xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision ac099daf6742ead81ea7ea86351a8ef4e783041b)
1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2018 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/time.h>
55 #include <sys/user.h>
56 
57 #include <vm/vm.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_page.h>
61 #include <vm/vm_pager.h>
62 
63 #include <machine/stdarg.h>
64 
65 #if defined(__i386__) || defined(__amd64__)
66 #include <machine/md_var.h>
67 #endif
68 
69 #include <linux/kobject.h>
70 #include <linux/device.h>
71 #include <linux/slab.h>
72 #include <linux/module.h>
73 #include <linux/moduleparam.h>
74 #include <linux/cdev.h>
75 #include <linux/file.h>
76 #include <linux/sysfs.h>
77 #include <linux/mm.h>
78 #include <linux/io.h>
79 #include <linux/vmalloc.h>
80 #include <linux/netdevice.h>
81 #include <linux/timer.h>
82 #include <linux/interrupt.h>
83 #include <linux/uaccess.h>
84 #include <linux/list.h>
85 #include <linux/kthread.h>
86 #include <linux/kernel.h>
87 #include <linux/compat.h>
88 #include <linux/poll.h>
89 #include <linux/smp.h>
90 #include <linux/wait_bit.h>
91 
92 #if defined(__i386__) || defined(__amd64__)
93 #include <asm/smp.h>
94 #endif
95 
96 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97     "LinuxKPI parameters");
98 
99 int linuxkpi_debug;
100 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
101     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
102 
103 static struct timeval lkpi_net_lastlog;
104 static int lkpi_net_curpps;
105 static int lkpi_net_maxpps = 99;
106 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
107     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
108 
109 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
110 
111 #include <linux/rbtree.h>
112 /* Undo Linux compat changes. */
113 #undef RB_ROOT
114 #undef file
115 #undef cdev
116 #define	RB_ROOT(head)	(head)->rbh_root
117 
118 static void linux_cdev_deref(struct linux_cdev *ldev);
119 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
120 
121 struct kobject linux_class_root;
122 struct device linux_root_device;
123 struct class linux_class_misc;
124 struct list_head pci_drivers;
125 struct list_head pci_devices;
126 spinlock_t pci_lock;
127 
128 unsigned long linux_timer_hz_mask;
129 
130 wait_queue_head_t linux_bit_waitq;
131 wait_queue_head_t linux_var_waitq;
132 
133 int
134 panic_cmp(struct rb_node *one, struct rb_node *two)
135 {
136 	panic("no cmp");
137 }
138 
139 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
140 
141 int
142 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
143 {
144 	va_list tmp_va;
145 	int len;
146 	char *old;
147 	char *name;
148 	char dummy;
149 
150 	old = kobj->name;
151 
152 	if (old && fmt == NULL)
153 		return (0);
154 
155 	/* compute length of string */
156 	va_copy(tmp_va, args);
157 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
158 	va_end(tmp_va);
159 
160 	/* account for zero termination */
161 	len++;
162 
163 	/* check for error */
164 	if (len < 1)
165 		return (-EINVAL);
166 
167 	/* allocate memory for string */
168 	name = kzalloc(len, GFP_KERNEL);
169 	if (name == NULL)
170 		return (-ENOMEM);
171 	vsnprintf(name, len, fmt, args);
172 	kobj->name = name;
173 
174 	/* free old string */
175 	kfree(old);
176 
177 	/* filter new string */
178 	for (; *name != '\0'; name++)
179 		if (*name == '/')
180 			*name = '!';
181 	return (0);
182 }
183 
184 int
185 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
186 {
187 	va_list args;
188 	int error;
189 
190 	va_start(args, fmt);
191 	error = kobject_set_name_vargs(kobj, fmt, args);
192 	va_end(args);
193 
194 	return (error);
195 }
196 
197 static int
198 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
199 {
200 	const struct kobj_type *t;
201 	int error;
202 
203 	kobj->parent = parent;
204 	error = sysfs_create_dir(kobj);
205 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
206 		struct attribute **attr;
207 		t = kobj->ktype;
208 
209 		for (attr = t->default_attrs; *attr != NULL; attr++) {
210 			error = sysfs_create_file(kobj, *attr);
211 			if (error)
212 				break;
213 		}
214 		if (error)
215 			sysfs_remove_dir(kobj);
216 	}
217 	return (error);
218 }
219 
220 int
221 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
222 {
223 	va_list args;
224 	int error;
225 
226 	va_start(args, fmt);
227 	error = kobject_set_name_vargs(kobj, fmt, args);
228 	va_end(args);
229 	if (error)
230 		return (error);
231 
232 	return kobject_add_complete(kobj, parent);
233 }
234 
235 void
236 linux_kobject_release(struct kref *kref)
237 {
238 	struct kobject *kobj;
239 	char *name;
240 
241 	kobj = container_of(kref, struct kobject, kref);
242 	sysfs_remove_dir(kobj);
243 	name = kobj->name;
244 	if (kobj->ktype && kobj->ktype->release)
245 		kobj->ktype->release(kobj);
246 	kfree(name);
247 }
248 
249 static void
250 linux_kobject_kfree(struct kobject *kobj)
251 {
252 	kfree(kobj);
253 }
254 
255 static void
256 linux_kobject_kfree_name(struct kobject *kobj)
257 {
258 	if (kobj) {
259 		kfree(kobj->name);
260 	}
261 }
262 
263 const struct kobj_type linux_kfree_type = {
264 	.release = linux_kobject_kfree
265 };
266 
267 static void
268 linux_device_release(struct device *dev)
269 {
270 	pr_debug("linux_device_release: %s\n", dev_name(dev));
271 	kfree(dev);
272 }
273 
274 static ssize_t
275 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
276 {
277 	struct class_attribute *dattr;
278 	ssize_t error;
279 
280 	dattr = container_of(attr, struct class_attribute, attr);
281 	error = -EIO;
282 	if (dattr->show)
283 		error = dattr->show(container_of(kobj, struct class, kobj),
284 		    dattr, buf);
285 	return (error);
286 }
287 
288 static ssize_t
289 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
290     size_t count)
291 {
292 	struct class_attribute *dattr;
293 	ssize_t error;
294 
295 	dattr = container_of(attr, struct class_attribute, attr);
296 	error = -EIO;
297 	if (dattr->store)
298 		error = dattr->store(container_of(kobj, struct class, kobj),
299 		    dattr, buf, count);
300 	return (error);
301 }
302 
303 static void
304 linux_class_release(struct kobject *kobj)
305 {
306 	struct class *class;
307 
308 	class = container_of(kobj, struct class, kobj);
309 	if (class->class_release)
310 		class->class_release(class);
311 }
312 
313 static const struct sysfs_ops linux_class_sysfs = {
314 	.show  = linux_class_show,
315 	.store = linux_class_store,
316 };
317 
318 const struct kobj_type linux_class_ktype = {
319 	.release = linux_class_release,
320 	.sysfs_ops = &linux_class_sysfs
321 };
322 
323 static void
324 linux_dev_release(struct kobject *kobj)
325 {
326 	struct device *dev;
327 
328 	dev = container_of(kobj, struct device, kobj);
329 	/* This is the precedence defined by linux. */
330 	if (dev->release)
331 		dev->release(dev);
332 	else if (dev->class && dev->class->dev_release)
333 		dev->class->dev_release(dev);
334 }
335 
336 static ssize_t
337 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
338 {
339 	struct device_attribute *dattr;
340 	ssize_t error;
341 
342 	dattr = container_of(attr, struct device_attribute, attr);
343 	error = -EIO;
344 	if (dattr->show)
345 		error = dattr->show(container_of(kobj, struct device, kobj),
346 		    dattr, buf);
347 	return (error);
348 }
349 
350 static ssize_t
351 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
352     size_t count)
353 {
354 	struct device_attribute *dattr;
355 	ssize_t error;
356 
357 	dattr = container_of(attr, struct device_attribute, attr);
358 	error = -EIO;
359 	if (dattr->store)
360 		error = dattr->store(container_of(kobj, struct device, kobj),
361 		    dattr, buf, count);
362 	return (error);
363 }
364 
365 static const struct sysfs_ops linux_dev_sysfs = {
366 	.show  = linux_dev_show,
367 	.store = linux_dev_store,
368 };
369 
370 const struct kobj_type linux_dev_ktype = {
371 	.release = linux_dev_release,
372 	.sysfs_ops = &linux_dev_sysfs
373 };
374 
375 struct device *
376 device_create(struct class *class, struct device *parent, dev_t devt,
377     void *drvdata, const char *fmt, ...)
378 {
379 	struct device *dev;
380 	va_list args;
381 
382 	dev = kzalloc(sizeof(*dev), M_WAITOK);
383 	dev->parent = parent;
384 	dev->class = class;
385 	dev->devt = devt;
386 	dev->driver_data = drvdata;
387 	dev->release = linux_device_release;
388 	va_start(args, fmt);
389 	kobject_set_name_vargs(&dev->kobj, fmt, args);
390 	va_end(args);
391 	device_register(dev);
392 
393 	return (dev);
394 }
395 
396 int
397 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
398     struct kobject *parent, const char *fmt, ...)
