On 05/03/2017 19:21, Alexander Boettcher wrote: > the SVM I/O permission bitmap for user-level (ring-3) VM code running in > SVM seems to be ignored and causes a GP-fault. (Actual the IO permission > was granted by the kernel via the TSS I/O port permission bitmap). > > After some debugging the GP code originates from target/i386/translate.c > gen_check_io() within the if(s->pe && (s->cpl > s->iopl || s->vm86)) > condition. However, the actual SVM IO permission bitmap is checked after > that condition, which succeeds and would permit the access.
>From your message it's not clear what is going wrong. The code as is written now matches the AMD manual: "Exceptions related to virtual x86 mode, IOPL, or the TSS-bitmap are checked before the SVM intercept check. All other exceptions are checked after the SVM intercept check". Please explain better what is going on: 1) does the TSS I/O permission bitmap grant permission to access the port (the answer seems to be yes here)? 2) does the SVM I/O permission bitmap grant permission to access the port? 3) you get a #GP, do you expect the access to be trapped to the hypervisor or not? 4) what is the exact instruction that the user-level code is executing? Paolo
