vchiq_2835_arm.c revision 278277
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/malloc.h> 37#include <sys/rwlock.h> 38 39#include <vm/vm.h> 40#include <vm/pmap.h> 41#include <vm/vm_extern.h> 42#include <vm/vm_kern.h> 43#include <vm/vm_map.h> 44#include <vm/vm_object.h> 45#include <vm/vm_page.h> 46#include <vm/vm_pager.h> 47#include <vm/vm_param.h> 48#include <vm/vm_phys.h> 49 50#include <machine/bus.h> 51#include <arm/broadcom/bcm2835/bcm2835_mbox.h> 52#include <arm/broadcom/bcm2835/bcm2835_vcbus.h> 53 54MALLOC_DEFINE(M_VCPAGELIST, "vcpagelist", "VideoCore pagelist memory"); 55 56#define TOTAL_SLOTS (VCHIQ_SLOT_ZERO_SLOTS + 2 * 32) 57 58#define VCHIQ_DOORBELL_IRQ IRQ_ARM_DOORBELL_0 59#define VCHIQ_ARM_ADDRESS(x) ((void *)PHYS_TO_VCBUS(pmap_kextract((vm_offset_t)(x)))) 60 61#include "vchiq_arm.h" 62#include "vchiq_2835.h" 63#include "vchiq_connected.h" 64 65#define MAX_FRAGMENTS (VCHIQ_NUM_CURRENT_BULKS * 2) 66 67typedef struct vchiq_2835_state_struct { 68 int inited; 69 VCHIQ_ARM_STATE_T arm_state; 70} VCHIQ_2835_ARM_STATE_T; 71 72static char *g_slot_mem; 73static int g_slot_mem_size; 74vm_paddr_t g_slot_phys; 75/* BSD DMA */ 76bus_dma_tag_t bcm_slots_dma_tag; 77bus_dmamap_t bcm_slots_dma_map; 78 79static FRAGMENTS_T *g_fragments_base; 80static FRAGMENTS_T *g_free_fragments; 81struct semaphore g_free_fragments_sema; 82 83static DEFINE_SEMAPHORE(g_free_fragments_mutex); 84 85typedef struct bulkinfo_struct { 86 PAGELIST_T *pagelist; 87 bus_dma_tag_t pagelist_dma_tag; 88 bus_dmamap_t pagelist_dma_map; 89 void *buf; 90 size_t size; 91} BULKINFO_T; 92 93static int 94create_pagelist(char __user *buf, size_t count, unsigned short type, 95 struct proc *p, BULKINFO_T *bi); 96 97static void 98free_pagelist(BULKINFO_T *bi, int actual); 99 100static void 101vchiq_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err) 102{ 103 bus_addr_t *addr; 104 105 if (err) 106 return; 107 108 addr = (bus_addr_t*)arg; 109 *addr = PHYS_TO_VCBUS(segs[0].ds_addr); 110} 111 112int __init 113vchiq_platform_init(VCHIQ_STATE_T *state) 114{ 115 VCHIQ_SLOT_ZERO_T *vchiq_slot_zero; 116 int frag_mem_size; 117 int err; 118 int i; 119 120 /* Allocate space for the channels in coherent memory */ 121 g_slot_mem_size = PAGE_ALIGN(TOTAL_SLOTS * VCHIQ_SLOT_SIZE); 122 frag_mem_size = PAGE_ALIGN(sizeof(FRAGMENTS_T) * MAX_FRAGMENTS); 123 124 err = bus_dma_tag_create( 125 NULL, 126 PAGE_SIZE, 0, /* alignment, boundary */ 127 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 128 BUS_SPACE_MAXADDR, /* highaddr */ 129 NULL, NULL, /* filter, filterarg */ 130 g_slot_mem_size + frag_mem_size, 1, /* maxsize, nsegments */ 131 g_slot_mem_size + frag_mem_size, 0, /* maxsegsize, flags */ 132 NULL, NULL, /* lockfunc, lockarg */ 133 &bcm_slots_dma_tag); 134 135 err = bus_dmamem_alloc(bcm_slots_dma_tag, (void **)&g_slot_mem, 136 BUS_DMA_COHERENT | BUS_DMA_WAITOK, &bcm_slots_dma_map); 137 if (err) { 138 vchiq_log_error(vchiq_core_log_level, "Unable to allocate channel memory"); 139 err = -ENOMEM; 140 goto failed_alloc; 141 } 