Re: [PATCH RFC 4/4] x86/entry/pti: don't switch PGD on tasks holding flag TIF_NOPTI

[Andy Lutomirski]

On 01/08/2018 08:12 AM, Willy Tarreau wrote:
> If a task has the TIF_NOPTI flag set, it doesn't want to experience
> page table isolation. In this case, returns from kernel to user will
> not switch the CR3, leaving it to the kernel one which already maps
> both user and kernel pages. Upon entry in the kernel, we can't check
> this flag so we simply check if CR3 was pointing to the kernel's PGD,
> indicating an earlier absence of switch, and in this case we don't
> change it.
>
> Thanks to these changes, haproxy running under KVM went back from
> 12400 conn/s to 21000 once loaded after calling prctl().
>
> Signed-off-by: Willy Tarreau <w@xxxxxx>
> ---
>   arch/x86/entry/calling.h | 23 +++++++++++++++++++++++
>   1 file changed, 23 insertions(+)
>
> diff --git a/arch/x86/entry/calling.h b/arch/x86/entry/calling.h
> index 45a63e0..054b8b7 100644
> --- a/arch/x86/entry/calling.h
> +++ b/arch/x86/entry/calling.h
> @@ -1,5 +1,6 @@
>   /* SPDX-License-Identifier: GPL-2.0 */
>   #include <linux/jump_label.h>
> +#include <asm/thread_info.h>
>   #include <asm/unwind_hints.h>
>   #include <asm/cpufeatures.h>
>   #include <asm/page_types.h>
> @@ -214,6 +215,11 @@
>   .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
>   	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
>   	mov	%cr3, \scratch_reg
> +
> +	/* if we're already on the kernel PGD, we don't switch */
> +	testq $(PTI_SWITCH_PGTABLES_MASK), \scratch_reg
> +	jz .Lend_\@
> +
>   	ADJUST_KERNEL_CR3 \scratch_reg
>   	mov	\scratch_reg, %cr3
>   .Lend_\@:
> @@ -224,6 +230,12 @@
>   
>   .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req
>   	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
> +
> +	/* "NOPTI" taskflag avoids the switch */
> +	movq	PER_CPU_VAR(current_task), \scratch_reg
> +	btq	$TIF_NOPTI, TASK_TI_flags(\scratch_reg)
> +	jc	.Lend_\@
> +
>   	mov	%cr3, \scratch_reg
>   
>   	ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
> @@ -262,6 +274,13 @@
>   	ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI
>   	movq	%cr3, \scratch_reg
>   	movq	\scratch_reg, \save_reg
> +
> +	/* if we're already on the kernel PGD, we don't switch,
> +	 * we just save the current cr3.
> +	 */
> +	testq $(PTI_SWITCH_PGTABLES_MASK), \scratch_reg
> +	jz .Ldone_\@
> +
>   	/*
>   	 * Is the "switch mask" all zero?  That means that both of
>   	 * these are zero:
> @@ -284,6 +303,10 @@
>   .macro RESTORE_CR3 scratch_reg:req save_reg:req
>   	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
>   
> +	/* if we saved a kernel context, we didn't switch so we don't switch */
> +	testq $(PTI_SWITCH_PGTABLES_MASK), \save_reg
> +	jz .Lend_\@
> +

I haven't looked that carefully, but this seems generally sane.