Guard Namespace Reference

Synchronizes the kernel with interrupts using the Prologue/Epilogue Model. More...

Functions
void	enter ()
	Entering the critical section from level 0. More...
void	leave ()
	Leaving the critical section. More...
void	relay (Gate &item)
	A prologue wants its epilogue to be processed (entering from level 1). More...

Detailed Description

Synchronizes the kernel with interrupts using the Prologue/Epilogue Model.

The Guard is used to synchronize between "normal" core activities (currently just the text output, later system calls) and interrupt handling routines. For this purpose, Guard has to contain one ore more queues, in which gate objects can be added. This is necessary if the critical section is occupied at the time when an interrupt occurs, and the epilogue() method cannot be executed immediately. The queued epilogues are processed when leaving the critical section.

Hints:

• The epilogue queue is a central data structure, whose consistency must be ensured. The implementation provided by the Queue is not safe against concurrency, i.e. there must never be accesses (neither enqueue nor dequeue) by two cores at the same time. You need to disable interrupts during operations on the queue.

• In MPStuBS, you need a separate epilogue queue for each core, in which each processor serializes its epilogues. However, epilogues on different cores could then be executed in parallel, since the critical section is managed separately on a per-core base. This must be prevented by using a global Ticketlock to avoid concurrent execution of epilogues – there must never be more than one epilogue active on the whole system at the same time!
  Please note: This giant lock (synchronizing all cores) should not be confused with the (core-specific) flag variable that marks only the entry to the epilogue level on the corresponding core!

• Gate objects must not be enqueued multiple times in the same queue. So if two interrupts of the same type occur so quick (for MPStuBS: on the same core) that the corresponding epilogue has not yet been handled, you must not enqueue the same gate object again. The enqueue method already prevents this.

• Interrupts should be disabled for as short as possible. Due to this reason, the prologue/epilogue model allows epilogues to be interrupted by prologues. This means that interrupts should be enabled again before the epilogue is executed (this includes notifying the APIC about the End-Of-Interrupt)

Function Documentation

enter()

void Guard::enter

(

)

Entering the critical section from level 0.

Entering the critical section has to be handled differently depending on the system: In a single-core system it is sufficient to mark the entry by just setting a flag variable (since only one control flow can enter the critical section at the same time). However, as soon as there are multiple cores, this is no longer the case. If a core wants to enter the critical section while another core is already in there, it should (actively) wait in this method until the critical area is released again.

leave()

void Guard::leave

(

)

Leaving the critical section.

Leaves the critical section and processes all remaining (enqueued) epilogues.

relay()

void Guard::relay

(

Gate &

item

)

A prologue wants its epilogue to be processed (entering from level 1).

This method is called by interrupt_handler if the previously executed Gate::prologue has indicated that it needs its epilogue to be run. Whether this is done immediately or the epilogue just enqueued to the epilogue queue depends on whether the critical section on this Core is accessible or not.