request.hh revision 13954:2f400a5f2627
1/* 2 * Copyright (c) 2012-2013,2017-2018 ARM Limited 3 * All rights reserved 4 * 5 * The license below extends only to copyright in the software and shall 6 * not be construed as granting a license to any other intellectual 7 * property including but not limited to intellectual property relating 8 * to a hardware implementation of the functionality of the software 9 * licensed hereunder. You may use the software subject to the license 10 * terms below provided that you ensure that this notice is replicated 11 * unmodified and in its entirety in all distributions of the software, 12 * modified or unmodified, in source code or in binary form. 13 * 14 * Copyright (c) 2002-2005 The Regents of The University of Michigan 15 * Copyright (c) 2010,2015 Advanced Micro Devices, Inc. 16 * All rights reserved. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions are 20 * met: redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer; 22 * redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution; 25 * neither the name of the copyright holders nor the names of its 26 * contributors may be used to endorse or promote products derived from 27 * this software without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40 * 41 * Authors: Ron Dreslinski 42 * Steve Reinhardt 43 * Ali Saidi 44 */ 45 46/** 47 * @file 48 * Declaration of a request, the overall memory request consisting of 49 the parts of the request that are persistent throughout the transaction. 50 */ 51 52#ifndef __MEM_REQUEST_HH__ 53#define __MEM_REQUEST_HH__ 54 55#include <cassert> 56#include <climits> 57 58#include "base/flags.hh" 59#include "base/logging.hh" 60#include "base/types.hh" 61#include "cpu/inst_seq.hh" 62#include "sim/core.hh" 63 64/** 65 * Special TaskIds that are used for per-context-switch stats dumps 66 * and Cache Occupancy. Having too many tasks seems to be a problem 67 * with vector stats. 1024 seems to be a reasonable number that 68 * doesn't cause a problem with stats and is large enough to realistic 69 * benchmarks (Linux/Android boot, BBench, etc.) 70 */ 71 72namespace ContextSwitchTaskId { 73 enum TaskId { 74 MaxNormalTaskId = 1021, /* Maximum number of normal tasks */ 75 Prefetcher = 1022, /* For cache lines brought in by prefetcher */ 76 DMA = 1023, /* Mostly Table Walker */ 77 Unknown = 1024, 78 NumTaskId 79 }; 80} 81 82class Request; 83 84typedef std::shared_ptr<Request> RequestPtr; 85typedef uint16_t MasterID; 86 87class Request 88{ 89 public: 90 typedef uint64_t FlagsType; 91 typedef uint8_t ArchFlagsType; 92 typedef ::Flags<FlagsType> Flags; 93 94 enum : FlagsType { 95 /** 96 * Architecture specific flags. 97 * 98 * These bits int the flag field are reserved for 99 * architecture-specific code. For example, SPARC uses them to 100 * represent ASIs. 101 */ 102 ARCH_BITS = 0x000000FF, 103 /** The request was an instruction fetch. */ 104 INST_FETCH = 0x00000100, 105 /** The virtual address is also the physical address. */ 106 PHYSICAL = 0x00000200, 107 /** 108 * The request is to an uncacheable address. 109 * 110 * @note Uncacheable accesses may be reordered by CPU models. The 111 * STRICT_ORDER flag should be set if such reordering is 112 * undesirable. 113 */ 114 UNCACHEABLE = 0x00000400, 115 /** 116 * The request is required to be strictly ordered by <i>CPU 117 * models</i> and is non-speculative. 