linux_compat.c revision 294636
1219820Sjeff/*-
2219820Sjeff * Copyright (c) 2010 Isilon Systems, Inc.
3219820Sjeff * Copyright (c) 2010 iX Systems, Inc.
4219820Sjeff * Copyright (c) 2010 Panasas, Inc.
5271127Shselasky * Copyright (c) 2013, 2014 Mellanox Technologies, Ltd.
6219820Sjeff * All rights reserved.
7219820Sjeff *
8219820Sjeff * Redistribution and use in source and binary forms, with or without
9219820Sjeff * modification, are permitted provided that the following conditions
10219820Sjeff * are met:
11219820Sjeff * 1. Redistributions of source code must retain the above copyright
12219820Sjeff *    notice unmodified, this list of conditions, and the following
13219820Sjeff *    disclaimer.
14219820Sjeff * 2. Redistributions in binary form must reproduce the above copyright
15219820Sjeff *    notice, this list of conditions and the following disclaimer in the
16219820Sjeff *    documentation and/or other materials provided with the distribution.
17219820Sjeff *
18219820Sjeff * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19219820Sjeff * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20219820Sjeff * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21219820Sjeff * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22219820Sjeff * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23219820Sjeff * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24219820Sjeff * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25219820Sjeff * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26219820Sjeff * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27219820Sjeff * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28219820Sjeff */
29219820Sjeff
30219820Sjeff#include <sys/param.h>
31219820Sjeff#include <sys/systm.h>
32219820Sjeff#include <sys/malloc.h>
33219820Sjeff#include <sys/kernel.h>
34219820Sjeff#include <sys/sysctl.h>
35282513Shselasky#include <sys/proc.h>
36287637Sjhb#include <sys/sglist.h>
37282513Shselasky#include <sys/sleepqueue.h>
38219820Sjeff#include <sys/lock.h>
39219820Sjeff#include <sys/mutex.h>
40219820Sjeff#include <sys/bus.h>
41219820Sjeff#include <sys/fcntl.h>
42219820Sjeff#include <sys/file.h>
43219820Sjeff#include <sys/filio.h>
44248084Sattilio#include <sys/rwlock.h>
45219820Sjeff
46219820Sjeff#include <vm/vm.h>
47219820Sjeff#include <vm/pmap.h>
48219820Sjeff
49219820Sjeff#include <machine/stdarg.h>
50219820Sjeff#include <machine/pmap.h>
51219820Sjeff
52219820Sjeff#include <linux/kobject.h>
53219820Sjeff#include <linux/device.h>
54219820Sjeff#include <linux/slab.h>
55219820Sjeff#include <linux/module.h>
56219820Sjeff#include <linux/cdev.h>
57219820Sjeff#include <linux/file.h>
58219820Sjeff#include <linux/sysfs.h>
59219820Sjeff#include <linux/mm.h>
60219820Sjeff#include <linux/io.h>
61219820Sjeff#include <linux/vmalloc.h>
62282513Shselasky#include <linux/timer.h>
63282513Shselasky#include <linux/netdevice.h>
64219820Sjeff
65219820Sjeff#include <vm/vm_pager.h>
66219820Sjeff
67293151Shselasky#include <linux/workqueue.h>
68293151Shselasky
69219820SjeffMALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
70219820Sjeff
71219820Sjeff#include <linux/rbtree.h>
72219820Sjeff/* Undo Linux compat changes. */
73219820Sjeff#undef RB_ROOT
74219820Sjeff#undef file
75219820Sjeff#undef cdev
76219820Sjeff#define	RB_ROOT(head)	(head)->rbh_root
77219820Sjeff
78219820Sjeffstruct kobject class_root;
79219820Sjeffstruct device linux_rootdev;
80219820Sjeffstruct class miscclass;
81219820Sjeffstruct list_head pci_drivers;
82219820Sjeffstruct list_head pci_devices;
83282513Shselaskystruct net init_net;
84219820Sjeffspinlock_t pci_lock;
85219820Sjeff
86282513Shselaskyunsigned long linux_timer_hz_mask;
87282513Shselasky
88219820Sjeffint
89219820Sjeffpanic_cmp(struct rb_node *one, struct rb_node *two)
90219820Sjeff{
91219820Sjeff	panic("no cmp");
92219820Sjeff}
93219820Sjeff
94219820SjeffRB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
95293151Shselasky
96219820Sjeffint
97293151Shselaskykobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
98293151Shselasky{
99293151Shselasky	va_list tmp_va;
100293151Shselasky	int len;
101293151Shselasky	char *old;
102293151Shselasky	char *name;
103293151Shselasky	char dummy;
104293151Shselasky
105293151Shselasky	old = kobj->name;
106293151Shselasky
107293151Shselasky	if (old && fmt == NULL)
108293151Shselasky		return (0);
109293151Shselasky
110293151Shselasky	/* compute length of string */
111293151Shselasky	va_copy(tmp_va, args);
112293151Shselasky	len = vsnprintf(&dummy, 0, fmt, tmp_va);
113293151Shselasky	va_end(tmp_va);
114293151Shselasky
115293151Shselasky	/* account for zero termination */
116293151Shselasky	len++;
117293151Shselasky
118293151Shselasky	/* check for error */
119293151Shselasky	if (len < 1)
120293151Shselasky		return (-EINVAL);
121293151Shselasky
122293151Shselasky	/* allocate memory for string */
123293151Shselasky	name = kzalloc(len, GFP_KERNEL);
124293151Shselasky	if (name == NULL)
125293151Shselasky		return (-ENOMEM);
126293151Shselasky	vsnprintf(name, len, fmt, args);
127293151Shselasky	kobj->name = name;
128293151Shselasky
129293151Shselasky	/* free old string */
130293151Shselasky	kfree(old);
131293151Shselasky
132293151Shselasky	/* filter new string */
133293151Shselasky	for (; *name != '\0'; name++)
134293151Shselasky		if (*name == '/')
135293151Shselasky			*name = '!';
136293151Shselasky	return (0);
137293151Shselasky}
138293151Shselasky
139293151Shselaskyint
140219820Sjeffkobject_set_name(struct kobject *kobj, const char *fmt, ...)
