KVM虚拟机IO处理过程(二) ----QEMU/KVM I/O 处理过程

   接着KVM虚拟机IO处理过程中Guest Vm IO处理过程(http://blog.csdn.net/dashulu/article/details/16820281),本篇文章主要描述IO从guest vm跳转到kvm和qemu后的处理过程.

    首先回顾一下kvm的启动过程(http://blog.csdn.net/dashulu/article/details/17074675).qemu通过调用kvm提供的一系列接口来启动kvm. qemu的入口为vl.c中的main函数,main函数通过调用kvm_init 和 machine->init来初始化kvm. 其中, machine->init会创建vcpu, 用一个线程去模拟vcpu, 该线程执行的函数为qemu_kvm_cpu_thread_fn, 并且该线程最终kvm_cpu_exec,该函数调用kvm_vcpu_ioctl切换到kvm中,下次从kvm中返回时,会接着执行kvm_vcpu_ioctl之后的代码,判断exit_reason,然后进行相应处理.

int kvm_cpu_exec(CPUState *cpu)
{
    struct kvm_run *run = cpu->kvm_run;
    int ret, run_ret;

    DPRINTF("kvm_cpu_exec()
");

    if (kvm_arch_process_async_events(cpu)) {
        cpu->exit_request = 0;
        return EXCP_HLT;
    }

    do {
        if (cpu->kvm_vcpu_dirty) {
            kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
            cpu->kvm_vcpu_dirty = false;
        }

        kvm_arch_pre_run(cpu, run);
        if (cpu->exit_request) {
            DPRINTF("interrupt exit requested
");
            /*
             * KVM requires us to reenter the kernel after IO exits to complete
             * instruction emulation. This self-signal will ensure that we
             * leave ASAP again.
             */
            qemu_cpu_kick_self();
        }
        qemu_mutex_unlock_iothread();

        run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);

        qemu_mutex_lock_iothread();
        kvm_arch_post_run(cpu, run);

        if (run_ret < 0) {
            if (run_ret == -EINTR || run_ret == -EAGAIN) {
                DPRINTF("io window exit
");
                ret = EXCP_INTERRUPT;
                break;
            }
            fprintf(stderr, "error: kvm run failed %s
",
                    strerror(-run_ret));
            abort();
        }

        trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
        switch (run->exit_reason) {
        case KVM_EXIT_IO:
            DPRINTF("handle_io
");
            kvm_handle_io(run->io.port,
                          (uint8_t *)run + run->io.data_offset,
                          run->io.direction,
                          run->io.size,
                          run->io.count);
            ret = 0;
            break;
        case KVM_EXIT_MMIO:
            DPRINTF("handle_mmio
");
            cpu_physical_memory_rw(run->mmio.phys_addr,
                                   run->mmio.data,
                                   run->mmio.len,
                                   run->mmio.is_write);
            ret = 0;
            break;
        case KVM_EXIT_IRQ_WINDOW_OPEN:
            DPRINTF("irq_window_open
");
            ret = EXCP_INTERRUPT;
            break;
        case KVM_EXIT_SHUTDOWN:
            DPRINTF("shutdown
");
            qemu_system_reset_request();
            ret = EXCP_INTERRUPT;
            break;
        case KVM_EXIT_UNKNOWN:
            fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "
",
                    (uint64_t)run->hw.hardware_exit_reason);
            ret = -1;
            break;
        case KVM_EXIT_INTERNAL_ERROR:
            ret = kvm_handle_internal_error(cpu, run);
            break;
        default:
            DPRINTF("kvm_arch_handle_exit
");
            ret = kvm_arch_handle_exit(cpu, run);
            break;
        }
    } while (ret == 0);

    if (ret < 0) {
        cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
        vm_stop(RUN_STATE_INTERNAL_ERROR);
    }

    cpu->exit_request = 0;
    return ret;
}

kvm_vcpu_ioctl执行时,调用的kvm函数是virt/kvm/kvm-main.c中的kvm_vcpu_ioctl.c函数.当传入参数为KVM_RUN时,最终会调用到vcpu_enter_guest函数, vcpu_enter_guest函数中调用了kvm_x86_ops->run(vcpu),在intel处理器架构中该函数对应的实现为vmx_vcpu_run, vmx_vcpu_run设置好寄存器状态之后调用VM_LAUNCH或者VM_RESUME进入guest vm, 一旦发生vm exit则从此处继续执行下去.

static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
{
    struct vcpu_vmx *vmx = to_vmx(vcpu);
    unsigned long debugctlmsr;

