1 /** 2 * Copyright (c) 2010-2012 Broadcom. 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 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions, and the following disclaimer, 9 * without modification. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. The names of the above-listed copyright holders may not be used 14 * to endorse or promote products derived from this software without 15 * specific prior written permission. 16 * 17 * ALTERNATIVELY, this software may be distributed under the terms of the 18 * GNU General Public License ("GPL") version 2, as published by the Free 19 * Software Foundation. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS 22 * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 23 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR 25 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 26 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 27 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 28 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <interface/compat/vchi_bsd.h> 35 36 #include <sys/param.h> 37 #include <sys/malloc.h> 38 #include <sys/bus.h> 39 #include <sys/kmem.h> 40 41 #include <linux/completion.h> 42 43 #include <uvm/uvm_extern.h> 44 45 #include <arm/cpufunc.h> 46 47 #include <arch/arm/broadcom/bcm2835_mbox.h> 48 #include <arch/arm/broadcom/bcm2835var.h> 49 50 #define TOTAL_SLOTS (VCHIQ_SLOT_ZERO_SLOTS + 2 * 32) 51 52 #define VCHIQ_DOORBELL_IRQ IRQ_ARM_DOORBELL_0 53 54 #define IS_USER_ADDRESS(va) \ 55 ((vaddr_t)(va) >= VM_MIN_ADDRESS && (vaddr_t)(va) < VM_MAX_ADDRESS) 56 57 #include "vchiq_arm.h" 58 #include "vchiq_2835.h" 59 #include "vchiq_netbsd.h" 60 #include "vchiq_connected.h" 61 62 #define VCPAGE_OFFSET 0x0fff 63 #define VCPAGE_SHIFT 12 64 65 #define MAX_FRAGMENTS (VCHIQ_NUM_CURRENT_BULKS * 2) 66 67 typedef struct vchiq_2835_state_struct { 68 int inited; 69 VCHIQ_ARM_STATE_T arm_state; 70 } VCHIQ_2835_ARM_STATE_T; 71 72 /* BSD DMA */ 73 static bus_dma_tag_t dma_tag; 74 static bus_dmamap_t dma_map; 75 76 static unsigned int g_cache_line_size = CACHE_LINE_SIZE; 77 static unsigned int g_fragments_size; 78 static char *g_fragments_base; 79 static char *g_free_fragments; 80 81 struct semaphore g_free_fragments_sema; 82 static struct semaphore g_free_fragments_mutex; 83 84 void 85 vchiq_platform_attach(bus_dma_tag_t tag) 86 { 87 dma_tag = tag; 88 } 89 90 int __init 91 vchiq_platform_init(VCHIQ_STATE_T *state) 92 { 93 VCHIQ_SLOT_ZERO_T *vchiq_slot_zero; 94 bus_dma_segment_t dma_segs[1]; 95 int dma_nsegs; 96 void *slot_mem; 97 bus_addr_t slot_phys; 98 int slot_mem_size, frag_mem_size; 99 int err; 100 int i; 101 102 _sema_init(&g_free_fragments_mutex, 1); 103 104 g_cache_line_size = 32; 105 106 g_fragments_size = 2 * g_cache_line_size; 107 108 /* Allocate space for the channels in coherent memory */ 109 slot_mem_size = PAGE_ALIGN(TOTAL_SLOTS * VCHIQ_SLOT_SIZE); 110 frag_mem_size = PAGE_ALIGN(g_fragments_size * MAX_FRAGMENTS); 111 112 dma_nsegs = __arraycount(dma_segs); 113 err = bus_dmamem_alloc(dma_tag, 114 slot_mem_size + frag_mem_size, PAGE_SIZE, 0, 115 dma_segs, dma_nsegs, &dma_nsegs, BUS_DMA_WAITOK); 116 if (err) { 117 vchiq_log_error(vchiq_core_log_level, "Unable to allocate channel memory"); 118 err = -ENOMEM; 119 goto failed_alloc; 