Documentation: Document array_index_nospec
authorMark Rutland <mark.rutland@arm.com>
Tue, 30 Jan 2018 01:02:16 +0000 (17:02 -0800)
committerBen Hutchings <ben@decadent.org.uk>
Mon, 19 Mar 2018 18:58:36 +0000 (18:58 +0000)
commit f84a56f73dddaeac1dba8045b007f742f61cd2da upstream.

Document the rationale and usage of the new array_index_nospec() helper.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: linux-arch@vger.kernel.org
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: gregkh@linuxfoundation.org
Cc: kernel-hardening@lists.openwall.com
Cc: torvalds@linux-foundation.org
Cc: alan@linux.intel.com
Link: https://lkml.kernel.org/r/151727413645.33451.15878817161436755393.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Documentation/speculation.txt [new file with mode: 0644]

diff --git a/Documentation/speculation.txt b/Documentation/speculation.txt
new file mode 100644 (file)
index 0000000..e9e6cba
--- /dev/null
@@ -0,0 +1,90 @@
+This document explains potential effects of speculation, and how undesirable
+effects can be mitigated portably using common APIs.
+
+===========
+Speculation
+===========
+
+To improve performance and minimize average latencies, many contemporary CPUs
+employ speculative execution techniques such as branch prediction, performing
+work which may be discarded at a later stage.
+
+Typically speculative execution cannot be observed from architectural state,
+such as the contents of registers. However, in some cases it is possible to
+observe its impact on microarchitectural state, such as the presence or
+absence of data in caches. Such state may form side-channels which can be
+observed to extract secret information.
+
+For example, in the presence of branch prediction, it is possible for bounds
+checks to be ignored by code which is speculatively executed. Consider the
+following code:
+
+       int load_array(int *array, unsigned int index)
+       {
+               if (index >= MAX_ARRAY_ELEMS)
+                       return 0;
+               else
+                       return array[index];
+       }
+
+Which, on arm64, may be compiled to an assembly sequence such as:
+
+       CMP     <index>, #MAX_ARRAY_ELEMS
+       B.LT    less
+       MOV     <returnval>, #0
+       RET
+  less:
+       LDR     <returnval>, [<array>, <index>]
+       RET
+
+It is possible that a CPU mis-predicts the conditional branch, and
+speculatively loads array[index], even if index >= MAX_ARRAY_ELEMS. This
+value will subsequently be discarded, but the speculated load may affect
+microarchitectural state which can be subsequently measured.
+
+More complex sequences involving multiple dependent memory accesses may
+result in sensitive information being leaked. Consider the following
+code, building on the prior example:
+
+       int load_dependent_arrays(int *arr1, int *arr2, int index)
+       {
+               int val1, val2,
+
+               val1 = load_array(arr1, index);
+               val2 = load_array(arr2, val1);
+
+               return val2;
+       }
+
+Under speculation, the first call to load_array() may return the value
+of an out-of-bounds address, while the second call will influence
+microarchitectural state dependent on this value. This may provide an
+arbitrary read primitive.
+
+====================================
+Mitigating speculation side-channels
+====================================
+
+The kernel provides a generic API to ensure that bounds checks are
+respected even under speculation. Architectures which are affected by
+speculation-based side-channels are expected to implement these
+primitives.
+
+The array_index_nospec() helper in <linux/nospec.h> can be used to
+prevent information from being leaked via side-channels.
+
+A call to array_index_nospec(index, size) returns a sanitized index
+value that is bounded to [0, size) even under cpu speculation
+conditions.
+
+This can be used to protect the earlier load_array() example:
+
+       int load_array(int *array, unsigned int index)
+       {
+               if (index >= MAX_ARRAY_ELEMS)
+                       return 0;
+               else {
+                       index = array_index_nospec(index, MAX_ARRAY_ELEMS);
+                       return array[index];
+               }
+       }