summaryrefslogtreecommitdiffstats
path: root/doc/user/bsp.t
blob: f1350c0c96d6c6b1e9c824b9c772032c4218693b (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
@c
@c  COPYRIGHT (c) 1988-1998.
@c  On-Line Applications Research Corporation (OAR).
@c  All rights reserved.
@c
@c  $Id$
@c

@chapter Board Support Packages

@section Introduction

A board support package (BSP) is a collection of
user-provided facilities which interface RTEMS and an
application with a specific hardware platform.  These facilities
may  include hardware initialization, device drivers, user
extensions, and a Multiprocessor Communications Interface
(MPCI).  However, a minimal BSP need only support processor
reset and initialization and, if needed, a clock tick.

@section Reset and Initialization

An RTEMS based application is initiated or
re-initiated when the processor is reset.  This initialization
code is responsible for preparing the target platform for the
RTEMS application.  Although the exact actions performed by the
initialization code are highly processor and target dependent,
the logical functionality of these actions are similar across a
variety of processors and target platforms.

Normally, the application's initialization is
performed at two separate times:  before the call to
@code{@value{DIRPREFIX}initialize_executive}
(reset application initialization) and
after @code{@value{DIRPREFIX}initialize_executive}
in the user's initialization tasks
(local and global application initialization).  The order of the
startup procedure is as follows:

@enumerate
@item Reset application initialization.
@item Call to @code{@value{DIRPREFIX}initialize_executive}
@item Local and global application initialization.
@end enumerate

The reset application initialization code is executed
first when the processor is reset.  All of the hardware must be
initialized to a quiescent state by this software before
initializing RTEMS.  When in quiescent state, devices do not
generate any interrupts or require any servicing by the
application.  Some of the hardware components may be initialized
in this code as well as any application initialization that does
not involve calls to RTEMS directives.

The processor's Interrupt Vector Table which will be
used by the application may need to be set to the required value
by the reset application initialization code.  Because
interrupts are enabled automatically by RTEMS as part of the
@code{@value{DIRPREFIX}initialize_executive} directive,
the Interrupt Vector Table MUST
be set before this directive is invoked to insure correct
interrupt vectoring.  The processor's Interrupt Vector Table
must be accessible by RTEMS as it will be modified by the
@code{@value{DIRPREFIX}interrupt_catch} directive.  
On some CPUs, RTEMS installs it's
own Interrupt Vector Table as part of initialization and thus
these requirements are met automatically.  The reset code which
is executed before the call to @code{@value{DIRPREFIX}initialize_executive}
has the following requirements:

@itemize @bullet
@item Must not make any RTEMS directive calls.

@item If the processor supports multiple privilege levels,
must leave the processor in the most privileged, or supervisory,
state.

@item Must allocate a stack of at least @code{@value{RPREFIX}MINIMUM_STACK_SIZE}
bytes and initialize the stack pointer for the
@code{@value{DIRPREFIX}initialize_executive} directive.

@item Must initialize the processor's Interrupt Vector Table.

@item Must disable all maskable interrupts.

@item If the processor supports a separate interrupt stack,
must allocate the interrupt stack and initialize the interrupt
stack pointer.
@end itemize

The @code{@value{DIRPREFIX}initialize_executive} directive does not return to
the initialization code, but causes the highest priority
initialization task to begin execution.  Initialization tasks
are used to perform both local and global application
initialization which is dependent on RTEMS facilities.  The user
initialization task facility is typically used to create the
application's set of tasks.

@subsection Interrupt Stack Requirements

The worst-case stack usage by interrupt service
routines must be taken into account when designing an
application.  If the processor supports interrupt nesting, the
stack usage must include the deepest nest level.  The worst-case
stack usage must account for the following requirements:

@itemize @bullet
@item Processor's interrupt stack frame

@item Processor's subroutine call stack frame

@item RTEMS system calls

@item Registers saved on stack

@item Application subroutine calls
@end itemize

The size of the interrupt stack must be greater than
or equal to the constant @code{@value{RPREFIX}MINIMUM_STACK_SIZE}.

@subsection Processors with a Separate Interrupt Stack

Some processors support a separate stack for
interrupts.  When an interrupt is vectored and the interrupt is
not nested, the processor will automatically switch from the
current stack to the interrupt stack.  The size of this stack is
based solely on the worst-case stack usage by interrupt service
routines.

The dedicated interrupt stack for the entire
application is supplied and initialized by the reset and
initialization code of the user's board support package.  Since
all ISRs use this stack, the stack size must take into account
the worst case stack usage by any combination of nested ISRs.

@subsection Processors without a Separate Interrupt Stack

Some processors do not support a separate stack for
interrupts.  In this case, without special assistance every
task's stack must include enough space to handle the task's
worst-case stack usage as well as the worst-case interrupt stack
usage.  This is necessary because the worst-case interrupt
nesting could occur while any task is executing.

On many processors without dedicated hardware managed
interrupt stacks, RTEMS manages a dedicated interrupt stack in
software.  If this capability is supported on a CPU, then it is
logically equivalent to the processor supporting a separate
interrupt stack in hardware.

@section Device Drivers

Device drivers consist of control software for
special peripheral devices and provide a logical interface for
the application developer.  The RTEMS I/O manager provides
directives which allow applications to access these device
drivers in a consistent fashion.  A Board Support Package may
include device drivers to access the hardware on the target
platform.  These devices typically include serial and parallel
ports, counter/timer peripherals, real-time clocks, disk
interfaces, and network controllers.

