/*
+ * Copyright (c) 2012-2013,2017 ARM Limited
+ * All rights reserved
+ *
+ * The license below extends only to copyright in the software and shall
+ * not be construed as granting a license to any other intellectual
+ * property including but not limited to intellectual property relating
+ * to a hardware implementation of the functionality of the software
+ * licensed hereunder. You may use the software subject to the license
+ * terms below provided that you ensure that this notice is replicated
+ * unmodified and in its entirety in all distributions of the software,
+ * modified or unmodified, in source code or in binary form.
+ *
* Copyright (c) 2002-2005 The Regents of The University of Michigan
+ * Copyright (c) 2010,2015 Advanced Micro Devices, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#define __MEM_REQUEST_HH__
#include <cassert>
+#include <climits>
-#include "base/fast_alloc.hh"
#include "base/flags.hh"
-#include "base/misc.hh"
-#include "sim/host.hh"
+#include "base/logging.hh"
+#include "base/types.hh"
+#include "cpu/inst_seq.hh"
#include "sim/core.hh"
+/**
+ * Special TaskIds that are used for per-context-switch stats dumps
+ * and Cache Occupancy. Having too many tasks seems to be a problem
+ * with vector stats. 1024 seems to be a reasonable number that
+ * doesn't cause a problem with stats and is large enough to realistic
+ * benchmarks (Linux/Android boot, BBench, etc.)
+ */
+
+namespace ContextSwitchTaskId {
+ enum TaskId {
+ MaxNormalTaskId = 1021, /* Maximum number of normal tasks */
+ Prefetcher = 1022, /* For cache lines brought in by prefetcher */
+ DMA = 1023, /* Mostly Table Walker */
+ Unknown = 1024,
+ NumTaskId
+ };
+}
+
class Request;
typedef Request* RequestPtr;
+typedef uint16_t MasterID;
-class Request : public FastAlloc
+class Request
{
- friend class Packet;
-
public:
- typedef uint32_t FlagsType;
+ typedef uint64_t FlagsType;
+ typedef uint8_t ArchFlagsType;
typedef ::Flags<FlagsType> Flags;
- /** ASI information for this request if it exists. */
- static const FlagsType ASI_BITS = 0x000000FF;
- /** The request is a Load locked/store conditional. */
- static const FlagsType LLSC = 0x00000100;
- /** The virtual address is also the physical address. */
- static const FlagsType PHYSICAL = 0x00000200;
- /** The request is an ALPHA VPTE pal access (hw_ld). */
- static const FlagsType VPTE = 0x00000400;
- /** Use the alternate mode bits in ALPHA. */
- static const FlagsType ALTMODE = 0x00000800;
- /** The request is to an uncacheable address. */
- static const FlagsType UNCACHEABLE = 0x00001000;
- /** The request should not cause a page fault. */
- static const FlagsType NO_FAULT = 0x00002000;
- /** The request should not cause a memory access. */
- static const FlagsType NO_ACCESS = 0x00004000;
- /** This request will lock or unlock the accessed memory. */
- static const FlagsType LOCKED = 0x00008000;
- /** The request should be prefetched into the exclusive state. */
- static const FlagsType PF_EXCLUSIVE = 0x00010000;
- /** The request should be marked as LRU. */
- static const FlagsType EVICT_NEXT = 0x00020000;
- /** The request should ignore unaligned access faults */
- static const FlagsType NO_ALIGN_FAULT = 0x00040000;
- /** The request was an instruction read. */
- static const FlagsType INST_READ = 0x00080000;
- /** This request is for a memory swap. */
- static const FlagsType MEM_SWAP = 0x00100000;
- static const FlagsType MEM_SWAP_COND = 0x00200000;
- /** The request should ignore unaligned access faults */
- static const FlagsType NO_HALF_WORD_ALIGN_FAULT = 0x00400000;
- /** This request is to a memory mapped register. */
- static const FlagsType MMAPED_IPR = 0x00800000;
+ enum : FlagsType {
+ /**
+ * Architecture specific flags.
+ *
+ * These bits int the flag field are reserved for
+ * architecture-specific code. For example, SPARC uses them to
+ * represent ASIs.
+ */
+ ARCH_BITS = 0x000000FF,
+ /** The request was an instruction fetch. */
+ INST_FETCH = 0x00000100,
+ /** The virtual address is also the physical address. */
+ PHYSICAL = 0x00000200,
+ /**
+ * The request is to an uncacheable address.
