I'm developping an application on an ARM-7 based platform which is running Linux 2.4.18 with the latest BlueZ version for that platform. In this application there is a master device that can connect to up to 7 slaves (master and slave devices are all the same, the external controller determines which is master).

The application does the following, the external controller on the master can send a single command to the Linux application, in the command is an address field identifying to which slaves the command must be passed on. A slave will pass the command on to its external controller, which will respond to the command by sending a reply back to the Linux slave, which in turn passes it on to the Linux master and is then handed over to the external controller again. As soon as all addressed slaves have responded, the sequence is repeated.

Now to what I see happening, when I run this with 5 slaves, I don't see any problems for more than an hour, but when I run it with 6 or 7 slaves, I see that the BlueCore resets itself, with 6 slaves after about 40 minutes and with 7 slaves already after 5 to 10 minutes. When I eliminate the CRC check I can run the 7 slaves configuration for up to about 40 minutes after it crashes.

What I think happens is that the BlueCore receives data to fast over the air and can not pass it on to the Linux CPU quick enough, so there is an internal overflow in the BlueCore.