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 #define pr_fmt(fmt) "pseries-hotplug-cpu: " fmt
23 #include <linux/kernel.h>
24 #include <linux/interrupt.h>
25 #include <linux/delay.h>
26 #include <linux/sched.h> /* for idle_task_exit */
27 #include <linux/sched/hotplug.h>
28 #include <linux/cpu.h>
30 #include <linux/slab.h>
33 #include <asm/firmware.h>
34 #include <asm/machdep.h>
35 #include <asm/vdso_datapage.h>
37 #include <asm/plpar_wrappers.h>
40 #include "offline_states.h"
42 /* This version can't take the spinlock, because it never returns */
43 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
45 static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
47 static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
49 static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
51 static bool cede_offline_enabled __read_mostly = true;
54 * Enable/disable cede_offline when available.
56 static int __init setup_cede_offline(char *str)
58 return (kstrtobool(str, &cede_offline_enabled) == 0);
61 __setup("cede_offline=", setup_cede_offline);
63 enum cpu_state_vals get_cpu_current_state(int cpu)
65 return per_cpu(current_state, cpu);
68 void set_cpu_current_state(int cpu, enum cpu_state_vals state)
70 per_cpu(current_state, cpu) = state;
73 enum cpu_state_vals get_preferred_offline_state(int cpu)
75 return per_cpu(preferred_offline_state, cpu);
78 void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
80 per_cpu(preferred_offline_state, cpu) = state;
83 void set_default_offline_state(int cpu)
85 per_cpu(preferred_offline_state, cpu) = default_offline_state;
88 static void rtas_stop_self(void)
90 static struct rtas_args args;
94 BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE);
96 printk("cpu %u (hwid %u) Ready to die...\n",
97 smp_processor_id(), hard_smp_processor_id());
99 rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL);
101 panic("Alas, I survived.\n");
104 static void pseries_mach_cpu_die(void)
106 unsigned int cpu = smp_processor_id();
107 unsigned int hwcpu = hard_smp_processor_id();
108 u8 cede_latency_hint = 0;
114 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
115 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
116 if (ppc_md.suspend_disable_cpu)
117 ppc_md.suspend_disable_cpu();
119 cede_latency_hint = 2;
121 get_lppaca()->idle = 1;
122 if (!lppaca_shared_proc(get_lppaca()))
123 get_lppaca()->donate_dedicated_cpu = 1;
125 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
126 while (!prep_irq_for_idle()) {
131 extended_cede_processor(cede_latency_hint);
136 if (!lppaca_shared_proc(get_lppaca()))
137 get_lppaca()->donate_dedicated_cpu = 0;
138 get_lppaca()->idle = 0;
140 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
141 unregister_slb_shadow(hwcpu);
145 * Call to start_secondary_resume() will not return.
146 * Kernel stack will be reset and start_secondary()
147 * will be called to continue the online operation.
149 start_secondary_resume();
153 /* Requested state is CPU_STATE_OFFLINE at this point */
154 WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
156 set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
157 unregister_slb_shadow(hwcpu);
160 /* Should never get here... */
165 static int pseries_cpu_disable(void)
167 int cpu = smp_processor_id();
169 set_cpu_online(cpu, false);
170 vdso_data->processorCount--;
172 /*fix boot_cpuid here*/
173 if (cpu == boot_cpuid)
174 boot_cpuid = cpumask_any(cpu_online_mask);
176 /* FIXME: abstract this to not be platform specific later on */
177 xics_migrate_irqs_away();
182 * pseries_cpu_die: Wait for the cpu to die.
183 * @cpu: logical processor id of the CPU whose death we're awaiting.
185 * This function is called from the context of the thread which is performing
186 * the cpu-offline. Here we wait for long enough to allow the cpu in question
187 * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
190 * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
193 static void pseries_cpu_die(unsigned int cpu)
197 unsigned int pcpu = get_hard_smp_processor_id(cpu);
199 if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
201 for (tries = 0; tries < 5000; tries++) {
202 if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
208 } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
210 for (tries = 0; tries < 25; tries++) {
211 cpu_status = smp_query_cpu_stopped(pcpu);
212 if (cpu_status == QCSS_STOPPED ||
213 cpu_status == QCSS_HARDWARE_ERROR)
219 if (cpu_status != 0) {
220 printk("Querying DEAD? cpu %i (%i) shows %i\n",
221 cpu, pcpu, cpu_status);
224 /* Isolation and deallocation are definitely done by
225 * drslot_chrp_cpu. If they were not they would be
226 * done here. Change isolate state to Isolate and
227 * change allocation-state to Unusable.
