|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH] x86emul: Zero-extend the XOP0A and MOV B8+r immediates when decoding
On 18.11.2026 09:57, Stephen Cheng wrote:
> The decoder sign-extends the 32-bit immediate of the XOP0A insns
> (BEXTR, LWPINS, LWPVAL), but this immediate is unsigned. It also
> sign-extends the immediate of MOV B8+r without REX.W, while with REX.W
> the low half is already zero-extended to build the 64-bit immediate.
> In 64-bit mode MOV r32,imm32 zero-extends into the full register, and
> no form of B8+r sign-extends its immediate. Emulation uses only the low
> bits, so only x86_insn_immediate() returns a wrong value. No current
> caller reads it for these insns.
>
> Zero-extend the XOP0A immediate, as commit aa6070d2c840 ("x86emul:
> Zero-extend unsigned 8-bit immediates when decoding") did for the 0F,
> 0F3A and XOP08 maps. For MOV B8+r, zero-extend the immediate for all
> operand sizes, and keep only the fetch of the high half in the REX.W
> path. Other insns' 16-bit immediates are left as they are for now.
>
> Fixes: 653cae725cea ("x86/32on64: use generic instruction decoding for call
> gate emulation")
> Requested-by: Jan Beulich <jbeulich@xxxxxxxx>
> Assisted-by: Claude:claude-opus-5-5
> Signed-off-by: Stephen Cheng <stephen.cheng@xxxxxxxxxx>
Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
> I left the 16-bit immediates of other insns as they are, because
> changing the shared fetch would affect many insns.
Sure. For ALU insns with 16-bit operand size it is ambiguous anyway whether
to treat the unused bits as sign- or zero-extended.
> --- a/xen/arch/x86/x86_emulate/decode.c
> +++ b/xen/arch/x86/x86_emulate/decode.c
> @@ -683,9 +683,10 @@ decode_onebyte(struct x86_emulate_state *s,
> break;
>
> case 0xb8 ... 0xbf: /* mov imm{16,32,64},r{16,32,64} */
> + /* The immediate is zero-extended, not sign-extended. */
> + s->imm1 &= s->op_bytes == 2 ? 0xffff : 0xffffffffU;
> if ( s->op_bytes == 8 ) /* Fetch more bytes to obtain imm64. */
> - s->imm1 = ((uint32_t)s->imm1 |
> - ((uint64_t)insn_fetch_type(uint32_t) << 32));
> + s->imm1 |= (uint64_t)insn_fetch_type(uint32_t) << 32;
> break;
You covering the 16-bit case here as well makes me wonder about the 8-bit
one. I think we want to be consistent: Either we leave the 16-bit form
alone here (then a v2 would be needed), and then we don't need to do
anything else for the 8-bit form. Or we make the 8-bit form match the 16-
bit one (then another small patch amending the earlier one would be
needed, while this patch could go in as-is). Personally I favor the
former option, as that's less change in behavior (and less churn) overall.
Jan
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |