1 |
227 |
jeremybenn |
|
2 |
|
|
SIS - Sparc Instruction Simulator README file (v2.0, 05-02-1996)
|
3 |
|
|
-------------------------------------------------------------------
|
4 |
|
|
|
5 |
|
|
1. Introduction
|
6 |
|
|
|
7 |
|
|
The SIS is a SPARC V7 architecture simulator. It consist of two parts,
|
8 |
|
|
the simulator core and a user defined memory module. The simulator
|
9 |
|
|
core executes the instructions while the memory module emulates memory
|
10 |
|
|
and peripherals.
|
11 |
|
|
|
12 |
|
|
2. Usage
|
13 |
|
|
|
14 |
|
|
The simulator is started as follows:
|
15 |
|
|
|
16 |
|
|
sis [-uart1 uart_device1] [-uart2 uart_device2]
|
17 |
|
|
[-nfp] [-freq frequency] [-c batch_file] [files]
|
18 |
|
|
|
19 |
|
|
The default uart devices for SIS are /dev/ptypc and /dev/ptypd. The
|
20 |
|
|
-uart[1,2] switch can be used to connect the uarts to other devices.
|
21 |
|
|
Use 'tip /dev/ttypc' to connect a terminal emulator to the uarts.
|
22 |
|
|
The '-nfp' will disable the simulated FPU, so each FPU instruction will
|
23 |
|
|
generate a FPU disabled trap. The '-freq' switch can be used to define
|
24 |
|
|
which "frequency" the simulator runs at. This is used by the 'perf'
|
25 |
|
|
command to calculated the MIPS figure for a particular configuration.
|
26 |
|
|
The give frequency must be an integer indicating the frequency in MHz.
|
27 |
|
|
|
28 |
|
|
The -c option indicates that sis commands should be read from 'batch_file'
|
29 |
|
|
at startup.
|
30 |
|
|
|
31 |
|
|
Files to be loaded must be in one of the supported formats (see INSTALLATION),
|
32 |
|
|
and will be loaded into the simulated memory. The file formats are
|
33 |
|
|
automatically recognised.
|
34 |
|
|
|
35 |
|
|
The script 'startsim' will start the simulator in one xterm window and
|
36 |
|
|
open a terminal emulator (tip) connected to the UART A in a second
|
37 |
|
|
xterm window. Below is description of commands that are recognized by
|
38 |
|
|
the simulator. The command-line is parsed using GNU readline. A command
|
39 |
|
|
history of 64 commands is maintained. Use the up/down arrows to recall
|
40 |
|
|
previous commands. For more details, see the readline documentation.
|
41 |
|
|
|
42 |
|
|
batch
|
43 |
|
|
|
44 |
|
|
Execute a batch file of SIS commands.
|
45 |
|
|
|
46 |
|
|
+bp
|
47 |
|
|
|
48 |
|
|
Adds an breakpoint at address .
|
49 |
|
|
|
50 |
|
|
bp
|
51 |
|
|
|
52 |
|
|
Prints all breakpoints
|
53 |
|
|
|
54 |
|
|
-bp
|
55 |
|
|
|
56 |
|
|
Deletes breakpoint . Use 'bp' to see which number is assigned to the
|
57 |
|
|
breakpoints.
|
58 |
|
|
|
59 |
|
|
cont [inst_count]
|
60 |
|
|
|
61 |
|
|
Continue execution at present position, optionally for [inst_count]
|
62 |
|
|
instructions.
|
63 |
|
|
|
64 |
|
|
dis [addr] [count]
|
65 |
|
|
|
66 |
|
|
Disassemble [count] instructions at address [addr]. Default values for
|
67 |
|
|
count is 16 and addr is the present address.
|
68 |
|
|
|
69 |
|
|
echo
|
70 |
|
|
|
71 |
|
|
Print to the simulator window.
|
72 |
|
|
|
73 |
|
|
float
|
74 |
|
|
|
75 |
|
|
Prints the FPU registers
|
76 |
|
|
|
77 |
|
|
go [inst_count]
|
78 |
|
|
|
79 |
|
|
The go command will set pc to and npc to + 4, and start
|
80 |
|
|
execution. No other initialisation will be done. If inst_count is given,
|
81 |
|
|
execution will stop after the specified number of instructions.
|
82 |
|
|
|
83 |
|
|
help
|
84 |
|
|
|
85 |
|
|
Print a small help menu for the SIS commands.
|
86 |
|
|
|
87 |
|
|
hist [trace_length]
|
88 |
|
|
|
89 |
|
|
Enable the instruction trace buffer. The 'trace_length' last executed
|
90 |
|
|
instructions will be placed in the trace buffer. A 'hist' command without
|
91 |
|
|
a trace_length will display the trace buffer. Specifying a zero trace
|
92 |
|
|
length will disable the trace buffer.
