[PATCH] ppc64: kexec support for ppc64
[linux-2.6] / arch / ppc64 / 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/hvcall.h>
17 #include <asm/prom.h>
18 #include <asm/systemcfg.h>
19 #include <asm/paca.h>
20 #include <asm/lppaca.h>
21 #include <asm/machdep.h>
22
23 static DEFINE_PER_CPU(struct cpu, cpu_devices);
24
25 /* SMT stuff */
26
27 #ifdef CONFIG_PPC_MULTIPLATFORM
28 /* default to snooze disabled */
29 DEFINE_PER_CPU(unsigned long, smt_snooze_delay);
30
31 static ssize_t store_smt_snooze_delay(struct sys_device *dev, const char *buf,
32                                       size_t count)
33 {
34         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
35         ssize_t ret;
36         unsigned long snooze;
37
38         ret = sscanf(buf, "%lu", &snooze);
39         if (ret != 1)
40                 return -EINVAL;
41
42         per_cpu(smt_snooze_delay, cpu->sysdev.id) = snooze;
43
44         return count;
45 }
46
47 static ssize_t show_smt_snooze_delay(struct sys_device *dev, char *buf)
48 {
49         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
50
51         return sprintf(buf, "%lu\n", per_cpu(smt_snooze_delay, cpu->sysdev.id));
52 }
53
54 static SYSDEV_ATTR(smt_snooze_delay, 0644, show_smt_snooze_delay,
55                    store_smt_snooze_delay);
56
57 /* Only parse OF options if the matching cmdline option was not specified */
58 static int smt_snooze_cmdline;
59
60 static int __init smt_setup(void)
61 {
62         struct device_node *options;
63         unsigned int *val;
64         unsigned int cpu;
65
66         if (!cpu_has_feature(CPU_FTR_SMT))
67                 return 1;
68
69         options = find_path_device("/options");
70         if (!options)
71                 return 1;
72
73         val = (unsigned int *)get_property(options, "ibm,smt-snooze-delay",
74                                            NULL);
75         if (!smt_snooze_cmdline && val) {
76                 for_each_cpu(cpu)
77                         per_cpu(smt_snooze_delay, cpu) = *val;
78         }
79
80         return 1;
81 }
82 __initcall(smt_setup);
83
84 static int __init setup_smt_snooze_delay(char *str)
85 {
86         unsigned int cpu;
87         int snooze;
88
89         if (!cpu_has_feature(CPU_FTR_SMT))
90                 return 1;
91
92         smt_snooze_cmdline = 1;
93
94         if (get_option(&str, &snooze)) {
95                 for_each_cpu(cpu)
96                         per_cpu(smt_snooze_delay, cpu) = snooze;
97         }
98
99         return 1;
100 }
101 __setup("smt-snooze-delay=", setup_smt_snooze_delay);
102
103 /*
104  * Enabling PMCs will slow partition context switch times so we only do
105  * it the first time we write to the PMCs.
106  */
107
108 static DEFINE_PER_CPU(char, pmcs_enabled);
109
110 void ppc64_enable_pmcs(void)
111 {
112         unsigned long hid0;
113 #ifdef CONFIG_PPC_PSERIES
114         unsigned long set, reset;
115         int ret;
116 #endif /* CONFIG_PPC_PSERIES */
117
118         /* Only need to enable them once */
119         if (__get_cpu_var(pmcs_enabled))
120                 return;
121
122         __get_cpu_var(pmcs_enabled) = 1;
123
124         switch (systemcfg->platform) {
125         case PLATFORM_PSERIES:
126         case PLATFORM_POWERMAC:
127                 hid0 = mfspr(HID0);
128                 hid0 |= 1UL << (63 - 20);
129
130                 /* POWER4 requires the following sequence */
131                 asm volatile(
132                              "sync\n"
133                              "mtspr     %1, %0\n"
134                              "mfspr     %0, %1\n"
135                              "mfspr     %0, %1\n"
136                              "mfspr     %0, %1\n"
137                              "mfspr     %0, %1\n"
138                              "mfspr     %0, %1\n"
139                              "mfspr     %0, %1\n"
140                              "isync" : "=&r" (hid0) : "i" (HID0), "0" (hid0):
141                              "memory");
142                 break;
143
144 #ifdef CONFIG_PPC_PSERIES
145         case PLATFORM_PSERIES_LPAR:
146                 set = 1UL << 63;
147                 reset = 0;
148                 ret = plpar_hcall_norets(H_PERFMON, set, reset);
149                 if (ret)
150                         printk(KERN_ERR "H_PERFMON call on cpu %u "
151                                "returned %d\n",
152                                smp_processor_id(), ret);
153                 break;
154 #endif /* CONFIG_PPC_PSERIES */
155
156         default:
157                 break;
158         }
159
160 #ifdef CONFIG_PPC_PSERIES
161         /* instruct hypervisor to maintain PMCs */
162         if (cur_cpu_spec->firmware_features & FW_FEATURE_SPLPAR)
163                 get_paca()->lppaca.pmcregs_in_use = 1;
164
165         /*
166          * On SMT machines we have to set the run latch in the ctrl register
167          * in order to make PMC6 spin.
168          */
169         if (cpu_has_feature(CPU_FTR_SMT))
170                 ppc64_runlatch_on();
171 #endif /* CONFIG_PPC_PSERIES */
172 }
173
174 #else
175
176 /* PMC stuff */
177 void ppc64_enable_pmcs(void)
178 {
179         /* XXX Implement for iseries */
180 }
181 #endif /* CONFIG_PPC_MULTIPLATFORM */
182
183 EXPORT_SYMBOL(ppc64_enable_pmcs);
184
185 /* XXX convert to rusty's on_one_cpu */
186 static unsigned long run_on_cpu(unsigned long cpu,
187                                 unsigned long (*func)(unsigned long),
188                                 unsigned long arg)
189 {
190         cpumask_t old_affinity = current->cpus_allowed;
191         unsigned long ret;
192
193         /* should return -EINVAL to userspace */
194         if (set_cpus_allowed(current, cpumask_of_cpu(cpu)))
195                 return 0;
196
197         ret = func(arg);
198
199         set_cpus_allowed(current, old_affinity);
200
201         return ret;
202 }
203
204 #define SYSFS_PMCSETUP(NAME, ADDRESS) \
205 static unsigned long read_##NAME(unsigned long junk) \
206 { \
207         return mfspr(ADDRESS); \
208 } \
209 static unsigned long write_##NAME(unsigned long val) \
210 { \
211         ppc64_enable_pmcs(); \
212         mtspr(ADDRESS, val); \
213         return 0; \
214 } \
215 static ssize_t show_##NAME(struct sys_device *dev, char *buf) \
216 { \
217         struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
218         unsigned long val = run_on_cpu(cpu->sysdev.id, read_##NAME, 0); \
219         return sprintf(buf, "%lx\n", val); \
220 } \
221 static ssize_t __attribute_used__ \
222         store_##NAME(struct sys_device *dev, const char *buf, size_t count) \
223 { \
224         struct cpu *cpu = container_of(dev, struct cpu, sysdev); \
225         unsigned long val; \
226         int ret = sscanf(buf, "%lx", &val); \
227         if (ret != 1) \
228                 return -EINVAL; \
229         run_on_cpu(cpu->sysdev.id, write_##NAME, val); \
230         return count; \
231 }
232
233 SYSFS_PMCSETUP(mmcr0, SPRN_MMCR0);
234 SYSFS_PMCSETUP(mmcr1, SPRN_MMCR1);
235 SYSFS_PMCSETUP(mmcra, SPRN_MMCRA);
236 SYSFS_PMCSETUP(pmc1, SPRN_PMC1);
237 SYSFS_PMCSETUP(pmc2, SPRN_PMC2);
238 SYSFS_PMCSETUP(pmc3, SPRN_PMC3);
239 SYSFS_PMCSETUP(pmc4, SPRN_PMC4);
240 SYSFS_PMCSETUP(pmc5, SPRN_PMC5);
241 SYSFS_PMCSETUP(pmc6, SPRN_PMC6);
242 SYSFS_PMCSETUP(pmc7, SPRN_PMC7);
243 SYSFS_PMCSETUP(pmc8, SPRN_PMC8);
244 SYSFS_PMCSETUP(purr, SPRN_PURR);
245
246 static SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0);
247 static SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1);
248 static SYSDEV_ATTR(mmcra, 0600, show_mmcra, store_mmcra);
249 static SYSDEV_ATTR(pmc1, 0600, show_pmc1, store_pmc1);
250 static SYSDEV_ATTR(pmc2, 0600, show_pmc2, store_pmc2);
251 static SYSDEV_ATTR(pmc3, 0600, show_pmc3, store_pmc3);
252 static SYSDEV_ATTR(pmc4, 0600, show_pmc4, store_pmc4);
253 static SYSDEV_ATTR(pmc5, 0600, show_pmc5, store_pmc5);
254 static SYSDEV_ATTR(pmc6, 0600, show_pmc6, store_pmc6);
255 static SYSDEV_ATTR(pmc7, 0600, show_pmc7, store_pmc7);
256 static SYSDEV_ATTR(pmc8, 0600, show_pmc8, store_pmc8);
257 static SYSDEV_ATTR(purr, 0600, show_purr, NULL);
258
259 static void register_cpu_online(unsigned int cpu)
260 {
261         struct cpu *c = &per_cpu(cpu_devices, cpu);
262         struct sys_device *s = &c->sysdev;
263
264 #ifndef CONFIG_PPC_ISERIES
265         if (cpu_has_feature(CPU_FTR_SMT))
266                 sysdev_create_file(s, &attr_smt_snooze_delay);
267 #endif
268
269         /* PMC stuff */
270
271         sysdev_create_file(s, &attr_mmcr0);
272         sysdev_create_file(s, &attr_mmcr1);
273
274         if (cpu_has_feature(CPU_FTR_MMCRA))
275                 sysdev_create_file(s, &attr_mmcra);
276
277         sysdev_create_file(s, &attr_pmc1);
278         sysdev_create_file(s, &attr_pmc2);
279         sysdev_create_file(s, &attr_pmc3);
280         sysdev_create_file(s, &attr_pmc4);
281         sysdev_create_file(s, &attr_pmc5);
282         sysdev_create_file(s, &attr_pmc6);
283
284         if (cpu_has_feature(CPU_FTR_PMC8)) {
285                 sysdev_create_file(s, &attr_pmc7);
286                 sysdev_create_file(s, &attr_pmc8);
287         }
288
289         if (cpu_has_feature(CPU_FTR_SMT))
