ARM: net: support BPF_ALU | BPF_MOD instructions in the BPF JIT.

For ARMv7 with UDIV instruction support, generate an UDIV instruction
followed by an MLS instruction.

For other ARM variants, generate code calling a C wrapper similar to
the jit_udiv() function used for BPF_ALU | BPF_DIV instructions.

Some performance numbers reported by the test_bpf module (the duration
per filter run is reported in nanoseconds, between "jitted:<x>" and
"PASS":

ARMv7 QEMU nojit:	test_bpf: #3 DIV_MOD_KX jited:0 2196 PASS
ARMv7 QEMU jit:		test_bpf: #3 DIV_MOD_KX jited:1 104 PASS
ARMv5 QEMU nojit:	test_bpf: #3 DIV_MOD_KX jited:0 2176 PASS
ARMv5 QEMU jit:		test_bpf: #3 DIV_MOD_KX jited:1 1104 PASS
ARMv5 kirkwood nojit:	test_bpf: #3 DIV_MOD_KX jited:0 1103 PASS
ARMv5 kirkwood jit:	test_bpf: #3 DIV_MOD_KX jited:1 311 PASS
Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

arch/arm/net/bpf_jit_32.c

@@ -125,7 +125,7 @@ static u64 jit_get_skb_w(struct sk_buff *skb, int offset)
 }
 
 /*
- * Wrapper that handles both OABI and EABI and assures Thumb2 interworking
+ * Wrappers which handle both OABI and EABI and assures Thumb2 interworking
  * (where the assembly routines like __aeabi_uidiv could cause problems).
  */
 static u32 jit_udiv(u32 dividend, u32 divisor)
@@ -133,6 +133,11 @@ static u32 jit_udiv(u32 dividend, u32 divisor)
 	return dividend / divisor;
 }
 
+static u32 jit_mod(u32 dividend, u32 divisor)
+{
+	return dividend % divisor;
+}
+
 static inline void _emit(int cond, u32 inst, struct jit_ctx *ctx)
 {
 	inst |= (cond << 28);
@@ -471,11 +476,17 @@ static inline void emit_blx_r(u8 tgt_reg, struct jit_ctx *ctx)
 #endif
 }
 
-static inline void emit_udiv(u8 rd, u8 rm, u8 rn, struct jit_ctx *ctx)
+static inline void emit_udivmod(u8 rd, u8 rm, u8 rn, struct jit_ctx *ctx,
+				int bpf_op)
 {
 #if __LINUX_ARM_ARCH__ == 7
 	if (elf_hwcap & HWCAP_IDIVA) {
-		emit(ARM_UDIV(rd, rm, rn), ctx);
+		if (bpf_op == BPF_DIV)
+			emit(ARM_UDIV(rd, rm, rn), ctx);
+		else {
+			emit(ARM_UDIV(ARM_R3, rm, rn), ctx);
+			emit(ARM_MLS(rd, rn, ARM_R3, rm), ctx);
+		}
 		return;
 	}
 #endif
@@ -496,7 +507,8 @@ static inline void emit_udiv(u8 rd, u8 rm, u8 rn, struct jit_ctx *ctx)
 		emit(ARM_MOV_R(ARM_R0, rm), ctx);
 
 	ctx->seen |= SEEN_CALL;
-	emit_mov_i(ARM_R3, (u32)jit_udiv, ctx);
+	emit_mov_i(ARM_R3, bpf_op == BPF_DIV ? (u32)jit_udiv : (u32)jit_mod,
+		   ctx);
 	emit_blx_r(ARM_R3, ctx);
 
 	if (rd != ARM_R0)
@@ -697,13 +709,27 @@ static int build_body(struct jit_ctx *ctx)
 			if (k == 1)
 				break;
 			emit_mov_i(r_scratch, k, ctx);
-			emit_udiv(r_A, r_A, r_scratch, ctx);
+			emit_udivmod(r_A, r_A, r_scratch, ctx, BPF_DIV);
 			break;
 		case BPF_ALU | BPF_DIV | BPF_X:
 			update_on_xread(ctx);
 			emit(ARM_CMP_I(r_X, 0), ctx);
 			emit_err_ret(ARM_COND_EQ, ctx);
-			emit_udiv(r_A, r_A, r_X, ctx);
+			emit_udivmod(r_A, r_A, r_X, ctx, BPF_DIV);
+			break;
+		case BPF_ALU | BPF_MOD | BPF_K:
+			if (k == 1) {
+				emit_mov_i(r_A, 0, ctx);
+				break;
+			}
+			emit_mov_i(r_scratch, k, ctx);
+			emit_udivmod(r_A, r_A, r_scratch, ctx, BPF_MOD);
+			break;
+		case BPF_ALU | BPF_MOD | BPF_X:
+			update_on_xread(ctx);
+			emit(ARM_CMP_I(r_X, 0), ctx);
+			emit_err_ret(ARM_COND_EQ, ctx);
+			emit_udivmod(r_A, r_A, r_X, ctx, BPF_MOD);
 			break;
 		case BPF_ALU | BPF_OR | BPF_K:
 			/* A |= K */

arch/arm/net/bpf_jit_32.h

@@ -115,6 +115,8 @@
 
 #define ARM_INST_UMULL		0x00800090
 
+#define ARM_INST_MLS		0x00600090
+
 /*
  * Use a suitable undefined instruction to use for ARM/Thumb2 faulting.
  * We need to be careful not to conflict with those used by other modules
@@ -210,4 +212,7 @@
 #define ARM_UMULL(rd_lo, rd_hi, rn, rm)	(ARM_INST_UMULL | (rd_hi) << 16 \
 					 | (rd_lo) << 12 | (rm) << 8 | rn)
 
+#define ARM_MLS(rd, rn, rm, ra)	(ARM_INST_MLS | (rd) << 16 | (rn) | (rm) << 8 \
+				 | (ra) << 12)
+
 #endif /* PFILTER_OPCODES_ARM_H */