1236 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1237 
1238 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1239 		    nprot, 0, td->td_ucred);
1240 
1241 		linux_cdev_handle_free(vmap);
1242 
1243 		if (*object == NULL) {
1244 			sglist_free(sg);
1245 			return (EINVAL);
1246 		}
1247 	}
1248 
1249 	if (attr != VM_MEMATTR_DEFAULT) {
1250 		VM_OBJECT_WLOCK(*object);
1251 		vm_object_set_memattr(*object, attr);
1252 		VM_OBJECT_WUNLOCK(*object);
1253 	}
1254 	*offset = 0;
1255 	return (0);
1256 }
1257 
1258 struct cdevsw linuxcdevsw = {
1259 	.d_version = D_VERSION,
1260 	.d_fdopen = linux_dev_fdopen,
1261 	.d_name = "lkpidev",
1262 };
1263 
1264 static int
1265 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1266     int flags, struct thread *td)
1267 {
1268 	struct linux_file *filp;
1269 	ssize_t bytes;
1270 	int error;
1271 
1272 	error = 0;
1273 	filp = (struct linux_file *)file->f_data;
1274 	filp->f_flags = file->f_flag;
1275 	/* XXX no support for I/O vectors currently */
1276 	if (uio->uio_iovcnt != 1)
1277 		return (EOPNOTSUPP);
1278 	linux_set_current(td);
1279 	if (filp->f_op->read) {
1280 		bytes = OPW(file, td, filp->f_op->read(filp, uio->uio_iov->iov_base,
1281 		    uio->uio_iov->iov_len, &uio->uio_offset));
1282 		if (bytes >= 0) {
1283 			uio->uio_iov->iov_base =
1284 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1285 			uio->uio_iov->iov_len -= bytes;
1286 			uio->uio_resid -= bytes;
1287 		} else {
1288 			error = -bytes;
1289 			if (error == ERESTARTSYS)
1290 				error = ERESTART;
1291 		}
1292 	} else
1293 		error = ENXIO;
1294 
1295 	/* update kqfilter status, if any */
1296 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1297 
1298 	return (error);
1299 }
1300 
1301 static int
1302 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1303     int flags, struct thread *td)
1304 {
1305 	struct linux_file *filp;
1306 	ssize_t bytes;
1307 	int error;
1308 
1309 	error = 0;
1310 	filp = (struct linux_file *)file->f_data;
1311 	filp->f_flags = file->f_flag;
1312 	/* XXX no support for I/O vectors currently */
1313 	if (uio->uio_iovcnt != 1)
1314 		return (EOPNOTSUPP);
1315 	linux_set_current(td);
1316 	if (filp->f_op->write) {
1317 		bytes = OPW(file, td, filp->f_op->write(filp, uio->uio_iov->iov_base,
1318 		    uio->uio_iov->iov_len, &uio->uio_offset));
1319 		if (bytes >= 0) {
1320 			uio->uio_iov->iov_base =
1321 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1322 			uio->uio_iov->iov_len -= bytes;
1323 			uio->uio_resid -= bytes;
1324 		} else {
1325 			error = -bytes;
1326 			if (error == ERESTARTSYS)
1327 				error = ERESTART;
1328 		}
1329 	} else
1330 		error = ENXIO;
1331 
1332 	/* update kqfilter status, if any */
1333 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1334 
1335 	return (error);
1336 }
1337 
1338 static int
1339 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1340     struct thread *td)
1341 {
1342 	struct linux_file *filp;
1343 	int revents;
1344 
1345 	filp = (struct linux_file *)file->f_data;
1346 	filp->f_flags = file->f_flag;
1347 	linux_set_current(td);
1348 	if (filp->f_op->poll != NULL)
1349 		revents = OPW(file, td, filp->f_op->poll(filp, LINUX_POLL_TABLE_NORMAL)) & events;
1350 	else
1351 		revents = 0;
1352 
1353 	return (revents);
1354 }
1355 
1356 static int
1357 linux_file_close(struct file *file, struct thread *td)
1358 {
1359 	struct linux_file *filp;