    /*...此处省略n行代码...*/
    vmx->__launched = vmx->loaded_vmcs->launched;
    asm(
        /* Store host registers */
        "push %%" _ASM_DX "; push %%" _ASM_BP ";"
        "push %%" _ASM_CX " 
	" /* placeholder for guest rcx */
        "push %%" _ASM_CX " 
	"
        "cmp %%" _ASM_SP ", %c[host_rsp](%0) 
	"
        "je 1f 
	"
        "mov %%" _ASM_SP ", %c[host_rsp](%0) 
	"
        __ex(ASM_VMX_VMWRITE_RSP_RDX) "
	"
        "1: 
	"
        /* Reload cr2 if changed */
        "mov %c[cr2](%0), %%" _ASM_AX " 
	"
        "mov %%cr2, %%" _ASM_DX " 
	"
        "cmp %%" _ASM_AX ", %%" _ASM_DX " 
	"
        "je 2f 
	"
        "mov %%" _ASM_AX", %%cr2 
	"
        "2: 
	"
        /* Check if vmlaunch of vmresume is needed */
        "cmpl $0, %c[launched](%0) 
	"
        /* Load guest registers.  Don't clobber flags. */
        "mov %c[rax](%0), %%" _ASM_AX " 
	"
        "mov %c[rbx](%0), %%" _ASM_BX " 
	"
        "mov %c[rdx](%0), %%" _ASM_DX " 
	"
        "mov %c[rsi](%0), %%" _ASM_SI " 
	"
        "mov %c[rdi](%0), %%" _ASM_DI " 
	"
        "mov %c[rbp](%0), %%" _ASM_BP " 
	"
#ifdef CONFIG_X86_64
        "mov %c[r8](%0),  %%r8  
	"
        "mov %c[r9](%0),  %%r9  
	"
        "mov %c[r10](%0), %%r10 
	"
        "mov %c[r11](%0), %%r11 
	"
        "mov %c[r12](%0), %%r12 
	"
        "mov %c[r13](%0), %%r13 
	"
        "mov %c[r14](%0), %%r14 
	"
        "mov %c[r15](%0), %%r15 
	"
#endif
        "mov %c[rcx](%0), %%" _ASM_CX " 
	" /* kills %0 (ecx) */

        /* Enter guest mode */
        "jne 1f 
	"
        __ex(ASM_VMX_VMLAUNCH) "
	"
        "jmp 2f 
	"
        "1: " __ex(ASM_VMX_VMRESUME) "
	"
        "2: "
        /* Save guest registers, load host registers, keep flags */
        "mov %0, %c[wordsize](%%" _ASM_SP ") 
	"
        "pop %0 
	"
        "mov %%" _ASM_AX ", %c[rax](%0) 
	"
        "mov %%" _ASM_BX ", %c[rbx](%0) 
	"
        __ASM_SIZE(pop) " %c[rcx](%0) 
	"
        "mov %%" _ASM_DX ", %c[rdx](%0) 
	"
        "mov %%" _ASM_SI ", %c[rsi](%0) 
	"
        "mov %%" _ASM_DI ", %c[rdi](%0) 
	"
        "mov %%" _ASM_BP ", %c[rbp](%0) 
	"
#ifdef CONFIG_X86_64
        "mov %%r8,  %c[r8](%0) 
	"
        "mov %%r9,  %c[r9](%0) 
	"
        "mov %%r10, %c[r10](%0) 
	"
        "mov %%r11, %c[r11](%0) 
	"
        "mov %%r12, %c[r12](%0) 
	"
        "mov %%r13, %c[r13](%0) 
	"
        "mov %%r14, %c[r14](%0) 
	"
        "mov %%r15, %c[r15](%0) 
	"
#endif
        "mov %%cr2, %%" _ASM_AX "   
	"
        "mov %%" _ASM_AX ", %c[cr2](%0) 
	"

        "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " 
	"
        "setbe %c[fail](%0) 
	"
        ".pushsection .rodata 
	"
        ".global vmx_return 
	"
        "vmx_return: " _ASM_PTR " 2b 
	"
        ".popsection"
          : : "c"(vmx), "d"((unsigned long)HOST_RSP),
        [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
        [fail]"i"(offsetof(struct vcpu_vmx, fail)),
        [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
        [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
        [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
        [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
        [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
        [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
        [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
        [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
#ifdef CONFIG_X86_64
        [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
        [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
        [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
        [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
        [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
        [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
        [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
        [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
#endif
        [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
        [wordsize]"i"(sizeof(ulong))
          : "cc", "memory"
#ifdef CONFIG_X86_64
        , "rax", "rbx", "rdi", "rsi"
        , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
#else
        , "eax", "ebx", "edi", "esi"
#endif
          );

    /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
    if (debugctlmsr)
        update_debugctlmsr(debugctlmsr);

