Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / staging / vme / devices / vme_user.c
diff --git a/kernel/drivers/staging/vme/devices/vme_user.c b/kernel/drivers/staging/vme/devices/vme_user.c
new file mode 100644 (file)
index 0000000..19ba749
--- /dev/null
@@ -0,0 +1,962 @@
+/*
+ * VMEbus User access driver
+ *
+ * Author: Martyn Welch <martyn.welch@ge.com>
+ * Copyright 2008 GE Intelligent Platforms Embedded Systems, Inc.
+ *
+ * Based on work by:
+ *   Tom Armistead and Ajit Prem
+ *     Copyright 2004 Motorola Inc.
+ *
+ *
+ * This program is free software; you can redistribute  it and/or modify it
+ * under  the terms of  the GNU General  Public License as published by the
+ * Free Software Foundation;  either version 2 of the  License, or (at your
+ * option) any later version.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/atomic.h>
+#include <linux/cdev.h>
+#include <linux/delay.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/ioctl.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/pagemap.h>
+#include <linux/pci.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/syscalls.h>
+#include <linux/types.h>
+
+#include <linux/io.h>
+#include <linux/uaccess.h>
+#include <linux/vme.h>
+
+#include "vme_user.h"
+
+static const char driver_name[] = "vme_user";
+
+static int bus[VME_USER_BUS_MAX];
+static unsigned int bus_num;
+
+/* Currently Documentation/devices.txt defines the following for VME:
+ *
+ * 221 char    VME bus
+ *               0 = /dev/bus/vme/m0           First master image
+ *               1 = /dev/bus/vme/m1           Second master image
+ *               2 = /dev/bus/vme/m2           Third master image
+ *               3 = /dev/bus/vme/m3           Fourth master image
+ *               4 = /dev/bus/vme/s0           First slave image
+ *               5 = /dev/bus/vme/s1           Second slave image
+ *               6 = /dev/bus/vme/s2           Third slave image
+ *               7 = /dev/bus/vme/s3           Fourth slave image
+ *               8 = /dev/bus/vme/ctl          Control
+ *
+ *             It is expected that all VME bus drivers will use the
+ *             same interface.  For interface documentation see
+ *             http://www.vmelinux.org/.
+ *
+ * However the VME driver at http://www.vmelinux.org/ is rather old and doesn't
+ * even support the tsi148 chipset (which has 8 master and 8 slave windows).
+ * We'll run with this for now as far as possible, however it probably makes
+ * sense to get rid of the old mappings and just do everything dynamically.
+ *
+ * So for now, we'll restrict the driver to providing 4 masters and 4 slaves as
+ * defined above and try to support at least some of the interface from
+ * http://www.vmelinux.org/ as an alternative the driver can be written
+ * providing a saner interface later.
+ *
+ * The vmelinux.org driver never supported slave images, the devices reserved
+ * for slaves were repurposed to support all 8 master images on the UniverseII!
+ * We shall support 4 masters and 4 slaves with this driver.
+ */
+#define VME_MAJOR      221     /* VME Major Device Number */
+#define VME_DEVS       9       /* Number of dev entries */
+
+#define MASTER_MINOR   0
+#define MASTER_MAX     3
+#define SLAVE_MINOR    4
+#define SLAVE_MAX      7
+#define CONTROL_MINOR  8
+
+#define PCI_BUF_SIZE  0x20000  /* Size of one slave image buffer */
+
+/*
+ * Structure to handle image related parameters.