1360 	int error;
1361 
1362 	filp = (struct linux_file *)file->f_data;
1363 
1364 	KASSERT(file_count(filp) == 0, ("File refcount(%d) is not zero", file_count(filp)));
1365 
1366 	filp->f_flags = file->f_flag;
1367 	linux_set_current(td);
1368 	linux_poll_wait_dequeue(filp);
1369 	error = -OPW(file, td, filp->f_op->release(filp->f_vnode, filp));
1370 	funsetown(&filp->f_sigio);
1371 	if (filp->f_vnode != NULL)
1372 		vdrop(filp->f_vnode);
1373 	kfree(filp);
1374 
1375 	return (error);
1376 }
1377 
1378 static int
1379 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1380     struct thread *td)
1381 {
1382 	struct linux_file *filp;
1383 	int error;
1384 
1385 	filp = (struct linux_file *)fp->f_data;
1386 	filp->f_flags = fp->f_flag;
1387 	error = 0;
1388 
1389 	linux_set_current(td);
1390 	switch (cmd) {
1391 	case FIONBIO:
1392 		break;
1393 	case FIOASYNC:
1394 		if (filp->f_op->fasync == NULL)
1395 			break;
1396 		error = -OPW(fp, td, filp->f_op->fasync(0, filp, fp->f_flag & FASYNC));
1397 		break;
1398 	case FIOSETOWN:
1399 		error = fsetown(*(int *)data, &filp->f_sigio);
1400 		if (error == 0) {
1401 			if (filp->f_op->fasync == NULL)
1402 				break;
1403 			error = -OPW(fp, td, filp->f_op->fasync(0, filp,
1404 			    fp->f_flag & FASYNC));
1405 		}
1406 		break;
1407 	case FIOGETOWN:
1408 		*(int *)data = fgetown(&filp->f_sigio);
1409 		break;
1410 	default:
1411 		error = linux_file_ioctl_sub(fp, filp, cmd, data, td);
1412 		break;
1413 	}
1414 	return (error);
1415 }
1416 
1417 static int
1418 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1419     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1420     vm_ooffset_t *foff, vm_object_t *objp)
1421 {
1422 	/*
1423 	 * Character devices do not provide private mappings
1424 	 * of any kind:
1425 	 */
1426 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1427 	    (prot & VM_PROT_WRITE) != 0)
1428 		return (EACCES);
1429 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1430 		return (EINVAL);
1431 
1432 	return (linux_file_mmap_single(fp, foff, objsize, objp, (int)prot, td));
1433 }
1434 
1435 static int
1436 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1437     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1438     struct thread *td)
1439 {
1440 	struct linux_file *filp;
1441 	struct mount *mp;
1442 	struct vnode *vp;
1443 	vm_object_t object;
1444 	vm_prot_t maxprot;
1445 	int error;
1446 
1447 	filp = (struct linux_file *)fp->f_data;
1448 
1449 	vp = filp->f_vnode;
1450 	if (vp == NULL)
1451 		return (EOPNOTSUPP);
1452 
1453 	/*
1454 	 * Ensure that file and memory protections are
1455 	 * compatible.
1456 	 */
1457 	mp = vp->v_mount;
1458 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1459 		maxprot = VM_PROT_NONE;
1460 		if ((prot & VM_PROT_EXECUTE) != 0)
1461 			return (EACCES);
1462 	} else
1463 		maxprot = VM_PROT_EXECUTE;
1464 	if ((fp->f_flag & FREAD) != 0)
1465 		maxprot |= VM_PROT_READ;
1466 	else if ((prot & VM_PROT_READ) != 0)
1467 		return (EACCES);
1468 
1469 	/*
1470 	 * If we are sharing potential changes via MAP_SHARED and we
1471 	 * are trying to get write permission although we opened it
1472 	 * without asking for it, bail out.
1473 	 *
1474 	 * Note that most character devices always share mappings.
1475 	 *
1476 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1477 	 * requests rather than doing it here.
