performance: 0.825
graphic: 0.807
device: 0.793
debug: 0.789
PID: 0.786
other: 0.778
files: 0.768
vnc: 0.762
permissions: 0.749
boot: 0.749
socket: 0.709
network: 0.693
KVM: 0.628
semantic: 0.520

illegal instructions for AArch64 ARMv8

The test case is in the attachment. To reproduce as following (I tried both GCC and Clang):
$aarch64-linux-gnu-gcc qemu.c -o test
$./test
qemu: uncaught target signal 4 (Illegal instruction) - core dumped
Illegal instruction (core dumped)

There are 3 intrinsics are tested in the test case: vqmovunh_s16,  vqmovuns_s32, vqmovund_s64. They will be compiled into instructions:
SQXTUN Bd, Hn
SQXTUN Hd, Sn
SQXTUN Sd, Dn.

It seems that these instructions are not supported in QEMU. Is this a bug?



Can you attach a statically linked test case binary, please?



Peter Maydell <email address hidden> writes:

> Can you attach a statically linked test case binary, please?

I can reproduce with the source file. It looks like:

@@ -7553,12 +7555,9 @@ static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
         }
         break;
     case 0x12: /* SQXTUN */
-        if (u) {
-            unallocated_encoding(s);
-            return;
-        }
         /* fall through */

Fixes it. Let me check why this slipped through the risu tests and
re-validate. I'll submit a patch once I've double checked.

-- 
Alex Bennée



On 16 April 2014 11:55, Alex Bennée <email address hidden> wrote:
>
> Peter Maydell <email address hidden> writes:
>
>> Can you attach a statically linked test case binary, please?
>
> I can reproduce with the source file. It looks like:
>
> @@ -7553,12 +7555,9 @@ static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
>          }
>          break;
>      case 0x12: /* SQXTUN */
> -        if (u) {
> -            unallocated_encoding(s);
> -            return;
> -        }
>          /* fall through */
>
> Fixes it.

However the ARM ARM, unless I'm misreading it, requires scalar-2-misc
SQXTUN to have U==1, so the correct fix should be to turn that "if (u)"
into "if (!u)" I think. (Opcode 0x12 u==0 isn't in the table so should undef.)

Better check we didn't make the same mistake in the vector-2-misc
decode as well.

thanks
-- PMM


Fix identified

I've sent this patch to the mailing list but it fixes the attached test case and has been tested with risu patterns.

@pmaydell: yeah vector is unaffected as U is used to select another opcode.

Patch had been included here:
http://git.qemu.org/?p=qemu.git;a=commitdiff;h=e44a90c59697cf98
==> Fix released