630b122b |
1 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * |
2 | * linkage_arm.s for PCSX * |
3 | * Copyright (C) 2009-2011 Ari64 * |
4 | * Copyright (C) 2021 notaz * |
5 | * * |
6 | * This program is free software; you can redistribute it and/or modify * |
7 | * it under the terms of the GNU General Public License as published by * |
8 | * the Free Software Foundation; either version 2 of the License, or * |
9 | * (at your option) any later version. * |
10 | * * |
11 | * This program is distributed in the hope that it will be useful, * |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of * |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
14 | * GNU General Public License for more details. * |
15 | * * |
16 | * You should have received a copy of the GNU General Public License * |
17 | * along with this program; if not, write to the * |
18 | * Free Software Foundation, Inc., * |
19 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * |
20 | * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ |
21 | |
22 | #include "arm_features.h" |
23 | #include "new_dynarec_config.h" |
24 | #include "assem_arm64.h" |
25 | #include "linkage_offsets.h" |
26 | |
27 | #if (LO_mem_wtab & 7) |
28 | #error misligned pointers |
29 | #endif |
30 | |
31 | .bss |
32 | .align 4 |
33 | .global dynarec_local |
34 | .type dynarec_local, %object |
35 | .size dynarec_local, LO_dynarec_local_size |
36 | dynarec_local: |
37 | .space LO_dynarec_local_size |
38 | |
39 | #define DRC_VAR_(name, vname, size_) \ |
40 | vname = dynarec_local + LO_##name; \ |
41 | .global vname; \ |
42 | .type vname, %object; \ |
43 | .size vname, size_ |
44 | |
45 | #define DRC_VAR(name, size_) \ |
46 | DRC_VAR_(name, ESYM(name), size_) |
47 | |
48 | DRC_VAR(next_interupt, 4) |
49 | DRC_VAR(cycle_count, 4) |
50 | DRC_VAR(last_count, 4) |
51 | DRC_VAR(pending_exception, 4) |
52 | DRC_VAR(stop, 4) |
53 | DRC_VAR(branch_target, 4) |
54 | DRC_VAR(address, 4) |
7c8454e3 |
55 | DRC_VAR(hack_addr, 4) |
630b122b |
56 | DRC_VAR(psxRegs, LO_psxRegs_end - LO_psxRegs) |
57 | |
58 | /* psxRegs */ |
59 | #DRC_VAR(reg, 128) |
60 | DRC_VAR(lo, 4) |
61 | DRC_VAR(hi, 4) |
62 | DRC_VAR(reg_cop0, 128) |
63 | DRC_VAR(reg_cop2d, 128) |
64 | DRC_VAR(reg_cop2c, 128) |
65 | DRC_VAR(pcaddr, 4) |
66 | #DRC_VAR(code, 4) |
67 | #DRC_VAR(cycle, 4) |
68 | #DRC_VAR(interrupt, 4) |
69 | #DRC_VAR(intCycle, 256) |
70 | |
71 | DRC_VAR(rcnts, 7*4*4) |
72 | DRC_VAR(inv_code_start, 4) |
73 | DRC_VAR(inv_code_end, 4) |
74 | DRC_VAR(mem_rtab, 8) |
75 | DRC_VAR(mem_wtab, 8) |
76 | DRC_VAR(psxH_ptr, 8) |
77 | DRC_VAR(invc_ptr, 8) |
78 | DRC_VAR(zeromem_ptr, 8) |
79 | DRC_VAR(scratch_buf_ptr, 8) |
80 | DRC_VAR(ram_offset, 8) |
81 | DRC_VAR(mini_ht, 256) |
630b122b |
