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 struct ioat_test *test; 69 TAILQ_ENTRY(test_transaction) entry; 70 }; 71 72 #define IT_LOCK() mtx_lock(&ioat_test_lk) 73 #define IT_UNLOCK() mtx_unlock(&ioat_test_lk) 74 #define IT_ASSERT() mtx_assert(&ioat_test_lk, MA_OWNED) 75 static struct mtx ioat_test_lk; 76 MTX_SYSINIT(ioat_test_lk, &ioat_test_lk, "test coordination mtx", MTX_DEF); 77 78 static int g_thread_index = 1; 79 static struct cdev *g_ioat_cdev = NULL; 80 81 #define ioat_test_log(v, ...) _ioat_test_log((v), "ioat_test: " __VA_ARGS__) 82 static inline void _ioat_test_log(int verbosity, const char *fmt, ...); 83 84 static void 85 ioat_test_transaction_destroy(struct test_transaction *tx) 86 { 87 struct ioat_test *test; 88 int i; 89 90 test = tx->test; 91 92 for (i = 0; i < IOAT_MAX_BUFS; i++) { 93 if (tx->buf[i] != NULL) { 94 if (test->testkind == IOAT_TEST_DMA_8K) 95 free(tx->buf[i], M_IOAT_TEST); 96 else 97 contigfree(tx->buf[i], tx->length, M_IOAT_TEST); 98 tx->buf[i] = NULL; 99 } 100 } 101 102 free(tx, M_IOAT_TEST); 103 } 104 105 static struct 106 test_transaction *ioat_test_transaction_create(struct ioat_test *test, 107 unsigned num_buffers) 108 { 109 struct test_transaction *tx; 110 unsigned i; 111 112 tx = malloc(sizeof(*tx), M_IOAT_TEST, M_NOWAIT | M_ZERO); 113 if (tx == NULL) 114 return (NULL); 115 116 tx->length = test->buffer_size; 117 118 for (i = 0; i < num_buffers; i++) { 119 if (test->testkind == IOAT_TEST_DMA_8K) 120 tx->buf[i] = malloc(test->buffer_size, M_IOAT_TEST, 121 M_NOWAIT); 122 else 123 tx->buf[i] = contigmalloc(test->buffer_size, 124 M_IOAT_TEST, M_NOWAIT, 0, BUS_SPACE_MAXADDR, 125 PAGE_SIZE, 0); 126 127 if (tx->buf[i] == NULL) { 128 ioat_test_transaction_destroy(tx); 129 return (NULL); 130 } 131 } 132 return (tx); 133 } 134 135 static void 136 dump_hex(void *p, size_t chunks) 137 { 138 size_t i, j; 139 140 for (i = 0; i < chunks; i++) { 141 for (j = 0; j < 8; j++) 142 printf("%08x ", ((uint32_t *)p)[i * 8 + j]); 143 printf("\n"); 144 } 145 } 146 147 static bool 148 ioat_compare_ok(struct test_transaction *tx) 149 { 150 struct ioat_test *test; 151 char *dst, *src; 152 uint32_t i, j; 153 154 test = tx->test; 155 156 for (i = 0; i < tx->depth; i++) { 157 dst = tx->buf[2 * i + 1]; 158 src = tx->buf[2 * i]; 159 160 if (test->testkind == IOAT_TEST_FILL) { 161 for (j = 0; j < tx->length; j += sizeof(uint64_t)) { 162 if (memcmp(src, &dst[j], 163 MIN(sizeof(uint64_t), tx->length - j)) 164 != 0) 165 return (false); 166 } 167 } else if (test->testkind == IOAT_TEST_DMA) { 168 if (memcmp(src, dst, tx->length) != 0) 169 return (false); 170 } else if (test->testkind == IOAT_TEST_RAW_DMA) { 171 if (test->raw_write) 172 dst = test->raw_vtarget; 173 dump_hex(dst, tx->length / 32); 174 } 175 } 176 return (true); 177 } 178 179 static void 180 ioat_dma_test_callback(void *arg, int error) 181 { 182 struct test_transaction *tx; 183 struct ioat_test *test; 184 185 if (error != 0) 186 ioat_test_log(0, "%s: Got error: %d\n", __func__, error); 187 188 tx = arg; 189 test = tx->test; 190 191 if (test->verify && !ioat_compare_ok(tx)) { 192 ioat_test_log(0, "miscompare found\n"); 193 atomic_add_32(&test->status[IOAT_TEST_MISCOMPARE], tx->depth); 194 } else if (!test->too_late) 195 atomic_add_32(&test->status[IOAT_TEST_OK], tx->depth); 196 197 IT_LOCK(); 198 TAILQ_REMOVE(&test->pend_q, tx, entry); 199 TAILQ_INSERT_TAIL(&test->free_q, tx, entry); 200 wakeup(&test->free_q); 