118 * 119 * A strictly ordered request is guaranteed to never be 120 * re-ordered or executed speculatively by a CPU model. The 121 * memory system may still reorder requests in caches unless 122 * the UNCACHEABLE flag is set as well. 123 */ 124 STRICT_ORDER = 0x00000800, 125 /** This request is to a memory mapped register. */ 126 MMAPPED_IPR = 0x00002000, 127 /** This request is made in privileged mode. */ 128 PRIVILEGED = 0x00008000, 129 130 /** 131 * This is a write that is targeted and zeroing an entire 132 * cache block. There is no need for a read/modify/write 133 */ 134 CACHE_BLOCK_ZERO = 0x00010000, 135 136 /** The request should not cause a memory access. */ 137 NO_ACCESS = 0x00080000, 138 /** 139 * This request will lock or unlock the accessed memory. When 140 * used with a load, the access locks the particular chunk of 141 * memory. When used with a store, it unlocks. The rule is 142 * that locked accesses have to be made up of a locked load, 143 * some operation on the data, and then a locked store. 144 */ 145 LOCKED_RMW = 0x00100000, 146 /** The request is a Load locked/store conditional. */ 147 LLSC = 0x00200000, 148 /** This request is for a memory swap. */ 149 MEM_SWAP = 0x00400000, 150 MEM_SWAP_COND = 0x00800000, 151 152 /** The request is a prefetch. */ 153 PREFETCH = 0x01000000, 154 /** The request should be prefetched into the exclusive state. */ 155 PF_EXCLUSIVE = 0x02000000, 156 /** The request should be marked as LRU. */ 157 EVICT_NEXT = 0x04000000, 158 /** The request should be marked with ACQUIRE. */ 159 ACQUIRE = 0x00020000, 160 /** The request should be marked with RELEASE. */ 161 RELEASE = 0x00040000, 162 163 /** The request is an atomic that returns data. */ 164 ATOMIC_RETURN_OP = 0x40000000, 165 /** The request is an atomic that does not return data. */ 166 ATOMIC_NO_RETURN_OP = 0x80000000, 167 168 /** The request should be marked with KERNEL. 169 * Used to indicate the synchronization associated with a GPU kernel 170 * launch or completion. 171 */ 172 KERNEL = 0x00001000, 173 174 /** 175 * The request should be handled by the generic IPR code (only 176 * valid together with MMAPPED_IPR) 177 */ 178 GENERIC_IPR = 0x08000000, 179 180 /** The request targets the secure memory space. */ 181 SECURE = 0x10000000, 182 /** The request is a page table walk */ 183 PT_WALK = 0x20000000, 184 185 /** The request invalidates a memory location */ 186 INVALIDATE = 0x0000000100000000, 187 /** The request cleans a memory location */ 188 CLEAN = 0x0000000200000000, 189 190 /** The request targets the point of unification */ 191 DST_POU = 0x0000001000000000, 192 193 /** The request targets the point of coherence */ 194 DST_POC = 0x0000002000000000, 195 196 /** Bits to define the destination of a request */ 197 DST_BITS = 0x0000003000000000, 198 199 /** 200 * These flags are *not* cleared when a Request object is 201 * reused (assigned a new address). 202 */ 203 STICKY_FLAGS = INST_FETCH 204 }; 205 static const FlagsType STORE_NO_DATA = CACHE_BLOCK_ZERO | 206 CLEAN | INVALIDATE; 207 208 /** Master Ids that are statically allocated 209 * @{*/ 210 enum : MasterID { 211 /** This master id is used for writeback requests by the caches */ 212 wbMasterId = 0, 213 /** 214 * This master id is used for functional requests that 215 * don't come from a particular device 216 */ 217 funcMasterId = 1, 218 /** This master id is used for message signaled interrupts */ 219 intMasterId = 2, 220 /** 221 * Invalid master id for assertion checking only. It is 222 * invalid behavior to ever send this id as part of a request. 