142 143 err = bus_dmamap_load(bcm_slots_dma_tag, bcm_slots_dma_map, g_slot_mem, 144 g_slot_mem_size + frag_mem_size, vchiq_dmamap_cb, 145 &g_slot_phys, 0); 146 147 if (err) { 148 vchiq_log_error(vchiq_core_log_level, "cannot load DMA map"); 149 err = -ENOMEM; 150 goto failed_load; 151 } 152 153 WARN_ON(((int)g_slot_mem & (PAGE_SIZE - 1)) != 0); 154 155 vchiq_slot_zero = vchiq_init_slots(g_slot_mem, g_slot_mem_size); 156 if (!vchiq_slot_zero) { 157 err = -EINVAL; 158 goto failed_init_slots; 159 } 160 161 vchiq_slot_zero->platform_data[VCHIQ_PLATFORM_FRAGMENTS_OFFSET_IDX] = 162 (int)g_slot_phys + g_slot_mem_size; 163 vchiq_slot_zero->platform_data[VCHIQ_PLATFORM_FRAGMENTS_COUNT_IDX] = 164 MAX_FRAGMENTS; 165 166 g_fragments_base = (FRAGMENTS_T *)(g_slot_mem + g_slot_mem_size); 167 g_slot_mem_size += frag_mem_size; 168 169 g_free_fragments = g_fragments_base; 170 for (i = 0; i < (MAX_FRAGMENTS - 1); i++) { 171 *(FRAGMENTS_T **)&g_fragments_base[i] = 172 &g_fragments_base[i + 1]; 173 } 174 *(FRAGMENTS_T **)&g_fragments_base[i] = NULL; 175 _sema_init(&g_free_fragments_sema, MAX_FRAGMENTS); 176 177 if (vchiq_init_state(state, vchiq_slot_zero, 0/*slave*/) != 178 VCHIQ_SUCCESS) { 179 err = -EINVAL; 180 goto failed_vchiq_init; 181 } 182 183 bcm_mbox_write(BCM2835_MBOX_CHAN_VCHIQ, (unsigned int)g_slot_phys); 184 185 vchiq_log_info(vchiq_arm_log_level, 186 "vchiq_init - done (slots %x, phys %x)", 187 (unsigned int)vchiq_slot_zero, g_slot_phys); 188 189 vchiq_call_connected_callbacks(); 190 191 return 0; 192 193failed_vchiq_init: 194failed_init_slots: 195failed_load: 196 bus_dmamap_unload(bcm_slots_dma_tag, bcm_slots_dma_map); 197failed_alloc: 198 bus_dmamap_destroy(bcm_slots_dma_tag, bcm_slots_dma_map); 199 bus_dma_tag_destroy(bcm_slots_dma_tag); 200 201 return err; 202} 203 204void __exit 205vchiq_platform_exit(VCHIQ_STATE_T *state) 206{ 207 208 bus_dmamap_unload(bcm_slots_dma_tag, bcm_slots_dma_map); 209 bus_dmamap_destroy(bcm_slots_dma_tag, bcm_slots_dma_map); 210 bus_dma_tag_destroy(bcm_slots_dma_tag); 211} 212 213VCHIQ_STATUS_T 214vchiq_platform_init_state(VCHIQ_STATE_T *state) 215{ 216 VCHIQ_STATUS_T status = VCHIQ_SUCCESS; 217 state->platform_state = kzalloc(sizeof(VCHIQ_2835_ARM_STATE_T), GFP_KERNEL); 218 ((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited = 1; 219 status = vchiq_arm_init_state(state, &((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->arm_state); 220 if(status != VCHIQ_SUCCESS) 221 { 222 ((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited = 0; 223 } 224 return status; 225} 226 227VCHIQ_ARM_STATE_T* 228vchiq_platform_get_arm_state(VCHIQ_STATE_T *state) 229{ 230 if(!