IRQ Routing
cover IRQ Routing
Following is my alternative to irq routing. The differences compared to Sheng's version are:
- A single ioctl to replace the entire routing table, instead of add/remove ioctls for individual routing entries. Routing changes are rare, and we need to track the entire table in userspace anyway (for save/restore, and for user irqchip). As a side effect changes are atomic.
- Interrupt numbers are allocated by userspace, instead of the kernel
- I implemented irqchip routings rather then MSIs, it should be easy to add MSIs later on.
Please review and comment.
Avi Kivity (1):
KVM: Userspace controlled irq routing
arch/ia64/kvm/kvm-ia64.c | 5 ++ arch/x86/kvm/x86.c | 6 ++ include/linux/kvm.h | 33 +++++++++ include/linux/kvm_host.h | 31 +++++++++ virt/kvm/irq_comm.c | 168 ++++++++++++++++++++++++++++++++++++++++++++- virt/kvm/kvm_main.c | 25 +++++++ 6 files changed, 264 insertions(+), 4 deletions(-)
PATCH 1/1 KVM: Userspace controlled irq routing
Currently KVM has a static routing from GSI numbers to interrupts (namely, 0-15 are mapped 1:1 to both PIC and IOAPIC, and 16:23 are mapped 1:1 to the IOAPIC). This is insufficient for several reasons:
- HPET requires non 1:1 mapping for the timer interrupt
- MSIs need a new method to assign interrupt numbers and dispatch them
- ACPI APIC mode needs to be able to reassign the PCI LINK interrupts to the ioapics
This patch implements an interrupt routing table (as a linked list, but this can be easily changed) and a userspace interface to replace the table. The routing table is initialized according to the current hardwired mapping.
--- arch/ia64/kvm/kvm-ia64.c | 5 ++ arch/x86/kvm/x86.c | 6 ++ include/linux/kvm.h | 33 +++++++++ include/linux/kvm_host.h | 31 +++++++++ virt/kvm/irq_comm.c | 168 ++++++++++++++++++++++++++++++++++++++++++++- virt/kvm/kvm_main.c | 25 +++++++ 6 files changed, 264 insertions(+), 4 deletions(-)
arch/ia64/kvm/kvm-ia64.c +5 -0
@@ -919,6 +919,11 @@ long kvm_arch_vm_ioctl(struct file *filp, r =kvm_ioapic_init (kvm);if (r) goto out; + r =kvm_setup_default_irq_routing (kvm); +if (r) { +kfree (kvm->arch.vioapic); + goto out; + }break ;case KVM_IRQ_LINE : {struct kvm_irq_level irq_event;
arch/x86/kvm/x86.c +6 -0
@@ -1833,6 +1833,12 @@ long kvm_arch_vm_ioctl(struct file *filp, } }else goto out; + r =kvm_setup_default_irq_routing (kvm); +if (r) { +kfree (kvm->arch.vpic); +kfree (kvm->arch.vioapic); + goto out; + }break ;case KVM_CREATE_PIT :mutex_lock (&kvm->lock);
include/linux/kvm.h +33 -0
@@ -399,6 +399,38 @@ struct kvm_trace_rec {#if defined(CONFIG_X86 )#define KVM_CAP_REINJECT_CONTROL 24#endif +#if defined(CONFIG_X86 )||defined(CONFIG_IA64 ) +#define KVM_CAP_IRQ_ROUTING 25 +#endif + +#ifdef KVM_CAP_IRQ_ROUTING + +struct kvm_irq_routing_irqchip { +__u32 irqchip; +__u32 pin; +}; + +/* gsi routing entry types */ +#define KVM_IRQ_ROUTING_IRQCHIP 1 + +struct kvm_irq_routing_entry { +__u32 gsi; +__u32 type; +__u32 flags; +__u32 pad; +union { +struct kvm_irq_routing_irqchip irqchip; +__u32 pad[8 ]; + } u; +}; + +struct kvm_irq_routing { +__u32 nr; +__u32 flags; +struct kvm_irq_routing_entry entries[0 ]; +}; + +#endif /* * ioctls for VM fds
@@ -430,6 +462,7 @@ struct kvm_trace_rec { _IOW(KVMIO, 0x68, struct kvm_coalesced_mmio_zone)#define KVM_ASSIGN_PCI_DEVICE _IOR(KVMIO, 0x69, \ struct kvm_assigned_pci_dev) +#define KVM_SET_GSI_ROUTING _IOW(KVMIO, 0x6a, struct kvm_irq_routing)#define KVM_ASSIGN_IRQ _IOR(KVMIO, 0x70, \ struct kvm_assigned_irq)#define KVM_REINJECT_CONTROL _IO(KVMIO, 0x71)
include/linux/kvm_host.h +31 -0
@@ -107,6 +107,19 @@ struct kvm_memory_slot {int user_alloc; }; +struct kvm_kernel_irq_routing_entry { +u32 gsi; +void (*set)(struct kvm_kernel_irq_routing_entry *e, +struct kvm *kvm,int level); +union { +struct { +unsigned irqchip; +unsigned pin; + } irqchip; + }; +struct list_head link; +}; +struct kvm {struct mutex lock;/* protects the vcpus array and APIC accesses */ spinlock_t mmu_lock;
@@ -128,6 +141,7 @@ struct kvm {#endif #ifdef CONFIG_HAVE_KVM_IRQCHIP +struct list_head irq_routing;/* of kvm_kernel_irq_routing_entry */ struct hlist_head mask_notifier_list;#endif
