| Message ID | 20190911155125.11932-3-eric.auger@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | KVM/ARM: Fix >256 vcpus | expand |
On Wed, Sep 11, 2019 at 05:51:24PM +0200, Eric Auger wrote: > Host kernels that expose the KVM_CAP_ARM_IRQ_LINE_LAYOUT_2 capability > allow injection of interrupts along with vcpu ids larger than 255. > Let's encode the vpcu id on 12 bits according to the upgraded KVM_IRQ_LINE > ABI when needed. > > Given that we have two callsites that need to assemble > the value for kvm_set_irq(), a new helper routine, kvm_arm_set_irq > is introduced. > > Without that patch qemu exits with "kvm_set_irq: Invalid argument" > message. > > Signed-off-by: Eric Auger <eric.auger@redhat.com> > Reported-by: Zenghui Yu <yuzenghui@huawei.com> > --- > hw/intc/arm_gic_kvm.c | 7 ++----- > target/arm/cpu.c | 10 ++++------ > target/arm/kvm.c | 16 ++++++++++++++++ > target/arm/kvm_arm.h | 1 + > 4 files changed, 23 insertions(+), 11 deletions(-) > > diff --git a/hw/intc/arm_gic_kvm.c b/hw/intc/arm_gic_kvm.c > index b56fda144f..9deb15e7e6 100644 > --- a/hw/intc/arm_gic_kvm.c > +++ b/hw/intc/arm_gic_kvm.c > @@ -55,7 +55,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) > * has separate fields in the irq number for type, > * CPU number and interrupt number. > */ > - int kvm_irq, irqtype, cpu; > + int irqtype, cpu; > > if (irq < (num_irq - GIC_INTERNAL)) { > /* External interrupt. The kernel numbers these like the GIC > @@ -72,10 +72,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) > cpu = irq / GIC_INTERNAL; > irq %= GIC_INTERNAL; > } > - kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) > - | (cpu << KVM_ARM_IRQ_VCPU_SHIFT) | irq; > - > - kvm_set_irq(kvm_state, kvm_irq, !!level); > + kvm_arm_set_irq(cpu, irqtype, irq, !!level); > } > > static void kvm_arm_gicv2_set_irq(void *opaque, int irq, int level) > diff --git a/target/arm/cpu.c b/target/arm/cpu.c > index 2399c14471..13813fb213 100644 > --- a/target/arm/cpu.c > +++ b/target/arm/cpu.c > @@ -576,16 +576,16 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) > ARMCPU *cpu = opaque; > CPUARMState *env = &cpu->env; > CPUState *cs = CPU(cpu); > - int kvm_irq = KVM_ARM_IRQ_TYPE_CPU << KVM_ARM_IRQ_TYPE_SHIFT; > uint32_t linestate_bit; > + int irq_id; > > switch (irq) { > case ARM_CPU_IRQ: > - kvm_irq |= KVM_ARM_IRQ_CPU_IRQ; > + irq_id = KVM_ARM_IRQ_CPU_IRQ; > linestate_bit = CPU_INTERRUPT_HARD; > break; > case ARM_CPU_FIQ: > - kvm_irq |= KVM_ARM_IRQ_CPU_FIQ; > + irq_id = KVM_ARM_IRQ_CPU_FIQ; > linestate_bit = CPU_INTERRUPT_FIQ; > break; > default: > @@ -597,9 +597,7 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) > } else { > env->irq_line_state &= ~linestate_bit; > } > - > - kvm_irq |= cs->cpu_index << KVM_ARM_IRQ_VCPU_SHIFT; > - kvm_set_irq(kvm_state, kvm_irq, level ? 