Re: [Qemu-devel] [RFC PATCH 2/3] tcg: Add support for fence generation in x86 backend

[Alex Bennée]

Pranith Kumar <address@hidden> writes:

> Hi Richard,
>
> Thank you for the helpful comments.
>
> On Wed, May 25, 2016 at 1:35 PM, Richard Henderson <address@hidden> wrote:
>> On 05/24/2016 10:18 AM, Pranith Kumar wrote:
>>> diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
>>> index 92be341..93ea42e 100644
>>> --- a/tcg/i386/tcg-target.h
>>> +++ b/tcg/i386/tcg-target.h
>>> @@ -100,6 +100,7 @@ extern bool have_bmi1;
>>>  #define TCG_TARGET_HAS_muls2_i32        1
>>>  #define TCG_TARGET_HAS_muluh_i32        0
>>>  #define TCG_TARGET_HAS_mulsh_i32        0
>>> +#define TCG_TARGET_HAS_fence            1
>>
>>
>> This has to be defined for all hosts.
>
> OK. I will add an entry in tcg.h with default 0 and override in
> individual architecture once it is implemented.
>
>>> @@ -347,6 +347,7 @@ static inline int
>>> tcg_target_const_match(tcg_target_long val, TCGType type,
>>>  #define OPC_SHRX        (0xf7 | P_EXT38 | P_SIMDF2)
>>>  #define OPC_TESTL      (0x85)
>>>  #define OPC_XCHG_ax_r32        (0x90)
>>> +#define OPC_MFENCE      (0xAE | P_EXT)
>>
>> Why define OPC_MFENCE if you're not going to use it?  Of course, it's not
>> exactly a complete and useful definition, so maybe just delete OPC_MFENCE.
>
> I want to use OPC_MFENCE instead of hard-coding the value in
> tcg_out_fence(), but as you said the definition is not complete(it
> currently generates only 0x0FAE). I am trying to figure out how to
> generate 0x0FAEF0 using the definition.

I think your going to have to just use tcg_out_fence() and the
tcg_out_opc() does black magic with extra flag bits and assumes it is
encoding rx registers (at least in 64 bit mode).

However I would suggest a comment and maybe breakdown of the different
fence types you can emit. Will pre-P4 processors never need mfences?

>
>>
>> Also, for 32-bit you need to check for sse2 before outputting this.  See
>> also the existing cpuid checks in tcg_target_init and the fallback smp_mb
>> definition for pre-gcc-4.4.
>
> OK, I'll check the current code and do something similar.
>
> Thanks,


--
Alex Bennée