For more information on device drivers, refer to the
I/O Manager chapter.

@subsection Clock Tick Device Driver

Most RTEMS applications will include a clock tick
device driver which invokes the @code{@value{DIRPREFIX}clock_tick}
directive at regular intervals.  The clock tick is necessary if
the application is to utilize timeslicing, the clock manager, the
timer manager, the rate monotonic manager, or the timeout option on blocking
directives.

The clock tick is usually provided as an interrupt
from a counter/timer or a real-time clock device.  When a
counter/timer is used to provide the clock tick, the device is
typically programmed to operate in continuous mode.  This mode
selection causes the device to automatically reload the initial
count and continue the countdown without programmer
intervention.  This reduces the overhead required to manipulate
the counter/timer in the clock tick ISR and increases the
accuracy of tick occurrences.  The initial count can be based on
the microseconds_per_tick field in the RTEMS Configuration
Table.  An alternate approach is to set the initial count for a
fixed time period (such as one millisecond) and have the ISR
invoke @code{@value{DIRPREFIX}clock_tick}
on the microseconds_per_tick boundaries.
Obviously, this can induce some error if the configured
microseconds_per_tick is not evenly divisible by the chosen
clock interrupt quantum.

It is important to note that the interval between
clock ticks directly impacts the granularity of RTEMS timing
operations.  In addition, the frequency of clock ticks is an
important factor in the overall level of system overhead.  A
high clock tick frequency results in less processor time being
available for task execution due to the increased number of
clock tick ISRs.

@section User Extensions

RTEMS allows the application developer to augment
selected features by invoking user-supplied extension routines
when the following system events occur:

@itemize @bullet
@item Task creation
@item Task initiation
@item Task reinitiation
@item Task deletion
@item Task context switch
@item Post task context switch
@item Task begin
@item Task exits
@item Fatal error detection
@end itemize

User extensions can be used to implement a wide variety of
functions including execution profiling, non-standard
coprocessor support, debug support, and error detection and
recovery.  For example, the context of a non-standard numeric
coprocessor may be maintained via the user extensions.  In this
example, the task creation and deletion extensions are
responsible for allocating and deallocating the context area,
the task initiation and reinitiation extensions would be
responsible for priming the context area, and the task context
switch extension would save and restore the context of the
device.

For more information on user extensions, refer to the
User Extensions chapter.

@section Multiprocessor Communications Interface (MPCI)

RTEMS requires that an MPCI layer be provided when a
multiple node application is developed.  This MPCI layer must
provide an efficient and reliable communications mechanism
between the multiple nodes.  Tasks on different nodes
communicate and synchronize with one another via the MPCI.  Each
MPCI layer must be tailored to support the architecture of the
target platform.

For more information on the MPCI, refer to the
Multiprocessing Manager chapter.

@subsection Tightly-Coupled Systems

A tightly-coupled system is a multiprocessor
configuration in which the processors communicate solely via
shared global memory.  The MPCI can simply place the RTEMS
packets in the shared memory space.  The two primary
considerations when designing an MPCI for a tightly-coupled
system are data consistency and informing another node of a
packet.

The data consistency problem may be solved using
atomic "test and set" operations to provide a "lock" in the
shared memory.  It is important to minimize the length of time
any particular processor locks a shared data structure.

The problem of informing another node of a packet can
be addressed using one of two techniques.  The first technique
is to use an interprocessor interrupt capability to cause an
interrupt on the receiving node.  This technique requires that
special support hardware be provided by either the processor
itself or the target platform.  The second technique is to have
a node poll for arrival of packets.  The drawback to this
technique is the overhead associated with polling.

@subsection Loosely-Coupled Systems

A loosely-coupled system is a multiprocessor
configuration in which the processors communicate via some type
of communications link which is not shared global memory.  The
MPCI sends the RTEMS packets across the communications link to
the destination node.  The characteristics of the communications
link vary widely and have a significant impact on the MPCI
layer.  For example, the bandwidth of the communications link
has an obvious impact on the maximum MPCI throughput.

The characteristics of a shared network, such as
Ethernet, lend themselves to supporting an MPCI layer.  These
networks provide both the point-to-point and broadcast
capabilities which are expected by RTEMS.

@subsection Systems with Mixed Coupling

A mixed-coupling system is a multiprocessor
configuration in which the processors communicate via both
shared memory and communications links.  A unique characteristic
of mixed-coupling systems is that a node may not have access to
all communication methods.  There may be multiple shared memory
areas and communication links.  Therefore, one of the primary
functions of the MPCI layer is to efficiently route RTEMS
packets between nodes.  This routing may be based on numerous
algorithms. In addition, the router may provide alternate
communications paths in the event of an overload or a partial
failure.

@subsection Heterogeneous Systems

Designing an MPCI layer for a heterogeneous system
requires special considerations by the developer.  RTEMS is
designed to eliminate many of the problems associated with
sharing data in a heterogeneous environment.  The MPCI layer
need only address the representation of thirty-two (32) bit
unsigned quantities.

For more information on supporting a heterogeneous
system, refer the Supporting Heterogeneous Environments in the
Multiprocessing Manager chapter.