QEMU源码全解析 —— CPU虚拟化（15）

于 2024-07-30 19:11:00 发布
阅读量448 收藏
点赞数 12
分类专栏： QEMU/KVM 文章标签： QEMU KVM CPU虚拟化
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接： https://blog.csdn.net/phmatthaus/article/details/140794594
版权
QEMU/KVM 专栏收录该内容
123 篇文章 36 订阅 ¥49.90 ¥99.00
已订阅 8折续费
您已是超级会员，正在免费阅读会员专享内容
查看更多超级会员权益

接前一篇文章：

本文内容参考：

《趣谈 Linux操作系统 》 —— 刘超， 极客时间

《 QEMU /KVM》 源码 解析与应用 —— 李强，机械工业出版社

《深度探索 Linux 系统 虚拟化 原理与实现》—— 王柏生 谢广军， 机械工业出版社

特此致谢！

三、KVM模块初始化介绍

2. KVM模块初始化

上一回讲到为了更好、更方便地理解KVM模块的初始化部分（先有一、后有二），笔者找了一个较早版本的Linux内核linux-6.1.10。再来看一下其arch/86/kvm/vmx/vmx.c中的vmx_init函数，代码如下： 

  1. static int __init vmx_init(void)
  2. {
  3. 	int r, cpu;
  4.  
  5. #if IS_ENABLED(CONFIG_HYPERV)
  6. 	/*
  7. 	 * Enlightened VMCS usage should be recommended and the host needs
  8. 	 * to support eVMCS v1 or above. We can also disable eVMCS support
  9. 	 * with module parameter.
  10. 	 */
  11. 	if (enlightened_vmcs &&
  12. 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
  13. 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
  14. 	    KVM_EVMCS_VERSION) {
  15.  
  16. 		/* Check that we have assist pages on all online CPUs */
  17. 		for_each_online_cpu(cpu) {
  18. 			if (!hv_get_vp_assist_page(cpu)) {
  19. 				enlightened_vmcs = false;
  20. 				break;
  21. 			}
  22. 		}
  23.  
  24. 		if (enlightened_vmcs) {
  25. 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
  26. 			static_branch_enable(&enable_evmcs);
  27. 		}
  28.  
  29. 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
  30. 			vmx_x86_ops.enable_direct_tlbflush
  31. 				= hv_enable_direct_tlbflush;
  32.  
  33. 	} else {
  34. 		enlightened_vmcs = false;
  35. 	}
  36. #endif
  37.  
  38. 	r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
  39. 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
  40. 	if (r)
  41. 		return r;
  42.  
  43. 	/*
  44. 	 * Must be called after kvm_init() so enable_ept is properly set
  45. 	 * up. Hand the parameter mitigation value in which was stored in
  46. 	 * the pre module init parser. If no parameter was given, it will
  47. 	 * contain 'auto' which will be turned into the default 'cond'
  48. 	 * mitigation mode.
  49. 	 */
  50. 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
  51. 	if (r) {
  52. 		vmx_exit();
  53. 		return r;
  54. 	}
  55.  
  56. 	vmx_setup_fb_clear_ctrl();
  57.  
  58. 	for_each_possible_cpu(cpu) {
  59. 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
  60.  
  61. 		pi_init_cpu(cpu);
  62. 	}
  63.  
  64. #ifdef CONFIG_KEXEC_CORE
  65. 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
  66. 			   crash_vmclear_local_loaded_vmcss);
  67. #endif
  68. 	vmx_check_vmcs12_offsets();
  69.  
  70. 	/*
  71. 	 * Shadow paging doesn't have a (further) performance penalty
  72. 	 * from GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable it
  73. 	 * by default
  74. 	 */
  75. 	if (!enable_ept)
  76. 		allow_smaller_maxphyaddr = true;
  77.  
  78. 	return 0;
  79. }
  80. module_init(vmx_init);

vmx_init_ops的定义和初始化在同文件（Linux内核/arch/x86/kvm/vmx/vmx.c）中，代码如下： 

  1. static struct kvm_x86_init_ops vmx_init_ops __initdata = {
  2. 	.cpu_has_kvm_support = cpu_has_kvm_support,
  3. 	.disabled_by_bios = vmx_disabled_by_bios,
  4. 	.check_processor_compatibility = vmx_check_processor_compat,
  5. 	.hardware_setup = hardware_setup,
  6. 	.handle_intel_pt_intr = NULL,
  7.  
  8. 	.runtime_ops = &vmx_x86_ops,
  9. 	.pmu_ops = &intel_pmu_ops,
  10. };

