u-boot: v2026.04: helios64: retry OTP read on transient SPI failure

The previous commit preserved has_been_read = 0 on SPI I/O failure so a
later call could retry. But in the actual boot flow no retry happens:
board_early_init_r() ignores the return of read_otp_data(), and the
later consumers (set_board_info, mac_read_from_otp) bail out early with
'if (!is_data_valid()) return;' — so a single transient SPI failure
leaves board_rev / serial# / ethaddr unset for the whole boot.

Add an ensure_otp_data_ready() helper that re-invokes read_otp_data()
when the cached data is not yet valid, and use it in both consumers.
Valid/invalid-data behavior is unchanged; only the retry path is now
actually exercised.

Addresses a CodeRabbit review comment on PR armbian#9675.

Author: iav

patch/u-boot/v2026.04/board_helios64/board/sys_otp.c

@@ -43,6 +43,21 @@ static inline int is_data_valid(void)
 	return data_valid;
 }
 
+static int read_otp_data(void);
+
+/*
+ * Ensure OTP data is cached and valid. If a previous read failed (for example
+ * on a transient SPI error during board_early_init_r()), retry here so
+ * consumers don't silently skip using OTP-provided values for the whole boot.
+ */
+static int ensure_otp_data_ready(void)
+{
+	if (is_data_valid())
+		return 0;
+
+	return read_otp_data();
+}
+
 /*
  * Decode the 6-byte serial number into a u64. Doing a casted *(uint64_t *)
  * read on otp.serial_num would over-read into otp.mfg_year and is unaligned
@@ -231,7 +246,7 @@ void set_board_info(void)
 {
 	char env_str[13];
 
-	if (!is_data_valid())
+	if (ensure_otp_data_ready())
 		return;
 
 	snprintf(env_str, sizeof(env_str), "%i.%i", (otp.revision & 0xf0) >> 4, otp.revision & 0x0f);
@@ -246,7 +261,7 @@ int mac_read_from_otp(void)
 {
 	unsigned int i;
 
-	if (!is_data_valid())
+	if (ensure_otp_data_ready())
 		return -1;
 
 	for (i = 0; i < MAX_NUM_PORTS; i++) {