82 | |
83 | |
84 | .text |
85 | .align 2 |
86 | |
630b122b |
87 | FUNCTION(dyna_linker): |
88 | /* r0 = virtual target address */ |
89 | /* r1 = instruction to patch */ |
048fcced |
90 | bl ndrc_get_addr_ht |
f9e37973 |
91 | br x0 |
630b122b |
92 | .size dyna_linker, .-dyna_linker |
93 | |
630b122b |
94 | .align 2 |
95 | FUNCTION(cc_interrupt): |
96 | ldr w0, [rFP, #LO_last_count] |
630b122b |
97 | add rCC, w0, rCC |
98 | str wzr, [rFP, #LO_pending_exception] |
630b122b |
99 | str rCC, [rFP, #LO_cycle] /* PCSX cycles */ |
100 | # str rCC, [rFP, #LO_reg_cop0+36] /* Count */ |
630b122b |
101 | mov x21, lr |
630b122b |
102 | 1: |
6c62131f |
103 | add x0, rFP, #(LO_psxRegs + 34*4) /* CP0 */ |
630b122b |
104 | bl gen_interupt |
105 | mov lr, x21 |
106 | ldr rCC, [rFP, #LO_cycle] |
107 | ldr w0, [rFP, #LO_next_interupt] |
108 | ldr w1, [rFP, #LO_pending_exception] |
109 | ldr w2, [rFP, #LO_stop] |
110 | str w0, [rFP, #LO_last_count] |
111 | sub rCC, rCC, w0 |
112 | cbnz w2, new_dyna_leave |
113 | cbnz w1, 2f |
114 | ret |
115 | 2: |
116 | ldr w0, [rFP, #LO_pcaddr] |
048fcced |
117 | bl ndrc_get_addr_ht |
630b122b |
118 | br x0 |
630b122b |
119 | .size cc_interrupt, .-cc_interrupt |
120 | |
121 | .align 2 |
f2e25348 |
122 | FUNCTION(jump_overflow_ds): |
123 | mov w0, #(12<<2) /* R3000E_Ov */ |
124 | mov w1, #1 |
125 | b call_psxException |
126 | FUNCTION(jump_overflow): |
127 | mov w0, #(12<<2) |
128 | mov w1, #0 |
129 | b call_psxException |
467357cc |
130 | FUNCTION(jump_break_ds): |
f2e25348 |
131 | mov w0, #(9<<2) /* R3000E_Bp */ |
467357cc |
132 | mov w1, #1 |
133 | b call_psxException |
134 | FUNCTION(jump_break): |
f2e25348 |
135 | mov w0, #(9<<2) |
467357cc |
136 | mov w1, #0 |
137 | b call_psxException |
138 | FUNCTION(jump_syscall_ds): |
f2e25348 |
139 | mov w0, #(8<<2) /* R3000E_Syscall */ |
6d79a06f |
140 | mov w1, #2 |
467357cc |
141 | b call_psxException |
630b122b |
142 | FUNCTION(jump_syscall): |
f2e25348 |
143 | mov w0, #(8<<2) |
467357cc |
144 | mov w1, #0 |
145 | |
146 | call_psxException: |
147 | ldr w3, [rFP, #LO_last_count] |
148 | str w2, [rFP, #LO_pcaddr] |
149 | add rCC, w3, rCC |
6c62131f |
150 | str rCC, [rFP, #LO_cycle] /* PCSX cycles */ |
151 | add x2, rFP, #(LO_psxRegs + 34*4) /* CP0 */ |
467357cc |
152 | bl psxException |
630b122b |
153 | |
154 | /* note: psxException might do recursive recompiler call from it's HLE code, |
155 | * so be ready for this */ |
156 | FUNCTION(jump_to_new_pc): |
157 | ldr w1, [rFP, #LO_next_interupt] |
158 | ldr rCC, [rFP, #LO_cycle] |
159 | ldr w0, [rFP, #LO_pcaddr] |
160 | sub rCC, rCC, w1 |
161 | str w1, [rFP, #LO_last_count] |
048fcced |
162 | bl ndrc_get_addr_ht |
630b122b |
163 | br x0 |
164 | .size jump_to_new_pc, .-jump_to_new_pc |
165 | |
166 | /* stack must be aligned by 16, and include space for save_regs() use */ |
167 | .align 2 |
168 | FUNCTION(new_dyna_start): |
169 | stp x29, x30, [sp, #-SSP_ALL]! |
170 | ldr w1, [x0, #LO_next_interupt] |
171 | ldr w2, [x0, #LO_cycle] |
172 | stp x19, x20, [sp, #16*1] |
173 | stp x21, x22, [sp, #16*2] |
174 | stp x23, x24, [sp, #16*3] |
175 | stp x25, x26, [sp, #16*4] |
176 | stp x27, x28, [sp, #16*5] |
177 | mov rFP, x0 |
178 | ldr w0, [rFP, #LO_pcaddr] |
179 | str w1, [rFP, #LO_last_count] |
180 | sub rCC, w2, w1 |
048fcced |
181 | bl ndrc_get_addr_ht |
630b122b |
182 | br x0 |
183 | .size new_dyna_start, .-new_dyna_start |
184 | |
185 | .align 2 |
186 | FUNCTION(new_dyna_leave): |
187 | ldr w0, [rFP, #LO_last_count] |
188 | add rCC, rCC, w0 |
189 | str rCC, [rFP, #LO_cycle] |
190 | ldp x19, x20, [sp, #16*1] |
191 | ldp x21, x22, [sp, #16*2] |
192 | ldp x23, x24, [sp, #16*3] |
193 | ldp x25, x26, [sp, #16*4] |
194 | ldp x27, x28, [sp, #16*5] |
195 | ldp x29, x30, [sp], #SSP_ALL |
196 | ret |
197 | .size new_dyna_leave, .-new_dyna_leave |
198 | |
199 | /* --------------------------------------- */ |
200 | |
201 | .align 2 |
202 | |
203 | .macro memhandler_pre |
204 | /* w0 = adddr/data, x1 = rhandler, w2 = cycles, x3 = whandler */ |
205 | ldr w4, [rFP, #LO_last_count] |
206 | add w4, w4, w2 |
207 | str w4, [rFP, #LO_cycle] |
208 | .endm |
209 | |
210 | .macro memhandler_post |
211 | ldr w0, [rFP, #LO_next_interupt] |
212 | ldr w2, [rFP, #LO_cycle] // memhandlers can modify cc, like dma |
213 | str w0, [rFP, #LO_last_count] |
214 | sub w0, w2, w0 |
215 | .endm |
216 | |
217 | FUNCTION(do_memhandler_pre): |
218 | memhandler_pre |
219 | ret |
220 | |
221 | FUNCTION(do_memhandler_post): |
222 | memhandler_post |
223 | ret |
224 | |
225 | .macro pcsx_read_mem readop tab_shift |
226 | /* w0 = address, x1 = handler_tab, w2 = cycles */ |
227 | ubfm w4, w0, #\tab_shift, #11 |
228 | ldr x3, [x1, w4, uxtw #3] |
229 | adds x3, x3, x3 |
230 | bcs 0f |
231 | \readop w0, [x3, w4, uxtw #\tab_shift] |
232 | ret |
233 | 0: |
234 | stp xzr, x30, [sp, #-16]! |
235 | memhandler_pre |
236 | blr x3 |
237 | .endm |
238 | |
239 | FUNCTION(jump_handler_read8): |
240 | add x1, x1, #0x1000/4*8 + 0x1000/2*8 /* shift to r8 part */ |
241 | pcsx_read_mem ldrb, 0 |
242 | b handler_read_end |
243 | |
244 | FUNCTION(jump_handler_read16): |
245 | add x1, x1, #0x1000/4*8 /* shift to r16 part */ |
246 | pcsx_read_mem ldrh, 1 |
247 | b handler_read_end |
248 | |
249 | FUNCTION(jump_handler_read32): |
250 | pcsx_read_mem ldr, 2 |
251 | |
252 | handler_read_end: |
253 | ldp xzr, x30, [sp], #16 |
254 | ret |
255 | |
256 | .macro pcsx_write_mem wrtop movop tab_shift |
257 | /* w0 = address, w1 = data, w2 = cycles, x3 = handler_tab */ |