@@ -480,4 +494,21 @@ static inline int mmu_notifier_retry(struct kvm_vcpu *vcpu, unsigned long mmu_se }#endif +#ifdef CONFIG_HAVE_KVM_IRQCHIP + +#define KVM_MAX_IRQ_ROUTES 1024 + +int kvm_setup_default_irq_routing (struct kvm *kvm); +int kvm_set_irq_routing (struct kvm *kvm, +const struct kvm_irq_routing_entry *entries, +unsigned nr, +unsigned flags); +void kvm_free_irq_routing (struct kvm *kvm); + +#else + +static inline void kvm_free_irq_routing (struct kvm *kvm) {} + +#endif +#endif
virt/kvm/irq_comm.c +164 -4
186 lines not on this page · raw message
virt/kvm/kvm_main.c +25 -0
@@ -840,6 +840,7 @@ static struct kvm *kvm_create_vm(void)if (IS_ERR (kvm)) goto out;#ifdef CONFIG_HAVE_KVM_IRQCHIP +INIT_LIST_HEAD (&kvm->irq_routing);INIT_HLIST_HEAD (&kvm->mask_notifier_list);#endif
@@ -923,6 +924,7 @@ static void kvm_destroy_vm(struct kvm *kvm)spin_lock (&kvm_lock);list_del (&kvm->vm_list);spin_unlock (&kvm_lock); +kvm_free_irq_routing (kvm);kvm_io_bus_destroy (&kvm->pio_bus);kvm_io_bus_destroy (&kvm->mmio_bus);#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
@@ -1943,6 +1945,29 @@ static long kvm_vm_ioctl(struct file *filp,break ; }#endif +#ifdef KVM_CAP_IRQ_ROUTING +case KVM_SET_GSI_ROUTING : { +struct kvm_irq_routing routing; +struct kvm_irq_routing_entry *entries; + + r = -EFAULT ; +if (copy_from_user (&routing, argp,sizeof (routing))) + goto out; + r = -EINVAL ; +if (routing.nr >=KVM_MAX_IRQ_ROUTES ) + goto out; +if (routing.flags) + goto out; + r = -ENOMEM ; + entries =vmalloc (routing.nr *sizeof (*entries)); +if (!entries) + goto out; + r =kvm_set_irq_routing (kvm, entries, routing.nr, + routing.flags); +vfree (entries); +break ; + } +#endif default : r =kvm_arch_vm_ioctl (filp, ioctl, arg); }
Signature · 1 line
1.6.0.6
Quoted 11 lines · On Wednesday 14 January 2009 04:02:29 Avi Kivity wrote:
Following is my alternative to irq routing. The differences compared to Sheng's version are:
- A single ioctl to replace the entire routing table, instead of add/remove ioctls for individual routing entries. Routing changes are rare, and we need to track the entire table in userspace anyway (for save/restore, and for user irqchip). As a side effect changes are atomic.
- Interrupt numbers are allocated by userspace, instead of the kernel
- I implemented irqchip routings rather then MSIs, it should be easy to add MSIs later on.
Quoted 1 line
Please review and comment.
Look nice to me now... Save/restore is a good reason to maintain a table in userspace.
And looking forward to userspace patch. Put a table in kvm_context?
Signature · 2 lines
regards Yang, Sheng
Quoted 12 lines
Avi Kivity (1):
KVM: Userspace controlled irq routingarch/ia64/kvm/kvm-ia64.c | 5 ++ arch/x86/kvm/x86.c | 6 ++ include/linux/kvm.h | 33 +++++++++ include/linux/kvm_host.h | 31 +++++++++ virt/kvm/irq_comm.c | 168 ++++++++++++++++++++++++++++++++++++++++++++- virt/kvm/kvm_main.c | 25 +++++++ 6 files changed, 264 insertions(+), 4 deletions(-)
Quoted 13 lines · Sheng Yang
Quoted 11 lines · On Wednesday 14 January 2009 04:02:29 Avi Kivity wrote:
Following is my alternative to irq routing. The differences compared to Sheng's version are:
- A single ioctl to replace the entire routing table, instead of add/remove ioctls for individual routing entries. Routing changes are rare, and we need to track the entire table in userspace anyway (for save/restore, and for user irqchip). As a side effect changes are atomic.
- Interrupt numbers are allocated by userspace, instead of the kernel
- I implemented irqchip routings rather then MSIs, it should be easy to add MSIs later on.
Quoted 7 lines
Quoted 2 lines
Please review and comment.
Look nice to me now... Save/restore is a good reason to maintain a table in userspace.
And looking forward to userspace patch. Put a table in
kvm_context?
Sent out. I indeed placed a table in kvm_context, with add/del route APIs.
Signature · 1 line
error compiling committee.c: too many arguments to function