linux/drivers/uio/uio.c

928 lines
20 KiB
C
Raw Normal View History

/*
* drivers/uio/uio.c
*
* Copyright(C) 2005, Benedikt Spranger <b.spranger@linutronix.de>
* Copyright(C) 2005, Thomas Gleixner <tglx@linutronix.de>
* Copyright(C) 2006, Hans J. Koch <hjk@linutronix.de>
* Copyright(C) 2006, Greg Kroah-Hartman <greg@kroah.com>
*
* Userspace IO
*
* Base Functions
*
* Licensed under the GPLv2 only.
*/
#include <linux/module.h>
#include <linux/init.h>
#include <linux/poll.h>
#include <linux/device.h>
#include <linux/mm.h>
#include <linux/idr.h>
#include <linux/string.h>
#include <linux/kobject.h>
#include <linux/uio_driver.h>
#define UIO_MAX_DEVICES 255
struct uio_device {
struct module *owner;
struct device *dev;
int minor;
atomic_t event;
struct fasync_struct *async_queue;
wait_queue_head_t wait;
int vma_count;
struct uio_info *info;
struct kobject *map_dir;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct kobject *portio_dir;
};
static int uio_major;
static DEFINE_IDR(uio_idr);
static const struct file_operations uio_fops;
/* UIO class infrastructure */
static struct uio_class {
struct kref kref;
struct class *class;
} *uio_class;
/* Protect idr accesses */
static DEFINE_MUTEX(minor_lock);
/*
* attributes
*/
struct uio_map {
struct kobject kobj;
struct uio_mem *mem;
};
#define to_map(map) container_of(map, struct uio_map, kobj)
static ssize_t map_name_show(struct uio_mem *mem, char *buf)
{
if (unlikely(!mem->name))
mem->name = "";
return sprintf(buf, "%s\n", mem->name);
}
static ssize_t map_addr_show(struct uio_mem *mem, char *buf)
{
return sprintf(buf, "0x%lx\n", mem->addr);
}
static ssize_t map_size_show(struct uio_mem *mem, char *buf)
{
return sprintf(buf, "0x%lx\n", mem->size);
}
static ssize_t map_offset_show(struct uio_mem *mem, char *buf)
{
return sprintf(buf, "0x%lx\n", mem->addr & ~PAGE_MASK);
}
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct map_sysfs_entry {
struct attribute attr;
ssize_t (*show)(struct uio_mem *, char *);
ssize_t (*store)(struct uio_mem *, const char *, size_t);
};
static struct map_sysfs_entry name_attribute =
__ATTR(name, S_IRUGO, map_name_show, NULL);
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static struct map_sysfs_entry addr_attribute =
__ATTR(addr, S_IRUGO, map_addr_show, NULL);
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static struct map_sysfs_entry size_attribute =
__ATTR(size, S_IRUGO, map_size_show, NULL);
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static struct map_sysfs_entry offset_attribute =
__ATTR(offset, S_IRUGO, map_offset_show, NULL);
static struct attribute *attrs[] = {
&name_attribute.attr,
&addr_attribute.attr,
&size_attribute.attr,
&offset_attribute.attr,
NULL, /* need to NULL terminate the list of attributes */
};
static void map_release(struct kobject *kobj)
{
struct uio_map *map = to_map(kobj);
kfree(map);
}
static ssize_t map_type_show(struct kobject *kobj, struct attribute *attr,
char *buf)
{
struct uio_map *map = to_map(kobj);
struct uio_mem *mem = map->mem;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct map_sysfs_entry *entry;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
entry = container_of(attr, struct map_sysfs_entry, attr);
if (!entry->show)
return -EIO;
return entry->show(mem, buf);
}
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static struct sysfs_ops map_sysfs_ops = {
.show = map_type_show,
};
static struct kobj_type map_attr_type = {
.release = map_release,
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
.sysfs_ops = &map_sysfs_ops,
.default_attrs = attrs,
};
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct uio_portio {
struct kobject kobj;
struct uio_port *port;
