blob: 349e0af9389ac4dd651c7301e1d961cc2e89257e [file] [log] [blame]
Michael Ellerman0332c2d2006-12-05 17:52:36 +11001/*
2 * pseries CPU Hotplug infrastructure.
3 *
Michael Ellerman413f7c42006-12-05 17:52:38 +11004 * Split out from arch/powerpc/platforms/pseries/setup.c
5 * arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c
Michael Ellerman0332c2d2006-12-05 17:52:36 +11006 *
7 * Peter Bergner, IBM March 2001.
8 * Copyright (C) 2001 IBM.
Michael Ellerman413f7c42006-12-05 17:52:38 +11009 * Dave Engebretsen, Peter Bergner, and
10 * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
11 * Plus various changes from other IBM teams...
Michael Ellerman0332c2d2006-12-05 17:52:36 +110012 *
13 * Copyright (C) 2006 Michael Ellerman, IBM Corporation
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 */
20
21#include <linux/kernel.h>
22#include <linux/delay.h>
23#include <linux/cpu.h>
24#include <asm/system.h>
25#include <asm/prom.h>
26#include <asm/rtas.h>
27#include <asm/firmware.h>
28#include <asm/machdep.h>
29#include <asm/vdso_datapage.h>
30#include <asm/pSeries_reconfig.h>
31#include "xics.h"
Michael Neuling473980a2008-01-11 14:02:47 +110032#include "plpar_wrappers.h"
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +000033#include "offline_states.h"
Michael Ellerman0332c2d2006-12-05 17:52:36 +110034
35/* This version can't take the spinlock, because it never returns */
36static struct rtas_args rtas_stop_self_args = {
37 .token = RTAS_UNKNOWN_SERVICE,
38 .nargs = 0,
39 .nret = 1,
40 .rets = &rtas_stop_self_args.args[0],
41};
42
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +000043static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
44 CPU_STATE_OFFLINE;
45static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
46
47static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
48
49static int cede_offline_enabled __read_mostly = 1;
50
51/*
52 * Enable/disable cede_offline when available.
53 */
54static int __init setup_cede_offline(char *str)
55{
56 if (!strcmp(str, "off"))
57 cede_offline_enabled = 0;
58 else if (!strcmp(str, "on"))
59 cede_offline_enabled = 1;
60 else
61 return 0;
62 return 1;
63}
64
65__setup("cede_offline=", setup_cede_offline);
66
67enum cpu_state_vals get_cpu_current_state(int cpu)
68{
69 return per_cpu(current_state, cpu);
70}
71
72void set_cpu_current_state(int cpu, enum cpu_state_vals state)
73{
74 per_cpu(current_state, cpu) = state;
75}
76
77enum cpu_state_vals get_preferred_offline_state(int cpu)
78{
79 return per_cpu(preferred_offline_state, cpu);
80}
81
82void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
83{
84 per_cpu(preferred_offline_state, cpu) = state;
85}
86
87void set_default_offline_state(int cpu)
88{
89 per_cpu(preferred_offline_state, cpu) = default_offline_state;
90}
91
Michael Ellerman04da6af2006-12-05 17:52:37 +110092static void rtas_stop_self(void)
Michael Ellerman0332c2d2006-12-05 17:52:36 +110093{
94 struct rtas_args *args = &rtas_stop_self_args;
95
96 local_irq_disable();
97
98 BUG_ON(args->token == RTAS_UNKNOWN_SERVICE);
99
100 printk("cpu %u (hwid %u) Ready to die...\n",
101 smp_processor_id(), hard_smp_processor_id());
102 enter_rtas(__pa(args));
103
104 panic("Alas, I survived.\n");
105}
106
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100107static void pseries_mach_cpu_die(void)
Michael Ellerman04da6af2006-12-05 17:52:37 +1100108{
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000109 unsigned int cpu = smp_processor_id();
110 unsigned int hwcpu = hard_smp_processor_id();
111 u8 cede_latency_hint = 0;
112
Michael Ellerman04da6af2006-12-05 17:52:37 +1100113 local_irq_disable();
114 idle_task_exit();
Nathan Fontenotc3e85062008-02-07 07:37:31 +1100115 xics_teardown_cpu();
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000116
117 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
118 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
119 cede_latency_hint = 2;
120
121 get_lppaca()->idle = 1;
122 if (!get_lppaca()->shared_proc)
123 get_lppaca()->donate_dedicated_cpu = 1;
124
125 printk(KERN_INFO
126 "cpu %u (hwid %u) ceding for offline with hint %d\n",
127 cpu, hwcpu, cede_latency_hint);
128 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
129 extended_cede_processor(cede_latency_hint);
130 printk(KERN_INFO "cpu %u (hwid %u) returned from cede.\n",
131 cpu, hwcpu);
132 printk(KERN_INFO
133 "Decrementer value = %x Timebase value = %llx\n",
134 get_dec(), get_tb());
135 }
136
137 printk(KERN_INFO "cpu %u (hwid %u) got prodded to go online\n",
138 cpu, hwcpu);
139
140 if (!get_lppaca()->shared_proc)
141 get_lppaca()->donate_dedicated_cpu = 0;
142 get_lppaca()->idle = 0;
143 }
144
145 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
146 unregister_slb_shadow(hwcpu, __pa(get_slb_shadow()));
147
148 /*
Vaidyanathan Srinivasan8dbce532010-03-01 02:58:09 +0000149 * Call to start_secondary_resume() will not return.
