Merge remote-tracking branch 'upstream/master' into kvm-devel
[linux-2.6/kvm.git] / arch / powerpc / platforms / pseries / dlpar.c
blobe9be25bc571bbb254b91ceea6639ff6ee0f053ef
1 /*
2 * Support for dynamic reconfiguration for PCI, Memory, and CPU
3 * Hotplug and Dynamic Logical Partitioning on RPA platforms.
5 * Copyright (C) 2009 Nathan Fontenot
6 * Copyright (C) 2009 IBM Corporation
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version
10 * 2 as published by the Free Software Foundation.
13 #include <linux/kernel.h>
14 #include <linux/kref.h>
15 #include <linux/notifier.h>
16 #include <linux/proc_fs.h>
17 #include <linux/spinlock.h>
18 #include <linux/cpu.h>
19 #include <linux/slab.h>
20 #include "offline_states.h"
22 #include <asm/prom.h>
23 #include <asm/machdep.h>
24 #include <asm/uaccess.h>
25 #include <asm/rtas.h>
26 #include <asm/pSeries_reconfig.h>
28 struct cc_workarea {
29 u32 drc_index;
30 u32 zero;
31 u32 name_offset;
32 u32 prop_length;
33 u32 prop_offset;
36 void dlpar_free_cc_property(struct property *prop)
38 kfree(prop->name);
39 kfree(prop->value);
40 kfree(prop);
43 static struct property *dlpar_parse_cc_property(struct cc_workarea *ccwa)
45 struct property *prop;
46 char *name;
47 char *value;
49 prop = kzalloc(sizeof(*prop), GFP_KERNEL);
50 if (!prop)
51 return NULL;
53 name = (char *)ccwa + ccwa->name_offset;
54 prop->name = kstrdup(name, GFP_KERNEL);
56 prop->length = ccwa->prop_length;
57 value = (char *)ccwa + ccwa->prop_offset;
58 prop->value = kmemdup(value, prop->length, GFP_KERNEL);
59 if (!prop->value) {
60 dlpar_free_cc_property(prop);
61 return NULL;
64 return prop;
67 static struct device_node *dlpar_parse_cc_node(struct cc_workarea *ccwa)
69 struct device_node *dn;
70 char *name;
72 dn = kzalloc(sizeof(*dn), GFP_KERNEL);
73 if (!dn)
74 return NULL;
76 /* The configure connector reported name does not contain a
77 * preceding '/', so we allocate a buffer large enough to
78 * prepend this to the full_name.
80 name = (char *)ccwa + ccwa->name_offset;
81 dn->full_name = kasprintf(GFP_KERNEL, "/%s", name);
82 if (!dn->full_name) {
83 kfree(dn);
84 return NULL;
87 return dn;
90 static void dlpar_free_one_cc_node(struct device_node *dn)
92 struct property *prop;
94 while (dn->properties) {
95 prop = dn->properties;
96 dn->properties = prop->next;
97 dlpar_free_cc_property(prop);
100 kfree(dn->full_name);
101 kfree(dn);
104 void dlpar_free_cc_nodes(struct device_node *dn)
106 if (dn->child)
107 dlpar_free_cc_nodes(dn->child);
109 if (dn->sibling)
110 dlpar_free_cc_nodes(dn->sibling);
112 dlpar_free_one_cc_node(dn);
115 #define NEXT_SIBLING 1
116 #define NEXT_CHILD 2
117 #define NEXT_PROPERTY 3
118 #define PREV_PARENT 4
119 #define MORE_MEMORY 5
120 #define CALL_AGAIN -2
121 #define ERR_CFG_USE -9003
123 struct device_node *dlpar_configure_connector(u32 drc_index)
125 struct device_node *dn;
126 struct device_node *first_dn = NULL;
127 struct device_node *last_dn = NULL;
128 struct property *property;
129 struct property *last_property = NULL;
130 struct cc_workarea *ccwa;
131 char *data_buf;
132 int cc_token;
133 int rc = -1;
135 cc_token = rtas_token("ibm,configure-connector");
136 if (cc_token == RTAS_UNKNOWN_SERVICE)
137 return NULL;
139 data_buf = kzalloc(RTAS_DATA_BUF_SIZE, GFP_KERNEL);
140 if (!data_buf)
141 return NULL;
143 ccwa = (struct cc_workarea *)&data_buf[0];
144 ccwa->drc_index = drc_index;
145 ccwa->zero = 0;
147 do {
148 /* Since we release the rtas_data_buf lock between configure
149 * connector calls we want to re-populate the rtas_data_buffer
150 * with the contents of the previous call.
