OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [rtos/] [ecos-2.0/] [packages/] [services/] [memalloc/] [common/] [v2_0/] [doc/] [notes.txt] - Blame information for rev 174

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 27 unneback
                Memory allocation package - Implementation Notes
2
                ------------------------------------------------
3
 
4
 
5
 
6
Made with loving care by Jonathan Larmour (jlarmour@redhat.com)
7
Initial version: 2000-07-03
8
Last updated:    2000-07-03
9
 
10
 
11
 
12
Meta
13
----
14
 
15
This document describes some interesting bits and pieces about the memory
16
allocation package - CYGPKG_MEMALLOC. It is intended as a guide to
17
developers, not users. This isn't (yet) in formal documentation format,
18
and probably should be.
19
 
20
 
21
Philosophy
22
----------
23
 
24
The object of this package is to provide everything required for dynamic
25
memory allocation, some sample implementations, the ability to plug in
26
more implementations, and a standard malloc() style interface to those
27
allocators.
28
 
29
The classic Unix-style view of a heap is using brk()/sbrk() to extend the
30
data segment of the application. However this is inappropriate for an
31
embedded system because:
32
 
33
- you may not have an MMU, which means memory may be disjoint, thus breaking
34
  this paradigm
35
 
36
- in a single process system there is no need to play tricks since there
37
  is only the one address space and therefore heap area to use.
38
 
39
Therefore instead, we base the heap on the idea of fixed size memory pools.
40
The size of each pool is known in advance.
41
 
42
 
43
Overview
44
--------
45
 
46
Most of the infrastructure this package provides is geared towards
47
supporting the ISO standard malloc() family of functions. A "standard"
48
eCos allocator should be able to plug in to this infrastructure and
49
transparently work. The interface is based on simple use of C++ - nothing
50
too advanced.
51
 
52
The allocator to use is dictated by the
53
CYGBLD_MEMALLOC_MALLOC_IMPLEMENTATION_HEADER option. Choosing the
54
allocator can be done by ensuring the CDL for the new allocator
55
has a "requires" that sets the location of the header to use when that
56
allocator is enabled. New allocators should default to disabled, so they
57
don't have to worry about which one is the default, thus causing CDL
58
conflicts. When enabled the new allocator should also claim to implement
59
CYGINT_MEMALLOC_MALLOC_ALLOCATORS.
60
 
61
The implementation header file that is set must have a special property
62
though - it may be included with __MALLOC_IMPL_WANTED defined. If this
63
is the case, then this means the infrastructure wants to find out the
64
name of the class that is implemented in this header file. This is done
65
by setting CYGCLS_MEMALLOC_MALLOC_IMPL. If __MALLOC_IMPL_WANTED is defined
66
then no non-preprocessor output should be generated, as this will be included
67
in a TCL script in due course. An existing example from this package would
68
be:
69
 
70
#define CYGCLS_MEMALLOC_MALLOC_IMPL Cyg_Mempool_dlmalloc
71
 
72
// if the implementation is all that's required, don't output anything else
73
#ifndef __MALLOC_IMPL_WANTED
74
 
75
class Cyg_Mempool_dlmalloc
76
{
77
[etc.]
78
 
79
To meet the expectations of malloc, the class should have the following
80
public interfaces (for details it is best to look at some of the
81
examples in this package):
82
 
83
- a constructor taking arguments of the form:
84
 
85
  ALLOCATORNAME( cyg_uint8 *base, cyg_int32 size );
86
 
87
  If you want to be able to support other arguments for when accessing
88
  the allocator directly you can add them, but give them default values,
89
  or use overloading
90
 
91
- a destructor
92
 
93
- a try_alloc() function that returns new memory, or NULL on failure:
94
 
95
    cyg_uint8 *
96
    try_alloc( cyg_int32 size );
97
 
98
- a free() function taking one pointer argument that returns a boolean
99
  for success or failure:
100
 
101
    cyg_bool
102
    free( cyg_uint8 *ptr );
103
 
104
  Again, extra arguments can be added, as long as they are defaulted.
105
 
106
 
107
- resize_alloc() which is designed purely to support realloc(). It
108
  has the prototype:
109
    cyg_uint8 *
110
    resize_alloc( cyg_uint8 *alloc_ptr, cyg_int32 newsize,
111
                  cyg_int32 *oldsize );
112
 
113
  The idea is that if alloc_ptr can be adjusted to newsize, then it will
114
  be. If oldsize is non-NULL the old size (possibly rounded) is placed
115
  there. However what this *doesn't* do (unlike the real realloc()) is
116
  fall back to doing a new malloc(). All it does is try to do tricks
117
  inside the allocator. It's up to higher layers to call malloc().
118
 
119
- get_status() allows the retrieval of info from the allocator. The idea
120
  is to pass in the bitmask OR of the flags defined in common.hxx, which
121
  selects what information is requested. If the request is supported by
122
  the allocator, the approriate structure fields are filled in; otherwise
123
  unsupported fields will be left with the value -1. (The constructor for
124
  Cyg_Mempool_Status initializes them to -1). If you want to reinitialize
125
  the structure and deliberately lose the data in a Cyg_Mempool_Status
126
  object, you need to invoke the init() method of the status object to
127
  reinitialize it.
128
 
