Re: [PATCH v4 13/16] KVM: TDX: Handle PAMT allocation in fault path

[Binbin Wu <binbin.wu@linux.intel.com>]

On 11/21/2025 8:51 AM, Rick Edgecombe wrote:
> From: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
>
> Install PAMT pages for TDX call backs called during the fault path.
>
> There are two distinct cases when the kernel needs to allocate PAMT memory
> in the fault path: for SEPT page tables in tdx_sept_link_private_spt() and
> for leaf pages in tdx_sept_set_private_spte().
>
> These code paths run in atomic context. Previous changes have made the
> fault path top up the per-VCPU pool for memory allocations. Use it to do
> tdx_pamt_get/put() for the fault path operations.
>
> In the generic MMU these ops are inside functions that don’t always
> operate from the vCPU contexts (for example zap paths), which means they
> don’t have a struct kvm_vcpu handy. But for TDX they are always in a vCPU
> context. Since the pool of pre-allocated pages is on the vCPU, use
> kvm_get_running_vcpu() to get the vCPU. In case a new path appears where
> this is not the  case, leave some KVM_BUG_ON()’s.
>
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> [Add feedback, update log]
> Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
> ---
> v4:
>   - Do prealloc.page_list initialization in tdx_td_vcpu_init() in case
>     userspace doesn't call KVM_TDX_INIT_VCPU.
>
> v3:
>   - Use new pre-allocation method
>   - Updated log
>   - Some extra safety around kvm_get_running_vcpu()
> ---
>   arch/x86/kvm/vmx/tdx.c | 44 ++++++++++++++++++++++++++++++++++++------
>   1 file changed, 38 insertions(+), 6 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
> index 61a058a8f159..24322263ac27 100644
> --- a/arch/x86/kvm/vmx/tdx.c
> +++ b/arch/x86/kvm/vmx/tdx.c
> @@ -683,6 +683,8 @@ int tdx_vcpu_create(struct kvm_vcpu *vcpu)
>   	if (!irqchip_split(vcpu->kvm))
>   		return -EINVAL;
>   
> +	INIT_LIST_HEAD(&tdx->prealloc.page_list);
> +

Should this change be moved to patch 12?
Because the pre-alloc page list has started to be used in patch 12 for external
page tables even without enabling dynamic PAMT.

>   	fpstate_set_confidential(&vcpu->arch.guest_fpu);
>   	vcpu->arch.apic->guest_apic_protected = true;
>   	INIT_LIST_HEAD(&tdx->vt.pi_wakeup_list);
>
[...]