[Stable-8.2.7,22/53] target/arm: Avoid shifts by -1 in tszimm_shr() and tszimm_shl()

Commit Message

Michael Tokarev Sept. 6, 2024, 6:53 a.m. UTC

From: Peter Maydell <[e-mail]>

The function tszimm_esz() returns a shift amount, or possibly -1 in
certain cases that correspond to unallocated encodings in the
instruction set.  We catch these later in the trans_ functions
(generally with an "a-esz < 0" check), but before we do the
decodetree-generated code will also call tszimm_shr() or tszimm_sl(),
which will use the tszimm_esz() return value as a shift count without
checking that it is not negative, which is undefined behaviour.

Avoid the UB by checking the return value in tszimm_shr() and
tszimm_shl().

Cc: [e-mail]
Resolves: Coverity CID 1547617, 1547694
Signed-off-by: Peter Maydell <[e-mail]>
Reviewed-by: Richard Henderson <[e-mail]>
Message-id: [e-mail]
(cherry picked from commit 76916dfa89e8900639c1055c07a295c06628a0bc)
Signed-off-by: Michael Tokarev <[e-mail]>

Patch

diff --git a/target/arm/tcg/translate-sve.c b/target/arm/tcg/translate-sve.c
index ada05aa530..466a19c25a 100644
--- a/target/arm/tcg/translate-sve.c
+++ b/target/arm/tcg/translate-sve.c
@@ -50,13 +50,27 @@  static int tszimm_esz(DisasContext *s, int x)
 
 static int tszimm_shr(DisasContext *s, int x)
 {
-    return (16 << tszimm_esz(s, x)) - x;
+    /*
+     * We won't use the tszimm_shr() value if tszimm_esz() returns -1 (the
+     * trans function will check for esz < 0), so we can return any
+     * value we like from here in that case as long as we avoid UB.
+     */
+    int esz = tszimm_esz(s, x);
+    if (esz < 0) {
+        return esz;
+    }
+    return (16 << esz) - x;
 }
 
 /* See e.g. LSL (immediate, predicated).  */
 static int tszimm_shl(DisasContext *s, int x)
 {
-    return x - (8 << tszimm_esz(s, x));
+    /* As with tszimm_shr(), value will be unused if esz < 0 */
+    int esz = tszimm_esz(s, x);
+    if (esz < 0) {
+        return esz;
+    }
+    return x - (8 << esz);
 }
 
 /* The SH bit is in bit 8.  Extract the low 8 and shift.  */