1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2009 Yahoo! Inc. 5 * Copyright (c) 2011-2015 LSI Corp. 6 * Copyright (c) 2013-2015 Avago Technologies 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD 31 * 32 * $FreeBSD$ 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 /* Communications core for Avago Technologies (LSI) MPT2 */ 39 40 /* TODO Move headers to mpsvar */ 41 #include <sys/types.h> 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/kernel.h> 45 #include <sys/selinfo.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/module.h> 49 #include <sys/bus.h> 50 #include <sys/conf.h> 51 #include <sys/bio.h> 52 #include <sys/malloc.h> 53 #include <sys/uio.h> 54 #include <sys/sysctl.h> 55 #include <sys/smp.h> 56 #include <sys/queue.h> 57 #include <sys/kthread.h> 58 #include <sys/taskqueue.h> 59 #include <sys/endian.h> 60 #include <sys/eventhandler.h> 61 #include <sys/sbuf.h> 62 #include <sys/priv.h> 63 64 #include <machine/bus.h> 65 #include <machine/resource.h> 66 #include <sys/rman.h> 67 #include <sys/proc.h> 68 69 #include <dev/pci/pcivar.h> 70 71 #include <cam/cam.h> 72 #include <cam/scsi/scsi_all.h> 73 74 #include <dev/mps/mpi/mpi2_type.h> 75 #include <dev/mps/mpi/mpi2.h> 76 #include <dev/mps/mpi/mpi2_ioc.h> 77 #include <dev/mps/mpi/mpi2_sas.h> 78 #include <dev/mps/mpi/mpi2_cnfg.h> 79 #include <dev/mps/mpi/mpi2_init.h> 80 #include <dev/mps/mpi/mpi2_tool.h> 81 #include <dev/mps/mps_ioctl.h> 82 #include <dev/mps/mpsvar.h> 83 #include <dev/mps/mps_table.h> 84 85 static int mps_diag_reset(struct mps_softc *sc, int sleep_flag); 86 static int mps_init_queues(struct mps_softc *sc); 87 static void mps_resize_queues(struct mps_softc *sc); 88 static int mps_message_unit_reset(struct mps_softc *sc, int sleep_flag); 89 static int mps_transition_operational(struct mps_softc *sc); 90 static int mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching); 91 static void mps_iocfacts_free(struct mps_softc *sc); 92 static void mps_startup(void *arg); 93 static int mps_send_iocinit(struct mps_softc *sc); 94 static int mps_alloc_queues(struct mps_softc *sc); 95 static int mps_alloc_hw_queues(struct mps_softc *sc); 96 static int mps_alloc_replies(struct mps_softc *sc); 97 static int mps_alloc_requests(struct mps_softc *sc); 98 static int mps_attach_log(struct mps_softc *sc); 99 static __inline void mps_complete_command(struct mps_softc *sc, 100 struct mps_command *cm); 101 static void mps_dispatch_event(struct mps_softc *sc, uintptr_t data, 102 MPI2_EVENT_NOTIFICATION_REPLY *reply); 103 static void mps_config_complete(struct mps_softc *sc, struct mps_command *cm); 104 static void mps_periodic(void *); 105 static int mps_reregister_events(struct mps_softc *sc); 106 static void mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm); 107 static int mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts); 108 static int mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag); 109 static int mps_debug_sysctl(SYSCTL_HANDLER_ARGS); 110 static int mps_dump_reqs(SYSCTL_HANDLER_ARGS); 111 static void mps_parse_debug(struct mps_softc *sc, char *list); 112 113 SYSCTL_NODE(_hw, OID_AUTO, mps, CTLFLAG_RD, 0, "MPS Driver Parameters"); 114 115 MALLOC_DEFINE(M_MPT2, "mps", "mpt2 driver memory"); 116 MALLOC_DECLARE(M_MPSUSER); 117 118 /* 119 * Do a "Diagnostic Reset" aka a hard reset. This should get the chip out of 120 * any state and back to its initialization state machine. 121 */ 122 static char mpt2_reset_magic[] = { 0x00, 0x0f, 0x04, 0x0b, 0x02, 0x07, 0x0d }; 123 124 /* Added this union to smoothly convert le64toh cm->cm_desc.Words. 125 * Compiler only support unint64_t to be passed as argument. 126 * Otherwise it will throw below error 127 * "aggregate value used where an integer was expected" 128 */ 129 130 typedef union _reply_descriptor { 131 u64 word; 132 struct { 133 u32 low; 134 u32 high; 135 } u; 136 }reply_descriptor,address_descriptor; 137 138 /* Rate limit chain-fail messages to 1 per minute */ 139 static struct timeval mps_chainfail_interval = { 60, 0 }; 140 141 /* 142 * sleep_flag can be either CAN_SLEEP or NO_SLEEP. 143 * If this function is called from process context, it can sleep 144 * and there is no harm to sleep, in case if this fuction is called 145 * from Interrupt handler, we can not sleep and need NO_SLEEP flag set. 146 * based on sleep flags driver will call either msleep, pause or DELAY. 147 * msleep and pause are of same variant, but pause is used when mps_mtx 148 * is not hold by driver. 149 * 150 */ 151 static int 152 mps_diag_reset(struct mps_softc *sc,int sleep_flag) 153 { 154 uint32_t reg; 155 int i, error, tries = 0; 156 uint8_t first_wait_done = FALSE; 157 158 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 159 160 /* Clear any pending interrupts */ 161 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 162 163 /* 164 * Force NO_SLEEP for threads prohibited to sleep 165 * e.a Thread from interrupt handler are prohibited to sleep. 166 */ 167 if (curthread->td_no_sleeping != 0) 168 sleep_flag = NO_SLEEP; 169 170 mps_dprint(sc, MPS_INIT, "sequence start, sleep_flag= %d\n", sleep_flag); 171 172 /* Push the magic sequence */ 173 error = ETIMEDOUT; 174 while (tries++ < 20) { 175 for (i = 0; i < sizeof(mpt2_reset_magic); i++) 176 mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 177 mpt2_reset_magic[i]); 178 /* wait 100 msec */ 179 if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) 180 msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, 181 "mpsdiag", hz/10); 182 else if (sleep_flag == CAN_SLEEP) 183 pause("mpsdiag", hz/10); 184 else 185 DELAY(100 * 1000); 186 187 reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); 188 if (reg & MPI2_DIAG_DIAG_WRITE_ENABLE) { 189 error = 0; 190 break; 191 } 192 } 193 if (error) { 194 mps_dprint(sc, MPS_INIT, "sequence failed, error=%d, exit\n", 195 error); 196 return (error); 197 } 198 199 /* Send the actual reset. XXX need to refresh the reg? */ 200 reg |= MPI2_DIAG_RESET_ADAPTER; 201 mps_dprint(sc, MPS_INIT, "sequence success, sending reset, reg= 0x%x\n", 202 reg); 203 mps_regwrite(sc, MPI2_HOST_DIAGNOSTIC_OFFSET, reg); 204 205 /* Wait up to 300 seconds in 50ms intervals */ 206 error = ETIMEDOUT; 207 for (i = 0; i < 6000; i++) { 208 /* 209 * Wait 50 msec. If this is the first time through, wait 256 210 * msec to satisfy Diag Reset timing requirements. 211 */ 212 if (first_wait_done) { 213 if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) 214 msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, 215 "mpsdiag", hz/20); 216 else if (sleep_flag == CAN_SLEEP) 217 pause("mpsdiag", hz/20); 218 else 219 DELAY(50 * 1000); 220 } else { 221 DELAY(256 * 1000); 222 first_wait_done = TRUE; 223 } 224 /* 225 * Check for the RESET_ADAPTER bit to be cleared first, then 226 * wait for the RESET state to be cleared, which takes a little 227 * longer. 228 */ 229 reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); 230 if (reg & MPI2_DIAG_RESET_ADAPTER) { 231 continue; 232 } 233 reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); 234 if ((reg & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_RESET) { 235 error = 0; 236 break; 237 } 238 } 239 if (error) { 240 mps_dprint(sc, MPS_INIT, "reset failed, error= %d, exit\n", 241 error); 242 return (error); 243 } 244 245 mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 0x0); 246 mps_dprint(sc, MPS_INIT, "diag reset success, exit\n"); 247 248 return (0); 249 } 250 251 static int 252 mps_message_unit_reset(struct mps_softc *sc, int sleep_flag) 253 { 254 int error; 255 256 MPS_FUNCTRACE(sc); 257 258 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 259 260 error = 0; 261 mps_regwrite(sc, MPI2_DOORBELL_OFFSET, 262 MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET << 263 MPI2_DOORBELL_FUNCTION_SHIFT); 264 265 if (mps_wait_db_ack(sc, 5, sleep_flag) != 0) { 266 mps_dprint(sc, MPS_INIT|MPS_FAULT, 267 "Doorbell handshake failed\n"); 268 error = ETIMEDOUT; 269 } 270 271 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 272 return (error); 273 } 274 275 static int 276 mps_transition_ready(struct mps_softc *sc) 277 { 278 uint32_t reg, state; 279 int error, tries = 0; 280 int sleep_flags; 281 282 MPS_FUNCTRACE(sc); 283 /* If we are in attach call, do not sleep */ 284 sleep_flags = (sc->mps_flags & MPS_FLAGS_ATTACH_DONE) 285 ? CAN_SLEEP:NO_SLEEP; 286 error = 0; 287 288 mps_dprint(sc, MPS_INIT, "%s entered, sleep_flags= %d\n", 289 __func__, sleep_flags); 290 291 while (tries++ < 1200) { 292 reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); 293 mps_dprint(sc, MPS_INIT, " Doorbell= 0x%x\n", reg); 294 295 /* 296 * Ensure the IOC is ready to talk. If it's not, try 297 * resetting it. 298 */ 299 if (reg & MPI2_DOORBELL_USED) { 300 mps_dprint(sc, MPS_INIT, " Not ready, sending diag " 301 "reset\n"); 302 mps_diag_reset(sc, sleep_flags); 303 DELAY(50000); 304 continue; 305 } 306 307 /* Is the adapter owned by another peer? */ 308 if ((reg & MPI2_DOORBELL_WHO_INIT_MASK) == 309 (MPI2_WHOINIT_PCI_PEER << MPI2_DOORBELL_WHO_INIT_SHIFT)) { 310 mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC is under the " 311 "control of another peer host, aborting " 312 "initialization.\n"); 313 error = ENXIO; 314 break; 315 } 316 317 state = reg & MPI2_IOC_STATE_MASK; 318 if (state == MPI2_IOC_STATE_READY) { 319 /* Ready to go! */ 320 error = 0; 321 break; 322 } else if (state == MPI2_IOC_STATE_FAULT) { 323 mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC in fault " 324 "state 0x%x, resetting\n", 325 state & MPI2_DOORBELL_FAULT_CODE_MASK); 326 mps_diag_reset(sc, sleep_flags); 327 } else if (state == MPI2_IOC_STATE_OPERATIONAL) { 328 /* Need to take ownership */ 329 mps_message_unit_reset(sc, sleep_flags); 330 } else if (state == MPI2_IOC_STATE_RESET) { 331 /* Wait a bit, IOC might be in transition */ 332 mps_dprint(sc, MPS_INIT|MPS_FAULT, 333 "IOC in unexpected reset state\n"); 334 } else { 335 mps_dprint(sc, MPS_INIT|MPS_FAULT, 336 "IOC in unknown state 0x%x\n", state); 337 error = EINVAL; 338 break; 339 } 340 341 /* Wait 50ms for things to settle down. */ 342 DELAY(50000); 343 } 344 345 if (error) 346 mps_dprint(sc, MPS_INIT|MPS_FAULT, 347 "Cannot transition IOC to ready\n"); 348 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 349 350 return (error); 351 } 352 353 static int 354 mps_transition_operational(struct mps_softc *sc) 355 { 356 uint32_t reg, state; 357 int error; 358 359 MPS_FUNCTRACE(sc); 360 361 error = 0; 362 reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); 363 mps_dprint(sc, MPS_INIT, "%s entered, Doorbell= 0x%x\n", __func__, reg); 364 365 state = reg & MPI2_IOC_STATE_MASK; 366 if (state != MPI2_IOC_STATE_READY) { 367 mps_dprint(sc, MPS_INIT, "IOC not ready\n"); 368 if ((error = mps_transition_ready(sc)) != 0) { 369 mps_dprint(sc, MPS_INIT|MPS_FAULT, 370 "failed to transition ready, exit\n"); 371 return (error); 372 } 373 } 374 375 error = mps_send_iocinit(sc); 376 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 377 378 return (error); 379 } 380 381 static void 382 mps_resize_queues(struct mps_softc *sc) 383 { 384 u_int reqcr, prireqcr, maxio, sges_per_frame; 385 386 /* 387 * Size the queues. Since the reply queues always need one free 388 * entry, we'll deduct one reply message here. The LSI documents 389 * suggest instead to add a count to the request queue, but I think 390 * that it's better to deduct from reply queue. 391 */ 392 prireqcr = MAX(1, sc->max_prireqframes); 393 prireqcr = MIN(prireqcr, sc->facts->HighPriorityCredit); 394 395 reqcr = MAX(2, sc->max_reqframes); 396 reqcr = MIN(reqcr, sc->facts->RequestCredit); 397 398 sc->num_reqs = prireqcr + reqcr; 399 sc->num_prireqs = prireqcr; 400 sc->num_replies = MIN(sc->max_replyframes + sc->max_evtframes, 401 sc->facts->MaxReplyDescriptorPostQueueDepth) - 1; 402 403 /* Store the request frame size in bytes rather than as 32bit words */ 404 sc->reqframesz = sc->facts->IOCRequestFrameSize * 4; 405 406 /* 407 * Max IO Size is Page Size * the following: 408 * ((SGEs per frame - 1 for chain element) * Max Chain Depth) 409 * + 1 for no chain needed in last frame 410 * 411 * If user suggests a Max IO size to use, use the smaller of the 412 * user's value and the calculated value as long as the user's 413 * value is larger than 0. The user's value is in pages. 414 */ 415 sges_per_frame = sc->reqframesz / sizeof(MPI2_SGE_SIMPLE64) - 1; 416 maxio = (sges_per_frame * sc->facts->MaxChainDepth + 1) * PAGE_SIZE; 417 418 /* 419 * If I/O size limitation requested, then use it and pass up to CAM. 420 * If not, use MAXPHYS as an optimization hint, but report HW limit. 421 */ 422 if (sc->max_io_pages > 0) { 423 maxio = min(maxio, sc->max_io_pages * PAGE_SIZE); 424 sc->maxio = maxio; 425 } else { 426 sc->maxio = maxio; 427 maxio = min(maxio, MAXPHYS); 428 } 429 430 sc->num_chains = (maxio / PAGE_SIZE + sges_per_frame - 2) / 431 sges_per_frame * reqcr; 432 if (sc->max_chains > 0 && sc->max_chains < sc->num_chains) 433 sc->num_chains = sc->max_chains; 434 435 /* 436 * Figure out the number of MSIx-based queues. If the firmware or 437 * user has done something crazy and not allowed enough credit for 438 * the queues to be useful then don't enable multi-queue. 