aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2016-07-29 16:55:30 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2016-07-29 16:55:30 -0400
commita6408f6cb63ac0958fee7dbce7861ffb540d8a49 (patch)
treec94a835d343974171951e3b805e6bbbb02852ebc
parent1a81a8f2a5918956e214bb718099a89e500e7ec5 (diff)
parent4fae16dffb812f0e0d98a0b2b0856ca48ca63e6c (diff)
Merge branch 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull smp hotplug updates from Thomas Gleixner: "This is the next part of the hotplug rework. - Convert all notifiers with a priority assigned - Convert all CPU_STARTING/DYING notifiers The final removal of the STARTING/DYING infrastructure will happen when the merge window closes. Another 700 hundred line of unpenetrable maze gone :)" * 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (70 commits) timers/core: Correct callback order during CPU hot plug leds/trigger/cpu: Move from CPU_STARTING to ONLINE level powerpc/numa: Convert to hotplug state machine arm/perf: Fix hotplug state machine conversion irqchip/armada: Avoid unused function warnings ARC/time: Convert to hotplug state machine clocksource/atlas7: Convert to hotplug state machine clocksource/armada-370-xp: Convert to hotplug state machine clocksource/exynos_mct: Convert to hotplug state machine clocksource/arm_global_timer: Convert to hotplug state machine rcu: Convert rcutree to hotplug state machine KVM/arm/arm64/vgic-new: Convert to hotplug state machine smp/cfd: Convert core to hotplug state machine x86/x2apic: Convert to CPU hotplug state machine profile: Convert to hotplug state machine timers/core: Convert to hotplug state machine hrtimer: Convert to hotplug state machine x86/tboot: Convert to hotplug state machine arm64/armv8 deprecated: Convert to hotplug state machine hwtracing/coresight-etm4x: Convert to hotplug state machine ...
Diffstat
-rw-r--r--arch/arc/kernel/time.c48
-rw-r--r--arch/arm/kernel/smp_twd.c31
-rw-r--r--arch/arm/mach-mvebu/coherency.c19
-rw-r--r--arch/arm/mm/cache-l2x0.c27
-rw-r--r--arch/arm/vfp/vfpmodule.c28
-rw-r--r--arch/arm/xen/enlighten.c41
-rw-r--r--arch/arm64/kernel/armv8_deprecated.c22
-rw-r--r--arch/blackfin/kernel/perf_event.c26
-rw-r--r--arch/metag/kernel/perf/perf_event.c26
-rw-r--r--arch/mips/oprofile/op_model_loongson3.c35
-rw-r--r--arch/powerpc/mm/numa.c48
-rw-r--r--arch/powerpc/perf/core-book3s.c30
-rw-r--r--arch/s390/kernel/perf_cpum_cf.c46
-rw-r--r--arch/s390/kernel/perf_cpum_sf.c45
-rw-r--r--arch/sh/kernel/perf_event.c23
-rw-r--r--arch/x86/entry/vdso/vma.c20
-rw-r--r--arch/x86/events/amd/core.c6
-rw-r--r--arch/x86/events/amd/ibs.c64
-rw-r--r--arch/x86/events/amd/power.c60
-rw-r--r--arch/x86/events/amd/uncore.c122
-rw-r--r--arch/x86/events/core.c103
-rw-r--r--arch/x86/events/intel/core.c4
-rw-r--r--arch/x86/events/intel/cqm.c49
-rw-r--r--arch/x86/events/intel/cstate.c51
-rw-r--r--arch/x86/events/intel/rapl.c84
-rw-r--r--arch/x86/events/intel/uncore.c133
-rw-r--r--arch/x86/kernel/apb_timer.c29
-rw-r--r--arch/x86/kernel/apic/x2apic_cluster.c80
-rw-r--r--arch/x86/kernel/hpet.c69
-rw-r--r--arch/x86/kernel/tboot.c25
-rw-r--r--arch/x86/kvm/x86.c35
-rw-r--r--arch/xtensa/kernel/perf_event.c26
-rw-r--r--drivers/acpi/processor_driver.c11
-rw-r--r--drivers/bus/arm-cci.c53
-rw-r--r--drivers/bus/arm-ccn.c57
-rw-r--r--drivers/clocksource/arm_arch_timer.c54
-rw-r--r--drivers/clocksource/arm_global_timer.c39
-rw-r--r--drivers/clocksource/dummy_timer.c36
-rw-r--r--drivers/clocksource/exynos_mct.c46
-rw-r--r--drivers/clocksource/metag_generic.c33
-rw-r--r--drivers/clocksource/mips-gic-timer.c38
-rw-r--r--drivers/clocksource/qcom-timer.c41
-rw-r--r--drivers/clocksource/time-armada-370-xp.c41
-rw-r--r--drivers/clocksource/timer-atlas7.c41
-rw-r--r--drivers/hwtracing/coresight/coresight-etm3x.c90
-rw-r--r--drivers/hwtracing/coresight/coresight-etm4x.c87
-rw-r--r--drivers/irqchip/irq-armada-370-xp.c44
-rw-r--r--drivers/irqchip/irq-bcm2836.c34
-rw-r--r--drivers/irqchip/irq-gic-v3.c22
-rw-r--r--drivers/irqchip/irq-gic.c23
-rw-r--r--drivers/irqchip/irq-hip04.c25
-rw-r--r--drivers/leds/trigger/ledtrig-cpu.c32
-rw-r--r--drivers/perf/arm_pmu.c59
-rw-r--r--include/linux/cpu.h11
-rw-r--r--include/linux/cpuhotplug.h76
-rw-r--r--include/linux/hrtimer.h7
-rw-r--r--include/linux/perf/arm_pmu.h2
-rw-r--r--include/linux/perf_event.h44
-rw-r--r--include/linux/rcutiny.h7
-rw-r--r--include/linux/rcutree.h7
-rw-r--r--include/linux/smp.h5
-rw-r--r--include/linux/timer.h6
-rw-r--r--include/linux/workqueue.h6
-rw-r--r--kernel/cpu.c66
-rw-r--r--kernel/events/core.c56
-rw-r--r--kernel/profile.c181
-rw-r--r--kernel/rcu/tree.c105
-rw-r--r--kernel/smp.c79
-rw-r--r--kernel/time/hrtimer.c40
-rw-r--r--kernel/time/timer.c25
-rw-r--r--kernel/workqueue.c108
-rw-r--r--virt/kvm/arm/arch_timer.c35
-rw-r--r--virt/kvm/arm/vgic.c39
-rw-r--r--virt/kvm/arm/vgic/vgic-init.c31
-rw-r--r--virt/kvm/kvm_main.c32
75 files changed, 1316 insertions, 2013 deletions
diff --git a/arch/arc/kernel/time.c b/arch/arc/kernel/time.c
index 98f22d2eb563..f927b8dc6edd 100644
--- a/arch/arc/kernel/time.c
+++ b/arch/arc/kernel/time.c
@@ -296,30 +296,23 @@ static irqreturn_t timer_irq_handler(int irq, void *dev_id)
296 return IRQ_HANDLED; 296 return IRQ_HANDLED;
297} 297}
298 298
299static int arc_timer_cpu_notify(struct notifier_block *self, 299
300 unsigned long action, void *hcpu) 300static int arc_timer_starting_cpu(unsigned int cpu)
301{ 301{
302 struct clock_event_device *evt = this_cpu_ptr(&arc_clockevent_device); 302 struct clock_event_device *evt = this_cpu_ptr(&arc_clockevent_device);
303 303
304 evt->cpumask = cpumask_of(smp_processor_id()); 304 evt->cpumask = cpumask_of(smp_processor_id());
305 305
306 switch (action & ~CPU_TASKS_FROZEN) { 306 clockevents_config_and_register(evt, arc_timer_freq, 0, ARC_TIMER_MAX);
307 case CPU_STARTING: 307 enable_percpu_irq(arc_timer_irq, 0);
308 clockevents_config_and_register(evt, arc_timer_freq, 308 return 0;
309 0, ULONG_MAX);
310 enable_percpu_irq(arc_timer_irq, 0);
311 break;
312 case CPU_DYING:
313 disable_percpu_irq(arc_timer_irq);
314 break;
315 }
316
317 return NOTIFY_OK;
318} 309}
319 310
320static struct notifier_block arc_timer_cpu_nb = { 311static int arc_timer_dying_cpu(unsigned int cpu)
321 .notifier_call = arc_timer_cpu_notify, 312{
322}; 313 disable_percpu_irq(arc_timer_irq);
314 return 0;
315}
323 316
324/* 317/*
325 * clockevent setup for boot CPU 318 * clockevent setup for boot CPU
@@ -329,12 +322,6 @@ static int __init arc_clockevent_setup(struct device_node *node)
329 struct clock_event_device *evt = this_cpu_ptr(&arc_clockevent_device); 322 struct clock_event_device *evt = this_cpu_ptr(&arc_clockevent_device);
330 int ret; 323 int ret;
331 324
332 ret = register_cpu_notifier(&arc_timer_cpu_nb);
333 if (ret) {
334 pr_err("Failed to register cpu notifier");
335 return ret;
336 }
337
338 arc_timer_irq = irq_of_parse_and_map(node, 0); 325 arc_timer_irq = irq_of_parse_and_map(node, 0);
339 if (arc_timer_irq <= 0) { 326 if (arc_timer_irq <= 0) {
340 pr_err("clockevent: missing irq"); 327 pr_err("clockevent: missing irq");
@@ -347,11 +334,6 @@ static int __init arc_clockevent_setup(struct device_node *node)
347 return ret; 334 return ret;
348 } 335 }
349 336
350 evt->irq = arc_timer_irq;
351 evt->cpumask = cpumask_of(smp_processor_id());
352 clockevents_config_and_register(evt, arc_timer_freq,
353 0, ARC_TIMER_MAX);
354
355 /* Needs apriori irq_set_percpu_devid() done in intc map function */ 337 /* Needs apriori irq_set_percpu_devid() done in intc map function */
356 ret = request_percpu_irq(arc_timer_irq, timer_irq_handler, 338 ret = request_percpu_irq(arc_timer_irq, timer_irq_handler,
357 "Timer0 (per-cpu-tick)", evt); 339 "Timer0 (per-cpu-tick)", evt);
@@ -360,8 +342,14 @@ static int __init arc_clockevent_setup(struct device_node *node)
360 return ret; 342 return ret;
361 } 343 }
362 344
363 enable_percpu_irq(arc_timer_irq, 0); 345 ret = cpuhp_setup_state(CPUHP_AP_ARC_TIMER_STARTING,
364 346 "AP_ARC_TIMER_STARTING",
347 arc_timer_starting_cpu,
348 arc_timer_dying_cpu);
349 if (ret) {
350 pr_err("Failed to setup hotplug state");
351 return ret;
352 }
365 return 0; 353 return 0;
366} 354}
367 355
diff --git a/arch/arm/kernel/smp_twd.c b/arch/arm/kernel/smp_twd.c
index b6ec65e68009..02d5e5e8d44c 100644
--- a/arch/arm/kernel/smp_twd.c
+++ b/arch/arm/kernel/smp_twd.c
@@ -310,24 +310,17 @@ static void twd_timer_setup(void)
310 enable_percpu_irq(clk->irq, 0); 310 enable_percpu_irq(clk->irq, 0);
311} 311}
312 312
313static int twd_timer_cpu_notify(struct notifier_block *self, 313static int twd_timer_starting_cpu(unsigned int cpu)
314 unsigned long action, void *hcpu)
315{ 314{
316 switch (action & ~CPU_TASKS_FROZEN) { 315 twd_timer_setup();
317 case CPU_STARTING: 316 return 0;
318 twd_timer_setup();
319 break;
320 case CPU_DYING:
321 twd_timer_stop();
322 break;
323 }
324
325 return NOTIFY_OK;
326} 317}
327 318
328static struct notifier_block twd_timer_cpu_nb = { 319static int twd_timer_dying_cpu(unsigned int cpu)
329 .notifier_call = twd_timer_cpu_notify, 320{
330}; 321 twd_timer_stop();
322 return 0;
323}
331 324
332static int __init twd_local_timer_common_register(struct device_node *np) 325static int __init twd_local_timer_common_register(struct device_node *np)
333{ 326{
@@ -345,9 +338,9 @@ static int __init twd_local_timer_common_register(struct device_node *np)
345 goto out_free; 338 goto out_free;
346 } 339 }
347 340
348 err = register_cpu_notifier(&twd_timer_cpu_nb); 341 cpuhp_setup_state_nocalls(CPUHP_AP_ARM_TWD_STARTING,
349 if (err) 342 "AP_ARM_TWD_STARTING",
350 goto out_irq; 343 twd_timer_starting_cpu, twd_timer_dying_cpu);
351 344
352 twd_get_clock(np); 345 twd_get_clock(np);
353 if (!of_property_read_bool(np, "always-on")) 346 if (!of_property_read_bool(np, "always-on"))
@@ -365,8 +358,6 @@ static int __init twd_local_timer_common_register(struct device_node *np)
365 358
366 return 0; 359 return 0;
367 360
368out_irq:
369 free_percpu_irq(twd_ppi, twd_evt);
370out_free: 361out_free:
371 iounmap(twd_base); 362 iounmap(twd_base);
372 twd_base = NULL; 363 twd_base = NULL;
diff --git a/arch/arm/mach-mvebu/coherency.c b/arch/arm/mach-mvebu/coherency.c
index e80f0dde2189..ae2a018b9305 100644
--- a/arch/arm/mach-mvebu/coherency.c
+++ b/arch/arm/mach-mvebu/coherency.c
@@ -111,20 +111,12 @@ static struct notifier_block mvebu_hwcc_pci_nb __maybe_unused = {
111 .notifier_call = mvebu_hwcc_notifier, 111 .notifier_call = mvebu_hwcc_notifier,
112}; 112};
113 113
114static int armada_xp_clear_shared_l2_notifier_func(struct notifier_block *nfb, 114static int armada_xp_clear_l2_starting(unsigned int cpu)
115 unsigned long action, void *hcpu)
116{ 115{
117 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) 116 armada_xp_clear_shared_l2();
118 armada_xp_clear_shared_l2(); 117 return 0;
119
120 return NOTIFY_OK;
121} 118}
122 119
123static struct notifier_block armada_xp_clear_shared_l2_notifier = {
124 .notifier_call = armada_xp_clear_shared_l2_notifier_func,
125 .priority = 100,
126};
127
128static void __init armada_370_coherency_init(struct device_node *np) 120static void __init armada_370_coherency_init(struct device_node *np)
129{ 121{
130 struct resource res; 122 struct resource res;
@@ -155,8 +147,9 @@ static void __init armada_370_coherency_init(struct device_node *np)
155 147
156 of_node_put(cpu_config_np); 148 of_node_put(cpu_config_np);
157 149
158 register_cpu_notifier(&armada_xp_clear_shared_l2_notifier); 150 cpuhp_setup_state_nocalls(CPUHP_AP_ARM_MVEBU_COHERENCY,
159 151 "AP_ARM_MVEBU_COHERENCY",
152 armada_xp_clear_l2_starting, NULL);
160exit: 153exit:
161 set_cpu_coherent(); 154 set_cpu_coherent();
162} 155}
diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c
index c61996c256cc..cc12905ae6f8 100644
--- a/arch/arm/mm/cache-l2x0.c
+++ b/arch/arm/mm/cache-l2x0.c
@@ -597,17 +597,16 @@ static void l2c310_configure(void __iomem *base)
597 L310_POWER_CTRL); 597 L310_POWER_CTRL);
598} 598}
599 599
600static int l2c310_cpu_enable_flz(struct notifier_block *nb, unsigned long act, void *data) 600static int l2c310_starting_cpu(unsigned int cpu)
601{ 601{
602 switch (act & ~CPU_TASKS_FROZEN) { 602 set_auxcr(get_auxcr() | BIT(3) | BIT(2) | BIT(1));
603 case CPU_STARTING: 603 return 0;
604 set_auxcr(get_auxcr() | BIT(3) | BIT(2) | BIT(1)); 604}
605 break; 605
606 case CPU_DYING: 606static int l2c310_dying_cpu(unsigned int cpu)
607 set_auxcr(get_auxcr() & ~(BIT(3) | BIT(2) | BIT(1))); 607{
608 break; 608 set_auxcr(get_auxcr() & ~(BIT(3) | BIT(2) | BIT(1)));
609 } 609 return 0;
610 return NOTIFY_OK;
611} 610}
612 611
613static void __init l2c310_enable(void __iomem *base, unsigned num_lock) 612static void __init l2c310_enable(void __iomem *base, unsigned num_lock)
@@ -678,10 +677,10 @@ static void __init l2c310_enable(void __iomem *base, unsigned num_lock)
678 power_ctrl & L310_STNDBY_MODE_EN ? "en" : "dis"); 677 power_ctrl & L310_STNDBY_MODE_EN ? "en" : "dis");
679 } 678 }
680 679
681 if (aux & L310_AUX_CTRL_FULL_LINE_ZERO) { 680 if (aux & L310_AUX_CTRL_FULL_LINE_ZERO)
682 set_auxcr(get_auxcr() | BIT(3) | BIT(2) | BIT(1)); 681 cpuhp_setup_state(CPUHP_AP_ARM_L2X0_STARTING,
683 cpu_notifier(l2c310_cpu_enable_flz, 0); 682 "AP_ARM_L2X0_STARTING", l2c310_starting_cpu,
684 } 683 l2c310_dying_cpu);
685} 684}
686 685
687static void __init l2c310_fixup(void __iomem *base, u32 cache_id, 686static void __init l2c310_fixup(void __iomem *base, u32 cache_id,
diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
index 73085d3482ed..da0b33deba6d 100644
--- a/arch/arm/vfp/vfpmodule.c
+++ b/arch/arm/vfp/vfpmodule.c
@@ -643,19 +643,19 @@ int vfp_restore_user_hwstate(struct user_vfp __user *ufp,
643 * hardware state at every thread switch. We clear our held state when 643 * hardware state at every thread switch. We clear our held state when
644 * a CPU has been killed, indicating that the VFP hardware doesn't contain 644 * a CPU has been killed, indicating that the VFP hardware doesn't contain
645 * a threads VFP state. When a CPU starts up, we re-enable access to the 645 * a threads VFP state. When a CPU starts up, we re-enable access to the
646 * VFP hardware. 646 * VFP hardware. The callbacks below are called on the CPU which
647 *
648 * Both CPU_DYING and CPU_STARTING are called on the CPU which
649 * is being offlined/onlined. 647 * is being offlined/onlined.
650 */ 648 */
651static int vfp_hotplug(struct notifier_block *b, unsigned long action, 649static int vfp_dying_cpu(unsigned int cpu)
652 void *hcpu)
653{ 650{
654 if (action == CPU_DYING || action == CPU_DYING_FROZEN) 651 vfp_force_reload(cpu, current_thread_info());
655 vfp_current_hw_state[(long)hcpu] = NULL; 652 return 0;
656 else if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) 653}
657 vfp_enable(NULL); 654
658 return NOTIFY_OK; 655static int vfp_starting_cpu(unsigned int unused)
656{
657 vfp_enable(NULL);
658 return 0;
659} 659}
660 660
661void vfp_kmode_exception(void) 661void vfp_kmode_exception(void)
@@ -732,6 +732,10 @@ static int __init vfp_init(void)
732 unsigned int vfpsid; 732 unsigned int vfpsid;
733 unsigned int cpu_arch = cpu_architecture(); 733 unsigned int cpu_arch = cpu_architecture();
734 734
735 /*
736 * Enable the access to the VFP on all online CPUs so the
737 * following test on FPSID will succeed.
738 */
735 if (cpu_arch >= CPU_ARCH_ARMv6) 739 if (cpu_arch >= CPU_ARCH_ARMv6)
736 on_each_cpu(vfp_enable, NULL, 1); 740 on_each_cpu(vfp_enable, NULL, 1);
737 741
@@ -794,7 +798,9 @@ static int __init vfp_init(void)
794 VFP_arch = (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT; 798 VFP_arch = (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT;
795 } 799 }
796 800
797 hotcpu_notifier(vfp_hotplug, 0); 801 cpuhp_setup_state_nocalls(CPUHP_AP_ARM_VFP_STARTING,
802 "AP_ARM_VFP_STARTING", vfp_starting_cpu,
803 vfp_dying_cpu);
798 804
799 vfp_vector = vfp_support_entry; 805 vfp_vector = vfp_support_entry;
800 806
diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
index 0bea3d271f6e..b0b82f5ea338 100644
--- a/arch/arm/xen/enlighten.c
+++ b/arch/arm/xen/enlighten.c
@@ -153,12 +153,11 @@ static struct notifier_block xen_pvclock_gtod_notifier = {
153 .notifier_call = xen_pvclock_gtod_notify, 153 .notifier_call = xen_pvclock_gtod_notify,
154}; 154};
155 155
156static void xen_percpu_init(void) 156static int xen_starting_cpu(unsigned int cpu)
157{ 157{
158 struct vcpu_register_vcpu_info info; 158 struct vcpu_register_vcpu_info info;
159 struct vcpu_info *vcpup; 159 struct vcpu_info *vcpup;
160 int err; 160 int err;
161 int cpu = get_cpu();
162 161
163 /* 162 /*
164 * VCPUOP_register_vcpu_info cannot be called twice for the same 163 * VCPUOP_register_vcpu_info cannot be called twice for the same
@@ -186,7 +185,13 @@ static void xen_percpu_init(void)
186 185
187after_register_vcpu_info: 186after_register_vcpu_info:
188 enable_percpu_irq(xen_events_irq, 0); 187 enable_percpu_irq(xen_events_irq, 0);
189 put_cpu(); 188 return 0;
189}
190
191static int xen_dying_cpu(unsigned int cpu)
192{
193 disable_percpu_irq(xen_events_irq);
194 return 0;
190} 195}
191 196
192static void xen_restart(enum reboot_mode reboot_mode, const char *cmd) 197static void xen_restart(enum reboot_mode reboot_mode, const char *cmd)
@@ -205,28 +210,6 @@ static void xen_power_off(void)
205 BUG_ON(rc); 210 BUG_ON(rc);
206} 211}
207 212
208static int xen_cpu_notification(struct notifier_block *self,
209 unsigned long action,
210 void *hcpu)
211{
212 switch (action) {
213 case CPU_STARTING:
214 xen_percpu_init();
215 break;
216 case CPU_DYING:
217 disable_percpu_irq(xen_events_irq);
218 break;
219 default:
220 break;
221 }
222
223 return NOTIFY_OK;
224}
225
226static struct notifier_block xen_cpu_notifier = {
227 .notifier_call = xen_cpu_notification,
228};
229
230static irqreturn_t xen_arm_callback(int irq, void *arg) 213static irqreturn_t xen_arm_callback(int irq, void *arg)
231{ 214{
232 xen_hvm_evtchn_do_upcall(); 215 xen_hvm_evtchn_do_upcall();
@@ -425,16 +408,14 @@ static int __init xen_guest_init(void)
425 return -EINVAL; 408 return -EINVAL;
426 } 409 }
427 410
428 xen_percpu_init();
429
430 register_cpu_notifier(&xen_cpu_notifier);
431
432 xen_time_setup_guest(); 411 xen_time_setup_guest();
433 412
434 if (xen_initial_domain()) 413 if (xen_initial_domain())
435 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier); 414 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
436 415
437 return 0; 416 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
417 "AP_ARM_XEN_STARTING", xen_starting_cpu,
418 xen_dying_cpu);
438} 419}
439early_initcall(xen_guest_init); 420early_initcall(xen_guest_init);
440 421
diff --git a/arch/arm64/kernel/armv8_deprecated.c b/arch/arm64/kernel/armv8_deprecated.c
index 5f72475e2e3b..42ffdb54e162 100644
--- a/arch/arm64/kernel/armv8_deprecated.c
+++ b/arch/arm64/kernel/armv8_deprecated.c
@@ -121,7 +121,7 @@ static int run_all_cpu_set_hw_mode(struct insn_emulation *insn, bool enable)
121 * 0 - If all the hooks ran successfully. 121 * 0 - If all the hooks ran successfully.
122 * -EINVAL - At least one hook is not supported by the CPU. 122 * -EINVAL - At least one hook is not supported by the CPU.
123 */ 123 */
124static int run_all_insn_set_hw_mode(unsigned long cpu) 124static int run_all_insn_set_hw_mode(unsigned int cpu)
125{ 125{
126 int rc = 0; 126 int rc = 0;
127 unsigned long flags; 127 unsigned long flags;
@@ -131,7 +131,7 @@ static int run_all_insn_set_hw_mode(unsigned long cpu)
131 list_for_each_entry(insn, &insn_emulation, node) { 131 list_for_each_entry(insn, &insn_emulation, node) {
132 bool enable = (insn->current_mode == INSN_HW); 132 bool enable = (insn->current_mode == INSN_HW);
133 if (insn->ops->set_hw_mode && insn->ops->set_hw_mode(enable)) { 133 if (insn->ops->set_hw_mode && insn->ops->set_hw_mode(enable)) {
134 pr_warn("CPU[%ld] cannot support the emulation of %s", 134 pr_warn("CPU[%u] cannot support the emulation of %s",
135 cpu, insn->ops->name); 135 cpu, insn->ops->name);
136 rc = -EINVAL; 136 rc = -EINVAL;
137 } 137 }
@@ -611,20 +611,6 @@ static struct insn_emulation_ops setend_ops = {
611 .set_hw_mode = setend_set_hw_mode, 611 .set_hw_mode = setend_set_hw_mode,
612}; 612};
613 613
614static int insn_cpu_hotplug_notify(struct notifier_block *b,
615 unsigned long action, void *hcpu)
616{
617 int rc = 0;
618 if ((action & ~CPU_TASKS_FROZEN) == CPU_STARTING)
619 rc = run_all_insn_set_hw_mode((unsigned long)hcpu);
620
621 return notifier_from_errno(rc);
622}
623
624static struct notifier_block insn_cpu_hotplug_notifier = {
625 .notifier_call = insn_cpu_hotplug_notify,
626};
627
628/* 614/*
629 * Invoked as late_initcall, since not needed before init spawned. 615 * Invoked as late_initcall, since not needed before init spawned.
630 */ 616 */
@@ -643,7 +629,9 @@ static int __init armv8_deprecated_init(void)
643 pr_info("setend instruction emulation is not supported on the system"); 629 pr_info("setend instruction emulation is not supported on the system");
644 } 630 }
645 631
646 register_cpu_notifier(&insn_cpu_hotplug_notifier); 632 cpuhp_setup_state_nocalls(CPUHP_AP_ARM64_ISNDEP_STARTING,
633 "AP_ARM64_ISNDEP_STARTING",
634 run_all_insn_set_hw_mode, NULL);
647 register_insn_emulation_sysctl(ctl_abi); 635 register_insn_emulation_sysctl(ctl_abi);
648 636
649 return 0; 637 return 0;
diff --git a/arch/blackfin/kernel/perf_event.c b/arch/blackfin/kernel/perf_event.c
index 170d786807c4..6355e97d22b9 100644
--- a/arch/blackfin/kernel/perf_event.c
+++ b/arch/blackfin/kernel/perf_event.c
@@ -453,29 +453,13 @@ static struct pmu pmu = {
453 .read = bfin_pmu_read, 453 .read = bfin_pmu_read,
454}; 454};
455 455
456static void bfin_pmu_setup(int cpu) 456static int bfin_pmu_prepare_cpu(unsigned int cpu)
457{ 457{
458 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu); 458 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
459 459
460 bfin_write_PFCTL(0);
460 memset(cpuhw, 0, sizeof(struct cpu_hw_events)); 461 memset(cpuhw, 0, sizeof(struct cpu_hw_events));
461} 462 return 0;
462
463static int
464bfin_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
465{
466 unsigned int cpu = (long)hcpu;
467
468 switch (action & ~CPU_TASKS_FROZEN) {
469 case CPU_UP_PREPARE:
470 bfin_write_PFCTL(0);
471 bfin_pmu_setup(cpu);
472 break;
473
474 default:
475 break;
476 }
477
478 return NOTIFY_OK;
479} 463}
480 464
481static int __init bfin_pmu_init(void) 465static int __init bfin_pmu_init(void)
@@ -491,8 +475,8 @@ static int __init bfin_pmu_init(void)
491 475
492 ret = perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW); 476 ret = perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);
493 if (!ret) 477 if (!ret)
494 perf_cpu_notifier(bfin_pmu_notifier); 478 cpuhp_setup_state(CPUHP_PERF_BFIN, "PERF_BFIN",
495 479 bfin_pmu_prepare_cpu, NULL);
496 return ret; 480 return ret;
497} 481}
498early_initcall(bfin_pmu_init); 482early_initcall(bfin_pmu_init);
diff --git a/arch/metag/kernel/perf/perf_event.c b/arch/metag/kernel/perf/perf_event.c
index 33a365f924be..052cba23708c 100644
--- a/arch/metag/kernel/perf/perf_event.c
+++ b/arch/metag/kernel/perf/perf_event.c
@@ -806,25 +806,16 @@ static struct metag_pmu _metag_pmu = {
806}; 806};
807 807
808/* PMU CPU hotplug notifier */ 808/* PMU CPU hotplug notifier */
809static int metag_pmu_cpu_notify(struct notifier_block *b, unsigned long action, 809static int metag_pmu_starting_cpu(unsigned int cpu)
810 void *hcpu)
811{ 810{
812 unsigned int cpu = (unsigned int)hcpu;
813 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); 811 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
814 812
815 if ((action & ~CPU_TASKS_FROZEN) != CPU_STARTING)
816 return NOTIFY_DONE;
817
818 memset(cpuc, 0, sizeof(struct cpu_hw_events)); 813 memset(cpuc, 0, sizeof(struct cpu_hw_events));
819 raw_spin_lock_init(&cpuc->pmu_lock); 814 raw_spin_lock_init(&cpuc->pmu_lock);
820 815
821 return NOTIFY_OK; 816 return 0;
822} 817}
823 818
824static struct notifier_block metag_pmu_notifier = {
825 .notifier_call = metag_pmu_cpu_notify,
826};
827
828/* PMU Initialisation */ 819/* PMU Initialisation */
829static int __init init_hw_perf_events(void) 820static int __init init_hw_perf_events(void)
830{ 821{
@@ -876,16 +867,13 @@ static int __init init_hw_perf_events(void)
876 metag_out32(0, PERF_COUNT(0)); 867 metag_out32(0, PERF_COUNT(0));
877 metag_out32(0, PERF_COUNT(1)); 868 metag_out32(0, PERF_COUNT(1));
878 869
879 for_each_possible_cpu(cpu) { 870 cpuhp_setup_state(CPUHP_AP_PERF_METAG_STARTING,
880 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); 871 "AP_PERF_METAG_STARTING", metag_pmu_starting_cpu,
872 NULL);
881 873
882 memset(cpuc, 0, sizeof(struct cpu_hw_events));
883 raw_spin_lock_init(&cpuc->pmu_lock);
884 }
885
886 register_cpu_notifier(&metag_pmu_notifier);
887 ret = perf_pmu_register(&pmu, metag_pmu->name, PERF_TYPE_RAW); 874 ret = perf_pmu_register(&pmu, metag_pmu->name, PERF_TYPE_RAW);
888out: 875 if (ret)
876 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_METAG_STARTING);
889 return ret; 877 return ret;
890} 878}
891early_initcall(init_hw_perf_events); 879early_initcall(init_hw_perf_events);
diff --git a/arch/mips/oprofile/op_model_loongson3.c b/arch/mips/oprofile/op_model_loongson3.c
index 8bcf7fc40f0d..85f3ee4ab456 100644
--- a/arch/mips/oprofile/op_model_loongson3.c
+++ b/arch/mips/oprofile/op_model_loongson3.c
@@ -168,33 +168,26 @@ static int loongson3_perfcount_handler(void)
168 return handled; 168 return handled;
169} 169}
170 170
171static int loongson3_cpu_callback(struct notifier_block *nfb, 171static int loongson3_starting_cpu(unsigned int cpu)
172 unsigned long action, void *hcpu)
173{ 172{
174 switch (action) { 173 write_c0_perflo1(reg.control1);
175 case CPU_STARTING: 174 write_c0_perflo2(reg.control2);
176 case CPU_STARTING_FROZEN: 175 return 0;
177 write_c0_perflo1(reg.control1);
178 write_c0_perflo2(reg.control2);
179 break;
180 case CPU_DYING:
181 case CPU_DYING_FROZEN:
182 write_c0_perflo1(0xc0000000);
183 write_c0_perflo2(0x40000000);
184 break;
185 }
186
187 return NOTIFY_OK;
188} 176}
189 177
190static struct notifier_block loongson3_notifier_block = { 178static int loongson3_dying_cpu(unsigned int cpu)
191 .notifier_call = loongson3_cpu_callback 179{
192}; 180 write_c0_perflo1(0xc0000000);
181 write_c0_perflo2(0x40000000);
182 return 0;
183}
193 184
194static int __init loongson3_init(void) 185static int __init loongson3_init(void)
195{ 186{
196 on_each_cpu(reset_counters, NULL, 1); 187 on_each_cpu(reset_counters, NULL, 1);
197 register_hotcpu_notifier(&loongson3_notifier_block); 188 cpuhp_setup_state_nocalls(CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
189 "AP_MIPS_OP_LOONGSON3_STARTING",
190 loongson3_starting_cpu, loongson3_dying_cpu);
198 save_perf_irq = perf_irq; 191 save_perf_irq = perf_irq;
199 perf_irq = loongson3_perfcount_handler; 192 perf_irq = loongson3_perfcount_handler;
200 193
@@ -204,7 +197,7 @@ static int __init loongson3_init(void)
204static void loongson3_exit(void) 197static void loongson3_exit(void)
205{ 198{
206 on_each_cpu(reset_counters, NULL, 1); 199 on_each_cpu(reset_counters, NULL, 1);
207 unregister_hotcpu_notifier(&loongson3_notifier_block); 200 cpuhp_remove_state_nocalls(CPUHP_AP_MIPS_OP_LOONGSON3_STARTING);
208 perf_irq = save_perf_irq; 201 perf_irq = save_perf_irq;
209} 202}
210 203
diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 669a15e7fa76..6dc07ddbfd04 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -581,30 +581,22 @@ static void verify_cpu_node_mapping(int cpu, int node)
581 } 581 }
582} 582}
583 583
584static int cpu_numa_callback(struct notifier_block *nfb, unsigned long action, 584/* Must run before sched domains notifier. */
585 void *hcpu) 585static int ppc_numa_cpu_prepare(unsigned int cpu)
586{ 586{
587 unsigned long lcpu = (unsigned long)hcpu; 587 int nid;
588 int ret = NOTIFY_DONE, nid;
589 588
590 switch (action) { 589 nid = numa_setup_cpu(cpu);
591 case CPU_UP_PREPARE: 590 verify_cpu_node_mapping(cpu, nid);
592 case CPU_UP_PREPARE_FROZEN: 591 return 0;
593 nid = numa_setup_cpu(lcpu); 592}
594 verify_cpu_node_mapping((int)lcpu, nid); 593
595 ret = NOTIFY_OK; 594static int ppc_numa_cpu_dead(unsigned int cpu)
596 break; 595{
597#ifdef CONFIG_HOTPLUG_CPU 596#ifdef CONFIG_HOTPLUG_CPU
598 case CPU_DEAD: 597 unmap_cpu_from_node(cpu);
599 case CPU_DEAD_FROZEN:
600 case CPU_UP_CANCELED:
601 case CPU_UP_CANCELED_FROZEN:
602 unmap_cpu_from_node(lcpu);
603 ret = NOTIFY_OK;
604 break;
605#endif 598#endif
606 } 599 return 0;
607 return ret;
608} 600}
609 601
610/* 602/*
@@ -913,11 +905,6 @@ static void __init dump_numa_memory_topology(void)
913 } 905 }
914} 906}
915 907
916static struct notifier_block ppc64_numa_nb = {
917 .notifier_call = cpu_numa_callback,
918 .priority = 1 /* Must run before sched domains notifier. */
919};
920
921/* Initialize NODE_DATA for a node on the local memory */ 908/* Initialize NODE_DATA for a node on the local memory */
922static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn) 909static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn)
923{ 910{
@@ -985,15 +972,18 @@ void __init initmem_init(void)
985 setup_node_to_cpumask_map(); 972 setup_node_to_cpumask_map();
986 973
987 reset_numa_cpu_lookup_table(); 974 reset_numa_cpu_lookup_table();
988 register_cpu_notifier(&ppc64_numa_nb); 975
989 /* 976 /*
990 * We need the numa_cpu_lookup_table to be accurate for all CPUs, 977 * We need the numa_cpu_lookup_table to be accurate for all CPUs,
991 * even before we online them, so that we can use cpu_to_{node,mem} 978 * even before we online them, so that we can use cpu_to_{node,mem}
992 * early in boot, cf. smp_prepare_cpus(). 979 * early in boot, cf. smp_prepare_cpus().