+ *
+ * @note Uncacheable accesses may be reordered by CPU models. The
+ * STRICT_ORDER flag should be set if such reordering is
+ * undesirable.
+ */
+ UNCACHEABLE = 0x00000400,
+ /**
+ * The request is required to be strictly ordered by <i>CPU
+ * models</i> and is non-speculative.
+ *
+ * A strictly ordered request is guaranteed to never be
+ * re-ordered or executed speculatively by a CPU model. The
+ * memory system may still reorder requests in caches unless
+ * the UNCACHEABLE flag is set as well.
+ */
+ STRICT_ORDER = 0x00000800,
+ /** This request is to a memory mapped register. */
+ MMAPPED_IPR = 0x00002000,
+ /** This request is made in privileged mode. */
+ PRIVILEGED = 0x00008000,
+
+ /**
+ * This is a write that is targeted and zeroing an entire
+ * cache block. There is no need for a read/modify/write
+ */
+ CACHE_BLOCK_ZERO = 0x00010000,
+
+ /** The request should not cause a memory access. */
+ NO_ACCESS = 0x00080000,
+ /**
+ * This request will lock or unlock the accessed memory. When
+ * used with a load, the access locks the particular chunk of
+ * memory. When used with a store, it unlocks. The rule is
+ * that locked accesses have to be made up of a locked load,
+ * some operation on the data, and then a locked store.
+ */
+ LOCKED_RMW = 0x00100000,
+ /** The request is a Load locked/store conditional. */
+ LLSC = 0x00200000,
+ /** This request is for a memory swap. */
+ MEM_SWAP = 0x00400000,
+ MEM_SWAP_COND = 0x00800000,
+
+ /** The request is a prefetch. */
+ PREFETCH = 0x01000000,
+ /** The request should be prefetched into the exclusive state. */
+ PF_EXCLUSIVE = 0x02000000,
+ /** The request should be marked as LRU. */
+ EVICT_NEXT = 0x04000000,
+ /** The request should be marked with ACQUIRE. */
+ ACQUIRE = 0x00020000,
+ /** The request should be marked with RELEASE. */
+ RELEASE = 0x00040000,
+
+ /** The request is an atomic that returns data. */
+ ATOMIC_RETURN_OP = 0x40000000,
+ /** The request is an atomic that does not return data. */
+ ATOMIC_NO_RETURN_OP = 0x80000000,
+
+ /** The request should be marked with KERNEL.
+ * Used to indicate the synchronization associated with a GPU kernel
+ * launch or completion.
+ */
+ KERNEL = 0x00001000,
+
+ /**
+ * The request should be handled by the generic IPR code (only
+ * valid together with MMAPPED_IPR)
+ */
+ GENERIC_IPR = 0x08000000,
+
+ /** The request targets the secure memory space. */
+ SECURE = 0x10000000,
+ /** The request is a page table walk */
+ PT_WALK = 0x20000000,
+
+ /** The request invalidates a memory location */
+ INVALIDATE = 0x0000000100000000,
+ /** The request cleans a memory location */
+ CLEAN = 0x0000000200000000,
+
+ /** The request targets the point of unification */
+ DST_POU = 0x0000001000000000,
+
+ /** The request targets the point of coherence */
+ DST_POC = 0x0000002000000000,
+
+ /** Bits to define the destination of a request */
+ DST_BITS = 0x0000003000000000,
+
+ /**
+ * These flags are *not* cleared when a Request object is
+ * reused (assigned a new address).
+ */
+ STICKY_FLAGS = INST_FETCH
+ };
+ static const FlagsType STORE_NO_DATA = CACHE_BLOCK_ZERO |
+ CLEAN | INVALIDATE;
+
+ /** Master Ids that are statically allocated
+ * @{*/
+ enum : MasterID {
+ /** This master id is used for writeback requests by the caches */
+ wbMasterId = 0,
+ /**
+ * This master id is used for functional requests that
+ * don't come from a particular device
+ */
+ funcMasterId = 1,
+ /** This master id is used for message signaled interrupts */
+ intMasterId = 2,
+ /**
+ * Invalid master id for assertion checking only. It is
+ * invalid behavior to ever send this id as part of a request.
