Hello!
> > + /*
> > + * When KVM is in use, psci is emulated in-kernel and not by qemu.
> > + * In order for it to work correctly we should use correct MPIDR
> > values,
> > + * which appear to be inherited from the host.
> why it must be inherited from host?
> Could you point out to KVM code that depends on it?
It is PSCI, i have tested this and experienced this problem. In KVM guest PSCI
is handled
by "hvc" call and is done entirely in kernel, no qemu code is involved. If i
supply wrong
IDs the processors will simply not power up.
I have checked how kvmtool (found here: git://linux-arm.org/linux-ap.git,
tools/kvm)
handles this. Related parts are:
tools/kvm/arm/fdt.c - generate_cpu_nodes() - "reg" property assignment:
--- cut ---
static void generate_cpu_nodes(void *fdt, struct kvm *kvm)
{
int cpu;
_FDT(fdt_begin_node(fdt, "cpus"));
_FDT(fdt_property_cell(fdt, "#address-cells", 0x1));
_FDT(fdt_property_cell(fdt, "#size-cells", 0x0));
for (cpu = 0; cpu < kvm->nrcpus; ++cpu) {
char cpu_name[CPU_NAME_MAX_LEN];
struct kvm_cpu *vcpu = kvm->cpus[cpu];
unsigned long mpidr = kvm_cpu__get_vcpu_mpidr(vcpu);
mpidr &= ARM_MPIDR_HWID_BITMASK;
snprintf(cpu_name, CPU_NAME_MAX_LEN, "cpu@%lx", mpidr);
_FDT(fdt_begin_node(fdt, cpu_name));
_FDT(fdt_property_string(fdt, "device_type", "cpu"));
_FDT(fdt_property_string(fdt, "compatible",
vcpu->cpu_compatible));
if (kvm->nrcpus > 1)
_FDT(fdt_property_string(fdt, "enable-method", "psci"));
_FDT(fdt_property_cell(fdt, "reg", mpidr));
_FDT(fdt_end_node(fdt));
}
_FDT(fdt_end_node(fdt));
}
--- cut ---
tools/kvm/arm/aarch64/kvm-cpu.c - kvm_cpu__get_vcpu_mpidr() - obtains CPU ID
from the
kernel:
--- cut ---
unsigned long kvm_cpu__get_vcpu_mpidr(struct kvm_cpu *vcpu)
{
struct kvm_one_reg reg;
u64 mpidr;
reg.id = ARM64_SYS_REG(ARM_CPU_ID, ARM_CPU_ID_MPIDR);
reg.addr = (u64)&mpidr;
if (ioctl(vcpu->vcpu_fd, KVM_GET_ONE_REG, ®) < 0)
die("KVM_GET_ONE_REG failed (get_mpidr vcpu%ld", vcpu->cpu_id);
return mpidr;
}
--- cut ---
So i just decided to do the same thing in KVM. However, to tell the truth, i
actually do
not know whether it is possible to do it in reverse and assign MPIDR to VCPUs
using
KVM_SET_ONE_REG ioctl instead. I can try it if you think it's more appropriate.
Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia
