ARM: KVM: drop arbitrary limitation to 4 CPU VMs
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts CONFIG_MAX_VCPUS.
Just remove this code altogether.
--- arch/arm/kvm/reset.c | 4 ---- 1 file changed, 4 deletions(-)
arch/arm/kvm/reset.c +0 -4
@@ -33,8 +33,6 @@* Cortex-A15 Reset Values */ -static const int a15_max_cpu_idx =3 ; -static struct kvm_regs a15_regs_reset = { .usr_regs.ARM_cpsr =SVC_MODE |PSR_A_BIT |PSR_I_BIT |PSR_F_BIT , };
@@ -63,8 +61,6 @@ int kvm_reset_vcpu(struct kvm_vcpu *vcpu)switch (vcpu->arch.target) {case KVM_ARM_TARGET_CORTEX_A15 : -if (vcpu->vcpu_id > a15_max_cpu_idx) -return -EINVAL ; reset_regs = &a15_regs_reset; vcpu->arch.midr =read_cpuid_id (); cpu_vtimer_irq = &a15_vtimer_irq;
Signature · 1 line
1.8.2.3
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 6 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster)
"Don't allow configurations that can't exist in hardware" makes sense as a restriction to me... QEMU will also enforce the 4-cpu limit.
-- PMM
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 11 lines · Peter Maydell
Quoted 6 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster)
"Don't allow configurations that can't exist in hardware" makes sense as a restriction to me... QEMU will also enforce the 4-cpu limit.
Nothing prevents anyone from building a multi-cluster A15 system:
http://www.lsi.com/products/mobile-communication-processors/pages/axxia-communication-processor-axm5500.aspx
If you want to limit QEMU to specific configurations, that's perfectly fine. The kernel doesn't need to, and kvmtool won't.
M.
Signature · 1 line
Fast, cheap, reliable. Pick two.
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
In reply to Marc ZyngierQuoted 7 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts
CONFIG_MAX_VCPUS.
So this comes from the early days where I looked at the A15 TRM and the MPIDR bit field for the CPU ID is limited to 2 bits. Exactly because I wasn't sure what remifications (if any) it would have to start populating this register with cluster id = (vcpu_id / 4) and cpu id = (vcpu_id % 4) I put this nice arbitrary restriction in there.
I think we need to fix how we show this register to the guest otherwise... No?
-Christoffer
Quoted 34 lines
Just remove this code altogether.
Signed-off-by Marc Zyngier <[email protected]>--- arch/arm/kvm/reset.c | 4 ---- 1 file changed, 4 deletions(-)arch/arm/kvm/reset.c +0
-4@@ -33,8 +33,6 @@* Cortex-A15 Reset Values */ -static const int a15_max_cpu_idx =3 ; -static struct kvm_regs a15_regs_reset = { .usr_regs.ARM_cpsr =SVC_MODE |PSR_A_BIT |PSR_I_BIT |PSR_F_BIT , };@@ -63,8 +61,6 @@ int kvm_reset_vcpu(struct kvm_vcpu *vcpu)switch (vcpu->arch.target) {case KVM_ARM_TARGET_CORTEX_A15 : -if (vcpu->vcpu_id > a15_max_cpu_idx) -return -EINVAL ; reset_regs = &a15_regs_reset; vcpu->arch.midr =read_cpuid_id (); cpu_vtimer_irq = &a15_vtimer_irq;Signature · 1 line
1.8.2.3
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 9 lines · Christoffer Dall
Quoted 7 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts
CONFIG_MAX_VCPUS.
Quoted 8 lines
So this comes from the early days where I looked at the A15 TRM and the MPIDR bit field for the CPU ID is limited to 2 bits. Exactly because I wasn't sure what remifications (if any) it would have to start populating this register with cluster id = (
vcpu_id/ 4) and cpu id = (vcpu_id% 4) I put this nice arbitrary restriction in there.I think we need to fix how we show this register to the guest otherwise... No?
I don't see this being an issue, but if we really want to be 100% true to the A15/A7 TRM, we can always compute MPIDR that way, and adjust L2CTLR as well.
That will require some userspace change in kvmtool though (need to change the DT generator to cope with the cluster ID).
