diff options
Diffstat (limited to 'src/mem/ruby/system/StoreBuffer.cc')
| -rw-r--r-- | src/mem/ruby/system/StoreBuffer.cc | 302 |
1 files changed, 302 insertions, 0 deletions
diff --git a/src/mem/ruby/system/StoreBuffer.cc b/src/mem/ruby/system/StoreBuffer.cc new file mode 100644 index 000000000..280decdd8 --- /dev/null +++ b/src/mem/ruby/system/StoreBuffer.cc @@ -0,0 +1,302 @@ + +/* + * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer; + * redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution; + * neither the name of the copyright holders nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +/* + * $Id$ + * + */ + +#include "mem/ruby/common/Global.hh" +#include "mem/ruby/config/RubyConfig.hh" +#include "mem/ruby/system/StoreBuffer.hh" +#include "mem/ruby/slicc_interface/AbstractChip.hh" +#include "mem/ruby/system/System.hh" +#include "mem/ruby/common/Driver.hh" +#include "mem/gems_common/Vector.hh" +#include "mem/ruby/eventqueue/RubyEventQueue.hh" +#include "mem/ruby/profiler/AddressProfiler.hh" +#include "mem/ruby/system/Sequencer.hh" +#include "mem/ruby/common/SubBlock.hh" +#include "mem/ruby/profiler/Profiler.hh" +#include "mem/packet.hh" + +// *** Begin Helper class *** +struct StoreBufferEntry { + StoreBufferEntry() {} // So we can allocate a vector of StoreBufferEntries + StoreBufferEntry(const SubBlock& block, CacheRequestType type, const Address& pc, AccessModeType access_mode, int size, int thread) : m_subblock(block) { + m_type = type; + m_pc = pc; + m_access_mode = access_mode; + m_size = size; + m_thread = thread; + m_time = g_eventQueue_ptr->getTime(); + } + + void print(ostream& out) const + { + out << "[StoreBufferEntry: " + << "SubBlock: " << m_subblock + << ", Type: " << m_type + << ", PC: " << m_pc + << ", AccessMode: " << m_access_mode + << ", Size: " << m_size + << ", Thread: " << m_thread + << ", Time: " << m_time + << "]"; + } + + SubBlock m_subblock; + CacheRequestType m_type; + Address m_pc; + AccessModeType m_access_mode; + int m_size; + int m_thread; + Time m_time; +}; + +extern inline +ostream& operator<<(ostream& out, const StoreBufferEntry& obj) +{ + obj.print(out); + out << flush; + return out; +} + +// *** End Helper class *** + +const int MAX_ENTRIES = 128; + +static void inc_index(int& index) +{ + index++; + if (index >= MAX_ENTRIES) { + index = 0; + } +} + +StoreBuffer::StoreBuffer(AbstractChip* chip_ptr, int version) : + m_store_cache() +{ + m_chip_ptr = chip_ptr; + m_version = version; + m_queue_ptr = new Vector<StoreBufferEntry>(MAX_ENTRIES); + m_queue_ptr->setSize(MAX_ENTRIES); + m_pending = false; + m_seen_atomic = false; + m_head = 0; + m_tail = 0; + m_size = 0; + m_deadlock_check_scheduled = false; +} + +StoreBuffer::~StoreBuffer() +{ + delete m_queue_ptr; +} + +// Used only to check for deadlock +void StoreBuffer::wakeup() +{ + // Check for deadlock of any of the requests + Time current_time = g_eventQueue_ptr->getTime(); + + int queue_pointer = m_head; + for (int i=0; i<m_size; i++) { + if (current_time - (getEntry(queue_pointer).m_time) >= g_DEADLOCK_THRESHOLD) { + WARN_EXPR(getEntry(queue_pointer)); + WARN_EXPR(m_chip_ptr->getID()); + WARN_EXPR(current_time); + ERROR_MSG("Possible Deadlock detected"); + } + inc_index(queue_pointer); + } + + if (m_size > 0) { // If there are still outstanding requests, keep checking + g_eventQueue_ptr->scheduleEvent(this, g_DEADLOCK_THRESHOLD); + } else { + m_deadlock_check_scheduled = false; + } +} + +void StoreBuffer::printConfig(ostream& out) +{ + out << "Store buffer entries: " << MAX_ENTRIES << " (Only valid if TSO is enabled)" << endl; +} + +// Handle an incoming store request, this method is responsible for +// calling hitCallback as needed +void +StoreBuffer::insertStore(Packet* pkt, const CacheMsg& request) +{ + Address addr = request.getAddress(); + CacheRequestType type = request.getType(); + Address pc = request.getProgramCounter(); + AccessModeType access_mode = request.getAccessMode(); + int size = request.getSize(); + int