+ */
+ invldMasterId = std::numeric_limits<MasterID>::max()
+ };
+ /** @} */
+
+ typedef uint32_t MemSpaceConfigFlagsType;
+ typedef ::Flags<MemSpaceConfigFlagsType> MemSpaceConfigFlags;
+
+ enum : MemSpaceConfigFlagsType {
+ /** Has a synchronization scope been set? */
+ SCOPE_VALID = 0x00000001,
+ /** Access has Wavefront scope visibility */
+ WAVEFRONT_SCOPE = 0x00000002,
+ /** Access has Workgroup scope visibility */
+ WORKGROUP_SCOPE = 0x00000004,
+ /** Access has Device (e.g., GPU) scope visibility */
+ DEVICE_SCOPE = 0x00000008,
+ /** Access has System (e.g., CPU + GPU) scope visibility */
+ SYSTEM_SCOPE = 0x00000010,
+
+ /** Global Segment */
+ GLOBAL_SEGMENT = 0x00000020,
+ /** Group Segment */
+ GROUP_SEGMENT = 0x00000040,
+ /** Private Segment */
+ PRIVATE_SEGMENT = 0x00000080,
+ /** Kergarg Segment */
+ KERNARG_SEGMENT = 0x00000100,
+ /** Readonly Segment */
+ READONLY_SEGMENT = 0x00000200,
+ /** Spill Segment */
+ SPILL_SEGMENT = 0x00000400,
+ /** Arg Segment */
+ ARG_SEGMENT = 0x00000800,
+ };
private:
- static const FlagsType PUBLIC_FLAGS = 0x00FFFFFF;
- static const FlagsType PRIVATE_FLAGS = 0xFF000000;
-
- /** Whether or not the size is valid. */
- static const FlagsType VALID_SIZE = 0x01000000;
- /** Whether or not paddr is valid (has been written yet). */
- static const FlagsType VALID_PADDR = 0x02000000;
- /** Whether or not the vaddr & asid are valid. */
- static const FlagsType VALID_VADDR = 0x04000000;
- /** Whether or not the pc is valid. */
- static const FlagsType VALID_PC = 0x10000000;
- /** Whether or not the context ID is valid. */
- static const FlagsType VALID_CONTEXT_ID = 0x20000000;
- static const FlagsType VALID_THREAD_ID = 0x40000000;
- /** Whether or not the sc result is valid. */
- static const FlagsType VALID_EXTRA_DATA = 0x80000000;
+ typedef uint8_t PrivateFlagsType;
+ typedef ::Flags<PrivateFlagsType> PrivateFlags;
+
+ enum : PrivateFlagsType {
+ /** Whether or not the size is valid. */
+ VALID_SIZE = 0x00000001,
+ /** Whether or not paddr is valid (has been written yet). */
+ VALID_PADDR = 0x00000002,
+ /** Whether or not the vaddr & asid are valid. */
+ VALID_VADDR = 0x00000004,
+ /** Whether or not the instruction sequence number is valid. */
+ VALID_INST_SEQ_NUM = 0x00000008,
+ /** Whether or not the pc is valid. */
+ VALID_PC = 0x00000010,
+ /** Whether or not the context ID is valid. */
+ VALID_CONTEXT_ID = 0x00000020,
+ /** Whether or not the sc result is valid. */
+ VALID_EXTRA_DATA = 0x00000080,
+ /**
+ * These flags are *not* cleared when a Request object is reused
+ * (assigned a new address).
+ */
+ STICKY_PRIVATE_FLAGS = VALID_CONTEXT_ID
+ };
private:
+
+ /**
+ * Set up a physical (e.g. device) request in a previously
+ * allocated Request object.
+ */
+ void
+ setPhys(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time)
+ {
+ _paddr = paddr;
+ _size = size;
+ _time = time;
+ _masterId = mid;
+ _flags.clear(~STICKY_FLAGS);
+ _flags.set(flags);
+ privateFlags.clear(~STICKY_PRIVATE_FLAGS);
+ privateFlags.set(VALID_PADDR|VALID_SIZE);
+ depth = 0;
+ accessDelta = 0;
+ //translateDelta = 0;
+ }
+
/**
* The physical address of the request. Valid only if validPaddr
* is set.
*/
- Addr paddr;
+ Addr _paddr;
/**
* The size of the request. This field must be set when vaddr or
* paddr is written via setVirt() or setPhys(), so it is always
* valid as long as one of the address fields is valid.
*/
- int size;
+ unsigned _size;
+
+ /** The requestor ID which is unique in the system for all ports
+ * that are capable of issuing a transaction
+ */
+ MasterID _masterId;
/** Flag structure for the request. */
- Flags flags;
+ Flags _flags;
+
+ /** Memory space configuraiton flag structure for the request. */
+ MemSpaceConfigFlags _memSpaceConfigFlags;
+
+ /** Private flags for field validity checking. */
+ PrivateFlags privateFlags;
/**
* The time this request was started. Used to calculate
- * latencies. This field is set to curTick any time paddr or vaddr
+ * latencies. This field is set to curTick() any time paddr or vaddr
* is written.
