Re: [Qemu-devel] [PATCH] target/m68k: fix set_cc_op()

[Laurent Vivier]

Le 21/12/2017 à 16:58, Richard Henderson a écrit :
> On 12/21/2017 06:29 AM, Laurent Vivier wrote:
>> The first call of set_cc_op() in a new translation sequence
>> is done with old_op set to CC_OP_DYNAMIC (-1).
>>
>> This will do an out of bound access to the array cc_op_live[].
>>
>> We fix that by adding an entry in cc_op_live[] for CC_OP_DYNAMIC,
>> and in this case we can also discard QREG_CC_OP.
>>
>> Reported-by: Thomas Huth <address@hidden>
>> Signed-off-by: Laurent Vivier <address@hidden>
>> ---
>>  target/m68k/cpu.h       | 2 +-
>>  target/m68k/translate.c | 6 ++++++
>>  2 files changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/target/m68k/cpu.h b/target/m68k/cpu.h
>> index afae5f68ac..5d03764eab 100644
>> --- a/target/m68k/cpu.h
>> +++ b/target/m68k/cpu.h
>> @@ -182,7 +182,7 @@ void cpu_m68k_set_fpcr(CPUM68KState *env, uint32_t val);
>>   */
>>  typedef enum {
>>      /* Translator only -- use env->cc_op.  */
>> -    CC_OP_DYNAMIC = -1,
>> +    CC_OP_DYNAMIC,
>>  
>>      /* Each flag bit computed into cc_[xcnvz].  */
>>      CC_OP_FLAGS,
>> diff --git a/target/m68k/translate.c b/target/m68k/translate.c
>> index b60909222c..61ac1a8e83 100644
>> --- a/target/m68k/translate.c
>> +++ b/target/m68k/translate.c
>> @@ -207,6 +207,7 @@ typedef void (*disas_proc)(CPUM68KState *env, 
>> DisasContext *s, uint16_t insn);
>>  #endif
>>  
>>  static const uint8_t cc_op_live[CC_OP_NB] = {
>> +    [CC_OP_DYNAMIC] = CCF_C | CCF_V | CCF_Z | CCF_N | CCF_X,
>>      [CC_OP_FLAGS] = CCF_C | CCF_V | CCF_Z | CCF_N | CCF_X,
>>      [CC_OP_ADDB ... CC_OP_ADDL] = CCF_X | CCF_N | CCF_V,
>>      [CC_OP_SUBB ... CC_OP_SUBL] = CCF_X | CCF_N | CCF_V,
> 
> These bits are fine.
> 
>> @@ -237,6 +238,11 @@ static void set_cc_op(DisasContext *s, CCOp op)
>>      if (dead & CCF_V) {
>>          tcg_gen_discard_i32(QREG_CC_V);
>>      }
>> +
>> +    /* Discard any computed CC_OP value */
>> +    if (old_op == CC_OP_DYNAMIC) {
>> +        tcg_gen_discard_i32(QREG_CC_OP);
>> +    }
> 
> This doesn't help, since the previous setting of CC_OP is certain to be in a
> different TB and therefore there is no code within the current TB to discard.

OK, I remove this.

Thanks,
Laurent