Re: [PATCH 0/9] x86/vvmx: Read instruction operands correctly on VM exit

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 26/10/17 18:03, Euan Harris wrote:
> decode_vmx_inst() does not read instruction operands correctly on VM exit:
>
>  * It incorrectly uses vmx_inst_info's address_size field to calculate
>    the sizes of the exit-causing instruction's operands.  The sizes of
>    the operands are specified in the SDM and might depend on whether the
>    guest is running in 32-bit or 64-bit mode, but they have nothing to do
>    with the address_size field.
>
>  * It includes its own segmentation logic, duplicating code elsewhere.
>    This segmentation logic is also incorrect and will raise #GP fault
>    rather than a #SS fault in response to an invalid memory access
>    through the stack segment.
>  
> Patches 1-6 (up to 'Remove operand decoding from decode_vmx_inst()')
> refactor decode_vmx_inst() in preparation for fixing the bugs mentioned
> above.  They remove unnecessary code and extract the logic for reading
> operands from decode_vmx_inst() into a new operand_read() function.
> These patches should not cause any functional changes.
>
> Patch 7 ('Use correct sizes when reading operands') replaces the incorrect
> operand size calculations based on address_size with the correct sizes
> from the SDM.
>
> Patches 8 and 9 add new hvm_copy_{to,from}_guest_virt() helpers and use
> them to read memory operands in place of the incorrect segmentation
> logic in decode_vmx_inst().
>
> Euan Harris (9):
>   x86/vvmx: Remove enum vmx_regs_enc
>   x86/vvmx: Unify operands in struct vmx_inst_decoded
>   x86/vvmx: Extract operand reading logic into operand_read()
>   x86/vvmx: Remove unnecessary VMX operand reads
>   x86/vvmx: Replace direct calls to reg_read() with operand_read()
>   x86/vvmx: Remove operand reading from decode_vmx_inst()
>   x86/vvmx: Use correct sizes when reading operands
>   x86/hvm: Add hvm_copy_{to,from}_guest_virt() helpers
>   x86/vvmx: Use hvm_copy_{to,from}_guest_virt() to read operands

All Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>.  I've
noticed a few trivial style issues which can be fixed up on commit if
there are no other issues.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel