1 /* 2 * DMA Engine test module 3 * 4 * Copyright (C) 2007 Atmel Corporation 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 version 2 as 8 * published by the Free Software Foundation. 9 */ 10 #include <linux/delay.h> 11 #include <linux/dma-mapping.h> 12 #include <linux/dmaengine.h> 13 #include <linux/init.h> 14 #include <linux/kthread.h> 15 #include <linux/module.h> 16 #include <linux/moduleparam.h> 17 #include <linux/random.h> 18 #include <linux/slab.h> 19 #include <linux/wait.h> 20 21 static unsigned int test_buf_size = 16384; 22 module_param(test_buf_size, uint, S_IRUGO); 23 MODULE_PARM_DESC(test_buf_size, "Size of the memcpy test buffer"); 24 25 static char test_channel[20]; 26 module_param_string(channel, test_channel, sizeof(test_channel), S_IRUGO); 27 MODULE_PARM_DESC(channel, "Bus ID of the channel to test (default: any)"); 28 29 static char test_device[20]; 30 module_param_string(device, test_device, sizeof(test_device), S_IRUGO); 31 MODULE_PARM_DESC(device, "Bus ID of the DMA Engine to test (default: any)"); 32 33 static unsigned int threads_per_chan = 1; 34 module_param(threads_per_chan, uint, S_IRUGO); 35 MODULE_PARM_DESC(threads_per_chan, 36 "Number of threads to start per channel (default: 1)"); 37 38 static unsigned int max_channels; 39 module_param(max_channels, uint, S_IRUGO); 40 MODULE_PARM_DESC(max_channels, 41 "Maximum number of channels to use (default: all)"); 42 43 static unsigned int iterations; 44 module_param(iterations, uint, S_IRUGO); 45 MODULE_PARM_DESC(iterations, 46 "Iterations before stopping test (default: infinite)"); 47 48 static unsigned int xor_sources = 3; 49 module_param(xor_sources, uint, S_IRUGO); 50 MODULE_PARM_DESC(xor_sources, 51 "Number of xor source buffers (default: 3)"); 52 53 static unsigned int pq_sources = 3; 54 module_param(pq_sources, uint, S_IRUGO); 55 MODULE_PARM_DESC(pq_sources, 56 "Number of p+q source buffers (default: 3)"); 57 58 static int timeout = 3000; 59 module_param(timeout, uint, S_IRUGO); 60 MODULE_PARM_DESC(timeout, "Transfer Timeout in msec (default: 3000), " 61 "Pass -1 for infinite timeout"); 62 63 /* 64 * Initialization patterns. All bytes in the source buffer has bit 7 65 * set, all bytes in the destination buffer has bit 7 cleared. 66 * 67 * Bit 6 is set for all bytes which are to be copied by the DMA 68 * engine. Bit 5 is set for all bytes which are to be overwritten by 69 * the DMA engine. 70 * 71 * The remaining bits are the inverse of a counter which increments by 72 * one for each byte address. 73 */ 74 #define PATTERN_SRC 0x80 75 #define PATTERN_DST 0x00 76 #define PATTERN_COPY 0x40 77 #define PATTERN_OVERWRITE 0x20 78 #define PATTERN_COUNT_MASK 0x1f 79 80 struct dmatest_thread { 81 struct list_head node; 82 struct task_struct *task; 83 struct dma_chan *chan; 84 u8 **srcs; 85 u8 **dsts; 86 enum dma_transaction_type type; 87 }; 88 89 struct dmatest_chan { 90 struct list_head node; 91 struct dma_chan *chan; 92 struct list_head threads; 93 }; 94 95 /* 96 * These are protected by dma_list_mutex since they're only used by 97 * the DMA filter function