1 /* 2 * QEMU PowerMac CUDA device support 3 * 4 * Copyright (c) 2004-2007 Fabrice Bellard 5 * Copyright (c) 2007 Jocelyn Mayer 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qemu-common.h" 28 #include "hw/hw.h" 29 #include "hw/ppc/mac.h" 30 #include "hw/input/adb.h" 31 #include "hw/misc/mos6522.h" 32 #include "hw/misc/macio/cuda.h" 33 #include "qemu/timer.h" 34 #include "sysemu/sysemu.h" 35 #include "qemu/cutils.h" 36 #include "qemu/log.h" 37 #include "qemu/module.h" 38 #include "trace.h" 39 40 /* Bits in B data register: all active low */ 41 #define TREQ 0x08 /* Transfer request (input) */ 42 #define TACK 0x10 /* Transfer acknowledge (output) */ 43 #define TIP 0x20 /* Transfer in progress (output) */ 44 45 /* commands (1st byte) */ 46 #define ADB_PACKET 0 47 #define CUDA_PACKET 1 48 #define ERROR_PACKET 2 49 #define TIMER_PACKET 3 50 #define POWER_PACKET 4 51 #define MACIIC_PACKET 5 52 #define PMU_PACKET 6 53 54 #define CUDA_TIMER_FREQ (4700000 / 6) 55 56 /* CUDA returns time_t's offset from Jan 1, 1904, not 1970 */ 57 #define RTC_OFFSET 2082844800 58 59 static void cuda_receive_packet_from_host(CUDAState *s, 60 const uint8_t *data, int len); 61 62 /* MacOS uses timer 1 for calibration on startup, so we use 63 * the timebase frequency and cuda_get_counter_value() with 64 * cuda_get_load_time() to steer MacOS to calculate calibrate its timers 65 * correctly for both TCG and KVM (see commit b981289c49 "PPC: Cuda: Use cuda 66 * timer to expose tbfreq to guest" for more information) */ 67 68 static uint64_t cuda_get_counter_value(MOS6522State *s, MOS6522Timer *ti) 69 { 70 MOS6522CUDAState *mcs = container_of(s, MOS6522CUDAState, parent_obj); 71 CUDAState *cs = container_of(mcs, CUDAState, mos6522_cuda); 72 73 /* Reverse of the tb calculation algorithm that Mac OS X uses on bootup */ 74 uint64_t tb_diff = muldiv64(qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL), 75 cs->tb_frequency, NANOSECONDS_PER_SECOND) - 76 ti->load_time; 77 78 return (tb_diff * 0xBF401675E5DULL) / (cs->tb_frequency << 24); 79 } 80 81 static uint64_t cuda_get_load_time(MOS6522State *s, MOS6522Timer *ti) 82 { 83 MOS6522CUDAState *mcs = container_of(s, MOS6522CUDAState, parent_obj); 84 CUDAState *cs = container_of(mcs, CUDAState, mos6522_cuda); 85 86 uint64_t load_time = muldiv64(qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL), 87 cs->tb_frequency, NANOSECONDS_PER_SECOND); 88 return load_time; 89 } 90 91 static void cuda_set_sr_int(void *opaque) 92 { 93 CUDAState *s = opaque; 94 MOS6522CUDAState *mcs = &s->mos6522_cuda; 95 MOS6522State *ms = MOS6522(mcs); 96 MOS6522DeviceClass *mdc = MOS6522_DEVICE_GET_CLASS(ms); 97 98 mdc->set_sr_int(ms); 99 } 100 101 static void cuda_delay_set_sr_int(CUDAState *s) 102 { 103 int64_t expire; 104 105 trace_cuda_delay_set_sr_int(); 106 107 expire = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + s->sr_delay_ns; 108 timer_mod(s->sr_delay_timer, expire); 109 } 110 111 /* NOTE: TIP and TREQ are negated */ 112 static void cuda_update(CUDAState *s) 113 { 114 MOS6522CUDAState *mcs = &s->mos6522_cuda; 115 MOS6522State *ms = MOS6522(mcs); 116 int packet_received, len; 117 118 packet_received = 0; 119 if (!