980 * _nocalls() + manual invocation is used because cpuhp is not yet
981 * initialized for the boot CPU.
993 */ 982 */
994 for_each_present_cpu(cpu) { 983 cpuhp_setup_state_nocalls(CPUHP_POWER_NUMA_PREPARE, "POWER_NUMA_PREPARE",
995 numa_setup_cpu((unsigned long)cpu); 984 ppc_numa_cpu_prepare, ppc_numa_cpu_dead);
996 } 985 for_each_present_cpu(cpu)
986 numa_setup_cpu(cpu);
997} 987}
998 988
999static int __init early_numa(char *p) 989static int __init early_numa(char *p)
diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index 97a1d40d8696..ffd61d55fb25 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -2158,31 +2158,15 @@ static void perf_event_interrupt(struct pt_regs *regs)
2158 irq_exit(); 2158 irq_exit();
2159} 2159}
2160 2160
2161static void power_pmu_setup(int cpu) 2161int power_pmu_prepare_cpu(unsigned int cpu)
2162{ 2162{
2163 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu); 2163 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
2164 2164
2165 if (!ppmu) 2165 if (ppmu) {
2166 return; 2166 memset(cpuhw, 0, sizeof(*cpuhw));
2167 memset(cpuhw, 0, sizeof(*cpuhw)); 2167 cpuhw->mmcr[0] = MMCR0_FC;
2168 cpuhw->mmcr[0] = MMCR0_FC;
2169}
2170
2171static int
2172power_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
2173{
2174 unsigned int cpu = (long)hcpu;
2175
2176 switch (action & ~CPU_TASKS_FROZEN) {
2177 case CPU_UP_PREPARE:
2178 power_pmu_setup(cpu);
2179 break;
2180
2181 default:
2182 break;
2183 } 2168 }
2184 2169 return 0;
2185 return NOTIFY_OK;
2186} 2170}
2187 2171
2188int register_power_pmu(struct power_pmu *pmu) 2172int register_power_pmu(struct power_pmu *pmu)
@@ -2205,7 +2189,7 @@ int register_power_pmu(struct power_pmu *pmu)
2205#endif /* CONFIG_PPC64 */ 2189#endif /* CONFIG_PPC64 */
2206 2190
2207 perf_pmu_register(&power_pmu, "cpu", PERF_TYPE_RAW); 2191 perf_pmu_register(&power_pmu, "cpu", PERF_TYPE_RAW);
2208 perf_cpu_notifier(power_pmu_notifier); 2192 cpuhp_setup_state(CPUHP_PERF_POWER, "PERF_POWER",
2209 2193 power_pmu_prepare_cpu, NULL);
2210 return 0; 2194 return 0;
2211} 2195}
diff --git a/arch/s390/kernel/perf_cpum_cf.c b/arch/s390/kernel/perf_cpum_cf.c
index 7ec63b1d920d..037c2a253ae4 100644
--- a/arch/s390/kernel/perf_cpum_cf.c
+++ b/arch/s390/kernel/perf_cpum_cf.c
@@ -664,30 +664,22 @@ static struct pmu cpumf_pmu = {
664 .cancel_txn = cpumf_pmu_cancel_txn, 664 .cancel_txn = cpumf_pmu_cancel_txn,
665}; 665};
666 666
667static int cpumf_pmu_notifier(struct notifier_block *self, unsigned long action, 667static int cpumf_pmf_setup(unsigned int cpu, int flags)
668 void *hcpu)
669{ 668{
670 int flags; 669 local_irq_disable();
671 670 setup_pmc_cpu(&flags);
672 switch (action & ~CPU_TASKS_FROZEN) { 671 local_irq_enable();
673 case CPU_ONLINE: 672 return 0;
674 case CPU_DOWN_FAILED: 673}
675 flags = PMC_INIT; 674
676 local_irq_disable(); 675static int s390_pmu_online_cpu(unsigned int cpu)
677 setup_pmc_cpu(&flags); 676{
678 local_irq_enable(); 677 return cpumf_pmf_setup(cpu, PMC_INIT);
679 break; 678}
680 case CPU_DOWN_PREPARE:
681 flags = PMC_RELEASE;
682 local_irq_disable();
683 setup_pmc_cpu(&flags);
684 local_irq_enable();
685 break;
686 default:
687 break;
688 }
689 679
690 return NOTIFY_OK; 680static int s390_pmu_offline_cpu(unsigned int cpu)
681{
682 return cpumf_pmf_setup(cpu, PMC_RELEASE);
691} 683}
692 684
693static int __init cpumf_pmu_init(void) 685static int __init cpumf_pmu_init(void)
@@ -707,7 +699,7 @@ static int __init cpumf_pmu_init(void)
707 if (rc) { 699 if (rc) {
708 pr_err("Registering for CPU-measurement alerts " 700 pr_err("Registering for CPU-measurement alerts "
709 "failed with rc=%i\n", rc); 701 "failed with rc=%i\n", rc);
710 goto out; 702 return rc;
711 } 703 }
712 704
713 cpumf_pmu.attr_groups = cpumf_cf_event_group(); 705 cpumf_pmu.attr_groups = cpumf_cf_event_group();
@@ -716,10 +708,10 @@ static int __init cpumf_pmu_init(void)
716 pr_err("Registering the cpum_cf PMU failed with rc=%i\n", rc); 708 pr_err("Registering the cpum_cf PMU failed with rc=%i\n", rc);
717 unregister_external_irq(EXT_IRQ_MEASURE_ALERT, 709 unregister_external_irq(EXT_IRQ_MEASURE_ALERT,
718 cpumf_measurement_alert); 710 cpumf_measurement_alert);
719 goto out; 711 return rc;
720 } 712 }
721 perf_cpu_notifier(cpumf_pmu_notifier); 713 return cpuhp_setup_state(CPUHP_AP_PERF_S390_CF_ONLINE,
722out: 714 "AP_PERF_S390_CF_ONLINE",
723 return rc; 715 s390_pmu_online_cpu, s390_pmu_offline_cpu);
724} 716}
725early_initcall(cpumf_pmu_init); 717early_initcall(cpumf_pmu_init);
diff --git a/arch/s390/kernel/perf_cpum_sf.c b/arch/s390/kernel/perf_cpum_sf.c
index 53acf2d76fa9..fcc634c1479a 100644
--- a/arch/s390/kernel/perf_cpum_sf.c
+++ b/arch/s390/kernel/perf_cpum_sf.c
@@ -1504,37 +1504,28 @@ static void cpumf_measurement_alert(struct ext_code ext_code,
1504 sf_disable(); 1504 sf_disable();
1505 } 1505 }
1506} 1506}
1507 1507static int cpusf_pmu_setup(unsigned int cpu, int flags)
1508static int cpumf_pmu_notifier(struct notifier_block *self,
1509 unsigned long action, void *hcpu)
1510{ 1508{
1511 int flags;
1512
1513 /* Ignore the notification if no events are scheduled on the PMU. 1509 /* Ignore the notification if no events are scheduled on the PMU.
1514 * This might be racy... 1510 * This might be racy...
1515 */ 1511 */
1516 if (!atomic_read(&num_events)) 1512 if (!atomic_read(&num_events))
1517 return NOTIFY_OK; 1513 return 0;
1518
1519 switch (action & ~CPU_TASKS_FROZEN) {
1520 case CPU_ONLINE:
1521 case CPU_DOWN_FAILED:
1522 flags = PMC_INIT;
1523 local_irq_disable();
1524 setup_pmc_cpu(&flags);
1525 local_irq_enable();
1526 break;
1527 case CPU_DOWN_PREPARE:
1528 flags = PMC_RELEASE;
1529 local_irq_disable();
1530 setup_pmc_cpu(&flags);
1531 local_irq_enable();
1532 break;
1533 default:
1534 break;
1535 }
1536 1514
1537 return NOTIFY_OK; 1515 local_irq_disable();
1516 setup_pmc_cpu(&flags);
1517 local_irq_enable();
1518 return 0;
1519}
1520
1521static int s390_pmu_sf_online_cpu(unsigned int cpu)
1522{
1523 return cpusf_pmu_setup(cpu, PMC_INIT);
1524}
1525
1526static int s390_pmu_sf_offline_cpu(unsigned int cpu)
1527{
1528 return cpusf_pmu_setup(cpu, PMC_RELEASE);
1538} 1529}
1539 1530
1540static int param_get_sfb_size(char *buffer, const struct kernel_param *kp) 1531static int param_get_sfb_size(char *buffer, const struct kernel_param *kp)
@@ -1634,7 +1625,9 @@ static int __init init_cpum_sampling_pmu(void)
1634 cpumf_measurement_alert); 1625 cpumf_measurement_alert);
1635 goto out; 1626 goto out;
1636 } 1627 }
1637 perf_cpu_notifier(cpumf_pmu_notifier); 1628
1629 cpuhp_setup_state(CPUHP_AP_PERF_S390_SF_ONLINE, "AP_PERF_S390_SF_ONLINE",
1630 s390_pmu_sf_online_cpu, s390_pmu_sf_offline_cpu);
1638out: 1631out:
1639 return err; 1632 return err;
1640} 1633}
diff --git a/arch/sh/kernel/perf_event.c b/arch/sh/kernel/perf_event.c
index 4dca18347ee9..ba3269a8304b 100644
--- a/arch/sh/kernel/perf_event.c
+++ b/arch/sh/kernel/perf_event.c
@@ -352,28 +352,12 @@ static struct pmu pmu = {
352 .read = sh_pmu_read, 352 .read = sh_pmu_read,
353}; 353};
354 354
355static void sh_pmu_setup(int cpu) 355static int sh_pmu_prepare_cpu(unsigned int cpu)
356{ 356{
357 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu); 357 struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
358 358
359 memset(cpuhw, 0, sizeof(struct cpu_hw_events)); 359 memset(cpuhw, 0, sizeof(struct cpu_hw_events));
360} 360 return 0;
361
362static int
363sh_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
364{
365 unsigned int cpu = (long)hcpu;
366
367 switch (action & ~CPU_TASKS_FROZEN) {
368 case CPU_UP_PREPARE:
369 sh_pmu_setup(cpu);
370 break;
371
372 default:
373 break;
374 }
375
376 return NOTIFY_OK;
377} 361}
378 362
379int register_sh_pmu(struct sh_pmu *_pmu) 363int register_sh_pmu(struct sh_pmu *_pmu)
@@ -394,6 +378,7 @@ int register_sh_pmu(struct sh_pmu *_pmu)
394 WARN_ON(_pmu->num_events > MAX_HWEVENTS); 378 WARN_ON(_pmu->num_events > MAX_HWEVENTS);
395 379
396 perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW); 380 perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);
397 perf_cpu_notifier(sh_pmu_notifier); 381 cpuhp_setup_state(CPUHP_PERF_SUPERH, "PERF_SUPERH", sh_pmu_prepare_cpu,
382 NULL);
398 return 0; 383 return 0;
399} 384}
diff --git a/arch/x86/entry/vdso/vma.c b/arch/x86/entry/vdso/vma.c
index 3329844e3c43..f840766659a8 100644
--- a/arch/x86/entry/vdso/vma.c
+++ b/arch/x86/entry/vdso/vma.c
@@ -331,15 +331,9 @@ static void vgetcpu_cpu_init(void *arg)
331 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S); 331 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
332} 332}
333 333
334static int 334static int vgetcpu_online(unsigned int cpu)
335vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
336{ 335{
337 long cpu = (long)arg; 336 return smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
338
339 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
340 smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
341
342 return NOTIFY_DONE;
343} 337}
344 338
345static int __init init_vdso(void) 339static int __init init_vdso(void)
@@ -350,15 +344,9 @@ static int __init init_vdso(void)
350 init_vdso_image(&vdso_image_x32); 344 init_vdso_image(&vdso_image_x32);
351#endif 345#endif
352 346
353 cpu_notifier_register_begin();
354
355 on_each_cpu(vgetcpu_cpu_init, NULL, 1);
356 /* notifier priority > KVM */ 347 /* notifier priority > KVM */
357 __hotcpu_notifier(vgetcpu_cpu_notifier, 30); 348 return cpuhp_setup_state(CPUHP_AP_X86_VDSO_VMA_ONLINE,
358 349 "AP_X86_VDSO_VMA_ONLINE", vgetcpu_online, NULL);
359 cpu_notifier_register_done();
360
361 return 0;
362} 350}
363subsys_initcall(init_vdso); 351subsys_initcall(init_vdso);
364#endif /* CONFIG_X86_64 */ 352#endif /* CONFIG_X86_64 */
diff --git a/arch/x86/events/amd/core.c b/arch/x86/events/amd/core.c
index bd3e8421b57c..e07a22bb9308 100644
--- a/arch/x86/events/amd/core.c
+++ b/arch/x86/events/amd/core.c
@@ -370,13 +370,13 @@ static int amd_pmu_cpu_prepare(int cpu)
370 WARN_ON_ONCE(cpuc->amd_nb); 370 WARN_ON_ONCE(cpuc->amd_nb);
371 371
372 if (!x86_pmu.amd_nb_constraints) 372 if (!x86_pmu.amd_nb_constraints)
373 return NOTIFY_OK; 373 return 0;
374 374
375 cpuc->amd_nb = amd_alloc_nb(cpu); 375 cpuc->amd_nb = amd_alloc_nb(cpu);
376 if (!cpuc->amd_nb) 376 if (!cpuc->amd_nb)
377 return NOTIFY_BAD; 377 return -ENOMEM;
378 378
379 return NOTIFY_OK; 379 return 0;
380} 380}
381 381
382static void amd_pmu_cpu_starting(int cpu) 382static void amd_pmu_cpu_starting(int cpu)
diff --git a/arch/x86/events/amd/ibs.c b/arch/x86/events/amd/ibs.c
index 72dea2f40fc4..155ea5324ae0 100644
--- a/arch/x86/events/amd/ibs.c
+++ b/arch/x86/events/amd/ibs.c
@@ -725,13 +725,10 @@ static __init int perf_ibs_pmu_init(struct perf_ibs *perf_ibs, char *name)
725 return ret; 725 return ret;
726} 726}
727 727
728static __init int perf_event_ibs_init(void) 728static __init void perf_event_ibs_init(void)
729{ 729{
730 struct attribute **attr = ibs_op_format_attrs; 730 struct attribute **attr = ibs_op_format_attrs;
731 731
732 if (!ibs_caps)
733 return -ENODEV; /* ibs not supported by the cpu */
734
735 perf_ibs_pmu_init(&perf_ibs_fetch, "ibs_fetch"); 732 perf_ibs_pmu_init(&perf_ibs_fetch, "ibs_fetch");
736 733
737 if (ibs_caps & IBS_CAPS_OPCNT) { 734 if (ibs_caps & IBS_CAPS_OPCNT) {
@@ -742,13 +739,11 @@ static __init int perf_event_ibs_init(void)
742 739
743 register_nmi_handler(NMI_LOCAL, perf_ibs_nmi_handler, 0, "perf_ibs"); 740 register_nmi_handler(NMI_LOCAL, perf_ibs_nmi_handler, 0, "perf_ibs");
744 pr_info("perf: AMD IBS detected (0x%08x)\n", ibs_caps); 741 pr_info("perf: AMD IBS detected (0x%08x)\n", ibs_caps);
745
746 return 0;
747} 742}
748 743
749#else /* defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD) */ 744#else /* defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_AMD) */
750 745
751static __init int perf_event_ibs_init(void) { return 0; } 746static __init void perf_event_ibs_init(void) { }
752 747
753#endif 748#endif
754 749
@@ -925,7 +920,7 @@ static inline int get_ibs_lvt_offset(void)
925 return val & IBSCTL_LVT_OFFSET_MASK; 920 return val & IBSCTL_LVT_OFFSET_MASK;
926} 921}
927 922
928static void setup_APIC_ibs(void *dummy) 923static void setup_APIC_ibs(void)
929{ 924{
930 int offset; 925 int offset;
931 926
@@ -940,7 +935,7 @@ failed:
940 smp_processor_id()); 935 smp_processor_id());
941} 936}
942 937
943static void clear_APIC_ibs(void *dummy) 938static void clear_APIC_ibs(void)
944{ 939{
945 int offset; 940 int offset;
946 941
@@ -949,18 +944,24 @@ static void clear_APIC_ibs(void *dummy)
949 setup_APIC_eilvt(offset, 0, APIC_EILVT_MSG_FIX, 1); 944 setup_APIC_eilvt(offset, 0, APIC_EILVT_MSG_FIX, 1);
950} 945}
951 946
947static int x86_pmu_amd_ibs_starting_cpu(unsigned int cpu)
948{
949 setup_APIC_ibs();
950 return 0;
951}
952
952#ifdef CONFIG_PM 953#ifdef CONFIG_PM
953 954
954static int perf_ibs_suspend(void) 955static int perf_ibs_suspend(void)
955{ 956{
956 clear_APIC_ibs(NULL); 957 clear_APIC_ibs();
957 return 0; 958 return 0;
958} 959}
959 960
960static void perf_ibs_resume(void) 961static void perf_ibs_resume(void)
961{ 962{
962 ibs_eilvt_setup(); 963 ibs_eilvt_setup();
963 setup_APIC_ibs(NULL); 964 setup_APIC_ibs();
964} 965}
965 966
966static struct syscore_ops perf_ibs_syscore_ops = { 967static struct syscore_ops perf_ibs_syscore_ops = {
@@ -979,27 +980,15 @@ static inline void perf_ibs_pm_init(void) { }
979 980
980#endif 981#endif
981 982
982static int 983static int x86_pmu_amd_ibs_dying_cpu(unsigned int cpu)
983perf_ibs_cpu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
984{ 984{
985 switch (action & ~CPU_TASKS_FROZEN) { 985 clear_APIC_ibs();
986 case CPU_STARTING: 986 return 0;
987 setup_APIC_ibs(NULL);
988 break;
989 case CPU_DYING:
990 clear_APIC_ibs(NULL);
991 break;
992 default:
993 break;
994 }
995
996 return NOTIFY_OK;
997} 987}
998 988
999static __init int amd_ibs_init(void) 989static __init int amd_ibs_init(void)
1000{ 990{
1001 u32 caps; 991 u32 caps;
1002 int ret = -EINVAL;
1003 992
1004 caps = __get_ibs_caps(); 993 caps = __get_ibs_caps();
1005 if (!caps) 994 if (!caps)
@@ -1008,22 +997,25 @@ static __init int amd_ibs_init(void)
1008 ibs_eilvt_setup(); 997 ibs_eilvt_setup();
1009 998
1010 if (!ibs_eilvt_valid()) 999 if (!ibs_eilvt_valid())
1011 goto out; 1000 return -EINVAL;
1012 1001
1013 perf_ibs_pm_init(); 1002 perf_ibs_pm_init();
1014 cpu_notifier_register_begin(); 1003
1015 ibs_caps = caps; 1004 ibs_caps = caps;
1016 /* make ibs_caps visible to other cpus: */ 1005 /* make ibs_caps visible to other cpus: */
1017 smp_mb(); 1006 smp_mb();
1018 smp_call_function(setup_APIC_ibs, NULL, 1); 1007 /*
1019 __perf_cpu_notifier(perf_ibs_cpu_notifier); 1008 * x86_pmu_amd_ibs_starting_cpu will be called from core on
1020 cpu_notifier_register_done(); 1009 * all online cpus.
1010 */
1011 cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_IBS_STARTING,
1012 "AP_PERF_X86_AMD_IBS_STARTING",
1013 x86_pmu_amd_ibs_starting_cpu,
1014 x86_pmu_amd_ibs_dying_cpu);
1021 1015
1022 ret = perf_event_ibs_init(); 1016 perf_event_ibs_init();
1023out: 1017
1024 if (ret) 1018 return 0;
1025 pr_err("Failed to setup IBS, %d\n", ret);
1026 return ret;
1027} 1019}
1028 1020
1029/* Since we need the pci subsystem to init ibs we can't do this earlier: */ 1021/* Since we need the pci subsystem to init ibs we can't do this earlier: */
diff --git a/arch/x86/events/amd/power.c b/arch/x86/events/amd/power.c
index 55a3529dbf12..9842270ed2f2 100644
--- a/arch/x86/events/amd/power.c
+++ b/arch/x86/events/amd/power.c
@@ -228,12 +228,12 @@ static struct pmu pmu_class = {
228 .read = pmu_event_read, 228 .read = pmu_event_read,
229}; 229};
230 230
231static void power_cpu_exit(int cpu) 231static int power_cpu_exit(unsigned int cpu)
232{ 232{
233 int target; 233 int target;
234 234
235 if (!cpumask_test_and_clear_cpu(cpu, &cpu_mask)) 235 if (!cpumask_test_and_clear_cpu(cpu, &cpu_mask))
236 return; 236 return 0;
237 237
238 /* 238 /*
239 * Find a new CPU on the same compute unit, if was set in cpumask 239 * Find a new CPU on the same compute unit, if was set in cpumask
@@ -245,9 +245,10 @@ static void power_cpu_exit(int cpu)
245 cpumask_set_cpu(target, &cpu_mask); 245 cpumask_set_cpu(target, &cpu_mask);
246 perf_pmu_migrate_context(&pmu_class, cpu, target); 246 perf_pmu_migrate_context(&pmu_class, cpu, target);
247 } 247 }
248 return 0;
248} 249}
249 250
250static void power_cpu_init(int cpu) 251static int power_cpu_init(unsigned int cpu)
251{ 252{
252 int target; 253 int target;
253 254
@@ -255,7 +256,7 @@ static void power_cpu_init(int cpu)
255 * 1) If any CPU is set at cpu_mask in the same compute unit, do 256 * 1) If any CPU is set at cpu_mask in the same compute unit, do
256 * nothing. 257 * nothing.
257 * 2) If no CPU is set at cpu_mask in the same compute unit, 258 * 2) If no CPU is set at cpu_mask in the same compute unit,
258 * set current STARTING CPU. 259 * set current ONLINE CPU.
259 * 260 *
260 * Note: if there is a CPU aside of the new one already in the 261 * Note: if there is a CPU aside of the new one already in the
261 * sibling mask, then it is also in cpu_mask. 262 * sibling mask, then it is also in cpu_mask.
@@ -263,33 +264,9 @@ static void power_cpu_init(int cpu)
263 target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu); 264 target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
264 if (target >= nr_cpumask_bits) 265 if (target >= nr_cpumask_bits)
265 cpumask_set_cpu(cpu, &cpu_mask); 266 cpumask_set_cpu(cpu, &cpu_mask);
267 return 0;
266} 268}
267 269
268static int
269power_cpu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
270{
271 unsigned int cpu = (long)hcpu;
272
273 switch (action & ~CPU_TASKS_FROZEN) {
274 case CPU_DOWN_FAILED:
275 case CPU_STARTING:
276 power_cpu_init(cpu);
277 break;
278 case CPU_DOWN_PREPARE:
279 power_cpu_exit(cpu);
280 break;
281 default:
282 break;
283 }
284
285 return NOTIFY_OK;
286}
287
288static struct notifier_block power_cpu_notifier_nb = {
289 .notifier_call = power_cpu_notifier,
290 .priority = CPU_PRI_PERF,
291};
292
293static const struct x86_cpu_id cpu_match[] = { 270static const struct x86_cpu_id cpu_match[] = {
294 { .vendor = X86_VENDOR_AMD, .family = 0x15 }, 271 { .vendor = X86_VENDOR_AMD, .family = 0x15 },
295 {}, 272 {},
@@ -297,7 +274,7 @@ static const struct x86_cpu_id cpu_match[] = {
297 274
298static int __init amd_power_pmu_init(void) 275static int __init amd_power_pmu_init(void)
299{ 276{
300 int cpu, target, ret; 277 int ret;
301 278
302 if (!x86_match_cpu(cpu_match)) 279 if (!x86_match_cpu(cpu_match))
303 return 0; 280 return 0;
@@ -312,38 +289,25 @@ static int __init amd_power_pmu_init(void)
312 return -ENODEV; 289 return -ENODEV;
313 } 290 }
314 291
315 cpu_notifier_register_begin();
316 292
317 /* Choose one online core of each compute unit. */ 293 cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_POWER_ONLINE,
318 for_each_online_cpu(cpu) { 294 "AP_PERF_X86_AMD_POWER_ONLINE",
319 target = cpumask_first(topology_sibling_cpumask(cpu)); 295 power_cpu_init, power_cpu_exit);
320 if (!cpumask_test_cpu(target, &cpu_mask))
321 cpumask_set_cpu(target, &cpu_mask);
322 }
323 296
324 ret = perf_pmu_register(&pmu_class, "power", -1); 297 ret = perf_pmu_register(&pmu_class, "power", -1);
325 if (WARN_ON(ret)) { 298 if (WARN_ON(ret)) {
326 pr_warn("AMD Power PMU registration failed\n"); 299 pr_warn("AMD Power PMU registration failed\n");
327 goto out; 300 return ret;
328 } 301 }
329 302
330 __register_cpu_notifier(&power_cpu_notifier_nb);
331
332 pr_info("AMD Power PMU detected\n"); 303 pr_info("AMD Power PMU detected\n");
333
334out:
335 cpu_notifier_register_done();
336
337 return ret; 304 return ret;
338} 305}
339module_init(amd_power_pmu_init); 306module_init(amd_power_pmu_init);
340 307
341static void __exit amd_power_pmu_exit(void) 308static void __exit amd_power_pmu_exit(void)
342{ 309{
343 cpu_notifier_register_begin(); 310 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_AMD_POWER_ONLINE);
344 __unregister_cpu_notifier(&power_cpu_notifier_nb);
345 cpu_notifier_register_done();
346
347 perf_pmu_unregister(&pmu_class); 311 perf_pmu_unregister(&pmu_class);
348} 312}
349module_exit(amd_power_pmu_exit); 313module_exit(amd_power_pmu_exit);
diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index 98ac57381bf9..e6131d4454e6 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -358,7 +358,7 @@ amd_uncore_find_online_sibling(struct amd_uncore *this,
358 return this; 358 return this;
359} 359}
360 360
361static void amd_uncore_cpu_starting(unsigned int cpu) 361static int amd_uncore_cpu_starting(unsigned int cpu)
362{ 362{
363 unsigned int eax, ebx, ecx, edx; 363 unsigned int eax, ebx, ecx, edx;
364 struct amd_uncore *uncore; 364 struct amd_uncore *uncore;
@@ -384,6 +384,8 @@ static void amd_uncore_cpu_starting(unsigned int cpu)
384 uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_l2); 384 uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_l2);
385 *per_cpu_ptr(amd_uncore_l2, cpu) = uncore; 385 *per_cpu_ptr(amd_uncore_l2, cpu) = uncore;
386 } 386 }
387
388 return 0;
387} 389}
388 390
389static void uncore_online(unsigned int cpu, 391static void uncore_online(unsigned int cpu,
@@ -398,13 +400,15 @@ static void uncore_online(unsigned int cpu,
398 cpumask_set_cpu(cpu, uncore->active_mask); 400 cpumask_set_cpu(cpu, uncore->active_mask);
399} 401}
400 402
401static void amd_uncore_cpu_online(unsigned int cpu) 403static int amd_uncore_cpu_online(unsigned int cpu)
402{ 404{
403 if (amd_uncore_nb) 405 if (amd_uncore_nb)
404 uncore_online(cpu, amd_uncore_nb); 406 uncore_online(cpu, amd_uncore_nb);
405 407
406 if (amd_uncore_l2) 408 if (amd_uncore_l2)
407 uncore_online(cpu, amd_uncore_l2); 409 uncore_online(cpu, amd_uncore_l2);
410
411 return 0;
408} 412}
409 413
410static void uncore_down_prepare(unsigned int cpu, 414static void uncore_down_prepare(unsigned int cpu,
@@ -433,13 +437,15 @@ static void uncore_down_prepare(unsigned int cpu,
433 } 437 }
434} 438}
435 439
436static void amd_uncore_cpu_down_prepare(unsigned int cpu) 440static int amd_uncore_cpu_down_prepare(unsigned int cpu)
437{ 441{
438 if (amd_uncore_nb) 442 if (amd_uncore_nb)
439 uncore_down_prepare(cpu, amd_uncore_nb); 443 uncore_down_prepare(cpu, amd_uncore_nb);
440 444
441 if (amd_uncore_l2) 445 if (amd_uncore_l2)
442 uncore_down_prepare(cpu, amd_uncore_l2); 446 uncore_down_prepare(cpu, amd_uncore_l2);
447
448 return 0;
443} 449}
444 450
445static void uncore_dead(unsigned int cpu, struct amd_uncore * __percpu *uncores) 451static void uncore_dead(unsigned int cpu, struct amd_uncore * __percpu *uncores)
@@ -454,74 +460,19 @@ static void uncore_dead(unsigned int cpu, struct amd_uncore * __percpu *uncores)
454 *per_cpu_ptr(uncores, cpu) = NULL; 460 *per_cpu_ptr(uncores, cpu) = NULL;
455} 461}
456 462
457static void amd_uncore_cpu_dead(unsigned int cpu) 463static int amd_uncore_cpu_dead(unsigned int cpu)
458{ 464{
459 if (amd_uncore_nb) 465 if (amd_uncore_nb)
460 uncore_dead(cpu, amd_uncore_nb); 466 uncore_dead(cpu, amd_uncore_nb);
461 467
462 if (amd_uncore_l2) 468 if (amd_uncore_l2)
463 uncore_dead(cpu, amd_uncore_l2); 469 uncore_dead(cpu, amd_uncore_l2);
464}
465
466static int
467amd_uncore_cpu_notifier(struct notifier_block *self, unsigned long action,
468 void *hcpu)
469{
470 unsigned int cpu = (long)hcpu;
471
472 switch (action & ~CPU_TASKS_FROZEN) {
473 case CPU_UP_PREPARE:
474 if (amd_uncore_cpu_up_prepare(cpu))
475 return notifier_from_errno(-ENOMEM);
476 break;
477
478 case CPU_STARTING:
479 amd_uncore_cpu_starting(cpu);
480 break;
481
482 case CPU_ONLINE:
483 amd_uncore_cpu_online(cpu);
484 break;
485
486 case CPU_DOWN_PREPARE:
487 amd_uncore_cpu_down_prepare(cpu);
488 break;
489
490 case CPU_UP_CANCELED:
491 case CPU_DEAD:
492 amd_uncore_cpu_dead(cpu);
493 break;
494
495 default:
496 break;
497 }
498
499 return NOTIFY_OK;
500}
501
502static struct notifier_block amd_uncore_cpu_notifier_block = {
503 .notifier_call = amd_uncore_cpu_notifier,
504 .priority = CPU_PRI_PERF + 1,
505};
506
507static void __init init_cpu_already_online(void *dummy)
508{
509 unsigned int cpu = smp_processor_id();
510
511 amd_uncore_cpu_starting(cpu);
512 amd_uncore_cpu_online(cpu);
513}
514 470
515static void cleanup_cpu_online(void *dummy) 471 return 0;
516{
517 unsigned int cpu = smp_processor_id();
518
519 amd_uncore_cpu_dead(cpu);
520} 472}
521 473
522static int __init amd_uncore_init(void) 474static int __init amd_uncore_init(void)
523{ 475{
524 unsigned int cpu, cpu2;
525 int ret = -ENODEV; 476 int ret = -ENODEV;
526 477
527 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) 478 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
@@ -558,38 +509,29 @@ static int __init amd_uncore_init(void)
558 ret = 0; 509 ret = 0;
559 } 510 }
560 511
561 if (ret) 512 /*
562 goto fail_nodev; 513 * Install callbacks. Core will call them for each online cpu.