1478 	 */
1479 	if ((flags & MAP_SHARED) != 0) {
1480 		if ((fp->f_flag & FWRITE) != 0)
1481 			maxprot |= VM_PROT_WRITE;
1482 		else if ((prot & VM_PROT_WRITE) != 0)
1483 			return (EACCES);
1484 	}
1485 	maxprot &= cap_maxprot;
1486 
1487 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, &foff,
1488 	    &object);
1489 	if (error != 0)
1490 		return (error);
1491 
1492 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1493 	    foff, FALSE, td);
1494 	if (error != 0)
1495 		vm_object_deallocate(object);
1496 	return (error);
1497 }
1498 
1499 static int
1500 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1501     struct thread *td)
1502 {
1503 	struct linux_file *filp;
1504 	struct vnode *vp;
1505 	int error;
1506 
1507 	filp = (struct linux_file *)fp->f_data;
1508 	if (filp->f_vnode == NULL)
1509 		return (EOPNOTSUPP);
1510 
1511 	vp = filp->f_vnode;
1512 
1513 	vn_lock(vp, LK_SHARED | LK_RETRY);
1514 	error = vn_stat(vp, sb, td->td_ucred, NOCRED, td);
1515 	VOP_UNLOCK(vp, 0);
1516 
1517 	return (error);
1518 }
1519 
1520 static int
1521 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1522     struct filedesc *fdp)
1523 {
1524 
1525 	return (0);
1526 }
1527 
1528 unsigned int
1529 linux_iminor(struct inode *inode)
1530 {
1531 	struct linux_cdev *ldev;
1532 
1533 	if (inode == NULL || inode->v_rdev == NULL ||
1534 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1535 		return (-1U);
1536 	ldev = inode->v_rdev->si_drv1;
1537 	if (ldev == NULL)
1538 		return (-1U);
1539 
1540 	return (minor(ldev->dev));
1541 }
1542 
1543 struct fileops linuxfileops = {
1544 	.fo_read = linux_file_read,
1545 	.fo_write = linux_file_write,
1546 	.fo_truncate = invfo_truncate,
1547 	.fo_kqfilter = linux_file_kqfilter,
1548 	.fo_stat = linux_file_stat,
1549 	.fo_fill_kinfo = linux_file_fill_kinfo,
1550 	.fo_poll = linux_file_poll,
1551 	.fo_close = linux_file_close,
1552 	.fo_ioctl = linux_file_ioctl,
1553 	.fo_mmap = linux_file_mmap,
1554 	.fo_chmod = invfo_chmod,
1555 	.fo_chown = invfo_chown,
1556 	.fo_sendfile = invfo_sendfile,
1557 };
1558 
1559 /*
1560  * Hash of vmmap addresses.  This is infrequently accessed and does not
1561  * need to be particularly large.  This is done because we must store the
1562  * caller's idea of the map size to properly unmap.
1563  */
1564 struct vmmap {
1565 	LIST_ENTRY(vmmap)	vm_next;
1566 	void 			*vm_addr;
1567 	unsigned long		vm_size;
1568 };
1569 
1570 struct vmmaphd {
1571 	struct vmmap *lh_first;
1572 };
1573 #define	VMMAP_HASH_SIZE	64
1574 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1575 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1576 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1577 static struct mtx vmmaplock;
1578 
1579 static void
1580 vmmap_add(void *addr, unsigned long size)
1581 {
1582 	struct vmmap *vmmap;
1583 
1584 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1585 	mtx_lock(&vmmaplock);
1586 	vmmap->vm_size = size;
1587 	vmmap->vm_addr = addr;
1588 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1589 	mtx_unlock(&vmmaplock);
1590 }
1591 
1592 static struct vmmap *
1593 vmmap_remove(void *addr)
1594 {
1595 	struct vmmap *vmmap;
1596 
1597 	mtx_lock(&vmmaplock);
1598 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1599 		if (vmmap->vm_addr == addr)
1600 			break;
1601 	if (vmmap)
1602 		LIST_REMOVE(vmmap, vm_next);
1603 	mtx_unlock(&vmmaplock);
1604 
1605 	return (vmmap);
1606 }
1607 
1608 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
1609 void *
1610 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1611 {
1612 	void *addr;
1613 
1614 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1615 	if (addr == NULL)
1616 		return (NULL);
1617 	vmmap_add(addr, size);
1618 
1619 	return (addr);
1620 }
1621 #endif
1622 
1623 void
1624 iounmap(void *addr)
1625 {
1626 	struct vmmap *vmmap;
1627 
1628 	vmmap = vmmap_remove(addr);
1629 	if (vmmap == NULL)
1630 		return;
1631 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__)
1632 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1633 #endif
1634 	kfree(vmmap);
1635 }
1636 
1637 
1638 void *
1639 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1640 {
1641 	vm_offset_t off;
1642 	size_t size;
1643 
1644 	size = count * PAGE_SIZE;
1645 	off = kva_alloc(size);
1646 	if (off == 0)
1647 		return (NULL);
1648 	vmmap_add((void *)off, size);
1649 	pmap_qenter(off, pages, count);
1650 
1651 	return ((void *)off);
1652 }
1653 
1654 void
1655 vunmap(void *addr)
1656 {
1657 	struct vmmap *vmmap;
1658 
1659 	vmmap = vmmap_remove(addr);
1660 	if (vmmap == NULL)
1661 		return;
1662 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1663 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1664 	kfree(vmmap);
1665 }
1666 
1667 char *
1668 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1669 {
1670 	unsigned int len;
1671 	char *p;
1672 	va_list aq;
1673 
1674 	va_copy(aq, ap);
1675 	len = vsnprintf(NULL, 0, fmt, aq);
1676 	va_end(aq);
1677 
1678 	p = kmalloc(len + 1, gfp);
1679 	if (p != NULL)
1680 		vsnprintf(p, len + 1, fmt, ap);
1681 
1682 	return (p);
1683 }
1684 
1685 char *
1686 kasprintf(gfp_t gfp, const char *fmt, ...)
1687 {
1688 	va_list ap;
1689 	char *p;
1690 
1691 	va_start(ap, fmt);
1692 	p = kvasprintf(gfp, fmt, ap);
1693 	va_end(ap);
1694 
1695 	return (p);
1696 }
1697 
1698 static void
1699 linux_timer_callback_wrapper(void *context)
1700 {
1701 	struct timer_list *timer;
1702 
1703 	linux_set_current(curthread);
1704 
1705 	timer = context;
1706 	timer->function(timer->data);
1707 }
1708 
1709 void
1710 mod_timer(struct timer_list *timer, int expires)
1711 {
1712 
1713 	timer->expires = expires;
1714 	callout_reset(&timer->timer_callout,
1715 	    linux_timer_jiffies_until(expires),
1716 	    &linux_timer_callback_wrapper, timer);
1717 }
1718 
1719 void
1720 add_timer(struct timer_list *timer)
1721 {
1722 
1723 	callout_reset(&timer->timer_callout,
1724 	    linux_timer_jiffies_until(timer->expires),
1725 	    &linux_timer_callback_wrapper, timer);
1726 }
1727 
1728 void
1729 add_timer_on(struct timer_list *timer, int cpu)
1730 {
1731 
1732 	callout_reset_on(&timer->timer_callout,
1733 	    linux_timer_jiffies_until(timer->expires),
1734 	    &linux_timer_callback_wrapper, timer, cpu);
1735 }
1736 
1737 static void
1738 linux_timer_init(void *arg)
1739 {
1740 
1741 	/*
1742 	 * Compute an internal HZ value which can divide 2**32 to
1743 	 * avoid timer rounding problems when the tick value wraps
1744 	 * around 2**32:
1745 	 */
1746 	linux_timer_hz_mask = 1;
1747 	while (linux_timer_hz_mask < (unsigned long)hz)
1748 		linux_timer_hz_mask *= 2;
1749 	linux_timer_hz_mask--;
1750 }
1751 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
1752 
1753 void
1754 linux_complete_common(struct completion *c, int all)
1755 {
1756 	int wakeup_swapper;
1757 
1758 	sleepq_lock(c);
1759 	c->done++;
1760 	if (all)
1761 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
1762 	else
1763 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
1764 	sleepq_release(c);
1765 	if (wakeup_swapper)
1766 		kick_proc0();
1767 }
1768 
1769 /*
1770  * Indefinite wait for done != 0 with or without signals.