(ms->b & TIP)) { 120 /* transfer requested from host */ 121 122 if (ms->acr & SR_OUT) { 123 /* data output */ 124 if ((ms->b & (TACK | TIP)) != (s->last_b & (TACK | TIP))) { 125 if (s->data_out_index < sizeof(s->data_out)) { 126 trace_cuda_data_send(ms->sr); 127 s->data_out[s->data_out_index++] = ms->sr; 128 cuda_delay_set_sr_int(s); 129 } 130 } 131 } else { 132 if (s->data_in_index < s->data_in_size) { 133 /* data input */ 134 if ((ms->b & (TACK | TIP)) != (s->last_b & (TACK | TIP))) { 135 ms->sr = s->data_in[s->data_in_index++]; 136 trace_cuda_data_recv(ms->sr); 137 /* indicate end of transfer */ 138 if (s->data_in_index >= s->data_in_size) { 139 ms->b = (ms->b | TREQ); 140 } 141 cuda_delay_set_sr_int(s); 142 } 143 } 144 } 145 } else { 146 /* no transfer requested: handle sync case */ 147 if ((s->last_b & TIP) && (ms->b & TACK) != (s->last_b & TACK)) { 148 /* update TREQ state each time TACK change state */ 149 if (ms->b & TACK) { 150 ms->b = (ms->b | TREQ); 151 } else { 152 ms->b = (ms->b & ~TREQ); 153 } 154 cuda_delay_set_sr_int(s); 155 } else { 156 if (!(s->last_b & TIP)) { 157 /* handle end of host to cuda transfer */ 158 packet_received = (s->data_out_index > 0); 159 /* always an IRQ at the end of transfer */ 160 cuda_delay_set_sr_int(s); 161 } 162 /* signal if there is data to read */ 163 if (s->data_in_index < s->data_in_size) { 164 ms->b = (ms->b & ~TREQ); 165 } 166 } 167 } 168 169 s->last_acr = ms->acr; 170 s->last_b = ms->b; 171 172 /* NOTE: cuda_receive_packet_from_host() can call cuda_update() 173 recursively */ 174 if (packet_received) { 175 len = s->data_out_index; 176 s->data_out_index = 0; 177 cuda_receive_packet_from_host(s, s->data_out, len); 178 } 179 } 180 181 static void cuda_send_packet_to_host(CUDAState *s, 182 const uint8_t *data, int len) 183 { 184 int i; 185 186 trace_cuda_packet_send(len); 187 for (i = 0; i < len; i++) { 188 trace_cuda_packet_send_data(i, data[i]); 189 } 190 191 memcpy(s->data_in, data, len); 192 s->data_in_size = len; 193 s->data_in_index = 0; 194 cuda_update(s); 195 cuda_delay_set_sr_int(s); 196 } 197 198 static void cuda_adb_poll(void *opaque) 199 { 200 CUDAState *s = opaque; 201 uint8_t obuf[ADB_MAX_OUT_LEN + 2]; 202 int olen; 203 204 olen = adb_poll(&s->adb_bus, obuf + 2, s->adb_poll_mask); 205 if (olen > 0) { 206 obuf[0] = ADB_PACKET; 207 obuf[1] = 0x40; /* polled data */ 208 cuda_send_packet_to_host(s, obuf, olen + 2); 209 } 210 timer_mod(s->adb_poll_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 211 (NANOSECONDS_PER_SECOND / (1000 / s->autopoll_rate_ms))); 212 } 213 214 /* description of commands */ 215 typedef struct CudaCommand { 216 uint8_t command; 217 const char *name; 218 bool (*handler)(CUDAState *s, 219 const uint8_t *in_args, int in_len, 220 uint8_t *out_args, int *out_len); 221 } CudaCommand; 222 223 static bool cuda_cmd_autopoll(CUDAState *s, 224 const uint8_t *in_data, int in_len, 225 uint8_t *out_data, int *out_len) 226 { 227 int autopoll; 228 229 if (in_len != 1) { 230 return false; 231 } 232 233 autopoll = (in_data[0] != 0); 234 if (autopoll != s->autopoll) { 235 s->autopoll = autopoll; 236 if (autopoll) { 237 timer_mod(s->adb_poll_timer, 238 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 239 (NANOSECONDS_PER_SECOND / (1000 / s->autopoll_rate_ms))); 240 } else { 241 timer_del(s->adb_poll_timer); 242 } 243 } 244 return true; 245 } 246 247 static bool cuda_cmd_set_autorate(CUDAState *s, 248 const uint8_t *in_data, int in_len, 249 uint8_t *out_data, int *out_len) 250 { 251 if (in_len != 1) { 252 return false; 253 } 254 255 /* we don't want a period of 0 ms */ 256 /* FIXME: check what real hardware does */ 257 if (in_data[0] == 0) { 258 return false; 259 } 260 261 s->autopoll_rate_ms = in_data[0]; 262 if (s->autopoll) { 263 timer_mod(s->adb_poll_timer, 264 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 265 (NANOSECONDS_PER_SECOND / (1000 / s->autopoll_rate_ms))); 266 } 267 return true; 268 } 269 270 static bool cuda_cmd_set_device_list(CUDAState *s, 271 const uint8_t *in_data, int in_len, 272 uint8_t *out_data, int *out_len) 273 { 274 if (in_len != 2) { 275 return false; 276 } 277 278 s->adb_poll_mask = (((uint16_t)in_data[0]) << 8) | in_data[1]; 279 return true; 280 } 281 282 static bool cuda_cmd_powerdown(CUDAState *s, 283 const uint8_t *in_data, int in_len, 284 uint8_t *out_data, int *out_len) 285 { 286 if (in_len != 0) { 287 return false; 288 } 289 290 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN); 291 return true; 292 } 293 294 static bool cuda_cmd_reset_system(CUDAState *s, 295 const uint8_t *in_data, int in_len, 296 uint8_t *out_data, int *out_len) 297 { 298 if (in_len != 0) { 299 return false; 300 } 301 302 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); 303 return true; 304 } 305 306 static bool cuda_cmd_set_file_server_flag(CUDAState *s, 307 const uint8_t *in_data, int in_len, 308 uint8_t *out_data, int *out_len) 309 { 310 if (in_len != 1) { 311 return false; 312 } 313 314 qemu_log_mask(LOG_UNIMP, 315 "CUDA: unimplemented command FILE_SERVER_FLAG %d\n", 316 in_data[0]); 317 return true; 318 } 319 320 static bool cuda_cmd_set_power_message(CUDAState *s, 321 const uint8_t *in_data, int in_len, 322 uint8_t *out_data, int *out_len) 323 { 324 if (in_len != 1) { 325 return false; 326 } 327 328 qemu_log_mask(LOG_UNIMP, 329 "CUDA: unimplemented command SET_POWER_MESSAGE %d\n", 330 in_data[0]); 331 return true; 332 } 333 334 static