-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathfileformat.html
584 lines (560 loc) · 23 KB
/
fileformat.html
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
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
<html><head><title>6502 binary relocation format</title>
<link rev=made href="mailto:[email protected]">
</head>
<body bgcolor="#ffffff">
<h1>6502 binary relocation format</h1>
<h2>V1.2 as of 27jan1998</h2>
<h3>(c) André Fachat ([email protected])</h3>
<hr>
<h3>Changes from V1.1</h3>
<p>
The <a href="#change12a">order</a> for saving the undefined reference and the low byte of a
high byte relocation entry has changed. This makes the OS/A65 lib6502
implementation easier. How many other people use this format anyway...?
</p>
<hr>
<p>
<h3>0) Preface</h3>
<p>
With some new 6502/C64/C128 operating systems comes the need for a new
binary format. In multitasking operating systems like Lunix, SMOS, or
OS/A65, a binary file cannot be loaded to a fixed location that might
already be used by another program. Therefore it must be possible to
relocate the program to an arbitrary address at least at load time.
In addition to that, more specific information might be stored in a
binary executable file, like interrupt vectors for example.
<p>
This text gives a good solution to this problem for the 6502 CPU and an
assembler source format to use this format in a general manner. The file
format can even be used as an object file format, i.e. a format a linker
can use as an input file. It is also usable as a 65816 file format.
Instead of zeropage addressing modes, the 65816 has direct addressing
modes, that add the contents of the direct register to the zeropage
address in the opcode.
<p>
<h3>1) 6502/65816 specifics</h3>
<p>
The 6502 has the special feature of a 'zeropage', i.e. a very limited
memory address range used for special addressing modes. So the format
should not only provide a means to relocate absolute addresses but also
zeropage addresses. The 65816 replaces zeropage addressing with direct
addressing modes.
<p>
The stack space is also very limited. A binary format has to provide a
measure of how much stack space is needed for the application.
<p>
Such limits should be defined as 2 byte values, even if the 6502 only has
a range of 8 address bits for zeropage and stack. But the 65816 behaves
differently, it has a 16 bit stack pointer for example. For further
expandability, a 32 bit format should be provided, although the 16 bit
format suffices for the 65816 already.
<p>
Another problem is, that an address can be 'split', i.e. you can just use
the high byte or the low byte separately in an opcode. This gives need
to a special relocation table format, that can cope with half-address
references. The 65816 can even have three byte addresses, i.e. address
in a segment and segment number.
<p>
<h3>2) binary format</h3>
<h4>2.1) General</h4>
<p>
The file differs from the known Commodore file formats, in that a lot
more information is stored in the file. First the data is structured
in separate segments to allow different handling of text (program code),
data (like tables) and bss (uninitialized data).
<p>
Also tables are included to allow late binding, i.e. linking the
file with other files at load time, and relocation, i.e. executing
the file at different addresses in 6502 address space.
<p>
<h4>2.2) Segments</h4>
<p>
As already used in other formats, the assembler uses three different
segment types, i.e. text (the actual program code), data (initialized
variables), and bss (uninitialized variables).
To have these different segments seems to be 'overdesigned', but they
actually make memory handling easier in more complex operating systems
on systems with virtual addresses (OS/A65, for example).
<p>
The text segment is defined to be read-only memory. This doesn't allow
self-modifying code in this segment, but allows memory sharing in virtual
memory architectures. The data segment actually is like the text segment,
only it is allocated writable. This segment might not be shared between
different processes. The contents of these two segments are loaded from
the file. The bss segment is uninitialized data, i.e. upon program start,
it is not defined - and not loaded from the file. This area is read-write
and can be used during program execution. It is also not shared between
processes. In addition to these segments, the 6502 format also includes a
zeropage segment type, to allow zeropage variables to be relocated. This
zeropage segment is like a bss segment, in that only the length, but
not the data is saved. For the 65816 the zeropage segment changes its
meaning to a bank zero segment.
<p>
The different segments hold different type of data and can be located
anywhere in memory (except zero segment, which has to be in the zeropage
resp. bank zero). The program must therefore not assume anything about
the relative addresses between different segments.
