--- /dev/null
+/* IBM POWER Barrier Synchronization Register Driver
+ *
+ * Copyright IBM Corporation 2008
+ *
+ * Author: Sonny Rao <sonnyrao@us.ibm.com>
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#include <linux/kernel.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/of_platform.h>
+#include <linux/module.h>
+#include <linux/cdev.h>
+#include <linux/list.h>
+#include <linux/mm.h>
+#include <asm/io.h>
+
+/*
+ This driver exposes a special register which can be used for fast
+ synchronization across a large SMP machine.  The hardware is exposed
+ as an array of bytes where each process will write to one of the bytes to
+ indicate it has finished the current stage and this update is broadcast to
+ all processors without having to bounce a cacheline between them. In
+ POWER5 and POWER6 there is one of these registers per SMP,  but it is
+ presented in two forms; first, it is given as a whole and then as a number
+ of smaller registers which alias to parts of the single whole register.
+ This can potentially allow multiple groups of processes to each have their
+ own private synchronization device.
+
+ Note that this hardware *must* be written to using *only* single byte writes.
+ It may be read using 1, 2, 4, or 8 byte loads which must be aligned since
+ this region is treated as cache-inhibited  processes should also use a
+ full sync before and after writing to the BSR to ensure all stores and
+ the BSR update have made it to all chips in the system
+*/
+
+/* This is arbitrary number, up to Power6 it's been 17 or fewer  */
+#define BSR_MAX_DEVS (32)
+
+struct bsr_dev {
+       u64      bsr_addr;     /* Real address */
+       u64      bsr_len;      /* length of mem region we can map */
+       unsigned bsr_bytes;    /* size of the BSR reg itself */
+       unsigned bsr_stride;   /* interval at which BSR repeats in the page */
+       unsigned bsr_type;     /* maps to enum below */
+       unsigned bsr_num;      /* bsr id number for its type */
+       int      bsr_minor;
+
+       dev_t    bsr_dev;
+       struct cdev bsr_cdev;
+       struct device *bsr_device;
+       char     bsr_name[32];
+
+};
+
+static unsigned num_bsr_devs;
+static struct bsr_dev *bsr_devs;
+static struct class *bsr_class;
+static int bsr_major;
+
+enum {
+       BSR_8   = 0,
+       BSR_16  = 1,
+       BSR_64  = 2,
+       BSR_128 = 3,
+       BSR_UNKNOWN = 4,
+       BSR_MAX = 5,
+};
+
+static unsigned bsr_types[BSR_MAX];
+
+static ssize_t
+bsr_size_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+       struct bsr_dev *bsr_dev = dev_get_drvdata(dev);
+       return sprintf(buf, "%u\n", bsr_dev->bsr_bytes);
+}
+
+static ssize_t
+bsr_stride_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+       struct bsr_dev *bsr_dev = dev_get_drvdata(dev);
+       return sprintf(buf, "%u\n", bsr_dev->bsr_stride);
+}
+
+static ssize_t
+bsr_len_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+       struct bsr_dev *bsr_dev = dev_get_drvdata(dev);
+       return sprintf(buf, "%lu\n", bsr_dev->bsr_len);
+}
+
+static struct device_attribute bsr_dev_attrs[] = {
+       __ATTR(bsr_size, S_IRUGO, bsr_size_show, NULL),
+       __ATTR(bsr_stride, S_IRUGO, bsr_stride_show, NULL),
+       __ATTR(bsr_length, S_IRUGO, bsr_len_show, NULL),
+       __ATTR_NULL
+};
+
+static int bsr_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+       unsigned long size   = vma->vm_end - vma->vm_start;
+       struct bsr_dev *dev = filp->private_data;
+
+       if (size > dev->bsr_len || (size & (PAGE_SIZE-1)))
+               return -EINVAL;
+
+       vma->vm_flags |= (VM_IO | VM_DONTEXPAND);
+       vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+
+       if (io_remap_pfn_range(vma, vma->vm_start, dev->bsr_addr >> PAGE_SHIFT,
+                              size, vma->vm_page_prot))
+               return -EAGAIN;
+
+       return 0;
+}
+
+static int bsr_open(struct inode * inode, struct file * filp)
+{
+       struct cdev *cdev = inode->i_cdev;
+       struct bsr_dev *dev = container_of(cdev, struct bsr_dev, bsr_cdev);
+
+       filp->private_data = dev;
+       return 0;
+}
+
+const static struct file_operations bsr_fops = {
+       .owner = THIS_MODULE,
+       .mmap  = bsr_mmap,
+       .open  = bsr_open,
+};
+
+static void bsr_cleanup_devs(void)
+{
+       int i;
+       for (i=0 ; i < num_bsr_devs; i++) {
+               struct bsr_dev *cur = bsr_devs + i;
+               if (cur->bsr_device) {
+                       cdev_del(&cur->bsr_cdev);
+                       device_del(cur->bsr_device);
+               }
+       }
+
+       kfree(bsr_devs);
+}
+
+static int bsr_create_devs(struct device_node *bn)