((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->inited) 231 { 232 BUG(); 233 } 234 return &((VCHIQ_2835_ARM_STATE_T*)state->platform_state)->arm_state; 235} 236 237int 238vchiq_copy_from_user(void *dst, const void *src, int size) 239{ 240 241 if (((vm_offset_t)(src)) < VM_MIN_KERNEL_ADDRESS) { 242 int error = copyin(src, dst, size); 243 return error ? VCHIQ_ERROR : VCHIQ_SUCCESS; 244 } 245 else 246 bcopy(src, dst, size); 247 248 return 0; 249} 250 251VCHIQ_STATUS_T 252vchiq_prepare_bulk_data(VCHIQ_BULK_T *bulk, VCHI_MEM_HANDLE_T memhandle, 253 void *offset, int size, int dir) 254{ 255 PAGELIST_T *pagelist; 256 BULKINFO_T *bi; 257 int ret; 258 259 WARN_ON(memhandle != VCHI_MEM_HANDLE_INVALID); 260 bi = malloc(sizeof(*bi), M_VCPAGELIST, M_WAITOK | M_ZERO); 261 if (bi == NULL) 262 return VCHIQ_ERROR; 263 264 ret = create_pagelist((char __user *)offset, size, 265 (dir == VCHIQ_BULK_RECEIVE) 266 ? PAGELIST_READ 267 : PAGELIST_WRITE, 268 current, 269 bi); 270 if (ret != 0) 271 return VCHIQ_ERROR; 272 273 bulk->handle = memhandle; 274 bulk->data = VCHIQ_ARM_ADDRESS(bi->pagelist); 275 276 /* Store the pagelist address in remote_data, which isn't used by the 277 slave. */ 278 bulk->remote_data = bi; 279 280 return VCHIQ_SUCCESS; 281} 282 283void 284vchiq_complete_bulk(VCHIQ_BULK_T *bulk) 285{ 286 if (bulk && bulk->remote_data && bulk->actual) 287 free_pagelist((BULKINFO_T *)bulk->remote_data, bulk->actual); 288} 289 290void 291vchiq_transfer_bulk(VCHIQ_BULK_T *bulk) 292{ 293 /* 294 * This should only be called on the master (VideoCore) side, but 295 * provide an implementation to avoid the need for ifdefery. 296 */ 297 BUG(); 298} 299 300void 301vchiq_dump_platform_state(void *dump_context) 302{ 303 char buf[80]; 304 int len; 305 len = snprintf(buf, sizeof(buf), 306 " Platform: 2835 (VC master)"); 307 vchiq_dump(dump_context, buf, len + 1); 308} 309 310VCHIQ_STATUS_T 311vchiq_platform_suspend(VCHIQ_STATE_T *state) 312{ 313 return VCHIQ_ERROR; 314} 315 316VCHIQ_STATUS_T 317vchiq_platform_resume(VCHIQ_STATE_T *state) 318{ 319 return VCHIQ_SUCCESS; 320} 321 322void 323vchiq_platform_paused(VCHIQ_STATE_T *state) 324{ 325} 326 327void 328vchiq_platform_resumed(VCHIQ_STATE_T *state) 329{ 330} 331 332int 333vchiq_platform_videocore_wanted(VCHIQ_STATE_T* state) 334{ 335 return 1; // autosuspend not supported - videocore always wanted 336} 337 338int 339vchiq_platform_use_suspend_timer(void) 340{ 341 return 0; 342} 343void 344vchiq_dump_platform_use_state(VCHIQ_STATE_T *state) 345{ 346 vchiq_log_info(vchiq_arm_log_level, "Suspend timer not in use"); 347} 348void 349vchiq_platform_handle_timeout(VCHIQ_STATE_T *state) 350{ 351 (void)state; 352} 353/* 354 * Local functions 355 */ 356 357/* There is a potential problem with partial cache lines (pages?) 358** at the ends of the block when reading. If the CPU accessed anything in 359** the same line (page?) then it may have pulled old data into the cache, 360** obscuring the new data underneath. We can solve this by transferring the 361** partial cache lines separately, and allowing the ARM to copy into the 362** cached area. 363 364** N.B. This implementation plays slightly fast and loose with the Linux 365** driver programming rules, e.g. its use of __virt_to_bus instead of 366** dma_map_single, but it isn't a multi-platform driver and it benefits 367** from increased speed as a result. 