1 /*- 2 * Copyright (C) 2012 Intel Corporation 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 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 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/bus.h> 33 #include <sys/conf.h> 34 #include <sys/ioccom.h> 35 #include <sys/kernel.h> 36 #include <sys/lock.h> 37 #include <sys/malloc.h> 38 #include <sys/module.h> 39 #include <sys/mutex.h> 40 #include <sys/rman.h> 41 #include <sys/sysctl.h> 42 #include <dev/pci/pcireg.h> 43 #include <dev/pci/pcivar.h> 44 #include <machine/bus.h> 45 #include <machine/resource.h> 46 #include <machine/stdarg.h> 47 #include <vm/vm.h> 48 #include <vm/vm_param.h> 49 #include <vm/pmap.h> 50 51 #include "ioat.h" 52 #include "ioat_hw.h" 53 #include "ioat_internal.h" 54 #include "ioat_test.h" 55 56 #ifndef time_after 57 #define time_after(a,b) ((long)(b) - (long)(a) < 0) 58 #endif 59 60 MALLOC_DEFINE(M_IOAT_TEST, "ioat_test", "ioat test allocations"); 61 62 #define IOAT_MAX_BUFS 256 63 64 struct test_transaction { 65 void *buf[IOAT_MAX_BUFS]; 66 uint32_t length; 67 uint32_t depth; 68 uint32_t crc[IOAT_MAX_BUFS]; 69 struct ioat_test *test; 70 TAILQ_ENTRY(test_transaction) entry; 71 }; 72 73 #define IT_LOCK() mtx_lock(&ioat_test_lk) 74 #define IT_UNLOCK() mtx_unlock(&ioat_test_lk) 75 #define IT_ASSERT() mtx_assert(&ioat_test_lk, MA_OWNED) 76 static struct mtx ioat_test_lk; 77 MTX_SYSINIT(ioat_test_lk, &ioat_test_lk, "test coordination mtx", MTX_DEF); 78 79 static int g_thread_index = 1; 80 static struct cdev *g_ioat_cdev = NULL; 81 82 #define ioat_test_log(v, ...) _ioat_test_log((v), "ioat_test: " __VA_ARGS__) 83 static void _ioat_test_log(int verbosity, const char *fmt, ...); 84 85 static void 86 ioat_test_transaction_destroy(struct test_transaction *tx) 87 { 88 struct ioat_test *test; 89 int i; 90 91 test = tx->test; 92 93 for (i = 0; i < IOAT_MAX_BUFS; i++) { 94 if (tx->buf[i] != NULL) { 95 if (test->testkind == IOAT_TEST_DMA_8K) 96 free(tx->buf[i], M_IOAT_TEST); 97 else 98 contigfree(tx->buf[i], tx->length, M_IOAT_TEST); 99 tx->buf[i] = NULL; 100 } 101 } 102 103 free(tx, M_IOAT_TEST); 104 } 105 106 static struct 107 test_transaction *ioat_test_transaction_create(struct ioat_test *test, 108 unsigned num_buffers) 109 { 110 struct test_transaction *tx; 111 unsigned i; 112 113 tx = malloc(sizeof(*tx), M_IOAT_TEST, M_NOWAIT | M_ZERO); 114 if (tx == NULL) 115 return (NULL); 116 117 tx->length = test->buffer_size; 118 119 for (i = 0; i < num_buffers; i++) { 120 if (test->testkind == IOAT_TEST_DMA_8K) 121 tx->buf[i] = malloc(test->buffer_size, M_IOAT_TEST, 122 M_NOWAIT); 123 else 124 tx->buf[i] = contigmalloc(test->buffer_size, 125 M_IOAT_TEST, M_NOWAIT, 0, BUS_SPACE_MAXADDR, 126 PAGE_SIZE, 0); 127 128 if (tx->buf[i] == NULL) { 129 ioat_test_transaction_destroy(tx); 130 return (NULL); 131 } 132 } 133 return (tx); 134 } 135 136 static void 137 dump_hex(void *p, size_t chunks) 138 { 139 size_t i, j; 140 141 for (i = 0; i < chunks; i++) { 142 for (j = 0; j < 8; j++) 143 printf("%08x ", ((uint32_t *)p)[i * 8 + j]); 144 printf("\n"); 145 } 146 } 147 148 static bool 149 ioat_compare_ok(struct test_transaction *tx) 150 { 151 struct ioat_test *test; 152 char *dst, *src; 153 uint32_t i, j; 154 155 test = tx->test; 156 157 for (i = 0; i < tx->depth; i++) { 158 dst = tx->buf[2 * i + 1]; 159 src = tx->buf[2 * i]; 160 161 if (test->testkind == IOAT_TEST_FILL) { 162 for (j = 0; j < tx->length; j += sizeof(uint64_t)) { 163 if (memcmp(src, &dst[j], 164 MIN(sizeof(uint64_t), tx->length - j)) 165 != 0) 166 return (false); 167 } 168 } else if (test->testkind == IOAT_TEST_DMA) { 169 if (memcmp(src, dst, tx->length) != 0) 170 return (false); 171 } else if (test->testkind == IOAT_TEST_RAW_DMA) { 172 if (test->raw_write) 173 dst = test->raw_vtarget; 174 dump_hex(dst, tx->length / 32); 175 } 176 } 177 return (true); 178 } 179 180 static void 181 ioat_dma_test_callback(void *arg, int error) 182 { 183 struct test_transaction *tx; 184 struct ioat_test *test; 185 186 if (error != 0) 187 ioat_test_log(0, "%s: Got error: %d\n", __func__, error); 188 189 tx = arg; 190 test = tx->test; 191 192 if (test->verify && !ioat_compare_ok(tx)) { 193 ioat_test_log(0, "miscompare found\n"); 194 atomic_add_32(&test->status[IOAT_TEST_MISCOMPARE], tx->depth); 195 } else if (!test->too_late) 196 atomic_add_32(&test->status[IOAT_TEST_OK], tx->depth); 197 198 IT_LOCK(); 199 TAILQ_REMOVE(&test->pend_q, tx, entry); 200 TAILQ_INSERT_TAIL(&test->free_q, tx, entry); 201 wakeup(&test->free_q); 202 IT_UNLOCK(); 203 } 204 205 static int 206 ioat_test_prealloc_memory(struct ioat_test *test, int index) 207 { 208 uint32_t i, j, k; 209 struct test_transaction *tx; 210 211 for (i = 0; i < test->transactions; i++) { 212 tx = ioat_test_transaction_create(test, test->chain_depth * 2); 213 if (tx == NULL) { 214 ioat_test_log(0, "tx == NULL - memory exhausted\n"); 215 test->status[IOAT_TEST_NO_MEMORY]++; 216 return (ENOMEM); 217 } 218 219 TAILQ_INSERT_HEAD(&test->free_q, tx, entry); 220 221 tx->test = test; 222 tx->depth = test->chain_depth; 223 224 /* fill in source buffers */ 225 for (j = 0; j < (tx->length / sizeof(uint32_t)); j++) { 226 uint32_t val = j + (index << 28); 227 228 for (k = 0; k < test->chain_depth; k++) { 229 ((uint32_t *)tx->buf[2*k])[j] = ~val; 230 ((uint32_t *)tx->buf[2*k+1])[j] = val; 231 } 232 } 233 } 234 return (0); 235 } 236 237 static void 238 ioat_test_release_memory(struct ioat_test *test) 239 { 240 struct test_transaction *tx, *s; 241 242 TAILQ_FOREACH_SAFE(tx, &test->free_q, entry, s) 243 ioat_test_transaction_destroy(tx); 244 TAILQ_INIT(&test->free_q); 245 246 TAILQ_FOREACH_SAFE(tx, &test->pend_q, entry, s) 247 ioat_test_transaction_destroy(tx); 248 TAILQ_INIT(&test->pend_q); 249 } 250 251 static void 252 ioat_test_submit_1_tx(struct ioat_test *test, bus_dmaengine_t dma) 253 { 254 struct test_transaction *tx; 255 struct bus_dmadesc *desc; 256 bus_dmaengine_callback_t cb; 257 bus_addr_t src, dest; 258 uint64_t fillpattern; 259 uint32_t i, flags; 260 261 desc = NULL; 262 263 IT_LOCK(); 264 while (TAILQ_EMPTY(&test->free_q)) 265 msleep(&test->free_q, &ioat_test_lk, 0, "test_submit", 0); 266 267 tx = TAILQ_FIRST(&test->free_q); 268 TAILQ_REMOVE(&test->free_q, tx, entry); 269 TAILQ_INSERT_HEAD(&test->pend_q, tx, entry); 