129
    void
130
    get_status( cyg_mempool_status_flag_t flags, Cyg_Mempool_Status &status );
131
 
132
  A subset of the available stats are exported via mallinfo()
133
 
134
 
135
Cyg_Mempolt2 template
136
---------------------
137
 
138
If using the eCos kernel with multiple threads accessing the allocators,
139
then obviously you need to be sure that the allocator is accessed in a
140
thread-safe way. The malloc() wrappers do not make any assumptions
141
about this. One helpful approach currently used by all the allocators
142
in this package is to (optionally) use a template (Cyg_Mempolt2) that
143
provides extra functions like a blocking alloc() that waits for memory
144
to be freed before returning, and a timed variant. Other calls are
145
generally passed straight through, but with the kernel scheduler locked
146
to prevent pre-emption.
147
 
148
You don't have to use this facility to fit into the infrastructure though,
149
and thread safety is not a prerequisite for the rest of the infrastructure.
150
And indeed certain allocators will be able to do scheduling at a finer
151
granularity than just locking the scheduler every time.
152
 
153
The odd name is because of an original desire to keep 8.3 filenames, which
154
was reflected in the class name to make it correspond to the filename.
155
There used to be an alternative Cyg_Mempoolt template, but that has fallen
156
into disuse and is no longer supported.
157
 
158
 
159
Automatic heap sizing
160
---------------------
161
 
162
This package contains infrastructure to allow the automatic definition
163
of memory pools that occupy all available memory. In order to do this
164
you must use the eCos Memory Layout Tool to define a user-defined section.
165
These sections *must* have the prefix "heap", for example "heap1", "heap2",
166
"heapdram" etc. otherwise they will be ignored.
167
 
168
The user-defined section may be of fixed size, or of unknown size. If it
169
has unknown size then its size is dictated by either the location of
170
the next following section with an absolute address, or if there are
171
no following sections, the end of the memory region. The latter should
172
be the norm.
173
 
174
If no user-defined sections starting with "heap" are found, a fallback
175
static array (i.e. allocated in the BSS) will be used, whose size can
176
be set in the configuration.
177
 
178
It is also possible to define multiple heap sections. This is
179
necessary when you have multiple disjoint memory regions, and no MMU
180
to join it up into one contiguous memory space. In which case
181
a special wrapper allocator object is automatically used. This object
182
is an instantiation of the Cyg_Mempool_Joined template class,
183
defined in memjoin.hxx. It is instantiated with a list of every heap
184
section, which it then records. It's sole purpose is to act as a go
185
between to the underlying implementation, and does the right thing by
186
using pointer addresses to determine which memory pool the pointer
187
allocator, and therefore which memory pool instantiation to use.
188
 
189
Obviously using the Cyg_Mempool_Joined class adds overhead, but if this
190
is a problem, then in that case you shouldn't define multiple disjoint
191
heaps!
192
 
193
 
194
Run-time heap sizing
195
--------------------
196
 
197
As a special case, some platforms support the addition of memory in the
198
field, in which case it is desirable to automatically make this
199
available to malloc. The mechanism for this is to define a macro in
200
the HAL, specifically, defined in hal_intr.h:
201
 
202
HAL_MEM_REAL_REGION_TOP( cyg_uint8 *regionend )
203
 
204
This macro takes the address of the "normal" end of the region. This
205
corresponds with the size of the memory region in the MLT, and would
206
be end of the "unexpanded" region. This makes sense because the memory
207
region must be determined by the "worst case" of what memory will be
208
installed.
209
 
210
This macro then returns a pointer which is the *real* region end,
211
as determined by the HAL at run-time.
212
 
213
By having the macro in this form, it is therefore flexible enough to
214
work with multiple memory regions.
215
 
216
There is an example in the ARM HAL - specifically the EBSA285.
217
 
218
 
219
How it works
220
------------
221
 
222
The MLT outputs macros providing information about user-defined sections
223
into a header file, available via system.h with the CYGHWR_MEMORY_LAYOUT_H
224
define. When the user-defined section has no known size, it determines
225
the size correctly relative to the end of the region, and sets the SIZE
226
macro accordingly.
227
 
228
A custom build rule preprocesses src/heapgen.cpp to generate heapgeninc.tcl
229
This contains TCL "set"s to allow access to the values of various
230
bits of configuration data. heapgen.cpp also includes the malloc
231
implementation header (as defined by
232
CYGBLD_MEMALLOC_MALLOC_IMPLEMENTATION_HEADER) with __MALLOC_IMPL_WANTED
233
defined. This tells the header that it should define the macro
234
CYGCLS_MEMALLOC_MALLOC_IMPL to be the name of the actual class. This
235
is then also exported with a TCL "set".
236
 
237
src/heapgen.tcl then includes heapgeninc.tcl which gives it access to
238
the configuration values. heapgen.tcl then searches the LDI file for
239
any sections beginning with "heap" (with possibly leading underscores).
240
It records each one it finds and then generates a file heaps.cxx in the
241
build tree to instantiate a memory pool object of the required class for
242
each heap. It also generates a list containing the addresses of each
243
pool that was instantiated. A header file heaps.hxx is then generated
244
that exports the number of pools, a reference to this list array and
245
includes the implementation header.
246
 