563 514 */
564 cpu_notifier_register_begin(); 515 if (cpuhp_setup_state(CPUHP_PERF_X86_AMD_UNCORE_PREP,
565 516 "PERF_X86_AMD_UNCORE_PREP",
566 /* init cpus already online before registering for hotplug notifier */ 517 amd_uncore_cpu_up_prepare, amd_uncore_cpu_dead))
567 for_each_online_cpu(cpu) { 518 goto fail_l2;
568 ret = amd_uncore_cpu_up_prepare(cpu); 519
569 if (ret) 520 if (cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING,
570 goto fail_online; 521 "AP_PERF_X86_AMD_UNCORE_STARTING",
571 smp_call_function_single(cpu, init_cpu_already_online, NULL, 1); 522 amd_uncore_cpu_starting, NULL))
572 } 523 goto fail_prep;
573 524 if (cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_UNCORE_ONLINE,
574 __register_cpu_notifier(&amd_uncore_cpu_notifier_block); 525 "AP_PERF_X86_AMD_UNCORE_ONLINE",
575 cpu_notifier_register_done(); 526 amd_uncore_cpu_online,
576 527 amd_uncore_cpu_down_prepare))
528 goto fail_start;
577 return 0; 529 return 0;
578 530
579 531fail_start:
580fail_online: 532 cpuhp_remove_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING);
581 for_each_online_cpu(cpu2) { 533fail_prep:
582 if (cpu2 == cpu) 534 cpuhp_remove_state(CPUHP_PERF_X86_AMD_UNCORE_PREP);
583 break;
584 smp_call_function_single(cpu, cleanup_cpu_online, NULL, 1);
585 }
586 cpu_notifier_register_done();
587
588 /* amd_uncore_nb/l2 should have been freed by cleanup_cpu_online */
589 amd_uncore_nb = amd_uncore_l2 = NULL;
590
591 if (boot_cpu_has(X86_FEATURE_PERFCTR_L2))
592 perf_pmu_unregister(&amd_l2_pmu);
593fail_l2: 535fail_l2:
594 if (boot_cpu_has(X86_FEATURE_PERFCTR_NB)) 536 if (boot_cpu_has(X86_FEATURE_PERFCTR_NB))
595 perf_pmu_unregister(&amd_nb_pmu); 537 perf_pmu_unregister(&amd_nb_pmu);
diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index dfebbde2a4cc..c17f0de5fd39 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -1477,49 +1477,49 @@ NOKPROBE_SYMBOL(perf_event_nmi_handler);
1477struct event_constraint emptyconstraint; 1477struct event_constraint emptyconstraint;
1478struct event_constraint unconstrained; 1478struct event_constraint unconstrained;
1479 1479
1480static int 1480static int x86_pmu_prepare_cpu(unsigned int cpu)
1481x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
1482{ 1481{
1483 unsigned int cpu = (long)hcpu;
1484 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); 1482 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
1485 int i, ret = NOTIFY_OK; 1483 int i;
1486
1487 switch (action & ~CPU_TASKS_FROZEN) {
1488 case CPU_UP_PREPARE:
1489 for (i = 0 ; i < X86_PERF_KFREE_MAX; i++)
1490 cpuc->kfree_on_online[i] = NULL;
1491 if (x86_pmu.cpu_prepare)
1492 ret = x86_pmu.cpu_prepare(cpu);
1493 break;
1494
1495 case CPU_STARTING:
1496 if (x86_pmu.cpu_starting)
1497 x86_pmu.cpu_starting(cpu);
1498 break;
1499 1484
1500 case CPU_ONLINE: 1485 for (i = 0 ; i < X86_PERF_KFREE_MAX; i++)
1501 for (i = 0 ; i < X86_PERF_KFREE_MAX; i++) { 1486 cpuc->kfree_on_online[i] = NULL;
1502 kfree(cpuc->kfree_on_online[i]); 1487 if (x86_pmu.cpu_prepare)
1503 cpuc->kfree_on_online[i] = NULL; 1488 return x86_pmu.cpu_prepare(cpu);
1504 } 1489 return 0;
1505 break; 1490}
1506 1491
1507 case CPU_DYING: 1492static int x86_pmu_dead_cpu(unsigned int cpu)
1508 if (x86_pmu.cpu_dying) 1493{
1509 x86_pmu.cpu_dying(cpu); 1494 if (x86_pmu.cpu_dead)
1510 break; 1495 x86_pmu.cpu_dead(cpu);
1496 return 0;
1497}
1511 1498
1512 case CPU_UP_CANCELED: 1499static int x86_pmu_online_cpu(unsigned int cpu)
1513 case CPU_DEAD: 1500{
1514 if (x86_pmu.cpu_dead) 1501 struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
1515 x86_pmu.cpu_dead(cpu); 1502 int i;
1516 break;
1517 1503
1518 default: 1504 for (i = 0 ; i < X86_PERF_KFREE_MAX; i++) {
1519 break; 1505 kfree(cpuc->kfree_on_online[i]);
1506 cpuc->kfree_on_online[i] = NULL;
1520 } 1507 }
1508 return 0;
1509}
1521 1510
1522 return ret; 1511static int x86_pmu_starting_cpu(unsigned int cpu)
1512{
1513 if (x86_pmu.cpu_starting)
1514 x86_pmu.cpu_starting(cpu);
1515 return 0;
1516}
1517
1518static int x86_pmu_dying_cpu(unsigned int cpu)
1519{
1520 if (x86_pmu.cpu_dying)
1521 x86_pmu.cpu_dying(cpu);
1522 return 0;
1523} 1523}
1524 1524
1525static void __init pmu_check_apic(void) 1525static void __init pmu_check_apic(void)
@@ -1787,10 +1787,39 @@ static int __init init_hw_perf_events(void)
1787 pr_info("... fixed-purpose events: %d\n", x86_pmu.num_counters_fixed); 1787 pr_info("... fixed-purpose events: %d\n", x86_pmu.num_counters_fixed);
1788 pr_info("... event mask: %016Lx\n", x86_pmu.intel_ctrl); 1788 pr_info("... event mask: %016Lx\n", x86_pmu.intel_ctrl);
1789 1789
1790 perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW); 1790 /*
1791 perf_cpu_notifier(x86_pmu_notifier); 1791 * Install callbacks. Core will call them for each online
1792 * cpu.
1793 */
1794 err = cpuhp_setup_state(CPUHP_PERF_X86_PREPARE, "PERF_X86_PREPARE",
1795 x86_pmu_prepare_cpu, x86_pmu_dead_cpu);
1796 if (err)
1797 return err;
1798
1799 err = cpuhp_setup_state(CPUHP_AP_PERF_X86_STARTING,
1800 "AP_PERF_X86_STARTING", x86_pmu_starting_cpu,
1801 x86_pmu_dying_cpu);
1802 if (err)
1803 goto out;
1804
1805 err = cpuhp_setup_state(CPUHP_AP_PERF_X86_ONLINE, "AP_PERF_X86_ONLINE",
1806 x86_pmu_online_cpu, NULL);
1807 if (err)
1808 goto out1;
1809
1810 err = perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW);
1811 if (err)
1812 goto out2;
1792 1813
1793 return 0; 1814 return 0;
1815
1816out2:
1817 cpuhp_remove_state(CPUHP_AP_PERF_X86_ONLINE);
1818out1:
1819 cpuhp_remove_state(CPUHP_AP_PERF_X86_STARTING);
1820out:
1821 cpuhp_remove_state(CPUHP_PERF_X86_PREPARE);
1822 return err;
1794} 1823}
1795early_initcall(init_hw_perf_events); 1824early_initcall(init_hw_perf_events);
1796 1825
diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
index 0974ba11e954..2cbde2f449aa 100644
--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -3109,7 +3109,7 @@ static int intel_pmu_cpu_prepare(int cpu)
3109 cpuc->excl_thread_id = 0; 3109 cpuc->excl_thread_id = 0;
3110 } 3110 }
3111 3111
3112 return NOTIFY_OK; 3112 return 0;
3113 3113
3114err_constraint_list: 3114err_constraint_list:
3115 kfree(cpuc->constraint_list); 3115 kfree(cpuc->constraint_list);
@@ -3120,7 +3120,7 @@ err_shared_regs:
3120 cpuc->shared_regs = NULL; 3120 cpuc->shared_regs = NULL;
3121 3121
3122err: 3122err:
3123 return NOTIFY_BAD; 3123 return -ENOMEM;
3124} 3124}
3125 3125
3126static void intel_pmu_cpu_starting(int cpu) 3126static void intel_pmu_cpu_starting(int cpu)
diff --git a/arch/x86/events/intel/cqm.c b/arch/x86/events/intel/cqm.c
index 7b5fd811ef45..783c49ddef29 100644
--- a/arch/x86/events/intel/cqm.c
+++ b/arch/x86/events/intel/cqm.c
@@ -1577,7 +1577,7 @@ static inline void cqm_pick_event_reader(int cpu)
1577 cpumask_set_cpu(cpu, &cqm_cpumask); 1577 cpumask_set_cpu(cpu, &cqm_cpumask);
1578} 1578}
1579 1579
1580static void intel_cqm_cpu_starting(unsigned int cpu) 1580static int intel_cqm_cpu_starting(unsigned int cpu)
1581{ 1581{
1582 struct intel_pqr_state *state = &per_cpu(pqr_state, cpu); 1582 struct intel_pqr_state *state = &per_cpu(pqr_state, cpu);
1583 struct cpuinfo_x86 *c = &cpu_data(cpu); 1583 struct cpuinfo_x86 *c = &cpu_data(cpu);
@@ -1588,39 +1588,26 @@ static void intel_cqm_cpu_starting(unsigned int cpu)
1588 1588
1589 WARN_ON(c->x86_cache_max_rmid != cqm_max_rmid); 1589 WARN_ON(c->x86_cache_max_rmid != cqm_max_rmid);
1590 WARN_ON(c->x86_cache_occ_scale != cqm_l3_scale); 1590 WARN_ON(c->x86_cache_occ_scale != cqm_l3_scale);
1591
1592 cqm_pick_event_reader(cpu);
1593 return 0;
1591} 1594}
1592 1595
1593static void intel_cqm_cpu_exit(unsigned int cpu) 1596static int intel_cqm_cpu_exit(unsigned int cpu)
1594{ 1597{
1595 int target; 1598 int target;
1596 1599
1597 /* Is @cpu the current cqm reader for this package ? */ 1600 /* Is @cpu the current cqm reader for this package ? */
1598 if (!cpumask_test_and_clear_cpu(cpu, &cqm_cpumask)) 1601 if (!cpumask_test_and_clear_cpu(cpu, &cqm_cpumask))
1599 return; 1602 return 0;
1600 1603
1601 /* Find another online reader in this package */ 1604 /* Find another online reader in this package */
1602 target = cpumask_any_but(topology_core_cpumask(cpu), cpu); 1605 target = cpumask_any_but(topology_core_cpumask(cpu), cpu);
1603 1606
1604 if (target < nr_cpu_ids) 1607 if (target < nr_cpu_ids)
1605 cpumask_set_cpu(target, &cqm_cpumask); 1608 cpumask_set_cpu(target, &cqm_cpumask);
1606}
1607
1608static int intel_cqm_cpu_notifier(struct notifier_block *nb,
1609 unsigned long action, void *hcpu)
1610{
1611 unsigned int cpu = (unsigned long)hcpu;
1612
1613 switch (action & ~CPU_TASKS_FROZEN) {
1614 case CPU_DOWN_PREPARE:
1615 intel_cqm_cpu_exit(cpu);
1616 break;
1617 case CPU_STARTING:
1618 intel_cqm_cpu_starting(cpu);
1619 cqm_pick_event_reader(cpu);
1620 break;
1621 }
1622 1609
1623 return NOTIFY_OK; 1610 return 0;
1624} 1611}
1625 1612
1626static const struct x86_cpu_id intel_cqm_match[] = { 1613static const struct x86_cpu_id intel_cqm_match[] = {
@@ -1682,7 +1669,7 @@ out:
1682static int __init intel_cqm_init(void) 1669static int __init intel_cqm_init(void)
1683{ 1670{
1684 char *str = NULL, scale[20]; 1671 char *str = NULL, scale[20];
1685 int i, cpu, ret; 1672 int cpu, ret;
1686 1673
1687 if (x86_match_cpu(intel_cqm_match)) 1674 if (x86_match_cpu(intel_cqm_match))
1688 cqm_enabled = true; 1675 cqm_enabled = true;
@@ -1705,8 +1692,7 @@ static int __init intel_cqm_init(void)
1705 * 1692 *
1706 * Also, check that the scales match on all cpus. 1693 * Also, check that the scales match on all cpus.
1707 */ 1694 */
1708 cpu_notifier_register_begin(); 1695 get_online_cpus();
1709
1710 for_each_online_cpu(cpu) { 1696 for_each_online_cpu(cpu) {
1711 struct cpuinfo_x86 *c = &cpu_data(cpu); 1697 struct cpuinfo_x86 *c = &cpu_data(cpu);
1712 1698
@@ -1743,11 +1729,6 @@ static int __init intel_cqm_init(void)
1743 if (ret) 1729 if (ret)
1744 goto out; 1730 goto out;
1745 1731
1746 for_each_online_cpu(i) {
1747 intel_cqm_cpu_starting(i);
1748 cqm_pick_event_reader(i);
1749 }
1750
1751 if (mbm_enabled) 1732 if (mbm_enabled)
1752 ret = intel_mbm_init(); 1733 ret = intel_mbm_init();
1753 if (ret && !cqm_enabled) 1734 if (ret && !cqm_enabled)
@@ -1772,12 +1753,18 @@ static int __init intel_cqm_init(void)
1772 pr_info("Intel MBM enabled\n"); 1753 pr_info("Intel MBM enabled\n");
1773 1754
1774 /* 1755 /*
1775 * Register the hot cpu notifier once we are sure cqm 1756 * Setup the hot cpu notifier once we are sure cqm
1776 * is enabled to avoid notifier leak. 1757 * is enabled to avoid notifier leak.
1777 */ 1758 */
1778 __perf_cpu_notifier(intel_cqm_cpu_notifier); 1759 cpuhp_setup_state(CPUHP_AP_PERF_X86_CQM_STARTING,
1760 "AP_PERF_X86_CQM_STARTING",
1761 intel_cqm_cpu_starting, NULL);
1762 cpuhp_setup_state(CPUHP_AP_PERF_X86_CQM_ONLINE, "AP_PERF_X86_CQM_ONLINE",
1763 NULL, intel_cqm_cpu_exit);
1764
1779out: 1765out:
1780 cpu_notifier_register_done(); 1766 put_online_cpus();
1767
1781 if (ret) { 1768 if (ret) {
1782 kfree(str); 1769 kfree(str);
1783 cqm_cleanup(); 1770 cqm_cleanup();
diff --git a/arch/x86/events/intel/cstate.c b/arch/x86/events/intel/cstate.c
index 4c7638b91fa5..3ca87b5a8677 100644
--- a/arch/x86/events/intel/cstate.c
+++ b/arch/x86/events/intel/cstate.c
@@ -366,7 +366,7 @@ static int cstate_pmu_event_add(struct perf_event *event, int mode)
366 * Check if exiting cpu is the designated reader. If so migrate the 366 * Check if exiting cpu is the designated reader. If so migrate the
367 * events when there is a valid target available 367 * events when there is a valid target available
368 */ 368 */
369static void cstate_cpu_exit(int cpu) 369static int cstate_cpu_exit(unsigned int cpu)
370{ 370{
371 unsigned int target; 371 unsigned int target;
372 372
@@ -391,9 +391,10 @@ static void cstate_cpu_exit(int cpu)
391 perf_pmu_migrate_context(&cstate_pkg_pmu, cpu, target); 391 perf_pmu_migrate_context(&cstate_pkg_pmu, cpu, target);
392 } 392 }
393 } 393 }
394 return 0;
394} 395}
395 396
396static void cstate_cpu_init(int cpu) 397static int cstate_cpu_init(unsigned int cpu)
397{ 398{
398 unsigned int target; 399 unsigned int target;
399 400
@@ -415,31 +416,10 @@ static void cstate_cpu_init(int cpu)
415 topology_core_cpumask(cpu)); 416 topology_core_cpumask(cpu));
416 if (has_cstate_pkg && target >= nr_cpu_ids) 417 if (has_cstate_pkg && target >= nr_cpu_ids)
417 cpumask_set_cpu(cpu, &cstate_pkg_cpu_mask); 418 cpumask_set_cpu(cpu, &cstate_pkg_cpu_mask);
418}
419 419
420static int cstate_cpu_notifier(struct notifier_block *self, 420 return 0;
421 unsigned long action, void *hcpu)
422{
423 unsigned int cpu = (long)hcpu;
424
425 switch (action & ~CPU_TASKS_FROZEN) {
426 case CPU_STARTING:
427 cstate_cpu_init(cpu);
428 break;
429 case CPU_DOWN_PREPARE:
430 cstate_cpu_exit(cpu);
431 break;
432 default:
433 break;
434 }
435 return NOTIFY_OK;
436} 421}
437 422
438static struct notifier_block cstate_cpu_nb = {
439 .notifier_call = cstate_cpu_notifier,
440 .priority = CPU_PRI_PERF + 1,
441};
442
443static struct pmu cstate_core_pmu = { 423static struct pmu cstate_core_pmu = {
444 .attr_groups = core_attr_groups, 424 .attr_groups = core_attr_groups,
445 .name = "cstate_core", 425 .name = "cstate_core",
@@ -600,18 +580,20 @@ static inline void cstate_cleanup(void)
600 580
601static int __init cstate_init(void) 581static int __init cstate_init(void)
602{ 582{
603 int cpu, err; 583 int err;
604 584
605 cpu_notifier_register_begin(); 585 cpuhp_setup_state(CPUHP_AP_PERF_X86_CSTATE_STARTING,
606 for_each_online_cpu(cpu) 586 "AP_PERF_X86_CSTATE_STARTING", cstate_cpu_init,
607 cstate_cpu_init(cpu); 587 NULL);
588 cpuhp_setup_state(CPUHP_AP_PERF_X86_CSTATE_ONLINE,
589 "AP_PERF_X86_CSTATE_ONLINE", NULL, cstate_cpu_exit);
608 590
609 if (has_cstate_core) { 591 if (has_cstate_core) {
610 err = perf_pmu_register(&cstate_core_pmu, cstate_core_pmu.name, -1); 592 err = perf_pmu_register(&cstate_core_pmu, cstate_core_pmu.name, -1);
611 if (err) { 593 if (err) {
612 has_cstate_core = false; 594 has_cstate_core = false;
613 pr_info("Failed to register cstate core pmu\n"); 595 pr_info("Failed to register cstate core pmu\n");
614 goto out; 596 return err;
615 } 597 }
616 } 598 }
617 599
@@ -621,12 +603,10 @@ static int __init cstate_init(void)
621 has_cstate_pkg = false; 603 has_cstate_pkg = false;
622 pr_info("Failed to register cstate pkg pmu\n"); 604 pr_info("Failed to register cstate pkg pmu\n");
623 cstate_cleanup(); 605 cstate_cleanup();
624 goto out; 606 return err;
625 } 607 }
626 } 608 }
627 __register_cpu_notifier(&cstate_cpu_nb); 609
628out:
629 cpu_notifier_register_done();
630 return err; 610 return err;
631} 611}
632 612
@@ -652,9 +632,8 @@ module_init(cstate_pmu_init);
652 632
653static void __exit cstate_pmu_exit(void) 633static void __exit cstate_pmu_exit(void)
654{ 634{
655 cpu_notifier_register_begin(); 635 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_CSTATE_ONLINE);
656 __unregister_cpu_notifier(&cstate_cpu_nb); 636 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_CSTATE_STARTING);
657 cstate_cleanup(); 637 cstate_cleanup();
658 cpu_notifier_register_done();
659} 638}
660module_exit(cstate_pmu_exit); 639module_exit(cstate_pmu_exit);
diff --git a/arch/x86/events/intel/rapl.c b/arch/x86/events/intel/rapl.c
index d0c58b35155f..28865938aadf 100644
--- a/arch/x86/events/intel/rapl.c
+++ b/arch/x86/events/intel/rapl.c
@@ -556,14 +556,14 @@ const struct attribute_group *rapl_attr_groups[] = {
556 NULL, 556 NULL,
557}; 557};
558 558
559static void rapl_cpu_exit(int cpu) 559static int rapl_cpu_offline(unsigned int cpu)
560{ 560{
561 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu); 561 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
562 int target; 562 int target;
563 563
564 /* Check if exiting cpu is used for collecting rapl events */ 564 /* Check if exiting cpu is used for collecting rapl events */
565 if (!cpumask_test_and_clear_cpu(cpu, &rapl_cpu_mask)) 565 if (!cpumask_test_and_clear_cpu(cpu, &rapl_cpu_mask))
566 return; 566 return 0;
567 567
568 pmu->cpu = -1; 568 pmu->cpu = -1;
569 /* Find a new cpu to collect rapl events */ 569 /* Find a new cpu to collect rapl events */
@@ -575,9 +575,10 @@ static void rapl_cpu_exit(int cpu)
575 pmu->cpu = target; 575 pmu->cpu = target;
576 perf_pmu_migrate_context(pmu->pmu, cpu, target); 576 perf_pmu_migrate_context(pmu->pmu, cpu, target);
577 } 577 }
578 return 0;
578} 579}
579 580
580static void rapl_cpu_init(int cpu) 581static int rapl_cpu_online(unsigned int cpu)
581{ 582{
582 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu); 583 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
583 int target; 584 int target;
@@ -588,13 +589,14 @@ static void rapl_cpu_init(int cpu)
588 */ 589 */
589 target = cpumask_any_and(&rapl_cpu_mask, topology_core_cpumask(cpu)); 590 target = cpumask_any_and(&rapl_cpu_mask, topology_core_cpumask(cpu));
590 if (target < nr_cpu_ids) 591 if (target < nr_cpu_ids)
591 return; 592 return 0;
592 593
593 cpumask_set_cpu(cpu, &rapl_cpu_mask); 594 cpumask_set_cpu(cpu, &rapl_cpu_mask);
594 pmu->cpu = cpu; 595 pmu->cpu = cpu;
596 return 0;
595} 597}
596 598
597static int rapl_cpu_prepare(int cpu) 599static int rapl_cpu_prepare(unsigned int cpu)
598{ 600{
599 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu); 601 struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
600 602
@@ -615,33 +617,6 @@ static int rapl_cpu_prepare(int cpu)
615 return 0; 617 return 0;
616} 618}
617 619
618static int rapl_cpu_notifier(struct notifier_block *self,
619 unsigned long action, void *hcpu)
620{
621 unsigned int cpu = (long)hcpu;
622
623 switch (action & ~CPU_TASKS_FROZEN) {
624 case CPU_UP_PREPARE:
625 rapl_cpu_prepare(cpu);
626 break;
627
628 case CPU_DOWN_FAILED:
629 case CPU_ONLINE:
630 rapl_cpu_init(cpu);
631 break;
632
633 case CPU_DOWN_PREPARE:
634 rapl_cpu_exit(cpu);
635 break;
636 }
637 return NOTIFY_OK;
638}
639
640static struct notifier_block rapl_cpu_nb = {
641 .notifier_call = rapl_cpu_notifier,
642 .priority = CPU_PRI_PERF + 1,
643};
644
645static int rapl_check_hw_unit(bool apply_quirk) 620static int rapl_check_hw_unit(bool apply_quirk)
646{ 621{
647 u64 msr_rapl_power_unit_bits; 622 u64 msr_rapl_power_unit_bits;
@@ -692,24 +667,6 @@ static void __init rapl_advertise(void)
692 } 667 }
693} 668}
694 669
695static int __init rapl_prepare_cpus(void)
696{
697 unsigned int cpu, pkg;
698 int ret;
699
700 for_each_online_cpu(cpu) {
701 pkg = topology_logical_package_id(cpu);
702 if (rapl_pmus->pmus[pkg])
703 continue;
704
705 ret = rapl_cpu_prepare(cpu);
706 if (ret)
707 return ret;
708 rapl_cpu_init(cpu);
709 }
710 return 0;
711}
712
713static void cleanup_rapl_pmus(void) 670static void cleanup_rapl_pmus(void)
714{ 671{
715 int i; 672 int i;
@@ -837,35 +794,44 @@ static int __init rapl_pmu_init(void)
837 if (ret) 794 if (ret)
838 return ret; 795 return ret;
839 796
840 cpu_notifier_register_begin(); 797 /*
798 * Install callbacks. Core will call them for each online cpu.
799 */
841 800
842 ret = rapl_prepare_cpus(); 801 ret = cpuhp_setup_state(CPUHP_PERF_X86_RAPL_PREP, "PERF_X86_RAPL_PREP",
802 rapl_cpu_prepare, NULL);
843 if (ret) 803 if (ret)
844 goto out; 804 goto out;
845 805
806 ret = cpuhp_setup_state(CPUHP_AP_PERF_X86_RAPL_ONLINE,
807 "AP_PERF_X86_RAPL_ONLINE",
808 rapl_cpu_online, rapl_cpu_offline);
809 if (ret)
810 goto out1;
811
846 ret = perf_pmu_register(&rapl_pmus->pmu, "power", -1); 812 ret = perf_pmu_register(&rapl_pmus->pmu, "power", -1);
847 if (ret) 813 if (ret)
848 goto out; 814 goto out2;
849 815
850 __register_cpu_notifier(&rapl_cpu_nb);
851 cpu_notifier_register_done();
852 rapl_advertise(); 816 rapl_advertise();
853 return 0; 817 return 0;
854 818
819out2:
820 cpuhp_remove_state(CPUHP_AP_PERF_X86_RAPL_ONLINE);
821out1:
822 cpuhp_remove_state(CPUHP_PERF_X86_RAPL_PREP);
855out: 823out:
856 pr_warn("Initialization failed (%d), disabled\n", ret); 824 pr_warn("Initialization failed (%d), disabled\n", ret);
857 cleanup_rapl_pmus(); 825 cleanup_rapl_pmus();
858 cpu_notifier_register_done();
859 return ret; 826 return ret;
860} 827}
861module_init(rapl_pmu_init); 828module_init(rapl_pmu_init);
862 829
863static void __exit intel_rapl_exit(void) 830static void __exit intel_rapl_exit(void)
864{ 831{
865 cpu_notifier_register_begin(); 832 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_RAPL_ONLINE);
866 __unregister_cpu_notifier(&rapl_cpu_nb); 833 cpuhp_remove_state_nocalls(CPUHP_PERF_X86_RAPL_PREP);
867 perf_pmu_unregister(&rapl_pmus->pmu); 834 perf_pmu_unregister(&rapl_pmus->pmu);
868 cleanup_rapl_pmus(); 835 cleanup_rapl_pmus();
869 cpu_notifier_register_done();
870} 836}
871module_exit(intel_rapl_exit); 837module_exit(intel_rapl_exit);
diff --git a/arch/x86/events/intel/uncore.c b/arch/x86/events/intel/uncore.c
index 59b4974c697f..3f3d0d67749b 100644
--- a/arch/x86/events/intel/uncore.c
+++ b/arch/x86/events/intel/uncore.c
@@ -1052,7 +1052,7 @@ static void uncore_pci_exit(void)
1052 } 1052 }
1053} 1053}
1054 1054
1055static void uncore_cpu_dying(int cpu) 1055static int uncore_cpu_dying(unsigned int cpu)
1056{ 1056{
1057 struct intel_uncore_type *type, **types = uncore_msr_uncores; 1057 struct intel_uncore_type *type, **types = uncore_msr_uncores;
1058 struct intel_uncore_pmu *pmu; 1058 struct intel_uncore_pmu *pmu;
@@ -1069,16 +1069,19 @@ static void uncore_cpu_dying(int cpu)
1069 uncore_box_exit(box); 1069 uncore_box_exit(box);
1070 } 1070 }
1071 } 1071 }
1072 return 0;
1072} 1073}
1073 1074
1074static void uncore_cpu_starting(int cpu, bool init) 1075static int first_init;
1076
1077static int uncore_cpu_starting(unsigned int cpu)
1075{ 1078{
1076 struct intel_uncore_type *type, **types = uncore_msr_uncores; 1079 struct intel_uncore_type *type, **types = uncore_msr_uncores;
1077 struct intel_uncore_pmu *pmu; 1080 struct intel_uncore_pmu *pmu;
1078 struct intel_uncore_box *box; 1081 struct intel_uncore_box *box;
1079 int i, pkg, ncpus = 1; 1082 int i, pkg, ncpus = 1;
1080 1083
1081 if (init) { 1084 if (first_init) {
1082 /* 1085 /*
1083 * On init we get the number of online cpus in the package 1086 * On init we get the number of online cpus in the package
1084 * and set refcount for all of them. 1087 * and set refcount for all of them.
@@ -1099,9 +1102,11 @@ static void uncore_cpu_starting(int cpu, bool init)
1099 uncore_box_init(box); 1102 uncore_box_init(box);
1100 } 1103 }
1101 } 1104 }
1105
1106 return 0;
1102} 1107}
1103 1108
1104static int uncore_cpu_prepare(int cpu) 1109static int uncore_cpu_prepare(unsigned int cpu)
1105{ 1110{
1106 struct intel_uncore_type *type, **types = uncore_msr_uncores; 1111 struct intel_uncore_type *type, **types = uncore_msr_uncores;
1107 struct intel_uncore_pmu *pmu; 1112 struct intel_uncore_pmu *pmu;
@@ -1164,13 +1169,13 @@ static void uncore_change_context(struct intel_uncore_type **uncores,
1164 uncore_change_type_ctx(*uncores, old_cpu, new_cpu); 1169 uncore_change_type_ctx(*uncores, old_cpu, new_cpu);
1165} 1170}
1166 1171
1167static void uncore_event_exit_cpu(int cpu) 1172static int uncore_event_cpu_offline(unsigned int cpu)
1168{ 1173{
1169 int target; 1174 int target;
1170 1175
1171 /* Check if exiting cpu is used for collecting uncore events */ 1176 /* Check if exiting cpu is used for collecting uncore events */
1172 if (!cpumask_test_and_clear_cpu(cpu, &uncore_cpu_mask)) 1177 if (!cpumask_test_and_clear_cpu(cpu, &uncore_cpu_mask))
1173 return; 1178 return 0;
1174 1179
1175 /* Find a new cpu to collect uncore events */ 1180 /* Find a new cpu to collect uncore events */
1176 target = cpumask_any_but(topology_core_cpumask(cpu), cpu); 1181 target = cpumask_any_but(topology_core_cpumask(cpu), cpu);
@@ -1183,9 +1188,10 @@ static void uncore_event_exit_cpu(int cpu)
1183 1188
1184 uncore_change_context(uncore_msr_uncores, cpu, target); 1189 uncore_change_context(uncore_msr_uncores, cpu, target);
1185 uncore_change_context(uncore_pci_uncores, cpu, target); 1190 uncore_change_context(uncore_pci_uncores, cpu, target);
1191 return 0;
1186} 1192}
1187 1193
1188static void uncore_event_init_cpu(int cpu) 1194static int uncore_event_cpu_online(unsigned int cpu)
1189{ 1195{
1190 int target; 1196 int target;
1191 1197
@@ -1195,50 +1201,15 @@ static void uncore_event_init_cpu(int cpu)
1195 */ 1201 */
1196 target = cpumask_any_and(&uncore_cpu_mask, topology_core_cpumask(cpu)); 1202 target = cpumask_any_and(&uncore_cpu_mask, topology_core_cpumask(cpu));
1197 if (target < nr_cpu_ids) 1203 if (target < nr_cpu_ids)
1198 return; 1204 return 0;
1199 1205
1200 cpumask_set_cpu(cpu, &uncore_cpu_mask); 1206 cpumask_set_cpu(cpu, &uncore_cpu_mask);
1201 1207
1202 uncore_change_context(uncore_msr_uncores, -1, cpu); 1208 uncore_change_context(uncore_msr_uncores, -1, cpu);
1203 uncore_change_context(uncore_pci_uncores, -1, cpu); 1209 uncore_change_context(uncore_pci_uncores, -1, cpu);
1210 return 0;
1204} 1211}
1205 1212
1206static int uncore_cpu_notifier(struct notifier_block *self,
1207 unsigned long action, void *hcpu)
1208{
1209 unsigned int cpu = (long)hcpu;
1210
1211 switch (action & ~CPU_TASKS_FROZEN) {
1212 case CPU_UP_PREPARE:
1213 return notifier_from_errno(uncore_cpu_prepare(cpu));
1214
1215 case CPU_STARTING:
1216 uncore_cpu_starting(cpu, false);
1217 case CPU_DOWN_FAILED:
1218 uncore_event_init_cpu(cpu);
1219 break;
1220
1221 case CPU_UP_CANCELED:
1222 case CPU_DYING:
1223 uncore_cpu_dying(cpu);
1224 break;
1225
1226 case CPU_DOWN_PREPARE:
1227 uncore_event_exit_cpu(cpu);
1228 break;
1229 }
1230 return NOTIFY_OK;
1231}
1232
1233static struct notifier_block uncore_cpu_nb = {
1234 .notifier_call = uncore_cpu_notifier,
1235 /*
1236 * to migrate uncore events, our notifier should be executed
1237 * before perf core's notifier.
1238 */
1239 .priority = CPU_PRI_PERF + 1,
1240};
1241
1242static int __init type_pmu_register(struct intel_uncore_type *type) 1213static int __init type_pmu_register(struct intel_uncore_type *type)
1243{ 1214{
1244 int i, ret; 1215 int i, ret;
@@ -1282,41 +1253,6 @@ err:
1282 return ret; 1253 return ret;
1283} 1254}
1284 1255
1285static void __init uncore_cpu_setup(void *dummy)
1286{
1287 uncore_cpu_starting(smp_processor_id(), true);
1288}
1289
1290/* Lazy to avoid allocation of a few bytes for the normal case */
1291static __initdata DECLARE_BITMAP(packages, MAX_LOCAL_APIC);
1292
1293static int __init uncore_cpumask_init(bool msr)
1294{
1295 unsigned int cpu;
1296
1297 for_each_online_cpu(cpu) {
1298 unsigned int pkg = topology_logical_package_id(cpu);
1299 int ret;
1300
1301 if (test_and_set_bit(pkg, packages))
1302 continue;
1303 /*
1304 * The first online cpu of each package allocates and takes
1305 * the refcounts for all other online cpus in that package.
1306 * If msrs are not enabled no allocation is required.
1307 */
1308 if (msr) {
1309 ret = uncore_cpu_prepare(cpu);
1310 if (ret)
1311 return ret;
1312 }
1313 uncore_event_init_cpu(cpu);
1314 smp_call_function_single(cpu, uncore_cpu_setup, NULL, 1);
1315 }
1316 __register_cpu_notifier(&uncore_cpu_nb);
1317 return 0;
1318}
1319
1320#define X86_UNCORE_MODEL_MATCH(model, init) \ 1256#define X86_UNCORE_MODEL_MATCH(model, init) \
1321 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&init } 1257 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&init }
1322 1258
@@ -1440,11 +1376,33 @@ static int __init intel_uncore_init(void)
1440 if (cret && pret) 1376 if (cret && pret)
1441 return -ENODEV; 1377 return -ENODEV;
1442 1378
1443 cpu_notifier_register_begin(); 1379 /*
1444 ret = uncore_cpumask_init(!cret); 1380 * Install callbacks. Core will call them for each online cpu.