229 paca[cpu].cpu_start = 0;
233 * Update cpu_present_mask and paca(s) for a new cpu node. The wrinkle
234 * here is that a cpu device node may represent up to two logical cpus
235 * in the SMT case. We must honor the assumption in other code that
236 * the logical ids for sibling SMT threads x and y are adjacent, such
237 * that x^1 == y and y^1 == x.
239 static int pseries_add_processor(struct device_node *np)
242 cpumask_var_t candidate_mask, tmp;
243 int err = -ENOSPC, len, nthreads, i;
244 const __be32 *intserv;
246 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
250 zalloc_cpumask_var(&candidate_mask, GFP_KERNEL);
251 zalloc_cpumask_var(&tmp, GFP_KERNEL);
253 nthreads = len / sizeof(u32);
254 for (i = 0; i < nthreads; i++)
255 cpumask_set_cpu(i, tmp);
257 cpu_maps_update_begin();
259 BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask));
261 /* Get a bitmap of unoccupied slots. */
262 cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask);
263 if (cpumask_empty(candidate_mask)) {
264 /* If we get here, it most likely means that NR_CPUS is
265 * less than the partition's max processors setting.
267 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
268 " supports %d logical cpus.\n", np->full_name,
269 num_possible_cpus());
273 while (!cpumask_empty(tmp))
274 if (cpumask_subset(tmp, candidate_mask))
275 /* Found a range where we can insert the new cpu(s) */
278 cpumask_shift_left(tmp, tmp, nthreads);
280 if (cpumask_empty(tmp)) {
281 printk(KERN_ERR "Unable to find space in cpu_present_mask for"
282 " processor %s with %d thread(s)\n", np->name,
287 for_each_cpu(cpu, tmp) {
288 BUG_ON(cpu_present(cpu));
289 set_cpu_present(cpu, true);
290 set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++));
294 cpu_maps_update_done();
295 free_cpumask_var(candidate_mask);
296 free_cpumask_var(tmp);
301 * Update the present map for a cpu node which is going away, and set
302 * the hard id in the paca(s) to -1 to be consistent with boot time
303 * convention for non-present cpus.
305 static void pseries_remove_processor(struct device_node *np)
308 int len, nthreads, i;
309 const __be32 *intserv;
312 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
316 nthreads = len / sizeof(u32);
318 cpu_maps_update_begin();
319 for (i = 0; i < nthreads; i++) {
320 thread = be32_to_cpu(intserv[i]);
321 for_each_present_cpu(cpu) {
322 if (get_hard_smp_processor_id(cpu) != thread)
324 BUG_ON(cpu_online(cpu));
325 set_cpu_present(cpu, false);
326 set_hard_smp_processor_id(cpu, -1);
329 if (cpu >= nr_cpu_ids)
330 printk(KERN_WARNING "Could not find cpu to remove "
331 "with physical id 0x%x\n", thread);
333 cpu_maps_update_done();
336 static int dlpar_online_cpu(struct device_node *dn)
340 int len, nthreads, i;
341 const __be32 *intserv;
344 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
348 nthreads = len / sizeof(u32);
350 cpu_maps_update_begin();
351 for (i = 0; i < nthreads; i++) {
352 thread = be32_to_cpu(intserv[i]);
353 for_each_present_cpu(cpu) {
354 if (get_hard_smp_processor_id(cpu) != thread)
356 BUG_ON(get_cpu_current_state(cpu)
357 != CPU_STATE_OFFLINE);
358 cpu_maps_update_done();
359 rc = device_online(get_cpu_device(cpu));
362 cpu_maps_update_begin();
366 if (cpu == num_possible_cpus())
367 printk(KERN_WARNING "Could not find cpu to online "
368 "with physical id 0x%x\n", thread);
370 cpu_maps_update_done();
377 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
379 struct device_node *child = NULL;
384 /* Assume cpu doesn't exist */
387 for_each_child_of_node(parent, child) {
388 rc = of_property_read_u32(child, "ibm,my-drc-index",
393 if (my_drc_index == drc_index) {
403 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
412 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
417 if (drc == drc_index)
424 static ssize_t dlpar_cpu_add(u32 drc_index)
426 struct device_node *dn, *parent;
429 pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
431 parent = of_find_node_by_path("/cpus");
433 pr_warn("Failed to find CPU root node \"/cpus\"\n");
437 if (dlpar_cpu_exists(parent, drc_index)) {
439 pr_warn("CPU with drc index %x already exists\n", drc_index);
443 if (!valid_cpu_drc_index(parent, drc_index)) {
445 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
449 rc = dlpar_acquire_drc(drc_index);
451 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