};
#define to_portio(portio) container_of(portio, struct uio_portio, kobj)
static ssize_t portio_name_show(struct uio_port *port, char *buf)
{
if (unlikely(!port->name))
port->name = "";
return sprintf(buf, "%s\n", port->name);
}
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static ssize_t portio_start_show(struct uio_port *port, char *buf)
{
return sprintf(buf, "0x%lx\n", port->start);
}
static ssize_t portio_size_show(struct uio_port *port, char *buf)
{
return sprintf(buf, "0x%lx\n", port->size);
}
static ssize_t portio_porttype_show(struct uio_port *port, char *buf)
{
const char *porttypes[] = {"none", "x86", "gpio", "other"};
if ((port->porttype < 0) || (port->porttype > UIO_PORT_OTHER))
return -EINVAL;
return sprintf(buf, "port_%s\n", porttypes[port->porttype]);
}
struct portio_sysfs_entry {
struct attribute attr;
ssize_t (*show)(struct uio_port *, char *);
ssize_t (*store)(struct uio_port *, const char *, size_t);
};
static struct portio_sysfs_entry portio_name_attribute =
__ATTR(name, S_IRUGO, portio_name_show, NULL);
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
static struct portio_sysfs_entry portio_start_attribute =
__ATTR(start, S_IRUGO, portio_start_show, NULL);
static struct portio_sysfs_entry portio_size_attribute =
__ATTR(size, S_IRUGO, portio_size_show, NULL);
static struct portio_sysfs_entry portio_porttype_attribute =
__ATTR(porttype, S_IRUGO, portio_porttype_show, NULL);
static struct attribute *portio_attrs[] = {
&portio_name_attribute.attr,
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
&portio_start_attribute.attr,
&portio_size_attribute.attr,
&portio_porttype_attribute.attr,
NULL,
};
static void portio_release(struct kobject *kobj)
{
struct uio_portio *portio = to_portio(kobj);
kfree(portio);
}
static ssize_t portio_type_show(struct kobject *kobj, struct attribute *attr,
char *buf)
{
struct uio_portio *portio = to_portio(kobj);
struct uio_port *port = portio->port;
struct portio_sysfs_entry *entry;
entry = container_of(attr, struct portio_sysfs_entry, attr);
if (!entry->show)
return -EIO;
return entry->show(port, buf);
}
static struct sysfs_ops portio_sysfs_ops = {
.show = portio_type_show,
};
static struct kobj_type portio_attr_type = {
.release = portio_release,
.sysfs_ops = &portio_sysfs_ops,
.default_attrs = portio_attrs,
};
static ssize_t show_name(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct uio_device *idev = dev_get_drvdata(dev);
if (idev)
return sprintf(buf, "%s\n", idev->info->name);
else
return -ENODEV;
}
static DEVICE_ATTR(name, S_IRUGO, show_name, NULL);
static ssize_t show_version(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct uio_device *idev = dev_get_drvdata(dev);
if (idev)
return sprintf(buf, "%s\n", idev->info->version);
else
return -ENODEV;
}
static DEVICE_ATTR(version, S_IRUGO, show_version, NULL);
static ssize_t show_event(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct uio_device *idev = dev_get_drvdata(dev);
if (idev)
return sprintf(buf, "%u\n",
(unsigned int)atomic_read(&idev->event));
else
return -ENODEV;
}
static DEVICE_ATTR(event, S_IRUGO, show_event, NULL);
static struct attribute *uio_attrs[] = {
&dev_attr_name.attr,
&dev_attr_version.attr,
&dev_attr_event.attr,
NULL,
};
static struct attribute_group uio_attr_grp = {
.attrs = uio_attrs,
};
/*
* device functions
*/
static int uio_dev_add_attributes(struct uio_device *idev)
{
int ret;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
int mi, pi;
int map_found = 0;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
int portio_found = 0;
struct uio_mem *mem;
struct uio_map *map;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct uio_port *port;