+ */
+struct image_desc {
+       void *kern_buf; /* Buffer address in kernel space */
+       dma_addr_t pci_buf;     /* Buffer address in PCI address space */
+       unsigned long long size_buf;    /* Buffer size */
+       struct mutex mutex;     /* Mutex for locking image */
+       struct device *device;  /* Sysfs device */
+       struct vme_resource *resource;  /* VME resource */
+       int users;              /* Number of current users */
+       int mmap_count;         /* Number of current mmap's */
+};
+static struct image_desc image[VME_DEVS];
+
+struct driver_stats {
+       unsigned long reads;
+       unsigned long writes;
+       unsigned long ioctls;
+       unsigned long irqs;
+       unsigned long berrs;
+       unsigned long dmaerrors;
+       unsigned long timeouts;
+       unsigned long external;
+};
+static struct driver_stats statistics;
+
+static struct cdev *vme_user_cdev;             /* Character device */
+static struct class *vme_user_sysfs_class;     /* Sysfs class */
+static struct vme_dev *vme_user_bridge;                /* Pointer to user device */
+
+
+static const int type[VME_DEVS] = {    MASTER_MINOR,   MASTER_MINOR,
+                                       MASTER_MINOR,   MASTER_MINOR,
+                                       SLAVE_MINOR,    SLAVE_MINOR,
+                                       SLAVE_MINOR,    SLAVE_MINOR,
+                                       CONTROL_MINOR
+                               };
+
+
+static int vme_user_open(struct inode *, struct file *);
+static int vme_user_release(struct inode *, struct file *);
+static ssize_t vme_user_read(struct file *, char __user *, size_t, loff_t *);
+static ssize_t vme_user_write(struct file *, const char __user *, size_t,
+       loff_t *);
+static loff_t vme_user_llseek(struct file *, loff_t, int);
+static long vme_user_unlocked_ioctl(struct file *, unsigned int, unsigned long);
+static int vme_user_mmap(struct file *file, struct vm_area_struct *vma);
+
+static void vme_user_vm_open(struct vm_area_struct *vma);
+static void vme_user_vm_close(struct vm_area_struct *vma);
+
+static int vme_user_match(struct vme_dev *);
+static int vme_user_probe(struct vme_dev *);
+static int vme_user_remove(struct vme_dev *);
+
+static const struct file_operations vme_user_fops = {
+       .open = vme_user_open,
+       .release = vme_user_release,
+       .read = vme_user_read,
+       .write = vme_user_write,
+       .llseek = vme_user_llseek,
+       .unlocked_ioctl = vme_user_unlocked_ioctl,
+       .compat_ioctl = vme_user_unlocked_ioctl,
+       .mmap = vme_user_mmap,
+};
+
+struct vme_user_vma_priv {
+       unsigned int minor;
+       atomic_t refcnt;
+};
+
+static const struct vm_operations_struct vme_user_vm_ops = {
+       .open = vme_user_vm_open,
+       .close = vme_user_vm_close,
+};
+
+
+/*
+ * Reset all the statistic counters
+ */
+static void reset_counters(void)
+{
+       statistics.reads = 0;
+       statistics.writes = 0;
+       statistics.ioctls = 0;
+       statistics.irqs = 0;
+       statistics.berrs = 0;
+       statistics.dmaerrors = 0;
+       statistics.timeouts = 0;
+}
+
+static int vme_user_open(struct inode *inode, struct file *file)
+{
+       int err;
+       unsigned int minor = MINOR(inode->i_rdev);
+
+       mutex_lock(&image[minor].mutex);
+       /* Allow device to be opened if a resource is needed and allocated. */
+       if (minor < CONTROL_MINOR && image[minor].resource == NULL) {
+               pr_err("No resources allocated for device\n");
+               err = -EINVAL;
+               goto err_res;
+       }
+
+       /* Increment user count */
+       image[minor].users++;
+
+       mutex_unlock(&image[minor].mutex);
+
+       return 0;
+
+err_res:
+       mutex_unlock(&image[minor].mutex);
+
+       return err;
+}
+
+static int vme_user_release(struct inode *inode, struct file *file)
+{
+       unsigned int minor = MINOR(inode->i_rdev);
+
+       mutex_lock(&image[minor].mutex);
+
+       /* Decrement user count */
+       image[minor].users--;
+
+       mutex_unlock(&image[minor].mutex);
+
+       return 0;
+}
+
+/*
+ * We are going ot alloc a page during init per window for small transfers.