368*/ 369 370static int 371create_pagelist(char __user *buf, size_t count, unsigned short type, 372 struct proc *p, BULKINFO_T *bi) 373{ 374 PAGELIST_T *pagelist; 375 vm_page_t* pages; 376 unsigned long *addrs; 377 unsigned int num_pages, i; 378 vm_offset_t offset; 379 int pagelist_size; 380 char *addr, *base_addr, *next_addr; 381 int run, addridx, actual_pages; 382 int err; 383 vm_paddr_t pagelist_phys; 384 385 offset = (vm_offset_t)buf & (PAGE_SIZE - 1); 386 num_pages = (count + offset + PAGE_SIZE - 1) / PAGE_SIZE; 387 388 bi->pagelist = NULL; 389 bi->buf = buf; 390 bi->size = count; 391 392 /* Allocate enough storage to hold the page pointers and the page 393 ** list 394 */ 395 pagelist_size = sizeof(PAGELIST_T) + 396 (num_pages * sizeof(unsigned long)) + 397 (num_pages * sizeof(pages[0])); 398 399 err = bus_dma_tag_create( 400 NULL, 401 PAGE_SIZE, 0, /* alignment, boundary */ 402 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 403 BUS_SPACE_MAXADDR, /* highaddr */ 404 NULL, NULL, /* filter, filterarg */ 405 pagelist_size, 1, /* maxsize, nsegments */ 406 pagelist_size, 0, /* maxsegsize, flags */ 407 NULL, NULL, /* lockfunc, lockarg */ 408 &bi->pagelist_dma_tag); 409 410 411 412 err = bus_dmamem_alloc(bi->pagelist_dma_tag, (void **)&pagelist, 413 BUS_DMA_COHERENT | BUS_DMA_WAITOK, &bi->pagelist_dma_map); 414 if (err) { 415 vchiq_log_error(vchiq_core_log_level, "Unable to allocate pagelist memory"); 416 err = -ENOMEM; 417 goto failed_alloc; 418 } 419 420 err = bus_dmamap_load(bi->pagelist_dma_tag, bi->pagelist_dma_map, pagelist, 421 pagelist_size, vchiq_dmamap_cb, 422 &pagelist_phys, 0); 423 424 if (err) { 425 vchiq_log_error(vchiq_core_log_level, "cannot load DMA map for pagelist memory"); 426 err = -ENOMEM; 427 goto failed_load; 428 } 429 430 vchiq_log_trace(vchiq_arm_log_level, 431 "create_pagelist - %x", (unsigned int)pagelist); 432 if (!pagelist) 433 return -ENOMEM; 434 435 addrs = pagelist->addrs; 436 pages = (vm_page_t*)(addrs + num_pages); 437 438 actual_pages = vm_fault_quick_hold_pages(&p->p_vmspace->vm_map, 439 (vm_offset_t)buf, count, 440 (type == PAGELIST_READ ? VM_PROT_WRITE : 0 ) | VM_PROT_READ, pages, num_pages); 441 442 if (actual_pages != num_pages) { 443 vm_page_unhold_pages(pages, actual_pages); 444 free(pagelist, M_VCPAGELIST); 445 return (-ENOMEM); 446 } 447 448 pagelist->length = count; 449 pagelist->type = type; 450 pagelist->offset = offset; 451 452 /* Group the pages into runs of contiguous pages */ 453 454 base_addr = (void *)PHYS_TO_VCBUS(VM_PAGE_TO_PHYS(pages[0])); 455 next_addr = base_addr + PAGE_SIZE; 456 addridx = 0; 457 run = 0; 458 459 for (i = 1; i < num_pages; i++) { 460 addr = (void *)PHYS_TO_VCBUS(VM_PAGE_TO_PHYS(pages[i])); 461 if ((addr == next_addr) && (run < (PAGE_SIZE - 1))) { 462 next_addr += PAGE_SIZE; 463 run++; 464 } else { 465 addrs[addridx] = (unsigned long)base_addr + run; 466 addridx++; 467 base_addr = addr; 468 next_addr = addr + PAGE_SIZE; 469 run = 0; 470 } 471 } 472 473 addrs[addridx] = (unsigned long)base_addr + run; 474 addridx++; 475 476 /* Partial