399 {
400 	va_list args;
401 	int error;
402 
403 	kobject_init(kobj, ktype);
404 	kobj->ktype = ktype;
405 	kobj->parent = parent;
406 	kobj->name = NULL;
407 
408 	va_start(args, fmt);
409 	error = kobject_set_name_vargs(kobj, fmt, args);
410 	va_end(args);
411 	if (error)
412 		return (error);
413 	return kobject_add_complete(kobj, parent);
414 }
415 
416 static void
417 linux_kq_lock(void *arg)
418 {
419 	spinlock_t *s = arg;
420 
421 	spin_lock(s);
422 }
423 static void
424 linux_kq_unlock(void *arg)
425 {
426 	spinlock_t *s = arg;
427 
428 	spin_unlock(s);
429 }
430 
431 static void
432 linux_kq_assert_lock(void *arg, int what)
433 {
434 #ifdef INVARIANTS
435 	spinlock_t *s = arg;
436 
437 	if (what == LA_LOCKED)
438 		mtx_assert(&s->m, MA_OWNED);
439 	else
440 		mtx_assert(&s->m, MA_NOTOWNED);
441 #endif
442 }
443 
444 static void
445 linux_file_kqfilter_poll(struct linux_file *, int);
446 
447 struct linux_file *
448 linux_file_alloc(void)
449 {
450 	struct linux_file *filp;
451 
452 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
453 
454 	/* set initial refcount */
455 	filp->f_count = 1;
456 
457 	/* setup fields needed by kqueue support */
458 	spin_lock_init(&filp->f_kqlock);
459 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
460 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
461 
462 	return (filp);
463 }
464 
465 void
466 linux_file_free(struct linux_file *filp)
467 {
468 	if (filp->_file == NULL) {
469 		if (filp->f_shmem != NULL)
470 			vm_object_deallocate(filp->f_shmem);
471 		kfree(filp);
472 	} else {
473 		/*
474 		 * The close method of the character device or file
475 		 * will free the linux_file structure:
476 		 */
477 		_fdrop(filp->_file, curthread);
478 	}
479 }
480 
481 static int
482 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
483     vm_page_t *mres)
484 {
485 	struct vm_area_struct *vmap;
486 
487 	vmap = linux_cdev_handle_find(vm_obj->handle);
488 
489 	MPASS(vmap != NULL);
490 	MPASS(vmap->vm_private_data == vm_obj->handle);
491 
492 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
493 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
494 		vm_page_t page;
495 
496 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
497 			/*
498 			 * If the passed in result page is a fake
499 			 * page, update it with the new physical
500 			 * address.
501 			 */
502 			page = *mres;
503 			vm_page_updatefake(page, paddr, vm_obj->memattr);
504 		} else {
505 			/*
506 			 * Replace the passed in "mres" page with our
507 			 * own fake page and free up the all of the
508 			 * original pages.
509 			 */
510 			VM_OBJECT_WUNLOCK(vm_obj);
511 			page = vm_page_getfake(paddr, vm_obj->memattr);
512 			VM_OBJECT_WLOCK(vm_obj);
513 
514 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
515 			*mres = page;
516 		}
517 		vm_page_valid(page);
518 		return (VM_PAGER_OK);
519 	}
520 	return (VM_PAGER_FAIL);
521 }
522 
523 static int
524 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
525     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
526 {
527 	struct vm_area_struct *vmap;
528 	int err;
529 
530 	/* get VM area structure */
531 	vmap = linux_cdev_handle_find(vm_obj->handle);
532 	MPASS(vmap != NULL);
533 	MPASS(vmap->vm_private_data == vm_obj->handle);
534 
535 	VM_OBJECT_WUNLOCK(vm_obj);
536 
537 	linux_set_current(curthread);
538 
539 	down_write(&vmap->vm_mm->mmap_sem);
540 	if (unlikely(vmap->vm_ops == NULL)) {
541 		err = VM_FAULT_SIGBUS;
542 	} else {
543 		struct vm_fault vmf;
544 
545 		/* fill out VM fault structure */
546 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
547 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
548 		vmf.pgoff = 0;
549 		vmf.page = NULL;
550 		vmf.vma = vmap;
551 
552 		vmap->vm_pfn_count = 0;
553 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
554 		vmap->vm_obj = vm_obj;
555 
556 		err = vmap->vm_ops->fault(vmap, &vmf);
557 
558 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
559 			kern_yield(PRI_USER);
560 			err = vmap->vm_ops->fault(vmap, &vmf);
561 		}
562 	}
563 
564 	/* translate return code */
565 	switch (err) {
566 	case VM_FAULT_OOM:
567 		err = VM_PAGER_AGAIN;
568 		break;
569 	case VM_FAULT_SIGBUS:
570 		err = VM_PAGER_BAD;
571 		break;
572 	case VM_FAULT_NOPAGE:
573 		/*
574 		 * By contract the fault handler will return having
575 		 * busied all the pages itself. If pidx is already
576 		 * found in the object, it will simply xbusy the first
577 		 * page and return with vm_pfn_count set to 1.
578 		 */
579 		*first = vmap->vm_pfn_first;
580 		*last = *first + vmap->vm_pfn_count - 1;
581 		err = VM_PAGER_OK;
582 		break;
583 	default:
584 		err = VM_PAGER_ERROR;
585 		break;
586 	}
587 	up_write(&vmap->vm_mm->mmap_sem);
588 	VM_OBJECT_WLOCK(vm_obj);
589 	return (err);
590 }
591 
592 static struct rwlock linux_vma_lock;
593 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
594     TAILQ_HEAD_INITIALIZER(linux_vma_head);
595 
596 static void
597 linux_cdev_handle_free(struct vm_area_struct *vmap)
598 {
599 	/* Drop reference on vm_file */
600 	if (vmap->vm_file != NULL)
601 		fput(vmap->vm_file);
602 
603 	/* Drop reference on mm_struct */
604 	mmput(vmap->vm_mm);
605 
606 	kfree(vmap);
607 }
608 
609 static void
610 linux_cdev_handle_remove(struct vm_area_struct *vmap)
611 {
612 	rw_wlock(&linux_vma_lock);
613 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
614 	rw_wunlock(&linux_vma_lock);
615 }
616 
617 static struct vm_area_struct *
618 linux_cdev_handle_find(void *handle)
619 {
620 	struct vm_area_struct *vmap;
621 
622 	rw_rlock(&linux_vma_lock);
623 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
624 		if (vmap->vm_private_data == handle)
625 			break;
626 	}
627 	rw_runlock(&linux_vma_lock);
628 	return (vmap);
629 }
630 
631 static int
632 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
633 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
634 {
635 
636 	MPASS(linux_cdev_handle_find(handle) != NULL);
637 	*color = 0;
638 	return (0);
639 }
640 
641 static void
642 linux_cdev_pager_dtor(void *handle)
643 {
644 	const struct vm_operations_struct *vm_ops;
645 	struct vm_area_struct *vmap;
646 
647 	vmap = linux_cdev_handle_find(handle);
648 	MPASS(vmap != NULL);
649 
650 	/*
651 	 * Remove handle before calling close operation to prevent
652 	 * other threads from reusing the handle pointer.
