On Mon, Oct 04, 2010 at 10:43:07AM -0700, David Daney wrote:
> On 10/03/2010 11:15 AM, Rabin Vincent wrote:
>> On Tue, Sep 28, 2010 at 11:40 PM, David Daney<ddaney@xxxxxxxxxxxxxxxxxx> wrote:
>>> +void arch_jump_label_text_poke_early(jump_label_t addr)
>>> +{
>>> + union mips_instruction *insn_p =
>>> + (union mips_instruction *)(unsigned long)addr;
>>> +
>>> + insn_p->word = 0; /* nop */
>>> + flush_icache_range((unsigned long)insn_p,
>>> + (unsigned long)insn_p + sizeof(*insn_p));
>>> +}
>>
>> Can't this function be a no-op on MIPS? This seems to be
>> used on x86 to patch in the optimal nop instruction, but
>> on MIPS the optimal/only nop instruction should already
>> be in place at build time. Same thing for the SPARC
>> implementation.
>>
>
> Yes, I think you are correct.
>
> On MIPS the NOP is already optimal. I will respin the MIPS patch to
> make arch_jump_label_text_poke_early() be empty.
>
> davem wasn't CCed on the original message, so I added him. I would
> defer to him on the SPARC version.
>
>
that's right, arch_jump_label_text_poke_early() can probably be a no-op
for most arches.
We can also look at adding an empty definition into the generic
header. So that arches don't have to provide an empty definition.
thanks,
-Jason