struct uio_portio *portio;
ret = sysfs_create_group(&idev->dev->kobj, &uio_attr_grp);
if (ret)
goto err_group;
for (mi = 0; mi < MAX_UIO_MAPS; mi++) {
mem = &idev->info->mem[mi];
if (mem->size == 0)
break;
if (!map_found) {
map_found = 1;
idev->map_dir = kobject_create_and_add("maps",
&idev->dev->kobj);
if (!idev->map_dir)
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
goto err_map;
}
map = kzalloc(sizeof(*map), GFP_KERNEL);
if (!map)
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
goto err_map;
kobject_init(&map->kobj, &map_attr_type);
map->mem = mem;
mem->map = map;
ret = kobject_add(&map->kobj, idev->map_dir, "map%d", mi);
if (ret)
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
goto err_map;
ret = kobject_uevent(&map->kobj, KOBJ_ADD);
if (ret)
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
goto err_map;
}
for (pi = 0; pi < MAX_UIO_PORT_REGIONS; pi++) {
port = &idev->info->port[pi];
if (port->size == 0)
break;
if (!portio_found) {
portio_found = 1;
idev->portio_dir = kobject_create_and_add("portio",
&idev->dev->kobj);
if (!idev->portio_dir)
goto err_portio;
}
portio = kzalloc(sizeof(*portio), GFP_KERNEL);
if (!portio)
goto err_portio;
kobject_init(&portio->kobj, &portio_attr_type);
portio->port = port;
port->portio = portio;
ret = kobject_add(&portio->kobj, idev->portio_dir,
"port%d", pi);
if (ret)
goto err_portio;
ret = kobject_uevent(&portio->kobj, KOBJ_ADD);
if (ret)
goto err_portio;
}
return 0;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
err_portio:
for (pi--; pi >= 0; pi--) {
port = &idev->info->port[pi];
portio = port->portio;
kobject_put(&portio->kobj);
}
kobject_put(idev->portio_dir);
err_map:
for (mi--; mi>=0; mi--) {
mem = &idev->info->mem[mi];
map = mem->map;
kobject_put(&map->kobj);
}
kobject_put(idev->map_dir);
sysfs_remove_group(&idev->dev->kobj, &uio_attr_grp);
err_group:
dev_err(idev->dev, "error creating sysfs files (%d)\n", ret);
return ret;
}
static void uio_dev_del_attributes(struct uio_device *idev)
{
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
int i;
struct uio_mem *mem;
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
struct uio_port *port;
for (i = 0; i < MAX_UIO_MAPS; i++) {
mem = &idev->info->mem[i];
if (mem->size == 0)
break;
kobject_put(&mem->map->kobj);
}
kobject_put(idev->map_dir);
UIO: Pass information about ioports to userspace (V2) Devices sometimes have memory where all or parts of it can not be mapped to userspace. But it might still be possible to access this memory from userspace by other means. An example are PCI cards that advertise not only mappable memory but also ioport ranges. On x86 architectures, these can be accessed with ioperm, iopl, inb, outb, and friends. Mike Frysinger (CCed) reported a similar problem on Blackfin arch where it doesn't seem to be easy to mmap non-cached memory but it can still be accessed from userspace. This patch allows kernel drivers to pass information about such ports to userspace. Similar to the existing mem[] array, it adds a port[] array to struct uio_info. Each port range is described by start, size, and porttype. If a driver fills in at least one such port range, the UIO core will simply pass this information to userspace by creating a new directory "portio" underneath /sys/class/uio/uioN/. Similar to the "mem" directory, it will contain a subdirectory (portX) for each port range given. Note that UIO simply passes this information to userspace, it performs no action whatsoever with this data. It's userspace's responsibility to obtain access to these ports and to solve arch dependent issues. The "porttype" attribute tells userspace what kind of port it is dealing with. This mechanism could also be used to give userspace information about GPIOs related to a device. You frequently find such hardware in embedded devices, so I added a UIO_PORT_GPIO definition. I'm not really sure if this is a good idea since there are other solutions to this problem, but it won't hurt much anyway. Signed-off-by: Hans J. Koch <hjk@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2008-12-06 01:23:13 +00:00