+ * Small transfers will go VME -> buffer -> user space. Larger (more than a
+ * page) transfers will lock the user space buffer into memory and then
+ * transfer the data directly into the user space buffers.
+ */
+static ssize_t resource_to_user(int minor, char __user *buf, size_t count,
+       loff_t *ppos)
+{
+       ssize_t retval;
+       ssize_t copied = 0;
+
+       if (count <= image[minor].size_buf) {
+               /* We copy to kernel buffer */
+               copied = vme_master_read(image[minor].resource,
+                       image[minor].kern_buf, count, *ppos);
+               if (copied < 0)
+                       return (int)copied;
+
+               retval = __copy_to_user(buf, image[minor].kern_buf,
+                       (unsigned long)copied);
+               if (retval != 0) {
+                       copied = (copied - retval);
+                       pr_info("User copy failed\n");
+                       return -EINVAL;
+               }
+
+       } else {
+               /* XXX Need to write this */
+               pr_info("Currently don't support large transfers\n");
+               /* Map in pages from userspace */
+
+               /* Call vme_master_read to do the transfer */
+               return -EINVAL;
+       }
+
+       return copied;
+}
+
+/*
+ * We are going to alloc a page during init per window for small transfers.
+ * Small transfers will go user space -> buffer -> VME. Larger (more than a
+ * page) transfers will lock the user space buffer into memory and then
+ * transfer the data directly from the user space buffers out to VME.
+ */
+static ssize_t resource_from_user(unsigned int minor, const char __user *buf,
+       size_t count, loff_t *ppos)
+{
+       ssize_t retval;
+       ssize_t copied = 0;
+
+       if (count <= image[minor].size_buf) {
+               retval = __copy_from_user(image[minor].kern_buf, buf,
+                       (unsigned long)count);
+               if (retval != 0)
+                       copied = (copied - retval);
+               else
+                       copied = count;
+
+               copied = vme_master_write(image[minor].resource,
+                       image[minor].kern_buf, copied, *ppos);
+       } else {
+               /* XXX Need to write this */
+               pr_info("Currently don't support large transfers\n");
+               /* Map in pages from userspace */
+
+               /* Call vme_master_write to do the transfer */
+               return -EINVAL;
+       }
+
+       return copied;
+}
+
+static ssize_t buffer_to_user(unsigned int minor, char __user *buf,
+       size_t count, loff_t *ppos)
+{
+       void *image_ptr;
+       ssize_t retval;
+
+       image_ptr = image[minor].kern_buf + *ppos;
+
+       retval = __copy_to_user(buf, image_ptr, (unsigned long)count);
+       if (retval != 0) {
+               retval = (count - retval);
+               pr_warn("Partial copy to userspace\n");
+       } else
+               retval = count;
+
+       /* Return number of bytes successfully read */
+       return retval;
+}
+
+static ssize_t buffer_from_user(unsigned int minor, const char __user *buf,
+       size_t count, loff_t *ppos)
+{
+       void *image_ptr;
+       size_t retval;
+
+       image_ptr = image[minor].kern_buf + *ppos;
+
+       retval = __copy_from_user(image_ptr, buf, (unsigned long)count);
+       if (retval != 0) {
+               retval = (count - retval);
+               pr_warn("Partial copy to userspace\n");
+       } else
+               retval = count;
+
+       /* Return number of bytes successfully read */
+       return retval;
+}
+