439 */ 440 if (sc->facts->MaxMSIxVectors < 2) 441 sc->msi_msgs = 1; 442 443 if (sc->msi_msgs > 1) { 444 sc->msi_msgs = MIN(sc->msi_msgs, mp_ncpus); 445 sc->msi_msgs = MIN(sc->msi_msgs, sc->facts->MaxMSIxVectors); 446 if (sc->num_reqs / sc->msi_msgs < 2) 447 sc->msi_msgs = 1; 448 } 449 450 mps_dprint(sc, MPS_INIT, "Sized queues to q=%d reqs=%d replies=%d\n", 451 sc->msi_msgs, sc->num_reqs, sc->num_replies); 452 } 453 454 /* 455 * This is called during attach and when re-initializing due to a Diag Reset. 456 * IOC Facts is used to allocate many of the structures needed by the driver. 457 * If called from attach, de-allocation is not required because the driver has 458 * not allocated any structures yet, but if called from a Diag Reset, previously 459 * allocated structures based on IOC Facts will need to be freed and re- 460 * allocated bases on the latest IOC Facts. 461 */ 462 static int 463 mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching) 464 { 465 int error; 466 Mpi2IOCFactsReply_t saved_facts; 467 uint8_t saved_mode, reallocating; 468 469 mps_dprint(sc, MPS_INIT|MPS_TRACE, "%s entered\n", __func__); 470 471 /* Save old IOC Facts and then only reallocate if Facts have changed */ 472 if (!attaching) { 473 bcopy(sc->facts, &saved_facts, sizeof(MPI2_IOC_FACTS_REPLY)); 474 } 475 476 /* 477 * Get IOC Facts. In all cases throughout this function, panic if doing 478 * a re-initialization and only return the error if attaching so the OS 479 * can handle it. 480 */ 481 if ((error = mps_get_iocfacts(sc, sc->facts)) != 0) { 482 if (attaching) { 483 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to get " 484 "IOC Facts with error %d, exit\n", error); 485 return (error); 486 } else { 487 panic("%s failed to get IOC Facts with error %d\n", 488 __func__, error); 489 } 490 } 491 492 MPS_DPRINT_PAGE(sc, MPS_XINFO, iocfacts, sc->facts); 493 494 snprintf(sc->fw_version, sizeof(sc->fw_version), 495 "%02d.%02d.%02d.%02d", 496 sc->facts->FWVersion.Struct.Major, 497 sc->facts->FWVersion.Struct.Minor, 498 sc->facts->FWVersion.Struct.Unit, 499 sc->facts->FWVersion.Struct.Dev); 500 501 snprintf(sc->msg_version, sizeof(sc->msg_version), "%d.%d", 502 (sc->facts->MsgVersion & MPI2_IOCFACTS_MSGVERSION_MAJOR_MASK) >> 503 MPI2_IOCFACTS_MSGVERSION_MAJOR_SHIFT, 504 (sc->facts->MsgVersion & MPI2_IOCFACTS_MSGVERSION_MINOR_MASK) >> 505 MPI2_IOCFACTS_MSGVERSION_MINOR_SHIFT); 506 507 mps_dprint(sc, MPS_INFO, "Firmware: %s, Driver: %s\n", sc->fw_version, 508 MPS_DRIVER_VERSION); 509 mps_dprint(sc, MPS_INFO, "IOCCapabilities: %b\n", 510 sc->facts->IOCCapabilities, 511 "\20" "\3ScsiTaskFull" "\4DiagTrace" "\5SnapBuf" "\6ExtBuf" 512 "\7EEDP" "\10BiDirTarg" "\11Multicast" "\14TransRetry" "\15IR" 513 "\16EventReplay" "\17RaidAccel" "\20MSIXIndex" "\21HostDisc"); 514 515 /* 516 * If the chip doesn't support event replay then a hard reset will be 517 * required to trigger a full discovery. Do the reset here then 518 * retransition to Ready. A hard reset might have already been done, 519 * but it doesn't hurt to do it again. Only do this if attaching, not 520 * for a Diag Reset. 521 */ 522 if (attaching && ((sc->facts->IOCCapabilities & 523 MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY) == 0)) { 524 mps_dprint(sc, MPS_INIT, "No event replay, reseting\n"); 525 mps_diag_reset(sc, NO_SLEEP); 526 if ((error = mps_transition_ready(sc)) != 0) { 527 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to " 528 "transition to ready with error %d, exit\n", 529 error); 530 return (error); 531 } 532 } 533 534 /* 535 * Set flag if IR Firmware is loaded. If the RAID Capability has 536 * changed from the previous IOC Facts, log a warning, but only if 537 * checking this after a Diag Reset and not during attach. 538 */ 539 saved_mode = sc->ir_firmware; 540 if (sc->facts->IOCCapabilities & 541 MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID) 542 sc->ir_firmware = 1; 543 if (!attaching) { 544 if (sc->ir_firmware != saved_mode) { 545 mps_dprint(sc, MPS_INIT|MPS_FAULT, "new IR/IT mode " 546 "in IOC Facts does not match previous mode\n"); 547 } 548 } 549 550 /* Only deallocate and reallocate if relevant IOC Facts have changed */ 551 reallocating = FALSE; 552 sc->mps_flags &= ~MPS_FLAGS_REALLOCATED; 553 554 if ((!attaching) && 555 ((saved_facts.MsgVersion != sc->facts->MsgVersion) || 556 (saved_facts.HeaderVersion != sc->facts->HeaderVersion) || 557 (saved_facts.MaxChainDepth != sc->facts->MaxChainDepth) || 558 (saved_facts.RequestCredit != sc->facts->RequestCredit) || 559 (saved_facts.ProductID != sc->facts->ProductID) || 560 (saved_facts.IOCCapabilities != sc->facts->IOCCapabilities) || 561 (saved_facts.IOCRequestFrameSize != 562 sc->facts->IOCRequestFrameSize) || 563 (saved_facts.MaxTargets != sc->facts->MaxTargets) || 564 (saved_facts.MaxSasExpanders != sc->facts->MaxSasExpanders) || 565 (saved_facts.MaxEnclosures != sc->facts->MaxEnclosures) || 566 (saved_facts.HighPriorityCredit != sc->facts->HighPriorityCredit) || 567 (saved_facts.MaxReplyDescriptorPostQueueDepth != 568 sc->facts->MaxReplyDescriptorPostQueueDepth) || 569 (saved_facts.ReplyFrameSize != sc->facts->ReplyFrameSize) || 570 (saved_facts.MaxVolumes != sc->facts->MaxVolumes) || 571 (saved_facts.MaxPersistentEntries != 572 sc->facts->MaxPersistentEntries))) { 573 reallocating = TRUE; 574 575 /* Record that we reallocated everything */ 576 sc->mps_flags |= MPS_FLAGS_REALLOCATED; 577 } 578 579 /* 580 * Some things should be done if attaching or re-allocating after a Diag 581 * Reset, but are not needed after a Diag Reset if the FW has not 582 * changed. 583 */ 584 if (attaching || reallocating) { 585 /* 586 * Check if controller supports FW diag buffers and set flag to 587 * enable each type. 588 */ 589 if (sc->facts->IOCCapabilities & 590 MPI2_IOCFACTS_CAPABILITY_DIAG_TRACE_BUFFER) 591 sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_TRACE]. 592 enabled = TRUE; 593 if (sc->facts->IOCCapabilities & 594 MPI2_IOCFACTS_CAPABILITY_SNAPSHOT_BUFFER) 595 sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_SNAPSHOT]. 596 enabled = TRUE; 597 if (sc->facts->IOCCapabilities & 598 MPI2_IOCFACTS_CAPABILITY_EXTENDED_BUFFER) 599 sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_EXTENDED]. 600 enabled = TRUE; 601 602 /* 603 * Set flag if EEDP is supported and if TLR is supported. 604 */ 605 if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EEDP) 606 sc->eedp_enabled = TRUE; 607 if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_TLR) 608 sc->control_TLR = TRUE; 609 610 mps_resize_queues(sc); 611 612 /* 613 * Initialize all Tail Queues 614 */ 615 TAILQ_INIT(&sc->req_list); 616 TAILQ_INIT(&sc->high_priority_req_list); 617 TAILQ_INIT(&sc->chain_list); 618 TAILQ_INIT(&sc->tm_list); 619 } 620 621 /* 622 * If doing a Diag Reset and the FW is significantly different 623 * (reallocating will be set above in IOC Facts comparison), then all 624 * buffers based on the IOC Facts will need to be freed before they are 625 * reallocated. 626 */ 627 if (reallocating) { 628 mps_iocfacts_free(sc); 629 mpssas_realloc_targets(sc, saved_facts.MaxTargets + 630 saved_facts.MaxVolumes); 631 } 632 633 /* 634 * Any deallocation has been completed. Now start reallocating 635 * if needed. Will only need to reallocate if attaching or if the new 636 * IOC Facts are different from the previous IOC Facts after a Diag 637 * Reset. Targets have already been allocated above if needed. 638 */ 639 error = 0; 640 while (attaching || reallocating) { 641 if ((error = mps_alloc_hw_queues(sc)) != 0) 642 break; 643 if ((error = mps_alloc_replies(sc)) != 0) 644 break; 645 if ((error = mps_alloc_requests(sc)) != 0) 646 break; 647 if ((error = mps_alloc_queues(sc)) != 0) 648 break; 649 650 break; 651 } 652 if (error) { 653 mps_dprint(sc, MPS_INIT|MPS_FAULT, 654 "Failed to alloc queues with error %d\n", error); 655 mps_free(sc); 656 return (error); 657 } 658 659 /* Always initialize the queues */ 660 bzero(sc->free_queue, sc->fqdepth * 4); 661 mps_init_queues(sc); 662 663 /* 664 * Always get the chip out of the reset state, but only panic if not 665 * attaching. If attaching and there is an error, that is handled by 666 * the OS. 667 */ 668 error = mps_transition_operational(sc); 669 if (error != 0) { 670 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to " 671 "transition to operational with error %d\n", error); 672 mps_free(sc); 673 return (error); 674 } 675 676 /* 677 * Finish the queue initialization. 678 * These are set here instead of in mps_init_queues() because the 679 * IOC resets these values during the state transition in 680 * mps_transition_operational(). The free index is set to 1 681 * because the corresponding index in the IOC is set to 0, and the 682 * IOC treats the queues as full if both are set to the same value. 683 * Hence the reason that the queue can't hold all of the possible 684 * replies. 685 */ 686 sc->replypostindex = 0; 687 mps_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); 688 mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, 0); 689 690 /* 691 * Attach the subsystems so they can prepare their event masks. 692 * XXX Should be dynamic so that IM/IR and user modules can attach 693 */ 694 error = 0; 695 while (attaching) { 696 mps_dprint(sc, MPS_INIT, "Attaching subsystems\n"); 697 if ((error = mps_attach_log(sc)) != 0) 698 break; 699 if ((error = mps_attach_sas(sc)) != 0) 700 break; 701 if ((error = mps_attach_user(sc)) != 0) 702 break; 703 break; 704 } 705 if (error) { 706 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to attach all " 707 "subsystems: error %d\n", error); 708 mps_free(sc); 709 return (error); 710 } 711 712 /* 713 * XXX If the number of MSI-X vectors changes during re-init, this 714 * won't see it and adjust. 715 */ 716 if (attaching && (error = mps_pci_setup_interrupts(sc)) != 0) { 717 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to setup " 718 "interrupts\n"); 719 mps_free(sc); 720 return (error); 721 } 722 723 /* 724 * Set flag if this is a WD controller. This shouldn't ever change, but 725 * reset it after a Diag Reset, just in case. 726 */ 727 sc->WD_available = FALSE; 728 if (pci_get_device(sc->mps_dev) == MPI2_MFGPAGE_DEVID_SSS6200) 729 sc->WD_available = TRUE; 730 731 return (error); 732 } 733 734 /* 735 * This is called if memory is being free (during detach for example) and when 736 * buffers need to be reallocated due to a Diag Reset. 737 */ 738 static void 739 mps_iocfacts_free(struct mps_softc *sc) 740 { 741 struct mps_command *cm; 742 int i; 743 744 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 745 746 if (sc->free_busaddr != 0) 747 bus_dmamap_unload(sc->queues_dmat, sc->queues_map); 748 if (sc->free_queue != NULL) 749 bus_dmamem_free(sc->queues_dmat, sc->free_queue, 750 sc->queues_map); 751 if (sc->queues_dmat != NULL) 752 bus_dma_tag_destroy(sc->queues_dmat); 753 754 if (sc->chain_frames != NULL) { 755 bus_dmamap_unload(sc->chain_dmat, sc->chain_map); 756 bus_dmamem_free(sc->chain_dmat, sc->chain_frames, 757 sc->chain_map); 758 } 759 if (sc->chain_dmat != NULL) 760 bus_dma_tag_destroy(sc->chain_dmat); 761 762 if (sc->sense_busaddr != 0) 763 bus_dmamap_unload(sc->sense_dmat, sc->sense_map); 764 if (sc->sense_frames != NULL) 765 bus_dmamem_free(sc->sense_dmat, sc->sense_frames, 766 sc->sense_map); 767 if (sc->sense_dmat != NULL) 768 bus_dma_tag_destroy(sc->sense_dmat); 769 770 if (sc->reply_busaddr != 0) 771 bus_dmamap_unload(sc->reply_dmat, sc->reply_map); 772 if (sc->reply_frames != NULL) 773 bus_dmamem_free(sc->reply_dmat, sc->reply_frames, 774 sc->reply_map); 775 if (sc->reply_dmat != NULL) 776 bus_dma_tag_destroy(sc->reply_dmat); 777 778 if (sc->req_busaddr != 0) 779 bus_dmamap_unload(sc->req_dmat, sc->req_map); 780 if (sc->req_frames != NULL) 781 bus_dmamem_free(sc->req_dmat, sc->req_frames, sc->req_map); 782 if (sc->req_dmat != NULL) 783 bus_dma_tag_destroy(sc->req_dmat); 784 785 if (sc->chains != NULL) 786 free(sc->chains, M_MPT2); 787 if (sc->commands != NULL) { 788 for (i = 1; i < sc->num_reqs; i++) { 789 cm = &sc->commands[i]; 790 bus_dmamap_destroy(sc->buffer_dmat, cm->cm_dmamap); 791 } 792 free(sc->commands, M_MPT2); 793 } 794 if (sc->buffer_dmat != NULL) 795 bus_dma_tag_destroy(sc->buffer_dmat); 796 797 mps_pci_free_interrupts(sc); 798 free(sc->queues, M_MPT2); 799 sc->queues = NULL; 800 } 801 802 /* 803 * The terms diag reset and hard reset are used interchangeably in the MPI 804 * docs to mean resetting the controller chip. In this code diag reset 805 * cleans everything up, and the hard reset function just sends the reset 806 * sequence to the chip. This should probably be refactored so that every 807 * subsystem gets a reset notification of some sort, and can clean up 808 * appropriately. 