request.hh revision 12346:9b1144d046ca
1/* 2 * Copyright (c) 2012-2013,2017 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 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 targets the point of unification */ 186 DST_POU = 0x0000001000000000, 187 188 /** The request targets the point of coherence */ 189 DST_POC = 0x0000002000000000, 190 191 /** Bits to define the destination of a request */ 192 DST_BITS = 0x0000003000000000, 193 194 /** 195 * These flags are *not* cleared when a Request object is 196 * reused (assigned a new address). 197 */ 198 STICKY_FLAGS = INST_FETCH 199 }; 200 201 /** Master Ids that are statically allocated 202 * @{*/ 203 enum : MasterID { 204 /** This master id is used for writeback requests by the caches */ 205 wbMasterId = 0, 206 /** 207 * This master id is used for functional requests that 208 * don't come from a particular device 209 */ 210 funcMasterId = 1, 211 /** This master id is used for message signaled interrupts */ 212 intMasterId = 2, 213 /** 214 * Invalid master id for assertion checking only. It is 215 * invalid behavior to ever send this id as part of a request. 216 */ 217 invldMasterId = std::numeric_limits<MasterID>::max() 218 }; 219 /** @} */ 220 221 typedef uint32_t MemSpaceConfigFlagsType; 222 typedef ::Flags<MemSpaceConfigFlagsType> MemSpaceConfigFlags; 223 224 enum : MemSpaceConfigFlagsType { 225 /** Has a synchronization scope been set? */ 226 SCOPE_VALID = 0x00000001, 227 /** Access has Wavefront scope visibility */ 228 WAVEFRONT_SCOPE = 0x00000002, 229 /** Access has Workgroup scope visibility */ 230 WORKGROUP_SCOPE = 0x00000004, 231 /** Access has Device (e.g., GPU) scope visibility */ 232 DEVICE_SCOPE = 0x00000008, 233 /** Access has System (e.g., CPU + GPU) scope visibility */ 234 SYSTEM_SCOPE = 0x00000010, 235 236 /** Global Segment */ 237 GLOBAL_SEGMENT = 0x00000020, 238 /** Group Segment */ 239 GROUP_SEGMENT = 0x00000040, 240 /** Private Segment */ 241 PRIVATE_SEGMENT = 0x00000080, 242 /** Kergarg Segment */ 243 KERNARG_SEGMENT = 0x00000100, 244 /** Readonly Segment */ 245 READONLY_SEGMENT = 0x00000200, 246 /** Spill Segment */ 247 SPILL_SEGMENT = 0x00000400, 248 /** Arg Segment */ 249 ARG_SEGMENT = 0x00000800, 250 }; 251 252 private: 253 typedef uint8_t PrivateFlagsType; 254 typedef ::Flags<PrivateFlagsType> PrivateFlags; 255 256 enum : PrivateFlagsType { 257 /** Whether or not the size is valid. */ 258 VALID_SIZE = 0x00000001, 259 /** Whether or not paddr is valid (has been written yet). */ 260 VALID_PADDR = 0x00000002, 261 /** Whether or not the vaddr & asid are valid. */ 262 VALID_VADDR = 0x00000004, 263 /** Whether or not the instruction sequence number is valid. */ 264 VALID_INST_SEQ_NUM = 0x00000008, 265 /** Whether or not the pc is valid. */ 266 VALID_PC = 0x00000010, 267 /** Whether or not the context ID is valid. */ 268 VALID_CONTEXT_ID = 0x00000020, 269 /** Whether or not the sc result is valid. */ 270 VALID_EXTRA_DATA = 0x00000080, 271 /** 272 * These flags are *not* cleared when a Request object is reused 273 * (assigned a new address). 274 */ 275 STICKY_PRIVATE_FLAGS = VALID_CONTEXT_ID 276 }; 277 278 private: 279 280 /** 281 * Set up a physical (e.g. device) request in a previously 282 * allocated Request object. 283 */ 284 void 285 setPhys(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time) 286 { 287 _paddr = paddr; 288 _size = size; 289 _time = time; 290 _masterId = mid; 291 _flags.clear(~STICKY_FLAGS); 292 _flags.set(flags); 293 privateFlags.clear(~STICKY_PRIVATE_FLAGS); 294 privateFlags.set(VALID_PADDR|VALID_SIZE); 295 depth = 0; 296 accessDelta = 0; 297 //translateDelta = 0; 298 } 299 300 /** 301 * The physical address of the request. Valid only if validPaddr 302 * is set. 303 */ 304 Addr _paddr; 305 306 /** 307 * The size of the request. This field must be set when vaddr or 308 * paddr is written via setVirt() or setPhys(), so it is always 309 * valid as long as one of the address fields is valid. 