150 * Kernel stack will be reset and start_secondary()
151 * will be called to continue the online operation.
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000152 */
Vaidyanathan Srinivasan8dbce532010-03-01 02:58:09 +0000153 start_secondary_resume();
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000154
155 } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
156
157 set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
158 unregister_slb_shadow(hard_smp_processor_id(),
159 __pa(get_slb_shadow()));
160 rtas_stop_self();
161 }
162
Michael Ellerman04da6af2006-12-05 17:52:37 +1100163 /* Should never get here... */
164 BUG();
165 for(;;);
166}
167
Michael Ellerman674fa672006-12-05 17:52:38 +1100168static int qcss_tok; /* query-cpu-stopped-state token */
169
Michael Ellerman413f7c42006-12-05 17:52:38 +1100170/* Get state of physical CPU.
171 * Return codes:
172 * 0 - The processor is in the RTAS stopped state
173 * 1 - stop-self is in progress
174 * 2 - The processor is not in the RTAS stopped state
175 * -1 - Hardware Error
176 * -2 - Hardware Busy, Try again later.
177 */
178static int query_cpu_stopped(unsigned int pcpu)
179{
Michael Ellerman674fa672006-12-05 17:52:38 +1100180 int cpu_status, status;
Michael Ellerman413f7c42006-12-05 17:52:38 +1100181
Michael Ellerman413f7c42006-12-05 17:52:38 +1100182 status = rtas_call(qcss_tok, 1, 2, &cpu_status, pcpu);
183 if (status != 0) {
184 printk(KERN_ERR
185 "RTAS query-cpu-stopped-state failed: %i\n", status);
186 return status;
187 }
188
189 return cpu_status;
190}
191
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100192static int pseries_cpu_disable(void)
Michael Ellerman413f7c42006-12-05 17:52:38 +1100193{
194 int cpu = smp_processor_id();
195
Rusty Russellea0f1ca2009-09-24 09:34:48 -0600196 set_cpu_online(cpu, false);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100197 vdso_data->processorCount--;
198
199 /*fix boot_cpuid here*/
200 if (cpu == boot_cpuid)
201 boot_cpuid = any_online_cpu(cpu_online_map);
202
203 /* FIXME: abstract this to not be platform specific later on */
204 xics_migrate_irqs_away();
205 return 0;
206}
207
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000208/*
209 * pseries_cpu_die: Wait for the cpu to die.
210 * @cpu: logical processor id of the CPU whose death we're awaiting.
211 *
212 * This function is called from the context of the thread which is performing
213 * the cpu-offline. Here we wait for long enough to allow the cpu in question
214 * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
215 * notifications.
216 *
217 * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
218 * self-destruct.
219 */
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100220static void pseries_cpu_die(unsigned int cpu)
Michael Ellerman413f7c42006-12-05 17:52:38 +1100221{
222 int tries;
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000223 int cpu_status = 1;
Michael Ellerman413f7c42006-12-05 17:52:38 +1100224 unsigned int pcpu = get_hard_smp_processor_id(cpu);
225
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000226 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
227 cpu_status = 1;
228 for (tries = 0; tries < 1000; tries++) {
229 if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
230 cpu_status = 0;
231 break;
232 }
233 cpu_relax();
234 }
235 } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
236
237 for (tries = 0; tries < 25; tries++) {
238 cpu_status = query_cpu_stopped(pcpu);
239 if (cpu_status == 0 || cpu_status == -1)
240 break;
241 cpu_relax();
242 }
Michael Ellerman413f7c42006-12-05 17:52:38 +1100243 }
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000244
Michael Ellerman413f7c42006-12-05 17:52:38 +1100245 if (cpu_status != 0) {
246 printk("Querying DEAD? cpu %i (%i) shows %i\n",
247 cpu, pcpu, cpu_status);
248 }
249
250 /* Isolation and deallocation are definatly done by
251 * drslot_chrp_cpu. If they were not they would be
252 * done here. Change isolate state to Isolate and
253 * change allocation-state to Unusable.
254 */
255 paca[cpu].cpu_start = 0;
256}
257
258/*
259 * Update cpu_present_map and paca(s) for a new cpu node. The wrinkle
260 * here is that a cpu device node may represent up to two logical cpus
261 * in the SMT case. We must honor the assumption in other code that
262 * the logical ids for sibling SMT threads x and y are adjacent, such
263 * that x^1 == y and y^1 == x.
264 */
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100265static int pseries_add_processor(struct device_node *np)
Michael Ellerman413f7c42006-12-05 17:52:38 +1100266{
267 unsigned int cpu;
268 cpumask_t candidate_map, tmp = CPU_MASK_NONE;
269 int err = -ENOSPC, len, nthreads, i;
270 const u32 *intserv;
271
Stephen Rothwelle2eb6392007-04-03 22:26:41 +1000272 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100273 if (!intserv)
274 return 0;
275
276 nthreads = len / sizeof(u32);
277 for (i = 0; i < nthreads; i++)
278 cpu_set(i, tmp);
279
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100280 cpu_maps_update_begin();
Michael Ellerman413f7c42006-12-05 17:52:38 +1100281
282 BUG_ON(!cpus_subset(cpu_present_map, cpu_possible_map));
283
284 /* Get a bitmap of unoccupied slots. */
285 cpus_xor(candidate_map, cpu_possible_map, cpu_present_map);
286 if (cpus_empty(candidate_map)) {
287 /* If we get here, it most likely means that NR_CPUS is
288 * less than the partition's max processors setting.
289 */
290 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
291 " supports %d logical cpus.\n", np->full_name,
292 cpus_weight(cpu_possible_map));
293 goto out_unlock;
294 }
295
296 while (!cpus_empty(tmp))
297 if (cpus_subset(tmp, candidate_map))
298 /* Found a range where we can insert the new cpu(s) */
299 break;
300 else
301 cpus_shift_left(tmp, tmp, nthreads);
302
303 if (cpus_empty(tmp)) {
304 printk(KERN_ERR "Unable to find space in cpu_present_map for"
305 " processor %s with %d thread(s)\n", np->name,
306 nthreads);
307 goto out_unlock;
308 }
309
310 for_each_cpu_mask(cpu, tmp) {
311 BUG_ON(cpu_isset(cpu, cpu_present_map));
Rusty Russellea0f1ca2009-09-24 09:34:48 -0600312 set_cpu_present(cpu, true);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100313 set_hard_smp_processor_id(cpu, *intserv++);
314 }
315 err = 0;
316out_unlock:
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100317 cpu_maps_update_done();
Michael Ellerman413f7c42006-12-05 17:52:38 +1100318 return err;
319}
320
321/*
322 * Update the present map for a cpu node which is going away, and set
323 * the hard id in the paca(s) to -1 to be consistent with boot time
324 * convention for non-present cpus.