120 } 121 122 err = bus_dmamem_map(dma_tag, 123 dma_segs, dma_nsegs, slot_mem_size + frag_mem_size, 124 (void **)&slot_mem, BUS_DMA_COHERENT | BUS_DMA_WAITOK); 125 if (err) { 126 vchiq_log_error(vchiq_core_log_level, "Unable to map channel memory"); 127 err = -ENOMEM; 128 goto failed_alloc; 129 } 130 131 err = bus_dmamap_create(dma_tag, 132 slot_mem_size + frag_mem_size, 1, /* maxsize, nsegments */ 133 slot_mem_size + frag_mem_size, 0, /* maxsegsize, boundary */ 134 BUS_DMA_WAITOK, 135 &dma_map); 136 if (err) { 137 vchiq_log_error(vchiq_core_log_level, "Unable to create DMA map"); 138 err = -ENOMEM; 139 goto failed_alloc; 140 } 141 142 err = bus_dmamap_load(dma_tag, dma_map, slot_mem, 143 slot_mem_size + frag_mem_size, NULL, BUS_DMA_WAITOK); 144 if (err) { 145 vchiq_log_error(vchiq_core_log_level, "cannot load DMA map (%d)", err); 146 err = -ENOMEM; 147 goto failed_load; 148 } 149 slot_phys = dma_map->dm_segs[0].ds_addr; 150 151 vchiq_log_info(vchiq_arm_log_level, 152 "%s: slot_phys = %lx\n", __func__, slot_phys); 153 154 WARN_ON(((uintptr_t)slot_mem & (PAGE_SIZE - 1)) != 0); 155 156 vchiq_slot_zero = vchiq_init_slots(slot_mem, slot_mem_size); 157 if (!vchiq_slot_zero) { 158 err = -EINVAL; 159 goto failed_init_slots; 160 } 161 162 vchiq_slot_zero->platform_data[VCHIQ_PLATFORM_FRAGMENTS_OFFSET_IDX] = 163 htole32((uint32_t)slot_phys + slot_mem_size); 164 vchiq_slot_zero->platform_data[VCHIQ_PLATFORM_FRAGMENTS_COUNT_IDX] = 165 htole32(MAX_FRAGMENTS); 166 167 g_fragments_base = (char *)slot_mem + slot_mem_size; 168 slot_mem_size += frag_mem_size; 169 170 g_free_fragments = g_fragments_base; 171 for (i = 0; i < (MAX_FRAGMENTS - 1); i++) { 172 *(char **)&g_fragments_base[i*g_fragments_size] = 173 &g_fragments_base[(i + 1)*g_fragments_size]; 174 } 175 *(char **)&g_fragments_base[i * g_fragments_size] = NULL; 176 177 _sema_init(&g_free_fragments_sema, MAX_FRAGMENTS); 178 179 if (vchiq_init_state(state, vchiq_slot_zero, 0/*slave*/) != 180 VCHIQ_SUCCESS) { 181 err = -EINVAL; 182 goto failed_vchiq_init; 183 } 184 185 /* Send the base address of the slots to VideoCore */ 186 dsb(sy); /* Ensure all writes have completed */ 187 188 bus_dmamap_sync(dma_tag, dma_map, 0, slot_mem_size, 189 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 190 bcm_mbox_write(BCM2835_MBOX_CHAN_VCHIQ, (uint32_t)slot_phys); 191 bus_dmamap_sync(dma_tag, dma_map, 0, slot_mem_size, 192 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 193 194 vchiq_log_info(vchiq_arm_log_level, 195 "vchiq_init - done (slots %p, phys %x)", 196 vchiq_slot_zero, (unsigned int)slot_phys); 197 198 vchiq_call_connected_callbacks(); 199 200 return 0; 201 202 failed_vchiq_init: 203 failed_init_slots: 204 failed_load: 205 bus_dmamap_unload(dma_tag, dma_map); 206 failed_alloc: 207 bus_dmamap_destroy(dma_tag, dma_map); 208 209 return err; 210 } 211 212 void __exit 213 vchiq_platform_exit(VCHIQ_STATE_T *state) 214 { 215 216 bus_dmamap_unload(dma_tag, dma_map); 217 bus_dmamap_destroy(dma_tag, dma_map); 218 } 219 220 221 VCHIQ_STATUS_T 222 vchiq_platform_init_state(VCHIQ_STATE_T *state) 223 { 224 VCHIQ_STATUS_T status = VCHIQ_SUCCESS; 225 state->platform_state = kzalloc(sizeof(VCHIQ_2835_ARM_STATE_T), GFP_KERNEL); 226 ((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited = 1; 227 status = vchiq_arm_init_state(state, &((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->arm_state); 228 if(status != VCHIQ_SUCCESS) 229 { 230 ((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited = 0; 231 } 232 return status; 233 } 234 235 VCHIQ_ARM_STATE_T* 236 vchiq_platform_get_arm_state(VCHIQ_STATE_T *state) 237 { 238 if(!