前边反复提到，KVM的所有虚拟化实现（Intel和AMD）都会向KVM模块注册一个kvm_x86_ops（这个内核版本中已经改为了struct kvm_x86_init_ops）结构体（实例）。这样，KVM中的一些函数就是一个外壳，首先会调用kvm_arch_xxx函数，表示的是调用CPU架构相关的回调函数。如果kvm_arch_xxx函数需要调用到实现相关的代码，则会调用kvm_x86_ops结构中的相关回调函数。

VMX_init()中调用kvm_init函数的代码片段如下： 

  1. 	r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
  2. 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
  3. 	if (r)
  4. 		return r;

其实这个版本的Linux内核代码也比较新了，也不是早先第一个参数直接传递&vmx_x86_ops的方式，而是又包了一层，将vmx_x86_ops封到了struct kvm_x86_init_ops vmx_init_ops中（参见上边代码）。

vmx_x86_ops的定义和初始化在同文件（arch/x86/kvm/vmx/vmx.c）中，如下： 

  1. static struct kvm_x86_ops vmx_x86_ops __initdata = {
  2. 	.name = "kvm_intel",
  3.  
  4. 	.hardware_unsetup = vmx_hardware_unsetup,
  5.  
  6. 	.hardware_enable = vmx_hardware_enable,
  7. 	.hardware_disable = vmx_hardware_disable,
  8. 	.has_emulated_msr = vmx_has_emulated_msr,
  9.  
  10. 	.vm_size = sizeof(struct kvm_vmx),
  11. 	.vm_init = vmx_vm_init,
  12. 	.vm_destroy = vmx_vm_destroy,
  13.  
  14. 	.vcpu_precreate = vmx_vcpu_precreate,
  15. 	.vcpu_create = vmx_vcpu_create,
  16. 	.vcpu_free = vmx_vcpu_free,
  17. 	.vcpu_reset = vmx_vcpu_reset,
  18.  
  19. 	.prepare_switch_to_guest = vmx_prepare_switch_to_guest,
  20. 	.vcpu_load = vmx_vcpu_load,
  21. 	.vcpu_put = vmx_vcpu_put,
  22.  
  23. 	.update_exception_bitmap = vmx_update_exception_bitmap,
  24. 	.get_msr_feature = vmx_get_msr_feature,
  25. 	.get_msr = vmx_get_msr,
  26. 	.set_msr = vmx_set_msr,
  27. 	.get_segment_base = vmx_get_segment_base,
  28. 	.get_segment = vmx_get_segment,
  29. 	.set_segment = vmx_set_segment,
  30. 	.get_cpl = vmx_get_cpl,
  31. 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
  32. 	.set_cr0 = vmx_set_cr0,
  33. 	.is_valid_cr4 = vmx_is_valid_cr4,
  34. 	.set_cr4 = vmx_set_cr4,
  35. 	.set_efer = vmx_set_efer,
  36. 	.get_idt = vmx_get_idt,
  37. 	.set_idt = vmx_set_idt,
  38. 	.get_gdt = vmx_get_gdt,
  39. 	.set_gdt = vmx_set_gdt,
  40. 	.set_dr7 = vmx_set_dr7,
  41. 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
  42. 	.cache_reg = vmx_cache_reg,
  43. 	.get_rflags = vmx_get_rflags,
  44. 	.set_rflags = vmx_set_rflags,
  45. 	.get_if_flag = vmx_get_if_flag,
  46.  
  47. 	.flush_tlb_all = vmx_flush_tlb_all,
  48. 	.flush_tlb_current = vmx_flush_tlb_current,
  49. 	.flush_tlb_gva = vmx_flush_tlb_gva,
  50. 	.flush_tlb_guest = vmx_flush_tlb_guest,
  51.  
  52. 	.vcpu_pre_run = vmx_vcpu_pre_run,
  53. 	.vcpu_run = vmx_vcpu_run,
  54. 	.handle_exit = vmx_handle_exit,
  55. 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
  56. 	.update_emulated_instruction = vmx_update_emulated_instruction,
  57. 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
  58. 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
  59. 	.patch_hypercall = vmx_patch_hypercall,
  60. 	.inject_irq = vmx_inject_irq,
  61. 	.inject_nmi = vmx_inject_nmi,
  62. 	.inject_exception = vmx_inject_exception,