1 : 0); > + kvm_arm_set_irq(cs->cpu_index, KVM_ARM_IRQ_TYPE_CPU, irq_id, !!level); > #endif > } > > diff --git a/target/arm/kvm.c b/target/arm/kvm.c > index b2eaa50b8d..6cdfa2204f 100644 > --- a/target/arm/kvm.c > +++ b/target/arm/kvm.c > @@ -744,6 +744,22 @@ int kvm_arm_vgic_probe(void) > } > } > > +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level) > +{ > + int kvm_irq = 0; No need to init to zero, and could just immediately init with the line below instead. > + > + kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) | irq; > + > + if (cpu != 0) { No need for this 'if' > + int cpu_idx2 = cpu / 256; > + int cpu_idx1 = cpu % 256; > + > + kvm_irq |= (cpu_idx1 << KVM_ARM_IRQ_VCPU_SHIFT) | > + ((cpu_idx2 & KVM_ARM_IRQ_VCPU2_MASK) << KVM_ARM_IRQ_VCPU2_SHIFT); Masking should be unnecessary as the only way it'll do anything is if we have vcpus >= 4096, which I imagine will never happen or will be guarded against happening elsewhere. Silently masking doesn't look right anyway, so I'd either add an assert(cpu_idx2 < 16) and drop the masking or just drop the masking. > + } > + return kvm_set_irq(kvm_state, kvm_irq, !!level); > +} > + > int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, > uint64_t address, uint32_t data, PCIDevice *dev) > { > diff --git a/target/arm/kvm_arm.h b/target/arm/kvm_arm.h > index b3106c8600..b4e19457a0 100644 > --- a/target/arm/kvm_arm.h > +++ b/target/arm/kvm_arm.h > @@ -253,6 +253,7 @@ int kvm_arm_vgic_probe(void); > > void kvm_arm_pmu_set_irq(CPUState *cs, int irq); > void kvm_arm_pmu_init(CPUState *cs); > +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level); > > #else > > -- > 2.20.1 > > Thanks, drew
Hi Drew, On 9/12/19 9:36 AM, Andrew Jones wrote: > On Wed, Sep 11, 2019 at 05:51:24PM +0200, Eric Auger wrote: >> Host kernels that expose the KVM_CAP_ARM_IRQ_LINE_LAYOUT_2 capability >> allow injection of interrupts along with vcpu ids larger than 255. >> Let's encode the vpcu id on 12 bits according to the upgraded KVM_IRQ_LINE >> ABI when needed. >> >> Given that we have two callsites that need to assemble >> the value for kvm_set_irq(), a new helper routine, kvm_arm_set_irq >> is introduced. >> >> Without that patch qemu exits with "kvm_set_irq: Invalid argument" >> message. >> >> Signed-off-by: Eric Auger <eric.auger@redhat.com> >> Reported-by: Zenghui Yu <yuzenghui@huawei.com> >> --- >> hw/intc/arm_gic_kvm.c | 7 ++----- >> target/arm/cpu.c | 10 ++++------ >> target/arm/kvm.c | 16 ++++++++++++++++ >> target/arm/kvm_arm.h | 1 + >> 4 files changed, 23 insertions(+), 11 deletions(-) >> >> diff --git a/hw/intc/arm_gic_kvm.c b/hw/intc/arm_gic_kvm.c >> index b56fda144f..9deb15e7e6 100644 >> --- a/hw/intc/arm_gic_kvm.c >> +++ b/hw/intc/arm_gic_kvm.c >> @@ -55,7 +55,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) >> * has separate fields in the irq number for type, >> * CPU number and interrupt number. >> */ >> - int kvm_irq, irqtype, cpu; >> + int irqtype, cpu; >> >> if (irq < (num_irq - GIC_INTERNAL)) { >> /* External interrupt. The kernel numbers these like the GIC >> @@ -72,10 +72,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) >> cpu = irq / GIC_INTERNAL; >> irq %= GIC_INTERNAL; >> } >> - kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) >> - | (cpu << KVM_ARM_IRQ_VCPU_SHIFT) | irq; >> - >> - kvm_set_irq(kvm_state, kvm_irq, !!level); >> + kvm_arm_set_irq(cpu, irqtype, irq, !!level); >> } >> >> static void kvm_arm_gicv2_set_irq(void *opaque, int irq, int level) >> diff --git a/target/arm/cpu.c b/target/arm/cpu.c >> index 2399c14471..13813fb213 100644 >> --- a/target/arm/cpu.c >> +++ b/target/arm/cpu.c >> @@ -576,16 +576,16 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) >> ARMCPU *cpu = opaque; >> CPUARMState *env = &cpu->env; >> CPUState *cs = CPU(cpu); >> - int kvm_irq = KVM_ARM_IRQ_TYPE_CPU << KVM_ARM_IRQ_TYPE_SHIFT; >> uint32_t linestate_bit; >> + int irq_id; >> >> switch (irq) { >> case ARM_CPU_IRQ: >> - kvm_irq |= KVM_ARM_IRQ_CPU_IRQ; >> + irq_id = KVM_ARM_IRQ_CPU_IRQ; >> linestate_bit = CPU_INTERRUPT_HARD; >> break; >> case ARM_CPU_FIQ: >> - kvm_irq |= KVM_ARM_IRQ_CPU_FIQ; >> + irq_id = KVM_ARM_IRQ_CPU_FIQ; >> linestate_bit = CPU_INTERRUPT_FIQ; >> break; >> default: >> @@ -597,9 +597,7 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) >> } else { >> env->irq_line_state &= ~linestate_bit; >> } >> - >> - kvm_irq |= cs->cpu_index << KVM_ARM_IRQ_VCPU_SHIFT; >> - kvm_set_irq(kvm_state, kvm_irq, level ? 1 : 0); >> + kvm_arm_set_irq(cs->cpu_index, KVM_ARM_IRQ_TYPE_CPU, irq_id, !!level); >> #endif >> } >> >> diff --git a/target/arm/kvm.c b/target/arm/kvm.c >> index b2eaa50b8d..6cdfa2204f 100644 >> --- a/target/arm/kvm.c >> +++ b/target/arm/kvm.c >> @@ -744,6 +744,22 @@ int kvm_arm_vgic_probe(void) >> } >> } >> >> +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level) >> +{ >> + int kvm_irq = 0; > > No need to init to zero, and could just immediately init with the > line below instead. > >> + >> + kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) | irq; >> + >> + if (cpu != 0) { > > No need for this 'if' > >> + int cpu_idx2 = cpu / 256; >> + int cpu_idx1 = cpu % 256; >> + >> + kvm_irq |= (cpu_idx1 << KVM_ARM_IRQ_VCPU_SHIFT) | >> + ((cpu_idx2 & KVM_ARM_IRQ_VCPU2_MASK) << KVM_ARM_IRQ_VCPU2_SHIFT); > > Masking should be unnecessary as the only way it'll do anything is if we > have vcpus >= 4096, which I imagine will never happen or will be guarded > against happening elsewhere. Silently masking doesn't look right anyway, > so I'd either add an assert(cpu_idx2 < 16) and drop the masking or just > drop the masking. All that makes sense. Thank you for the review. Eric > >> + } >> + return kvm_set_irq(kvm_state, kvm_irq, !!level); >> +} >> + >> int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, >> uint64_t address, uint32_t data, PCIDevice *dev) >> { >> diff --git a/target/arm/kvm_arm.h b/target/arm/kvm_arm.h >> index b3106c8600..b4e19457a0 100644 >> --- a/target/arm/kvm_arm.h >> +++ b/target/arm/kvm_arm.h >> @@ -253,6 +253,7 @@ int kvm_arm_vgic_probe(void); >> >> void kvm_arm_pmu_set_irq(CPUState *cs, int irq); >> void kvm_arm_pmu_init(CPUState *cs); >> +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level); >> >> #else >> >> -- >> 2.20.1 >> >> > > Thanks, > drew >