270 IT_UNLOCK(); 271 272 if (test->testkind != IOAT_TEST_MEMCPY) 273 ioat_acquire(dma); 274 for (i = 0; i < tx->depth; i++) { 275 if (test->testkind == IOAT_TEST_MEMCPY) { 276 memcpy(tx->buf[2 * i + 1], tx->buf[2 * i], tx->length); 277 if (i == tx->depth - 1) 278 ioat_dma_test_callback(tx, 0); 279 continue; 280 } 281 282 src = vtophys((vm_offset_t)tx->buf[2*i]); 283 dest = vtophys((vm_offset_t)tx->buf[2*i+1]); 284 285 if (test->testkind == IOAT_TEST_RAW_DMA) { 286 if (test->raw_write) 287 dest = test->raw_target; 288 else 289 src = test->raw_target; 290 } 291 292 if (i == tx->depth - 1) { 293 cb = ioat_dma_test_callback; 294 flags = DMA_INT_EN; 295 } else { 296 cb = NULL; 297 flags = 0; 298 } 299 300 if (test->testkind == IOAT_TEST_DMA || 301 test->testkind == IOAT_TEST_RAW_DMA) 302 desc = ioat_copy(dma, dest, src, tx->length, cb, tx, 303 flags); 304 else if (test->testkind == IOAT_TEST_FILL) { 305 fillpattern = *(uint64_t *)tx->buf[2*i]; 306 desc = ioat_blockfill(dma, dest, fillpattern, 307 tx->length, cb, tx, flags); 308 } else if (test->testkind == IOAT_TEST_DMA_8K) { 309 bus_addr_t src2, dst2; 310 311 src2 = vtophys((vm_offset_t)tx->buf[2*i] + PAGE_SIZE); 312 dst2 = vtophys((vm_offset_t)tx->buf[2*i+1] + PAGE_SIZE); 313 314 desc = ioat_copy_8k_aligned(dma, dest, dst2, src, src2, 315 cb, tx, flags); 316 } else if (test->testkind == IOAT_TEST_DMA_8K_PB) { 317 bus_addr_t src2, dst2; 318 319 src2 = vtophys((vm_offset_t)tx->buf[2*i+1] + PAGE_SIZE); 320 dst2 = vtophys((vm_offset_t)tx->buf[2*i] + PAGE_SIZE); 321 322 desc = ioat_copy_8k_aligned(dma, dest, dst2, src, src2, 323 cb, tx, flags); 324 } else if (test->testkind == IOAT_TEST_DMA_CRC) { 325 bus_addr_t crc; 326 327 tx->crc[i] = 0; 328 crc = vtophys((vm_offset_t)&tx->crc[i]); 329 desc = ioat_crc(dma, src, tx->length, 330 NULL, crc, cb, tx, flags | DMA_CRC_STORE); 331 } else if (test->testkind == IOAT_TEST_DMA_CRC_COPY) { 332 bus_addr_t crc; 333 334 tx->crc[i] = 0; 335 crc = vtophys((vm_offset_t)&tx->crc[i]); 336 desc = ioat_copy_crc(dma, dest, src, tx->length, 337 NULL, crc, cb, tx, flags | DMA_CRC_STORE); 338 } 339 if (desc == NULL) 340 break; 341 } 342 if (test->testkind == IOAT_TEST_MEMCPY) 343 return; 344 ioat_release(dma); 345 346 /* 347 * We couldn't issue an IO -- either the device is being detached or 348 * the HW reset. Essentially spin until the device comes back up or 349 * our timer expires. 350 */ 351 if (desc == NULL && tx->depth > 0) { 352 atomic_add_32(&test->status[IOAT_TEST_NO_DMA_ENGINE], tx->depth); 353 IT_LOCK(); 354 TAILQ_REMOVE(&test->pend_q, tx, entry); 355 TAILQ_INSERT_HEAD(&test->free_q, tx, entry); 356 IT_UNLOCK(); 357 } 358 } 359 360 static void 361 ioat_dma_test(void *arg) 362 { 363 struct ioat_softc *ioat; 364 struct ioat_test *test; 365 bus_dmaengine_t dmaengine; 366 uint32_t loops; 367 int index, rc, start, end, error; 368 369 test = arg; 370 memset(__DEVOLATILE(void *, test->status), 0, sizeof(test->status)); 371 372 if ((test->testkind == IOAT_TEST_DMA_8K || 373 test->testkind == IOAT_TEST_DMA_8K_PB) && 374 test->buffer_size != 2 * PAGE_SIZE) { 375 ioat_test_log(0, "Asked for 8k test and