((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited) 239 { 240 BUG(); 241 } 242 return &((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->arm_state; 243 } 244 245 typedef struct bulkinfo_struct { 246 void *pagelist; 247 bus_dma_segment_t pagelist_sgs[1]; 248 bus_size_t pagelist_size; 249 bus_dmamap_t pagelist_map; 250 bus_dmamap_t dmamap; 251 struct proc *proc; 252 void *buf; 253 int size; 254 } BULKINFO_T; 255 256 /* There is a potential problem with partial cache lines (pages?) 257 ** at the ends of the block when reading. If the CPU accessed anything in 258 ** the same line (page?) then it may have pulled old data into the cache, 259 ** obscuring the new data underneath. We can solve this by transferring the 260 ** partial cache lines separately, and allowing the ARM to copy into the 261 ** cached area. 262 */ 263 VCHIQ_STATUS_T 264 vchiq_prepare_bulk_data(VCHIQ_BULK_T *bulk, VCHI_MEM_HANDLE_T memhandle, 265 void *buf, int size, int dir) 266 { 267 PAGELIST_T *pagelist; 268 BULKINFO_T *bi; 269 int nsegs; 270 int ret; 271 272 vchiq_log_info(vchiq_arm_log_level, 273 "%s: buf %p size %08x dir %s", __func__, buf, size, 274 dir == VCHIQ_BULK_RECEIVE ? "read" : "write"); 275 276 vaddr_t va = (vaddr_t)buf; 277 const size_t maxsegs = atop(round_page(va + size) - trunc_page(va)); 278 const int uvmflags = (dir == VCHIQ_BULK_RECEIVE ? 279 VM_PROT_READ : VM_PROT_WRITE); 280 const int dmaflags = (dir == VCHIQ_BULK_RECEIVE ? 281 BUS_DMA_READ : BUS_DMA_WRITE); 282 283 WARN_ON(memhandle != VCHI_MEM_HANDLE_INVALID); 284 285 bi = kmem_alloc(sizeof(*bi), KM_SLEEP); 286 bi->buf = buf; 287 bi->size = size; 288 bi->pagelist_size = sizeof(PAGELIST_T) + 289 (maxsegs * sizeof(uint32_t)); 290 bi->proc = curproc; 291 292 ret = bus_dmamem_alloc(dma_tag, bi->pagelist_size, 293 0 /*CACHE_LINE_SIZE*/, 0, bi->pagelist_sgs, 294 __arraycount(bi->pagelist_sgs), &nsegs, BUS_DMA_WAITOK); 295 296 if (ret != 0) 297 goto fail1; 298 299 ret = bus_dmamem_map(dma_tag, bi->pagelist_sgs, nsegs, 300 bi->pagelist_size, &bi->pagelist, BUS_DMA_COHERENT | BUS_DMA_WAITOK); 301 if (ret != 0) 302 goto fail2; 303 304 pagelist = bi->pagelist; 305 306 ret = bus_dmamap_create(dma_tag, bi->pagelist_size, 307 nsegs, bi->pagelist_size, 0, BUS_DMA_WAITOK, &bi->pagelist_map); 308 if (ret != 0) 309 goto fail3; 310 311 ret = bus_dmamap_load(dma_tag, bi->pagelist_map, pagelist, 312 bi->pagelist_size, NULL, BUS_DMA_WAITOK | BUS_DMA_WRITE); 313 if (ret != 0) 314 goto fail4; 315 316 /* 317 * Need to wire the buffer pages in. 318 */ 319 if (IS_USER_ADDRESS(buf)) { 320 ret = uvm_vslock(bi->proc->p_vmspace, buf, size, uvmflags); 321 if (ret != 0) { 322 printf("%s: uvm_vslock failed (%d)\n", __func__, ret); 323 goto fail5; 324 } 325 } 326 327 ret = bus_dmamap_create(dma_tag, size, maxsegs, size, 0, 328 BUS_DMA_WAITOK, &bi->dmamap); 329 330 if (ret != 0) 331 goto fail6; 332 333 ret = bus_dmamap_load(dma_tag, bi->dmamap, buf, size, 334 curproc, BUS_DMA_WAITOK | dmaflags); 335 336 if (ret != 0) 337 goto fail7; 338 339 bulk->handle = memhandle; 340 /* 341 * We've now got the bus_addr_t for the pagelist we want the transfer 342 * to use. 343 */ 344 bulk->data = (void *)bi->pagelist_map->dm_segs[0].ds_addr; 345 346 pagelist->type = htole16((dir == VCHIQ_BULK_RECEIVE) ? 