xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 5945da0bc9ac42f531b1079a246eb8ce4f0d63db)
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 enum system_states system_state = SYSTEM_RUNNING;
171 
172 int
173 panic_cmp(struct rb_node *one, struct rb_node *two)
174 {
175 	panic("no cmp");
176 }
177 
178 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
179 
180 #define	START(node)	((node)->start)
181 #define	LAST(node)	((node)->last)
182 
183 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
184     LAST,, lkpi_interval_tree)
185 
186 static void
187 linux_device_release(struct device *dev)
188 {
189 	pr_debug("linux_device_release: %s\n", dev_name(dev));
190 	kfree(dev);
191 }
192 
193 static ssize_t
194 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
195 {
196 	struct class_attribute *dattr;
197 	ssize_t error;
198 
199 	dattr = container_of(attr, struct class_attribute, attr);
200 	error = -EIO;
201 	if (dattr->show)
202 		error = dattr->show(container_of(kobj, struct class, kobj),
203 		    dattr, buf);
204 	return (error);
205 }
206 
207 static ssize_t
208 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
209     size_t count)
210 {
211 	struct class_attribute *dattr;
212 	ssize_t error;
213 
214 	dattr = container_of(attr, struct class_attribute, attr);
215 	error = -EIO;
216 	if (dattr->store)
217 		error = dattr->store(container_of(kobj, struct class, kobj),
218 		    dattr, buf, count);
219 	return (error);
220 }
221 
222 static void
223 linux_class_release(struct kobject *kobj)
224 {
225 	struct class *class;
226 
227 	class = container_of(kobj, struct class, kobj);
228 	if (class->class_release)
229 		class->class_release(class);
230 }
231 
232 static const struct sysfs_ops linux_class_sysfs = {
233 	.show  = linux_class_show,
234 	.store = linux_class_store,
235 };
236 
237 const struct kobj_type linux_class_ktype = {
238 	.release = linux_class_release,
239 	.sysfs_ops = &linux_class_sysfs
240 };
241 
242 static void
243 linux_dev_release(struct kobject *kobj)
244 {
245 	struct device *dev;
246 
247 	dev = container_of(kobj, struct device, kobj);
248 	/* This is the precedence defined by linux. */
249 	if (dev->release)
250 		dev->release(dev);
251 	else if (dev->class && dev->class->dev_release)
252 		dev->class->dev_release(dev);
253 }
254 
255 static ssize_t
256 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
257 {
258 	struct device_attribute *dattr;
259 	ssize_t error;
260 
261 	dattr = container_of(attr, struct device_attribute, attr);
262 	error = -EIO;
263 	if (dattr->show)
264 		error = dattr->show(container_of(kobj, struct device, kobj),
265 		    dattr, buf);
266 	return (error);
267 }
268 
269 static ssize_t
270 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
271     size_t count)
272 {
273 	struct device_attribute *dattr;
274 	ssize_t error;
275 
276 	dattr = container_of(attr, struct device_attribute, attr);
277 	error = -EIO;
278 	if (dattr->store)
279 		error = dattr->store(container_of(kobj, struct device, kobj),
280 		    dattr, buf, count);
281 	return (error);
282 }
283 
284 static const struct sysfs_ops linux_dev_sysfs = {
285 	.show  = linux_dev_show,
286 	.store = linux_dev_store,
287 };
288 
289 const struct kobj_type linux_dev_ktype = {
290 	.release = linux_dev_release,
291 	.sysfs_ops = &linux_dev_sysfs
292 };
293 
294 struct device *
295 device_create(struct class *class, struct device *parent, dev_t devt,
296     void *drvdata, const char *fmt, ...)
297 {
298 	struct device *dev;
299 	va_list args;
300 
301 	dev = kzalloc(sizeof(*dev), M_WAITOK);
302 	dev->parent = parent;
303 	dev->class = class;
304 	dev->devt = devt;
305 	dev->driver_data = drvdata;
306 	dev->release = linux_device_release;
307 	va_start(args, fmt);
308 	kobject_set_name_vargs(&dev->kobj, fmt, args);
309 	va_end(args);
310 	device_register(dev);
311 
312 	return (dev);
313 }
314 
315 struct device *
316 device_create_groups_vargs(struct class *class, struct device *parent,
317     dev_t devt, void *drvdata, const struct attribute_group **groups,
318     const char *fmt, va_list args)
319 {
320 	struct device *dev = NULL;
321 	int retval = -ENODEV;
322 
323 	if (class == NULL || IS_ERR(class))
324 		goto error;
325 
326 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
327 	if (!dev) {
328 		retval = -ENOMEM;
329 		goto error;
330 	}
331 
332 	dev->devt = devt;
333 	dev->class = class;
334 	dev->parent = parent;
335 	dev->groups = groups;
336 	dev->release = device_create_release;
337 	/* device_initialize() needs the class and parent to be set */
338 	device_initialize(dev);
339 	dev_set_drvdata(dev, drvdata);
340 
341 	retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
342 	if (retval)
343 		goto error;
344 
345 	retval = device_add(dev);
346 	if (retval)
347 		goto error;
348 
349 	return dev;
350 
351 error:
352 	put_device(dev);
353 	return ERR_PTR(retval);
354 }
355 
356 struct class *
357 lkpi_class_create(const char *name)
358 {
359 	struct class *class;
360 	int error;
361 
362 	class = kzalloc(sizeof(*class), M_WAITOK);
363 	class->name = name;
364 	class->class_release = linux_class_kfree;
365 	error = class_register(class);
366 	if (error) {
367 		kfree(class);
368 		return (NULL);
369 	}
370 
371 	return (class);
372 }
373 
374 static void
375 linux_kq_lock(void *arg)
376 {
377 	spinlock_t *s = arg;
378 
379 	spin_lock(s);
380 }
381 static void
382 linux_kq_unlock(void *arg)
383 {
384 	spinlock_t *s = arg;
385 
386 	spin_unlock(s);
387 }
388 
389 static void
390 linux_kq_assert_lock(void *arg, int what)
391 {
392 #ifdef INVARIANTS
393 	spinlock_t *s = arg;
394 
395 	if (what == LA_LOCKED)
396 		mtx_assert(s, MA_OWNED);
397 	else
398 		mtx_assert(s, MA_NOTOWNED);
399 #endif
400 }
401 
402 static void
403 linux_file_kqfilter_poll(struct linux_file *, int);
404 
405 struct linux_file *
406 linux_file_alloc(void)
407 {
408 	struct linux_file *filp;
409 
410 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
411 
412 	/* set initial refcount */
413 	filp->f_count = 1;
414 
415 	/* setup fields needed by kqueue support */
416 	spin_lock_init(&filp->f_kqlock);
417 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
418 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
419 
420 	return (filp);
421 }
422 
423 void
424 linux_file_free(struct linux_file *filp)
425 {
426 	if (filp->_file == NULL) {
427 		if (filp->f_op != NULL && filp->f_op->release != NULL)
428 			filp->f_op->release(filp->f_vnode, filp);
429 		if (filp->f_shmem != NULL)
430 			vm_object_deallocate(filp->f_shmem);
431 		kfree_rcu(filp, rcu);
432 	} else {
433 		/*
434 		 * The close method of the character device or file
435 		 * will free the linux_file structure:
436 		 */
437 		_fdrop(filp->_file, curthread);
438 	}
439 }
440 
441 struct linux_cdev *
442 cdev_alloc(void)
443 {
444 	struct linux_cdev *cdev;
445 
446 	cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
447 	kobject_init(&cdev->kobj, &linux_cdev_ktype);
448 	cdev->refs = 1;
449 	return (cdev);
450 }
451 
452 static int
453 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
454     vm_page_t *mres)
455 {
456 	struct vm_area_struct *vmap;
457 
458 	vmap = linux_cdev_handle_find(vm_obj->handle);
459 
460 	MPASS(vmap != NULL);
461 	MPASS(vmap->vm_private_data == vm_obj->handle);
462 
463 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
464 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
465 		vm_page_t page;
466 
467 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
468 			/*
469 			 * If the passed in result page is a fake
470 			 * page, update it with the new physical
471 			 * address.
472 			 */
473 			page = *mres;
474 			vm_page_updatefake(page, paddr, vm_obj->memattr);
475 		} else {
476 			/*
477 			 * Replace the passed in "mres" page with our
478 			 * own fake page and free up the all of the
479 			 * original pages.
480 			 */
481 			VM_OBJECT_WUNLOCK(vm_obj);
482 			page = vm_page_getfake(paddr, vm_obj->memattr);
483 			VM_OBJECT_WLOCK(vm_obj);
484 
485 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
486 			*mres = page;
487 		}
488 		vm_page_valid(page);
489 		return (VM_PAGER_OK);
490 	}
491 	return (VM_PAGER_FAIL);
492 }
493 
494 static int
495 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
496     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
497 {
498 	struct vm_area_struct *vmap;
499 	int err;
500 
501 	/* get VM area structure */
502 	vmap = linux_cdev_handle_find(vm_obj->handle);
503 	MPASS(vmap != NULL);
504 	MPASS(vmap->vm_private_data == vm_obj->handle);
505 
506 	VM_OBJECT_WUNLOCK(vm_obj);
507 
508 	linux_set_current(curthread);
509 
510 	down_write(&vmap->vm_mm->mmap_sem);
511 	if (unlikely(vmap->vm_ops == NULL)) {
512 		err = VM_FAULT_SIGBUS;
513 	} else {
514 		struct vm_fault vmf;
515 
516 		/* fill out VM fault structure */
517 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
518 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
519 		vmf.pgoff = 0;
520 		vmf.page = NULL;
521 		vmf.vma = vmap;
522 
523 		vmap->vm_pfn_count = 0;
524 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
525 		vmap->vm_obj = vm_obj;
526 
527 		err = vmap->vm_ops->fault(&vmf);
528 
529 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
530 			kern_yield(PRI_USER);
531 			err = vmap->vm_ops->fault(&vmf);
532 		}
533 	}
534 
535 	/* translate return code */
536 	switch (err) {
537 	case VM_FAULT_OOM:
538 		err = VM_PAGER_AGAIN;
539 		break;
540 	case VM_FAULT_SIGBUS:
541 		err = VM_PAGER_BAD;
542 		break;
543 	case VM_FAULT_NOPAGE:
544 		/*
545 		 * By contract the fault handler will return having
546 		 * busied all the pages itself. If pidx is already
547 		 * found in the object, it will simply xbusy the first
548 		 * page and return with vm_pfn_count set to 1.