152 spin_lock(&rtas_data_buf_lock);
154 memcpy(rtas_data_buf, data_buf, RTAS_DATA_BUF_SIZE);
155 rc = rtas_call(cc_token, 2, 1, NULL, rtas_data_buf, NULL);
156 memcpy(data_buf, rtas_data_buf, RTAS_DATA_BUF_SIZE);
158 spin_unlock(&rtas_data_buf_lock);
160 switch (rc) {
161 case NEXT_SIBLING:
162 dn = dlpar_parse_cc_node(ccwa);
163 if (!dn)
164 goto cc_error;
166 dn->parent = last_dn->parent;
167 last_dn->sibling = dn;
168 last_dn = dn;
169 break;
171 case NEXT_CHILD:
172 dn = dlpar_parse_cc_node(ccwa);
173 if (!dn)
174 goto cc_error;
176 if (!first_dn)
177 first_dn = dn;
178 else {
179 dn->parent = last_dn;
180 if (last_dn)
181 last_dn->child = dn;
184 last_dn = dn;
185 break;
187 case NEXT_PROPERTY:
188 property = dlpar_parse_cc_property(ccwa);
189 if (!property)
190 goto cc_error;
192 if (!last_dn->properties)
193 last_dn->properties = property;
194 else
195 last_property->next = property;
197 last_property = property;
198 break;
200 case PREV_PARENT:
201 last_dn = last_dn->parent;
202 break;
204 case CALL_AGAIN:
205 break;
207 case MORE_MEMORY:
208 case ERR_CFG_USE:
209 default:
210 printk(KERN_ERR "Unexpected Error (%d) "
211 "returned from configure-connector\n", rc);
212 goto cc_error;
214 } while (rc);
216 cc_error:
217 kfree(data_buf);
219 if (rc) {
220 if (first_dn)
221 dlpar_free_cc_nodes(first_dn);
223 return NULL;
226 return first_dn;
229 static struct device_node *derive_parent(const char *path)
231 struct device_node *parent;
232 char *last_slash;
234 last_slash = strrchr(path, '/');
235 if (last_slash == path) {
236 parent = of_find_node_by_path("/");
237 } else {
238 char *parent_path;
239 int parent_path_len = last_slash - path + 1;
240 parent_path = kmalloc(parent_path_len, GFP_KERNEL);
241 if (!parent_path)
242 return NULL;
244 strlcpy(parent_path, path, parent_path_len);
245 parent = of_find_node_by_path(parent_path);
246 kfree(parent_path);
249 return parent;
252 int dlpar_attach_node(struct device_node *dn)
254 #ifdef CONFIG_PROC_DEVICETREE
255 struct proc_dir_entry *ent;
256 #endif
257 int rc;
259 of_node_set_flag(dn, OF_DYNAMIC);
260 kref_init(&dn->kref);
261 dn->parent = derive_parent(dn->full_name);
262 if (!dn->parent)
263 return -ENOMEM;
265 rc = pSeries_reconfig_notify(PSERIES_RECONFIG_ADD, dn);
266 if (rc) {
267 printk(KERN_ERR "Failed to add device node %s\n",
268 dn->full_name);
269 return rc;
272 of_attach_node(dn);
274 #ifdef CONFIG_PROC_DEVICETREE
275 ent = proc_mkdir(strrchr(dn->full_name, '/') + 1, dn->parent->pde);
276 if (ent)
277 proc_device_tree_add_node(dn, ent);
278 #endif
280 of_node_put(dn->parent);
281 return 0;
284 int dlpar_detach_node(struct device_node *dn)
286 #ifdef CONFIG_PROC_DEVICETREE
287 struct device_node *parent = dn->parent;
288 struct property *prop = dn->properties;
290 while (prop) {
291 remove_proc_entry(prop->name, dn->pde);
292 prop = prop->next;
295 if (dn->pde)