*/
- Tick time;
+ Tick _time;
+
+ /**
+ * The task id associated with this request
+ */
+ uint32_t _taskId;
/** The address space ID. */
- int asid;
+ int _asid;
/** The virtual address of the request. */
- Addr vaddr;
+ Addr _vaddr;
/**
* Extra data for the request, such as the return value of
* store conditional or the compare value for a CAS. */
- uint64_t extraData;
+ uint64_t _extraData;
- /** The context ID (for statistics, typically). */
- int _contextId;
- /** The thread ID (id within this CPU) */
- int _threadId;
+ /** The context ID (for statistics, locks, and wakeups). */
+ ContextID _contextId;
/** program counter of initiating access; for tracing/debugging */
- Addr pc;
+ Addr _pc;
+
+ /** Sequence number of the instruction that creates the request */
+ InstSeqNum _reqInstSeqNum;
+
+ /** A pointer to an atomic operation */
+ AtomicOpFunctor *atomicOpFunctor;
public:
- /** Minimal constructor. No fields are initialized. */
+
+ /**
+ * Minimal constructor. No fields are initialized. (Note that
+ * _flags and privateFlags are cleared by Flags default
+ * constructor.)
+ */
Request()
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(0),
+ _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
{}
+ Request(Addr paddr, unsigned size, Flags flags, MasterID mid,
+ InstSeqNum seq_num, ContextID cid)
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(0),
+ _reqInstSeqNum(seq_num), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
+ {
+ setPhys(paddr, size, flags, mid, curTick());
+ setContext(cid);
+ privateFlags.set(VALID_INST_SEQ_NUM);
+ }
+
/**
* Constructor for physical (e.g. device) requests. Initializes
- * just physical address, size, flags, and timestamp (to curTick).
- * These fields are adequate to perform a request.
+ * just physical address, size, flags, and timestamp (to curTick()).
+ * These fields are adequate to perform a request.
*/
- Request(Addr paddr, int size, Flags flags)
+ Request(Addr paddr, unsigned size, Flags flags, MasterID mid)
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(0),
+ _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
{
- setPhys(paddr, size, flags);
+ setPhys(paddr, size, flags, mid, curTick());
}
- Request(int asid, Addr vaddr, int size, Flags flags, Addr pc,
- int cid, int tid)
+ Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time)
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(0),
+ _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
{
- setThreadContext(cid, tid);
- setVirt(asid, vaddr, size, flags, pc);
+ setPhys(paddr, size, flags, mid, time);
}
- ~Request() {} // for FastAlloc
+ Request(Addr paddr, unsigned size, Flags flags, MasterID mid, Tick time,
+ Addr pc)
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(pc),
+ _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
+ {
+ setPhys(paddr, size, flags, mid, time);
+ privateFlags.set(VALID_PC);
+ }
- /**
- * Set up CPU and thread numbers.
- */
- void
- setThreadContext(int context_id, int thread_id)
+ Request(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid,
+ Addr pc, ContextID cid)
+ : _paddr(0), _size(0), _masterId(invldMasterId), _time(0),
+ _taskId(ContextSwitchTaskId::Unknown), _asid(0), _vaddr(0),
+ _extraData(0), _contextId(0), _pc(0),
+ _reqInstSeqNum(0), atomicOpFunctor(nullptr), translateDelta(0),
+ accessDelta(0), depth(0)
{
- _contextId = context_id;
- _threadId = thread_id;
- flags.set(VALID_CONTEXT_ID|VALID_THREAD_ID);
+ setVirt(asid, vaddr, size, flags, mid, pc);
+ setContext(cid);
+ }
+
+ Request(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid,
+ Addr pc, ContextID cid, AtomicOpFunctor *atomic_op)
+ : atomicOpFunctor(atomic_op)
+ {
+ setVirt(asid, vaddr, size, flags, mid, pc);
+ setContext(cid);
+ }
+
+ ~Request()
+ {
+ if (hasAtomicOpFunctor()) {
+ delete atomicOpFunctor;
+ }
}
/**
- * Set up a physical (e.g. device) request in a previously
- * allocated Request object.
+ * Set up Context numbers.