223 */ 224 invldMasterId = std::numeric_limits<MasterID>::max() 225 }; 226 /** @} */ 227 228 typedef uint32_t MemSpaceConfigFlagsType; 229 typedef ::Flags<MemSpaceConfigFlagsType> MemSpaceConfigFlags; 230 231 enum : MemSpaceConfigFlagsType { 232 /** Has a synchronization scope been set? */ 233 SCOPE_VALID = 0x00000001, 234 /** Access has Wavefront scope visibility */ 235 WAVEFRONT_SCOPE = 0x00000002, 236 /** Access has Workgroup scope visibility */ 237 WORKGROUP_SCOPE = 0x00000004, 238 /** Access has Device (e.g., GPU) scope visibility */ 239 DEVICE_SCOPE = 0x00000008, 240 /** Access has System (e.g., CPU + GPU) scope visibility */ 241 SYSTEM_SCOPE = 0x00000010, 242 243 /** Global Segment */ 244 GLOBAL_SEGMENT = 0x00000020, 245 /** Group Segment */ 246 GROUP_SEGMENT = 0x00000040, 247 /** Private Segment */ 248 PRIVATE_SEGMENT = 0x00000080, 249 /** Kergarg Segment */ 250 KERNARG_SEGMENT = 0x00000100, 251 /** Readonly Segment */ 252 READONLY_SEGMENT = 0x00000200, 253 /** Spill Segment */ 254 SPILL_SEGMENT = 0x00000400, 255 /** Arg Segment */ 256 ARG_SEGMENT = 0x00000800, 257 }; 258 259 private: 260 typedef uint16_t PrivateFlagsType; 261 typedef ::Flags<PrivateFlagsType> PrivateFlags; 262 263 enum : PrivateFlagsType { 264 /** Whether or not the size is valid. */ 265 VALID_SIZE = 0x00000001, 266 /** Whether or not paddr is valid (has been written yet). */ 267 VALID_PADDR = 0x00000002, 268 /** Whether or not the vaddr & asid are valid. */ 269 VALID_VADDR = 0x00000004, 270 /** Whether or not the instruction sequence number is valid. */ 271 VALID_INST_SEQ_NUM = 0x00000008, 272 /** Whether or not the pc is valid. */ 273 VALID_PC = 0x00000010, 274 /** Whether or not the context ID is valid. */ 275 VALID_CONTEXT_ID = 0x00000020, 276 /** Whether or not the sc result is valid. */ 277 VALID_EXTRA_DATA = 0x00000080, 278 /** Whether or not the stream ID and substream ID is valid. */ 279 VALID_STREAM_ID = 0x00000100, 280 VALID_SUBSTREAM_ID = 0x00000200, 281 /** 282 * These flags are *not* cleared when a Request object is reused 283 * (assigned a new address). 284 */ 285 STICKY_PRIVATE_FLAGS = VALID_CONTEXT_ID 286 }; 287 288 private: 289 290 /** 291 * Set up a physical (e.g. device) request in a previously 292 * allocated Request object. 293 */ 294 void 295 setPhys(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time) 296 { 297 _paddr = paddr; 298 _size = size; 299 _time = time; 300 _masterId = mid; 301 _flags.clear(~STICKY_FLAGS); 302 _flags.set(flags); 303 privateFlags.clear(~STICKY_PRIVATE_FLAGS); 304 privateFlags.set(VALID_PADDR|VALID_SIZE); 305 depth = 0; 306 accessDelta = 0; 307 //translateDelta = 0; 308 } 309 310 /** 311 * The physical address of the request. Valid only if validPaddr 312 * is set. 313 */ 314 Addr _paddr; 315 316 /** 317 * The size of the request. This field must be set when vaddr or 318 * paddr is written via setVirt() or setPhys(), so it is always 319 * valid as long as one of the address fields is valid. 320 */ 321 unsigned _size; 322 323 /** Byte-enable mask for writes. */ 324 std::vector<bool> _byteEnable; 325 326 /** The requestor ID which is unique in the system for all ports 327 * that are capable of issuing a transaction 328 */ 329 MasterID _masterId; 330 331 /** Flag structure for the request. */ 332 Flags _flags; 333 334 /** Memory space configuraiton flag structure for the request. */ 335 MemSpaceConfigFlags _memSpaceConfigFlags; 336 337 /** Private flags for field validity checking. */ 338 PrivateFlags privateFlags; 339 340 /** 341 * The time this request was started. Used to calculate 342 * latencies. This field is set to curTick() any time paddr or vaddr 343 * is written. 