bpf: arm64: remove callee-save registers use for tmp registers

In the current implementation of ARM64 eBPF JIT, R23 and R24 are used for
tmp registers, which are callee-saved registers. This leads to variable size
of JIT prologue and epilogue. The latest blinding constant change prefers to
constant size of prologue and epilogue. AAPCS reserves R9 ~ R15 for temp
registers which not need to be saved/restored during function call. So, replace
R23 and R24 to R10 and R11, and remove tmp_used flag to save 2 instructions for
some jited BPF program.

CC: Daniel Borkmann <[e-mail]>
Acked-by: Zi Shen Lim <[e-mail]>
Signed-off-by: Yang Shi <[e-mail]>
Acked-by: Catalin Marinas <[e-mail]>
Acked-by: Daniel Borkmann <[e-mail]>
Signed-off-by: David S. Miller <[e-mail]>

arch/arm64/net/bpf_jit_comp.c

···
 	[BPF_REG_9] = A64_R(22),
 	/* read-only frame pointer to access stack */
 	[BPF_REG_FP] = A64_R(25),
-	/* temporary register for internal BPF JIT */
-	[TMP_REG_1] = A64_R(23),
-	[TMP_REG_2] = A64_R(24),
+	/* temporary registers for internal BPF JIT */
+	[TMP_REG_1] = A64_R(10),
+	[TMP_REG_2] = A64_R(11),
 	/* temporary register for blinding constants */
 	[BPF_REG_AX] = A64_R(9),
 };
···
 struct jit_ctx {
 	const struct bpf_prog *prog;
 	int idx;
-	int tmp_used;
 	int epilogue_offset;
 	int *offset;
 	u32 *image;
···
 	const u8 r8 = bpf2a64[BPF_REG_8];
 	const u8 r9 = bpf2a64[BPF_REG_9];
 	const u8 fp = bpf2a64[BPF_REG_FP];
-	const u8 tmp1 = bpf2a64[TMP_REG_1];
-	const u8 tmp2 = bpf2a64[TMP_REG_2];
 
 	/*
 	 * BPF prog stack layout
···
 	 *                        | ... | callee saved registers
 	 *                        +-----+
 	 *                        |     | x25/x26
-	 * BPF fp register => -80:+-----+ <= (BPF_FP)
+	 * BPF fp register => -64:+-----+ <= (BPF_FP)
 	 *                        |     |
 	 *                        | ... | BPF prog stack
 	 *                        |     |
···
 	/* Save callee-saved register */
 	emit(A64_PUSH(r6, r7, A64_SP), ctx);
 	emit(A64_PUSH(r8, r9, A64_SP), ctx);
-	if (ctx->tmp_used)
-		emit(A64_PUSH(tmp1, tmp2, A64_SP), ctx);
 
 	/* Save fp (x25) and x26. SP requires 16 bytes alignment */
 	emit(A64_PUSH(fp, A64_R(26), A64_SP), ctx);
···
 	const u8 r8 = bpf2a64[BPF_REG_8];
 	const u8 r9 = bpf2a64[BPF_REG_9];
 	const u8 fp = bpf2a64[BPF_REG_FP];
-	const u8 tmp1 = bpf2a64[TMP_REG_1];
-	const u8 tmp2 = bpf2a64[TMP_REG_2];
 
 	/* We're done with BPF stack */
 	emit(A64_ADD_I(1, A64_SP, A64_SP, STACK_SIZE), ctx);
···
 	emit(A64_POP(fp, A64_R(26), A64_SP), ctx);
 
 	/* Restore callee-saved register */
-	if (ctx->tmp_used)
-		emit(A64_POP(tmp1, tmp2, A64_SP), ctx);
 	emit(A64_POP(r8, r9, A64_SP), ctx);
 	emit(A64_POP(r6, r7, A64_SP), ctx);
 
···
 			emit(A64_UDIV(is64, dst, dst, src), ctx);
 			break;
 		case BPF_MOD:
-			ctx->tmp_used = 1;
 			emit(A64_UDIV(is64, tmp, dst, src), ctx);
 			emit(A64_MUL(is64, tmp, tmp, src), ctx);
 			emit(A64_SUB(is64, dst, dst, tmp), ctx);
···
 	/* dst = dst OP imm */
 	case BPF_ALU | BPF_ADD | BPF_K:
 	case BPF_ALU64 | BPF_ADD | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_ADD(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_SUB | BPF_K:
 	case BPF_ALU64 | BPF_SUB | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_SUB(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_AND | BPF_K:
 	case BPF_ALU64 | BPF_AND | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_AND(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_OR | BPF_K:
 	case BPF_ALU64 | BPF_OR | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_ORR(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_XOR | BPF_K:
 	case BPF_ALU64 | BPF_XOR | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_EOR(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_MUL | BPF_K:
 	case BPF_ALU64 | BPF_MUL | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_MUL(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_DIV | BPF_K:
 	case BPF_ALU64 | BPF_DIV | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp, imm, ctx);
 		emit(A64_UDIV(is64, dst, dst, tmp), ctx);
 		break;
 	case BPF_ALU | BPF_MOD | BPF_K:
 	case BPF_ALU64 | BPF_MOD | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(is64, tmp2, imm, ctx);
 		emit(A64_UDIV(is64, tmp, dst, tmp2), ctx);
 		emit(A64_MUL(is64, tmp, tmp, tmp2), ctx);
···
 	case BPF_JMP | BPF_JNE | BPF_K:
 	case BPF_JMP | BPF_JSGT | BPF_K:
 	case BPF_JMP | BPF_JSGE | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(1, tmp, imm, ctx);
 		emit(A64_CMP(1, dst, tmp), ctx);
 		goto emit_cond_jmp;
 	case BPF_JMP | BPF_JSET | BPF_K:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(1, tmp, imm, ctx);
 		emit(A64_TST(1, dst, tmp), ctx);
 		goto emit_cond_jmp;
···
 		const u8 r0 = bpf2a64[BPF_REG_0];
 		const u64 func = (u64)__bpf_call_base + imm;
 
-		ctx->tmp_used = 1;
 		emit_a64_mov_i64(tmp, func, ctx);
 		emit(A64_PUSH(A64_FP, A64_LR, A64_SP), ctx);
 		emit(A64_MOV(1, A64_FP, A64_SP), ctx);
···
 	case BPF_LDX | BPF_MEM | BPF_H:
 	case BPF_LDX | BPF_MEM | BPF_B:
 	case BPF_LDX | BPF_MEM | BPF_DW:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(1, tmp, off, ctx);
 		switch (BPF_SIZE(code)) {
 		case BPF_W:
···
 	case BPF_ST | BPF_MEM | BPF_B:
 	case BPF_ST | BPF_MEM | BPF_DW:
 		/* Load imm to a register then store it */
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(1, tmp2, off, ctx);
 		emit_a64_mov_i(1, tmp, imm, ctx);
 		switch (BPF_SIZE(code)) {
···
 	case BPF_STX | BPF_MEM | BPF_H:
 	case BPF_STX | BPF_MEM | BPF_B:
 	case BPF_STX | BPF_MEM | BPF_DW:
-		ctx->tmp_used = 1;
 		emit_a64_mov_i(1, tmp, off, ctx);
 		switch (BPF_SIZE(code)) {
 		case BPF_W:
···
 
 	/* 1. Initial fake pass to compute ctx->idx. */
 
-	/* Fake pass to fill in ctx->offset and ctx->tmp_used. */
+	/* Fake pass to fill in ctx->offset. */
 	if (build_body(&ctx)) {
 		prog = orig_prog;
 		goto out_off;