310 */ 311 unsigned _size; 312 313 /** The requestor ID which is unique in the system for all ports 314 * that are capable of issuing a transaction 315 */ 316 MasterID _masterId; 317 318 /** Flag structure for the request. */ 319 Flags _flags; 320 321 /** Memory space configuraiton flag structure for the request. */ 322 MemSpaceConfigFlags _memSpaceConfigFlags; 323 324 /** Private flags for field validity checking. */ 325 PrivateFlags privateFlags; 326 327 /** 328 * The time this request was started. Used to calculate 329 * latencies. This field is set to curTick() any time paddr or vaddr 330 * is written. 331 */ 332 Tick _time; 333 334 /** 335 * The task id associated with this request 336 */ 337 uint32_t _taskId; 338 339 /** The address space ID. */ 340 int _asid; 341 342 /** The virtual address of the request. */ 343 Addr _vaddr; 344 345 /** 346 * Extra data for the request, such as the return value of 347 * store conditional or the compare value for a CAS. */ 348 uint64_t _extraData; 349 350 /** The context ID (for statistics, locks, and wakeups). */ 351 ContextID _contextId; 352 353 /** program counter of initiating access; for tracing/debugging */ 354 Addr _pc; 355 356 /** Sequence number of the instruction that creates the request */ 357 InstSeqNum _reqInstSeqNum; 358 359 /** A pointer to an atomic operation */ 360 AtomicOpFunctor *atomicOpFunctor; 361 362 public: 363 364 /** 365 * Minimal constructor. No fields are initialized. (Note that 366 * _flags and privateFlags are cleared by Flags default 367 * constructor.) 368 */ 369 Request() 370 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 371 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 372 _extraData(0), _contextId(0), _pc(0), 373 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 374 accessDelta(0), depth(0) 375 {} 376 377 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, 378 InstSeqNum seq_num, ContextID cid) 379 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 380 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 381 _extraData(0), _contextId(0), _pc(0), 382 _reqInstSeqNum(seq_num), atomicOpFunctor(nullptr), translateDelta(0), 383 accessDelta(0), depth(0) 384 { 385 setPhys(paddr, size, flags, mid, curTick()); 386 setContext(cid); 387 privateFlags.set(VALID_INST_SEQ_NUM); 388 } 389 390 /** 391 * Constructor for physical (e.g. device) requests. Initializes 392 * just physical address, size, flags, and timestamp (to curTick()). 393 * These fields are adequate to perform a request. 394 */ 395 Request(Addr paddr, unsigned size, Flags flags, MasterID mid) 396 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 397 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 398 _extraData(0), _contextId(0), _pc(0), 399 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 400 accessDelta(0), depth(0) 401 { 402 setPhys(paddr, size, flags, mid, curTick()); 403 } 404 405 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time) 406 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 407 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 408 _extraData(0), _contextId(0), _pc(0), 409 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 410 accessDelta(0), depth(0) 411 { 412 setPhys(paddr, size, flags, mid, time); 413 } 414 415 Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time, 416 Addr pc) 417 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 418 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 419 _extraData(0), _contextId(0), _pc(pc), 420 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 421 accessDelta(0), depth(0) 422 { 423 setPhys(paddr, size, flags, mid, time); 424 privateFlags.set(VALID_PC); 425 } 426 427 