Merge git://git.infradead.org/~dwmw2/rbtree-2.6
[linux-2.6] / arch / powerpc / kernel / sysfs.c
1 #include <linux/config.h>
2 #include <linux/sysdev.h>
3 #include <linux/cpu.h>
4 #include <linux/smp.h>
5 #include <linux/percpu.h>
6 #include <linux/init.h>
7 #include <linux/sched.h>
8 #include <linux/module.h>
9 #include <linux/nodemask.h>
10 #include <linux/cpumask.h>
11 #include <linux/notifier.h>
12
13 #include <asm/current.h>
14 #include <asm/processor.h>
15 #include <asm/cputable.h>
16 #include <asm/firmware.h>
17 #include <asm/hvcall.h>
18 #include <asm/prom.h>
19 #include <asm/paca.h>
20 #include <asm/lppaca.h>
21 #include <asm/machdep.h>
22 #include <asm/smp.h>
23
24 static DEFINE_PER_CPU(struct cpu, cpu_devices);
25
26 /* SMT stuff */
27
28 #ifdef CONFIG_PPC_MULTIPLATFORM
29 /* default to snooze disabled */
30 DEFINE_PER_CPU(unsigned long, smt_snooze_delay);
31
32 static ssize_t store_smt_snooze_delay(struct sys_device *dev, const char *buf,
33                                       size_t count)
34 {
35         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
36         ssize_t ret;
37         unsigned long snooze;
38
39         ret = sscanf(buf, "%lu", &snooze);
40         if (ret != 1)
41                 return -EINVAL;
42
43         per_cpu(smt_snooze_delay, cpu->sysdev.id) = snooze;
44
45         return count;
46 }
47
48 static ssize_t show_smt_snooze_delay(struct sys_device *dev, char *buf)
49 {
50         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
51
52         return sprintf(buf, "%lu\n", per_cpu(smt_snooze_delay, cpu->sysdev.id));
53 }
54
55 static SYSDEV_ATTR(smt_snooze_delay, 0644, show_smt_snooze_delay,
56                    store_smt_snooze_delay);
57
58 /* Only parse OF options if the matching cmdline option was not specified */
59 static int smt_snooze_cmdline;
60
61 static int __init smt_setup(void)
62 {
63         struct device_node *options;
64         unsigned int *val;
65         unsigned int cpu;
66
67         if (!cpu_has_feature(CPU_FTR_SMT))
68                 return -ENODEV;
69
70         options = find_path_device("/options");
71         if (!options)
72                 return -ENODEV;
73
74         val = (unsigned int *)get_property(options, "ibm,smt-snooze-delay",
75                                            NULL);
76         if (!smt_snooze_cmdline && val) {
77                 for_each_possible_cpu(cpu)
78                         per_cpu(smt_snooze_delay, cpu) = *val;
79         }
80
81         return 0;
82 }
83 __initcall(smt_setup);
84
85 static int __init setup_smt_snooze_delay(char *str)
86 {
87         unsigned int cpu;
88         int snooze;
89
90         if (!cpu_has_feature(CPU_FTR_SMT))
91                 return 1;
92
93         smt_snooze_cmdline = 1;
94
95         if (get_option(&str, &snooze)) {
96                 for_each_possible_cpu(cpu)
97                         per_cpu(smt_snooze_delay, cpu) = snooze;
98         }
99
100         return 1;
101 }
102 __setup("smt-snooze-delay=", setup_smt_snooze_delay);
103
104 #endif /* CONFIG_PPC_MULTIPLATFORM */
105
106 /*
107  * Enabling PMCs will slow partition context switch times so we only do
108  * it the first time we write to the PMCs.
