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
);
98 static DEVICE_ATTR_RO(bsr_size
);
101 bsr_stride_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
103 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
104 return sprintf(buf
, "%u\n", bsr_dev
->bsr_stride
);
106 static DEVICE_ATTR_RO(bsr_stride
);
109 bsr_length_show(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
111 struct bsr_dev
*bsr_dev
= dev_get_drvdata(dev
);
112 return sprintf(buf
, "%llu\n", bsr_dev
->bsr_len
);
114 static DEVICE_ATTR_RO(bsr_length
);
116 static struct attribute
*bsr_dev_attrs
[] = {
117 &dev_attr_bsr_size
.attr
,
118 &dev_attr_bsr_stride
.attr
,
119 &dev_attr_bsr_length
.attr
,
122 ATTRIBUTE_GROUPS(bsr_dev
);
124 static int bsr_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
126 unsigned long size
= vma
->vm_end
- vma
->vm_start
;
127 struct bsr_dev
*dev
= filp
->private_data
;
130 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
132 /* check for the case of a small BSR device and map one 4k page for it*/
133 if (dev
->bsr_len
< PAGE_SIZE
&& size
== PAGE_SIZE
)
134 ret
= remap_4k_pfn(vma
, vma
->vm_start
, dev
->bsr_addr
>> 12,
136 else if (size
<= dev
->bsr_len
)
137 ret
= io_remap_pfn_range(vma
, vma
->vm_start
,
138 dev
->bsr_addr
>> PAGE_SHIFT
,
139 size
, vma
->vm_page_prot
);
149 static int bsr_open(struct inode
* inode
, struct file
* filp
)
151 struct cdev
*cdev
= inode
->i_cdev
;
152 struct bsr_dev
*dev
= container_of(cdev
, struct bsr_dev
, bsr_cdev
);
154 filp
->private_data
= dev
;
158 static const struct file_operations bsr_fops
= {
159 .owner
= THIS_MODULE
,
162 .llseek
= noop_llseek
,
165 static void bsr_cleanup_devs(void)
167 struct bsr_dev
*cur
, *n
;
169 list_for_each_entry_safe(cur
, n
, &bsr_devs
, bsr_list
) {
170 if (cur
->bsr_device
) {
171 cdev_del(&cur
->bsr_cdev
);
172 device_del(cur
->bsr_device
);
174 list_del(&cur
->bsr_list
);
179 static int bsr_add_node(struct device_node
*bn
)
181 int bsr_stride_len
, bsr_bytes_len
, num_bsr_devs
;
182 const u32
*bsr_stride
;
183 const u32
*bsr_bytes
;
187 bsr_stride
= of_get_property(bn
, "ibm,lock-stride", &bsr_stride_len
);
188 bsr_bytes
= of_get_property(bn
, "ibm,#lock-bytes", &bsr_bytes_len
);
190 if (!bsr_stride
|| !bsr_bytes
||
191 (bsr_stride_len
!= bsr_bytes_len
)) {
192 printk(KERN_ERR
"bsr of-node has missing/incorrect property\n");
196 num_bsr_devs
= bsr_bytes_len
/ sizeof(u32
);
198 for (i
= 0 ; i
< num_bsr_devs
; i
++) {
199 struct bsr_dev
*cur
= kzalloc(sizeof(struct bsr_dev
),
205 printk(KERN_ERR
"Unable to alloc bsr dev\n");
210 result
= of_address_to_resource(bn
, i
, &res
);
212 printk(KERN_ERR
"bsr of-node has invalid reg property, skipping\n");
217 cur
->bsr_minor
= i
+ total_bsr_devs
;
218 cur
->bsr_addr
= res
.start
;
219 cur
->bsr_len
= resource_size(&res
);
220 cur
->bsr_bytes
= bsr_bytes
[i
];
221 cur
->bsr_stride
= bsr_stride
[i
];
222 cur
->bsr_dev
= MKDEV(bsr_major
, i
+ total_bsr_devs
);
224 /* if we have a bsr_len of > 4k and less then PAGE_SIZE (64k pages) */
225 /* we can only map 4k of it, so only advertise the 4k in sysfs */
226 if (cur
->bsr_len
> 4096 && cur
->bsr_len
< PAGE_SIZE
)
229 switch(cur
->bsr_bytes
) {
231 cur
->bsr_type
= BSR_8
;
234 cur
->bsr_type
= BSR_16
;
237 cur
->bsr_type
= BSR_64
;
240 cur
->bsr_type
= BSR_128
;
243 cur
->bsr_type
= BSR_4096
;
246 cur
->bsr_type
= BSR_UNKNOWN
;
249 cur
->bsr_num
= bsr_types
[cur
->bsr_type
];
250 snprintf(cur
->bsr_name
, 32, "bsr%d_%d",
251 cur
->bsr_bytes
, cur
->bsr_num
);
253 cdev_init(&cur
->bsr_cdev
, &bsr_fops
);
254 result
= cdev_add(&cur
->bsr_cdev
, cur
->bsr_dev
, 1);
260 cur
->bsr_device
= device_create(bsr_class
, NULL
, cur
->bsr_dev
,
262 if (IS_ERR(cur
->bsr_device
)) {
263 printk(KERN_ERR
"device_create failed for %s\n",
265 cdev_del(&cur
->bsr_cdev
);
270 bsr_types
[cur
->bsr_type
] = cur
->bsr_num
+ 1;
271 list_add_tail(&cur
->bsr_list
, &bsr_devs
);
274 total_bsr_devs
+= num_bsr_devs
;
284 static int bsr_create_devs(struct device_node
*bn
)
289 ret
= bsr_add_node(bn
);
294 bn
= of_find_compatible_node(bn
, NULL
, "ibm,bsr");
299 static int __init
bsr_init(void)
301 struct device_node
*np
;
305 np
= of_find_compatible_node(NULL
, NULL
, "ibm,bsr");
309 bsr_class
= class_create(THIS_MODULE
, "bsr");
310 if (IS_ERR(bsr_class
)) {
311 printk(KERN_ERR
"class_create() failed for bsr_class\n");
312 ret
= PTR_ERR(bsr_class
);
315 bsr_class
->dev_groups
= bsr_dev_groups
;
317 ret
= alloc_chrdev_region(&bsr_dev
, 0, BSR_MAX_DEVS
, "bsr");
318 bsr_major
= MAJOR(bsr_dev
);
320 printk(KERN_ERR
"alloc_chrdev_region() failed for bsr\n");
324 if ((ret
= bsr_create_devs(np
)) < 0) {
332 unregister_chrdev_region(bsr_dev
, BSR_MAX_DEVS
);
335 class_destroy(bsr_class
);
345 static void __exit
bsr_exit(void)
351 class_destroy(bsr_class
);
354 unregister_chrdev_region(MKDEV(bsr_major
, 0), BSR_MAX_DEVS
);
357 module_init(bsr_init
);
358 module_exit(bsr_exit
);
359 MODULE_LICENSE("GPL");
360 MODULE_AUTHOR("Sonny Rao <sonnyrao@us.ibm.com>");