296 remove_proc_entry(dn->pde->name, parent->pde);
297 #endif
299 pSeries_reconfig_notify(PSERIES_RECONFIG_REMOVE, dn);
300 of_detach_node(dn);
301 of_node_put(dn); /* Must decrement the refcount */
303 return 0;
306 #define DR_ENTITY_SENSE 9003
307 #define DR_ENTITY_PRESENT 1
308 #define DR_ENTITY_UNUSABLE 2
309 #define ALLOCATION_STATE 9003
310 #define ALLOC_UNUSABLE 0
311 #define ALLOC_USABLE 1
312 #define ISOLATION_STATE 9001
313 #define ISOLATE 0
314 #define UNISOLATE 1
316 int dlpar_acquire_drc(u32 drc_index)
318 int dr_status, rc;
320 rc = rtas_call(rtas_token("get-sensor-state"), 2, 2, &dr_status,
321 DR_ENTITY_SENSE, drc_index);
322 if (rc || dr_status != DR_ENTITY_UNUSABLE)
323 return -1;
325 rc = rtas_set_indicator(ALLOCATION_STATE, drc_index, ALLOC_USABLE);
326 if (rc)
327 return rc;
329 rc = rtas_set_indicator(ISOLATION_STATE, drc_index, UNISOLATE);
330 if (rc) {
331 rtas_set_indicator(ALLOCATION_STATE, drc_index, ALLOC_UNUSABLE);
332 return rc;
335 return 0;
338 int dlpar_release_drc(u32 drc_index)
340 int dr_status, rc;
342 rc = rtas_call(rtas_token("get-sensor-state"), 2, 2, &dr_status,
343 DR_ENTITY_SENSE, drc_index);
344 if (rc || dr_status != DR_ENTITY_PRESENT)
345 return -1;
347 rc = rtas_set_indicator(ISOLATION_STATE, drc_index, ISOLATE);
348 if (rc)
349 return rc;
351 rc = rtas_set_indicator(ALLOCATION_STATE, drc_index, ALLOC_UNUSABLE);
352 if (rc) {
353 rtas_set_indicator(ISOLATION_STATE, drc_index, UNISOLATE);
354 return rc;
357 return 0;
360 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
362 static int dlpar_online_cpu(struct device_node *dn)
364 int rc = 0;
365 unsigned int cpu;
366 int len, nthreads, i;
367 const u32 *intserv;
369 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
370 if (!intserv)
371 return -EINVAL;
373 nthreads = len / sizeof(u32);
375 cpu_maps_update_begin();
376 for (i = 0; i < nthreads; i++) {
377 for_each_present_cpu(cpu) {
378 if (get_hard_smp_processor_id(cpu) != intserv[i])
379 continue;
380 BUG_ON(get_cpu_current_state(cpu)
381 != CPU_STATE_OFFLINE);
382 cpu_maps_update_done();
383 rc = cpu_up(cpu);
384 if (rc)
385 goto out;
386 cpu_maps_update_begin();
388 break;
390 if (cpu == num_possible_cpus())
391 printk(KERN_WARNING "Could not find cpu to online "
392 "with physical id 0x%x\n", intserv[i]);
394 cpu_maps_update_done();
396 out:
397 return rc;
401 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
403 struct device_node *dn;
404 unsigned long drc_index;
405 char *cpu_name;
406 int rc;
408 cpu_hotplug_driver_lock();
409 rc = strict_strtoul(buf, 0, &drc_index);
410 if (rc) {
411 rc = -EINVAL;
412 goto out;
415 dn = dlpar_configure_connector(drc_index);
416 if (!dn) {
417 rc = -EINVAL;
418 goto out;
421 /* configure-connector reports cpus as living in the base
422 * directory of the device tree. CPUs actually live in the
423 * cpus directory so we need to fixup the full_name.