653 	 */
654 	linux_cdev_handle_remove(vmap);
655 
656 	down_write(&vmap->vm_mm->mmap_sem);
657 	vm_ops = vmap->vm_ops;
658 	if (likely(vm_ops != NULL))
659 		vm_ops->close(vmap);
660 	up_write(&vmap->vm_mm->mmap_sem);
661 
662 	linux_cdev_handle_free(vmap);
663 }
664 
665 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
666   {
667 	/* OBJT_MGTDEVICE */
668 	.cdev_pg_populate	= linux_cdev_pager_populate,
669 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
670 	.cdev_pg_dtor	= linux_cdev_pager_dtor
671   },
672   {
673 	/* OBJT_DEVICE */
674 	.cdev_pg_fault	= linux_cdev_pager_fault,
675 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
676 	.cdev_pg_dtor	= linux_cdev_pager_dtor
677   },
678 };
679 
680 int
681 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
682     unsigned long size)
683 {
684 	vm_object_t obj;
685 	vm_page_t m;
686 
687 	obj = vma->vm_obj;
688 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
689 		return (-ENOTSUP);
690 	VM_OBJECT_RLOCK(obj);
691 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
692 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
693 	    m = TAILQ_NEXT(m, listq))
694 		pmap_remove_all(m);
695 	VM_OBJECT_RUNLOCK(obj);
696 	return (0);
697 }
698 
699 static struct file_operations dummy_ldev_ops = {
700 	/* XXXKIB */
701 };
702 
703 static struct linux_cdev dummy_ldev = {
704 	.ops = &dummy_ldev_ops,
705 };
706 
707 #define	LDEV_SI_DTR	0x0001
708 #define	LDEV_SI_REF	0x0002
709 
710 static void
711 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
712     struct linux_cdev **dev)
713 {
714 	struct linux_cdev *ldev;
715 	u_int siref;
716 
717 	ldev = filp->f_cdev;
718 	*fop = filp->f_op;
719 	if (ldev != NULL) {
720 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
721 			refcount_acquire(&ldev->refs);
722 		} else {
723 			for (siref = ldev->siref;;) {
724 				if ((siref & LDEV_SI_DTR) != 0) {
725 					ldev = &dummy_ldev;
726 					*fop = ldev->ops;
727 					siref = ldev->siref;
728 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
729 				} else if (atomic_fcmpset_int(&ldev->siref,
730 				    &siref, siref + LDEV_SI_REF)) {
731 					break;
732 				}
733 			}
734 		}
735 	}
736 	*dev = ldev;
737 }
738 
739 static void
740 linux_drop_fop(struct linux_cdev *ldev)
741 {
742 
743 	if (ldev == NULL)
744 		return;
745 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
746 		linux_cdev_deref(ldev);
747 	} else {
748 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
749 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
750 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
751 	}
752 }
753 
754 #define	OPW(fp,td,code) ({			\
755 	struct file *__fpop;			\
756 	__typeof(code) __retval;		\
757 						\
758 	__fpop = (td)->td_fpop;			\
759 	(td)->td_fpop = (fp);			\
760 	__retval = (code);			\
761 	(td)->td_fpop = __fpop;			\
762 	__retval;				\
763 })
764 
765 static int
766 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
767     struct file *file)
768 {
769 	struct linux_cdev *ldev;
770 	struct linux_file *filp;
771 	const struct file_operations *fop;
772 	int error;
773 
774 	ldev = dev->si_drv1;
775 
776 	filp = linux_file_alloc();
777 	filp->f_dentry = &filp->f_dentry_store;
778 	filp->f_op = ldev->ops;
779 	filp->f_mode = file->f_flag;
780 	filp->f_flags = file->f_flag;
781 	filp->f_vnode = file->f_vnode;
782 	filp->_file = file;
783 	refcount_acquire(&ldev->refs);
784 	filp->f_cdev = ldev;
785 
786 	linux_set_current(td);
787 	linux_get_fop(filp, &fop, &ldev);
788 
789 	if (fop->open != NULL) {
790 		error = -fop->open(file->f_vnode, filp);
791 		if (error != 0) {
792 			linux_drop_fop(ldev);
793 			linux_cdev_deref(filp->f_cdev);
794 			kfree(filp);
795 			return (error);
796 		}
797 	}
798 
799 	/* hold on to the vnode - used for fstat() */
800 	vhold(filp->f_vnode);
801 
802 	/* release the file from devfs */
803 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
804 	linux_drop_fop(ldev);
805 	return (ENXIO);
806 }
807 
808 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
809 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
810 
811 static inline int
812 linux_remap_address(void **uaddr, size_t len)
813 {
814 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
815 
816 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
817 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
818 		struct task_struct *pts = current;
819 		if (pts == NULL) {
820 			*uaddr = NULL;
821 			return (1);
822 		}
823 
824 		/* compute data offset */
825 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
826 
827 		/* check that length is within bounds */
828 		if ((len > IOCPARM_MAX) ||
829 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
830 			*uaddr = NULL;
831 			return (1);
832 		}
833 
834 		/* re-add kernel buffer address */
835 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
836 
837 		/* update address location */
838 		*uaddr = (void *)uaddr_val;
839 		return (1);
840 	}
841 	return (0);
842 }
843 
844 int
845 linux_copyin(const void *uaddr, void *kaddr, size_t len)
846 {
847 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
848 		if (uaddr == NULL)
849 			return (-EFAULT);
850 		memcpy(kaddr, uaddr, len);
851 		return (0);
852 	}
853 	return (-copyin(uaddr, kaddr, len));
854 }
855 
856 int
857 linux_copyout(const void *kaddr, void *uaddr, size_t len)
858 {
859 	if (linux_remap_address(&uaddr, len)) {
860 		if (uaddr == NULL)
861 			return (-EFAULT);
862 		memcpy(uaddr, kaddr, len);
863 		return (0);
864 	}
865 	return (-copyout(kaddr, uaddr, len));
866 }
867 
868 size_t
869 linux_clear_user(void *_uaddr, size_t _len)
870 {
871 	uint8_t *uaddr = _uaddr;
872 	size_t len = _len;
873 
874 	/* make sure uaddr is aligned before going into the fast loop */
875 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
876 		if (subyte(uaddr, 0))
877 			return (_len);
878 		uaddr++;
879 		len--;
880 	}
881 
882 	/* zero 8 bytes at a time */
883 	while (len > 7) {
884 #ifdef __LP64__
885 		if (suword64(uaddr, 0))
886 			return (_len);
887 #else
888 		if (suword32(uaddr, 0))
889 			return (_len);
890 		if (suword32(uaddr + 4, 0))
891 			return (_len);
892 #endif
893 		uaddr += 8;
894 		len -= 8;
895 	}
896 
897 	/* zero fill end, if any */
898 	while (len > 0) {
899 		if (subyte(uaddr, 0))
900 			return (_len);
901 		uaddr++;
902 		len--;
903 	}
904 	return (0);
905 }
906 
907 int
908 linux_access_ok(const void *uaddr, size_t len)
909 {
910 	uintptr_t saddr;
911 	uintptr_t eaddr;
912 
913 	/* get start and end address */
914 	saddr = (uintptr_t)uaddr;
915 	eaddr = (uintptr_t)uaddr + len;
916 
917 	/* verify addresses are valid for userspace */
918 	return ((saddr == eaddr) ||
919 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
920 }
921 
922 /*
923  * This function should return either EINTR or ERESTART depending on
924  * the signal type sent to this thread:
925  */
926 static int
927 linux_get_error(struct task_struct *task, int error)
928 {
929 	/* check for signal type interrupt code */
930 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
931 		error = -linux_schedule_get_interrupt_value(task);
932 		if (error == 0)
933 			error = EINTR;
934 	}
935 	return (error);
936 }
937 
938 static int
939 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
940     const struct file_operations *fop, u_long cmd, caddr_t data,
941     struct thread *td)
942 {
943 	struct task_struct *task = current;
944 	unsigned size;
945 	int error;
946 
947 	size = IOCPARM_LEN(cmd);
948 	/* refer to logic in sys_ioctl() */
949 	if (size > 0) {
950 		/*
951 		 * Setup hint for linux_copyin() and linux_copyout().
952 		 *
953 		 * Background: Linux code expects a user-space address
954 		 * while FreeBSD supplies a kernel-space address.
955 		 */
956 		task->bsd_ioctl_data = data;
957 		task->bsd_ioctl_len = size;
958 		data = (void *)LINUX_IOCTL_MIN_PTR;
959 	} else {
960 		/* fetch user-space pointer */
961 		data = *(void **)data;
962 	}
963 #if defined(__amd64__)
964 	if (td->td_proc->p_elf_machine == EM_386) {
965 		/* try the compat IOCTL handler first */
966 		if (fop->compat_ioctl != NULL) {
967 			error = -OPW(fp, td, fop->compat_ioctl(filp,
968 			    cmd, (u_long)data));
969 		} else {
970 			error = ENOTTY;
971 		}
972 
973 		/* fallback to the regular IOCTL handler, if any */
974 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
975 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
976 			    cmd, (u_long)data));
977 		}
978 	} else
979 #endif
980 	{
981 		if (fop->unlocked_ioctl != NULL) {
982 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
983 			    cmd, (u_long)data));
984 		} else {
985 			error = ENOTTY;
986 		}
987 	}
988 	if (size > 0) {
989 		task->bsd_ioctl_data = NULL;
990 		task->bsd_ioctl_len = 0;
991 	}
992 
993 	if (error == EWOULDBLOCK) {
994 		/* update kqfilter status, if any */
995 		linux_file_kqfilter_poll(filp,
996 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
997 	} else {
998 		error = linux_get_error(task, error);
999 	}
1000 	return (error);
1001 }
1002 
1003 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
1004 
1005 /*
1006  * This function atomically updates the poll wakeup state and returns
1007  * the previous state at the time of update.
1008  */
1009 static uint8_t
1010 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1011 {
1012 	int c, old;
1013 
1014 	c = v->counter;
1015 
1016 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1017 		c = old;
1018 
1019 	return (c);
1020 }
1021 
1022 static int
1023 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1024 {
1025 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1026 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1027 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1028 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1029 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1030 	};
1031 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1032 
1033 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1034 	case LINUX_FWQ_STATE_QUEUED:
1035 		linux_poll_wakeup(filp);
1036 		return (1);
1037 	default:
1038 		return (0);
1039 	}
1040 }
1041 
1042 void
1043 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1044 {
1045 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1046 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1047 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1048 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1049 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1050 	};
1051 
1052 	/* check if we are called inside the select system call */
1053 	if (p == LINUX_POLL_TABLE_NORMAL)
1054 		selrecord(curthread, &filp->f_selinfo);
1055 
1056 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1057 	case LINUX_FWQ_STATE_INIT:
1058 		/* NOTE: file handles can only belong to one wait-queue */
1059 		filp->f_wait_queue.wqh = wqh;
1060 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1061 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1062 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1063 		break;
1064 	default:
1065 		break;
1066 	}
1067 }
1068 
1069 static void
1070 linux_poll_wait_dequeue(struct linux_file *filp)
1071 {
1072 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1073 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1074 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1075 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1076 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1077 	};
1078 
1079 	seldrain(&filp->f_selinfo);
1080 
1081 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1082 	case LINUX_FWQ_STATE_NOT_READY:
1083 	case LINUX_FWQ_STATE_QUEUED:
1084 	case LINUX_FWQ_STATE_READY:
1085 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1086 		break;
1087 	default:
1088 		break;
1089 	}
1090 }
1091 
1092 void
1093 linux_poll_wakeup(struct linux_file *filp)
1094 {
1095 	/* this function should be NULL-safe */
1096 	if (filp == NULL)
1097 		return;
1098 
1099 	selwakeup(&filp->f_selinfo);
1100 
1101 	spin_lock(&filp->f_kqlock);
1102 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1103 	    LINUX_KQ_FLAG_NEED_WRITE;
1104 
1105 	/* make sure the "knote" gets woken up */
1106 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1107 	spin_unlock(&filp->f_kqlock);
1108 }
1109 
1110 static void
1111 linux_file_kqfilter_detach(struct knote *kn)
1112 {
1113 	struct linux_file *filp = kn->kn_hook;
1114 
1115 	spin_lock(&filp->f_kqlock);
1116 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1117 	spin_unlock(&filp->f_kqlock);
1118 }
1119 
1120 static int
1121 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1122 {
1123 	struct linux_file *filp = kn->kn_hook;
1124 
1125 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1126 
1127 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1128 }
1129 
1130 static int
1131 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1132 {
1133 	struct linux_file *filp = kn->kn_hook;
1134 
1135 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1136 
1137 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1138 }
1139 
1140 static struct filterops linux_dev_kqfiltops_read = {
1141 	.f_isfd = 1,
1142 	.f_detach = linux_file_kqfilter_detach,
1143 	.f_event = linux_file_kqfilter_read_event,
1144 };
1145 
1146 static struct filterops linux_dev_kqfiltops_write = {
1147 	.f_isfd = 1,
1148 	.f_detach = linux_file_kqfilter_detach,
1149 	.f_event = linux_file_kqfilter_write_event,
1150 };
1151 
1152 static void
1153 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1154 {
1155 	struct thread *td;
1156 	const struct file_operations *fop;
1157 	struct linux_cdev *ldev;
1158 	int temp;
1159 
1160 	if ((filp->f_kqflags & kqflags) == 0)
1161 		return;
1162 
1163 	td = curthread;
1164 
1165 	linux_get_fop(filp, &fop, &ldev);
1166 	/* get the latest polling state */
1167 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1168 	linux_drop_fop(ldev);
1169 
1170 	spin_lock(&filp->f_kqlock);
1171 	/* clear kqflags */
1172 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1173 	    LINUX_KQ_FLAG_NEED_WRITE);
1174 	/* update kqflags */
1175 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1176 		if ((temp & POLLIN) != 0)
1177 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1178 		if ((temp & POLLOUT) != 0)
1179 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1180 
1181 		/* make sure the "knote" gets woken up */
1182 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1183 	}
1184 	spin_unlock(&filp->f_kqlock);
1185 }
1186 
1187 static int
1188 linux_file_kqfilter(struct file *file, struct knote *kn)
1189 {
1190 	struct linux_file *filp;
1191 	struct thread *td;
1192 	int error;
1193 
1194 	td = curthread;
1195 	filp = (struct linux_file *)file->f_data;
1196 	filp->f_flags = file->f_flag;
1197 	if (filp->f_op->poll == NULL)
1198 		return (EINVAL);
1199 
1200 	spin_lock(&filp->f_kqlock);
1201 	switch (kn->kn_filter) {
1202 	case EVFILT_READ:
1203 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1204 		kn->kn_fop = &linux_dev_kqfiltops_read;
1205 		kn->kn_hook = filp;
1206 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1207 		error = 0;
1208 		break;
1209 	case EVFILT_WRITE:
1210 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1211 		kn->kn_fop = &linux_dev_kqfiltops_write;
1212 		kn->kn_hook = filp;
1213 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1214 		error = 0;
1215 		break;
1216 	default:
1217 		error = EINVAL;
1218 		break;
1219 	}
1220 	spin_unlock(&filp->f_kqlock);
1221 
1222 	if (error == 0) {
1223 		linux_set_current(td);
1224 
1225 		/* update kqfilter status, if any */
1226 		linux_file_kqfilter_poll(filp,
1227 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1228 	}
1229 	return (error);
1230 }
1231 
1232 static int
1233 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1234     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1235     int nprot, struct thread *td)
1236 {
1237 	struct task_struct *task;
1238 	struct vm_area_struct *vmap;
1239 	struct mm_struct *mm;
1240 	struct linux_file *filp;
1241 	vm_memattr_t attr;
1242 	int error;
1243 
1244 	filp = (struct linux_file *)fp->f_data;
1245 	filp->f_flags = fp->f_flag;
1246 
1247 	if (fop->mmap == NULL)
1248 		return (EOPNOTSUPP);
1249 
1250 	linux_set_current(td);
1251 
1252 	/*
1253 	 * The same VM object might be shared by multiple processes
1254 	 * and the mm_struct is usually freed when a process exits.
1255 	 *
1256 	 * The atomic reference below makes sure the mm_struct is
1257 	 * available as long as the vmap is in the linux_vma_head.
1258 	 */
1259 	task = current;
1260 	mm = task->mm;
1261 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1262 		return (EINVAL);
1263 
1264 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1265 	vmap->vm_start = 0;
1266 	vmap->vm_end = size;
1267 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1268 	vmap->vm_pfn = 0;
1269 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1270 	vmap->vm_ops = NULL;
1271 	vmap->vm_file = get_file(filp);
1272 	vmap->vm_mm = mm;
1273 
1274 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1275 		error = linux_get_error(task, EINTR);
1276 	} else {
1277 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1278 		error = linux_get_error(task, error);
1279 		up_write(&vmap->vm_mm->mmap_sem);
1280 	}
1281 
1282 	if (error != 0) {
1283 		linux_cdev_handle_free(vmap);
1284 		return (error);
1285 	}
1286 
1287 	attr = pgprot2cachemode(vmap->vm_page_prot);
1288 
1289 	if (vmap->vm_ops != NULL) {
1290 		struct vm_area_struct *ptr;
1291 		void *vm_private_data;
1292 		bool vm_no_fault;
1293 
1294 		if (vmap->vm_ops->open == NULL ||
1295 		    vmap->vm_ops->close == NULL ||
1296 		    vmap->vm_private_data == NULL) {
1297 			/* free allocated VM area struct */
1298 			linux_cdev_handle_free(vmap);
1299 			return (EINVAL);
1300 		}
1301 
1302 		vm_private_data = vmap->vm_private_data;
1303 
1304 		rw_wlock(&linux_vma_lock);
1305 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1306 			if (ptr->vm_private_data == vm_private_data)
1307 				break;
1308 		}
1309 		/* check if there is an existing VM area struct */
1310 		if (ptr != NULL) {
1311 			/* check if the VM area structure is invalid */
1312 			if (ptr->vm_ops == NULL ||
1313 			    ptr->vm_ops->open == NULL ||
1314 			    ptr->vm_ops->close == NULL) {
1315 				error = ESTALE;
1316 				vm_no_fault = 1;
1317 			} else {