1771  */
1772 int
1773 linux_wait_for_common(struct completion *c, int flags)
1774 {
1775 	int error;
1776 
1777 	if (SCHEDULER_STOPPED())
1778 		return (0);
1779 
1780 	DROP_GIANT();
1781 
1782 	if (flags != 0)
1783 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
1784 	else
1785 		flags = SLEEPQ_SLEEP;
1786 	error = 0;
1787 	for (;;) {
1788 		sleepq_lock(c);
1789 		if (c->done)
1790 			break;
1791 		sleepq_add(c, NULL, "completion", flags, 0);
1792 		if (flags & SLEEPQ_INTERRUPTIBLE) {
1793 			if (sleepq_wait_sig(c, 0) != 0) {
1794 				error = -ERESTARTSYS;
1795 				goto intr;
1796 			}
1797 		} else
1798 			sleepq_wait(c, 0);
1799 	}
1800 	c->done--;
1801 	sleepq_release(c);
1802 
1803 intr:
1804 	PICKUP_GIANT();
1805 
1806 	return (error);
1807 }
1808 
1809 /*
1810  * Time limited wait for done != 0 with or without signals.
1811  */
1812 int
1813 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
1814 {
1815 	int end = jiffies + timeout;
1816 	int error;
1817 	int ret;
1818 
1819 	if (SCHEDULER_STOPPED())
1820 		return (0);
1821 
1822 	DROP_GIANT();
1823 
1824 	if (flags != 0)
1825 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
1826 	else
1827 		flags = SLEEPQ_SLEEP;
1828 
1829 	error = 0;
1830 	ret = 0;
1831 	for (;;) {
1832 		sleepq_lock(c);
1833 		if (c->done)
1834 			break;
1835 		sleepq_add(c, NULL, "completion", flags, 0);
1836 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
1837 		if (flags & SLEEPQ_INTERRUPTIBLE)
1838 			ret = sleepq_timedwait_sig(c, 0);
1839 		else
1840 			ret = sleepq_timedwait(c, 0);
1841 		if (ret != 0) {
1842 			/* check for timeout or signal */
1843 			if (ret == EWOULDBLOCK)
1844 				error = 0;
1845 			else
1846 				error = -ERESTARTSYS;
1847 			goto intr;
1848 		}
1849 	}
1850 	c->done--;
1851 	sleepq_release(c);
1852 
1853 intr:
1854 	PICKUP_GIANT();
1855 
1856 	/* return how many jiffies are left */
1857 	return (ret != 0 ? error : linux_timer_jiffies_until(end));
1858 }
1859 
1860 int
1861 linux_try_wait_for_completion(struct completion *c)
1862 {
1863 	int isdone;
1864 
1865 	isdone = 1;
1866 	sleepq_lock(c);
1867 	if (c->done)
1868 		c->done--;
1869 	else
1870 		isdone = 0;
1871 	sleepq_release(c);
1872 	return (isdone);
1873 }
1874 
1875 int
1876 linux_completion_done(struct completion *c)
1877 {
1878 	int isdone;
1879 
1880 	isdone = 1;
1881 	sleepq_lock(c);
1882 	if (c->done == 0)
1883 		isdone = 0;
1884 	sleepq_release(c);
1885 	return (isdone);
1886 }
1887 
1888 static void
1889 linux_cdev_release(struct kobject *kobj)
1890 {
1891 	struct linux_cdev *cdev;
1892 	struct kobject *parent;
1893 
1894 	cdev = container_of(kobj, struct linux_cdev, kobj);
1895 	parent = kobj->parent;
1896 	if (cdev->cdev)
1897 		destroy_dev(cdev->cdev);
1898 	kfree(cdev);
1899 	kobject_put(parent);
1900 }
1901 
1902 static void
1903 linux_cdev_static_release(struct kobject *kobj)
1904 {
1905 	struct linux_cdev *cdev;
1906 	struct kobject *parent;
1907 
1908 	cdev = container_of(kobj, struct linux_cdev, kobj);
1909 	parent = kobj->parent;
1910 	if (cdev->cdev)
1911 		destroy_dev(cdev->cdev);
1912 	kobject_put(parent);
1913 }
1914 
1915 const struct kobj_type linux_cdev_ktype = {
1916 	.release = linux_cdev_release,
1917 };
1918 
1919 const struct kobj_type linux_cdev_static_ktype = {
1920 	.release = linux_cdev_static_release,
1921 };
1922 
1923 static void
1924 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
1925 {
1926 	struct notifier_block *nb;
1927 
1928 	nb = arg;
1929 	if (linkstate == LINK_STATE_UP)
1930 		nb->notifier_call(nb, NETDEV_UP, ifp);
1931 	else
1932 		nb->notifier_call(nb, NETDEV_DOWN, ifp);
1933 }
1934 
1935 static void
1936 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
1937 {
1938 	struct notifier_block *nb;
1939 
1940 	nb = arg;
1941 	nb->notifier_call(nb, NETDEV_REGISTER, ifp);
1942 }
1943 
1944 static void
1945 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
1946 {
1947 	struct notifier_block *nb;
1948 
1949 	nb = arg;
1950 	nb->notifier_call(nb, NETDEV_UNREGISTER, ifp);
