/[gxemul]/trunk/src/cpus/cpu_arm_instr_loadstore.c
This is repository of my old source code which isn't updated any more. Go to git.rot13.org for current projects!
ViewVC logotype

Contents of /trunk/src/cpus/cpu_arm_instr_loadstore.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 18 - (show annotations)
Mon Oct 8 16:19:11 2007 UTC (16 years, 5 months ago) by dpavlin
File MIME type: text/plain
File size: 13756 byte(s)
++ trunk/HISTORY	(local)
$Id: HISTORY,v 1.1004 2005/10/27 14:01:10 debug Exp $
20051011        Passing -A as the default boot arg for CATS (works fine with
                OpenBSD/cats).
20051012	Fixing the VGA cursor offset bug, and speeding up framebuffer
		redraws if character cells contain the same thing as during
		the last redraw.
20051013	Adding a slow strd ARM instruction hack.
20051017	Minor updates: Adding a dummy i80321 Verde controller (for
		XScale emulation), fixing the disassembly of the ARM "ldrd"
		instruction, adding "support" for less-than-4KB pages for ARM
		(by not adding them to translation tables).
20051020	Continuing on some HPCarm stuff. A NetBSD/hpcarm kernel prints
		some boot messages on an emulated Jornada 720.
		Making dev_ram work better with dyntrans (speeds up some things
		quite a bit).
20051021	Automatically generating some of the most common ARM load/store
		multiple instructions.
20051022	Better statistics gathering for the ARM load/store multiple.
		Various other dyntrans and device updates.
20051023	Various minor updates.
20051024	Continuing; minor device and dyntrans fine-tuning. Adding the
		first "reasonable" instruction combination hacks for ARM (the
		cores of NetBSD/cats' memset and memcpy).
20051025	Fixing a dyntrans-related bug in dev_vga. Also changing the
		dyntrans low/high access notification to only be updated on
		writes, not reads. Hopefully it will be enough. (dev_vga in
		charcell mode now seems to work correctly with both reads and
		writes.)
		Experimenting with gathering dyntrans statistics (which parts
		of emulated RAM that are actually executed), and adding
		instruction combination hacks for cache cleaning and a part of
		NetBSD's scanc() function.
20051026	Adding a bitmap for ARM emulation which indicates if a page is
		(specifically) user accessible; loads and stores with the t-
		flag set can now use the translation arrays, which results in
		a measurable speedup.
20051027	Dyntrans updates; adding an extra bitmap array for 32-bit
		emulation modes, speeding up the check whether a physical page
		has any code translations or not (O(n) -> O(1)). Doing a
		similar reduction of O(n) to O(1) by avoiding the scan through
		the translation entries on a translation update (32-bit mode
		only).
		Various other minor hacks.
20051029	Quick release, without any testing at all.

==============  RELEASE 0.3.6.2  ==============


1 /*
2 * Copyright (C) 2005 Anders Gavare. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions are met:
6 *
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
12 * 3. The name of the author may not be used to endorse or promote products
13 * derived from this software without specific prior written permission.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 *
27 *
28 * $Id: cpu_arm_instr_loadstore.c,v 1.15 2005/10/27 14:01:13 debug Exp $
29 *
30 *
31 * TODO: Many things...
32 *
33 * o) Big-endian ARM loads/stores.
34 *
35 * o) Alignment checks!
36 *
37 * o) Native load/store if the endianness is the same as the host's
38 * (only implemented for little endian, so far, and it assumes that
39 * alignment is correct!)
40 *
41 * o) "Base Updated Abort Model", which updates the base register
42 * even if the memory access failed.
43 *
44 * o) Some ARM implementations use pc+8, some use pc+12 for stores?
45 *
46 * o) All load/store variants with the PC register are not really
47 * valid. (E.g. a byte load into the PC register. What should that
48 * accomplish?)
49 *
50 * o) Perhaps an optimization for the case when offset = 0, because
51 * that's quite common, and also when the Reg expression is just
52 * a simple, non-rotated register (0..14).
53 */
54
55
56 #if defined(A__SIGNED) && defined(A__H) && !defined(A__L)
57 #define A__STRD
58 #endif
59
60
61 /*
62 * General load/store, by using memory_rw(). If at all possible, memory_rw()
63 * then inserts the page into the translation array, so that the fast
64 * load/store routine below can be used for further accesses.
