xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision a3cefe7f2b4df0f70ff92d4570ce18e517af43ec)
1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2021 Mellanox Technologies, Ltd.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice unmodified, this list of conditions, and the following
13  *    disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 #include "opt_global.h"
32 #include "opt_stack.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
39 #include <sys/proc.h>
40 #include <sys/sglist.h>
41 #include <sys/sleepqueue.h>
42 #include <sys/refcount.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/bus.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filio.h>
50 #include <sys/rwlock.h>
51 #include <sys/mman.h>
52 #include <sys/stack.h>
53 #include <sys/stdarg.h>
54 #include <sys/sysent.h>
55 #include <sys/time.h>
56 #include <sys/user.h>
57 
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pager.h>
63 #include <vm/vm_radix.h>
64 
65 #if defined(__i386__) || defined(__amd64__)
66 #include <machine/cputypes.h>
67 #include <machine/md_var.h>
68 #endif
69 
70 #include <linux/kobject.h>
71 #include <linux/cpu.h>
72 #include <linux/device.h>
73 #include <linux/slab.h>
74 #include <linux/module.h>
75 #include <linux/moduleparam.h>
76 #include <linux/cdev.h>
77 #include <linux/file.h>
78 #include <linux/fs.h>
79 #include <linux/sysfs.h>
80 #include <linux/mm.h>
81 #include <linux/io.h>
82 #include <linux/vmalloc.h>
83 #include <linux/netdevice.h>
84 #include <linux/timer.h>
85 #include <linux/interrupt.h>
86 #include <linux/uaccess.h>
87 #include <linux/utsname.h>
88 #include <linux/list.h>
89 #include <linux/kthread.h>
90 #include <linux/kernel.h>
91 #include <linux/compat.h>
92 #include <linux/io-mapping.h>
93 #include <linux/poll.h>
94 #include <linux/smp.h>
95 #include <linux/wait_bit.h>
96 #include <linux/rcupdate.h>
97 #include <linux/interval_tree.h>
98 #include <linux/interval_tree_generic.h>
99 #include <linux/printk.h>
100 #include <linux/seq_file.h>
101 #include <linux/uuid.h>
102 
103 #if defined(__i386__) || defined(__amd64__)
104 #include <asm/smp.h>
105 #include <asm/processor.h>
106 #endif
107 
108 #include <xen/xen.h>
109 #ifdef XENHVM
110 #undef xen_pv_domain
111 #undef xen_initial_domain
112 /* xen/xen-os.h redefines __must_check */
113 #undef __must_check
114 #include <xen/xen-os.h>
115 #endif
116 
117 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
118     "LinuxKPI parameters");
119 
120 int linuxkpi_debug;
121 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
122     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
123 
124 int linuxkpi_rcu_debug;
125 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, rcu_debug, CTLFLAG_RWTUN,
126     &linuxkpi_rcu_debug, 0, "Set to enable RCU warning. Clear to disable.");
127 
128 int linuxkpi_warn_dump_stack = 0;
129 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN,
130     &linuxkpi_warn_dump_stack, 0,
131     "Set to enable stack traces from WARN_ON(). Clear to disable.");
132 
133 static struct timeval lkpi_net_lastlog;
134 static int lkpi_net_curpps;
135 static int lkpi_net_maxpps = 99;
136 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
137     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
138 
139 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat");
140 
141 #include <linux/rbtree.h>
142 /* Undo Linux compat changes. */
143 #undef RB_ROOT
144 #undef file
145 #undef cdev
146 #define	RB_ROOT(head)	(head)->rbh_root
147 
148 static void linux_destroy_dev(struct linux_cdev *);
149 static void linux_cdev_deref(struct linux_cdev *ldev);
150 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
151 
152 cpumask_t cpu_online_mask;
153 static cpumask_t **static_single_cpu_mask;
154 static cpumask_t *static_single_cpu_mask_lcs;
155 struct kobject linux_class_root;
156 struct device linux_root_device;
157 struct class linux_class_misc;
158 struct list_head pci_drivers;
159 struct list_head pci_devices;
160 spinlock_t pci_lock;
161 struct uts_namespace init_uts_ns;
162 
163 unsigned long linux_timer_hz_mask;
164 
165 wait_queue_head_t linux_bit_waitq;
166 wait_queue_head_t linux_var_waitq;
167 
168 const guid_t guid_null;
169 
170 int
171 panic_cmp(struct rb_node *one, struct rb_node *two)
172 {
173 	panic("no cmp");
174 }
175 
176 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
177 
178 #define	START(node)	((node)->start)
179 #define	LAST(node)	((node)->last)
180 
181 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
182     LAST,, lkpi_interval_tree)
183 
184 static void
185 linux_device_release(struct device *dev)
186 {
187 	pr_debug("linux_device_release: %s\n", dev_name(dev));
188 	kfree(dev);
189 }
190 
191 static ssize_t
192 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
193 {
194 	struct class_attribute *dattr;
195 	ssize_t error;
196 
197 	dattr = container_of(attr, struct class_attribute, attr);
198 	error = -EIO;
199 	if (dattr->show)
200 		error = dattr->show(container_of(kobj, struct class, kobj),
201 		    dattr, buf);
202 	return (error);
203 }
204 
205 static ssize_t
206 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
207     size_t count)
208 {
209 	struct class_attribute *dattr;
210 	ssize_t error;
211 
212 	dattr = container_of(attr, struct class_attribute, attr);
213 	error = -EIO;
214 	if (dattr->store)
215 		error = dattr->store(container_of(kobj, struct class, kobj),
216 		    dattr, buf, count);
217 	return (error);
218 }
219 
220 static void
221 linux_class_release(struct kobject *kobj)
222 {
223 	struct class *class;
224 
225 	class = container_of(kobj, struct class, kobj);
226 	if (class->class_release)
227 		class->class_release(class);
228 }
229 
230 static const struct sysfs_ops linux_class_sysfs = {
231 	.show  = linux_class_show,
232 	.store = linux_class_store,
233 };
234 
235 const struct kobj_type linux_class_ktype = {
236 	.release = linux_class_release,
237 	.sysfs_ops = &linux_class_sysfs
238 };
239 
240 static void
241 linux_dev_release(struct kobject *kobj)
242 {
243 	struct device *dev;
244 
245 	dev = container_of(kobj, struct device, kobj);
246 	/* This is the precedence defined by linux. */
247 	if (dev->release)
248 		dev->release(dev);
249 	else if (dev->class && dev->class->dev_release)
250 		dev->class->dev_release(dev);
251 }
252 
253 static ssize_t
254 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
255 {
256 	struct device_attribute *dattr;
257 	ssize_t error;
258 
259 	dattr = container_of(attr, struct device_attribute, attr);
260 	error = -EIO;
261 	if (dattr->show)
262 		error = dattr->show(container_of(kobj, struct device, kobj),
263 		    dattr, buf);
264 	return (error);
265 }
266 
267 static ssize_t
268 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
269     size_t count)
270 {
271 	struct device_attribute *dattr;
272 	ssize_t error;
273 
274 	dattr = container_of(attr, struct device_attribute, attr);
275 	error = -EIO;
276 	if (dattr->store)
277 		error = dattr->store(container_of(kobj, struct device, kobj),
278 		    dattr, buf, count);
279 	return (error);
280 }
281 
282 static const struct sysfs_ops linux_dev_sysfs = {
283 	.show  = linux_dev_show,
284 	.store = linux_dev_store,
285 };
286 
287 const struct kobj_type linux_dev_ktype = {
288 	.release = linux_dev_release,
289 	.sysfs_ops = &linux_dev_sysfs
290 };
291 
292 struct device *
293 device_create(struct class *class, struct device *parent, dev_t devt,
294     void *drvdata, const char *fmt, ...)
295 {
296 	struct device *dev;
297 	va_list args;
298 
299 	dev = kzalloc(sizeof(*dev), M_WAITOK);
300 	dev->parent = parent;
301 	dev->class = class;
302 	dev->devt = devt;
303 	dev->driver_data = drvdata;
304 	dev->release = linux_device_release;
305 	va_start(args, fmt);
306 	kobject_set_name_vargs(&dev->kobj, fmt, args);
307 	va_end(args);
308 	device_register(dev);
309 
310 	return (dev);
311 }
312 
313 struct device *
314 device_create_groups_vargs(struct class *class, struct device *parent,
315     dev_t devt, void *drvdata, const struct attribute_group **groups,
316     const char *fmt, va_list args)
317 {
318 	struct device *dev = NULL;
319 	int retval = -ENODEV;
320 
321 	if (class == NULL || IS_ERR(class))
322 		goto error;
323 
324 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
325 	if (!dev) {
326 		retval = -ENOMEM;
327 		goto error;
328 	}
329 
330 	dev->devt = devt;
331 	dev->class = class;
332 	dev->parent = parent;
333 	dev->groups = groups;
334 	dev->release = device_create_release;
335 	/* device_initialize() needs the class and parent to be set */
336 	device_initialize(dev);
337 	dev_set_drvdata(dev, drvdata);
338 
339 	retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
340 	if (retval)
341 		goto error;
342 
343 	retval = device_add(dev);
344 	if (retval)
345 		goto error;
346 
347 	return dev;
348 
349 error:
350 	put_device(dev);
351 	return ERR_PTR(retval);
352 }
353 
354 struct class *
355 lkpi_class_create(const char *name)
356 {
357 	struct class *class;
358 	int error;
359 
360 	class = kzalloc(sizeof(*class), M_WAITOK);
361 	class->name = name;
362 	class->class_release = linux_class_kfree;
363 	error = class_register(class);
364 	if (error) {
365 		kfree(class);
366 		return (NULL);
367 	}
368 
369 	return (class);
370 }
371 
372 static void
373 linux_kq_lock(void *arg)
374 {
375 	spinlock_t *s = arg;
376 
377 	spin_lock(s);
378 }
379 static void
380 linux_kq_unlock(void *arg)
381 {
382 	spinlock_t *s = arg;
383 
384 	spin_unlock(s);
385 }
386 
387 static void
388 linux_kq_assert_lock(void *arg, int what)
389 {
390 #ifdef INVARIANTS
391 	spinlock_t *s = arg;
392 
393 	if (what == LA_LOCKED)
394 		mtx_assert(s, MA_OWNED);
395 	else
396 		mtx_assert(s, MA_NOTOWNED);
397 #endif
398 }
399 
400 static void
401 linux_file_kqfilter_poll(struct linux_file *, int);
402 
403 struct linux_file *
404 linux_file_alloc(void)
405 {
406 	struct linux_file *filp;
407 
408 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
409 
410 	/* set initial refcount */
411 	filp->f_count = 1;
412 
413 	/* setup fields needed by kqueue support */
414 	spin_lock_init(&filp->f_kqlock);
415 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
416 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
417 
418 	return (filp);
419 }
420 
421 void
422 linux_file_free(struct linux_file *filp)
423 {
424 	if (filp->_file == NULL) {
425 		if (filp->f_op != NULL && filp->f_op->release != NULL)
426 			filp->f_op->release(filp->f_vnode, filp);
427 		if (filp->f_shmem != NULL)
428 			vm_object_deallocate(filp->f_shmem);
429 		kfree_rcu(filp, rcu);
430 	} else {
431 		/*
432 		 * The close method of the character device or file
433 		 * will free the linux_file structure:
434 		 */
435 		_fdrop(filp->_file, curthread);
436 	}
437 }
438 
439 struct linux_cdev *
440 cdev_alloc(void)
441 {
442 	struct linux_cdev *cdev;
443 
444 	cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
445 	kobject_init(&cdev->kobj, &linux_cdev_ktype);
446 	cdev->refs = 1;
447 	return (cdev);
448 }
449 
450 static int
451 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
452     vm_page_t *mres)
453 {
454 	struct vm_area_struct *vmap;
455 
456 	vmap = linux_cdev_handle_find(vm_obj->handle);
457 
458 	MPASS(vmap != NULL);
459 	MPASS(vmap->vm_private_data == vm_obj->handle);
460 
461 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
462 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
463 		vm_page_t page;
464 
465 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
466 			/*
467 			 * If the passed in result page is a fake
468 			 * page, update it with the new physical
469 			 * address.
