Re: [Xen-devel] [PATCH v11 2/5] x86emul: New return code for unimplemented instruction

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 20.09.17 at 23:47, <ppircalabu@xxxxxxxxxxxxxxx> wrote:
> On Ma, 2017-09-19 at 09:19 -0600, Jan Beulich wrote:
>> > > > On 12.09.17 at 16:32, <ppircalabu@xxxxxxxxxxxxxxx> wrote:
>> > @@ -2102,6 +2112,7 @@ void hvm_emulate_one_vm_event(enum emul_kind
>> > kind, unsigned int trapnr,
>> >           * consistent with X86EMUL_RETRY.
>> >           */
>> >          return;
>> > +    case X86EMUL_UNIMPLEMENTED:
>> >      case X86EMUL_UNHANDLEABLE:
>> >          hvm_dump_emulation_state(XENLOG_G_DEBUG, "Mem event",
>> > &ctx);
>> ... you had added the new case label below existing ones uniformly.
>> But anyway.
> The order is reversed in this case because the patch 4/5 from this
> series adds the hvm_monitor_emul_unimplemented call and the "Fall-
> through" in case of failure. If I change the default order here, I will
> have to reverse it when adding the monitor notification handling.

Ah, okay, that's fine then.

>> > @@ -6195,7 +6196,7 @@ x86_emulate(
>> >                  /* vpsll{w,d} $imm8,{x,y}mm,{x,y}mm */
>> >              break;
>> >          default:
>> > -            goto cannot_emulate;
>> > +            goto unimplemented_insn;
>> >          }
>> This again wants to be "unrecognized".
> In this case "unrecognized" cannot be returned (yet) as instructions
> VPRORD and VPROLD are not implemented.
> http://sandpile.org/x86/opc_grp.htm (group #13 (EVEX 66h) (0Fh,72h) )

As you say yourself - these are EVEX-prefixed. We don't support
any AVX-512 instructions so far. As you're certainly aware even
the completion of AVX, AVX2, etc is pending review (and now
unlikely to make it into 4.10). Hence the "unimplemented" status
is supposed to be coming from a much earlier (decode stage)
place for these insns.

Jan


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