2 * pseries CPU Hotplug infrastructure.
4 * Split out from arch/powerpc/platforms/pseries/setup.c
5 * arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c
7 * Peter Bergner, IBM March 2001.
8 * Copyright (C) 2001 IBM.
9 * Dave Engebretsen, Peter Bergner, and
10 * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
11 * Plus various changes from other IBM teams...
13 * Copyright (C) 2006 Michael Ellerman, IBM Corporation
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.
21 #include <linux/kernel.h>
22 #include <linux/delay.h>
23 #include <linux/cpu.h>
24 #include <asm/system.h>
27 #include <asm/firmware.h>
28 #include <asm/machdep.h>
29 #include <asm/vdso_datapage.h>
30 #include <asm/pSeries_reconfig.h>
32 #include "plpar_wrappers.h"
33 #include "offline_states.h"
35 /* This version can't take the spinlock, because it never returns */
36 static struct rtas_args rtas_stop_self_args = {
37 .token = RTAS_UNKNOWN_SERVICE,
40 .rets = &rtas_stop_self_args.args[0],
43 static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
45 static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
47 static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
49 static int cede_offline_enabled __read_mostly = 1;
52 * Enable/disable cede_offline when available.
54 static int __init setup_cede_offline(char *str)
56 if (!strcmp(str, "off"))
57 cede_offline_enabled = 0;
58 else if (!strcmp(str, "on"))
59 cede_offline_enabled = 1;
65 __setup("cede_offline=", setup_cede_offline);
67 enum cpu_state_vals get_cpu_current_state(int cpu)
69 return per_cpu(current_state, cpu);
72 void set_cpu_current_state(int cpu, enum cpu_state_vals state)
74 per_cpu(current_state, cpu) = state;
77 enum cpu_state_vals get_preferred_offline_state(int cpu)
79 return per_cpu(preferred_offline_state, cpu);
82 void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
84 per_cpu(preferred_offline_state, cpu) = state;
87 void set_default_offline_state(int cpu)
89 per_cpu(preferred_offline_state, cpu) = default_offline_state;
92 static void rtas_stop_self(void)
94 struct rtas_args *args = &rtas_stop_self_args;
98 BUG_ON(args->token == RTAS_UNKNOWN_SERVICE);
100 printk("cpu %u (hwid %u) Ready to die...\n",
101 smp_processor_id(), hard_smp_processor_id());
102 enter_rtas(__pa(args));
104 panic("Alas, I survived.\n");
107 static void pseries_mach_cpu_die(void)
109 unsigned int cpu = smp_processor_id();
110 unsigned int hwcpu = hard_smp_processor_id();
111 u8 cede_latency_hint = 0;
117 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
118 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
119 cede_latency_hint = 2;
121 get_lppaca()->idle = 1;
122 if (!get_lppaca()->shared_proc)
123 get_lppaca()->donate_dedicated_cpu = 1;
125 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
126 extended_cede_processor(cede_latency_hint);
129 if (!get_lppaca()->shared_proc)
130 get_lppaca()->donate_dedicated_cpu = 0;
131 get_lppaca()->idle = 0;
133 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
134 unregister_slb_shadow(hwcpu, __pa(get_slb_shadow()));
137 * Call to start_secondary_resume() will not return.
138 * Kernel stack will be reset and start_secondary()
139 * will be called to continue the online operation.
141 start_secondary_resume();
145 /* Requested state is CPU_STATE_OFFLINE at this point */
146 WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
148 set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
149 unregister_slb_shadow(hwcpu, __pa(get_slb_shadow()));
152 /* Should never get here... */
157 static int qcss_tok; /* query-cpu-stopped-state token */
159 /* Get state of physical CPU.
161 * 0 - The processor is in the RTAS stopped state
162 * 1 - stop-self is in progress
163 * 2 - The processor is not in the RTAS stopped state
164 * -1 - Hardware Error
165 * -2 - Hardware Busy, Try again later.
167 static int query_cpu_stopped(unsigned int pcpu)
169 int cpu_status, status;
171 status = rtas_call(qcss_tok, 1, 2, &cpu_status, pcpu);
174 "RTAS query-cpu-stopped-state failed: %i\n", status);
181 static int pseries_cpu_disable(void)
183 int cpu = smp_processor_id();
185 set_cpu_online(cpu, false);
186 vdso_data->processorCount--;
188 /*fix boot_cpuid here*/
189 if (cpu == boot_cpuid)
190 boot_cpuid = any_online_cpu(cpu_online_map);
192 /* FIXME: abstract this to not be platform specific later on */
193 xics_migrate_irqs_away();
198 * pseries_cpu_die: Wait for the cpu to die.
199 * @cpu: logical processor id of the CPU whose death we're awaiting.
201 * This function is called from the context of the thread which is performing
202 * the cpu-offline. Here we wait for long enough to allow the cpu in question
203 * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
206 * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
209 static void pseries_cpu_die(unsigned int cpu)
213 unsigned int pcpu = get_hard_smp_processor_id(cpu);
215 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
217 for (tries = 0; tries < 1000; tries++) {
218 if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
224 } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
226 for (tries = 0; tries < 25; tries++) {
227 cpu_status = query_cpu_stopped(pcpu);
228 if (cpu_status == 0 || cpu_status == -1)
234 if (cpu_status != 0) {
235 printk("Querying DEAD? cpu %i (%i) shows %i\n",
236 cpu, pcpu, cpu_status);
239 /* Isolation and deallocation are definatly done by
240 * drslot_chrp_cpu. If they were not they would be
241 * done here. Change isolate state to Isolate and
242 * change allocation-state to Unusable.