callback 98 */ 99 static LIST_HEAD(dmatest_channels); 100 static unsigned int nr_channels; 101 102 static bool dmatest_match_channel(struct dma_chan *chan) 103 { 104 if (test_channel[0] == '\0') 105 return true; 106 return strcmp(dma_chan_name(chan), test_channel) == 0; 107 } 108 109 static bool dmatest_match_device(struct dma_device *device) 110 { 111 if (test_device[0] == '\0') 112 return true; 113 return strcmp(dev_name(device->dev), test_device) == 0; 114 } 115 116 static unsigned long dmatest_random(void) 117 { 118 unsigned long buf; 119 120 get_random_bytes(&buf, sizeof(buf)); 121 return buf; 122 } 123 124 static void dmatest_init_srcs(u8 **bufs, unsigned int start, unsigned int len) 125 { 126 unsigned int i; 127 u8 *buf; 128 129 for (; (buf = *bufs); bufs++) { 130 for (i = 0; i < start; i++) 131 buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK); 132 for ( ; i < start + len; i++) 133 buf[i] = PATTERN_SRC | PATTERN_COPY 134 | (~i & PATTERN_COUNT_MASK); 135 for ( ; i < test_buf_size; i++) 136 buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK); 137 buf++; 138 } 139 } 140 141 static void dmatest_init_dsts(u8 **bufs, unsigned int start, unsigned int len) 142 { 143 unsigned int i; 144 u8 *buf; 145 146 for (; (buf = *bufs); bufs++) { 147 for (i = 0; i < start; i++) 148 buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK); 149 for ( ; i < start + len; i++) 150 buf[i] = PATTERN_DST | PATTERN_OVERWRITE 151 | (~i & PATTERN_COUNT_MASK); 152 for ( ; i < test_buf_size; i++) 153 buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK); 154 } 155 } 156 157 static void dmatest_mismatch(u8 actual, u8 pattern, unsigned int index, 158 unsigned int counter, bool is_srcbuf) 159 { 160 u8 diff = actual ^ pattern; 161 u8 expected = pattern | (~counter & PATTERN_COUNT_MASK); 162 const char *thread_name = current->comm; 163 164 if (is_srcbuf) 165 pr_warning("%s: srcbuf[0x%x] overwritten!" 166 " Expected %02x, got %02x\n", 167 thread_name, index, expected, actual); 168 else if ((pattern & PATTERN_COPY) 169 && (diff & (PATTERN_COPY | PATTERN_OVERWRITE))) 170 pr_warning("%s: dstbuf[0x%x] not copied!" 171 " Expected %02x, got %02x\n", 172 thread_name, index, expected, actual); 173 else if (diff & PATTERN_SRC) 174 pr_warning("%s: dstbuf[0x%x] was copied!" 175 " Expected %02x, got %02x\n", 176 thread_name, index, expected, actual); 177 else 178 pr_warning("%s: dstbuf[0x%x] mismatch!" 179 " Expected %02x, got %02x\n", 180 thread_name, index, expected, actual); 181 } 182 183 static unsigned int dmatest_verify(u8 **bufs, unsigned int start, 184 unsigned int end, unsigned int counter, u8 pattern, 185 bool is_srcbuf) 186 { 187 unsigned int i; 188 unsigned int error_count = 0; 189 u8 actual; 190 u8 expected; 191 u8 *buf; 192 unsigned int counter_orig = counter; 193 194 for (; (buf = *bufs); bufs++) { 195 counter = counter_orig; 196 for (i = start; i < end; i++) { 197 actual = buf[i]; 198 expected = pattern | (~counter & PATTERN_COUNT_MASK); 199 if (actual != expected) { 200 if (error_count < 32) 201 dmatest_mismatch(actual, pattern, i, 202 