325 */
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100326static void pseries_remove_processor(struct device_node *np)
Michael Ellerman413f7c42006-12-05 17:52:38 +1100327{
328 unsigned int cpu;
329 int len, nthreads, i;
330 const u32 *intserv;
331
Stephen Rothwelle2eb6392007-04-03 22:26:41 +1000332 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100333 if (!intserv)
334 return;
335
336 nthreads = len / sizeof(u32);
337
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100338 cpu_maps_update_begin();
Michael Ellerman413f7c42006-12-05 17:52:38 +1100339 for (i = 0; i < nthreads; i++) {
340 for_each_present_cpu(cpu) {
341 if (get_hard_smp_processor_id(cpu) != intserv[i])
342 continue;
343 BUG_ON(cpu_online(cpu));
Rusty Russellea0f1ca2009-09-24 09:34:48 -0600344 set_cpu_present(cpu, false);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100345 set_hard_smp_processor_id(cpu, -1);
346 break;
347 }
348 if (cpu == NR_CPUS)
349 printk(KERN_WARNING "Could not find cpu to remove "
350 "with physical id 0x%x\n", intserv[i]);
351 }
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100352 cpu_maps_update_done();
Michael Ellerman413f7c42006-12-05 17:52:38 +1100353}
354
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100355static int pseries_smp_notifier(struct notifier_block *nb,
356 unsigned long action, void *node)
Michael Ellerman413f7c42006-12-05 17:52:38 +1100357{
358 int err = NOTIFY_OK;
359
360 switch (action) {
361 case PSERIES_RECONFIG_ADD:
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100362 if (pseries_add_processor(node))
Michael Ellerman413f7c42006-12-05 17:52:38 +1100363 err = NOTIFY_BAD;
364 break;
365 case PSERIES_RECONFIG_REMOVE:
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100366 pseries_remove_processor(node);
Michael Ellerman413f7c42006-12-05 17:52:38 +1100367 break;
368 default:
369 err = NOTIFY_DONE;
370 break;
371 }
372 return err;
373}
374
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100375static struct notifier_block pseries_smp_nb = {
376 .notifier_call = pseries_smp_notifier,
Michael Ellerman413f7c42006-12-05 17:52:38 +1100377};
378
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000379#define MAX_CEDE_LATENCY_LEVELS 4
380#define CEDE_LATENCY_PARAM_LENGTH 10
381#define CEDE_LATENCY_PARAM_MAX_LENGTH \
382 (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
383#define CEDE_LATENCY_TOKEN 45
384
385static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
386
387static int parse_cede_parameters(void)
388{
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000389 memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
Anton Blanchard20a8ab92010-02-07 13:52:05 +0000390 return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
391 NULL,
392 CEDE_LATENCY_TOKEN,
393 __pa(cede_parameters),
394 CEDE_LATENCY_PARAM_MAX_LENGTH);
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000395}
396
Michael Ellerman0332c2d2006-12-05 17:52:36 +1100397static int __init pseries_cpu_hotplug_init(void)
398{
Olof Johansson64f27582007-10-10 10:38:24 +1000399 struct device_node *np;
400 const char *typep;
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000401 int cpu;
Olof Johansson64f27582007-10-10 10:38:24 +1000402
403 for_each_node_by_name(np, "interrupt-controller") {
404 typep = of_get_property(np, "compatible", NULL);
405 if (strstr(typep, "open-pic")) {
406 of_node_put(np);
407
408 printk(KERN_INFO "CPU Hotplug not supported on "
409 "systems using MPIC\n");
410 return 0;
411 }
412 }
413
Michael Ellerman0332c2d2006-12-05 17:52:36 +1100414 rtas_stop_self_args.token = rtas_token("stop-self");
Michael Ellerman674fa672006-12-05 17:52:38 +1100415 qcss_tok = rtas_token("query-cpu-stopped-state");
416
417 if (rtas_stop_self_args.token == RTAS_UNKNOWN_SERVICE ||
418 qcss_tok == RTAS_UNKNOWN_SERVICE) {
419 printk(KERN_INFO "CPU Hotplug not supported by firmware "
420 "- disabling.\n");
421 return 0;
422 }
Michael Ellerman0332c2d2006-12-05 17:52:36 +1100423
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100424 ppc_md.cpu_die = pseries_mach_cpu_die;
425 smp_ops->cpu_disable = pseries_cpu_disable;
426 smp_ops->cpu_die = pseries_cpu_die;
Michael Ellerman413f7c42006-12-05 17:52:38 +1100427
428 /* Processors can be added/removed only on LPAR */
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000429 if (firmware_has_feature(FW_FEATURE_LPAR)) {
Michael Ellerman06ba30b2006-12-05 17:52:39 +1100430 pSeries_reconfig_notifier_register(&pseries_smp_nb);
Gautham R Shenoy3aa565f2009-10-29 19:22:53 +0000431 cpu_maps_update_begin();
432 if (cede_offline_enabled && parse_cede_parameters() == 0) {
433 default_offline_state = CPU_STATE_INACTIVE;
434 for_each_online_cpu(cpu)
435 set_default_offline_state(cpu);
436 }
437 cpu_maps_update_done();
438 }
Michael Ellerman413f7c42006-12-05 17:52:38 +1100439
Michael Ellerman0332c2d2006-12-05 17:52:36 +1100440 return 0;
441}
442arch_initcall(pseries_cpu_hotplug_init);