258 | ubfm w4, w0, #\tab_shift, #11 |
259 | ldr x3, [x3, w4, uxtw #3] |
260 | adds x3, x3, x3 |
261 | bcs 0f |
262 | mov w0, w2 /* cycle return */ |
263 | \wrtop w1, [x3, w4, uxtw #\tab_shift] |
264 | ret |
265 | 0: |
266 | stp xzr, x30, [sp, #-16]! |
267 | str w0, [rFP, #LO_address] /* some handlers still need it... */ |
268 | \movop w0, w1 |
269 | memhandler_pre |
270 | blr x3 |
271 | .endm |
272 | |
273 | FUNCTION(jump_handler_write8): |
274 | add x3, x3, #0x1000/4*8 + 0x1000/2*8 /* shift to r8 part */ |
275 | pcsx_write_mem strb uxtb 0 |
276 | b handler_write_end |
277 | |
278 | FUNCTION(jump_handler_write16): |
279 | add x3, x3, #0x1000/4*8 /* shift to r16 part */ |
280 | pcsx_write_mem strh uxth 1 |
281 | b handler_write_end |
282 | |
283 | FUNCTION(jump_handler_write32): |
284 | pcsx_write_mem str mov 2 |
285 | |
286 | handler_write_end: |
287 | memhandler_post |
288 | ldp xzr, x30, [sp], #16 |
289 | ret |
290 | |
291 | FUNCTION(jump_handle_swl): |
292 | /* w0 = address, w1 = data, w2 = cycles */ |
293 | ldr x3, [rFP, #LO_mem_wtab] |
294 | orr w4, wzr, w0, lsr #12 |
295 | ldr x3, [x3, w4, uxtw #3] |
296 | adds x3, x3, x3 |
297 | bcs 4f |
298 | add x3, x0, x3 |
299 | mov w0, w2 |
300 | tbz x3, #1, 10f // & 2 |
301 | tbz x3, #0, 2f // & 1 |
302 | 3: |
303 | stur w1, [x3, #-3] |
304 | ret |
305 | 2: |
306 | lsr w2, w1, #8 |
307 | lsr w1, w1, #24 |
308 | sturh w2, [x3, #-2] |
309 | strb w1, [x3] |
310 | ret |
311 | 10: |
312 | tbz x3, #0, 0f // & 1 |
313 | 1: |
314 | lsr w1, w1, #16 |
315 | sturh w1, [x3, #-1] |
316 | ret |
317 | 0: |
318 | lsr w2, w1, #24 |
319 | strb w2, [x3] |
320 | ret |
321 | 4: |
322 | mov w0, w2 // todo |
323 | bl abort |
324 | ret |
325 | |
326 | FUNCTION(jump_handle_swr): |
327 | /* w0 = address, w1 = data, w2 = cycles */ |
328 | ldr x3, [rFP, #LO_mem_wtab] |
329 | orr w4, wzr, w0, lsr #12 |
330 | ldr x3, [x3, w4, uxtw #3] |
331 | adds x3, x3, x3 |
332 | bcs 4f |
333 | add x3, x0, x3 |
334 | mov w0, w2 |
335 | tbz x3, #1, 10f // & 2 |
336 | tbz x3, #0, 2f // & 1 |
337 | 3: |
338 | strb w1, [x3] |
339 | ret |
340 | 2: |
341 | strh w1, [x3] |
342 | ret |
343 | 10: |
344 | tbz x3, #0, 0f // & 1 |
345 | 1: |
346 | lsr w2, w1, #8 |
347 | strb w1, [x3] |
348 | sturh w2, [x3, #1] |
349 | ret |
350 | 0: |
351 | str w1, [x3] |
352 | ret |
353 | 4: |
354 | mov w0, w2 // todo |
355 | bl abort |
356 | ret |
357 | |
358 | FUNCTION(call_gteStall): |
359 | /* w0 = op_cycles, w1 = cycles */ |
360 | ldr w2, [rFP, #LO_last_count] |
361 | str lr, [rFP, #LO_saved_lr] |
362 | add w1, w1, w2 |
363 | str w1, [rFP, #LO_cycle] |
364 | add x1, rFP, #LO_psxRegs |
365 | bl gteCheckStallRaw |
366 | ldr lr, [rFP, #LO_saved_lr] |
367 | add rCC, rCC, w0 |
368 | ret |
369 | |