470 			 */
471 			page = *mres;
472 			vm_page_updatefake(page, paddr, vm_obj->memattr);
473 		} else {
474 			/*
475 			 * Replace the passed in "mres" page with our
476 			 * own fake page and free up the all of the
477 			 * original pages.
478 			 */
479 			VM_OBJECT_WUNLOCK(vm_obj);
480 			page = vm_page_getfake(paddr, vm_obj->memattr);
481 			VM_OBJECT_WLOCK(vm_obj);
482 
483 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
484 			*mres = page;
485 		}
486 		vm_page_valid(page);
487 		return (VM_PAGER_OK);
488 	}
489 	return (VM_PAGER_FAIL);
490 }
491 
492 static int
493 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
494     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
495 {
496 	struct vm_area_struct *vmap;
497 	int err;
498 
499 	/* get VM area structure */
500 	vmap = linux_cdev_handle_find(vm_obj->handle);
501 	MPASS(vmap != NULL);
502 	MPASS(vmap->vm_private_data == vm_obj->handle);
503 
504 	VM_OBJECT_WUNLOCK(vm_obj);
505 
506 	linux_set_current(curthread);
507 
508 	down_write(&vmap->vm_mm->mmap_sem);
509 	if (unlikely(vmap->vm_ops == NULL)) {
510 		err = VM_FAULT_SIGBUS;
511 	} else {
512 		struct vm_fault vmf;
513 
514 		/* fill out VM fault structure */
515 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
516 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
517 		vmf.pgoff = 0;
518 		vmf.page = NULL;
519 		vmf.vma = vmap;
520 
521 		vmap->vm_pfn_count = 0;
522 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
523 		vmap->vm_obj = vm_obj;
524 
525 		err = vmap->vm_ops->fault(&vmf);
526 
527 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
528 			kern_yield(PRI_USER);
529 			err = vmap->vm_ops->fault(&vmf);
530 		}
531 	}
532 
533 	/* translate return code */
534 	switch (err) {
535 	case VM_FAULT_OOM:
536 		err = VM_PAGER_AGAIN;
537 		break;
538 	case VM_FAULT_SIGBUS:
539 		err = VM_PAGER_BAD;
540 		break;
541 	case VM_FAULT_NOPAGE:
542 		/*
543 		 * By contract the fault handler will return having
544 		 * busied all the pages itself. If pidx is already
545 		 * found in the object, it will simply xbusy the first
546 		 * page and return with vm_pfn_count set to 1.
547 		 */
548 		*first = vmap->vm_pfn_first;
549 		*last = *first + vmap->vm_pfn_count - 1;
550 		err = VM_PAGER_OK;
551 		break;
552 	default:
553 		err = VM_PAGER_ERROR;
554 		break;
555 	}
556 	up_write(&vmap->vm_mm->mmap_sem);
557 	VM_OBJECT_WLOCK(vm_obj);
558 	return (err);
559 }
560 
561 static struct rwlock linux_vma_lock;
562 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
563     TAILQ_HEAD_INITIALIZER(linux_vma_head);
564 
565 static void
566 linux_cdev_handle_free(struct vm_area_struct *vmap)
567 {
568 	/* Drop reference on vm_file */
569 	if (vmap->vm_file != NULL)
570 		fput(vmap->vm_file);
571 
572 	/* Drop reference on mm_struct */
573 	mmput(vmap->vm_mm);
574 
575 	kfree(vmap);
576 }
577 
578 static void
579 linux_cdev_handle_remove(struct vm_area_struct *vmap)
580 {
581 	rw_wlock(&linux_vma_lock);
582 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
583 	rw_wunlock(&linux_vma_lock);
584 }
585 
586 static struct vm_area_struct *
587 linux_cdev_handle_find(void *handle)
588 {
589 	struct vm_area_struct *vmap;
590 
591 	rw_rlock(&linux_vma_lock);
592 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
593 		if (vmap->vm_private_data == handle)
594 			break;
595 	}
596 	rw_runlock(&linux_vma_lock);
597 	return (vmap);
598 }
599 
600 static int
601 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
602 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
603 {
604 
605 	MPASS(linux_cdev_handle_find(handle) != NULL);
606 	*color = 0;
607 	return (0);
608 }
609 
610 static void
611 linux_cdev_pager_dtor(void *handle)
612 {
613 	const struct vm_operations_struct *vm_ops;
614 	struct vm_area_struct *vmap;
615 
616 	vmap = linux_cdev_handle_find(handle);
617 	MPASS(vmap != NULL);
618 
619 	/*
620 	 * Remove handle before calling close operation to prevent
621 	 * other threads from reusing the handle pointer.
622 	 */
623 	linux_cdev_handle_remove(vmap);
624 
625 	down_write(&vmap->vm_mm->mmap_sem);
626 	vm_ops = vmap->vm_ops;
627 	if (likely(vm_ops != NULL))
628 		vm_ops->close(vmap);
629 	up_write(&vmap->vm_mm->mmap_sem);
630 
631 	linux_cdev_handle_free(vmap);
632 }
633 
634 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
635   {
636 	/* OBJT_MGTDEVICE */
637 	.cdev_pg_populate	= linux_cdev_pager_populate,
638 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
639 	.cdev_pg_dtor	= linux_cdev_pager_dtor
640   },
641   {
642 	/* OBJT_DEVICE */
643 	.cdev_pg_fault	= linux_cdev_pager_fault,
644 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
645 	.cdev_pg_dtor	= linux_cdev_pager_dtor
646   },
647 };
648 
649 int
650 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
651     unsigned long size)
652 {
653 	struct pctrie_iter pages;
654 	vm_object_t obj;
655 	vm_page_t m;
656 
657 	obj = vma->vm_obj;
658 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
659 		return (-ENOTSUP);
660 	VM_OBJECT_RLOCK(obj);
661 	vm_page_iter_limit_init(&pages, obj, OFF_TO_IDX(address + size));
662 	VM_RADIX_FOREACH_FROM(m, &pages, OFF_TO_IDX(address))
663 		pmap_remove_all(m);
664 	VM_OBJECT_RUNLOCK(obj);
665 	return (0);
666 }
667 
668 void
669 vma_set_file(struct vm_area_struct *vma, struct linux_file *file)
670 {
671 	struct linux_file *tmp;
672 
673 	/* Changing an anonymous vma with this is illegal */
674 	get_file(file);
675 	tmp = vma->vm_file;
676 	vma->vm_file = file;
677 	fput(tmp);
678 }
679 
680 static struct file_operations dummy_ldev_ops = {
681 	/* XXXKIB */
682 };
683 
684 static struct linux_cdev dummy_ldev = {
685 	.ops = &dummy_ldev_ops,
686 };
687 
688 #define	LDEV_SI_DTR	0x0001
689 #define	LDEV_SI_REF	0x0002
690 
691 static void
692 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
693     struct linux_cdev **dev)
694 {
695 	struct linux_cdev *ldev;
696 	u_int siref;
697 
698 	ldev = filp->f_cdev;
699 	*fop = filp->f_op;
700 	if (ldev != NULL) {
701 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
702 			refcount_acquire(&ldev->refs);
703 		} else {
704 			for (siref = ldev->siref;;) {
705 				if ((siref & LDEV_SI_DTR) != 0) {
706 					ldev = &dummy_ldev;
707 					*fop = ldev->ops;
708 					siref = ldev->siref;
709 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
710 				} else if (atomic_fcmpset_int(&ldev->siref,
711 				    &siref, siref + LDEV_SI_REF)) {
712 					break;
713 				}
714 			}
715 		}
716 	}
717 	*dev = ldev;
718 }
719 
720 static void
721 linux_drop_fop(struct linux_cdev *ldev)
722 {
723 
724 	if (ldev == NULL)
725 		return;
726 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
727 		linux_cdev_deref(ldev);
728 	} else {
729 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
730 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
731 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
732 	}
733 }
734 
735 #define	OPW(fp,td,code) ({			\
736 	struct file *__fpop;			\
737 	__typeof(code) __retval;		\
738 						\
739 	__fpop = (td)->td_fpop;			\
740 	(td)->td_fpop = (fp);			\
741 	__retval = (code);			\
742 	(td)->td_fpop = __fpop;			\
743 	__retval;				\
744 })
745 
746 static int
747 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
748     struct file *file)
749 {
750 	struct linux_cdev *ldev;
751 	struct linux_file *filp;
752 	const struct file_operations *fop;
753 	int error;
754 
755 	ldev = dev->si_drv1;
756 
757 	filp = linux_file_alloc();
758 	filp->f_dentry = &filp->f_dentry_store;
759 	filp->f_op = ldev->ops;
760 	filp->f_mode = file->f_flag;
761 	filp->f_flags = file->f_flag;
762 	filp->f_vnode = file->f_vnode;
763 	filp->_file = file;
764 	refcount_acquire(&ldev->refs);
765 	filp->f_cdev = ldev;
766 
767 	linux_set_current(td);
768 	linux_get_fop(filp, &fop, &ldev);
769 
770 	if (fop->open != NULL) {
771 		error = -fop->open(file->f_vnode, filp);
772 		if (error != 0) {
773 			linux_drop_fop(ldev);
774 			linux_cdev_deref(filp->f_cdev);
775 			kfree(filp);
776 			return (error);
777 		}
778 	}
779 
780 	/* hold on to the vnode - used for fstat() */
781 	vref(filp->f_vnode);
782 
783 	/* release the file from devfs */
784 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
785 	linux_drop_fop(ldev);
786 	return (ENXIO);
787 }
788 
789 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
790 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
791 
792 static inline int
793 linux_remap_address(void **uaddr, size_t len)
794 {
795 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
796 
797 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
798 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
799 		struct task_struct *pts = current;
800 		if (pts == NULL) {
801 			*uaddr = NULL;
802 			return (1);
803 		}
804 
805 		/* compute data offset */
806 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
807 
808 		/* check that length is within bounds */
809 		if ((len > IOCPARM_MAX) ||
810 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
811 			*uaddr = NULL;
812 			return (1);
813 		}
814 
815 		/* re-add kernel buffer address */
816 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
817 
818 		/* update address location */
819 		*uaddr = (void *)uaddr_val;
820 		return (1);
821 	}
822 	return (0);
823 }
824 
825 int
826 linux_copyin(const void *uaddr, void *kaddr, size_t len)
827 {
828 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
829 		if (uaddr == NULL)
830 			return (-EFAULT);
831 		memcpy(kaddr, uaddr, len);
832 		return (0);
833 	}
834 	return (-copyin(uaddr, kaddr, len));
835 }
836 
837 int
838 linux_copyout(const void *kaddr, void *uaddr, size_t len)
839 {
840 	if (linux_remap_address(&uaddr, len)) {
841 		if (uaddr == NULL)
842 			return (-EFAULT);
843 		memcpy(uaddr, kaddr, len);
844 		return (0);
845 	}
846 	return (-copyout(kaddr, uaddr, len));
847 }
848 
849 size_t
850 linux_clear_user(void *_uaddr, size_t _len)
851 {
852 	uint8_t *uaddr = _uaddr;
853 	size_t len = _len;
854 
855 	/* make sure uaddr is aligned before going into the fast loop */
856 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
857 		if (subyte(uaddr, 0))
858 			return (_len);
859 		uaddr++;
860 		len--;
861 	}
862 
863 	/* zero 8 bytes at a time */
864 	while (len > 7) {
865 #ifdef __LP64__
866 		if (suword64(uaddr, 0))
867 			return (_len);
868 #else
869 		if (suword32(uaddr, 0))
870 			return (_len);
871 		if (suword32(uaddr + 4, 0))
872 			return (_len);
873 #endif
874 		uaddr += 8;
875 		len -= 8;
876 	}
877 
878 	/* zero fill end, if any */
879 	while (len > 0) {
880 		if (subyte(uaddr, 0))
881 			return (_len);
882 		uaddr++;
883 		len--;
884 	}
885 	return (0);
886 }
887 
888 int
889 linux_access_ok(const void *uaddr, size_t len)
890 {
891 	uintptr_t saddr;
892 	uintptr_t eaddr;
893 
894 	/* get start and end address */
895 	saddr = (uintptr_t)uaddr;
896 	eaddr = (uintptr_t)uaddr + len;
897 
898 	/* verify addresses are valid for userspace */
899 	return ((saddr == eaddr) ||
900 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
901 }
902 
903 /*
904  * This function should return either EINTR or ERESTART depending on
905  * the signal type sent to this thread:
906  */
907 static int
908 linux_get_error(struct task_struct *task, int error)
909 {
910 	/* check for signal type interrupt code */
911 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
912 		error = -linux_schedule_get_interrupt_value(task);
913 		if (error == 0)
914 			error = EINTR;
915 	}
916 	return (error);
917 }
918 
919 static int
920 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
921     const struct file_operations *fop, u_long cmd, caddr_t data,
922     struct thread *td)
923 {
924 	struct task_struct *task = current;
925 	unsigned size;
926 	int error;
927 
928 	size = IOCPARM_LEN(cmd);
929 	/* refer to logic in sys_ioctl() */
930 	if (size > 0) {
931 		/*
932 		 * Setup hint for linux_copyin() and linux_copyout().