*/
void
- setPhys(Addr _paddr, int _size, Flags _flags)
+ setContext(ContextID context_id)
{
- assert(_size >= 0);
- paddr = _paddr;
- size = _size;
- time = curTick;
-
- flags.set(VALID_PADDR|VALID_SIZE);
- flags.clear(VALID_VADDR|VALID_PC|VALID_EXTRA_DATA|MMAPED_IPR);
- flags.update(_flags, PUBLIC_FLAGS);
+ _contextId = context_id;
+ privateFlags.set(VALID_CONTEXT_ID);
}
/**
* allocated Request object.
*/
void
- setVirt(int _asid, Addr _vaddr, int _size, Flags _flags, Addr _pc)
+ setVirt(int asid, Addr vaddr, unsigned size, Flags flags, MasterID mid,
+ Addr pc)
{
- assert(_size >= 0);
- asid = _asid;
- vaddr = _vaddr;
- size = _size;
- pc = _pc;
- time = curTick;
+ _asid = asid;
+ _vaddr = vaddr;
+ _size = size;
+ _masterId = mid;
+ _pc = pc;
+ _time = curTick();
- flags.set(VALID_VADDR|VALID_SIZE|VALID_PC);
- flags.clear(VALID_PADDR|VALID_EXTRA_DATA|MMAPED_IPR);
- flags.update(_flags, PUBLIC_FLAGS);
+ _flags.clear(~STICKY_FLAGS);
+ _flags.set(flags);
+ privateFlags.clear(~STICKY_PRIVATE_FLAGS);
+ privateFlags.set(VALID_VADDR|VALID_SIZE|VALID_PC);
+ depth = 0;
+ accessDelta = 0;
+ translateDelta = 0;
}
/**
- * Set just the physical address. This should only be used to
- * record the result of a translation, and thus the vaddr must be
- * valid before this method is called. Otherwise, use setPhys()
- * to guarantee that the size and flags are also set.
+ * Set just the physical address. This usually used to record the
+ * result of a translation. However, when using virtualized CPUs
+ * setPhys() is sometimes called to finalize a physical address
+ * without a virtual address, so we can't check if the virtual
+ * address is valid.
*/
void
- setPaddr(Addr _paddr)
+ setPaddr(Addr paddr)
{
- assert(flags.isSet(VALID_VADDR));
- paddr = _paddr;
- flags.set(VALID_PADDR);
+ _paddr = paddr;
+ privateFlags.set(VALID_PADDR);
}
/**
*/
void splitOnVaddr(Addr split_addr, RequestPtr &req1, RequestPtr &req2)
{
- assert(flags.isSet(VALID_VADDR));
- assert(flags.noneSet(VALID_PADDR));
- assert(split_addr > vaddr && split_addr < vaddr + size);
- req1 = new Request;
- *req1 = *this;
- req2 = new Request;
- *req2 = *this;
- req1->size = split_addr - vaddr;
- req2->vaddr = split_addr;
- req2->size = size - req1->size;
+ assert(privateFlags.isSet(VALID_VADDR));
+ assert(privateFlags.noneSet(VALID_PADDR));
+ assert(split_addr > _vaddr && split_addr < _vaddr + _size);
+ req1 = new Request(*this);
+ req2 = new Request(*this);
+ req1->_size = split_addr - _vaddr;
+ req2->_vaddr = split_addr;
+ req2->_size = _size - req1->_size;
}
/**
* Accessor for paddr.
*/
+ bool
+ hasPaddr() const
+ {
+ return privateFlags.isSet(VALID_PADDR);
+ }
+
Addr
- getPaddr()
+ getPaddr() const
{
- assert(flags.isSet(VALID_PADDR));
- return paddr;
+ assert(privateFlags.isSet(VALID_PADDR));
+ return _paddr;
}
+ /**
+ * Time for the TLB/table walker to successfully translate this request.
+ */
+ Tick translateDelta;
+
+ /**
+ * Access latency to complete this memory transaction not including
+ * translation time.
+ */
+ Tick accessDelta;
+
+ /**
+ * Level of the cache hierachy where this request was responded to
+ * (e.g. 0 = L1; 1 = L2).
+ */
+ mutable int depth;
+
/**
* Accessor for size.
*/
- int
- getSize()
+ bool
+ hasSize() const
{
- assert(flags.isSet(VALID_SIZE));
- return size;
+ return privateFlags.isSet(VALID_SIZE);
}
- /** Accessor for time. */
- Tick
- getTime()
+ unsigned
+ getSize() const
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- return time;
+ assert(privateFlags.isSet(VALID_SIZE));
+ return _size;
}
- void
- setTime(Tick when)
+ /** Accessor for time. */
+ Tick
+ time() const
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- time = when;
+ assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR));
+ return _time;
}
- /** Accessor for flags. */
- Flags
- getFlags()
+ /**
+ * Accessor for atomic-op functor.