<p>
<h4>2.3) Relocation</h4>
<p>
In general, there are three ways to handle the relocation problem so far:
<pre>
- Tables: have a relocation table for a text segment
if the relocation table is put in front of code
you have to save the table in a side-storage
if table is behind, you still cannot relocate 'on the fly'.
- Deassembling: go through the code, deassemble it and change all absolute
addresses. Problem: needs to know or have hints about where some
data is in the code.
- Relocation info in the code: here each address is preceeded with an
'escape' code and is relocated when loading. But this disallows block
oriented transfer from storage media to memory.
</pre>
This binary format uses the first method, with the table after the
code/data. This way block oriented transfer for the text/data segment can
be used. And while reading the relocation tables bytewise, the relocation
can be done without the need to save the table somewhere.
<p>
<h4>2.4) External References & Exported Globals</h4>
<p>
As this file format should not only be used as an executable format, but
also as object file format, it must provide a way to define references
- references exported from this object and labels referenced in this
object. The external references list (also called 'undefined list') lists
the addresses where labels not defined in this object are referenced.
The exported globals list lists the addresses that are available for
other objects. The labels are named by null-terminated ASCII strings.
<p>
Even an executable file can have non-empty globals and externals lists,
but only if the operating system allows this. In this case, so called
'late binding' is used to link the object with some global libraries
at link time.
<p>
<h4>2.5) File extension</h4>
<p>
The proposed standard extension for the described format is ".o65" when
used as an object file.
<p>
<h4>2.6) Format description</h4>
<p>
The binary format is the following:
<pre>
(
header
text segment
data segment
external references list
relocation table for text segment
relocation table for data segment
exported globals list
)
</pre>
The description of the parts follows:
<p>
<h5>2.6.1) Header</h5>
<p>
The header contains the minimum needed data in a fixed struct.
The rest of the necessary information is put into the header options.
[Note: .word is a 16 bit value, low byte first, .byt is a simple byte.
.long is a 32 bit value, low byte first. .size is a 16 or 32 bit
value according to .word and .long, depending on the size bit in the
mode field ]
<p>
This is the fixed struct:
<pre>
(
.byt $01,$00 ; non-C64 marker
.asc "o65" ; o65 MAGIC!
.byt 0 ; version
.word mode ; mode word
.size tbase ; address to which text is assembled to
; originally
.size tlen ; length of text segment
.size dbase ; originating address for data segment
.size dlen ; length of data segment
.size bbase ; originating address for bss segment
.size blen ; length of bss segment
.size zbase ; originating address for zero segment
.size zlen ; length of zero segment
.size stack ; minimum needed stack size, 0= not known.
; the OS should add reasonable values for
; interrupt handling before allocating
; stack space
)
</pre>
The mode word currently has these defined bits:
<pre>
mode.15 : CPU 0= 6502 1= 65816
mode.14 : reloc 0= bytewise... 1= page(256byte)wise relocation
allowed
mode.13 : size 0= size=16 bit, 1= size=32 bit
mode.12 : obj 0= executable 1= object file
mode.0-1: align 0= byte align,
1= word (i.e. 2 byte) align
2= long (4 byte) align
3= block (256 byte) align
</pre>
The CPU bit tells the loader for which CPU the file was made. This has
implications on the zero segment, for example. Also a system can check
if the program will run at all (on a 6502 that is).
The reloc bit defines if an object file can be relocated bytewise,
or if it must be page-aligned. A page has 256 bytes. The restriction to
pagewise relocation simplifies the relocation table and also allows
simpler compilers/assemblers.
The size bit determines the size of the segment base address and length
entries. Currently the 16 bit size (size bit = 0) works for 6502 and
65816 CPUs.
<p>
The obj bit distinguishes between object files and executables.
An object file is used as assembler output that can be linked with
other object files to build an executable or an object library.
The two align bits give the address boundary the segments can be
placed. Even the 6502 needs this, as, for example, "jmp ($xxFF)" is
broken. The align bits are valid for all of the segments.
[Note: if reloc=1, then align should be 3. But if align=3, reloc need
not be 1, because reloc switches to a simpler version of the relocation
table. The reloc bit might be obsoleted in newer versions of this
format. Though it should be set, if necessary.]
