kvm feautres

CPUID 40000000:00 = 40000001 4b4d564b 564b4d56 0000004d | ...@KVMKVMKVM...
CPUID 40000001:00 = 01007afb 00000000 00000000 00000000 | .z..............

在 tools/testing/selftests/kvm/include/x86_64/processor.h 可以找到:

/* This CPUID returns two feature bitmaps in eax, edx. Before enabling
 * a particular paravirtualization, the appropriate feature bit should
 * be checked in eax. The performance hint feature bit should be checked
 * in edx.
 */
#define KVM_CPUID_FEATURES	0x40000001
#define KVM_FEATURE_CLOCKSOURCE		0
#define KVM_FEATURE_NOP_IO_DELAY	1
#define KVM_FEATURE_MMU_OP		2
/* This indicates that the new set of kvmclock msrs
 * are available. The use of 0x11 and 0x12 is deprecated
 */
#define KVM_FEATURE_CLOCKSOURCE2        3
#define KVM_FEATURE_ASYNC_PF		4
#define KVM_FEATURE_STEAL_TIME		5
#define KVM_FEATURE_PV_EOI		6
#define KVM_FEATURE_PV_UNHALT		7
#define KVM_FEATURE_PV_TLB_FLUSH	9
#define KVM_FEATURE_ASYNC_PF_VMEXIT	10
#define KVM_FEATURE_PV_SEND_IPI	11
#define KVM_FEATURE_POLL_CONTROL	12
#define KVM_FEATURE_PV_SCHED_YIELD	13
#define KVM_FEATURE_ASYNC_PF_INT	14
#define KVM_FEATURE_MSI_EXT_DEST_ID	15
#define KVM_FEATURE_HC_MAP_GPA_RANGE	16
#define KVM_FEATURE_MIGRATION_CONTROL	17

在虚拟机中执行 cpuid

   hypervisor features (0x40000001/eax):
      kvmclock available at MSR 0x11           = true
      delays unnecessary for PIO ops           = true
      mmu_op                                   = false
      kvmclock available at MSR 0x4b564d00     = true
      async pf enable available by MSR         = true
      steal clock supported                    = true
      guest EOI optimization enabled           = true
      guest spinlock optimization enabled      = true
      guest TLB flush optimization enabled     = true
      async PF VM exit enable available by MSR = true
      guest send IPI optimization enabled      = true
      host HLT poll disable at MSR 0x4b564d05  = true
      guest sched yield optimization enabled   = true
      guest uses intrs for page ready APF evs  = true
      extended destination ID                  = false
      map gpa range hypercall supported        = false
      MSR_KVM_MIGRATION_CONTROL supported      = false
      stable: no guest per-cpu warps expected  = true
   hypervisor features (0x40000001/edx):
      realtime hint: no unbound preemption = false
   hypervisor_id (0x40000100) = "\0\0\0\0\0\0\0\0\0\0\0\0"

qemu 的支持

在 qemu 的定义:

target/i386/cpu.c

    [FEAT_KVM] = {
        .type = CPUID_FEATURE_WORD,
        .feat_names = {
            "kvmclock", "kvm-nopiodelay", "kvm-mmu", "kvmclock",
            "kvm-asyncpf", "kvm-steal-time", "kvm-pv-eoi", "kvm-pv-unhalt",
            NULL, "kvm-pv-tlb-flush", "kvm-asyncpf-vmexit", "kvm-pv-ipi",
            "kvm-poll-control", "kvm-pv-sched-yield", "kvm-asyncpf-int", "kvm-msi-ext-dest-id",
            NULL, NULL, NULL, NULL,
            NULL, NULL, NULL, NULL,
            "kvmclock-stable-bit", NULL, NULL, NULL,
            NULL, NULL, NULL, NULL,
        },
        .cpuid = { .eax = KVM_CPUID_FEATURES, .reg = R_EAX, },
        .tcg_features = TCG_KVM_FEATURES,
    },

kvm_emulate_hypercall -> __kvm_emulate_hypercall

KVM_HC_KICK_CPU

好几个 kvm_sched_yield 是看

• directed_yield_attempted
• directed_yield_successful

每一个都是需要仔细看看的:

	case KVM_HC_KICK_CPU:
		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
			break;

		kvm_pv_kick_cpu_op(vcpu->kvm, a1);
		kvm_sched_yield(vcpu, a1);
		ret = 0;
		break;
#ifdef CONFIG_X86_64
	case KVM_HC_CLOCK_PAIRING:
		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
		break;
#endif
	case KVM_HC_SEND_IPI:
		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
			break;

		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
		break;
	case KVM_HC_SCHED_YIELD:
		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
			break;

		kvm_sched_yield(vcpu, a0);
		ret = 0;
		break;

vcpu stall detector

内核中存在 : drivers/misc/vcpu_stall_detector.c

对应的公开实现在 crosvm：devices/src/vmwdt.rs (https://chromium.googlesource.com/crosvm/crosvm/+/refs/heads/main/devices/src/vmwdt.rs)。

• impl BusDevice for Vmwdt → write()：接收 guest 的 MMIO 写操作，处理启停、频率配置和喂狗。寄存器偏移 0x00/0x04/0x08/0x0c、每个 vCPU 占 0x10 字节，都与 Linux 驱动一致。

• vmwdt_worker_thread()：等待各 vCPU 的定时器，判断是否超时；首次超时触发 guest 中断，后续仍未恢复则发送 WatchdogReset。

• get_guest_time_ms()：读取 /proc//task//stat 的 guest_time，获取 vCPU 实际执行 guest 的时间。

设备创建和 MMIO/IRQ 注册在 aarch64/src/lib.rs (https://android.googlesource.com/platform/external/crosvm/+/refs/heads/master/aarch64/src/lib.rs) 的 add_arch_devs()，搜索 Vmwdt::new 即可。

本站所有文章转发 CSDN 将按侵权追究法律责任，其它情况随意。