+static ssize_t vme_user_read(struct file *file, char __user *buf, size_t count,
+                       loff_t *ppos)
+{
+       unsigned int minor = MINOR(file_inode(file)->i_rdev);
+       ssize_t retval;
+       size_t image_size;
+       size_t okcount;
+
+       if (minor == CONTROL_MINOR)
+               return 0;
+
+       mutex_lock(&image[minor].mutex);
+
+       /* XXX Do we *really* want this helper - we can use vme_*_get ? */
+       image_size = vme_get_size(image[minor].resource);
+
+       /* Ensure we are starting at a valid location */
+       if ((*ppos < 0) || (*ppos > (image_size - 1))) {
+               mutex_unlock(&image[minor].mutex);
+               return 0;
+       }
+
+       /* Ensure not reading past end of the image */
+       if (*ppos + count > image_size)
+               okcount = image_size - *ppos;
+       else
+               okcount = count;
+
+       switch (type[minor]) {
+       case MASTER_MINOR:
+               retval = resource_to_user(minor, buf, okcount, ppos);
+               break;
+       case SLAVE_MINOR:
+               retval = buffer_to_user(minor, buf, okcount, ppos);
+               break;
+       default:
+               retval = -EINVAL;
+       }
+
+       mutex_unlock(&image[minor].mutex);
+       if (retval > 0)
+               *ppos += retval;
+
+       return retval;
+}
+
+static ssize_t vme_user_write(struct file *file, const char __user *buf,
+                       size_t count, loff_t *ppos)
+{
+       unsigned int minor = MINOR(file_inode(file)->i_rdev);
+       ssize_t retval;
+       size_t image_size;
+       size_t okcount;
+
+       if (minor == CONTROL_MINOR)
+               return 0;
+
+       mutex_lock(&image[minor].mutex);
+
+       image_size = vme_get_size(image[minor].resource);
+
+       /* Ensure we are starting at a valid location */
+       if ((*ppos < 0) || (*ppos > (image_size - 1))) {
+               mutex_unlock(&image[minor].mutex);
+               return 0;
+       }
+
+       /* Ensure not reading past end of the image */
+       if (*ppos + count > image_size)
+               okcount = image_size - *ppos;
+       else
+               okcount = count;
+
+       switch (type[minor]) {
+       case MASTER_MINOR:
+               retval = resource_from_user(minor, buf, okcount, ppos);
+               break;
+       case SLAVE_MINOR:
+               retval = buffer_from_user(minor, buf, okcount, ppos);
+               break;
+       default:
+               retval = -EINVAL;
+       }
+
+       mutex_unlock(&image[minor].mutex);
+
+       if (retval > 0)
+               *ppos += retval;
+
+       return retval;
+}
+
+static loff_t vme_user_llseek(struct file *file, loff_t off, int whence)
+{
+       unsigned int minor = MINOR(file_inode(file)->i_rdev);
+       size_t image_size;
+       loff_t res;
+
+       if (minor == CONTROL_MINOR)
+               return -EINVAL;
+
+       mutex_lock(&image[minor].mutex);
+       image_size = vme_get_size(image[minor].resource);
+       res = fixed_size_llseek(file, off, whence, image_size);
+       mutex_unlock(&image[minor].mutex);
+
+       return res;
+}
+
+/*
+ * The ioctls provided by the old VME access method (the one at vmelinux.org)
+ * are most certainly wrong as the effectively push the registers layout
+ * through to user space. Given that the VME core can handle multiple bridges,
+ * with different register layouts this is most certainly not the way to go.
+ *
+ * We aren't using the structures defined in the Motorola driver either - these
+ * are also quite low level, however we should use the definitions that have
+ * already been defined.