<p>
All unused bits in the mode field must be zero.
<p>
Note that the header size is 26 if the size bit is zero and 44 if the
size bit is one.
<p>
The fixed sized struct is immediately followed by a list of header options.
Each header option consists of a single byte total length, a type byte
and some data bytes if needed. A single length byte of $00 ends the
header option list.
<pre>
(
{ ; optional options, more than one allowed
.byt olen ; overall length (including length and type
; byte
.byt otype ; option type
[ .byt option_bytes ]
}
.byt $00 ; end of options marker (i.e. option len=0)
)
</pre>
The header options currently defined/proposed are:
<pre>
- Filename:
type=0; len=strlen(filename_in_ascii)+3; content="filename_in_ascii",0
The string contains the name of the object.
- Operating System Header
type=1; len=?
the first data byte is the OS type:
1 OSA/65 header supplement
2 Lunix header supplement
[others to follow?]
the following data contains OS specific information.
A suggested data byte is the OS version as second byte.
- Assemblerprogram:
type=2; len=strlen(ass)+3; content="ass",0
The string contains the name of the assembler resp. linker that produced
this file/object.
For example (syntax see below)
.fopt 2, "xa 2.1.1g",0
becomes
0c 02 78 61 20 32 2e 31 2e 31 67 00
in the file.
- Author:
type=3; len=strlen(author)+3; content="author",0
The string contains the author of the file.
- Creation data:
type=4; len=strlen(date)+3; content="date_string",0
The string contains the creation date in format like:
"Sat Dec 21 14:00:23 MET 1996", where we have the day, Month, date,
time, timezone and year. See output of `date`...
</pre>
<h5>2.6.2) text and data segments</h5>
<p>
The text and data segments are just the assembled code.
The only difference between text and data segments is the read/write mode
of the two segments. Therefore, to be compliant to this file format,
self-modifying code goes into the data segment.
<p>
<h5>2.6.3) Undefined references list</h5>
<p>
The next list is an ASCII list of labels that are referenced in this file
but not defined. The lists is preceeded with the number of undefined labels
(16 or 32 bits, according to the mode.size bit).
<pre>
undef_list: number_of_undefined_labels.s
"undefined_label1",0
"undefined_label2",0
...
</pre>
<h5>2.6.4) Relocation tables</h5>
<p>
The relocation tables are the same format for the two segments, text and
data. In general a relocation entry consists of the offset from the
previous relocation address to the next one, the type of the relocation
and additional info. Relocation not only defines the relocation when
moving object code to a different address, but also filling in the
undefined references.
<p>
Each table starts at relocation address = segment base address -1.
I.e. if the segment base address is $1000 for example, the first entry
has an offset computed from base address-1 = $0fff.
The offset to the next relocation address is the first byte of each
entry. If the offset is larger than 254 (i.e. 255 or above), than a
255 is set as offset byte, the offset is decremented by 254 (note the
difference) and the entry is started again.
<pre>
{ [255,...,255,] offset of next relocation (b), typebyte|segmentID [, low_byte] }+
</pre>
where typebyte has the bits 5, 6 and 7 and is one of
<pre>
WORD $80 2 byte address
HIGH $40 high byte of an address
LOW $20 low byte of an address
SEGADR $c0 3 byte address (65816)
SEG $a0 segment byte of 3 byte address
</pre>
The segmentID stands for the segment the reference points to:
<pre>
0 undefined
1 absolute value
2 text segment
3 data segment
4 bss segment
5 zero segment
</pre>
(Of course the absolute value will never appear in a relocation table,
but this value is necessary for the exported list)
<p>
If the type is HIGH, the low byte of the value is stored behind the
relocation table entry, if bytewise relocation is allowed (header mode
field bit 14). If only pagewise relocation is allowed, then only HIGH
relocation entries can occur, and the low byte is implicitely set zero
(i.e. it is _not_ saved in the relocation table).
<p>
If the type is SEG, then the two lower bytes of the three byte segment
address are stored behind the entry in the relocation table, lower byte
first.
<p>
<!---
If the segment is "undefined", then the whole relocation entry is followed
by a two (mode.size=0) or four (mode.size=1) byte value index in the
undefined references list.