#ifndef CONFIG_X86_64
    /*
     * The sysexit path does not restore ds/es, so we must set them to
     * a reasonable value ourselves.
     *
     * We can't defer this to vmx_load_host_state() since that function
     * may be executed in interrupt context, which saves and restore segments
     * around it, nullifying its effect.
     */
    loadsegment(ds, __USER_DS);
    loadsegment(es, __USER_DS);
#endif

    vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
                  | (1 << VCPU_EXREG_RFLAGS)
                  | (1 << VCPU_EXREG_CPL)
                  | (1 << VCPU_EXREG_PDPTR)
                  | (1 << VCPU_EXREG_SEGMENTS)
                  | (1 << VCPU_EXREG_CR3));
    vcpu->arch.regs_dirty = 0;

    vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);

    if (is_guest_mode(vcpu)) {
        struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
        vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
        if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
            vmcs12->idt_vectoring_error_code =
                vmcs_read32(IDT_VECTORING_ERROR_CODE);
            vmcs12->vm_exit_instruction_len =
                vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
        }
    }

    vmx->loaded_vmcs->launched = 1;

    vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
    trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);

    vmx_complete_atomic_exit(vmx);
    vmx_recover_nmi_blocking(vmx);
    vmx_complete_interrupts(vmx);
}

    介绍完初始化的流程,可以介绍IO在kvm和qemu中的处理流程了. 当Guest Vm进行IO操作需要访问设备时,就会触发vm exit 返回到vmx_vcpu_run, vmx保存好vmcs并且记录下VM_ExIT_REASON后返回到调用该函数的vcpu_enter_guest, 在vcpu_enter_guest函数末尾调用了r = kvm_x86_ops->handle_exit(vcpu), 该函数对应于vmx_handle_exit函数(intel cpu架构对应关系可以查看vmx.c文件中static struct kvm_x86_ops vmx_x86_ops), vmx_handle_exit 调用kvm_vmx_exit_handlers[exit_reason](vcpu),该语句根据exit_reason调用不同的函数,该数据结构定义如下:

static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
    [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
    [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
    [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
    [EXIT_REASON_NMI_WINDOW]          = handle_nmi_window,
    [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
    [EXIT_REASON_CR_ACCESS]               = handle_cr,
    [EXIT_REASON_DR_ACCESS]               = handle_dr,
    [EXIT_REASON_CPUID]                   = handle_cpuid,
    [EXIT_REASON_MSR_READ]                = handle_rdmsr,
    [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
    [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
    [EXIT_REASON_HLT]                     = handle_halt,
    [EXIT_REASON_INVD]            = handle_invd,
    [EXIT_REASON_INVLPG]              = handle_invlpg,
    [EXIT_REASON_RDPMC]                   = handle_rdpmc,
    [EXIT_REASON_VMCALL]                  = handle_vmcall,
    [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
    [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
    [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
    [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
    [EXIT_REASON_VMREAD]                  = handle_vmread,
    [EXIT_REASON_VMRESUME]                = handle_vmresume,
    [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
    [EXIT_REASON_VMOFF]                   = handle_vmoff,
    [EXIT_REASON_VMON]                    = handle_vmon,
    [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
    [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
    [EXIT_REASON_WBINVD]                  = handle_wbinvd,
    [EXIT_REASON_XSETBV]                  = handle_xsetbv,
    [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
    [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
    [EXIT_REASON_EPT_VIOLATION]       = handle_ept_violation,
    [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
    [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
    [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
    [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
};

如果是因为IO原因导致的vm exit,则调用的处理函数为handle_io,handle_io的处理可以查看(http://blog.csdn.net/fanwenyi/article/details/12748613), 该过程结束之后需要qemu去处理IO,这时候会返回到qemu, 在kvm_cpu_exec中继续执行下去,看上面kvm_cpu_exec的代码,如果是因为IO原因返回到qemu,会调用kvm_handle_io函数.

switch (run->exit_reason) {
        case KVM_EXIT_IO:
            DPRINTF("handle_io
");
            kvm_handle_io(run->io.port,
                          (uint8_t *)run + run->io.data_offset,
                          run->io.direction,
                          run->io.size,
                          run->io.count);
            ret = 0;
            break;

kvm_handle_io调用cpu_outb, cpu_outw等指令处理IO操作.

    假设虚拟机是用raw格式的磁盘,则IO在qemu中处理时经过的函数栈如下所示:

