kvm: fix race with level interrupts
PATCH 1/1 kvm: fix race with level interrupts
When more than 1 source id is in use for the same GSI, we have the following race related to handling irq_states:
CPU 0 clears bit in irq_states. CPU 0 reads irq_state as 0. CPU 1 sets bit in irq_states. CPU 1 calls kvm_ioapic_set_irq(1). CPU 0 calls kvm_ioapic_set_irq(0).
Now ioapic thinks the level is 0 but irq_state is not 0.
Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all irq_states bitmap handling under pic/ioapic lock. This also removes the need for atomics with irq_states handling.
---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.
arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)
arch/x86/include/asm/kvm_host.h +14 -1
@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level) +{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);
arch/x86/kvm/i8259.c +12 -2
@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,
@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */
virt/kvm/ioapic.c +12 -1
@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;
@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {
virt/kvm/ioapic.h +3 -1
@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);
virt/kvm/irq_comm.c +4 -27
@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif
@@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)
@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);
@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Signature · 1 line
MST
Quoted 65 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +14
-1@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level) +{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +0
-0
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 191
Quoted 5 lines
arch/x86/kvm/i8259.c +1
-1@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l)
Please use "irq_source_id" and keep "level". I hate reviewing code where I have to differentiate 'l' vs '1'. Same below, mixing src_id and irq_source_id in various places.
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 194
Quoted 4 lines
arch/x86/kvm/i8259.c +1
-1{ -struct kvm_pic *s = opaque;int ret =-1 ; +int level;
And you'd avoid this variable
Quoted 149 lines
arch/x86/kvm/i8259.c +10
-0pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 69 lines · Alex Williamson
Quoted 68 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +14
-1@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level) +{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +0
-0@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); }
Quoted 5 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 191
Quoted 2 lines
arch/x86/kvm/i8259.c +1
-1-int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l)Please use "
irq_source_id" and keep "level". I hate reviewing code where I have to differentiate 'l' vs '1'.
l is an illegal variable name? Switch to a different font.
Quoted 2 lines
Same below, mixing
src_idandirq_source_idin various places.
irq_source_id is too long, coding style says variable names should be short:
LOCAL variable names should be short, and to the point.
Quoted 6 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 194
Quoted 4 lines
arch/x86/kvm/i8259.c +1
-1{ -struct kvm_pic *s = opaque;int ret =-1 ; +int level;And you'd avoid this variable
by giving a different meaning to the same variable in different parts of a function? Now *that* is a bad idea.
l is level for a specific source, level is the resulting irq value.
Quoted 151 lines
Quoted 149 lines
arch/x86/kvm/i8259.c +10
-0pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 70 lines · Michael S. Tsirkin
Quoted 69 lines · Alex Williamson
Quoted 68 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +14
-1@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level) +{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +0
-0@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); }
Quoted 7 lines
Quoted 5 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 191
Quoted 2 lines
arch/x86/kvm/i8259.c +1
-1-int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l)Please use "
irq_source_id" and keep "level". I hate reviewing code where I have to differentiate 'l' vs '1'.l is an illegal variable name? Switch to a different font.
WTF
Quoted 6 lines
Quoted 2 lines
Same below, mixing
src_idandirq_source_idin various places.
irq_source_idis too long, coding style says variable names should be short:
LOCAL variable names should be short, and to the point.
irq_source_id is used all over the code, so either convert them all of be consistent.
Quoted 11 lines
Quoted 6 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 194
Quoted 4 lines
arch/x86/kvm/i8259.c +1
-1{ -struct kvm_pic *s = opaque;int ret =-1 ; +int level;And you'd avoid this variable
by giving a different meaning to the same variable in different parts of a function? Now *that* is a bad idea.
l is level for a specific source, level is the resulting irq value.
Then maybe 'l' isn't a meaningful variable name.