933 		 *
934 		 * Background: Linux code expects a user-space address
935 		 * while FreeBSD supplies a kernel-space address.
936 		 */
937 		task->bsd_ioctl_data = data;
938 		task->bsd_ioctl_len = size;
939 		data = (void *)LINUX_IOCTL_MIN_PTR;
940 	} else {
941 		/* fetch user-space pointer */
942 		data = *(void **)data;
943 	}
944 #ifdef COMPAT_FREEBSD32
945 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
946 		/* try the compat IOCTL handler first */
947 		if (fop->compat_ioctl != NULL) {
948 			error = -OPW(fp, td, fop->compat_ioctl(filp,
949 			    cmd, (u_long)data));
950 		} else {
951 			error = ENOTTY;
952 		}
953 
954 		/* fallback to the regular IOCTL handler, if any */
955 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
956 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
957 			    cmd, (u_long)data));
958 		}
959 	} else
960 #endif
961 	{
962 		if (fop->unlocked_ioctl != NULL) {
963 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
964 			    cmd, (u_long)data));
965 		} else {
966 			error = ENOTTY;
967 		}
968 	}
969 	if (size > 0) {
970 		task->bsd_ioctl_data = NULL;
971 		task->bsd_ioctl_len = 0;
972 	}
973 
974 	if (error == EWOULDBLOCK) {
975 		/* update kqfilter status, if any */
976 		linux_file_kqfilter_poll(filp,
977 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
978 	} else {
979 		error = linux_get_error(task, error);
980 	}
981 	return (error);
982 }
983 
984 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
985 
986 /*
987  * This function atomically updates the poll wakeup state and returns
988  * the previous state at the time of update.
989  */
990 static uint8_t
991 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
992 {
993 	int c, old;
994 
995 	c = v->counter;
996 
997 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
998 		c = old;
999 
1000 	return (c);
1001 }
1002 
1003 static int
1004 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1005 {
1006 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1007 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1008 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1009 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1010 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1011 	};
1012 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1013 
1014 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1015 	case LINUX_FWQ_STATE_QUEUED:
1016 		linux_poll_wakeup(filp);
1017 		return (1);
1018 	default:
1019 		return (0);
1020 	}
1021 }
1022 
1023 void
1024 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1025 {
1026 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1027 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1028 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1029 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1030 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1031 	};
1032 
1033 	/* check if we are called inside the select system call */
1034 	if (p == LINUX_POLL_TABLE_NORMAL)
1035 		selrecord(curthread, &filp->f_selinfo);
1036 
1037 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1038 	case LINUX_FWQ_STATE_INIT:
1039 		/* NOTE: file handles can only belong to one wait-queue */
1040 		filp->f_wait_queue.wqh = wqh;
1041 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1042 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1043 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1044 		break;
1045 	default:
1046 		break;
1047 	}
1048 }
1049 
1050 static void
1051 linux_poll_wait_dequeue(struct linux_file *filp)
1052 {
1053 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1054 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1055 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1056 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1057 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1058 	};
1059 
1060 	seldrain(&filp->f_selinfo);
1061 
1062 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1063 	case LINUX_FWQ_STATE_NOT_READY:
1064 	case LINUX_FWQ_STATE_QUEUED:
1065 	case LINUX_FWQ_STATE_READY:
1066 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1067 		break;
1068 	default:
1069 		break;
1070 	}
1071 }
1072 
1073 void
1074 linux_poll_wakeup(struct linux_file *filp)
1075 {
1076 	/* this function should be NULL-safe */
1077 	if (filp == NULL)
1078 		return;
1079 
1080 	selwakeup(&filp->f_selinfo);
1081 
1082 	spin_lock(&filp->f_kqlock);
1083 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1084 	    LINUX_KQ_FLAG_NEED_WRITE;
1085 
1086 	/* make sure the "knote" gets woken up */
1087 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1088 	spin_unlock(&filp->f_kqlock);
1089 }
1090 
1091 static struct linux_file *
1092 __get_file_rcu(struct linux_file **f)
1093 {
1094 	struct linux_file *file1, *file2;
1095 
1096 	file1 = READ_ONCE(*f);
1097 	if (file1 == NULL)
1098 		return (NULL);
1099 
1100 	if (!refcount_acquire_if_not_zero(
1101 	    file1->_file == NULL ? &file1->f_count : &file1->_file->f_count))
1102 		return (ERR_PTR(-EAGAIN));
1103 
1104 	file2 = READ_ONCE(*f);
1105 	if (file2 == file1)
1106 		return (file2);
1107 
1108 	fput(file1);
1109 	return (ERR_PTR(-EAGAIN));
1110 }
1111 
1112 struct linux_file *
1113 linux_get_file_rcu(struct linux_file **f)
1114 {
1115 	struct linux_file *file1;
1116 
1117 	for (;;) {
1118 		file1 = __get_file_rcu(f);
1119 		if (file1 == NULL)
1120 			return (NULL);
1121 
1122 		if (IS_ERR(file1))
1123 			continue;
1124 
1125 		return (file1);
1126 	}
1127 }
1128 
1129 struct linux_file *
1130 get_file_active(struct linux_file **f)
1131 {
1132 	struct linux_file *file1;
1133 
1134 	rcu_read_lock();
1135 	file1 = __get_file_rcu(f);
1136 	rcu_read_unlock();
1137 	if (IS_ERR(file1))
1138 		file1 = NULL;
1139 
1140 	return (file1);
1141 }
1142 
1143 static void
1144 linux_file_kqfilter_detach(struct knote *kn)
1145 {
1146 	struct linux_file *filp = kn->kn_hook;
1147 
1148 	spin_lock(&filp->f_kqlock);
1149 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1150 	spin_unlock(&filp->f_kqlock);
1151 }
1152 
1153 static int
1154 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1155 {
1156 	struct linux_file *filp = kn->kn_hook;
1157 
1158 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1159 
1160 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1161 }
1162 
1163 static int
1164 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1165 {
1166 	struct linux_file *filp = kn->kn_hook;
1167 
1168 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1169 
1170 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1171 }
1172 
1173 static const struct filterops linux_dev_kqfiltops_read = {
1174 	.f_isfd = 1,
1175 	.f_detach = linux_file_kqfilter_detach,
1176 	.f_event = linux_file_kqfilter_read_event,
1177 	.f_copy = knote_triv_copy,
1178 };
1179 
1180 static const struct filterops linux_dev_kqfiltops_write = {
1181 	.f_isfd = 1,
1182 	.f_detach = linux_file_kqfilter_detach,
1183 	.f_event = linux_file_kqfilter_write_event,
1184 	.f_copy = knote_triv_copy,
1185 };
1186 
1187 static void
1188 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1189 {
1190 	struct thread *td;
1191 	const struct file_operations *fop;
1192 	struct linux_cdev *ldev;
1193 	int temp;
1194 
1195 	if ((filp->f_kqflags & kqflags) == 0)
1196 		return;
1197 
1198 	td = curthread;
1199 
1200 	linux_get_fop(filp, &fop, &ldev);
1201 	/* get the latest polling state */
1202 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1203 	linux_drop_fop(ldev);
1204 
1205 	spin_lock(&filp->f_kqlock);
1206 	/* clear kqflags */
1207 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1208 	    LINUX_KQ_FLAG_NEED_WRITE);
1209 	/* update kqflags */
1210 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1211 		if ((temp & POLLIN) != 0)
1212 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1213 		if ((temp & POLLOUT) != 0)
1214 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1215 
1216 		/* make sure the "knote" gets woken up */
1217 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1218 	}
1219 	spin_unlock(&filp->f_kqlock);
1220 }
1221 
1222 static int
1223 linux_file_kqfilter(struct file *file, struct knote *kn)
1224 {
1225 	struct linux_file *filp;
1226 	struct thread *td;
1227 	int error;
1228 
1229 	td = curthread;
1230 	filp = (struct linux_file *)file->f_data;
1231 	filp->f_flags = file->f_flag;
1232 	if (filp->f_op->poll == NULL)
1233 		return (EINVAL);
1234 
1235 	spin_lock(&filp->f_kqlock);
1236 	switch (kn->kn_filter) {
1237 	case EVFILT_READ:
1238 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1239 		kn->kn_fop = &linux_dev_kqfiltops_read;
1240 		kn->kn_hook = filp;
1241 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1242 		error = 0;
1243 		break;
1244 	case EVFILT_WRITE:
1245 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1246 		kn->kn_fop = &linux_dev_kqfiltops_write;
1247 		kn->kn_hook = filp;
1248 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1249 		error = 0;
1250 		break;
1251 	default:
1252 		error = EINVAL;
1253 		break;
1254 	}
1255 	spin_unlock(&filp->f_kqlock);
1256 
1257 	if (error == 0) {
1258 		linux_set_current(td);
1259 
1260 		/* update kqfilter status, if any */
1261 		linux_file_kqfilter_poll(filp,
1262 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1263 	}
1264 	return (error);
1265 }
1266 
1267 static int
1268 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1269     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1270     int nprot, bool is_shared, struct thread *td)
1271 {
1272 	struct task_struct *task;
1273 	struct vm_area_struct *vmap;
1274 	struct mm_struct *mm;
1275 	struct linux_file *filp;
1276 	vm_memattr_t attr;
1277 	int error;
1278 
1279 	filp = (struct linux_file *)fp->f_data;
1280 	filp->f_flags = fp->f_flag;
1281 
1282 	if (fop->mmap == NULL)
1283 		return (EOPNOTSUPP);
1284 
1285 	linux_set_current(td);
1286 
1287 	/*
1288 	 * The same VM object might be shared by multiple processes
1289 	 * and the mm_struct is usually freed when a process exits.