290                 sysdev_create_file(s, &attr_purr);
291 }
292
293 #ifdef CONFIG_HOTPLUG_CPU
294 static void unregister_cpu_online(unsigned int cpu)
295 {
296         struct cpu *c = &per_cpu(cpu_devices, cpu);
297         struct sys_device *s = &c->sysdev;
298
299         BUG_ON(c->no_control);
300
301 #ifndef CONFIG_PPC_ISERIES
302         if (cpu_has_feature(CPU_FTR_SMT))
303                 sysdev_remove_file(s, &attr_smt_snooze_delay);
304 #endif
305
306         /* PMC stuff */
307
308         sysdev_remove_file(s, &attr_mmcr0);
309         sysdev_remove_file(s, &attr_mmcr1);
310
311         if (cpu_has_feature(CPU_FTR_MMCRA))
312                 sysdev_remove_file(s, &attr_mmcra);
313
314         sysdev_remove_file(s, &attr_pmc1);
315         sysdev_remove_file(s, &attr_pmc2);
316         sysdev_remove_file(s, &attr_pmc3);
317         sysdev_remove_file(s, &attr_pmc4);
318         sysdev_remove_file(s, &attr_pmc5);
319         sysdev_remove_file(s, &attr_pmc6);
320
321         if (cpu_has_feature(CPU_FTR_PMC8)) {
322                 sysdev_remove_file(s, &attr_pmc7);
323                 sysdev_remove_file(s, &attr_pmc8);
324         }
325
326         if (cpu_has_feature(CPU_FTR_SMT))
327                 sysdev_remove_file(s, &attr_purr);
328 }
329 #endif /* CONFIG_HOTPLUG_CPU */
330
331 static int __devinit sysfs_cpu_notify(struct notifier_block *self,
332                                       unsigned long action, void *hcpu)
333 {
334         unsigned int cpu = (unsigned int)(long)hcpu;
335
336         switch (action) {
337         case CPU_ONLINE:
338                 register_cpu_online(cpu);
339                 break;
340 #ifdef CONFIG_HOTPLUG_CPU
341         case CPU_DEAD:
342                 unregister_cpu_online(cpu);
343                 break;
344 #endif
345         }
346         return NOTIFY_OK;
347 }
348
349 static struct notifier_block __devinitdata sysfs_cpu_nb = {
350         .notifier_call  = sysfs_cpu_notify,
351 };
352
353 /* NUMA stuff */
354
355 #ifdef CONFIG_NUMA
356 static struct node node_devices[MAX_NUMNODES];
357
358 static void register_nodes(void)
359 {
360         int i;
361
362         for (i = 0; i < MAX_NUMNODES; i++) {
363                 if (node_online(i)) {
364                         int p_node = parent_node(i);
365                         struct node *parent = NULL;
366
367                         if (p_node != i)
368                                 parent = &node_devices[p_node];
369
370                         register_node(&node_devices[i], i, parent);
371                 }
372         }
373 }
374 #else
375 static void register_nodes(void)
376 {
377         return;
378 }
379 #endif
380
381 /* Only valid if CPU is present. */
382 static ssize_t show_physical_id(struct sys_device *dev, char *buf)
383 {
384         struct cpu *cpu = container_of(dev, struct cpu, sysdev);
385
386         return sprintf(buf, "%d\n", get_hard_smp_processor_id(cpu->sysdev.id));
387 }
388 static SYSDEV_ATTR(physical_id, 0444, show_physical_id, NULL);
389
390 static int __init topology_init(void)
391 {
392         int cpu;
393         struct node *parent = NULL;
394
395         register_nodes();
396
397         register_cpu_notifier(&sysfs_cpu_nb);
398
399         for_each_cpu(cpu) {
400                 struct cpu *c = &per_cpu(cpu_devices, cpu);
401
402 #ifdef CONFIG_NUMA
403                 parent = &node_devices[cpu_to_node(cpu)];
404 #endif
405                 /*
406                  * For now, we just see if the system supports making
407                  * the RTAS calls for CPU hotplug.  But, there may be a
408                  * more comprehensive way to do this for an individual
409                  * CPU.  For instance, the boot cpu might never be valid
410                  * for hotplugging.
411                  */
412                 if (!ppc_md.cpu_die)
413                         c->no_control = 1;
414
415                 if (cpu_online(cpu) || (c->no_control == 0)) {
416                         register_cpu(c, cpu, parent);
417
418                         sysdev_create_file(&c->sysdev, &attr_physical_id);
419                 }
420
421                 if (cpu_online(cpu))
422                         register_cpu_online(cpu);
423         }
424
425         return 0;
426 }
427 __initcall(topology_init);