109  */
110
111 static DEFINE_PER_CPU(char, pmcs_enabled);
112
113 void ppc64_enable_pmcs(void)
114 {
115         /* Only need to enable them once */
116         if (__get_cpu_var(pmcs_enabled))
117                 return;
118
119         __get_cpu_var(pmcs_enabled) = 1;
120
121         if (ppc_md.enable_pmcs)
122                 ppc_md.enable_pmcs();
123 }
124 EXPORT_SYMBOL(ppc64_enable_pmcs);
125
126 /* XXX convert to rusty's on_one_cpu */
127 static unsigned long run_on_cpu(unsigned long cpu,
128                                 unsigned long (*func)(unsigned long),
129                                 unsigned long arg)
130 {
131         cpumask_t old_affinity = current->cpus_allowed;
132         unsigned long ret;
133
134         /* should return -EINVAL to userspace */
135         if (set_cpus_allowed(current, cpumask_of_cpu(cpu)))
136                 return 0;
137
138         ret = func(arg);
139
140         set_cpus_allowed(current, old_affinity);
141
142         return ret;
143 }
144
145 #define SYSFS_PMCSETUP(NAME, ADDRESS) \
146 static unsigned long read_##NAME(unsigned long junk) \
147 { \
148         return mfspr(ADDRESS); \
149 } \
150 static unsigned long write_##NAME(unsigned long val) \
151 { \
152         ppc64_enable_pmcs(); \
153         mtspr(ADDRESS, val); \
154         return 0; \
155 } \
156 static ssize_t show_##NAME(struct sys_device *dev, char *buf) \
157 { \
158         struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
159         unsigned long val = run_on_cpu(cpu->sysdev.id, read_##NAME, 0); \
160         return sprintf(buf, "%lx\n", val); \
161 } \
162 static ssize_t __attribute_used__ \
163         store_##NAME(struct sys_device *dev, const char *buf, size_t count) \
164 { \
165         struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
166         unsigned long val; \
167         int ret = sscanf(buf, "%lx", &val); \
168         if (ret != 1) \
169                 return -EINVAL; \
170         run_on_cpu(cpu->sysdev.id, write_##NAME, val); \
171         return count; \
172 }
173
174 SYSFS_PMCSETUP(mmcr0, SPRN_MMCR0);
175 SYSFS_PMCSETUP(mmcr1, SPRN_MMCR1);
176 SYSFS_PMCSETUP(mmcra, SPRN_MMCRA);
177 SYSFS_PMCSETUP(pmc1, SPRN_PMC1);
178 SYSFS_PMCSETUP(pmc2, SPRN_PMC2);
179 SYSFS_PMCSETUP(pmc3, SPRN_PMC3);
180 SYSFS_PMCSETUP(pmc4, SPRN_PMC4);
181 SYSFS_PMCSETUP(pmc5, SPRN_PMC5);
182 SYSFS_PMCSETUP(pmc6, SPRN_PMC6);
183 SYSFS_PMCSETUP(pmc7, SPRN_PMC7);
184 SYSFS_PMCSETUP(pmc8, SPRN_PMC8);
185 SYSFS_PMCSETUP(purr, SPRN_PURR);
186
187 static SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0);
188 static SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1);
189 static SYSDEV_ATTR(mmcra, 0600, show_mmcra, store_mmcra);
190 static SYSDEV_ATTR(pmc1, 0600, show_pmc1, store_pmc1);
191 static SYSDEV_ATTR(pmc2, 0600, show_pmc2, store_pmc2);
192 static SYSDEV_ATTR(pmc3, 0600, show_pmc3, store_pmc3);
193 static SYSDEV_ATTR(pmc4, 0600, show_pmc4, store_pmc4);
194 static SYSDEV_ATTR(pmc5, 0600, show_pmc5, store_pmc5);
195 static SYSDEV_ATTR(pmc6, 0600, show_pmc6, store_pmc6);
196 static SYSDEV_ATTR(pmc7, 0600, show_pmc7, store_pmc7);
197 static SYSDEV_ATTR(pmc8, 0600, show_pmc8, store_pmc8);
198 static SYSDEV_ATTR(purr, 0600, show_purr, NULL);
199
200 static void register_cpu_online(unsigned int cpu)
201 {
202         struct cpu *c = &per_cpu(cpu_devices, cpu);