Quoted 155 lines
Quoted 136 lines
Quoted 136 lines
arch/x86/kvm/i8259.c +10
-0pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +2
-21@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock;Quoted 15 lines
virt/kvm/irq_comm.c void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id) · line 253
Quoted 13 lines
virt/kvm/irq_comm.c +2
-6-for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }--
To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 6 lines · Alex Williamson
I hate reviewing code
Quoted 3 lines
Quoted 1 line
where I have to differentiate 'l' vs '1'.
l is an illegal variable name? Switch to a different font.
WTF
Really, you should use a font where these differ. I currently use Monospace but there are many other good fonts. If you don't bugs sneak in.
int x = 11;
and
int x = 1l;
Should look very different.
Signature · 1 line
MST
Quoted 12 lines · Michael S. Tsirkin
Quoted 6 lines · Alex Williamson
I hate reviewing code
Quoted 3 lines
Quoted 1 line
where I have to differentiate 'l' vs '1'.
l is an illegal variable name? Switch to a different font.
WTF
Really, you should use a font where these differ. I currently use Monospace but there are many other good fonts. If you don't bugs sneak in.
Quoted 5 lines
int x = 11;
and
int x = 1l;Should look very different.
Don't be an asshole, my dog could tell the difference between those side-by-side. That's not what you posted. I use monospace as well and find it offensive that you suggest someone change fonts to review your code. Clearly, any time spent trying to identify a single character in isolation is not well spent.
Quoted 17 lines · Alex Williamson
Quoted 16 lines · Michael S. Tsirkin
Quoted 6 lines · Alex Williamson
I hate reviewing code
Quoted 3 lines
Quoted 1 line
where I have to differentiate 'l' vs '1'.
l is an illegal variable name? Switch to a different font.
WTF
Really, you should use a font where these differ. I currently use Monospace but there are many other good fonts. If you don't bugs sneak in.
int x = 11;
and
int x = 1l;
Quoted 8 lines
Quoted 1 line
Should look very different.
Don't be an asshole, my dog could tell the difference between those side-by-side. That's not what you posted. I use monospace as well and find it offensive that you suggest someone change fonts to review your code. Clearly, any time spent trying to identify a single character in isolation is not well spent.
I apologize for offending you, this was not my intent. I sent v2 to address your comments, thanks.
Signature · 1 line
MST
Quoted 70 lines · Michael S. Tsirkin
Quoted 69 lines · Alex Williamson
Quoted 68 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +14
-1@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level) +{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +0
-0@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); }
Quoted 8 lines
Quoted 5 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 191
Quoted 2 lines
arch/x86/kvm/i8259.c +1
-1-int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l)Please use "
irq_source_id" and keep "level". I hate reviewing code where I have to differentiate 'l' vs '1'.l is an illegal variable name? Switch to a different font.
No, but it does not make it good variable name. I agree with Alex here.
Quoted 6 lines
Quoted 2 lines
Same below, mixing
src_idandirq_source_idin various places.
irq_source_idis too long, coding style says variable names should be short:
LOCAL variable names should be short, and to the point.
Again I agree with Alex about consistency.
Quoted 8 lines
Quoted 7 lines
arch/x86/kvm/i8259.c void kvm_pic_update_irq(struct kvm_pic *s) · line 194
Quoted 5 lines
arch/x86/kvm/i8259.c +1
-1{ -struct kvm_pic *s = opaque;int ret =-1 ; +int level;And you'd avoid this variable
Quoted 5 lines
by giving a different meaning to the same variable in different parts of a function? Now *that* is a bad idea.
l is level for a specific source, level is the resulting irq value.
You can call it shared_level if you wish, but what's wrong with doing:
ret = pic_set_irq1(&s->pics[irq >> 3], irq & 7,
kvm_irq_line_state(&s->irq_states[irq], src_id, l));
Quoted 136 lines
Quoted 136 lines
Quoted 136 lines
arch/x86/kvm/i8259.c +10
-0pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +2
-21@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock;
Quoted 15 lines
Quoted 5 lines
Quoted 5 lines
virt/kvm/irq_comm.c +1
-4-for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);Quoted 10 lines
virt/kvm/irq_comm.c void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id) · line 253
Quoted 8 lines
virt/kvm/irq_comm.c +1
-2#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
--
Gleb.
