bpf, arm32: Simplify tailcall handling

 - treat tailcall count as 32-bit for access and update
 - change out_offset scope from file to function
 - minor format/structure changes for consistency

Testing: (skipping fentry, fexit, freplace)
========

root@qemu-armhf:/usr/libexec/kselftests-bpf# modprobe test_bpf test_suite=test_tail_calls
test_bpf: #0 Tail call leaf jited:1 967 PASS
test_bpf: #1 Tail call 2 jited:1 1427 PASS
test_bpf: #2 Tail call 3 jited:1 2373 PASS
test_bpf: #3 Tail call 4 jited:1 2304 PASS
test_bpf: #4 Tail call load/store leaf jited:1 1684 PASS
test_bpf: #5 Tail call load/store jited:1 2249 PASS
test_bpf: torvalds#6 Tail call error path, max count reached jited:1 22538 PASS
test_bpf: torvalds#7 Tail call count preserved across function calls jited:1 1055668 PASS
test_bpf: torvalds#8 Tail call error path, NULL target jited:1 513 PASS
test_bpf: torvalds#9 Tail call error path, index out of range jited:1 392 PASS
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]

root@qemu-armhf:/usr/libexec/kselftests-bpf# ./test_progs -n 397/1-12,17-18,23-24,27-31
397/1   tailcalls/tailcall_1:OK
397/2   tailcalls/tailcall_2:OK
397/3   tailcalls/tailcall_3:OK
397/4   tailcalls/tailcall_4:OK
397/5   tailcalls/tailcall_5:OK
397/6   tailcalls/tailcall_6:OK
397/7   tailcalls/tailcall_bpf2bpf_1:OK
397/8   tailcalls/tailcall_bpf2bpf_2:OK
397/9   tailcalls/tailcall_bpf2bpf_3:OK
397/10  tailcalls/tailcall_bpf2bpf_4:OK
397/11  tailcalls/tailcall_bpf2bpf_5:OK
397/12  tailcalls/tailcall_bpf2bpf_6:OK
397/17  tailcalls/tailcall_poke:OK
397/18  tailcalls/tailcall_bpf2bpf_hierarchy_1:OK
397/23  tailcalls/tailcall_bpf2bpf_hierarchy_2:OK
397/24  tailcalls/tailcall_bpf2bpf_hierarchy_3:OK
397/27  tailcalls/tailcall_failure:OK
397/28  tailcalls/reject_tail_call_spin_lock:OK
397/29  tailcalls/reject_tail_call_rcu_lock:OK
397/30  tailcalls/reject_tail_call_preempt_lock:OK
397/31  tailcalls/reject_tail_call_ref:OK
397     tailcalls:OK
Summary: 1/21 PASSED, 0 SKIPPED, 0 FAILED

Signed-off-by: Tony Ambardar <[email protected]>

Author: guidosarducci

arch/arm/net/bpf_jit_32.c

@@ -1403,23 +1403,21 @@ static inline void emit_ar_r(const u8 rd, const u8 rt, const u8 rm,
 	}
 }
 
-static int out_offset = -1; /* initialized on the first pass of build_body() */
 
+/* Helper bpf_tail_call(void *prog_ctx, struct bpf_array *array, u32 index) */
+#define cur_offset (ctx->idx - idx0)
+#define jmp_offset (out_offset - (cur_offset) - 2)
 static int emit_bpf_tail_call(struct jit_ctx *ctx)
 {
-
-	/* bpf_tail_call(void *prog_ctx, struct bpf_array *array, u64 index) */
 	const s8 *r2 = bpf2a32[BPF_REG_2];
 	const s8 *r3 = bpf2a32[BPF_REG_3];
 	const s8 *tmp = bpf2a32[TMP_REG_1];
 	const s8 *tmp2 = bpf2a32[TMP_REG_2];
 	const s8 *tcc = bpf2a32[TCALL_CNT];
-	const s8 *tc;
+	static int out_offset = -1; /* initialized on JIT 1st pass */
 	const int idx0 = ctx->idx;
-#define cur_offset (ctx->idx - idx0)
-#define jmp_offset (out_offset - (cur_offset) - 2)
-	u32 lo, hi;
-	s8 r_array, r_index;
+	u32 tc_max;
+	s8 tc, r_array, r_index;
 	int off;
 
 	/* if (index >= array->map.max_entries)
@@ -1444,15 +1442,12 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx)
 	 *	goto out;
 	 * tail_call_cnt++;
 	 */
-	lo = (u32)MAX_TAIL_CALL_CNT;
-	hi = (u32)((u64)MAX_TAIL_CALL_CNT >> 32);
-	tc = arm_bpf_get_reg64(tcc, tmp, ctx);
-	emit(ARM_CMP_I(tc[0], hi), ctx);
-	_emit(ARM_COND_EQ, ARM_CMP_I(tc[1], lo), ctx);
+	tc_max = (u32)MAX_TAIL_CALL_CNT;
+	tc = arm_bpf_get_reg32(tcc[1], tmp[1], ctx);
+	emit(ARM_CMP_I(tc, tc_max), ctx);
 	_emit(ARM_COND_CS, ARM_B(jmp_offset), ctx);
-	emit(ARM_ADDS_I(tc[1], tc[1], 1), ctx);
-	emit(ARM_ADC_I(tc[0], tc[0], 0), ctx);
-	arm_bpf_put_reg64(tcc, tmp, ARM_LR, ctx);
+	emit(ARM_ADD_I(tc, tc, 1), ctx);
+	arm_bpf_put_reg32(tcc[1], tc, tmp[0], ctx);
 
 	/* prog = array->ptrs[index]
 	 * if (prog == NULL)
@@ -1485,9 +1480,9 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx)
 		return -1;
 	}
 	return 0;
+}
 #undef cur_offset
 #undef jmp_offset
-}
 
 /* 0xabcd => 0xcdab */
 static inline void emit_rev16(const u8 rd, const u8 rn, struct jit_ctx *ctx)