x86/x86: KVM: Fixes for KVM's PI wakeup handler
Sean Christopherson
maintainercover x86/x86: KVM: Fixes for KVM's PI wakeup handler
Two fixes and two cleanups related to KVM's posted interrupt wakeup handler. Fix #1 ensures any in-flight IRQs finish after changing the handler. Fix #2 actually uninstalls KVM's handler so that a spurious IRQ won't jump into a freed module.
AFAIK, no one has actually hit these bugs as it would require a really spurious IRQ, or a bug+race elsehwere that caused a device to post an interrupt well after a KVM guest is torn down.
Sean Christopherson (4):
x86/irq: Ensure PI wakeup handler is unregistered before module unload
KVM: VMX: Unregister posted interrupt wakeup handler on hardware
unsetup
x86/irq: KVM: Harden posted interrupt (un)registration paths
KVM: VMX: Register posted interrupt wakeup handler iff APICv is
enabled
arch/x86/include/asm/irq.h | 3 ++- arch/x86/kernel/irq.c | 30 +++++++++++++++++++++--------- arch/x86/kvm/vmx/vmx.c | 9 +++++++-- 3 files changed, 30 insertions(+), 12 deletions(-)
Signature · 1 line
2.33.0.882.g93a45727a2-goog
Sean Christopherson
maintainerPATCH 1/4 x86/irq: Ensure PI wakeup handler is unregistered before module unload
Add a synchronize_rcu() after setting the posted interrupt wakeup handler to ensure all readers, i.e. in-flight IRQ handlers, see the new handler before returning to the caller. If the caller is an exiting module and is unregistering its handler, failure to wait could result in the IRQ handler jumping into an unloaded module.
--- arch/x86/kernel/irq.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/kernel/irq.c +1 -0
@@ -293,6 +293,7 @@ void kvm_set_posted_intr_wakeup_handler(void (*handler)(void)) kvm_posted_intr_wakeup_handler = handler; else kvm_posted_intr_wakeup_handler = dummy_handler; +synchronize_rcu (); }EXPORT_SYMBOL_GPL (kvm_set_posted_intr_wakeup_handler);
Signature · 1 line
2.33.0.882.g93a45727a2-goog
Sean Christopherson
maintainerPATCH 2/4 KVM: VMX: Unregister posted interrupt wakeup handler on hardware unsetup
In reply to Sean ChristophersonUnregister KVM's posted interrupt wakeup handler during unsetup so that a spurious interrupt that arrives after kvm_intel.ko is unloaded doesn't call into freed memory.
--- arch/x86/kvm/vmx/vmx.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/kvm/vmx/vmx.c +5 -2
@@ -7553,6 +7553,8 @@ static void vmx_migrate_timers(struct kvm_vcpu *vcpu)static void hardware_unsetup (void ) { +kvm_set_posted_intr_wakeup_handler (NULL ); +if (nested)nested_vmx_hardware_unsetup ();
@@ -7881,8 +7883,6 @@ static __init int hardware_setup(void) vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit; } -kvm_set_posted_intr_wakeup_handler (pi_wakeup_handler); - kvm_mce_cap_supported |=MCG_LMCE_P ;if (pt_mode !=PT_MODE_SYSTEM && pt_mode !=PT_MODE_HOST_GUEST )
@@ -7906,6 +7906,9 @@ static __init int hardware_setup(void) r =alloc_kvm_area ();if (r)nested_vmx_hardware_unsetup (); + +kvm_set_posted_intr_wakeup_handler (pi_wakeup_handler); +return r; }
Signature · 1 line
2.33.0.882.g93a45727a2-goog
Sean Christopherson
maintainerPATCH 3/4 x86/irq: KVM: Harden posted interrupt (un)registration paths
In reply to Sean ChristophersonSplit the register and unregister paths for the posted interrupt wakeup handler, and WARN on conditions that are blatant bugs, e.g. attempting to overwrite an existing handler, unregistering the wrong handler, etc... This is very much a "low hanging fruit" hardening, e.g. a broken module could foul things up by doing concurrent registration from multiple CPUs.