65 */
66 void A__NAME__general(struct cpu *cpu, struct arm_instr_call *ic)
67 {
68 #if !defined(A__P) && defined(A__W)
69 const int memory_rw_flags = CACHE_DATA | MEMORY_USER_ACCESS;
70 #else
71 const int memory_rw_flags = CACHE_DATA;
72 #endif
73
74 #ifdef A__REG
75 uint32_t (*reg_func)(struct cpu *, struct arm_instr_call *)
76 = (void *)(size_t)ic->arg[1];
77 #endif
78
79 #ifdef A__STRD
80 unsigned char data[8];
81 const int datalen = 8;
82 #else
83 #ifdef A__B
84 unsigned char data[1];
85 const int datalen = 1;
86 #else
87 #ifdef A__H
88 unsigned char data[2];
89 const int datalen = 2;
90 #else
91 const int datalen = 4;
92 #ifdef HOST_LITTLE_ENDIAN
93 unsigned char *data = (unsigned char *) ic->arg[2];
94 #else
95 unsigned char data[4];
96 #endif
97 #endif
98 #endif
99 #endif
100
101 uint32_t addr, low_pc, offset =
102 #ifndef A__U
103 -
104 #endif
105 #ifdef A__REG
106 reg_func(cpu, ic);
107 #else
108 ic->arg[1];
109 #endif
110
111 low_pc = ((size_t)ic - (size_t)cpu->cd.arm.
112 cur_ic_page) / sizeof(struct arm_instr_call);
113 cpu->cd.arm.r[ARM_PC] &= ~((ARM_IC_ENTRIES_PER_PAGE-1)
114 << ARM_INSTR_ALIGNMENT_SHIFT);
115 cpu->cd.arm.r[ARM_PC] += (low_pc << ARM_INSTR_ALIGNMENT_SHIFT);
116 cpu->pc = cpu->cd.arm.r[ARM_PC];
117
118 addr = reg(ic->arg[0])
119 #ifdef A__P
120 + offset
121 #endif
122 ;
123
124 #ifdef A__L
125 /* Load: */
126 if (!cpu->memory_rw(cpu, cpu->mem, addr, data, datalen,
127 MEM_READ, memory_rw_flags)) {
128 /* load failed, an exception was generated */
129 return;
130 }
131 #ifdef A__B
132 reg(ic->arg[2]) =
133 #ifdef A__SIGNED
134 (int32_t)(int8_t)
135 #endif
136 data[0];
137 #else
138 #ifdef A__H
139 reg(ic->arg[2]) =
140 #ifdef A__SIGNED
141 (int32_t)(int16_t)
142 #endif
143 (data[0] + (data[1] << 8));
144 #else
145 #ifdef HOST_LITTLE_ENDIAN
146 /* Nothing. */
147 #else
148 reg(ic->arg[2]) = data[0] + (data[1] << 8) +
149 (data[2] << 16) + (data[3] << 24);
150 #endif
151 #endif
152 #endif
153 #else
154 /* Store: */
155 #if !defined(A__B) && !defined(A__H) && defined(HOST_LITTLE_ENDIAN)
156 #ifdef A__STRD
157 *(uint32_t *)data = reg(ic->arg[2]);
158 *(uint32_t *)(data + 4) = reg(ic->arg[2] + 4);
159 #endif
160 #else
161 data[0] = reg(ic->arg[2]);
162 #ifndef A__B
163 data[1] = reg(ic->arg[2]) >> 8;
164 #if !defined(A__H) || defined(A__STRD)
165 data[1] = reg(ic->arg[2]) >> 8;
166 data[2] = reg(ic->arg[2]) >> 16;
167 data[3] = reg(ic->arg[2]) >> 24;
168 #ifdef A__STRD
169 data[4] = reg(ic->arg[2] + 4);
170 data[5] = reg(ic->arg[2] + 4) >> 8;
171 data[6] = reg(ic->arg[2] + 4) >> 16;
172 data[7] = reg(ic->arg[2] + 4) >> 24;
173 #endif
174 #endif
175 #endif
176 #endif
177 if (!cpu->memory_rw(cpu, cpu->mem, addr, data, datalen,
178 MEM_WRITE, memory_rw_flags)) {
179 /* store failed, an exception was generated */
180 return;
181 }
182 #endif
183
184 #ifdef A__P
185 #ifdef A__W
186 reg(ic->arg[0]) = addr;
187 #endif
188 #else /* post-index writeback */
189 reg(ic->arg[0]) = addr + offset;
190 #endif
191 }
192
193
194 /*
195 * Fast load/store, if the page is in the translation array.
196 */
197 void A__NAME(struct cpu *cpu, struct arm_instr_call *ic)
198 {
199 #ifdef A__STRD
200 /* Chicken out, let's do this unoptimized for now: */
201 A__NAME__general(cpu, ic);
202 #else
203 #ifdef A__REG
204 uint32_t (*reg_func)(struct cpu *, struct arm_instr_call *)
205 = (void *)(size_t)ic->arg[1];
206 #endif
207 uint32_t offset =
208 #ifndef A__U
209 -
210 #endif
211 #ifdef A__REG
212 reg_func(cpu, ic);
213 #else
214 ic->arg[1];
215 #endif
216 uint32_t addr = reg(ic->arg[0])
217 #ifdef A__P
218 + offset
219 #endif
220 ;
221 unsigned char *page = cpu->cd.arm.
222 #ifdef A__L
223 host_load
224 #else
225 host_store
226 #endif
227 [addr >> 12];
228
229
230 #if !defined(A__P) && defined(A__W)
231 /*
232 * T-bit: userland access: check the corresponding bit in the
233 * is_userpage array. If it is set, then we're ok. Otherwise: use the
234 * generic function.
235 */
236 unsigned char x = cpu->cd.arm.is_userpage[addr >> 15];
237 if (!(x & (1 << ((addr >> 12) & 7))))
238 A__NAME__general(cpu, ic);
239 else
240 #endif
241
242
243 if (page == NULL) {
244 A__NAME__general(cpu, ic);
245 } else {
246 #ifdef A__L
247 #ifdef A__B
248 reg(ic->arg[2]) =
249 #ifdef A__SIGNED
250 (int32_t)(int8_t)
251 #endif
252 page[addr & 0xfff];
253 #else
254 #ifdef A__H
255 reg(ic->arg[2]) =
256 #ifdef A__SIGNED
257 (int32_t)(int16_t)
258 #endif
259 (page[addr & 0xfff] + (page[(addr & 0xfff) + 1] << 8));
260 #else
261 #ifdef HOST_LITTLE_ENDIAN
262 reg(ic->arg[2]) = *(uint32_t *)(page + (addr & 0xffc));
263 #else
264 reg(ic->arg[2]) = page[addr & 0xfff] +
265 (page[(addr & 0xfff) + 1] << 8) +
266 (page[(addr & 0xfff) + 2] << 16) +
267 (page[(addr & 0xfff) + 3] << 24);
268 #endif
269 #endif
270 #endif
271 #else
272 #ifdef A__B
273 page[addr & 0xfff] = reg(ic->arg[2]);
274 #else
275 #ifdef A__H
276 page[addr & 0xfff] = reg(ic->arg[2]);
277 page[(addr & 0xfff)+1] = reg(ic->arg[2]) >> 8;
278 #else
279 #ifdef HOST_LITTLE_ENDIAN
280 *(uint32_t *)(page + (addr & 0xffc)) = reg(ic->arg[2]);
281 #else
282 page[addr & 0xfff] = reg(ic->arg[2]);
283 page[(addr & 0xfff)+1] = reg(ic->arg[2]) >> 8;
284 page[(addr & 0xfff)+2] = reg(ic->arg[2]) >> 16;
285 page[(addr & 0xfff)+3] = reg(ic->arg[2]) >> 24;
286 #endif
287 #endif
288 #endif
289 #endif
290
291 /* Index Write-back: */
292 #ifdef A__P
293 #ifdef A__W
294 reg(ic->arg[0]) = addr;
295 #endif
296 #else
297 /* post-index writeback */
298 reg(ic->arg[0]) = addr + offset;
299 #endif
300 }
301 #endif /* not STRD */
302 }
303
304
305 /*
306 * Special case when loading or storing the ARM's PC register, or when the PC
307 * register is used as the base address register.