bool cuda_cmd_get_time(CUDAState *s, 335 const uint8_t *in_data, int in_len, 336 uint8_t *out_data, int *out_len) 337 { 338 uint32_t ti; 339 340 if (in_len != 0) { 341 return false; 342 } 343 344 ti = s->tick_offset + (qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) 345 / NANOSECONDS_PER_SECOND); 346 out_data[0] = ti >> 24; 347 out_data[1] = ti >> 16; 348 out_data[2] = ti >> 8; 349 out_data[3] = ti; 350 *out_len = 4; 351 return true; 352 } 353 354 static bool cuda_cmd_set_time(CUDAState *s, 355 const uint8_t *in_data, int in_len, 356 uint8_t *out_data, int *out_len) 357 { 358 uint32_t ti; 359 360 if (in_len != 4) { 361 return false; 362 } 363 364 ti = (((uint32_t)in_data[0]) << 24) + (((uint32_t)in_data[1]) << 16) 365 + (((uint32_t)in_data[2]) << 8) + in_data[3]; 366 s->tick_offset = ti - (qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) 367 / NANOSECONDS_PER_SECOND); 368 return true; 369 } 370 371 static const CudaCommand handlers[] = { 372 { CUDA_AUTOPOLL, "AUTOPOLL", cuda_cmd_autopoll }, 373 { CUDA_SET_AUTO_RATE, "SET_AUTO_RATE", cuda_cmd_set_autorate }, 374 { CUDA_SET_DEVICE_LIST, "SET_DEVICE_LIST", cuda_cmd_set_device_list }, 375 { CUDA_POWERDOWN, "POWERDOWN", cuda_cmd_powerdown }, 376 { CUDA_RESET_SYSTEM, "RESET_SYSTEM", cuda_cmd_reset_system }, 377 { CUDA_FILE_SERVER_FLAG, "FILE_SERVER_FLAG", 378 cuda_cmd_set_file_server_flag }, 379 { CUDA_SET_POWER_MESSAGES, "SET_POWER_MESSAGES", 380 cuda_cmd_set_power_message }, 381 { CUDA_GET_TIME, "GET_TIME", cuda_cmd_get_time }, 382 { CUDA_SET_TIME, "SET_TIME", cuda_cmd_set_time }, 383 }; 384 385 static void cuda_receive_packet(CUDAState *s, 386 const uint8_t *data, int len) 387 { 388 uint8_t obuf[16] = { CUDA_PACKET, 0, data[0] }; 389 int i, out_len = 0; 390 391 for (i = 0; i < ARRAY_SIZE(handlers); i++) { 392 const CudaCommand *desc = &handlers[i]; 393 if (desc->command == data[0]) { 394 trace_cuda_receive_packet_cmd(desc->name); 395 out_len = 0; 396 if (desc->handler(s, data + 1, len - 1, obuf + 3, &out_len)) { 397 cuda_send_packet_to_host(s, obuf, 3 + out_len); 398 } else { 399 qemu_log_mask(LOG_GUEST_ERROR, 400 "CUDA: %s: wrong parameters %d\n", 401 desc->name, len); 402 obuf[0] = ERROR_PACKET; 403 obuf[1] = 0x5; /* bad parameters */ 404 obuf[2] = CUDA_PACKET; 405 obuf[3] = data[0]; 406 cuda_send_packet_to_host(s, obuf, 4); 407 } 408 return; 409 } 410 } 411 412 qemu_log_mask(LOG_GUEST_ERROR, "CUDA: unknown command 0x%02x\n", data[0]); 413 obuf[0] = ERROR_PACKET; 414 obuf[1] = 0x2; /* unknown command */ 415 obuf[2] = CUDA_PACKET; 416 obuf[3] = data[0]; 417 cuda_send_packet_to_host(s, obuf, 4); 418 } 419 420 static void cuda_receive_packet_from_host(CUDAState *s, 421 const uint8_t *data, int len) 422 { 423 int i; 424 425 trace_cuda_packet_receive(len); 426 for (i = 0; i < len; i++) { 427 trace_cuda_packet_receive_data(i, data[i]); 428 } 429 430 switch(data[0]) { 