344 */ 345 Tick _time; 346 347 /** 348 * The task id associated with this request 349 */ 350 uint32_t _taskId; 351 352 union { 353 struct { 354 /** 355 * The stream ID uniquely identifies a device behind the 356 * SMMU/IOMMU Each transaction arriving at the SMMU/IOMMU is 357 * associated with exactly one stream ID. 358 */ 359 uint32_t _streamId; 360 361 /** 362 * The substream ID identifies an "execution context" within a 363 * device behind an SMMU/IOMMU. It's intended to map 1-to-1 to 364 * PCIe PASID (Process Address Space ID). The presence of a 365 * substream ID is optional. 366 */ 367 uint32_t _substreamId; 368 }; 369 370 /** The address space ID. */ 371 uint64_t _asid; 372 }; 373 374 /** The virtual address of the request. */ 375 Addr _vaddr; 376 377 /** 378 * Extra data for the request, such as the return value of 379 * store conditional or the compare value for a CAS. */ 380 uint64_t _extraData; 381 382 /** The context ID (for statistics, locks, and wakeups). */ 383 ContextID _contextId; 384 385 /** program counter of initiating access; for tracing/debugging */ 386 Addr _pc; 387 388 /** Sequence number of the instruction that creates the request */ 389 InstSeqNum _reqInstSeqNum; 390 391 /** A pointer to an atomic operation */ 392 AtomicOpFunctor *atomicOpFunctor; 393 394 public: 395 396 /** 397 * Minimal constructor. No fields are initialized. (Note that 398 * _flags and privateFlags are cleared by Flags default 399 * constructor.) 400 */ 401 Request() 402 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 403 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 404 _extraData(0), _contextId(0), _pc(0), 405 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 406 accessDelta(0), depth(0) 407 {} 408 409 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, 410 InstSeqNum seq_num, ContextID cid) 411 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 412 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 413 _extraData(0), _contextId(0), _pc(0), 414 _reqInstSeqNum(seq_num), atomicOpFunctor(nullptr), translateDelta(0), 415 accessDelta(0), depth(0) 416 { 417 setPhys(paddr, size, flags, mid, curTick()); 418 setContext(cid); 419 privateFlags.set(VALID_INST_SEQ_NUM); 420 } 421 422 /** 423 * Constructor for physical (e.g. device) requests. Initializes 424 * just physical address, size, flags, and timestamp (to curTick()). 425 * These fields are adequate to perform a request. 426 */ 427 Request(Addr paddr, unsigned size, Flags flags, MasterID mid) 428 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 429 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 430 _extraData(0), _contextId(0), _pc(0), 431 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 432 accessDelta(0), depth(0) 433 { 434 setPhys(paddr, size, flags, mid, curTick()); 435 } 436 437 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time) 438 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 439 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 440 _extraData(0), _contextId(0), _pc(0), 441 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 442 accessDelta(0), depth(0) 443 { 444 setPhys(paddr, size, flags, mid, time); 445 } 446 447 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time, 448 Addr pc) 449 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 450 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 451 _extraData(0), _contextId(0), _pc(pc), 452 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 453 accessDelta(0), depth(0) 454 { 455 setPhys(paddr, size, flags, mid, time); 456 privateFlags.set(VALID_PC); 457 } 458 459 Request(uint64_t asid, Addr vaddr, unsigned size, Flags flags, 460 MasterID mid, Addr pc, ContextID cid) 461 