counter, is_srcbuf); 203 error_count++; 204 } 205 counter++; 206 } 207 } 208 209 if (error_count > 32) 210 pr_warning("%s: %u errors suppressed\n", 211 current->comm, error_count - 32); 212 213 return error_count; 214 } 215 216 static void dmatest_callback(void *completion) 217 { 218 complete(completion); 219 } 220 221 /* 222 * This function repeatedly tests DMA transfers of various lengths and 223 * offsets for a given operation type until it is told to exit by 224 * kthread_stop(). There may be multiple threads running this function 225 * in parallel for a single channel, and there may be multiple channels 226 * being tested in parallel. 227 * 228 * Before each test, the source and destination buffer is initialized 229 * with a known pattern. This pattern is different depending on 230 * whether it's in an area which is supposed to be copied or 231 * overwritten, and different in the source and destination buffers. 232 * So if the DMA engine doesn't copy exactly what we tell it to copy, 233 * we'll notice. 234 */ 235 static int dmatest_func(void *data) 236 { 237 struct dmatest_thread *thread = data; 238 struct dma_chan *chan; 239 const char *thread_name; 240 unsigned int src_off, dst_off, len; 241 unsigned int error_count; 242 unsigned int failed_tests = 0; 243 unsigned int total_tests = 0; 244 dma_cookie_t cookie; 245 enum dma_status status; 246 enum dma_ctrl_flags flags; 247 u8 pq_coefs[pq_sources + 1]; 248 int ret; 249 int src_cnt; 250 int dst_cnt; 251 int i; 252 253 thread_name = current->comm; 254 255 ret = -ENOMEM; 256 257 smp_rmb(); 258 chan = thread->chan; 259 if (thread->type == DMA_MEMCPY) 260 src_cnt = dst_cnt = 1; 261 else if (thread->type == DMA_XOR) { 262 src_cnt = xor_sources | 1; /* force odd to ensure dst = src */ 263 dst_cnt = 1; 264 } else if (thread->type == DMA_PQ) { 265 src_cnt = pq_sources | 1; /* force odd to ensure dst = src */ 266 dst_cnt = 2; 267 for (i = 0; i < src_cnt; i++) 268 pq_coefs[i] = 1; 269 } else 270 goto err_srcs; 271 272 thread->srcs = kcalloc(src_cnt+1, sizeof(u8 *), GFP_KERNEL); 273 if (!thread->srcs) 274 goto err_srcs; 275 for (i = 0; i < src_cnt; i++) { 276 thread->srcs[i] = kmalloc(test_buf_size, GFP_KERNEL); 277 if (!thread->srcs[i]) 278 goto err_srcbuf; 279 } 280 thread->srcs[i] = NULL; 281 282 thread->dsts = kcalloc(dst_cnt+1, sizeof(u8 *), GFP_KERNEL); 283 if (!thread->dsts) 284 goto err_dsts; 285 for (i = 0; i < dst_cnt; i++) { 286 thread->dsts[i] = kmalloc(test_buf_size, GFP_KERNEL); 287 if (!thread->dsts[i]) 288 goto err_dstbuf; 289 } 290 thread->dsts[i] = NULL; 291 292 set_user_nice(current, 10); 293 294 /* 295 * src buffers are freed by the DMAEngine code with dma_unmap_single() 296 * dst buffers are freed by ourselves below 297 */ 298 flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT 299 | DMA_COMPL_SKIP_DEST_UNMAP | DMA_COMPL_SRC_UNMAP_SINGLE; 300 301 while (!kthread_should_stop() 302 && !