809 */ 810 int 811 mps_reinit(struct mps_softc *sc) 812 { 813 int error; 814 struct mpssas_softc *sassc; 815 816 sassc = sc->sassc; 817 818 MPS_FUNCTRACE(sc); 819 820 mtx_assert(&sc->mps_mtx, MA_OWNED); 821 822 mps_dprint(sc, MPS_INIT|MPS_INFO, "Reinitializing controller\n"); 823 if (sc->mps_flags & MPS_FLAGS_DIAGRESET) { 824 mps_dprint(sc, MPS_INIT, "Reset already in progress\n"); 825 return 0; 826 } 827 828 /* make sure the completion callbacks can recognize they're getting 829 * a NULL cm_reply due to a reset. 830 */ 831 sc->mps_flags |= MPS_FLAGS_DIAGRESET; 832 833 /* 834 * Mask interrupts here. 835 */ 836 mps_dprint(sc, MPS_INIT, "masking interrupts and resetting\n"); 837 mps_mask_intr(sc); 838 839 error = mps_diag_reset(sc, CAN_SLEEP); 840 if (error != 0) { 841 /* XXXSL No need to panic here */ 842 panic("%s hard reset failed with error %d\n", 843 __func__, error); 844 } 845 846 /* Restore the PCI state, including the MSI-X registers */ 847 mps_pci_restore(sc); 848 849 /* Give the I/O subsystem special priority to get itself prepared */ 850 mpssas_handle_reinit(sc); 851 852 /* 853 * Get IOC Facts and allocate all structures based on this information. 854 * The attach function will also call mps_iocfacts_allocate at startup. 855 * If relevant values have changed in IOC Facts, this function will free 856 * all of the memory based on IOC Facts and reallocate that memory. 857 */ 858 if ((error = mps_iocfacts_allocate(sc, FALSE)) != 0) { 859 panic("%s IOC Facts based allocation failed with error %d\n", 860 __func__, error); 861 } 862 863 /* 864 * Mapping structures will be re-allocated after getting IOC Page8, so 865 * free these structures here. 866 */ 867 mps_mapping_exit(sc); 868 869 /* 870 * The static page function currently read is IOC Page8. Others can be 871 * added in future. It's possible that the values in IOC Page8 have 872 * changed after a Diag Reset due to user modification, so always read 873 * these. Interrupts are masked, so unmask them before getting config 874 * pages. 875 */ 876 mps_unmask_intr(sc); 877 sc->mps_flags &= ~MPS_FLAGS_DIAGRESET; 878 mps_base_static_config_pages(sc); 879 880 /* 881 * Some mapping info is based in IOC Page8 data, so re-initialize the 882 * mapping tables. 883 */ 884 mps_mapping_initialize(sc); 885 886 /* 887 * Restart will reload the event masks clobbered by the reset, and 888 * then enable the port. 889 */ 890 mps_reregister_events(sc); 891 892 /* the end of discovery will release the simq, so we're done. */ 893 mps_dprint(sc, MPS_INIT|MPS_XINFO, "Finished sc %p post %u free %u\n", 894 sc, sc->replypostindex, sc->replyfreeindex); 895 896 mpssas_release_simq_reinit(sassc); 897 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 898 899 return 0; 900 } 901 902 /* Wait for the chip to ACK a word that we've put into its FIFO 903 * Wait for <timeout> seconds. In single loop wait for busy loop 904 * for 500 microseconds. 905 * Total is [ 0.5 * (2000 * <timeout>) ] in miliseconds. 906 * */ 907 static int 908 mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag) 909 { 910 911 u32 cntdn, count; 912 u32 int_status; 913 u32 doorbell; 914 915 count = 0; 916 cntdn = (sleep_flag == CAN_SLEEP) ? 1000*timeout : 2000*timeout; 917 do { 918 int_status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); 919 if (!(int_status & MPI2_HIS_SYS2IOC_DB_STATUS)) { 920 mps_dprint(sc, MPS_TRACE, 921 "%s: successful count(%d), timeout(%d)\n", 922 __func__, count, timeout); 923 return 0; 924 } else if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) { 925 doorbell = mps_regread(sc, MPI2_DOORBELL_OFFSET); 926 if ((doorbell & MPI2_IOC_STATE_MASK) == 927 MPI2_IOC_STATE_FAULT) { 928 mps_dprint(sc, MPS_FAULT, 929 "fault_state(0x%04x)!\n", doorbell); 930 return (EFAULT); 931 } 932 } else if (int_status == 0xFFFFFFFF) 933 goto out; 934 935 /* If it can sleep, sleep for 1 milisecond, else busy loop for 936 * 0.5 milisecond */ 937 if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) 938 msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, 939 "mpsdba", hz/1000); 940 else if (sleep_flag == CAN_SLEEP) 941 pause("mpsdba", hz/1000); 942 else 943 DELAY(500); 944 count++; 945 } while (--cntdn); 946 947 out: 948 mps_dprint(sc, MPS_FAULT, "%s: failed due to timeout count(%d), " 949 "int_status(%x)!\n", __func__, count, int_status); 950 return (ETIMEDOUT); 951 952 } 953 954 /* Wait for the chip to signal that the next word in its FIFO can be fetched */ 955 static int 956 mps_wait_db_int(struct mps_softc *sc) 957 { 958 int retry; 959 960 for (retry = 0; retry < MPS_DB_MAX_WAIT; retry++) { 961 if ((mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET) & 962 MPI2_HIS_IOC2SYS_DB_STATUS) != 0) 963 return (0); 964 DELAY(2000); 965 } 966 return (ETIMEDOUT); 967 } 968 969 /* Step through the synchronous command state machine, i.e. "Doorbell mode" */ 970 static int 971 mps_request_sync(struct mps_softc *sc, void *req, MPI2_DEFAULT_REPLY *reply, 972 int req_sz, int reply_sz, int timeout) 973 { 974 uint32_t *data32; 975 uint16_t *data16; 976 int i, count, ioc_sz, residual; 977 int sleep_flags = CAN_SLEEP; 978 979 if (curthread->td_no_sleeping != 0) 980 sleep_flags = NO_SLEEP; 981 982 /* Step 1 */ 983 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 984 985 /* Step 2 */ 986 if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) 987 return (EBUSY); 988 989 /* Step 3 990 * Announce that a message is coming through the doorbell. Messages 991 * are pushed at 32bit words, so round up if needed. 992 */ 993 count = (req_sz + 3) / 4; 994 mps_regwrite(sc, MPI2_DOORBELL_OFFSET, 995 (MPI2_FUNCTION_HANDSHAKE << MPI2_DOORBELL_FUNCTION_SHIFT) | 996 (count << MPI2_DOORBELL_ADD_DWORDS_SHIFT)); 997 998 /* Step 4 */ 999 if (mps_wait_db_int(sc) || 1000 (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) == 0) { 1001 mps_dprint(sc, MPS_FAULT, "Doorbell failed to activate\n"); 1002 return (ENXIO); 1003 } 1004 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1005 if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) { 1006 mps_dprint(sc, MPS_FAULT, "Doorbell handshake failed\n"); 1007 return (ENXIO); 1008 } 1009 1010 /* Step 5 */ 1011 /* Clock out the message data synchronously in 32-bit dwords*/ 1012 data32 = (uint32_t *)req; 1013 for (i = 0; i < count; i++) { 1014 mps_regwrite(sc, MPI2_DOORBELL_OFFSET, htole32(data32[i])); 1015 if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) { 1016 mps_dprint(sc, MPS_FAULT, 1017 "Timeout while writing doorbell\n"); 1018 return (ENXIO); 1019 } 1020 } 1021 1022 /* Step 6 */ 1023 /* Clock in the reply in 16-bit words. The total length of the 1024 * message is always in the 4th byte, so clock out the first 2 words 1025 * manually, then loop the rest. 1026 */ 1027 data16 = (uint16_t *)reply; 1028 if (mps_wait_db_int(sc) != 0) { 1029 mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 0\n"); 1030 return (ENXIO); 1031 } 1032 data16[0] = 1033 mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; 1034 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1035 if (mps_wait_db_int(sc) != 0) { 1036 mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 1\n"); 1037 return (ENXIO); 1038 } 1039 data16[1] = 1040 mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; 1041 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1042 1043 /* Number of 32bit words in the message */ 1044 ioc_sz = reply->MsgLength; 1045 1046 /* 1047 * Figure out how many 16bit words to clock in without overrunning. 1048 * The precision loss with dividing reply_sz can safely be 1049 * ignored because the messages can only be multiples of 32bits. 1050 */ 1051 residual = 0; 1052 count = MIN((reply_sz / 4), ioc_sz) * 2; 1053 if (count < ioc_sz * 2) { 1054 residual = ioc_sz * 2 - count; 1055 mps_dprint(sc, MPS_ERROR, "Driver error, throwing away %d " 1056 "residual message words\n", residual); 1057 } 1058 1059 for (i = 2; i < count; i++) { 1060 if (mps_wait_db_int(sc) != 0) { 1061 mps_dprint(sc, MPS_FAULT, 1062 "Timeout reading doorbell %d\n", i); 1063 return (ENXIO); 1064 } 1065 data16[i] = mps_regread(sc, MPI2_DOORBELL_OFFSET) & 1066 MPI2_DOORBELL_DATA_MASK; 1067 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1068 } 1069 1070 /* 1071 * Pull out residual words that won't fit into the provided buffer. 1072 * This keeps the chip from hanging due to a driver programming 1073 * error. 1074 */ 1075 while (residual--) { 1076 if (mps_wait_db_int(sc) != 0) { 1077 mps_dprint(sc, MPS_FAULT, 1078 "Timeout reading doorbell\n"); 1079 return (ENXIO); 1080 } 1081 (void)mps_regread(sc, MPI2_DOORBELL_OFFSET); 1082 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1083 } 1084 1085 /* Step 7 */ 1086 if (mps_wait_db_int(sc) != 0) { 1087 mps_dprint(sc, MPS_FAULT, "Timeout waiting to exit doorbell\n"); 1088 return (ENXIO); 1089 } 1090 if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) 1091 mps_dprint(sc, MPS_FAULT, "Warning, doorbell still active\n"); 1092 mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); 1093 1094 return (0); 1095 } 1096 1097 static void 1098 mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm) 1099 { 1100 reply_descriptor rd; 1101 MPS_FUNCTRACE(sc); 1102 mps_dprint(sc, MPS_TRACE, "SMID %u cm %p ccb %p\n", 1103 cm->cm_desc.Default.SMID, cm, cm->cm_ccb); 1104 1105 if (sc->mps_flags & MPS_FLAGS_ATTACH_DONE && !(sc->mps_flags & MPS_FLAGS_SHUTDOWN)) 1106 mtx_assert(&sc->mps_mtx, MA_OWNED); 1107 1108 if (++sc->io_cmds_active > sc->io_cmds_highwater) 1109 sc->io_cmds_highwater++; 1110 rd.u.low = cm->cm_desc.Words.Low; 1111 rd.u.high = cm->cm_desc.Words.High; 1112 rd.word = htole64(rd.word); 1113 1114 KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, ("command not busy\n")); 1115 cm->cm_state = MPS_CM_STATE_INQUEUE; 1116 1117 /* TODO-We may need to make below regwrite atomic */ 1118 mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_LOW_OFFSET, 1119 rd.u.low); 1120 mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_HIGH_OFFSET, 1121 rd.u.high); 1122 } 1123 1124 /* 1125 * Just the FACTS, ma'am. 1126 */ 1127 static int 1128 mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts) 1129 { 1130 MPI2_DEFAULT_REPLY *reply; 1131 MPI2_IOC_FACTS_REQUEST request; 1132 int error, req_sz, reply_sz; 1133 1134 MPS_FUNCTRACE(sc); 1135 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 1136 1137 req_sz = sizeof(MPI2_IOC_FACTS_REQUEST); 1138 reply_sz = sizeof(MPI2_IOC_FACTS_REPLY); 1139 reply = (MPI2_DEFAULT_REPLY *)facts; 1140 1141 bzero(&request, req_sz); 1142 request.Function = MPI2_FUNCTION_IOC_FACTS; 1143 error = mps_request_sync(sc, &request, reply, req_sz, reply_sz, 5); 1144 mps_dprint(sc, MPS_INIT, "%s exit error= %d\n", __func__, error); 1145 1146 return (error); 1147 } 1148 1149 static int 1150 mps_send_iocinit(struct mps_softc *sc) 1151 { 1152 MPI2_IOC_INIT_REQUEST init; 1153 MPI2_DEFAULT_REPLY reply; 1154 int req_sz, reply_sz, error; 1155 struct timeval now; 1156 uint64_t time_in_msec; 1157 1158 MPS_FUNCTRACE(sc); 1159 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 1160 1161 /* Do a quick sanity check on proper initialization */ 1162 if ((sc->pqdepth == 0) || (sc->fqdepth == 0) || (sc->reqframesz == 0) 1163 || (sc->replyframesz == 0)) { 1164 mps_dprint(sc, MPS_INIT|MPS_ERROR, 1165 "Driver not fully initialized for IOCInit\n"); 1166 return (EINVAL); 1167 } 1168 1169 req_sz = sizeof(MPI2_IOC_INIT_REQUEST); 1170 reply_sz = sizeof(MPI2_IOC_INIT_REPLY); 1171 bzero(&init, req_sz); 1172 bzero(&reply, reply_sz); 1173 1174 /* 1175 * Fill in the init block. Note that most addresses are 1176 * deliberately in the lower 32bits of memory. This is a micro- 1177 * optimzation for PCI/PCIX, though it's not clear if it helps PCIe. 1178 */ 1179 init.Function = MPI2_FUNCTION_IOC_INIT; 1180 init.WhoInit = MPI2_WHOINIT_HOST_DRIVER; 1181 init.MsgVersion = htole16(MPI2_VERSION); 1182 init.HeaderVersion = htole16(MPI2_HEADER_VERSION); 1183 init.SystemRequestFrameSize = htole16((uint16_t)(sc->reqframesz / 4)); 1184 init.ReplyDescriptorPostQueueDepth = htole16(sc->pqdepth); 1185 init.ReplyFreeQueueDepth = htole16(sc->fqdepth); 1186 init.SenseBufferAddressHigh = 0; 1187 init.SystemReplyAddressHigh = 0; 1188 init.SystemRequestFrameBaseAddress.High = 0; 1189 init.SystemRequestFrameBaseAddress.Low = htole32((uint32_t)sc->req_busaddr); 1190 init.ReplyDescriptorPostQueueAddress.High = 0; 1191 init.ReplyDescriptorPostQueueAddress.Low = htole32((uint32_t)sc->post_busaddr); 1192 init.ReplyFreeQueueAddress.High = 0; 1193 init.ReplyFreeQueueAddress.Low = htole32((uint32_t)sc->free_busaddr); 1194 getmicrotime(&now); 1195 time_in_msec = (now.tv_sec * 1000 + now.tv_usec/1000); 1196 init.TimeStamp.High = htole32((time_in_msec >> 32) & 0xFFFFFFFF); 1197 init.TimeStamp.Low = htole32(time_in_msec & 0xFFFFFFFF); 1198 1199 error = mps_request_sync(sc, &init, &reply, req_sz, reply_sz, 5); 1200 if ((reply.IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) 1201 error = ENXIO; 1202 1203 mps_dprint(sc, MPS_INIT, "IOCInit status= 0x%x\n", reply.IOCStatus); 1204 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 1205 return (error); 1206 } 1207 1208 void 1209 mps_memaddr_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) 1210 { 1211 bus_addr_t *addr; 1212 1213 addr = arg; 1214 *addr = segs[0].ds_addr; 1215 } 1216 1217 void 1218 mps_memaddr_wait_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) 1219 { 1220 struct mps_busdma_context *ctx; 1221 int need_unload, need_free; 1222 1223 ctx = (struct mps_busdma_context *)arg; 1224 need_unload = 0; 1225 need_free = 0; 1226 1227 mps_lock(ctx->softc); 1228 ctx->error = error; 1229 ctx->completed = 1; 1230 if ((error == 0) && (ctx->abandoned == 0)) { 1231 *ctx->addr = segs[0].ds_addr; 1232 } else { 1233 if (nsegs != 0) 1234 need_unload = 1; 1235 if (ctx->abandoned != 0) 1236 need_free = 1; 1237 } 1238 if (need_free == 0) 1239 wakeup(ctx); 1240 1241 mps_unlock(ctx->softc); 1242 1243 if (need_unload != 0) { 1244 bus_dmamap_unload(ctx->buffer_dmat, 1245 ctx->buffer_dmamap); 1246 *ctx->addr = 0; 1247 } 1248 1249 if (need_free != 0) 1250 free(ctx, M_MPSUSER); 1251 } 1252 1253 static int 1254 mps_alloc_queues(struct mps_softc *sc) 1255 { 1256 struct mps_queue *q; 1257 u_int nq, i; 1258 1259 nq = sc->msi_msgs; 1260 mps_dprint(sc, MPS_INIT|MPS_XINFO, "Allocating %d I/O queues\n", nq); 1261 1262 sc->queues = malloc(sizeof(struct mps_queue) * nq, M_MPT2, 1263 M_NOWAIT|M_ZERO); 1264 if (sc->queues == NULL) 1265 return (ENOMEM); 1266 1267 for (i = 0; i < nq; i++) { 1268 q = &sc->queues[i]; 1269 mps_dprint(sc, MPS_INIT, "Configuring queue %d %p\n", i, q); 1270 q->sc = sc; 1271 q->qnum = i; 1272 } 1273 1274 return (0); 1275 } 1276 1277 static int 1278 mps_alloc_hw_queues(struct mps_softc *sc) 1279 { 1280 bus_addr_t queues_busaddr; 1281 uint8_t *queues; 1282 int qsize, fqsize, pqsize; 1283 1284 /* 1285 * The reply free queue contains 4 byte entries in multiples of 16 and 1286 * aligned on a 16 byte boundary. There must always be an unused entry. 