Re: [PATCH RFC V2 3/5] jump_label: if a key has already been initialized, don't nop it out

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 10/05/2011 05:17 PM, H. Peter Anvin wrote:
> On 10/05/2011 05:16 PM, Jeremy Fitzhardinge wrote:
>> On 10/04/2011 09:30 AM, H. Peter Anvin wrote:
>>> On 10/04/2011 07:10 AM, Jason Baron wrote:
>>>> 1) The jmp +0, is a 'safe' no-op that I know is going to initially
>>>> boot for all x86. I'm not sure if there is a 5-byte nop that works on
>>>> all x86 variants - but by using jmp +0, we make it much easier to debug
>>>> cases where we may be using broken no-ops.
>>>>
>>> There are *plenty*.  jmp+0 is about as pessimal as you can get.
>> As an aside, do you know if a 2-byte unconditional jmp is any more
>> efficient than 5-byte, aside from just being a smaller instruction and
>> taking less icache?
>>
> I don't know for sure, no.  I probably depends on the CPU.

I was thinking about making the jump-label stuff generate a small jmp if
the offset is small (specifically "jmp; nop3", or perhaps "jmp; ud2a;
nop" to make absolutely sure there's no speculation beyond the jmp), on
the grounds that, while it probably doesn't matter for any modern
Intel/AMD processor, it may help for others.  But I couldn't find any
concrete evidence to support it, and there's already enough questions
about doing "live" code updates without adding more instruction patterns.

    J