457 dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
460 pr_warn("Failed call to configure-connector, drc index: %x\n",
462 dlpar_release_drc(drc_index);
466 rc = dlpar_attach_node(dn);
469 pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n",
470 dn->name, rc, drc_index);
472 rc = dlpar_release_drc(drc_index);
474 dlpar_free_cc_nodes(dn);
479 rc = dlpar_online_cpu(dn);
482 pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n",
483 dn->name, rc, drc_index);
485 rc = dlpar_detach_node(dn);
487 dlpar_release_drc(drc_index);
492 pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name,
497 static int dlpar_offline_cpu(struct device_node *dn)
501 int len, nthreads, i;
502 const __be32 *intserv;
505 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
509 nthreads = len / sizeof(u32);
511 cpu_maps_update_begin();
512 for (i = 0; i < nthreads; i++) {
513 thread = be32_to_cpu(intserv[i]);
514 for_each_present_cpu(cpu) {
515 if (get_hard_smp_processor_id(cpu) != thread)
518 if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
521 if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
522 set_preferred_offline_state(cpu,
524 cpu_maps_update_done();
525 rc = device_offline(get_cpu_device(cpu));
528 cpu_maps_update_begin();
534 * The cpu is in CPU_STATE_INACTIVE.
535 * Upgrade it's state to CPU_STATE_OFFLINE.
537 set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
538 BUG_ON(plpar_hcall_norets(H_PROD, thread)
543 if (cpu == num_possible_cpus())
544 printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread);
546 cpu_maps_update_done();
553 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
557 pr_debug("Attempting to remove CPU %s, drc index: %x\n",
558 dn->name, drc_index);
560 rc = dlpar_offline_cpu(dn);
562 pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc);
566 rc = dlpar_release_drc(drc_index);
568 pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n",
569 drc_index, dn->name, rc);
570 dlpar_online_cpu(dn);
574 rc = dlpar_detach_node(dn);
578 pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc);
580 rc = dlpar_acquire_drc(drc_index);
582 dlpar_online_cpu(dn);
587 pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
591 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
593 struct device_node *dn;
597 for_each_node_by_type(dn, "cpu") {
598 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
602 if (my_index == drc_index)
609 static int dlpar_cpu_remove_by_index(u32 drc_index)
611 struct device_node *dn;
614 dn = cpu_drc_index_to_dn(drc_index);
616 pr_warn("Cannot find CPU (drc index %x) to remove\n",
621 rc = dlpar_cpu_remove(dn, drc_index);
626 static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove)
628 struct device_node *dn;
632 /* We want to find cpus_to_remove + 1 CPUs to ensure we do not
633 * remove the last CPU.
635 for_each_node_by_type(dn, "cpu") {
638 if (cpus_found > cpus_to_remove) {
643 /* Note that cpus_found is always 1 ahead of the index
644 * into the cpu_drcs array, so we use cpus_found - 1
646 rc = of_property_read_u32(dn, "ibm,my-drc-index",
647 &cpu_drcs[cpus_found - 1]);
649 pr_warn("Error occurred getting drc-index for %s\n",
656 if (cpus_found < cpus_to_remove) {
657 pr_warn("Failed to find enough CPUs (%d of %d) to remove\n",
658 cpus_found, cpus_to_remove);
659 } else if (cpus_found == cpus_to_remove) {
660 pr_warn("Cannot remove all CPUs\n");
666 static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
670 int cpus_removed = 0;
673 pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove);
675 cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL);
679 cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove);
680 if (cpus_found <= cpus_to_remove) {
685 for (i = 0; i < cpus_to_remove; i++) {
686 rc = dlpar_cpu_remove_by_index(cpu_drcs[i]);
693 if (cpus_removed != cpus_to_remove) {
694 pr_warn("CPU hot-remove failed, adding back removed CPUs\n");
696 for (i = 0; i < cpus_removed; i++)
697 dlpar_cpu_add(cpu_drcs[i]);
708 static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
710 struct device_node *parent;
714 parent = of_find_node_by_path("/cpus");
716 pr_warn("Could not find CPU root node in device tree\n");
721 /* Search the ibm,drc-indexes array for possible CPU drcs to
722 * add. Note that the format of the ibm,drc-indexes array is
723 * the number of entries in the array followed by the array
724 * of drc values so we start looking at index = 1.