Request(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid, 428 Addr pc, ContextID cid) 429 : _paddr(0), _size(0), _masterId(invldMasterId), _time(0), 430 _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0), 431 _extraData(0), _contextId(0), _pc(0), 432 _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0), 433 accessDelta(0), depth(0) 434 { 435 setVirt(asid, vaddr, size, flags, mid, pc); 436 setContext(cid); 437 } 438 439 Request(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid, 440 Addr pc, ContextID cid, AtomicOpFunctor *atomic_op) 441 : atomicOpFunctor(atomic_op) 442 { 443 setVirt(asid, vaddr, size, flags, mid, pc); 444 setContext(cid); 445 } 446 447 ~Request() 448 { 449 if (hasAtomicOpFunctor()) { 450 delete atomicOpFunctor; 451 } 452 } 453 454 /** 455 * Set up Context numbers. 456 */ 457 void 458 setContext(ContextID context_id) 459 { 460 _contextId = context_id; 461 privateFlags.set(VALID_CONTEXT_ID); 462 } 463 464 /** 465 * Set up a virtual (e.g., CPU) request in a previously 466 * allocated Request object. 467 */ 468 void 469 setVirt(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid, 470 Addr pc) 471 { 472 _asid = asid; 473 _vaddr = vaddr; 474 _size = size; 475 _masterId = mid; 476 _pc = pc; 477 _time = curTick(); 478 479 _flags.clear(~STICKY_FLAGS); 480 _flags.set(flags); 481 privateFlags.clear(~STICKY_PRIVATE_FLAGS); 482 privateFlags.set(VALID_VADDR|VALID_SIZE|VALID_PC); 483 depth = 0; 484 accessDelta = 0; 485 translateDelta = 0; 486 } 487 488 /** 489 * Set just the physical address. This usually used to record the 490 * result of a translation. However, when using virtualized CPUs 491 * setPhys() is sometimes called to finalize a physical address 492 * without a virtual address, so we can't check if the virtual 493 * address is valid. 494 */ 495 void 496 setPaddr(Addr paddr) 497 { 498 _paddr = paddr; 499 privateFlags.set(VALID_PADDR); 500 } 501 502 /** 503 * Generate two requests as if this request had been split into two 504 * pieces. The original request can't have been translated already. 505 */ 506 void splitOnVaddr(Addr split_addr, RequestPtr &req1, RequestPtr &req2) 507 { 508 assert(privateFlags.isSet(VALID_VADDR)); 509 assert(privateFlags.noneSet(VALID_PADDR)); 510 assert(split_addr > _vaddr && split_addr < _vaddr + _size); 511 req1 = new Request(*this); 512 req2 = new Request(*this); 513 req1->_size = split_addr - _vaddr; 514 req2->_vaddr = split_addr; 515 req2->_size = _size - req1->_size; 516 } 517 518 /** 519 * Accessor for paddr. 520 */ 521 bool 522 hasPaddr() const 523 { 524 return privateFlags.isSet(VALID_PADDR); 525 } 526 527 Addr 528 getPaddr() const 529 { 530 assert(privateFlags.isSet(VALID_PADDR)); 531 return _paddr; 532 } 533 534 /** 535 * Time for the TLB/table walker to successfully translate this request. 536 */ 537 Tick translateDelta; 538 539 /** 540 * Access latency to complete this memory transaction not including 541 * translation time. 542 */ 543 Tick accessDelta; 544 545 /** 546 * Level of the cache hierachy where this request was responded to 547 * (e.g. 0 = L1; 1 = L2). 548 */ 549 mutable int depth; 550 551 /** 552 * Accessor for size. 553 */ 554 bool 555 hasSize() const 556 { 557 return privateFlags.isSet(VALID_SIZE); 558 } 559 560 unsigned 561 getSize() const 562 { 563 assert(privateFlags.isSet(VALID_SIZE)); 564 return _size; 565 } 566 567 /** Accessor for time. */ 568 Tick 569 time() const 570 { 571 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 572 return _time; 573 } 574 575 /** 576 * Accessor for atomic-op functor. 