diff options
-rw-r--r-- | xen/arch/x86/cpu/mtrr/main.c | 2 | ||||
-rw-r--r-- | xen/arch/x86/domain.c | 3 | ||||
-rw-r--r-- | xen/arch/x86/smpboot.c | 25 | ||||
-rw-r--r-- | xen/common/Makefile | 1 | ||||
-rw-r--r-- | xen/common/stop_machine.c | 168 | ||||
-rw-r--r-- | xen/include/asm-x86/smp.h | 3 | ||||
-rw-r--r-- | xen/include/xen/smp.h | 17 | ||||
-rw-r--r-- | xen/include/xen/softirq.h | 3 | ||||
-rw-r--r-- | xen/include/xen/stop_machine.h | 30 |
9 files changed, 239 insertions, 13 deletions
diff --git a/xen/arch/x86/cpu/mtrr/main.c b/xen/arch/x86/cpu/mtrr/main.c index 4881055d1a..2a606f0320 100644 --- a/xen/arch/x86/cpu/mtrr/main.c +++ b/xen/arch/x86/cpu/mtrr/main.c @@ -46,8 +46,6 @@ #define DEFINE_MUTEX(_m) DEFINE_SPINLOCK(_m) #define mutex_lock(_m) spin_lock(_m) #define mutex_unlock(_m) spin_unlock(_m) -#define lock_cpu_hotplug() ((void)0) -#define unlock_cpu_hotplug() ((void)0) #define dump_stack() ((void)0) #define get_cpu() smp_processor_id() #define put_cpu() do {} while(0) diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c index 9d0537a9f9..25bc2c7148 100644 --- a/xen/arch/x86/domain.c +++ b/xen/arch/x86/domain.c @@ -82,7 +82,6 @@ static void default_idle(void) static void play_dead(void) { - __cpu_disable(); /* This must be done before dead CPU ack */ cpu_exit_clear(); hvm_cpu_down(); @@ -101,7 +100,7 @@ void idle_loop(void) { for ( ; ; ) { - if (cpu_is_offline(smp_processor_id())) + if ( cpu_is_offline(smp_processor_id()) ) play_dead(); page_scrub_schedule_work(); default_idle(); diff --git a/xen/arch/x86/smpboot.c b/xen/arch/x86/smpboot.c index 32668e9455..914ec62689 100644 --- a/xen/arch/x86/smpboot.c +++ b/xen/arch/x86/smpboot.c @@ -54,6 +54,7 @@ #include <mach_apic.h> #include <mach_wakecpu.h> #include <smpboot_hooks.h> +#include <xen/stop_machine.h> #define set_kernel_exec(x, y) (0) #define setup_trampoline() (bootsym_phys(trampoline_realmode_entry)) @@ -1208,6 +1209,15 @@ int __cpu_disable(void) if (cpu == 0) return -EBUSY; + /* + * Only S3 is using this path, and thus idle vcpus are running on all + * APs when we are called. To support full cpu hotplug, other + * notification mechanisms should be introduced (e.g., migrate vcpus + * off this physical cpu before rendezvous point). + */ + if (!is_idle_vcpu(current)) + return -EINVAL; + local_irq_disable(); clear_local_APIC(); /* Allow any queued timer interrupts to get serviced */ @@ -1244,6 +1254,11 @@ void __cpu_die(unsigned int cpu) printk(KERN_ERR "CPU %u didn't die...\n", cpu); } +static int take_cpu_down(void *unused) +{ + return __cpu_disable(); +} + /* * XXX: One important thing missed here is to migrate vcpus * from dead cpu to other online ones and then put whole @@ -1269,7 +1284,6 @@ void __cpu_die(unsigned int cpu) int cpu_down(unsigned int cpu) { int err = 0; - cpumask_t mask; spin_lock(&cpu_add_remove_lock); if (num_online_cpus() == 1) { @@ -1283,11 +1297,10 @@ int cpu_down(unsigned int cpu) } printk("Prepare to bring CPU%d down...\n", cpu); - /* Send notification to remote idle vcpu */ - cpus_clear(mask); - cpu_set(cpu, mask); - per_cpu(cpu_state, cpu) = CPU_DYING; - smp_send_event_check_mask(mask); + + err = __stop_machine_run(take_cpu_down, NULL, cpu); + if ( err < 0 ) + goto out; __cpu_die(cpu); diff --git a/xen/common/Makefile b/xen/common/Makefile index 1bf38310bf..631ac384bd 100644 --- a/xen/common/Makefile +++ b/xen/common/Makefile @@ -16,6 +16,7 @@ obj-y += sched_sedf.o obj-y += schedule.o obj-y += shutdown.o obj-y += softirq.o +obj-y += stop_machine.o obj-y += string.o obj-y += symbols.o obj-y += sysctl.o diff --git a/xen/common/stop_machine.c b/xen/common/stop_machine.c new file mode 100644 index 0000000000..b109167ef7 --- /dev/null +++ b/xen/common/stop_machine.c @@ -0,0 +1,168 @@ +/****************************************************************************** + * common/stop_machine.c + * + * Facilities to put whole machine in a safe 'stop' state + * + * Copyright 2005 Rusty Russell rusty@rustcorp.com.au IBM Corporation + * Copyright 2008 Kevin Tian <kevin.tian@intel.com>, Intel Corporation. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 59 Temple + * Place - Suite 330, Boston, MA 02111-1307 USA. + */ + +#include <xen/config.h> +#include <xen/init.h> +#include <xen/spinlock.h> +#include <asm/smp.h> +#include <asm/current.h> +#include <xen/softirq.h> +#include <asm/processor.h> +#include <xen/errno.h> + +enum stopmachine_state { + STOPMACHINE_START, + STOPMACHINE_PREPARE, + STOPMACHINE_DISABLE_IRQ, + STOPMACHINE_INVOKE, + STOPMACHINE_EXIT +}; + +struct stopmachine_data { + unsigned int nr_cpus; + + enum stopmachine_state state; + atomic_t done; + + unsigned int fn_cpu; + int fn_result; + int (*fn)(void *); + void *fn_data; +}; + +static struct stopmachine_data stopmachine_data; +static DEFINE_SPINLOCK(stopmachine_lock); + +static void stopmachine_set_state(enum stopmachine_state state) +{ + atomic_set(&stopmachine_data.done, 0); + smp_wmb(); + stopmachine_data.state = state; + while ( atomic_read(&stopmachine_data.done) != stopmachine_data.nr_cpus ) + cpu_relax(); +} + +int __stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) +{ + cpumask_t allbutself; + unsigned int i, nr_cpus; + int ret; + + BUG_ON(!local_irq_is_enabled()); + + allbutself = cpu_online_map; + cpu_clear(smp_processor_id(), allbutself); + nr_cpus = cpus_weight(allbutself); + + if ( nr_cpus == 0 ) + { + BUG_ON(cpu != smp_processor_id()); + return (*fn)(data); + } + + /* Note: We shouldn't spin on lock when it's held by others since others + * is expecting this cpus to enter softirq context. Or else deadlock + * is caused. + */ + if ( !spin_trylock(&stopmachine_lock) ) + return -EBUSY; + + stopmachine_data.fn = fn; + stopmachine_data.fn_data = data; + stopmachine_data.nr_cpus = nr_cpus; + stopmachine_data.fn_cpu = cpu; + atomic_set(&stopmachine_data.done, 0); + stopmachine_data.state = STOPMACHINE_START; + + smp_wmb(); + + for_each_cpu_mask ( i, allbutself ) + cpu_raise_softirq(i, STOPMACHINE_SOFTIRQ); + + stopmachine_set_state(STOPMACHINE_PREPARE); + + local_irq_disable(); + stopmachine_set_state(STOPMACHINE_DISABLE_IRQ); + + if ( cpu == smp_processor_id() ) + stopmachine_data.fn_result = (*fn)(data); + stopmachine_set_state(STOPMACHINE_INVOKE); + ret = stopmachine_data.fn_result; + + stopmachine_set_state(STOPMACHINE_EXIT); + local_irq_enable(); + + spin_unlock(&stopmachine_lock); + + return ret; +} + +int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) +{ + int ret; + + lock_cpu_hotplug(); + ret = __stop_machine_run(fn, data, cpu); + unlock_cpu_hotplug(); + + return ret; +} + +static void stopmachine_softirq(void) +{ + enum stopmachine_state state = STOPMACHINE_START; + + smp_mb(); + + while ( state != STOPMACHINE_EXIT ) + { + while ( stopmachine_data.state == state ) + cpu_relax(); + + state = stopmachine_data.state; + switch ( state ) + { + case STOPMACHINE_DISABLE_IRQ: + local_irq_disable(); + break; + case STOPMACHINE_INVOKE: + if ( stopmachine_data.fn_cpu == smp_processor_id() ) + stopmachine_data.fn_result = + stopmachine_data.fn(stopmachine_data.fn_data); + break; + default: + break; + } + + smp_mb(); + atomic_inc(&stopmachine_data.done); + } + + local_irq_enable(); +} + +static int __init cpu_stopmachine_init(void) +{ + open_softirq(STOPMACHINE_SOFTIRQ, stopmachine_softirq); + return 0; +} +__initcall(cpu_stopmachine_init); diff --git a/xen/include/asm-x86/smp.h b/xen/include/asm-x86/smp.h index 55b8cf7705..cf5159603d 100644 --- a/xen/include/asm-x86/smp.h +++ b/xen/include/asm-x86/smp.h @@ -51,12 +51,11 @@ extern u8 x86_cpu_to_apicid[]; /* State of each CPU. */ #define CPU_ONLINE 0x0002 /* CPU is up */ -#define CPU_DYING 0x0003 /* CPU is requested to die */ #define CPU_DEAD 0x0004 /* CPU is dead */ DECLARE_PER_CPU(int, cpu_state); #ifdef CONFIG_HOTPLUG_CPU -#define cpu_is_offline(cpu) unlikely(per_cpu(cpu_state,cpu) == CPU_DYING) +#define cpu_is_offline(cpu) unlikely(!cpu_online(cpu)) extern int cpu_down(unsigned int cpu); extern int cpu_up(unsigned int cpu); extern void cpu_exit_clear(void); diff --git a/xen/include/xen/smp.h b/xen/include/xen/smp.h index c02ecd4dc3..3e7de77ec8 100644 --- a/xen/include/xen/smp.h +++ b/xen/include/xen/smp.h @@ -112,4 +112,21 @@ static inline int on_each_cpu( #define smp_processor_id() raw_smp_processor_id() +#ifdef CONFIG_HOTPLUG_CPU +extern spinlock_t cpu_add_remove_lock; +/* + * FIXME: need a better lock mechanism when real cpu hotplug is later + * supported, since spinlock may cause dead lock: + * cpu0: in stop_machine with lock held. Wait for cpu1 to respond + * to stop request + * cpu1: spin loop on lock upon cpu hotplug request from guest, + * without chance to handle softirq + * ... + */ +#define lock_cpu_hotplug() spin_lock(&cpu_add_remove_lock); +#define unlock_cpu_hotplug() spin_unlock(&cpu_add_remove_lock); +#else +#define lock_cpu_hotplug() do { } while ( 0 ) +#define unlock_cpu_hotplug() do { } while ( 0 ) +#endif #endif diff --git a/xen/include/xen/softirq.h b/xen/include/xen/softirq.h index 9a66e0fb03..7734b3374d 100644 --- a/xen/include/xen/softirq.h +++ b/xen/include/xen/softirq.h @@ -10,8 +10,9 @@ #define PAGE_SCRUB_SOFTIRQ 5 #define TRACE_SOFTIRQ 6 #define RCU_SOFTIRQ 7 +#define STOPMACHINE_SOFTIRQ 8 -#define NR_COMMON_SOFTIRQS 8 +#define NR_COMMON_SOFTIRQS 9 #include <asm/softirq.h> diff --git a/xen/include/xen/stop_machine.h b/xen/include/xen/stop_machine.h new file mode 100644 index 0000000000..f379f6705a --- /dev/null +++ b/xen/include/xen/stop_machine.h @@ -0,0 +1,30 @@ +#ifndef __XEN_STOP_MACHINE_H__ +#define __XEN_STOP_MACHINE_H__ + +/** + * stop_machine_run: freeze the machine on all CPUs and run this function + * @fn: the function to run + * @data: the data ptr for the @fn() + * @cpu: the cpu to run @fn() on (or any, if @cpu == NR_CPUS). + * + * Description: This causes every other cpu to enter a safe point, with + * each of which disables interrupts, and finally interrupts are disabled + * on the current CPU. The result is that none is holding a spinlock + * or inside any other preempt-disabled region when @fn() runs. + * + * This can be thought of as a very heavy write lock, equivalent to + * grabbing every spinlock in the kernel. */ +int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu); + +/** + * __stop_machine_run: freeze the machine on all CPUs and run this function + * @fn: the function to run + * @data: the data ptr for the @fn + * @cpu: the cpu to run @fn on (or any, if @cpu == NR_CPUS. + * + * Description: This is a special version of the above, without explicit + * lock acquisition. Used by hotplug cpu. + */ +int __stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu); + +#endif /* __XEN_STOP_MACHINE_H__ */ |