Drop the use of a dummy handler so that the rejection logic can use a simple NULL check. There is zero benefit to blindly calling into a dummy handler.
Note, the registration path doesn't require synchronization, as it's the caller's responsibility to not generate interrupts it cares about until after its handler is registered, i.e. there can't be a relevant in-flight interrupt.
--- arch/x86/include/asm/irq.h | 3 ++- arch/x86/kernel/irq.c | 29 ++++++++++++++++++++--------- arch/x86/kvm/vmx/vmx.c | 4 ++-- 3 files changed, 24 insertions(+), 12 deletions(-)
arch/x86/include/asm/irq.h +2 -1
@@ -30,7 +30,8 @@ struct irq_desc;extern void fixup_irqs (void );#ifdef CONFIG_HAVE_KVM -extern void kvm_set_posted_intr_wakeup_handler (void (*handler)(void )); +extern void kvm_register_posted_intr_wakeup_handler (void (*handler)(void )); +extern void kvm_unregister_posted_intr_wakeup_handler (void (*handler)(void ));#endif extern void (*x86_platform_ipi_callback)(void );
arch/x86/kernel/irq.c +20 -9
@@ -284,18 +284,26 @@ DEFINE_IDTENTRY_SYSVEC(sysvec_x86_platform_ipi)#endif #ifdef CONFIG_HAVE_KVM -static void dummy_handler (void ) {} -static void (*kvm_posted_intr_wakeup_handler)(void ) = dummy_handler; +static void (*kvm_posted_intr_wakeup_handler)(void ); -void kvm_set_posted_intr_wakeup_handler (void (*handler)(void )) +void kvm_register_posted_intr_wakeup_handler (void (*handler)(void )) { -if (handler) - kvm_posted_intr_wakeup_handler = handler; -else - kvm_posted_intr_wakeup_handler = dummy_handler; +if (WARN_ON_ONCE (!handler || kvm_posted_intr_wakeup_handler)) +return ; + +WRITE_ONCE (kvm_posted_intr_wakeup_handler, handler); +} +EXPORT_SYMBOL_GPL (kvm_register_posted_intr_wakeup_handler); + +void kvm_unregister_posted_intr_wakeup_handler (void (*handler)(void )) +{ +if (WARN_ON_ONCE (!handler || handler != kvm_posted_intr_wakeup_handler)) +return ; + +WRITE_ONCE (kvm_posted_intr_wakeup_handler,NULL );synchronize_rcu (); } -EXPORT_SYMBOL_GPL (kvm_set_posted_intr_wakeup_handler); +EXPORT_SYMBOL_GPL (kvm_unregister_posted_intr_wakeup_handler);/* * Handler for POSTED_INTERRUPT_VECTOR.
@@ -311,9 +319,12 @@ DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_ipi)*/ DEFINE_IDTENTRY_SYSVEC (sysvec_kvm_posted_intr_wakeup_ipi) { +void (*handler)(void ) =READ_ONCE (kvm_posted_intr_wakeup_handler); +ack_APIC_irq ();inc_irq_stat (kvm_posted_intr_wakeup_ipis); -kvm_posted_intr_wakeup_handler (); +if (handler) +handler (); }/*
arch/x86/kvm/vmx/vmx.c +2 -2
@@ -7553,7 +7553,7 @@ static void vmx_migrate_timers(struct kvm_vcpu *vcpu)static void hardware_unsetup (void ) { -kvm_set_posted_intr_wakeup_handler (NULL ); +kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler);if (nested)nested_vmx_hardware_unsetup ();
@@ -7907,7 +7907,7 @@ static __init int hardware_setup(void)if (r)nested_vmx_hardware_unsetup (); -kvm_set_posted_intr_wakeup_handler (pi_wakeup_handler); +kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler);return r; }
Signature · 1 line
2.33.0.882.g93a45727a2-goog
Paolo Bonzini
maintainerRe: [PATCH 3/4] x86/irq: KVM: Harden posted interrupt (un)registration paths
Quoted 10 lines · Sean Christopherson
Split the register and unregister paths for the posted interrupt wakeup handler, and WARN on conditions that are blatant bugs, e.g. attempting to overwrite an existing handler, unregistering the wrong handler, etc... This is very much a "low hanging fruit" hardening, e.g. a broken module could foul things up by doing concurrent registration from multiple CPUs.