1287 * This queue supplies fresh reply frames for the firmware to use. 1288 * 1289 * The reply descriptor post queue contains 8 byte entries in 1290 * multiples of 16 and aligned on a 16 byte boundary. This queue 1291 * contains filled-in reply frames sent from the firmware to the host. 1292 * 1293 * These two queues are allocated together for simplicity. 1294 */ 1295 sc->fqdepth = roundup2(sc->num_replies + 1, 16); 1296 sc->pqdepth = roundup2(sc->num_replies + 1, 16); 1297 fqsize= sc->fqdepth * 4; 1298 pqsize = sc->pqdepth * 8; 1299 qsize = fqsize + pqsize; 1300 1301 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1302 16, 0, /* algnmnt, boundary */ 1303 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ 1304 BUS_SPACE_MAXADDR, /* highaddr */ 1305 NULL, NULL, /* filter, filterarg */ 1306 qsize, /* maxsize */ 1307 1, /* nsegments */ 1308 qsize, /* maxsegsize */ 1309 0, /* flags */ 1310 NULL, NULL, /* lockfunc, lockarg */ 1311 &sc->queues_dmat)) { 1312 mps_dprint(sc, MPS_ERROR, "Cannot allocate queues DMA tag\n"); 1313 return (ENOMEM); 1314 } 1315 if (bus_dmamem_alloc(sc->queues_dmat, (void **)&queues, BUS_DMA_NOWAIT, 1316 &sc->queues_map)) { 1317 mps_dprint(sc, MPS_ERROR, "Cannot allocate queues memory\n"); 1318 return (ENOMEM); 1319 } 1320 bzero(queues, qsize); 1321 bus_dmamap_load(sc->queues_dmat, sc->queues_map, queues, qsize, 1322 mps_memaddr_cb, &queues_busaddr, 0); 1323 1324 sc->free_queue = (uint32_t *)queues; 1325 sc->free_busaddr = queues_busaddr; 1326 sc->post_queue = (MPI2_REPLY_DESCRIPTORS_UNION *)(queues + fqsize); 1327 sc->post_busaddr = queues_busaddr + fqsize; 1328 mps_dprint(sc, MPS_INIT, "free queue busaddr= %#016jx size= %d\n", 1329 (uintmax_t)sc->free_busaddr, fqsize); 1330 mps_dprint(sc, MPS_INIT, "reply queue busaddr= %#016jx size= %d\n", 1331 (uintmax_t)sc->post_busaddr, pqsize); 1332 1333 return (0); 1334 } 1335 1336 static int 1337 mps_alloc_replies(struct mps_softc *sc) 1338 { 1339 int rsize, num_replies; 1340 1341 /* Store the reply frame size in bytes rather than as 32bit words */ 1342 sc->replyframesz = sc->facts->ReplyFrameSize * 4; 1343 1344 /* 1345 * sc->num_replies should be one less than sc->fqdepth. We need to 1346 * allocate space for sc->fqdepth replies, but only sc->num_replies 1347 * replies can be used at once. 1348 */ 1349 num_replies = max(sc->fqdepth, sc->num_replies); 1350 1351 rsize = sc->replyframesz * num_replies; 1352 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1353 4, 0, /* algnmnt, boundary */ 1354 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ 1355 BUS_SPACE_MAXADDR, /* highaddr */ 1356 NULL, NULL, /* filter, filterarg */ 1357 rsize, /* maxsize */ 1358 1, /* nsegments */ 1359 rsize, /* maxsegsize */ 1360 0, /* flags */ 1361 NULL, NULL, /* lockfunc, lockarg */ 1362 &sc->reply_dmat)) { 1363 mps_dprint(sc, MPS_ERROR, "Cannot allocate replies DMA tag\n"); 1364 return (ENOMEM); 1365 } 1366 if (bus_dmamem_alloc(sc->reply_dmat, (void **)&sc->reply_frames, 1367 BUS_DMA_NOWAIT, &sc->reply_map)) { 1368 mps_dprint(sc, MPS_ERROR, "Cannot allocate replies memory\n"); 1369 return (ENOMEM); 1370 } 1371 bzero(sc->reply_frames, rsize); 1372 bus_dmamap_load(sc->reply_dmat, sc->reply_map, sc->reply_frames, rsize, 1373 mps_memaddr_cb, &sc->reply_busaddr, 0); 1374 1375 mps_dprint(sc, MPS_INIT, "reply frames busaddr= %#016jx size= %d\n", 1376 (uintmax_t)sc->reply_busaddr, rsize); 1377 1378 return (0); 1379 } 1380 1381 static void 1382 mps_load_chains_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) 1383 { 1384 struct mps_softc *sc = arg; 1385 struct mps_chain *chain; 1386 bus_size_t bo; 1387 int i, o, s; 1388 1389 if (error != 0) 1390 return; 1391 1392 for (i = 0, o = 0, s = 0; s < nsegs; s++) { 1393 for (bo = 0; bo + sc->reqframesz <= segs[s].ds_len; 1394 bo += sc->reqframesz) { 1395 chain = &sc->chains[i++]; 1396 chain->chain =(MPI2_SGE_IO_UNION *)(sc->chain_frames+o); 1397 chain->chain_busaddr = segs[s].ds_addr + bo; 1398 o += sc->reqframesz; 1399 mps_free_chain(sc, chain); 1400 } 1401 if (bo != segs[s].ds_len) 1402 o += segs[s].ds_len - bo; 1403 } 1404 sc->chain_free_lowwater = i; 1405 } 1406 1407 static int 1408 mps_alloc_requests(struct mps_softc *sc) 1409 { 1410 struct mps_command *cm; 1411 int i, rsize, nsegs; 1412 1413 rsize = sc->reqframesz * sc->num_reqs; 1414 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1415 16, 0, /* algnmnt, boundary */ 1416 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ 1417 BUS_SPACE_MAXADDR, /* highaddr */ 1418 NULL, NULL, /* filter, filterarg */ 1419 rsize, /* maxsize */ 1420 1, /* nsegments */ 1421 rsize, /* maxsegsize */ 1422 0, /* flags */ 1423 NULL, NULL, /* lockfunc, lockarg */ 1424 &sc->req_dmat)) { 1425 mps_dprint(sc, MPS_ERROR, "Cannot allocate request DMA tag\n"); 1426 return (ENOMEM); 1427 } 1428 if (bus_dmamem_alloc(sc->req_dmat, (void **)&sc->req_frames, 1429 BUS_DMA_NOWAIT, &sc->req_map)) { 1430 mps_dprint(sc, MPS_ERROR, "Cannot allocate request memory\n"); 1431 return (ENOMEM); 1432 } 1433 bzero(sc->req_frames, rsize); 1434 bus_dmamap_load(sc->req_dmat, sc->req_map, sc->req_frames, rsize, 1435 mps_memaddr_cb, &sc->req_busaddr, 0); 1436 mps_dprint(sc, MPS_INIT, "request frames busaddr= %#016jx size= %d\n", 1437 (uintmax_t)sc->req_busaddr, rsize); 1438 1439 sc->chains = malloc(sizeof(struct mps_chain) * sc->num_chains, M_MPT2, 1440 M_NOWAIT | M_ZERO); 1441 if (!sc->chains) { 1442 mps_dprint(sc, MPS_ERROR, "Cannot allocate chain memory\n"); 1443 return (ENOMEM); 1444 } 1445 rsize = sc->reqframesz * sc->num_chains; 1446 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1447 16, 0, /* algnmnt, boundary */ 1448 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ 1449 BUS_SPACE_MAXADDR, /* highaddr */ 1450 NULL, NULL, /* filter, filterarg */ 1451 rsize, /* maxsize */ 1452 howmany(rsize, PAGE_SIZE), /* nsegments */ 1453 rsize, /* maxsegsize */ 1454 0, /* flags */ 1455 NULL, NULL, /* lockfunc, lockarg */ 1456 &sc->chain_dmat)) { 1457 mps_dprint(sc, MPS_ERROR, "Cannot allocate chain DMA tag\n"); 1458 return (ENOMEM); 1459 } 1460 if (bus_dmamem_alloc(sc->chain_dmat, (void **)&sc->chain_frames, 1461 BUS_DMA_NOWAIT | BUS_DMA_ZERO, &sc->chain_map)) { 1462 mps_dprint(sc, MPS_ERROR, "Cannot allocate chain memory\n"); 1463 return (ENOMEM); 1464 } 1465 if (bus_dmamap_load(sc->chain_dmat, sc->chain_map, sc->chain_frames, 1466 rsize, mps_load_chains_cb, sc, BUS_DMA_NOWAIT)) { 1467 mps_dprint(sc, MPS_ERROR, "Cannot load chain memory\n"); 1468 bus_dmamem_free(sc->chain_dmat, sc->chain_frames, 1469 sc->chain_map); 1470 return (ENOMEM); 1471 } 1472 1473 rsize = MPS_SENSE_LEN * sc->num_reqs; 1474 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1475 1, 0, /* algnmnt, boundary */ 1476 BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ 1477 BUS_SPACE_MAXADDR, /* highaddr */ 1478 NULL, NULL, /* filter, filterarg */ 1479 rsize, /* maxsize */ 1480 1, /* nsegments */ 1481 rsize, /* maxsegsize */ 1482 0, /* flags */ 1483 NULL, NULL, /* lockfunc, lockarg */ 1484 &sc->sense_dmat)) { 1485 mps_dprint(sc, MPS_ERROR, "Cannot allocate sense DMA tag\n"); 1486 return (ENOMEM); 1487 } 1488 if (bus_dmamem_alloc(sc->sense_dmat, (void **)&sc->sense_frames, 1489 BUS_DMA_NOWAIT, &sc->sense_map)) { 1490 mps_dprint(sc, MPS_ERROR, "Cannot allocate sense memory\n"); 1491 return (ENOMEM); 1492 } 1493 bzero(sc->sense_frames, rsize); 1494 bus_dmamap_load(sc->sense_dmat, sc->sense_map, sc->sense_frames, rsize, 1495 mps_memaddr_cb, &sc->sense_busaddr, 0); 1496 mps_dprint(sc, MPS_INIT, "sense frames busaddr= %#016jx size= %d\n", 1497 (uintmax_t)sc->sense_busaddr, rsize); 1498 1499 nsegs = (sc->maxio / PAGE_SIZE) + 1; 1500 if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1501 1, 0, /* algnmnt, boundary */ 1502 BUS_SPACE_MAXADDR, /* lowaddr */ 1503 BUS_SPACE_MAXADDR, /* highaddr */ 1504 NULL, NULL, /* filter, filterarg */ 1505 BUS_SPACE_MAXSIZE_32BIT,/* maxsize */ 1506 nsegs, /* nsegments */ 1507 BUS_SPACE_MAXSIZE_24BIT,/* maxsegsize */ 1508 BUS_DMA_ALLOCNOW, /* flags */ 1509 busdma_lock_mutex, /* lockfunc */ 1510 &sc->mps_mtx, /* lockarg */ 1511 &sc->buffer_dmat)) { 1512 mps_dprint(sc, MPS_ERROR, "Cannot allocate buffer DMA tag\n"); 1513 return (ENOMEM); 1514 } 1515 1516 /* 1517 * SMID 0 cannot be used as a free command per the firmware spec. 1518 * Just drop that command instead of risking accounting bugs. 1519 */ 1520 sc->commands = malloc(sizeof(struct mps_command) * sc->num_reqs, 1521 M_MPT2, M_WAITOK | M_ZERO); 1522 if(!sc->commands) { 1523 mps_dprint(sc, MPS_ERROR, "Cannot allocate command memory\n"); 1524 return (ENOMEM); 1525 } 1526 for (i = 1; i < sc->num_reqs; i++) { 1527 cm = &sc->commands[i]; 1528 cm->cm_req = sc->req_frames + i * sc->reqframesz; 1529 cm->cm_req_busaddr = sc->req_busaddr + i * sc->reqframesz; 1530 cm->cm_sense = &sc->sense_frames[i]; 1531 cm->cm_sense_busaddr = sc->sense_busaddr + i * MPS_SENSE_LEN; 1532 cm->cm_desc.Default.SMID = i; 1533 cm->cm_sc = sc; 1534 cm->cm_state = MPS_CM_STATE_BUSY; 1535 TAILQ_INIT(&cm->cm_chain_list); 1536 callout_init_mtx(&cm->cm_callout, &sc->mps_mtx, 0); 1537 1538 /* XXX Is a failure here a critical problem? */ 1539 if (bus_dmamap_create(sc->buffer_dmat, 0, &cm->cm_dmamap) == 0) 1540 if (i <= sc->num_prireqs) 1541 mps_free_high_priority_command(sc, cm); 1542 else 1543 mps_free_command(sc, cm); 1544 else { 1545 panic("failed to allocate command %d\n", i); 1546 sc->num_reqs = i; 1547 break; 1548 } 1549 } 1550 1551 return (0); 1552 } 1553 1554 static int 1555 mps_init_queues(struct mps_softc *sc) 1556 { 1557 int i; 1558 1559 memset((uint8_t *)sc->post_queue, 0xff, sc->pqdepth * 8); 1560 1561 /* 1562 * According to the spec, we need to use one less reply than we 1563 * have space for on the queue. So sc->num_replies (the number we 1564 * use) should be less than sc->fqdepth (allocated size). 1565 */ 1566 if (sc->num_replies >= sc->fqdepth) 1567 return (EINVAL); 1568 1569 /* 1570 * Initialize all of the free queue entries. 1571 */ 1572 for (i = 0; i < sc->fqdepth; i++) 1573 sc->free_queue[i] = sc->reply_busaddr + (i * sc->replyframesz); 1574 sc->replyfreeindex = sc->num_replies; 1575 1576 return (0); 1577 } 1578 1579 /* Get the driver parameter tunables. Lowest priority are the driver defaults. 1580 * Next are the global settings, if they exist. Highest are the per-unit 1581 * settings, if they exist. 1582 */ 1583 void 1584 mps_get_tunables(struct mps_softc *sc) 1585 { 1586 char tmpstr[80], mps_debug[80]; 1587 1588 /* XXX default to some debugging for now */ 1589 sc->mps_debug = MPS_INFO|MPS_FAULT; 1590 sc->disable_msix = 0; 1591 sc->disable_msi = 0; 1592 sc->max_msix = MPS_MSIX_MAX; 1593 sc->max_chains = MPS_CHAIN_FRAMES; 1594 sc->max_io_pages = MPS_MAXIO_PAGES; 1595 sc->enable_ssu = MPS_SSU_ENABLE_SSD_DISABLE_HDD; 1596 sc->spinup_wait_time = DEFAULT_SPINUP_WAIT; 1597 sc->use_phynum = 1; 1598 sc->max_reqframes = MPS_REQ_FRAMES; 1599 sc->max_prireqframes = MPS_PRI_REQ_FRAMES; 1600 sc->max_replyframes = MPS_REPLY_FRAMES; 1601 sc->max_evtframes = MPS_EVT_REPLY_FRAMES; 1602 1603 /* 1604 * Grab the global variables. 