Commit | Line | Data |
---|---|---|
fe9e8d53 SR |
1 | /* IBM POWER Barrier Synchronization Register Driver |
2 | * | |
3 | * Copyright IBM Corporation 2008 | |
4 | * | |
5 | * Author: Sonny Rao <sonnyrao@us.ibm.com> | |
6 | * | |
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. | |
11 | * | |
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. | |
16 | * | |
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 | |
20 | */ | |
21 | ||
22 | #include <linux/kernel.h> | |
23 | #include <linux/of.h> | |
24 | #include <linux/of_device.h> | |
25 | #include <linux/of_platform.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/cdev.h> | |
28 | #include <linux/list.h> | |
29 | #include <linux/mm.h> | |
30 | #include <asm/io.h> | |
31 | ||
32 | /* | |
33 | This driver exposes a special register which can be used for fast | |
34 | synchronization across a large SMP machine. The hardware is exposed | |
35 | as an array of bytes where each process will write to one of the bytes to | |
36 | indicate it has finished the current stage and this update is broadcast to | |
37 | all processors without having to bounce a cacheline between them. In | |
38 | POWER5 and POWER6 there is one of these registers per SMP, but it is | |
39 | presented in two forms; first, it is given as a whole and then as a number | |
40 | of smaller registers which alias to parts of the single whole register. | |
41 | This can potentially allow multiple groups of processes to each have their | |
42 | own private synchronization device. | |
43 | ||
44 | Note that this hardware *must* be written to using *only* single byte writes. | |
45 | It may be read using 1, 2, 4, or 8 byte loads which must be aligned since | |
46 | this region is treated as cache-inhibited processes should also use a | |
47 | full sync before and after writing to the BSR to ensure all stores and | |
48 | the BSR update have made it to all chips in the system | |
49 | */ | |
50 | ||
51 | /* This is arbitrary number, up to Power6 it's been 17 or fewer */ | |
52 | #define BSR_MAX_DEVS (32) | |
53 | ||
54 | struct bsr_dev { | |
55 | u64 bsr_addr; /* Real address */ | |
56 | u64 bsr_len; /* length of mem region we can map */ | |
57 | unsigned bsr_bytes; /* size of the BSR reg itself */ | |
58 | unsigned bsr_stride; /* interval at which BSR repeats in the page */ | |
59 | unsigned bsr_type; /* maps to enum below */ | |
60 | unsigned bsr_num; /* bsr id number for its type */ | |
61 | int bsr_minor; | |
62 | ||
a0e2f9f4 SR |
63 | struct list_head bsr_list; |
64 | ||
fe9e8d53 SR |
65 | dev_t bsr_dev; |
66 | struct cdev bsr_cdev; | |
67 | struct device *bsr_device; | |
68 | char bsr_name[32]; | |
69 | ||
70 | }; | |
71 | ||
a0e2f9f4 SR |
72 | static unsigned total_bsr_devs; |
73 | static struct list_head bsr_devs = LIST_HEAD_INIT(bsr_devs); | |
fe9e8d53 SR |
74 | static struct class *bsr_class; |
75 | static int bsr_major; | |
76 | ||
77 | enum { | |
78 | BSR_8 = 0, | |
79 | BSR_16 = 1, | |
80 | BSR_64 = 2, | |
81 | BSR_128 = 3, | |
82 | BSR_UNKNOWN = 4, | |
83 | BSR_MAX = 5, | |
84 | }; | |
85 | ||
86 | static unsigned bsr_types[BSR_MAX]; | |
87 | ||
88 | static ssize_t | |
89 | bsr_size_show(struct device *dev, struct device_attribute *attr, char *buf) | |
90 | { | |
91 | struct bsr_dev *bsr_dev = dev_get_drvdata(dev); | |
92 | return sprintf(buf, "%u\n", bsr_dev->bsr_bytes); | |
93 | } | |
94 | ||
95 | static ssize_t | |
96 | bsr_stride_show(struct device *dev, struct device_attribute *attr, char *buf) | |
97 | { | |
98 | struct bsr_dev *bsr_dev = dev_get_drvdata(dev); | |
99 | return sprintf(buf, "%u\n", bsr_dev->bsr_stride); | |
100 | } | |
101 | ||
102 | static ssize_t | |
103 | bsr_len_show(struct device *dev, struct device_attribute *attr, char *buf) | |
