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 <vm/vm.h> 47 #include <vm/pmap.h> 48 49 #include "ioat.h" 50 #include "ioat_hw.h" 51 #include "ioat_internal.h" 52 #include "ioat_test.h" 53 54 #ifndef time_after 55 #define time_after(a,b) ((long)(b) - (long)(a) < 0) 56 #endif 57 58 MALLOC_DEFINE(M_IOAT_TEST, "ioat_test", "ioat test allocations"); 59 60 #define IOAT_MAX_BUFS 256 61 62 struct test_transaction { 63 void *buf[IOAT_MAX_BUFS]; 64 uint32_t length; 65 uint32_t depth; 66 struct ioat_test *test; 67 TAILQ_ENTRY(test_transaction) entry; 68 }; 69 70 #define IT_LOCK() mtx_lock(&ioat_test_lk) 71 #define IT_UNLOCK() mtx_unlock(&ioat_test_lk) 72 #define IT_ASSERT() mtx_assert(&ioat_test_lk, MA_OWNED) 73 static struct mtx ioat_test_lk; 74 MTX_SYSINIT(ioat_test_lk, &ioat_test_lk, "test coordination mtx", MTX_DEF); 75 76 static int g_thread_index = 1; 77 static struct cdev *g_ioat_cdev = NULL; 78 79 static void 80 ioat_test_transaction_destroy(struct test_transaction *tx) 81 { 82 int i; 83 84 for (i = 0; i < IOAT_MAX_BUFS; i++) { 85 if (tx->buf[i] != NULL) { 86 contigfree(tx->buf[i], tx->length, M_IOAT_TEST); 87 tx->buf[i] = NULL; 88 } 89 } 90 91 free(tx, M_IOAT_TEST); 92 } 93 94 static struct 95 test_transaction *ioat_test_transaction_create(unsigned num_buffers, 96 uint32_t buffer_size) 97 { 98 struct test_transaction *tx; 99 unsigned i; 100 101 tx = malloc(sizeof(*tx), M_IOAT_TEST, M_NOWAIT | M_ZERO); 102 if (tx == NULL) 103 return (NULL); 104 105 tx->length = buffer_size; 106 107 for (i = 0; i < num_buffers; i++) { 108 tx->buf[i] = contigmalloc(buffer_size, M_IOAT_TEST, M_NOWAIT, 109 0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); 110 111 if (tx->buf[i] == NULL) { 112 ioat_test_transaction_destroy(tx); 113 return (NULL); 114 } 115 } 116 return (tx); 117 } 118 119 static bool 120 ioat_compare_ok(struct test_transaction *tx) 121 { 122 uint32_t i; 123 124 for (i = 0; i < tx->depth; i++) { 125 if (memcmp(tx->buf[2*i], tx->buf[2*i+1], tx->length) != 0) 126 return (false); 127 } 128 return (true); 129 } 130 131 static void 132 ioat_dma_test_callback(void *arg) 133 { 134 struct test_transaction *tx; 135 struct ioat_test *test; 136 137 tx = arg; 138 test = tx->test; 139 140 if (test->verify && !ioat_compare_ok(tx)) { 141 ioat_log_message(0, "miscompare found\n"); 142 atomic_add_32(&test->status[IOAT_TEST_MISCOMPARE], tx->depth); 143 } else if (!test->too_late) 144 atomic_add_32(&test->status[IOAT_TEST_OK], tx->depth); 145 146 IT_LOCK(); 147 TAILQ_REMOVE(&test->pend_q, tx, entry); 148 TAILQ_INSERT_TAIL(&test->free_q, tx, entry); 149 wakeup(&test->free_q); 150 IT_UNLOCK(); 151 } 152 153 static int 154 ioat_test_prealloc_memory(struct ioat_test *test, int index) 155 { 156 uint32_t i, j, k; 157 struct test_transaction *tx; 158 159 for (i = 0; i < test->transactions; i++) { 160 tx = ioat_test_transaction_create(test->chain_depth * 2, 161 test->buffer_size); 162 if (tx == NULL) { 163 ioat_log_message(0, "tx == NULL - memory exhausted\n"); 164 test->status[IOAT_TEST_NO_MEMORY]++; 165 return (ENOMEM); 166 } 167 168 TAILQ_INSERT_HEAD(&test->free_q, tx, entry); 169 170 tx->test = test; 171 tx->depth = test->chain_depth; 