1951 }
1952 
1953 static void
1954 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
1955 {
1956 	struct notifier_block *nb;
1957 
1958 	nb = arg;
1959 	nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp);
1960 }
1961 
1962 static void
1963 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
1964 {
1965 	struct notifier_block *nb;
1966 
1967 	nb = arg;
1968 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp);
1969 }
1970 
1971 int
1972 register_netdevice_notifier(struct notifier_block *nb)
1973 {
1974 
1975 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
1976 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
1977 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
1978 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
1979 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
1980 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
1981 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
1982 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
1983 
1984 	return (0);
1985 }
1986 
1987 int
1988 register_inetaddr_notifier(struct notifier_block *nb)
1989 {
1990 
1991         nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
1992             ifaddr_event, linux_handle_ifaddr_event, nb, 0);
1993         return (0);
1994 }
1995 
1996 int
1997 unregister_netdevice_notifier(struct notifier_block *nb)
1998 {
1999 
2000         EVENTHANDLER_DEREGISTER(ifnet_link_event,
2001 	    nb->tags[NETDEV_UP]);
2002         EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2003 	    nb->tags[NETDEV_REGISTER]);
2004         EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2005 	    nb->tags[NETDEV_UNREGISTER]);
2006         EVENTHANDLER_DEREGISTER(iflladdr_event,
2007 	    nb->tags[NETDEV_CHANGEADDR]);
2008 
2009 	return (0);
2010 }
2011 
2012 int
2013 unregister_inetaddr_notifier(struct notifier_block *nb)
2014 {
2015 
2016         EVENTHANDLER_DEREGISTER(ifaddr_event,
2017             nb->tags[NETDEV_CHANGEIFADDR]);
2018 
2019         return (0);
2020 }
2021 
2022 struct list_sort_thunk {
2023 	int (*cmp)(void *, struct list_head *, struct list_head *);
2024 	void *priv;
2025 };
2026 
2027 static inline int
2028 linux_le_cmp(void *priv, const void *d1, const void *d2)
2029 {
2030 	struct list_head *le1, *le2;
2031 	struct list_sort_thunk *thunk;
2032 
2033 	thunk = priv;
2034 	le1 = *(__DECONST(struct list_head **, d1));
2035 	le2 = *(__DECONST(struct list_head **, d2));
2036 	return ((thunk->cmp)(thunk->priv, le1, le2));
2037 }
2038 
2039 void
2040 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2041     struct list_head *a, struct list_head *b))
2042 {
2043 	struct list_sort_thunk thunk;
2044 	struct list_head **ar, *le;
2045 	size_t count, i;
2046 
2047 	count = 0;
2048 	list_for_each(le, head)
2049 		count++;
2050 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2051 	i = 0;
2052 	list_for_each(le, head)
2053 		ar[i++] = le;
2054 	thunk.cmp = cmp;
2055 	thunk.priv = priv;
2056 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2057 	INIT_LIST_HEAD(head);
2058 	for (i = 0; i < count; i++)
2059 		list_add_tail(ar[i], head);
2060 	free(ar, M_KMALLOC);
2061 }
2062 
2063 void
2064 linux_irq_handler(void *ent)
2065 {
2066 	struct irq_ent *irqe;
2067 
2068 	linux_set_current(curthread);
2069 
2070 	irqe = ent;
2071 	irqe->handler(irqe->irq, irqe->arg);
2072 }
2073 
2074 #if defined(__i386__) || defined(__amd64__)
2075 int
2076 linux_wbinvd_on_all_cpus(void)
2077 {
2078 
2079 	pmap_invalidate_cache();
2080 	return (0);
2081 }
2082 #endif
2083 
2084 int
2085 linux_on_each_cpu(void callback(void *), void *data)
2086 {
2087 
2088 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2089 	    smp_no_rendezvous_barrier, data);
2090 	return (0);
2091 }
2092 
2093 int
2094 linux_in_atomic(void)
2095 {
2096 
2097 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2098 }
2099 
2100 struct linux_cdev *
2101 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2102 {
2103 	dev_t dev = MKDEV(major, minor);