buffer size isn't 8k\n"); 376 test->status[IOAT_TEST_INVALID_INPUT]++; 377 return; 378 } 379 380 if (test->buffer_size > 1024 * 1024) { 381 ioat_test_log(0, "Buffer size too large >1MB\n"); 382 test->status[IOAT_TEST_NO_MEMORY]++; 383 return; 384 } 385 386 if (test->chain_depth * 2 > IOAT_MAX_BUFS) { 387 ioat_test_log(0, "Depth too large (> %u)\n", 388 (unsigned)IOAT_MAX_BUFS / 2); 389 test->status[IOAT_TEST_NO_MEMORY]++; 390 return; 391 } 392 393 if (btoc((uint64_t)test->buffer_size * test->chain_depth * 394 test->transactions) > (physmem / 4)) { 395 ioat_test_log(0, "Sanity check failed -- test would " 396 "use more than 1/4 of phys mem.\n"); 397 test->status[IOAT_TEST_NO_MEMORY]++; 398 return; 399 } 400 401 if ((uint64_t)test->transactions * test->chain_depth > (1<<16)) { 402 ioat_test_log(0, "Sanity check failed -- test would " 403 "use more than available IOAT ring space.\n"); 404 test->status[IOAT_TEST_NO_MEMORY]++; 405 return; 406 } 407 408 if (test->testkind >= IOAT_NUM_TESTKINDS) { 409 ioat_test_log(0, "Invalid kind %u\n", 410 (unsigned)test->testkind); 411 test->status[IOAT_TEST_INVALID_INPUT]++; 412 return; 413 } 414 415 dmaengine = ioat_get_dmaengine(test->channel_index, M_NOWAIT); 416 if (dmaengine == NULL) { 417 ioat_test_log(0, "Couldn't acquire dmaengine\n"); 418 test->status[IOAT_TEST_NO_DMA_ENGINE]++; 419 return; 420 } 421 ioat = to_ioat_softc(dmaengine); 422 423 if (test->testkind == IOAT_TEST_FILL && 424 (ioat->capabilities & IOAT_DMACAP_BFILL) == 0) 425 { 426 ioat_test_log(0, 427 "Hardware doesn't support block fill, aborting test\n"); 428 test->status[IOAT_TEST_INVALID_INPUT]++; 429 goto out; 430 } 431 432 if (test->coalesce_period > ioat->intrdelay_max) { 433 ioat_test_log(0, 434 "Hardware doesn't support intrdelay of %u us.\n", 435 (unsigned)test->coalesce_period); 436 test->status[IOAT_TEST_INVALID_INPUT]++; 437 goto out; 438 } 439 error = ioat_set_interrupt_coalesce(dmaengine, test->coalesce_period); 440 if (error == ENODEV && test->coalesce_period == 0) 441 error = 0; 442 if (error != 0) { 443 ioat_test_log(0, "ioat_set_interrupt_coalesce: %d\n", error); 444 test->status[IOAT_TEST_INVALID_INPUT]++; 445 goto out; 446 } 447 448 if (test->zero_stats) 449 memset(&ioat->stats, 0, sizeof(ioat->stats)); 450 451 if (test->testkind == IOAT_TEST_RAW_DMA) { 452 if (test->raw_is_virtual) { 453 test->raw_vtarget = (void *)test->raw_target; 454 test->raw_target = vtophys(test->raw_vtarget); 455 } else { 456 test->raw_vtarget = pmap_mapdev(test->raw_target, 457 test->buffer_size); 458 } 459 } 460 461 index = g_thread_index++; 462 TAILQ_INIT(&test->free_q); 463 TAILQ_INIT(&test->pend_q); 464 465 if (test->duration == 0) 466 ioat_test_log(1, "Thread %d: num_loops remaining: 0x%08x\n", 467 index, test->transactions); 468 else 469 ioat_test_log(1, "Thread %d: starting\n", index); 470 471 rc = ioat_test_prealloc_memory(test, index); 472 if (rc != 0) { 473 ioat_test_log(0, "prealloc_memory: %d\n", rc); 474 goto out; 475 } 476 wmb(); 477 478 test->too_late = false; 479 start = ticks; 480 end = start + (((sbintime_t)test->duration * hz) / 1000); 481 482 for (loops = 0;; loops++) { 483 if (test->duration == 0 && loops >= test->transactions) 484 break; 485 else if (test->duration != 0 && time_after(ticks, end)) { 486 test->too_late = true; 487 break; 488 } 489 490 ioat_test_submit_1_tx(test, dmaengine); 491 } 492 493 ioat_test_log(1, "Test Elapsed: %d ticks (overrun %d), %d sec.