summaryrefslogtreecommitdiff
path: root/arch/x86/net/bpf_jit_comp.c
diff options
context:
space:
mode:
authorDaniel Borkmann <daniel@iogearbox.net>2018-02-24 01:07:59 +0100
committerAlexei Starovoitov <ast@kernel.org>2018-02-23 22:50:00 -0800
commit6fe8b9c1f41dfe3209dabc5bd0726e003a065288 (patch)
treed252cdd2b889a6b785b499456edda2b0f2759fde /arch/x86/net/bpf_jit_comp.c
parent88e69a1fcc1e67dec3025af64736a84532528242 (diff)
bpf, x64: save several bytes by using mov over movabsq when possible
While analyzing some of the more complex BPF programs from Cilium, I found that LLVM generally prefers to emit LD_IMM64 instead of MOV32 BPF instructions for loading unsigned 32-bit immediates into a register. Given we cannot change the current/stable LLVM versions that are already out there, lets optimize this case such that the JIT prefers to emit 'mov %eax, imm32' over 'movabsq %rax, imm64' whenever suitable in order to reduce the image size by 4-5 bytes per such load in the typical case, reducing image size on some of the bigger programs by up to 4%. emit_mov_imm32() and emit_mov_imm64() have been added as helpers. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Diffstat (limited to 'arch/x86/net/bpf_jit_comp.c')
-rw-r--r--arch/x86/net/bpf_jit_comp.c125
1 files changed, 74 insertions, 51 deletions
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index 4bc36bd1b97a..f3e5cd8c1e68 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -60,7 +60,12 @@ static bool is_imm8(int value)
static bool is_simm32(s64 value)
{
- return value == (s64) (s32) value;
+ return value == (s64)(s32)value;
+}
+
+static bool is_uimm32(u64 value)
+{
+ return value == (u64)(u32)value;
}
/* mov dst, src */
@@ -355,6 +360,68 @@ static void emit_load_skb_data_hlen(u8 **pprog)
*pprog = prog;
}
+static void emit_mov_imm32(u8 **pprog, bool sign_propagate,
+ u32 dst_reg, const u32 imm32)
+{
+ u8 *prog = *pprog;
+ u8 b1, b2, b3;
+ int cnt = 0;
+
+ /* optimization: if imm32 is positive, use 'mov %eax, imm32'
+ * (which zero-extends imm32) to save 2 bytes.
+ */
+ if (sign_propagate && (s32)imm32 < 0) {
+ /* 'mov %rax, imm32' sign extends imm32 */
+ b1 = add_1mod(0x48, dst_reg);
+ b2 = 0xC7;
+ b3 = 0xC0;
+ EMIT3_off32(b1, b2, add_1reg(b3, dst_reg), imm32);
+ goto done;
+ }
+
+ /* optimization: if imm32 is zero, use 'xor %eax, %eax'
+ * to save 3 bytes.
+ */
+ if (imm32 == 0) {
+ if (is_ereg(dst_reg))
+ EMIT1(add_2mod(0x40, dst_reg, dst_reg));
+ b2 = 0x31; /* xor */
+ b3 = 0xC0;
+ EMIT2(b2, add_2reg(b3, dst_reg, dst_reg));
+ goto done;
+ }
+
+ /* mov %eax, imm32 */
+ if (is_ereg(dst_reg))
+ EMIT1(add_1mod(0x40, dst_reg));
+ EMIT1_off32(add_1reg(0xB8, dst_reg), imm32);
+done:
+ *pprog = prog;
+}
+
+static void emit_mov_imm64(u8 **pprog, u32 dst_reg,
+ const u32 imm32_hi, const u32 imm32_lo)
+{
+ u8 *prog = *pprog;
+ int cnt = 0;
+
+ if (is_uimm32(((u64)imm32_hi << 32) | (u32)imm32_lo)) {
+ /* For emitting plain u32, where sign bit must not be
+ * propagated LLVM tends to load imm64 over mov32
+ * directly, so save couple of bytes by just doing
+ * 'mov %eax, imm32' instead.
+ */
+ emit_mov_imm32(&prog, false, dst_reg, imm32_lo);
+ } else {
+ /* movabsq %rax, imm64 */
+ EMIT2(add_1mod(0x48, dst_reg), add_1reg(0xB8, dst_reg));
+ EMIT(imm32_lo, 4);
+ EMIT(imm32_hi, 4);
+ }
+
+ *pprog = prog;
+}
+
static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
int oldproglen, struct jit_context *ctx)
{
@@ -377,7 +444,7 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
const s32 imm32 = insn->imm;
u32 dst_reg = insn->dst_reg;
u32 src_reg = insn->src_reg;
- u8 b1 = 0, b2 = 0, b3 = 0;
+ u8 b2 = 0, b3 = 0;
s64 jmp_offset;
u8 jmp_cond;
bool reload_skb_data;
@@ -485,58 +552,13 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
break;
case BPF_ALU64 | BPF_MOV | BPF_K:
- /* optimization: if imm32 is positive,
- * use 'mov eax, imm32' (which zero-extends imm32)
- * to save 2 bytes
- */
- if (imm32 < 0) {
- /* 'mov rax, imm32' sign extends imm32 */
- b1 = add_1mod(0x48, dst_reg);
- b2 = 0xC7;
- b3 = 0xC0;
- EMIT3_off32(b1, b2, add_1reg(b3, dst_reg), imm32);
- break;
- }
-
case BPF_ALU | BPF_MOV | BPF_K:
- /* optimization: if imm32 is zero, use 'xor <dst>,<dst>'
- * to save 3 bytes.
- */
- if (imm32 == 0) {
- if (is_ereg(dst_reg))
- EMIT1(add_2mod(0x40, dst_reg, dst_reg));
- b2 = 0x31; /* xor */
- b3 = 0xC0;
- EMIT2(b2, add_2reg(b3, dst_reg, dst_reg));
- break;
- }
-
- /* mov %eax, imm32 */
- if (is_ereg(dst_reg))
- EMIT1(add_1mod(0x40, dst_reg));
- EMIT1_off32(add_1reg(0xB8, dst_reg), imm32);
+ emit_mov_imm32(&prog, BPF_CLASS(insn->code) == BPF_ALU64,
+ dst_reg, imm32);
break;
case BPF_LD | BPF_IMM | BPF_DW:
- /* optimization: if imm64 is zero, use 'xor <dst>,<dst>'
- * to save 7 bytes.
- */
- if (insn[0].imm == 0 && insn[1].imm == 0) {
- b1 = add_2mod(0x48, dst_reg, dst_reg);
- b2 = 0x31; /* xor */
- b3 = 0xC0;
- EMIT3(b1, b2, add_2reg(b3, dst_reg, dst_reg));
-
- insn++;
- i++;
- break;
- }
-
- /* movabsq %rax, imm64 */
- EMIT2(add_1mod(0x48, dst_reg), add_1reg(0xB8, dst_reg));
- EMIT(insn[0].imm, 4);
- EMIT(insn[1].imm, 4);
-
+ emit_mov_imm64(&prog, dst_reg, insn[1].imm, insn[0].imm);
insn++;
i++;
break;
@@ -604,7 +626,8 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
EMIT_mov(BPF_REG_0, src_reg);
else
/* mov rax, imm32 */
- EMIT3_off32(0x48, 0xC7, 0xC0, imm32);
+ emit_mov_imm32(&prog, true,
+ BPF_REG_0, imm32);
if (BPF_CLASS(insn->code) == BPF_ALU64)
EMIT1(add_1mod(0x48, AUX_REG));