1318 				error = EEXIST;
1319 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1320 			}
1321 		} else {
1322 			/* insert VM area structure into list */
1323 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1324 			error = 0;
1325 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1326 		}
1327 		rw_wunlock(&linux_vma_lock);
1328 
1329 		if (error != 0) {
1330 			/* free allocated VM area struct */
1331 			linux_cdev_handle_free(vmap);
1332 			/* check for stale VM area struct */
1333 			if (error != EEXIST)
1334 				return (error);
1335 		}
1336 
1337 		/* check if there is no fault handler */
1338 		if (vm_no_fault) {
1339 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1340 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1341 			    td->td_ucred);
1342 		} else {
1343 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1344 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1345 			    td->td_ucred);
1346 		}
1347 
1348 		/* check if allocating the VM object failed */
1349 		if (*object == NULL) {
1350 			if (error == 0) {
1351 				/* remove VM area struct from list */
1352 				linux_cdev_handle_remove(vmap);
1353 				/* free allocated VM area struct */
1354 				linux_cdev_handle_free(vmap);
1355 			}
1356 			return (EINVAL);
1357 		}
1358 	} else {
1359 		struct sglist *sg;
1360 
1361 		sg = sglist_alloc(1, M_WAITOK);
1362 		sglist_append_phys(sg,
1363 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1364 
1365 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1366 		    nprot, 0, td->td_ucred);
1367 
1368 		linux_cdev_handle_free(vmap);
1369 
1370 		if (*object == NULL) {
1371 			sglist_free(sg);
1372 			return (EINVAL);
1373 		}
1374 	}
1375 
1376 	if (attr != VM_MEMATTR_DEFAULT) {
1377 		VM_OBJECT_WLOCK(*object);
1378 		vm_object_set_memattr(*object, attr);
1379 		VM_OBJECT_WUNLOCK(*object);
1380 	}
1381 	*offset = 0;
1382 	return (0);
1383 }
1384 
1385 struct cdevsw linuxcdevsw = {
1386 	.d_version = D_VERSION,
1387 	.d_fdopen = linux_dev_fdopen,
1388 	.d_name = "lkpidev",
1389 };
1390 
1391 static int
1392 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1393     int flags, struct thread *td)
1394 {
1395 	struct linux_file *filp;
1396 	const struct file_operations *fop;
1397 	struct linux_cdev *ldev;
1398 	ssize_t bytes;
1399 	int error;
1400 
1401 	error = 0;
1402 	filp = (struct linux_file *)file->f_data;
1403 	filp->f_flags = file->f_flag;
1404 	/* XXX no support for I/O vectors currently */
1405 	if (uio->uio_iovcnt != 1)
1406 		return (EOPNOTSUPP);
1407 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1408 		return (EINVAL);
1409 	linux_set_current(td);
1410 	linux_get_fop(filp, &fop, &ldev);
1411 	if (fop->read != NULL) {
1412 		bytes = OPW(file, td, fop->read(filp,
1413 		    uio->uio_iov->iov_base,
1414 		    uio->uio_iov->iov_len, &uio->uio_offset));
1415 		if (bytes >= 0) {
1416 			uio->uio_iov->iov_base =
1417 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1418 			uio->uio_iov->iov_len -= bytes;
1419 			uio->uio_resid -= bytes;
1420 		} else {
1421 			error = linux_get_error(current, -bytes);
1422 		}
1423 	} else
1424 		error = ENXIO;
1425 
1426 	/* update kqfilter status, if any */
1427 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1428 	linux_drop_fop(ldev);
1429 
1430 	return (error);
1431 }
1432 
1433 static int
1434 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1435     int flags, struct thread *td)
1436 {
1437 	struct linux_file *filp;
1438 	const struct file_operations *fop;
1439 	struct linux_cdev *ldev;
1440 	ssize_t bytes;
1441 	int error;
1442 
1443 	filp = (struct linux_file *)file->f_data;
1444 	filp->f_flags = file->f_flag;
1445 	/* XXX no support for I/O vectors currently */
1446 	if (uio->uio_iovcnt != 1)
1447 		return (EOPNOTSUPP);
1448 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1449 		return (EINVAL);
1450 	linux_set_current(td);
1451 	linux_get_fop(filp, &fop, &ldev);
1452 	if (fop->write != NULL) {
1453 		bytes = OPW(file, td, fop->write(filp,
1454 		    uio->uio_iov->iov_base,
1455 		    uio->uio_iov->iov_len, &uio->uio_offset));
1456 		if (bytes >= 0) {
1457 			uio->uio_iov->iov_base =
1458 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1459 			uio->uio_iov->iov_len -= bytes;
1460 			uio->uio_resid -= bytes;
1461 			error = 0;
1462 		} else {
1463 			error = linux_get_error(current, -bytes);
1464 		}
1465 	} else
1466 		error = ENXIO;
1467 
1468 	/* update kqfilter status, if any */
1469 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1470 
1471 	linux_drop_fop(ldev);
1472 
1473 	return (error);
1474 }
1475 
1476 static int
1477 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1478     struct thread *td)
1479 {
1480 	struct linux_file *filp;
1481 	const struct file_operations *fop;
1482 	struct linux_cdev *ldev;
1483 	int revents;
1484 
1485 	filp = (struct linux_file *)file->f_data;
1486 	filp->f_flags = file->f_flag;
1487 	linux_set_current(td);
1488 	linux_get_fop(filp, &fop, &ldev);
1489 	if (fop->poll != NULL) {
1490 		revents = OPW(file, td, fop->poll(filp,
1491 		    LINUX_POLL_TABLE_NORMAL)) & events;
1492 	} else {
1493 		revents = 0;
1494 	}
1495 	linux_drop_fop(ldev);
1496 	return (revents);
1497 }
1498 
1499 static int
1500 linux_file_close(struct file *file, struct thread *td)
1501 {
1502 	struct linux_file *filp;
1503 	int (*release)(struct inode *, struct linux_file *);
1504 	const struct file_operations *fop;
1505 	struct linux_cdev *ldev;
1506 	int error;
1507 
1508 	filp = (struct linux_file *)file->f_data;
1509 
1510 	KASSERT(file_count(filp) == 0,
1511 	    ("File refcount(%d) is not zero", file_count(filp)));
1512 
1513 	if (td == NULL)
1514 		td = curthread;
1515 
1516 	error = 0;
1517 	filp->f_flags = file->f_flag;
1518 	linux_set_current(td);
1519 	linux_poll_wait_dequeue(filp);
1520 	linux_get_fop(filp, &fop, &ldev);
1521 	/*
1522 	 * Always use the real release function, if any, to avoid
1523 	 * leaking device resources:
1524 	 */
1525 	release = filp->f_op->release;
1526 	if (release != NULL)
1527 		error = -OPW(file, td, release(filp->f_vnode, filp));
1528 	funsetown(&filp->f_sigio);
1529 	if (filp->f_vnode != NULL)
1530 		vdrop(filp->f_vnode);
1531 	linux_drop_fop(ldev);
1532 	ldev = filp->f_cdev;
1533 	if (ldev != NULL)
1534 		linux_cdev_deref(ldev);
1535 	kfree(filp);
1536 
1537 	return (error);
1538 }
1539 
1540 static int
1541 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1542     struct thread *td)
1543 {
1544 	struct linux_file *filp;
1545 	const struct file_operations *fop;
1546 	struct linux_cdev *ldev;
1547 	struct fiodgname_arg *fgn;
1548 	const char *p;
1549 	int error, i;
1550 
1551 	error = 0;
1552 	filp = (struct linux_file *)fp->f_data;
1553 	filp->f_flags = fp->f_flag;
1554 	linux_get_fop(filp, &fop, &ldev);
1555 
1556 	linux_set_current(td);
1557 	switch (cmd) {
1558 	case FIONBIO:
1559 		break;
1560 	case FIOASYNC:
1561 		if (fop->fasync == NULL)
1562 			break;
1563 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1564 		break;
1565 	case FIOSETOWN:
1566 		error = fsetown(*(int *)data, &filp->f_sigio);
1567 		if (error == 0) {
1568 			if (fop->fasync == NULL)
1569 				break;
1570 			error = -OPW(fp, td, fop->fasync(0, filp,
1571 			    fp->f_flag & FASYNC));
1572 		}
1573 		break;
1574 	case FIOGETOWN:
1575 		*(int *)data = fgetown(&filp->f_sigio);
1576 		break;
1577 	case FIODGNAME:
1578 #ifdef	COMPAT_FREEBSD32
1579 	case FIODGNAME_32:
1580 #endif
1581 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1582 			error = ENXIO;
1583 			break;
1584 		}
1585 		fgn = data;
1586 		p = devtoname(filp->f_cdev->cdev);
1587 		i = strlen(p) + 1;
1588 		if (i > fgn->len) {
1589 			error = EINVAL;
1590 			break;
1591 		}
1592 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1593 		break;
1594 	default:
1595 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1596 		break;
1597 	}
1598 	linux_drop_fop(ldev);
1599 	return (error);
1600 }
1601 
1602 static int
1603 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1604     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1605     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1606 {
1607 	/*
1608 	 * Character devices do not provide private mappings
1609 	 * of any kind:
1610 	 */
1611 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1612 	    (prot & VM_PROT_WRITE) != 0)
1613 		return (EACCES);
1614 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1615 		return (EINVAL);
1616 
1617 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1618 	    (int)prot, td));
1619 }
1620 
1621 static int
1622 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1623     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1624     struct thread *td)
1625 {
1626 	struct linux_file *filp;
1627 	const struct file_operations *fop;
1628 	struct linux_cdev *ldev;
1629 	struct mount *mp;
1630 	struct vnode *vp;
1631 	vm_object_t object;
1632 	vm_prot_t maxprot;
1633 	int error;
1634 
1635 	filp = (struct linux_file *)fp->f_data;
1636 
1637 	vp = filp->f_vnode;
1638 	if (vp == NULL)
1639 		return (EOPNOTSUPP);
1640 
1641 	/*
1642 	 * Ensure that file and memory protections are
1643 	 * compatible.