Quoted 39 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 6 lines
arch/x86/include/asm/kvm_host.h +2
-1@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl); -int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)
This should probably be __kvm_irq_line_state given the calling restrictions.
Quoted 178 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);
Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.
It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.
Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Quoted 180 lines
Quoted 178 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 51 lines · Michael S. Tsirkin
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.
Quoted 8 lines
It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Please get off your high horse, this bug has existed for a long time and nobody has noticed. __ indicates locking and makes people think twice about arbitrarily calling it. Correct naming prevents future bugs, which is something you've been riding me on in other areas...
Quoted 180 lines
Quoted 180 lines
Quoted 178 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 58 lines · Alex Williamson
Quoted 57 lines · Michael S. Tsirkin
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.
Quoted 8 lines
Quoted 2 lines
Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Please get off your high horse, this bug has existed for a long time and nobody has noticed. __ indicates locking and makes people think twice about arbitrarily calling it. Correct naming prevents future bugs, which is something you've been riding me on in other areas...
Agree, this is not a recent regression and in fact the bug cannot be triggered without device assignment that uses level interrupt (is this works at all?).
--
Gleb.
Quoted 58 lines · Alex Williamson
Quoted 57 lines · Michael S. Tsirkin
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.
Quoted 7 lines
Quoted 2 lines
Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Please get off your high horse, this bug has existed for a long time and nobody has noticed. __ indicates locking and makes people think twice about arbitrarily calling it. Correct naming prevents future bugs, which is something you've been riding me on in other areas...
It's all fine but you already sent two messages about improving coding style in the same patch. What I really hoped for is a thorough review that looks at all issues at once, including the functional side as well. Or maybe style comments is all there is ... we'll see.
Quoted 182 lines
Quoted 180 lines
Quoted 180 lines
Quoted 178 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 67 lines · Michael S. Tsirkin
Quoted 58 lines · Alex Williamson
Quoted 57 lines · Michael S. Tsirkin
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.Quoted 7 lines
Quoted 2 lines
Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Please get off your high horse, this bug has existed for a long time and nobody has noticed. __ indicates locking and makes people think twice about arbitrarily calling it. Correct naming prevents future bugs, which is something you've been riding me on in other areas...
Quoted 4 lines
It's all fine but you already sent two messages about improving coding style in the same patch. What I really hoped for is a thorough review that looks at all issues at once, including the functional side as well. Or maybe style comments is all there is ... we'll see.
So you're accusing me of an superfluous review because I noted coding style errors? Very professional. My comments on your rfc should validate the review was thorough.
Quoted 182 lines
Quoted 182 lines
Quoted 180 lines
Quoted 180 lines
Quoted 178 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }
Quoted 68 lines · Alex Williamson
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Alex Williamson
Quoted 51 lines · Michael S. Tsirkin
Quoted 43 lines · Alex Williamson
Quoted 42 lines · Michael S. Tsirkin
When more than 1 source id is in use for the same GSI, we have the following race related to handling
irq_states:CPU 0 clears bit in
irq_states. CPU 0 readsirq_stateas 0. CPU 1 sets bit inirq_states. CPU 1 callskvm_ioapic_set_irq(1). CPU 0 callskvm_ioapic_set_irq(0).Now ioapic thinks the level is 0 but
irq_stateis not 0.Note that above is valid behaviour if CPU0 and CPU1 are using different source ids.
Fix by performing all
irq_statesbitmap handling under pic/ioapic lock. This also removes the need for atomics withirq_stateshandling.Reported-by Gleb Natapov <[email protected]>Signed-off-by Michael S. Tsirkin <[email protected]>---
This survives stress testing for me (though I only tried virtio, not device assignment).
Avi, Marcelo, though we have not observed this in the field, it's a bugfix so probably 3.5 material?
I assume yes so the patch is against 3.5-rc7.