1445 if (ret) 1381 *
1446 goto err; 1382 * The first online cpu of each package allocates and takes
1447 cpu_notifier_register_done(); 1383 * the refcounts for all other online cpus in that package.
1384 * If msrs are not enabled no allocation is required and
1385 * uncore_cpu_prepare() is not called for each online cpu.
1386 */
1387 if (!cret) {
1388 ret = cpuhp_setup_state(CPUHP_PERF_X86_UNCORE_PREP,
1389 "PERF_X86_UNCORE_PREP",
1390 uncore_cpu_prepare, NULL);
1391 if (ret)
1392 goto err;
1393 } else {
1394 cpuhp_setup_state_nocalls(CPUHP_PERF_X86_UNCORE_PREP,
1395 "PERF_X86_UNCORE_PREP",
1396 uncore_cpu_prepare, NULL);
1397 }
1398 first_init = 1;
1399 cpuhp_setup_state(CPUHP_AP_PERF_X86_UNCORE_STARTING,
1400 "AP_PERF_X86_UNCORE_STARTING",
1401 uncore_cpu_starting, uncore_cpu_dying);
1402 first_init = 0;
1403 cpuhp_setup_state(CPUHP_AP_PERF_X86_UNCORE_ONLINE,
1404 "AP_PERF_X86_UNCORE_ONLINE",
1405 uncore_event_cpu_online, uncore_event_cpu_offline);
1448 return 0; 1406 return 0;
1449 1407
1450err: 1408err:
@@ -1452,17 +1410,16 @@ err:
1452 on_each_cpu_mask(&uncore_cpu_mask, uncore_exit_boxes, NULL, 1); 1410 on_each_cpu_mask(&uncore_cpu_mask, uncore_exit_boxes, NULL, 1);
1453 uncore_types_exit(uncore_msr_uncores); 1411 uncore_types_exit(uncore_msr_uncores);
1454 uncore_pci_exit(); 1412 uncore_pci_exit();
1455 cpu_notifier_register_done();
1456 return ret; 1413 return ret;
1457} 1414}
1458module_init(intel_uncore_init); 1415module_init(intel_uncore_init);
1459 1416
1460static void __exit intel_uncore_exit(void) 1417static void __exit intel_uncore_exit(void)
1461{ 1418{
1462 cpu_notifier_register_begin(); 1419 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_UNCORE_ONLINE);
1463 __unregister_cpu_notifier(&uncore_cpu_nb); 1420 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_UNCORE_STARTING);
1421 cpuhp_remove_state_nocalls(CPUHP_PERF_X86_UNCORE_PREP);
1464 uncore_types_exit(uncore_msr_uncores); 1422 uncore_types_exit(uncore_msr_uncores);
1465 uncore_pci_exit(); 1423 uncore_pci_exit();
1466 cpu_notifier_register_done();
1467} 1424}
1468module_exit(intel_uncore_exit); 1425module_exit(intel_uncore_exit);
diff --git a/arch/x86/kernel/apb_timer.c b/arch/x86/kernel/apb_timer.c
index cefacbad1531..456316f6c868 100644
--- a/arch/x86/kernel/apb_timer.c
+++ b/arch/x86/kernel/apb_timer.c
@@ -215,26 +215,18 @@ void apbt_setup_secondary_clock(void)
215 * cpu timers during the offline process due to the ordering of notification. 215 * cpu timers during the offline process due to the ordering of notification.
216 * the extra interrupt is harmless. 216 * the extra interrupt is harmless.
217 */ 217 */
218static int apbt_cpuhp_notify(struct notifier_block *n, 218static int apbt_cpu_dead(unsigned int cpu)
219 unsigned long action, void *hcpu)
220{ 219{
221 unsigned long cpu = (unsigned long)hcpu;
222 struct apbt_dev *adev = &per_cpu(cpu_apbt_dev, cpu); 220 struct apbt_dev *adev = &per_cpu(cpu_apbt_dev, cpu);
223 221
224 switch (action & ~CPU_TASKS_FROZEN) { 222 dw_apb_clockevent_pause(adev->timer);
225 case CPU_DEAD: 223 if (system_state == SYSTEM_RUNNING) {
226 dw_apb_clockevent_pause(adev->timer); 224 pr_debug("skipping APBT CPU %u offline\n", cpu);
227 if (system_state == SYSTEM_RUNNING) { 225 } else {
228 pr_debug("skipping APBT CPU %lu offline\n", cpu); 226 pr_debug("APBT clockevent for cpu %u offline\n", cpu);
229 } else { 227 dw_apb_clockevent_stop(adev->timer);
230 pr_debug("APBT clockevent for cpu %lu offline\n", cpu);
231 dw_apb_clockevent_stop(adev->timer);
232 }
233 break;
234 default:
235 pr_debug("APBT notified %lu, no action\n", action);
236 } 228 }
237 return NOTIFY_OK; 229 return 0;
238} 230}
239 231
240static __init int apbt_late_init(void) 232static __init int apbt_late_init(void)
@@ -242,9 +234,8 @@ static __init int apbt_late_init(void)
242 if (intel_mid_timer_options == INTEL_MID_TIMER_LAPIC_APBT || 234 if (intel_mid_timer_options == INTEL_MID_TIMER_LAPIC_APBT ||
243 !apb_timer_block_enabled) 235 !apb_timer_block_enabled)
244 return 0; 236 return 0;
245 /* This notifier should be called after workqueue is ready */ 237 return cpuhp_setup_state(CPUHP_X86_APB_DEAD, "X86_APB_DEAD", NULL,
246 hotcpu_notifier(apbt_cpuhp_notify, -20); 238 apbt_cpu_dead);
247 return 0;
248} 239}
249fs_initcall(apbt_late_init); 240fs_initcall(apbt_late_init);
250#else 241#else
diff --git a/arch/x86/kernel/apic/x2apic_cluster.c b/arch/x86/kernel/apic/x2apic_cluster.c
index 24170d0809ba..6368fa69d2af 100644
--- a/arch/x86/kernel/apic/x2apic_cluster.c
+++ b/arch/x86/kernel/apic/x2apic_cluster.c
@@ -152,68 +152,48 @@ static void init_x2apic_ldr(void)
152 } 152 }
153} 153}
154 154
155 /* 155/*
156 * At CPU state changes, update the x2apic cluster sibling info. 156 * At CPU state changes, update the x2apic cluster sibling info.
157 */ 157 */
158static int 158int x2apic_prepare_cpu(unsigned int cpu)
159update_clusterinfo(struct notifier_block *nfb, unsigned long action, void *hcpu)
160{ 159{
161 unsigned int this_cpu = (unsigned long)hcpu; 160 if (!zalloc_cpumask_var(&per_cpu(cpus_in_cluster, cpu), GFP_KERNEL))
162 unsigned int cpu; 161 return -ENOMEM;
163 int err = 0; 162
164 163 if (!zalloc_cpumask_var(&per_cpu(ipi_mask, cpu), GFP_KERNEL)) {
165 switch (action) { 164 free_cpumask_var(per_cpu(cpus_in_cluster, cpu));
166 case CPU_UP_PREPARE: 165 return -ENOMEM;
167 if (!zalloc_cpumask_var(&per_cpu(cpus_in_cluster, this_cpu),
168 GFP_KERNEL)) {
169 err = -ENOMEM;
170 } else if (!zalloc_cpumask_var(&per_cpu(ipi_mask, this_cpu),
171 GFP_KERNEL)) {
172 free_cpumask_var(per_cpu(cpus_in_cluster, this_cpu));
173 err = -ENOMEM;
174 }
175 break;
176 case CPU_UP_CANCELED:
177 case CPU_UP_CANCELED_FROZEN:
178 case CPU_DEAD:
179 for_each_online_cpu(cpu) {
180 if (x2apic_cluster(this_cpu) != x2apic_cluster(cpu))
181 continue;
182 cpumask_clear_cpu(this_cpu, per_cpu(cpus_in_cluster, cpu));
183 cpumask_clear_cpu(cpu, per_cpu(cpus_in_cluster, this_cpu));
184 }
185 free_cpumask_var(per_cpu(cpus_in_cluster, this_cpu));
186 free_cpumask_var(per_cpu(ipi_mask, this_cpu));
187 break;
188 } 166 }
189 167
190 return notifier_from_errno(err); 168 return 0;
191} 169}
192 170
193static struct notifier_block x2apic_cpu_notifier = { 171int x2apic_dead_cpu(unsigned int this_cpu)
194 .notifier_call = update_clusterinfo,
195};
196
197static int x2apic_init_cpu_notifier(void)
198{ 172{
199 int cpu = smp_processor_id(); 173 int cpu;
200
201 zalloc_cpumask_var(&per_cpu(cpus_in_cluster, cpu), GFP_KERNEL);
202 zalloc_cpumask_var(&per_cpu(ipi_mask, cpu), GFP_KERNEL);
203 174
204 BUG_ON(!per_cpu(cpus_in_cluster, cpu) || !per_cpu(ipi_mask, cpu)); 175 for_each_online_cpu(cpu) {
205 176 if (x2apic_cluster(this_cpu) != x2apic_cluster(cpu))
206 cpumask_set_cpu(cpu, per_cpu(cpus_in_cluster, cpu)); 177 continue;
207 register_hotcpu_notifier(&x2apic_cpu_notifier); 178 cpumask_clear_cpu(this_cpu, per_cpu(cpus_in_cluster, cpu));
208 return 1; 179 cpumask_clear_cpu(cpu, per_cpu(cpus_in_cluster, this_cpu));
180 }
181 free_cpumask_var(per_cpu(cpus_in_cluster, this_cpu));
182 free_cpumask_var(per_cpu(ipi_mask, this_cpu));
183 return 0;
209} 184}
210 185
211static int x2apic_cluster_probe(void) 186static int x2apic_cluster_probe(void)
212{ 187{
213 if (x2apic_mode) 188 int cpu = smp_processor_id();
214 return x2apic_init_cpu_notifier(); 189
215 else 190 if (!x2apic_mode)
216 return 0; 191 return 0;
192
193 cpumask_set_cpu(cpu, per_cpu(cpus_in_cluster, cpu));
194 cpuhp_setup_state(CPUHP_X2APIC_PREPARE, "X2APIC_PREPARE",
195 x2apic_prepare_cpu, x2apic_dead_cpu);
196 return 1;
217} 197}
218 198
219static const struct cpumask *x2apic_cluster_target_cpus(void) 199static const struct cpumask *x2apic_cluster_target_cpus(void)
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index f112af7aa62e..3d747070fe67 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -710,31 +710,29 @@ static void hpet_work(struct work_struct *w)
710 complete(&hpet_work->complete); 710 complete(&hpet_work->complete);
711} 711}
712 712
713static int hpet_cpuhp_notify(struct notifier_block *n, 713static int hpet_cpuhp_online(unsigned int cpu)
714 unsigned long action, void *hcpu)
715{ 714{
716 unsigned long cpu = (unsigned long)hcpu;
717 struct hpet_work_struct work; 715 struct hpet_work_struct work;
716
717 INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
718 init_completion(&work.complete);
719 /* FIXME: add schedule_work_on() */
720 schedule_delayed_work_on(cpu, &work.work, 0);
721 wait_for_completion(&work.complete);
722 destroy_delayed_work_on_stack(&work.work);
723 return 0;
724}
725
726static int hpet_cpuhp_dead(unsigned int cpu)
727{
718 struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu); 728 struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
719 729
720 switch (action & ~CPU_TASKS_FROZEN) { 730 if (!hdev)
721 case CPU_ONLINE: 731 return 0;
722 INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work); 732 free_irq(hdev->irq, hdev);
723 init_completion(&work.complete); 733 hdev->flags &= ~HPET_DEV_USED;
724 /* FIXME: add schedule_work_on() */ 734 per_cpu(cpu_hpet_dev, cpu) = NULL;
725 schedule_delayed_work_on(cpu, &work.work, 0); 735 return 0;
726 wait_for_completion(&work.complete);
727 destroy_delayed_work_on_stack(&work.work);
728 break;
729 case CPU_DEAD:
730 if (hdev) {
731 free_irq(hdev->irq, hdev);
732 hdev->flags &= ~HPET_DEV_USED;
733 per_cpu(cpu_hpet_dev, cpu) = NULL;
734 }
735 break;
736 }
737 return NOTIFY_OK;
738} 736}
739#else 737#else
740 738
@@ -750,11 +748,8 @@ static void hpet_reserve_msi_timers(struct hpet_data *hd)
750} 748}
751#endif 749#endif
752 750
753static int hpet_cpuhp_notify(struct notifier_block *n, 751#define hpet_cpuhp_online NULL
754 unsigned long action, void *hcpu) 752#define hpet_cpuhp_dead NULL
755{
756 return NOTIFY_OK;
757}
758 753
759#endif 754#endif
760 755
@@ -931,7 +926,7 @@ out_nohpet:
931 */ 926 */
932static __init int hpet_late_init(void) 927static __init int hpet_late_init(void)
933{ 928{
934 int cpu; 929 int ret;
935 930
936 if (boot_hpet_disable) 931 if (boot_hpet_disable)
937 return -ENODEV; 932 return -ENODEV;
@@ -961,16 +956,20 @@ static __init int hpet_late_init(void)
961 if (boot_cpu_has(X86_FEATURE_ARAT)) 956 if (boot_cpu_has(X86_FEATURE_ARAT))
962 return 0; 957 return 0;
963 958
964 cpu_notifier_register_begin();
965 for_each_online_cpu(cpu) {
966 hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
967 }
968
969 /* This notifier should be called after workqueue is ready */ 959 /* This notifier should be called after workqueue is ready */
970 __hotcpu_notifier(hpet_cpuhp_notify, -20); 960 ret = cpuhp_setup_state(CPUHP_AP_X86_HPET_ONLINE, "AP_X86_HPET_ONLINE",
971 cpu_notifier_register_done(); 961 hpet_cpuhp_online, NULL);
972 962 if (ret)
963 return ret;
964 ret = cpuhp_setup_state(CPUHP_X86_HPET_DEAD, "X86_HPET_DEAD", NULL,
965 hpet_cpuhp_dead);
966 if (ret)
967 goto err_cpuhp;
973 return 0; 968 return 0;
969
970err_cpuhp:
971 cpuhp_remove_state(CPUHP_AP_X86_HPET_ONLINE);
972 return ret;
974} 973}
975fs_initcall(hpet_late_init); 974fs_initcall(hpet_late_init);
976 975
diff --git a/arch/x86/kernel/tboot.c b/arch/x86/kernel/tboot.c
index 9b0185fbe3eb..654f6c66fe45 100644
--- a/arch/x86/kernel/tboot.c
+++ b/arch/x86/kernel/tboot.c
@@ -323,25 +323,16 @@ static int tboot_wait_for_aps(int num_aps)
323 return !(atomic_read((atomic_t *)&tboot->num_in_wfs) == num_aps); 323 return !(atomic_read((atomic_t *)&tboot->num_in_wfs) == num_aps);
324} 324}
325 325
326static int tboot_cpu_callback(struct notifier_block *nfb, unsigned long action, 326static int tboot_dying_cpu(unsigned int cpu)
327 void *hcpu)
328{ 327{
329 switch (action) { 328 atomic_inc(&ap_wfs_count);
330 case CPU_DYING: 329 if (num_online_cpus() == 1) {
331 atomic_inc(&ap_wfs_count); 330 if (tboot_wait_for_aps(atomic_read(&ap_wfs_count)))
332 if (num_online_cpus() == 1) 331 return -EBUSY;
333 if (tboot_wait_for_aps(atomic_read(&ap_wfs_count)))
334 return NOTIFY_BAD;
335 break;
336 } 332 }
337 return NOTIFY_OK; 333 return 0;
338} 334}
339 335
340static struct notifier_block tboot_cpu_notifier =
341{
342 .notifier_call = tboot_cpu_callback,
343};
344
345#ifdef CONFIG_DEBUG_FS 336#ifdef CONFIG_DEBUG_FS
346 337
347#define TBOOT_LOG_UUID { 0x26, 0x25, 0x19, 0xc0, 0x30, 0x6b, 0xb4, 0x4d, \ 338#define TBOOT_LOG_UUID { 0x26, 0x25, 0x19, 0xc0, 0x30, 0x6b, 0xb4, 0x4d, \
@@ -417,8 +408,8 @@ static __init int tboot_late_init(void)
417 tboot_create_trampoline(); 408 tboot_create_trampoline();
418 409
419 atomic_set(&ap_wfs_count, 0); 410 atomic_set(&ap_wfs_count, 0);
420 register_hotcpu_notifier(&tboot_cpu_notifier); 411 cpuhp_setup_state(CPUHP_AP_X86_TBOOT_DYING, "AP_X86_TBOOT_DYING", NULL,
421 412 tboot_dying_cpu);
422#ifdef CONFIG_DEBUG_FS 413#ifdef CONFIG_DEBUG_FS
423 debugfs_create_file("tboot_log", S_IRUSR, 414 debugfs_create_file("tboot_log", S_IRUSR,
424 arch_debugfs_dir, NULL, &tboot_log_fops); 415 arch_debugfs_dir, NULL, &tboot_log_fops);
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index b2766723c951..45608a7da9b3 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -5552,9 +5552,10 @@ int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5552} 5552}
5553EXPORT_SYMBOL_GPL(kvm_fast_pio_out); 5553EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5554 5554
5555static void tsc_bad(void *info) 5555static int kvmclock_cpu_down_prep(unsigned int cpu)
5556{ 5556{
5557 __this_cpu_write(cpu_tsc_khz, 0); 5557 __this_cpu_write(cpu_tsc_khz, 0);
5558 return 0;
5558} 5559}
5559 5560
5560static void tsc_khz_changed(void *data) 5561static void tsc_khz_changed(void *data)
@@ -5659,35 +5660,18 @@ static struct notifier_block kvmclock_cpufreq_notifier_block = {
5659 .notifier_call = kvmclock_cpufreq_notifier 5660 .notifier_call = kvmclock_cpufreq_notifier
5660}; 5661};
5661 5662
5662static int kvmclock_cpu_notifier(struct notifier_block *nfb, 5663static int kvmclock_cpu_online(unsigned int cpu)
5663 unsigned long action, void *hcpu)
5664{ 5664{
5665 unsigned int cpu = (unsigned long)hcpu; 5665 tsc_khz_changed(NULL);
5666 5666 return 0;
5667 switch (action) {
5668 case CPU_ONLINE:
5669 case CPU_DOWN_FAILED:
5670 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5671 break;
5672 case CPU_DOWN_PREPARE:
5673 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5674 break;
5675 }
5676 return NOTIFY_OK;
5677} 5667}
5678 5668
5679static struct notifier_block kvmclock_cpu_notifier_block = {
5680 .notifier_call = kvmclock_cpu_notifier,
5681 .priority = -INT_MAX
5682};
5683
5684static void kvm_timer_init(void) 5669static void kvm_timer_init(void)
5685{ 5670{
5686 int cpu; 5671 int cpu;
5687 5672
5688 max_tsc_khz = tsc_khz; 5673 max_tsc_khz = tsc_khz;
5689 5674
5690 cpu_notifier_register_begin();
5691 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { 5675 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5692#ifdef CONFIG_CPU_FREQ 5676#ifdef CONFIG_CPU_FREQ
5693 struct cpufreq_policy policy; 5677 struct cpufreq_policy policy;
@@ -5702,12 +5686,9 @@ static void kvm_timer_init(void)
5702 CPUFREQ_TRANSITION_NOTIFIER); 5686 CPUFREQ_TRANSITION_NOTIFIER);
5703 } 5687 }
5704 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz); 5688 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5705 for_each_online_cpu(cpu)
5706 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5707
5708 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5709 cpu_notifier_register_done();
5710 5689
5690 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "AP_X86_KVM_CLK_ONLINE",
5691 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5711} 5692}
5712 5693
5713static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); 5694static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
@@ -5896,7 +5877,7 @@ void kvm_arch_exit(void)
5896 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) 5877 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5897 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, 5878 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5898 CPUFREQ_TRANSITION_NOTIFIER); 5879 CPUFREQ_TRANSITION_NOTIFIER);
5899 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block); 5880 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
5900#ifdef CONFIG_X86_64 5881#ifdef CONFIG_X86_64
5901 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); 5882 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5902#endif 5883#endif
diff --git a/arch/xtensa/kernel/perf_event.c b/arch/xtensa/kernel/perf_event.c
index ef90479e0397..0fecc8a2c0b5 100644
--- a/arch/xtensa/kernel/perf_event.c
+++ b/arch/xtensa/kernel/perf_event.c
@@ -404,7 +404,7 @@ static struct pmu xtensa_pmu = {
404 .read = xtensa_pmu_read, 404 .read = xtensa_pmu_read,
405}; 405};
406 406
407static void xtensa_pmu_setup(void) 407static int xtensa_pmu_setup(int cpu)
408{ 408{
409 unsigned i; 409 unsigned i;
410 410
@@ -413,21 +413,7 @@ static void xtensa_pmu_setup(void)
413 set_er(0, XTENSA_PMU_PMCTRL(i)); 413 set_er(0, XTENSA_PMU_PMCTRL(i));
414 set_er(get_er(XTENSA_PMU_PMSTAT(i)), XTENSA_PMU_PMSTAT(i)); 414 set_er(get_er(XTENSA_PMU_PMSTAT(i)), XTENSA_PMU_PMSTAT(i));
415 } 415 }
416} 416 return 0;
417
418static int xtensa_pmu_notifier(struct notifier_block *self,
419 unsigned long action, void *data)
420{
421 switch (action & ~CPU_TASKS_FROZEN) {
422 case CPU_STARTING:
423 xtensa_pmu_setup();
424 break;
425
426 default:
427 break;
428 }
429
430 return NOTIFY_OK;
431} 417}
432 418
433static int __init xtensa_pmu_init(void) 419static int __init xtensa_pmu_init(void)
@@ -435,7 +421,13 @@ static int __init xtensa_pmu_init(void)
435 int ret; 421 int ret;
436 int irq = irq_create_mapping(NULL, XCHAL_PROFILING_INTERRUPT); 422 int irq = irq_create_mapping(NULL, XCHAL_PROFILING_INTERRUPT);
437 423
438 perf_cpu_notifier(xtensa_pmu_notifier); 424 ret = cpuhp_setup_state(CPUHP_AP_PERF_XTENSA_STARTING,
425 "AP_PERF_XTENSA_STARTING", xtensa_pmu_setup,
426 NULL);
427 if (ret) {
428 pr_err("xtensa_pmu: failed to register CPU-hotplug.\n");
429 return ret;
430 }
439#if XTENSA_FAKE_NMI 431#if XTENSA_FAKE_NMI
440 enable_irq(irq); 432 enable_irq(irq);
441#else 433#else
diff --git a/drivers/acpi/processor_driver.c b/drivers/acpi/processor_driver.c
index 0ca14ac7bb28..0553aeebb228 100644
--- a/drivers/acpi/processor_driver.c
+++ b/drivers/acpi/processor_driver.c
@@ -118,12 +118,13 @@ static int acpi_cpu_soft_notify(struct notifier_block *nfb,
118 struct acpi_device *device; 118 struct acpi_device *device;
119 action &= ~CPU_TASKS_FROZEN; 119 action &= ~CPU_TASKS_FROZEN;
120 120
121 /* 121 switch (action) {
122 * CPU_STARTING and CPU_DYING must not sleep. Return here since 122 case CPU_ONLINE:
123 * acpi_bus_get_device() may sleep. 123 case CPU_DEAD:
124 */ 124 break;
125 if (action == CPU_STARTING || action == CPU_DYING) 125 default:
126 return NOTIFY_DONE; 126 return NOTIFY_DONE;
127 }
127 128
128 if (!pr || acpi_bus_get_device(pr->handle, &device)) 129 if (!pr || acpi_bus_get_device(pr->handle, &device))
129 return NOTIFY_DONE; 130 return NOTIFY_DONE;
diff --git a/drivers/bus/arm-cci.c b/drivers/bus/arm-cci.c
index a49b28378d59..5755907f836f 100644
--- a/drivers/bus/arm-cci.c
+++ b/drivers/bus/arm-cci.c
@@ -144,12 +144,15 @@ struct cci_pmu {
144 int num_cntrs; 144 int num_cntrs;
145 atomic_t active_events; 145 atomic_t active_events;
146 struct mutex reserve_mutex; 146 struct mutex reserve_mutex;
147 struct notifier_block cpu_nb; 147 struct list_head entry;
148 cpumask_t cpus; 148 cpumask_t cpus;
149}; 149};
150 150
151#define to_cci_pmu(c) (container_of(c, struct cci_pmu, pmu)) 151#define to_cci_pmu(c) (container_of(c, struct cci_pmu, pmu))
152 152
153static DEFINE_MUTEX(cci_pmu_mutex);
154static LIST_HEAD(cci_pmu_list);
155
153enum cci_models { 156enum cci_models {
154#ifdef CONFIG_ARM_CCI400_PMU 157#ifdef CONFIG_ARM_CCI400_PMU
155 CCI400_R0, 158 CCI400_R0,
@@ -1503,31 +1506,26 @@ static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
1503 return perf_pmu_register(&cci_pmu->pmu, name, -1); 1506 return perf_pmu_register(&cci_pmu->pmu, name, -1);
1504} 1507}
1505 1508
1506static int cci_pmu_cpu_notifier(struct notifier_block *self, 1509static int cci_pmu_offline_cpu(unsigned int cpu)
1507 unsigned long action, void *hcpu)
1508{ 1510{
1509 struct cci_pmu *cci_pmu = container_of(self, 1511 struct cci_pmu *cci_pmu;
1510 struct cci_pmu, cpu_nb);
1511 unsigned int cpu = (long)hcpu;
1512 unsigned int target; 1512 unsigned int target;
1513 1513
1514 switch (action & ~CPU_TASKS_FROZEN) { 1514 mutex_lock(&cci_pmu_mutex);
1515 case CPU_DOWN_PREPARE: 1515 list_for_each_entry(cci_pmu, &cci_pmu_list, entry) {
1516 if (!cpumask_test_and_clear_cpu(cpu, &cci_pmu->cpus)) 1516 if (!cpumask_test_and_clear_cpu(cpu, &cci_pmu->cpus))
1517 break; 1517 continue;
1518 target = cpumask_any_but(cpu_online_mask, cpu); 1518 target = cpumask_any_but(cpu_online_mask, cpu);
1519 if (target >= nr_cpu_ids) // UP, last CPU 1519 if (target >= nr_cpu_ids)
1520 break; 1520 continue;
1521 /* 1521 /*
1522 * TODO: migrate context once core races on event->ctx have 1522 * TODO: migrate context once core races on event->ctx have
1523 * been fixed. 1523 * been fixed.
1524 */ 1524 */
1525 cpumask_set_cpu(target, &cci_pmu->cpus); 1525 cpumask_set_cpu(target, &cci_pmu->cpus);
1526 default:
1527 break;
1528 } 1526 }
1529 1527 mutex_unlock(&cci_pmu_mutex);
1530 return NOTIFY_OK; 1528 return 0;
1531} 1529}
1532 1530
1533static struct cci_pmu_model cci_pmu_models[] = { 1531static struct cci_pmu_model cci_pmu_models[] = {
@@ -1766,24 +1764,13 @@ static int cci_pmu_probe(struct platform_device *pdev)
1766 atomic_set(&cci_pmu->active_events, 0); 1764 atomic_set(&cci_pmu->active_events, 0);
1767 cpumask_set_cpu(smp_processor_id(), &cci_pmu->cpus); 1765 cpumask_set_cpu(smp_processor_id(), &cci_pmu->cpus);
1768 1766
1769 cci_pmu->cpu_nb = (struct notifier_block) { 1767 ret = cci_pmu_init(cci_pmu, pdev);
1770 .notifier_call = cci_pmu_cpu_notifier,
1771 /*
1772 * to migrate uncore events, our notifier should be executed
1773 * before perf core's notifier.
1774 */
1775 .priority = CPU_PRI_PERF + 1,
1776 };
1777
1778 ret = register_cpu_notifier(&cci_pmu->cpu_nb);
1779 if (ret) 1768 if (ret)
1780 return ret; 1769 return ret;
1781 1770
1782 ret = cci_pmu_init(cci_pmu, pdev); 1771 mutex_lock(&cci_pmu_mutex);
1783 if (ret) { 1772 list_add(&cci_pmu->entry, &cci_pmu_list);
1784 unregister_cpu_notifier(&cci_pmu->cpu_nb); 1773 mutex_unlock(&cci_pmu_mutex);
1785 return ret;
1786 }
1787 1774
1788 pr_info("ARM %s PMU driver probed", cci_pmu->model->name); 1775 pr_info("ARM %s PMU driver probed", cci_pmu->model->name);
1789 return 0; 1776 return 0;
@@ -1817,6 +1804,12 @@ static int __init cci_platform_init(void)
1817{ 1804{
1818 int ret; 1805 int ret;
1819 1806
1807 ret = cpuhp_setup_state_nocalls(CPUHP_AP_PERF_ARM_CCI_ONLINE,
1808 "AP_PERF_ARM_CCI_ONLINE", NULL,
1809 cci_pmu_offline_cpu);
1810 if (ret)
1811 return ret;
1812
1820 ret = platform_driver_register(&cci_pmu_driver); 1813 ret = platform_driver_register(&cci_pmu_driver);
1821 if (ret) 1814 if (ret)
1822 return ret; 1815 return ret;
diff --git a/drivers/bus/arm-ccn.c b/drivers/bus/arm-ccn.c
index acc3eb542c74..97a9185af433 100644
--- a/drivers/bus/arm-ccn.c
+++ b/drivers/bus/arm-ccn.c
@@ -167,7 +167,7 @@ struct arm_ccn_dt {
167 struct hrtimer hrtimer; 167 struct hrtimer hrtimer;
168 168
169 cpumask_t cpu; 169 cpumask_t cpu;
170 struct notifier_block cpu_nb; 170 struct list_head entry;
171 171
172 struct pmu pmu; 172 struct pmu pmu;
173}; 173};
@@ -189,6 +189,8 @@ struct arm_ccn {
189 struct arm_ccn_dt dt; 189 struct arm_ccn_dt dt;
190}; 190};
191 191
192static DEFINE_MUTEX(arm_ccn_mutex);
193static LIST_HEAD(arm_ccn_list);
192 194
193static int arm_ccn_node_to_xp(int node) 195static int arm_ccn_node_to_xp(int node)
194{ 196{
@@ -1171,30 +1173,27 @@ static enum hrtimer_restart arm_ccn_pmu_timer_handler(struct hrtimer *hrtimer)
1171} 1173}
1172 1174
1173 1175
1174static int arm_ccn_pmu_cpu_notifier(struct notifier_block *nb, 1176static int arm_ccn_pmu_offline_cpu(unsigned int cpu)
1175 unsigned long action, void *hcpu)
1176{ 1177{
1177 struct arm_ccn_dt *dt = container_of(nb, struct arm_ccn_dt, cpu_nb); 1178 struct arm_ccn_dt *dt;
1178 struct arm_ccn *ccn = container_of(dt, struct arm_ccn, dt);
1179 unsigned int cpu = (long)hcpu; /* for (long) see kernel/cpu.c */
1180 unsigned int target; 1179 unsigned int target;
1181 1180
1182 switch (action & ~CPU_TASKS_FROZEN) { 1181 mutex_lock(&arm_ccn_mutex);
1183 case CPU_DOWN_PREPARE: 1182 list_for_each_entry(dt, &arm_ccn_list, entry) {
1183 struct arm_ccn *ccn = container_of(dt, struct arm_ccn, dt);
1184
1184 if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu)) 1185 if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu))
1185 break; 1186 continue;
1186 target = cpumask_any_but(cpu_online_mask, cpu); 1187 target = cpumask_any_but(cpu_online_mask, cpu);
1187 if (target >= nr_cpu_ids) 1188 if (target >= nr_cpu_ids)
1188 break; 1189 continue;
1189 perf_pmu_migrate_context(&dt->pmu, cpu, target); 1190 perf_pmu_migrate_context(&dt->pmu, cpu, target);
1190 cpumask_set_cpu(target, &dt->cpu); 1191 cpumask_set_cpu(target, &dt->cpu);
1191 if (ccn->irq) 1192 if (ccn->irq)
1192 WARN_ON(irq_set_affinity_hint(ccn->irq, &dt->cpu) != 0); 1193 WARN_ON(irq_set_affinity_hint(ccn->irq, &dt->cpu) != 0);
1193 default:
1194 break;
1195 } 1194 }
1196 1195 mutex_unlock(&arm_ccn_mutex);
1197 return NOTIFY_OK; 1196 return 0;
1198} 1197}
1199 1198
1200 1199
@@ -1266,16 +1265,6 @@ static int arm_ccn_pmu_init(struct arm_ccn *ccn)
1266 /* Pick one CPU which we will use to collect data from CCN... */ 1265 /* Pick one CPU which we will use to collect data from CCN... */
1267 cpumask_set_cpu(smp_processor_id(), &ccn->dt.cpu); 1266 cpumask_set_cpu(smp_processor_id(), &ccn->dt.cpu);
1268 1267
1269 /*
1270 * ... and change the selection when it goes offline. Priority is
1271 * picked to have a chance to migrate events before perf is notified.
