Re: [RFC PATCH] x86/mm: Flush more aggressively in lazy TLB mode

From: Borislav Petkov <bp@alien8.de>
To: Andy Lutomirski <luto@kernel.org>
Cc: X86 ML <x86@kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Markus Trippelsdorf <markus@trippelsdorf.de>,
	Adam Borowski <kilobyte@angband.pl>,
	Brian Gerst <brgerst@gmail.com>,
	Johannes Hirte <johannes.hirte@datenkhaos.de>
Subject: Re: [RFC PATCH] x86/mm: Flush more aggressively in lazy TLB mode
Date: Mon, 9 Oct 2017 20:08:10 +0200	[thread overview]
Message-ID: <20171009180810.zkcugzxicgpsegh3@pd.tnic> (raw)
In-Reply-To: <CALCETrXY+fKsqE9ZKSJaFuw0EUh9zmt342VdG6fxDjx96LswWw@mail.gmail.com>

On Mon, Oct 09, 2017 at 10:50:34AM -0700, Andy Lutomirski wrote:
> The choices are somewhat lazy and not lazy at all.

Yeah, you probably should explain those choices somewhere and what
exactly they mean.

> The degree of simplification I would get by removing it is basically
> nil.  The debugfs code itself goes away, and a
> static_branch_unlikely() turns into a static_cpu_has(), and that's it.

Sure. But it is one variable less which is not really needed by the
widest audience.

> The real reason I added it is because Chris Mason volunteered to
> benchmark it, and I'll send it to him once it survives a bit of
> review.

Sure but it still doesn't need to be upstream. You can do all the
measurements with a patch ontop. You don't need the permanent knob in
debugfs either. After a year, no one would really need that anymore,
since the majority will be PCID machines.

> This is non-lazy.  It's roughtly what our state was in old kernels
> when we went lazy and then called leave_mm().

non-lazy when we went lazy?!

Now I'm confused :)

-- 
Regards/Gruss,
    Boris.

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