1290 	 *
1291 	 * The atomic reference below makes sure the mm_struct is
1292 	 * available as long as the vmap is in the linux_vma_head.
1293 	 */
1294 	task = current;
1295 	mm = task->mm;
1296 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1297 		return (EINVAL);
1298 
1299 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1300 	vmap->vm_start = 0;
1301 	vmap->vm_end = size;
1302 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1303 	vmap->vm_pfn = 0;
1304 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1305 	if (is_shared)
1306 		vmap->vm_flags |= VM_SHARED;
1307 	vmap->vm_ops = NULL;
1308 	vmap->vm_file = get_file(filp);
1309 	vmap->vm_mm = mm;
1310 
1311 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1312 		error = linux_get_error(task, EINTR);
1313 	} else {
1314 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1315 		error = linux_get_error(task, error);
1316 		up_write(&vmap->vm_mm->mmap_sem);
1317 	}
1318 
1319 	if (error != 0) {
1320 		linux_cdev_handle_free(vmap);
1321 		return (error);
1322 	}
1323 
1324 	attr = pgprot2cachemode(vmap->vm_page_prot);
1325 
1326 	if (vmap->vm_ops != NULL) {
1327 		struct vm_area_struct *ptr;
1328 		void *vm_private_data;
1329 		bool vm_no_fault;
1330 
1331 		if (vmap->vm_ops->open == NULL ||
1332 		    vmap->vm_ops->close == NULL ||
1333 		    vmap->vm_private_data == NULL) {
1334 			/* free allocated VM area struct */
1335 			linux_cdev_handle_free(vmap);
1336 			return (EINVAL);
1337 		}
1338 
1339 		vm_private_data = vmap->vm_private_data;
1340 
1341 		rw_wlock(&linux_vma_lock);
1342 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1343 			if (ptr->vm_private_data == vm_private_data)
1344 				break;
1345 		}
1346 		/* check if there is an existing VM area struct */
1347 		if (ptr != NULL) {
1348 			/* check if the VM area structure is invalid */
1349 			if (ptr->vm_ops == NULL ||
1350 			    ptr->vm_ops->open == NULL ||
1351 			    ptr->vm_ops->close == NULL) {
1352 				error = ESTALE;
1353 				vm_no_fault = 1;
1354 			} else {
1355 				error = EEXIST;
1356 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1357 			}
1358 		} else {
1359 			/* insert VM area structure into list */
1360 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1361 			error = 0;
1362 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1363 		}
1364 		rw_wunlock(&linux_vma_lock);
1365 
1366 		if (error != 0) {
1367 			/* free allocated VM area struct */
1368 			linux_cdev_handle_free(vmap);
1369 			/* check for stale VM area struct */
1370 			if (error != EEXIST)
1371 				return (error);
1372 		}
1373 
1374 		/* check if there is no fault handler */
1375 		if (vm_no_fault) {
1376 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1377 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1378 			    td->td_ucred);
1379 		} else {
1380 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1381 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1382 			    td->td_ucred);
1383 		}
1384 
1385 		/* check if allocating the VM object failed */
1386 		if (*object == NULL) {
1387 			if (error == 0) {
1388 				/* remove VM area struct from list */
1389 				linux_cdev_handle_remove(vmap);
1390 				/* free allocated VM area struct */
1391 				linux_cdev_handle_free(vmap);
1392 			}
1393 			return (EINVAL);
1394 		}
1395 	} else {
1396 		struct sglist *sg;
1397 
1398 		sg = sglist_alloc(1, M_WAITOK);
1399 		sglist_append_phys(sg,
1400 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1401 
1402 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1403 		    nprot, 0, td->td_ucred);
1404 
1405 		linux_cdev_handle_free(vmap);
1406 
1407 		if (*object == NULL) {
1408 			sglist_free(sg);
1409 			return (EINVAL);
1410 		}
1411 	}
1412 
1413 	if (attr != VM_MEMATTR_DEFAULT) {
1414 		VM_OBJECT_WLOCK(*object);
1415 		vm_object_set_memattr(*object, attr);
1416 		VM_OBJECT_WUNLOCK(*object);
1417 	}
1418 	*offset = 0;
1419 	return (0);
1420 }
1421 
1422 struct cdevsw linuxcdevsw = {
1423 	.d_version = D_VERSION,
1424 	.d_fdopen = linux_dev_fdopen,
1425 	.d_name = "lkpidev",
1426 };
1427 
1428 static int
1429 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1430     int flags, struct thread *td)
1431 {
1432 	struct linux_file *filp;
1433 	const struct file_operations *fop;
1434 	struct linux_cdev *ldev;
1435 	ssize_t bytes;
1436 	int error;
1437 
1438 	error = 0;
1439 	filp = (struct linux_file *)file->f_data;
1440 	filp->f_flags = file->f_flag;
1441 	/* XXX no support for I/O vectors currently */
1442 	if (uio->uio_iovcnt != 1)
1443 		return (EOPNOTSUPP);
1444 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1445 		return (EINVAL);
1446 	linux_set_current(td);
1447 	linux_get_fop(filp, &fop, &ldev);
1448 	if (fop->read != NULL) {
1449 		bytes = OPW(file, td, fop->read(filp,
1450 		    uio->uio_iov->iov_base,
1451 		    uio->uio_iov->iov_len, &uio->uio_offset));
1452 		if (bytes >= 0) {
1453 			uio->uio_iov->iov_base =
1454 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1455 			uio->uio_iov->iov_len -= bytes;
1456 			uio->uio_resid -= bytes;
1457 		} else {
1458 			error = linux_get_error(current, -bytes);
1459 		}
1460 	} else
1461 		error = ENXIO;
1462 
1463 	/* update kqfilter status, if any */
1464 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1465 	linux_drop_fop(ldev);
1466 
1467 	return (error);
1468 }
1469 
1470 static int
1471 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1472     int flags, struct thread *td)
1473 {
1474 	struct linux_file *filp;
1475 	const struct file_operations *fop;
1476 	struct linux_cdev *ldev;
1477 	ssize_t bytes;
1478 	int error;
1479 
1480 	filp = (struct linux_file *)file->f_data;
1481 	filp->f_flags = file->f_flag;
1482 	/* XXX no support for I/O vectors currently */
1483 	if (uio->uio_iovcnt != 1)
1484 		return (EOPNOTSUPP);
1485 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1486 		return (EINVAL);
1487 	linux_set_current(td);
1488 	linux_get_fop(filp, &fop, &ldev);
1489 	if (fop->write != NULL) {
1490 		bytes = OPW(file, td, fop->write(filp,
1491 		    uio->uio_iov->iov_base,
1492 		    uio->uio_iov->iov_len, &uio->uio_offset));
1493 		if (bytes >= 0) {
1494 			uio->uio_iov->iov_base =
1495 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1496 			uio->uio_iov->iov_len -= bytes;
1497 			uio->uio_resid -= bytes;
1498 			error = 0;
1499 		} else {
1500 			error = linux_get_error(current, -bytes);
1501 		}
1502 	} else
1503 		error = ENXIO;
1504 
1505 	/* update kqfilter status, if any */
1506 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1507 
1508 	linux_drop_fop(ldev);
1509 
1510 	return (error);
1511 }
1512 
1513 static int
1514 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1515     struct thread *td)
1516 {
1517 	struct linux_file *filp;
1518 	const struct file_operations *fop;
1519 	struct linux_cdev *ldev;
1520 	int revents;
1521 
1522 	filp = (struct linux_file *)file->f_data;
1523 	filp->f_flags = file->f_flag;
1524 	linux_set_current(td);
1525 	linux_get_fop(filp, &fop, &ldev);
1526 	if (fop->poll != NULL) {
1527 		revents = OPW(file, td, fop->poll(filp,
1528 		    LINUX_POLL_TABLE_NORMAL)) & events;
1529 	} else {
1530 		revents = 0;
1531 	}
1532 	linux_drop_fop(ldev);
1533 	return (revents);
1534 }
1535 
1536 static int
1537 linux_file_close(struct file *file, struct thread *td)
1538 {
1539 	struct linux_file *filp;
1540 	int (*release)(struct inode *, struct linux_file *);
1541 	const struct file_operations *fop;
1542 	struct linux_cdev *ldev;
1543 	int error;
1544 
1545 	filp = (struct linux_file *)file->f_data;
1546 
1547 	KASSERT(file_count(filp) == 0,
1548 	    ("File refcount(%d) is not zero", file_count(filp)));
1549 
1550 	if (td == NULL)
1551 		td = curthread;
1552 
1553 	error = 0;
1554 	filp->f_flags = file->f_flag;
1555 	linux_set_current(td);
1556 	linux_poll_wait_dequeue(filp);
1557 	linux_get_fop(filp, &fop, &ldev);
1558 	/*
1559 	 * Always use the real release function, if any, to avoid
1560 	 * leaking device resources:
1561 	 */
1562 	release = filp->f_op->release;
1563 	if (release != NULL)
1564 		error = -OPW(file, td, release(filp->f_vnode, filp));
1565 	funsetown(&filp->f_sigio);
1566 	if (filp->f_vnode != NULL)
1567 		vrele(filp->f_vnode);
1568 	linux_drop_fop(ldev);
1569 	ldev = filp->f_cdev;
1570 	if (ldev != NULL)
1571 		linux_cdev_deref(ldev);
1572 	linux_synchronize_rcu(RCU_TYPE_REGULAR);
1573 	kfree(filp);
1574 
1575 	return (error);
1576 }
1577 
1578 static int
1579 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1580     struct thread *td)
1581 {
1582 	struct linux_file *filp;
1583 	const struct file_operations *fop;
1584 	struct linux_cdev *ldev;
1585 	struct fiodgname_arg *fgn;
1586 	const char *p;
1587 	int error, i;
1588 
1589 	error = 0;
1590 	filp = (struct linux_file *)fp->f_data;
1591 	filp->f_flags = fp->f_flag;
1592 	linux_get_fop(filp, &fop, &ldev);
1593 
1594 	linux_set_current(td);
1595 	switch (cmd) {
1596 	case FIONBIO:
1597 		break;
1598 	case FIOASYNC:
1599 		if (fop->fasync == NULL)
1600 			break;
1601 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1602 		break;
1603 	case FIOSETOWN:
1604 		error = fsetown(*(int *)data, &filp->f_sigio);
1605 		if (error == 0) {
1606 			if (fop->fasync == NULL)
1607 				break;
1608 			error = -OPW(fp, td, fop->fasync(0, filp,
1609 			    fp->f_flag & FASYNC));
1610 		}
1611 		break;
1612 	case FIOGETOWN:
1613 		*(int *)data = fgetown(&filp->f_sigio);
1614 		break;
1615 	case FIODGNAME:
1616 #ifdef	COMPAT_FREEBSD32
1617 	case FIODGNAME_32:
1618 #endif
1619 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1620 			error = ENXIO;
1621 			break;
1622 		}
1623 		fgn = data;
1624 		p = devtoname(filp->f_cdev->cdev);
1625 		i = strlen(p) + 1;
1626 		if (i > fgn->len) {
1627 			error = EINVAL;
1628 			break;
1629 		}
1630 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1631 		break;
1632 	default:
1633 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1634 		break;
1635 	}
1636 	linux_drop_fop(ldev);
1637 	return (error);
1638 }
1639 
1640 static int
1641 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1642     vm_prot_t maxprot, int flags, struct file *fp,
1643     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1644 {
1645 	/*
1646 	 * Character devices do not provide private mappings
1647 	 * of any kind:
1648 	 */
1649 	if ((maxprot & VM_PROT_WRITE) == 0 &&
1650 	    (prot & VM_PROT_WRITE) != 0)
1651 		return (EACCES);
1652 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1653 		return (EINVAL);
1654 
1655 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1656 	    (int)prot, (flags & MAP_SHARED) ? true : false, td));
1657 }
1658 
1659 static int
1660 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1661     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1662     struct thread *td)
1663 {
1664 	struct linux_file *filp;
1665 	const struct file_operations *fop;
1666 	struct linux_cdev *ldev;
1667 	struct mount *mp;
1668 	struct vnode *vp;
1669 	vm_object_t object;
1670 	vm_prot_t maxprot;
1671 	int error;
1672 
1673 	filp = (struct linux_file *)fp->f_data;
1674 
1675 	vp = filp->f_vnode;
1676 	if (vp == NULL)
1677 		return (EOPNOTSUPP);
1678 
1679 	/*
1680 	 * Ensure that file and memory protections are
1681 	 * compatible.
