[patch 00/24] x86/speculation: Remedy the STIBP/IBPB overhead

From: Thomas Gleixner <tglx@linutronix.de>
To: LKML <linux-kernel@vger.kernel.org>
Cc: x86@kernel.org, Peter Zijlstra <peterz@infradead.org>,
	Andy Lutomirski <luto@kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Jiri Kosina <jkosina@suse.cz>,
	Tom Lendacky <thomas.lendacky@amd.com>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	Andrea Arcangeli <aarcange@redhat.com>,
	David Woodhouse <dwmw@amazon.co.uk>,
	Andi Kleen <ak@linux.intel.com>,
	Dave Hansen <dave.hansen@intel.com>,
	Casey Schaufler <casey.schaufler@intel.com>,
	Asit Mallick <asit.k.mallick@intel.com>,
	Arjan van de Ven <arjan@linux.intel.com>,
	Jon Masters <jcm@redhat.com>, Waiman Long <longman9394@gmail.com>,
	Greg KH <gregkh@linuxfoundation.org>,
	Dave Stewart <david.c.stewart@intel.com>,
	Kees Cook <keescook@chromium.org>
Subject: [patch 00/24] x86/speculation: Remedy the STIBP/IBPB overhead
Date: Wed, 21 Nov 2018 21:14:30 +0100
Message-ID: <20181121201430.559770965@linutronix.de> (raw)

This is based on Tim Chen's V5 patch series. The following changes have
been made:

   - Control STIPB evaluation with a single static key

   - Move IBPB out from switch_mm() into switch_to() and control the
     always and the conditional mode with static keys.

     The mainline implementation is wrong in a few aspects, e.g. it fails
     to protect tasks within the same process, which breaks
     sandboxing. That same process optimization was the sole reason to have
     it in switch_mm().

     The new always mode is just issuing the barrier unconditionally when
     switching to a user task, but that also leaves STIPB always on. So
     really paranoid people get the highest possible protection and the
     highest overhead.

     The conditional mode issues the barrier when a task which is mitigated
     is scheduling out or scheduling in. That is required to support proper
     sandboxing.

   - Remove the ptrace_may_access_sched() code as it's unused now. It was
     ugly anyway and would have given people ideas how to slow down
     switch_mm() some more.

   - Rename TIF_STIPB to TIF_SPEC_IB because it controls both STIBP and
     IBPB.

   - Fix all the corner cases vs. UP and SMT disabled.

   - Limit the overhead when conditional STIPB is not enabled so
     switch_to_xtra() is not invoked for nothing when the TIF bit would
     trigger the entry and nothing else is to do. That can happen when SMT
     is off and a task has the TIF bit set. On UP STIPB is never enabled.

   - Dropped the dumpable part

TODO: Write documentation

It's avaiable from git:

   git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git WIP.x86/pti

It's based on the x86/pti branch unfortunately, which contains the removal
of the minimal asm retpoline hackery. I noticed too late. If the minimal
asm stuff should not be backported it's trivial to rebase that series on
Linus tree.

Thanks,

	tglx