104 | { | |
105 | struct bsr_dev *bsr_dev = dev_get_drvdata(dev); | |
106 | return sprintf(buf, "%lu\n", bsr_dev->bsr_len); | |
107 | } | |
108 | ||
109 | static struct device_attribute bsr_dev_attrs[] = { | |
110 | __ATTR(bsr_size, S_IRUGO, bsr_size_show, NULL), | |
111 | __ATTR(bsr_stride, S_IRUGO, bsr_stride_show, NULL), | |
112 | __ATTR(bsr_length, S_IRUGO, bsr_len_show, NULL), | |
113 | __ATTR_NULL | |
114 | }; | |
115 | ||
116 | static int bsr_mmap(struct file *filp, struct vm_area_struct *vma) | |
117 | { | |
118 | unsigned long size = vma->vm_end - vma->vm_start; | |
119 | struct bsr_dev *dev = filp->private_data; | |
120 | ||
121 | if (size > dev->bsr_len || (size & (PAGE_SIZE-1))) | |
122 | return -EINVAL; | |
123 | ||
124 | vma->vm_flags |= (VM_IO | VM_DONTEXPAND); | |
125 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
126 | ||
127 | if (io_remap_pfn_range(vma, vma->vm_start, dev->bsr_addr >> PAGE_SHIFT, | |
128 | size, vma->vm_page_prot)) | |
129 | return -EAGAIN; | |
130 | ||
131 | return 0; | |
132 | } | |
133 | ||
134 | static int bsr_open(struct inode * inode, struct file * filp) | |
135 | { | |
136 | struct cdev *cdev = inode->i_cdev; | |
137 | struct bsr_dev *dev = container_of(cdev, struct bsr_dev, bsr_cdev); | |
138 | ||
139 | filp->private_data = dev; | |
140 | return 0; | |
141 | } | |
142 | ||
143 | const static struct file_operations bsr_fops = { | |
144 | .owner = THIS_MODULE, | |
145 | .mmap = bsr_mmap, | |
146 | .open = bsr_open, | |
147 | }; | |
148 | ||
149 | static void bsr_cleanup_devs(void) | |
150 | { | |
a0e2f9f4 SR |
151 | struct bsr_dev *cur, *n; |
152 | ||
153 | list_for_each_entry_safe(cur, n, &bsr_devs, bsr_list) { | |
fe9e8d53 SR |
154 | if (cur->bsr_device) { |
155 | cdev_del(&cur->bsr_cdev); | |
156 | device_del(cur->bsr_device); | |
157 | } | |
a0e2f9f4 SR |
158 | list_del(&cur->bsr_list); |
159 | kfree(cur); | |
fe9e8d53 | 160 | } |
fe9e8d53 SR |
161 | } |
162 | ||
a0e2f9f4 | 163 | static int bsr_add_node(struct device_node *bn) |
fe9e8d53 | 164 | { |
a0e2f9f4 | 165 | int bsr_stride_len, bsr_bytes_len, num_bsr_devs; |
fe9e8d53 SR |
166 | const u32 *bsr_stride; |
167 | const u32 *bsr_bytes; | |
168 | unsigned i; | |
a0e2f9f4 | 169 | int ret = -ENODEV; |
fe9e8d53 SR |
170 | |
171 | bsr_stride = of_get_property(bn, "ibm,lock-stride", &bsr_stride_len); | |
172 | bsr_bytes = of_get_property(bn, "ibm,#lock-bytes", &bsr_bytes_len); | |
173 | ||
174 | if (!bsr_stride || !bsr_bytes || | |
175 | (bsr_stride_len != bsr_bytes_len)) { | |
176 | printk(KERN_ERR "bsr of-node has missing/incorrect property\n"); | |
a0e2f9f4 | 177 | return ret; |
fe9e8d53 SR |
178 | } |
179 | ||
180 | num_bsr_devs = bsr_bytes_len / sizeof(u32); | |
181 | ||
fe9e8d53 | 182 | for (i = 0 ; i < num_bsr_devs; i++) { |
a0e2f9f4 SR |
183 | struct bsr_dev *cur = kzalloc(sizeof(struct bsr_dev), |
184 | GFP_KERNEL); | |
fe9e8d53 SR |
185 | struct resource res; |
186 | int result; | |
187 | ||
a0e2f9f4 SR |
188 | if (!cur) { |
189 | printk(KERN_ERR "Unable to alloc bsr dev\n"); | |
190 | ret = -ENOMEM; | |
191 | goto out_err; | |
192 | } | |
193 | ||
fe9e8d53 SR |
194 | result = of_address_to_resource(bn, i, &res); |
195 | if (result < 0) { | |
a0e2f9f4 SR |
196 | printk(KERN_ERR "bsr of-node has invalid reg property, skipping\n"); |
197 | kfree(cur); | |
198 | continue; | |
fe9e8d53 SR |
199 | } |
200 | ||
a0e2f9f4 | 201 | cur->bsr_minor = i + total_bsr_devs; |
fe9e8d53 SR |
202 | cur->bsr_addr = res.start; |
203 | cur->bsr_len = res.end - res.start + 1; | |
204 | cur->bsr_bytes = bsr_bytes[i]; | |
205 | cur->bsr_stride = bsr_stride[i]; | |
a0e2f9f4 | 206 | cur->bsr_dev = MKDEV(bsr_major, i + total_bsr_devs); |
fe9e8d53 SR |
207 | |