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 462 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 463 _extraData(0), _contextId(0), _pc(0), 464 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 465 accessDelta(0), depth(0) 466 { 467 setVirt(asid, vaddr, size, flags, mid, pc); 468 setContext(cid); 469 } 470 471 Request(uint64_t asid, Addr vaddr, unsigned size, Flags flags, 472 MasterID mid, Addr pc, ContextID cid, 473 AtomicOpFunctor *atomic_op) 474 { 475 setVirt(asid, vaddr, size, flags, mid, pc, atomic_op); 476 setContext(cid); 477 } 478 479 Request(const Request& other) 480 : _paddr(other._paddr), _size(other._size), 481 _masterId(other._masterId), 482 _flags(other._flags), 483 _memSpaceConfigFlags(other._memSpaceConfigFlags), 484 privateFlags(other.privateFlags), 485 _time(other._time), 486 _taskId(other._taskId), _asid(other._asid), _vaddr(other._vaddr), 487 _extraData(other._extraData), _contextId(other._contextId), 488 _pc(other._pc), _reqInstSeqNum(other._reqInstSeqNum), 489 translateDelta(other.translateDelta), 490 accessDelta(other.accessDelta), depth(other.depth) 491 { 492 if (other.atomicOpFunctor) 493 atomicOpFunctor = (other.atomicOpFunctor)->clone(); 494 else 495 atomicOpFunctor = nullptr; 496 } 497 498 ~Request() 499 { 500 if (hasAtomicOpFunctor()) { 501 delete atomicOpFunctor; 502 } 503 } 504 505 /** 506 * Set up Context numbers. 507 */ 508 void 509 setContext(ContextID context_id) 510 { 511 _contextId = context_id; 512 privateFlags.set(VALID_CONTEXT_ID); 513 } 514 515 void 516 setStreamId(uint32_t sid) 517 { 518 _streamId = sid; 519 privateFlags.set(VALID_STREAM_ID); 520 } 521 522 void 523 setSubStreamId(uint32_t ssid) 524 { 525 assert(privateFlags.isSet(VALID_STREAM_ID)); 526 _substreamId = ssid; 527 privateFlags.set(VALID_SUBSTREAM_ID); 528 } 529 530 /** 531 * Set up a virtual (e.g., CPU) request in a previously 532 * allocated Request object. 533 */ 534 void 535 setVirt(uint64_t asid, Addr vaddr, unsigned size, Flags flags, 536 MasterID mid, Addr pc, AtomicOpFunctor *amo_op = nullptr) 537 { 538 _asid = asid; 539 _vaddr = vaddr; 540 _size = size; 541 _masterId = mid; 542 _pc = pc; 543 _time = curTick(); 544 545 _flags.clear(~STICKY_FLAGS); 546 _flags.set(flags); 547 privateFlags.clear(~STICKY_PRIVATE_FLAGS); 548 privateFlags.set(VALID_VADDR|VALID_SIZE|VALID_PC); 549 depth = 0; 550 accessDelta = 0; 551 translateDelta = 0; 552 atomicOpFunctor = amo_op; 553 } 554 555 /** 556 * Set just the physical address. This usually used to record the 557 * result of a translation. However, when using virtualized CPUs 558 * setPhys() is sometimes called to finalize a physical address 559 * without a virtual address, so we can't check if the virtual 560 * address is valid. 561 */ 562 void 563 setPaddr(Addr paddr) 564 { 565 _paddr = paddr; 566 privateFlags.set(VALID_PADDR); 567 } 568 569 /** 570 * Generate two requests as if this request had been split into two 571 * pieces. The original request can't have been translated already. 572 */ 573 // TODO: this function is still required by TimingSimpleCPU - should be 574 // removed once TimingSimpleCPU will support arbitrarily long multi-line 575 // mem. accesses 576 void splitOnVaddr(Addr split_addr, RequestPtr &req1, RequestPtr &req2) 577 { 578 assert(privateFlags.isSet(VALID_VADDR)); 579 assert(privateFlags.noneSet(VALID_PADDR)); 580 assert(split_addr > _vaddr && split_addr < _vaddr + _size); 581 req1 = std::make_shared<Request>(*this); 582 req2 = std::make_shared<Request>(*this); 583 req1->_size = split_addr - _vaddr; 584 req2->_vaddr = split_addr; 585 req2->_size = _size - req1->_size; 586 if (!