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_device.h>
25 #include <linux/of_platform.h>
27 #include <linux/module.h>
28 #include <linux/cdev.h>
29 #include <linux/list.h>
31 #include <linux/slab.h>
32 #include <asm/pgtable.h>
36 This driver exposes a special register which can be used for fast
37 synchronization across a large SMP machine. The hardware is exposed
38 as an array of bytes where each process will write to one of the bytes to
39 indicate it has finished the current stage and this update is broadcast to
40 all processors without having to bounce a cacheline between them. In
41 POWER5 and POWER6 there is one of these registers per SMP, but it is
42 presented in two forms; first, it is given as a whole and then as a number
43 of smaller registers which alias to parts of the single whole register.
44 This can potentially allow multiple groups of processes to each have their
45 own private synchronization device.
47 Note that this hardware *must* be written to using *only* single byte writes.
48 It may be read using 1, 2, 4, or 8 byte loads which must be aligned since
49 this region is treated as cache-inhibited processes should also use a
50 full sync before and after writing to the BSR to ensure all stores and
51 the BSR update have made it to all chips in the system
54 /* This is arbitrary number, up to Power6 it's been 17 or fewer */
55 #define BSR_MAX_DEVS (32)
58 u64 bsr_addr
; /* Real address */
59 u64 bsr_len
; /* length of mem region we can map */
60 unsigned bsr_bytes
; /* size of the BSR reg itself */
61 unsigned bsr_stride
; /* interval at which BSR repeats in the page */
62 unsigned bsr_type
; /* maps to enum below */
63 unsigned bsr_num
; /* bsr id number for its type */
66 struct list_head bsr_list
;
70 struct device
*bsr_device
;
75 static unsigned total_bsr_devs
;
76 static struct list_head bsr_devs
= LIST_HEAD_INIT(bsr_devs
);
77 static struct class *bsr_class
;
90 static unsigned bsr_types
[BSR_MAX
];
93 bsr_size_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
95 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
96 return sprintf(buf
, "%u\n", bsr_dev
->bsr_bytes
);
100 bsr_stride_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
102 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
103 return sprintf(buf
, "%u\n", bsr_dev
->bsr_stride
);
107 bsr_len_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
109 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
110 return sprintf(buf
, "%llu\n", bsr_dev
->bsr_len
);
113 static struct device_attribute bsr_dev_attrs
[] = {
114 __ATTR(bsr_size
, S_IRUGO
, bsr_size_show
, NULL
),
115 __ATTR(bsr_stride
, S_IRUGO
, bsr_stride_show
, NULL
),
116 __ATTR(bsr_length
, S_IRUGO
, bsr_len_show
, NULL
),
120 static int bsr_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
122 unsigned long size
= vma
->vm_end
- vma
->vm_start
;
123 struct bsr_dev
*dev
= filp
->private_data
;
126 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
128 /* check for the case of a small BSR device and map one 4k page for it*/
129 if (dev
->bsr_len
< PAGE_SIZE
&& size
== PAGE_SIZE
)
130 ret
= remap_4k_pfn(vma
, vma
->vm_start
, dev
->bsr_addr
>> 12,
132 else if (size
<= dev
->bsr_len
)
133 ret
= io_remap_pfn_range(vma
, vma
->vm_start
,
134 dev
->bsr_addr
>> PAGE_SHIFT
,
135 size
, vma
->vm_page_prot
);
145 static int bsr_open(struct inode
* inode
, struct file
* filp
)
147 struct cdev
*cdev
= inode
->i_cdev
;
148 struct bsr_dev
*dev
= container_of(cdev
, struct bsr_dev