(iterations && total_tests >= iterations)) { 303 struct dma_device *dev = chan->device; 304 struct dma_async_tx_descriptor *tx = NULL; 305 dma_addr_t dma_srcs[src_cnt]; 306 dma_addr_t dma_dsts[dst_cnt]; 307 struct completion cmp; 308 unsigned long tmo = msecs_to_jiffies(timeout); 309 u8 align = 0; 310 311 total_tests++; 312 313 /* honor alignment restrictions */ 314 if (thread->type == DMA_MEMCPY) 315 align = dev->copy_align; 316 else if (thread->type == DMA_XOR) 317 align = dev->xor_align; 318 else if (thread->type == DMA_PQ) 319 align = dev->pq_align; 320 321 if (1 << align > test_buf_size) { 322 pr_err("%u-byte buffer too small for %d-byte alignment\n", 323 test_buf_size, 1 << align); 324 break; 325 } 326 327 len = dmatest_random() % test_buf_size + 1; 328 len = (len >> align) << align; 329 if (!len) 330 len = 1 << align; 331 src_off = dmatest_random() % (test_buf_size - len + 1); 332 dst_off = dmatest_random() % (test_buf_size - len + 1); 333 334 src_off = (src_off >> align) << align; 335 dst_off = (dst_off >> align) << align; 336 337 dmatest_init_srcs(thread->srcs, src_off, len); 338 dmatest_init_dsts(thread->dsts, dst_off, len); 339 340 for (i = 0; i < src_cnt; i++) { 341 u8 *buf = thread->srcs[i] + src_off; 342 343 dma_srcs[i] = dma_map_single(dev->dev, buf, len, 344 DMA_TO_DEVICE); 345 } 346 /* map with DMA_BIDIRECTIONAL to force writeback/invalidate */ 347 for (i = 0; i < dst_cnt; i++) { 348 dma_dsts[i] = dma_map_single(dev->dev, thread->dsts[i], 349 test_buf_size, 350 DMA_BIDIRECTIONAL); 351 } 352 353 354 if (thread->type == DMA_MEMCPY) 355 tx = dev->device_prep_dma_memcpy(chan, 356 dma_dsts[0] + dst_off, 357 dma_srcs[0], len, 358 flags); 359 else if (thread->type == DMA_XOR) 360 tx = dev->device_prep_dma_xor(chan, 361 dma_dsts[0] + dst_off, 362 dma_srcs, src_cnt, 363 len, flags); 364 else if (thread->type == DMA_PQ) { 365 dma_addr_t dma_pq[dst_cnt]; 366 367 for (i = 0; i < dst_cnt; i++) 368 dma_pq[i] = dma_dsts[i] + dst_off; 369 tx = dev->device_prep_dma_pq(chan, dma_pq, dma_srcs, 370 src_cnt, pq_coefs, 371 len, flags); 372 } 373 374 if (!tx) { 375 for (i = 0; i < src_cnt; i++) 376 dma_unmap_single(dev->dev, dma_srcs[i], len, 377 DMA_TO_DEVICE); 378 for (i = 0; i < dst_cnt; i++) 379 dma_unmap_single(dev->dev, dma_dsts[i], 380 test_buf_size, 381 DMA_BIDIRECTIONAL); 382 pr_warning("%s: #%u: prep error with src_off=0x%x " 383 "dst_off=0x%x len=0x%x\n", 384 thread_name, total_tests - 1, 385 src_off, dst_off, len); 386 msleep(100); 387 failed_tests++; 388 continue; 389 } 390 391 init_completion(&cmp); 392 tx->callback = dmatest_callback; 393 tx->callback_param = &cmp; 394 cookie = tx->tx_submit(tx); 395 396 if (dma_submit_error(cookie)) { 397 pr_warning("%s: #%u: submit error %d with src_off=0x%x " 398 "dst_off=0x%x len=0x%x\n", 399 thread_name, total_tests - 1, cookie, 400 src_off, dst_off, len); 401 msleep(100); 402 failed_tests++; 403 continue; 404 } 405 dma_async_issue_pending(chan); 406 407 tmo = wait_for_completion_timeout(&cmp, tmo); 408 status = dma_async_is_tx_complete(chan, cookie, NULL, NULL); 409 410 if (tmo == 0) { 411 pr_warning("%s: #%u: test timed out\n", 412 thread_name, total_tests - 1); 413 failed_tests++; 414 continue; 415 } else if (status != DMA_SUCCESS) { 416 pr_warning("%s: #%u: got completion callback," 417 " but status is \'%s\'\n", 418 thread_name, total_tests - 1, 419 status == DMA_ERROR ? "error" : "in progress"); 420 failed_tests++; 421 continue; 422 } 423 424 /* Unmap by myself (see DMA_COMPL_SKIP_DEST_UNMAP above) */ 425 for (i = 0; i < dst_cnt; i++) 426 dma_unmap_single(dev->dev, dma_dsts[i], test_buf_size, 427 DMA_BIDIRECTIONAL); 428 429 error_count = 0; 430 431 pr_debug("%s: verifying source buffer...