--->
<a name="change12a"></a>
If the segment is "undefined", the typebyte is immediately followed
by the two (mode size=0) or four (mode size=1) byte value index
in the undefined references list. If it is a high byte relocation,
the low byte is saved behind the index value. The index value
determines the undefined reference, which must be looked up by the
loader.
<p>
The value taken from the relocation address in the segment, together with
the low byte from the relocation table (if HIGH entry) form the address
used if the segment would be used unrelocated. To relocate the segment,
the difference between the relocated segment base address and the segment
base address from the file is then added to the above address. The result
is again saved in the segment.
<p>
A zero offset byte ends the relocation table. The first offset is computed
from the segment base address-1, to avoid a 0 value in the first entry.
<p>
Note that direct addressing modes do not generate entries in the
relocation table. instead it is assumed that the 65816 direct register
holds the correct value (i.e. zero segment base address) when running
this program.
<p>
Example:
<p>
Segment Base address in file (header.tbase) is $1000.
The start address of the text segment after relocation is real.tbase = $1234.
<p>
Now the first (unrelocated) address at which a relocation should take
place is here:
<pre>
$1222 A9 23 lda #>vector
</pre>
This generates the offset: $1222-($1000-1) = $223. This is larger than
254 ($fe), so the first byte is 255 ($ff). The offset is decremented
by $fe, and gives $125. This again is larger than $fe, so the next byte
is $ff again. After substracting $fe again, we have $27. But this is
the address of the opcode. To get the address of the address byte, we
have to add 1 to get $28, which becomes the third byte.
The next offset is then computed from $1223, because this is the
last relocation address.
<p>
Now we reference the high byte of an address, lets say vector=$23d0 (not
relocated), in the text segment. Therefore the relocation type becomes
'HIGH | text_segmentID = $42', which is the next byte. Because we are
referencing a high byte of an address, the low byte of the unrelocated
address is saved behind the typebyte in the relocation entry. This byte
is missing when referencing a low byte or address.
<p>
The relocation table entry is now:
<pre>
$ff, $ff, $28, $42, $d0.
</pre>
When actually doing the relocation, the relocation pointer is initialized
to real.tbase-1 = $1233. Then we compute the offset to $224, which brings
us to $1457, where the address byte of the above opcode is after loading
the file to $1234. We now have to compute the new address, where vector
is after relocation. So we take the unrelocated low byte from the
relocation table ($d0) and the high byte from $1457 ($23).
<pre>
vector_file = ($23 << 8) + $c0
</pre>
To this value we add
the difference between the address the program is assembled to and the
real load address:
<pre>
vector_relocated = vector_file + (real.tbase - header.tbase)
= $23d0 + ($1234 - $1000)
= $23d0 + $234
= $2604
</pre>
From this value the high byte is then written back to the address $1457.
Had we not saved the low byte in the relocation table, and only added
the high bytes, we would have missed the carry bit that increments
the high byte in this case!
<p>
Had "vector" now been an undefined reference, and "vector" would be
the second label in the undefined references list, we would get the
following relocation table entry (assuming mode.size=0):
<pre>
$ff, $ff, $28, $40, $02, $00, $00
</pre>
The value computed with the above formula for vector_file is now added
to the address the label "vector" now really has (This must of course
be looked up into an external table or list).
Had the opcode been "LDA #>vector+$567", then the low byte in the relocation
table would be $67, while the high byte in the opcode would be $05.
This value would result in vector_file and the real address of "vector"
would be added before wrting back the high byte to the opcode.
<p>
<h5>2.6.5) exported globals list</h5>
<p>
The global list is a list of names, together with the target segment
and the offset in the segment for each name. It is preceeded with the
number of exported labels. This allows the loader to allocate a table
large enough, if needed. The number of labels and the offset value
are 16 bit or 32 bit values according to the size bit in the header mode
field. The segmentID is a byte value and the same as in the relocation
table entry (see section 2.6.3).
<pre>
number_of_exported_labels.s
"global_label_name_in_asc1",0, segmentID.b, value.s
...