431 case ADB_PACKET: 432 { 433 uint8_t obuf[ADB_MAX_OUT_LEN + 3]; 434 int olen; 435 olen = adb_request(&s->adb_bus, obuf + 2, data + 1, len - 1); 436 if (olen > 0) { 437 obuf[0] = ADB_PACKET; 438 obuf[1] = 0x00; 439 cuda_send_packet_to_host(s, obuf, olen + 2); 440 } else { 441 /* error */ 442 obuf[0] = ADB_PACKET; 443 obuf[1] = -olen; 444 obuf[2] = data[1]; 445 olen = 0; 446 cuda_send_packet_to_host(s, obuf, olen + 3); 447 } 448 } 449 break; 450 case CUDA_PACKET: 451 cuda_receive_packet(s, data + 1, len - 1); 452 break; 453 } 454 } 455 456 static uint64_t mos6522_cuda_read(void *opaque, hwaddr addr, unsigned size) 457 { 458 CUDAState *s = opaque; 459 MOS6522CUDAState *mcs = &s->mos6522_cuda; 460 MOS6522State *ms = MOS6522(mcs); 461 462 addr = (addr >> 9) & 0xf; 463 return mos6522_read(ms, addr, size); 464 } 465 466 static void mos6522_cuda_write(void *opaque, hwaddr addr, uint64_t val, 467 unsigned size) 468 { 469 CUDAState *s = opaque; 470 MOS6522CUDAState *mcs = &s->mos6522_cuda; 471 MOS6522State *ms = MOS6522(mcs); 472 473 addr = (addr >> 9) & 0xf; 474 mos6522_write(ms, addr, val, size); 475 } 476 477 static const MemoryRegionOps mos6522_cuda_ops = { 478 .read = mos6522_cuda_read, 479 .write = mos6522_cuda_write, 480 .endianness = DEVICE_BIG_ENDIAN, 481 .valid = { 482 .min_access_size = 1, 483 .max_access_size = 1, 484 }, 485 }; 486 487 static const VMStateDescription vmstate_cuda = { 488 .name = "cuda", 489 .version_id = 5, 490 .minimum_version_id = 5, 491 .fields = (VMStateField[]) { 492 VMSTATE_STRUCT(mos6522_cuda.parent_obj, CUDAState, 0, vmstate_mos6522, 493 MOS6522State), 494 VMSTATE_UINT8(last_b, CUDAState), 495 VMSTATE_UINT8(last_acr, CUDAState), 496 VMSTATE_INT32(data_in_size, CUDAState), 497 VMSTATE_INT32(data_in_index, CUDAState), 498 VMSTATE_INT32(data_out_index, CUDAState), 499 VMSTATE_UINT8(autopoll, CUDAState), 500 VMSTATE_UINT8(autopoll_rate_ms, CUDAState), 501 VMSTATE_UINT16(adb_poll_mask, CUDAState), 502 VMSTATE_BUFFER(data_in, CUDAState), 503 VMSTATE_BUFFER(data_out, CUDAState), 504 VMSTATE_UINT32(tick_offset, CUDAState), 505 VMSTATE_TIMER_PTR(adb_poll_timer, CUDAState), 506 VMSTATE_TIMER_PTR(sr_delay_timer, CUDAState), 507 VMSTATE_END_OF_LIST() 508 } 509 }; 510 511 static void cuda_reset(DeviceState *dev) 512 { 513 CUDAState *s = CUDA(dev); 514 515 s->data_in_size = 0; 516 s->data_in_index = 0; 517 s->data_out_index = 0; 518 s->autopoll = 0; 519 } 520 521 static void cuda_realize(DeviceState *dev, Error **errp) 522 { 523 CUDAState *s = CUDA(dev); 524 SysBusDevice *sbd; 525 MOS6522State *ms; 526 DeviceState *d; 527 struct tm tm; 528 529 /* Pass IRQ from 6522 */ 530 d = DEVICE(&s->mos6522_cuda); 531 ms = MOS6522(d); 532 sbd = SYS_BUS_DEVICE(s); 533 sysbus_pass_irq(sbd, SYS_BUS_DEVICE(ms)); 534 535 qemu_get_timedate(&tm, 0); 536 s->tick_offset = (uint32_t)mktimegm(&tm) + RTC_OFFSET; 537 538 