_byteEnable.empty()) { 587 req1->_byteEnable = std::vector<bool>( 588 _byteEnable.begin(), 589 _byteEnable.begin() + req1->_size); 590 req2->_byteEnable = std::vector<bool>( 591 _byteEnable.begin() + req1->_size, 592 _byteEnable.end()); 593 } 594 } 595 596 /** 597 * Accessor for paddr. 598 */ 599 bool 600 hasPaddr() const 601 { 602 return privateFlags.isSet(VALID_PADDR); 603 } 604 605 Addr 606 getPaddr() const 607 { 608 assert(privateFlags.isSet(VALID_PADDR)); 609 return _paddr; 610 } 611 612 /** 613 * Time for the TLB/table walker to successfully translate this request. 614 */ 615 Tick translateDelta; 616 617 /** 618 * Access latency to complete this memory transaction not including 619 * translation time. 620 */ 621 Tick accessDelta; 622 623 /** 624 * Level of the cache hierachy where this request was responded to 625 * (e.g. 0 = L1; 1 = L2). 626 */ 627 mutable int depth; 628 629 /** 630 * Accessor for size. 631 */ 632 bool 633 hasSize() const 634 { 635 return privateFlags.isSet(VALID_SIZE); 636 } 637 638 unsigned 639 getSize() const 640 { 641 assert(privateFlags.isSet(VALID_SIZE)); 642 return _size; 643 } 644 645 const std::vector<bool>& 646 getByteEnable() const 647 { 648 return _byteEnable; 649 } 650 651 void 652 setByteEnable(const std::vector<bool>& be) 653 { 654 assert(be.empty() || be.size() == _size); 655 _byteEnable = be; 656 } 657 658 /** Accessor for time. */ 659 Tick 660 time() const 661 { 662 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 663 return _time; 664 } 665 666 /** 667 * Accessor for atomic-op functor. 668 */ 669 bool 670 hasAtomicOpFunctor() 671 { 672 return atomicOpFunctor != NULL; 673 } 674 675 AtomicOpFunctor * 676 getAtomicOpFunctor() 677 { 678 assert(atomicOpFunctor != NULL); 679 return atomicOpFunctor; 680 } 681 682 /** Accessor for flags. */ 683 Flags 684 getFlags() 685 { 686 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 687 return _flags; 688 } 689 690 /** Note that unlike other accessors, this function sets *specific 691 flags* (ORs them in); it does not assign its argument to the 692 _flags field. Thus this method should rightly be called 693 setFlags() and not just flags(). */ 694 void 695 setFlags(Flags flags) 696 { 697 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 698 _flags.set(flags); 699 } 700 701 void 702 setMemSpaceConfigFlags(MemSpaceConfigFlags extraFlags) 703 { 704 assert(privateFlags.isSet(VALID_PADDR | VALID_VADDR)); 705 _memSpaceConfigFlags.set(extraFlags); 706 } 707 708 /** Accessor function for vaddr.*/ 709 bool 710 hasVaddr() const 711 { 712 return privateFlags.isSet(VALID_VADDR); 713 } 714 715 Addr 716 getVaddr() const 717 { 718 assert(privateFlags.isSet(VALID_VADDR)); 719 return _vaddr; 720 } 721 722 /** Accesssor for the requestor id. */ 723 MasterID 724 masterId() const 725 { 726 return _masterId; 727 } 728 729 uint32_t 730 taskId() const 731 { 732 return _taskId; 733 } 734 735 void 736 taskId(uint32_t id) { 737 _taskId = id; 738 } 739 740 /** Accessor function for asid.*/ 741 uint64_t 742 getAsid() const 743 { 744 assert(privateFlags.isSet(VALID_VADDR)); 745 return _asid; 746 } 747 748 /** Accessor function for asid.*/ 749 void 750 setAsid(uint64_t asid) 751 { 752 _asid = asid; 753 } 754 755 /** Accessor function for architecture-specific flags.*/ 756 ArchFlagsType 757 getArchFlags() const 758 { 759 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 760 return _flags & ARCH_BITS; 761 } 762 763 /** Accessor function to check if sc result is valid. */ 764 bool 765 extraDataValid() const 766 { 767 return privateFlags.isSet(VALID_EXTRA_DATA); 768 } 769 770 /** Accessor function for store conditional return value.