tcg_target_long to pointer conversion?

tcg_target_long to pointer conversion?

[Michael Tokarev]

Hi!

This commit:

commit 705ed477d54c5bccf51d924c403a52049586c3d7
Author: Richard Henderson <richard.henderson@linaro.org>
Date:   Wed Oct 28 23:42:12 2020 -0700

     tcg/i386: Support split-wx code generation

contains the following change:

-    diff = arg - ((uintptr_t)s->code_ptr + 7);
+    diff = tcg_pcrel_diff(s, (const void *)arg) - 7;

where arg is of type tcg_target_long.  It looks like this is
the only place in qemu where this type is used as the pointer.

Does it make sense?

This warning is produced when building qemu on x32:

  In file included from tcg/tcg.c:432:
  tcg/i386/tcg-target.c.inc: In function ‘tcg_out_movi_int’:
  tcg/i386/tcg-target.c.inc:1041:30: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
   1041 |     diff = tcg_pcrel_diff(s, (const void *)arg) - 7;

yes, x32 itself isn't a supported target, but it has some
rather unique properties which might catch possible issues.

Thanks,

/mjt

Re: tcg_target_long to pointer conversion?

[Richard Henderson]

On 4/23/23 10:55, Michael Tokarev wrote:
> Hi!
> 
> This commit:
> 
> commit 705ed477d54c5bccf51d924c403a52049586c3d7
> Author: Richard Henderson <richard.henderson@linaro.org>
> Date:   Wed Oct 28 23:42:12 2020 -0700
> 
>      tcg/i386: Support split-wx code generation
> 
> contains the following change:
> 
> -    diff = arg - ((uintptr_t)s->code_ptr + 7);
> +    diff = tcg_pcrel_diff(s, (const void *)arg) - 7;
> 
> where arg is of type tcg_target_long.  It looks like this is
> the only place in qemu where this type is used as the pointer.
> 
> Does it make sense?

Hmm.  No, I don't think it does.  We can't truncate arg from uint64_t and have the test 
work properly.

> 
> This warning is produced when building qemu on x32:
> 
>   In file included from tcg/tcg.c:432:
>   tcg/i386/tcg-target.c.inc: In function ‘tcg_out_movi_int’:
>   tcg/i386/tcg-target.c.inc:1041:30: warning: cast to pointer from integer of different 
> size [-Wint-to-pointer-cast]
>    1041 |     diff = tcg_pcrel_diff(s, (const void *)arg) - 7;
> 
> yes, x32 itself isn't a supported target, but it has some
> rather unique properties which might catch possible issues.

Maybe

     diff = arg - ((uintptr_t)tcg_splitwx_to_rx(s->code_ptr) + 7);


r~

Re: tcg_target_long to pointer conversion?

[Michael Tokarev]

Just found an old thread marked "ToDo"..

23.04.2023 13:40, Richard Henderson wrote:
> On 4/23/23 10:55, Michael Tokarev wrote:
>> Hi!
>>
>> This commit:
>>
>> commit 705ed477d54c5bccf51d924c403a52049586c3d7
>> Author: Richard Henderson <richard.henderson@linaro.org>
>> Date:   Wed Oct 28 23:42:12 2020 -0700
>>
>>      tcg/i386: Support split-wx code generation
>>
>> contains the following change:
>>
>> -    diff = arg - ((uintptr_t)s->code_ptr + 7);
>> +    diff = tcg_pcrel_diff(s, (const void *)arg) - 7;
>>
>> where arg is of type tcg_target_long.  It looks like this is
>> the only place in qemu where this type is used as the pointer.
>>
>> Does it make sense?
> 
> Hmm.  No, I don't think it does.  We can't truncate arg from uint64_t and have the test work properly.
> 
>>
>> This warning is produced when building qemu on x32:
>>
>>   In file included from tcg/tcg.c:432:
>>   tcg/i386/tcg-target.c.inc: In function ‘tcg_out_movi_int’:
>>   tcg/i386/tcg-target.c.inc:1041:30: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
>>    1041 |     diff = tcg_pcrel_diff(s, (const void *)arg) - 7;
>>
>> yes, x32 itself isn't a supported target, but it has some
>> rather unique properties which might catch possible issues.
> 
> Maybe
> 
>      diff = arg - ((uintptr_t)tcg_splitwx_to_rx(s->code_ptr) + 7);

I know almost nothing about tcg.. :)

/mjt