</pre>
<h3>3) assembler source format</h3>
<p>
The assembler source format is a suggestion only. It will be implemented
in xa65, a cross assembler for 6502 CPUs running on Unix/Atari ST/Amiga
as a reference platform.
<p>
The assembler provides a way to embed absolute address code in relocatable
code. This is needed when code should be copied to a specific location
known at assemble time.
There also is a way to make a file 'romable'. You can give the start
address of the _file_ in ROM, and the assembler automatically sets
the text segment start address to where the code will be in the ROM.
Of course, the other segments must be taken care of with -b? command
line parameter, that set the segment start address.
<p>
<h4>3.1) embed absolute code in relocatable files</h4>
<p>
When the assembler is started in relocatable mode, everything is put into
a .o65 relocatable file. All address references generate relocation table
entries. If a "*= value" pseudo opcode is encountered,
then the assembler switches to absolute mode. The following opcodes don't
generate relocation table entries. If a "*=" without a value is read,
then the assembler switches back to relocatable mode. The relocation
program counter is increased with the length of the absolute part and
the absolute code is embedded between the relocatable parts.
<p>
<h4>3.2) embed relocatable code in absolute files</h4>
<p>
This is dropped - too complicated. Should better be done with some
objdump or linker programs or so.
<p>
<h4>3.2) Header options</h4>
<p>
Before any opcode (after starting in relocatable mode, or after a .reloc
opcode), a header option can be set by:
<pre>
.fopt byte1, byte2, ...
</pre>
The header option length is automatically set by the assembler.
An example for an file author entry:
<pre>
.fopt 3, "Andre Fachat",0
</pre>
The 3 is the type byte for the author header option. The last zero ends
the name. The assembler can be configured to automatically include an
assembler header option into a file header.
<p>
<h4>3.3) allocation of data segment/zeropage segment address space</h4>
<p>
The assembler switches between the different segments by the means of
".text", ".data", ".bss" and ".zero" pseudo opcodes. After starting in
relocatable mode, the assembler is in the text segment.
<p>
The text segment contains the program code. Data holds the initialized data,
while bss and zero segments contain uninitialized data for normal/zeropage
address space.
Everything that is between one of these segment opcodes and the next segment
opcode gets into the corresponding segment, i.e. labels, assembled code etc.
The text and data segments are saved in the file, while for the bss and
zero segments only the length is saved in the file.
<p>
The assembler should issue a warning when a direct addressing mode
is used without a zero segment address and vice versa for 65816 CPUs.
<p>
<h4>3.4) referencing data/bss/zeropage addresses</h4>
<p>
One problem with the 6502 is, that it cannot load an address within one
step or assembler opcode. So an address is loaded with standard byte
opcodes, like "lda #<label". But how do we decide, whether "label"
is an address or not, and what do we if we get something like
"lda #zp_label + 12 * label2"?
<p>
The assembler is now intelligent enough to evaluate such expressions
and check for:
<pre>
- no address label : ok, absolute
- one address label, only add to label : ok, relocate
- difference between two addresses : If addresses in same segment, compute
diff and set absolute, otherwise bail
- everything else : warning
</pre>
This way there is no change in syntax. Address labels are distinguished
by using the "label:" syntax, as opposed to "label = value".
Also, if the assembler is capable of doing so, an address label may be
defined by "label opcode", i.e. without a colon.
<p>
<h4>3.5) aligning code</h4>
<p>
The 6502 has the problem that some opcodes (e.g. "JMP ($xxFF)" are
broken, if the address given is at some (odd) address. But when loading
a relocatable file, one cannot know if an address will be odd or even.
Therefore there is a new opcode,
<pre>
.align 2
</pre>
that aligns the next address at the given address boundary. Valid
values are 2, 4, and 256.
<p>
<h3>4) Clearance</h3>
<p>
This file is surely not the optimum and could be improved. Also the
header option "assigned numbers" should be added here.
<p>
For this reason the author, André Fachat, will function as a
clearing point, where problems can be discussed and number can be assigned.
<p>
Dec. 22, 1996,<br>
André Fachat <br>
([email protected])<br>
<p>
<h2>Appendix</h2>
<h3>A) File examples</h3>
<p>
(to be done with reference assembler)
</body></html>