\n", thread_name); 432 error_count += dmatest_verify(thread->srcs, 0, src_off, 433 0, PATTERN_SRC, true); 434 error_count += dmatest_verify(thread->srcs, src_off, 435 src_off + len, src_off, 436 PATTERN_SRC | PATTERN_COPY, true); 437 error_count += dmatest_verify(thread->srcs, src_off + len, 438 test_buf_size, src_off + len, 439 PATTERN_SRC, true); 440 441 pr_debug("%s: verifying dest buffer...\n", 442 thread->task->comm); 443 error_count += dmatest_verify(thread->dsts, 0, dst_off, 444 0, PATTERN_DST, false); 445 error_count += dmatest_verify(thread->dsts, dst_off, 446 dst_off + len, src_off, 447 PATTERN_SRC | PATTERN_COPY, false); 448 error_count += dmatest_verify(thread->dsts, dst_off + len, 449 test_buf_size, dst_off + len, 450 PATTERN_DST, false); 451 452 if (error_count) { 453 pr_warning("%s: #%u: %u errors with " 454 "src_off=0x%x dst_off=0x%x len=0x%x\n", 455 thread_name, total_tests - 1, error_count, 456 src_off, dst_off, len); 457 failed_tests++; 458 } else { 459 pr_debug("%s: #%u: No errors with " 460 "src_off=0x%x dst_off=0x%x len=0x%x\n", 461 thread_name, total_tests - 1, 462 src_off, dst_off, len); 463 } 464 } 465 466 ret = 0; 467 for (i = 0; thread->dsts[i]; i++) 468 kfree(thread->dsts[i]); 469 err_dstbuf: 470 kfree(thread->dsts); 471 err_dsts: 472 for (i = 0; thread->srcs[i]; i++) 473 kfree(thread->srcs[i]); 474 err_srcbuf: 475 kfree(thread->srcs); 476 err_srcs: 477 pr_notice("%s: terminating after %u tests, %u failures (status %d)\n", 478 thread_name, total_tests, failed_tests, ret); 479 480 if (iterations > 0) 481 while (!kthread_should_stop()) { 482 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wait_dmatest_exit); 483 interruptible_sleep_on(&wait_dmatest_exit); 484 } 485 486 return ret; 487 } 488 489 static void dmatest_cleanup_channel(struct dmatest_chan *dtc) 490 { 491 struct dmatest_thread *thread; 492 struct dmatest_thread *_thread; 493 int ret; 494 495 list_for_each_entry_safe(thread, _thread, &dtc->threads, node) { 496 ret = kthread_stop(thread->task); 497 pr_debug("dmatest: thread %s exited with status %d\n", 498 thread->task->comm, ret); 499 list_del(&thread->node); 500 kfree(thread); 501 } 502 kfree(dtc); 503 } 504 505 static int dmatest_add_threads(struct dmatest_chan *dtc, enum dma_transaction_type type) 506 { 507 struct dmatest_thread *thread; 508 struct dma_chan *chan = dtc->chan; 509 char *op; 510 unsigned int i; 511 512 if (type == DMA_MEMCPY) 513 op = "copy"; 514 else if (type == DMA_XOR) 515 op = "xor"; 516 else if (type == DMA_PQ) 517 op = "pq"; 518 else 519 return -EINVAL; 520 521 for (i = 0; i < threads_per_chan; i++) { 522 thread = kzalloc(sizeof(struct