1682 	 */
1683 	mp = vp->v_mount;
1684 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1685 		maxprot = VM_PROT_NONE;
1686 		if ((prot & VM_PROT_EXECUTE) != 0)
1687 			return (EACCES);
1688 	} else
1689 		maxprot = VM_PROT_EXECUTE;
1690 	if ((fp->f_flag & FREAD) != 0)
1691 		maxprot |= VM_PROT_READ;
1692 	else if ((prot & VM_PROT_READ) != 0)
1693 		return (EACCES);
1694 
1695 	/*
1696 	 * If we are sharing potential changes via MAP_SHARED and we
1697 	 * are trying to get write permission although we opened it
1698 	 * without asking for it, bail out.
1699 	 *
1700 	 * Note that most character devices always share mappings.
1701 	 *
1702 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1703 	 * requests rather than doing it here.
1704 	 */
1705 	if ((flags & MAP_SHARED) != 0) {
1706 		if ((fp->f_flag & FWRITE) != 0)
1707 			maxprot |= VM_PROT_WRITE;
1708 		else if ((prot & VM_PROT_WRITE) != 0)
1709 			return (EACCES);
1710 	}
1711 	maxprot &= cap_maxprot;
1712 
1713 	linux_get_fop(filp, &fop, &ldev);
1714 	error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1715 	    &foff, fop, &object);
1716 	if (error != 0)
1717 		goto out;
1718 
1719 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1720 	    foff, FALSE, td);
1721 	if (error != 0)
1722 		vm_object_deallocate(object);
1723 out:
1724 	linux_drop_fop(ldev);
1725 	return (error);
1726 }
1727 
1728 static int
1729 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1730 {
1731 	struct linux_file *filp;
1732 	struct vnode *vp;
1733 	int error;
1734 
1735 	filp = (struct linux_file *)fp->f_data;
1736 	if (filp->f_vnode == NULL)
1737 		return (EOPNOTSUPP);
1738 
1739 	vp = filp->f_vnode;
1740 
1741 	vn_lock(vp, LK_SHARED | LK_RETRY);
1742 	error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1743 	VOP_UNLOCK(vp);
1744 
1745 	return (error);
1746 }
1747 
1748 static int
1749 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1750     struct filedesc *fdp)
1751 {
1752 	struct linux_file *filp;
1753 	struct vnode *vp;
1754 	int error;
1755 
1756 	filp = fp->f_data;
1757 	vp = filp->f_vnode;
1758 	if (vp == NULL) {
1759 		error = 0;
1760 		kif->kf_type = KF_TYPE_DEV;
1761 	} else {
1762 		vref(vp);
1763 		FILEDESC_SUNLOCK(fdp);
1764 		error = vn_fill_kinfo_vnode(vp, kif);
1765 		vrele(vp);
1766 		kif->kf_type = KF_TYPE_VNODE;
1767 		FILEDESC_SLOCK(fdp);
1768 	}
1769 	return (error);
1770 }
1771 
1772 unsigned int
1773 linux_iminor(struct inode *inode)
1774 {
1775 	struct linux_cdev *ldev;
1776 
1777 	if (inode == NULL || inode->v_rdev == NULL ||
1778 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1779 		return (-1U);
1780 	ldev = inode->v_rdev->si_drv1;
1781 	if (ldev == NULL)
1782 		return (-1U);
1783 
1784 	return (minor(ldev->dev));
1785 }
1786 
1787 static int
1788 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td)
1789 {
1790 	struct linux_file *filp1, *filp2;
1791 
1792 	if (fp2->f_type != DTYPE_DEV)
1793 		return (3);
1794 
1795 	filp1 = fp1->f_data;
1796 	filp2 = fp2->f_data;
1797 	return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev));
1798 }
1799 
1800 const struct fileops linuxfileops = {
1801 	.fo_read = linux_file_read,
1802 	.fo_write = linux_file_write,
1803 	.fo_truncate = invfo_truncate,
1804 	.fo_kqfilter = linux_file_kqfilter,
1805 	.fo_stat = linux_file_stat,
1806 	.fo_fill_kinfo = linux_file_fill_kinfo,
1807 	.fo_poll = linux_file_poll,
1808 	.fo_close = linux_file_close,
1809 	.fo_ioctl = linux_file_ioctl,
1810 	.fo_mmap = linux_file_mmap,
1811 	.fo_chmod = invfo_chmod,
1812 	.fo_chown = invfo_chown,
1813 	.fo_sendfile = invfo_sendfile,
1814 	.fo_cmp = linux_file_kcmp,
1815 	.fo_flags = DFLAG_PASSABLE,
1816 };
1817 
1818 static char *
1819 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1820 {
1821 	unsigned int len;
1822 	char *p;
1823 	va_list aq;
1824 
1825 	va_copy(aq, ap);
1826 	len = vsnprintf(NULL, 0, fmt, aq);
1827 	va_end(aq);
1828 
1829 	if (dev != NULL)
1830 		p = devm_kmalloc(dev, len + 1, gfp);
1831 	else
1832 		p = kmalloc(len + 1, gfp);
1833 	if (p != NULL)
1834 		vsnprintf(p, len + 1, fmt, ap);
1835 
1836 	return (p);
1837 }
1838 
1839 char *
1840 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1841 {
1842 
1843 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1844 }
1845 
1846 char *
1847 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1848 {
1849 	va_list ap;
1850 	char *p;
1851 
1852 	va_start(ap, fmt);
1853 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1854 	va_end(ap);
1855 
1856 	return (p);
1857 }
1858 
1859 char *
1860 kasprintf(gfp_t gfp, const char *fmt, ...)
1861 {
1862 	va_list ap;
1863 	char *p;
1864 
1865 	va_start(ap, fmt);
1866 	p = kvasprintf(gfp, fmt, ap);
1867 	va_end(ap);
1868 
1869 	return (p);
1870 }
1871 
1872 int
1873 __lkpi_hexdump_printf(void *arg1 __unused, const char *fmt, ...)
1874 {
1875 	va_list ap;
1876 	int result;
1877 
1878 	va_start(ap, fmt);
1879 	result = vprintf(fmt, ap);
1880 	va_end(ap);
1881 	return (result);
1882 }
1883 
1884 int
1885 __lkpi_hexdump_sbuf_printf(void *arg1, const char *fmt, ...)