347 PAGELIST_READ : PAGELIST_WRITE); 348 pagelist->length = htole32(size); 349 pagelist->offset = htole16(va & VCPAGE_OFFSET); 350 351 /* 352 * busdma already coalesces contiguous pages for us 353 */ 354 for (int i = 0; i < bi->dmamap->dm_nsegs; i++) { 355 bus_addr_t addr = bi->dmamap->dm_segs[i].ds_addr; 356 bus_size_t len = bi->dmamap->dm_segs[i].ds_len; 357 bus_size_t off = addr & VCPAGE_OFFSET; 358 int npgs = ((off + len + VCPAGE_OFFSET) >> VCPAGE_SHIFT); 359 360 pagelist->addrs[i] = htole32((addr & ~VCPAGE_OFFSET) | (npgs - 1)); 361 } 362 363 /* Partial cache lines (fragments) require special measures */ 364 if ((le16toh(pagelist->type) == PAGELIST_READ) && 365 ((le16toh(pagelist->offset) & (g_cache_line_size - 1)) || 366 ((le16toh(pagelist->offset) + le32toh(pagelist->length)) & (g_cache_line_size - 1)))) { 367 char *fragments; 368 369 if (down_interruptible(&g_free_fragments_sema) != 0) { 370 goto fail7; 371 } 372 373 WARN_ON(g_free_fragments == NULL); 374 375 down(&g_free_fragments_mutex); 376 fragments = g_free_fragments; 377 WARN_ON(fragments == NULL); 378 g_free_fragments = *(char **) g_free_fragments; 379 up(&g_free_fragments_mutex); 380 pagelist->type = htole16(PAGELIST_READ_WITH_FRAGMENTS + 381 (fragments - g_fragments_base) / g_fragments_size); 382 bus_dmamap_sync(dma_tag, dma_map, 383 (char *)fragments - g_fragments_base, sizeof(*fragments), 384 BUS_DMASYNC_PREREAD); 385 } 386 387 /* 388 * Store the BULKINFO_T address in remote_data, which isn't used by the 389 * slave. 390 */ 391 bulk->remote_data = bi; 392 393 bus_dmamap_sync(dma_tag, bi->pagelist_map, 0, 394 bi->pagelist_size, BUS_DMASYNC_PREWRITE); 395 396 bus_dmamap_sync(dma_tag, bi->dmamap, 0, bi->size, 397 le16toh(pagelist->type) == PAGELIST_WRITE ? 398 BUS_DMASYNC_PREWRITE : BUS_DMASYNC_PREREAD); 399 400 return VCHIQ_SUCCESS; 401 402 fail7: 403 bus_dmamap_destroy(dma_tag, bi->dmamap); 404 405 fail6: 406 if (IS_USER_ADDRESS(bi->buf)) 407 uvm_vsunlock(curproc->p_vmspace, bi->buf, bi->size); 408 409 fail5: 410 bus_dmamap_unload(dma_tag, bi->pagelist_map); 411 412 fail4: 413 bus_dmamap_destroy(dma_tag, bi->pagelist_map); 414 415 fail3: 416 bus_dmamem_unmap(dma_tag, bi->pagelist, bi->pagelist_size); 417 418 fail2: 419 bus_dmamem_free(dma_tag, bi->pagelist_sgs, 420 __arraycount(bi->pagelist_sgs)); 421 422 fail1: 423 kmem_free(bi, sizeof(*bi)); 424 return VCHIQ_ERROR; 425 } 426 427 void 428 vchiq_complete_bulk(VCHIQ_BULK_T *bulk) 429 { 430 if (bulk && bulk->remote_data && bulk->actual) { 431 int actual = bulk->actual; 432 BULKINFO_T *bi = bulk->remote_data; 433 PAGELIST_T *pagelist = bi->pagelist; 434 435 vchiq_log_trace(vchiq_arm_log_level, 436 "free_pagelist - %p, %d", pagelist, actual); 437 438 bus_dmamap_sync(dma_tag, bi->pagelist_map, 0, 439 bi->pagelist_size, BUS_DMASYNC_POSTWRITE); 440 441 bus_dmamap_sync(dma_tag, bi->dmamap, 0, bi->size, 442 le16toh(pagelist->type) == PAGELIST_WRITE ? 