247
Custom build rules then copy the heaps.hxx into the include/pkgconf
248
subdir of the install tree, and compile the heaps.cxx.
249
 
250
To access the generated information, you must #include 
251
The number of heaps is given by the CYGMEM_HEAP_COUNT macro. The type of
252
the pools is given by CYGCLS_MEMALLOC_MALLOC_IMPL, and the array of
253
instantiated pools is available with cygmem_memalloc_heaps. For example,
254
here is a sample heaps.hxx:
255
 
256
#ifndef CYGONCE_PKGCONF_HEAPS_HXX
257
#define CYGONCE_PKGCONF_HEAPS_HXX
258
/*  */
259
 
260
/* This is a generated file - do not edit! */
261
 
262
#define CYGMEM_HEAP_COUNT 1
263
#include 
264
 
265
extern Cyg_Mempool_dlmalloc *cygmem_memalloc_heaps[ 2 ];
266
 
267
#endif
268
/* EOF  */
269
 
270
The array has size 2 because it consists of one pool, plus a terminating
271
NULL.
272
 
273
In future the addition of cdl_get() available from TCL scripts contained
274
within the CDL scripts will remove the need for a lot of this magic.
275
 
276
 
277
dlmalloc
278
--------
279
 
280
A port of dlmalloc is included. Far too many changes were required to make
281
it fit within the scheme above, so therefore there was no point
282
trying to preserve the layout to make it easier to merge in new versions.
283
However dlmalloc rarely changes any more - it is very stable.
284
 
285
The version of dlmalloc used was a mixture of 2.6.6 and the dlmalloc from
286
newlib (based on 2.6.4). In the event, most of the patches merged were
287
of no consequence to the final version.
288
 
289
For reference, the various versions examined are included in the
290
doc/dlmalloc subdirectory: dlmalloc-2.6.4.c, dlmalloc-2.6.6.c,
291
dlmalloc-newlib.c and dlmalloc-merged.c (which is the result of merging
292
the changes between 2.6.4 and the newlib version into 2.6.6). Note it
293
was not tested at that point.
294
 
295
 
296
Remaining issues
297
----------------
298
 
299
You should be allowed to have different allocators for different memory
300
regions. The biggest hurdle here is host tools support to express this.
301
 
302
Currently the "joined" allocator wrapper simply treats each memory pool
303
as an equal. It doesn't understand that some memory pools may be faster
304
than others, and cannot make decisions about which pools (and therefore
305
regions and therefore possibly speeds of memory) to use on the basis
306
of allocation size. This should be (configurably) possible.
307
 
308
 
309
History
310
-------
311
 
312
 
313
A long, long time ago, in a galaxy far far away.... the situation used to
314
be that the kernel package contained the fixed block and simple variable
315
block memory allocators, and those were the only memory allocator
316
implementations. This was all a bit incongruous as it meant that any code
317
wanting dynamic memory allocation had to include the whole kernel, even
318
though the dependencies could be encapsulated. This was particularly silly
319
because the implementation of malloc() (etc.) in the C library didn't use
320
any of the features that *did* depend on the kernel, such as timed waits
321
while allocating memory, etc.
322
 
323
The C library malloc was pretty naff then too. It used a static buffer
324
as the basis of the memory pool, with a hard-coded size, set in the
325
configuration. You couldn't make it fit into all of memory.
326
 
327
Jifl
328
2000-07-03
329
 
330
//####ECOSGPLCOPYRIGHTBEGIN####
331
// -------------------------------------------
332
// This file is part of eCos, the Embedded Configurable Operating System.
333
// Copyright (C) 1998, 1999, 2000, 2001, 2002 Red Hat, Inc.
334
//
335
// eCos is free software; you can redistribute it and/or modify it under
336
// the terms of the GNU General Public License as published by the Free
337
// Software Foundation; either version 2 or (at your option) any later version.
338
//
339
// eCos is distributed in the hope that it will be useful, but WITHOUT ANY
340
// WARRANTY; without even the implied warranty of MERCHANTABILITY or
341
// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
342
// for more details.
343
//
344
// You should have received a copy of the GNU General Public License along
345
// with eCos; if not, write to the Free Software Foundation, Inc.,
346
// 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
347
//
348
// As a special exception, if other files instantiate templates or use macros
349
// or inline functions from this file, or you compile this file and link it
350
// with other works to produce a work based on this file, this file does not
351
// by itself cause the resulting work to be covered by the GNU General Public
352
// License. However the source code for this file must still be made available
353
// in accordance with section (3) of the GNU General Public License.
354
//
355
// This exception does not invalidate any other reasons why a work based on
356
// this file might be covered by the GNU General Public License.
357
//
358
// Alternative licenses for eCos may be arranged by contacting Red Hat, Inc.
359
// at http://sources.redhat.com/ecos/ecos-license/
360
// -------------------------------------------
361
//####ECOSGPLCOPYRIGHTEND####

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.