1886 {
1887 	va_list ap;
1888 	int result;
1889 
1890 	va_start(ap, fmt);
1891 	result = sbuf_vprintf(arg1, fmt, ap);
1892 	va_end(ap);
1893 	return (result);
1894 }
1895 
1896 void
1897 lkpi_hex_dump(int(*_fpf)(void *, const char *, ...), void *arg1,
1898     const char *level, const char *prefix_str,
1899     const int prefix_type, const int rowsize, const int groupsize,
1900     const void *buf, size_t len, const bool ascii, const bool trailing_newline)
1901 {
1902 	typedef const struct { long long value; } __packed *print_64p_t;
1903 	typedef const struct { uint32_t value; } __packed *print_32p_t;
1904 	typedef const struct { uint16_t value; } __packed *print_16p_t;
1905 	const void *buf_old = buf;
1906 	int row, linelen, ret;
1907 
1908 	while (len > 0) {
1909 		linelen = 0;
1910 		if (level != NULL) {
1911 			ret = _fpf(arg1, "%s", level);
1912 			if (ret < 0)
1913 				break;
1914 			linelen += ret;
1915 		}
1916 		if (prefix_str != NULL) {
1917 			ret = _fpf(
1918 			    arg1, "%s%s", linelen ? " " : "", prefix_str);
1919 			if (ret < 0)
1920 				break;
1921 			linelen += ret;
1922 		}
1923 
1924 		switch (prefix_type) {
1925 		case DUMP_PREFIX_ADDRESS:
1926 			ret = _fpf(
1927 			    arg1, "%s[%p]", linelen ? " " : "", buf);
1928 			if (ret < 0)
1929 				return;
1930 			linelen += ret;
1931 			break;
1932 		case DUMP_PREFIX_OFFSET:
1933 			ret = _fpf(
1934 			    arg1, "%s[%#tx]", linelen ? " " : "",
1935 			    ((const char *)buf - (const char *)buf_old));
1936 			if (ret < 0)
1937 				return;
1938 			linelen += ret;
1939 			break;
1940 		default:
1941 			break;
1942 		}
1943 		for (row = 0; row != rowsize; row++) {
1944 			if (groupsize == 8 && len > 7) {
1945 				ret = _fpf(
1946 				    arg1, "%s%016llx", linelen ? " " : "",
1947 				    ((print_64p_t)buf)->value);
1948 				if (ret < 0)
1949 					return;
1950 				linelen += ret;
1951 				buf = (const uint8_t *)buf + 8;
1952 				len -= 8;
1953 			} else if (groupsize == 4 && len > 3) {
1954 				ret = _fpf(
1955 				    arg1, "%s%08x", linelen ? " " : "",
1956 				    ((print_32p_t)buf)->value);
1957 				if (ret < 0)
1958 					return;
1959 				linelen += ret;
1960 				buf = (const uint8_t *)buf + 4;
1961 				len -= 4;
1962 			} else if (groupsize == 2 && len > 1) {
1963 				ret = _fpf(
1964 				    arg1, "%s%04x", linelen ? " " : "",
1965 				    ((print_16p_t)buf)->value);
1966 				if (ret < 0)
1967 					return;
1968 				linelen += ret;
1969 				buf = (const uint8_t *)buf + 2;
1970 				len -= 2;
1971 			} else if (len > 0) {
1972 				ret = _fpf(
1973 				    arg1, "%s%02x", linelen ? " " : "",
1974 				    *(const uint8_t *)buf);
1975 				if (ret < 0)
1976 					return;
1977 				linelen += ret;
1978 				buf = (const uint8_t *)buf + 1;
1979 				len--;
1980 			} else {
1981 				break;
1982 			}
1983 		}
1984 		if (len > 0 && trailing_newline) {
1985 			ret = _fpf(arg1, "\n");
1986 			if (ret < 0)
1987 				break;
1988 		}
1989 	}
1990 }
1991 
1992 struct hdtb_context {
1993 	char	*linebuf;
1994 	size_t	 linebuflen;
1995 	int	 written;
1996 };
1997 
1998 static int
1999 hdtb_cb(void *arg, const char *format, ...)
2000 {
2001 	struct hdtb_context *context;
2002 	int written;
2003 	va_list args;
2004 
2005 	context = arg;
2006 
2007 	va_start(args, format);
2008 	written = vsnprintf(
2009 	    context->linebuf, context->linebuflen, format, args);
2010 	va_end(args);
2011 
2012 	if (written < 0)
2013 		return (written);
2014 
2015 	/*
2016 	 * Linux' hex_dump_to_buffer() function has the same behaviour as
2017 	 * snprintf() basically. Therefore, it returns the number of bytes it
2018 	 * would have written if the destination buffer was large enough.
2019 	 *
2020 	 * If the destination buffer was exhausted, lkpi_hex_dump() will
2021 	 * continue to call this callback but it will only compute the bytes it
2022 	 * would have written but write nothing to that buffer.
2023 	 */
2024 	context->written += written;
2025 
2026 	if (written < context->linebuflen) {
2027 		context->linebuf += written;
2028 		context->linebuflen -= written;
2029 	} else {
2030 		context->linebuf += context->linebuflen;
2031 		context->linebuflen = 0;
2032 	}
2033 
2034 	return (written);
2035 }
2036 
2037 int
2038 lkpi_hex_dump_to_buffer(const void *buf, size_t len, int rowsize,
2039     int groupsize, char *linebuf, size_t linebuflen, bool ascii)
2040 {
2041 	int written;
2042 	struct hdtb_context context;
2043 
2044 	context.linebuf = linebuf;
2045 	context.linebuflen = linebuflen;
2046 	context.written = 0;
2047 
2048 	if (rowsize != 16 && rowsize != 32)
2049 		rowsize = 16;
2050 
2051 	len = min(len, rowsize);
2052 
2053 	lkpi_hex_dump(
2054 	    hdtb_cb, &context, NULL, NULL, DUMP_PREFIX_NONE,
2055 	    rowsize, groupsize, buf, len, ascii, false);
2056 
2057 	written = context.written;
2058 
2059 	return (written);
2060 }
2061 
2062 static void
2063 linux_timer_callback_wrapper(void *context)
2064 {
2065 	struct timer_list *timer;
2066 
2067 	timer = context;
2068 
2069 	/* the timer is about to be shutdown permanently */
2070 	if (timer->function == NULL)
2071 		return;
2072 
2073 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
2074 		/* try again later */
2075 		callout_reset(&timer->callout, 1,
2076 		    &linux_timer_callback_wrapper, timer);
2077 		return;
2078 	}
2079 
2080 	timer->function(timer->data);
2081 }
2082 
2083 static int
2084 linux_timer_jiffies_until(unsigned long expires)
2085 {
2086 	unsigned long delta = expires - jiffies;
2087 
2088 	/*
2089 	 * Guard against already expired values and make sure that the value can
2090 	 * be used as a tick count, rather than a jiffies count.
2091 	 */
2092 	if ((long)delta < 1)
2093 		delta = 1;
2094 	else if (delta > INT_MAX)
2095 		delta = INT_MAX;
2096 	return ((int)delta);
2097 }
2098 
2099 int
2100 mod_timer(struct timer_list *timer, unsigned long expires)
2101 {
2102 	int ret;
2103 
2104 	timer->expires = expires;
2105 	ret = callout_reset(&timer->callout,
2106 	    linux_timer_jiffies_until(expires),
2107 	    &linux_timer_callback_wrapper, timer);
2108 
2109 	MPASS(ret == 0 || ret == 1);
2110 
2111 	return (ret == 1);
2112 }
2113 
2114 void
2115 add_timer(struct timer_list *timer)
2116 {
2117 
2118 	callout_reset(&timer->callout,
2119 	    linux_timer_jiffies_until(timer->expires),
2120 	    &linux_timer_callback_wrapper, timer);
2121 }
2122 
2123 void
2124 add_timer_on(struct timer_list *timer, int cpu)
2125 {
2126 
2127 	callout_reset_on(&timer->callout,
2128 	    linux_timer_jiffies_until(timer->expires),
2129 	    &linux_timer_callback_wrapper, timer, cpu);
2130 }
2131 
2132 int
2133 timer_delete(struct timer_list *timer)
2134 {
2135 
2136 	if (callout_stop(&(timer)->callout) == -1)
2137 		return (0);
2138 	return (1);
2139 }
2140 
2141 int
2142 timer_delete_sync(struct timer_list *timer)
2143 {
2144 
2145 	if (callout_drain(&(timer)->callout) == -1)
2146 		return (0);
2147 	return (1);
2148 }
2149 
2150 int
2151 timer_shutdown_sync(struct timer_list *timer)
2152 {
2153 
2154 	timer->function = NULL;
2155 	return (del_timer_sync(timer));
2156 }
2157 
2158 /* greatest common divisor, Euclid equation */
2159 static uint64_t
2160 lkpi_gcd_64(uint64_t a, uint64_t b)
2161 {
2162 	uint64_t an;
2163 	uint64_t bn;
2164 
2165 	while (b != 0) {
2166 		an = b;
2167 		bn = a % b;
2168 		a = an;
2169 		b = bn;
2170 	}
2171 	return (a);
2172 }
2173 
2174 uint64_t lkpi_nsec2hz_rem;
2175 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2176 uint64_t lkpi_nsec2hz_max;
2177 
2178 uint64_t lkpi_usec2hz_rem;
2179 uint64_t lkpi_usec2hz_div = 1000000ULL;
2180 uint64_t lkpi_usec2hz_max;
2181 
2182 uint64_t lkpi_msec2hz_rem;
2183 uint64_t lkpi_msec2hz_div = 1000ULL;
2184 uint64_t lkpi_msec2hz_max;
2185 
2186 static void
2187 linux_timer_init(void *arg)
2188 {
2189 	uint64_t gcd;
2190 
2191 	/*
2192 	 * Compute an internal HZ value which can divide 2**32 to
2193 	 * avoid timer rounding problems when the tick value wraps
2194 	 * around 2**32:
2195 	 */
2196 	linux_timer_hz_mask = 1;
2197 	while (linux_timer_hz_mask < (unsigned long)hz)
2198 		linux_timer_hz_mask *= 2;
2199 	linux_timer_hz_mask--;
2200 
2201 	/* compute some internal constants */
2202 
2203 	lkpi_nsec2hz_rem = hz;
2204 	lkpi_usec2hz_rem = hz;
2205 	lkpi_msec2hz_rem = hz;
2206 
2207 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2208 	lkpi_nsec2hz_rem /= gcd;
2209 	lkpi_nsec2hz_div /= gcd;
2210 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2211 
2212 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2213 	lkpi_usec2hz_rem /= gcd;
2214 	lkpi_usec2hz_div /= gcd;
2215 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2216 
2217 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2218 	lkpi_msec2hz_rem /= gcd;
2219 	lkpi_msec2hz_div /= gcd;
2220 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2221 }
2222 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2223 
2224 void
2225 linux_complete_common(struct completion *c, int all)
2226 {
2227 	sleepq_lock(c);
2228 	if (all) {
2229 		c->done = UINT_MAX;
2230 		sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2231 	} else {
2232 		if (c->done != UINT_MAX)
2233 			c->done++;
2234 		sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2235 	}
2236 	sleepq_release(c);
2237 }
2238 
2239 /*
2240  * Indefinite wait for done != 0 with or without signals.
2241  */
2242 int
2243 linux_wait_for_common(struct completion *c, int flags)
2244 {
2245 	struct task_struct *task;
2246 	int error;
2247 
2248 	if (SCHEDULER_STOPPED())
2249 		return (0);
2250 
2251 	task = current;
2252 
2253 	if (flags != 0)
2254 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2255 	else
2256 		flags = SLEEPQ_SLEEP;
2257 	error = 0;
2258 	for (;;) {
2259 		sleepq_lock(c);
2260 		if (c->done)
2261 			break;
2262 		sleepq_add(c, NULL, "completion", flags, 0);
2263 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2264 			DROP_GIANT();
2265 			error = -sleepq_wait_sig(c, 0);
2266 			PICKUP_GIANT();
2267 			if (error != 0) {
2268 				linux_schedule_save_interrupt_value(task, error);
2269 				error = -ERESTARTSYS;
2270 				goto intr;
2271 			}
2272 		} else {
2273 			DROP_GIANT();
2274 			sleepq_wait(c, 0);
2275 			PICKUP_GIANT();
2276 		}
2277 	}
2278 	if (c->done != UINT_MAX)
2279 		c->done--;
2280 	sleepq_release(c);
2281 
2282 intr:
2283 	return (error);
2284 }
2285 
2286 /*
2287  * Time limited wait for done != 0 with or without signals.