308 *
309 * o) Loads into the PC register cause a branch. If an exception occured
310 * during the load, then the PC register should already point to the
311 * exception handler, in which case we simply recalculate the pointers a
312 * second time (no harm is done by doing that).
313 *
314 * TODO: A tiny performance optimization would be to separate the two
315 * cases: a load where arg[0] = PC, and the case where arg[2] = PC.
316 *
317 * o) Stores store "PC of the current instruction + 12". The solution I have
318 * choosen is to calculate this value and place it into a temporary
319 * variable (tmp_pc), which is then used for the store.
320 */
321 void A__NAME_PC(struct cpu *cpu, struct arm_instr_call *ic)
322 {
323 #ifdef A__L
324 /* Load: */
325 if (ic->arg[0] == (size_t)(&cpu->cd.arm.tmp_pc)) {
326 /* tmp_pc = current PC + 8: */
327 uint32_t low_pc, tmp;
328 low_pc = ((size_t)ic - (size_t) cpu->cd.arm.cur_ic_page) /
329 sizeof(struct arm_instr_call);
330 tmp = cpu->cd.arm.r[ARM_PC] & ~((ARM_IC_ENTRIES_PER_PAGE-1) <<
331 ARM_INSTR_ALIGNMENT_SHIFT);
332 tmp += (low_pc << ARM_INSTR_ALIGNMENT_SHIFT);
333 cpu->cd.arm.tmp_pc = tmp + 8;
334 }
335 A__NAME(cpu, ic);
336 if (ic->arg[2] == (size_t)(&cpu->cd.arm.r[ARM_PC])) {
337 cpu->pc = cpu->cd.arm.r[ARM_PC];
338 if (cpu->machine->show_trace_tree)
339 cpu_functioncall_trace(cpu, cpu->pc);
340 quick_pc_to_pointers(cpu);
341 }
342 #else
343 /* Store: */
344 uint32_t low_pc, tmp;
345 /* Calculate tmp from this instruction's PC + 12 */
346 low_pc = ((size_t)ic - (size_t) cpu->cd.arm.cur_ic_page) /
347 sizeof(struct arm_instr_call);
348 tmp = cpu->cd.arm.r[ARM_PC] & ~((ARM_IC_ENTRIES_PER_PAGE-1) <<
349 ARM_INSTR_ALIGNMENT_SHIFT);
350 tmp += (low_pc << ARM_INSTR_ALIGNMENT_SHIFT);
351 cpu->cd.arm.tmp_pc = tmp + 12;
352 A__NAME(cpu, ic);
353 #endif
354 }
355
356
357 #ifndef A__NOCONDITIONS
358 /* Load/stores with all registers except the PC register: */
359 void A__NAME__eq(struct cpu *cpu, struct arm_instr_call *ic)
360 { if (cpu->cd.arm.cpsr & ARM_FLAG_Z) A__NAME(cpu, ic); }
361 void A__NAME__ne(struct cpu *cpu, struct arm_instr_call *ic)
362 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME(cpu, ic); }
363 void A__NAME__cs(struct cpu *cpu, struct arm_instr_call *ic)
364 { if (cpu->cd.arm.cpsr & ARM_FLAG_C) A__NAME(cpu, ic); }
365 void A__NAME__cc(struct cpu *cpu, struct arm_instr_call *ic)
366 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_C)) A__NAME(cpu, ic); }
367 void A__NAME__mi(struct cpu *cpu, struct arm_instr_call *ic)
368 { if (cpu->cd.arm.cpsr & ARM_FLAG_N) A__NAME(cpu, ic); }
369 void A__NAME__pl(struct cpu *cpu, struct arm_instr_call *ic)
370 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_N)) A__NAME(cpu, ic); }
371 void A__NAME__vs(struct cpu *cpu, struct arm_instr_call *ic)
372 { if (cpu->cd.arm.cpsr & ARM_FLAG_V) A__NAME(cpu, ic); }
373 void A__NAME__vc(struct cpu *cpu, struct arm_instr_call *ic)
374 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_V)) A__NAME(cpu, ic); }
375
376 void A__NAME__hi(struct cpu *cpu, struct arm_instr_call *ic)
377 { if (cpu->cd.arm.cpsr & ARM_FLAG_C &&
378 !(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME(cpu, ic); }
379 void A__NAME__ls(struct cpu *cpu, struct arm_instr_call *ic)
380 { if (cpu->cd.arm.cpsr & ARM_FLAG_Z ||
381 !(cpu->cd.arm.cpsr & ARM_FLAG_C)) A__NAME(cpu, ic); }
382 void A__NAME__ge(struct cpu *cpu, struct arm_instr_call *ic)