1272 */
1273 ccn->dt.cpu_nb.notifier_call = arm_ccn_pmu_cpu_notifier;
1274 ccn->dt.cpu_nb.priority = CPU_PRI_PERF + 1,
1275 err = register_cpu_notifier(&ccn->dt.cpu_nb);
1276 if (err)
1277 goto error_cpu_notifier;
1278
1279 /* Also make sure that the overflow interrupt is handled by this CPU */ 1268 /* Also make sure that the overflow interrupt is handled by this CPU */
1280 if (ccn->irq) { 1269 if (ccn->irq) {
1281 err = irq_set_affinity_hint(ccn->irq, &ccn->dt.cpu); 1270 err = irq_set_affinity_hint(ccn->irq, &ccn->dt.cpu);
@@ -1289,12 +1278,13 @@ static int arm_ccn_pmu_init(struct arm_ccn *ccn)
1289 if (err) 1278 if (err)
1290 goto error_pmu_register; 1279 goto error_pmu_register;
1291 1280
1281 mutex_lock(&arm_ccn_mutex);
1282 list_add(&ccn->dt.entry, &arm_ccn_list);
1283 mutex_unlock(&arm_ccn_mutex);
1292 return 0; 1284 return 0;
1293 1285
1294error_pmu_register: 1286error_pmu_register:
1295error_set_affinity: 1287error_set_affinity:
1296 unregister_cpu_notifier(&ccn->dt.cpu_nb);
1297error_cpu_notifier:
1298 ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id); 1288 ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id);
1299 for (i = 0; i < ccn->num_xps; i++) 1289 for (i = 0; i < ccn->num_xps; i++)
1300 writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL); 1290 writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL);
@@ -1306,9 +1296,12 @@ static void arm_ccn_pmu_cleanup(struct arm_ccn *ccn)
1306{ 1296{
1307 int i; 1297 int i;
1308 1298
1299 mutex_lock(&arm_ccn_mutex);
1300 list_del(&ccn->dt.entry);
1301 mutex_unlock(&arm_ccn_mutex);
1302
1309 if (ccn->irq) 1303 if (ccn->irq)
1310 irq_set_affinity_hint(ccn->irq, NULL); 1304 irq_set_affinity_hint(ccn->irq, NULL);
1311 unregister_cpu_notifier(&ccn->dt.cpu_nb);
1312 for (i = 0; i < ccn->num_xps; i++) 1305 for (i = 0; i < ccn->num_xps; i++)
1313 writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL); 1306 writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL);
1314 writel(0, ccn->dt.base + CCN_DT_PMCR); 1307 writel(0, ccn->dt.base + CCN_DT_PMCR);
@@ -1316,7 +1309,6 @@ static void arm_ccn_pmu_cleanup(struct arm_ccn *ccn)
1316 ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id); 1309 ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id);
1317} 1310}
1318 1311
1319
1320static int arm_ccn_for_each_valid_region(struct arm_ccn *ccn, 1312static int arm_ccn_for_each_valid_region(struct arm_ccn *ccn,
1321 int (*callback)(struct arm_ccn *ccn, int region, 1313 int (*callback)(struct arm_ccn *ccn, int region,
1322 void __iomem *base, u32 type, u32 id)) 1314 void __iomem *base, u32 type, u32 id))
@@ -1533,7 +1525,13 @@ static struct platform_driver arm_ccn_driver = {
1533 1525
1534static int __init arm_ccn_init(void) 1526static int __init arm_ccn_init(void)
1535{ 1527{
1536 int i; 1528 int i, ret;
1529
1530 ret = cpuhp_setup_state_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE,
1531 "AP_PERF_ARM_CCN_ONLINE", NULL,
1532 arm_ccn_pmu_offline_cpu);
1533 if (ret)
1534 return ret;
1537 1535
1538 for (i = 0; i < ARRAY_SIZE(arm_ccn_pmu_events); i++) 1536 for (i = 0; i < ARRAY_SIZE(arm_ccn_pmu_events); i++)
1539 arm_ccn_pmu_events_attrs[i] = &arm_ccn_pmu_events[i].attr.attr; 1537 arm_ccn_pmu_events_attrs[i] = &arm_ccn_pmu_events[i].attr.attr;
@@ -1543,6 +1541,7 @@ static int __init arm_ccn_init(void)
1543 1541
1544static void __exit arm_ccn_exit(void) 1542static void __exit arm_ccn_exit(void)
1545{ 1543{
1544 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE);
1546 platform_driver_unregister(&arm_ccn_driver); 1545 platform_driver_unregister(&arm_ccn_driver);
1547} 1546}
1548 1547
diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
index 5effd3027319..28bce3f4f81d 100644
--- a/drivers/clocksource/arm_arch_timer.c
+++ b/drivers/clocksource/arm_arch_timer.c
@@ -370,8 +370,10 @@ static bool arch_timer_has_nonsecure_ppi(void)
370 arch_timer_ppi[PHYS_NONSECURE_PPI]); 370 arch_timer_ppi[PHYS_NONSECURE_PPI]);
371} 371}
372 372
373static int arch_timer_setup(struct clock_event_device *clk) 373static int arch_timer_starting_cpu(unsigned int cpu)
374{ 374{
375 struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
376
375 __arch_timer_setup(ARCH_CP15_TIMER, clk); 377 __arch_timer_setup(ARCH_CP15_TIMER, clk);
376 378
377 enable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], 0); 379 enable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], 0);
@@ -527,29 +529,14 @@ static void arch_timer_stop(struct clock_event_device *clk)
527 clk->set_state_shutdown(clk); 529 clk->set_state_shutdown(clk);
528} 530}
529 531
530static int arch_timer_cpu_notify(struct notifier_block *self, 532static int arch_timer_dying_cpu(unsigned int cpu)
531 unsigned long action, void *hcpu)
532{ 533{
533 /* 534 struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
534 * Grab cpu pointer in each case to avoid spurious
535 * preemptible warnings
536 */
537 switch (action & ~CPU_TASKS_FROZEN) {
538 case CPU_STARTING:
539 arch_timer_setup(this_cpu_ptr(arch_timer_evt));
540 break;
541 case CPU_DYING:
542 arch_timer_stop(this_cpu_ptr(arch_timer_evt));
543 break;
544 }
545 535
546 return NOTIFY_OK; 536 arch_timer_stop(clk);
537 return 0;
547} 538}
548 539
549static struct notifier_block arch_timer_cpu_nb = {
550 .notifier_call = arch_timer_cpu_notify,
551};
552
553#ifdef CONFIG_CPU_PM 540#ifdef CONFIG_CPU_PM
554static unsigned int saved_cntkctl; 541static unsigned int saved_cntkctl;
555static int arch_timer_cpu_pm_notify(struct notifier_block *self, 542static int arch_timer_cpu_pm_notify(struct notifier_block *self,
@@ -570,11 +557,21 @@ static int __init arch_timer_cpu_pm_init(void)
570{ 557{
571 return cpu_pm_register_notifier(&arch_timer_cpu_pm_notifier); 558 return cpu_pm_register_notifier(&arch_timer_cpu_pm_notifier);
572} 559}
560
561static void __init arch_timer_cpu_pm_deinit(void)
562{
563 WARN_ON(cpu_pm_unregister_notifier(&arch_timer_cpu_pm_notifier));
564}
565
573#else 566#else
574static int __init arch_timer_cpu_pm_init(void) 567static int __init arch_timer_cpu_pm_init(void)
575{ 568{
576 return 0; 569 return 0;
577} 570}
571
572static void __init arch_timer_cpu_pm_deinit(void)
573{
574}
578#endif 575#endif
579 576
580static int __init arch_timer_register(void) 577static int __init arch_timer_register(void)
@@ -621,22 +618,23 @@ static int __init arch_timer_register(void)
621 goto out_free; 618 goto out_free;
622 } 619 }
623 620
624 err = register_cpu_notifier(&arch_timer_cpu_nb);
625 if (err)
626 goto out_free_irq;
627
628 err = arch_timer_cpu_pm_init(); 621 err = arch_timer_cpu_pm_init();
629 if (err) 622 if (err)
630 goto out_unreg_notify; 623 goto out_unreg_notify;
631 624
632 /* Immediately configure the timer on the boot CPU */
633 arch_timer_setup(this_cpu_ptr(arch_timer_evt));
634 625
626 /* Register and immediately configure the timer on the boot CPU */
627 err = cpuhp_setup_state(CPUHP_AP_ARM_ARCH_TIMER_STARTING,
628 "AP_ARM_ARCH_TIMER_STARTING",
629 arch_timer_starting_cpu, arch_timer_dying_cpu);
630 if (err)
631 goto out_unreg_cpupm;
635 return 0; 632 return 0;
636 633
634out_unreg_cpupm:
635 arch_timer_cpu_pm_deinit();
636
637out_unreg_notify: 637out_unreg_notify:
638 unregister_cpu_notifier(&arch_timer_cpu_nb);
639out_free_irq:
640 free_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], arch_timer_evt); 638 free_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], arch_timer_evt);
641 if (arch_timer_has_nonsecure_ppi()) 639 if (arch_timer_has_nonsecure_ppi())
642 free_percpu_irq(arch_timer_ppi[PHYS_NONSECURE_PPI], 640 free_percpu_irq(arch_timer_ppi[PHYS_NONSECURE_PPI],
diff --git a/drivers/clocksource/arm_global_timer.c b/drivers/clocksource/arm_global_timer.c
index 2a9ceb6e93f9..8da03298f844 100644
--- a/drivers/clocksource/arm_global_timer.c
+++ b/drivers/clocksource/arm_global_timer.c
@@ -165,9 +165,9 @@ static irqreturn_t gt_clockevent_interrupt(int irq, void *dev_id)
165 return IRQ_HANDLED; 165 return IRQ_HANDLED;
166} 166}
167 167
168static int gt_clockevents_init(struct clock_event_device *clk) 168static int gt_starting_cpu(unsigned int cpu)
169{ 169{
170 int cpu = smp_processor_id(); 170 struct clock_event_device *clk = this_cpu_ptr(gt_evt);
171 171
172 clk->name = "arm_global_timer"; 172 clk->name = "arm_global_timer";
173 clk->features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT | 173 clk->features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT |
@@ -186,10 +186,13 @@ static int gt_clockevents_init(struct clock_event_device *clk)
186 return 0; 186 return 0;
187} 187}
188 188
189static void gt_clockevents_stop(struct clock_event_device *clk) 189static int gt_dying_cpu(unsigned int cpu)
190{ 190{
191 struct clock_event_device *clk = this_cpu_ptr(gt_evt);
192
191 gt_clockevent_shutdown(clk); 193 gt_clockevent_shutdown(clk);
192 disable_percpu_irq(clk->irq); 194 disable_percpu_irq(clk->irq);
195 return 0;
193} 196}
194 197
195static cycle_t gt_clocksource_read(struct clocksource *cs) 198static cycle_t gt_clocksource_read(struct clocksource *cs)
@@ -252,24 +255,6 @@ static int __init gt_clocksource_init(void)
252 return clocksource_register_hz(&gt_clocksource, gt_clk_rate); 255 return clocksource_register_hz(&gt_clocksource, gt_clk_rate);
253} 256}
254 257
255static int gt_cpu_notify(struct notifier_block *self, unsigned long action,
256 void *hcpu)
257{
258 switch (action & ~CPU_TASKS_FROZEN) {
259 case CPU_STARTING:
260 gt_clockevents_init(this_cpu_ptr(gt_evt));
261 break;
262 case CPU_DYING:
263 gt_clockevents_stop(this_cpu_ptr(gt_evt));
264 break;
265 }
266
267 return NOTIFY_OK;
268}
269static struct notifier_block gt_cpu_nb = {
270 .notifier_call = gt_cpu_notify,
271};
272
273static int __init global_timer_of_register(struct device_node *np) 258static int __init global_timer_of_register(struct device_node *np)
274{ 259{
275 struct clk *gt_clk; 260 struct clk *gt_clk;
@@ -325,18 +310,14 @@ static int __init global_timer_of_register(struct device_node *np)
325 goto out_free; 310 goto out_free;
326 } 311 }
327 312
328 err = register_cpu_notifier(&gt_cpu_nb); 313 /* Register and immediately configure the timer on the boot CPU */
329 if (err) {
330 pr_warn("global-timer: unable to register cpu notifier.\n");
331 goto out_irq;
332 }
333
334 /* Immediately configure the timer on the boot CPU */
335 err = gt_clocksource_init(); 314 err = gt_clocksource_init();
336 if (err) 315 if (err)
337 goto out_irq; 316 goto out_irq;
338 317
339 err = gt_clockevents_init(this_cpu_ptr(gt_evt)); 318 err = cpuhp_setup_state(CPUHP_AP_ARM_GLOBAL_TIMER_STARTING,
319 "AP_ARM_GLOBAL_TIMER_STARTING",
320 gt_starting_cpu, gt_dying_cpu);
340 if (err) 321 if (err)
341 goto out_irq; 322 goto out_irq;
342 323
diff --git a/drivers/clocksource/dummy_timer.c b/drivers/clocksource/dummy_timer.c
index 776b6c86dcd5..89f1c2edbe02 100644
--- a/drivers/clocksource/dummy_timer.c
+++ b/drivers/clocksource/dummy_timer.c
@@ -16,10 +16,9 @@
16 16
17static DEFINE_PER_CPU(struct clock_event_device, dummy_timer_evt); 17static DEFINE_PER_CPU(struct clock_event_device, dummy_timer_evt);
18 18
19static void dummy_timer_setup(void) 19static int dummy_timer_starting_cpu(unsigned int cpu)
20{ 20{
21 int cpu = smp_processor_id(); 21 struct clock_event_device *evt = per_cpu_ptr(&dummy_timer_evt, cpu);
22 struct clock_event_device *evt = raw_cpu_ptr(&dummy_timer_evt);
23 22
24 evt->name = "dummy_timer"; 23 evt->name = "dummy_timer";
25 evt->features = CLOCK_EVT_FEAT_PERIODIC | 24 evt->features = CLOCK_EVT_FEAT_PERIODIC |
@@ -29,36 +28,13 @@ static void dummy_timer_setup(void)
29 evt->cpumask = cpumask_of(cpu); 28 evt->cpumask = cpumask_of(cpu);
30 29
31 clockevents_register_device(evt); 30 clockevents_register_device(evt);
31 return 0;
32} 32}
33 33
34static int dummy_timer_cpu_notify(struct notifier_block *self,
35 unsigned long action, void *hcpu)
36{
37 if ((action & ~CPU_TASKS_FROZEN) == CPU_STARTING)
38 dummy_timer_setup();
39
40 return NOTIFY_OK;
41}
42
43static struct notifier_block dummy_timer_cpu_nb = {
44 .notifier_call = dummy_timer_cpu_notify,
45};
46
47static int __init dummy_timer_register(void) 34static int __init dummy_timer_register(void)
48{ 35{
49 int err = 0; 36 return cpuhp_setup_state(CPUHP_AP_DUMMY_TIMER_STARTING,
50 37 "AP_DUMMY_TIMER_STARTING",
51 cpu_notifier_register_begin(); 38 dummy_timer_starting_cpu, NULL);
52 err = __register_cpu_notifier(&dummy_timer_cpu_nb);
53 if (err)
54 goto out;
55
56 /* We won't get a call on the boot CPU, so register immediately */
57 if (num_possible_cpus() > 1)
58 dummy_timer_setup();
59
60out:
61 cpu_notifier_register_done();
62 return err;
63} 39}
64early_initcall(dummy_timer_register); 40early_initcall(dummy_timer_register);
diff --git a/drivers/clocksource/exynos_mct.c b/drivers/clocksource/exynos_mct.c
index 0d18dd4b3bd2..41840d02c331 100644
--- a/drivers/clocksource/exynos_mct.c
+++ b/drivers/clocksource/exynos_mct.c
@@ -443,10 +443,11 @@ static irqreturn_t exynos4_mct_tick_isr(int irq, void *dev_id)
443 return IRQ_HANDLED; 443 return IRQ_HANDLED;
444} 444}
445 445
446static int exynos4_local_timer_setup(struct mct_clock_event_device *mevt) 446static int exynos4_mct_starting_cpu(unsigned int cpu)
447{ 447{
448 struct mct_clock_event_device *mevt =
449 per_cpu_ptr(&percpu_mct_tick, cpu);
448 struct clock_event_device *evt = &mevt->evt; 450 struct clock_event_device *evt = &mevt->evt;
449 unsigned int cpu = smp_processor_id();
450 451
451 mevt->base = EXYNOS4_MCT_L_BASE(cpu); 452 mevt->base = EXYNOS4_MCT_L_BASE(cpu);
452 snprintf(mevt->name, sizeof(mevt->name), "mct_tick%d", cpu); 453 snprintf(mevt->name, sizeof(mevt->name), "mct_tick%d", cpu);
@@ -480,8 +481,10 @@ static int exynos4_local_timer_setup(struct mct_clock_event_device *mevt)
480 return 0; 481 return 0;
481} 482}
482 483
483static void exynos4_local_timer_stop(struct mct_clock_event_device *mevt) 484static int exynos4_mct_dying_cpu(unsigned int cpu)
484{ 485{
486 struct mct_clock_event_device *mevt =
487 per_cpu_ptr(&percpu_mct_tick, cpu);
485 struct clock_event_device *evt = &mevt->evt; 488 struct clock_event_device *evt = &mevt->evt;
486 489
487 evt->set_state_shutdown(evt); 490 evt->set_state_shutdown(evt);
@@ -491,39 +494,12 @@ static void exynos4_local_timer_stop(struct mct_clock_event_device *mevt)
491 } else { 494 } else {
492 disable_percpu_irq(mct_irqs[MCT_L0_IRQ]); 495 disable_percpu_irq(mct_irqs[MCT_L0_IRQ]);
493 } 496 }
497 return 0;
494} 498}
495 499
496static int exynos4_mct_cpu_notify(struct notifier_block *self,
497 unsigned long action, void *hcpu)
498{
499 struct mct_clock_event_device *mevt;
500
501 /*
502 * Grab cpu pointer in each case to avoid spurious
503 * preemptible warnings
504 */
505 switch (action & ~CPU_TASKS_FROZEN) {
506 case CPU_STARTING:
507 mevt = this_cpu_ptr(&percpu_mct_tick);
508 exynos4_local_timer_setup(mevt);
509 break;
510 case CPU_DYING:
511 mevt = this_cpu_ptr(&percpu_mct_tick);
512 exynos4_local_timer_stop(mevt);
513 break;
514 }
515
516 return NOTIFY_OK;
517}
518
519static struct notifier_block exynos4_mct_cpu_nb = {
520 .notifier_call = exynos4_mct_cpu_notify,
521};
522
523static int __init exynos4_timer_resources(struct device_node *np, void __iomem *base) 500static int __init exynos4_timer_resources(struct device_node *np, void __iomem *base)
524{ 501{
525 int err, cpu; 502 int err, cpu;
526 struct mct_clock_event_device *mevt = this_cpu_ptr(&percpu_mct_tick);
527 struct clk *mct_clk, *tick_clk; 503 struct clk *mct_clk, *tick_clk;
528 504
529 tick_clk = np ? of_clk_get_by_name(np, "fin_pll") : 505 tick_clk = np ? of_clk_get_by_name(np, "fin_pll") :
@@ -570,12 +546,14 @@ static int __init exynos4_timer_resources(struct device_node *np, void __iomem *
570 } 546 }
571 } 547 }
572 548
573 err = register_cpu_notifier(&exynos4_mct_cpu_nb); 549 /* Install hotplug callbacks which configure the timer on this CPU */
550 err = cpuhp_setup_state(CPUHP_AP_EXYNOS4_MCT_TIMER_STARTING,
551 "AP_EXYNOS4_MCT_TIMER_STARTING",
552 exynos4_mct_starting_cpu,
553 exynos4_mct_dying_cpu);
574 if (err) 554 if (err)
575 goto out_irq; 555 goto out_irq;
576 556
577 /* Immediately configure the timer on the boot CPU */
578 exynos4_local_timer_setup(mevt);
579 return 0; 557 return 0;
580 558
581out_irq: 559out_irq:
diff --git a/drivers/clocksource/metag_generic.c b/drivers/clocksource/metag_generic.c
index bcd5c0d602a0..a80ab3e446b7 100644
--- a/drivers/clocksource/metag_generic.c
+++ b/drivers/clocksource/metag_generic.c
@@ -90,7 +90,7 @@ unsigned long long sched_clock(void)
90 return ticks << HARDWARE_TO_NS_SHIFT; 90 return ticks << HARDWARE_TO_NS_SHIFT;
91} 91}
92 92
93static void arch_timer_setup(unsigned int cpu) 93static int arch_timer_starting_cpu(unsigned int cpu)
94{ 94{
95 unsigned int txdivtime; 95 unsigned int txdivtime;
96 struct clock_event_device *clk = &per_cpu(local_clockevent, cpu); 96 struct clock_event_device *clk = &per_cpu(local_clockevent, cpu);
@@ -132,27 +132,9 @@ static void arch_timer_setup(unsigned int cpu)
132 val = core_reg_read(TXUCT_ID, TXTIMER_REGNUM, thread0); 132 val = core_reg_read(TXUCT_ID, TXTIMER_REGNUM, thread0);
133 __core_reg_set(TXTIMER, val); 133 __core_reg_set(TXTIMER, val);
134 } 134 }
135 return 0;
135} 136}
136 137
137static int arch_timer_cpu_notify(struct notifier_block *self,
138 unsigned long action, void *hcpu)
139{
140 int cpu = (long)hcpu;
141
142 switch (action) {
143 case CPU_STARTING:
144 case CPU_STARTING_FROZEN:
145 arch_timer_setup(cpu);
146 break;
147 }
148
149 return NOTIFY_OK;
150}
151
152static struct notifier_block arch_timer_cpu_nb = {
153 .notifier_call = arch_timer_cpu_notify,
154};
155
156int __init metag_generic_timer_init(void) 138int __init metag_generic_timer_init(void)
157{ 139{
158 /* 140 /*
@@ -170,11 +152,8 @@ int __init metag_generic_timer_init(void)
170 152
171 setup_irq(tbisig_map(TBID_SIGNUM_TRT), &metag_timer_irq); 153 setup_irq(tbisig_map(TBID_SIGNUM_TRT), &metag_timer_irq);
172 154
173 /* Configure timer on boot CPU */ 155 /* Hook cpu boot to configure the CPU's timers */
174 arch_timer_setup(smp_processor_id()); 156 return cpuhp_setup_state(CPUHP_AP_METAG_TIMER_STARTING,
175 157 "AP_METAG_TIMER_STARTING",
176 /* Hook cpu boot to configure other CPU's timers */ 158 arch_timer_starting_cpu, NULL);
177 register_cpu_notifier(&arch_timer_cpu_nb);
178
179 return 0;
180} 159}
diff --git a/drivers/clocksource/mips-gic-timer.c b/drivers/clocksource/mips-gic-timer.c
index 1572c7a778ab..d91e8725917c 100644
--- a/drivers/clocksource/mips-gic-timer.c
+++ b/drivers/clocksource/mips-gic-timer.c
@@ -49,10 +49,9 @@ struct irqaction gic_compare_irqaction = {
49 .name = "timer", 49 .name = "timer",
50}; 50};
51 51
52static void gic_clockevent_cpu_init(struct clock_event_device *cd) 52static void gic_clockevent_cpu_init(unsigned int cpu,
53 struct clock_event_device *cd)
53{ 54{
54 unsigned int cpu = smp_processor_id();
55
56 cd->name = "MIPS GIC"; 55 cd->name = "MIPS GIC";
57 cd->features = CLOCK_EVT_FEAT_ONESHOT | 56 cd->features = CLOCK_EVT_FEAT_ONESHOT |
58 CLOCK_EVT_FEAT_C3STOP; 57 CLOCK_EVT_FEAT_C3STOP;
@@ -79,19 +78,10 @@ static void gic_update_frequency(void *data)
79 clockevents_update_freq(this_cpu_ptr(&gic_clockevent_device), rate); 78 clockevents_update_freq(this_cpu_ptr(&gic_clockevent_device), rate);
80} 79}
81 80
82static int gic_cpu_notifier(struct notifier_block *nb, unsigned long action, 81static int gic_starting_cpu(unsigned int cpu)
83 void *data)
84{ 82{
85 switch (action & ~CPU_TASKS_FROZEN) { 83 gic_clockevent_cpu_init(cpu, this_cpu_ptr(&gic_clockevent_device));
86 case CPU_STARTING: 84 return 0;
87 gic_clockevent_cpu_init(this_cpu_ptr(&gic_clockevent_device));
88 break;
89 case CPU_DYING:
90 gic_clockevent_cpu_exit(this_cpu_ptr(&gic_clockevent_device));
91 break;
92 }
93
94 return NOTIFY_OK;
95} 85}
96 86
97static int gic_clk_notifier(struct notifier_block *nb, unsigned long action, 87static int gic_clk_notifier(struct notifier_block *nb, unsigned long action,
@@ -105,10 +95,11 @@ static int gic_clk_notifier(struct notifier_block *nb, unsigned long action,
105 return NOTIFY_OK; 95 return NOTIFY_OK;
106} 96}
107 97
108 98static int gic_dying_cpu(unsigned int cpu)
109static struct notifier_block gic_cpu_nb = { 99{
110 .notifier_call = gic_cpu_notifier, 100 gic_clockevent_cpu_exit(this_cpu_ptr(&gic_clockevent_device));
111}; 101 return 0;
102}
112 103
113static struct notifier_block gic_clk_nb = { 104static struct notifier_block gic_clk_nb = {
114 .notifier_call = gic_clk_notifier, 105 .notifier_call = gic_clk_notifier,
@@ -125,12 +116,9 @@ static int gic_clockevent_init(void)
125 if (ret < 0) 116 if (ret < 0)
126 return ret; 117 return ret;
127 118
128 ret = register_cpu_notifier(&gic_cpu_nb); 119 cpuhp_setup_state(CPUHP_AP_MIPS_GIC_TIMER_STARTING,
129 if (ret < 0) 120 "AP_MIPS_GIC_TIMER_STARTING", gic_starting_cpu,
130 pr_warn("GIC: Unable to register CPU notifier\n"); 121 gic_dying_cpu);
131
132 gic_clockevent_cpu_init(this_cpu_ptr(&gic_clockevent_device));
133
134 return 0; 122 return 0;
135} 123}
136 124
diff --git a/drivers/clocksource/qcom-timer.c b/drivers/clocksource/qcom-timer.c
index 662576339049..3283cfa2aa52 100644
--- a/drivers/clocksource/qcom-timer.c
+++ b/drivers/clocksource/qcom-timer.c
@@ -105,9 +105,9 @@ static struct clocksource msm_clocksource = {
105static int msm_timer_irq; 105static int msm_timer_irq;
106static int msm_timer_has_ppi; 106static int msm_timer_has_ppi;
107 107
108static int msm_local_timer_setup(struct clock_event_device *evt) 108static int msm_local_timer_starting_cpu(unsigned int cpu)
109{ 109{
110 int cpu = smp_processor_id(); 110 struct clock_event_device *evt = per_cpu_ptr(msm_evt, cpu);
111 int err; 111 int err;
112 112
113 evt->irq = msm_timer_irq; 113 evt->irq = msm_timer_irq;
@@ -135,35 +135,15 @@ static int msm_local_timer_setup(struct clock_event_device *evt)
135 return 0; 135 return 0;
136} 136}
137 137
138static void msm_local_timer_stop(struct clock_event_device *evt) 138static int msm_local_timer_dying_cpu(unsigned int cpu)
139{ 139{
140 struct clock_event_device *evt = per_cpu_ptr(msm_evt, cpu);
141
140 evt->set_state_shutdown(evt); 142 evt->set_state_shutdown(evt);
141 disable_percpu_irq(evt->irq); 143 disable_percpu_irq(evt->irq);
144 return 0;
142} 145}
143 146
144static int msm_timer_cpu_notify(struct notifier_block *self,
145 unsigned long action, void *hcpu)
146{
147 /*
148 * Grab cpu pointer in each case to avoid spurious
149 * preemptible warnings
150 */
151 switch (action & ~CPU_TASKS_FROZEN) {
152 case CPU_STARTING:
153 msm_local_timer_setup(this_cpu_ptr(msm_evt));
154 break;
155 case CPU_DYING:
156 msm_local_timer_stop(this_cpu_ptr(msm_evt));
157 break;
158 }
159
160 return NOTIFY_OK;
161}
162
163static struct notifier_block msm_timer_cpu_nb = {
164 .notifier_call = msm_timer_cpu_notify,
165};
166
167static u64 notrace msm_sched_clock_read(void) 147static u64 notrace msm_sched_clock_read(void)
168{ 148{
169 return msm_clocksource.read(&msm_clocksource); 149 return msm_clocksource.read(&msm_clocksource);
@@ -200,14 +180,15 @@ static int __init msm_timer_init(u32 dgt_hz, int sched_bits, int irq,
200 if (res) { 180 if (res) {
201 pr_err("request_percpu_irq failed\n"); 181 pr_err("request_percpu_irq failed\n");
202 } else { 182 } else {
203 res = register_cpu_notifier(&msm_timer_cpu_nb); 183 /* Install and invoke hotplug callbacks */
184 res = cpuhp_setup_state(CPUHP_AP_QCOM_TIMER_STARTING,
185 "AP_QCOM_TIMER_STARTING",
186 msm_local_timer_starting_cpu,
187 msm_local_timer_dying_cpu);
204 if (res) { 188 if (res) {
205 free_percpu_irq(irq, msm_evt); 189 free_percpu_irq(irq, msm_evt);
206 goto err; 190 goto err;
207 } 191 }
208
209 /* Immediately configure the timer on the boot CPU */
210 msm_local_timer_setup(raw_cpu_ptr(msm_evt));
211 } 192 }
212 193
213err: 194err:
diff --git a/drivers/clocksource/time-armada-370-xp.c b/drivers/clocksource/time-armada-370-xp.c
index 20ec066481fe..719b478d136e 100644
--- a/drivers/clocksource/time-armada-370-xp.c
+++ b/drivers/clocksource/time-armada-370-xp.c
@@ -170,10 +170,10 @@ static irqreturn_t armada_370_xp_timer_interrupt(int irq, void *dev_id)
170/* 170/*
171 * Setup the local clock events for a CPU. 171 * Setup the local clock events for a CPU.
172 */ 172 */
173static int armada_370_xp_timer_setup(struct clock_event_device *evt) 173static int armada_370_xp_timer_starting_cpu(unsigned int cpu)
174{ 174{
175 struct clock_event_device *evt = per_cpu_ptr(armada_370_xp_evt, cpu);
175 u32 clr = 0, set = 0; 176 u32 clr = 0, set = 0;
176 int cpu = smp_processor_id();
177 177
178 if (timer25Mhz) 178 if (timer25Mhz)
179 set = TIMER0_25MHZ; 179 set = TIMER0_25MHZ;
@@ -200,35 +200,15 @@ static int armada_370_xp_timer_setup(struct clock_event_device *evt)
200 return 0; 200 return 0;
201} 201}
202 202
203static void armada_370_xp_timer_stop(struct clock_event_device *evt) 203static int armada_370_xp_timer_dying_cpu(unsigned int cpu)
204{ 204{
205 struct clock_event_device *evt = per_cpu_ptr(armada_370_xp_evt, cpu);
206
205 evt->set_state_shutdown(evt); 207 evt->set_state_shutdown(evt);
206 disable_percpu_irq(evt->irq); 208 disable_percpu_irq(evt->irq);
209 return 0;
207} 210}
208 211
209static int armada_370_xp_timer_cpu_notify(struct notifier_block *self,
210 unsigned long action, void *hcpu)
211{
212 /*
213 * Grab cpu pointer in each case to avoid spurious
214 * preemptible warnings
215 */
216 switch (action & ~CPU_TASKS_FROZEN) {
217 case CPU_STARTING:
218 armada_370_xp_timer_setup(this_cpu_ptr(armada_370_xp_evt));
219 break;
220 case CPU_DYING:
221 armada_370_xp_timer_stop(this_cpu_ptr(armada_370_xp_evt));
222 break;
223 }
224
225 return NOTIFY_OK;
226}
227
228static struct notifier_block armada_370_xp_timer_cpu_nb = {
229 .notifier_call = armada_370_xp_timer_cpu_notify,
230};
231
232static u32 timer0_ctrl_reg, timer0_local_ctrl_reg; 212static u32 timer0_ctrl_reg, timer0_local_ctrl_reg;
233 213
234static int armada_370_xp_timer_suspend(void) 214static int armada_370_xp_timer_suspend(void)
@@ -322,8 +302,6 @@ static int __init armada_370_xp_timer_common_init(struct device_node *np)
322 return res; 302 return res;
323 } 303 }
324 304
325 register_cpu_notifier(&armada_370_xp_timer_cpu_nb);
326
327 armada_370_xp_evt = alloc_percpu(struct clock_event_device); 305 armada_370_xp_evt = alloc_percpu(struct clock_event_device);
328 if (!armada_370_xp_evt) 306 if (!armada_370_xp_evt)
329 return -ENOMEM; 307 return -ENOMEM;
@@ -341,9 +319,12 @@ static int __init armada_370_xp_timer_common_init(struct device_node *np)
341 return res; 319 return res;
342 } 320 }
343 321
344 res = armada_370_xp_timer_setup(this_cpu_ptr(armada_370_xp_evt)); 322 res = cpuhp_setup_state(CPUHP_AP_ARMADA_TIMER_STARTING,
323 "AP_ARMADA_TIMER_STARTING",
324 armada_370_xp_timer_starting_cpu,
325 armada_370_xp_timer_dying_cpu);
345 if (res) { 326 if (res) {
346 pr_err("Failed to setup timer"); 327 pr_err("Failed to setup hotplug state and timer");
347 return res; 328 return res;
348 } 329 }
349 330
diff --git a/drivers/clocksource/timer-atlas7.c b/drivers/clocksource/timer-atlas7.c
index 90f8fbc154a4..4334e0330ada 100644
--- a/drivers/clocksource/timer-atlas7.c
+++ b/drivers/clocksource/timer-atlas7.c
@@ -172,9 +172,9 @@ static struct irqaction sirfsoc_timer1_irq = {
172 .handler = sirfsoc_timer_interrupt, 172 .handler = sirfsoc_timer_interrupt,
173}; 173};
174 174
175static int sirfsoc_local_timer_setup(struct clock_event_device *ce) 175static int sirfsoc_local_timer_starting_cpu(unsigned int cpu)
176{ 176{
177 int cpu = smp_processor_id(); 177 struct clock_event_device *ce = per_cpu_ptr(sirfsoc_clockevent, cpu);
178 struct irqaction *action; 178 struct irqaction *action;
179 179
180 if (cpu == 0) 180 if (cpu == 0)
@@ -203,50 +203,27 @@ static int sirfsoc_local_timer_setup(struct clock_event_device *ce)
203 return 0; 203 return 0;
204} 204}
205 205
206static void sirfsoc_local_timer_stop(struct clock_event_device *ce) 206static int sirfsoc_local_timer_dying_cpu(unsigned int cpu)
207{ 207{
208 int cpu = smp_processor_id();
209
210 sirfsoc_timer_count_disable(1); 208 sirfsoc_timer_count_disable(1);
211 209
212 if (cpu == 0) 210 if (cpu == 0)
213 remove_irq(sirfsoc_timer_irq.irq, &sirfsoc_timer_irq); 211 remove_irq(sirfsoc_timer_irq.irq, &sirfsoc_timer_irq);
214 else 212 else
215 remove_irq(sirfsoc_timer1_irq.irq, &sirfsoc_timer1_irq); 213 remove_irq(sirfsoc_timer1_irq.irq, &sirfsoc_timer1_irq);
214 return 0;
216} 215}
217 216
218static int sirfsoc_cpu_notify(struct notifier_block *self,
219 unsigned long action, void *hcpu)
220{
221 /*
222 * Grab cpu pointer in each case to avoid spurious
223 * preemptible warnings
224 */
225 switch (action & ~CPU_TASKS_FROZEN) {
226 case CPU_STARTING:
227 sirfsoc_local_timer_setup(this_cpu_ptr(sirfsoc_clockevent));
228 break;
229 case CPU_DYING:
230 sirfsoc_local_timer_stop(this_cpu_ptr(sirfsoc_clockevent));
231 break;
232 }
233
234 return NOTIFY_OK;
235}
236
237static struct notifier_block sirfsoc_cpu_nb = {
238 .notifier_call = sirfsoc_cpu_notify,
239};
240
241static int __init sirfsoc_clockevent_init(void) 217static int __init sirfsoc_clockevent_init(void)
242{ 218{
243 sirfsoc_clockevent = alloc_percpu(struct clock_event_device); 219 sirfsoc_clockevent = alloc_percpu(struct clock_event_device);
244 BUG_ON(!sirfsoc_clockevent); 220 BUG_ON(!sirfsoc_clockevent);
245 221
246 BUG_ON(register_cpu_notifier(&sirfsoc_cpu_nb)); 222 /* Install and invoke hotplug callbacks */
247 223 return cpuhp_setup_state(CPUHP_AP_MARCO_TIMER_STARTING,
248 /* Immediately configure the timer on the boot CPU */ 224 "AP_MARCO_TIMER_STARTING",
249 return sirfsoc_local_timer_setup(this_cpu_ptr(sirfsoc_clockevent)); 225 sirfsoc_local_timer_starting_cpu,
226 sirfsoc_local_timer_dying_cpu);
250} 227}
251 228
252/* initialize the kernel jiffy timer source */ 229/* initialize the kernel jiffy timer source */
diff --git a/drivers/hwtracing/coresight/coresight-etm3x.c b/drivers/hwtracing/coresight/coresight-etm3x.c
index d83ab82672e4..2de4cad9c5ed 100644
--- a/drivers/hwtracing/coresight/coresight-etm3x.c
+++ b/drivers/hwtracing/coresight/coresight-etm3x.c
@@ -51,6 +51,8 @@ module_param_named(boot_enable, boot_enable, int, S_IRUGO);
51static int etm_count; 51static int etm_count;
52static struct etm_drvdata *etmdrvdata[NR_CPUS]; 52static struct etm_drvdata *etmdrvdata[NR_CPUS];
53 53
54static enum cpuhp_state hp_online;
55
54/* 56/*
55 * Memory mapped writes to clear os lock are not supported on some processors 57 * Memory mapped writes to clear os lock are not supported on some processors
56 * and OS lock must be unlocked before any memory mapped access on such 58 * and OS lock must be unlocked before any memory mapped access on such
@@ -481,8 +483,7 @@ static int etm_enable_sysfs(struct coresight_device *csdev)
481 483
482 /* 484 /*
483 * Configure the ETM only if the CPU is online. If it isn't online 485 * Configure the ETM only if the CPU is online. If it isn't online
484 * hw configuration will take place when 'CPU_STARTING' is received 486 * hw configuration will take place on the local CPU during bring up.
