The situation is similar to commit
bf6376642fe8 ("board: ti: common:
board_detect: Fix EEPROM read quirk"). This is seen on a variant of
eeproms seen on some BeagleBone-AI64 which now has a mix of both 1 byte
addressing and 2 byte addressing eeproms.
Unlike the am335x (ti_i2c_eeprom_am_get) and dra7
(ti_i2c_eeprom_dra7_get) which use constant data structure which allows
us to do a complete read of the data, the
am6(ti_i2c_eeprom_am6_get) eeprom parse operation is dynamic.
This removes the option of being able to read the complete eeprom data
in one single shot.
Fortunately, on the I2C bus, we do see the following behavior: In 1
byte mode, if we attempt to read the first header data yet again, the
misbehaving 2 byte addressing device acts in constant addressing mode
which results in the header not matching up and follow on attempt at 2
byte addressing scheme grabs the correct data.
This costs us an extra ~3 milliseconds, which is a minor penalty
compared to the consistent image support we need to have.
Reported-by: Jason Kridner <jkridner@beagleboard.org>
Fixes:
a58147c2dbbf ("board: ti: common: board_detect: Do 1byte address checks first.")
Signed-off-by: Nishanth Menon <nm@ti.com>
if (rc)
return rc;
+ /*
+ * Handle case of bad 2 byte eeproms that responds to 1 byte addressing
+ * but gets stuck in const addressing when read requests are performed
+ * on offsets. We re-read the board ID to ensure we have sane data back
+ */
+ rc = ti_i2c_eeprom_get(bus_addr, dev_addr, TI_EEPROM_HEADER_MAGIC,
+ sizeof(board_id), (uint8_t *)&board_id);
+ if (rc)
+ return rc;
+
if (board_id.header.id != TI_AM6_EEPROM_RECORD_BOARD_ID) {
pr_err("%s: Invalid board ID record!\n", __func__);
return -EINVAL;