141219820Sjeff{
142219820Sjeff	va_list args;
143219820Sjeff	int error;
144219820Sjeff
145219820Sjeff	va_start(args, fmt);
146219820Sjeff	error = kobject_set_name_vargs(kobj, fmt, args);
147219820Sjeff	va_end(args);
148219820Sjeff
149219820Sjeff	return (error);
150219820Sjeff}
151219820Sjeff
152219820Sjeffstatic inline int
153219820Sjeffkobject_add_complete(struct kobject *kobj, struct kobject *parent)
154219820Sjeff{
155219820Sjeff	struct kobj_type *t;
156219820Sjeff	int error;
157219820Sjeff
158219820Sjeff	kobj->parent = kobject_get(parent);
159219820Sjeff	error = sysfs_create_dir(kobj);
160219820Sjeff	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
161219820Sjeff		struct attribute **attr;
162219820Sjeff		t = kobj->ktype;
163219820Sjeff
164219820Sjeff		for (attr = t->default_attrs; *attr != NULL; attr++) {
165219820Sjeff			error = sysfs_create_file(kobj, *attr);
166219820Sjeff			if (error)
167219820Sjeff				break;
168219820Sjeff		}
169219820Sjeff		if (error)
170219820Sjeff			sysfs_remove_dir(kobj);
171219820Sjeff
172219820Sjeff	}
173219820Sjeff	return (error);
174219820Sjeff}
175219820Sjeff
176219820Sjeffint
177219820Sjeffkobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
178219820Sjeff{
179219820Sjeff	va_list args;
180219820Sjeff	int error;
181219820Sjeff
182219820Sjeff	va_start(args, fmt);
183219820Sjeff	error = kobject_set_name_vargs(kobj, fmt, args);
184219820Sjeff	va_end(args);
185219820Sjeff	if (error)
186219820Sjeff		return (error);
187219820Sjeff
188219820Sjeff	return kobject_add_complete(kobj, parent);
189219820Sjeff}
190219820Sjeff
191219820Sjeffvoid
192219820Sjeffkobject_release(struct kref *kref)
193219820Sjeff{
194219820Sjeff	struct kobject *kobj;
195219820Sjeff	char *name;
196219820Sjeff
197219820Sjeff	kobj = container_of(kref, struct kobject, kref);
198219820Sjeff	sysfs_remove_dir(kobj);
199219820Sjeff	if (kobj->parent)
200219820Sjeff		kobject_put(kobj->parent);
201219820Sjeff	kobj->parent = NULL;
202219820Sjeff	name = kobj->name;
203219820Sjeff	if (kobj->ktype && kobj->ktype->release)
204219820Sjeff		kobj->ktype->release(kobj);
205219820Sjeff	kfree(name);
206219820Sjeff}
207219820Sjeff
208219820Sjeffstatic void
209219820Sjeffkobject_kfree(struct kobject *kobj)
210219820Sjeff{
211219820Sjeff	kfree(kobj);
212219820Sjeff}
213219820Sjeff
214271127Shselaskystatic void
215271127Shselaskykobject_kfree_name(struct kobject *kobj)
216271127Shselasky{
217271127Shselasky	if (kobj) {
218271127Shselasky		kfree(kobj->name);
219271127Shselasky	}
220271127Shselasky}
221271127Shselasky
222219820Sjeffstruct kobj_type kfree_type = { .release = kobject_kfree };
223219820Sjeff
224277139Shselaskystatic void
225277139Shselaskydev_release(struct device *dev)
226277139Shselasky{
227277139Shselasky	pr_debug("dev_release: %s\n", dev_name(dev));
228277139Shselasky	kfree(dev);
229277139Shselasky}
230277139Shselasky
231219820Sjeffstruct device *
232219820Sjeffdevice_create(struct class *class, struct device *parent, dev_t devt,
233219820Sjeff    void *drvdata, const char *fmt, ...)
234219820Sjeff{
235219820Sjeff	struct device *dev;
236219820Sjeff	va_list args;
237219820Sjeff
238219820Sjeff	dev = kzalloc(sizeof(*dev), M_WAITOK);
239219820Sjeff	dev->parent = parent;
240219820Sjeff	dev->class = class;
241219820Sjeff	dev->devt = devt;
242219820Sjeff	dev->driver_data = drvdata;
243277139Shselasky	dev->release = dev_release;
244219820Sjeff	va_start(args, fmt);
245219820Sjeff	kobject_set_name_vargs(&dev->kobj, fmt, args);
246219820Sjeff	va_end(args);
247219820Sjeff	device_register(dev);
248219820Sjeff
249219820Sjeff	return (dev);
250219820Sjeff}
251219820Sjeff
252219820Sjeffint
253219820Sjeffkobject_init_and_add(struct kobject *kobj, struct kobj_type *ktype,
254219820Sjeff    struct kobject *parent, const char *fmt, ...)