172 173 /* fill in source buffers */ 174 for (j = 0; j < (tx->length / sizeof(uint32_t)); j++) { 175 uint32_t val = j + (index << 28); 176 177 for (k = 0; k < test->chain_depth; k++) { 178 ((uint32_t *)tx->buf[2*k])[j] = ~val; 179 ((uint32_t *)tx->buf[2*k+1])[j] = val; 180 } 181 } 182 } 183 return (0); 184 } 185 186 static void 187 ioat_test_release_memory(struct ioat_test *test) 188 { 189 struct test_transaction *tx, *s; 190 191 TAILQ_FOREACH_SAFE(tx, &test->free_q, entry, s) 192 ioat_test_transaction_destroy(tx); 193 TAILQ_INIT(&test->free_q); 194 195 TAILQ_FOREACH_SAFE(tx, &test->pend_q, entry, s) 196 ioat_test_transaction_destroy(tx); 197 TAILQ_INIT(&test->pend_q); 198 } 199 200 static void 201 ioat_test_submit_1_tx(struct ioat_test *test, bus_dmaengine_t dma) 202 { 203 struct test_transaction *tx; 204 struct bus_dmadesc *desc; 205 bus_dmaengine_callback_t cb; 206 bus_addr_t src, dest; 207 uint32_t i, flags; 208 209 IT_LOCK(); 210 while (TAILQ_EMPTY(&test->free_q)) 211 msleep(&test->free_q, &ioat_test_lk, 0, "test_submit", 0); 212 213 tx = TAILQ_FIRST(&test->free_q); 214 TAILQ_REMOVE(&test->free_q, tx, entry); 215 TAILQ_INSERT_HEAD(&test->pend_q, tx, entry); 216 IT_UNLOCK(); 217 218 ioat_acquire(dma); 219 for (i = 0; i < tx->depth; i++) { 220 src = vtophys((vm_offset_t)tx->buf[2*i]); 221 dest = vtophys((vm_offset_t)tx->buf[2*i+1]); 222 223 if (i == tx->depth - 1) { 224 cb = ioat_dma_test_callback; 225 flags = DMA_INT_EN; 226 } else { 227 cb = NULL; 228 flags = 0; 229 } 230 231 desc = ioat_copy(dma, src, dest, tx->length, cb, tx, flags); 232 if (desc == NULL) 233 panic("Failed to allocate a ring slot " 234 "-- this shouldn't happen!"); 235 } 236 ioat_release(dma); 237 } 238 239 static void 240 ioat_dma_test(void *arg) 241 { 242 struct ioat_test *test; 243 bus_dmaengine_t dmaengine; 244 uint32_t loops; 245 int index, rc, start, end; 246 247 test = arg; 248 memset(__DEVOLATILE(void *, test->status), 0, sizeof(test->status)); 249 250 if (test->buffer_size > 1024 * 1024) { 251 ioat_log_message(0, "Buffer size too large >1MB\n"); 252 test->status[IOAT_TEST_NO_MEMORY]++; 253 return; 254 } 255 256 if (test->chain_depth * 2 > IOAT_MAX_BUFS) { 257 ioat_log_message(0, "Depth too large (> %u)\n", 258 (unsigned)IOAT_MAX_BUFS / 2); 259 test->status[IOAT_TEST_NO_MEMORY]++; 260 return; 261 } 262 263 if (btoc((uint64_t)test->buffer_size * test->chain_depth * 264 test->transactions) > (physmem / 4)) { 265 ioat_log_message(0, "Sanity check failed -- test would " 266 "use more than 1/4 of phys mem.\n"); 267 test->status[IOAT_TEST_NO_MEMORY]++; 268 return; 269 } 270 271 if ((uint64_t)test->transactions * test->chain_depth > (1<<16)) { 272 ioat_log_message(0, "Sanity check failed -- test would " 273 "use more than available IOAT ring space.