Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <1559116604-23105-1-git-send-email-zhenzhong.duan@oracle.com>
In-Reply-To: <1559116604-23105-1-git-send-email-zhenzhong.duan@oracle.com>
From:   Andy Lutomirski <luto@kernel.org>
Date:   Thu, 30 May 2019 07:15:02 -0700
Message-ID: <CALCETrU96Pjw5AEy_Aju_hMkv=QdE3YVfx5aY24B8WwDqM1A9Q@mail.gmail.com>
Subject: Re: [PATCH] x86/mm/tlb: Do partial TLB flush when possible
To:     Zhenzhong Duan <zhenzhong.duan@oracle.com>
Cc:     LKML <linux-kernel@vger.kernel.org>, srinivas.eeda@oracle.com,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Andy Lutomirski <luto@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        X86 ML <x86@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, May 30, 2019 at 12:56 AM Zhenzhong Duan
<zhenzhong.duan@oracle.com> wrote:
>
> This is a small optimization to stale TLB flush, if there is one new TLB
> flush, let it choose to do partial or full flush. or else, the stale
> flush take over and do full flush.

I think this is invalid because:

>
> +       if (unlikely(f->new_tlb_gen <= local_tlb_gen &&
> +           local_tlb_gen + 1 == mm_tlb_gen)) {
> +               /*
> +                * For stale TLB flush request, if there will be one new TLB
> +                * flush coming, we leave the work to the new IPI as it knows
> +                * partial or full TLB flush to take, or else we do the full
> +                * flush.
> +                */
> +               trace_tlb_flush(reason, 0);
> +               return;

We do indeed know that the TLB will get flushed eventually, but we're
actually providing a stronger guarantee that the TLB will be
adequately flushed by the time we return.  Otherwise, after
flush_tlb_mm_range(), there will be a window in which the TLB isn't
flushed yet.