+ */
+static int vme_user_ioctl(struct inode *inode, struct file *file,
+       unsigned int cmd, unsigned long arg)
+{
+       struct vme_master master;
+       struct vme_slave slave;
+       struct vme_irq_id irq_req;
+       unsigned long copied;
+       unsigned int minor = MINOR(inode->i_rdev);
+       int retval;
+       dma_addr_t pci_addr;
+       void __user *argp = (void __user *)arg;
+
+       statistics.ioctls++;
+
+       switch (type[minor]) {
+       case CONTROL_MINOR:
+               switch (cmd) {
+               case VME_IRQ_GEN:
+                       copied = copy_from_user(&irq_req, argp,
+                                               sizeof(struct vme_irq_id));
+                       if (copied != 0) {
+                               pr_warn("Partial copy from userspace\n");
+                               return -EFAULT;
+                       }
+
+                       return vme_irq_generate(vme_user_bridge,
+                                                 irq_req.level,
+                                                 irq_req.statid);
+               }
+               break;
+       case MASTER_MINOR:
+               switch (cmd) {
+               case VME_GET_MASTER:
+                       memset(&master, 0, sizeof(struct vme_master));
+
+                       /* XXX  We do not want to push aspace, cycle and width
+                        *      to userspace as they are
+                        */
+                       retval = vme_master_get(image[minor].resource,
+                               &master.enable, &master.vme_addr,
+                               &master.size, &master.aspace,
+                               &master.cycle, &master.dwidth);
+
+                       copied = copy_to_user(argp, &master,
+                               sizeof(struct vme_master));
+                       if (copied != 0) {
+                               pr_warn("Partial copy to userspace\n");
+                               return -EFAULT;
+                       }
+
+                       return retval;
+
+               case VME_SET_MASTER:
+
+                       if (image[minor].mmap_count != 0) {
+                               pr_warn("Can't adjust mapped window\n");
+                               return -EPERM;
+                       }
+
+                       copied = copy_from_user(&master, argp, sizeof(master));
+                       if (copied != 0) {
+                               pr_warn("Partial copy from userspace\n");
+                               return -EFAULT;
+                       }
+
+                       /* XXX  We do not want to push aspace, cycle and width
+                        *      to userspace as they are
+                        */
+                       return vme_master_set(image[minor].resource,
+                               master.enable, master.vme_addr, master.size,
+                               master.aspace, master.cycle, master.dwidth);
+
+                       break;
+               }
+               break;
+       case SLAVE_MINOR:
+               switch (cmd) {
+               case VME_GET_SLAVE:
+                       memset(&slave, 0, sizeof(struct vme_slave));
+
+                       /* XXX  We do not want to push aspace, cycle and width
+                        *      to userspace as they are
+                        */
+                       retval = vme_slave_get(image[minor].resource,
+                               &slave.enable, &slave.vme_addr,
+                               &slave.size, &pci_addr, &slave.aspace,
+                               &slave.cycle);
+
+                       copied = copy_to_user(argp, &slave,
+                               sizeof(struct vme_slave));
+                       if (copied != 0) {
+                               pr_warn("Partial copy to userspace\n");
+                               return -EFAULT;
+                       }
+
+                       return retval;
+
+               case VME_SET_SLAVE:
+
+                       copied = copy_from_user(&slave, argp, sizeof(slave));
+                       if (copied != 0) {
+                               pr_warn("Partial copy from userspace\n");
+                               return -EFAULT;
+                       }
+
+                       /* XXX  We do not want to push aspace, cycle and width
+                        *      to userspace as they are
+                        */
+                       return vme_slave_set(image[minor].resource,
+                               slave.enable, slave.vme_addr, slave.size,
+                               image[minor].pci_buf, slave.aspace,