1605 */ 1606 bzero(mps_debug, 80); 1607 if (TUNABLE_STR_FETCH("hw.mps.debug_level", mps_debug, 80) != 0) 1608 mps_parse_debug(sc, mps_debug); 1609 TUNABLE_INT_FETCH("hw.mps.disable_msix", &sc->disable_msix); 1610 TUNABLE_INT_FETCH("hw.mps.disable_msi", &sc->disable_msi); 1611 TUNABLE_INT_FETCH("hw.mps.max_msix", &sc->max_msix); 1612 TUNABLE_INT_FETCH("hw.mps.max_chains", &sc->max_chains); 1613 TUNABLE_INT_FETCH("hw.mps.max_io_pages", &sc->max_io_pages); 1614 TUNABLE_INT_FETCH("hw.mps.enable_ssu", &sc->enable_ssu); 1615 TUNABLE_INT_FETCH("hw.mps.spinup_wait_time", &sc->spinup_wait_time); 1616 TUNABLE_INT_FETCH("hw.mps.use_phy_num", &sc->use_phynum); 1617 TUNABLE_INT_FETCH("hw.mps.max_reqframes", &sc->max_reqframes); 1618 TUNABLE_INT_FETCH("hw.mps.max_prireqframes", &sc->max_prireqframes); 1619 TUNABLE_INT_FETCH("hw.mps.max_replyframes", &sc->max_replyframes); 1620 TUNABLE_INT_FETCH("hw.mps.max_evtframes", &sc->max_evtframes); 1621 1622 /* Grab the unit-instance variables */ 1623 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.debug_level", 1624 device_get_unit(sc->mps_dev)); 1625 bzero(mps_debug, 80); 1626 if (TUNABLE_STR_FETCH(tmpstr, mps_debug, 80) != 0) 1627 mps_parse_debug(sc, mps_debug); 1628 1629 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msix", 1630 device_get_unit(sc->mps_dev)); 1631 TUNABLE_INT_FETCH(tmpstr, &sc->disable_msix); 1632 1633 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msi", 1634 device_get_unit(sc->mps_dev)); 1635 TUNABLE_INT_FETCH(tmpstr, &sc->disable_msi); 1636 1637 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_msix", 1638 device_get_unit(sc->mps_dev)); 1639 TUNABLE_INT_FETCH(tmpstr, &sc->max_msix); 1640 1641 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_chains", 1642 device_get_unit(sc->mps_dev)); 1643 TUNABLE_INT_FETCH(tmpstr, &sc->max_chains); 1644 1645 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_io_pages", 1646 device_get_unit(sc->mps_dev)); 1647 TUNABLE_INT_FETCH(tmpstr, &sc->max_io_pages); 1648 1649 bzero(sc->exclude_ids, sizeof(sc->exclude_ids)); 1650 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.exclude_ids", 1651 device_get_unit(sc->mps_dev)); 1652 TUNABLE_STR_FETCH(tmpstr, sc->exclude_ids, sizeof(sc->exclude_ids)); 1653 1654 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.enable_ssu", 1655 device_get_unit(sc->mps_dev)); 1656 TUNABLE_INT_FETCH(tmpstr, &sc->enable_ssu); 1657 1658 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.spinup_wait_time", 1659 device_get_unit(sc->mps_dev)); 1660 TUNABLE_INT_FETCH(tmpstr, &sc->spinup_wait_time); 1661 1662 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.use_phy_num", 1663 device_get_unit(sc->mps_dev)); 1664 TUNABLE_INT_FETCH(tmpstr, &sc->use_phynum); 1665 1666 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_reqframes", 1667 device_get_unit(sc->mps_dev)); 1668 TUNABLE_INT_FETCH(tmpstr, &sc->max_reqframes); 1669 1670 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_prireqframes", 1671 device_get_unit(sc->mps_dev)); 1672 TUNABLE_INT_FETCH(tmpstr, &sc->max_prireqframes); 1673 1674 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_replyframes", 1675 device_get_unit(sc->mps_dev)); 1676 TUNABLE_INT_FETCH(tmpstr, &sc->max_replyframes); 1677 1678 snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_evtframes", 1679 device_get_unit(sc->mps_dev)); 1680 TUNABLE_INT_FETCH(tmpstr, &sc->max_evtframes); 1681 1682 } 1683 1684 static void 1685 mps_setup_sysctl(struct mps_softc *sc) 1686 { 1687 struct sysctl_ctx_list *sysctl_ctx = NULL; 1688 struct sysctl_oid *sysctl_tree = NULL; 1689 char tmpstr[80], tmpstr2[80]; 1690 1691 /* 1692 * Setup the sysctl variable so the user can change the debug level 1693 * on the fly. 1694 */ 1695 snprintf(tmpstr, sizeof(tmpstr), "MPS controller %d", 1696 device_get_unit(sc->mps_dev)); 1697 snprintf(tmpstr2, sizeof(tmpstr2), "%d", device_get_unit(sc->mps_dev)); 1698 1699 sysctl_ctx = device_get_sysctl_ctx(sc->mps_dev); 1700 if (sysctl_ctx != NULL) 1701 sysctl_tree = device_get_sysctl_tree(sc->mps_dev); 1702 1703 if (sysctl_tree == NULL) { 1704 sysctl_ctx_init(&sc->sysctl_ctx); 1705 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx, 1706 SYSCTL_STATIC_CHILDREN(_hw_mps), OID_AUTO, tmpstr2, 1707 CTLFLAG_RD, 0, tmpstr); 1708 if (sc->sysctl_tree == NULL) 1709 return; 1710 sysctl_ctx = &sc->sysctl_ctx; 1711 sysctl_tree = sc->sysctl_tree; 1712 } 1713 1714 SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1715 OID_AUTO, "debug_level", CTLTYPE_STRING | CTLFLAG_RW |CTLFLAG_MPSAFE, 1716 sc, 0, mps_debug_sysctl, "A", "mps debug level"); 1717 1718 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1719 OID_AUTO, "disable_msix", CTLFLAG_RD, &sc->disable_msix, 0, 1720 "Disable the use of MSI-X interrupts"); 1721 1722 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1723 OID_AUTO, "disable_msi", CTLFLAG_RD, &sc->disable_msi, 0, 1724 "Disable the use of MSI interrupts"); 1725 1726 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1727 OID_AUTO, "max_msix", CTLFLAG_RD, &sc->max_msix, 0, 1728 "User-defined maximum number of MSIX queues"); 1729 1730 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1731 OID_AUTO, "msix_msgs", CTLFLAG_RD, &sc->msi_msgs, 0, 1732 "Negotiated number of MSIX queues"); 1733 1734 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1735 OID_AUTO, "max_reqframes", CTLFLAG_RD, &sc->max_reqframes, 0, 1736 "Total number of allocated request frames"); 1737 1738 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1739 OID_AUTO, "max_prireqframes", CTLFLAG_RD, &sc->max_prireqframes, 0, 1740 "Total number of allocated high priority request frames"); 1741 1742 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1743 OID_AUTO, "max_replyframes", CTLFLAG_RD, &sc->max_replyframes, 0, 1744 "Total number of allocated reply frames"); 1745 1746 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1747 OID_AUTO, "max_evtframes", CTLFLAG_RD, &sc->max_evtframes, 0, 1748 "Total number of event frames allocated"); 1749 1750 SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1751 OID_AUTO, "firmware_version", CTLFLAG_RD, sc->fw_version, 1752 strlen(sc->fw_version), "firmware version"); 1753 1754 SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1755 OID_AUTO, "driver_version", CTLFLAG_RD, MPS_DRIVER_VERSION, 1756 strlen(MPS_DRIVER_VERSION), "driver version"); 1757 1758 SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1759 OID_AUTO, "msg_version", CTLFLAG_RD, sc->msg_version, 1760 strlen(sc->msg_version), "message interface version"); 1761 1762 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1763 OID_AUTO, "io_cmds_active", CTLFLAG_RD, 1764 &sc->io_cmds_active, 0, "number of currently active commands"); 1765 1766 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1767 OID_AUTO, "io_cmds_highwater", CTLFLAG_RD, 1768 &sc->io_cmds_highwater, 0, "maximum active commands seen"); 1769 1770 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1771 OID_AUTO, "chain_free", CTLFLAG_RD, 1772 &sc->chain_free, 0, "number of free chain elements"); 1773 1774 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1775 OID_AUTO, "chain_free_lowwater", CTLFLAG_RD, 1776 &sc->chain_free_lowwater, 0,"lowest number of free chain elements"); 1777 1778 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1779 OID_AUTO, "max_chains", CTLFLAG_RD, 1780 &sc->max_chains, 0,"maximum chain frames that will be allocated"); 1781 1782 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1783 OID_AUTO, "max_io_pages", CTLFLAG_RD, 1784 &sc->max_io_pages, 0,"maximum pages to allow per I/O (if <1 use " 1785 "IOCFacts)"); 1786 1787 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1788 OID_AUTO, "enable_ssu", CTLFLAG_RW, &sc->enable_ssu, 0, 1789 "enable SSU to SATA SSD/HDD at shutdown"); 1790 1791 SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1792 OID_AUTO, "chain_alloc_fail", CTLFLAG_RD, 1793 &sc->chain_alloc_fail, "chain allocation failures"); 1794 1795 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1796 OID_AUTO, "spinup_wait_time", CTLFLAG_RD, 1797 &sc->spinup_wait_time, DEFAULT_SPINUP_WAIT, "seconds to wait for " 1798 "spinup after SATA ID error"); 1799 1800 SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1801 OID_AUTO, "mapping_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, 1802 mps_mapping_dump, "A", "Mapping Table Dump"); 1803 1804 SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1805 OID_AUTO, "encl_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, 1806 mps_mapping_encl_dump, "A", "Enclosure Table Dump"); 1807 1808 SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1809 OID_AUTO, "dump_reqs", CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_SKIP, sc, 0, 1810 mps_dump_reqs, "I", "Dump Active Requests"); 1811 1812 SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), 1813 OID_AUTO, "use_phy_num", CTLFLAG_RD, &sc->use_phynum, 0, 1814 "Use the phy number for enumeration"); 1815 } 1816 1817 static struct mps_debug_string { 1818 char *name; 1819 int flag; 1820 } mps_debug_strings[] = { 1821 {"info", MPS_INFO}, 1822 {"fault", MPS_FAULT}, 1823 {"event", MPS_EVENT}, 1824 {"log", MPS_LOG}, 1825 {"recovery", MPS_RECOVERY}, 1826 {"error", MPS_ERROR}, 1827 {"init", MPS_INIT}, 1828 {"xinfo", MPS_XINFO}, 1829 {"user", MPS_USER}, 1830 {"mapping", MPS_MAPPING}, 1831 {"trace", MPS_TRACE} 1832 }; 1833 1834 enum mps_debug_level_combiner { 1835 COMB_NONE, 1836 COMB_ADD, 1837 COMB_SUB 1838 }; 1839 1840 static int 1841 mps_debug_sysctl(SYSCTL_HANDLER_ARGS) 1842 { 1843 struct mps_softc *sc; 1844 struct mps_debug_string *string; 1845 struct sbuf *sbuf; 1846 char *buffer; 1847 size_t sz; 1848 int i, len, debug, error; 1849 1850 sc = (struct mps_softc *)arg1; 1851 1852 error = sysctl_wire_old_buffer(req, 0); 1853 if (error != 0) 1854 return (error); 1855 1856 sbuf = sbuf_new_for_sysctl(NULL, NULL, 128, req); 1857 debug = sc->mps_debug; 1858 1859 sbuf_printf(sbuf, "%#x", debug); 1860 1861 sz = sizeof(mps_debug_strings) / sizeof(mps_debug_strings[0]); 1862 for (i = 0; i < sz; i++) { 1863 string = &mps_debug_strings[i]; 1864 if (debug & string->flag) 1865 sbuf_printf(sbuf, ",%s", string->name); 1866 } 1867 1868 error = sbuf_finish(sbuf); 1869 sbuf_delete(sbuf); 1870 1871 if (error || req->newptr == NULL) 1872 return (error); 1873 1874 len = req->newlen - req->newidx; 1875 if (len == 0) 1876 return (0); 1877 1878 buffer = malloc(len, M_MPT2, M_ZERO|M_WAITOK); 1879 error = SYSCTL_IN(req, buffer, len); 1880 1881 mps_parse_debug(sc, buffer); 1882 1883 free(buffer, M_MPT2); 1884 return (error); 1885 } 1886 1887 static void 1888 mps_parse_debug(struct mps_softc *sc, char *list) 1889 { 1890 struct mps_debug_string *string; 1891 enum mps_debug_level_combiner op; 1892 char *token, *endtoken; 1893 size_t sz; 1894 int flags, i; 1895 1896 if (list == NULL || *list == '\0') 1897 return; 1898 1899 if (*list == '+') { 1900 op = COMB_ADD; 1901 list++; 1902 } else if (*list == '-') { 1903 op = COMB_SUB; 1904 list++; 1905 } else 1906 op = COMB_NONE; 1907 if (*list == '\0') 1908 return; 1909 1910 flags = 0; 1911 sz = sizeof(mps_debug_strings) / sizeof(mps_debug_strings[0]); 1912 while ((token = strsep(&list, ":,")) != NULL) { 1913 1914 /* Handle integer flags */ 1915 flags |= strtol(token, &endtoken, 0); 1916 if (token != endtoken) 1917 continue; 1918 1919 /* Handle text flags */ 1920 for (i = 0; i < sz; i++) { 1921 string = &mps_debug_strings[i]; 1922 if (strcasecmp(token, string->name) == 0) { 1923 flags |= string->flag; 1924 break; 1925 } 1926 } 1927 } 1928 1929 switch (op) { 1930 case COMB_NONE: 1931 sc->mps_debug = flags; 1932 break; 1933 case COMB_ADD: 1934 sc->mps_debug |= flags; 1935 break; 1936 case COMB_SUB: 1937 sc->mps_debug &= (~flags); 1938 break; 1939 } 1940 1941 return; 1942 } 1943 1944 struct mps_dumpreq_hdr { 1945 uint32_t smid; 1946 uint32_t state; 1947 uint32_t numframes; 1948 uint32_t deschi; 1949 uint32_t desclo; 1950 }; 1951 1952 static int 1953 mps_dump_reqs(SYSCTL_HANDLER_ARGS) 1954 { 1955 struct mps_softc *sc; 1956 struct mps_chain *chain, *chain1; 1957 struct mps_command *cm; 1958 struct mps_dumpreq_hdr hdr; 1959 struct sbuf *sb; 1960 uint32_t smid, state; 1961 int i, numreqs, error = 0; 1962 1963 sc = (struct mps_softc *)arg1; 1964 1965 if ((error = priv_check(curthread, PRIV_DRIVER)) != 0) { 1966 printf("priv check error %d\n", error); 1967 return (error); 1968 } 1969 1970 state = MPS_CM_STATE_INQUEUE; 1971 smid = 1; 1972 numreqs = sc->num_reqs; 1973 1974 if (req->newptr != NULL) 1975 return (EINVAL); 1976 1977 if (smid == 0 || smid > sc->num_reqs) 1978 return (EINVAL); 1979 if (numreqs <= 0 || (numreqs + smid > sc->num_reqs)) 1980 numreqs = sc->num_reqs; 1981 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req); 1982 1983 /* Best effort, no locking */ 1984 for (i = smid; i < numreqs; i++) { 1985 cm = &sc->commands[i]; 1986 if (cm->cm_state != state) 1987 continue; 1988 hdr.smid = i; 1989 hdr.state = cm->cm_state; 1990 hdr.numframes = 1; 1991 hdr.deschi = cm->cm_desc.Words.High; 1992 hdr.desclo = cm->cm_desc.Words.Low; 1993 TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, 1994 chain1) 1995 hdr.numframes++; 1996 sbuf_bcat(sb, &hdr, sizeof(hdr)); 1997 sbuf_bcat(sb, cm->cm_req, 128); 1998 TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, 1999 chain1) 2000 sbuf_bcat(sb, chain->chain, 128); 2001 } 2002 2003 error = sbuf_finish(sb); 2004 sbuf_delete(sb); 2005 return (error); 2006 } 2007 2008 int 2009 mps_attach(struct mps_softc *sc) 2010 { 2011 int error; 2012 2013 MPS_FUNCTRACE(sc); 2014 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 2015 2016 mtx_init(&sc->mps_mtx, "MPT2SAS lock", NULL, MTX_DEF); 2017 callout_init_mtx(&sc->periodic, &sc->mps_mtx, 0); 2018 callout_init_mtx(&sc->device_check_callout, &sc->mps_mtx, 0); 2019 TAILQ_INIT(&sc->event_list); 2020 timevalclear(&sc->lastfail); 2021 2022 if ((error = mps_transition_ready(sc)) != 0) { 2023 mps_dprint(sc, MPS_INIT|MPS_FAULT, "failed to transition " 2024 "ready\n"); 2025 return (error); 2026 } 2027 2028 sc->facts = malloc(sizeof(MPI2_IOC_FACTS_REPLY), M_MPT2, 2029 M_ZERO|M_NOWAIT); 2030 if(!sc->facts) { 2031 mps_dprint(sc, MPS_INIT|MPS_FAULT, "Cannot allocate memory, " 2032 "exit\n"); 2033 return (ENOMEM); 2034 } 2035 2036 /* 2037 * Get IOC Facts and allocate all structures based on this information. 