549 		 */
550 		*first = vmap->vm_pfn_first;
551 		*last = *first + vmap->vm_pfn_count - 1;
552 		err = VM_PAGER_OK;
553 		break;
554 	default:
555 		err = VM_PAGER_ERROR;
556 		break;
557 	}
558 	up_write(&vmap->vm_mm->mmap_sem);
559 	VM_OBJECT_WLOCK(vm_obj);
560 	return (err);
561 }
562 
563 static struct rwlock linux_vma_lock;
564 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
565     TAILQ_HEAD_INITIALIZER(linux_vma_head);
566 
567 static void
568 linux_cdev_handle_free(struct vm_area_struct *vmap)
569 {
570 	/* Drop reference on vm_file */
571 	if (vmap->vm_file != NULL)
572 		fput(vmap->vm_file);
573 
574 	/* Drop reference on mm_struct */
575 	mmput(vmap->vm_mm);
576 
577 	kfree(vmap);
578 }
579 
580 static void
581 linux_cdev_handle_remove(struct vm_area_struct *vmap)
582 {
583 	rw_wlock(&linux_vma_lock);
584 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
585 	rw_wunlock(&linux_vma_lock);
586 }
587 
588 static struct vm_area_struct *
589 linux_cdev_handle_find(void *handle)
590 {
591 	struct vm_area_struct *vmap;
592 
593 	rw_rlock(&linux_vma_lock);
594 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
595 		if (vmap->vm_private_data == handle)
596 			break;
597 	}
598 	rw_runlock(&linux_vma_lock);
599 	return (vmap);
600 }
601 
602 static int
603 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
604 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
605 {
606 
607 	MPASS(linux_cdev_handle_find(handle) != NULL);
608 	*color = 0;
609 	return (0);
610 }
611 
612 static void
613 linux_cdev_pager_dtor(void *handle)
614 {
615 	const struct vm_operations_struct *vm_ops;
616 	struct vm_area_struct *vmap;
617 
618 	vmap = linux_cdev_handle_find(handle);
619 	MPASS(vmap != NULL);
620 
621 	/*
622 	 * Remove handle before calling close operation to prevent
623 	 * other threads from reusing the handle pointer.
624 	 */
625 	linux_cdev_handle_remove(vmap);
626 
627 	down_write(&vmap->vm_mm->mmap_sem);
628 	vm_ops = vmap->vm_ops;
629 	if (likely(vm_ops != NULL))
630 		vm_ops->close(vmap);
631 	up_write(&vmap->vm_mm->mmap_sem);
632 
633 	linux_cdev_handle_free(vmap);
634 }
635 
636 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
637   {
638 	/* OBJT_MGTDEVICE */
639 	.cdev_pg_populate	= linux_cdev_pager_populate,
640 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
641 	.cdev_pg_dtor	= linux_cdev_pager_dtor
642   },
643   {
644 	/* OBJT_DEVICE */
645 	.cdev_pg_fault	= linux_cdev_pager_fault,
646 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
647 	.cdev_pg_dtor	= linux_cdev_pager_dtor
648   },
649 };
650 
651 int
652 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
653     unsigned long size)
654 {
655 	struct pctrie_iter pages;
656 	vm_object_t obj;
657 	vm_page_t m;
658 
659 	obj = vma->vm_obj;
660 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
661 		return (-ENOTSUP);
662 	VM_OBJECT_RLOCK(obj);
663 	vm_page_iter_limit_init(&pages, obj, OFF_TO_IDX(address + size));
664 	VM_RADIX_FOREACH_FROM(m, &pages, OFF_TO_IDX(address))
665 		pmap_remove_all(m);
666 	VM_OBJECT_RUNLOCK(obj);
667 	return (0);
668 }
669 
670 void
671 vma_set_file(struct vm_area_struct *vma, struct linux_file *file)
672 {
673 	struct linux_file *tmp;
674 
675 	/* Changing an anonymous vma with this is illegal */
676 	get_file(file);
677 	tmp = vma->vm_file;
678 	vma->vm_file = file;
679 	fput(tmp);
680 }
681 
682 static struct file_operations dummy_ldev_ops = {
683 	/* XXXKIB */
684 };
685 
686 static struct linux_cdev dummy_ldev = {
687 	.ops = &dummy_ldev_ops,
688 };
689 
690 #define	LDEV_SI_DTR	0x0001
691 #define	LDEV_SI_REF	0x0002
692 
693 static void
694 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
695     struct linux_cdev **dev)
696 {
697 	struct linux_cdev *ldev;
698 	u_int siref;
699 
700 	ldev = filp->f_cdev;
701 	*fop = filp->f_op;
702 	if (ldev != NULL) {
703 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
704 			refcount_acquire(&ldev->refs);
705 		} else {
706 			for (siref = ldev->siref;;) {
707 				if ((siref & LDEV_SI_DTR) != 0) {
708 					ldev = &dummy_ldev;
709 					*fop = ldev->ops;
710 					siref = ldev->siref;
711 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
712 				} else if (atomic_fcmpset_int(&ldev->siref,
713 				    &siref, siref + LDEV_SI_REF)) {
714 					break;
715 				}
716 			}
717 		}
718 	}
719 	*dev = ldev;
720 }
721 
722 static void
723 linux_drop_fop(struct linux_cdev *ldev)
724 {
725 
726 	if (ldev == NULL)
727 		return;
728 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
729 		linux_cdev_deref(ldev);
730 	} else {
731 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
732 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
733 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
734 	}
735 }
736 
737 #define	OPW(fp,td,code) ({			\
738 	struct file *__fpop;			\
739 	__typeof(code) __retval;		\
740 						\
741 	__fpop = (td)->td_fpop;			\
742 	(td)->td_fpop = (fp);			\
743 	__retval = (code);			\
744 	(td)->td_fpop = __fpop;			\
745 	__retval;				\
746 })
747 
748 static int
749 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
750     struct file *file)
751 {
752 	struct linux_cdev *ldev;
753 	struct linux_file *filp;
754 	const struct file_operations *fop;
755 	int error;
756 
757 	ldev = dev->si_drv1;
758 
759 	filp = linux_file_alloc();
760 	filp->f_dentry = &filp->f_dentry_store;
761 	filp->f_op = ldev->ops;
762 	filp->f_mode = file->f_flag;
763 	filp->f_flags = file->f_flag;
764 	filp->f_vnode = file->f_vnode;
765 	filp->_file = file;
766 	refcount_acquire(&ldev->refs);
767 	filp->f_cdev = ldev;
768 
769 	linux_set_current(td);
770 	linux_get_fop(filp, &fop, &ldev);
771 
772 	if (fop->open != NULL) {
773 		error = -fop->open(file->f_vnode, filp);
774 		if (error != 0) {
775 			linux_drop_fop(ldev);
776 			linux_cdev_deref(filp->f_cdev);
777 			kfree(filp);
778 			return (error);
779 		}
780 	}
781 
782 	/* hold on to the vnode - used for fstat() */
783 	vref(filp->f_vnode);
784 
785 	/* release the file from devfs */
786 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
787 	linux_drop_fop(ldev);
788 	return (ENXIO);
789 }
790 
791 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
792 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
793 
794 static inline int
795 linux_remap_address(void **uaddr, size_t len)
796 {
797 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
798 
799 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
800 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
801 		struct task_struct *pts = current;
802 		if (pts == NULL) {
803 			*uaddr = NULL;
804 			return (1);
805 		}
806 
807 		/* compute data offset */
808 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
809 
810 		/* check that length is within bounds */
811 		if ((len > IOCPARM_MAX) ||
812 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
813 			*uaddr = NULL;
814 			return (1);
815 		}
816 
817 		/* re-add kernel buffer address */
818 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
819 
820 		/* update address location */
821 		*uaddr = (void *)uaddr_val;
822 		return (1);
823 	}
824 	return (0);
825 }
826 
827 int
828 linux_copyin(const void *uaddr, void *kaddr, size_t len)
829 {
830 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
831 		if (uaddr == NULL)
832 			return (-EFAULT);
833 		memcpy(kaddr, uaddr, len);
834 		return (0);
835 	}
836 	return (-copyin(uaddr, kaddr, len));
837 }
838 
839 int
840 linux_copyout(const void *kaddr, void *uaddr, size_t len)
841 {
842 	if (linux_remap_address(&uaddr, len)) {
843 		if (uaddr == NULL)
844 			return (-EFAULT);
845 		memcpy(uaddr, kaddr, len);
846 		return (0);
847 	}
848 	return (-copyout(kaddr, uaddr, len));
849 }
850 
851 size_t
852 linux_clear_user(void *_uaddr, size_t _len)
853 {
854 	uint8_t *uaddr = _uaddr;
855 	size_t len = _len;
856 
857 	/* make sure uaddr is aligned before going into the fast loop */
858 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
859 		if (subyte(uaddr, 0))
860 			return (_len);
861 		uaddr++;
862 		len--;
863 	}
864 
865 	/* zero 8 bytes at a time */
866 	while (len > 7) {
867 #ifdef __LP64__
868 		if (suword64(uaddr, 0))
869 			return (_len);
870 #else
871 		if (suword32(uaddr, 0))
872 			return (_len);
873 		if (suword32(uaddr + 4, 0))
874 			return (_len);
875 #endif
876 		uaddr += 8;
877 		len -= 8;
878 	}
879 
880 	/* zero fill end, if any */
881 	while (len > 0) {
882 		if (subyte(uaddr, 0))
883 			return (_len);
884 		uaddr++;
885 		len--;
886 	}
887 	return (0);
888 }
889 
890 int
891 linux_access_ok(const void *uaddr, size_t len)
892 {
893 	uintptr_t saddr;
894 	uintptr_t eaddr;
895 
896 	/* get start and end address */
897 	saddr = (uintptr_t)uaddr;
898 	eaddr = (uintptr_t)uaddr + len;
899 
900 	/* verify addresses are valid for userspace */
901 	return ((saddr == eaddr) ||
902 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
903 }
904 
905 /*
906  * This function should return either EINTR or ERESTART depending on
907  * the signal type sent to this thread:
908  */
909 static int
910 linux_get_error(struct task_struct *task, int error)
911 {
912 	/* check for signal type interrupt code */
913 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
914 		error = -linux_schedule_get_interrupt_value(task);
915 		if (error == 0)
916 			error = EINTR;
917 	}
918 	return (error);
919 }
920 
921 static int
922 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
923     const struct file_operations *fop, u_long cmd, caddr_t data,
924     struct thread *td)
925 {
926 	struct task_struct *task = current;
927 	unsigned size;
928 	int error;
929 
930 	size = IOCPARM_LEN(cmd);
931 	/* refer to logic in sys_ioctl() */
932 	if (size > 0) {
933 		/*
934 		 * Setup hint for linux_copyin() and linux_copyout().