1644 	 */
1645 	mp = vp->v_mount;
1646 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1647 		maxprot = VM_PROT_NONE;
1648 		if ((prot & VM_PROT_EXECUTE) != 0)
1649 			return (EACCES);
1650 	} else
1651 		maxprot = VM_PROT_EXECUTE;
1652 	if ((fp->f_flag & FREAD) != 0)
1653 		maxprot |= VM_PROT_READ;
1654 	else if ((prot & VM_PROT_READ) != 0)
1655 		return (EACCES);
1656 
1657 	/*
1658 	 * If we are sharing potential changes via MAP_SHARED and we
1659 	 * are trying to get write permission although we opened it
1660 	 * without asking for it, bail out.
1661 	 *
1662 	 * Note that most character devices always share mappings.
1663 	 *
1664 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1665 	 * requests rather than doing it here.
1666 	 */
1667 	if ((flags & MAP_SHARED) != 0) {
1668 		if ((fp->f_flag & FWRITE) != 0)
1669 			maxprot |= VM_PROT_WRITE;
1670 		else if ((prot & VM_PROT_WRITE) != 0)
1671 			return (EACCES);
1672 	}
1673 	maxprot &= cap_maxprot;
1674 
1675 	linux_get_fop(filp, &fop, &ldev);
1676 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp,
1677 	    &foff, fop, &object);
1678 	if (error != 0)
1679 		goto out;
1680 
1681 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1682 	    foff, FALSE, td);
1683 	if (error != 0)
1684 		vm_object_deallocate(object);
1685 out:
1686 	linux_drop_fop(ldev);
1687 	return (error);
1688 }
1689 
1690 static int
1691 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1692     struct thread *td)
1693 {
1694 	struct linux_file *filp;
1695 	struct vnode *vp;
1696 	int error;
1697 
1698 	filp = (struct linux_file *)fp->f_data;
1699 	if (filp->f_vnode == NULL)
1700 		return (EOPNOTSUPP);
1701 
1702 	vp = filp->f_vnode;
1703 
1704 	vn_lock(vp, LK_SHARED | LK_RETRY);
1705 	error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td);
1706 	VOP_UNLOCK(vp);
1707 
1708 	return (error);
1709 }
1710 
1711 static int
1712 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1713     struct filedesc *fdp)
1714 {
1715 	struct linux_file *filp;
1716 	struct vnode *vp;
1717 	int error;
1718 
1719 	filp = fp->f_data;
1720 	vp = filp->f_vnode;
1721 	if (vp == NULL) {
1722 		error = 0;
1723 		kif->kf_type = KF_TYPE_DEV;
1724 	} else {
1725 		vref(vp);
1726 		FILEDESC_SUNLOCK(fdp);
1727 		error = vn_fill_kinfo_vnode(vp, kif);
1728 		vrele(vp);
1729 		kif->kf_type = KF_TYPE_VNODE;
1730 		FILEDESC_SLOCK(fdp);
1731 	}
1732 	return (error);
1733 }
1734 
1735 unsigned int
1736 linux_iminor(struct inode *inode)
1737 {
1738 	struct linux_cdev *ldev;
1739 
1740 	if (inode == NULL || inode->v_rdev == NULL ||
1741 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1742 		return (-1U);
1743 	ldev = inode->v_rdev->si_drv1;
1744 	if (ldev == NULL)
1745 		return (-1U);
1746 
1747 	return (minor(ldev->dev));
1748 }
1749 
1750 struct fileops linuxfileops = {
1751 	.fo_read = linux_file_read,
1752 	.fo_write = linux_file_write,
1753 	.fo_truncate = invfo_truncate,
1754 	.fo_kqfilter = linux_file_kqfilter,
1755 	.fo_stat = linux_file_stat,
1756 	.fo_fill_kinfo = linux_file_fill_kinfo,
1757 	.fo_poll = linux_file_poll,
1758 	.fo_close = linux_file_close,
1759 	.fo_ioctl = linux_file_ioctl,
1760 	.fo_mmap = linux_file_mmap,
1761 	.fo_chmod = invfo_chmod,
1762 	.fo_chown = invfo_chown,
1763 	.fo_sendfile = invfo_sendfile,
1764 	.fo_flags = DFLAG_PASSABLE,
1765 };
1766 
1767 /*
1768  * Hash of vmmap addresses.  This is infrequently accessed and does not
1769  * need to be particularly large.  This is done because we must store the
1770  * caller's idea of the map size to properly unmap.
1771  */
1772 struct vmmap {
1773 	LIST_ENTRY(vmmap)	vm_next;
1774 	void 			*vm_addr;
1775 	unsigned long		vm_size;
1776 };
1777 
1778 struct vmmaphd {
1779 	struct vmmap *lh_first;
1780 };
1781 #define	VMMAP_HASH_SIZE	64
1782 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1783 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1784 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1785 static struct mtx vmmaplock;
1786 
1787 static void
1788 vmmap_add(void *addr, unsigned long size)
1789 {
1790 	struct vmmap *vmmap;
1791 
1792 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1793 	mtx_lock(&vmmaplock);
1794 	vmmap->vm_size = size;
1795 	vmmap->vm_addr = addr;
1796 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1797 	mtx_unlock(&vmmaplock);
1798 }
1799 
1800 static struct vmmap *
1801 vmmap_remove(void *addr)
1802 {
1803 	struct vmmap *vmmap;
1804 
1805 	mtx_lock(&vmmaplock);
1806 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1807 		if (vmmap->vm_addr == addr)
1808 			break;
1809 	if (vmmap)
1810 		LIST_REMOVE(vmmap, vm_next);
1811 	mtx_unlock(&vmmaplock);
1812 
1813 	return (vmmap);
1814 }
1815 
1816 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1817 void *
1818 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1819 {
1820 	void *addr;
1821 
1822 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1823 	if (addr == NULL)
1824 		return (NULL);
1825 	vmmap_add(addr, size);
1826 
1827 	return (addr);
1828 }
1829 #endif
1830 
1831 void
1832 iounmap(void *addr)
1833 {
1834 	struct vmmap *vmmap;
1835 
1836 	vmmap = vmmap_remove(addr);
1837 	if (vmmap == NULL)
1838 		return;
1839 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1840 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1841 #endif
1842 	kfree(vmmap);
1843 }
1844 
1845 void *
1846 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1847 {
1848 	vm_offset_t off;
1849 	size_t size;
1850 
1851 	size = count * PAGE_SIZE;
1852 	off = kva_alloc(size);
1853 	if (off == 0)
1854 		return (NULL);
1855 	vmmap_add((void *)off, size);
1856 	pmap_qenter(off, pages, count);
1857 
1858 	return ((void *)off);
1859 }
1860 
1861 void
1862 vunmap(void *addr)
1863 {
1864 	struct vmmap *vmmap;
1865 
1866 	vmmap = vmmap_remove(addr);
1867 	if (vmmap == NULL)
1868 		return;
1869 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1870 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1871 	kfree(vmmap);
1872 }
1873 
1874 static char *
1875 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1876 {
1877 	unsigned int len;
1878 	char *p;
1879 	va_list aq;
1880 
1881 	va_copy(aq, ap);
1882 	len = vsnprintf(NULL, 0, fmt, aq);
1883 	va_end(aq);
1884 
1885 	if (dev != NULL)
1886 		p = devm_kmalloc(dev, len + 1, gfp);
1887 	else
1888 		p = kmalloc(len + 1, gfp);
1889 	if (p != NULL)
1890 		vsnprintf(p, len + 1, fmt, ap);
1891 
1892 	return (p);
1893 }
1894 
1895 char *
1896 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1897 {
1898 
1899 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1900 }
1901 
1902 char *
1903 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1904 {
1905 	va_list ap;
1906 	char *p;
1907 
1908 	va_start(ap, fmt);
1909 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1910 	va_end(ap);
1911 
1912 	return (p);
1913 }
1914 
1915 char *
1916 kasprintf(gfp_t gfp, const char *fmt, ...)