383 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) ==
384 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0)) A__NAME(cpu, ic); }
385 void A__NAME__lt(struct cpu *cpu, struct arm_instr_call *ic)
386 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) !=
387 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0)) A__NAME(cpu, ic); }
388 void A__NAME__gt(struct cpu *cpu, struct arm_instr_call *ic)
389 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) ==
390 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0) &&
391 !(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME(cpu, ic); }
392 void A__NAME__le(struct cpu *cpu, struct arm_instr_call *ic)
393 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) !=
394 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0) ||
395 (cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME(cpu, ic); }
396
397
398 /* Load/stores with the PC register: */
399 void A__NAME_PC__eq(struct cpu *cpu, struct arm_instr_call *ic)
400 { if (cpu->cd.arm.cpsr & ARM_FLAG_Z) A__NAME_PC(cpu, ic); }
401 void A__NAME_PC__ne(struct cpu *cpu, struct arm_instr_call *ic)
402 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME_PC(cpu, ic); }
403 void A__NAME_PC__cs(struct cpu *cpu, struct arm_instr_call *ic)
404 { if (cpu->cd.arm.cpsr & ARM_FLAG_C) A__NAME_PC(cpu, ic); }
405 void A__NAME_PC__cc(struct cpu *cpu, struct arm_instr_call *ic)
406 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_C)) A__NAME_PC(cpu, ic); }
407 void A__NAME_PC__mi(struct cpu *cpu, struct arm_instr_call *ic)
408 { if (cpu->cd.arm.cpsr & ARM_FLAG_N) A__NAME_PC(cpu, ic); }
409 void A__NAME_PC__pl(struct cpu *cpu, struct arm_instr_call *ic)
410 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_N)) A__NAME_PC(cpu, ic); }
411 void A__NAME_PC__vs(struct cpu *cpu, struct arm_instr_call *ic)
412 { if (cpu->cd.arm.cpsr & ARM_FLAG_V) A__NAME_PC(cpu, ic); }
413 void A__NAME_PC__vc(struct cpu *cpu, struct arm_instr_call *ic)
414 { if (!(cpu->cd.arm.cpsr & ARM_FLAG_V)) A__NAME_PC(cpu, ic); }
415
416 void A__NAME_PC__hi(struct cpu *cpu, struct arm_instr_call *ic)
417 { if (cpu->cd.arm.cpsr & ARM_FLAG_C &&
418 !(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME_PC(cpu, ic); }
419 void A__NAME_PC__ls(struct cpu *cpu, struct arm_instr_call *ic)
420 { if (cpu->cd.arm.cpsr & ARM_FLAG_Z ||
421 !(cpu->cd.arm.cpsr & ARM_FLAG_C)) A__NAME_PC(cpu, ic); }
422 void A__NAME_PC__ge(struct cpu *cpu, struct arm_instr_call *ic)
423 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) ==
424 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0)) A__NAME_PC(cpu, ic); }
425 void A__NAME_PC__lt(struct cpu *cpu, struct arm_instr_call *ic)
426 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) !=
427 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0)) A__NAME_PC(cpu, ic); }
428 void A__NAME_PC__gt(struct cpu *cpu, struct arm_instr_call *ic)
429 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) ==
430 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0) &&
431 !(cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME_PC(cpu, ic); }
432 void A__NAME_PC__le(struct cpu *cpu, struct arm_instr_call *ic)
433 { if (((cpu->cd.arm.cpsr & ARM_FLAG_N)?1:0) !=
434 ((cpu->cd.arm.cpsr & ARM_FLAG_V)?1:0) ||
435 (cpu->cd.arm.cpsr & ARM_FLAG_Z)) A__NAME_PC(cpu, ic); }
436 #endif
437
438
439 #ifdef A__STRD
440 #undef A__STRD
441 #endif

  ViewVC Help
Powered by ViewVC 1.1.26