935 		 *
936 		 * Background: Linux code expects a user-space address
937 		 * while FreeBSD supplies a kernel-space address.
938 		 */
939 		task->bsd_ioctl_data = data;
940 		task->bsd_ioctl_len = size;
941 		data = (void *)LINUX_IOCTL_MIN_PTR;
942 	} else {
943 		/* fetch user-space pointer */
944 		data = *(void **)data;
945 	}
946 #ifdef COMPAT_FREEBSD32
947 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
948 		/* try the compat IOCTL handler first */
949 		if (fop->compat_ioctl != NULL) {
950 			error = -OPW(fp, td, fop->compat_ioctl(filp,
951 			    cmd, (u_long)data));
952 		} else {
953 			error = ENOTTY;
954 		}
955 
956 		/* fallback to the regular IOCTL handler, if any */
957 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
958 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
959 			    cmd, (u_long)data));
960 		}
961 	} else
962 #endif
963 	{
964 		if (fop->unlocked_ioctl != NULL) {
965 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
966 			    cmd, (u_long)data));
967 		} else {
968 			error = ENOTTY;
969 		}
970 	}
971 	if (size > 0) {
972 		task->bsd_ioctl_data = NULL;
973 		task->bsd_ioctl_len = 0;
974 	}
975 
976 	if (error == EWOULDBLOCK) {
977 		/* update kqfilter status, if any */
978 		linux_file_kqfilter_poll(filp,
979 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
980 	} else {
981 		error = linux_get_error(task, error);
982 	}
983 	return (error);
984 }
985 
986 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
987 
988 /*
989  * This function atomically updates the poll wakeup state and returns
990  * the previous state at the time of update.
991  */
992 static uint8_t
993 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
994 {
995 	int c, old;
996 
997 	c = v->counter;
998 
999 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1000 		c = old;
1001 
1002 	return (c);
1003 }
1004 
1005 static int
1006 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1007 {
1008 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1009 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1010 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1011 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1012 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1013 	};
1014 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1015 
1016 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1017 	case LINUX_FWQ_STATE_QUEUED:
1018 		linux_poll_wakeup(filp);
1019 		return (1);
1020 	default:
1021 		return (0);
1022 	}
1023 }
1024 
1025 void
1026 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1027 {
1028 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1029 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1030 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1031 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1032 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1033 	};
1034 
1035 	/* check if we are called inside the select system call */
1036 	if (p == LINUX_POLL_TABLE_NORMAL)
1037 		selrecord(curthread, &filp->f_selinfo);
1038 
1039 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1040 	case LINUX_FWQ_STATE_INIT:
1041 		/* NOTE: file handles can only belong to one wait-queue */
1042 		filp->f_wait_queue.wqh = wqh;
1043 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1044 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1045 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1046 		break;
1047 	default:
1048 		break;
1049 	}
1050 }
1051 
1052 static void
1053 linux_poll_wait_dequeue(struct linux_file *filp)
1054 {
1055 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1056 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1057 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1058 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1059 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1060 	};
1061 
1062 	seldrain(&filp->f_selinfo);
1063 
1064 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1065 	case LINUX_FWQ_STATE_NOT_READY:
1066 	case LINUX_FWQ_STATE_QUEUED:
1067 	case LINUX_FWQ_STATE_READY:
1068 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1069 		break;
1070 	default:
1071 		break;
1072 	}
1073 }
1074 
1075 void
1076 linux_poll_wakeup(struct linux_file *filp)
1077 {
1078 	/* this function should be NULL-safe */
1079 	if (filp == NULL)
1080 		return;
1081 
1082 	selwakeup(&filp->f_selinfo);
1083 
1084 	spin_lock(&filp->f_kqlock);
1085 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1086 	    LINUX_KQ_FLAG_NEED_WRITE;
1087 
1088 	/* make sure the "knote" gets woken up */
1089 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1090 	spin_unlock(&filp->f_kqlock);
1091 }
1092 
1093 static struct linux_file *
1094 __get_file_rcu(struct linux_file **f)
1095 {
1096 	struct linux_file *file1, *file2;
1097 
1098 	file1 = READ_ONCE(*f);
1099 	if (file1 == NULL)
1100 		return (NULL);
1101 
1102 	if (!refcount_acquire_if_not_zero(
1103 	    file1->_file == NULL ? &file1->f_count : &file1->_file->f_count))
1104 		return (ERR_PTR(-EAGAIN));
1105 
1106 	file2 = READ_ONCE(*f);
1107 	if (file2 == file1)
1108 		return (file2);
1109 
1110 	fput(file1);
1111 	return (ERR_PTR(-EAGAIN));
1112 }
1113 
1114 struct linux_file *
1115 linux_get_file_rcu(struct linux_file **f)
1116 {
1117 	struct linux_file *file1;
1118 
1119 	for (;;) {
1120 		file1 = __get_file_rcu(f);
1121 		if (file1 == NULL)
1122 			return (NULL);
1123 
1124 		if (IS_ERR(file1))
1125 			continue;
1126 
1127 		return (file1);
1128 	}
1129 }
1130 
1131 struct linux_file *
1132 get_file_active(struct linux_file **f)
1133 {
1134 	struct linux_file *file1;
1135 
1136 	rcu_read_lock();
1137 	file1 = __get_file_rcu(f);
1138 	rcu_read_unlock();
1139 	if (IS_ERR(file1))
1140 		file1 = NULL;
1141 
1142 	return (file1);
1143 }
1144 
1145 static void
1146 linux_file_kqfilter_detach(struct knote *kn)
1147 {
1148 	struct linux_file *filp = kn->kn_hook;
1149 
1150 	spin_lock(&filp->f_kqlock);
1151 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1152 	spin_unlock(&filp->f_kqlock);
1153 }
1154 
1155 static int
1156 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1157 {
1158 	struct linux_file *filp = kn->kn_hook;
1159 
1160 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1161 
1162 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1163 }
1164 
1165 static int
1166 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1167 {
1168 	struct linux_file *filp = kn->kn_hook;
1169 
1170 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1171 
1172 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1173 }
1174 
1175 static const struct filterops linux_dev_kqfiltops_read = {
1176 	.f_isfd = 1,
1177 	.f_detach = linux_file_kqfilter_detach,
1178 	.f_event = linux_file_kqfilter_read_event,
1179 	.f_copy = knote_triv_copy,
1180 };
1181 
1182 static const struct filterops linux_dev_kqfiltops_write = {
1183 	.f_isfd = 1,
1184 	.f_detach = linux_file_kqfilter_detach,
1185 	.f_event = linux_file_kqfilter_write_event,
1186 	.f_copy = knote_triv_copy,
1187 };
1188 
1189 static void
1190 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1191 {
1192 	struct thread *td;
1193 	const struct file_operations *fop;
1194 	struct linux_cdev *ldev;
1195 	int temp;
1196 
1197 	if ((filp->f_kqflags & kqflags) == 0)
1198 		return;
1199 
1200 	td = curthread;
1201 
1202 	linux_get_fop(filp, &fop, &ldev);
1203 	/* get the latest polling state */
1204 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1205 	linux_drop_fop(ldev);
1206 
1207 	spin_lock(&filp->f_kqlock);
1208 	/* clear kqflags */
1209 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1210 	    LINUX_KQ_FLAG_NEED_WRITE);
1211 	/* update kqflags */
1212 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1213 		if ((temp & POLLIN) != 0)
1214 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1215 		if ((temp & POLLOUT) != 0)
1216 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1217 
1218 		/* make sure the "knote" gets woken up */
1219 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1220 	}
1221 	spin_unlock(&filp->f_kqlock);
1222 }
1223 
1224 static int
1225 linux_file_kqfilter(struct file *file, struct knote *kn)
1226 {
1227 	struct linux_file *filp;
1228 	struct thread *td;
1229 	int error;
1230 
1231 	td = curthread;
1232 	filp = (struct linux_file *)file->f_data;
1233 	filp->f_flags = file->f_flag;
1234 	if (filp->f_op->poll == NULL)
1235 		return (EINVAL);
1236 
1237 	spin_lock(&filp->f_kqlock);
1238 	switch (kn->kn_filter) {
1239 	case EVFILT_READ:
1240 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1241 		kn->kn_fop = &linux_dev_kqfiltops_read;
1242 		kn->kn_hook = filp;
1243 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1244 		error = 0;
1245 		break;
1246 	case EVFILT_WRITE:
1247 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1248 		kn->kn_fop = &linux_dev_kqfiltops_write;
1249 		kn->kn_hook = filp;
1250 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1251 		error = 0;
1252 		break;
1253 	default:
1254 		error = EINVAL;
1255 		break;
1256 	}
1257 	spin_unlock(&filp->f_kqlock);
1258 
1259 	if (error == 0) {
1260 		linux_set_current(td);
1261 
1262 		/* update kqfilter status, if any */
1263 		linux_file_kqfilter_poll(filp,
1264 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1265 	}
1266 	return (error);
1267 }
1268 
1269 static int
1270 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1271     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1272     int nprot, bool is_shared, struct thread *td)
1273 {
1274 	struct task_struct *task;
1275 	struct vm_area_struct *vmap;
1276 	struct mm_struct *mm;
1277 	struct linux_file *filp;
1278 	vm_memattr_t attr;
1279 	int error;
1280 
1281 	filp = (struct linux_file *)fp->f_data;
1282 	filp->f_flags = fp->f_flag;
1283 
1284 	if (fop->mmap == NULL)
1285 		return (EOPNOTSUPP);
1286 
1287 	linux_set_current(td);
1288 
1289 	/*
1290 	 * The same VM object might be shared by multiple processes
1291 	 * and the mm_struct is usually freed when a process exits.
1292 	 *
1293 	 * The atomic reference below makes sure the mm_struct is
1294 	 * available as long as the vmap is in the linux_vma_head.
