question about nested ntp

I have a question about nested ntp code. Comment in init_kvm_nested_mmu() function says "arch.mmu.gva_to_gpa translates l2_gva to l1_gpa". If I read the code correctly at the time of nested guest execution arch.mmu.gva_to_gpa() points to one of paging*_gva_to_gpa() variants. The function calls walk_addr() with provided address which should be l2_gva according to the comment. Now when NPF happens while nested guest runs arch.mmu.page_fault() is called which translates to call to one of paging*_page_fault() variants. paging*_page_fault() calls walk_addr() too, but now it gets l2_gpa as a parameter since during NPF svm->vmcb->control.exit_info_2 will not point to l2_gva, but l2_gpa instead. Do I miss something here?

Hi Gleb,

Quoted 11 lines · Gleb Natapov

I have a question about nested ntp code. Comment in init_kvm_nested_mmu() function says "arch.mmu.gva_to_gpa translates l2_gva to l1_gpa". If I read the code correctly at the time of nested guest execution arch.mmu.gva_to_gpa() points to one of paging*_gva_to_gpa() variants. The function calls walk_addr() with provided address which should be l2_gva according to the comment. Now when NPF happens while nested guest runs arch.mmu.page_fault() is called which translates to call to one of paging*_page_fault() variants. paging*_page_fault() calls walk_addr() too, but now it gets l2_gpa as a parameter since during NPF svm->vmcb->control.exit_info_2 will not point to l2_gva, but l2_gpa instead. Do I miss something here?

No, the comment is wrong. The comment actually was right in -v1 of the patch-series. But I changed that in -v2 and forgot the update/remove the comment. Thanks for mentioning that :-)