+                               slave.cycle);
+
+                       break;
+               }
+               break;
+       }
+
+       return -EINVAL;
+}
+
+static long
+vme_user_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       int ret;
+       struct inode *inode = file_inode(file);
+       unsigned int minor = MINOR(inode->i_rdev);
+
+       mutex_lock(&image[minor].mutex);
+       ret = vme_user_ioctl(inode, file, cmd, arg);
+       mutex_unlock(&image[minor].mutex);
+
+       return ret;
+}
+
+static void vme_user_vm_open(struct vm_area_struct *vma)
+{
+       struct vme_user_vma_priv *vma_priv = vma->vm_private_data;
+
+       atomic_inc(&vma_priv->refcnt);
+}
+
+static void vme_user_vm_close(struct vm_area_struct *vma)
+{
+       struct vme_user_vma_priv *vma_priv = vma->vm_private_data;
+       unsigned int minor = vma_priv->minor;
+
+       if (!atomic_dec_and_test(&vma_priv->refcnt))
+               return;
+
+       mutex_lock(&image[minor].mutex);
+       image[minor].mmap_count--;
+       mutex_unlock(&image[minor].mutex);
+
+       kfree(vma_priv);
+}
+
+static int vme_user_master_mmap(unsigned int minor, struct vm_area_struct *vma)
+{
+       int err;
+       struct vme_user_vma_priv *vma_priv;
+
+       mutex_lock(&image[minor].mutex);
+
+       err = vme_master_mmap(image[minor].resource, vma);
+       if (err) {
+               mutex_unlock(&image[minor].mutex);
+               return err;
+       }
+
+       vma_priv = kmalloc(sizeof(struct vme_user_vma_priv), GFP_KERNEL);
+       if (vma_priv == NULL) {
+               mutex_unlock(&image[minor].mutex);
+               return -ENOMEM;
+       }
+
+       vma_priv->minor = minor;
+       atomic_set(&vma_priv->refcnt, 1);
+       vma->vm_ops = &vme_user_vm_ops;
+       vma->vm_private_data = vma_priv;
+
+       image[minor].mmap_count++;
+
+       mutex_unlock(&image[minor].mutex);
+
+       return 0;
+}
+
+static int vme_user_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       unsigned int minor = MINOR(file_inode(file)->i_rdev);
+
+       if (type[minor] == MASTER_MINOR)
+               return vme_user_master_mmap(minor, vma);
+
+       return -ENODEV;
+}
+
+
+/*
+ * Unallocate a previously allocated buffer
+ */
+static void buf_unalloc(int num)
+{
+       if (image[num].kern_buf) {
+#ifdef VME_DEBUG
+               pr_debug("UniverseII:Releasing buffer at %p\n",
+                        image[num].pci_buf);
+#endif
+
+               vme_free_consistent(image[num].resource, image[num].size_buf,
+                       image[num].kern_buf, image[num].pci_buf);
+
+               image[num].kern_buf = NULL;
+               image[num].pci_buf = 0;
+               image[num].size_buf = 0;
+
+#ifdef VME_DEBUG
+       } else {
+               pr_debug("UniverseII: Buffer not allocated\n");
+#endif
+       }
+}
+
+static struct vme_driver vme_user_driver = {
+       .name = driver_name,
+       .match = vme_user_match,
+       .probe = vme_user_probe,
+       .remove = vme_user_remove,
+};
+
+
+static int __init vme_user_init(void)
+{
+       int retval = 0;
+
+       pr_info("VME User Space Access Driver\n");
+
+       if (bus_num == 0) {
+               pr_err("No cards, skipping registration\n");
+               retval = -ENODEV;
+               goto err_nocard;
+       }
+
+       /* Let's start by supporting one bus, we can support more than one
+        * in future revisions if that ever becomes necessary.
+        */
+       if (bus_num > VME_USER_BUS_MAX) {
+               pr_err("Driver only able to handle %d buses\n",
+                      VME_USER_BUS_MAX);
+               bus_num = VME_USER_BUS_MAX;
+       }
+
+       /*
+        * Here we just register the maximum number of devices we can and
+        * leave vme_user_match() to allow only 1 to go through to probe().
+        * This way, if we later want to allow multiple user access devices,
+        * we just change the code in vme_user_match().
+        */
+       retval = vme_register_driver(&vme_user_driver, VME_MAX_SLOTS);
+       if (retval != 0)
+               goto err_reg;
+
+       return retval;
+
+err_reg:
+err_nocard:
+       return retval;
+}
+
+static int vme_user_match(struct vme_dev *vdev)
+{
+       int i;
+
+       int cur_bus = vme_bus_num(vdev);
+       int cur_slot = vme_slot_num(vdev);
+
+       for (i = 0; i < bus_num; i++)
+               if ((cur_bus == bus[i]) && (cur_slot == vdev->num))
+                       return 1;
+
+       return 0;
+}
+
+/*
+ * In this simple access driver, the old behaviour is being preserved as much
+ * as practical. We will therefore reserve the buffers and request the images
+ * here so that we don't have to do it later.