1295 	 */
1296 	task = current;
1297 	mm = task->mm;
1298 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1299 		return (EINVAL);
1300 
1301 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1302 	vmap->vm_start = 0;
1303 	vmap->vm_end = size;
1304 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1305 	vmap->vm_pfn = 0;
1306 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1307 	if (is_shared)
1308 		vmap->vm_flags |= VM_SHARED;
1309 	vmap->vm_ops = NULL;
1310 	vmap->vm_file = get_file(filp);
1311 	vmap->vm_mm = mm;
1312 
1313 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1314 		error = linux_get_error(task, EINTR);
1315 	} else {
1316 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1317 		error = linux_get_error(task, error);
1318 		up_write(&vmap->vm_mm->mmap_sem);
1319 	}
1320 
1321 	if (error != 0) {
1322 		linux_cdev_handle_free(vmap);
1323 		return (error);
1324 	}
1325 
1326 	attr = pgprot2cachemode(vmap->vm_page_prot);
1327 
1328 	if (vmap->vm_ops != NULL) {
1329 		struct vm_area_struct *ptr;
1330 		void *vm_private_data;
1331 		bool vm_no_fault;
1332 
1333 		if (vmap->vm_ops->open == NULL ||
1334 		    vmap->vm_ops->close == NULL ||
1335 		    vmap->vm_private_data == NULL) {
1336 			/* free allocated VM area struct */
1337 			linux_cdev_handle_free(vmap);
1338 			return (EINVAL);
1339 		}
1340 
1341 		vm_private_data = vmap->vm_private_data;
1342 
1343 		rw_wlock(&linux_vma_lock);
1344 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1345 			if (ptr->vm_private_data == vm_private_data)
1346 				break;
1347 		}
1348 		/* check if there is an existing VM area struct */
1349 		if (ptr != NULL) {
1350 			/* check if the VM area structure is invalid */
1351 			if (ptr->vm_ops == NULL ||
1352 			    ptr->vm_ops->open == NULL ||
1353 			    ptr->vm_ops->close == NULL) {
1354 				error = ESTALE;
1355 				vm_no_fault = 1;
1356 			} else {
1357 				error = EEXIST;
1358 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1359 			}
1360 		} else {
1361 			/* insert VM area structure into list */
1362 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1363 			error = 0;
1364 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1365 		}
1366 		rw_wunlock(&linux_vma_lock);
1367 
1368 		if (error != 0) {
1369 			/* free allocated VM area struct */
1370 			linux_cdev_handle_free(vmap);
1371 			/* check for stale VM area struct */
1372 			if (error != EEXIST)
1373 				return (error);
1374 		}
1375 
1376 		/* check if there is no fault handler */
1377 		if (vm_no_fault) {
1378 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1379 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1380 			    td->td_ucred);
1381 		} else {
1382 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1383 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1384 			    td->td_ucred);
1385 		}
1386 
1387 		/* check if allocating the VM object failed */
1388 		if (*object == NULL) {
1389 			if (error == 0) {
1390 				/* remove VM area struct from list */
1391 				linux_cdev_handle_remove(vmap);
1392 				/* free allocated VM area struct */
1393 				linux_cdev_handle_free(vmap);
1394 			}
1395 			return (EINVAL);
1396 		}
1397 	} else {
1398 		struct sglist *sg;
1399 
1400 		sg = sglist_alloc(1, M_WAITOK);
1401 		sglist_append_phys(sg,
1402 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1403 
1404 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1405 		    nprot, 0, td->td_ucred);
1406 
1407 		linux_cdev_handle_free(vmap);
1408 
1409 		if (*object == NULL) {
1410 			sglist_free(sg);
1411 			return (EINVAL);
1412 		}
1413 	}
1414 
1415 	if (attr != VM_MEMATTR_DEFAULT) {
1416 		VM_OBJECT_WLOCK(*object);
1417 		vm_object_set_memattr(*object, attr);
1418 		VM_OBJECT_WUNLOCK(*object);
1419 	}
1420 	*offset = 0;
1421 	return (0);
1422 }
1423 
1424 struct cdevsw linuxcdevsw = {
1425 	.d_version = D_VERSION,
1426 	.d_fdopen = linux_dev_fdopen,
1427 	.d_name = "lkpidev",
1428 };
1429 
1430 static int
1431 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1432     int flags, struct thread *td)
1433 {
1434 	struct linux_file *filp;
1435 	const struct file_operations *fop;
1436 	struct linux_cdev *ldev;
1437 	ssize_t bytes;
1438 	int error;
1439 
1440 	error = 0;
1441 	filp = (struct linux_file *)file->f_data;
1442 	filp->f_flags = file->f_flag;
1443 	/* XXX no support for I/O vectors currently */
1444 	if (uio->uio_iovcnt != 1)
1445 		return (EOPNOTSUPP);
1446 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1447 		return (EINVAL);
1448 	linux_set_current(td);
1449 	linux_get_fop(filp, &fop, &ldev);
1450 	if (fop->read != NULL) {
1451 		bytes = OPW(file, td, fop->read(filp,
1452 		    uio->uio_iov->iov_base,
1453 		    uio->uio_iov->iov_len, &uio->uio_offset));
1454 		if (bytes >= 0) {
1455 			uio->uio_iov->iov_base =
1456 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1457 			uio->uio_iov->iov_len -= bytes;
1458 			uio->uio_resid -= bytes;
1459 		} else {
1460 			error = linux_get_error(current, -bytes);
1461 		}
1462 	} else
1463 		error = ENXIO;
1464 
1465 	/* update kqfilter status, if any */
1466 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1467 	linux_drop_fop(ldev);
1468 
1469 	return (error);
1470 }
1471 
1472 static int
1473 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1474     int flags, struct thread *td)
1475 {
1476 	struct linux_file *filp;
1477 	const struct file_operations *fop;
1478 	struct linux_cdev *ldev;
1479 	ssize_t bytes;
1480 	int error;
1481 
1482 	filp = (struct linux_file *)file->f_data;
1483 	filp->f_flags = file->f_flag;
1484 	/* XXX no support for I/O vectors currently */
1485 	if (uio->uio_iovcnt != 1)
1486 		return (EOPNOTSUPP);
1487 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1488 		return (EINVAL);
1489 	linux_set_current(td);
1490 	linux_get_fop(filp, &fop, &ldev);
1491 	if (fop->write != NULL) {
1492 		bytes = OPW(file, td, fop->write(filp,
1493 		    uio->uio_iov->iov_base,
1494 		    uio->uio_iov->iov_len, &uio->uio_offset));
1495 		if (bytes >= 0) {
1496 			uio->uio_iov->iov_base =
1497 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1498 			uio->uio_iov->iov_len -= bytes;
1499 			uio->uio_resid -= bytes;
1500 			error = 0;
1501 		} else {
1502 			error = linux_get_error(current, -bytes);
1503 		}
1504 	} else
1505 		error = ENXIO;
1506 
1507 	/* update kqfilter status, if any */
1508 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1509 
1510 	linux_drop_fop(ldev);
1511 
1512 	return (error);
1513 }
1514 
1515 static int
1516 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1517     struct thread *td)
1518 {
1519 	struct linux_file *filp;
1520 	const struct file_operations *fop;
1521 	struct linux_cdev *ldev;
1522 	int revents;
1523 
1524 	filp = (struct linux_file *)file->f_data;
1525 	filp->f_flags = file->f_flag;
1526 	linux_set_current(td);
1527 	linux_get_fop(filp, &fop, &ldev);
1528 	if (fop->poll != NULL) {
1529 		revents = OPW(file, td, fop->poll(filp,
1530 		    LINUX_POLL_TABLE_NORMAL)) & events;
1531 	} else {
1532 		revents = 0;
1533 	}
1534 	linux_drop_fop(ldev);
1535 	return (revents);
1536 }
1537 
1538 static int
1539 linux_file_close(struct file *file, struct thread *td)
1540 {
1541 	struct linux_file *filp;
1542 	int (*release)(struct inode *, struct linux_file *);
1543 	const struct file_operations *fop;
1544 	struct linux_cdev *ldev;
1545 	int error;
1546 
1547 	filp = (struct linux_file *)file->f_data;
1548 
1549 	KASSERT(file_count(filp) == 0,
1550 	    ("File refcount(%d) is not zero", file_count(filp)));
1551 
1552 	if (td == NULL)
1553 		td = curthread;
1554 
1555 	error = 0;
1556 	filp->f_flags = file->f_flag;
1557 	linux_set_current(td);
1558 	linux_poll_wait_dequeue(filp);
1559 	linux_get_fop(filp, &fop, &ldev);
1560 	/*
1561 	 * Always use the real release function, if any, to avoid
1562 	 * leaking device resources:
1563 	 */
1564 	release = filp->f_op->release;
1565 	if (release != NULL)
1566 		error = -OPW(file, td, release(filp->f_vnode, filp));
1567 	funsetown(&filp->f_sigio);
1568 	if (filp->f_vnode != NULL)
1569 		vrele(filp->f_vnode);
1570 	linux_drop_fop(ldev);
1571 	ldev = filp->f_cdev;
1572 	if (ldev != NULL)
1573 		linux_cdev_deref(ldev);
1574 	linux_synchronize_rcu(RCU_TYPE_REGULAR);
1575 	kfree(filp);
1576 
1577 	return (error);
1578 }
1579 
1580 static int
1581 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1582     struct thread *td)
1583 {
1584 	struct linux_file *filp;
1585 	const struct file_operations *fop;
1586 	struct linux_cdev *ldev;
1587 	struct fiodgname_arg *fgn;
1588 	const char *p;
1589 	int error, i;
1590 
1591 	error = 0;
1592 	filp = (struct linux_file *)fp->f_data;
1593 	filp->f_flags = fp->f_flag;
1594 	linux_get_fop(filp, &fop, &ldev);
1595 
1596 	linux_set_current(td);
1597 	switch (cmd) {
1598 	case FIONBIO:
1599 		break;
1600 	case FIOASYNC:
1601 		if (fop->fasync == NULL)
1602 			break;
1603 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1604 		break;
1605 	case FIOSETOWN:
1606 		error = fsetown(*(int *)data, &filp->f_sigio);
1607 		if (error == 0) {
1608 			if (fop->fasync == NULL)
1609 				break;
1610 			error = -OPW(fp, td, fop->fasync(0, filp,
1611 			    fp->f_flag & FASYNC));
1612 		}
1613 		break;
1614 	case FIOGETOWN:
1615 		*(int *)data = fgetown(&filp->f_sigio);
1616 		break;
1617 	case FIODGNAME:
1618 #ifdef	COMPAT_FREEBSD32
1619 	case FIODGNAME_32:
1620 #endif
1621 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1622 			error = ENXIO;
1623 			break;
1624 		}
1625 		fgn = data;
1626 		p = devtoname(filp->f_cdev->cdev);
1627 		i = strlen(p) + 1;
1628 		if (i > fgn->len) {
1629 			error = EINVAL;
1630 			break;
1631 		}
1632 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1633 		break;
1634 	default:
1635 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1636 		break;
1637 	}
1638 	linux_drop_fop(ldev);
1639 	return (error);
1640 }
1641 
1642 static int
1643 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1644     vm_prot_t maxprot, int flags, struct file *fp,
1645     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1646 {
1647 	/*
1648 	 * Character devices do not provide private mappings
1649 	 * of any kind:
1650 	 */
1651 	if ((maxprot & VM_PROT_WRITE) == 0 &&
1652 	    (prot & VM_PROT_WRITE) != 0)
1653 		return (EACCES);
1654 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1655 		return (EINVAL);
1656 
1657 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1658 	    (int)prot, (flags & MAP_SHARED) ? true : false, td));
1659 }
1660 
1661 static int
1662 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1663     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1664     struct thread *td)
1665 {
1666 	struct linux_file *filp;
1667 	const struct file_operations *fop;
1668 	struct linux_cdev *ldev;
1669 	struct mount *mp;
1670 	struct vnode *vp;
1671 	vm_object_t object;
1672 	vm_prot_t maxprot;
1673 	int error;
1674 
1675 	filp = (struct linux_file *)fp->f_data;
1676 
1677 	vp = filp->f_vnode;
1678 	if (vp == NULL)
1679 		return (EOPNOTSUPP);
1680 
1681 	/*
1682 	 * Ensure that file and memory protections are
1683 	 * compatible.