1917 {
1918 	va_list ap;
1919 	char *p;
1920 
1921 	va_start(ap, fmt);
1922 	p = kvasprintf(gfp, fmt, ap);
1923 	va_end(ap);
1924 
1925 	return (p);
1926 }
1927 
1928 static void
1929 linux_timer_callback_wrapper(void *context)
1930 {
1931 	struct timer_list *timer;
1932 
1933 	timer = context;
1934 
1935 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
1936 		/* try again later */
1937 		callout_reset(&timer->callout, 1,
1938 		    &linux_timer_callback_wrapper, timer);
1939 		return;
1940 	}
1941 
1942 	timer->function(timer->data);
1943 }
1944 
1945 int
1946 mod_timer(struct timer_list *timer, int expires)
1947 {
1948 	int ret;
1949 
1950 	timer->expires = expires;
1951 	ret = callout_reset(&timer->callout,
1952 	    linux_timer_jiffies_until(expires),
1953 	    &linux_timer_callback_wrapper, timer);
1954 
1955 	MPASS(ret == 0 || ret == 1);
1956 
1957 	return (ret == 1);
1958 }
1959 
1960 void
1961 add_timer(struct timer_list *timer)
1962 {
1963 
1964 	callout_reset(&timer->callout,
1965 	    linux_timer_jiffies_until(timer->expires),
1966 	    &linux_timer_callback_wrapper, timer);
1967 }
1968 
1969 void
1970 add_timer_on(struct timer_list *timer, int cpu)
1971 {
1972 
1973 	callout_reset_on(&timer->callout,
1974 	    linux_timer_jiffies_until(timer->expires),
1975 	    &linux_timer_callback_wrapper, timer, cpu);
1976 }
1977 
1978 int
1979 del_timer(struct timer_list *timer)
1980 {
1981 
1982 	if (callout_stop(&(timer)->callout) == -1)
1983 		return (0);
1984 	return (1);
1985 }
1986 
1987 int
1988 del_timer_sync(struct timer_list *timer)
1989 {
1990 
1991 	if (callout_drain(&(timer)->callout) == -1)
1992 		return (0);
1993 	return (1);
1994 }
1995 
1996 /* greatest common divisor, Euclid equation */
1997 static uint64_t
1998 lkpi_gcd_64(uint64_t a, uint64_t b)
1999 {
2000 	uint64_t an;
2001 	uint64_t bn;
2002 
2003 	while (b != 0) {
2004 		an = b;
2005 		bn = a % b;
2006 		a = an;
2007 		b = bn;
2008 	}
2009 	return (a);
2010 }
2011 
2012 uint64_t lkpi_nsec2hz_rem;
2013 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2014 uint64_t lkpi_nsec2hz_max;
2015 
2016 uint64_t lkpi_usec2hz_rem;
2017 uint64_t lkpi_usec2hz_div = 1000000ULL;
2018 uint64_t lkpi_usec2hz_max;
2019 
2020 uint64_t lkpi_msec2hz_rem;
2021 uint64_t lkpi_msec2hz_div = 1000ULL;
2022 uint64_t lkpi_msec2hz_max;
2023 
2024 static void
2025 linux_timer_init(void *arg)
2026 {
2027 	uint64_t gcd;
2028 
2029 	/*
2030 	 * Compute an internal HZ value which can divide 2**32 to
2031 	 * avoid timer rounding problems when the tick value wraps
2032 	 * around 2**32:
2033 	 */
2034 	linux_timer_hz_mask = 1;
2035 	while (linux_timer_hz_mask < (unsigned long)hz)
2036 		linux_timer_hz_mask *= 2;
2037 	linux_timer_hz_mask--;
2038 
2039 	/* compute some internal constants */
2040 
2041 	lkpi_nsec2hz_rem = hz;
2042 	lkpi_usec2hz_rem = hz;
2043 	lkpi_msec2hz_rem = hz;
2044 
2045 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2046 	lkpi_nsec2hz_rem /= gcd;
2047 	lkpi_nsec2hz_div /= gcd;
2048 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2049 
2050 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2051 	lkpi_usec2hz_rem /= gcd;
2052 	lkpi_usec2hz_div /= gcd;
2053 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2054 
2055 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2056 	lkpi_msec2hz_rem /= gcd;
2057 	lkpi_msec2hz_div /= gcd;
2058 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2059 }
2060 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2061 
2062 void
2063 linux_complete_common(struct completion *c, int all)
2064 {
2065 	int wakeup_swapper;
2066 
2067 	sleepq_lock(c);
2068 	if (all) {
2069 		c->done = UINT_MAX;
2070 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2071 	} else {
2072 		if (c->done != UINT_MAX)
2073 			c->done++;
2074 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2075 	}
2076 	sleepq_release(c);
2077 	if (wakeup_swapper)
2078 		kick_proc0();
2079 }
2080 
2081 /*
2082  * Indefinite wait for done != 0 with or without signals.
2083  */
2084 int
2085 linux_wait_for_common(struct completion *c, int flags)
2086 {
2087 	struct task_struct *task;
2088 	int error;
2089 
2090 	if (SCHEDULER_STOPPED())
2091 		return (0);
2092 
2093 	task = current;
2094 
2095 	if (flags != 0)
2096 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2097 	else
2098 		flags = SLEEPQ_SLEEP;
2099 	error = 0;
2100 	for (;;) {
2101 		sleepq_lock(c);
2102 		if (c->done)
2103 			break;
2104 		sleepq_add(c, NULL, "completion", flags, 0);
2105 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2106 			DROP_GIANT();
2107 			error = -sleepq_wait_sig(c, 0);
2108 			PICKUP_GIANT();
2109 			if (error != 0) {
2110 				linux_schedule_save_interrupt_value(task, error);
2111 				error = -ERESTARTSYS;
2112 				goto intr;
2113 			}
2114 		} else {
2115 			DROP_GIANT();
2116 			sleepq_wait(c, 0);
2117 			PICKUP_GIANT();
2118 		}
2119 	}
2120 	if (c->done != UINT_MAX)
2121 		c->done--;
2122 	sleepq_release(c);
2123 
2124 intr:
2125 	return (error);
2126 }
2127 
2128 /*
2129  * Time limited wait for done != 0 with or without signals.
2130  */
2131 int
2132 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2133 {
2134 	struct task_struct *task;
2135 	int end = jiffies + timeout;
2136 	int error;
2137 
2138 	if (SCHEDULER_STOPPED())
2139 		return (0);
2140 
2141 	task = current;
2142 
2143 	if (flags != 0)
2144 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2145 	else
2146 		flags = SLEEPQ_SLEEP;
2147 
2148 	for (;;) {
2149 		sleepq_lock(c);
2150 		if (c->done)
2151 			break;
2152 		sleepq_add(c, NULL, "completion", flags, 0);
2153 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2154 
2155 		DROP_GIANT();
2156 		if (flags & SLEEPQ_INTERRUPTIBLE)
2157 			error = -sleepq_timedwait_sig(c, 0);
2158 		else
2159 			error = -sleepq_timedwait(c, 0);
2160 		PICKUP_GIANT();
2161 
2162 		if (error != 0) {
2163 			/* check for timeout */
2164 			if (error == -EWOULDBLOCK) {
2165 				error = 0;	/* timeout */
2166 			} else {
2167 				/* signal happened */
2168 				linux_schedule_save_interrupt_value(task, error);
2169 				error = -ERESTARTSYS;
2170 			}
2171 			goto done;
2172 		}
2173 	}
2174 	if (c->done != UINT_MAX)
2175 		c->done--;
2176 	sleepq_release(c);
2177 
2178 	/* return how many jiffies are left */
2179 	error = linux_timer_jiffies_until(end);
2180 done:
2181 	return (error);
2182 }
2183 
2184 int
2185 linux_try_wait_for_completion(struct completion *c)
2186 {
2187 	int isdone;
2188 
2189 	sleepq_lock(c);
2190 	isdone = (c->done != 0);
2191 	if (c->done != 0 && c->done != UINT_MAX)
2192 		c->done--;
2193 	sleepq_release(c);
2194 	return (isdone);
2195 }
2196 
2197 int
2198 linux_completion_done(struct completion *c)
2199 {
2200 	int isdone;
2201 
2202 	sleepq_lock(c);
2203 	isdone = (c->done != 0);
2204 	sleepq_release(c);
2205 	return (isdone);
2206 }
2207 
2208 static void
2209 linux_cdev_deref(struct linux_cdev *ldev)
2210 {
2211 	if (refcount_release(&ldev->refs) &&
2212 	    ldev->kobj.ktype == &linux_cdev_ktype)
2213 		kfree(ldev);
2214 }
2215 
2216 static void
2217 linux_cdev_release(struct kobject *kobj)
2218 {
2219 	struct linux_cdev *cdev;
2220 	struct kobject *parent;
2221 
2222 	cdev = container_of(kobj, struct linux_cdev, kobj);
2223 	parent = kobj->parent;
2224 	linux_destroy_dev(cdev);
2225 	linux_cdev_deref(cdev);
2226 	kobject_put(parent);
2227 }
2228 
2229 static void
2230 linux_cdev_static_release(struct kobject *kobj)
2231 {
2232 	struct cdev *cdev;
2233 	struct linux_cdev *ldev;
2234 
2235 	ldev = container_of(kobj, struct linux_cdev, kobj);
2236 	cdev = ldev->cdev;
2237 	if (cdev != NULL) {
2238 		destroy_dev(cdev);
2239 		ldev->cdev = NULL;
2240 	}
2241 	kobject_put(kobj->parent);
2242 }
2243 
2244 void
2245 linux_destroy_dev(struct linux_cdev *ldev)
2246 {
2247 
2248 	if (ldev->cdev == NULL)
2249 		return;
2250 
2251 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2252 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2253 
2254 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2255 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2256 		pause("ldevdtr", hz / 4);
2257 
2258 	destroy_dev(ldev->cdev);
2259 	ldev->cdev = NULL;
2260 }
2261 
2262 const struct kobj_type linux_cdev_ktype = {
2263 	.release = linux_cdev_release,
2264 };
2265 
2266 const struct kobj_type linux_cdev_static_ktype = {
2267 	.release = linux_cdev_static_release,
2268 };
2269 
2270 static void
2271 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2272 {
2273 	struct notifier_block *nb;
2274 	struct netdev_notifier_info ni;
2275 
2276 	nb = arg;
2277 	ni.dev = (struct net_device *)ifp;
2278 	if (linkstate == LINK_STATE_UP)
2279 		nb->notifier_call(nb, NETDEV_UP, &ni);
2280 	else
2281 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2282 }
2283 
2284 static void
2285 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2286 {
2287 	struct notifier_block *nb;
2288 	struct netdev_notifier_info ni;
2289 
2290 	nb = arg;