2038 * A Diag Reset will also call mps_iocfacts_allocate and re-read the IOC 2039 * Facts. If relevant values have changed in IOC Facts, this function 2040 * will free all of the memory based on IOC Facts and reallocate that 2041 * memory. If this fails, any allocated memory should already be freed. 2042 */ 2043 if ((error = mps_iocfacts_allocate(sc, TRUE)) != 0) { 2044 mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC Facts based allocation " 2045 "failed with error %d, exit\n", error); 2046 return (error); 2047 } 2048 2049 /* Start the periodic watchdog check on the IOC Doorbell */ 2050 mps_periodic(sc); 2051 2052 /* 2053 * The portenable will kick off discovery events that will drive the 2054 * rest of the initialization process. The CAM/SAS module will 2055 * hold up the boot sequence until discovery is complete. 2056 */ 2057 sc->mps_ich.ich_func = mps_startup; 2058 sc->mps_ich.ich_arg = sc; 2059 if (config_intrhook_establish(&sc->mps_ich) != 0) { 2060 mps_dprint(sc, MPS_INIT|MPS_ERROR, 2061 "Cannot establish MPS config hook\n"); 2062 error = EINVAL; 2063 } 2064 2065 /* 2066 * Allow IR to shutdown gracefully when shutdown occurs. 2067 */ 2068 sc->shutdown_eh = EVENTHANDLER_REGISTER(shutdown_final, 2069 mpssas_ir_shutdown, sc, SHUTDOWN_PRI_DEFAULT); 2070 2071 if (sc->shutdown_eh == NULL) 2072 mps_dprint(sc, MPS_INIT|MPS_ERROR, 2073 "shutdown event registration failed\n"); 2074 2075 mps_setup_sysctl(sc); 2076 2077 sc->mps_flags |= MPS_FLAGS_ATTACH_DONE; 2078 mps_dprint(sc, MPS_INIT, "%s exit error= %d\n", __func__, error); 2079 2080 return (error); 2081 } 2082 2083 /* Run through any late-start handlers. */ 2084 static void 2085 mps_startup(void *arg) 2086 { 2087 struct mps_softc *sc; 2088 2089 sc = (struct mps_softc *)arg; 2090 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 2091 2092 mps_lock(sc); 2093 mps_unmask_intr(sc); 2094 2095 /* initialize device mapping tables */ 2096 mps_base_static_config_pages(sc); 2097 mps_mapping_initialize(sc); 2098 mpssas_startup(sc); 2099 mps_unlock(sc); 2100 2101 mps_dprint(sc, MPS_INIT, "disestablish config intrhook\n"); 2102 config_intrhook_disestablish(&sc->mps_ich); 2103 sc->mps_ich.ich_arg = NULL; 2104 2105 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 2106 } 2107 2108 /* Periodic watchdog. Is called with the driver lock already held. */ 2109 static void 2110 mps_periodic(void *arg) 2111 { 2112 struct mps_softc *sc; 2113 uint32_t db; 2114 2115 sc = (struct mps_softc *)arg; 2116 if (sc->mps_flags & MPS_FLAGS_SHUTDOWN) 2117 return; 2118 2119 db = mps_regread(sc, MPI2_DOORBELL_OFFSET); 2120 if ((db & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) { 2121 mps_dprint(sc, MPS_FAULT, "IOC Fault 0x%08x, Resetting\n", db); 2122 mps_reinit(sc); 2123 } 2124 2125 callout_reset(&sc->periodic, MPS_PERIODIC_DELAY * hz, mps_periodic, sc); 2126 } 2127 2128 static void 2129 mps_log_evt_handler(struct mps_softc *sc, uintptr_t data, 2130 MPI2_EVENT_NOTIFICATION_REPLY *event) 2131 { 2132 MPI2_EVENT_DATA_LOG_ENTRY_ADDED *entry; 2133 2134 MPS_DPRINT_EVENT(sc, generic, event); 2135 2136 switch (event->Event) { 2137 case MPI2_EVENT_LOG_DATA: 2138 mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_DATA:\n"); 2139 if (sc->mps_debug & MPS_EVENT) 2140 hexdump(event->EventData, event->EventDataLength, NULL, 0); 2141 break; 2142 case MPI2_EVENT_LOG_ENTRY_ADDED: 2143 entry = (MPI2_EVENT_DATA_LOG_ENTRY_ADDED *)event->EventData; 2144 mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_ENTRY_ADDED event " 2145 "0x%x Sequence %d:\n", entry->LogEntryQualifier, 2146 entry->LogSequence); 2147 break; 2148 default: 2149 break; 2150 } 2151 return; 2152 } 2153 2154 static int 2155 mps_attach_log(struct mps_softc *sc) 2156 { 2157 u32 events[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS]; 2158 2159 bzero(events, 16); 2160 setbit(events, MPI2_EVENT_LOG_DATA); 2161 setbit(events, MPI2_EVENT_LOG_ENTRY_ADDED); 2162 2163 mps_register_events(sc, events, mps_log_evt_handler, NULL, 2164 &sc->mps_log_eh); 2165 2166 return (0); 2167 } 2168 2169 static int 2170 mps_detach_log(struct mps_softc *sc) 2171 { 2172 2173 if (sc->mps_log_eh != NULL) 2174 mps_deregister_events(sc, sc->mps_log_eh); 2175 return (0); 2176 } 2177 2178 /* 2179 * Free all of the driver resources and detach submodules. Should be called 2180 * without the lock held. 2181 */ 2182 int 2183 mps_free(struct mps_softc *sc) 2184 { 2185 int error; 2186 2187 mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); 2188 /* Turn off the watchdog */ 2189 mps_lock(sc); 2190 sc->mps_flags |= MPS_FLAGS_SHUTDOWN; 2191 mps_unlock(sc); 2192 /* Lock must not be held for this */ 2193 callout_drain(&sc->periodic); 2194 callout_drain(&sc->device_check_callout); 2195 2196 if (((error = mps_detach_log(sc)) != 0) || 2197 ((error = mps_detach_sas(sc)) != 0)) { 2198 mps_dprint(sc, MPS_INIT|MPS_FAULT, "failed to detach " 2199 "subsystems, exit\n"); 2200 return (error); 2201 } 2202 2203 mps_detach_user(sc); 2204 2205 /* Put the IOC back in the READY state. */ 2206 mps_lock(sc); 2207 if ((error = mps_transition_ready(sc)) != 0) { 2208 mps_unlock(sc); 2209 return (error); 2210 } 2211 mps_unlock(sc); 2212 2213 if (sc->facts != NULL) 2214 free(sc->facts, M_MPT2); 2215 2216 /* 2217 * Free all buffers that are based on IOC Facts. A Diag Reset may need 2218 * to free these buffers too. 2219 */ 2220 mps_iocfacts_free(sc); 2221 2222 if (sc->sysctl_tree != NULL) 2223 sysctl_ctx_free(&sc->sysctl_ctx); 2224 2225 /* Deregister the shutdown function */ 2226 if (sc->shutdown_eh != NULL) 2227 EVENTHANDLER_DEREGISTER(shutdown_final, sc->shutdown_eh); 2228 2229 mtx_destroy(&sc->mps_mtx); 2230 mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); 2231 2232 return (0); 2233 } 2234 2235 static __inline void 2236 mps_complete_command(struct mps_softc *sc, struct mps_command *cm) 2237 { 2238 MPS_FUNCTRACE(sc); 2239 2240 if (cm == NULL) { 2241 mps_dprint(sc, MPS_ERROR, "Completing NULL command\n"); 2242 return; 2243 } 2244 2245 if (cm->cm_flags & MPS_CM_FLAGS_POLLED) 2246 cm->cm_flags |= MPS_CM_FLAGS_COMPLETE; 2247 2248 if (cm->cm_complete != NULL) { 2249 mps_dprint(sc, MPS_TRACE, 2250 "%s cm %p calling cm_complete %p data %p reply %p\n", 2251 __func__, cm, cm->cm_complete, cm->cm_complete_data, 2252 cm->cm_reply); 2253 cm->cm_complete(sc, cm); 2254 } 2255 2256 if (cm->cm_flags & MPS_CM_FLAGS_WAKEUP) { 2257 mps_dprint(sc, MPS_TRACE, "waking up %p\n", cm); 2258 wakeup(cm); 2259 } 2260 2261 if (cm->cm_sc->io_cmds_active != 0) { 2262 cm->cm_sc->io_cmds_active--; 2263 } else { 2264 mps_dprint(sc, MPS_ERROR, "Warning: io_cmds_active is " 2265 "out of sync - resynching to 0\n"); 2266 } 2267 } 2268 2269 2270 static void 2271 mps_sas_log_info(struct mps_softc *sc , u32 log_info) 2272 { 2273 union loginfo_type { 2274 u32 loginfo; 2275 struct { 2276 u32 subcode:16; 2277 u32 code:8; 2278 u32 originator:4; 2279 u32 bus_type:4; 2280 } dw; 2281 }; 2282 union loginfo_type sas_loginfo; 2283 char *originator_str = NULL; 2284 2285 sas_loginfo.loginfo = log_info; 2286 if (sas_loginfo.dw.bus_type != 3 /*SAS*/) 2287 return; 2288 2289 /* each nexus loss loginfo */ 2290 if (log_info == 0x31170000) 2291 return; 2292 2293 /* eat the loginfos associated with task aborts */ 2294 if ((log_info == 30050000 || log_info == 2295 0x31140000 || log_info == 0x31130000)) 2296 return; 2297 2298 switch (sas_loginfo.dw.originator) { 2299 case 0: 2300 originator_str = "IOP"; 2301 break; 2302 case 1: 2303 originator_str = "PL"; 2304 break; 2305 case 2: 2306 originator_str = "IR"; 2307 break; 2308 } 2309 2310 mps_dprint(sc, MPS_LOG, "log_info(0x%08x): originator(%s), " 2311 "code(0x%02x), sub_code(0x%04x)\n", log_info, 2312 originator_str, sas_loginfo.dw.code, 2313 sas_loginfo.dw.subcode); 2314 } 2315 2316 static void 2317 mps_display_reply_info(struct mps_softc *sc, uint8_t *reply) 2318 { 2319 MPI2DefaultReply_t *mpi_reply; 2320 u16 sc_status; 2321 2322 mpi_reply = (MPI2DefaultReply_t*)reply; 2323 sc_status = le16toh(mpi_reply->IOCStatus); 2324 if (sc_status & MPI2_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE) 2325 mps_sas_log_info(sc, le32toh(mpi_reply->IOCLogInfo)); 2326 } 2327 void 2328 mps_intr(void *data) 2329 { 2330 struct mps_softc *sc; 2331 uint32_t status; 2332 2333 sc = (struct mps_softc *)data; 2334 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 2335 2336 /* 2337 * Check interrupt status register to flush the bus. This is 2338 * needed for both INTx interrupts and driver-driven polling 2339 */ 2340 status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); 2341 if ((status & MPI2_HIS_REPLY_DESCRIPTOR_INTERRUPT) == 0) 2342 return; 2343 2344 mps_lock(sc); 2345 mps_intr_locked(data); 2346 mps_unlock(sc); 2347 return; 2348 } 2349 2350 /* 2351 * In theory, MSI/MSIX interrupts shouldn't need to read any registers on the 2352 * chip. Hopefully this theory is correct. 2353 */ 2354 void 2355 mps_intr_msi(void *data) 2356 { 2357 struct mps_softc *sc; 2358 2359 sc = (struct mps_softc *)data; 2360 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 2361 mps_lock(sc); 2362 mps_intr_locked(data); 2363 mps_unlock(sc); 2364 return; 2365 } 2366 2367 /* 2368 * The locking is overly broad and simplistic, but easy to deal with for now. 2369 */ 2370 void 2371 mps_intr_locked(void *data) 2372 { 2373 MPI2_REPLY_DESCRIPTORS_UNION *desc; 2374 MPI2_DIAG_RELEASE_REPLY *rel_rep; 2375 mps_fw_diagnostic_buffer_t *pBuffer; 2376 struct mps_softc *sc; 2377 struct mps_command *cm = NULL; 2378 uint64_t tdesc; 2379 uint8_t flags; 2380 u_int pq; 2381 2382 sc = (struct mps_softc *)data; 2383 2384 pq = sc->replypostindex; 2385 mps_dprint(sc, MPS_TRACE, 2386 "%s sc %p starting with replypostindex %u\n", 2387 __func__, sc, sc->replypostindex); 2388 2389 for ( ;; ) { 2390 cm = NULL; 2391 desc = &sc->post_queue[sc->replypostindex]; 2392 2393 /* 2394 * Copy and clear out the descriptor so that any reentry will 2395 * immediately know that this descriptor has already been 2396 * looked at. There is unfortunate casting magic because the 2397 * MPI API doesn't have a cardinal 64bit type. 2398 */ 2399 tdesc = 0xffffffffffffffff; 2400 tdesc = atomic_swap_64((uint64_t *)desc, tdesc); 2401 desc = (MPI2_REPLY_DESCRIPTORS_UNION *)&tdesc; 2402 2403 flags = desc->Default.ReplyFlags & 2404 MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK; 2405 if ((flags == MPI2_RPY_DESCRIPT_FLAGS_UNUSED) 2406 || (le32toh(desc->Words.High) == 0xffffffff)) 2407 break; 2408 2409 /* increment the replypostindex now, so that event handlers 2410 * and cm completion handlers which decide to do a diag 2411 * reset can zero it without it getting incremented again 2412 * afterwards, and we break out of this loop on the next 2413 * iteration since the reply post queue has been cleared to 2414 * 0xFF and all descriptors look unused (which they are). 2415 */ 2416 if (++sc->replypostindex >= sc->pqdepth) 2417 sc->replypostindex = 0; 2418 2419 switch (flags) { 2420 case MPI2_RPY_DESCRIPT_FLAGS_SCSI_IO_SUCCESS: 2421 cm = &sc->commands[le16toh(desc->SCSIIOSuccess.SMID)]; 2422 KASSERT(cm->cm_state == MPS_CM_STATE_INQUEUE, 2423 ("command not inqueue\n")); 2424 cm->cm_state = MPS_CM_STATE_BUSY; 2425 cm->cm_reply = NULL; 2426 break; 2427 case MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY: 2428 { 2429 uint32_t baddr; 2430 uint8_t *reply; 2431 2432 /* 2433 * Re-compose the reply address from the address 2434 * sent back from the chip. The ReplyFrameAddress 2435 * is the lower 32 bits of the physical address of 2436 * particular reply frame. Convert that address to 2437 * host format, and then use that to provide the 2438 * offset against the virtual address base 2439 * (sc->reply_frames). 2440 */ 2441 baddr = le32toh(desc->AddressReply.ReplyFrameAddress); 2442 reply = sc->reply_frames + 2443 (baddr - ((uint32_t)sc->reply_busaddr)); 2444 /* 2445 * Make sure the reply we got back is in a valid 2446 * range. If not, go ahead and panic here, since 2447 * we'll probably panic as soon as we deference the 2448 * reply pointer anyway. 2449 */ 2450 if ((reply < sc->reply_frames) 2451 || (reply > (sc->reply_frames + 2452 (sc->fqdepth * sc->replyframesz)))) { 2453 printf("%s: WARNING: reply %p out of range!