Communicate Device Ownership without External Communication Path (02-Jul-2009)

Communicate Device Ownership without External Communication Path

1

Redundant Device Ownership

Today, it is often necessary, that two controllers are able to access a single device. This aids in redundant solutions, where high availability is a key factor. In those environments, two or more embedded controllers access a single device. If one controller can not access the device, the other controller (called buddy controller) can try and access the device. Only the controller that owns the local bus can access the target chip. Special hardware in the device entity allows each controller to force the local bus, named LBUS in Fig.1, and gain access to the target chip.

This does not help when the target device itself is broken, but helps in case one controller or its connection to the target device is broken. The controller may be broken due to software of hardware problems.

If a device entity can be controlled by two controllers, both controllers coordinate themselves through some external communication path. This can be any communication vehicle such as Ethernet, UART or I²C. Here Ethernet is used.

Fig 1. State of the Art Solution. Ethernet Connection between both Controllers to communicate Chip Ownership.

The black lines (also called links) are the communication link between both

2

controllers to a single device unit. Many device entities can be connected to a pair of controllers, here shown as gray lines.

State of the Art and Problem Statement

A communication between the two controllers is needed to communicate the device ownership and other status information between them. This is done by sending heartbeat messages between each other. This requires an additional external communication path. This additional path adds more complexity to the hardware and software stack and is another source of failure. Furthermore there is no relation between communication path and the target device itself. If a controller can not access the device anymore, but did not notice, the controller could still communicate a good status.

Proposed Solution

The discussed solution builds on a mailbox, named MBX in Fig 2 below. This is a bi-directional communication device located on the device entity. As the mailbox is located in front of the local bus, each controller can receive and send messages without owning the local bus. Each controller can write to the mailbox, as soon as with write is completed, the buddy controller is notified via interrupt that data is available to be read. This is a simple full duplex communication path comparable to Ethernet.

3

Fig 2. Hardware Layout on Device Entity to communicate Chip Ownership between both Controllers.

Hardware Prerequisites

The device entity with the mailbox and local bus offers the following hardware...