Re: [Qemu-devel] [PATCH v3 07/16] tcg-ppc64: DefineTCG_TARGET_INSN_UNIT_SIZEE

[Richard Henderson]

On 04/29/2014 08:41 AM, Alex Bennée wrote:
> 
> Richard Henderson <address@hidden> writes:
> 
>> On 04/29/2014 04:25 AM, Alex Bennée wrote:
>>>> +static void reloc_pc24(tcg_insn_unit *pc, tcg_insn_unit *target)
>>>>  {
>>>> -    *(uint32_t *)pc = (*(uint32_t *)pc & ~0x3fffffc)
>>>> -        | reloc_pc24_val(pc, target);
>>>> +    *pc = (*pc & ~0x3fffffc) | reloc_pc24_val(pc, target);
>>>>  }
>>>
>>> Should 0x3fffffc be a #define'd mask? Can the PC ever actually be
>>> non-word aligned?
>>
>> No, it can't be unaligned.  The 0x3fffffc is really the field into
>> which the offset is being placed.  I don't think a define is helpful
>> really, since this is the only place it's used.
> 
> Ahh ok that makes sense. I was thrown because the masking pattern seems
> to occur all around the ppc tcg code:
> 
> 16:40 address@hidden/x86_64 [qemu.git] >git grep "0x3fffffc"
> disas/ppc.c:  { 0x3fffffc, 0, NULL, NULL, PPC_OPERAND_RELATIVE | 
> PPC_OPERAND_SIGNED },
> disas/ppc.c:  { 0x3fffffc, 0, NULL, NULL, PPC_OPERAND_ABSOLUTE | 
> PPC_OPERAND_SIGNED },
> tcg/ppc/tcg-target.c:    return disp & 0x3fffffc;
> tcg/ppc/tcg-target.c:    *(uint32_t *) pc = (*(uint32_t *) pc & ~0x3fffffc)
> tcg/ppc/tcg-target.c:        tcg_out32 (s, B | (disp & 0x3fffffc) | mask);
> tcg/ppc/tcg-target.c:                tcg_out32 (s, B | (val & 0x3fffffc));
> tcg/ppc64/tcg-target.c:    return disp & 0x3fffffc;
> tcg/ppc64/tcg-target.c:    *(uint32_t *)pc = (*(uint32_t *)pc & ~0x3fffffc)
> tcg/ppc64/tcg-target.c:    unsigned retrans = *(uint32_t *)s->code_ptr & 
> 0x3fffffc;
> tcg/ppc64/tcg-target.c:        tcg_out32(s, B | (disp & 0x3fffffc) | mask);

Well, its true that it's going to be replicated between the disassembler and
the two ppc backends.  I'm slightly surprised that it appears more than twice
for each ppc backend, but I suppose that just means there's more room to tidy 
up.

I do think that's out of scope for this patch set though.


r~