255219820Sjeff{
256219820Sjeff	va_list args;
257219820Sjeff	int error;
258219820Sjeff
259219820Sjeff	kobject_init(kobj, ktype);
260219820Sjeff	kobj->ktype = ktype;
261219820Sjeff	kobj->parent = parent;
262219820Sjeff	kobj->name = NULL;
263219820Sjeff
264219820Sjeff	va_start(args, fmt);
265219820Sjeff	error = kobject_set_name_vargs(kobj, fmt, args);
266219820Sjeff	va_end(args);
267219820Sjeff	if (error)
268219820Sjeff		return (error);
269219820Sjeff	return kobject_add_complete(kobj, parent);
270219820Sjeff}
271219820Sjeff
272219820Sjeffstatic void
273219820Sjefflinux_file_dtor(void *cdp)
274219820Sjeff{
275219820Sjeff	struct linux_file *filp;
276219820Sjeff
277219820Sjeff	filp = cdp;
278251617Sjhb	filp->f_op->release(filp->f_vnode, filp);
279251617Sjhb	vdrop(filp->f_vnode);
280219820Sjeff	kfree(filp);
281219820Sjeff}
282219820Sjeff
283219820Sjeffstatic int
284219820Sjefflinux_dev_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
285219820Sjeff{
286219820Sjeff	struct linux_cdev *ldev;
287219820Sjeff	struct linux_file *filp;
288219820Sjeff	struct file *file;
289219820Sjeff	int error;
290219820Sjeff
291219820Sjeff	file = curthread->td_fpop;
292219820Sjeff	ldev = dev->si_drv1;
293219820Sjeff	if (ldev == NULL)
294219820Sjeff		return (ENODEV);
295219820Sjeff	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
296219820Sjeff	filp->f_dentry = &filp->f_dentry_store;
297219820Sjeff	filp->f_op = ldev->ops;
298219820Sjeff	filp->f_flags = file->f_flag;
299251617Sjhb	vhold(file->f_vnode);
300251617Sjhb	filp->f_vnode = file->f_vnode;
301219820Sjeff	if (filp->f_op->open) {
302219820Sjeff		error = -filp->f_op->open(file->f_vnode, filp);
303219820Sjeff		if (error) {
304219820Sjeff			kfree(filp);
305219820Sjeff			return (error);
306219820Sjeff		}
307219820Sjeff	}
308219820Sjeff	error = devfs_set_cdevpriv(filp, linux_file_dtor);
309219820Sjeff	if (error) {
310219820Sjeff		filp->f_op->release(file->f_vnode, filp);
311219820Sjeff		kfree(filp);
312219820Sjeff		return (error);
313219820Sjeff	}
314219820Sjeff
315219820Sjeff	return 0;
316219820Sjeff}
317219820Sjeff
318219820Sjeffstatic int
319219820Sjefflinux_dev_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
320219820Sjeff{
321219820Sjeff	struct linux_cdev *ldev;
322219820Sjeff	struct linux_file *filp;
323219820Sjeff	struct file *file;
324219820Sjeff	int error;
325219820Sjeff
326219820Sjeff	file = curthread->td_fpop;
327219820Sjeff	ldev = dev->si_drv1;
328219820Sjeff	if (ldev == NULL)
329219820Sjeff		return (0);
330219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
331219820Sjeff		return (error);
332219820Sjeff	filp->f_flags = file->f_flag;
333255932Salfred        devfs_clear_cdevpriv();
334255932Salfred
335219820Sjeff
336219820Sjeff	return (0);
337219820Sjeff}
338219820Sjeff
339219820Sjeffstatic int
340219820Sjefflinux_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
341219820Sjeff    struct thread *td)
342219820Sjeff{
343219820Sjeff	struct linux_cdev *ldev;
344219820Sjeff	struct linux_file *filp;
345219820Sjeff	struct file *file;
346219820Sjeff	int error;
347219820Sjeff
348219820Sjeff	file = curthread->td_fpop;
349219820Sjeff	ldev = dev->si_drv1;
350219820Sjeff	if (ldev == NULL)
351219820Sjeff		return (0);
352219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
353219820Sjeff		return (error);
354219820Sjeff	filp->f_flags = file->f_flag;
355219820Sjeff	/*
356219820Sjeff	 * Linux does not have a generic ioctl copyin/copyout layer.  All
357219820Sjeff	 * linux ioctls must be converted to void ioctls which pass a
358219820Sjeff	 * pointer to the address of the data.  We want the actual user
359219820Sjeff	 * address so we dereference here.