425 cpu_name = kasprintf(GFP_KERNEL, "/cpus%s", dn->full_name);
426 if (!cpu_name) {
427 dlpar_free_cc_nodes(dn);
428 rc = -ENOMEM;
429 goto out;
432 kfree(dn->full_name);
433 dn->full_name = cpu_name;
435 rc = dlpar_acquire_drc(drc_index);
436 if (rc) {
437 dlpar_free_cc_nodes(dn);
438 rc = -EINVAL;
439 goto out;
442 rc = dlpar_attach_node(dn);
443 if (rc) {
444 dlpar_release_drc(drc_index);
445 dlpar_free_cc_nodes(dn);
446 goto out;
449 rc = dlpar_online_cpu(dn);
450 out:
451 cpu_hotplug_driver_unlock();
453 return rc ? rc : count;
456 static int dlpar_offline_cpu(struct device_node *dn)
458 int rc = 0;
459 unsigned int cpu;
460 int len, nthreads, i;
461 const u32 *intserv;
463 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
464 if (!intserv)
465 return -EINVAL;
467 nthreads = len / sizeof(u32);
469 cpu_maps_update_begin();
470 for (i = 0; i < nthreads; i++) {
471 for_each_present_cpu(cpu) {
472 if (get_hard_smp_processor_id(cpu) != intserv[i])
473 continue;
475 if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
476 break;
478 if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
479 set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
480 cpu_maps_update_done();
481 rc = cpu_down(cpu);
482 if (rc)
483 goto out;
484 cpu_maps_update_begin();
485 break;
490 * The cpu is in CPU_STATE_INACTIVE.
491 * Upgrade it's state to CPU_STATE_OFFLINE.
493 set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
494 BUG_ON(plpar_hcall_norets(H_PROD, intserv[i])
495 != H_SUCCESS);
496 __cpu_die(cpu);
497 break;
499 if (cpu == num_possible_cpus())
500 printk(KERN_WARNING "Could not find cpu to offline "
501 "with physical id 0x%x\n", intserv[i]);
503 cpu_maps_update_done();
505 out:
506 return rc;
510 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
512 struct device_node *dn;
513 const u32 *drc_index;
514 int rc;
516 dn = of_find_node_by_path(buf);
517 if (!dn)
518 return -EINVAL;
520 drc_index = of_get_property(dn, "ibm,my-drc-index", NULL);
521 if (!drc_index) {
522 of_node_put(dn);
523 return -EINVAL;
526 cpu_hotplug_driver_lock();
527 rc = dlpar_offline_cpu(dn);
528 if (rc) {
529 of_node_put(dn);
530 rc = -EINVAL;
531 goto out;
534 rc = dlpar_release_drc(*drc_index);
535 if (rc) {
536 of_node_put(dn);
537 goto out;
540 rc = dlpar_detach_node(dn);
541 if (rc) {
542 dlpar_acquire_drc(*drc_index);
543 goto out;
546 of_node_put(dn);
547 out:
548 cpu_hotplug_driver_unlock();
549 return rc ? rc : count;
552 static int __init pseries_dlpar_init(void)
554 ppc_md.cpu_probe = dlpar_cpu_probe;
555 ppc_md.cpu_release = dlpar_cpu_release;
557 return 0;
559 machine_device_initcall(pseries, pseries_dlpar_init);
561 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */