Re: [RFC PATCH 4/4] x86/TSC: Use RDTSCP

From: Borislav Petkov <bp@alien8.de>
To: Andy Lutomirski <luto@kernel.org>
Cc: Tom Lendacky <Thomas.Lendacky@amd.com>,
	LKML <linux-kernel@vger.kernel.org>, X86 ML <x86@kernel.org>,
	"H. Peter Anvin" <hpa@zytor.com>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	John Stultz <john.stultz@linaro.org>
Subject: Re: [RFC PATCH 4/4] x86/TSC: Use RDTSCP
Date: Wed, 12 Dec 2018 11:08:14 +0100	[thread overview]
Message-ID: <20181212100814.GB6653@zn.tnic> (raw)
In-Reply-To: <CALCETrXvZ4avdUh2bTLVTTz3Jxogo3zK5akWTsJ=rkJpf1XiDg@mail.gmail.com>

On Tue, Dec 11, 2018 at 06:24:44PM -0800, Andy Lutomirski wrote:
> This makes me nervous, since no one knows what “serializing” means.

Why no one? If you wanna say that X86_FEATURE_LFENCE_SERIALIZING is not
really telling, so is X86_FEATURE_LFENCE_RDTSC, TBH. :)

> IIRC AMD specifically documents that MFENCE is required before RDTSC
> to get sensible ordering.  So it’s entirely plausible to me that
> LFENCE is okay for Spectre mitigation but MFENCE is needed for RDTSC
> on some CPU.

Look at init_amd(), the if (cpu_has(c, X86_FEATURE_XMM2)) branch where
we make LFENCE serializing. The logic with the new names would be:

        asm volatile(ALTERNATIVE_3("rdtsc",
                                   "mfence; rdtsc", X86_FEATURE_MFENCE,
                                   "lfence; rdtsc", X86_FEATURE_LFENCE_SERIALIZING,
                                   "rdtscp", X86_FEATURE_RDTSCP)

RDTSC is put there during build. At boot time:

	if CPU has MFENCE
		use MFENCE to stop RDTSC speculation
	if LFENCE is serializing
		use LFENCE...
	if CPU has RDTSCP
		even better, use that as it is the fastest or on par.

Of course the order of those is important.

Ok?

-- 
Regards/Gruss,
    Boris.

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