244 paca[cpu].cpu_start = 0;
248 * Update cpu_present_map and paca(s) for a new cpu node. The wrinkle
249 * here is that a cpu device node may represent up to two logical cpus
250 * in the SMT case. We must honor the assumption in other code that
251 * the logical ids for sibling SMT threads x and y are adjacent, such
252 * that x^1 == y and y^1 == x.
254 static int pseries_add_processor(struct device_node *np)
257 cpumask_t candidate_map, tmp = CPU_MASK_NONE;
258 int err = -ENOSPC, len, nthreads, i;
261 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
265 nthreads = len / sizeof(u32);
266 for (i = 0; i < nthreads; i++)
269 cpu_maps_update_begin();
271 BUG_ON(!cpus_subset(cpu_present_map, cpu_possible_map));
273 /* Get a bitmap of unoccupied slots. */
274 cpus_xor(candidate_map, cpu_possible_map, cpu_present_map);
275 if (cpus_empty(candidate_map)) {
276 /* If we get here, it most likely means that NR_CPUS is
277 * less than the partition's max processors setting.
279 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
280 " supports %d logical cpus.\n", np->full_name,
281 cpus_weight(cpu_possible_map));
285 while (!cpus_empty(tmp))
286 if (cpus_subset(tmp, candidate_map))
287 /* Found a range where we can insert the new cpu(s) */
290 cpus_shift_left(tmp, tmp, nthreads);
292 if (cpus_empty(tmp)) {
293 printk(KERN_ERR "Unable to find space in cpu_present_map for"
294 " processor %s with %d thread(s)\n", np->name,
299 for_each_cpu_mask(cpu, tmp) {
300 BUG_ON(cpu_isset(cpu, cpu_present_map));
301 set_cpu_present(cpu, true);
302 set_hard_smp_processor_id(cpu, *intserv++);
306 cpu_maps_update_done();
311 * Update the present map for a cpu node which is going away, and set
312 * the hard id in the paca(s) to -1 to be consistent with boot time
313 * convention for non-present cpus.
315 static void pseries_remove_processor(struct device_node *np)
318 int len, nthreads, i;
321 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
325 nthreads = len / sizeof(u32);
327 cpu_maps_update_begin();
328 for (i = 0; i < nthreads; i++) {
329 for_each_present_cpu(cpu) {
330 if (get_hard_smp_processor_id(cpu) != intserv[i])
332 BUG_ON(cpu_online(cpu));
333 set_cpu_present(cpu, false);
334 set_hard_smp_processor_id(cpu, -1);
338 printk(KERN_WARNING "Could not find cpu to remove "
339 "with physical id 0x%x\n", intserv[i]);
341 cpu_maps_update_done();
344 static int pseries_smp_notifier(struct notifier_block *nb,
345 unsigned long action, void *node)
350 case PSERIES_RECONFIG_ADD:
351 if (pseries_add_processor(node))
354 case PSERIES_RECONFIG_REMOVE:
355 pseries_remove_processor(node);
364 static struct notifier_block pseries_smp_nb = {
365 .notifier_call = pseries_smp_notifier,
368 #define MAX_CEDE_LATENCY_LEVELS 4
369 #define CEDE_LATENCY_PARAM_LENGTH 10
370 #define CEDE_LATENCY_PARAM_MAX_LENGTH \
371 (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
372 #define CEDE_LATENCY_TOKEN 45
374 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
376 static int parse_cede_parameters(void)
378 memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
379 return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
382 __pa(cede_parameters),
383 CEDE_LATENCY_PARAM_MAX_LENGTH);
386 static int __init pseries_cpu_hotplug_init(void)
388 struct device_node *np;
392 for_each_node_by_name(np, "interrupt-controller") {
393 typep = of_get_property(np, "compatible", NULL);
394 if (strstr(typep, "open-pic")) {
397 printk(KERN_INFO "CPU Hotplug not supported on "
398 "systems using MPIC\n");
403 rtas_stop_self_args.token = rtas_token("stop-self");
404 qcss_tok = rtas_token("query-cpu-stopped-state");
406 if (rtas_stop_self_args.token == RTAS_UNKNOWN_SERVICE ||
407 qcss_tok == RTAS_UNKNOWN_SERVICE) {
408 printk(KERN_INFO "CPU Hotplug not supported by firmware "
413 ppc_md.cpu_die = pseries_mach_cpu_die;
414 smp_ops->cpu_disable = pseries_cpu_disable;
415 smp_ops->cpu_die = pseries_cpu_die;
417 /* Processors can be added/removed only on LPAR */
418 if (firmware_has_feature(FW_FEATURE_LPAR)) {
419 pSeries_reconfig_notifier_register(&pseries_smp_nb);
420 cpu_maps_update_begin();
421 if (cede_offline_enabled && parse_cede_parameters() == 0) {
422 default_offline_state = CPU_STATE_INACTIVE;
423 for_each_online_cpu(cpu)
424 set_default_offline_state(cpu);
426 cpu_maps_update_done();
431 arch_initcall(pseries_cpu_hotplug_init);