360219820Sjeff	 */
361219820Sjeff	data = *(void **)data;
362219820Sjeff	if (filp->f_op->unlocked_ioctl)
363219820Sjeff		error = -filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data);
364219820Sjeff	else
365219820Sjeff		error = ENOTTY;
366219820Sjeff
367219820Sjeff	return (error);
368219820Sjeff}
369219820Sjeff
370219820Sjeffstatic int
371219820Sjefflinux_dev_read(struct cdev *dev, struct uio *uio, int ioflag)
372219820Sjeff{
373219820Sjeff	struct linux_cdev *ldev;
374219820Sjeff	struct linux_file *filp;
375219820Sjeff	struct file *file;
376219820Sjeff	ssize_t bytes;
377219820Sjeff	int error;
378219820Sjeff
379219820Sjeff	file = curthread->td_fpop;
380219820Sjeff	ldev = dev->si_drv1;
381219820Sjeff	if (ldev == NULL)
382219820Sjeff		return (0);
383219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
384219820Sjeff		return (error);
385219820Sjeff	filp->f_flags = file->f_flag;
386219820Sjeff	if (uio->uio_iovcnt != 1)
387219820Sjeff		panic("linux_dev_read: uio %p iovcnt %d",
388219820Sjeff		    uio, uio->uio_iovcnt);
389219820Sjeff	if (filp->f_op->read) {
390219820Sjeff		bytes = filp->f_op->read(filp, uio->uio_iov->iov_base,
391219820Sjeff		    uio->uio_iov->iov_len, &uio->uio_offset);
392219820Sjeff		if (bytes >= 0) {
393219820Sjeff			uio->uio_iov->iov_base += bytes;
394219820Sjeff			uio->uio_iov->iov_len -= bytes;
395219820Sjeff			uio->uio_resid -= bytes;
396219820Sjeff		} else
397219820Sjeff			error = -bytes;
398219820Sjeff	} else
399219820Sjeff		error = ENXIO;
400219820Sjeff
401219820Sjeff	return (error);
402219820Sjeff}
403219820Sjeff
404219820Sjeffstatic int
405219820Sjefflinux_dev_write(struct cdev *dev, struct uio *uio, int ioflag)
406219820Sjeff{
407219820Sjeff	struct linux_cdev *ldev;
408219820Sjeff	struct linux_file *filp;
409219820Sjeff	struct file *file;
410219820Sjeff	ssize_t bytes;
411219820Sjeff	int error;
412219820Sjeff
413219820Sjeff	file = curthread->td_fpop;
414219820Sjeff	ldev = dev->si_drv1;
415219820Sjeff	if (ldev == NULL)
416219820Sjeff		return (0);
417219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
418219820Sjeff		return (error);
419219820Sjeff	filp->f_flags = file->f_flag;
420219820Sjeff	if (uio->uio_iovcnt != 1)
421219820Sjeff		panic("linux_dev_write: uio %p iovcnt %d",
422219820Sjeff		    uio, uio->uio_iovcnt);
423219820Sjeff	if (filp->f_op->write) {
424219820Sjeff		bytes = filp->f_op->write(filp, uio->uio_iov->iov_base,
425219820Sjeff		    uio->uio_iov->iov_len, &uio->uio_offset);
426219820Sjeff		if (bytes >= 0) {
427219820Sjeff			uio->uio_iov->iov_base += bytes;
428219820Sjeff			uio->uio_iov->iov_len -= bytes;
429219820Sjeff			uio->uio_resid -= bytes;
430219820Sjeff		} else
431219820Sjeff			error = -bytes;
432219820Sjeff	} else
433219820Sjeff		error = ENXIO;
434219820Sjeff
435219820Sjeff	return (error);
436219820Sjeff}
437219820Sjeff
438219820Sjeffstatic int
439219820Sjefflinux_dev_poll(struct cdev *dev, int events, struct thread *td)
440219820Sjeff{
441219820Sjeff	struct linux_cdev *ldev;
442219820Sjeff	struct linux_file *filp;
443219820Sjeff	struct file *file;
444219820Sjeff	int revents;
445219820Sjeff	int error;
446219820Sjeff
447219820Sjeff	file = curthread->td_fpop;
448219820Sjeff	ldev = dev->si_drv1;
449219820Sjeff	if (ldev == NULL)
450219820Sjeff		return (0);
451219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
452219820Sjeff		return (error);
453219820Sjeff	filp->f_flags = file->f_flag;
454219820Sjeff	if (filp->f_op->poll)
455219820Sjeff		revents = filp->f_op->poll(filp, NULL) & events;
456219820Sjeff	else
457219820Sjeff		revents = 0;
458219820Sjeff
459219820Sjeff	return (revents);
460219820Sjeff}
461219820Sjeff
462219820Sjeffstatic int
463219820Sjefflinux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
464219820Sjeff    vm_size_t size, struct vm_object **object, int nprot)
465219820Sjeff{
466219820Sjeff	struct linux_cdev *ldev;
467219820Sjeff	struct linux_file *filp;
468219820Sjeff	struct file *file;
469219820Sjeff	struct vm_area_struct vma;
470219820Sjeff	int error;
471219820Sjeff
472219820Sjeff	file = curthread->td_fpop;
473219820Sjeff	ldev = dev->si_drv1;
474219820Sjeff	if (ldev == NULL)
475219820Sjeff		return (ENODEV);
476219820Sjeff	if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
477219820Sjeff		return (error);
478219820Sjeff	filp->f_flags = file->f_flag;
479219820Sjeff	vma.vm_start = 0;
480287637Sjhb	vma.vm_end = size;
481219820Sjeff	vma.vm_pgoff = *offset / PAGE_SIZE;
482219820Sjeff	vma.vm_pfn = 0;
483294636Sjhb	vma.vm_page_prot = VM_MEMATTR_DEFAULT;
484219820Sjeff	if (filp->f_op->mmap) {
485219820Sjeff		error = -filp->f_op->mmap(filp, &vma);
486219820Sjeff		if (error == 0) {
487287637Sjhb			struct sglist *sg;
488287637Sjhb
489287637Sjhb			sg = sglist_alloc(1, M_WAITOK);
490287637Sjhb			sglist_append_phys(sg,
491287637Sjhb			    (vm_paddr_t)vma.vm_pfn << PAGE_SHIFT, vma.vm_len);
492287637Sjhb			*object = vm_pager_allocate(OBJT_SG, sg, vma.vm_len,
493287637Sjhb			    nprot, 0, curthread->td_ucred);
