| Message ID | 20190829151027.9930-1-joao.m.martins@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] cpuidle-haltpoll: vcpu hotplug support | expand |
On Thu, Aug 29, 2019 at 04:10:27PM +0100, Joao Martins wrote: > When cpus != maxcpus cpuidle-haltpoll will fail to register all vcpus > past the online ones and thus fail to register the idle driver. > This is because cpuidle_add_sysfs() will return with -ENODEV as a > consequence from get_cpu_device() return no device for a non-existing > CPU. > > Instead switch to cpuidle_register_driver() and manually register each > of the present cpus through cpuhp_setup_state() callback and future > ones that get onlined. This mimmics similar logic that intel_idle does. > > Fixes: fa86ee90eb11 ("add cpuidle-haltpoll driver") > Signed-off-by: Joao Martins <joao.m.martins@oracle.com> > Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> > --- > v2: > * move cpus_read_unlock() right after unregistering all cpuidle_devices; > (Marcello Tosatti) > * redundant usage of cpuidle_unregister() when only > cpuidle_unregister_driver() suffices; (Marcelo Tosatti) > * cpuhp_setup_state() returns a state (> 0) on success with CPUHP_AP_ONLINE_DYN > thus we set @ret to 0 > --- > arch/x86/include/asm/cpuidle_haltpoll.h | 4 +- > arch/x86/kernel/kvm.c | 18 +++---- > drivers/cpuidle/cpuidle-haltpoll.c | 67 +++++++++++++++++++++++-- > include/linux/cpuidle_haltpoll.h | 4 +- > 4 files changed, 72 insertions(+), 21 deletions(-) > > diff --git a/arch/x86/include/asm/cpuidle_haltpoll.h b/arch/x86/include/asm/cpuidle_haltpoll.h > index ff8607d81526..c8b39c6716ff 100644 > --- a/arch/x86/include/asm/cpuidle_haltpoll.h > +++ b/arch/x86/include/asm/cpuidle_haltpoll.h > @@ -2,7 +2,7 @@ > #ifndef _ARCH_HALTPOLL_H > #define _ARCH_HALTPOLL_H > > -void arch_haltpoll_enable(void); > -void arch_haltpoll_disable(void); > +void arch_haltpoll_enable(unsigned int cpu); > +void arch_haltpoll_disable(unsigned int cpu); > > #endif > diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c > index 8d150e3732d9..a9b6c4e2446d 100644 > --- a/arch/x86/kernel/kvm.c > +++ b/arch/x86/kernel/kvm.c > @@ -880,32 +880,26 @@ static void kvm_enable_host_haltpoll(void *i) > wrmsrl(MSR_KVM_POLL_CONTROL, 1); > } > > -void arch_haltpoll_enable(void) > +void arch_haltpoll_enable(unsigned int cpu) > { > if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) { > - printk(KERN_ERR "kvm: host does not support poll control\n"); > - printk(KERN_ERR "kvm: host upgrade recommended\n"); > + pr_err_once("kvm: host does not support poll control\n"); > + pr_err_once("kvm: host upgrade recommended\n"); > return; > } > > - preempt_disable(); > /* Enable guest halt poll disables host halt poll */ > - kvm_disable_host_haltpoll(NULL); > - smp_call_function(kvm_disable_host_haltpoll, NULL, 1); > - preempt_enable(); > + smp_call_function_single(cpu, kvm_disable_host_haltpoll, NULL, 1); > } > EXPORT_SYMBOL_GPL(arch_haltpoll_enable); > > -void arch_haltpoll_disable(void) > +void arch_haltpoll_disable(unsigned int cpu) > { > if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) > return; > > - preempt_disable(); > /* Enable guest halt poll disables host halt poll */ > - kvm_enable_host_haltpoll(NULL); > - smp_call_function(kvm_enable_host_haltpoll, NULL, 1); > - preempt_enable(); > + smp_call_function_single(cpu, kvm_enable_host_haltpoll, NULL, 1); > } > EXPORT_SYMBOL_GPL(arch_haltpoll_disable); > #endif > diff --git a/drivers/cpuidle/cpuidle-haltpoll.c