for (i = 0; i < MAX_UIO_PORT_REGIONS; i++) {
port = &idev->info->port[i];
if (port->size == 0)
break;
kobject_put(&port->portio->kobj);
}
kobject_put(idev->portio_dir);
sysfs_remove_group(&idev->dev->kobj, &uio_attr_grp);
}
static int uio_get_minor(struct uio_device *idev)
{
int retval = -ENOMEM;
int id;
mutex_lock(&minor_lock);
if (idr_pre_get(&uio_idr, GFP_KERNEL) == 0)
goto exit;
retval = idr_get_new(&uio_idr, idev, &id);
if (retval < 0) {
if (retval == -EAGAIN)
retval = -ENOMEM;
goto exit;
}
idev->minor = id & MAX_ID_MASK;
exit:
mutex_unlock(&minor_lock);
return retval;
}
static void uio_free_minor(struct uio_device *idev)
{
mutex_lock(&minor_lock);
idr_remove(&uio_idr, idev->minor);
mutex_unlock(&minor_lock);
}
/**
* uio_event_notify - trigger an interrupt event
* @info: UIO device capabilities
*/
void uio_event_notify(struct uio_info *info)
{
struct uio_device *idev = info->uio_dev;
atomic_inc(&idev->event);
wake_up_interruptible(&idev->wait);
kill_fasync(&idev->async_queue, SIGIO, POLL_IN);
}
EXPORT_SYMBOL_GPL(uio_event_notify);
/**
* uio_interrupt - hardware interrupt handler
* @irq: IRQ number, can be UIO_IRQ_CYCLIC for cyclic timer
* @dev_id: Pointer to the devices uio_device structure
*/
static irqreturn_t uio_interrupt(int irq, void *dev_id)
{
struct uio_device *idev = (struct uio_device *)dev_id;
irqreturn_t ret = idev->info->handler(irq, idev->info);
if (ret == IRQ_HANDLED)
uio_event_notify(idev->info);
return ret;
}
struct uio_listener {
struct uio_device *dev;
s32 event_count;
};
static int uio_open(struct inode *inode, struct file *filep)
{
struct uio_device *idev;
struct uio_listener *listener;
int ret = 0;
mutex_lock(&minor_lock);
idev = idr_find(&uio_idr, iminor(inode));
mutex_unlock(&minor_lock);
if (!idev) {
ret = -ENODEV;
goto out;
}
if (!try_module_get(idev->owner)) {
ret = -ENODEV;
goto out;
}
listener = kmalloc(sizeof(*listener), GFP_KERNEL);
if (!listener) {
ret = -ENOMEM;
goto err_alloc_listener;
}
listener->dev = idev;
listener->event_count = atomic_read(&idev->event);
filep->private_data = listener;
if (idev->info->open) {
ret = idev->info->open(idev->info, inode);
if (ret)
goto err_infoopen;
}
return 0;
err_infoopen:
kfree(listener);
err_alloc_listener:
module_put(idev->owner);
out:
return ret;
}
static int uio_fasync(int fd, struct file *filep, int on)
{
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
return fasync_helper(fd, filep, on, &idev->async_queue);
}
static int uio_release(struct inode *inode, struct file *filep)
{
int ret = 0;
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
if (idev->info->release)
ret = idev->info->release(idev->info, inode);
module_put(idev->owner);
kfree(listener);
return ret;
}
static unsigned int uio_poll(struct file *filep, poll_table *wait)
{
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
if (idev->info->irq == UIO_IRQ_NONE)
return -EIO;
poll_wait(filep, &idev->wait, wait);
if (listener->event_count != atomic_read(&idev->event))
return POLLIN | POLLRDNORM;
return 0;
}
static ssize_t uio_read(struct file *filep, char __user *buf,
size_t count, loff_t *ppos)
{
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
DECLARE_WAITQUEUE(wait, current);
ssize_t retval;
s32 event_count;