2104 	struct cdev *cdev;
2105 
2106 	dev_lock();
2107 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2108 		struct linux_cdev *ldev = cdev->si_drv1;
2109 		if (ldev->dev == dev &&
2110 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2111 			break;
2112 		}
2113 	}
2114 	dev_unlock();
2115 
2116 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2117 }
2118 
2119 int
2120 __register_chrdev(unsigned int major, unsigned int baseminor,
2121     unsigned int count, const char *name,
2122     const struct file_operations *fops)
2123 {
2124 	struct linux_cdev *cdev;
2125 	int ret = 0;
2126 	int i;
2127 
2128 	for (i = baseminor; i < baseminor + count; i++) {
2129 		cdev = cdev_alloc();
2130 		cdev_init(cdev, fops);
2131 		kobject_set_name(&cdev->kobj, name);
2132 
2133 		ret = cdev_add(cdev, makedev(major, i), 1);
2134 		if (ret != 0)
2135 			break;
2136 	}
2137 	return (ret);
2138 }
2139 
2140 int
2141 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2142     unsigned int count, const char *name,
2143     const struct file_operations *fops, uid_t uid,
2144     gid_t gid, int mode)
2145 {
2146 	struct linux_cdev *cdev;
2147 	int ret = 0;
2148 	int i;
2149 
2150 	for (i = baseminor; i < baseminor + count; i++) {
2151 		cdev = cdev_alloc();
2152 		cdev_init(cdev, fops);
2153 		kobject_set_name(&cdev->kobj, name);
2154 
2155 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2156 		if (ret != 0)
2157 			break;
2158 	}
2159 	return (ret);
2160 }
2161 
2162 void
2163 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2164     unsigned int count, const char *name)
2165 {
2166 	struct linux_cdev *cdevp;
2167 	int i;
2168 
2169 	for (i = baseminor; i < baseminor + count; i++) {
2170 		cdevp = linux_find_cdev(name, major, i);
2171 		if (cdevp != NULL)
2172 			cdev_del(cdevp);
2173 	}
2174 }
2175 
2176 #if defined(__i386__) || defined(__amd64__)
2177 bool linux_cpu_has_clflush;
2178 #endif
2179 
2180 static void
2181 linux_compat_init(void *arg)
2182 {
2183 	struct sysctl_oid *rootoid;
2184 	int i;
2185 
2186 #if defined(__i386__) || defined(__amd64__)
2187 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2188 #endif
2189 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2190 
2191 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2192 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2193 	kobject_init(&linux_class_root, &linux_class_ktype);
2194 	kobject_set_name(&linux_class_root, "class");
2195 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2196 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2197 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2198 	kobject_set_name(&linux_root_device.kobj, "device");
2199 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2200 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL,
2201 	    "device");
2202 	linux_root_device.bsddev = root_bus;
2203 	linux_class_misc.name = "misc";
2204 	class_register(&linux_class_misc);
2205 	INIT_LIST_HEAD(&pci_drivers);
2206 	INIT_LIST_HEAD(&pci_devices);
2207 	spin_lock_init(&pci_lock);
2208 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2209 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2210 		LIST_INIT(&vmmaphead[i]);
2211 }
2212 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2213 
2214 static void
2215 linux_compat_uninit(void *arg)
2216 {
2217 	linux_kobject_kfree_name(&linux_class_root);
2218 	linux_kobject_kfree_name(&linux_root_device.kobj);
2219 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2220 
2221 	mtx_destroy(&vmmaplock);
2222 	spin_lock_destroy(&pci_lock);
2223 	rw_destroy(&linux_vma_lock);
2224 }
2225 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2226 
2227 /*
2228  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2229  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2230  * used. Assert these types have the same size, else some parts of the
2231  * LinuxKPI may not work like expected:
2232  */
2233 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2234