\n", 2454 __func__, reply); 2455 printf("%s: reply_frames %p, fqdepth %d, " 2456 "frame size %d\n", __func__, 2457 sc->reply_frames, sc->fqdepth, 2458 sc->replyframesz); 2459 printf("%s: baddr %#x,\n", __func__, baddr); 2460 /* LSI-TODO. See Linux Code for Graceful exit */ 2461 panic("Reply address out of range"); 2462 } 2463 if (le16toh(desc->AddressReply.SMID) == 0) { 2464 if (((MPI2_DEFAULT_REPLY *)reply)->Function == 2465 MPI2_FUNCTION_DIAG_BUFFER_POST) { 2466 /* 2467 * If SMID is 0 for Diag Buffer Post, 2468 * this implies that the reply is due to 2469 * a release function with a status that 2470 * the buffer has been released. Set 2471 * the buffer flags accordingly. 2472 */ 2473 rel_rep = 2474 (MPI2_DIAG_RELEASE_REPLY *)reply; 2475 if ((le16toh(rel_rep->IOCStatus) & 2476 MPI2_IOCSTATUS_MASK) == 2477 MPI2_IOCSTATUS_DIAGNOSTIC_RELEASED) 2478 { 2479 pBuffer = 2480 &sc->fw_diag_buffer_list[ 2481 rel_rep->BufferType]; 2482 pBuffer->valid_data = TRUE; 2483 pBuffer->owned_by_firmware = 2484 FALSE; 2485 pBuffer->immediate = FALSE; 2486 } 2487 } else 2488 mps_dispatch_event(sc, baddr, 2489 (MPI2_EVENT_NOTIFICATION_REPLY *) 2490 reply); 2491 } else { 2492 /* 2493 * Ignore commands not in INQUEUE state 2494 * since they've already been completed 2495 * via another path. 2496 */ 2497 cm = &sc->commands[ 2498 le16toh(desc->AddressReply.SMID)]; 2499 if (cm->cm_state == MPS_CM_STATE_INQUEUE) { 2500 cm->cm_state = MPS_CM_STATE_BUSY; 2501 cm->cm_reply = reply; 2502 cm->cm_reply_data = le32toh( 2503 desc->AddressReply.ReplyFrameAddress); 2504 } else { 2505 mps_dprint(sc, MPS_RECOVERY, 2506 "Bad state for ADDRESS_REPLY status," 2507 " ignoring state %d cm %p\n", 2508 cm->cm_state, cm); 2509 } 2510 } 2511 break; 2512 } 2513 case MPI2_RPY_DESCRIPT_FLAGS_TARGETASSIST_SUCCESS: 2514 case MPI2_RPY_DESCRIPT_FLAGS_TARGET_COMMAND_BUFFER: 2515 case MPI2_RPY_DESCRIPT_FLAGS_RAID_ACCELERATOR_SUCCESS: 2516 default: 2517 /* Unhandled */ 2518 mps_dprint(sc, MPS_ERROR, "Unhandled reply 0x%x\n", 2519 desc->Default.ReplyFlags); 2520 cm = NULL; 2521 break; 2522 } 2523 2524 2525 if (cm != NULL) { 2526 // Print Error reply frame 2527 if (cm->cm_reply) 2528 mps_display_reply_info(sc,cm->cm_reply); 2529 mps_complete_command(sc, cm); 2530 } 2531 } 2532 2533 if (pq != sc->replypostindex) { 2534 mps_dprint(sc, MPS_TRACE, "%s sc %p writing postindex %d\n", 2535 __func__, sc, sc->replypostindex); 2536 mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, 2537 sc->replypostindex); 2538 } 2539 2540 return; 2541 } 2542 2543 static void 2544 mps_dispatch_event(struct mps_softc *sc, uintptr_t data, 2545 MPI2_EVENT_NOTIFICATION_REPLY *reply) 2546 { 2547 struct mps_event_handle *eh; 2548 int event, handled = 0; 2549 2550 event = le16toh(reply->Event); 2551 TAILQ_FOREACH(eh, &sc->event_list, eh_list) { 2552 if (isset(eh->mask, event)) { 2553 eh->callback(sc, data, reply); 2554 handled++; 2555 } 2556 } 2557 2558 if (handled == 0) 2559 mps_dprint(sc, MPS_EVENT, "Unhandled event 0x%x\n", le16toh(event)); 2560 2561 /* 2562 * This is the only place that the event/reply should be freed. 2563 * Anything wanting to hold onto the event data should have 2564 * already copied it into their own storage. 2565 */ 2566 mps_free_reply(sc, data); 2567 } 2568 2569 static void 2570 mps_reregister_events_complete(struct mps_softc *sc, struct mps_command *cm) 2571 { 2572 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 2573 2574 if (cm->cm_reply) 2575 MPS_DPRINT_EVENT(sc, generic, 2576 (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply); 2577 2578 mps_free_command(sc, cm); 2579 2580 /* next, send a port enable */ 2581 mpssas_startup(sc); 2582 } 2583 2584 /* 2585 * For both register_events and update_events, the caller supplies a bitmap 2586 * of events that it _wants_. These functions then turn that into a bitmask 2587 * suitable for the controller. 2588 */ 2589 int 2590 mps_register_events(struct mps_softc *sc, u32 *mask, 2591 mps_evt_callback_t *cb, void *data, struct mps_event_handle **handle) 2592 { 2593 struct mps_event_handle *eh; 2594 int error = 0; 2595 2596 eh = malloc(sizeof(struct mps_event_handle), M_MPT2, M_WAITOK|M_ZERO); 2597 if(!eh) { 2598 mps_dprint(sc, MPS_ERROR, "Cannot allocate event memory\n"); 2599 return (ENOMEM); 2600 } 2601 eh->callback = cb; 2602 eh->data = data; 2603 TAILQ_INSERT_TAIL(&sc->event_list, eh, eh_list); 2604 if (mask != NULL) 2605 error = mps_update_events(sc, eh, mask); 2606 *handle = eh; 2607 2608 return (error); 2609 } 2610 2611 int 2612 mps_update_events(struct mps_softc *sc, struct mps_event_handle *handle, 2613 u32 *mask) 2614 { 2615 MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; 2616 MPI2_EVENT_NOTIFICATION_REPLY *reply = NULL; 2617 struct mps_command *cm; 2618 int error, i; 2619 2620 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 2621 2622 if ((mask != NULL) && (handle != NULL)) 2623 bcopy(mask, &handle->mask[0], sizeof(u32) * 2624 MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); 2625 2626 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2627 sc->event_mask[i] = -1; 2628 2629 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2630 sc->event_mask[i] &= ~handle->mask[i]; 2631 2632 2633 if ((cm = mps_alloc_command(sc)) == NULL) 2634 return (EBUSY); 2635 evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; 2636 evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; 2637 evtreq->MsgFlags = 0; 2638 evtreq->SASBroadcastPrimitiveMasks = 0; 2639 #ifdef MPS_DEBUG_ALL_EVENTS 2640 { 2641 u_char fullmask[16]; 2642 memset(fullmask, 0x00, 16); 2643 bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) * 2644 MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); 2645 } 2646 #else 2647 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2648 evtreq->EventMasks[i] = 2649 htole32(sc->event_mask[i]); 2650 #endif 2651 cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; 2652 cm->cm_data = NULL; 2653 2654 error = mps_wait_command(sc, &cm, 60, 0); 2655 if (cm != NULL) 2656 reply = (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply; 2657 if ((reply == NULL) || 2658 (reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) 2659 error = ENXIO; 2660 2661 if (reply) 2662 MPS_DPRINT_EVENT(sc, generic, reply); 2663 2664 mps_dprint(sc, MPS_TRACE, "%s finished error %d\n", __func__, error); 2665 2666 if (cm != NULL) 2667 mps_free_command(sc, cm); 2668 return (error); 2669 } 2670 2671 static int 2672 mps_reregister_events(struct mps_softc *sc) 2673 { 2674 MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; 2675 struct mps_command *cm; 2676 struct mps_event_handle *eh; 2677 int error, i; 2678 2679 mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 2680 2681 /* first, reregister events */ 2682 2683 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2684 sc->event_mask[i] = -1; 2685 2686 TAILQ_FOREACH(eh, &sc->event_list, eh_list) { 2687 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2688 sc->event_mask[i] &= ~eh->mask[i]; 2689 } 2690 2691 if ((cm = mps_alloc_command(sc)) == NULL) 2692 return (EBUSY); 2693 evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; 2694 evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; 2695 evtreq->MsgFlags = 0; 2696 evtreq->SASBroadcastPrimitiveMasks = 0; 2697 #ifdef MPS_DEBUG_ALL_EVENTS 2698 { 2699 u_char fullmask[16]; 2700 memset(fullmask, 0x00, 16); 2701 bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) * 2702 MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); 2703 } 2704 #else 2705 for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) 2706 evtreq->EventMasks[i] = 2707 htole32(sc->event_mask[i]); 2708 #endif 2709 cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; 2710 cm->cm_data = NULL; 2711 cm->cm_complete = mps_reregister_events_complete; 2712 2713 error = mps_map_command(sc, cm); 2714 2715 mps_dprint(sc, MPS_TRACE, "%s finished with error %d\n", __func__, 2716 error); 2717 return (error); 2718 } 2719 2720 void 2721 mps_deregister_events(struct mps_softc *sc, struct mps_event_handle *handle) 2722 { 2723 2724 TAILQ_REMOVE(&sc->event_list, handle, eh_list); 2725 free(handle, M_MPT2); 2726 } 2727 2728 /* 2729 * Add a chain element as the next SGE for the specified command. 2730 * Reset cm_sge and cm_sgesize to indicate all the available space. 2731 */ 2732 static int 2733 mps_add_chain(struct mps_command *cm) 2734 { 2735 MPI2_SGE_CHAIN32 *sgc; 2736 struct mps_chain *chain; 2737 u_int space; 2738 2739 if (cm->cm_sglsize < MPS_SGC_SIZE) 2740 panic("MPS: Need SGE Error Code\n"); 2741 2742 chain = mps_alloc_chain(cm->cm_sc); 2743 if (chain == NULL) 2744 return (ENOBUFS); 2745 2746 space = cm->cm_sc->reqframesz; 2747 2748 /* 2749 * Note: a double-linked list is used to make it easier to 2750 * walk for debugging. 2751 */ 2752 TAILQ_INSERT_TAIL(&cm->cm_chain_list, chain, chain_link); 2753 2754 sgc = (MPI2_SGE_CHAIN32 *)&cm->cm_sge->MpiChain; 2755 sgc->Length = htole16(space); 2756 sgc->NextChainOffset = 0; 2757 /* TODO Looks like bug in Setting sgc->Flags. 2758 * sgc->Flags = ( MPI2_SGE_FLAGS_CHAIN_ELEMENT | MPI2_SGE_FLAGS_64_BIT_ADDRESSING | 2759 * MPI2_SGE_FLAGS_SYSTEM_ADDRESS) << MPI2_SGE_FLAGS_SHIFT 2760 * This is fine.. because we are not using simple element. In case of 2761 * MPI2_SGE_CHAIN32, we have separate Length and Flags feild. 2762 */ 2763 sgc->Flags = MPI2_SGE_FLAGS_CHAIN_ELEMENT; 2764 sgc->Address = htole32(chain->chain_busaddr); 2765 2766 cm->cm_sge = (MPI2_SGE_IO_UNION *)&chain->chain->MpiSimple; 2767 cm->cm_sglsize = space; 2768 return (0); 2769 } 2770 2771 /* 2772 * Add one scatter-gather element (chain, simple, transaction context) 2773 * to the scatter-gather list for a command. Maintain cm_sglsize and 2774 * cm_sge as the remaining size and pointer to the next SGE to fill 2775 * in, respectively. 2776 */ 2777 int 2778 mps_push_sge(struct mps_command *cm, void *sgep, size_t len, int segsleft) 2779 { 2780 MPI2_SGE_TRANSACTION_UNION *tc = sgep; 2781 MPI2_SGE_SIMPLE64 *sge = sgep; 2782 int error, type; 2783 uint32_t saved_buf_len, saved_address_low, saved_address_high; 2784 2785 type = (tc->Flags & MPI2_SGE_FLAGS_ELEMENT_MASK); 2786 2787 #ifdef INVARIANTS 2788 switch (type) { 2789 case MPI2_SGE_FLAGS_TRANSACTION_ELEMENT: { 2790 if (len != tc->DetailsLength + 4) 2791 panic("TC %p length %u or %zu?", tc, 2792 tc->DetailsLength + 4, len); 2793 } 2794 break; 2795 case MPI2_SGE_FLAGS_CHAIN_ELEMENT: 2796 /* Driver only uses 32-bit chain elements */ 2797 if (len != MPS_SGC_SIZE) 2798 panic("CHAIN %p length %u or %zu?", sgep, 2799 MPS_SGC_SIZE, len); 2800 break; 2801 case MPI2_SGE_FLAGS_SIMPLE_ELEMENT: 2802 /* Driver only uses 64-bit SGE simple elements */ 2803 if (len != MPS_SGE64_SIZE) 2804 panic("SGE simple %p length %u or %zu?", sge, 2805 MPS_SGE64_SIZE, len); 2806 if (((le32toh(sge->FlagsLength) >> MPI2_SGE_FLAGS_SHIFT) & 2807 MPI2_SGE_FLAGS_ADDRESS_SIZE) == 0) 2808 panic("SGE simple %p not marked 64-bit?", sge); 2809 2810 break; 2811 default: 2812 panic("Unexpected SGE %p, flags %02x", tc, tc->Flags); 2813 } 2814 #endif 2815 2816 /* 2817 * case 1: 1 more segment, enough room for it 2818 * case 2: 2 more segments, enough room for both 2819 * case 3: >=2 more segments, only enough room for 1 and a chain 2820 * case 4: >=1 more segment, enough room for only a chain 2821 * case 5: >=1 more segment, no room for anything (error) 2822 */ 2823 2824 /* 2825 * There should be room for at least a chain element, or this 2826 * code is buggy. Case (5). 2827 */ 2828 if (cm->cm_sglsize < MPS_SGC_SIZE) 2829 panic("MPS: Need SGE Error Code\n"); 2830 2831 if (segsleft >= 1 && cm->cm_sglsize < len + MPS_SGC_SIZE) { 2832 /* 2833 * 1 or more segment, enough room for only a chain. 2834 * Hope the previous element wasn't a Simple entry 2835 * that needed to be marked with 2836 * MPI2_SGE_FLAGS_LAST_ELEMENT. Case (4). 2837 */ 2838 if ((error = mps_add_chain(cm)) != 0) 2839 return (error); 2840 } 2841 2842 if (segsleft >= 2 && 2843 cm->cm_sglsize < len + MPS_SGC_SIZE + MPS_SGE64_SIZE) { 2844 /* 2845 * There are 2 or more segments left to add, and only 2846 * enough room for 1 and a chain. Case (3). 2847 * 2848 * Mark as last element in this chain if necessary. 2849 */ 2850 if (type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) { 2851 sge->FlagsLength |= htole32( 2852 MPI2_SGE_FLAGS_LAST_ELEMENT << MPI2_SGE_FLAGS_SHIFT); 2853 } 2854 2855 /* 2856 * Add the item then a chain. Do the chain now, 2857 * rather than on the next iteration, to simplify 2858 * understanding the code. 2859 */ 2860 cm->cm_sglsize -= len; 2861 bcopy(sgep, cm->cm_sge, len); 2862 cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); 2863 return (mps_add_chain(cm)); 2864 } 2865 2866 #ifdef INVARIANTS 2867 /* Case 1: 1 more segment, enough room for it. */ 2868 if (segsleft == 1 && cm->cm_sglsize < len) 2869 panic("1 seg left and no room? %u versus %zu", 2870 cm->cm_sglsize, len); 2871 2872 /* Case 2: 2 more segments, enough room for both */ 2873 if (segsleft == 2 && cm->cm_sglsize < len + MPS_SGE64_SIZE) 2874 panic("2 segs left and no room? %u versus %zu", 2875 cm->cm_sglsize, len); 2876 #endif 2877 2878 if (segsleft == 1 && type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) { 2879 /* 2880 * If this is a bi-directional request, need to account for that 2881 * here. Save the pre-filled sge values. These will be used 2882 * either for the 2nd SGL or for a single direction SGL. If 2883 * cm_out_len is non-zero, this is a bi-directional request, so 2884 * fill in the OUT SGL first, then the IN SGL, otherwise just 2885 * fill in the IN SGL. Note that at this time, when filling in 2886 * 2 SGL's for a bi-directional request, they both use the same 2887 * DMA buffer (same cm command). 