485 * in @etm_cpu_callback.
486 */ 487 */
487 if (cpu_online(drvdata->cpu)) { 488 if (cpu_online(drvdata->cpu)) {
488 ret = smp_call_function_single(drvdata->cpu, 489 ret = smp_call_function_single(drvdata->cpu,
@@ -641,47 +642,44 @@ static const struct coresight_ops etm_cs_ops = {
641 .source_ops = &etm_source_ops, 642 .source_ops = &etm_source_ops,
642}; 643};
643 644
644static int etm_cpu_callback(struct notifier_block *nfb, unsigned long action, 645static int etm_online_cpu(unsigned int cpu)
645 void *hcpu)
646{ 646{
647 unsigned int cpu = (unsigned long)hcpu;
648
649 if (!etmdrvdata[cpu]) 647 if (!etmdrvdata[cpu])
650 goto out; 648 return 0;
651 649
652 switch (action & (~CPU_TASKS_FROZEN)) { 650 if (etmdrvdata[cpu]->boot_enable && !etmdrvdata[cpu]->sticky_enable)
653 case CPU_STARTING: 651 coresight_enable(etmdrvdata[cpu]->csdev);
654 spin_lock(&etmdrvdata[cpu]->spinlock); 652 return 0;
655 if (!etmdrvdata[cpu]->os_unlock) { 653}
656 etm_os_unlock(etmdrvdata[cpu]);
657 etmdrvdata[cpu]->os_unlock = true;
658 }
659
660 if (local_read(&etmdrvdata[cpu]->mode))
661 etm_enable_hw(etmdrvdata[cpu]);
662 spin_unlock(&etmdrvdata[cpu]->spinlock);
663 break;
664 654
665 case CPU_ONLINE: 655static int etm_starting_cpu(unsigned int cpu)
666 if (etmdrvdata[cpu]->boot_enable && 656{
667 !etmdrvdata[cpu]->sticky_enable) 657 if (!etmdrvdata[cpu])
668 coresight_enable(etmdrvdata[cpu]->csdev); 658 return 0;
669 break;
670 659
671 case CPU_DYING: 660 spin_lock(&etmdrvdata[cpu]->spinlock);
672 spin_lock(&etmdrvdata[cpu]->spinlock); 661 if (!etmdrvdata[cpu]->os_unlock) {
673 if (local_read(&etmdrvdata[cpu]->mode)) 662 etm_os_unlock(etmdrvdata[cpu]);
674 etm_disable_hw(etmdrvdata[cpu]); 663 etmdrvdata[cpu]->os_unlock = true;
675 spin_unlock(&etmdrvdata[cpu]->spinlock);
676 break;
677 } 664 }
678out: 665
679 return NOTIFY_OK; 666 if (local_read(&etmdrvdata[cpu]->mode))
667 etm_enable_hw(etmdrvdata[cpu]);
668 spin_unlock(&etmdrvdata[cpu]->spinlock);
669 return 0;
680} 670}
681 671
682static struct notifier_block etm_cpu_notifier = { 672static int etm_dying_cpu(unsigned int cpu)
683 .notifier_call = etm_cpu_callback, 673{
684}; 674 if (!etmdrvdata[cpu])
675 return 0;
676
677 spin_lock(&etmdrvdata[cpu]->spinlock);
678 if (local_read(&etmdrvdata[cpu]->mode))
679 etm_disable_hw(etmdrvdata[cpu]);
680 spin_unlock(&etmdrvdata[cpu]->spinlock);
681 return 0;
682}
685 683
686static bool etm_arch_supported(u8 arch) 684static bool etm_arch_supported(u8 arch)
687{ 685{
@@ -806,9 +804,17 @@ static int etm_probe(struct amba_device *adev, const struct amba_id *id)
806 etm_init_arch_data, drvdata, 1)) 804 etm_init_arch_data, drvdata, 1))
807 dev_err(dev, "ETM arch init failed\n"); 805 dev_err(dev, "ETM arch init failed\n");
808 806
809 if (!etm_count++) 807 if (!etm_count++) {
810 register_hotcpu_notifier(&etm_cpu_notifier); 808 cpuhp_setup_state_nocalls(CPUHP_AP_ARM_CORESIGHT_STARTING,
811 809 "AP_ARM_CORESIGHT_STARTING",
810 etm_starting_cpu, etm_dying_cpu);
811 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
812 "AP_ARM_CORESIGHT_ONLINE",
813 etm_online_cpu, NULL);
814 if (ret < 0)
815 goto err_arch_supported;
816 hp_online = ret;
817 }
812 put_online_cpus(); 818 put_online_cpus();
813 819
814 if (etm_arch_supported(drvdata->arch) == false) { 820 if (etm_arch_supported(drvdata->arch) == false) {
@@ -839,7 +845,6 @@ static int etm_probe(struct amba_device *adev, const struct amba_id *id)
839 845
840 pm_runtime_put(&adev->dev); 846 pm_runtime_put(&adev->dev);
841 dev_info(dev, "%s initialized\n", (char *)id->data); 847 dev_info(dev, "%s initialized\n", (char *)id->data);
842
843 if (boot_enable) { 848 if (boot_enable) {
844 coresight_enable(drvdata->csdev); 849 coresight_enable(drvdata->csdev);
845 drvdata->boot_enable = true; 850 drvdata->boot_enable = true;
@@ -848,8 +853,11 @@ static int etm_probe(struct amba_device *adev, const struct amba_id *id)
848 return 0; 853 return 0;
849 854
850err_arch_supported: 855err_arch_supported:
851 if (--etm_count == 0) 856 if (--etm_count == 0) {
852 unregister_hotcpu_notifier(&etm_cpu_notifier); 857 cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_STARTING);
858 if (hp_online)
859 cpuhp_remove_state_nocalls(hp_online);
860 }
853 return ret; 861 return ret;
854} 862}
855 863
diff --git a/drivers/hwtracing/coresight/coresight-etm4x.c b/drivers/hwtracing/coresight/coresight-etm4x.c
index 462f0dc15757..1a5e0d14c1dd 100644
--- a/drivers/hwtracing/coresight/coresight-etm4x.c
+++ b/drivers/hwtracing/coresight/coresight-etm4x.c
@@ -48,6 +48,8 @@ static int etm4_count;
48static struct etmv4_drvdata *etmdrvdata[NR_CPUS]; 48static struct etmv4_drvdata *etmdrvdata[NR_CPUS];
49static void etm4_set_default(struct etmv4_config *config); 49static void etm4_set_default(struct etmv4_config *config);
50 50
51static enum cpuhp_state hp_online;
52
51static void etm4_os_unlock(struct etmv4_drvdata *drvdata) 53static void etm4_os_unlock(struct etmv4_drvdata *drvdata)
52{ 54{
53 /* Writing any value to ETMOSLAR unlocks the trace registers */ 55 /* Writing any value to ETMOSLAR unlocks the trace registers */
@@ -673,47 +675,44 @@ void etm4_config_trace_mode(struct etmv4_config *config)
673 config->addr_acc[ETM_DEFAULT_ADDR_COMP + 1] = addr_acc; 675 config->addr_acc[ETM_DEFAULT_ADDR_COMP + 1] = addr_acc;
674} 676}
675 677
676static int etm4_cpu_callback(struct notifier_block *nfb, unsigned long action, 678static int etm4_online_cpu(unsigned int cpu)
677 void *hcpu)
678{ 679{
679 unsigned int cpu = (unsigned long)hcpu;
680
681 if (!etmdrvdata[cpu]) 680 if (!etmdrvdata[cpu])
682 goto out; 681 return 0;
683
684 switch (action & (~CPU_TASKS_FROZEN)) {
685 case CPU_STARTING:
686 spin_lock(&etmdrvdata[cpu]->spinlock);
687 if (!etmdrvdata[cpu]->os_unlock) {
688 etm4_os_unlock(etmdrvdata[cpu]);
689 etmdrvdata[cpu]->os_unlock = true;
690 }
691
692 if (local_read(&etmdrvdata[cpu]->mode))
693 etm4_enable_hw(etmdrvdata[cpu]);
694 spin_unlock(&etmdrvdata[cpu]->spinlock);
695 break;
696 682
697 case CPU_ONLINE: 683 if (etmdrvdata[cpu]->boot_enable && !etmdrvdata[cpu]->sticky_enable)
698 if (etmdrvdata[cpu]->boot_enable && 684 coresight_enable(etmdrvdata[cpu]->csdev);
699 !etmdrvdata[cpu]->sticky_enable) 685 return 0;
700 coresight_enable(etmdrvdata[cpu]->csdev); 686}
701 break;
702 687
703 case CPU_DYING: 688static int etm4_starting_cpu(unsigned int cpu)
704 spin_lock(&etmdrvdata[cpu]->spinlock); 689{
705 if (local_read(&etmdrvdata[cpu]->mode)) 690 if (!etmdrvdata[cpu])
706 etm4_disable_hw(etmdrvdata[cpu]); 691 return 0;
707 spin_unlock(&etmdrvdata[cpu]->spinlock); 692
708 break; 693 spin_lock(&etmdrvdata[cpu]->spinlock);
694 if (!etmdrvdata[cpu]->os_unlock) {
695 etm4_os_unlock(etmdrvdata[cpu]);
696 etmdrvdata[cpu]->os_unlock = true;
709 } 697 }
710out: 698
711 return NOTIFY_OK; 699 if (local_read(&etmdrvdata[cpu]->mode))
700 etm4_enable_hw(etmdrvdata[cpu]);
701 spin_unlock(&etmdrvdata[cpu]->spinlock);
702 return 0;
712} 703}
713 704
714static struct notifier_block etm4_cpu_notifier = { 705static int etm4_dying_cpu(unsigned int cpu)
715 .notifier_call = etm4_cpu_callback, 706{
716}; 707 if (!etmdrvdata[cpu])
708 return 0;
709
710 spin_lock(&etmdrvdata[cpu]->spinlock);
711 if (local_read(&etmdrvdata[cpu]->mode))
712 etm4_disable_hw(etmdrvdata[cpu]);
713 spin_unlock(&etmdrvdata[cpu]->spinlock);
714 return 0;
715}
717 716
718static void etm4_init_trace_id(struct etmv4_drvdata *drvdata) 717static void etm4_init_trace_id(struct etmv4_drvdata *drvdata)
719{ 718{
@@ -767,8 +766,17 @@ static int etm4_probe(struct amba_device *adev, const struct amba_id *id)
767 etm4_init_arch_data, drvdata, 1)) 766 etm4_init_arch_data, drvdata, 1))
768 dev_err(dev, "ETM arch init failed\n"); 767 dev_err(dev, "ETM arch init failed\n");
769 768
770 if (!etm4_count++) 769 if (!etm4_count++) {
771 register_hotcpu_notifier(&etm4_cpu_notifier); 770 cpuhp_setup_state_nocalls(CPUHP_AP_ARM_CORESIGHT4_STARTING,
771 "AP_ARM_CORESIGHT4_STARTING",
772 etm4_starting_cpu, etm4_dying_cpu);
773 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
774 "AP_ARM_CORESIGHT4_ONLINE",
775 etm4_online_cpu, NULL);
776 if (ret < 0)
777 goto err_arch_supported;
778 hp_online = ret;
779 }
772 780
773 put_online_cpus(); 781 put_online_cpus();
774 782
@@ -809,8 +817,11 @@ static int etm4_probe(struct amba_device *adev, const struct amba_id *id)
809 return 0; 817 return 0;
810 818
811err_arch_supported: 819err_arch_supported:
812 if (--etm4_count == 0) 820 if (--etm4_count == 0) {
813 unregister_hotcpu_notifier(&etm4_cpu_notifier); 821 cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT4_STARTING);
822 if (hp_online)
823 cpuhp_remove_state_nocalls(hp_online);
824 }
814 return ret; 825 return ret;
815} 826}
816 827
diff --git a/drivers/irqchip/irq-armada-370-xp.c b/drivers/irqchip/irq-armada-370-xp.c
index 7c42b1d13faf..8bcee65a0b8c 100644
--- a/drivers/irqchip/irq-armada-370-xp.c
+++ b/drivers/irqchip/irq-armada-370-xp.c
@@ -345,38 +345,20 @@ static void armada_mpic_send_doorbell(const struct cpumask *mask,
345 ARMADA_370_XP_SW_TRIG_INT_OFFS); 345 ARMADA_370_XP_SW_TRIG_INT_OFFS);
346} 346}
347 347
348static int armada_xp_mpic_secondary_init(struct notifier_block *nfb, 348static int armada_xp_mpic_starting_cpu(unsigned int cpu)
349 unsigned long action, void *hcpu)
350{ 349{
351 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) { 350 armada_xp_mpic_perf_init();
352 armada_xp_mpic_perf_init(); 351 armada_xp_mpic_smp_cpu_init();
353 armada_xp_mpic_smp_cpu_init(); 352 return 0;
354 }
355
356 return NOTIFY_OK;
357} 353}
358 354
359static struct notifier_block armada_370_xp_mpic_cpu_notifier = { 355static int mpic_cascaded_starting_cpu(unsigned int cpu)
360 .notifier_call = armada_xp_mpic_secondary_init,
361 .priority = 100,
362};
363
364static int mpic_cascaded_secondary_init(struct notifier_block *nfb,
365 unsigned long action, void *hcpu)
366{ 356{
367 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) { 357 armada_xp_mpic_perf_init();
368 armada_xp_mpic_perf_init(); 358 enable_percpu_irq(parent_irq, IRQ_TYPE_NONE);
369 enable_percpu_irq(parent_irq, IRQ_TYPE_NONE); 359 return 0;
370 }
371
372 return NOTIFY_OK;
373} 360}
374 361#endif
375static struct notifier_block mpic_cascaded_cpu_notifier = {
376 .notifier_call = mpic_cascaded_secondary_init,
377 .priority = 100,
378};
379#endif /* CONFIG_SMP */
380 362
381static const struct irq_domain_ops armada_370_xp_mpic_irq_ops = { 363static const struct irq_domain_ops armada_370_xp_mpic_irq_ops = {
382 .map = armada_370_xp_mpic_irq_map, 364 .map = armada_370_xp_mpic_irq_map,
@@ -595,11 +577,15 @@ static int __init armada_370_xp_mpic_of_init(struct device_node *node,
595 set_handle_irq(armada_370_xp_handle_irq); 577 set_handle_irq(armada_370_xp_handle_irq);
596#ifdef CONFIG_SMP 578#ifdef CONFIG_SMP
597 set_smp_cross_call(armada_mpic_send_doorbell); 579 set_smp_cross_call(armada_mpic_send_doorbell);
598 register_cpu_notifier(&armada_370_xp_mpic_cpu_notifier); 580 cpuhp_setup_state_nocalls(CPUHP_AP_IRQ_ARMADA_XP_STARTING,
581 "AP_IRQ_ARMADA_XP_STARTING",
582 armada_xp_mpic_starting_cpu, NULL);
599#endif 583#endif
600 } else { 584 } else {
601#ifdef CONFIG_SMP 585#ifdef CONFIG_SMP
602 register_cpu_notifier(&mpic_cascaded_cpu_notifier); 586 cpuhp_setup_state_nocalls(CPUHP_AP_IRQ_ARMADA_CASC_STARTING,
587 "AP_IRQ_ARMADA_CASC_STARTING",
588 mpic_cascaded_starting_cpu, NULL);
603#endif 589#endif
604 irq_set_chained_handler(parent_irq, 590 irq_set_chained_handler(parent_irq,
605 armada_370_xp_mpic_handle_cascade_irq); 591 armada_370_xp_mpic_handle_cascade_irq);
diff --git a/drivers/irqchip/irq-bcm2836.c b/drivers/irqchip/irq-bcm2836.c
index df1949c0aa23..d96b2c947e74 100644
--- a/drivers/irqchip/irq-bcm2836.c
+++ b/drivers/irqchip/irq-bcm2836.c
@@ -202,26 +202,19 @@ static void bcm2836_arm_irqchip_send_ipi(const struct cpumask *mask,
202 } 202 }
203} 203}
204 204
205/* Unmasks the IPI on the CPU when it's online. */ 205static int bcm2836_cpu_starting(unsigned int cpu)
206static int bcm2836_arm_irqchip_cpu_notify(struct notifier_block *nfb,
207 unsigned long action, void *hcpu)
208{ 206{
209 unsigned int cpu = (unsigned long)hcpu; 207 bcm2836_arm_irqchip_unmask_per_cpu_irq(LOCAL_MAILBOX_INT_CONTROL0, 0,
210 unsigned int int_reg = LOCAL_MAILBOX_INT_CONTROL0; 208 cpu);
211 unsigned int mailbox = 0; 209 return 0;
212
213 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN)
214 bcm2836_arm_irqchip_unmask_per_cpu_irq(int_reg, mailbox, cpu);
215 else if (action == CPU_DYING)
216 bcm2836_arm_irqchip_mask_per_cpu_irq(int_reg, mailbox, cpu);
217
218 return NOTIFY_OK;
219} 210}
220 211
221static struct notifier_block bcm2836_arm_irqchip_cpu_notifier = { 212static int bcm2836_cpu_dying(unsigned int cpu)
222 .notifier_call = bcm2836_arm_irqchip_cpu_notify, 213{
223 .priority = 100, 214 bcm2836_arm_irqchip_mask_per_cpu_irq(LOCAL_MAILBOX_INT_CONTROL0, 0,
224}; 215 cpu);
216 return 0;
217}
225 218
226#ifdef CONFIG_ARM 219#ifdef CONFIG_ARM
227static int __init bcm2836_smp_boot_secondary(unsigned int cpu, 220static int __init bcm2836_smp_boot_secondary(unsigned int cpu,
@@ -251,10 +244,9 @@ bcm2836_arm_irqchip_smp_init(void)
251{ 244{
252#ifdef CONFIG_SMP 245#ifdef CONFIG_SMP
253 /* Unmask IPIs to the boot CPU. */ 246 /* Unmask IPIs to the boot CPU. */
254 bcm2836_arm_irqchip_cpu_notify(&bcm2836_arm_irqchip_cpu_notifier, 247 cpuhp_setup_state(CPUHP_AP_IRQ_BCM2836_STARTING,
255 CPU_STARTING, 248 "AP_IRQ_BCM2836_STARTING", bcm2836_cpu_starting,
256 (void *)(uintptr_t)smp_processor_id()); 249 bcm2836_cpu_dying);
257 register_cpu_notifier(&bcm2836_arm_irqchip_cpu_notifier);
258 250
259 set_smp_cross_call(bcm2836_arm_irqchip_send_ipi); 251 set_smp_cross_call(bcm2836_arm_irqchip_send_ipi);
260 252
diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
index 2c5ba0e704bf..6fc56c3466b0 100644
--- a/drivers/irqchip/irq-gic-v3.c
+++ b/drivers/irqchip/irq-gic-v3.c
@@ -538,23 +538,13 @@ static void gic_cpu_init(void)
538} 538}
539 539
540#ifdef CONFIG_SMP 540#ifdef CONFIG_SMP
541static int gic_secondary_init(struct notifier_block *nfb, 541
542 unsigned long action, void *hcpu) 542static int gic_starting_cpu(unsigned int cpu)
543{ 543{
544 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) 544 gic_cpu_init();
545 gic_cpu_init(); 545 return 0;
546 return NOTIFY_OK;
547} 546}
548 547
549/*
550 * Notifier for enabling the GIC CPU interface. Set an arbitrarily high
551 * priority because the GIC needs to be up before the ARM generic timers.
552 */
553static struct notifier_block gic_cpu_notifier = {
554 .notifier_call = gic_secondary_init,
555 .priority = 100,
556};
557
558static u16 gic_compute_target_list(int *base_cpu, const struct cpumask *mask, 548static u16 gic_compute_target_list(int *base_cpu, const struct cpumask *mask,
559 unsigned long cluster_id) 549 unsigned long cluster_id)
560{ 550{
@@ -634,7 +624,9 @@ static void gic_raise_softirq(const struct cpumask *mask, unsigned int irq)
634static void gic_smp_init(void) 624static void gic_smp_init(void)
635{ 625{
636 set_smp_cross_call(gic_raise_softirq); 626 set_smp_cross_call(gic_raise_softirq);
637 register_cpu_notifier(&gic_cpu_notifier); 627 cpuhp_setup_state_nocalls(CPUHP_AP_IRQ_GICV3_STARTING,
628 "AP_IRQ_GICV3_STARTING", gic_starting_cpu,
629 NULL);
638} 630}
639 631
640static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val, 632static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
diff --git a/drivers/irqchip/irq-gic.c b/drivers/irqchip/irq-gic.c
index 1de07eb5839c..c2cab572c511 100644
--- a/drivers/irqchip/irq-gic.c
+++ b/drivers/irqchip/irq-gic.c
@@ -984,25 +984,12 @@ static int gic_irq_domain_translate(struct irq_domain *d,
984 return -EINVAL; 984 return -EINVAL;
985} 985}
986 986
987#ifdef CONFIG_SMP 987static int gic_starting_cpu(unsigned int cpu)
988static int gic_secondary_init(struct notifier_block *nfb, unsigned long action,
989 void *hcpu)
990{ 988{
991 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) 989 gic_cpu_init(&gic_data[0]);
992 gic_cpu_init(&gic_data[0]); 990 return 0;
993 return NOTIFY_OK;
994} 991}
995 992
996/*
997 * Notifier for enabling the GIC CPU interface. Set an arbitrarily high
998 * priority because the GIC needs to be up before the ARM generic timers.
999 */
1000static struct notifier_block gic_cpu_notifier = {
1001 .notifier_call = gic_secondary_init,
1002 .priority = 100,
1003};
1004#endif
1005
1006static int gic_irq_domain_alloc(struct irq_domain *domain, unsigned int virq, 993static int gic_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
1007 unsigned int nr_irqs, void *arg) 994 unsigned int nr_irqs, void *arg)
1008{ 995{
@@ -1177,8 +1164,10 @@ static int __init __gic_init_bases(struct gic_chip_data *gic,
1177 gic_cpu_map[i] = 0xff; 1164 gic_cpu_map[i] = 0xff;
1178#ifdef CONFIG_SMP 1165#ifdef CONFIG_SMP
1179 set_smp_cross_call(gic_raise_softirq); 1166 set_smp_cross_call(gic_raise_softirq);
1180 register_cpu_notifier(&gic_cpu_notifier);
1181#endif 1167#endif
1168 cpuhp_setup_state_nocalls(CPUHP_AP_IRQ_GIC_STARTING,
1169 "AP_IRQ_GIC_STARTING",
1170 gic_starting_cpu, NULL);
1182 set_handle_irq(gic_handle_irq); 1171 set_handle_irq(gic_handle_irq);
1183 if (static_key_true(&supports_deactivate)) 1172 if (static_key_true(&supports_deactivate))
1184 pr_info("GIC: Using split EOI/Deactivate mode\n"); 1173 pr_info("GIC: Using split EOI/Deactivate mode\n");
diff --git a/drivers/irqchip/irq-hip04.c b/drivers/irqchip/irq-hip04.c
index 9e25d8ce08e5..021b0e0833c1 100644
--- a/drivers/irqchip/irq-hip04.c
+++ b/drivers/irqchip/irq-hip04.c
@@ -342,26 +342,12 @@ static int hip04_irq_domain_xlate(struct irq_domain *d,
342 return ret; 342 return ret;
343} 343}
344 344
345#ifdef CONFIG_SMP 345static int hip04_irq_starting_cpu(unsigned int cpu)
346static int hip04_irq_secondary_init(struct notifier_block *nfb,
347 unsigned long action,
348 void *hcpu)
349{ 346{
350 if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) 347 hip04_irq_cpu_init(&hip04_data);
351 hip04_irq_cpu_init(&hip04_data); 348 return 0;
352 return NOTIFY_OK;
353} 349}
354 350
355/*
356 * Notifier for enabling the INTC CPU interface. Set an arbitrarily high
357 * priority because the GIC needs to be up before the ARM generic timers.
358 */
359static struct notifier_block hip04_irq_cpu_notifier = {
360 .notifier_call = hip04_irq_secondary_init,
361 .priority = 100,
362};
363#endif
364
365static const struct irq_domain_ops hip04_irq_domain_ops = { 351static const struct irq_domain_ops hip04_irq_domain_ops = {
366 .map = hip04_irq_domain_map, 352 .map = hip04_irq_domain_map,
367 .xlate = hip04_irq_domain_xlate, 353 .xlate = hip04_irq_domain_xlate,
@@ -417,13 +403,12 @@ hip04_of_init(struct device_node *node, struct device_node *parent)
417 403
418#ifdef CONFIG_SMP 404#ifdef CONFIG_SMP
419 set_smp_cross_call(hip04_raise_softirq); 405 set_smp_cross_call(hip04_raise_softirq);
420 register_cpu_notifier(&hip04_irq_cpu_notifier);
421#endif 406#endif
422 set_handle_irq(hip04_handle_irq); 407 set_handle_irq(hip04_handle_irq);
423 408
424 hip04_irq_dist_init(&hip04_data); 409 hip04_irq_dist_init(&hip04_data);
425 hip04_irq_cpu_init(&hip04_data); 410 cpuhp_setup_state(CPUHP_AP_IRQ_HIP04_STARTING, "AP_IRQ_HIP04_STARTING",
426 411 hip04_irq_starting_cpu, NULL);
427 return 0; 412 return 0;
428} 413}
429IRQCHIP_DECLARE(hip04_intc, "hisilicon,hip04-intc", hip04_of_init); 414IRQCHIP_DECLARE(hip04_intc, "hisilicon,hip04-intc", hip04_of_init);
diff --git a/drivers/leds/trigger/ledtrig-cpu.c b/drivers/leds/trigger/ledtrig-cpu.c
index 938467fb82be..22f0634dd3fa 100644
--- a/drivers/leds/trigger/ledtrig-cpu.c
+++ b/drivers/leds/trigger/ledtrig-cpu.c
@@ -92,29 +92,22 @@ static struct syscore_ops ledtrig_cpu_syscore_ops = {
92 .resume = ledtrig_cpu_syscore_resume, 92 .resume = ledtrig_cpu_syscore_resume,
93}; 93};
94 94
95static int ledtrig_cpu_notify(struct notifier_block *self, 95static int ledtrig_online_cpu(unsigned int cpu)
96 unsigned long action, void *hcpu)
97{ 96{
98 switch (action & ~CPU_TASKS_FROZEN) { 97 ledtrig_cpu(CPU_LED_START);
99 case CPU_STARTING: 98 return 0;
100 ledtrig_cpu(CPU_LED_START);
101 break;
102 case CPU_DYING:
103 ledtrig_cpu(CPU_LED_STOP);
104 break;
105 }
106
107 return NOTIFY_OK;
108} 99}
109 100
110 101static int ledtrig_prepare_down_cpu(unsigned int cpu)
111static struct notifier_block ledtrig_cpu_nb = { 102{
112 .notifier_call = ledtrig_cpu_notify, 103 ledtrig_cpu(CPU_LED_STOP);
113}; 104 return 0;
105}
114 106
115static int __init ledtrig_cpu_init(void) 107static int __init ledtrig_cpu_init(void)
116{ 108{
117 int cpu; 109 int cpu;
110 int ret;
118 111
119 /* Supports up to 9999 cpu cores */ 112 /* Supports up to 9999 cpu cores */
120 BUILD_BUG_ON(CONFIG_NR_CPUS > 9999); 113 BUILD_BUG_ON(CONFIG_NR_CPUS > 9999);
@@ -133,7 +126,12 @@ static int __init ledtrig_cpu_init(void)
133 } 126 }
134 127
135 register_syscore_ops(&ledtrig_cpu_syscore_ops); 128 register_syscore_ops(&ledtrig_cpu_syscore_ops);
136 register_cpu_notifier(&ledtrig_cpu_nb); 129
130 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "AP_LEDTRIG_STARTING",
131 ledtrig_online_cpu, ledtrig_prepare_down_cpu);
132 if (ret < 0)
133 pr_err("CPU hotplug notifier for ledtrig-cpu could not be registered: %d\n",
134 ret);
137 135
138 pr_info("ledtrig-cpu: registered to indicate activity on CPUs\n"); 136 pr_info("ledtrig-cpu: registered to indicate activity on CPUs\n");
139 137
diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
index 8e4d7f590b06..6ccb994bdfcb 100644
--- a/drivers/perf/arm_pmu.c
+++ b/drivers/perf/arm_pmu.c
@@ -688,30 +688,29 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
688 return 0; 688 return 0;
689} 689}
690 690
691static DEFINE_MUTEX(arm_pmu_mutex);
692static LIST_HEAD(arm_pmu_list);
693
691/* 694/*
692 * PMU hardware loses all context when a CPU goes offline. 695 * PMU hardware loses all context when a CPU goes offline.
693 * When a CPU is hotplugged back in, since some hardware registers are 696 * When a CPU is hotplugged back in, since some hardware registers are
694 * UNKNOWN at reset, the PMU must be explicitly reset to avoid reading 697 * UNKNOWN at reset, the PMU must be explicitly reset to avoid reading
695 * junk values out of them. 698 * junk values out of them.
696 */ 699 */
697static int cpu_pmu_notify(struct notifier_block *b, unsigned long action, 700static int arm_perf_starting_cpu(unsigned int cpu)
698 void *hcpu)
699{ 701{
700 int cpu = (unsigned long)hcpu; 702 struct arm_pmu *pmu;
701 struct arm_pmu *pmu = container_of(b, struct arm_pmu, hotplug_nb);
702
703 if ((action & ~CPU_TASKS_FROZEN) != CPU_STARTING)
704 return NOTIFY_DONE;
705
706 if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
707 return NOTIFY_DONE;
708 703
709 if (pmu->reset) 704 mutex_lock(&arm_pmu_mutex);
710 pmu->reset(pmu); 705 list_for_each_entry(pmu, &arm_pmu_list, entry) {
711 else
712 return NOTIFY_DONE;
713 706
714 return NOTIFY_OK; 707 if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
708 continue;
709 if (pmu->reset)
710 pmu->reset(pmu);
711 }
712 mutex_unlock(&arm_pmu_mutex);
713 return 0;
715} 714}
716 715
717#ifdef CONFIG_CPU_PM 716#ifdef CONFIG_CPU_PM
@@ -822,10 +821,9 @@ static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
822 if (!cpu_hw_events) 821 if (!cpu_hw_events)
823 return -ENOMEM; 822 return -ENOMEM;
824 823
825 cpu_pmu->hotplug_nb.notifier_call = cpu_pmu_notify; 824 mutex_lock(&arm_pmu_mutex);
826 err = register_cpu_notifier(&cpu_pmu->hotplug_nb); 825 list_add_tail(&cpu_pmu->entry, &arm_pmu_list);
827 if (err) 826 mutex_unlock(&arm_pmu_mutex);
828 goto out_hw_events;
829 827
830 err = cpu_pm_pmu_register(cpu_pmu); 828 err = cpu_pm_pmu_register(cpu_pmu);
831 if (err) 829 if (err)
@@ -861,8 +859,9 @@ static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
861 return 0; 859 return 0;
862 860
863out_unregister: 861out_unregister:
864 unregister_cpu_notifier(&cpu_pmu->hotplug_nb); 862 mutex_lock(&arm_pmu_mutex);
865out_hw_events: 863 list_del(&cpu_pmu->entry);
864 mutex_unlock(&arm_pmu_mutex);
866 free_percpu(cpu_hw_events); 865 free_percpu(cpu_hw_events);
867 return err; 866 return err;
868} 867}
@@ -870,7 +869,9 @@ out_hw_events:
870static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu) 869static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu)
871{ 870{
872 cpu_pm_pmu_unregister(cpu_pmu); 871 cpu_pm_pmu_unregister(cpu_pmu);
873 unregister_cpu_notifier(&cpu_pmu->hotplug_nb); 872 mutex_lock(&arm_pmu_mutex);
873 list_del(&cpu_pmu->entry);
874 mutex_unlock(&arm_pmu_mutex);
874 free_percpu(cpu_pmu->hw_events); 875 free_percpu(cpu_pmu->hw_events);
875} 876}
876 877
@@ -1061,3 +1062,17 @@ out_free:
1061 kfree(pmu); 1062 kfree(pmu);
1062 return ret; 1063 return ret;
1063} 1064}
1065
1066static int arm_pmu_hp_init(void)
1067{
1068 int ret;
1069
1070 ret = cpuhp_setup_state_nocalls(CPUHP_AP_PERF_ARM_STARTING,
1071 "AP_PERF_ARM_STARTING",
1072 arm_perf_starting_cpu, NULL);
1073 if (ret)
1074 pr_err("CPU hotplug notifier for ARM PMU could not be registered: %d\n",
1075 ret);
1076 return ret;
1077}
1078subsys_initcall(arm_pmu_hp_init);
diff --git a/include/linux/cpu.h b/include/linux/cpu.h
index 21597dcac0e2..797d9c8e9a1b 100644
--- a/include/linux/cpu.h
+++ b/include/linux/cpu.h
@@ -55,17 +55,6 @@ extern ssize_t arch_cpu_release(const char *, size_t);
55#endif 55#endif
56struct notifier_block; 56struct notifier_block;
57 57
58/*
59 * CPU notifier priorities.
