the logic will assign the token arbitrarily to one of the ports. It should be noted, however,
that semaphores alone do not guarantee that access to a resource is secure. As with
any powerful programming technique, errors can be introduced if semaphores are mis-
used or misinterpreted. Code integrity is of the utmost performance when semaphores
are being used instead of slower, more restrictive hardware-intensive systems.
Semaphore initialization is not automatic and must therefore be incorporated in the
power up initialization procedures. Since any semaphore flag containing a zero must be
reset to one, initialization should write a one to all request flags from both sides to
ensure that they will be available when required.
Using Semaphores –
Some Examples
Perhaps the simplest application of semaphores is their use as resource markers for the
M67025E’s dual-port RAM. If it is necessary to split the 8 Kb
×
16 RAM into two 4 Kb
×
16
blocks which are to be dedicated to serving either the left or right port at any one time. Sema-
phore 0 can be used to indicate which side is controlling the lower segment of memory and
semaphore 1 can be defined as indicating the upper segment of memory.
To take control of a resource, in this case the lower 4 Kb of a dual-port RAM, the left port
processor would then write a zero into semaphore flag 0 and then read it back. If suc-
cessful in taking the token (reading back a zero rather than a one), the left processor
could then take control of the lower 4 Kb of RAM. If the right processor attempts to per-
form the same function to take control of the resource after the left processor has
already done so, it will read back a one in response to the attempted write of a zero into
semaphore 0. At this point the software may choose to attempt to gain control of the
second 4 Kb segment of RAM by writing and then reading a zero in semaphore 1. If suc-
cessful, it will lock out the left processor.
Once the left side has completed its task it will write a one to semaphore 0 and may then
attempt to access semaphore 1. If semaphore 1 is still occupied by the right side, the left
side may abandon its semaphore request and perform other operations until it is able to
write and then read a zero in semaphore 1. If the right processor performs the same
operation with semaphore 0, this protocol would then allow the two processes to swap
4 Kb blocks of dual-port RAM between one another.
The blocks do not have to be any particular size, and may even be of variable size
depending on the complexity of the software using the semaphore flags. All eight sema-
phores could be used to divide the dual-port RAM or other shared resources into eight
parts. Semaphores can even be assigned different meanings on each side, rather than
having a common meaning as is described in the above example.
Semaphores are a useful form of arbitration in systems such as disk interfaces where
the CPU must be locked out of a segment of memory during a data transfer operation,
and the I/0 device cannot tolerate any wait states. If semaphores are used, both the
CPU and the I/0 device can access assigned memory segments, without the need for
wait states, once the two devices have determined which memory area is barred to the
CPU.
Semaphores are also useful in applications where no memory WAIT state is available
on one or both sides. On a semaphore handshake has been performed, both proces-
sors can access their assigned RAM segments at full speed.
Another application is in complex data structures. Block arbitration is very important in
this case, since one processor may be responsible for building and updating a data
structure whilst the other processor reads and interprets it. A major error condition may
be created if the interpreting processor reads an incomplete data structure. Some sort of
arbitration between the two different processors is therefore necessary. The building
processor requests access to the block, locks it and is then able to enter the block to
7
M67025E
4146J–AERO–06/03
ATMEL [ ATMEL CORPORATION ]