577 */ 578 bool 579 hasAtomicOpFunctor() 580 { 581 return atomicOpFunctor != NULL; 582 } 583 584 AtomicOpFunctor * 585 getAtomicOpFunctor() 586 { 587 assert(atomicOpFunctor != NULL); 588 return atomicOpFunctor; 589 } 590 591 /** Accessor for flags. */ 592 Flags 593 getFlags() 594 { 595 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 596 return _flags; 597 } 598 599 /** Note that unlike other accessors, this function sets *specific 600 flags* (ORs them in); it does not assign its argument to the 601 _flags field. Thus this method should rightly be called 602 setFlags() and not just flags(). */ 603 void 604 setFlags(Flags flags) 605 { 606 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 607 _flags.set(flags); 608 } 609 610 void 611 setMemSpaceConfigFlags(MemSpaceConfigFlags extraFlags) 612 { 613 assert(privateFlags.isSet(VALID_PADDR | VALID_VADDR)); 614 _memSpaceConfigFlags.set(extraFlags); 615 } 616 617 /** Accessor function for vaddr.*/ 618 bool 619 hasVaddr() const 620 { 621 return privateFlags.isSet(VALID_VADDR); 622 } 623 624 Addr 625 getVaddr() const 626 { 627 assert(privateFlags.isSet(VALID_VADDR)); 628 return _vaddr; 629 } 630 631 /** Accesssor for the requestor id. */ 632 MasterID 633 masterId() const 634 { 635 return _masterId; 636 } 637 638 uint32_t 639 taskId() const 640 { 641 return _taskId; 642 } 643 644 void 645 taskId(uint32_t id) { 646 _taskId = id; 647 } 648 649 /** Accessor function for asid.*/ 650 int 651 getAsid() const 652 { 653 assert(privateFlags.isSet(VALID_VADDR)); 654 return _asid; 655 } 656 657 /** Accessor function for asid.*/ 658 void 659 setAsid(int asid) 660 { 661 _asid = asid; 662 } 663 664 /** Accessor function for architecture-specific flags.*/ 665 ArchFlagsType 666 getArchFlags() const 667 { 668 assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR)); 669 return _flags & ARCH_BITS; 670 } 671 672 /** Accessor function to check if sc result is valid. */ 673 bool 674 extraDataValid() const 675 { 676 return privateFlags.isSet(VALID_EXTRA_DATA); 677 } 678 679 /** Accessor function for store conditional return value.*/ 680 uint64_t 681 getExtraData() const 682 { 683 assert(privateFlags.isSet(VALID_EXTRA_DATA)); 684 return _extraData; 685 } 686 687 /** Accessor function for store conditional return value.*/ 688 void 689 setExtraData(uint64_t extraData) 690 { 691 _extraData = extraData; 692 privateFlags.set(VALID_EXTRA_DATA); 693 } 694 695 bool 696 hasContextId() const 697 { 698 return privateFlags.isSet(VALID_CONTEXT_ID); 699 } 700 701 /** Accessor function for context ID.*/ 702 ContextID 703 contextId() const 704 { 705 assert(privateFlags.isSet(VALID_CONTEXT_ID)); 706 return _contextId; 707 } 708 709 void 710 setPC(Addr pc) 711 { 712 privateFlags.set(VALID_PC); 713 _pc = pc; 714 } 715 716 bool 717 hasPC() const 718 { 719 return privateFlags.isSet(VALID_PC); 720 } 721 722 /** Accessor function for pc.*/ 723 Addr 724 getPC() const 725 { 726 assert(privateFlags.isSet(VALID_PC)); 727 return _pc; 728 } 729 730 /** 731 * Increment/Get the depth at which this request is responded to. 732 * This currently happens when the request misses in any cache level. 733 */ 734 void incAccessDepth() const { depth++; } 735 int getAccessDepth() const { return depth; } 736 737 /** 738 * Set/Get the time taken for this request to be successfully translated. 739 */ 740 void setTranslateLatency() { translateDelta = curTick() - _time; } 741 Tick getTranslateLatency() const { return translateDelta; } 742 743 /** 744 * Set/Get the time taken to complete this request's access, not including 745 * the time to successfully translate the request. 746 */ 747 void setAccessLatency() { accessDelta = curTick() - _time - translateDelta; } 748 Tick getAccessLatency() const { return accessDelta; } 749 750 /** 751 * Accessor for the sequence number of instruction that creates the 752 * request. 