\n", 494 ticks - start, ticks - end, (ticks - start) / hz); 495 496 IT_LOCK(); 497 while (!TAILQ_EMPTY(&test->pend_q)) 498 msleep(&test->free_q, &ioat_test_lk, 0, "ioattestcompl", hz); 499 IT_UNLOCK(); 500 501 ioat_test_log(1, "Test Elapsed2: %d ticks (overrun %d), %d sec.\n", 502 ticks - start, ticks - end, (ticks - start) / hz); 503 504 ioat_test_release_memory(test); 505 out: 506 if (test->testkind == IOAT_TEST_RAW_DMA && !test->raw_is_virtual) 507 pmap_unmapdev((vm_offset_t)test->raw_vtarget, 508 test->buffer_size); 509 ioat_put_dmaengine(dmaengine); 510 } 511 512 static int 513 ioat_test_open(struct cdev *dev, int flags, int fmt, struct thread *td) 514 { 515 516 return (0); 517 } 518 519 static int 520 ioat_test_close(struct cdev *dev, int flags, int fmt, struct thread *td) 521 { 522 523 return (0); 524 } 525 526 static int 527 ioat_test_ioctl(struct cdev *dev, unsigned long cmd, caddr_t arg, int flag, 528 struct thread *td) 529 { 530 531 switch (cmd) { 532 case IOAT_DMATEST: 533 ioat_dma_test(arg); 534 break; 535 default: 536 return (EINVAL); 537 } 538 return (0); 539 } 540 541 static struct cdevsw ioat_cdevsw = { 542 .d_version = D_VERSION, 543 .d_flags = 0, 544 .d_open = ioat_test_open, 545 .d_close = ioat_test_close, 546 .d_ioctl = ioat_test_ioctl, 547 .d_name = "ioat_test", 548 }; 549 550 static int 551 enable_ioat_test(bool enable) 552 { 553 554 mtx_assert(&Giant, MA_OWNED); 555 556 if (enable && g_ioat_cdev == NULL) { 557 g_ioat_cdev = make_dev(&ioat_cdevsw, 0, UID_ROOT, GID_WHEEL, 558 0600, "ioat_test"); 559 } else if (!enable && g_ioat_cdev != NULL) { 560 destroy_dev(g_ioat_cdev); 561 g_ioat_cdev = NULL; 562 } 563 return (0); 564 } 565 566 static int 567 sysctl_enable_ioat_test(SYSCTL_HANDLER_ARGS) 568 { 569 int error, enabled; 570 571 enabled = (g_ioat_cdev != NULL); 572 error = sysctl_handle_int(oidp, &enabled, 0, req); 573 if (error != 0 || req->newptr == NULL) 574 return (error); 575 576 enable_ioat_test(enabled); 577 return (0); 578 } 579 SYSCTL_PROC(_hw_ioat, OID_AUTO, enable_ioat_test, 580 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0, 581 sysctl_enable_ioat_test, "I", 582 "Non-zero: Enable the /dev/ioat_test device"); 583 584 void 585 ioat_test_attach(void) 586 { 587 char *val; 588 589 val = kern_getenv("hw.ioat.enable_ioat_test"); 590 if (val != NULL && strcmp(val, "0") != 0) { 591 mtx_lock(&Giant); 592 enable_ioat_test(true); 593 mtx_unlock(&Giant); 594 } 595 freeenv(val); 596 } 597 598 void 599 ioat_test_detach(void) 600 { 601 602 mtx_lock(&Giant); 603 enable_ioat_test(false); 604 mtx_unlock(&Giant); 605 } 606 607 static void 608 _ioat_test_log(int verbosity, const char *fmt, ...) 609 { 610 va_list argp; 611 612 if (verbosity > g_ioat_debug_level) 613 return; 614 615 va_start(argp, fmt); 616 vprintf(fmt, argp); 617 va_end(argp); 618 } 619