+ */
+ bool
+ hasAtomicOpFunctor()
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- return flags & PUBLIC_FLAGS;
+ return atomicOpFunctor != NULL;
}
- Flags
- anyFlags(Flags _flags)
+ AtomicOpFunctor *
+ getAtomicOpFunctor()
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- assert(_flags.noneSet(~PUBLIC_FLAGS));
- return flags.isSet(_flags);
+ assert(atomicOpFunctor != NULL);
+ return atomicOpFunctor;
}
+ /** Accessor for flags. */
Flags
- allFlags(Flags _flags)
+ getFlags()
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- assert(_flags.noneSet(~PUBLIC_FLAGS));
- return flags.allSet(_flags);
+ assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR));
+ return _flags;
}
- /** Accessor for flags. */
+ /** Note that unlike other accessors, this function sets *specific
+ flags* (ORs them in); it does not assign its argument to the
+ _flags field. Thus this method should rightly be called
+ setFlags() and not just flags(). */
void
- setFlags(Flags _flags)
+ setFlags(Flags flags)
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- assert(_flags.noneSet(~PUBLIC_FLAGS));
- flags.set(_flags);
+ assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR));
+ _flags.set(flags);
}
void
- clearFlags(Flags _flags)
+ setMemSpaceConfigFlags(MemSpaceConfigFlags extraFlags)
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- assert(_flags.noneSet(~PUBLIC_FLAGS));
- flags.clear(_flags);
+ assert(privateFlags.isSet(VALID_PADDR | VALID_VADDR));
+ _memSpaceConfigFlags.set(extraFlags);
}
- void
- clearFlags()
+ /** Accessor function for vaddr.*/
+ bool
+ hasVaddr() const
{
- assert(flags.isSet(VALID_PADDR|VALID_VADDR));
- flags.clear(PUBLIC_FLAGS);
+ return privateFlags.isSet(VALID_VADDR);
}
- /** Accessor function for vaddr.*/
Addr
- getVaddr()
+ getVaddr() const
{
- assert(flags.isSet(VALID_VADDR));
- return vaddr;
+ assert(privateFlags.isSet(VALID_VADDR));
+ return _vaddr;
}
- /** Accessor function for asid.*/
- int
- getAsid()
+ /** Accesssor for the requestor id. */
+ MasterID
+ masterId() const
{
- assert(flags.isSet(VALID_VADDR));
- return asid;
+ return _masterId;
}
- /** Accessor function for asi.*/
- uint8_t
- getAsi()
+ uint32_t
+ taskId() const
{
- assert(flags.isSet(VALID_VADDR));
- return flags & ASI_BITS;
+ return _taskId;
}
- /** Accessor function for asi.*/
void
- setAsi(uint8_t a)
- {
- assert(flags.isSet(VALID_VADDR));
- flags.update(a, ASI_BITS);
+ taskId(uint32_t id) {
+ _taskId = id;
}
- /** Accessor function for asi.*/
- bool
- isMmapedIpr()
+ /** Accessor function for asid.*/
+ int
+ getAsid() const
{
- assert(flags.isSet(VALID_PADDR));
- return flags.isSet(MMAPED_IPR);
+ assert(privateFlags.isSet(VALID_VADDR));
+ return _asid;
}
- /** Accessor function for asi.*/
+ /** Accessor function for asid.*/
void
- setMmapedIpr(bool r)
+ setAsid(int asid)
+ {
+ _asid = asid;
+ }
+
+ /** Accessor function for architecture-specific flags.*/
+ ArchFlagsType
+ getArchFlags() const
{
- assert(VALID_VADDR);
- flags.set(MMAPED_IPR);
+ assert(privateFlags.isSet(VALID_PADDR|VALID_VADDR));
+ return _flags & ARCH_BITS;
}
/** Accessor function to check if sc result is valid. */
bool
- extraDataValid()
+ extraDataValid() const
{
- return flags.isSet(VALID_EXTRA_DATA);
+ return privateFlags.isSet(VALID_EXTRA_DATA);
}
/** Accessor function for store conditional return value.*/
uint64_t
getExtraData() const
{
- assert(flags.isSet(VALID_EXTRA_DATA));
- return extraData;
+ assert(privateFlags.isSet(VALID_EXTRA_DATA));
+ return _extraData;
}
/** Accessor function for store conditional return value.*/
void
- setExtraData(uint64_t _extraData)
+ setExtraData(uint64_t extraData)
{
- extraData = _extraData;