b/drivers/cpuidle/cpuidle-haltpoll.c > index 9ac093dcbb01..8baade23f8d0 100644 > --- a/drivers/cpuidle/cpuidle-haltpoll.c > +++ b/drivers/cpuidle/cpuidle-haltpoll.c > @@ -11,12 +11,15 @@ > */ > > #include <linux/init.h> > +#include <linux/cpu.h> > #include <linux/cpuidle.h> > #include <linux/module.h> > #include <linux/sched/idle.h> > #include <linux/kvm_para.h> > #include <linux/cpuidle_haltpoll.h> > > +static struct cpuidle_device __percpu *haltpoll_cpuidle_devices; > + > static int default_enter_idle(struct cpuidle_device *dev, > struct cpuidle_driver *drv, int index) > { > @@ -46,6 +49,48 @@ static struct cpuidle_driver haltpoll_driver = { > .state_count = 2, > }; > > +static int haltpoll_cpu_online(unsigned int cpu) > +{ > + struct cpuidle_device *dev; > + > + dev = per_cpu_ptr(haltpoll_cpuidle_devices, cpu); > + if (!dev->registered) { > + dev->cpu = cpu; > + if (cpuidle_register_device(dev)) { > + pr_notice("cpuidle_register_device %d failed!\n", cpu); > + return -EIO; > + } > + arch_haltpoll_enable(cpu); > + } > + > + return 0; > +} > + > +static void haltpoll_uninit(void) > +{ > + unsigned int cpu; > + > + cpus_read_lock(); > + > + for_each_online_cpu(cpu) { > + struct cpuidle_device *dev = > + per_cpu_ptr(haltpoll_cpuidle_devices, cpu); > + > + if (!dev->registered) > + continue; > + > + arch_haltpoll_disable(cpu); > + cpuidle_unregister_device(dev); > + } > + > + cpus_read_unlock(); > + > + cpuidle_unregister_driver(&haltpoll_driver); > + > + free_percpu(haltpoll_cpuidle_devices); > + haltpoll_cpuidle_devices = NULL; > +} > + > static int __init haltpoll_init(void) > { > int ret; > @@ -56,17 +101,29 @@ static int __init haltpoll_init(void) > if (!kvm_para_available()) > return 0; > > - ret = cpuidle_register(&haltpoll_driver, NULL); > - if (ret == 0) > - arch_haltpoll_enable(); > + ret = cpuidle_register_driver(drv); > + if (ret < 0) > + return ret; > + > + haltpoll_cpuidle_devices = alloc_percpu(struct cpuidle_device); > + if (haltpoll_cpuidle_devices == NULL) { > + cpuidle_unregister_driver(drv); > + return -ENOMEM; > + } > + > + ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/haltpoll:online", > + haltpoll_cpu_online, NULL); > + if (ret < 0) > + haltpoll_uninit(); > + else > + ret = 0; > > return ret; > } > > static void __exit haltpoll_exit(void) > { > - arch_haltpoll_disable(); > - cpuidle_unregister(&haltpoll_driver); > + haltpoll_uninit(); > } > > module_init(haltpoll_init); > diff --git a/include/linux/cpuidle_haltpoll.h b/include/linux/cpuidle_haltpoll.h > index fe5954c2409e..d50c1e0411a2 100644 > --- a/include/linux/cpuidle_haltpoll.h > +++ b/include/linux/cpuidle_haltpoll.h > @@ -5,11 +5,11 @@ > #ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL > #include <asm/cpuidle_haltpoll.h> > #else > -static inline void arch_haltpoll_enable(void) > +static inline void arch_haltpoll_enable(unsigned int cpu) > { > } > > -static inline void arch_haltpoll_disable(void) > +static inline void arch_haltpoll_disable(unsigned int cpu) > { > } > #endif > -- > 2.17.1 Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
