WO2003091878A2 - Reduced overhead context-saving in static priority scheduled operating systems - Google Patents

Abstract

There are disclosed methods, computing devices and software products that seek to reduce memory/processor overheads in static priority scheduled operating systems by reducing the number of resource contexts that need to be saved and restored when running a plurality of tasks. This may be achieved by forming mutually-exclusive non-pre-emption groups of tasks, and/or by determining which tasks share a common resource.

Description

IMPROVEMENTS RELATING TO REDUCED-OVERHEAD CONTEXT- SAVING IN STATIC PRIORITY SCHEDULED OPERATING SYSTEMS

The present invention relates to methods and systems for reducing memory and processor usage in operating systems that make use of static priority-based scheduling, specifically, in reducing the need for resource saving on inter-task context switches. Embodiments of the present invention are of particular relevance to the technical field of embedded devices and the control thereof, since these generally use static priority-based scheduling given the limited range of functions required. Embodiments of the present invention may be used in conjunction with an invention disclosed in the present applicant's copending UK patent application no 0209800.2 of 30^th April 2002, a copy of which is being filed with the present application and the full contents of which are hereby incorporated into the present application by reference.

hi an operating system that is based around static priority based scheduling (systems in which the priority ordering of tasks is determined offline and remains constant) there exists some external description of the tasking structure of the system that is used to generate data structures describing the tasks and their priorities. These data structures are most advantageously calculated offline by a "configuration tool" or other program and used to initialise the system when it starts up. Tasks in such an operating system may be activated at any time, but do not start to run until they are the highest-priority runnable task in the system. They continue to run (though they may be pre-empted by other higher-priority tasks at which point a context-switch takes place) until their work is done and then terminate.

Applications built using such operating systems typically consist of a plurality of concurrent tasks each set at a predetermined priority. Embodiments of the present invention relate to reducing the amount of context that needs to be saved when making a context switch between any two tasks. The present invention is described in terms of its application to an operating system with single-stack, single-shot, run-to-completion task semantics though is generally applicable to other forms of tasking model.

Consider a system that makes use of any software or hardware resource which must "appear" to each task to be under its exclusive control (for example, a non-reentrant software floating point implementation; a set of hardware floating-point registers; or a set of digital signal processor accumulators). We refer to the collection of data that must be loaded and saved in order to give each task its own view of such resources as a "resource context", since this is additional to the normal task context.

In a conventional system, these resources would be saved on entry to any task and restored when the task exits. This means that for any system there would be an additional memory usage of one "resource context" per task in the system.

According to a first aspect of the present invention, there is provided a method of reducing memory or processor usage in an operating system that uses static priority- based scheduling for tasks, wherein a reduced number of resource contexts required by the operating system is calculated by determining which tasks make use of resource contexts that must be saved and restored and wherein wrappers or dispatchers are provided for tasks that must save and restore such resource contexts.

According to a second aspect of the present invention, there is provided a computing device having an operating system that uses static priority-based scheduling for tasks, the computing device being operable to calculate a reduced number of resource contexts required by the operating system by determining which tasks make use of resource contexts that must be saved and restored, and wherein the computing device provides wrappers or dispatchers for tasks that must save and restore such resource contexts. According to a third aspect of the present invention, there is provided a software product for use with a computing device having an operating system that uses static priority-based scheduling for tasks, the software product being operable to calculate a reduced number of resource contexts required by the operating system by determining which tasks make use of resource contexts that must be saved and restored, and wherein the software product provides wrappers or dispatchers for tasks that must save and restore such resource contexts.

Preferably, it is also determined which tasks will never run together (i.e. which tasks can be determined to be in mutual exclusion, thus forming a mutually exclusive non- pre-emption group of tasks), either by timing analysis or by schedulability analysis so as further to reduce the number of resource contexts required.

The software product may be in any appropriate form or language, and may be stored on a data carrier such as a floppy disk, optical disk, flash memory device or any other type of data carrier.

We shall first discuss the meaning of "wrapper" and "dispatcher" as used subsequently in the present application before discussing potential implementations of embodiments of the invention.

A "wrapper" in this application is defined to be a fragment of program code that encapsulates a task and is called in its place by the underlying operating system - for example, if we consider a task "task_l" whose "entry" (the function that carries out useful work) is defined as follows:

void task_l_entry(void)

{

/* do lots of useful work */ } we may under some circumstances (for example if loading or saving of context is required) wish to replace it with a wrapper:

void task_l_entry_wrapper (void)

{

' do_stuff_that_must_be_done_before_calling_task () task_l_entry () ; do_stu f_that_must_be_done_after_running_tas () ; }

A "dispatcher" performs essentially the same activities, but rather than the task entry function being replaced, the operating system itself is modified to make the pre- and post-task calls. For example, we might find the following code inside the operating system:

void os_dispatcher (void)

task *t = os_find_ ighest_priority_runnable_task() ; if (t->priority > os_current_task->priority) { do_stuff_that_must_be_done_before_calling_task () ; t->entry () ; do_stuf f_that_must_be_done_before_calling_task ( ) ;

} }

The description of this invention concentrates on the use of wrappers although an approach similar to this but using modified forms of the operating system dispatch function is also possible.

Let us consider an example of an unoptimised configuration of a system that uses task wrappers. Let us assume that a system is declared as follows (where higher numbers imply higher priority):

task a priority 1; task b priority 2; task c priority 3 task d priority 4 task e priority 5

Trivially, this system would require five "resource contexts" - one for each of tasks a, b, c, d and e to save their resources into when they are pre-empted by higher priority tasks. Each task would be protected by a "wrapper" that ensures that the "resource context" would be saved into the appropriate buffer and the operating system dispatcher would call this instead of the task entry function - for example, instead of jumping direct to "task_d_entry" the system dispatcher would call a wrapper:

resource_context re;

void task_d_entry_wrapper (void) /* Generated by config tool */

{ save_context (&rc) ; /* save context */ task_d_entry () ; /* call entry */ restore_context (&rc) ; /* restore context */

}

(in this case we show the wrapper being generated in a C-like language and including the declaration of the "resource context" itself- though this could also be generated in assembly language with appropriate code and data declarations, or could be brought about by calling different variants or branches of the operating system dispatcher function).

The system dispatcher in this case could use a table of function pointers to decide which function to call for each task, hi a hypothetical assembly language this might appear as below: task_entry_tab: PUBLIC

.pointer task_a_entry_wrapper .pointer task__b_entry_wrapper .pointer task_c_entry_ rapper .pointer task_d_entry_ rapper

.pointer task_e_entry_wrapper

But let us assume that only three of the tasks make use of resources that must be saved and restored - the system description is modified to:

task a priority 1 uses fp; task b priority 2 uses fp; task c priority 3 ; task d priority 4 uses fp; task e priority 5;

In this system, the "uses fp" declaration is added to indicate that a task uses the floating-point coprocessor. If a task does not explicitly mention that it uses something that requires a "resource context", then one is not provided (and no wrapper is needed - the OS can call the task's entry function directly).

A logical implication of the above is that the lowest-priority task that uses a shared resource need not provide a wrapper to load or save that context - since the pre- empter saves the context rather than the pre-emptee. The ultimate extension of this is that any system in which only one task uses a shared resource needs no wrapper or storage as there is nothing else that uses the resource to pre-empt it.

This means that only two resource contexts and wrappers need be generated (for task b and task d). In this case the dispatch table might look like this:

t a s k_ent ry_t ab : PUBL I C

.pointer task_a_entry # no need for wrapper - preemptor saves .pointer task_b_entry_wrapper # needs wrapper, can pre-empt a

.pointer task_c_entry # no need for wrapper- no FP .pointer task_d_entry_wrapper # needs wrapper, can pre-empt b

.pointer task_e_entry # no need for a wrapper

A further optimisation can be made to this. Suppose we have a mechanism for declaring that two tasks execute in mutual exclusion to each other (i.e. a for a pair of tasks x and y, even though x is of higher priority than y, x can never pre-empt y).

Consider this system description that places tasks a and b in mutual exclusion:

task a priority 1 uses fp; task b priority 2 uses fp; task c priority 3; task d priority 4 uses fp; task e priority 5; mutex a, b;

Given that tasks a and b are in mutual exclusion, we can therefore remove another resource context and wrapper - only task d needs to save and restore the "resource context" since it is the only higher-priority task that may affect the state of the shared resources, so in this case only task d would be generated with a wrapper and a "resource context".

It is to be noted that when tasks or the like share the same priority, they can be treated in the same way as if they are in mutual exclusion.

Consider also this variant: task a priority 1 uses fp; task b priority 2; task c priority 3 uses fp; task d priority 4 uses fp; task e priority 5; mutex c, d;

this system, it is clear that both tasks c and d need a wrapper to save the context, but since they are in mutual exclusion they can share the same resource context. The wrappers might be of the form:

resource context re c and d;

void task_c_entry_wrapper (void) /* Generated by config tool */

{ save_context (&rc_c_and_d) ; /* save context*/ task_c_entry () ; /* call entry */ restore_context (&rc_c_and_d) ; /* restore context */ }

void task_d_entry_wrapper (void) /* Generated by config tool */

{ save_context (&rc_c_and_d) ; /* save context*/ task_d_entry () ; /* call entry */ restore_context (&rc_c_and_d) ; /* restore context */ }

Although we discuss only tasks in this description, this scheme is uniformly extensible to Interrupt Service Routines provided that we can determine interrupt priority ordering - so any ISR that uses a particular resource would need to save and restore that resource if it could be interrupted by another higher-priority interrupt. (Interrupts at the same priority but at a different arbitration order would be considered to be in a mutual exclusion group).

An algorithm that could be used by an offline configuration tool for determining resource contexts and wrappers might be as follows:

global list of task Tasks; global list of mutex Mutexes;

-- Preconditions: Mutexes is a list of all mutexes in the system -- Tasks is a list of tasks sorted into priority order.

procedure Generate_Contexts is list of context Resource_Contexts = nil; list of wrapper Wrappers = nil;

-- Phase 1. Generate unique wrapper and context for every task -- that locks the resource. Point the context at the wrapper. for t in Tasks do if uses_shared_resource (t) then wrapper w = new Wrapper (t); append (w, Wrappers); context c = new Context (t) ; append (c, Contexts); wrapper_set_context (c,w) ; end if; end for;

-- Phase 2. Merge all contexts that can be merged, and -- amend the wrappers to point to these merged contexts. -- Delete any contexts that we don't use any more.

for t in Tasks do for u in Tasks do

-- merge if t & u wrapped but exist in a mutex if t <> u and exists_wrapper (Wrappers, t) and exists_wrapper (Wrappers, u) and exists_mutex(t,u) then -- get the wrappers for the tasks . w = get_wrapper (Wrappers, t) ; x = get_wrapper (Wrappers, u) ; -- get the contexts for the tasks. cl = get_context (t) ; c2 = get_context (u) ;

-- make a new context that combines both existing ones. c3 = new Context (merge_context (cl, c2) ) ; append (c3 , Contexts) ;

-- get rid of the old contexts. delete_contex (Contexts, cl) ; delete_context (Contexts, c2) ; -- point the wrappers at the new context. wrapper_set_context (w, c3) ; wrapper_set_context (x, c3) ; end if; end for; end for;

-- phase 3. The lowest-priority task using the resource -- AND ANYTHING IN A MUTEX WITH IT

-- (which means anything sharing the same context) will NOT need -- a wrapper or a context, w = head Wrappers; first_wrapped_task = get_task (w) ; c = get_context (first_wrapped_task) ; for t in tasks do w = get_wrapper (t) ; if get_context (t) = c then delete_wrapper (Wrappers, w) ; end if; end for; delete context (Contexts, c) ;

end procedure;

It will be appreciated that this algorithm is merely an example of a way to determine resource contexts; embodiments of the present invention may use this algorithm or any other appropriate algorithm or method that achieves the same or a similar effect.

A simple extension of this approach would allow each device to use multiple shared resources, with optimal wrappers being generated for each task based on the resource(s) it uses - for example consider the following system which has floating point registers and a digital signal processor (DSP) unit:

task a priority 1 uses fp; task b priority 2 uses dsp, fp: task c priority 3 uses fp; task d priority 4 uses dsp; task e priority 5 uses dsp, fp; task f priority 6 uses fp;

In this system:

Task a needs no wrapper (since the pre-empter saves the resources)

Task b needs a wrapper that saves only the "fp" context (it is the lowest priority task that uses the "dsp" context, so does not need to save it)

Task c needs a wrapper that saves only the "fp" context

Task d needs a wrapper that saves only the "dsp" context (it is the first preempting task to use it)

Task e needs a wrapper that saves both contexts

Task f needs a wrapper that saves only the "fp" context.

Such wrappers can be automatically generated from code skeletons provided by the user of the configuration tool.

According to a fourth aspect of the present invention, there is provided a method of integrating results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task or tasks with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

According to a fifth aspect of the present invention, there is provided a computing device adapted to integrate results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

According to a sixth aspect of the present invention, there is provided a software product for a computing device, the software product being adapted to integrate results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

h other words, the results of a timing analysis, which may be heuristic in nature, can be used in place of or in addition to results from a formal schedulability analysis so as to designate tasks as being effectively in mutual exclusion. In this way, duplication of resource contexts and/or wrappers is avoided.