1684 	 */
1685 	mp = vp->v_mount;
1686 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1687 		maxprot = VM_PROT_NONE;
1688 		if ((prot & VM_PROT_EXECUTE) != 0)
1689 			return (EACCES);
1690 	} else
1691 		maxprot = VM_PROT_EXECUTE;
1692 	if ((fp->f_flag & FREAD) != 0)
1693 		maxprot |= VM_PROT_READ;
1694 	else if ((prot & VM_PROT_READ) != 0)
1695 		return (EACCES);
1696 
1697 	/*
1698 	 * If we are sharing potential changes via MAP_SHARED and we
1699 	 * are trying to get write permission although we opened it
1700 	 * without asking for it, bail out.
1701 	 *
1702 	 * Note that most character devices always share mappings.
1703 	 *
1704 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1705 	 * requests rather than doing it here.
1706 	 */
1707 	if ((flags & MAP_SHARED) != 0) {
1708 		if ((fp->f_flag & FWRITE) != 0)
1709 			maxprot |= VM_PROT_WRITE;
1710 		else if ((prot & VM_PROT_WRITE) != 0)
1711 			return (EACCES);
1712 	}
1713 	maxprot &= cap_maxprot;
1714 
1715 	linux_get_fop(filp, &fop, &ldev);
1716 	error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1717 	    &foff, fop, &object);
1718 	if (error != 0)
1719 		goto out;
1720 
1721 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1722 	    foff, FALSE, td);
1723 	if (error != 0)
1724 		vm_object_deallocate(object);
1725 out:
1726 	linux_drop_fop(ldev);
1727 	return (error);
1728 }
1729 
1730 static int
1731 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1732 {
1733 	struct linux_file *filp;
1734 	struct vnode *vp;
1735 	int error;
1736 
1737 	filp = (struct linux_file *)fp->f_data;
1738 	if (filp->f_vnode == NULL)
1739 		return (EOPNOTSUPP);
1740 
1741 	vp = filp->f_vnode;
1742 
1743 	vn_lock(vp, LK_SHARED | LK_RETRY);
1744 	error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1745 	VOP_UNLOCK(vp);
1746 
1747 	return (error);
1748 }
1749 
1750 static int
1751 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1752     struct filedesc *fdp)
1753 {
1754 	struct linux_file *filp;
1755 	struct vnode *vp;
1756 	int error;
1757 
1758 	filp = fp->f_data;
1759 	vp = filp->f_vnode;
1760 	if (vp == NULL) {
1761 		error = 0;
1762 		kif->kf_type = KF_TYPE_DEV;
1763 	} else {
1764 		vref(vp);
1765 		FILEDESC_SUNLOCK(fdp);
1766 		error = vn_fill_kinfo_vnode(vp, kif);
1767 		vrele(vp);
1768 		kif->kf_type = KF_TYPE_VNODE;
1769 		FILEDESC_SLOCK(fdp);
1770 	}
1771 	return (error);
1772 }
1773 
1774 unsigned int
1775 linux_iminor(struct inode *inode)
1776 {
1777 	struct linux_cdev *ldev;
1778 
1779 	if (inode == NULL || inode->v_rdev == NULL ||
1780 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1781 		return (-1U);
1782 	ldev = inode->v_rdev->si_drv1;
1783 	if (ldev == NULL)
1784 		return (-1U);
1785 
1786 	return (minor(ldev->dev));
1787 }
1788 
1789 static int
1790 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td)
1791 {
1792 	struct linux_file *filp1, *filp2;
1793 
1794 	if (fp2->f_type != DTYPE_DEV)
1795 		return (3);
1796 
1797 	filp1 = fp1->f_data;
1798 	filp2 = fp2->f_data;
1799 	return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev));
1800 }
1801 
1802 const struct fileops linuxfileops = {
1803 	.fo_read = linux_file_read,
1804 	.fo_write = linux_file_write,
1805 	.fo_truncate = invfo_truncate,
1806 	.fo_kqfilter = linux_file_kqfilter,
1807 	.fo_stat = linux_file_stat,
1808 	.fo_fill_kinfo = linux_file_fill_kinfo,
1809 	.fo_poll = linux_file_poll,
1810 	.fo_close = linux_file_close,
1811 	.fo_ioctl = linux_file_ioctl,
1812 	.fo_mmap = linux_file_mmap,
1813 	.fo_chmod = invfo_chmod,
1814 	.fo_chown = invfo_chown,
1815 	.fo_sendfile = invfo_sendfile,
1816 	.fo_cmp = linux_file_kcmp,
1817 	.fo_flags = DFLAG_PASSABLE,
1818 };
1819 
1820 static char *
1821 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1822 {
1823 	unsigned int len;
1824 	char *p;
1825 	va_list aq;
1826 
1827 	va_copy(aq, ap);
1828 	len = vsnprintf(NULL, 0, fmt, aq);
1829 	va_end(aq);
1830 
1831 	if (dev != NULL)
1832 		p = devm_kmalloc(dev, len + 1, gfp);
1833 	else
1834 		p = kmalloc(len + 1, gfp);
1835 	if (p != NULL)
1836 		vsnprintf(p, len + 1, fmt, ap);
1837 
1838 	return (p);
1839 }
1840 
1841 char *
1842 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1843 {
1844 
1845 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1846 }
1847 
1848 char *
1849 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1850 {
1851 	va_list ap;
1852 	char *p;
1853 
1854 	va_start(ap, fmt);
1855 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1856 	va_end(ap);
1857 
1858 	return (p);
1859 }
1860 
1861 char *
1862 kasprintf(gfp_t gfp, const char *fmt, ...)
1863 {
1864 	va_list ap;
1865 	char *p;
1866 
1867 	va_start(ap, fmt);
1868 	p = kvasprintf(gfp, fmt, ap);
1869 	va_end(ap);
1870 
1871 	return (p);
1872 }
1873 
1874 int
1875 __lkpi_hexdump_printf(void *arg1 __unused, const char *fmt, ...)
1876 {
1877 	va_list ap;
1878 	int result;
1879 
1880 	va_start(ap, fmt);
1881 	result = vprintf(fmt, ap);
1882 	va_end(ap);
1883 	return (result);
1884 }
1885 
1886 int
1887 __lkpi_hexdump_sbuf_printf(void *arg1, const char *fmt, ...)
1888 {
1889 	va_list ap;
1890 	int result;
1891 
1892 	va_start(ap, fmt);
1893 	result = sbuf_vprintf(arg1, fmt, ap);
1894 	va_end(ap);
1895 	return (result);
1896 }
1897 
1898 void
1899 lkpi_hex_dump(int(*_fpf)(void *, const char *, ...), void *arg1,
1900     const char *level, const char *prefix_str,
1901     const int prefix_type, const int rowsize, const int groupsize,
1902     const void *buf, size_t len, const bool ascii, const bool trailing_newline)
1903 {
1904 	typedef const struct { long long value; } __packed *print_64p_t;
1905 	typedef const struct { uint32_t value; } __packed *print_32p_t;
1906 	typedef const struct { uint16_t value; } __packed *print_16p_t;
1907 	const void *buf_old = buf;
1908 	int row, linelen, ret;
1909 
1910 	while (len > 0) {
1911 		linelen = 0;
1912 		if (level != NULL) {
1913 			ret = _fpf(arg1, "%s", level);
1914 			if (ret < 0)
1915 				break;
1916 			linelen += ret;
1917 		}
1918 		if (prefix_str != NULL) {
1919 			ret = _fpf(
1920 			    arg1, "%s%s", linelen ? " " : "", prefix_str);
1921 			if (ret < 0)
1922 				break;
1923 			linelen += ret;
1924 		}
1925 
1926 		switch (prefix_type) {
1927 		case DUMP_PREFIX_ADDRESS:
1928 			ret = _fpf(
1929 			    arg1, "%s[%p]", linelen ? " " : "", buf);
1930 			if (ret < 0)
1931 				return;
1932 			linelen += ret;
1933 			break;
1934 		case DUMP_PREFIX_OFFSET:
1935 			ret = _fpf(
1936 			    arg1, "%s[%#tx]", linelen ? " " : "",
1937 			    ((const char *)buf - (const char *)buf_old));
1938 			if (ret < 0)
1939 				return;
1940 			linelen += ret;
1941 			break;
1942 		default:
1943 			break;
1944 		}
1945 		for (row = 0; row != rowsize; row++) {
1946 			if (groupsize == 8 && len > 7) {
1947 				ret = _fpf(
1948 				    arg1, "%s%016llx", linelen ? " " : "",
1949 				    ((print_64p_t)buf)->value);
1950 				if (ret < 0)
1951 					return;
1952 				linelen += ret;
1953 				buf = (const uint8_t *)buf + 8;
1954 				len -= 8;
1955 			} else if (groupsize == 4 && len > 3) {
1956 				ret = _fpf(
1957 				    arg1, "%s%08x", linelen ? " " : "",
1958 				    ((print_32p_t)buf)->value);
1959 				if (ret < 0)
1960 					return;
1961 				linelen += ret;
1962 				buf = (const uint8_t *)buf + 4;
1963 				len -= 4;
1964 			} else if (groupsize == 2 && len > 1) {
1965 				ret = _fpf(
1966 				    arg1, "%s%04x", linelen ? " " : "",
1967 				    ((print_16p_t)buf)->value);
1968 				if (ret < 0)
1969 					return;
1970 				linelen += ret;
1971 				buf = (const uint8_t *)buf + 2;
1972 				len -= 2;
1973 			} else if (len > 0) {
1974 				ret = _fpf(
1975 				    arg1, "%s%02x", linelen ? " " : "",
1976 				    *(const uint8_t *)buf);
1977 				if (ret < 0)
1978 					return;
1979 				linelen += ret;
1980 				buf = (const uint8_t *)buf + 1;
1981 				len--;
1982 			} else {
1983 				break;
1984 			}
1985 		}
1986 		if (len > 0 && trailing_newline) {
1987 			ret = _fpf(arg1, "\n");
1988 			if (ret < 0)
1989 				break;
1990 		}
1991 	}
1992 }
1993 
1994 struct hdtb_context {
1995 	char	*linebuf;
1996 	size_t	 linebuflen;
1997 	int	 written;
1998 };
1999 
2000 static int
2001 hdtb_cb(void *arg, const char *format, ...)
2002 {
2003 	struct hdtb_context *context;
2004 	int written;
2005 	va_list args;
2006 
2007 	context = arg;
2008 
2009 	va_start(args, format);
2010 	written = vsnprintf(
2011 	    context->linebuf, context->linebuflen, format, args);
2012 	va_end(args);
2013 
2014 	if (written < 0)
2015 		return (written);
2016 
2017 	/*
2018 	 * Linux' hex_dump_to_buffer() function has the same behaviour as
2019 	 * snprintf() basically. Therefore, it returns the number of bytes it
2020 	 * would have written if the destination buffer was large enough.
2021 	 *
2022 	 * If the destination buffer was exhausted, lkpi_hex_dump() will
2023 	 * continue to call this callback but it will only compute the bytes it
2024 	 * would have written but write nothing to that buffer.
2025 	 */
2026 	context->written += written;
2027 
2028 	if (written < context->linebuflen) {
2029 		context->linebuf += written;
2030 		context->linebuflen -= written;
2031 	} else {
2032 		context->linebuf += context->linebuflen;
2033 		context->linebuflen = 0;
2034 	}
2035 
2036 	return (written);
2037 }
2038 
2039 int
2040 lkpi_hex_dump_to_buffer(const void *buf, size_t len, int rowsize,
2041     int groupsize, char *linebuf, size_t linebuflen, bool ascii)
2042 {
2043 	int written;
2044 	struct hdtb_context context;
2045 
2046 	context.linebuf = linebuf;
2047 	context.linebuflen = linebuflen;
2048 	context.written = 0;
2049 
2050 	if (rowsize != 16 && rowsize != 32)
2051 		rowsize = 16;
2052 
2053 	len = min(len, rowsize);
2054 
2055 	lkpi_hex_dump(
2056 	    hdtb_cb, &context, NULL, NULL, DUMP_PREFIX_NONE,
2057 	    rowsize, groupsize, buf, len, ascii, false);
2058 
2059 	written = context.written;
2060 
2061 	return (written);
2062 }
2063 
2064 static void
2065 linux_timer_callback_wrapper(void *context)
2066 {
2067 	struct timer_list *timer;
2068 
2069 	timer = context;
2070 
2071 	/* the timer is about to be shutdown permanently */
2072 	if (timer->function == NULL)
2073 		return;
2074 
2075 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
2076 		/* try again later */
2077 		callout_reset(&timer->callout, 1,
2078 		    &linux_timer_callback_wrapper, timer);
2079 		return;
2080 	}
2081 
2082 	timer->function(timer->data);
2083 }
2084 
2085 static int
2086 linux_timer_jiffies_until(unsigned long expires)
2087 {
2088 	unsigned long delta = expires - jiffies;
2089 
2090 	/*
2091 	 * Guard against already expired values and make sure that the value can
2092 	 * be used as a tick count, rather than a jiffies count.
2093 	 */
2094 	if ((long)delta < 1)
2095 		delta = 1;
2096 	else if (delta > INT_MAX)
2097 		delta = INT_MAX;
2098 	return ((int)delta);
2099 }
2100 
2101 int
2102 mod_timer(struct timer_list *timer, unsigned long expires)
2103 {
2104 	int ret;
2105 
2106 	timer->expires = expires;
2107 	ret = callout_reset(&timer->callout,
2108 	    linux_timer_jiffies_until(expires),
2109 	    &linux_timer_callback_wrapper, timer);
2110 
2111 	MPASS(ret == 0 || ret == 1);
2112 
2113 	return (ret == 1);
2114 }
2115 
2116 void
2117 add_timer(struct timer_list *timer)
2118 {
2119 
2120 	callout_reset(&timer->callout,
2121 	    linux_timer_jiffies_until(timer->expires),
2122 	    &linux_timer_callback_wrapper, timer);
2123 }
2124 
2125 void
2126 add_timer_on(struct timer_list *timer, int cpu)
2127 {
2128 
2129 	callout_reset_on(&timer->callout,
2130 	    linux_timer_jiffies_until(timer->expires),
2131 	    &linux_timer_callback_wrapper, timer, cpu);
2132 }
2133 
2134 int
2135 timer_delete(struct timer_list *timer)
2136 {
2137 
2138 	if (callout_stop(&(timer)->callout) == -1)
2139 		return (0);
2140 	return (1);
2141 }
2142 
2143 int
2144 timer_delete_sync(struct timer_list *timer)
2145 {
2146 
2147 	if (callout_drain(&(timer)->callout) == -1)
2148 		return (0);
2149 	return (1);
2150 }
2151 
2152 int
2153 timer_shutdown_sync(struct timer_list *timer)
2154 {
2155 
2156 	timer->function = NULL;
2157 	return (del_timer_sync(timer));
2158 }
2159 
2160 /* greatest common divisor, Euclid equation */
2161 static uint64_t
2162 lkpi_gcd_64(uint64_t a, uint64_t b)
2163 {
2164 	uint64_t an;
2165 	uint64_t bn;
2166 
2167 	while (b != 0) {
2168 		an = b;
2169 		bn = a % b;
2170 		a = an;
2171 		b = bn;
2172 	}
2173 	return (a);
2174 }
2175 
2176 uint64_t lkpi_nsec2hz_rem;
2177 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2178 uint64_t lkpi_nsec2hz_max;
2179 
2180 uint64_t lkpi_usec2hz_rem;
2181 uint64_t lkpi_usec2hz_div = 1000000ULL;
2182 uint64_t lkpi_usec2hz_max;
2183 
2184 uint64_t lkpi_msec2hz_rem;
2185 uint64_t lkpi_msec2hz_div = 1000ULL;
2186 uint64_t lkpi_msec2hz_max;
2187 
2188 static void
2189 linux_timer_init(void *arg)
2190 {
2191 	uint64_t gcd;
2192 
2193 	/*
2194 	 * Compute an internal HZ value which can divide 2**32 to
2195 	 * avoid timer rounding problems when the tick value wraps
2196 	 * around 2**32:
2197 	 */
2198 	linux_timer_hz_mask = 1;
2199 	while (linux_timer_hz_mask < (unsigned long)hz)
2200 		linux_timer_hz_mask *= 2;
2201 	linux_timer_hz_mask--;
2202 
2203 	/* compute some internal constants */
2204 
2205 	lkpi_nsec2hz_rem = hz;
2206 	lkpi_usec2hz_rem = hz;
2207 	lkpi_msec2hz_rem = hz;
2208 
2209 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2210 	lkpi_nsec2hz_rem /= gcd;
2211 	lkpi_nsec2hz_div /= gcd;
2212 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2213 
2214 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2215 	lkpi_usec2hz_rem /= gcd;
2216 	lkpi_usec2hz_div /= gcd;
2217 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2218 
2219 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2220 	lkpi_msec2hz_rem /= gcd;
2221 	lkpi_msec2hz_div /= gcd;
2222 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2223 }
2224 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2225 
2226 void
2227 linux_complete_common(struct completion *c, int all)
2228 {
2229 	sleepq_lock(c);
2230 	if (all) {
2231 		c->done = UINT_MAX;
2232 		sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2233 	} else {
2234 		if (c->done != UINT_MAX)
2235 			c->done++;
2236 		sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2237 	}
2238 	sleepq_release(c);
2239 }
2240 
2241 /*
2242  * Indefinite wait for done != 0 with or without signals.
2243  */
2244 int
2245 linux_wait_for_common(struct completion *c, int flags)
2246 {
2247 	struct task_struct *task;
2248 	int error;
2249 
2250 	if (SCHEDULER_STOPPED())
2251 		return (0);
2252 
2253 	task = current;
2254 
2255 	if (flags != 0)
2256 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2257 	else
2258 		flags = SLEEPQ_SLEEP;
2259 	error = 0;
2260 	for (;;) {
2261 		sleepq_lock(c);
2262 		if (c->done)
2263 			break;
2264 		sleepq_add(c, NULL, "completion", flags, 0);
2265 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2266 			DROP_GIANT();
2267 			error = -sleepq_wait_sig(c, 0);
2268 			PICKUP_GIANT();
2269 			if (error != 0) {
2270 				linux_schedule_save_interrupt_value(task, error);
2271 				error = -ERESTARTSYS;
2272 				goto intr;
2273 			}
2274 		} else {
2275 			DROP_GIANT();
2276 			sleepq_wait(c, 0);
2277 			PICKUP_GIANT();
2278 		}
2279 	}
2280 	if (c->done != UINT_MAX)
2281 		c->done--;
2282 	sleepq_release(c);
2283 
2284 intr:
2285 	return (error);
2286 }
2287 
2288 /*
2289  * Time limited wait for done != 0 with or without signals.
2290  */
2291 unsigned long
2292 linux_wait_for_timeout_common(struct completion *c, unsigned long timeout,
2293     int flags)
2294 {
2295 	struct task_struct *task;
2296 	unsigned long end = jiffies + timeout, error;
2297 
2298 	if (SCHEDULER_STOPPED())
2299 		return (0);
2300 
2301 	task = current;
2302 
2303 	if (flags != 0)
2304 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2305 	else
2306 		flags = SLEEPQ_SLEEP;
2307 
2308 	for (;;) {
2309 		sleepq_lock(c);
2310 		if (c->done)
2311 			break;
2312 		sleepq_add(c, NULL, "completion", flags, 0);
2313 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2314 
2315 		DROP_GIANT();
2316 		if (flags & SLEEPQ_INTERRUPTIBLE)
2317 			error = -sleepq_timedwait_sig(c, 0);
2318 		else
2319 			error = -sleepq_timedwait(c, 0);
2320 		PICKUP_GIANT();
2321 
2322 		if (error != 0) {
2323 			/* check for timeout */
2324 			if (error == -EWOULDBLOCK) {
2325 				error = 0;	/* timeout */
2326 			} else {
2327 				/* signal happened */
2328 				linux_schedule_save_interrupt_value(task, error);
2329 				error = -ERESTARTSYS;
2330 			}
2331 			goto done;
2332 		}
2333 	}
2334 	if (c->done != UINT_MAX)
2335 		c->done--;
2336 	sleepq_release(c);
2337 
2338 	/* return how many jiffies are left */
2339 	error = linux_timer_jiffies_until(end);
2340 done:
2341 	return (error);
2342 }
2343 
2344 int
2345 linux_try_wait_for_completion(struct completion *c)
2346 {
2347 	int isdone;
2348 
2349 	sleepq_lock(c);
2350 	isdone = (c->done != 0);
2351 	if (c->done != 0 && c->done != UINT_MAX)
2352 		c->done--;
2353 	sleepq_release(c);
2354 	return (isdone);
2355 }
2356 
2357 int
2358 linux_completion_done(struct completion *c)
2359 {
2360 	int isdone;
2361 
2362 	sleepq_lock(c);
2363 	isdone = (c->done != 0);
2364 	sleepq_release(c);
2365 	return (isdone);
2366 }
2367 
2368 static void
2369 linux_cdev_deref(struct linux_cdev *ldev)
2370 {
2371 	if (refcount_release(&ldev->refs) &&
2372 	    ldev->kobj.ktype == &linux_cdev_ktype)
2373 		kfree(ldev);
2374 }
2375 
2376 static void
2377 linux_cdev_release(struct kobject *kobj)
2378 {
2379 	struct linux_cdev *cdev;
2380 	struct kobject *parent;
2381 
2382 	cdev = container_of(kobj, struct linux_cdev, kobj);
2383 	parent = kobj->parent;
2384 	linux_destroy_dev(cdev);
2385 	linux_cdev_deref(cdev);
2386 	kobject_put(parent);
2387 }
2388 
2389 static void
2390 linux_cdev_static_release(struct kobject *kobj)
2391 {
2392 	struct cdev *cdev;
2393 	struct linux_cdev *ldev;
2394 
2395 	ldev = container_of(kobj, struct linux_cdev, kobj);
2396 	cdev = ldev->cdev;
2397 	if (cdev != NULL) {
2398 		destroy_dev(cdev);
2399 		ldev->cdev = NULL;
2400 	}
2401 	kobject_put(kobj->parent);
2402 }
2403 
2404 int
2405 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2406 {
2407 	int ret;
2408 
2409 	if (dev->devt != 0) {
2410 		/* Set parent kernel object. */
2411 		ldev->kobj.parent = &dev->kobj;
2412 
2413 		/*
2414 		 * Unlike Linux we require the kobject of the
2415 		 * character device structure to have a valid name
2416 		 * before calling this function:
2417 		 */
2418 		if (ldev->kobj.name == NULL)
2419 			return (-EINVAL);
2420 
2421 		ret = cdev_add(ldev, dev->devt, 1);
2422 		if (ret)
2423 			return (ret);
2424 	}
2425 	ret = device_add(dev);
2426 	if (ret != 0 && dev->devt != 0)
2427 		cdev_del(ldev);
2428 	return (ret);
2429 }
2430 
2431 void
2432 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2433 {
2434 	device_del(dev);
2435 
2436 	if (dev->devt != 0)
2437 		cdev_del(ldev);
2438 }
2439 
2440 static void
2441 linux_destroy_dev(struct linux_cdev *ldev)
2442 {
2443 
2444 	if (ldev->cdev == NULL)
2445 		return;
2446 
2447 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2448 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2449 
2450 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2451 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2452 		pause("ldevdtr", hz / 4);
2453 
2454 	destroy_dev(ldev->cdev);
2455 	ldev->cdev = NULL;
2456 }
2457 
2458 const struct kobj_type linux_cdev_ktype = {
2459 	.release = linux_cdev_release,
2460 };
2461 
2462 const struct kobj_type linux_cdev_static_ktype = {
2463 	.release = linux_cdev_static_release,
2464 };
2465 
2466 static void
2467 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2468 {
2469 	struct notifier_block *nb;
2470 	struct netdev_notifier_info ni;
2471 
2472 	nb = arg;
2473 	ni.ifp = ifp;
2474 	ni.dev = (struct net_device *)ifp;
2475 	if (linkstate == LINK_STATE_UP)
2476 		nb->notifier_call(nb, NETDEV_UP, &ni);
2477 	else
2478 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2479 }
2480 
2481 static void
2482 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2483 {
2484 	struct notifier_block *nb;
2485 	struct netdev_notifier_info ni;
2486 
2487 	nb = arg;
2488 	ni.ifp = ifp;
2489 	ni.dev = (struct net_device *)ifp;
2490 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2491 }
2492 
2493 static void
2494 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2495 {
2496 	struct notifier_block *nb;
2497 	struct netdev_notifier_info ni;
2498 
2499 	nb = arg;
2500 	ni.ifp = ifp;
2501 	ni.dev = (struct net_device *)ifp;
2502 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2503 }
2504 
2505 static void
2506 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2507 {
2508 	struct notifier_block *nb;
2509 	struct netdev_notifier_info ni;
2510 
2511 	nb = arg;
2512 	ni.ifp = ifp;
2513 	ni.dev = (struct net_device *)ifp;
2514 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2515 }
2516 
2517 static void
2518 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2519 {
2520 	struct notifier_block *nb;
2521 	struct netdev_notifier_info ni;
2522 
2523 	nb = arg;
2524 	ni.ifp = ifp;
2525 	ni.dev = (struct net_device *)ifp;
2526 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2527 }
2528 
2529 int
2530 register_netdevice_notifier(struct notifier_block *nb)
2531 {
2532 
2533 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2534 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2535 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2536 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2537 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2538 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2539 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2540 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2541 
2542 	return (0);
2543 }
2544 
2545 int
2546 register_inetaddr_notifier(struct notifier_block *nb)
2547 {
2548 
2549 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2550 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2551 	return (0);
2552 }
2553 
2554 int
2555 unregister_netdevice_notifier(struct notifier_block *nb)
2556 {
2557 
2558 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2559 	    nb->tags[NETDEV_UP]);
2560 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2561 	    nb->tags[NETDEV_REGISTER]);
2562 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2563 	    nb->tags[NETDEV_UNREGISTER]);
2564 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2565 	    nb->tags[NETDEV_CHANGEADDR]);
2566 
2567 	return (0);
2568 }
2569 
2570 int
2571 unregister_inetaddr_notifier(struct notifier_block *nb)
2572 {
2573 
2574 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2575 	    nb->tags[NETDEV_CHANGEIFADDR]);
2576 
2577 	return (0);
2578 }
2579 
2580 struct list_sort_thunk {
2581 	int (*cmp)(void *, struct list_head *, struct list_head *);
2582 	void *priv;
2583 };
2584 
2585 static inline int
2586 linux_le_cmp(const void *d1, const void *d2, void *priv)
2587 {
2588 	struct list_head *le1, *le2;
2589 	struct list_sort_thunk *thunk;
2590 
2591 	thunk = priv;
2592 	le1 = *(__DECONST(struct list_head **, d1));
2593 	le2 = *(__DECONST(struct list_head **, d2));
2594 	return ((thunk->cmp)(thunk->priv, le1, le2));
2595 }
2596 
2597 void
2598 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2599     struct list_head *a, struct list_head *b))
2600 {
2601 	struct list_sort_thunk thunk;
2602 	struct list_head **ar, *le;
2603 	size_t count, i;
2604 
2605 	count = 0;
2606 	list_for_each(le, head)
2607 		count++;
2608 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2609 	i = 0;
2610 	list_for_each(le, head)
2611 		ar[i++] = le;
2612 	thunk.cmp = cmp;
2613 	thunk.priv = priv;
2614 	qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk);
2615 	INIT_LIST_HEAD(head);
2616 	for (i = 0; i < count; i++)
2617 		list_add_tail(ar[i], head);
2618 	free(ar, M_KMALLOC);
2619 }
2620 
2621 #if defined(__i386__) || defined(__amd64__)
2622 int
2623 linux_wbinvd_on_all_cpus(void)
2624 {
2625 
2626 	pmap_invalidate_cache();
2627 	return (0);
2628 }
2629 #endif
2630 
2631 int
2632 linux_on_each_cpu(void callback(void *), void *data)
2633 {
2634 
2635 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2636 	    smp_no_rendezvous_barrier, data);
2637 	return (0);
2638 }
2639 
2640 int
2641 linux_in_atomic(void)
2642 {
2643 
2644 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2645 }
2646 
2647 struct linux_cdev *
2648 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2649 {
2650 	dev_t dev = MKDEV(major, minor);
2651 	struct cdev *cdev;
2652 
2653 	dev_lock();
2654 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2655 		struct linux_cdev *ldev = cdev->si_drv1;
2656 		if (ldev->dev == dev &&
2657 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2658 			break;
2659 		}
2660 	}
2661 	dev_unlock();
2662 
2663 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2664 }
2665 
2666 int
2667 __register_chrdev(unsigned int major, unsigned int baseminor,
2668     unsigned int count, const char *name,
2669     const struct file_operations *fops)
2670 {
2671 	struct linux_cdev *cdev;
2672 	int ret = 0;
2673 	int i;
2674 
2675 	for (i = baseminor; i < baseminor + count; i++) {
2676 		cdev = cdev_alloc();
2677 		cdev->ops = fops;
2678 		kobject_set_name(&cdev->kobj, name);
2679 
2680 		ret = cdev_add(cdev, makedev(major, i), 1);
2681 		if (ret != 0)
2682 			break;
2683 	}
2684 	return (ret);
2685 }
2686 
2687 int
2688 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2689     unsigned int count, const char *name,
2690     const struct file_operations *fops, uid_t uid,
2691     gid_t gid, int mode)
2692 {
2693 	struct linux_cdev *cdev;
2694 	int ret = 0;
2695 	int i;
2696 
2697 	for (i = baseminor; i < baseminor + count; i++) {
2698 		cdev = cdev_alloc();
2699 		cdev->ops = fops;
2700 		kobject_set_name(&cdev->kobj, name);
2701 
2702 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2703 		if (ret != 0)
2704 			break;
2705 	}
2706 	return (ret);
2707 }
2708 
2709 void
2710 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2711     unsigned int count, const char *name)
2712 {
2713 	struct linux_cdev *cdevp;
2714 	int i;
2715 
2716 	for (i = baseminor; i < baseminor + count; i++) {
2717 		cdevp = linux_find_cdev(name, major, i);
2718 		if (cdevp != NULL)
2719 			cdev_del(cdevp);
2720 	}
2721 }
2722 
2723 void
2724 linux_dump_stack(void)
2725 {
2726 #ifdef STACK
2727 	struct stack st;
2728 
2729 	stack_save(&st);
2730 	stack_print(&st);
2731 #endif
2732 }
2733 
2734 int
2735 linuxkpi_net_ratelimit(void)
2736 {
2737 
2738 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2739 	   lkpi_net_maxpps));
2740 }
2741 
2742 struct io_mapping *
2743 io_mapping_create_wc(resource_size_t base, unsigned long size)
2744 {
2745 	struct io_mapping *mapping;
2746 
2747 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2748 	if (mapping == NULL)
2749 		return (NULL);
2750 	return (io_mapping_init_wc(mapping, base, size));
2751 }
2752 
2753 /* We likely want a linuxkpi_device.c at some point. */
2754 bool
2755 device_can_wakeup(struct device *dev)
2756 {
2757 
2758 	if (dev == NULL)
2759 		return (false);
2760 	/*
2761 	 * XXX-BZ iwlwifi queries it as part of enabling WoWLAN.
