Re: [PATCH v8 1/2] x86emul: New return code for unimplemented instruction

[Petre Ovidiu PIRCALABU <ppircalabu@bitdefender.com>]

[-- Attachment #1.1: Type: text/plain, Size: 2563 bytes --]

Hi Jan,


The main use-case for the new return code is to have a clear distinction between an instruction not implemented by the emulator (e.g. ?fldenv or fnstenv) and the failure to emulate .


- hvm_process_io_incercept returns X86EMUL_UNHANDLEABLE if one of the hvm_io_ops (read/write) failed or one of the hvm_copy_to(_from)_guest_phys returned an error code which is not handled in their correspondent switch statement. In either cases this is not caused by an unimplemented instruction.

- hvm_do_io / hvm_do_io_buffer call hvm_process_io_incercept which cannot return unimplemented.

- Thank-you very much for pointing out the invoke_stub issue. I have added a new label "unimplemented_insn" and updated the patch.


I will re-send a new patchset with the changes and a more detailed description of the places where the new return value was not handled.


Many thanks,

Petre


________________________________
From: Jan Beulich <JBeulich@suse.com>
Sent: Tuesday, August 22, 2017 11:09 AM
To: Petre Ovidiu PIRCALABU
Cc: rcojocaru@bitdefender.com; andrew.cooper3@citrix.com; paul.durrant@citrix.com; wei.liu2@citrix.com; George.Dunlap@eu.citrix.com; ian.jackson@eu.citrix.com; jun.nakajima@intel.com; kevin.tian@intel.com; sstabellini@kernel.org; xen-devel@lists.xen.org; konrad.wilk@oracle.com; tamas@tklengyel.com; tim@xen.org
Subject: Re: [PATCH v8 1/2] x86emul: New return code for unimplemented instruction

>>> On 08.08.17 at 20:06, <ppircalabu@bitdefender.com> wrote:
> --- a/xen/arch/x86/hvm/emulate.c
> +++ b/xen/arch/x86/hvm/emulate.c

What about the use in a switch() statement in hvmemul_do_io()
in this file? And the use in hvmemul_do_io_buffer()?

> @@ -2044,6 +2044,8 @@ int hvm_emulate_one_mmio(unsigned long mfn, unsigned long gla)
>      switch ( rc )
>      {
>      case X86EMUL_UNHANDLEABLE:
> +        /* fall-through */
> +    case X86EMUL_UNIMPLEMENTED:

The fall-through comment is pointless in such a case.

hvm/intercept.c has a use in hvm_process_io_intercept() which
looks like it needs dealing with too. And there are more. Any
places you perhaps leave alone intentionally should be reasoned
about in the description.

> @@ -7717,7 +7717,7 @@ x86_emulate(
>
>      default:
>      cannot_emulate:
> -        rc = X86EMUL_UNHANDLEABLE;
> +        rc = X86EMUL_UNIMPLEMENTED;

There's at least one goto to the label here which can't stay as is
(in invoke_stub()). Did you really audit them all?

Jan


________________________
This email was scanned by Bitdefender

[-- Attachment #1.2: Type: text/html, Size: 4073 bytes --]

[-- Attachment #2: Type: text/plain, Size: 127 bytes --]

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