Embedded Software for SoC - Grupo de Mecatrônica EESC/USP

Software Architectural Transformations 473
Device drivers are represented by a parallelogram (e.g., the timer device
driver in Figure 34-3).
Application processes are represented by an ellipse with an arrow on the
perimeter (e.g., process read_heart_rate in Figure 34-3). Signal
handlers are represented by a dotted circle attached to the corresponding
application process (e.g., signal_handler1, which is attached to process
update_operation_mode in Figure 34-3).
The association between hardware devices and device drivers is depicted
by a solid line connecting them. A box with a crossbar represents an active
hardware device, whereas a box without a crossbar represents a passive
hardware device. Explanation of the difference between an active device and
a passive device is given later. Arrowed edges between any two vertices
represent the communication of data or control messages, and are annotated
with the selected IPC mechanisms. Since the system is OS-driven, they are
to be implemented as calls to the system functions. For example, the edge
from process main_upsem to process read_heart_rate in Figure 34-3 is a
control message that is implemented using the semaphore service provided by
the OS. Naturally, an edge from or to a hardware device represents the transfer
of data using OS system functions. A small solid square at the termination of
an edge indicates a blocking communication. Otherwise, the communication
is assumed to be non-blocking.
Behind the software architectural representation is a model of computation
featuring the following details:
1.
2.
A hardware device can be either active or passive. An active device initiates
data transfer spontaneously. A timer is a special active device that
spontaneously initiates data (signal) transfer at regular intervals. A passive
device, on the other hand, waits for the processor to initiate data transfer.
There are two types of passive devices. Fast devices respond to the request
from the processor immediately, slow devices require some time (typically
more than 1 ms) to respond to the processor. To efficiently use the
processor, data transfer with a slow passive device is always blocking, so
that the process requesting the data transfer can be suspended and the OS
can bring in other processes for execution. On the other hand, data transfer
with a fast passive device is naturally non-blocking since the device will
respond to the processor request immediately.
Instead of responding to the processor’s request, an active device can
initiate a data transfer. Therefore, data transfer with an active device has
to be blocking, waiting for the active device to take the initiative.
Moreover, to avoid data loss, no process in the SAG should be blocked
by more than one active device, either directly, or indirectly through other
processes. On the other hand, to efficiently use the processor (no busy
waiting), every process in the SAG should be blocked (directly or indirectly)
by at least one active device. Therefore, the natural consequence
of these requirements restricts the number of “active” blocking points for