60 */
61enum {
62 CPU_PRI_PERF = 20,
63
64 /* bring up workqueues before normal notifiers and down after */
65 CPU_PRI_WORKQUEUE_UP = 5,
66 CPU_PRI_WORKQUEUE_DOWN = -5,
67};
68
69#define CPU_ONLINE 0x0002 /* CPU (unsigned)v is up */ 58#define CPU_ONLINE 0x0002 /* CPU (unsigned)v is up */
70#define CPU_UP_PREPARE 0x0003 /* CPU (unsigned)v coming up */ 59#define CPU_UP_PREPARE 0x0003 /* CPU (unsigned)v coming up */
71#define CPU_UP_CANCELED 0x0004 /* CPU (unsigned)v NOT coming up */ 60#define CPU_UP_CANCELED 0x0004 /* CPU (unsigned)v NOT coming up */
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index 386374d19987..242bf530edfc 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -4,19 +4,95 @@
4enum cpuhp_state { 4enum cpuhp_state {
5 CPUHP_OFFLINE, 5 CPUHP_OFFLINE,
6 CPUHP_CREATE_THREADS, 6 CPUHP_CREATE_THREADS,
7 CPUHP_PERF_PREPARE,
8 CPUHP_PERF_X86_PREPARE,
9 CPUHP_PERF_X86_UNCORE_PREP,
10 CPUHP_PERF_X86_AMD_UNCORE_PREP,
11 CPUHP_PERF_X86_RAPL_PREP,
12 CPUHP_PERF_BFIN,
13 CPUHP_PERF_POWER,
14 CPUHP_PERF_SUPERH,
15 CPUHP_X86_HPET_DEAD,
16 CPUHP_X86_APB_DEAD,
17 CPUHP_WORKQUEUE_PREP,
18 CPUHP_POWER_NUMA_PREPARE,
19 CPUHP_HRTIMERS_PREPARE,
20 CPUHP_PROFILE_PREPARE,
21 CPUHP_X2APIC_PREPARE,
22 CPUHP_SMPCFD_PREPARE,
23 CPUHP_RCUTREE_PREP,
7 CPUHP_NOTIFY_PREPARE, 24 CPUHP_NOTIFY_PREPARE,
25 CPUHP_TIMERS_DEAD,
8 CPUHP_BRINGUP_CPU, 26 CPUHP_BRINGUP_CPU,
9 CPUHP_AP_IDLE_DEAD, 27 CPUHP_AP_IDLE_DEAD,
10 CPUHP_AP_OFFLINE, 28 CPUHP_AP_OFFLINE,
11 CPUHP_AP_SCHED_STARTING, 29 CPUHP_AP_SCHED_STARTING,
30 CPUHP_AP_RCUTREE_DYING,
31 CPUHP_AP_IRQ_GIC_STARTING,
32 CPUHP_AP_IRQ_GICV3_STARTING,
33 CPUHP_AP_IRQ_HIP04_STARTING,
34 CPUHP_AP_IRQ_ARMADA_XP_STARTING,
35 CPUHP_AP_IRQ_ARMADA_CASC_STARTING,
36 CPUHP_AP_IRQ_BCM2836_STARTING,
37 CPUHP_AP_ARM_MVEBU_COHERENCY,
38 CPUHP_AP_PERF_X86_UNCORE_STARTING,
39 CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING,
40 CPUHP_AP_PERF_X86_STARTING,
41 CPUHP_AP_PERF_X86_AMD_IBS_STARTING,
42 CPUHP_AP_PERF_X86_CQM_STARTING,
43 CPUHP_AP_PERF_X86_CSTATE_STARTING,
44 CPUHP_AP_PERF_XTENSA_STARTING,
45 CPUHP_AP_PERF_METAG_STARTING,
46 CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
47 CPUHP_AP_ARM_VFP_STARTING,
48 CPUHP_AP_PERF_ARM_STARTING,
49 CPUHP_AP_ARM_L2X0_STARTING,
50 CPUHP_AP_ARM_ARCH_TIMER_STARTING,
51 CPUHP_AP_ARM_GLOBAL_TIMER_STARTING,
52 CPUHP_AP_DUMMY_TIMER_STARTING,
53 CPUHP_AP_EXYNOS4_MCT_TIMER_STARTING,
54 CPUHP_AP_ARM_TWD_STARTING,
55 CPUHP_AP_METAG_TIMER_STARTING,
56 CPUHP_AP_QCOM_TIMER_STARTING,
57 CPUHP_AP_ARMADA_TIMER_STARTING,
58 CPUHP_AP_MARCO_TIMER_STARTING,
59 CPUHP_AP_MIPS_GIC_TIMER_STARTING,
60 CPUHP_AP_ARC_TIMER_STARTING,
61 CPUHP_AP_KVM_STARTING,
62 CPUHP_AP_KVM_ARM_VGIC_INIT_STARTING,
63 CPUHP_AP_KVM_ARM_VGIC_STARTING,
64 CPUHP_AP_KVM_ARM_TIMER_STARTING,
65 CPUHP_AP_ARM_XEN_STARTING,
66 CPUHP_AP_ARM_CORESIGHT_STARTING,
67 CPUHP_AP_ARM_CORESIGHT4_STARTING,
68 CPUHP_AP_ARM64_ISNDEP_STARTING,
69 CPUHP_AP_SMPCFD_DYING,
70 CPUHP_AP_X86_TBOOT_DYING,
12 CPUHP_AP_NOTIFY_STARTING, 71 CPUHP_AP_NOTIFY_STARTING,
13 CPUHP_AP_ONLINE, 72 CPUHP_AP_ONLINE,
14 CPUHP_TEARDOWN_CPU, 73 CPUHP_TEARDOWN_CPU,
15 CPUHP_AP_ONLINE_IDLE, 74 CPUHP_AP_ONLINE_IDLE,
16 CPUHP_AP_SMPBOOT_THREADS, 75 CPUHP_AP_SMPBOOT_THREADS,
76 CPUHP_AP_X86_VDSO_VMA_ONLINE,
77 CPUHP_AP_PERF_ONLINE,
78 CPUHP_AP_PERF_X86_ONLINE,
79 CPUHP_AP_PERF_X86_UNCORE_ONLINE,
80 CPUHP_AP_PERF_X86_AMD_UNCORE_ONLINE,
81 CPUHP_AP_PERF_X86_AMD_POWER_ONLINE,
82 CPUHP_AP_PERF_X86_RAPL_ONLINE,
83 CPUHP_AP_PERF_X86_CQM_ONLINE,
84 CPUHP_AP_PERF_X86_CSTATE_ONLINE,
85 CPUHP_AP_PERF_S390_CF_ONLINE,
86 CPUHP_AP_PERF_S390_SF_ONLINE,
87 CPUHP_AP_PERF_ARM_CCI_ONLINE,
88 CPUHP_AP_PERF_ARM_CCN_ONLINE,
89 CPUHP_AP_WORKQUEUE_ONLINE,
90 CPUHP_AP_RCUTREE_ONLINE,
17 CPUHP_AP_NOTIFY_ONLINE, 91 CPUHP_AP_NOTIFY_ONLINE,
18 CPUHP_AP_ONLINE_DYN, 92 CPUHP_AP_ONLINE_DYN,
19 CPUHP_AP_ONLINE_DYN_END = CPUHP_AP_ONLINE_DYN + 30, 93 CPUHP_AP_ONLINE_DYN_END = CPUHP_AP_ONLINE_DYN + 30,
94 CPUHP_AP_X86_HPET_ONLINE,
95 CPUHP_AP_X86_KVM_CLK_ONLINE,
20 CPUHP_AP_ACTIVE, 96 CPUHP_AP_ACTIVE,
21 CPUHP_ONLINE, 97 CPUHP_ONLINE,
22}; 98};
diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index c98c6539e2c2..5e00f80b1535 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -494,4 +494,11 @@ extern void __init hrtimers_init(void);
494/* Show pending timers: */ 494/* Show pending timers: */
495extern void sysrq_timer_list_show(void); 495extern void sysrq_timer_list_show(void);
496 496
497int hrtimers_prepare_cpu(unsigned int cpu);
498#ifdef CONFIG_HOTPLUG_CPU
499int hrtimers_dead_cpu(unsigned int cpu);
500#else
501#define hrtimers_dead_cpu NULL
502#endif
503
497#endif 504#endif
diff --git a/include/linux/perf/arm_pmu.h b/include/linux/perf/arm_pmu.h
index d28ac05c7f92..e18843809eec 100644
--- a/include/linux/perf/arm_pmu.h
+++ b/include/linux/perf/arm_pmu.h
@@ -109,7 +109,7 @@ struct arm_pmu {
109 DECLARE_BITMAP(pmceid_bitmap, ARMV8_PMUV3_MAX_COMMON_EVENTS); 109 DECLARE_BITMAP(pmceid_bitmap, ARMV8_PMUV3_MAX_COMMON_EVENTS);
110 struct platform_device *plat_device; 110 struct platform_device *plat_device;
111 struct pmu_hw_events __percpu *hw_events; 111 struct pmu_hw_events __percpu *hw_events;
112 struct notifier_block hotplug_nb; 112 struct list_head entry;
113 struct notifier_block cpu_pm_nb; 113 struct notifier_block cpu_pm_nb;
114}; 114};
115 115
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index e1f921c2e4e0..8ed4326164cc 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1309,41 +1309,6 @@ static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag)
1309 1309
1310#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) 1310#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
1311 1311
1312/*
1313 * This has to have a higher priority than migration_notifier in sched/core.c.
1314 */
1315#define perf_cpu_notifier(fn) \
1316do { \
1317 static struct notifier_block fn##_nb = \
1318 { .notifier_call = fn, .priority = CPU_PRI_PERF }; \
1319 unsigned long cpu = smp_processor_id(); \
1320 unsigned long flags; \
1321 \
1322 cpu_notifier_register_begin(); \
1323 fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \
1324 (void *)(unsigned long)cpu); \
1325 local_irq_save(flags); \
1326 fn(&fn##_nb, (unsigned long)CPU_STARTING, \
1327 (void *)(unsigned long)cpu); \
1328 local_irq_restore(flags); \
1329 fn(&fn##_nb, (unsigned long)CPU_ONLINE, \
1330 (void *)(unsigned long)cpu); \
1331 __register_cpu_notifier(&fn##_nb); \
1332 cpu_notifier_register_done(); \
1333} while (0)
1334
1335/*
1336 * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the
1337 * callback for already online CPUs.
1338 */
1339#define __perf_cpu_notifier(fn) \
1340do { \
1341 static struct notifier_block fn##_nb = \
1342 { .notifier_call = fn, .priority = CPU_PRI_PERF }; \
1343 \
1344 __register_cpu_notifier(&fn##_nb); \
1345} while (0)
1346
1347struct perf_pmu_events_attr { 1312struct perf_pmu_events_attr {
1348 struct device_attribute attr; 1313 struct device_attribute attr;
1349 u64 id; 1314 u64 id;
@@ -1385,4 +1350,13 @@ _name##_show(struct device *dev, \
1385 \ 1350 \
1386static struct device_attribute format_attr_##_name = __ATTR_RO(_name) 1351static struct device_attribute format_attr_##_name = __ATTR_RO(_name)
1387 1352
1353/* Performance counter hotplug functions */
1354#ifdef CONFIG_PERF_EVENTS
1355int perf_event_init_cpu(unsigned int cpu);
1356int perf_event_exit_cpu(unsigned int cpu);
1357#else
1358#define perf_event_init_cpu NULL
1359#define perf_event_exit_cpu NULL
1360#endif
1361
1388#endif /* _LINUX_PERF_EVENT_H */ 1362#endif /* _LINUX_PERF_EVENT_H */
diff --git a/include/linux/rcutiny.h b/include/linux/rcutiny.h
index 93aea75029fb..ac81e4063b40 100644
--- a/include/linux/rcutiny.h
+++ b/include/linux/rcutiny.h
@@ -243,4 +243,11 @@ static inline void rcu_all_qs(void)
243 barrier(); /* Avoid RCU read-side critical sections leaking across. */ 243 barrier(); /* Avoid RCU read-side critical sections leaking across. */
244} 244}
245 245
246/* RCUtree hotplug events */
247#define rcutree_prepare_cpu NULL
248#define rcutree_online_cpu NULL
249#define rcutree_offline_cpu NULL
250#define rcutree_dead_cpu NULL
251#define rcutree_dying_cpu NULL
252
246#endif /* __LINUX_RCUTINY_H */ 253#endif /* __LINUX_RCUTINY_H */
diff --git a/include/linux/rcutree.h b/include/linux/rcutree.h
index 5043cb823fb2..63a4e4cf40a5 100644
--- a/include/linux/rcutree.h
+++ b/include/linux/rcutree.h
@@ -111,4 +111,11 @@ bool rcu_is_watching(void);
111 111
112void rcu_all_qs(void); 112void rcu_all_qs(void);
113 113
114/* RCUtree hotplug events */
115int rcutree_prepare_cpu(unsigned int cpu);
116int rcutree_online_cpu(unsigned int cpu);
117int rcutree_offline_cpu(unsigned int cpu);
118int rcutree_dead_cpu(unsigned int cpu);
119int rcutree_dying_cpu(unsigned int cpu);
120
114#endif /* __LINUX_RCUTREE_H */ 121#endif /* __LINUX_RCUTREE_H */
diff --git a/include/linux/smp.h b/include/linux/smp.h
index c4414074bd88..eccae4690f41 100644
--- a/include/linux/smp.h
+++ b/include/linux/smp.h
@@ -196,4 +196,9 @@ extern void arch_enable_nonboot_cpus_end(void);
196 196
197void smp_setup_processor_id(void); 197void smp_setup_processor_id(void);
198 198
199/* SMP core functions */
200int smpcfd_prepare_cpu(unsigned int cpu);
201int smpcfd_dead_cpu(unsigned int cpu);
202int smpcfd_dying_cpu(unsigned int cpu);
203
199#endif /* __LINUX_SMP_H */ 204#endif /* __LINUX_SMP_H */
diff --git a/include/linux/timer.h b/include/linux/timer.h
index 4419506b564e..51d601f192d4 100644
--- a/include/linux/timer.h
+++ b/include/linux/timer.h
@@ -273,4 +273,10 @@ unsigned long __round_jiffies_up_relative(unsigned long j, int cpu);
273unsigned long round_jiffies_up(unsigned long j); 273unsigned long round_jiffies_up(unsigned long j);
274unsigned long round_jiffies_up_relative(unsigned long j); 274unsigned long round_jiffies_up_relative(unsigned long j);
275 275
276#ifdef CONFIG_HOTPLUG_CPU
277int timers_dead_cpu(unsigned int cpu);
278#else
279#define timers_dead_cpu NULL
280#endif
281
276#endif 282#endif
diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index ca73c503b92a..26cc1df280d6 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -625,4 +625,10 @@ void wq_watchdog_touch(int cpu);
625static inline void wq_watchdog_touch(int cpu) { } 625static inline void wq_watchdog_touch(int cpu) { }
626#endif /* CONFIG_WQ_WATCHDOG */ 626#endif /* CONFIG_WQ_WATCHDOG */
627 627
628#ifdef CONFIG_SMP
629int workqueue_prepare_cpu(unsigned int cpu);
630int workqueue_online_cpu(unsigned int cpu);
631int workqueue_offline_cpu(unsigned int cpu);
632#endif
633
628#endif 634#endif
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 7b61887f7ccd..341bf80f80bd 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -517,6 +517,13 @@ static int cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state,
517 if (!cpu_online(cpu)) 517 if (!cpu_online(cpu))
518 return 0; 518 return 0;
519 519
520 /*
521 * If we are up and running, use the hotplug thread. For early calls
522 * we invoke the thread function directly.
523 */
524 if (!st->thread)
525 return cpuhp_invoke_callback(cpu, state, cb);
526
520 st->cb_state = state; 527 st->cb_state = state;
521 st->cb = cb; 528 st->cb = cb;
522 /* 529 /*
@@ -1173,6 +1180,31 @@ static struct cpuhp_step cpuhp_bp_states[] = {
1173 .teardown = NULL, 1180 .teardown = NULL,
1174 .cant_stop = true, 1181 .cant_stop = true,
1175 }, 1182 },
1183 [CPUHP_PERF_PREPARE] = {
1184 .name = "perf prepare",
1185 .startup = perf_event_init_cpu,
1186 .teardown = perf_event_exit_cpu,
1187 },
1188 [CPUHP_WORKQUEUE_PREP] = {
1189 .name = "workqueue prepare",
1190 .startup = workqueue_prepare_cpu,
1191 .teardown = NULL,
1192 },
1193 [CPUHP_HRTIMERS_PREPARE] = {
1194 .name = "hrtimers prepare",
1195 .startup = hrtimers_prepare_cpu,
1196 .teardown = hrtimers_dead_cpu,
1197 },
1198 [CPUHP_SMPCFD_PREPARE] = {
1199 .name = "SMPCFD prepare",
1200 .startup = smpcfd_prepare_cpu,
1201 .teardown = smpcfd_dead_cpu,
1202 },
1203 [CPUHP_RCUTREE_PREP] = {
1204 .name = "RCU-tree prepare",
1205 .startup = rcutree_prepare_cpu,
1206 .teardown = rcutree_dead_cpu,
1207 },
1176 /* 1208 /*
1177 * Preparatory and dead notifiers. Will be replaced once the notifiers 1209 * Preparatory and dead notifiers. Will be replaced once the notifiers
1178 * are converted to states. 1210 * are converted to states.
@@ -1184,6 +1216,16 @@ static struct cpuhp_step cpuhp_bp_states[] = {
1184 .skip_onerr = true, 1216 .skip_onerr = true,
1185 .cant_stop = true, 1217 .cant_stop = true,
1186 }, 1218 },
1219 /*
1220 * On the tear-down path, timers_dead_cpu() must be invoked
1221 * before blk_mq_queue_reinit_notify() from notify_dead(),
1222 * otherwise a RCU stall occurs.
1223 */
1224 [CPUHP_TIMERS_DEAD] = {
1225 .name = "timers dead",
1226 .startup = NULL,
1227 .teardown = timers_dead_cpu,
1228 },
1187 /* Kicks the plugged cpu into life */ 1229 /* Kicks the plugged cpu into life */
1188 [CPUHP_BRINGUP_CPU] = { 1230 [CPUHP_BRINGUP_CPU] = {
1189 .name = "cpu:bringup", 1231 .name = "cpu:bringup",
@@ -1191,6 +1233,10 @@ static struct cpuhp_step cpuhp_bp_states[] = {
1191 .teardown = NULL, 1233 .teardown = NULL,
1192 .cant_stop = true, 1234 .cant_stop = true,
1193 }, 1235 },
1236 [CPUHP_AP_SMPCFD_DYING] = {
1237 .startup = NULL,
1238 .teardown = smpcfd_dying_cpu,
1239 },
1194 /* 1240 /*
1195 * Handled on controll processor until the plugged processor manages 1241 * Handled on controll processor until the plugged processor manages
1196 * this itself. 1242 * this itself.
@@ -1227,6 +1273,10 @@ static struct cpuhp_step cpuhp_ap_states[] = {
1227 .startup = sched_cpu_starting, 1273 .startup = sched_cpu_starting,
1228 .teardown = sched_cpu_dying, 1274 .teardown = sched_cpu_dying,
1229 }, 1275 },
1276 [CPUHP_AP_RCUTREE_DYING] = {
1277 .startup = NULL,
1278 .teardown = rcutree_dying_cpu,
1279 },
1230 /* 1280 /*
1231 * Low level startup/teardown notifiers. Run with interrupts 1281 * Low level startup/teardown notifiers. Run with interrupts
1232 * disabled. Will be removed once the notifiers are converted to 1282 * disabled. Will be removed once the notifiers are converted to
@@ -1250,6 +1300,22 @@ static struct cpuhp_step cpuhp_ap_states[] = {
1250 .startup = smpboot_unpark_threads, 1300 .startup = smpboot_unpark_threads,
1251 .teardown = NULL, 1301 .teardown = NULL,
1252 }, 1302 },
1303 [CPUHP_AP_PERF_ONLINE] = {
1304 .name = "perf online",
1305 .startup = perf_event_init_cpu,
1306 .teardown = perf_event_exit_cpu,
1307 },
1308 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1309 .name = "workqueue online",
1310 .startup = workqueue_online_cpu,
1311 .teardown = workqueue_offline_cpu,
1312 },
1313 [CPUHP_AP_RCUTREE_ONLINE] = {
1314 .name = "RCU-tree online",
1315 .startup = rcutree_online_cpu,
1316 .teardown = rcutree_offline_cpu,
1317 },
1318
1253 /* 1319 /*
1254 * Online/down_prepare notifiers. Will be removed once the notifiers 1320 * Online/down_prepare notifiers. Will be removed once the notifiers
1255 * are converted to states. 1321 * are converted to states.
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 09ae27b353c1..356a6c7cb52a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -10357,7 +10357,7 @@ static void __init perf_event_init_all_cpus(void)
10357 } 10357 }
10358} 10358}
10359 10359
10360static void perf_event_init_cpu(int cpu) 10360int perf_event_init_cpu(unsigned int cpu)
10361{ 10361{
10362 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu); 10362 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
10363 10363
@@ -10370,6 +10370,7 @@ static void perf_event_init_cpu(int cpu)
10370 rcu_assign_pointer(swhash->swevent_hlist, hlist); 10370 rcu_assign_pointer(swhash->swevent_hlist, hlist);
10371 } 10371 }
10372 mutex_unlock(&swhash->hlist_mutex); 10372 mutex_unlock(&swhash->hlist_mutex);
10373 return 0;
10373} 10374}
10374 10375
10375#if defined CONFIG_HOTPLUG_CPU || defined CONFIG_KEXEC_CORE 10376#if defined CONFIG_HOTPLUG_CPU || defined CONFIG_KEXEC_CORE
@@ -10401,14 +10402,17 @@ static void perf_event_exit_cpu_context(int cpu)
10401 } 10402 }
10402 srcu_read_unlock(&pmus_srcu, idx); 10403 srcu_read_unlock(&pmus_srcu, idx);
10403} 10404}
10405#else
10406
10407static void perf_event_exit_cpu_context(int cpu) { }
10408
10409#endif
10404 10410
10405static void perf_event_exit_cpu(int cpu) 10411int perf_event_exit_cpu(unsigned int cpu)
10406{ 10412{
10407 perf_event_exit_cpu_context(cpu); 10413 perf_event_exit_cpu_context(cpu);
10414 return 0;
10408} 10415}
10409#else
10410static inline void perf_event_exit_cpu(int cpu) { }
10411#endif
10412 10416
10413static int 10417static int
10414perf_reboot(struct notifier_block *notifier, unsigned long val, void *v) 10418perf_reboot(struct notifier_block *notifier, unsigned long val, void *v)
@@ -10430,46 +10434,6 @@ static struct notifier_block perf_reboot_notifier = {
10430 .priority = INT_MIN, 10434 .priority = INT_MIN,
10431}; 10435};
10432 10436
10433static int
10434perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
10435{
10436 unsigned int cpu = (long)hcpu;
10437
10438 switch (action & ~CPU_TASKS_FROZEN) {
10439
10440 case CPU_UP_PREPARE:
10441 /*
10442 * This must be done before the CPU comes alive, because the
10443 * moment we can run tasks we can encounter (software) events.
10444 *
10445 * Specifically, someone can have inherited events on kthreadd
10446 * or a pre-existing worker thread that gets re-bound.
10447 */
10448 perf_event_init_cpu(cpu);
10449 break;
10450
10451 case CPU_DOWN_PREPARE:
10452 /*
10453 * This must be done before the CPU dies because after that an
10454 * active event might want to IPI the CPU and that'll not work
10455 * so great for dead CPUs.
10456 *
10457 * XXX smp_call_function_single() return -ENXIO without a warn
10458 * so we could possibly deal with this.
10459 *
10460 * This is safe against new events arriving because
10461 * sys_perf_event_open() serializes against hotplug using
10462 * get_online_cpus().
10463 */
10464 perf_event_exit_cpu(cpu);
10465 break;
10466 default:
10467 break;
10468 }
10469
10470 return NOTIFY_OK;
10471}
10472
10473void __init perf_event_init(void) 10437void __init perf_event_init(void)
10474{ 10438{
10475 int ret; 10439 int ret;
@@ -10482,7 +10446,7 @@ void __init perf_event_init(void)
10482 perf_pmu_register(&perf_cpu_clock, NULL, -1); 10446 perf_pmu_register(&perf_cpu_clock, NULL, -1);
10483 perf_pmu_register(&perf_task_clock, NULL, -1); 10447 perf_pmu_register(&perf_task_clock, NULL, -1);
10484 perf_tp_register(); 10448 perf_tp_register();
10485 perf_cpu_notifier(perf_cpu_notify); 10449 perf_event_init_cpu(smp_processor_id());
10486 register_reboot_notifier(&perf_reboot_notifier); 10450 register_reboot_notifier(&perf_reboot_notifier);
10487 10451
10488 ret = init_hw_breakpoint(); 10452 ret = init_hw_breakpoint();
diff --git a/kernel/profile.c b/kernel/profile.c
index c2199e9901c9..2dbccf2d806c 100644
--- a/kernel/profile.c
+++ b/kernel/profile.c
@@ -328,68 +328,57 @@ out:
328 put_cpu(); 328 put_cpu();
329} 329}
330 330
331static int profile_cpu_callback(struct notifier_block *info, 331static int profile_dead_cpu(unsigned int cpu)
332 unsigned long action, void *__cpu)
333{ 332{
334 int node, cpu = (unsigned long)__cpu;
335 struct page *page; 333 struct page *page;
334 int i;
336 335
337 switch (action) { 336 if (prof_cpu_mask != NULL)
338 case CPU_UP_PREPARE: 337 cpumask_clear_cpu(cpu, prof_cpu_mask);
339 case CPU_UP_PREPARE_FROZEN: 338
340 node = cpu_to_mem(cpu); 339 for (i = 0; i < 2; i++) {
341 per_cpu(cpu_profile_flip, cpu) = 0; 340 if (per_cpu(cpu_profile_hits, cpu)[i]) {
342 if (!per_cpu(cpu_profile_hits, cpu)[1]) { 341 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[i]);
343 page = __alloc_pages_node(node, 342 per_cpu(cpu_profile_hits, cpu)[i] = NULL;
344 GFP_KERNEL | __GFP_ZERO,
345 0);
346 if (!page)
347 return notifier_from_errno(-ENOMEM);
348 per_cpu(cpu_profile_hits, cpu)[1] = page_address(page);
349 }
350 if (!per_cpu(cpu_profile_hits, cpu)[0]) {
351 page = __alloc_pages_node(node,
352 GFP_KERNEL | __GFP_ZERO,
353 0);
354 if (!page)
355 goto out_free;
356 per_cpu(cpu_profile_hits, cpu)[0] = page_address(page);
357 }
358 break;
359out_free:
360 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[1]);
361 per_cpu(cpu_profile_hits, cpu)[1] = NULL;
362 __free_page(page);
363 return notifier_from_errno(-ENOMEM);
364 case CPU_ONLINE:
365 case CPU_ONLINE_FROZEN:
366 if (prof_cpu_mask != NULL)
367 cpumask_set_cpu(cpu, prof_cpu_mask);
368 break;
369 case CPU_UP_CANCELED:
370 case CPU_UP_CANCELED_FROZEN:
371 case CPU_DEAD:
372 case CPU_DEAD_FROZEN:
373 if (prof_cpu_mask != NULL)
374 cpumask_clear_cpu(cpu, prof_cpu_mask);
375 if (per_cpu(cpu_profile_hits, cpu)[0]) {
376 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[0]);
377 per_cpu(cpu_profile_hits, cpu)[0] = NULL;
378 __free_page(page); 343 __free_page(page);
379 } 344 }
380 if (per_cpu(cpu_profile_hits, cpu)[1]) { 345 }
381 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[1]); 346 return 0;
382 per_cpu(cpu_profile_hits, cpu)[1] = NULL; 347}
383 __free_page(page); 348
349static int profile_prepare_cpu(unsigned int cpu)
350{
351 int i, node = cpu_to_mem(cpu);
352 struct page *page;
353
354 per_cpu(cpu_profile_flip, cpu) = 0;
355
356 for (i = 0; i < 2; i++) {
357 if (per_cpu(cpu_profile_hits, cpu)[i])
358 continue;
359
360 page = __alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
361 if (!page) {
362 profile_dead_cpu(cpu);
363 return -ENOMEM;
384 } 364 }
385 break; 365 per_cpu(cpu_profile_hits, cpu)[i] = page_address(page);
366
386 } 367 }
387 return NOTIFY_OK; 368 return 0;
369}
370
371static int profile_online_cpu(unsigned int cpu)
372{
373 if (prof_cpu_mask != NULL)
374 cpumask_set_cpu(cpu, prof_cpu_mask);
375
376 return 0;
388} 377}
378
389#else /* !CONFIG_SMP */ 379#else /* !CONFIG_SMP */
390#define profile_flip_buffers() do { } while (0) 380#define profile_flip_buffers() do { } while (0)
391#define profile_discard_flip_buffers() do { } while (0) 381#define profile_discard_flip_buffers() do { } while (0)
392#define profile_cpu_callback NULL
393 382
394static void do_profile_hits(int type, void *__pc, unsigned int nr_hits) 383static void do_profile_hits(int type, void *__pc, unsigned int nr_hits)
395{ 384{
@@ -531,83 +520,43 @@ static const struct file_operations proc_profile_operations = {
531 .llseek = default_llseek, 520 .llseek = default_llseek,
532}; 521};
533 522
534#ifdef CONFIG_SMP 523int __ref create_proc_profile(void)
535static void profile_nop(void *unused)
536{
537}
538
539static int create_hash_tables(void)
540{ 524{
541 int cpu; 525 struct proc_dir_entry *entry;
542 526#ifdef CONFIG_SMP
543 for_each_online_cpu(cpu) { 527 enum cpuhp_state online_state;
544 int node = cpu_to_mem(cpu);
545 struct page *page;
546
547 page = __alloc_pages_node(node,
548 GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
549 0);
550 if (!page)
551 goto out_cleanup;
552 per_cpu(cpu_profile_hits, cpu)[1]
553 = (struct profile_hit *)page_address(page);
554 page = __alloc_pages_node(node,
555 GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
556 0);
557 if (!page)
558 goto out_cleanup;
559 per_cpu(cpu_profile_hits, cpu)[0]
560 = (struct profile_hit *)page_address(page);
561 }
562 return 0;
563out_cleanup:
564 prof_on = 0;
565 smp_mb();
566 on_each_cpu(profile_nop, NULL, 1);
567 for_each_online_cpu(cpu) {
568 struct page *page;
569
570 if (per_cpu(cpu_profile_hits, cpu)[0]) {
571 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[0]);
572 per_cpu(cpu_profile_hits, cpu)[0] = NULL;
573 __free_page(page);
574 }
575 if (per_cpu(cpu_profile_hits, cpu)[1]) {
576 page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[1]);
577 per_cpu(cpu_profile_hits, cpu)[1] = NULL;
578 __free_page(page);
579 }
580 }
581 return -1;
582}
583#else
584#define create_hash_tables() ({ 0; })
585#endif 528#endif
586 529
587int __ref create_proc_profile(void) /* false positive from hotcpu_notifier */
588{
589 struct proc_dir_entry *entry;
590 int err = 0; 530 int err = 0;
591 531
592 if (!prof_on) 532 if (!prof_on)
593 return 0; 533 return 0;
594 534#ifdef CONFIG_SMP
595 cpu_notifier_register_begin(); 535 err = cpuhp_setup_state(CPUHP_PROFILE_PREPARE, "PROFILE_PREPARE",
596 536 profile_prepare_cpu, profile_dead_cpu);
597 if (create_hash_tables()) { 537 if (err)
598 err = -ENOMEM; 538 return err;
599 goto out; 539
600 } 540 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "AP_PROFILE_ONLINE",
601 541 profile_online_cpu, NULL);
542 if (err < 0)
543 goto err_state_prep;
544 online_state = err;
545 err = 0;
546#endif
602 entry = proc_create("profile", S_IWUSR | S_IRUGO, 547 entry = proc_create("profile", S_IWUSR | S_IRUGO,
603 NULL, &proc_profile_operations); 548 NULL, &proc_profile_operations);
604 if (!entry) 549 if (!entry)
605 goto out; 550 goto err_state_onl;
606 proc_set_size(entry, (1 + prof_len) * sizeof(atomic_t)); 551 proc_set_size(entry, (1 + prof_len) * sizeof(atomic_t));
607 __hotcpu_notifier(profile_cpu_callback, 0);
608 552
609out: 553 return err;
610 cpu_notifier_register_done(); 554err_state_onl:
555#ifdef CONFIG_SMP
556 cpuhp_remove_state(online_state);
557err_state_prep:
558 cpuhp_remove_state(CPUHP_PROFILE_PREPARE);
559#endif
611 return err; 560 return err;
612} 561}
613subsys_initcall(create_proc_profile); 562subsys_initcall(create_proc_profile);
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index f433959e9322..5d80925e7fc8 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1073,11 +1073,11 @@ EXPORT_SYMBOL_GPL(rcu_is_watching);
1073 * offline to continue to use RCU for one jiffy after marking itself 1073 * offline to continue to use RCU for one jiffy after marking itself
1074 * offline in the cpu_online_mask. This leniency is necessary given the 1074 * offline in the cpu_online_mask. This leniency is necessary given the
1075 * non-atomic nature of the online and offline processing, for example, 1075 * non-atomic nature of the online and offline processing, for example,
1076 * the fact that a CPU enters the scheduler after completing the CPU_DYING 1076 * the fact that a CPU enters the scheduler after completing the teardown
1077 * notifiers. 1077 * of the CPU.