+ */
+static int vme_user_probe(struct vme_dev *vdev)
+{
+       int i, err;
+       char *name;
+
+       /* Save pointer to the bridge device */
+       if (vme_user_bridge != NULL) {
+               dev_err(&vdev->dev, "Driver can only be loaded for 1 device\n");
+               err = -EINVAL;
+               goto err_dev;
+       }
+       vme_user_bridge = vdev;
+
+       /* Initialise descriptors */
+       for (i = 0; i < VME_DEVS; i++) {
+               image[i].kern_buf = NULL;
+               image[i].pci_buf = 0;
+               mutex_init(&image[i].mutex);
+               image[i].device = NULL;
+               image[i].resource = NULL;
+               image[i].users = 0;
+       }
+
+       /* Initialise statistics counters */
+       reset_counters();
+
+       /* Assign major and minor numbers for the driver */
+       err = register_chrdev_region(MKDEV(VME_MAJOR, 0), VME_DEVS,
+               driver_name);
+       if (err) {
+               dev_warn(&vdev->dev, "Error getting Major Number %d for driver.\n",
+                        VME_MAJOR);
+               goto err_region;
+       }
+
+       /* Register the driver as a char device */
+       vme_user_cdev = cdev_alloc();
+       if (!vme_user_cdev) {
+               err = -ENOMEM;
+               goto err_char;
+       }
+       vme_user_cdev->ops = &vme_user_fops;
+       vme_user_cdev->owner = THIS_MODULE;
+       err = cdev_add(vme_user_cdev, MKDEV(VME_MAJOR, 0), VME_DEVS);
+       if (err) {
+               dev_warn(&vdev->dev, "cdev_all failed\n");
+               goto err_char;
+       }
+
+       /* Request slave resources and allocate buffers (128kB wide) */
+       for (i = SLAVE_MINOR; i < (SLAVE_MAX + 1); i++) {
+               /* XXX Need to properly request attributes */
+               /* For ca91cx42 bridge there are only two slave windows
+                * supporting A16 addressing, so we request A24 supported
+                * by all windows.
+                */
+               image[i].resource = vme_slave_request(vme_user_bridge,
+                       VME_A24, VME_SCT);
+               if (image[i].resource == NULL) {
+                       dev_warn(&vdev->dev,
+                                "Unable to allocate slave resource\n");
+                       err = -ENOMEM;
+                       goto err_slave;
+               }
+               image[i].size_buf = PCI_BUF_SIZE;
+               image[i].kern_buf = vme_alloc_consistent(image[i].resource,
+                       image[i].size_buf, &image[i].pci_buf);
+               if (image[i].kern_buf == NULL) {
+                       dev_warn(&vdev->dev,
+                                "Unable to allocate memory for buffer\n");
+                       image[i].pci_buf = 0;
+                       vme_slave_free(image[i].resource);
+                       err = -ENOMEM;
+                       goto err_slave;
+               }
+       }
+
+       /*
+        * Request master resources allocate page sized buffers for small
+        * reads and writes
+        */
+       for (i = MASTER_MINOR; i < (MASTER_MAX + 1); i++) {
+               /* XXX Need to properly request attributes */
+               image[i].resource = vme_master_request(vme_user_bridge,
+                       VME_A32, VME_SCT, VME_D32);
+               if (image[i].resource == NULL) {
+                       dev_warn(&vdev->dev,
+                                "Unable to allocate master resource\n");
+                       err = -ENOMEM;
+                       goto err_master;
+               }
+               image[i].size_buf = PCI_BUF_SIZE;
+               image[i].kern_buf = kmalloc(image[i].size_buf, GFP_KERNEL);
+               if (image[i].kern_buf == NULL) {
+                       err = -ENOMEM;
+                       vme_master_free(image[i].resource);