diff --git a/hw/intc/arm_gic_kvm.c b/hw/intc/arm_gic_kvm.c index b56fda144f..9deb15e7e6 100644 --- a/hw/intc/arm_gic_kvm.c +++ b/hw/intc/arm_gic_kvm.c @@ -55,7 +55,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) * has separate fields in the irq number for type, * CPU number and interrupt number. */ - int kvm_irq, irqtype, cpu; + int irqtype, cpu; if (irq < (num_irq - GIC_INTERNAL)) { /* External interrupt. The kernel numbers these like the GIC @@ -72,10 +72,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq, int level) cpu = irq / GIC_INTERNAL; irq %= GIC_INTERNAL; } - kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) - | (cpu << KVM_ARM_IRQ_VCPU_SHIFT) | irq; - - kvm_set_irq(kvm_state, kvm_irq, !!level); + kvm_arm_set_irq(cpu, irqtype, irq, !!level); } static void kvm_arm_gicv2_set_irq(void *opaque, int irq, int level) diff --git a/target/arm/cpu.c b/target/arm/cpu.c index 2399c14471..13813fb213 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -576,16 +576,16 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) ARMCPU *cpu = opaque; CPUARMState *env = &cpu->env; CPUState *cs = CPU(cpu); - int kvm_irq = KVM_ARM_IRQ_TYPE_CPU << KVM_ARM_IRQ_TYPE_SHIFT; uint32_t linestate_bit; + int irq_id; switch (irq) { case ARM_CPU_IRQ: - kvm_irq |= KVM_ARM_IRQ_CPU_IRQ; + irq_id = KVM_ARM_IRQ_CPU_IRQ; linestate_bit = CPU_INTERRUPT_HARD; break; case ARM_CPU_FIQ: - kvm_irq |= KVM_ARM_IRQ_CPU_FIQ; + irq_id = KVM_ARM_IRQ_CPU_FIQ; linestate_bit = CPU_INTERRUPT_FIQ; break; default: @@ -597,9 +597,7 @@ static void arm_cpu_kvm_set_irq(void *opaque, int irq, int level) } else { env->irq_line_state &= ~linestate_bit; } - - kvm_irq |= cs->cpu_index << KVM_ARM_IRQ_VCPU_SHIFT; - kvm_set_irq(kvm_state, kvm_irq, level ? 1 : 0); + kvm_arm_set_irq(cs->cpu_index, KVM_ARM_IRQ_TYPE_CPU, irq_id, !!level); #endif } diff --git a/target/arm/kvm.c b/target/arm/kvm.c index b2eaa50b8d..6cdfa2204f 100644 --- a/target/arm/kvm.c +++ b/target/arm/kvm.c @@ -744,6 +744,22 @@ int kvm_arm_vgic_probe(void) } } +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level) +{ + int kvm_irq = 0; + + kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) | irq; + + if (cpu != 0) { + int cpu_idx2 = cpu / 256; + int cpu_idx1 = cpu % 256; + + kvm_irq |= (cpu_idx1 << KVM_ARM_IRQ_VCPU_SHIFT) | + ((cpu_idx2 & KVM_ARM_IRQ_VCPU2_MASK) << KVM_ARM_IRQ_VCPU2_SHIFT); + } + return kvm_set_irq(kvm_state, kvm_irq, !!level); +} + int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, uint64_t address, uint32_t data, PCIDevice *dev) { diff --git a/target/arm/kvm_arm.h b/target/arm/kvm_arm.h index b3106c8600..b4e19457a0 100644 --- a/target/arm/kvm_arm.h +++ b/target/arm/kvm_arm.h @@ -253,6 +253,7 @@ int kvm_arm_vgic_probe(void); void kvm_arm_pmu_set_irq(CPUState *cs, int irq); void kvm_arm_pmu_init(CPUState *cs); +int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level); #else
Host kernels that expose the KVM_CAP_ARM_IRQ_LINE_LAYOUT_2 capability allow injection of interrupts along with vcpu ids larger than 255. Let's encode the vpcu id on 12 bits according to the upgraded KVM_IRQ_LINE ABI when needed. Given that we have two callsites that need to assemble the value for kvm_set_irq(), a new helper routine, kvm_arm_set_irq is introduced. Without that patch qemu exits with "kvm_set_irq: Invalid argument" message. Signed-off-by: Eric Auger <eric.auger@redhat.com> Reported-by: Zenghui Yu <yuzenghui@huawei.com> --- hw/intc/arm_gic_kvm.c | 7 ++----- target/arm/cpu.c | 10 ++++------ target/arm/kvm.c | 16 ++++++++++++++++ target/arm/kvm_arm.h | 1 + 4 files changed, 23 insertions(+), 11 deletions(-)