diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh index d83dc868f..2076d67d1 100644 --- a/src/cpu/o3/lsq_unit.hh +++ b/src/cpu/o3/lsq_unit.hh @@ -39,6 +39,7 @@ #include "arch/faults.hh" #include "arch/locked_mem.hh" +#include "arch/mmapped_ipr.hh" #include "base/fast_alloc.hh" #include "base/hashmap.hh" #include "config/full_system.hh" @@ -578,6 +579,43 @@ LSQUnit::read(Request *req, Request *sreqLow, Request *sreqHigh, load_inst->recordResult = true; } + if (req->isMmappedIpr()) { + assert(!load_inst->memData); + load_inst->memData = new uint8_t[64]; + + ThreadContext *thread = cpu->tcBase(lsqID); + Tick delay; + PacketPtr data_pkt = + new Packet(req, MemCmd::ReadReq, Packet::Broadcast); + + if (!TheISA::HasUnalignedMemAcc || !sreqLow) { + data_pkt->dataStatic(load_inst->memData); + delay = TheISA::handleIprRead(thread, data_pkt); + } else { + assert(sreqLow->isMmappedIpr() && sreqHigh->isMmappedIpr()); + PacketPtr fst_data_pkt = + new Packet(sreqLow, MemCmd::ReadReq, Packet::Broadcast); + PacketPtr snd_data_pkt = + new Packet(sreqHigh, MemCmd::ReadReq, Packet::Broadcast); + + fst_data_pkt->dataStatic(load_inst->memData); + snd_data_pkt->dataStatic(load_inst->memData + sreqLow->getSize()); + + delay = TheISA::handleIprRead(thread, fst_data_pkt); + unsigned delay2 = TheISA::handleIprRead(thread, snd_data_pkt); + if (delay2 > delay) + delay = delay2; + + delete sreqLow; + delete sreqHigh; + delete fst_data_pkt; + delete snd_data_pkt; + } + WritebackEvent *wb = new WritebackEvent(load_inst, data_pkt, this); + cpu->schedule(wb, curTick() + delay); + return NoFault; + } + while (store_idx != -1) { // End once we've reached the top of the LSQ if (store_idx == storeWBIdx) { diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh index 2c2b30b8a..79a20a673 100644 --- a/src/cpu/o3/lsq_unit_impl.hh +++ b/src/cpu/o3/lsq_unit_impl.hh @@ -716,6 +716,9 @@ LSQUnit::writebackStores() DynInstPtr inst = storeQueue[storeWBIdx].inst; Request *req = storeQueue[storeWBIdx].req; + RequestPtr sreqLow = storeQueue[storeWBIdx].sreqLow; + RequestPtr sreqHigh = storeQueue[storeWBIdx].sreqHigh; + storeQueue[storeWBIdx].committed = true; assert(!inst->memData); @@ -741,9 +744,6 @@ LSQUnit::writebackStores() data_pkt->dataStatic(inst->memData); data_pkt->senderState = state; } else { - RequestPtr sreqLow = storeQueue[storeWBIdx].sreqLow; - RequestPtr sreqHigh = storeQueue[storeWBIdx].sreqHigh; - // Create two packets if the store is split in two. data_pkt = new Packet(sreqLow, command, Packet::Broadcast); snd_data_pkt = new Packet(sreqHigh, command, Packet::Broadcast); @@ -794,20 +794,40 @@ LSQUnit::writebackStores() state->noWB = true; } - if (!sendStore(data_pkt)) { + bool split = + TheISA::HasUnalignedMemAcc && storeQueue[storeWBIdx].isSplit; + + ThreadContext *thread = cpu->tcBase(lsqID); + + if (req->isMmappedIpr()) { + assert(!inst->isStoreConditional()); + TheISA::handleIprWrite(thread, data_pkt); + delete data_pkt; + if (split) { + assert(snd_data_pkt->req->isMmappedIpr()); + TheISA::handleIprWrite(thread, snd_data_pkt); + delete snd_data_pkt; + delete sreqLow; + delete sreqHigh; + } + delete state; + delete req; + completeStore(storeWBIdx); + incrStIdx(storeWBIdx); + } else if (!sendStore(data_pkt)) { DPRINTF(IEW, "D-Cache became blocked when writing [sn:%lli], will" "retry later\n", inst->seqNum); // Need to store the second packet, if split. - if (TheISA::HasUnalignedMemAcc && storeQueue[storeWBIdx].isSplit) { + if (split) { state->pktToSend = true; state->pendingPacket = snd_data_pkt; } } else { // If split, try to send the second packet too - if (TheISA::HasUnalignedMemAcc && storeQueue[storeWBIdx].isSplit) { + if (split) { assert(snd_data_pkt); // Ensure there are enough ports to use.