linux-kernel - [PATCH] tools/memory-model: Clarify syntactic and semantic dependencies

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 25 Jan 2022 17:28:19 +0000
From:   Paul Heidekrüger <paul.heidekrueger@...tum.de>
To:     Alan Stern <stern@...land.harvard.edu>,
        Andrea Parri <parri.andrea@...il.com>,
        Will Deacon <will@...nel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Boqun Feng <boqun.feng@...il.com>,
        Nicholas Piggin <npiggin@...il.com>,
        David Howells <dhowells@...hat.com>,
        Jade Alglave <j.alglave@....ac.uk>,
        Luc Maranget <luc.maranget@...ia.fr>,
        "Paul E. McKenney" <paulmck@...nel.org>,
        Akira Yokosawa <akiyks@...il.com>,
        Daniel Lustig <dlustig@...dia.com>,
        Joel Fernandes <joel@...lfernandes.org>,
        Paul Heidekrüger <paul.heidekrueger@...tum.de>,
        Björn Töpel <bjorn@...nel.org>,
        linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org
Cc:     Marco Elver <elver@...gle.com>,
        Charalampos Mainas <charalampos.mainas@...il.com>,
        Pramod Bhatotia <pramod.bhatotia@...tum.de>
Subject: [PATCH] tools/memory-model: Clarify syntactic and semantic dependencies

Dependencies which are purely syntactic, i.e. not semantic, might imply
ordering at first glance. However, since they do not affect defined
behavior, compilers are within their rights to remove such dependencies
when optimizing code.

Since syntactic dependencies are not related to any kind of dependency
in particular, explicitly distinguish syntactic and semantic
dependencies as part of the 'A WARNING' section in explanation.txt,
which gives examples of how compilers might affect the LKMM's dependency
orderings in general.

Link: https://lore.kernel.org/all/20211102190138.GA1497378@rowland.harvard.edu/
Signed-off-by: Paul Heidekrüger <paul.heidekrueger@...tum.de>
Cc: Marco Elver <elver@...gle.com>
Cc: Charalampos Mainas <charalampos.mainas@...il.com>
Cc: Pramod Bhatotia <pramod.bhatotia@...tum.de>
---
 .../Documentation/explanation.txt             | 25 +++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/tools/memory-model/Documentation/explanation.txt b/tools/memory-model/Documentation/explanation.txt
index 5d72f3112e56..6d679e5ebdf9 100644
--- a/tools/memory-model/Documentation/explanation.txt
+++ b/tools/memory-model/Documentation/explanation.txt
@@ -411,6 +411,31 @@ Given this version of the code, the LKMM would predict that the load
 from x could be executed after the store to y.  Thus, the memory
 model's original prediction could be invalidated by the compiler.
 
+Caution is also advised when dependencies are purely syntactic, i.e.
+not semantic.  A dependency between two marked accesses is purely
+syntactic iff the defined behavior of the second access is unaffected
+by its dependency.
+
+Compilers are aware of syntactic dependencies and are within their
+rights to remove them as part of optimizations, thereby breaking any
+guarantees of ordering.
+
+Notable cases are dependencies eliminated through constant propagation
+or those where only one value leads to defined behavior as in the
+following example:
+
+	int a[1];
+	int i;
+
+	r1 = READ_ONCE(i);
+	r2 = READ_ONCE(a[r1]);
+
+The formal LKMM is unaware of syntactic dependencies and therefore
+predicts ordering.  However, since any other value than 0 for r1 would
+result in an out-of-bounds access, which is undefined behavior, r2 is
+not affected by its dependency to r1, making the above a purely
+syntactic dependency.
+
 Another issue arises from the fact that in C, arguments to many
 operators and function calls can be evaluated in any order.  For
 example:
-- 
2.33.1