, bsr_cdev
);
150 filp
->private_data
= dev
;
154 static const struct file_operations bsr_fops
= {
155 .owner
= THIS_MODULE
,
158 .llseek
= noop_llseek
,
161 static void bsr_cleanup_devs(void)
163 struct bsr_dev
*cur
, *n
;
165 list_for_each_entry_safe(cur
, n
, &bsr_devs
, bsr_list
) {
166 if (cur
->bsr_device
) {
167 cdev_del(&cur
->bsr_cdev
);
168 device_del(cur
->bsr_device
);
170 list_del(&cur
->bsr_list
);
175 static int bsr_add_node(struct device_node
*bn
)
177 int bsr_stride_len
, bsr_bytes_len
, num_bsr_devs
;
178 const u32
*bsr_stride
;
179 const u32
*bsr_bytes
;
183 bsr_stride
= of_get_property(bn
, "ibm,lock-stride", &bsr_stride_len
);
184 bsr_bytes
= of_get_property(bn
, "ibm,#lock-bytes", &bsr_bytes_len
);
186 if (!bsr_stride
|| !bsr_bytes
||
187 (bsr_stride_len
!= bsr_bytes_len
)) {
188 printk(KERN_ERR
"bsr of-node has missing/incorrect property\n");
192 num_bsr_devs
= bsr_bytes_len
/ sizeof(u32
);
194 for (i
= 0 ; i
< num_bsr_devs
; i
++) {
195 struct bsr_dev
*cur
= kzalloc(sizeof(struct bsr_dev
),
201 printk(KERN_ERR
"Unable to alloc bsr dev\n");
206 result
= of_address_to_resource(bn
, i
, &res
);
208 printk(KERN_ERR
"bsr of-node has invalid reg property, skipping\n");
213 cur
->bsr_minor
= i
+ total_bsr_devs
;
214 cur
->bsr_addr
= res
.start
;
215 cur
->bsr_len
= resource_size(&res
);
216 cur
->bsr_bytes
= bsr_bytes
[i
];
217 cur
->bsr_stride
= bsr_stride
[i
];
218 cur
->bsr_dev
= MKDEV(bsr_major
, i
+ total_bsr_devs
);
220 /* if we have a bsr_len of > 4k and less then PAGE_SIZE (64k pages) */
221 /* we can only map 4k of it, so only advertise the 4k in sysfs */
222 if (cur
->bsr_len
> 4096 && cur
->bsr_len
< PAGE_SIZE
)
225 switch(cur
->bsr_bytes
) {
227 cur
->bsr_type
= BSR_8
;
230 cur
->bsr_type
= BSR_16
;
233 cur
->bsr_type
= BSR_64
;
236 cur
->bsr_type
= BSR_128
;
239 cur
->bsr_type
= BSR_4096
;
242 cur
->bsr_type
= BSR_UNKNOWN
;
245 cur
->bsr_num
= bsr_types
[cur
->bsr_type
];
246 snprintf(cur
->bsr_name
, 32, "bsr%d_%d",
247 cur
->bsr_bytes
, cur
->bsr_num
);
249 cdev_init(&cur
->bsr_cdev
, &bsr_fops
);
250 result
= cdev_add(&cur
->bsr_cdev
, cur
->bsr_dev
, 1);
256 cur
->bsr_device
= device_create(bsr_class
, NULL
, cur
->bsr_dev
,
258 if (IS_ERR(cur
->bsr_device
)) {
259 printk(KERN_ERR
"device_create failed for %s\n",
261 cdev_del(&cur
->bsr_cdev
);
266 bsr_types
[cur
->bsr_type
] = cur
->bsr_num
+ 1;
267 list_add_tail(&cur
->bsr_list
, &bsr_devs
);
270 total_bsr_devs
+= num_bsr_devs
;
280 static int bsr_create_devs(struct device_node
*bn
)
285 ret
= bsr_add_node(bn
);
290 bn
= of_find_compatible_node(bn
, NULL
, "ibm,bsr");
295 static int __init
bsr_init(void)
297 struct device_node
*np
;
301 np
= of_find_compatible_node(NULL
, NULL
, "ibm,bsr");
305 bsr_class
= class_create(THIS_MODULE
, "bsr");
306 if (IS_ERR(bsr_class
)) {
307 printk(KERN_ERR
"class_create() failed for bsr_class\n");
308 ret
= PTR_ERR(bsr_class
);
311 bsr_class
->dev_attrs
= bsr_dev_attrs
;
313 ret
= alloc_chrdev_region(&bsr_dev
, 0, BSR_MAX_DEVS
, "bsr");
314 bsr_major
= MAJOR(bsr_dev
);
316 printk(KERN_ERR
"alloc_chrdev_region() failed for bsr\n");
320 if ((ret
= bsr_create_devs(np
)) < 0) {
328 unregister_chrdev_region(bsr_dev
, BSR_MAX_DEVS
);
331 class_destroy(bsr_class
);
341 static void __exit
bsr_exit(void)
347 class_destroy(bsr_class
);
350 unregister_chrdev_region(MKDEV(bsr_major
, 0), BSR_MAX_DEVS
);
353 module_init(bsr_init
);
354 module_exit(bsr_exit
);
355 MODULE_LICENSE("GPL");
356 MODULE_AUTHOR("Sonny Rao <sonnyrao@us.ibm.com>");