Also stable? It's a very old bug.arch/x86/include/asm/kvm_host.h | 15 ++++++++++++++- arch/x86/kvm/i8259.c | 14 ++++++++++++-- virt/kvm/ioapic.c | 13 ++++++++++++- virt/kvm/ioapic.h | 4 +++- virt/kvm/irq_comm.c | 31 ++++--------------------------- 5 files changed, 45 insertions(+), 32 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -802,7 +802,20 @@ int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,void kvm_propagate_fault (struct kvm_vcpu *vcpu,struct x86_exception *fault);bool kvm_require_cpl (struct kvm_vcpu *vcpu,int required_cpl);Quoted 6 lines
arch/x86/include/asm/kvm_host.h int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, · line 806
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +2
-1-int kvm_pic_set_irq (void *opaque,int irq,int level); +static inline int kvm_irq_line_state (unsigned long *irq_state, +int irq_source_id,int level)This should probably be
__kvm_irq_line_stategiven the calling restrictions.Quoted 8 lines
It's such a trivial helper, do we need to split hairs about this?
Look it's not a good time for minor coding style nits.
3.5 is imminent, it's about 1am here and I really don't have time to retest today, so we'll release another kernel with a bug.Could you focus on reviewing the functionality and correctness, and leave ideas for better variable naming aside until 3.6?
Please get off your high horse, this bug has existed for a long time and nobody has noticed. __ indicates locking and makes people think twice about arbitrarily calling it. Correct naming prevents future bugs, which is something you've been riding me on in other areas...
Quoted 7 lines
Quoted 4 lines
It's all fine but you already sent two messages about improving coding style in the same patch. What I really hoped for is a thorough review that looks at all issues at once, including the functional side as well. Or maybe style comments is all there is ... we'll see.
So you're accusing me of an superfluous review because I noted coding style errors?
No, what I mean is that you sent multiple messages on different parts of the same patch which is confusing. It seems reasonable to ask for a single mail that looks at all issues at once with a small patch like this.
Quoted 2 lines
Very professional. My comments on your rfc should validate the review was thorough.
But you gave me no indication whether you have completed the review or only looked at style and more comments are forthcoming.
Quoted 185 lines
Quoted 183 lines
Quoted 183 lines
Quoted 181 lines
Quoted 181 lines
Quoted 179 lines
arch/x86/include/asm/kvm_host.h +12
-0+{ +/* Logical OR for level trig interrupt */ +if (level) +__set_bit (irq_source_id, irq_state); +else +__clear_bit (irq_source_id, irq_state); + +return !!(*irq_state); +} + +int kvm_pic_set_irq (struct kvm_pic *pic,int irq,int irq_source_id,int level); +void kvm_pic_clear_all (struct kvm_pic *pic,int irq_source_id);void kvm_inject_nmi (struct kvm_vcpu *vcpu);arch/x86/kvm/i8259.c +12
-2@@ -188,13 +188,14 @@ void kvm_pic_update_irq(struct kvm_pic *s)pic_unlock (s); } -int kvm_pic_set_irq (void *opaque,int irq,int level) +int kvm_pic_set_irq (struct kvm_pic *s,int irq,int src_id,int l) { -struct kvm_pic *s = opaque;int ret =-1 ; +int level;pic_lock (s);if (irq >=0 && irq <PIC_NUM_PINS ) { + level =kvm_irq_line_state (&s->irq_states[irq], src_id, l); ret =pic_set_irq1 (&s->pics[irq >>3 ], irq &7 , level);pic_update_irq (s); trace_kvm_pic_set_irq(irq >> 3, irq & 7, s->pics[irq >> 3].elcr,@@ -205,6 +206,15 @@ int kvm_pic_set_irq(void *opaque, int irq, int level)return ret; } +void kvm_pic_clear_all (struct kvm_pic *s,int src_id) +{ +int i; +pic_lock (s); +for (i =0 ; i <PIC_NUM_PINS ; i++) +__clear_bit (src_id, &s->irq_states[i]); +pic_unlock (s); +} +/* * acknowledge interrupt 'irq' */ virt/kvm/ioapic.c +12