*/ 771 uint64_t 772 getExtraData() const 773 { 774 assert(privateFlags.isSet(VALID_EXTRA_DATA)); 775 return _extraData; 776 } 777 778 /** Accessor function for store conditional return value.*/ 779 void 780 setExtraData(uint64_t extraData) 781 { 782 _extraData = extraData; 783 privateFlags.set(VALID_EXTRA_DATA); 784 } 785 786 bool 787 hasContextId() const 788 { 789 return privateFlags.isSet(VALID_CONTEXT_ID); 790 } 791 792 /** Accessor function for context ID.*/ 793 ContextID 794 contextId() const 795 { 796 assert(privateFlags.isSet(VALID_CONTEXT_ID)); 797 return _contextId; 798 } 799 800 uint32_t 801 streamId() const 802 { 803 assert(privateFlags.isSet(VALID_STREAM_ID)); 804 return _streamId; 805 } 806 807 bool 808 hasSubstreamId() const 809 { 810 return privateFlags.isSet(VALID_SUBSTREAM_ID); 811 } 812 813 uint32_t 814 substreamId() const 815 { 816 assert(privateFlags.isSet(VALID_SUBSTREAM_ID)); 817 return _substreamId; 818 } 819 820 void 821 setPC(Addr pc) 822 { 823 privateFlags.set(VALID_PC); 824 _pc = pc; 825 } 826 827 bool 828 hasPC() const 829 { 830 return privateFlags.isSet(VALID_PC); 831 } 832 833 /** Accessor function for pc.*/ 834 Addr 835 getPC() const 836 { 837 assert(privateFlags.isSet(VALID_PC)); 838 return _pc; 839 } 840 841 /** 842 * Increment/Get the depth at which this request is responded to. 843 * This currently happens when the request misses in any cache level. 844 */ 845 void incAccessDepth() const { depth++; } 846 int getAccessDepth() const { return depth; } 847 848 /** 849 * Set/Get the time taken for this request to be successfully translated. 850 */ 851 void setTranslateLatency() { translateDelta = curTick() - _time; } 852 Tick getTranslateLatency() const { return translateDelta; } 853 854 /** 855 * Set/Get the time taken to complete this request's access, not including 856 * the time to successfully translate the request. 857 */ 858 void setAccessLatency() { accessDelta = curTick() - _time - translateDelta; } 859 Tick getAccessLatency() const { return accessDelta; } 860 861 /** 862 * Accessor for the sequence number of instruction that creates the 863 * request. 864 */ 865 bool 866 hasInstSeqNum() const 867 { 868 return privateFlags.isSet(VALID_INST_SEQ_NUM); 869 } 870 871 InstSeqNum 872 getReqInstSeqNum() const 873 { 874 assert(privateFlags.isSet(VALID_INST_SEQ_NUM)); 875 return _reqInstSeqNum; 876 } 877 878 void 879 setReqInstSeqNum(const InstSeqNum seq_num) 880 { 881 privateFlags.set(VALID_INST_SEQ_NUM); 882 _reqInstSeqNum = seq_num; 883 } 884 885 /** Accessor functions for flags. Note that these are for testing 886 only; setting flags should be done via setFlags(). */ 887 bool isUncacheable() const { return _flags.isSet(UNCACHEABLE); } 888 bool isStrictlyOrdered() const { return _flags.isSet(STRICT_ORDER); } 889 bool isInstFetch() const { return _flags.isSet(INST_FETCH); } 890 bool isPrefetch() const { return (_flags.isSet(PREFETCH) || 891 _flags.isSet(PF_EXCLUSIVE)); } 892 bool isPrefetchEx() const { return _flags.isSet(PF_EXCLUSIVE); } 893 bool isLLSC() const { return _flags.isSet(LLSC); } 894 bool isPriv() const { return _flags.isSet(PRIVILEGED); } 895 bool isLockedRMW() const { return _flags.isSet(LOCKED_RMW); } 896 bool isSwap() const { return _flags.isSet(MEM_SWAP|MEM_SWAP_COND); } 897 bool isCondSwap() const { return _flags.isSet(MEM_SWAP_COND); } 898 bool isMmappedIpr() const { return _flags.isSet(MMAPPED_IPR); } 899 bool isSecure() const { return _flags.isSet(SECURE); } 900 bool isPTWalk() const { return _flags.isSet(PT_WALK); } 901 bool isAcquire() const { return _flags.isSet(ACQUIRE); } 902 bool isRelease() const { return _flags.isSet(RELEASE); } 903 bool isKernel() const { return _flags.isSet(KERNEL); } 904 bool isAtomicReturn() const { return _flags.isSet(ATOMIC_RETURN_OP); } 905 bool isAtomicNoReturn() const { return _flags.isSet(ATOMIC_NO_RETURN_OP); } 906 907 bool 908 isAtomic() const 909 { 910 return _flags.isSet(ATOMIC_RETURN_OP) || 911 _flags.isSet(ATOMIC_NO_RETURN_OP); 912 } 913 914 /** 915 * Accessor functions for the destination of a memory request. The 916 * destination flag can specify a point of reference for the 917 * operation (e.g. a cache block clean to the the point of 918 * unification). At the moment the destination is only used by the 919 * cache maintenance operations. 920 */ 921 bool isToPOU() const { return _flags.isSet(DST_POU); } 922 bool isToPOC() const { return _flags.isSet(DST_POC); } 923 Flags getDest() const { return _flags & DST_BITS; } 924 925 /** 926 * Accessor functions for the memory space configuration flags and used by 927 * GPU ISAs such as the Heterogeneous System Architecture (HSA). Note that 928 * these are for testing only; setting extraFlags should be done via 929 * setMemSpaceConfigFlags(). 930 */ 931 bool isScoped() const { return _memSpaceConfigFlags.isSet(SCOPE_VALID); } 932 933 bool 934 isWavefrontScope() const 935 { 936 assert(isScoped()); 937 return _memSpaceConfigFlags.isSet(WAVEFRONT_SCOPE); 938 } 939 940 bool 941 isWorkgroupScope() const 942 { 943 assert(isScoped()); 944 return _memSpaceConfigFlags.isSet(WORKGROUP_SCOPE); 945 } 946 947 bool 948 isDeviceScope() const 949 { 950 assert(isScoped()); 951 return _memSpaceConfigFlags.isSet(DEVICE_SCOPE); 952 } 953 954 bool 955 isSystemScope() const 956 { 957 assert(isScoped()); 958 return _memSpaceConfigFlags.isSet(SYSTEM_SCOPE); 959 } 960 961 bool 962 isGlobalSegment() const 963 { 964 return _memSpaceConfigFlags.isSet(GLOBAL_SEGMENT) || 965 (!isGroupSegment() && !isPrivateSegment() && 966 !isKernargSegment() && !isReadonlySegment() && 967 !isSpillSegment() && !isArgSegment()); 968 } 969 970 bool 971 isGroupSegment() const 972 { 973 return _memSpaceConfigFlags.isSet(GROUP_SEGMENT); 974 } 975 976 bool 977 isPrivateSegment() const 978 { 979 return _memSpaceConfigFlags.isSet(PRIVATE_SEGMENT); 980 } 981 982 bool 983 isKernargSegment() const 984 { 985 return _memSpaceConfigFlags.isSet(KERNARG_SEGMENT); 986 } 987 988 bool 989 isReadonlySegment() const 990 { 991 return _memSpaceConfigFlags.isSet(READONLY_SEGMENT); 992 } 993 994 bool 995 isSpillSegment() const 996 { 997 return _memSpaceConfigFlags.isSet(SPILL_SEGMENT); 998 } 999 1000 bool 1001 isArgSegment() const 1002 { 1003 return _memSpaceConfigFlags.isSet(ARG_SEGMENT); 1004 } 1005 1006 /** 1007 * Accessor functions to determine whether this request is part of 1008 * a cache maintenance operation. At the moment three operations 1009 * are supported: 1010 1011 * 1) A cache clean operation updates all copies of a memory 1012 * location to the point of reference, 1013 * 2) A cache invalidate operation invalidates all copies of the 1014 * specified block in the memory above the point of reference, 1015 * 3) A clean and invalidate operation is a combination of the two 1016 * operations. 1017 * @{ */ 1018 bool isCacheClean() const { return _flags.isSet(CLEAN); } 1019 bool isCacheInvalidate() const { return _flags.isSet(INVALIDATE); } 1020 bool isCacheMaintenance() const { return _flags.isSet(CLEAN|INVALIDATE); } 1021 /** @} */ 1022}; 1023 1024#endif // __MEM_REQUEST_HH__ 1025