\n"); 274 test->status[IOAT_TEST_NO_MEMORY]++; 275 return; 276 } 277 278 dmaengine = ioat_get_dmaengine(test->channel_index); 279 if (dmaengine == NULL) { 280 ioat_log_message(0, "Couldn't acquire dmaengine\n"); 281 test->status[IOAT_TEST_NO_DMA_ENGINE]++; 282 return; 283 } 284 285 index = g_thread_index++; 286 TAILQ_INIT(&test->free_q); 287 TAILQ_INIT(&test->pend_q); 288 289 if (test->duration == 0) 290 ioat_log_message(1, "Thread %d: num_loops remaining: 0x%08x\n", 291 index, test->transactions); 292 else 293 ioat_log_message(1, "Thread %d: starting\n", index); 294 295 rc = ioat_test_prealloc_memory(test, index); 296 if (rc != 0) { 297 ioat_log_message(0, "prealloc_memory: %d\n", rc); 298 return; 299 } 300 wmb(); 301 302 test->too_late = false; 303 start = ticks; 304 end = start + (((sbintime_t)test->duration * hz) / 1000); 305 306 for (loops = 0;; loops++) { 307 if (test->duration == 0 && loops >= test->transactions) 308 break; 309 else if (test->duration != 0 && time_after(ticks, end)) { 310 test->too_late = true; 311 break; 312 } 313 314 ioat_test_submit_1_tx(test, dmaengine); 315 } 316 317 ioat_log_message(1, "Test Elapsed: %d ticks (overrun %d), %d sec.\n", 318 ticks - start, ticks - end, (ticks - start) / hz); 319 320 IT_LOCK(); 321 while (!TAILQ_EMPTY(&test->pend_q)) 322 msleep(&test->free_q, &ioat_test_lk, 0, "ioattestcompl", hz); 323 IT_UNLOCK(); 324 325 ioat_log_message(1, "Test Elapsed2: %d ticks (overrun %d), %d sec.\n", 326 ticks - start, ticks - end, (ticks - start) / hz); 327 328 ioat_test_release_memory(test); 329 } 330 331 static int 332 ioat_test_open(struct cdev *dev, int flags, int fmt, struct thread *td) 333 { 334 335 return (0); 336 } 337 338 static int 339 ioat_test_close(struct cdev *dev, int flags, int fmt, struct thread *td) 340 { 341 342 return (0); 343 } 344 345 static int 346 ioat_test_ioctl(struct cdev *dev, unsigned long cmd, caddr_t arg, int flag, 347 struct thread *td) 348 { 349 350 switch (cmd) { 351 case IOAT_DMATEST: 352 ioat_dma_test(arg); 353 break; 354 default: 355 return (EINVAL); 356 } 357 return (0); 358 } 359 360 static struct cdevsw ioat_cdevsw = { 361 .d_version = D_VERSION, 362 .d_flags = 0, 363 .d_open = ioat_test_open, 364 .d_close = ioat_test_close, 365 .d_ioctl = ioat_test_ioctl, 366 .d_name = "ioat_test", 367 }; 368 369 static int 370 enable_ioat_test(bool enable) 371 { 372 373 mtx_assert(&Giant, MA_OWNED); 374 375 if (enable && g_ioat_cdev == NULL) { 376 g_ioat_cdev = make_dev(&ioat_cdevsw, 0, UID_ROOT, GID_WHEEL, 377 0600, "ioat_test"); 378 } else if (!enable && g_ioat_cdev != NULL) { 379 destroy_dev(g_ioat_cdev); 380 g_ioat_cdev = NULL; 381 } 382 return (0); 383 } 384 385 static int 386 sysctl_enable_ioat_test(SYSCTL_HANDLER_ARGS) 387 { 388 int error, enabled; 389 390 enabled = (g_ioat_cdev != NULL); 391 error = sysctl_handle_int(oidp, &enabled, 0, req); 392 if (error != 0 || req->newptr == NULL) 393 return (error); 394 395 enable_ioat_test(enabled); 396 return (0); 397 } 398 SYSCTL_PROC(_hw_ioat, OID_AUTO, enable_ioat_test, CTLTYPE_INT | CTLFLAG_RW, 399 0, 0, sysctl_enable_ioat_test, "I", 400 "Non-zero: Enable the /dev/ioat_test device"); 401 402 void 403 ioat_test_attach(void) 404 { 405 char *val; 406 407 val = kern_getenv("hw.ioat.enable_ioat_test"); 408 if (val != NULL && strcmp(val, "0") != 0) { 409 mtx_lock(&Giant); 410 enable_ioat_test(true); 411 mtx_unlock(&Giant); 412 } 413 freeenv(val); 414 } 415 416 void 417 ioat_test_detach(void) 418 { 419 420 mtx_lock(&Giant); 421 enable_ioat_test(false); 422 mtx_unlock(&Giant); 423 } 424