  63. 	.cancel_injection = vmx_cancel_injection,
  64. 	.interrupt_allowed = vmx_interrupt_allowed,
  65. 	.nmi_allowed = vmx_nmi_allowed,
  66. 	.get_nmi_mask = vmx_get_nmi_mask,
  67. 	.set_nmi_mask = vmx_set_nmi_mask,
  68. 	.enable_nmi_window = vmx_enable_nmi_window,
  69. 	.enable_irq_window = vmx_enable_irq_window,
  70. 	.update_cr8_intercept = vmx_update_cr8_intercept,
  71. 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
  72. 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
  73. 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
  74. 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
  75. 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
  76. 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
  77. 	.hwapic_irr_update = vmx_hwapic_irr_update,
  78. 	.hwapic_isr_update = vmx_hwapic_isr_update,
  79. 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
  80. 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
  81. 	.deliver_interrupt = vmx_deliver_interrupt,
  82. 	.dy_apicv_has_pending_interrupt = pi_has_pending_interrupt,
  83.  
  84. 	.set_tss_addr = vmx_set_tss_addr,
  85. 	.set_identity_map_addr = vmx_set_identity_map_addr,
  86. 	.get_mt_mask = vmx_get_mt_mask,
  87.  
  88. 	.get_exit_info = vmx_get_exit_info,
  89.  
  90. 	.vcpu_after_set_cpuid = vmx_vcpu_after_set_cpuid,
  91.  
  92. 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
  93.  
  94. 	.get_l2_tsc_offset = vmx_get_l2_tsc_offset,
  95. 	.get_l2_tsc_multiplier = vmx_get_l2_tsc_multiplier,
  96. 	.write_tsc_offset = vmx_write_tsc_offset,
  97. 	.write_tsc_multiplier = vmx_write_tsc_multiplier,
  98.  
  99. 	.load_mmu_pgd = vmx_load_mmu_pgd,
  100.  
  101. 	.check_intercept = vmx_check_intercept,
  102. 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
  103.  
  104. 	.request_immediate_exit = vmx_request_immediate_exit,
  105.  
  106. 	.sched_in = vmx_sched_in,
  107.  
  108. 	.cpu_dirty_log_size = PML_ENTITY_NUM,
  109. 	.update_cpu_dirty_logging = vmx_update_cpu_dirty_logging,
  110.  
  111. 	.nested_ops = &vmx_nested_ops,
  112.  
  113. 	.pi_update_irte = vmx_pi_update_irte,
  114. 	.pi_start_assignment = vmx_pi_start_assignment,
  115.  
  116. #ifdef CONFIG_X86_64
  117. 	.set_hv_timer = vmx_set_hv_timer,
  118. 	.cancel_hv_timer = vmx_cancel_hv_timer,
  119. #endif
  120.  
  121. 	.setup_mce = vmx_setup_mce,
  122.  
  123. 	.smi_allowed = vmx_smi_allowed,
  124. 	.enter_smm = vmx_enter_smm,
  125. 	.leave_smm = vmx_leave_smm,
  126. 	.enable_smi_window = vmx_enable_smi_window,
  127.  
  128. 	.can_emulate_instruction = vmx_can_emulate_instruction,
  129. 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
  130. 	.migrate_timers = vmx_migrate_timers,
  131.  
  132. 	.msr_filter_changed = vmx_msr_filter_changed,
  133. 	.complete_emulated_msr = kvm_complete_insn_gp,
  134.  
  135. 	.vcpu_deliver_sipi_vector = kvm_vcpu_deliver_sipi_vector,
  136. };