201 IT_UNLOCK(); 202 } 203 204 static int 205 ioat_test_prealloc_memory(struct ioat_test *test, int index) 206 { 207 uint32_t i, j, k; 208 struct test_transaction *tx; 209 210 for (i = 0; i < test->transactions; i++) { 211 tx = ioat_test_transaction_create(test, test->chain_depth * 2); 212 if (tx == NULL) { 213 ioat_test_log(0, "tx == NULL - memory exhausted\n"); 214 test->status[IOAT_TEST_NO_MEMORY]++; 215 return (ENOMEM); 216 } 217 218 TAILQ_INSERT_HEAD(&test->free_q, tx, entry); 219 220 tx->test = test; 221 tx->depth = test->chain_depth; 222 223 /* fill in source buffers */ 224 for (j = 0; j < (tx->length / sizeof(uint32_t)); j++) { 225 uint32_t val = j + (index << 28); 226 227 for (k = 0; k < test->chain_depth; k++) { 228 ((uint32_t *)tx->buf[2*k])[j] = ~val; 229 ((uint32_t *)tx->buf[2*k+1])[j] = val; 230 } 231 } 232 } 233 return (0); 234 } 235 236 static void 237 ioat_test_release_memory(struct ioat_test *test) 238 { 239 struct test_transaction *tx, *s; 240 241 TAILQ_FOREACH_SAFE(tx, &test->free_q, entry, s) 242 ioat_test_transaction_destroy(tx); 243 TAILQ_INIT(&test->free_q); 244 245 TAILQ_FOREACH_SAFE(tx, &test->pend_q, entry, s) 246 ioat_test_transaction_destroy(tx); 247 TAILQ_INIT(&test->pend_q); 248 } 249 250 static void 251 ioat_test_submit_1_tx(struct ioat_test *test, bus_dmaengine_t dma) 252 { 253 struct test_transaction *tx; 254 struct bus_dmadesc *desc; 255 bus_dmaengine_callback_t cb; 256 bus_addr_t src, dest; 257 uint64_t fillpattern; 258 uint32_t i, flags; 259 260 desc = NULL; 261 262 IT_LOCK(); 263 while (TAILQ_EMPTY(&test->free_q)) 264 msleep(&test->free_q, &ioat_test_lk, 0, "test_submit", 0); 265 266 tx = TAILQ_FIRST(&test->free_q); 267 TAILQ_REMOVE(&test->free_q, tx, entry); 268 TAILQ_INSERT_HEAD(&test->pend_q, tx, entry); 269 IT_UNLOCK(); 270 271 if (test->testkind != IOAT_TEST_MEMCPY) 272 ioat_acquire(dma); 273 for (i = 0; i < tx->depth; i++) { 274 if (test->testkind == IOAT_TEST_MEMCPY) { 275 memcpy(tx->buf[2 * i + 1], tx->buf[2 * i], tx->length); 276 if (i == tx->depth - 1) 277 ioat_dma_test_callback(tx, 0); 278 continue; 279 } 280 281 src = vtophys((vm_offset_t)tx->buf[2*i]); 282 dest = vtophys((vm_offset_t)tx->buf[2*i+1]); 283 284 if (test->testkind == IOAT_TEST_RAW_DMA) { 285 if (test->raw_write) 286 dest = test->raw_target; 287 else 288 src = test->raw_target; 289 } 290 291 if (i == tx->depth - 1) { 292 cb = ioat_dma_test_callback; 293 flags = DMA_INT_EN; 294 } else { 295 cb = NULL; 296 flags = 0; 297 } 298 299 if (test->testkind == IOAT_TEST_DMA || 300 test->testkind == IOAT_TEST_RAW_DMA) 301 desc = ioat_copy(dma, dest, src, tx->length, cb, tx, 302 flags); 303 else if (test->testkind == IOAT_TEST_FILL) { 304 fillpattern = *(uint64_t *)tx->buf[2*i]; 305 desc = ioat_blockfill(dma, dest, fillpattern, 306 tx->length, cb, tx, flags); 307 } else if (test->testkind == IOAT_TEST_DMA_8K) { 308 bus_addr_t src2, dst2; 309 310 src2 = vtophys((vm_offset_t)tx->buf[2*i] + PAGE_SIZE); 311 dst2 = vtophys((vm_offset_t)tx->buf[2*i+1] + PAGE_SIZE); 312 313 desc = ioat_copy_8k_aligned(dma, dest, dst2, src, src2, 314 cb, tx, flags); 315 } 316 if (desc == NULL) 317 break; 318 } 319 if (test->testkind == IOAT_TEST_MEMCPY) 320 return; 321 ioat_release(dma); 322 323 /* 324 * We couldn't issue an IO -- either the device is being detached or 325 * the HW reset. Essentially spin until the device comes back up or 326 * our timer expires. 