1 /* IBM POWER Barrier Synchronization Register Driver
3 * Copyright IBM Corporation 2008
5 * Author: Sonny Rao <sonnyrao@us.ibm.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 #include <linux/kernel.h>
24 #include <linux/of_address.h>
25 #include <linux/of_device.h>
26 #include <linux/of_platform.h>
28 #include <linux/module.h>
29 #include <linux/cdev.h>
30 #include <linux/list.h>
32 #include <linux/slab.h>
33 #include <asm/pgtable.h>
37 This driver exposes a special register which can be used for fast
38 synchronization across a large SMP machine. The hardware is exposed
39 as an array of bytes where each process will write to one of the bytes to
40 indicate it has finished the current stage and this update is broadcast to
41 all processors without having to bounce a cacheline between them. In
42 POWER5 and POWER6 there is one of these registers per SMP, but it is
43 presented in two forms; first, it is given as a whole and then as a number
44 of smaller registers which alias to parts of the single whole register.
45 This can potentially allow multiple groups of processes to each have their
46 own private synchronization device.
48 Note that this hardware *must* be written to using *only* single byte writes.
49 It may be read using 1, 2, 4, or 8 byte loads which must be aligned since
50 this region is treated as cache-inhibited processes should also use a
51 full sync before and after writing to the BSR to ensure all stores and
52 the BSR update have made it to all chips in the system
55 /* This is arbitrary number, up to Power6 it's been 17 or fewer */
56 #define BSR_MAX_DEVS (32)
59 u64 bsr_addr
; /* Real address */
60 u64 bsr_len
; /* length of mem region we can map */
61 unsigned bsr_bytes
; /* size of the BSR reg itself */
62 unsigned bsr_stride
; /* interval at which BSR repeats in the page */
63 unsigned bsr_type
; /* maps to enum below */
64 unsigned bsr_num
; /* bsr id number for its type */
67 struct list_head bsr_list
;
71 struct device
*bsr_device
;
76 static unsigned total_bsr_devs
;
77 static struct list_head bsr_devs
= LIST_HEAD_INIT(bsr_devs
);
78 static struct class *bsr_class
;
91 static unsigned bsr_types
[BSR_MAX
];
94 bsr_size_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
96 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
97 return sprintf(buf
, "%u\n", bsr_dev
->bsr_bytes
);
99 static DEVICE_ATTR_RO(bsr_size
);
102 bsr_stride_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
104 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
105 return sprintf(buf
, "%u\n", bsr_dev
->bsr_stride
);
107 static DEVICE_ATTR_RO(bsr_stride
);
110 bsr_length_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
112 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
113 return sprintf(buf
, "%llu\n", bsr_dev
->bsr_len
);
115 static DEVICE_ATTR_RO(bsr_length
);
117 static struct attribute
*bsr_dev_attrs
[] = {
118 &dev_attr_bsr_size
.attr
,
119 &dev_attr_bsr_stride
.attr
,
120 &dev_attr_bsr_length
.attr
,
123 ATTRIBUTE_GROUPS(bsr_dev
);
125 static int bsr_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
127 unsigned long size
= vma
->vm_end
- vma
->vm_start
;
128 struct bsr_dev
*dev
= filp
->private_data
;
131 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
133 /* check for the case of a small BSR device and map one 4k page for it*/
134 if (dev
->bsr_len
< PAGE_SIZE
&& size
== PAGE_SIZE
)
135 ret
= remap_4k_pfn(vma
, vma
->vm_start
, dev
->bsr_addr
>> 12,
137 else if (size
<= dev
->bsr_len
)
138 ret
= io_remap_pfn_range(vma
, vma
->vm_start
,
139 dev
->bsr_addr
>> PAGE_SHIFT
,
140 size
, vma
->vm_page_prot
);
150 static int bsr_open(struct inode
