On Thu, 23 Oct 2025 09:41:29 GMT, Andrew Dinn <[email protected]> wrote:

> > I meant to refer to this check implementation, which reads 8 bytes from 
> > where LR points to and checks whether they contain the NOP+MOVK sequence.
> 
> Ah, my apologies for misunderstanding you and also and what was going on in 
> this code. It is indeed reading two instructions at once and checking them 
> both in one go.
> 
> I agree with you that this is best split up into two separate reads+checks to 
> deal with the case where we don't have a NOP and might drop off the end of 
> the current buffer. I think this will still be less work than using SafeFetch 
> but @theRealAph may want to override me here.

I think I do want to. We shouldn't leave such fragile code in once we've 
noticed it. IMO it's a false economy to avoid SafeFetch on efficiency grounds. 
If needs be, there are ways to make it faster.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/26678#issuecomment-3466810293

Reply via email to