203         struct sys_device *s = &c->sysdev;
204
205 #ifndef CONFIG_PPC_ISERIES
206         if (cpu_has_feature(CPU_FTR_SMT))
207                 sysdev_create_file(s, &attr_smt_snooze_delay);
208 #endif
209
210         /* PMC stuff */
211
212         sysdev_create_file(s, &attr_mmcr0);
213         sysdev_create_file(s, &attr_mmcr1);
214
215         if (cpu_has_feature(CPU_FTR_MMCRA))
216                 sysdev_create_file(s, &attr_mmcra);
217
218         if (cur_cpu_spec->num_pmcs >= 1)
219                 sysdev_create_file(s, &attr_pmc1);
220         if (cur_cpu_spec->num_pmcs >= 2)
221                 sysdev_create_file(s, &attr_pmc2);
222         if (cur_cpu_spec->num_pmcs >= 3)
223                 sysdev_create_file(s, &attr_pmc3);
224         if (cur_cpu_spec->num_pmcs >= 4)
225                 sysdev_create_file(s, &attr_pmc4);
226         if (cur_cpu_spec->num_pmcs >= 5)
227                 sysdev_create_file(s, &attr_pmc5);
228         if (cur_cpu_spec->num_pmcs >= 6)
229                 sysdev_create_file(s, &attr_pmc6);
230         if (cur_cpu_spec->num_pmcs >= 7)
231                 sysdev_create_file(s, &attr_pmc7);
232         if (cur_cpu_spec->num_pmcs >= 8)
233                 sysdev_create_file(s, &attr_pmc8);
234
235         if (cpu_has_feature(CPU_FTR_SMT))
236                 sysdev_create_file(s, &attr_purr);
237 }
238
239 #ifdef CONFIG_HOTPLUG_CPU
240 static void unregister_cpu_online(unsigned int cpu)
241 {
242         struct cpu *c = &per_cpu(cpu_devices, cpu);
243         struct sys_device *s = &c->sysdev;
244
245         BUG_ON(c->no_control);
246
247 #ifndef CONFIG_PPC_ISERIES
248         if (cpu_has_feature(CPU_FTR_SMT))
249                 sysdev_remove_file(s, &attr_smt_snooze_delay);
250 #endif
251
252         /* PMC stuff */
253
254         sysdev_remove_file(s, &attr_mmcr0);
255         sysdev_remove_file(s, &attr_mmcr1);
256
257         if (cpu_has_feature(CPU_FTR_MMCRA))
258                 sysdev_remove_file(s, &attr_mmcra);
259
260         if (cur_cpu_spec->num_pmcs >= 1)
261                 sysdev_remove_file(s, &attr_pmc1);
262         if (cur_cpu_spec->num_pmcs >= 2)
263                 sysdev_remove_file(s, &attr_pmc2);
264         if (cur_cpu_spec->num_pmcs >= 3)
265                 sysdev_remove_file(s, &attr_pmc3);
266         if (cur_cpu_spec->num_pmcs >= 4)
267                 sysdev_remove_file(s, &attr_pmc4);
268         if (cur_cpu_spec->num_pmcs >= 5)
269                 sysdev_remove_file(s, &attr_pmc5);
270         if (cur_cpu_spec->num_pmcs >= 6)
271                 sysdev_remove_file(s, &attr_pmc6);
272         if (cur_cpu_spec->num_pmcs >= 7)
273                 sysdev_remove_file(s, &attr_pmc7);
274         if (cur_cpu_spec->num_pmcs >= 8)
275                 sysdev_remove_file(s, &attr_pmc8);
276
277         if (cpu_has_feature(CPU_FTR_SMT))
278                 sysdev_remove_file(s, &attr_purr);
279 }
280 #endif /* CONFIG_HOTPLUG_CPU */
281
282 static int sysfs_cpu_notify(struct notifier_block *self,
283                                       unsigned long action, void *hcpu)
284 {
285         unsigned int cpu = (unsigned int)(long)hcpu;
286
287         switch (action) {
288         case CPU_ONLINE:
289                 register_cpu_online(cpu);
290                 break;
291 #ifdef CONFIG_HOTPLUG_CPU
292         case CPU_DEAD:
293                 unregister_cpu_online(cpu);