327 */ 328 if (desc == NULL && tx->depth > 0) { 329 atomic_add_32(&test->status[IOAT_TEST_NO_DMA_ENGINE], tx->depth); 330 IT_LOCK(); 331 TAILQ_REMOVE(&test->pend_q, tx, entry); 332 TAILQ_INSERT_HEAD(&test->free_q, tx, entry); 333 IT_UNLOCK(); 334 } 335 } 336 337 static void 338 ioat_dma_test(void *arg) 339 { 340 struct ioat_softc *ioat; 341 struct ioat_test *test; 342 bus_dmaengine_t dmaengine; 343 uint32_t loops; 344 int index, rc, start, end, error; 345 346 test = arg; 347 memset(__DEVOLATILE(void *, test->status), 0, sizeof(test->status)); 348 349 if (test->testkind == IOAT_TEST_DMA_8K && 350 test->buffer_size != 2 * PAGE_SIZE) { 351 ioat_test_log(0, "Asked for 8k test and buffer size isn't 8k\n"); 352 test->status[IOAT_TEST_INVALID_INPUT]++; 353 return; 354 } 355 356 if (test->buffer_size > 1024 * 1024) { 357 ioat_test_log(0, "Buffer size too large >1MB\n"); 358 test->status[IOAT_TEST_NO_MEMORY]++; 359 return; 360 } 361 362 if (test->chain_depth * 2 > IOAT_MAX_BUFS) { 363 ioat_test_log(0, "Depth too large (> %u)\n", 364 (unsigned)IOAT_MAX_BUFS / 2); 365 test->status[IOAT_TEST_NO_MEMORY]++; 366 return; 367 } 368 369 if (btoc((uint64_t)test->buffer_size * test->chain_depth * 370 test->transactions) > (physmem / 4)) { 371 ioat_test_log(0, "Sanity check failed -- test would " 372 "use more than 1/4 of phys mem.\n"); 373 test->status[IOAT_TEST_NO_MEMORY]++; 374 return; 375 } 376 377 if ((uint64_t)test->transactions * test->chain_depth > (1<<16)) { 378 ioat_test_log(0, "Sanity check failed -- test would " 379 "use more than available IOAT ring space.\n"); 380 test->status[IOAT_TEST_NO_MEMORY]++; 381 return; 382 } 383 384 if (test->testkind >= IOAT_NUM_TESTKINDS) { 385 ioat_test_log(0, "Invalid kind %u\n", 386 (unsigned)test->testkind); 387 test->status[IOAT_TEST_INVALID_INPUT]++; 388 return; 389 } 390 391 dmaengine = ioat_get_dmaengine(test->channel_index, M_NOWAIT); 392 if (dmaengine == NULL) { 393 ioat_test_log(0, "Couldn't acquire dmaengine\n"); 394 test->status[IOAT_TEST_NO_DMA_ENGINE]++; 395 return; 396 } 397 ioat = to_ioat_softc(dmaengine); 398 399 if (test->testkind == IOAT_TEST_FILL && 400 (ioat->capabilities & IOAT_DMACAP_BFILL) == 0) 401 { 402 ioat_test_log(0, 403 "Hardware doesn't support block fill, aborting test\n"); 404 test->status[IOAT_TEST_INVALID_INPUT]++; 405 goto out; 406 } 407 408 if (test->coalesce_period > ioat->intrdelay_max) { 409 ioat_test_log(0, 410 "Hardware doesn't support intrdelay of %u us.\n", 411 (unsigned)test->coalesce_period); 412 test->status[IOAT_TEST_INVALID_INPUT]++; 413 goto out; 414 } 415 error = ioat_set_interrupt_coalesce(dmaengine, test->coalesce_period); 416 if (error == ENODEV && test->coalesce_period == 0) 417 error = 0; 418 if (error != 0) { 419 ioat_test_log(0, "ioat_set_interrupt_coalesce: %d\n", error); 420 test->status[IOAT_TEST_INVALID_INPUT]++; 421 goto out; 422 } 423 424 if (test->zero_stats) 425 memset(&ioat->stats, 0, sizeof(ioat->stats)); 426 427 if (test->testkind == IOAT_TEST_RAW_DMA) { 428 if (test->raw_is_virtual) { 429 test->raw_vtarget = (void *)test->raw_target; 430 test->raw_target = vtophys(test->raw_vtarget); 431 } else { 432 test->raw_vtarget = pmap_mapdev(test->raw_target, 433 test->buffer_size); 434 } 435 } 436 437 index = g_thread_index++; 438 TAILQ_INIT(&test->free_q); 439 TAILQ_INIT(&test->pend_q); 440 441 if (test->duration == 0) 442 ioat_test_log(1, "Thread %d: num_loops remaining: 0x%08x\n", 443 index, test->transactions); 444 else 445 