target/i386: Fix BEXTR instruction

There were two problems here: not limiting the input to operand bits,
and not correctly handling large extraction length.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1372
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230114230542.3116013-3-richard.henderson@linaro.org>
Cc: qemu-stable@nongnu.org
Fixes: 1d0b926150e5 ("target/i386: move scalar 0F 38 and 0F 3A instruction to new decoder", 2022-10-18)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

2 files changed, 23 insertions, 11 deletions

diff --git a/target/i386/tcg/emit.c.inc b/target/i386/tcg/emit.c.inc
index 7037ff91c6..99f6ba6e19 100644
--- a/target/i386/tcg/emit.c.inc
+++ b/target/i386/tcg/emit.c.inc
@@ -1078,30 +1078,30 @@ static void gen_ANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 static void gen_BEXTR(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 {
     MemOp ot = decode->op[0].ot;
-    TCGv bound, zero;
+    TCGv bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
+    TCGv zero = tcg_constant_tl(0);
+    TCGv mone = tcg_constant_tl(-1);
 
     /*
      * Extract START, and shift the operand.
      * Shifts larger than operand size get zeros.
      */
     tcg_gen_ext8u_tl(s->A0, s->T1);
+    if (TARGET_LONG_BITS == 64 && ot == MO_32) {
+        tcg_gen_ext32u_tl(s->T0, s->T0);
+    }
     tcg_gen_shr_tl(s->T0, s->T0, s->A0);
 
-    bound = tcg_constant_tl(ot == MO_64 ? 63 : 31);
-    zero = tcg_constant_tl(0);
     tcg_gen_movcond_tl(TCG_COND_LEU, s->T0, s->A0, bound, s->T0, zero);
 
     /*
-     * Extract the LEN into a mask.  Lengths larger than
-     * operand size get all ones.
+     * Extract the LEN into an inverse mask.  Lengths larger than
+     * operand size get all zeros, length 0 gets all ones.
      */
     tcg_gen_extract_tl(s->A0, s->T1, 8, 8);
-    tcg_gen_movcond_tl(TCG_COND_LEU, s->A0, s->A0, bound, s->A0, bound);
-
-    tcg_gen_movi_tl(s->T1, 1);
-    tcg_gen_shl_tl(s->T1, s->T1, s->A0);
-    tcg_gen_subi_tl(s->T1, s->T1, 1);
-    tcg_gen_and_tl(s->T0, s->T0, s->T1);
+    tcg_gen_shl_tl(s->T1, mone, s->A0);
+    tcg_gen_movcond_tl(TCG_COND_LEU, s->T1, s->A0, bound, s->T1, zero);
+    tcg_gen_andc_tl(s->T0, s->T0, s->T1);
 
     gen_op_update1_cc(s);
     set_cc_op(s, CC_OP_LOGICB + ot);
diff --git a/tests/tcg/i386/test-i386-bmi2.c b/tests/tcg/i386/test-i386-bmi2.c
index 3c3ef85513..982d4abda4 100644
--- a/tests/tcg/i386/test-i386-bmi2.c
+++ b/tests/tcg/i386/test-i386-bmi2.c
@@ -99,6 +99,9 @@ int main(int argc, char *argv[]) {
     result = bextrq(mask, 0x10f8);
     assert(result == 0);
 
+    result = bextrq(0xfedcba9876543210ull, 0x7f00);
+    assert(result == 0xfedcba9876543210ull);
+
     result = blsiq(0x30);
     assert(result == 0x10);
 
@@ -164,6 +167,15 @@ int main(int argc, char *argv[]) {
     result = bextrl(mask, 0x1038);
     assert(result == 0);
 
+    result = bextrl((reg_t)0x8f635a775ad3b9b4ull, 0x3018);
+    assert(result == 0x5a);
+
+    result = bextrl((reg_t)0xfedcba9876543210ull, 0x7f00);
+    assert(result == 0x76543210u);
+
+    result = bextrl(-1, 0);
+    assert(result == 0);
+
     result = blsil(0xffff);
     assert(result == 1);