494287637Sjhb		        if (*object == NULL) {
495287637Sjhb				sglist_free(sg);
496287637Sjhb				return (EINVAL);
497287637Sjhb			}
498287637Sjhb			*offset = 0;
499287637Sjhb			if (vma.vm_page_prot != VM_MEMATTR_DEFAULT) {
500287637Sjhb				VM_OBJECT_WLOCK(*object);
501287637Sjhb				vm_object_set_memattr(*object,
502287637Sjhb				    vma.vm_page_prot);
503287637Sjhb				VM_OBJECT_WUNLOCK(*object);
504287637Sjhb			}
505219820Sjeff		}
506219820Sjeff	} else
507219820Sjeff		error = ENODEV;
508219820Sjeff
509219820Sjeff	return (error);
510219820Sjeff}
511219820Sjeff
512219820Sjeffstruct cdevsw linuxcdevsw = {
513219820Sjeff	.d_version = D_VERSION,
514219820Sjeff	.d_flags = D_TRACKCLOSE,
515219820Sjeff	.d_open = linux_dev_open,
516219820Sjeff	.d_close = linux_dev_close,
517219820Sjeff	.d_read = linux_dev_read,
518219820Sjeff	.d_write = linux_dev_write,
519219820Sjeff	.d_ioctl = linux_dev_ioctl,
520219820Sjeff	.d_mmap_single = linux_dev_mmap_single,
521219820Sjeff	.d_poll = linux_dev_poll,
522219820Sjeff};
523219820Sjeff
524219820Sjeffstatic int
525219820Sjefflinux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
526219820Sjeff    int flags, struct thread *td)
527219820Sjeff{
528219820Sjeff	struct linux_file *filp;
529219820Sjeff	ssize_t bytes;
530219820Sjeff	int error;
531219820Sjeff
532219820Sjeff	error = 0;
533219820Sjeff	filp = (struct linux_file *)file->f_data;
534219820Sjeff	filp->f_flags = file->f_flag;
535219820Sjeff	if (uio->uio_iovcnt != 1)
536219820Sjeff		panic("linux_file_read: uio %p iovcnt %d",
537219820Sjeff		    uio, uio->uio_iovcnt);
538219820Sjeff	if (filp->f_op->read) {
539219820Sjeff		bytes = filp->f_op->read(filp, uio->uio_iov->iov_base,
540219820Sjeff		    uio->uio_iov->iov_len, &uio->uio_offset);
541219820Sjeff		if (bytes >= 0) {
542219820Sjeff			uio->uio_iov->iov_base += bytes;
543219820Sjeff			uio->uio_iov->iov_len -= bytes;
544219820Sjeff			uio->uio_resid -= bytes;
545219820Sjeff		} else
546219820Sjeff			error = -bytes;
547219820Sjeff	} else
548219820Sjeff		error = ENXIO;
549219820Sjeff
550219820Sjeff	return (error);
551219820Sjeff}
552219820Sjeff
553219820Sjeffstatic int
554219820Sjefflinux_file_poll(struct file *file, int events, struct ucred *active_cred,
555219820Sjeff    struct thread *td)
556219820Sjeff{
557219820Sjeff	struct linux_file *filp;
558219820Sjeff	int revents;
559219820Sjeff
560219820Sjeff	filp = (struct linux_file *)file->f_data;
561219820Sjeff	filp->f_flags = file->f_flag;
562219820Sjeff	if (filp->f_op->poll)
563219820Sjeff		revents = filp->f_op->poll(filp, NULL) & events;
564219820Sjeff	else
565219820Sjeff		revents = 0;
566219820Sjeff
567219820Sjeff	return (0);
568219820Sjeff}
569219820Sjeff
570219820Sjeffstatic int
571219820Sjefflinux_file_close(struct file *file, struct thread *td)
572219820Sjeff{
573219820Sjeff	struct linux_file *filp;
574219820Sjeff	int error;
575219820Sjeff
576219820Sjeff	filp = (struct linux_file *)file->f_data;
577219820Sjeff	filp->f_flags = file->f_flag;
578219820Sjeff	error = -filp->f_op->release(NULL, filp);
579219820Sjeff	funsetown(&filp->f_sigio);
580219820Sjeff	kfree(filp);
581219820Sjeff
582219820Sjeff	return (error);
583219820Sjeff}
584219820Sjeff
585219820Sjeffstatic int
586219820Sjefflinux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
587219820Sjeff    struct thread *td)
588219820Sjeff{
589219820Sjeff	struct linux_file *filp;
590219820Sjeff	int error;
591219820Sjeff
592219820Sjeff	filp = (struct linux_file *)fp->f_data;
593219820Sjeff	filp->f_flags = fp->f_flag;
594219820Sjeff	error = 0;
595219820Sjeff
596219820Sjeff	switch (cmd) {
597219820Sjeff	case FIONBIO:
598219820Sjeff		break;
599219820Sjeff	case FIOASYNC:
600219820Sjeff		if (filp->f_op->fasync == NULL)
601219820Sjeff			break;
602219820Sjeff		error = filp->f_op->fasync(0, filp, fp->f_flag & FASYNC);
603219820Sjeff		break;
604219820Sjeff	case FIOSETOWN:
605219820Sjeff		error = fsetown(*(int *)data, &filp->f_sigio);
606219820Sjeff		if (error == 0)
607219820Sjeff			error = filp->f_op->fasync(0, filp,
608219820Sjeff			    fp->f_flag & FASYNC);
609219820Sjeff		break;
610219820Sjeff	case FIOGETOWN:
611219820Sjeff		*(int *)data = fgetown(&filp->f_sigio);
612219820Sjeff		break;
613219820Sjeff	default:
614219820Sjeff		error = ENOTTY;
615219820Sjeff		break;
616219820Sjeff	}
617219820Sjeff	return (error);
618219820Sjeff}
619219820Sjeff
620219820Sjeffstruct fileops linuxfileops = {
621219820Sjeff	.fo_read = linux_file_read,
622219820Sjeff	.fo_poll = linux_file_poll,
623219820Sjeff	.fo_close = linux_file_close,
624224914Skib	.fo_ioctl = linux_file_ioctl,
625224914Skib	.fo_chmod = invfo_chmod,
626224914Skib	.fo_chown = invfo_chown,
627254356Sglebius	.fo_sendfile = invfo_sendfile,
628219820Sjeff};
629219820Sjeff
630219820Sjeff/*
631219820Sjeff * Hash of vmmap addresses.  This is infrequently accessed and does not
632219820Sjeff * need to be particularly large.  This is done because we must store the
633219820Sjeff * caller's idea of the map size to properly unmap.