threadID = request.getThreadID(); + + DEBUG_MSG(STOREBUFFER_COMP, MedPrio, "insertStore"); + DEBUG_EXPR(STOREBUFFER_COMP, MedPrio, g_eventQueue_ptr->getTime()); + assert((type == CacheRequestType_ST) || (type == CacheRequestType_ATOMIC)); + assert(isReady()); + + // See if we should schedule a deadlock check + if (m_deadlock_check_scheduled == false) { + g_eventQueue_ptr->scheduleEvent(this, g_DEADLOCK_THRESHOLD); + m_deadlock_check_scheduled = true; + } + + // Perform the hit-callback for the store + SubBlock subblock(addr, size); + if(type == CacheRequestType_ST) { + g_system_ptr->getDriver()->hitCallback(pkt); + assert(subblock.getSize() != 0); + } else { + // wait to perform the hitCallback until later for Atomics + } + + // Perform possible pre-fetch + if(!isEmpty()) { + Packet new_pkt(pkt); + pkt->req->setFlags(Request::PREFETCH); + m_chip_ptr->getSequencer(m_version)->makeRequest(&new_pkt); + } + + // Update the StoreCache + m_store_cache.add(subblock); + + // Enqueue the entry + StoreBufferEntry entry(subblock, type, pc, access_mode, size, threadID); // FIXME + enqueue(entry); + + if(type == CacheRequestType_ATOMIC) { + m_seen_atomic = true; + } + + processHeadOfQueue(); +} + +void StoreBuffer::callBack(const Address& addr, DataBlock& data, Packet* pkt) +{ + DEBUG_MSG(STOREBUFFER_COMP, MedPrio, "callBack"); + DEBUG_EXPR(STOREBUFFER_COMP, MedPrio, g_eventQueue_ptr->getTime()); + assert(!isEmpty()); + assert(m_pending == true); + assert(line_address(addr) == addr); + assert(line_address(m_pending_address) == addr); + assert(line_address(peek().m_subblock.getAddress()) == addr); + CacheRequestType type = peek().m_type; + //int threadID = peek().m_thread; + assert((type == CacheRequestType_ST) || (type == CacheRequestType_ATOMIC)); + m_pending = false; + + // If oldest entry was ATOMIC, perform the callback + if(type == CacheRequestType_ST) { + // We already performed the call back for the store at insert time + } else { + // We waited to perform the hitCallback until now for Atomics + peek().m_subblock.mergeFrom(data); // copy the correct bytes from DataBlock into the SubBlock for the Load part of the atomic Load/Store + g_system_ptr->getDriver()->hitCallback(pkt); + m_seen_atomic = false; + + /// FIXME - record the time spent in the store buffer - split out ST vs ATOMIC + } + assert(peek().m_subblock.getSize() != 0); + + // Apply the head entry to the datablock + peek().m_subblock.mergeTo(data); // For both the Store and Atomic cases + + // Update the StoreCache + m_store_cache.remove(peek().m_subblock); + + // Dequeue the entry from the store buffer + dequeue(); + + if (isEmpty()) { + assert(m_store_cache.isEmpty()); + } + + if(type == CacheRequestType_ATOMIC) { + assert(isEmpty()); + } + + // See if we can remove any more entries + processHeadOfQueue(); +} + +void StoreBuffer::processHeadOfQueue() +{ + if(!isEmpty() && !m_pending) { + StoreBufferEntry& entry = peek(); + assert(m_pending == false); + m_pending = true; + m_pending_address = entry.m_subblock.getAddress(); + CacheMsg request(entry.m_subblock.getAddress(), entry.m_subblock.getAddress(), entry.m_type, entry.m_pc, entry.m_access_mode, entry.m_size, PrefetchBit_No, 0, Address(0), entry.m_thread); + m_chip_ptr->getSequencer(m_version)->doRequest(request); + } +} + +bool StoreBuffer::isReady() const +{ + return ((m_size < MAX_ENTRIES) && (!m_seen_atomic)); +} + +// Queue implementation methods + +StoreBufferEntry& StoreBuffer::peek() +{ + return getEntry(m_head); +} + +void StoreBuffer::dequeue() +{ + assert(m_size > 0); + m_size--; + inc_index(m_head); +} + +void StoreBuffer::enqueue(const StoreBufferEntry& entry) +{ + // assert(isReady()); + (*m_queue_ptr)[m_tail] = entry; + m_size++; + g_system_ptr->getProfiler()->storeBuffer(m_size, m_store_cache.size()); + inc_index(m_tail); +} + +StoreBufferEntry& StoreBuffer::getEntry(int index) +{ + return (*m_queue_ptr)[index]; +} + +void StoreBuffer::print(ostream& out) const +{ + out << "[StoreBuffer]"; +} + |