|
93 |
|
|
|
94 |
|
|
load
|
95 |
|
|
|
96 |
|
|
Loads a file into simulator memory.
|
97 |
|
|
|
98 |
|
|
mem [addr] [count]
|
99 |
|
|
|
100 |
|
|
Display memory at [addr] for [count] bytes. Same default values as above.
|
101 |
|
|
|
102 |
|
|
quit
|
103 |
|
|
|
104 |
|
|
Exits the simulator.
|
105 |
|
|
|
106 |
|
|
perf [reset]
|
107 |
|
|
|
108 |
|
|
The 'perf' command will display various execution statistics. A 'perf reset'
|
109 |
|
|
command will reset the statistics. This can be used if statistics shall
|
110 |
|
|
be calculated only over a part of the program. The 'run' and 'reset'
|
111 |
|
|
command also resets the statistic information.
|
112 |
|
|
|
113 |
|
|
reg [reg_name] [value]
|
114 |
|
|
|
115 |
|
|
Prints and sets the IU regiters. 'reg' without parameters prints the IU
|
116 |
|
|
registers. 'reg [reg_name] [value]' sets the corresponding register to
|
117 |
|
|
[value]. Valid register names are psr, tbr, wim, y, g1-g7, o0-o7 and
|
118 |
|
|
l0-l7.
|
119 |
|
|
|
120 |
|
|
reset
|
121 |
|
|
|
122 |
|
|
Performs a power-on reset. This command is equal to 'run 0'.
|
123 |
|
|
|
124 |
|
|
run [inst_count]
|
125 |
|
|
|
126 |
|
|
Resets the simulator and starts execution from address 0. If an instruction
|
127 |
|
|
count is given (inst_count), the simulator will stop after the specified
|
128 |
|
|
number of instructions. The event queue is emptied but any set breakpoints
|
129 |
|
|
remain.
|
130 |
|
|
|
131 |
|
|
step
|
132 |
|
|
|
133 |
|
|
Equal to 'trace 1'
|
134 |
|
|
|
135 |
|
|
tra [inst_count]
|
136 |
|
|
|
137 |
|
|
Starts the simulator at the present position and prints each instruction
|
138 |
|
|
it executes. If an instruction count is given (inst_count), the simulator
|
139 |
|
|
will stop after the specified number of instructions.
|
140 |
|
|
|
141 |
|
|
Typing a 'Ctrl-C' will interrupt a running simulator.
|
142 |
|
|
|
143 |
|
|
Short forms of the commands are allowed, e.g 'c' 'co' or 'con' are all
|
144 |
|
|
interpreted as 'cont'.
|
145 |
|
|
|
146 |
|
|
|
147 |
|
|
3. Simulator core
|
148 |
|
|
|
149 |
|
|
The SIS emulates the behavior of the 90C601E and 90C602E sparc IU and
|
150 |
|
|
FPU from Matra MHS. These are roughly equivalent to the Cypress C601
|
151 |
|
|
and C602. The simulator is cycle true, i.e a simulator time is
|
152 |
|
|
maintained and inremented according the IU and FPU instruction timing.
|
153 |
|
|
The parallel execution between the IU and FPU is modelled, as well as
|
154 |
|
|
stalls due to operand dependencies (FPU). The core interacts with the
|
155 |
|
|
user-defined memory modules through a number of functions. The memory
|
156 |
|
|
module must provide the following functions:
|
157 |
|
|
|
158 |
|
|
int memory_read(asi,addr,data,ws)
|
159 |
|
|
int asi;
|
160 |
|
|
unsigned int addr;
|
161 |
|
|
unsigned int *data;
|
162 |
|
|
int *ws;
|
163 |
|
|
|
164 |
|
|
int memory_write(asi,addr,data,sz,ws)
|
165 |
|
|
int asi;
|
166 |
|
|
unsigned int addr;
|
167 |
|
|
unsigned int *data;
|
168 |
|
|
int sz;
|
169 |
|
|
int *ws;
|
170 |
|
|
|
171 |
|
|
int sis_memory_read(addr, data, length)
|
172 |
|
|
unsigned int addr;
|
173 |
|
|
char *data;
|
174 |
|
|
unsigned int length;
|
175 |
|
|
|
176 |
|
|
int sis_memory_write(addr, data, length)
|
177 |
|
|
unsigned int addr;
|
178 |
|
|
char *data;
|
179 |
|
|
unsigned int length;
|
180 |
|
|
|
181 |
|
|
int init_sim()
|
182 |
|
|
|
183 |
|
|
int reset()
|
184 |
|
|
|
185 |
|
|
int error_mode(pc)
|
186 |
|
|
unsigned int pc;
|
187 |
|
|
|
188 |
|
|
memory_read() is used by the simulator to fetch instructions and
|
189 |
|
|
operands. The address space identifier (asi) and address is passed as
|
190 |
|
|
parameters. The read data should be assigned to the data pointer
|
191 |
|
|
(*data) and the number of waitstate to *ws. 'memory_read' should return
|
192 |
|
|
|
193 |
|
|
instruction fetch trap. memory_read() always reads one 32-bit word.