ioat_test_log(1, "Thread %d: starting\n", index); 446 447 rc = ioat_test_prealloc_memory(test, index); 448 if (rc != 0) { 449 ioat_test_log(0, "prealloc_memory: %d\n", rc); 450 goto out; 451 } 452 wmb(); 453 454 test->too_late = false; 455 start = ticks; 456 end = start + (((sbintime_t)test->duration * hz) / 1000); 457 458 for (loops = 0;; loops++) { 459 if (test->duration == 0 && loops >= test->transactions) 460 break; 461 else if (test->duration != 0 && time_after(ticks, end)) { 462 test->too_late = true; 463 break; 464 } 465 466 ioat_test_submit_1_tx(test, dmaengine); 467 } 468 469 ioat_test_log(1, "Test Elapsed: %d ticks (overrun %d), %d sec.\n", 470 ticks - start, ticks - end, (ticks - start) / hz); 471 472 IT_LOCK(); 473 while (!TAILQ_EMPTY(&test->pend_q)) 474 msleep(&test->free_q, &ioat_test_lk, 0, "ioattestcompl", hz); 475 IT_UNLOCK(); 476 477 ioat_test_log(1, "Test Elapsed2: %d ticks (overrun %d), %d sec.\n", 478 ticks - start, ticks - end, (ticks - start) / hz); 479 480 ioat_test_release_memory(test); 481 out: 482 if (test->testkind == IOAT_TEST_RAW_DMA && !test->raw_is_virtual) 483 pmap_unmapdev((vm_offset_t)test->raw_vtarget, 484 test->buffer_size); 485 ioat_put_dmaengine(dmaengine); 486 } 487 488 static int 489 ioat_test_open(struct cdev *dev, int flags, int fmt, struct thread *td) 490 { 491 492 return (0); 493 } 494 495 static int 496 ioat_test_close(struct cdev *dev, int flags, int fmt, struct thread *td) 497 { 498 499 return (0); 500 } 501 502 static int 503 ioat_test_ioctl(struct cdev *dev, unsigned long cmd, caddr_t arg, int flag, 504 struct thread *td) 505 { 506 507 switch (cmd) { 508 case IOAT_DMATEST: 509 ioat_dma_test(arg); 510 break; 511 default: 512 return (EINVAL); 513 } 514 return (0); 515 } 516 517 static struct cdevsw ioat_cdevsw = { 518 .d_version = D_VERSION, 519 .d_flags = 0, 520 .d_open = ioat_test_open, 521 .d_close = ioat_test_close, 522 .d_ioctl = ioat_test_ioctl, 523 .d_name = "ioat_test", 524 }; 525 526 static int 527 enable_ioat_test(bool enable) 528 { 529 530 mtx_assert(&Giant, MA_OWNED); 531 532 if (enable && g_ioat_cdev == NULL) { 533 g_ioat_cdev = make_dev(&ioat_cdevsw, 0, UID_ROOT, GID_WHEEL, 534 0600, "ioat_test"); 535 } else if (!enable && g_ioat_cdev != NULL) { 536 destroy_dev(g_ioat_cdev); 537 g_ioat_cdev = NULL; 538 } 539 return (0); 540 } 541 542 static int 543 sysctl_enable_ioat_test(SYSCTL_HANDLER_ARGS) 544 { 545 int error, enabled; 546 547 enabled = (g_ioat_cdev != NULL); 548 error = sysctl_handle_int(oidp, &enabled, 0, req); 549 if (error != 0 || req->newptr == NULL) 550 return (error); 551 552 enable_ioat_test(enabled); 553 return (0); 554 } 555 SYSCTL_PROC(_hw_ioat, OID_AUTO, enable_ioat_test, CTLTYPE_INT | CTLFLAG_RW, 556 0, 0, sysctl_enable_ioat_test, "I", 557 "Non-zero: Enable the /dev/ioat_test device"); 558 559 void 560 ioat_test_attach(void) 561 { 562 char *val; 563 564 val = kern_getenv("hw.ioat.enable_ioat_test"); 565 if (val != NULL && strcmp(val, "0") != 0) { 566 mtx_lock(&Giant); 567 enable_ioat_test(true); 568 mtx_unlock(&Giant); 569 } 570 freeenv(val); 571 } 572 573 void 574 ioat_test_detach(void) 575 { 576 577 mtx_lock(&Giant); 578 enable_ioat_test(false); 579 mtx_unlock(&Giant); 580 } 581 582 static inline void 583 _ioat_test_log(int verbosity, const char *fmt, ...) 584 { 585 va_list argp; 586 587 if (verbosity > g_ioat_debug_level) 588 return; 589 590 va_start(argp, fmt); 591 vprintf(fmt, argp); 592 va_end(argp); 593 } 594