+                       goto err_master;
+               }
+       }
+
+       /* Create sysfs entries - on udev systems this creates the dev files */
+       vme_user_sysfs_class = class_create(THIS_MODULE, driver_name);
+       if (IS_ERR(vme_user_sysfs_class)) {
+               dev_err(&vdev->dev, "Error creating vme_user class.\n");
+               err = PTR_ERR(vme_user_sysfs_class);
+               goto err_class;
+       }
+
+       /* Add sysfs Entries */
+       for (i = 0; i < VME_DEVS; i++) {
+               int num;
+
+               switch (type[i]) {
+               case MASTER_MINOR:
+                       name = "bus/vme/m%d";
+                       break;
+               case CONTROL_MINOR:
+                       name = "bus/vme/ctl";
+                       break;
+               case SLAVE_MINOR:
+                       name = "bus/vme/s%d";
+                       break;
+               default:
+                       err = -EINVAL;
+                       goto err_sysfs;
+               }
+
+               num = (type[i] == SLAVE_MINOR) ? i - (MASTER_MAX + 1) : i;
+               image[i].device = device_create(vme_user_sysfs_class, NULL,
+                                       MKDEV(VME_MAJOR, i), NULL, name, num);
+               if (IS_ERR(image[i].device)) {
+                       dev_info(&vdev->dev, "Error creating sysfs device\n");
+                       err = PTR_ERR(image[i].device);
+                       goto err_sysfs;
+               }
+       }
+
+       return 0;
+
+err_sysfs:
+       while (i > 0) {
+               i--;
+               device_destroy(vme_user_sysfs_class, MKDEV(VME_MAJOR, i));
+       }
+       class_destroy(vme_user_sysfs_class);
+
+       /* Ensure counter set correcty to unalloc all master windows */
+       i = MASTER_MAX + 1;
+err_master:
+       while (i > MASTER_MINOR) {
+               i--;
+               kfree(image[i].kern_buf);
+               vme_master_free(image[i].resource);
+       }
+
+       /*
+        * Ensure counter set correcty to unalloc all slave windows and buffers
+        */
+       i = SLAVE_MAX + 1;
+err_slave:
+       while (i > SLAVE_MINOR) {
+               i--;
+               buf_unalloc(i);
+               vme_slave_free(image[i].resource);
+       }
+err_class:
+       cdev_del(vme_user_cdev);
+err_char:
+       unregister_chrdev_region(MKDEV(VME_MAJOR, 0), VME_DEVS);
+err_region:
+err_dev:
+       return err;
+}
+
+static int vme_user_remove(struct vme_dev *dev)
+{
+       int i;
+
+       /* Remove sysfs Entries */
+       for (i = 0; i < VME_DEVS; i++) {
+               mutex_destroy(&image[i].mutex);
+               device_destroy(vme_user_sysfs_class, MKDEV(VME_MAJOR, i));
+       }
+       class_destroy(vme_user_sysfs_class);
+
+       for (i = MASTER_MINOR; i < (MASTER_MAX + 1); i++) {
+               kfree(image[i].kern_buf);
+               vme_master_free(image[i].resource);
+       }
+
+       for (i = SLAVE_MINOR; i < (SLAVE_MAX + 1); i++) {
+               vme_slave_set(image[i].resource, 0, 0, 0, 0, VME_A32, 0);
+               buf_unalloc(i);
+               vme_slave_free(image[i].resource);
+       }
+
+       /* Unregister device driver */
+       cdev_del(vme_user_cdev);
+
+       /* Unregiser the major and minor device numbers */
+       unregister_chrdev_region(MKDEV(VME_MAJOR, 0), VME_DEVS);
+
+       return 0;
+}
+
+static void __exit vme_user_exit(void)
+{
+       vme_unregister_driver(&vme_user_driver);
+}
+
+
+MODULE_PARM_DESC(bus, "Enumeration of VMEbus to which the driver is connected");
+module_param_array(bus, int, &bus_num, 0);
+
+MODULE_DESCRIPTION("VME User Space Access Driver");
+MODULE_AUTHOR("Martyn Welch <martyn.welch@ge.com");
+MODULE_LICENSE("GPL");
+
+module_init(vme_user_init);
+module_exit(vme_user_exit);