#0 bdrv_aio_writev (bs=0x55555629e9b0, sector_num=870456,
qiov=0x555556715ab0, nb_sectors=1,
cb=0x55555570161b <ide_sector_write_cb>, opaque=0x5555567157b8)
at block.c:3408
#1 0x0000555555701960 in ide_sector_write (s=0x5555567157b8)
at hw/ide/core.c:798
#2 0x00005555557047ae in ide_data_writew (opaque=0x555556715740, addr=496,
val=8995) at hw/ide/core.c:1907
#3 0x00005555558d9e4c in portio_write (opaque=0x5555565c0670, addr=0,
data=8995, size=2) at /home/dashu/kvm/qemu/qemu-dev-zwu/ioport.c:174
#4 0x00005555558e13d5 in memory_region_write_accessor (mr=0x5555565c0670,
addr=0, value=0x7fffb4dbd528, size=2, shift=0, mask=65535)
at /home/dashu/kvm/qemu/qemu-dev-zwu/memory.c:440
#5 0x00005555558e151d in access_with_adjusted_size (addr=0,
value=0x7fffb4dbd528, size=2, access_size_min=1, access_size_max=4,
access=0x5555558e1341 <memory_region_write_accessor>, mr=0x5555565c0670)
at /home/dashu/kvm/qemu/qemu-dev-zwu/memory.c:477
#6 0x00005555558e3dfb in memory_region_dispatch_write (mr=0x5555565c0670,
addr=0, data=8995, size=2)
at /home/dashu/kvm/qemu/qemu-dev-zwu/memory.c:984
#7 0x00005555558e7384 in io_mem_write (mr=0x5555565c0670, addr=0, val=8995,
size=2) at /home/dashu/kvm/qemu/qemu-dev-zwu/memory.c:1748
#8 0x000055555586a18e in address_space_rw (as=0x555556216d80, addr=496,
buf=0x7fffb4dbd670 "##", len=2, is_write=true)
at /home/dashu/kvm/qemu/qemu-dev-zwu/exec.c:1968
#9 0x000055555586a474 in address_space_write (as=0x555556216d80, addr=496,
buf=0x7fffb4dbd670 "##", len=2)
at /home/dashu/kvm/qemu/qemu-dev-zwu/exec.c:2030
#10 0x00005555558d98c9 in cpu_outw (addr=496, val=8995)
at /home/dashu/kvm/qemu/qemu-dev-zwu/ioport.c:61

bdrv_aio_writev最终调用bdrv_co_aio_rw_vector函数, 该函数调用co = qemu_coroutine_create(bdrv_co_do_rw) 创建一个协程去执行bdrv_co_do_rw函数,bdrv_co_wo_rw函数的函数栈如下:

#1 0x000055555563653c in paio_submit (bs=0x5555562a13d0, fd=10, sector_num=2,
qiov=0x555556715ab0, nb_sectors=1,
cb=0x5555556028b1 <bdrv_co_io_em_complete>, opaque=0x555556964e30, type=1)
at block/raw-posix.c:825
#2 0x0000555555636659 in raw_aio_submit (bs=0x5555562a13d0, sector_num=2,
qiov=0x555556715ab0, nb_sectors=1,
cb=0x5555556028b1 <bdrv_co_io_em_complete>, opaque=0x555556964e30, type=1)
at block/raw-posix.c:853
#3 0x00005555556366c9 in raw_aio_readv (bs=0x5555562a13d0, sector_num=2,
qiov=0x555556715ab0, nb_sectors=1,
cb=0x5555556028b1 <bdrv_co_io_em_complete>, opaque=0x555556964e30)
at block/raw-posix.c:861
#4 0x00005555556029b8 in bdrv_co_io_em (bs=0x5555562a13d0, sector_num=2,
nb_sectors=1, iov=0x555556715ab0, is_write=false) at block.c:4038
#5 0x0000555555602a49 in bdrv_co_readv_em (bs=0x5555562a13d0, sector_num=2,
nb_sectors=1, iov=0x555556715ab0) at block.c:4055
#6 0x00005555555fed61 in bdrv_co_do_readv (bs=0x5555562a13d0, sector_num=2,
nb_sectors=1, qiov=0x555556715ab0, flags=0) at block.c:2547
#7 0x00005555555fee03 in bdrv_co_readv (bs=0x5555562a13d0, sector_num=2,
nb_sectors=1, qiov=0x555556715ab0) at block.c:2573
#8 0x0000555555637d8c in raw_co_readv (bs=0x55555629e9b0, sector_num=2,
nb_sectors=1, qiov=0x555556715ab0) at block/raw.c:47
#9 0x00005555555fed61 in bdrv_co_do_readv (bs=0x55555629e9b0, sector_num=2,
nb_sectors=1, qiov=0x555556715ab0, flags=0) at block.c:2547
#10 0x00005555556023af in bdrv_co_do_rw

最终在paio_summit中会往线程池中提交一个请求thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque), 由调度器去执行aio_worker函数,aio_worker是真正做IO操作的函数,它通过pwrite和pread去读取磁盘.

    当qemu完成IO操作后,会在kvm_cpu_exec函数的循环中,调用kvm_vcpu_ioctl重新进入kvm.

    以上阐述了IO操作在kvm和qemu中处理的整个过程.

参考资料:

1. kvm代码解析连载(二):io的虚拟化:http://blog.csdn.net/fanwenyi/article/details/12748613