QEMU源码全解析 —— CPU虚拟化（16）

已于 2024-07-31 19:40:56 修改
阅读量225 收藏 1
点赞数 3
分类专栏： QEMU/KVM 文章标签： QEMU KVM CPU虚拟化
于 2024-07-31 16:48:01 首次发布
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/140815904
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

本回开始对于kvm_init函数进行解析。首先贴出Linux 6.1.10内核版本下该函数的源码，在Linux内核源码根目录/virt/kvm/kvm_main.c中，代码如下： 

  1. int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
  2. 		  struct module *module)
  3. {
  4. 	struct kvm_cpu_compat_check c;
  5. 	int r;
  6. 	int cpu;
  7.  
  8. 	r = kvm_arch_init(opaque);
  9. 	if (r)
  10. 		goto out_fail;
  11.  
  12. 	/*
  13. 	 * kvm_arch_init makes sure there's at most one caller
  14. 	 * for architectures that support multiple implementations,
  15. 	 * like intel and amd on x86.
  16. 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
  17. 	 * conflicts in case kvm is already setup for another implementation.
  18. 	 */
  19. 	r = kvm_irqfd_init();
  20. 	if (r)
  21. 		goto out_irqfd;
  22.  
  23. 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
  24. 		r = -ENOMEM;
  25. 		goto out_free_0;
  26. 	}
  27.  
  28. 	r = kvm_arch_hardware_setup(opaque);
  29. 	if (r < 0)
  30. 		goto out_free_1;
  31.  
  32. 	c.ret = &r;
  33. 	c.opaque = opaque;
  34. 	for_each_online_cpu(cpu) {
  35. 		smp_call_function_single(cpu, check_processor_compat, &c, 1);
  36. 		if (r < 0)
  37. 			goto out_free_2;
  38. 	}
  39.  
  40. 	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
  41. 				      kvm_starting_cpu, kvm_dying_cpu);
  42. 	if (r)
  43. 		goto out_free_2;
  44. 	register_reboot_notifier(&kvm_reboot_notifier);
  45.  
  46. 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
  47. 	if (!vcpu_align)
  48. 		vcpu_align = __alignof__(struct kvm_vcpu);
  49. 	kvm_vcpu_cache =
  50. 		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
  51. 					   SLAB_ACCOUNT,
  52. 					   offsetof(struct kvm_vcpu, arch),
  53. 					   offsetofend(struct kvm_vcpu, stats_id)
  54. 					   - offsetof(struct kvm_vcpu, arch),
  55. 					   NULL);
  56. 	if (!kvm_vcpu_cache) {
  57. 		r = -ENOMEM;
  58. 		goto out_free_3;
  59. 	}
  60.  
  61. 	for_each_possible_cpu(cpu) {
  62. 		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
  63. 					    GFP_KERNEL, cpu_to_node(cpu))) {
  64. 			r = -ENOMEM;
  65. 			goto out_free_4;
  66. 		}
  67. 	}
  68.  
  69. 	r = kvm_async_pf_init();
  70. 	if (r)
  71. 		goto out_free_4;
  72.  
  73. 	kvm_chardev_ops.owner = module;
  74.  
  75. 	r = misc_register(&kvm_dev);
  76. 	if (r) {
  77. 		pr_err("kvm: misc device register failed\n");
  78. 		goto out_unreg;
  79. 	}
  80.  
  81. 	register_syscore_ops(&kvm_syscore_ops);
  82.  
  83. 	kvm_preempt_ops.sched_in = kvm_sched_in;
  84. 	kvm_preempt_ops.sched_out = kvm_sched_out;
  85.  
  86. 	kvm_init_debug();
  87.  
  88. 	r = kvm_vfio_ops_init();
  89. 	WARN_ON(r);
  90.  
  91. 	return 0;
  92.  
  93. out_unreg:
  94. 	kvm_async_pf_deinit();
  95. out_free_4:
  96. 	for_each_possible_cpu(cpu)
  97. 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
  98. 	kmem_cache_destroy(kvm_vcpu_cache);
  99. out_free_3:
  100. 	unregister_reboot_notifier(&kvm_reboot_notifier);
  101. 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
  102. out_free_2:
  103. 	kvm_arch_hardware_unsetup();
  104. out_free_1:
  105. 	free_cpumask_var(cpus_hardware_enabled);
  106. out_free_0:
  107. 	kvm_irqfd_exit();
  108. out_irqfd:
  109. 	kvm_arch_exit();
  110. out_fail:
  111. 	return r;
  112. }
  113. EXPORT_SYMBOL_GPL(kvm_init);

为了便于理解，再次贴出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;

kvm_init函数总体调用的函数如下图所示：

一个函数一个函数来看。

（1）kvm_arch_init函数

代码片段如下： 

  1. 	r = kvm_arch_init(opaque);
  2. 	if (r)
  3. 		goto out_fail;

