On Thu, 30 Aug 2012 14:21:28 -0700, Christoffer Dall
<c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote:
> When the guest uses mmio operations that do not have decode information
> in the HSR, we decode these instructions manually, but the instruction
> may be used to access the vgic distributor interface emulated inside the
> kernel so we have to go through that path for both values of HSR.ISV.
>
> Signed-off-by: Christoffer Dall <c.dall@xxxxxxxxxxxxxxxxxxxxxx>
One nit, see below:
> @@ -758,34 +757,55 @@ static int io_mem_abort(struct kvm_vcpu *vcpu,
> struct kvm_run *run,
> return 1;
> }
>
> + mmio->is_write = is_write;
> + mmio->phys_addr = fault_ipa;
> + mmio->len = len;
> + vcpu->arch.mmio.sign_extend = sign_extend;
> + vcpu->arch.mmio.rd = rd;
> +
> + /*
> + * The MMIO instruction is emulated and should not be re-executed
> + * in the guest.
> + */
> + kvm_skip_instr(vcpu, (vcpu->arch.hsr >> 25) & 1);
> + return 0;
> +}
> +
> +static int io_mem_abort(struct kvm_vcpu *vcpu, struct kvm_run *run,
> + phys_addr_t fault_ipa, struct kvm_memory_slot *memslot)
> +{
> + struct kvm_exit_mmio mmio;
> + unsigned long rd;
> + int ret;
> +
> /*
> * Prepare MMIO operation. First stash it in a private
> * structure that we can use for in-kernel emulation. If the
> * kernel can't handle it, copy it into run->mmio and let user
> * space do its magic.
> */
> - mmio.is_write = is_write;
> - mmio.phys_addr = fault_ipa;
> - mmio.len = len;
> - vcpu->arch.mmio.sign_extend = sign_extend;
> - vcpu->arch.mmio.rd = rd;
>
> - trace_kvm_mmio((is_write) ? KVM_TRACE_MMIO_WRITE :
> - KVM_TRACE_MMIO_READ_UNSATISFIED,
> - len, fault_ipa, (is_write) ? *vcpu_reg(vcpu, rd) : 0);
> + if (vcpu->arch.hsr & HSR_ISV)
> + ret = decode_hsr(vcpu, fault_ipa, &mmio);
> + else
> + ret = invalid_io_mem_abort(vcpu, fault_ipa, &mmio);
>
> - if (is_write)
> - memcpy(mmio.data, vcpu_reg(vcpu, rd), len);
> + if (ret != 0)
> + return ret;
>
> - /*
> - * The MMIO instruction is emulated and should not be re-executed
> - * in the guest.
> - */
> - kvm_skip_instr(vcpu, (vcpu->arch.hsr >> 25) & 1);
> + rd = vcpu->arch.mmio.rd;
> + trace_kvm_mmio((mmio.is_write) ? KVM_TRACE_MMIO_WRITE :
> + KVM_TRACE_MMIO_READ_UNSATISFIED,
> + mmio.len, fault_ipa,
> + (mmio.is_write) ? *vcpu_reg(vcpu, rd) : 0);
> +
> + if (mmio.is_write)
> + memcpy(mmio.data, vcpu_reg(vcpu, rd), mmio.len);
>
> if (vgic_handle_mmio(vcpu, run, &mmio))
> return 1;
>
> + run->exit_reason = KVM_EXIT_MMIO;
> kvm_prepare_mmio(run, &mmio);
kvm_prepare_mmio() already set the exit_reason field.
M.
--
Who you jivin' with that Cosmik Debris?
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