634219820Sjeff */
635219820Sjeffstruct vmmap {
636219820Sjeff	LIST_ENTRY(vmmap)	vm_next;
637219820Sjeff	void 			*vm_addr;
638219820Sjeff	unsigned long		vm_size;
639219820Sjeff};
640219820Sjeff
641282513Shselaskystruct vmmaphd {
642282513Shselasky	struct vmmap *lh_first;
643282513Shselasky};
644219820Sjeff#define	VMMAP_HASH_SIZE	64
645219820Sjeff#define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
646219820Sjeff#define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
647219820Sjeffstatic struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
648219820Sjeffstatic struct mtx vmmaplock;
649219820Sjeff
650219820Sjeffstatic void
651219820Sjeffvmmap_add(void *addr, unsigned long size)
652219820Sjeff{
653219820Sjeff	struct vmmap *vmmap;
654219820Sjeff
655219820Sjeff	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
656219820Sjeff	mtx_lock(&vmmaplock);
657219820Sjeff	vmmap->vm_size = size;
658219820Sjeff	vmmap->vm_addr = addr;
659219820Sjeff	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
660219820Sjeff	mtx_unlock(&vmmaplock);
661219820Sjeff}
662219820Sjeff
663219820Sjeffstatic struct vmmap *
664219820Sjeffvmmap_remove(void *addr)
665219820Sjeff{
666219820Sjeff	struct vmmap *vmmap;
667219820Sjeff
668219820Sjeff	mtx_lock(&vmmaplock);
669219820Sjeff	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
670219820Sjeff		if (vmmap->vm_addr == addr)
671219820Sjeff			break;
672219820Sjeff	if (vmmap)
673219820Sjeff		LIST_REMOVE(vmmap, vm_next);
674219820Sjeff	mtx_unlock(&vmmaplock);
675219820Sjeff
676219820Sjeff	return (vmmap);
677219820Sjeff}
678219820Sjeff
679219820Sjeffvoid *
680219820Sjeff_ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
681219820Sjeff{
682219820Sjeff	void *addr;
683219820Sjeff
684219820Sjeff	addr = pmap_mapdev_attr(phys_addr, size, attr);
685219820Sjeff	if (addr == NULL)
686219820Sjeff		return (NULL);
687219820Sjeff	vmmap_add(addr, size);
688219820Sjeff
689219820Sjeff	return (addr);
690219820Sjeff}
691219820Sjeff
692219820Sjeffvoid
693219820Sjeffiounmap(void *addr)
694219820Sjeff{
695219820Sjeff	struct vmmap *vmmap;
696219820Sjeff
697219820Sjeff	vmmap = vmmap_remove(addr);
698219820Sjeff	if (vmmap == NULL)
699219820Sjeff		return;
700219820Sjeff	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
701219820Sjeff	kfree(vmmap);
702219820Sjeff}
703219820Sjeff
704219820Sjeff
705219820Sjeffvoid *
706219820Sjeffvmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
707219820Sjeff{
708219820Sjeff	vm_offset_t off;
709219820Sjeff	size_t size;
710219820Sjeff
711219820Sjeff	size = count * PAGE_SIZE;
712254025Sjeff	off = kva_alloc(size);
713219820Sjeff	if (off == 0)
714219820Sjeff		return (NULL);
715219820Sjeff	vmmap_add((void *)off, size);
716219820Sjeff	pmap_qenter(off, pages, count);
717219820Sjeff
718219820Sjeff	return ((void *)off);
719219820Sjeff}
720219820Sjeff
721219820Sjeffvoid
722219820Sjeffvunmap(void *addr)
723219820Sjeff{
724219820Sjeff	struct vmmap *vmmap;
725219820Sjeff
726219820Sjeff	vmmap = vmmap_remove(addr);
727219820Sjeff	if (vmmap == NULL)
728219820Sjeff		return;
729219820Sjeff	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
730254025Sjeff	kva_free((vm_offset_t)addr, vmmap->vm_size);
731219820Sjeff	kfree(vmmap);
732219820Sjeff}
733219820Sjeff
734285410Shselaskychar *
735285410Shselaskykvasprintf(gfp_t gfp, const char *fmt, va_list ap)
736285410Shselasky{
737285410Shselasky	unsigned int len;
738285410Shselasky	char *p;
739285410Shselasky	va_list aq;
740282513Shselasky
741285410Shselasky	va_copy(aq, ap);
742285410Shselasky	len = vsnprintf(NULL, 0, fmt, aq);
743285410Shselasky	va_end(aq);
744285410Shselasky
745285410Shselasky	p = kmalloc(len + 1, gfp);
746285410Shselasky	if (p != NULL)
747285410Shselasky		vsnprintf(p, len + 1, fmt, ap);
748285410Shselasky
749285410Shselasky	return (p);
750285410Shselasky}
751285410Shselasky
752282513Shselaskychar *
753282513Shselaskykasprintf(gfp_t gfp, const char *fmt, ...)
754282513Shselasky{
755282513Shselasky	va_list ap;
756282513Shselasky	char *p;
757282513Shselasky
758282513Shselasky	va_start(ap, fmt);
759282513Shselasky	p = kvasprintf(gfp, fmt, ap);
760282513Shselasky	va_end(ap);
761282513Shselasky
762285410Shselasky	return (p);
763282513Shselasky}
764282513Shselasky
765282513Shselaskystatic int
766282513Shselaskylinux_timer_jiffies_until(unsigned long expires)
767282513Shselasky{
768282513Shselasky	int delta = expires - jiffies;
769282513Shselasky	/* guard against already expired values */
770282513Shselasky	if (delta < 1)
771282513Shselasky		delta = 1;
772282513Shselasky	return (delta);
773282513Shselasky}
774282513Shselasky
775219820Sjeffstatic void
776282513Shselaskylinux_timer_callback_wrapper(void *context)
777282513Shselasky{
778282513Shselasky	struct timer_list *timer;
779282513Shselasky
780282513Shselasky	timer = context;
781282513Shselasky	timer->function(timer->data);
782282513Shselasky}
783282513Shselasky
784282513Shselaskyvoid
785282513Shselaskymod_timer(struct timer_list *timer, unsigned long expires)
786282513Shselasky{
787282513Shselasky
788282513Shselasky	timer->expires = expires;
789282513Shselasky	callout_reset(&timer->timer_callout,
790282513Shselasky	    linux_timer_jiffies_until(expires),
791282513Shselasky	    &linux_timer_callback_wrapper, timer);
792282513Shselasky}
793282513Shselasky
794282513Shselaskyvoid
795282513Shselaskyadd_timer(struct timer_list *timer)
796282513Shselasky{
797282513Shselasky
798282513Shselasky	callout_reset(&timer->timer_callout,
799282513Shselasky	    linux_timer_jiffies_until(timer->expires),
800282513Shselasky	    &linux_timer_callback_wrapper, timer);
801282513Shselasky}
802282513Shselasky
803282513Shselaskystatic void
804282513Shselaskylinux_timer_init(void *arg)
805282513Shselasky{
806282513Shselasky
807282513Shselasky	/*
808282513Shselasky	 * Compute an internal HZ value which can divide 2**32 to
809282513Shselasky	 * avoid timer rounding problems when the tick value wraps
810282513Shselasky	 * around 2**32:
811282513Shselasky	 */
812282513Shselasky	linux_timer_hz_mask = 1;
813282513Shselasky	while (linux_timer_hz_mask < (unsigned long)hz)
814282513Shselasky		linux_timer_hz_mask *= 2;
815282513Shselasky	linux_timer_hz_mask--;
816282513Shselasky}
817282513ShselaskySYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
818282513Shselasky
819282513Shselaskyvoid
820282513Shselaskylinux_complete_common(struct completion *c, int all)
821282513Shselasky{
822282513Shselasky	int wakeup_swapper;
823282513Shselasky
824282513Shselasky	sleepq_lock(c);
825282513Shselasky	c->done++;
826282513Shselasky	if (all)
827282513Shselasky		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
828282513Shselasky	else
829282513Shselasky		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
830282513Shselasky	sleepq_release(c);
831282513Shselasky	if (wakeup_swapper)
832282513Shselasky		kick_proc0();
833282513Shselasky}
834282513Shselasky
835282513Shselasky/*
836282513Shselasky * Indefinite wait for done != 0 with or without signals.
837282513Shselasky */
838282513Shselaskylong
839282513Shselaskylinux_wait_for_common(struct completion *c, int flags)
840282513Shselasky{
841282513Shselasky
842282513Shselasky	if (flags != 0)
843282513Shselasky		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
844282513Shselasky	else
845282513Shselasky		flags = SLEEPQ_SLEEP;
846282513Shselasky	for (;;) {
847282513Shselasky		sleepq_lock(c);
848282513Shselasky		if (c->done)
849282513Shselasky			break;
850282513Shselasky		sleepq_add(c, NULL, "completion", flags, 0);
851282513Shselasky		if (flags & SLEEPQ_INTERRUPTIBLE) {
852282513Shselasky			if (sleepq_wait_sig(c, 0) != 0)
853282513Shselasky				return (-ERESTARTSYS);
854282513Shselasky		} else
855282513Shselasky			sleepq_wait(c, 0);
856282513Shselasky	}
857282513Shselasky	c->done--;
858282513Shselasky	sleepq_release(c);
859282513Shselasky
860282513Shselasky	return (0);
861282513Shselasky}
862282513Shselasky
863282513Shselasky/*
864282513Shselasky * Time limited wait for done != 0 with or without signals.
865282513Shselasky */
866282513Shselaskylong
867282513Shselaskylinux_wait_for_timeout_common(struct completion *c, long timeout, int flags)
868282513Shselasky{
869282513Shselasky	long end = jiffies + timeout;
870282513Shselasky
871282513Shselasky	if (flags != 0)
872282513Shselasky		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
873282513Shselasky	else
874282513Shselasky		flags = SLEEPQ_SLEEP;
875282513Shselasky	for (;;) {
876282513Shselasky		int ret;
877282513Shselasky
878282513Shselasky		sleepq_lock(c);
879282513Shselasky		if (c->done)
880282513Shselasky			break;
881282513Shselasky		sleepq_add(c, NULL, "completion", flags, 0);
882282513Shselasky		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
883282513Shselasky		if (flags & SLEEPQ_INTERRUPTIBLE)
884282513Shselasky			ret = sleepq_timedwait_sig(c, 0);
885282513Shselasky		else
886282513Shselasky			ret = sleepq_timedwait(c, 0);
887282513Shselasky		if (ret != 0) {
888282513Shselasky			/* check for timeout or signal */
889282513Shselasky			if (ret == EWOULDBLOCK)
890282513Shselasky				return (0);
891282513Shselasky			else
892282513Shselasky				return (-ERESTARTSYS);
893282513Shselasky		}
894282513Shselasky	}
895282513Shselasky	c->done--;
896282513Shselasky	sleepq_release(c);
897282513Shselasky
898282513Shselasky	/* return how many jiffies are left */
899282513Shselasky	return (linux_timer_jiffies_until(end));
900282513Shselasky}
901282513Shselasky
902282513Shselaskyint
903282513Shselaskylinux_try_wait_for_completion(struct completion *c)
904282513Shselasky{
905282513Shselasky	int isdone;
906282513Shselasky
907282513Shselasky	isdone = 1;
908282513Shselasky	sleepq_lock(c);
909282513Shselasky	if (c->done)
910282513Shselasky		c->done--;
911282513Shselasky	else
912282513Shselasky		isdone = 0;
913282513Shselasky	sleepq_release(c);
914282513Shselasky	return (isdone);
915282513Shselasky}
916282513Shselasky
917282513Shselaskyint
918282513Shselaskylinux_completion_done(struct completion *c)
919282513Shselasky{
920282513Shselasky	int isdone;
921282513Shselasky
922282513Shselasky	isdone = 1;
923282513Shselasky	sleepq_lock(c);
924282513Shselasky	if (c->done == 0)
925282513Shselasky		isdone = 0;
926282513Shselasky	sleepq_release(c);
927282513Shselasky	return (isdone);
928282513Shselasky}
929282513Shselasky
930293151Shselaskyvoid
931293151Shselaskylinux_delayed_work_fn(void *arg)
932293151Shselasky{
933293151Shselasky	struct delayed_work *work;
934293151Shselasky
935293151Shselasky	work = arg;
936293151Shselasky	taskqueue_enqueue(work->work.taskqueue, &work->work.work_task);
937293151Shselasky}
938293151Shselasky
939293151Shselaskyvoid
940293151Shselaskylinux_work_fn(void *context, int pending)
941293151Shselasky{
942293151Shselasky	struct work_struct *work;
943293151Shselasky
944293151Shselasky	work = context;
945293151Shselasky	work->fn(work);
946293151Shselasky}
947293151Shselasky
948293151Shselaskyvoid
949293151Shselaskylinux_flush_fn(void *context, int pending)
950293151Shselasky{
951293151Shselasky}
952293151Shselasky
953293151Shselaskystruct workqueue_struct *
954293151Shselaskylinux_create_workqueue_common(const char *name, int cpus)
955293151Shselasky{
956293151Shselasky	struct workqueue_struct *wq;
957293151Shselasky
958293151Shselasky	wq = kmalloc(sizeof(*wq), M_WAITOK);
959293151Shselasky	wq->taskqueue = taskqueue_create(name, M_WAITOK,
960293151Shselasky	    taskqueue_thread_enqueue,  &wq->taskqueue);
961293151Shselasky	atomic_set(&wq->draining, 0);
962293151Shselasky	taskqueue_start_threads(&wq->taskqueue, cpus, PWAIT, "%s", name);
963293151Shselasky
964293151Shselasky	return (wq);
965293151Shselasky}
966293151Shselasky
967293151Shselaskyvoid
968293151Shselaskydestroy_workqueue(struct workqueue_struct *wq)
969293151Shselasky{
970293151Shselasky	taskqueue_free(wq->taskqueue);
971293151Shselasky	kfree(wq);
972293151Shselasky}
973293151Shselasky
974282513Shselaskystatic void
975280540Shselaskylinux_compat_init(void *arg)
976219820Sjeff{
977219820Sjeff	struct sysctl_oid *rootoid;
978219820Sjeff	int i;
979219820Sjeff
980219820Sjeff	rootoid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(),
981219820Sjeff	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
982219820Sjeff	kobject_init(&class_root, &class_ktype);
983219820Sjeff	kobject_set_name(&class_root, "class");
984219820Sjeff	class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
985219820Sjeff	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
986219820Sjeff	kobject_init(&linux_rootdev.kobj, &dev_ktype);
987219820Sjeff	kobject_set_name(&linux_rootdev.kobj, "device");
988219820Sjeff	linux_rootdev.kobj.oidp = SYSCTL_ADD_NODE(NULL,
989219820Sjeff	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL,
990219820Sjeff	    "device");
991219820Sjeff	linux_rootdev.bsddev = root_bus;
992219820Sjeff	miscclass.name = "misc";
993219820Sjeff	class_register(&miscclass);
994219820Sjeff	INIT_LIST_HEAD(&pci_drivers);
995219820Sjeff	INIT_LIST_HEAD(&pci_devices);
996219820Sjeff	spin_lock_init(&pci_lock);
997219820Sjeff	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
998219820Sjeff	for (i = 0; i < VMMAP_HASH_SIZE; i++)
999219820Sjeff		LIST_INIT(&vmmaphead[i]);
1000219820Sjeff}
1001219820Sjeff
1002219820SjeffSYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
1003271127Shselasky
1004271127Shselaskystatic void
1005280540Shselaskylinux_compat_uninit(void *arg)
1006271127Shselasky{
1007271127Shselasky	kobject_kfree_name(&class_root);
1008271127Shselasky	kobject_kfree_name(&linux_rootdev.kobj);
1009271127Shselasky	kobject_kfree_name(&miscclass.kobj);
1010271127Shselasky}
1011271127ShselaskySYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
1012