Re: [tip:x86/pti] x86/retpoline: Fill return stack buffer on vmexit

[Borislav Petkov]

On Fri, Jan 12, 2018 at 03:37:49AM -0800, tip-bot for David Woodhouse wrote:
> Commit-ID:  117cc7a908c83697b0b737d15ae1eb5943afe35b
> Gitweb:     https://git.kernel.org/tip/117cc7a908c83697b0b737d15ae1eb5943afe35b
> Author:     David Woodhouse <dwmw@xxxxxxxxxxxx>
> AuthorDate: Fri, 12 Jan 2018 11:11:27 +0000
> Committer:  Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> CommitDate: Fri, 12 Jan 2018 12:33:37 +0100
> 
> x86/retpoline: Fill return stack buffer on vmexit

...

> +/*
> + * Google experimented with loop-unrolling and this turned out to be
> + * the optimal version â two calls, each with their own speculation
> + * trap should their return address end up getting used, in a loop.
> + */
> +#define __FILL_RETURN_BUFFER(reg, nr, sp)	\
> +	mov	$(nr/2), reg;			\
> +771:						\
> +	call	772f;				\
> +773:	/* speculation trap */			\
> +	pause;					\
> +	jmp	773b;				\
> +772:						\
> +	call	774f;				\
> +775:	/* speculation trap */			\
> +	pause;					\
> +	jmp	775b;				\
> +774:						\
> +	dec	reg;				\
> +	jnz	771b;				\
> +	add	$(BITS_PER_LONG/8) * nr, sp;
> +

Btw, just a minor nit: one could finish the lines with \n\t - the lines
preceding labels only with \n - so that the asm output looks readable.

Right now it is a single line of instructions. But I can fix that later.

---
--- a/arch/x86/include/asm/nospec-branch.h
+++ b/arch/x86/include/asm/nospec-branch.h
@@ -33,23 +33,23 @@
  * trap should their return address end up getting used, in a loop.
  */
 #define __FILL_RETURN_BUFFER(reg, nr, sp)	\
-	mov	$(nr/2), reg;			\
-771:						\
-	call	772f;				\
-773:	/* speculation trap */			\
-	pause;					\
-	lfence;					\
-	jmp	773b;				\
-772:						\
-	call	774f;				\
-775:	/* speculation trap */			\
-	pause;					\
-	lfence;					\
-	jmp	775b;				\
-774:						\
-	dec	reg;				\
-	jnz	771b;				\
-	add	$(BITS_PER_LONG/8) * nr, sp;
+	mov	$(nr/2), reg\n			\
+771:\n\t					\
+	call	772f\n				\
+773:/* speculation trap */\n\t			\
+	pause\n\t				\
+	lfence\n\t				\
+	jmp	773b\n				\
+772:\n\t					\
+	call	774f\n				\
+775:	/* speculation trap */\n\t		\
+	pause\n\t				\
+	lfence\n\t				\
+	jmp	775b\n				\
+774:\n\t					\
+	dec	reg\n\t				\
+	jnz	771b\n\t			\
+	add	$(BITS_PER_LONG/8) * nr, sp\n
 
 #ifdef __ASSEMBLY__

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.