* inode
, struct file
* filp
)
152 struct cdev
*cdev
= inode
->i_cdev
;
153 struct bsr_dev
*dev
= container_of(cdev
, struct bsr_dev
, bsr_cdev
);
155 filp
->private_data
= dev
;
159 static const struct file_operations bsr_fops
= {
160 .owner
= THIS_MODULE
,
163 .llseek
= noop_llseek
,
166 static void bsr_cleanup_devs(void)
168 struct bsr_dev
*cur
, *n
;
170 list_for_each_entry_safe(cur
, n
, &bsr_devs
, bsr_list
) {
171 if (cur
->bsr_device
) {
172 cdev_del(&cur
->bsr_cdev
);
173 device_del(cur
->bsr_device
);
175 list_del(&cur
->bsr_list
);
180 static int bsr_add_node(struct device_node
*bn
)
182 int bsr_stride_len
, bsr_bytes_len
, num_bsr_devs
;
183 const u32
*bsr_stride
;
184 const u32
*bsr_bytes
;
188 bsr_stride
= of_get_property(bn
, "ibm,lock-stride", &bsr_stride_len
);
189 bsr_bytes
= of_get_property(bn
, "ibm,#lock-bytes", &bsr_bytes_len
);
191 if (!bsr_stride
|| !bsr_bytes
||
192 (bsr_stride_len
!= bsr_bytes_len
)) {
193 printk(KERN_ERR
"bsr of-node has missing/incorrect property\n");
197 num_bsr_devs
= bsr_bytes_len
/ sizeof(u32
);
199 for (i
= 0 ; i
< num_bsr_devs
; i
++) {
200 struct bsr_dev
*cur
= kzalloc(sizeof(struct bsr_dev
),
206 printk(KERN_ERR
"Unable to alloc bsr dev\n");
211 result
= of_address_to_resource(bn
, i
, &res
);
213 printk(KERN_ERR
"bsr of-node has invalid reg property, skipping\n");
218 cur
->bsr_minor
= i
+ total_bsr_devs
;
219 cur
->bsr_addr
= res
.start
;
220 cur
->bsr_len
= resource_size(&res
);
221 cur
->bsr_bytes
= bsr_bytes
[i
];
222 cur
->bsr_stride
= bsr_stride
[i
];
223 cur
->bsr_dev
= MKDEV(bsr_major
, i
+ total_bsr_devs
);
225 /* if we have a bsr_len of > 4k and less then PAGE_SIZE (64k pages) */
226 /* we can only map 4k of it, so only advertise the 4k in sysfs */
227 if (cur
->bsr_len
> 4096 && cur
->bsr_len
< PAGE_SIZE
)
230 switch(cur
->bsr_bytes
) {
232 cur
->bsr_type
= BSR_8
;
235 cur
->bsr_type
= BSR_16
;
238 cur
->bsr_type
= BSR_64
;
241 cur
->bsr_type
= BSR_128
;
244 cur
->bsr_type
= BSR_4096
;
247 cur
->bsr_type
= BSR_UNKNOWN
;
250 cur
->bsr_num
= bsr_types
[cur
->bsr_type
];
251 snprintf(cur
->bsr_name
, 32, "bsr%d_%d",
252 cur
->bsr_bytes
, cur
->bsr_num
);
254 cdev_init(&cur
->bsr_cdev
, &bsr_fops
);
255 result
= cdev_add(&cur
->bsr_cdev
, cur
->bsr_dev
, 1);
261 cur
->bsr_device
= device_create(bsr_class
, NULL
, cur
->bsr_dev
,
263 if (IS_ERR(cur
->bsr_device
)) {
264 printk(KERN_ERR
"device_create failed for %s\n",
266 cdev_del(&cur
->bsr_cdev
);
271 bsr_types
[cur
->bsr_type
] = cur
->bsr_num
+ 1;
272 list_add_tail(&cur
->bsr_list
, &bsr_devs
);
275 total_bsr_devs
+= num_bsr_devs
;
285 static int bsr_create_devs(struct device_node
*bn
)
290 ret
= bsr_add_node(bn
);
295 bn
= of_find_compatible_node(bn
, NULL
, "ibm,bsr");
300 static int __init
bsr_init(void)
302 struct device_node
*np
;
306 np
= of_find_compatible_node(NULL
, NULL
, "ibm,bsr");
310 bsr_class
= class_create(THIS_MODULE
, "bsr");
311 if (IS_ERR(bsr_class
)) {
312 printk(KERN_ERR
"class_create() failed for bsr_class\n");
313 ret
= PTR_ERR(bsr_class
);
316 bsr_class
->dev_groups
= bsr_dev_groups
;
318 ret
= alloc_chrdev_region(&bsr_dev
, 0, BSR_MAX_DEVS
, "bsr");
319 bsr_major
= MAJOR(bsr_dev
);
321 printk(KERN_ERR
"alloc_chrdev_region() failed for bsr\n");
325 if ((ret
= bsr_create_devs(np
)) < 0) {
333 unregister_chrdev_region(bsr_dev
, BSR_MAX_DEVS
);
336 class_destroy(bsr_class
);
346 static void __exit
bsr_exit(void)
352 class_destroy(bsr_class
);
355 unregister_chrdev_region(MKDEV(bsr_major
, 0), BSR_MAX_DEVS
);
358 module_init(bsr_init
);
359 module_exit(bsr_exit
);
360 MODULE_LICENSE("GPL");
361 MODULE_AUTHOR("Sonny Rao <sonnyrao@us.ibm.com>");