Drop the use of a dummy handler so that the rejection logic can use a simple NULL check. There is zero benefit to blindly calling into a dummy handler.
Quoted 6 lines
Note, the registration path doesn't require synchronization, as it's the caller's responsibility to not generate interrupts it cares about until after its handler is registered, i.e. there can't be a relevant in-flight interrupt.
Signed-off-by Sean Christopherson <[email protected]>
Seems a bit overengineered, considering there are just two callers. Doing
arch/x86/kernel/irq.c +3 -2
@@ -291,9 +291,10 @@ void kvm_set_posted_intr_wakeup_handler(void (*handler)(void)) {if (handler) kvm_posted_intr_wakeup_handler = handler; -else +else { kvm_posted_intr_wakeup_handler = dummy_handler; -synchronize_rcu (); +synchronize_rcu (); + } }EXPORT_SYMBOL_GPL (kvm_set_posted_intr_wakeup_handler);
on top of path 1 ensures that the synchronization is only done on the unregistration path.
Paolo
Quoted 96 lines
Signature · 4 lines
arch/x86/include/asm/irq.h | 3 ++- arch/x86/kernel/irq.c | 29 ++++++++++++++++++++--------- arch/x86/kvm/vmx/vmx.c | 4 ++-- 3 files changed, 24 insertions(+), 12 deletions(-)arch/x86/include/asm/irq.h +2
-1@@ -30,7 +30,8 @@ struct irq_desc;extern void fixup_irqs (void );#ifdef CONFIG_HAVE_KVM -extern void kvm_set_posted_intr_wakeup_handler (void (*handler)(void )); +extern void kvm_register_posted_intr_wakeup_handler (void (*handler)(void )); +extern void kvm_unregister_posted_intr_wakeup_handler (void (*handler)(void ));#endif extern void (*x86_platform_ipi_callback)(void );arch/x86/kernel/irq.c +20
-9@@ -284,18 +284,26 @@ DEFINE_IDTENTRY_SYSVEC(sysvec_x86_platform_ipi)#endif #ifdef CONFIG_HAVE_KVM -static void dummy_handler (void ) {} -static void (*kvm_posted_intr_wakeup_handler)(void ) = dummy_handler; +static void (*kvm_posted_intr_wakeup_handler)(void ); -void kvm_set_posted_intr_wakeup_handler (void (*handler)(void )) +void kvm_register_posted_intr_wakeup_handler (void (*handler)(void )) { -if (handler) - kvm_posted_intr_wakeup_handler = handler; -else - kvm_posted_intr_wakeup_handler = dummy_handler; +if (WARN_ON_ONCE (!handler || kvm_posted_intr_wakeup_handler)) +return ; + +WRITE_ONCE (kvm_posted_intr_wakeup_handler, handler); +} +EXPORT_SYMBOL_GPL (kvm_register_posted_intr_wakeup_handler); + +void kvm_unregister_posted_intr_wakeup_handler (void (*handler)(void )) +{ +if (WARN_ON_ONCE (!handler || handler != kvm_posted_intr_wakeup_handler)) +return ; + +WRITE_ONCE (kvm_posted_intr_wakeup_handler,NULL );synchronize_rcu (); } -EXPORT_SYMBOL_GPL (kvm_set_posted_intr_wakeup_handler); +EXPORT_SYMBOL_GPL (kvm_unregister_posted_intr_wakeup_handler);/* * Handler for POSTED_INTERRUPT_VECTOR. @@ -311,9 +319,12 @@ DEFINE_IDTENTRY_SYSVEC_SIMPLE(sysvec_kvm_posted_intr_ipi)*/ DEFINE_IDTENTRY_SYSVEC (sysvec_kvm_posted_intr_wakeup_ipi) { +void (*handler)(void ) =READ_ONCE (kvm_posted_intr_wakeup_handler); +ack_APIC_irq ();inc_irq_stat (kvm_posted_intr_wakeup_ipis); -kvm_posted_intr_wakeup_handler (); +if (handler) +handler (); }/* arch/x86/kvm/vmx/vmx.c +2