727 while (cpus_found < cpus_to_add) {
730 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
735 if (dlpar_cpu_exists(parent, drc))
738 cpu_drcs[cpus_found++] = drc;
745 static int dlpar_cpu_add_by_count(u32 cpus_to_add)
752 pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add);
754 cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL);
758 cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add);
759 if (cpus_found < cpus_to_add) {
760 pr_warn("Failed to find enough CPUs (%d of %d) to add\n",
761 cpus_found, cpus_to_add);
766 for (i = 0; i < cpus_to_add; i++) {
767 rc = dlpar_cpu_add(cpu_drcs[i]);
774 if (cpus_added < cpus_to_add) {
775 pr_warn("CPU hot-add failed, removing any added CPUs\n");
777 for (i = 0; i < cpus_added; i++)
778 dlpar_cpu_remove_by_index(cpu_drcs[i]);
789 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
791 u32 count, drc_index;
794 count = hp_elog->_drc_u.drc_count;
795 drc_index = hp_elog->_drc_u.drc_index;
797 lock_device_hotplug();
799 switch (hp_elog->action) {
800 case PSERIES_HP_ELOG_ACTION_REMOVE:
801 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
802 rc = dlpar_cpu_remove_by_count(count);
803 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
804 rc = dlpar_cpu_remove_by_index(drc_index);
808 case PSERIES_HP_ELOG_ACTION_ADD:
809 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
810 rc = dlpar_cpu_add_by_count(count);
811 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
812 rc = dlpar_cpu_add(drc_index);
817 pr_err("Invalid action (%d) specified\n", hp_elog->action);
822 unlock_device_hotplug();
826 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
828 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
833 rc = kstrtou32(buf, 0, &drc_index);
837 rc = dlpar_cpu_add(drc_index);
839 return rc ? rc : count;
842 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
844 struct device_node *dn;
848 dn = of_find_node_by_path(buf);
852 rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
858 rc = dlpar_cpu_remove(dn, drc_index);
861 return rc ? rc : count;
864 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
866 static int pseries_smp_notifier(struct notifier_block *nb,
867 unsigned long action, void *data)
869 struct of_reconfig_data *rd = data;
873 case OF_RECONFIG_ATTACH_NODE:
874 err = pseries_add_processor(rd->dn);
876 case OF_RECONFIG_DETACH_NODE:
877 pseries_remove_processor(rd->dn);
880 return notifier_from_errno(err);
883 static struct notifier_block pseries_smp_nb = {
884 .notifier_call = pseries_smp_notifier,
887 #define MAX_CEDE_LATENCY_LEVELS 4
888 #define CEDE_LATENCY_PARAM_LENGTH 10
889 #define CEDE_LATENCY_PARAM_MAX_LENGTH \
890 (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
891 #define CEDE_LATENCY_TOKEN 45
893 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
895 static int parse_cede_parameters(void)
897 memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
898 return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
901 __pa(cede_parameters),
902 CEDE_LATENCY_PARAM_MAX_LENGTH);
905 static int __init pseries_cpu_hotplug_init(void)
910 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
911 ppc_md.cpu_probe = dlpar_cpu_probe;
912 ppc_md.cpu_release = dlpar_cpu_release;
913 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
915 rtas_stop_self_token = rtas_token("stop-self");
916 qcss_tok = rtas_token("query-cpu-stopped-state");
918 if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
919 qcss_tok == RTAS_UNKNOWN_SERVICE) {
920 printk(KERN_INFO "CPU Hotplug not supported by firmware "
925 ppc_md.cpu_die = pseries_mach_cpu_die;
926 smp_ops->cpu_disable = pseries_cpu_disable;
927 smp_ops->cpu_die = pseries_cpu_die;
929 /* Processors can be added/removed only on LPAR */
930 if (firmware_has_feature(FW_FEATURE_LPAR)) {
931 of_reconfig_notifier_register(&pseries_smp_nb);
932 cpu_maps_update_begin();
933 if (cede_offline_enabled && parse_cede_parameters() == 0) {
934 default_offline_state = CPU_STATE_INACTIVE;
935 for_each_online_cpu(cpu)
936 set_default_offline_state(cpu);
938 cpu_maps_update_done();
943 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);