gmbedded/Real-Time Operating System Concepts 191 Ina preemptive kernel, response time = interrupt latency + time to save CPU registers context. syem’s worst-case interrupt response time has to syieroperating systenvembedded sotware bbe considered while evaluating the performance - Interrupt Recovery Time: Time required for CPU to 0 the ‘odethighest pri ei called interrupt recovery time, to return to the interrupted codefhighest priority In non-preemprive kernel, interrupt recovery time = time to restore the CPU context + time to execute the return instruction from the interrupted instruction. In preemptive kernel, interrupt recovery time = time tovcheck whether a high priority task is ready + time to restore CPU context of the highest priority task + time to execute the return instruction from the interrupt instruction. Task CPU cpu Task Running Context Context Running Saving Saving | interrust Lag Recovery 1 1 1 Interrupt 1 Request 1 1 Interrupt hag Latency | g— —> r interrupt 1 Response Time Fig. 7.7 Interrupt Latency, Interrupt Response Time and Interrupt Recovery Time The interrupt latency, interrupt response time and interrupr recovery time are shown in Fig. 7.7. 7.4 Semaphores When multiple tasks are running, two or mote tasks may need ¢o share the same resource. As shown in Fig. 7.8(a), consider a case where two tasks want to write to a display. Assume that taskl wants to display the message “temperature is 50” and task2 has to display the message “humidity is 4096" The display is a shared resource and if there is no synchronization between the asks, then a garbled message is displayed such as “cemphumieratiditurey is is 50 40%", To access a shared resource, there should be a mechanism so that there is discipline, This is known as resource synchronization. Scanned with CamScar192 Display Task 1 Task 2 Fig, 78 (a): Resource Synchronization Fig. 7.8 (b): Task Synchronization Now consider another situation shown in Fig, 7.8(b), In this case, one task reads the data from an ADC and writes it co memory. Another task reads that data and sends it to a DAC. The read operation takes place only after write operation and it has to be done very fast with minimal time delay. In this case, there should be a mechanism for task to inform task2 that it has done its job. This has to°be done through a well-defined procedure. This is known as task synchronization ‘Semaphor is-a-kernel abject that is used for both resource synchronization and task synchronization. ~Scannea with vamscalEmbedded Real-Time Operating System Concepts 193 Q TO Semeohors La Se © © Task 1 Task 2 © © © © Task 1 Task 1 Task 1 Task 2 Task 2 Task 2 Acquires Uses Releases Acquires Uses Releases ‘Semaphore Display Semaphore ‘Semaphore Display + Semaphore Fig. 7.9: Display Semaphore Consider a situation shown in Fig. 7.9. Two tasks want to access a display. Display is a shared resource, To control the access, is created. The semaphore is like a key to enter a house, and hence the semaphore is represented as a ‘key’. If taskl wants to access the printer, it acquires the semaphore, uses the printer and then releases the semaphore. If both the tasks want to access a resource simultaneously, the kernel has to give the semaphore only to one of the tasks. This allocation may be based on the priority of the task or on first-come-first-served basis. If a number of tasks have to access the same resource then the tasks are kept in a queue and each task can acquire the semaphore semaphor one by one. Pool of Butlers f=] | tf fg Fig. 7.10: Counting Semaphore 4 — canned with CamscarConsider another example in which a pool of 10 buffers is available, The buffers need to be shared by a number of tasks, as shown in Fig. 7.10. Any task can write to a buffer. Using a binary semaphore Semaphore is just an integer. does not work in this case. So, a counting semaphore is used. The Semaphores are of two types: initial value of the semaphore is set to 10. Whenever a task acquires counting semaphore and binary | the semaphore, the value is decremented by 1 and whenever a task semaphore. Counting releases the semaphore, it is incremented by 1. When the value is semaphore will have an i we wae A ilable. Pr an integer 0, it is an indication that the shared resource is no longer available. value greater than 1. Binary take the values semaphore wi of either 0 or 1. A counting semaphore is like having multiple keys to enter a house. CamScai ocanne