2888 */ 2889 saved_buf_len = le32toh(sge->FlagsLength) & 0x00FFFFFF; 2890 saved_address_low = sge->Address.Low; 2891 saved_address_high = sge->Address.High; 2892 if (cm->cm_out_len) { 2893 sge->FlagsLength = htole32(cm->cm_out_len | 2894 ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | 2895 MPI2_SGE_FLAGS_END_OF_BUFFER | 2896 MPI2_SGE_FLAGS_HOST_TO_IOC | 2897 MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << 2898 MPI2_SGE_FLAGS_SHIFT)); 2899 cm->cm_sglsize -= len; 2900 bcopy(sgep, cm->cm_sge, len); 2901 cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge 2902 + len); 2903 } 2904 saved_buf_len |= 2905 ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | 2906 MPI2_SGE_FLAGS_END_OF_BUFFER | 2907 MPI2_SGE_FLAGS_LAST_ELEMENT | 2908 MPI2_SGE_FLAGS_END_OF_LIST | 2909 MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << 2910 MPI2_SGE_FLAGS_SHIFT); 2911 if (cm->cm_flags & MPS_CM_FLAGS_DATAIN) { 2912 saved_buf_len |= 2913 ((uint32_t)(MPI2_SGE_FLAGS_IOC_TO_HOST) << 2914 MPI2_SGE_FLAGS_SHIFT); 2915 } else { 2916 saved_buf_len |= 2917 ((uint32_t)(MPI2_SGE_FLAGS_HOST_TO_IOC) << 2918 MPI2_SGE_FLAGS_SHIFT); 2919 } 2920 sge->FlagsLength = htole32(saved_buf_len); 2921 sge->Address.Low = saved_address_low; 2922 sge->Address.High = saved_address_high; 2923 } 2924 2925 cm->cm_sglsize -= len; 2926 bcopy(sgep, cm->cm_sge, len); 2927 cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); 2928 return (0); 2929 } 2930 2931 /* 2932 * Add one dma segment to the scatter-gather list for a command. 2933 */ 2934 int 2935 mps_add_dmaseg(struct mps_command *cm, vm_paddr_t pa, size_t len, u_int flags, 2936 int segsleft) 2937 { 2938 MPI2_SGE_SIMPLE64 sge; 2939 2940 /* 2941 * This driver always uses 64-bit address elements for simplicity. 2942 */ 2943 bzero(&sge, sizeof(sge)); 2944 flags |= MPI2_SGE_FLAGS_SIMPLE_ELEMENT | 2945 MPI2_SGE_FLAGS_64_BIT_ADDRESSING; 2946 sge.FlagsLength = htole32(len | (flags << MPI2_SGE_FLAGS_SHIFT)); 2947 mps_from_u64(pa, &sge.Address); 2948 2949 return (mps_push_sge(cm, &sge, sizeof sge, segsleft)); 2950 } 2951 2952 static void 2953 mps_data_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) 2954 { 2955 struct mps_softc *sc; 2956 struct mps_command *cm; 2957 u_int i, dir, sflags; 2958 2959 cm = (struct mps_command *)arg; 2960 sc = cm->cm_sc; 2961 2962 /* 2963 * In this case, just print out a warning and let the chip tell the 2964 * user they did the wrong thing. 2965 */ 2966 if ((cm->cm_max_segs != 0) && (nsegs > cm->cm_max_segs)) { 2967 mps_dprint(sc, MPS_ERROR, 2968 "%s: warning: busdma returned %d segments, " 2969 "more than the %d allowed\n", __func__, nsegs, 2970 cm->cm_max_segs); 2971 } 2972 2973 /* 2974 * Set up DMA direction flags. Bi-directional requests are also handled 2975 * here. In that case, both direction flags will be set. 2976 */ 2977 sflags = 0; 2978 if (cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) { 2979 /* 2980 * We have to add a special case for SMP passthrough, there 2981 * is no easy way to generically handle it. The first 2982 * S/G element is used for the command (therefore the 2983 * direction bit needs to be set). The second one is used 2984 * for the reply. We'll leave it to the caller to make 2985 * sure we only have two buffers. 2986 */ 2987 /* 2988 * Even though the busdma man page says it doesn't make 2989 * sense to have both direction flags, it does in this case. 2990 * We have one s/g element being accessed in each direction. 2991 */ 2992 dir = BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD; 2993 2994 /* 2995 * Set the direction flag on the first buffer in the SMP 2996 * passthrough request. We'll clear it for the second one. 2997 */ 2998 sflags |= MPI2_SGE_FLAGS_DIRECTION | 2999 MPI2_SGE_FLAGS_END_OF_BUFFER; 3000 } else if (cm->cm_flags & MPS_CM_FLAGS_DATAOUT) { 3001 sflags |= MPI2_SGE_FLAGS_HOST_TO_IOC; 3002 dir = BUS_DMASYNC_PREWRITE; 3003 } else 3004 dir = BUS_DMASYNC_PREREAD; 3005 3006 for (i = 0; i < nsegs; i++) { 3007 if ((cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) && (i != 0)) { 3008 sflags &= ~MPI2_SGE_FLAGS_DIRECTION; 3009 } 3010 error = mps_add_dmaseg(cm, segs[i].ds_addr, segs[i].ds_len, 3011 sflags, nsegs - i); 3012 if (error != 0) { 3013 /* Resource shortage, roll back! */ 3014 if (ratecheck(&sc->lastfail, &mps_chainfail_interval)) 3015 mps_dprint(sc, MPS_INFO, "Out of chain frames, " 3016 "consider increasing hw.mps.max_chains.\n"); 3017 cm->cm_flags |= MPS_CM_FLAGS_CHAIN_FAILED; 3018 mps_complete_command(sc, cm); 3019 return; 3020 } 3021 } 3022 3023 bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir); 3024 mps_enqueue_request(sc, cm); 3025 3026 return; 3027 } 3028 3029 static void 3030 mps_data_cb2(void *arg, bus_dma_segment_t *segs, int nsegs, bus_size_t mapsize, 3031 int error) 3032 { 3033 mps_data_cb(arg, segs, nsegs, error); 3034 } 3035 3036 /* 3037 * This is the routine to enqueue commands ansynchronously. 3038 * Note that the only error path here is from bus_dmamap_load(), which can 3039 * return EINPROGRESS if it is waiting for resources. Other than this, it's 3040 * assumed that if you have a command in-hand, then you have enough credits 3041 * to use it. 3042 */ 3043 int 3044 mps_map_command(struct mps_softc *sc, struct mps_command *cm) 3045 { 3046 int error = 0; 3047 3048 if (cm->cm_flags & MPS_CM_FLAGS_USE_UIO) { 3049 error = bus_dmamap_load_uio(sc->buffer_dmat, cm->cm_dmamap, 3050 &cm->cm_uio, mps_data_cb2, cm, 0); 3051 } else if (cm->cm_flags & MPS_CM_FLAGS_USE_CCB) { 3052 error = bus_dmamap_load_ccb(sc->buffer_dmat, cm->cm_dmamap, 3053 cm->cm_data, mps_data_cb, cm, 0); 3054 } else if ((cm->cm_data != NULL) && (cm->cm_length != 0)) { 3055 error = bus_dmamap_load(sc->buffer_dmat, cm->cm_dmamap, 3056 cm->cm_data, cm->cm_length, mps_data_cb, cm, 0); 3057 } else { 3058 /* Add a zero-length element as needed */ 3059 if (cm->cm_sge != NULL) 3060 mps_add_dmaseg(cm, 0, 0, 0, 1); 3061 mps_enqueue_request(sc, cm); 3062 } 3063 3064 return (error); 3065 } 3066 3067 /* 3068 * This is the routine to enqueue commands synchronously. An error of 3069 * EINPROGRESS from mps_map_command() is ignored since the command will 3070 * be executed and enqueued automatically. Other errors come from msleep(). 3071 */ 3072 int 3073 mps_wait_command(struct mps_softc *sc, struct mps_command **cmp, int timeout, 3074 int sleep_flag) 3075 { 3076 int error, rc; 3077 struct timeval cur_time, start_time; 3078 struct mps_command *cm = *cmp; 3079 3080 if (sc->mps_flags & MPS_FLAGS_DIAGRESET) 3081 return EBUSY; 3082 3083 cm->cm_complete = NULL; 3084 cm->cm_flags |= MPS_CM_FLAGS_POLLED; 3085 error = mps_map_command(sc, cm); 3086 if ((error != 0) && (error != EINPROGRESS)) 3087 return (error); 3088 3089 /* 3090 * Check for context and wait for 50 mSec at a time until time has 3091 * expired or the command has finished. If msleep can't be used, need 3092 * to poll. 3093 */ 3094 if (curthread->td_no_sleeping != 0) 3095 sleep_flag = NO_SLEEP; 3096 getmicrouptime(&start_time); 3097 if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) { 3098 cm->cm_flags |= MPS_CM_FLAGS_WAKEUP; 3099 error = msleep(cm, &sc->mps_mtx, 0, "mpswait", timeout*hz); 3100 if (error == EWOULDBLOCK) { 3101 /* 3102 * Record the actual elapsed time in the case of a 3103 * timeout for the message below. 3104 */ 3105 getmicrouptime(&cur_time); 3106 timevalsub(&cur_time, &start_time); 3107 } 3108 } else { 3109 while ((cm->cm_flags & MPS_CM_FLAGS_COMPLETE) == 0) { 3110 mps_intr_locked(sc); 3111 if (sleep_flag == CAN_SLEEP) 3112 pause("mpswait", hz/20); 3113 else 3114 DELAY(50000); 3115 3116 getmicrouptime(&cur_time); 3117 timevalsub(&cur_time, &start_time); 3118 if (cur_time.tv_sec > timeout) { 3119 error = EWOULDBLOCK; 3120 break; 3121 } 3122 } 3123 } 3124 3125 if (error == EWOULDBLOCK) { 3126 if (cm->cm_timeout_handler == NULL) { 3127 mps_dprint(sc, MPS_FAULT, "Calling Reinit from %s, timeout=%d," 3128 " elapsed=%jd\n", __func__, timeout, 3129 (intmax_t)cur_time.tv_sec); 3130 rc = mps_reinit(sc); 3131 mps_dprint(sc, MPS_FAULT, "Reinit %s\n", (rc == 0) ? "success" : 3132 "failed"); 3133 } else 3134 cm->cm_timeout_handler(sc, cm); 3135 if (sc->mps_flags & MPS_FLAGS_REALLOCATED) { 3136 /* 3137 * Tell the caller that we freed the command in a 3138 * reinit. 3139 */ 3140 *cmp = NULL; 3141 } 3142 error = ETIMEDOUT; 3143 } 3144 return (error); 3145 } 3146 3147 /* 3148 * The MPT driver had a verbose interface for config pages. In this driver, 3149 * reduce it to much simpler terms, similar to the Linux driver. 3150 */ 3151 int 3152 mps_read_config_page(struct mps_softc *sc, struct mps_config_params *params) 3153 { 3154 MPI2_CONFIG_REQUEST *req; 3155 struct mps_command *cm; 3156 int error; 3157 3158 if (sc->mps_flags & MPS_FLAGS_BUSY) { 3159 return (EBUSY); 3160 } 3161 3162 cm = mps_alloc_command(sc); 3163 if (cm == NULL) { 3164 return (EBUSY); 3165 } 3166 3167 req = (MPI2_CONFIG_REQUEST *)cm->cm_req; 3168 req->Function = MPI2_FUNCTION_CONFIG; 3169 req->Action = params->action; 3170 req->SGLFlags = 0; 3171 req->ChainOffset = 0; 3172 req->PageAddress = params->page_address; 3173 if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { 3174 MPI2_CONFIG_EXTENDED_PAGE_HEADER *hdr; 3175 3176 hdr = ¶ms->hdr.Ext; 3177 req->ExtPageType = hdr->ExtPageType; 3178 req->ExtPageLength = hdr->ExtPageLength; 3179 req->Header.PageType = MPI2_CONFIG_PAGETYPE_EXTENDED; 3180 req->Header.PageLength = 0; /* Must be set to zero */ 3181 req->Header.PageNumber = hdr->PageNumber; 3182 req->Header.PageVersion = hdr->PageVersion; 3183 } else { 3184 MPI2_CONFIG_PAGE_HEADER *hdr; 3185 3186 hdr = ¶ms->hdr.Struct; 3187 req->Header.PageType = hdr->PageType; 3188 req->Header.PageNumber = hdr->PageNumber; 3189 req->Header.PageLength = hdr->PageLength; 3190 req->Header.PageVersion = hdr->PageVersion; 3191 } 3192 3193 cm->cm_data = params->buffer; 3194 cm->cm_length = params->length; 3195 if (cm->cm_data != NULL) { 3196 cm->cm_sge = &req->PageBufferSGE; 3197 cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION); 3198 cm->cm_flags = MPS_CM_FLAGS_SGE_SIMPLE | MPS_CM_FLAGS_DATAIN; 3199 } else 3200 cm->cm_sge = NULL; 3201 cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; 3202 3203 cm->cm_complete_data = params; 3204 if (params->callback != NULL) { 3205 cm->cm_complete = mps_config_complete; 3206 return (mps_map_command(sc, cm)); 3207 } else { 3208 error = mps_wait_command(sc, &cm, 0, CAN_SLEEP); 3209 if (error) { 3210 mps_dprint(sc, MPS_FAULT, 3211 "Error %d reading config page\n", error); 3212 if (cm != NULL) 3213 mps_free_command(sc, cm); 3214 return (error); 3215 } 3216 mps_config_complete(sc, cm); 3217 } 3218 3219 return (0); 3220 } 3221 3222 int 3223 mps_write_config_page(struct mps_softc *sc, struct mps_config_params *params) 3224 { 3225 return (EINVAL); 3226 } 3227 3228 static void 3229 mps_config_complete(struct mps_softc *sc, struct mps_command *cm) 3230 { 3231 MPI2_CONFIG_REPLY *reply; 3232 struct mps_config_params *params; 3233 3234 MPS_FUNCTRACE(sc); 3235 params = cm->cm_complete_data; 3236 3237 if (cm->cm_data != NULL) { 3238 bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, 3239 BUS_DMASYNC_POSTREAD); 3240 bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); 3241 } 3242 3243 /* 3244 * XXX KDM need to do more error recovery? This results in the 3245 * device in question not getting probed. 3246 */ 3247 if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { 3248 params->status = MPI2_IOCSTATUS_BUSY; 3249 goto done; 3250 } 3251 3252 reply = (MPI2_CONFIG_REPLY *)cm->cm_reply; 3253 if (reply == NULL) { 3254 params->status = MPI2_IOCSTATUS_BUSY; 3255 goto done; 3256 } 3257 params->status = reply->IOCStatus; 3258 if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { 3259 params->hdr.Ext.ExtPageType = reply->ExtPageType; 3260 params->hdr.Ext.ExtPageLength = reply->ExtPageLength; 3261 params->hdr.Ext.PageType = reply->Header.PageType; 3262 params->hdr.Ext.PageNumber = reply->Header.PageNumber; 3263 params->hdr.Ext.PageVersion = reply->Header.PageVersion; 3264 } else { 3265 params->hdr.Struct.PageType = reply->Header.PageType; 3266 params->hdr.Struct.PageNumber = reply->Header.PageNumber; 3267 params->hdr.Struct.PageLength = reply->Header.PageLength; 3268 params->hdr.Struct.PageVersion = reply->Header.PageVersion; 3269 } 3270 3271 done: 3272 mps_free_command(sc, cm); 3273 if (params->callback != NULL) 3274 params->callback(sc, params); 3275 3276 return; 3277 } 3278