|
194 |
|
|
|
195 |
|
|
sis_memory_read() is used by the simulator to display and disassemble
|
196 |
|
|
memory contants. The function should copy 'length' bytes of the simulated
|
197 |
|
|
memory starting at 'addr' to '*data'.
|
198 |
|
|
The sis_memory_read() should return 1 on success and 0 on failure.
|
199 |
|
|
Failure should only be indicated if access to unimplemented memory is attempted.
|
200 |
|
|
|
201 |
|
|
memory_write() is used to write to memory. In addition to the asi
|
202 |
|
|
and address parameters, the size of the written data is given by 'sz'.
|
203 |
|
|
The pointer *data points to the data to be written. The 'sz' is coded
|
204 |
|
|
as follows:
|
205 |
|
|
|
206 |
|
|
sz access type
|
207 |
|
|
|
208 |
|
|
1 halfword
|
209 |
|
|
2 word
|
210 |
|
|
3 double-word
|
211 |
|
|
|
212 |
|
|
If a double word is written, the most significant word is in data[0] and
|
213 |
|
|
the least significant in data[1].
|
214 |
|
|
|
215 |
|
|
sis_memory_write() is used by the simulator during loading of programs.
|
216 |
|
|
The function should copy 'length' bytes from *data to the simulated
|
217 |
|
|
memory starting at 'addr'. sis_memory_write() should return 1 on
|
218 |
|
|
success and 0 on failure. Failure should only be indicated if access
|
219 |
|
|
to unimplemented memory is attempted. See erc32.c for more details
|
220 |
|
|
on how to define the memory emulation functions.
|
221 |
|
|
|
222 |
|
|
The 'init_sim' is called once when the simulator is started. This function
|
223 |
|
|
should be used to perform initialisations of user defined memory or
|
224 |
|
|
peripherals that only have to be done once, such as opening files etc.
|
225 |
|
|
|
226 |
|
|
The 'reset' is called every time the simulator is reset, i.e. when a
|
227 |
|
|
'run' command is given. This function should be used to simulate a power
|
228 |
|
|
on reset of memory and peripherals.
|
229 |
|
|
|
230 |
|
|
error_mode() is called by the simulator when the IU goes into error mode,
|
231 |
|
|
typically if a trap is caused when traps are disabled. The memory module
|
232 |
|
|
can then take actions, such as issue a reset.
|
233 |
|
|
|
234 |
|
|
sys_reset() can be called by the memory module to reset the simulator. A
|
235 |
|
|
reset will empty the event queue and perform a power-on reset.
|
236 |
|
|
|
237 |
|
|
4. Events and interrupts
|
238 |
|
|
|
239 |
|
|
The simulator supports an event queue and the generation of processor
|
240 |
|
|
interrupts. The following functions are available to the user-defined
|
241 |
|
|
memory module:
|
242 |
|
|
|
243 |
|
|
event(cfunc,arg,delta)
|
244 |
|
|
void (*cfunc)();
|
245 |
|
|
int arg;
|
246 |
|
|
unsigned int delta;
|
247 |
|
|
|
248 |
|
|
set_int(level,callback,arg)
|
249 |
|
|
int level;
|
250 |
|
|
void (*callback)();
|
251 |
|
|
int arg;
|
252 |
|
|
|
253 |
|
|
clear_int(level)
|
254 |
|
|
int level;
|
255 |
|
|
|
256 |
|
|
sim_stop()
|
257 |
|
|
|
258 |
|
|
The 'event' functions will schedule the execution of the function 'cfunc'
|
259 |
|
|
at time 'now + delta' clock cycles. The parameter 'arg' is passed as a
|
260 |
|
|
parameter to 'cfunc'.
|
261 |
|
|
|
262 |
|
|
The 'set_int' function set the processor interrupt 'level'. When the interrupt
|
263 |
|
|
is taken, the function 'callback' is called with the argument 'arg'. This
|
264 |
|
|
will also clear the interrupt. An interrupt can be cleared before it is
|
265 |
|
|
taken by calling 'clear_int' with the appropriate interrupt level.
|
266 |
|
|
|
267 |
|
|
The sim_stop function is called each time the simulator stops execution.