2291 	ni.dev = (struct net_device *)ifp;
2292 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2293 }
2294 
2295 static void
2296 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2297 {
2298 	struct notifier_block *nb;
2299 	struct netdev_notifier_info ni;
2300 
2301 	nb = arg;
2302 	ni.dev = (struct net_device *)ifp;
2303 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2304 }
2305 
2306 static void
2307 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2308 {
2309 	struct notifier_block *nb;
2310 	struct netdev_notifier_info ni;
2311 
2312 	nb = arg;
2313 	ni.dev = (struct net_device *)ifp;
2314 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2315 }
2316 
2317 static void
2318 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2319 {
2320 	struct notifier_block *nb;
2321 	struct netdev_notifier_info ni;
2322 
2323 	nb = arg;
2324 	ni.dev = (struct net_device *)ifp;
2325 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2326 }
2327 
2328 int
2329 register_netdevice_notifier(struct notifier_block *nb)
2330 {
2331 
2332 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2333 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2334 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2335 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2336 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2337 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2338 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2339 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2340 
2341 	return (0);
2342 }
2343 
2344 int
2345 register_inetaddr_notifier(struct notifier_block *nb)
2346 {
2347 
2348 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2349 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2350 	return (0);
2351 }
2352 
2353 int
2354 unregister_netdevice_notifier(struct notifier_block *nb)
2355 {
2356 
2357 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2358 	    nb->tags[NETDEV_UP]);
2359 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2360 	    nb->tags[NETDEV_REGISTER]);
2361 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2362 	    nb->tags[NETDEV_UNREGISTER]);
2363 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2364 	    nb->tags[NETDEV_CHANGEADDR]);
2365 
2366 	return (0);
2367 }
2368 
2369 int
2370 unregister_inetaddr_notifier(struct notifier_block *nb)
2371 {
2372 
2373 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2374 	    nb->tags[NETDEV_CHANGEIFADDR]);
2375 
2376 	return (0);
2377 }
2378 
2379 struct list_sort_thunk {
2380 	int (*cmp)(void *, struct list_head *, struct list_head *);
2381 	void *priv;
2382 };
2383 
2384 static inline int
2385 linux_le_cmp(void *priv, const void *d1, const void *d2)
2386 {
2387 	struct list_head *le1, *le2;
2388 	struct list_sort_thunk *thunk;
2389 
2390 	thunk = priv;
2391 	le1 = *(__DECONST(struct list_head **, d1));
2392 	le2 = *(__DECONST(struct list_head **, d2));
2393 	return ((thunk->cmp)(thunk->priv, le1, le2));
2394 }
2395 
2396 void
2397 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2398     struct list_head *a, struct list_head *b))
2399 {
2400 	struct list_sort_thunk thunk;
2401 	struct list_head **ar, *le;
2402 	size_t count, i;
2403 
2404 	count = 0;
2405 	list_for_each(le, head)
2406 		count++;
2407 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2408 	i = 0;
2409 	list_for_each(le, head)
2410 		ar[i++] = le;
2411 	thunk.cmp = cmp;
2412 	thunk.priv = priv;
2413 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2414 	INIT_LIST_HEAD(head);
2415 	for (i = 0; i < count; i++)
2416 		list_add_tail(ar[i], head);
2417 	free(ar, M_KMALLOC);
2418 }
2419 
2420 void
2421 linux_irq_handler(void *ent)
2422 {
2423 	struct irq_ent *irqe;
2424 
2425 	if (linux_set_current_flags(curthread, M_NOWAIT))
2426 		return;
2427 
2428 	irqe = ent;
2429 	irqe->handler(irqe->irq, irqe->arg);
2430 }
2431 
2432 #if defined(__i386__) || defined(__amd64__)
2433 int
2434 linux_wbinvd_on_all_cpus(void)
2435 {
2436 
2437 	pmap_invalidate_cache();
2438 	return (0);
2439 }
2440 #endif
2441 
2442 int
2443 linux_on_each_cpu(void callback(void *), void *data)
2444 {
2445 
2446 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2447 	    smp_no_rendezvous_barrier, data);
2448 	return (0);
2449 }
2450 
2451 int
2452 linux_in_atomic(void)
2453 {
2454 
2455 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2456 }
2457 
2458 struct linux_cdev *
2459 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2460 {
2461 	dev_t dev = MKDEV(major, minor);
2462 	struct cdev *cdev;
2463 
2464 	dev_lock();
2465 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2466 		struct linux_cdev *ldev = cdev->si_drv1;
2467 		if (ldev->dev == dev &&
2468 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2469 			break;
2470 		}
2471 	}
2472 	dev_unlock();
2473 
2474 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2475 }
2476 
2477 int
2478 __register_chrdev(unsigned int major, unsigned int baseminor,
2479     unsigned int count, const char *name,
2480     const struct file_operations *fops)
2481 {
2482 	struct linux_cdev *cdev;
2483 	int ret = 0;
2484 	int i;
2485 
2486 	for (i = baseminor; i < baseminor + count; i++) {
2487 		cdev = cdev_alloc();
2488 		cdev->ops = fops;
2489 		kobject_set_name(&cdev->kobj, name);
2490 
2491 		ret = cdev_add(cdev, makedev(major, i), 1);
2492 		if (ret != 0)
2493 			break;
2494 	}
2495 	return (ret);
2496 }
2497 
2498 int
2499 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2500     unsigned int count, const char *name,
2501     const struct file_operations *fops, uid_t uid,
2502     gid_t gid, int mode)
2503 {
2504 	struct linux_cdev *cdev;
2505 	int ret = 0;
2506 	int i;
2507 
2508 	for (i = baseminor; i < baseminor + count; i++) {
2509 		cdev = cdev_alloc();
2510 		cdev->ops = fops;
2511 		kobject_set_name(&cdev->kobj, name);
2512 
2513 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2514 		if (ret != 0)
2515 			break;
2516 	}
2517 	return (ret);
2518 }
2519 
2520 void
2521 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2522     unsigned int count, const char *name)
2523 {
2524 	struct linux_cdev *cdevp;
2525 	int i;
2526 
2527 	for (i = baseminor; i < baseminor + count; i++) {
2528 		cdevp = linux_find_cdev(name, major, i);
2529 		if (cdevp != NULL)
2530 			cdev_del(cdevp);
2531 	}
2532 }
2533 
2534 void
2535 linux_dump_stack(void)
2536 {
2537 #ifdef STACK
2538 	struct stack st;
2539 
2540 	stack_zero(&st);
2541 	stack_save(&st);
2542 	stack_print(&st);
2543 #endif
2544 }
2545 
2546 int
2547 linuxkpi_net_ratelimit(void)
2548 {
2549 
2550 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2551 	   lkpi_net_maxpps));
2552 }
2553 
2554 #if defined(__i386__) || defined(__amd64__)
2555 bool linux_cpu_has_clflush;
2556 #endif
2557 
2558 static void
2559 linux_compat_init(void *arg)
2560 {
2561 	struct sysctl_oid *rootoid;
2562 	int i;
2563 
2564 #if defined(__i386__) || defined(__amd64__)
2565 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2566 #endif
2567 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2568 
2569 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2570 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2571 	kobject_init(&linux_class_root, &linux_class_ktype);
2572 	kobject_set_name(&linux_class_root, "class");
2573 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2574 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2575 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2576 	kobject_set_name(&linux_root_device.kobj, "device");
2577 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2578 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2579 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2580 	linux_root_device.bsddev = root_bus;
2581 	linux_class_misc.name = "misc";
2582 	class_register(&linux_class_misc);
2583 	INIT_LIST_HEAD(&pci_drivers);
2584 	INIT_LIST_HEAD(&pci_devices);
2585 	spin_lock_init(&pci_lock);
2586 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2587 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2588 		LIST_INIT(&vmmaphead[i]);
2589 	init_waitqueue_head(&linux_bit_waitq);
2590 	init_waitqueue_head(&linux_var_waitq);
2591 }
2592 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2593 
2594 static void
2595 linux_compat_uninit(void *arg)
2596 {
2597 	linux_kobject_kfree_name(&linux_class_root);
2598 	linux_kobject_kfree_name(&linux_root_device.kobj);
2599 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2600 
2601 	mtx_destroy(&vmmaplock);
2602 	spin_lock_destroy(&pci_lock);
2603 	rw_destroy(&linux_vma_lock);
2604 }
2605 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2606 
2607 /*
2608  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2609  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2610  * used. Assert these types have the same size, else some parts of the
2611  * LinuxKPI may not work like expected:
2612  */
2613 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2614