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 16 lines · Marc Zyngier
Quoted 15 lines · Christoffer Dall
Quoted 7 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts
CONFIG_MAX_VCPUS.So this comes from the early days where I looked at the A15 TRM and the MPIDR bit field for the CPU ID is limited to 2 bits. Exactly because I wasn't sure what remifications (if any) it would have to start populating this register with cluster id = (
vcpu_id/ 4) and cpu id = (vcpu_id% 4) I put this nice arbitrary restriction in there.
Quoted 6 lines
Quoted 2 lines
I think we need to fix how we show this register to the guest otherwise... No?
I don't see this being an issue, but if we really want to be 100% true to the A15/A7 TRM, we can always compute MPIDR that way, and adjust L2CTLR as well.
Even with mach-virt we're still pretending to be an A15/A7 right? I think we should adhere to that. Is there some other reason why people shouldn't generally expect MPIDR to be correct (merging this code to KVM and running in a VM notwithstanding)?
Quoted 4 lines
That will require some userspace change in kvmtool though (need to change the DT generator to cope with the cluster ID).
That would be the less fun part...
-Christoffer
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 20 lines · Christoffer Dall
Quoted 19 lines · Marc Zyngier
Quoted 17 lines · Christoffer Dall
Quoted 7 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts
CONFIG_MAX_VCPUS.So this comes from the early days where I looked at the A15 TRM and the MPIDR bit field for the CPU ID is limited to 2 bits. Exactly because I wasn't sure what remifications (if any) it would have to start populating this register with cluster id = (
vcpu_id/ 4) and cpu id = (vcpu_id% 4) I put this nice arbitrary restriction in there.I think we need to fix how we show this register to the guest otherwise... No?
Quoted 8 lines
Quoted 3 lines
I don't see this being an issue, but if we really want to be 100% true to the A15/A7 TRM, we can always compute MPIDR that way, and adjust L2CTLR as well.
Even with mach-virt we're still pretending to be an A15/A7 right? I think we should adhere to that. Is there some other reason why people shouldn't generally expect MPIDR to be correct (merging this code to KVM and running in a VM notwithstanding)?
Yes, it is probably better to adhere to the law of least surprise. I'm reworking this patch and will repost (slightly longer) the series.
Quoted 5 lines
Quoted 4 lines
That will require some userspace change in kvmtool though (need to change the DT generator to cope with the cluster ID).
That would be the less fun part...
Actually, it is surprisingly easy once the kernel does the right thing.
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 1/1 ARM: KVM: drop arbitrary limitation to 4 CPU VMs
Quoted 34 lines · Marc Zyngier
Quoted 20 lines · Christoffer Dall
Quoted 19 lines · Marc Zyngier
Quoted 17 lines · Christoffer Dall
Quoted 7 lines · Marc Zyngier
It appears we have an arbitrary limitation where we refuse to create more than 4 virtual A15 in a single VM.
This limitation doesn't make much sense (the number 4 probably comes from the maximum number of CPUs in a A15 cluster, but KVM doesn't have any notion of cluster), and directly contradicts
CONFIG_MAX_VCPUS.So this comes from the early days where I looked at the A15 TRM and the MPIDR bit field for the CPU ID is limited to 2 bits. Exactly because I wasn't sure what remifications (if any) it would have to start populating this register with cluster id = (
vcpu_id/ 4) and cpu id = (vcpu_id% 4) I put this nice arbitrary restriction in there.I think we need to fix how we show this register to the guest otherwise... No?
Quoted 8 lines
Quoted 3 lines
I don't see this being an issue, but if we really want to be 100% true to the A15/A7 TRM, we can always compute MPIDR that way, and adjust L2CTLR as well.
Even with mach-virt we're still pretending to be an A15/A7 right? I think we should adhere to that. Is there some other reason why people shouldn't generally expect MPIDR to be correct (merging this code to KVM and running in a VM notwithstanding)?
Yes, it is probably better to adhere to the law of least surprise. I'm reworking this patch and will repost (slightly longer) the series.
Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 4 lines
Quoted 3 lines
That will require some userspace change in kvmtool though (need to change the DT generator to cope with the cluster ID).
That would be the less fun part...
Actually, it is surprisingly easy once the kernel does the right thing.
Great ;)