On 8/29/19 4:27 PM, Marcelo Tosatti wrote: > On Thu, Aug 29, 2019 at 04:10:27PM +0100, Joao Martins wrote: >> When cpus != maxcpus cpuidle-haltpoll will fail to register all vcpus >> past the online ones and thus fail to register the idle driver. >> This is because cpuidle_add_sysfs() will return with -ENODEV as a >> consequence from get_cpu_device() return no device for a non-existing >> CPU. >> >> Instead switch to cpuidle_register_driver() and manually register each >> of the present cpus through cpuhp_setup_state() callback and future >> ones that get onlined. This mimmics similar logic that intel_idle does. >> >> Fixes: fa86ee90eb11 ("add cpuidle-haltpoll driver") >> Signed-off-by: Joao Martins <joao.m.martins@oracle.com> >> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> >> --- >> v2: >> * move cpus_read_unlock() right after unregistering all cpuidle_devices; >> (Marcello Tosatti) >> * redundant usage of cpuidle_unregister() when only >> cpuidle_unregister_driver() suffices; (Marcelo Tosatti) >> * cpuhp_setup_state() returns a state (> 0) on success with CPUHP_AP_ONLINE_DYN >> thus we set @ret to 0 [ ... ] > > Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com> > Thanks! Meanwhile upon re-reading cpuhp_setup_state() I found out the teardown/offlining and haltpoll_uninit() could be a bit simplified. So I sent out a new version[0], but didn't add your Rb because there's was some very slight functional changes. [0] https://lore.kernel.org/kvm/20190902104031.9296-1-joao.m.martins@oracle.com/ Joao
diff --git a/arch/x86/include/asm/cpuidle_haltpoll.h b/arch/x86/include/asm/cpuidle_haltpoll.h index ff8607d81526..c8b39c6716ff 100644 --- a/arch/x86/include/asm/cpuidle_haltpoll.h +++ b/arch/x86/include/asm/cpuidle_haltpoll.h @@ -2,7 +2,7 @@ #ifndef _ARCH_HALTPOLL_H #define _ARCH_HALTPOLL_H -void arch_haltpoll_enable(void); -void arch_haltpoll_disable(void); +void arch_haltpoll_enable(unsigned int cpu); +void arch_haltpoll_disable(unsigned int cpu); #endif diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c index 8d150e3732d9..a9b6c4e2446d 100644 --- a/arch/x86/kernel/kvm.c +++ b/arch/x86/kernel/kvm.c @@ -880,32 +880,26 @@ static void kvm_enable_host_haltpoll(void *i) wrmsrl(MSR_KVM_POLL_CONTROL, 1); } -void arch_haltpoll_enable(void) +void arch_haltpoll_enable(unsigned int cpu) { if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) { - printk(KERN_ERR "kvm: host does not support poll control\n"); - printk(KERN_ERR "kvm: host upgrade recommended\n"); + pr_err_once("kvm: host does not support poll control\n"); + pr_err_once("kvm: host upgrade recommended\n"); return; } - preempt_disable(); /* Enable guest halt poll disables host halt poll */ - kvm_disable_host_haltpoll(NULL); - smp_call_function(kvm_disable_host_haltpoll, NULL, 1); - preempt_enable(); + smp_call_function_single(cpu, kvm_disable_host_haltpoll, NULL, 1); } EXPORT_SYMBOL_GPL(arch_haltpoll_enable); -void arch_haltpoll_disable(void) +void arch_haltpoll_disable(unsigned int cpu) { if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) return; - preempt_disable(); /* Enable guest halt poll disables host halt poll */ - kvm_enable_host_haltpoll(NULL); - smp_call_function(kvm_enable_host_haltpoll, NULL, 1); - preempt_enable(); + smp_call_function_single(cpu, kvm_enable_host_haltpoll, NULL, 1); } EXPORT_SYMBOL_GPL(arch_haltpoll_disable); #endif diff --git a/drivers/cpuidle/cpuidle-haltpoll.c b/drivers/cpuidle/cpuidle-haltpoll.c index 9ac093dcbb01..8baade23f8d0 100644 --- a/drivers/cpuidle/cpuidle-haltpoll.c +++ b/drivers/cpuidle/cpuidle-haltpoll.c @@ -11,12 +11,15 @@ */ #include <linux/init.h> +#include <linux/cpu.h> #include <linux/cpuidle.h> #include <linux/module.h> #include <linux/sched/idle.h> #include <linux/kvm_para.h> #include <linux/cpuidle_haltpoll.h> +static struct cpuidle_device __percpu *haltpoll_cpuidle_devices; + static int default_enter_idle(struct cpuidle_device *dev, struct cpuidle_driver *drv, int index) { @@ -46,6 +49,48 @@ static struct cpuidle_driver haltpoll_driver = { .state_count = 2, }; +static int haltpoll_cpu_online(unsigned int cpu) +{ + struct cpuidle_device *dev; + + dev = per_cpu_ptr(haltpoll_cpuidle_devices, cpu); + if (!dev->registered) { + dev->cpu = cpu; + if (cpuidle_register_device(dev)) { + pr_notice("cpuidle_register_device %d failed!