605010 - JM: Add detection for multibyte NOP

Status: RESOLVED WONTFIX

(Core :: JavaScript Engine, defect)

Description

[Alex Miller]

User-Agent:       Mozilla/5.0 (X11; U; Linux x86_64; en-US) AppleWebKit/533.4 (KHTML, like Gecko) Chrome/5.0.375.127 Safari/533.4
Build Identifier: 

Multibyte NOP's are not supported on every CPU, so we need to query CPUID to see if they can be used.

Reproducible: Always

Comment 1

[Alex Miller]

Attachment: patch v1

Shifted the actual CPUID call out to a function shared between the multibyte NOP detection code and the SSE detection code and added multibyte NOP detection modeled after the existing SSE detection.

Comment 2

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

Are we seeing crashes due to this?  Seems like we could be, and an easy win for stability for those users. b8 fodder?

Comment 3

[David Mandelin [:dmandelin]]

(In reply to comment #2)
> Are we seeing crashes due to this?  Seems like we could be, and an easy win for
> stability for those users. b8 fodder?

I don't think we have a way of knowing that. It's a good prospective stability fix, though.

Comment 4

[David Mandelin [:dmandelin]]

(In reply to comment #0)
> Multibyte NOP's are not supported on every CPU, so we need to query CPUID to
> see if they can be used.

Where do the multibyte nops come into play? dvander says JM doesn't generate any.

Comment 5

[Alex Miller]

This patch was created at sstangl's request. I believe he was using it to do testing to see if forcing alignment of jump or call destinations would increase performance. I'm not sure what the status of his work is.

To directly address the questions above, this patch is not meant in any way to address crashes or stability of existing code.

Comment 6

[Sean Stangl [:sstangl]]

There are no crashes due to this patch (or any related concept). Multibyte NOPs are useful for alignment purposes. This patch is support code for another patch, yet to be added to bugzilla, that uses multibyte NOPs to align jump targets to pointer-size.

Comment 7

[Ryan VanderMeulen [:RyanVM]]

Sean, is this still wanted?

Comment 8

[Sean Stangl [:sstangl]]

No, it doesn't look like we'll need multi-byte NOPs in the future.