if (idev->info->irq == UIO_IRQ_NONE)
return -EIO;
if (count != sizeof(s32))
return -EINVAL;
add_wait_queue(&idev->wait, &wait);
do {
set_current_state(TASK_INTERRUPTIBLE);
event_count = atomic_read(&idev->event);
if (event_count != listener->event_count) {
if (copy_to_user(buf, &event_count, count))
retval = -EFAULT;
else {
listener->event_count = event_count;
retval = count;
}
break;
}
if (filep->f_flags & O_NONBLOCK) {
retval = -EAGAIN;
break;
}
if (signal_pending(current)) {
retval = -ERESTARTSYS;
break;
}
schedule();
} while (1);
__set_current_state(TASK_RUNNING);
remove_wait_queue(&idev->wait, &wait);
return retval;
}
static ssize_t uio_write(struct file *filep, const char __user *buf,
size_t count, loff_t *ppos)
{
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
ssize_t retval;
s32 irq_on;
if (idev->info->irq == UIO_IRQ_NONE)
return -EIO;
if (count != sizeof(s32))
return -EINVAL;
if (!idev->info->irqcontrol)
return -ENOSYS;
if (copy_from_user(&irq_on, buf, count))
return -EFAULT;
retval = idev->info->irqcontrol(idev->info, irq_on);
return retval ? retval : sizeof(s32);
}
static int uio_find_mem_index(struct vm_area_struct *vma)
{
int mi;
struct uio_device *idev = vma->vm_private_data;
for (mi = 0; mi < MAX_UIO_MAPS; mi++) {
if (idev->info->mem[mi].size == 0)
return -1;
if (vma->vm_pgoff == mi)
return mi;
}
return -1;
}
static void uio_vma_open(struct vm_area_struct *vma)
{
struct uio_device *idev = vma->vm_private_data;
idev->vma_count++;
}
static void uio_vma_close(struct vm_area_struct *vma)
{
struct uio_device *idev = vma->vm_private_data;
idev->vma_count--;
}
static int uio_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
struct uio_device *idev = vma->vm_private_data;
struct page *page;
unsigned long offset;
int mi = uio_find_mem_index(vma);
if (mi < 0)
return VM_FAULT_SIGBUS;
/*
* We need to subtract mi because userspace uses offset = N*PAGE_SIZE
* to use mem[N].
*/
offset = (vmf->pgoff - mi) << PAGE_SHIFT;
if (idev->info->mem[mi].memtype == UIO_MEM_LOGICAL)
page = virt_to_page(idev->info->mem[mi].addr + offset);
else
page = vmalloc_to_page((void *)idev->info->mem[mi].addr
+ offset);
get_page(page);
vmf->page = page;
return 0;
}
static const struct vm_operations_struct uio_vm_ops = {
.open = uio_vma_open,
.close = uio_vma_close,
.fault = uio_vma_fault,
};
static int uio_mmap_physical(struct vm_area_struct *vma)
{
struct uio_device *idev = vma->vm_private_data;
int mi = uio_find_mem_index(vma);
if (mi < 0)
return -EINVAL;
vma->vm_flags |= VM_IO | VM_RESERVED;
vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
return remap_pfn_range(vma,
vma->vm_start,
idev->info->mem[mi].addr >> PAGE_SHIFT,
vma->vm_end - vma->vm_start,
vma->vm_page_prot);
}
static int uio_mmap_logical(struct vm_area_struct *vma)
{
vma->vm_flags |= VM_RESERVED;
vma->vm_ops = &uio_vm_ops;
uio_vma_open(vma);
return 0;
}
static int uio_mmap(struct file *filep, struct vm_area_struct *vma)
{
struct uio_listener *listener = filep->private_data;
struct uio_device *idev = listener->dev;
int mi;
unsigned long requested_pages, actual_pages;
int ret = 0;
if (vma->vm_end < vma->vm_start)
return -EINVAL;
vma->vm_private_data = idev;
mi = uio_find_mem_index(vma);
if (mi < 0)
return -EINVAL;
requested_pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
actual_pages = ((idev->info->mem[mi].addr & ~PAGE_MASK)
+ idev->info->mem[mi].size + PAGE_SIZE -1) >> PAGE_SHIFT;
if (requested_pages > actual_pages)
return -EINVAL;
if (idev->info->mmap) {
ret = idev->info->mmap(idev->info, vma);
return ret;
}
switch (idev->info->mem[mi].memtype) {
case UIO_MEM_PHYS:
return uio_mmap_physical(vma);
case UIO_MEM_LOGICAL:
case UIO_MEM_VIRTUAL:
