[MODERATED] Re: [patch V5 05/10] KVM magic # 5

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

..snip..
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -9576,9 +9576,20 @@ static int vmx_handle_exit(struct kvm_vc
>  #define L1D_CACHE_ORDER 4
>  static void *vmx_l1d_flush_pages;
>  
> -static void __maybe_unused vmx_l1d_flush(void)
> +static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
>  {
>  	int size = PAGE_SIZE << L1D_CACHE_ORDER;
> +	bool always;
> +
> +	/*
> +	 * If the mitigation mode is 'flush always', keep the flush bit
> +	 * set, otherwise clear it. It gets set again either from
> +	 * vcpu_run() or from one of the unsafe VMEXIT handlers.
> +	 */
> +	always = vmentry_l1d_flush == VMENTER_L1D_FLUSH_ALWAYS;
> +	vcpu->arch.l1tf_flush_l1d = always;

<blinks> You did the reset of arch.l1tf_flush_l1d _after_ we have done
this vmx_l1d_flush call, nice!! So obvious in retrospect.

See below one tiny comment.
..snip..
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -195,6 +195,7 @@ struct kvm_stats_debugfs_item debugfs_en
>  	{ "irq_injections", VCPU_STAT(irq_injections) },
>  	{ "nmi_injections", VCPU_STAT(nmi_injections) },
>  	{ "req_event", VCPU_STAT(req_event) },
> +	{ "l1d_flush", VCPU_STAT(l1d_flush) },
>  	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
>  	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
>  	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
> @@ -4799,6 +4800,8 @@ int kvm_read_guest_virt(struct kvm_vcpu
>  {
>  	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
>  
> +	/* The gva_to_pa walker can pull in tons of pages. */
> +	vcpu->arch.l1tf_flush_l1d = true;


I think also kvm_write_guest_virt_system ? That covers vmptrs and vmread.


>  	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
>  					  exception);
>  }
> @@ -6050,6 +6053,8 @@ int x86_emulate_instruction(struct kvm_v
>  	bool writeback = true;
>  	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
>  
> +	vcpu->arch.l1tf_flush_l1d = true;
> +
>  	/*
>  	 * Clear write_fault_to_shadow_pgtable here to ensure it is
>  	 * never reused.
> @@ -7579,6 +7584,7 @@ static int vcpu_run(struct kvm_vcpu *vcp
>  	struct kvm *kvm = vcpu->kvm;
>  
>  	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
> +	vcpu->arch.l1tf_flush_l1d = true;
>  
>  	for (;;) {
>  		if (kvm_vcpu_running(vcpu)) {
> @@ -8698,6 +8704,7 @@ void kvm_arch_vcpu_uninit(struct kvm_vcp
>  
>  void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
>  {
> +	vcpu->arch.l1tf_flush_l1d = true;
>  	kvm_x86_ops->sched_in(vcpu, cpu);
>  }
>  
>