1078 * 1078 *
1079 * This is also why RCU internally marks CPUs online during the 1079 * This is also why RCU internally marks CPUs online during in the
1080 * CPU_UP_PREPARE phase and offline during the CPU_DEAD phase. 1080 * preparation phase and offline after the CPU has been taken down.
1081 * 1081 *
1082 * Disable checking if in an NMI handler because we cannot safely report 1082 * Disable checking if in an NMI handler because we cannot safely report
1083 * errors from NMI handlers anyway. 1083 * errors from NMI handlers anyway.
@@ -3806,12 +3806,58 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3806 raw_spin_unlock_irqrestore_rcu_node(rnp, flags); 3806 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3807} 3807}
3808 3808
3809static void rcu_prepare_cpu(int cpu) 3809int rcutree_prepare_cpu(unsigned int cpu)
3810{ 3810{
3811 struct rcu_state *rsp; 3811 struct rcu_state *rsp;
3812 3812
3813 for_each_rcu_flavor(rsp) 3813 for_each_rcu_flavor(rsp)
3814 rcu_init_percpu_data(cpu, rsp); 3814 rcu_init_percpu_data(cpu, rsp);
3815
3816 rcu_prepare_kthreads(cpu);
3817 rcu_spawn_all_nocb_kthreads(cpu);
3818
3819 return 0;
3820}
3821
3822static void rcutree_affinity_setting(unsigned int cpu, int outgoing)
3823{
3824 struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3825
3826 rcu_boost_kthread_setaffinity(rdp->mynode, outgoing);
3827}
3828
3829int rcutree_online_cpu(unsigned int cpu)
3830{
3831 sync_sched_exp_online_cleanup(cpu);
3832 rcutree_affinity_setting(cpu, -1);
3833 return 0;
3834}
3835
3836int rcutree_offline_cpu(unsigned int cpu)
3837{
3838 rcutree_affinity_setting(cpu, cpu);
3839 return 0;
3840}
3841
3842
3843int rcutree_dying_cpu(unsigned int cpu)
3844{
3845 struct rcu_state *rsp;
3846
3847 for_each_rcu_flavor(rsp)
3848 rcu_cleanup_dying_cpu(rsp);
3849 return 0;
3850}
3851
3852int rcutree_dead_cpu(unsigned int cpu)
3853{
3854 struct rcu_state *rsp;
3855
3856 for_each_rcu_flavor(rsp) {
3857 rcu_cleanup_dead_cpu(cpu, rsp);
3858 do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu));
3859 }
3860 return 0;
3815} 3861}
3816 3862
3817#ifdef CONFIG_HOTPLUG_CPU 3863#ifdef CONFIG_HOTPLUG_CPU
@@ -3851,52 +3897,6 @@ void rcu_report_dead(unsigned int cpu)
3851} 3897}
3852#endif 3898#endif
3853 3899
3854/*
3855 * Handle CPU online/offline notification events.
3856 */
3857int rcu_cpu_notify(struct notifier_block *self,
3858 unsigned long action, void *hcpu)
3859{
3860 long cpu = (long)hcpu;
3861 struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3862 struct rcu_node *rnp = rdp->mynode;
3863 struct rcu_state *rsp;
3864
3865 switch (action) {
3866 case CPU_UP_PREPARE:
3867 case CPU_UP_PREPARE_FROZEN:
3868 rcu_prepare_cpu(cpu);
3869 rcu_prepare_kthreads(cpu);
3870 rcu_spawn_all_nocb_kthreads(cpu);
3871 break;
3872 case CPU_ONLINE:
3873 case CPU_DOWN_FAILED:
3874 sync_sched_exp_online_cleanup(cpu);
3875 rcu_boost_kthread_setaffinity(rnp, -1);
3876 break;
3877 case CPU_DOWN_PREPARE:
3878 rcu_boost_kthread_setaffinity(rnp, cpu);
3879 break;
3880 case CPU_DYING:
3881 case CPU_DYING_FROZEN:
3882 for_each_rcu_flavor(rsp)
3883 rcu_cleanup_dying_cpu(rsp);
3884 break;
3885 case CPU_DEAD:
3886 case CPU_DEAD_FROZEN:
3887 case CPU_UP_CANCELED:
3888 case CPU_UP_CANCELED_FROZEN:
3889 for_each_rcu_flavor(rsp) {
3890 rcu_cleanup_dead_cpu(cpu, rsp);
3891 do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu));
3892 }
3893 break;
3894 default:
3895 break;
3896 }
3897 return NOTIFY_OK;
3898}
3899
3900static int rcu_pm_notify(struct notifier_block *self, 3900static int rcu_pm_notify(struct notifier_block *self,
3901 unsigned long action, void *hcpu) 3901 unsigned long action, void *hcpu)
3902{ 3902{
@@ -4208,10 +4208,9 @@ void __init rcu_init(void)
4208 * this is called early in boot, before either interrupts 4208 * this is called early in boot, before either interrupts
4209 * or the scheduler are operational. 4209 * or the scheduler are operational.
4210 */ 4210 */
4211 cpu_notifier(rcu_cpu_notify, 0);
4212 pm_notifier(rcu_pm_notify, 0); 4211 pm_notifier(rcu_pm_notify, 0);
4213 for_each_online_cpu(cpu) 4212 for_each_online_cpu(cpu)
4214 rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); 4213 rcutree_prepare_cpu(cpu);
4215} 4214}
4216 4215
4217#include "tree_exp.h" 4216#include "tree_exp.h"
diff --git a/kernel/smp.c b/kernel/smp.c
index 36552beed397..3aa642d39c03 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -33,69 +33,54 @@ static DEFINE_PER_CPU_SHARED_ALIGNED(struct llist_head, call_single_queue);
33 33
34static void flush_smp_call_function_queue(bool warn_cpu_offline); 34static void flush_smp_call_function_queue(bool warn_cpu_offline);
35 35
36static int 36int smpcfd_prepare_cpu(unsigned int cpu)
37hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu)
38{ 37{
39 long cpu = (long)hcpu;
40 struct call_function_data *cfd = &per_cpu(cfd_data, cpu); 38 struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
41 39
42 switch (action) { 40 if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL,
43 case CPU_UP_PREPARE: 41 cpu_to_node(cpu)))
44 case CPU_UP_PREPARE_FROZEN: 42 return -ENOMEM;
45 if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL, 43 cfd->csd = alloc_percpu(struct call_single_data);
46 cpu_to_node(cpu))) 44 if (!cfd->csd) {
47 return notifier_from_errno(-ENOMEM);
48 cfd->csd = alloc_percpu(struct call_single_data);
49 if (!cfd->csd) {
50 free_cpumask_var(cfd->cpumask);
51 return notifier_from_errno(-ENOMEM);
52 }
53 break;
54
55#ifdef CONFIG_HOTPLUG_CPU
56 case CPU_UP_CANCELED:
57 case CPU_UP_CANCELED_FROZEN:
58 /* Fall-through to the CPU_DEAD[_FROZEN] case. */
59
60 case CPU_DEAD:
61 case CPU_DEAD_FROZEN:
62 free_cpumask_var(cfd->cpumask); 45 free_cpumask_var(cfd->cpumask);
63 free_percpu(cfd->csd); 46 return -ENOMEM;
64 break; 47 }
65 48
66 case CPU_DYING: 49 return 0;
67 case CPU_DYING_FROZEN: 50}
68 /* 51
69 * The IPIs for the smp-call-function callbacks queued by other 52int smpcfd_dead_cpu(unsigned int cpu)
70 * CPUs might arrive late, either due to hardware latencies or 53{
71 * because this CPU disabled interrupts (inside stop-machine) 54 struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
72 * before the IPIs were sent. So flush out any pending callbacks
73 * explicitly (without waiting for the IPIs to arrive), to
74 * ensure that the outgoing CPU doesn't go offline with work
75 * still pending.
76 */
77 flush_smp_call_function_queue(false);
78 break;
79#endif
80 };
81 55
82 return NOTIFY_OK; 56 free_cpumask_var(cfd->cpumask);
57 free_percpu(cfd->csd);
58 return 0;
83} 59}
84 60
85static struct notifier_block hotplug_cfd_notifier = { 61int smpcfd_dying_cpu(unsigned int cpu)
86 .notifier_call = hotplug_cfd, 62{
87}; 63 /*
64 * The IPIs for the smp-call-function callbacks queued by other
65 * CPUs might arrive late, either due to hardware latencies or
66 * because this CPU disabled interrupts (inside stop-machine)
67 * before the IPIs were sent. So flush out any pending callbacks
68 * explicitly (without waiting for the IPIs to arrive), to
69 * ensure that the outgoing CPU doesn't go offline with work
70 * still pending.
71 */
72 flush_smp_call_function_queue(false);
73 return 0;
74}
88 75
89void __init call_function_init(void) 76void __init call_function_init(void)
90{ 77{
91 void *cpu = (void *)(long)smp_processor_id();
92 int i; 78 int i;
93 79
94 for_each_possible_cpu(i) 80 for_each_possible_cpu(i)
95 init_llist_head(&per_cpu(call_single_queue, i)); 81 init_llist_head(&per_cpu(call_single_queue, i));
96 82
97 hotplug_cfd(&hotplug_cfd_notifier, CPU_UP_PREPARE, cpu); 83 smpcfd_prepare_cpu(smp_processor_id());
98 register_cpu_notifier(&hotplug_cfd_notifier);
99} 84}
100 85
101/* 86/*
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index d13c9aebf7a3..9ba7c820fc23 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1590,7 +1590,7 @@ SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1590/* 1590/*
1591 * Functions related to boot-time initialization: 1591 * Functions related to boot-time initialization:
1592 */ 1592 */
1593static void init_hrtimers_cpu(int cpu) 1593int hrtimers_prepare_cpu(unsigned int cpu)
1594{ 1594{
1595 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); 1595 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1596 int i; 1596 int i;
@@ -1602,6 +1602,7 @@ static void init_hrtimers_cpu(int cpu)
1602 1602
1603 cpu_base->cpu = cpu; 1603 cpu_base->cpu = cpu;
1604 hrtimer_init_hres(cpu_base); 1604 hrtimer_init_hres(cpu_base);
1605 return 0;
1605} 1606}
1606 1607
1607#ifdef CONFIG_HOTPLUG_CPU 1608#ifdef CONFIG_HOTPLUG_CPU
@@ -1636,7 +1637,7 @@ static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1636 } 1637 }
1637} 1638}
1638 1639
1639static void migrate_hrtimers(int scpu) 1640int hrtimers_dead_cpu(unsigned int scpu)
1640{ 1641{
1641 struct hrtimer_cpu_base *old_base, *new_base; 1642 struct hrtimer_cpu_base *old_base, *new_base;
1642 int i; 1643 int i;
@@ -1665,45 +1666,14 @@ static void migrate_hrtimers(int scpu)
1665 /* Check, if we got expired work to do */ 1666 /* Check, if we got expired work to do */
1666 __hrtimer_peek_ahead_timers(); 1667 __hrtimer_peek_ahead_timers();
1667 local_irq_enable(); 1668 local_irq_enable();
1669 return 0;
1668} 1670}
1669 1671
1670#endif /* CONFIG_HOTPLUG_CPU */ 1672#endif /* CONFIG_HOTPLUG_CPU */
1671 1673
1672static int hrtimer_cpu_notify(struct notifier_block *self,
1673 unsigned long action, void *hcpu)
1674{
1675 int scpu = (long)hcpu;
1676
1677 switch (action) {
1678
1679 case CPU_UP_PREPARE:
1680 case CPU_UP_PREPARE_FROZEN:
1681 init_hrtimers_cpu(scpu);
1682 break;
1683
1684#ifdef CONFIG_HOTPLUG_CPU
1685 case CPU_DEAD:
1686 case CPU_DEAD_FROZEN:
1687 migrate_hrtimers(scpu);
1688 break;
1689#endif
1690
1691 default:
1692 break;
1693 }
1694
1695 return NOTIFY_OK;
1696}
1697
1698static struct notifier_block hrtimers_nb = {
1699 .notifier_call = hrtimer_cpu_notify,
1700};
1701
1702void __init hrtimers_init(void) 1674void __init hrtimers_init(void)
1703{ 1675{
1704 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, 1676 hrtimers_prepare_cpu(smp_processor_id());
1705 (void *)(long)smp_processor_id());
1706 register_cpu_notifier(&hrtimers_nb);
1707} 1677}
1708 1678
1709/** 1679/**
diff --git a/kernel/time/timer.c b/kernel/time/timer.c
index cb9ab401e2d9..555670a5143c 100644
--- a/kernel/time/timer.c
+++ b/kernel/time/timer.c
@@ -1804,7 +1804,7 @@ static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *h
1804 } 1804 }
1805} 1805}
1806 1806
1807static void migrate_timers(int cpu) 1807int timers_dead_cpu(unsigned int cpu)
1808{ 1808{
1809 struct timer_base *old_base; 1809 struct timer_base *old_base;
1810 struct timer_base *new_base; 1810 struct timer_base *new_base;
@@ -1831,29 +1831,9 @@ static void migrate_timers(int cpu)
1831 spin_unlock_irq(&new_base->lock); 1831 spin_unlock_irq(&new_base->lock);
1832 put_cpu_ptr(&timer_bases); 1832 put_cpu_ptr(&timer_bases);
1833 } 1833 }
1834 return 0;
1834} 1835}
1835 1836
1836static int timer_cpu_notify(struct notifier_block *self,
1837 unsigned long action, void *hcpu)
1838{
1839 switch (action) {
1840 case CPU_DEAD:
1841 case CPU_DEAD_FROZEN:
1842 migrate_timers((long)hcpu);
1843 break;
1844 default:
1845 break;
1846 }
1847
1848 return NOTIFY_OK;
1849}
1850
1851static inline void timer_register_cpu_notifier(void)
1852{
1853 cpu_notifier(timer_cpu_notify, 0);
1854}
1855#else
1856static inline void timer_register_cpu_notifier(void) { }
1857#endif /* CONFIG_HOTPLUG_CPU */ 1837#endif /* CONFIG_HOTPLUG_CPU */
1858 1838
1859static void __init init_timer_cpu(int cpu) 1839static void __init init_timer_cpu(int cpu)
@@ -1881,7 +1861,6 @@ void __init init_timers(void)
1881{ 1861{
1882 init_timer_cpus(); 1862 init_timer_cpus();
1883 init_timer_stats(); 1863 init_timer_stats();
1884 timer_register_cpu_notifier();
1885 open_softirq(TIMER_SOFTIRQ, run_timer_softirq); 1864 open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1886} 1865}
1887 1866
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index d12bd958077e..ef071ca73fc3 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -4607,84 +4607,65 @@ static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu)
4607 WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, &cpumask) < 0); 4607 WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, &cpumask) < 0);
4608} 4608}
4609 4609
4610/* 4610int workqueue_prepare_cpu(unsigned int cpu)
4611 * Workqueues should be brought up before normal priority CPU notifiers. 4611{
4612 * This will be registered high priority CPU notifier. 4612 struct worker_pool *pool;
4613 */ 4613
4614static int workqueue_cpu_up_callback(struct notifier_block *nfb, 4614 for_each_cpu_worker_pool(pool, cpu) {
4615 unsigned long action, 4615 if (pool->nr_workers)
4616 void *hcpu) 4616 continue;
4617 if (!create_worker(pool))
4618 return -ENOMEM;
4619 }
4620 return 0;
4621}
4622
4623int workqueue_online_cpu(unsigned int cpu)
4617{ 4624{
4618 int cpu = (unsigned long)hcpu;
4619 struct worker_pool *pool; 4625 struct worker_pool *pool;
4620 struct workqueue_struct *wq; 4626 struct workqueue_struct *wq;
4621 int pi; 4627 int pi;
4622 4628
4623 switch (action & ~CPU_TASKS_FROZEN) { 4629 mutex_lock(&wq_pool_mutex);
4624 case CPU_UP_PREPARE:
4625 for_each_cpu_worker_pool(pool, cpu) {
4626 if (pool->nr_workers)
4627 continue;
4628 if (!create_worker(pool))
4629 return NOTIFY_BAD;
4630 }
4631 break;
4632
4633 case CPU_DOWN_FAILED:
4634 case CPU_ONLINE:
4635 mutex_lock(&wq_pool_mutex);
4636 4630
4637 for_each_pool(pool, pi) { 4631 for_each_pool(pool, pi) {
4638 mutex_lock(&pool->attach_mutex); 4632 mutex_lock(&pool->attach_mutex);
4639 4633
4640 if (pool->cpu == cpu) 4634 if (pool->cpu == cpu)
4641 rebind_workers(pool); 4635 rebind_workers(pool);
4642 else if (pool->cpu < 0) 4636 else if (pool->cpu < 0)
4643 restore_unbound_workers_cpumask(pool, cpu); 4637 restore_unbound_workers_cpumask(pool, cpu);
4644 4638
4645 mutex_unlock(&pool->attach_mutex); 4639 mutex_unlock(&pool->attach_mutex);
4646 } 4640 }
4647 4641
4648 /* update NUMA affinity of unbound workqueues */ 4642 /* update NUMA affinity of unbound workqueues */
4649 list_for_each_entry(wq, &workqueues, list) 4643 list_for_each_entry(wq, &workqueues, list)
4650 wq_update_unbound_numa(wq, cpu, true); 4644 wq_update_unbound_numa(wq, cpu, true);
4651 4645
4652 mutex_unlock(&wq_pool_mutex); 4646 mutex_unlock(&wq_pool_mutex);
4653 break; 4647 return 0;
4654 }
4655 return NOTIFY_OK;
4656} 4648}
4657 4649
4658/* 4650int workqueue_offline_cpu(unsigned int cpu)
4659 * Workqueues should be brought down after normal priority CPU notifiers.
4660 * This will be registered as low priority CPU notifier.
4661 */
4662static int workqueue_cpu_down_callback(struct notifier_block *nfb,
4663 unsigned long action,
4664 void *hcpu)
4665{ 4651{
4666 int cpu = (unsigned long)hcpu;
4667 struct work_struct unbind_work; 4652 struct work_struct unbind_work;
4668 struct workqueue_struct *wq; 4653 struct workqueue_struct *wq;
4669 4654
4670 switch (action & ~CPU_TASKS_FROZEN) { 4655 /* unbinding per-cpu workers should happen on the local CPU */
4671 case CPU_DOWN_PREPARE: 4656 INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn);
4672 /* unbinding per-cpu workers should happen on the local CPU */ 4657 queue_work_on(cpu, system_highpri_wq, &unbind_work);
4673 INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); 4658
4674 queue_work_on(cpu, system_highpri_wq, &unbind_work); 4659 /* update NUMA affinity of unbound workqueues */
4675 4660 mutex_lock(&wq_pool_mutex);
4676 /* update NUMA affinity of unbound workqueues */ 4661 list_for_each_entry(wq, &workqueues, list)
4677 mutex_lock(&wq_pool_mutex); 4662 wq_update_unbound_numa(wq, cpu, false);
4678 list_for_each_entry(wq, &workqueues, list) 4663 mutex_unlock(&wq_pool_mutex);
4679 wq_update_unbound_numa(wq, cpu, false); 4664
4680 mutex_unlock(&wq_pool_mutex); 4665 /* wait for per-cpu unbinding to finish */
4681 4666 flush_work(&unbind_work);
4682 /* wait for per-cpu unbinding to finish */ 4667 destroy_work_on_stack(&unbind_work);
4683 flush_work(&unbind_work); 4668 return 0;
4684 destroy_work_on_stack(&unbind_work);
4685 break;
4686 }
4687 return NOTIFY_OK;
4688} 4669}
4689 4670
4690#ifdef CONFIG_SMP 4671#ifdef CONFIG_SMP
@@ -5486,9 +5467,6 @@ static int __init init_workqueues(void)
5486 5467
5487 pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); 5468 pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC);
5488 5469
5489 cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP);
5490 hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN);
5491
5492 wq_numa_init(); 5470 wq_numa_init();
5493 5471
5494 /* initialize CPU pools */ 5472 /* initialize CPU pools */
diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
index e2d5b6f988fb..4fde8c7dfcfe 100644
--- a/virt/kvm/arm/arch_timer.c
+++ b/virt/kvm/arm/arch_timer.c
@@ -405,26 +405,17 @@ u64 kvm_arm_timer_get_reg(struct kvm_vcpu *vcpu, u64 regid)
405 return (u64)-1; 405 return (u64)-1;
406} 406}
407 407
408static int kvm_timer_cpu_notify(struct notifier_block *self, 408static int kvm_timer_starting_cpu(unsigned int cpu)
409 unsigned long action, void *cpu)
410{ 409{
411 switch (action) { 410 kvm_timer_init_interrupt(NULL);
412 case CPU_STARTING: 411 return 0;
413 case CPU_STARTING_FROZEN:
414 kvm_timer_init_interrupt(NULL);
415 break;
416 case CPU_DYING:
417 case CPU_DYING_FROZEN:
418 disable_percpu_irq(host_vtimer_irq);
419 break;
420 }
421
422 return NOTIFY_OK;
423} 412}
424 413
425static struct notifier_block kvm_timer_cpu_nb = { 414static int kvm_timer_dying_cpu(unsigned int cpu)
426 .notifier_call = kvm_timer_cpu_notify, 415{
427}; 416 disable_percpu_irq(host_vtimer_irq);
417 return 0;
418}
428 419
429int kvm_timer_hyp_init(void) 420int kvm_timer_hyp_init(void)
430{ 421{
@@ -449,12 +440,6 @@ int kvm_timer_hyp_init(void)
449 goto out; 440 goto out;
450 } 441 }
451 442
452 err = __register_cpu_notifier(&kvm_timer_cpu_nb);
453 if (err) {
454 kvm_err("Cannot register timer CPU notifier\n");
455 goto out_free;
456 }
457
458 wqueue = create_singlethread_workqueue("kvm_arch_timer"); 443 wqueue = create_singlethread_workqueue("kvm_arch_timer");
459 if (!wqueue) { 444 if (!wqueue) {
460 err = -ENOMEM; 445 err = -ENOMEM;
@@ -462,8 +447,10 @@ int kvm_timer_hyp_init(void)
462 } 447 }
463 448
464 kvm_info("virtual timer IRQ%d\n", host_vtimer_irq); 449 kvm_info("virtual timer IRQ%d\n", host_vtimer_irq);
465 on_each_cpu(kvm_timer_init_interrupt, NULL, 1);
466 450
451 cpuhp_setup_state(CPUHP_AP_KVM_ARM_TIMER_STARTING,
452 "AP_KVM_ARM_TIMER_STARTING", kvm_timer_starting_cpu,
453 kvm_timer_dying_cpu);
467 goto out; 454 goto out;
468out_free: 455out_free:
469 free_percpu_irq(host_vtimer_irq, kvm_get_running_vcpus()); 456 free_percpu_irq(host_vtimer_irq, kvm_get_running_vcpus());
diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
index c3bfbb981e73..67cb5e948be2 100644
--- a/virt/kvm/arm/vgic.c
+++ b/virt/kvm/arm/vgic.c
@@ -2326,32 +2326,18 @@ int vgic_has_attr_regs(const struct vgic_io_range *ranges, phys_addr_t offset)
2326 return -ENXIO; 2326 return -ENXIO;
2327} 2327}
2328 2328
2329static void vgic_init_maintenance_interrupt(void *info) 2329static int vgic_starting_cpu(unsigned int cpu)
2330{ 2330{
2331 enable_percpu_irq(vgic->maint_irq, 0); 2331 enable_percpu_irq(vgic->maint_irq, 0);
2332 return 0;
2332} 2333}
2333 2334
2334static int vgic_cpu_notify(struct notifier_block *self, 2335static int vgic_dying_cpu(unsigned int cpu)
2335 unsigned long action, void *cpu)
2336{ 2336{
2337 switch (action) { 2337 disable_percpu_irq(vgic->maint_irq);
2338 case CPU_STARTING: 2338 return 0;
2339 case CPU_STARTING_FROZEN:
2340 vgic_init_maintenance_interrupt(NULL);
2341 break;
2342 case CPU_DYING:
2343 case CPU_DYING_FROZEN:
2344 disable_percpu_irq(vgic->maint_irq);
2345 break;
2346 }
2347
2348 return NOTIFY_OK;
2349} 2339}
2350 2340
2351static struct notifier_block vgic_cpu_nb = {
2352 .notifier_call = vgic_cpu_notify,
2353};
2354
2355static int kvm_vgic_probe(void) 2341static int kvm_vgic_probe(void)
2356{ 2342{
2357 const struct gic_kvm_info *gic_kvm_info; 2343 const struct gic_kvm_info *gic_kvm_info;
@@ -2392,19 +2378,10 @@ int kvm_vgic_hyp_init(void)
2392 return ret; 2378 return ret;
2393 } 2379 }
2394 2380
2395 ret = __register_cpu_notifier(&vgic_cpu_nb); 2381 cpuhp_setup_state(CPUHP_AP_KVM_ARM_VGIC_STARTING,
2396 if (ret) { 2382 "AP_KVM_ARM_VGIC_STARTING", vgic_starting_cpu,
2397 kvm_err("Cannot register vgic CPU notifier\n"); 2383 vgic_dying_cpu);
2398 goto out_free_irq;
2399 }
2400
2401 on_each_cpu(vgic_init_maintenance_interrupt, NULL, 1);
2402
2403 return 0; 2384 return 0;
2404
2405out_free_irq:
2406 free_percpu_irq(vgic->maint_irq, kvm_get_running_vcpus());
2407 return ret;
2408} 2385}
2409 2386
2410int kvm_irq_map_gsi(struct kvm *kvm, 2387int kvm_irq_map_gsi(struct kvm *kvm,
diff --git a/virt/kvm/arm/vgic/vgic-init.c b/virt/kvm/arm/vgic/vgic-init.c
index a1442f7c9c4d..2c7f0d5a62ea 100644
--- a/virt/kvm/arm/vgic/vgic-init.c
+++ b/virt/kvm/arm/vgic/vgic-init.c
@@ -353,32 +353,19 @@ out:
353 353
354/* GENERIC PROBE */ 354/* GENERIC PROBE */
355 355
356static void vgic_init_maintenance_interrupt(void *info) 356static int vgic_init_cpu_starting(unsigned int cpu)
357{ 357{
358 enable_percpu_irq(kvm_vgic_global_state.maint_irq, 0); 358 enable_percpu_irq(kvm_vgic_global_state.maint_irq, 0);
359 return 0;
359} 360}
360 361
361static int vgic_cpu_notify(struct notifier_block *self,
362 unsigned long action, void *cpu)
363{
364 switch (action) {
365 case CPU_STARTING:
366 case CPU_STARTING_FROZEN:
367 vgic_init_maintenance_interrupt(NULL);
368 break;
369 case CPU_DYING:
370 case CPU_DYING_FROZEN:
371 disable_percpu_irq(kvm_vgic_global_state.maint_irq);
372 break;
373 }
374 362
375 return NOTIFY_OK; 363static int vgic_init_cpu_dying(unsigned int cpu)
364{
365 disable_percpu_irq(kvm_vgic_global_state.maint_irq);
366 return 0;
376} 367}
377 368
378static struct notifier_block vgic_cpu_nb = {
379 .notifier_call = vgic_cpu_notify,
380};
381
382static irqreturn_t vgic_maintenance_handler(int irq, void *data) 369static irqreturn_t vgic_maintenance_handler(int irq, void *data)
383{ 370{
384 /* 371 /*
@@ -434,14 +421,14 @@ int kvm_vgic_hyp_init(void)
434 return ret; 421 return ret;
435 } 422 }
436 423
437 ret = __register_cpu_notifier(&vgic_cpu_nb); 424 ret = cpuhp_setup_state(CPUHP_AP_KVM_ARM_VGIC_INIT_STARTING,
425 "AP_KVM_ARM_VGIC_INIT_STARTING",
426 vgic_init_cpu_starting, vgic_init_cpu_dying);
438 if (ret) { 427 if (ret) {
439 kvm_err("Cannot register vgic CPU notifier\n"); 428 kvm_err("Cannot register vgic CPU notifier\n");
440 goto out_free_irq; 429 goto out_free_irq;
441 } 430 }
442 431
443 on_each_cpu(vgic_init_maintenance_interrupt, NULL, 1);
444
445 kvm_info("vgic interrupt IRQ%d\n", kvm_vgic_global_state.maint_irq); 432 kvm_info("vgic interrupt IRQ%d\n", kvm_vgic_global_state.maint_irq);
446 return 0; 433 return 0;
447 434
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index ce3d8e5be73e..2e791367c576 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -3155,12 +3155,13 @@ static void hardware_enable_nolock(void *junk)
3155 } 3155 }
3156} 3156}
3157 3157
3158static void hardware_enable(void) 3158static int kvm_starting_cpu(unsigned int cpu)
3159{ 3159{
3160 raw_spin_lock(&kvm_count_lock); 3160 raw_spin_lock(&kvm_count_lock);
3161 if (kvm_usage_count) 3161 if (kvm_usage_count)
3162 hardware_enable_nolock(NULL); 3162 hardware_enable_nolock(NULL);
3163 raw_spin_unlock(&kvm_count_lock); 3163 raw_spin_unlock(&kvm_count_lock);
3164 return 0;
3164} 3165}
3165 3166
3166static void hardware_disable_nolock(void *junk) 3167static void hardware_disable_nolock(void *junk)
@@ -3173,12 +3174,13 @@ static void hardware_disable_nolock(void *junk)
3173 kvm_arch_hardware_disable(); 3174 kvm_arch_hardware_disable();
3174} 3175}
3175 3176
3176static void hardware_disable(void) 3177static int kvm_dying_cpu(unsigned int cpu)
3177{ 3178{
3178 raw_spin_lock(&kvm_count_lock); 3179 raw_spin_lock(&kvm_count_lock);
3179 if (kvm_usage_count) 3180 if (kvm_usage_count)
3180 hardware_disable_nolock(NULL); 3181 hardware_disable_nolock(NULL);
3181 raw_spin_unlock(&kvm_count_lock); 3182 raw_spin_unlock(&kvm_count_lock);
3183 return 0;
3182} 3184}
3183 3185
3184static void hardware_disable_all_nolock(void) 3186static void hardware_disable_all_nolock(void)
@@ -3219,21 +3221,6 @@ static int hardware_enable_all(void)
3219 return r; 3221 return r;
3220} 3222}
3221 3223
3222static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3223 void *v)
3224{
3225 val &= ~CPU_TASKS_FROZEN;
3226 switch (val) {
3227 case CPU_DYING:
3228 hardware_disable();
3229 break;
3230 case CPU_STARTING:
3231 hardware_enable();
3232 break;
3233 }
3234 return NOTIFY_OK;
3235}
3236
3237static int kvm_reboot(struct notifier_block *notifier, unsigned long val, 3224static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3238 void *v) 3225 void *v)
3239{ 3226{
@@ -3500,10 +3487,6 @@ int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3500 return r; 3487 return r;
3501} 3488}
3502 3489
3503static struct notifier_block kvm_cpu_notifier = {
3504 .notifier_call = kvm_cpu_hotplug,
3505};
3506
3507static int kvm_debugfs_open(struct inode *inode, struct file *file, 3490static int kvm_debugfs_open(struct inode *inode, struct file *file,
3508 int (*get)(void *, u64 *), int (*set)(void *, u64), 3491 int (*get)(void *, u64 *), int (*set)(void *, u64),
3509 const char *fmt) 3492 const char *fmt)
@@ -3754,7 +3737,8 @@ int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3754 goto out_free_1; 3737 goto out_free_1;
3755 } 3738 }
3756 3739
3757 r = register_cpu_notifier(&kvm_cpu_notifier); 3740 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "AP_KVM_STARTING",
3741 kvm_starting_cpu, kvm_dying_cpu);
3758 if (r) 3742 if (r)
3759 goto out_free_2; 3743 goto out_free_2;
3760 register_reboot_notifier(&kvm_reboot_notifier); 3744 register_reboot_notifier(&kvm_reboot_notifier);
@@ -3808,7 +3792,7 @@ out_free:
3808 kmem_cache_destroy(kvm_vcpu_cache); 3792 kmem_cache_destroy(kvm_vcpu_cache);
3809out_free_3: 3793out_free_3:
3810 unregister_reboot_notifier(&kvm_reboot_notifier); 3794 unregister_reboot_notifier(&kvm_reboot_notifier);
3811 unregister_cpu_notifier(&kvm_cpu_notifier); 3795 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
3812out_free_2: 3796out_free_2:
3813out_free_1: 3797out_free_1:
3814 kvm_arch_hardware_unsetup(); 3798 kvm_arch_hardware_unsetup();
@@ -3831,7 +3815,7 @@ void kvm_exit(void)
3831 kvm_async_pf_deinit(); 3815 kvm_async_pf_deinit();
3832 unregister_syscore_ops(&kvm_syscore_ops); 3816 unregister_syscore_ops(&kvm_syscore_ops);
3833 unregister_reboot_notifier(&kvm_reboot_notifier); 3817 unregister_reboot_notifier(&kvm_reboot_notifier);
3834 unregister_cpu_notifier(&kvm_cpu_notifier); 3818 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
3835 on_each_cpu(hardware_disable_nolock, NULL, 1); 3819 on_each_cpu(hardware_disable_nolock, NULL, 1);
3836 kvm_arch_hardware_unsetup(); 3820 kvm_arch_hardware_unsetup();
3837 kvm_arch_exit(); 3821 kvm_arch_exit();
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-01 11:25:50 -0400