- flags.set(VALID_EXTRA_DATA);
+ _extraData = extraData;
+ privateFlags.set(VALID_EXTRA_DATA);
}
bool
hasContextId() const
{
- return flags.isSet(VALID_CONTEXT_ID);
+ return privateFlags.isSet(VALID_CONTEXT_ID);
}
/** Accessor function for context ID.*/
- int
+ ContextID
contextId() const
{
- assert(flags.isSet(VALID_CONTEXT_ID));
+ assert(privateFlags.isSet(VALID_CONTEXT_ID));
return _contextId;
}
- /** Accessor function for thread ID. */
- int
- threadId() const
+ void
+ setPC(Addr pc)
{
- assert(flags.isSet(VALID_THREAD_ID));
- return _threadId;
+ privateFlags.set(VALID_PC);
+ _pc = pc;
}
bool
hasPC() const
{
- return flags.isSet(VALID_PC);
+ return privateFlags.isSet(VALID_PC);
}
/** Accessor function for pc.*/
Addr
getPC() const
{
- assert(flags.isSet(VALID_PC));
- return pc;
+ assert(privateFlags.isSet(VALID_PC));
+ return _pc;
+ }
+
+ /**
+ * Increment/Get the depth at which this request is responded to.
+ * This currently happens when the request misses in any cache level.
+ */
+ void incAccessDepth() const { depth++; }
+ int getAccessDepth() const { return depth; }
+
+ /**
+ * Set/Get the time taken for this request to be successfully translated.
+ */
+ void setTranslateLatency() { translateDelta = curTick() - _time; }
+ Tick getTranslateLatency() const { return translateDelta; }
+
+ /**
+ * Set/Get the time taken to complete this request's access, not including
+ * the time to successfully translate the request.
+ */
+ void setAccessLatency() { accessDelta = curTick() - _time - translateDelta; }
+ Tick getAccessLatency() const { return accessDelta; }
+
+ /**
+ * Accessor for the sequence number of instruction that creates the
+ * request.
+ */
+ bool
+ hasInstSeqNum() const
+ {
+ return privateFlags.isSet(VALID_INST_SEQ_NUM);
+ }
+
+ InstSeqNum
+ getReqInstSeqNum() const
+ {
+ assert(privateFlags.isSet(VALID_INST_SEQ_NUM));
+ return _reqInstSeqNum;
+ }
+
+ void
+ setReqInstSeqNum(const InstSeqNum seq_num)
+ {
+ privateFlags.set(VALID_INST_SEQ_NUM);
+ _reqInstSeqNum = seq_num;
+ }
+
+ /** Accessor functions for flags. Note that these are for testing
+ only; setting flags should be done via setFlags(). */
+ bool isUncacheable() const { return _flags.isSet(UNCACHEABLE); }
+ bool isStrictlyOrdered() const { return _flags.isSet(STRICT_ORDER); }
+ bool isInstFetch() const { return _flags.isSet(INST_FETCH); }
+ bool isPrefetch() const { return _flags.isSet(PREFETCH); }
+ bool isLLSC() const { return _flags.isSet(LLSC); }
+ bool isPriv() const { return _flags.isSet(PRIVILEGED); }
+ bool isLockedRMW() const { return _flags.isSet(LOCKED_RMW); }
+ bool isSwap() const { return _flags.isSet(MEM_SWAP|MEM_SWAP_COND); }
+ bool isCondSwap() const { return _flags.isSet(MEM_SWAP_COND); }
+ bool isMmappedIpr() const { return _flags.isSet(MMAPPED_IPR); }
+ bool isSecure() const { return _flags.isSet(SECURE); }
+ bool isPTWalk() const { return _flags.isSet(PT_WALK); }
+ bool isAcquire() const { return _flags.isSet(ACQUIRE); }
+ bool isRelease() const { return _flags.isSet(RELEASE); }
+ bool isKernel() const { return _flags.isSet(KERNEL); }
+ bool isAtomicReturn() const { return _flags.isSet(ATOMIC_RETURN_OP); }
+ bool isAtomicNoReturn() const { return _flags.isSet(ATOMIC_NO_RETURN_OP); }
+
+ bool
+ isAtomic() const
+ {
+ return _flags.isSet(ATOMIC_RETURN_OP) ||
+ _flags.isSet(ATOMIC_NO_RETURN_OP);
+ }
+
+ /**
+ * Accessor functions for the destination of a memory request. The
+ * destination flag can specify a point of reference for the
+ * operation (e.g. a cache block clean to the the point of
+ * unification). At the moment the destination is only used by the
+ * cache maintenance operations.
