A call to i2c_read() for a non-existent address followed by an i2c_read()
to a valid address results in a false abort status for the 2nd call.
i2c_read(1, 0x40, 0, buf, sizeof(buf)) => 0x2000000 (I2C_ERR_TIMEOUT)
i2c_read(1, 0x74, 0, buf, sizeof(buf)) => 0x4000000 (I2C_ERR_ABORT)
Because the abort status register is cleared on read and wait_tx_fifo()
reads it twice, the returned status does not contain the abort status.
Fixing that changed the 2nd read to reflect the abort status.
i2c_read(1, 0x40, 0, buf, sizeof(buf)) => 0x2000000 (I2C_ERR_TIMEOUT)
i2c_read(1, 0x74, 0, buf, sizeof(buf)) => 0x4000001 (I2C_ERR_ABORT)
Bit 0 indicates that the address was not acknowledged by any slave.
That's the abort status from the previous transaction.
So I added a read of the abort status before starting a transaction in
both i2c_read() and i2c_write().
i2c_read(1, 0x40, 0, buf, sizeof(buf)) => 0x2000000 (I2C_ERR_TIMEOUT)
i2c_read(1, 0x74, 0, buf, sizeof(buf)) => 0 (I2C_SUCCESS)
Tested on a Bay Trail E3845 SoC.
Change-Id: I39e4ff4206587267b6fceef58f4a567bf162fbbe
Signed-off-by: Ben Gardner <gardner.ben@gmail.com>
Reviewed-on: https://review.coreboot.org/14160
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>
Reviewed-by: Werner Zeh <werner.zeh@siemens.com>
77e351d9d1