208 | switch(cur->bsr_bytes) { | |
209 | case 8: | |
210 | cur->bsr_type = BSR_8; | |
211 | break; | |
212 | case 16: | |
213 | cur->bsr_type = BSR_16; | |
214 | break; | |
215 | case 64: | |
216 | cur->bsr_type = BSR_64; | |
217 | break; | |
218 | case 128: | |
219 | cur->bsr_type = BSR_128; | |
220 | break; | |
221 | default: | |
222 | cur->bsr_type = BSR_UNKNOWN; | |
223 | printk(KERN_INFO "unknown BSR size %d\n",cur->bsr_bytes); | |
224 | } | |
225 | ||
226 | cur->bsr_num = bsr_types[cur->bsr_type]; | |
fe9e8d53 SR |
227 | snprintf(cur->bsr_name, 32, "bsr%d_%d", |
228 | cur->bsr_bytes, cur->bsr_num); | |
229 | ||
230 | cdev_init(&cur->bsr_cdev, &bsr_fops); | |
231 | result = cdev_add(&cur->bsr_cdev, cur->bsr_dev, 1); | |
a0e2f9f4 SR |
232 | if (result) { |
233 | kfree(cur); | |
fe9e8d53 | 234 | goto out_err; |
a0e2f9f4 | 235 | } |
fe9e8d53 | 236 | |
03457cd4 GKH |
237 | cur->bsr_device = device_create(bsr_class, NULL, cur->bsr_dev, |
238 | cur, cur->bsr_name); | |
fe9e8d53 SR |
239 | if (!cur->bsr_device) { |
240 | printk(KERN_ERR "device_create failed for %s\n", | |
241 | cur->bsr_name); | |
242 | cdev_del(&cur->bsr_cdev); | |
a0e2f9f4 | 243 | kfree(cur); |
fe9e8d53 SR |
244 | goto out_err; |
245 | } | |
a0e2f9f4 SR |
246 | |
247 | bsr_types[cur->bsr_type] = cur->bsr_num + 1; | |
248 | list_add_tail(&cur->bsr_list, &bsr_devs); | |
fe9e8d53 SR |
249 | } |
250 | ||
a0e2f9f4 SR |
251 | total_bsr_devs += num_bsr_devs; |
252 | ||
fe9e8d53 SR |
253 | return 0; |
254 | ||
255 | out_err: | |
256 | ||
257 | bsr_cleanup_devs(); | |
a0e2f9f4 SR |
258 | return ret; |
259 | } | |
260 | ||
261 | static int bsr_create_devs(struct device_node *bn) | |
262 | { | |
263 | int ret; | |
264 | ||
265 | while (bn) { | |
266 | ret = bsr_add_node(bn); | |
267 | if (ret) { | |
268 | of_node_put(bn); | |
269 | return ret; | |
270 | } | |
271 | bn = of_find_compatible_node(bn, NULL, "ibm,bsr"); | |
272 | } | |
273 | return 0; | |
fe9e8d53 SR |
274 | } |
275 | ||
276 | static int __init bsr_init(void) | |
277 | { | |
278 | struct device_node *np; | |
279 | dev_t bsr_dev = MKDEV(bsr_major, 0); | |
280 | int ret = -ENODEV; | |
281 | int result; | |
282 | ||
a0e2f9f4 | 283 | np = of_find_compatible_node(NULL, NULL, "ibm,bsr"); |
fe9e8d53 SR |
284 | if (!np) |
285 | goto out_err; | |
286 | ||
287 | bsr_class = class_create(THIS_MODULE, "bsr"); | |
288 | if (IS_ERR(bsr_class)) { | |
289 | printk(KERN_ERR "class_create() failed for bsr_class\n"); | |
290 | goto out_err_1; | |
291 | } | |
292 | bsr_class->dev_attrs = bsr_dev_attrs; | |
293 | ||
294 | result = alloc_chrdev_region(&bsr_dev, 0, BSR_MAX_DEVS, "bsr"); | |
295 | bsr_major = MAJOR(bsr_dev); | |
296 | if (result < 0) { | |
297 | printk(KERN_ERR "alloc_chrdev_region() failed for bsr\n"); | |
298 | goto out_err_2; | |
299 | } | |
300 | ||
a0e2f9f4 SR |
301 | if ((ret = bsr_create_devs(np)) < 0) { |
302 | np = NULL; | |
fe9e8d53 | 303 | goto out_err_3; |
a0e2f9f4 | 304 | } |
fe9e8d53 SR |
305 | |
306 | return 0; | |
307 | ||
308 | out_err_3: | |
309 | unregister_chrdev_region(bsr_dev, BSR_MAX_DEVS); | |
310 | ||
311 | out_err_2: | |
312 | class_destroy(bsr_class); | |
313 | ||
314 | out_err_1: | |
315 | of_node_put(np); | |
316 | ||
317 | out_err: | |
318 | ||
319 | return ret; | |
320 | } | |
321 | ||
322 | static void __exit bsr_exit(void) | |
323 | { | |
324 | ||
325 | bsr_cleanup_devs(); | |
326 | ||
327 | if (bsr_class) | |
328 | class_destroy(bsr_class); | |
329 | ||
330 | if (bsr_major) | |
331 | unregister_chrdev_region(MKDEV(bsr_major, 0), BSR_MAX_DEVS); | |
332 | } | |
333 | ||
334 | module_init(bsr_init); | |
335 | module_exit(bsr_exit); | |
336 | MODULE_LICENSE("GPL"); | |
337 | MODULE_AUTHOR("Sonny Rao <sonnyrao@us.ibm.com>"); |