Return-Path: Received: from mail.linutronix.de (146.0.238.70:993) by crypto-ml.lab.linutronix.de with IMAP4-SSL for ; 05 Feb 2019 18:30:25 -0000 Received: from merlin.infradead.org ([2001:8b0:10b:1231::1]) by Galois.linutronix.de with esmtps (TLS1.2:RSA_AES_256_CBC_SHA256:256) (Exim 4.80) (envelope-from ) id 1gr5UG-0005Lu-C9 for speck@linutronix.de; Tue, 05 Feb 2019 19:30:24 +0100 Received: from j217100.upc-j.chello.nl ([24.132.217.100] helo=hirez.programming.kicks-ass.net) by merlin.infradead.org with esmtpsa (Exim 4.90_1 #2 (Red Hat Linux)) id 1gr5UF-0005qs-4P for speck@linutronix.de; Tue, 05 Feb 2019 18:30:23 +0000 Date: Tue, 5 Feb 2019 19:30:20 +0100 From: Peter Zijlstra Subject: [MODERATED] Re: [PATCH v1 9/9] PERFv1 3 Message-ID: <20190205183020.GJ17550@hirez.programming.kicks-ass.net> References: <20190205152804.GY17528@hirez.programming.kicks-ass.net> <20190205164829.GI31598@tassilo.jf.intel.com> <20190205171326.GI17550@hirez.programming.kicks-ass.net> <20190205172613.GK31598@tassilo.jf.intel.com> MIME-Version: 1.0 In-Reply-To: <20190205172613.GK31598@tassilo.jf.intel.com> Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit To: speck@linutronix.de List-ID: On Tue, Feb 05, 2019 at 09:26:13AM -0800, speck for Andi Kleen wrote: > > We should probably have a notsx boot option that completely disables the > > feature including its userspace enumeration. Some security consious > > folks might actually like that. TSX has been a boon to all the fault > > driven exploit muck. > > TSX generally only messes with exploit detection approaches, which > are IMHO all snake oil anyways because they can be always circumvented > in other ways. If you dig into it more deeply there's no clear security case > for actually disabling TSX. In some cases it can make exploit > easier to write, but that alone isn't a good reason to disable it. Making exploits harder is enough for some. > > And userspace should be able to do this; we killed TSX on some early > > models before, right? > > On parts of the models without many cores. The large core count parts > always had it. Doesn't matter some always had it; userspace now knows how to handle no TSX. > Functionally yes, but there are workloads that rely on it for performance, > which would be effectively broken too. Which is an choice the admin should be able to make.