+ */
+ bool isToPOU() const { return _flags.isSet(DST_POU); }
+ bool isToPOC() const { return _flags.isSet(DST_POC); }
+ Flags getDest() const { return _flags & DST_BITS; }
+
+ /**
+ * Accessor functions for the memory space configuration flags and used by
+ * GPU ISAs such as the Heterogeneous System Architecture (HSA). Note that
+ * these are for testing only; setting extraFlags should be done via
+ * setMemSpaceConfigFlags().
+ */
+ bool isScoped() const { return _memSpaceConfigFlags.isSet(SCOPE_VALID); }
+
+ bool
+ isWavefrontScope() const
+ {
+ assert(isScoped());
+ return _memSpaceConfigFlags.isSet(WAVEFRONT_SCOPE);
+ }
+
+ bool
+ isWorkgroupScope() const
+ {
+ assert(isScoped());
+ return _memSpaceConfigFlags.isSet(WORKGROUP_SCOPE);
+ }
+
+ bool
+ isDeviceScope() const
+ {
+ assert(isScoped());
+ return _memSpaceConfigFlags.isSet(DEVICE_SCOPE);
+ }
+
+ bool
+ isSystemScope() const
+ {
+ assert(isScoped());
+ return _memSpaceConfigFlags.isSet(SYSTEM_SCOPE);
+ }
+
+ bool
+ isGlobalSegment() const
+ {
+ return _memSpaceConfigFlags.isSet(GLOBAL_SEGMENT) ||
+ (!isGroupSegment() && !isPrivateSegment() &&
+ !isKernargSegment() && !isReadonlySegment() &&
+ !isSpillSegment() && !isArgSegment());
}
- /** Accessor Function to Check Cacheability. */
- bool isUncacheable() const { return flags.isSet(UNCACHEABLE); }
- bool isInstRead() const { return flags.isSet(INST_READ); }
- bool isLlsc() const { return flags.isSet(LLSC); }
- bool isLocked() const { return flags.isSet(LOCKED); }
- bool isSwap() const { return flags.isSet(MEM_SWAP|MEM_SWAP_COND); }
- bool isCondSwap() const { return flags.isSet(MEM_SWAP_COND); }
+ bool
+ isGroupSegment() const
+ {
+ return _memSpaceConfigFlags.isSet(GROUP_SEGMENT);
+ }
bool
- isMisaligned() const
+ isPrivateSegment() const
{
- if (flags.isSet(NO_ALIGN_FAULT))
- return false;
+ return _memSpaceConfigFlags.isSet(PRIVATE_SEGMENT);
+ }
- if ((vaddr & 0x1))
- return true;
+ bool
+ isKernargSegment() const
+ {
+ return _memSpaceConfigFlags.isSet(KERNARG_SEGMENT);
+ }
- if (flags.isSet(NO_HALF_WORD_ALIGN_FAULT))
- return false;
+ bool
+ isReadonlySegment() const
+ {
+ return _memSpaceConfigFlags.isSet(READONLY_SEGMENT);
+ }
- if ((vaddr & 0x2))
- return true;
+ bool
+ isSpillSegment() const
+ {
+ return _memSpaceConfigFlags.isSet(SPILL_SEGMENT);
+ }
- return false;
+ bool
+ isArgSegment() const
+ {
+ return _memSpaceConfigFlags.isSet(ARG_SEGMENT);
}
+
+ /**
+ * Accessor functions to determine whether this request is part of
+ * a cache maintenance operation. At the moment three operations
+ * are supported:
+
+ * 1) A cache clean operation updates all copies of a memory
+ * location to the point of reference,
+ * 2) A cache invalidate operation invalidates all copies of the
+ * specified block in the memory above the point of reference,
+ * 3) A clean and invalidate operation is a combination of the two
+ * operations.
+ * @{ */
+ bool isCacheClean() const { return _flags.isSet(CLEAN); }
+ bool isCacheInvalidate() const { return _flags.isSet(INVALIDATE); }
+ bool isCacheMaintenance() const { return _flags.isSet(CLEAN|INVALIDATE); }
+ /** @} */
};
#endif // __MEM_REQUEST_HH__
projects / gem5.git / blobdiff