It is to be noted that a timing analysis may be obtained by way of a timing analysis tool that applies techniques such as Rate Monotonic Analysis and extensions thereof in order to determine, among other things, when a given task might be executing, the latest time at which the given task might complete, and which other tasks can possibly pre-empt the given task.

Take for example the following simple periodic system: task tl priority 1 every 20 ticks duration 5 ticks offset 0 ticks uses fp; task t2 priority 2 every 20 ticks duration 5 ticks offset 2 ticks uses fp;

The execution profile of this system is shown in Figure 1 of the accompanying drawings. In this system, task t2 will always pre-empt task tl two "ticks" into its run, the system returning to task tl after t2 has finished running after five "ticks". Since both tasks share use of the floating point registers, task t2 must therefore load and save their context.

Consider, though, the following system:

task tl priority 1 every 20 ticks duration 5 ticks offset 0 ticks uses fp; task t2 priority 2 every 20 ticks duration 5 ticks offset 10 ticks uses fp;

The execution profile of this system is shown in Figure 2 of the accompanying drawings. In this system, task t2 never pre-empts task tl - it is therefore legitimate for a timing analysis tool to infer that the tasks can execute in mutual exclusion and by automatically adding the declaration

mutex tl , t2 ;

to the system declaration this can inhibit the generation of wrappers and context-save information for these two tasks.

The approach is extensible to any form of analysis that can demonstrate that for any pair of tasks x, y, there exists no condition in which the tasks can pre-empt each other. For example, a published prior art paper entitled: "How embedded applications using an RTOS can stay within on-chip memory limits" [Robert Davis, Nick Merriam, and Nigel Tracey; Proceedings of the Work in Progress and Industrial Experience Session, Euromicro Conference on Real-Time Systems, June 2000.] shows how schedulability analysis can be used to collect tasks into mutexes while ensuring that all task deadlines are met. It is simple to extend these techniques so that mutex creation minimises the combination of stack size and context save areas by grouping into mutexes those tasks which would otherwise require separate context save areas.

The preferred features of the invention are applicable to all aspects of the invention and may be used in any possible combination.

Throughout the description and claims of this specification, the words "comprise" and "contain" and variations of the words, for example "comprising" and "comprises", mean "including but not limited to", and are not intended to (and do not) exclude other components, integers, moieties, additives or steps.

Claims

CLAIMS:

1. A method of reducing memory or processor usage in an operating system that uses static priority-based scheduling for tasks, wherein a reduced number of resource contexts required by the operating system is calculated by determining which tasks make use of resource contexts that must be saved and restored and wherein wrappers or dispatchers are provided for tasks that must save and restore such resource contexts.

2. A method according to claim 1, wherein it is determined which tasks will never run together, either by way of timing analysis or schedulability analysis of the operating system description or by way of a timing analysis so as further to reduce the number of resource contexts required.

3. A computing device having an operating system that uses static priority-based schedulmg for tasks, the computing device being operable to calculate a reduced number of resource contexts required by the operating system by determining which tasks make use of resource contexts that must be saved and restored, and wherein the computing device provides wrappers or dispatchers for tasks that must save and restore such resource contexts.

4. A device as claimed in claim 3, the device being operable to determine which tasks will never run together, either by way of timing analysis or schedulability analysis so as further to reduce the number of resource contexts required.

5. A software product for use with a computing device having an operating system that uses static priority-based scheduling for tasks, the software product being operable to calculate a reduced number of resource contexts required by the operating system by determining which tasks make use of resource contexts that must be saved and restored, and wherein the software product provides wrappers or dispatchers for tasks that must save and restore such resource contexts.

6. A software product as claimed in claim 5, the software product being operable to determine which tasks will never run together, either by way of timing analysis or schedulability analysis so as further to reduce the number of resource contexts required.

7. A method of integrating results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

8. A computing device adapted to integrate results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

9. A software product for a computing device, the software product being adapted to integrate results from a timing analysis tool with an operating system configuration tool in such a way that when it is shown by the timing analysis tool that a particular task never pre-empts another task with which it contends for a resource, only a single resource context is generated, the single resource context being shared by the tasks.

10 A method according to any one of claims 1, 2 or 7, wherein the tasks include Interrupt Service Routines.

11. A device as claimed in any one of claims 3, 4 or 8, wherein the tasks include Interrupt Service Routines.

12. A software product as claimed in any one of claims 5, 6 or 9, wherein the tasks include Interrupt Service Routines.