2762 	 * Normally this would be based on a bool in dev->power.XXX.
2763 	 * Check such as PCI PCIM_PCAP_*PME.  We have no way to enable this yet.
2764 	 * We may get away by directly calling into bsddev for as long as
2765 	 * we can assume PCI only avoiding changing struct device breaking KBI.
2766 	 */
2767 	pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__);
2768 	return (false);
2769 }
2770 
2771 static void
2772 devm_device_group_remove(struct device *dev, void *p)
2773 {
2774 	const struct attribute_group **dr = p;
2775 	const struct attribute_group *group = *dr;
2776 
2777 	sysfs_remove_group(&dev->kobj, group);
2778 }
2779 
2780 int
2781 lkpi_devm_device_add_group(struct device *dev,
2782     const struct attribute_group *group)
2783 {
2784 	const struct attribute_group **dr;
2785 	int ret;
2786 
2787 	dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL);
2788 	if (dr == NULL)
2789 		return (-ENOMEM);
2790 
2791 	ret = sysfs_create_group(&dev->kobj, group);
2792 	if (ret == 0) {
2793 		*dr = group;
2794 		devres_add(dev, dr);
2795 	} else
2796 		devres_free(dr);
2797 
2798 	return (ret);
2799 }
2800 
2801 #if defined(__i386__) || defined(__amd64__)
2802 bool linux_cpu_has_clflush;
2803 struct cpuinfo_x86 boot_cpu_data;
2804 struct cpuinfo_x86 *__cpu_data;
2805 #endif
2806 
2807 cpumask_t *
2808 lkpi_get_static_single_cpu_mask(int cpuid)
2809 {
2810 
2811 	KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n",
2812 	    __func__, cpuid));
2813 	KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n",
2814 	    __func__, cpuid));
2815 
2816 	return (static_single_cpu_mask[cpuid]);
2817 }
2818 
2819 bool
2820 lkpi_xen_initial_domain(void)
2821 {
2822 #ifdef XENHVM
2823 	return (xen_initial_domain());
2824 #else
2825 	return (false);
2826 #endif
2827 }
2828 
2829 bool
2830 lkpi_xen_pv_domain(void)
2831 {
2832 #ifdef XENHVM
2833 	return (xen_pv_domain());
2834 #else
2835 	return (false);
2836 #endif
2837 }
2838 
2839 static void
2840 linux_compat_init(void *arg)
2841 {
2842 	struct sysctl_oid *rootoid;
2843 	int i;
2844 
2845 #if defined(__i386__) || defined(__amd64__)
2846 	static const uint32_t x86_vendors[X86_VENDOR_NUM] = {
2847 		[X86_VENDOR_INTEL] = CPU_VENDOR_INTEL,
2848 		[X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX,
2849 		[X86_VENDOR_AMD] = CPU_VENDOR_AMD,
2850 		[X86_VENDOR_UMC] = CPU_VENDOR_UMC,
2851 		[X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR,
2852 		[X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA,
2853 		[X86_VENDOR_NSC] = CPU_VENDOR_NSC,
2854 		[X86_VENDOR_HYGON] = CPU_VENDOR_HYGON,
2855 	};
2856 	uint8_t x86_vendor = X86_VENDOR_UNKNOWN;
2857 
2858 	for (i = 0; i < X86_VENDOR_NUM; i++) {
2859 		if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) {
2860 			x86_vendor = i;
2861 			break;
2862 		}
2863 	}
2864 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2865 	boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2866 	boot_cpu_data.x86_max_cores = mp_ncpus;
2867 	boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id);
2868 	boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id);
2869 	boot_cpu_data.x86_vendor = x86_vendor;
2870 
2871 	__cpu_data = kmalloc_array(mp_maxid + 1,
2872 	    sizeof(*__cpu_data), M_WAITOK | M_ZERO);
2873 	CPU_FOREACH(i) {
2874 		__cpu_data[i].x86_clflush_size = cpu_clflush_line_size;
2875 		__cpu_data[i].x86_max_cores = mp_ncpus;
2876 		__cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id);
2877 		__cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id);
2878 		__cpu_data[i].x86_vendor = x86_vendor;
2879 	}
2880 #endif
2881 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2882 
2883 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2884 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2885 	kobject_init(&linux_class_root, &linux_class_ktype);
2886 	kobject_set_name(&linux_class_root, "class");
2887 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2888 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2889 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2890 	kobject_set_name(&linux_root_device.kobj, "device");
2891 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2892 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2893 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2894 	linux_root_device.bsddev = root_bus;
2895 	linux_class_misc.name = "misc";
2896 	class_register(&linux_class_misc);
2897 	INIT_LIST_HEAD(&pci_drivers);
2898 	INIT_LIST_HEAD(&pci_devices);
2899 	spin_lock_init(&pci_lock);
2900 	init_waitqueue_head(&linux_bit_waitq);
2901 	init_waitqueue_head(&linux_var_waitq);
2902 
2903 	CPU_COPY(&all_cpus, &cpu_online_mask);
2904 	/*
2905 	 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system.
2906 	 * CPUs are indexed from 0..(mp_maxid).  The entry for cpuid 0 will only
2907 	 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on.
2908 	 * This is used by cpumask_of() (and possibly others in the future) for,
2909 	 * e.g., drivers to pass hints to irq_set_affinity_hint().
2910 	 */
2911 	static_single_cpu_mask = kmalloc_array(mp_maxid + 1,
2912 	    sizeof(static_single_cpu_mask), M_WAITOK | M_ZERO);
2913 
2914 	/*
2915 	 * When the number of CPUs reach a threshold, we start to save memory
2916 	 * given the sets are static by overlapping those having their single
2917 	 * bit set at same position in a bitset word.  Asymptotically, this
2918 	 * regular scheme is in O(n²) whereas the overlapping one is in O(n)
2919 	 * only with n being the maximum number of CPUs, so the gain will become
2920 	 * huge quite quickly.  The threshold for 64-bit architectures is 128
2921 	 * CPUs.
2922 	 */
2923 	if (mp_ncpus < (2 * _BITSET_BITS)) {
2924 		cpumask_t *sscm_ptr;
2925 
2926 		/*
2927 		 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) *
2928 		 * (_BITSET_BITS / 8)' bytes (for comparison with the
2929 		 * overlapping scheme).
2930 		 */
2931 		static_single_cpu_mask_lcs = kmalloc_array(mp_ncpus,
2932 		    sizeof(*static_single_cpu_mask_lcs),
2933 		    M_WAITOK | M_ZERO);
2934 
2935 		sscm_ptr = static_single_cpu_mask_lcs;
2936 		CPU_FOREACH(i) {
2937 			static_single_cpu_mask[i] = sscm_ptr++;
2938 			CPU_SET(i, static_single_cpu_mask[i]);
2939 		}
2940 	} else {
2941 		/* Pointer to a bitset word. */
2942 		__typeof(((cpuset_t *)NULL)->__bits[0]) *bwp;
2943 
2944 		/*
2945 		 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t'
2946 		 * really) with a single bit set that can be reused for all
2947 		 * single CPU masks by making them start at different offsets.
2948 		 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before
2949 		 * the word having its single bit set, and the same amount
2950 		 * after.
2951 		 */
2952 		static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS,
2953 		    (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8),
2954 		    M_KMALLOC, M_WAITOK | M_ZERO);
2955 
2956 		/*
2957 		 * We rely below on cpuset_t and the bitset generic
2958 		 * implementation assigning words in the '__bits' array in the
2959 		 * same order of bits (i.e., little-endian ordering, not to be
2960 		 * confused with machine endianness, which concerns bits in
2961 		 * words and other integers).  This is an imperfect test, but it
2962 		 * will detect a change to big-endian ordering.
2963 		 */
2964 		_Static_assert(
2965 		    __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1,
2966 		    "Assumes a bitset implementation that is little-endian "
2967 		    "on its words");
2968 
2969 		/* Initialize the single bit of each static span. */
2970 		bwp = (__typeof(bwp))static_single_cpu_mask_lcs +
2971 		    (__bitset_words(CPU_SETSIZE) - 1);
2972 		for (i = 0; i < _BITSET_BITS; i++) {
2973 			CPU_SET(i, (cpuset_t *)bwp);
2974 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1);
2975 		}
2976 
2977 		/*
2978 		 * Finally set all CPU masks to the proper word in their
2979 		 * relevant span.
2980 		 */
2981 		CPU_FOREACH(i) {
2982 			bwp = (__typeof(bwp))static_single_cpu_mask_lcs;
2983 			/* Find the non-zero word of the relevant span. */
2984 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) *
2985 			    (i % _BITSET_BITS) +
2986 			    __bitset_words(CPU_SETSIZE) - 1;
2987 			/* Shift to find the CPU mask start. */
2988 			bwp -= (i / _BITSET_BITS);
2989 			static_single_cpu_mask[i] = (cpuset_t *)bwp;
2990 		}
2991 	}
2992 
2993 	strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release));
2994 }
2995 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2996 
2997 static void
2998 linux_compat_uninit(void *arg)
2999 {
3000 	linux_kobject_kfree_name(&linux_class_root);
3001 	linux_kobject_kfree_name(&linux_root_device.kobj);
3002 	linux_kobject_kfree_name(&linux_class_misc.kobj);
3003 
3004 	free(static_single_cpu_mask_lcs, M_KMALLOC);
3005 	free(static_single_cpu_mask, M_KMALLOC);
3006 #if defined(__i386__) || defined(__amd64__)
3007 	free(__cpu_data, M_KMALLOC);
3008 #endif
3009 
3010 	spin_lock_destroy(&pci_lock);
3011 	rw_destroy(&linux_vma_lock);
3012 }
3013 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
3014 
3015 /*
3016  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
3017  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
3018  * used. Assert these types have the same size, else some parts of the
3019  * LinuxKPI may not work like expected:
3020  */
3021 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
3022