443 BUS_DMASYNC_POSTWRITE : BUS_DMASYNC_POSTREAD); 444 445 /* Deal with any partial cache lines (fragments) */ 446 if (le16toh(pagelist->type) >= PAGELIST_READ_WITH_FRAGMENTS) { 447 char *fragments = g_fragments_base + 448 (le16toh(pagelist->type) - PAGELIST_READ_WITH_FRAGMENTS) * 449 g_fragments_size; 450 int head_bytes, tail_bytes; 451 452 bus_dmamap_sync(dma_tag, dma_map, 453 (char *)fragments - g_fragments_base, g_fragments_size, 454 BUS_DMASYNC_POSTREAD); 455 456 head_bytes = (g_cache_line_size - le16toh(pagelist->offset)) & 457 (g_cache_line_size - 1); 458 tail_bytes = (le16toh(pagelist->offset) + actual) & 459 (g_cache_line_size - 1); 460 461 if ((actual >= 0) && (head_bytes != 0)) { 462 if (head_bytes > actual) 463 head_bytes = actual; 464 465 if (IS_USER_ADDRESS(bi->buf)) { 466 copyout_proc(bi->proc, fragments, 467 bi->buf, head_bytes); 468 } else { 469 kcopy(fragments, bi->buf, head_bytes); 470 } 471 } 472 if ((actual >= 0) && (head_bytes < actual) && 473 (tail_bytes != 0)) { 474 void *t = (char *)bi->buf + bi->size - 475 tail_bytes; 476 477 if (IS_USER_ADDRESS(bi->buf)) { 478 copyout_proc(bi->proc, 479 fragments + g_cache_line_size, t, 480 tail_bytes); 481 } else { 482 kcopy(fragments + g_cache_line_size, t, 483 tail_bytes); 484 } 485 } 486 487 down(&g_free_fragments_mutex); 488 *(char **)fragments = g_free_fragments; 489 g_free_fragments = fragments; 490 up(&g_free_fragments_mutex); 491 up(&g_free_fragments_sema); 492 } 493 bus_dmamap_unload(dma_tag, bi->dmamap); 494 bus_dmamap_destroy(dma_tag, bi->dmamap); 495 if (IS_USER_ADDRESS(bi->buf)) 496 uvm_vsunlock(bi->proc->p_vmspace, bi->buf, bi->size); 497 498 bus_dmamap_unload(dma_tag, bi->pagelist_map); 499 bus_dmamap_destroy(dma_tag, bi->pagelist_map); 500 bus_dmamem_unmap(dma_tag, bi->pagelist, 501 bi->pagelist_size); 502 bus_dmamem_free(dma_tag, bi->pagelist_sgs, 503 __arraycount(bi->pagelist_sgs)); 504 kmem_free(bi, sizeof(*bi)); 505 } 506 } 507 508 void 509 vchiq_transfer_bulk(VCHIQ_BULK_T *bulk) 510 { 511 /* 512 * This should only be called on the master (VideoCore) side, but 513 * provide an implementation to avoid the need for ifdefery. 514 */ 515 BUG(); 516 } 517 518 void 519 vchiq_dump_platform_state(void *dump_context) 520 { 521 char buf[80]; 522 int len; 523 len = snprintf(buf, sizeof(buf), 524 " Platform: 2835 (VC master)"); 525 vchiq_dump(dump_context, buf, len + 1); 526 } 527 528 VCHIQ_STATUS_T 529 vchiq_platform_suspend(VCHIQ_STATE_T *state) 530 { 531 return VCHIQ_ERROR; 532 } 533 534 VCHIQ_STATUS_T 535 vchiq_platform_resume(VCHIQ_STATE_T *state) 536 { 537 return VCHIQ_SUCCESS; 538 } 539 540 void 541 vchiq_platform_paused(VCHIQ_STATE_T *state) 542 { 543 } 544 545 void 546 vchiq_platform_resumed(VCHIQ_STATE_T *state) 547 { 548 } 549 550 int 551 vchiq_platform_videocore_wanted(VCHIQ_STATE_T* state) 552 { 553 return 1; // autosuspend not supported - videocore always wanted 554 } 555 556 int 557 vchiq_platform_use_suspend_timer(void) 558 { 559 return 0; 560 } 561 void 562 vchiq_dump_platform_use_state(VCHIQ_STATE_T *state) 563 { 564 vchiq_log_info(vchiq_arm_log_level, "Suspend timer not in use"); 565 } 566 void 567 vchiq_platform_handle_timeout(VCHIQ_STATE_T *state) 568 { 569 (void)state; 570 } 571
Indexes created Sat Oct 10 00:27:21 UTC 2026