struct kvm_x86_ops vmx_x86_ops是一个非常大的结构，包括具体硬件检测、 虚拟机 创建VCPU的实现、一些寄存器的设置、虚拟机退出的处理函数等。

kvm_init函数的第2个参数为sizeof(struct vcpu_vmx)，表示VMX实现的VCPU结构体的大小。struct vcpu_vmx的定义在Linux内核源码/arch/x86/kvm/vmx/vmx.h中，如下： 

  1. struct vcpu_vmx {
  2. 	struct kvm_vcpu       vcpu;
  3. 	u8                    fail;
  4. 	u8		      x2apic_msr_bitmap_mode;
  5.  
  6. 	/*
  7. 	 * If true, host state has been stored in vmx->loaded_vmcs for
  8. 	 * the CPU registers that only need to be switched when transitioning
  9. 	 * to/from the kernel, and the registers have been loaded with guest
  10. 	 * values.  If false, host state is loaded in the CPU registers
  11. 	 * and vmx->loaded_vmcs->host_state is invalid.
  12. 	 */
  13. 	bool		      guest_state_loaded;
  14.  
  15. 	unsigned long         exit_qualification;
  16. 	u32                   exit_intr_info;
  17. 	u32                   idt_vectoring_info;
  18. 	ulong                 rflags;
  19.  
  20. 	/*
  21. 	 * User return MSRs are always emulated when enabled in the guest, but
  22. 	 * only loaded into hardware when necessary, e.g. SYSCALL #UDs outside
  23. 	 * of 64-bit mode or if EFER.SCE=1, thus the SYSCALL MSRs don't need to
  24. 	 * be loaded into hardware if those conditions aren't met.
  25. 	 */
  26. 	struct vmx_uret_msr   guest_uret_msrs[MAX_NR_USER_RETURN_MSRS];
  27. 	bool                  guest_uret_msrs_loaded;
  28. #ifdef CONFIG_X86_64
  29. 	u64		      msr_host_kernel_gs_base;
  30. 	u64		      msr_guest_kernel_gs_base;
  31. #endif
  32.  
  33. 	u64		      spec_ctrl;
  34. 	u32		      msr_ia32_umwait_control;
  35.  
  36. 	/*
  37. 	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
  38. 	 * non-nested (L1) guest, it always points to vmcs01. For a nested
  39. 	 * guest (L2), it points to a different VMCS.
  40. 	 */
  41. 	struct loaded_vmcs    vmcs01;
  42. 	struct loaded_vmcs   *loaded_vmcs;
  43.  
  44. 	struct msr_autoload {
  45. 		struct vmx_msrs guest;
  46. 		struct vmx_msrs host;
  47. 	} msr_autoload;
  48.  
  49. 	struct msr_autostore {
  50. 		struct vmx_msrs guest;
  51. 	} msr_autostore;
  52.  
  53. 	struct {
  54. 		int vm86_active;
  55. 		ulong save_rflags;
  56. 		struct kvm_segment segs[8];
  57. 	} rmode;
  58. 	struct {
  59. 		u32 bitmask; /* 4 bits per segment (1 bit per field) */
  60. 		struct kvm_save_segment {
  61. 			u16 selector;
  62. 			unsigned long base;
  63. 			u32 limit;
  64. 			u32 ar;
  65. 		} seg[8];
  66. 	} segment_cache;
  67. 	int vpid;
  68. 	bool emulation_required;
  69.  
  70. 	union vmx_exit_reason exit_reason;
  71.  
  72. 	/* Posted interrupt descriptor */
  73. 	struct pi_desc pi_desc;
  74.  
  75. 	/* Used if this vCPU is waiting for PI notification wakeup. */
  76. 	struct list_head pi_wakeup_list;
  77.  
  78. 	/* Support for a guest hypervisor (nested VMX) */
  79. 	struct nested_vmx nested;
  80.  
  81. 	/* Dynamic PLE window. */
  82. 	unsigned int ple_window;
  83. 	bool ple_window_dirty;
  84.  
  85. 	bool req_immediate_exit;
  86.  
  87. 	/* Support for PML */
  88. #define PML_ENTITY_NUM		512
  89. 	struct page *pml_pg;
  90.  
  91. 	/* apic deadline value in host tsc */
  92. 	u64 hv_deadline_tsc;
  93.  
  94. 	unsigned long host_debugctlmsr;
  95.  
  96. 	/*
  97. 	 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
  98. 	 * msr_ia32_feature_control. FEAT_CTL_LOCKED is always included
  99. 	 * in msr_ia32_feature_control_valid_bits.
  100. 	 */
  101. 	u64 msr_ia32_feature_control;
  102. 	u64 msr_ia32_feature_control_valid_bits;
  103. 	/* SGX Launch Control public key hash */
  104. 	u64 msr_ia32_sgxlepubkeyhash[4];
  105. 	u64 msr_ia32_mcu_opt_ctrl;
  106. 	bool disable_fb_clear;
  107.  
  108. 	struct pt_desc pt_desc;
  109. 	struct lbr_desc lbr_desc;
  110.  
  111. 	/* Save desired MSR intercept (read: pass-through) state */
  112. #define MAX_POSSIBLE_PASSTHROUGH_MSRS	15
  113. 	struct {
  114. 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
  115. 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
  116. 	} shadow_msr_intercept;
  117. };

对于kvm_init函数的讲解，请看下回。

举报

选择你想要举报的内容（必选）
  • 内容涉黄
  • 政治相关
  • 内容抄袭
  • 涉嫌广告
  • 内容侵权
  • 侮辱谩骂
  • 样式问题
  • 其他

取消

确定