+{
+       int bsr_stride_len, bsr_bytes_len;
+       const u32 *bsr_stride;
+       const u32 *bsr_bytes;
+       unsigned i;
+
+       bsr_stride = of_get_property(bn, "ibm,lock-stride", &bsr_stride_len);
+       bsr_bytes  = of_get_property(bn, "ibm,#lock-bytes", &bsr_bytes_len);
+
+       if (!bsr_stride || !bsr_bytes ||
+           (bsr_stride_len != bsr_bytes_len)) {
+               printk(KERN_ERR "bsr of-node has missing/incorrect property\n");
+               return -ENODEV;
+       }
+
+       num_bsr_devs = bsr_bytes_len / sizeof(u32);
+
+       /* only a warning, its informational since we'll fail and exit */
+       WARN_ON(num_bsr_devs > BSR_MAX_DEVS);
+
+       bsr_devs = kzalloc(sizeof(struct bsr_dev) * num_bsr_devs, GFP_KERNEL);
+       if (!bsr_devs)
+               return -ENOMEM;
+
+       for (i = 0 ; i < num_bsr_devs; i++) {
+               struct bsr_dev *cur = bsr_devs + i;
+               struct resource res;
+               int result;
+
+               result = of_address_to_resource(bn, i, &res);
+               if (result < 0) {
+                       printk(KERN_ERR "bsr of-node has invalid reg property\n");
+                       goto out_err;
+               }
+
+               cur->bsr_minor  = i;
+               cur->bsr_addr   = res.start;
+               cur->bsr_len    = res.end - res.start + 1;
+               cur->bsr_bytes  = bsr_bytes[i];
+               cur->bsr_stride = bsr_stride[i];
+               cur->bsr_dev    = MKDEV(bsr_major, i);
+
+               switch(cur->bsr_bytes) {
+               case 8:
+                       cur->bsr_type = BSR_8;
+                       break;
+               case 16:
+                       cur->bsr_type = BSR_16;
+                       break;
+               case 64:
+                       cur->bsr_type = BSR_64;
+                       break;
+               case 128:
+                       cur->bsr_type = BSR_128;
+                       break;
+               default:
+                       cur->bsr_type = BSR_UNKNOWN;
+                       printk(KERN_INFO "unknown BSR size %d\n",cur->bsr_bytes);
+               }
+
+               cur->bsr_num = bsr_types[cur->bsr_type];
+               bsr_types[cur->bsr_type] = cur->bsr_num + 1;
+               snprintf(cur->bsr_name, 32, "bsr%d_%d",
+                        cur->bsr_bytes, cur->bsr_num);
+
+               cdev_init(&cur->bsr_cdev, &bsr_fops);
+               result = cdev_add(&cur->bsr_cdev, cur->bsr_dev, 1);
+               if (result)
+                       goto out_err;
+
+               cur->bsr_device = device_create_drvdata(bsr_class, NULL,
+                                                       cur->bsr_dev,
+                                                       cur, cur->bsr_name);
+               if (!cur->bsr_device) {
+                       printk(KERN_ERR "device_create failed for %s\n",
+                              cur->bsr_name);
+                       cdev_del(&cur->bsr_cdev);
+                       goto out_err;
+               }
+       }
+
+       return 0;
+
+ out_err:
+
+       bsr_cleanup_devs();
+       return -ENODEV;
+}
+
+static int __init bsr_init(void)
+{
+       struct device_node *np;
+       dev_t bsr_dev = MKDEV(bsr_major, 0);
+       int ret = -ENODEV;
+       int result;
+
+       np = of_find_compatible_node(NULL, "ibm,bsr", "ibm,bsr");
+       if (!np)
+               goto out_err;
+
+       bsr_class = class_create(THIS_MODULE, "bsr");
+       if (IS_ERR(bsr_class)) {
+               printk(KERN_ERR "class_create() failed for bsr_class\n");
+               goto out_err_1;
+       }
+       bsr_class->dev_attrs = bsr_dev_attrs;
+
+       result = alloc_chrdev_region(&bsr_dev, 0, BSR_MAX_DEVS, "bsr");
+       bsr_major = MAJOR(bsr_dev);
+       if (result < 0) {
+               printk(KERN_ERR "alloc_chrdev_region() failed for bsr\n");
+               goto out_err_2;
+       }
+
+       if ((ret = bsr_create_devs(np)) < 0)
+               goto out_err_3;
+
+       of_node_put(np);
+
+       return 0;
+
+ out_err_3:
+       unregister_chrdev_region(bsr_dev, BSR_MAX_DEVS);
+
+ out_err_2:
+       class_destroy(bsr_class);
+
+ out_err_1:
+       of_node_put(np);
+
+ out_err:
+
+       return ret;
+}
+
+static void __exit  bsr_exit(void)
+{
+
+       bsr_cleanup_devs();
+
+       if (bsr_class)
+               class_destroy(bsr_class);
+
+       if (bsr_major)
+               unregister_chrdev_region(MKDEV(bsr_major, 0), BSR_MAX_DEVS);
+}
+
+module_init(bsr_init);
+module_exit(bsr_exit);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Sonny Rao <sonnyrao@us.ibm.com>");