-2@@ -7553,7 +7553,7 @@ static void vmx_migrate_timers(struct kvm_vcpu *vcpu)static void hardware_unsetup (void ) { -kvm_set_posted_intr_wakeup_handler (NULL ); +kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler);if (nested)nested_vmx_hardware_unsetup ();@@ -7907,7 +7907,7 @@ static __init int hardware_setup(void)if (r)nested_vmx_hardware_unsetup (); -kvm_set_posted_intr_wakeup_handler (pi_wakeup_handler); +kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler);return r; }
Sean Christopherson
maintainerPATCH 4/4 KVM: VMX: Register posted interrupt wakeup handler iff APICv is enabled
In reply to Sean ChristophersonDon't bother registering a posted interrupt wakeup handler if APICv is disabled, KVM utilizes the wakeup vector if and only if APICv is enabled. Practically speaking, there's no meaningful functional change as KVM's wakeup handler is a glorified nop if there are no vCPUs using posted interrupts, not to mention that nothing in the system should be sending wakeup interrupts when APICv is disabled.
--- arch/x86/kvm/vmx/vmx.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/x86/kvm/vmx/vmx.c +4 -2
@@ -7553,7 +7553,8 @@ static void vmx_migrate_timers(struct kvm_vcpu *vcpu)static void hardware_unsetup (void ) { -kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler); +if (enable_apicv) +kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler);if (nested)nested_vmx_hardware_unsetup ();
@@ -7907,7 +7908,8 @@ static __init int hardware_setup(void)if (r)nested_vmx_hardware_unsetup (); -kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler); +if (enable_apicv) +kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler);return r; }
Signature · 1 line
2.33.0.882.g93a45727a2-goog
Paolo Bonzini
maintainerRe: [PATCH 4/4] KVM: VMX: Register posted interrupt wakeup handler iff APICv is enabled
Quoted 27 lines · Sean Christopherson
Don't bother registering a posted interrupt wakeup handler if APICv is disabled, KVM utilizes the wakeup vector if and only if APICv is enabled. Practically speaking, there's no meaningful functional change as KVM's wakeup handler is a glorified nop if there are no vCPUs using posted interrupts, not to mention that nothing in the system should be sending wakeup interrupts when APICv is disabled.
Signed-off-by Sean Christopherson <[email protected]>--- arch/x86/kvm/vmx/vmx.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)arch/x86/kvm/vmx/vmx.c +2
-1@@ -7553,7 +7553,8 @@ static void vmx_migrate_timers(struct kvm_vcpu *vcpu)static void hardware_unsetup (void ) { -kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler); +if (enable_apicv) +kvm_unregister_posted_intr_wakeup_handler (pi_wakeup_handler);if (nested)nested_vmx_hardware_unsetup ();
arch/x86/kvm/vmx/vmx.c static __init int hardware_setup(void) · line 7915
Quoted 9 lines
arch/x86/kvm/vmx/vmx.c +2
-1@@ -7907,7 +7908,8 @@ static __init int hardware_setup(void)if (r)nested_vmx_hardware_unsetup (); -kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler); +if (enable_apicv) +kvm_register_posted_intr_wakeup_handler (pi_wakeup_handler);return r; }
Also adds unnecessary complexity (you have to ensure the condition cannot change from one call to the other), so I am only queuing patches 1 and 2.
Paolo