-1@@ -191,17 +191,19 @@ static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)return kvm_irq_delivery_to_apic (ioapic->kvm,NULL , &irqe); } -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level) +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int src_id,int l) {u32 old_irr;u32 mask =1 << irq;union kvm_ioapic_redirect_entry entry;int ret =1 ; +int level;spin_lock (&ioapic->lock); old_irr = ioapic->irr;if (irq >=0 && irq <IOAPIC_NUM_PINS ) { entry = ioapic->redirtbl[irq]; + level =kvm_irq_line_state (&ioapic->irq_states[irq], src_id, l); level ^= entry.fields.polarity;if (!level) ioapic->irr &= ~mask;@@ -221,6 +223,15 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)return ret; } +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int src_id) +{ +int i; +spin_lock (&ioapic->lock); +for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) +__clear_bit (src_id, &ioapic->irq_states[i]); +spin_unlock (&ioapic->lock); +} +static void __kvm_ioapic_update_eoi (struct kvm_ioapic *ioapic,int vector,int trigger_mode) {virt/kvm/ioapic.h +3
-1@@ -74,7 +74,9 @@ void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode);bool kvm_ioapic_handles_vector (struct kvm *kvm,int vector);int kvm_ioapic_init (struct kvm *kvm);void kvm_ioapic_destroy (struct kvm *kvm); -int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int level); +int kvm_ioapic_set_irq (struct kvm_ioapic *ioapic,int irq,int irq_source_id, +int level); +void kvm_ioapic_clear_all (struct kvm_ioapic *ioapic,int irq_source_id);void kvm_ioapic_reset (struct kvm_ioapic *ioapic);int kvm_irq_delivery_to_apic (struct kvm *kvm,struct kvm_lapic *src,struct kvm_lapic_irq *irq);virt/kvm/irq_comm.c +4
-27@@ -33,26 +33,12 @@#include "ioapic.h" -static inline int kvm_irq_line_state (unsigned long *irq_state, -int irq_source_id,int level) -{ -/* Logical OR for level trig interrupt */ -if (level) -set_bit (irq_source_id, irq_state); -else -clear_bit (irq_source_id, irq_state); - -return !!(*irq_state); -} -static int kvm_set_pic_irq (struct kvm_kernel_irq_routing_entry *e,struct kvm *kvm,int irq_source_id,int level) {#ifdef CONFIG_X86 struct kvm_pic *pic =pic_irqchip (kvm); - level =kvm_irq_line_state (&pic->irq_states[e->irqchip.pin], - irq_source_id, level); -return kvm_pic_set_irq (pic, e->irqchip.pin, level); +return kvm_pic_set_irq (pic, e->irqchip.pin, irq_source_id, level);#else return -1 ;#endif @@ -62,10 +48,7 @@ static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, int irq_source_id, int level) { struct kvm_ioapic *ioapic = kvm->arch.vioapic; - level =kvm_irq_line_state (&ioapic->irq_states[e->irqchip.pin], - irq_source_id, level); - -return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, level); +return kvm_ioapic_set_irq (ioapic, e->irqchip.pin, irq_source_id, level); }inline static bool kvm_is_dm_lowest_prio (struct kvm_lapic_irq *irq)@@ -249,8 +232,6 @@ unlock:void kvm_free_irq_source_id (struct kvm *kvm,int irq_source_id) { -int i; -ASSERT (irq_source_id !=KVM_USERSPACE_IRQ_SOURCE_ID );mutex_lock (&kvm->irq_lock);@@ -263,14 +244,10 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)if (!irqchip_in_kernel (kvm)) goto unlock; -for (i =0 ; i <KVM_IOAPIC_NUM_PINS ; i++) { -clear_bit (irq_source_id, &kvm->arch.vioapic->irq_states[i]); -if (i >=16 ) -continue ; +kvm_ioapic_clear_all (kvm->arch.vioapic, irq_source_id);#ifdef CONFIG_X86 -clear_bit (irq_source_id, &pic_irqchip (kvm)->irq_states[i]); +kvm_pic_clear_all (pic_irqchip (kvm), irq_source_id);#endif - } unlock:mutex_unlock (&kvm->irq_lock); }