753 */ 754 bool 755 hasInstSeqNum() const 756 { 757 return privateFlags.isSet(VALID_INST_SEQ_NUM); 758 } 759 760 InstSeqNum 761 getReqInstSeqNum() const 762 { 763 assert(privateFlags.isSet(VALID_INST_SEQ_NUM)); 764 return _reqInstSeqNum; 765 } 766 767 void 768 setReqInstSeqNum(const InstSeqNum seq_num) 769 { 770 privateFlags.set(VALID_INST_SEQ_NUM); 771 _reqInstSeqNum = seq_num; 772 } 773 774 /** Accessor functions for flags. Note that these are for testing 775 only; setting flags should be done via setFlags(). */ 776 bool isUncacheable() const { return _flags.isSet(UNCACHEABLE); } 777 bool isStrictlyOrdered() const { return _flags.isSet(STRICT_ORDER); } 778 bool isInstFetch() const { return _flags.isSet(INST_FETCH); } 779 bool isPrefetch() const { return _flags.isSet(PREFETCH); } 780 bool isLLSC() const { return _flags.isSet(LLSC); } 781 bool isPriv() const { return _flags.isSet(PRIVILEGED); } 782 bool isLockedRMW() const { return _flags.isSet(LOCKED_RMW); } 783 bool isSwap() const { return _flags.isSet(MEM_SWAP|MEM_SWAP_COND); } 784 bool isCondSwap() const { return _flags.isSet(MEM_SWAP_COND); } 785 bool isMmappedIpr() const { return _flags.isSet(MMAPPED_IPR); } 786 bool isSecure() const { return _flags.isSet(SECURE); } 787 bool isPTWalk() const { return _flags.isSet(PT_WALK); } 788 bool isAcquire() const { return _flags.isSet(ACQUIRE); } 789 bool isRelease() const { return _flags.isSet(RELEASE); } 790 bool isKernel() const { return _flags.isSet(KERNEL); } 791 bool isAtomicReturn() const { return _flags.isSet(ATOMIC_RETURN_OP); } 792 bool isAtomicNoReturn() const { return _flags.isSet(ATOMIC_NO_RETURN_OP); } 793 794 bool 795 isAtomic() const 796 { 797 return _flags.isSet(ATOMIC_RETURN_OP) || 798 _flags.isSet(ATOMIC_NO_RETURN_OP); 799 } 800 801 /** 802 * Accessor functions for the destination of a memory request. The 803 * destination flag can specify a point of reference for the 804 * operation (e.g. a cache block clean to the the point of 805 * unification). At the moment the destination is only used by the 806 * cache maintenance operations. 807 */ 808 bool isToPOU() const { return _flags.isSet(DST_POU); } 809 bool isToPOC() const { return _flags.isSet(DST_POC); } 810 Flags getDest() const { return _flags & DST_BITS; } 811 812 /** 813 * Accessor functions for the memory space configuration flags and used by 814 * GPU ISAs such as the Heterogeneous System Architecture (HSA). Note that 815 * these are for testing only; setting extraFlags should be done via 816 * setMemSpaceConfigFlags(). 817 */ 818 bool isScoped() const { return _memSpaceConfigFlags.isSet(SCOPE_VALID); } 819 820 bool 821 isWavefrontScope() const 822 { 823 assert(isScoped()); 824 return _memSpaceConfigFlags.isSet(WAVEFRONT_SCOPE); 825 } 826 827 bool 828 isWorkgroupScope() const 829 { 830 assert(isScoped()); 831 return _memSpaceConfigFlags.isSet(WORKGROUP_SCOPE); 832 } 833 834 bool 835 isDeviceScope() const 836 { 837 assert(isScoped()); 838 return _memSpaceConfigFlags.isSet(DEVICE_SCOPE); 839 } 840 841 bool 842 isSystemScope() const 843 { 844 assert(isScoped()); 845 return _memSpaceConfigFlags.isSet(SYSTEM_SCOPE); 846 } 847 848 bool 849 isGlobalSegment() const 850 { 851 return _memSpaceConfigFlags.isSet(GLOBAL_SEGMENT) || 852 (!isGroupSegment() && !isPrivateSegment() && 853 !isKernargSegment() && !isReadonlySegment() && 854 !isSpillSegment() && !isArgSegment()); 855 } 856 857 bool 858 isGroupSegment() const 859 { 860 return _memSpaceConfigFlags.isSet(GROUP_SEGMENT); 861 } 862 863 bool 864 isPrivateSegment() const 865 { 866 return _memSpaceConfigFlags.isSet(PRIVATE_SEGMENT); 867 } 868 869 bool 870 isKernargSegment() const 871 { 872 return _memSpaceConfigFlags.isSet(KERNARG_SEGMENT); 873 } 874 875 bool 876 isReadonlySegment() const 877 { 878 return _memSpaceConfigFlags.isSet(READONLY_SEGMENT); 879 } 880 881 bool 882 isSpillSegment() const 883 { 884 return _memSpaceConfigFlags.isSet(SPILL_SEGMENT); 885 } 886 887 bool 888 isArgSegment() const 889 { 890 return _memSpaceConfigFlags.isSet(ARG_SEGMENT); 891 } 892}; 893 894#endif // __MEM_REQUEST_HH__ 895