2288  */
2289 unsigned long
2290 linux_wait_for_timeout_common(struct completion *c, unsigned long timeout,
2291     int flags)
2292 {
2293 	struct task_struct *task;
2294 	unsigned long end = jiffies + timeout, error;
2295 
2296 	if (SCHEDULER_STOPPED())
2297 		return (0);
2298 
2299 	task = current;
2300 
2301 	if (flags != 0)
2302 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2303 	else
2304 		flags = SLEEPQ_SLEEP;
2305 
2306 	for (;;) {
2307 		sleepq_lock(c);
2308 		if (c->done)
2309 			break;
2310 		sleepq_add(c, NULL, "completion", flags, 0);
2311 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2312 
2313 		DROP_GIANT();
2314 		if (flags & SLEEPQ_INTERRUPTIBLE)
2315 			error = -sleepq_timedwait_sig(c, 0);
2316 		else
2317 			error = -sleepq_timedwait(c, 0);
2318 		PICKUP_GIANT();
2319 
2320 		if (error != 0) {
2321 			/* check for timeout */
2322 			if (error == -EWOULDBLOCK) {
2323 				error = 0;	/* timeout */
2324 			} else {
2325 				/* signal happened */
2326 				linux_schedule_save_interrupt_value(task, error);
2327 				error = -ERESTARTSYS;
2328 			}
2329 			goto done;
2330 		}
2331 	}
2332 	if (c->done != UINT_MAX)
2333 		c->done--;
2334 	sleepq_release(c);
2335 
2336 	/* return how many jiffies are left */
2337 	error = linux_timer_jiffies_until(end);
2338 done:
2339 	return (error);
2340 }
2341 
2342 int
2343 linux_try_wait_for_completion(struct completion *c)
2344 {
2345 	int isdone;
2346 
2347 	sleepq_lock(c);
2348 	isdone = (c->done != 0);
2349 	if (c->done != 0 && c->done != UINT_MAX)
2350 		c->done--;
2351 	sleepq_release(c);
2352 	return (isdone);
2353 }
2354 
2355 int
2356 linux_completion_done(struct completion *c)
2357 {
2358 	int isdone;
2359 
2360 	sleepq_lock(c);
2361 	isdone = (c->done != 0);
2362 	sleepq_release(c);
2363 	return (isdone);
2364 }
2365 
2366 static void
2367 linux_cdev_deref(struct linux_cdev *ldev)
2368 {
2369 	if (refcount_release(&ldev->refs) &&
2370 	    ldev->kobj.ktype == &linux_cdev_ktype)
2371 		kfree(ldev);
2372 }
2373 
2374 static void
2375 linux_cdev_release(struct kobject *kobj)
2376 {
2377 	struct linux_cdev *cdev;
2378 	struct kobject *parent;
2379 
2380 	cdev = container_of(kobj, struct linux_cdev, kobj);
2381 	parent = kobj->parent;
2382 	linux_destroy_dev(cdev);
2383 	linux_cdev_deref(cdev);
2384 	kobject_put(parent);
2385 }
2386 
2387 static void
2388 linux_cdev_static_release(struct kobject *kobj)
2389 {
2390 	struct cdev *cdev;
2391 	struct linux_cdev *ldev;
2392 
2393 	ldev = container_of(kobj, struct linux_cdev, kobj);
2394 	cdev = ldev->cdev;
2395 	if (cdev != NULL) {
2396 		destroy_dev(cdev);
2397 		ldev->cdev = NULL;
2398 	}
2399 	kobject_put(kobj->parent);
2400 }
2401 
2402 int
2403 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2404 {
2405 	int ret;
2406 
2407 	if (dev->devt != 0) {
2408 		/* Set parent kernel object. */
2409 		ldev->kobj.parent = &dev->kobj;
2410 
2411 		/*
2412 		 * Unlike Linux we require the kobject of the
2413 		 * character device structure to have a valid name
2414 		 * before calling this function:
2415 		 */
2416 		if (ldev->kobj.name == NULL)
2417 			return (-EINVAL);
2418 
2419 		ret = cdev_add(ldev, dev->devt, 1);
2420 		if (ret)
2421 			return (ret);
2422 	}
2423 	ret = device_add(dev);
2424 	if (ret != 0 && dev->devt != 0)
2425 		cdev_del(ldev);
2426 	return (ret);
2427 }
2428 
2429 void
2430 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2431 {
2432 	device_del(dev);
2433 
2434 	if (dev->devt != 0)
2435 		cdev_del(ldev);
2436 }
2437 
2438 static void
2439 linux_destroy_dev(struct linux_cdev *ldev)
2440 {
2441 
2442 	if (ldev->cdev == NULL)
2443 		return;
2444 
2445 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2446 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2447 
2448 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2449 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2450 		pause("ldevdtr", hz / 4);
2451 
2452 	destroy_dev(ldev->cdev);
2453 	ldev->cdev = NULL;
2454 }
2455 
2456 const struct kobj_type linux_cdev_ktype = {
2457 	.release = linux_cdev_release,
2458 };
2459 
2460 const struct kobj_type linux_cdev_static_ktype = {
2461 	.release = linux_cdev_static_release,
2462 };
2463 
2464 static void
2465 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2466 {
2467 	struct notifier_block *nb;
2468 	struct netdev_notifier_info ni;
2469 
2470 	nb = arg;
2471 	ni.ifp = ifp;
2472 	ni.dev = (struct net_device *)ifp;
2473 	if (linkstate == LINK_STATE_UP)
2474 		nb->notifier_call(nb, NETDEV_UP, &ni);
2475 	else
2476 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2477 }
2478 
2479 static void
2480 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2481 {
2482 	struct notifier_block *nb;
2483 	struct netdev_notifier_info ni;
2484 
2485 	nb = arg;
2486 	ni.ifp = ifp;
2487 	ni.dev = (struct net_device *)ifp;
2488 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2489 }
2490 
2491 static void
2492 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2493 {
2494 	struct notifier_block *nb;
2495 	struct netdev_notifier_info ni;
2496 
2497 	nb = arg;
2498 	ni.ifp = ifp;
2499 	ni.dev = (struct net_device *)ifp;
2500 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2501 }
2502 
2503 static void
2504 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2505 {
2506 	struct notifier_block *nb;
2507 	struct netdev_notifier_info ni;
2508 
2509 	nb = arg;
2510 	ni.ifp = ifp;
2511 	ni.dev = (struct net_device *)ifp;
2512 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2513 }
2514 
2515 static void
2516 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2517 {
2518 	struct notifier_block *nb;
2519 	struct netdev_notifier_info ni;
2520 
2521 	nb = arg;
2522 	ni.ifp = ifp;
2523 	ni.dev = (struct net_device *)ifp;
2524 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2525 }
2526 
2527 int
2528 register_netdevice_notifier(struct notifier_block *nb)
2529 {
2530 
2531 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2532 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2533 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2534 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2535 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2536 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2537 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2538 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2539 
2540 	return (0);
2541 }
2542 
2543 int
2544 register_inetaddr_notifier(struct notifier_block *nb)
2545 {
2546 
2547 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2548 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2549 	return (0);
2550 }
2551 
2552 int
2553 unregister_netdevice_notifier(struct notifier_block *nb)
2554 {
2555 
2556 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2557 	    nb->tags[NETDEV_UP]);
2558 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2559 	    nb->tags[NETDEV_REGISTER]);
2560 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2561 	    nb->tags[NETDEV_UNREGISTER]);
2562 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2563 	    nb->tags[NETDEV_CHANGEADDR]);
2564 
2565 	return (0);
2566 }
2567 
2568 int
2569 unregister_inetaddr_notifier(struct notifier_block *nb)
2570 {
2571 
2572 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2573 	    nb->tags[NETDEV_CHANGEIFADDR]);
2574 
2575 	return (0);
2576 }
2577 
2578 struct list_sort_thunk {
2579 	int (*cmp)(void *, struct list_head *, struct list_head *);
2580 	void *priv;
2581 };
2582 
2583 static inline int
2584 linux_le_cmp(const void *d1, const void *d2, void *priv)
2585 {
2586 	struct list_head *le1, *le2;
2587 	struct list_sort_thunk *thunk;
2588 
2589 	thunk = priv;
2590 	le1 = *(__DECONST(struct list_head **, d1));
2591 	le2 = *(__DECONST(struct list_head **, d2));
2592 	return ((thunk->cmp)(thunk->priv, le1, le2));
2593 }
2594 
2595 void
2596 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2597     struct list_head *a, struct list_head *b))
2598 {
2599 	struct list_sort_thunk thunk;
2600 	struct list_head **ar, *le;
2601 	size_t count, i;
2602 
2603 	count = 0;
2604 	list_for_each(le, head)
2605 		count++;
2606 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2607 	i = 0;
2608 	list_for_each(le, head)
2609 		ar[i++] = le;
2610 	thunk.cmp = cmp;
2611 	thunk.priv = priv;
2612 	qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk);
2613 	INIT_LIST_HEAD(head);
2614 	for (i = 0; i < count; i++)
2615 		list_add_tail(ar[i], head);
2616 	free(ar, M_KMALLOC);
2617 }
2618 
2619 #if defined(__i386__) || defined(__amd64__)
2620 int
2621 linux_wbinvd_on_all_cpus(void)
2622 {
2623 
2624 	pmap_invalidate_cache();
2625 	return (0);
2626 }
2627 #endif
2628 
2629 int
2630 linux_on_each_cpu(void callback(void *), void *data)
2631 {
2632 
2633 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2634 	    smp_no_rendezvous_barrier, data);
2635 	return (0);
2636 }
2637 
2638 int
2639 linux_in_atomic(void)
2640 {
2641 
2642 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2643 }
2644 
2645 struct linux_cdev *
2646 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2647 {
2648 	dev_t dev = MKDEV(major, minor);
2649 	struct cdev *cdev;
2650 
2651 	dev_lock();
2652 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2653 		struct linux_cdev *ldev = cdev->si_drv1;
2654 		if (ldev->dev == dev &&
2655 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2656 			break;
2657 		}
2658 	}
2659 	dev_unlock();
2660 
2661 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2662 }
2663 
2664 int
2665 __register_chrdev(unsigned int major, unsigned int baseminor,
2666     unsigned int count, const char *name,
2667     const struct file_operations *fops)
2668 {
2669 	struct linux_cdev *cdev;
2670 	int ret = 0;
2671 	int i;
2672 
2673 	for (i = baseminor; i < baseminor + count; i++) {
2674 		cdev = cdev_alloc();
2675 		cdev->ops = fops;
2676 		kobject_set_name(&cdev->kobj, name);
2677 
2678 		ret = cdev_add(cdev, makedev(major, i), 1);
2679 		if (ret != 0)
2680 			break;
2681 	}
2682 	return (ret);
2683 }
2684 
2685 int
2686 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2687     unsigned int count, const char *name,
2688     const struct file_operations *fops, uid_t uid,
2689     gid_t gid, int mode)
2690 {
2691 	struct linux_cdev *cdev;
2692 	int ret = 0;
2693 	int i;
2694 
2695 	for (i = baseminor; i < baseminor + count; i++) {
2696 		cdev = cdev_alloc();
2697 		cdev->ops = fops;
2698 		kobject_set_name(&cdev->kobj, name);
2699 
2700 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2701 		if (ret != 0)
2702 			break;
2703 	}
2704 	return (ret);
2705 }
2706 
2707 void
2708 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2709     unsigned int count, const char *name)
2710 {
2711 	struct linux_cdev *cdevp;
2712 	int i;
2713 
2714 	for (i = baseminor; i < baseminor + count; i++) {
2715 		cdevp = linux_find_cdev(name, major, i);
2716 		if (cdevp != NULL)
2717 			cdev_del(cdevp);
2718 	}
2719 }
2720 
2721 void
2722 linux_dump_stack(void)
2723 {
2724 #ifdef STACK
2725 	struct stack st;
2726 
2727 	stack_save(&st);
2728 	stack_print(&st);
2729 #endif
2730 }
2731 
2732 int
2733 linuxkpi_net_ratelimit(void)
2734 {
2735 
2736 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2737 	   lkpi_net_maxpps));
2738 }
2739 
2740 struct io_mapping *
2741 io_mapping_create_wc(resource_size_t base, unsigned long size)
2742 {
2743 	struct io_mapping *mapping;
2744 
2745 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2746 	if (mapping == NULL)
2747 		return (NULL);
2748 	return (io_mapping_init_wc(mapping, base, size));
2749 }
2750 
2751 /* We likely want a linuxkpi_device.c at some point. */
2752 bool
2753 device_can_wakeup(struct device *dev)
2754 {
2755 
2756 	if (dev == NULL)
2757 		return (false);
2758 	/*
2759 	 * XXX-BZ iwlwifi queries it as part of enabling WoWLAN.