s->sr_delay_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, cuda_set_sr_int, s); 539 s->sr_delay_ns = 20 * SCALE_US; 540 541 s->adb_poll_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, cuda_adb_poll, s); 542 s->adb_poll_mask = 0xffff; 543 s->autopoll_rate_ms = 20; 544 } 545 546 static void cuda_init(Object *obj) 547 { 548 CUDAState *s = CUDA(obj); 549 SysBusDevice *sbd = SYS_BUS_DEVICE(obj); 550 551 sysbus_init_child_obj(obj, "mos6522-cuda", &s->mos6522_cuda, 552 sizeof(s->mos6522_cuda), TYPE_MOS6522_CUDA); 553 554 memory_region_init_io(&s->mem, obj, &mos6522_cuda_ops, s, "cuda", 0x2000); 555 sysbus_init_mmio(sbd, &s->mem); 556 557 qbus_create_inplace(&s->adb_bus, sizeof(s->adb_bus), TYPE_ADB_BUS, 558 DEVICE(obj), "adb.0"); 559 } 560 561 static Property cuda_properties[] = { 562 DEFINE_PROP_UINT64("timebase-frequency", CUDAState, tb_frequency, 0), 563 DEFINE_PROP_END_OF_LIST() 564 }; 565 566 static void cuda_class_init(ObjectClass *oc, void *data) 567 { 568 DeviceClass *dc = DEVICE_CLASS(oc); 569 570 dc->realize = cuda_realize; 571 dc->reset = cuda_reset; 572 dc->vmsd = &vmstate_cuda; 573 dc->props = cuda_properties; 574 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories); 575 } 576 577 static const TypeInfo cuda_type_info = { 578 .name = TYPE_CUDA, 579 .parent = TYPE_SYS_BUS_DEVICE, 580 .instance_size = sizeof(CUDAState), 581 .instance_init = cuda_init, 582 .class_init = cuda_class_init, 583 }; 584 585 static void mos6522_cuda_portB_write(MOS6522State *s) 586 { 587 MOS6522CUDAState *mcs = container_of(s, MOS6522CUDAState, parent_obj); 588 CUDAState *cs = container_of(mcs, CUDAState, mos6522_cuda); 589 590 cuda_update(cs); 591 } 592 593 static void mos6522_cuda_reset(DeviceState *dev) 594 { 595 MOS6522State *ms = MOS6522(dev); 596 MOS6522DeviceClass *mdc = MOS6522_DEVICE_GET_CLASS(ms); 597 598 mdc->parent_reset(dev); 599 600 ms->timers[0].frequency = CUDA_TIMER_FREQ; 601 ms->timers[1].frequency = (SCALE_US * 6000) / 4700; 602 } 603 604 static void mos6522_cuda_class_init(ObjectClass *oc, void *data) 605 { 606 DeviceClass *dc = DEVICE_CLASS(oc); 607 MOS6522DeviceClass *mdc = MOS6522_DEVICE_CLASS(oc); 608 609 dc->reset = mos6522_cuda_reset; 610 mdc->portB_write = mos6522_cuda_portB_write; 611 mdc->get_timer1_counter_value = cuda_get_counter_value; 612 mdc->get_timer2_counter_value = cuda_get_counter_value; 613 mdc->get_timer1_load_time = cuda_get_load_time; 614 mdc->get_timer2_load_time = cuda_get_load_time; 615 } 616 617 static const TypeInfo mos6522_cuda_type_info = { 618 .name = TYPE_MOS6522_CUDA, 619 .parent = TYPE_MOS6522, 620 .instance_size = sizeof(MOS6522CUDAState), 621 .class_init = mos6522_cuda_class_init, 622 }; 623 624 static void cuda_register_types(void) 625 { 626 type_register_static(&mos6522_cuda_type_info); 627 type_register_static(&cuda_type_info); 628 } 629 630 type_init(cuda_register_types) 631