Re: [PATCH v2 03/12] x86/retpoline/entry: Convert entry assembler indirect jumps

[David Woodhouse]

On Wed, 2018-01-03 at 18:00 -0800, Andi Kleen wrote:
> @@ -269,8 +270,9 @@ entry_SYSCALL_64_fastpath:
> ÂÂÂÂÂÂÂÂ * This call instruction is handled specially in stub_ptregs_64.
>  * It might end up jumping to the slow path. If it jumps, RAX
> ÂÂÂÂÂÂÂÂ * and all argument registers are clobbered.
> -ÂÂÂÂÂÂÂ */
> -ÂÂÂÂÂÂÂcallÂÂÂÂ*sys_call_table(, %rax, 8)
> +ÂÂÂÂÂÂÂ*/
> +ÂÂÂÂÂÂÂmovqÂÂÂÂsys_call_table(, %rax, 8), %rax
> +ÂÂÂÂÂÂÂNOSPEC_CALLÂÂÂÂÂ%rax
> Â.Lentry_SYSCALL_64_after_fastpath_call:
> Â
> ÂÂÂÂÂÂÂÂmovqÂÂÂÂ%rax, RAX(%rsp)

Now I *know* you're working from an older version of my patches and
haven't just deliberately reverted the CET support (and alternatives?).

I fixed that mis-indentation of the closing */ and Intel were even
shipping that version in the latest pre-embargo patch tarball.

Should I post a new series and call it 'v3'? Was there anything
specific you changed other than the cosmetics like
s/CALL_THUNK/NOSPEC_CALL/ ? And obviously the fact that it stands alone
instead of being based on the IBRS microcode support?

Attachment: smime.p7s
Description: S/MIME cryptographic signature