|
268 |
|
|
It can be used to flush buffered devices to get a clean state during
|
269 |
|
|
single stepping etc.
|
270 |
|
|
|
271 |
|
|
See 'erc32.c' for examples on how to use events and interrupts.
|
272 |
|
|
|
273 |
|
|
5. Memory module
|
274 |
|
|
|
275 |
|
|
The supplied memory module (erc32.c) emulates the functions of memory and
|
276 |
|
|
the MEC asic developed for the 90C601/2. It includes the following functions:
|
277 |
|
|
|
278 |
|
|
* UART A & B
|
279 |
|
|
* Real-time clock
|
280 |
|
|
* General purpose timer
|
281 |
|
|
* Interrupt controller
|
282 |
|
|
* Breakpoint register
|
283 |
|
|
* Watchpoint register
|
284 |
|
|
* 512 Kbyte ROM
|
285 |
|
|
* 4 Mbyte RAM
|
286 |
|
|
|
287 |
|
|
See README.erc32 on how the MEC functions are emulated. For a detailed MEC
|
288 |
|
|
specification, look at the ERC32 home page at URL:
|
289 |
|
|
|
290 |
|
|
http://www.estec.esa.nl/wsmwww/erc32
|
291 |
|
|
|
292 |
|
|
6. Compile and linking programs
|
293 |
|
|
|
294 |
|
|
The directory 'examples' contain some code fragments for SIS.
|
295 |
|
|
The script gccx indicates how the native sunos gcc and linker can be used
|
296 |
|
|
to produce executables for the simulator. To compile and link the provided
|
297 |
|
|
'hello.c', type 'gccx hello.c'. This will build the executable 'hello'.
|
298 |
|
|
Start the simulator by running 'startsim hello', and issue the command 'run.
|
299 |
|
|
After the program is terminated, the IU will be force to error mode through
|
300 |
|
|
a software trap and halt.
|
301 |
|
|
|
302 |
|
|
The programs are linked with a start-up file, srt0.S. This file includes
|
303 |
|
|
the traptable and window underflow/overflow trap routines.
|
304 |
|
|
|
305 |
|
|
7. IU and FPU instruction timing.
|
306 |
|
|
|
307 |
|
|
The simulator provides cycle true simulation. The following table shows
|
308 |
|
|
the emulated instruction timing for 90C601E & 90C602E:
|
309 |
|
|
|
310 |
|
|
Instructions Cycles
|
311 |
|
|
|
312 |
|
|
jmpl, rett 2
|
313 |
|
|
load 2
|
314 |
|
|
store 3
|
315 |
|
|
load double 3
|
316 |
|
|
store double 4
|
317 |
|
|
other integer ops 1
|
318 |
|
|
fabs 2
|
319 |
|
|
fadds 4
|
320 |
|
|
faddd 4
|
321 |
|
|
fcmps 4
|
322 |
|
|
fcmpd 4
|
323 |
|
|
fdivs 20
|
324 |
|
|
fdivd 35
|
325 |
|
|
fmovs 2
|
326 |
|
|
fmuls 5
|
327 |
|
|
fmuld 9
|
328 |
|
|
fnegs 2
|
329 |
|
|
fsqrts 37
|
330 |
|
|
fsqrtd 65
|
331 |
|
|
fsubs 4
|
332 |
|
|
fsubd 4
|
333 |
|
|
fdtoi 7
|
334 |
|
|
fdots 3
|
335 |
|
|
fitos 6
|
336 |
|
|
fitod 6
|
337 |
|
|
fstoi 6
|
338 |
|
|
fstod 2
|
339 |
|
|
|
340 |
|
|
The parallel operation between the IU and FPU is modelled. This means
|
341 |
|
|
that a FPU instruction will execute in parallel with other instructions as
|
342 |
|
|
long as no data or resource dependency is detected. See the 90C602E data
|
343 |
|
|
sheet for the various types of dependencies. Tracing using the 'trace'
|
344 |
|
|
command will display the current simulator time in the left column. This
|
345 |
|
|
time indicates when the instruction is fetched. If a dependency is detetected,
|
346 |
|
|
the following fetch will be delayed until the conflict is resolved.
|
347 |
|
|
|
348 |
|
|
The load dependency in the 90C601E is also modelled - if the destination
|
349 |
|
|
register of a load instruction is used by the following instruction, an
|
350 |
|
|
idle cycle is inserted.
|
351 |
|
|
|
352 |
|
|
8. FPU implementation
|
353 |
|
|
|
354 |
|
|
The simulator maps floating-point operations on the hosts floating point
|
355 |
|
|
capabilities. This means that accuracy and generation of IEEE exceptions is
|
356 |
|
|
host dependent.
|