cache lines (fragments) require special measures */ 477 if ((type == PAGELIST_READ) && 478 ((pagelist->offset & (CACHE_LINE_SIZE - 1)) || 479 ((pagelist->offset + pagelist->length) & 480 (CACHE_LINE_SIZE - 1)))) { 481 FRAGMENTS_T *fragments; 482 483 if (down_interruptible(&g_free_fragments_sema) != 0) { 484 free(pagelist, M_VCPAGELIST); 485 return -EINTR; 486 } 487 488 WARN_ON(g_free_fragments == NULL); 489 490 down(&g_free_fragments_mutex); 491 fragments = (FRAGMENTS_T *) g_free_fragments; 492 WARN_ON(fragments == NULL); 493 g_free_fragments = *(FRAGMENTS_T **) g_free_fragments; 494 up(&g_free_fragments_mutex); 495 pagelist->type = 496 PAGELIST_READ_WITH_FRAGMENTS + (fragments - 497 g_fragments_base); 498 } 499 500 /* XXX: optimize? INV operation for read WBINV for write? */ 501 cpu_dcache_wbinv_range((vm_offset_t)buf, count); 502 503 bi->pagelist = pagelist; 504 505 return 0; 506 507failed_load: 508 bus_dmamap_unload(bi->pagelist_dma_tag, bi->pagelist_dma_map); 509failed_alloc: 510 bus_dmamap_destroy(bi->pagelist_dma_tag, bi->pagelist_dma_map); 511 bus_dma_tag_destroy(bi->pagelist_dma_tag); 512 513 return err; 514} 515 516static void 517free_pagelist(BULKINFO_T *bi, int actual) 518{ 519 vm_page_t*pages; 520 unsigned int num_pages, i; 521 void *page_address; 522 PAGELIST_T *pagelist; 523 524 pagelist = bi->pagelist; 525 526 vchiq_log_trace(vchiq_arm_log_level, 527 "free_pagelist - %x, %d", (unsigned int)pagelist, actual); 528 529 num_pages = 530 (pagelist->length + pagelist->offset + PAGE_SIZE - 1) / 531 PAGE_SIZE; 532 533 pages = (vm_page_t*)(pagelist->addrs + num_pages); 534 535 /* Deal with any partial cache lines (fragments) */ 536 if (pagelist->type >= PAGELIST_READ_WITH_FRAGMENTS) { 537 FRAGMENTS_T *fragments = g_fragments_base + 538 (pagelist->type - PAGELIST_READ_WITH_FRAGMENTS); 539 int head_bytes, tail_bytes; 540 head_bytes = (CACHE_LINE_SIZE - pagelist->offset) & 541 (CACHE_LINE_SIZE - 1); 542 tail_bytes = (pagelist->offset + actual) & 543 (CACHE_LINE_SIZE - 1); 544 545 if ((actual >= 0) && (head_bytes != 0)) { 546 if (head_bytes > actual) 547 head_bytes = actual; 548 549 memcpy((char *)bi->buf, 550 fragments->headbuf, 551 head_bytes); 552 } 553 554 if ((actual >= 0) && (head_bytes < actual) && 555 (tail_bytes != 0)) { 556 memcpy((char *)bi->buf + actual - tail_bytes, 557 fragments->tailbuf, tail_bytes); 558 } 559 560 down(&g_free_fragments_mutex); 561 *(FRAGMENTS_T **) fragments = g_free_fragments; 562 g_free_fragments = fragments; 563 up(&g_free_fragments_mutex); 564 up(&g_free_fragments_sema); 565 } 566 567 for (i = 0; i < num_pages; i++) { 568 if (pagelist->type != PAGELIST_WRITE) 569 vm_page_dirty(pages[i]); 570 } 571 572 vm_page_unhold_pages(pages, num_pages); 573 574 bus_dmamap_unload(bi->pagelist_dma_tag, bi->pagelist_dma_map); 575 bus_dmamem_free(bi->pagelist_dma_tag, bi->pagelist, bi->pagelist_dma_map); 576 bus_dmamap_destroy(bi->pagelist_dma_tag, bi->pagelist_dma_map); 577 bus_dma_tag_destroy(bi->pagelist_dma_tag); 578 579 free(bi, M_VCPAGELIST); 580} 581