return uio_mmap_logical(vma);
default:
return -EINVAL;
}
}
static const struct file_operations uio_fops = {
.owner = THIS_MODULE,
.open = uio_open,
.release = uio_release,
.read = uio_read,
.write = uio_write,
.mmap = uio_mmap,
.poll = uio_poll,
.fasync = uio_fasync,
};
static int uio_major_init(void)
{
uio_major = register_chrdev(0, "uio", &uio_fops);
if (uio_major < 0)
return uio_major;
return 0;
}
static void uio_major_cleanup(void)
{
unregister_chrdev(uio_major, "uio");
}
static int init_uio_class(void)
{
int ret = 0;
if (uio_class != NULL) {
kref_get(&uio_class->kref);
goto exit;
}
/* This is the first time in here, set everything up properly */
ret = uio_major_init();
if (ret)
goto exit;
uio_class = kzalloc(sizeof(*uio_class), GFP_KERNEL);
if (!uio_class) {
ret = -ENOMEM;
goto err_kzalloc;
}
kref_init(&uio_class->kref);
uio_class->class = class_create(THIS_MODULE, "uio");
if (IS_ERR(uio_class->class)) {
ret = IS_ERR(uio_class->class);
printk(KERN_ERR "class_create failed for uio\n");
goto err_class_create;
}
return 0;
err_class_create:
kfree(uio_class);
uio_class = NULL;
err_kzalloc:
uio_major_cleanup();
exit:
return ret;
}
static void release_uio_class(struct kref *kref)
{
/* Ok, we cheat as we know we only have one uio_class */
class_destroy(uio_class->class);
kfree(uio_class);
uio_major_cleanup();
uio_class = NULL;
}
static void uio_class_destroy(void)
{
if (uio_class)
kref_put(&uio_class->kref, release_uio_class);
}
/**
* uio_register_device - register a new userspace IO device
* @owner: module that creates the new device
* @parent: parent device
* @info: UIO device capabilities
*
* returns zero on success or a negative error code.
*/
int __uio_register_device(struct module *owner,
struct device *parent,
struct uio_info *info)
{
struct uio_device *idev;
int ret = 0;
if (!parent || !info || !info->name || !info->version)
return -EINVAL;
info->uio_dev = NULL;
ret = init_uio_class();
if (ret)
return ret;
idev = kzalloc(sizeof(*idev), GFP_KERNEL);
if (!idev) {
ret = -ENOMEM;
goto err_kzalloc;
}
idev->owner = owner;
idev->info = info;
init_waitqueue_head(&idev->wait);
atomic_set(&idev->event, 0);
ret = uio_get_minor(idev);
if (ret)
goto err_get_minor;
idev->dev = device_create(uio_class->class, parent,
MKDEV(uio_major, idev->minor), idev,
"uio%d", idev->minor);
if (IS_ERR(idev->dev)) {
printk(KERN_ERR "UIO: device register failed\n");
ret = PTR_ERR(idev->dev);
goto err_device_create;
}
ret = uio_dev_add_attributes(idev);
if (ret)
goto err_uio_dev_add_attributes;
info->uio_dev = idev;
if (idev->info->irq >= 0) {
ret = request_irq(idev->info->irq, uio_interrupt,
idev->info->irq_flags, idev->info->name, idev);
if (ret)
goto err_request_irq;
}
return 0;
err_request_irq:
uio_dev_del_attributes(idev);
err_uio_dev_add_attributes:
device_destroy(uio_class->class, MKDEV(uio_major, idev->minor));
err_device_create:
uio_free_minor(idev);
err_get_minor:
kfree(idev);
err_kzalloc:
uio_class_destroy();
return ret;
}
EXPORT_SYMBOL_GPL(__uio_register_device);
/**
* uio_unregister_device - unregister a industrial IO device
* @info: UIO device capabilities
*
*/
void uio_unregister_device(struct uio_info *info)
{
struct uio_device *idev;
if (!info || !info->uio_dev)
return;
idev = info->uio_dev;
uio_free_minor(idev);
if (info->irq >= 0)
free_irq(info->irq, idev);
uio_dev_del_attributes(idev);
dev_set_drvdata(idev->dev, NULL);
device_destroy(uio_class->class, MKDEV(uio_major, idev->minor));
kfree(idev);
uio_class_destroy();
return;
}
EXPORT_SYMBOL_GPL(uio_unregister_device);
static int __init uio_init(void)
{
return 0;
}
static void __exit uio_exit(void)
{
}
module_init(uio_init)
module_exit(uio_exit)
MODULE_LICENSE("GPL v2");