2760 	 * Normally this would be based on a bool in dev->power.XXX.
2761 	 * Check such as PCI PCIM_PCAP_*PME.  We have no way to enable this yet.
2762 	 * We may get away by directly calling into bsddev for as long as
2763 	 * we can assume PCI only avoiding changing struct device breaking KBI.
2764 	 */
2765 	pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__);
2766 	return (false);
2767 }
2768 
2769 static void
2770 devm_device_group_remove(struct device *dev, void *p)
2771 {
2772 	const struct attribute_group **dr = p;
2773 	const struct attribute_group *group = *dr;
2774 
2775 	sysfs_remove_group(&dev->kobj, group);
2776 }
2777 
2778 int
2779 lkpi_devm_device_add_group(struct device *dev,
2780     const struct attribute_group *group)
2781 {
2782 	const struct attribute_group **dr;
2783 	int ret;
2784 
2785 	dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL);
2786 	if (dr == NULL)
2787 		return (-ENOMEM);
2788 
2789 	ret = sysfs_create_group(&dev->kobj, group);
2790 	if (ret == 0) {
2791 		*dr = group;
2792 		devres_add(dev, dr);
2793 	} else
2794 		devres_free(dr);
2795 
2796 	return (ret);
2797 }
2798 
2799 #if defined(__i386__) || defined(__amd64__)
2800 bool linux_cpu_has_clflush;
2801 struct cpuinfo_x86 boot_cpu_data;
2802 struct cpuinfo_x86 *__cpu_data;
2803 #endif
2804 
2805 cpumask_t *
2806 lkpi_get_static_single_cpu_mask(int cpuid)
2807 {
2808 
2809 	KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n",
2810 	    __func__, cpuid));
2811 	KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n",
2812 	    __func__, cpuid));
2813 
2814 	return (static_single_cpu_mask[cpuid]);
2815 }
2816 
2817 bool
2818 lkpi_xen_initial_domain(void)
2819 {
2820 #ifdef XENHVM
2821 	return (xen_initial_domain());
2822 #else
2823 	return (false);
2824 #endif
2825 }
2826 
2827 bool
2828 lkpi_xen_pv_domain(void)
2829 {
2830 #ifdef XENHVM
2831 	return (xen_pv_domain());
2832 #else
2833 	return (false);
2834 #endif
2835 }
2836 
2837 static void
2838 linux_compat_init(void *arg)
2839 {
2840 	struct sysctl_oid *rootoid;
2841 	int i;
2842 
2843 #if defined(__i386__) || defined(__amd64__)
2844 	static const uint32_t x86_vendors[X86_VENDOR_NUM] = {
2845 		[X86_VENDOR_INTEL] = CPU_VENDOR_INTEL,
2846 		[X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX,
2847 		[X86_VENDOR_AMD] = CPU_VENDOR_AMD,
2848 		[X86_VENDOR_UMC] = CPU_VENDOR_UMC,
2849 		[X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR,
2850 		[X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA,
2851 		[X86_VENDOR_NSC] = CPU_VENDOR_NSC,
2852 		[X86_VENDOR_HYGON] = CPU_VENDOR_HYGON,
2853 	};
2854 	uint8_t x86_vendor = X86_VENDOR_UNKNOWN;
2855 
2856 	for (i = 0; i < X86_VENDOR_NUM; i++) {
2857 		if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) {
2858 			x86_vendor = i;
2859 			break;
2860 		}
2861 	}
2862 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2863 	boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2864 	boot_cpu_data.x86_max_cores = mp_ncpus;
2865 	boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id);
2866 	boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id);
2867 	boot_cpu_data.x86_vendor = x86_vendor;
2868 
2869 	__cpu_data = kmalloc_array(mp_maxid + 1,
2870 	    sizeof(*__cpu_data), M_WAITOK | M_ZERO);
2871 	CPU_FOREACH(i) {
2872 		__cpu_data[i].x86_clflush_size = cpu_clflush_line_size;
2873 		__cpu_data[i].x86_max_cores = mp_ncpus;
2874 		__cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id);
2875 		__cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id);
2876 		__cpu_data[i].x86_vendor = x86_vendor;
2877 	}
2878 #endif
2879 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2880 
2881 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2882 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2883 	kobject_init(&linux_class_root, &linux_class_ktype);
2884 	kobject_set_name(&linux_class_root, "class");
2885 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2886 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2887 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2888 	kobject_set_name(&linux_root_device.kobj, "device");
2889 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2890 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2891 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2892 	linux_root_device.bsddev = root_bus;
2893 	linux_class_misc.name = "misc";
2894 	class_register(&linux_class_misc);
2895 	INIT_LIST_HEAD(&pci_drivers);
2896 	INIT_LIST_HEAD(&pci_devices);
2897 	spin_lock_init(&pci_lock);
2898 	init_waitqueue_head(&linux_bit_waitq);
2899 	init_waitqueue_head(&linux_var_waitq);
2900 
2901 	CPU_COPY(&all_cpus, &cpu_online_mask);
2902 	/*
2903 	 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system.
2904 	 * CPUs are indexed from 0..(mp_maxid).  The entry for cpuid 0 will only
2905 	 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on.
2906 	 * This is used by cpumask_of() (and possibly others in the future) for,
2907 	 * e.g., drivers to pass hints to irq_set_affinity_hint().
2908 	 */
2909 	static_single_cpu_mask = kmalloc_array(mp_maxid + 1,
2910 	    sizeof(static_single_cpu_mask), M_WAITOK | M_ZERO);
2911 
2912 	/*
2913 	 * When the number of CPUs reach a threshold, we start to save memory
2914 	 * given the sets are static by overlapping those having their single
2915 	 * bit set at same position in a bitset word.  Asymptotically, this
2916 	 * regular scheme is in O(n²) whereas the overlapping one is in O(n)
2917 	 * only with n being the maximum number of CPUs, so the gain will become
2918 	 * huge quite quickly.  The threshold for 64-bit architectures is 128
2919 	 * CPUs.
2920 	 */
2921 	if (mp_ncpus < (2 * _BITSET_BITS)) {
2922 		cpumask_t *sscm_ptr;
2923 
2924 		/*
2925 		 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) *
2926 		 * (_BITSET_BITS / 8)' bytes (for comparison with the
2927 		 * overlapping scheme).
2928 		 */
2929 		static_single_cpu_mask_lcs = kmalloc_array(mp_ncpus,
2930 		    sizeof(*static_single_cpu_mask_lcs),
2931 		    M_WAITOK | M_ZERO);
2932 
2933 		sscm_ptr = static_single_cpu_mask_lcs;
2934 		CPU_FOREACH(i) {
2935 			static_single_cpu_mask[i] = sscm_ptr++;
2936 			CPU_SET(i, static_single_cpu_mask[i]);
2937 		}
2938 	} else {
2939 		/* Pointer to a bitset word. */
2940 		__typeof(((cpuset_t *)NULL)->__bits[0]) *bwp;
2941 
2942 		/*
2943 		 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t'
2944 		 * really) with a single bit set that can be reused for all
2945 		 * single CPU masks by making them start at different offsets.
2946 		 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before
2947 		 * the word having its single bit set, and the same amount
2948 		 * after.
2949 		 */
2950 		static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS,
2951 		    (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8),
2952 		    M_KMALLOC, M_WAITOK | M_ZERO);
2953 
2954 		/*
2955 		 * We rely below on cpuset_t and the bitset generic
2956 		 * implementation assigning words in the '__bits' array in the
2957 		 * same order of bits (i.e., little-endian ordering, not to be
2958 		 * confused with machine endianness, which concerns bits in
2959 		 * words and other integers).  This is an imperfect test, but it
2960 		 * will detect a change to big-endian ordering.
2961 		 */
2962 		_Static_assert(
2963 		    __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1,
2964 		    "Assumes a bitset implementation that is little-endian "
2965 		    "on its words");
2966 
2967 		/* Initialize the single bit of each static span. */
2968 		bwp = (__typeof(bwp))static_single_cpu_mask_lcs +
2969 		    (__bitset_words(CPU_SETSIZE) - 1);
2970 		for (i = 0; i < _BITSET_BITS; i++) {
2971 			CPU_SET(i, (cpuset_t *)bwp);
2972 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1);
2973 		}
2974 
2975 		/*
2976 		 * Finally set all CPU masks to the proper word in their
2977 		 * relevant span.
2978 		 */
2979 		CPU_FOREACH(i) {
2980 			bwp = (__typeof(bwp))static_single_cpu_mask_lcs;
2981 			/* Find the non-zero word of the relevant span. */
2982 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) *
2983 			    (i % _BITSET_BITS) +
2984 			    __bitset_words(CPU_SETSIZE) - 1;
2985 			/* Shift to find the CPU mask start. */
2986 			bwp -= (i / _BITSET_BITS);
2987 			static_single_cpu_mask[i] = (cpuset_t *)bwp;
2988 		}
2989 	}
2990 
2991 	strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release));
2992 }
2993 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2994 
2995 static void
2996 linux_compat_uninit(void *arg)
2997 {
2998 	linux_kobject_kfree_name(&linux_class_root);
2999 	linux_kobject_kfree_name(&linux_root_device.kobj);
3000 	linux_kobject_kfree_name(&linux_class_misc.kobj);
3001 
3002 	free(static_single_cpu_mask_lcs, M_KMALLOC);
3003 	free(static_single_cpu_mask, M_KMALLOC);
3004 #if defined(__i386__) || defined(__amd64__)
3005 	free(__cpu_data, M_KMALLOC);
3006 #endif
3007 
3008 	spin_lock_destroy(&pci_lock);
3009 	rw_destroy(&linux_vma_lock);
3010 }
3011 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
3012 
3013 /*
3014  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
3015  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
3016  * used. Assert these types have the same size, else some parts of the
3017  * LinuxKPI may not work like expected:
3018  */
3019 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
3020