294                 break;
295 #endif
296         }
297         return NOTIFY_OK;
298 }
299
300 static struct notifier_block sysfs_cpu_nb = {
301         .notifier_call  = sysfs_cpu_notify,
302 };
303
304 /* NUMA stuff */
305
306 #ifdef CONFIG_NUMA
307 static struct node node_devices[MAX_NUMNODES];
308
309 static void register_nodes(void)
310 {
311         int i;
312
313         for (i = 0; i < MAX_NUMNODES; i++) {
314                 if (node_online(i)) {
315                         int p_node = parent_node(i);
316                         struct node *parent = NULL;
317
318                         if (p_node != i)
319                                 parent = &node_devices[p_node];
320
321                         register_node(&node_devices[i], i, parent);
322                 }
323         }
324 }
325
326 int sysfs_add_device_to_node(struct sys_device *dev, int nid)
327 {
328         struct node *node = &node_devices[nid];
329         return sysfs_create_link(&node->sysdev.kobj, &dev->kobj,
330                         kobject_name(&dev->kobj));
331 }
332
333 void sysfs_remove_device_from_node(struct sys_device *dev, int nid)
334 {
335         struct node *node = &node_devices[nid];
336         sysfs_remove_link(&node->sysdev.kobj, kobject_name(&dev->kobj));
337 }
338
339 #else
340 static void register_nodes(void)
341 {
342         return;
343 }
344
345 #endif
346
347 EXPORT_SYMBOL_GPL(sysfs_add_device_to_node);
348 EXPORT_SYMBOL_GPL(sysfs_remove_device_from_node);
349
350 /* Only valid if CPU is present. */
351 static ssize_t show_physical_id(struct sys_device *dev, char *buf)
352 {
353         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
354
355         return sprintf(buf, "%d\n", get_hard_smp_processor_id(cpu->sysdev.id));
356 }
357 static SYSDEV_ATTR(physical_id, 0444, show_physical_id, NULL);
358
359 static int __init topology_init(void)
360 {
361         int cpu;
362         struct node *parent = NULL;
363
364         register_nodes();
365
366         register_cpu_notifier(&sysfs_cpu_nb);
367
368         for_each_possible_cpu(cpu) {
369                 struct cpu *c = &per_cpu(cpu_devices, cpu);
370
371 #ifdef CONFIG_NUMA
372                 /* The node to which a cpu belongs can't be known
373                  * until the cpu is made present.
374                  */
375                 parent = NULL;
376                 if (cpu_present(cpu))
377                         parent = &node_devices[cpu_to_node(cpu)];
378 #endif
379                 /*
380                  * For now, we just see if the system supports making
381                  * the RTAS calls for CPU hotplug.  But, there may be a
382                  * more comprehensive way to do this for an individual
383                  * CPU.  For instance, the boot cpu might never be valid
384                  * for hotplugging.
385                  */
386                 if (!ppc_md.cpu_die)
387                         c->no_control = 1;
388
389                 if (cpu_online(cpu) || (c->no_control == 0)) {
390                         register_cpu(c, cpu, parent);
391
392                         sysdev_create_file(&c->sysdev, &attr_physical_id);
393                 }
394
395                 if (cpu_online(cpu))
396                         register_cpu_online(cpu);
397         }
398
399         return 0;
400 }
401 __initcall(topology_init);