[Qemu-devel] [PATCH] x86-64: Fix 64-bit lgs/lfs/lss

[Qemu-devel] [PATCH] x86-64: Fix 64-bit lgs/lfs/lss

[Jan Kiszka]

Reading the code while porting my segment limit and type checks to
latest SVN made me stumble over this bug in the translator: 64-bit
lgs/lfs/lss was incorrectly reading only 32-bit offsets.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---
 target-i386/translate.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: b/target-i386/translate.c
===================================================================
--- a/target-i386/translate.c
+++ b/target-i386/translate.c
@@ -4864,7 +4864,7 @@ static target_ulong disas_insn(DisasCont
     case 0x1b5: /* lgs Gv */
         op = R_GS;
     do_lxx:
-        ot = dflag ? OT_LONG : OT_WORD;
+        ot = dflag + OT_WORD;
         modrm = ldub_code(s->pc++);
         reg = ((modrm >> 3) & 7) | rex_r;
         mod = (modrm >> 6) & 3;
@@ -4872,7 +4872,7 @@ static target_ulong disas_insn(DisasCont
             goto illegal_op;
         gen_lea_modrm(s, modrm, &reg_addr, &offset_addr);
         gen_op_ld_T1_A0(ot + s->mem_index);
-        gen_add_A0_im(s, 1 << (ot - OT_WORD + 1));
+        gen_add_A0_im(s, 1 << ot);
         /* load the segment first to handle exceptions properly */
         gen_op_ldu_T0_A0(OT_WORD + s->mem_index);
         gen_movl_seg_T0(s, op, pc_start - s->cs_base);

Re: [Qemu-devel] [PATCH] x86-64: Fix 64-bit lgs/lfs/lss

[Fabrice Bellard]

Hi,

OK, at least for Intel CPUs. From the AMD & Intel specs, the behavior 
seems to differ between the two. If it is really the case, a CPU 
specific config bit is needed. Unfortunately there may be other corner 
cases where AMD and Intel differ in 64 bit mode.

Fabrice.

Jan Kiszka wrote:
> Reading the code while porting my segment limit and type checks to
> latest SVN made me stumble over this bug in the translator: 64-bit
> lgs/lfs/lss was incorrectly reading only 32-bit offsets.
> 
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>  target-i386/translate.c |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> Index: b/target-i386/translate.c
> ===================================================================
> --- a/target-i386/translate.c
> +++ b/target-i386/translate.c
> @@ -4864,7 +4864,7 @@ static target_ulong disas_insn(DisasCont
>      case 0x1b5: /* lgs Gv */
>          op = R_GS;
>      do_lxx:
> -        ot = dflag ? OT_LONG : OT_WORD;
> +        ot = dflag + OT_WORD;
>          modrm = ldub_code(s->pc++);
>          reg = ((modrm >> 3) & 7) | rex_r;
>          mod = (modrm >> 6) & 3;
> @@ -4872,7 +4872,7 @@ static target_ulong disas_insn(DisasCont
>              goto illegal_op;
>          gen_lea_modrm(s, modrm, &reg_addr, &offset_addr);
>          gen_op_ld_T1_A0(ot + s->mem_index);
> -        gen_add_A0_im(s, 1 << (ot - OT_WORD + 1));
> +        gen_add_A0_im(s, 1 << ot);
>          /* load the segment first to handle exceptions properly */
>          gen_op_ldu_T0_A0(OT_WORD + s->mem_index);
>          gen_movl_seg_T0(s, op, pc_start - s->cs_base);
> 
> 
> 
> 

[Qemu-devel] Re: [PATCH] x86-64: Fix 64-bit lgs/lfs/lss

[Jan Kiszka]

Fabrice Bellard wrote:
> Hi,
> 
> OK, at least for Intel CPUs. From the AMD & Intel specs, the behavior
> seems to differ between the two. If it is really the case, a CPU
> specific config bit is needed. Unfortunately there may be other corner
> cases where AMD and Intel differ in 64 bit mode.

Indeed, just checked: there is a difference in the specs. Will update
this patch to make it CPU-specific.

Jan

[Qemu-devel] Re: [PATCH] x86-64: Fix 64-bit lgs/lfs/lss

[Jan Kiszka]

Jan Kiszka wrote:
> Fabrice Bellard wrote:
>> Hi,
>>
>> OK, at least for Intel CPUs. From the AMD & Intel specs, the behavior
>> seems to differ between the two. If it is really the case, a CPU
>> specific config bit is needed. Unfortunately there may be other corner
>> cases where AMD and Intel differ in 64 bit mode.
> 
> Indeed, just checked: there is a difference in the specs. Will update
> this patch to make it CPU-specific.

Hmm, there is no Intel cpu type yet that supports 64-bit...

So I guess we better keeps things as-is for now before claiming to be
Intel-compatible while other broken corner cases are waiting to bite us.

Jan