\n", cpu); + return -EIO; + } + arch_haltpoll_enable(cpu); + } + + return 0; +} + +static void haltpoll_uninit(void) +{ + unsigned int cpu; + + cpus_read_lock(); + + for_each_online_cpu(cpu) { + struct cpuidle_device *dev = + per_cpu_ptr(haltpoll_cpuidle_devices, cpu); + + if (!dev->registered) + continue; + + arch_haltpoll_disable(cpu); + cpuidle_unregister_device(dev); + } + + cpus_read_unlock(); + + cpuidle_unregister_driver(&haltpoll_driver); + + free_percpu(haltpoll_cpuidle_devices); + haltpoll_cpuidle_devices = NULL; +} + static int __init haltpoll_init(void) { int ret; @@ -56,17 +101,29 @@ static int __init haltpoll_init(void) if (!kvm_para_available()) return 0; - ret = cpuidle_register(&haltpoll_driver, NULL); - if (ret == 0) - arch_haltpoll_enable(); + ret = cpuidle_register_driver(drv); + if (ret < 0) + return ret; + + haltpoll_cpuidle_devices = alloc_percpu(struct cpuidle_device); + if (haltpoll_cpuidle_devices == NULL) { + cpuidle_unregister_driver(drv); + return -ENOMEM; + } + + ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/haltpoll:online", + haltpoll_cpu_online, NULL); + if (ret < 0) + haltpoll_uninit(); + else + ret = 0; return ret; } static void __exit haltpoll_exit(void) { - arch_haltpoll_disable(); - cpuidle_unregister(&haltpoll_driver); + haltpoll_uninit(); } module_init(haltpoll_init); diff --git a/include/linux/cpuidle_haltpoll.h b/include/linux/cpuidle_haltpoll.h index fe5954c2409e..d50c1e0411a2 100644 --- a/include/linux/cpuidle_haltpoll.h +++ b/include/linux/cpuidle_haltpoll.h @@ -5,11 +5,11 @@ #ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL #include <asm/cpuidle_haltpoll.h> #else -static inline void arch_haltpoll_enable(void) +static inline void arch_haltpoll_enable(unsigned int cpu) { } -static inline void arch_haltpoll_disable(void) +static inline void arch_haltpoll_disable(unsigned int cpu) { } #endif
When cpus != maxcpus cpuidle-haltpoll will fail to register all vcpus past the online ones and thus fail to register the idle driver. This is because cpuidle_add_sysfs() will return with -ENODEV as a consequence from get_cpu_device() return no device for a non-existing CPU. Instead switch to cpuidle_register_driver() and manually register each of the present cpus through cpuhp_setup_state() callback and future ones that get onlined. This mimmics similar logic that intel_idle does. Fixes: fa86ee90eb11 ("add cpuidle-haltpoll driver") Signed-off-by: Joao Martins <joao.m.martins@oracle.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> --- v2: * move cpus_read_unlock() right after unregistering all cpuidle_devices; (Marcello Tosatti) * redundant usage of cpuidle_unregister() when only cpuidle_unregister_driver() suffices; (Marcelo Tosatti) * cpuhp_setup_state() returns a state (> 0) on success with CPUHP_AP_ONLINE_DYN thus we set @ret to 0 --- arch/x86/include/asm/cpuidle_haltpoll.h | 4 +- arch/x86/kernel/kvm.c | 18 +++---- drivers/cpuidle/cpuidle-haltpoll.c | 67 +++++++++++++++++++++++-- include/linux/cpuidle_haltpoll.h | 4 +- 4 files changed, 72 insertions(+), 21 deletions(-)