dmatest_thread), GFP_KERNEL); 523 if (!thread) { 524 pr_warning("dmatest: No memory for %s-%s%u\n", 525 dma_chan_name(chan), op, i); 526 527 break; 528 } 529 thread->chan = dtc->chan; 530 thread->type = type; 531 smp_wmb(); 532 thread->task = kthread_run(dmatest_func, thread, "%s-%s%u", 533 dma_chan_name(chan), op, i); 534 if (IS_ERR(thread->task)) { 535 pr_warning("dmatest: Failed to run thread %s-%s%u\n", 536 dma_chan_name(chan), op, i); 537 kfree(thread); 538 break; 539 } 540 541 /* srcbuf and dstbuf are allocated by the thread itself */ 542 543 list_add_tail(&thread->node, &dtc->threads); 544 } 545 546 return i; 547 } 548 549 static int dmatest_add_channel(struct dma_chan *chan) 550 { 551 struct dmatest_chan *dtc; 552 struct dma_device *dma_dev = chan->device; 553 unsigned int thread_count = 0; 554 int cnt; 555 556 dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL); 557 if (!dtc) { 558 pr_warning("dmatest: No memory for %s\n", dma_chan_name(chan)); 559 return -ENOMEM; 560 } 561 562 dtc->chan = chan; 563 INIT_LIST_HEAD(&dtc->threads); 564 565 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) { 566 cnt = dmatest_add_threads(dtc, DMA_MEMCPY); 567 thread_count += cnt > 0 ? cnt : 0; 568 } 569 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) { 570 cnt = dmatest_add_threads(dtc, DMA_XOR); 571 thread_count += cnt > 0 ? cnt : 0; 572 } 573 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) { 574 cnt = dmatest_add_threads(dtc, DMA_PQ); 575 thread_count += cnt > 0 ?: 0; 576 } 577 578 pr_info("dmatest: Started %u threads using %s\n", 579 thread_count, dma_chan_name(chan)); 580 581 list_add_tail(&dtc->node, &dmatest_channels); 582 nr_channels++; 583 584 return 0; 585 } 586 587 static bool filter(struct dma_chan *chan, void *param) 588 { 589 if (!dmatest_match_channel(chan) || !dmatest_match_device(chan->device)) 590 return false; 591 else 592 return true; 593 } 594 595 static int __init dmatest_init(void) 596 { 597 dma_cap_mask_t mask; 598 struct dma_chan *chan; 599 int err = 0; 600 601 dma_cap_zero(mask); 602 dma_cap_set(DMA_MEMCPY, mask); 603 for (;;) { 604 chan = dma_request_channel(mask, filter, NULL); 605 if (chan) { 606 err = dmatest_add_channel(chan); 607 if (err) { 608 dma_release_channel(chan); 609 break; /* add_channel failed, punt */ 610 } 611 } else 612 break; /* no more channels available */ 613 if (max_channels && nr_channels >= max_channels) 614 break; /* we have all we need */ 615 } 616 617 return err; 618 } 619 /* when compiled-in wait for drivers to load first */ 620 late_initcall(dmatest_init); 621 622 static void __exit dmatest_exit(void) 623 { 624 struct dmatest_chan *dtc, *_dtc; 625 struct dma_chan *chan; 626 627 list_for_each_entry_safe(dtc, _dtc, &dmatest_channels, node) { 628 list_del(&dtc->node); 629 chan = dtc->chan; 630 dmatest_cleanup_channel(dtc); 631 pr_debug("dmatest: dropped channel %s\n", 632 dma_chan_name(chan)); 633 dma_release_channel(chan); 634 } 635 } 636 module_exit(dmatest_exit); 637 638 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)"); 639 MODULE_LICENSE("GPL v2"); 640