| Message ID | 20210728073700.120449-3-suleiman@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | KVM: Support Heterogeneous RT VCPU Configurations. | expand |
Suleiman Souhlal <suleiman@google.com> writes: > When KVM_PREEMPT_COUNT_REPORTING is enabled, the host can use > preempt_count to determine if the guest is in a critical section, > if it also has CONFIG_KVM_HETEROGENEOUS_RT enabled, in order to > use heterogeneous RT VCPU configurations. > > Signed-off-by: Suleiman Souhlal <suleiman@google.com> > --- > arch/x86/Kconfig | 11 +++++++++++ > arch/x86/kernel/kvm.c | 10 ++++++++++ > 2 files changed, 21 insertions(+) > > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig > index 49270655e827..d8b62789df57 100644 > --- a/arch/x86/Kconfig > +++ b/arch/x86/Kconfig > @@ -846,6 +846,17 @@ config PARAVIRT_TIME_ACCOUNTING > config PARAVIRT_CLOCK > bool > > +config KVM_PREEMPT_COUNT_REPORTING > + bool "KVM preempt_count reporting to the host" > + depends on KVM_GUEST && PREEMPT_COUNT > + default n > + help > + Select this option to enable KVM preempt_count reporting to the host, > + which can be useful in cases where some VCPUs are RT and the rest > + aren't. > + > + If in doubt, say N here. > + > config JAILHOUSE_GUEST > bool "Jailhouse non-root cell support" > depends on X86_64 && PCI > diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c > index a26643dc6bd6..7ec53ea3f979 100644 > --- a/arch/x86/kernel/kvm.c > +++ b/arch/x86/kernel/kvm.c > @@ -363,6 +363,16 @@ static void kvm_guest_cpu_init(void) > > if (has_steal_clock) > kvm_register_steal_time(); > + > +#ifdef CONFIG_KVM_PREEMPT_COUNT_REPORTING > + if (kvm_para_has_feature(KVM_FEATURE_PREEMPT_COUNT)) { > + unsigned long pa; > + > + pa = slow_virt_to_phys(this_cpu_ptr(&__preempt_count)) | > + KVM_MSR_ENABLED; > + wrmsrl(MSR_KVM_PREEMPT_COUNT, pa); > + } > +#endif > } Please also disable the feature in kvm_guest_cpu_offline() as e.g. upon kexec the memory address looses its meaning. > > static void kvm_pv_disable_apf(void)
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 49270655e827..d8b62789df57 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -846,6 +846,17 @@ config PARAVIRT_TIME_ACCOUNTING config PARAVIRT_CLOCK bool +config KVM_PREEMPT_COUNT_REPORTING + bool "KVM preempt_count reporting to the host" + depends on KVM_GUEST && PREEMPT_COUNT + default n + help + Select this option to enable KVM preempt_count reporting to the host, + which can be useful in cases where some VCPUs are RT and the rest + aren't. + + If in doubt, say N here. + config JAILHOUSE_GUEST bool "Jailhouse non-root cell support" depends on X86_64 && PCI diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c index a26643dc6bd6..7ec53ea3f979 100644 --- a/arch/x86/kernel/kvm.c +++ b/arch/x86/kernel/kvm.c @@ -363,6 +363,16 @@ static void kvm_guest_cpu_init(void) if (has_steal_clock) kvm_register_steal_time(); + +#ifdef CONFIG_KVM_PREEMPT_COUNT_REPORTING + if (kvm_para_has_feature(KVM_FEATURE_PREEMPT_COUNT)) { + unsigned long pa; + + pa = slow_virt_to_phys(this_cpu_ptr(&__preempt_count)) | + KVM_MSR_ENABLED; + wrmsrl(MSR_KVM_PREEMPT_COUNT, pa); + } +#endif } static void kvm_pv_disable_apf(void)
When KVM_PREEMPT_COUNT_REPORTING is enabled, the host can use preempt_count to determine if the guest is in a critical section, if it also has CONFIG_KVM_HETEROGENEOUS_RT enabled, in order to use heterogeneous RT VCPU configurations. Signed-off-by: Suleiman Souhlal <suleiman@google.com> --- arch/x86/Kconfig | 11 +++++++++++ arch/x86/kernel/kvm.c | 10 ++++++++++ 2 files changed, 21 insertions(+)