Gleb Natapov wrote on 2013-01-17: > On Thu, Jan 17, 2013 at 01:36:36AM +0000, Zhang, Yang Z wrote: >> Gleb Natapov wrote on 2013-01-16: >>> On Wed, Jan 16, 2013 at 05:00:48PM +0800, Yang Zhang wrote: >>>> From: Yang Zhang <[email protected]> >>>> >>>> The "acknowledge interrupt on exit" feature controls processor behavior >>>> for external interrupt acknowledgement. When this control is set, the >>>> processor acknowledges the interrupt controller to acquire the >>>> interrupt vector on VM exit. >>>> >>>> This feature is required by Posted Interrupt. It will be turnned on only >>>> when posted interrupt is enabled. >>>> >>> Why? Always enable it. It is faster than current approach. >> I don't think it will faster. It should be same. >> > Interrupt delivered->vmexit->interrupt enabled->interrupt > redelivered->acked->dispatched through idt. > > vs > > Interrupt delivered->acked->vmexit->dispatched through idt. > > Anyway even if it is not slower it is better to use the same code path > to make sure it is well tested. Ok.
>>>> Refer to Intel SDM volum 3, chapter 33.2. >>>> >>> Please CC H. Peter Anvin <[email protected]> on the next version. >>> >>>> Signed-off-by: Yang Zhang <[email protected]> >>>> --- >>>> arch/x86/kvm/vmx.c | 51 >>>> +++++++++++++++++++++++++++++++++++++++++++++++++-- 1 files changed, >>>> 49 insertions(+), 2 deletions(-) >>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c >>>> index dd2a85c..d1ed9ae 100644 >>>> --- a/arch/x86/kvm/vmx.c >>>> +++ b/arch/x86/kvm/vmx.c >>>> @@ -2565,7 +2565,7 @@ static __init int setup_vmcs_config(struct >>> vmcs_config *vmcs_conf) >>>> #ifdef CONFIG_X86_64 >>>> min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; >>>> #endif >>>> - opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT; >>>> + opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT | >>> VM_EXIT_ACK_INTR_ON_EXIT; >>>> if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS, >>>> &_vmexit_control) < 0) return -EIO; @@ >>>> -3926,7 +3926,7 @@ >>>> static int vmx_vcpu_setup(struct vcpu_vmx *vmx) ++vmx->nmsrs; > } >>>> - vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl); >>>> + vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl & >>>> ~VM_EXIT_ACK_INTR_ON_EXIT); >>>> >>>> /* 22.2.1, 20.8.1 */ >>>> vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl); >>>> @@ -6096,6 +6096,52 @@ static void vmx_complete_atomic_exit(struct >>> vcpu_vmx *vmx) >>>> } >>>> } >>>> + +static noinline void vmx_handle_external_intr(struct kvm_vcpu >>>> *vcpu) +{ + u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); + >>>> + if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == >>>> INTR_TYPE_EXT_INTR && + (exit_intr_info & INTR_INFO_VALID_MASK) ) >>>> { >>> if (exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK) >>> == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) >>> >>>> + unsigned int vector; >>>> + unsigned long entry; >>>> + struct desc_ptr dt; >>>> + gate_desc *desc; >>>> + >>>> + native_store_idt(&dt); >>> This does not change. Store it in vcpu during vcpu creation instead of >>> reading it each time. >> Right. It will reduce the cost. >> >>>> + >>>> + vector = exit_intr_info & INTR_INFO_VECTOR_MASK; >>>> + desc = (void *)dt.address + vector * 16; >>>> + >>>> + entry = gate_offset(*desc); >>>> + asm( >>>> + "mov %0, %%" _ASM_DX "\n\t" >>>> + "mov %%" _ASM_SP ", %%" _ASM_BX "\n\t" >>>> +#ifdef CONFIG_X86_64 >>>> + "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t" >>>> +#endif >>>> + "mov %%ss, %%" _ASM_AX "\n\t" >>>> + "push %%" _ASM_AX "\n\t" >>>> + "push %%" _ASM_BX "\n\t" >>> For 32bit you do not need to save ss:esp. >> You are right. No need to save it for 32bit. >> >>>> + "pushf\n\t" >>>> + "mov %%cs, %%" _ASM_AX "\n\t" >>>> + "push %%" _ASM_AX "\n\t" >>>> + "push intr_return\n\t" >>>> + "jmp *%% " _ASM_DX "\n\t" >>>> + ".pushsection .rodata \n\t" >>>> + ".global intr_return \n\t" >>>> + "intr_return: " _ASM_PTR " 1b \n\t" >>>> + ".popsection\n\t" >>>> + : :"m"(entry) : >>>> +#ifdef CONFIG_X86_64 >>>> + "rax", "rbx", "rdx" >>>> +#else >>>> + "eax", "ebx", "edx" >>>> +#endif >>>> + ); >>>> + } >>>> +} >>>> + >>>> static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx) >>>> { >>>> u32 exit_intr_info; >>>> @@ -6431,6 +6477,7 @@ static void __noclone vmx_vcpu_run(struct >>> kvm_vcpu *vcpu) >>>> vmx_complete_atomic_exit(vmx); >>>> vmx_recover_nmi_blocking(vmx); >>>> vmx_complete_interrupts(vmx); >>>> + vmx_handle_external_intr(vcpu); >>> This should be done just before enabling interrupts, otherwise we are >>> taking an interrupt before vmexit is fully complete. >> How about to put it just after set vcpu->mode to OUTSIDE_GUEST_MODE? At >> that point, the vmexit is considered as fully complete. >> > As I said, right before local_irq_enable(). Ok. Best regards, Yang -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
