The NASA report [1] goes into the details. The procedure was to fire the service module reaction control thrusters to separate it from the command module, spin

The NASA report [1] goes into the details. The procedure was to fire the service module reaction control thrusters to separate it from the command module, spin it to give it stability, and then exhaust their fuel in slowing it down. As the service module was not symmetric around its longitudinal axis, however, the initial spin was not exactly aligned along that axis, and would precess with a period of about 10 seconds. This would lead to sloshing of the residual service module propellants, exacerbating the misalignment markedly -- sometimes wobbling beyond 90 degrees. This meant that the thrusters were not achieving the desired change in velocity, and might sometimes even accelerate towards the command module. After modeling the problem for various residual fuel masses, a solution was found in reducing the time spent on spinning the service module to two seconds, and cutting off all thrusters after 25 seconds. [1] https://archive.org/details/nasa_techdoc_19710017109/page/n3 https://archive.org/details/nasa_techdoc_19710017109/page/n3