Re: [RFC 04/10] x86/mm: Only flush indirect branches when switching into non dumpable process

["Woodhouse, David" <dwmw@amazon.co.uk>]

[-- Attachment #1: Type: text/plain, Size: 2300 bytes --]

On Sat, 2018-01-20 at 20:22 +0100, KarimAllah Ahmed wrote:
> From: Tim Chen <tim.c.chen@linux.intel.com>

I think this is probably From: Andi now rather than From: Tim?

We do need the series this far in order to have a full retpoline-based
mitigation, and I'd like to see that go in sooner rather than later.
There's a little more discussion to be had about the IBRS parts which
come later in the series (and the final one or two which weren't posted
yet).

I think this is the one patch of the "we want this now" IBPB set that
we expect serious debate on, which is why it's still a separate
"optimisation" patch on top of the previous one which just does IBPB
unconditionally.


> Flush indirect branches when switching into a process that marked
> itself non dumpable.  This protects high value processes like gpg
> better, without having too high performance overhead.
> 
> Signed-off-by: Andi Kleen <ak@linux.intel.com>
> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> Signed-off-by: KarimAllah Ahmed <karahmed@amazon.de>
> ---
>  arch/x86/mm/tlb.c | 13 ++++++++++++-
>  1 file changed, 12 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
> index 304de7d..f64e80c 100644
> --- a/arch/x86/mm/tlb.c
> +++ b/arch/x86/mm/tlb.c
> @@ -225,8 +225,19 @@ void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
>  		 * Avoid user/user BTB poisoning by flushing the branch predictor
>  		 * when switching between processes. This stops one process from
>  		 * doing Spectre-v2 attacks on another.
> +		 *
> +		 * As an optimization: Flush indirect branches only when
> +		 * switching into processes that disable dumping.
> +		 *
> +		 * This will not flush when switching into kernel threads.
> +		 * But it would flush when switching into idle and back
> +		 *
> +		 * It might be useful to have a one-off cache here
> +		 * to also not flush the idle case, but we would need some
> +		 * kind of stable sequence number to remember the previous mm.
>  		 */
> -		indirect_branch_prediction_barrier();
> +		if (tsk && tsk->mm && get_dumpable(tsk->mm) != SUID_DUMP_USER)
> +			indirect_branch_prediction_barrier();
>  
>  		if (IS_ENABLED(CONFIG_VMAP_STACK)) {
>  			/*

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5210 bytes --]