linux-kernel - Re: [PATCH 2/4] mm/tlb: Remove tlb_remove

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 23 Aug 2018 13:31:03 +1000
From:   Nicholas Piggin <npiggin@...il.com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     torvalds@...ux-foundation.org, luto@...nel.org, x86@...nel.org,
        bp@...en8.de, will.deacon@....com, riel@...riel.com,
        jannh@...gle.com, ascannell@...gle.com, dave.hansen@...el.com,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        David Miller <davem@...emloft.net>,
        Martin Schwidefsky <schwidefsky@...ibm.com>,
        Michael Ellerman <mpe@...erman.id.au>
Subject: Re: [PATCH 2/4] mm/tlb: Remove tlb_remove_table() non-concurrent
 condition

On Wed, 22 Aug 2018 17:30:14 +0200
Peter Zijlstra <peterz@...radead.org> wrote:

> Will noted that only checking mm_users is incorrect; we should also
> check mm_count in order to cover CPUs that have a lazy reference to
> this mm (and could do speculative TLB operations).

Why is that incorrect?

This shortcut has nothing to do with no TLBs -- not sure about x86, but
other CPUs can certainly have remaining TLBs here, speculative
operations or not (even if they don't have an mm_count ref they can
have TLBs here).

So that leaves speculative operations. I don't see where the problem is
with those either -- this shortcut needs to ensure there are no other
*non speculative* operations. mm_users is correct for that.

If there is a speculation security problem here it should be carefully
documented otherwise it's going to be re-introduced...

I actually have a patch to extend this optimisation further that I'm
going to send out again today. It's nice to avoid the double handling
of the pages.

Thanks,
Nick

> 
> If removing this turns out to be a performance issue, we can
> re-instate a more complete check, but in tlb_table_flush() eliding the
> call_rcu_sched().
> 
> Cc: stable@...nel.org
> Cc: Nicholas Piggin <npiggin@...il.com>
> Cc: David Miller <davem@...emloft.net>
> Cc: Will Deacon <will.deacon@....com>
> Cc: Martin Schwidefsky <schwidefsky@...ibm.com>
> Cc: Michael Ellerman <mpe@...erman.id.au>
> Fixes: 267239116987 ("mm, powerpc: move the RCU page-table freeing into generic code")
> Reported-by: Will Deacon <will.deacon@....com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
> ---
>  mm/memory.c |    9 ---------
>  1 file changed, 9 deletions(-)
> 
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -375,15 +375,6 @@ void tlb_remove_table(struct mmu_gather
>  {
>  	struct mmu_table_batch **batch = &tlb->batch;
>  
> -	/*
> -	 * When there's less then two users of this mm there cannot be a
> -	 * concurrent page-table walk.
> -	 */
> -	if (atomic_read(&tlb->mm->mm_users) < 2) {
> -		__tlb_remove_table(table);
> -		return;
> -	}
> -
>  	if (*batch == NULL) {
>  		*batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
>  		if (*batch == NULL) {
> 
>