kvm_arch_init函数用来初始化架构相关的代码。它是一个体系结构相关的函数，在不同的体系结构中均有实现，视不同体系结构不同而不同。

这里仍以x86为例。x86体系结构下的kvm_arch_init函数在Linux内核源码根目录/arch/x86/kvm/x86.c中，代码如下： 

  1. int kvm_arch_init(void *opaque)
  2. {
  3. 	struct kvm_x86_init_ops *ops = opaque;
  4. 	u64 host_pat;
  5. 	int r;
  6.  
  7. 	if (kvm_x86_ops.hardware_enable) {
  8. 		pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
  9. 		return -EEXIST;
  10. 	}
  11.  
  12. 	if (!ops->cpu_has_kvm_support()) {
  13. 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
  14. 				   ops->runtime_ops->name);
  15. 		return -EOPNOTSUPP;
  16. 	}
  17. 	if (ops->disabled_by_bios()) {
  18. 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
  19. 				   ops->runtime_ops->name);
  20. 		return -EOPNOTSUPP;
  21. 	}
  22.  
  23. 	/*
  24. 	 * KVM explicitly assumes that the guest has an FPU and
  25. 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
  26. 	 * vCPU's FPU state as a fxregs_state struct.
  27. 	 */
  28. 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
  29. 		printk(KERN_ERR "kvm: inadequate fpu\n");
  30. 		return -EOPNOTSUPP;
  31. 	}
  32.  
  33. 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
  34. 		pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
  35. 		return -EOPNOTSUPP;
  36. 	}
  37.  
  38. 	/*
  39. 	 * KVM assumes that PAT entry '0' encodes WB memtype and simply zeroes
  40. 	 * the PAT bits in SPTEs.  Bail if PAT[0] is programmed to something
  41. 	 * other than WB.  Note, EPT doesn't utilize the PAT, but don't bother
  42. 	 * with an exception.  PAT[0] is set to WB on RESET and also by the
  43. 	 * kernel, i.e. failure indicates a kernel bug or broken firmware.
  44. 	 */
  45. 	if (rdmsrl_safe(MSR_IA32_CR_PAT, &host_pat) ||
  46. 	    (host_pat & GENMASK(2, 0)) != 6) {
  47. 		pr_err("kvm: host PAT[0] is not WB\n");
  48. 		return -EIO;
  49. 	}
  50.  
  51. 	x86_emulator_cache = kvm_alloc_emulator_cache();
  52. 	if (!x86_emulator_cache) {
  53. 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
  54. 		return -ENOMEM;
  55. 	}
  56.  
  57. 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
  58. 	if (!user_return_msrs) {
  59. 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
  60. 		r = -ENOMEM;
  61. 		goto out_free_x86_emulator_cache;
  62. 	}
  63. 	kvm_nr_uret_msrs = 0;
  64.  
  65. 	r = kvm_mmu_vendor_module_init();
  66. 	if (r)
  67. 		goto out_free_percpu;
  68.  
  69. 	kvm_timer_init();
  70.  
  71. 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
  72. 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
  73. 		kvm_caps.supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
  74. 	}
  75.  
  76. 	if (pi_inject_timer == -1)
  77. 		pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
  78. #ifdef CONFIG_X86_64
  79. 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
  80.  
  81. 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
  82. 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
  83. #endif
  84.  
  85. 	return 0;
  86.  
  87. out_free_percpu:
  88. 	free_percpu(user_return_msrs);
  89. out_free_x86_emulator_cache:
  90. 	kmem_cache_destroy(x86_emulator_cache);
  91. 	return r;
  92. }

kvm_arch_init函数做了一些初始化的工作，确保只有一个KVM能够加载到内核。KVM实现的结构体会被赋值到全局变量kvm_x86_ops中（如前文所述），这里传递到kvm_arch_init的参数opaque就是vmx_x86_ops（新版本中封了一层，意思一样），在其中存放了Intel CPU下KVM实现的各类回调函数。

再来回顾一下vmx_x86_ops的相关代码，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. };

kvm_arch_init函数调用实现相关的回调函数cpu_has_kvm_support()和disable_by_bios()。代码片段如下： 

  1.     if (!ops->cpu_has_kvm_support()) {
  2. 		pr_err_ratelimited("kvm: no hardware support for '%s'\n",
  3. 				   ops->runtime_ops->name);
  4. 		return -EOPNOTSUPP;
  5. 	}
  6. 	if (ops->disabled_by_bios()) {
  7. 		pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
  8. 				   ops->runtime_ops->name);
  9. 		return -EOPNOTSUPP;
  10. 	}

cpu_has_kvm_support()用来检测CPU是否支持VMX模式（对应开启条件1），通过CPUID指令的返回值判断；disable_by_bios()用来检测是否被BIOS关闭（对应开启条件7），通过读取MSR寄存器判断。

kvm_arch_init函数其它的初始化工作包括：

  • 分配一个percpu变量user_return_msrs

代码片段如下： 

  1. 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
  2. 	if (!user_return_msrs) {
  3. 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
  4. 		r = -ENOMEM;
  5. 		goto out_free_x86_emulator_cache;
  6. 	}
  7. 	kvm_nr_uret_msrs = 0;
  • 完成内存虚拟化的初始化工作

代码片段如下： 

  1. 	r = kvm_mmu_vendor_module_init();
  2. 	if (r)
  3. 		goto out_free_percpu;
  • 完成timer的初始化

代码片段如下： 

    kvm_timer_init();

当然，还有其它一些功能，这里不一一详述。

这里提一下，在更早的内核版本中，kvm_arch_init函数中还有以下函数：

  • kvm_set_mmio_spte_mask()

设置MMIO内存的标识符（该标识符通过shadow_mmio_mask表示）。

  • kvm_lapic_init()

完成lapic的初始化。

但在6.1.10内核版本中，这几个函数已经没有了。

本回内容较多，暂告一段落，下一回继续解析kvm_arch_init函数。

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